]> git.proxmox.com Git - mirror_zfs.git/blob - module/os/linux/zfs/zfs_znode.c
projects / mirror_zfs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Remove atomics from zh_refcount
[mirror_zfs.git] / module / os / linux / zfs / zfs_znode.c
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 https://opensource.org/licenses/CDDL-1.0.
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 static 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
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 static int
117 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
118 {
119 (void) arg, (void) kmflags;
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 static void
144 zfs_znode_cache_destructor(void *buf, void *arg)
145 {
146 (void) arg;
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
166 zfs_znode_hold_cache_constructor(void *buf, void *arg, int kmflags)
167 {
168 (void) arg, (void) kmflags;
169 znode_hold_t *zh = buf;
170
171 mutex_init(&zh->zh_lock, NULL, MUTEX_DEFAULT, NULL);
172 zh->zh_refcount = 0;
173
174 return (0);
175 }
176
177 static void
178 zfs_znode_hold_cache_destructor(void *buf, void *arg)
179 {
180 (void) arg;
181 znode_hold_t *zh = buf;
182
183 mutex_destroy(&zh->zh_lock);
184 }
185
186 void
187 zfs_znode_init(void)
188 {
189 /*
190 * Initialize zcache. The KMC_SLAB hint is used in order that it be
191 * backed by kmalloc() when on the Linux slab in order that any
192 * wait_on_bit() operations on the related inode operate properly.
193 */
194 ASSERT(znode_cache == NULL);
195 znode_cache = kmem_cache_create("zfs_znode_cache",
196 sizeof (znode_t), 0, zfs_znode_cache_constructor,
197 zfs_znode_cache_destructor, NULL, NULL, NULL, KMC_SLAB);
198
199 ASSERT(znode_hold_cache == NULL);
200 znode_hold_cache = kmem_cache_create("zfs_znode_hold_cache",
201 sizeof (znode_hold_t), 0, zfs_znode_hold_cache_constructor,
202 zfs_znode_hold_cache_destructor, NULL, NULL, NULL, 0);
203 }
204
205 void
206 zfs_znode_fini(void)
207 {
208 /*
209 * Cleanup zcache
210 */
211 if (znode_cache)
212 kmem_cache_destroy(znode_cache);
213 znode_cache = NULL;
214
215 if (znode_hold_cache)
216 kmem_cache_destroy(znode_hold_cache);
217 znode_hold_cache = NULL;
218 }
219
220 /*
221 * The zfs_znode_hold_enter() / zfs_znode_hold_exit() functions are used to
222 * serialize access to a znode and its SA buffer while the object is being
223 * created or destroyed. This kind of locking would normally reside in the
224 * znode itself but in this case that's impossible because the znode and SA
225 * buffer may not yet exist. Therefore the locking is handled externally
226 * with an array of mutexes and AVLs trees which contain per-object locks.
227 *
228 * In zfs_znode_hold_enter() a per-object lock is created as needed, inserted
229 * in to the correct AVL tree and finally the per-object lock is held. In
230 * zfs_znode_hold_exit() the process is reversed. The per-object lock is
231 * released, removed from the AVL tree and destroyed if there are no waiters.
232 *
233 * This scheme has two important properties:
234 *
235 * 1) No memory allocations are performed while holding one of the z_hold_locks.
236 * This ensures evict(), which can be called from direct memory reclaim, will
237 * never block waiting on a z_hold_locks which just happens to have hashed
238 * to the same index.
239 *
240 * 2) All locks used to serialize access to an object are per-object and never
241 * shared. This minimizes lock contention without creating a large number
242 * of dedicated locks.
243 *
244 * On the downside it does require znode_lock_t structures to be frequently
245 * allocated and freed. However, because these are backed by a kmem cache
246 * and very short lived this cost is minimal.
247 */
248 int
249 zfs_znode_hold_compare(const void *a, const void *b)
250 {
251 const znode_hold_t *zh_a = (const znode_hold_t *)a;
252 const znode_hold_t *zh_b = (const znode_hold_t *)b;
253
254 return (TREE_CMP(zh_a->zh_obj, zh_b->zh_obj));
255 }
256
257 static boolean_t __maybe_unused
258 zfs_znode_held(zfsvfs_t *zfsvfs, uint64_t obj)
259 {
260 znode_hold_t *zh, search;
261 int i = ZFS_OBJ_HASH(zfsvfs, obj);
262 boolean_t held;
263
264 search.zh_obj = obj;
265
266 mutex_enter(&zfsvfs->z_hold_locks[i]);
267 zh = avl_find(&zfsvfs->z_hold_trees[i], &search, NULL);
268 held = (zh && MUTEX_HELD(&zh->zh_lock)) ? B_TRUE : B_FALSE;
269 mutex_exit(&zfsvfs->z_hold_locks[i]);
270
271 return (held);
272 }
273
274 static znode_hold_t *
275 zfs_znode_hold_enter(zfsvfs_t *zfsvfs, uint64_t obj)
276 {
277 znode_hold_t *zh, *zh_new, search;
278 int i = ZFS_OBJ_HASH(zfsvfs, obj);
279 boolean_t found = B_FALSE;
280
281 zh_new = kmem_cache_alloc(znode_hold_cache, KM_SLEEP);
282 search.zh_obj = obj;
283
284 mutex_enter(&zfsvfs->z_hold_locks[i]);
285 zh = avl_find(&zfsvfs->z_hold_trees[i], &search, NULL);
286 if (likely(zh == NULL)) {
287 zh = zh_new;
288 zh->zh_obj = obj;
289 avl_add(&zfsvfs->z_hold_trees[i], zh);
290 } else {
291 ASSERT3U(zh->zh_obj, ==, obj);
292 found = B_TRUE;
293 }
294 zh->zh_refcount++;
295 ASSERT3S(zh->zh_refcount, >, 0);
296 mutex_exit(&zfsvfs->z_hold_locks[i]);
297
298 if (found == B_TRUE)
299 kmem_cache_free(znode_hold_cache, zh_new);
300
301 ASSERT(MUTEX_NOT_HELD(&zh->zh_lock));
302 mutex_enter(&zh->zh_lock);
303
304 return (zh);
305 }
306
307 static void
308 zfs_znode_hold_exit(zfsvfs_t *zfsvfs, znode_hold_t *zh)
309 {
310 int i = ZFS_OBJ_HASH(zfsvfs, zh->zh_obj);
311 boolean_t remove = B_FALSE;
312
313 ASSERT(zfs_znode_held(zfsvfs, zh->zh_obj));
314 mutex_exit(&zh->zh_lock);
315
316 mutex_enter(&zfsvfs->z_hold_locks[i]);
317 ASSERT3S(zh->zh_refcount, >, 0);
318 if (--zh->zh_refcount == 0) {
319 avl_remove(&zfsvfs->z_hold_trees[i], zh);
320 remove = B_TRUE;
321 }
322 mutex_exit(&zfsvfs->z_hold_locks[i]);
323
324 if (remove == B_TRUE)
325 kmem_cache_free(znode_hold_cache, zh);
326 }
327
328 dev_t
329 zfs_cmpldev(uint64_t dev)
330 {
331 return (dev);
332 }
333
334 static void
335 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
336 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
337 {
338 ASSERT(zfs_znode_held(zfsvfs, zp->z_id));
339
340 mutex_enter(&zp->z_lock);
341
342 ASSERT(zp->z_sa_hdl == NULL);
343 ASSERT(zp->z_acl_cached == NULL);
344 if (sa_hdl == NULL) {
345 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
346 SA_HDL_SHARED, &zp->z_sa_hdl));
347 } else {
348 zp->z_sa_hdl = sa_hdl;
349 sa_set_userp(sa_hdl, zp);
350 }
351
352 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
353
354 mutex_exit(&zp->z_lock);
355 }
356
357 void
358 zfs_znode_dmu_fini(znode_t *zp)
359 {
360 ASSERT(zfs_znode_held(ZTOZSB(zp), zp->z_id) || zp->z_unlinked ||
361 RW_WRITE_HELD(&ZTOZSB(zp)->z_teardown_inactive_lock));
362
363 sa_handle_destroy(zp->z_sa_hdl);
364 zp->z_sa_hdl = NULL;
365 }
366
367 /*
368 * Called by new_inode() to allocate a new inode.
