1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
26 /*
27 * Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T.
28 * All Rights Reserved
29 */
30
31 /*
32 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
33 */
34
35 #include <sys/param.h>
36 #include <sys/types.h>
37 #include <sys/systm.h>
38 #include <sys/cred.h>
39 #include <sys/proc.h>
40 #include <sys/user.h>
41 #include <sys/time.h>
42 #include <sys/buf.h>
43 #include <sys/vfs.h>
44 #include <sys/vnode.h>
45 #include <sys/socket.h>
46 #include <sys/uio.h>
47 #include <sys/tiuser.h>
48 #include <sys/swap.h>
49 #include <sys/errno.h>
50 #include <sys/debug.h>
51 #include <sys/kmem.h>
52 #include <sys/kstat.h>
53 #include <sys/cmn_err.h>
54 #include <sys/vtrace.h>
55 #include <sys/session.h>
56 #include <sys/dnlc.h>
57 #include <sys/bitmap.h>
58 #include <sys/acl.h>
59 #include <sys/ddi.h>
60 #include <sys/pathname.h>
61 #include <sys/flock.h>
62 #include <sys/dirent.h>
63 #include <sys/flock.h>
64 #include <sys/callb.h>
65 #include <sys/sdt.h>
66
67 #include <vm/pvn.h>
68
69 #include <rpc/types.h>
70 #include <rpc/xdr.h>
71 #include <rpc/auth.h>
72 #include <rpc/rpcsec_gss.h>
73 #include <rpc/clnt.h>
74
75 #include <nfs/nfs.h>
76 #include <nfs/nfs_clnt.h>
77 #include <nfs/nfs_acl.h>
78
79 #include <nfs/nfs4.h>
80 #include <nfs/rnode4.h>
81 #include <nfs/nfs4_clnt.h>
82
83 /*
84 * The hash queues for the access to active and cached rnodes
85 * are organized as doubly linked lists. A reader/writer lock
86 * for each hash bucket is used to control access and to synchronize
87 * lookups, additions, and deletions from the hash queue.
88 *
89 * The rnode freelist is organized as a doubly linked list with
90 * a head pointer. Additions and deletions are synchronized via
91 * a single mutex.
92 *
93 * In order to add an rnode to the free list, it must be hashed into
94 * a hash queue and the exclusive lock to the hash queue be held.
95 * If an rnode is not hashed into a hash queue, then it is destroyed
96 * because it represents no valuable information that can be reused
97 * about the file. The exclusive lock to the hash queue must be
98 * held in order to prevent a lookup in the hash queue from finding
99 * the rnode and using it and assuming that the rnode is not on the
100 * freelist. The lookup in the hash queue will have the hash queue
101 * locked, either exclusive or shared.
102 *
103 * The vnode reference count for each rnode is not allowed to drop
104 * below 1. This prevents external entities, such as the VM
105 * subsystem, from acquiring references to vnodes already on the
106 * freelist and then trying to place them back on the freelist
107 * when their reference is released. This means that the when an
108 * rnode is looked up in the hash queues, then either the rnode
109 * is removed from the freelist and that reference is transferred to
110 * the new reference or the vnode reference count must be incremented
111 * accordingly. The mutex for the freelist must be held in order to
112 * accurately test to see if the rnode is on the freelist or not.
113 * The hash queue lock might be held shared and it is possible that
114 * two different threads may race to remove the rnode from the
115 * freelist. This race can be resolved by holding the mutex for the
116 * freelist. Please note that the mutex for the freelist does not
117 * need to be held if the rnode is not on the freelist. It can not be
118 * placed on the freelist due to the requirement that the thread
119 * putting the rnode on the freelist must hold the exclusive lock
120 * to the hash queue and the thread doing the lookup in the hash
121 * queue is holding either a shared or exclusive lock to the hash
122 * queue.
123 *
124 * The lock ordering is:
125 *
126 * hash bucket lock -> vnode lock
127 * hash bucket lock -> freelist lock -> r_statelock
128 */
129 r4hashq_t *rtable4;
130
131 static kmutex_t rp4freelist_lock;
132 static rnode4_t *rp4freelist = NULL;
133 static long rnode4_new = 0;
134 int rtable4size;
135 static int rtable4mask;
136 static struct kmem_cache *rnode4_cache;
137 static int rnode4_hashlen = 4;
138
139 static void r4inactive(rnode4_t *, cred_t *);
140 static vnode_t *make_rnode4(nfs4_sharedfh_t *, r4hashq_t *, struct vfs *,
141 struct vnodeops *,
142 int (*)(vnode_t *, page_t *, u_offset_t *, size_t *, int,
143 cred_t *),
144 int *, cred_t *);
145 static void rp4_rmfree(rnode4_t *);
146 int nfs4_free_data_reclaim(rnode4_t *);
147 static int nfs4_active_data_reclaim(rnode4_t *);
148 static int nfs4_free_reclaim(void);
149 static int nfs4_active_reclaim(void);
150 static int nfs4_rnode_reclaim(void);
151 static void nfs4_reclaim(void *);
152 static int isrootfh(nfs4_sharedfh_t *, rnode4_t *);
153 static void uninit_rnode4(rnode4_t *);
154 static void destroy_rnode4(rnode4_t *);
155 static void r4_stub_set(rnode4_t *, nfs4_stub_type_t);
156
157 #ifdef DEBUG
158 static int r4_check_for_dups = 0; /* Flag to enable dup rnode detection. */
159 static int nfs4_rnode_debug = 0;
160 /* if nonzero, kmem_cache_free() rnodes rather than place on freelist */
161 static int nfs4_rnode_nofreelist = 0;
162 /* give messages on colliding shared filehandles */
163 static void r4_dup_check(rnode4_t *, vfs_t *);
164 #endif
165
166 /*
167 * If the vnode has pages, run the list and check for any that are
168 * still dangling. We call this routine before putting an rnode on
169 * the free list.
170 */
171 static int
172 nfs4_dross_pages(vnode_t *vp)
173 {
174 page_t *pp;
175 kmutex_t *vphm;
176
177 vphm = page_vnode_mutex(vp);
178 mutex_enter(vphm);
179 if ((pp = vp->v_pages) != NULL) {
180 do {
181 if (pp->p_hash != PVN_VPLIST_HASH_TAG &&
182 pp->p_fsdata != C_NOCOMMIT) {
183 mutex_exit(vphm);
184 return (1);
185 }
186 } while ((pp = pp->p_vpnext) != vp->v_pages);
187 }
188 mutex_exit(vphm);
189
190 return (0);
191 }
192
193 /*
194 * Flush any pages left on this rnode.
195 */
196 static void
197 r4flushpages(rnode4_t *rp, cred_t *cr)
198 {
199 vnode_t *vp;
200 int error;
201
202 /*
203 * Before freeing anything, wait until all asynchronous
204 * activity is done on this rnode. This will allow all
205 * asynchronous read ahead and write behind i/o's to
206 * finish.
207 */
208 mutex_enter(&rp->r_statelock);
209 while (rp->r_count > 0)
210 cv_wait(&rp->r_cv, &rp->r_statelock);
211 mutex_exit(&rp->r_statelock);
212
213 /*
214 * Flush and invalidate all pages associated with the vnode.
215 */
216 vp = RTOV4(rp);
217 if (nfs4_has_pages(vp)) {
218 ASSERT(vp->v_type != VCHR);
219 if ((rp->r_flags & R4DIRTY) && !rp->r_error) {
220 error = VOP_PUTPAGE(vp, (u_offset_t)0, 0, 0, cr, NULL);
221 if (error && (error == ENOSPC || error == EDQUOT)) {
222 mutex_enter(&rp->r_statelock);
223 if (!rp->r_error)
224 rp->r_error = error;
225 mutex_exit(&rp->r_statelock);
226 }
227 }
228 nfs4_invalidate_pages(vp, (u_offset_t)0, cr);
229 }
230 }
231
232 /*
233 * Free the resources associated with an rnode.
234 */
235 static void
236 r4inactive(rnode4_t *rp, cred_t *cr)
237 {
238 vnode_t *vp;
239 char *contents;
240 int size;
241 vsecattr_t *vsp;
242 vnode_t *xattr;
243
244 r4flushpages(rp, cr);
245
246 vp = RTOV4(rp);
247
248 /*
249 * Free any held caches which may be
250 * associated with this rnode.
251 */
252 mutex_enter(&rp->r_statelock);
253 contents = rp->r_symlink.contents;
254 size = rp->r_symlink.size;
255 rp->r_symlink.contents = NULL;
256 vsp = rp->r_secattr;
257 rp->r_secattr = NULL;
258 xattr = rp->r_xattr_dir;
259 rp->r_xattr_dir = NULL;
260 mutex_exit(&rp->r_statelock);
261
262 /*
263 * Free the access cache entries.
264 */
265 (void) nfs4_access_purge_rp(rp);
266
267 /*
268 * Free the readdir cache entries.
