Print this page
7127 remove -Wno-missing-braces from Makefile.uts
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/nfs/nfs4_vnops.c
+++ new/usr/src/uts/common/fs/nfs/nfs4_vnops.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
24 24 */
25 25
26 26 /*
27 27 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
28 28 * Use is subject to license terms.
29 29 */
30 30
31 31 /*
32 32 * Copyright 1983,1984,1985,1986,1987,1988,1989 AT&T.
33 33 * All Rights Reserved
34 34 */
35 35
36 36 /*
37 37 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
38 38 */
39 39
40 40 #include <sys/param.h>
41 41 #include <sys/types.h>
42 42 #include <sys/systm.h>
43 43 #include <sys/cred.h>
44 44 #include <sys/time.h>
45 45 #include <sys/vnode.h>
46 46 #include <sys/vfs.h>
47 47 #include <sys/vfs_opreg.h>
48 48 #include <sys/file.h>
49 49 #include <sys/filio.h>
50 50 #include <sys/uio.h>
51 51 #include <sys/buf.h>
52 52 #include <sys/mman.h>
53 53 #include <sys/pathname.h>
54 54 #include <sys/dirent.h>
55 55 #include <sys/debug.h>
56 56 #include <sys/vmsystm.h>
57 57 #include <sys/fcntl.h>
58 58 #include <sys/flock.h>
59 59 #include <sys/swap.h>
60 60 #include <sys/errno.h>
61 61 #include <sys/strsubr.h>
62 62 #include <sys/sysmacros.h>
63 63 #include <sys/kmem.h>
64 64 #include <sys/cmn_err.h>
65 65 #include <sys/pathconf.h>
66 66 #include <sys/utsname.h>
67 67 #include <sys/dnlc.h>
68 68 #include <sys/acl.h>
69 69 #include <sys/systeminfo.h>
70 70 #include <sys/policy.h>
71 71 #include <sys/sdt.h>
72 72 #include <sys/list.h>
73 73 #include <sys/stat.h>
74 74 #include <sys/zone.h>
75 75
76 76 #include <rpc/types.h>
77 77 #include <rpc/auth.h>
78 78 #include <rpc/clnt.h>
79 79
80 80 #include <nfs/nfs.h>
81 81 #include <nfs/nfs_clnt.h>
82 82 #include <nfs/nfs_acl.h>
83 83 #include <nfs/lm.h>
84 84 #include <nfs/nfs4.h>
85 85 #include <nfs/nfs4_kprot.h>
86 86 #include <nfs/rnode4.h>
87 87 #include <nfs/nfs4_clnt.h>
88 88
89 89 #include <vm/hat.h>
90 90 #include <vm/as.h>
91 91 #include <vm/page.h>
92 92 #include <vm/pvn.h>
93 93 #include <vm/seg.h>
94 94 #include <vm/seg_map.h>
95 95 #include <vm/seg_kpm.h>
96 96 #include <vm/seg_vn.h>
97 97
98 98 #include <fs/fs_subr.h>
99 99
100 100 #include <sys/ddi.h>
101 101 #include <sys/int_fmtio.h>
102 102 #include <sys/fs/autofs.h>
103 103
104 104 typedef struct {
105 105 nfs4_ga_res_t *di_garp;
106 106 cred_t *di_cred;
107 107 hrtime_t di_time_call;
108 108 } dirattr_info_t;
109 109
110 110 typedef enum nfs4_acl_op {
111 111 NFS4_ACL_GET,
112 112 NFS4_ACL_SET
113 113 } nfs4_acl_op_t;
114 114
115 115 static struct lm_sysid *nfs4_find_sysid(mntinfo4_t *mi);
116 116
117 117 static void nfs4_update_dircaches(change_info4 *, vnode_t *, vnode_t *,
118 118 char *, dirattr_info_t *);
119 119
120 120 static void nfs4close_otw(rnode4_t *, cred_t *, nfs4_open_owner_t *,
121 121 nfs4_open_stream_t *, int *, int *, nfs4_close_type_t,
122 122 nfs4_error_t *, int *);
123 123 static int nfs4_rdwrlbn(vnode_t *, page_t *, u_offset_t, size_t, int,
124 124 cred_t *);
125 125 static int nfs4write(vnode_t *, caddr_t, u_offset_t, int, cred_t *,
126 126 stable_how4 *);
127 127 static int nfs4read(vnode_t *, caddr_t, offset_t, int, size_t *,
128 128 cred_t *, bool_t, struct uio *);
129 129 static int nfs4setattr(vnode_t *, struct vattr *, int, cred_t *,
130 130 vsecattr_t *);
131 131 static int nfs4openattr(vnode_t *, vnode_t **, int, cred_t *);
132 132 static int nfs4lookup(vnode_t *, char *, vnode_t **, cred_t *, int);
133 133 static int nfs4lookup_xattr(vnode_t *, char *, vnode_t **, int, cred_t *);
134 134 static int nfs4lookupvalidate_otw(vnode_t *, char *, vnode_t **, cred_t *);
135 135 static int nfs4lookupnew_otw(vnode_t *, char *, vnode_t **, cred_t *);
136 136 static int nfs4mknod(vnode_t *, char *, struct vattr *, enum vcexcl,
137 137 int, vnode_t **, cred_t *);
138 138 static int nfs4open_otw(vnode_t *, char *, struct vattr *, vnode_t **,
139 139 cred_t *, int, int, enum createmode4, int);
140 140 static int nfs4rename(vnode_t *, char *, vnode_t *, char *, cred_t *,
141 141 caller_context_t *);
142 142 static int nfs4rename_persistent_fh(vnode_t *, char *, vnode_t *,
143 143 vnode_t *, char *, cred_t *, nfsstat4 *);
144 144 static int nfs4rename_volatile_fh(vnode_t *, char *, vnode_t *,
145 145 vnode_t *, char *, cred_t *, nfsstat4 *);
146 146 static int do_nfs4readdir(vnode_t *, rddir4_cache *, cred_t *);
147 147 static void nfs4readdir(vnode_t *, rddir4_cache *, cred_t *);
148 148 static int nfs4_bio(struct buf *, stable_how4 *, cred_t *, bool_t);
149 149 static int nfs4_getapage(vnode_t *, u_offset_t, size_t, uint_t *,
150 150 page_t *[], size_t, struct seg *, caddr_t,
151 151 enum seg_rw, cred_t *);
152 152 static void nfs4_readahead(vnode_t *, u_offset_t, caddr_t, struct seg *,
153 153 cred_t *);
154 154 static int nfs4_sync_putapage(vnode_t *, page_t *, u_offset_t, size_t,
155 155 int, cred_t *);
156 156 static int nfs4_sync_pageio(vnode_t *, page_t *, u_offset_t, size_t,
157 157 int, cred_t *);
158 158 static int nfs4_commit(vnode_t *, offset4, count4, cred_t *);
159 159 static void nfs4_set_mod(vnode_t *);
160 160 static void nfs4_get_commit(vnode_t *);
161 161 static void nfs4_get_commit_range(vnode_t *, u_offset_t, size_t);
162 162 static int nfs4_putpage_commit(vnode_t *, offset_t, size_t, cred_t *);
163 163 static int nfs4_commit_vp(vnode_t *, u_offset_t, size_t, cred_t *, int);
164 164 static int nfs4_sync_commit(vnode_t *, page_t *, offset3, count3,
165 165 cred_t *);
166 166 static void do_nfs4_async_commit(vnode_t *, page_t *, offset3, count3,
167 167 cred_t *);
168 168 static int nfs4_update_attrcache(nfsstat4, nfs4_ga_res_t *,
169 169 hrtime_t, vnode_t *, cred_t *);
170 170 static int nfs4_open_non_reg_file(vnode_t **, int, cred_t *);
171 171 static int nfs4_safelock(vnode_t *, const struct flock64 *, cred_t *);
172 172 static void nfs4_register_lock_locally(vnode_t *, struct flock64 *, int,
173 173 u_offset_t);
174 174 static int nfs4_lockrelease(vnode_t *, int, offset_t, cred_t *);
175 175 static int nfs4_block_and_wait(clock_t *, rnode4_t *);
176 176 static cred_t *state_to_cred(nfs4_open_stream_t *);
177 177 static void denied_to_flk(LOCK4denied *, flock64_t *, LOCKT4args *);
178 178 static pid_t lo_to_pid(lock_owner4 *);
179 179 static void nfs4_reinstitute_local_lock_state(vnode_t *, flock64_t *,
180 180 cred_t *, nfs4_lock_owner_t *);
181 181 static void push_reinstate(vnode_t *, int, flock64_t *, cred_t *,
182 182 nfs4_lock_owner_t *);
183 183 static int open_and_get_osp(vnode_t *, cred_t *, nfs4_open_stream_t **);
184 184 static void nfs4_delmap_callback(struct as *, void *, uint_t);
185 185 static void nfs4_free_delmapcall(nfs4_delmapcall_t *);
186 186 static nfs4_delmapcall_t *nfs4_init_delmapcall();
187 187 static int nfs4_find_and_delete_delmapcall(rnode4_t *, int *);
188 188 static int nfs4_is_acl_mask_valid(uint_t, nfs4_acl_op_t);
189 189 static int nfs4_create_getsecattr_return(vsecattr_t *, vsecattr_t *,
190 190 uid_t, gid_t, int);
191 191
192 192 /*
193 193 * Routines that implement the setting of v4 args for the misc. ops
194 194 */
195 195 static void nfs4args_lock_free(nfs_argop4 *);
196 196 static void nfs4args_lockt_free(nfs_argop4 *);
197 197 static void nfs4args_setattr(nfs_argop4 *, vattr_t *, vsecattr_t *,
198 198 int, rnode4_t *, cred_t *, bitmap4, int *,
199 199 nfs4_stateid_types_t *);
200 200 static void nfs4args_setattr_free(nfs_argop4 *);
201 201 static int nfs4args_verify(nfs_argop4 *, vattr_t *, enum nfs_opnum4,
202 202 bitmap4);
203 203 static void nfs4args_verify_free(nfs_argop4 *);
204 204 static void nfs4args_write(nfs_argop4 *, stable_how4, rnode4_t *, cred_t *,
205 205 WRITE4args **, nfs4_stateid_types_t *);
206 206
207 207 /*
208 208 * These are the vnode ops functions that implement the vnode interface to
209 209 * the networked file system. See more comments below at nfs4_vnodeops.
210 210 */
211 211 static int nfs4_open(vnode_t **, int, cred_t *, caller_context_t *);
212 212 static int nfs4_close(vnode_t *, int, int, offset_t, cred_t *,
213 213 caller_context_t *);
214 214 static int nfs4_read(vnode_t *, struct uio *, int, cred_t *,
215 215 caller_context_t *);
216 216 static int nfs4_write(vnode_t *, struct uio *, int, cred_t *,
217 217 caller_context_t *);
218 218 static int nfs4_ioctl(vnode_t *, int, intptr_t, int, cred_t *, int *,
219 219 caller_context_t *);
220 220 static int nfs4_setattr(vnode_t *, struct vattr *, int, cred_t *,
221 221 caller_context_t *);
222 222 static int nfs4_access(vnode_t *, int, int, cred_t *, caller_context_t *);
223 223 static int nfs4_readlink(vnode_t *, struct uio *, cred_t *,
224 224 caller_context_t *);
225 225 static int nfs4_fsync(vnode_t *, int, cred_t *, caller_context_t *);
226 226 static int nfs4_create(vnode_t *, char *, struct vattr *, enum vcexcl,
227 227 int, vnode_t **, cred_t *, int, caller_context_t *,
228 228 vsecattr_t *);
229 229 static int nfs4_remove(vnode_t *, char *, cred_t *, caller_context_t *,
230 230 int);
231 231 static int nfs4_link(vnode_t *, vnode_t *, char *, cred_t *,
232 232 caller_context_t *, int);
233 233 static int nfs4_rename(vnode_t *, char *, vnode_t *, char *, cred_t *,
234 234 caller_context_t *, int);
235 235 static int nfs4_mkdir(vnode_t *, char *, struct vattr *, vnode_t **,
236 236 cred_t *, caller_context_t *, int, vsecattr_t *);
237 237 static int nfs4_rmdir(vnode_t *, char *, vnode_t *, cred_t *,
238 238 caller_context_t *, int);
239 239 static int nfs4_symlink(vnode_t *, char *, struct vattr *, char *,
240 240 cred_t *, caller_context_t *, int);
241 241 static int nfs4_readdir(vnode_t *, struct uio *, cred_t *, int *,
242 242 caller_context_t *, int);
243 243 static int nfs4_seek(vnode_t *, offset_t, offset_t *, caller_context_t *);
244 244 static int nfs4_getpage(vnode_t *, offset_t, size_t, uint_t *,
245 245 page_t *[], size_t, struct seg *, caddr_t,
246 246 enum seg_rw, cred_t *, caller_context_t *);
247 247 static int nfs4_putpage(vnode_t *, offset_t, size_t, int, cred_t *,
248 248 caller_context_t *);
249 249 static int nfs4_map(vnode_t *, offset_t, struct as *, caddr_t *, size_t,
250 250 uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *);
251 251 static int nfs4_addmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
252 252 uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *);
253 253 static int nfs4_cmp(vnode_t *, vnode_t *, caller_context_t *);
254 254 static int nfs4_frlock(vnode_t *, int, struct flock64 *, int, offset_t,
255 255 struct flk_callback *, cred_t *, caller_context_t *);
256 256 static int nfs4_space(vnode_t *, int, struct flock64 *, int, offset_t,
257 257 cred_t *, caller_context_t *);
258 258 static int nfs4_delmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
259 259 uint_t, uint_t, uint_t, cred_t *, caller_context_t *);
260 260 static int nfs4_pageio(vnode_t *, page_t *, u_offset_t, size_t, int,
261 261 cred_t *, caller_context_t *);
262 262 static void nfs4_dispose(vnode_t *, page_t *, int, int, cred_t *,
263 263 caller_context_t *);
264 264 static int nfs4_setsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
265 265 caller_context_t *);
266 266 /*
267 267 * These vnode ops are required to be called from outside this source file,
268 268 * e.g. by ephemeral mount stub vnode ops, and so may not be declared
269 269 * as static.
270 270 */
271 271 int nfs4_getattr(vnode_t *, struct vattr *, int, cred_t *,
272 272 caller_context_t *);
273 273 void nfs4_inactive(vnode_t *, cred_t *, caller_context_t *);
274 274 int nfs4_lookup(vnode_t *, char *, vnode_t **,
275 275 struct pathname *, int, vnode_t *, cred_t *,
276 276 caller_context_t *, int *, pathname_t *);
277 277 int nfs4_fid(vnode_t *, fid_t *, caller_context_t *);
278 278 int nfs4_rwlock(vnode_t *, int, caller_context_t *);
279 279 void nfs4_rwunlock(vnode_t *, int, caller_context_t *);
280 280 int nfs4_realvp(vnode_t *, vnode_t **, caller_context_t *);
281 281 int nfs4_pathconf(vnode_t *, int, ulong_t *, cred_t *,
282 282 caller_context_t *);
283 283 int nfs4_getsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
284 284 caller_context_t *);
285 285 int nfs4_shrlock(vnode_t *, int, struct shrlock *, int, cred_t *,
286 286 caller_context_t *);
287 287
288 288 /*
289 289 * Used for nfs4_commit_vp() to indicate if we should
290 290 * wait on pending writes.
291 291 */
292 292 #define NFS4_WRITE_NOWAIT 0
293 293 #define NFS4_WRITE_WAIT 1
294 294
295 295 #define NFS4_BASE_WAIT_TIME 1 /* 1 second */
296 296
297 297 /*
298 298 * Error flags used to pass information about certain special errors
299 299 * which need to be handled specially.
300 300 */
301 301 #define NFS_EOF -98
302 302 #define NFS_VERF_MISMATCH -97
303 303
304 304 /*
305 305 * Flags used to differentiate between which operation drove the
306 306 * potential CLOSE OTW. (see nfs4_close_otw_if_necessary)
307 307 */
308 308 #define NFS4_CLOSE_OP 0x1
309 309 #define NFS4_DELMAP_OP 0x2
310 310 #define NFS4_INACTIVE_OP 0x3
311 311
312 312 #define ISVDEV(t) ((t == VBLK) || (t == VCHR) || (t == VFIFO))
313 313
314 314 /* ALIGN64 aligns the given buffer and adjust buffer size to 64 bit */
315 315 #define ALIGN64(x, ptr, sz) \
316 316 x = ((uintptr_t)(ptr)) & (sizeof (uint64_t) - 1); \
317 317 if (x) { \
318 318 x = sizeof (uint64_t) - (x); \
319 319 sz -= (x); \
320 320 ptr += (x); \
321 321 }
322 322
323 323 #ifdef DEBUG
324 324 int nfs4_client_attr_debug = 0;
325 325 int nfs4_client_state_debug = 0;
326 326 int nfs4_client_shadow_debug = 0;
327 327 int nfs4_client_lock_debug = 0;
328 328 int nfs4_seqid_sync = 0;
329 329 int nfs4_client_map_debug = 0;
330 330 static int nfs4_pageio_debug = 0;
331 331 int nfs4_client_inactive_debug = 0;
332 332 int nfs4_client_recov_debug = 0;
333 333 int nfs4_client_failover_debug = 0;
334 334 int nfs4_client_call_debug = 0;
335 335 int nfs4_client_lookup_debug = 0;
336 336 int nfs4_client_zone_debug = 0;
337 337 int nfs4_lost_rqst_debug = 0;
338 338 int nfs4_rdattrerr_debug = 0;
339 339 int nfs4_open_stream_debug = 0;
340 340
341 341 int nfs4read_error_inject;
342 342
343 343 static int nfs4_create_misses = 0;
344 344
345 345 static int nfs4_readdir_cache_shorts = 0;
346 346 static int nfs4_readdir_readahead = 0;
347 347
348 348 static int nfs4_bio_do_stop = 0;
349 349
350 350 static int nfs4_lostpage = 0; /* number of times we lost original page */
351 351
352 352 int nfs4_mmap_debug = 0;
353 353
354 354 static int nfs4_pathconf_cache_hits = 0;
355 355 static int nfs4_pathconf_cache_misses = 0;
356 356
357 357 int nfs4close_all_cnt;
358 358 int nfs4close_one_debug = 0;
359 359 int nfs4close_notw_debug = 0;
360 360
361 361 int denied_to_flk_debug = 0;
362 362 void *lockt_denied_debug;
363 363
364 364 #endif
365 365
366 366 /*
367 367 * How long to wait before trying again if OPEN_CONFIRM gets ETIMEDOUT
368 368 * or NFS4ERR_RESOURCE.
369 369 */
370 370 static int confirm_retry_sec = 30;
371 371
372 372 static int nfs4_lookup_neg_cache = 1;
373 373
374 374 /*
375 375 * number of pages to read ahead
376 376 * optimized for 100 base-T.
377 377 */
378 378 static int nfs4_nra = 4;
379 379
380 380 static int nfs4_do_symlink_cache = 1;
381 381
382 382 static int nfs4_pathconf_disable_cache = 0;
383 383
384 384 /*
385 385 * These are the vnode ops routines which implement the vnode interface to
386 386 * the networked file system. These routines just take their parameters,
387 387 * make them look networkish by putting the right info into interface structs,
388 388 * and then calling the appropriate remote routine(s) to do the work.
|
↓ open down ↓ |
388 lines elided |
↑ open up ↑ |
389 389 *
390 390 * Note on directory name lookup cacheing: If we detect a stale fhandle,
391 391 * we purge the directory cache relative to that vnode. This way, the
392 392 * user won't get burned by the cache repeatedly. See <nfs/rnode4.h> for
393 393 * more details on rnode locking.
394 394 */
395 395
396 396 struct vnodeops *nfs4_vnodeops;
397 397
398 398 const fs_operation_def_t nfs4_vnodeops_template[] = {
399 - VOPNAME_OPEN, { .vop_open = nfs4_open },
400 - VOPNAME_CLOSE, { .vop_close = nfs4_close },
401 - VOPNAME_READ, { .vop_read = nfs4_read },
402 - VOPNAME_WRITE, { .vop_write = nfs4_write },
403 - VOPNAME_IOCTL, { .vop_ioctl = nfs4_ioctl },
404 - VOPNAME_GETATTR, { .vop_getattr = nfs4_getattr },
405 - VOPNAME_SETATTR, { .vop_setattr = nfs4_setattr },
406 - VOPNAME_ACCESS, { .vop_access = nfs4_access },
407 - VOPNAME_LOOKUP, { .vop_lookup = nfs4_lookup },
408 - VOPNAME_CREATE, { .vop_create = nfs4_create },
409 - VOPNAME_REMOVE, { .vop_remove = nfs4_remove },
410 - VOPNAME_LINK, { .vop_link = nfs4_link },
411 - VOPNAME_RENAME, { .vop_rename = nfs4_rename },
412 - VOPNAME_MKDIR, { .vop_mkdir = nfs4_mkdir },
413 - VOPNAME_RMDIR, { .vop_rmdir = nfs4_rmdir },
414 - VOPNAME_READDIR, { .vop_readdir = nfs4_readdir },
415 - VOPNAME_SYMLINK, { .vop_symlink = nfs4_symlink },
416 - VOPNAME_READLINK, { .vop_readlink = nfs4_readlink },
417 - VOPNAME_FSYNC, { .vop_fsync = nfs4_fsync },
418 - VOPNAME_INACTIVE, { .vop_inactive = nfs4_inactive },
419 - VOPNAME_FID, { .vop_fid = nfs4_fid },
420 - VOPNAME_RWLOCK, { .vop_rwlock = nfs4_rwlock },
421 - VOPNAME_RWUNLOCK, { .vop_rwunlock = nfs4_rwunlock },
422 - VOPNAME_SEEK, { .vop_seek = nfs4_seek },
423 - VOPNAME_FRLOCK, { .vop_frlock = nfs4_frlock },
424 - VOPNAME_SPACE, { .vop_space = nfs4_space },
425 - VOPNAME_REALVP, { .vop_realvp = nfs4_realvp },
426 - VOPNAME_GETPAGE, { .vop_getpage = nfs4_getpage },
427 - VOPNAME_PUTPAGE, { .vop_putpage = nfs4_putpage },
428 - VOPNAME_MAP, { .vop_map = nfs4_map },
429 - VOPNAME_ADDMAP, { .vop_addmap = nfs4_addmap },
430 - VOPNAME_DELMAP, { .vop_delmap = nfs4_delmap },
399 + { VOPNAME_OPEN, { .vop_open = nfs4_open } },
400 + { VOPNAME_CLOSE, { .vop_close = nfs4_close } },
401 + { VOPNAME_READ, { .vop_read = nfs4_read } },
402 + { VOPNAME_WRITE, { .vop_write = nfs4_write } },
403 + { VOPNAME_IOCTL, { .vop_ioctl = nfs4_ioctl } },
404 + { VOPNAME_GETATTR, { .vop_getattr = nfs4_getattr } },
405 + { VOPNAME_SETATTR, { .vop_setattr = nfs4_setattr } },
406 + { VOPNAME_ACCESS, { .vop_access = nfs4_access } },
407 + { VOPNAME_LOOKUP, { .vop_lookup = nfs4_lookup } },
408 + { VOPNAME_CREATE, { .vop_create = nfs4_create } },
409 + { VOPNAME_REMOVE, { .vop_remove = nfs4_remove } },
410 + { VOPNAME_LINK, { .vop_link = nfs4_link } },
411 + { VOPNAME_RENAME, { .vop_rename = nfs4_rename } },
412 + { VOPNAME_MKDIR, { .vop_mkdir = nfs4_mkdir } },
413 + { VOPNAME_RMDIR, { .vop_rmdir = nfs4_rmdir } },
414 + { VOPNAME_READDIR, { .vop_readdir = nfs4_readdir } },
415 + { VOPNAME_SYMLINK, { .vop_symlink = nfs4_symlink } },
416 + { VOPNAME_READLINK, { .vop_readlink = nfs4_readlink } },
417 + { VOPNAME_FSYNC, { .vop_fsync = nfs4_fsync } },
418 + { VOPNAME_INACTIVE, { .vop_inactive = nfs4_inactive } },
419 + { VOPNAME_FID, { .vop_fid = nfs4_fid } },
420 + { VOPNAME_RWLOCK, { .vop_rwlock = nfs4_rwlock } },
421 + { VOPNAME_RWUNLOCK, { .vop_rwunlock = nfs4_rwunlock } },
422 + { VOPNAME_SEEK, { .vop_seek = nfs4_seek } },
423 + { VOPNAME_FRLOCK, { .vop_frlock = nfs4_frlock } },
424 + { VOPNAME_SPACE, { .vop_space = nfs4_space } },
425 + { VOPNAME_REALVP, { .vop_realvp = nfs4_realvp } },
426 + { VOPNAME_GETPAGE, { .vop_getpage = nfs4_getpage } },
427 + { VOPNAME_PUTPAGE, { .vop_putpage = nfs4_putpage } },
428 + { VOPNAME_MAP, { .vop_map = nfs4_map } },
429 + { VOPNAME_ADDMAP, { .vop_addmap = nfs4_addmap } },
430 + { VOPNAME_DELMAP, { .vop_delmap = nfs4_delmap } },
431 431 /* no separate nfs4_dump */
432 - VOPNAME_DUMP, { .vop_dump = nfs_dump },
433 - VOPNAME_PATHCONF, { .vop_pathconf = nfs4_pathconf },
434 - VOPNAME_PAGEIO, { .vop_pageio = nfs4_pageio },
435 - VOPNAME_DISPOSE, { .vop_dispose = nfs4_dispose },
436 - VOPNAME_SETSECATTR, { .vop_setsecattr = nfs4_setsecattr },
437 - VOPNAME_GETSECATTR, { .vop_getsecattr = nfs4_getsecattr },
438 - VOPNAME_SHRLOCK, { .vop_shrlock = nfs4_shrlock },
439 - VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
440 - NULL, NULL
432 + { VOPNAME_DUMP, { .vop_dump = nfs_dump } },
433 + { VOPNAME_PATHCONF, { .vop_pathconf = nfs4_pathconf } },
434 + { VOPNAME_PAGEIO, { .vop_pageio = nfs4_pageio } },
435 + { VOPNAME_DISPOSE, { .vop_dispose = nfs4_dispose } },
436 + { VOPNAME_SETSECATTR, { .vop_setsecattr = nfs4_setsecattr } },
437 + { VOPNAME_GETSECATTR, { .vop_getsecattr = nfs4_getsecattr } },
438 + { VOPNAME_SHRLOCK, { .vop_shrlock = nfs4_shrlock } },
439 + { VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support } },
440 + { NULL, { NULL } }
441 441 };
442 442
443 443 /*
444 444 * The following are subroutines and definitions to set args or get res
445 445 * for the different nfsv4 ops
446 446 */
447 447
448 448 void
449 449 nfs4args_lookup_free(nfs_argop4 *argop, int arglen)
450 450 {
451 451 int i;
452 452
453 453 for (i = 0; i < arglen; i++) {
454 454 if (argop[i].argop == OP_LOOKUP) {
455 455 kmem_free(
456 456 argop[i].nfs_argop4_u.oplookup.
457 457 objname.utf8string_val,
458 458 argop[i].nfs_argop4_u.oplookup.
459 459 objname.utf8string_len);
460 460 }
461 461 }
462 462 }
463 463
464 464 static void
465 465 nfs4args_lock_free(nfs_argop4 *argop)
466 466 {
467 467 locker4 *locker = &argop->nfs_argop4_u.oplock.locker;
468 468
469 469 if (locker->new_lock_owner == TRUE) {
470 470 open_to_lock_owner4 *open_owner;
471 471
472 472 open_owner = &locker->locker4_u.open_owner;
473 473 if (open_owner->lock_owner.owner_val != NULL) {
474 474 kmem_free(open_owner->lock_owner.owner_val,
475 475 open_owner->lock_owner.owner_len);
476 476 }
477 477 }
478 478 }
479 479
480 480 static void
481 481 nfs4args_lockt_free(nfs_argop4 *argop)
482 482 {
483 483 lock_owner4 *lowner = &argop->nfs_argop4_u.oplockt.owner;
484 484
485 485 if (lowner->owner_val != NULL) {
486 486 kmem_free(lowner->owner_val, lowner->owner_len);
487 487 }
488 488 }
489 489
490 490 static void
491 491 nfs4args_setattr(nfs_argop4 *argop, vattr_t *vap, vsecattr_t *vsap, int flags,
492 492 rnode4_t *rp, cred_t *cr, bitmap4 supp, int *error,
493 493 nfs4_stateid_types_t *sid_types)
494 494 {
495 495 fattr4 *attr = &argop->nfs_argop4_u.opsetattr.obj_attributes;
496 496 mntinfo4_t *mi;
497 497
498 498 argop->argop = OP_SETATTR;
499 499 /*
500 500 * The stateid is set to 0 if client is not modifying the size
501 501 * and otherwise to whatever nfs4_get_stateid() returns.
502 502 *
503 503 * XXX Note: nfs4_get_stateid() returns 0 if no lockowner and/or no
504 504 * state struct could be found for the process/file pair. We may
505 505 * want to change this in the future (by OPENing the file). See
506 506 * bug # 4474852.
507 507 */
508 508 if (vap->va_mask & AT_SIZE) {
509 509
510 510 ASSERT(rp != NULL);
511 511 mi = VTOMI4(RTOV4(rp));
512 512
513 513 argop->nfs_argop4_u.opsetattr.stateid =
514 514 nfs4_get_stateid(cr, rp, curproc->p_pidp->pid_id, mi,
515 515 OP_SETATTR, sid_types, FALSE);
516 516 } else {
517 517 bzero(&argop->nfs_argop4_u.opsetattr.stateid,
518 518 sizeof (stateid4));
519 519 }
520 520
521 521 *error = vattr_to_fattr4(vap, vsap, attr, flags, OP_SETATTR, supp);
522 522 if (*error)
523 523 bzero(attr, sizeof (*attr));
524 524 }
525 525
526 526 static void
527 527 nfs4args_setattr_free(nfs_argop4 *argop)
528 528 {
529 529 nfs4_fattr4_free(&argop->nfs_argop4_u.opsetattr.obj_attributes);
530 530 }
531 531
532 532 static int
533 533 nfs4args_verify(nfs_argop4 *argop, vattr_t *vap, enum nfs_opnum4 op,
534 534 bitmap4 supp)
535 535 {
536 536 fattr4 *attr;
537 537 int error = 0;
538 538
539 539 argop->argop = op;
540 540 switch (op) {
541 541 case OP_VERIFY:
542 542 attr = &argop->nfs_argop4_u.opverify.obj_attributes;
543 543 break;
544 544 case OP_NVERIFY:
545 545 attr = &argop->nfs_argop4_u.opnverify.obj_attributes;
546 546 break;
547 547 default:
548 548 return (EINVAL);
549 549 }
550 550 if (!error)
551 551 error = vattr_to_fattr4(vap, NULL, attr, 0, op, supp);
552 552 if (error)
553 553 bzero(attr, sizeof (*attr));
554 554 return (error);
555 555 }
556 556
557 557 static void
558 558 nfs4args_verify_free(nfs_argop4 *argop)
559 559 {
560 560 switch (argop->argop) {
561 561 case OP_VERIFY:
562 562 nfs4_fattr4_free(&argop->nfs_argop4_u.opverify.obj_attributes);
563 563 break;
564 564 case OP_NVERIFY:
565 565 nfs4_fattr4_free(&argop->nfs_argop4_u.opnverify.obj_attributes);
566 566 break;
567 567 default:
568 568 break;
569 569 }
570 570 }
571 571
572 572 static void
573 573 nfs4args_write(nfs_argop4 *argop, stable_how4 stable, rnode4_t *rp, cred_t *cr,
574 574 WRITE4args **wargs_pp, nfs4_stateid_types_t *sid_tp)
575 575 {
576 576 WRITE4args *wargs = &argop->nfs_argop4_u.opwrite;
577 577 mntinfo4_t *mi = VTOMI4(RTOV4(rp));
578 578
579 579 argop->argop = OP_WRITE;
580 580 wargs->stable = stable;
581 581 wargs->stateid = nfs4_get_w_stateid(cr, rp, curproc->p_pidp->pid_id,
582 582 mi, OP_WRITE, sid_tp);
583 583 wargs->mblk = NULL;
584 584 *wargs_pp = wargs;
585 585 }
586 586
587 587 void
588 588 nfs4args_copen_free(OPEN4cargs *open_args)
589 589 {
590 590 if (open_args->owner.owner_val) {
591 591 kmem_free(open_args->owner.owner_val,
592 592 open_args->owner.owner_len);
593 593 }
594 594 if ((open_args->opentype == OPEN4_CREATE) &&
595 595 (open_args->mode != EXCLUSIVE4)) {
596 596 nfs4_fattr4_free(&open_args->createhow4_u.createattrs);
597 597 }
598 598 }
599 599
600 600 /*
601 601 * XXX: This is referenced in modstubs.s
602 602 */
603 603 struct vnodeops *
604 604 nfs4_getvnodeops(void)
605 605 {
606 606 return (nfs4_vnodeops);
607 607 }
608 608
609 609 /*
610 610 * The OPEN operation opens a regular file.
611 611 */
612 612 /*ARGSUSED3*/
613 613 static int
614 614 nfs4_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
615 615 {
616 616 vnode_t *dvp = NULL;
617 617 rnode4_t *rp, *drp;
618 618 int error;
619 619 int just_been_created;
620 620 char fn[MAXNAMELEN];
621 621
622 622 NFS4_DEBUG(nfs4_client_state_debug, (CE_NOTE, "nfs4_open: "));
623 623 if (nfs_zone() != VTOMI4(*vpp)->mi_zone)
624 624 return (EIO);
625 625 rp = VTOR4(*vpp);
626 626
627 627 /*
628 628 * Check to see if opening something besides a regular file;
629 629 * if so skip the OTW call
630 630 */
631 631 if ((*vpp)->v_type != VREG) {
632 632 error = nfs4_open_non_reg_file(vpp, flag, cr);
633 633 return (error);
634 634 }
635 635
636 636 /*
637 637 * XXX - would like a check right here to know if the file is
638 638 * executable or not, so as to skip OTW
639 639 */
640 640
641 641 if ((error = vtodv(*vpp, &dvp, cr, TRUE)) != 0)
642 642 return (error);
643 643
644 644 drp = VTOR4(dvp);
645 645 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR4(dvp)))
646 646 return (EINTR);
647 647
648 648 if ((error = vtoname(*vpp, fn, MAXNAMELEN)) != 0) {
649 649 nfs_rw_exit(&drp->r_rwlock);
650 650 return (error);
651 651 }
652 652
653 653 /*
654 654 * See if this file has just been CREATEd.
655 655 * If so, clear the flag and update the dnlc, which was previously
656 656 * skipped in nfs4_create.
657 657 * XXX need better serilization on this.
658 658 * XXX move this into the nf4open_otw call, after we have
659 659 * XXX acquired the open owner seqid sync.
660 660 */
661 661 mutex_enter(&rp->r_statev4_lock);
662 662 if (rp->created_v4) {
663 663 rp->created_v4 = 0;
664 664 mutex_exit(&rp->r_statev4_lock);
665 665
666 666 dnlc_update(dvp, fn, *vpp);
667 667 /* This is needed so we don't bump the open ref count */
668 668 just_been_created = 1;
669 669 } else {
670 670 mutex_exit(&rp->r_statev4_lock);
671 671 just_been_created = 0;
672 672 }
673 673
674 674 /*
675 675 * If caller specified O_TRUNC/FTRUNC, then be sure to set
676 676 * FWRITE (to drive successful setattr(size=0) after open)
677 677 */
678 678 if (flag & FTRUNC)
679 679 flag |= FWRITE;
680 680
681 681 error = nfs4open_otw(dvp, fn, NULL, vpp, cr, 0, flag, 0,
682 682 just_been_created);
683 683
684 684 if (!error && !((*vpp)->v_flag & VROOT))
685 685 dnlc_update(dvp, fn, *vpp);
686 686
687 687 nfs_rw_exit(&drp->r_rwlock);
688 688
689 689 /* release the hold from vtodv */
690 690 VN_RELE(dvp);
691 691
692 692 /* exchange the shadow for the master vnode, if needed */
693 693
694 694 if (error == 0 && IS_SHADOW(*vpp, rp))
695 695 sv_exchange(vpp);
696 696
697 697 return (error);
698 698 }
699 699
700 700 /*
701 701 * See if there's a "lost open" request to be saved and recovered.
702 702 */
703 703 static void
704 704 nfs4open_save_lost_rqst(int error, nfs4_lost_rqst_t *lost_rqstp,
705 705 nfs4_open_owner_t *oop, cred_t *cr, vnode_t *vp,
706 706 vnode_t *dvp, OPEN4cargs *open_args)
707 707 {
708 708 vfs_t *vfsp;
709 709 char *srccfp;
710 710
711 711 vfsp = (dvp ? dvp->v_vfsp : vp->v_vfsp);
712 712
713 713 if (error != ETIMEDOUT && error != EINTR &&
714 714 !NFS4_FRC_UNMT_ERR(error, vfsp)) {
715 715 lost_rqstp->lr_op = 0;
716 716 return;
717 717 }
718 718
719 719 NFS4_DEBUG(nfs4_lost_rqst_debug, (CE_NOTE,
720 720 "nfs4open_save_lost_rqst: error %d", error));
721 721
722 722 lost_rqstp->lr_op = OP_OPEN;
723 723
724 724 /*
725 725 * The vp (if it is not NULL) and dvp are held and rele'd via
726 726 * the recovery code. See nfs4_save_lost_rqst.
727 727 */
728 728 lost_rqstp->lr_vp = vp;
729 729 lost_rqstp->lr_dvp = dvp;
730 730 lost_rqstp->lr_oop = oop;
731 731 lost_rqstp->lr_osp = NULL;
732 732 lost_rqstp->lr_lop = NULL;
733 733 lost_rqstp->lr_cr = cr;
734 734 lost_rqstp->lr_flk = NULL;
735 735 lost_rqstp->lr_oacc = open_args->share_access;
736 736 lost_rqstp->lr_odeny = open_args->share_deny;
737 737 lost_rqstp->lr_oclaim = open_args->claim;
738 738 if (open_args->claim == CLAIM_DELEGATE_CUR) {
739 739 lost_rqstp->lr_ostateid =
740 740 open_args->open_claim4_u.delegate_cur_info.delegate_stateid;
741 741 srccfp = open_args->open_claim4_u.delegate_cur_info.cfile;
742 742 } else {
743 743 srccfp = open_args->open_claim4_u.cfile;
744 744 }
745 745 lost_rqstp->lr_ofile.utf8string_len = 0;
746 746 lost_rqstp->lr_ofile.utf8string_val = NULL;
747 747 (void) str_to_utf8(srccfp, &lost_rqstp->lr_ofile);
748 748 lost_rqstp->lr_putfirst = FALSE;
749 749 }
750 750
751 751 struct nfs4_excl_time {
752 752 uint32 seconds;
753 753 uint32 nseconds;
754 754 };
755 755
756 756 /*
757 757 * The OPEN operation creates and/or opens a regular file
758 758 *
759 759 * ARGSUSED
760 760 */
761 761 static int
762 762 nfs4open_otw(vnode_t *dvp, char *file_name, struct vattr *in_va,
763 763 vnode_t **vpp, cred_t *cr, int create_flag, int open_flag,
764 764 enum createmode4 createmode, int file_just_been_created)
765 765 {
766 766 rnode4_t *rp;
767 767 rnode4_t *drp = VTOR4(dvp);
768 768 vnode_t *vp = NULL;
769 769 vnode_t *vpi = *vpp;
770 770 bool_t needrecov = FALSE;
771 771
772 772 int doqueue = 1;
773 773
774 774 COMPOUND4args_clnt args;
775 775 COMPOUND4res_clnt res;
776 776 nfs_argop4 *argop;
777 777 nfs_resop4 *resop;
778 778 int argoplist_size;
779 779 int idx_open, idx_fattr;
780 780
781 781 GETFH4res *gf_res = NULL;
782 782 OPEN4res *op_res = NULL;
783 783 nfs4_ga_res_t *garp;
784 784 fattr4 *attr = NULL;
785 785 struct nfs4_excl_time verf;
786 786 bool_t did_excl_setup = FALSE;
787 787 int created_osp;
788 788
789 789 OPEN4cargs *open_args;
790 790 nfs4_open_owner_t *oop = NULL;
791 791 nfs4_open_stream_t *osp = NULL;
792 792 seqid4 seqid = 0;
793 793 bool_t retry_open = FALSE;
794 794 nfs4_recov_state_t recov_state;
795 795 nfs4_lost_rqst_t lost_rqst;
796 796 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
797 797 hrtime_t t;
798 798 int acc = 0;
799 799 cred_t *cred_otw = NULL; /* cred used to do the RPC call */
800 800 cred_t *ncr = NULL;
801 801
802 802 nfs4_sharedfh_t *otw_sfh;
803 803 nfs4_sharedfh_t *orig_sfh;
804 804 int fh_differs = 0;
805 805 int numops, setgid_flag;
806 806 int num_bseqid_retry = NFS4_NUM_RETRY_BAD_SEQID + 1;
807 807
808 808 /*
809 809 * Make sure we properly deal with setting the right gid on
810 810 * a newly created file to reflect the parent's setgid bit
811 811 */
812 812 setgid_flag = 0;
813 813 if (create_flag && in_va) {
814 814
815 815 /*
816 816 * If there is grpid mount flag used or
817 817 * the parent's directory has the setgid bit set
818 818 * _and_ the client was able to get a valid mapping
819 819 * for the parent dir's owner_group, we want to
820 820 * append NVERIFY(owner_group == dva.va_gid) and
821 821 * SETATTR to the CREATE compound.
822 822 */
823 823 mutex_enter(&drp->r_statelock);
824 824 if ((VTOMI4(dvp)->mi_flags & MI4_GRPID ||
825 825 drp->r_attr.va_mode & VSGID) &&
826 826 drp->r_attr.va_gid != GID_NOBODY) {
827 827 in_va->va_mask |= AT_GID;
828 828 in_va->va_gid = drp->r_attr.va_gid;
829 829 setgid_flag = 1;
830 830 }
831 831 mutex_exit(&drp->r_statelock);
832 832 }
833 833
834 834 /*
835 835 * Normal/non-create compound:
836 836 * PUTFH(dfh) + OPEN(create) + GETFH + GETATTR(new)
837 837 *
838 838 * Open(create) compound no setgid:
839 839 * PUTFH(dfh) + SAVEFH + OPEN(create) + GETFH + GETATTR(new) +
840 840 * RESTOREFH + GETATTR
841 841 *
842 842 * Open(create) setgid:
843 843 * PUTFH(dfh) + OPEN(create) + GETFH + GETATTR(new) +
844 844 * SAVEFH + PUTFH(dfh) + GETATTR(dvp) + RESTOREFH +
845 845 * NVERIFY(grp) + SETATTR
846 846 */
847 847 if (setgid_flag) {
848 848 numops = 10;
849 849 idx_open = 1;
850 850 idx_fattr = 3;
851 851 } else if (create_flag) {
852 852 numops = 7;
853 853 idx_open = 2;
854 854 idx_fattr = 4;
855 855 } else {
856 856 numops = 4;
857 857 idx_open = 1;
858 858 idx_fattr = 3;
859 859 }
860 860
861 861 args.array_len = numops;
862 862 argoplist_size = numops * sizeof (nfs_argop4);
863 863 argop = kmem_alloc(argoplist_size, KM_SLEEP);
864 864
865 865 NFS4_DEBUG(nfs4_client_state_debug, (CE_NOTE, "nfs4open_otw: "
866 866 "open %s open flag 0x%x cred %p", file_name, open_flag,
867 867 (void *)cr));
868 868
869 869 ASSERT(nfs_zone() == VTOMI4(dvp)->mi_zone);
870 870 if (create_flag) {
871 871 /*
872 872 * We are to create a file. Initialize the passed in vnode
873 873 * pointer.
874 874 */
875 875 vpi = NULL;
876 876 } else {
877 877 /*
878 878 * Check to see if the client owns a read delegation and is
879 879 * trying to open for write. If so, then return the delegation
880 880 * to avoid the server doing a cb_recall and returning DELAY.
881 881 * NB - we don't use the statev4_lock here because we'd have
882 882 * to drop the lock anyway and the result would be stale.
883 883 */
884 884 if ((open_flag & FWRITE) &&
885 885 VTOR4(vpi)->r_deleg_type == OPEN_DELEGATE_READ)
886 886 (void) nfs4delegreturn(VTOR4(vpi), NFS4_DR_REOPEN);
887 887
888 888 /*
889 889 * If the file has a delegation, then do an access check up
890 890 * front. This avoids having to an access check later after
891 891 * we've already done start_op, which could deadlock.
892 892 */
893 893 if (VTOR4(vpi)->r_deleg_type != OPEN_DELEGATE_NONE) {
894 894 if (open_flag & FREAD &&
895 895 nfs4_access(vpi, VREAD, 0, cr, NULL) == 0)
896 896 acc |= VREAD;
897 897 if (open_flag & FWRITE &&
898 898 nfs4_access(vpi, VWRITE, 0, cr, NULL) == 0)
899 899 acc |= VWRITE;
900 900 }
901 901 }
902 902
903 903 drp = VTOR4(dvp);
904 904
905 905 recov_state.rs_flags = 0;
906 906 recov_state.rs_num_retry_despite_err = 0;
907 907 cred_otw = cr;
908 908
909 909 recov_retry:
910 910 fh_differs = 0;
911 911 nfs4_error_zinit(&e);
912 912
913 913 e.error = nfs4_start_op(VTOMI4(dvp), dvp, vpi, &recov_state);
914 914 if (e.error) {
915 915 if (ncr != NULL)
916 916 crfree(ncr);
917 917 kmem_free(argop, argoplist_size);
918 918 return (e.error);
919 919 }
920 920
921 921 args.ctag = TAG_OPEN;
922 922 args.array_len = numops;
923 923 args.array = argop;
924 924
925 925 /* putfh directory fh */
926 926 argop[0].argop = OP_CPUTFH;
927 927 argop[0].nfs_argop4_u.opcputfh.sfh = drp->r_fh;
928 928
929 929 /* OPEN: either op 1 or op 2 depending upon create/setgid flags */
930 930 argop[idx_open].argop = OP_COPEN;
931 931 open_args = &argop[idx_open].nfs_argop4_u.opcopen;
932 932 open_args->claim = CLAIM_NULL;
933 933
934 934 /* name of file */
935 935 open_args->open_claim4_u.cfile = file_name;
936 936 open_args->owner.owner_len = 0;
937 937 open_args->owner.owner_val = NULL;
938 938
939 939 if (create_flag) {
940 940 /* CREATE a file */
941 941 open_args->opentype = OPEN4_CREATE;
942 942 open_args->mode = createmode;
943 943 if (createmode == EXCLUSIVE4) {
944 944 if (did_excl_setup == FALSE) {
945 945 verf.seconds = zone_get_hostid(NULL);
946 946 if (verf.seconds != 0)
947 947 verf.nseconds = newnum();
948 948 else {
949 949 timestruc_t now;
950 950
951 951 gethrestime(&now);
952 952 verf.seconds = now.tv_sec;
953 953 verf.nseconds = now.tv_nsec;
954 954 }
955 955 /*
956 956 * Since the server will use this value for the
957 957 * mtime, make sure that it can't overflow. Zero
958 958 * out the MSB. The actual value does not matter
959 959 * here, only its uniqeness.
960 960 */
961 961 verf.seconds &= INT32_MAX;
962 962 did_excl_setup = TRUE;
963 963 }
964 964
965 965 /* Now copy over verifier to OPEN4args. */
966 966 open_args->createhow4_u.createverf = *(uint64_t *)&verf;
967 967 } else {
968 968 int v_error;
969 969 bitmap4 supp_attrs;
970 970 servinfo4_t *svp;
971 971
972 972 attr = &open_args->createhow4_u.createattrs;
973 973
974 974 svp = drp->r_server;
975 975 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
976 976 supp_attrs = svp->sv_supp_attrs;
977 977 nfs_rw_exit(&svp->sv_lock);
978 978
979 979 /* GUARDED4 or UNCHECKED4 */
980 980 v_error = vattr_to_fattr4(in_va, NULL, attr, 0, OP_OPEN,
981 981 supp_attrs);
982 982 if (v_error) {
983 983 bzero(attr, sizeof (*attr));
984 984 nfs4args_copen_free(open_args);
985 985 nfs4_end_op(VTOMI4(dvp), dvp, vpi,
986 986 &recov_state, FALSE);
987 987 if (ncr != NULL)
988 988 crfree(ncr);
989 989 kmem_free(argop, argoplist_size);
990 990 return (v_error);
991 991 }
992 992 }
993 993 } else {
994 994 /* NO CREATE */
995 995 open_args->opentype = OPEN4_NOCREATE;
996 996 }
997 997
998 998 if (recov_state.rs_sp != NULL) {
999 999 mutex_enter(&recov_state.rs_sp->s_lock);
1000 1000 open_args->owner.clientid = recov_state.rs_sp->clientid;
1001 1001 mutex_exit(&recov_state.rs_sp->s_lock);
1002 1002 } else {
1003 1003 /* XXX should we just fail here? */
1004 1004 open_args->owner.clientid = 0;
1005 1005 }
1006 1006
1007 1007 /*
1008 1008 * This increments oop's ref count or creates a temporary 'just_created'
1009 1009 * open owner that will become valid when this OPEN/OPEN_CONFIRM call
1010 1010 * completes.
1011 1011 */
1012 1012 mutex_enter(&VTOMI4(dvp)->mi_lock);
1013 1013
1014 1014 /* See if a permanent or just created open owner exists */
1015 1015 oop = find_open_owner_nolock(cr, NFS4_JUST_CREATED, VTOMI4(dvp));
1016 1016 if (!oop) {
1017 1017 /*
1018 1018 * This open owner does not exist so create a temporary
1019 1019 * just created one.
1020 1020 */
1021 1021 oop = create_open_owner(cr, VTOMI4(dvp));
1022 1022 ASSERT(oop != NULL);
1023 1023 }
1024 1024 mutex_exit(&VTOMI4(dvp)->mi_lock);
1025 1025
1026 1026 /* this length never changes, do alloc before seqid sync */
1027 1027 open_args->owner.owner_len = sizeof (oop->oo_name);
1028 1028 open_args->owner.owner_val =
1029 1029 kmem_alloc(open_args->owner.owner_len, KM_SLEEP);
1030 1030
1031 1031 e.error = nfs4_start_open_seqid_sync(oop, VTOMI4(dvp));
1032 1032 if (e.error == EAGAIN) {
1033 1033 open_owner_rele(oop);
1034 1034 nfs4args_copen_free(open_args);
1035 1035 nfs4_end_op(VTOMI4(dvp), dvp, vpi, &recov_state, TRUE);
1036 1036 if (ncr != NULL) {
1037 1037 crfree(ncr);
1038 1038 ncr = NULL;
1039 1039 }
1040 1040 goto recov_retry;
1041 1041 }
1042 1042
1043 1043 /* Check to see if we need to do the OTW call */
1044 1044 if (!create_flag) {
1045 1045 if (!nfs4_is_otw_open_necessary(oop, open_flag, vpi,
1046 1046 file_just_been_created, &e.error, acc, &recov_state)) {
1047 1047
1048 1048 /*
1049 1049 * The OTW open is not necessary. Either
1050 1050 * the open can succeed without it (eg.
1051 1051 * delegation, error == 0) or the open
1052 1052 * must fail due to an access failure
1053 1053 * (error != 0). In either case, tidy
1054 1054 * up and return.
1055 1055 */
1056 1056
1057 1057 nfs4_end_open_seqid_sync(oop);
1058 1058 open_owner_rele(oop);
1059 1059 nfs4args_copen_free(open_args);
1060 1060 nfs4_end_op(VTOMI4(dvp), dvp, vpi, &recov_state, FALSE);
1061 1061 if (ncr != NULL)
1062 1062 crfree(ncr);
1063 1063 kmem_free(argop, argoplist_size);
1064 1064 return (e.error);
1065 1065 }
1066 1066 }
1067 1067
1068 1068 bcopy(&oop->oo_name, open_args->owner.owner_val,
1069 1069 open_args->owner.owner_len);
1070 1070
1071 1071 seqid = nfs4_get_open_seqid(oop) + 1;
1072 1072 open_args->seqid = seqid;
1073 1073 open_args->share_access = 0;
1074 1074 if (open_flag & FREAD)
1075 1075 open_args->share_access |= OPEN4_SHARE_ACCESS_READ;
1076 1076 if (open_flag & FWRITE)
1077 1077 open_args->share_access |= OPEN4_SHARE_ACCESS_WRITE;
1078 1078 open_args->share_deny = OPEN4_SHARE_DENY_NONE;
1079 1079
1080 1080
1081 1081
1082 1082 /*
1083 1083 * getfh w/sanity check for idx_open/idx_fattr
1084 1084 */
1085 1085 ASSERT((idx_open + 1) == (idx_fattr - 1));
1086 1086 argop[idx_open + 1].argop = OP_GETFH;
1087 1087
1088 1088 /* getattr */
1089 1089 argop[idx_fattr].argop = OP_GETATTR;
1090 1090 argop[idx_fattr].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
1091 1091 argop[idx_fattr].nfs_argop4_u.opgetattr.mi = VTOMI4(dvp);
1092 1092
1093 1093 if (setgid_flag) {
1094 1094 vattr_t _v;
1095 1095 servinfo4_t *svp;
1096 1096 bitmap4 supp_attrs;
1097 1097
1098 1098 svp = drp->r_server;
1099 1099 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
1100 1100 supp_attrs = svp->sv_supp_attrs;
1101 1101 nfs_rw_exit(&svp->sv_lock);
1102 1102
1103 1103 /*
1104 1104 * For setgid case, we need to:
1105 1105 * 4:savefh(new) 5:putfh(dir) 6:getattr(dir) 7:restorefh(new)
1106 1106 */
1107 1107 argop[4].argop = OP_SAVEFH;
1108 1108
1109 1109 argop[5].argop = OP_CPUTFH;
1110 1110 argop[5].nfs_argop4_u.opcputfh.sfh = drp->r_fh;
1111 1111
1112 1112 argop[6].argop = OP_GETATTR;
1113 1113 argop[6].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
1114 1114 argop[6].nfs_argop4_u.opgetattr.mi = VTOMI4(dvp);
1115 1115
1116 1116 argop[7].argop = OP_RESTOREFH;
1117 1117
1118 1118 /*
1119 1119 * nverify
1120 1120 */
1121 1121 _v.va_mask = AT_GID;
1122 1122 _v.va_gid = in_va->va_gid;
1123 1123 if (!(e.error = nfs4args_verify(&argop[8], &_v, OP_NVERIFY,
1124 1124 supp_attrs))) {
1125 1125
1126 1126 /*
1127 1127 * setattr
1128 1128 *
1129 1129 * We _know_ we're not messing with AT_SIZE or
1130 1130 * AT_XTIME, so no need for stateid or flags.
1131 1131 * Also we specify NULL rp since we're only
1132 1132 * interested in setting owner_group attributes.
1133 1133 */
1134 1134 nfs4args_setattr(&argop[9], &_v, NULL, 0, NULL, cr,
1135 1135 supp_attrs, &e.error, 0);
1136 1136 if (e.error)
1137 1137 nfs4args_verify_free(&argop[8]);
1138 1138 }
1139 1139
1140 1140 if (e.error) {
1141 1141 /*
1142 1142 * XXX - Revisit the last argument to nfs4_end_op()
1143 1143 * once 5020486 is fixed.
1144 1144 */
1145 1145 nfs4_end_open_seqid_sync(oop);
1146 1146 open_owner_rele(oop);
1147 1147 nfs4args_copen_free(open_args);
1148 1148 nfs4_end_op(VTOMI4(dvp), dvp, vpi, &recov_state, TRUE);
1149 1149 if (ncr != NULL)
1150 1150 crfree(ncr);
1151 1151 kmem_free(argop, argoplist_size);
1152 1152 return (e.error);
1153 1153 }
1154 1154 } else if (create_flag) {
1155 1155 argop[1].argop = OP_SAVEFH;
1156 1156
1157 1157 argop[5].argop = OP_RESTOREFH;
1158 1158
1159 1159 argop[6].argop = OP_GETATTR;
1160 1160 argop[6].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
1161 1161 argop[6].nfs_argop4_u.opgetattr.mi = VTOMI4(dvp);
1162 1162 }
1163 1163
1164 1164 NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE,
1165 1165 "nfs4open_otw: %s call, nm %s, rp %s",
1166 1166 needrecov ? "recov" : "first", file_name,
1167 1167 rnode4info(VTOR4(dvp))));
1168 1168
1169 1169 t = gethrtime();
1170 1170
1171 1171 rfs4call(VTOMI4(dvp), &args, &res, cred_otw, &doqueue, 0, &e);
1172 1172
1173 1173 if (!e.error && nfs4_need_to_bump_seqid(&res))
1174 1174 nfs4_set_open_seqid(seqid, oop, args.ctag);
1175 1175
1176 1176 needrecov = nfs4_needs_recovery(&e, TRUE, dvp->v_vfsp);
1177 1177
1178 1178 if (e.error || needrecov) {
1179 1179 bool_t abort = FALSE;
1180 1180
1181 1181 if (needrecov) {
1182 1182 nfs4_bseqid_entry_t *bsep = NULL;
1183 1183
1184 1184 nfs4open_save_lost_rqst(e.error, &lost_rqst, oop,
1185 1185 cred_otw, vpi, dvp, open_args);
1186 1186
1187 1187 if (!e.error && res.status == NFS4ERR_BAD_SEQID) {
1188 1188 bsep = nfs4_create_bseqid_entry(oop, NULL,
1189 1189 vpi, 0, args.ctag, open_args->seqid);
1190 1190 num_bseqid_retry--;
1191 1191 }
1192 1192
1193 1193 abort = nfs4_start_recovery(&e, VTOMI4(dvp), dvp, vpi,
1194 1194 NULL, lost_rqst.lr_op == OP_OPEN ?
1195 1195 &lost_rqst : NULL, OP_OPEN, bsep, NULL, NULL);
1196 1196
1197 1197 if (bsep)
1198 1198 kmem_free(bsep, sizeof (*bsep));
1199 1199 /* give up if we keep getting BAD_SEQID */
1200 1200 if (num_bseqid_retry == 0)
1201 1201 abort = TRUE;
1202 1202 if (abort == TRUE && e.error == 0)
1203 1203 e.error = geterrno4(res.status);
1204 1204 }
1205 1205 nfs4_end_open_seqid_sync(oop);
1206 1206 open_owner_rele(oop);
1207 1207 nfs4_end_op(VTOMI4(dvp), dvp, vpi, &recov_state, needrecov);
1208 1208 nfs4args_copen_free(open_args);
1209 1209 if (setgid_flag) {
1210 1210 nfs4args_verify_free(&argop[8]);
1211 1211 nfs4args_setattr_free(&argop[9]);
1212 1212 }
1213 1213 if (!e.error)
1214 1214 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1215 1215 if (ncr != NULL) {
1216 1216 crfree(ncr);
1217 1217 ncr = NULL;
1218 1218 }
1219 1219 if (!needrecov || abort == TRUE || e.error == EINTR ||
1220 1220 NFS4_FRC_UNMT_ERR(e.error, dvp->v_vfsp)) {
1221 1221 kmem_free(argop, argoplist_size);
1222 1222 return (e.error);
1223 1223 }
1224 1224 goto recov_retry;
1225 1225 }
1226 1226
1227 1227 /*
1228 1228 * Will check and update lease after checking the rflag for
1229 1229 * OPEN_CONFIRM in the successful OPEN call.
1230 1230 */
1231 1231 if (res.status != NFS4_OK && res.array_len <= idx_fattr + 1) {
1232 1232
1233 1233 /*
1234 1234 * XXX what if we're crossing mount points from server1:/drp
1235 1235 * to server2:/drp/rp.
1236 1236 */
1237 1237
1238 1238 /* Signal our end of use of the open seqid */
1239 1239 nfs4_end_open_seqid_sync(oop);
1240 1240
1241 1241 /*
1242 1242 * This will destroy the open owner if it was just created,
1243 1243 * and no one else has put a reference on it.
1244 1244 */
1245 1245 open_owner_rele(oop);
1246 1246 if (create_flag && (createmode != EXCLUSIVE4) &&
1247 1247 res.status == NFS4ERR_BADOWNER)
1248 1248 nfs4_log_badowner(VTOMI4(dvp), OP_OPEN);
1249 1249
1250 1250 e.error = geterrno4(res.status);
1251 1251 nfs4args_copen_free(open_args);
1252 1252 if (setgid_flag) {
1253 1253 nfs4args_verify_free(&argop[8]);
1254 1254 nfs4args_setattr_free(&argop[9]);
1255 1255 }
1256 1256 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1257 1257 nfs4_end_op(VTOMI4(dvp), dvp, vpi, &recov_state, needrecov);
1258 1258 /*
1259 1259 * If the reply is NFS4ERR_ACCESS, it may be because
1260 1260 * we are root (no root net access). If the real uid
1261 1261 * is not root, then retry with the real uid instead.
1262 1262 */
1263 1263 if (ncr != NULL) {
1264 1264 crfree(ncr);
1265 1265 ncr = NULL;
1266 1266 }
1267 1267 if (res.status == NFS4ERR_ACCESS &&
1268 1268 (ncr = crnetadjust(cred_otw)) != NULL) {
1269 1269 cred_otw = ncr;
1270 1270 goto recov_retry;
1271 1271 }
1272 1272 kmem_free(argop, argoplist_size);
1273 1273 return (e.error);
1274 1274 }
1275 1275
1276 1276 resop = &res.array[idx_open]; /* open res */
1277 1277 op_res = &resop->nfs_resop4_u.opopen;
1278 1278
1279 1279 #ifdef DEBUG
1280 1280 /*
1281 1281 * verify attrset bitmap
1282 1282 */
1283 1283 if (create_flag &&
1284 1284 (createmode == UNCHECKED4 || createmode == GUARDED4)) {
1285 1285 /* make sure attrset returned is what we asked for */
1286 1286 /* XXX Ignore this 'error' for now */
1287 1287 if (attr->attrmask != op_res->attrset)
1288 1288 /* EMPTY */;
1289 1289 }
1290 1290 #endif
1291 1291
1292 1292 if (op_res->rflags & OPEN4_RESULT_LOCKTYPE_POSIX) {
1293 1293 mutex_enter(&VTOMI4(dvp)->mi_lock);
1294 1294 VTOMI4(dvp)->mi_flags |= MI4_POSIX_LOCK;
1295 1295 mutex_exit(&VTOMI4(dvp)->mi_lock);
1296 1296 }
1297 1297
1298 1298 resop = &res.array[idx_open + 1]; /* getfh res */
1299 1299 gf_res = &resop->nfs_resop4_u.opgetfh;
1300 1300
1301 1301 otw_sfh = sfh4_get(&gf_res->object, VTOMI4(dvp));
1302 1302
1303 1303 /*
1304 1304 * The open stateid has been updated on the server but not
1305 1305 * on the client yet. There is a path: makenfs4node->nfs4_attr_cache->
1306 1306 * flush_pages->VOP_PUTPAGE->...->nfs4write where we will issue an OTW
1307 1307 * WRITE call. That, however, will use the old stateid, so go ahead
1308 1308 * and upate the open stateid now, before any call to makenfs4node.
1309 1309 */
1310 1310 if (vpi) {
1311 1311 nfs4_open_stream_t *tmp_osp;
1312 1312 rnode4_t *tmp_rp = VTOR4(vpi);
1313 1313
1314 1314 tmp_osp = find_open_stream(oop, tmp_rp);
1315 1315 if (tmp_osp) {
1316 1316 tmp_osp->open_stateid = op_res->stateid;
1317 1317 mutex_exit(&tmp_osp->os_sync_lock);
1318 1318 open_stream_rele(tmp_osp, tmp_rp);
1319 1319 }
1320 1320
1321 1321 /*
1322 1322 * We must determine if the file handle given by the otw open
1323 1323 * is the same as the file handle which was passed in with
1324 1324 * *vpp. This case can be reached if the file we are trying
1325 1325 * to open has been removed and another file has been created
1326 1326 * having the same file name. The passed in vnode is released
1327 1327 * later.
1328 1328 */
1329 1329 orig_sfh = VTOR4(vpi)->r_fh;
1330 1330 fh_differs = nfs4cmpfh(&orig_sfh->sfh_fh, &otw_sfh->sfh_fh);
1331 1331 }
1332 1332
1333 1333 garp = &res.array[idx_fattr].nfs_resop4_u.opgetattr.ga_res;
1334 1334
1335 1335 if (create_flag || fh_differs) {
1336 1336 int rnode_err = 0;
1337 1337
1338 1338 vp = makenfs4node(otw_sfh, garp, dvp->v_vfsp, t, cr,
1339 1339 dvp, fn_get(VTOSV(dvp)->sv_name, file_name, otw_sfh));
1340 1340
1341 1341 if (e.error)
1342 1342 PURGE_ATTRCACHE4(vp);
1343 1343 /*
1344 1344 * For the newly created vp case, make sure the rnode
1345 1345 * isn't bad before using it.
1346 1346 */
1347 1347 mutex_enter(&(VTOR4(vp))->r_statelock);
1348 1348 if (VTOR4(vp)->r_flags & R4RECOVERR)
1349 1349 rnode_err = EIO;
1350 1350 mutex_exit(&(VTOR4(vp))->r_statelock);
1351 1351
1352 1352 if (rnode_err) {
1353 1353 nfs4_end_open_seqid_sync(oop);
1354 1354 nfs4args_copen_free(open_args);
1355 1355 if (setgid_flag) {
1356 1356 nfs4args_verify_free(&argop[8]);
1357 1357 nfs4args_setattr_free(&argop[9]);
1358 1358 }
1359 1359 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1360 1360 nfs4_end_op(VTOMI4(dvp), dvp, vpi, &recov_state,
1361 1361 needrecov);
1362 1362 open_owner_rele(oop);
1363 1363 VN_RELE(vp);
1364 1364 if (ncr != NULL)
1365 1365 crfree(ncr);
1366 1366 sfh4_rele(&otw_sfh);
1367 1367 kmem_free(argop, argoplist_size);
1368 1368 return (EIO);
1369 1369 }
1370 1370 } else {
1371 1371 vp = vpi;
1372 1372 }
1373 1373 sfh4_rele(&otw_sfh);
1374 1374
1375 1375 /*
1376 1376 * It seems odd to get a full set of attrs and then not update
1377 1377 * the object's attrcache in the non-create case. Create case uses
1378 1378 * the attrs since makenfs4node checks to see if the attrs need to
1379 1379 * be updated (and then updates them). The non-create case should
1380 1380 * update attrs also.
1381 1381 */
1382 1382 if (! create_flag && ! fh_differs && !e.error) {
1383 1383 nfs4_attr_cache(vp, garp, t, cr, TRUE, NULL);
1384 1384 }
1385 1385
1386 1386 nfs4_error_zinit(&e);
1387 1387 if (op_res->rflags & OPEN4_RESULT_CONFIRM) {
1388 1388 /* This does not do recovery for vp explicitly. */
1389 1389 nfs4open_confirm(vp, &seqid, &op_res->stateid, cred_otw, FALSE,
1390 1390 &retry_open, oop, FALSE, &e, &num_bseqid_retry);
1391 1391
1392 1392 if (e.error || e.stat) {
1393 1393 nfs4_end_open_seqid_sync(oop);
1394 1394 nfs4args_copen_free(open_args);
1395 1395 if (setgid_flag) {
1396 1396 nfs4args_verify_free(&argop[8]);
1397 1397 nfs4args_setattr_free(&argop[9]);
1398 1398 }
1399 1399 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1400 1400 nfs4_end_op(VTOMI4(dvp), dvp, vpi, &recov_state,
1401 1401 needrecov);
1402 1402 open_owner_rele(oop);
1403 1403 if (create_flag || fh_differs) {
1404 1404 /* rele the makenfs4node */
1405 1405 VN_RELE(vp);
1406 1406 }
1407 1407 if (ncr != NULL) {
1408 1408 crfree(ncr);
1409 1409 ncr = NULL;
1410 1410 }
1411 1411 if (retry_open == TRUE) {
1412 1412 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
1413 1413 "nfs4open_otw: retry the open since OPEN "
1414 1414 "CONFIRM failed with error %d stat %d",
1415 1415 e.error, e.stat));
1416 1416 if (create_flag && createmode == GUARDED4) {
1417 1417 NFS4_DEBUG(nfs4_client_recov_debug,
1418 1418 (CE_NOTE, "nfs4open_otw: switch "
1419 1419 "createmode from GUARDED4 to "
1420 1420 "UNCHECKED4"));
1421 1421 createmode = UNCHECKED4;
1422 1422 }
1423 1423 goto recov_retry;
1424 1424 }
1425 1425 if (!e.error) {
1426 1426 if (create_flag && (createmode != EXCLUSIVE4) &&
1427 1427 e.stat == NFS4ERR_BADOWNER)
1428 1428 nfs4_log_badowner(VTOMI4(dvp), OP_OPEN);
1429 1429
1430 1430 e.error = geterrno4(e.stat);
1431 1431 }
1432 1432 kmem_free(argop, argoplist_size);
1433 1433 return (e.error);
1434 1434 }
1435 1435 }
1436 1436
1437 1437 rp = VTOR4(vp);
1438 1438
1439 1439 mutex_enter(&rp->r_statev4_lock);
1440 1440 if (create_flag)
1441 1441 rp->created_v4 = 1;
1442 1442 mutex_exit(&rp->r_statev4_lock);
1443 1443
1444 1444 mutex_enter(&oop->oo_lock);
1445 1445 /* Doesn't matter if 'oo_just_created' already was set as this */
1446 1446 oop->oo_just_created = NFS4_PERM_CREATED;
1447 1447 if (oop->oo_cred_otw)
1448 1448 crfree(oop->oo_cred_otw);
1449 1449 oop->oo_cred_otw = cred_otw;
1450 1450 crhold(oop->oo_cred_otw);
1451 1451 mutex_exit(&oop->oo_lock);
1452 1452
1453 1453 /* returns with 'os_sync_lock' held */
1454 1454 osp = find_or_create_open_stream(oop, rp, &created_osp);
1455 1455 if (!osp) {
1456 1456 NFS4_DEBUG(nfs4_client_state_debug, (CE_NOTE,
1457 1457 "nfs4open_otw: failed to create an open stream"));
1458 1458 NFS4_DEBUG(nfs4_seqid_sync, (CE_NOTE, "nfs4open_otw: "
1459 1459 "signal our end of use of the open seqid"));
1460 1460
1461 1461 nfs4_end_open_seqid_sync(oop);
1462 1462 open_owner_rele(oop);
1463 1463 nfs4args_copen_free(open_args);
1464 1464 if (setgid_flag) {
1465 1465 nfs4args_verify_free(&argop[8]);
1466 1466 nfs4args_setattr_free(&argop[9]);
1467 1467 }
1468 1468 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1469 1469 nfs4_end_op(VTOMI4(dvp), dvp, vpi, &recov_state, needrecov);
1470 1470 if (create_flag || fh_differs)
1471 1471 VN_RELE(vp);
1472 1472 if (ncr != NULL)
1473 1473 crfree(ncr);
1474 1474
1475 1475 kmem_free(argop, argoplist_size);
1476 1476 return (EINVAL);
1477 1477
1478 1478 }
1479 1479
1480 1480 osp->open_stateid = op_res->stateid;
1481 1481
1482 1482 if (open_flag & FREAD)
1483 1483 osp->os_share_acc_read++;
1484 1484 if (open_flag & FWRITE)
1485 1485 osp->os_share_acc_write++;
1486 1486 osp->os_share_deny_none++;
1487 1487
1488 1488 /*
1489 1489 * Need to reset this bitfield for the possible case where we were
1490 1490 * going to OTW CLOSE the file, got a non-recoverable error, and before
1491 1491 * we could retry the CLOSE, OPENed the file again.
1492 1492 */
1493 1493 ASSERT(osp->os_open_owner->oo_seqid_inuse);
1494 1494 osp->os_final_close = 0;
1495 1495 osp->os_force_close = 0;
1496 1496 #ifdef DEBUG
1497 1497 if (osp->os_failed_reopen)
1498 1498 NFS4_DEBUG(nfs4_open_stream_debug, (CE_NOTE, "nfs4open_otw:"
1499 1499 " clearing os_failed_reopen for osp %p, cr %p, rp %s",
1500 1500 (void *)osp, (void *)cr, rnode4info(rp)));
1501 1501 #endif
1502 1502 osp->os_failed_reopen = 0;
1503 1503
1504 1504 mutex_exit(&osp->os_sync_lock);
1505 1505
1506 1506 nfs4_end_open_seqid_sync(oop);
1507 1507
1508 1508 if (created_osp && recov_state.rs_sp != NULL) {
1509 1509 mutex_enter(&recov_state.rs_sp->s_lock);
1510 1510 nfs4_inc_state_ref_count_nolock(recov_state.rs_sp, VTOMI4(dvp));
1511 1511 mutex_exit(&recov_state.rs_sp->s_lock);
1512 1512 }
1513 1513
1514 1514 /* get rid of our reference to find oop */
1515 1515 open_owner_rele(oop);
1516 1516
1517 1517 open_stream_rele(osp, rp);
1518 1518
1519 1519 /* accept delegation, if any */
1520 1520 nfs4_delegation_accept(rp, CLAIM_NULL, op_res, garp, cred_otw);
1521 1521
1522 1522 nfs4_end_op(VTOMI4(dvp), dvp, vpi, &recov_state, needrecov);
1523 1523
1524 1524 if (createmode == EXCLUSIVE4 &&
1525 1525 (in_va->va_mask & ~(AT_GID | AT_SIZE))) {
1526 1526 NFS4_DEBUG(nfs4_client_state_debug, (CE_NOTE, "nfs4open_otw:"
1527 1527 " EXCLUSIVE4: sending a SETATTR"));
1528 1528 /*
1529 1529 * If doing an exclusive create, then generate
1530 1530 * a SETATTR to set the initial attributes.
1531 1531 * Try to set the mtime and the atime to the
1532 1532 * server's current time. It is somewhat
1533 1533 * expected that these fields will be used to
1534 1534 * store the exclusive create cookie. If not,
1535 1535 * server implementors will need to know that
1536 1536 * a SETATTR will follow an exclusive create
1537 1537 * and the cookie should be destroyed if
1538 1538 * appropriate.
1539 1539 *
1540 1540 * The AT_GID and AT_SIZE bits are turned off
1541 1541 * so that the SETATTR request will not attempt
1542 1542 * to process these. The gid will be set
1543 1543 * separately if appropriate. The size is turned
1544 1544 * off because it is assumed that a new file will
1545 1545 * be created empty and if the file wasn't empty,
1546 1546 * then the exclusive create will have failed
1547 1547 * because the file must have existed already.
1548 1548 * Therefore, no truncate operation is needed.
1549 1549 */
1550 1550 in_va->va_mask &= ~(AT_GID | AT_SIZE);
1551 1551 in_va->va_mask |= (AT_MTIME | AT_ATIME);
1552 1552
1553 1553 e.error = nfs4setattr(vp, in_va, 0, cr, NULL);
1554 1554 if (e.error) {
1555 1555 /*
1556 1556 * Couldn't correct the attributes of
1557 1557 * the newly created file and the
1558 1558 * attributes are wrong. Remove the
1559 1559 * file and return an error to the
1560 1560 * application.
1561 1561 */
1562 1562 /* XXX will this take care of client state ? */
1563 1563 NFS4_DEBUG(nfs4_client_state_debug, (CE_NOTE,
1564 1564 "nfs4open_otw: EXCLUSIVE4: error %d on SETATTR:"
1565 1565 " remove file", e.error));
1566 1566 VN_RELE(vp);
1567 1567 (void) nfs4_remove(dvp, file_name, cr, NULL, 0);
1568 1568 /*
1569 1569 * Since we've reled the vnode and removed
1570 1570 * the file we now need to return the error.
1571 1571 * At this point we don't want to update the
1572 1572 * dircaches, call nfs4_waitfor_purge_complete
1573 1573 * or set vpp to vp so we need to skip these
1574 1574 * as well.
1575 1575 */
1576 1576 goto skip_update_dircaches;
1577 1577 }
1578 1578 }
1579 1579
1580 1580 /*
1581 1581 * If we created or found the correct vnode, due to create_flag or
1582 1582 * fh_differs being set, then update directory cache attribute, readdir
1583 1583 * and dnlc caches.
1584 1584 */
1585 1585 if (create_flag || fh_differs) {
1586 1586 dirattr_info_t dinfo, *dinfop;
1587 1587
1588 1588 /*
1589 1589 * Make sure getattr succeeded before using results.
1590 1590 * note: op 7 is getattr(dir) for both flavors of
1591 1591 * open(create).
1592 1592 */
1593 1593 if (create_flag && res.status == NFS4_OK) {
1594 1594 dinfo.di_time_call = t;
1595 1595 dinfo.di_cred = cr;
1596 1596 dinfo.di_garp =
1597 1597 &res.array[6].nfs_resop4_u.opgetattr.ga_res;
1598 1598 dinfop = &dinfo;
1599 1599 } else {
1600 1600 dinfop = NULL;
1601 1601 }
1602 1602
1603 1603 nfs4_update_dircaches(&op_res->cinfo, dvp, vp, file_name,
1604 1604 dinfop);
1605 1605 }
1606 1606
1607 1607 /*
1608 1608 * If the page cache for this file was flushed from actions
1609 1609 * above, it was done asynchronously and if that is true,
1610 1610 * there is a need to wait here for it to complete. This must
1611 1611 * be done outside of start_fop/end_fop.
1612 1612 */
1613 1613 (void) nfs4_waitfor_purge_complete(vp);
1614 1614
1615 1615 /*
1616 1616 * It is implicit that we are in the open case (create_flag == 0) since
1617 1617 * fh_differs can only be set to a non-zero value in the open case.
1618 1618 */
1619 1619 if (fh_differs != 0 && vpi != NULL)
1620 1620 VN_RELE(vpi);
1621 1621
1622 1622 /*
1623 1623 * Be sure to set *vpp to the correct value before returning.
1624 1624 */
1625 1625 *vpp = vp;
1626 1626
1627 1627 skip_update_dircaches:
1628 1628
1629 1629 nfs4args_copen_free(open_args);
1630 1630 if (setgid_flag) {
1631 1631 nfs4args_verify_free(&argop[8]);
1632 1632 nfs4args_setattr_free(&argop[9]);
1633 1633 }
1634 1634 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1635 1635
1636 1636 if (ncr)
1637 1637 crfree(ncr);
1638 1638 kmem_free(argop, argoplist_size);
1639 1639 return (e.error);
1640 1640 }
1641 1641
1642 1642 /*
1643 1643 * Reopen an open instance. cf. nfs4open_otw().
1644 1644 *
1645 1645 * Errors are returned by the nfs4_error_t parameter.
1646 1646 * - ep->error contains an errno value or zero.
1647 1647 * - if it is zero, ep->stat is set to an NFS status code, if any.
1648 1648 * If the file could not be reopened, but the caller should continue, the
1649 1649 * file is marked dead and no error values are returned. If the caller
1650 1650 * should stop recovering open files and start over, either the ep->error
1651 1651 * value or ep->stat will indicate an error (either something that requires
1652 1652 * recovery or EAGAIN). Note that some recovery (e.g., expired volatile
1653 1653 * filehandles) may be handled silently by this routine.
1654 1654 * - if it is EINTR, ETIMEDOUT, or NFS4_FRC_UNMT_ERR, recovery for lost state
1655 1655 * will be started, so the caller should not do it.
1656 1656 *
1657 1657 * Gotos:
1658 1658 * - kill_file : reopen failed in such a fashion to constitute marking the
1659 1659 * file dead and setting the open stream's 'os_failed_reopen' as 1. This
1660 1660 * is for cases where recovery is not possible.
1661 1661 * - failed_reopen : same as above, except that the file has already been
1662 1662 * marked dead, so no need to do it again.
1663 1663 * - bailout : reopen failed but we are able to recover and retry the reopen -
1664 1664 * either within this function immediately or via the calling function.
1665 1665 */
1666 1666
1667 1667 void
1668 1668 nfs4_reopen(vnode_t *vp, nfs4_open_stream_t *osp, nfs4_error_t *ep,
1669 1669 open_claim_type4 claim, bool_t frc_use_claim_previous,
1670 1670 bool_t is_recov)
1671 1671 {
1672 1672 COMPOUND4args_clnt args;
1673 1673 COMPOUND4res_clnt res;
1674 1674 nfs_argop4 argop[4];
1675 1675 nfs_resop4 *resop;
1676 1676 OPEN4res *op_res = NULL;
1677 1677 OPEN4cargs *open_args;
1678 1678 GETFH4res *gf_res;
1679 1679 rnode4_t *rp = VTOR4(vp);
1680 1680 int doqueue = 1;
1681 1681 cred_t *cr = NULL, *cred_otw = NULL;
1682 1682 nfs4_open_owner_t *oop = NULL;
1683 1683 seqid4 seqid;
1684 1684 nfs4_ga_res_t *garp;
1685 1685 char fn[MAXNAMELEN];
1686 1686 nfs4_recov_state_t recov = {NULL, 0};
1687 1687 nfs4_lost_rqst_t lost_rqst;
1688 1688 mntinfo4_t *mi = VTOMI4(vp);
1689 1689 bool_t abort;
1690 1690 char *failed_msg = "";
1691 1691 int fh_different;
1692 1692 hrtime_t t;
1693 1693 nfs4_bseqid_entry_t *bsep = NULL;
1694 1694
1695 1695 ASSERT(nfs4_consistent_type(vp));
1696 1696 ASSERT(nfs_zone() == mi->mi_zone);
1697 1697
1698 1698 nfs4_error_zinit(ep);
1699 1699
1700 1700 /* this is the cred used to find the open owner */
1701 1701 cr = state_to_cred(osp);
1702 1702 if (cr == NULL) {
1703 1703 failed_msg = "Couldn't reopen: no cred";
1704 1704 goto kill_file;
1705 1705 }
1706 1706 /* use this cred for OTW operations */
1707 1707 cred_otw = nfs4_get_otw_cred(cr, mi, osp->os_open_owner);
1708 1708
1709 1709 top:
1710 1710 nfs4_error_zinit(ep);
1711 1711
1712 1712 if (mi->mi_vfsp->vfs_flag & VFS_UNMOUNTED) {
1713 1713 /* File system has been unmounted, quit */
1714 1714 ep->error = EIO;
1715 1715 failed_msg = "Couldn't reopen: file system has been unmounted";
1716 1716 goto kill_file;
1717 1717 }
1718 1718
1719 1719 oop = osp->os_open_owner;
1720 1720
1721 1721 ASSERT(oop != NULL);
1722 1722 if (oop == NULL) { /* be defensive in non-DEBUG */
1723 1723 failed_msg = "can't reopen: no open owner";
1724 1724 goto kill_file;
1725 1725 }
1726 1726 open_owner_hold(oop);
1727 1727
1728 1728 ep->error = nfs4_start_open_seqid_sync(oop, mi);
1729 1729 if (ep->error) {
1730 1730 open_owner_rele(oop);
1731 1731 oop = NULL;
1732 1732 goto bailout;
1733 1733 }
1734 1734
1735 1735 /*
1736 1736 * If the rnode has a delegation and the delegation has been
1737 1737 * recovered and the server didn't request a recall and the caller
1738 1738 * didn't specifically ask for CLAIM_PREVIOUS (nfs4frlock during
1739 1739 * recovery) and the rnode hasn't been marked dead, then install
1740 1740 * the delegation stateid in the open stream. Otherwise, proceed
1741 1741 * with a CLAIM_PREVIOUS or CLAIM_NULL OPEN.
1742 1742 */
1743 1743 mutex_enter(&rp->r_statev4_lock);
1744 1744 if (rp->r_deleg_type != OPEN_DELEGATE_NONE &&
1745 1745 !rp->r_deleg_return_pending &&
1746 1746 (rp->r_deleg_needs_recovery == OPEN_DELEGATE_NONE) &&
1747 1747 !rp->r_deleg_needs_recall &&
1748 1748 claim != CLAIM_DELEGATE_CUR && !frc_use_claim_previous &&
1749 1749 !(rp->r_flags & R4RECOVERR)) {
1750 1750 mutex_enter(&osp->os_sync_lock);
1751 1751 osp->os_delegation = 1;
1752 1752 osp->open_stateid = rp->r_deleg_stateid;
1753 1753 mutex_exit(&osp->os_sync_lock);
1754 1754 mutex_exit(&rp->r_statev4_lock);
1755 1755 goto bailout;
1756 1756 }
1757 1757 mutex_exit(&rp->r_statev4_lock);
1758 1758
1759 1759 /*
1760 1760 * If the file failed recovery, just quit. This failure need not
1761 1761 * affect other reopens, so don't return an error.
1762 1762 */
1763 1763 mutex_enter(&rp->r_statelock);
1764 1764 if (rp->r_flags & R4RECOVERR) {
1765 1765 mutex_exit(&rp->r_statelock);
1766 1766 ep->error = 0;
1767 1767 goto failed_reopen;
1768 1768 }
1769 1769 mutex_exit(&rp->r_statelock);
1770 1770
1771 1771 /*
1772 1772 * argop is empty here
1773 1773 *
1774 1774 * PUTFH, OPEN, GETATTR
1775 1775 */
1776 1776 args.ctag = TAG_REOPEN;
1777 1777 args.array_len = 4;
1778 1778 args.array = argop;
1779 1779
1780 1780 NFS4_DEBUG(nfs4_client_failover_debug, (CE_NOTE,
1781 1781 "nfs4_reopen: file is type %d, id %s",
1782 1782 vp->v_type, rnode4info(VTOR4(vp))));
1783 1783
1784 1784 argop[0].argop = OP_CPUTFH;
1785 1785
1786 1786 if (claim != CLAIM_PREVIOUS) {
1787 1787 /*
1788 1788 * if this is a file mount then
1789 1789 * use the mntinfo parentfh
1790 1790 */
1791 1791 argop[0].nfs_argop4_u.opcputfh.sfh =
1792 1792 (vp->v_flag & VROOT) ? mi->mi_srvparentfh :
1793 1793 VTOSV(vp)->sv_dfh;
1794 1794 } else {
1795 1795 /* putfh fh to reopen */
1796 1796 argop[0].nfs_argop4_u.opcputfh.sfh = rp->r_fh;
1797 1797 }
1798 1798
1799 1799 argop[1].argop = OP_COPEN;
1800 1800 open_args = &argop[1].nfs_argop4_u.opcopen;
1801 1801 open_args->claim = claim;
1802 1802
1803 1803 if (claim == CLAIM_NULL) {
1804 1804
1805 1805 if ((ep->error = vtoname(vp, fn, MAXNAMELEN)) != 0) {
1806 1806 nfs_cmn_err(ep->error, CE_WARN, "nfs4_reopen: vtoname "
1807 1807 "failed for vp 0x%p for CLAIM_NULL with %m",
1808 1808 (void *)vp);
1809 1809 failed_msg = "Couldn't reopen: vtoname failed for "
1810 1810 "CLAIM_NULL";
1811 1811 /* nothing allocated yet */
1812 1812 goto kill_file;
1813 1813 }
1814 1814
1815 1815 open_args->open_claim4_u.cfile = fn;
1816 1816 } else if (claim == CLAIM_PREVIOUS) {
1817 1817
1818 1818 /*
1819 1819 * We have two cases to deal with here:
1820 1820 * 1) We're being called to reopen files in order to satisfy
1821 1821 * a lock operation request which requires us to explicitly
1822 1822 * reopen files which were opened under a delegation. If
1823 1823 * we're in recovery, we *must* use CLAIM_PREVIOUS. In
1824 1824 * that case, frc_use_claim_previous is TRUE and we must
1825 1825 * use the rnode's current delegation type (r_deleg_type).
1826 1826 * 2) We're reopening files during some form of recovery.
1827 1827 * In this case, frc_use_claim_previous is FALSE and we
1828 1828 * use the delegation type appropriate for recovery
1829 1829 * (r_deleg_needs_recovery).
1830 1830 */
1831 1831 mutex_enter(&rp->r_statev4_lock);
1832 1832 open_args->open_claim4_u.delegate_type =
1833 1833 frc_use_claim_previous ?
