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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 */
25
26 /* Portions Copyright 2007 Jeremy Teo */
27
28 #ifdef _KERNEL
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/time.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
35 #include <sys/mntent.h>
36 #include <sys/mkdev.h>
37 #include <sys/u8_textprep.h>
38 #include <sys/dsl_dataset.h>
39 #include <sys/vfs.h>
40 #include <sys/vfs_opreg.h>
41 #include <sys/vnode.h>
42 #include <sys/file.h>
43 #include <sys/kmem.h>
44 #include <sys/errno.h>
45 #include <sys/unistd.h>
46 #include <sys/mode.h>
47 #include <sys/atomic.h>
48 #include <vm/pvn.h>
49 #include "fs/fs_subr.h"
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_acl.h>
52 #include <sys/zfs_ioctl.h>
53 #include <sys/zfs_rlock.h>
54 #include <sys/zfs_fuid.h>
55 #include <sys/dnode.h>
56 #include <sys/fs/zfs.h>
57 #include <sys/kidmap.h>
58 #endif /* _KERNEL */
59
60 #include <sys/dmu.h>
61 #include <sys/refcount.h>
62 #include <sys/stat.h>
63 #include <sys/zap.h>
64 #include <sys/zfs_znode.h>
65 #include <sys/sa.h>
66 #include <sys/zfs_sa.h>
67 #include <sys/zfs_stat.h>
68
69 #include "zfs_prop.h"
70 #include "zfs_comutil.h"
71
72 /*
73 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
74 * turned on when DEBUG is also defined.
75 */
76 #ifdef DEBUG
77 #define ZNODE_STATS
78 #endif /* DEBUG */
79
80 #ifdef ZNODE_STATS
81 #define ZNODE_STAT_ADD(stat) ((stat)++)
82 #else
83 #define ZNODE_STAT_ADD(stat) /* nothing */
84 #endif /* ZNODE_STATS */
85
86 /*
87 * Functions needed for userland (ie: libzpool) are not put under
88 * #ifdef_KERNEL; the rest of the functions have dependencies
89 * (such as VFS logic) that will not compile easily in userland.
90 */
91 #ifdef _KERNEL
92 /*
93 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to
94 * be freed before it can be safely accessed.
95 */
96 krwlock_t zfsvfs_lock;
97
98 static kmem_cache_t *znode_cache = NULL;
99
100 /*ARGSUSED*/
101 static void
102 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
103 {
104 /*
105 * We should never drop all dbuf refs without first clearing
106 * the eviction callback.
107 */
108 panic("evicting znode %p\n", user_ptr);
109 }
110
111 /*ARGSUSED*/
112 static int
113 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
114 {
115 znode_t *zp = buf;
116
117 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
118
119 zp->z_vnode = vn_alloc(kmflags);
120 if (zp->z_vnode == NULL) {
121 return (-1);
122 }
123 ZTOV(zp)->v_data = zp;
124
125 list_link_init(&zp->z_link_node);
126
127 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
128 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
129 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
130 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
131
132 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
133 avl_create(&zp->z_range_avl, zfs_range_compare,
134 sizeof (rl_t), offsetof(rl_t, r_node));
135
136 zp->z_dirlocks = NULL;
137 zp->z_acl_cached = NULL;
138 zp->z_moved = 0;
139 return (0);
140 }
141
142 /*ARGSUSED*/
143 static void
144 zfs_znode_cache_destructor(void *buf, void *arg)
145 {
146 znode_t *zp = buf;
147
148 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
149 ASSERT(ZTOV(zp)->v_data == zp);
150 vn_free(ZTOV(zp));
151 ASSERT(!list_link_active(&zp->z_link_node));
152 mutex_destroy(&zp->z_lock);
153 rw_destroy(&zp->z_parent_lock);
154 rw_destroy(&zp->z_name_lock);
155 mutex_destroy(&zp->z_acl_lock);
156 avl_destroy(&zp->z_range_avl);
157 mutex_destroy(&zp->z_range_lock);
158
159 ASSERT(zp->z_dirlocks == NULL);
160 ASSERT(zp->z_acl_cached == NULL);
161 }
162
163 #ifdef ZNODE_STATS
164 static struct {
165 uint64_t zms_zfsvfs_invalid;
166 uint64_t zms_zfsvfs_recheck1;
167 uint64_t zms_zfsvfs_unmounted;
168 uint64_t zms_zfsvfs_recheck2;
169 uint64_t zms_obj_held;
170 uint64_t zms_vnode_locked;
171 uint64_t zms_not_only_dnlc;
172 } znode_move_stats;
173 #endif /* ZNODE_STATS */
174
175 static void
176 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
177 {
178 vnode_t *vp;
179
180 /* Copy fields. */
181 nzp->z_zfsvfs = ozp->z_zfsvfs;
182
183 /* Swap vnodes. */
184 vp = nzp->z_vnode;
185 nzp->z_vnode = ozp->z_vnode;
186 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
187 ZTOV(ozp)->v_data = ozp;
188 ZTOV(nzp)->v_data = nzp;
189
190 nzp->z_id = ozp->z_id;
191 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
192 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0);
193 nzp->z_unlinked = ozp->z_unlinked;
194 nzp->z_atime_dirty = ozp->z_atime_dirty;
195 nzp->z_zn_prefetch = ozp->z_zn_prefetch;
196 nzp->z_blksz = ozp->z_blksz;
197 nzp->z_seq = ozp->z_seq;
198 nzp->z_mapcnt = ozp->z_mapcnt;
199 nzp->z_gen = ozp->z_gen;
200 nzp->z_sync_cnt = ozp->z_sync_cnt;
201 nzp->z_is_sa = ozp->z_is_sa;
202 nzp->z_sa_hdl = ozp->z_sa_hdl;
203 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
204 nzp->z_links = ozp->z_links;
205 nzp->z_size = ozp->z_size;
206 nzp->z_pflags = ozp->z_pflags;
207 nzp->z_uid = ozp->z_uid;
208 nzp->z_gid = ozp->z_gid;
209 nzp->z_mode = ozp->z_mode;
210
211 /*
212 * Since this is just an idle znode and kmem is already dealing with
213 * memory pressure, release any cached ACL.
214 */
215 if (ozp->z_acl_cached) {
216 zfs_acl_free(ozp->z_acl_cached);
217 ozp->z_acl_cached = NULL;
218 }
219
220 sa_set_userp(nzp->z_sa_hdl, nzp);
221
222 /*
223 * Invalidate the original znode by clearing fields that provide a
224 * pointer back to the znode. Set the low bit of the vfs pointer to
225 * ensure that zfs_znode_move() recognizes the znode as invalid in any
226 * subsequent callback.
227 */
228 ozp->z_sa_hdl = NULL;
229 POINTER_INVALIDATE(&ozp->z_zfsvfs);
230
231 /*
232 * Mark the znode.
233 */
234 nzp->z_moved = 1;
235 ozp->z_moved = (uint8_t)-1;
236 }
237
238 /*ARGSUSED*/
239 static kmem_cbrc_t
240 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
241 {
242 znode_t *ozp = buf, *nzp = newbuf;
243 zfsvfs_t *zfsvfs;
244 vnode_t *vp;
245
246 /*
247 * The znode is on the file system's list of known znodes if the vfs
248 * pointer is valid. We set the low bit of the vfs pointer when freeing
249 * the znode to invalidate it, and the memory patterns written by kmem
250 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
251 * created znode sets the vfs pointer last of all to indicate that the
252 * znode is known and in a valid state to be moved by this function.