269 */
270 nfs4_purge_rddir_cache(vp);
271
272 /*
273 * Free the symbolic link cache.
274 */
275 if (contents != NULL) {
276
277 kmem_free((void *)contents, size);
278 }
279
280 /*
281 * Free any cached ACL.
282 */
283 if (vsp != NULL)
284 nfs4_acl_free_cache(vsp);
285
286 /*
287 * Release the cached xattr_dir
288 */
289 if (xattr != NULL)
290 VN_RELE(xattr);
291 }
292
293 /*
294 * We have seen a case that the fh passed in is for "." which
295 * should be a VROOT node, however, the fh is different from the
296 * root fh stored in the mntinfo4_t. The invalid fh might be
297 * from a misbehaved server and will panic the client system at
298 * a later time. To avoid the panic, we drop the bad fh, use
299 * the root fh from mntinfo4_t, and print an error message
300 * for attention.
301 */
302 nfs4_sharedfh_t *
303 badrootfh_check(nfs4_sharedfh_t *fh, nfs4_fname_t *nm, mntinfo4_t *mi,
304 int *wasbad)
305 {
306 char *s;
307
308 *wasbad = 0;
309 s = fn_name(nm);
310 ASSERT(strcmp(s, "..") != 0);
311
312 if ((s[0] == '.' && s[1] == '\0') && fh &&
313 !SFH4_SAME(mi->mi_rootfh, fh)) {
314 #ifdef DEBUG
315 nfs4_fhandle_t fhandle;
316
317 zcmn_err(mi->mi_zone->zone_id, CE_WARN,
318 "Server %s returns a different "
319 "root filehandle for the path %s:",
320 mi->mi_curr_serv->sv_hostname,
321 mi->mi_curr_serv->sv_path);
322
323 /* print the bad fh */
324 fhandle.fh_len = fh->sfh_fh.nfs_fh4_len;
325 bcopy(fh->sfh_fh.nfs_fh4_val, fhandle.fh_buf,
326 fhandle.fh_len);
327 nfs4_printfhandle(&fhandle);
328
329 /* print mi_rootfh */
330 fhandle.fh_len = mi->mi_rootfh->sfh_fh.nfs_fh4_len;
331 bcopy(mi->mi_rootfh->sfh_fh.nfs_fh4_val, fhandle.fh_buf,
332 fhandle.fh_len);
333 nfs4_printfhandle(&fhandle);
334 #endif
335 /* use mi_rootfh instead; fh will be rele by the caller */
336 fh = mi->mi_rootfh;
337 *wasbad = 1;
338 }
339
340 kmem_free(s, MAXNAMELEN);
341 return (fh);
342 }
343
344 void
345 r4_do_attrcache(vnode_t *vp, nfs4_ga_res_t *garp, int newnode,
346 hrtime_t t, cred_t *cr, int index)
347 {
348 int is_stub;
349 vattr_t *attr;
350 /*
351 * Don't add to attrcache if time overflow, but
352 * no need to check because either attr is null or the time
353 * values in it were processed by nfs4_time_ntov(), which checks
354 * for time overflows.
355 */
356 attr = garp ? &garp->n4g_va : NULL;
357
358 if (attr) {
359 if (!newnode) {
360 rw_exit(&rtable4[index].r_lock);
361 #ifdef DEBUG
362 if (vp->v_type != attr->va_type &&
363 vp->v_type != VNON && attr->va_type != VNON) {
364 zcmn_err(VTOMI4(vp)->mi_zone->zone_id, CE_WARN,
365 "makenfs4node: type (%d) doesn't "
366 "match type of found node at %p (%d)",
367 attr->va_type, (void *)vp, vp->v_type);
368 }
369 #endif
370 nfs4_attr_cache(vp, garp, t, cr, TRUE, NULL);
371 } else {
372 rnode4_t *rp = VTOR4(vp);
373
374 vp->v_type = attr->va_type;
375 vp->v_rdev = attr->va_rdev;
376
377 /*
378 * Turn this object into a "stub" object if we
379 * crossed an underlying server fs boundary.
380 * To make this check, during mount we save the
381 * fsid of the server object being mounted.
382 * Here we compare this object's server fsid
383 * with the fsid we saved at mount. If they
384 * are different, we crossed server fs boundary.
385 *
386 * The stub type is set (or not) at rnode
387 * creation time and it never changes for life
388 * of the rnode.
389 *
390 * This stub will be for a mirror-mount, rather than
391 * a referral (the latter also sets R4SRVSTUB).
392 *
393 * The stub type is also set during RO failover,
394 * nfs4_remap_file().
395 *
396 * We don't bother with taking r_state_lock to
397 * set the stub type because this is a new rnode
398 * and we're holding the hash bucket r_lock RW_WRITER.
399 * No other thread could have obtained access
400 * to this rnode.
401 */
402 is_stub = 0;
403 if (garp->n4g_fsid_valid) {
404 fattr4_fsid ga_fsid = garp->n4g_fsid;
405 servinfo4_t *svp = rp->r_server;
406
407 rp->r_srv_fsid = ga_fsid;
408
409 (void) nfs_rw_enter_sig(&svp->sv_lock,
410 RW_READER, 0);
411 if (!FATTR4_FSID_EQ(&ga_fsid, &svp->sv_fsid))
412 is_stub = 1;
413 nfs_rw_exit(&svp->sv_lock);
414 }
415
416 if (is_stub)
417 r4_stub_mirrormount(rp);
418 else
419 r4_stub_none(rp);
420
421 /* Can not cache partial attr */
422 if (attr->va_mask == AT_ALL)
423 nfs4_attrcache_noinval(vp, garp, t);
424 else
425 PURGE_ATTRCACHE4(vp);
426
427 rw_exit(&rtable4[index].r_lock);
428 }
429 } else {
430 if (newnode) {
431 PURGE_ATTRCACHE4(vp);
432 }
433 rw_exit(&rtable4[index].r_lock);
434 }
435 }
436
437 /*
438 * Find or create an rnode based primarily on filehandle. To be
439 * used when dvp (vnode for parent directory) is not available;
440 * otherwise, makenfs4node() should be used.
441 *
442 * The nfs4_fname_t argument *npp is consumed and nulled out.
443 */
444
445 vnode_t *
446 makenfs4node_by_fh(nfs4_sharedfh_t *sfh, nfs4_sharedfh_t *psfh,
447 nfs4_fname_t **npp, nfs4_ga_res_t *garp,
448 mntinfo4_t *mi, cred_t *cr, hrtime_t t)
449 {
450 vfs_t *vfsp = mi->mi_vfsp;
451 int newnode = 0;
452 vnode_t *vp;
453 rnode4_t *rp;
454 svnode_t *svp;
455 nfs4_fname_t *name, *svpname;
456 int index;
457
458 ASSERT(npp && *npp);
459 name = *npp;
460 *npp = NULL;
461
462 index = rtable4hash(sfh);
463 rw_enter(&rtable4[index].r_lock, RW_READER);
464
465 vp = make_rnode4(sfh, &rtable4[index], vfsp,
466 nfs4_vnodeops, nfs4_putapage, &newnode, cr);
467
468 svp = VTOSV(vp);
469 rp = VTOR4(vp);
470 if (newnode) {
471 svp->sv_forw = svp->sv_back = svp;
472 svp->sv_name = name;
473 if (psfh != NULL)
474 sfh4_hold(psfh);
475 svp->sv_dfh = psfh;
476 } else {
477 /*
478 * It is possible that due to a server
479 * side rename fnames have changed.
480 * update the fname here.
481 */
482 mutex_enter(&rp->r_svlock);
483 svpname = svp->sv_name;
484 if (svp->sv_name != name) {
485 svp->sv_name = name;
486 mutex_exit(&rp->r_svlock);
487 fn_rele(&svpname);
488 } else {
489 mutex_exit(&rp->r_svlock);
490 fn_rele(&name);
491 }
492 }
493
494 ASSERT(RW_LOCK_HELD(&rtable4[index].r_lock));
495 r4_do_attrcache(vp, garp, newnode, t, cr, index);
496 ASSERT(rw_owner(&rtable4[index].r_lock) != curthread);
497
498 return (vp);
499 }
500
501 /*
502 * Find or create a vnode for the given filehandle, filesystem, parent, and
503 * name. The reference to nm is consumed, so the caller must first do an
504 * fn_hold() if it wants to continue using nm after this call.
505 */
506 vnode_t *
507 makenfs4node(nfs4_sharedfh_t *fh, nfs4_ga_res_t *garp, struct vfs *vfsp,
508 hrtime_t t, cred_t *cr, vnode_t *dvp, nfs4_fname_t *nm)
509 {
510 vnode_t *vp;
511 int newnode;
512 int index;
513 mntinfo4_t *mi = VFTOMI4(vfsp);
514 int had_badfh = 0;
515 rnode4_t *rp;
516
517 ASSERT(dvp != NULL);
518
519 fh = badrootfh_check(fh, nm, mi, &had_badfh);
520
521 index = rtable4hash(fh);
522 rw_enter(&rtable4[index].r_lock, RW_READER);
523
524 /*
525 * Note: make_rnode4() may upgrade the hash bucket lock to exclusive.