369 */
370 int
371 zfs_inode_alloc(struct super_block *sb, struct inode **ip)
372 {
373 znode_t *zp;
374
375 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
376 *ip = ZTOI(zp);
377
378 return (0);
379 }
380
381 /*
382 * Called in multiple places when an inode should be destroyed.
383 */
384 void
385 zfs_inode_destroy(struct inode *ip)
386 {
387 znode_t *zp = ITOZ(ip);
388 zfsvfs_t *zfsvfs = ZTOZSB(zp);
389
390 mutex_enter(&zfsvfs->z_znodes_lock);
391 if (list_link_active(&zp->z_link_node)) {
392 list_remove(&zfsvfs->z_all_znodes, zp);
393 zfsvfs->z_nr_znodes--;
394 }
395 mutex_exit(&zfsvfs->z_znodes_lock);
396
397 if (zp->z_acl_cached) {
398 zfs_acl_free(zp->z_acl_cached);
399 zp->z_acl_cached = NULL;
400 }
401
402 if (zp->z_xattr_cached) {
403 nvlist_free(zp->z_xattr_cached);
404 zp->z_xattr_cached = NULL;
405 }
406
407 kmem_cache_free(znode_cache, zp);
408 }
409
410 static void
411 zfs_inode_set_ops(zfsvfs_t *zfsvfs, struct inode *ip)
412 {
413 uint64_t rdev = 0;
414
415 switch (ip->i_mode & S_IFMT) {
416 case S_IFREG:
417 ip->i_op = &zpl_inode_operations;
418 ip->i_fop = &zpl_file_operations;
419 ip->i_mapping->a_ops = &zpl_address_space_operations;
420 break;
421
422 case S_IFDIR:
423 #ifdef HAVE_RENAME2_OPERATIONS_WRAPPER
424 ip->i_flags |= S_IOPS_WRAPPER;
425 ip->i_op = &zpl_dir_inode_operations.ops;
426 #else
427 ip->i_op = &zpl_dir_inode_operations;
428 #endif
429 ip->i_fop = &zpl_dir_file_operations;
430 ITOZ(ip)->z_zn_prefetch = B_TRUE;
431 break;
432
433 case S_IFLNK:
434 ip->i_op = &zpl_symlink_inode_operations;
435 break;
436
437 /*
438 * rdev is only stored in a SA only for device files.
439 */
440 case S_IFCHR:
441 case S_IFBLK:
442 (void) sa_lookup(ITOZ(ip)->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), &rdev,
443 sizeof (rdev));
444 zfs_fallthrough;
445 case S_IFIFO:
446 case S_IFSOCK:
447 init_special_inode(ip, ip->i_mode, rdev);
448 ip->i_op = &zpl_special_inode_operations;
449 break;
450
451 default:
452 zfs_panic_recover("inode %llu has invalid mode: 0x%x\n",
453 (u_longlong_t)ip->i_ino, ip->i_mode);
454
455 /* Assume the inode is a file and attempt to continue */
456 ip->i_mode = S_IFREG | 0644;
457 ip->i_op = &zpl_inode_operations;
458 ip->i_fop = &zpl_file_operations;
459 ip->i_mapping->a_ops = &zpl_address_space_operations;
460 break;
461 }
462 }
463
464 static void
465 zfs_set_inode_flags(znode_t *zp, struct inode *ip)
466 {
467 /*
468 * Linux and Solaris have different sets of file attributes, so we
469 * restrict this conversion to the intersection of the two.
470 */
471 #ifdef HAVE_INODE_SET_FLAGS
472 unsigned int flags = 0;
473 if (zp->z_pflags & ZFS_IMMUTABLE)
474 flags |= S_IMMUTABLE;
475 if (zp->z_pflags & ZFS_APPENDONLY)
476 flags |= S_APPEND;
477
478 inode_set_flags(ip, flags, S_IMMUTABLE|S_APPEND);
479 #else
480 if (zp->z_pflags & ZFS_IMMUTABLE)
481 ip->i_flags |= S_IMMUTABLE;
482 else
483 ip->i_flags &= ~S_IMMUTABLE;
484
485 if (zp->z_pflags & ZFS_APPENDONLY)
486 ip->i_flags |= S_APPEND;
487 else
488 ip->i_flags &= ~S_APPEND;
489 #endif
490 }
491
492 /*
493 * Update the embedded inode given the znode.
494 */
495 void
496 zfs_znode_update_vfs(znode_t *zp)
497 {
498 zfsvfs_t *zfsvfs;
499 struct inode *ip;
500 uint32_t blksize;
501 u_longlong_t i_blocks;
502
503 ASSERT(zp != NULL);
504 zfsvfs = ZTOZSB(zp);
505 ip = ZTOI(zp);
506
507 /* Skip .zfs control nodes which do not exist on disk. */
508 if (zfsctl_is_node(ip))
509 return;
510
511 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &blksize, &i_blocks);
512
513 spin_lock(&ip->i_lock);
514 ip->i_mode = zp->z_mode;
515 ip->i_blocks = i_blocks;
516 i_size_write(ip, zp->z_size);
517 spin_unlock(&ip->i_lock);
518 }
519
520
521 /*
522 * Construct a znode+inode and initialize.
523 *
524 * This does not do a call to dmu_set_user() that is
525 * up to the caller to do, in case you don't want to
526 * return the znode
527 */
528 static znode_t *
529 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
530 dmu_object_type_t obj_type, sa_handle_t *hdl)
531 {
532 znode_t *zp;
533 struct inode *ip;
534 uint64_t mode;
535 uint64_t parent;
536 uint64_t tmp_gen;
537 uint64_t links;
538 uint64_t z_uid, z_gid;
539 uint64_t atime[2], mtime[2], ctime[2], btime[2];
540 uint64_t projid = ZFS_DEFAULT_PROJID;
541 sa_bulk_attr_t bulk[12];
542 int count = 0;
543
544 ASSERT(zfsvfs != NULL);
545
546 ip = new_inode(zfsvfs->z_sb);
547 if (ip == NULL)
548 return (NULL);
549
550 zp = ITOZ(ip);
551 ASSERT(zp->z_dirlocks == NULL);
552 ASSERT3P(zp->z_acl_cached, ==, NULL);
553 ASSERT3P(zp->z_xattr_cached, ==, NULL);
554 zp->z_unlinked = B_FALSE;
555 zp->z_atime_dirty = B_FALSE;
556 zp->z_is_mapped = B_FALSE;
557 zp->z_is_ctldir = B_FALSE;
558 zp->z_suspended = B_FALSE;
559 zp->z_sa_hdl = NULL;
560 zp->z_mapcnt = 0;
561 zp->z_id = db->db_object;
562 zp->z_blksz = blksz;
563 zp->z_seq = 0x7A4653;
564 zp->z_sync_cnt = 0;
565 zp->z_sync_writes_cnt = 0;
566 zp->z_async_writes_cnt = 0;
567
568 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
569
570 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
571 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &tmp_gen, 8);
572 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
573 &zp->z_size, 8);
574 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
575 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
576 &zp->z_pflags, 8);
577 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
578 &parent, 8);
579 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &z_uid, 8);
580 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &z_gid, 8);
581 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
582 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
583 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
584 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &btime, 16);
585
586 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || tmp_gen == 0 ||
587 (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
588 (zp->z_pflags & ZFS_PROJID) &&
589 sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) {
590 if (hdl == NULL)
591 sa_handle_destroy(zp->z_sa_hdl);
592 zp->z_sa_hdl = NULL;
593 goto error;
594 }
595
596 zp->z_projid = projid;
597 zp->z_mode = ip->i_mode = mode;
598 ip->i_generation = (uint32_t)tmp_gen;
599 ip->i_blkbits = SPA_MINBLOCKSHIFT;
600 set_nlink(ip, (uint32_t)links);
601 zfs_uid_write(ip, z_uid);
602 zfs_gid_write(ip, z_gid);
603 zfs_set_inode_flags(zp, ip);
604
605 /* Cache the xattr parent id */
606 if (zp->z_pflags & ZFS_XATTR)
607 zp->z_xattr_parent = parent;
608
609 ZFS_TIME_DECODE(&ip->i_atime, atime);
610 ZFS_TIME_DECODE(&ip->i_mtime, mtime);
611 ZFS_TIME_DECODE(&ip->i_ctime, ctime);
612 ZFS_TIME_DECODE(&zp->z_btime, btime);
613
614 ip->i_ino = zp->z_id;
615 zfs_znode_update_vfs(zp);
616 zfs_inode_set_ops(zfsvfs, ip);
617
618 /*
619 * The only way insert_inode_locked() can fail is if the ip->i_ino
620 * number is already hashed for this super block. This can never
621 * happen because the inode numbers map 1:1 with the object numbers.