253 */
254 zfsvfs = ozp->z_zfsvfs;
255 if (!POINTER_IS_VALID(zfsvfs)) {
256 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
257 return (KMEM_CBRC_DONT_KNOW);
258 }
259
260 /*
261 * Close a small window in which it's possible that the filesystem could
262 * be unmounted and freed, and zfsvfs, though valid in the previous
263 * statement, could point to unrelated memory by the time we try to
264 * prevent the filesystem from being unmounted.
265 */
266 rw_enter(&zfsvfs_lock, RW_WRITER);
267 if (zfsvfs != ozp->z_zfsvfs) {
268 rw_exit(&zfsvfs_lock);
269 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
270 return (KMEM_CBRC_DONT_KNOW);
271 }
272
273 /*
274 * If the znode is still valid, then so is the file system. We know that
275 * no valid file system can be freed while we hold zfsvfs_lock, so we
276 * can safely ensure that the filesystem is not and will not be
277 * unmounted. The next statement is equivalent to ZFS_ENTER().
278 */
279 rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
280 if (zfsvfs->z_unmounted) {
281 ZFS_EXIT(zfsvfs);
282 rw_exit(&zfsvfs_lock);
283 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
284 return (KMEM_CBRC_DONT_KNOW);
285 }
286 rw_exit(&zfsvfs_lock);
287
288 mutex_enter(&zfsvfs->z_znodes_lock);
289 /*
290 * Recheck the vfs pointer in case the znode was removed just before
291 * acquiring the lock.
292 */
293 if (zfsvfs != ozp->z_zfsvfs) {
294 mutex_exit(&zfsvfs->z_znodes_lock);
295 ZFS_EXIT(zfsvfs);
296 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
297 return (KMEM_CBRC_DONT_KNOW);
298 }
299
300 /*
301 * At this point we know that as long as we hold z_znodes_lock, the
302 * znode cannot be freed and fields within the znode can be safely
303 * accessed. Now, prevent a race with zfs_zget().
304 */
305 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
306 mutex_exit(&zfsvfs->z_znodes_lock);
307 ZFS_EXIT(zfsvfs);
308 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
309 return (KMEM_CBRC_LATER);
310 }
311
312 vp = ZTOV(ozp);
313 if (mutex_tryenter(&vp->v_lock) == 0) {
314 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
315 mutex_exit(&zfsvfs->z_znodes_lock);
316 ZFS_EXIT(zfsvfs);
317 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
318 return (KMEM_CBRC_LATER);
319 }
320
321 /* Only move znodes that are referenced _only_ by the DNLC. */
322 if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
323 mutex_exit(&vp->v_lock);
324 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
325 mutex_exit(&zfsvfs->z_znodes_lock);
326 ZFS_EXIT(zfsvfs);
327 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
328 return (KMEM_CBRC_LATER);
329 }
330
331 /*
332 * The znode is known and in a valid state to move. We're holding the
333 * locks needed to execute the critical section.
334 */
335 zfs_znode_move_impl(ozp, nzp);
336 mutex_exit(&vp->v_lock);
337 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
338
339 list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
340 mutex_exit(&zfsvfs->z_znodes_lock);
341 ZFS_EXIT(zfsvfs);
342
343 return (KMEM_CBRC_YES);
344 }
345
346 void
347 zfs_znode_init(void)
348 {
349 /*
350 * Initialize zcache
351 */
352 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
353 ASSERT(znode_cache == NULL);
354 znode_cache = kmem_cache_create("zfs_znode_cache",
355 sizeof (znode_t), 0, zfs_znode_cache_constructor,
356 zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
357 kmem_cache_set_move(znode_cache, zfs_znode_move);
358 }
359
360 void
361 zfs_znode_fini(void)
362 {
363 /*
364 * Cleanup vfs & vnode ops
365 */
366 zfs_remove_op_tables();
367
368 /*
369 * Cleanup zcache
370 */
371 if (znode_cache)
372 kmem_cache_destroy(znode_cache);
373 znode_cache = NULL;
374 rw_destroy(&zfsvfs_lock);
375 }
376
377 struct vnodeops *zfs_dvnodeops;
378 struct vnodeops *zfs_fvnodeops;
379 struct vnodeops *zfs_symvnodeops;
380 struct vnodeops *zfs_xdvnodeops;
381 struct vnodeops *zfs_evnodeops;
382 struct vnodeops *zfs_sharevnodeops;
383
384 void
385 zfs_remove_op_tables()
386 {
387 /*
388 * Remove vfs ops
389 */
390 ASSERT(zfsfstype);
391 (void) vfs_freevfsops_by_type(zfsfstype);
392 zfsfstype = 0;
393
394 /*
395 * Remove vnode ops
396 */
397 if (zfs_dvnodeops)
398 vn_freevnodeops(zfs_dvnodeops);
399 if (zfs_fvnodeops)
400 vn_freevnodeops(zfs_fvnodeops);
401 if (zfs_symvnodeops)
402 vn_freevnodeops(zfs_symvnodeops);
403 if (zfs_xdvnodeops)
404 vn_freevnodeops(zfs_xdvnodeops);
405 if (zfs_evnodeops)
406 vn_freevnodeops(zfs_evnodeops);
407 if (zfs_sharevnodeops)
408 vn_freevnodeops(zfs_sharevnodeops);
409
410 zfs_dvnodeops = NULL;
411 zfs_fvnodeops = NULL;
412 zfs_symvnodeops = NULL;
413 zfs_xdvnodeops = NULL;
414 zfs_evnodeops = NULL;
415 zfs_sharevnodeops = NULL;
416 }
417
418 extern const fs_operation_def_t zfs_dvnodeops_template[];
419 extern const fs_operation_def_t zfs_fvnodeops_template[];
420 extern const fs_operation_def_t zfs_xdvnodeops_template[];
421 extern const fs_operation_def_t zfs_symvnodeops_template[];
422 extern const fs_operation_def_t zfs_evnodeops_template[];
423 extern const fs_operation_def_t zfs_sharevnodeops_template[];
424
425 int
426 zfs_create_op_tables()
427 {
428 int error;
429
430 /*
431 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
432 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
433 * In this case we just return as the ops vectors are already set up.
434 */
435 if (zfs_dvnodeops)
436 return (0);
437
438 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
439 &zfs_dvnodeops);
440 if (error)
441 return (error);
442
443 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
444 &zfs_fvnodeops);
445 if (error)
446 return (error);
447
448 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
449 &zfs_symvnodeops);
450 if (error)
451 return (error);
452
453 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
454 &zfs_xdvnodeops);
455 if (error)
456 return (error);
457
458 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
459 &zfs_evnodeops);
460 if (error)
461 return (error);
462
463 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
464 &zfs_sharevnodeops);
465
466 return (error);
467 }
468
469 int
470 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
471 {
472 zfs_acl_ids_t acl_ids;
473 vattr_t vattr;
474 znode_t *sharezp;
475 vnode_t *vp;
476 znode_t *zp;
477 int error;
478
479 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
480 vattr.va_type = VDIR;
481 vattr.va_mode = S_IFDIR|0555;
482 vattr.va_uid = crgetuid(kcred);
483 vattr.va_gid = crgetgid(kcred);
484
485 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
486 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
487 sharezp->z_moved = 0;
488 sharezp->z_unlinked = 0;
489 sharezp->z_atime_dirty = 0;
490 sharezp->z_zfsvfs = zfsvfs;
491 sharezp->z_is_sa = zfsvfs->z_use_sa;
492
493 vp = ZTOV(sharezp);
494 vn_reinit(vp);
495 vp->v_type = VDIR;
496
497 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
498 kcred, NULL, &acl_ids));
499 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
500 ASSERT3P(zp, ==, sharezp);
501 ASSERT(!vn_in_dnlc(ZTOV(sharezp))); /* not valid to move */
502 POINTER_INVALIDATE(&sharezp->z_zfsvfs);
503 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
504 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
505 zfsvfs->z_shares_dir = sharezp->z_id;
506
507 zfs_acl_ids_free(&acl_ids);
508 ZTOV(sharezp)->v_count = 0;
509 sa_handle_destroy(sharezp->z_sa_hdl);
510 kmem_cache_free(znode_cache, sharezp);
511
512 return (error);
513 }
514
515 /*
516 * define a couple of values we need available
517 * for both 64 and 32 bit environments.