526 */
527 vp = make_rnode4(fh, &rtable4[index], vfsp, nfs4_vnodeops,
528 nfs4_putapage, &newnode, cr);
529
530 rp = VTOR4(vp);
531 sv_activate(&vp, dvp, &nm, newnode);
532 if (dvp->v_flag & V_XATTRDIR) {
533 mutex_enter(&rp->r_statelock);
534 rp->r_flags |= R4ISXATTR;
535 mutex_exit(&rp->r_statelock);
536 }
537
538 /* if getting a bad file handle, do not cache the attributes. */
539 if (had_badfh) {
540 rw_exit(&rtable4[index].r_lock);
541 return (vp);
542 }
543
544 ASSERT(RW_LOCK_HELD(&rtable4[index].r_lock));
545 r4_do_attrcache(vp, garp, newnode, t, cr, index);
546 ASSERT(rw_owner(&rtable4[index].r_lock) != curthread);
547
548 return (vp);
549 }
550
551 /*
552 * Hash on address of filehandle object.
553 * XXX totally untuned.
554 */
555
556 int
557 rtable4hash(nfs4_sharedfh_t *fh)
558 {
559 return (((uintptr_t)fh / sizeof (*fh)) & rtable4mask);
560 }
561
562 /*
563 * Find or create the vnode for the given filehandle and filesystem.
564 * *newnode is set to zero if the vnode already existed; non-zero if it had
565 * to be created.
566 *
567 * Note: make_rnode4() may upgrade the hash bucket lock to exclusive.
568 */
569
570 static vnode_t *
571 make_rnode4(nfs4_sharedfh_t *fh, r4hashq_t *rhtp, struct vfs *vfsp,
572 struct vnodeops *vops,
573 int (*putapage)(vnode_t *, page_t *, u_offset_t *, size_t *, int, cred_t *),
574 int *newnode, cred_t *cr)
575 {
576 rnode4_t *rp;
577 rnode4_t *trp;
578 vnode_t *vp;
579 mntinfo4_t *mi;
580
581 ASSERT(RW_READ_HELD(&rhtp->r_lock));
582
583 mi = VFTOMI4(vfsp);
584
585 start:
586 if ((rp = r4find(rhtp, fh, vfsp)) != NULL) {
587 vp = RTOV4(rp);
588 *newnode = 0;
589 return (vp);
590 }
591 rw_exit(&rhtp->r_lock);
592
593 mutex_enter(&rp4freelist_lock);
594
595 if (rp4freelist != NULL && rnode4_new >= nrnode) {
596 rp = rp4freelist;
597 rp4_rmfree(rp);
598 mutex_exit(&rp4freelist_lock);
599
600 vp = RTOV4(rp);
601
602 if (rp->r_flags & R4HASHED) {
603 rw_enter(&rp->r_hashq->r_lock, RW_WRITER);
604 mutex_enter(&vp->v_lock);
605 if (vp->v_count > 1) {
606 vp->v_count--;
607 mutex_exit(&vp->v_lock);
608 rw_exit(&rp->r_hashq->r_lock);
609 rw_enter(&rhtp->r_lock, RW_READER);
610 goto start;
611 }
612 mutex_exit(&vp->v_lock);
613 rp4_rmhash_locked(rp);
614 rw_exit(&rp->r_hashq->r_lock);
615 }
616
617 r4inactive(rp, cr);
618
619 mutex_enter(&vp->v_lock);
620 if (vp->v_count > 1) {
621 vp->v_count--;
622 mutex_exit(&vp->v_lock);
623 rw_enter(&rhtp->r_lock, RW_READER);
624 goto start;
625 }
626 mutex_exit(&vp->v_lock);
627 vn_invalid(vp);
628
629 /*
630 * destroy old locks before bzero'ing and
631 * recreating the locks below.
632 */
633 uninit_rnode4(rp);
634
635 /*
636 * Make sure that if rnode is recycled then
637 * VFS count is decremented properly before
638 * reuse.
639 */
640 VFS_RELE(vp->v_vfsp);
641 vn_reinit(vp);
642 } else {
643 vnode_t *new_vp;
644
645 mutex_exit(&rp4freelist_lock);
646
647 rp = kmem_cache_alloc(rnode4_cache, KM_SLEEP);
648 new_vp = vn_alloc(KM_SLEEP);
649
650 atomic_add_long((ulong_t *)&rnode4_new, 1);
651 #ifdef DEBUG
652 clstat4_debug.nrnode.value.ui64++;
653 #endif
654 vp = new_vp;
655 }
656
657 bzero(rp, sizeof (*rp));
658 rp->r_vnode = vp;
659 nfs_rw_init(&rp->r_rwlock, NULL, RW_DEFAULT, NULL);
660 nfs_rw_init(&rp->r_lkserlock, NULL, RW_DEFAULT, NULL);
661 mutex_init(&rp->r_svlock, NULL, MUTEX_DEFAULT, NULL);
662 mutex_init(&rp->r_statelock, NULL, MUTEX_DEFAULT, NULL);
663 mutex_init(&rp->r_statev4_lock, NULL, MUTEX_DEFAULT, NULL);
664 mutex_init(&rp->r_os_lock, NULL, MUTEX_DEFAULT, NULL);
665 rp->created_v4 = 0;
666 list_create(&rp->r_open_streams, sizeof (nfs4_open_stream_t),
667 offsetof(nfs4_open_stream_t, os_node));
668 rp->r_lo_head.lo_prev_rnode = &rp->r_lo_head;
669 rp->r_lo_head.lo_next_rnode = &rp->r_lo_head;
670 cv_init(&rp->r_cv, NULL, CV_DEFAULT, NULL);
671 cv_init(&rp->r_commit.c_cv, NULL, CV_DEFAULT, NULL);
672 rp->r_flags = R4READDIRWATTR;
673 rp->r_fh = fh;
674 rp->r_hashq = rhtp;
675 sfh4_hold(rp->r_fh);
676 rp->r_server = mi->mi_curr_serv;
677 rp->r_deleg_type = OPEN_DELEGATE_NONE;
678 rp->r_deleg_needs_recovery = OPEN_DELEGATE_NONE;
679 nfs_rw_init(&rp->r_deleg_recall_lock, NULL, RW_DEFAULT, NULL);
680
681 rddir4_cache_create(rp);
682 rp->r_putapage = putapage;
683 vn_setops(vp, vops);
684 vp->v_data = (caddr_t)rp;
685 vp->v_vfsp = vfsp;
686 VFS_HOLD(vfsp);
687 vp->v_type = VNON;
688 vp->v_flag |= VMODSORT;
689 if (isrootfh(fh, rp))
690 vp->v_flag = VROOT;
691 vn_exists(vp);
692
693 /*
694 * There is a race condition if someone else
695 * alloc's the rnode while no locks are held, so we
696 * check again and recover if found.
697 */
698 rw_enter(&rhtp->r_lock, RW_WRITER);
699 if ((trp = r4find(rhtp, fh, vfsp)) != NULL) {
700 vp = RTOV4(trp);
701 *newnode = 0;
702 rw_exit(&rhtp->r_lock);
703 rp4_addfree(rp, cr);
704 rw_enter(&rhtp->r_lock, RW_READER);
705 return (vp);
706 }
707 rp4_addhash(rp);
708 *newnode = 1;
709 return (vp);
710 }
711
712 static void
713 uninit_rnode4(rnode4_t *rp)
714 {
715 vnode_t *vp = RTOV4(rp);
716
717 ASSERT(rp != NULL);
718 ASSERT(vp != NULL);
719 ASSERT(vp->v_count == 1);
720 ASSERT(rp->r_count == 0);
721 ASSERT(rp->r_mapcnt == 0);
722 if (rp->r_flags & R4LODANGLERS) {
723 nfs4_flush_lock_owners(rp);
724 }
725 ASSERT(rp->r_lo_head.lo_next_rnode == &rp->r_lo_head);
726 ASSERT(rp->r_lo_head.lo_prev_rnode == &rp->r_lo_head);
727 ASSERT(!(rp->r_flags & R4HASHED));
728 ASSERT(rp->r_freef == NULL && rp->r_freeb == NULL);
729 nfs4_clear_open_streams(rp);
730 list_destroy(&rp->r_open_streams);
731
732 /*
733 * Destroy the rddir cache first since we need to grab the r_statelock.
