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 https://opensource.org/licenses/CDDL-1.0.
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>
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>
55 #include <sys/dmu_objset.h>
56 #include <sys/dmu_tx.h>
57 #include <sys/zfs_refcount.h>
60 #include <sys/zfs_znode.h>
62 #include <sys/zfs_sa.h>
63 #include <sys/zfs_stat.h>
66 #include "zfs_comutil.h"
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.
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
;
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.
83 static int zfs_unlink_suspend_progress
= 0;
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.
92 zfs_rangelock_cb(zfs_locked_range_t
*new, void *arg
)
97 * If in append mode, convert to writer and lock starting at the
98 * current end of file.
100 if (new->lr_type
== RL_APPEND
) {
101 new->lr_offset
= zp
->z_size
;
102 new->lr_type
= RL_WRITER
;
106 * If we need to grow the block size then lock the whole file range.
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
)) {
112 new->lr_length
= UINT64_MAX
;
117 zfs_znode_cache_constructor(void *buf
, void *arg
, int kmflags
)
119 (void) arg
, (void) kmflags
;
122 inode_init_once(ZTOI(zp
));
123 list_link_init(&zp
->z_link_node
);
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
);
131 zfs_rangelock_init(&zp
->z_rangelock
, zfs_rangelock_cb
, zp
);
133 zp
->z_dirlocks
= NULL
;
134 zp
->z_acl_cached
= NULL
;
135 zp
->z_xattr_cached
= NULL
;
136 zp
->z_xattr_parent
= 0;
137 zp
->z_sync_writes_cnt
= 0;
138 zp
->z_async_writes_cnt
= 0;
144 zfs_znode_cache_destructor(void *buf
, void *arg
)
149 ASSERT(!list_link_active(&zp
->z_link_node
));
150 mutex_destroy(&zp
->z_lock
);
151 rw_destroy(&zp
->z_parent_lock
);
152 rw_destroy(&zp
->z_name_lock
);
153 mutex_destroy(&zp
->z_acl_lock
);
154 rw_destroy(&zp
->z_xattr_lock
);
155 zfs_rangelock_fini(&zp
->z_rangelock
);
157 ASSERT3P(zp
->z_dirlocks
, ==, NULL
);
158 ASSERT3P(zp
->z_acl_cached
, ==, NULL
);
159 ASSERT3P(zp
->z_xattr_cached
, ==, NULL
);
161 ASSERT0(atomic_load_32(&zp
->z_sync_writes_cnt
));
162 ASSERT0(atomic_load_32(&zp
->z_async_writes_cnt
));
166 zfs_znode_hold_cache_constructor(void *buf
, void *arg
, int kmflags
)
168 (void) arg
, (void) kmflags
;
169 znode_hold_t
*zh
= buf
;
171 mutex_init(&zh
->zh_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
178 zfs_znode_hold_cache_destructor(void *buf
, void *arg
)
181 znode_hold_t
*zh
= buf
;
183 mutex_destroy(&zh
->zh_lock
);
190 * Initialize zcache. The KMC_SLAB hint is used in order that it be
191 * backed by kmalloc() when on the Linux slab in order that any
192 * wait_on_bit() operations on the related inode operate properly.
194 ASSERT(znode_cache
== NULL
);
195 znode_cache
= kmem_cache_create("zfs_znode_cache",
196 sizeof (znode_t
), 0, zfs_znode_cache_constructor
,
197 zfs_znode_cache_destructor
, NULL
, NULL
, NULL
, KMC_SLAB
);
199 ASSERT(znode_hold_cache
== NULL
);
200 znode_hold_cache
= kmem_cache_create("zfs_znode_hold_cache",
201 sizeof (znode_hold_t
), 0, zfs_znode_hold_cache_constructor
,
202 zfs_znode_hold_cache_destructor
, NULL
, NULL
, NULL
, 0);
212 kmem_cache_destroy(znode_cache
);
215 if (znode_hold_cache
)
216 kmem_cache_destroy(znode_hold_cache
);
217 znode_hold_cache
= NULL
;
221 * The zfs_znode_hold_enter() / zfs_znode_hold_exit() functions are used to
222 * serialize access to a znode and its SA buffer while the object is being
223 * created or destroyed. This kind of locking would normally reside in the
224 * znode itself but in this case that's impossible because the znode and SA
225 * buffer may not yet exist. Therefore the locking is handled externally
226 * with an array of mutexes and AVLs trees which contain per-object locks.
