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) 2012, 2018 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/sysmacros.h>
33 #include <sys/mntent.h>
34 #include <sys/u8_textprep.h>
35 #include <sys/dsl_dataset.h>
36 #include <sys/vfs.h>
37 #include <sys/vnode.h>
38 #include <sys/file.h>
39 #include <sys/kmem.h>
40 #include <sys/errno.h>
41 #include <sys/atomic.h>
42 #include <sys/zfs_dir.h>
43 #include <sys/zfs_acl.h>
44 #include <sys/zfs_ioctl.h>
45 #include <sys/zfs_rlock.h>
46 #include <sys/zfs_fuid.h>
47 #include <sys/zfs_vnops.h>
48 #include <sys/zfs_ctldir.h>
49 #include <sys/dnode.h>
50 #include <sys/fs/zfs.h>
51 #include <sys/zpl.h>
52 #endif /* _KERNEL */
53
54 #include <sys/dmu.h>
55 #include <sys/dmu_objset.h>
56 #include <sys/dmu_tx.h>
57 #include <sys/zfs_refcount.h>
58 #include <sys/stat.h>
59 #include <sys/zap.h>
60 #include <sys/zfs_znode.h>
61 #include <sys/sa.h>
62 #include <sys/zfs_sa.h>
63 #include <sys/zfs_stat.h>
64
65 #include "zfs_prop.h"
66 #include "zfs_comutil.h"
67
68 /*
69 * Functions needed for userland (ie: libzpool) are not put under
70 * #ifdef_KERNEL; the rest of the functions have dependencies
71 * (such as VFS logic) that will not compile easily in userland.
72 */
73 #ifdef _KERNEL
74
75 static kmem_cache_t *znode_cache = NULL;
76 static kmem_cache_t *znode_hold_cache = NULL;
77 unsigned int zfs_object_mutex_size = ZFS_OBJ_MTX_SZ;
78
79 /*
80 * This is used by the test suite so that it can delay znodes from being
81 * freed in order to inspect the unlinked set.
82 */
83 int zfs_unlink_suspend_progress = 0;
84
85 /*
86 * This callback is invoked when acquiring a RL_WRITER or RL_APPEND lock on
87 * z_rangelock. It will modify the offset and length of the lock to reflect
88 * znode-specific information, and convert RL_APPEND to RL_WRITER. This is
89 * called with the rangelock_t's rl_lock held, which avoids races.
90 */
91 static void
zfs_rangelock_cb(zfs_locked_range_t * new,void * arg)92 zfs_rangelock_cb(zfs_locked_range_t *new, void *arg)
93 {
94 znode_t *zp = arg;
95
96 /*
97 * If in append mode, convert to writer and lock starting at the
98 * current end of file.
99 */
100 if (new->lr_type == RL_APPEND) {
101 new->lr_offset = zp->z_size;
102 new->lr_type = RL_WRITER;
103 }
104
105 /*
106 * If we need to grow the block size then lock the whole file range.
107 */
108 uint64_t end_size = MAX(zp->z_size, new->lr_offset + new->lr_length);
109 if (end_size > zp->z_blksz && (!ISP2(zp->z_blksz) ||
110 zp->z_blksz < ZTOZSB(zp)->z_max_blksz)) {
111 new->lr_offset = 0;
112 new->lr_length = UINT64_MAX;
113 }
114 }
115
116 /*ARGSUSED*/
117 static int
zfs_znode_cache_constructor(void * buf,void * arg,int kmflags)118 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
119 {
120 znode_t *zp = buf;
121
122 inode_init_once(ZTOI(zp));
123 list_link_init(&zp->z_link_node);
124
125 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
126 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
127 rw_init(&zp->z_name_lock, NULL, RW_NOLOCKDEP, NULL);
128 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
129 rw_init(&zp->z_xattr_lock, NULL, RW_DEFAULT, NULL);
130
131 zfs_rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp);
132
133 zp->z_dirlocks = NULL;
134 zp->z_acl_cached = NULL;
135 zp->z_xattr_cached = NULL;
136 zp->z_xattr_parent = 0;
137 zp->z_sync_writes_cnt = 0;
138 zp->z_async_writes_cnt = 0;
139
140 return (0);
141 }
142
143 /*ARGSUSED*/
144 static void
zfs_znode_cache_destructor(void * buf,void * arg)145 zfs_znode_cache_destructor(void *buf, void *arg)
146 {
147 znode_t *zp = buf;
148
149 ASSERT(!list_link_active(&zp->z_link_node));
150 mutex_destroy(&zp->z_lock);
151 rw_destroy(&zp->z_parent_lock);
152 rw_destroy(&zp->z_name_lock);
153 mutex_destroy(&zp->z_acl_lock);
154 rw_destroy(&zp->z_xattr_lock);
155 zfs_rangelock_fini(&zp->z_rangelock);
156
157 ASSERT3P(zp->z_dirlocks, ==, NULL);
158 ASSERT3P(zp->z_acl_cached, ==, NULL);
159 ASSERT3P(zp->z_xattr_cached, ==, NULL);
160
161 ASSERT0(atomic_load_32(&zp->z_sync_writes_cnt));
162 ASSERT0(atomic_load_32(&zp->z_async_writes_cnt));
163 }
164
165 static int
zfs_znode_hold_cache_constructor(void * buf,void * arg,int kmflags)166 zfs_znode_hold_cache_constructor(void *buf, void *arg, int kmflags)
167 {
168 znode_hold_t *zh = buf;
169
170 mutex_init(&zh->zh_lock, NULL, MUTEX_DEFAULT, NULL);
171 zh->zh_refcount = 0;
172
173 return (0);
174 }
175
176 static void
zfs_znode_hold_cache_destructor(void * buf,void * arg)177 zfs_znode_hold_cache_destructor(void *buf, void *arg)
178 {
179 znode_hold_t *zh = buf;
180
181 mutex_destroy(&zh->zh_lock);
182 }
183
184 void
zfs_znode_init(void)185 zfs_znode_init(void)
186 {
187 /*
188 * Initialize zcache. The KMC_SLAB hint is used in order that it be
189 * backed by kmalloc() when on the Linux slab in order that any
190 * wait_on_bit() operations on the related inode operate properly.
191 */
192 ASSERT(znode_cache == NULL);
193 znode_cache = kmem_cache_create("zfs_znode_cache",
194 sizeof (znode_t), 0, zfs_znode_cache_constructor,
195 zfs_znode_cache_destructor, NULL, NULL, NULL, KMC_SLAB);
196
197 ASSERT(znode_hold_cache == NULL);
198 znode_hold_cache = kmem_cache_create("zfs_znode_hold_cache",
199 sizeof (znode_hold_t), 0, zfs_znode_hold_cache_constructor,
200 zfs_znode_hold_cache_destructor, NULL, NULL, NULL, 0);
201 }
202
203 void
zfs_znode_fini(void)204 zfs_znode_fini(void)
205 {
206 /*
207 * Cleanup zcache
208 */
209 if (znode_cache)
210 kmem_cache_destroy(znode_cache);
211 znode_cache = NULL;
212
213 if (znode_hold_cache)
214 kmem_cache_destroy(znode_hold_cache);
215 znode_hold_cache = NULL;
216 }
217
218 /*
219 * The zfs_znode_hold_enter() / zfs_znode_hold_exit() functions are used to
220 * serialize access to a znode and its SA buffer while the object is being
221 * created or destroyed. This kind of locking would normally reside in the
222 * znode itself but in this case that's impossible because the znode and SA
223 * buffer may not yet exist. Therefore the locking is handled externally
224 * with an array of mutexes and AVLs trees which contain per-object locks.
225 *
226 * In zfs_znode_hold_enter() a per-object lock is created as needed, inserted
227 * in to the correct AVL tree and finally the per-object lock is held. In
228 * zfs_znode_hold_exit() the process is reversed. The per-object lock is
229 * released, removed from the AVL tree and destroyed if there are no waiters.
230 *
231 * This scheme has two important properties:
232 *
233 * 1) No memory allocations are performed while holding one of the z_hold_locks.
234 * This ensures evict(), which can be called from direct memory reclaim, will
235 * never block waiting on a z_hold_locks which just happens to have hashed
236 * to the same index.
237 *
238 * 2) All locks used to serialize access to an object are per-object and never
239 * shared. This minimizes lock contention without creating a large number
240 * of dedicated locks.
241 *
242 * On the downside it does require znode_lock_t structures to be frequently
243 * allocated and freed. However, because these are backed by a kmem cache
244 * and very short lived this cost is minimal.