734 */
735 mutex_enter(&rp->r_statelock);
736 rddir4_cache_destroy(rp);
737 mutex_exit(&rp->r_statelock);
738 sv_uninit(&rp->r_svnode);
739 sfh4_rele(&rp->r_fh);
740 nfs_rw_destroy(&rp->r_rwlock);
741 nfs_rw_destroy(&rp->r_lkserlock);
742 mutex_destroy(&rp->r_statelock);
743 mutex_destroy(&rp->r_statev4_lock);
744 mutex_destroy(&rp->r_os_lock);
745 cv_destroy(&rp->r_cv);
746 cv_destroy(&rp->r_commit.c_cv);
747 nfs_rw_destroy(&rp->r_deleg_recall_lock);
748 if (rp->r_flags & R4DELMAPLIST)
749 list_destroy(&rp->r_indelmap);
750 }
751
752 /*
753 * Put an rnode on the free list.
754 *
755 * Rnodes which were allocated above and beyond the normal limit
756 * are immediately freed.
757 */
758 void
759 rp4_addfree(rnode4_t *rp, cred_t *cr)
760 {
761 vnode_t *vp;
762 vnode_t *xattr;
763 struct vfs *vfsp;
764
765 vp = RTOV4(rp);
766 ASSERT(vp->v_count >= 1);
767 ASSERT(rp->r_freef == NULL && rp->r_freeb == NULL);
768
769 /*
770 * If we have too many rnodes allocated and there are no
771 * references to this rnode, or if the rnode is no longer
772 * accessible by it does not reside in the hash queues,
773 * or if an i/o error occurred while writing to the file,
774 * then just free it instead of putting it on the rnode
775 * freelist.
776 */
777 vfsp = vp->v_vfsp;
778 if (((rnode4_new > nrnode || !(rp->r_flags & R4HASHED) ||
779 #ifdef DEBUG
780 (nfs4_rnode_nofreelist != 0) ||
781 #endif
782 rp->r_error || (rp->r_flags & R4RECOVERR) ||
783 (vfsp->vfs_flag & VFS_UNMOUNTED)) && rp->r_count == 0)) {
784 if (rp->r_flags & R4HASHED) {
785 rw_enter(&rp->r_hashq->r_lock, RW_WRITER);
786 mutex_enter(&vp->v_lock);
787 if (vp->v_count > 1) {
788 vp->v_count--;
789 mutex_exit(&vp->v_lock);
790 rw_exit(&rp->r_hashq->r_lock);
791 return;
792 }
793 mutex_exit(&vp->v_lock);
794 rp4_rmhash_locked(rp);
795 rw_exit(&rp->r_hashq->r_lock);
796 }
797
798 /*
799 * Make sure we don't have a delegation on this rnode
800 * before destroying it.
801 */
802 if (rp->r_deleg_type != OPEN_DELEGATE_NONE) {
803 (void) nfs4delegreturn(rp,
804 NFS4_DR_FORCE|NFS4_DR_PUSH|NFS4_DR_REOPEN);
805 }
806
807 r4inactive(rp, cr);
808
809 /*
810 * Recheck the vnode reference count. We need to
811 * make sure that another reference has not been
812 * acquired while we were not holding v_lock. The
813 * rnode is not in the rnode hash queues; one
814 * way for a reference to have been acquired
815 * is for a VOP_PUTPAGE because the rnode was marked
816 * with R4DIRTY or for a modified page. This
817 * reference may have been acquired before our call
818 * to r4inactive. The i/o may have been completed,
819 * thus allowing r4inactive to complete, but the
820 * reference to the vnode may not have been released
821 * yet. In any case, the rnode can not be destroyed
822 * until the other references to this vnode have been
823 * released. The other references will take care of
824 * either destroying the rnode or placing it on the
825 * rnode freelist. If there are no other references,
826 * then the rnode may be safely destroyed.
827 */
828 mutex_enter(&vp->v_lock);
829 if (vp->v_count > 1) {
830 vp->v_count--;
831 mutex_exit(&vp->v_lock);
832 return;
833 }
834 mutex_exit(&vp->v_lock);
835
836 destroy_rnode4(rp);
837 return;
838 }
839
840 /*
841 * Lock the hash queue and then recheck the reference count
842 * to ensure that no other threads have acquired a reference
843 * to indicate that the rnode should not be placed on the
844 * freelist. If another reference has been acquired, then
845 * just release this one and let the other thread complete
846 * the processing of adding this rnode to the freelist.
847 */
848 again:
849 rw_enter(&rp->r_hashq->r_lock, RW_WRITER);
850
851 mutex_enter(&vp->v_lock);
852 if (vp->v_count > 1) {
853 vp->v_count--;
854 mutex_exit(&vp->v_lock);
855 rw_exit(&rp->r_hashq->r_lock);
856 return;
857 }
858 mutex_exit(&vp->v_lock);
859
860 /*
861 * Make sure we don't put an rnode with a delegation
862 * on the free list.
863 */
864 if (rp->r_deleg_type != OPEN_DELEGATE_NONE) {
865 rw_exit(&rp->r_hashq->r_lock);
866 (void) nfs4delegreturn(rp,
867 NFS4_DR_FORCE|NFS4_DR_PUSH|NFS4_DR_REOPEN);
868 goto again;
869 }
870
871 /*
872 * Now that we have the hash queue lock, and we know there
873 * are not anymore references on the vnode, check to make
874 * sure there aren't any open streams still on the rnode.
875 * If so, drop the hash queue lock, remove the open streams,
876 * and recheck the v_count.
877 */
878 mutex_enter(&rp->r_os_lock);
879 if (list_head(&rp->r_open_streams) != NULL) {
880 mutex_exit(&rp->r_os_lock);
881 rw_exit(&rp->r_hashq->r_lock);
882 if (nfs_zone() != VTOMI4(vp)->mi_zone)
883 nfs4_clear_open_streams(rp);
884 else
885 (void) nfs4close_all(vp, cr);
886 goto again;
887 }
888 mutex_exit(&rp->r_os_lock);
889
890 /*
891 * Before we put it on the freelist, make sure there are no pages.
892 * If there are, flush and commit of all of the dirty and
893 * uncommitted pages, assuming the file system isn't read only.
894 */
895 if (!(vp->v_vfsp->vfs_flag & VFS_RDONLY) && nfs4_dross_pages(vp)) {
896 rw_exit(&rp->r_hashq->r_lock);
897 r4flushpages(rp, cr);
898 goto again;
899 }
900
901 /*
902 * Before we put it on the freelist, make sure there is no
903 * active xattr directory cached, the freelist will not
904 * have its entries r4inactive'd if there is still an active
905 * rnode, thus nothing in the freelist can hold another
906 * rnode active.
907 */
908 xattr = rp->r_xattr_dir;
909 rp->r_xattr_dir = NULL;
910
911 /*
912 * If there is no cached data or metadata for this file, then
913 * put the rnode on the front of the freelist so that it will
914 * be reused before other rnodes which may have cached data or
915 * metadata associated with them.
916 */
917 mutex_enter(&rp4freelist_lock);
918 if (rp4freelist == NULL) {
919 rp->r_freef = rp;
920 rp->r_freeb = rp;
921 rp4freelist = rp;
922 } else {
923 rp->r_freef = rp4freelist;
924 rp->r_freeb = rp4freelist->r_freeb;
925 rp4freelist->r_freeb->r_freef = rp;
926 rp4freelist->r_freeb = rp;
927 if (!nfs4_has_pages(vp) && rp->r_dir == NULL &&
928 rp->r_symlink.contents == NULL && rp->r_secattr == NULL)
929 rp4freelist = rp;
930 }
931 mutex_exit(&rp4freelist_lock);
932
933 rw_exit(&rp->r_hashq->r_lock);
934
935 if (xattr)
936 VN_RELE(xattr);
937 }
938
939 /*
940 * Remove an rnode from the free list.
941 *
942 * The caller must be holding rp4freelist_lock and the rnode
943 * must be on the freelist.
944 */
945 static void
946 rp4_rmfree(rnode4_t *rp)
947 {
948
949 ASSERT(MUTEX_HELD(&rp4freelist_lock));
950 ASSERT(rp->r_freef != NULL && rp->r_freeb != NULL);
951
952 if (rp == rp4freelist) {
953 rp4freelist = rp->r_freef;
954 if (rp == rp4freelist)
955 rp4freelist = NULL;
956 }
957 rp->r_freeb->r_freef = rp->r_freef;
958 rp->r_freef->r_freeb = rp->r_freeb;
959
960 rp->r_freef = rp->r_freeb = NULL;
961 }
962
963 /*
964 * Put a rnode in the hash table.