1834 1834 rp->r_deleg_type :
1835 1835 rp->r_deleg_needs_recovery;
1836 1836 mutex_exit(&rp->r_statev4_lock);
1837 1837
1838 1838 } else if (claim == CLAIM_DELEGATE_CUR) {
1839 1839
1840 1840 if ((ep->error = vtoname(vp, fn, MAXNAMELEN)) != 0) {
1841 1841 nfs_cmn_err(ep->error, CE_WARN, "nfs4_reopen: vtoname "
1842 1842 "failed for vp 0x%p for CLAIM_DELEGATE_CUR "
1843 1843 "with %m", (void *)vp);
1844 1844 failed_msg = "Couldn't reopen: vtoname failed for "
1845 1845 "CLAIM_DELEGATE_CUR";
1846 1846 /* nothing allocated yet */
1847 1847 goto kill_file;
1848 1848 }
1849 1849
1850 1850 mutex_enter(&rp->r_statev4_lock);
1851 1851 open_args->open_claim4_u.delegate_cur_info.delegate_stateid =
1852 1852 rp->r_deleg_stateid;
1853 1853 mutex_exit(&rp->r_statev4_lock);
1854 1854
1855 1855 open_args->open_claim4_u.delegate_cur_info.cfile = fn;
1856 1856 }
1857 1857 open_args->opentype = OPEN4_NOCREATE;
1858 1858 open_args->owner.clientid = mi2clientid(mi);
1859 1859 open_args->owner.owner_len = sizeof (oop->oo_name);
1860 1860 open_args->owner.owner_val =
1861 1861 kmem_alloc(open_args->owner.owner_len, KM_SLEEP);
1862 1862 bcopy(&oop->oo_name, open_args->owner.owner_val,
1863 1863 open_args->owner.owner_len);
1864 1864 open_args->share_access = 0;
1865 1865 open_args->share_deny = 0;
1866 1866
1867 1867 mutex_enter(&osp->os_sync_lock);
1868 1868 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE, "nfs4_reopen: osp %p rp "
1869 1869 "%p: read acc %"PRIu64" write acc %"PRIu64": open ref count %d: "
1870 1870 "mmap read %"PRIu64" mmap write %"PRIu64" claim %d ",
1871 1871 (void *)osp, (void *)rp, osp->os_share_acc_read,
1872 1872 osp->os_share_acc_write, osp->os_open_ref_count,
1873 1873 osp->os_mmap_read, osp->os_mmap_write, claim));
1874 1874
1875 1875 if (osp->os_share_acc_read || osp->os_mmap_read)
1876 1876 open_args->share_access |= OPEN4_SHARE_ACCESS_READ;
1877 1877 if (osp->os_share_acc_write || osp->os_mmap_write)
1878 1878 open_args->share_access |= OPEN4_SHARE_ACCESS_WRITE;
1879 1879 if (osp->os_share_deny_read)
1880 1880 open_args->share_deny |= OPEN4_SHARE_DENY_READ;
1881 1881 if (osp->os_share_deny_write)
1882 1882 open_args->share_deny |= OPEN4_SHARE_DENY_WRITE;
1883 1883 mutex_exit(&osp->os_sync_lock);
1884 1884
1885 1885 seqid = nfs4_get_open_seqid(oop) + 1;
1886 1886 open_args->seqid = seqid;
1887 1887
1888 1888 /* Construct the getfh part of the compound */
1889 1889 argop[2].argop = OP_GETFH;
1890 1890
1891 1891 /* Construct the getattr part of the compound */
1892 1892 argop[3].argop = OP_GETATTR;
1893 1893 argop[3].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
1894 1894 argop[3].nfs_argop4_u.opgetattr.mi = mi;
1895 1895
1896 1896 t = gethrtime();
1897 1897
1898 1898 rfs4call(mi, &args, &res, cred_otw, &doqueue, 0, ep);
1899 1899
1900 1900 if (ep->error) {
1901 1901 if (!is_recov && !frc_use_claim_previous &&
1902 1902 (ep->error == EINTR || ep->error == ETIMEDOUT ||
1903 1903 NFS4_FRC_UNMT_ERR(ep->error, vp->v_vfsp))) {
1904 1904 nfs4open_save_lost_rqst(ep->error, &lost_rqst, oop,
1905 1905 cred_otw, vp, NULL, open_args);
1906 1906 abort = nfs4_start_recovery(ep,
1907 1907 VTOMI4(vp), vp, NULL, NULL,
1908 1908 lost_rqst.lr_op == OP_OPEN ?
1909 1909 &lost_rqst : NULL, OP_OPEN, NULL, NULL, NULL);
1910 1910 nfs4args_copen_free(open_args);
1911 1911 goto bailout;
1912 1912 }
1913 1913
1914 1914 nfs4args_copen_free(open_args);
1915 1915
1916 1916 if (ep->error == EACCES && cred_otw != cr) {
1917 1917 crfree(cred_otw);
1918 1918 cred_otw = cr;
1919 1919 crhold(cred_otw);
1920 1920 nfs4_end_open_seqid_sync(oop);
1921 1921 open_owner_rele(oop);
1922 1922 oop = NULL;
1923 1923 goto top;
1924 1924 }
1925 1925 if (ep->error == ETIMEDOUT)
1926 1926 goto bailout;
1927 1927 failed_msg = "Couldn't reopen: rpc error";
1928 1928 goto kill_file;
1929 1929 }
1930 1930
1931 1931 if (nfs4_need_to_bump_seqid(&res))
1932 1932 nfs4_set_open_seqid(seqid, oop, args.ctag);
1933 1933
1934 1934 switch (res.status) {
1935 1935 case NFS4_OK:
1936 1936 if (recov.rs_flags & NFS4_RS_DELAY_MSG) {
1937 1937 mutex_enter(&rp->r_statelock);
1938 1938 rp->r_delay_interval = 0;
1939 1939 mutex_exit(&rp->r_statelock);
1940 1940 }
1941 1941 break;
1942 1942 case NFS4ERR_BAD_SEQID:
1943 1943 bsep = nfs4_create_bseqid_entry(oop, NULL, vp, 0,
1944 1944 args.ctag, open_args->seqid);
1945 1945
1946 1946 abort = nfs4_start_recovery(ep, VTOMI4(vp), vp, NULL,
1947 1947 NULL, lost_rqst.lr_op == OP_OPEN ? &lost_rqst :
1948 1948 NULL, OP_OPEN, bsep, NULL, NULL);
1949 1949
1950 1950 nfs4args_copen_free(open_args);
1951 1951 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1952 1952 nfs4_end_open_seqid_sync(oop);
1953 1953 open_owner_rele(oop);
1954 1954 oop = NULL;
1955 1955 kmem_free(bsep, sizeof (*bsep));
1956 1956
1957 1957 goto kill_file;
1958 1958 case NFS4ERR_NO_GRACE:
1959 1959 nfs4args_copen_free(open_args);
1960 1960 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1961 1961 nfs4_end_open_seqid_sync(oop);
1962 1962 open_owner_rele(oop);
1963 1963 oop = NULL;
1964 1964 if (claim == CLAIM_PREVIOUS) {
1965 1965 /*
1966 1966 * Retry as a plain open. We don't need to worry about
1967 1967 * checking the changeinfo: it is acceptable for a
1968 1968 * client to re-open a file and continue processing
1969 1969 * (in the absence of locks).
1970 1970 */
1971 1971 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
1972 1972 "nfs4_reopen: CLAIM_PREVIOUS: NFS4ERR_NO_GRACE; "
1973 1973 "will retry as CLAIM_NULL"));
1974 1974 claim = CLAIM_NULL;
1975 1975 nfs4_mi_kstat_inc_no_grace(mi);
1976 1976 goto top;
1977 1977 }
1978 1978 failed_msg =
1979 1979 "Couldn't reopen: tried reclaim outside grace period. ";
1980 1980 goto kill_file;
1981 1981 case NFS4ERR_GRACE:
1982 1982 nfs4_set_grace_wait(mi);
1983 1983 nfs4args_copen_free(open_args);
1984 1984 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1985 1985 nfs4_end_open_seqid_sync(oop);
1986 1986 open_owner_rele(oop);
1987 1987 oop = NULL;
1988 1988 ep->error = nfs4_wait_for_grace(mi, &recov);
1989 1989 if (ep->error != 0)
1990 1990 goto bailout;
1991 1991 goto top;
1992 1992 case NFS4ERR_DELAY:
1993 1993 nfs4_set_delay_wait(vp);
1994 1994 nfs4args_copen_free(open_args);
1995 1995 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1996 1996 nfs4_end_open_seqid_sync(oop);
1997 1997 open_owner_rele(oop);
1998 1998 oop = NULL;
1999 1999 ep->error = nfs4_wait_for_delay(vp, &recov);
2000 2000 nfs4_mi_kstat_inc_delay(mi);
2001 2001 if (ep->error != 0)
2002 2002 goto bailout;
2003 2003 goto top;
2004 2004 case NFS4ERR_FHEXPIRED:
2005 2005 /* recover filehandle and retry */
2006 2006 abort = nfs4_start_recovery(ep,
2007 2007 mi, vp, NULL, NULL, NULL, OP_OPEN, NULL, NULL, NULL);
2008 2008 nfs4args_copen_free(open_args);
2009 2009 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2010 2010 nfs4_end_open_seqid_sync(oop);
2011 2011 open_owner_rele(oop);
2012 2012 oop = NULL;
2013 2013 if (abort == FALSE)
2014 2014 goto top;
2015 2015 failed_msg = "Couldn't reopen: recovery aborted";
2016 2016 goto kill_file;
2017 2017 case NFS4ERR_RESOURCE:
2018 2018 case NFS4ERR_STALE_CLIENTID:
2019 2019 case NFS4ERR_WRONGSEC:
2020 2020 case NFS4ERR_EXPIRED:
2021 2021 /*
2022 2022 * Do not mark the file dead and let the calling
2023 2023 * function initiate recovery.
2024 2024 */
2025 2025 nfs4args_copen_free(open_args);
2026 2026 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2027 2027 nfs4_end_open_seqid_sync(oop);
2028 2028 open_owner_rele(oop);
2029 2029 oop = NULL;
2030 2030 goto bailout;
2031 2031 case NFS4ERR_ACCESS:
2032 2032 if (cred_otw != cr) {
2033 2033 crfree(cred_otw);
2034 2034 cred_otw = cr;
2035 2035 crhold(cred_otw);
2036 2036 nfs4args_copen_free(open_args);
2037 2037 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2038 2038 nfs4_end_open_seqid_sync(oop);
2039 2039 open_owner_rele(oop);
2040 2040 oop = NULL;
2041 2041 goto top;
2042 2042 }
2043 2043 /* fall through */
2044 2044 default:
2045 2045 NFS4_DEBUG(nfs4_client_failover_debug, (CE_NOTE,
2046 2046 "nfs4_reopen: r_server 0x%p, mi_curr_serv 0x%p, rnode %s",
2047 2047 (void*)VTOR4(vp)->r_server, (void*)mi->mi_curr_serv,
2048 2048 rnode4info(VTOR4(vp))));
2049 2049 failed_msg = "Couldn't reopen: NFSv4 error";
2050 2050 nfs4args_copen_free(open_args);
2051 2051 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2052 2052 goto kill_file;
2053 2053 }
2054 2054
2055 2055 resop = &res.array[1]; /* open res */
2056 2056 op_res = &resop->nfs_resop4_u.opopen;
2057 2057
2058 2058 garp = &res.array[3].nfs_resop4_u.opgetattr.ga_res;
2059 2059
2060 2060 /*
2061 2061 * Check if the path we reopened really is the same
2062 2062 * file. We could end up in a situation where the file
2063 2063 * was removed and a new file created with the same name.
2064 2064 */
2065 2065 resop = &res.array[2];
2066 2066 gf_res = &resop->nfs_resop4_u.opgetfh;
2067 2067 (void) nfs_rw_enter_sig(&mi->mi_fh_lock, RW_READER, 0);
2068 2068 fh_different = (nfs4cmpfh(&rp->r_fh->sfh_fh, &gf_res->object) != 0);
2069 2069 if (fh_different) {
2070 2070 if (mi->mi_fh_expire_type == FH4_PERSISTENT ||
2071 2071 mi->mi_fh_expire_type & FH4_NOEXPIRE_WITH_OPEN) {
2072 2072 /* Oops, we don't have the same file */
2073 2073 if (mi->mi_fh_expire_type == FH4_PERSISTENT)
2074 2074 failed_msg = "Couldn't reopen: Persistent "
2075 2075 "file handle changed";
2076 2076 else
2077 2077 failed_msg = "Couldn't reopen: Volatile "
2078 2078 "(no expire on open) file handle changed";
2079 2079
2080 2080 nfs4args_copen_free(open_args);
2081 2081 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2082 2082 nfs_rw_exit(&mi->mi_fh_lock);
2083 2083 goto kill_file;
2084 2084
2085 2085 } else {
2086 2086 /*
2087 2087 * We have volatile file handles that don't compare.
2088 2088 * If the fids are the same then we assume that the
2089 2089 * file handle expired but the rnode still refers to
2090 2090 * the same file object.
2091 2091 *
2092 2092 * First check that we have fids or not.
2093 2093 * If we don't we have a dumb server so we will
2094 2094 * just assume every thing is ok for now.
2095 2095 */
2096 2096 if (!ep->error && garp->n4g_va.va_mask & AT_NODEID &&
2097 2097 rp->r_attr.va_mask & AT_NODEID &&
2098 2098 rp->r_attr.va_nodeid != garp->n4g_va.va_nodeid) {
2099 2099 /*
2100 2100 * We have fids, but they don't
2101 2101 * compare. So kill the file.
2102 2102 */
2103 2103 failed_msg =
2104 2104 "Couldn't reopen: file handle changed"
2105 2105 " due to mismatched fids";
2106 2106 nfs4args_copen_free(open_args);
2107 2107 (void) xdr_free(xdr_COMPOUND4res_clnt,
2108 2108 (caddr_t)&res);
2109 2109 nfs_rw_exit(&mi->mi_fh_lock);
2110 2110 goto kill_file;
2111 2111 } else {
2112 2112 /*
2113 2113 * We have volatile file handles that refers
2114 2114 * to the same file (at least they have the
2115 2115 * same fid) or we don't have fids so we
2116 2116 * can't tell. :(. We'll be a kind and accepting
2117 2117 * client so we'll update the rnode's file
2118 2118 * handle with the otw handle.
2119 2119 *
2120 2120 * We need to drop mi->mi_fh_lock since
2121 2121 * sh4_update acquires it. Since there is
2122 2122 * only one recovery thread there is no
2123 2123 * race.
2124 2124 */
2125 2125 nfs_rw_exit(&mi->mi_fh_lock);
2126 2126 sfh4_update(rp->r_fh, &gf_res->object);
2127 2127 }
2128 2128 }
2129 2129 } else {
2130 2130 nfs_rw_exit(&mi->mi_fh_lock);
2131 2131 }
2132 2132
2133 2133 ASSERT(nfs4_consistent_type(vp));
2134 2134
2135 2135 /*
2136 2136 * If the server wanted an OPEN_CONFIRM but that fails, just start
2137 2137 * over. Presumably if there is a persistent error it will show up
2138 2138 * when we resend the OPEN.
2139 2139 */
2140 2140 if (op_res->rflags & OPEN4_RESULT_CONFIRM) {
2141 2141 bool_t retry_open = FALSE;
2142 2142
2143 2143 nfs4open_confirm(vp, &seqid, &op_res->stateid,
2144 2144 cred_otw, is_recov, &retry_open,
2145 2145 oop, FALSE, ep, NULL);
2146 2146 if (ep->error || ep->stat) {
2147 2147 nfs4args_copen_free(open_args);
2148 2148 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2149 2149 nfs4_end_open_seqid_sync(oop);
2150 2150 open_owner_rele(oop);
2151 2151 oop = NULL;
2152 2152 goto top;
2153 2153 }
2154 2154 }
2155 2155
2156 2156 mutex_enter(&osp->os_sync_lock);
2157 2157 osp->open_stateid = op_res->stateid;
2158 2158 osp->os_delegation = 0;
2159 2159 /*
2160 2160 * Need to reset this bitfield for the possible case where we were
2161 2161 * going to OTW CLOSE the file, got a non-recoverable error, and before
2162 2162 * we could retry the CLOSE, OPENed the file again.
2163 2163 */
2164 2164 ASSERT(osp->os_open_owner->oo_seqid_inuse);
2165 2165 osp->os_final_close = 0;
2166 2166 osp->os_force_close = 0;
2167 2167 if (claim == CLAIM_DELEGATE_CUR || claim == CLAIM_PREVIOUS)
2168 2168 osp->os_dc_openacc = open_args->share_access;
2169 2169 mutex_exit(&osp->os_sync_lock);
2170 2170
2171 2171 nfs4_end_open_seqid_sync(oop);
2172 2172
2173 2173 /* accept delegation, if any */
2174 2174 nfs4_delegation_accept(rp, claim, op_res, garp, cred_otw);
2175 2175
2176 2176 nfs4args_copen_free(open_args);
2177 2177
2178 2178 nfs4_attr_cache(vp, garp, t, cr, TRUE, NULL);
2179 2179
2180 2180 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2181 2181
2182 2182 ASSERT(nfs4_consistent_type(vp));
2183 2183
2184 2184 open_owner_rele(oop);
2185 2185 crfree(cr);
2186 2186 crfree(cred_otw);
2187 2187 return;
2188 2188
2189 2189 kill_file:
2190 2190 nfs4_fail_recov(vp, failed_msg, ep->error, ep->stat);
2191 2191 failed_reopen:
2192 2192 NFS4_DEBUG(nfs4_open_stream_debug, (CE_NOTE,
2193 2193 "nfs4_reopen: setting os_failed_reopen for osp %p, cr %p, rp %s",
2194 2194 (void *)osp, (void *)cr, rnode4info(rp)));
2195 2195 mutex_enter(&osp->os_sync_lock);
2196 2196 osp->os_failed_reopen = 1;
2197 2197 mutex_exit(&osp->os_sync_lock);
2198 2198 bailout:
2199 2199 if (oop != NULL) {
2200 2200 nfs4_end_open_seqid_sync(oop);
2201 2201 open_owner_rele(oop);
2202 2202 }
2203 2203 if (cr != NULL)
2204 2204 crfree(cr);
2205 2205 if (cred_otw != NULL)
2206 2206 crfree(cred_otw);
2207 2207 }
2208 2208
2209 2209 /* for . and .. OPENs */
2210 2210 /* ARGSUSED */
2211 2211 static int
2212 2212 nfs4_open_non_reg_file(vnode_t **vpp, int flag, cred_t *cr)
2213 2213 {
2214 2214 rnode4_t *rp;
2215 2215 nfs4_ga_res_t gar;
2216 2216
2217 2217 ASSERT(nfs_zone() == VTOMI4(*vpp)->mi_zone);
2218 2218
2219 2219 /*
2220 2220 * If close-to-open consistency checking is turned off or
2221 2221 * if there is no cached data, we can avoid
2222 2222 * the over the wire getattr. Otherwise, force a
2223 2223 * call to the server to get fresh attributes and to
2224 2224 * check caches. This is required for close-to-open
2225 2225 * consistency.
2226 2226 */
2227 2227 rp = VTOR4(*vpp);
2228 2228 if (VTOMI4(*vpp)->mi_flags & MI4_NOCTO ||
2229 2229 (rp->r_dir == NULL && !nfs4_has_pages(*vpp)))
2230 2230 return (0);
2231 2231
2232 2232 gar.n4g_va.va_mask = AT_ALL;
2233 2233 return (nfs4_getattr_otw(*vpp, &gar, cr, 0));
2234 2234 }
2235 2235
2236 2236 /*
2237 2237 * CLOSE a file
2238 2238 */
2239 2239 /* ARGSUSED */
2240 2240 static int
2241 2241 nfs4_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
2242 2242 caller_context_t *ct)
2243 2243 {
2244 2244 rnode4_t *rp;
2245 2245 int error = 0;
2246 2246 int r_error = 0;
2247 2247 int n4error = 0;
2248 2248 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
2249 2249
2250 2250 /*
2251 2251 * Remove client state for this (lockowner, file) pair.
2252 2252 * Issue otw v4 call to have the server do the same.
2253 2253 */
2254 2254
2255 2255 rp = VTOR4(vp);
2256 2256
2257 2257 /*
2258 2258 * zone_enter(2) prevents processes from changing zones with NFS files
2259 2259 * open; if we happen to get here from the wrong zone we can't do
2260 2260 * anything over the wire.
2261 2261 */
2262 2262 if (VTOMI4(vp)->mi_zone != nfs_zone()) {
2263 2263 /*
2264 2264 * We could attempt to clean up locks, except we're sure
2265 2265 * that the current process didn't acquire any locks on
2266 2266 * the file: any attempt to lock a file belong to another zone
2267 2267 * will fail, and one can't lock an NFS file and then change
2268 2268 * zones, as that fails too.
2269 2269 *
2270 2270 * Returning an error here is the sane thing to do. A
2271 2271 * subsequent call to VN_RELE() which translates to a
2272 2272 * nfs4_inactive() will clean up state: if the zone of the
2273 2273 * vnode's origin is still alive and kicking, the inactive
2274 2274 * thread will handle the request (from the correct zone), and
2275 2275 * everything (minus the OTW close call) should be OK. If the
2276 2276 * zone is going away nfs4_async_inactive() will throw away
2277 2277 * delegations, open streams and cached pages inline.
2278 2278 */
2279 2279 return (EIO);
2280 2280 }
2281 2281
2282 2282 /*
2283 2283 * If we are using local locking for this filesystem, then
2284 2284 * release all of the SYSV style record locks. Otherwise,
2285 2285 * we are doing network locking and we need to release all
2286 2286 * of the network locks. All of the locks held by this
2287 2287 * process on this file are released no matter what the
2288 2288 * incoming reference count is.
2289 2289 */
2290 2290 if (VTOMI4(vp)->mi_flags & MI4_LLOCK) {
2291 2291 cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
2292 2292 cleanshares(vp, ttoproc(curthread)->p_pid);
2293 2293 } else
2294 2294 e.error = nfs4_lockrelease(vp, flag, offset, cr);
2295 2295
2296 2296 if (e.error) {
2297 2297 struct lm_sysid *lmsid;
2298 2298 lmsid = nfs4_find_sysid(VTOMI4(vp));
2299 2299 if (lmsid == NULL) {
2300 2300 DTRACE_PROBE2(unknown__sysid, int, e.error,
2301 2301 vnode_t *, vp);
2302 2302 } else {
2303 2303 cleanlocks(vp, ttoproc(curthread)->p_pid,
2304 2304 (lm_sysidt(lmsid) | LM_SYSID_CLIENT));
2305 2305 }
2306 2306 return (e.error);
2307 2307 }
2308 2308
2309 2309 if (count > 1)
2310 2310 return (0);
2311 2311
2312 2312 /*
2313 2313 * If the file has been `unlinked', then purge the
2314 2314 * DNLC so that this vnode will get reycled quicker
2315 2315 * and the .nfs* file on the server will get removed.
2316 2316 */
2317 2317 if (rp->r_unldvp != NULL)
2318 2318 dnlc_purge_vp(vp);
2319 2319
2320 2320 /*
2321 2321 * If the file was open for write and there are pages,
2322 2322 * do a synchronous flush and commit of all of the
2323 2323 * dirty and uncommitted pages.
2324 2324 */
2325 2325 ASSERT(!e.error);
2326 2326 if ((flag & FWRITE) && nfs4_has_pages(vp))
2327 2327 error = nfs4_putpage_commit(vp, 0, 0, cr);
2328 2328
2329 2329 mutex_enter(&rp->r_statelock);
2330 2330 r_error = rp->r_error;
2331 2331 rp->r_error = 0;
2332 2332 mutex_exit(&rp->r_statelock);
2333 2333
2334 2334 /*
2335 2335 * If this file type is one for which no explicit 'open' was
2336 2336 * done, then bail now (ie. no need for protocol 'close'). If
2337 2337 * there was an error w/the vm subsystem, return _that_ error,
2338 2338 * otherwise, return any errors that may've been reported via
2339 2339 * the rnode.
2340 2340 */
2341 2341 if (vp->v_type != VREG)
2342 2342 return (error ? error : r_error);
2343 2343
2344 2344 /*
2345 2345 * The sync putpage commit may have failed above, but since
2346 2346 * we're working w/a regular file, we need to do the protocol
2347 2347 * 'close' (nfs4close_one will figure out if an otw close is
2348 2348 * needed or not). Report any errors _after_ doing the protocol
2349 2349 * 'close'.
2350 2350 */
2351 2351 nfs4close_one(vp, NULL, cr, flag, NULL, &e, CLOSE_NORM, 0, 0, 0);
2352 2352 n4error = e.error ? e.error : geterrno4(e.stat);
2353 2353
2354 2354 /*
2355 2355 * Error reporting prio (Hi -> Lo)
2356 2356 *
2357 2357 * i) nfs4_putpage_commit (error)
2358 2358 * ii) rnode's (r_error)
2359 2359 * iii) nfs4close_one (n4error)
2360 2360 */
2361 2361 return (error ? error : (r_error ? r_error : n4error));
2362 2362 }
2363 2363
2364 2364 /*
2365 2365 * Initialize *lost_rqstp.
2366 2366 */
2367 2367
2368 2368 static void
2369 2369 nfs4close_save_lost_rqst(int error, nfs4_lost_rqst_t *lost_rqstp,
2370 2370 nfs4_open_owner_t *oop, nfs4_open_stream_t *osp, cred_t *cr,
2371 2371 vnode_t *vp)
2372 2372 {
2373 2373 if (error != ETIMEDOUT && error != EINTR &&
2374 2374 !NFS4_FRC_UNMT_ERR(error, vp->v_vfsp)) {
2375 2375 lost_rqstp->lr_op = 0;
2376 2376 return;
2377 2377 }
2378 2378
2379 2379 NFS4_DEBUG(nfs4_lost_rqst_debug, (CE_NOTE,
2380 2380 "nfs4close_save_lost_rqst: error %d", error));
2381 2381
2382 2382 lost_rqstp->lr_op = OP_CLOSE;
2383 2383 /*
2384 2384 * The vp is held and rele'd via the recovery code.
2385 2385 * See nfs4_save_lost_rqst.
2386 2386 */
2387 2387 lost_rqstp->lr_vp = vp;
2388 2388 lost_rqstp->lr_dvp = NULL;
2389 2389 lost_rqstp->lr_oop = oop;
2390 2390 lost_rqstp->lr_osp = osp;
2391 2391 ASSERT(osp != NULL);
2392 2392 ASSERT(mutex_owned(&osp->os_sync_lock));
2393 2393 osp->os_pending_close = 1;
2394 2394 lost_rqstp->lr_lop = NULL;
2395 2395 lost_rqstp->lr_cr = cr;
2396 2396 lost_rqstp->lr_flk = NULL;
2397 2397 lost_rqstp->lr_putfirst = FALSE;
2398 2398 }
2399 2399
2400 2400 /*
2401 2401 * Assumes you already have the open seqid sync grabbed as well as the
2402 2402 * 'os_sync_lock'. Note: this will release the open seqid sync and
2403 2403 * 'os_sync_lock' if client recovery starts. Calling functions have to
2404 2404 * be prepared to handle this.
2405 2405 *
2406 2406 * 'recov' is returned as 1 if the CLOSE operation detected client recovery
2407 2407 * was needed and was started, and that the calling function should retry
2408 2408 * this function; otherwise it is returned as 0.
2409 2409 *
2410 2410 * Errors are returned via the nfs4_error_t parameter.
2411 2411 */
2412 2412 static void
2413 2413 nfs4close_otw(rnode4_t *rp, cred_t *cred_otw, nfs4_open_owner_t *oop,
2414 2414 nfs4_open_stream_t *osp, int *recov, int *did_start_seqid_syncp,
2415 2415 nfs4_close_type_t close_type, nfs4_error_t *ep, int *have_sync_lockp)
2416 2416 {
2417 2417 COMPOUND4args_clnt args;
2418 2418 COMPOUND4res_clnt res;
2419 2419 CLOSE4args *close_args;
2420 2420 nfs_resop4 *resop;
2421 2421 nfs_argop4 argop[3];
2422 2422 int doqueue = 1;
2423 2423 mntinfo4_t *mi;
2424 2424 seqid4 seqid;
2425 2425 vnode_t *vp;
2426 2426 bool_t needrecov = FALSE;
2427 2427 nfs4_lost_rqst_t lost_rqst;
2428 2428 hrtime_t t;
2429 2429
2430 2430 ASSERT(nfs_zone() == VTOMI4(RTOV4(rp))->mi_zone);
2431 2431
2432 2432 ASSERT(MUTEX_HELD(&osp->os_sync_lock));
2433 2433
2434 2434 NFS4_DEBUG(nfs4_client_state_debug, (CE_NOTE, "nfs4close_otw"));
2435 2435
2436 2436 /* Only set this to 1 if recovery is started */
2437 2437 *recov = 0;
2438 2438
2439 2439 /* do the OTW call to close the file */
2440 2440
2441 2441 if (close_type == CLOSE_RESEND)
2442 2442 args.ctag = TAG_CLOSE_LOST;
2443 2443 else if (close_type == CLOSE_AFTER_RESEND)
2444 2444 args.ctag = TAG_CLOSE_UNDO;
2445 2445 else
2446 2446 args.ctag = TAG_CLOSE;
2447 2447
2448 2448 args.array_len = 3;
2449 2449 args.array = argop;
2450 2450
2451 2451 vp = RTOV4(rp);
2452 2452
2453 2453 mi = VTOMI4(vp);
2454 2454
2455 2455 /* putfh target fh */
2456 2456 argop[0].argop = OP_CPUTFH;
2457 2457 argop[0].nfs_argop4_u.opcputfh.sfh = rp->r_fh;
2458 2458
2459 2459 argop[1].argop = OP_GETATTR;
2460 2460 argop[1].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
2461 2461 argop[1].nfs_argop4_u.opgetattr.mi = mi;
2462 2462
2463 2463 argop[2].argop = OP_CLOSE;
2464 2464 close_args = &argop[2].nfs_argop4_u.opclose;
2465 2465
2466 2466 seqid = nfs4_get_open_seqid(oop) + 1;
2467 2467
2468 2468 close_args->seqid = seqid;
2469 2469 close_args->open_stateid = osp->open_stateid;
2470 2470
2471 2471 NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE,
2472 2472 "nfs4close_otw: %s call, rp %s", needrecov ? "recov" : "first",
2473 2473 rnode4info(rp)));
2474 2474
2475 2475 t = gethrtime();
2476 2476
2477 2477 rfs4call(mi, &args, &res, cred_otw, &doqueue, 0, ep);
2478 2478
2479 2479 if (!ep->error && nfs4_need_to_bump_seqid(&res)) {
2480 2480 nfs4_set_open_seqid(seqid, oop, args.ctag);
2481 2481 }
2482 2482
2483 2483 needrecov = nfs4_needs_recovery(ep, TRUE, mi->mi_vfsp);
2484 2484 if (ep->error && !needrecov) {
2485 2485 /*
2486 2486 * if there was an error and no recovery is to be done
2487 2487 * then then set up the file to flush its cache if
2488 2488 * needed for the next caller.
2489 2489 */
2490 2490 mutex_enter(&rp->r_statelock);
2491 2491 PURGE_ATTRCACHE4_LOCKED(rp);
2492 2492 rp->r_flags &= ~R4WRITEMODIFIED;
2493 2493 mutex_exit(&rp->r_statelock);
2494 2494 return;
2495 2495 }
2496 2496
2497 2497 if (needrecov) {
2498 2498 bool_t abort;
2499 2499 nfs4_bseqid_entry_t *bsep = NULL;
2500 2500
2501 2501 if (close_type != CLOSE_RESEND)
2502 2502 nfs4close_save_lost_rqst(ep->error, &lost_rqst, oop,
2503 2503 osp, cred_otw, vp);
2504 2504
2505 2505 if (!ep->error && res.status == NFS4ERR_BAD_SEQID)
2506 2506 bsep = nfs4_create_bseqid_entry(oop, NULL, vp,
2507 2507 0, args.ctag, close_args->seqid);
2508 2508
2509 2509 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
2510 2510 "nfs4close_otw: initiating recovery. error %d "
2511 2511 "res.status %d", ep->error, res.status));
2512 2512
2513 2513 /*
2514 2514 * Drop the 'os_sync_lock' here so we don't hit
2515 2515 * a potential recursive mutex_enter via an
2516 2516 * 'open_stream_hold()'.
2517 2517 */
2518 2518 mutex_exit(&osp->os_sync_lock);
2519 2519 *have_sync_lockp = 0;
2520 2520 abort = nfs4_start_recovery(ep, VTOMI4(vp), vp, NULL, NULL,
2521 2521 (close_type != CLOSE_RESEND &&
2522 2522 lost_rqst.lr_op == OP_CLOSE) ? &lost_rqst : NULL,
2523 2523 OP_CLOSE, bsep, NULL, NULL);
2524 2524
2525 2525 /* drop open seq sync, and let the calling function regrab it */
2526 2526 nfs4_end_open_seqid_sync(oop);
2527 2527 *did_start_seqid_syncp = 0;
2528 2528
2529 2529 if (bsep)
2530 2530 kmem_free(bsep, sizeof (*bsep));
2531 2531 /*
2532 2532 * For signals, the caller wants to quit, so don't say to
2533 2533 * retry. For forced unmount, if it's a user thread, it
2534 2534 * wants to quit. If it's a recovery thread, the retry
2535 2535 * will happen higher-up on the call stack. Either way,
2536 2536 * don't say to retry.
2537 2537 */
2538 2538 if (abort == FALSE && ep->error != EINTR &&
2539 2539 !NFS4_FRC_UNMT_ERR(ep->error, mi->mi_vfsp) &&
2540 2540 close_type != CLOSE_RESEND &&
2541 2541 close_type != CLOSE_AFTER_RESEND)
2542 2542 *recov = 1;
2543 2543 else
2544 2544 *recov = 0;
2545 2545
2546 2546 if (!ep->error)
2547 2547 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2548 2548 return;
2549 2549 }
2550 2550
2551 2551 if (res.status) {
2552 2552 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2553 2553 return;
2554 2554 }
2555 2555
2556 2556 mutex_enter(&rp->r_statev4_lock);
2557 2557 rp->created_v4 = 0;
2558 2558 mutex_exit(&rp->r_statev4_lock);
2559 2559
2560 2560 resop = &res.array[2];
2561 2561 osp->open_stateid = resop->nfs_resop4_u.opclose.open_stateid;
2562 2562 osp->os_valid = 0;
2563 2563
2564 2564 /*
2565 2565 * This removes the reference obtained at OPEN; ie, when the
2566 2566 * open stream structure was created.
2567 2567 *
2568 2568 * We don't have to worry about calling 'open_stream_rele'
2569 2569 * since we our currently holding a reference to the open
2570 2570 * stream which means the count cannot go to 0 with this
2571 2571 * decrement.
2572 2572 */
2573 2573 ASSERT(osp->os_ref_count >= 2);
2574 2574 osp->os_ref_count--;
2575 2575
2576 2576 if (!ep->error)
2577 2577 nfs4_attr_cache(vp,
2578 2578 &res.array[1].nfs_resop4_u.opgetattr.ga_res,
2579 2579 t, cred_otw, TRUE, NULL);
2580 2580
2581 2581 NFS4_DEBUG(nfs4_client_state_debug, (CE_NOTE, "nfs4close_otw:"
2582 2582 " returning %d", ep->error));
2583 2583
2584 2584 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
2585 2585 }
2586 2586
2587 2587 /* ARGSUSED */
2588 2588 static int
2589 2589 nfs4_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
2590 2590 caller_context_t *ct)
2591 2591 {
2592 2592 rnode4_t *rp;
2593 2593 u_offset_t off;
2594 2594 offset_t diff;
2595 2595 uint_t on;
2596 2596 uint_t n;
2597 2597 caddr_t base;
2598 2598 uint_t flags;
2599 2599 int error;
2600 2600 mntinfo4_t *mi;
2601 2601
2602 2602 rp = VTOR4(vp);
2603 2603
2604 2604 ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER));
2605 2605
2606 2606 if (IS_SHADOW(vp, rp))
2607 2607 vp = RTOV4(rp);
2608 2608
2609 2609 if (vp->v_type != VREG)
2610 2610 return (EISDIR);
2611 2611
2612 2612 mi = VTOMI4(vp);
2613 2613
2614 2614 if (nfs_zone() != mi->mi_zone)
2615 2615 return (EIO);
2616 2616
2617 2617 if (uiop->uio_resid == 0)
2618 2618 return (0);
2619 2619
2620 2620 if (uiop->uio_loffset < 0 || uiop->uio_loffset + uiop->uio_resid < 0)
2621 2621 return (EINVAL);
2622 2622
2623 2623 mutex_enter(&rp->r_statelock);
2624 2624 if (rp->r_flags & R4RECOVERRP)
2625 2625 error = (rp->r_error ? rp->r_error : EIO);
2626 2626 else
2627 2627 error = 0;
2628 2628 mutex_exit(&rp->r_statelock);
2629 2629 if (error)
2630 2630 return (error);
2631 2631
2632 2632 /*
2633 2633 * Bypass VM if caching has been disabled (e.g., locking) or if
2634 2634 * using client-side direct I/O and the file is not mmap'd and
2635 2635 * there are no cached pages.
2636 2636 */
2637 2637 if ((vp->v_flag & VNOCACHE) ||
2638 2638 (((rp->r_flags & R4DIRECTIO) || (mi->mi_flags & MI4_DIRECTIO)) &&
2639 2639 rp->r_mapcnt == 0 && rp->r_inmap == 0 && !nfs4_has_pages(vp))) {
2640 2640 size_t resid = 0;
2641 2641
2642 2642 return (nfs4read(vp, NULL, uiop->uio_loffset,
2643 2643 uiop->uio_resid, &resid, cr, FALSE, uiop));
2644 2644 }
2645 2645
2646 2646 error = 0;
2647 2647
2648 2648 do {
2649 2649 off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
2650 2650 on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
2651 2651 n = MIN(MAXBSIZE - on, uiop->uio_resid);
2652 2652
2653 2653 if (error = nfs4_validate_caches(vp, cr))
2654 2654 break;
2655 2655
2656 2656 mutex_enter(&rp->r_statelock);
2657 2657 while (rp->r_flags & R4INCACHEPURGE) {
2658 2658 if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) {
2659 2659 mutex_exit(&rp->r_statelock);
2660 2660 return (EINTR);
2661 2661 }
2662 2662 }
2663 2663 diff = rp->r_size - uiop->uio_loffset;
2664 2664 mutex_exit(&rp->r_statelock);
2665 2665 if (diff <= 0)
2666 2666 break;
2667 2667 if (diff < n)
2668 2668 n = (uint_t)diff;
2669 2669
2670 2670 if (vpm_enable) {
2671 2671 /*
2672 2672 * Copy data.
2673 2673 */
2674 2674 error = vpm_data_copy(vp, off + on, n, uiop,
2675 2675 1, NULL, 0, S_READ);
2676 2676 } else {
2677 2677 base = segmap_getmapflt(segkmap, vp, off + on, n, 1,
2678 2678 S_READ);
2679 2679
2680 2680 error = uiomove(base + on, n, UIO_READ, uiop);
2681 2681 }
2682 2682
2683 2683 if (!error) {
2684 2684 /*
2685 2685 * If read a whole block or read to eof,
2686 2686 * won't need this buffer again soon.
2687 2687 */
2688 2688 mutex_enter(&rp->r_statelock);
2689 2689 if (n + on == MAXBSIZE ||
2690 2690 uiop->uio_loffset == rp->r_size)
2691 2691 flags = SM_DONTNEED;
2692 2692 else
2693 2693 flags = 0;
2694 2694 mutex_exit(&rp->r_statelock);
2695 2695 if (vpm_enable) {
2696 2696 error = vpm_sync_pages(vp, off, n, flags);
2697 2697 } else {
2698 2698 error = segmap_release(segkmap, base, flags);
2699 2699 }
2700 2700 } else {
2701 2701 if (vpm_enable) {
2702 2702 (void) vpm_sync_pages(vp, off, n, 0);
2703 2703 } else {
2704 2704 (void) segmap_release(segkmap, base, 0);
2705 2705 }
2706 2706 }
2707 2707 } while (!error && uiop->uio_resid > 0);
2708 2708
2709 2709 return (error);
2710 2710 }
2711 2711
2712 2712 /* ARGSUSED */
2713 2713 static int
2714 2714 nfs4_write(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
2715 2715 caller_context_t *ct)
2716 2716 {
2717 2717 rlim64_t limit = uiop->uio_llimit;
2718 2718 rnode4_t *rp;
2719 2719 u_offset_t off;
2720 2720 caddr_t base;
2721 2721 uint_t flags;
2722 2722 int remainder;
2723 2723 size_t n;
2724 2724 int on;
2725 2725 int error;
2726 2726 int resid;
2727 2727 u_offset_t offset;
2728 2728 mntinfo4_t *mi;
2729 2729 uint_t bsize;
2730 2730
2731 2731 rp = VTOR4(vp);
2732 2732
2733 2733 if (IS_SHADOW(vp, rp))
2734 2734 vp = RTOV4(rp);
2735 2735
2736 2736 if (vp->v_type != VREG)
2737 2737 return (EISDIR);
2738 2738
2739 2739 mi = VTOMI4(vp);
2740 2740
2741 2741 if (nfs_zone() != mi->mi_zone)
2742 2742 return (EIO);
2743 2743
2744 2744 if (uiop->uio_resid == 0)
2745 2745 return (0);
2746 2746
2747 2747 mutex_enter(&rp->r_statelock);
2748 2748 if (rp->r_flags & R4RECOVERRP)
2749 2749 error = (rp->r_error ? rp->r_error : EIO);
2750 2750 else
2751 2751 error = 0;
2752 2752 mutex_exit(&rp->r_statelock);
2753 2753 if (error)
2754 2754 return (error);
2755 2755
2756 2756 if (ioflag & FAPPEND) {
2757 2757 struct vattr va;
2758 2758
2759 2759 /*
2760 2760 * Must serialize if appending.
2761 2761 */
2762 2762 if (nfs_rw_lock_held(&rp->r_rwlock, RW_READER)) {
2763 2763 nfs_rw_exit(&rp->r_rwlock);
2764 2764 if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER,
2765 2765 INTR4(vp)))
2766 2766 return (EINTR);
2767 2767 }
2768 2768
2769 2769 va.va_mask = AT_SIZE;
2770 2770 error = nfs4getattr(vp, &va, cr);
2771 2771 if (error)
2772 2772 return (error);
2773 2773 uiop->uio_loffset = va.va_size;
2774 2774 }
2775 2775
2776 2776 offset = uiop->uio_loffset + uiop->uio_resid;
2777 2777
2778 2778 if (uiop->uio_loffset < (offset_t)0 || offset < 0)
2779 2779 return (EINVAL);
2780 2780
2781 2781 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
2782 2782 limit = MAXOFFSET_T;
2783 2783
2784 2784 /*
2785 2785 * Check to make sure that the process will not exceed
2786 2786 * its limit on file size. It is okay to write up to
2787 2787 * the limit, but not beyond. Thus, the write which
2788 2788 * reaches the limit will be short and the next write
2789 2789 * will return an error.
2790 2790 */
2791 2791 remainder = 0;
2792 2792 if (offset > uiop->uio_llimit) {
2793 2793 remainder = offset - uiop->uio_llimit;
2794 2794 uiop->uio_resid = uiop->uio_llimit - uiop->uio_loffset;
2795 2795 if (uiop->uio_resid <= 0) {
2796 2796 proc_t *p = ttoproc(curthread);
2797 2797
2798 2798 uiop->uio_resid += remainder;
2799 2799 mutex_enter(&p->p_lock);
2800 2800 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE],
2801 2801 p->p_rctls, p, RCA_UNSAFE_SIGINFO);
2802 2802 mutex_exit(&p->p_lock);
2803 2803 return (EFBIG);
2804 2804 }
2805 2805 }
2806 2806
2807 2807 /* update the change attribute, if we have a write delegation */
2808 2808
2809 2809 mutex_enter(&rp->r_statev4_lock);
2810 2810 if (rp->r_deleg_type == OPEN_DELEGATE_WRITE)
2811 2811 rp->r_deleg_change++;
2812 2812
2813 2813 mutex_exit(&rp->r_statev4_lock);
2814 2814
2815 2815 if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR4(vp)))
2816 2816 return (EINTR);
2817 2817
2818 2818 /*
2819 2819 * Bypass VM if caching has been disabled (e.g., locking) or if
2820 2820 * using client-side direct I/O and the file is not mmap'd and
2821 2821 * there are no cached pages.
2822 2822 */
2823 2823 if ((vp->v_flag & VNOCACHE) ||
2824 2824 (((rp->r_flags & R4DIRECTIO) || (mi->mi_flags & MI4_DIRECTIO)) &&
2825 2825 rp->r_mapcnt == 0 && rp->r_inmap == 0 && !nfs4_has_pages(vp))) {
2826 2826 size_t bufsize;
2827 2827 int count;
2828 2828 u_offset_t org_offset;
2829 2829 stable_how4 stab_comm;
2830 2830 nfs4_fwrite:
2831 2831 if (rp->r_flags & R4STALE) {
2832 2832 resid = uiop->uio_resid;
2833 2833 offset = uiop->uio_loffset;
2834 2834 error = rp->r_error;
2835 2835 /*
2836 2836 * A close may have cleared r_error, if so,
2837 2837 * propagate ESTALE error return properly
2838 2838 */
2839 2839 if (error == 0)
2840 2840 error = ESTALE;
2841 2841 goto bottom;
2842 2842 }
2843 2843
2844 2844 bufsize = MIN(uiop->uio_resid, mi->mi_stsize);
2845 2845 base = kmem_alloc(bufsize, KM_SLEEP);
2846 2846 do {
2847 2847 if (ioflag & FDSYNC)
2848 2848 stab_comm = DATA_SYNC4;
2849 2849 else
2850 2850 stab_comm = FILE_SYNC4;
2851 2851 resid = uiop->uio_resid;
2852 2852 offset = uiop->uio_loffset;
2853 2853 count = MIN(uiop->uio_resid, bufsize);
2854 2854 org_offset = uiop->uio_loffset;
2855 2855 error = uiomove(base, count, UIO_WRITE, uiop);
2856 2856 if (!error) {
2857 2857 error = nfs4write(vp, base, org_offset,
2858 2858 count, cr, &stab_comm);
2859 2859 if (!error) {
2860 2860 mutex_enter(&rp->r_statelock);
2861 2861 if (rp->r_size < uiop->uio_loffset)
2862 2862 rp->r_size = uiop->uio_loffset;
2863 2863 mutex_exit(&rp->r_statelock);
2864 2864 }
2865 2865 }
2866 2866 } while (!error && uiop->uio_resid > 0);
2867 2867 kmem_free(base, bufsize);
2868 2868 goto bottom;
2869 2869 }
2870 2870
2871 2871 bsize = vp->v_vfsp->vfs_bsize;
2872 2872
2873 2873 do {
2874 2874 off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
2875 2875 on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
2876 2876 n = MIN(MAXBSIZE - on, uiop->uio_resid);
2877 2877
2878 2878 resid = uiop->uio_resid;
2879 2879 offset = uiop->uio_loffset;
2880 2880
2881 2881 if (rp->r_flags & R4STALE) {
2882 2882 error = rp->r_error;
2883 2883 /*
2884 2884 * A close may have cleared r_error, if so,
2885 2885 * propagate ESTALE error return properly
2886 2886 */
2887 2887 if (error == 0)
2888 2888 error = ESTALE;
2889 2889 break;
2890 2890 }
2891 2891
2892 2892 /*
2893 2893 * Don't create dirty pages faster than they
2894 2894 * can be cleaned so that the system doesn't
2895 2895 * get imbalanced. If the async queue is
2896 2896 * maxed out, then wait for it to drain before
2897 2897 * creating more dirty pages. Also, wait for
2898 2898 * any threads doing pagewalks in the vop_getattr
2899 2899 * entry points so that they don't block for
2900 2900 * long periods.
2901 2901 */
2902 2902 mutex_enter(&rp->r_statelock);
2903 2903 while ((mi->mi_max_threads != 0 &&
2904 2904 rp->r_awcount > 2 * mi->mi_max_threads) ||
2905 2905 rp->r_gcount > 0) {
2906 2906 if (INTR4(vp)) {
2907 2907 klwp_t *lwp = ttolwp(curthread);
2908 2908
2909 2909 if (lwp != NULL)
2910 2910 lwp->lwp_nostop++;
2911 2911 if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) {
2912 2912 mutex_exit(&rp->r_statelock);
2913 2913 if (lwp != NULL)
2914 2914 lwp->lwp_nostop--;
2915 2915 error = EINTR;
2916 2916 goto bottom;
2917 2917 }
2918 2918 if (lwp != NULL)
2919 2919 lwp->lwp_nostop--;
2920 2920 } else
2921 2921 cv_wait(&rp->r_cv, &rp->r_statelock);
2922 2922 }
2923 2923 mutex_exit(&rp->r_statelock);
2924 2924
2925 2925 /*
2926 2926 * Touch the page and fault it in if it is not in core
2927 2927 * before segmap_getmapflt or vpm_data_copy can lock it.
2928 2928 * This is to avoid the deadlock if the buffer is mapped
2929 2929 * to the same file through mmap which we want to write.
2930 2930 */
2931 2931 uio_prefaultpages((long)n, uiop);
2932 2932
2933 2933 if (vpm_enable) {
2934 2934 /*
2935 2935 * It will use kpm mappings, so no need to
2936 2936 * pass an address.
2937 2937 */
2938 2938 error = writerp4(rp, NULL, n, uiop, 0);
2939 2939 } else {
2940 2940 if (segmap_kpm) {
2941 2941 int pon = uiop->uio_loffset & PAGEOFFSET;
2942 2942 size_t pn = MIN(PAGESIZE - pon,
2943 2943 uiop->uio_resid);
2944 2944 int pagecreate;
2945 2945
2946 2946 mutex_enter(&rp->r_statelock);
2947 2947 pagecreate = (pon == 0) && (pn == PAGESIZE ||
2948 2948 uiop->uio_loffset + pn >= rp->r_size);
2949 2949 mutex_exit(&rp->r_statelock);
2950 2950
2951 2951 base = segmap_getmapflt(segkmap, vp, off + on,
2952 2952 pn, !pagecreate, S_WRITE);
2953 2953
2954 2954 error = writerp4(rp, base + pon, n, uiop,
2955 2955 pagecreate);
2956 2956
2957 2957 } else {
2958 2958 base = segmap_getmapflt(segkmap, vp, off + on,
2959 2959 n, 0, S_READ);
2960 2960 error = writerp4(rp, base + on, n, uiop, 0);
2961 2961 }
2962 2962 }
2963 2963
2964 2964 if (!error) {
2965 2965 if (mi->mi_flags & MI4_NOAC)
2966 2966 flags = SM_WRITE;
2967 2967 else if ((uiop->uio_loffset % bsize) == 0 ||
2968 2968 IS_SWAPVP(vp)) {
2969 2969 /*
2970 2970 * Have written a whole block.
2971 2971 * Start an asynchronous write
2972 2972 * and mark the buffer to
2973 2973 * indicate that it won't be
2974 2974 * needed again soon.
2975 2975 */
2976 2976 flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
2977 2977 } else
2978 2978 flags = 0;
2979 2979 if ((ioflag & (FSYNC|FDSYNC)) ||
2980 2980 (rp->r_flags & R4OUTOFSPACE)) {
2981 2981 flags &= ~SM_ASYNC;
2982 2982 flags |= SM_WRITE;
2983 2983 }
2984 2984 if (vpm_enable) {
2985 2985 error = vpm_sync_pages(vp, off, n, flags);
2986 2986 } else {
2987 2987 error = segmap_release(segkmap, base, flags);
2988 2988 }
2989 2989 } else {
2990 2990 if (vpm_enable) {
2991 2991 (void) vpm_sync_pages(vp, off, n, 0);
2992 2992 } else {
2993 2993 (void) segmap_release(segkmap, base, 0);
2994 2994 }
2995 2995 /*
2996 2996 * In the event that we got an access error while
2997 2997 * faulting in a page for a write-only file just
2998 2998 * force a write.
2999 2999 */
3000 3000 if (error == EACCES)
3001 3001 goto nfs4_fwrite;
3002 3002 }
3003 3003 } while (!error && uiop->uio_resid > 0);
3004 3004
3005 3005 bottom:
3006 3006 if (error) {
3007 3007 uiop->uio_resid = resid + remainder;
3008 3008 uiop->uio_loffset = offset;
3009 3009 } else {
3010 3010 uiop->uio_resid += remainder;
3011 3011
3012 3012 mutex_enter(&rp->r_statev4_lock);
3013 3013 if (rp->r_deleg_type == OPEN_DELEGATE_WRITE) {
3014 3014 gethrestime(&rp->r_attr.va_mtime);
3015 3015 rp->r_attr.va_ctime = rp->r_attr.va_mtime;
3016 3016 }
3017 3017 mutex_exit(&rp->r_statev4_lock);
3018 3018 }
3019 3019
3020 3020 nfs_rw_exit(&rp->r_lkserlock);
3021 3021
3022 3022 return (error);
3023 3023 }
3024 3024
3025 3025 /*
3026 3026 * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED}
3027 3027 */
3028 3028 static int
3029 3029 nfs4_rdwrlbn(vnode_t *vp, page_t *pp, u_offset_t off, size_t len,
3030 3030 int flags, cred_t *cr)
3031 3031 {
3032 3032 struct buf *bp;
3033 3033 int error;
3034 3034 page_t *savepp;
3035 3035 uchar_t fsdata;
3036 3036 stable_how4 stab_comm;
3037 3037
3038 3038 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
3039 3039 bp = pageio_setup(pp, len, vp, flags);
3040 3040 ASSERT(bp != NULL);
3041 3041
3042 3042 /*
3043 3043 * pageio_setup should have set b_addr to 0. This
3044 3044 * is correct since we want to do I/O on a page
3045 3045 * boundary. bp_mapin will use this addr to calculate
3046 3046 * an offset, and then set b_addr to the kernel virtual
3047 3047 * address it allocated for us.
3048 3048 */
3049 3049 ASSERT(bp->b_un.b_addr == 0);
3050 3050
3051 3051 bp->b_edev = 0;
3052 3052 bp->b_dev = 0;
3053 3053 bp->b_lblkno = lbtodb(off);
3054 3054 bp->b_file = vp;
3055 3055 bp->b_offset = (offset_t)off;
3056 3056 bp_mapin(bp);
3057 3057
3058 3058 if ((flags & (B_WRITE|B_ASYNC)) == (B_WRITE|B_ASYNC) &&
3059 3059 freemem > desfree)
3060 3060 stab_comm = UNSTABLE4;
3061 3061 else
3062 3062 stab_comm = FILE_SYNC4;
3063 3063
3064 3064 error = nfs4_bio(bp, &stab_comm, cr, FALSE);
3065 3065
3066 3066 bp_mapout(bp);
3067 3067 pageio_done(bp);
3068 3068
3069 3069 if (stab_comm == UNSTABLE4)
3070 3070 fsdata = C_DELAYCOMMIT;
3071 3071 else
3072 3072 fsdata = C_NOCOMMIT;
3073 3073
3074 3074 savepp = pp;
3075 3075 do {
3076 3076 pp->p_fsdata = fsdata;
3077 3077 } while ((pp = pp->p_next) != savepp);
3078 3078
3079 3079 return (error);
3080 3080 }
3081 3081
3082 3082 /*
3083 3083 */
3084 3084 static int
3085 3085 nfs4rdwr_check_osid(vnode_t *vp, nfs4_error_t *ep, cred_t *cr)
3086 3086 {
3087 3087 nfs4_open_owner_t *oop;
3088 3088 nfs4_open_stream_t *osp;
3089 3089 rnode4_t *rp = VTOR4(vp);
3090 3090 mntinfo4_t *mi = VTOMI4(vp);
3091 3091 int reopen_needed;
3092 3092
3093 3093 ASSERT(nfs_zone() == mi->mi_zone);
3094 3094
3095 3095
3096 3096 oop = find_open_owner(cr, NFS4_PERM_CREATED, mi);
3097 3097 if (!oop)
3098 3098 return (EIO);
3099 3099
3100 3100 /* returns with 'os_sync_lock' held */
3101 3101 osp = find_open_stream(oop, rp);
3102 3102 if (!osp) {
3103 3103 open_owner_rele(oop);
3104 3104 return (EIO);
3105 3105 }
3106 3106
3107 3107 if (osp->os_failed_reopen) {
3108 3108 mutex_exit(&osp->os_sync_lock);
3109 3109 open_stream_rele(osp, rp);
3110 3110 open_owner_rele(oop);
3111 3111 return (EIO);
3112 3112 }
3113 3113
3114 3114 /*
3115 3115 * Determine whether a reopen is needed. If this
3116 3116 * is a delegation open stream, then the os_delegation bit
3117 3117 * should be set.
3118 3118 */
3119 3119
3120 3120 reopen_needed = osp->os_delegation;
3121 3121
3122 3122 mutex_exit(&osp->os_sync_lock);
3123 3123 open_owner_rele(oop);
3124 3124
3125 3125 if (reopen_needed) {
3126 3126 nfs4_error_zinit(ep);
3127 3127 nfs4_reopen(vp, osp, ep, CLAIM_NULL, FALSE, FALSE);
3128 3128 mutex_enter(&osp->os_sync_lock);
3129 3129 if (ep->error || ep->stat || osp->os_failed_reopen) {
3130 3130 mutex_exit(&osp->os_sync_lock);
3131 3131 open_stream_rele(osp, rp);
3132 3132 return (EIO);
3133 3133 }
3134 3134 mutex_exit(&osp->os_sync_lock);
3135 3135 }
3136 3136 open_stream_rele(osp, rp);
3137 3137
3138 3138 return (0);
3139 3139 }
3140 3140
3141 3141 /*
3142 3142 * Write to file. Writes to remote server in largest size
3143 3143 * chunks that the server can handle. Write is synchronous.
3144 3144 */
3145 3145 static int
3146 3146 nfs4write(vnode_t *vp, caddr_t base, u_offset_t offset, int count, cred_t *cr,
3147 3147 stable_how4 *stab_comm)
3148 3148 {
3149 3149 mntinfo4_t *mi;
3150 3150 COMPOUND4args_clnt args;
3151 3151 COMPOUND4res_clnt res;
3152 3152 WRITE4args *wargs;
3153 3153 WRITE4res *wres;
3154 3154 nfs_argop4 argop[2];
3155 3155 nfs_resop4 *resop;
3156 3156 int tsize;
3157 3157 stable_how4 stable;
3158 3158 rnode4_t *rp;
3159 3159 int doqueue = 1;
3160 3160 bool_t needrecov;
3161 3161 nfs4_recov_state_t recov_state;
3162 3162 nfs4_stateid_types_t sid_types;
3163 3163 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
3164 3164 int recov;
3165 3165
3166 3166 rp = VTOR4(vp);
3167 3167 mi = VTOMI4(vp);
3168 3168
3169 3169 ASSERT(nfs_zone() == mi->mi_zone);
3170 3170
3171 3171 stable = *stab_comm;
3172 3172 *stab_comm = FILE_SYNC4;
3173 3173
3174 3174 needrecov = FALSE;
3175 3175 recov_state.rs_flags = 0;
3176 3176 recov_state.rs_num_retry_despite_err = 0;
3177 3177 nfs4_init_stateid_types(&sid_types);
3178 3178
3179 3179 /* Is curthread the recovery thread? */
3180 3180 mutex_enter(&mi->mi_lock);
3181 3181 recov = (mi->mi_recovthread == curthread);
3182 3182 mutex_exit(&mi->mi_lock);
3183 3183
3184 3184 recov_retry:
3185 3185 args.ctag = TAG_WRITE;
3186 3186 args.array_len = 2;
3187 3187 args.array = argop;
3188 3188
3189 3189 if (!recov) {
3190 3190 e.error = nfs4_start_fop(VTOMI4(vp), vp, NULL, OH_WRITE,
3191 3191 &recov_state, NULL);
3192 3192 if (e.error)
3193 3193 return (e.error);
3194 3194 }
3195 3195
3196 3196 /* 0. putfh target fh */
3197 3197 argop[0].argop = OP_CPUTFH;
3198 3198 argop[0].nfs_argop4_u.opcputfh.sfh = rp->r_fh;
3199 3199
3200 3200 /* 1. write */
3201 3201 nfs4args_write(&argop[1], stable, rp, cr, &wargs, &sid_types);
3202 3202
3203 3203 do {
3204 3204
3205 3205 wargs->offset = (offset4)offset;
3206 3206 wargs->data_val = base;
3207 3207
3208 3208 if (mi->mi_io_kstats) {
3209 3209 mutex_enter(&mi->mi_lock);
3210 3210 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
3211 3211 mutex_exit(&mi->mi_lock);
3212 3212 }
3213 3213
3214 3214 if ((vp->v_flag & VNOCACHE) ||
3215 3215 (rp->r_flags & R4DIRECTIO) ||
3216 3216 (mi->mi_flags & MI4_DIRECTIO))
3217 3217 tsize = MIN(mi->mi_stsize, count);
3218 3218 else
3219 3219 tsize = MIN(mi->mi_curwrite, count);
3220 3220 wargs->data_len = (uint_t)tsize;
3221 3221 rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
3222 3222
3223 3223 if (mi->mi_io_kstats) {
3224 3224 mutex_enter(&mi->mi_lock);
3225 3225 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
3226 3226 mutex_exit(&mi->mi_lock);
3227 3227 }
3228 3228
3229 3229 if (!recov) {
3230 3230 needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
3231 3231 if (e.error && !needrecov) {
3232 3232 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_WRITE,
3233 3233 &recov_state, needrecov);
3234 3234 return (e.error);
3235 3235 }
3236 3236 } else {
3237 3237 if (e.error)
3238 3238 return (e.error);
3239 3239 }
3240 3240
3241 3241 /*
3242 3242 * Do handling of OLD_STATEID outside
3243 3243 * of the normal recovery framework.
3244 3244 *
3245 3245 * If write receives a BAD stateid error while using a
3246 3246 * delegation stateid, retry using the open stateid (if it
3247 3247 * exists). If it doesn't have an open stateid, reopen the
3248 3248 * file first, then retry.
3249 3249 */
3250 3250 if (!e.error && res.status == NFS4ERR_OLD_STATEID &&
3251 3251 sid_types.cur_sid_type != SPEC_SID) {
3252 3252 nfs4_save_stateid(&wargs->stateid, &sid_types);
3253 3253 if (!recov)
3254 3254 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_WRITE,
3255 3255 &recov_state, needrecov);
3256 3256 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3257 3257 goto recov_retry;
3258 3258 } else if (e.error == 0 && res.status == NFS4ERR_BAD_STATEID &&
3259 3259 sid_types.cur_sid_type == DEL_SID) {
3260 3260 nfs4_save_stateid(&wargs->stateid, &sid_types);
3261 3261 mutex_enter(&rp->r_statev4_lock);
3262 3262 rp->r_deleg_return_pending = TRUE;
3263 3263 mutex_exit(&rp->r_statev4_lock);
3264 3264 if (nfs4rdwr_check_osid(vp, &e, cr)) {
3265 3265 if (!recov)
3266 3266 nfs4_end_fop(mi, vp, NULL, OH_WRITE,
3267 3267 &recov_state, needrecov);
3268 3268 (void) xdr_free(xdr_COMPOUND4res_clnt,
3269 3269 (caddr_t)&res);
3270 3270 return (EIO);
3271 3271 }
3272 3272 if (!recov)
3273 3273 nfs4_end_fop(mi, vp, NULL, OH_WRITE,
3274 3274 &recov_state, needrecov);
3275 3275 /* hold needed for nfs4delegreturn_thread */
3276 3276 VN_HOLD(vp);
3277 3277 nfs4delegreturn_async(rp, (NFS4_DR_PUSH|NFS4_DR_REOPEN|
3278 3278 NFS4_DR_DISCARD), FALSE);
3279 3279 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3280 3280 goto recov_retry;
3281 3281 }
3282 3282
3283 3283 if (needrecov) {
3284 3284 bool_t abort;
3285 3285
3286 3286 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
3287 3287 "nfs4write: client got error %d, res.status %d"
3288 3288 ", so start recovery", e.error, res.status));
3289 3289
3290 3290 abort = nfs4_start_recovery(&e,
3291 3291 VTOMI4(vp), vp, NULL, &wargs->stateid,
3292 3292 NULL, OP_WRITE, NULL, NULL, NULL);
3293 3293 if (!e.error) {
3294 3294 e.error = geterrno4(res.status);
3295 3295 (void) xdr_free(xdr_COMPOUND4res_clnt,
3296 3296 (caddr_t)&res);
3297 3297 }
3298 3298 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_WRITE,
3299 3299 &recov_state, needrecov);
3300 3300 if (abort == FALSE)
3301 3301 goto recov_retry;
3302 3302 return (e.error);
3303 3303 }
3304 3304
3305 3305 if (res.status) {
3306 3306 e.error = geterrno4(res.status);
3307 3307 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3308 3308 if (!recov)
3309 3309 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_WRITE,
3310 3310 &recov_state, needrecov);
3311 3311 return (e.error);
3312 3312 }
3313 3313
3314 3314 resop = &res.array[1]; /* write res */
3315 3315 wres = &resop->nfs_resop4_u.opwrite;
3316 3316
3317 3317 if ((int)wres->count > tsize) {
3318 3318 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3319 3319
3320 3320 zcmn_err(getzoneid(), CE_WARN,
3321 3321 "nfs4write: server wrote %u, requested was %u",
3322 3322 (int)wres->count, tsize);
3323 3323 if (!recov)
3324 3324 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_WRITE,
3325 3325 &recov_state, needrecov);
3326 3326 return (EIO);
3327 3327 }
3328 3328 if (wres->committed == UNSTABLE4) {
3329 3329 *stab_comm = UNSTABLE4;
3330 3330 if (wargs->stable == DATA_SYNC4 ||
3331 3331 wargs->stable == FILE_SYNC4) {
3332 3332 (void) xdr_free(xdr_COMPOUND4res_clnt,
3333 3333 (caddr_t)&res);
3334 3334 zcmn_err(getzoneid(), CE_WARN,
3335 3335 "nfs4write: server %s did not commit "
3336 3336 "to stable storage",
3337 3337 rp->r_server->sv_hostname);
3338 3338 if (!recov)
3339 3339 nfs4_end_fop(VTOMI4(vp), vp, NULL,
3340 3340 OH_WRITE, &recov_state, needrecov);
3341 3341 return (EIO);
3342 3342 }
3343 3343 }
3344 3344
3345 3345 tsize = (int)wres->count;
3346 3346 count -= tsize;
3347 3347 base += tsize;
3348 3348 offset += tsize;
3349 3349 if (mi->mi_io_kstats) {
3350 3350 mutex_enter(&mi->mi_lock);
3351 3351 KSTAT_IO_PTR(mi->mi_io_kstats)->writes++;
3352 3352 KSTAT_IO_PTR(mi->mi_io_kstats)->nwritten +=
3353 3353 tsize;
3354 3354 mutex_exit(&mi->mi_lock);
3355 3355 }
3356 3356 lwp_stat_update(LWP_STAT_OUBLK, 1);
3357 3357 mutex_enter(&rp->r_statelock);
3358 3358 if (rp->r_flags & R4HAVEVERF) {
3359 3359 if (rp->r_writeverf != wres->writeverf) {
3360 3360 nfs4_set_mod(vp);
3361 3361 rp->r_writeverf = wres->writeverf;
3362 3362 }
3363 3363 } else {
3364 3364 rp->r_writeverf = wres->writeverf;
3365 3365 rp->r_flags |= R4HAVEVERF;
3366 3366 }
3367 3367 PURGE_ATTRCACHE4_LOCKED(rp);
3368 3368 rp->r_flags |= R4WRITEMODIFIED;
3369 3369 gethrestime(&rp->r_attr.va_mtime);
3370 3370 rp->r_attr.va_ctime = rp->r_attr.va_mtime;
3371 3371 mutex_exit(&rp->r_statelock);
3372 3372 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3373 3373 } while (count);
3374 3374
3375 3375 if (!recov)
3376 3376 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_WRITE, &recov_state,
3377 3377 needrecov);
3378 3378
3379 3379 return (e.error);
3380 3380 }
3381 3381
3382 3382 /*
3383 3383 * Read from a file. Reads data in largest chunks our interface can handle.
3384 3384 */
3385 3385 static int
3386 3386 nfs4read(vnode_t *vp, caddr_t base, offset_t offset, int count,
3387 3387 size_t *residp, cred_t *cr, bool_t async, struct uio *uiop)
3388 3388 {
3389 3389 mntinfo4_t *mi;
3390 3390 COMPOUND4args_clnt args;
3391 3391 COMPOUND4res_clnt res;
3392 3392 READ4args *rargs;
3393 3393 nfs_argop4 argop[2];
3394 3394 int tsize;
3395 3395 int doqueue;
3396 3396 rnode4_t *rp;
3397 3397 int data_len;
3398 3398 bool_t is_eof;
3399 3399 bool_t needrecov = FALSE;
3400 3400 nfs4_recov_state_t recov_state;
3401 3401 nfs4_stateid_types_t sid_types;
3402 3402 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
3403 3403
3404 3404 rp = VTOR4(vp);
3405 3405 mi = VTOMI4(vp);
3406 3406 doqueue = 1;
3407 3407
3408 3408 ASSERT(nfs_zone() == mi->mi_zone);
3409 3409
3410 3410 args.ctag = async ? TAG_READAHEAD : TAG_READ;
3411 3411
3412 3412 args.array_len = 2;
3413 3413 args.array = argop;
3414 3414
3415 3415 nfs4_init_stateid_types(&sid_types);
3416 3416
3417 3417 recov_state.rs_flags = 0;
3418 3418 recov_state.rs_num_retry_despite_err = 0;
3419 3419
3420 3420 recov_retry:
3421 3421 e.error = nfs4_start_fop(mi, vp, NULL, OH_READ,
3422 3422 &recov_state, NULL);
3423 3423 if (e.error)
3424 3424 return (e.error);
3425 3425
3426 3426 /* putfh target fh */
3427 3427 argop[0].argop = OP_CPUTFH;
3428 3428 argop[0].nfs_argop4_u.opcputfh.sfh = rp->r_fh;
3429 3429
3430 3430 /* read */
3431 3431 argop[1].argop = OP_READ;
3432 3432 rargs = &argop[1].nfs_argop4_u.opread;
3433 3433 rargs->stateid = nfs4_get_stateid(cr, rp, curproc->p_pidp->pid_id, mi,
3434 3434 OP_READ, &sid_types, async);
3435 3435
3436 3436 do {
3437 3437 if (mi->mi_io_kstats) {
3438 3438 mutex_enter(&mi->mi_lock);
3439 3439 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
3440 3440 mutex_exit(&mi->mi_lock);
3441 3441 }
3442 3442
3443 3443 NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE,
3444 3444 "nfs4read: %s call, rp %s",
3445 3445 needrecov ? "recov" : "first",
3446 3446 rnode4info(rp)));
3447 3447
3448 3448 if ((vp->v_flag & VNOCACHE) ||
3449 3449 (rp->r_flags & R4DIRECTIO) ||
3450 3450 (mi->mi_flags & MI4_DIRECTIO))
3451 3451 tsize = MIN(mi->mi_tsize, count);
3452 3452 else
3453 3453 tsize = MIN(mi->mi_curread, count);
3454 3454
3455 3455 rargs->offset = (offset4)offset;
3456 3456 rargs->count = (count4)tsize;
3457 3457 rargs->res_data_val_alt = NULL;
3458 3458 rargs->res_mblk = NULL;
3459 3459 rargs->res_uiop = NULL;
3460 3460 rargs->res_maxsize = 0;
3461 3461 rargs->wlist = NULL;
3462 3462
3463 3463 if (uiop)
3464 3464 rargs->res_uiop = uiop;
3465 3465 else
3466 3466 rargs->res_data_val_alt = base;
3467 3467 rargs->res_maxsize = tsize;
3468 3468
3469 3469 rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
3470 3470 #ifdef DEBUG
3471 3471 if (nfs4read_error_inject) {
3472 3472 res.status = nfs4read_error_inject;
3473 3473 nfs4read_error_inject = 0;
3474 3474 }
3475 3475 #endif
3476 3476
3477 3477 if (mi->mi_io_kstats) {
3478 3478 mutex_enter(&mi->mi_lock);
3479 3479 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
3480 3480 mutex_exit(&mi->mi_lock);
3481 3481 }
3482 3482
3483 3483 needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
3484 3484 if (e.error != 0 && !needrecov) {
3485 3485 nfs4_end_fop(mi, vp, NULL, OH_READ,
3486 3486 &recov_state, needrecov);
3487 3487 return (e.error);
3488 3488 }
3489 3489
3490 3490 /*
3491 3491 * Do proper retry for OLD and BAD stateid errors outside
3492 3492 * of the normal recovery framework. There are two differences
3493 3493 * between async and sync reads. The first is that we allow
3494 3494 * retry on BAD_STATEID for async reads, but not sync reads.
3495 3495 * The second is that we mark the file dead for a failed
3496 3496 * attempt with a special stateid for sync reads, but just
3497 3497 * return EIO for async reads.
3498 3498 *
3499 3499 * If a sync read receives a BAD stateid error while using a
3500 3500 * delegation stateid, retry using the open stateid (if it
3501 3501 * exists). If it doesn't have an open stateid, reopen the
3502 3502 * file first, then retry.
3503 3503 */
3504 3504 if (e.error == 0 && (res.status == NFS4ERR_OLD_STATEID ||
3505 3505 res.status == NFS4ERR_BAD_STATEID) && async) {
3506 3506 nfs4_end_fop(mi, vp, NULL, OH_READ,
3507 3507 &recov_state, needrecov);
3508 3508 if (sid_types.cur_sid_type == SPEC_SID) {
3509 3509 (void) xdr_free(xdr_COMPOUND4res_clnt,
3510 3510 (caddr_t)&res);
3511 3511 return (EIO);
3512 3512 }
3513 3513 nfs4_save_stateid(&rargs->stateid, &sid_types);
3514 3514 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3515 3515 goto recov_retry;
3516 3516 } else if (e.error == 0 && res.status == NFS4ERR_OLD_STATEID &&
3517 3517 !async && sid_types.cur_sid_type != SPEC_SID) {
3518 3518 nfs4_save_stateid(&rargs->stateid, &sid_types);
3519 3519 nfs4_end_fop(mi, vp, NULL, OH_READ,
3520 3520 &recov_state, needrecov);
3521 3521 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3522 3522 goto recov_retry;
3523 3523 } else if (e.error == 0 && res.status == NFS4ERR_BAD_STATEID &&
3524 3524 sid_types.cur_sid_type == DEL_SID) {
3525 3525 nfs4_save_stateid(&rargs->stateid, &sid_types);
3526 3526 mutex_enter(&rp->r_statev4_lock);
3527 3527 rp->r_deleg_return_pending = TRUE;
3528 3528 mutex_exit(&rp->r_statev4_lock);
3529 3529 if (nfs4rdwr_check_osid(vp, &e, cr)) {
3530 3530 nfs4_end_fop(mi, vp, NULL, OH_READ,
3531 3531 &recov_state, needrecov);
3532 3532 (void) xdr_free(xdr_COMPOUND4res_clnt,
3533 3533 (caddr_t)&res);
3534 3534 return (EIO);
3535 3535 }
3536 3536 nfs4_end_fop(mi, vp, NULL, OH_READ,
3537 3537 &recov_state, needrecov);
3538 3538 /* hold needed for nfs4delegreturn_thread */
3539 3539 VN_HOLD(vp);
3540 3540 nfs4delegreturn_async(rp, (NFS4_DR_PUSH|NFS4_DR_REOPEN|
3541 3541 NFS4_DR_DISCARD), FALSE);
3542 3542 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3543 3543 goto recov_retry;
3544 3544 }
3545 3545 if (needrecov) {
3546 3546 bool_t abort;
3547 3547
3548 3548 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
3549 3549 "nfs4read: initiating recovery\n"));
3550 3550 abort = nfs4_start_recovery(&e,
3551 3551 mi, vp, NULL, &rargs->stateid,
3552 3552 NULL, OP_READ, NULL, NULL, NULL);
3553 3553 nfs4_end_fop(mi, vp, NULL, OH_READ,
3554 3554 &recov_state, needrecov);
3555 3555 /*
3556 3556 * Do not retry if we got OLD_STATEID using a special
3557 3557 * stateid. This avoids looping with a broken server.
3558 3558 */
3559 3559 if (e.error == 0 && res.status == NFS4ERR_OLD_STATEID &&
3560 3560 sid_types.cur_sid_type == SPEC_SID)
3561 3561 abort = TRUE;
3562 3562
3563 3563 if (abort == FALSE) {
3564 3564 /*
3565 3565 * Need to retry all possible stateids in
3566 3566 * case the recovery error wasn't stateid
3567 3567 * related or the stateids have become
3568 3568 * stale (server reboot).
3569 3569 */
3570 3570 nfs4_init_stateid_types(&sid_types);
3571 3571 (void) xdr_free(xdr_COMPOUND4res_clnt,
3572 3572 (caddr_t)&res);
3573 3573 goto recov_retry;
3574 3574 }
3575 3575
3576 3576 if (!e.error) {
3577 3577 e.error = geterrno4(res.status);
3578 3578 (void) xdr_free(xdr_COMPOUND4res_clnt,
3579 3579 (caddr_t)&res);
3580 3580 }
3581 3581 return (e.error);
3582 3582 }
3583 3583
3584 3584 if (res.status) {
3585 3585 e.error = geterrno4(res.status);
3586 3586 nfs4_end_fop(mi, vp, NULL, OH_READ,
3587 3587 &recov_state, needrecov);
3588 3588 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3589 3589 return (e.error);
3590 3590 }
3591 3591
3592 3592 data_len = res.array[1].nfs_resop4_u.opread.data_len;
3593 3593 count -= data_len;
3594 3594 if (base)
3595 3595 base += data_len;
3596 3596 offset += data_len;
3597 3597 if (mi->mi_io_kstats) {
3598 3598 mutex_enter(&mi->mi_lock);
3599 3599 KSTAT_IO_PTR(mi->mi_io_kstats)->reads++;
3600 3600 KSTAT_IO_PTR(mi->mi_io_kstats)->nread += data_len;
3601 3601 mutex_exit(&mi->mi_lock);
3602 3602 }
3603 3603 lwp_stat_update(LWP_STAT_INBLK, 1);
3604 3604 is_eof = res.array[1].nfs_resop4_u.opread.eof;
3605 3605 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3606 3606
3607 3607 } while (count && !is_eof);
3608 3608
3609 3609 *residp = count;
3610 3610
3611 3611 nfs4_end_fop(mi, vp, NULL, OH_READ, &recov_state, needrecov);
3612 3612
3613 3613 return (e.error);
3614 3614 }
3615 3615
3616 3616 /* ARGSUSED */
3617 3617 static int
3618 3618 nfs4_ioctl(vnode_t *vp, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp,
3619 3619 caller_context_t *ct)
3620 3620 {
3621 3621 if (nfs_zone() != VTOMI4(vp)->mi_zone)
3622 3622 return (EIO);
3623 3623 switch (cmd) {
3624 3624 case _FIODIRECTIO:
3625 3625 return (nfs4_directio(vp, (int)arg, cr));
3626 3626 default:
3627 3627 return (ENOTTY);
3628 3628 }
3629 3629 }
3630 3630
3631 3631 /* ARGSUSED */
3632 3632 int
3633 3633 nfs4_getattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
3634 3634 caller_context_t *ct)
3635 3635 {
3636 3636 int error;
3637 3637 rnode4_t *rp = VTOR4(vp);
3638 3638
3639 3639 if (nfs_zone() != VTOMI4(vp)->mi_zone)
3640 3640 return (EIO);
3641 3641 /*
3642 3642 * If it has been specified that the return value will
3643 3643 * just be used as a hint, and we are only being asked
3644 3644 * for size, fsid or rdevid, then return the client's
3645 3645 * notion of these values without checking to make sure
3646 3646 * that the attribute cache is up to date.
3647 3647 * The whole point is to avoid an over the wire GETATTR
3648 3648 * call.
3649 3649 */
3650 3650 if (flags & ATTR_HINT) {
3651 3651 if (!(vap->va_mask & ~(AT_SIZE | AT_FSID | AT_RDEV))) {
3652 3652 mutex_enter(&rp->r_statelock);
3653 3653 if (vap->va_mask & AT_SIZE)
3654 3654 vap->va_size = rp->r_size;
3655 3655 if (vap->va_mask & AT_FSID)
3656 3656 vap->va_fsid = rp->r_attr.va_fsid;
3657 3657 if (vap->va_mask & AT_RDEV)
3658 3658 vap->va_rdev = rp->r_attr.va_rdev;
3659 3659 mutex_exit(&rp->r_statelock);
3660 3660 return (0);
3661 3661 }
3662 3662 }
3663 3663
3664 3664 /*
3665 3665 * Only need to flush pages if asking for the mtime
3666 3666 * and if there any dirty pages or any outstanding
3667 3667 * asynchronous (write) requests for this file.
3668 3668 */
3669 3669 if (vap->va_mask & AT_MTIME) {
3670 3670 rp = VTOR4(vp);
3671 3671 if (nfs4_has_pages(vp)) {
3672 3672 mutex_enter(&rp->r_statev4_lock);
3673 3673 if (rp->r_deleg_type != OPEN_DELEGATE_WRITE) {
3674 3674 mutex_exit(&rp->r_statev4_lock);
3675 3675 if (rp->r_flags & R4DIRTY ||
3676 3676 rp->r_awcount > 0) {
3677 3677 mutex_enter(&rp->r_statelock);
3678 3678 rp->r_gcount++;
3679 3679 mutex_exit(&rp->r_statelock);
3680 3680 error =
3681 3681 nfs4_putpage(vp, (u_offset_t)0,
3682 3682 0, 0, cr, NULL);
3683 3683 mutex_enter(&rp->r_statelock);
3684 3684 if (error && (error == ENOSPC ||
3685 3685 error == EDQUOT)) {
3686 3686 if (!rp->r_error)
3687 3687 rp->r_error = error;
3688 3688 }
3689 3689 if (--rp->r_gcount == 0)
3690 3690 cv_broadcast(&rp->r_cv);
3691 3691 mutex_exit(&rp->r_statelock);
3692 3692 }
3693 3693 } else {
3694 3694 mutex_exit(&rp->r_statev4_lock);
3695 3695 }
3696 3696 }
3697 3697 }
3698 3698 return (nfs4getattr(vp, vap, cr));
3699 3699 }
3700 3700
3701 3701 int
3702 3702 nfs4_compare_modes(mode_t from_server, mode_t on_client)
3703 3703 {
3704 3704 /*
3705 3705 * If these are the only two bits cleared
3706 3706 * on the server then return 0 (OK) else
3707 3707 * return 1 (BAD).
3708 3708 */
3709 3709 on_client &= ~(S_ISUID|S_ISGID);
3710 3710 if (on_client == from_server)
3711 3711 return (0);
3712 3712 else
3713 3713 return (1);
3714 3714 }
3715 3715
3716 3716 /*ARGSUSED4*/
3717 3717 static int
3718 3718 nfs4_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
3719 3719 caller_context_t *ct)
3720 3720 {
3721 3721 int error;
3722 3722
3723 3723 if (vap->va_mask & AT_NOSET)
3724 3724 return (EINVAL);
3725 3725
3726 3726 if (nfs_zone() != VTOMI4(vp)->mi_zone)
3727 3727 return (EIO);
3728 3728
3729 3729 /*
3730 3730 * Don't call secpolicy_vnode_setattr, the client cannot
3731 3731 * use its cached attributes to make security decisions
3732 3732 * as the server may be faking mode bits or mapping uid/gid.
3733 3733 * Always just let the server to the checking.
3734 3734 * If we provide the ability to remove basic priviledges
3735 3735 * to setattr (e.g. basic without chmod) then we will
3736 3736 * need to add a check here before calling the server.
3737 3737 */
3738 3738 error = nfs4setattr(vp, vap, flags, cr, NULL);
3739 3739
3740 3740 if (error == 0 && (vap->va_mask & AT_SIZE) && vap->va_size == 0)
3741 3741 vnevent_truncate(vp, ct);
3742 3742
3743 3743 return (error);
3744 3744 }
3745 3745
3746 3746 /*
3747 3747 * To replace the "guarded" version 3 setattr, we use two types of compound
3748 3748 * setattr requests:
3749 3749 * 1. The "normal" setattr, used when the size of the file isn't being
3750 3750 * changed - { Putfh <fh>; Setattr; Getattr }/
3751 3751 * 2. If the size is changed, precede Setattr with: Getattr; Verify
3752 3752 * with only ctime as the argument. If the server ctime differs from
3753 3753 * what is cached on the client, the verify will fail, but we would
3754 3754 * already have the ctime from the preceding getattr, so just set it
3755 3755 * and retry. Thus the compound here is - { Putfh <fh>; Getattr; Verify;
3756 3756 * Setattr; Getattr }.
3757 3757 *
3758 3758 * The vsecattr_t * input parameter will be non-NULL if ACLs are being set in
3759 3759 * this setattr and NULL if they are not.
3760 3760 */
3761 3761 static int
3762 3762 nfs4setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
3763 3763 vsecattr_t *vsap)
3764 3764 {
3765 3765 COMPOUND4args_clnt args;
3766 3766 COMPOUND4res_clnt res, *resp = NULL;
3767 3767 nfs4_ga_res_t *garp = NULL;
3768 3768 int numops = 3; /* { Putfh; Setattr; Getattr } */
3769 3769 nfs_argop4 argop[5];
3770 3770 int verify_argop = -1;
3771 3771 int setattr_argop = 1;
3772 3772 nfs_resop4 *resop;
3773 3773 vattr_t va;
3774 3774 rnode4_t *rp;
3775 3775 int doqueue = 1;
3776 3776 uint_t mask = vap->va_mask;
3777 3777 mode_t omode;
3778 3778 vsecattr_t *vsp;
3779 3779 timestruc_t ctime;
3780 3780 bool_t needrecov = FALSE;
3781 3781 nfs4_recov_state_t recov_state;
3782 3782 nfs4_stateid_types_t sid_types;
3783 3783 stateid4 stateid;
3784 3784 hrtime_t t;
3785 3785 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
3786 3786 servinfo4_t *svp;
3787 3787 bitmap4 supp_attrs;
3788 3788
3789 3789 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
3790 3790 rp = VTOR4(vp);
3791 3791 nfs4_init_stateid_types(&sid_types);
3792 3792
3793 3793 /*
3794 3794 * Only need to flush pages if there are any pages and
3795 3795 * if the file is marked as dirty in some fashion. The
3796 3796 * file must be flushed so that we can accurately
3797 3797 * determine the size of the file and the cached data
3798 3798 * after the SETATTR returns. A file is considered to
3799 3799 * be dirty if it is either marked with R4DIRTY, has
3800 3800 * outstanding i/o's active, or is mmap'd. In this
3801 3801 * last case, we can't tell whether there are dirty
3802 3802 * pages, so we flush just to be sure.
3803 3803 */
3804 3804 if (nfs4_has_pages(vp) &&
3805 3805 ((rp->r_flags & R4DIRTY) ||
3806 3806 rp->r_count > 0 ||
3807 3807 rp->r_mapcnt > 0)) {
3808 3808 ASSERT(vp->v_type != VCHR);
3809 3809 e.error = nfs4_putpage(vp, (offset_t)0, 0, 0, cr, NULL);
3810 3810 if (e.error && (e.error == ENOSPC || e.error == EDQUOT)) {
3811 3811 mutex_enter(&rp->r_statelock);
3812 3812 if (!rp->r_error)
3813 3813 rp->r_error = e.error;
3814 3814 mutex_exit(&rp->r_statelock);
3815 3815 }
3816 3816 }
3817 3817
3818 3818 if (mask & AT_SIZE) {
3819 3819 /*
3820 3820 * Verification setattr compound for non-deleg AT_SIZE:
3821 3821 * { Putfh; Getattr; Verify; Setattr; Getattr }
3822 3822 * Set ctime local here (outside the do_again label)
3823 3823 * so that subsequent retries (after failed VERIFY)
3824 3824 * will use ctime from GETATTR results (from failed
3825 3825 * verify compound) as VERIFY arg.
3826 3826 * If file has delegation, then VERIFY(time_metadata)
3827 3827 * is of little added value, so don't bother.
3828 3828 */
3829 3829 mutex_enter(&rp->r_statev4_lock);
3830 3830 if (rp->r_deleg_type == OPEN_DELEGATE_NONE ||
3831 3831 rp->r_deleg_return_pending) {
3832 3832 numops = 5;
3833 3833 ctime = rp->r_attr.va_ctime;
3834 3834 }
3835 3835 mutex_exit(&rp->r_statev4_lock);
3836 3836 }
3837 3837
3838 3838 recov_state.rs_flags = 0;
3839 3839 recov_state.rs_num_retry_despite_err = 0;
3840 3840
3841 3841 args.ctag = TAG_SETATTR;
3842 3842 do_again:
3843 3843 recov_retry:
3844 3844 setattr_argop = numops - 2;
3845 3845
3846 3846 args.array = argop;
3847 3847 args.array_len = numops;
3848 3848
3849 3849 e.error = nfs4_start_op(VTOMI4(vp), vp, NULL, &recov_state);
3850 3850 if (e.error)
3851 3851 return (e.error);
3852 3852
3853 3853
3854 3854 /* putfh target fh */
3855 3855 argop[0].argop = OP_CPUTFH;
3856 3856 argop[0].nfs_argop4_u.opcputfh.sfh = rp->r_fh;
3857 3857
3858 3858 if (numops == 5) {
3859 3859 /*
3860 3860 * We only care about the ctime, but need to get mtime
3861 3861 * and size for proper cache update.
3862 3862 */
3863 3863 /* getattr */
3864 3864 argop[1].argop = OP_GETATTR;
3865 3865 argop[1].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
3866 3866 argop[1].nfs_argop4_u.opgetattr.mi = VTOMI4(vp);
3867 3867
3868 3868 /* verify - set later in loop */
3869 3869 verify_argop = 2;
3870 3870 }
3871 3871
3872 3872 /* setattr */
3873 3873 svp = rp->r_server;
3874 3874 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
3875 3875 supp_attrs = svp->sv_supp_attrs;
3876 3876 nfs_rw_exit(&svp->sv_lock);
3877 3877
3878 3878 nfs4args_setattr(&argop[setattr_argop], vap, vsap, flags, rp, cr,
3879 3879 supp_attrs, &e.error, &sid_types);
3880 3880 stateid = argop[setattr_argop].nfs_argop4_u.opsetattr.stateid;
3881 3881 if (e.error) {
3882 3882 /* req time field(s) overflow - return immediately */
3883 3883 nfs4_end_op(VTOMI4(vp), vp, NULL, &recov_state, needrecov);
3884 3884 nfs4_fattr4_free(&argop[setattr_argop].nfs_argop4_u.
3885 3885 opsetattr.obj_attributes);
3886 3886 return (e.error);
3887 3887 }
3888 3888 omode = rp->r_attr.va_mode;
3889 3889
3890 3890 /* getattr */
3891 3891 argop[numops-1].argop = OP_GETATTR;
3892 3892 argop[numops-1].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
3893 3893 /*
3894 3894 * If we are setting the ACL (indicated only by vsap != NULL), request
3895 3895 * the ACL in this getattr. The ACL returned from this getattr will be
3896 3896 * used in updating the ACL cache.
3897 3897 */
3898 3898 if (vsap != NULL)
3899 3899 argop[numops-1].nfs_argop4_u.opgetattr.attr_request |=
3900 3900 FATTR4_ACL_MASK;
3901 3901 argop[numops-1].nfs_argop4_u.opgetattr.mi = VTOMI4(vp);
3902 3902
3903 3903 /*
3904 3904 * setattr iterates if the object size is set and the cached ctime
3905 3905 * does not match the file ctime. In that case, verify the ctime first.