622 *
623 * Exceptions include rolling back a mounted file system, either
624 * from the zfs rollback or zfs recv command.
625 *
626 * Active inodes are unhashed during the rollback, but since zrele
627 * can happen asynchronously, we can't guarantee they've been
628 * unhashed. This can cause hash collisions in unlinked drain
629 * processing so do not hash unlinked znodes.
630 */
631 if (links > 0)
632 VERIFY3S(insert_inode_locked(ip), ==, 0);
633
634 mutex_enter(&zfsvfs->z_znodes_lock);
635 list_insert_tail(&zfsvfs->z_all_znodes, zp);
636 zfsvfs->z_nr_znodes++;
637 mutex_exit(&zfsvfs->z_znodes_lock);
638
639 if (links > 0)
640 unlock_new_inode(ip);
641 return (zp);
642
643 error:
644 iput(ip);
645 return (NULL);
646 }
647
648 /*
649 * Safely mark an inode dirty. Inodes which are part of a read-only
650 * file system or snapshot may not be dirtied.
651 */
652 void
653 zfs_mark_inode_dirty(struct inode *ip)
654 {
655 zfsvfs_t *zfsvfs = ITOZSB(ip);
656
657 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
658 return;
659
660 mark_inode_dirty(ip);
661 }
662
663 static uint64_t empty_xattr;
664 static uint64_t pad[4];
665 static zfs_acl_phys_t acl_phys;
666 /*
667 * Create a new DMU object to hold a zfs znode.
668 *
669 * IN: dzp - parent directory for new znode
670 * vap - file attributes for new znode
671 * tx - dmu transaction id for zap operations
672 * cr - credentials of caller
673 * flag - flags:
674 * IS_ROOT_NODE - new object will be root
675 * IS_TMPFILE - new object is of O_TMPFILE
676 * IS_XATTR - new object is an attribute
677 * acl_ids - ACL related attributes
678 *
679 * OUT: zpp - allocated znode (set to dzp if IS_ROOT_NODE)
680 *
681 */
682 void
683 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
684 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
685 {
686 uint64_t crtime[2], atime[2], mtime[2], ctime[2];
687 uint64_t mode, size, links, parent, pflags;
688 uint64_t projid = ZFS_DEFAULT_PROJID;
689 uint64_t rdev = 0;
690 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
691 dmu_buf_t *db;
692 inode_timespec_t now;
693 uint64_t gen, obj;
694 int bonuslen;
695 int dnodesize;
696 sa_handle_t *sa_hdl;
697 dmu_object_type_t obj_type;
698 sa_bulk_attr_t *sa_attrs;
699 int cnt = 0;
700 zfs_acl_locator_cb_t locate = { 0 };
701 znode_hold_t *zh;
702
703 if (zfsvfs->z_replay) {
704 obj = vap->va_nodeid;
705 now = vap->va_ctime; /* see zfs_replay_create() */
706 gen = vap->va_nblocks; /* ditto */
707 dnodesize = vap->va_fsid; /* ditto */
708 } else {
709 obj = 0;
710 gethrestime(&now);
711 gen = dmu_tx_get_txg(tx);
712 dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
713 }
714
715 if (dnodesize == 0)
716 dnodesize = DNODE_MIN_SIZE;
717
718 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
719
720 bonuslen = (obj_type == DMU_OT_SA) ?
721 DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
722
723 /*
724 * Create a new DMU object.
725 */
726 /*
727 * There's currently no mechanism for pre-reading the blocks that will
728 * be needed to allocate a new object, so we accept the small chance
729 * that there will be an i/o error and we will fail one of the
730 * assertions below.
731 */
732 if (S_ISDIR(vap->va_mode)) {
733 if (zfsvfs->z_replay) {
734 VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj,
735 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
736 obj_type, bonuslen, dnodesize, tx));
737 } else {
738 obj = zap_create_norm_dnsize(zfsvfs->z_os,
739 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
740 obj_type, bonuslen, dnodesize, tx);
741 }
742 } else {
743 if (zfsvfs->z_replay) {
744 VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj,
745 DMU_OT_PLAIN_FILE_CONTENTS, 0,
746 obj_type, bonuslen, dnodesize, tx));
747 } else {
748 obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
749 DMU_OT_PLAIN_FILE_CONTENTS, 0,
750 obj_type, bonuslen, dnodesize, tx);
751 }
752 }
753
754 zh = zfs_znode_hold_enter(zfsvfs, obj);
755 VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
756
757 /*
758 * If this is the root, fix up the half-initialized parent pointer
759 * to reference the just-allocated physical data area.
760 */
761 if (flag & IS_ROOT_NODE) {
762 dzp->z_id = obj;
763 }
764
765 /*
766 * If parent is an xattr, so am I.
767 */
768 if (dzp->z_pflags & ZFS_XATTR) {
769 flag |= IS_XATTR;
770 }
771
772 if (zfsvfs->z_use_fuids)
773 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
774 else
775 pflags = 0;
776
777 if (S_ISDIR(vap->va_mode)) {
778 size = 2; /* contents ("." and "..") */
779 links = 2;
780 } else {
781 size = 0;
782 links = (flag & IS_TMPFILE) ? 0 : 1;
783 }
784
785 if (S_ISBLK(vap->va_mode) || S_ISCHR(vap->va_mode))
786 rdev = vap->va_rdev;
787
788 parent = dzp->z_id;
789 mode = acl_ids->z_mode;
790 if (flag & IS_XATTR)
791 pflags |= ZFS_XATTR;
792
793 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode)) {
794 /*
795 * With ZFS_PROJID flag, we can easily know whether there is
796 * project ID stored on disk or not. See zfs_space_delta_cb().
797 */
798 if (obj_type != DMU_OT_ZNODE &&
799 dmu_objset_projectquota_enabled(zfsvfs->z_os))
800 pflags |= ZFS_PROJID;
801
802 /*
803 * Inherit project ID from parent if required.
804 */
805 projid = zfs_inherit_projid(dzp);
806 if (dzp->z_pflags & ZFS_PROJINHERIT)
807 pflags |= ZFS_PROJINHERIT;
808 }
809
810 /*
811 * No execs denied will be determined when zfs_mode_compute() is called.
