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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
29 #include <sys/types.h>
30 #include <sys/param.h>
32 #include <sys/sysmacros.h>
33 #include <sys/mntent.h>
34 #include <sys/u8_textprep.h>
35 #include <sys/dsl_dataset.h>
37 #include <sys/vnode.h>
40 #include <sys/errno.h>
42 #include <sys/atomic.h>
43 #include <sys/zfs_dir.h>
44 #include <sys/zfs_acl.h>
45 #include <sys/zfs_ioctl.h>
46 #include <sys/zfs_rlock.h>
47 #include <sys/zfs_fuid.h>
48 #include <sys/zfs_vnops.h>
49 #include <sys/zfs_ctldir.h>
50 #include <sys/dnode.h>
51 #include <sys/fs/zfs.h>
56 #include <sys/dmu_objset.h>
57 #include <sys/dmu_tx.h>
58 #include <sys/refcount.h>
61 #include <sys/zfs_znode.h>
63 #include <sys/zfs_sa.h>
64 #include <sys/zfs_stat.h>
67 #include "zfs_comutil.h"
70 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
71 * turned on when DEBUG is also defined.
78 #define ZNODE_STAT_ADD(stat) ((stat)++)
80 #define ZNODE_STAT_ADD(stat) /* nothing */
81 #endif /* ZNODE_STATS */
84 * Functions needed for userland (ie: libzpool) are not put under
85 * #ifdef_KERNEL; the rest of the functions have dependencies
86 * (such as VFS logic) that will not compile easily in userland.
90 static kmem_cache_t *znode_cache = NULL;
91 static kmem_cache_t *znode_hold_cache = NULL;
92 unsigned int zfs_object_mutex_size = ZFS_OBJ_MTX_SZ;
95 * This callback is invoked when acquiring a RL_WRITER or RL_APPEND lock on
96 * z_rangelock. It will modify the offset and length of the lock to reflect
97 * znode-specific information, and convert RL_APPEND to RL_WRITER. This is
98 * called with the rangelock_t's rl_lock held, which avoids races.
101 zfs_rangelock_cb(locked_range_t *new, void *arg)
106 * If in append mode, convert to writer and lock starting at the
107 * current end of file.
109 if (new->lr_type == RL_APPEND) {
110 new->lr_offset = zp->z_size;
111 new->lr_type = RL_WRITER;
115 * If we need to grow the block size then lock the whole file range.
117 uint64_t end_size = MAX(zp->z_size, new->lr_offset + new->lr_length);
118 if (end_size > zp->z_blksz && (!ISP2(zp->z_blksz) ||
119 zp->z_blksz < ZTOZSB(zp)->z_max_blksz)) {
121 new->lr_length = UINT64_MAX;
127 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
131 inode_init_once(ZTOI(zp));
132 list_link_init(&zp->z_link_node);
134 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
135 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
136 rw_init(&zp->z_name_lock, NULL, RW_NOLOCKDEP, NULL);
137 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
138 rw_init(&zp->z_xattr_lock, NULL, RW_DEFAULT, NULL);
140 rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp);
142 zp->z_dirlocks = NULL;
143 zp->z_acl_cached = NULL;
144 zp->z_xattr_cached = NULL;
145 zp->z_xattr_parent = 0;
152 zfs_znode_cache_destructor(void *buf, void *arg)
156 ASSERT(!list_link_active(&zp->z_link_node));
157 mutex_destroy(&zp->z_lock);
158 rw_destroy(&zp->z_parent_lock);
159 rw_destroy(&zp->z_name_lock);
160 mutex_destroy(&zp->z_acl_lock);
161 rw_destroy(&zp->z_xattr_lock);
162 rangelock_fini(&zp->z_rangelock);
164 ASSERT(zp->z_dirlocks == NULL);
165 ASSERT(zp->z_acl_cached == NULL);
166 ASSERT(zp->z_xattr_cached == NULL);
170 zfs_znode_hold_cache_constructor(void *buf, void *arg, int kmflags)
172 znode_hold_t *zh = buf;
174 mutex_init(&zh->zh_lock, NULL, MUTEX_DEFAULT, NULL);
175 zfs_refcount_create(&zh->zh_refcount);
176 zh->zh_obj = ZFS_NO_OBJECT;
182 zfs_znode_hold_cache_destructor(void *buf, void *arg)
184 znode_hold_t *zh = buf;
186 mutex_destroy(&zh->zh_lock);
187 zfs_refcount_destroy(&zh->zh_refcount);
194 * Initialize zcache. The KMC_SLAB hint is used in order that it be
195 * backed by kmalloc() when on the Linux slab in order that any
196 * wait_on_bit() operations on the related inode operate properly.
198 ASSERT(znode_cache == NULL);
199 znode_cache = kmem_cache_create("zfs_znode_cache",
200 sizeof (znode_t), 0, zfs_znode_cache_constructor,
201 zfs_znode_cache_destructor, NULL, NULL, NULL, KMC_SLAB);
203 ASSERT(znode_hold_cache == NULL);
204 znode_hold_cache = kmem_cache_create("zfs_znode_hold_cache",
205 sizeof (znode_hold_t), 0, zfs_znode_hold_cache_constructor,
206 zfs_znode_hold_cache_destructor, NULL, NULL, NULL, 0);
216 kmem_cache_destroy(znode_cache);
219 if (znode_hold_cache)
220 kmem_cache_destroy(znode_hold_cache);
221 znode_hold_cache = NULL;
225 * The zfs_znode_hold_enter() / zfs_znode_hold_exit() functions are used to
226 * serialize access to a znode and its SA buffer while the object is being
227 * created or destroyed. This kind of locking would normally reside in the
228 * znode itself but in this case that's impossible because the znode and SA
229 * buffer may not yet exist. Therefore the locking is handled externally
230 * with an array of mutexs and AVLs trees which contain per-object locks.
232 * In zfs_znode_hold_enter() a per-object lock is created as needed, inserted
233 * in to the correct AVL tree and finally the per-object lock is held. In
234 * zfs_znode_hold_exit() the process is reversed. The per-object lock is
235 * released, removed from the AVL tree and destroyed if there are no waiters.