3906 3906 */
3907 3907
3908 3908 do {
3909 3909 if (verify_argop != -1) {
3910 3910 /*
3911 3911 * Verify that the ctime match before doing setattr.
3912 3912 */
3913 3913 va.va_mask = AT_CTIME;
3914 3914 va.va_ctime = ctime;
3915 3915 svp = rp->r_server;
3916 3916 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
3917 3917 supp_attrs = svp->sv_supp_attrs;
3918 3918 nfs_rw_exit(&svp->sv_lock);
3919 3919 e.error = nfs4args_verify(&argop[verify_argop], &va,
3920 3920 OP_VERIFY, supp_attrs);
3921 3921 if (e.error) {
3922 3922 /* req time field(s) overflow - return */
3923 3923 nfs4_end_op(VTOMI4(vp), vp, NULL, &recov_state,
3924 3924 needrecov);
3925 3925 break;
3926 3926 }
3927 3927 }
3928 3928
3929 3929 doqueue = 1;
3930 3930
3931 3931 t = gethrtime();
3932 3932
3933 3933 rfs4call(VTOMI4(vp), &args, &res, cr, &doqueue, 0, &e);
3934 3934
3935 3935 /*
3936 3936 * Purge the access cache and ACL cache if changing either the
3937 3937 * owner of the file, the group owner, or the mode. These may
3938 3938 * change the access permissions of the file, so purge old
3939 3939 * information and start over again.
3940 3940 */
3941 3941 if (mask & (AT_UID | AT_GID | AT_MODE)) {
3942 3942 (void) nfs4_access_purge_rp(rp);
3943 3943 if (rp->r_secattr != NULL) {
3944 3944 mutex_enter(&rp->r_statelock);
3945 3945 vsp = rp->r_secattr;
3946 3946 rp->r_secattr = NULL;
3947 3947 mutex_exit(&rp->r_statelock);
3948 3948 if (vsp != NULL)
3949 3949 nfs4_acl_free_cache(vsp);
3950 3950 }
3951 3951 }
3952 3952
3953 3953 /*
3954 3954 * If res.array_len == numops, then everything succeeded,
3955 3955 * except for possibly the final getattr. If only the
3956 3956 * last getattr failed, give up, and don't try recovery.
3957 3957 */
3958 3958 if (res.array_len == numops) {
3959 3959 nfs4_end_op(VTOMI4(vp), vp, NULL, &recov_state,
3960 3960 needrecov);
3961 3961 if (! e.error)
3962 3962 resp = &res;
3963 3963 break;
3964 3964 }
3965 3965
3966 3966 /*
3967 3967 * if either rpc call failed or completely succeeded - done
3968 3968 */
3969 3969 needrecov = nfs4_needs_recovery(&e, FALSE, vp->v_vfsp);
3970 3970 if (e.error) {
3971 3971 PURGE_ATTRCACHE4(vp);
3972 3972 if (!needrecov) {
3973 3973 nfs4_end_op(VTOMI4(vp), vp, NULL, &recov_state,
3974 3974 needrecov);
3975 3975 break;
3976 3976 }
3977 3977 }
3978 3978
3979 3979 /*
3980 3980 * Do proper retry for OLD_STATEID outside of the normal
3981 3981 * recovery framework.
3982 3982 */
3983 3983 if (e.error == 0 && res.status == NFS4ERR_OLD_STATEID &&
3984 3984 sid_types.cur_sid_type != SPEC_SID &&
3985 3985 sid_types.cur_sid_type != NO_SID) {
3986 3986 nfs4_end_op(VTOMI4(vp), vp, NULL, &recov_state,
3987 3987 needrecov);
3988 3988 nfs4_save_stateid(&stateid, &sid_types);
3989 3989 nfs4_fattr4_free(&argop[setattr_argop].nfs_argop4_u.
3990 3990 opsetattr.obj_attributes);
3991 3991 if (verify_argop != -1) {
3992 3992 nfs4args_verify_free(&argop[verify_argop]);
3993 3993 verify_argop = -1;
3994 3994 }
3995 3995 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3996 3996 goto recov_retry;
3997 3997 }
3998 3998
3999 3999 if (needrecov) {
4000 4000 bool_t abort;
4001 4001
4002 4002 abort = nfs4_start_recovery(&e,
4003 4003 VTOMI4(vp), vp, NULL, NULL, NULL,
4004 4004 OP_SETATTR, NULL, NULL, NULL);
4005 4005 nfs4_end_op(VTOMI4(vp), vp, NULL, &recov_state,
4006 4006 needrecov);
4007 4007 /*
4008 4008 * Do not retry if we failed with OLD_STATEID using
4009 4009 * a special stateid. This is done to avoid looping
4010 4010 * with a broken server.
4011 4011 */
4012 4012 if (e.error == 0 && res.status == NFS4ERR_OLD_STATEID &&
4013 4013 (sid_types.cur_sid_type == SPEC_SID ||
4014 4014 sid_types.cur_sid_type == NO_SID))
4015 4015 abort = TRUE;
4016 4016 if (!e.error) {
4017 4017 if (res.status == NFS4ERR_BADOWNER)
4018 4018 nfs4_log_badowner(VTOMI4(vp),
4019 4019 OP_SETATTR);
4020 4020
4021 4021 e.error = geterrno4(res.status);
4022 4022 (void) xdr_free(xdr_COMPOUND4res_clnt,
4023 4023 (caddr_t)&res);
4024 4024 }
4025 4025 nfs4_fattr4_free(&argop[setattr_argop].nfs_argop4_u.
4026 4026 opsetattr.obj_attributes);
4027 4027 if (verify_argop != -1) {
4028 4028 nfs4args_verify_free(&argop[verify_argop]);
4029 4029 verify_argop = -1;
4030 4030 }
4031 4031 if (abort == FALSE) {
4032 4032 /*
4033 4033 * Need to retry all possible stateids in
4034 4034 * case the recovery error wasn't stateid
4035 4035 * related or the stateids have become
4036 4036 * stale (server reboot).
4037 4037 */
4038 4038 nfs4_init_stateid_types(&sid_types);
4039 4039 goto recov_retry;
4040 4040 }
4041 4041 return (e.error);
4042 4042 }
4043 4043
4044 4044 /*
4045 4045 * Need to call nfs4_end_op before nfs4getattr to
4046 4046 * avoid potential nfs4_start_op deadlock. See RFE
4047 4047 * 4777612. Calls to nfs4_invalidate_pages() and
4048 4048 * nfs4_purge_stale_fh() might also generate over the
4049 4049 * wire calls which my cause nfs4_start_op() deadlock.
4050 4050 */
4051 4051 nfs4_end_op(VTOMI4(vp), vp, NULL, &recov_state, needrecov);
4052 4052
4053 4053 /*
4054 4054 * Check to update lease.
4055 4055 */
4056 4056 resp = &res;
4057 4057 if (res.status == NFS4_OK) {
4058 4058 break;
4059 4059 }
4060 4060
4061 4061 /*
4062 4062 * Check if verify failed to see if try again
4063 4063 */
4064 4064 if ((verify_argop == -1) || (res.array_len != 3)) {
4065 4065 /*
4066 4066 * can't continue...
4067 4067 */
4068 4068 if (res.status == NFS4ERR_BADOWNER)
4069 4069 nfs4_log_badowner(VTOMI4(vp), OP_SETATTR);
4070 4070
4071 4071 e.error = geterrno4(res.status);
4072 4072 } else {
4073 4073 /*
4074 4074 * When the verify request fails, the client ctime is
4075 4075 * not in sync with the server. This is the same as
4076 4076 * the version 3 "not synchronized" error, and we
4077 4077 * handle it in a similar manner (XXX do we need to???).
4078 4078 * Use the ctime returned in the first getattr for
4079 4079 * the input to the next verify.
4080 4080 * If we couldn't get the attributes, then we give up
4081 4081 * because we can't complete the operation as required.
4082 4082 */
4083 4083 garp = &res.array[1].nfs_resop4_u.opgetattr.ga_res;
4084 4084 }
4085 4085 if (e.error) {
4086 4086 PURGE_ATTRCACHE4(vp);
4087 4087 nfs4_purge_stale_fh(e.error, vp, cr);
4088 4088 } else {
4089 4089 /*
4090 4090 * retry with a new verify value
4091 4091 */
4092 4092 ctime = garp->n4g_va.va_ctime;
4093 4093 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
4094 4094 resp = NULL;
4095 4095 }
4096 4096 if (!e.error) {
4097 4097 nfs4_fattr4_free(&argop[setattr_argop].nfs_argop4_u.
4098 4098 opsetattr.obj_attributes);
4099 4099 if (verify_argop != -1) {
4100 4100 nfs4args_verify_free(&argop[verify_argop]);
4101 4101 verify_argop = -1;
4102 4102 }
4103 4103 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
4104 4104 goto do_again;
4105 4105 }
4106 4106 } while (!e.error);
4107 4107
4108 4108 if (e.error) {
4109 4109 /*
4110 4110 * If we are here, rfs4call has an irrecoverable error - return
4111 4111 */
4112 4112 nfs4_fattr4_free(&argop[setattr_argop].nfs_argop4_u.
4113 4113 opsetattr.obj_attributes);
4114 4114 if (verify_argop != -1) {
4115 4115 nfs4args_verify_free(&argop[verify_argop]);
4116 4116 verify_argop = -1;
4117 4117 }
4118 4118 if (resp)
4119 4119 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)resp);
4120 4120 return (e.error);
4121 4121 }
4122 4122
4123 4123
4124 4124
4125 4125 /*
4126 4126 * If changing the size of the file, invalidate
4127 4127 * any local cached data which is no longer part
4128 4128 * of the file. We also possibly invalidate the
4129 4129 * last page in the file. We could use
4130 4130 * pvn_vpzero(), but this would mark the page as
4131 4131 * modified and require it to be written back to
4132 4132 * the server for no particularly good reason.
4133 4133 * This way, if we access it, then we bring it
4134 4134 * back in. A read should be cheaper than a
4135 4135 * write.
4136 4136 */
4137 4137 if (mask & AT_SIZE) {
4138 4138 nfs4_invalidate_pages(vp, (vap->va_size & PAGEMASK), cr);
4139 4139 }
4140 4140
4141 4141 /* either no error or one of the postop getattr failed */
4142 4142
4143 4143 /*
4144 4144 * XXX Perform a simplified version of wcc checking. Instead of
4145 4145 * have another getattr to get pre-op, just purge cache if
4146 4146 * any of the ops prior to and including the getattr failed.
4147 4147 * If the getattr succeeded then update the attrcache accordingly.
4148 4148 */
4149 4149
4150 4150 garp = NULL;
4151 4151 if (res.status == NFS4_OK) {
4152 4152 /*
4153 4153 * Last getattr
4154 4154 */
4155 4155 resop = &res.array[numops - 1];
4156 4156 garp = &resop->nfs_resop4_u.opgetattr.ga_res;
4157 4157 }
4158 4158 /*
4159 4159 * In certain cases, nfs4_update_attrcache() will purge the attrcache,
4160 4160 * rather than filling it. See the function itself for details.
4161 4161 */
4162 4162 e.error = nfs4_update_attrcache(res.status, garp, t, vp, cr);
4163 4163 if (garp != NULL) {
4164 4164 if (garp->n4g_resbmap & FATTR4_ACL_MASK) {
4165 4165 nfs4_acl_fill_cache(rp, &garp->n4g_vsa);
4166 4166 vs_ace4_destroy(&garp->n4g_vsa);
4167 4167 } else {
4168 4168 if (vsap != NULL) {
4169 4169 /*
4170 4170 * The ACL was supposed to be set and to be
4171 4171 * returned in the last getattr of this
4172 4172 * compound, but for some reason the getattr
4173 4173 * result doesn't contain the ACL. In this
4174 4174 * case, purge the ACL cache.
4175 4175 */
4176 4176 if (rp->r_secattr != NULL) {
4177 4177 mutex_enter(&rp->r_statelock);
4178 4178 vsp = rp->r_secattr;
4179 4179 rp->r_secattr = NULL;
4180 4180 mutex_exit(&rp->r_statelock);
4181 4181 if (vsp != NULL)
4182 4182 nfs4_acl_free_cache(vsp);
4183 4183 }
4184 4184 }
4185 4185 }
4186 4186 }
4187 4187
4188 4188 if (res.status == NFS4_OK && (mask & AT_SIZE)) {
4189 4189 /*
4190 4190 * Set the size, rather than relying on getting it updated
4191 4191 * via a GETATTR. With delegations the client tries to
4192 4192 * suppress GETATTR calls.
4193 4193 */
4194 4194 mutex_enter(&rp->r_statelock);
4195 4195 rp->r_size = vap->va_size;
4196 4196 mutex_exit(&rp->r_statelock);
4197 4197 }
4198 4198
4199 4199 /*
4200 4200 * Can free up request args and res
4201 4201 */
4202 4202 nfs4_fattr4_free(&argop[setattr_argop].nfs_argop4_u.
4203 4203 opsetattr.obj_attributes);
4204 4204 if (verify_argop != -1) {
4205 4205 nfs4args_verify_free(&argop[verify_argop]);
4206 4206 verify_argop = -1;
4207 4207 }
4208 4208 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
4209 4209
4210 4210 /*
4211 4211 * Some servers will change the mode to clear the setuid
4212 4212 * and setgid bits when changing the uid or gid. The
4213 4213 * client needs to compensate appropriately.
4214 4214 */
4215 4215 if (mask & (AT_UID | AT_GID)) {
4216 4216 int terror, do_setattr;
4217 4217
4218 4218 do_setattr = 0;
4219 4219 va.va_mask = AT_MODE;
4220 4220 terror = nfs4getattr(vp, &va, cr);
4221 4221 if (!terror &&
4222 4222 (((mask & AT_MODE) && va.va_mode != vap->va_mode) ||
4223 4223 (!(mask & AT_MODE) && va.va_mode != omode))) {
4224 4224 va.va_mask = AT_MODE;
4225 4225 if (mask & AT_MODE) {
4226 4226 /*
4227 4227 * We asked the mode to be changed and what
4228 4228 * we just got from the server in getattr is
4229 4229 * not what we wanted it to be, so set it now.
4230 4230 */
4231 4231 va.va_mode = vap->va_mode;
4232 4232 do_setattr = 1;
4233 4233 } else {
4234 4234 /*
4235 4235 * We did not ask the mode to be changed,
4236 4236 * Check to see that the server just cleared
4237 4237 * I_SUID and I_GUID from it. If not then
4238 4238 * set mode to omode with UID/GID cleared.
4239 4239 */
4240 4240 if (nfs4_compare_modes(va.va_mode, omode)) {
4241 4241 omode &= ~(S_ISUID|S_ISGID);
4242 4242 va.va_mode = omode;
4243 4243 do_setattr = 1;
4244 4244 }
4245 4245 }
4246 4246
4247 4247 if (do_setattr)
4248 4248 (void) nfs4setattr(vp, &va, 0, cr, NULL);
4249 4249 }
4250 4250 }
4251 4251
4252 4252 return (e.error);
4253 4253 }
4254 4254
4255 4255 /* ARGSUSED */
4256 4256 static int
4257 4257 nfs4_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
4258 4258 {
4259 4259 COMPOUND4args_clnt args;
4260 4260 COMPOUND4res_clnt res;
4261 4261 int doqueue;
4262 4262 uint32_t acc, resacc, argacc;
4263 4263 rnode4_t *rp;
4264 4264 cred_t *cred, *ncr, *ncrfree = NULL;
4265 4265 nfs4_access_type_t cacc;
4266 4266 int num_ops;
4267 4267 nfs_argop4 argop[3];
4268 4268 nfs_resop4 *resop;
4269 4269 bool_t needrecov = FALSE, do_getattr;
4270 4270 nfs4_recov_state_t recov_state;
4271 4271 int rpc_error;
4272 4272 hrtime_t t;
4273 4273 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
4274 4274 mntinfo4_t *mi = VTOMI4(vp);
4275 4275
4276 4276 if (nfs_zone() != mi->mi_zone)
4277 4277 return (EIO);
4278 4278
4279 4279 acc = 0;
4280 4280 if (mode & VREAD)
4281 4281 acc |= ACCESS4_READ;
4282 4282 if (mode & VWRITE) {
4283 4283 if ((vp->v_vfsp->vfs_flag & VFS_RDONLY) && !ISVDEV(vp->v_type))
4284 4284 return (EROFS);
4285 4285 if (vp->v_type == VDIR)
4286 4286 acc |= ACCESS4_DELETE;
4287 4287 acc |= ACCESS4_MODIFY | ACCESS4_EXTEND;
4288 4288 }
4289 4289 if (mode & VEXEC) {
4290 4290 if (vp->v_type == VDIR)
4291 4291 acc |= ACCESS4_LOOKUP;
4292 4292 else
4293 4293 acc |= ACCESS4_EXECUTE;
4294 4294 }
4295 4295
4296 4296 if (VTOR4(vp)->r_acache != NULL) {
4297 4297 e.error = nfs4_validate_caches(vp, cr);
4298 4298 if (e.error)
4299 4299 return (e.error);
4300 4300 }
4301 4301
4302 4302 rp = VTOR4(vp);
4303 4303 if (vp->v_type == VDIR)
4304 4304 argacc = ACCESS4_READ | ACCESS4_DELETE | ACCESS4_MODIFY |
4305 4305 ACCESS4_EXTEND | ACCESS4_LOOKUP;
4306 4306 else
4307 4307 argacc = ACCESS4_READ | ACCESS4_MODIFY | ACCESS4_EXTEND |
4308 4308 ACCESS4_EXECUTE;
4309 4309 recov_state.rs_flags = 0;
4310 4310 recov_state.rs_num_retry_despite_err = 0;
4311 4311
4312 4312 cred = cr;
4313 4313 /*
4314 4314 * ncr and ncrfree both initially
4315 4315 * point to the memory area returned
4316 4316 * by crnetadjust();
4317 4317 * ncrfree not NULL when exiting means
4318 4318 * that we need to release it
4319 4319 */
4320 4320 ncr = crnetadjust(cred);
4321 4321 ncrfree = ncr;
4322 4322
4323 4323 tryagain:
4324 4324 cacc = nfs4_access_check(rp, acc, cred);
4325 4325 if (cacc == NFS4_ACCESS_ALLOWED) {
4326 4326 if (ncrfree != NULL)
4327 4327 crfree(ncrfree);
4328 4328 return (0);
4329 4329 }
4330 4330 if (cacc == NFS4_ACCESS_DENIED) {
4331 4331 /*
4332 4332 * If the cred can be adjusted, try again
4333 4333 * with the new cred.
4334 4334 */
4335 4335 if (ncr != NULL) {
4336 4336 cred = ncr;
4337 4337 ncr = NULL;
4338 4338 goto tryagain;
4339 4339 }
4340 4340 if (ncrfree != NULL)
4341 4341 crfree(ncrfree);
4342 4342 return (EACCES);
4343 4343 }
4344 4344
4345 4345 recov_retry:
4346 4346 /*
4347 4347 * Don't take with r_statev4_lock here. r_deleg_type could
4348 4348 * change as soon as lock is released. Since it is an int,
4349 4349 * there is no atomicity issue.
4350 4350 */
4351 4351 do_getattr = (rp->r_deleg_type == OPEN_DELEGATE_NONE);
4352 4352 num_ops = do_getattr ? 3 : 2;
4353 4353
4354 4354 args.ctag = TAG_ACCESS;
4355 4355
4356 4356 args.array_len = num_ops;
4357 4357 args.array = argop;
4358 4358
4359 4359 if (e.error = nfs4_start_fop(mi, vp, NULL, OH_ACCESS,
4360 4360 &recov_state, NULL)) {
4361 4361 if (ncrfree != NULL)
4362 4362 crfree(ncrfree);
4363 4363 return (e.error);
4364 4364 }
4365 4365
4366 4366 /* putfh target fh */
4367 4367 argop[0].argop = OP_CPUTFH;
4368 4368 argop[0].nfs_argop4_u.opcputfh.sfh = VTOR4(vp)->r_fh;
4369 4369
4370 4370 /* access */
4371 4371 argop[1].argop = OP_ACCESS;
4372 4372 argop[1].nfs_argop4_u.opaccess.access = argacc;
4373 4373
4374 4374 /* getattr */
4375 4375 if (do_getattr) {
4376 4376 argop[2].argop = OP_GETATTR;
4377 4377 argop[2].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
4378 4378 argop[2].nfs_argop4_u.opgetattr.mi = mi;
4379 4379 }
4380 4380
4381 4381 NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE,
4382 4382 "nfs4_access: %s call, rp %s", needrecov ? "recov" : "first",
4383 4383 rnode4info(VTOR4(vp))));
4384 4384
4385 4385 doqueue = 1;
4386 4386 t = gethrtime();
4387 4387 rfs4call(VTOMI4(vp), &args, &res, cred, &doqueue, 0, &e);
4388 4388 rpc_error = e.error;
4389 4389
4390 4390 needrecov = nfs4_needs_recovery(&e, FALSE, vp->v_vfsp);
4391 4391 if (needrecov) {
4392 4392 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
4393 4393 "nfs4_access: initiating recovery\n"));
4394 4394
4395 4395 if (nfs4_start_recovery(&e, VTOMI4(vp), vp, NULL, NULL,
4396 4396 NULL, OP_ACCESS, NULL, NULL, NULL) == FALSE) {
4397 4397 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_ACCESS,
4398 4398 &recov_state, needrecov);
4399 4399 if (!e.error)
4400 4400 (void) xdr_free(xdr_COMPOUND4res_clnt,
4401 4401 (caddr_t)&res);
4402 4402 goto recov_retry;
4403 4403 }
4404 4404 }
4405 4405 nfs4_end_fop(mi, vp, NULL, OH_ACCESS, &recov_state, needrecov);
4406 4406
4407 4407 if (e.error)
4408 4408 goto out;
4409 4409
4410 4410 if (res.status) {
4411 4411 e.error = geterrno4(res.status);
4412 4412 /*
4413 4413 * This might generate over the wire calls throught
4414 4414 * nfs4_invalidate_pages. Hence we need to call nfs4_end_op()
4415 4415 * here to avoid a deadlock.
4416 4416 */
4417 4417 nfs4_purge_stale_fh(e.error, vp, cr);
4418 4418 goto out;
4419 4419 }
4420 4420 resop = &res.array[1]; /* access res */
4421 4421
4422 4422 resacc = resop->nfs_resop4_u.opaccess.access;
4423 4423
4424 4424 if (do_getattr) {
4425 4425 resop++; /* getattr res */
4426 4426 nfs4_attr_cache(vp, &resop->nfs_resop4_u.opgetattr.ga_res,
4427 4427 t, cr, FALSE, NULL);
4428 4428 }
4429 4429
4430 4430 if (!e.error) {
4431 4431 nfs4_access_cache(rp, argacc, resacc, cred);
4432 4432 /*
4433 4433 * we just cached results with cred; if cred is the
4434 4434 * adjusted credentials from crnetadjust, we do not want
4435 4435 * to release them before exiting: hence setting ncrfree
4436 4436 * to NULL
4437 4437 */
4438 4438 if (cred != cr)
4439 4439 ncrfree = NULL;
4440 4440 /* XXX check the supported bits too? */
4441 4441 if ((acc & resacc) != acc) {
4442 4442 /*
4443 4443 * The following code implements the semantic
4444 4444 * that a setuid root program has *at least* the
4445 4445 * permissions of the user that is running the
4446 4446 * program. See rfs3call() for more portions
4447 4447 * of the implementation of this functionality.
4448 4448 */
4449 4449 /* XXX-LP */
4450 4450 if (ncr != NULL) {
4451 4451 (void) xdr_free(xdr_COMPOUND4res_clnt,
4452 4452 (caddr_t)&res);
4453 4453 cred = ncr;
4454 4454 ncr = NULL;
4455 4455 goto tryagain;
4456 4456 }
4457 4457 e.error = EACCES;
4458 4458 }
4459 4459 }
4460 4460
4461 4461 out:
4462 4462 if (!rpc_error)
4463 4463 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
4464 4464
4465 4465 if (ncrfree != NULL)
4466 4466 crfree(ncrfree);
4467 4467
4468 4468 return (e.error);
4469 4469 }
4470 4470
4471 4471 /* ARGSUSED */
4472 4472 static int
4473 4473 nfs4_readlink(vnode_t *vp, struct uio *uiop, cred_t *cr, caller_context_t *ct)
4474 4474 {
4475 4475 COMPOUND4args_clnt args;
4476 4476 COMPOUND4res_clnt res;
4477 4477 int doqueue;
4478 4478 rnode4_t *rp;
4479 4479 nfs_argop4 argop[3];
4480 4480 nfs_resop4 *resop;
4481 4481 READLINK4res *lr_res;
4482 4482 nfs4_ga_res_t *garp;
4483 4483 uint_t len;
4484 4484 char *linkdata;
4485 4485 bool_t needrecov = FALSE;
4486 4486 nfs4_recov_state_t recov_state;
4487 4487 hrtime_t t;
4488 4488 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
4489 4489
4490 4490 if (nfs_zone() != VTOMI4(vp)->mi_zone)
4491 4491 return (EIO);
4492 4492 /*
4493 4493 * Can't readlink anything other than a symbolic link.
4494 4494 */
4495 4495 if (vp->v_type != VLNK)
4496 4496 return (EINVAL);
4497 4497
4498 4498 rp = VTOR4(vp);
4499 4499 if (nfs4_do_symlink_cache && rp->r_symlink.contents != NULL) {
4500 4500 e.error = nfs4_validate_caches(vp, cr);
4501 4501 if (e.error)
4502 4502 return (e.error);
4503 4503 mutex_enter(&rp->r_statelock);
4504 4504 if (rp->r_symlink.contents != NULL) {
4505 4505 e.error = uiomove(rp->r_symlink.contents,
4506 4506 rp->r_symlink.len, UIO_READ, uiop);
4507 4507 mutex_exit(&rp->r_statelock);
4508 4508 return (e.error);
4509 4509 }
4510 4510 mutex_exit(&rp->r_statelock);
4511 4511 }
4512 4512 recov_state.rs_flags = 0;
4513 4513 recov_state.rs_num_retry_despite_err = 0;
4514 4514
4515 4515 recov_retry:
4516 4516 args.array_len = 3;
4517 4517 args.array = argop;
4518 4518 args.ctag = TAG_READLINK;
4519 4519
4520 4520 e.error = nfs4_start_op(VTOMI4(vp), vp, NULL, &recov_state);
4521 4521 if (e.error) {
4522 4522 return (e.error);
4523 4523 }
4524 4524
4525 4525 /* 0. putfh symlink fh */
4526 4526 argop[0].argop = OP_CPUTFH;
4527 4527 argop[0].nfs_argop4_u.opcputfh.sfh = VTOR4(vp)->r_fh;
4528 4528
4529 4529 /* 1. readlink */
4530 4530 argop[1].argop = OP_READLINK;
4531 4531
4532 4532 /* 2. getattr */
4533 4533 argop[2].argop = OP_GETATTR;
4534 4534 argop[2].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
4535 4535 argop[2].nfs_argop4_u.opgetattr.mi = VTOMI4(vp);
4536 4536
4537 4537 doqueue = 1;
4538 4538
4539 4539 NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE,
4540 4540 "nfs4_readlink: %s call, rp %s", needrecov ? "recov" : "first",
4541 4541 rnode4info(VTOR4(vp))));
4542 4542
4543 4543 t = gethrtime();
4544 4544
4545 4545 rfs4call(VTOMI4(vp), &args, &res, cr, &doqueue, 0, &e);
4546 4546
4547 4547 needrecov = nfs4_needs_recovery(&e, FALSE, vp->v_vfsp);
4548 4548 if (needrecov) {
4549 4549 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
4550 4550 "nfs4_readlink: initiating recovery\n"));
4551 4551
4552 4552 if (nfs4_start_recovery(&e, VTOMI4(vp), vp, NULL, NULL,
4553 4553 NULL, OP_READLINK, NULL, NULL, NULL) == FALSE) {
4554 4554 if (!e.error)
4555 4555 (void) xdr_free(xdr_COMPOUND4res_clnt,
4556 4556 (caddr_t)&res);
4557 4557
4558 4558 nfs4_end_op(VTOMI4(vp), vp, NULL, &recov_state,
4559 4559 needrecov);
4560 4560 goto recov_retry;
4561 4561 }
4562 4562 }
4563 4563
4564 4564 nfs4_end_op(VTOMI4(vp), vp, NULL, &recov_state, needrecov);
4565 4565
4566 4566 if (e.error)
4567 4567 return (e.error);
4568 4568
4569 4569 /*
4570 4570 * There is an path in the code below which calls
4571 4571 * nfs4_purge_stale_fh(), which may generate otw calls through
4572 4572 * nfs4_invalidate_pages. Hence we need to call nfs4_end_op()
4573 4573 * here to avoid nfs4_start_op() deadlock.
4574 4574 */
4575 4575
4576 4576 if (res.status && (res.array_len < args.array_len)) {
4577 4577 /*
4578 4578 * either Putfh or Link failed
4579 4579 */
4580 4580 e.error = geterrno4(res.status);
4581 4581 nfs4_purge_stale_fh(e.error, vp, cr);
4582 4582 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
4583 4583 return (e.error);
4584 4584 }
4585 4585
4586 4586 resop = &res.array[1]; /* readlink res */
4587 4587 lr_res = &resop->nfs_resop4_u.opreadlink;
4588 4588
4589 4589 /*
4590 4590 * treat symlink names as data
4591 4591 */
4592 4592 linkdata = utf8_to_str((utf8string *)&lr_res->link, &len, NULL);
4593 4593 if (linkdata != NULL) {
4594 4594 int uio_len = len - 1;
4595 4595 /* len includes null byte, which we won't uiomove */
4596 4596 e.error = uiomove(linkdata, uio_len, UIO_READ, uiop);
4597 4597 if (nfs4_do_symlink_cache && rp->r_symlink.contents == NULL) {
4598 4598 mutex_enter(&rp->r_statelock);
4599 4599 if (rp->r_symlink.contents == NULL) {
4600 4600 rp->r_symlink.contents = linkdata;
4601 4601 rp->r_symlink.len = uio_len;
4602 4602 rp->r_symlink.size = len;
4603 4603 mutex_exit(&rp->r_statelock);
4604 4604 } else {
4605 4605 mutex_exit(&rp->r_statelock);
4606 4606 kmem_free(linkdata, len);
4607 4607 }
4608 4608 } else {
4609 4609 kmem_free(linkdata, len);
4610 4610 }
4611 4611 }
4612 4612 if (res.status == NFS4_OK) {
4613 4613 resop++; /* getattr res */
4614 4614 garp = &resop->nfs_resop4_u.opgetattr.ga_res;
4615 4615 }
4616 4616 e.error = nfs4_update_attrcache(res.status, garp, t, vp, cr);
4617 4617
4618 4618 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
4619 4619
4620 4620 /*
4621 4621 * The over the wire error for attempting to readlink something
4622 4622 * other than a symbolic link is ENXIO. However, we need to
4623 4623 * return EINVAL instead of ENXIO, so we map it here.
4624 4624 */
4625 4625 return (e.error == ENXIO ? EINVAL : e.error);
4626 4626 }
4627 4627
4628 4628 /*
4629 4629 * Flush local dirty pages to stable storage on the server.
4630 4630 *
4631 4631 * If FNODSYNC is specified, then there is nothing to do because
4632 4632 * metadata changes are not cached on the client before being
4633 4633 * sent to the server.
4634 4634 */
4635 4635 /* ARGSUSED */
4636 4636 static int
4637 4637 nfs4_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
4638 4638 {
4639 4639 int error;
4640 4640
4641 4641 if ((syncflag & FNODSYNC) || IS_SWAPVP(vp))
4642 4642 return (0);
4643 4643 if (nfs_zone() != VTOMI4(vp)->mi_zone)
4644 4644 return (EIO);
4645 4645 error = nfs4_putpage_commit(vp, (offset_t)0, 0, cr);
4646 4646 if (!error)
4647 4647 error = VTOR4(vp)->r_error;
4648 4648 return (error);
4649 4649 }
4650 4650
4651 4651 /*
4652 4652 * Weirdness: if the file was removed or the target of a rename
4653 4653 * operation while it was open, it got renamed instead. Here we
4654 4654 * remove the renamed file.
4655 4655 */
4656 4656 /* ARGSUSED */
4657 4657 void
4658 4658 nfs4_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4659 4659 {
4660 4660 rnode4_t *rp;
4661 4661
4662 4662 ASSERT(vp != DNLC_NO_VNODE);
4663 4663
4664 4664 rp = VTOR4(vp);
4665 4665
4666 4666 if (IS_SHADOW(vp, rp)) {
4667 4667 sv_inactive(vp);
4668 4668 return;
4669 4669 }
4670 4670
4671 4671 /*
4672 4672 * If this is coming from the wrong zone, we let someone in the right
4673 4673 * zone take care of it asynchronously. We can get here due to
4674 4674 * VN_RELE() being called from pageout() or fsflush(). This call may
4675 4675 * potentially turn into an expensive no-op if, for instance, v_count
4676 4676 * gets incremented in the meantime, but it's still correct.
4677 4677 */
4678 4678 if (nfs_zone() != VTOMI4(vp)->mi_zone) {
4679 4679 nfs4_async_inactive(vp, cr);
4680 4680 return;
4681 4681 }
4682 4682
4683 4683 /*
4684 4684 * Some of the cleanup steps might require over-the-wire
4685 4685 * operations. Since VOP_INACTIVE can get called as a result of
4686 4686 * other over-the-wire operations (e.g., an attribute cache update
4687 4687 * can lead to a DNLC purge), doing those steps now would lead to a
4688 4688 * nested call to the recovery framework, which can deadlock. So
4689 4689 * do any over-the-wire cleanups asynchronously, in a separate
4690 4690 * thread.
4691 4691 */
4692 4692
4693 4693 mutex_enter(&rp->r_os_lock);
4694 4694 mutex_enter(&rp->r_statelock);
4695 4695 mutex_enter(&rp->r_statev4_lock);
4696 4696
4697 4697 if (vp->v_type == VREG && list_head(&rp->r_open_streams) != NULL) {
4698 4698 mutex_exit(&rp->r_statev4_lock);
4699 4699 mutex_exit(&rp->r_statelock);
4700 4700 mutex_exit(&rp->r_os_lock);
4701 4701 nfs4_async_inactive(vp, cr);
4702 4702 return;
4703 4703 }
4704 4704
4705 4705 if (rp->r_deleg_type == OPEN_DELEGATE_READ ||
4706 4706 rp->r_deleg_type == OPEN_DELEGATE_WRITE) {
4707 4707 mutex_exit(&rp->r_statev4_lock);
4708 4708 mutex_exit(&rp->r_statelock);
4709 4709 mutex_exit(&rp->r_os_lock);
4710 4710 nfs4_async_inactive(vp, cr);
4711 4711 return;
4712 4712 }
4713 4713
4714 4714 if (rp->r_unldvp != NULL) {
4715 4715 mutex_exit(&rp->r_statev4_lock);
4716 4716 mutex_exit(&rp->r_statelock);
4717 4717 mutex_exit(&rp->r_os_lock);
4718 4718 nfs4_async_inactive(vp, cr);
4719 4719 return;
4720 4720 }
4721 4721 mutex_exit(&rp->r_statev4_lock);
4722 4722 mutex_exit(&rp->r_statelock);
4723 4723 mutex_exit(&rp->r_os_lock);
4724 4724
4725 4725 rp4_addfree(rp, cr);
4726 4726 }
4727 4727
4728 4728 /*
4729 4729 * nfs4_inactive_otw - nfs4_inactive, plus over-the-wire calls to free up
4730 4730 * various bits of state. The caller must not refer to vp after this call.
4731 4731 */
4732 4732
4733 4733 void
4734 4734 nfs4_inactive_otw(vnode_t *vp, cred_t *cr)
4735 4735 {
4736 4736 rnode4_t *rp = VTOR4(vp);
4737 4737 nfs4_recov_state_t recov_state;
4738 4738 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
4739 4739 vnode_t *unldvp;
4740 4740 char *unlname;
4741 4741 cred_t *unlcred;
4742 4742 COMPOUND4args_clnt args;
4743 4743 COMPOUND4res_clnt res, *resp;
4744 4744 nfs_argop4 argop[2];
4745 4745 int doqueue;
4746 4746 #ifdef DEBUG
4747 4747 char *name;
4748 4748 #endif
4749 4749
4750 4750 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
4751 4751 ASSERT(!IS_SHADOW(vp, rp));
4752 4752
4753 4753 #ifdef DEBUG
4754 4754 name = fn_name(VTOSV(vp)->sv_name);
4755 4755 NFS4_DEBUG(nfs4_client_inactive_debug, (CE_NOTE, "nfs4_inactive_otw: "
4756 4756 "release vnode %s", name));
4757 4757 kmem_free(name, MAXNAMELEN);
4758 4758 #endif
4759 4759
4760 4760 if (vp->v_type == VREG) {
4761 4761 bool_t recov_failed = FALSE;
4762 4762
4763 4763 e.error = nfs4close_all(vp, cr);
4764 4764 if (e.error) {
4765 4765 /* Check to see if recovery failed */
4766 4766 mutex_enter(&(VTOMI4(vp)->mi_lock));
4767 4767 if (VTOMI4(vp)->mi_flags & MI4_RECOV_FAIL)
4768 4768 recov_failed = TRUE;
4769 4769 mutex_exit(&(VTOMI4(vp)->mi_lock));
4770 4770 if (!recov_failed) {
4771 4771 mutex_enter(&rp->r_statelock);
4772 4772 if (rp->r_flags & R4RECOVERR)
4773 4773 recov_failed = TRUE;
4774 4774 mutex_exit(&rp->r_statelock);
4775 4775 }
4776 4776 if (recov_failed) {
4777 4777 NFS4_DEBUG(nfs4_client_recov_debug,
4778 4778 (CE_NOTE, "nfs4_inactive_otw: "
4779 4779 "close failed (recovery failure)"));
4780 4780 }
4781 4781 }
4782 4782 }
4783 4783
4784 4784 redo:
4785 4785 if (rp->r_unldvp == NULL) {
4786 4786 rp4_addfree(rp, cr);
4787 4787 return;
4788 4788 }
4789 4789
4790 4790 /*
4791 4791 * Save the vnode pointer for the directory where the
4792 4792 * unlinked-open file got renamed, then set it to NULL
4793 4793 * to prevent another thread from getting here before
4794 4794 * we're done with the remove. While we have the
4795 4795 * statelock, make local copies of the pertinent rnode
4796 4796 * fields. If we weren't to do this in an atomic way, the
4797 4797 * the unl* fields could become inconsistent with respect
4798 4798 * to each other due to a race condition between this
4799 4799 * code and nfs_remove(). See bug report 1034328.
4800 4800 */
4801 4801 mutex_enter(&rp->r_statelock);
4802 4802 if (rp->r_unldvp == NULL) {
4803 4803 mutex_exit(&rp->r_statelock);
4804 4804 rp4_addfree(rp, cr);
4805 4805 return;
4806 4806 }
4807 4807
4808 4808 unldvp = rp->r_unldvp;
4809 4809 rp->r_unldvp = NULL;
4810 4810 unlname = rp->r_unlname;
4811 4811 rp->r_unlname = NULL;
4812 4812 unlcred = rp->r_unlcred;
4813 4813 rp->r_unlcred = NULL;
4814 4814 mutex_exit(&rp->r_statelock);
4815 4815
4816 4816 /*
4817 4817 * If there are any dirty pages left, then flush
4818 4818 * them. This is unfortunate because they just
4819 4819 * may get thrown away during the remove operation,
4820 4820 * but we have to do this for correctness.
4821 4821 */
4822 4822 if (nfs4_has_pages(vp) &&
4823 4823 ((rp->r_flags & R4DIRTY) || rp->r_count > 0)) {
4824 4824 ASSERT(vp->v_type != VCHR);
4825 4825 e.error = nfs4_putpage(vp, (u_offset_t)0, 0, 0, cr, NULL);
4826 4826 if (e.error) {
4827 4827 mutex_enter(&rp->r_statelock);
4828 4828 if (!rp->r_error)
4829 4829 rp->r_error = e.error;
4830 4830 mutex_exit(&rp->r_statelock);
4831 4831 }
4832 4832 }
4833 4833
4834 4834 recov_state.rs_flags = 0;
4835 4835 recov_state.rs_num_retry_despite_err = 0;
4836 4836 recov_retry_remove:
4837 4837 /*
4838 4838 * Do the remove operation on the renamed file
4839 4839 */
4840 4840 args.ctag = TAG_INACTIVE;
4841 4841
4842 4842 /*
4843 4843 * Remove ops: putfh dir; remove
4844 4844 */
4845 4845 args.array_len = 2;
4846 4846 args.array = argop;
4847 4847
4848 4848 e.error = nfs4_start_op(VTOMI4(unldvp), unldvp, NULL, &recov_state);
4849 4849 if (e.error) {
4850 4850 kmem_free(unlname, MAXNAMELEN);
4851 4851 crfree(unlcred);
4852 4852 VN_RELE(unldvp);
4853 4853 /*
4854 4854 * Try again; this time around r_unldvp will be NULL, so we'll
4855 4855 * just call rp4_addfree() and return.
4856 4856 */
4857 4857 goto redo;
4858 4858 }
4859 4859
4860 4860 /* putfh directory */
4861 4861 argop[0].argop = OP_CPUTFH;
4862 4862 argop[0].nfs_argop4_u.opcputfh.sfh = VTOR4(unldvp)->r_fh;
4863 4863
4864 4864 /* remove */
4865 4865 argop[1].argop = OP_CREMOVE;
4866 4866 argop[1].nfs_argop4_u.opcremove.ctarget = unlname;
4867 4867
4868 4868 doqueue = 1;
4869 4869 resp = &res;
4870 4870
4871 4871 #if 0 /* notyet */
4872 4872 /*
4873 4873 * Can't do this yet. We may be being called from
4874 4874 * dnlc_purge_XXX while that routine is holding a
4875 4875 * mutex lock to the nc_rele list. The calls to
4876 4876 * nfs3_cache_wcc_data may result in calls to
4877 4877 * dnlc_purge_XXX. This will result in a deadlock.
4878 4878 */
4879 4879 rfs4call(VTOMI4(unldvp), &args, &res, unlcred, &doqueue, 0, &e);
4880 4880 if (e.error) {
4881 4881 PURGE_ATTRCACHE4(unldvp);
4882 4882 resp = NULL;
4883 4883 } else if (res.status) {
4884 4884 e.error = geterrno4(res.status);
4885 4885 PURGE_ATTRCACHE4(unldvp);
4886 4886 /*
4887 4887 * This code is inactive right now
4888 4888 * but if made active there should
4889 4889 * be a nfs4_end_op() call before
4890 4890 * nfs4_purge_stale_fh to avoid start_op()
4891 4891 * deadlock. See BugId: 4948726
4892 4892 */
4893 4893 nfs4_purge_stale_fh(error, unldvp, cr);
4894 4894 } else {
4895 4895 nfs_resop4 *resop;
4896 4896 REMOVE4res *rm_res;
4897 4897
4898 4898 resop = &res.array[1];
4899 4899 rm_res = &resop->nfs_resop4_u.opremove;
4900 4900 /*
4901 4901 * Update directory cache attribute,
4902 4902 * readdir and dnlc caches.
4903 4903 */
4904 4904 nfs4_update_dircaches(&rm_res->cinfo, unldvp, NULL, NULL, NULL);
4905 4905 }
4906 4906 #else
4907 4907 rfs4call(VTOMI4(unldvp), &args, &res, unlcred, &doqueue, 0, &e);
4908 4908
4909 4909 PURGE_ATTRCACHE4(unldvp);
4910 4910 #endif
4911 4911
4912 4912 if (nfs4_needs_recovery(&e, FALSE, unldvp->v_vfsp)) {
4913 4913 if (nfs4_start_recovery(&e, VTOMI4(unldvp), unldvp, NULL,
4914 4914 NULL, NULL, OP_REMOVE, NULL, NULL, NULL) == FALSE) {
4915 4915 if (!e.error)
4916 4916 (void) xdr_free(xdr_COMPOUND4res_clnt,
4917 4917 (caddr_t)&res);
4918 4918 nfs4_end_op(VTOMI4(unldvp), unldvp, NULL,
4919 4919 &recov_state, TRUE);
4920 4920 goto recov_retry_remove;
4921 4921 }
4922 4922 }
4923 4923 nfs4_end_op(VTOMI4(unldvp), unldvp, NULL, &recov_state, FALSE);
4924 4924
4925 4925 /*
4926 4926 * Release stuff held for the remove
4927 4927 */
4928 4928 VN_RELE(unldvp);
4929 4929 if (!e.error && resp)
4930 4930 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)resp);
4931 4931
4932 4932 kmem_free(unlname, MAXNAMELEN);
4933 4933 crfree(unlcred);
4934 4934 goto redo;
4935 4935 }
4936 4936
4937 4937 /*
4938 4938 * Remote file system operations having to do with directory manipulation.
4939 4939 */
4940 4940 /* ARGSUSED3 */
4941 4941 int
4942 4942 nfs4_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
4943 4943 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
4944 4944 int *direntflags, pathname_t *realpnp)
4945 4945 {
4946 4946 int error;
4947 4947 vnode_t *vp, *avp = NULL;
4948 4948 rnode4_t *drp;
4949 4949
4950 4950 *vpp = NULL;
4951 4951 if (nfs_zone() != VTOMI4(dvp)->mi_zone)
4952 4952 return (EPERM);
4953 4953 /*
4954 4954 * if LOOKUP_XATTR, must replace dvp (object) with
4955 4955 * object's attrdir before continuing with lookup
4956 4956 */
4957 4957 if (flags & LOOKUP_XATTR) {
4958 4958 error = nfs4lookup_xattr(dvp, nm, &avp, flags, cr);
4959 4959 if (error)
4960 4960 return (error);
4961 4961
4962 4962 dvp = avp;
4963 4963
4964 4964 /*
4965 4965 * If lookup is for "", just return dvp now. The attrdir
4966 4966 * has already been activated (from nfs4lookup_xattr), and
4967 4967 * the caller will RELE the original dvp -- not
4968 4968 * the attrdir. So, set vpp and return.
4969 4969 * Currently, when the LOOKUP_XATTR flag is
4970 4970 * passed to VOP_LOOKUP, the name is always empty, and
4971 4971 * shortcircuiting here avoids 3 unneeded lock/unlock
4972 4972 * pairs.
4973 4973 *
4974 4974 * If a non-empty name was provided, then it is the
4975 4975 * attribute name, and it will be looked up below.
4976 4976 */
4977 4977 if (*nm == '\0') {
4978 4978 *vpp = dvp;
4979 4979 return (0);
4980 4980 }
4981 4981
4982 4982 /*
4983 4983 * The vfs layer never sends a name when asking for the
4984 4984 * attrdir, so we should never get here (unless of course
4985 4985 * name is passed at some time in future -- at which time
4986 4986 * we'll blow up here).
4987 4987 */
4988 4988 ASSERT(0);
4989 4989 }
4990 4990
4991 4991 drp = VTOR4(dvp);
4992 4992 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR4(dvp)))
4993 4993 return (EINTR);
4994 4994
4995 4995 error = nfs4lookup(dvp, nm, vpp, cr, 0);
4996 4996 nfs_rw_exit(&drp->r_rwlock);
4997 4997
4998 4998 /*
4999 4999 * If vnode is a device, create special vnode.
5000 5000 */
5001 5001 if (!error && ISVDEV((*vpp)->v_type)) {
5002 5002 vp = *vpp;
5003 5003 *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr);
5004 5004 VN_RELE(vp);
5005 5005 }
5006 5006
5007 5007 return (error);
5008 5008 }
5009 5009
5010 5010 /* ARGSUSED */
5011 5011 static int
5012 5012 nfs4lookup_xattr(vnode_t *dvp, char *nm, vnode_t **vpp, int flags, cred_t *cr)
5013 5013 {
5014 5014 int error;
5015 5015 rnode4_t *drp;
5016 5016 int cflag = ((flags & CREATE_XATTR_DIR) != 0);
5017 5017 mntinfo4_t *mi;
5018 5018
5019 5019 mi = VTOMI4(dvp);
5020 5020 if (!(mi->mi_vfsp->vfs_flag & VFS_XATTR) &&
5021 5021 !vfs_has_feature(mi->mi_vfsp, VFSFT_SYSATTR_VIEWS))
5022 5022 return (EINVAL);
5023 5023
5024 5024 drp = VTOR4(dvp);
5025 5025 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR4(dvp)))
5026 5026 return (EINTR);
5027 5027
5028 5028 mutex_enter(&drp->r_statelock);
5029 5029 /*
5030 5030 * If the server doesn't support xattrs just return EINVAL
5031 5031 */
5032 5032 if (drp->r_xattr_dir == NFS4_XATTR_DIR_NOTSUPP) {
5033 5033 mutex_exit(&drp->r_statelock);
5034 5034 nfs_rw_exit(&drp->r_rwlock);
5035 5035 return (EINVAL);
5036 5036 }
5037 5037
5038 5038 /*
5039 5039 * If there is a cached xattr directory entry,
5040 5040 * use it as long as the attributes are valid. If the
5041 5041 * attributes are not valid, take the simple approach and
5042 5042 * free the cached value and re-fetch a new value.
5043 5043 *
5044 5044 * We don't negative entry cache for now, if we did we
5045 5045 * would need to check if the file has changed on every
5046 5046 * lookup. But xattrs don't exist very often and failing
5047 5047 * an openattr is not much more expensive than and NVERIFY or GETATTR
5048 5048 * so do an openattr over the wire for now.
5049 5049 */
5050 5050 if (drp->r_xattr_dir != NULL) {
5051 5051 if (ATTRCACHE4_VALID(dvp)) {
5052 5052 VN_HOLD(drp->r_xattr_dir);
5053 5053 *vpp = drp->r_xattr_dir;
5054 5054 mutex_exit(&drp->r_statelock);
5055 5055 nfs_rw_exit(&drp->r_rwlock);
5056 5056 return (0);
5057 5057 }
5058 5058 VN_RELE(drp->r_xattr_dir);
5059 5059 drp->r_xattr_dir = NULL;
5060 5060 }
5061 5061 mutex_exit(&drp->r_statelock);
5062 5062
5063 5063 error = nfs4openattr(dvp, vpp, cflag, cr);
5064 5064
5065 5065 nfs_rw_exit(&drp->r_rwlock);
5066 5066
5067 5067 return (error);
5068 5068 }
5069 5069
5070 5070 static int
5071 5071 nfs4lookup(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr, int skipdnlc)
5072 5072 {
5073 5073 int error;
5074 5074 rnode4_t *drp;
5075 5075
5076 5076 ASSERT(nfs_zone() == VTOMI4(dvp)->mi_zone);
5077 5077
5078 5078 /*
5079 5079 * If lookup is for "", just return dvp. Don't need
5080 5080 * to send it over the wire, look it up in the dnlc,
5081 5081 * or perform any access checks.
5082 5082 */
5083 5083 if (*nm == '\0') {
5084 5084 VN_HOLD(dvp);
5085 5085 *vpp = dvp;
5086 5086 return (0);
5087 5087 }
5088 5088
5089 5089 /*
5090 5090 * Can't do lookups in non-directories.
5091 5091 */
5092 5092 if (dvp->v_type != VDIR)
5093 5093 return (ENOTDIR);
5094 5094
5095 5095 /*
5096 5096 * If lookup is for ".", just return dvp. Don't need
5097 5097 * to send it over the wire or look it up in the dnlc,
5098 5098 * just need to check access.
5099 5099 */
5100 5100 if (nm[0] == '.' && nm[1] == '\0') {
5101 5101 error = nfs4_access(dvp, VEXEC, 0, cr, NULL);
5102 5102 if (error)
5103 5103 return (error);
5104 5104 VN_HOLD(dvp);
5105 5105 *vpp = dvp;
5106 5106 return (0);
5107 5107 }
5108 5108
5109 5109 drp = VTOR4(dvp);
5110 5110 if (!(drp->r_flags & R4LOOKUP)) {
5111 5111 mutex_enter(&drp->r_statelock);
5112 5112 drp->r_flags |= R4LOOKUP;
5113 5113 mutex_exit(&drp->r_statelock);
5114 5114 }
5115 5115
5116 5116 *vpp = NULL;
5117 5117 /*
5118 5118 * Lookup this name in the DNLC. If there is no entry
5119 5119 * lookup over the wire.
5120 5120 */
5121 5121 if (!skipdnlc)
5122 5122 *vpp = dnlc_lookup(dvp, nm);
5123 5123 if (*vpp == NULL) {
5124 5124 /*
5125 5125 * We need to go over the wire to lookup the name.
5126 5126 */
5127 5127 return (nfs4lookupnew_otw(dvp, nm, vpp, cr));
5128 5128 }
5129 5129
5130 5130 /*
5131 5131 * We hit on the dnlc
5132 5132 */
5133 5133 if (*vpp != DNLC_NO_VNODE ||
5134 5134 (dvp->v_vfsp->vfs_flag & VFS_RDONLY)) {
5135 5135 /*
5136 5136 * But our attrs may not be valid.
5137 5137 */
5138 5138 if (ATTRCACHE4_VALID(dvp)) {
5139 5139 error = nfs4_waitfor_purge_complete(dvp);
5140 5140 if (error) {
5141 5141 VN_RELE(*vpp);
5142 5142 *vpp = NULL;
5143 5143 return (error);
5144 5144 }
5145 5145
5146 5146 /*
5147 5147 * If after the purge completes, check to make sure
5148 5148 * our attrs are still valid.
5149 5149 */
5150 5150 if (ATTRCACHE4_VALID(dvp)) {
5151 5151 /*
5152 5152 * If we waited for a purge we may have
5153 5153 * lost our vnode so look it up again.
5154 5154 */
5155 5155 VN_RELE(*vpp);
5156 5156 *vpp = dnlc_lookup(dvp, nm);
5157 5157 if (*vpp == NULL)
5158 5158 return (nfs4lookupnew_otw(dvp,
5159 5159 nm, vpp, cr));
5160 5160
5161 5161 /*
5162 5162 * The access cache should almost always hit
5163 5163 */
5164 5164 error = nfs4_access(dvp, VEXEC, 0, cr, NULL);
5165 5165
5166 5166 if (error) {
5167 5167 VN_RELE(*vpp);
5168 5168 *vpp = NULL;
5169 5169 return (error);
5170 5170 }
5171 5171 if (*vpp == DNLC_NO_VNODE) {
5172 5172 VN_RELE(*vpp);
5173 5173 *vpp = NULL;
5174 5174 return (ENOENT);
5175 5175 }
5176 5176 return (0);
5177 5177 }
5178 5178 }
5179 5179 }
5180 5180
5181 5181 ASSERT(*vpp != NULL);
5182 5182
5183 5183 /*
5184 5184 * We may have gotten here we have one of the following cases:
5185 5185 * 1) vpp != DNLC_NO_VNODE, our attrs have timed out so we
5186 5186 * need to validate them.
5187 5187 * 2) vpp == DNLC_NO_VNODE, a negative entry that we always
5188 5188 * must validate.
5189 5189 *
5190 5190 * Go to the server and check if the directory has changed, if
5191 5191 * it hasn't we are done and can use the dnlc entry.
5192 5192 */
5193 5193 return (nfs4lookupvalidate_otw(dvp, nm, vpp, cr));
5194 5194 }
5195 5195
5196 5196 /*
5197 5197 * Go to the server and check if the directory has changed, if
5198 5198 * it hasn't we are done and can use the dnlc entry. If it
5199 5199 * has changed we get a new copy of its attributes and check
5200 5200 * the access for VEXEC, then relookup the filename and
5201 5201 * get its filehandle and attributes.
5202 5202 *
5203 5203 * PUTFH dfh NVERIFY GETATTR ACCESS LOOKUP GETFH GETATTR
5204 5204 * if the NVERIFY failed we must
5205 5205 * purge the caches
5206 5206 * cache new attributes (will set r_time_attr_inval)
5207 5207 * cache new access
5208 5208 * recheck VEXEC access
5209 5209 * add name to dnlc, possibly negative
5210 5210 * if LOOKUP succeeded
5211 5211 * cache new attributes
5212 5212 * else
5213 5213 * set a new r_time_attr_inval for dvp
5214 5214 * check to make sure we have access
5215 5215 *
5216 5216 * The vpp returned is the vnode passed in if the directory is valid,
5217 5217 * a new vnode if successful lookup, or NULL on error.
5218 5218 */
5219 5219 static int
5220 5220 nfs4lookupvalidate_otw(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr)
5221 5221 {
5222 5222 COMPOUND4args_clnt args;
5223 5223 COMPOUND4res_clnt res;
5224 5224 fattr4 *ver_fattr;
5225 5225 fattr4_change dchange;
5226 5226 int32_t *ptr;
5227 5227 int argoplist_size = 7 * sizeof (nfs_argop4);
5228 5228 nfs_argop4 *argop;
5229 5229 int doqueue;
5230 5230 mntinfo4_t *mi;
5231 5231 nfs4_recov_state_t recov_state;
5232 5232 hrtime_t t;
5233 5233 int isdotdot;
5234 5234 vnode_t *nvp;
5235 5235 nfs_fh4 *fhp;
5236 5236 nfs4_sharedfh_t *sfhp;
5237 5237 nfs4_access_type_t cacc;
5238 5238 rnode4_t *nrp;
5239 5239 rnode4_t *drp = VTOR4(dvp);
5240 5240 nfs4_ga_res_t *garp = NULL;
5241 5241 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
5242 5242
5243 5243 ASSERT(nfs_zone() == VTOMI4(dvp)->mi_zone);
5244 5244 ASSERT(nm != NULL);
5245 5245 ASSERT(nm[0] != '\0');
5246 5246 ASSERT(dvp->v_type == VDIR);
5247 5247 ASSERT(nm[0] != '.' || nm[1] != '\0');
5248 5248 ASSERT(*vpp != NULL);
5249 5249
5250 5250 if (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0') {
5251 5251 isdotdot = 1;
5252 5252 args.ctag = TAG_LOOKUP_VPARENT;
5253 5253 } else {
5254 5254 /*
5255 5255 * If dvp were a stub, it should have triggered and caused
5256 5256 * a mount for us to get this far.
5257 5257 */
5258 5258 ASSERT(!RP_ISSTUB(VTOR4(dvp)));
5259 5259
5260 5260 isdotdot = 0;
5261 5261 args.ctag = TAG_LOOKUP_VALID;
5262 5262 }
5263 5263
5264 5264 mi = VTOMI4(dvp);
5265 5265 recov_state.rs_flags = 0;
5266 5266 recov_state.rs_num_retry_despite_err = 0;
5267 5267
5268 5268 nvp = NULL;
5269 5269
5270 5270 /* Save the original mount point security information */
5271 5271 (void) save_mnt_secinfo(mi->mi_curr_serv);
5272 5272
5273 5273 recov_retry:
5274 5274 e.error = nfs4_start_fop(mi, dvp, NULL, OH_LOOKUP,
5275 5275 &recov_state, NULL);
5276 5276 if (e.error) {
5277 5277 (void) check_mnt_secinfo(mi->mi_curr_serv, nvp);
5278 5278 VN_RELE(*vpp);
5279 5279 *vpp = NULL;
5280 5280 return (e.error);
5281 5281 }
5282 5282
5283 5283 argop = kmem_alloc(argoplist_size, KM_SLEEP);
5284 5284
5285 5285 /* PUTFH dfh NVERIFY GETATTR ACCESS LOOKUP GETFH GETATTR */
5286 5286 args.array_len = 7;
5287 5287 args.array = argop;
5288 5288
5289 5289 /* 0. putfh file */
5290 5290 argop[0].argop = OP_CPUTFH;
5291 5291 argop[0].nfs_argop4_u.opcputfh.sfh = VTOR4(dvp)->r_fh;
5292 5292
5293 5293 /* 1. nverify the change info */
5294 5294 argop[1].argop = OP_NVERIFY;
5295 5295 ver_fattr = &argop[1].nfs_argop4_u.opnverify.obj_attributes;
5296 5296 ver_fattr->attrmask = FATTR4_CHANGE_MASK;
5297 5297 ver_fattr->attrlist4 = (char *)&dchange;
5298 5298 ptr = (int32_t *)&dchange;
5299 5299 IXDR_PUT_HYPER(ptr, VTOR4(dvp)->r_change);
5300 5300 ver_fattr->attrlist4_len = sizeof (fattr4_change);
5301 5301
5302 5302 /* 2. getattr directory */
5303 5303 argop[2].argop = OP_GETATTR;
5304 5304 argop[2].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
5305 5305 argop[2].nfs_argop4_u.opgetattr.mi = VTOMI4(dvp);
5306 5306
5307 5307 /* 3. access directory */
5308 5308 argop[3].argop = OP_ACCESS;
5309 5309 argop[3].nfs_argop4_u.opaccess.access = ACCESS4_READ | ACCESS4_DELETE |
5310 5310 ACCESS4_MODIFY | ACCESS4_EXTEND | ACCESS4_LOOKUP;
5311 5311
5312 5312 /* 4. lookup name */
5313 5313 if (isdotdot) {
5314 5314 argop[4].argop = OP_LOOKUPP;
5315 5315 } else {
5316 5316 argop[4].argop = OP_CLOOKUP;
5317 5317 argop[4].nfs_argop4_u.opclookup.cname = nm;
5318 5318 }
5319 5319
5320 5320 /* 5. resulting file handle */
5321 5321 argop[5].argop = OP_GETFH;
5322 5322
5323 5323 /* 6. resulting file attributes */
5324 5324 argop[6].argop = OP_GETATTR;
5325 5325 argop[6].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
5326 5326 argop[6].nfs_argop4_u.opgetattr.mi = VTOMI4(dvp);
5327 5327
5328 5328 doqueue = 1;
5329 5329 t = gethrtime();
5330 5330
5331 5331 rfs4call(VTOMI4(dvp), &args, &res, cr, &doqueue, 0, &e);
5332 5332
5333 5333 if (!isdotdot && res.status == NFS4ERR_MOVED) {
5334 5334 e.error = nfs4_setup_referral(dvp, nm, vpp, cr);
5335 5335 if (e.error != 0 && *vpp != NULL)
5336 5336 VN_RELE(*vpp);
5337 5337 nfs4_end_fop(mi, dvp, NULL, OH_LOOKUP,
5338 5338 &recov_state, FALSE);
5339 5339 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
5340 5340 kmem_free(argop, argoplist_size);
5341 5341 return (e.error);
5342 5342 }
5343 5343
5344 5344 if (nfs4_needs_recovery(&e, FALSE, dvp->v_vfsp)) {
5345 5345 /*
5346 5346 * For WRONGSEC of a non-dotdot case, send secinfo directly
5347 5347 * from this thread, do not go thru the recovery thread since
5348 5348 * we need the nm information.
5349 5349 *
5350 5350 * Not doing dotdot case because there is no specification
5351 5351 * for (PUTFH, SECINFO "..") yet.
5352 5352 */
5353 5353 if (!isdotdot && res.status == NFS4ERR_WRONGSEC) {
5354 5354 if ((e.error = nfs4_secinfo_vnode_otw(dvp, nm, cr)))
5355 5355 nfs4_end_fop(mi, dvp, NULL, OH_LOOKUP,
5356 5356 &recov_state, FALSE);
5357 5357 else
5358 5358 nfs4_end_fop(mi, dvp, NULL, OH_LOOKUP,
5359 5359 &recov_state, TRUE);
5360 5360 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
5361 5361 kmem_free(argop, argoplist_size);
5362 5362 if (!e.error)
5363 5363 goto recov_retry;
5364 5364 (void) check_mnt_secinfo(mi->mi_curr_serv, nvp);
5365 5365 VN_RELE(*vpp);
5366 5366 *vpp = NULL;
5367 5367 return (e.error);
5368 5368 }
5369 5369
5370 5370 if (nfs4_start_recovery(&e, mi, dvp, NULL, NULL, NULL,
5371 5371 OP_LOOKUP, NULL, NULL, NULL) == FALSE) {
5372 5372 nfs4_end_fop(mi, dvp, NULL, OH_LOOKUP,
5373 5373 &recov_state, TRUE);
5374 5374
5375 5375 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
5376 5376 kmem_free(argop, argoplist_size);
5377 5377 goto recov_retry;
5378 5378 }
5379 5379 }
5380 5380
5381 5381 nfs4_end_fop(mi, dvp, NULL, OH_LOOKUP, &recov_state, FALSE);
5382 5382
5383 5383 if (e.error || res.array_len == 0) {
5384 5384 /*
5385 5385 * If e.error isn't set, then reply has no ops (or we couldn't
5386 5386 * be here). The only legal way to reply without an op array
5387 5387 * is via NFS4ERR_MINOR_VERS_MISMATCH. An ops array should
5388 5388 * be in the reply for all other status values.
5389 5389 *
5390 5390 * For valid replies without an ops array, return ENOTSUP
5391 5391 * (geterrno4 xlation of VERS_MISMATCH). For illegal replies,
5392 5392 * return EIO -- don't trust status.
5393 5393 */
5394 5394 if (e.error == 0)
5395 5395 e.error = (res.status == NFS4ERR_MINOR_VERS_MISMATCH) ?
5396 5396 ENOTSUP : EIO;
5397 5397 VN_RELE(*vpp);
5398 5398 *vpp = NULL;
5399 5399 kmem_free(argop, argoplist_size);
5400 5400 (void) check_mnt_secinfo(mi->mi_curr_serv, nvp);
5401 5401 return (e.error);
5402 5402 }
5403 5403
5404 5404 if (res.status != NFS4ERR_SAME) {
5405 5405 e.error = geterrno4(res.status);
5406 5406
5407 5407 /*
5408 5408 * The NVERIFY "failed" so the directory has changed
5409 5409 * First make sure PUTFH succeeded and NVERIFY "failed"
5410 5410 * cleanly.
5411 5411 */
5412 5412 if ((res.array[0].nfs_resop4_u.opputfh.status != NFS4_OK) ||
5413 5413 (res.array[1].nfs_resop4_u.opnverify.status != NFS4_OK)) {
5414 5414 nfs4_purge_stale_fh(e.error, dvp, cr);
5415 5415 VN_RELE(*vpp);
5416 5416 *vpp = NULL;
5417 5417 goto exit;
5418 5418 }
5419 5419
5420 5420 /*
5421 5421 * We know the NVERIFY "failed" so we must:
5422 5422 * purge the caches (access and indirectly dnlc if needed)
5423 5423 */
5424 5424 nfs4_purge_caches(dvp, NFS4_NOPURGE_DNLC, cr, TRUE);
5425 5425
5426 5426 if (res.array[2].nfs_resop4_u.opgetattr.status != NFS4_OK) {
5427 5427 nfs4_purge_stale_fh(e.error, dvp, cr);
5428 5428 VN_RELE(*vpp);
5429 5429 *vpp = NULL;
5430 5430 goto exit;
5431 5431 }
5432 5432
5433 5433 /*
5434 5434 * Install new cached attributes for the directory
5435 5435 */
5436 5436 nfs4_attr_cache(dvp,
5437 5437 &res.array[2].nfs_resop4_u.opgetattr.ga_res,
5438 5438 t, cr, FALSE, NULL);
5439 5439
5440 5440 if (res.array[3].nfs_resop4_u.opaccess.status != NFS4_OK) {
5441 5441 nfs4_purge_stale_fh(e.error, dvp, cr);
5442 5442 VN_RELE(*vpp);
5443 5443 *vpp = NULL;
5444 5444 e.error = geterrno4(res.status);
5445 5445 goto exit;
5446 5446 }
5447 5447
5448 5448 /*
5449 5449 * Now we know the directory is valid,
5450 5450 * cache new directory access
5451 5451 */
5452 5452 nfs4_access_cache(drp,
5453 5453 args.array[3].nfs_argop4_u.opaccess.access,
5454 5454 res.array[3].nfs_resop4_u.opaccess.access, cr);
5455 5455
5456 5456 /*
5457 5457 * recheck VEXEC access
5458 5458 */
5459 5459 cacc = nfs4_access_check(drp, ACCESS4_LOOKUP, cr);
5460 5460 if (cacc != NFS4_ACCESS_ALLOWED) {
5461 5461 /*
5462 5462 * Directory permissions might have been revoked
5463 5463 */
5464 5464 if (cacc == NFS4_ACCESS_DENIED) {
5465 5465 e.error = EACCES;
5466 5466 VN_RELE(*vpp);
5467 5467 *vpp = NULL;
5468 5468 goto exit;
5469 5469 }
5470 5470
5471 5471 /*
5472 5472 * Somehow we must not have asked for enough
5473 5473 * so try a singleton ACCESS, should never happen.
5474 5474 */
5475 5475 e.error = nfs4_access(dvp, VEXEC, 0, cr, NULL);
5476 5476 if (e.error) {
5477 5477 VN_RELE(*vpp);
5478 5478 *vpp = NULL;
5479 5479 goto exit;
5480 5480 }
5481 5481 }
5482 5482
5483 5483 e.error = geterrno4(res.status);
5484 5484 if (res.array[4].nfs_resop4_u.oplookup.status != NFS4_OK) {
5485 5485 /*
5486 5486 * The lookup failed, probably no entry
5487 5487 */
5488 5488 if (e.error == ENOENT && nfs4_lookup_neg_cache) {
5489 5489 dnlc_update(dvp, nm, DNLC_NO_VNODE);
5490 5490 } else {
5491 5491 /*
5492 5492 * Might be some other error, so remove
5493 5493 * the dnlc entry to make sure we start all
5494 5494 * over again, next time.
5495 5495 */
5496 5496 dnlc_remove(dvp, nm);
5497 5497 }
5498 5498 VN_RELE(*vpp);
5499 5499 *vpp = NULL;
5500 5500 goto exit;
5501 5501 }
5502 5502
5503 5503 if (res.array[5].nfs_resop4_u.opgetfh.status != NFS4_OK) {
5504 5504 /*
5505 5505 * The file exists but we can't get its fh for
5506 5506 * some unknown reason. Remove it from the dnlc
5507 5507 * and error out to be safe.
5508 5508 */
5509 5509 dnlc_remove(dvp, nm);
5510 5510 VN_RELE(*vpp);
5511 5511 *vpp = NULL;
5512 5512 goto exit;
5513 5513 }
5514 5514 fhp = &res.array[5].nfs_resop4_u.opgetfh.object;
5515 5515 if (fhp->nfs_fh4_len == 0) {
5516 5516 /*
5517 5517 * The file exists but a bogus fh
5518 5518 * some unknown reason. Remove it from the dnlc
5519 5519 * and error out to be safe.
5520 5520 */
5521 5521 e.error = ENOENT;
5522 5522 dnlc_remove(dvp, nm);
5523 5523 VN_RELE(*vpp);
5524 5524 *vpp = NULL;
5525 5525 goto exit;
5526 5526 }
5527 5527 sfhp = sfh4_get(fhp, mi);
5528 5528
5529 5529 if (res.array[6].nfs_resop4_u.opgetattr.status == NFS4_OK)
5530 5530 garp = &res.array[6].nfs_resop4_u.opgetattr.ga_res;
5531 5531
5532 5532 /*
5533 5533 * Make the new rnode
5534 5534 */
5535 5535 if (isdotdot) {
5536 5536 e.error = nfs4_make_dotdot(sfhp, t, dvp, cr, &nvp, 1);
5537 5537 if (e.error) {
5538 5538 sfh4_rele(&sfhp);
5539 5539 VN_RELE(*vpp);
5540 5540 *vpp = NULL;
5541 5541 goto exit;
5542 5542 }
5543 5543 /*
5544 5544 * XXX if nfs4_make_dotdot uses an existing rnode
5545 5545 * XXX it doesn't update the attributes.
5546 5546 * XXX for now just save them again to save an OTW
5547 5547 */
5548 5548 nfs4_attr_cache(nvp, garp, t, cr, FALSE, NULL);
5549 5549 } else {
5550 5550 nvp = makenfs4node(sfhp, garp, dvp->v_vfsp, t, cr,
5551 5551 dvp, fn_get(VTOSV(dvp)->sv_name, nm, sfhp));
5552 5552 /*
5553 5553 * If v_type == VNON, then garp was NULL because
5554 5554 * the last op in the compound failed and makenfs4node
5555 5555 * could not find the vnode for sfhp. It created
5556 5556 * a new vnode, so we have nothing to purge here.
5557 5557 */
5558 5558 if (nvp->v_type == VNON) {
5559 5559 vattr_t vattr;
5560 5560
5561 5561 vattr.va_mask = AT_TYPE;
5562 5562 /*
5563 5563 * N.B. We've already called nfs4_end_fop above.
5564 5564 */
5565 5565 e.error = nfs4getattr(nvp, &vattr, cr);
5566 5566 if (e.error) {
5567 5567 sfh4_rele(&sfhp);
5568 5568 VN_RELE(*vpp);
5569 5569 *vpp = NULL;
5570 5570 VN_RELE(nvp);
5571 5571 goto exit;
5572 5572 }
5573 5573 nvp->v_type = vattr.va_type;
5574 5574 }
5575 5575 }
5576 5576 sfh4_rele(&sfhp);
5577 5577
5578 5578 nrp = VTOR4(nvp);
5579 5579 mutex_enter(&nrp->r_statev4_lock);
5580 5580 if (!nrp->created_v4) {
5581 5581 mutex_exit(&nrp->r_statev4_lock);
5582 5582 dnlc_update(dvp, nm, nvp);
5583 5583 } else
5584 5584 mutex_exit(&nrp->r_statev4_lock);
5585 5585
5586 5586 VN_RELE(*vpp);
5587 5587 *vpp = nvp;
5588 5588 } else {
5589 5589 hrtime_t now;
5590 5590 hrtime_t delta = 0;
5591 5591
5592 5592 e.error = 0;
5593 5593
5594 5594 /*
5595 5595 * Because the NVERIFY "succeeded" we know that the
5596 5596 * directory attributes are still valid
5597 5597 * so update r_time_attr_inval
5598 5598 */
5599 5599 now = gethrtime();
5600 5600 mutex_enter(&drp->r_statelock);
5601 5601 if (!(mi->mi_flags & MI4_NOAC) && !(dvp->v_flag & VNOCACHE)) {
5602 5602 delta = now - drp->r_time_attr_saved;
5603 5603 if (delta < mi->mi_acdirmin)
5604 5604 delta = mi->mi_acdirmin;
5605 5605 else if (delta > mi->mi_acdirmax)
5606 5606 delta = mi->mi_acdirmax;
5607 5607 }
5608 5608 drp->r_time_attr_inval = now + delta;
5609 5609 mutex_exit(&drp->r_statelock);
5610 5610 dnlc_update(dvp, nm, *vpp);
5611 5611
5612 5612 /*
5613 5613 * Even though we have a valid directory attr cache
5614 5614 * and dnlc entry, we may not have access.
5615 5615 * This should almost always hit the cache.
5616 5616 */
5617 5617 e.error = nfs4_access(dvp, VEXEC, 0, cr, NULL);
5618 5618 if (e.error) {
5619 5619 VN_RELE(*vpp);
5620 5620 *vpp = NULL;
5621 5621 }
5622 5622
5623 5623 if (*vpp == DNLC_NO_VNODE) {
5624 5624 VN_RELE(*vpp);
5625 5625 *vpp = NULL;
5626 5626 e.error = ENOENT;
5627 5627 }
5628 5628 }
5629 5629
5630 5630 exit:
5631 5631 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
5632 5632 kmem_free(argop, argoplist_size);
5633 5633 (void) check_mnt_secinfo(mi->mi_curr_serv, nvp);
5634 5634 return (e.error);
5635 5635 }
5636 5636
5637 5637 /*
5638 5638 * We need to go over the wire to lookup the name, but
5639 5639 * while we are there verify the directory has not
5640 5640 * changed but if it has, get new attributes and check access
5641 5641 *
5642 5642 * PUTFH dfh SAVEFH LOOKUP nm GETFH GETATTR RESTOREFH
5643 5643 * NVERIFY GETATTR ACCESS
5644 5644 *
5645 5645 * With the results:
5646 5646 * if the NVERIFY failed we must purge the caches, add new attributes,
5647 5647 * and cache new access.
5648 5648 * set a new r_time_attr_inval
5649 5649 * add name to dnlc, possibly negative
5650 5650 * if LOOKUP succeeded
5651 5651 * cache new attributes
5652 5652 */
5653 5653 static int
5654 5654 nfs4lookupnew_otw(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr)
5655 5655 {
5656 5656 COMPOUND4args_clnt args;
5657 5657 COMPOUND4res_clnt res;
5658 5658 fattr4 *ver_fattr;
5659 5659 fattr4_change dchange;
5660 5660 int32_t *ptr;
5661 5661 nfs4_ga_res_t *garp = NULL;
5662 5662 int argoplist_size = 9 * sizeof (nfs_argop4);
5663 5663 nfs_argop4 *argop;
5664 5664 int doqueue;
5665 5665 mntinfo4_t *mi;
5666 5666 nfs4_recov_state_t recov_state;
5667 5667 hrtime_t t;
5668 5668 int isdotdot;
5669 5669 vnode_t *nvp;
5670 5670 nfs_fh4 *fhp;
5671 5671 nfs4_sharedfh_t *sfhp;
5672 5672 nfs4_access_type_t cacc;
5673 5673 rnode4_t *nrp;
5674 5674 rnode4_t *drp = VTOR4(dvp);
5675 5675 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
5676 5676
5677 5677 ASSERT(nfs_zone() == VTOMI4(dvp)->mi_zone);
5678 5678 ASSERT(nm != NULL);
5679 5679 ASSERT(nm[0] != '\0');
5680 5680 ASSERT(dvp->v_type == VDIR);
5681 5681 ASSERT(nm[0] != '.' || nm[1] != '\0');
5682 5682 ASSERT(*vpp == NULL);
5683 5683
5684 5684 if (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0') {
5685 5685 isdotdot = 1;
5686 5686 args.ctag = TAG_LOOKUP_PARENT;
5687 5687 } else {
5688 5688 /*
5689 5689 * If dvp were a stub, it should have triggered and caused
5690 5690 * a mount for us to get this far.
5691 5691 */
5692 5692 ASSERT(!RP_ISSTUB(VTOR4(dvp)));
5693 5693
5694 5694 isdotdot = 0;
5695 5695 args.ctag = TAG_LOOKUP;
5696 5696 }
5697 5697
5698 5698 mi = VTOMI4(dvp);
5699 5699 recov_state.rs_flags = 0;
5700 5700 recov_state.rs_num_retry_despite_err = 0;
5701 5701
5702 5702 nvp = NULL;
5703 5703
5704 5704 /* Save the original mount point security information */
5705 5705 (void) save_mnt_secinfo(mi->mi_curr_serv);
5706 5706
5707 5707 recov_retry:
5708 5708 e.error = nfs4_start_fop(mi, dvp, NULL, OH_LOOKUP,
5709 5709 &recov_state, NULL);
5710 5710 if (e.error) {
5711 5711 (void) check_mnt_secinfo(mi->mi_curr_serv, nvp);
5712 5712 return (e.error);
5713 5713 }
5714 5714
5715 5715 argop = kmem_alloc(argoplist_size, KM_SLEEP);
5716 5716
5717 5717 /* PUTFH SAVEFH LOOKUP GETFH GETATTR RESTOREFH NVERIFY GETATTR ACCESS */
5718 5718 args.array_len = 9;
5719 5719 args.array = argop;
5720 5720
5721 5721 /* 0. putfh file */
5722 5722 argop[0].argop = OP_CPUTFH;
5723 5723 argop[0].nfs_argop4_u.opcputfh.sfh = VTOR4(dvp)->r_fh;
5724 5724
5725 5725 /* 1. savefh for the nverify */
5726 5726 argop[1].argop = OP_SAVEFH;
5727 5727
5728 5728 /* 2. lookup name */
5729 5729 if (isdotdot) {
5730 5730 argop[2].argop = OP_LOOKUPP;
5731 5731 } else {
5732 5732 argop[2].argop = OP_CLOOKUP;
5733 5733 argop[2].nfs_argop4_u.opclookup.cname = nm;
5734 5734 }
5735 5735
5736 5736 /* 3. resulting file handle */
5737 5737 argop[3].argop = OP_GETFH;
5738 5738
5739 5739 /* 4. resulting file attributes */
5740 5740 argop[4].argop = OP_GETATTR;
5741 5741 argop[4].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
5742 5742 argop[4].nfs_argop4_u.opgetattr.mi = VTOMI4(dvp);
5743 5743
5744 5744 /* 5. restorefh back the directory for the nverify */
5745 5745 argop[5].argop = OP_RESTOREFH;
5746 5746
5747 5747 /* 6. nverify the change info */
5748 5748 argop[6].argop = OP_NVERIFY;
5749 5749 ver_fattr = &argop[6].nfs_argop4_u.opnverify.obj_attributes;
5750 5750 ver_fattr->attrmask = FATTR4_CHANGE_MASK;
5751 5751 ver_fattr->attrlist4 = (char *)&dchange;
5752 5752 ptr = (int32_t *)&dchange;
5753 5753 IXDR_PUT_HYPER(ptr, VTOR4(dvp)->r_change);
5754 5754 ver_fattr->attrlist4_len = sizeof (fattr4_change);
5755 5755
5756 5756 /* 7. getattr directory */
5757 5757 argop[7].argop = OP_GETATTR;
5758 5758 argop[7].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
5759 5759 argop[7].nfs_argop4_u.opgetattr.mi = VTOMI4(dvp);
5760 5760
5761 5761 /* 8. access directory */
5762 5762 argop[8].argop = OP_ACCESS;
5763 5763 argop[8].nfs_argop4_u.opaccess.access = ACCESS4_READ | ACCESS4_DELETE |
5764 5764 ACCESS4_MODIFY | ACCESS4_EXTEND | ACCESS4_LOOKUP;
5765 5765
5766 5766 doqueue = 1;
5767 5767 t = gethrtime();
5768 5768
5769 5769 rfs4call(VTOMI4(dvp), &args, &res, cr, &doqueue, 0, &e);
5770 5770
5771 5771 if (!isdotdot && res.status == NFS4ERR_MOVED) {
5772 5772 e.error = nfs4_setup_referral(dvp, nm, vpp, cr);
5773 5773 if (e.error != 0 && *vpp != NULL)
5774 5774 VN_RELE(*vpp);
5775 5775 nfs4_end_fop(mi, dvp, NULL, OH_LOOKUP,
5776 5776 &recov_state, FALSE);
5777 5777 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
5778 5778 kmem_free(argop, argoplist_size);
5779 5779 return (e.error);
5780 5780 }
5781 5781
5782 5782 if (nfs4_needs_recovery(&e, FALSE, dvp->v_vfsp)) {
5783 5783 /*
5784 5784 * For WRONGSEC of a non-dotdot case, send secinfo directly
5785 5785 * from this thread, do not go thru the recovery thread since
5786 5786 * we need the nm information.
5787 5787 *
5788 5788 * Not doing dotdot case because there is no specification
5789 5789 * for (PUTFH, SECINFO "..") yet.
5790 5790 */
5791 5791 if (!isdotdot && res.status == NFS4ERR_WRONGSEC) {
5792 5792 if ((e.error = nfs4_secinfo_vnode_otw(dvp, nm, cr)))
5793 5793 nfs4_end_fop(mi, dvp, NULL, OH_LOOKUP,
5794 5794 &recov_state, FALSE);
5795 5795 else
5796 5796 nfs4_end_fop(mi, dvp, NULL, OH_LOOKUP,
5797 5797 &recov_state, TRUE);
5798 5798 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
5799 5799 kmem_free(argop, argoplist_size);
5800 5800 if (!e.error)
5801 5801 goto recov_retry;
5802 5802 (void) check_mnt_secinfo(mi->mi_curr_serv, nvp);
5803 5803 return (e.error);
5804 5804 }
5805 5805
5806 5806 if (nfs4_start_recovery(&e, mi, dvp, NULL, NULL, NULL,
5807 5807 OP_LOOKUP, NULL, NULL, NULL) == FALSE) {
5808 5808 nfs4_end_fop(mi, dvp, NULL, OH_LOOKUP,
5809 5809 &recov_state, TRUE);
5810 5810
5811 5811 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
5812 5812 kmem_free(argop, argoplist_size);
5813 5813 goto recov_retry;
5814 5814 }
5815 5815 }
5816 5816
5817 5817 nfs4_end_fop(mi, dvp, NULL, OH_LOOKUP, &recov_state, FALSE);
5818 5818
5819 5819 if (e.error || res.array_len == 0) {
5820 5820 /*
5821 5821 * If e.error isn't set, then reply has no ops (or we couldn't
5822 5822 * be here). The only legal way to reply without an op array
5823 5823 * is via NFS4ERR_MINOR_VERS_MISMATCH. An ops array should
5824 5824 * be in the reply for all other status values.
5825 5825 *
5826 5826 * For valid replies without an ops array, return ENOTSUP
5827 5827 * (geterrno4 xlation of VERS_MISMATCH). For illegal replies,
5828 5828 * return EIO -- don't trust status.
5829 5829 */
5830 5830 if (e.error == 0)
5831 5831 e.error = (res.status == NFS4ERR_MINOR_VERS_MISMATCH) ?
5832 5832 ENOTSUP : EIO;
5833 5833
5834 5834 kmem_free(argop, argoplist_size);
5835 5835 (void) check_mnt_secinfo(mi->mi_curr_serv, nvp);
5836 5836 return (e.error);
5837 5837 }
5838 5838
5839 5839 e.error = geterrno4(res.status);
5840 5840
5841 5841 /*
5842 5842 * The PUTFH and SAVEFH may have failed.
5843 5843 */
5844 5844 if ((res.array[0].nfs_resop4_u.opputfh.status != NFS4_OK) ||
5845 5845 (res.array[1].nfs_resop4_u.opsavefh.status != NFS4_OK)) {
5846 5846 nfs4_purge_stale_fh(e.error, dvp, cr);
5847 5847 goto exit;
5848 5848 }
5849 5849
5850 5850 /*
5851 5851 * Check if the file exists, if it does delay entering
5852 5852 * into the dnlc until after we update the directory
5853 5853 * attributes so we don't cause it to get purged immediately.
5854 5854 */
5855 5855 if (res.array[2].nfs_resop4_u.oplookup.status != NFS4_OK) {
5856 5856 /*
5857 5857 * The lookup failed, probably no entry
5858 5858 */
5859 5859 if (e.error == ENOENT && nfs4_lookup_neg_cache)
5860 5860 dnlc_update(dvp, nm, DNLC_NO_VNODE);
5861 5861 goto exit;
5862 5862 }
5863 5863
5864 5864 if (res.array[3].nfs_resop4_u.opgetfh.status != NFS4_OK) {
5865 5865 /*
5866 5866 * The file exists but we can't get its fh for
5867 5867 * some unknown reason. Error out to be safe.
5868 5868 */
5869 5869 goto exit;
5870 5870 }
5871 5871
5872 5872 fhp = &res.array[3].nfs_resop4_u.opgetfh.object;
5873 5873 if (fhp->nfs_fh4_len == 0) {
5874 5874 /*
5875 5875 * The file exists but a bogus fh
5876 5876 * some unknown reason. Error out to be safe.
5877 5877 */
5878 5878 e.error = EIO;
5879 5879 goto exit;
5880 5880 }
5881 5881 sfhp = sfh4_get(fhp, mi);
5882 5882
5883 5883 if (res.array[4].nfs_resop4_u.opgetattr.status != NFS4_OK) {
5884 5884 sfh4_rele(&sfhp);
5885 5885 goto exit;
5886 5886 }
5887 5887 garp = &res.array[4].nfs_resop4_u.opgetattr.ga_res;
5888 5888
5889 5889 /*
5890 5890 * The RESTOREFH may have failed
5891 5891 */
5892 5892 if (res.array[5].nfs_resop4_u.oprestorefh.status != NFS4_OK) {
5893 5893 sfh4_rele(&sfhp);
5894 5894 e.error = EIO;
5895 5895 goto exit;
5896 5896 }
5897 5897
5898 5898 if (res.array[6].nfs_resop4_u.opnverify.status != NFS4ERR_SAME) {
5899 5899 /*
5900 5900 * First make sure the NVERIFY failed as we expected,
5901 5901 * if it didn't then be conservative and error out
5902 5902 * as we can't trust the directory.