518 */
519 #ifndef NBITSMINOR64
520 #define NBITSMINOR64 32
521 #endif
522 #ifndef MAXMAJ64
523 #define MAXMAJ64 0xffffffffUL
524 #endif
525 #ifndef MAXMIN64
526 #define MAXMIN64 0xffffffffUL
527 #endif
528
529 /*
530 * Create special expldev for ZFS private use.
531 * Can't use standard expldev since it doesn't do
532 * what we want. The standard expldev() takes a
533 * dev32_t in LP64 and expands it to a long dev_t.
534 * We need an interface that takes a dev32_t in ILP32
535 * and expands it to a long dev_t.
536 */
537 static uint64_t
538 zfs_expldev(dev_t dev)
539 {
540 #ifndef _LP64
541 major_t major = (major_t)dev >> NBITSMINOR32 & MAXMAJ32;
542 return (((uint64_t)major << NBITSMINOR64) |
543 ((minor_t)dev & MAXMIN32));
544 #else
545 return (dev);
546 #endif
547 }
548
549 /*
550 * Special cmpldev for ZFS private use.
551 * Can't use standard cmpldev since it takes
552 * a long dev_t and compresses it to dev32_t in
553 * LP64. We need to do a compaction of a long dev_t
554 * to a dev32_t in ILP32.
555 */
556 dev_t
557 zfs_cmpldev(uint64_t dev)
558 {
559 #ifndef _LP64
560 minor_t minor = (minor_t)dev & MAXMIN64;
561 major_t major = (major_t)(dev >> NBITSMINOR64) & MAXMAJ64;
562
563 if (major > MAXMAJ32 || minor > MAXMIN32)
564 return (NODEV32);
565
566 return (((dev32_t)major << NBITSMINOR32) | minor);
567 #else
568 return (dev);
569 #endif
570 }
571
572 static void
573 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
574 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
575 {
576 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
577 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
578
579 mutex_enter(&zp->z_lock);
580
581 ASSERT(zp->z_sa_hdl == NULL);
582 ASSERT(zp->z_acl_cached == NULL);
583 if (sa_hdl == NULL) {
584 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
585 SA_HDL_SHARED, &zp->z_sa_hdl));
586 } else {
587 zp->z_sa_hdl = sa_hdl;
588 sa_set_userp(sa_hdl, zp);
589 }
590
591 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
592
593 /*
594 * Slap on VROOT if we are the root znode
595 */
596 if (zp->z_id == zfsvfs->z_root)
597 ZTOV(zp)->v_flag |= VROOT;
598
599 mutex_exit(&zp->z_lock);
600 vn_exists(ZTOV(zp));
601 }
602
603 void
604 zfs_znode_dmu_fini(znode_t *zp)
605 {
606 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
607 zp->z_unlinked ||
608 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
609
610 sa_handle_destroy(zp->z_sa_hdl);
611 zp->z_sa_hdl = NULL;
612 }
613
614 /*
615 * Construct a new znode/vnode and intialize.
616 *
617 * This does not do a call to dmu_set_user() that is
618 * up to the caller to do, in case you don't want to
619 * return the znode
620 */
621 static znode_t *
622 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
623 dmu_object_type_t obj_type, sa_handle_t *hdl)
624 {
625 znode_t *zp;
626 vnode_t *vp;
627 uint64_t mode;
628 uint64_t parent;
629 sa_bulk_attr_t bulk[9];
630 int count = 0;
631
632 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
633
634 ASSERT(zp->z_dirlocks == NULL);
635 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
636 zp->z_moved = 0;
637
638 /*
639 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
640 * the zfs_znode_move() callback.
641 */
642 zp->z_sa_hdl = NULL;
643 zp->z_unlinked = 0;
644 zp->z_atime_dirty = 0;
645 zp->z_mapcnt = 0;
646 zp->z_id = db->db_object;
647 zp->z_blksz = blksz;
648 zp->z_seq = 0x7A4653;
649 zp->z_sync_cnt = 0;
650
651 vp = ZTOV(zp);
652 vn_reinit(vp);
653
654 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
655
656 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
657 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
658 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
659 &zp->z_size, 8);
660 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
661 &zp->z_links, 8);
662 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
663 &zp->z_pflags, 8);
664 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
665 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
666 &zp->z_atime, 16);
667 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
668 &zp->z_uid, 8);
669 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
670 &zp->z_gid, 8);
671
672 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) {
673 if (hdl == NULL)
674 sa_handle_destroy(zp->z_sa_hdl);
675 kmem_cache_free(znode_cache, zp);
676 return (NULL);
677 }
678
679 zp->z_mode = mode;
680 vp->v_vfsp = zfsvfs->z_parent->z_vfs;
681
682 vp->v_type = IFTOVT((mode_t)mode);
683
684 switch (vp->v_type) {
685 case VDIR:
686 if (zp->z_pflags & ZFS_XATTR) {
687 vn_setops(vp, zfs_xdvnodeops);
688 vp->v_flag |= V_XATTRDIR;
689 } else {
690 vn_setops(vp, zfs_dvnodeops);
691 }
692 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
693 break;
694 case VBLK:
695 case VCHR:
696 {
697 uint64_t rdev;
698 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
699 &rdev, sizeof (rdev)) == 0);
700
701 vp->v_rdev = zfs_cmpldev(rdev);
702 }
703 /*FALLTHROUGH*/
704 case VFIFO:
705 case VSOCK:
706 case VDOOR:
707 vn_setops(vp, zfs_fvnodeops);
708 break;
709 case VREG:
710 vp->v_flag |= VMODSORT;
711 if (parent == zfsvfs->z_shares_dir) {
712 ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
713 vn_setops(vp, zfs_sharevnodeops);
714 } else {
715 vn_setops(vp, zfs_fvnodeops);
716 }
717 break;
718 case VLNK:
719 vn_setops(vp, zfs_symvnodeops);
720 break;
721 default:
722 vn_setops(vp, zfs_evnodeops);
723 break;
724 }
725
726 mutex_enter(&zfsvfs->z_znodes_lock);
727 list_insert_tail(&zfsvfs->z_all_znodes, zp);
728 membar_producer();
729 /*
730 * Everything else must be valid before assigning z_zfsvfs makes the
731 * znode eligible for zfs_znode_move().
732 */
733 zp->z_zfsvfs = zfsvfs;
734 mutex_exit(&zfsvfs->z_znodes_lock);
735
736 VFS_HOLD(zfsvfs->z_vfs);
737 return (zp);
738 }
739
740 static uint64_t empty_xattr;
741 static uint64_t pad[4];
742 static zfs_acl_phys_t acl_phys;
743 /*
744 * Create a new DMU object to hold a zfs znode.