237 * This scheme has two important properties:
239 * 1) No memory allocations are performed while holding one of the z_hold_locks.
240 * This ensures evict(), which can be called from direct memory reclaim, will
241 * never block waiting on a z_hold_locks which just happens to have hashed
244 * 2) All locks used to serialize access to an object are per-object and never
245 * shared. This minimizes lock contention without creating a large number
246 * of dedicated locks.
248 * On the downside it does require znode_lock_t structures to be frequently
249 * allocated and freed. However, because these are backed by a kmem cache
250 * and very short lived this cost is minimal.
253 zfs_znode_hold_compare(const void *a, const void *b)
255 const znode_hold_t *zh_a = (const znode_hold_t *)a;
256 const znode_hold_t *zh_b = (const znode_hold_t *)b;
258 return (AVL_CMP(zh_a->zh_obj, zh_b->zh_obj));
262 zfs_znode_held(zfsvfs_t *zfsvfs, uint64_t obj)
264 znode_hold_t *zh, search;
265 int i = ZFS_OBJ_HASH(zfsvfs, obj);
270 mutex_enter(&zfsvfs->z_hold_locks[i]);
271 zh = avl_find(&zfsvfs->z_hold_trees[i], &search, NULL);
272 held = (zh && MUTEX_HELD(&zh->zh_lock)) ? B_TRUE : B_FALSE;
273 mutex_exit(&zfsvfs->z_hold_locks[i]);
278 static znode_hold_t *
279 zfs_znode_hold_enter(zfsvfs_t *zfsvfs, uint64_t obj)
281 znode_hold_t *zh, *zh_new, search;
282 int i = ZFS_OBJ_HASH(zfsvfs, obj);
283 boolean_t found = B_FALSE;
285 zh_new = kmem_cache_alloc(znode_hold_cache, KM_SLEEP);
286 zh_new->zh_obj = obj;
289 mutex_enter(&zfsvfs->z_hold_locks[i]);
290 zh = avl_find(&zfsvfs->z_hold_trees[i], &search, NULL);
291 if (likely(zh == NULL)) {
293 avl_add(&zfsvfs->z_hold_trees[i], zh);
295 ASSERT3U(zh->zh_obj, ==, obj);
298 zfs_refcount_add(&zh->zh_refcount, NULL);
299 mutex_exit(&zfsvfs->z_hold_locks[i]);
302 kmem_cache_free(znode_hold_cache, zh_new);
304 ASSERT(MUTEX_NOT_HELD(&zh->zh_lock));
305 ASSERT3S(zfs_refcount_count(&zh->zh_refcount), >, 0);
306 mutex_enter(&zh->zh_lock);
312 zfs_znode_hold_exit(zfsvfs_t *zfsvfs, znode_hold_t *zh)
314 int i = ZFS_OBJ_HASH(zfsvfs, zh->zh_obj);
315 boolean_t remove = B_FALSE;
317 ASSERT(zfs_znode_held(zfsvfs, zh->zh_obj));
318 ASSERT3S(zfs_refcount_count(&zh->zh_refcount), >, 0);
319 mutex_exit(&zh->zh_lock);
321 mutex_enter(&zfsvfs->z_hold_locks[i]);
322 if (zfs_refcount_remove(&zh->zh_refcount, NULL) == 0) {
323 avl_remove(&zfsvfs->z_hold_trees[i], zh);
326 mutex_exit(&zfsvfs->z_hold_locks[i]);
328 if (remove == B_TRUE)
329 kmem_cache_free(znode_hold_cache, zh);
333 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
334 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
336 ASSERT(zfs_znode_held(zfsvfs, zp->z_id));
338 mutex_enter(&zp->z_lock);
340 ASSERT(zp->z_sa_hdl == NULL);
341 ASSERT(zp->z_acl_cached == NULL);
342 if (sa_hdl == NULL) {
343 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
344 SA_HDL_SHARED, &zp->z_sa_hdl));
346 zp->z_sa_hdl = sa_hdl;
347 sa_set_userp(sa_hdl, zp);
350 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
352 mutex_exit(&zp->z_lock);
356 zfs_znode_dmu_fini(znode_t *zp)
358 ASSERT(zfs_znode_held(ZTOZSB(zp), zp->z_id) || zp->z_unlinked ||
359 RW_WRITE_HELD(&ZTOZSB(zp)->z_teardown_inactive_lock));
361 sa_handle_destroy(zp->z_sa_hdl);
366 * Called by new_inode() to allocate a new inode.
369 zfs_inode_alloc(struct super_block *sb, struct inode **ip)
373 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
380 * Called in multiple places when an inode should be destroyed.
383 zfs_inode_destroy(struct inode *ip)
385 znode_t *zp = ITOZ(ip);
386 zfsvfs_t *zfsvfs = ZTOZSB(zp);
388 mutex_enter(&zfsvfs->z_znodes_lock);
389 if (list_link_active(&zp->z_link_node)) {
390 list_remove(&zfsvfs->z_all_znodes, zp);
391 zfsvfs->z_nr_znodes--;
393 mutex_exit(&zfsvfs->z_znodes_lock);
395 if (zp->z_acl_cached) {
396 zfs_acl_free(zp->z_acl_cached);
397 zp->z_acl_cached = NULL;
400 if (zp->z_xattr_cached) {
401 nvlist_free(zp->z_xattr_cached);
402 zp->z_xattr_cached = NULL;
405 kmem_cache_free(znode_cache, zp);
409 zfs_inode_set_ops(zfsvfs_t *zfsvfs, struct inode *ip)
413 switch (ip->i_mode & S_IFMT) {
415 ip->i_op = &zpl_inode_operations;
416 ip->i_fop = &zpl_file_operations;
417 ip->i_mapping->a_ops = &zpl_address_space_operations;
421 ip->i_op = &zpl_dir_inode_operations;
422 ip->i_fop = &zpl_dir_file_operations;
423 ITOZ(ip)->z_zn_prefetch = B_TRUE;
427 ip->i_op = &zpl_symlink_inode_operations;
431 * rdev is only stored in a SA only for device files.