228 * In zfs_znode_hold_enter() a per-object lock is created as needed, inserted
229 * in to the correct AVL tree and finally the per-object lock is held. In
230 * zfs_znode_hold_exit() the process is reversed. The per-object lock is
231 * released, removed from the AVL tree and destroyed if there are no waiters.
233 * This scheme has two important properties:
235 * 1) No memory allocations are performed while holding one of the z_hold_locks.
236 * This ensures evict(), which can be called from direct memory reclaim, will
237 * never block waiting on a z_hold_locks which just happens to have hashed
240 * 2) All locks used to serialize access to an object are per-object and never
241 * shared. This minimizes lock contention without creating a large number
242 * of dedicated locks.
244 * On the downside it does require znode_lock_t structures to be frequently
245 * allocated and freed. However, because these are backed by a kmem cache
246 * and very short lived this cost is minimal.
249 zfs_znode_hold_compare(const void *a
, const void *b
)
251 const znode_hold_t
*zh_a
= (const znode_hold_t
*)a
;
252 const znode_hold_t
*zh_b
= (const znode_hold_t
*)b
;
254 return (TREE_CMP(zh_a
->zh_obj
, zh_b
->zh_obj
));
257 static boolean_t __maybe_unused
258 zfs_znode_held(zfsvfs_t
*zfsvfs
, uint64_t obj
)
260 znode_hold_t
*zh
, search
;
261 int i
= ZFS_OBJ_HASH(zfsvfs
, obj
);
266 mutex_enter(&zfsvfs
->z_hold_locks
[i
]);
267 zh
= avl_find(&zfsvfs
->z_hold_trees
[i
], &search
, NULL
);
268 held
= (zh
&& MUTEX_HELD(&zh
->zh_lock
)) ? B_TRUE
: B_FALSE
;
269 mutex_exit(&zfsvfs
->z_hold_locks
[i
]);
275 zfs_znode_hold_enter(zfsvfs_t
*zfsvfs
, uint64_t obj
)
277 znode_hold_t
*zh
, *zh_new
, search
;
278 int i
= ZFS_OBJ_HASH(zfsvfs
, obj
);
279 boolean_t found
= B_FALSE
;
281 zh_new
= kmem_cache_alloc(znode_hold_cache
, KM_SLEEP
);
284 mutex_enter(&zfsvfs
->z_hold_locks
[i
]);
285 zh
= avl_find(&zfsvfs
->z_hold_trees
[i
], &search
, NULL
);
286 if (likely(zh
== NULL
)) {
289 avl_add(&zfsvfs
->z_hold_trees
[i
], zh
);
291 ASSERT3U(zh
->zh_obj
, ==, obj
);
295 ASSERT3S(zh
->zh_refcount
, >, 0);
296 mutex_exit(&zfsvfs
->z_hold_locks
[i
]);
299 kmem_cache_free(znode_hold_cache
, zh_new
);
301 ASSERT(MUTEX_NOT_HELD(&zh
->zh_lock
));
302 mutex_enter(&zh
->zh_lock
);
308 zfs_znode_hold_exit(zfsvfs_t
*zfsvfs
, znode_hold_t
*zh
)
310 int i
= ZFS_OBJ_HASH(zfsvfs
, zh
->zh_obj
);
311 boolean_t remove
= B_FALSE
;
313 ASSERT(zfs_znode_held(zfsvfs
, zh
->zh_obj
));
314 mutex_exit(&zh
->zh_lock
);
316 mutex_enter(&zfsvfs