745 *
746 * IN: dzp - parent directory for new znode
747 * vap - file attributes for new znode
748 * tx - dmu transaction id for zap operations
749 * cr - credentials of caller
750 * flag - flags:
751 * IS_ROOT_NODE - new object will be root
752 * IS_XATTR - new object is an attribute
753 * bonuslen - length of bonus buffer
754 * setaclp - File/Dir initial ACL
755 * fuidp - Tracks fuid allocation.
756 *
757 * OUT: zpp - allocated znode
758 *
759 */
760 void
761 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
762 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
763 {
764 uint64_t crtime[2], atime[2], mtime[2], ctime[2];
765 uint64_t mode, size, links, parent, pflags;
766 uint64_t dzp_pflags = 0;
767 uint64_t rdev = 0;
768 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
769 dmu_buf_t *db;
770 timestruc_t now;
771 uint64_t gen, obj;
772 int err;
773 int bonuslen;
774 sa_handle_t *sa_hdl;
775 dmu_object_type_t obj_type;
776 sa_bulk_attr_t sa_attrs[ZPL_END];
777 int cnt = 0;
778 zfs_acl_locator_cb_t locate = { 0 };
779
780 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
781
782 if (zfsvfs->z_replay) {
783 obj = vap->va_nodeid;
784 now = vap->va_ctime; /* see zfs_replay_create() */
785 gen = vap->va_nblocks; /* ditto */
786 } else {
787 obj = 0;
788 gethrestime(&now);
789 gen = dmu_tx_get_txg(tx);
790 }
791
792 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
793 bonuslen = (obj_type == DMU_OT_SA) ?
794 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
795
796 /*
797 * Create a new DMU object.
798 */
799 /*
800 * There's currently no mechanism for pre-reading the blocks that will
801 * be needed to allocate a new object, so we accept the small chance
802 * that there will be an i/o error and we will fail one of the
803 * assertions below.
804 */
805 if (vap->va_type == VDIR) {
806 if (zfsvfs->z_replay) {
807 err = zap_create_claim_norm(zfsvfs->z_os, obj,
808 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
809 obj_type, bonuslen, tx);
810 ASSERT0(err);
811 } else {
812 obj = zap_create_norm(zfsvfs->z_os,
813 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
814 obj_type, bonuslen, tx);
815 }
816 } else {
817 if (zfsvfs->z_replay) {
818 err = dmu_object_claim(zfsvfs->z_os, obj,
819 DMU_OT_PLAIN_FILE_CONTENTS, 0,
820 obj_type, bonuslen, tx);
821 ASSERT0(err);
822 } else {
823 obj = dmu_object_alloc(zfsvfs->z_os,
824 DMU_OT_PLAIN_FILE_CONTENTS, 0,
825 obj_type, bonuslen, tx);
826 }
827 }
828
829 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
830 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
831
832 /*
833 * If this is the root, fix up the half-initialized parent pointer
834 * to reference the just-allocated physical data area.
835 */
836 if (flag & IS_ROOT_NODE) {
837 dzp->z_id = obj;
838 } else {
839 dzp_pflags = dzp->z_pflags;
840 }
841
842 /*
843 * If parent is an xattr, so am I.
844 */
845 if (dzp_pflags & ZFS_XATTR) {
846 flag |= IS_XATTR;
847 }
848
849 if (zfsvfs->z_use_fuids)
850 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
851 else
852 pflags = 0;
853
854 if (vap->va_type == VDIR) {
855 size = 2; /* contents ("." and "..") */
856 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
857 } else {
858 size = links = 0;
859 }
860
861 if (vap->va_type == VBLK || vap->va_type == VCHR) {
862 rdev = zfs_expldev(vap->va_rdev);
863 }
864
865 parent = dzp->z_id;
866 mode = acl_ids->z_mode;
867 if (flag & IS_XATTR)
868 pflags |= ZFS_XATTR;
869
870 /*
871 * No execs denied will be deterimed when zfs_mode_compute() is called.
872 */
873 pflags |= acl_ids->z_aclp->z_hints &
874 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
875 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
876
877 ZFS_TIME_ENCODE(&now, crtime);
878 ZFS_TIME_ENCODE(&now, ctime);
879
880 if (vap->va_mask & AT_ATIME) {
881 ZFS_TIME_ENCODE(&vap->va_atime, atime);
882 } else {
883 ZFS_TIME_ENCODE(&now, atime);
884 }
885
886 if (vap->va_mask & AT_MTIME) {
887 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
888 } else {
889 ZFS_TIME_ENCODE(&now, mtime);
890 }
891
892 /* Now add in all of the "SA" attributes */
893 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
894 &sa_hdl));
895
896 /*
897 * Setup the array of attributes to be replaced/set on the new file
898 *
899 * order for DMU_OT_ZNODE is critical since it needs to be constructed
900 * in the old znode_phys_t format. Don't change this ordering
901 */
902
903 if (obj_type == DMU_OT_ZNODE) {
904 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
905 NULL, &atime, 16);
906 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
907 NULL, &mtime, 16);
908 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
909 NULL, &ctime, 16);
910 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
911 NULL, &crtime, 16);
912 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
913 NULL, &gen, 8);
914 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
915 NULL, &mode, 8);
916 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
917 NULL, &size, 8);
918 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
919 NULL, &parent, 8);
920 } else {
921 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
922 NULL, &mode, 8);
923 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
924 NULL, &size, 8);
925 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
926 NULL, &gen, 8);
927 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
928 &acl_ids->z_fuid, 8);
929 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
930 &acl_ids->z_fgid, 8);
931 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
932 NULL, &parent, 8);
933 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
934 NULL, &pflags, 8);
935 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
936 NULL, &atime, 16);
937 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
938 NULL, &mtime, 16);
939 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
940 NULL, &ctime, 16);
941 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
942 NULL, &crtime, 16);
943 }
944
945 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
946
947 if (obj_type == DMU_OT_ZNODE) {
948 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
949 &empty_xattr, 8);
950 }
951 if (obj_type == DMU_OT_ZNODE ||
952 (vap->va_type == VBLK || vap->va_type == VCHR)) {
953 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
954 NULL, &rdev, 8);
955
956 }
957 if (obj_type == DMU_OT_ZNODE) {
958 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
959 NULL, &pflags, 8);
960 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
961 &acl_ids->z_fuid, 8);
962 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
963 &acl_ids->z_fgid, 8);
964 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
965 sizeof (uint64_t) * 4);
966 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
967 &acl_phys, sizeof (zfs_acl_phys_t));
968 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
969 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
970 &acl_ids->z_aclp->z_acl_count, 8);
971 locate.cb_aclp = acl_ids->z_aclp;
972 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
973 zfs_acl_data_locator, &locate,
974 acl_ids->z_aclp->z_acl_bytes);
975 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
976 acl_ids->z_fuid, acl_ids->z_fgid);
977 }
978
979 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
980
981 if (!(flag & IS_ROOT_NODE)) {
982 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
983 ASSERT(*zpp != NULL);
984 } else {
985 /*
986 * If we are creating the root node, the "parent" we
987 * passed in is the znode for the root.
988 */
989 *zpp = dzp;
990
991 (*zpp)->z_sa_hdl = sa_hdl;
992 }
993
994 (*zpp)->z_pflags = pflags;
995 (*zpp)->z_mode = mode;
996
997 if (vap->va_mask & AT_XVATTR)
998 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
999
1000 if (obj_type == DMU_OT_ZNODE ||
1001 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1002 err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx);
1003 ASSERT0(err);
1004 }
1005 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1006 }
1007
1008 /*
1009 * Update in-core attributes. It is assumed the caller will be doing an
1010 * sa_bulk_update to push the changes out.