435 (void) sa_lookup(ITOZ(ip)->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), &rdev,
440 init_special_inode(ip, ip->i_mode, rdev);
441 ip->i_op = &zpl_special_inode_operations;
445 zfs_panic_recover("inode %llu has invalid mode: 0x%x\n",
446 (u_longlong_t)ip->i_ino, ip->i_mode);
448 /* Assume the inode is a file and attempt to continue */
449 ip->i_mode = S_IFREG | 0644;
450 ip->i_op = &zpl_inode_operations;
451 ip->i_fop = &zpl_file_operations;
452 ip->i_mapping->a_ops = &zpl_address_space_operations;
458 zfs_set_inode_flags(znode_t *zp, struct inode *ip)
461 * Linux and Solaris have different sets of file attributes, so we
462 * restrict this conversion to the intersection of the two.
464 #ifdef HAVE_INODE_SET_FLAGS
465 unsigned int flags = 0;
466 if (zp->z_pflags & ZFS_IMMUTABLE)
467 flags |= S_IMMUTABLE;
468 if (zp->z_pflags & ZFS_APPENDONLY)
471 inode_set_flags(ip, flags, S_IMMUTABLE|S_APPEND);
473 if (zp->z_pflags & ZFS_IMMUTABLE)
474 ip->i_flags |= S_IMMUTABLE;
476 ip->i_flags &= ~S_IMMUTABLE;
478 if (zp->z_pflags & ZFS_APPENDONLY)
479 ip->i_flags |= S_APPEND;
481 ip->i_flags &= ~S_APPEND;
486 * Update the embedded inode given the znode. We should work toward
487 * eliminating this function as soon as possible by removing values
488 * which are duplicated between the znode and inode. If the generic
489 * inode has the correct field it should be used, and the ZFS code
490 * updated to access the inode. This can be done incrementally.
493 zfs_inode_update(znode_t *zp)
498 u_longlong_t i_blocks;
504 /* Skip .zfs control nodes which do not exist on disk. */
505 if (zfsctl_is_node(ip))
508 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &blksize, &i_blocks);
510 spin_lock(&ip->i_lock);
511 ip->i_blocks = i_blocks;
512 i_size_write(ip, zp->z_size);
513 spin_unlock(&ip->i_lock);
518 * Construct a znode+inode and initialize.
520 * This does not do a call to dmu_set_user() that is
521 * up to the caller to do, in case you don't want to
525 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
526 dmu_object_type_t obj_type, uint64_t obj, sa_handle_t *hdl)
534 uint64_t z_uid, z_gid;
535 uint64_t atime[2], mtime[2], ctime[2];
536 uint64_t projid = ZFS_DEFAULT_PROJID;
537 sa_bulk_attr_t bulk[11];
540 ASSERT(zfsvfs != NULL);
542 ip = new_inode(zfsvfs->z_sb);
547 ASSERT(zp->z_dirlocks == NULL);
548 ASSERT3P(zp->z_acl_cached, ==, NULL);
549 ASSERT3P(zp->z_xattr_cached, ==, NULL);
553 zp->z_atime_dirty = 0;
555 zp->z_id = db->db_object;
557 zp->z_seq = 0x7A4653;
559 zp->z_is_mapped = B_FALSE;
560 zp->z_is_ctldir = B_FALSE;
561 zp->z_is_stale = B_FALSE;
563 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
565 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
566 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &tmp_gen, 8);
567 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
569 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
570 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
572 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
574 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, &z_uid, 8);
575 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, &z_gid, 8);
576 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
577 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
578 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
580 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || tmp_gen == 0 ||
581 (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
582 (zp->z_pflags & ZFS_PROJID) &&
583 sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) {
585 sa_handle_destroy(zp->z_sa_hdl);
590 zp->z_projid = projid;
591 zp->z_mode = ip->i_mode = mode;
592 ip->i_generation = (uint32_t)tmp_gen;
593 ip->i_blkbits = SPA_MINBLOCKSHIFT;
594 set_nlink(ip, (uint32_t)links);
595 zfs_uid_write(ip, z_uid);
596 zfs_gid_write(ip, z_gid);
597 zfs_set_inode_flags(zp, ip);
599 /* Cache the xattr parent id */
600 if (zp->z_pflags & ZFS_XATTR)
601 zp->z_xattr_parent = parent;
603 ZFS_TIME_DECODE(&ip->i_atime, atime);
604 ZFS_TIME_DECODE(&ip->i_mtime, mtime);
605 ZFS_TIME_DECODE(&ip->i_ctime, ctime);
608 zfs_inode_update(zp);
609 zfs_inode_set_ops(zfsvfs, ip);
612 * The only way insert_inode_locked() can fail is if the ip->i_ino
613 * number is already hashed for this super block. This can never
614 * happen because the inode numbers map 1:1 with the object numbers.
616 * The one exception is rolling back a mounted file system, but in
617 * this case all the active inode are unhashed during the rollback.
619 VERIFY3S(insert_inode_locked(ip), ==, 0);
621 mutex_enter(&zfsvfs->z_znodes_lock);
622 list_insert_tail(&zfsvfs->z_all_znodes, zp);
623 zfsvfs->z_nr_znodes++;
625 mutex_exit(&zfsvfs->z_znodes_lock);
627 unlock_new_inode(ip);
636 * Safely mark an inode dirty. Inodes which are part of a read-only
637 * file system or snapshot may not be dirtied.
640 zfs_mark_inode_dirty(struct inode *ip)
642 zfsvfs_t *zfsvfs = ITOZSB(ip);
644 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
647 mark_inode_dirty(ip);
650 static uint64_t empty_xattr;
651 static uint64_t pad[4];
652 static zfs_acl_phys_t acl_phys;
654 * Create a new DMU object to hold a zfs znode.
656 * IN: dzp - parent directory for new znode
657 * vap - file attributes for new znode
658 * tx - dmu transaction id for zap operations
659 * cr - credentials of caller
661 * IS_ROOT_NODE - new object will be root
662 * IS_XATTR - new object is an attribute
663 * bonuslen - length of bonus buffer
664 * setaclp - File/Dir initial ACL
665 * fuidp - Tracks fuid allocation.