965 *
966 * The caller must be holding the exclusive hash queue lock
967 */
968 void
969 rp4_addhash(rnode4_t *rp)
970 {
971 ASSERT(RW_WRITE_HELD(&rp->r_hashq->r_lock));
972 ASSERT(!(rp->r_flags & R4HASHED));
973
974 #ifdef DEBUG
975 r4_dup_check(rp, RTOV4(rp)->v_vfsp);
976 #endif
977
978 rp->r_hashf = rp->r_hashq->r_hashf;
979 rp->r_hashq->r_hashf = rp;
980 rp->r_hashb = (rnode4_t *)rp->r_hashq;
981 rp->r_hashf->r_hashb = rp;
982
983 mutex_enter(&rp->r_statelock);
984 rp->r_flags |= R4HASHED;
985 mutex_exit(&rp->r_statelock);
986 }
987
988 /*
989 * Remove a rnode from the hash table.
990 *
991 * The caller must be holding the hash queue lock.
992 */
993 void
994 rp4_rmhash_locked(rnode4_t *rp)
995 {
996 ASSERT(RW_WRITE_HELD(&rp->r_hashq->r_lock));
997 ASSERT(rp->r_flags & R4HASHED);
998
999 rp->r_hashb->r_hashf = rp->r_hashf;
1000 rp->r_hashf->r_hashb = rp->r_hashb;
1001
1002 mutex_enter(&rp->r_statelock);
1003 rp->r_flags &= ~R4HASHED;
1004 mutex_exit(&rp->r_statelock);
1005 }
1006
1007 /*
1008 * Remove a rnode from the hash table.
1009 *
1010 * The caller must not be holding the hash queue lock.
1011 */
1012 void
1013 rp4_rmhash(rnode4_t *rp)
1014 {
1015 rw_enter(&rp->r_hashq->r_lock, RW_WRITER);
1016 rp4_rmhash_locked(rp);
1017 rw_exit(&rp->r_hashq->r_lock);
1018 }
1019
1020 /*
1021 * Lookup a rnode by fhandle. Ignores rnodes that had failed recovery.
1022 * Returns NULL if no match. If an rnode is returned, the reference count
1023 * on the master vnode is incremented.
1024 *
1025 * The caller must be holding the hash queue lock, either shared or exclusive.
1026 */
1027 rnode4_t *
1028 r4find(r4hashq_t *rhtp, nfs4_sharedfh_t *fh, struct vfs *vfsp)
1029 {
1030 rnode4_t *rp;
1031 vnode_t *vp;
1032
1033 ASSERT(RW_LOCK_HELD(&rhtp->r_lock));
1034
1035 for (rp = rhtp->r_hashf; rp != (rnode4_t *)rhtp; rp = rp->r_hashf) {
1036 vp = RTOV4(rp);
1037 if (vp->v_vfsp == vfsp && SFH4_SAME(rp->r_fh, fh)) {
1038
1039 mutex_enter(&rp->r_statelock);
1040 if (rp->r_flags & R4RECOVERR) {
1041 mutex_exit(&rp->r_statelock);
1042 continue;
1043 }
1044 mutex_exit(&rp->r_statelock);
1045 #ifdef DEBUG
1046 r4_dup_check(rp, vfsp);
1047 #endif
1048 if (rp->r_freef != NULL) {
1049 mutex_enter(&rp4freelist_lock);
1050 /*
1051 * If the rnode is on the freelist,
1052 * then remove it and use that reference
1053 * as the new reference. Otherwise,
1054 * need to increment the reference count.
1055 */
1056 if (rp->r_freef != NULL) {
1057 rp4_rmfree(rp);
1058 mutex_exit(&rp4freelist_lock);
1059 } else {
1060 mutex_exit(&rp4freelist_lock);
1061 VN_HOLD(vp);
1062 }
1063 } else
1064 VN_HOLD(vp);
1065
1066 /*
1067 * if root vnode, set v_flag to indicate that
1068 */
1069 if (isrootfh(fh, rp)) {
1070 if (!(vp->v_flag & VROOT)) {
1071 mutex_enter(&vp->v_lock);
1072 vp->v_flag |= VROOT;
1073 mutex_exit(&vp->v_lock);
1074 }
1075 }
1076 return (rp);
1077 }
1078 }
1079 return (NULL);
1080 }
1081
1082 /*
1083 * Lookup an rnode by fhandle. Just a wrapper for r4find()
1084 * that assumes the caller hasn't already got the lock
1085 * on the hash bucket.
1086 */
1087 rnode4_t *
1088 r4find_unlocked(nfs4_sharedfh_t *fh, struct vfs *vfsp)
1089 {
1090 rnode4_t *rp;
1091 int index;
1092
1093 index = rtable4hash(fh);
1094 rw_enter(&rtable4[index].r_lock, RW_READER);
1095 rp = r4find(&rtable4[index], fh, vfsp);
1096 rw_exit(&rtable4[index].r_lock);
1097
1098 return (rp);
1099 }
1100
1101 /*
1102 * Return >0 if there is a active vnode belonging to this vfs in the
1103 * rtable4 cache.
1104 *
1105 * Several of these checks are done without holding the usual
1106 * locks. This is safe because destroy_rtable(), rp_addfree(),
1107 * etc. will redo the necessary checks before actually destroying
1108 * any rnodes.
1109 */
1110 int
1111 check_rtable4(struct vfs *vfsp)
1112 {
1113 rnode4_t *rp;
1114 vnode_t *vp;
1115 int busy = NFSV4_RTABLE4_OK;
1116 int index;
1117
1118 for (index = 0; index < rtable4size; index++) {
1119 rw_enter(&rtable4[index].r_lock, RW_READER);
1120
1121 for (rp = rtable4[index].r_hashf;
1122 rp != (rnode4_t *)(&rtable4[index]);
1123 rp = rp->r_hashf) {
1124
1125 vp = RTOV4(rp);
1126 if (vp->v_vfsp == vfsp) {
1127 if (rp->r_freef == NULL) {
1128 busy = NFSV4_RTABLE4_NOT_FREE_LIST;
1129 } else if (nfs4_has_pages(vp) &&
1130 (rp->r_flags & R4DIRTY)) {
1131 busy = NFSV4_RTABLE4_DIRTY_PAGES;
1132 } else if (rp->r_count > 0) {
1133 busy = NFSV4_RTABLE4_POS_R_COUNT;
1134 }
1135
1136 if (busy != NFSV4_RTABLE4_OK) {
1137 #ifdef DEBUG
1138 char *path;
1139
1140 path = fn_path(rp->r_svnode.sv_name);
1141 DTRACE_NFSV4_3(rnode__e__debug,
1142 int, busy, char *, path,
1143 rnode4_t *, rp);
1144 kmem_free(path, strlen(path)+1);
1145 #endif
1146 rw_exit(&rtable4[index].r_lock);
1147 return (busy);
1148 }
1149 }
1150 }
1151 rw_exit(&rtable4[index].r_lock);
1152 }
1153 return (busy);
1154 }
1155
1156 /*
1157 * Destroy inactive vnodes from the hash queues which
1158 * belong to this vfs. All of the vnodes should be inactive.
1159 * It is essential that we destroy all rnodes in case of
1160 * forced unmount as well as in normal unmount case.
1161 */
1162
1163 void
1164 destroy_rtable4(struct vfs *vfsp, cred_t *cr)
1165 {
1166 int index;
1167 vnode_t *vp;
1168 rnode4_t *rp, *r_hashf, *rlist;
1169
1170 rlist = NULL;
1171
1172 for (index = 0; index < rtable4size; index++) {
1173 rw_enter(&rtable4[index].r_lock, RW_WRITER);
1174 for (rp = rtable4[index].r_hashf;
1175 rp != (rnode4_t *)(&rtable4[index]);
1176 rp = r_hashf) {
1177 /* save the hash pointer before destroying */
1178 r_hashf = rp->r_hashf;
1179
1180 vp = RTOV4(rp);
1181 if (vp->v_vfsp == vfsp) {
1182 mutex_enter(&rp4freelist_lock);
1183 if (rp->r_freef != NULL) {
1184 rp4_rmfree(rp);
1185 mutex_exit(&rp4freelist_lock);
1186 rp4_rmhash_locked(rp);
1187 rp->r_hashf = rlist;
1188 rlist = rp;
1189 } else
1190 mutex_exit(&rp4freelist_lock);
1191 }
1192 }
1193 rw_exit(&rtable4[index].r_lock);
1194 }
1195
1196 for (rp = rlist; rp != NULL; rp = r_hashf) {
1197 r_hashf = rp->r_hashf;
1198 /*
1199 * This call to rp4_addfree will end up destroying the
1200 * rnode, but in a safe way with the appropriate set
1201 * of checks done.
1202 */
1203 rp4_addfree(rp, cr);
1204 }
1205 }
1206
1207 /*
1208 * This routine destroys all the resources of an rnode
1209 * and finally the rnode itself.