->z_hold_locks
[i
]);
317 ASSERT3S(zh
->zh_refcount
, >, 0);
318 if (--zh
->zh_refcount
== 0) {
319 avl_remove(&zfsvfs
->z_hold_trees
[i
], zh
);
322 mutex_exit(&zfsvfs
->z_hold_locks
[i
]);
324 if (remove
== B_TRUE
)
325 kmem_cache_free(znode_hold_cache
, zh
);
329 zfs_cmpldev(uint64_t dev
)
335 zfs_znode_sa_init(zfsvfs_t
*zfsvfs
, znode_t
*zp
,
336 dmu_buf_t
*db
, dmu_object_type_t obj_type
, sa_handle_t
*sa_hdl
)
338 ASSERT(zfs_znode_held(zfsvfs
, zp
->z_id
));
340 mutex_enter(&zp
->z_lock
);
342 ASSERT(zp
->z_sa_hdl
== NULL
);
343 ASSERT(zp
->z_acl_cached
== NULL
);
344 if (sa_hdl
== NULL
) {
345 VERIFY(0 == sa_handle_get_from_db(zfsvfs
->z_os
, db
, zp
,
346 SA_HDL_SHARED
, &zp
->z_sa_hdl
));
348 zp
->z_sa_hdl
= sa_hdl
;
349 sa_set_userp(sa_hdl
, zp
);
352 zp
->z_is_sa
= (obj_type
== DMU_OT_SA
) ? B_TRUE
: B_FALSE
;
354 mutex_exit(&zp
->z_lock
);
358 zfs_znode_dmu_fini(znode_t
*zp
)
360 ASSERT(zfs_znode_held(ZTOZSB(zp
), zp
->z_id
) ||
361 RW_WRITE_HELD(&ZTOZSB(zp
)->z_teardown_inactive_lock
));
363 sa_handle_destroy(zp
->z_sa_hdl
);
368 * Called by new_inode() to allocate a new inode.
371 zfs_inode_alloc(struct super_block
*sb
, struct inode
**ip
)
375 zp
= kmem_cache_alloc(znode_cache
, KM_SLEEP
);
382 * Called in multiple places when an inode should be destroyed.
385 zfs_inode_destroy(struct inode
*ip
)
387 znode_t
*zp
= ITOZ(ip
);
388 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
390 mutex_enter(&zfsvfs
->z_znodes_lock
);
391 if (list_link_active(&zp
->z_link_node
)) {
392 list_remove(&zfsvfs
->z_all_znodes
, zp
);
394 mutex_exit(&zfsvfs
->z_znodes_lock
);
396 if (zp
->z_acl_cached
) {
397 zfs_acl_free(zp
->z_acl_cached
);
398 zp
->z_acl_cached
= NULL
;
401 if (zp
->z_xattr_cached
) {
402 nvlist_free(zp
->z_xattr_cached
);
403 zp
->z_xattr_cached
= NULL
;
406 kmem_cache_free(znode_cache
, zp
);
410 zfs_inode_set_ops(zfsvfs_t
*zfsvfs
, struct inode
*ip
)
414 switch (ip
->i_mode
& S_IFMT
) {
416 ip
->i_op
= &zpl_inode_operations
;
417 #ifdef HAVE_VFS_FILE_OPERATIONS_EXTEND
418 ip
->i_fop
= &zpl_file_operations
.kabi_fops
;
420 ip
->i_fop
= &zpl_file_operations
;
422 ip
->i_mapping
->a_ops
= &zpl_address_space_operations
;
426 #ifdef HAVE_RENAME2_OPERATIONS_WRAPPER
427 ip
->i_flags
|= S_IOPS_WRAPPER
;
428 ip
->i_op
= &zpl_dir_inode_operations
.ops
;
430 ip
->i_op
= &zpl_dir_inode_operations
;
432 ip
->i_fop
= &zpl_dir_file_operations
;
433 ITOZ(ip
)->z_zn_prefetch
= B_TRUE
;
437 ip
->i_op
= &zpl_symlink_inode_operations
;
441 * rdev is only stored in a SA only for device files.
