?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Fix O_APPEND for Linux 3.15 and older kernels
[zfs.git] / module / os / linux / zfs / zfs_znode.c
blobb76e65d168223e9a9ccb74220977179fbcf54778
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 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 return (0);
138 }
139
140 static void
141 zfs_znode_cache_destructor(void *buf, void *arg)
142 {
143 (void) arg;
144 znode_t *zp = buf;
145
146 ASSERT(!list_link_active(&zp->z_link_node));
147 mutex_destroy(&zp->z_lock);
148 rw_destroy(&zp->z_parent_lock);
149 rw_destroy(&zp->z_name_lock);
150 mutex_destroy(&zp->z_acl_lock);
151 rw_destroy(&zp->z_xattr_lock);
152 zfs_rangelock_fini(&zp->z_rangelock);
153
154 ASSERT3P(zp->z_dirlocks, ==, NULL);
155 ASSERT3P(zp->z_acl_cached, ==, NULL);
156 ASSERT3P(zp->z_xattr_cached, ==, NULL);
157 }
158
159 static int
160 zfs_znode_hold_cache_constructor(void *buf, void *arg, int kmflags)
161 {
162 (void) arg, (void) kmflags;
163 znode_hold_t *zh = buf;
164
165 mutex_init(&zh->zh_lock, NULL, MUTEX_DEFAULT, NULL);
166 zfs_refcount_create(&zh->zh_refcount);
167 zh->zh_obj = ZFS_NO_OBJECT;
168
169 return (0);
170 }
171
172 static void
173 zfs_znode_hold_cache_destructor(void *buf, void *arg)
174 {
175 (void) arg;
176 znode_hold_t *zh = buf;
177
178 mutex_destroy(&zh->zh_lock);
179 zfs_refcount_destroy(&zh->zh_refcount);
180 }
181
182 void
183 zfs_znode_init(void)
184 {
185 /*
186 * Initialize zcache. The KMC_SLAB hint is used in order that it be
187 * backed by kmalloc() when on the Linux slab in order that any
188 * wait_on_bit() operations on the related inode operate properly.
189 */
190 ASSERT(znode_cache == NULL);
191 znode_cache = kmem_cache_create("zfs_znode_cache",
192 sizeof (znode_t), 0, zfs_znode_cache_constructor,
193 zfs_znode_cache_destructor, NULL, NULL, NULL, KMC_SLAB);
194
195 ASSERT(znode_hold_cache == NULL);
196 znode_hold_cache = kmem_cache_create("zfs_znode_hold_cache",
197 sizeof (znode_hold_t), 0, zfs_znode_hold_cache_constructor,
198 zfs_znode_hold_cache_destructor, NULL, NULL, NULL, 0);
199 }
200
201 void
202 zfs_znode_fini(void)
203 {
204 /*
205 * Cleanup zcache
206 */
207 if (znode_cache)
208 kmem_cache_destroy(znode_cache);
209 znode_cache = NULL;
210
211 if (znode_hold_cache)
212 kmem_cache_destroy(znode_hold_cache);
213 znode_hold_cache = NULL;
214 }
215
216 /*
217 * The zfs_znode_hold_enter() / zfs_znode_hold_exit() functions are used to
218 * serialize access to a znode and its SA buffer while the object is being
219 * created or destroyed. This kind of locking would normally reside in the
220 * znode itself but in this case that's impossible because the znode and SA
221 * buffer may not yet exist. Therefore the locking is handled externally
222 * with an array of mutexes and AVLs trees which contain per-object locks.
223 *
224 * In zfs_znode_hold_enter() a per-object lock is created as needed, inserted
225 * in to the correct AVL tree and finally the per-object lock is held. In
226 * zfs_znode_hold_exit() the process is reversed. The per-object lock is
227 * released, removed from the AVL tree and destroyed if there are no waiters.
228 *
229 * This scheme has two important properties:
230 *
231 * 1) No memory allocations are performed while holding one of the z_hold_locks.
232 * This ensures evict(), which can be called from direct memory reclaim, will
233 * never block waiting on a z_hold_locks which just happens to have hashed
234 * to the same index.
235 *
236 * 2) All locks used to serialize access to an object are per-object and never
237 * shared. This minimizes lock contention without creating a large number
238 * of dedicated locks.
239 *
240 * On the downside it does require znode_lock_t structures to be frequently
241 * allocated and freed. However, because these are backed by a kmem cache
242 * and very short lived this cost is minimal.
243 */
244 int
245 zfs_znode_hold_compare(const void *a, const void *b)
246 {
247 const znode_hold_t *zh_a = (const znode_hold_t *)a;
248 const znode_hold_t *zh_b = (const znode_hold_t *)b;
249
250 return (TREE_CMP(zh_a->zh_obj, zh_b->zh_obj));
251 }
252
253 static boolean_t __maybe_unused
254 zfs_znode_held(zfsvfs_t *zfsvfs, uint64_t obj)
255 {
256 znode_hold_t *zh, search;
257 int i = ZFS_OBJ_HASH(zfsvfs, obj);
258 boolean_t held;
259
260 search.zh_obj = obj;
261
262 mutex_enter(&zfsvfs->z_hold_locks[i]);
263 zh = avl_find(&zfsvfs->z_hold_trees[i], &search, NULL);
264 held = (zh && MUTEX_HELD(&zh->zh_lock)) ? B_TRUE : B_FALSE;
265 mutex_exit(&zfsvfs->z_hold_locks[i]);
266
267 return (held);
268 }
269
270 static znode_hold_t *
271 zfs_znode_hold_enter(zfsvfs_t *zfsvfs, uint64_t obj)
272 {
273 znode_hold_t *zh, *zh_new, search;
274 int i = ZFS_OBJ_HASH(zfsvfs, obj);
275 boolean_t found = B_FALSE;
276
277 zh_new = kmem_cache_alloc(znode_hold_cache, KM_SLEEP);
278 zh_new->zh_obj = obj;
279 search.zh_obj = obj;
280
281 mutex_enter(&zfsvfs->z_hold_locks[i]);
282 zh = avl_find(&zfsvfs->z_hold_trees[i], &search, NULL);
283 if (likely(zh == NULL)) {
284 zh = zh_new;
285 avl_add(&zfsvfs->z_hold_trees[i], zh);
286 } else {
287 ASSERT3U(zh->zh_obj, ==, obj);
288 found = B_TRUE;
289 }
290 zfs_refcount_add(&zh->zh_refcount, NULL);
291 mutex_exit(&zfsvfs->z_hold_locks[i]);
292
293 if (found == B_TRUE)
294 kmem_cache_free(znode_hold_cache, zh_new);
295
296 ASSERT(MUTEX_NOT_HELD(&zh->zh_lock));
297 ASSERT3S(zfs_refcount_count(&zh->zh_refcount), >, 0);
298 mutex_enter(&zh->zh_lock);
299
300 return (zh);
301 }
302
303 static void
304 zfs_znode_hold_exit(zfsvfs_t *zfsvfs, znode_hold_t *zh)
305 {
306 int i = ZFS_OBJ_HASH(zfsvfs, zh->zh_obj);
307 boolean_t remove = B_FALSE;
308
309 ASSERT(zfs_znode_held(zfsvfs, zh->zh_obj));
310 ASSERT3S(zfs_refcount_count(&zh->zh_refcount), >, 0);
311 mutex_exit(&zh->zh_lock);
312
313 mutex_enter(&zfsvfs->z_hold_locks[i]);
314 if (zfs_refcount_remove(&zh->zh_refcount, NULL) == 0) {
315 avl_remove(&zfsvfs->z_hold_trees[i], zh);
316 remove = B_TRUE;
317 }
318 mutex_exit(&zfsvfs->z_hold_locks[i]);
319
320 if (remove == B_TRUE)
321 kmem_cache_free(znode_hold_cache, zh);
322 }
323
324 dev_t
325 zfs_cmpldev(uint64_t dev)
326 {
327 return (dev);
328 }
329
330 static void
331 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
332 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
333 {
334 ASSERT(zfs_znode_held(zfsvfs, zp->z_id));
335
336 mutex_enter(&zp->z_lock);
337
338 ASSERT(zp->z_sa_hdl == NULL);
339 ASSERT(zp->z_acl_cached == NULL);
340 if (sa_hdl == NULL) {
341 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
342 SA_HDL_SHARED, &zp->z_sa_hdl));
343 } else {
344 zp->z_sa_hdl = sa_hdl;
345 sa_set_userp(sa_hdl, zp);
346 }
347
348 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
349
350 mutex_exit(&zp->z_lock);
351 }
352
353 void
354 zfs_znode_dmu_fini(znode_t *zp)
355 {
356 ASSERT(zfs_znode_held(ZTOZSB(zp), zp->z_id) || zp->z_unlinked ||
357 RW_WRITE_HELD(&ZTOZSB(zp)->z_teardown_inactive_lock));
358
359 sa_handle_destroy(zp->z_sa_hdl);
360 zp->z_sa_hdl = NULL;
361 }
362
363 /*
364 * Called by new_inode() to allocate a new inode.