1011 */
1012 void
1013 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1014 {
1015 xoptattr_t *xoap;
1016
1017 xoap = xva_getxoptattr(xvap);
1018 ASSERT(xoap);
1019
1020 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1021 uint64_t times[2];
1022 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1023 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1024 ×, sizeof (times), tx);
1025 XVA_SET_RTN(xvap, XAT_CREATETIME);
1026 }
1027 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1028 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1029 zp->z_pflags, tx);
1030 XVA_SET_RTN(xvap, XAT_READONLY);
1031 }
1032 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1033 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1034 zp->z_pflags, tx);
1035 XVA_SET_RTN(xvap, XAT_HIDDEN);
1036 }
1037 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1038 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1039 zp->z_pflags, tx);
1040 XVA_SET_RTN(xvap, XAT_SYSTEM);
1041 }
1042 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1043 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1044 zp->z_pflags, tx);
1045 XVA_SET_RTN(xvap, XAT_ARCHIVE);
1046 }
1047 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1048 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1049 zp->z_pflags, tx);
1050 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1051 }
1052 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1053 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1054 zp->z_pflags, tx);
1055 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1056 }
1057 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1058 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1059 zp->z_pflags, tx);
1060 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1061 }
1062 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1063 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1064 zp->z_pflags, tx);
1065 XVA_SET_RTN(xvap, XAT_NODUMP);
1066 }
1067 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1068 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1069 zp->z_pflags, tx);
1070 XVA_SET_RTN(xvap, XAT_OPAQUE);
1071 }
1072 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1073 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1074 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1075 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1076 }
1077 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1078 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1079 zp->z_pflags, tx);
1080 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1081 }
1082 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1083 zfs_sa_set_scanstamp(zp, xvap, tx);
1084 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1085 }
1086 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1087 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1088 zp->z_pflags, tx);
1089 XVA_SET_RTN(xvap, XAT_REPARSE);
1090 }
1091 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1092 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1093 zp->z_pflags, tx);
1094 XVA_SET_RTN(xvap, XAT_OFFLINE);
1095 }
1096 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1097 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1098 zp->z_pflags, tx);
1099 XVA_SET_RTN(xvap, XAT_SPARSE);
1100 }
1101 }
1102
1103 int
1104 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1105 {
1106 dmu_object_info_t doi;
1107 dmu_buf_t *db;
1108 znode_t *zp;
1109 int err;
1110 sa_handle_t *hdl;
1111
1112 *zpp = NULL;
1113
1114 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1115
1116 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1117 if (err) {
1118 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1119 return (err);
1120 }
1121
1122 dmu_object_info_from_db(db, &doi);
1123 if (doi.doi_bonus_type != DMU_OT_SA &&
1124 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1125 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1126 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1127 sa_buf_rele(db, NULL);
1128 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1129 return (SET_ERROR(EINVAL));
1130 }
1131
1132 hdl = dmu_buf_get_user(db);
1133 if (hdl != NULL) {
1134 zp = sa_get_userdata(hdl);
1135
1136
1137 /*
1138 * Since "SA" does immediate eviction we
1139 * should never find a sa handle that doesn't
1140 * know about the znode.
1141 */
1142
1143 ASSERT3P(zp, !=, NULL);
1144
1145 mutex_enter(&zp->z_lock);
1146 ASSERT3U(zp->z_id, ==, obj_num);
1147 if (zp->z_unlinked) {
1148 err = SET_ERROR(ENOENT);
1149 } else {
1150 VN_HOLD(ZTOV(zp));
1151 *zpp = zp;
1152 err = 0;
1153 }
1154 sa_buf_rele(db, NULL);
1155 mutex_exit(&zp->z_lock);
1156 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1157 return (err);
1158 }
1159
1160 /*
1161 * Not found create new znode/vnode
1162 * but only if file exists.
1163 *
1164 * There is a small window where zfs_vget() could
1165 * find this object while a file create is still in
1166 * progress. This is checked for in zfs_znode_alloc()
1167 *
1168 * if zfs_znode_alloc() fails it will drop the hold on the
1169 * bonus buffer.
1170 */
1171 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1172 doi.doi_bonus_type, NULL);
1173 if (zp == NULL) {
1174 err = SET_ERROR(ENOENT);
1175 } else {
1176 *zpp = zp;
1177 }
1178 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1179 return (err);
1180 }
1181
1182 int
1183 zfs_rezget(znode_t *zp)
1184 {
1185 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1186 dmu_object_info_t doi;
1187 dmu_buf_t *db;
1188 uint64_t obj_num = zp->z_id;
1189 uint64_t mode;
1190 sa_bulk_attr_t bulk[8];
1191 int err;
1192 int count = 0;
1193 uint64_t gen;
1194
1195 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1196
1197 mutex_enter(&zp->z_acl_lock);
1198 if (zp->z_acl_cached) {
1199 zfs_acl_free(zp->z_acl_cached);
1200 zp->z_acl_cached = NULL;
1201 }
1202
1203 mutex_exit(&zp->z_acl_lock);
1204 ASSERT(zp->z_sa_hdl == NULL);
1205 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1206 if (err) {
1207 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1208 return (err);
1209 }
1210
1211 dmu_object_info_from_db(db, &doi);
1212 if (doi.doi_bonus_type != DMU_OT_SA &&
1213 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1214 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1215 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1216 sa_buf_rele(db, NULL);
1217 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1218 return (SET_ERROR(EINVAL));
1219 }
1220
1221 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1222
1223 /* reload cached values */
1224 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1225 &gen, sizeof (gen));
1226 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1227 &zp->z_size, sizeof (zp->z_size));
1228 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1229 &zp->z_links, sizeof (zp->z_links));
1230 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1231 &zp->z_pflags, sizeof (zp->z_pflags));
1232 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1233 &zp->z_atime, sizeof (zp->z_atime));
1234 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1235 &zp->z_uid, sizeof (zp->z_uid));
1236 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1237 &zp->z_gid, sizeof (zp->z_gid));
1238 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1239 &mode, sizeof (mode));
1240
1241 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1242 zfs_znode_dmu_fini(zp);
1243 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1244 return (SET_ERROR(EIO));
1245 }
1246
1247 zp->z_mode = mode;
1248
1249 if (gen != zp->z_gen) {
1250 zfs_znode_dmu_fini(zp);
1251 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1252 return (SET_ERROR(EIO));
1253 }
1254
1255 zp->z_unlinked = (zp->z_links == 0);
1256 zp->z_blksz = doi.doi_data_block_size;
1257
1258 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1259
1260 return (0);
1261 }
1262
1263 void
1264 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1265 {
1266 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1267 objset_t *os = zfsvfs->z_os;
1268 uint64_t obj = zp->z_id;
1269 uint64_t acl_obj = zfs_external_acl(zp);
1270
1271 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1272 if (acl_obj) {
1273 VERIFY(!zp->z_is_sa);
1274 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1275 }
1276 VERIFY(0 == dmu_object_free(os, obj, tx));
1277 zfs_znode_dmu_fini(zp);
1278 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1279 zfs_znode_free(zp);
1280 }
1281
1282 void
1283 zfs_zinactive(znode_t *zp)
1284 {
1285 vnode_t *vp = ZTOV(zp);
1286 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1287 uint64_t z_id = zp->z_id;
1288
1289 ASSERT(zp->z_sa_hdl);
1290
1291 /*
1292 * Don't allow a zfs_zget() while were trying to release this znode
1293 */
1294 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1295
1296 mutex_enter(&zp->z_lock);
1297 mutex_enter(&vp->v_lock);
1298 vp->v_count--;
1299 if (vp->v_count > 0 || vn_has_cached_data(vp)) {
1300 /*
1301 * If the hold count is greater than zero, somebody has
1302 * obtained a new reference on this znode while we were
1303 * processing it here, so we are done. If we still have
1304 * mapped pages then we are also done, since we don't
1305 * want to inactivate the znode until the pages get pushed.