1210 */
1211 static void
1212 destroy_rnode4(rnode4_t *rp)
1213 {
1214 vnode_t *vp;
1215 vfs_t *vfsp;
1216
1217 ASSERT(rp->r_deleg_type == OPEN_DELEGATE_NONE);
1218
1219 vp = RTOV4(rp);
1220 vfsp = vp->v_vfsp;
1221
1222 uninit_rnode4(rp);
1223 atomic_add_long((ulong_t *)&rnode4_new, -1);
1224 #ifdef DEBUG
1225 clstat4_debug.nrnode.value.ui64--;
1226 #endif
1227 kmem_cache_free(rnode4_cache, rp);
1228 vn_invalid(vp);
1229 vn_free(vp);
1230 VFS_RELE(vfsp);
1231 }
1232
1233 /*
1234 * Invalidate the attributes on all rnodes forcing the next getattr
1235 * to go over the wire. Used to flush stale uid and gid mappings.
1236 * Maybe done on a per vfsp, or all rnodes (vfsp == NULL)
1237 */
1238 void
1239 nfs4_rnode_invalidate(struct vfs *vfsp)
1240 {
1241 int index;
1242 rnode4_t *rp;
1243 vnode_t *vp;
1244
1245 /*
1246 * Walk the hash queues looking for rnodes.
1247 */
1248 for (index = 0; index < rtable4size; index++) {
1249 rw_enter(&rtable4[index].r_lock, RW_READER);
1250 for (rp = rtable4[index].r_hashf;
1251 rp != (rnode4_t *)(&rtable4[index]);
1252 rp = rp->r_hashf) {
1253 vp = RTOV4(rp);
1254 if (vfsp != NULL && vp->v_vfsp != vfsp)
1255 continue;
1256
1257 if (!mutex_tryenter(&rp->r_statelock))
1258 continue;
1259
1260 /*
1261 * Expire the attributes by resetting the change
1262 * and attr timeout.
1263 */
1264 rp->r_change = 0;
1265 PURGE_ATTRCACHE4_LOCKED(rp);
1266 mutex_exit(&rp->r_statelock);
1267 }
1268 rw_exit(&rtable4[index].r_lock);
1269 }
1270 }
1271
1272 /*
1273 * Flush all vnodes in this (or every) vfs.
1274 * Used by nfs_sync and by nfs_unmount.
1275 */
1276 void
1277 r4flush(struct vfs *vfsp, cred_t *cr)
1278 {
1279 int index;
1280 rnode4_t *rp;
1281 vnode_t *vp, **vplist;
1282 long num, cnt;
1283
1284 /*
1285 * Check to see whether there is anything to do.
1286 */
1287 num = rnode4_new;
1288 if (num == 0)
1289 return;
1290
1291 /*
1292 * Allocate a slot for all currently active rnodes on the
1293 * supposition that they all may need flushing.
1294 */
1295 vplist = kmem_alloc(num * sizeof (*vplist), KM_SLEEP);
1296 cnt = 0;
1297
1298 /*
1299 * Walk the hash queues looking for rnodes with page
1300 * lists associated with them. Make a list of these
1301 * files.
1302 */
1303 for (index = 0; index < rtable4size; index++) {
1304 rw_enter(&rtable4[index].r_lock, RW_READER);
1305 for (rp = rtable4[index].r_hashf;
1306 rp != (rnode4_t *)(&rtable4[index]);
1307 rp = rp->r_hashf) {
1308 vp = RTOV4(rp);
1309 /*
1310 * Don't bother sync'ing a vp if it
1311 * is part of virtual swap device or
1312 * if VFS is read-only
1313 */
1314 if (IS_SWAPVP(vp) || vn_is_readonly(vp))
1315 continue;
1316 /*
1317 * If flushing all mounted file systems or
1318 * the vnode belongs to this vfs, has pages
1319 * and is marked as either dirty or mmap'd,
1320 * hold and add this vnode to the list of
1321 * vnodes to flush.
1322 */
1323 if ((vfsp == NULL || vp->v_vfsp == vfsp) &&
1324 nfs4_has_pages(vp) &&
1325 ((rp->r_flags & R4DIRTY) || rp->r_mapcnt > 0)) {
1326 VN_HOLD(vp);
1327 vplist[cnt++] = vp;
1328 if (cnt == num) {
1329 rw_exit(&rtable4[index].r_lock);
1330 goto toomany;
1331 }
1332 }
1333 }
1334 rw_exit(&rtable4[index].r_lock);
1335 }
1336 toomany:
1337
1338 /*
1339 * Flush and release all of the files on the list.
1340 */
1341 while (cnt-- > 0) {
1342 vp = vplist[cnt];
1343 (void) VOP_PUTPAGE(vp, (u_offset_t)0, 0, B_ASYNC, cr, NULL);
1344 VN_RELE(vp);
1345 }
1346
1347 /*
1348 * Free the space allocated to hold the list.
1349 */
1350 kmem_free(vplist, num * sizeof (*vplist));
1351 }
1352
1353 int
1354 nfs4_free_data_reclaim(rnode4_t *rp)
1355 {
1356 char *contents;
1357 vnode_t *xattr;
1358 int size;
1359 vsecattr_t *vsp;
1360 int freed;
1361 bool_t rdc = FALSE;
1362
1363 /*
1364 * Free any held caches which may
1365 * be associated with this rnode.
1366 */
1367 mutex_enter(&rp->r_statelock);
1368 if (rp->r_dir != NULL)
1369 rdc = TRUE;
1370 contents = rp->r_symlink.contents;
1371 size = rp->r_symlink.size;
1372 rp->r_symlink.contents = NULL;
1373 vsp = rp->r_secattr;
1374 rp->r_secattr = NULL;
1375 xattr = rp->r_xattr_dir;
1376 rp->r_xattr_dir = NULL;
1377 mutex_exit(&rp->r_statelock);
1378
1379 /*
1380 * Free the access cache entries.
1381 */
1382 freed = nfs4_access_purge_rp(rp);
1383
1384 if (rdc == FALSE && contents == NULL && vsp == NULL && xattr == NULL)
1385 return (freed);
1386
1387 /*
1388 * Free the readdir cache entries, incompletely if we can't block.
1389 */
1390 nfs4_purge_rddir_cache(RTOV4(rp));
1391
1392 /*
1393 * Free the symbolic link cache.
1394 */
1395 if (contents != NULL) {
1396
1397 kmem_free((void *)contents, size);
1398 }
1399
1400 /*
1401 * Free any cached ACL.
1402 */
1403 if (vsp != NULL)
1404 nfs4_acl_free_cache(vsp);
1405
1406 /*
1407 * Release the xattr directory vnode
1408 */
1409 if (xattr != NULL)
1410 VN_RELE(xattr);
1411
1412 return (1);
1413 }
1414
1415 static int
1416 nfs4_active_data_reclaim(rnode4_t *rp)
1417 {
1418 char *contents;
1419 vnode_t *xattr = NULL;
1420 int size;
1421 vsecattr_t *vsp;
1422 int freed;
1423 bool_t rdc = FALSE;
1424
1425 /*
1426 * Free any held credentials and caches which
1427 * may be associated with this rnode.
1428 */
1429 if (!mutex_tryenter(&rp->r_statelock))
1430 return (0);
1431 contents = rp->r_symlink.contents;
1432 size = rp->r_symlink.size;
1433 rp->r_symlink.contents = NULL;
1434 vsp = rp->r_secattr;
1435 rp->r_secattr = NULL;
1436 if (rp->r_dir != NULL)
1437 rdc = TRUE;
1438 /*
1439 * To avoid a deadlock, do not free r_xattr_dir cache if it is hashed
1440 * on the same r_hashq queue. We are not mandated to free all caches.
1441 * VN_RELE(rp->r_xattr_dir) will be done sometime later - e.g. when the
1442 * rnode 'rp' is freed or put on the free list.
1443 *
1444 * We will retain NFS4_XATTR_DIR_NOTSUPP because:
1445 * - it has no associated rnode4_t (its v_data is NULL),
1446 * - it is preallocated statically and will never go away,
1447 * so we cannot save anything by releasing it.
1448 */
1449 if (rp->r_xattr_dir && rp->r_xattr_dir != NFS4_XATTR_DIR_NOTSUPP &&
1450 VTOR4(rp->r_xattr_dir)->r_hashq != rp->r_hashq) {
1451 xattr = rp->r_xattr_dir;
1452 rp->r_xattr_dir = NULL;
1453 }
1454 mutex_exit(&rp->r_statelock);
1455
1456 /*
1457 * Free the access cache entries.
1458 */
1459 freed = nfs4_access_purge_rp(rp);
1460
1461 if (contents == NULL && vsp == NULL && rdc == FALSE && xattr == NULL)
1462 return (freed);
1463
1464 /*
1465 * Free the symbolic link cache.
1466 */
1467 if (contents != NULL) {
1468
1469 kmem_free((void *)contents, size);
1470 }
1471
1472 /*
1473 * Free any cached ACL.