812 */
813 pflags |= acl_ids->z_aclp->z_hints &
814 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
815 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
816
817 ZFS_TIME_ENCODE(&now, crtime);
818 ZFS_TIME_ENCODE(&now, ctime);
819
820 if (vap->va_mask & ATTR_ATIME) {
821 ZFS_TIME_ENCODE(&vap->va_atime, atime);
822 } else {
823 ZFS_TIME_ENCODE(&now, atime);
824 }
825
826 if (vap->va_mask & ATTR_MTIME) {
827 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
828 } else {
829 ZFS_TIME_ENCODE(&now, mtime);
830 }
831
832 /* Now add in all of the "SA" attributes */
833 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
834 &sa_hdl));
835
836 /*
837 * Setup the array of attributes to be replaced/set on the new file
838 *
839 * order for DMU_OT_ZNODE is critical since it needs to be constructed
840 * in the old znode_phys_t format. Don't change this ordering
841 */
842 sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
843
844 if (obj_type == DMU_OT_ZNODE) {
845 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
846 NULL, &atime, 16);
847 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
848 NULL, &mtime, 16);
849 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
850 NULL, &ctime, 16);
851 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
852 NULL, &crtime, 16);
853 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
854 NULL, &gen, 8);
855 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
856 NULL, &mode, 8);
857 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
858 NULL, &size, 8);
859 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
860 NULL, &parent, 8);
861 } else {
862 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
863 NULL, &mode, 8);
864 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
865 NULL, &size, 8);
866 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
867 NULL, &gen, 8);
868 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs),
869 NULL, &acl_ids->z_fuid, 8);
870 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs),
871 NULL, &acl_ids->z_fgid, 8);
872 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
873 NULL, &parent, 8);
874 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
875 NULL, &pflags, 8);
876 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
877 NULL, &atime, 16);
878 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
879 NULL, &mtime, 16);
880 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
881 NULL, &ctime, 16);
882 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
883 NULL, &crtime, 16);
884 }
885
886 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
887
888 if (obj_type == DMU_OT_ZNODE) {
889 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
890 &empty_xattr, 8);
891 } else if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
892 pflags & ZFS_PROJID) {
893 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PROJID(zfsvfs),
894 NULL, &projid, 8);
895 }
896 if (obj_type == DMU_OT_ZNODE ||
897 (S_ISBLK(vap->va_mode) || S_ISCHR(vap->va_mode))) {
898 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
899 NULL, &rdev, 8);
900 }
901 if (obj_type == DMU_OT_ZNODE) {
902 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
903 NULL, &pflags, 8);
904 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
905 &acl_ids->z_fuid, 8);
906 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
907 &acl_ids->z_fgid, 8);
908 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
909 sizeof (uint64_t) * 4);
910 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
911 &acl_phys, sizeof (zfs_acl_phys_t));
912 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
913 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
914 &acl_ids->z_aclp->z_acl_count, 8);
915 locate.cb_aclp = acl_ids->z_aclp;
916 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
917 zfs_acl_data_locator, &locate,
918 acl_ids->z_aclp->z_acl_bytes);
919 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
920 acl_ids->z_fuid, acl_ids->z_fgid);
921 }
922
923 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
924
925 if (!(flag & IS_ROOT_NODE)) {
926 /*
927 * The call to zfs_znode_alloc() may fail if memory is low
928 * via the call path: alloc_inode() -> inode_init_always() ->
929 * security_inode_alloc() -> inode_alloc_security(). Since
930 * the existing code is written such that zfs_mknode() can
931 * not fail retry until sufficient memory has been reclaimed.
932 */
933 do {
934 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
935 } while (*zpp == NULL);
936
937 VERIFY(*zpp != NULL);
938 VERIFY(dzp != NULL);
939 } else {
940 /*
941 * If we are creating the root node, the "parent" we
942 * passed in is the znode for the root.
943 */
944 *zpp = dzp;
945
946 (*zpp)->z_sa_hdl = sa_hdl;
947 }
948
949 (*zpp)->z_pflags = pflags;
950 (*zpp)->z_mode = ZTOI(*zpp)->i_mode = mode;
951 (*zpp)->z_dnodesize = dnodesize;
952 (*zpp)->z_projid = projid;
953
954 if (obj_type == DMU_OT_ZNODE ||
955 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
956 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
957 }
958 kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
959 zfs_znode_hold_exit(zfsvfs, zh);
960 }
961
962 /*
963 * Update in-core attributes. It is assumed the caller will be doing an
964 * sa_bulk_update to push the changes out.
965 */
966 void
967 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
968 {
969 xoptattr_t *xoap;
970 boolean_t update_inode = B_FALSE;
971
972 xoap = xva_getxoptattr(xvap);
973 ASSERT(xoap);
974
975 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
976 uint64_t times[2];
977 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
978 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(ZTOZSB(zp)),
979 &times, sizeof (times), tx);
980 XVA_SET_RTN(xvap, XAT_CREATETIME);
981 }
982 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
983 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
984 zp->z_pflags, tx);
985 XVA_SET_RTN(xvap, XAT_READONLY);
986 }
987 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
988 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
989 zp->z_pflags, tx);
990 XVA_SET_RTN(xvap, XAT_HIDDEN);
991 }
992 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
993 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
994 zp->z_pflags, tx);
995 XVA_SET_RTN(xvap, XAT_SYSTEM);
996 }
997 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
998 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
999 zp->z_pflags, tx);
1000 XVA_SET_RTN(xvap, XAT_ARCHIVE);
1001 }
1002 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1003 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1004 zp->z_pflags, tx);
1005 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1006
1007 update_inode = B_TRUE;
1008 }
1009 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1010 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1011 zp->z_pflags, tx);
1012 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1013 }
1014 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1015 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1016 zp->z_pflags, tx);
1017 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1018
1019 update_inode = B_TRUE;
1020 }
1021 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1022 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1023 zp->z_pflags, tx);
1024 XVA_SET_RTN(xvap, XAT_NODUMP);
1025 }
1026 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1027 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1028 zp->z_pflags, tx);
1029 XVA_SET_RTN(xvap, XAT_OPAQUE);
1030 }
1031 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1032 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1033 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1034 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1035 }
1036 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1037 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1038 zp->z_pflags, tx);
1039 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1040 }
1041 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1042 zfs_sa_set_scanstamp(zp, xvap, tx);
1043 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1044 }
1045 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1046 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1047 zp->z_pflags, tx);
1048 XVA_SET_RTN(xvap, XAT_REPARSE);
1049 }
1050 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1051 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1052 zp->z_pflags, tx);
1053 XVA_SET_RTN(xvap, XAT_OFFLINE);
1054 }
1055 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1056 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1057 zp->z_pflags, tx);
1058 XVA_SET_RTN(xvap, XAT_SPARSE);
1059 }
1060 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
1061 ZFS_ATTR_SET(zp, ZFS_PROJINHERIT, xoap->xoa_projinherit,
1062 zp->z_pflags, tx);
1063 XVA_SET_RTN(xvap, XAT_PROJINHERIT);
1064 }
1065
1066 if (update_inode)
1067 zfs_set_inode_flags(zp, ZTOI(zp));
1068 }
1069
1070 int
1071 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1072 {
1073 dmu_object_info_t doi;
1074 dmu_buf_t *db;
1075 znode_t *zp;
1076 znode_hold_t *zh;
1077 int err;
1078 sa_handle_t *hdl;
1079
1080 *zpp = NULL;
1081
1082 again:
1083 zh = zfs_znode_hold_enter(zfsvfs, obj_num);
1084
1085 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1086 if (err) {
1087 zfs_znode_hold_exit(zfsvfs, zh);
1088 return (err);
1089 }
1090
1091 dmu_object_info_from_db(db, &doi);
1092 if (doi.doi_bonus_type != DMU_OT_SA &&
1093 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1094 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1095 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1096 sa_buf_rele(db, NULL);
1097 zfs_znode_hold_exit(zfsvfs, zh);
1098 return (SET_ERROR(EINVAL));
1099 }
1100
1101 hdl = dmu_buf_get_user(db);
1102 if (hdl != NULL) {
1103 zp = sa_get_userdata(hdl);
1104
1105
1106 /*
1107 * Since "SA" does immediate eviction we
1108 * should never find a sa handle that doesn't
1109 * know about the znode.
1110 */
1111
1112 ASSERT3P(zp, !=, NULL);
1113
1114 mutex_enter(&zp->z_lock);
1115 ASSERT3U(zp->z_id, ==, obj_num);
1116 /*
1117 * If zp->z_unlinked is set, the znode is already marked
1118 * for deletion and should not be discovered. Check this
1119 * after checking igrab() due to fsetxattr() & O_TMPFILE.
1120 *
1121 * If igrab() returns NULL the VFS has independently
1122 * determined the inode should be evicted and has
1123 * called iput_final() to start the eviction process.