365 */
366 int
367 zfs_inode_alloc(struct super_block *sb, struct inode **ip)
368 {
369 znode_t *zp;
370
371 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
372 *ip = ZTOI(zp);
373
374 return (0);
375 }
376
377 /*
378 * Called in multiple places when an inode should be destroyed.
379 */
380 void
381 zfs_inode_destroy(struct inode *ip)
382 {
383 znode_t *zp = ITOZ(ip);
384 zfsvfs_t *zfsvfs = ZTOZSB(zp);
385
386 mutex_enter(&zfsvfs->z_znodes_lock);
387 if (list_link_active(&zp->z_link_node)) {
388 list_remove(&zfsvfs->z_all_znodes, zp);
389 zfsvfs->z_nr_znodes--;
390 }
391 mutex_exit(&zfsvfs->z_znodes_lock);
392
393 if (zp->z_acl_cached) {
394 zfs_acl_free(zp->z_acl_cached);
395 zp->z_acl_cached = NULL;
396 }
397
398 if (zp->z_xattr_cached) {
399 nvlist_free(zp->z_xattr_cached);
400 zp->z_xattr_cached = NULL;
401 }
402
403 kmem_cache_free(znode_cache, zp);
404 }
405
406 static void
407 zfs_inode_set_ops(zfsvfs_t *zfsvfs, struct inode *ip)
408 {
409 uint64_t rdev = 0;
410
411 switch (ip->i_mode & S_IFMT) {
412 case S_IFREG:
413 ip->i_op = &zpl_inode_operations;
414 ip->i_fop = &zpl_file_operations;
415 ip->i_mapping->a_ops = &zpl_address_space_operations;
416 break;
417
418 case S_IFDIR:
419 ip->i_op = &zpl_dir_inode_operations;
420 ip->i_fop = &zpl_dir_file_operations;
421 ITOZ(ip)->z_zn_prefetch = B_TRUE;
422 break;
423
424 case S_IFLNK:
425 ip->i_op = &zpl_symlink_inode_operations;
426 break;
427
428 /*
429 * rdev is only stored in a SA only for device files.
430 */
431 case S_IFCHR:
432 case S_IFBLK:
433 (void) sa_lookup(ITOZ(ip)->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), &rdev,
434 sizeof (rdev));
435 zfs_fallthrough;
436 case S_IFIFO:
437 case S_IFSOCK:
438 init_special_inode(ip, ip->i_mode, rdev);
439 ip->i_op = &zpl_special_inode_operations;
440 break;
441
442 default:
443 zfs_panic_recover("inode %llu has invalid mode: 0x%x\n",
444 (u_longlong_t)ip->i_ino, ip->i_mode);
445
446 /* Assume the inode is a file and attempt to continue */
447 ip->i_mode = S_IFREG | 0644;
448 ip->i_op = &zpl_inode_operations;
449 ip->i_fop = &zpl_file_operations;
450 ip->i_mapping->a_ops = &zpl_address_space_operations;
451 break;
452 }
453 }
454
455 static void
456 zfs_set_inode_flags(znode_t *zp, struct inode *ip)
457 {
458 /*
459 * Linux and Solaris have different sets of file attributes, so we
460 * restrict this conversion to the intersection of the two.
461 */
462 #ifdef HAVE_INODE_SET_FLAGS
463 unsigned int flags = 0;
464 if (zp->z_pflags & ZFS_IMMUTABLE)
465 flags |= S_IMMUTABLE;
466 if (zp->z_pflags & ZFS_APPENDONLY)
467 flags |= S_APPEND;
468
469 inode_set_flags(ip, flags, S_IMMUTABLE|S_APPEND);
470 #else
471 if (zp->z_pflags & ZFS_IMMUTABLE)
472 ip->i_flags |= S_IMMUTABLE;
473 else
474 ip->i_flags &= ~S_IMMUTABLE;
475
476 if (zp->z_pflags & ZFS_APPENDONLY)
477 ip->i_flags |= S_APPEND;
478 else
479 ip->i_flags &= ~S_APPEND;
480 #endif
481 }
482
483 /*
484 * Update the embedded inode given the znode.
485 */
486 void
487 zfs_znode_update_vfs(znode_t *zp)
488 {
489 zfsvfs_t *zfsvfs;
490 struct inode *ip;
491 uint32_t blksize;
492 u_longlong_t i_blocks;
493
494 ASSERT(zp != NULL);
495 zfsvfs = ZTOZSB(zp);
496 ip = ZTOI(zp);
497
498 /* Skip .zfs control nodes which do not exist on disk. */
499 if (zfsctl_is_node(ip))
500 return;
501
502 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &blksize, &i_blocks);
503
504 spin_lock(&ip->i_lock);
505 ip->i_mode = zp->z_mode;
506 ip->i_blocks = i_blocks;
507 i_size_write(ip, zp->z_size);
508 spin_unlock(&ip->i_lock);
509 }
510
511
512 /*
513 * Construct a znode+inode and initialize.
514 *
515 * This does not do a call to dmu_set_user() that is
516 * up to the caller to do, in case you don't want to
517 * return the znode
518 */
519 static znode_t *
520 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
521 dmu_object_type_t obj_type, sa_handle_t *hdl)
522 {
523 znode_t *zp;
524 struct inode *ip;
525 uint64_t mode;
526 uint64_t parent;
527 uint64_t tmp_gen;
528 uint64_t links;
529 uint64_t z_uid, z_gid;
530 uint64_t atime[2], mtime[2], ctime[2], btime[2];
531 uint64_t projid = ZFS_DEFAULT_PROJID;
532 sa_bulk_attr_t bulk[12];
533 int count = 0;
534
535 ASSERT(zfsvfs != NULL);
536
537 ip = new_inode(zfsvfs->z_sb);
538 if (ip == NULL)
539 return (NULL);
540
541 zp = ITOZ(ip);
542 ASSERT(zp->z_dirlocks == NULL);
543 ASSERT3P(zp->z_acl_cached, ==, NULL);
544 ASSERT3P(zp->z_xattr_cached, ==, NULL);
545 zp->z_unlinked = B_FALSE;
546 zp->z_atime_dirty = B_FALSE;
547 zp->z_is_mapped = B_FALSE;
548 zp->z_is_ctldir = B_FALSE;
549 zp->z_is_stale = B_FALSE;
550 zp->z_suspended = B_FALSE;
551 zp->z_sa_hdl = NULL;
552 zp->z_mapcnt = 0;
553 zp->z_id = db->db_object;
554 zp->z_blksz = blksz;
555 zp->z_seq = 0x7A4653;
556 zp->z_sync_cnt = 0;
557
558 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
559
560 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
561 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &tmp_gen, 8);
562 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
563 &zp->z_size, 8);
564 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
565 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
566 &zp->z_pflags, 8);
567 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
568 &parent, 8);
569 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &z_uid, 8);
570 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &z_gid, 8);
571 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
572 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
573 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
574 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &btime, 16);
575
576 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || tmp_gen == 0 ||
577 (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
578 (zp->z_pflags & ZFS_PROJID) &&
579 sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) {
580 if (hdl == NULL)
581 sa_handle_destroy(zp->z_sa_hdl);
582 zp->z_sa_hdl = NULL;
583 goto error;
584 }
585
586 zp->z_projid = projid;
587 zp->z_mode = ip->i_mode = mode;
588 ip->i_generation = (uint32_t)tmp_gen;
589 ip->i_blkbits = SPA_MINBLOCKSHIFT;
590 set_nlink(ip, (uint32_t)links);
591 zfs_uid_write(ip, z_uid);
592 zfs_gid_write(ip, z_gid);
593 zfs_set_inode_flags(zp, ip);
594
595 /* Cache the xattr parent id */
596 if (zp->z_pflags & ZFS_XATTR)
597 zp->z_xattr_parent = parent;
598
599 ZFS_TIME_DECODE(&ip->i_atime, atime);
600 ZFS_TIME_DECODE(&ip->i_mtime, mtime);
601 ZFS_TIME_DECODE(&ip->i_ctime, ctime);
602 ZFS_TIME_DECODE(&zp->z_btime, btime);
603
604 ip->i_ino = zp->z_id;
605 zfs_znode_update_vfs(zp);
606 zfs_inode_set_ops(zfsvfs, ip);
607
608 /*
609 * The only way insert_inode_locked() can fail is if the ip->i_ino
610 * number is already hashed for this super block. This can never
611 * happen because the inode numbers map 1:1 with the object numbers.