1474 */
1475 if (vsp != NULL)
1476 nfs4_acl_free_cache(vsp);
1477
1478 nfs4_purge_rddir_cache(RTOV4(rp));
1479
1480 /*
1481 * Release the xattr directory vnode
1482 */
1483 if (xattr != NULL)
1484 VN_RELE(xattr);
1485
1486 return (1);
1487 }
1488
1489 static int
1490 nfs4_free_reclaim(void)
1491 {
1492 int freed;
1493 rnode4_t *rp;
1494
1495 #ifdef DEBUG
1496 clstat4_debug.f_reclaim.value.ui64++;
1497 #endif
1498 freed = 0;
1499 mutex_enter(&rp4freelist_lock);
1500 rp = rp4freelist;
1501 if (rp != NULL) {
1502 do {
1503 if (nfs4_free_data_reclaim(rp))
1504 freed = 1;
1505 } while ((rp = rp->r_freef) != rp4freelist);
1506 }
1507 mutex_exit(&rp4freelist_lock);
1508 return (freed);
1509 }
1510
1511 static int
1512 nfs4_active_reclaim(void)
1513 {
1514 int freed;
1515 int index;
1516 rnode4_t *rp;
1517
1518 #ifdef DEBUG
1519 clstat4_debug.a_reclaim.value.ui64++;
1520 #endif
1521 freed = 0;
1522 for (index = 0; index < rtable4size; index++) {
1523 rw_enter(&rtable4[index].r_lock, RW_READER);
1524 for (rp = rtable4[index].r_hashf;
1525 rp != (rnode4_t *)(&rtable4[index]);
1526 rp = rp->r_hashf) {
1527 if (nfs4_active_data_reclaim(rp))
1528 freed = 1;
1529 }
1530 rw_exit(&rtable4[index].r_lock);
1531 }
1532 return (freed);
1533 }
1534
1535 static int
1536 nfs4_rnode_reclaim(void)
1537 {
1538 int freed;
1539 rnode4_t *rp;
1540 vnode_t *vp;
1541
1542 #ifdef DEBUG
1543 clstat4_debug.r_reclaim.value.ui64++;
1544 #endif
1545 freed = 0;
1546 mutex_enter(&rp4freelist_lock);
1547 while ((rp = rp4freelist) != NULL) {
1548 rp4_rmfree(rp);
1549 mutex_exit(&rp4freelist_lock);
1550 if (rp->r_flags & R4HASHED) {
1551 vp = RTOV4(rp);
1552 rw_enter(&rp->r_hashq->r_lock, RW_WRITER);
1553 mutex_enter(&vp->v_lock);
1554 if (vp->v_count > 1) {
1555 vp->v_count--;
1556 mutex_exit(&vp->v_lock);
1557 rw_exit(&rp->r_hashq->r_lock);
1558 mutex_enter(&rp4freelist_lock);
1559 continue;
1560 }
1561 mutex_exit(&vp->v_lock);
1562 rp4_rmhash_locked(rp);
1563 rw_exit(&rp->r_hashq->r_lock);
1564 }
1565 /*
1566 * This call to rp_addfree will end up destroying the
1567 * rnode, but in a safe way with the appropriate set
1568 * of checks done.
1569 */
1570 rp4_addfree(rp, CRED());
1571 mutex_enter(&rp4freelist_lock);
1572 }
1573 mutex_exit(&rp4freelist_lock);
1574 return (freed);
1575 }
1576
1577 /*ARGSUSED*/
1578 static void
1579 nfs4_reclaim(void *cdrarg)
1580 {
1581 #ifdef DEBUG
1582 clstat4_debug.reclaim.value.ui64++;
1583 #endif
1584 if (nfs4_free_reclaim())
1585 return;
1586
1587 if (nfs4_active_reclaim())
1588 return;
1589
1590 (void) nfs4_rnode_reclaim();
1591 }
1592
1593 /*
1594 * Returns the clientid4 to use for the given mntinfo4. Note that the
1595 * clientid can change if the caller drops mi_recovlock.
1596 */
1597
1598 clientid4
1599 mi2clientid(mntinfo4_t *mi)
1600 {
1601 nfs4_server_t *sp;
1602 clientid4 clientid = 0;
1603
1604 /* this locks down sp if it is found */
1605 sp = find_nfs4_server(mi);
1606 if (sp != NULL) {
1607 clientid = sp->clientid;
1608 mutex_exit(&sp->s_lock);
1609 nfs4_server_rele(sp);
1610 }
1611 return (clientid);
1612 }
1613
1614 /*
1615 * Return a list with information about all the known open instances for
1616 * a filesystem. The caller must call r4releopenlist() when done with the
1617 * list.
1618 *
1619 * We are safe at looking at os_valid and os_pending_close across dropping
1620 * the 'os_sync_lock' to count up the number of open streams and then
1621 * allocate memory for the osp list due to:
1622 * -Looking at os_pending_close is safe since this routine is
1623 * only called via recovery, and os_pending_close can only be set via
1624 * a non-recovery operation (which are all blocked when recovery
1625 * is active).
1626 *
1627 * -Examining os_valid is safe since non-recovery operations, which
1628 * could potentially switch os_valid to 0, are blocked (via
1629 * nfs4_start_fop) and recovery is single-threaded per mntinfo4_t
1630 * (which means we are the only recovery thread potentially acting
1631 * on this open stream).
1632 */
1633
1634 nfs4_opinst_t *
1635 r4mkopenlist(mntinfo4_t *mi)
1636 {
1637 nfs4_opinst_t *reopenlist, *rep;
1638 rnode4_t *rp;
1639 vnode_t *vp;
1640 vfs_t *vfsp = mi->mi_vfsp;
1641 int numosp;
1642 nfs4_open_stream_t *osp;
1643 int index;
1644 open_delegation_type4 dtype;
1645 int hold_vnode;
1646
1647 reopenlist = NULL;
1648
1649 for (index = 0; index < rtable4size; index++) {
1650 rw_enter(&rtable4[index].r_lock, RW_READER);
1651 for (rp = rtable4[index].r_hashf;
1652 rp != (rnode4_t *)(&rtable4[index]);
1653 rp = rp->r_hashf) {
1654
1655 vp = RTOV4(rp);
1656 if (vp->v_vfsp != vfsp)
1657 continue;
1658 hold_vnode = 0;
1659
1660 mutex_enter(&rp->r_os_lock);
1661
1662 /* Count the number of valid open_streams of the file */
1663 numosp = 0;
1664 for (osp = list_head(&rp->r_open_streams); osp != NULL;
1665 osp = list_next(&rp->r_open_streams, osp)) {
1666 mutex_enter(&osp->os_sync_lock);
1667 if (osp->os_valid && !osp->os_pending_close)
1668 numosp++;
1669 mutex_exit(&osp->os_sync_lock);
1670 }
1671
1672 /* Fill in the valid open streams per vp */
1673 if (numosp > 0) {
1674 int j;
1675
1676 hold_vnode = 1;
1677
1678 /*
1679 * Add a new open instance to the list
1680 */
1681 rep = kmem_zalloc(sizeof (*reopenlist),
1682 KM_SLEEP);
1683 rep->re_next = reopenlist;
1684 reopenlist = rep;
1685
1686 rep->re_vp = vp;
1687 rep->re_osp = kmem_zalloc(
1688 numosp * sizeof (*(rep->re_osp)),
1689 KM_SLEEP);
1690 rep->re_numosp = numosp;
1691
1692 j = 0;
1693 for (osp = list_head(&rp->r_open_streams);
1694 osp != NULL;
1695 osp = list_next(&rp->r_open_streams, osp)) {
1696
1697 mutex_enter(&osp->os_sync_lock);
1698 if (osp->os_valid &&
1699 !osp->os_pending_close) {
1700 osp->os_ref_count++;
1701 rep->re_osp[j] = osp;
1702 j++;
1703 }
1704 mutex_exit(&osp->os_sync_lock);
1705 }
1706 /*
1707 * Assuming valid osp(s) stays valid between
1708 * the time obtaining j and numosp.
1709 */
1710 ASSERT(j == numosp);
1711 }
1712
1713 mutex_exit(&rp->r_os_lock);
1714 /* do this here to keep v_lock > r_os_lock */
1715 if (hold_vnode)
1716 VN_HOLD(vp);
1717 mutex_enter(&rp->r_statev4_lock);
1718 if (rp->r_deleg_type != OPEN_DELEGATE_NONE) {
1719 /*
1720 * If this rnode holds a delegation,
1721 * but if there are no valid open streams,
1722 * then just discard the delegation
1723 * without doing delegreturn.
1724 */
1725 if (numosp > 0)
1726 rp->r_deleg_needs_recovery =
1727 rp->r_deleg_type;
1728 }
1729 /* Save the delegation type for use outside the lock */
1730 dtype = rp->r_deleg_type;
1731 mutex_exit(&rp->r_statev4_lock);
1732
1733 /*
1734 * If we have a delegation then get rid of it.