1124 * The SA handle is still valid but because the VFS
1125 * requires that the eviction succeed we must drop
1126 * our locks and references to allow the eviction to
1127 * complete. The zfs_zget() may then be retried.
1128 *
1129 * This unlikely case could be optimized by registering
1130 * a sops->drop_inode() callback. The callback would
1131 * need to detect the active SA hold thereby informing
1132 * the VFS that this inode should not be evicted.
1133 */
1134 if (igrab(ZTOI(zp)) == NULL) {
1135 if (zp->z_unlinked)
1136 err = SET_ERROR(ENOENT);
1137 else
1138 err = SET_ERROR(EAGAIN);
1139 } else {
1140 *zpp = zp;
1141 err = 0;
1142 }
1143
1144 mutex_exit(&zp->z_lock);
1145 sa_buf_rele(db, NULL);
1146 zfs_znode_hold_exit(zfsvfs, zh);
1147
1148 if (err == EAGAIN) {
1149 /* inode might need this to finish evict */
1150 cond_resched();
1151 goto again;
1152 }
1153 return (err);
1154 }
1155
1156 /*
1157 * Not found create new znode/vnode but only if file exists.
1158 *
1159 * There is a small window where zfs_vget() could
1160 * find this object while a file create is still in
1161 * progress. This is checked for in zfs_znode_alloc()
1162 *
1163 * if zfs_znode_alloc() fails it will drop the hold on the
1164 * bonus buffer.
1165 */
1166 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1167 doi.doi_bonus_type, NULL);
1168 if (zp == NULL) {
1169 err = SET_ERROR(ENOENT);
1170 } else {
1171 *zpp = zp;
1172 }
1173 zfs_znode_hold_exit(zfsvfs, zh);
1174 return (err);
1175 }
1176
1177 int
1178 zfs_rezget(znode_t *zp)
1179 {
1180 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1181 dmu_object_info_t doi;
1182 dmu_buf_t *db;
1183 uint64_t obj_num = zp->z_id;
1184 uint64_t mode;
1185 uint64_t links;
1186 sa_bulk_attr_t bulk[11];
1187 int err;
1188 int count = 0;
1189 uint64_t gen;
1190 uint64_t z_uid, z_gid;
1191 uint64_t atime[2], mtime[2], ctime[2], btime[2];
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(&ZTOI(zp)->i_atime, atime);
1285 ZFS_TIME_DECODE(&ZTOI(zp)->i_mtime, mtime);
1286 ZFS_TIME_DECODE(&ZTOI(zp)->i_ctime, ctime);
1287 ZFS_TIME_DECODE(&zp->z_btime, btime);
1288
1289 if ((uint32_t)gen != ZTOI(zp)->i_generation) {
1290 zfs_znode_dmu_fini(zp);
1291 zfs_znode_hold_exit(zfsvfs, zh);
1292 return (SET_ERROR(EIO));
1293 }
1294
1295 set_nlink(ZTOI(zp), (uint32_t)links);
1296 zfs_set_inode_flags(zp, ZTOI(zp));
1297
1298 zp->z_blksz = doi.doi_data_block_size;
1299 zp->z_atime_dirty = B_FALSE;
1300 zfs_znode_update_vfs(zp);
1301
1302 /*
1303 * If the file has zero links, then it has been unlinked on the send
1304 * side and it must be in the received unlinked set.
1305 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1306 * stale data and to prevent automatic removal of the file in
1307 * zfs_zinactive(). The file will be removed either when it is removed
1308 * on the send side and the next incremental stream is received or
1309 * when the unlinked set gets processed.
1310 */
1311 zp->z_unlinked = (ZTOI(zp)->i_nlink == 0);
1312 if (zp->z_unlinked)
1313 zfs_znode_dmu_fini(zp);
1314
1315 zfs_znode_hold_exit(zfsvfs, zh);
1316
1317 return (0);
1318 }
1319
1320 void
1321 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1322 {
1323 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1324 objset_t *os = zfsvfs->z_os;
1325 uint64_t obj = zp->z_id;
1326 uint64_t acl_obj = zfs_external_acl(zp);
1327 znode_hold_t *zh;
1328
1329 zh = zfs_znode_hold_enter(zfsvfs, obj);
1330 if (acl_obj) {
1331 VERIFY(!zp->z_is_sa);
1332 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1333 }
1334 VERIFY(0 == dmu_object_free(os, obj, tx));
1335 zfs_znode_dmu_fini(zp);
1336 zfs_znode_hold_exit(zfsvfs, zh);
1337 }
1338
1339 void
1340 zfs_zinactive(znode_t *zp)
1341 {
1342 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1343 uint64_t z_id = zp->z_id;
1344 znode_hold_t *zh;
1345
1346 ASSERT(zp->z_sa_hdl);
1347
1348 /*
1349 * Don't allow a zfs_zget() while were trying to release this znode.
1350 */
1351 zh = zfs_znode_hold_enter(zfsvfs, z_id);
1352
1353 mutex_enter(&zp->z_lock);
1354
1355 /*
1356 * If this was the last reference to a file with no links, remove
1357 * the file from the file system unless the file system is mounted
1358 * read-only. That can happen, for example, if the file system was
1359 * originally read-write, the file was opened, then unlinked and
1360 * the file system was made read-only before the file was finally
1361 * closed. The file will remain in the unlinked set.
1362 */
1363 if (zp->z_unlinked) {
1364 ASSERT(!zfsvfs->z_issnap);
1365 if (!zfs_is_readonly(zfsvfs) && !zfs_unlink_suspend_progress) {
1366 mutex_exit(&zp->z_lock);
1367 zfs_znode_hold_exit(zfsvfs, zh);
1368 zfs_rmnode(zp);
1369 return;
1370 }
1371 }
1372
1373 mutex_exit(&zp->z_lock);
1374 zfs_znode_dmu_fini(zp);
1375
1376 zfs_znode_hold_exit(zfsvfs, zh);
1377 }
1378
1379 #if defined(HAVE_INODE_TIMESPEC64_TIMES)
1380 #define zfs_compare_timespec timespec64_compare
1381 #else
1382 #define zfs_compare_timespec timespec_compare
1383 #endif
1384
1385 /*
1386 * Determine whether the znode's atime must be updated. The logic mostly
1387 * duplicates the Linux kernel's relatime_need_update() functionality.
1388 * This function is only called if the underlying filesystem actually has
1389 * atime updates enabled.
1390 */
1391 boolean_t
1392 zfs_relatime_need_update(const struct inode *ip)
1393 {
1394 inode_timespec_t now;
1395
1396 gethrestime(&now);
1397 /*
1398 * In relatime mode, only update the atime if the previous atime
1399 * is earlier than either the ctime or mtime or if at least a day
1400 * has passed since the last update of atime.
1401 */
1402 if (zfs_compare_timespec(&ip->i_mtime, &ip->i_atime) >= 0)
1403 return (B_TRUE);
1404
1405 if (zfs_compare_timespec(&ip->i_ctime, &ip->i_atime) >= 0)
1406 return (B_TRUE);
1407
1408 if ((hrtime_t)now.tv_sec - (hrtime_t)ip->i_atime.tv_sec >= 24*60*60)
1409 return (B_TRUE);
1410
1411 return (B_FALSE);
1412 }
1413
1414 /*
1415 * Prepare to update znode time stamps.
1416 *
1417 * IN: zp - znode requiring timestamp update
1418 * flag - ATTR_MTIME, ATTR_CTIME flags
1419 *
1420 * OUT: zp - z_seq
1421 * mtime - new mtime
1422 * ctime - new ctime
1423 *
1424 * Note: We don't update atime here, because we rely on Linux VFS to do
1425 * atime updating.