1306 *
1307 * XXX - if vn_has_cached_data(vp) is true, but count == 0,
1308 * this seems like it would leave the znode hanging with
1309 * no chance to go inactive...
1310 */
1311 mutex_exit(&vp->v_lock);
1312 mutex_exit(&zp->z_lock);
1313 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1314 return;
1315 }
1316 mutex_exit(&vp->v_lock);
1317
1318 /*
1319 * If this was the last reference to a file with no links,
1320 * remove the file from the file system.
1321 */
1322 if (zp->z_unlinked) {
1323 mutex_exit(&zp->z_lock);
1324 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1325 zfs_rmnode(zp);
1326 return;
1327 }
1328
1329 mutex_exit(&zp->z_lock);
1330 zfs_znode_dmu_fini(zp);
1331 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1332 zfs_znode_free(zp);
1333 }
1334
1335 void
1336 zfs_znode_free(znode_t *zp)
1337 {
1338 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1339
1340 vn_invalid(ZTOV(zp));
1341
1342 ASSERT(ZTOV(zp)->v_count == 0);
1343
1344 mutex_enter(&zfsvfs->z_znodes_lock);
1345 POINTER_INVALIDATE(&zp->z_zfsvfs);
1346 list_remove(&zfsvfs->z_all_znodes, zp);
1347 mutex_exit(&zfsvfs->z_znodes_lock);
1348
1349 if (zp->z_acl_cached) {
1350 zfs_acl_free(zp->z_acl_cached);
1351 zp->z_acl_cached = NULL;
1352 }
1353
1354 kmem_cache_free(znode_cache, zp);
1355
1356 VFS_RELE(zfsvfs->z_vfs);
1357 }
1358
1359 void
1360 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1361 uint64_t ctime[2], boolean_t have_tx)
1362 {
1363 timestruc_t now;
1364
1365 gethrestime(&now);
1366
1367 if (have_tx) { /* will sa_bulk_update happen really soon? */
1368 zp->z_atime_dirty = 0;
1369 zp->z_seq++;
1370 } else {
1371 zp->z_atime_dirty = 1;
1372 }
1373
1374 if (flag & AT_ATIME) {
1375 ZFS_TIME_ENCODE(&now, zp->z_atime);
1376 }
1377
1378 if (flag & AT_MTIME) {
1379 ZFS_TIME_ENCODE(&now, mtime);
1380 if (zp->z_zfsvfs->z_use_fuids) {
1381 zp->z_pflags |= (ZFS_ARCHIVE |
1382 ZFS_AV_MODIFIED);
1383 }
1384 }
1385
1386 if (flag & AT_CTIME) {
1387 ZFS_TIME_ENCODE(&now, ctime);
1388 if (zp->z_zfsvfs->z_use_fuids)
1389 zp->z_pflags |= ZFS_ARCHIVE;
1390 }
1391 }
1392
1393 /*
1394 * Grow the block size for a file.
1395 *
1396 * IN: zp - znode of file to free data in.
1397 * size - requested block size
1398 * tx - open transaction.
1399 *
1400 * NOTE: this function assumes that the znode is write locked.
1401 */
1402 void
1403 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1404 {
1405 int error;
1406 u_longlong_t dummy;
1407
1408 if (size <= zp->z_blksz)
1409 return;
1410 /*
1411 * If the file size is already greater than the current blocksize,
1412 * we will not grow. If there is more than one block in a file,
1413 * the blocksize cannot change.
1414 */
1415 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1416 return;
1417
1418 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1419 size, 0, tx);
1420
1421 if (error == ENOTSUP)
1422 return;
1423 ASSERT0(error);
1424
1425 /* What blocksize did we actually get? */
1426 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1427 }
1428
1429 /*
1430 * This is a dummy interface used when pvn_vplist_dirty() should *not*
1431 * be calling back into the fs for a putpage(). E.g.: when truncating
1432 * a file, the pages being "thrown away* don't need to be written out.
1433 */
1434 /* ARGSUSED */
1435 static int
1436 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1437 int flags, cred_t *cr)
1438 {
1439 ASSERT(0);
1440 return (0);
1441 }
1442
1443 /*
1444 * Increase the file length
1445 *
1446 * IN: zp - znode of file to free data in.
1447 * end - new end-of-file
1448 *
1449 * RETURN: 0 on success, error code on failure
1450 */
1451 static int
1452 zfs_extend(znode_t *zp, uint64_t end)
1453 {
1454 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1455 dmu_tx_t *tx;
1456 rl_t *rl;
1457 uint64_t newblksz;
1458 int error;
1459
1460 /*
1461 * We will change zp_size, lock the whole file.
1462 */
1463 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1464
1465 /*
1466 * Nothing to do if file already at desired length.
1467 */
1468 if (end <= zp->z_size) {
1469 zfs_range_unlock(rl);
1470 return (0);
1471 }
1472 tx = dmu_tx_create(zfsvfs->z_os);
1473 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1474 zfs_sa_upgrade_txholds(tx, zp);
1475 if (end > zp->z_blksz &&
1476 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1477 /*
1478 * We are growing the file past the current block size.
1479 */
1480 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1481 ASSERT(!ISP2(zp->z_blksz));
1482 newblksz = MIN(end, SPA_MAXBLOCKSIZE);
1483 } else {
1484 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1485 }
1486 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1487 } else {
1488 newblksz = 0;
1489 }
1490
1491 error = dmu_tx_assign(tx, TXG_WAIT);
1492 if (error) {
1493 dmu_tx_abort(tx);
1494 zfs_range_unlock(rl);
1495 return (error);
1496 }
1497
1498 if (newblksz)
1499 zfs_grow_blocksize(zp, newblksz, tx);
1500
1501 zp->z_size = end;
1502
1503 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1504 &zp->z_size, sizeof (zp->z_size), tx));
1505
1506 zfs_range_unlock(rl);
1507
1508 dmu_tx_commit(tx);
1509
1510 return (0);
1511 }
1512
1513 /*
1514 * Free space in a file.
1515 *
1516 * IN: zp - znode of file to free data in.
1517 * off - start of section to free.
1518 * len - length of section to free.
1519 *
1520 * RETURN: 0 on success, error code on failure
1521 */
1522 static int
1523 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1524 {
1525 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1526 rl_t *rl;
1527 int error;
1528
1529 /*
1530 * Lock the range being freed.
1531 */
1532 rl = zfs_range_lock(zp, off, len, RL_WRITER);
1533
1534 /*
1535 * Nothing to do if file already at desired length.
1536 */
1537 if (off >= zp->z_size) {
1538 zfs_range_unlock(rl);
1539 return (0);
1540 }
1541
1542 if (off + len > zp->z_size)
1543 len = zp->z_size - off;
1544
1545 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1546
1547 zfs_range_unlock(rl);
1548
1549 return (error);
1550 }
1551
1552 /*
1553 * Truncate a file
1554 *
1555 * IN: zp - znode of file to free data in.
1556 * end - new end-of-file.
1557 *
1558 * RETURN: 0 on success, error code on failure
1559 */
1560 static int
1561 zfs_trunc(znode_t *zp, uint64_t end)
1562 {
1563 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1564 vnode_t *vp = ZTOV(zp);
1565 dmu_tx_t *tx;
1566 rl_t *rl;
1567 int error;
1568 sa_bulk_attr_t bulk[2];
1569 int count = 0;
1570
1571 /*
1572 * We will change zp_size, lock the whole file.