1735 * We've set rp->r_deleg_needs_recovery so we have
1736 * enough information to recover.
1737 */
1738 if (dtype != OPEN_DELEGATE_NONE) {
1739 (void) nfs4delegreturn(rp, NFS4_DR_DISCARD);
1740 }
1741 }
1742 rw_exit(&rtable4[index].r_lock);
1743 }
1744 return (reopenlist);
1745 }
1746
1747 /*
1748 * Given a filesystem id, check to see if any rnodes
1749 * within this fsid reside in the rnode cache, other
1750 * than one we know about.
1751 *
1752 * Return 1 if an rnode is found, 0 otherwise
1753 */
1754 int
1755 r4find_by_fsid(mntinfo4_t *mi, fattr4_fsid *moved_fsid)
1756 {
1757 rnode4_t *rp;
1758 vnode_t *vp;
1759 vfs_t *vfsp = mi->mi_vfsp;
1760 fattr4_fsid *fsid;
1761 int index, found = 0;
1762
1763 for (index = 0; index < rtable4size; index++) {
1764 rw_enter(&rtable4[index].r_lock, RW_READER);
1765 for (rp = rtable4[index].r_hashf;
1766 rp != (rnode4_t *)(&rtable4[index]);
1767 rp = rp->r_hashf) {
1768
1769 vp = RTOV4(rp);
1770 if (vp->v_vfsp != vfsp)
1771 continue;
1772
1773 /*
1774 * XXX there might be a case where a
1775 * replicated fs may have the same fsid
1776 * across two different servers. This
1777 * check isn't good enough in that case
1778 */
1779 fsid = &rp->r_srv_fsid;
1780 if (FATTR4_FSID_EQ(moved_fsid, fsid)) {
1781 found = 1;
1782 break;
1783 }
1784 }
1785 rw_exit(&rtable4[index].r_lock);
1786
1787 if (found)
1788 break;
1789 }
1790 return (found);
1791 }
1792
1793 /*
1794 * Release the list of open instance references.
1795 */
1796
1797 void
1798 r4releopenlist(nfs4_opinst_t *reopenp)
1799 {
1800 nfs4_opinst_t *rep, *next;
1801 int i;
1802
1803 for (rep = reopenp; rep; rep = next) {
1804 next = rep->re_next;
1805
1806 for (i = 0; i < rep->re_numosp; i++)
1807 open_stream_rele(rep->re_osp[i], VTOR4(rep->re_vp));
1808
1809 VN_RELE(rep->re_vp);
1810 kmem_free(rep->re_osp,
1811 rep->re_numosp * sizeof (*(rep->re_osp)));
1812
1813 kmem_free(rep, sizeof (*rep));
1814 }
1815 }
1816
1817 int
1818 nfs4_rnode_init(void)
1819 {
1820 ulong_t nrnode4_max;
1821 int i;
1822
1823 /*
1824 * Compute the size of the rnode4 hash table
1825 */
1826 if (nrnode <= 0)
1827 nrnode = ncsize;
1828 nrnode4_max =
1829 (ulong_t)((kmem_maxavail() >> 2) / sizeof (struct rnode4));
1830 if (nrnode > nrnode4_max || (nrnode == 0 && ncsize == 0)) {
1831 zcmn_err(GLOBAL_ZONEID, CE_NOTE,
1832 "!setting nrnode to max value of %ld", nrnode4_max);
1833 nrnode = nrnode4_max;
1834 }
1835 rtable4size = 1 << highbit(nrnode / rnode4_hashlen);
1836 rtable4mask = rtable4size - 1;
1837
1838 /*
1839 * Allocate and initialize the hash buckets
1840 */
1841 rtable4 = kmem_alloc(rtable4size * sizeof (*rtable4), KM_SLEEP);
1842 for (i = 0; i < rtable4size; i++) {
1843 rtable4[i].r_hashf = (rnode4_t *)(&rtable4[i]);
1844 rtable4[i].r_hashb = (rnode4_t *)(&rtable4[i]);
1845 rw_init(&rtable4[i].r_lock, NULL, RW_DEFAULT, NULL);
1846 }
1847
1848 rnode4_cache = kmem_cache_create("rnode4_cache", sizeof (rnode4_t),
1849 0, NULL, NULL, nfs4_reclaim, NULL, NULL, 0);
1850
1851 return (0);
1852 }
1853
1854 int
1855 nfs4_rnode_fini(void)
1856 {
1857 int i;
1858
1859 /*
1860 * Deallocate the rnode hash queues
1861 */
1862 kmem_cache_destroy(rnode4_cache);
1863
1864 for (i = 0; i < rtable4size; i++)
1865 rw_destroy(&rtable4[i].r_lock);
1866
1867 kmem_free(rtable4, rtable4size * sizeof (*rtable4));
1868
1869 return (0);
1870 }
1871
1872 /*
1873 * Return non-zero if the given filehandle refers to the root filehandle
1874 * for the given rnode.
1875 */
1876
1877 static int
1878 isrootfh(nfs4_sharedfh_t *fh, rnode4_t *rp)
1879 {
1880 int isroot;
1881
1882 isroot = 0;
1883 if (SFH4_SAME(VTOMI4(RTOV4(rp))->mi_rootfh, fh))
1884 isroot = 1;
1885
1886 return (isroot);
1887 }
1888
1889 /*
1890 * The r4_stub_* routines assume that the rnode is newly activated, and
1891 * that the caller either holds the hash bucket r_lock for this rnode as
1892 * RW_WRITER, or holds r_statelock.
1893 */
1894 static void
1895 r4_stub_set(rnode4_t *rp, nfs4_stub_type_t type)
1896 {
1897 vnode_t *vp = RTOV4(rp);
1898 krwlock_t *hash_lock = &rp->r_hashq->r_lock;
1899
1900 ASSERT(RW_WRITE_HELD(hash_lock) || MUTEX_HELD(&rp->r_statelock));
1901
1902 rp->r_stub_type = type;
1903
1904 /*
1905 * Safely switch this vnode to the trigger vnodeops.
1906 *
1907 * Currently, we don't ever switch a trigger vnode back to using
1908 * "regular" v4 vnodeops. NFS4_STUB_NONE is only used to note that
1909 * a new v4 object is not a trigger, and it will already have the
1910 * correct v4 vnodeops by default. So, no "else" case required here.
1911 */
1912 if (type != NFS4_STUB_NONE)
1913 vn_setops(vp, nfs4_trigger_vnodeops);
1914 }
1915
1916 void
1917 r4_stub_mirrormount(rnode4_t *rp)
1918 {
1919 r4_stub_set(rp, NFS4_STUB_MIRRORMOUNT);
1920 }
1921
1922 void
1923 r4_stub_referral(rnode4_t *rp)
1924 {
1925 DTRACE_PROBE1(nfs4clnt__func__referral__moved,
1926 vnode_t *, RTOV4(rp));
1927 r4_stub_set(rp, NFS4_STUB_REFERRAL);
1928 }
1929
1930 void
1931 r4_stub_none(rnode4_t *rp)
1932 {
1933 r4_stub_set(rp, NFS4_STUB_NONE);
1934 }
1935
1936 #ifdef DEBUG
1937
1938 /*
1939 * Look in the rnode table for other rnodes that have the same filehandle.
1940 * Assume the lock is held for the hash chain of checkrp
1941 */
1942
1943 static void
1944 r4_dup_check(rnode4_t *checkrp, vfs_t *vfsp)
1945 {
1946 rnode4_t *rp;
1947 vnode_t *tvp;
1948 nfs4_fhandle_t fh, fh2;
1949 int index;
1950
1951 if (!r4_check_for_dups)
1952 return;
1953
1954 ASSERT(RW_LOCK_HELD(&checkrp->r_hashq->r_lock));
1955
1956 sfh4_copyval(checkrp->r_fh, &fh);
1957
1958 for (index = 0; index < rtable4size; index++) {
1959
1960 if (&rtable4[index] != checkrp->r_hashq)
1961 rw_enter(&rtable4[index].r_lock, RW_READER);
1962
1963 for (rp = rtable4[index].r_hashf;
1964 rp != (rnode4_t *)(&rtable4[index]);
1965 rp = rp->r_hashf) {
1966
1967 if (rp == checkrp)
1968 continue;
1969
1970 tvp = RTOV4(rp);
1971 if (tvp->v_vfsp != vfsp)
1972 continue;
1973
1974 sfh4_copyval(rp->r_fh, &fh2);
1975 if (nfs4cmpfhandle(&fh, &fh2) == 0) {
1976 cmn_err(CE_PANIC, "rnodes with same fs, fh "
1977 "(%p, %p)", (void *)checkrp, (void *)rp);
1978 }
1979 }
1980
1981 if (&rtable4[index] != checkrp->r_hashq)
1982 rw_exit(&rtable4[index].r_lock);
1983 }
1984 }
1985
1986 #endif /* DEBUG */