1426 */
1427 void
1428 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1429 uint64_t ctime[2])
1430 {
1431 inode_timespec_t now;
1432
1433 gethrestime(&now);
1434
1435 zp->z_seq++;
1436
1437 if (flag & ATTR_MTIME) {
1438 ZFS_TIME_ENCODE(&now, mtime);
1439 ZFS_TIME_DECODE(&(ZTOI(zp)->i_mtime), mtime);
1440 if (ZTOZSB(zp)->z_use_fuids) {
1441 zp->z_pflags |= (ZFS_ARCHIVE |
1442 ZFS_AV_MODIFIED);
1443 }
1444 }
1445
1446 if (flag & ATTR_CTIME) {
1447 ZFS_TIME_ENCODE(&now, ctime);
1448 ZFS_TIME_DECODE(&(ZTOI(zp)->i_ctime), ctime);
1449 if (ZTOZSB(zp)->z_use_fuids)
1450 zp->z_pflags |= ZFS_ARCHIVE;
1451 }
1452 }
1453
1454 /*
1455 * Grow the block size for a file.
1456 *
1457 * IN: zp - znode of file to free data in.
1458 * size - requested block size
1459 * tx - open transaction.
1460 *
1461 * NOTE: this function assumes that the znode is write locked.
1462 */
1463 void
1464 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1465 {
1466 int error;
1467 u_longlong_t dummy;
1468
1469 if (size <= zp->z_blksz)
1470 return;
1471 /*
1472 * If the file size is already greater than the current blocksize,
1473 * we will not grow. If there is more than one block in a file,
1474 * the blocksize cannot change.
1475 */
1476 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1477 return;
1478
1479 error = dmu_object_set_blocksize(ZTOZSB(zp)->z_os, zp->z_id,
1480 size, 0, tx);
1481
1482 if (error == ENOTSUP)
1483 return;
1484 ASSERT0(error);
1485
1486 /* What blocksize did we actually get? */
1487 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1488 }
1489
1490 /*
1491 * Increase the file length
1492 *
1493 * IN: zp - znode of file to free data in.
1494 * end - new end-of-file
1495 *
1496 * RETURN: 0 on success, error code on failure
1497 */
1498 static int
1499 zfs_extend(znode_t *zp, uint64_t end)
1500 {
1501 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1502 dmu_tx_t *tx;
1503 zfs_locked_range_t *lr;
1504 uint64_t newblksz;
1505 int error;
1506
1507 /*
1508 * We will change zp_size, lock the whole file.
1509 */
1510 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1511
1512 /*
1513 * Nothing to do if file already at desired length.
1514 */
1515 if (end <= zp->z_size) {
1516 zfs_rangelock_exit(lr);
1517 return (0);
1518 }
1519 tx = dmu_tx_create(zfsvfs->z_os);
1520 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1521 zfs_sa_upgrade_txholds(tx, zp);
1522 if (end > zp->z_blksz &&
1523 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1524 /*
1525 * We are growing the file past the current block size.
1526 */
1527 if (zp->z_blksz > ZTOZSB(zp)->z_max_blksz) {
1528 /*
1529 * File's blocksize is already larger than the
1530 * "recordsize" property. Only let it grow to
1531 * the next power of 2.
1532 */
1533 ASSERT(!ISP2(zp->z_blksz));
1534 newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1535 } else {
1536 newblksz = MIN(end, ZTOZSB(zp)->z_max_blksz);
1537 }
1538 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1539 } else {
1540 newblksz = 0;
1541 }
1542
1543 error = dmu_tx_assign(tx, TXG_WAIT);
1544 if (error) {
1545 dmu_tx_abort(tx);
1546 zfs_rangelock_exit(lr);
1547 return (error);
1548 }
1549
1550 if (newblksz)
1551 zfs_grow_blocksize(zp, newblksz, tx);
1552
1553 zp->z_size = end;
1554
1555 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(ZTOZSB(zp)),
1556 &zp->z_size, sizeof (zp->z_size), tx));
1557
1558 zfs_rangelock_exit(lr);
1559
1560 dmu_tx_commit(tx);
1561
1562 return (0);
1563 }
1564
1565 /*
1566 * zfs_zero_partial_page - Modeled after update_pages() but
1567 * with different arguments and semantics for use by zfs_freesp().
1568 *
1569 * Zeroes a piece of a single page cache entry for zp at offset
1570 * start and length len.
1571 *
1572 * Caller must acquire a range lock on the file for the region
1573 * being zeroed in order that the ARC and page cache stay in sync.
1574 */
1575 static void
1576 zfs_zero_partial_page(znode_t *zp, uint64_t start, uint64_t len)
1577 {
1578 struct address_space *mp = ZTOI(zp)->i_mapping;
1579 struct page *pp;
1580 int64_t off;
1581 void *pb;
1582
1583 ASSERT((start & PAGE_MASK) == ((start + len - 1) & PAGE_MASK));
1584
1585 off = start & (PAGE_SIZE - 1);
1586 start &= PAGE_MASK;
1587
1588 pp = find_lock_page(mp, start >> PAGE_SHIFT);
1589 if (pp) {
1590 if (mapping_writably_mapped(mp))
1591 flush_dcache_page(pp);
1592
1593 pb = kmap(pp);
1594 memset(pb + off, 0, len);
1595 kunmap(pp);
1596
1597 if (mapping_writably_mapped(mp))
1598 flush_dcache_page(pp);
1599
1600 mark_page_accessed(pp);
1601 SetPageUptodate(pp);
1602 ClearPageError(pp);
1603 unlock_page(pp);
1604 put_page(pp);
1605 }
1606 }
1607
1608 /*
1609 * Free space in a file.
1610 *
1611 * IN: zp - znode of file to free data in.
1612 * off - start of section to free.
1613 * len - length of section to free.
1614 *
1615 * RETURN: 0 on success, error code on failure
1616 */
1617 static int
1618 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1619 {
1620 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1621 zfs_locked_range_t *lr;
1622 int error;
1623
1624 /*
1625 * Lock the range being freed.
1626 */
1627 lr = zfs_rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER);
1628
1629 /*
1630 * Nothing to do if file already at desired length.
1631 */
1632 if (off >= zp->z_size) {
1633 zfs_rangelock_exit(lr);
1634 return (0);
1635 }
1636
1637 if (off + len > zp->z_size)
1638 len = zp->z_size - off;
1639
1640 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1641
1642 /*
1643 * Zero partial page cache entries. This must be done under a
1644 * range lock in order to keep the ARC and page cache in sync.
1645 */
1646 if (zp->z_is_mapped) {
1647 loff_t first_page, last_page, page_len;
1648 loff_t first_page_offset, last_page_offset;
1649
1650 /* first possible full page in hole */
1651 first_page = (off + PAGE_SIZE - 1) >> PAGE_SHIFT;
1652 /* last page of hole */
1653 last_page = (off + len) >> PAGE_SHIFT;
1654
1655 /* offset of first_page */
1656 first_page_offset = first_page << PAGE_SHIFT;
1657 /* offset of last_page */
1658 last_page_offset = last_page << PAGE_SHIFT;
1659
1660 /* truncate whole pages */
1661 if (last_page_offset > first_page_offset) {
1662 truncate_inode_pages_range(ZTOI(zp)->i_mapping,
1663 first_page_offset, last_page_offset - 1);
1664 }
1665
1666 /* truncate sub-page ranges */
1667 if (first_page > last_page) {
1668 /* entire punched area within a single page */
1669 zfs_zero_partial_page(zp, off, len);
1670 } else {
1671 /* beginning of punched area at the end of a page */
1672 page_len = first_page_offset - off;
1673 if (page_len > 0)
1674 zfs_zero_partial_page(zp, off, page_len);
1675
1676 /* end of punched area at the beginning of a page */
1677 page_len = off + len - last_page_offset;
1678 if (page_len > 0)
1679 zfs_zero_partial_page(zp, last_page_offset,
1680 page_len);
1681 }
1682 }
1683 zfs_rangelock_exit(lr);
1684
1685 return (error);
1686 }
1687
1688 /*
1689 * Truncate a file
1690 *
1691 * IN: zp - znode of file to free data in.
1692 * end - new end-of-file.