5903 5903 */
5904 5904 if (res.array[6].nfs_resop4_u.opnverify.status != NFS4_OK) {
5905 5905 sfh4_rele(&sfhp);
5906 5906 e.error = EIO;
5907 5907 goto exit;
5908 5908 }
5909 5909
5910 5910 /*
5911 5911 * We know the NVERIFY "failed" so the directory has changed,
5912 5912 * so we must:
5913 5913 * purge the caches (access and indirectly dnlc if needed)
5914 5914 */
5915 5915 nfs4_purge_caches(dvp, NFS4_NOPURGE_DNLC, cr, TRUE);
5916 5916
5917 5917 if (res.array[7].nfs_resop4_u.opgetattr.status != NFS4_OK) {
5918 5918 sfh4_rele(&sfhp);
5919 5919 goto exit;
5920 5920 }
5921 5921 nfs4_attr_cache(dvp,
5922 5922 &res.array[7].nfs_resop4_u.opgetattr.ga_res,
5923 5923 t, cr, FALSE, NULL);
5924 5924
5925 5925 if (res.array[8].nfs_resop4_u.opaccess.status != NFS4_OK) {
5926 5926 nfs4_purge_stale_fh(e.error, dvp, cr);
5927 5927 sfh4_rele(&sfhp);
5928 5928 e.error = geterrno4(res.status);
5929 5929 goto exit;
5930 5930 }
5931 5931
5932 5932 /*
5933 5933 * Now we know the directory is valid,
5934 5934 * cache new directory access
5935 5935 */
5936 5936 nfs4_access_cache(drp,
5937 5937 args.array[8].nfs_argop4_u.opaccess.access,
5938 5938 res.array[8].nfs_resop4_u.opaccess.access, cr);
5939 5939
5940 5940 /*
5941 5941 * recheck VEXEC access
5942 5942 */
5943 5943 cacc = nfs4_access_check(drp, ACCESS4_LOOKUP, cr);
5944 5944 if (cacc != NFS4_ACCESS_ALLOWED) {
5945 5945 /*
5946 5946 * Directory permissions might have been revoked
5947 5947 */
5948 5948 if (cacc == NFS4_ACCESS_DENIED) {
5949 5949 sfh4_rele(&sfhp);
5950 5950 e.error = EACCES;
5951 5951 goto exit;
5952 5952 }
5953 5953
5954 5954 /*
5955 5955 * Somehow we must not have asked for enough
5956 5956 * so try a singleton ACCESS should never happen
5957 5957 */
5958 5958 e.error = nfs4_access(dvp, VEXEC, 0, cr, NULL);
5959 5959 if (e.error) {
5960 5960 sfh4_rele(&sfhp);
5961 5961 goto exit;
5962 5962 }
5963 5963 }
5964 5964
5965 5965 e.error = geterrno4(res.status);
5966 5966 } else {
5967 5967 hrtime_t now;
5968 5968 hrtime_t delta = 0;
5969 5969
5970 5970 e.error = 0;
5971 5971
5972 5972 /*
5973 5973 * Because the NVERIFY "succeeded" we know that the
5974 5974 * directory attributes are still valid
5975 5975 * so update r_time_attr_inval
5976 5976 */
5977 5977 now = gethrtime();
5978 5978 mutex_enter(&drp->r_statelock);
5979 5979 if (!(mi->mi_flags & MI4_NOAC) && !(dvp->v_flag & VNOCACHE)) {
5980 5980 delta = now - drp->r_time_attr_saved;
5981 5981 if (delta < mi->mi_acdirmin)
5982 5982 delta = mi->mi_acdirmin;
5983 5983 else if (delta > mi->mi_acdirmax)
5984 5984 delta = mi->mi_acdirmax;
5985 5985 }
5986 5986 drp->r_time_attr_inval = now + delta;
5987 5987 mutex_exit(&drp->r_statelock);
5988 5988
5989 5989 /*
5990 5990 * Even though we have a valid directory attr cache,
5991 5991 * we may not have access.
5992 5992 * This should almost always hit the cache.
5993 5993 */
5994 5994 e.error = nfs4_access(dvp, VEXEC, 0, cr, NULL);
5995 5995 if (e.error) {
5996 5996 sfh4_rele(&sfhp);
5997 5997 goto exit;
5998 5998 }
5999 5999 }
6000 6000
6001 6001 /*
6002 6002 * Now we have successfully completed the lookup, if the
6003 6003 * directory has changed we now have the valid attributes.
6004 6004 * We also know we have directory access.
6005 6005 * Create the new rnode and insert it in the dnlc.
6006 6006 */
6007 6007 if (isdotdot) {
6008 6008 e.error = nfs4_make_dotdot(sfhp, t, dvp, cr, &nvp, 1);
6009 6009 if (e.error) {
6010 6010 sfh4_rele(&sfhp);
6011 6011 goto exit;
6012 6012 }
6013 6013 /*
6014 6014 * XXX if nfs4_make_dotdot uses an existing rnode
6015 6015 * XXX it doesn't update the attributes.
6016 6016 * XXX for now just save them again to save an OTW
6017 6017 */
6018 6018 nfs4_attr_cache(nvp, garp, t, cr, FALSE, NULL);
6019 6019 } else {
6020 6020 nvp = makenfs4node(sfhp, garp, dvp->v_vfsp, t, cr,
6021 6021 dvp, fn_get(VTOSV(dvp)->sv_name, nm, sfhp));
6022 6022 }
6023 6023 sfh4_rele(&sfhp);
6024 6024
6025 6025 nrp = VTOR4(nvp);
6026 6026 mutex_enter(&nrp->r_statev4_lock);
6027 6027 if (!nrp->created_v4) {
6028 6028 mutex_exit(&nrp->r_statev4_lock);
6029 6029 dnlc_update(dvp, nm, nvp);
6030 6030 } else
6031 6031 mutex_exit(&nrp->r_statev4_lock);
6032 6032
6033 6033 *vpp = nvp;
6034 6034
6035 6035 exit:
6036 6036 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
6037 6037 kmem_free(argop, argoplist_size);
6038 6038 (void) check_mnt_secinfo(mi->mi_curr_serv, nvp);
6039 6039 return (e.error);
6040 6040 }
6041 6041
6042 6042 #ifdef DEBUG
6043 6043 void
6044 6044 nfs4lookup_dump_compound(char *where, nfs_argop4 *argbase, int argcnt)
6045 6045 {
6046 6046 uint_t i, len;
6047 6047 zoneid_t zoneid = getzoneid();
6048 6048 char *s;
6049 6049
6050 6050 zcmn_err(zoneid, CE_NOTE, "%s: dumping cmpd", where);
6051 6051 for (i = 0; i < argcnt; i++) {
6052 6052 nfs_argop4 *op = &argbase[i];
6053 6053 switch (op->argop) {
6054 6054 case OP_CPUTFH:
6055 6055 case OP_PUTFH:
6056 6056 zcmn_err(zoneid, CE_NOTE, "\t op %d, putfh", i);
6057 6057 break;
6058 6058 case OP_PUTROOTFH:
6059 6059 zcmn_err(zoneid, CE_NOTE, "\t op %d, putrootfh", i);
6060 6060 break;
6061 6061 case OP_CLOOKUP:
6062 6062 s = op->nfs_argop4_u.opclookup.cname;
6063 6063 zcmn_err(zoneid, CE_NOTE, "\t op %d, lookup %s", i, s);
6064 6064 break;
6065 6065 case OP_LOOKUP:
6066 6066 s = utf8_to_str(&op->nfs_argop4_u.oplookup.objname,
6067 6067 &len, NULL);
6068 6068 zcmn_err(zoneid, CE_NOTE, "\t op %d, lookup %s", i, s);
6069 6069 kmem_free(s, len);
6070 6070 break;
6071 6071 case OP_LOOKUPP:
6072 6072 zcmn_err(zoneid, CE_NOTE, "\t op %d, lookupp ..", i);
6073 6073 break;
6074 6074 case OP_GETFH:
6075 6075 zcmn_err(zoneid, CE_NOTE, "\t op %d, getfh", i);
6076 6076 break;
6077 6077 case OP_GETATTR:
6078 6078 zcmn_err(zoneid, CE_NOTE, "\t op %d, getattr", i);
6079 6079 break;
6080 6080 case OP_OPENATTR:
6081 6081 zcmn_err(zoneid, CE_NOTE, "\t op %d, openattr", i);
6082 6082 break;
6083 6083 default:
6084 6084 zcmn_err(zoneid, CE_NOTE, "\t op %d, opcode %d", i,
6085 6085 op->argop);
6086 6086 break;
6087 6087 }
6088 6088 }
6089 6089 }
6090 6090 #endif
6091 6091
6092 6092 /*
6093 6093 * nfs4lookup_setup - constructs a multi-lookup compound request.
6094 6094 *
6095 6095 * Given the path "nm1/nm2/.../nmn", the following compound requests
6096 6096 * may be created:
6097 6097 *
6098 6098 * Note: Getfh is not be needed because filehandle attr is mandatory, but it
6099 6099 * is faster, for now.
6100 6100 *
6101 6101 * l4_getattrs indicates the type of compound requested.
6102 6102 *
6103 6103 * LKP4_NO_ATTRIBUTE - no attributes (used by secinfo):
6104 6104 *
6105 6105 * compound { Put*fh; Lookup {nm1}; Lookup {nm2}; ... Lookup {nmn} }
6106 6106 *
6107 6107 * total number of ops is n + 1.
6108 6108 *
6109 6109 * LKP4_LAST_NAMED_ATTR - multi-component path for a named
6110 6110 * attribute: create lookups plus one OPENATTR/GETFH/GETATTR
6111 6111 * before the last component, and only get attributes
6112 6112 * for the last component. Note that the second-to-last
6113 6113 * pathname component is XATTR_RPATH, which does NOT go
6114 6114 * over-the-wire as a lookup.
6115 6115 *
6116 6116 * compound { Put*fh; Lookup {nm1}; Lookup {nm2}; ... Lookup {nmn-2};
6117 6117 * Openattr; Getfh; Getattr; Lookup {nmn}; Getfh; Getattr }
6118 6118 *
6119 6119 * and total number of ops is n + 5.
6120 6120 *
6121 6121 * LKP4_LAST_ATTRDIR - multi-component path for the hidden named
6122 6122 * attribute directory: create lookups plus an OPENATTR
6123 6123 * replacing the last lookup. Note that the last pathname
6124 6124 * component is XATTR_RPATH, which does NOT go over-the-wire
6125 6125 * as a lookup.
6126 6126 *
6127 6127 * compound { Put*fh; Lookup {nm1}; Lookup {nm2}; ... Getfh; Getattr;
6128 6128 * Openattr; Getfh; Getattr }
6129 6129 *
6130 6130 * and total number of ops is n + 5.
6131 6131 *
6132 6132 * LKP4_ALL_ATTRIBUTES - create lookups and get attributes for intermediate
6133 6133 * nodes too.
6134 6134 *
6135 6135 * compound { Put*fh; Lookup {nm1}; Getfh; Getattr;
6136 6136 * Lookup {nm2}; ... Lookup {nmn}; Getfh; Getattr }
6137 6137 *
6138 6138 * and total number of ops is 3*n + 1.
6139 6139 *
6140 6140 * All cases: returns the index in the arg array of the final LOOKUP op, or
6141 6141 * -1 if no LOOKUPs were used.
6142 6142 */
6143 6143 int
6144 6144 nfs4lookup_setup(char *nm, lookup4_param_t *lookupargp, int needgetfh)
6145 6145 {
6146 6146 enum lkp4_attr_setup l4_getattrs = lookupargp->l4_getattrs;
6147 6147 nfs_argop4 *argbase, *argop;
6148 6148 int arglen, argcnt;
6149 6149 int n = 1; /* number of components */
6150 6150 int nga = 1; /* number of Getattr's in request */
6151 6151 char c = '\0', *s, *p;
6152 6152 int lookup_idx = -1;
6153 6153 int argoplist_size;
6154 6154
6155 6155 /* set lookuparg response result to 0 */
6156 6156 lookupargp->resp->status = NFS4_OK;
6157 6157
6158 6158 /* skip leading "/" or "." e.g. ".//./" if there is */
6159 6159 for (; ; nm++) {
6160 6160 if (*nm != '/' && *nm != '.')
6161 6161 break;
6162 6162
6163 6163 /* ".." is counted as 1 component */
6164 6164 if (*nm == '.' && *(nm + 1) != '/')
6165 6165 break;
6166 6166 }
6167 6167
6168 6168 /*
6169 6169 * Find n = number of components - nm must be null terminated
6170 6170 * Skip "." components.
6171 6171 */
6172 6172 if (*nm != '\0')
6173 6173 for (n = 1, s = nm; *s != '\0'; s++) {
6174 6174 if ((*s == '/') && (*(s + 1) != '/') &&
6175 6175 (*(s + 1) != '\0') &&
6176 6176 !(*(s + 1) == '.' && (*(s + 2) == '/' ||
6177 6177 *(s + 2) == '\0')))
6178 6178 n++;
6179 6179 }
6180 6180 else
6181 6181 n = 0;
6182 6182
6183 6183 /*
6184 6184 * nga is number of components that need Getfh+Getattr
6185 6185 */
6186 6186 switch (l4_getattrs) {
6187 6187 case LKP4_NO_ATTRIBUTES:
6188 6188 nga = 0;
6189 6189 break;
6190 6190 case LKP4_ALL_ATTRIBUTES:
6191 6191 nga = n;
6192 6192 /*
6193 6193 * Always have at least 1 getfh, getattr pair
6194 6194 */
6195 6195 if (nga == 0)
6196 6196 nga++;
6197 6197 break;
6198 6198 case LKP4_LAST_ATTRDIR:
6199 6199 case LKP4_LAST_NAMED_ATTR:
6200 6200 nga = n+1;
6201 6201 break;
6202 6202 }
6203 6203
6204 6204 /*
6205 6205 * If change to use the filehandle attr instead of getfh
6206 6206 * the following line can be deleted.
6207 6207 */
6208 6208 nga *= 2;
6209 6209
6210 6210 /*
6211 6211 * calculate number of ops in request as
6212 6212 * header + trailer + lookups + getattrs
6213 6213 */
6214 6214 arglen = lookupargp->header_len + lookupargp->trailer_len + n + nga;
6215 6215
6216 6216 argoplist_size = arglen * sizeof (nfs_argop4);
6217 6217 argop = argbase = kmem_alloc(argoplist_size, KM_SLEEP);
6218 6218 lookupargp->argsp->array = argop;
6219 6219
6220 6220 argcnt = lookupargp->header_len;
6221 6221 argop += argcnt;
6222 6222
6223 6223 /*
6224 6224 * loop and create a lookup op and possibly getattr/getfh for
6225 6225 * each component. Skip "." components.
6226 6226 */
6227 6227 for (s = nm; *s != '\0'; s = p) {
6228 6228 /*
6229 6229 * Set up a pathname struct for each component if needed
6230 6230 */
6231 6231 while (*s == '/')
6232 6232 s++;
6233 6233 if (*s == '\0')
6234 6234 break;
6235 6235
6236 6236 for (p = s; (*p != '/') && (*p != '\0'); p++)
6237 6237 ;
6238 6238 c = *p;
6239 6239 *p = '\0';
6240 6240
6241 6241 if (s[0] == '.' && s[1] == '\0') {
6242 6242 *p = c;
6243 6243 continue;
6244 6244 }
6245 6245 if (l4_getattrs == LKP4_LAST_ATTRDIR &&
6246 6246 strcmp(s, XATTR_RPATH) == 0) {
6247 6247 /* getfh XXX may not be needed in future */
6248 6248 argop->argop = OP_GETFH;
6249 6249 argop++;
6250 6250 argcnt++;
6251 6251
6252 6252 /* getattr */
6253 6253 argop->argop = OP_GETATTR;
6254 6254 argop->nfs_argop4_u.opgetattr.attr_request =
6255 6255 lookupargp->ga_bits;
6256 6256 argop->nfs_argop4_u.opgetattr.mi =
6257 6257 lookupargp->mi;
6258 6258 argop++;
6259 6259 argcnt++;
6260 6260
6261 6261 /* openattr */
6262 6262 argop->argop = OP_OPENATTR;
6263 6263 } else if (l4_getattrs == LKP4_LAST_NAMED_ATTR &&
6264 6264 strcmp(s, XATTR_RPATH) == 0) {
6265 6265 /* openattr */
6266 6266 argop->argop = OP_OPENATTR;
6267 6267 argop++;
6268 6268 argcnt++;
6269 6269
6270 6270 /* getfh XXX may not be needed in future */
6271 6271 argop->argop = OP_GETFH;
6272 6272 argop++;
6273 6273 argcnt++;
6274 6274
6275 6275 /* getattr */
6276 6276 argop->argop = OP_GETATTR;
6277 6277 argop->nfs_argop4_u.opgetattr.attr_request =
6278 6278 lookupargp->ga_bits;
6279 6279 argop->nfs_argop4_u.opgetattr.mi =
6280 6280 lookupargp->mi;
6281 6281 argop++;
6282 6282 argcnt++;
6283 6283 *p = c;
6284 6284 continue;
6285 6285 } else if (s[0] == '.' && s[1] == '.' && s[2] == '\0') {
6286 6286 /* lookupp */
6287 6287 argop->argop = OP_LOOKUPP;
6288 6288 } else {
6289 6289 /* lookup */
6290 6290 argop->argop = OP_LOOKUP;
6291 6291 (void) str_to_utf8(s,
6292 6292 &argop->nfs_argop4_u.oplookup.objname);
6293 6293 }
6294 6294 lookup_idx = argcnt;
6295 6295 argop++;
6296 6296 argcnt++;
6297 6297
6298 6298 *p = c;
6299 6299
6300 6300 if (l4_getattrs == LKP4_ALL_ATTRIBUTES) {
6301 6301 /* getfh XXX may not be needed in future */
6302 6302 argop->argop = OP_GETFH;
6303 6303 argop++;
6304 6304 argcnt++;
6305 6305
6306 6306 /* getattr */
6307 6307 argop->argop = OP_GETATTR;
6308 6308 argop->nfs_argop4_u.opgetattr.attr_request =
6309 6309 lookupargp->ga_bits;
6310 6310 argop->nfs_argop4_u.opgetattr.mi =
6311 6311 lookupargp->mi;
6312 6312 argop++;
6313 6313 argcnt++;
6314 6314 }
6315 6315 }
6316 6316
6317 6317 if ((l4_getattrs != LKP4_NO_ATTRIBUTES) &&
6318 6318 ((l4_getattrs != LKP4_ALL_ATTRIBUTES) || (lookup_idx < 0))) {
6319 6319 if (needgetfh) {
6320 6320 /* stick in a post-lookup getfh */
6321 6321 argop->argop = OP_GETFH;
6322 6322 argcnt++;
6323 6323 argop++;
6324 6324 }
6325 6325 /* post-lookup getattr */
6326 6326 argop->argop = OP_GETATTR;
6327 6327 argop->nfs_argop4_u.opgetattr.attr_request =
6328 6328 lookupargp->ga_bits;
6329 6329 argop->nfs_argop4_u.opgetattr.mi = lookupargp->mi;
6330 6330 argcnt++;
6331 6331 }
6332 6332 argcnt += lookupargp->trailer_len; /* actual op count */
6333 6333 lookupargp->argsp->array_len = argcnt;
6334 6334 lookupargp->arglen = arglen;
6335 6335
6336 6336 #ifdef DEBUG
6337 6337 if (nfs4_client_lookup_debug)
6338 6338 nfs4lookup_dump_compound("nfs4lookup_setup", argbase, argcnt);
6339 6339 #endif
6340 6340
6341 6341 return (lookup_idx);
6342 6342 }
6343 6343
6344 6344 static int
6345 6345 nfs4openattr(vnode_t *dvp, vnode_t **avp, int cflag, cred_t *cr)
6346 6346 {
6347 6347 COMPOUND4args_clnt args;
6348 6348 COMPOUND4res_clnt res;
6349 6349 GETFH4res *gf_res = NULL;
6350 6350 nfs_argop4 argop[4];
6351 6351 nfs_resop4 *resop = NULL;
6352 6352 nfs4_sharedfh_t *sfhp;
6353 6353 hrtime_t t;
6354 6354 nfs4_error_t e;
6355 6355
6356 6356 rnode4_t *drp;
6357 6357 int doqueue = 1;
6358 6358 vnode_t *vp;
6359 6359 int needrecov = 0;
6360 6360 nfs4_recov_state_t recov_state;
6361 6361
6362 6362 ASSERT(nfs_zone() == VTOMI4(dvp)->mi_zone);
6363 6363
6364 6364 *avp = NULL;
6365 6365 recov_state.rs_flags = 0;
6366 6366 recov_state.rs_num_retry_despite_err = 0;
6367 6367
6368 6368 recov_retry:
6369 6369 /* COMPOUND: putfh, openattr, getfh, getattr */
6370 6370 args.array_len = 4;
6371 6371 args.array = argop;
6372 6372 args.ctag = TAG_OPENATTR;
6373 6373
6374 6374 e.error = nfs4_start_op(VTOMI4(dvp), dvp, NULL, &recov_state);
6375 6375 if (e.error)
6376 6376 return (e.error);
6377 6377
6378 6378 drp = VTOR4(dvp);
6379 6379
6380 6380 /* putfh */
6381 6381 argop[0].argop = OP_CPUTFH;
6382 6382 argop[0].nfs_argop4_u.opcputfh.sfh = drp->r_fh;
6383 6383
6384 6384 /* openattr */
6385 6385 argop[1].argop = OP_OPENATTR;
6386 6386 argop[1].nfs_argop4_u.opopenattr.createdir = (cflag ? TRUE : FALSE);
6387 6387
6388 6388 /* getfh */
6389 6389 argop[2].argop = OP_GETFH;
6390 6390
6391 6391 /* getattr */
6392 6392 argop[3].argop = OP_GETATTR;
6393 6393 argop[3].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
6394 6394 argop[3].nfs_argop4_u.opgetattr.mi = VTOMI4(dvp);
6395 6395
6396 6396 NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE,
6397 6397 "nfs4openattr: %s call, drp %s", needrecov ? "recov" : "first",
6398 6398 rnode4info(drp)));
6399 6399
6400 6400 t = gethrtime();
6401 6401
6402 6402 rfs4call(VTOMI4(dvp), &args, &res, cr, &doqueue, 0, &e);
6403 6403
6404 6404 needrecov = nfs4_needs_recovery(&e, FALSE, dvp->v_vfsp);
6405 6405 if (needrecov) {
6406 6406 bool_t abort;
6407 6407
6408 6408 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
6409 6409 "nfs4openattr: initiating recovery\n"));
6410 6410
6411 6411 abort = nfs4_start_recovery(&e,
6412 6412 VTOMI4(dvp), dvp, NULL, NULL, NULL,
6413 6413 OP_OPENATTR, NULL, NULL, NULL);
6414 6414 nfs4_end_op(VTOMI4(dvp), dvp, NULL, &recov_state, needrecov);
6415 6415 if (!e.error) {
6416 6416 e.error = geterrno4(res.status);
6417 6417 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
6418 6418 }
6419 6419 if (abort == FALSE)
6420 6420 goto recov_retry;
6421 6421 return (e.error);
6422 6422 }
6423 6423
6424 6424 if (e.error) {
6425 6425 nfs4_end_op(VTOMI4(dvp), dvp, NULL, &recov_state, needrecov);
6426 6426 return (e.error);
6427 6427 }
6428 6428
6429 6429 if (res.status) {
6430 6430 /*
6431 6431 * If OTW errro is NOTSUPP, then it should be
6432 6432 * translated to EINVAL. All Solaris file system
6433 6433 * implementations return EINVAL to the syscall layer
6434 6434 * when the attrdir cannot be created due to an
6435 6435 * implementation restriction or noxattr mount option.
6436 6436 */
6437 6437 if (res.status == NFS4ERR_NOTSUPP) {
6438 6438 mutex_enter(&drp->r_statelock);
6439 6439 if (drp->r_xattr_dir)
6440 6440 VN_RELE(drp->r_xattr_dir);
6441 6441 VN_HOLD(NFS4_XATTR_DIR_NOTSUPP);
6442 6442 drp->r_xattr_dir = NFS4_XATTR_DIR_NOTSUPP;
6443 6443 mutex_exit(&drp->r_statelock);
6444 6444
6445 6445 e.error = EINVAL;
6446 6446 } else {
6447 6447 e.error = geterrno4(res.status);
6448 6448 }
6449 6449
6450 6450 if (e.error) {
6451 6451 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
6452 6452 nfs4_end_op(VTOMI4(dvp), dvp, NULL, &recov_state,
6453 6453 needrecov);
6454 6454 return (e.error);
6455 6455 }
6456 6456 }
6457 6457
6458 6458 resop = &res.array[0]; /* putfh res */
6459 6459 ASSERT(resop->nfs_resop4_u.opgetfh.status == NFS4_OK);
6460 6460
6461 6461 resop = &res.array[1]; /* openattr res */
6462 6462 ASSERT(resop->nfs_resop4_u.opopenattr.status == NFS4_OK);
6463 6463
6464 6464 resop = &res.array[2]; /* getfh res */
6465 6465 gf_res = &resop->nfs_resop4_u.opgetfh;
6466 6466 if (gf_res->object.nfs_fh4_len == 0) {
6467 6467 *avp = NULL;
6468 6468 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
6469 6469 nfs4_end_op(VTOMI4(dvp), dvp, NULL, &recov_state, needrecov);
6470 6470 return (ENOENT);
6471 6471 }
6472 6472
6473 6473 sfhp = sfh4_get(&gf_res->object, VTOMI4(dvp));
6474 6474 vp = makenfs4node(sfhp, &res.array[3].nfs_resop4_u.opgetattr.ga_res,
6475 6475 dvp->v_vfsp, t, cr, dvp,
6476 6476 fn_get(VTOSV(dvp)->sv_name, XATTR_RPATH, sfhp));
6477 6477 sfh4_rele(&sfhp);
6478 6478
6479 6479 if (e.error)
6480 6480 PURGE_ATTRCACHE4(vp);
6481 6481
6482 6482 mutex_enter(&vp->v_lock);
6483 6483 vp->v_flag |= V_XATTRDIR;
6484 6484 mutex_exit(&vp->v_lock);
6485 6485
6486 6486 *avp = vp;
6487 6487
6488 6488 mutex_enter(&drp->r_statelock);
6489 6489 if (drp->r_xattr_dir)
6490 6490 VN_RELE(drp->r_xattr_dir);
6491 6491 VN_HOLD(vp);
6492 6492 drp->r_xattr_dir = vp;
6493 6493
6494 6494 /*
6495 6495 * Invalidate pathconf4 cache because r_xattr_dir is no longer
6496 6496 * NULL. xattrs could be created at any time, and we have no
6497 6497 * way to update pc4_xattr_exists in the base object if/when
6498 6498 * it happens.
6499 6499 */
6500 6500 drp->r_pathconf.pc4_xattr_valid = 0;
6501 6501
6502 6502 mutex_exit(&drp->r_statelock);
6503 6503
6504 6504 nfs4_end_op(VTOMI4(dvp), dvp, NULL, &recov_state, needrecov);
6505 6505
6506 6506 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
6507 6507
6508 6508 return (0);
6509 6509 }
6510 6510
6511 6511 /* ARGSUSED */
6512 6512 static int
6513 6513 nfs4_create(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive,
6514 6514 int mode, vnode_t **vpp, cred_t *cr, int flags, caller_context_t *ct,
6515 6515 vsecattr_t *vsecp)
6516 6516 {
6517 6517 int error;
6518 6518 vnode_t *vp = NULL;
6519 6519 rnode4_t *rp;
6520 6520 struct vattr vattr;
6521 6521 rnode4_t *drp;
6522 6522 vnode_t *tempvp;
6523 6523 enum createmode4 createmode;
6524 6524 bool_t must_trunc = FALSE;
6525 6525 int truncating = 0;
6526 6526
6527 6527 if (nfs_zone() != VTOMI4(dvp)->mi_zone)
6528 6528 return (EPERM);
6529 6529 if (exclusive == EXCL && (dvp->v_flag & V_XATTRDIR)) {
6530 6530 return (EINVAL);
6531 6531 }
6532 6532
6533 6533 /* . and .. have special meaning in the protocol, reject them. */
6534 6534
6535 6535 if (nm[0] == '.' && (nm[1] == '\0' || (nm[1] == '.' && nm[2] == '\0')))
6536 6536 return (EISDIR);
6537 6537
6538 6538 drp = VTOR4(dvp);
6539 6539
6540 6540 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR4(dvp)))
6541 6541 return (EINTR);
6542 6542
6543 6543 top:
6544 6544 /*
6545 6545 * We make a copy of the attributes because the caller does not
6546 6546 * expect us to change what va points to.
6547 6547 */
6548 6548 vattr = *va;
6549 6549
6550 6550 /*
6551 6551 * If the pathname is "", then dvp is the root vnode of
6552 6552 * a remote file mounted over a local directory.
6553 6553 * All that needs to be done is access
6554 6554 * checking and truncation. Note that we avoid doing
6555 6555 * open w/ create because the parent directory might
6556 6556 * be in pseudo-fs and the open would fail.
6557 6557 */
6558 6558 if (*nm == '\0') {
6559 6559 error = 0;
6560 6560 VN_HOLD(dvp);
6561 6561 vp = dvp;
6562 6562 must_trunc = TRUE;
6563 6563 } else {
6564 6564 /*
6565 6565 * We need to go over the wire, just to be sure whether the
6566 6566 * file exists or not. Using the DNLC can be dangerous in
6567 6567 * this case when making a decision regarding existence.
6568 6568 */
6569 6569 error = nfs4lookup(dvp, nm, &vp, cr, 1);
6570 6570 }
6571 6571
6572 6572 if (exclusive)
6573 6573 createmode = EXCLUSIVE4;
6574 6574 else
6575 6575 createmode = GUARDED4;
6576 6576
6577 6577 /*
6578 6578 * error would be set if the file does not exist on the
6579 6579 * server, so lets go create it.
6580 6580 */
6581 6581 if (error) {
6582 6582 goto create_otw;
6583 6583 }
6584 6584
6585 6585 /*
6586 6586 * File does exist on the server
6587 6587 */
6588 6588 if (exclusive == EXCL)
6589 6589 error = EEXIST;
6590 6590 else if (vp->v_type == VDIR && (mode & VWRITE))
6591 6591 error = EISDIR;
6592 6592 else {
6593 6593 /*
6594 6594 * If vnode is a device, create special vnode.
6595 6595 */
6596 6596 if (ISVDEV(vp->v_type)) {
6597 6597 tempvp = vp;
6598 6598 vp = specvp(vp, vp->v_rdev, vp->v_type, cr);
6599 6599 VN_RELE(tempvp);
6600 6600 }
6601 6601 if (!(error = VOP_ACCESS(vp, mode, 0, cr, ct))) {
6602 6602 if ((vattr.va_mask & AT_SIZE) &&
6603 6603 vp->v_type == VREG) {
6604 6604 rp = VTOR4(vp);
6605 6605 /*
6606 6606 * Check here for large file handled
6607 6607 * by LF-unaware process (as
6608 6608 * ufs_create() does)
6609 6609 */
6610 6610 if (!(flags & FOFFMAX)) {
6611 6611 mutex_enter(&rp->r_statelock);
6612 6612 if (rp->r_size > MAXOFF32_T)
6613 6613 error = EOVERFLOW;
6614 6614 mutex_exit(&rp->r_statelock);
6615 6615 }
6616 6616
6617 6617 /* if error is set then we need to return */
6618 6618 if (error) {
6619 6619 nfs_rw_exit(&drp->r_rwlock);
6620 6620 VN_RELE(vp);
6621 6621 return (error);
6622 6622 }
6623 6623
6624 6624 if (must_trunc) {
6625 6625 vattr.va_mask = AT_SIZE;
6626 6626 error = nfs4setattr(vp, &vattr, 0, cr,
6627 6627 NULL);
6628 6628 } else {
6629 6629 /*
6630 6630 * we know we have a regular file that already
6631 6631 * exists and we may end up truncating the file
6632 6632 * as a result of the open_otw, so flush out
6633 6633 * any dirty pages for this file first.
6634 6634 */
6635 6635 if (nfs4_has_pages(vp) &&
6636 6636 ((rp->r_flags & R4DIRTY) ||
6637 6637 rp->r_count > 0 ||
6638 6638 rp->r_mapcnt > 0)) {
6639 6639 error = nfs4_putpage(vp,
6640 6640 (offset_t)0, 0, 0, cr, ct);
6641 6641 if (error && (error == ENOSPC ||
6642 6642 error == EDQUOT)) {
6643 6643 mutex_enter(
6644 6644 &rp->r_statelock);
6645 6645 if (!rp->r_error)
6646 6646 rp->r_error =
6647 6647 error;
6648 6648 mutex_exit(
6649 6649 &rp->r_statelock);
6650 6650 }
6651 6651 }
6652 6652 vattr.va_mask = (AT_SIZE |
6653 6653 AT_TYPE | AT_MODE);
6654 6654 vattr.va_type = VREG;
6655 6655 createmode = UNCHECKED4;
6656 6656 truncating = 1;
6657 6657 goto create_otw;
6658 6658 }
6659 6659 }
6660 6660 }
6661 6661 }
6662 6662 nfs_rw_exit(&drp->r_rwlock);
6663 6663 if (error) {
6664 6664 VN_RELE(vp);
6665 6665 } else {
6666 6666 vnode_t *tvp;
6667 6667 rnode4_t *trp;
6668 6668 tvp = vp;
6669 6669 if (vp->v_type == VREG) {
6670 6670 trp = VTOR4(vp);
6671 6671 if (IS_SHADOW(vp, trp))
6672 6672 tvp = RTOV4(trp);
6673 6673 }
6674 6674
6675 6675 if (must_trunc) {
6676 6676 /*
6677 6677 * existing file got truncated, notify.
6678 6678 */
6679 6679 vnevent_create(tvp, ct);
6680 6680 }
6681 6681
6682 6682 *vpp = vp;
6683 6683 }
6684 6684 return (error);
6685 6685
6686 6686 create_otw:
6687 6687 dnlc_remove(dvp, nm);
6688 6688
6689 6689 ASSERT(vattr.va_mask & AT_TYPE);
6690 6690
6691 6691 /*
6692 6692 * If not a regular file let nfs4mknod() handle it.
6693 6693 */
6694 6694 if (vattr.va_type != VREG) {
6695 6695 error = nfs4mknod(dvp, nm, &vattr, exclusive, mode, vpp, cr);
6696 6696 nfs_rw_exit(&drp->r_rwlock);
6697 6697 return (error);
6698 6698 }
6699 6699
6700 6700 /*
6701 6701 * It _is_ a regular file.
6702 6702 */
6703 6703 ASSERT(vattr.va_mask & AT_MODE);
6704 6704 if (MANDMODE(vattr.va_mode)) {
6705 6705 nfs_rw_exit(&drp->r_rwlock);
6706 6706 return (EACCES);
6707 6707 }
6708 6708
6709 6709 /*
6710 6710 * If this happens to be a mknod of a regular file, then flags will
6711 6711 * have neither FREAD or FWRITE. However, we must set at least one
6712 6712 * for the call to nfs4open_otw. If it's open(O_CREAT) driving
6713 6713 * nfs4_create, then either FREAD, FWRITE, or FRDWR has already been
6714 6714 * set (based on openmode specified by app).
6715 6715 */
6716 6716 if ((flags & (FREAD|FWRITE)) == 0)
6717 6717 flags |= (FREAD|FWRITE);
6718 6718
6719 6719 error = nfs4open_otw(dvp, nm, &vattr, vpp, cr, 1, flags, createmode, 0);
6720 6720
6721 6721 if (vp != NULL) {
6722 6722 /* if create was successful, throw away the file's pages */
6723 6723 if (!error && (vattr.va_mask & AT_SIZE))
6724 6724 nfs4_invalidate_pages(vp, (vattr.va_size & PAGEMASK),
6725 6725 cr);
6726 6726 /* release the lookup hold */
6727 6727 VN_RELE(vp);
6728 6728 vp = NULL;
6729 6729 }
6730 6730
6731 6731 /*
6732 6732 * validate that we opened a regular file. This handles a misbehaving
6733 6733 * server that returns an incorrect FH.
6734 6734 */
6735 6735 if ((error == 0) && *vpp && (*vpp)->v_type != VREG) {
6736 6736 error = EISDIR;
6737 6737 VN_RELE(*vpp);
6738 6738 }
6739 6739
6740 6740 /*
6741 6741 * If this is not an exclusive create, then the CREATE
6742 6742 * request will be made with the GUARDED mode set. This
6743 6743 * means that the server will return EEXIST if the file
6744 6744 * exists. The file could exist because of a retransmitted
6745 6745 * request. In this case, we recover by starting over and
6746 6746 * checking to see whether the file exists. This second
6747 6747 * time through it should and a CREATE request will not be
6748 6748 * sent.
6749 6749 *
6750 6750 * This handles the problem of a dangling CREATE request
6751 6751 * which contains attributes which indicate that the file
6752 6752 * should be truncated. This retransmitted request could
6753 6753 * possibly truncate valid data in the file if not caught
6754 6754 * by the duplicate request mechanism on the server or if
6755 6755 * not caught by other means. The scenario is:
6756 6756 *
6757 6757 * Client transmits CREATE request with size = 0
6758 6758 * Client times out, retransmits request.
6759 6759 * Response to the first request arrives from the server
6760 6760 * and the client proceeds on.
6761 6761 * Client writes data to the file.
6762 6762 * The server now processes retransmitted CREATE request
6763 6763 * and truncates file.
6764 6764 *
6765 6765 * The use of the GUARDED CREATE request prevents this from
6766 6766 * happening because the retransmitted CREATE would fail
6767 6767 * with EEXIST and would not truncate the file.
6768 6768 */
6769 6769 if (error == EEXIST && exclusive == NONEXCL) {
6770 6770 #ifdef DEBUG
6771 6771 nfs4_create_misses++;
6772 6772 #endif
6773 6773 goto top;
6774 6774 }
6775 6775 nfs_rw_exit(&drp->r_rwlock);
6776 6776 if (truncating && !error && *vpp) {
6777 6777 vnode_t *tvp;
6778 6778 rnode4_t *trp;
6779 6779 /*
6780 6780 * existing file got truncated, notify.
6781 6781 */
6782 6782 tvp = *vpp;
6783 6783 trp = VTOR4(tvp);
6784 6784 if (IS_SHADOW(tvp, trp))
6785 6785 tvp = RTOV4(trp);
6786 6786 vnevent_create(tvp, ct);
6787 6787 }
6788 6788 return (error);
6789 6789 }
6790 6790
6791 6791 /*
6792 6792 * Create compound (for mkdir, mknod, symlink):
6793 6793 * { Putfh <dfh>; Create; Getfh; Getattr }
6794 6794 * It's okay if setattr failed to set gid - this is not considered
6795 6795 * an error, but purge attrs in that case.
6796 6796 */
6797 6797 static int
6798 6798 call_nfs4_create_req(vnode_t *dvp, char *nm, void *data, struct vattr *va,
6799 6799 vnode_t **vpp, cred_t *cr, nfs_ftype4 type)
6800 6800 {
6801 6801 int need_end_op = FALSE;
6802 6802 COMPOUND4args_clnt args;
6803 6803 COMPOUND4res_clnt res, *resp = NULL;
6804 6804 nfs_argop4 *argop;
6805 6805 nfs_resop4 *resop;
6806 6806 int doqueue;
6807 6807 mntinfo4_t *mi;
6808 6808 rnode4_t *drp = VTOR4(dvp);
6809 6809 change_info4 *cinfo;
6810 6810 GETFH4res *gf_res;
6811 6811 struct vattr vattr;
6812 6812 vnode_t *vp;
6813 6813 fattr4 *crattr;
6814 6814 bool_t needrecov = FALSE;
6815 6815 nfs4_recov_state_t recov_state;
6816 6816 nfs4_sharedfh_t *sfhp = NULL;
6817 6817 hrtime_t t;
6818 6818 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
6819 6819 int numops, argoplist_size, setgid_flag, idx_create, idx_fattr;
6820 6820 dirattr_info_t dinfo, *dinfop;
6821 6821 servinfo4_t *svp;
6822 6822 bitmap4 supp_attrs;
6823 6823
6824 6824 ASSERT(type == NF4DIR || type == NF4LNK || type == NF4BLK ||
6825 6825 type == NF4CHR || type == NF4SOCK || type == NF4FIFO);
6826 6826
6827 6827 mi = VTOMI4(dvp);
6828 6828
6829 6829 /*
6830 6830 * Make sure we properly deal with setting the right gid
6831 6831 * on a new directory to reflect the parent's setgid bit
6832 6832 */
6833 6833 setgid_flag = 0;
6834 6834 if (type == NF4DIR) {
6835 6835 struct vattr dva;
6836 6836
6837 6837 va->va_mode &= ~VSGID;
6838 6838 dva.va_mask = AT_MODE | AT_GID;
6839 6839 if (VOP_GETATTR(dvp, &dva, 0, cr, NULL) == 0) {
6840 6840
6841 6841 /*
6842 6842 * If the parent's directory has the setgid bit set
6843 6843 * _and_ the client was able to get a valid mapping
6844 6844 * for the parent dir's owner_group, we want to
6845 6845 * append NVERIFY(owner_group == dva.va_gid) and
6846 6846 * SETTATTR to the CREATE compound.
6847 6847 */
6848 6848 if (mi->mi_flags & MI4_GRPID || dva.va_mode & VSGID) {
6849 6849 setgid_flag = 1;
6850 6850 va->va_mode |= VSGID;
6851 6851 if (dva.va_gid != GID_NOBODY) {
6852 6852 va->va_mask |= AT_GID;
6853 6853 va->va_gid = dva.va_gid;
6854 6854 }
6855 6855 }
6856 6856 }
6857 6857 }
6858 6858
6859 6859 /*
6860 6860 * Create ops:
6861 6861 * 0:putfh(dir) 1:savefh(dir) 2:create 3:getfh(new) 4:getattr(new)
6862 6862 * 5:restorefh(dir) 6:getattr(dir)
6863 6863 *
6864 6864 * if (setgid)
6865 6865 * 0:putfh(dir) 1:create 2:getfh(new) 3:getattr(new)
6866 6866 * 4:savefh(new) 5:putfh(dir) 6:getattr(dir) 7:restorefh(new)
6867 6867 * 8:nverify 9:setattr
6868 6868 */
6869 6869 if (setgid_flag) {
6870 6870 numops = 10;
6871 6871 idx_create = 1;
6872 6872 idx_fattr = 3;
6873 6873 } else {
6874 6874 numops = 7;
6875 6875 idx_create = 2;
6876 6876 idx_fattr = 4;
6877 6877 }
6878 6878
6879 6879 ASSERT(nfs_zone() == mi->mi_zone);
6880 6880 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR4(dvp))) {
6881 6881 return (EINTR);
6882 6882 }
6883 6883 recov_state.rs_flags = 0;
6884 6884 recov_state.rs_num_retry_despite_err = 0;
6885 6885
6886 6886 argoplist_size = numops * sizeof (nfs_argop4);
6887 6887 argop = kmem_alloc(argoplist_size, KM_SLEEP);
6888 6888
6889 6889 recov_retry:
6890 6890 if (type == NF4LNK)
6891 6891 args.ctag = TAG_SYMLINK;
6892 6892 else if (type == NF4DIR)
6893 6893 args.ctag = TAG_MKDIR;
6894 6894 else
6895 6895 args.ctag = TAG_MKNOD;
6896 6896
6897 6897 args.array_len = numops;
6898 6898 args.array = argop;
6899 6899
6900 6900 if (e.error = nfs4_start_op(mi, dvp, NULL, &recov_state)) {
6901 6901 nfs_rw_exit(&drp->r_rwlock);
6902 6902 kmem_free(argop, argoplist_size);
6903 6903 return (e.error);
6904 6904 }
6905 6905 need_end_op = TRUE;
6906 6906
6907 6907
6908 6908 /* 0: putfh directory */
6909 6909 argop[0].argop = OP_CPUTFH;
6910 6910 argop[0].nfs_argop4_u.opcputfh.sfh = drp->r_fh;
6911 6911
6912 6912 /* 1/2: Create object */
6913 6913 argop[idx_create].argop = OP_CCREATE;
6914 6914 argop[idx_create].nfs_argop4_u.opccreate.cname = nm;
6915 6915 argop[idx_create].nfs_argop4_u.opccreate.type = type;
6916 6916 if (type == NF4LNK) {
6917 6917 /*
6918 6918 * symlink, treat name as data
6919 6919 */
6920 6920 ASSERT(data != NULL);
6921 6921 argop[idx_create].nfs_argop4_u.opccreate.ftype4_u.clinkdata =
6922 6922 (char *)data;
6923 6923 }
6924 6924 if (type == NF4BLK || type == NF4CHR) {
6925 6925 ASSERT(data != NULL);
6926 6926 argop[idx_create].nfs_argop4_u.opccreate.ftype4_u.devdata =
6927 6927 *((specdata4 *)data);
6928 6928 }
6929 6929
6930 6930 crattr = &argop[idx_create].nfs_argop4_u.opccreate.createattrs;
6931 6931
6932 6932 svp = drp->r_server;
6933 6933 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
6934 6934 supp_attrs = svp->sv_supp_attrs;
6935 6935 nfs_rw_exit(&svp->sv_lock);
6936 6936
6937 6937 if (vattr_to_fattr4(va, NULL, crattr, 0, OP_CREATE, supp_attrs)) {
6938 6938 nfs_rw_exit(&drp->r_rwlock);
6939 6939 nfs4_end_op(mi, dvp, NULL, &recov_state, needrecov);
6940 6940 e.error = EINVAL;
6941 6941 kmem_free(argop, argoplist_size);
6942 6942 return (e.error);
6943 6943 }
6944 6944
6945 6945 /* 2/3: getfh fh of created object */
6946 6946 ASSERT(idx_create + 1 == idx_fattr - 1);
6947 6947 argop[idx_create + 1].argop = OP_GETFH;
6948 6948
6949 6949 /* 3/4: getattr of new object */
6950 6950 argop[idx_fattr].argop = OP_GETATTR;
6951 6951 argop[idx_fattr].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
6952 6952 argop[idx_fattr].nfs_argop4_u.opgetattr.mi = mi;
6953 6953
6954 6954 if (setgid_flag) {
6955 6955 vattr_t _v;
6956 6956
6957 6957 argop[4].argop = OP_SAVEFH;
6958 6958
6959 6959 argop[5].argop = OP_CPUTFH;
6960 6960 argop[5].nfs_argop4_u.opcputfh.sfh = drp->r_fh;
6961 6961
6962 6962 argop[6].argop = OP_GETATTR;
6963 6963 argop[6].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
6964 6964 argop[6].nfs_argop4_u.opgetattr.mi = mi;
6965 6965
6966 6966 argop[7].argop = OP_RESTOREFH;
6967 6967
6968 6968 /*
6969 6969 * nverify
6970 6970 *
6971 6971 * XXX - Revisit the last argument to nfs4_end_op()
6972 6972 * once 5020486 is fixed.
6973 6973 */
6974 6974 _v.va_mask = AT_GID;
6975 6975 _v.va_gid = va->va_gid;
6976 6976 if (e.error = nfs4args_verify(&argop[8], &_v, OP_NVERIFY,
6977 6977 supp_attrs)) {
6978 6978 nfs4_end_op(mi, dvp, *vpp, &recov_state, TRUE);
6979 6979 nfs_rw_exit(&drp->r_rwlock);
6980 6980 nfs4_fattr4_free(crattr);
6981 6981 kmem_free(argop, argoplist_size);
6982 6982 return (e.error);
6983 6983 }
6984 6984
6985 6985 /*
6986 6986 * setattr
6987 6987 *
6988 6988 * We _know_ we're not messing with AT_SIZE or AT_XTIME,
6989 6989 * so no need for stateid or flags. Also we specify NULL
6990 6990 * rp since we're only interested in setting owner_group
6991 6991 * attributes.
6992 6992 */
6993 6993 nfs4args_setattr(&argop[9], &_v, NULL, 0, NULL, cr, supp_attrs,
6994 6994 &e.error, 0);
6995 6995
6996 6996 if (e.error) {
6997 6997 nfs4_end_op(mi, dvp, *vpp, &recov_state, TRUE);
6998 6998 nfs_rw_exit(&drp->r_rwlock);
6999 6999 nfs4_fattr4_free(crattr);
7000 7000 nfs4args_verify_free(&argop[8]);
7001 7001 kmem_free(argop, argoplist_size);
7002 7002 return (e.error);
7003 7003 }
7004 7004 } else {
7005 7005 argop[1].argop = OP_SAVEFH;
7006 7006
7007 7007 argop[5].argop = OP_RESTOREFH;
7008 7008
7009 7009 argop[6].argop = OP_GETATTR;
7010 7010 argop[6].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
7011 7011 argop[6].nfs_argop4_u.opgetattr.mi = mi;
7012 7012 }
7013 7013
7014 7014 dnlc_remove(dvp, nm);
7015 7015
7016 7016 doqueue = 1;
7017 7017 t = gethrtime();
7018 7018 rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
7019 7019
7020 7020 needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
7021 7021 if (e.error) {
7022 7022 PURGE_ATTRCACHE4(dvp);
7023 7023 if (!needrecov)
7024 7024 goto out;
7025 7025 }
7026 7026
7027 7027 if (needrecov) {
7028 7028 if (nfs4_start_recovery(&e, mi, dvp, NULL, NULL, NULL,
7029 7029 OP_CREATE, NULL, NULL, NULL) == FALSE) {
7030 7030 nfs4_end_op(mi, dvp, NULL, &recov_state,
7031 7031 needrecov);
7032 7032 need_end_op = FALSE;
7033 7033 nfs4_fattr4_free(crattr);
7034 7034 if (setgid_flag) {
7035 7035 nfs4args_verify_free(&argop[8]);
7036 7036 nfs4args_setattr_free(&argop[9]);
7037 7037 }
7038 7038 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
7039 7039 goto recov_retry;
7040 7040 }
7041 7041 }
7042 7042
7043 7043 resp = &res;
7044 7044
7045 7045 if (res.status != NFS4_OK && res.array_len <= idx_fattr + 1) {
7046 7046
7047 7047 if (res.status == NFS4ERR_BADOWNER)
7048 7048 nfs4_log_badowner(mi, OP_CREATE);
7049 7049
7050 7050 e.error = geterrno4(res.status);
7051 7051
7052 7052 /*
7053 7053 * This check is left over from when create was implemented
7054 7054 * using a setattr op (instead of createattrs). If the
7055 7055 * putfh/create/getfh failed, the error was returned. If
7056 7056 * setattr/getattr failed, we keep going.
7057 7057 *
7058 7058 * It might be better to get rid of the GETFH also, and just
7059 7059 * do PUTFH/CREATE/GETATTR since the FH attr is mandatory.
7060 7060 * Then if any of the operations failed, we could return the
7061 7061 * error now, and remove much of the error code below.
7062 7062 */
7063 7063 if (res.array_len <= idx_fattr) {
7064 7064 /*
7065 7065 * Either Putfh, Create or Getfh failed.
7066 7066 */
7067 7067 PURGE_ATTRCACHE4(dvp);
7068 7068 /*
7069 7069 * nfs4_purge_stale_fh() may generate otw calls through
7070 7070 * nfs4_invalidate_pages. Hence the need to call
7071 7071 * nfs4_end_op() here to avoid nfs4_start_op() deadlock.
7072 7072 */
7073 7073 nfs4_end_op(mi, dvp, NULL, &recov_state,
7074 7074 needrecov);
7075 7075 need_end_op = FALSE;
7076 7076 nfs4_purge_stale_fh(e.error, dvp, cr);
7077 7077 goto out;
7078 7078 }
7079 7079 }
7080 7080
7081 7081 resop = &res.array[idx_create]; /* create res */
7082 7082 cinfo = &resop->nfs_resop4_u.opcreate.cinfo;
7083 7083
7084 7084 resop = &res.array[idx_create + 1]; /* getfh res */
7085 7085 gf_res = &resop->nfs_resop4_u.opgetfh;
7086 7086
7087 7087 sfhp = sfh4_get(&gf_res->object, mi);
7088 7088 if (e.error) {
7089 7089 *vpp = vp = makenfs4node(sfhp, NULL, dvp->v_vfsp, t, cr, dvp,
7090 7090 fn_get(VTOSV(dvp)->sv_name, nm, sfhp));
7091 7091 if (vp->v_type == VNON) {
7092 7092 vattr.va_mask = AT_TYPE;
7093 7093 /*
7094 7094 * Need to call nfs4_end_op before nfs4getattr to avoid
7095 7095 * potential nfs4_start_op deadlock. See RFE 4777612.
7096 7096 */
7097 7097 nfs4_end_op(mi, dvp, NULL, &recov_state,
7098 7098 needrecov);
7099 7099 need_end_op = FALSE;
7100 7100 e.error = nfs4getattr(vp, &vattr, cr);
7101 7101 if (e.error) {
7102 7102 VN_RELE(vp);
7103 7103 *vpp = NULL;
7104 7104 goto out;
7105 7105 }
7106 7106 vp->v_type = vattr.va_type;
7107 7107 }
7108 7108 e.error = 0;
7109 7109 } else {
7110 7110 *vpp = vp = makenfs4node(sfhp,
7111 7111 &res.array[idx_fattr].nfs_resop4_u.opgetattr.ga_res,
7112 7112 dvp->v_vfsp, t, cr,
7113 7113 dvp, fn_get(VTOSV(dvp)->sv_name, nm, sfhp));
7114 7114 }
7115 7115
7116 7116 /*
7117 7117 * If compound succeeded, then update dir attrs
7118 7118 */
7119 7119 if (res.status == NFS4_OK) {
7120 7120 dinfo.di_garp = &res.array[6].nfs_resop4_u.opgetattr.ga_res;
7121 7121 dinfo.di_cred = cr;
7122 7122 dinfo.di_time_call = t;
7123 7123 dinfop = &dinfo;
7124 7124 } else
7125 7125 dinfop = NULL;
7126 7126
7127 7127 /* Update directory cache attribute, readdir and dnlc caches */
7128 7128 nfs4_update_dircaches(cinfo, dvp, vp, nm, dinfop);
7129 7129
7130 7130 out:
7131 7131 if (sfhp != NULL)
7132 7132 sfh4_rele(&sfhp);
7133 7133 nfs_rw_exit(&drp->r_rwlock);
7134 7134 nfs4_fattr4_free(crattr);
7135 7135 if (setgid_flag) {
7136 7136 nfs4args_verify_free(&argop[8]);
7137 7137 nfs4args_setattr_free(&argop[9]);
7138 7138 }
7139 7139 if (resp)
7140 7140 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)resp);
7141 7141 if (need_end_op)
7142 7142 nfs4_end_op(mi, dvp, NULL, &recov_state, needrecov);
7143 7143
7144 7144 kmem_free(argop, argoplist_size);
7145 7145 return (e.error);
7146 7146 }
7147 7147
7148 7148 /* ARGSUSED */
7149 7149 static int
7150 7150 nfs4mknod(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive,
7151 7151 int mode, vnode_t **vpp, cred_t *cr)
7152 7152 {
7153 7153 int error;
7154 7154 vnode_t *vp;
7155 7155 nfs_ftype4 type;
7156 7156 specdata4 spec, *specp = NULL;
7157 7157
7158 7158 ASSERT(nfs_zone() == VTOMI4(dvp)->mi_zone);
7159 7159
7160 7160 switch (va->va_type) {
7161 7161 case VCHR:
7162 7162 case VBLK:
7163 7163 type = (va->va_type == VCHR) ? NF4CHR : NF4BLK;
7164 7164 spec.specdata1 = getmajor(va->va_rdev);
7165 7165 spec.specdata2 = getminor(va->va_rdev);
7166 7166 specp = &spec;
7167 7167 break;
7168 7168
7169 7169 case VFIFO:
7170 7170 type = NF4FIFO;
7171 7171 break;
7172 7172 case VSOCK:
7173 7173 type = NF4SOCK;
7174 7174 break;
7175 7175
7176 7176 default:
7177 7177 return (EINVAL);
7178 7178 }
7179 7179
7180 7180 error = call_nfs4_create_req(dvp, nm, specp, va, &vp, cr, type);
7181 7181 if (error) {
7182 7182 return (error);
7183 7183 }
7184 7184
7185 7185 /*
7186 7186 * This might not be needed any more; special case to deal
7187 7187 * with problematic v2/v3 servers. Since create was unable
7188 7188 * to set group correctly, not sure what hope setattr has.
7189 7189 */
7190 7190 if (va->va_gid != VTOR4(vp)->r_attr.va_gid) {
7191 7191 va->va_mask = AT_GID;
7192 7192 (void) nfs4setattr(vp, va, 0, cr, NULL);
7193 7193 }
7194 7194
7195 7195 /*
7196 7196 * If vnode is a device create special vnode
7197 7197 */
7198 7198 if (ISVDEV(vp->v_type)) {
7199 7199 *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr);
7200 7200 VN_RELE(vp);
7201 7201 } else {
7202 7202 *vpp = vp;
7203 7203 }
7204 7204 return (error);
7205 7205 }
7206 7206
7207 7207 /*
7208 7208 * Remove requires that the current fh be the target directory.
7209 7209 * After the operation, the current fh is unchanged.
7210 7210 * The compound op structure is:
7211 7211 * PUTFH(targetdir), REMOVE
7212 7212 *
7213 7213 * Weirdness: if the vnode to be removed is open
7214 7214 * we rename it instead of removing it and nfs_inactive
7215 7215 * will remove the new name.
7216 7216 */
7217 7217 /* ARGSUSED */
7218 7218 static int
7219 7219 nfs4_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct, int flags)
7220 7220 {
7221 7221 COMPOUND4args_clnt args;
7222 7222 COMPOUND4res_clnt res, *resp = NULL;
7223 7223 REMOVE4res *rm_res;
7224 7224 nfs_argop4 argop[3];
7225 7225 nfs_resop4 *resop;
7226 7226 vnode_t *vp;
7227 7227 char *tmpname;
7228 7228 int doqueue;
7229 7229 mntinfo4_t *mi;
7230 7230 rnode4_t *rp;
7231 7231 rnode4_t *drp;
7232 7232 int needrecov = 0;
7233 7233 nfs4_recov_state_t recov_state;
7234 7234 int isopen;
7235 7235 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
7236 7236 dirattr_info_t dinfo;
7237 7237
7238 7238 if (nfs_zone() != VTOMI4(dvp)->mi_zone)
7239 7239 return (EPERM);
7240 7240 drp = VTOR4(dvp);
7241 7241 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR4(dvp)))
7242 7242 return (EINTR);
7243 7243
7244 7244 e.error = nfs4lookup(dvp, nm, &vp, cr, 0);
7245 7245 if (e.error) {
7246 7246 nfs_rw_exit(&drp->r_rwlock);
7247 7247 return (e.error);
7248 7248 }
7249 7249
7250 7250 if (vp->v_type == VDIR) {
7251 7251 VN_RELE(vp);
7252 7252 nfs_rw_exit(&drp->r_rwlock);
7253 7253 return (EISDIR);
7254 7254 }
7255 7255
7256 7256 /*
7257 7257 * First just remove the entry from the name cache, as it
7258 7258 * is most likely the only entry for this vp.
7259 7259 */
7260 7260 dnlc_remove(dvp, nm);
7261 7261
7262 7262 rp = VTOR4(vp);
7263 7263
7264 7264 /*
7265 7265 * For regular file types, check to see if the file is open by looking
7266 7266 * at the open streams.
7267 7267 * For all other types, check the reference count on the vnode. Since
7268 7268 * they are not opened OTW they never have an open stream.
7269 7269 *
7270 7270 * If the file is open, rename it to .nfsXXXX.
7271 7271 */
7272 7272 if (vp->v_type != VREG) {
7273 7273 /*
7274 7274 * If the file has a v_count > 1 then there may be more than one
7275 7275 * entry in the name cache due multiple links or an open file,
7276 7276 * but we don't have the real reference count so flush all
7277 7277 * possible entries.
7278 7278 */
7279 7279 if (vp->v_count > 1)
7280 7280 dnlc_purge_vp(vp);
7281 7281
7282 7282 /*
7283 7283 * Now we have the real reference count.
7284 7284 */
7285 7285 isopen = vp->v_count > 1;
7286 7286 } else {
7287 7287 mutex_enter(&rp->r_os_lock);
7288 7288 isopen = list_head(&rp->r_open_streams) != NULL;
7289 7289 mutex_exit(&rp->r_os_lock);
7290 7290 }
7291 7291
7292 7292 mutex_enter(&rp->r_statelock);
7293 7293 if (isopen &&
7294 7294 (rp->r_unldvp == NULL || strcmp(nm, rp->r_unlname) == 0)) {
7295 7295 mutex_exit(&rp->r_statelock);
7296 7296 tmpname = newname();
7297 7297 e.error = nfs4rename(dvp, nm, dvp, tmpname, cr, ct);
7298 7298 if (e.error)
7299 7299 kmem_free(tmpname, MAXNAMELEN);
7300 7300 else {
7301 7301 mutex_enter(&rp->r_statelock);
7302 7302 if (rp->r_unldvp == NULL) {
7303 7303 VN_HOLD(dvp);
7304 7304 rp->r_unldvp = dvp;
7305 7305 if (rp->r_unlcred != NULL)
7306 7306 crfree(rp->r_unlcred);
7307 7307 crhold(cr);
7308 7308 rp->r_unlcred = cr;
7309 7309 rp->r_unlname = tmpname;
7310 7310 } else {
7311 7311 kmem_free(rp->r_unlname, MAXNAMELEN);
7312 7312 rp->r_unlname = tmpname;
7313 7313 }
7314 7314 mutex_exit(&rp->r_statelock);
7315 7315 }
7316 7316 VN_RELE(vp);
7317 7317 nfs_rw_exit(&drp->r_rwlock);
7318 7318 return (e.error);
7319 7319 }
7320 7320 /*
7321 7321 * Actually remove the file/dir
7322 7322 */
7323 7323 mutex_exit(&rp->r_statelock);
7324 7324
7325 7325 /*
7326 7326 * We need to flush any dirty pages which happen to
7327 7327 * be hanging around before removing the file.
7328 7328 * This shouldn't happen very often since in NFSv4
7329 7329 * we should be close to open consistent.
7330 7330 */
7331 7331 if (nfs4_has_pages(vp) &&
7332 7332 ((rp->r_flags & R4DIRTY) || rp->r_count > 0)) {
7333 7333 e.error = nfs4_putpage(vp, (u_offset_t)0, 0, 0, cr, ct);
7334 7334 if (e.error && (e.error == ENOSPC || e.error == EDQUOT)) {
7335 7335 mutex_enter(&rp->r_statelock);
7336 7336 if (!rp->r_error)
7337 7337 rp->r_error = e.error;
7338 7338 mutex_exit(&rp->r_statelock);
7339 7339 }
7340 7340 }
7341 7341
7342 7342 mi = VTOMI4(dvp);
7343 7343
7344 7344 (void) nfs4delegreturn(rp, NFS4_DR_REOPEN);
7345 7345 recov_state.rs_flags = 0;
7346 7346 recov_state.rs_num_retry_despite_err = 0;
7347 7347
7348 7348 recov_retry:
7349 7349 /*
7350 7350 * Remove ops: putfh dir; remove
7351 7351 */
7352 7352 args.ctag = TAG_REMOVE;
7353 7353 args.array_len = 3;
7354 7354 args.array = argop;
7355 7355
7356 7356 e.error = nfs4_start_op(VTOMI4(dvp), dvp, NULL, &recov_state);
7357 7357 if (e.error) {
7358 7358 nfs_rw_exit(&drp->r_rwlock);
7359 7359 VN_RELE(vp);
7360 7360 return (e.error);
7361 7361 }
7362 7362
7363 7363 /* putfh directory */
7364 7364 argop[0].argop = OP_CPUTFH;
7365 7365 argop[0].nfs_argop4_u.opcputfh.sfh = drp->r_fh;
7366 7366
7367 7367 /* remove */
7368 7368 argop[1].argop = OP_CREMOVE;
7369 7369 argop[1].nfs_argop4_u.opcremove.ctarget = nm;
7370 7370
7371 7371 /* getattr dir */
7372 7372 argop[2].argop = OP_GETATTR;
7373 7373 argop[2].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
7374 7374 argop[2].nfs_argop4_u.opgetattr.mi = mi;
7375 7375
7376 7376 doqueue = 1;
7377 7377 dinfo.di_time_call = gethrtime();
7378 7378 rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
7379 7379
7380 7380 PURGE_ATTRCACHE4(vp);
7381 7381
7382 7382 needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
7383 7383 if (e.error)
7384 7384 PURGE_ATTRCACHE4(dvp);
7385 7385
7386 7386 if (needrecov) {
7387 7387 if (nfs4_start_recovery(&e, VTOMI4(dvp), dvp,
7388 7388 NULL, NULL, NULL, OP_REMOVE, NULL, NULL, NULL) == FALSE) {
7389 7389 if (!e.error)
7390 7390 (void) xdr_free(xdr_COMPOUND4res_clnt,
7391 7391 (caddr_t)&res);
7392 7392 nfs4_end_op(VTOMI4(dvp), dvp, NULL, &recov_state,
7393 7393 needrecov);
7394 7394 goto recov_retry;
7395 7395 }
7396 7396 }
7397 7397
7398 7398 /*
7399 7399 * Matching nfs4_end_op() for start_op() above.
7400 7400 * There is a path in the code below which calls
7401 7401 * nfs4_purge_stale_fh(), which may generate otw calls through
7402 7402 * nfs4_invalidate_pages. Hence we need to call nfs4_end_op()
7403 7403 * here to avoid nfs4_start_op() deadlock.
7404 7404 */
7405 7405 nfs4_end_op(VTOMI4(dvp), dvp, NULL, &recov_state, needrecov);
7406 7406
7407 7407 if (!e.error) {
7408 7408 resp = &res;
7409 7409
7410 7410 if (res.status) {
7411 7411 e.error = geterrno4(res.status);
7412 7412 PURGE_ATTRCACHE4(dvp);
7413 7413 nfs4_purge_stale_fh(e.error, dvp, cr);
7414 7414 } else {
7415 7415 resop = &res.array[1]; /* remove res */
7416 7416 rm_res = &resop->nfs_resop4_u.opremove;
7417 7417
7418 7418 dinfo.di_garp =
7419 7419 &res.array[2].nfs_resop4_u.opgetattr.ga_res;
7420 7420 dinfo.di_cred = cr;
7421 7421
7422 7422 /* Update directory attr, readdir and dnlc caches */
7423 7423 nfs4_update_dircaches(&rm_res->cinfo, dvp, NULL, NULL,
7424 7424 &dinfo);
7425 7425 }
7426 7426 }
7427 7427 nfs_rw_exit(&drp->r_rwlock);
7428 7428 if (resp)
7429 7429 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)resp);
7430 7430
7431 7431 if (e.error == 0) {
7432 7432 vnode_t *tvp;
7433 7433 rnode4_t *trp;
7434 7434 trp = VTOR4(vp);
7435 7435 tvp = vp;
7436 7436 if (IS_SHADOW(vp, trp))
7437 7437 tvp = RTOV4(trp);
7438 7438 vnevent_remove(tvp, dvp, nm, ct);
7439 7439 }
7440 7440 VN_RELE(vp);
7441 7441 return (e.error);
7442 7442 }
7443 7443
7444 7444 /*
7445 7445 * Link requires that the current fh be the target directory and the
7446 7446 * saved fh be the source fh. After the operation, the current fh is unchanged.
7447 7447 * Thus the compound op structure is:
7448 7448 * PUTFH(file), SAVEFH, PUTFH(targetdir), LINK, RESTOREFH,
7449 7449 * GETATTR(file)
7450 7450 */
7451 7451 /* ARGSUSED */
7452 7452 static int
7453 7453 nfs4_link(vnode_t *tdvp, vnode_t *svp, char *tnm, cred_t *cr,
7454 7454 caller_context_t *ct, int flags)
7455 7455 {
7456 7456 COMPOUND4args_clnt args;
7457 7457 COMPOUND4res_clnt res, *resp = NULL;
7458 7458 LINK4res *ln_res;
7459 7459 int argoplist_size = 7 * sizeof (nfs_argop4);
7460 7460 nfs_argop4 *argop;
7461 7461 nfs_resop4 *resop;
7462 7462 vnode_t *realvp, *nvp;
7463 7463 int doqueue;
7464 7464 mntinfo4_t *mi;
7465 7465 rnode4_t *tdrp;
7466 7466 bool_t needrecov = FALSE;
7467 7467 nfs4_recov_state_t recov_state;
7468 7468 hrtime_t t;
7469 7469 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
7470 7470 dirattr_info_t dinfo;
7471 7471
7472 7472 ASSERT(*tnm != '\0');
7473 7473 ASSERT(tdvp->v_type == VDIR);
7474 7474 ASSERT(nfs4_consistent_type(tdvp));
7475 7475 ASSERT(nfs4_consistent_type(svp));
7476 7476
7477 7477 if (nfs_zone() != VTOMI4(tdvp)->mi_zone)
7478 7478 return (EPERM);
7479 7479 if (VOP_REALVP(svp, &realvp, ct) == 0) {
7480 7480 svp = realvp;
7481 7481 ASSERT(nfs4_consistent_type(svp));
7482 7482 }
7483 7483
7484 7484 tdrp = VTOR4(tdvp);
7485 7485 mi = VTOMI4(svp);
7486 7486
7487 7487 if (!(mi->mi_flags & MI4_LINK)) {
7488 7488 return (EOPNOTSUPP);
7489 7489 }
7490 7490 recov_state.rs_flags = 0;
7491 7491 recov_state.rs_num_retry_despite_err = 0;
7492 7492
7493 7493 if (nfs_rw_enter_sig(&tdrp->r_rwlock, RW_WRITER, INTR4(tdvp)))
7494 7494 return (EINTR);
7495 7495
7496 7496 recov_retry:
7497 7497 argop = kmem_alloc(argoplist_size, KM_SLEEP);
7498 7498
7499 7499 args.ctag = TAG_LINK;
7500 7500
7501 7501 /*
7502 7502 * Link ops: putfh fl; savefh; putfh tdir; link; getattr(dir);
7503 7503 * restorefh; getattr(fl)
7504 7504 */
7505 7505 args.array_len = 7;
7506 7506 args.array = argop;
7507 7507
7508 7508 e.error = nfs4_start_op(VTOMI4(svp), svp, tdvp, &recov_state);
7509 7509 if (e.error) {
7510 7510 kmem_free(argop, argoplist_size);
7511 7511 nfs_rw_exit(&tdrp->r_rwlock);
7512 7512 return (e.error);
7513 7513 }
7514 7514
7515 7515 /* 0. putfh file */
7516 7516 argop[0].argop = OP_CPUTFH;
7517 7517 argop[0].nfs_argop4_u.opcputfh.sfh = VTOR4(svp)->r_fh;
7518 7518
7519 7519 /* 1. save current fh to free up the space for the dir */
7520 7520 argop[1].argop = OP_SAVEFH;
7521 7521
7522 7522 /* 2. putfh targetdir */
7523 7523 argop[2].argop = OP_CPUTFH;
7524 7524 argop[2].nfs_argop4_u.opcputfh.sfh = tdrp->r_fh;
7525 7525
7526 7526 /* 3. link: current_fh is targetdir, saved_fh is source */
7527 7527 argop[3].argop = OP_CLINK;
7528 7528 argop[3].nfs_argop4_u.opclink.cnewname = tnm;
7529 7529
7530 7530 /* 4. Get attributes of dir */
7531 7531 argop[4].argop = OP_GETATTR;
7532 7532 argop[4].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
7533 7533 argop[4].nfs_argop4_u.opgetattr.mi = mi;
7534 7534
7535 7535 /* 5. If link was successful, restore current vp to file */
7536 7536 argop[5].argop = OP_RESTOREFH;
7537 7537
7538 7538 /* 6. Get attributes of linked object */
7539 7539 argop[6].argop = OP_GETATTR;
7540 7540 argop[6].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
7541 7541 argop[6].nfs_argop4_u.opgetattr.mi = mi;
7542 7542
7543 7543 dnlc_remove(tdvp, tnm);
7544 7544
7545 7545 doqueue = 1;
7546 7546 t = gethrtime();
7547 7547
7548 7548 rfs4call(VTOMI4(svp), &args, &res, cr, &doqueue, 0, &e);
7549 7549
7550 7550 needrecov = nfs4_needs_recovery(&e, FALSE, svp->v_vfsp);
7551 7551 if (e.error != 0 && !needrecov) {
7552 7552 PURGE_ATTRCACHE4(tdvp);
7553 7553 PURGE_ATTRCACHE4(svp);
7554 7554 nfs4_end_op(VTOMI4(svp), svp, tdvp, &recov_state, needrecov);
7555 7555 goto out;
7556 7556 }
7557 7557
7558 7558 if (needrecov) {
7559 7559 bool_t abort;
7560 7560
7561 7561 abort = nfs4_start_recovery(&e, VTOMI4(svp), svp, tdvp,
7562 7562 NULL, NULL, OP_LINK, NULL, NULL, NULL);
7563 7563 if (abort == FALSE) {
7564 7564 nfs4_end_op(VTOMI4(svp), svp, tdvp, &recov_state,
7565 7565 needrecov);
7566 7566 kmem_free(argop, argoplist_size);
7567 7567 if (!e.error)
7568 7568 (void) xdr_free(xdr_COMPOUND4res_clnt,
7569 7569 (caddr_t)&res);
7570 7570 goto recov_retry;
7571 7571 } else {
7572 7572 if (e.error != 0) {
7573 7573 PURGE_ATTRCACHE4(tdvp);
7574 7574 PURGE_ATTRCACHE4(svp);
7575 7575 nfs4_end_op(VTOMI4(svp), svp, tdvp,
7576 7576 &recov_state, needrecov);
7577 7577 goto out;
7578 7578 }
7579 7579 /* fall through for res.status case */
7580 7580 }
7581 7581 }
7582 7582
7583 7583 nfs4_end_op(VTOMI4(svp), svp, tdvp, &recov_state, needrecov);
7584 7584
7585 7585 resp = &res;
7586 7586 if (res.status) {
7587 7587 /* If link succeeded, then don't return error */
7588 7588 e.error = geterrno4(res.status);
7589 7589 if (res.array_len <= 4) {
7590 7590 /*
7591 7591 * Either Putfh, Savefh, Putfh dir, or Link failed
7592 7592 */
7593 7593 PURGE_ATTRCACHE4(svp);
7594 7594 PURGE_ATTRCACHE4(tdvp);
7595 7595 if (e.error == EOPNOTSUPP) {
7596 7596 mutex_enter(&mi->mi_lock);
7597 7597 mi->mi_flags &= ~MI4_LINK;
7598 7598 mutex_exit(&mi->mi_lock);
7599 7599 }
7600 7600 /* Remap EISDIR to EPERM for non-root user for SVVS */
7601 7601 /* XXX-LP */
7602 7602 if (e.error == EISDIR && crgetuid(cr) != 0)
7603 7603 e.error = EPERM;
7604 7604 goto out;
7605 7605 }
7606 7606 }
7607 7607
7608 7608 /* either no error or one of the postop getattr failed */
7609 7609
7610 7610 /*
7611 7611 * XXX - if LINK succeeded, but no attrs were returned for link
7612 7612 * file, purge its cache.
7613 7613 *
7614 7614 * XXX Perform a simplified version of wcc checking. Instead of
7615 7615 * have another getattr to get pre-op, just purge cache if
7616 7616 * any of the ops prior to and including the getattr failed.
7617 7617 * If the getattr succeeded then update the attrcache accordingly.
7618 7618 */
7619 7619
7620 7620 /*
7621 7621 * update cache with link file postattrs.
7622 7622 * Note: at this point resop points to link res.
7623 7623 */
7624 7624 resop = &res.array[3]; /* link res */
7625 7625 ln_res = &resop->nfs_resop4_u.oplink;
7626 7626 if (res.status == NFS4_OK)
7627 7627 e.error = nfs4_update_attrcache(res.status,
7628 7628 &res.array[6].nfs_resop4_u.opgetattr.ga_res,
7629 7629 t, svp, cr);
7630 7630
7631 7631 /*
7632 7632 * Call makenfs4node to create the new shadow vp for tnm.
7633 7633 * We pass NULL attrs because we just cached attrs for
7634 7634 * the src object. All we're trying to accomplish is to
7635 7635 * to create the new shadow vnode.
7636 7636 */
7637 7637 nvp = makenfs4node(VTOR4(svp)->r_fh, NULL, tdvp->v_vfsp, t, cr,
7638 7638 tdvp, fn_get(VTOSV(tdvp)->sv_name, tnm, VTOR4(svp)->r_fh));
7639 7639
7640 7640 /* Update target cache attribute, readdir and dnlc caches */
7641 7641 dinfo.di_garp = &res.array[4].nfs_resop4_u.opgetattr.ga_res;
7642 7642 dinfo.di_time_call = t;
7643 7643 dinfo.di_cred = cr;
7644 7644
7645 7645 nfs4_update_dircaches(&ln_res->cinfo, tdvp, nvp, tnm, &dinfo);
7646 7646 ASSERT(nfs4_consistent_type(tdvp));
7647 7647 ASSERT(nfs4_consistent_type(svp));
7648 7648 ASSERT(nfs4_consistent_type(nvp));
7649 7649 VN_RELE(nvp);
7650 7650
7651 7651 if (!e.error) {
7652 7652 vnode_t *tvp;
7653 7653 rnode4_t *trp;
7654 7654 /*
7655 7655 * Notify the source file of this link operation.
7656 7656 */
7657 7657 trp = VTOR4(svp);
7658 7658 tvp = svp;
7659 7659 if (IS_SHADOW(svp, trp))
7660 7660 tvp = RTOV4(trp);
7661 7661 vnevent_link(tvp, ct);
7662 7662 }
7663 7663 out:
7664 7664 kmem_free(argop, argoplist_size);
7665 7665 if (resp)
7666 7666 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)resp);
7667 7667
7668 7668 nfs_rw_exit(&tdrp->r_rwlock);
7669 7669
7670 7670 return (e.error);
7671 7671 }
7672 7672
7673 7673 /* ARGSUSED */
7674 7674 static int
7675 7675 nfs4_rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
7676 7676 caller_context_t *ct, int flags)
7677 7677 {
7678 7678 vnode_t *realvp;
7679 7679
7680 7680 if (nfs_zone() != VTOMI4(odvp)->mi_zone)
7681 7681 return (EPERM);
7682 7682 if (VOP_REALVP(ndvp, &realvp, ct) == 0)
7683 7683 ndvp = realvp;
7684 7684
7685 7685 return (nfs4rename(odvp, onm, ndvp, nnm, cr, ct));
7686 7686 }
7687 7687
7688 7688 /*
7689 7689 * nfs4rename does the real work of renaming in NFS Version 4.
7690 7690 *
7691 7691 * A file handle is considered volatile for renaming purposes if either
7692 7692 * of the volatile bits are turned on. However, the compound may differ
7693 7693 * based on the likelihood of the filehandle to change during rename.
7694 7694 */
7695 7695 static int
7696 7696 nfs4rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
7697 7697 caller_context_t *ct)
7698 7698 {
7699 7699 int error;
7700 7700 mntinfo4_t *mi;
7701 7701 vnode_t *nvp = NULL;
7702 7702 vnode_t *ovp = NULL;
7703 7703 char *tmpname = NULL;
7704 7704 rnode4_t *rp;
7705 7705 rnode4_t *odrp;
7706 7706 rnode4_t *ndrp;
7707 7707 int did_link = 0;
7708 7708 int do_link = 1;
7709 7709 nfsstat4 stat = NFS4_OK;
7710 7710
7711 7711 ASSERT(nfs_zone() == VTOMI4(odvp)->mi_zone);
7712 7712 ASSERT(nfs4_consistent_type(odvp));
7713 7713 ASSERT(nfs4_consistent_type(ndvp));
7714 7714
7715 7715 if (onm[0] == '.' && (onm[1] == '\0' ||
7716 7716 (onm[1] == '.' && onm[2] == '\0')))
7717 7717 return (EINVAL);
7718 7718
7719 7719 if (nnm[0] == '.' && (nnm[1] == '\0' ||
7720 7720 (nnm[1] == '.' && nnm[2] == '\0')))
7721 7721 return (EINVAL);
7722 7722
7723 7723 odrp = VTOR4(odvp);
7724 7724 ndrp = VTOR4(ndvp);
7725 7725 if ((intptr_t)odrp < (intptr_t)ndrp) {
7726 7726 if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR4(odvp)))
7727 7727 return (EINTR);
7728 7728 if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR4(ndvp))) {
7729 7729 nfs_rw_exit(&odrp->r_rwlock);
7730 7730 return (EINTR);
7731 7731 }
7732 7732 } else {
7733 7733 if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR4(ndvp)))
7734 7734 return (EINTR);
7735 7735 if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR4(odvp))) {
7736 7736 nfs_rw_exit(&ndrp->r_rwlock);
7737 7737 return (EINTR);
7738 7738 }
7739 7739 }
7740 7740
7741 7741 /*
7742 7742 * Lookup the target file. If it exists, it needs to be
7743 7743 * checked to see whether it is a mount point and whether
7744 7744 * it is active (open).
7745 7745 */
7746 7746 error = nfs4lookup(ndvp, nnm, &nvp, cr, 0);
7747 7747 if (!error) {
7748 7748 int isactive;
7749 7749
7750 7750 ASSERT(nfs4_consistent_type(nvp));
7751 7751 /*
7752 7752 * If this file has been mounted on, then just
7753 7753 * return busy because renaming to it would remove
7754 7754 * the mounted file system from the name space.
7755 7755 */
7756 7756 if (vn_ismntpt(nvp)) {
7757 7757 VN_RELE(nvp);
7758 7758 nfs_rw_exit(&odrp->r_rwlock);
7759 7759 nfs_rw_exit(&ndrp->r_rwlock);
7760 7760 return (EBUSY);
7761 7761 }
7762 7762
7763 7763 /*
7764 7764 * First just remove the entry from the name cache, as it
7765 7765 * is most likely the only entry for this vp.
7766 7766 */
7767 7767 dnlc_remove(ndvp, nnm);
7768 7768
7769 7769 rp = VTOR4(nvp);
7770 7770
7771 7771 if (nvp->v_type != VREG) {
7772 7772 /*
7773 7773 * Purge the name cache of all references to this vnode
7774 7774 * so that we can check the reference count to infer
7775 7775 * whether it is active or not.
7776 7776 */
7777 7777 if (nvp->v_count > 1)
7778 7778 dnlc_purge_vp(nvp);
7779 7779
7780 7780 isactive = nvp->v_count > 1;
7781 7781 } else {
7782 7782 mutex_enter(&rp->r_os_lock);
7783 7783 isactive = list_head(&rp->r_open_streams) != NULL;
7784 7784 mutex_exit(&rp->r_os_lock);
7785 7785 }
7786 7786
7787 7787 /*
7788 7788 * If the vnode is active and is not a directory,
7789 7789 * arrange to rename it to a
7790 7790 * temporary file so that it will continue to be
7791 7791 * accessible. This implements the "unlink-open-file"
7792 7792 * semantics for the target of a rename operation.
7793 7793 * Before doing this though, make sure that the
7794 7794 * source and target files are not already the same.
7795 7795 */
7796 7796 if (isactive && nvp->v_type != VDIR) {
7797 7797 /*
7798 7798 * Lookup the source name.
7799 7799 */
7800 7800 error = nfs4lookup(odvp, onm, &ovp, cr, 0);
7801 7801
7802 7802 /*
7803 7803 * The source name *should* already exist.
7804 7804 */
7805 7805 if (error) {
7806 7806 VN_RELE(nvp);
7807 7807 nfs_rw_exit(&odrp->r_rwlock);
7808 7808 nfs_rw_exit(&ndrp->r_rwlock);
7809 7809 return (error);
7810 7810 }
7811 7811
7812 7812 ASSERT(nfs4_consistent_type(ovp));
7813 7813
7814 7814 /*
7815 7815 * Compare the two vnodes. If they are the same,
7816 7816 * just release all held vnodes and return success.
7817 7817 */
7818 7818 if (VN_CMP(ovp, nvp)) {
7819 7819 VN_RELE(ovp);
7820 7820 VN_RELE(nvp);
7821 7821 nfs_rw_exit(&odrp->r_rwlock);
7822 7822 nfs_rw_exit(&ndrp->r_rwlock);
7823 7823 return (0);
7824 7824 }
7825 7825
7826 7826 /*
7827 7827 * Can't mix and match directories and non-
7828 7828 * directories in rename operations. We already
7829 7829 * know that the target is not a directory. If
7830 7830 * the source is a directory, return an error.
7831 7831 */
7832 7832 if (ovp->v_type == VDIR) {
7833 7833 VN_RELE(ovp);
7834 7834 VN_RELE(nvp);
7835 7835 nfs_rw_exit(&odrp->r_rwlock);
7836 7836 nfs_rw_exit(&ndrp->r_rwlock);
7837 7837 return (ENOTDIR);
7838 7838 }
7839 7839 link_call:
7840 7840 /*
7841 7841 * The target file exists, is not the same as
7842 7842 * the source file, and is active. We first
7843 7843 * try to Link it to a temporary filename to
7844 7844 * avoid having the server removing the file
7845 7845 * completely (which could cause data loss to
7846 7846 * the user's POV in the event the Rename fails
7847 7847 * -- see bug 1165874).
7848 7848 */
7849 7849 /*
7850 7850 * The do_link and did_link booleans are
7851 7851 * introduced in the event we get NFS4ERR_FILE_OPEN
7852 7852 * returned for the Rename. Some servers can
7853 7853 * not Rename over an Open file, so they return
7854 7854 * this error. The client needs to Remove the
7855 7855 * newly created Link and do two Renames, just
7856 7856 * as if the server didn't support LINK.
7857 7857 */
7858 7858 tmpname = newname();
7859 7859 error = 0;
7860 7860
7861 7861 if (do_link) {
7862 7862 error = nfs4_link(ndvp, nvp, tmpname, cr,
7863 7863 NULL, 0);
7864 7864 }
7865 7865 if (error == EOPNOTSUPP || !do_link) {
7866 7866 error = nfs4_rename(ndvp, nnm, ndvp, tmpname,
7867 7867 cr, NULL, 0);
7868 7868 did_link = 0;
7869 7869 } else {
7870 7870 did_link = 1;
7871 7871 }
7872 7872 if (error) {
7873 7873 kmem_free(tmpname, MAXNAMELEN);
7874 7874 VN_RELE(ovp);
7875 7875 VN_RELE(nvp);
7876 7876 nfs_rw_exit(&odrp->r_rwlock);
7877 7877 nfs_rw_exit(&ndrp->r_rwlock);
7878 7878 return (error);
7879 7879 }
7880 7880
7881 7881 mutex_enter(&rp->r_statelock);
7882 7882 if (rp->r_unldvp == NULL) {
7883 7883 VN_HOLD(ndvp);
7884 7884 rp->r_unldvp = ndvp;
7885 7885 if (rp->r_unlcred != NULL)
7886 7886 crfree(rp->r_unlcred);
7887 7887 crhold(cr);
7888 7888 rp->r_unlcred = cr;
7889 7889 rp->r_unlname = tmpname;
7890 7890 } else {
7891 7891 if (rp->r_unlname)
7892 7892 kmem_free(rp->r_unlname, MAXNAMELEN);
7893 7893 rp->r_unlname = tmpname;
7894 7894 }
7895 7895 mutex_exit(&rp->r_statelock);
7896 7896 }
7897 7897
7898 7898 (void) nfs4delegreturn(VTOR4(nvp), NFS4_DR_PUSH|NFS4_DR_REOPEN);
7899 7899
7900 7900 ASSERT(nfs4_consistent_type(nvp));
7901 7901 }
7902 7902
7903 7903 if (ovp == NULL) {
7904 7904 /*
7905 7905 * When renaming directories to be a subdirectory of a
7906 7906 * different parent, the dnlc entry for ".." will no
7907 7907 * longer be valid, so it must be removed.
7908 7908 *
7909 7909 * We do a lookup here to determine whether we are renaming
7910 7910 * a directory and we need to check if we are renaming
7911 7911 * an unlinked file. This might have already been done
7912 7912 * in previous code, so we check ovp == NULL to avoid
7913 7913 * doing it twice.
7914 7914 */
7915 7915 error = nfs4lookup(odvp, onm, &ovp, cr, 0);
7916 7916 /*
7917 7917 * The source name *should* already exist.
7918 7918 */
7919 7919 if (error) {
7920 7920 nfs_rw_exit(&odrp->r_rwlock);
7921 7921 nfs_rw_exit(&ndrp->r_rwlock);
7922 7922 if (nvp) {
7923 7923 VN_RELE(nvp);
7924 7924 }
7925 7925 return (error);
7926 7926 }
7927 7927 ASSERT(ovp != NULL);
7928 7928 ASSERT(nfs4_consistent_type(ovp));
7929 7929 }
7930 7930
7931 7931 /*
7932 7932 * Is the object being renamed a dir, and if so, is
7933 7933 * it being renamed to a child of itself? The underlying
7934 7934 * fs should ultimately return EINVAL for this case;
7935 7935 * however, buggy beta non-Solaris NFSv4 servers at
7936 7936 * interop testing events have allowed this behavior,
7937 7937 * and it caused our client to panic due to a recursive
7938 7938 * mutex_enter in fn_move.
7939 7939 *
7940 7940 * The tedious locking in fn_move could be changed to
7941 7941 * deal with this case, and the client could avoid the
7942 7942 * panic; however, the client would just confuse itself
7943 7943 * later and misbehave. A better way to handle the broken
7944 7944 * server is to detect this condition and return EINVAL
7945 7945 * without ever sending the the bogus rename to the server.
7946 7946 * We know the rename is invalid -- just fail it now.
7947 7947 */
7948 7948 if (ovp->v_type == VDIR && VN_CMP(ndvp, ovp)) {
7949 7949 VN_RELE(ovp);
7950 7950 nfs_rw_exit(&odrp->r_rwlock);
7951 7951 nfs_rw_exit(&ndrp->r_rwlock);
7952 7952 if (nvp) {
7953 7953 VN_RELE(nvp);
7954 7954 }
7955 7955 return (EINVAL);
7956 7956 }
7957 7957
7958 7958 (void) nfs4delegreturn(VTOR4(ovp), NFS4_DR_PUSH|NFS4_DR_REOPEN);
7959 7959
7960 7960 /*
7961 7961 * If FH4_VOL_RENAME or FH4_VOLATILE_ANY bits are set, it is
7962 7962 * possible for the filehandle to change due to the rename.
7963 7963 * If neither of these bits is set, but FH4_VOL_MIGRATION is set,
7964 7964 * the fh will not change because of the rename, but we still need
7965 7965 * to update its rnode entry with the new name for
7966 7966 * an eventual fh change due to migration. The FH4_NOEXPIRE_ON_OPEN
7967 7967 * has no effect on these for now, but for future improvements,
7968 7968 * we might want to use it too to simplify handling of files
7969 7969 * that are open with that flag on. (XXX)
7970 7970 */
7971 7971 mi = VTOMI4(odvp);
7972 7972 if (NFS4_VOLATILE_FH(mi))
7973 7973 error = nfs4rename_volatile_fh(odvp, onm, ovp, ndvp, nnm, cr,
7974 7974 &stat);
7975 7975 else
7976 7976 error = nfs4rename_persistent_fh(odvp, onm, ovp, ndvp, nnm, cr,
7977 7977 &stat);
7978 7978
7979 7979 ASSERT(nfs4_consistent_type(odvp));
7980 7980 ASSERT(nfs4_consistent_type(ndvp));
7981 7981 ASSERT(nfs4_consistent_type(ovp));
7982 7982
7983 7983 if (stat == NFS4ERR_FILE_OPEN && did_link) {
7984 7984 do_link = 0;
7985 7985 /*
7986 7986 * Before the 'link_call' code, we did a nfs4_lookup
7987 7987 * that puts a VN_HOLD on nvp. After the nfs4_link
7988 7988 * call we call VN_RELE to match that hold. We need
7989 7989 * to place an additional VN_HOLD here since we will
7990 7990 * be hitting that VN_RELE again.