667 * OUT: zpp - allocated znode
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)
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;
678 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
680 inode_timespec_t now;
685 dmu_object_type_t obj_type;
686 sa_bulk_attr_t *sa_attrs;
688 zfs_acl_locator_cb_t locate = { 0 };
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 */
699 gen = dmu_tx_get_txg(tx);
700 dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
704 dnodesize = DNODE_MIN_SIZE;
706 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
708 bonuslen = (obj_type == DMU_OT_SA) ?
709 DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
712 * Create a new DMU object.
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
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));
726 obj = zap_create_norm_dnsize(zfsvfs->z_os,
727 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
728 obj_type, bonuslen, dnodesize, tx);
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));
736 obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
737 DMU_OT_PLAIN_FILE_CONTENTS, 0,
738 obj_type, bonuslen, dnodesize, tx);
742 zh = zfs_znode_hold_enter(zfsvfs, obj);
743 VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
746 * If this is the root, fix up the half-initialized parent pointer
747 * to reference the just-allocated physical data area.
749 if (flag & IS_ROOT_NODE) {
754 * If parent is an xattr, so am I.
756 if (dzp->z_pflags & ZFS_XATTR) {
760 if (zfsvfs->z_use_fuids)
761 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
765 if (S_ISDIR(vap->va_mode)) {
766 size = 2; /* contents ("." and "..") */
770 links = (flag & IS_TMPFILE) ? 0 : 1;
773 if (S_ISBLK(vap->va_mode) || S_ISCHR(vap->va_mode))
777 mode = acl_ids->z_mode;
781 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode)) {
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().
786 if (obj_type != DMU_OT_ZNODE &&
787 dmu_objset_projectquota_enabled(zfsvfs->z_os))
788 pflags |= ZFS_PROJID;
791 * Inherit project ID from parent if required.
793 projid = zfs_inherit_projid(dzp);
794 if (dzp->z_pflags & ZFS_PROJINHERIT)
795 pflags |= ZFS_PROJINHERIT;
799 * No execs denied will be deterimed when zfs_mode_compute() is called.
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);
805 ZFS_TIME_ENCODE(&now, crtime);
806 ZFS_TIME_ENCODE(&now, ctime);
808 if (vap->va_mask & ATTR_ATIME) {
809 ZFS_TIME_ENCODE(&vap->va_atime, atime);
811 ZFS_TIME_ENCODE(&now, atime);
814 if (vap->va_mask & ATTR_MTIME) {
815 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
817 ZFS_TIME_ENCODE(&now, mtime);
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,
825 * Setup the array of attributes to be replaced/set on the new file
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
830 sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
832 if (obj_type == DMU_OT_ZNODE) {
833 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
835 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
837 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
839 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
841 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
843 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
845 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
847 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
850 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
852 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
854 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
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),
862 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
864 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
866 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
868 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
870 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
874 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
876 if (obj_type == DMU_OT_ZNODE) {
877 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
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),
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),
889 if (obj_type == DMU_OT_ZNODE) {
890 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
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);
911 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
913 if (!(flag & IS_ROOT_NODE)) {
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.
922 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, obj,
924 } while (*zpp == NULL);
926 VERIFY(*zpp != NULL);
930 * If we are creating the root node, the "parent" we
931 * passed in is the znode for the root.
935 (*zpp)->z_sa_hdl = sa_hdl;
938 (*zpp)->z_pflags = pflags;
939 (*zpp)->z_mode = ZTOI(*zpp)->i_mode = mode;
940 (*zpp)->z_dnodesize = dnodesize;
941 (*zpp)->z_projid = projid;
943 if (obj_type == DMU_OT_ZNODE ||
944 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
945 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
947 kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
948 zfs_znode_hold_exit(zfsvfs, zh);
952 * Update in-core attributes. It is assumed the caller will be doing an
953 * sa_bulk_update to push the changes out.
956 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
959 boolean_t update_inode = B_FALSE;
961 xoap = xva_getxoptattr(xvap);
964 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
966 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
967 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(ZTOZSB(zp)),
968 ×, sizeof (times), tx);
969 XVA_SET_RTN(xvap, XAT_CREATETIME);
971 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
972 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
974 XVA_SET_RTN(xvap, XAT_READONLY);
976 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
977 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
979 XVA_SET_RTN(xvap, XAT_HIDDEN);
981 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
982 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
984 XVA_SET_RTN(xvap, XAT_SYSTEM);
986 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
987 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
989 XVA_SET_RTN(xvap, XAT_ARCHIVE);
991 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
992 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
994 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
996 update_inode = B_TRUE;
998 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
999 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1001 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1003 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1004 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1006 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1008 update_inode = B_TRUE;
1010 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1011 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1013 XVA_SET_RTN(xvap, XAT_NODUMP);
1015 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1016 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1018 XVA_SET_RTN(xvap, XAT_OPAQUE);
1020 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1021 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1022 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1023 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1025 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1026 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1028 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1030 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1031 zfs_sa_set_scanstamp(zp, xvap, tx);
1032 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1034 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1035 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1037 XVA_SET_RTN(xvap, XAT_REPARSE);
1039 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1040 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1042 XVA_SET_RTN(xvap, XAT_OFFLINE);
1044 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1045 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1047 XVA_SET_RTN(xvap, XAT_SPARSE);
1049 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
1050 ZFS_ATTR_SET(zp, ZFS_PROJINHERIT, xoap->xoa_projinherit,
1052 XVA_SET_RTN(xvap, XAT_PROJINHERIT);
1056 zfs_set_inode_flags(zp, ZTOI(zp));
1060 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1062 dmu_object_info_t doi;
1072 zh = zfs_znode_hold_enter(zfsvfs, obj_num);
1074 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1076 zfs_znode_hold_exit(zfsvfs, zh);
1080 dmu_object_info_from_db(db, &doi);
1081 if (doi.doi_bonus_type != DMU_OT_SA &&
1082 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1083 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1084 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1085 sa_buf_rele(db, NULL);
1086 zfs_znode_hold_exit(zfsvfs, zh);
1087 return (SET_ERROR(EINVAL));
1090 hdl = dmu_buf_get_user(db);
1092 zp = sa_get_userdata(hdl);
1096 * Since "SA" does immediate eviction we
1097 * should never find a sa handle that doesn't
1098 * know about the znode.
1101 ASSERT3P(zp, !=, NULL);
1103 mutex_enter(&zp->z_lock);
1104 ASSERT3U(zp->z_id, ==, obj_num);
1106 * If igrab() returns NULL the VFS has independently
1107 * determined the inode should be evicted and has
1108 * called iput_final() to start the eviction process.
1109 * The SA handle is still valid but because the VFS
1110 * requires that the eviction succeed we must drop
1111 * our locks and references to allow the eviction to
1112 * complete. The zfs_zget() may then be retried.
1114 * This unlikely case could be optimized by registering
1115 * a sops->drop_inode() callback. The callback would
1116 * need to detect the active SA hold thereby informing
1117 * the VFS that this inode should not be evicted.
1119 if (igrab(ZTOI(zp)) == NULL) {
1120 mutex_exit(&zp->z_lock);
1121 sa_buf_rele(db, NULL);
1122 zfs_znode_hold_exit(zfsvfs, zh);
1123 /* inode might need this to finish evict */
1129 mutex_exit(&zp->z_lock);
1130 sa_buf_rele(db, NULL);
1131 zfs_znode_hold_exit(zfsvfs, zh);
1136 * Not found create new znode/vnode but only if file exists.
1138 * There is a small window where zfs_vget() could
1139 * find this object while a file create is still in
1140 * progress. This is checked for in zfs_znode_alloc()
1142 * if zfs_znode_alloc() fails it will drop the hold on the
1145 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1146 doi.doi_bonus_type, obj_num, NULL);
1148 err = SET_ERROR(ENOENT);
1152 zfs_znode_hold_exit(zfsvfs, zh);
1157 zfs_rezget(znode_t *zp)
1159 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1160 dmu_object_info_t doi;
1162 uint64_t obj_num = zp->z_id;
1165 sa_bulk_attr_t bulk[10];
1169 uint64_t z_uid, z_gid;
1170 uint64_t atime[2], mtime[2], ctime[2];
1171 uint64_t projid = ZFS_DEFAULT_PROJID;
1175 * skip ctldir, otherwise they will always get invalidated. This will
1176 * cause funny behaviour for the mounted snapdirs. Especially for
1177 * Linux >= 3.18, d_invalidate will detach the mountpoint and prevent
1178 * anyone automount it again as long as someone is still using the
1181 if (zp->z_is_ctldir)
1184 zh = zfs_znode_hold_enter(zfsvfs, obj_num);
1186 mutex_enter(&zp->z_acl_lock);
1187 if (zp->z_acl_cached) {
1188 zfs_acl_free(zp->z_acl_cached);
1189 zp->z_acl_cached = NULL;
1191 mutex_exit(&zp->z_acl_lock);
1193 rw_enter(&zp->z_xattr_lock, RW_WRITER);
1194 if (zp->z_xattr_cached) {
1195 nvlist_free(zp->z_xattr_cached);
1196 zp->z_xattr_cached = NULL;
1198 rw_exit(&zp->z_xattr_lock);
1200 ASSERT(zp->z_sa_hdl == NULL);
1201 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1203 zfs_znode_hold_exit(zfsvfs, zh);
1207 dmu_object_info_from_db(db, &doi);
1208 if (doi.doi_bonus_type != DMU_OT_SA &&
1209 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1210 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1211 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1212 sa_buf_rele(db, NULL);
1213 zfs_znode_hold_exit(zfsvfs, zh);
1214 return (SET_ERROR(EINVAL));
1217 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1219 /* reload cached values */
1220 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1221 &gen, sizeof (gen));
1222 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1223 &zp->z_size, sizeof (zp->z_size));
1224 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1225 &links, sizeof (links));
1226 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1227 &zp->z_pflags, sizeof (zp->z_pflags));
1228 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1229 &z_uid, sizeof (z_uid));
1230 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1231 &z_gid, sizeof (z_gid));
1232 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1233 &mode, sizeof (mode));
1234 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1236 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
1238 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1241 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1242 zfs_znode_dmu_fini(zp);
1243 zfs_znode_hold_exit(zfsvfs, zh);
1244 return (SET_ERROR(EIO));
1247 if (dmu_objset_projectquota_enabled(zfsvfs->z_os)) {
1248 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs),
1250 if (err != 0 && err != ENOENT) {
1251 zfs_znode_dmu_fini(zp);
1252 zfs_znode_hold_exit(zfsvfs, zh);
1253 return (SET_ERROR(err));
1257 zp->z_projid = projid;
1258 zp->z_mode = ZTOI(zp)->i_mode = mode;
1259 zfs_uid_write(ZTOI(zp), z_uid);
1260 zfs_gid_write(ZTOI(zp), z_gid);
1262 ZFS_TIME_DECODE(&ZTOI(zp)->i_atime, atime);
1263 ZFS_TIME_DECODE(&ZTOI(zp)->i_mtime, mtime);
1264 ZFS_TIME_DECODE(&ZTOI(zp)->i_ctime, ctime);
1266 if (gen != ZTOI(zp)->i_generation) {
1267 zfs_znode_dmu_fini(zp);
1268 zfs_znode_hold_exit(zfsvfs, zh);
1269 return (SET_ERROR(EIO));
1272 set_nlink(ZTOI(zp), (uint32_t)links);
1273 zfs_set_inode_flags(zp, ZTOI(zp));
1275 zp->z_blksz = doi.doi_data_block_size;
1276 zp->z_atime_dirty = 0;
1277 zfs_inode_update(zp);
1280 * If the file has zero links, then it has been unlinked on the send
1281 * side and it must be in the received unlinked set.
1282 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1283 * stale data and to prevent automatical removal of the file in
1284 * zfs_zinactive(). The file will be removed either when it is removed
1285 * on the send side and the next incremental stream is received or
1286 * when the unlinked set gets processed.
1288 zp->z_unlinked = (ZTOI(zp)->i_nlink == 0);
1290 zfs_znode_dmu_fini(zp);
1292 zfs_znode_hold_exit(zfsvfs, zh);
1298 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1300 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1301 objset_t *os = zfsvfs->z_os;
1302 uint64_t obj = zp->z_id;
1303 uint64_t acl_obj = zfs_external_acl(zp);
1306 zh = zfs_znode_hold_enter(zfsvfs, obj);
1308 VERIFY(!zp->z_is_sa);
1309 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1311 VERIFY(0 == dmu_object_free(os, obj, tx));
1312 zfs_znode_dmu_fini(zp);
1313 zfs_znode_hold_exit(zfsvfs, zh);
1317 zfs_zinactive(znode_t *zp)
1319 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1320 uint64_t z_id = zp->z_id;
1323 ASSERT(zp->z_sa_hdl);
1326 * Don't allow a zfs_zget() while were trying to release this znode.
1328 zh = zfs_znode_hold_enter(zfsvfs, z_id);
1330 mutex_enter(&zp->z_lock);
1333 * If this was the last reference to a file with no links, remove
1334 * the file from the file system unless the file system is mounted
1335 * read-only. That can happen, for example, if the file system was
1336 * originally read-write, the file was opened, then unlinked and
1337 * the file system was made read-only before the file was finally
1338 * closed. The file will remain in the unlinked set.
1340 if (zp->z_unlinked) {
1341 ASSERT(!zfsvfs->z_issnap);
1342 if (!zfs_is_readonly(zfsvfs)) {
1343 mutex_exit(&zp->z_lock);
1344 zfs_znode_hold_exit(zfsvfs, zh);
1350 mutex_exit(&zp->z_lock);
1351 zfs_znode_dmu_fini(zp);
1353 zfs_znode_hold_exit(zfsvfs, zh);
1357 zfs_compare_timespec(struct timespec *t1, struct timespec *t2)
1359 if (t1->tv_sec < t2->tv_sec)
1362 if (t1->tv_sec > t2->tv_sec)
1365 return (t1->tv_nsec - t2->tv_nsec);
1369 * Prepare to update znode time stamps.
1371 * IN: zp - znode requiring timestamp update
1372 * flag - ATTR_MTIME, ATTR_CTIME flags
1378 * Note: We don't update atime here, because we rely on Linux VFS to do
1382 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1385 inode_timespec_t now;
1391 if (flag & ATTR_MTIME) {
1392 ZFS_TIME_ENCODE(&now, mtime);
1393 ZFS_TIME_DECODE(&(ZTOI(zp)->i_mtime), mtime);
1394 if (ZTOZSB(zp)->z_use_fuids) {
1395 zp->z_pflags |= (ZFS_ARCHIVE |
1400 if (flag & ATTR_CTIME) {
1401 ZFS_TIME_ENCODE(&now, ctime);
1402 ZFS_TIME_DECODE(&(ZTOI(zp)->i_ctime), ctime);
1403 if (ZTOZSB(zp)->z_use_fuids)
1404 zp->z_pflags |= ZFS_ARCHIVE;
1409 * Grow the block size for a file.
1411 * IN: zp - znode of file to free data in.
1412 * size - requested block size
1413 * tx - open transaction.
1415 * NOTE: this function assumes that the znode is write locked.
1418 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1423 if (size <= zp->z_blksz)
1426 * If the file size is already greater than the current blocksize,
1427 * we will not grow. If there is more than one block in a file,
1428 * the blocksize cannot change.
1430 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1433 error = dmu_object_set_blocksize(ZTOZSB(zp)->z_os, zp->z_id,
1436 if (error == ENOTSUP)
1440 /* What blocksize did we actually get? */
1441 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1445 * Increase the file length
1447 * IN: zp - znode of file to free data in.
1448 * end - new end-of-file
1450 * RETURN: 0 on success, error code on failure
1453 zfs_extend(znode_t *zp, uint64_t end)
1455 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1462 * We will change zp_size, lock the whole file.
1464 lr = rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1467 * Nothing to do if file already at desired length.
1469 if (end <= zp->z_size) {
1473 tx = dmu_tx_create(zfsvfs->z_os);
1474 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1475 zfs_sa_upgrade_txholds(tx, zp);
1476 if (end > zp->z_blksz &&
1477 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1479 * We are growing the file past the current block size.
1481 if (zp->z_blksz > ZTOZSB(zp)->z_max_blksz) {
1483 * File's blocksize is already larger than the
1484 * "recordsize" property. Only let it grow to
1485 * the next power of 2.
1487 ASSERT(!ISP2(zp->z_blksz));
1488 newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1490 newblksz = MIN(end, ZTOZSB(zp)->z_max_blksz);
1492 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1497 error = dmu_tx_assign(tx, TXG_WAIT);
1505 zfs_grow_blocksize(zp, newblksz, tx);
1509 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(ZTOZSB(zp)),
1510 &zp->z_size, sizeof (zp->z_size), tx));
1520 * zfs_zero_partial_page - Modeled after update_pages() but
1521 * with different arguments and semantics for use by zfs_freesp().
1523 * Zeroes a piece of a single page cache entry for zp at offset
1524 * start and length len.
1526 * Caller must acquire a range lock on the file for the region
1527 * being zeroed in order that the ARC and page cache stay in sync.
1530 zfs_zero_partial_page(znode_t *zp, uint64_t start, uint64_t len)
1532 struct address_space *mp = ZTOI(zp)->i_mapping;
1537 ASSERT((start & PAGE_MASK) == ((start + len - 1) & PAGE_MASK));
1539 off = start & (PAGE_SIZE - 1);
1542 pp = find_lock_page(mp, start >> PAGE_SHIFT);
1544 if (mapping_writably_mapped(mp))
1545 flush_dcache_page(pp);
1548 bzero(pb + off, len);
1551 if (mapping_writably_mapped(mp))
1552 flush_dcache_page(pp);
1554 mark_page_accessed(pp);
1555 SetPageUptodate(pp);
1563 * Free space in a file.