445 (void) sa_lookup(ITOZ(ip
)->z_sa_hdl
, SA_ZPL_RDEV(zfsvfs
), &rdev
,
450 init_special_inode(ip
, ip
->i_mode
, rdev
);
451 ip
->i_op
= &zpl_special_inode_operations
;
455 zfs_panic_recover("inode %llu has invalid mode: 0x%x\n",
456 (u_longlong_t
)ip
->i_ino
, ip
->i_mode
);
458 /* Assume the inode is a file and attempt to continue */
459 ip
->i_mode
= S_IFREG
| 0644;
460 ip
->i_op
= &zpl_inode_operations
;
461 #ifdef HAVE_VFS_FILE_OPERATIONS_EXTEND
462 ip
->i_fop
= &zpl_file_operations
.kabi_fops
;
464 ip
->i_fop
= &zpl_file_operations
;
466 ip
->i_mapping
->a_ops
= &zpl_address_space_operations
;
472 zfs_set_inode_flags(znode_t
*zp
, struct inode
*ip
)
475 * Linux and Solaris have different sets of file attributes, so we
476 * restrict this conversion to the intersection of the two.
478 #ifdef HAVE_INODE_SET_FLAGS
479 unsigned int flags
= 0;
480 if (zp
->z_pflags
& ZFS_IMMUTABLE
)
481 flags
|= S_IMMUTABLE
;
482 if (zp
->z_pflags
& ZFS_APPENDONLY
)
485 inode_set_flags(ip
, flags
, S_IMMUTABLE
|S_APPEND
);
487 if (zp
->z_pflags
& ZFS_IMMUTABLE
)
488 ip
->i_flags
|= S_IMMUTABLE
;
490 ip
->i_flags
&= ~S_IMMUTABLE
;
492 if (zp
->z_pflags
& ZFS_APPENDONLY
)
493 ip
->i_flags
|= S_APPEND
;
495 ip
->i_flags
&= ~S_APPEND
;
500 * Update the embedded inode given the znode.
503 zfs_znode_update_vfs(znode_t
*zp
)
507 u_longlong_t i_blocks
;
512 /* Skip .zfs control nodes which do not exist on disk. */
513 if (zfsctl_is_node(ip
))
516 dmu_object_size_from_db(sa_get_db(zp
->z_sa_hdl
), &blksize
, &i_blocks
);
518 spin_lock(&ip
->i_lock
);
519 ip
->i_mode
= zp
->z_mode
;
520 ip
->i_blocks
= i_blocks
;
521 i_size_write(ip
, zp
->z_size
);
522 spin_unlock(&ip
->i_lock
);
527 * Construct a znode+inode and initialize.