7991 7991 */
7992 7992 VN_HOLD(nvp);
7993 7993
7994 7994 (void) nfs4_remove(ndvp, tmpname, cr, NULL, 0);
7995 7995
7996 7996 /* Undo the unlinked file naming stuff we just did */
7997 7997 mutex_enter(&rp->r_statelock);
7998 7998 if (rp->r_unldvp) {
7999 7999 VN_RELE(ndvp);
8000 8000 rp->r_unldvp = NULL;
8001 8001 if (rp->r_unlcred != NULL)
8002 8002 crfree(rp->r_unlcred);
8003 8003 rp->r_unlcred = NULL;
8004 8004 /* rp->r_unlanme points to tmpname */
8005 8005 if (rp->r_unlname)
8006 8006 kmem_free(rp->r_unlname, MAXNAMELEN);
8007 8007 rp->r_unlname = NULL;
8008 8008 }
8009 8009 mutex_exit(&rp->r_statelock);
8010 8010
8011 8011 if (nvp) {
8012 8012 VN_RELE(nvp);
8013 8013 }
8014 8014 goto link_call;
8015 8015 }
8016 8016
8017 8017 if (error) {
8018 8018 VN_RELE(ovp);
8019 8019 nfs_rw_exit(&odrp->r_rwlock);
8020 8020 nfs_rw_exit(&ndrp->r_rwlock);
8021 8021 if (nvp) {
8022 8022 VN_RELE(nvp);
8023 8023 }
8024 8024 return (error);
8025 8025 }
8026 8026
8027 8027 /*
8028 8028 * when renaming directories to be a subdirectory of a
8029 8029 * different parent, the dnlc entry for ".." will no
8030 8030 * longer be valid, so it must be removed
8031 8031 */
8032 8032 rp = VTOR4(ovp);
8033 8033 if (ndvp != odvp) {
8034 8034 if (ovp->v_type == VDIR) {
8035 8035 dnlc_remove(ovp, "..");
8036 8036 if (rp->r_dir != NULL)
8037 8037 nfs4_purge_rddir_cache(ovp);
8038 8038 }
8039 8039 }
8040 8040
8041 8041 /*
8042 8042 * If we are renaming the unlinked file, update the
8043 8043 * r_unldvp and r_unlname as needed.
8044 8044 */
8045 8045 mutex_enter(&rp->r_statelock);
8046 8046 if (rp->r_unldvp != NULL) {
8047 8047 if (strcmp(rp->r_unlname, onm) == 0) {
8048 8048 (void) strncpy(rp->r_unlname, nnm, MAXNAMELEN);
8049 8049 rp->r_unlname[MAXNAMELEN - 1] = '\0';
8050 8050 if (ndvp != rp->r_unldvp) {
8051 8051 VN_RELE(rp->r_unldvp);
8052 8052 rp->r_unldvp = ndvp;
8053 8053 VN_HOLD(ndvp);
8054 8054 }
8055 8055 }
8056 8056 }
8057 8057 mutex_exit(&rp->r_statelock);
8058 8058
8059 8059 /*
8060 8060 * Notify the rename vnevents to source vnode, and to the target
8061 8061 * vnode if it already existed.
8062 8062 */
8063 8063 if (error == 0) {
8064 8064 vnode_t *tvp;
8065 8065 rnode4_t *trp;
8066 8066 /*
8067 8067 * Notify the vnode. Each links is represented by
8068 8068 * a different vnode, in nfsv4.
8069 8069 */
8070 8070 if (nvp) {
8071 8071 trp = VTOR4(nvp);
8072 8072 tvp = nvp;
8073 8073 if (IS_SHADOW(nvp, trp))
8074 8074 tvp = RTOV4(trp);
8075 8075 vnevent_rename_dest(tvp, ndvp, nnm, ct);
8076 8076 }
8077 8077
8078 8078 /*
8079 8079 * if the source and destination directory are not the
8080 8080 * same notify the destination directory.
8081 8081 */
8082 8082 if (VTOR4(odvp) != VTOR4(ndvp)) {
8083 8083 trp = VTOR4(ndvp);
8084 8084 tvp = ndvp;
8085 8085 if (IS_SHADOW(ndvp, trp))
8086 8086 tvp = RTOV4(trp);
8087 8087 vnevent_rename_dest_dir(tvp, ct);
8088 8088 }
8089 8089
8090 8090 trp = VTOR4(ovp);
8091 8091 tvp = ovp;
8092 8092 if (IS_SHADOW(ovp, trp))
8093 8093 tvp = RTOV4(trp);
8094 8094 vnevent_rename_src(tvp, odvp, onm, ct);
8095 8095 }
8096 8096
8097 8097 if (nvp) {
8098 8098 VN_RELE(nvp);
8099 8099 }
8100 8100 VN_RELE(ovp);
8101 8101
8102 8102 nfs_rw_exit(&odrp->r_rwlock);
8103 8103 nfs_rw_exit(&ndrp->r_rwlock);
8104 8104
8105 8105 return (error);
8106 8106 }
8107 8107
8108 8108 /*
8109 8109 * When the parent directory has changed, sv_dfh must be updated
8110 8110 */
8111 8111 static void
8112 8112 update_parentdir_sfh(vnode_t *vp, vnode_t *ndvp)
8113 8113 {
8114 8114 svnode_t *sv = VTOSV(vp);
8115 8115 nfs4_sharedfh_t *old_dfh = sv->sv_dfh;
8116 8116 nfs4_sharedfh_t *new_dfh = VTOR4(ndvp)->r_fh;
8117 8117
8118 8118 sfh4_hold(new_dfh);
8119 8119 sv->sv_dfh = new_dfh;
8120 8120 sfh4_rele(&old_dfh);
8121 8121 }
8122 8122
8123 8123 /*
8124 8124 * nfs4rename_persistent does the otw portion of renaming in NFS Version 4,
8125 8125 * when it is known that the filehandle is persistent through rename.
8126 8126 *
8127 8127 * Rename requires that the current fh be the target directory and the
8128 8128 * saved fh be the source directory. After the operation, the current fh
8129 8129 * is unchanged.
8130 8130 * The compound op structure for persistent fh rename is:
8131 8131 * PUTFH(sourcdir), SAVEFH, PUTFH(targetdir), RENAME
8132 8132 * Rather than bother with the directory postop args, we'll simply
8133 8133 * update that a change occurred in the cache, so no post-op getattrs.
8134 8134 */
8135 8135 static int
8136 8136 nfs4rename_persistent_fh(vnode_t *odvp, char *onm, vnode_t *renvp,
8137 8137 vnode_t *ndvp, char *nnm, cred_t *cr, nfsstat4 *statp)
8138 8138 {
8139 8139 COMPOUND4args_clnt args;
8140 8140 COMPOUND4res_clnt res, *resp = NULL;
8141 8141 nfs_argop4 *argop;
8142 8142 nfs_resop4 *resop;
8143 8143 int doqueue, argoplist_size;
8144 8144 mntinfo4_t *mi;
8145 8145 rnode4_t *odrp = VTOR4(odvp);
8146 8146 rnode4_t *ndrp = VTOR4(ndvp);
8147 8147 RENAME4res *rn_res;
8148 8148 bool_t needrecov;
8149 8149 nfs4_recov_state_t recov_state;
8150 8150 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
8151 8151 dirattr_info_t dinfo, *dinfop;
8152 8152
8153 8153 ASSERT(nfs_zone() == VTOMI4(odvp)->mi_zone);
8154 8154
8155 8155 recov_state.rs_flags = 0;
8156 8156 recov_state.rs_num_retry_despite_err = 0;
8157 8157
8158 8158 /*
8159 8159 * Rename ops: putfh sdir; savefh; putfh tdir; rename; getattr tdir
8160 8160 *
8161 8161 * If source/target are different dirs, then append putfh(src); getattr
8162 8162 */
8163 8163 args.array_len = (odvp == ndvp) ? 5 : 7;
8164 8164 argoplist_size = args.array_len * sizeof (nfs_argop4);
8165 8165 args.array = argop = kmem_alloc(argoplist_size, KM_SLEEP);
8166 8166
8167 8167 recov_retry:
8168 8168 *statp = NFS4_OK;
8169 8169
8170 8170 /* No need to Lookup the file, persistent fh */
8171 8171 args.ctag = TAG_RENAME;
8172 8172
8173 8173 mi = VTOMI4(odvp);
8174 8174 e.error = nfs4_start_op(mi, odvp, ndvp, &recov_state);
8175 8175 if (e.error) {
8176 8176 kmem_free(argop, argoplist_size);
8177 8177 return (e.error);
8178 8178 }
8179 8179
8180 8180 /* 0: putfh source directory */
8181 8181 argop[0].argop = OP_CPUTFH;
8182 8182 argop[0].nfs_argop4_u.opcputfh.sfh = odrp->r_fh;
8183 8183
8184 8184 /* 1: Save source fh to free up current for target */
8185 8185 argop[1].argop = OP_SAVEFH;
8186 8186
8187 8187 /* 2: putfh targetdir */
8188 8188 argop[2].argop = OP_CPUTFH;
8189 8189 argop[2].nfs_argop4_u.opcputfh.sfh = ndrp->r_fh;
8190 8190
8191 8191 /* 3: current_fh is targetdir, saved_fh is sourcedir */
8192 8192 argop[3].argop = OP_CRENAME;
8193 8193 argop[3].nfs_argop4_u.opcrename.coldname = onm;
8194 8194 argop[3].nfs_argop4_u.opcrename.cnewname = nnm;
8195 8195
8196 8196 /* 4: getattr (targetdir) */
8197 8197 argop[4].argop = OP_GETATTR;
8198 8198 argop[4].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
8199 8199 argop[4].nfs_argop4_u.opgetattr.mi = mi;
8200 8200
8201 8201 if (ndvp != odvp) {
8202 8202
8203 8203 /* 5: putfh (sourcedir) */
8204 8204 argop[5].argop = OP_CPUTFH;
8205 8205 argop[5].nfs_argop4_u.opcputfh.sfh = ndrp->r_fh;
8206 8206
8207 8207 /* 6: getattr (sourcedir) */
8208 8208 argop[6].argop = OP_GETATTR;
8209 8209 argop[6].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
8210 8210 argop[6].nfs_argop4_u.opgetattr.mi = mi;
8211 8211 }
8212 8212
8213 8213 dnlc_remove(odvp, onm);
8214 8214 dnlc_remove(ndvp, nnm);
8215 8215
8216 8216 doqueue = 1;
8217 8217 dinfo.di_time_call = gethrtime();
8218 8218 rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
8219 8219
8220 8220 needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
8221 8221 if (e.error) {
8222 8222 PURGE_ATTRCACHE4(odvp);
8223 8223 PURGE_ATTRCACHE4(ndvp);
8224 8224 } else {
8225 8225 *statp = res.status;
8226 8226 }
8227 8227
8228 8228 if (needrecov) {
8229 8229 if (nfs4_start_recovery(&e, mi, odvp, ndvp, NULL, NULL,
8230 8230 OP_RENAME, NULL, NULL, NULL) == FALSE) {
8231 8231 nfs4_end_op(mi, odvp, ndvp, &recov_state, needrecov);
8232 8232 if (!e.error)
8233 8233 (void) xdr_free(xdr_COMPOUND4res_clnt,
8234 8234 (caddr_t)&res);
8235 8235 goto recov_retry;
8236 8236 }
8237 8237 }
8238 8238
8239 8239 if (!e.error) {
8240 8240 resp = &res;
8241 8241 /*
8242 8242 * as long as OP_RENAME
8243 8243 */
8244 8244 if (res.status != NFS4_OK && res.array_len <= 4) {
8245 8245 e.error = geterrno4(res.status);
8246 8246 PURGE_ATTRCACHE4(odvp);
8247 8247 PURGE_ATTRCACHE4(ndvp);
8248 8248 /*
8249 8249 * System V defines rename to return EEXIST, not
8250 8250 * ENOTEMPTY if the target directory is not empty.
8251 8251 * Over the wire, the error is NFSERR_ENOTEMPTY
8252 8252 * which geterrno4 maps to ENOTEMPTY.
8253 8253 */
8254 8254 if (e.error == ENOTEMPTY)
8255 8255 e.error = EEXIST;
8256 8256 } else {
8257 8257
8258 8258 resop = &res.array[3]; /* rename res */
8259 8259 rn_res = &resop->nfs_resop4_u.oprename;
8260 8260
8261 8261 if (res.status == NFS4_OK) {
8262 8262 /*
8263 8263 * Update target attribute, readdir and dnlc
8264 8264 * caches.
8265 8265 */
8266 8266 dinfo.di_garp =
8267 8267 &res.array[4].nfs_resop4_u.opgetattr.ga_res;
8268 8268 dinfo.di_cred = cr;
8269 8269 dinfop = &dinfo;
8270 8270 } else
8271 8271 dinfop = NULL;
8272 8272
8273 8273 nfs4_update_dircaches(&rn_res->target_cinfo,
8274 8274 ndvp, NULL, NULL, dinfop);
8275 8275
8276 8276 /*
8277 8277 * Update source attribute, readdir and dnlc caches
8278 8278 *
8279 8279 */
8280 8280 if (ndvp != odvp) {
8281 8281 update_parentdir_sfh(renvp, ndvp);
8282 8282
8283 8283 if (dinfop)
8284 8284 dinfo.di_garp =
8285 8285 &(res.array[6].nfs_resop4_u.
8286 8286 opgetattr.ga_res);
8287 8287
8288 8288 nfs4_update_dircaches(&rn_res->source_cinfo,
8289 8289 odvp, NULL, NULL, dinfop);
8290 8290 }
8291 8291
8292 8292 fn_move(VTOSV(renvp)->sv_name, VTOSV(ndvp)->sv_name,
8293 8293 nnm);
8294 8294 }
8295 8295 }
8296 8296
8297 8297 if (resp)
8298 8298 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)resp);
8299 8299 nfs4_end_op(mi, odvp, ndvp, &recov_state, needrecov);
8300 8300 kmem_free(argop, argoplist_size);
8301 8301
8302 8302 return (e.error);
8303 8303 }
8304 8304
8305 8305 /*
8306 8306 * nfs4rename_volatile_fh does the otw part of renaming in NFS Version 4, when
8307 8307 * it is possible for the filehandle to change due to the rename.
8308 8308 *
8309 8309 * The compound req in this case includes a post-rename lookup and getattr
8310 8310 * to ensure that we have the correct fh and attributes for the object.
8311 8311 *
8312 8312 * Rename requires that the current fh be the target directory and the
8313 8313 * saved fh be the source directory. After the operation, the current fh
8314 8314 * is unchanged.
8315 8315 *
8316 8316 * We need the new filehandle (hence a LOOKUP and GETFH) so that we can
8317 8317 * update the filehandle for the renamed object. We also get the old
8318 8318 * filehandle for historical reasons; this should be taken out sometime.
8319 8319 * This results in a rather cumbersome compound...
8320 8320 *
8321 8321 * PUTFH(sourcdir), SAVEFH, LOOKUP(src), GETFH(old),
8322 8322 * PUTFH(targetdir), RENAME, LOOKUP(trgt), GETFH(new), GETATTR
8323 8323 *
8324 8324 */
8325 8325 static int
8326 8326 nfs4rename_volatile_fh(vnode_t *odvp, char *onm, vnode_t *ovp,
8327 8327 vnode_t *ndvp, char *nnm, cred_t *cr, nfsstat4 *statp)
8328 8328 {
8329 8329 COMPOUND4args_clnt args;
8330 8330 COMPOUND4res_clnt res, *resp = NULL;
8331 8331 int argoplist_size;
8332 8332 nfs_argop4 *argop;
8333 8333 nfs_resop4 *resop;
8334 8334 int doqueue;
8335 8335 mntinfo4_t *mi;
8336 8336 rnode4_t *odrp = VTOR4(odvp); /* old directory */
8337 8337 rnode4_t *ndrp = VTOR4(ndvp); /* new directory */
8338 8338 rnode4_t *orp = VTOR4(ovp); /* object being renamed */
8339 8339 RENAME4res *rn_res;
8340 8340 GETFH4res *ngf_res;
8341 8341 bool_t needrecov;
8342 8342 nfs4_recov_state_t recov_state;
8343 8343 hrtime_t t;
8344 8344 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
8345 8345 dirattr_info_t dinfo, *dinfop = &dinfo;
8346 8346
8347 8347 ASSERT(nfs_zone() == VTOMI4(odvp)->mi_zone);
8348 8348
8349 8349 recov_state.rs_flags = 0;
8350 8350 recov_state.rs_num_retry_despite_err = 0;
8351 8351
8352 8352 recov_retry:
8353 8353 *statp = NFS4_OK;
8354 8354
8355 8355 /*
8356 8356 * There is a window between the RPC and updating the path and
8357 8357 * filehandle stored in the rnode. Lock out the FHEXPIRED recovery
8358 8358 * code, so that it doesn't try to use the old path during that
8359 8359 * window.
8360 8360 */
8361 8361 mutex_enter(&orp->r_statelock);
8362 8362 while (orp->r_flags & R4RECEXPFH) {
8363 8363 klwp_t *lwp = ttolwp(curthread);
8364 8364
8365 8365 if (lwp != NULL)
8366 8366 lwp->lwp_nostop++;
8367 8367 if (cv_wait_sig(&orp->r_cv, &orp->r_statelock) == 0) {
8368 8368 mutex_exit(&orp->r_statelock);
8369 8369 if (lwp != NULL)
8370 8370 lwp->lwp_nostop--;
8371 8371 return (EINTR);
8372 8372 }
8373 8373 if (lwp != NULL)
8374 8374 lwp->lwp_nostop--;
8375 8375 }
8376 8376 orp->r_flags |= R4RECEXPFH;
8377 8377 mutex_exit(&orp->r_statelock);
8378 8378
8379 8379 mi = VTOMI4(odvp);
8380 8380
8381 8381 args.ctag = TAG_RENAME_VFH;
8382 8382 args.array_len = (odvp == ndvp) ? 10 : 12;
8383 8383 argoplist_size = args.array_len * sizeof (nfs_argop4);
8384 8384 argop = kmem_alloc(argoplist_size, KM_SLEEP);
8385 8385
8386 8386 /*
8387 8387 * Rename ops:
8388 8388 * PUTFH(sourcdir), SAVEFH, LOOKUP(src), GETFH(old),
8389 8389 * PUTFH(targetdir), RENAME, GETATTR(targetdir)
8390 8390 * LOOKUP(trgt), GETFH(new), GETATTR,
8391 8391 *
8392 8392 * if (odvp != ndvp)
8393 8393 * add putfh(sourcedir), getattr(sourcedir) }
8394 8394 */
8395 8395 args.array = argop;
8396 8396
8397 8397 e.error = nfs4_start_fop(mi, odvp, ndvp, OH_VFH_RENAME,
8398 8398 &recov_state, NULL);
8399 8399 if (e.error) {
8400 8400 kmem_free(argop, argoplist_size);
8401 8401 mutex_enter(&orp->r_statelock);
8402 8402 orp->r_flags &= ~R4RECEXPFH;
8403 8403 cv_broadcast(&orp->r_cv);
8404 8404 mutex_exit(&orp->r_statelock);
8405 8405 return (e.error);
8406 8406 }
8407 8407
8408 8408 /* 0: putfh source directory */
8409 8409 argop[0].argop = OP_CPUTFH;
8410 8410 argop[0].nfs_argop4_u.opcputfh.sfh = odrp->r_fh;
8411 8411
8412 8412 /* 1: Save source fh to free up current for target */
8413 8413 argop[1].argop = OP_SAVEFH;
8414 8414
8415 8415 /* 2: Lookup pre-rename fh of renamed object */
8416 8416 argop[2].argop = OP_CLOOKUP;
8417 8417 argop[2].nfs_argop4_u.opclookup.cname = onm;
8418 8418
8419 8419 /* 3: getfh fh of renamed object (before rename) */
8420 8420 argop[3].argop = OP_GETFH;
8421 8421
8422 8422 /* 4: putfh targetdir */
8423 8423 argop[4].argop = OP_CPUTFH;
8424 8424 argop[4].nfs_argop4_u.opcputfh.sfh = ndrp->r_fh;
8425 8425
8426 8426 /* 5: current_fh is targetdir, saved_fh is sourcedir */
8427 8427 argop[5].argop = OP_CRENAME;
8428 8428 argop[5].nfs_argop4_u.opcrename.coldname = onm;
8429 8429 argop[5].nfs_argop4_u.opcrename.cnewname = nnm;
8430 8430
8431 8431 /* 6: getattr of target dir (post op attrs) */
8432 8432 argop[6].argop = OP_GETATTR;
8433 8433 argop[6].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
8434 8434 argop[6].nfs_argop4_u.opgetattr.mi = mi;
8435 8435
8436 8436 /* 7: Lookup post-rename fh of renamed object */
8437 8437 argop[7].argop = OP_CLOOKUP;
8438 8438 argop[7].nfs_argop4_u.opclookup.cname = nnm;
8439 8439
8440 8440 /* 8: getfh fh of renamed object (after rename) */
8441 8441 argop[8].argop = OP_GETFH;
8442 8442
8443 8443 /* 9: getattr of renamed object */
8444 8444 argop[9].argop = OP_GETATTR;
8445 8445 argop[9].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
8446 8446 argop[9].nfs_argop4_u.opgetattr.mi = mi;
8447 8447
8448 8448 /*
8449 8449 * If source/target dirs are different, then get new post-op
8450 8450 * attrs for source dir also.
8451 8451 */
8452 8452 if (ndvp != odvp) {
8453 8453 /* 10: putfh (sourcedir) */
8454 8454 argop[10].argop = OP_CPUTFH;
8455 8455 argop[10].nfs_argop4_u.opcputfh.sfh = ndrp->r_fh;
8456 8456
8457 8457 /* 11: getattr (sourcedir) */
8458 8458 argop[11].argop = OP_GETATTR;
8459 8459 argop[11].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
8460 8460 argop[11].nfs_argop4_u.opgetattr.mi = mi;
8461 8461 }
8462 8462
8463 8463 dnlc_remove(odvp, onm);
8464 8464 dnlc_remove(ndvp, nnm);
8465 8465
8466 8466 doqueue = 1;
8467 8467 t = gethrtime();
8468 8468 rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
8469 8469
8470 8470 needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
8471 8471 if (e.error) {
8472 8472 PURGE_ATTRCACHE4(odvp);
8473 8473 PURGE_ATTRCACHE4(ndvp);
8474 8474 if (!needrecov) {
8475 8475 nfs4_end_fop(mi, odvp, ndvp, OH_VFH_RENAME,
8476 8476 &recov_state, needrecov);
8477 8477 goto out;
8478 8478 }
8479 8479 } else {
8480 8480 *statp = res.status;
8481 8481 }
8482 8482
8483 8483 if (needrecov) {
8484 8484 bool_t abort;
8485 8485
8486 8486 abort = nfs4_start_recovery(&e, mi, odvp, ndvp, NULL, NULL,
8487 8487 OP_RENAME, NULL, NULL, NULL);
8488 8488 if (abort == FALSE) {
8489 8489 nfs4_end_fop(mi, odvp, ndvp, OH_VFH_RENAME,
8490 8490 &recov_state, needrecov);
8491 8491 kmem_free(argop, argoplist_size);
8492 8492 if (!e.error)
8493 8493 (void) xdr_free(xdr_COMPOUND4res_clnt,
8494 8494 (caddr_t)&res);
8495 8495 mutex_enter(&orp->r_statelock);
8496 8496 orp->r_flags &= ~R4RECEXPFH;
8497 8497 cv_broadcast(&orp->r_cv);
8498 8498 mutex_exit(&orp->r_statelock);
8499 8499 goto recov_retry;
8500 8500 } else {
8501 8501 if (e.error != 0) {
8502 8502 nfs4_end_fop(mi, odvp, ndvp, OH_VFH_RENAME,
8503 8503 &recov_state, needrecov);
8504 8504 goto out;
8505 8505 }
8506 8506 /* fall through for res.status case */
8507 8507 }
8508 8508 }
8509 8509
8510 8510 resp = &res;
8511 8511 /*
8512 8512 * If OP_RENAME (or any prev op) failed, then return an error.
8513 8513 * OP_RENAME is index 5, so if array len <= 6 we return an error.
8514 8514 */
8515 8515 if ((res.status != NFS4_OK) && (res.array_len <= 6)) {
8516 8516 /*
8517 8517 * Error in an op other than last Getattr
8518 8518 */
8519 8519 e.error = geterrno4(res.status);
8520 8520 PURGE_ATTRCACHE4(odvp);
8521 8521 PURGE_ATTRCACHE4(ndvp);
8522 8522 /*
8523 8523 * System V defines rename to return EEXIST, not
8524 8524 * ENOTEMPTY if the target directory is not empty.
8525 8525 * Over the wire, the error is NFSERR_ENOTEMPTY
8526 8526 * which geterrno4 maps to ENOTEMPTY.
8527 8527 */
8528 8528 if (e.error == ENOTEMPTY)
8529 8529 e.error = EEXIST;
8530 8530 nfs4_end_fop(mi, odvp, ndvp, OH_VFH_RENAME, &recov_state,
8531 8531 needrecov);
8532 8532 goto out;
8533 8533 }
8534 8534
8535 8535 /* rename results */
8536 8536 rn_res = &res.array[5].nfs_resop4_u.oprename;
8537 8537
8538 8538 if (res.status == NFS4_OK) {
8539 8539 /* Update target attribute, readdir and dnlc caches */
8540 8540 dinfo.di_garp =
8541 8541 &res.array[6].nfs_resop4_u.opgetattr.ga_res;
8542 8542 dinfo.di_cred = cr;
8543 8543 dinfo.di_time_call = t;
8544 8544 } else
8545 8545 dinfop = NULL;
8546 8546
8547 8547 /* Update source cache attribute, readdir and dnlc caches */
8548 8548 nfs4_update_dircaches(&rn_res->target_cinfo, ndvp, NULL, NULL, dinfop);
8549 8549
8550 8550 /* Update source cache attribute, readdir and dnlc caches */
8551 8551 if (ndvp != odvp) {
8552 8552 update_parentdir_sfh(ovp, ndvp);
8553 8553
8554 8554 /*
8555 8555 * If dinfop is non-NULL, then compound succeded, so
8556 8556 * set di_garp to attrs for source dir. dinfop is only
8557 8557 * set to NULL when compound fails.
8558 8558 */
8559 8559 if (dinfop)
8560 8560 dinfo.di_garp =
8561 8561 &res.array[11].nfs_resop4_u.opgetattr.ga_res;
8562 8562 nfs4_update_dircaches(&rn_res->source_cinfo, odvp, NULL, NULL,
8563 8563 dinfop);
8564 8564 }
8565 8565
8566 8566 /*
8567 8567 * Update the rnode with the new component name and args,
8568 8568 * and if the file handle changed, also update it with the new fh.
8569 8569 * This is only necessary if the target object has an rnode
8570 8570 * entry and there is no need to create one for it.
8571 8571 */
8572 8572 resop = &res.array[8]; /* getfh new res */
8573 8573 ngf_res = &resop->nfs_resop4_u.opgetfh;
8574 8574
8575 8575 /*
8576 8576 * Update the path and filehandle for the renamed object.
8577 8577 */
8578 8578 nfs4rename_update(ovp, ndvp, &ngf_res->object, nnm);
8579 8579
8580 8580 nfs4_end_fop(mi, odvp, ndvp, OH_VFH_RENAME, &recov_state, needrecov);
8581 8581
8582 8582 if (res.status == NFS4_OK) {
8583 8583 resop++; /* getattr res */
8584 8584 e.error = nfs4_update_attrcache(res.status,
8585 8585 &resop->nfs_resop4_u.opgetattr.ga_res,
8586 8586 t, ovp, cr);
8587 8587 }
8588 8588
8589 8589 out:
8590 8590 kmem_free(argop, argoplist_size);
8591 8591 if (resp)
8592 8592 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)resp);
8593 8593 mutex_enter(&orp->r_statelock);
8594 8594 orp->r_flags &= ~R4RECEXPFH;
8595 8595 cv_broadcast(&orp->r_cv);
8596 8596 mutex_exit(&orp->r_statelock);
8597 8597
8598 8598 return (e.error);
8599 8599 }
8600 8600
8601 8601 /* ARGSUSED */
8602 8602 static int
8603 8603 nfs4_mkdir(vnode_t *dvp, char *nm, struct vattr *va, vnode_t **vpp, cred_t *cr,
8604 8604 caller_context_t *ct, int flags, vsecattr_t *vsecp)
8605 8605 {
8606 8606 int error;
8607 8607 vnode_t *vp;
8608 8608
8609 8609 if (nfs_zone() != VTOMI4(dvp)->mi_zone)
8610 8610 return (EPERM);
8611 8611 /*
8612 8612 * As ".." has special meaning and rather than send a mkdir
8613 8613 * over the wire to just let the server freak out, we just
8614 8614 * short circuit it here and return EEXIST
8615 8615 */
8616 8616 if (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0')
8617 8617 return (EEXIST);
8618 8618
8619 8619 /*
8620 8620 * Decision to get the right gid and setgid bit of the
8621 8621 * new directory is now made in call_nfs4_create_req.
8622 8622 */
8623 8623 va->va_mask |= AT_MODE;
8624 8624 error = call_nfs4_create_req(dvp, nm, NULL, va, &vp, cr, NF4DIR);
8625 8625 if (error)
8626 8626 return (error);
8627 8627
8628 8628 *vpp = vp;
8629 8629 return (0);
8630 8630 }
8631 8631
8632 8632
8633 8633 /*
8634 8634 * rmdir is using the same remove v4 op as does remove.
8635 8635 * Remove requires that the current fh be the target directory.
8636 8636 * After the operation, the current fh is unchanged.
8637 8637 * The compound op structure is:
8638 8638 * PUTFH(targetdir), REMOVE
8639 8639 */
8640 8640 /*ARGSUSED4*/
8641 8641 static int
8642 8642 nfs4_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr,
8643 8643 caller_context_t *ct, int flags)
8644 8644 {
8645 8645 int need_end_op = FALSE;
8646 8646 COMPOUND4args_clnt args;
8647 8647 COMPOUND4res_clnt res, *resp = NULL;
8648 8648 REMOVE4res *rm_res;
8649 8649 nfs_argop4 argop[3];
8650 8650 nfs_resop4 *resop;
8651 8651 vnode_t *vp;
8652 8652 int doqueue;
8653 8653 mntinfo4_t *mi;
8654 8654 rnode4_t *drp;
8655 8655 bool_t needrecov = FALSE;
8656 8656 nfs4_recov_state_t recov_state;
8657 8657 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
8658 8658 dirattr_info_t dinfo, *dinfop;
8659 8659
8660 8660 if (nfs_zone() != VTOMI4(dvp)->mi_zone)
8661 8661 return (EPERM);
8662 8662 /*
8663 8663 * As ".." has special meaning and rather than send a rmdir
8664 8664 * over the wire to just let the server freak out, we just
8665 8665 * short circuit it here and return EEXIST
8666 8666 */
8667 8667 if (nm[0] == '.' && nm[1] == '.' && nm[2] == '\0')
8668 8668 return (EEXIST);
8669 8669
8670 8670 drp = VTOR4(dvp);
8671 8671 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR4(dvp)))
8672 8672 return (EINTR);
8673 8673
8674 8674 /*
8675 8675 * Attempt to prevent a rmdir(".") from succeeding.
8676 8676 */
8677 8677 e.error = nfs4lookup(dvp, nm, &vp, cr, 0);
8678 8678 if (e.error) {
8679 8679 nfs_rw_exit(&drp->r_rwlock);
8680 8680 return (e.error);
8681 8681 }
8682 8682 if (vp == cdir) {
8683 8683 VN_RELE(vp);
8684 8684 nfs_rw_exit(&drp->r_rwlock);
8685 8685 return (EINVAL);
8686 8686 }
8687 8687
8688 8688 /*
8689 8689 * Since nfsv4 remove op works on both files and directories,
8690 8690 * check that the removed object is indeed a directory.
8691 8691 */
8692 8692 if (vp->v_type != VDIR) {
8693 8693 VN_RELE(vp);
8694 8694 nfs_rw_exit(&drp->r_rwlock);
8695 8695 return (ENOTDIR);
8696 8696 }
8697 8697
8698 8698 /*
8699 8699 * First just remove the entry from the name cache, as it
8700 8700 * is most likely an entry for this vp.
8701 8701 */
8702 8702 dnlc_remove(dvp, nm);
8703 8703
8704 8704 /*
8705 8705 * If there vnode reference count is greater than one, then
8706 8706 * there may be additional references in the DNLC which will
8707 8707 * need to be purged. First, trying removing the entry for
8708 8708 * the parent directory and see if that removes the additional
8709 8709 * reference(s). If that doesn't do it, then use dnlc_purge_vp
8710 8710 * to completely remove any references to the directory which
8711 8711 * might still exist in the DNLC.
8712 8712 */
8713 8713 if (vp->v_count > 1) {
8714 8714 dnlc_remove(vp, "..");
8715 8715 if (vp->v_count > 1)
8716 8716 dnlc_purge_vp(vp);
8717 8717 }
8718 8718
8719 8719 mi = VTOMI4(dvp);
8720 8720 recov_state.rs_flags = 0;
8721 8721 recov_state.rs_num_retry_despite_err = 0;
8722 8722
8723 8723 recov_retry:
8724 8724 args.ctag = TAG_RMDIR;
8725 8725
8726 8726 /*
8727 8727 * Rmdir ops: putfh dir; remove
8728 8728 */
8729 8729 args.array_len = 3;
8730 8730 args.array = argop;
8731 8731
8732 8732 e.error = nfs4_start_op(VTOMI4(dvp), dvp, NULL, &recov_state);
8733 8733 if (e.error) {
8734 8734 nfs_rw_exit(&drp->r_rwlock);
8735 8735 return (e.error);
8736 8736 }
8737 8737 need_end_op = TRUE;
8738 8738
8739 8739 /* putfh directory */
8740 8740 argop[0].argop = OP_CPUTFH;
8741 8741 argop[0].nfs_argop4_u.opcputfh.sfh = drp->r_fh;
8742 8742
8743 8743 /* remove */
8744 8744 argop[1].argop = OP_CREMOVE;
8745 8745 argop[1].nfs_argop4_u.opcremove.ctarget = nm;
8746 8746
8747 8747 /* getattr (postop attrs for dir that contained removed dir) */
8748 8748 argop[2].argop = OP_GETATTR;
8749 8749 argop[2].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
8750 8750 argop[2].nfs_argop4_u.opgetattr.mi = mi;
8751 8751
8752 8752 dinfo.di_time_call = gethrtime();
8753 8753 doqueue = 1;
8754 8754 rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
8755 8755
8756 8756 PURGE_ATTRCACHE4(vp);
8757 8757
8758 8758 needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
8759 8759 if (e.error) {
8760 8760 PURGE_ATTRCACHE4(dvp);
8761 8761 }
8762 8762
8763 8763 if (needrecov) {
8764 8764 if (nfs4_start_recovery(&e, VTOMI4(dvp), dvp, NULL, NULL,
8765 8765 NULL, OP_REMOVE, NULL, NULL, NULL) == FALSE) {
8766 8766 if (!e.error)
8767 8767 (void) xdr_free(xdr_COMPOUND4res_clnt,
8768 8768 (caddr_t)&res);
8769 8769
8770 8770 nfs4_end_op(VTOMI4(dvp), dvp, NULL, &recov_state,
8771 8771 needrecov);
8772 8772 need_end_op = FALSE;
8773 8773 goto recov_retry;
8774 8774 }
8775 8775 }
8776 8776
8777 8777 if (!e.error) {
8778 8778 resp = &res;
8779 8779
8780 8780 /*
8781 8781 * Only return error if first 2 ops (OP_REMOVE or earlier)
8782 8782 * failed.
8783 8783 */
8784 8784 if (res.status != NFS4_OK && res.array_len <= 2) {
8785 8785 e.error = geterrno4(res.status);
8786 8786 PURGE_ATTRCACHE4(dvp);
8787 8787 nfs4_end_op(VTOMI4(dvp), dvp, NULL,
8788 8788 &recov_state, needrecov);
8789 8789 need_end_op = FALSE;
8790 8790 nfs4_purge_stale_fh(e.error, dvp, cr);
8791 8791 /*
8792 8792 * System V defines rmdir to return EEXIST, not
8793 8793 * ENOTEMPTY if the directory is not empty. Over
8794 8794 * the wire, the error is NFSERR_ENOTEMPTY which
8795 8795 * geterrno4 maps to ENOTEMPTY.
8796 8796 */
8797 8797 if (e.error == ENOTEMPTY)
8798 8798 e.error = EEXIST;
8799 8799 } else {
8800 8800 resop = &res.array[1]; /* remove res */
8801 8801 rm_res = &resop->nfs_resop4_u.opremove;
8802 8802
8803 8803 if (res.status == NFS4_OK) {
8804 8804 resop = &res.array[2]; /* dir attrs */
8805 8805 dinfo.di_garp =
8806 8806 &resop->nfs_resop4_u.opgetattr.ga_res;
8807 8807 dinfo.di_cred = cr;
8808 8808 dinfop = &dinfo;
8809 8809 } else
8810 8810 dinfop = NULL;
8811 8811
8812 8812 /* Update dir attribute, readdir and dnlc caches */
8813 8813 nfs4_update_dircaches(&rm_res->cinfo, dvp, NULL, NULL,
8814 8814 dinfop);
8815 8815
8816 8816 /* destroy rddir cache for dir that was removed */
8817 8817 if (VTOR4(vp)->r_dir != NULL)
8818 8818 nfs4_purge_rddir_cache(vp);
8819 8819 }
8820 8820 }
8821 8821
8822 8822 if (need_end_op)
8823 8823 nfs4_end_op(VTOMI4(dvp), dvp, NULL, &recov_state, needrecov);
8824 8824
8825 8825 nfs_rw_exit(&drp->r_rwlock);
8826 8826
8827 8827 if (resp)
8828 8828 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)resp);
8829 8829
8830 8830 if (e.error == 0) {
8831 8831 vnode_t *tvp;
8832 8832 rnode4_t *trp;
8833 8833 trp = VTOR4(vp);
8834 8834 tvp = vp;
8835 8835 if (IS_SHADOW(vp, trp))
8836 8836 tvp = RTOV4(trp);
8837 8837 vnevent_rmdir(tvp, dvp, nm, ct);
8838 8838 }
8839 8839
8840 8840 VN_RELE(vp);
8841 8841
8842 8842 return (e.error);
8843 8843 }
8844 8844
8845 8845 /* ARGSUSED */
8846 8846 static int
8847 8847 nfs4_symlink(vnode_t *dvp, char *lnm, struct vattr *tva, char *tnm, cred_t *cr,
8848 8848 caller_context_t *ct, int flags)
8849 8849 {
8850 8850 int error;
8851 8851 vnode_t *vp;
8852 8852 rnode4_t *rp;
8853 8853 char *contents;
8854 8854 mntinfo4_t *mi = VTOMI4(dvp);
8855 8855
8856 8856 if (nfs_zone() != mi->mi_zone)
8857 8857 return (EPERM);
8858 8858 if (!(mi->mi_flags & MI4_SYMLINK))
8859 8859 return (EOPNOTSUPP);
8860 8860
8861 8861 error = call_nfs4_create_req(dvp, lnm, tnm, tva, &vp, cr, NF4LNK);
8862 8862 if (error)
8863 8863 return (error);
8864 8864
8865 8865 ASSERT(nfs4_consistent_type(vp));
8866 8866 rp = VTOR4(vp);
8867 8867 if (nfs4_do_symlink_cache && rp->r_symlink.contents == NULL) {
8868 8868
8869 8869 contents = kmem_alloc(MAXPATHLEN, KM_SLEEP);
8870 8870
8871 8871 if (contents != NULL) {
8872 8872 mutex_enter(&rp->r_statelock);
8873 8873 if (rp->r_symlink.contents == NULL) {
8874 8874 rp->r_symlink.len = strlen(tnm);
8875 8875 bcopy(tnm, contents, rp->r_symlink.len);
8876 8876 rp->r_symlink.contents = contents;
8877 8877 rp->r_symlink.size = MAXPATHLEN;
8878 8878 mutex_exit(&rp->r_statelock);
8879 8879 } else {
8880 8880 mutex_exit(&rp->r_statelock);
8881 8881 kmem_free((void *)contents, MAXPATHLEN);
8882 8882 }
8883 8883 }
8884 8884 }
8885 8885 VN_RELE(vp);
8886 8886
8887 8887 return (error);
8888 8888 }
8889 8889
8890 8890
8891 8891 /*
8892 8892 * Read directory entries.
8893 8893 * There are some weird things to look out for here. The uio_loffset
8894 8894 * field is either 0 or it is the offset returned from a previous
8895 8895 * readdir. It is an opaque value used by the server to find the
8896 8896 * correct directory block to read. The count field is the number
8897 8897 * of blocks to read on the server. This is advisory only, the server
8898 8898 * may return only one block's worth of entries. Entries may be compressed
8899 8899 * on the server.
8900 8900 */
8901 8901 /* ARGSUSED */
8902 8902 static int
8903 8903 nfs4_readdir(vnode_t *vp, struct uio *uiop, cred_t *cr, int *eofp,
8904 8904 caller_context_t *ct, int flags)
8905 8905 {
8906 8906 int error;
8907 8907 uint_t count;
8908 8908 rnode4_t *rp;
8909 8909 rddir4_cache *rdc;
8910 8910 rddir4_cache *rrdc;
8911 8911
8912 8912 if (nfs_zone() != VTOMI4(vp)->mi_zone)
8913 8913 return (EIO);
8914 8914 rp = VTOR4(vp);
8915 8915
8916 8916 ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER));
8917 8917
8918 8918 /*
8919 8919 * Make sure that the directory cache is valid.
8920 8920 */
8921 8921 if (rp->r_dir != NULL) {
8922 8922 if (nfs_disable_rddir_cache != 0) {
8923 8923 /*
8924 8924 * Setting nfs_disable_rddir_cache in /etc/system
8925 8925 * allows interoperability with servers that do not
8926 8926 * properly update the attributes of directories.
8927 8927 * Any cached information gets purged before an
8928 8928 * access is made to it.
8929 8929 */
8930 8930 nfs4_purge_rddir_cache(vp);
8931 8931 }
8932 8932
8933 8933 error = nfs4_validate_caches(vp, cr);
8934 8934 if (error)
8935 8935 return (error);
8936 8936 }
8937 8937
8938 8938 count = MIN(uiop->uio_iov->iov_len, MAXBSIZE);
8939 8939
8940 8940 /*
8941 8941 * Short circuit last readdir which always returns 0 bytes.
8942 8942 * This can be done after the directory has been read through
8943 8943 * completely at least once. This will set r_direof which
8944 8944 * can be used to find the value of the last cookie.
8945 8945 */
8946 8946 mutex_enter(&rp->r_statelock);
8947 8947 if (rp->r_direof != NULL &&
8948 8948 uiop->uio_loffset == rp->r_direof->nfs4_ncookie) {
8949 8949 mutex_exit(&rp->r_statelock);
8950 8950 #ifdef DEBUG
8951 8951 nfs4_readdir_cache_shorts++;
8952 8952 #endif
8953 8953 if (eofp)
8954 8954 *eofp = 1;
8955 8955 return (0);
8956 8956 }
8957 8957
8958 8958 /*
8959 8959 * Look for a cache entry. Cache entries are identified
8960 8960 * by the NFS cookie value and the byte count requested.
8961 8961 */
8962 8962 rdc = rddir4_cache_lookup(rp, uiop->uio_loffset, count);
8963 8963
8964 8964 /*
8965 8965 * If rdc is NULL then the lookup resulted in an unrecoverable error.
8966 8966 */
8967 8967 if (rdc == NULL) {
8968 8968 mutex_exit(&rp->r_statelock);
8969 8969 return (EINTR);
8970 8970 }
8971 8971
8972 8972 /*
8973 8973 * Check to see if we need to fill this entry in.
8974 8974 */
8975 8975 if (rdc->flags & RDDIRREQ) {
8976 8976 rdc->flags &= ~RDDIRREQ;
8977 8977 rdc->flags |= RDDIR;
8978 8978 mutex_exit(&rp->r_statelock);
8979 8979
8980 8980 /*
8981 8981 * Do the readdir.
8982 8982 */
8983 8983 nfs4readdir(vp, rdc, cr);
8984 8984
8985 8985 /*
8986 8986 * Reacquire the lock, so that we can continue
8987 8987 */
8988 8988 mutex_enter(&rp->r_statelock);
8989 8989 /*
8990 8990 * The entry is now complete
8991 8991 */
8992 8992 rdc->flags &= ~RDDIR;
8993 8993 }
8994 8994
8995 8995 ASSERT(!(rdc->flags & RDDIR));
8996 8996
8997 8997 /*
8998 8998 * If an error occurred while attempting
8999 8999 * to fill the cache entry, mark the entry invalid and
9000 9000 * just return the error.
9001 9001 */
9002 9002 if (rdc->error) {
9003 9003 error = rdc->error;
9004 9004 rdc->flags |= RDDIRREQ;
9005 9005 rddir4_cache_rele(rp, rdc);
9006 9006 mutex_exit(&rp->r_statelock);
9007 9007 return (error);
9008 9008 }
9009 9009
9010 9010 /*
9011 9011 * The cache entry is complete and good,
9012 9012 * copyout the dirent structs to the calling
9013 9013 * thread.
9014 9014 */
9015 9015 error = uiomove(rdc->entries, rdc->actlen, UIO_READ, uiop);
9016 9016
9017 9017 /*
9018 9018 * If no error occurred during the copyout,
9019 9019 * update the offset in the uio struct to
9020 9020 * contain the value of the next NFS 4 cookie
9021 9021 * and set the eof value appropriately.
9022 9022 */
9023 9023 if (!error) {
9024 9024 uiop->uio_loffset = rdc->nfs4_ncookie;
9025 9025 if (eofp)
9026 9026 *eofp = rdc->eof;
9027 9027 }
9028 9028
9029 9029 /*
9030 9030 * Decide whether to do readahead. Don't if we
9031 9031 * have already read to the end of directory.
9032 9032 */
9033 9033 if (rdc->eof) {
9034 9034 /*
9035 9035 * Make the entry the direof only if it is cached
9036 9036 */
9037 9037 if (rdc->flags & RDDIRCACHED)
9038 9038 rp->r_direof = rdc;
9039 9039 rddir4_cache_rele(rp, rdc);
9040 9040 mutex_exit(&rp->r_statelock);
9041 9041 return (error);
9042 9042 }
9043 9043
9044 9044 /* Determine if a readdir readahead should be done */
9045 9045 if (!(rp->r_flags & R4LOOKUP)) {
9046 9046 rddir4_cache_rele(rp, rdc);
9047 9047 mutex_exit(&rp->r_statelock);
9048 9048 return (error);
9049 9049 }
9050 9050
9051 9051 /*
9052 9052 * Now look for a readahead entry.
9053 9053 *
9054 9054 * Check to see whether we found an entry for the readahead.
9055 9055 * If so, we don't need to do anything further, so free the new
9056 9056 * entry if one was allocated. Otherwise, allocate a new entry, add
9057 9057 * it to the cache, and then initiate an asynchronous readdir
9058 9058 * operation to fill it.
9059 9059 */
9060 9060 rrdc = rddir4_cache_lookup(rp, rdc->nfs4_ncookie, count);
9061 9061
9062 9062 /*
9063 9063 * A readdir cache entry could not be obtained for the readahead. In
9064 9064 * this case we skip the readahead and return.
9065 9065 */
9066 9066 if (rrdc == NULL) {
9067 9067 rddir4_cache_rele(rp, rdc);
9068 9068 mutex_exit(&rp->r_statelock);
9069 9069 return (error);
9070 9070 }
9071 9071
9072 9072 /*
9073 9073 * Check to see if we need to fill this entry in.
9074 9074 */
9075 9075 if (rrdc->flags & RDDIRREQ) {
9076 9076 rrdc->flags &= ~RDDIRREQ;
9077 9077 rrdc->flags |= RDDIR;
9078 9078 rddir4_cache_rele(rp, rdc);
9079 9079 mutex_exit(&rp->r_statelock);
9080 9080 #ifdef DEBUG
9081 9081 nfs4_readdir_readahead++;
9082 9082 #endif
9083 9083 /*
9084 9084 * Do the readdir.
9085 9085 */
9086 9086 nfs4_async_readdir(vp, rrdc, cr, do_nfs4readdir);
9087 9087 return (error);
9088 9088 }
9089 9089
9090 9090 rddir4_cache_rele(rp, rrdc);
9091 9091 rddir4_cache_rele(rp, rdc);
9092 9092 mutex_exit(&rp->r_statelock);
9093 9093 return (error);
9094 9094 }
9095 9095
9096 9096 static int
9097 9097 do_nfs4readdir(vnode_t *vp, rddir4_cache *rdc, cred_t *cr)
9098 9098 {
9099 9099 int error;
9100 9100 rnode4_t *rp;
9101 9101
9102 9102 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
9103 9103
9104 9104 rp = VTOR4(vp);
9105 9105
9106 9106 /*
9107 9107 * Obtain the readdir results for the caller.
9108 9108 */
9109 9109 nfs4readdir(vp, rdc, cr);
9110 9110
9111 9111 mutex_enter(&rp->r_statelock);
9112 9112 /*
9113 9113 * The entry is now complete
9114 9114 */
9115 9115 rdc->flags &= ~RDDIR;
9116 9116
9117 9117 error = rdc->error;
9118 9118 if (error)
9119 9119 rdc->flags |= RDDIRREQ;
9120 9120 rddir4_cache_rele(rp, rdc);
9121 9121 mutex_exit(&rp->r_statelock);
9122 9122
9123 9123 return (error);
9124 9124 }
9125 9125
9126 9126 /*
9127 9127 * Read directory entries.
9128 9128 * There are some weird things to look out for here. The uio_loffset
9129 9129 * field is either 0 or it is the offset returned from a previous
9130 9130 * readdir. It is an opaque value used by the server to find the
9131 9131 * correct directory block to read. The count field is the number
9132 9132 * of blocks to read on the server. This is advisory only, the server
9133 9133 * may return only one block's worth of entries. Entries may be compressed
9134 9134 * on the server.
9135 9135 *
9136 9136 * Generates the following compound request:
9137 9137 * 1. If readdir offset is zero and no dnlc entry for parent exists,
9138 9138 * must include a Lookupp as well. In this case, send:
9139 9139 * { Putfh <fh>; Readdir; Lookupp; Getfh; Getattr }
9140 9140 * 2. Otherwise just do: { Putfh <fh>; Readdir }
9141 9141 *
9142 9142 * Get complete attributes and filehandles for entries if this is the
9143 9143 * first read of the directory. Otherwise, just get fileid's.
9144 9144 */
9145 9145 static void
9146 9146 nfs4readdir(vnode_t *vp, rddir4_cache *rdc, cred_t *cr)
9147 9147 {
9148 9148 COMPOUND4args_clnt args;
9149 9149 COMPOUND4res_clnt res;
9150 9150 READDIR4args *rargs;
9151 9151 READDIR4res_clnt *rd_res;
9152 9152 bitmap4 rd_bitsval;
9153 9153 nfs_argop4 argop[5];
9154 9154 nfs_resop4 *resop;
9155 9155 rnode4_t *rp = VTOR4(vp);
9156 9156 mntinfo4_t *mi = VTOMI4(vp);
9157 9157 int doqueue;
9158 9158 u_longlong_t nodeid, pnodeid; /* id's of dir and its parents */
9159 9159 vnode_t *dvp;
9160 9160 nfs_cookie4 cookie = (nfs_cookie4)rdc->nfs4_cookie;
9161 9161 int num_ops, res_opcnt;
9162 9162 bool_t needrecov = FALSE;
9163 9163 nfs4_recov_state_t recov_state;
9164 9164 hrtime_t t;
9165 9165 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
9166 9166
9167 9167 ASSERT(nfs_zone() == mi->mi_zone);
9168 9168 ASSERT(rdc->flags & RDDIR);
9169 9169 ASSERT(rdc->entries == NULL);
9170 9170
9171 9171 /*
9172 9172 * If rp were a stub, it should have triggered and caused
9173 9173 * a mount for us to get this far.
9174 9174 */
9175 9175 ASSERT(!RP_ISSTUB(rp));
9176 9176
9177 9177 num_ops = 2;
9178 9178 if (cookie == (nfs_cookie4)0 || cookie == (nfs_cookie4)1) {
9179 9179 /*
9180 9180 * Since nfsv4 readdir may not return entries for "." and "..",
9181 9181 * the client must recreate them:
9182 9182 * To find the correct nodeid, do the following:
9183 9183 * For current node, get nodeid from dnlc.
9184 9184 * - if current node is rootvp, set pnodeid to nodeid.
9185 9185 * - else if parent is in the dnlc, get its nodeid from there.
9186 9186 * - else add LOOKUPP+GETATTR to compound.
9187 9187 */
9188 9188 nodeid = rp->r_attr.va_nodeid;
9189 9189 if (vp->v_flag & VROOT) {
9190 9190 pnodeid = nodeid; /* root of mount point */
9191 9191 } else {
9192 9192 dvp = dnlc_lookup(vp, "..");
9193 9193 if (dvp != NULL && dvp != DNLC_NO_VNODE) {
9194 9194 /* parent in dnlc cache - no need for otw */
9195 9195 pnodeid = VTOR4(dvp)->r_attr.va_nodeid;
9196 9196 } else {
9197 9197 /*
9198 9198 * parent not in dnlc cache,
9199 9199 * do lookupp to get its id
9200 9200 */
9201 9201 num_ops = 5;
9202 9202 pnodeid = 0; /* set later by getattr parent */
9203 9203 }
9204 9204 if (dvp)
9205 9205 VN_RELE(dvp);
9206 9206 }
9207 9207 }
9208 9208 recov_state.rs_flags = 0;
9209 9209 recov_state.rs_num_retry_despite_err = 0;
9210 9210
9211 9211 /* Save the original mount point security flavor */
9212 9212 (void) save_mnt_secinfo(mi->mi_curr_serv);
9213 9213
9214 9214 recov_retry:
9215 9215 args.ctag = TAG_READDIR;
9216 9216
9217 9217 args.array = argop;
9218 9218 args.array_len = num_ops;
9219 9219
9220 9220 if (e.error = nfs4_start_fop(VTOMI4(vp), vp, NULL, OH_READDIR,
9221 9221 &recov_state, NULL)) {
9222 9222 /*
9223 9223 * If readdir a node that is a stub for a crossed mount point,
9224 9224 * keep the original secinfo flavor for the current file
9225 9225 * system, not the crossed one.
9226 9226 */
9227 9227 (void) check_mnt_secinfo(mi->mi_curr_serv, vp);
9228 9228 rdc->error = e.error;
9229 9229 return;
9230 9230 }
9231 9231
9232 9232 /*
9233 9233 * Determine which attrs to request for dirents. This code
9234 9234 * must be protected by nfs4_start/end_fop because of r_server
9235 9235 * (which will change during failover recovery).
9236 9236 *
9237 9237 */
9238 9238 if (rp->r_flags & (R4LOOKUP | R4READDIRWATTR)) {
9239 9239 /*
9240 9240 * Get all vattr attrs plus filehandle and rdattr_error
9241 9241 */
9242 9242 rd_bitsval = NFS4_VATTR_MASK |
9243 9243 FATTR4_RDATTR_ERROR_MASK |
9244 9244 FATTR4_FILEHANDLE_MASK;
9245 9245
9246 9246 if (rp->r_flags & R4READDIRWATTR) {
9247 9247 mutex_enter(&rp->r_statelock);
9248 9248 rp->r_flags &= ~R4READDIRWATTR;
9249 9249 mutex_exit(&rp->r_statelock);
9250 9250 }
9251 9251 } else {
9252 9252 servinfo4_t *svp = rp->r_server;
9253 9253
9254 9254 /*
9255 9255 * Already read directory. Use readdir with
9256 9256 * no attrs (except for mounted_on_fileid) for updates.
9257 9257 */
9258 9258 rd_bitsval = FATTR4_RDATTR_ERROR_MASK;
9259 9259
9260 9260 /*
9261 9261 * request mounted on fileid if supported, else request
9262 9262 * fileid. maybe we should verify that fileid is supported
9263 9263 * and request something else if not.
9264 9264 */
9265 9265 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
9266 9266 if (svp->sv_supp_attrs & FATTR4_MOUNTED_ON_FILEID_MASK)
9267 9267 rd_bitsval |= FATTR4_MOUNTED_ON_FILEID_MASK;
9268 9268 nfs_rw_exit(&svp->sv_lock);
9269 9269 }
9270 9270
9271 9271 /* putfh directory fh */
9272 9272 argop[0].argop = OP_CPUTFH;
9273 9273 argop[0].nfs_argop4_u.opcputfh.sfh = rp->r_fh;
9274 9274
9275 9275 argop[1].argop = OP_READDIR;
9276 9276 rargs = &argop[1].nfs_argop4_u.opreaddir;
9277 9277 /*
9278 9278 * 1 and 2 are reserved for client "." and ".." entry offset.
9279 9279 * cookie 0 should be used over-the-wire to start reading at
9280 9280 * the beginning of the directory excluding "." and "..".
9281 9281 */
9282 9282 if (rdc->nfs4_cookie == 0 ||
9283 9283 rdc->nfs4_cookie == 1 ||
9284 9284 rdc->nfs4_cookie == 2) {
9285 9285 rargs->cookie = (nfs_cookie4)0;
9286 9286 rargs->cookieverf = 0;
9287 9287 } else {
9288 9288 rargs->cookie = (nfs_cookie4)rdc->nfs4_cookie;
9289 9289 mutex_enter(&rp->r_statelock);
9290 9290 rargs->cookieverf = rp->r_cookieverf4;
9291 9291 mutex_exit(&rp->r_statelock);
9292 9292 }
9293 9293 rargs->dircount = MIN(rdc->buflen, mi->mi_tsize);
9294 9294 rargs->maxcount = mi->mi_tsize;
9295 9295 rargs->attr_request = rd_bitsval;
9296 9296 rargs->rdc = rdc;
9297 9297 rargs->dvp = vp;
9298 9298 rargs->mi = mi;
9299 9299 rargs->cr = cr;
9300 9300
9301 9301
9302 9302 /*
9303 9303 * If count < than the minimum required, we return no entries
9304 9304 * and fail with EINVAL
9305 9305 */
9306 9306 if (rargs->dircount < (DIRENT64_RECLEN(1) + DIRENT64_RECLEN(2))) {
9307 9307 rdc->error = EINVAL;
9308 9308 goto out;
9309 9309 }
9310 9310
9311 9311 if (args.array_len == 5) {
9312 9312 /*
9313 9313 * Add lookupp and getattr for parent nodeid.
9314 9314 */
9315 9315 argop[2].argop = OP_LOOKUPP;
9316 9316
9317 9317 argop[3].argop = OP_GETFH;
9318 9318
9319 9319 /* getattr parent */
9320 9320 argop[4].argop = OP_GETATTR;
9321 9321 argop[4].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
9322 9322 argop[4].nfs_argop4_u.opgetattr.mi = mi;
9323 9323 }
9324 9324
9325 9325 doqueue = 1;
9326 9326
9327 9327 if (mi->mi_io_kstats) {
9328 9328 mutex_enter(&mi->mi_lock);
9329 9329 kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
9330 9330 mutex_exit(&mi->mi_lock);
9331 9331 }
9332 9332
9333 9333 /* capture the time of this call */
9334 9334 rargs->t = t = gethrtime();
9335 9335
9336 9336 rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
9337 9337
9338 9338 if (mi->mi_io_kstats) {
9339 9339 mutex_enter(&mi->mi_lock);
9340 9340 kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
9341 9341 mutex_exit(&mi->mi_lock);
9342 9342 }
9343 9343
9344 9344 needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
9345 9345
9346 9346 /*
9347 9347 * If RPC error occurred and it isn't an error that
9348 9348 * triggers recovery, then go ahead and fail now.
9349 9349 */
9350 9350 if (e.error != 0 && !needrecov) {
9351 9351 rdc->error = e.error;
9352 9352 goto out;
9353 9353 }
9354 9354
9355 9355 if (needrecov) {
9356 9356 bool_t abort;
9357 9357
9358 9358 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
9359 9359 "nfs4readdir: initiating recovery.\n"));
9360 9360
9361 9361 abort = nfs4_start_recovery(&e, VTOMI4(vp), vp, NULL, NULL,
9362 9362 NULL, OP_READDIR, NULL, NULL, NULL);
9363 9363 if (abort == FALSE) {
9364 9364 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_READDIR,
9365 9365 &recov_state, needrecov);
9366 9366 if (!e.error)
9367 9367 (void) xdr_free(xdr_COMPOUND4res_clnt,
9368 9368 (caddr_t)&res);
9369 9369 if (rdc->entries != NULL) {
9370 9370 kmem_free(rdc->entries, rdc->entlen);
9371 9371 rdc->entries = NULL;
9372 9372 }
9373 9373 goto recov_retry;
9374 9374 }
9375 9375
9376 9376 if (e.error != 0) {
9377 9377 rdc->error = e.error;
9378 9378 goto out;
9379 9379 }
9380 9380
9381 9381 /* fall through for res.status case */
9382 9382 }
9383 9383
9384 9384 res_opcnt = res.array_len;
9385 9385
9386 9386 /*
9387 9387 * If compound failed first 2 ops (PUTFH+READDIR), then return
9388 9388 * failure here. Subsequent ops are for filling out dot-dot
9389 9389 * dirent, and if they fail, we still want to give the caller
9390 9390 * the dirents returned by (the successful) READDIR op, so we need
9391 9391 * to silently ignore failure for subsequent ops (LOOKUPP+GETATTR).
9392 9392 *
9393 9393 * One example where PUTFH+READDIR ops would succeed but
9394 9394 * LOOKUPP+GETATTR would fail would be a dir that has r perm
9395 9395 * but lacks x. In this case, a POSIX server's VOP_READDIR
9396 9396 * would succeed; however, VOP_LOOKUP(..) would fail since no
9397 9397 * x perm. We need to come up with a non-vendor-specific way
9398 9398 * for a POSIX server to return d_ino from dotdot's dirent if
9399 9399 * client only requests mounted_on_fileid, and just say the
9400 9400 * LOOKUPP succeeded and fill out the GETATTR. However, if
9401 9401 * client requested any mandatory attrs, server would be required
9402 9402 * to fail the GETATTR op because it can't call VOP_LOOKUP+VOP_GETATTR
9403 9403 * for dotdot.
9404 9404 */
9405 9405
9406 9406 if (res.status) {
9407 9407 if (res_opcnt <= 2) {
9408 9408 e.error = geterrno4(res.status);
9409 9409 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_READDIR,
9410 9410 &recov_state, needrecov);
9411 9411 nfs4_purge_stale_fh(e.error, vp, cr);
9412 9412 rdc->error = e.error;
9413 9413 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
9414 9414 if (rdc->entries != NULL) {
9415 9415 kmem_free(rdc->entries, rdc->entlen);
9416 9416 rdc->entries = NULL;
9417 9417 }
9418 9418 /*
9419 9419 * If readdir a node that is a stub for a
9420 9420 * crossed mount point, keep the original
9421 9421 * secinfo flavor for the current file system,
9422 9422 * not the crossed one.
9423 9423 */
9424 9424 (void) check_mnt_secinfo(mi->mi_curr_serv, vp);
9425 9425 return;
9426 9426 }
9427 9427 }
9428 9428
9429 9429 resop = &res.array[1]; /* readdir res */
9430 9430 rd_res = &resop->nfs_resop4_u.opreaddirclnt;
9431 9431
9432 9432 mutex_enter(&rp->r_statelock);
9433 9433 rp->r_cookieverf4 = rd_res->cookieverf;
9434 9434 mutex_exit(&rp->r_statelock);
9435 9435
9436 9436 /*
9437 9437 * For "." and ".." entries
9438 9438 * e.g.
9439 9439 * seek(cookie=0) -> "." entry with d_off = 1
9440 9440 * seek(cookie=1) -> ".." entry with d_off = 2
9441 9441 */
9442 9442 if (cookie == (nfs_cookie4) 0) {
9443 9443 if (rd_res->dotp)
9444 9444 rd_res->dotp->d_ino = nodeid;
9445 9445 if (rd_res->dotdotp)
9446 9446 rd_res->dotdotp->d_ino = pnodeid;
9447 9447 }
9448 9448 if (cookie == (nfs_cookie4) 1) {
9449 9449 if (rd_res->dotdotp)
9450 9450 rd_res->dotdotp->d_ino = pnodeid;
9451 9451 }
9452 9452
9453 9453
9454 9454 /* LOOKUPP+GETATTR attemped */
9455 9455 if (args.array_len == 5 && rd_res->dotdotp) {
9456 9456 if (res.status == NFS4_OK && res_opcnt == 5) {
9457 9457 nfs_fh4 *fhp;
9458 9458 nfs4_sharedfh_t *sfhp;
9459 9459 vnode_t *pvp;
9460 9460 nfs4_ga_res_t *garp;
9461 9461
9462 9462 resop++; /* lookupp */
9463 9463 resop++; /* getfh */
9464 9464 fhp = &resop->nfs_resop4_u.opgetfh.object;
9465 9465
9466 9466 resop++; /* getattr of parent */
9467 9467
9468 9468 /*
9469 9469 * First, take care of finishing the
9470 9470 * readdir results.
9471 9471 */
9472 9472 garp = &resop->nfs_resop4_u.opgetattr.ga_res;
9473 9473 /*
9474 9474 * The d_ino of .. must be the inode number
9475 9475 * of the mounted filesystem.
9476 9476 */
9477 9477 if (garp->n4g_va.va_mask & AT_NODEID)
9478 9478 rd_res->dotdotp->d_ino =
9479 9479 garp->n4g_va.va_nodeid;
9480 9480
9481 9481
9482 9482 /*
9483 9483 * Next, create the ".." dnlc entry
9484 9484 */
9485 9485 sfhp = sfh4_get(fhp, mi);
9486 9486 if (!nfs4_make_dotdot(sfhp, t, vp, cr, &pvp, 0)) {
9487 9487 dnlc_update(vp, "..", pvp);
9488 9488 VN_RELE(pvp);
9489 9489 }
9490 9490 sfh4_rele(&sfhp);
9491 9491 }
9492 9492 }
9493 9493
9494 9494 if (mi->mi_io_kstats) {
9495 9495 mutex_enter(&mi->mi_lock);
9496 9496 KSTAT_IO_PTR(mi->mi_io_kstats)->reads++;
9497 9497 KSTAT_IO_PTR(mi->mi_io_kstats)->nread += rdc->actlen;
9498 9498 mutex_exit(&mi->mi_lock);
9499 9499 }
9500 9500
9501 9501 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
9502 9502
9503 9503 out:
9504 9504 /*
9505 9505 * If readdir a node that is a stub for a crossed mount point,
9506 9506 * keep the original secinfo flavor for the current file system,
9507 9507 * not the crossed one.
9508 9508 */
9509 9509 (void) check_mnt_secinfo(mi->mi_curr_serv, vp);
9510 9510
9511 9511 nfs4_end_fop(mi, vp, NULL, OH_READDIR, &recov_state, needrecov);
9512 9512 }
9513 9513
9514 9514
9515 9515 static int
9516 9516 nfs4_bio(struct buf *bp, stable_how4 *stab_comm, cred_t *cr, bool_t readahead)
9517 9517 {
9518 9518 rnode4_t *rp = VTOR4(bp->b_vp);
9519 9519 int count;
9520 9520 int error;
9521 9521 cred_t *cred_otw = NULL;
9522 9522 offset_t offset;
9523 9523 nfs4_open_stream_t *osp = NULL;
9524 9524 bool_t first_time = TRUE; /* first time getting otw cred */
9525 9525 bool_t last_time = FALSE; /* last time getting otw cred */
9526 9526
9527 9527 ASSERT(nfs_zone() == VTOMI4(bp->b_vp)->mi_zone);
9528 9528
9529 9529 DTRACE_IO1(start, struct buf *, bp);
9530 9530 offset = ldbtob(bp->b_lblkno);
9531 9531
9532 9532 if (bp->b_flags & B_READ) {
9533 9533 read_again:
9534 9534 /*
9535 9535 * Releases the osp, if it is provided.
9536 9536 * Puts a hold on the cred_otw and the new osp (if found).
9537 9537 */
9538 9538 cred_otw = nfs4_get_otw_cred_by_osp(rp, cr, &osp,
9539 9539 &first_time, &last_time);
9540 9540 error = bp->b_error = nfs4read(bp->b_vp, bp->b_un.b_addr,
9541 9541 offset, bp->b_bcount, &bp->b_resid, cred_otw,
9542 9542 readahead, NULL);
9543 9543 crfree(cred_otw);
9544 9544 if (!error) {
9545 9545 if (bp->b_resid) {
9546 9546 /*
9547 9547 * Didn't get it all because we hit EOF,
9548 9548 * zero all the memory beyond the EOF.
9549 9549 */
9550 9550 /* bzero(rdaddr + */
9551 9551 bzero(bp->b_un.b_addr +
9552 9552 bp->b_bcount - bp->b_resid, bp->b_resid);
9553 9553 }
9554 9554 mutex_enter(&rp->r_statelock);
9555 9555 if (bp->b_resid == bp->b_bcount &&
9556 9556 offset >= rp->r_size) {
9557 9557 /*
9558 9558 * We didn't read anything at all as we are
9559 9559 * past EOF. Return an error indicator back
9560 9560 * but don't destroy the pages (yet).
9561 9561 */
9562 9562 error = NFS_EOF;
9563 9563 }
9564 9564 mutex_exit(&rp->r_statelock);
9565 9565 } else if (error == EACCES && last_time == FALSE) {
9566 9566 goto read_again;
9567 9567 }
9568 9568 } else {
9569 9569 if (!(rp->r_flags & R4STALE)) {
9570 9570 write_again:
9571 9571 /*
9572 9572 * Releases the osp, if it is provided.
9573 9573 * Puts a hold on the cred_otw and the new
9574 9574 * osp (if found).
9575 9575 */
9576 9576 cred_otw = nfs4_get_otw_cred_by_osp(rp, cr, &osp,
9577 9577 &first_time, &last_time);
9578 9578 mutex_enter(&rp->r_statelock);
9579 9579 count = MIN(bp->b_bcount, rp->r_size - offset);
9580 9580 mutex_exit(&rp->r_statelock);
9581 9581 if (count < 0)
9582 9582 cmn_err(CE_PANIC, "nfs4_bio: write count < 0");
9583 9583 #ifdef DEBUG
9584 9584 if (count == 0) {
9585 9585 zoneid_t zoneid = getzoneid();
9586 9586
9587 9587 zcmn_err(zoneid, CE_WARN,
9588 9588 "nfs4_bio: zero length write at %lld",
9589 9589 offset);
9590 9590 zcmn_err(zoneid, CE_CONT, "flags=0x%x, "
9591 9591 "b_bcount=%ld, file size=%lld",
9592 9592 rp->r_flags, (long)bp->b_bcount,
9593 9593 rp->r_size);
9594 9594 sfh4_printfhandle(VTOR4(bp->b_vp)->r_fh);
9595 9595 if (nfs4_bio_do_stop)
9596 9596 debug_enter("nfs4_bio");
9597 9597 }
9598 9598 #endif
9599 9599 error = nfs4write(bp->b_vp, bp->b_un.b_addr, offset,
9600 9600 count, cred_otw, stab_comm);
9601 9601 if (error == EACCES && last_time == FALSE) {
9602 9602 crfree(cred_otw);
9603 9603 goto write_again;
9604 9604 }
9605 9605 bp->b_error = error;
9606 9606 if (error && error != EINTR &&
9607 9607 !(bp->b_vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)) {
9608 9608 /*
9609 9609 * Don't print EDQUOT errors on the console.
9610 9610 * Don't print asynchronous EACCES errors.
9611 9611 * Don't print EFBIG errors.
9612 9612 * Print all other write errors.
9613 9613 */
9614 9614 if (error != EDQUOT && error != EFBIG &&
9615 9615 (error != EACCES ||
9616 9616 !(bp->b_flags & B_ASYNC)))
9617 9617 nfs4_write_error(bp->b_vp,
9618 9618 error, cred_otw);
9619 9619 /*
9620 9620 * Update r_error and r_flags as appropriate.
9621 9621 * If the error was ESTALE, then mark the
9622 9622 * rnode as not being writeable and save
9623 9623 * the error status. Otherwise, save any
9624 9624 * errors which occur from asynchronous
9625 9625 * page invalidations. Any errors occurring
9626 9626 * from other operations should be saved
9627 9627 * by the caller.
9628 9628 */
9629 9629 mutex_enter(&rp->r_statelock);
9630 9630 if (error == ESTALE) {
9631 9631 rp->r_flags |= R4STALE;
9632 9632 if (!rp->r_error)
9633 9633 rp->r_error = error;
9634 9634 } else if (!rp->r_error &&
9635 9635 (bp->b_flags &
9636 9636 (B_INVAL|B_FORCE|B_ASYNC)) ==
9637 9637 (B_INVAL|B_FORCE|B_ASYNC)) {
9638 9638 rp->r_error = error;
9639 9639 }
9640 9640 mutex_exit(&rp->r_statelock);
9641 9641 }
9642 9642 crfree(cred_otw);
9643 9643 } else {
9644 9644 error = rp->r_error;
9645 9645 /*
9646 9646 * A close may have cleared r_error, if so,
9647 9647 * propagate ESTALE error return properly
9648 9648 */
9649 9649 if (error == 0)
9650 9650 error = ESTALE;
9651 9651 }
9652 9652 }
9653 9653
9654 9654 if (error != 0 && error != NFS_EOF)
9655 9655 bp->b_flags |= B_ERROR;
9656 9656
9657 9657 if (osp)
9658 9658 open_stream_rele(osp, rp);
9659 9659
9660 9660 DTRACE_IO1(done, struct buf *, bp);
9661 9661
9662 9662 return (error);
9663 9663 }
9664 9664
9665 9665 /* ARGSUSED */
9666 9666 int
9667 9667 nfs4_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
9668 9668 {
9669 9669 return (EREMOTE);
9670 9670 }
9671 9671
9672 9672 /* ARGSUSED2 */
9673 9673 int
9674 9674 nfs4_rwlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
9675 9675 {
9676 9676 rnode4_t *rp = VTOR4(vp);
9677 9677
9678 9678 if (!write_lock) {
9679 9679 (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE);
9680 9680 return (V_WRITELOCK_FALSE);
9681 9681 }
9682 9682
9683 9683 if ((rp->r_flags & R4DIRECTIO) ||
9684 9684 (VTOMI4(vp)->mi_flags & MI4_DIRECTIO)) {
9685 9685 (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE);
9686 9686 if (rp->r_mapcnt == 0 && !nfs4_has_pages(vp))
9687 9687 return (V_WRITELOCK_FALSE);
9688 9688 nfs_rw_exit(&rp->r_rwlock);
9689 9689 }
9690 9690
9691 9691 (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, FALSE);
9692 9692 return (V_WRITELOCK_TRUE);
9693 9693 }
9694 9694
9695 9695 /* ARGSUSED */
9696 9696 void
9697 9697 nfs4_rwunlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
9698 9698 {
9699 9699 rnode4_t *rp = VTOR4(vp);
9700 9700
9701 9701 nfs_rw_exit(&rp->r_rwlock);
9702 9702 }
9703 9703
9704 9704 /* ARGSUSED */
9705 9705 static int
9706 9706 nfs4_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct)
9707 9707 {
9708 9708 if (nfs_zone() != VTOMI4(vp)->mi_zone)
9709 9709 return (EIO);
9710 9710
9711 9711 /*
9712 9712 * Because we stuff the readdir cookie into the offset field
9713 9713 * someone may attempt to do an lseek with the cookie which
9714 9714 * we want to succeed.
9715 9715 */
9716 9716 if (vp->v_type == VDIR)
9717 9717 return (0);
9718 9718 if (*noffp < 0)
9719 9719 return (EINVAL);
9720 9720 return (0);
9721 9721 }
9722 9722
9723 9723
9724 9724 /*
9725 9725 * Return all the pages from [off..off+len) in file
9726 9726 */
9727 9727 /* ARGSUSED */
9728 9728 static int
9729 9729 nfs4_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
9730 9730 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
9731 9731 enum seg_rw rw, cred_t *cr, caller_context_t *ct)
9732 9732 {
9733 9733 rnode4_t *rp;
9734 9734 int error;
9735 9735 mntinfo4_t *mi;
9736 9736
9737 9737 if (nfs_zone() != VTOMI4(vp)->mi_zone)
9738 9738 return (EIO);
9739 9739 rp = VTOR4(vp);
9740 9740 if (IS_SHADOW(vp, rp))
9741 9741 vp = RTOV4(rp);
9742 9742
9743 9743 if (vp->v_flag & VNOMAP)
9744 9744 return (ENOSYS);
9745 9745
9746 9746 if (protp != NULL)
9747 9747 *protp = PROT_ALL;
9748 9748
9749 9749 /*
9750 9750 * Now validate that the caches are up to date.
9751 9751 */
9752 9752 if (error = nfs4_validate_caches(vp, cr))
9753 9753 return (error);
9754 9754
9755 9755 mi = VTOMI4(vp);
9756 9756 retry:
9757 9757 mutex_enter(&rp->r_statelock);
9758 9758
9759 9759 /*
9760 9760 * Don't create dirty pages faster than they
9761 9761 * can be cleaned so that the system doesn't
9762 9762 * get imbalanced. If the async queue is
9763 9763 * maxed out, then wait for it to drain before
9764 9764 * creating more dirty pages. Also, wait for
9765 9765 * any threads doing pagewalks in the vop_getattr
9766 9766 * entry points so that they don't block for
9767 9767 * long periods.
9768 9768 */
9769 9769 if (rw == S_CREATE) {
9770 9770 while ((mi->mi_max_threads != 0 &&
9771 9771 rp->r_awcount > 2 * mi->mi_max_threads) ||
9772 9772 rp->r_gcount > 0)
9773 9773 cv_wait(&rp->r_cv, &rp->r_statelock);
9774 9774 }
9775 9775
9776 9776 /*
9777 9777 * If we are getting called as a side effect of an nfs_write()
9778 9778 * operation the local file size might not be extended yet.
9779 9779 * In this case we want to be able to return pages of zeroes.
9780 9780 */
9781 9781 if (off + len > rp->r_size + PAGEOFFSET && seg != segkmap) {
9782 9782 NFS4_DEBUG(nfs4_pageio_debug,
9783 9783 (CE_NOTE, "getpage beyond EOF: off=%lld, "
9784 9784 "len=%llu, size=%llu, attrsize =%llu", off,
9785 9785 (u_longlong_t)len, rp->r_size, rp->r_attr.va_size));
9786 9786 mutex_exit(&rp->r_statelock);
9787 9787 return (EFAULT); /* beyond EOF */
9788 9788 }
9789 9789
9790 9790 mutex_exit(&rp->r_statelock);
9791 9791
9792 9792 error = pvn_getpages(nfs4_getapage, vp, off, len, protp,
9793 9793 pl, plsz, seg, addr, rw, cr);
9794 9794 NFS4_DEBUG(nfs4_pageio_debug && error,
9795 9795 (CE_NOTE, "getpages error %d; off=%lld, len=%lld",
9796 9796 error, off, (u_longlong_t)len));
9797 9797
9798 9798 switch (error) {
9799 9799 case NFS_EOF:
9800 9800 nfs4_purge_caches(vp, NFS4_NOPURGE_DNLC, cr, FALSE);
9801 9801 goto retry;
9802 9802 case ESTALE:
9803 9803 nfs4_purge_stale_fh(error, vp, cr);
9804 9804 }
9805 9805
9806 9806 return (error);
9807 9807 }
9808 9808
9809 9809 /*
9810 9810 * Called from pvn_getpages to get a particular page.
9811 9811 */
9812 9812 /* ARGSUSED */
9813 9813 static int
9814 9814 nfs4_getapage(vnode_t *vp, u_offset_t off, size_t len, uint_t *protp,
9815 9815 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
9816 9816 enum seg_rw rw, cred_t *cr)
9817 9817 {
9818 9818 rnode4_t *rp;
9819 9819 uint_t bsize;
9820 9820 struct buf *bp;
9821 9821 page_t *pp;
9822 9822 u_offset_t lbn;
9823 9823 u_offset_t io_off;
9824 9824 u_offset_t blkoff;
9825 9825 u_offset_t rablkoff;
9826 9826 size_t io_len;
9827 9827 uint_t blksize;
9828 9828 int error;
9829 9829 int readahead;
9830 9830 int readahead_issued = 0;
9831 9831 int ra_window; /* readahead window */
9832 9832 page_t *pagefound;
9833 9833 page_t *savepp;
9834 9834
9835 9835 if (nfs_zone() != VTOMI4(vp)->mi_zone)
9836 9836 return (EIO);
9837 9837
9838 9838 rp = VTOR4(vp);
9839 9839 ASSERT(!IS_SHADOW(vp, rp));
9840 9840 bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
9841 9841
9842 9842 reread:
9843 9843 bp = NULL;
9844 9844 pp = NULL;
9845 9845 pagefound = NULL;
9846 9846
9847 9847 if (pl != NULL)
9848 9848 pl[0] = NULL;
9849 9849
9850 9850 error = 0;
9851 9851 lbn = off / bsize;
9852 9852 blkoff = lbn * bsize;
9853 9853
9854 9854 /*
9855 9855 * Queueing up the readahead before doing the synchronous read
9856 9856 * results in a significant increase in read throughput because
9857 9857 * of the increased parallelism between the async threads and
9858 9858 * the process context.
9859 9859 */
9860 9860 if ((off & ((vp->v_vfsp->vfs_bsize) - 1)) == 0 &&
9861 9861 rw != S_CREATE &&
9862 9862 !(vp->v_flag & VNOCACHE)) {
9863 9863 mutex_enter(&rp->r_statelock);
9864 9864
9865 9865 /*
9866 9866 * Calculate the number of readaheads to do.
9867 9867 * a) No readaheads at offset = 0.
9868 9868 * b) Do maximum(nfs4_nra) readaheads when the readahead
9869 9869 * window is closed.
9870 9870 * c) Do readaheads between 1 to (nfs4_nra - 1) depending
9871 9871 * upon how far the readahead window is open or close.
9872 9872 * d) No readaheads if rp->r_nextr is not within the scope
9873 9873 * of the readahead window (random i/o).
9874 9874 */
9875 9875
9876 9876 if (off == 0)
9877 9877 readahead = 0;
9878 9878 else if (blkoff == rp->r_nextr)
9879 9879 readahead = nfs4_nra;
9880 9880 else if (rp->r_nextr > blkoff &&
9881 9881 ((ra_window = (rp->r_nextr - blkoff) / bsize)
9882 9882 <= (nfs4_nra - 1)))
9883 9883 readahead = nfs4_nra - ra_window;
9884 9884 else
9885 9885 readahead = 0;
9886 9886
9887 9887 rablkoff = rp->r_nextr;
9888 9888 while (readahead > 0 && rablkoff + bsize < rp->r_size) {
9889 9889 mutex_exit(&rp->r_statelock);
9890 9890 if (nfs4_async_readahead(vp, rablkoff + bsize,
9891 9891 addr + (rablkoff + bsize - off),
9892 9892 seg, cr, nfs4_readahead) < 0) {
9893 9893 mutex_enter(&rp->r_statelock);
9894 9894 break;
9895 9895 }
9896 9896 readahead--;
9897 9897 rablkoff += bsize;
9898 9898 /*
9899 9899 * Indicate that we did a readahead so
9900 9900 * readahead offset is not updated
9901 9901 * by the synchronous read below.
9902 9902 */
9903 9903 readahead_issued = 1;
9904 9904 mutex_enter(&rp->r_statelock);
9905 9905 /*
9906 9906 * set readahead offset to
9907 9907 * offset of last async readahead
9908 9908 * request.
9909 9909 */
9910 9910 rp->r_nextr = rablkoff;
9911 9911 }
9912 9912 mutex_exit(&rp->r_statelock);
9913 9913 }
9914 9914
9915 9915 again:
9916 9916 if ((pagefound = page_exists(vp, off)) == NULL) {
9917 9917 if (pl == NULL) {
9918 9918 (void) nfs4_async_readahead(vp, blkoff, addr, seg, cr,
9919 9919 nfs4_readahead);
9920 9920 } else if (rw == S_CREATE) {
9921 9921 /*
9922 9922 * Block for this page is not allocated, or the offset
9923 9923 * is beyond the current allocation size, or we're
9924 9924 * allocating a swap slot and the page was not found,
9925 9925 * so allocate it and return a zero page.
9926 9926 */
9927 9927 if ((pp = page_create_va(vp, off,
9928 9928 PAGESIZE, PG_WAIT, seg, addr)) == NULL)
9929 9929 cmn_err(CE_PANIC, "nfs4_getapage: page_create");
9930 9930 io_len = PAGESIZE;
9931 9931 mutex_enter(&rp->r_statelock);
9932 9932 rp->r_nextr = off + PAGESIZE;
9933 9933 mutex_exit(&rp->r_statelock);
9934 9934 } else {
9935 9935 /*
9936 9936 * Need to go to server to get a block
9937 9937 */
9938 9938 mutex_enter(&rp->r_statelock);
9939 9939 if (blkoff < rp->r_size &&
9940 9940 blkoff + bsize > rp->r_size) {
9941 9941 /*
9942 9942 * If less than a block left in
9943 9943 * file read less than a block.
9944 9944 */
9945 9945 if (rp->r_size <= off) {
9946 9946 /*
9947 9947 * Trying to access beyond EOF,
9948 9948 * set up to get at least one page.
9949 9949 */
9950 9950 blksize = off + PAGESIZE - blkoff;
9951 9951 } else
9952 9952 blksize = rp->r_size - blkoff;
9953 9953 } else if ((off == 0) ||
9954 9954 (off != rp->r_nextr && !readahead_issued)) {
9955 9955 blksize = PAGESIZE;
9956 9956 blkoff = off; /* block = page here */
9957 9957 } else
9958 9958 blksize = bsize;
9959 9959 mutex_exit(&rp->r_statelock);
9960 9960
9961 9961 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
9962 9962 &io_len, blkoff, blksize, 0);
9963 9963
9964 9964 /*
9965 9965 * Some other thread has entered the page,
9966 9966 * so just use it.
9967 9967 */
9968 9968 if (pp == NULL)
9969 9969 goto again;
9970 9970
9971 9971 /*
9972 9972 * Now round the request size up to page boundaries.
9973 9973 * This ensures that the entire page will be
9974 9974 * initialized to zeroes if EOF is encountered.
9975 9975 */
9976 9976 io_len = ptob(btopr(io_len));
9977 9977
9978 9978 bp = pageio_setup(pp, io_len, vp, B_READ);
9979 9979 ASSERT(bp != NULL);
9980 9980
9981 9981 /*
9982 9982 * pageio_setup should have set b_addr to 0. This
9983 9983 * is correct since we want to do I/O on a page
9984 9984 * boundary. bp_mapin will use this addr to calculate
9985 9985 * an offset, and then set b_addr to the kernel virtual
9986 9986 * address it allocated for us.
9987 9987 */
9988 9988 ASSERT(bp->b_un.b_addr == 0);
9989 9989
9990 9990 bp->b_edev = 0;
9991 9991 bp->b_dev = 0;
9992 9992 bp->b_lblkno = lbtodb(io_off);
9993 9993 bp->b_file = vp;
9994 9994 bp->b_offset = (offset_t)off;
9995 9995 bp_mapin(bp);
9996 9996
9997 9997 /*
9998 9998 * If doing a write beyond what we believe is EOF,
9999 9999 * don't bother trying to read the pages from the
10000 10000 * server, we'll just zero the pages here. We
10001 10001 * don't check that the rw flag is S_WRITE here
10002 10002 * because some implementations may attempt a
10003 10003 * read access to the buffer before copying data.
10004 10004 */
10005 10005 mutex_enter(&rp->r_statelock);
10006 10006 if (io_off >= rp->r_size && seg == segkmap) {
10007 10007 mutex_exit(&rp->r_statelock);
10008 10008 bzero(bp->b_un.b_addr, io_len);
10009 10009 } else {
10010 10010 mutex_exit(&rp->r_statelock);
10011 10011 error = nfs4_bio(bp, NULL, cr, FALSE);
10012 10012 }
10013 10013
10014 10014 /*
10015 10015 * Unmap the buffer before freeing it.
10016 10016 */
10017 10017 bp_mapout(bp);
10018 10018 pageio_done(bp);
10019 10019
10020 10020 savepp = pp;
10021 10021 do {
10022 10022 pp->p_fsdata = C_NOCOMMIT;
10023 10023 } while ((pp = pp->p_next) != savepp);
10024 10024
10025 10025 if (error == NFS_EOF) {
10026 10026 /*
10027 10027 * If doing a write system call just return
10028 10028 * zeroed pages, else user tried to get pages
10029 10029 * beyond EOF, return error. We don't check
10030 10030 * that the rw flag is S_WRITE here because
10031 10031 * some implementations may attempt a read
10032 10032 * access to the buffer before copying data.
10033 10033 */
10034 10034 if (seg == segkmap)
10035 10035 error = 0;
10036 10036 else
10037 10037 error = EFAULT;
10038 10038 }
10039 10039
10040 10040 if (!readahead_issued && !error) {
10041 10041 mutex_enter(&rp->r_statelock);
10042 10042 rp->r_nextr = io_off + io_len;
10043 10043 mutex_exit(&rp->r_statelock);
10044 10044 }
10045 10045 }
10046 10046 }
10047 10047
10048 10048 out:
10049 10049 if (pl == NULL)
10050 10050 return (error);
10051 10051
10052 10052 if (error) {
10053 10053 if (pp != NULL)
10054 10054 pvn_read_done(pp, B_ERROR);
10055 10055 return (error);
10056 10056 }
10057 10057
10058 10058 if (pagefound) {
10059 10059 se_t se = (rw == S_CREATE ? SE_EXCL : SE_SHARED);
10060 10060
10061 10061 /*
10062 10062 * Page exists in the cache, acquire the appropriate lock.