1565 * IN: zp - znode of file to free data in.
1566 * off - start of section to free.
1567 * len - length of section to free.
1569 * RETURN: 0 on success, error code on failure
1572 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1574 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1579 * Lock the range being freed.
1581 lr = rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER);
1584 * Nothing to do if file already at desired length.
1586 if (off >= zp->z_size) {
1591 if (off + len > zp->z_size)
1592 len = zp->z_size - off;
1594 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1597 * Zero partial page cache entries. This must be done under a
1598 * range lock in order to keep the ARC and page cache in sync.
1600 if (zp->z_is_mapped) {
1601 loff_t first_page, last_page, page_len;
1602 loff_t first_page_offset, last_page_offset;
1604 /* first possible full page in hole */
1605 first_page = (off + PAGE_SIZE - 1) >> PAGE_SHIFT;
1606 /* last page of hole */
1607 last_page = (off + len) >> PAGE_SHIFT;
1609 /* offset of first_page */
1610 first_page_offset = first_page << PAGE_SHIFT;
1611 /* offset of last_page */
1612 last_page_offset = last_page << PAGE_SHIFT;
1614 /* truncate whole pages */
1615 if (last_page_offset > first_page_offset) {
1616 truncate_inode_pages_range(ZTOI(zp)->i_mapping,
1617 first_page_offset, last_page_offset - 1);
1620 /* truncate sub-page ranges */
1621 if (first_page > last_page) {
1622 /* entire punched area within a single page */
1623 zfs_zero_partial_page(zp, off, len);
1625 /* beginning of punched area at the end of a page */
1626 page_len = first_page_offset - off;
1628 zfs_zero_partial_page(zp, off, page_len);
1630 /* end of punched area at the beginning of a page */
1631 page_len = off + len - last_page_offset;
1633 zfs_zero_partial_page(zp, last_page_offset,
1645 * IN: zp - znode of file to free data in.
1646 * end - new end-of-file.
1648 * RETURN: 0 on success, error code on failure
1651 zfs_trunc(znode_t *zp, uint64_t end)
1653 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1657 sa_bulk_attr_t bulk[2];
1661 * We will change zp_size, lock the whole file.
1663 lr = rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1666 * Nothing to do if file already at desired length.
1668 if (end >= zp->z_size) {
1673 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1679 tx = dmu_tx_create(zfsvfs->z_os);
1680 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1681 zfs_sa_upgrade_txholds(tx, zp);
1682 dmu_tx_mark_netfree(tx);
1683 error = dmu_tx_assign(tx, TXG_WAIT);
1691 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1692 NULL, &zp->z_size, sizeof (zp->z_size));
1695 zp->z_pflags &= ~ZFS_SPARSE;
1696 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1697 NULL, &zp->z_pflags, 8);
1699 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1708 * Free space in a file
1710 * IN: zp - znode of file to free data in.
1711 * off - start of range
1712 * len - end of range (0 => EOF)
1713 * flag - current file open mode flags.
1714 * log - TRUE if this action should be logged
1716 * RETURN: 0 on success, error code on failure
1719 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1722 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1723 zilog_t *zilog = zfsvfs->z_log;
1725 uint64_t mtime[2], ctime[2];
1726 sa_bulk_attr_t bulk[3];
1730 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1731 sizeof (mode))) != 0)
1734 if (off > zp->z_size) {
1735 error = zfs_extend(zp, off+len);
1736 if (error == 0 && log)
1742 error = zfs_trunc(zp, off);
1744 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1745 off + len > zp->z_size)
1746 error = zfs_extend(zp, off+len);
1751 tx = dmu_tx_create(zfsvfs->z_os);
1752 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1753 zfs_sa_upgrade_txholds(tx, zp);
1754 error = dmu_tx_assign(tx, TXG_WAIT);
1760 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1761 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1762 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1763 NULL, &zp->z_pflags, 8);
1764 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime);
1765 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1768 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1772 zfs_inode_update(zp);
1777 * Truncate the page cache - for file truncate operations, use
1778 * the purpose-built API for truncations. For punching operations,
1779 * the truncation is handled under a range lock in zfs_free_range.
1782 truncate_setsize(ZTOI(zp), off);
1787 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1789 struct super_block *sb;
1791 uint64_t moid, obj, sa_obj, version;
1792 uint64_t sense = ZFS_CASE_SENSITIVE;
1798 znode_t *rootzp = NULL;
1801 zfs_acl_ids_t acl_ids;
1804 * First attempt to create master node.
1807 * In an empty objset, there are no blocks to read and thus
1808 * there can be no i/o errors (which we assert below).
1810 moid = MASTER_NODE_OBJ;
1811 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1812 DMU_OT_NONE, 0, tx);
1816 * Set starting attributes.
1818 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1820 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1821 /* For the moment we expect all zpl props to be uint64_ts */
1825 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1826 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1827 name = nvpair_name(elem);
1828 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1832 error = zap_update(os, moid, name, 8, 1, &val, tx);
1835 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1837 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1840 ASSERT(version != 0);
1841 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1844 * Create zap object used for SA attribute registration
1847 if (version >= ZPL_VERSION_SA) {
1848 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1849 DMU_OT_NONE, 0, tx);
1850 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1856 * Create a delete queue.
1858 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1860 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1864 * Create root znode. Create minimal znode/inode/zfsvfs/sb
1865 * to allow zfs_mknode to work.