1573 */
1574 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1575
1576 /*
1577 * Nothing to do if file already at desired length.
1578 */
1579 if (end >= zp->z_size) {
1580 zfs_range_unlock(rl);
1581 return (0);
1582 }
1583
1584 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1);
1585 if (error) {
1586 zfs_range_unlock(rl);
1587 return (error);
1588 }
1589 tx = dmu_tx_create(zfsvfs->z_os);
1590 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1591 zfs_sa_upgrade_txholds(tx, zp);
1592 error = dmu_tx_assign(tx, TXG_WAIT);
1593 if (error) {
1594 dmu_tx_abort(tx);
1595 zfs_range_unlock(rl);
1596 return (error);
1597 }
1598
1599 zp->z_size = end;
1600 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1601 NULL, &zp->z_size, sizeof (zp->z_size));
1602
1603 if (end == 0) {
1604 zp->z_pflags &= ~ZFS_SPARSE;
1605 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1606 NULL, &zp->z_pflags, 8);
1607 }
1608 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1609
1610 dmu_tx_commit(tx);
1611
1612 /*
1613 * Clear any mapped pages in the truncated region. This has to
1614 * happen outside of the transaction to avoid the possibility of
1615 * a deadlock with someone trying to push a page that we are
1616 * about to invalidate.
1617 */
1618 if (vn_has_cached_data(vp)) {
1619 page_t *pp;
1620 uint64_t start = end & PAGEMASK;
1621 int poff = end & PAGEOFFSET;
1622
1623 if (poff != 0 && (pp = page_lookup(vp, start, SE_SHARED))) {
1624 /*
1625 * We need to zero a partial page.
1626 */
1627 pagezero(pp, poff, PAGESIZE - poff);
1628 start += PAGESIZE;
1629 page_unlock(pp);
1630 }
1631 error = pvn_vplist_dirty(vp, start, zfs_no_putpage,
1632 B_INVAL | B_TRUNC, NULL);
1633 ASSERT(error == 0);
1634 }
1635
1636 zfs_range_unlock(rl);
1637
1638 return (0);
1639 }
1640
1641 /*
1642 * Free space in a file
1643 *
1644 * IN: zp - znode of file to free data in.
1645 * off - start of range
1646 * len - end of range (0 => EOF)
1647 * flag - current file open mode flags.
1648 * log - TRUE if this action should be logged
1649 *
1650 * RETURN: 0 on success, error code on failure
1651 */
1652 int
1653 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1654 {
1655 vnode_t *vp = ZTOV(zp);
1656 dmu_tx_t *tx;
1657 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1658 zilog_t *zilog = zfsvfs->z_log;
1659 uint64_t mode;
1660 uint64_t mtime[2], ctime[2];
1661 sa_bulk_attr_t bulk[3];
1662 int count = 0;
1663 int error;
1664
1665 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1666 sizeof (mode))) != 0)
1667 return (error);
1668
1669 if (off > zp->z_size) {
1670 error = zfs_extend(zp, off+len);
1671 if (error == 0 && log)
1672 goto log;
1673 else
1674 return (error);
1675 }
1676
1677 /*
1678 * Check for any locks in the region to be freed.
1679 */
1680
1681 if (MANDLOCK(vp, (mode_t)mode)) {
1682 uint64_t length = (len ? len : zp->z_size - off);
1683 if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1684 return (error);
1685 }
1686
1687 if (len == 0) {
1688 error = zfs_trunc(zp, off);
1689 } else {
1690 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1691 off + len > zp->z_size)
1692 error = zfs_extend(zp, off+len);
1693 }
1694 if (error || !log)
1695 return (error);
1696 log:
1697 tx = dmu_tx_create(zfsvfs->z_os);
1698 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1699 zfs_sa_upgrade_txholds(tx, zp);
1700 error = dmu_tx_assign(tx, TXG_WAIT);
1701 if (error) {
1702 dmu_tx_abort(tx);
1703 return (error);
1704 }
1705
1706 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1707 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1708 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1709 NULL, &zp->z_pflags, 8);
1710 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1711 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1712 ASSERT(error == 0);
1713
1714 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1715
1716 dmu_tx_commit(tx);
1717 return (0);
1718 }
1719
1720 void
1721 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1722 {
1723 zfsvfs_t zfsvfs;
1724 uint64_t moid, obj, sa_obj, version;
1725 uint64_t sense = ZFS_CASE_SENSITIVE;
1726 uint64_t norm = 0;
1727 nvpair_t *elem;
1728 int error;
1729 int i;
1730 znode_t *rootzp = NULL;
1731 vnode_t *vp;
1732 vattr_t vattr;
1733 znode_t *zp;
1734 zfs_acl_ids_t acl_ids;
1735
1736 /*
1737 * First attempt to create master node.
1738 */
1739 /*
1740 * In an empty objset, there are no blocks to read and thus
1741 * there can be no i/o errors (which we assert below).
1742 */
1743 moid = MASTER_NODE_OBJ;
1744 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1745 DMU_OT_NONE, 0, tx);
1746 ASSERT(error == 0);
1747
1748 /*
1749 * Set starting attributes.
1750 */
1751 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1752 elem = NULL;
1753 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1754 /* For the moment we expect all zpl props to be uint64_ts */
1755 uint64_t val;
1756 char *name;
1757
1758 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1759 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1760 name = nvpair_name(elem);
1761 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1762 if (val < version)
1763 version = val;
1764 } else {
1765 error = zap_update(os, moid, name, 8, 1, &val, tx);
1766 }
1767 ASSERT(error == 0);
1768 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1769 norm = val;
1770 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1771 sense = val;
1772 }
1773 ASSERT(version != 0);
1774 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1775
1776 /*
1777 * Create zap object used for SA attribute registration
1778 */
1779
1780 if (version >= ZPL_VERSION_SA) {
1781 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1782 DMU_OT_NONE, 0, tx);
1783 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1784 ASSERT(error == 0);
1785 } else {
1786 sa_obj = 0;
1787 }
1788 /*
1789 * Create a delete queue.
1790 */
1791 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1792
1793 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1794 ASSERT(error == 0);
1795
1796 /*
1797 * Create root znode. Create minimal znode/vnode/zfsvfs
1798 * to allow zfs_mknode to work.
1799 */
1800 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1801 vattr.va_type = VDIR;
1802 vattr.va_mode = S_IFDIR|0755;
1803 vattr.va_uid = crgetuid(cr);
1804 vattr.va_gid = crgetgid(cr);
1805
1806 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1807 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1808 rootzp->z_moved = 0;
1809 rootzp->z_unlinked = 0;
1810 rootzp->z_atime_dirty = 0;
1811 rootzp->z_is_sa = USE_SA(version, os);
1812
1813 vp = ZTOV(rootzp);
1814 vn_reinit(vp);
1815 vp->v_type = VDIR;
1816
1817 bzero(&zfsvfs, sizeof (zfsvfs_t));
1818
1819 zfsvfs.z_os = os;
1820 zfsvfs.z_parent = &zfsvfs;
1821 zfsvfs.z_version = version;
1822 zfsvfs.z_use_fuids = USE_FUIDS(version, os);
1823 zfsvfs.z_use_sa = USE_SA(version, os);
1824 zfsvfs.z_norm = norm;
1825
1826 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1827 &zfsvfs.z_attr_table);
1828
1829 ASSERT(error == 0);
1830
1831 /*
1832 * Fold case on file systems that are always or sometimes case
1833 * insensitive.