10063 10063 * If this fails, start all over again.
10064 10064 */
10065 10065 if ((pp = page_lookup(vp, off, se)) == NULL) {
10066 10066 #ifdef DEBUG
10067 10067 nfs4_lostpage++;
10068 10068 #endif
10069 10069 goto reread;
10070 10070 }
10071 10071 pl[0] = pp;
10072 10072 pl[1] = NULL;
10073 10073 return (0);
10074 10074 }
10075 10075
10076 10076 if (pp != NULL)
10077 10077 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
10078 10078
10079 10079 return (error);
10080 10080 }
10081 10081
10082 10082 static void
10083 10083 nfs4_readahead(vnode_t *vp, u_offset_t blkoff, caddr_t addr, struct seg *seg,
10084 10084 cred_t *cr)
10085 10085 {
10086 10086 int error;
10087 10087 page_t *pp;
10088 10088 u_offset_t io_off;
10089 10089 size_t io_len;
10090 10090 struct buf *bp;
10091 10091 uint_t bsize, blksize;
10092 10092 rnode4_t *rp = VTOR4(vp);
10093 10093 page_t *savepp;
10094 10094
10095 10095 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
10096 10096
10097 10097 bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
10098 10098
10099 10099 mutex_enter(&rp->r_statelock);
10100 10100 if (blkoff < rp->r_size && blkoff + bsize > rp->r_size) {
10101 10101 /*
10102 10102 * If less than a block left in file read less
10103 10103 * than a block.
10104 10104 */
10105 10105 blksize = rp->r_size - blkoff;
10106 10106 } else
10107 10107 blksize = bsize;
10108 10108 mutex_exit(&rp->r_statelock);
10109 10109
10110 10110 pp = pvn_read_kluster(vp, blkoff, segkmap, addr,
10111 10111 &io_off, &io_len, blkoff, blksize, 1);
10112 10112 /*
10113 10113 * The isra flag passed to the kluster function is 1, we may have
10114 10114 * gotten a return value of NULL for a variety of reasons (# of free
10115 10115 * pages < minfree, someone entered the page on the vnode etc). In all
10116 10116 * cases, we want to punt on the readahead.
10117 10117 */
10118 10118 if (pp == NULL)
10119 10119 return;
10120 10120
10121 10121 /*
10122 10122 * Now round the request size up to page boundaries.
10123 10123 * This ensures that the entire page will be
10124 10124 * initialized to zeroes if EOF is encountered.
10125 10125 */
10126 10126 io_len = ptob(btopr(io_len));
10127 10127
10128 10128 bp = pageio_setup(pp, io_len, vp, B_READ);
10129 10129 ASSERT(bp != NULL);
10130 10130
10131 10131 /*
10132 10132 * pageio_setup should have set b_addr to 0. This is correct since
10133 10133 * we want to do I/O on a page boundary. bp_mapin() will use this addr
10134 10134 * to calculate an offset, and then set b_addr to the kernel virtual
10135 10135 * address it allocated for us.
10136 10136 */
10137 10137 ASSERT(bp->b_un.b_addr == 0);
10138 10138
10139 10139 bp->b_edev = 0;
10140 10140 bp->b_dev = 0;
10141 10141 bp->b_lblkno = lbtodb(io_off);
10142 10142 bp->b_file = vp;
10143 10143 bp->b_offset = (offset_t)blkoff;
10144 10144 bp_mapin(bp);
10145 10145
10146 10146 /*
10147 10147 * If doing a write beyond what we believe is EOF, don't bother trying
10148 10148 * to read the pages from the server, we'll just zero the pages here.
10149 10149 * We don't check that the rw flag is S_WRITE here because some
10150 10150 * implementations may attempt a read access to the buffer before
10151 10151 * copying data.
10152 10152 */
10153 10153 mutex_enter(&rp->r_statelock);
10154 10154 if (io_off >= rp->r_size && seg == segkmap) {
10155 10155 mutex_exit(&rp->r_statelock);
10156 10156 bzero(bp->b_un.b_addr, io_len);
10157 10157 error = 0;
10158 10158 } else {
10159 10159 mutex_exit(&rp->r_statelock);
10160 10160 error = nfs4_bio(bp, NULL, cr, TRUE);
10161 10161 if (error == NFS_EOF)
10162 10162 error = 0;
10163 10163 }
10164 10164
10165 10165 /*
10166 10166 * Unmap the buffer before freeing it.
10167 10167 */
10168 10168 bp_mapout(bp);
10169 10169 pageio_done(bp);
10170 10170
10171 10171 savepp = pp;
10172 10172 do {
10173 10173 pp->p_fsdata = C_NOCOMMIT;
10174 10174 } while ((pp = pp->p_next) != savepp);
10175 10175
10176 10176 pvn_read_done(pp, error ? B_READ | B_ERROR : B_READ);
10177 10177
10178 10178 /*
10179 10179 * In case of error set readahead offset
10180 10180 * to the lowest offset.
10181 10181 * pvn_read_done() calls VN_DISPOSE to destroy the pages
10182 10182 */
10183 10183 if (error && rp->r_nextr > io_off) {
10184 10184 mutex_enter(&rp->r_statelock);
10185 10185 if (rp->r_nextr > io_off)
10186 10186 rp->r_nextr = io_off;
10187 10187 mutex_exit(&rp->r_statelock);
10188 10188 }
10189 10189 }
10190 10190
10191 10191 /*
10192 10192 * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE}
10193 10193 * If len == 0, do from off to EOF.
10194 10194 *
10195 10195 * The normal cases should be len == 0 && off == 0 (entire vp list) or
10196 10196 * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
10197 10197 * (from pageout).
10198 10198 */
10199 10199 /* ARGSUSED */
10200 10200 static int
10201 10201 nfs4_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
10202 10202 caller_context_t *ct)
10203 10203 {
10204 10204 int error;
10205 10205 rnode4_t *rp;
10206 10206
10207 10207 ASSERT(cr != NULL);
10208 10208
10209 10209 if (!(flags & B_ASYNC) && nfs_zone() != VTOMI4(vp)->mi_zone)
10210 10210 return (EIO);
10211 10211
10212 10212 rp = VTOR4(vp);
10213 10213 if (IS_SHADOW(vp, rp))
10214 10214 vp = RTOV4(rp);
10215 10215
10216 10216 /*
10217 10217 * XXX - Why should this check be made here?
10218 10218 */
10219 10219 if (vp->v_flag & VNOMAP)
10220 10220 return (ENOSYS);
10221 10221
10222 10222 if (len == 0 && !(flags & B_INVAL) &&
10223 10223 (vp->v_vfsp->vfs_flag & VFS_RDONLY))
10224 10224 return (0);
10225 10225
10226 10226 mutex_enter(&rp->r_statelock);
10227 10227 rp->r_count++;
10228 10228 mutex_exit(&rp->r_statelock);
10229 10229 error = nfs4_putpages(vp, off, len, flags, cr);
10230 10230 mutex_enter(&rp->r_statelock);
10231 10231 rp->r_count--;
10232 10232 cv_broadcast(&rp->r_cv);
10233 10233 mutex_exit(&rp->r_statelock);
10234 10234
10235 10235 return (error);
10236 10236 }
10237 10237
10238 10238 /*
10239 10239 * Write out a single page, possibly klustering adjacent dirty pages.
10240 10240 */
10241 10241 int
10242 10242 nfs4_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
10243 10243 int flags, cred_t *cr)
10244 10244 {
10245 10245 u_offset_t io_off;
10246 10246 u_offset_t lbn_off;
10247 10247 u_offset_t lbn;
10248 10248 size_t io_len;
10249 10249 uint_t bsize;
10250 10250 int error;
10251 10251 rnode4_t *rp;
10252 10252
10253 10253 ASSERT(!(vp->v_vfsp->vfs_flag & VFS_RDONLY));
10254 10254 ASSERT(pp != NULL);
10255 10255 ASSERT(cr != NULL);
10256 10256 ASSERT((flags & B_ASYNC) || nfs_zone() == VTOMI4(vp)->mi_zone);
10257 10257
10258 10258 rp = VTOR4(vp);
10259 10259 ASSERT(rp->r_count > 0);
10260 10260 ASSERT(!IS_SHADOW(vp, rp));
10261 10261
10262 10262 bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
10263 10263 lbn = pp->p_offset / bsize;
10264 10264 lbn_off = lbn * bsize;
10265 10265
10266 10266 /*
10267 10267 * Find a kluster that fits in one block, or in
10268 10268 * one page if pages are bigger than blocks. If
10269 10269 * there is less file space allocated than a whole
10270 10270 * page, we'll shorten the i/o request below.
10271 10271 */
10272 10272 pp = pvn_write_kluster(vp, pp, &io_off, &io_len, lbn_off,
10273 10273 roundup(bsize, PAGESIZE), flags);
10274 10274
10275 10275 /*
10276 10276 * pvn_write_kluster shouldn't have returned a page with offset
10277 10277 * behind the original page we were given. Verify that.
10278 10278 */
10279 10279 ASSERT((pp->p_offset / bsize) >= lbn);
10280 10280
10281 10281 /*
10282 10282 * Now pp will have the list of kept dirty pages marked for
10283 10283 * write back. It will also handle invalidation and freeing
10284 10284 * of pages that are not dirty. Check for page length rounding
10285 10285 * problems.
10286 10286 */
10287 10287 if (io_off + io_len > lbn_off + bsize) {
10288 10288 ASSERT((io_off + io_len) - (lbn_off + bsize) < PAGESIZE);
10289 10289 io_len = lbn_off + bsize - io_off;
10290 10290 }
10291 10291 /*
10292 10292 * The R4MODINPROGRESS flag makes sure that nfs4_bio() sees a
10293 10293 * consistent value of r_size. R4MODINPROGRESS is set in writerp4().
10294 10294 * When R4MODINPROGRESS is set it indicates that a uiomove() is in
10295 10295 * progress and the r_size has not been made consistent with the
10296 10296 * new size of the file. When the uiomove() completes the r_size is
10297 10297 * updated and the R4MODINPROGRESS flag is cleared.
10298 10298 *
10299 10299 * The R4MODINPROGRESS flag makes sure that nfs4_bio() sees a
10300 10300 * consistent value of r_size. Without this handshaking, it is
10301 10301 * possible that nfs4_bio() picks up the old value of r_size
10302 10302 * before the uiomove() in writerp4() completes. This will result
10303 10303 * in the write through nfs4_bio() being dropped.
10304 10304 *
10305 10305 * More precisely, there is a window between the time the uiomove()
10306 10306 * completes and the time the r_size is updated. If a VOP_PUTPAGE()
10307 10307 * operation intervenes in this window, the page will be picked up,
10308 10308 * because it is dirty (it will be unlocked, unless it was
10309 10309 * pagecreate'd). When the page is picked up as dirty, the dirty
10310 10310 * bit is reset (pvn_getdirty()). In nfs4write(), r_size is
10311 10311 * checked. This will still be the old size. Therefore the page will
10312 10312 * not be written out. When segmap_release() calls VOP_PUTPAGE(),
10313 10313 * the page will be found to be clean and the write will be dropped.
10314 10314 */
10315 10315 if (rp->r_flags & R4MODINPROGRESS) {
10316 10316 mutex_enter(&rp->r_statelock);
10317 10317 if ((rp->r_flags & R4MODINPROGRESS) &&
10318 10318 rp->r_modaddr + MAXBSIZE > io_off &&
10319 10319 rp->r_modaddr < io_off + io_len) {
10320 10320 page_t *plist;
10321 10321 /*
10322 10322 * A write is in progress for this region of the file.
10323 10323 * If we did not detect R4MODINPROGRESS here then this
10324 10324 * path through nfs_putapage() would eventually go to
10325 10325 * nfs4_bio() and may not write out all of the data
10326 10326 * in the pages. We end up losing data. So we decide
10327 10327 * to set the modified bit on each page in the page
10328 10328 * list and mark the rnode with R4DIRTY. This write
10329 10329 * will be restarted at some later time.
10330 10330 */
10331 10331 plist = pp;
10332 10332 while (plist != NULL) {
10333 10333 pp = plist;
10334 10334 page_sub(&plist, pp);
10335 10335 hat_setmod(pp);
10336 10336 page_io_unlock(pp);
10337 10337 page_unlock(pp);
10338 10338 }
10339 10339 rp->r_flags |= R4DIRTY;
10340 10340 mutex_exit(&rp->r_statelock);
10341 10341 if (offp)
10342 10342 *offp = io_off;
10343 10343 if (lenp)
10344 10344 *lenp = io_len;
10345 10345 return (0);
10346 10346 }
10347 10347 mutex_exit(&rp->r_statelock);
10348 10348 }
10349 10349
10350 10350 if (flags & B_ASYNC) {
10351 10351 error = nfs4_async_putapage(vp, pp, io_off, io_len, flags, cr,
10352 10352 nfs4_sync_putapage);
10353 10353 } else
10354 10354 error = nfs4_sync_putapage(vp, pp, io_off, io_len, flags, cr);
10355 10355
10356 10356 if (offp)
10357 10357 *offp = io_off;
10358 10358 if (lenp)
10359 10359 *lenp = io_len;
10360 10360 return (error);
10361 10361 }
10362 10362
10363 10363 static int
10364 10364 nfs4_sync_putapage(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
10365 10365 int flags, cred_t *cr)
10366 10366 {
10367 10367 int error;
10368 10368 rnode4_t *rp;
10369 10369
10370 10370 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
10371 10371
10372 10372 flags |= B_WRITE;
10373 10373
10374 10374 error = nfs4_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
10375 10375
10376 10376 rp = VTOR4(vp);
10377 10377
10378 10378 if ((error == ENOSPC || error == EDQUOT || error == EFBIG ||
10379 10379 error == EACCES) &&
10380 10380 (flags & (B_INVAL|B_FORCE)) != (B_INVAL|B_FORCE)) {
10381 10381 if (!(rp->r_flags & R4OUTOFSPACE)) {
10382 10382 mutex_enter(&rp->r_statelock);
10383 10383 rp->r_flags |= R4OUTOFSPACE;
10384 10384 mutex_exit(&rp->r_statelock);
10385 10385 }
10386 10386 flags |= B_ERROR;
10387 10387 pvn_write_done(pp, flags);
10388 10388 /*
10389 10389 * If this was not an async thread, then try again to
10390 10390 * write out the pages, but this time, also destroy
10391 10391 * them whether or not the write is successful. This
10392 10392 * will prevent memory from filling up with these
10393 10393 * pages and destroying them is the only alternative
10394 10394 * if they can't be written out.
10395 10395 *
10396 10396 * Don't do this if this is an async thread because
10397 10397 * when the pages are unlocked in pvn_write_done,
10398 10398 * some other thread could have come along, locked
10399 10399 * them, and queued for an async thread. It would be
10400 10400 * possible for all of the async threads to be tied
10401 10401 * up waiting to lock the pages again and they would
10402 10402 * all already be locked and waiting for an async
10403 10403 * thread to handle them. Deadlock.
10404 10404 */
10405 10405 if (!(flags & B_ASYNC)) {
10406 10406 error = nfs4_putpage(vp, io_off, io_len,
10407 10407 B_INVAL | B_FORCE, cr, NULL);
10408 10408 }
10409 10409 } else {
10410 10410 if (error)
10411 10411 flags |= B_ERROR;
10412 10412 else if (rp->r_flags & R4OUTOFSPACE) {
10413 10413 mutex_enter(&rp->r_statelock);
10414 10414 rp->r_flags &= ~R4OUTOFSPACE;
10415 10415 mutex_exit(&rp->r_statelock);
10416 10416 }
10417 10417 pvn_write_done(pp, flags);
10418 10418 if (freemem < desfree)
10419 10419 (void) nfs4_commit_vp(vp, (u_offset_t)0, 0, cr,
10420 10420 NFS4_WRITE_NOWAIT);
10421 10421 }
10422 10422
10423 10423 return (error);
10424 10424 }
10425 10425
10426 10426 #ifdef DEBUG
10427 10427 int nfs4_force_open_before_mmap = 0;
10428 10428 #endif
10429 10429
10430 10430 /* ARGSUSED */
10431 10431 static int
10432 10432 nfs4_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
10433 10433 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
10434 10434 caller_context_t *ct)
10435 10435 {
10436 10436 struct segvn_crargs vn_a;
10437 10437 int error = 0;
10438 10438 rnode4_t *rp = VTOR4(vp);
10439 10439 mntinfo4_t *mi = VTOMI4(vp);
10440 10440
10441 10441 if (nfs_zone() != VTOMI4(vp)->mi_zone)
10442 10442 return (EIO);
10443 10443
10444 10444 if (vp->v_flag & VNOMAP)
10445 10445 return (ENOSYS);
10446 10446
10447 10447 if (off < 0 || (off + len) < 0)
10448 10448 return (ENXIO);
10449 10449
10450 10450 if (vp->v_type != VREG)
10451 10451 return (ENODEV);
10452 10452
10453 10453 /*
10454 10454 * If the file is delegated to the client don't do anything.
10455 10455 * If the file is not delegated, then validate the data cache.
10456 10456 */
10457 10457 mutex_enter(&rp->r_statev4_lock);
10458 10458 if (rp->r_deleg_type == OPEN_DELEGATE_NONE) {
10459 10459 mutex_exit(&rp->r_statev4_lock);
10460 10460 error = nfs4_validate_caches(vp, cr);
10461 10461 if (error)
10462 10462 return (error);
10463 10463 } else {
10464 10464 mutex_exit(&rp->r_statev4_lock);
10465 10465 }
10466 10466
10467 10467 /*
10468 10468 * Check to see if the vnode is currently marked as not cachable.
10469 10469 * This means portions of the file are locked (through VOP_FRLOCK).
10470 10470 * In this case the map request must be refused. We use
10471 10471 * rp->r_lkserlock to avoid a race with concurrent lock requests.
10472 10472 *
10473 10473 * Atomically increment r_inmap after acquiring r_rwlock. The
10474 10474 * idea here is to acquire r_rwlock to block read/write and
10475 10475 * not to protect r_inmap. r_inmap will inform nfs4_read/write()
10476 10476 * that we are in nfs4_map(). Now, r_rwlock is acquired in order
10477 10477 * and we can prevent the deadlock that would have occurred
10478 10478 * when nfs4_addmap() would have acquired it out of order.
10479 10479 *
10480 10480 * Since we are not protecting r_inmap by any lock, we do not
10481 10481 * hold any lock when we decrement it. We atomically decrement
10482 10482 * r_inmap after we release r_lkserlock.
10483 10483 */
10484 10484
10485 10485 if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, INTR4(vp)))
10486 10486 return (EINTR);
10487 10487 atomic_inc_uint(&rp->r_inmap);
10488 10488 nfs_rw_exit(&rp->r_rwlock);
10489 10489
10490 10490 if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR4(vp))) {
10491 10491 atomic_dec_uint(&rp->r_inmap);
10492 10492 return (EINTR);
10493 10493 }
10494 10494
10495 10495
10496 10496 if (vp->v_flag & VNOCACHE) {
10497 10497 error = EAGAIN;
10498 10498 goto done;
10499 10499 }
10500 10500
10501 10501 /*
10502 10502 * Don't allow concurrent locks and mapping if mandatory locking is
10503 10503 * enabled.
10504 10504 */
10505 10505 if (flk_has_remote_locks(vp)) {
10506 10506 struct vattr va;
10507 10507 va.va_mask = AT_MODE;
10508 10508 error = nfs4getattr(vp, &va, cr);
10509 10509 if (error != 0)
10510 10510 goto done;
10511 10511 if (MANDLOCK(vp, va.va_mode)) {
10512 10512 error = EAGAIN;
10513 10513 goto done;
10514 10514 }
10515 10515 }
10516 10516
10517 10517 /*
10518 10518 * It is possible that the rnode has a lost lock request that we
10519 10519 * are still trying to recover, and that the request conflicts with
10520 10520 * this map request.
10521 10521 *
10522 10522 * An alternative approach would be for nfs4_safemap() to consider
10523 10523 * queued lock requests when deciding whether to set or clear
10524 10524 * VNOCACHE. This would require the frlock code path to call
10525 10525 * nfs4_safemap() after enqueing a lost request.
10526 10526 */
10527 10527 if (nfs4_map_lost_lock_conflict(vp)) {
10528 10528 error = EAGAIN;
10529 10529 goto done;
10530 10530 }
10531 10531
10532 10532 as_rangelock(as);
10533 10533 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
10534 10534 if (error != 0) {
10535 10535 as_rangeunlock(as);
10536 10536 goto done;
10537 10537 }
10538 10538
10539 10539 if (vp->v_type == VREG) {
10540 10540 /*
10541 10541 * We need to retrieve the open stream
10542 10542 */
10543 10543 nfs4_open_stream_t *osp = NULL;
10544 10544 nfs4_open_owner_t *oop = NULL;
10545 10545
10546 10546 oop = find_open_owner(cr, NFS4_PERM_CREATED, mi);
10547 10547 if (oop != NULL) {
10548 10548 /* returns with 'os_sync_lock' held */
10549 10549 osp = find_open_stream(oop, rp);
10550 10550 open_owner_rele(oop);
10551 10551 }
10552 10552 if (osp == NULL) {
10553 10553 #ifdef DEBUG
10554 10554 if (nfs4_force_open_before_mmap) {
10555 10555 error = EIO;
10556 10556 goto done;
10557 10557 }
10558 10558 #endif
10559 10559 /* returns with 'os_sync_lock' held */
10560 10560 error = open_and_get_osp(vp, cr, &osp);
10561 10561 if (osp == NULL) {
10562 10562 NFS4_DEBUG(nfs4_mmap_debug, (CE_NOTE,
10563 10563 "nfs4_map: we tried to OPEN the file "
10564 10564 "but again no osp, so fail with EIO"));
10565 10565 goto done;
10566 10566 }
10567 10567 }
10568 10568
10569 10569 if (osp->os_failed_reopen) {
10570 10570 mutex_exit(&osp->os_sync_lock);
10571 10571 open_stream_rele(osp, rp);
10572 10572 NFS4_DEBUG(nfs4_open_stream_debug, (CE_NOTE,
10573 10573 "nfs4_map: os_failed_reopen set on "
10574 10574 "osp %p, cr %p, rp %s", (void *)osp,
10575 10575 (void *)cr, rnode4info(rp)));
10576 10576 error = EIO;
10577 10577 goto done;
10578 10578 }
10579 10579 mutex_exit(&osp->os_sync_lock);
10580 10580 open_stream_rele(osp, rp);
10581 10581 }
10582 10582
10583 10583 vn_a.vp = vp;
10584 10584 vn_a.offset = off;
10585 10585 vn_a.type = (flags & MAP_TYPE);
10586 10586 vn_a.prot = (uchar_t)prot;
10587 10587 vn_a.maxprot = (uchar_t)maxprot;
10588 10588 vn_a.flags = (flags & ~MAP_TYPE);
10589 10589 vn_a.cred = cr;
10590 10590 vn_a.amp = NULL;
10591 10591 vn_a.szc = 0;
10592 10592 vn_a.lgrp_mem_policy_flags = 0;
10593 10593
10594 10594 error = as_map(as, *addrp, len, segvn_create, &vn_a);
10595 10595 as_rangeunlock(as);
10596 10596
10597 10597 done:
10598 10598 nfs_rw_exit(&rp->r_lkserlock);
10599 10599 atomic_dec_uint(&rp->r_inmap);
10600 10600 return (error);
10601 10601 }
10602 10602
10603 10603 /*
10604 10604 * We're most likely dealing with a kernel module that likes to READ
10605 10605 * and mmap without OPENing the file (ie: lookup/read/mmap), so lets
10606 10606 * officially OPEN the file to create the necessary client state
10607 10607 * for bookkeeping of os_mmap_read/write counts.
10608 10608 *
10609 10609 * Since VOP_MAP only passes in a pointer to the vnode rather than
10610 10610 * a double pointer, we can't handle the case where nfs4open_otw()
10611 10611 * returns a different vnode than the one passed into VOP_MAP (since
10612 10612 * VOP_DELMAP will not see the vnode nfs4open_otw used). In this case,
10613 10613 * we return NULL and let nfs4_map() fail. Note: the only case where
10614 10614 * this should happen is if the file got removed and replaced with the
10615 10615 * same name on the server (in addition to the fact that we're trying
10616 10616 * to VOP_MAP withouth VOP_OPENing the file in the first place).
10617 10617 */
10618 10618 static int
10619 10619 open_and_get_osp(vnode_t *map_vp, cred_t *cr, nfs4_open_stream_t **ospp)
10620 10620 {
10621 10621 rnode4_t *rp, *drp;
10622 10622 vnode_t *dvp, *open_vp;
10623 10623 char file_name[MAXNAMELEN];
10624 10624 int just_created;
10625 10625 nfs4_open_stream_t *osp;
10626 10626 nfs4_open_owner_t *oop;
10627 10627 int error;
10628 10628
10629 10629 *ospp = NULL;
10630 10630 open_vp = map_vp;
10631 10631
10632 10632 rp = VTOR4(open_vp);
10633 10633 if ((error = vtodv(open_vp, &dvp, cr, TRUE)) != 0)
10634 10634 return (error);
10635 10635 drp = VTOR4(dvp);
10636 10636
10637 10637 if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR4(dvp))) {
10638 10638 VN_RELE(dvp);
10639 10639 return (EINTR);
10640 10640 }
10641 10641
10642 10642 if ((error = vtoname(open_vp, file_name, MAXNAMELEN)) != 0) {
10643 10643 nfs_rw_exit(&drp->r_rwlock);
10644 10644 VN_RELE(dvp);
10645 10645 return (error);
10646 10646 }
10647 10647
10648 10648 mutex_enter(&rp->r_statev4_lock);
10649 10649 if (rp->created_v4) {
10650 10650 rp->created_v4 = 0;
10651 10651 mutex_exit(&rp->r_statev4_lock);
10652 10652
10653 10653 dnlc_update(dvp, file_name, open_vp);
10654 10654 /* This is needed so we don't bump the open ref count */
10655 10655 just_created = 1;
10656 10656 } else {
10657 10657 mutex_exit(&rp->r_statev4_lock);
10658 10658 just_created = 0;
10659 10659 }
10660 10660
10661 10661 VN_HOLD(map_vp);
10662 10662
10663 10663 error = nfs4open_otw(dvp, file_name, NULL, &open_vp, cr, 0, FREAD, 0,
10664 10664 just_created);
10665 10665 if (error) {
10666 10666 nfs_rw_exit(&drp->r_rwlock);
10667 10667 VN_RELE(dvp);
10668 10668 VN_RELE(map_vp);
10669 10669 return (error);
10670 10670 }
10671 10671
10672 10672 nfs_rw_exit(&drp->r_rwlock);
10673 10673 VN_RELE(dvp);
10674 10674
10675 10675 /*
10676 10676 * If nfs4open_otw() returned a different vnode then "undo"
10677 10677 * the open and return failure to the caller.
10678 10678 */
10679 10679 if (!VN_CMP(open_vp, map_vp)) {
10680 10680 nfs4_error_t e;
10681 10681
10682 10682 NFS4_DEBUG(nfs4_mmap_debug, (CE_NOTE, "open_and_get_osp: "
10683 10683 "open returned a different vnode"));
10684 10684 /*
10685 10685 * If there's an error, ignore it,
10686 10686 * and let VOP_INACTIVE handle it.
10687 10687 */
10688 10688 (void) nfs4close_one(open_vp, NULL, cr, FREAD, NULL, &e,
10689 10689 CLOSE_NORM, 0, 0, 0);
10690 10690 VN_RELE(map_vp);
10691 10691 return (EIO);
10692 10692 }
10693 10693
10694 10694 VN_RELE(map_vp);
10695 10695
10696 10696 oop = find_open_owner(cr, NFS4_PERM_CREATED, VTOMI4(open_vp));
10697 10697 if (!oop) {
10698 10698 nfs4_error_t e;
10699 10699
10700 10700 NFS4_DEBUG(nfs4_mmap_debug, (CE_NOTE, "open_and_get_osp: "
10701 10701 "no open owner"));
10702 10702 /*
10703 10703 * If there's an error, ignore it,
10704 10704 * and let VOP_INACTIVE handle it.
10705 10705 */
10706 10706 (void) nfs4close_one(open_vp, NULL, cr, FREAD, NULL, &e,
10707 10707 CLOSE_NORM, 0, 0, 0);
10708 10708 return (EIO);
10709 10709 }
10710 10710 osp = find_open_stream(oop, rp);
10711 10711 open_owner_rele(oop);
10712 10712 *ospp = osp;
10713 10713 return (0);
10714 10714 }
10715 10715
10716 10716 /*
10717 10717 * Please be aware that when this function is called, the address space write
10718 10718 * a_lock is held. Do not put over the wire calls in this function.
10719 10719 */
10720 10720 /* ARGSUSED */
10721 10721 static int
10722 10722 nfs4_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
10723 10723 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
10724 10724 caller_context_t *ct)
10725 10725 {
10726 10726 rnode4_t *rp;
10727 10727 int error = 0;
10728 10728 mntinfo4_t *mi;
10729 10729
10730 10730 mi = VTOMI4(vp);
10731 10731 rp = VTOR4(vp);
10732 10732
10733 10733 if (nfs_zone() != mi->mi_zone)
10734 10734 return (EIO);
10735 10735 if (vp->v_flag & VNOMAP)
10736 10736 return (ENOSYS);
10737 10737
10738 10738 /*
10739 10739 * Don't need to update the open stream first, since this
10740 10740 * mmap can't add any additional share access that isn't
10741 10741 * already contained in the open stream (for the case where we
10742 10742 * open/mmap/only update rp->r_mapcnt/server reboots/reopen doesn't
10743 10743 * take into account os_mmap_read[write] counts).
10744 10744 */
10745 10745 atomic_add_long((ulong_t *)&rp->r_mapcnt, btopr(len));
10746 10746
10747 10747 if (vp->v_type == VREG) {
10748 10748 /*
10749 10749 * We need to retrieve the open stream and update the counts.
10750 10750 * If there is no open stream here, something is wrong.
10751 10751 */
10752 10752 nfs4_open_stream_t *osp = NULL;
10753 10753 nfs4_open_owner_t *oop = NULL;
10754 10754
10755 10755 oop = find_open_owner(cr, NFS4_PERM_CREATED, mi);
10756 10756 if (oop != NULL) {
10757 10757 /* returns with 'os_sync_lock' held */
10758 10758 osp = find_open_stream(oop, rp);
10759 10759 open_owner_rele(oop);
10760 10760 }
10761 10761 if (osp == NULL) {
10762 10762 NFS4_DEBUG(nfs4_mmap_debug, (CE_NOTE,
10763 10763 "nfs4_addmap: we should have an osp"
10764 10764 "but we don't, so fail with EIO"));
10765 10765 error = EIO;
10766 10766 goto out;
10767 10767 }
10768 10768
10769 10769 NFS4_DEBUG(nfs4_mmap_debug, (CE_NOTE, "nfs4_addmap: osp %p,"
10770 10770 " pages %ld, prot 0x%x", (void *)osp, btopr(len), prot));
10771 10771
10772 10772 /*
10773 10773 * Update the map count in the open stream.
10774 10774 * This is necessary in the case where we
10775 10775 * open/mmap/close/, then the server reboots, and we
10776 10776 * attempt to reopen. If the mmap doesn't add share
10777 10777 * access then we send an invalid reopen with
10778 10778 * access = NONE.
10779 10779 *
10780 10780 * We need to specifically check each PROT_* so a mmap
10781 10781 * call of (PROT_WRITE | PROT_EXEC) will ensure us both
10782 10782 * read and write access. A simple comparison of prot
10783 10783 * to ~PROT_WRITE to determine read access is insufficient
10784 10784 * since prot can be |= with PROT_USER, etc.
10785 10785 */
10786 10786
10787 10787 /*
10788 10788 * Unless we're MAP_SHARED, no sense in adding os_mmap_write
10789 10789 */
10790 10790 if ((flags & MAP_SHARED) && (maxprot & PROT_WRITE))
10791 10791 osp->os_mmap_write += btopr(len);
10792 10792 if (maxprot & PROT_READ)
10793 10793 osp->os_mmap_read += btopr(len);
10794 10794 if (maxprot & PROT_EXEC)
10795 10795 osp->os_mmap_read += btopr(len);
10796 10796 /*
10797 10797 * Ensure that os_mmap_read gets incremented, even if
10798 10798 * maxprot were to look like PROT_NONE.
10799 10799 */
10800 10800 if (!(maxprot & PROT_READ) && !(maxprot & PROT_WRITE) &&
10801 10801 !(maxprot & PROT_EXEC))
10802 10802 osp->os_mmap_read += btopr(len);
10803 10803 osp->os_mapcnt += btopr(len);
10804 10804 mutex_exit(&osp->os_sync_lock);
10805 10805 open_stream_rele(osp, rp);
10806 10806 }
10807 10807
10808 10808 out:
10809 10809 /*
10810 10810 * If we got an error, then undo our
10811 10811 * incrementing of 'r_mapcnt'.
10812 10812 */
10813 10813
10814 10814 if (error) {
10815 10815 atomic_add_long((ulong_t *)&rp->r_mapcnt, -btopr(len));
10816 10816 ASSERT(rp->r_mapcnt >= 0);
10817 10817 }
10818 10818 return (error);
10819 10819 }
10820 10820
10821 10821 /* ARGSUSED */
10822 10822 static int
10823 10823 nfs4_cmp(vnode_t *vp1, vnode_t *vp2, caller_context_t *ct)
10824 10824 {
10825 10825
10826 10826 return (VTOR4(vp1) == VTOR4(vp2));
10827 10827 }
10828 10828
10829 10829 /* ARGSUSED */
10830 10830 static int
10831 10831 nfs4_frlock(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
10832 10832 offset_t offset, struct flk_callback *flk_cbp, cred_t *cr,
10833 10833 caller_context_t *ct)
10834 10834 {
10835 10835 int rc;
10836 10836 u_offset_t start, end;
10837 10837 rnode4_t *rp;
10838 10838 int error = 0, intr = INTR4(vp);
10839 10839 nfs4_error_t e;
10840 10840
10841 10841 if (nfs_zone() != VTOMI4(vp)->mi_zone)
10842 10842 return (EIO);
10843 10843
10844 10844 /* check for valid cmd parameter */
10845 10845 if (cmd != F_GETLK && cmd != F_SETLK && cmd != F_SETLKW)
10846 10846 return (EINVAL);
10847 10847
10848 10848 /* Verify l_type. */
10849 10849 switch (bfp->l_type) {
10850 10850 case F_RDLCK:
10851 10851 if (cmd != F_GETLK && !(flag & FREAD))
10852 10852 return (EBADF);
10853 10853 break;
10854 10854 case F_WRLCK:
10855 10855 if (cmd != F_GETLK && !(flag & FWRITE))
10856 10856 return (EBADF);
10857 10857 break;
10858 10858 case F_UNLCK:
10859 10859 intr = 0;
10860 10860 break;
10861 10861
10862 10862 default:
10863 10863 return (EINVAL);
10864 10864 }
10865 10865
10866 10866 /* check the validity of the lock range */
10867 10867 if (rc = flk_convert_lock_data(vp, bfp, &start, &end, offset))
10868 10868 return (rc);
10869 10869 if (rc = flk_check_lock_data(start, end, MAXEND))
10870 10870 return (rc);
10871 10871
10872 10872 /*
10873 10873 * If the filesystem is mounted using local locking, pass the
10874 10874 * request off to the local locking code.
10875 10875 */
10876 10876 if (VTOMI4(vp)->mi_flags & MI4_LLOCK || vp->v_type != VREG) {
10877 10877 if (cmd == F_SETLK || cmd == F_SETLKW) {
10878 10878 /*
10879 10879 * For complete safety, we should be holding
10880 10880 * r_lkserlock. However, we can't call
10881 10881 * nfs4_safelock and then fs_frlock while
10882 10882 * holding r_lkserlock, so just invoke
10883 10883 * nfs4_safelock and expect that this will
10884 10884 * catch enough of the cases.
10885 10885 */
10886 10886 if (!nfs4_safelock(vp, bfp, cr))
10887 10887 return (EAGAIN);
10888 10888 }
10889 10889 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
10890 10890 }
10891 10891
10892 10892 rp = VTOR4(vp);
10893 10893
10894 10894 /*
10895 10895 * Check whether the given lock request can proceed, given the
10896 10896 * current file mappings.
10897 10897 */
10898 10898 if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_WRITER, intr))
10899 10899 return (EINTR);
10900 10900 if (cmd == F_SETLK || cmd == F_SETLKW) {
10901 10901 if (!nfs4_safelock(vp, bfp, cr)) {
10902 10902 rc = EAGAIN;
10903 10903 goto done;
10904 10904 }
10905 10905 }
10906 10906
10907 10907 /*
10908 10908 * Flush the cache after waiting for async I/O to finish. For new
10909 10909 * locks, this is so that the process gets the latest bits from the
10910 10910 * server. For unlocks, this is so that other clients see the
10911 10911 * latest bits once the file has been unlocked. If currently dirty
10912 10912 * pages can't be flushed, then don't allow a lock to be set. But
10913 10913 * allow unlocks to succeed, to avoid having orphan locks on the
10914 10914 * server.
10915 10915 */
10916 10916 if (cmd != F_GETLK) {
10917 10917 mutex_enter(&rp->r_statelock);
10918 10918 while (rp->r_count > 0) {
10919 10919 if (intr) {
10920 10920 klwp_t *lwp = ttolwp(curthread);
10921 10921
10922 10922 if (lwp != NULL)
10923 10923 lwp->lwp_nostop++;
10924 10924 if (cv_wait_sig(&rp->r_cv,
10925 10925 &rp->r_statelock) == 0) {
10926 10926 if (lwp != NULL)
10927 10927 lwp->lwp_nostop--;
10928 10928 rc = EINTR;
10929 10929 break;
10930 10930 }
10931 10931 if (lwp != NULL)
10932 10932 lwp->lwp_nostop--;
10933 10933 } else
10934 10934 cv_wait(&rp->r_cv, &rp->r_statelock);
10935 10935 }
10936 10936 mutex_exit(&rp->r_statelock);
10937 10937 if (rc != 0)
10938 10938 goto done;
10939 10939 error = nfs4_putpage(vp, (offset_t)0, 0, B_INVAL, cr, ct);
10940 10940 if (error) {
10941 10941 if (error == ENOSPC || error == EDQUOT) {
10942 10942 mutex_enter(&rp->r_statelock);
10943 10943 if (!rp->r_error)
10944 10944 rp->r_error = error;
10945 10945 mutex_exit(&rp->r_statelock);
10946 10946 }
10947 10947 if (bfp->l_type != F_UNLCK) {
10948 10948 rc = ENOLCK;
10949 10949 goto done;
10950 10950 }
10951 10951 }
10952 10952 }
10953 10953
10954 10954 /*
10955 10955 * Call the lock manager to do the real work of contacting
10956 10956 * the server and obtaining the lock.
10957 10957 */
10958 10958 nfs4frlock(NFS4_LCK_CTYPE_NORM, vp, cmd, bfp, flag, offset,
10959 10959 cr, &e, NULL, NULL);
10960 10960 rc = e.error;
10961 10961
10962 10962 if (rc == 0)
10963 10963 nfs4_lockcompletion(vp, cmd);
10964 10964
10965 10965 done:
10966 10966 nfs_rw_exit(&rp->r_lkserlock);
10967 10967
10968 10968 return (rc);
10969 10969 }
10970 10970
10971 10971 /*
10972 10972 * Free storage space associated with the specified vnode. The portion
10973 10973 * to be freed is specified by bfp->l_start and bfp->l_len (already
10974 10974 * normalized to a "whence" of 0).
10975 10975 *
10976 10976 * This is an experimental facility whose continued existence is not
10977 10977 * guaranteed. Currently, we only support the special case
10978 10978 * of l_len == 0, meaning free to end of file.
10979 10979 */
10980 10980 /* ARGSUSED */
10981 10981 static int
10982 10982 nfs4_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
10983 10983 offset_t offset, cred_t *cr, caller_context_t *ct)
10984 10984 {
10985 10985 int error;
10986 10986
10987 10987 if (nfs_zone() != VTOMI4(vp)->mi_zone)
10988 10988 return (EIO);
10989 10989 ASSERT(vp->v_type == VREG);
10990 10990 if (cmd != F_FREESP)
10991 10991 return (EINVAL);
10992 10992
10993 10993 error = convoff(vp, bfp, 0, offset);
10994 10994 if (!error) {
10995 10995 ASSERT(bfp->l_start >= 0);
10996 10996 if (bfp->l_len == 0) {
10997 10997 struct vattr va;
10998 10998
10999 10999 va.va_mask = AT_SIZE;
11000 11000 va.va_size = bfp->l_start;
11001 11001 error = nfs4setattr(vp, &va, 0, cr, NULL);
11002 11002
11003 11003 if (error == 0 && bfp->l_start == 0)
11004 11004 vnevent_truncate(vp, ct);
11005 11005 } else
11006 11006 error = EINVAL;
11007 11007 }
11008 11008
11009 11009 return (error);
11010 11010 }
11011 11011
11012 11012 /* ARGSUSED */
11013 11013 int
11014 11014 nfs4_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct)
11015 11015 {
11016 11016 rnode4_t *rp;
11017 11017 rp = VTOR4(vp);
11018 11018
11019 11019 if (vp->v_type == VREG && IS_SHADOW(vp, rp)) {
11020 11020 vp = RTOV4(rp);
11021 11021 }
11022 11022 *vpp = vp;
11023 11023 return (0);
11024 11024 }
11025 11025
11026 11026 /*
11027 11027 * Setup and add an address space callback to do the work of the delmap call.
11028 11028 * The callback will (and must be) deleted in the actual callback function.
11029 11029 *
11030 11030 * This is done in order to take care of the problem that we have with holding
11031 11031 * the address space's a_lock for a long period of time (e.g. if the NFS server
11032 11032 * is down). Callbacks will be executed in the address space code while the
11033 11033 * a_lock is not held. Holding the address space's a_lock causes things such
11034 11034 * as ps and fork to hang because they are trying to acquire this lock as well.
11035 11035 */
11036 11036 /* ARGSUSED */
11037 11037 static int
11038 11038 nfs4_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
11039 11039 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
11040 11040 caller_context_t *ct)
11041 11041 {
11042 11042 int caller_found;
11043 11043 int error;
11044 11044 rnode4_t *rp;
11045 11045 nfs4_delmap_args_t *dmapp;
11046 11046 nfs4_delmapcall_t *delmap_call;
11047 11047
11048 11048 if (vp->v_flag & VNOMAP)
11049 11049 return (ENOSYS);
11050 11050
11051 11051 /*
11052 11052 * A process may not change zones if it has NFS pages mmap'ed
11053 11053 * in, so we can't legitimately get here from the wrong zone.
11054 11054 */
11055 11055 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
11056 11056
11057 11057 rp = VTOR4(vp);
11058 11058
11059 11059 /*
11060 11060 * The way that the address space of this process deletes its mapping
11061 11061 * of this file is via the following call chains:
11062 11062 * - as_free()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs4_delmap()
11063 11063 * - as_unmap()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs4_delmap()
11064 11064 *
11065 11065 * With the use of address space callbacks we are allowed to drop the
11066 11066 * address space lock, a_lock, while executing the NFS operations that
11067 11067 * need to go over the wire. Returning EAGAIN to the caller of this
11068 11068 * function is what drives the execution of the callback that we add
11069 11069 * below. The callback will be executed by the address space code
11070 11070 * after dropping the a_lock. When the callback is finished, since
11071 11071 * we dropped the a_lock, it must be re-acquired and segvn_unmap()
11072 11072 * is called again on the same segment to finish the rest of the work
11073 11073 * that needs to happen during unmapping.
11074 11074 *
11075 11075 * This action of calling back into the segment driver causes
11076 11076 * nfs4_delmap() to get called again, but since the callback was
11077 11077 * already executed at this point, it already did the work and there
11078 11078 * is nothing left for us to do.
11079 11079 *
11080 11080 * To Summarize:
11081 11081 * - The first time nfs4_delmap is called by the current thread is when
11082 11082 * we add the caller associated with this delmap to the delmap caller
11083 11083 * list, add the callback, and return EAGAIN.
11084 11084 * - The second time in this call chain when nfs4_delmap is called we
11085 11085 * will find this caller in the delmap caller list and realize there
11086 11086 * is no more work to do thus removing this caller from the list and
11087 11087 * returning the error that was set in the callback execution.
11088 11088 */
11089 11089 caller_found = nfs4_find_and_delete_delmapcall(rp, &error);
11090 11090 if (caller_found) {
11091 11091 /*
11092 11092 * 'error' is from the actual delmap operations. To avoid
11093 11093 * hangs, we need to handle the return of EAGAIN differently
11094 11094 * since this is what drives the callback execution.
11095 11095 * In this case, we don't want to return EAGAIN and do the
11096 11096 * callback execution because there are none to execute.
11097 11097 */
11098 11098 if (error == EAGAIN)
11099 11099 return (0);
11100 11100 else
11101 11101 return (error);
11102 11102 }
11103 11103
11104 11104 /* current caller was not in the list */
11105 11105 delmap_call = nfs4_init_delmapcall();
11106 11106
11107 11107 mutex_enter(&rp->r_statelock);
11108 11108 list_insert_tail(&rp->r_indelmap, delmap_call);
11109 11109 mutex_exit(&rp->r_statelock);
11110 11110
11111 11111 dmapp = kmem_alloc(sizeof (nfs4_delmap_args_t), KM_SLEEP);
11112 11112
11113 11113 dmapp->vp = vp;
11114 11114 dmapp->off = off;
11115 11115 dmapp->addr = addr;
11116 11116 dmapp->len = len;
11117 11117 dmapp->prot = prot;
11118 11118 dmapp->maxprot = maxprot;
11119 11119 dmapp->flags = flags;
11120 11120 dmapp->cr = cr;
11121 11121 dmapp->caller = delmap_call;
11122 11122
11123 11123 error = as_add_callback(as, nfs4_delmap_callback, dmapp,
11124 11124 AS_UNMAP_EVENT, addr, len, KM_SLEEP);
11125 11125
11126 11126 return (error ? error : EAGAIN);
11127 11127 }
11128 11128
11129 11129 static nfs4_delmapcall_t *
11130 11130 nfs4_init_delmapcall()
11131 11131 {
11132 11132 nfs4_delmapcall_t *delmap_call;
11133 11133
11134 11134 delmap_call = kmem_alloc(sizeof (nfs4_delmapcall_t), KM_SLEEP);
11135 11135 delmap_call->call_id = curthread;
11136 11136 delmap_call->error = 0;
11137 11137
11138 11138 return (delmap_call);
11139 11139 }
11140 11140
11141 11141 static void
11142 11142 nfs4_free_delmapcall(nfs4_delmapcall_t *delmap_call)
11143 11143 {
11144 11144 kmem_free(delmap_call, sizeof (nfs4_delmapcall_t));
11145 11145 }
11146 11146
11147 11147 /*
11148 11148 * Searches for the current delmap caller (based on curthread) in the list of
11149 11149 * callers. If it is found, we remove it and free the delmap caller.
11150 11150 * Returns:
11151 11151 * 0 if the caller wasn't found
11152 11152 * 1 if the caller was found, removed and freed. *errp will be set
11153 11153 * to what the result of the delmap was.
11154 11154 */
11155 11155 static int
11156 11156 nfs4_find_and_delete_delmapcall(rnode4_t *rp, int *errp)
11157 11157 {
11158 11158 nfs4_delmapcall_t *delmap_call;
11159 11159
11160 11160 /*
11161 11161 * If the list doesn't exist yet, we create it and return
11162 11162 * that the caller wasn't found. No list = no callers.
11163 11163 */
11164 11164 mutex_enter(&rp->r_statelock);
11165 11165 if (!(rp->r_flags & R4DELMAPLIST)) {
11166 11166 /* The list does not exist */
11167 11167 list_create(&rp->r_indelmap, sizeof (nfs4_delmapcall_t),
11168 11168 offsetof(nfs4_delmapcall_t, call_node));
11169 11169 rp->r_flags |= R4DELMAPLIST;
11170 11170 mutex_exit(&rp->r_statelock);
11171 11171 return (0);
11172 11172 } else {
11173 11173 /* The list exists so search it */
11174 11174 for (delmap_call = list_head(&rp->r_indelmap);
11175 11175 delmap_call != NULL;
11176 11176 delmap_call = list_next(&rp->r_indelmap, delmap_call)) {
11177 11177 if (delmap_call->call_id == curthread) {
11178 11178 /* current caller is in the list */
11179 11179 *errp = delmap_call->error;
11180 11180 list_remove(&rp->r_indelmap, delmap_call);
11181 11181 mutex_exit(&rp->r_statelock);
11182 11182 nfs4_free_delmapcall(delmap_call);
11183 11183 return (1);
11184 11184 }
11185 11185 }
11186 11186 }
11187 11187 mutex_exit(&rp->r_statelock);
11188 11188 return (0);
11189 11189 }
11190 11190
11191 11191 /*
11192 11192 * Remove some pages from an mmap'd vnode. Just update the
11193 11193 * count of pages. If doing close-to-open, then flush and
11194 11194 * commit all of the pages associated with this file.
11195 11195 * Otherwise, start an asynchronous page flush to write out
11196 11196 * any dirty pages. This will also associate a credential
11197 11197 * with the rnode which can be used to write the pages.
11198 11198 */
11199 11199 /* ARGSUSED */
11200 11200 static void
11201 11201 nfs4_delmap_callback(struct as *as, void *arg, uint_t event)
11202 11202 {
11203 11203 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
11204 11204 rnode4_t *rp;
11205 11205 mntinfo4_t *mi;
11206 11206 nfs4_delmap_args_t *dmapp = (nfs4_delmap_args_t *)arg;
11207 11207
11208 11208 rp = VTOR4(dmapp->vp);
11209 11209 mi = VTOMI4(dmapp->vp);
11210 11210
11211 11211 atomic_add_long((ulong_t *)&rp->r_mapcnt, -btopr(dmapp->len));
11212 11212 ASSERT(rp->r_mapcnt >= 0);
11213 11213
11214 11214 /*
11215 11215 * Initiate a page flush and potential commit if there are
11216 11216 * pages, the file system was not mounted readonly, the segment
11217 11217 * was mapped shared, and the pages themselves were writeable.
11218 11218 */
11219 11219 if (nfs4_has_pages(dmapp->vp) &&
11220 11220 !(dmapp->vp->v_vfsp->vfs_flag & VFS_RDONLY) &&
11221 11221 dmapp->flags == MAP_SHARED && (dmapp->maxprot & PROT_WRITE)) {
11222 11222 mutex_enter(&rp->r_statelock);
11223 11223 rp->r_flags |= R4DIRTY;
11224 11224 mutex_exit(&rp->r_statelock);
11225 11225 e.error = nfs4_putpage_commit(dmapp->vp, dmapp->off,
11226 11226 dmapp->len, dmapp->cr);
11227 11227 if (!e.error) {
11228 11228 mutex_enter(&rp->r_statelock);
11229 11229 e.error = rp->r_error;
11230 11230 rp->r_error = 0;
11231 11231 mutex_exit(&rp->r_statelock);
11232 11232 }
11233 11233 } else
11234 11234 e.error = 0;
11235 11235
11236 11236 if ((rp->r_flags & R4DIRECTIO) || (mi->mi_flags & MI4_DIRECTIO))
11237 11237 (void) nfs4_putpage(dmapp->vp, dmapp->off, dmapp->len,
11238 11238 B_INVAL, dmapp->cr, NULL);
11239 11239
11240 11240 if (e.error) {
11241 11241 e.stat = puterrno4(e.error);
11242 11242 nfs4_queue_fact(RF_DELMAP_CB_ERR, mi, e.stat, 0,
11243 11243 OP_COMMIT, FALSE, NULL, 0, dmapp->vp);
11244 11244 dmapp->caller->error = e.error;
11245 11245 }
11246 11246
11247 11247 /* Check to see if we need to close the file */
11248 11248
11249 11249 if (dmapp->vp->v_type == VREG) {
11250 11250 nfs4close_one(dmapp->vp, NULL, dmapp->cr, 0, NULL, &e,
11251 11251 CLOSE_DELMAP, dmapp->len, dmapp->maxprot, dmapp->flags);
11252 11252
11253 11253 if (e.error != 0 || e.stat != NFS4_OK) {
11254 11254 /*
11255 11255 * Since it is possible that e.error == 0 and
11256 11256 * e.stat != NFS4_OK (and vice versa),
11257 11257 * we do the proper checking in order to get both
11258 11258 * e.error and e.stat reporting the correct info.
11259 11259 */
11260 11260 if (e.stat == NFS4_OK)
11261 11261 e.stat = puterrno4(e.error);
11262 11262 if (e.error == 0)
11263 11263 e.error = geterrno4(e.stat);
11264 11264
11265 11265 nfs4_queue_fact(RF_DELMAP_CB_ERR, mi, e.stat, 0,
11266 11266 OP_CLOSE, FALSE, NULL, 0, dmapp->vp);
11267 11267 dmapp->caller->error = e.error;
11268 11268 }
11269 11269 }
11270 11270
11271 11271 (void) as_delete_callback(as, arg);
11272 11272 kmem_free(dmapp, sizeof (nfs4_delmap_args_t));
11273 11273 }
11274 11274
11275 11275
11276 11276 static uint_t
11277 11277 fattr4_maxfilesize_to_bits(uint64_t ll)
11278 11278 {
11279 11279 uint_t l = 1;
11280 11280
11281 11281 if (ll == 0) {
11282 11282 return (0);
11283 11283 }
11284 11284
11285 11285 if (ll & 0xffffffff00000000) {
11286 11286 l += 32; ll >>= 32;
11287 11287 }
11288 11288 if (ll & 0xffff0000) {
11289 11289 l += 16; ll >>= 16;
11290 11290 }
11291 11291 if (ll & 0xff00) {
11292 11292 l += 8; ll >>= 8;
11293 11293 }
11294 11294 if (ll & 0xf0) {
11295 11295 l += 4; ll >>= 4;
11296 11296 }
11297 11297 if (ll & 0xc) {
11298 11298 l += 2; ll >>= 2;
11299 11299 }
11300 11300 if (ll & 0x2) {
11301 11301 l += 1;
11302 11302 }
11303 11303 return (l);
11304 11304 }
11305 11305
11306 11306 static int
11307 11307 nfs4_have_xattrs(vnode_t *vp, ulong_t *valp, cred_t *cr)
11308 11308 {
11309 11309 vnode_t *avp = NULL;
11310 11310 int error;
11311 11311
11312 11312 if ((error = nfs4lookup_xattr(vp, "", &avp,
11313 11313 LOOKUP_XATTR, cr)) == 0)
11314 11314 error = do_xattr_exists_check(avp, valp, cr);
11315 11315 if (avp)
11316 11316 VN_RELE(avp);
11317 11317
11318 11318 return (error);
11319 11319 }
11320 11320
11321 11321 /* ARGSUSED */
11322 11322 int
11323 11323 nfs4_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
11324 11324 caller_context_t *ct)
11325 11325 {
11326 11326 int error;
11327 11327 hrtime_t t;
11328 11328 rnode4_t *rp;
11329 11329 nfs4_ga_res_t gar;
11330 11330 nfs4_ga_ext_res_t ger;
11331 11331
11332 11332 gar.n4g_ext_res = &ger;
11333 11333
11334 11334 if (nfs_zone() != VTOMI4(vp)->mi_zone)
11335 11335 return (EIO);
11336 11336 if (cmd == _PC_PATH_MAX || cmd == _PC_SYMLINK_MAX) {
11337 11337 *valp = MAXPATHLEN;
11338 11338 return (0);
11339 11339 }
11340 11340 if (cmd == _PC_ACL_ENABLED) {
11341 11341 *valp = _ACL_ACE_ENABLED;
11342 11342 return (0);
11343 11343 }
11344 11344
11345 11345 rp = VTOR4(vp);
11346 11346 if (cmd == _PC_XATTR_EXISTS) {
11347 11347 /*
11348 11348 * The existence of the xattr directory is not sufficient
11349 11349 * for determining whether generic user attributes exists.
11350 11350 * The attribute directory could only be a transient directory
11351 11351 * used for Solaris sysattr support. Do a small readdir
11352 11352 * to verify if the only entries are sysattrs or not.
11353 11353 *
11354 11354 * pc4_xattr_valid can be only be trusted when r_xattr_dir
11355 11355 * is NULL. Once the xadir vp exists, we can create xattrs,
11356 11356 * and we don't have any way to update the "base" object's
11357 11357 * pc4_xattr_exists from the xattr or xadir. Maybe FEM
11358 11358 * could help out.
11359 11359 */
11360 11360 if (ATTRCACHE4_VALID(vp) && rp->r_pathconf.pc4_xattr_valid &&
11361 11361 rp->r_xattr_dir == NULL) {
11362 11362 return (nfs4_have_xattrs(vp, valp, cr));
11363 11363 }
11364 11364 } else { /* OLD CODE */
11365 11365 if (ATTRCACHE4_VALID(vp)) {
11366 11366 mutex_enter(&rp->r_statelock);
11367 11367 if (rp->r_pathconf.pc4_cache_valid) {
11368 11368 error = 0;
11369 11369 switch (cmd) {
11370 11370 case _PC_FILESIZEBITS:
11371 11371 *valp =
11372 11372 rp->r_pathconf.pc4_filesizebits;
11373 11373 break;
11374 11374 case _PC_LINK_MAX:
11375 11375 *valp =
11376 11376 rp->r_pathconf.pc4_link_max;
11377 11377 break;
11378 11378 case _PC_NAME_MAX:
11379 11379 *valp =
11380 11380 rp->r_pathconf.pc4_name_max;
11381 11381 break;
11382 11382 case _PC_CHOWN_RESTRICTED:
11383 11383 *valp =
11384 11384 rp->r_pathconf.pc4_chown_restricted;
11385 11385 break;
11386 11386 case _PC_NO_TRUNC:
11387 11387 *valp =
11388 11388 rp->r_pathconf.pc4_no_trunc;
11389 11389 break;
11390 11390 default:
11391 11391 error = EINVAL;
11392 11392 break;
11393 11393 }
11394 11394 mutex_exit(&rp->r_statelock);
11395 11395 #ifdef DEBUG
11396 11396 nfs4_pathconf_cache_hits++;
11397 11397 #endif
11398 11398 return (error);
11399 11399 }
11400 11400 mutex_exit(&rp->r_statelock);
11401 11401 }
11402 11402 }
11403 11403 #ifdef DEBUG
11404 11404 nfs4_pathconf_cache_misses++;
11405 11405 #endif
11406 11406
11407 11407 t = gethrtime();
11408 11408
11409 11409 error = nfs4_attr_otw(vp, TAG_PATHCONF, &gar, NFS4_PATHCONF_MASK, cr);
11410 11410
11411 11411 if (error) {
11412 11412 mutex_enter(&rp->r_statelock);
11413 11413 rp->r_pathconf.pc4_cache_valid = FALSE;
11414 11414 rp->r_pathconf.pc4_xattr_valid = FALSE;
11415 11415 mutex_exit(&rp->r_statelock);
11416 11416 return (error);
11417 11417 }
11418 11418
11419 11419 /* interpret the max filesize */
11420 11420 gar.n4g_ext_res->n4g_pc4.pc4_filesizebits =
11421 11421 fattr4_maxfilesize_to_bits(gar.n4g_ext_res->n4g_maxfilesize);
11422 11422
11423 11423 /* Store the attributes we just received */
11424 11424 nfs4_attr_cache(vp, &gar, t, cr, TRUE, NULL);
11425 11425
11426 11426 switch (cmd) {
11427 11427 case _PC_FILESIZEBITS:
11428 11428 *valp = gar.n4g_ext_res->n4g_pc4.pc4_filesizebits;
11429 11429 break;
11430 11430 case _PC_LINK_MAX:
11431 11431 *valp = gar.n4g_ext_res->n4g_pc4.pc4_link_max;
11432 11432 break;
11433 11433 case _PC_NAME_MAX:
11434 11434 *valp = gar.n4g_ext_res->n4g_pc4.pc4_name_max;
11435 11435 break;
11436 11436 case _PC_CHOWN_RESTRICTED:
11437 11437 *valp = gar.n4g_ext_res->n4g_pc4.pc4_chown_restricted;
11438 11438 break;
11439 11439 case _PC_NO_TRUNC:
11440 11440 *valp = gar.n4g_ext_res->n4g_pc4.pc4_no_trunc;
11441 11441 break;
11442 11442 case _PC_XATTR_EXISTS:
11443 11443 if (gar.n4g_ext_res->n4g_pc4.pc4_xattr_exists) {
11444 11444 if (error = nfs4_have_xattrs(vp, valp, cr))
11445 11445 return (error);
11446 11446 }
11447 11447 break;
11448 11448 default:
11449 11449 return (EINVAL);
11450 11450 }
11451 11451
11452 11452 return (0);
11453 11453 }
11454 11454
11455 11455 /*
11456 11456 * Called by async thread to do synchronous pageio. Do the i/o, wait
11457 11457 * for it to complete, and cleanup the page list when done.
11458 11458 */
11459 11459 static int
11460 11460 nfs4_sync_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
11461 11461 int flags, cred_t *cr)
11462 11462 {
11463 11463 int error;
11464 11464
11465 11465 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
11466 11466
11467 11467 error = nfs4_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
11468 11468 if (flags & B_READ)
11469 11469 pvn_read_done(pp, (error ? B_ERROR : 0) | flags);
11470 11470 else
11471 11471 pvn_write_done(pp, (error ? B_ERROR : 0) | flags);
11472 11472 return (error);
11473 11473 }
11474 11474
11475 11475 /* ARGSUSED */
11476 11476 static int
11477 11477 nfs4_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
11478 11478 int flags, cred_t *cr, caller_context_t *ct)
11479 11479 {
11480 11480 int error;
11481 11481 rnode4_t *rp;
11482 11482
11483 11483 if (!(flags & B_ASYNC) && nfs_zone() != VTOMI4(vp)->mi_zone)
11484 11484 return (EIO);
11485 11485
11486 11486 if (pp == NULL)
11487 11487 return (EINVAL);
11488 11488
11489 11489 rp = VTOR4(vp);
11490 11490 mutex_enter(&rp->r_statelock);
11491 11491 rp->r_count++;
11492 11492 mutex_exit(&rp->r_statelock);
11493 11493
11494 11494 if (flags & B_ASYNC) {
11495 11495 error = nfs4_async_pageio(vp, pp, io_off, io_len, flags, cr,
11496 11496 nfs4_sync_pageio);
11497 11497 } else
11498 11498 error = nfs4_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
11499 11499 mutex_enter(&rp->r_statelock);
11500 11500 rp->r_count--;
11501 11501 cv_broadcast(&rp->r_cv);
11502 11502 mutex_exit(&rp->r_statelock);
11503 11503 return (error);
11504 11504 }
11505 11505
11506 11506 /* ARGSUSED */
11507 11507 static void
11508 11508 nfs4_dispose(vnode_t *vp, page_t *pp, int fl, int dn, cred_t *cr,
11509 11509 caller_context_t *ct)
11510 11510 {
11511 11511 int error;
11512 11512 rnode4_t *rp;
11513 11513 page_t *plist;
11514 11514 page_t *pptr;
11515 11515 offset3 offset;
11516 11516 count3 len;
11517 11517 k_sigset_t smask;
11518 11518
11519 11519 /*
11520 11520 * We should get called with fl equal to either B_FREE or
11521 11521 * B_INVAL. Any other value is illegal.
11522 11522 *
11523 11523 * The page that we are either supposed to free or destroy
11524 11524 * should be exclusive locked and its io lock should not
11525 11525 * be held.
11526 11526 */
11527 11527 ASSERT(fl == B_FREE || fl == B_INVAL);
11528 11528 ASSERT((PAGE_EXCL(pp) && !page_iolock_assert(pp)) || panicstr);
11529 11529
11530 11530 rp = VTOR4(vp);
11531 11531
11532 11532 /*
11533 11533 * If the page doesn't need to be committed or we shouldn't
11534 11534 * even bother attempting to commit it, then just make sure
11535 11535 * that the p_fsdata byte is clear and then either free or
11536 11536 * destroy the page as appropriate.
11537 11537 */
11538 11538 if (pp->p_fsdata == C_NOCOMMIT || (rp->r_flags & R4STALE)) {
11539 11539 pp->p_fsdata = C_NOCOMMIT;
11540 11540 if (fl == B_FREE)
11541 11541 page_free(pp, dn);
11542 11542 else
11543 11543 page_destroy(pp, dn);
11544 11544 return;
11545 11545 }
11546 11546
11547 11547 /*
11548 11548 * If there is a page invalidation operation going on, then
11549 11549 * if this is one of the pages being destroyed, then just
11550 11550 * clear the p_fsdata byte and then either free or destroy
11551 11551 * the page as appropriate.
11552 11552 */
11553 11553 mutex_enter(&rp->r_statelock);
11554 11554 if ((rp->r_flags & R4TRUNCATE) && pp->p_offset >= rp->r_truncaddr) {
11555 11555 mutex_exit(&rp->r_statelock);
11556 11556 pp->p_fsdata = C_NOCOMMIT;
11557 11557 if (fl == B_FREE)
11558 11558 page_free(pp, dn);
11559 11559 else
11560 11560 page_destroy(pp, dn);
11561 11561 return;
11562 11562 }
11563 11563
11564 11564 /*
11565 11565 * If we are freeing this page and someone else is already
11566 11566 * waiting to do a commit, then just unlock the page and
11567 11567 * return. That other thread will take care of commiting
11568 11568 * this page. The page can be freed sometime after the
11569 11569 * commit has finished. Otherwise, if the page is marked
11570 11570 * as delay commit, then we may be getting called from
11571 11571 * pvn_write_done, one page at a time. This could result
11572 11572 * in one commit per page, so we end up doing lots of small
11573 11573 * commits instead of fewer larger commits. This is bad,
11574 11574 * we want do as few commits as possible.
11575 11575 */
11576 11576 if (fl == B_FREE) {
11577 11577 if (rp->r_flags & R4COMMITWAIT) {
11578 11578 page_unlock(pp);
11579 11579 mutex_exit(&rp->r_statelock);
11580 11580 return;
11581 11581 }
11582 11582 if (pp->p_fsdata == C_DELAYCOMMIT) {
11583 11583 pp->p_fsdata = C_COMMIT;
11584 11584 page_unlock(pp);
11585 11585 mutex_exit(&rp->r_statelock);
11586 11586 return;
11587 11587 }
11588 11588 }
11589 11589
11590 11590 /*
11591 11591 * Check to see if there is a signal which would prevent an
11592 11592 * attempt to commit the pages from being successful. If so,
11593 11593 * then don't bother with all of the work to gather pages and
11594 11594 * generate the unsuccessful RPC. Just return from here and
11595 11595 * let the page be committed at some later time.
11596 11596 */
11597 11597 sigintr(&smask, VTOMI4(vp)->mi_flags & MI4_INT);
11598 11598 if (ttolwp(curthread) != NULL && ISSIG(curthread, JUSTLOOKING)) {
11599 11599 sigunintr(&smask);
11600 11600 page_unlock(pp);
11601 11601 mutex_exit(&rp->r_statelock);
11602 11602 return;
11603 11603 }
11604 11604 sigunintr(&smask);
11605 11605
11606 11606 /*
11607 11607 * We are starting to need to commit pages, so let's try
11608 11608 * to commit as many as possible at once to reduce the
11609 11609 * overhead.
11610 11610 *
11611 11611 * Set the `commit inprogress' state bit. We must
11612 11612 * first wait until any current one finishes. Then
11613 11613 * we initialize the c_pages list with this page.
11614 11614 */
11615 11615 while (rp->r_flags & R4COMMIT) {
11616 11616 rp->r_flags |= R4COMMITWAIT;
11617 11617 cv_wait(&rp->r_commit.c_cv, &rp->r_statelock);
11618 11618 rp->r_flags &= ~R4COMMITWAIT;
11619 11619 }
11620 11620 rp->r_flags |= R4COMMIT;
11621 11621 mutex_exit(&rp->r_statelock);
11622 11622 ASSERT(rp->r_commit.c_pages == NULL);
11623 11623 rp->r_commit.c_pages = pp;
11624 11624 rp->r_commit.c_commbase = (offset3)pp->p_offset;
11625 11625 rp->r_commit.c_commlen = PAGESIZE;
11626 11626
11627 11627 /*
11628 11628 * Gather together all other pages which can be committed.
11629 11629 * They will all be chained off r_commit.c_pages.
11630 11630 */
11631 11631 nfs4_get_commit(vp);
11632 11632
11633 11633 /*
11634 11634 * Clear the `commit inprogress' status and disconnect
11635 11635 * the list of pages to be committed from the rnode.
11636 11636 * At this same time, we also save the starting offset
11637 11637 * and length of data to be committed on the server.
11638 11638 */
11639 11639 plist = rp->r_commit.c_pages;
11640 11640 rp->r_commit.c_pages = NULL;
11641 11641 offset = rp->r_commit.c_commbase;
11642 11642 len = rp->r_commit.c_commlen;
11643 11643 mutex_enter(&rp->r_statelock);
11644 11644 rp->r_flags &= ~R4COMMIT;
11645 11645 cv_broadcast(&rp->r_commit.c_cv);
11646 11646 mutex_exit(&rp->r_statelock);
11647 11647
11648 11648 if (curproc == proc_pageout || curproc == proc_fsflush ||
11649 11649 nfs_zone() != VTOMI4(vp)->mi_zone) {
11650 11650 nfs4_async_commit(vp, plist, offset, len,
11651 11651 cr, do_nfs4_async_commit);
11652 11652 return;
11653 11653 }
11654 11654
11655 11655 /*
11656 11656 * Actually generate the COMMIT op over the wire operation.
11657 11657 */
11658 11658 error = nfs4_commit(vp, (offset4)offset, (count4)len, cr);
11659 11659
11660 11660 /*
11661 11661 * If we got an error during the commit, just unlock all
11662 11662 * of the pages. The pages will get retransmitted to the
11663 11663 * server during a putpage operation.
11664 11664 */
11665 11665 if (error) {
11666 11666 while (plist != NULL) {
11667 11667 pptr = plist;
11668 11668 page_sub(&plist, pptr);
11669 11669 page_unlock(pptr);
11670 11670 }
11671 11671 return;
11672 11672 }
11673 11673
11674 11674 /*
11675 11675 * We've tried as hard as we can to commit the data to stable
11676 11676 * storage on the server. We just unlock the rest of the pages
11677 11677 * and clear the commit required state. They will be put
11678 11678 * onto the tail of the cachelist if they are nolonger
11679 11679 * mapped.
11680 11680 */
11681 11681 while (plist != pp) {
11682 11682 pptr = plist;
11683 11683 page_sub(&plist, pptr);
11684 11684 pptr->p_fsdata = C_NOCOMMIT;
11685 11685 page_unlock(pptr);
11686 11686 }
11687 11687
11688 11688 /*
11689 11689 * It is possible that nfs4_commit didn't return error but
11690 11690 * some other thread has modified the page we are going
11691 11691 * to free/destroy.
11692 11692 * In this case we need to rewrite the page. Do an explicit check
11693 11693 * before attempting to free/destroy the page. If modified, needs to
11694 11694 * be rewritten so unlock the page and return.
11695 11695 */
11696 11696 if (hat_ismod(pp)) {
11697 11697 pp->p_fsdata = C_NOCOMMIT;
11698 11698 page_unlock(pp);
11699 11699 return;
11700 11700 }
11701 11701
11702 11702 /*
11703 11703 * Now, as appropriate, either free or destroy the page
11704 11704 * that we were called with.
11705 11705 */
11706 11706 pp->p_fsdata = C_NOCOMMIT;
11707 11707 if (fl == B_FREE)
11708 11708 page_free(pp, dn);
11709 11709 else
11710 11710 page_destroy(pp, dn);
11711 11711 }
11712 11712
11713 11713 /*
11714 11714 * Commit requires that the current fh be the file written to.
11715 11715 * The compound op structure is:
11716 11716 * PUTFH(file), COMMIT
11717 11717 */
11718 11718 static int
11719 11719 nfs4_commit(vnode_t *vp, offset4 offset, count4 count, cred_t *cr)
11720 11720 {
11721 11721 COMPOUND4args_clnt args;
11722 11722 COMPOUND4res_clnt res;
11723 11723 COMMIT4res *cm_res;
11724 11724 nfs_argop4 argop[2];
11725 11725 nfs_resop4 *resop;
11726 11726 int doqueue;
11727 11727 mntinfo4_t *mi;
11728 11728 rnode4_t *rp;
11729 11729 cred_t *cred_otw = NULL;
11730 11730 bool_t needrecov = FALSE;
11731 11731 nfs4_recov_state_t recov_state;
11732 11732 nfs4_open_stream_t *osp = NULL;
11733 11733 bool_t first_time = TRUE; /* first time getting OTW cred */
11734 11734 bool_t last_time = FALSE; /* last time getting OTW cred */
11735 11735 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
11736 11736
11737 11737 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
11738 11738
11739 11739 rp = VTOR4(vp);
11740 11740
11741 11741 mi = VTOMI4(vp);
11742 11742 recov_state.rs_flags = 0;
11743 11743 recov_state.rs_num_retry_despite_err = 0;
11744 11744 get_commit_cred:
11745 11745 /*
11746 11746 * Releases the osp, if a valid open stream is provided.
11747 11747 * Puts a hold on the cred_otw and the new osp (if found).
11748 11748 */
11749 11749 cred_otw = nfs4_get_otw_cred_by_osp(rp, cr, &osp,
11750 11750 &first_time, &last_time);
11751 11751 args.ctag = TAG_COMMIT;
11752 11752 recov_retry:
11753 11753 /*
11754 11754 * Commit ops: putfh file; commit
11755 11755 */
11756 11756 args.array_len = 2;
11757 11757 args.array = argop;
11758 11758
11759 11759 e.error = nfs4_start_fop(VTOMI4(vp), vp, NULL, OH_COMMIT,
11760 11760 &recov_state, NULL);
11761 11761 if (e.error) {
11762 11762 crfree(cred_otw);
11763 11763 if (osp != NULL)
11764 11764 open_stream_rele(osp, rp);
11765 11765 return (e.error);
11766 11766 }
11767 11767
11768 11768 /* putfh directory */
11769 11769 argop[0].argop = OP_CPUTFH;
11770 11770 argop[0].nfs_argop4_u.opcputfh.sfh = rp->r_fh;
11771 11771
11772 11772 /* commit */
11773 11773 argop[1].argop = OP_COMMIT;
11774 11774 argop[1].nfs_argop4_u.opcommit.offset = offset;
11775 11775 argop[1].nfs_argop4_u.opcommit.count = count;
11776 11776
11777 11777 doqueue = 1;
11778 11778 rfs4call(mi, &args, &res, cred_otw, &doqueue, 0, &e);
11779 11779
11780 11780 needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
11781 11781 if (!needrecov && e.error) {
11782 11782 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_COMMIT, &recov_state,
11783 11783 needrecov);
11784 11784 crfree(cred_otw);
11785 11785 if (e.error == EACCES && last_time == FALSE)
11786 11786 goto get_commit_cred;
11787 11787 if (osp != NULL)
11788 11788 open_stream_rele(osp, rp);
11789 11789 return (e.error);
11790 11790 }
11791 11791
11792 11792 if (needrecov) {
11793 11793 if (nfs4_start_recovery(&e, VTOMI4(vp), vp, NULL, NULL,
11794 11794 NULL, OP_COMMIT, NULL, NULL, NULL) == FALSE) {
11795 11795 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_COMMIT,
11796 11796 &recov_state, needrecov);
11797 11797 if (!e.error)
11798 11798 (void) xdr_free(xdr_COMPOUND4res_clnt,
11799 11799 (caddr_t)&res);
11800 11800 goto recov_retry;
11801 11801 }
11802 11802 if (e.error) {
11803 11803 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_COMMIT,
11804 11804 &recov_state, needrecov);
11805 11805 crfree(cred_otw);
11806 11806 if (osp != NULL)
11807 11807 open_stream_rele(osp, rp);
11808 11808 return (e.error);
11809 11809 }
11810 11810 /* fall through for res.status case */
11811 11811 }
11812 11812
11813 11813 if (res.status) {
11814 11814 e.error = geterrno4(res.status);
11815 11815 if (e.error == EACCES && last_time == FALSE) {
11816 11816 crfree(cred_otw);
11817 11817 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_COMMIT,
11818 11818 &recov_state, needrecov);
11819 11819 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
11820 11820 goto get_commit_cred;
11821 11821 }
11822 11822 /*
11823 11823 * Can't do a nfs4_purge_stale_fh here because this
11824 11824 * can cause a deadlock. nfs4_commit can
11825 11825 * be called from nfs4_dispose which can be called
11826 11826 * indirectly via pvn_vplist_dirty. nfs4_purge_stale_fh
11827 11827 * can call back to pvn_vplist_dirty.
11828 11828 */
11829 11829 if (e.error == ESTALE) {
11830 11830 mutex_enter(&rp->r_statelock);
11831 11831 rp->r_flags |= R4STALE;
11832 11832 if (!rp->r_error)
11833 11833 rp->r_error = e.error;
11834 11834 mutex_exit(&rp->r_statelock);
11835 11835 PURGE_ATTRCACHE4(vp);
11836 11836 } else {
11837 11837 mutex_enter(&rp->r_statelock);
11838 11838 if (!rp->r_error)
11839 11839 rp->r_error = e.error;
11840 11840 mutex_exit(&rp->r_statelock);
11841 11841 }
11842 11842 } else {
11843 11843 ASSERT(rp->r_flags & R4HAVEVERF);
11844 11844 resop = &res.array[1]; /* commit res */
11845 11845 cm_res = &resop->nfs_resop4_u.opcommit;
11846 11846 mutex_enter(&rp->r_statelock);
11847 11847 if (cm_res->writeverf == rp->r_writeverf) {
11848 11848 mutex_exit(&rp->r_statelock);
11849 11849 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
11850 11850 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_COMMIT,
11851 11851 &recov_state, needrecov);
11852 11852 crfree(cred_otw);
11853 11853 if (osp != NULL)
11854 11854 open_stream_rele(osp, rp);
11855 11855 return (0);
11856 11856 }
11857 11857 nfs4_set_mod(vp);
11858 11858 rp->r_writeverf = cm_res->writeverf;
11859 11859 mutex_exit(&rp->r_statelock);
11860 11860 e.error = NFS_VERF_MISMATCH;
11861 11861 }
11862 11862
11863 11863 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
11864 11864 nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_COMMIT, &recov_state, needrecov);
11865 11865 crfree(cred_otw);
11866 11866 if (osp != NULL)
11867 11867 open_stream_rele(osp, rp);
11868 11868
11869 11869 return (e.error);
11870 11870 }
11871 11871
11872 11872 static void
11873 11873 nfs4_set_mod(vnode_t *vp)
11874 11874 {
11875 11875 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
11876 11876
11877 11877 /* make sure we're looking at the master vnode, not a shadow */
11878 11878 pvn_vplist_setdirty(RTOV4(VTOR4(vp)), nfs_setmod_check);
11879 11879 }
11880 11880
11881 11881 /*
11882 11882 * This function is used to gather a page list of the pages which
11883 11883 * can be committed on the server.
11884 11884 *
11885 11885 * The calling thread must have set R4COMMIT. This bit is used to
11886 11886 * serialize access to the commit structure in the rnode. As long
11887 11887 * as the thread has set R4COMMIT, then it can manipulate the commit
11888 11888 * structure without requiring any other locks.
11889 11889 *
11890 11890 * When this function is called from nfs4_dispose() the page passed
11891 11891 * into nfs4_dispose() will be SE_EXCL locked, and so this function
11892 11892 * will skip it. This is not a problem since we initially add the
11893 11893 * page to the r_commit page list.
11894 11894 *
11895 11895 */
11896 11896 static void
11897 11897 nfs4_get_commit(vnode_t *vp)
11898 11898 {
11899 11899 rnode4_t *rp;
11900 11900 page_t *pp;
11901 11901 kmutex_t *vphm;
11902 11902
11903 11903 rp = VTOR4(vp);
11904 11904
11905 11905 ASSERT(rp->r_flags & R4COMMIT);
11906 11906
11907 11907 /* make sure we're looking at the master vnode, not a shadow */
11908 11908
11909 11909 if (IS_SHADOW(vp, rp))
11910 11910 vp = RTOV4(rp);
11911 11911
11912 11912 vphm = page_vnode_mutex(vp);
11913 11913 mutex_enter(vphm);
11914 11914
11915 11915 /*
11916 11916 * If there are no pages associated with this vnode, then
11917 11917 * just return.
11918 11918 */
11919 11919 if ((pp = vp->v_pages) == NULL) {
11920 11920 mutex_exit(vphm);
11921 11921 return;
11922 11922 }
11923 11923
11924 11924 /*
11925 11925 * Step through all of the pages associated with this vnode
11926 11926 * looking for pages which need to be committed.
11927 11927 */
11928 11928 do {
11929 11929 /* Skip marker pages. */
11930 11930 if (pp->p_hash == PVN_VPLIST_HASH_TAG)
11931 11931 continue;
11932 11932
11933 11933 /*
11934 11934 * First short-cut everything (without the page_lock)
11935 11935 * and see if this page does not need to be committed
11936 11936 * or is modified if so then we'll just skip it.
11937 11937 */
11938 11938 if (pp->p_fsdata == C_NOCOMMIT || hat_ismod(pp))
11939 11939 continue;
11940 11940
11941 11941 /*
11942 11942 * Attempt to lock the page. If we can't, then
11943 11943 * someone else is messing with it or we have been
11944 11944 * called from nfs4_dispose and this is the page that
11945 11945 * nfs4_dispose was called with.. anyway just skip it.
11946 11946 */
11947 11947 if (!page_trylock(pp, SE_EXCL))
11948 11948 continue;
11949 11949
11950 11950 /*
11951 11951 * Lets check again now that we have the page lock.
11952 11952 */
11953 11953 if (pp->p_fsdata == C_NOCOMMIT || hat_ismod(pp)) {
11954 11954 page_unlock(pp);
11955 11955 continue;
11956 11956 }
11957 11957
11958 11958 /* this had better not be a free page */
11959 11959 ASSERT(PP_ISFREE(pp) == 0);
11960 11960
11961 11961 /*
11962 11962 * The page needs to be committed and we locked it.
11963 11963 * Update the base and length parameters and add it
11964 11964 * to r_pages.
11965 11965 */
11966 11966 if (rp->r_commit.c_pages == NULL) {
11967 11967 rp->r_commit.c_commbase = (offset3)pp->p_offset;
11968 11968 rp->r_commit.c_commlen = PAGESIZE;
11969 11969 } else if (pp->p_offset < rp->r_commit.c_commbase) {
11970 11970 rp->r_commit.c_commlen = rp->r_commit.c_commbase -
11971 11971 (offset3)pp->p_offset + rp->r_commit.c_commlen;
11972 11972 rp->r_commit.c_commbase = (offset3)pp->p_offset;
11973 11973 } else if ((rp->r_commit.c_commbase + rp->r_commit.c_commlen)
11974 11974 <= pp->p_offset) {
11975 11975 rp->r_commit.c_commlen = (offset3)pp->p_offset -
11976 11976 rp->r_commit.c_commbase + PAGESIZE;
11977 11977 }
11978 11978 page_add(&rp->r_commit.c_pages, pp);
11979 11979 } while ((pp = pp->p_vpnext) != vp->v_pages);
11980 11980
11981 11981 mutex_exit(vphm);
11982 11982 }
11983 11983
11984 11984 /*
11985 11985 * This routine is used to gather together a page list of the pages
11986 11986 * which are to be committed on the server. This routine must not
11987 11987 * be called if the calling thread holds any locked pages.
11988 11988 *
11989 11989 * The calling thread must have set R4COMMIT. This bit is used to
11990 11990 * serialize access to the commit structure in the rnode. As long
11991 11991 * as the thread has set R4COMMIT, then it can manipulate the commit
11992 11992 * structure without requiring any other locks.
11993 11993 */
11994 11994 static void
11995 11995 nfs4_get_commit_range(vnode_t *vp, u_offset_t soff, size_t len)
11996 11996 {
11997 11997
11998 11998 rnode4_t *rp;
11999 11999 page_t *pp;
12000 12000 u_offset_t end;
12001 12001 u_offset_t off;
12002 12002 ASSERT(len != 0);
12003 12003 rp = VTOR4(vp);
12004 12004 ASSERT(rp->r_flags & R4COMMIT);
12005 12005
12006 12006 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
12007 12007
12008 12008 /* make sure we're looking at the master vnode, not a shadow */
12009 12009
12010 12010 if (IS_SHADOW(vp, rp))
12011 12011 vp = RTOV4(rp);
12012 12012
12013 12013 /*
12014 12014 * If there are no pages associated with this vnode, then
12015 12015 * just return.
12016 12016 */
12017 12017 if ((pp = vp->v_pages) == NULL)
12018 12018 return;
12019 12019 /*
12020 12020 * Calculate the ending offset.
12021 12021 */
12022 12022 end = soff + len;
12023 12023 for (off = soff; off < end; off += PAGESIZE) {
12024 12024 /*
12025 12025 * Lookup each page by vp, offset.
12026 12026 */
12027 12027 if ((pp = page_lookup_nowait(vp, off, SE_EXCL)) == NULL)
12028 12028 continue;
12029 12029 /*
12030 12030 * If this page does not need to be committed or is
12031 12031 * modified, then just skip it.
12032 12032 */
12033 12033 if (pp->p_fsdata == C_NOCOMMIT || hat_ismod(pp)) {
12034 12034 page_unlock(pp);
12035 12035 continue;
12036 12036 }
12037 12037
12038 12038 ASSERT(PP_ISFREE(pp) == 0);
12039 12039 /*
12040 12040 * The page needs to be committed and we locked it.
12041 12041 * Update the base and length parameters and add it
12042 12042 * to r_pages.
12043 12043 */
12044 12044 if (rp->r_commit.c_pages == NULL) {
12045 12045 rp->r_commit.c_commbase = (offset3)pp->p_offset;
12046 12046 rp->r_commit.c_commlen = PAGESIZE;
12047 12047 } else {
12048 12048 rp->r_commit.c_commlen = (offset3)pp->p_offset -
12049 12049 rp->r_commit.c_commbase + PAGESIZE;
12050 12050 }
12051 12051 page_add(&rp->r_commit.c_pages, pp);
12052 12052 }
12053 12053 }
12054 12054
12055 12055 /*
12056 12056 * Called from nfs4_close(), nfs4_fsync() and nfs4_delmap().
12057 12057 * Flushes and commits data to the server.
12058 12058 */
12059 12059 static int
12060 12060 nfs4_putpage_commit(vnode_t *vp, offset_t poff, size_t plen, cred_t *cr)
12061 12061 {
12062 12062 int error;
12063 12063 verifier4 write_verf;
12064 12064 rnode4_t *rp = VTOR4(vp);
12065 12065
12066 12066 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
12067 12067
12068 12068 /*
12069 12069 * Flush the data portion of the file and then commit any
12070 12070 * portions which need to be committed. This may need to
12071 12071 * be done twice if the server has changed state since
12072 12072 * data was last written. The data will need to be
12073 12073 * rewritten to the server and then a new commit done.
12074 12074 *
12075 12075 * In fact, this may need to be done several times if the
12076 12076 * server is having problems and crashing while we are
12077 12077 * attempting to do this.
12078 12078 */
12079 12079
12080 12080 top:
12081 12081 /*
12082 12082 * Do a flush based on the poff and plen arguments. This
12083 12083 * will synchronously write out any modified pages in the
12084 12084 * range specified by (poff, plen). This starts all of the
12085 12085 * i/o operations which will be waited for in the next
12086 12086 * call to nfs4_putpage
12087 12087 */
12088 12088
12089 12089 mutex_enter(&rp->r_statelock);
12090 12090 write_verf = rp->r_writeverf;
12091 12091 mutex_exit(&rp->r_statelock);
12092 12092
12093 12093 error = nfs4_putpage(vp, poff, plen, B_ASYNC, cr, NULL);
12094 12094 if (error == EAGAIN)
12095 12095 error = 0;
12096 12096
12097 12097 /*
12098 12098 * Do a flush based on the poff and plen arguments. This
12099 12099 * will synchronously write out any modified pages in the
12100 12100 * range specified by (poff, plen) and wait until all of
12101 12101 * the asynchronous i/o's in that range are done as well.
12102 12102 */
12103 12103 if (!error)
12104 12104 error = nfs4_putpage(vp, poff, plen, 0, cr, NULL);
12105 12105
12106 12106 if (error)
12107 12107 return (error);
12108 12108
12109 12109 mutex_enter(&rp->r_statelock);
12110 12110 if (rp->r_writeverf != write_verf) {
12111 12111 mutex_exit(&rp->r_statelock);
12112 12112 goto top;
12113 12113 }
12114 12114 mutex_exit(&rp->r_statelock);
12115 12115
12116 12116 /*
12117 12117 * Now commit any pages which might need to be committed.
12118 12118 * If the error, NFS_VERF_MISMATCH, is returned, then
12119 12119 * start over with the flush operation.