529 * This does not do a call to dmu_set_user() that is
530 * up to the caller to do, in case you don't want to
534 zfs_znode_alloc(zfsvfs_t
*zfsvfs
, dmu_buf_t
*db
, int blksz
,
535 dmu_object_type_t obj_type
, sa_handle_t
*hdl
)
543 uint64_t z_uid
, z_gid
;
544 uint64_t atime
[2], mtime
[2], ctime
[2], btime
[2];
545 inode_timespec_t tmp_ts
;
546 uint64_t projid
= ZFS_DEFAULT_PROJID
;
547 sa_bulk_attr_t bulk
[12];
550 ASSERT(zfsvfs
!= NULL
);
552 ip
= new_inode(zfsvfs
->z_sb
);
557 ASSERT(zp
->z_dirlocks
== NULL
);
558 ASSERT3P(zp
->z_acl_cached
, ==, NULL
);
559 ASSERT3P(zp
->z_xattr_cached
, ==, NULL
);
560 zp
->z_unlinked
= B_FALSE
;
561 zp
->z_atime_dirty
= B_FALSE
;
562 #if !defined(HAVE_FILEMAP_RANGE_HAS_PAGE)
563 zp
->z_is_mapped
= B_FALSE
;
565 zp
->z_is_ctldir
= B_FALSE
;
566 zp
->z_suspended
= B_FALSE
;
569 zp
->z_id
= db
->db_object
;
571 zp
->z_seq
= 0x7A4653;
573 zp
->z_sync_writes_cnt
= 0;
574 zp
->z_async_writes_cnt
= 0;
576 zfs_znode_sa_init(zfsvfs
, zp
, db
, obj_type
, hdl
);
578 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
), NULL
, &mode
, 8);
579 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GEN(zfsvfs
), NULL
, &tmp_gen
, 8);
580 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
), NULL
,
582 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zfsvfs
), NULL
, &links
, 8);
583 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
585 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_PARENT(zfsvfs
), NULL
,
587 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zfsvfs
), NULL
, &z_uid
, 8);
588 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zfsvfs
), NULL
, &z_gid
, 8);
589 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
, &atime
, 16);
590 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
591 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
592 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CRTIME(zfsvfs
), NULL
, &btime
, 16);
594 if (sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
) != 0 || tmp_gen
== 0 ||
595 (dmu_objset_projectquota_enabled(zfsvfs
->z_os
) &&
596 (zp
->z_pflags
& ZFS_PROJID
) &&
597 sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PROJID(zfsvfs
), &projid
, 8) != 0)) {
599 sa_handle_destroy(zp
->z_sa_hdl
);
604 zp
->z_projid
= projid
;
605 zp
->z_mode
= ip
->i_mode
= mode
;
606 ip
->i_generation
= (uint32_t)tmp_gen
;
607 ip
->i_blkbits
= SPA_MINBLOCKSHIFT
;
608 set_nlink(ip
, (uint32_t)links
);
609 zfs_uid_write(ip
, z_uid
);
610 zfs_gid_write(ip
, z_gid
);
611 zfs_set_inode_flags(zp
, ip
);
613 /* Cache the xattr parent id */
614 if (zp
->z_pflags
& ZFS_XATTR
)
615 zp
->z_xattr_parent
= parent
;
617 ZFS_TIME_DECODE(&tmp_ts
, atime
);
618 zpl_inode_set_atime_to_ts(ip
, tmp_ts
);
619 ZFS_TIME_DECODE(&tmp_ts
, mtime
);
620 zpl_inode_set_mtime_to_ts(ip
, tmp_ts
);
621 ZFS_TIME_DECODE(&tmp_ts
, ctime
);
622 zpl_inode_set_ctime_to_ts(ip
, tmp_ts
);
623 ZFS_TIME_DECODE(&zp
->z_btime
, btime
);
625 ip
->i_ino
= zp
->z_id
;
626 zfs_znode_update_vfs(zp
);
627 zfs_inode_set_ops(zfsvfs
, ip
);
630 * The only way insert_inode_locked() can fail is if the ip->i_ino
631 * number is already hashed for this super block. This can never
632 * happen because the inode numbers map 1:1 with the object numbers.
634 * Exceptions include rolling back a mounted file system, either
635 * from the zfs rollback or zfs recv command.
637 * Active inodes are unhashed during the rollback, but since zrele
638 * can happen asynchronously, we can't guarantee they've been
639 * unhashed. This can cause hash collisions in unlinked drain
640 * processing so do not hash unlinked znodes.