612 *
613 * Exceptions include rolling back a mounted file system, either
614 * from the zfs rollback or zfs recv command.
615 *
616 * Active inodes are unhashed during the rollback, but since zrele
617 * can happen asynchronously, we can't guarantee they've been
618 * unhashed. This can cause hash collisions in unlinked drain
619 * processing so do not hash unlinked znodes.
620 */
621 if (links > 0)
622 VERIFY3S(insert_inode_locked(ip), ==, 0);
623
624 mutex_enter(&zfsvfs->z_znodes_lock);
625 list_insert_tail(&zfsvfs->z_all_znodes, zp);
626 zfsvfs->z_nr_znodes++;
627 mutex_exit(&zfsvfs->z_znodes_lock);
628
629 if (links > 0)
630 unlock_new_inode(ip);
631 return (zp);
632
633 error:
634 iput(ip);
635 return (NULL);
636 }
637
638 /*
639 * Safely mark an inode dirty. Inodes which are part of a read-only
640 * file system or snapshot may not be dirtied.
641 */
642 void
643 zfs_mark_inode_dirty(struct inode *ip)
644 {
645 zfsvfs_t *zfsvfs = ITOZSB(ip);
646
647 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
648 return;
649
650 mark_inode_dirty(ip);
651 }
652
653 static uint64_t empty_xattr;
654 static uint64_t pad[4];
655 static zfs_acl_phys_t acl_phys;
656 /*
657 * Create a new DMU object to hold a zfs znode.
658 *
659 * IN: dzp - parent directory for new znode
660 * vap - file attributes for new znode
661 * tx - dmu transaction id for zap operations
662 * cr - credentials of caller
663 * flag - flags:
664 * IS_ROOT_NODE - new object will be root
665 * IS_TMPFILE - new object is of O_TMPFILE
666 * IS_XATTR - new object is an attribute
667 * acl_ids - ACL related attributes
668 *
669 * OUT: zpp - allocated znode (set to dzp if IS_ROOT_NODE)
670 *
671 */
672 void
673 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
674 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
675 {
676 uint64_t crtime[2], atime[2], mtime[2], ctime[2];
677 uint64_t mode, size, links, parent, pflags;
678 uint64_t projid = ZFS_DEFAULT_PROJID;
679 uint64_t rdev = 0;
680 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
681 dmu_buf_t *db;
682 inode_timespec_t now;
683 uint64_t gen, obj;
684 int bonuslen;
685 int dnodesize;
686 sa_handle_t *sa_hdl;
687 dmu_object_type_t obj_type;
688 sa_bulk_attr_t *sa_attrs;
689 int cnt = 0;
690 zfs_acl_locator_cb_t locate = { 0 };
691 znode_hold_t *zh;
692
693 if (zfsvfs->z_replay) {
694 obj = vap->va_nodeid;
695 now = vap->va_ctime; /* see zfs_replay_create() */
696 gen = vap->va_nblocks; /* ditto */
697 dnodesize = vap->va_fsid; /* ditto */
698 } else {
699 obj = 0;
700 gethrestime(&now);
701 gen = dmu_tx_get_txg(tx);
702 dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
703 }
704
705 if (dnodesize == 0)
706 dnodesize = DNODE_MIN_SIZE;
707
708 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
709
710 bonuslen = (obj_type == DMU_OT_SA) ?
711 DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
712
713 /*
714 * Create a new DMU object.
715 */
716 /*
717 * There's currently no mechanism for pre-reading the blocks that will
718 * be needed to allocate a new object, so we accept the small chance
719 * that there will be an i/o error and we will fail one of the
720 * assertions below.
721 */
722 if (S_ISDIR(vap->va_mode)) {
723 if (zfsvfs->z_replay) {
724 VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj,
725 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
726 obj_type, bonuslen, dnodesize, tx));
727 } else {
728 obj = zap_create_norm_dnsize(zfsvfs->z_os,
729 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
730 obj_type, bonuslen, dnodesize, tx);
731 }
732 } else {
733 if (zfsvfs->z_replay) {
734 VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj,
735 DMU_OT_PLAIN_FILE_CONTENTS, 0,
736 obj_type, bonuslen, dnodesize, tx));
737 } else {
738 obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
739 DMU_OT_PLAIN_FILE_CONTENTS, 0,
740 obj_type, bonuslen, dnodesize, tx);
741 }
742 }
743
744 zh = zfs_znode_hold_enter(zfsvfs, obj);
745 VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
746
747 /*
748 * If this is the root, fix up the half-initialized parent pointer
749 * to reference the just-allocated physical data area.
750 */
751 if (flag & IS_ROOT_NODE) {
752 dzp->z_id = obj;
753 }
754
755 /*
756 * If parent is an xattr, so am I.
757 */
758 if (dzp->z_pflags & ZFS_XATTR) {
759 flag |= IS_XATTR;
760 }
761
762 if (zfsvfs->z_use_fuids)
763 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
764 else
765 pflags = 0;
766
767 if (S_ISDIR(vap->va_mode)) {
768 size = 2; /* contents ("." and "..") */
769 links = 2;
770 } else {
771 size = 0;
772 links = (flag & IS_TMPFILE) ? 0 : 1;
773 }
774
775 if (S_ISBLK(vap->va_mode) || S_ISCHR(vap->va_mode))
776 rdev = vap->va_rdev;
777
778 parent = dzp->z_id;
779 mode = acl_ids->z_mode;
780 if (flag & IS_XATTR)
781 pflags |= ZFS_XATTR;
782
783 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode)) {
784 /*
785 * With ZFS_PROJID flag, we can easily know whether there is
786 * project ID stored on disk or not. See zfs_space_delta_cb().
787 */
788 if (obj_type != DMU_OT_ZNODE &&
789 dmu_objset_projectquota_enabled(zfsvfs->z_os))
790 pflags |= ZFS_PROJID;
791
792 /*
793 * Inherit project ID from parent if required.
794 */
795 projid = zfs_inherit_projid(dzp);
796 if (dzp->z_pflags & ZFS_PROJINHERIT)
797 pflags |= ZFS_PROJINHERIT;
798 }
799
800 /*
801 * No execs denied will be determined when zfs_mode_compute() is called.
802 */
803 pflags |= acl_ids->z_aclp->z_hints &
804 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
805 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
806
807 ZFS_TIME_ENCODE(&now, crtime);
808 ZFS_TIME_ENCODE(&now, ctime);
809
810 if (vap->va_mask & ATTR_ATIME) {
811 ZFS_TIME_ENCODE(&vap->va_atime, atime);
812 } else {
813 ZFS_TIME_ENCODE(&now, atime);
814 }
815
816 if (vap->va_mask & ATTR_MTIME) {
817 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
818 } else {
819 ZFS_TIME_ENCODE(&now, mtime);
820 }
821
822 /* Now add in all of the "SA" attributes */
823 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
824 &sa_hdl));
825
826 /*
827 * Setup the array of attributes to be replaced/set on the new file
828 *
829 * order for DMU_OT_ZNODE is critical since it needs to be constructed
830 * in the old znode_phys_t format. Don't change this ordering
831 */
832 sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
833
834 if (obj_type == DMU_OT_ZNODE) {
835 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
836 NULL, &atime, 16);
837 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
838 NULL, &mtime, 16);
839 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
840 NULL, &ctime, 16);
841 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
842 NULL, &crtime, 16);
843 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
844 NULL, &gen, 8);
845 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
846 NULL, &mode, 8);
847 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
848 NULL, &size, 8);
849 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
850 NULL, &parent, 8);
851 } else {
852 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
853 NULL, &mode, 8);
854 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
855 NULL, &size, 8);
856 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
857 NULL, &gen, 8);
858 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs),
859 NULL, &acl_ids->z_fuid, 8);
860 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs),
861 NULL, &acl_ids->z_fgid, 8);
862 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
863 NULL, &parent, 8);
864 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
865 NULL, &pflags, 8);
866 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
867 NULL, &atime, 16);
868 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
869 NULL, &mtime, 16);
870 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
871 NULL, &ctime, 16);
872 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
873 NULL, &crtime, 16);
874 }
875
876 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
877
878 if (obj_type == DMU_OT_ZNODE) {
879 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
880 &empty_xattr, 8);
881 } else if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
882 pflags & ZFS_PROJID) {
883 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PROJID(zfsvfs),
884 NULL, &projid, 8);
885 }
886 if (obj_type == DMU_OT_ZNODE ||
887 (S_ISBLK(vap->va_mode) || S_ISCHR(vap->va_mode))) {
888 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
889 NULL, &rdev, 8);
890 }
891 if (obj_type == DMU_OT_ZNODE) {
892 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
893 NULL, &pflags, 8);
894 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
895 &acl_ids->z_fuid, 8);
896 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
897 &acl_ids->z_fgid, 8);
898 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
899 sizeof (uint64_t) * 4);
900 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
901 &acl_phys, sizeof (zfs_acl_phys_t));
902 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
903 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
904 &acl_ids->z_aclp->z_acl_count, 8);
905 locate.cb_aclp = acl_ids->z_aclp;
906 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
907 zfs_acl_data_locator, &locate,
908 acl_ids->z_aclp->z_acl_bytes);
909 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
910 acl_ids->z_fuid, acl_ids->z_fgid);
911 }
912
913 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
914
915 if (!(flag & IS_ROOT_NODE)) {
916 /*
917 * The call to zfs_znode_alloc() may fail if memory is low
918 * via the call path: alloc_inode() -> inode_init_always() ->
919 * security_inode_alloc() -> inode_alloc_security(). Since
920 * the existing code is written such that zfs_mknode() can
921 * not fail retry until sufficient memory has been reclaimed.
922 */
923 do {
924 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
925 } while (*zpp == NULL);
926
927 VERIFY(*zpp != NULL);
928 VERIFY(dzp != NULL);
929 } else {
930 /*
931 * If we are creating the root node, the "parent" we
932 * passed in is the znode for the root.
933 */
934 *zpp = dzp;
935
936 (*zpp)->z_sa_hdl = sa_hdl;
937 }
938
939 (*zpp)->z_pflags = pflags;
940 (*zpp)->z_mode = ZTOI(*zpp)->i_mode = mode;
941 (*zpp)->z_dnodesize = dnodesize;
942 (*zpp)->z_projid = projid;
943
944 if (obj_type == DMU_OT_ZNODE ||
945 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
946 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
947 }
948 kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
949 zfs_znode_hold_exit(zfsvfs, zh);
950 }
951
952 /*
953 * Update in-core attributes. It is assumed the caller will be doing an
954 * sa_bulk_update to push the changes out.
955 */
956 void
957 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
958 {
959 xoptattr_t *xoap;
960 boolean_t update_inode = B_FALSE;
961
962 xoap = xva_getxoptattr(xvap);
963 ASSERT(xoap);
964
965 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
966 uint64_t times[2];
967 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
968 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(ZTOZSB(zp)),
969 &times, sizeof (times), tx);
970 XVA_SET_RTN(xvap, XAT_CREATETIME);
971 }
972 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
973 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
974 zp->z_pflags, tx);
975 XVA_SET_RTN(xvap, XAT_READONLY);
976 }
977 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
978 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
979 zp->z_pflags, tx);
980 XVA_SET_RTN(xvap, XAT_HIDDEN);
981 }
982 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
983 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
984 zp->z_pflags, tx);
985 XVA_SET_RTN(xvap, XAT_SYSTEM);
986 }
987 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
988 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
989 zp->z_pflags, tx);
990 XVA_SET_RTN(xvap, XAT_ARCHIVE);
991 }
992 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
993 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
994 zp->z_pflags, tx);
995 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
996
997 update_inode = B_TRUE;
998 }
999 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1000 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1001 zp->z_pflags, tx);
1002 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1003 }
1004 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1005 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1006 zp->z_pflags, tx);
1007 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1008
1009 update_inode = B_TRUE;
1010 }
1011 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1012 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1013 zp->z_pflags, tx);
1014 XVA_SET_RTN(xvap, XAT_NODUMP);
1015 }
1016 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1017 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1018 zp->z_pflags, tx);
1019 XVA_SET_RTN(xvap, XAT_OPAQUE);
1020 }
1021 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1022 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1023 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1024 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1025 }
1026 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1027 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1028 zp->z_pflags, tx);
1029 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1030 }
1031 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1032 zfs_sa_set_scanstamp(zp, xvap, tx);
1033 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1034 }
1035 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1036 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1037 zp->z_pflags, tx);
1038 XVA_SET_RTN(xvap, XAT_REPARSE);
1039 }
1040 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1041 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1042 zp->z_pflags, tx);
1043 XVA_SET_RTN(xvap, XAT_OFFLINE);
1044 }
1045 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1046 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1047 zp->z_pflags, tx);
1048 XVA_SET_RTN(xvap, XAT_SPARSE);
1049 }
1050 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
1051 ZFS_ATTR_SET(zp, ZFS_PROJINHERIT, xoap->xoa_projinherit,
1052 zp->z_pflags, tx);
1053 XVA_SET_RTN(xvap, XAT_PROJINHERIT);
1054 }
1055
1056 if (update_inode)
1057 zfs_set_inode_flags(zp, ZTOI(zp));
1058 }
1059
1060 int
1061 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1062 {
1063 dmu_object_info_t doi;
1064 dmu_buf_t *db;
1065 znode_t *zp;
1066 znode_hold_t *zh;
1067 int err;
1068 sa_handle_t *hdl;
1069
1070 *zpp = NULL;
1071
1072 again:
1073 zh = zfs_znode_hold_enter(zfsvfs, obj_num);
1074
1075 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1076 if (err) {
1077 zfs_znode_hold_exit(zfsvfs, zh);
1078 return (err);
1079 }
1080
1081 dmu_object_info_from_db(db, &doi);
1082 if (doi.doi_bonus_type != DMU_OT_SA &&
1083 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1084 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1085 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1086 sa_buf_rele(db, NULL);
1087 zfs_znode_hold_exit(zfsvfs, zh);
1088 return (SET_ERROR(EINVAL));
1089 }
1090
1091 hdl = dmu_buf_get_user(db);
1092 if (hdl != NULL) {
1093 zp = sa_get_userdata(hdl);
1094
1095
1096 /*
1097 * Since "SA" does immediate eviction we
1098 * should never find a sa handle that doesn't
1099 * know about the znode.
1100 */
1101
1102 ASSERT3P(zp, !=, NULL);
1103
1104 mutex_enter(&zp->z_lock);
1105 ASSERT3U(zp->z_id, ==, obj_num);
1106 /*
1107 * If zp->z_unlinked is set, the znode is already marked
1108 * for deletion and should not be discovered. Check this
1109 * after checking igrab() due to fsetxattr() & O_TMPFILE.
1110 *
1111 * If igrab() returns NULL the VFS has independently
1112 * determined the inode should be evicted and has
1113 * called iput_final() to start the eviction process.
1114 * The SA handle is still valid but because the VFS
1115 * requires that the eviction succeed we must drop
1116 * our locks and references to allow the eviction to
1117 * complete. The zfs_zget() may then be retried.
1118 *
1119 * This unlikely case could be optimized by registering
1120 * a sops->drop_inode() callback. The callback would
1121 * need to detect the active SA hold thereby informing
1122 * the VFS that this inode should not be evicted.
1123 */
1124 if (igrab(ZTOI(zp)) == NULL) {
1125 if (zp->z_unlinked)
1126 err = SET_ERROR(ENOENT);
1127 else
1128 err = SET_ERROR(EAGAIN);
1129 } else {
1130 *zpp = zp;
1131 err = 0;
1132 }
1133
1134 mutex_exit(&zp->z_lock);
1135 sa_buf_rele(db, NULL);
1136 zfs_znode_hold_exit(zfsvfs, zh);
1137
1138 if (err == EAGAIN) {
1139 /* inode might need this to finish evict */
1140 cond_resched();
1141 goto again;
1142 }
1143 return (err);
1144 }
1145
1146 /*
1147 * Not found create new znode/vnode but only if file exists.
1148 *
1149 * There is a small window where zfs_vget() could
1150 * find this object while a file create is still in
1151 * progress. This is checked for in zfs_znode_alloc()
1152 *
1153 * if zfs_znode_alloc() fails it will drop the hold on the
1154 * bonus buffer.
1155 */
1156 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1157 doi.doi_bonus_type, NULL);
1158 if (zp == NULL) {
1159 err = SET_ERROR(ENOENT);
1160 } else {
1161 *zpp = zp;
1162 }
1163 zfs_znode_hold_exit(zfsvfs, zh);
1164 return (err);
1165 }
1166
1167 int
1168 zfs_rezget(znode_t *zp)
1169 {
1170 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1171 dmu_object_info_t doi;
1172 dmu_buf_t *db;
1173 uint64_t obj_num = zp->z_id;
1174 uint64_t mode;
1175 uint64_t links;
1176 sa_bulk_attr_t bulk[11];
1177 int err;
1178 int count = 0;
1179 uint64_t gen;
1180 uint64_t z_uid, z_gid;
1181 uint64_t atime[2], mtime[2], ctime[2], btime[2];
1182 uint64_t projid = ZFS_DEFAULT_PROJID;
1183 znode_hold_t *zh;
1184
1185 /*
1186 * skip ctldir, otherwise they will always get invalidated. This will
1187 * cause funny behaviour for the mounted snapdirs. Especially for
1188 * Linux >= 3.18, d_invalidate will detach the mountpoint and prevent
1189 * anyone automount it again as long as someone is still using the
1190 * detached mount.
1191 */
1192 if (zp->z_is_ctldir)
1193 return (0);
1194
1195 zh = zfs_znode_hold_enter(zfsvfs, obj_num);
1196
1197 mutex_enter(&zp->z_acl_lock);
1198 if (zp->z_acl_cached) {
1199 zfs_acl_free(zp->z_acl_cached);
1200 zp->z_acl_cached = NULL;
1201 }
1202 mutex_exit(&zp->z_acl_lock);
1203
1204 rw_enter(&zp->z_xattr_lock, RW_WRITER);
1205 if (zp->z_xattr_cached) {
1206 nvlist_free(zp->z_xattr_cached);
1207 zp->z_xattr_cached = NULL;
1208 }
1209 rw_exit(&zp->z_xattr_lock);
1210
1211 ASSERT(zp->z_sa_hdl == NULL);
1212 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1213 if (err) {
1214 zfs_znode_hold_exit(zfsvfs, zh);
1215 return (err);
1216 }
1217
1218 dmu_object_info_from_db(db, &doi);
1219 if (doi.doi_bonus_type != DMU_OT_SA &&
1220 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1221 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1222 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1223 sa_buf_rele(db, NULL);
1224 zfs_znode_hold_exit(zfsvfs, zh);
1225 return (SET_ERROR(EINVAL));
1226 }
1227
1228 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1229
1230 /* reload cached values */
1231 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1232 &gen, sizeof (gen));
1233 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1234 &zp->z_size, sizeof (zp->z_size));
1235 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1236 &links, sizeof (links));
1237 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1238 &zp->z_pflags, sizeof (zp->z_pflags));
1239 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1240 &z_uid, sizeof (z_uid));
1241 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1242 &z_gid, sizeof (z_gid));
1243 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1244 &mode, sizeof (mode));
1245 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1246 &atime, 16);
1247 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
1248 &mtime, 16);
1249 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1250 &ctime, 16);
1251 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &btime, 16);
1252
1253 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1254 zfs_znode_dmu_fini(zp);
1255 zfs_znode_hold_exit(zfsvfs, zh);
1256 return (SET_ERROR(EIO));
1257 }
1258
1259 if (dmu_objset_projectquota_enabled(zfsvfs->z_os)) {
1260 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs),
1261 &projid, 8);
1262 if (err != 0 && err != ENOENT) {
1263 zfs_znode_dmu_fini(zp);
1264 zfs_znode_hold_exit(zfsvfs, zh);
1265 return (SET_ERROR(err));
1266 }
1267 }
1268
1269 zp->z_projid = projid;
1270 zp->z_mode = ZTOI(zp)->i_mode = mode;
1271 zfs_uid_write(ZTOI(zp), z_uid);
1272 zfs_gid_write(ZTOI(zp), z_gid);
1273
1274 ZFS_TIME_DECODE(&ZTOI(zp)->i_atime, atime);
1275 ZFS_TIME_DECODE(&ZTOI(zp)->i_mtime, mtime);
1276 ZFS_TIME_DECODE(&ZTOI(zp)->i_ctime, ctime);
1277 ZFS_TIME_DECODE(&zp->z_btime, btime);
1278
1279 if ((uint32_t)gen != ZTOI(zp)->i_generation) {
1280 zfs_znode_dmu_fini(zp);
1281 zfs_znode_hold_exit(zfsvfs, zh);
1282 return (SET_ERROR(EIO));
1283 }
1284
1285 set_nlink(ZTOI(zp), (uint32_t)links);
1286 zfs_set_inode_flags(zp, ZTOI(zp));
1287
1288 zp->z_blksz = doi.doi_data_block_size;
1289 zp->z_atime_dirty = B_FALSE;
1290 zfs_znode_update_vfs(zp);
1291
1292 /*
1293 * If the file has zero links, then it has been unlinked on the send
1294 * side and it must be in the received unlinked set.
1295 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1296 * stale data and to prevent automatic removal of the file in
1297 * zfs_zinactive(). The file will be removed either when it is removed
1298 * on the send side and the next incremental stream is received or
1299 * when the unlinked set gets processed.
1300 */
1301 zp->z_unlinked = (ZTOI(zp)->i_nlink == 0);
1302 if (zp->z_unlinked)
1303 zfs_znode_dmu_fini(zp);
1304
1305 zfs_znode_hold_exit(zfsvfs, zh);
1306
1307 return (0);
1308 }
1309
1310 void
1311 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1312 {
1313 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1314 objset_t *os = zfsvfs->z_os;
1315 uint64_t obj = zp->z_id;
1316 uint64_t acl_obj = zfs_external_acl(zp);
1317 znode_hold_t *zh;
1318
1319 zh = zfs_znode_hold_enter(zfsvfs, obj);
1320 if (acl_obj) {
1321 VERIFY(!zp->z_is_sa);
1322 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1323 }
1324 VERIFY(0 == dmu_object_free(os, obj, tx));
1325 zfs_znode_dmu_fini(zp);
1326 zfs_znode_hold_exit(zfsvfs, zh);
1327 }
1328
1329 void
1330 zfs_zinactive(znode_t *zp)
1331 {
1332 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1333 uint64_t z_id = zp->z_id;
1334 znode_hold_t *zh;
1335
1336 ASSERT(zp->z_sa_hdl);
1337
1338 /*
1339 * Don't allow a zfs_zget() while were trying to release this znode.
1340 */
1341 zh = zfs_znode_hold_enter(zfsvfs, z_id);
1342
1343 mutex_enter(&zp->z_lock);
1344
1345 /*
1346 * If this was the last reference to a file with no links, remove
1347 * the file from the file system unless the file system is mounted
1348 * read-only. That can happen, for example, if the file system was
1349 * originally read-write, the file was opened, then unlinked and
1350 * the file system was made read-only before the file was finally
1351 * closed. The file will remain in the unlinked set.
1352 */
1353 if (zp->z_unlinked) {
1354 ASSERT(!zfsvfs->z_issnap);
1355 if (!zfs_is_readonly(zfsvfs) && !zfs_unlink_suspend_progress) {
1356 mutex_exit(&zp->z_lock);
1357 zfs_znode_hold_exit(zfsvfs, zh);
1358 zfs_rmnode(zp);
1359 return;
1360 }
1361 }
1362
1363 mutex_exit(&zp->z_lock);
1364 zfs_znode_dmu_fini(zp);
1365
1366 zfs_znode_hold_exit(zfsvfs, zh);
1367 }
1368
1369 #if defined(HAVE_INODE_TIMESPEC64_TIMES)
1370 #define zfs_compare_timespec timespec64_compare
1371 #else
1372 #define zfs_compare_timespec timespec_compare
1373 #endif
1374
1375 /*
1376 * Determine whether the znode's atime must be updated. The logic mostly
1377 * duplicates the Linux kernel's relatime_need_update() functionality.
1378 * This function is only called if the underlying filesystem actually has
1379 * atime updates enabled.
1380 */
1381 boolean_t
1382 zfs_relatime_need_update(const struct inode *ip)
1383 {
1384 inode_timespec_t now;
1385
1386 gethrestime(&now);
1387 /*
1388 * In relatime mode, only update the atime if the previous atime
1389 * is earlier than either the ctime or mtime or if at least a day
1390 * has passed since the last update of atime.
1391 */
1392 if (zfs_compare_timespec(&ip->i_mtime, &ip->i_atime) >= 0)
1393 return (B_TRUE);
1394
1395 if (zfs_compare_timespec(&ip->i_ctime, &ip->i_atime) >= 0)
1396 return (B_TRUE);
1397
1398 if ((hrtime_t)now.tv_sec - (hrtime_t)ip->i_atime.tv_sec >= 24*60*60)
1399 return (B_TRUE);
1400
1401 return (B_FALSE);
1402 }
1403
1404 /*
1405 * Prepare to update znode time stamps.
1406 *
1407 * IN: zp - znode requiring timestamp update
1408 * flag - ATTR_MTIME, ATTR_CTIME flags
1409 *
1410 * OUT: zp - z_seq
1411 * mtime - new mtime
1412 * ctime - new ctime
1413 *
1414 * Note: We don't update atime here, because we rely on Linux VFS to do
1415 * atime updating.
1416 */
1417 void
1418 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1419 uint64_t ctime[2])
1420 {
1421 inode_timespec_t now;
1422
1423 gethrestime(&now);
1424
1425 zp->z_seq++;
1426
1427 if (flag & ATTR_MTIME) {
1428 ZFS_TIME_ENCODE(&now, mtime);
1429 ZFS_TIME_DECODE(&(ZTOI(zp)->i_mtime), mtime);
1430 if (ZTOZSB(zp)->z_use_fuids) {
1431 zp->z_pflags |= (ZFS_ARCHIVE |
1432 ZFS_AV_MODIFIED);
1433 }
1434 }
1435
1436 if (flag & ATTR_CTIME) {
1437 ZFS_TIME_ENCODE(&now, ctime);
1438 ZFS_TIME_DECODE(&(ZTOI(zp)->i_ctime), ctime);
1439 if (ZTOZSB(zp)->z_use_fuids)
1440 zp->z_pflags |= ZFS_ARCHIVE;
1441 }
1442 }
1443
1444 /*
1445 * Grow the block size for a file.
1446 *
1447 * IN: zp - znode of file to free data in.
1448 * size - requested block size
1449 * tx - open transaction.
1450 *
1451 * NOTE: this function assumes that the znode is write locked.
1452 */
1453 void
1454 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1455 {
1456 int error;
1457 u_longlong_t dummy;
1458
1459 if (size <= zp->z_blksz)
1460 return;
1461 /*
1462 * If the file size is already greater than the current blocksize,
1463 * we will not grow. If there is more than one block in a file,
1464 * the blocksize cannot change.
1465 */
1466 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1467 return;
1468
1469 error = dmu_object_set_blocksize(ZTOZSB(zp)->z_os, zp->z_id,
1470 size, 0, tx);
1471
1472 if (error == ENOTSUP)
1473 return;
1474 ASSERT0(error);
1475
1476 /* What blocksize did we actually get? */
1477 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1478 }
1479
1480 /*
1481 * Increase the file length
1482 *
1483 * IN: zp - znode of file to free data in.
1484 * end - new end-of-file
1485 *
1486 * RETURN: 0 on success, error code on failure
1487 */
1488 static int
1489 zfs_extend(znode_t *zp, uint64_t end)
1490 {
1491 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1492 dmu_tx_t *tx;
1493 zfs_locked_range_t *lr;
1494 uint64_t newblksz;
1495 int error;
1496
1497 /*
1498 * We will change zp_size, lock the whole file.
1499 */
1500 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1501
1502 /*
1503 * Nothing to do if file already at desired length.
1504 */
1505 if (end <= zp->z_size) {
1506 zfs_rangelock_exit(lr);
1507 return (0);
1508 }
1509 tx = dmu_tx_create(zfsvfs->z_os);
1510 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1511 zfs_sa_upgrade_txholds(tx, zp);
1512 if (end > zp->z_blksz &&
1513 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1514 /*
1515 * We are growing the file past the current block size.
1516 */
1517 if (zp->z_blksz > ZTOZSB(zp)->z_max_blksz) {
1518 /*
1519 * File's blocksize is already larger than the
1520 * "recordsize" property. Only let it grow to
1521 * the next power of 2.
1522 */
1523 ASSERT(!ISP2(zp->z_blksz));
1524 newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1525 } else {
1526 newblksz = MIN(end, ZTOZSB(zp)->z_max_blksz);
1527 }
1528 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1529 } else {
1530 newblksz = 0;
1531 }
1532
1533 error = dmu_tx_assign(tx, TXG_WAIT);
1534 if (error) {
1535 dmu_tx_abort(tx);
1536 zfs_rangelock_exit(lr);
1537 return (error);
1538 }
1539
1540 if (newblksz)
1541 zfs_grow_blocksize(zp, newblksz, tx);
1542
1543 zp->z_size = end;
1544
1545 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(ZTOZSB(zp)),
1546 &zp->z_size, sizeof (zp->z_size), tx));
1547
1548 zfs_rangelock_exit(lr);
1549
1550 dmu_tx_commit(tx);
1551
1552 return (0);
1553 }
1554
1555 /*
1556 * zfs_zero_partial_page - Modeled after update_pages() but
1557 * with different arguments and semantics for use by zfs_freesp().
1558 *
1559 * Zeroes a piece of a single page cache entry for zp at offset
1560 * start and length len.
1561 *
1562 * Caller must acquire a range lock on the file for the region
1563 * being zeroed in order that the ARC and page cache stay in sync.
1564 */
1565 static void
1566 zfs_zero_partial_page(znode_t *zp, uint64_t start, uint64_t len)
1567 {
1568 struct address_space *mp = ZTOI(zp)->i_mapping;
1569 struct page *pp;
1570 int64_t off;
1571 void *pb;
1572
1573 ASSERT((start & PAGE_MASK) == ((start + len - 1) & PAGE_MASK));
1574
1575 off = start & (PAGE_SIZE - 1);
1576 start &= PAGE_MASK;
1577
1578 pp = find_lock_page(mp, start >> PAGE_SHIFT);
1579 if (pp) {
1580 if (mapping_writably_mapped(mp))
1581 flush_dcache_page(pp);
1582
1583 pb = kmap(pp);
1584 memset(pb + off, 0, len);
1585 kunmap(pp);
1586
1587 if (mapping_writably_mapped(mp))
1588 flush_dcache_page(pp);
1589
1590 mark_page_accessed(pp);
1591 SetPageUptodate(pp);
1592 ClearPageError(pp);
1593 unlock_page(pp);
1594 put_page(pp);
1595 }
1596 }
1597
1598 /*
1599 * Free space in a file.
1600 *
1601 * IN: zp - znode of file to free data in.
1602 * off - start of section to free.
1603 * len - length of section to free.
1604 *
1605 * RETURN: 0 on success, error code on failure
1606 */
1607 static int
1608 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1609 {
1610 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1611 zfs_locked_range_t *lr;
1612 int error;
1613
1614 /*
1615 * Lock the range being freed.
1616 */
1617 lr = zfs_rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER);
1618
1619 /*
1620 * Nothing to do if file already at desired length.
1621 */
1622 if (off >= zp->z_size) {
1623 zfs_rangelock_exit(lr);
1624 return (0);
1625 }
1626
1627 if (off + len > zp->z_size)
1628 len = zp->z_size - off;
1629
1630 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1631
1632 /*
1633 * Zero partial page cache entries. This must be done under a
1634 * range lock in order to keep the ARC and page cache in sync.