1693 *
1694 * RETURN: 0 on success, error code on failure
1695 */
1696 static int
1697 zfs_trunc(znode_t *zp, uint64_t end)
1698 {
1699 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1700 dmu_tx_t *tx;
1701 zfs_locked_range_t *lr;
1702 int error;
1703 sa_bulk_attr_t bulk[2];
1704 int count = 0;
1705
1706 /*
1707 * We will change zp_size, lock the whole file.
1708 */
1709 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1710
1711 /*
1712 * Nothing to do if file already at desired length.
1713 */
1714 if (end >= zp->z_size) {
1715 zfs_rangelock_exit(lr);
1716 return (0);
1717 }
1718
1719 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1720 DMU_OBJECT_END);
1721 if (error) {
1722 zfs_rangelock_exit(lr);
1723 return (error);
1724 }
1725 tx = dmu_tx_create(zfsvfs->z_os);
1726 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1727 zfs_sa_upgrade_txholds(tx, zp);
1728 dmu_tx_mark_netfree(tx);
1729 error = dmu_tx_assign(tx, TXG_WAIT);
1730 if (error) {
1731 dmu_tx_abort(tx);
1732 zfs_rangelock_exit(lr);
1733 return (error);
1734 }
1735
1736 zp->z_size = end;
1737 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1738 NULL, &zp->z_size, sizeof (zp->z_size));
1739
1740 if (end == 0) {
1741 zp->z_pflags &= ~ZFS_SPARSE;
1742 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1743 NULL, &zp->z_pflags, 8);
1744 }
1745 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1746
1747 dmu_tx_commit(tx);
1748 zfs_rangelock_exit(lr);
1749
1750 return (0);
1751 }
1752
1753 /*
1754 * Free space in a file
1755 *
1756 * IN: zp - znode of file to free data in.
1757 * off - start of range
1758 * len - end of range (0 => EOF)
1759 * flag - current file open mode flags.
1760 * log - TRUE if this action should be logged
1761 *
1762 * RETURN: 0 on success, error code on failure
1763 */
1764 int
1765 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1766 {
1767 dmu_tx_t *tx;
1768 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1769 zilog_t *zilog = zfsvfs->z_log;
1770 uint64_t mode;
1771 uint64_t mtime[2], ctime[2];
1772 sa_bulk_attr_t bulk[3];
1773 int count = 0;
1774 int error;
1775
1776 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1777 sizeof (mode))) != 0)
1778 return (error);
1779
1780 if (off > zp->z_size) {
1781 error = zfs_extend(zp, off+len);
1782 if (error == 0 && log)
1783 goto log;
1784 goto out;
1785 }
1786
1787 if (len == 0) {
1788 error = zfs_trunc(zp, off);
1789 } else {
1790 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1791 off + len > zp->z_size)
1792 error = zfs_extend(zp, off+len);
1793 }
1794 if (error || !log)
1795 goto out;
1796 log:
1797 tx = dmu_tx_create(zfsvfs->z_os);
1798 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1799 zfs_sa_upgrade_txholds(tx, zp);
1800 error = dmu_tx_assign(tx, TXG_WAIT);
1801 if (error) {
1802 dmu_tx_abort(tx);
1803 goto out;
1804 }
1805
1806 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1807 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1808 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1809 NULL, &zp->z_pflags, 8);
1810 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
1811 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1812 ASSERT(error == 0);
1813
1814 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1815
1816 dmu_tx_commit(tx);
1817
1818 zfs_znode_update_vfs(zp);
1819 error = 0;
1820
1821 out:
1822 /*
1823 * Truncate the page cache - for file truncate operations, use
1824 * the purpose-built API for truncations. For punching operations,
1825 * the truncation is handled under a range lock in zfs_free_range.
1826 */
1827 if (len == 0)
1828 truncate_setsize(ZTOI(zp), off);
1829 return (error);
1830 }
1831
1832 void
1833 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1834 {
1835 struct super_block *sb;
1836 zfsvfs_t *zfsvfs;
1837 uint64_t moid, obj, sa_obj, version;
1838 uint64_t sense = ZFS_CASE_SENSITIVE;
1839 uint64_t norm = 0;
1840 nvpair_t *elem;
1841 int size;
1842 int error;
1843 int i;
1844 znode_t *rootzp = NULL;
1845 vattr_t vattr;
1846 znode_t *zp;
1847 zfs_acl_ids_t acl_ids;
1848
1849 /*
1850 * First attempt to create master node.
1851 */
1852 /*
1853 * In an empty objset, there are no blocks to read and thus
1854 * there can be no i/o errors (which we assert below).
1855 */
1856 moid = MASTER_NODE_OBJ;
1857 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1858 DMU_OT_NONE, 0, tx);
1859 ASSERT(error == 0);
1860
1861 /*
1862 * Set starting attributes.
1863 */
1864 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1865 elem = NULL;
1866 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1867 /* For the moment we expect all zpl props to be uint64_ts */
1868 uint64_t val;
1869 char *name;
1870
1871 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1872 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1873 name = nvpair_name(elem);
1874 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1875 if (val < version)
1876 version = val;
1877 } else {
1878 error = zap_update(os, moid, name, 8, 1, &val, tx);
1879 }
1880 ASSERT(error == 0);
1881 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1882 norm = val;
1883 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1884 sense = val;
1885 }
1886 ASSERT(version != 0);
1887 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1888
1889 /*
1890 * Create zap object used for SA attribute registration
1891 */
1892
1893 if (version >= ZPL_VERSION_SA) {
1894 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1895 DMU_OT_NONE, 0, tx);
1896 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1897 ASSERT(error == 0);
1898 } else {
1899 sa_obj = 0;
1900 }
1901 /*
1902 * Create a delete queue.
1903 */
1904 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1905
1906 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1907 ASSERT(error == 0);
1908
1909 /*
1910 * Create root znode. Create minimal znode/inode/zfsvfs/sb
1911 * to allow zfs_mknode to work.
1912 */
1913 vattr.va_mask = ATTR_MODE|ATTR_UID|ATTR_GID;
1914 vattr.va_mode = S_IFDIR|0755;
1915 vattr.va_uid = crgetuid(cr);
1916 vattr.va_gid = crgetgid(cr);
1917
1918 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1919 rootzp->z_unlinked = B_FALSE;
1920 rootzp->z_atime_dirty = B_FALSE;
1921 rootzp->z_is_sa = USE_SA(version, os);
1922 rootzp->z_pflags = 0;
1923
1924 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1925 zfsvfs->z_os = os;
1926 zfsvfs->z_parent = zfsvfs;
1927 zfsvfs->z_version = version;
1928 zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1929 zfsvfs->z_use_sa = USE_SA(version, os);
1930 zfsvfs->z_norm = norm;
1931
1932 sb = kmem_zalloc(sizeof (struct super_block), KM_SLEEP);
1933 sb->s_fs_info = zfsvfs;
1934
1935 ZTOI(rootzp)->i_sb = sb;
1936
1937 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1938 &zfsvfs->z_attr_table);
1939
1940 ASSERT(error == 0);
1941
1942 /*
1943 * Fold case on file systems that are always or sometimes case
1944 * insensitive.