245 */
246 int
zfs_znode_hold_compare(const void * a,const void * b)247 zfs_znode_hold_compare(const void *a, const void *b)
248 {
249 const znode_hold_t *zh_a = (const znode_hold_t *)a;
250 const znode_hold_t *zh_b = (const znode_hold_t *)b;
251
252 return (TREE_CMP(zh_a->zh_obj, zh_b->zh_obj));
253 }
254
255 static boolean_t __maybe_unused
zfs_znode_held(zfsvfs_t * zfsvfs,uint64_t obj)256 zfs_znode_held(zfsvfs_t *zfsvfs, uint64_t obj)
257 {
258 znode_hold_t *zh, search;
259 int i = ZFS_OBJ_HASH(zfsvfs, obj);
260 boolean_t held;
261
262 search.zh_obj = obj;
263
264 mutex_enter(&zfsvfs->z_hold_locks[i]);
265 zh = avl_find(&zfsvfs->z_hold_trees[i], &search, NULL);
266 held = (zh && MUTEX_HELD(&zh->zh_lock)) ? B_TRUE : B_FALSE;
267 mutex_exit(&zfsvfs->z_hold_locks[i]);
268
269 return (held);
270 }
271
272 static znode_hold_t *
zfs_znode_hold_enter(zfsvfs_t * zfsvfs,uint64_t obj)273 zfs_znode_hold_enter(zfsvfs_t *zfsvfs, uint64_t obj)
274 {
275 znode_hold_t *zh, *zh_new, search;
276 int i = ZFS_OBJ_HASH(zfsvfs, obj);
277 boolean_t found = B_FALSE;
278
279 zh_new = kmem_cache_alloc(znode_hold_cache, KM_SLEEP);
280 search.zh_obj = obj;
281
282 mutex_enter(&zfsvfs->z_hold_locks[i]);
283 zh = avl_find(&zfsvfs->z_hold_trees[i], &search, NULL);
284 if (likely(zh == NULL)) {
285 zh = zh_new;
286 zh->zh_obj = obj;
287 avl_add(&zfsvfs->z_hold_trees[i], zh);
288 } else {
289 ASSERT3U(zh->zh_obj, ==, obj);
290 found = B_TRUE;
291 }
292 zh->zh_refcount++;
293 ASSERT3S(zh->zh_refcount, >, 0);
294 mutex_exit(&zfsvfs->z_hold_locks[i]);
295
296 if (found == B_TRUE)
297 kmem_cache_free(znode_hold_cache, zh_new);
298
299 ASSERT(MUTEX_NOT_HELD(&zh->zh_lock));
300 mutex_enter(&zh->zh_lock);
301
302 return (zh);
303 }
304
305 static void
zfs_znode_hold_exit(zfsvfs_t * zfsvfs,znode_hold_t * zh)306 zfs_znode_hold_exit(zfsvfs_t *zfsvfs, znode_hold_t *zh)
307 {
308 int i = ZFS_OBJ_HASH(zfsvfs, zh->zh_obj);
309 boolean_t remove = B_FALSE;
310
311 ASSERT(zfs_znode_held(zfsvfs, zh->zh_obj));
312 mutex_exit(&zh->zh_lock);
313
314 mutex_enter(&zfsvfs->z_hold_locks[i]);
315 ASSERT3S(zh->zh_refcount, >, 0);
316 if (--zh->zh_refcount == 0) {
317 avl_remove(&zfsvfs->z_hold_trees[i], zh);
318 remove = B_TRUE;
319 }
320 mutex_exit(&zfsvfs->z_hold_locks[i]);
321
322 if (remove == B_TRUE)
323 kmem_cache_free(znode_hold_cache, zh);
324 }
325
326 dev_t
zfs_cmpldev(uint64_t dev)327 zfs_cmpldev(uint64_t dev)
328 {
329 return (dev);
330 }
331
332 static void
zfs_znode_sa_init(zfsvfs_t * zfsvfs,znode_t * zp,dmu_buf_t * db,dmu_object_type_t obj_type,sa_handle_t * sa_hdl)333 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
334 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
335 {
336 ASSERT(zfs_znode_held(zfsvfs, zp->z_id));
337
338 mutex_enter(&zp->z_lock);
339
340 ASSERT(zp->z_sa_hdl == NULL);
341 ASSERT(zp->z_acl_cached == NULL);
342 if (sa_hdl == NULL) {
343 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
344 SA_HDL_SHARED, &zp->z_sa_hdl));
345 } else {
346 zp->z_sa_hdl = sa_hdl;
347 sa_set_userp(sa_hdl, zp);
348 }
349
350 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
351
352 mutex_exit(&zp->z_lock);
353 }
354
355 void
zfs_znode_dmu_fini(znode_t * zp)356 zfs_znode_dmu_fini(znode_t *zp)
357 {
358 ASSERT(zfs_znode_held(ZTOZSB(zp), zp->z_id) || zp->z_unlinked ||
359 RW_WRITE_HELD(&ZTOZSB(zp)->z_teardown_inactive_lock));
360
361 sa_handle_destroy(zp->z_sa_hdl);
362 zp->z_sa_hdl = NULL;
363 }
364
365 /*
366 * Called by new_inode() to allocate a new inode.
367 */
368 int
zfs_inode_alloc(struct super_block * sb,struct inode ** ip)369 zfs_inode_alloc(struct super_block *sb, struct inode **ip)
370 {
371 znode_t *zp;
372
373 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
374 *ip = ZTOI(zp);
375
376 return (0);
377 }
378
379 /*
380 * Called in multiple places when an inode should be destroyed.
381 */
382 void
zfs_inode_destroy(struct inode * ip)383 zfs_inode_destroy(struct inode *ip)
384 {
385 znode_t *zp = ITOZ(ip);
386 zfsvfs_t *zfsvfs = ZTOZSB(zp);
387
388 mutex_enter(&zfsvfs->z_znodes_lock);
389 if (list_link_active(&zp->z_link_node)) {
390 list_remove(&zfsvfs->z_all_znodes, zp);
391 zfsvfs->z_nr_znodes--;
392 }
393 mutex_exit(&zfsvfs->z_znodes_lock);
394
395 if (zp->z_acl_cached) {
396 zfs_acl_free(zp->z_acl_cached);
397 zp->z_acl_cached = NULL;
398 }
399
400 if (zp->z_xattr_cached) {
401 nvlist_free(zp->z_xattr_cached);
402 zp->z_xattr_cached = NULL;
403 }
404
405 kmem_cache_free(znode_cache, zp);
406 }
407
408 static void
zfs_inode_set_ops(zfsvfs_t * zfsvfs,struct inode * ip)409 zfs_inode_set_ops(zfsvfs_t *zfsvfs, struct inode *ip)
410 {
411 uint64_t rdev = 0;
412
413 switch (ip->i_mode & S_IFMT) {
414 case S_IFREG:
415 ip->i_op = &zpl_inode_operations;
416 ip->i_fop = &zpl_file_operations;
417 ip->i_mapping->a_ops = &zpl_address_space_operations;
418 break;
419
420 case S_IFDIR:
421 ip->i_op = &zpl_dir_inode_operations;
422 ip->i_fop = &zpl_dir_file_operations;
423 ITOZ(ip)->z_zn_prefetch = B_TRUE;
424 break;
425
426 case S_IFLNK:
427 ip->i_op = &zpl_symlink_inode_operations;
428 break;
429
430 /*
431 * rdev is only stored in a SA only for device files.
432 */
433 case S_IFCHR:
434 case S_IFBLK:
435 (void) sa_lookup(ITOZ(ip)->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), &rdev,
436 sizeof (rdev));
437 fallthrough;
438 case S_IFIFO:
439 case S_IFSOCK:
440 init_special_inode(ip, ip->i_mode, rdev);
441 ip->i_op = &zpl_special_inode_operations;
442 break;
443
444 default:
445 zfs_panic_recover("inode %llu has invalid mode: 0x%x\n",
446 (u_longlong_t)ip->i_ino, ip->i_mode);
447
448 /* Assume the inode is a file and attempt to continue */
449 ip->i_mode = S_IFREG | 0644;
450 ip->i_op = &zpl_inode_operations;
451 ip->i_fop = &zpl_file_operations;
452 ip->i_mapping->a_ops = &zpl_address_space_operations;
453 break;
454 }
455 }
456
457 static void
zfs_set_inode_flags(znode_t * zp,struct inode * ip)458 zfs_set_inode_flags(znode_t *zp, struct inode *ip)
459 {
460 /*
461 * Linux and Solaris have different sets of file attributes, so we
462 * restrict this conversion to the intersection of the two.
463 */
464 #ifdef HAVE_INODE_SET_FLAGS
465 unsigned int flags = 0;
466 if (zp->z_pflags & ZFS_IMMUTABLE)
467 flags |= S_IMMUTABLE;
468 if (zp->z_pflags & ZFS_APPENDONLY)
469 flags |= S_APPEND;
470
471 inode_set_flags(ip, flags, S_IMMUTABLE|S_APPEND);
472 #else
473 if (zp->z_pflags & ZFS_IMMUTABLE)
474 ip->i_flags |= S_IMMUTABLE;
475 else
476 ip->i_flags &= ~S_IMMUTABLE;
477
478 if (zp->z_pflags & ZFS_APPENDONLY)
479 ip->i_flags |= S_APPEND;
480 else
481 ip->i_flags &= ~S_APPEND;
482 #endif
483 }
484
485 /*
486 * Update the embedded inode given the znode.
487 */
488 void
zfs_znode_update_vfs(znode_t * zp)489 zfs_znode_update_vfs(znode_t *zp)
490 {
491 zfsvfs_t *zfsvfs;
492 struct inode *ip;
493 uint32_t blksize;
494 u_longlong_t i_blocks;
495
496 ASSERT(zp != NULL);
497 zfsvfs = ZTOZSB(zp);
498 ip = ZTOI(zp);
499
500 /* Skip .zfs control nodes which do not exist on disk. */
501 if (zfsctl_is_node(ip))
502 return;
503
504 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &blksize, &i_blocks);
505
506 spin_lock(&ip->i_lock);
507 ip->i_mode = zp->z_mode;
508 ip->i_blocks = i_blocks;
509 i_size_write(ip, zp->z_size);
510 spin_unlock(&ip->i_lock);
511 }
512
513
514 /*
515 * Construct a znode+inode and initialize.