12120 12120 */
12121 12121 error = nfs4_commit_vp(vp, poff, plen, cr, NFS4_WRITE_WAIT);
12122 12122
12123 12123 if (error == NFS_VERF_MISMATCH)
12124 12124 goto top;
12125 12125
12126 12126 return (error);
12127 12127 }
12128 12128
12129 12129 /*
12130 12130 * nfs4_commit_vp() will wait for other pending commits and
12131 12131 * will either commit the whole file or a range, plen dictates
12132 12132 * if we commit whole file. a value of zero indicates the whole
12133 12133 * file. Called from nfs4_putpage_commit() or nfs4_sync_putapage()
12134 12134 */
12135 12135 static int
12136 12136 nfs4_commit_vp(vnode_t *vp, u_offset_t poff, size_t plen,
12137 12137 cred_t *cr, int wait_on_writes)
12138 12138 {
12139 12139 rnode4_t *rp;
12140 12140 page_t *plist;
12141 12141 offset3 offset;
12142 12142 count3 len;
12143 12143
12144 12144 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
12145 12145
12146 12146 rp = VTOR4(vp);
12147 12147
12148 12148 /*
12149 12149 * before we gather commitable pages make
12150 12150 * sure there are no outstanding async writes
12151 12151 */
12152 12152 if (rp->r_count && wait_on_writes == NFS4_WRITE_WAIT) {
12153 12153 mutex_enter(&rp->r_statelock);
12154 12154 while (rp->r_count > 0) {
12155 12155 cv_wait(&rp->r_cv, &rp->r_statelock);
12156 12156 }
12157 12157 mutex_exit(&rp->r_statelock);
12158 12158 }
12159 12159
12160 12160 /*
12161 12161 * Set the `commit inprogress' state bit. We must
12162 12162 * first wait until any current one finishes.
12163 12163 */
12164 12164 mutex_enter(&rp->r_statelock);
12165 12165 while (rp->r_flags & R4COMMIT) {
12166 12166 rp->r_flags |= R4COMMITWAIT;
12167 12167 cv_wait(&rp->r_commit.c_cv, &rp->r_statelock);
12168 12168 rp->r_flags &= ~R4COMMITWAIT;
12169 12169 }
12170 12170 rp->r_flags |= R4COMMIT;
12171 12171 mutex_exit(&rp->r_statelock);
12172 12172
12173 12173 /*
12174 12174 * Gather all of the pages which need to be
12175 12175 * committed.
12176 12176 */
12177 12177 if (plen == 0)
12178 12178 nfs4_get_commit(vp);
12179 12179 else
12180 12180 nfs4_get_commit_range(vp, poff, plen);
12181 12181
12182 12182 /*
12183 12183 * Clear the `commit inprogress' bit and disconnect the
12184 12184 * page list which was gathered by nfs4_get_commit.
12185 12185 */
12186 12186 plist = rp->r_commit.c_pages;
12187 12187 rp->r_commit.c_pages = NULL;
12188 12188 offset = rp->r_commit.c_commbase;
12189 12189 len = rp->r_commit.c_commlen;
12190 12190 mutex_enter(&rp->r_statelock);
12191 12191 rp->r_flags &= ~R4COMMIT;
12192 12192 cv_broadcast(&rp->r_commit.c_cv);
12193 12193 mutex_exit(&rp->r_statelock);
12194 12194
12195 12195 /*
12196 12196 * If any pages need to be committed, commit them and
12197 12197 * then unlock them so that they can be freed some
12198 12198 * time later.
12199 12199 */
12200 12200 if (plist == NULL)
12201 12201 return (0);
12202 12202
12203 12203 /*
12204 12204 * No error occurred during the flush portion
12205 12205 * of this operation, so now attempt to commit
12206 12206 * the data to stable storage on the server.
12207 12207 *
12208 12208 * This will unlock all of the pages on the list.
12209 12209 */
12210 12210 return (nfs4_sync_commit(vp, plist, offset, len, cr));
12211 12211 }
12212 12212
12213 12213 static int
12214 12214 nfs4_sync_commit(vnode_t *vp, page_t *plist, offset3 offset, count3 count,
12215 12215 cred_t *cr)
12216 12216 {
12217 12217 int error;
12218 12218 page_t *pp;
12219 12219
12220 12220 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
12221 12221
12222 12222 error = nfs4_commit(vp, (offset4)offset, (count3)count, cr);
12223 12223
12224 12224 /*
12225 12225 * If we got an error, then just unlock all of the pages
12226 12226 * on the list.
12227 12227 */
12228 12228 if (error) {
12229 12229 while (plist != NULL) {
12230 12230 pp = plist;
12231 12231 page_sub(&plist, pp);
12232 12232 page_unlock(pp);
12233 12233 }
12234 12234 return (error);
12235 12235 }
12236 12236 /*
12237 12237 * We've tried as hard as we can to commit the data to stable
12238 12238 * storage on the server. We just unlock the pages and clear
12239 12239 * the commit required state. They will get freed later.
12240 12240 */
12241 12241 while (plist != NULL) {
12242 12242 pp = plist;
12243 12243 page_sub(&plist, pp);
12244 12244 pp->p_fsdata = C_NOCOMMIT;
12245 12245 page_unlock(pp);
12246 12246 }
12247 12247
12248 12248 return (error);
12249 12249 }
12250 12250
12251 12251 static void
12252 12252 do_nfs4_async_commit(vnode_t *vp, page_t *plist, offset3 offset, count3 count,
12253 12253 cred_t *cr)
12254 12254 {
12255 12255
12256 12256 (void) nfs4_sync_commit(vp, plist, offset, count, cr);
12257 12257 }
12258 12258
12259 12259 /*ARGSUSED*/
12260 12260 static int
12261 12261 nfs4_setsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr,
12262 12262 caller_context_t *ct)
12263 12263 {
12264 12264 int error = 0;
12265 12265 mntinfo4_t *mi;
12266 12266 vattr_t va;
12267 12267 vsecattr_t nfsace4_vsap;
12268 12268
12269 12269 mi = VTOMI4(vp);
12270 12270 if (nfs_zone() != mi->mi_zone)
12271 12271 return (EIO);
12272 12272 if (mi->mi_flags & MI4_ACL) {
12273 12273 /* if we have a delegation, return it */
12274 12274 if (VTOR4(vp)->r_deleg_type != OPEN_DELEGATE_NONE)
12275 12275 (void) nfs4delegreturn(VTOR4(vp),
12276 12276 NFS4_DR_REOPEN|NFS4_DR_PUSH);
12277 12277
12278 12278 error = nfs4_is_acl_mask_valid(vsecattr->vsa_mask,
12279 12279 NFS4_ACL_SET);
12280 12280 if (error) /* EINVAL */
12281 12281 return (error);
12282 12282
12283 12283 if (vsecattr->vsa_mask & (VSA_ACL | VSA_DFACL)) {
12284 12284 /*
12285 12285 * These are aclent_t type entries.
12286 12286 */
12287 12287 error = vs_aent_to_ace4(vsecattr, &nfsace4_vsap,
12288 12288 vp->v_type == VDIR, FALSE);
12289 12289 if (error)
12290 12290 return (error);
12291 12291 } else {
12292 12292 /*
12293 12293 * These are ace_t type entries.
12294 12294 */
12295 12295 error = vs_acet_to_ace4(vsecattr, &nfsace4_vsap,
12296 12296 FALSE);
12297 12297 if (error)
12298 12298 return (error);
12299 12299 }
12300 12300 bzero(&va, sizeof (va));
12301 12301 error = nfs4setattr(vp, &va, flag, cr, &nfsace4_vsap);
12302 12302 vs_ace4_destroy(&nfsace4_vsap);
12303 12303 return (error);
12304 12304 }
12305 12305 return (ENOSYS);
12306 12306 }
12307 12307
12308 12308 /* ARGSUSED */
12309 12309 int
12310 12310 nfs4_getsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr,
12311 12311 caller_context_t *ct)
12312 12312 {
12313 12313 int error;
12314 12314 mntinfo4_t *mi;
12315 12315 nfs4_ga_res_t gar;
12316 12316 rnode4_t *rp = VTOR4(vp);
12317 12317
12318 12318 mi = VTOMI4(vp);
12319 12319 if (nfs_zone() != mi->mi_zone)
12320 12320 return (EIO);
12321 12321
12322 12322 bzero(&gar, sizeof (gar));
12323 12323 gar.n4g_vsa.vsa_mask = vsecattr->vsa_mask;
12324 12324
12325 12325 /*
12326 12326 * vsecattr->vsa_mask holds the original acl request mask.
12327 12327 * This is needed when determining what to return.
12328 12328 * (See: nfs4_create_getsecattr_return())
12329 12329 */
12330 12330 error = nfs4_is_acl_mask_valid(vsecattr->vsa_mask, NFS4_ACL_GET);
12331 12331 if (error) /* EINVAL */
12332 12332 return (error);
12333 12333
12334 12334 /*
12335 12335 * If this is a referral stub, don't try to go OTW for an ACL
12336 12336 */
12337 12337 if (RP_ISSTUB_REFERRAL(VTOR4(vp)))
12338 12338 return (fs_fab_acl(vp, vsecattr, flag, cr, ct));
12339 12339
12340 12340 if (mi->mi_flags & MI4_ACL) {
12341 12341 /*
12342 12342 * Check if the data is cached and the cache is valid. If it
12343 12343 * is we don't go over the wire.
12344 12344 */
12345 12345 if (rp->r_secattr != NULL && ATTRCACHE4_VALID(vp)) {
12346 12346 mutex_enter(&rp->r_statelock);
12347 12347 if (rp->r_secattr != NULL) {
12348 12348 error = nfs4_create_getsecattr_return(
12349 12349 rp->r_secattr, vsecattr, rp->r_attr.va_uid,
12350 12350 rp->r_attr.va_gid,
12351 12351 vp->v_type == VDIR);
12352 12352 if (!error) { /* error == 0 - Success! */
12353 12353 mutex_exit(&rp->r_statelock);
12354 12354 return (error);
12355 12355 }
12356 12356 }
12357 12357 mutex_exit(&rp->r_statelock);
12358 12358 }
12359 12359
12360 12360 /*
12361 12361 * The getattr otw call will always get both the acl, in
12362 12362 * the form of a list of nfsace4's, and the number of acl
12363 12363 * entries; independent of the value of gar.n4g_vsa.vsa_mask.
12364 12364 */
12365 12365 gar.n4g_va.va_mask = AT_ALL;
12366 12366 error = nfs4_getattr_otw(vp, &gar, cr, 1);
12367 12367 if (error) {
12368 12368 vs_ace4_destroy(&gar.n4g_vsa);
12369 12369 if (error == ENOTSUP || error == EOPNOTSUPP)
12370 12370 error = fs_fab_acl(vp, vsecattr, flag, cr, ct);
12371 12371 return (error);
12372 12372 }
12373 12373
12374 12374 if (!(gar.n4g_resbmap & FATTR4_ACL_MASK)) {
12375 12375 /*
12376 12376 * No error was returned, but according to the response
12377 12377 * bitmap, neither was an acl.
12378 12378 */
12379 12379 vs_ace4_destroy(&gar.n4g_vsa);
12380 12380 error = fs_fab_acl(vp, vsecattr, flag, cr, ct);
12381 12381 return (error);
12382 12382 }
12383 12383
12384 12384 /*
12385 12385 * Update the cache with the ACL.
12386 12386 */
12387 12387 nfs4_acl_fill_cache(rp, &gar.n4g_vsa);
12388 12388
12389 12389 error = nfs4_create_getsecattr_return(&gar.n4g_vsa,
12390 12390 vsecattr, gar.n4g_va.va_uid, gar.n4g_va.va_gid,
12391 12391 vp->v_type == VDIR);
12392 12392 vs_ace4_destroy(&gar.n4g_vsa);
12393 12393 if ((error) && (vsecattr->vsa_mask &
12394 12394 (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT)) &&
12395 12395 (error != EACCES)) {
12396 12396 error = fs_fab_acl(vp, vsecattr, flag, cr, ct);
12397 12397 }
12398 12398 return (error);
12399 12399 }
12400 12400 error = fs_fab_acl(vp, vsecattr, flag, cr, ct);
12401 12401 return (error);
12402 12402 }
12403 12403
12404 12404 /*
12405 12405 * The function returns:
12406 12406 * - 0 (zero) if the passed in "acl_mask" is a valid request.
12407 12407 * - EINVAL if the passed in "acl_mask" is an invalid request.
12408 12408 *
12409 12409 * In the case of getting an acl (op == NFS4_ACL_GET) the mask is invalid if:
12410 12410 * - We have a mixture of ACE and ACL requests (e.g. VSA_ACL | VSA_ACE)
12411 12411 *
12412 12412 * In the case of setting an acl (op == NFS4_ACL_SET) the mask is invalid if:
12413 12413 * - We have a mixture of ACE and ACL requests (e.g. VSA_ACL | VSA_ACE)
12414 12414 * - We have a count field set without the corresponding acl field set. (e.g. -
12415 12415 * VSA_ACECNT is set, but VSA_ACE is not)
12416 12416 */
12417 12417 static int
12418 12418 nfs4_is_acl_mask_valid(uint_t acl_mask, nfs4_acl_op_t op)
12419 12419 {
12420 12420 /* Shortcut the masks that are always valid. */
12421 12421 if (acl_mask == (VSA_ACE | VSA_ACECNT))
12422 12422 return (0);
12423 12423 if (acl_mask == (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT))
12424 12424 return (0);
12425 12425
12426 12426 if (acl_mask & (VSA_ACE | VSA_ACECNT)) {
12427 12427 /*
12428 12428 * We can't have any VSA_ACL type stuff in the mask now.
12429 12429 */
12430 12430 if (acl_mask & (VSA_ACL | VSA_ACLCNT | VSA_DFACL |
12431 12431 VSA_DFACLCNT))
12432 12432 return (EINVAL);
12433 12433
12434 12434 if (op == NFS4_ACL_SET) {
12435 12435 if ((acl_mask & VSA_ACECNT) && !(acl_mask & VSA_ACE))
12436 12436 return (EINVAL);
12437 12437 }
12438 12438 }
12439 12439
12440 12440 if (acl_mask & (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT)) {
12441 12441 /*
12442 12442 * We can't have any VSA_ACE type stuff in the mask now.
12443 12443 */
12444 12444 if (acl_mask & (VSA_ACE | VSA_ACECNT))
12445 12445 return (EINVAL);
12446 12446
12447 12447 if (op == NFS4_ACL_SET) {
12448 12448 if ((acl_mask & VSA_ACLCNT) && !(acl_mask & VSA_ACL))
12449 12449 return (EINVAL);
12450 12450
12451 12451 if ((acl_mask & VSA_DFACLCNT) &&
12452 12452 !(acl_mask & VSA_DFACL))
12453 12453 return (EINVAL);
12454 12454 }
12455 12455 }
12456 12456 return (0);
12457 12457 }
12458 12458
12459 12459 /*
12460 12460 * The theory behind creating the correct getsecattr return is simply this:
12461 12461 * "Don't return anything that the caller is not expecting to have to free."
12462 12462 */
12463 12463 static int
12464 12464 nfs4_create_getsecattr_return(vsecattr_t *filled_vsap, vsecattr_t *vsap,
12465 12465 uid_t uid, gid_t gid, int isdir)
12466 12466 {
12467 12467 int error = 0;
12468 12468 /* Save the mask since the translators modify it. */
12469 12469 uint_t orig_mask = vsap->vsa_mask;
12470 12470
12471 12471 if (orig_mask & (VSA_ACE | VSA_ACECNT)) {
12472 12472 error = vs_ace4_to_acet(filled_vsap, vsap, uid, gid, FALSE);
12473 12473
12474 12474 if (error)
12475 12475 return (error);
12476 12476
12477 12477 /*
12478 12478 * If the caller only asked for the ace count (VSA_ACECNT)
12479 12479 * don't give them the full acl (VSA_ACE), free it.
12480 12480 */
12481 12481 if (!orig_mask & VSA_ACE) {
12482 12482 if (vsap->vsa_aclentp != NULL) {
12483 12483 kmem_free(vsap->vsa_aclentp,
12484 12484 vsap->vsa_aclcnt * sizeof (ace_t));
12485 12485 vsap->vsa_aclentp = NULL;
12486 12486 }
12487 12487 }
12488 12488 vsap->vsa_mask = orig_mask;
12489 12489
12490 12490 } else if (orig_mask & (VSA_ACL | VSA_ACLCNT | VSA_DFACL |
12491 12491 VSA_DFACLCNT)) {
12492 12492 error = vs_ace4_to_aent(filled_vsap, vsap, uid, gid,
12493 12493 isdir, FALSE);
12494 12494
12495 12495 if (error)
12496 12496 return (error);
12497 12497
12498 12498 /*
12499 12499 * If the caller only asked for the acl count (VSA_ACLCNT)
12500 12500 * and/or the default acl count (VSA_DFACLCNT) don't give them
12501 12501 * the acl (VSA_ACL) or default acl (VSA_DFACL), free it.
12502 12502 */
12503 12503 if (!orig_mask & VSA_ACL) {
12504 12504 if (vsap->vsa_aclentp != NULL) {
12505 12505 kmem_free(vsap->vsa_aclentp,
12506 12506 vsap->vsa_aclcnt * sizeof (aclent_t));
12507 12507 vsap->vsa_aclentp = NULL;
12508 12508 }
12509 12509 }
12510 12510
12511 12511 if (!orig_mask & VSA_DFACL) {
12512 12512 if (vsap->vsa_dfaclentp != NULL) {
12513 12513 kmem_free(vsap->vsa_dfaclentp,
12514 12514 vsap->vsa_dfaclcnt * sizeof (aclent_t));
12515 12515 vsap->vsa_dfaclentp = NULL;
12516 12516 }
12517 12517 }
12518 12518 vsap->vsa_mask = orig_mask;
12519 12519 }
12520 12520 return (0);
12521 12521 }
12522 12522
12523 12523 /* ARGSUSED */
12524 12524 int
12525 12525 nfs4_shrlock(vnode_t *vp, int cmd, struct shrlock *shr, int flag, cred_t *cr,
12526 12526 caller_context_t *ct)
12527 12527 {
12528 12528 int error;
12529 12529
12530 12530 if (nfs_zone() != VTOMI4(vp)->mi_zone)
12531 12531 return (EIO);
12532 12532 /*
12533 12533 * check for valid cmd parameter
12534 12534 */
12535 12535 if (cmd != F_SHARE && cmd != F_UNSHARE && cmd != F_HASREMOTELOCKS)
12536 12536 return (EINVAL);
12537 12537
12538 12538 /*
12539 12539 * Check access permissions
12540 12540 */
12541 12541 if ((cmd & F_SHARE) &&
12542 12542 (((shr->s_access & F_RDACC) && (flag & FREAD) == 0) ||
12543 12543 (shr->s_access == F_WRACC && (flag & FWRITE) == 0)))
12544 12544 return (EBADF);
12545 12545
12546 12546 /*
12547 12547 * If the filesystem is mounted using local locking, pass the
12548 12548 * request off to the local share code.
12549 12549 */
12550 12550 if (VTOMI4(vp)->mi_flags & MI4_LLOCK)
12551 12551 return (fs_shrlock(vp, cmd, shr, flag, cr, ct));
12552 12552
12553 12553 switch (cmd) {
12554 12554 case F_SHARE:
12555 12555 case F_UNSHARE:
12556 12556 /*
12557 12557 * This will be properly implemented later,
12558 12558 * see RFE: 4823948 .
12559 12559 */
12560 12560 error = EAGAIN;
12561 12561 break;
12562 12562
12563 12563 case F_HASREMOTELOCKS:
12564 12564 /*
12565 12565 * NFS client can't store remote locks itself
12566 12566 */
12567 12567 shr->s_access = 0;
12568 12568 error = 0;
12569 12569 break;
12570 12570
12571 12571 default:
12572 12572 error = EINVAL;
12573 12573 break;
12574 12574 }
12575 12575
12576 12576 return (error);
12577 12577 }
12578 12578
12579 12579 /*
12580 12580 * Common code called by directory ops to update the attrcache
12581 12581 */
12582 12582 static int
12583 12583 nfs4_update_attrcache(nfsstat4 status, nfs4_ga_res_t *garp,
12584 12584 hrtime_t t, vnode_t *vp, cred_t *cr)
12585 12585 {
12586 12586 int error = 0;
12587 12587
12588 12588 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
12589 12589
12590 12590 if (status != NFS4_OK) {
12591 12591 /* getattr not done or failed */
12592 12592 PURGE_ATTRCACHE4(vp);
12593 12593 return (error);
12594 12594 }
12595 12595
12596 12596 if (garp) {
12597 12597 nfs4_attr_cache(vp, garp, t, cr, FALSE, NULL);
12598 12598 } else {
12599 12599 PURGE_ATTRCACHE4(vp);
12600 12600 }
12601 12601 return (error);
12602 12602 }
12603 12603
12604 12604 /*
12605 12605 * Update directory caches for directory modification ops (link, rename, etc.)
12606 12606 * When dinfo is NULL, manage dircaches in the old way.
12607 12607 */
12608 12608 static void
12609 12609 nfs4_update_dircaches(change_info4 *cinfo, vnode_t *dvp, vnode_t *vp, char *nm,
12610 12610 dirattr_info_t *dinfo)
12611 12611 {
12612 12612 rnode4_t *drp = VTOR4(dvp);
12613 12613
12614 12614 ASSERT(nfs_zone() == VTOMI4(dvp)->mi_zone);
12615 12615
12616 12616 /* Purge rddir cache for dir since it changed */
12617 12617 if (drp->r_dir != NULL)
12618 12618 nfs4_purge_rddir_cache(dvp);
12619 12619
12620 12620 /*
12621 12621 * If caller provided dinfo, then use it to manage dir caches.
12622 12622 */
12623 12623 if (dinfo != NULL) {
12624 12624 if (vp != NULL) {
12625 12625 mutex_enter(&VTOR4(vp)->r_statev4_lock);
12626 12626 if (!VTOR4(vp)->created_v4) {
12627 12627 mutex_exit(&VTOR4(vp)->r_statev4_lock);
12628 12628 dnlc_update(dvp, nm, vp);
12629 12629 } else {
12630 12630 /*
12631 12631 * XXX don't update if the created_v4 flag is
12632 12632 * set
12633 12633 */
12634 12634 mutex_exit(&VTOR4(vp)->r_statev4_lock);
12635 12635 NFS4_DEBUG(nfs4_client_state_debug,
12636 12636 (CE_NOTE, "nfs4_update_dircaches: "
12637 12637 "don't update dnlc: created_v4 flag"));
12638 12638 }
12639 12639 }
12640 12640
12641 12641 nfs4_attr_cache(dvp, dinfo->di_garp, dinfo->di_time_call,
12642 12642 dinfo->di_cred, FALSE, cinfo);
12643 12643
12644 12644 return;
12645 12645 }
12646 12646
12647 12647 /*
12648 12648 * Caller didn't provide dinfo, then check change_info4 to update DNLC.
12649 12649 * Since caller modified dir but didn't receive post-dirmod-op dir
12650 12650 * attrs, the dir's attrs must be purged.
12651 12651 *
12652 12652 * XXX this check and dnlc update/purge should really be atomic,
12653 12653 * XXX but can't use rnode statelock because it'll deadlock in
12654 12654 * XXX dnlc_purge_vp, however, the risk is minimal even if a race
12655 12655 * XXX does occur.
12656 12656 *
12657 12657 * XXX We also may want to check that atomic is true in the
12658 12658 * XXX change_info struct. If it is not, the change_info may
12659 12659 * XXX reflect changes by more than one clients which means that
12660 12660 * XXX our cache may not be valid.
12661 12661 */
12662 12662 PURGE_ATTRCACHE4(dvp);
12663 12663 if (drp->r_change == cinfo->before) {
12664 12664 /* no changes took place in the directory prior to our link */
12665 12665 if (vp != NULL) {
12666 12666 mutex_enter(&VTOR4(vp)->r_statev4_lock);
12667 12667 if (!VTOR4(vp)->created_v4) {
12668 12668 mutex_exit(&VTOR4(vp)->r_statev4_lock);
12669 12669 dnlc_update(dvp, nm, vp);
12670 12670 } else {
12671 12671 /*
12672 12672 * XXX dont' update if the created_v4 flag
12673 12673 * is set
12674 12674 */
12675 12675 mutex_exit(&VTOR4(vp)->r_statev4_lock);
12676 12676 NFS4_DEBUG(nfs4_client_state_debug, (CE_NOTE,
12677 12677 "nfs4_update_dircaches: don't"
12678 12678 " update dnlc: created_v4 flag"));
12679 12679 }
12680 12680 }
12681 12681 } else {
12682 12682 /* Another client modified directory - purge its dnlc cache */
12683 12683 dnlc_purge_vp(dvp);
12684 12684 }
12685 12685 }
12686 12686
12687 12687 /*
12688 12688 * The OPEN_CONFIRM operation confirms the sequence number used in OPENing a
12689 12689 * file.
12690 12690 *
12691 12691 * The 'reopening_file' boolean should be set to TRUE if we are reopening this
12692 12692 * file (ie: client recovery) and otherwise set to FALSE.
12693 12693 *
12694 12694 * 'nfs4_start/end_op' should have been called by the proper (ie: not recovery
12695 12695 * initiated) calling functions.
12696 12696 *
12697 12697 * 'resend' is set to TRUE if this is a OPEN_CONFIRM issued as a result
12698 12698 * of resending a 'lost' open request.
12699 12699 *
12700 12700 * 'num_bseqid_retryp' makes sure we don't loop forever on a broken
12701 12701 * server that hands out BAD_SEQID on open confirm.
12702 12702 *
12703 12703 * Errors are returned via the nfs4_error_t parameter.
12704 12704 */
12705 12705 void
12706 12706 nfs4open_confirm(vnode_t *vp, seqid4 *seqid, stateid4 *stateid, cred_t *cr,
12707 12707 bool_t reopening_file, bool_t *retry_open, nfs4_open_owner_t *oop,
12708 12708 bool_t resend, nfs4_error_t *ep, int *num_bseqid_retryp)
12709 12709 {
12710 12710 COMPOUND4args_clnt args;
12711 12711 COMPOUND4res_clnt res;
12712 12712 nfs_argop4 argop[2];
12713 12713 nfs_resop4 *resop;
12714 12714 int doqueue = 1;
12715 12715 mntinfo4_t *mi;
12716 12716 OPEN_CONFIRM4args *open_confirm_args;
12717 12717 int needrecov;
12718 12718
12719 12719 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
12720 12720 #if DEBUG
12721 12721 mutex_enter(&oop->oo_lock);
12722 12722 ASSERT(oop->oo_seqid_inuse);
12723 12723 mutex_exit(&oop->oo_lock);
12724 12724 #endif
12725 12725
12726 12726 recov_retry_confirm:
12727 12727 nfs4_error_zinit(ep);
12728 12728 *retry_open = FALSE;
12729 12729
12730 12730 if (resend)
12731 12731 args.ctag = TAG_OPEN_CONFIRM_LOST;
12732 12732 else
12733 12733 args.ctag = TAG_OPEN_CONFIRM;
12734 12734
12735 12735 args.array_len = 2;
12736 12736 args.array = argop;
12737 12737
12738 12738 /* putfh target fh */
12739 12739 argop[0].argop = OP_CPUTFH;
12740 12740 argop[0].nfs_argop4_u.opcputfh.sfh = VTOR4(vp)->r_fh;
12741 12741
12742 12742 argop[1].argop = OP_OPEN_CONFIRM;
12743 12743 open_confirm_args = &argop[1].nfs_argop4_u.opopen_confirm;
12744 12744
12745 12745 (*seqid) += 1;
12746 12746 open_confirm_args->seqid = *seqid;
12747 12747 open_confirm_args->open_stateid = *stateid;
12748 12748
12749 12749 mi = VTOMI4(vp);
12750 12750
12751 12751 rfs4call(mi, &args, &res, cr, &doqueue, 0, ep);
12752 12752
12753 12753 if (!ep->error && nfs4_need_to_bump_seqid(&res)) {
12754 12754 nfs4_set_open_seqid((*seqid), oop, args.ctag);
12755 12755 }
12756 12756
12757 12757 needrecov = nfs4_needs_recovery(ep, FALSE, mi->mi_vfsp);
12758 12758 if (!needrecov && ep->error)
12759 12759 return;
12760 12760
12761 12761 if (needrecov) {
12762 12762 bool_t abort = FALSE;
12763 12763
12764 12764 if (reopening_file == FALSE) {
12765 12765 nfs4_bseqid_entry_t *bsep = NULL;
12766 12766
12767 12767 if (!ep->error && res.status == NFS4ERR_BAD_SEQID)
12768 12768 bsep = nfs4_create_bseqid_entry(oop, NULL,
12769 12769 vp, 0, args.ctag,
12770 12770 open_confirm_args->seqid);
12771 12771
12772 12772 abort = nfs4_start_recovery(ep, VTOMI4(vp), vp, NULL,
12773 12773 NULL, NULL, OP_OPEN_CONFIRM, bsep, NULL, NULL);
12774 12774 if (bsep) {
12775 12775 kmem_free(bsep, sizeof (*bsep));
12776 12776 if (num_bseqid_retryp &&
12777 12777 --(*num_bseqid_retryp) == 0)
12778 12778 abort = TRUE;
12779 12779 }
12780 12780 }
12781 12781 if ((ep->error == ETIMEDOUT ||
12782 12782 res.status == NFS4ERR_RESOURCE) &&
12783 12783 abort == FALSE && resend == FALSE) {
12784 12784 if (!ep->error)
12785 12785 (void) xdr_free(xdr_COMPOUND4res_clnt,
12786 12786 (caddr_t)&res);
12787 12787
12788 12788 delay(SEC_TO_TICK(confirm_retry_sec));
12789 12789 goto recov_retry_confirm;
12790 12790 }
12791 12791 /* State may have changed so retry the entire OPEN op */
12792 12792 if (abort == FALSE)
12793 12793 *retry_open = TRUE;
12794 12794 else
12795 12795 *retry_open = FALSE;
12796 12796 if (!ep->error)
12797 12797 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
12798 12798 return;
12799 12799 }
12800 12800
12801 12801 if (res.status) {
12802 12802 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
12803 12803 return;
12804 12804 }
12805 12805
12806 12806 resop = &res.array[1]; /* open confirm res */
12807 12807 bcopy(&resop->nfs_resop4_u.opopen_confirm.open_stateid,
12808 12808 stateid, sizeof (*stateid));
12809 12809
12810 12810 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
12811 12811 }
12812 12812
12813 12813 /*
12814 12814 * Return the credentials associated with a client state object. The
12815 12815 * caller is responsible for freeing the credentials.
12816 12816 */
12817 12817
12818 12818 static cred_t *
12819 12819 state_to_cred(nfs4_open_stream_t *osp)
12820 12820 {
12821 12821 cred_t *cr;
12822 12822
12823 12823 /*
12824 12824 * It's ok to not lock the open stream and open owner to get
12825 12825 * the oo_cred since this is only written once (upon creation)
12826 12826 * and will not change.
12827 12827 */
12828 12828 cr = osp->os_open_owner->oo_cred;
12829 12829 crhold(cr);
12830 12830
12831 12831 return (cr);
12832 12832 }
12833 12833
12834 12834 /*
12835 12835 * nfs4_find_sysid
12836 12836 *
12837 12837 * Find the sysid for the knetconfig associated with the given mi.
12838 12838 */
12839 12839 static struct lm_sysid *
12840 12840 nfs4_find_sysid(mntinfo4_t *mi)
12841 12841 {
12842 12842 ASSERT(nfs_zone() == mi->mi_zone);
12843 12843
12844 12844 /*
12845 12845 * Switch from RDMA knconf to original mount knconf
12846 12846 */
12847 12847 return (lm_get_sysid(ORIG_KNCONF(mi), &mi->mi_curr_serv->sv_addr,
12848 12848 mi->mi_curr_serv->sv_hostname, NULL));
12849 12849 }
12850 12850
12851 12851 #ifdef DEBUG
12852 12852 /*
12853 12853 * Return a string version of the call type for easy reading.
12854 12854 */
12855 12855 static char *
12856 12856 nfs4frlock_get_call_type(nfs4_lock_call_type_t ctype)
12857 12857 {
12858 12858 switch (ctype) {
12859 12859 case NFS4_LCK_CTYPE_NORM:
12860 12860 return ("NORMAL");
12861 12861 case NFS4_LCK_CTYPE_RECLAIM:
12862 12862 return ("RECLAIM");
12863 12863 case NFS4_LCK_CTYPE_RESEND:
12864 12864 return ("RESEND");
12865 12865 case NFS4_LCK_CTYPE_REINSTATE:
12866 12866 return ("REINSTATE");
12867 12867 default:
12868 12868 cmn_err(CE_PANIC, "nfs4frlock_get_call_type: got illegal "
12869 12869 "type %d", ctype);
12870 12870 return ("");
12871 12871 }
12872 12872 }
12873 12873 #endif
12874 12874
12875 12875 /*
12876 12876 * Map the frlock cmd and lock type to the NFSv4 over-the-wire lock type
12877 12877 * Unlock requests don't have an over-the-wire locktype, so we just return
12878 12878 * something non-threatening.
12879 12879 */
12880 12880
12881 12881 static nfs_lock_type4
12882 12882 flk_to_locktype(int cmd, int l_type)
12883 12883 {
12884 12884 ASSERT(l_type == F_RDLCK || l_type == F_WRLCK || l_type == F_UNLCK);
12885 12885
12886 12886 switch (l_type) {
12887 12887 case F_UNLCK:
12888 12888 return (READ_LT);
12889 12889 case F_RDLCK:
12890 12890 if (cmd == F_SETLK)
12891 12891 return (READ_LT);
12892 12892 else
12893 12893 return (READW_LT);
12894 12894 case F_WRLCK:
12895 12895 if (cmd == F_SETLK)
12896 12896 return (WRITE_LT);
12897 12897 else
12898 12898 return (WRITEW_LT);
12899 12899 }
12900 12900 panic("flk_to_locktype");
12901 12901 /*NOTREACHED*/
12902 12902 }
12903 12903
12904 12904 /*
12905 12905 * Do some preliminary checks for nfs4frlock.
12906 12906 */
12907 12907 static int
12908 12908 nfs4frlock_validate_args(int cmd, flock64_t *flk, int flag, vnode_t *vp,
12909 12909 u_offset_t offset)
12910 12910 {
12911 12911 int error = 0;
12912 12912
12913 12913 /*
12914 12914 * If we are setting a lock, check that the file is opened
12915 12915 * with the correct mode.
12916 12916 */
12917 12917 if (cmd == F_SETLK || cmd == F_SETLKW) {
12918 12918 if ((flk->l_type == F_RDLCK && (flag & FREAD) == 0) ||
12919 12919 (flk->l_type == F_WRLCK && (flag & FWRITE) == 0)) {
12920 12920 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
12921 12921 "nfs4frlock_validate_args: file was opened with "
12922 12922 "incorrect mode"));
12923 12923 return (EBADF);
12924 12924 }
12925 12925 }
12926 12926
12927 12927 /* Convert the offset. It may need to be restored before returning. */
12928 12928 if (error = convoff(vp, flk, 0, offset)) {
12929 12929 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
12930 12930 "nfs4frlock_validate_args: convoff => error= %d\n",
12931 12931 error));
12932 12932 return (error);
12933 12933 }
12934 12934
12935 12935 return (error);
12936 12936 }
12937 12937
12938 12938 /*
12939 12939 * Set the flock64's lm_sysid for nfs4frlock.
12940 12940 */
12941 12941 static int
12942 12942 nfs4frlock_get_sysid(struct lm_sysid **lspp, vnode_t *vp, flock64_t *flk)
12943 12943 {
12944 12944 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
12945 12945
12946 12946 /* Find the lm_sysid */
12947 12947 *lspp = nfs4_find_sysid(VTOMI4(vp));
12948 12948
12949 12949 if (*lspp == NULL) {
12950 12950 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
12951 12951 "nfs4frlock_get_sysid: no sysid, return ENOLCK"));
12952 12952 return (ENOLCK);
12953 12953 }
12954 12954
12955 12955 flk->l_sysid = lm_sysidt(*lspp);
12956 12956
12957 12957 return (0);
12958 12958 }
12959 12959
12960 12960 /*
12961 12961 * Do the remaining preliminary setup for nfs4frlock.
12962 12962 */
12963 12963 static void
12964 12964 nfs4frlock_pre_setup(clock_t *tick_delayp, nfs4_recov_state_t *recov_statep,
12965 12965 flock64_t *flk, short *whencep, vnode_t *vp, cred_t *search_cr,
12966 12966 cred_t **cred_otw)
12967 12967 {
12968 12968 /*
12969 12969 * set tick_delay to the base delay time.
12970 12970 * (NFS4_BASE_WAIT_TIME is in secs)
12971 12971 */
12972 12972
12973 12973 *tick_delayp = drv_usectohz(NFS4_BASE_WAIT_TIME * 1000 * 1000);
12974 12974
12975 12975 /*
12976 12976 * If lock is relative to EOF, we need the newest length of the
12977 12977 * file. Therefore invalidate the ATTR_CACHE.
12978 12978 */
12979 12979
12980 12980 *whencep = flk->l_whence;
12981 12981
12982 12982 if (*whencep == 2) /* SEEK_END */
12983 12983 PURGE_ATTRCACHE4(vp);
12984 12984
12985 12985 recov_statep->rs_flags = 0;
12986 12986 recov_statep->rs_num_retry_despite_err = 0;
12987 12987 *cred_otw = nfs4_get_otw_cred(search_cr, VTOMI4(vp), NULL);
12988 12988 }
12989 12989
12990 12990 /*
12991 12991 * Initialize and allocate the data structures necessary for
12992 12992 * the nfs4frlock call.
12993 12993 * Allocates argsp's op array, frees up the saved_rqstpp if there is one.
12994 12994 */
12995 12995 static void
12996 12996 nfs4frlock_call_init(COMPOUND4args_clnt *argsp, COMPOUND4args_clnt **argspp,
12997 12997 nfs_argop4 **argopp, nfs4_op_hint_t *op_hintp, flock64_t *flk, int cmd,
12998 12998 bool_t *retry, bool_t *did_start_fop, COMPOUND4res_clnt **respp,
12999 12999 bool_t *skip_get_err, nfs4_lost_rqst_t *lost_rqstp)
13000 13000 {
13001 13001 int argoplist_size;
13002 13002 int num_ops = 2;
13003 13003
13004 13004 *retry = FALSE;
13005 13005 *did_start_fop = FALSE;
13006 13006 *skip_get_err = FALSE;
13007 13007 lost_rqstp->lr_op = 0;
13008 13008 argoplist_size = num_ops * sizeof (nfs_argop4);
13009 13009 /* fill array with zero */
13010 13010 *argopp = kmem_zalloc(argoplist_size, KM_SLEEP);
13011 13011
13012 13012 *argspp = argsp;
13013 13013 *respp = NULL;
13014 13014
13015 13015 argsp->array_len = num_ops;
13016 13016 argsp->array = *argopp;
13017 13017
13018 13018 /* initialize in case of error; will get real value down below */
13019 13019 argsp->ctag = TAG_NONE;
13020 13020
13021 13021 if ((cmd == F_SETLK || cmd == F_SETLKW) && flk->l_type == F_UNLCK)
13022 13022 *op_hintp = OH_LOCKU;
13023 13023 else
13024 13024 *op_hintp = OH_OTHER;
13025 13025 }
13026 13026
13027 13027 /*
13028 13028 * Call the nfs4_start_fop() for nfs4frlock, if necessary. Assign
13029 13029 * the proper nfs4_server_t for this instance of nfs4frlock.
13030 13030 * Returns 0 (success) or an errno value.
13031 13031 */
13032 13032 static int
13033 13033 nfs4frlock_start_call(nfs4_lock_call_type_t ctype, vnode_t *vp,
13034 13034 nfs4_op_hint_t op_hint, nfs4_recov_state_t *recov_statep,
13035 13035 bool_t *did_start_fop, bool_t *startrecovp)
13036 13036 {
13037 13037 int error = 0;
13038 13038 rnode4_t *rp;
13039 13039
13040 13040 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
13041 13041
13042 13042 if (ctype == NFS4_LCK_CTYPE_NORM) {
13043 13043 error = nfs4_start_fop(VTOMI4(vp), vp, NULL, op_hint,
13044 13044 recov_statep, startrecovp);
13045 13045 if (error)
13046 13046 return (error);
13047 13047 *did_start_fop = TRUE;
13048 13048 } else {
13049 13049 *did_start_fop = FALSE;
13050 13050 *startrecovp = FALSE;
13051 13051 }
13052 13052
13053 13053 if (!error) {
13054 13054 rp = VTOR4(vp);
13055 13055
13056 13056 /* If the file failed recovery, just quit. */
13057 13057 mutex_enter(&rp->r_statelock);
13058 13058 if (rp->r_flags & R4RECOVERR) {
13059 13059 error = EIO;
13060 13060 }
13061 13061 mutex_exit(&rp->r_statelock);
13062 13062 }
13063 13063
13064 13064 return (error);
13065 13065 }
13066 13066
13067 13067 /*
13068 13068 * Setup the LOCK4/LOCKU4 arguments for resending a lost lock request. A
13069 13069 * resend nfs4frlock call is initiated by the recovery framework.
13070 13070 * Acquires the lop and oop seqid synchronization.
13071 13071 */
13072 13072 static void
13073 13073 nfs4frlock_setup_resend_lock_args(nfs4_lost_rqst_t *resend_rqstp,
13074 13074 COMPOUND4args_clnt *argsp, nfs_argop4 *argop, nfs4_lock_owner_t **lopp,
13075 13075 nfs4_open_owner_t **oopp, nfs4_open_stream_t **ospp,
13076 13076 LOCK4args **lock_argsp, LOCKU4args **locku_argsp)
13077 13077 {
13078 13078 mntinfo4_t *mi = VTOMI4(resend_rqstp->lr_vp);
13079 13079 int error;
13080 13080
13081 13081 NFS4_DEBUG((nfs4_lost_rqst_debug || nfs4_client_lock_debug),
13082 13082 (CE_NOTE,
13083 13083 "nfs4frlock_setup_resend_lock_args: have lost lock to resend"));
13084 13084 ASSERT(resend_rqstp != NULL);
13085 13085 ASSERT(resend_rqstp->lr_op == OP_LOCK ||
13086 13086 resend_rqstp->lr_op == OP_LOCKU);
13087 13087
13088 13088 *oopp = resend_rqstp->lr_oop;
13089 13089 if (resend_rqstp->lr_oop) {
13090 13090 open_owner_hold(resend_rqstp->lr_oop);
13091 13091 error = nfs4_start_open_seqid_sync(resend_rqstp->lr_oop, mi);
13092 13092 ASSERT(error == 0); /* recov thread always succeeds */
13093 13093 }
13094 13094
13095 13095 /* Must resend this lost lock/locku request. */
13096 13096 ASSERT(resend_rqstp->lr_lop != NULL);
13097 13097 *lopp = resend_rqstp->lr_lop;
13098 13098 lock_owner_hold(resend_rqstp->lr_lop);
13099 13099 error = nfs4_start_lock_seqid_sync(resend_rqstp->lr_lop, mi);
13100 13100 ASSERT(error == 0); /* recov thread always succeeds */
13101 13101
13102 13102 *ospp = resend_rqstp->lr_osp;
13103 13103 if (*ospp)
13104 13104 open_stream_hold(resend_rqstp->lr_osp);
13105 13105
13106 13106 if (resend_rqstp->lr_op == OP_LOCK) {
13107 13107 LOCK4args *lock_args;
13108 13108
13109 13109 argop->argop = OP_LOCK;
13110 13110 *lock_argsp = lock_args = &argop->nfs_argop4_u.oplock;
13111 13111 lock_args->locktype = resend_rqstp->lr_locktype;
13112 13112 lock_args->reclaim =
13113 13113 (resend_rqstp->lr_ctype == NFS4_LCK_CTYPE_RECLAIM);
13114 13114 lock_args->offset = resend_rqstp->lr_flk->l_start;
13115 13115 lock_args->length = resend_rqstp->lr_flk->l_len;
13116 13116 if (lock_args->length == 0)
13117 13117 lock_args->length = ~lock_args->length;
13118 13118 nfs4_setup_lock_args(*lopp, *oopp, *ospp,
13119 13119 mi2clientid(mi), &lock_args->locker);
13120 13120
13121 13121 switch (resend_rqstp->lr_ctype) {
13122 13122 case NFS4_LCK_CTYPE_RESEND:
13123 13123 argsp->ctag = TAG_LOCK_RESEND;
13124 13124 break;
13125 13125 case NFS4_LCK_CTYPE_REINSTATE:
13126 13126 argsp->ctag = TAG_LOCK_REINSTATE;
13127 13127 break;
13128 13128 case NFS4_LCK_CTYPE_RECLAIM:
13129 13129 argsp->ctag = TAG_LOCK_RECLAIM;
13130 13130 break;
13131 13131 default:
13132 13132 argsp->ctag = TAG_LOCK_UNKNOWN;
13133 13133 break;
13134 13134 }
13135 13135 } else {
13136 13136 LOCKU4args *locku_args;
13137 13137 nfs4_lock_owner_t *lop = resend_rqstp->lr_lop;
13138 13138
13139 13139 argop->argop = OP_LOCKU;
13140 13140 *locku_argsp = locku_args = &argop->nfs_argop4_u.oplocku;
13141 13141 locku_args->locktype = READ_LT;
13142 13142 locku_args->seqid = lop->lock_seqid + 1;
13143 13143 mutex_enter(&lop->lo_lock);
13144 13144 locku_args->lock_stateid = lop->lock_stateid;
13145 13145 mutex_exit(&lop->lo_lock);
13146 13146 locku_args->offset = resend_rqstp->lr_flk->l_start;
13147 13147 locku_args->length = resend_rqstp->lr_flk->l_len;
13148 13148 if (locku_args->length == 0)
13149 13149 locku_args->length = ~locku_args->length;
13150 13150
13151 13151 switch (resend_rqstp->lr_ctype) {
13152 13152 case NFS4_LCK_CTYPE_RESEND:
13153 13153 argsp->ctag = TAG_LOCKU_RESEND;
13154 13154 break;
13155 13155 case NFS4_LCK_CTYPE_REINSTATE:
13156 13156 argsp->ctag = TAG_LOCKU_REINSTATE;
13157 13157 break;
13158 13158 default:
13159 13159 argsp->ctag = TAG_LOCK_UNKNOWN;
13160 13160 break;
13161 13161 }
13162 13162 }
13163 13163 }
13164 13164
13165 13165 /*
13166 13166 * Setup the LOCKT4 arguments.
13167 13167 */
13168 13168 static void
13169 13169 nfs4frlock_setup_lockt_args(nfs4_lock_call_type_t ctype, nfs_argop4 *argop,
13170 13170 LOCKT4args **lockt_argsp, COMPOUND4args_clnt *argsp, flock64_t *flk,
13171 13171 rnode4_t *rp)
13172 13172 {
13173 13173 LOCKT4args *lockt_args;
13174 13174
13175 13175 ASSERT(nfs_zone() == VTOMI4(RTOV4(rp))->mi_zone);
13176 13176 ASSERT(ctype == NFS4_LCK_CTYPE_NORM);
13177 13177 argop->argop = OP_LOCKT;
13178 13178 argsp->ctag = TAG_LOCKT;
13179 13179 lockt_args = &argop->nfs_argop4_u.oplockt;
13180 13180
13181 13181 /*
13182 13182 * The locktype will be READ_LT unless it's
13183 13183 * a write lock. We do this because the Solaris
13184 13184 * system call allows the combination of
13185 13185 * F_UNLCK and F_GETLK* and so in that case the
13186 13186 * unlock is mapped to a read.
13187 13187 */
13188 13188 if (flk->l_type == F_WRLCK)
13189 13189 lockt_args->locktype = WRITE_LT;
13190 13190 else
13191 13191 lockt_args->locktype = READ_LT;
13192 13192
13193 13193 lockt_args->owner.clientid = mi2clientid(VTOMI4(RTOV4(rp)));
13194 13194 /* set the lock owner4 args */
13195 13195 nfs4_setlockowner_args(&lockt_args->owner, rp,
13196 13196 ctype == NFS4_LCK_CTYPE_NORM ? curproc->p_pidp->pid_id :
13197 13197 flk->l_pid);
13198 13198 lockt_args->offset = flk->l_start;
13199 13199 lockt_args->length = flk->l_len;
13200 13200 if (flk->l_len == 0)
13201 13201 lockt_args->length = ~lockt_args->length;
13202 13202
13203 13203 *lockt_argsp = lockt_args;
13204 13204 }
13205 13205
13206 13206 /*
13207 13207 * If the client is holding a delegation, and the open stream to be used
13208 13208 * with this lock request is a delegation open stream, then re-open the stream.
13209 13209 * Sets the nfs4_error_t to all zeros unless the open stream has already
13210 13210 * failed a reopen or we couldn't find the open stream. NFS4ERR_DELAY
13211 13211 * means the caller should retry (like a recovery retry).
13212 13212 */
13213 13213 static void
13214 13214 nfs4frlock_check_deleg(vnode_t *vp, nfs4_error_t *ep, cred_t *cr, int lt)
13215 13215 {
13216 13216 open_delegation_type4 dt;
13217 13217 bool_t reopen_needed, force;
13218 13218 nfs4_open_stream_t *osp;
13219 13219 open_claim_type4 oclaim;
13220 13220 rnode4_t *rp = VTOR4(vp);
13221 13221 mntinfo4_t *mi = VTOMI4(vp);
13222 13222
13223 13223 ASSERT(nfs_zone() == mi->mi_zone);
13224 13224
13225 13225 nfs4_error_zinit(ep);
13226 13226
13227 13227 mutex_enter(&rp->r_statev4_lock);
13228 13228 dt = rp->r_deleg_type;
13229 13229 mutex_exit(&rp->r_statev4_lock);
13230 13230
13231 13231 if (dt != OPEN_DELEGATE_NONE) {
13232 13232 nfs4_open_owner_t *oop;
13233 13233
13234 13234 oop = find_open_owner(cr, NFS4_PERM_CREATED, mi);
13235 13235 if (!oop) {
13236 13236 ep->stat = NFS4ERR_IO;
13237 13237 return;
13238 13238 }
13239 13239 /* returns with 'os_sync_lock' held */
13240 13240 osp = find_open_stream(oop, rp);
13241 13241 if (!osp) {
13242 13242 open_owner_rele(oop);
13243 13243 ep->stat = NFS4ERR_IO;
13244 13244 return;
13245 13245 }
13246 13246
13247 13247 if (osp->os_failed_reopen) {
13248 13248 NFS4_DEBUG((nfs4_open_stream_debug ||
13249 13249 nfs4_client_lock_debug), (CE_NOTE,
13250 13250 "nfs4frlock_check_deleg: os_failed_reopen set "
13251 13251 "for osp %p, cr %p, rp %s", (void *)osp,
13252 13252 (void *)cr, rnode4info(rp)));
13253 13253 mutex_exit(&osp->os_sync_lock);
13254 13254 open_stream_rele(osp, rp);
13255 13255 open_owner_rele(oop);
13256 13256 ep->stat = NFS4ERR_IO;
13257 13257 return;
13258 13258 }
13259 13259
13260 13260 /*
13261 13261 * Determine whether a reopen is needed. If this
13262 13262 * is a delegation open stream, then send the open
13263 13263 * to the server to give visibility to the open owner.
13264 13264 * Even if it isn't a delegation open stream, we need
13265 13265 * to check if the previous open CLAIM_DELEGATE_CUR
13266 13266 * was sufficient.
13267 13267 */
13268 13268
13269 13269 reopen_needed = osp->os_delegation ||
13270 13270 ((lt == F_RDLCK &&
13271 13271 !(osp->os_dc_openacc & OPEN4_SHARE_ACCESS_READ)) ||
13272 13272 (lt == F_WRLCK &&
13273 13273 !(osp->os_dc_openacc & OPEN4_SHARE_ACCESS_WRITE)));
13274 13274
13275 13275 mutex_exit(&osp->os_sync_lock);
13276 13276 open_owner_rele(oop);
13277 13277
13278 13278 if (reopen_needed) {
13279 13279 /*
13280 13280 * Always use CLAIM_PREVIOUS after server reboot.
13281 13281 * The server will reject CLAIM_DELEGATE_CUR if
13282 13282 * it is used during the grace period.
13283 13283 */
13284 13284 mutex_enter(&mi->mi_lock);
13285 13285 if (mi->mi_recovflags & MI4R_SRV_REBOOT) {
13286 13286 oclaim = CLAIM_PREVIOUS;
13287 13287 force = TRUE;
13288 13288 } else {
13289 13289 oclaim = CLAIM_DELEGATE_CUR;
13290 13290 force = FALSE;
13291 13291 }
13292 13292 mutex_exit(&mi->mi_lock);
13293 13293
13294 13294 nfs4_reopen(vp, osp, ep, oclaim, force, FALSE);
13295 13295 if (ep->error == EAGAIN) {
13296 13296 nfs4_error_zinit(ep);
13297 13297 ep->stat = NFS4ERR_DELAY;
13298 13298 }
13299 13299 }
13300 13300 open_stream_rele(osp, rp);
13301 13301 osp = NULL;
13302 13302 }
13303 13303 }
13304 13304
13305 13305 /*
13306 13306 * Setup the LOCKU4 arguments.
13307 13307 * Returns errors via the nfs4_error_t.
13308 13308 * NFS4_OK no problems. *go_otwp is TRUE if call should go
13309 13309 * over-the-wire. The caller must release the
13310 13310 * reference on *lopp.
13311 13311 * NFS4ERR_DELAY caller should retry (like recovery retry)
13312 13312 * (other) unrecoverable error.
13313 13313 */
13314 13314 static void
13315 13315 nfs4frlock_setup_locku_args(nfs4_lock_call_type_t ctype, nfs_argop4 *argop,
13316 13316 LOCKU4args **locku_argsp, flock64_t *flk,
13317 13317 nfs4_lock_owner_t **lopp, nfs4_error_t *ep, COMPOUND4args_clnt *argsp,
13318 13318 vnode_t *vp, int flag, u_offset_t offset, cred_t *cr,
13319 13319 bool_t *skip_get_err, bool_t *go_otwp)
13320 13320 {
13321 13321 nfs4_lock_owner_t *lop = NULL;
13322 13322 LOCKU4args *locku_args;
13323 13323 pid_t pid;
13324 13324 bool_t is_spec = FALSE;
13325 13325 rnode4_t *rp = VTOR4(vp);
13326 13326
13327 13327 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
13328 13328 ASSERT(ctype == NFS4_LCK_CTYPE_NORM);
13329 13329
13330 13330 nfs4frlock_check_deleg(vp, ep, cr, F_UNLCK);
13331 13331 if (ep->error || ep->stat)
13332 13332 return;
13333 13333
13334 13334 argop->argop = OP_LOCKU;
13335 13335 if (ctype == NFS4_LCK_CTYPE_REINSTATE)
13336 13336 argsp->ctag = TAG_LOCKU_REINSTATE;
13337 13337 else
13338 13338 argsp->ctag = TAG_LOCKU;
13339 13339 locku_args = &argop->nfs_argop4_u.oplocku;
13340 13340 *locku_argsp = locku_args;
13341 13341
13342 13342 /*
13343 13343 * XXX what should locku_args->locktype be?
13344 13344 * setting to ALWAYS be READ_LT so at least
13345 13345 * it is a valid locktype.
13346 13346 */
13347 13347
13348 13348 locku_args->locktype = READ_LT;
13349 13349
13350 13350 pid = ctype == NFS4_LCK_CTYPE_NORM ? curproc->p_pidp->pid_id :
13351 13351 flk->l_pid;
13352 13352
13353 13353 /*
13354 13354 * Get the lock owner stateid. If no lock owner
13355 13355 * exists, return success.
13356 13356 */
13357 13357 lop = find_lock_owner(rp, pid, LOWN_ANY);
13358 13358 *lopp = lop;
13359 13359 if (lop && CLNT_ISSPECIAL(&lop->lock_stateid))
13360 13360 is_spec = TRUE;
13361 13361 if (!lop || is_spec) {
13362 13362 /*
13363 13363 * No lock owner so no locks to unlock.
13364 13364 * Return success. If there was a failed
13365 13365 * reclaim earlier, the lock might still be
13366 13366 * registered with the local locking code,
13367 13367 * so notify it of the unlock.
13368 13368 *
13369 13369 * If the lockowner is using a special stateid,
13370 13370 * then the original lock request (that created
13371 13371 * this lockowner) was never successful, so we
13372 13372 * have no lock to undo OTW.
13373 13373 */
13374 13374 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
13375 13375 "nfs4frlock_setup_locku_args: LOCKU: no lock owner "
13376 13376 "(%ld) so return success", (long)pid));
13377 13377
13378 13378 if (ctype == NFS4_LCK_CTYPE_NORM)
13379 13379 flk->l_pid = curproc->p_pid;
13380 13380 nfs4_register_lock_locally(vp, flk, flag, offset);
13381 13381 /*
13382 13382 * Release our hold and NULL out so final_cleanup
13383 13383 * doesn't try to end a lock seqid sync we
13384 13384 * never started.
13385 13385 */
13386 13386 if (is_spec) {
13387 13387 lock_owner_rele(lop);
13388 13388 *lopp = NULL;
13389 13389 }
13390 13390 *skip_get_err = TRUE;
13391 13391 *go_otwp = FALSE;
13392 13392 return;
13393 13393 }
13394 13394
13395 13395 ep->error = nfs4_start_lock_seqid_sync(lop, VTOMI4(vp));
13396 13396 if (ep->error == EAGAIN) {
13397 13397 lock_owner_rele(lop);
13398 13398 *lopp = NULL;
13399 13399 return;
13400 13400 }
13401 13401
13402 13402 mutex_enter(&lop->lo_lock);
13403 13403 locku_args->lock_stateid = lop->lock_stateid;
13404 13404 mutex_exit(&lop->lo_lock);
13405 13405 locku_args->seqid = lop->lock_seqid + 1;
13406 13406
13407 13407 /* leave the ref count on lop, rele after RPC call */
13408 13408
13409 13409 locku_args->offset = flk->l_start;
13410 13410 locku_args->length = flk->l_len;
13411 13411 if (flk->l_len == 0)
13412 13412 locku_args->length = ~locku_args->length;
13413 13413
13414 13414 *go_otwp = TRUE;
13415 13415 }
13416 13416
13417 13417 /*
13418 13418 * Setup the LOCK4 arguments.
13419 13419 *
13420 13420 * Returns errors via the nfs4_error_t.
13421 13421 * NFS4_OK no problems
13422 13422 * NFS4ERR_DELAY caller should retry (like recovery retry)
13423 13423 * (other) unrecoverable error
13424 13424 */
13425 13425 static void
13426 13426 nfs4frlock_setup_lock_args(nfs4_lock_call_type_t ctype, LOCK4args **lock_argsp,
13427 13427 nfs4_open_owner_t **oopp, nfs4_open_stream_t **ospp,
13428 13428 nfs4_lock_owner_t **lopp, nfs_argop4 *argop, COMPOUND4args_clnt *argsp,
13429 13429 flock64_t *flk, int cmd, vnode_t *vp, cred_t *cr, nfs4_error_t *ep)
13430 13430 {
13431 13431 LOCK4args *lock_args;
13432 13432 nfs4_open_owner_t *oop = NULL;
13433 13433 nfs4_open_stream_t *osp = NULL;
13434 13434 nfs4_lock_owner_t *lop = NULL;
13435 13435 pid_t pid;
13436 13436 rnode4_t *rp = VTOR4(vp);
13437 13437
13438 13438 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
13439 13439
13440 13440 nfs4frlock_check_deleg(vp, ep, cr, flk->l_type);
13441 13441 if (ep->error || ep->stat != NFS4_OK)
13442 13442 return;
13443 13443
13444 13444 argop->argop = OP_LOCK;
13445 13445 if (ctype == NFS4_LCK_CTYPE_NORM)
13446 13446 argsp->ctag = TAG_LOCK;
13447 13447 else if (ctype == NFS4_LCK_CTYPE_RECLAIM)
13448 13448 argsp->ctag = TAG_RELOCK;
13449 13449 else
13450 13450 argsp->ctag = TAG_LOCK_REINSTATE;
13451 13451 lock_args = &argop->nfs_argop4_u.oplock;
13452 13452 lock_args->locktype = flk_to_locktype(cmd, flk->l_type);
13453 13453 lock_args->reclaim = ctype == NFS4_LCK_CTYPE_RECLAIM ? 1 : 0;
13454 13454 /*
13455 13455 * Get the lock owner. If no lock owner exists,
13456 13456 * create a 'temporary' one and grab the open seqid
13457 13457 * synchronization (which puts a hold on the open
13458 13458 * owner and open stream).
13459 13459 * This also grabs the lock seqid synchronization.
13460 13460 */
13461 13461 pid = ctype == NFS4_LCK_CTYPE_NORM ? curproc->p_pid : flk->l_pid;
13462 13462 ep->stat =
13463 13463 nfs4_find_or_create_lock_owner(pid, rp, cr, &oop, &osp, &lop);
13464 13464
13465 13465 if (ep->stat != NFS4_OK)
13466 13466 goto out;
13467 13467
13468 13468 nfs4_setup_lock_args(lop, oop, osp, mi2clientid(VTOMI4(vp)),
13469 13469 &lock_args->locker);
13470 13470
13471 13471 lock_args->offset = flk->l_start;
13472 13472 lock_args->length = flk->l_len;
13473 13473 if (flk->l_len == 0)
13474 13474 lock_args->length = ~lock_args->length;
13475 13475 *lock_argsp = lock_args;
13476 13476 out:
13477 13477 *oopp = oop;
13478 13478 *ospp = osp;
13479 13479 *lopp = lop;
13480 13480 }
13481 13481
13482 13482 /*
13483 13483 * After we get the reply from the server, record the proper information
13484 13484 * for possible resend lock requests.
13485 13485 *
13486 13486 * Allocates memory for the saved_rqstp if we have a lost lock to save.
13487 13487 */
13488 13488 static void
13489 13489 nfs4frlock_save_lost_rqst(nfs4_lock_call_type_t ctype, int error,
13490 13490 nfs_lock_type4 locktype, nfs4_open_owner_t *oop,
13491 13491 nfs4_open_stream_t *osp, nfs4_lock_owner_t *lop, flock64_t *flk,
13492 13492 nfs4_lost_rqst_t *lost_rqstp, cred_t *cr, vnode_t *vp)
13493 13493 {
13494 13494 bool_t unlock = (flk->l_type == F_UNLCK);
13495 13495
13496 13496 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
13497 13497 ASSERT(ctype == NFS4_LCK_CTYPE_NORM ||
13498 13498 ctype == NFS4_LCK_CTYPE_REINSTATE);
13499 13499
13500 13500 if (error != 0 && !unlock) {
13501 13501 NFS4_DEBUG((nfs4_lost_rqst_debug ||
13502 13502 nfs4_client_lock_debug), (CE_NOTE,
13503 13503 "nfs4frlock_save_lost_rqst: set lo_pending_rqsts to 1 "
13504 13504 " for lop %p", (void *)lop));
13505 13505 ASSERT(lop != NULL);
13506 13506 mutex_enter(&lop->lo_lock);
13507 13507 lop->lo_pending_rqsts = 1;
13508 13508 mutex_exit(&lop->lo_lock);
13509 13509 }
13510 13510
13511 13511 lost_rqstp->lr_putfirst = FALSE;
13512 13512 lost_rqstp->lr_op = 0;
13513 13513
13514 13514 /*
13515 13515 * For lock/locku requests, we treat EINTR as ETIMEDOUT for
13516 13516 * recovery purposes so that the lock request that was sent
13517 13517 * can be saved and re-issued later. Ditto for EIO from a forced
13518 13518 * unmount. This is done to have the client's local locking state
13519 13519 * match the v4 server's state; that is, the request was
13520 13520 * potentially received and accepted by the server but the client
13521 13521 * thinks it was not.
13522 13522 */
13523 13523 if (error == ETIMEDOUT || error == EINTR ||
13524 13524 NFS4_FRC_UNMT_ERR(error, vp->v_vfsp)) {
13525 13525 NFS4_DEBUG((nfs4_lost_rqst_debug ||
13526 13526 nfs4_client_lock_debug), (CE_NOTE,
13527 13527 "nfs4frlock_save_lost_rqst: got a lost %s lock for "
13528 13528 "lop %p oop %p osp %p", unlock ? "LOCKU" : "LOCK",
13529 13529 (void *)lop, (void *)oop, (void *)osp));
13530 13530 if (unlock)
13531 13531 lost_rqstp->lr_op = OP_LOCKU;
13532 13532 else {
13533 13533 lost_rqstp->lr_op = OP_LOCK;
13534 13534 lost_rqstp->lr_locktype = locktype;
13535 13535 }
13536 13536 /*
13537 13537 * Objects are held and rele'd via the recovery code.
13538 13538 * See nfs4_save_lost_rqst.
13539 13539 */
13540 13540 lost_rqstp->lr_vp = vp;
13541 13541 lost_rqstp->lr_dvp = NULL;
13542 13542 lost_rqstp->lr_oop = oop;
13543 13543 lost_rqstp->lr_osp = osp;
13544 13544 lost_rqstp->lr_lop = lop;
13545 13545 lost_rqstp->lr_cr = cr;
13546 13546 switch (ctype) {
13547 13547 case NFS4_LCK_CTYPE_NORM:
13548 13548 flk->l_pid = ttoproc(curthread)->p_pid;
13549 13549 lost_rqstp->lr_ctype = NFS4_LCK_CTYPE_RESEND;
13550 13550 break;
13551 13551 case NFS4_LCK_CTYPE_REINSTATE:
13552 13552 lost_rqstp->lr_putfirst = TRUE;
13553 13553 lost_rqstp->lr_ctype = ctype;
13554 13554 break;
13555 13555 default:
13556 13556 break;
13557 13557 }
13558 13558 lost_rqstp->lr_flk = flk;
13559 13559 }
13560 13560 }
13561 13561
13562 13562 /*
13563 13563 * Update lop's seqid. Also update the seqid stored in a resend request,
13564 13564 * if any. (Some recovery errors increment the seqid, and we may have to
13565 13565 * send the resend request again.)
13566 13566 */
13567 13567
13568 13568 static void
13569 13569 nfs4frlock_bump_seqid(LOCK4args *lock_args, LOCKU4args *locku_args,
13570 13570 nfs4_open_owner_t *oop, nfs4_lock_owner_t *lop, nfs4_tag_type_t tag_type)
13571 13571 {
13572 13572 if (lock_args) {
13573 13573 if (lock_args->locker.new_lock_owner == TRUE)
13574 13574 nfs4_get_and_set_next_open_seqid(oop, tag_type);
13575 13575 else {
13576 13576 ASSERT(lop->lo_flags & NFS4_LOCK_SEQID_INUSE);
13577 13577 nfs4_set_lock_seqid(lop->lock_seqid + 1, lop);
13578 13578 }
13579 13579 } else if (locku_args) {
13580 13580 ASSERT(lop->lo_flags & NFS4_LOCK_SEQID_INUSE);
13581 13581 nfs4_set_lock_seqid(lop->lock_seqid +1, lop);
13582 13582 }
13583 13583 }
13584 13584
13585 13585 /*
13586 13586 * Calls nfs4_end_fop, drops the seqid syncs, and frees up the
13587 13587 * COMPOUND4 args/res for calls that need to retry.
13588 13588 * Switches the *cred_otwp to base_cr.
13589 13589 */
13590 13590 static void
13591 13591 nfs4frlock_check_access(vnode_t *vp, nfs4_op_hint_t op_hint,
13592 13592 nfs4_recov_state_t *recov_statep, int needrecov, bool_t *did_start_fop,
13593 13593 COMPOUND4args_clnt **argspp, COMPOUND4res_clnt **respp, int error,
13594 13594 nfs4_lock_owner_t **lopp, nfs4_open_owner_t **oopp,
13595 13595 nfs4_open_stream_t **ospp, cred_t *base_cr, cred_t **cred_otwp)
13596 13596 {
13597 13597 nfs4_open_owner_t *oop = *oopp;
13598 13598 nfs4_open_stream_t *osp = *ospp;
13599 13599 nfs4_lock_owner_t *lop = *lopp;
13600 13600 nfs_argop4 *argop = (*argspp)->array;
13601 13601
13602 13602 if (*did_start_fop) {
13603 13603 nfs4_end_fop(VTOMI4(vp), vp, NULL, op_hint, recov_statep,
13604 13604 needrecov);
13605 13605 *did_start_fop = FALSE;
13606 13606 }
13607 13607 ASSERT((*argspp)->array_len == 2);
13608 13608 if (argop[1].argop == OP_LOCK)
13609 13609 nfs4args_lock_free(&argop[1]);
13610 13610 else if (argop[1].argop == OP_LOCKT)
13611 13611 nfs4args_lockt_free(&argop[1]);
13612 13612 kmem_free(argop, 2 * sizeof (nfs_argop4));
13613 13613 if (!error)
13614 13614 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)*respp);
13615 13615 *argspp = NULL;
13616 13616 *respp = NULL;
13617 13617
13618 13618 if (lop) {
13619 13619 nfs4_end_lock_seqid_sync(lop);
13620 13620 lock_owner_rele(lop);
13621 13621 *lopp = NULL;
13622 13622 }
13623 13623
13624 13624 /* need to free up the reference on osp for lock args */
13625 13625 if (osp != NULL) {
13626 13626 open_stream_rele(osp, VTOR4(vp));
13627 13627 *ospp = NULL;
13628 13628 }
13629 13629
13630 13630 /* need to free up the reference on oop for lock args */
13631 13631 if (oop != NULL) {
13632 13632 nfs4_end_open_seqid_sync(oop);
13633 13633 open_owner_rele(oop);
13634 13634 *oopp = NULL;
13635 13635 }
13636 13636
13637 13637 crfree(*cred_otwp);
13638 13638 *cred_otwp = base_cr;
13639 13639 crhold(*cred_otwp);
13640 13640 }
13641 13641
13642 13642 /*
13643 13643 * Function to process the client's recovery for nfs4frlock.
13644 13644 * Returns TRUE if we should retry the lock request; FALSE otherwise.
13645 13645 *
13646 13646 * Calls nfs4_end_fop, drops the seqid syncs, and frees up the
13647 13647 * COMPOUND4 args/res for calls that need to retry.
13648 13648 *
13649 13649 * Note: the rp's r_lkserlock is *not* dropped during this path.
13650 13650 */
13651 13651 static bool_t
13652 13652 nfs4frlock_recovery(int needrecov, nfs4_error_t *ep,
13653 13653 COMPOUND4args_clnt **argspp, COMPOUND4res_clnt **respp,
13654 13654 LOCK4args *lock_args, LOCKU4args *locku_args,
13655 13655 nfs4_open_owner_t **oopp, nfs4_open_stream_t **ospp,
13656 13656 nfs4_lock_owner_t **lopp, rnode4_t *rp, vnode_t *vp,
13657 13657 nfs4_recov_state_t *recov_statep, nfs4_op_hint_t op_hint,
13658 13658 bool_t *did_start_fop, nfs4_lost_rqst_t *lost_rqstp, flock64_t *flk)
13659 13659 {
13660 13660 nfs4_open_owner_t *oop = *oopp;
13661 13661 nfs4_open_stream_t *osp = *ospp;
13662 13662 nfs4_lock_owner_t *lop = *lopp;
13663 13663
13664 13664 bool_t abort, retry;
13665 13665
13666 13666 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
13667 13667 ASSERT((*argspp) != NULL);
13668 13668 ASSERT((*respp) != NULL);
13669 13669 if (lock_args || locku_args)
13670 13670 ASSERT(lop != NULL);
13671 13671
13672 13672 NFS4_DEBUG((nfs4_client_lock_debug || nfs4_client_recov_debug),
13673 13673 (CE_NOTE, "nfs4frlock_recovery: initiating recovery\n"));
13674 13674
13675 13675 retry = TRUE;
13676 13676 abort = FALSE;
13677 13677 if (needrecov) {
13678 13678 nfs4_bseqid_entry_t *bsep = NULL;
13679 13679 nfs_opnum4 op;
13680 13680
13681 13681 op = lock_args ? OP_LOCK : locku_args ? OP_LOCKU : OP_LOCKT;
13682 13682
13683 13683 if (!ep->error && ep->stat == NFS4ERR_BAD_SEQID) {
13684 13684 seqid4 seqid;
13685 13685
13686 13686 if (lock_args) {
13687 13687 if (lock_args->locker.new_lock_owner == TRUE)
13688 13688 seqid = lock_args->locker.locker4_u.
13689 13689 open_owner.open_seqid;
13690 13690 else
13691 13691 seqid = lock_args->locker.locker4_u.
13692 13692 lock_owner.lock_seqid;
13693 13693 } else if (locku_args) {
13694 13694 seqid = locku_args->seqid;
13695 13695 } else {
13696 13696 seqid = 0;
13697 13697 }
13698 13698
13699 13699 bsep = nfs4_create_bseqid_entry(oop, lop, vp,
13700 13700 flk->l_pid, (*argspp)->ctag, seqid);
13701 13701 }
13702 13702
13703 13703 abort = nfs4_start_recovery(ep, VTOMI4(vp), vp, NULL, NULL,
13704 13704 (lost_rqstp && (lost_rqstp->lr_op == OP_LOCK ||
13705 13705 lost_rqstp->lr_op == OP_LOCKU)) ? lost_rqstp :
13706 13706 NULL, op, bsep, NULL, NULL);
13707 13707
13708 13708 if (bsep)
13709 13709 kmem_free(bsep, sizeof (*bsep));
13710 13710 }
13711 13711
13712 13712 /*
13713 13713 * Return that we do not want to retry the request for 3 cases:
13714 13714 * 1. If we received EINTR or are bailing out because of a forced
13715 13715 * unmount, we came into this code path just for the sake of
13716 13716 * initiating recovery, we now need to return the error.
13717 13717 * 2. If we have aborted recovery.
13718 13718 * 3. We received NFS4ERR_BAD_SEQID.
13719 13719 */
13720 13720 if (ep->error == EINTR || NFS4_FRC_UNMT_ERR(ep->error, vp->v_vfsp) ||
13721 13721 abort == TRUE || (ep->error == 0 && ep->stat == NFS4ERR_BAD_SEQID))
13722 13722 retry = FALSE;
13723 13723
13724 13724 if (*did_start_fop == TRUE) {
13725 13725 nfs4_end_fop(VTOMI4(vp), vp, NULL, op_hint, recov_statep,
13726 13726 needrecov);
13727 13727 *did_start_fop = FALSE;
13728 13728 }
13729 13729
13730 13730 if (retry == TRUE) {
13731 13731 nfs_argop4 *argop;
13732 13732
13733 13733 argop = (*argspp)->array;
13734 13734 ASSERT((*argspp)->array_len == 2);
13735 13735
13736 13736 if (argop[1].argop == OP_LOCK)
13737 13737 nfs4args_lock_free(&argop[1]);
13738 13738 else if (argop[1].argop == OP_LOCKT)
13739 13739 nfs4args_lockt_free(&argop[1]);
13740 13740 kmem_free(argop, 2 * sizeof (nfs_argop4));
13741 13741 if (!ep->error)
13742 13742 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)*respp);
13743 13743 *respp = NULL;
13744 13744 *argspp = NULL;
13745 13745 }
13746 13746
13747 13747 if (lop != NULL) {
13748 13748 nfs4_end_lock_seqid_sync(lop);
13749 13749 lock_owner_rele(lop);
13750 13750 }
13751 13751
13752 13752 *lopp = NULL;
13753 13753
13754 13754 /* need to free up the reference on osp for lock args */
13755 13755 if (osp != NULL) {
13756 13756 open_stream_rele(osp, rp);
13757 13757 *ospp = NULL;
13758 13758 }
13759 13759
13760 13760 /* need to free up the reference on oop for lock args */
13761 13761 if (oop != NULL) {
13762 13762 nfs4_end_open_seqid_sync(oop);
13763 13763 open_owner_rele(oop);
13764 13764 *oopp = NULL;
13765 13765 }
13766 13766
13767 13767 return (retry);
13768 13768 }
13769 13769
13770 13770 /*
13771 13771 * Handles the successful reply from the server for nfs4frlock.
13772 13772 */
13773 13773 static void
13774 13774 nfs4frlock_results_ok(nfs4_lock_call_type_t ctype, int cmd, flock64_t *flk,
13775 13775 vnode_t *vp, int flag, u_offset_t offset,
13776 13776 nfs4_lost_rqst_t *resend_rqstp)
13777 13777 {
13778 13778 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
13779 13779 if ((cmd == F_SETLK || cmd == F_SETLKW) &&
13780 13780 (flk->l_type == F_RDLCK || flk->l_type == F_WRLCK)) {
13781 13781 if (ctype == NFS4_LCK_CTYPE_NORM) {
13782 13782 flk->l_pid = ttoproc(curthread)->p_pid;
13783 13783 /*
13784 13784 * We do not register lost locks locally in
13785 13785 * the 'resend' case since the user/application
13786 13786 * doesn't think we have the lock.
13787 13787 */
13788 13788 ASSERT(!resend_rqstp);
13789 13789 nfs4_register_lock_locally(vp, flk, flag, offset);
13790 13790 }
13791 13791 }
13792 13792 }
13793 13793
13794 13794 /*
13795 13795 * Handle the DENIED reply from the server for nfs4frlock.
13796 13796 * Returns TRUE if we should retry the request; FALSE otherwise.
13797 13797 *
13798 13798 * Calls nfs4_end_fop, drops the seqid syncs, and frees up the
13799 13799 * COMPOUND4 args/res for calls that need to retry. Can also
13800 13800 * drop and regrab the r_lkserlock.
13801 13801 */
13802 13802 static bool_t
13803 13803 nfs4frlock_results_denied(nfs4_lock_call_type_t ctype, LOCK4args *lock_args,
13804 13804 LOCKT4args *lockt_args, nfs4_open_owner_t **oopp,
13805 13805 nfs4_open_stream_t **ospp, nfs4_lock_owner_t **lopp, int cmd,
13806 13806 vnode_t *vp, flock64_t *flk, nfs4_op_hint_t op_hint,
13807 13807 nfs4_recov_state_t *recov_statep, int needrecov,
13808 13808 COMPOUND4args_clnt **argspp, COMPOUND4res_clnt **respp,
13809 13809 clock_t *tick_delayp, short *whencep, int *errorp,
13810 13810 nfs_resop4 *resop, cred_t *cr, bool_t *did_start_fop,
13811 13811 bool_t *skip_get_err)
13812 13812 {
13813 13813 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
13814 13814
13815 13815 if (lock_args) {
13816 13816 nfs4_open_owner_t *oop = *oopp;
13817 13817 nfs4_open_stream_t *osp = *ospp;
13818 13818 nfs4_lock_owner_t *lop = *lopp;
13819 13819 int intr;
13820 13820
13821 13821 /*
13822 13822 * Blocking lock needs to sleep and retry from the request.
13823 13823 *
13824 13824 * Do not block and wait for 'resend' or 'reinstate'
13825 13825 * lock requests, just return the error.
13826 13826 *
13827 13827 * Note: reclaim requests have cmd == F_SETLK, not F_SETLKW.
13828 13828 */
13829 13829 if (cmd == F_SETLKW) {
13830 13830 rnode4_t *rp = VTOR4(vp);
13831 13831 nfs_argop4 *argop = (*argspp)->array;
13832 13832
13833 13833 ASSERT(ctype == NFS4_LCK_CTYPE_NORM);
13834 13834
13835 13835 nfs4_end_fop(VTOMI4(vp), vp, NULL, op_hint,
13836 13836 recov_statep, needrecov);
13837 13837 *did_start_fop = FALSE;
13838 13838 ASSERT((*argspp)->array_len == 2);
13839 13839 if (argop[1].argop == OP_LOCK)
13840 13840 nfs4args_lock_free(&argop[1]);
13841 13841 else if (argop[1].argop == OP_LOCKT)
13842 13842 nfs4args_lockt_free(&argop[1]);
13843 13843 kmem_free(argop, 2 * sizeof (nfs_argop4));
13844 13844 if (*respp)
13845 13845 (void) xdr_free(xdr_COMPOUND4res_clnt,
13846 13846 (caddr_t)*respp);
13847 13847 *argspp = NULL;
13848 13848 *respp = NULL;
13849 13849 nfs4_end_lock_seqid_sync(lop);
13850 13850 lock_owner_rele(lop);
13851 13851 *lopp = NULL;
13852 13852 if (osp != NULL) {
13853 13853 open_stream_rele(osp, rp);
13854 13854 *ospp = NULL;
13855 13855 }
13856 13856 if (oop != NULL) {
13857 13857 nfs4_end_open_seqid_sync(oop);
13858 13858 open_owner_rele(oop);
13859 13859 *oopp = NULL;
13860 13860 }
13861 13861
13862 13862 nfs_rw_exit(&rp->r_lkserlock);
13863 13863
13864 13864 intr = nfs4_block_and_wait(tick_delayp, rp);
13865 13865
13866 13866 if (intr) {
13867 13867 (void) nfs_rw_enter_sig(&rp->r_lkserlock,
13868 13868 RW_WRITER, FALSE);
13869 13869 *errorp = EINTR;
13870 13870 return (FALSE);
13871 13871 }
13872 13872
13873 13873 (void) nfs_rw_enter_sig(&rp->r_lkserlock,
13874 13874 RW_WRITER, FALSE);
13875 13875
13876 13876 /*
13877 13877 * Make sure we are still safe to lock with
13878 13878 * regards to mmapping.
13879 13879 */
13880 13880 if (!nfs4_safelock(vp, flk, cr)) {
13881 13881 *errorp = EAGAIN;
13882 13882 return (FALSE);
13883 13883 }
13884 13884
13885 13885 return (TRUE);
13886 13886 }
13887 13887 if (ctype == NFS4_LCK_CTYPE_NORM)
13888 13888 *errorp = EAGAIN;
13889 13889 *skip_get_err = TRUE;
13890 13890 flk->l_whence = 0;
13891 13891 *whencep = 0;
13892 13892 return (FALSE);
13893 13893 } else if (lockt_args) {
13894 13894 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
13895 13895 "nfs4frlock_results_denied: OP_LOCKT DENIED"));
13896 13896
13897 13897 denied_to_flk(&resop->nfs_resop4_u.oplockt.denied,
13898 13898 flk, lockt_args);
13899 13899
13900 13900 /* according to NLM code */
13901 13901 *errorp = 0;
13902 13902 *whencep = 0;
13903 13903 *skip_get_err = TRUE;
13904 13904 return (FALSE);
13905 13905 }
13906 13906 return (FALSE);
13907 13907 }
13908 13908
13909 13909 /*
13910 13910 * Handles all NFS4 errors besides NFS4_OK and NFS4ERR_DENIED for nfs4frlock.
13911 13911 */
13912 13912 static void
13913 13913 nfs4frlock_results_default(COMPOUND4res_clnt *resp, int *errorp)
13914 13914 {
13915 13915 switch (resp->status) {
13916 13916 case NFS4ERR_ACCESS:
13917 13917 case NFS4ERR_ADMIN_REVOKED:
13918 13918 case NFS4ERR_BADHANDLE:
13919 13919 case NFS4ERR_BAD_RANGE:
13920 13920 case NFS4ERR_BAD_SEQID:
13921 13921 case NFS4ERR_BAD_STATEID:
13922 13922 case NFS4ERR_BADXDR:
13923 13923 case NFS4ERR_DEADLOCK:
13924 13924 case NFS4ERR_DELAY:
13925 13925 case NFS4ERR_EXPIRED:
13926 13926 case NFS4ERR_FHEXPIRED:
13927 13927 case NFS4ERR_GRACE:
13928 13928 case NFS4ERR_INVAL:
13929 13929 case NFS4ERR_ISDIR:
13930 13930 case NFS4ERR_LEASE_MOVED:
13931 13931 case NFS4ERR_LOCK_NOTSUPP:
13932 13932 case NFS4ERR_LOCK_RANGE:
13933 13933 case NFS4ERR_MOVED:
13934 13934 case NFS4ERR_NOFILEHANDLE:
13935 13935 case NFS4ERR_NO_GRACE:
13936 13936 case NFS4ERR_OLD_STATEID:
13937 13937 case NFS4ERR_OPENMODE:
13938 13938 case NFS4ERR_RECLAIM_BAD:
13939 13939 case NFS4ERR_RECLAIM_CONFLICT:
13940 13940 case NFS4ERR_RESOURCE:
13941 13941 case NFS4ERR_SERVERFAULT:
13942 13942 case NFS4ERR_STALE:
13943 13943 case NFS4ERR_STALE_CLIENTID:
13944 13944 case NFS4ERR_STALE_STATEID:
13945 13945 return;
13946 13946 default:
13947 13947 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
13948 13948 "nfs4frlock_results_default: got unrecognizable "
13949 13949 "res.status %d", resp->status));
13950 13950 *errorp = NFS4ERR_INVAL;
13951 13951 }
13952 13952 }
13953 13953
13954 13954 /*
13955 13955 * The lock request was successful, so update the client's state.
13956 13956 */
13957 13957 static void
13958 13958 nfs4frlock_update_state(LOCK4args *lock_args, LOCKU4args *locku_args,
13959 13959 LOCKT4args *lockt_args, nfs_resop4 *resop, nfs4_lock_owner_t *lop,
13960 13960 vnode_t *vp, flock64_t *flk, cred_t *cr,
13961 13961 nfs4_lost_rqst_t *resend_rqstp)
13962 13962 {
13963 13963 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
13964 13964
13965 13965 if (lock_args) {
13966 13966 LOCK4res *lock_res;
13967 13967
13968 13968 lock_res = &resop->nfs_resop4_u.oplock;
13969 13969 /* update the stateid with server's response */
13970 13970
13971 13971 if (lock_args->locker.new_lock_owner == TRUE) {
13972 13972 mutex_enter(&lop->lo_lock);
13973 13973 lop->lo_just_created = NFS4_PERM_CREATED;
13974 13974 mutex_exit(&lop->lo_lock);
13975 13975 }
13976 13976
13977 13977 nfs4_set_lock_stateid(lop, lock_res->LOCK4res_u.lock_stateid);
13978 13978
13979 13979 /*
13980 13980 * If the lock was the result of a resending a lost
13981 13981 * request, we've synched up the stateid and seqid
13982 13982 * with the server, but now the server might be out of sync
13983 13983 * with what the application thinks it has for locks.
13984 13984 * Clean that up here. It's unclear whether we should do
13985 13985 * this even if the filesystem has been forcibly unmounted.
13986 13986 * For most servers, it's probably wasted effort, but
13987 13987 * RFC3530 lets servers require that unlocks exactly match
13988 13988 * the locks that are held.
13989 13989 */
13990 13990 if (resend_rqstp != NULL &&
13991 13991 resend_rqstp->lr_ctype != NFS4_LCK_CTYPE_REINSTATE) {
13992 13992 nfs4_reinstitute_local_lock_state(vp, flk, cr, lop);
13993 13993 } else {
13994 13994 flk->l_whence = 0;
13995 13995 }
13996 13996 } else if (locku_args) {
13997 13997 LOCKU4res *locku_res;
13998 13998
13999 13999 locku_res = &resop->nfs_resop4_u.oplocku;
14000 14000
14001 14001 /* Update the stateid with the server's response */
14002 14002 nfs4_set_lock_stateid(lop, locku_res->lock_stateid);
14003 14003 } else if (lockt_args) {
14004 14004 /* Switch the lock type to express success, see fcntl */
14005 14005 flk->l_type = F_UNLCK;
14006 14006 flk->l_whence = 0;
14007 14007 }
14008 14008 }
14009 14009
14010 14010 /*
14011 14011 * Do final cleanup before exiting nfs4frlock.
14012 14012 * Calls nfs4_end_fop, drops the seqid syncs, and frees up the
14013 14013 * COMPOUND4 args/res for calls that haven't already.
14014 14014 */
14015 14015 static void
14016 14016 nfs4frlock_final_cleanup(nfs4_lock_call_type_t ctype, COMPOUND4args_clnt *argsp,
14017 14017 COMPOUND4res_clnt *resp, vnode_t *vp, nfs4_op_hint_t op_hint,
14018 14018 nfs4_recov_state_t *recov_statep, int needrecov, nfs4_open_owner_t *oop,
14019 14019 nfs4_open_stream_t *osp, nfs4_lock_owner_t *lop, flock64_t *flk,
14020 14020 short whence, u_offset_t offset, struct lm_sysid *ls,
14021 14021 int *errorp, LOCK4args *lock_args, LOCKU4args *locku_args,
14022 14022 bool_t did_start_fop, bool_t skip_get_err,
14023 14023 cred_t *cred_otw, cred_t *cred)
14024 14024 {
14025 14025 mntinfo4_t *mi = VTOMI4(vp);
14026 14026 rnode4_t *rp = VTOR4(vp);
14027 14027 int error = *errorp;
14028 14028 nfs_argop4 *argop;
14029 14029 int do_flush_pages = 0;
14030 14030
14031 14031 ASSERT(nfs_zone() == mi->mi_zone);
14032 14032 /*
14033 14033 * The client recovery code wants the raw status information,
14034 14034 * so don't map the NFS status code to an errno value for
14035 14035 * non-normal call types.
14036 14036 */
14037 14037 if (ctype == NFS4_LCK_CTYPE_NORM) {
14038 14038 if (*errorp == 0 && resp != NULL && skip_get_err == FALSE)
14039 14039 *errorp = geterrno4(resp->status);
14040 14040 if (did_start_fop == TRUE)
14041 14041 nfs4_end_fop(mi, vp, NULL, op_hint, recov_statep,
14042 14042 needrecov);
14043 14043
14044 14044 /*
14045 14045 * We've established a new lock on the server, so invalidate
14046 14046 * the pages associated with the vnode to get the most up to
14047 14047 * date pages from the server after acquiring the lock. We
14048 14048 * want to be sure that the read operation gets the newest data.
14049 14049 * N.B.
14050 14050 * We used to do this in nfs4frlock_results_ok but that doesn't
14051 14051 * work since VOP_PUTPAGE can call nfs4_commit which calls
14052 14052 * nfs4_start_fop. We flush the pages below after calling
14053 14053 * nfs4_end_fop above
14054 14054 * The flush of the page cache must be done after
14055 14055 * nfs4_end_open_seqid_sync() to avoid a 4-way hang.