1867 vattr.va_mask = ATTR_MODE|ATTR_UID|ATTR_GID;
1868 vattr.va_mode = S_IFDIR|0755;
1869 vattr.va_uid = crgetuid(cr);
1870 vattr.va_gid = crgetgid(cr);
1872 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1873 rootzp->z_moved = 0;
1874 rootzp->z_unlinked = 0;
1875 rootzp->z_atime_dirty = 0;
1876 rootzp->z_is_sa = USE_SA(version, os);
1877 rootzp->z_pflags = 0;
1879 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1881 zfsvfs->z_parent = zfsvfs;
1882 zfsvfs->z_version = version;
1883 zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1884 zfsvfs->z_use_sa = USE_SA(version, os);
1885 zfsvfs->z_norm = norm;
1887 sb = kmem_zalloc(sizeof (struct super_block), KM_SLEEP);
1888 sb->s_fs_info = zfsvfs;
1890 ZTOI(rootzp)->i_sb = sb;
1892 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1893 &zfsvfs->z_attr_table);
1898 * Fold case on file systems that are always or sometimes case
1901 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1902 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1904 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1905 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1906 offsetof(znode_t, z_link_node));
1908 size = MIN(1 << (highbit64(zfs_object_mutex_size)-1), ZFS_OBJ_MTX_MAX);
1909 zfsvfs->z_hold_size = size;
1910 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
1912 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
1913 for (i = 0; i != size; i++) {
1914 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
1915 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
1916 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
1919 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1920 cr, NULL, &acl_ids));
1921 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1922 ASSERT3P(zp, ==, rootzp);
1923 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1925 zfs_acl_ids_free(&acl_ids);
1927 atomic_set(&ZTOI(rootzp)->i_count, 0);
1928 sa_handle_destroy(rootzp->z_sa_hdl);
1929 kmem_cache_free(znode_cache, rootzp);
1931 for (i = 0; i != size; i++) {
1932 avl_destroy(&zfsvfs->z_hold_trees[i]);
1933 mutex_destroy(&zfsvfs->z_hold_locks[i]);
1936 mutex_destroy(&zfsvfs->z_znodes_lock);
1938 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
1939 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
1940 kmem_free(sb, sizeof (struct super_block));
1941 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1943 #endif /* _KERNEL */
1946 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1948 uint64_t sa_obj = 0;
1951 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1952 if (error != 0 && error != ENOENT)
1955 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1960 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1961 dmu_buf_t **db, void *tag)
1963 dmu_object_info_t doi;
1966 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
1969 dmu_object_info_from_db(*db, &doi);
1970 if ((doi.doi_bonus_type != DMU_OT_SA &&
1971 doi.doi_bonus_type != DMU_OT_ZNODE) ||
1972 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1973 doi.doi_bonus_size < sizeof (znode_phys_t))) {
1974 sa_buf_rele(*db, tag);
1975 return (SET_ERROR(ENOTSUP));
1978 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
1980 sa_buf_rele(*db, tag);
1988 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
1990 sa_handle_destroy(hdl);
1991 sa_buf_rele(db, tag);
1995 * Given an object number, return its parent object number and whether
1996 * or not the object is an extended attribute directory.
1999 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
2000 uint64_t *pobjp, int *is_xattrdir)
2005 uint64_t parent_mode;
2006 sa_bulk_attr_t bulk[3];
2007 sa_handle_t *sa_hdl;
2012 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
2013 &parent, sizeof (parent));
2014 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
2015 &pflags, sizeof (pflags));
2016 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2017 &mode, sizeof (mode));
2019 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
2023 * When a link is removed its parent pointer is not changed and will
2024 * be invalid. There are two cases where a link is removed but the
2025 * file stays around, when it goes to the delete queue and when there
2026 * are additional links.
2028 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
2032 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
2033 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2037 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
2040 * Extended attributes can be applied to files, directories, etc.
2041 * Otherwise the parent must be a directory.
2043 if (!*is_xattrdir && !S_ISDIR(parent_mode))
2044 return (SET_ERROR(EINVAL));
2052 * Given an object number, return some zpl level statistics
2055 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2058 sa_bulk_attr_t bulk[4];
2061 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2062 &sb->zs_mode, sizeof (sb->zs_mode));
2063 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2064 &sb->zs_gen, sizeof (sb->zs_gen));
2065 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2066 &sb->zs_links, sizeof (sb->zs_links));
2067 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2068 &sb->zs_ctime, sizeof (sb->zs_ctime));
2070 return (sa_bulk_lookup(hdl, bulk, count));
2074 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2075 sa_attr_type_t *sa_table, char *buf, int len)
2077 sa_handle_t *sa_hdl;
2078 sa_handle_t *prevhdl = NULL;
2079 dmu_buf_t *prevdb = NULL;
2080 dmu_buf_t *sa_db = NULL;
2081 char *path = buf + len - 1;
2087 uint64_t deleteq_obj;
2088 VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ,
2089 ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj));
2090 error = zap_lookup_int(osp, deleteq_obj, obj);
2093 } else if (error != ENOENT) {
2100 char component[MAXNAMELEN + 2];
2102 int is_xattrdir = 0;
2105 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2107 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2108 &is_xattrdir)) != 0)
2119 (void) sprintf(component + 1, "<xattrdir>");
2121 error = zap_value_search(osp, pobj, obj,
2122 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2127 complen = strlen(component);
2129 ASSERT(path >= buf);
2130 bcopy(component, path, complen);
2133 if (sa_hdl != hdl) {
2137 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2145 if (sa_hdl != NULL && sa_hdl != hdl) {
2146 ASSERT(sa_db != NULL);
2147 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2151 (void) memmove(buf, path, buf + len - path);
2157 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2159 sa_attr_type_t *sa_table;
2164 error = zfs_sa_setup(osp, &sa_table);
2168 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2172 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2174 zfs_release_sa_handle(hdl, db, FTAG);
2179 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2182 char *path = buf + len - 1;
2183 sa_attr_type_t *sa_table;
2190 error = zfs_sa_setup(osp, &sa_table);
2194 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2198 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2200 zfs_release_sa_handle(hdl, db, FTAG);
2204 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2206 zfs_release_sa_handle(hdl, db, FTAG);
2210 #if defined(_KERNEL)
2211 EXPORT_SYMBOL(zfs_create_fs);
2212 EXPORT_SYMBOL(zfs_obj_to_path);
2215 module_param(zfs_object_mutex_size, uint, 0644);
2216 MODULE_PARM_DESC(zfs_object_mutex_size, "Size of znode hold array");