516 *
517 * This does not do a call to dmu_set_user() that is
518 * up to the caller to do, in case you don't want to
519 * return the znode
520 */
521 static znode_t *
zfs_znode_alloc(zfsvfs_t * zfsvfs,dmu_buf_t * db,int blksz,dmu_object_type_t obj_type,sa_handle_t * hdl)522 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
523 dmu_object_type_t obj_type, sa_handle_t *hdl)
524 {
525 znode_t *zp;
526 struct inode *ip;
527 uint64_t mode;
528 uint64_t parent;
529 uint64_t tmp_gen;
530 uint64_t links;
531 uint64_t z_uid, z_gid;
532 uint64_t atime[2], mtime[2], ctime[2], btime[2];
533 inode_timespec_t tmp_ts;
534 uint64_t projid = ZFS_DEFAULT_PROJID;
535 sa_bulk_attr_t bulk[12];
536 int count = 0;
537
538 ASSERT(zfsvfs != NULL);
539
540 ip = new_inode(zfsvfs->z_sb);
541 if (ip == NULL)
542 return (NULL);
543
544 zp = ITOZ(ip);
545 ASSERT(zp->z_dirlocks == NULL);
546 ASSERT3P(zp->z_acl_cached, ==, NULL);
547 ASSERT3P(zp->z_xattr_cached, ==, NULL);
548 zp->z_unlinked = B_FALSE;
549 zp->z_atime_dirty = B_FALSE;
550 #if !defined(HAVE_FILEMAP_RANGE_HAS_PAGE)
551 zp->z_is_mapped = B_FALSE;
552 #endif
553 zp->z_is_ctldir = B_FALSE;
554 zp->z_suspended = B_FALSE;
555 zp->z_sa_hdl = NULL;
556 zp->z_mapcnt = 0;
557 zp->z_id = db->db_object;
558 zp->z_blksz = blksz;
559 zp->z_seq = 0x7A4653;
560 zp->z_sync_cnt = 0;
561 zp->z_sync_writes_cnt = 0;
562 zp->z_async_writes_cnt = 0;
563
564 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
565
566 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
567 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &tmp_gen, 8);
568 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
569 &zp->z_size, 8);
570 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
571 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
572 &zp->z_pflags, 8);
573 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
574 &parent, 8);
575 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &z_uid, 8);
576 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &z_gid, 8);
577 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
578 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
579 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
580 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &btime, 16);
581
582 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || tmp_gen == 0 ||
583 (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
584 (zp->z_pflags & ZFS_PROJID) &&
585 sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) {
586 if (hdl == NULL)
587 sa_handle_destroy(zp->z_sa_hdl);
588 zp->z_sa_hdl = NULL;
589 goto error;
590 }
591
592 zp->z_projid = projid;
593 zp->z_mode = ip->i_mode = mode;
594 ip->i_generation = (uint32_t)tmp_gen;
595 ip->i_blkbits = SPA_MINBLOCKSHIFT;
596 set_nlink(ip, (uint32_t)links);
597 zfs_uid_write(ip, z_uid);
598 zfs_gid_write(ip, z_gid);
599 zfs_set_inode_flags(zp, ip);
600
601 /* Cache the xattr parent id */
602 if (zp->z_pflags & ZFS_XATTR)
603 zp->z_xattr_parent = parent;
604
605 ZFS_TIME_DECODE(&tmp_ts, atime);
606 zpl_inode_set_atime_to_ts(ip, tmp_ts);
607 ZFS_TIME_DECODE(&tmp_ts, mtime);
608 zpl_inode_set_mtime_to_ts(ip, tmp_ts);
609 ZFS_TIME_DECODE(&tmp_ts, ctime);
610 zpl_inode_set_ctime_to_ts(ip, tmp_ts);
611 ZFS_TIME_DECODE(&zp->z_btime, btime);
612
613 ip->i_ino = zp->z_id;
614 zfs_znode_update_vfs(zp);
615 zfs_inode_set_ops(zfsvfs, ip);
616
617 /*
618 * The only way insert_inode_locked() can fail is if the ip->i_ino
619 * number is already hashed for this super block. This can never
620 * happen because the inode numbers map 1:1 with the object numbers.
621 *
622 * Exceptions include rolling back a mounted file system, either
623 * from the zfs rollback or zfs recv command.
624 *
625 * Active inodes are unhashed during the rollback, but since zrele
626 * can happen asynchronously, we can't guarantee they've been
627 * unhashed. This can cause hash collisions in unlinked drain
628 * processing so do not hash unlinked znodes.
629 */
630 if (links > 0)
631 VERIFY3S(insert_inode_locked(ip), ==, 0);
632
633 mutex_enter(&zfsvfs->z_znodes_lock);
634 list_insert_tail(&zfsvfs->z_all_znodes, zp);
635 zfsvfs->z_nr_znodes++;
636 mutex_exit(&zfsvfs->z_znodes_lock);
637
638 if (links > 0)
639 unlock_new_inode(ip);
640 return (zp);
641
642 error:
643 iput(ip);
644 return (NULL);
645 }
646
647 /*
648 * Safely mark an inode dirty. Inodes which are part of a read-only
649 * file system or snapshot may not be dirtied.
650 */
651 void
zfs_mark_inode_dirty(struct inode * ip)652 zfs_mark_inode_dirty(struct inode *ip)
653 {
654 zfsvfs_t *zfsvfs = ITOZSB(ip);
655
656 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
657 return;
658
659 mark_inode_dirty(ip);
660 }
661
662 static uint64_t empty_xattr;
663 static uint64_t pad[4];
664 static zfs_acl_phys_t acl_phys;
665 /*
666 * Create a new DMU object to hold a zfs znode.
667 *
668 * IN: dzp - parent directory for new znode
669 * vap - file attributes for new znode
670 * tx - dmu transaction id for zap operations
671 * cr - credentials of caller
672 * flag - flags:
673 * IS_ROOT_NODE - new object will be root
674 * IS_TMPFILE - new object is of O_TMPFILE
675 * IS_XATTR - new object is an attribute
676 * acl_ids - ACL related attributes
677 *
678 * OUT: zpp - allocated znode (set to dzp if IS_ROOT_NODE)
679 *
680 */
681 void
zfs_mknode(znode_t * dzp,vattr_t * vap,dmu_tx_t * tx,cred_t * cr,uint_t flag,znode_t ** zpp,zfs_acl_ids_t * acl_ids)682 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
683 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
684 {
685 uint64_t crtime[2], atime[2], mtime[2], ctime[2];
686 uint64_t mode, size, links, parent, pflags;
687 uint64_t projid = ZFS_DEFAULT_PROJID;
688 uint64_t rdev = 0;
689 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
690 dmu_buf_t *db;
691 inode_timespec_t now;
692 uint64_t gen, obj;
693 int bonuslen;
694 int dnodesize;
695 sa_handle_t *sa_hdl;
696 dmu_object_type_t obj_type;
697 sa_bulk_attr_t *sa_attrs;
698 int cnt = 0;
699 zfs_acl_locator_cb_t locate = { 0 };
700 znode_hold_t *zh;
701
702 if (zfsvfs->z_replay) {
703 obj = vap->va_nodeid;
704 now = vap->va_ctime; /* see zfs_replay_create() */
705 gen = vap->va_nblocks; /* ditto */
706 dnodesize = vap->va_fsid; /* ditto */
707 } else {
708 obj = 0;
709 gethrestime(&now);
710 gen = dmu_tx_get_txg(tx);
711 dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
712 }
713
714 if (dnodesize == 0)
715 dnodesize = DNODE_MIN_SIZE;
716
717 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
718
719 bonuslen = (obj_type == DMU_OT_SA) ?
720 DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
721
722 /*
723 * Create a new DMU object.
724 */
725 /*
726 * There's currently no mechanism for pre-reading the blocks that will
727 * be needed to allocate a new object, so we accept the small chance
728 * that there will be an i/o error and we will fail one of the
729 * assertions below.
730 */
731 if (S_ISDIR(vap->va_mode)) {
732 if (zfsvfs->z_replay) {
733 VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj,
734 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
735 obj_type, bonuslen, dnodesize, tx));
736 } else {
737 obj = zap_create_norm_dnsize(zfsvfs->z_os,
738 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
739 obj_type, bonuslen, dnodesize, tx);
740 }
741 } else {
742 if (zfsvfs->z_replay) {
743 VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj,
744 DMU_OT_PLAIN_FILE_CONTENTS, 0,
745 obj_type, bonuslen, dnodesize, tx));
746 } else {
747 obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
748 DMU_OT_PLAIN_FILE_CONTENTS, 0,
749 obj_type, bonuslen, dnodesize, tx);
750 }
751 }
752
753 zh = zfs_znode_hold_enter(zfsvfs, obj);
754 VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
755
756 /*
757 * If this is the root, fix up the half-initialized parent pointer
758 * to reference the just-allocated physical data area.
759 */
760 if (flag & IS_ROOT_NODE) {
761 dzp->z_id = obj;
762 }
763
764 /*
765 * If parent is an xattr, so am I.
766 */
767 if (dzp->z_pflags & ZFS_XATTR) {
768 flag |= IS_XATTR;
769 }
770
771 if (zfsvfs->z_use_fuids)
772 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
773 else
774 pflags = 0;
775
776 if (S_ISDIR(vap->va_mode)) {
777 size = 2; /* contents ("." and "..") */
778 links = 2;
779 } else {
780 size = 0;
781 links = (flag & IS_TMPFILE) ? 0 : 1;
782 }
783
784 if (S_ISBLK(vap->va_mode) || S_ISCHR(vap->va_mode))
785 rdev = vap->va_rdev;
786
787 parent = dzp->z_id;
788 mode = acl_ids->z_mode;
789 if (flag & IS_XATTR)
790 pflags |= ZFS_XATTR;
791
792 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode)) {
793 /*
794 * With ZFS_PROJID flag, we can easily know whether there is
795 * project ID stored on disk or not. See zfs_space_delta_cb().
796 */
797 if (obj_type != DMU_OT_ZNODE &&
798 dmu_objset_projectquota_enabled(zfsvfs->z_os))
799 pflags |= ZFS_PROJID;
800
801 /*
802 * Inherit project ID from parent if required.
803 */
804 projid = zfs_inherit_projid(dzp);
805 if (dzp->z_pflags & ZFS_PROJINHERIT)
806 pflags |= ZFS_PROJINHERIT;
807 }
808
809 /*
810 * No execs denied will be determined when zfs_mode_compute() is called.
811 */
812 pflags |= acl_ids->z_aclp->z_hints &
813 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
814 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
815
816 ZFS_TIME_ENCODE(&now, crtime);
817 ZFS_TIME_ENCODE(&now, ctime);
818
819 if (vap->va_mask & ATTR_ATIME) {
820 ZFS_TIME_ENCODE(&vap->va_atime, atime);
821 } else {
822 ZFS_TIME_ENCODE(&now, atime);
823 }
824
825 if (vap->va_mask & ATTR_MTIME) {
826 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
827 } else {
828 ZFS_TIME_ENCODE(&now, mtime);
829 }
830
831 /* Now add in all of the "SA" attributes */
832 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
833 &sa_hdl));
834
835 /*
836 * Setup the array of attributes to be replaced/set on the new file
837 *
838 * order for DMU_OT_ZNODE is critical since it needs to be constructed
839 * in the old znode_phys_t format. Don't change this ordering
840 */
841 sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
842
843 if (obj_type == DMU_OT_ZNODE) {
844 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
845 NULL, &atime, 16);
846 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
847 NULL, &mtime, 16);
848 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
849 NULL, &ctime, 16);
850 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
851 NULL, &crtime, 16);
852 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
853 NULL, &gen, 8);
854 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
855 NULL, &mode, 8);
856 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
857 NULL, &size, 8);
858 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
859 NULL, &parent, 8);
860 } else {
861 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
862 NULL, &mode, 8);
863 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
864 NULL, &size, 8);
865 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
866 NULL, &gen, 8);
867 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs),
868 NULL, &acl_ids->z_fuid, 8);
869 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs),
870 NULL, &acl_ids->z_fgid, 8);
871 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
872 NULL, &parent, 8);
873 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
874 NULL, &pflags, 8);
875 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
876 NULL, &atime, 16);
877 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
878 NULL, &mtime, 16);
879 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
880 NULL, &ctime, 16);
881 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
882 NULL, &crtime, 16);
883 }
884
885 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
886
887 if (obj_type == DMU_OT_ZNODE) {
888 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
889 &empty_xattr, 8);
890 } else if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
891 pflags & ZFS_PROJID) {
892 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PROJID(zfsvfs),
893 NULL, &projid, 8);
894 }
895 if (obj_type == DMU_OT_ZNODE ||
896 (S_ISBLK(vap->va_mode) || S_ISCHR(vap->va_mode))) {
897 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
898 NULL, &rdev, 8);
899 }
900 if (obj_type == DMU_OT_ZNODE) {
901 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
902 NULL, &pflags, 8);
903 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
904 &acl_ids->z_fuid, 8);
905 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
906 &acl_ids->z_fgid, 8);
907 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
908 sizeof (uint64_t) * 4);
909 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
910 &acl_phys, sizeof (zfs_acl_phys_t));
911 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
912 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
913 &acl_ids->z_aclp->z_acl_count, 8);
914 locate.cb_aclp = acl_ids->z_aclp;
915 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
916 zfs_acl_data_locator, &locate,
917 acl_ids->z_aclp->z_acl_bytes);
918 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
919 acl_ids->z_fuid, acl_ids->z_fgid);
920 }
921
922 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
923
924 if (!(flag & IS_ROOT_NODE)) {
925 /*
926 * The call to zfs_znode_alloc() may fail if memory is low
927 * via the call path: alloc_inode() -> inode_init_always() ->
928 * security_inode_alloc() -> inode_alloc_security(). Since
929 * the existing code is written such that zfs_mknode() can
930 * not fail retry until sufficient memory has been reclaimed.
931 */
932 do {
933 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
934 } while (*zpp == NULL);
935
936 VERIFY(*zpp != NULL);
937 VERIFY(dzp != NULL);
938 } else {
939 /*
940 * If we are creating the root node, the "parent" we
941 * passed in is the znode for the root.
942 */
943 *zpp = dzp;
944
945 (*zpp)->z_sa_hdl = sa_hdl;
946 }
947
948 (*zpp)->z_pflags = pflags;
949 (*zpp)->z_mode = ZTOI(*zpp)->i_mode = mode;
950 (*zpp)->z_dnodesize = dnodesize;
951 (*zpp)->z_projid = projid;
952
953 if (obj_type == DMU_OT_ZNODE ||
954 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
955 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
956 }
957 kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
958 zfs_znode_hold_exit(zfsvfs, zh);
959 }
960
961 /*
962 * Update in-core attributes. It is assumed the caller will be doing an
963 * sa_bulk_update to push the changes out.
964 */
965 void
zfs_xvattr_set(znode_t * zp,xvattr_t * xvap,dmu_tx_t * tx)966 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
967 {
968 xoptattr_t *xoap;
969 boolean_t update_inode = B_FALSE;
970
971 xoap = xva_getxoptattr(xvap);
972 ASSERT(xoap);
973
974 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
975 uint64_t times[2];
976 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
977 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(ZTOZSB(zp)),
978 ×, sizeof (times), tx);
979 XVA_SET_RTN(xvap, XAT_CREATETIME);
980 }
981 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
982 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
983 zp->z_pflags, tx);
984 XVA_SET_RTN(xvap, XAT_READONLY);
985 }
986 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
987 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
988 zp->z_pflags, tx);
989 XVA_SET_RTN(xvap, XAT_HIDDEN);
990 }
991 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
992 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
993 zp->z_pflags, tx);
994 XVA_SET_RTN(xvap, XAT_SYSTEM);
995 }
996 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
997 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
998 zp->z_pflags, tx);
999 XVA_SET_RTN(xvap, XAT_ARCHIVE);
1000 }
1001 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1002 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1003 zp->z_pflags, tx);
1004 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1005
1006 update_inode = B_TRUE;
1007 }
1008 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1009 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1010 zp->z_pflags, tx);
1011 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1012 }
1013 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1014 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1015 zp->z_pflags, tx);
1016 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1017
1018 update_inode = B_TRUE;
1019 }
1020 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1021 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1022 zp->z_pflags, tx);
1023 XVA_SET_RTN(xvap, XAT_NODUMP);
1024 }
1025 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1026 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1027 zp->z_pflags, tx);
1028 XVA_SET_RTN(xvap, XAT_OPAQUE);
1029 }
1030 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1031 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1032 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1033 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1034 }
1035 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1036 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1037 zp->z_pflags, tx);
1038 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1039 }
1040 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1041 zfs_sa_set_scanstamp(zp, xvap, tx);
1042 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1043 }
1044 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1045 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1046 zp->z_pflags, tx);
1047 XVA_SET_RTN(xvap, XAT_REPARSE);
1048 }
1049 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1050 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1051 zp->z_pflags, tx);
1052 XVA_SET_RTN(xvap, XAT_OFFLINE);
1053 }
1054 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1055 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1056 zp->z_pflags, tx);
1057 XVA_SET_RTN(xvap, XAT_SPARSE);
1058 }
1059 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
1060 ZFS_ATTR_SET(zp, ZFS_PROJINHERIT, xoap->xoa_projinherit,
1061 zp->z_pflags, tx);
1062 XVA_SET_RTN(xvap, XAT_PROJINHERIT);
1063 }
1064
1065 if (update_inode)
1066 zfs_set_inode_flags(zp, ZTOI(zp));
1067 }
1068
1069 int
zfs_zget(zfsvfs_t * zfsvfs,uint64_t obj_num,znode_t ** zpp)1070 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1071 {
1072 dmu_object_info_t doi;
1073 dmu_buf_t *db;
1074 znode_t *zp;
1075 znode_hold_t *zh;
1076 int err;
1077 sa_handle_t *hdl;
1078
1079 *zpp = NULL;
1080
1081 again:
1082 zh = zfs_znode_hold_enter(zfsvfs, obj_num);
1083
1084 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1085 if (err) {
1086 zfs_znode_hold_exit(zfsvfs, zh);
1087 return (err);
1088 }
1089
1090 dmu_object_info_from_db(db, &doi);
1091 if (doi.doi_bonus_type != DMU_OT_SA &&
1092 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1093 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1094 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1095 sa_buf_rele(db, NULL);
1096 zfs_znode_hold_exit(zfsvfs, zh);
1097 return (SET_ERROR(EINVAL));
1098 }
1099
1100 hdl = dmu_buf_get_user(db);
1101 if (hdl != NULL) {
1102 zp = sa_get_userdata(hdl);
1103
1104
1105 /*
1106 * Since "SA" does immediate eviction we
1107 * should never find a sa handle that doesn't
1108 * know about the znode.