14056 14056 */
14057 14057 if (!error && resp && resp->status == NFS4_OK)
14058 14058 do_flush_pages = 1;
14059 14059 }
14060 14060 if (argsp) {
14061 14061 ASSERT(argsp->array_len == 2);
14062 14062 argop = argsp->array;
14063 14063 if (argop[1].argop == OP_LOCK)
14064 14064 nfs4args_lock_free(&argop[1]);
14065 14065 else if (argop[1].argop == OP_LOCKT)
14066 14066 nfs4args_lockt_free(&argop[1]);
14067 14067 kmem_free(argop, 2 * sizeof (nfs_argop4));
14068 14068 if (resp)
14069 14069 (void) xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)resp);
14070 14070 }
14071 14071
14072 14072 /* free the reference on the lock owner */
14073 14073 if (lop != NULL) {
14074 14074 nfs4_end_lock_seqid_sync(lop);
14075 14075 lock_owner_rele(lop);
14076 14076 }
14077 14077
14078 14078 /* need to free up the reference on osp for lock args */
14079 14079 if (osp != NULL)
14080 14080 open_stream_rele(osp, rp);
14081 14081
14082 14082 /* need to free up the reference on oop for lock args */
14083 14083 if (oop != NULL) {
14084 14084 nfs4_end_open_seqid_sync(oop);
14085 14085 open_owner_rele(oop);
14086 14086 }
14087 14087
14088 14088 if (do_flush_pages)
14089 14089 nfs4_flush_pages(vp, cred);
14090 14090
14091 14091 (void) convoff(vp, flk, whence, offset);
14092 14092
14093 14093 lm_rel_sysid(ls);
14094 14094
14095 14095 /*
14096 14096 * Record debug information in the event we get EINVAL.
14097 14097 */
14098 14098 mutex_enter(&mi->mi_lock);
14099 14099 if (*errorp == EINVAL && (lock_args || locku_args) &&
14100 14100 (!(mi->mi_flags & MI4_POSIX_LOCK))) {
14101 14101 if (!(mi->mi_flags & MI4_LOCK_DEBUG)) {
14102 14102 zcmn_err(getzoneid(), CE_NOTE,
14103 14103 "%s operation failed with "
14104 14104 "EINVAL probably since the server, %s,"
14105 14105 " doesn't support POSIX style locking",
14106 14106 lock_args ? "LOCK" : "LOCKU",
14107 14107 mi->mi_curr_serv->sv_hostname);
14108 14108 mi->mi_flags |= MI4_LOCK_DEBUG;
14109 14109 }
14110 14110 }
14111 14111 mutex_exit(&mi->mi_lock);
14112 14112
14113 14113 if (cred_otw)
14114 14114 crfree(cred_otw);
14115 14115 }
14116 14116
14117 14117 /*
14118 14118 * This calls the server and the local locking code.
14119 14119 *
14120 14120 * Client locks are registerred locally by oring the sysid with
14121 14121 * LM_SYSID_CLIENT. The server registers locks locally using just the sysid.
14122 14122 * We need to distinguish between the two to avoid collision in case one
14123 14123 * machine is used as both client and server.
14124 14124 *
14125 14125 * Blocking lock requests will continually retry to acquire the lock
14126 14126 * forever.
14127 14127 *
14128 14128 * The ctype is defined as follows:
14129 14129 * NFS4_LCK_CTYPE_NORM: normal lock request.
14130 14130 *
14131 14131 * NFS4_LCK_CTYPE_RECLAIM: bypass the usual calls for synchronizing with client
14132 14132 * recovery, get the pid from flk instead of curproc, and don't reregister
14133 14133 * the lock locally.
14134 14134 *
14135 14135 * NFS4_LCK_CTYPE_RESEND: same as NFS4_LCK_CTYPE_RECLAIM, with the addition
14136 14136 * that we will use the information passed in via resend_rqstp to setup the
14137 14137 * lock/locku request. This resend is the exact same request as the 'lost
14138 14138 * lock', and is initiated by the recovery framework. A successful resend
14139 14139 * request can initiate one or more reinstate requests.
14140 14140 *
14141 14141 * NFS4_LCK_CTYPE_REINSTATE: same as NFS4_LCK_CTYPE_RESEND, except that it
14142 14142 * does not trigger additional reinstate requests. This lock call type is
14143 14143 * set for setting the v4 server's locking state back to match what the
14144 14144 * client's local locking state is in the event of a received 'lost lock'.
14145 14145 *
14146 14146 * Errors are returned via the nfs4_error_t parameter.
14147 14147 */
14148 14148 void
14149 14149 nfs4frlock(nfs4_lock_call_type_t ctype, vnode_t *vp, int cmd, flock64_t *flk,
14150 14150 int flag, u_offset_t offset, cred_t *cr, nfs4_error_t *ep,
14151 14151 nfs4_lost_rqst_t *resend_rqstp, int *did_reclaimp)
14152 14152 {
14153 14153 COMPOUND4args_clnt args, *argsp = NULL;
14154 14154 COMPOUND4res_clnt res, *resp = NULL;
14155 14155 nfs_argop4 *argop;
14156 14156 nfs_resop4 *resop;
14157 14157 rnode4_t *rp;
14158 14158 int doqueue = 1;
14159 14159 clock_t tick_delay; /* delay in clock ticks */
14160 14160 struct lm_sysid *ls;
14161 14161 LOCK4args *lock_args = NULL;
14162 14162 LOCKU4args *locku_args = NULL;
14163 14163 LOCKT4args *lockt_args = NULL;
14164 14164 nfs4_open_owner_t *oop = NULL;
14165 14165 nfs4_open_stream_t *osp = NULL;
14166 14166 nfs4_lock_owner_t *lop = NULL;
14167 14167 bool_t needrecov = FALSE;
14168 14168 nfs4_recov_state_t recov_state;
14169 14169 short whence;
14170 14170 nfs4_op_hint_t op_hint;
14171 14171 nfs4_lost_rqst_t lost_rqst;
14172 14172 bool_t retry = FALSE;
14173 14173 bool_t did_start_fop = FALSE;
14174 14174 bool_t skip_get_err = FALSE;
14175 14175 cred_t *cred_otw = NULL;
14176 14176 bool_t recovonly; /* just queue request */
14177 14177 int frc_no_reclaim = 0;
14178 14178 #ifdef DEBUG
14179 14179 char *name;
14180 14180 #endif
14181 14181
14182 14182 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
14183 14183
14184 14184 #ifdef DEBUG
14185 14185 name = fn_name(VTOSV(vp)->sv_name);
14186 14186 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE, "nfs4frlock: "
14187 14187 "%s: cmd %d, type %d, offset %llu, start %"PRIx64", "
14188 14188 "length %"PRIu64", pid %d, sysid %d, call type %s, "
14189 14189 "resend request %s", name, cmd, flk->l_type, offset, flk->l_start,
14190 14190 flk->l_len, ctype == NFS4_LCK_CTYPE_NORM ? curproc->p_pid :
14191 14191 flk->l_pid, flk->l_sysid, nfs4frlock_get_call_type(ctype),
14192 14192 resend_rqstp ? "TRUE" : "FALSE"));
14193 14193 kmem_free(name, MAXNAMELEN);
14194 14194 #endif
14195 14195
14196 14196 nfs4_error_zinit(ep);
14197 14197 ep->error = nfs4frlock_validate_args(cmd, flk, flag, vp, offset);
14198 14198 if (ep->error)
14199 14199 return;
14200 14200 ep->error = nfs4frlock_get_sysid(&ls, vp, flk);
14201 14201 if (ep->error)
14202 14202 return;
14203 14203 nfs4frlock_pre_setup(&tick_delay, &recov_state, flk, &whence,
14204 14204 vp, cr, &cred_otw);
14205 14205
14206 14206 recov_retry:
14207 14207 nfs4frlock_call_init(&args, &argsp, &argop, &op_hint, flk, cmd,
14208 14208 &retry, &did_start_fop, &resp, &skip_get_err, &lost_rqst);
14209 14209 rp = VTOR4(vp);
14210 14210
14211 14211 ep->error = nfs4frlock_start_call(ctype, vp, op_hint, &recov_state,
14212 14212 &did_start_fop, &recovonly);
14213 14213
14214 14214 if (ep->error)
14215 14215 goto out;
14216 14216
14217 14217 if (recovonly) {
14218 14218 /*
14219 14219 * Leave the request for the recovery system to deal with.
14220 14220 */
14221 14221 ASSERT(ctype == NFS4_LCK_CTYPE_NORM);
14222 14222 ASSERT(cmd != F_GETLK);
14223 14223 ASSERT(flk->l_type == F_UNLCK);
14224 14224
14225 14225 nfs4_error_init(ep, EINTR);
14226 14226 needrecov = TRUE;
14227 14227 lop = find_lock_owner(rp, curproc->p_pid, LOWN_ANY);
14228 14228 if (lop != NULL) {
14229 14229 nfs4frlock_save_lost_rqst(ctype, ep->error, READ_LT,
14230 14230 NULL, NULL, lop, flk, &lost_rqst, cr, vp);
14231 14231 (void) nfs4_start_recovery(ep,
14232 14232 VTOMI4(vp), vp, NULL, NULL,
14233 14233 (lost_rqst.lr_op == OP_LOCK ||
14234 14234 lost_rqst.lr_op == OP_LOCKU) ?
14235 14235 &lost_rqst : NULL, OP_LOCKU, NULL, NULL, NULL);
14236 14236 lock_owner_rele(lop);
14237 14237 lop = NULL;
14238 14238 }
14239 14239 flk->l_pid = curproc->p_pid;
14240 14240 nfs4_register_lock_locally(vp, flk, flag, offset);
14241 14241 goto out;
14242 14242 }
14243 14243
14244 14244 /* putfh directory fh */
14245 14245 argop[0].argop = OP_CPUTFH;
14246 14246 argop[0].nfs_argop4_u.opcputfh.sfh = rp->r_fh;
14247 14247
14248 14248 /*
14249 14249 * Set up the over-the-wire arguments and get references to the
14250 14250 * open owner, etc.
14251 14251 */
14252 14252
14253 14253 if (ctype == NFS4_LCK_CTYPE_RESEND ||
14254 14254 ctype == NFS4_LCK_CTYPE_REINSTATE) {
14255 14255 nfs4frlock_setup_resend_lock_args(resend_rqstp, argsp,
14256 14256 &argop[1], &lop, &oop, &osp, &lock_args, &locku_args);
14257 14257 } else {
14258 14258 bool_t go_otw = TRUE;
14259 14259
14260 14260 ASSERT(resend_rqstp == NULL);
14261 14261
14262 14262 switch (cmd) {
14263 14263 case F_GETLK:
14264 14264 case F_O_GETLK:
14265 14265 nfs4frlock_setup_lockt_args(ctype, &argop[1],
14266 14266 &lockt_args, argsp, flk, rp);
14267 14267 break;
14268 14268 case F_SETLKW:
14269 14269 case F_SETLK:
14270 14270 if (flk->l_type == F_UNLCK)
14271 14271 nfs4frlock_setup_locku_args(ctype,
14272 14272 &argop[1], &locku_args, flk,
14273 14273 &lop, ep, argsp,
14274 14274 vp, flag, offset, cr,
14275 14275 &skip_get_err, &go_otw);
14276 14276 else
14277 14277 nfs4frlock_setup_lock_args(ctype,
14278 14278 &lock_args, &oop, &osp, &lop, &argop[1],
14279 14279 argsp, flk, cmd, vp, cr, ep);
14280 14280
14281 14281 if (ep->error)
14282 14282 goto out;
14283 14283
14284 14284 switch (ep->stat) {
14285 14285 case NFS4_OK:
14286 14286 break;
14287 14287 case NFS4ERR_DELAY:
14288 14288 /* recov thread never gets this error */
14289 14289 ASSERT(resend_rqstp == NULL);
14290 14290 ASSERT(did_start_fop);
14291 14291
14292 14292 nfs4_end_fop(VTOMI4(vp), vp, NULL, op_hint,
14293 14293 &recov_state, TRUE);
14294 14294 did_start_fop = FALSE;
14295 14295 if (argop[1].argop == OP_LOCK)
14296 14296 nfs4args_lock_free(&argop[1]);
14297 14297 else if (argop[1].argop == OP_LOCKT)
14298 14298 nfs4args_lockt_free(&argop[1]);
14299 14299 kmem_free(argop, 2 * sizeof (nfs_argop4));
14300 14300 argsp = NULL;
14301 14301 goto recov_retry;
14302 14302 default:
14303 14303 ep->error = EIO;
14304 14304 goto out;
14305 14305 }
14306 14306 break;
14307 14307 default:
14308 14308 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
14309 14309 "nfs4_frlock: invalid cmd %d", cmd));
14310 14310 ep->error = EINVAL;
14311 14311 goto out;
14312 14312 }
14313 14313
14314 14314 if (!go_otw)
14315 14315 goto out;
14316 14316 }
14317 14317
14318 14318 /* XXX should we use the local reclock as a cache ? */
14319 14319 /*
14320 14320 * Unregister the lock with the local locking code before
14321 14321 * contacting the server. This avoids a potential race where
14322 14322 * another process gets notified that it has been granted a lock
14323 14323 * before we can unregister ourselves locally.
14324 14324 */
14325 14325 if ((cmd == F_SETLK || cmd == F_SETLKW) && flk->l_type == F_UNLCK) {
14326 14326 if (ctype == NFS4_LCK_CTYPE_NORM)
14327 14327 flk->l_pid = ttoproc(curthread)->p_pid;
14328 14328 nfs4_register_lock_locally(vp, flk, flag, offset);
14329 14329 }
14330 14330
14331 14331 /*
14332 14332 * Send the server the lock request. Continually loop with a delay
14333 14333 * if get error NFS4ERR_DENIED (for blocking locks) or NFS4ERR_GRACE.
14334 14334 */
14335 14335 resp = &res;
14336 14336
14337 14337 NFS4_DEBUG((nfs4_client_call_debug || nfs4_client_lock_debug),
14338 14338 (CE_NOTE,
14339 14339 "nfs4frlock: %s call, rp %s", needrecov ? "recov" : "first",
14340 14340 rnode4info(rp)));
14341 14341
14342 14342 if (lock_args && frc_no_reclaim) {
14343 14343 ASSERT(ctype == NFS4_LCK_CTYPE_RECLAIM);
14344 14344 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
14345 14345 "nfs4frlock: frc_no_reclaim: clearing reclaim"));
14346 14346 lock_args->reclaim = FALSE;
14347 14347 if (did_reclaimp)
14348 14348 *did_reclaimp = 0;
14349 14349 }
14350 14350
14351 14351 /*
14352 14352 * Do the OTW call.
14353 14353 */
14354 14354 rfs4call(VTOMI4(vp), argsp, resp, cred_otw, &doqueue, 0, ep);
14355 14355
14356 14356 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
14357 14357 "nfs4frlock: error %d, status %d", ep->error, resp->status));
14358 14358
14359 14359 needrecov = nfs4_needs_recovery(ep, TRUE, vp->v_vfsp);
14360 14360 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
14361 14361 "nfs4frlock: needrecov %d", needrecov));
14362 14362
14363 14363 if (ep->error == 0 && nfs4_need_to_bump_seqid(resp))
14364 14364 nfs4frlock_bump_seqid(lock_args, locku_args, oop, lop,
14365 14365 args.ctag);
14366 14366
14367 14367 /*
14368 14368 * Check if one of these mutually exclusive error cases has
14369 14369 * happened:
14370 14370 * need to swap credentials due to access error
14371 14371 * recovery is needed
14372 14372 * different error (only known case is missing Kerberos ticket)
14373 14373 */
14374 14374
14375 14375 if ((ep->error == EACCES ||
14376 14376 (ep->error == 0 && resp->status == NFS4ERR_ACCESS)) &&
14377 14377 cred_otw != cr) {
14378 14378 nfs4frlock_check_access(vp, op_hint, &recov_state, needrecov,
14379 14379 &did_start_fop, &argsp, &resp, ep->error, &lop, &oop, &osp,
14380 14380 cr, &cred_otw);
14381 14381 goto recov_retry;
14382 14382 }
14383 14383
14384 14384 if (needrecov) {
14385 14385 /*
14386 14386 * LOCKT requests don't need to recover from lost
14387 14387 * requests since they don't create/modify state.
14388 14388 */
14389 14389 if ((ep->error == EINTR ||
14390 14390 NFS4_FRC_UNMT_ERR(ep->error, vp->v_vfsp)) &&
14391 14391 lockt_args)
14392 14392 goto out;
14393 14393 /*
14394 14394 * Do not attempt recovery for requests initiated by
14395 14395 * the recovery framework. Let the framework redrive them.
14396 14396 */
14397 14397 if (ctype != NFS4_LCK_CTYPE_NORM)
14398 14398 goto out;
14399 14399 else {
14400 14400 ASSERT(resend_rqstp == NULL);
14401 14401 }
14402 14402
14403 14403 nfs4frlock_save_lost_rqst(ctype, ep->error,
14404 14404 flk_to_locktype(cmd, flk->l_type),
14405 14405 oop, osp, lop, flk, &lost_rqst, cred_otw, vp);
14406 14406
14407 14407 retry = nfs4frlock_recovery(needrecov, ep, &argsp,
14408 14408 &resp, lock_args, locku_args, &oop, &osp, &lop,
14409 14409 rp, vp, &recov_state, op_hint, &did_start_fop,
14410 14410 cmd != F_GETLK ? &lost_rqst : NULL, flk);
14411 14411
14412 14412 if (retry) {
14413 14413 ASSERT(oop == NULL);
14414 14414 ASSERT(osp == NULL);
14415 14415 ASSERT(lop == NULL);
14416 14416 goto recov_retry;
14417 14417 }
14418 14418 goto out;
14419 14419 }
14420 14420
14421 14421 /*
14422 14422 * Bail out if have reached this point with ep->error set. Can
14423 14423 * happen if (ep->error == EACCES && !needrecov && cred_otw == cr).
14424 14424 * This happens if Kerberos ticket has expired or has been
14425 14425 * destroyed.
14426 14426 */
14427 14427 if (ep->error != 0)
14428 14428 goto out;
14429 14429
14430 14430 /*
14431 14431 * Process the reply.
14432 14432 */
14433 14433 switch (resp->status) {
14434 14434 case NFS4_OK:
14435 14435 resop = &resp->array[1];
14436 14436 nfs4frlock_results_ok(ctype, cmd, flk, vp, flag, offset,
14437 14437 resend_rqstp);
14438 14438 /*
14439 14439 * Have a successful lock operation, now update state.
14440 14440 */
14441 14441 nfs4frlock_update_state(lock_args, locku_args, lockt_args,
14442 14442 resop, lop, vp, flk, cr, resend_rqstp);
14443 14443 break;
14444 14444
14445 14445 case NFS4ERR_DENIED:
14446 14446 resop = &resp->array[1];
14447 14447 retry = nfs4frlock_results_denied(ctype, lock_args, lockt_args,
14448 14448 &oop, &osp, &lop, cmd, vp, flk, op_hint,
14449 14449 &recov_state, needrecov, &argsp, &resp,
14450 14450 &tick_delay, &whence, &ep->error, resop, cr,
14451 14451 &did_start_fop, &skip_get_err);
14452 14452
14453 14453 if (retry) {
14454 14454 ASSERT(oop == NULL);
14455 14455 ASSERT(osp == NULL);
14456 14456 ASSERT(lop == NULL);
14457 14457 goto recov_retry;
14458 14458 }
14459 14459 break;
14460 14460 /*
14461 14461 * If the server won't let us reclaim, fall-back to trying to lock
14462 14462 * the file from scratch. Code elsewhere will check the changeinfo
14463 14463 * to ensure the file hasn't been changed.
14464 14464 */
14465 14465 case NFS4ERR_NO_GRACE:
14466 14466 if (lock_args && lock_args->reclaim == TRUE) {
14467 14467 ASSERT(ctype == NFS4_LCK_CTYPE_RECLAIM);
14468 14468 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
14469 14469 "nfs4frlock: reclaim: NFS4ERR_NO_GRACE"));
14470 14470 frc_no_reclaim = 1;
14471 14471 /* clean up before retrying */
14472 14472 needrecov = 0;
14473 14473 (void) nfs4frlock_recovery(needrecov, ep, &argsp, &resp,
14474 14474 lock_args, locku_args, &oop, &osp, &lop, rp, vp,
14475 14475 &recov_state, op_hint, &did_start_fop, NULL, flk);
14476 14476 goto recov_retry;
14477 14477 }
14478 14478 /* FALLTHROUGH */
14479 14479
14480 14480 default:
14481 14481 nfs4frlock_results_default(resp, &ep->error);
14482 14482 break;
14483 14483 }
14484 14484 out:
14485 14485 /*
14486 14486 * Process and cleanup from error. Make interrupted unlock
14487 14487 * requests look successful, since they will be handled by the
14488 14488 * client recovery code.
14489 14489 */
14490 14490 nfs4frlock_final_cleanup(ctype, argsp, resp, vp, op_hint, &recov_state,
14491 14491 needrecov, oop, osp, lop, flk, whence, offset, ls, &ep->error,
14492 14492 lock_args, locku_args, did_start_fop,
14493 14493 skip_get_err, cred_otw, cr);
14494 14494
14495 14495 if (ep->error == EINTR && flk->l_type == F_UNLCK &&
14496 14496 (cmd == F_SETLK || cmd == F_SETLKW))
14497 14497 ep->error = 0;
14498 14498 }
14499 14499
14500 14500 /*
14501 14501 * nfs4_safelock:
14502 14502 *
14503 14503 * Return non-zero if the given lock request can be handled without
14504 14504 * violating the constraints on concurrent mapping and locking.
14505 14505 */
14506 14506
14507 14507 static int
14508 14508 nfs4_safelock(vnode_t *vp, const struct flock64 *bfp, cred_t *cr)
14509 14509 {
14510 14510 rnode4_t *rp = VTOR4(vp);
14511 14511 struct vattr va;
14512 14512 int error;
14513 14513
14514 14514 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
14515 14515 ASSERT(rp->r_mapcnt >= 0);
14516 14516 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE, "nfs4_safelock %s: "
14517 14517 "(%"PRIx64", %"PRIx64"); mapcnt = %ld", bfp->l_type == F_WRLCK ?
14518 14518 "write" : bfp->l_type == F_RDLCK ? "read" : "unlock",
14519 14519 bfp->l_start, bfp->l_len, rp->r_mapcnt));
14520 14520
14521 14521 if (rp->r_mapcnt == 0)
14522 14522 return (1); /* always safe if not mapped */
14523 14523
14524 14524 /*
14525 14525 * If the file is already mapped and there are locks, then they
14526 14526 * should be all safe locks. So adding or removing a lock is safe
14527 14527 * as long as the new request is safe (i.e., whole-file, meaning
14528 14528 * length and starting offset are both zero).
14529 14529 */
14530 14530
14531 14531 if (bfp->l_start != 0 || bfp->l_len != 0) {
14532 14532 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE, "nfs4_safelock: "
14533 14533 "cannot lock a memory mapped file unless locking the "
14534 14534 "entire file: start %"PRIx64", len %"PRIx64,
14535 14535 bfp->l_start, bfp->l_len));
14536 14536 return (0);
14537 14537 }
14538 14538
14539 14539 /* mandatory locking and mapping don't mix */
14540 14540 va.va_mask = AT_MODE;
14541 14541 error = VOP_GETATTR(vp, &va, 0, cr, NULL);
14542 14542 if (error != 0) {
14543 14543 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE, "nfs4_safelock: "
14544 14544 "getattr error %d", error));
14545 14545 return (0); /* treat errors conservatively */
14546 14546 }
14547 14547 if (MANDLOCK(vp, va.va_mode)) {
14548 14548 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE, "nfs4_safelock: "
14549 14549 "cannot mandatory lock and mmap a file"));
14550 14550 return (0);
14551 14551 }
14552 14552
14553 14553 return (1);
14554 14554 }
14555 14555
14556 14556
14557 14557 /*
14558 14558 * Register the lock locally within Solaris.
14559 14559 * As the client, we "or" the sysid with LM_SYSID_CLIENT when
14560 14560 * recording locks locally.
14561 14561 *
14562 14562 * This should handle conflicts/cooperation with NFS v2/v3 since all locks
14563 14563 * are registered locally.
14564 14564 */
14565 14565 void
14566 14566 nfs4_register_lock_locally(vnode_t *vp, struct flock64 *flk, int flag,
14567 14567 u_offset_t offset)
14568 14568 {
14569 14569 int oldsysid;
14570 14570 int error;
14571 14571 #ifdef DEBUG
14572 14572 char *name;
14573 14573 #endif
14574 14574
14575 14575 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
14576 14576
14577 14577 #ifdef DEBUG
14578 14578 name = fn_name(VTOSV(vp)->sv_name);
14579 14579 NFS4_DEBUG(nfs4_client_lock_debug,
14580 14580 (CE_NOTE, "nfs4_register_lock_locally: %s: type %d, "
14581 14581 "start %"PRIx64", length %"PRIx64", pid %ld, sysid %d",
14582 14582 name, flk->l_type, flk->l_start, flk->l_len, (long)flk->l_pid,
14583 14583 flk->l_sysid));
14584 14584 kmem_free(name, MAXNAMELEN);
14585 14585 #endif
14586 14586
14587 14587 /* register the lock with local locking */
14588 14588 oldsysid = flk->l_sysid;
14589 14589 flk->l_sysid |= LM_SYSID_CLIENT;
14590 14590 error = reclock(vp, flk, SETFLCK, flag, offset, NULL);
14591 14591 #ifdef DEBUG
14592 14592 if (error != 0) {
14593 14593 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
14594 14594 "nfs4_register_lock_locally: could not register with"
14595 14595 " local locking"));
14596 14596 NFS4_DEBUG(nfs4_client_lock_debug, (CE_CONT,
14597 14597 "error %d, vp 0x%p, pid %d, sysid 0x%x",
14598 14598 error, (void *)vp, flk->l_pid, flk->l_sysid));
14599 14599 NFS4_DEBUG(nfs4_client_lock_debug, (CE_CONT,
14600 14600 "type %d off 0x%" PRIx64 " len 0x%" PRIx64,
14601 14601 flk->l_type, flk->l_start, flk->l_len));
14602 14602 (void) reclock(vp, flk, 0, flag, offset, NULL);
14603 14603 NFS4_DEBUG(nfs4_client_lock_debug, (CE_CONT,
14604 14604 "blocked by pid %d sysid 0x%x type %d "
14605 14605 "off 0x%" PRIx64 " len 0x%" PRIx64,
14606 14606 flk->l_pid, flk->l_sysid, flk->l_type, flk->l_start,
14607 14607 flk->l_len));
14608 14608 }
14609 14609 #endif
14610 14610 flk->l_sysid = oldsysid;
14611 14611 }
14612 14612
14613 14613 /*
14614 14614 * nfs4_lockrelease:
14615 14615 *
14616 14616 * Release any locks on the given vnode that are held by the current
14617 14617 * process. Also removes the lock owner (if one exists) from the rnode's
14618 14618 * list.
14619 14619 */
14620 14620 static int
14621 14621 nfs4_lockrelease(vnode_t *vp, int flag, offset_t offset, cred_t *cr)
14622 14622 {
14623 14623 flock64_t ld;
14624 14624 int ret, error;
14625 14625 rnode4_t *rp;
14626 14626 nfs4_lock_owner_t *lop;
14627 14627 nfs4_recov_state_t recov_state;
14628 14628 mntinfo4_t *mi;
14629 14629 bool_t possible_orphan = FALSE;
14630 14630 bool_t recovonly;
14631 14631
14632 14632 ASSERT((uintptr_t)vp > KERNELBASE);
14633 14633 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
14634 14634
14635 14635 rp = VTOR4(vp);
14636 14636 mi = VTOMI4(vp);
14637 14637
14638 14638 /*
14639 14639 * If we have not locked anything then we can
14640 14640 * just return since we have no work to do.
14641 14641 */
14642 14642 if (rp->r_lo_head.lo_next_rnode == &rp->r_lo_head) {
14643 14643 return (0);
14644 14644 }
14645 14645
14646 14646 /*
14647 14647 * We need to comprehend that another thread may
14648 14648 * kick off recovery and the lock_owner we have stashed
14649 14649 * in lop might be invalid so we should NOT cache it
14650 14650 * locally!
14651 14651 */
14652 14652 recov_state.rs_flags = 0;
14653 14653 recov_state.rs_num_retry_despite_err = 0;
14654 14654 error = nfs4_start_fop(mi, vp, NULL, OH_LOCKU, &recov_state,
14655 14655 &recovonly);
14656 14656 if (error) {
14657 14657 mutex_enter(&rp->r_statelock);
14658 14658 rp->r_flags |= R4LODANGLERS;
14659 14659 mutex_exit(&rp->r_statelock);
14660 14660 return (error);
14661 14661 }
14662 14662
14663 14663 lop = find_lock_owner(rp, curproc->p_pid, LOWN_ANY);
14664 14664
14665 14665 /*
14666 14666 * Check if the lock owner might have a lock (request was sent but
14667 14667 * no response was received). Also check if there are any remote
14668 14668 * locks on the file. (In theory we shouldn't have to make this
14669 14669 * second check if there's no lock owner, but for now we'll be
14670 14670 * conservative and do it anyway.) If either condition is true,
14671 14671 * send an unlock for the entire file to the server.
14672 14672 *
14673 14673 * Note that no explicit synchronization is needed here. At worst,
14674 14674 * flk_has_remote_locks() will return a false positive, in which case
14675 14675 * the unlock call wastes time but doesn't harm correctness.
14676 14676 */
14677 14677
14678 14678 if (lop) {
14679 14679 mutex_enter(&lop->lo_lock);
14680 14680 possible_orphan = lop->lo_pending_rqsts;
14681 14681 mutex_exit(&lop->lo_lock);
14682 14682 lock_owner_rele(lop);
14683 14683 }
14684 14684
14685 14685 nfs4_end_fop(mi, vp, NULL, OH_LOCKU, &recov_state, 0);
14686 14686
14687 14687 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
14688 14688 "nfs4_lockrelease: possible orphan %d, remote locks %d, for "
14689 14689 "lop %p.", possible_orphan, flk_has_remote_locks(vp),
14690 14690 (void *)lop));
14691 14691
14692 14692 if (possible_orphan || flk_has_remote_locks(vp)) {
14693 14693 ld.l_type = F_UNLCK; /* set to unlock entire file */
14694 14694 ld.l_whence = 0; /* unlock from start of file */
14695 14695 ld.l_start = 0;
14696 14696 ld.l_len = 0; /* do entire file */
14697 14697
14698 14698 ret = VOP_FRLOCK(vp, F_SETLK, &ld, flag, offset, NULL,
14699 14699 cr, NULL);
14700 14700
14701 14701 if (ret != 0) {
14702 14702 /*
14703 14703 * If VOP_FRLOCK fails, make sure we unregister
14704 14704 * local locks before we continue.
14705 14705 */
14706 14706 ld.l_pid = ttoproc(curthread)->p_pid;
14707 14707 nfs4_register_lock_locally(vp, &ld, flag, offset);
14708 14708 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
14709 14709 "nfs4_lockrelease: lock release error on vp"
14710 14710 " %p: error %d.\n", (void *)vp, ret));
14711 14711 }
14712 14712 }
14713 14713
14714 14714 recov_state.rs_flags = 0;
14715 14715 recov_state.rs_num_retry_despite_err = 0;
14716 14716 error = nfs4_start_fop(mi, vp, NULL, OH_LOCKU, &recov_state,
14717 14717 &recovonly);
14718 14718 if (error) {
14719 14719 mutex_enter(&rp->r_statelock);
14720 14720 rp->r_flags |= R4LODANGLERS;
14721 14721 mutex_exit(&rp->r_statelock);
14722 14722 return (error);
14723 14723 }
14724 14724
14725 14725 /*
14726 14726 * So, here we're going to need to retrieve the lock-owner
14727 14727 * again (in case recovery has done a switch-a-roo) and
14728 14728 * remove it because we can.
14729 14729 */
14730 14730 lop = find_lock_owner(rp, curproc->p_pid, LOWN_ANY);
14731 14731
14732 14732 if (lop) {
14733 14733 nfs4_rnode_remove_lock_owner(rp, lop);
14734 14734 lock_owner_rele(lop);
14735 14735 }
14736 14736
14737 14737 nfs4_end_fop(mi, vp, NULL, OH_LOCKU, &recov_state, 0);
14738 14738 return (0);
14739 14739 }
14740 14740
14741 14741 /*
14742 14742 * Wait for 'tick_delay' clock ticks.
14743 14743 * Implement exponential backoff until hit the lease_time of this nfs4_server.
14744 14744 * NOTE: lock_lease_time is in seconds.
14745 14745 *
14746 14746 * XXX For future improvements, should implement a waiting queue scheme.
14747 14747 */
14748 14748 static int
14749 14749 nfs4_block_and_wait(clock_t *tick_delay, rnode4_t *rp)
14750 14750 {
14751 14751 long milliseconds_delay;
14752 14752 time_t lock_lease_time;
14753 14753
14754 14754 /* wait tick_delay clock ticks or siginteruptus */
14755 14755 if (delay_sig(*tick_delay)) {
14756 14756 return (EINTR);
14757 14757 }
14758 14758 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE, "nfs4_block_and_wait: "
14759 14759 "reissue the lock request: blocked for %ld clock ticks: %ld "
14760 14760 "milliseconds", *tick_delay, drv_hztousec(*tick_delay) / 1000));
14761 14761
14762 14762 /* get the lease time */
14763 14763 lock_lease_time = r2lease_time(rp);
14764 14764
14765 14765 /* drv_hztousec converts ticks to microseconds */
14766 14766 milliseconds_delay = drv_hztousec(*tick_delay) / 1000;
14767 14767 if (milliseconds_delay < lock_lease_time * 1000) {
14768 14768 *tick_delay = 2 * *tick_delay;
14769 14769 if (drv_hztousec(*tick_delay) > lock_lease_time * 1000 * 1000)
14770 14770 *tick_delay = drv_usectohz(lock_lease_time*1000*1000);
14771 14771 }
14772 14772 return (0);
14773 14773 }
14774 14774
14775 14775
14776 14776 void
14777 14777 nfs4_vnops_init(void)
14778 14778 {
14779 14779 }
14780 14780
14781 14781 void
14782 14782 nfs4_vnops_fini(void)
14783 14783 {
14784 14784 }
14785 14785
14786 14786 /*
14787 14787 * Return a reference to the directory (parent) vnode for a given vnode,
14788 14788 * using the saved pathname information and the directory file handle. The
14789 14789 * caller is responsible for disposing of the reference.
14790 14790 * Returns zero or an errno value.
14791 14791 *
14792 14792 * Caller should set need_start_op to FALSE if it is the recovery
14793 14793 * thread, or if a start_fop has already been done. Otherwise, TRUE.
14794 14794 */
14795 14795 int
14796 14796 vtodv(vnode_t *vp, vnode_t **dvpp, cred_t *cr, bool_t need_start_op)
14797 14797 {
14798 14798 svnode_t *svnp;
14799 14799 vnode_t *dvp = NULL;
14800 14800 servinfo4_t *svp;
14801 14801 nfs4_fname_t *mfname;
14802 14802 int error;
14803 14803
14804 14804 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
14805 14805
14806 14806 if (vp->v_flag & VROOT) {
14807 14807 nfs4_sharedfh_t *sfh;
14808 14808 nfs_fh4 fh;
14809 14809 mntinfo4_t *mi;
14810 14810
14811 14811 ASSERT(vp->v_type == VREG);
14812 14812
14813 14813 mi = VTOMI4(vp);
14814 14814 svp = mi->mi_curr_serv;
14815 14815 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
14816 14816 fh.nfs_fh4_len = svp->sv_pfhandle.fh_len;
14817 14817 fh.nfs_fh4_val = svp->sv_pfhandle.fh_buf;
14818 14818 sfh = sfh4_get(&fh, VTOMI4(vp));
14819 14819 nfs_rw_exit(&svp->sv_lock);
14820 14820 mfname = mi->mi_fname;
14821 14821 fn_hold(mfname);
14822 14822 dvp = makenfs4node_by_fh(sfh, NULL, &mfname, NULL, mi, cr, 0);
14823 14823 sfh4_rele(&sfh);
14824 14824
14825 14825 if (dvp->v_type == VNON)
14826 14826 dvp->v_type = VDIR;
14827 14827 *dvpp = dvp;
14828 14828 return (0);
14829 14829 }
14830 14830
14831 14831 svnp = VTOSV(vp);
14832 14832
14833 14833 if (svnp == NULL) {
14834 14834 NFS4_DEBUG(nfs4_client_shadow_debug, (CE_NOTE, "vtodv: "
14835 14835 "shadow node is NULL"));
14836 14836 return (EINVAL);
14837 14837 }
14838 14838
14839 14839 if (svnp->sv_name == NULL || svnp->sv_dfh == NULL) {
14840 14840 NFS4_DEBUG(nfs4_client_shadow_debug, (CE_NOTE, "vtodv: "
14841 14841 "shadow node name or dfh val == NULL"));
14842 14842 return (EINVAL);
14843 14843 }
14844 14844
14845 14845 error = nfs4_make_dotdot(svnp->sv_dfh, 0, vp, cr, &dvp,
14846 14846 (int)need_start_op);
14847 14847 if (error != 0) {
14848 14848 NFS4_DEBUG(nfs4_client_shadow_debug, (CE_NOTE, "vtodv: "
14849 14849 "nfs4_make_dotdot returned %d", error));
14850 14850 return (error);
14851 14851 }
14852 14852 if (!dvp) {
14853 14853 NFS4_DEBUG(nfs4_client_shadow_debug, (CE_NOTE, "vtodv: "
14854 14854 "nfs4_make_dotdot returned a NULL dvp"));
14855 14855 return (EIO);
14856 14856 }
14857 14857 if (dvp->v_type == VNON)
14858 14858 dvp->v_type = VDIR;
14859 14859 ASSERT(dvp->v_type == VDIR);
14860 14860 if (VTOR4(vp)->r_flags & R4ISXATTR) {
14861 14861 mutex_enter(&dvp->v_lock);
14862 14862 dvp->v_flag |= V_XATTRDIR;
14863 14863 mutex_exit(&dvp->v_lock);
14864 14864 }
14865 14865 *dvpp = dvp;
14866 14866 return (0);
14867 14867 }
14868 14868
14869 14869 /*
14870 14870 * Copy the (final) component name of vp to fnamep. maxlen is the maximum
14871 14871 * length that fnamep can accept, including the trailing null.
14872 14872 * Returns 0 if okay, returns an errno value if there was a problem.
14873 14873 */
14874 14874
14875 14875 int
14876 14876 vtoname(vnode_t *vp, char *fnamep, ssize_t maxlen)
14877 14877 {
14878 14878 char *fn;
14879 14879 int err = 0;
14880 14880 servinfo4_t *svp;
14881 14881 svnode_t *shvp;
14882 14882
14883 14883 /*
14884 14884 * If the file being opened has VROOT set, then this is
14885 14885 * a "file" mount. sv_name will not be interesting, so
14886 14886 * go back to the servinfo4 to get the original mount
14887 14887 * path and strip off all but the final edge. Otherwise
14888 14888 * just return the name from the shadow vnode.
14889 14889 */
14890 14890
14891 14891 if (vp->v_flag & VROOT) {
14892 14892
14893 14893 svp = VTOMI4(vp)->mi_curr_serv;
14894 14894 (void) nfs_rw_enter_sig(&svp->sv_lock, RW_READER, 0);
14895 14895
14896 14896 fn = strrchr(svp->sv_path, '/');
14897 14897 if (fn == NULL)
14898 14898 err = EINVAL;
14899 14899 else
14900 14900 fn++;
14901 14901 } else {
14902 14902 shvp = VTOSV(vp);
14903 14903 fn = fn_name(shvp->sv_name);
14904 14904 }
14905 14905
14906 14906 if (err == 0)
14907 14907 if (strlen(fn) < maxlen)
14908 14908 (void) strcpy(fnamep, fn);
14909 14909 else
14910 14910 err = ENAMETOOLONG;
14911 14911
14912 14912 if (vp->v_flag & VROOT)
14913 14913 nfs_rw_exit(&svp->sv_lock);
14914 14914 else
14915 14915 kmem_free(fn, MAXNAMELEN);
14916 14916
14917 14917 return (err);
14918 14918 }
14919 14919
14920 14920 /*
14921 14921 * Bookkeeping for a close that doesn't need to go over the wire.
14922 14922 * *have_lockp is set to 0 if 'os_sync_lock' is released; otherwise
14923 14923 * it is left at 1.
14924 14924 */
14925 14925 void
14926 14926 nfs4close_notw(vnode_t *vp, nfs4_open_stream_t *osp, int *have_lockp)
14927 14927 {
14928 14928 rnode4_t *rp;
14929 14929 mntinfo4_t *mi;
14930 14930
14931 14931 mi = VTOMI4(vp);
14932 14932 rp = VTOR4(vp);
14933 14933
14934 14934 NFS4_DEBUG(nfs4close_notw_debug, (CE_NOTE, "nfs4close_notw: "
14935 14935 "rp=%p osp=%p", (void *)rp, (void *)osp));
14936 14936 ASSERT(nfs_zone() == mi->mi_zone);
14937 14937 ASSERT(mutex_owned(&osp->os_sync_lock));
14938 14938 ASSERT(*have_lockp);
14939 14939
14940 14940 if (!osp->os_valid ||
14941 14941 osp->os_open_ref_count > 0 || osp->os_mapcnt > 0) {
14942 14942 return;
14943 14943 }
14944 14944
14945 14945 /*
14946 14946 * This removes the reference obtained at OPEN; ie,
14947 14947 * when the open stream structure was created.
14948 14948 *
14949 14949 * We don't have to worry about calling 'open_stream_rele'
14950 14950 * since we our currently holding a reference to this
14951 14951 * open stream which means the count can not go to 0 with
14952 14952 * this decrement.
14953 14953 */
14954 14954 ASSERT(osp->os_ref_count >= 2);
14955 14955 osp->os_ref_count--;
14956 14956 osp->os_valid = 0;
14957 14957 mutex_exit(&osp->os_sync_lock);
14958 14958 *have_lockp = 0;
14959 14959
14960 14960 nfs4_dec_state_ref_count(mi);
14961 14961 }
14962 14962
14963 14963 /*
14964 14964 * Close all remaining open streams on the rnode. These open streams
14965 14965 * could be here because:
14966 14966 * - The close attempted at either close or delmap failed
14967 14967 * - Some kernel entity did VOP_OPEN but never did VOP_CLOSE
14968 14968 * - Someone did mknod on a regular file but never opened it
14969 14969 */
14970 14970 int
14971 14971 nfs4close_all(vnode_t *vp, cred_t *cr)
14972 14972 {
14973 14973 nfs4_open_stream_t *osp;
14974 14974 int error;
14975 14975 nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
14976 14976 rnode4_t *rp;
14977 14977
14978 14978 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
14979 14979
14980 14980 error = 0;
14981 14981 rp = VTOR4(vp);
14982 14982
14983 14983 /*
14984 14984 * At this point, all we know is that the last time
14985 14985 * someone called vn_rele, the count was 1. Since then,
14986 14986 * the vnode could have been re-activated. We want to
14987 14987 * loop through the open streams and close each one, but
14988 14988 * we have to be careful since once we release the rnode
14989 14989 * hash bucket lock, someone else is free to come in and
14990 14990 * re-activate the rnode and add new open streams. The
14991 14991 * strategy is take the rnode hash bucket lock, verify that
14992 14992 * the count is still 1, grab the open stream off the
14993 14993 * head of the list and mark it invalid, then release the
14994 14994 * rnode hash bucket lock and proceed with that open stream.
14995 14995 * This is ok because nfs4close_one() will acquire the proper
14996 14996 * open/create to close/destroy synchronization for open
14997 14997 * streams, and will ensure that if someone has reopened
14998 14998 * the open stream after we've dropped the hash bucket lock
14999 14999 * then we'll just simply return without destroying the
15000 15000 * open stream.
15001 15001 * Repeat until the list is empty.
15002 15002 */
15003 15003
15004 15004 for (;;) {
15005 15005
15006 15006 /* make sure vnode hasn't been reactivated */
15007 15007 rw_enter(&rp->r_hashq->r_lock, RW_READER);
15008 15008 mutex_enter(&vp->v_lock);
15009 15009 if (vp->v_count > 1) {
15010 15010 mutex_exit(&vp->v_lock);
15011 15011 rw_exit(&rp->r_hashq->r_lock);
15012 15012 break;
15013 15013 }
15014 15014 /*
15015 15015 * Grabbing r_os_lock before releasing v_lock prevents
15016 15016 * a window where the rnode/open stream could get
15017 15017 * reactivated (and os_force_close set to 0) before we
15018 15018 * had a chance to set os_force_close to 1.
15019 15019 */
15020 15020 mutex_enter(&rp->r_os_lock);
15021 15021 mutex_exit(&vp->v_lock);
15022 15022
15023 15023 osp = list_head(&rp->r_open_streams);
15024 15024 if (!osp) {
15025 15025 /* nothing left to CLOSE OTW, so return */
15026 15026 mutex_exit(&rp->r_os_lock);
15027 15027 rw_exit(&rp->r_hashq->r_lock);
15028 15028 break;
15029 15029 }
15030 15030
15031 15031 mutex_enter(&rp->r_statev4_lock);
15032 15032 /* the file can't still be mem mapped */
15033 15033 ASSERT(rp->r_mapcnt == 0);
15034 15034 if (rp->created_v4)
15035 15035 rp->created_v4 = 0;
15036 15036 mutex_exit(&rp->r_statev4_lock);
15037 15037
15038 15038 /*
15039 15039 * Grab a ref on this open stream; nfs4close_one
15040 15040 * will mark it as invalid
15041 15041 */
15042 15042 mutex_enter(&osp->os_sync_lock);
15043 15043 osp->os_ref_count++;
15044 15044 osp->os_force_close = 1;
15045 15045 mutex_exit(&osp->os_sync_lock);
15046 15046 mutex_exit(&rp->r_os_lock);
15047 15047 rw_exit(&rp->r_hashq->r_lock);
15048 15048
15049 15049 nfs4close_one(vp, osp, cr, 0, NULL, &e, CLOSE_FORCE, 0, 0, 0);
15050 15050
15051 15051 /* Update error if it isn't already non-zero */
15052 15052 if (error == 0) {
15053 15053 if (e.error)
15054 15054 error = e.error;
15055 15055 else if (e.stat)
15056 15056 error = geterrno4(e.stat);
15057 15057 }
15058 15058
15059 15059 #ifdef DEBUG
15060 15060 nfs4close_all_cnt++;
15061 15061 #endif
15062 15062 /* Release the ref on osp acquired above. */
15063 15063 open_stream_rele(osp, rp);
15064 15064
15065 15065 /* Proceed to the next open stream, if any */
15066 15066 }
15067 15067 return (error);
15068 15068 }
15069 15069
15070 15070 /*
15071 15071 * nfs4close_one - close one open stream for a file if needed.
15072 15072 *
15073 15073 * "close_type" indicates which close path this is:
15074 15074 * CLOSE_NORM: close initiated via VOP_CLOSE.
15075 15075 * CLOSE_DELMAP: close initiated via VOP_DELMAP.
15076 15076 * CLOSE_FORCE: close initiated via VOP_INACTIVE. This path forces
15077 15077 * the close and release of client state for this open stream
15078 15078 * (unless someone else has the open stream open).
15079 15079 * CLOSE_RESEND: indicates the request is a replay of an earlier request
15080 15080 * (e.g., due to abort because of a signal).
15081 15081 * CLOSE_AFTER_RESEND: close initiated to "undo" a successful resent OPEN.
15082 15082 *
15083 15083 * CLOSE_RESEND and CLOSE_AFTER_RESEND will not attempt to retry after client
15084 15084 * recovery. Instead, the caller is expected to deal with retries.
15085 15085 *
15086 15086 * The caller can either pass in the osp ('provided_osp') or not.
15087 15087 *
15088 15088 * 'access_bits' represents the access we are closing/downgrading.
15089 15089 *
15090 15090 * 'len', 'prot', and 'mmap_flags' are used for CLOSE_DELMAP. 'len' is the
15091 15091 * number of bytes we are unmapping, 'maxprot' is the mmap protection, and
15092 15092 * 'mmap_flags' tells us the type of sharing (MAP_PRIVATE or MAP_SHARED).
15093 15093 *
15094 15094 * Errors are returned via the nfs4_error_t.
15095 15095 */
15096 15096 void
15097 15097 nfs4close_one(vnode_t *vp, nfs4_open_stream_t *provided_osp, cred_t *cr,
15098 15098 int access_bits, nfs4_lost_rqst_t *lrp, nfs4_error_t *ep,
15099 15099 nfs4_close_type_t close_type, size_t len, uint_t maxprot,
15100 15100 uint_t mmap_flags)
15101 15101 {
15102 15102 nfs4_open_owner_t *oop;
15103 15103 nfs4_open_stream_t *osp = NULL;
15104 15104 int retry = 0;
15105 15105 int num_retries = NFS4_NUM_RECOV_RETRIES;
15106 15106 rnode4_t *rp;
15107 15107 mntinfo4_t *mi;
15108 15108 nfs4_recov_state_t recov_state;
15109 15109 cred_t *cred_otw = NULL;
15110 15110 bool_t recovonly = FALSE;
15111 15111 int isrecov;
15112 15112 int force_close;
15113 15113 int close_failed = 0;
15114 15114 int did_dec_count = 0;
15115 15115 int did_start_op = 0;
15116 15116 int did_force_recovlock = 0;
15117 15117 int did_start_seqid_sync = 0;
15118 15118 int have_sync_lock = 0;
15119 15119
15120 15120 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
15121 15121
15122 15122 NFS4_DEBUG(nfs4close_one_debug, (CE_NOTE, "closing vp %p osp %p, "
15123 15123 "lrp %p, close type %d len %ld prot %x mmap flags %x bits %x",
15124 15124 (void *)vp, (void *)provided_osp, (void *)lrp, close_type,
15125 15125 len, maxprot, mmap_flags, access_bits));
15126 15126
15127 15127 nfs4_error_zinit(ep);
15128 15128 rp = VTOR4(vp);
15129 15129 mi = VTOMI4(vp);
15130 15130 isrecov = (close_type == CLOSE_RESEND ||
15131 15131 close_type == CLOSE_AFTER_RESEND);
15132 15132
15133 15133 /*
15134 15134 * First get the open owner.
15135 15135 */
15136 15136 if (!provided_osp) {
15137 15137 oop = find_open_owner(cr, NFS4_PERM_CREATED, mi);
15138 15138 } else {
15139 15139 oop = provided_osp->os_open_owner;
15140 15140 ASSERT(oop != NULL);
15141 15141 open_owner_hold(oop);
15142 15142 }
15143 15143
15144 15144 if (!oop) {
15145 15145 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
15146 15146 "nfs4close_one: no oop, rp %p, mi %p, cr %p, osp %p, "
15147 15147 "close type %d", (void *)rp, (void *)mi, (void *)cr,
15148 15148 (void *)provided_osp, close_type));
15149 15149 ep->error = EIO;
15150 15150 goto out;
15151 15151 }
15152 15152
15153 15153 cred_otw = nfs4_get_otw_cred(cr, mi, oop);
15154 15154 recov_retry:
15155 15155 osp = NULL;
15156 15156 close_failed = 0;
15157 15157 force_close = (close_type == CLOSE_FORCE);
15158 15158 retry = 0;
15159 15159 did_start_op = 0;
15160 15160 did_force_recovlock = 0;
15161 15161 did_start_seqid_sync = 0;
15162 15162 have_sync_lock = 0;
15163 15163 recovonly = FALSE;
15164 15164 recov_state.rs_flags = 0;
15165 15165 recov_state.rs_num_retry_despite_err = 0;
15166 15166
15167 15167 /*
15168 15168 * Second synchronize with recovery.
15169 15169 */
15170 15170 if (!isrecov) {
15171 15171 ep->error = nfs4_start_fop(mi, vp, NULL, OH_CLOSE,
15172 15172 &recov_state, &recovonly);
15173 15173 if (!ep->error) {
15174 15174 did_start_op = 1;
15175 15175 } else {
15176 15176 close_failed = 1;
15177 15177 /*
15178 15178 * If we couldn't get start_fop, but have to
15179 15179 * cleanup state, then at least acquire the
15180 15180 * mi_recovlock so we can synchronize with
15181 15181 * recovery.
15182 15182 */
15183 15183 if (close_type == CLOSE_FORCE) {
15184 15184 (void) nfs_rw_enter_sig(&mi->mi_recovlock,
15185 15185 RW_READER, FALSE);
15186 15186 did_force_recovlock = 1;
15187 15187 } else
15188 15188 goto out;
15189 15189 }
15190 15190 }
15191 15191
15192 15192 /*
15193 15193 * We cannot attempt to get the open seqid sync if nfs4_start_fop
15194 15194 * set 'recovonly' to TRUE since most likely this is due to
15195 15195 * reovery being active (MI4_RECOV_ACTIV). If recovery is active,
15196 15196 * nfs4_start_open_seqid_sync() will fail with EAGAIN asking us
15197 15197 * to retry, causing us to loop until recovery finishes. Plus we
15198 15198 * don't need protection over the open seqid since we're not going
15199 15199 * OTW, hence don't need to use the seqid.
15200 15200 */
15201 15201 if (recovonly == FALSE) {
15202 15202 /* need to grab the open owner sync before 'os_sync_lock' */
15203 15203 ep->error = nfs4_start_open_seqid_sync(oop, mi);
15204 15204 if (ep->error == EAGAIN) {
15205 15205 ASSERT(!isrecov);
15206 15206 if (did_start_op)
15207 15207 nfs4_end_fop(mi, vp, NULL, OH_CLOSE,
15208 15208 &recov_state, TRUE);
15209 15209 if (did_force_recovlock)
15210 15210 nfs_rw_exit(&mi->mi_recovlock);
15211 15211 goto recov_retry;
15212 15212 }
15213 15213 did_start_seqid_sync = 1;
15214 15214 }
15215 15215
15216 15216 /*
15217 15217 * Third get an open stream and acquire 'os_sync_lock' to
15218 15218 * sychronize the opening/creating of an open stream with the
15219 15219 * closing/destroying of an open stream.
15220 15220 */
15221 15221 if (!provided_osp) {
15222 15222 /* returns with 'os_sync_lock' held */
15223 15223 osp = find_open_stream(oop, rp);
15224 15224 if (!osp) {
15225 15225 ep->error = EIO;
15226 15226 goto out;
15227 15227 }
15228 15228 } else {
15229 15229 osp = provided_osp;
15230 15230 open_stream_hold(osp);
15231 15231 mutex_enter(&osp->os_sync_lock);
15232 15232 }
15233 15233 have_sync_lock = 1;
15234 15234
15235 15235 ASSERT(oop == osp->os_open_owner);
15236 15236
15237 15237 /*
15238 15238 * Fourth, do any special pre-OTW CLOSE processing
15239 15239 * based on the specific close type.
15240 15240 */
15241 15241 if ((close_type == CLOSE_NORM || close_type == CLOSE_AFTER_RESEND) &&
15242 15242 !did_dec_count) {
15243 15243 ASSERT(osp->os_open_ref_count > 0);
15244 15244 osp->os_open_ref_count--;
15245 15245 did_dec_count = 1;
15246 15246 if (osp->os_open_ref_count == 0)
15247 15247 osp->os_final_close = 1;
15248 15248 }
15249 15249
15250 15250 if (close_type == CLOSE_FORCE) {
15251 15251 /* see if somebody reopened the open stream. */
15252 15252 if (!osp->os_force_close) {
15253 15253 NFS4_DEBUG(nfs4close_one_debug, (CE_NOTE,
15254 15254 "nfs4close_one: skip CLOSE_FORCE as osp %p "
15255 15255 "was reopened, vp %p", (void *)osp, (void *)vp));
15256 15256 ep->error = 0;
15257 15257 ep->stat = NFS4_OK;
15258 15258 goto out;
15259 15259 }
15260 15260
15261 15261 if (!osp->os_final_close && !did_dec_count) {
15262 15262 osp->os_open_ref_count--;
15263 15263 did_dec_count = 1;
15264 15264 }
15265 15265
15266 15266 /*
15267 15267 * We can't depend on os_open_ref_count being 0 due to the
15268 15268 * way executables are opened (VN_RELE to match a VOP_OPEN).
15269 15269 */
15270 15270 #ifdef NOTYET
15271 15271 ASSERT(osp->os_open_ref_count == 0);
15272 15272 #endif
15273 15273 if (osp->os_open_ref_count != 0) {
15274 15274 NFS4_DEBUG(nfs4close_one_debug, (CE_NOTE,
15275 15275 "nfs4close_one: should panic here on an "
15276 15276 "ASSERT(osp->os_open_ref_count == 0). Ignoring "
15277 15277 "since this is probably the exec problem."));
15278 15278
15279 15279 osp->os_open_ref_count = 0;
15280 15280 }
15281 15281
15282 15282 /*
15283 15283 * There is the possibility that nfs4close_one()
15284 15284 * for close_type == CLOSE_DELMAP couldn't find the
15285 15285 * open stream, thus couldn't decrement its os_mapcnt;
15286 15286 * therefore we can't use this ASSERT yet.
15287 15287 */
15288 15288 #ifdef NOTYET
15289 15289 ASSERT(osp->os_mapcnt == 0);
15290 15290 #endif
15291 15291 osp->os_mapcnt = 0;
15292 15292 }
15293 15293
15294 15294 if (close_type == CLOSE_DELMAP && !did_dec_count) {
15295 15295 ASSERT(osp->os_mapcnt >= btopr(len));
15296 15296
15297 15297 if ((mmap_flags & MAP_SHARED) && (maxprot & PROT_WRITE))
15298 15298 osp->os_mmap_write -= btopr(len);
15299 15299 if (maxprot & PROT_READ)
15300 15300 osp->os_mmap_read -= btopr(len);
15301 15301 if (maxprot & PROT_EXEC)
15302 15302 osp->os_mmap_read -= btopr(len);
15303 15303 /* mirror the PROT_NONE check in nfs4_addmap() */
15304 15304 if (!(maxprot & PROT_READ) && !(maxprot & PROT_WRITE) &&
15305 15305 !(maxprot & PROT_EXEC))
15306 15306 osp->os_mmap_read -= btopr(len);
15307 15307 osp->os_mapcnt -= btopr(len);
15308 15308 did_dec_count = 1;
15309 15309 }
15310 15310
15311 15311 if (recovonly) {
15312 15312 nfs4_lost_rqst_t lost_rqst;
15313 15313
15314 15314 /* request should not already be in recovery queue */
15315 15315 ASSERT(lrp == NULL);
15316 15316 nfs4_error_init(ep, EINTR);
15317 15317 nfs4close_save_lost_rqst(ep->error, &lost_rqst, oop,
15318 15318 osp, cred_otw, vp);
15319 15319 mutex_exit(&osp->os_sync_lock);
15320 15320 have_sync_lock = 0;
15321 15321 (void) nfs4_start_recovery(ep, mi, vp, NULL, NULL,
15322 15322 lost_rqst.lr_op == OP_CLOSE ?
15323 15323 &lost_rqst : NULL, OP_CLOSE, NULL, NULL, NULL);
15324 15324 close_failed = 1;
15325 15325 force_close = 0;
15326 15326 goto close_cleanup;
15327 15327 }
15328 15328
15329 15329 /*
15330 15330 * If a previous OTW call got NFS4ERR_BAD_SEQID, then
15331 15331 * we stopped operating on the open owner's <old oo_name, old seqid>
15332 15332 * space, which means we stopped operating on the open stream
15333 15333 * too. So don't go OTW (as the seqid is likely bad, and the
15334 15334 * stateid could be stale, potentially triggering a false
15335 15335 * setclientid), and just clean up the client's internal state.
15336 15336 */
15337 15337 if (osp->os_orig_oo_name != oop->oo_name) {
15338 15338 NFS4_DEBUG(nfs4close_one_debug || nfs4_client_recov_debug,
15339 15339 (CE_NOTE, "nfs4close_one: skip OTW close for osp %p "
15340 15340 "oop %p due to bad seqid (orig oo_name %" PRIx64 " current "
15341 15341 "oo_name %" PRIx64")",
15342 15342 (void *)osp, (void *)oop, osp->os_orig_oo_name,
15343 15343 oop->oo_name));
15344 15344 close_failed = 1;
15345 15345 }
15346 15346
15347 15347 /* If the file failed recovery, just quit. */
15348 15348 mutex_enter(&rp->r_statelock);
15349 15349 if (rp->r_flags & R4RECOVERR) {
15350 15350 close_failed = 1;
15351 15351 }
15352 15352 mutex_exit(&rp->r_statelock);
15353 15353
15354 15354 /*
15355 15355 * If the force close path failed to obtain start_fop
15356 15356 * then skip the OTW close and just remove the state.
15357 15357 */
15358 15358 if (close_failed)
15359 15359 goto close_cleanup;
15360 15360
15361 15361 /*
15362 15362 * Fifth, check to see if there are still mapped pages or other
15363 15363 * opens using this open stream. If there are then we can't
15364 15364 * close yet but we can see if an OPEN_DOWNGRADE is necessary.
15365 15365 */
15366 15366 if (osp->os_open_ref_count > 0 || osp->os_mapcnt > 0) {
15367 15367 nfs4_lost_rqst_t new_lost_rqst;
15368 15368 bool_t needrecov = FALSE;
15369 15369 cred_t *odg_cred_otw = NULL;
15370 15370 seqid4 open_dg_seqid = 0;
15371 15371
15372 15372 if (osp->os_delegation) {
15373 15373 /*
15374 15374 * If this open stream was never OPENed OTW then we
15375 15375 * surely can't DOWNGRADE it (especially since the
15376 15376 * osp->open_stateid is really a delegation stateid
15377 15377 * when os_delegation is 1).
15378 15378 */
15379 15379 if (access_bits & FREAD)
15380 15380 osp->os_share_acc_read--;
15381 15381 if (access_bits & FWRITE)
15382 15382 osp->os_share_acc_write--;
15383 15383 osp->os_share_deny_none--;
15384 15384 nfs4_error_zinit(ep);
15385 15385 goto out;
15386 15386 }
15387 15387 nfs4_open_downgrade(access_bits, 0, oop, osp, vp, cr,
15388 15388 lrp, ep, &odg_cred_otw, &open_dg_seqid);
15389 15389 needrecov = nfs4_needs_recovery(ep, TRUE, mi->mi_vfsp);
15390 15390 if (needrecov && !isrecov) {
15391 15391 bool_t abort;
15392 15392 nfs4_bseqid_entry_t *bsep = NULL;
15393 15393
15394 15394 if (!ep->error && ep->stat == NFS4ERR_BAD_SEQID)
15395 15395 bsep = nfs4_create_bseqid_entry(oop, NULL,
15396 15396 vp, 0,
15397 15397 lrp ? TAG_OPEN_DG_LOST : TAG_OPEN_DG,
15398 15398 open_dg_seqid);
15399 15399
15400 15400 nfs4open_dg_save_lost_rqst(ep->error, &new_lost_rqst,
15401 15401 oop, osp, odg_cred_otw, vp, access_bits, 0);
15402 15402 mutex_exit(&osp->os_sync_lock);
15403 15403 have_sync_lock = 0;
15404 15404 abort = nfs4_start_recovery(ep, mi, vp, NULL, NULL,
15405 15405 new_lost_rqst.lr_op == OP_OPEN_DOWNGRADE ?
15406 15406 &new_lost_rqst : NULL, OP_OPEN_DOWNGRADE,
15407 15407 bsep, NULL, NULL);
15408 15408 if (odg_cred_otw)
15409 15409 crfree(odg_cred_otw);
15410 15410 if (bsep)
15411 15411 kmem_free(bsep, sizeof (*bsep));
15412 15412
15413 15413 if (abort == TRUE)
15414 15414 goto out;
15415 15415
15416 15416 if (did_start_seqid_sync) {
15417 15417 nfs4_end_open_seqid_sync(oop);
15418 15418 did_start_seqid_sync = 0;
15419 15419 }
15420 15420 open_stream_rele(osp, rp);
15421 15421
15422 15422 if (did_start_op)
15423 15423 nfs4_end_fop(mi, vp, NULL, OH_CLOSE,
15424 15424 &recov_state, FALSE);
15425 15425 if (did_force_recovlock)
15426 15426 nfs_rw_exit(&mi->mi_recovlock);
15427 15427
15428 15428 goto recov_retry;
15429 15429 } else {
15430 15430 if (odg_cred_otw)
15431 15431 crfree(odg_cred_otw);
15432 15432 }
15433 15433 goto out;
15434 15434 }
15435 15435
15436 15436 /*
15437 15437 * If this open stream was created as the results of an open
15438 15438 * while holding a delegation, then just release it; no need
15439 15439 * to do an OTW close. Otherwise do a "normal" OTW close.
15440 15440 */
15441 15441 if (osp->os_delegation) {
15442 15442 nfs4close_notw(vp, osp, &have_sync_lock);
15443 15443 nfs4_error_zinit(ep);
15444 15444 goto out;
15445 15445 }
15446 15446
15447 15447 /*
15448 15448 * If this stream is not valid, we're done.
15449 15449 */
15450 15450 if (!osp->os_valid) {
15451 15451 nfs4_error_zinit(ep);
15452 15452 goto out;
15453 15453 }
15454 15454
15455 15455 /*
15456 15456 * Last open or mmap ref has vanished, need to do an OTW close.
15457 15457 * First check to see if a close is still necessary.
15458 15458 */
15459 15459 if (osp->os_failed_reopen) {
15460 15460 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
15461 15461 "don't close OTW osp %p since reopen failed.",
15462 15462 (void *)osp));
15463 15463 /*
15464 15464 * Reopen of the open stream failed, hence the
15465 15465 * stateid of the open stream is invalid/stale, and
15466 15466 * sending this OTW would incorrectly cause another
15467 15467 * round of recovery. In this case, we need to set
15468 15468 * the 'os_valid' bit to 0 so another thread doesn't
15469 15469 * come in and re-open this open stream before
15470 15470 * this "closing" thread cleans up state (decrementing
15471 15471 * the nfs4_server_t's state_ref_count and decrementing
15472 15472 * the os_ref_count).
15473 15473 */
15474 15474 osp->os_valid = 0;
15475 15475 /*
15476 15476 * This removes the reference obtained at OPEN; ie,
15477 15477 * when the open stream structure was created.
15478 15478 *
15479 15479 * We don't have to worry about calling 'open_stream_rele'
15480 15480 * since we our currently holding a reference to this
15481 15481 * open stream which means the count can not go to 0 with
15482 15482 * this decrement.
15483 15483 */
15484 15484 ASSERT(osp->os_ref_count >= 2);
15485 15485 osp->os_ref_count--;
15486 15486 nfs4_error_zinit(ep);
15487 15487 close_failed = 0;
15488 15488 goto close_cleanup;
15489 15489 }
15490 15490
15491 15491 ASSERT(osp->os_ref_count > 1);
15492 15492
15493 15493 /*
15494 15494 * Sixth, try the CLOSE OTW.
15495 15495 */
15496 15496 nfs4close_otw(rp, cred_otw, oop, osp, &retry, &did_start_seqid_sync,
15497 15497 close_type, ep, &have_sync_lock);
15498 15498
15499 15499 if (ep->error == EINTR || NFS4_FRC_UNMT_ERR(ep->error, vp->v_vfsp)) {
15500 15500 /*
15501 15501 * Let the recovery thread be responsible for
15502 15502 * removing the state for CLOSE.
15503 15503 */
15504 15504 close_failed = 1;
15505 15505 force_close = 0;
15506 15506 retry = 0;
15507 15507 }
15508 15508
15509 15509 /* See if we need to retry with a different cred */
15510 15510 if ((ep->error == EACCES ||
15511 15511 (ep->error == 0 && ep->stat == NFS4ERR_ACCESS)) &&
15512 15512 cred_otw != cr) {
15513 15513 crfree(cred_otw);
15514 15514 cred_otw = cr;
15515 15515 crhold(cred_otw);
15516 15516 retry = 1;
15517 15517 }
15518 15518
15519 15519 if (ep->error || ep->stat)
15520 15520 close_failed = 1;
15521 15521
15522 15522 if (retry && !isrecov && num_retries-- > 0) {
15523 15523 if (have_sync_lock) {
15524 15524 mutex_exit(&osp->os_sync_lock);
15525 15525 have_sync_lock = 0;
15526 15526 }
15527 15527 if (did_start_seqid_sync) {
15528 15528 nfs4_end_open_seqid_sync(oop);
15529 15529 did_start_seqid_sync = 0;
15530 15530 }
15531 15531 open_stream_rele(osp, rp);
15532 15532
15533 15533 if (did_start_op)
15534 15534 nfs4_end_fop(mi, vp, NULL, OH_CLOSE,
15535 15535 &recov_state, FALSE);
15536 15536 if (did_force_recovlock)
15537 15537 nfs_rw_exit(&mi->mi_recovlock);
15538 15538 NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
15539 15539 "nfs4close_one: need to retry the close "
15540 15540 "operation"));
15541 15541 goto recov_retry;
15542 15542 }
15543 15543 close_cleanup:
15544 15544 /*
15545 15545 * Seventh and lastly, process our results.
15546 15546 */
15547 15547 if (close_failed && force_close) {
15548 15548 /*
15549 15549 * It's ok to drop and regrab the 'os_sync_lock' since
15550 15550 * nfs4close_notw() will recheck to make sure the
15551 15551 * "close"/removal of state should happen.
15552 15552 */
15553 15553 if (!have_sync_lock) {
15554 15554 mutex_enter(&osp->os_sync_lock);
15555 15555 have_sync_lock = 1;
15556 15556 }
15557 15557 /*
15558 15558 * This is last call, remove the ref on the open
15559 15559 * stream created by open and clean everything up.
15560 15560 */
15561 15561 osp->os_pending_close = 0;
15562 15562 nfs4close_notw(vp, osp, &have_sync_lock);
15563 15563 nfs4_error_zinit(ep);
15564 15564 }
15565 15565
15566 15566 if (!close_failed) {
15567 15567 if (have_sync_lock) {
15568 15568 osp->os_pending_close = 0;
15569 15569 mutex_exit(&osp->os_sync_lock);
15570 15570 have_sync_lock = 0;
15571 15571 } else {
15572 15572 mutex_enter(&osp->os_sync_lock);
15573 15573 osp->os_pending_close = 0;
15574 15574 mutex_exit(&osp->os_sync_lock);
15575 15575 }
15576 15576 if (did_start_op && recov_state.rs_sp != NULL) {
15577 15577 mutex_enter(&recov_state.rs_sp->s_lock);
15578 15578 nfs4_dec_state_ref_count_nolock(recov_state.rs_sp, mi);
15579 15579 mutex_exit(&recov_state.rs_sp->s_lock);
15580 15580 } else {
15581 15581 nfs4_dec_state_ref_count(mi);
15582 15582 }
15583 15583 nfs4_error_zinit(ep);
15584 15584 }
15585 15585
15586 15586 out:
15587 15587 if (have_sync_lock)
15588 15588 mutex_exit(&osp->os_sync_lock);
15589 15589 if (did_start_op)
15590 15590 nfs4_end_fop(mi, vp, NULL, OH_CLOSE, &recov_state,
15591 15591 recovonly ? TRUE : FALSE);
15592 15592 if (did_force_recovlock)
15593 15593 nfs_rw_exit(&mi->mi_recovlock);
15594 15594 if (cred_otw)
15595 15595 crfree(cred_otw);
15596 15596 if (osp)
15597 15597 open_stream_rele(osp, rp);
15598 15598 if (oop) {
15599 15599 if (did_start_seqid_sync)
15600 15600 nfs4_end_open_seqid_sync(oop);
15601 15601 open_owner_rele(oop);
15602 15602 }
15603 15603 }
15604 15604
15605 15605 /*
15606 15606 * Convert information returned by the server in the LOCK4denied
15607 15607 * structure to the form required by fcntl.
15608 15608 */
15609 15609 static void
15610 15610 denied_to_flk(LOCK4denied *lockt_denied, flock64_t *flk, LOCKT4args *lockt_args)
15611 15611 {
15612 15612 nfs4_lo_name_t *lo;
15613 15613
15614 15614 #ifdef DEBUG
15615 15615 if (denied_to_flk_debug) {
15616 15616 lockt_denied_debug = lockt_denied;
15617 15617 debug_enter("lockt_denied");
15618 15618 }
15619 15619 #endif
15620 15620
15621 15621 flk->l_type = lockt_denied->locktype == READ_LT ? F_RDLCK : F_WRLCK;
15622 15622 flk->l_whence = 0; /* aka SEEK_SET */
15623 15623 flk->l_start = lockt_denied->offset;
15624 15624 flk->l_len = lockt_denied->length;
15625 15625
15626 15626 /*
15627 15627 * If the blocking clientid matches our client id, then we can
15628 15628 * interpret the lockowner (since we built it). If not, then
15629 15629 * fabricate a sysid and pid. Note that the l_sysid field
15630 15630 * in *flk already has the local sysid.
15631 15631 */
15632 15632
15633 15633 if (lockt_denied->owner.clientid == lockt_args->owner.clientid) {
15634 15634
15635 15635 if (lockt_denied->owner.owner_len == sizeof (*lo)) {
15636 15636 lo = (nfs4_lo_name_t *)
15637 15637 lockt_denied->owner.owner_val;
15638 15638
15639 15639 flk->l_pid = lo->ln_pid;
15640 15640 } else {
15641 15641 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
15642 15642 "denied_to_flk: bad lock owner length\n"));
15643 15643
15644 15644 flk->l_pid = lo_to_pid(&lockt_denied->owner);
15645 15645 }
15646 15646 } else {
15647 15647 NFS4_DEBUG(nfs4_client_lock_debug, (CE_NOTE,
15648 15648 "denied_to_flk: foreign clientid\n"));
15649 15649
15650 15650 /*
15651 15651 * Construct a new sysid which should be different from
15652 15652 * sysids of other systems.
15653 15653 */
15654 15654
15655 15655 flk->l_sysid++;
15656 15656 flk->l_pid = lo_to_pid(&lockt_denied->owner);
15657 15657 }
15658 15658 }
15659 15659
15660 15660 static pid_t
15661 15661 lo_to_pid(lock_owner4 *lop)
15662 15662 {
15663 15663 pid_t pid = 0;
15664 15664 uchar_t *cp;
15665 15665 int i;
15666 15666
15667 15667 cp = (uchar_t *)&lop->clientid;
15668 15668
15669 15669 for (i = 0; i < sizeof (lop->clientid); i++)
15670 15670 pid += (pid_t)*cp++;
15671 15671
15672 15672 cp = (uchar_t *)lop->owner_val;
15673 15673
15674 15674 for (i = 0; i < lop->owner_len; i++)
15675 15675 pid += (pid_t)*cp++;
15676 15676
15677 15677 return (pid);
15678 15678 }
15679 15679
15680 15680 /*
15681 15681 * Given a lock pointer, returns the length of that lock.
15682 15682 * "end" is the last locked offset the "l_len" covers from
15683 15683 * the start of the lock.
15684 15684 */
15685 15685 static off64_t
15686 15686 lock_to_end(flock64_t *lock)
15687 15687 {
15688 15688 off64_t lock_end;
15689 15689
15690 15690 if (lock->l_len == 0)
15691 15691 lock_end = (off64_t)MAXEND;
15692 15692 else
15693 15693 lock_end = lock->l_start + lock->l_len - 1;
15694 15694
15695 15695 return (lock_end);
15696 15696 }
15697 15697
15698 15698 /*
15699 15699 * Given the end of a lock, it will return you the length "l_len" for that lock.
15700 15700 */
15701 15701 static off64_t
15702 15702 end_to_len(off64_t start, off64_t end)
15703 15703 {
15704 15704 off64_t lock_len;
15705 15705
15706 15706 ASSERT(end >= start);
15707 15707 if (end == MAXEND)
15708 15708 lock_len = 0;
15709 15709 else
15710 15710 lock_len = end - start + 1;
15711 15711
15712 15712 return (lock_len);
15713 15713 }
15714 15714
15715 15715 /*
15716 15716 * On given end for a lock it determines if it is the last locked offset
15717 15717 * or not, if so keeps it as is, else adds one to return the length for
15718 15718 * valid start.
15719 15719 */
15720 15720 static off64_t
15721 15721 start_check(off64_t x)
15722 15722 {
15723 15723 if (x == MAXEND)
15724 15724 return (x);
15725 15725 else
15726 15726 return (x + 1);
15727 15727 }
15728 15728
15729 15729 /*
15730 15730 * See if these two locks overlap, and if so return 1;
15731 15731 * otherwise, return 0.
15732 15732 */
15733 15733 static int
15734 15734 locks_intersect(flock64_t *llfp, flock64_t *curfp)
15735 15735 {
15736 15736 off64_t llfp_end, curfp_end;
15737 15737
15738 15738 llfp_end = lock_to_end(llfp);
15739 15739 curfp_end = lock_to_end(curfp);
15740 15740
15741 15741 if (((llfp_end >= curfp->l_start) &&
15742 15742 (llfp->l_start <= curfp->l_start)) ||
15743 15743 ((curfp->l_start <= llfp->l_start) && (curfp_end >= llfp->l_start)))
15744 15744 return (1);
15745 15745 return (0);
15746 15746 }
15747 15747
15748 15748 /*
15749 15749 * Determine what the intersecting lock region is, and add that to the
15750 15750 * 'nl_llpp' locklist in increasing order (by l_start).
15751 15751 */
15752 15752 static void
15753 15753 nfs4_add_lock_range(flock64_t *lost_flp, flock64_t *local_flp,
15754 15754 locklist_t **nl_llpp, vnode_t *vp)
15755 15755 {
15756 15756 locklist_t *intersect_llp, *tmp_fllp, *cur_fllp;
15757 15757 off64_t lost_flp_end, local_flp_end, len, start;
15758 15758
15759 15759 NFS4_DEBUG(nfs4_lost_rqst_debug, (CE_NOTE, "nfs4_add_lock_range:"));
15760 15760
15761 15761 if (!locks_intersect(lost_flp, local_flp))
15762 15762 return;
15763 15763
15764 15764 NFS4_DEBUG(nfs4_lost_rqst_debug, (CE_NOTE, "nfs4_add_lock_range: "
15765 15765 "locks intersect"));
15766 15766
15767 15767 lost_flp_end = lock_to_end(lost_flp);
15768 15768 local_flp_end = lock_to_end(local_flp);
15769 15769
15770 15770 /* Find the starting point of the intersecting region */
15771 15771 if (local_flp->l_start > lost_flp->l_start)
15772 15772 start = local_flp->l_start;
15773 15773 else
15774 15774 start = lost_flp->l_start;
15775 15775
15776 15776 /* Find the lenght of the intersecting region */
15777 15777 if (lost_flp_end < local_flp_end)
15778 15778 len = end_to_len(start, lost_flp_end);
15779 15779 else
15780 15780 len = end_to_len(start, local_flp_end);
15781 15781
15782 15782 /*
15783 15783 * Prepare the flock structure for the intersection found and insert
15784 15784 * it into the new list in increasing l_start order. This list contains
15785 15785 * intersections of locks registered by the client with the local host
15786 15786 * and the lost lock.
15787 15787 * The lock type of this lock is the same as that of the local_flp.
15788 15788 */
15789 15789 intersect_llp = (locklist_t *)kmem_alloc(sizeof (locklist_t), KM_SLEEP);
15790 15790 intersect_llp->ll_flock.l_start = start;
15791 15791 intersect_llp->ll_flock.l_len = len;
15792 15792 intersect_llp->ll_flock.l_type = local_flp->l_type;
15793 15793 intersect_llp->ll_flock.l_pid = local_flp->l_pid;
15794 15794 intersect_llp->ll_flock.l_sysid = local_flp->l_sysid;
15795 15795 intersect_llp->ll_flock.l_whence = 0; /* aka SEEK_SET */
15796 15796 intersect_llp->ll_vp = vp;
15797 15797
15798 15798 tmp_fllp = *nl_llpp;
15799 15799 cur_fllp = NULL;
15800 15800 while (tmp_fllp != NULL && tmp_fllp->ll_flock.l_start <
15801 15801 intersect_llp->ll_flock.l_start) {
15802 15802 cur_fllp = tmp_fllp;
15803 15803 tmp_fllp = tmp_fllp->ll_next;
15804 15804 }
15805 15805 if (cur_fllp == NULL) {
15806 15806 /* first on the list */
15807 15807 intersect_llp->ll_next = *nl_llpp;
15808 15808 *nl_llpp = intersect_llp;
15809 15809 } else {
15810 15810 intersect_llp->ll_next = cur_fllp->ll_next;
15811 15811 cur_fllp->ll_next = intersect_llp;
15812 15812 }
15813 15813
15814 15814 NFS4_DEBUG(nfs4_lost_rqst_debug, (CE_NOTE, "nfs4_add_lock_range: "
15815 15815 "created lock region: start %"PRIx64" end %"PRIx64" : %s\n",
15816 15816 intersect_llp->ll_flock.l_start,
15817 15817 intersect_llp->ll_flock.l_start + intersect_llp->ll_flock.l_len,
15818 15818 intersect_llp->ll_flock.l_type == F_RDLCK ? "READ" : "WRITE"));
15819 15819 }
15820 15820
15821 15821 /*
15822 15822 * Our local locking current state is potentially different than
15823 15823 * what the NFSv4 server thinks we have due to a lost lock that was
15824 15824 * resent and then received. We need to reset our "NFSv4" locking
15825 15825 * state to match the current local locking state for this pid since
15826 15826 * that is what the user/application sees as what the world is.
15827 15827 *
15828 15828 * We cannot afford to drop the open/lock seqid sync since then we can
15829 15829 * get confused about what the current local locking state "is" versus
15830 15830 * "was".
15831 15831 *
15832 15832 * If we are unable to fix up the locks, we send SIGLOST to the affected
15833 15833 * process. This is not done if the filesystem has been forcibly
15834 15834 * unmounted, in case the process has already exited and a new process
15835 15835 * exists with the same pid.
15836 15836 */
15837 15837 static void
15838 15838 nfs4_reinstitute_local_lock_state(vnode_t *vp, flock64_t *lost_flp, cred_t *cr,
15839 15839 nfs4_lock_owner_t *lop)
15840 15840 {
15841 15841 locklist_t *locks, *llp, *ri_llp, *tmp_llp;
15842 15842 mntinfo4_t *mi = VTOMI4(vp);
15843 15843 const int cmd = F_SETLK;
15844 15844 off64_t cur_start, llp_ll_flock_end, lost_flp_end;
15845 15845 flock64_t ul_fl;
15846 15846
15847 15847 NFS4_DEBUG(nfs4_lost_rqst_debug, (CE_NOTE,
15848 15848 "nfs4_reinstitute_local_lock_state"));
15849 15849
15850 15850 /*
15851 15851 * Find active locks for this vp from the local locking code.
15852 15852 * Scan through this list and find out the locks that intersect with
15853 15853 * the lost lock. Once we find the lock that intersects, add the
15854 15854 * intersection area as a new lock to a new list "ri_llp". The lock
15855 15855 * type of the intersection region lock added to ri_llp is the same
15856 15856 * as that found in the active lock list, "list". The intersecting
15857 15857 * region locks are added to ri_llp in increasing l_start order.
15858 15858 */
15859 15859 ASSERT(nfs_zone() == mi->mi_zone);
15860 15860
15861 15861 locks = flk_active_locks_for_vp(vp);
15862 15862 ri_llp = NULL;
15863 15863
15864 15864 for (llp = locks; llp != NULL; llp = llp->ll_next) {
15865 15865 ASSERT(llp->ll_vp == vp);
15866 15866 /*
15867 15867 * Pick locks that belong to this pid/lockowner
15868 15868 */
15869 15869 if (llp->ll_flock.l_pid != lost_flp->l_pid)
15870 15870 continue;
15871 15871
15872 15872 nfs4_add_lock_range(lost_flp, &llp->ll_flock, &ri_llp, vp);
15873 15873 }
15874 15874
15875 15875 /*
15876 15876 * Now we have the list of intersections with the lost lock. These are
15877 15877 * the locks that were/are active before the server replied to the
15878 15878 * last/lost lock. Issue these locks to the server here. Playing these
15879 15879 * locks to the server will re-establish aur current local locking state
15880 15880 * with the v4 server.
15881 15881 * If we get an error, send SIGLOST to the application for that lock.
15882 15882 */
15883 15883
15884 15884 for (llp = ri_llp; llp != NULL; llp = llp->ll_next) {
15885 15885 NFS4_DEBUG(nfs4_lost_rqst_debug, (CE_NOTE,
15886 15886 "nfs4_reinstitute_local_lock_state: need to issue "
15887 15887 "flock: [%"PRIx64" - %"PRIx64"] : %s",
15888 15888 llp->ll_flock.l_start,
15889 15889 llp->ll_flock.l_start + llp->ll_flock.l_len,
15890 15890 llp->ll_flock.l_type == F_RDLCK ? "READ" :
15891 15891 llp->ll_flock.l_type == F_WRLCK ? "WRITE" : "INVALID"));
15892 15892 /*
15893 15893 * No need to relock what we already have
15894 15894 */
15895 15895 if (llp->ll_flock.l_type == lost_flp->l_type)
15896 15896 continue;
15897 15897
15898 15898 push_reinstate(vp, cmd, &llp->ll_flock, cr, lop);
15899 15899 }
15900 15900
15901 15901 /*
15902 15902 * Now keeping the start of the lost lock as our reference parse the
15903 15903 * newly created ri_llp locklist to find the ranges that we have locked
15904 15904 * with the v4 server but not in the current local locking. We need
15905 15905 * to unlock these ranges.
15906 15906 * These ranges can also be reffered to as those ranges, where the lost
15907 15907 * lock does not overlap with the locks in the ri_llp but are locked
15908 15908 * since the server replied to the lost lock.
15909 15909 */
15910 15910 cur_start = lost_flp->l_start;
15911 15911 lost_flp_end = lock_to_end(lost_flp);
15912 15912
15913 15913 ul_fl.l_type = F_UNLCK;
15914 15914 ul_fl.l_whence = 0; /* aka SEEK_SET */
15915 15915 ul_fl.l_sysid = lost_flp->l_sysid;
15916 15916 ul_fl.l_pid = lost_flp->l_pid;
15917 15917
15918 15918 for (llp = ri_llp; llp != NULL; llp = llp->ll_next) {
15919 15919 llp_ll_flock_end = lock_to_end(&llp->ll_flock);
15920 15920
15921 15921 if (llp->ll_flock.l_start <= cur_start) {
15922 15922 cur_start = start_check(llp_ll_flock_end);
15923 15923 continue;
15924 15924 }
15925 15925 NFS4_DEBUG(nfs4_lost_rqst_debug, (CE_NOTE,
15926 15926 "nfs4_reinstitute_local_lock_state: "
15927 15927 "UNLOCK [%"PRIx64" - %"PRIx64"]",
15928 15928 cur_start, llp->ll_flock.l_start));
15929 15929
15930 15930 ul_fl.l_start = cur_start;
15931 15931 ul_fl.l_len = end_to_len(cur_start,
15932 15932 (llp->ll_flock.l_start - 1));
15933 15933
15934 15934 push_reinstate(vp, cmd, &ul_fl, cr, lop);
15935 15935 cur_start = start_check(llp_ll_flock_end);
15936 15936 }
15937 15937
15938 15938 /*
15939 15939 * In the case where the lost lock ends after all intersecting locks,
15940 15940 * unlock the last part of the lost lock range.
15941 15941 */
15942 15942 if (cur_start != start_check(lost_flp_end)) {
15943 15943 NFS4_DEBUG(nfs4_lost_rqst_debug, (CE_NOTE,
15944 15944 "nfs4_reinstitute_local_lock_state: UNLOCK end of the "
15945 15945 "lost lock region [%"PRIx64" - %"PRIx64"]",
15946 15946 cur_start, lost_flp->l_start + lost_flp->l_len));
15947 15947
15948 15948 ul_fl.l_start = cur_start;
15949 15949 /*
15950 15950 * Is it an to-EOF lock? if so unlock till the end
15951 15951 */
15952 15952 if (lost_flp->l_len == 0)
15953 15953 ul_fl.l_len = 0;
15954 15954 else
15955 15955 ul_fl.l_len = start_check(lost_flp_end) - cur_start;
15956 15956
15957 15957 push_reinstate(vp, cmd, &ul_fl, cr, lop);
15958 15958 }
15959 15959
15960 15960 if (locks != NULL)
15961 15961 flk_free_locklist(locks);
15962 15962
15963 15963 /* Free up our newly created locklist */
15964 15964 for (llp = ri_llp; llp != NULL; ) {
15965 15965 tmp_llp = llp->ll_next;
15966 15966 kmem_free(llp, sizeof (locklist_t));
15967 15967 llp = tmp_llp;
15968 15968 }
15969 15969
15970 15970 /*
15971 15971 * Now return back to the original calling nfs4frlock()
15972 15972 * and let us naturally drop our seqid syncs.
15973 15973 */
15974 15974 }
15975 15975
15976 15976 /*
15977 15977 * Create a lost state record for the given lock reinstantiation request
15978 15978 * and push it onto the lost state queue.
15979 15979 */
15980 15980 static void
15981 15981 push_reinstate(vnode_t *vp, int cmd, flock64_t *flk, cred_t *cr,
15982 15982 nfs4_lock_owner_t *lop)
15983 15983 {
15984 15984 nfs4_lost_rqst_t req;
15985 15985 nfs_lock_type4 locktype;
15986 15986 nfs4_error_t e = { EINTR, NFS4_OK, RPC_SUCCESS };
15987 15987
15988 15988 ASSERT(nfs_zone() == VTOMI4(vp)->mi_zone);
15989 15989
15990 15990 locktype = flk_to_locktype(cmd, flk->l_type);
15991 15991 nfs4frlock_save_lost_rqst(NFS4_LCK_CTYPE_REINSTATE, EINTR, locktype,
15992 15992 NULL, NULL, lop, flk, &req, cr, vp);
15993 15993 (void) nfs4_start_recovery(&e, VTOMI4(vp), vp, NULL, NULL,
15994 15994 (req.lr_op == OP_LOCK || req.lr_op == OP_LOCKU) ?
15995 15995 &req : NULL, flk->l_type == F_UNLCK ? OP_LOCKU : OP_LOCK,
15996 15996 NULL, NULL, NULL);
15997 15997 }
|
↓ open down ↓ |
15547 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX