1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
24 */
25
26 /* Portions Copyright 2007 Jeremy Teo */
27
28 #ifdef _KERNEL
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/time.h>
32 #include <sys/sysmacros.h>
33 #include <sys/mntent.h>
34 #include <sys/u8_textprep.h>
35 #include <sys/dsl_dataset.h>
36 #include <sys/vfs.h>
37 #include <sys/vnode.h>
38 #include <sys/file.h>
39 #include <sys/kmem.h>
40 #include <sys/errno.h>
41 #include <sys/atomic.h>
42 #include <sys/zfs_dir.h>
43 #include <sys/zfs_acl.h>
44 #include <sys/zfs_ioctl.h>
45 #include <sys/zfs_rlock.h>
46 #include <sys/zfs_fuid.h>
47 #include <sys/zfs_vnops.h>
48 #include <sys/zfs_ctldir.h>
49 #include <sys/dnode.h>
50 #include <sys/fs/zfs.h>
51 #include <sys/zpl.h>
52 #endif /* _KERNEL */
53
54 #include <sys/dmu.h>
55 #include <sys/dmu_objset.h>
56 #include <sys/dmu_tx.h>
57 #include <sys/zfs_refcount.h>
58 #include <sys/stat.h>
59 #include <sys/zap.h>
60 #include <sys/zfs_znode.h>
61 #include <sys/sa.h>
62 #include <sys/zfs_sa.h>
63 #include <sys/zfs_stat.h>
64
65 #include "zfs_prop.h"
66 #include "zfs_comutil.h"
67
68 /*
69 * Functions needed for userland (ie: libzpool) are not put under
70 * #ifdef_KERNEL; the rest of the functions have dependencies
71 * (such as VFS logic) that will not compile easily in userland.
72 */
73 #ifdef _KERNEL
74
75 static kmem_cache_t *znode_cache = NULL;
76 static kmem_cache_t *znode_hold_cache = NULL;
77 unsigned int zfs_object_mutex_size = ZFS_OBJ_MTX_SZ;
78
79 /*
80 * This is used by the test suite so that it can delay znodes from being
81 * freed in order to inspect the unlinked set.
82 */
83 int zfs_unlink_suspend_progress = 0;
84
85 /*
86 * This callback is invoked when acquiring a RL_WRITER or RL_APPEND lock on
87 * z_rangelock. It will modify the offset and length of the lock to reflect
88 * znode-specific information, and convert RL_APPEND to RL_WRITER. This is
89 * called with the rangelock_t's rl_lock held, which avoids races.
90 */
91 static void
zfs_rangelock_cb(zfs_locked_range_t * new,void * arg)92 zfs_rangelock_cb(zfs_locked_range_t *new, void *arg)
93 {
94 znode_t *zp = arg;
95
96 /*
97 * If in append mode, convert to writer and lock starting at the
98 * current end of file.
99 */
100 if (new->lr_type == RL_APPEND) {
101 new->lr_offset = zp->z_size;
102 new->lr_type = RL_WRITER;
103 }
104
105 /*
106 * If we need to grow the block size then lock the whole file range.
107 */
108 uint64_t end_size = MAX(zp->z_size, new->lr_offset + new->lr_length);
109 if (end_size > zp->z_blksz && (!ISP2(zp->z_blksz) ||
110 zp->z_blksz < ZTOZSB(zp)->z_max_blksz)) {
111 new->lr_offset = 0;
112 new->lr_length = UINT64_MAX;
113 }
114 }
115
116 /*ARGSUSED*/
117 static int
zfs_znode_cache_constructor(void * buf,void * arg,int kmflags)118 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
119 {
120 znode_t *zp = buf;
121
122 inode_init_once(ZTOI(zp));
123 list_link_init(&zp->z_link_node);
124
125 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
126 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
127 rw_init(&zp->z_name_lock, NULL, RW_NOLOCKDEP, NULL);
128 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
129 rw_init(&zp->z_xattr_lock, NULL, RW_DEFAULT, NULL);
130
131 zfs_rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp);
132
133 zp->z_dirlocks = NULL;
134 zp->z_acl_cached = NULL;
135 zp->z_xattr_cached = NULL;
136 zp->z_xattr_parent = 0;
137 return (0);
138 }
139
140 /*ARGSUSED*/
141 static void
zfs_znode_cache_destructor(void * buf,void * arg)142 zfs_znode_cache_destructor(void *buf, void *arg)
143 {
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 ASSERT(zp->z_dirlocks == NULL);
155 ASSERT(zp->z_acl_cached == NULL);
156 ASSERT(zp->z_xattr_cached == NULL);
157 }
158
159 static int
zfs_znode_hold_cache_constructor(void * buf,void * arg,int kmflags)160 zfs_znode_hold_cache_constructor(void *buf, void *arg, int kmflags)
161 {
162 znode_hold_t *zh = buf;
163
164 mutex_init(&zh->zh_lock, NULL, MUTEX_DEFAULT, NULL);
165 zfs_refcount_create(&zh->zh_refcount);
166 zh->zh_obj = ZFS_NO_OBJECT;
167
168 return (0);
169 }
170
171 static void
zfs_znode_hold_cache_destructor(void * buf,void * arg)172 zfs_znode_hold_cache_destructor(void *buf, void *arg)
173 {
174 znode_hold_t *zh = buf;
175
176 mutex_destroy(&zh->zh_lock);
177 zfs_refcount_destroy(&zh->zh_refcount);
178 }
179
180 void
zfs_znode_init(void)181 zfs_znode_init(void)
182 {
183 /*
184 * Initialize zcache. The KMC_SLAB hint is used in order that it be
185 * backed by kmalloc() when on the Linux slab in order that any
186 * wait_on_bit() operations on the related inode operate properly.
187 */
188 ASSERT(znode_cache == NULL);
189 znode_cache = kmem_cache_create("zfs_znode_cache",
190 sizeof (znode_t), 0, zfs_znode_cache_constructor,
191 zfs_znode_cache_destructor, NULL, NULL, NULL, KMC_SLAB);
192
193 ASSERT(znode_hold_cache == NULL);
194 znode_hold_cache = kmem_cache_create("zfs_znode_hold_cache",
195 sizeof (znode_hold_t), 0, zfs_znode_hold_cache_constructor,
196 zfs_znode_hold_cache_destructor, NULL, NULL, NULL, 0);
197 }
198
199 void
zfs_znode_fini(void)200 zfs_znode_fini(void)
201 {
202 /*
203 * Cleanup zcache
204 */
205 if (znode_cache)
206 kmem_cache_destroy(znode_cache);
207 znode_cache = NULL;
208
209 if (znode_hold_cache)
210 kmem_cache_destroy(znode_hold_cache);
211 znode_hold_cache = NULL;
212 }
213
214 /*
215 * The zfs_znode_hold_enter() / zfs_znode_hold_exit() functions are used to
216 * serialize access to a znode and its SA buffer while the object is being
217 * created or destroyed. This kind of locking would normally reside in the
218 * znode itself but in this case that's impossible because the znode and SA
219 * buffer may not yet exist. Therefore the locking is handled externally
220 * with an array of mutexes and AVLs trees which contain per-object locks.
221 *
222 * In zfs_znode_hold_enter() a per-object lock is created as needed, inserted
223 * in to the correct AVL tree and finally the per-object lock is held. In
224 * zfs_znode_hold_exit() the process is reversed. The per-object lock is
225 * released, removed from the AVL tree and destroyed if there are no waiters.
226 *
227 * This scheme has two important properties:
228 *
229 * 1) No memory allocations are performed while holding one of the z_hold_locks.
230 * This ensures evict(), which can be called from direct memory reclaim, will
231 * never block waiting on a z_hold_locks which just happens to have hashed
232 * to the same index.
233 *
234 * 2) All locks used to serialize access to an object are per-object and never
235 * shared. This minimizes lock contention without creating a large number
236 * of dedicated locks.
237 *
238 * On the downside it does require znode_lock_t structures to be frequently
239 * allocated and freed. However, because these are backed by a kmem cache
240 * and very short lived this cost is minimal.
241 */
242 int
zfs_znode_hold_compare(const void * a,const void * b)243 zfs_znode_hold_compare(const void *a, const void *b)
244 {
245 const znode_hold_t *zh_a = (const znode_hold_t *)a;
246 const znode_hold_t *zh_b = (const znode_hold_t *)b;
247
248 return (TREE_CMP(zh_a->zh_obj, zh_b->zh_obj));
249 }
250
251 static boolean_t __maybe_unused
zfs_znode_held(zfsvfs_t * zfsvfs,uint64_t obj)252 zfs_znode_held(zfsvfs_t *zfsvfs, uint64_t obj)
253 {
254 znode_hold_t *zh, search;
255 int i = ZFS_OBJ_HASH(zfsvfs, obj);
256 boolean_t held;
257
258 search.zh_obj = obj;
259
260 mutex_enter(&zfsvfs->z_hold_locks[i]);
261 zh = avl_find(&zfsvfs->z_hold_trees[i], &search, NULL);
262 held = (zh && MUTEX_HELD(&zh->zh_lock)) ? B_TRUE : B_FALSE;
263 mutex_exit(&zfsvfs->z_hold_locks[i]);
264
265 return (held);
266 }
267
268 static znode_hold_t *
zfs_znode_hold_enter(zfsvfs_t * zfsvfs,uint64_t obj)269 zfs_znode_hold_enter(zfsvfs_t *zfsvfs, uint64_t obj)
270 {
271 znode_hold_t *zh, *zh_new, search;
272 int i = ZFS_OBJ_HASH(zfsvfs, obj);
273 boolean_t found = B_FALSE;
274
275 zh_new = kmem_cache_alloc(znode_hold_cache, KM_SLEEP);
276 zh_new->zh_obj = obj;
277 search.zh_obj = obj;
278
279 mutex_enter(&zfsvfs->z_hold_locks[i]);
280 zh = avl_find(&zfsvfs->z_hold_trees[i], &search, NULL);
281 if (likely(zh == NULL)) {
282 zh = zh_new;
283 avl_add(&zfsvfs->z_hold_trees[i], zh);
284 } else {
285 ASSERT3U(zh->zh_obj, ==, obj);
286 found = B_TRUE;
287 }
288 zfs_refcount_add(&zh->zh_refcount, NULL);
289 mutex_exit(&zfsvfs->z_hold_locks[i]);
290
291 if (found == B_TRUE)
292 kmem_cache_free(znode_hold_cache, zh_new);
293
294 ASSERT(MUTEX_NOT_HELD(&zh->zh_lock));
295 ASSERT3S(zfs_refcount_count(&zh->zh_refcount), >, 0);
296 mutex_enter(&zh->zh_lock);
297
298 return (zh);
299 }
300
301 static void
zfs_znode_hold_exit(zfsvfs_t * zfsvfs,znode_hold_t * zh)302 zfs_znode_hold_exit(zfsvfs_t *zfsvfs, znode_hold_t *zh)
303 {
304 int i = ZFS_OBJ_HASH(zfsvfs, zh->zh_obj);
305 boolean_t remove = B_FALSE;
306
307 ASSERT(zfs_znode_held(zfsvfs, zh->zh_obj));
308 ASSERT3S(zfs_refcount_count(&zh->zh_refcount), >, 0);
309 mutex_exit(&zh->zh_lock);
310
311 mutex_enter(&zfsvfs->z_hold_locks[i]);
312 if (zfs_refcount_remove(&zh->zh_refcount, NULL) == 0) {
313 avl_remove(&zfsvfs->z_hold_trees[i], zh);
314 remove = B_TRUE;
315 }
316 mutex_exit(&zfsvfs->z_hold_locks[i]);
317
318 if (remove == B_TRUE)
319 kmem_cache_free(znode_hold_cache, zh);
320 }
321
322 dev_t
zfs_cmpldev(uint64_t dev)323 zfs_cmpldev(uint64_t dev)
324 {
325 return (dev);
326 }
327
328 static void
zfs_znode_sa_init(zfsvfs_t * zfsvfs,znode_t * zp,dmu_buf_t * db,dmu_object_type_t obj_type,sa_handle_t * sa_hdl)329 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
330 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
331 {
332 ASSERT(zfs_znode_held(zfsvfs, zp->z_id));
333
334 mutex_enter(&zp->z_lock);
335
336 ASSERT(zp->z_sa_hdl == NULL);
337 ASSERT(zp->z_acl_cached == NULL);
338 if (sa_hdl == NULL) {
339 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
340 SA_HDL_SHARED, &zp->z_sa_hdl));
341 } else {
342 zp->z_sa_hdl = sa_hdl;
343 sa_set_userp(sa_hdl, zp);
344 }
345
346 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
347
348 mutex_exit(&zp->z_lock);
349 }
350
351 void
zfs_znode_dmu_fini(znode_t * zp)352 zfs_znode_dmu_fini(znode_t *zp)
353 {
354 ASSERT(zfs_znode_held(ZTOZSB(zp), zp->z_id) || zp->z_unlinked ||
355 RW_WRITE_HELD(&ZTOZSB(zp)->z_teardown_inactive_lock));
356
357 sa_handle_destroy(zp->z_sa_hdl);
358 zp->z_sa_hdl = NULL;
359 }
360
361 /*
362 * Called by new_inode() to allocate a new inode.
363 */
364 int
zfs_inode_alloc(struct super_block * sb,struct inode ** ip)365 zfs_inode_alloc(struct super_block *sb, struct inode **ip)
366 {
367 znode_t *zp;
368
369 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
370 *ip = ZTOI(zp);
371
372 return (0);
373 }
374
375 /*
376 * Called in multiple places when an inode should be destroyed.
377 */
378 void
zfs_inode_destroy(struct inode * ip)379 zfs_inode_destroy(struct inode *ip)
380 {
381 znode_t *zp = ITOZ(ip);
382 zfsvfs_t *zfsvfs = ZTOZSB(zp);
383
384 mutex_enter(&zfsvfs->z_znodes_lock);
385 if (list_link_active(&zp->z_link_node)) {
386 list_remove(&zfsvfs->z_all_znodes, zp);
387 zfsvfs->z_nr_znodes--;
388 }
389 mutex_exit(&zfsvfs->z_znodes_lock);
390
391 if (zp->z_acl_cached) {
392 zfs_acl_free(zp->z_acl_cached);
393 zp->z_acl_cached = NULL;
394 }
395
396 if (zp->z_xattr_cached) {
397 nvlist_free(zp->z_xattr_cached);
398 zp->z_xattr_cached = NULL;
399 }
400
401 kmem_cache_free(znode_cache, zp);
402 }
403
404 static void
zfs_inode_set_ops(zfsvfs_t * zfsvfs,struct inode * ip)405 zfs_inode_set_ops(zfsvfs_t *zfsvfs, struct inode *ip)
406 {
407 uint64_t rdev = 0;
408
409 switch (ip->i_mode & S_IFMT) {
410 case S_IFREG:
411 ip->i_op = &zpl_inode_operations;
412 ip->i_fop = &zpl_file_operations;
413 ip->i_mapping->a_ops = &zpl_address_space_operations;
414 break;
415
416 case S_IFDIR:
417 ip->i_op = &zpl_dir_inode_operations;
418 ip->i_fop = &zpl_dir_file_operations;
419 ITOZ(ip)->z_zn_prefetch = B_TRUE;
420 break;
421
422 case S_IFLNK:
423 ip->i_op = &zpl_symlink_inode_operations;
424 break;
425
426 /*
427 * rdev is only stored in a SA only for device files.
428 */
429 case S_IFCHR:
430 case S_IFBLK:
431 (void) sa_lookup(ITOZ(ip)->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), &rdev,
432 sizeof (rdev));
433 fallthrough;
434 case S_IFIFO:
435 case S_IFSOCK:
436 init_special_inode(ip, ip->i_mode, rdev);
437 ip->i_op = &zpl_special_inode_operations;
438 break;
439
440 default:
441 zfs_panic_recover("inode %llu has invalid mode: 0x%x\n",
442 (u_longlong_t)ip->i_ino, ip->i_mode);
443
444 /* Assume the inode is a file and attempt to continue */
445 ip->i_mode = S_IFREG | 0644;
446 ip->i_op = &zpl_inode_operations;
447 ip->i_fop = &zpl_file_operations;
448 ip->i_mapping->a_ops = &zpl_address_space_operations;
449 break;
450 }
451 }
452
453 static void
zfs_set_inode_flags(znode_t * zp,struct inode * ip)454 zfs_set_inode_flags(znode_t *zp, struct inode *ip)
455 {
456 /*
457 * Linux and Solaris have different sets of file attributes, so we
458 * restrict this conversion to the intersection of the two.
459 */
460 #ifdef HAVE_INODE_SET_FLAGS
461 unsigned int flags = 0;
462 if (zp->z_pflags & ZFS_IMMUTABLE)
463 flags |= S_IMMUTABLE;
464 if (zp->z_pflags & ZFS_APPENDONLY)
465 flags |= S_APPEND;
466
467 inode_set_flags(ip, flags, S_IMMUTABLE|S_APPEND);
468 #else
469 if (zp->z_pflags & ZFS_IMMUTABLE)
470 ip->i_flags |= S_IMMUTABLE;
471 else
472 ip->i_flags &= ~S_IMMUTABLE;
473
474 if (zp->z_pflags & ZFS_APPENDONLY)
475 ip->i_flags |= S_APPEND;
476 else
477 ip->i_flags &= ~S_APPEND;
478 #endif
479 }
480
481 /*
482 * Update the embedded inode given the znode.
483 */
484 void
zfs_znode_update_vfs(znode_t * zp)485 zfs_znode_update_vfs(znode_t *zp)
486 {
487 zfsvfs_t *zfsvfs;
488 struct inode *ip;
489 uint32_t blksize;
490 u_longlong_t i_blocks;
491
492 ASSERT(zp != NULL);
493 zfsvfs = ZTOZSB(zp);
494 ip = ZTOI(zp);
495
496 /* Skip .zfs control nodes which do not exist on disk. */
497 if (zfsctl_is_node(ip))
498 return;
499
500 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &blksize, &i_blocks);
501
502 spin_lock(&ip->i_lock);
503 ip->i_mode = zp->z_mode;
504 ip->i_blocks = i_blocks;
505 i_size_write(ip, zp->z_size);
506 spin_unlock(&ip->i_lock);
507 }
508
509
510 /*
511 * Construct a znode+inode and initialize.
512 *
513 * This does not do a call to dmu_set_user() that is
514 * up to the caller to do, in case you don't want to
515 * return the znode
516 */
517 static znode_t *
zfs_znode_alloc(zfsvfs_t * zfsvfs,dmu_buf_t * db,int blksz,dmu_object_type_t obj_type,sa_handle_t * hdl)518 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
519 dmu_object_type_t obj_type, sa_handle_t *hdl)
520 {
521 znode_t *zp;
522 struct inode *ip;
523 uint64_t mode;
524 uint64_t parent;
525 uint64_t tmp_gen;
526 uint64_t links;
527 uint64_t z_uid, z_gid;
528 uint64_t atime[2], mtime[2], ctime[2], btime[2];
529 uint64_t projid = ZFS_DEFAULT_PROJID;
530 sa_bulk_attr_t bulk[12];
531 int count = 0;
532
533 ASSERT(zfsvfs != NULL);
534
535 ip = new_inode(zfsvfs->z_sb);
536 if (ip == NULL)
537 return (NULL);
538
539 zp = ITOZ(ip);
540 ASSERT(zp->z_dirlocks == NULL);
541 ASSERT3P(zp->z_acl_cached, ==, NULL);
542 ASSERT3P(zp->z_xattr_cached, ==, NULL);
543 zp->z_unlinked = B_FALSE;
544 zp->z_atime_dirty = B_FALSE;
545 zp->z_is_mapped = B_FALSE;
546 zp->z_is_ctldir = B_FALSE;
547 zp->z_is_stale = B_FALSE;
548 zp->z_suspended = B_FALSE;
549 zp->z_sa_hdl = NULL;
550 zp->z_mapcnt = 0;
551 zp->z_id = db->db_object;
552 zp->z_blksz = blksz;
553 zp->z_seq = 0x7A4653;
554 zp->z_sync_cnt = 0;
555
556 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
557
558 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
559 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &tmp_gen, 8);
560 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
561 &zp->z_size, 8);
562 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
563 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
564 &zp->z_pflags, 8);
565 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
566 &parent, 8);
567 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &z_uid, 8);
568 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &z_gid, 8);
569 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
570 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
571 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
572 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &btime, 16);
573
574 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || tmp_gen == 0 ||
575 (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
576 (zp->z_pflags & ZFS_PROJID) &&
577 sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) {
578 if (hdl == NULL)
579 sa_handle_destroy(zp->z_sa_hdl);
580 zp->z_sa_hdl = NULL;
581 goto error;
582 }
583
584 zp->z_projid = projid;
585 zp->z_mode = ip->i_mode = mode;
586 ip->i_generation = (uint32_t)tmp_gen;
587 ip->i_blkbits = SPA_MINBLOCKSHIFT;
588 set_nlink(ip, (uint32_t)links);
589 zfs_uid_write(ip, z_uid);
590 zfs_gid_write(ip, z_gid);
591 zfs_set_inode_flags(zp, ip);
592
593 /* Cache the xattr parent id */
594 if (zp->z_pflags & ZFS_XATTR)
595 zp->z_xattr_parent = parent;
596
597 ZFS_TIME_DECODE(&ip->i_atime, atime);
598 ZFS_TIME_DECODE(&ip->i_mtime, mtime);
599 ZFS_TIME_DECODE(&ip->i_ctime, ctime);
600 ZFS_TIME_DECODE(&zp->z_btime, btime);
601
602 ip->i_ino = zp->z_id;
603 zfs_znode_update_vfs(zp);
604 zfs_inode_set_ops(zfsvfs, ip);
605
606 /*
607 * The only way insert_inode_locked() can fail is if the ip->i_ino
608 * number is already hashed for this super block. This can never
609 * happen because the inode numbers map 1:1 with the object numbers.
610 *
611 * Exceptions include rolling back a mounted file system, either
612 * from the zfs rollback or zfs recv command.
613 *
614 * Active inodes are unhashed during the rollback, but since zrele
615 * can happen asynchronously, we can't guarantee they've been
616 * unhashed. This can cause hash collisions in unlinked drain
617 * processing so do not hash unlinked znodes.
618 */
619 if (links > 0)
620 VERIFY3S(insert_inode_locked(ip), ==, 0);
621
622 mutex_enter(&zfsvfs->z_znodes_lock);
623 list_insert_tail(&zfsvfs->z_all_znodes, zp);
624 zfsvfs->z_nr_znodes++;
625 mutex_exit(&zfsvfs->z_znodes_lock);
626
627 if (links > 0)
628 unlock_new_inode(ip);
629 return (zp);
630
631 error:
632 iput(ip);
633 return (NULL);
634 }
635
636 /*
637 * Safely mark an inode dirty. Inodes which are part of a read-only
638 * file system or snapshot may not be dirtied.
639 */
640 void
zfs_mark_inode_dirty(struct inode * ip)641 zfs_mark_inode_dirty(struct inode *ip)
642 {
643 zfsvfs_t *zfsvfs = ITOZSB(ip);
644
645 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
646 return;
647
648 mark_inode_dirty(ip);
649 }
650
651 static uint64_t empty_xattr;
652 static uint64_t pad[4];
653 static zfs_acl_phys_t acl_phys;
654 /*
655 * Create a new DMU object to hold a zfs znode.
656 *
657 * IN: dzp - parent directory for new znode
658 * vap - file attributes for new znode
659 * tx - dmu transaction id for zap operations
660 * cr - credentials of caller
661 * flag - flags:
662 * IS_ROOT_NODE - new object will be root
663 * IS_TMPFILE - new object is of O_TMPFILE
664 * IS_XATTR - new object is an attribute
665 * acl_ids - ACL related attributes
666 *
667 * OUT: zpp - allocated znode (set to dzp if IS_ROOT_NODE)
668 *
669 */
670 void
zfs_mknode(znode_t * dzp,vattr_t * vap,dmu_tx_t * tx,cred_t * cr,uint_t flag,znode_t ** zpp,zfs_acl_ids_t * acl_ids)671 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
672 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
673 {
674 uint64_t crtime[2], atime[2], mtime[2], ctime[2];
675 uint64_t mode, size, links, parent, pflags;
676 uint64_t projid = ZFS_DEFAULT_PROJID;
677 uint64_t rdev = 0;
678 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
679 dmu_buf_t *db;
680 inode_timespec_t now;
681 uint64_t gen, obj;
682 int bonuslen;
683 int dnodesize;
684 sa_handle_t *sa_hdl;
685 dmu_object_type_t obj_type;
686 sa_bulk_attr_t *sa_attrs;
687 int cnt = 0;
688 zfs_acl_locator_cb_t locate = { 0 };
689 znode_hold_t *zh;
690
691 if (zfsvfs->z_replay) {
692 obj = vap->va_nodeid;
693 now = vap->va_ctime; /* see zfs_replay_create() */
694 gen = vap->va_nblocks; /* ditto */
695 dnodesize = vap->va_fsid; /* ditto */
696 } else {
697 obj = 0;
698 gethrestime(&now);
699 gen = dmu_tx_get_txg(tx);
700 dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
701 }
702
703 if (dnodesize == 0)
704 dnodesize = DNODE_MIN_SIZE;
705
706 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
707
708 bonuslen = (obj_type == DMU_OT_SA) ?
709 DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
710
711 /*
712 * Create a new DMU object.
713 */
714 /*
715 * There's currently no mechanism for pre-reading the blocks that will
716 * be needed to allocate a new object, so we accept the small chance
717 * that there will be an i/o error and we will fail one of the
718 * assertions below.
719 */
720 if (S_ISDIR(vap->va_mode)) {
721 if (zfsvfs->z_replay) {
722 VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj,
723 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
724 obj_type, bonuslen, dnodesize, tx));
725 } else {
726 obj = zap_create_norm_dnsize(zfsvfs->z_os,
727 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
728 obj_type, bonuslen, dnodesize, tx);
729 }
730 } else {
731 if (zfsvfs->z_replay) {
732 VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj,
733 DMU_OT_PLAIN_FILE_CONTENTS, 0,
734 obj_type, bonuslen, dnodesize, tx));
735 } else {
736 obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
737 DMU_OT_PLAIN_FILE_CONTENTS, 0,
738 obj_type, bonuslen, dnodesize, tx);
739 }
740 }
741
742 zh = zfs_znode_hold_enter(zfsvfs, obj);
743 VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
744
745 /*
746 * If this is the root, fix up the half-initialized parent pointer
747 * to reference the just-allocated physical data area.
748 */
749 if (flag & IS_ROOT_NODE) {
750 dzp->z_id = obj;
751 }
752
753 /*
754 * If parent is an xattr, so am I.
755 */
756 if (dzp->z_pflags & ZFS_XATTR) {
757 flag |= IS_XATTR;
758 }
759
760 if (zfsvfs->z_use_fuids)
761 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
762 else
763 pflags = 0;
764
765 if (S_ISDIR(vap->va_mode)) {
766 size = 2; /* contents ("." and "..") */
767 links = 2;
768 } else {
769 size = 0;
770 links = (flag & IS_TMPFILE) ? 0 : 1;
771 }
772
773 if (S_ISBLK(vap->va_mode) || S_ISCHR(vap->va_mode))
774 rdev = vap->va_rdev;
775
776 parent = dzp->z_id;
777 mode = acl_ids->z_mode;
778 if (flag & IS_XATTR)
779 pflags |= ZFS_XATTR;
780
781 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode)) {
782 /*
783 * With ZFS_PROJID flag, we can easily know whether there is
784 * project ID stored on disk or not. See zfs_space_delta_cb().
785 */
786 if (obj_type != DMU_OT_ZNODE &&
787 dmu_objset_projectquota_enabled(zfsvfs->z_os))
788 pflags |= ZFS_PROJID;
789
790 /*
791 * Inherit project ID from parent if required.
792 */
793 projid = zfs_inherit_projid(dzp);
794 if (dzp->z_pflags & ZFS_PROJINHERIT)
795 pflags |= ZFS_PROJINHERIT;
796 }
797
798 /*
799 * No execs denied will be determined when zfs_mode_compute() is called.
800 */
801 pflags |= acl_ids->z_aclp->z_hints &
802 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
803 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
804
805 ZFS_TIME_ENCODE(&now, crtime);
806 ZFS_TIME_ENCODE(&now, ctime);
807
808 if (vap->va_mask & ATTR_ATIME) {
809 ZFS_TIME_ENCODE(&vap->va_atime, atime);
810 } else {
811 ZFS_TIME_ENCODE(&now, atime);
812 }
813
814 if (vap->va_mask & ATTR_MTIME) {
815 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
816 } else {
817 ZFS_TIME_ENCODE(&now, mtime);
818 }
819
820 /* Now add in all of the "SA" attributes */
821 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
822 &sa_hdl));
823
824 /*
825 * Setup the array of attributes to be replaced/set on the new file
826 *
827 * order for DMU_OT_ZNODE is critical since it needs to be constructed
828 * in the old znode_phys_t format. Don't change this ordering
829 */
830 sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
831
832 if (obj_type == DMU_OT_ZNODE) {
833 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
834 NULL, &atime, 16);
835 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
836 NULL, &mtime, 16);
837 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
838 NULL, &ctime, 16);
839 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
840 NULL, &crtime, 16);
841 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
842 NULL, &gen, 8);
843 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
844 NULL, &mode, 8);
845 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
846 NULL, &size, 8);
847 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
848 NULL, &parent, 8);
849 } else {
850 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
851 NULL, &mode, 8);
852 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
853 NULL, &size, 8);
854 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
855 NULL, &gen, 8);
856 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs),
857 NULL, &acl_ids->z_fuid, 8);
858 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs),
859 NULL, &acl_ids->z_fgid, 8);
860 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
861 NULL, &parent, 8);
862 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
863 NULL, &pflags, 8);
864 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
865 NULL, &atime, 16);
866 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
867 NULL, &mtime, 16);
868 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
869 NULL, &ctime, 16);
870 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
871 NULL, &crtime, 16);
872 }
873
874 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
875
876 if (obj_type == DMU_OT_ZNODE) {
877 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
878 &empty_xattr, 8);
879 } else if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
880 pflags & ZFS_PROJID) {
881 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PROJID(zfsvfs),
882 NULL, &projid, 8);
883 }
884 if (obj_type == DMU_OT_ZNODE ||
885 (S_ISBLK(vap->va_mode) || S_ISCHR(vap->va_mode))) {
886 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
887 NULL, &rdev, 8);
888 }
889 if (obj_type == DMU_OT_ZNODE) {
890 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
891 NULL, &pflags, 8);
892 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
893 &acl_ids->z_fuid, 8);
894 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
895 &acl_ids->z_fgid, 8);
896 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
897 sizeof (uint64_t) * 4);
898 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
899 &acl_phys, sizeof (zfs_acl_phys_t));
900 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
901 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
902 &acl_ids->z_aclp->z_acl_count, 8);
903 locate.cb_aclp = acl_ids->z_aclp;
904 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
905 zfs_acl_data_locator, &locate,
906 acl_ids->z_aclp->z_acl_bytes);
907 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
908 acl_ids->z_fuid, acl_ids->z_fgid);
909 }
910
911 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
912
913 if (!(flag & IS_ROOT_NODE)) {
914 /*
915 * The call to zfs_znode_alloc() may fail if memory is low
916 * via the call path: alloc_inode() -> inode_init_always() ->
917 * security_inode_alloc() -> inode_alloc_security(). Since
918 * the existing code is written such that zfs_mknode() can
919 * not fail retry until sufficient memory has been reclaimed.
920 */
921 do {
922 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
923 } while (*zpp == NULL);
924
925 VERIFY(*zpp != NULL);
926 VERIFY(dzp != NULL);
927 } else {
928 /*
929 * If we are creating the root node, the "parent" we
930 * passed in is the znode for the root.
931 */
932 *zpp = dzp;
933
934 (*zpp)->z_sa_hdl = sa_hdl;
935 }
936
937 (*zpp)->z_pflags = pflags;
938 (*zpp)->z_mode = ZTOI(*zpp)->i_mode = mode;
939 (*zpp)->z_dnodesize = dnodesize;
940 (*zpp)->z_projid = projid;
941
942 if (obj_type == DMU_OT_ZNODE ||
943 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
944 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
945 }
946 kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
947 zfs_znode_hold_exit(zfsvfs, zh);
948 }
949
950 /*
951 * Update in-core attributes. It is assumed the caller will be doing an
952 * sa_bulk_update to push the changes out.
953 */
954 void
zfs_xvattr_set(znode_t * zp,xvattr_t * xvap,dmu_tx_t * tx)955 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
956 {
957 xoptattr_t *xoap;
958 boolean_t update_inode = B_FALSE;
959
960 xoap = xva_getxoptattr(xvap);
961 ASSERT(xoap);
962
963 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
964 uint64_t times[2];
965 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
966 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(ZTOZSB(zp)),
967 ×, sizeof (times), tx);
968 XVA_SET_RTN(xvap, XAT_CREATETIME);
969 }
970 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
971 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
972 zp->z_pflags, tx);
973 XVA_SET_RTN(xvap, XAT_READONLY);
974 }
975 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
976 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
977 zp->z_pflags, tx);
978 XVA_SET_RTN(xvap, XAT_HIDDEN);
979 }
980 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
981 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
982 zp->z_pflags, tx);
983 XVA_SET_RTN(xvap, XAT_SYSTEM);
984 }
985 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
986 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
987 zp->z_pflags, tx);
988 XVA_SET_RTN(xvap, XAT_ARCHIVE);
989 }
990 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
991 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
992 zp->z_pflags, tx);
993 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
994
995 update_inode = B_TRUE;
996 }
997 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
998 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
999 zp->z_pflags, tx);
1000 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1001 }
1002 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1003 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1004 zp->z_pflags, tx);
1005 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1006
1007 update_inode = B_TRUE;
1008 }
1009 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1010 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1011 zp->z_pflags, tx);
1012 XVA_SET_RTN(xvap, XAT_NODUMP);
1013 }
1014 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1015 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1016 zp->z_pflags, tx);
1017 XVA_SET_RTN(xvap, XAT_OPAQUE);
1018 }
1019 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1020 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1021 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1022 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1023 }
1024 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1025 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1026 zp->z_pflags, tx);
1027 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1028 }
1029 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1030 zfs_sa_set_scanstamp(zp, xvap, tx);
1031 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1032 }
1033 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1034 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1035 zp->z_pflags, tx);
1036 XVA_SET_RTN(xvap, XAT_REPARSE);
1037 }
1038 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1039 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1040 zp->z_pflags, tx);
1041 XVA_SET_RTN(xvap, XAT_OFFLINE);
1042 }
1043 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1044 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1045 zp->z_pflags, tx);
1046 XVA_SET_RTN(xvap, XAT_SPARSE);
1047 }
1048 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
1049 ZFS_ATTR_SET(zp, ZFS_PROJINHERIT, xoap->xoa_projinherit,
1050 zp->z_pflags, tx);
1051 XVA_SET_RTN(xvap, XAT_PROJINHERIT);
1052 }
1053
1054 if (update_inode)
1055 zfs_set_inode_flags(zp, ZTOI(zp));
1056 }
1057
1058 int
zfs_zget(zfsvfs_t * zfsvfs,uint64_t obj_num,znode_t ** zpp)1059 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1060 {
1061 dmu_object_info_t doi;
1062 dmu_buf_t *db;
1063 znode_t *zp;
1064 znode_hold_t *zh;
1065 int err;
1066 sa_handle_t *hdl;
1067
1068 *zpp = NULL;
1069
1070 again:
1071 zh = zfs_znode_hold_enter(zfsvfs, obj_num);
1072
1073 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1074 if (err) {
1075 zfs_znode_hold_exit(zfsvfs, zh);
1076 return (err);
1077 }
1078
1079 dmu_object_info_from_db(db, &doi);
1080 if (doi.doi_bonus_type != DMU_OT_SA &&
1081 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1082 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1083 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1084 sa_buf_rele(db, NULL);
1085 zfs_znode_hold_exit(zfsvfs, zh);
1086 return (SET_ERROR(EINVAL));
1087 }
1088
1089 hdl = dmu_buf_get_user(db);
1090 if (hdl != NULL) {
1091 zp = sa_get_userdata(hdl);
1092
1093
1094 /*
1095 * Since "SA" does immediate eviction we
1096 * should never find a sa handle that doesn't
1097 * know about the znode.
1098 */
1099
1100 ASSERT3P(zp, !=, NULL);
1101
1102 mutex_enter(&zp->z_lock);
1103 ASSERT3U(zp->z_id, ==, obj_num);
1104 /*
1105 * If zp->z_unlinked is set, the znode is already marked
1106 * for deletion and should not be discovered. Check this
1107 * after checking igrab() due to fsetxattr() & O_TMPFILE.
1108 *
1109 * If igrab() returns NULL the VFS has independently
1110 * determined the inode should be evicted and has
1111 * called iput_final() to start the eviction process.
1112 * The SA handle is still valid but because the VFS
1113 * requires that the eviction succeed we must drop
1114 * our locks and references to allow the eviction to
1115 * complete. The zfs_zget() may then be retried.
1116 *
1117 * This unlikely case could be optimized by registering
1118 * a sops->drop_inode() callback. The callback would
1119 * need to detect the active SA hold thereby informing
1120 * the VFS that this inode should not be evicted.
1121 */
1122 if (igrab(ZTOI(zp)) == NULL) {
1123 if (zp->z_unlinked)
1124 err = SET_ERROR(ENOENT);
1125 else
1126 err = SET_ERROR(EAGAIN);
1127 } else {
1128 *zpp = zp;
1129 err = 0;
1130 }
1131
1132 mutex_exit(&zp->z_lock);
1133 sa_buf_rele(db, NULL);
1134 zfs_znode_hold_exit(zfsvfs, zh);
1135
1136 if (err == EAGAIN) {
1137 /* inode might need this to finish evict */
1138 cond_resched();
1139 goto again;
1140 }
1141 return (err);
1142 }
1143
1144 /*
1145 * Not found create new znode/vnode but only if file exists.
1146 *
1147 * There is a small window where zfs_vget() could
1148 * find this object while a file create is still in
1149 * progress. This is checked for in zfs_znode_alloc()
1150 *
1151 * if zfs_znode_alloc() fails it will drop the hold on the
1152 * bonus buffer.
1153 */
1154 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1155 doi.doi_bonus_type, NULL);
1156 if (zp == NULL) {
1157 err = SET_ERROR(ENOENT);
1158 } else {
1159 *zpp = zp;
1160 }
1161 zfs_znode_hold_exit(zfsvfs, zh);
1162 return (err);
1163 }
1164
1165 int
zfs_rezget(znode_t * zp)1166 zfs_rezget(znode_t *zp)
1167 {
1168 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1169 dmu_object_info_t doi;
1170 dmu_buf_t *db;
1171 uint64_t obj_num = zp->z_id;
1172 uint64_t mode;
1173 uint64_t links;
1174 sa_bulk_attr_t bulk[11];
1175 int err;
1176 int count = 0;
1177 uint64_t gen;
1178 uint64_t z_uid, z_gid;
1179 uint64_t atime[2], mtime[2], ctime[2], btime[2];
1180 uint64_t projid = ZFS_DEFAULT_PROJID;
1181 znode_hold_t *zh;
1182
1183 /*
1184 * skip ctldir, otherwise they will always get invalidated. This will
1185 * cause funny behaviour for the mounted snapdirs. Especially for
1186 * Linux >= 3.18, d_invalidate will detach the mountpoint and prevent
1187 * anyone automount it again as long as someone is still using the
1188 * detached mount.
1189 */
1190 if (zp->z_is_ctldir)
1191 return (0);
1192
1193 zh = zfs_znode_hold_enter(zfsvfs, obj_num);
1194
1195 mutex_enter(&zp->z_acl_lock);
1196 if (zp->z_acl_cached) {
1197 zfs_acl_free(zp->z_acl_cached);
1198 zp->z_acl_cached = NULL;
1199 }
1200 mutex_exit(&zp->z_acl_lock);
1201
1202 rw_enter(&zp->z_xattr_lock, RW_WRITER);
1203 if (zp->z_xattr_cached) {
1204 nvlist_free(zp->z_xattr_cached);
1205 zp->z_xattr_cached = NULL;
1206 }
1207 rw_exit(&zp->z_xattr_lock);
1208
1209 ASSERT(zp->z_sa_hdl == NULL);
1210 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1211 if (err) {
1212 zfs_znode_hold_exit(zfsvfs, zh);
1213 return (err);
1214 }
1215
1216 dmu_object_info_from_db(db, &doi);
1217 if (doi.doi_bonus_type != DMU_OT_SA &&
1218 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1219 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1220 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1221 sa_buf_rele(db, NULL);
1222 zfs_znode_hold_exit(zfsvfs, zh);
1223 return (SET_ERROR(EINVAL));
1224 }
1225
1226 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1227
1228 /* reload cached values */
1229 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1230 &gen, sizeof (gen));
1231 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1232 &zp->z_size, sizeof (zp->z_size));
1233 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1234 &links, sizeof (links));
1235 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1236 &zp->z_pflags, sizeof (zp->z_pflags));
1237 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1238 &z_uid, sizeof (z_uid));
1239 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1240 &z_gid, sizeof (z_gid));
1241 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1242 &mode, sizeof (mode));
1243 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1244 &atime, 16);
1245 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
1246 &mtime, 16);
1247 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1248 &ctime, 16);
1249 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CRTIME(zfsvfs), NULL, &btime, 16);
1250
1251 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1252 zfs_znode_dmu_fini(zp);
1253 zfs_znode_hold_exit(zfsvfs, zh);
1254 return (SET_ERROR(EIO));
1255 }
1256
1257 if (dmu_objset_projectquota_enabled(zfsvfs->z_os)) {
1258 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs),
1259 &projid, 8);
1260 if (err != 0 && err != ENOENT) {
1261 zfs_znode_dmu_fini(zp);
1262 zfs_znode_hold_exit(zfsvfs, zh);
1263 return (SET_ERROR(err));
1264 }
1265 }
1266
1267 zp->z_projid = projid;
1268 zp->z_mode = ZTOI(zp)->i_mode = mode;
1269 zfs_uid_write(ZTOI(zp), z_uid);
1270 zfs_gid_write(ZTOI(zp), z_gid);
1271
1272 ZFS_TIME_DECODE(&ZTOI(zp)->i_atime, atime);
1273 ZFS_TIME_DECODE(&ZTOI(zp)->i_mtime, mtime);
1274 ZFS_TIME_DECODE(&ZTOI(zp)->i_ctime, ctime);
1275 ZFS_TIME_DECODE(&zp->z_btime, btime);
1276
1277 if ((uint32_t)gen != ZTOI(zp)->i_generation) {
1278 zfs_znode_dmu_fini(zp);
1279 zfs_znode_hold_exit(zfsvfs, zh);
1280 return (SET_ERROR(EIO));
1281 }
1282
1283 set_nlink(ZTOI(zp), (uint32_t)links);
1284 zfs_set_inode_flags(zp, ZTOI(zp));
1285
1286 zp->z_blksz = doi.doi_data_block_size;
1287 zp->z_atime_dirty = B_FALSE;
1288 zfs_znode_update_vfs(zp);
1289
1290 /*
1291 * If the file has zero links, then it has been unlinked on the send
1292 * side and it must be in the received unlinked set.
1293 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1294 * stale data and to prevent automatic removal of the file in
1295 * zfs_zinactive(). The file will be removed either when it is removed
1296 * on the send side and the next incremental stream is received or
1297 * when the unlinked set gets processed.
1298 */
1299 zp->z_unlinked = (ZTOI(zp)->i_nlink == 0);
1300 if (zp->z_unlinked)
1301 zfs_znode_dmu_fini(zp);
1302
1303 zfs_znode_hold_exit(zfsvfs, zh);
1304
1305 return (0);
1306 }
1307
1308 void
zfs_znode_delete(znode_t * zp,dmu_tx_t * tx)1309 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1310 {
1311 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1312 objset_t *os = zfsvfs->z_os;
1313 uint64_t obj = zp->z_id;
1314 uint64_t acl_obj = zfs_external_acl(zp);
1315 znode_hold_t *zh;
1316
1317 zh = zfs_znode_hold_enter(zfsvfs, obj);
1318 if (acl_obj) {
1319 VERIFY(!zp->z_is_sa);
1320 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1321 }
1322 VERIFY(0 == dmu_object_free(os, obj, tx));
1323 zfs_znode_dmu_fini(zp);
1324 zfs_znode_hold_exit(zfsvfs, zh);
1325 }
1326
1327 void
zfs_zinactive(znode_t * zp)1328 zfs_zinactive(znode_t *zp)
1329 {
1330 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1331 uint64_t z_id = zp->z_id;
1332 znode_hold_t *zh;
1333
1334 ASSERT(zp->z_sa_hdl);
1335
1336 /*
1337 * Don't allow a zfs_zget() while were trying to release this znode.
1338 */
1339 zh = zfs_znode_hold_enter(zfsvfs, z_id);
1340
1341 mutex_enter(&zp->z_lock);
1342
1343 /*
1344 * If this was the last reference to a file with no links, remove
1345 * the file from the file system unless the file system is mounted
1346 * read-only. That can happen, for example, if the file system was
1347 * originally read-write, the file was opened, then unlinked and
1348 * the file system was made read-only before the file was finally
1349 * closed. The file will remain in the unlinked set.
1350 */
1351 if (zp->z_unlinked) {
1352 ASSERT(!zfsvfs->z_issnap);
1353 if (!zfs_is_readonly(zfsvfs) && !zfs_unlink_suspend_progress) {
1354 mutex_exit(&zp->z_lock);
1355 zfs_znode_hold_exit(zfsvfs, zh);
1356 zfs_rmnode(zp);
1357 return;
1358 }
1359 }
1360
1361 mutex_exit(&zp->z_lock);
1362 zfs_znode_dmu_fini(zp);
1363
1364 zfs_znode_hold_exit(zfsvfs, zh);
1365 }
1366
1367 #if defined(HAVE_INODE_TIMESPEC64_TIMES)
1368 #define zfs_compare_timespec timespec64_compare
1369 #else
1370 #define zfs_compare_timespec timespec_compare
1371 #endif
1372
1373 /*
1374 * Determine whether the znode's atime must be updated. The logic mostly
1375 * duplicates the Linux kernel's relatime_need_update() functionality.
1376 * This function is only called if the underlying filesystem actually has
1377 * atime updates enabled.
1378 */
1379 boolean_t
zfs_relatime_need_update(const struct inode * ip)1380 zfs_relatime_need_update(const struct inode *ip)
1381 {
1382 inode_timespec_t now;
1383
1384 gethrestime(&now);
1385 /*
1386 * In relatime mode, only update the atime if the previous atime
1387 * is earlier than either the ctime or mtime or if at least a day
1388 * has passed since the last update of atime.
1389 */
1390 if (zfs_compare_timespec(&ip->i_mtime, &ip->i_atime) >= 0)
1391 return (B_TRUE);
1392
1393 if (zfs_compare_timespec(&ip->i_ctime, &ip->i_atime) >= 0)
1394 return (B_TRUE);
1395
1396 if ((hrtime_t)now.tv_sec - (hrtime_t)ip->i_atime.tv_sec >= 24*60*60)
1397 return (B_TRUE);
1398
1399 return (B_FALSE);
1400 }
1401
1402 /*
1403 * Prepare to update znode time stamps.
1404 *
1405 * IN: zp - znode requiring timestamp update
1406 * flag - ATTR_MTIME, ATTR_CTIME flags
1407 *
1408 * OUT: zp - z_seq
1409 * mtime - new mtime
1410 * ctime - new ctime
1411 *
1412 * Note: We don't update atime here, because we rely on Linux VFS to do
1413 * atime updating.
1414 */
1415 void
zfs_tstamp_update_setup(znode_t * zp,uint_t flag,uint64_t mtime[2],uint64_t ctime[2])1416 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1417 uint64_t ctime[2])
1418 {
1419 inode_timespec_t now;
1420
1421 gethrestime(&now);
1422
1423 zp->z_seq++;
1424
1425 if (flag & ATTR_MTIME) {
1426 ZFS_TIME_ENCODE(&now, mtime);
1427 ZFS_TIME_DECODE(&(ZTOI(zp)->i_mtime), mtime);
1428 if (ZTOZSB(zp)->z_use_fuids) {
1429 zp->z_pflags |= (ZFS_ARCHIVE |
1430 ZFS_AV_MODIFIED);
1431 }
1432 }
1433
1434 if (flag & ATTR_CTIME) {
1435 ZFS_TIME_ENCODE(&now, ctime);
1436 ZFS_TIME_DECODE(&(ZTOI(zp)->i_ctime), ctime);
1437 if (ZTOZSB(zp)->z_use_fuids)
1438 zp->z_pflags |= ZFS_ARCHIVE;
1439 }
1440 }
1441
1442 /*
1443 * Grow the block size for a file.
1444 *
1445 * IN: zp - znode of file to free data in.
1446 * size - requested block size
1447 * tx - open transaction.
1448 *
1449 * NOTE: this function assumes that the znode is write locked.
1450 */
1451 void
zfs_grow_blocksize(znode_t * zp,uint64_t size,dmu_tx_t * tx)1452 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1453 {
1454 int error;
1455 u_longlong_t dummy;
1456
1457 if (size <= zp->z_blksz)
1458 return;
1459 /*
1460 * If the file size is already greater than the current blocksize,
1461 * we will not grow. If there is more than one block in a file,
1462 * the blocksize cannot change.
1463 */
1464 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1465 return;
1466
1467 error = dmu_object_set_blocksize(ZTOZSB(zp)->z_os, zp->z_id,
1468 size, 0, tx);
1469
1470 if (error == ENOTSUP)
1471 return;
1472 ASSERT0(error);
1473
1474 /* What blocksize did we actually get? */
1475 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1476 }
1477
1478 /*
1479 * Increase the file length
1480 *
1481 * IN: zp - znode of file to free data in.
1482 * end - new end-of-file
1483 *
1484 * RETURN: 0 on success, error code on failure
1485 */
1486 static int
zfs_extend(znode_t * zp,uint64_t end)1487 zfs_extend(znode_t *zp, uint64_t end)
1488 {
1489 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1490 dmu_tx_t *tx;
1491 zfs_locked_range_t *lr;
1492 uint64_t newblksz;
1493 int error;
1494
1495 /*
1496 * We will change zp_size, lock the whole file.
1497 */
1498 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1499
1500 /*
1501 * Nothing to do if file already at desired length.
1502 */
1503 if (end <= zp->z_size) {
1504 zfs_rangelock_exit(lr);
1505 return (0);
1506 }
1507 tx = dmu_tx_create(zfsvfs->z_os);
1508 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1509 zfs_sa_upgrade_txholds(tx, zp);
1510 if (end > zp->z_blksz &&
1511 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1512 /*
1513 * We are growing the file past the current block size.
1514 */
1515 if (zp->z_blksz > ZTOZSB(zp)->z_max_blksz) {
1516 /*
1517 * File's blocksize is already larger than the
1518 * "recordsize" property. Only let it grow to
1519 * the next power of 2.
1520 */
1521 ASSERT(!ISP2(zp->z_blksz));
1522 newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1523 } else {
1524 newblksz = MIN(end, ZTOZSB(zp)->z_max_blksz);
1525 }
1526 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1527 } else {
1528 newblksz = 0;
1529 }
1530
1531 error = dmu_tx_assign(tx, TXG_WAIT);
1532 if (error) {
1533 dmu_tx_abort(tx);
1534 zfs_rangelock_exit(lr);
1535 return (error);
1536 }
1537
1538 if (newblksz)
1539 zfs_grow_blocksize(zp, newblksz, tx);
1540
1541 zp->z_size = end;
1542
1543 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(ZTOZSB(zp)),
1544 &zp->z_size, sizeof (zp->z_size), tx));
1545
1546 zfs_rangelock_exit(lr);
1547
1548 dmu_tx_commit(tx);
1549
1550 return (0);
1551 }
1552
1553 /*
1554 * zfs_zero_partial_page - Modeled after update_pages() but
1555 * with different arguments and semantics for use by zfs_freesp().
1556 *
1557 * Zeroes a piece of a single page cache entry for zp at offset
1558 * start and length len.
1559 *
1560 * Caller must acquire a range lock on the file for the region
1561 * being zeroed in order that the ARC and page cache stay in sync.
1562 */
1563 static void
zfs_zero_partial_page(znode_t * zp,uint64_t start,uint64_t len)1564 zfs_zero_partial_page(znode_t *zp, uint64_t start, uint64_t len)
1565 {
1566 struct address_space *mp = ZTOI(zp)->i_mapping;
1567 struct page *pp;
1568 int64_t off;
1569 void *pb;
1570
1571 ASSERT((start & PAGE_MASK) == ((start + len - 1) & PAGE_MASK));
1572
1573 off = start & (PAGE_SIZE - 1);
1574 start &= PAGE_MASK;
1575
1576 pp = find_lock_page(mp, start >> PAGE_SHIFT);
1577 if (pp) {
1578 if (mapping_writably_mapped(mp))
1579 flush_dcache_page(pp);
1580
1581 pb = kmap(pp);
1582 bzero(pb + off, len);
1583 kunmap(pp);
1584
1585 if (mapping_writably_mapped(mp))
1586 flush_dcache_page(pp);
1587
1588 mark_page_accessed(pp);
1589 SetPageUptodate(pp);
1590 ClearPageError(pp);
1591 unlock_page(pp);
1592 put_page(pp);
1593 }
1594 }
1595
1596 /*
1597 * Free space in a file.
1598 *
1599 * IN: zp - znode of file to free data in.
1600 * off - start of section to free.
1601 * len - length of section to free.
1602 *
1603 * RETURN: 0 on success, error code on failure
1604 */
1605 static int
zfs_free_range(znode_t * zp,uint64_t off,uint64_t len)1606 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1607 {
1608 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1609 zfs_locked_range_t *lr;
1610 int error;
1611
1612 /*
1613 * Lock the range being freed.
1614 */
1615 lr = zfs_rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER);
1616
1617 /*
1618 * Nothing to do if file already at desired length.
1619 */
1620 if (off >= zp->z_size) {
1621 zfs_rangelock_exit(lr);
1622 return (0);
1623 }
1624
1625 if (off + len > zp->z_size)
1626 len = zp->z_size - off;
1627
1628 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1629
1630 /*
1631 * Zero partial page cache entries. This must be done under a
1632 * range lock in order to keep the ARC and page cache in sync.
1633 */
1634 if (zp->z_is_mapped) {
1635 loff_t first_page, last_page, page_len;
1636 loff_t first_page_offset, last_page_offset;
1637
1638 /* first possible full page in hole */
1639 first_page = (off + PAGE_SIZE - 1) >> PAGE_SHIFT;
1640 /* last page of hole */
1641 last_page = (off + len) >> PAGE_SHIFT;
1642
1643 /* offset of first_page */
1644 first_page_offset = first_page << PAGE_SHIFT;
1645 /* offset of last_page */
1646 last_page_offset = last_page << PAGE_SHIFT;
1647
1648 /* truncate whole pages */
1649 if (last_page_offset > first_page_offset) {
1650 truncate_inode_pages_range(ZTOI(zp)->i_mapping,
1651 first_page_offset, last_page_offset - 1);
1652 }
1653
1654 /* truncate sub-page ranges */
1655 if (first_page > last_page) {
1656 /* entire punched area within a single page */
1657 zfs_zero_partial_page(zp, off, len);
1658 } else {
1659 /* beginning of punched area at the end of a page */
1660 page_len = first_page_offset - off;
1661 if (page_len > 0)
1662 zfs_zero_partial_page(zp, off, page_len);
1663
1664 /* end of punched area at the beginning of a page */
1665 page_len = off + len - last_page_offset;
1666 if (page_len > 0)
1667 zfs_zero_partial_page(zp, last_page_offset,
1668 page_len);
1669 }
1670 }
1671 zfs_rangelock_exit(lr);
1672
1673 return (error);
1674 }
1675
1676 /*
1677 * Truncate a file
1678 *
1679 * IN: zp - znode of file to free data in.
1680 * end - new end-of-file.
1681 *
1682 * RETURN: 0 on success, error code on failure
1683 */
1684 static int
zfs_trunc(znode_t * zp,uint64_t end)1685 zfs_trunc(znode_t *zp, uint64_t end)
1686 {
1687 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1688 dmu_tx_t *tx;
1689 zfs_locked_range_t *lr;
1690 int error;
1691 sa_bulk_attr_t bulk[2];
1692 int count = 0;
1693
1694 /*
1695 * We will change zp_size, lock the whole file.
1696 */
1697 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1698
1699 /*
1700 * Nothing to do if file already at desired length.
1701 */
1702 if (end >= zp->z_size) {
1703 zfs_rangelock_exit(lr);
1704 return (0);
1705 }
1706
1707 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1708 DMU_OBJECT_END);
1709 if (error) {
1710 zfs_rangelock_exit(lr);
1711 return (error);
1712 }
1713 tx = dmu_tx_create(zfsvfs->z_os);
1714 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1715 zfs_sa_upgrade_txholds(tx, zp);
1716 dmu_tx_mark_netfree(tx);
1717 error = dmu_tx_assign(tx, TXG_WAIT);
1718 if (error) {
1719 dmu_tx_abort(tx);
1720 zfs_rangelock_exit(lr);
1721 return (error);
1722 }
1723
1724 zp->z_size = end;
1725 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1726 NULL, &zp->z_size, sizeof (zp->z_size));
1727
1728 if (end == 0) {
1729 zp->z_pflags &= ~ZFS_SPARSE;
1730 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1731 NULL, &zp->z_pflags, 8);
1732 }
1733 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1734
1735 dmu_tx_commit(tx);
1736 zfs_rangelock_exit(lr);
1737
1738 return (0);
1739 }
1740
1741 /*
1742 * Free space in a file
1743 *
1744 * IN: zp - znode of file to free data in.
1745 * off - start of range
1746 * len - end of range (0 => EOF)
1747 * flag - current file open mode flags.
1748 * log - TRUE if this action should be logged
1749 *
1750 * RETURN: 0 on success, error code on failure
1751 */
1752 int
zfs_freesp(znode_t * zp,uint64_t off,uint64_t len,int flag,boolean_t log)1753 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1754 {
1755 dmu_tx_t *tx;
1756 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1757 zilog_t *zilog = zfsvfs->z_log;
1758 uint64_t mode;
1759 uint64_t mtime[2], ctime[2];
1760 sa_bulk_attr_t bulk[3];
1761 int count = 0;
1762 int error;
1763
1764 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1765 sizeof (mode))) != 0)
1766 return (error);
1767
1768 if (off > zp->z_size) {
1769 error = zfs_extend(zp, off+len);
1770 if (error == 0 && log)
1771 goto log;
1772 goto out;
1773 }
1774
1775 if (len == 0) {
1776 error = zfs_trunc(zp, off);
1777 } else {
1778 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1779 off + len > zp->z_size)
1780 error = zfs_extend(zp, off+len);
1781 }
1782 if (error || !log)
1783 goto out;
1784 log:
1785 tx = dmu_tx_create(zfsvfs->z_os);
1786 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1787 zfs_sa_upgrade_txholds(tx, zp);
1788 error = dmu_tx_assign(tx, TXG_WAIT);
1789 if (error) {
1790 dmu_tx_abort(tx);
1791 goto out;
1792 }
1793
1794 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1795 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1796 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1797 NULL, &zp->z_pflags, 8);
1798 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
1799 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1800 ASSERT(error == 0);
1801
1802 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1803
1804 dmu_tx_commit(tx);
1805
1806 zfs_znode_update_vfs(zp);
1807 error = 0;
1808
1809 out:
1810 /*
1811 * Truncate the page cache - for file truncate operations, use
1812 * the purpose-built API for truncations. For punching operations,
1813 * the truncation is handled under a range lock in zfs_free_range.
1814 */
1815 if (len == 0)
1816 truncate_setsize(ZTOI(zp), off);
1817 return (error);
1818 }
1819
1820 void
zfs_create_fs(objset_t * os,cred_t * cr,nvlist_t * zplprops,dmu_tx_t * tx)1821 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1822 {
1823 struct super_block *sb;
1824 zfsvfs_t *zfsvfs;
1825 uint64_t moid, obj, sa_obj, version;
1826 uint64_t sense = ZFS_CASE_SENSITIVE;
1827 uint64_t norm = 0;
1828 nvpair_t *elem;
1829 int size;
1830 int error;
1831 int i;
1832 znode_t *rootzp = NULL;
1833 vattr_t vattr;
1834 znode_t *zp;
1835 zfs_acl_ids_t acl_ids;
1836
1837 /*
1838 * First attempt to create master node.
1839 */
1840 /*
1841 * In an empty objset, there are no blocks to read and thus
1842 * there can be no i/o errors (which we assert below).
1843 */
1844 moid = MASTER_NODE_OBJ;
1845 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1846 DMU_OT_NONE, 0, tx);
1847 ASSERT(error == 0);
1848
1849 /*
1850 * Set starting attributes.
1851 */
1852 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1853 elem = NULL;
1854 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1855 /* For the moment we expect all zpl props to be uint64_ts */
1856 uint64_t val;
1857 char *name;
1858
1859 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1860 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1861 name = nvpair_name(elem);
1862 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1863 if (val < version)
1864 version = val;
1865 } else {
1866 error = zap_update(os, moid, name, 8, 1, &val, tx);
1867 }
1868 ASSERT(error == 0);
1869 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1870 norm = val;
1871 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1872 sense = val;
1873 }
1874 ASSERT(version != 0);
1875 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1876
1877 /*
1878 * Create zap object used for SA attribute registration
1879 */
1880
1881 if (version >= ZPL_VERSION_SA) {
1882 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1883 DMU_OT_NONE, 0, tx);
1884 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1885 ASSERT(error == 0);
1886 } else {
1887 sa_obj = 0;
1888 }
1889 /*
1890 * Create a delete queue.
1891 */
1892 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1893
1894 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1895 ASSERT(error == 0);
1896
1897 /*
1898 * Create root znode. Create minimal znode/inode/zfsvfs/sb
1899 * to allow zfs_mknode to work.
1900 */
1901 vattr.va_mask = ATTR_MODE|ATTR_UID|ATTR_GID;
1902 vattr.va_mode = S_IFDIR|0755;
1903 vattr.va_uid = crgetuid(cr);
1904 vattr.va_gid = crgetgid(cr);
1905
1906 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1907 rootzp->z_unlinked = B_FALSE;
1908 rootzp->z_atime_dirty = B_FALSE;
1909 rootzp->z_is_sa = USE_SA(version, os);
1910 rootzp->z_pflags = 0;
1911
1912 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1913 zfsvfs->z_os = os;
1914 zfsvfs->z_parent = zfsvfs;
1915 zfsvfs->z_version = version;
1916 zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1917 zfsvfs->z_use_sa = USE_SA(version, os);
1918 zfsvfs->z_norm = norm;
1919
1920 sb = kmem_zalloc(sizeof (struct super_block), KM_SLEEP);
1921 sb->s_fs_info = zfsvfs;
1922
1923 ZTOI(rootzp)->i_sb = sb;
1924
1925 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1926 &zfsvfs->z_attr_table);
1927
1928 ASSERT(error == 0);
1929
1930 /*
1931 * Fold case on file systems that are always or sometimes case
1932 * insensitive.
1933 */
1934 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1935 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1936
1937 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1938 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1939 offsetof(znode_t, z_link_node));
1940
1941 size = MIN(1 << (highbit64(zfs_object_mutex_size)-1), ZFS_OBJ_MTX_MAX);
1942 zfsvfs->z_hold_size = size;
1943 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
1944 KM_SLEEP);
1945 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
1946 for (i = 0; i != size; i++) {
1947 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
1948 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
1949 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
1950 }
1951
1952 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1953 cr, NULL, &acl_ids));
1954 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1955 ASSERT3P(zp, ==, rootzp);
1956 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1957 ASSERT(error == 0);
1958 zfs_acl_ids_free(&acl_ids);
1959
1960 atomic_set(&ZTOI(rootzp)->i_count, 0);
1961 sa_handle_destroy(rootzp->z_sa_hdl);
1962 kmem_cache_free(znode_cache, rootzp);
1963
1964 for (i = 0; i != size; i++) {
1965 avl_destroy(&zfsvfs->z_hold_trees[i]);
1966 mutex_destroy(&zfsvfs->z_hold_locks[i]);
1967 }
1968
1969 mutex_destroy(&zfsvfs->z_znodes_lock);
1970
1971 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
1972 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
1973 kmem_free(sb, sizeof (struct super_block));
1974 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1975 }
1976 #endif /* _KERNEL */
1977
1978 static int
zfs_sa_setup(objset_t * osp,sa_attr_type_t ** sa_table)1979 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1980 {
1981 uint64_t sa_obj = 0;
1982 int error;
1983
1984 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1985 if (error != 0 && error != ENOENT)
1986 return (error);
1987
1988 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1989 return (error);
1990 }
1991
1992 static int
zfs_grab_sa_handle(objset_t * osp,uint64_t obj,sa_handle_t ** hdlp,dmu_buf_t ** db,void * tag)1993 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1994 dmu_buf_t **db, void *tag)
1995 {
1996 dmu_object_info_t doi;
1997 int error;
1998
1999 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
2000 return (error);
2001
2002 dmu_object_info_from_db(*db, &doi);
2003 if ((doi.doi_bonus_type != DMU_OT_SA &&
2004 doi.doi_bonus_type != DMU_OT_ZNODE) ||
2005 (doi.doi_bonus_type == DMU_OT_ZNODE &&
2006 doi.doi_bonus_size < sizeof (znode_phys_t))) {
2007 sa_buf_rele(*db, tag);
2008 return (SET_ERROR(ENOTSUP));
2009 }
2010
2011 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
2012 if (error != 0) {
2013 sa_buf_rele(*db, tag);
2014 return (error);
2015 }
2016
2017 return (0);
2018 }
2019
2020 static void
zfs_release_sa_handle(sa_handle_t * hdl,dmu_buf_t * db,void * tag)2021 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
2022 {
2023 sa_handle_destroy(hdl);
2024 sa_buf_rele(db, tag);
2025 }
2026
2027 /*
2028 * Given an object number, return its parent object number and whether
2029 * or not the object is an extended attribute directory.
2030 */
2031 static int
zfs_obj_to_pobj(objset_t * osp,sa_handle_t * hdl,sa_attr_type_t * sa_table,uint64_t * pobjp,int * is_xattrdir)2032 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
2033 uint64_t *pobjp, int *is_xattrdir)
2034 {
2035 uint64_t parent;
2036 uint64_t pflags;
2037 uint64_t mode;
2038 uint64_t parent_mode;
2039 sa_bulk_attr_t bulk[3];
2040 sa_handle_t *sa_hdl;
2041 dmu_buf_t *sa_db;
2042 int count = 0;
2043 int error;
2044
2045 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
2046 &parent, sizeof (parent));
2047 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
2048 &pflags, sizeof (pflags));
2049 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2050 &mode, sizeof (mode));
2051
2052 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
2053 return (error);
2054
2055 /*
2056 * When a link is removed its parent pointer is not changed and will
2057 * be invalid. There are two cases where a link is removed but the
2058 * file stays around, when it goes to the delete queue and when there
2059 * are additional links.
2060 */
2061 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
2062 if (error != 0)
2063 return (error);
2064
2065 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
2066 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2067 if (error != 0)
2068 return (error);
2069
2070 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
2071
2072 /*
2073 * Extended attributes can be applied to files, directories, etc.
2074 * Otherwise the parent must be a directory.
2075 */
2076 if (!*is_xattrdir && !S_ISDIR(parent_mode))
2077 return (SET_ERROR(EINVAL));
2078
2079 *pobjp = parent;
2080
2081 return (0);
2082 }
2083
2084 /*
2085 * Given an object number, return some zpl level statistics
2086 */
2087 static int
zfs_obj_to_stats_impl(sa_handle_t * hdl,sa_attr_type_t * sa_table,zfs_stat_t * sb)2088 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2089 zfs_stat_t *sb)
2090 {
2091 sa_bulk_attr_t bulk[4];
2092 int count = 0;
2093
2094 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2095 &sb->zs_mode, sizeof (sb->zs_mode));
2096 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2097 &sb->zs_gen, sizeof (sb->zs_gen));
2098 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2099 &sb->zs_links, sizeof (sb->zs_links));
2100 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2101 &sb->zs_ctime, sizeof (sb->zs_ctime));
2102
2103 return (sa_bulk_lookup(hdl, bulk, count));
2104 }
2105
2106 static int
zfs_obj_to_path_impl(objset_t * osp,uint64_t obj,sa_handle_t * hdl,sa_attr_type_t * sa_table,char * buf,int len)2107 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2108 sa_attr_type_t *sa_table, char *buf, int len)
2109 {
2110 sa_handle_t *sa_hdl;
2111 sa_handle_t *prevhdl = NULL;
2112 dmu_buf_t *prevdb = NULL;
2113 dmu_buf_t *sa_db = NULL;
2114 char *path = buf + len - 1;
2115 int error;
2116
2117 *path = '\0';
2118 sa_hdl = hdl;
2119
2120 uint64_t deleteq_obj;
2121 VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ,
2122 ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj));
2123 error = zap_lookup_int(osp, deleteq_obj, obj);
2124 if (error == 0) {
2125 return (ESTALE);
2126 } else if (error != ENOENT) {
2127 return (error);
2128 }
2129 error = 0;
2130
2131 for (;;) {
2132 uint64_t pobj = 0;
2133 char component[MAXNAMELEN + 2];
2134 size_t complen;
2135 int is_xattrdir = 0;
2136
2137 if (prevdb) {
2138 ASSERT(prevhdl != NULL);
2139 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2140 }
2141
2142 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2143 &is_xattrdir)) != 0)
2144 break;
2145
2146 if (pobj == obj) {
2147 if (path[0] != '/')
2148 *--path = '/';
2149 break;
2150 }
2151
2152 component[0] = '/';
2153 if (is_xattrdir) {
2154 (void) sprintf(component + 1, "<xattrdir>");
2155 } else {
2156 error = zap_value_search(osp, pobj, obj,
2157 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2158 if (error != 0)
2159 break;
2160 }
2161
2162 complen = strlen(component);
2163 path -= complen;
2164 ASSERT(path >= buf);
2165 bcopy(component, path, complen);
2166 obj = pobj;
2167
2168 if (sa_hdl != hdl) {
2169 prevhdl = sa_hdl;
2170 prevdb = sa_db;
2171 }
2172 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2173 if (error != 0) {
2174 sa_hdl = prevhdl;
2175 sa_db = prevdb;
2176 break;
2177 }
2178 }
2179
2180 if (sa_hdl != NULL && sa_hdl != hdl) {
2181 ASSERT(sa_db != NULL);
2182 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2183 }
2184
2185 if (error == 0)
2186 (void) memmove(buf, path, buf + len - path);
2187
2188 return (error);
2189 }
2190
2191 int
zfs_obj_to_path(objset_t * osp,uint64_t obj,char * buf,int len)2192 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2193 {
2194 sa_attr_type_t *sa_table;
2195 sa_handle_t *hdl;
2196 dmu_buf_t *db;
2197 int error;
2198
2199 error = zfs_sa_setup(osp, &sa_table);
2200 if (error != 0)
2201 return (error);
2202
2203 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2204 if (error != 0)
2205 return (error);
2206
2207 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2208
2209 zfs_release_sa_handle(hdl, db, FTAG);
2210 return (error);
2211 }
2212
2213 int
zfs_obj_to_stats(objset_t * osp,uint64_t obj,zfs_stat_t * sb,char * buf,int len)2214 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2215 char *buf, int len)
2216 {
2217 char *path = buf + len - 1;
2218 sa_attr_type_t *sa_table;
2219 sa_handle_t *hdl;
2220 dmu_buf_t *db;
2221 int error;
2222
2223 *path = '\0';
2224
2225 error = zfs_sa_setup(osp, &sa_table);
2226 if (error != 0)
2227 return (error);
2228
2229 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2230 if (error != 0)
2231 return (error);
2232
2233 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2234 if (error != 0) {
2235 zfs_release_sa_handle(hdl, db, FTAG);
2236 return (error);
2237 }
2238
2239 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2240
2241 zfs_release_sa_handle(hdl, db, FTAG);
2242 return (error);
2243 }
2244
2245 #if defined(_KERNEL)
2246 EXPORT_SYMBOL(zfs_create_fs);
2247 EXPORT_SYMBOL(zfs_obj_to_path);
2248
2249 /* CSTYLED */
2250 module_param(zfs_object_mutex_size, uint, 0644);
2251 MODULE_PARM_DESC(zfs_object_mutex_size, "Size of znode hold array");
2252 module_param(zfs_unlink_suspend_progress, int, 0644);
2253 MODULE_PARM_DESC(zfs_unlink_suspend_progress, "Set to prevent async unlinks "
2254 "(debug - leaks space into the unlinked set)");
2255 #endif
2256