1109 */
1110
1111 ASSERT3P(zp, !=, NULL);
1112
1113 mutex_enter(&zp->z_lock);
1114 ASSERT3U(zp->z_id, ==, obj_num);
1115 /*
1116 * If zp->z_unlinked is set, the znode is already marked
1117 * for deletion and should not be discovered. Check this
1118 * after checking igrab() due to fsetxattr() & O_TMPFILE.
1119 *
1120 * If igrab() returns NULL the VFS has independently
1121 * determined the inode should be evicted and has
1122 * called iput_final() to start the eviction process.
1123 * The SA handle is still valid but because the VFS
1124 * requires that the eviction succeed we must drop
1125 * our locks and references to allow the eviction to
1126 * complete. The zfs_zget() may then be retried.
1127 *
1128 * This unlikely case could be optimized by registering
1129 * a sops->drop_inode() callback. The callback would
1130 * need to detect the active SA hold thereby informing
1131 * the VFS that this inode should not be evicted.
1132 */
1133 if (igrab(ZTOI(zp)) == NULL) {
1134 if (zp->z_unlinked)
1135 err = SET_ERROR(ENOENT);
1136 else
1137 err = SET_ERROR(EAGAIN);
1138 } else {
1139 *zpp = zp;
1140 err = 0;
1141 }
1142
1143 mutex_exit(&zp->z_lock);
1144 sa_buf_rele(db, NULL);
1145 zfs_znode_hold_exit(zfsvfs, zh);
1146
1147 if (err == EAGAIN) {
1148 /* inode might need this to finish evict */
1149 cond_resched();
1150 goto again;
1151 }
1152 return (err);
1153 }
1154
1155 /*
1156 * Not found create new znode/vnode but only if file exists.
1157 *
1158 * There is a small window where zfs_vget() could
1159 * find this object while a file create is still in
1160 * progress. This is checked for in zfs_znode_alloc()
1161 *
1162 * if zfs_znode_alloc() fails it will drop the hold on the
1163 * bonus buffer.
1164 */
1165 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1166 doi.doi_bonus_type, NULL);
1167 if (zp == NULL) {
1168 err = SET_ERROR(ENOENT);
1169 } else {
1170 *zpp = zp;
1171 }
1172 zfs_znode_hold_exit(zfsvfs, zh);
1173 return (err);
1174 }
1175
1176 int
zfs_rezget(znode_t * zp)1177 zfs_rezget(znode_t *zp)
1178 {
1179 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1180 dmu_object_info_t doi;
1181 dmu_buf_t *db;
1182 uint64_t obj_num = zp->z_id;
1183 uint64_t mode;
1184 uint64_t links;
1185 sa_bulk_attr_t bulk[11];
1186 int err;
1187 int count = 0;
1188 uint64_t gen;
1189 uint64_t z_uid, z_gid;
1190 uint64_t atime[2], mtime[2], ctime[2], btime[2];
1191 inode_timespec_t tmp_ts;
1192 uint64_t projid = ZFS_DEFAULT_PROJID;
1193 znode_hold_t *zh;
1194
1195 /*
1196 * skip ctldir, otherwise they will always get invalidated. This will
1197 * cause funny behaviour for the mounted snapdirs. Especially for
1198 * Linux >= 3.18, d_invalidate will detach the mountpoint and prevent
1199 * anyone automount it again as long as someone is still using the
1200 * detached mount.
1201 */
1202 if (zp->z_is_ctldir)
1203 return (0);
1204
1205 zh = zfs_znode_hold_enter(zfsvfs, obj_num);
1206
1207 mutex_enter(&zp->z_acl_lock);
1208 if (zp->z_acl_cached) {
1209 zfs_acl_free(zp->z_acl_cached);
1210 zp->z_acl_cached = NULL;
1211 }
1212 mutex_exit(&zp->z_acl_lock);
1213
1214 rw_enter(&zp->z_xattr_lock, RW_WRITER);
1215 if (zp->z_xattr_cached) {
1216 nvlist_free(zp->z_xattr_cached);
1217 zp->z_xattr_cached = NULL;
1218 }
1219 rw_exit(&zp->z_xattr_lock);
1220
1221 ASSERT(zp->z_sa_hdl == NULL);
1222 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1223 if (err) {
1224 zfs_znode_hold_exit(zfsvfs, zh);
1225 return (err);
1226 }
1227
1228 dmu_object_info_from_db(db, &doi);
1229 if (doi.doi_bonus_type != DMU_OT_SA &&
1230 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1231 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1232 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1233 sa_buf_rele(db, NULL);
1234 zfs_znode_hold_exit(zfsvfs, zh);
1235 return (SET_ERROR(EINVAL));
1236 }
1237
1238 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1239
1240 /* reload cached values */
1241 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1242 &gen, sizeof (gen));
1243 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1244 &zp->z_size, sizeof (zp->z_size));
1245 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1246 &links, sizeof (links));
1247 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1248 &zp->z_pflags, sizeof (zp->z_pflags));
1249 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1250 &z_uid, sizeof (z_uid));
1251 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1252 &z_gid, sizeof (z_gid));
1253 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1254 &mode, sizeof (mode));
1255 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1256 &atime, 16);
1257 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
1258 &mtime, 16);
1259 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1260 &ctime, 16);
1261 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &btime, 16);
1262
1263 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1264 zfs_znode_dmu_fini(zp);
1265 zfs_znode_hold_exit(zfsvfs, zh);
1266 return (SET_ERROR(EIO));
1267 }
1268
1269 if (dmu_objset_projectquota_enabled(zfsvfs->z_os)) {
1270 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs),
1271 &projid, 8);
1272 if (err != 0 && err != ENOENT) {
1273 zfs_znode_dmu_fini(zp);
1274 zfs_znode_hold_exit(zfsvfs, zh);
1275 return (SET_ERROR(err));
1276 }
1277 }
1278
1279 zp->z_projid = projid;
1280 zp->z_mode = ZTOI(zp)->i_mode = mode;
1281 zfs_uid_write(ZTOI(zp), z_uid);
1282 zfs_gid_write(ZTOI(zp), z_gid);
1283
1284 ZFS_TIME_DECODE(&tmp_ts, atime);
1285 zpl_inode_set_atime_to_ts(ZTOI(zp), tmp_ts);
1286 ZFS_TIME_DECODE(&tmp_ts, mtime);
1287 zpl_inode_set_mtime_to_ts(ZTOI(zp), tmp_ts);
1288 ZFS_TIME_DECODE(&tmp_ts, ctime);
1289 zpl_inode_set_ctime_to_ts(ZTOI(zp), tmp_ts);
1290 ZFS_TIME_DECODE(&zp->z_btime, btime);
1291
1292 if ((uint32_t)gen != ZTOI(zp)->i_generation) {
1293 zfs_znode_dmu_fini(zp);
1294 zfs_znode_hold_exit(zfsvfs, zh);
1295 return (SET_ERROR(EIO));
1296 }
1297
1298 set_nlink(ZTOI(zp), (uint32_t)links);
1299 zfs_set_inode_flags(zp, ZTOI(zp));
1300
1301 zp->z_blksz = doi.doi_data_block_size;
1302 zp->z_atime_dirty = B_FALSE;
1303 zfs_znode_update_vfs(zp);
1304
1305 /*
1306 * If the file has zero links, then it has been unlinked on the send
1307 * side and it must be in the received unlinked set.
1308 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1309 * stale data and to prevent automatic removal of the file in
1310 * zfs_zinactive(). The file will be removed either when it is removed
1311 * on the send side and the next incremental stream is received or
1312 * when the unlinked set gets processed.
1313 */
1314 zp->z_unlinked = (ZTOI(zp)->i_nlink == 0);
1315 if (zp->z_unlinked)
1316 zfs_znode_dmu_fini(zp);
1317
1318 zfs_znode_hold_exit(zfsvfs, zh);
1319
1320 return (0);
1321 }
1322
1323 void
zfs_znode_delete(znode_t * zp,dmu_tx_t * tx)1324 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1325 {
1326 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1327 objset_t *os = zfsvfs->z_os;
1328 uint64_t obj = zp->z_id;
1329 uint64_t acl_obj = zfs_external_acl(zp);
1330 znode_hold_t *zh;
1331
1332 zh = zfs_znode_hold_enter(zfsvfs, obj);
1333 if (acl_obj) {
1334 VERIFY(!zp->z_is_sa);
1335 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1336 }
1337 VERIFY(0 == dmu_object_free(os, obj, tx));
1338 zfs_znode_dmu_fini(zp);
1339 zfs_znode_hold_exit(zfsvfs, zh);
1340 }
1341
1342 void
zfs_zinactive(znode_t * zp)1343 zfs_zinactive(znode_t *zp)
1344 {
1345 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1346 uint64_t z_id = zp->z_id;
1347 znode_hold_t *zh;
1348
1349 ASSERT(zp->z_sa_hdl);
1350
1351 /*
1352 * Don't allow a zfs_zget() while were trying to release this znode.
1353 */
1354 zh = zfs_znode_hold_enter(zfsvfs, z_id);
1355
1356 mutex_enter(&zp->z_lock);
1357
1358 /*
1359 * If this was the last reference to a file with no links, remove
1360 * the file from the file system unless the file system is mounted
1361 * read-only. That can happen, for example, if the file system was
1362 * originally read-write, the file was opened, then unlinked and
1363 * the file system was made read-only before the file was finally
1364 * closed. The file will remain in the unlinked set.
1365 */
1366 if (zp->z_unlinked) {
1367 ASSERT(!zfsvfs->z_issnap);
1368 if (!zfs_is_readonly(zfsvfs) && !zfs_unlink_suspend_progress) {
1369 mutex_exit(&zp->z_lock);
1370 zfs_znode_hold_exit(zfsvfs, zh);
1371 zfs_rmnode(zp);
1372 return;
1373 }
1374 }
1375
1376 mutex_exit(&zp->z_lock);
1377 zfs_znode_dmu_fini(zp);
1378
1379 zfs_znode_hold_exit(zfsvfs, zh);
1380 }
1381
1382 #if defined(HAVE_INODE_TIMESPEC64_TIMES)
1383 #define zfs_compare_timespec timespec64_compare
1384 #else
1385 #define zfs_compare_timespec timespec_compare
1386 #endif
1387
1388 /*
1389 * Determine whether the znode's atime must be updated. The logic mostly
1390 * duplicates the Linux kernel's relatime_need_update() functionality.
1391 * This function is only called if the underlying filesystem actually has
1392 * atime updates enabled.
1393 */
1394 boolean_t
zfs_relatime_need_update(const struct inode * ip)1395 zfs_relatime_need_update(const struct inode *ip)
1396 {
1397 inode_timespec_t now, tmp_atime, tmp_ts;
1398
1399 gethrestime(&now);
1400 tmp_atime = zpl_inode_get_atime(ip);
1401 /*
1402 * In relatime mode, only update the atime if the previous atime
1403 * is earlier than either the ctime or mtime or if at least a day
1404 * has passed since the last update of atime.
1405 */
1406 tmp_ts = zpl_inode_get_mtime(ip);
1407 if (zfs_compare_timespec(&tmp_ts, &tmp_atime) >= 0)
1408 return (B_TRUE);
1409
1410 tmp_ts = zpl_inode_get_ctime(ip);
1411 if (zfs_compare_timespec(&tmp_ts, &tmp_atime) >= 0)
1412 return (B_TRUE);
1413
1414 if ((hrtime_t)now.tv_sec - (hrtime_t)tmp_atime.tv_sec >= 24*60*60)
1415 return (B_TRUE);
1416
1417 return (B_FALSE);
1418 }
1419
1420 /*
1421 * Prepare to update znode time stamps.
1422 *
1423 * IN: zp - znode requiring timestamp update
1424 * flag - ATTR_MTIME, ATTR_CTIME flags
1425 *
1426 * OUT: zp - z_seq
1427 * mtime - new mtime
1428 * ctime - new ctime
1429 *
1430 * Note: We don't update atime here, because we rely on Linux VFS to do
1431 * atime updating.
1432 */
1433 void
zfs_tstamp_update_setup(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2])1434 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1435 uint64_t ctime[2])
1436 {
1437 inode_timespec_t now, tmp_ts;
1438
1439 gethrestime(&now);
1440
1441 zp->z_seq++;
1442
1443 if (flag & ATTR_MTIME) {
1444 ZFS_TIME_ENCODE(&now, mtime);
1445 ZFS_TIME_DECODE(&tmp_ts, mtime);
1446 zpl_inode_set_mtime_to_ts(ZTOI(zp), tmp_ts);
1447 if (ZTOZSB(zp)->z_use_fuids) {
1448 zp->z_pflags |= (ZFS_ARCHIVE |
1449 ZFS_AV_MODIFIED);
1450 }
1451 }
1452
1453 if (flag & ATTR_CTIME) {
1454 ZFS_TIME_ENCODE(&now, ctime);
1455 ZFS_TIME_DECODE(&tmp_ts, ctime);
1456 zpl_inode_set_ctime_to_ts(ZTOI(zp), tmp_ts);
1457 if (ZTOZSB(zp)->z_use_fuids)
1458 zp->z_pflags |= ZFS_ARCHIVE;
1459 }
1460 }
1461
1462 /*
1463 * Grow the block size for a file.
1464 *
1465 * IN: zp - znode of file to free data in.
1466 * size - requested block size
1467 * tx - open transaction.
1468 *
1469 * NOTE: this function assumes that the znode is write locked.
1470 */
1471 void
zfs_grow_blocksize(znode_t * zp,uint64_t size,dmu_tx_t * tx)1472 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1473 {
1474 int error;
1475 u_longlong_t dummy;
1476
1477 if (size <= zp->z_blksz)
1478 return;
1479 /*
1480 * If the file size is already greater than the current blocksize,
1481 * we will not grow. If there is more than one block in a file,
1482 * the blocksize cannot change.
1483 */
1484 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1485 return;
1486
1487 error = dmu_object_set_blocksize(ZTOZSB(zp)->z_os, zp->z_id,
1488 size, 0, tx);
1489
1490 if (error == ENOTSUP)
1491 return;
1492 ASSERT0(error);
1493
1494 /* What blocksize did we actually get? */
1495 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1496 }
1497
1498 /*
1499 * Increase the file length
1500 *
1501 * IN: zp - znode of file to free data in.
1502 * end - new end-of-file
1503 *
1504 * RETURN: 0 on success, error code on failure
1505 */
1506 static int
zfs_extend(znode_t * zp,uint64_t end)1507 zfs_extend(znode_t *zp, uint64_t end)
1508 {
1509 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1510 dmu_tx_t *tx;
1511 zfs_locked_range_t *lr;
1512 uint64_t newblksz;
1513 int error;
1514
1515 /*
1516 * We will change zp_size, lock the whole file.
1517 */
1518 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1519
1520 /*
1521 * Nothing to do if file already at desired length.
1522 */
1523 if (end <= zp->z_size) {
1524 zfs_rangelock_exit(lr);
1525 return (0);
1526 }
1527 tx = dmu_tx_create(zfsvfs->z_os);
1528 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1529 zfs_sa_upgrade_txholds(tx, zp);
1530 if (end > zp->z_blksz &&
1531 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1532 /*
1533 * We are growing the file past the current block size.
1534 */
1535 if (zp->z_blksz > ZTOZSB(zp)->z_max_blksz) {
1536 /*
1537 * File's blocksize is already larger than the
1538 * "recordsize" property. Only let it grow to
1539 * the next power of 2.
1540 */
1541 ASSERT(!ISP2(zp->z_blksz));
1542 newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1543 } else {
1544 newblksz = MIN(end, ZTOZSB(zp)->z_max_blksz);
1545 }
1546 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1547 } else {
1548 newblksz = 0;
1549 }
1550
1551 error = dmu_tx_assign(tx, TXG_WAIT);
1552 if (error) {
1553 dmu_tx_abort(tx);
1554 zfs_rangelock_exit(lr);
1555 return (error);
1556 }
1557
1558 if (newblksz)
1559 zfs_grow_blocksize(zp, newblksz, tx);
1560
1561 zp->z_size = end;
1562
1563 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(ZTOZSB(zp)),
1564 &zp->z_size, sizeof (zp->z_size), tx));
1565
1566 zfs_rangelock_exit(lr);
1567
1568 dmu_tx_commit(tx);
1569
1570 return (0);
1571 }
1572
1573 /*
1574 * zfs_zero_partial_page - Modeled after update_pages() but
1575 * with different arguments and semantics for use by zfs_freesp().
1576 *
1577 * Zeroes a piece of a single page cache entry for zp at offset
1578 * start and length len.
1579 *
1580 * Caller must acquire a range lock on the file for the region
1581 * being zeroed in order that the ARC and page cache stay in sync.
1582 */
1583 static void
zfs_zero_partial_page(znode_t * zp,uint64_t start,uint64_t len)1584 zfs_zero_partial_page(znode_t *zp, uint64_t start, uint64_t len)
1585 {
1586 struct address_space *mp = ZTOI(zp)->i_mapping;
1587 struct page *pp;
1588 int64_t off;
1589 void *pb;
1590
1591 ASSERT((start & PAGE_MASK) == ((start + len - 1) & PAGE_MASK));
1592
1593 off = start & (PAGE_SIZE - 1);
1594 start &= PAGE_MASK;
1595
1596 pp = find_lock_page(mp, start >> PAGE_SHIFT);
1597 if (pp) {
1598 if (mapping_writably_mapped(mp))
1599 flush_dcache_page(pp);
1600
1601 pb = kmap(pp);
1602 bzero(pb + off, len);
1603 kunmap(pp);
1604
1605 if (mapping_writably_mapped(mp))
1606 flush_dcache_page(pp);
1607
1608 mark_page_accessed(pp);
1609 SetPageUptodate(pp);
1610 ClearPageError(pp);
1611 unlock_page(pp);
1612 put_page(pp);
1613 }
1614 }
1615
1616 /*
1617 * Free space in a file.
1618 *
1619 * IN: zp - znode of file to free data in.
1620 * off - start of section to free.
1621 * len - length of section to free.
1622 *
1623 * RETURN: 0 on success, error code on failure
1624 */
1625 static int
zfs_free_range(znode_t * zp,uint64_t off,uint64_t len)1626 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1627 {
1628 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1629 zfs_locked_range_t *lr;
1630 int error;
1631
1632 /*
1633 * Lock the range being freed.
1634 */
1635 lr = zfs_rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER);
1636
1637 /*
1638 * Nothing to do if file already at desired length.
1639 */
1640 if (off >= zp->z_size) {
1641 zfs_rangelock_exit(lr);
1642 return (0);
1643 }
1644
1645 if (off + len > zp->z_size)
1646 len = zp->z_size - off;
1647
1648 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1649
1650 /*
1651 * Zero partial page cache entries. This must be done under a
1652 * range lock in order to keep the ARC and page cache in sync.