1945 */
1946 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1947 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1948
1949 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1950 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1951 offsetof(znode_t, z_link_node));
1952
1953 size = MIN(1 << (highbit64(zfs_object_mutex_size)-1), ZFS_OBJ_MTX_MAX);
1954 zfsvfs->z_hold_size = size;
1955 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
1956 KM_SLEEP);
1957 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
1958 for (i = 0; i != size; i++) {
1959 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
1960 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
1961 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
1962 }
1963
1964 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1965 cr, NULL, &acl_ids, kcred->user_ns));
1966 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1967 ASSERT3P(zp, ==, rootzp);
1968 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1969 ASSERT(error == 0);
1970 zfs_acl_ids_free(&acl_ids);
1971
1972 atomic_set(&ZTOI(rootzp)->i_count, 0);
1973 sa_handle_destroy(rootzp->z_sa_hdl);
1974 kmem_cache_free(znode_cache, rootzp);
1975
1976 for (i = 0; i != size; i++) {
1977 avl_destroy(&zfsvfs->z_hold_trees[i]);
1978 mutex_destroy(&zfsvfs->z_hold_locks[i]);
1979 }
1980
1981 mutex_destroy(&zfsvfs->z_znodes_lock);
1982
1983 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
1984 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
1985 kmem_free(sb, sizeof (struct super_block));
1986 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1987 }
1988 #endif /* _KERNEL */
1989
1990 static int
1991 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1992 {
1993 uint64_t sa_obj = 0;
1994 int error;
1995
1996 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1997 if (error != 0 && error != ENOENT)
1998 return (error);
1999
2000 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
2001 return (error);
2002 }
2003
2004 static int
2005 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
2006 dmu_buf_t **db, const void *tag)
2007 {
2008 dmu_object_info_t doi;
2009 int error;
2010
2011 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
2012 return (error);
2013
2014 dmu_object_info_from_db(*db, &doi);
2015 if ((doi.doi_bonus_type != DMU_OT_SA &&
2016 doi.doi_bonus_type != DMU_OT_ZNODE) ||
2017 (doi.doi_bonus_type == DMU_OT_ZNODE &&
2018 doi.doi_bonus_size < sizeof (znode_phys_t))) {
2019 sa_buf_rele(*db, tag);
2020 return (SET_ERROR(ENOTSUP));
2021 }
2022
2023 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
2024 if (error != 0) {
2025 sa_buf_rele(*db, tag);
2026 return (error);
2027 }
2028
2029 return (0);
2030 }
2031
2032 static void
2033 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, const void *tag)
2034 {
2035 sa_handle_destroy(hdl);
2036 sa_buf_rele(db, tag);
2037 }
2038
2039 /*
2040 * Given an object number, return its parent object number and whether
2041 * or not the object is an extended attribute directory.
2042 */
2043 static int
2044 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
2045 uint64_t *pobjp, int *is_xattrdir)
2046 {
2047 uint64_t parent;
2048 uint64_t pflags;
2049 uint64_t mode;
2050 uint64_t parent_mode;
2051 sa_bulk_attr_t bulk[3];
2052 sa_handle_t *sa_hdl;
2053 dmu_buf_t *sa_db;
2054 int count = 0;
2055 int error;
2056
2057 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
2058 &parent, sizeof (parent));
2059 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
2060 &pflags, sizeof (pflags));
2061 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2062 &mode, sizeof (mode));
2063
2064 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
2065 return (error);
2066
2067 /*
2068 * When a link is removed its parent pointer is not changed and will
2069 * be invalid. There are two cases where a link is removed but the
2070 * file stays around, when it goes to the delete queue and when there
2071 * are additional links.
2072 */
2073 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
2074 if (error != 0)
2075 return (error);
2076
2077 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
2078 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2079 if (error != 0)
2080 return (error);
2081
2082 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
2083
2084 /*
2085 * Extended attributes can be applied to files, directories, etc.
2086 * Otherwise the parent must be a directory.
2087 */
2088 if (!*is_xattrdir && !S_ISDIR(parent_mode))
2089 return (SET_ERROR(EINVAL));
2090
2091 *pobjp = parent;
2092
2093 return (0);
2094 }
2095
2096 /*
2097 * Given an object number, return some zpl level statistics
2098 */
2099 static int
2100 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2101 zfs_stat_t *sb)
2102 {
2103 sa_bulk_attr_t bulk[4];
2104 int count = 0;
2105
2106 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2107 &sb->zs_mode, sizeof (sb->zs_mode));
2108 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2109 &sb->zs_gen, sizeof (sb->zs_gen));
2110 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2111 &sb->zs_links, sizeof (sb->zs_links));
2112 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2113 &sb->zs_ctime, sizeof (sb->zs_ctime));
2114
2115 return (sa_bulk_lookup(hdl, bulk, count));
2116 }
2117
2118 static int
2119 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2120 sa_attr_type_t *sa_table, char *buf, int len)
2121 {
2122 sa_handle_t *sa_hdl;
2123 sa_handle_t *prevhdl = NULL;
2124 dmu_buf_t *prevdb = NULL;
2125 dmu_buf_t *sa_db = NULL;
2126 char *path = buf + len - 1;
2127 int error;
2128
2129 *path = '\0';
2130 sa_hdl = hdl;
2131
2132 uint64_t deleteq_obj;
2133 VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ,
2134 ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj));
2135 error = zap_lookup_int(osp, deleteq_obj, obj);
2136 if (error == 0) {
2137 return (ESTALE);
2138 } else if (error != ENOENT) {
2139 return (error);
2140 }
2141
2142 for (;;) {
2143 uint64_t pobj = 0;
2144 char component[MAXNAMELEN + 2];
2145 size_t complen;
2146 int is_xattrdir = 0;
2147
2148 if (prevdb) {
2149 ASSERT(prevhdl != NULL);
2150 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2151 }
2152
2153 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2154 &is_xattrdir)) != 0)
2155 break;
2156
2157 if (pobj == obj) {
2158 if (path[0] != '/')
2159 *--path = '/';
2160 break;
2161 }
2162
2163 component[0] = '/';
2164 if (is_xattrdir) {
2165 strcpy(component + 1, "<xattrdir>");
2166 } else {
2167 error = zap_value_search(osp, pobj, obj,
2168 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2169 if (error != 0)
2170 break;
2171 }
2172
2173 complen = strlen(component);
2174 path -= complen;
2175 ASSERT(path >= buf);
2176 memcpy(path, component, complen);
2177 obj = pobj;
2178
2179 if (sa_hdl != hdl) {
2180 prevhdl = sa_hdl;
2181 prevdb = sa_db;
2182 }
2183 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2184 if (error != 0) {
2185 sa_hdl = prevhdl;
2186 sa_db = prevdb;
2187 break;
2188 }
2189 }
2190
2191 if (sa_hdl != NULL && sa_hdl != hdl) {
2192 ASSERT(sa_db != NULL);
2193 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2194 }
2195
2196 if (error == 0)
2197 (void) memmove(buf, path, buf + len - path);
2198
2199 return (error);
2200 }
2201
2202 int
2203 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2204 {
2205 sa_attr_type_t *sa_table;
2206 sa_handle_t *hdl;
2207 dmu_buf_t *db;
2208 int error;
2209
2210 error = zfs_sa_setup(osp, &sa_table);
2211 if (error != 0)
2212 return (error);
2213
2214 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2215 if (error != 0)
2216 return (error);
2217
2218 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2219
2220 zfs_release_sa_handle(hdl, db, FTAG);
2221 return (error);
2222 }
2223
2224 int
2225 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2226 char *buf, int len)
2227 {
2228 char *path = buf + len - 1;
2229 sa_attr_type_t *sa_table;
2230 sa_handle_t *hdl;
2231 dmu_buf_t *db;
2232 int error;
2233
2234 *path = '\0';
2235
2236 error = zfs_sa_setup(osp, &sa_table);
2237 if (error != 0)
2238 return (error);
2239
2240 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2241 if (error != 0)
2242 return (error);
2243
2244 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2245 if (error != 0) {
2246 zfs_release_sa_handle(hdl, db, FTAG);
2247 return (error);
2248 }
2249
2250 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2251
2252 zfs_release_sa_handle(hdl, db, FTAG);
2253 return (error);
2254 }
2255
2256 #if defined(_KERNEL)
2257 EXPORT_SYMBOL(zfs_create_fs);
2258 EXPORT_SYMBOL(zfs_obj_to_path);
2259
2260 /* CSTYLED */
2261 module_param(zfs_object_mutex_size, uint, 0644);
2262 MODULE_PARM_DESC(zfs_object_mutex_size, "Size of znode hold array");
2263 module_param(zfs_unlink_suspend_progress, int, 0644);
2264 MODULE_PARM_DESC(zfs_unlink_suspend_progress, "Set to prevent async unlinks "
2265 "(debug - leaks space into the unlinked set)");
2266 #endif
ZFS On Linux mirror for PVE