1635 */
1636 if (zp->z_is_mapped) {
1637 loff_t first_page, last_page, page_len;
1638 loff_t first_page_offset, last_page_offset;
1639
1640 /* first possible full page in hole */
1641 first_page = (off + PAGE_SIZE - 1) >> PAGE_SHIFT;
1642 /* last page of hole */
1643 last_page = (off + len) >> PAGE_SHIFT;
1644
1645 /* offset of first_page */
1646 first_page_offset = first_page << PAGE_SHIFT;
1647 /* offset of last_page */
1648 last_page_offset = last_page << PAGE_SHIFT;
1649
1650 /* truncate whole pages */
1651 if (last_page_offset > first_page_offset) {
1652 truncate_inode_pages_range(ZTOI(zp)->i_mapping,
1653 first_page_offset, last_page_offset - 1);
1654 }
1655
1656 /* truncate sub-page ranges */
1657 if (first_page > last_page) {
1658 /* entire punched area within a single page */
1659 zfs_zero_partial_page(zp, off, len);
1660 } else {
1661 /* beginning of punched area at the end of a page */
1662 page_len = first_page_offset - off;
1663 if (page_len > 0)
1664 zfs_zero_partial_page(zp, off, page_len);
1665
1666 /* end of punched area at the beginning of a page */
1667 page_len = off + len - last_page_offset;
1668 if (page_len > 0)
1669 zfs_zero_partial_page(zp, last_page_offset,
1670 page_len);
1671 }
1672 }
1673 zfs_rangelock_exit(lr);
1674
1675 return (error);
1676 }
1677
1678 /*
1679 * Truncate a file
1680 *
1681 * IN: zp - znode of file to free data in.
1682 * end - new end-of-file.
1683 *
1684 * RETURN: 0 on success, error code on failure
1685 */
1686 static int
1687 zfs_trunc(znode_t *zp, uint64_t end)
1688 {
1689 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1690 dmu_tx_t *tx;
1691 zfs_locked_range_t *lr;
1692 int error;
1693 sa_bulk_attr_t bulk[2];
1694 int count = 0;
1695
1696 /*
1697 * We will change zp_size, lock the whole file.
1698 */
1699 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1700
1701 /*
1702 * Nothing to do if file already at desired length.
1703 */
1704 if (end >= zp->z_size) {
1705 zfs_rangelock_exit(lr);
1706 return (0);
1707 }
1708
1709 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1710 DMU_OBJECT_END);
1711 if (error) {
1712 zfs_rangelock_exit(lr);
1713 return (error);
1714 }
1715 tx = dmu_tx_create(zfsvfs->z_os);
1716 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1717 zfs_sa_upgrade_txholds(tx, zp);
1718 dmu_tx_mark_netfree(tx);
1719 error = dmu_tx_assign(tx, TXG_WAIT);
1720 if (error) {
1721 dmu_tx_abort(tx);
1722 zfs_rangelock_exit(lr);
1723 return (error);
1724 }
1725
1726 zp->z_size = end;
1727 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1728 NULL, &zp->z_size, sizeof (zp->z_size));
1729
1730 if (end == 0) {
1731 zp->z_pflags &= ~ZFS_SPARSE;
1732 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1733 NULL, &zp->z_pflags, 8);
1734 }
1735 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1736
1737 dmu_tx_commit(tx);
1738 zfs_rangelock_exit(lr);
1739
1740 return (0);
1741 }
1742
1743 /*
1744 * Free space in a file
1745 *
1746 * IN: zp - znode of file to free data in.
1747 * off - start of range
1748 * len - end of range (0 => EOF)
1749 * flag - current file open mode flags.
1750 * log - TRUE if this action should be logged
1751 *
1752 * RETURN: 0 on success, error code on failure
1753 */
1754 int
1755 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1756 {
1757 dmu_tx_t *tx;
1758 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1759 zilog_t *zilog = zfsvfs->z_log;
1760 uint64_t mode;
1761 uint64_t mtime[2], ctime[2];
1762 sa_bulk_attr_t bulk[3];
1763 int count = 0;
1764 int error;
1765
1766 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1767 sizeof (mode))) != 0)
1768 return (error);
1769
1770 if (off > zp->z_size) {
1771 error = zfs_extend(zp, off+len);
1772 if (error == 0 && log)
1773 goto log;
1774 goto out;
1775 }
1776
1777 if (len == 0) {
1778 error = zfs_trunc(zp, off);
1779 } else {
1780 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1781 off + len > zp->z_size)
1782 error = zfs_extend(zp, off+len);
1783 }
1784 if (error || !log)
1785 goto out;
1786 log:
1787 tx = dmu_tx_create(zfsvfs->z_os);
1788 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1789 zfs_sa_upgrade_txholds(tx, zp);
1790 error = dmu_tx_assign(tx, TXG_WAIT);
1791 if (error) {
1792 dmu_tx_abort(tx);
1793 goto out;
1794 }
1795
1796 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1797 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1798 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1799 NULL, &zp->z_pflags, 8);
1800 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
1801 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1802 ASSERT(error == 0);
1803
1804 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1805
1806 dmu_tx_commit(tx);
1807
1808 zfs_znode_update_vfs(zp);
1809 error = 0;
1810
1811 out:
1812 /*
1813 * Truncate the page cache - for file truncate operations, use
1814 * the purpose-built API for truncations. For punching operations,
1815 * the truncation is handled under a range lock in zfs_free_range.
1816 */
1817 if (len == 0)
1818 truncate_setsize(ZTOI(zp), off);
1819 return (error);
1820 }
1821
1822 void
1823 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1824 {
1825 struct super_block *sb;
1826 zfsvfs_t *zfsvfs;
1827 uint64_t moid, obj, sa_obj, version;
1828 uint64_t sense = ZFS_CASE_SENSITIVE;
1829 uint64_t norm = 0;
1830 nvpair_t *elem;
1831 int size;
1832 int error;
1833 int i;
1834 znode_t *rootzp = NULL;
1835 vattr_t vattr;
1836 znode_t *zp;
1837 zfs_acl_ids_t acl_ids;
1838
1839 /*
1840 * First attempt to create master node.
1841 */
1842 /*
1843 * In an empty objset, there are no blocks to read and thus
1844 * there can be no i/o errors (which we assert below).
1845 */
1846 moid = MASTER_NODE_OBJ;
1847 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1848 DMU_OT_NONE, 0, tx);
1849 ASSERT(error == 0);
1850
1851 /*
1852 * Set starting attributes.
1853 */
1854 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1855 elem = NULL;
1856 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1857 /* For the moment we expect all zpl props to be uint64_ts */
1858 uint64_t val;
1859 char *name;
1860
1861 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1862 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1863 name = nvpair_name(elem);
1864 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1865 if (val < version)
1866 version = val;
1867 } else {
1868 error = zap_update(os, moid, name, 8, 1, &val, tx);
1869 }
1870 ASSERT(error == 0);
1871 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1872 norm = val;
1873 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1874 sense = val;
1875 }
1876 ASSERT(version != 0);
1877 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1878
1879 /*
1880 * Create zap object used for SA attribute registration
1881 */
1882
1883 if (version >= ZPL_VERSION_SA) {
1884 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1885 DMU_OT_NONE, 0, tx);
1886 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1887 ASSERT(error == 0);
1888 } else {
1889 sa_obj = 0;
1890 }
1891 /*
1892 * Create a delete queue.
1893 */
1894 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1895
1896 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1897 ASSERT(error == 0);
1898
1899 /*
1900 * Create root znode. Create minimal znode/inode/zfsvfs/sb
1901 * to allow zfs_mknode to work.
1902 */
1903 vattr.va_mask = ATTR_MODE|ATTR_UID|ATTR_GID;
1904 vattr.va_mode = S_IFDIR|0755;
1905 vattr.va_uid = crgetuid(cr);
1906 vattr.va_gid = crgetgid(cr);
1907
1908 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1909 rootzp->z_unlinked = B_FALSE;
1910 rootzp->z_atime_dirty = B_FALSE;
1911 rootzp->z_is_sa = USE_SA(version, os);
1912 rootzp->z_pflags = 0;
1913
1914 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1915 zfsvfs->z_os = os;
1916 zfsvfs->z_parent = zfsvfs;
1917 zfsvfs->z_version = version;
1918 zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1919 zfsvfs->z_use_sa = USE_SA(version, os);
1920 zfsvfs->z_norm = norm;
1921
1922 sb = kmem_zalloc(sizeof (struct super_block), KM_SLEEP);
1923 sb->s_fs_info = zfsvfs;
1924
1925 ZTOI(rootzp)->i_sb = sb;
1926
1927 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1928 &zfsvfs->z_attr_table);
1929
1930 ASSERT(error == 0);
1931
1932 /*
1933 * Fold case on file systems that are always or sometimes case
1934 * insensitive.
1935 */
1936 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1937 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1938
1939 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1940 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1941 offsetof(znode_t, z_link_node));
1942
1943 size = MIN(1 << (highbit64(zfs_object_mutex_size)-1), ZFS_OBJ_MTX_MAX);
1944 zfsvfs->z_hold_size = size;
1945 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
1946 KM_SLEEP);
1947 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
1948 for (i = 0; i != size; i++) {
1949 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
1950 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
1951 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
1952 }
1953
1954 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1955 cr, NULL, &acl_ids));
1956 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1957 ASSERT3P(zp, ==, rootzp);
1958 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1959 ASSERT(error == 0);
1960 zfs_acl_ids_free(&acl_ids);
1961
1962 atomic_set(&ZTOI(rootzp)->i_count, 0);
1963 sa_handle_destroy(rootzp->z_sa_hdl);
1964 kmem_cache_free(znode_cache, rootzp);
1965
1966 for (i = 0; i != size; i++) {
1967 avl_destroy(&zfsvfs->z_hold_trees[i]);
1968 mutex_destroy(&zfsvfs->z_hold_locks[i]);
1969 }
1970
1971 mutex_destroy(&zfsvfs->z_znodes_lock);
1972
1973 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
1974 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
1975 kmem_free(sb, sizeof (struct super_block));
1976 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1977 }
1978 #endif /* _KERNEL */
1979
1980 static int
1981 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1982 {
1983 uint64_t sa_obj = 0;
1984 int error;
1985
1986 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1987 if (error != 0 && error != ENOENT)
1988 return (error);
1989
1990 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1991 return (error);
1992 }
1993
1994 static int
1995 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1996 dmu_buf_t **db, void *tag)
1997 {
1998 dmu_object_info_t doi;
1999 int error;
2000
2001 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
2002 return (error);
2003
2004 dmu_object_info_from_db(*db, &doi);
2005 if ((doi.doi_bonus_type != DMU_OT_SA &&
2006 doi.doi_bonus_type != DMU_OT_ZNODE) ||
2007 (doi.doi_bonus_type == DMU_OT_ZNODE &&
2008 doi.doi_bonus_size < sizeof (znode_phys_t))) {
2009 sa_buf_rele(*db, tag);
2010 return (SET_ERROR(ENOTSUP));
2011 }
2012
2013 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
2014 if (error != 0) {
2015 sa_buf_rele(*db, tag);
2016 return (error);
2017 }
2018
2019 return (0);
2020 }
2021
2022 static void
2023 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
2024 {
2025 sa_handle_destroy(hdl);
2026 sa_buf_rele(db, tag);
2027 }
2028
2029 /*
2030 * Given an object number, return its parent object number and whether
2031 * or not the object is an extended attribute directory.
2032 */
2033 static int
2034 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
2035 uint64_t *pobjp, int *is_xattrdir)
2036 {
2037 uint64_t parent;
2038 uint64_t pflags;
2039 uint64_t mode;
2040 uint64_t parent_mode;
2041 sa_bulk_attr_t bulk[3];
2042 sa_handle_t *sa_hdl;
2043 dmu_buf_t *sa_db;
2044 int count = 0;
2045 int error;
2046
2047 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
2048 &parent, sizeof (parent));
2049 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
2050 &pflags, sizeof (pflags));
2051 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2052 &mode, sizeof (mode));
2053
2054 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
2055 return (error);
2056
2057 /*
2058 * When a link is removed its parent pointer is not changed and will
2059 * be invalid. There are two cases where a link is removed but the
2060 * file stays around, when it goes to the delete queue and when there
2061 * are additional links.
2062 */
2063 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
2064 if (error != 0)
2065 return (error);
2066
2067 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
2068 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2069 if (error != 0)
2070 return (error);
2071
2072 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
2073
2074 /*
2075 * Extended attributes can be applied to files, directories, etc.
2076 * Otherwise the parent must be a directory.
2077 */
2078 if (!*is_xattrdir && !S_ISDIR(parent_mode))
2079 return (SET_ERROR(EINVAL));
2080
2081 *pobjp = parent;
2082
2083 return (0);
2084 }
2085
2086 /*
2087 * Given an object number, return some zpl level statistics
2088 */
2089 static int
2090 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2091 zfs_stat_t *sb)
2092 {
2093 sa_bulk_attr_t bulk[4];
2094 int count = 0;
2095
2096 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2097 &sb->zs_mode, sizeof (sb->zs_mode));
2098 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2099 &sb->zs_gen, sizeof (sb->zs_gen));
2100 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2101 &sb->zs_links, sizeof (sb->zs_links));
2102 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2103 &sb->zs_ctime, sizeof (sb->zs_ctime));
2104
2105 return (sa_bulk_lookup(hdl, bulk, count));
2106 }
2107
2108 static int
2109 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2110 sa_attr_type_t *sa_table, char *buf, int len)
2111 {
2112 sa_handle_t *sa_hdl;
2113 sa_handle_t *prevhdl = NULL;
2114 dmu_buf_t *prevdb = NULL;
2115 dmu_buf_t *sa_db = NULL;
2116 char *path = buf + len - 1;
2117 int error;
2118
2119 *path = '\0';
2120 sa_hdl = hdl;
2121
2122 uint64_t deleteq_obj;
2123 VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ,
2124 ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj));
2125 error = zap_lookup_int(osp, deleteq_obj, obj);
2126 if (error == 0) {
2127 return (ESTALE);
2128 } else if (error != ENOENT) {
2129 return (error);
2130 }
2131 error = 0;
2132
2133 for (;;) {
2134 uint64_t pobj = 0;
2135 char component[MAXNAMELEN + 2];
2136 size_t complen;
2137 int is_xattrdir = 0;
2138
2139 if (prevdb) {
2140 ASSERT(prevhdl != NULL);
2141 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2142 }
2143
2144 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2145 &is_xattrdir)) != 0)
2146 break;
2147
2148 if (pobj == obj) {
2149 if (path[0] != '/')
2150 *--path = '/';
2151 break;
2152 }
2153
2154 component[0] = '/';
2155 if (is_xattrdir) {
2156 strcpy(component + 1, "<xattrdir>");
2157 } else {
2158 error = zap_value_search(osp, pobj, obj,
2159 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2160 if (error != 0)
2161 break;
2162 }
2163
2164 complen = strlen(component);
2165 path -= complen;
2166 ASSERT(path >= buf);
2167 memcpy(path, component, complen);
2168 obj = pobj;
2169
2170 if (sa_hdl != hdl) {
2171 prevhdl = sa_hdl;
2172 prevdb = sa_db;
2173 }
2174 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2175 if (error != 0) {
2176 sa_hdl = prevhdl;
2177 sa_db = prevdb;
2178 break;
2179 }
2180 }
2181
2182 if (sa_hdl != NULL && sa_hdl != hdl) {
2183 ASSERT(sa_db != NULL);
2184 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2185 }
2186
2187 if (error == 0)
2188 (void) memmove(buf, path, buf + len - path);
2189
2190 return (error);
2191 }
2192
2193 int
2194 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2195 {
2196 sa_attr_type_t *sa_table;
2197 sa_handle_t *hdl;
2198 dmu_buf_t *db;
2199 int error;
2200
2201 error = zfs_sa_setup(osp, &sa_table);
2202 if (error != 0)
2203 return (error);
2204
2205 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2206 if (error != 0)
2207 return (error);
2208
2209 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2210
2211 zfs_release_sa_handle(hdl, db, FTAG);
2212 return (error);
2213 }
2214
2215 int
2216 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2217 char *buf, int len)
2218 {
2219 char *path = buf + len - 1;
2220 sa_attr_type_t *sa_table;
2221 sa_handle_t *hdl;
2222 dmu_buf_t *db;
2223 int error;
2224
2225 *path = '\0';
2226
2227 error = zfs_sa_setup(osp, &sa_table);
2228 if (error != 0)
2229 return (error);
2230
2231 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2232 if (error != 0)
2233 return (error);
2234
2235 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2236 if (error != 0) {
2237 zfs_release_sa_handle(hdl, db, FTAG);
2238 return (error);
2239 }
2240
2241 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2242
2243 zfs_release_sa_handle(hdl, db, FTAG);
2244 return (error);
2245 }
2246
2247 #if defined(_KERNEL)
2248 EXPORT_SYMBOL(zfs_create_fs);
2249 EXPORT_SYMBOL(zfs_obj_to_path);
2250
2251 /* CSTYLED */
2252 module_param(zfs_object_mutex_size, uint, 0644);
2253 MODULE_PARM_DESC(zfs_object_mutex_size, "Size of znode hold array");
2254 module_param(zfs_unlink_suspend_progress, int, 0644);
2255 MODULE_PARM_DESC(zfs_unlink_suspend_progress, "Set to prevent async unlinks "
2256 "(debug - leaks space into the unlinked set)");
2257 #endif
OpenZFS on Linux and FreeBSD