1653 */
1654 if (zn_has_cached_data(zp, off, off + len - 1)) {
1655 loff_t first_page, last_page, page_len;
1656 loff_t first_page_offset, last_page_offset;
1657
1658 /* first possible full page in hole */
1659 first_page = (off + PAGE_SIZE - 1) >> PAGE_SHIFT;
1660 /* last page of hole */
1661 last_page = (off + len) >> PAGE_SHIFT;
1662
1663 /* offset of first_page */
1664 first_page_offset = first_page << PAGE_SHIFT;
1665 /* offset of last_page */
1666 last_page_offset = last_page << PAGE_SHIFT;
1667
1668 /* truncate whole pages */
1669 if (last_page_offset > first_page_offset) {
1670 truncate_inode_pages_range(ZTOI(zp)->i_mapping,
1671 first_page_offset, last_page_offset - 1);
1672 }
1673
1674 /* truncate sub-page ranges */
1675 if (first_page > last_page) {
1676 /* entire punched area within a single page */
1677 zfs_zero_partial_page(zp, off, len);
1678 } else {
1679 /* beginning of punched area at the end of a page */
1680 page_len = first_page_offset - off;
1681 if (page_len > 0)
1682 zfs_zero_partial_page(zp, off, page_len);
1683
1684 /* end of punched area at the beginning of a page */
1685 page_len = off + len - last_page_offset;
1686 if (page_len > 0)
1687 zfs_zero_partial_page(zp, last_page_offset,
1688 page_len);
1689 }
1690 }
1691 zfs_rangelock_exit(lr);
1692
1693 return (error);
1694 }
1695
1696 /*
1697 * Truncate a file
1698 *
1699 * IN: zp - znode of file to free data in.
1700 * end - new end-of-file.
1701 *
1702 * RETURN: 0 on success, error code on failure
1703 */
1704 static int
zfs_trunc(znode_t * zp,uint64_t end)1705 zfs_trunc(znode_t *zp, uint64_t end)
1706 {
1707 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1708 dmu_tx_t *tx;
1709 zfs_locked_range_t *lr;
1710 int error;
1711 sa_bulk_attr_t bulk[2];
1712 int count = 0;
1713
1714 /*
1715 * We will change zp_size, lock the whole file.
1716 */
1717 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1718
1719 /*
1720 * Nothing to do if file already at desired length.
1721 */
1722 if (end >= zp->z_size) {
1723 zfs_rangelock_exit(lr);
1724 return (0);
1725 }
1726
1727 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1728 DMU_OBJECT_END);
1729 if (error) {
1730 zfs_rangelock_exit(lr);
1731 return (error);
1732 }
1733 tx = dmu_tx_create(zfsvfs->z_os);
1734 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1735 zfs_sa_upgrade_txholds(tx, zp);
1736 dmu_tx_mark_netfree(tx);
1737 error = dmu_tx_assign(tx, TXG_WAIT);
1738 if (error) {
1739 dmu_tx_abort(tx);
1740 zfs_rangelock_exit(lr);
1741 return (error);
1742 }
1743
1744 zp->z_size = end;
1745 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1746 NULL, &zp->z_size, sizeof (zp->z_size));
1747
1748 if (end == 0) {
1749 zp->z_pflags &= ~ZFS_SPARSE;
1750 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1751 NULL, &zp->z_pflags, 8);
1752 }
1753 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1754
1755 dmu_tx_commit(tx);
1756 zfs_rangelock_exit(lr);
1757
1758 return (0);
1759 }
1760
1761 /*
1762 * Free space in a file
1763 *
1764 * IN: zp - znode of file to free data in.
1765 * off - start of range
1766 * len - end of range (0 => EOF)
1767 * flag - current file open mode flags.
1768 * log - TRUE if this action should be logged
1769 *
1770 * RETURN: 0 on success, error code on failure
1771 */
1772 int
zfs_freesp(znode_t * zp,uint64_t off,uint64_t len,int flag,boolean_t log)1773 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1774 {
1775 dmu_tx_t *tx;
1776 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1777 zilog_t *zilog = zfsvfs->z_log;
1778 uint64_t mode;
1779 uint64_t mtime[2], ctime[2];
1780 sa_bulk_attr_t bulk[3];
1781 int count = 0;
1782 int error;
1783
1784 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1785 sizeof (mode))) != 0)
1786 return (error);
1787
1788 if (off > zp->z_size) {
1789 error = zfs_extend(zp, off+len);
1790 if (error == 0 && log)
1791 goto log;
1792 goto out;
1793 }
1794
1795 if (len == 0) {
1796 error = zfs_trunc(zp, off);
1797 } else {
1798 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1799 off + len > zp->z_size)
1800 error = zfs_extend(zp, off+len);
1801 }
1802 if (error || !log)
1803 goto out;
1804 log:
1805 tx = dmu_tx_create(zfsvfs->z_os);
1806 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1807 zfs_sa_upgrade_txholds(tx, zp);
1808 error = dmu_tx_assign(tx, TXG_WAIT);
1809 if (error) {
1810 dmu_tx_abort(tx);
1811 goto out;
1812 }
1813
1814 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1815 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1816 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1817 NULL, &zp->z_pflags, 8);
1818 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
1819 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1820 ASSERT(error == 0);
1821
1822 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1823
1824 dmu_tx_commit(tx);
1825
1826 zfs_znode_update_vfs(zp);
1827 error = 0;
1828
1829 out:
1830 /*
1831 * Truncate the page cache - for file truncate operations, use
1832 * the purpose-built API for truncations. For punching operations,
1833 * the truncation is handled under a range lock in zfs_free_range.
1834 */
1835 if (len == 0)
1836 truncate_setsize(ZTOI(zp), off);
1837 return (error);
1838 }
1839
1840 void
zfs_create_fs(objset_t * os,cred_t * cr,nvlist_t * zplprops,dmu_tx_t * tx)1841 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1842 {
1843 struct super_block *sb;
1844 zfsvfs_t *zfsvfs;
1845 uint64_t moid, obj, sa_obj, version;
1846 uint64_t sense = ZFS_CASE_SENSITIVE;
1847 uint64_t norm = 0;
1848 nvpair_t *elem;
1849 int size;
1850 int error;
1851 int i;
1852 znode_t *rootzp = NULL;
1853 vattr_t vattr;
1854 znode_t *zp;
1855 zfs_acl_ids_t acl_ids;
1856
1857 /*
1858 * First attempt to create master node.
1859 */
1860 /*
1861 * In an empty objset, there are no blocks to read and thus
1862 * there can be no i/o errors (which we assert below).
1863 */
1864 moid = MASTER_NODE_OBJ;
1865 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1866 DMU_OT_NONE, 0, tx);
1867 ASSERT(error == 0);
1868
1869 /*
1870 * Set starting attributes.
1871 */
1872 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1873 elem = NULL;
1874 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1875 /* For the moment we expect all zpl props to be uint64_ts */
1876 uint64_t val;
1877 char *name;
1878
1879 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1880 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1881 name = nvpair_name(elem);
1882 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1883 if (val < version)
1884 version = val;
1885 } else {
1886 error = zap_update(os, moid, name, 8, 1, &val, tx);
1887 }
1888 ASSERT(error == 0);
1889 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1890 norm = val;
1891 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1892 sense = val;
1893 }
1894 ASSERT(version != 0);
1895 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1896
1897 /*
1898 * Create zap object used for SA attribute registration
1899 */
1900
1901 if (version >= ZPL_VERSION_SA) {
1902 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1903 DMU_OT_NONE, 0, tx);
1904 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1905 ASSERT(error == 0);
1906 } else {
1907 sa_obj = 0;
1908 }
1909 /*
1910 * Create a delete queue.
1911 */
1912 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1913
1914 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1915 ASSERT(error == 0);
1916
1917 /*
1918 * Create root znode. Create minimal znode/inode/zfsvfs/sb
1919 * to allow zfs_mknode to work.
1920 */
1921 vattr.va_mask = ATTR_MODE|ATTR_UID|ATTR_GID;
1922 vattr.va_mode = S_IFDIR|0755;
1923 vattr.va_uid = crgetuid(cr);
1924 vattr.va_gid = crgetgid(cr);
1925
1926 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1927 rootzp->z_unlinked = B_FALSE;
1928 rootzp->z_atime_dirty = B_FALSE;
1929 rootzp->z_is_sa = USE_SA(version, os);
1930 rootzp->z_pflags = 0;
1931
1932 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1933 zfsvfs->z_os = os;
1934 zfsvfs->z_parent = zfsvfs;
1935 zfsvfs->z_version = version;
1936 zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1937 zfsvfs->z_use_sa = USE_SA(version, os);
1938 zfsvfs->z_norm = norm;
1939
1940 sb = kmem_zalloc(sizeof (struct super_block), KM_SLEEP);
1941 sb->s_fs_info = zfsvfs;
1942
1943 ZTOI(rootzp)->i_sb = sb;
1944
1945 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1946 &zfsvfs->z_attr_table);
1947
1948 ASSERT(error == 0);
1949
1950 /*
1951 * Fold case on file systems that are always or sometimes case
1952 * insensitive.
1953 */
1954 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1955 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1956
1957 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1958 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1959 offsetof(znode_t, z_link_node));
1960
1961 size = MIN(1 << (highbit64(zfs_object_mutex_size)-1), ZFS_OBJ_MTX_MAX);
1962 zfsvfs->z_hold_size = size;
1963 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
1964 KM_SLEEP);
1965 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
1966 for (i = 0; i != size; i++) {
1967 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
1968 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
1969 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
1970 }
1971
1972 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1973 cr, NULL, &acl_ids));
1974 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1975 ASSERT3P(zp, ==, rootzp);
1976 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1977 ASSERT(error == 0);
1978 zfs_acl_ids_free(&acl_ids);
1979
1980 atomic_set(&ZTOI(rootzp)->i_count, 0);
1981 sa_handle_destroy(rootzp->z_sa_hdl);
1982 kmem_cache_free(znode_cache, rootzp);
1983
1984 for (i = 0; i != size; i++) {
1985 avl_destroy(&zfsvfs->z_hold_trees[i]);
1986 mutex_destroy(&zfsvfs->z_hold_locks[i]);
1987 }
1988
1989 mutex_destroy(&zfsvfs->z_znodes_lock);
1990
1991 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
1992 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
1993 kmem_free(sb, sizeof (struct super_block));
1994 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1995 }
1996 #endif /* _KERNEL */
1997
1998 static int
zfs_sa_setup(objset_t * osp,sa_attr_type_t ** sa_table)1999 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
2000 {
2001 uint64_t sa_obj = 0;
2002 int error;
2003
2004 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
2005 if (error != 0 && error != ENOENT)
2006 return (error);
2007
2008 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
2009 return (error);
2010 }
2011
2012 static int
zfs_grab_sa_handle(objset_t * osp,uint64_t obj,sa_handle_t ** hdlp,dmu_buf_t ** db,void * tag)2013 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
2014 dmu_buf_t **db, void *tag)
2015 {
2016 dmu_object_info_t doi;
2017 int error;
2018
2019 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
2020 return (error);
2021
2022 dmu_object_info_from_db(*db, &doi);
2023 if ((doi.doi_bonus_type != DMU_OT_SA &&
2024 doi.doi_bonus_type != DMU_OT_ZNODE) ||
2025 (doi.doi_bonus_type == DMU_OT_ZNODE &&
2026 doi.doi_bonus_size < sizeof (znode_phys_t))) {
2027 sa_buf_rele(*db, tag);
2028 return (SET_ERROR(ENOTSUP));
2029 }
2030
2031 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
2032 if (error != 0) {
2033 sa_buf_rele(*db, tag);
2034 return (error);
2035 }
2036
2037 return (0);
2038 }
2039
2040 static void
zfs_release_sa_handle(sa_handle_t * hdl,dmu_buf_t * db,void * tag)2041 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
2042 {
2043 sa_handle_destroy(hdl);
2044 sa_buf_rele(db, tag);
2045 }
2046
2047 /*
2048 * Given an object number, return its parent object number and whether
2049 * or not the object is an extended attribute directory.
2050 */
2051 static int
zfs_obj_to_pobj(objset_t * osp,sa_handle_t * hdl,sa_attr_type_t * sa_table,uint64_t * pobjp,int * is_xattrdir)2052 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
2053 uint64_t *pobjp, int *is_xattrdir)
2054 {
2055 uint64_t parent;
2056 uint64_t pflags;
2057 uint64_t mode;
2058 uint64_t parent_mode;
2059 sa_bulk_attr_t bulk[3];
2060 sa_handle_t *sa_hdl;
2061 dmu_buf_t *sa_db;
2062 int count = 0;
2063 int error;
2064
2065 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
2066 &parent, sizeof (parent));
2067 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
2068 &pflags, sizeof (pflags));
2069 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2070 &mode, sizeof (mode));
2071
2072 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
2073 return (error);
2074
2075 /*
2076 * When a link is removed its parent pointer is not changed and will
2077 * be invalid. There are two cases where a link is removed but the
2078 * file stays around, when it goes to the delete queue and when there
2079 * are additional links.
2080 */
2081 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
2082 if (error != 0)
2083 return (error);
2084
2085 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
2086 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2087 if (error != 0)
2088 return (error);
2089
2090 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
2091
2092 /*
2093 * Extended attributes can be applied to files, directories, etc.
2094 * Otherwise the parent must be a directory.
2095 */
2096 if (!*is_xattrdir && !S_ISDIR(parent_mode))
2097 return (SET_ERROR(EINVAL));
2098
2099 *pobjp = parent;
2100
2101 return (0);
2102 }
2103
2104 /*
2105 * Given an object number, return some zpl level statistics
2106 */
2107 static int
zfs_obj_to_stats_impl(sa_handle_t * hdl,sa_attr_type_t * sa_table,zfs_stat_t * sb)2108 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2109 zfs_stat_t *sb)
2110 {
2111 sa_bulk_attr_t bulk[4];
2112 int count = 0;
2113
2114 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2115 &sb->zs_mode, sizeof (sb->zs_mode));
2116 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2117 &sb->zs_gen, sizeof (sb->zs_gen));
2118 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2119 &sb->zs_links, sizeof (sb->zs_links));
2120 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2121 &sb->zs_ctime, sizeof (sb->zs_ctime));
2122
2123 return (sa_bulk_lookup(hdl, bulk, count));
2124 }
2125
2126 static int
zfs_obj_to_path_impl(objset_t * osp,uint64_t obj,sa_handle_t * hdl,sa_attr_type_t * sa_table,char * buf,int len)2127 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2128 sa_attr_type_t *sa_table, char *buf, int len)
2129 {
2130 sa_handle_t *sa_hdl;
2131 sa_handle_t *prevhdl = NULL;
2132 dmu_buf_t *prevdb = NULL;
2133 dmu_buf_t *sa_db = NULL;
2134 char *path = buf + len - 1;
2135 int error;
2136
2137 *path = '\0';
2138 sa_hdl = hdl;
2139
2140 uint64_t deleteq_obj;
2141 VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ,
2142 ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj));
2143 error = zap_lookup_int(osp, deleteq_obj, obj);
2144 if (error == 0) {
2145 return (ESTALE);
2146 } else if (error != ENOENT) {
2147 return (error);
2148 }
2149 error = 0;
2150
2151 for (;;) {
2152 uint64_t pobj = 0;
2153 char component[MAXNAMELEN + 2];
2154 size_t complen;
2155 int is_xattrdir = 0;
2156
2157 if (prevdb) {
2158 ASSERT(prevhdl != NULL);
2159 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2160 }
2161
2162 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2163 &is_xattrdir)) != 0)
2164 break;
2165
2166 if (pobj == obj) {
2167 if (path[0] != '/')
2168 *--path = '/';
2169 break;
2170 }
2171
2172 component[0] = '/';
2173 if (is_xattrdir) {
2174 (void) sprintf(component + 1, "<xattrdir>");
2175 } else {
2176 error = zap_value_search(osp, pobj, obj,
2177 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2178 if (error != 0)
2179 break;
2180 }
2181
2182 complen = strlen(component);
2183 path -= complen;
2184 ASSERT(path >= buf);
2185 bcopy(component, path, complen);
2186 obj = pobj;
2187
2188 if (sa_hdl != hdl) {
2189 prevhdl = sa_hdl;
2190 prevdb = sa_db;
2191 }
2192 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2193 if (error != 0) {
2194 sa_hdl = prevhdl;
2195 sa_db = prevdb;
2196 break;
2197 }
2198 }
2199
2200 if (sa_hdl != NULL && sa_hdl != hdl) {
2201 ASSERT(sa_db != NULL);
2202 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2203 }
2204
2205 if (error == 0)
2206 (void) memmove(buf, path, buf + len - path);
2207
2208 return (error);
2209 }
2210
2211 int
zfs_obj_to_path(objset_t * osp,uint64_t obj,char * buf,int len)2212 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2213 {
2214 sa_attr_type_t *sa_table;
2215 sa_handle_t *hdl;
2216 dmu_buf_t *db;
2217 int error;
2218
2219 error = zfs_sa_setup(osp, &sa_table);
2220 if (error != 0)
2221 return (error);
2222
2223 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2224 if (error != 0)
2225 return (error);
2226
2227 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2228
2229 zfs_release_sa_handle(hdl, db, FTAG);
2230 return (error);
2231 }
2232
2233 int
zfs_obj_to_stats(objset_t * osp,uint64_t obj,zfs_stat_t * sb,char * buf,int len)2234 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2235 char *buf, int len)
2236 {
2237 char *path = buf + len - 1;
2238 sa_attr_type_t *sa_table;
2239 sa_handle_t *hdl;
2240 dmu_buf_t *db;
2241 int error;
2242
2243 *path = '\0';
2244
2245 error = zfs_sa_setup(osp, &sa_table);
2246 if (error != 0)
2247 return (error);
2248
2249 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2250 if (error != 0)
2251 return (error);
2252
2253 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2254 if (error != 0) {
2255 zfs_release_sa_handle(hdl, db, FTAG);
2256 return (error);
2257 }
2258
2259 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2260
2261 zfs_release_sa_handle(hdl, db, FTAG);
2262 return (error);
2263 }
2264
2265 #if defined(_KERNEL)
2266 EXPORT_SYMBOL(zfs_create_fs);
2267 EXPORT_SYMBOL(zfs_obj_to_path);
2268
2269 /* CSTYLED */
2270 module_param(zfs_object_mutex_size, uint, 0644);
2271 MODULE_PARM_DESC(zfs_object_mutex_size, "Size of znode hold array");
2272 module_param(zfs_unlink_suspend_progress, int, 0644);
2273 MODULE_PARM_DESC(zfs_unlink_suspend_progress, "Set to prevent async unlinks "
2274 "(debug - leaks space into the unlinked set)");
2275 #endif
2276