1834 */
1835 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1836 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER;
1837
1838 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1839 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t),
1840 offsetof(znode_t, z_link_node));
1841
1842 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1843 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1844
1845 rootzp->z_zfsvfs = &zfsvfs;
1846 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1847 cr, NULL, &acl_ids));
1848 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1849 ASSERT3P(zp, ==, rootzp);
1850 ASSERT(!vn_in_dnlc(ZTOV(rootzp))); /* not valid to move */
1851 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1852 ASSERT(error == 0);
1853 zfs_acl_ids_free(&acl_ids);
1854 POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1855
1856 ZTOV(rootzp)->v_count = 0;
1857 sa_handle_destroy(rootzp->z_sa_hdl);
1858 kmem_cache_free(znode_cache, rootzp);
1859
1860 /*
1861 * Create shares directory
1862 */
1863
1864 error = zfs_create_share_dir(&zfsvfs, tx);
1865
1866 ASSERT(error == 0);
1867
1868 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1869 mutex_destroy(&zfsvfs.z_hold_mtx[i]);
1870 }
1871
1872 #endif /* _KERNEL */
1873
1874 static int
1875 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1876 {
1877 uint64_t sa_obj = 0;
1878 int error;
1879
1880 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1881 if (error != 0 && error != ENOENT)
1882 return (error);
1883
1884 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1885 return (error);
1886 }
1887
1888 static int
1889 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1890 dmu_buf_t **db, void *tag)
1891 {
1892 dmu_object_info_t doi;
1893 int error;
1894
1895 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
1896 return (error);
1897
1898 dmu_object_info_from_db(*db, &doi);
1899 if ((doi.doi_bonus_type != DMU_OT_SA &&
1900 doi.doi_bonus_type != DMU_OT_ZNODE) ||
1901 doi.doi_bonus_type == DMU_OT_ZNODE &&
1902 doi.doi_bonus_size < sizeof (znode_phys_t)) {
1903 sa_buf_rele(*db, tag);
1904 return (SET_ERROR(ENOTSUP));
1905 }
1906
1907 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
1908 if (error != 0) {
1909 sa_buf_rele(*db, tag);
1910 return (error);
1911 }
1912
1913 return (0);
1914 }
1915
1916 void
1917 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
1918 {
1919 sa_handle_destroy(hdl);
1920 sa_buf_rele(db, tag);
1921 }
1922
1923 /*
1924 * Given an object number, return its parent object number and whether
1925 * or not the object is an extended attribute directory.
1926 */
1927 static int
1928 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
1929 uint64_t *pobjp, int *is_xattrdir)
1930 {
1931 uint64_t parent;
1932 uint64_t pflags;
1933 uint64_t mode;
1934 uint64_t parent_mode;
1935 sa_bulk_attr_t bulk[3];
1936 sa_handle_t *sa_hdl;
1937 dmu_buf_t *sa_db;
1938 int count = 0;
1939 int error;
1940
1941 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
1942 &parent, sizeof (parent));
1943 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
1944 &pflags, sizeof (pflags));
1945 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1946 &mode, sizeof (mode));
1947
1948 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
1949 return (error);
1950
1951 /*
1952 * When a link is removed its parent pointer is not changed and will
1953 * be invalid. There are two cases where a link is removed but the
1954 * file stays around, when it goes to the delete queue and when there
1955 * are additional links.
1956 */
1957 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
1958 if (error != 0)
1959 return (error);
1960
1961 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
1962 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
1963 if (error != 0)
1964 return (error);
1965
1966 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
1967
1968 /*
1969 * Extended attributes can be applied to files, directories, etc.
1970 * Otherwise the parent must be a directory.
1971 */
1972 if (!*is_xattrdir && !S_ISDIR(parent_mode))
1973 return (SET_ERROR(EINVAL));
1974
1975 *pobjp = parent;
1976
1977 return (0);
1978 }
1979
1980 /*
1981 * Given an object number, return some zpl level statistics
1982 */
1983 static int
1984 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
1985 zfs_stat_t *sb)
1986 {
1987 sa_bulk_attr_t bulk[4];
1988 int count = 0;
1989
1990 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1991 &sb->zs_mode, sizeof (sb->zs_mode));
1992 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
1993 &sb->zs_gen, sizeof (sb->zs_gen));
1994 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
1995 &sb->zs_links, sizeof (sb->zs_links));
1996 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
1997 &sb->zs_ctime, sizeof (sb->zs_ctime));
1998
1999 return (sa_bulk_lookup(hdl, bulk, count));
2000 }
2001
2002 static int
2003 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2004 sa_attr_type_t *sa_table, char *buf, int len)
2005 {
2006 sa_handle_t *sa_hdl;
2007 sa_handle_t *prevhdl = NULL;
2008 dmu_buf_t *prevdb = NULL;
2009 dmu_buf_t *sa_db = NULL;
2010 char *path = buf + len - 1;
2011 int error;
2012
2013 *path = '\0';
2014 sa_hdl = hdl;
2015
2016 for (;;) {
2017 uint64_t pobj;
2018 char component[MAXNAMELEN + 2];
2019 size_t complen;
2020 int is_xattrdir;
2021
2022 if (prevdb)
2023 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2024
2025 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2026 &is_xattrdir)) != 0)
2027 break;
2028
2029 if (pobj == obj) {
2030 if (path[0] != '/')
2031 *--path = '/';
2032 break;
2033 }
2034
2035 component[0] = '/';
2036 if (is_xattrdir) {
2037 (void) sprintf(component + 1, "<xattrdir>");
2038 } else {
2039 error = zap_value_search(osp, pobj, obj,
2040 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2041 if (error != 0)
2042 break;
2043 }
2044
2045 complen = strlen(component);
2046 path -= complen;
2047 ASSERT(path >= buf);
2048 bcopy(component, path, complen);
2049 obj = pobj;
2050
2051 if (sa_hdl != hdl) {
2052 prevhdl = sa_hdl;
2053 prevdb = sa_db;
2054 }
2055 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2056 if (error != 0) {
2057 sa_hdl = prevhdl;
2058 sa_db = prevdb;
2059 break;
2060 }
2061 }
2062
2063 if (sa_hdl != NULL && sa_hdl != hdl) {
2064 ASSERT(sa_db != NULL);
2065 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2066 }
2067
2068 if (error == 0)
2069 (void) memmove(buf, path, buf + len - path);
2070
2071 return (error);
2072 }
2073
2074 int
2075 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2076 {
2077 sa_attr_type_t *sa_table;
2078 sa_handle_t *hdl;
2079 dmu_buf_t *db;
2080 int error;
2081
2082 error = zfs_sa_setup(osp, &sa_table);
2083 if (error != 0)
2084 return (error);
2085
2086 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2087 if (error != 0)
2088 return (error);
2089
2090 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2091
2092 zfs_release_sa_handle(hdl, db, FTAG);
2093 return (error);
2094 }
2095
2096 int
2097 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2098 char *buf, int len)
2099 {
2100 char *path = buf + len - 1;
2101 sa_attr_type_t *sa_table;
2102 sa_handle_t *hdl;
2103 dmu_buf_t *db;
2104 int error;
2105
2106 *path = '\0';
2107
2108 error = zfs_sa_setup(osp, &sa_table);
2109 if (error != 0)
2110 return (error);
2111
2112 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2113 if (error != 0)
2114 return (error);
2115
2116 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2117 if (error != 0) {
2118 zfs_release_sa_handle(hdl, db, FTAG);
2119 return (error);
2120 }
2121
2122 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2123
2124 zfs_release_sa_handle(hdl, db, FTAG);
2125 return (error);
2126 }