643 VERIFY3S(insert_inode_locked(ip
), ==, 0);
645 mutex_enter(&zfsvfs
->z_znodes_lock
);
646 list_insert_tail(&zfsvfs
->z_all_znodes
, zp
);
647 mutex_exit(&zfsvfs
->z_znodes_lock
);
650 unlock_new_inode(ip
);
659 * Safely mark an inode dirty. Inodes which are part of a read-only
660 * file system or snapshot may not be dirtied.
663 zfs_mark_inode_dirty(struct inode
*ip
)
665 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
667 if (zfs_is_readonly(zfsvfs
) || dmu_objset_is_snapshot(zfsvfs
->z_os
))
670 mark_inode_dirty(ip
);
673 static uint64_t empty_xattr
;
674 static uint64_t pad
[4];
675 static zfs_acl_phys_t acl_phys
;
677 * Create a new DMU object to hold a zfs znode.
679 * IN: dzp - parent directory for new znode
680 * vap - file attributes for new znode
681 * tx - dmu transaction id for zap operations
682 * cr - credentials of caller
684 * IS_ROOT_NODE - new object will be root
685 * IS_TMPFILE - new object is of O_TMPFILE
686 * IS_XATTR - new object is an attribute
687 * acl_ids - ACL related attributes
689 * OUT: zpp - allocated znode (set to dzp if IS_ROOT_NODE)
693 zfs_mknode(znode_t
*dzp
, vattr_t
*vap
, dmu_tx_t
*tx
, cred_t
*cr
,
694 uint_t flag
, znode_t
**zpp
, zfs_acl_ids_t
*acl_ids
)
696 uint64_t crtime
[2], atime
[2], mtime
[2], ctime
[2];
697 uint64_t mode
, size
, links
, parent
, pflags
;
698 uint64_t projid
= ZFS_DEFAULT_PROJID
;
700 zfsvfs_t
*zfsvfs
= ZTOZSB(dzp
);
702 inode_timespec_t now
;
707 dmu_object_type_t obj_type
;
708 sa_bulk_attr_t
*sa_attrs
;
710 zfs_acl_locator_cb_t locate
= { 0 };
713 if (zfsvfs
->z_replay
) {
714 obj
= vap
->va_nodeid
;
715 now
= vap
->va_ctime
; /* see zfs_replay_create() */
716 gen
= vap
->va_nblocks
; /* ditto */
717 dnodesize
= vap
->va_fsid
; /* ditto */
721 gen
= dmu_tx_get_txg(tx
);
722 dnodesize
= dmu_objset_dnodesize(zfsvfs
->z_os
);
726 dnodesize
= DNODE_MIN_SIZE
;
728 obj_type
= zfsvfs
->z_use_sa
? DMU_OT_SA
: DMU_OT_ZNODE
;
730 bonuslen
= (obj_type
== DMU_OT_SA
) ?
731 DN_BONUS_SIZE(dnodesize
) : ZFS_OLD_ZNODE_PHYS_SIZE
;
734 * Create a new DMU object.
737 * There's currently no mechanism for pre-reading the blocks that will
738 * be needed to allocate a new object, so we accept the small chance
739 * that there will be an i/o error and we will fail one of the
742 if (S_ISDIR(vap
->va_mode
)) {
743 if (zfsvfs
->z_replay
) {
744 VERIFY0(zap_create_claim_norm_dnsize(zfsvfs
->z_os
, obj
,
745 zfsvfs
->z_norm
, DMU_OT_DIRECTORY_CONTENTS
,
746 obj_type
, bonuslen
, dnodesize
, tx
));
748 obj
= zap_create_norm_dnsize(zfsvfs
->z_os
,
749 zfsvfs
->z_norm
, DMU_OT_DIRECTORY_CONTENTS
,
750 obj_type
, bonuslen
, dnodesize
, tx
);
753 if (zfsvfs
->z_replay
) {
754 VERIFY0(dmu_object_claim_dnsize(zfsvfs
->z_os
, obj
,
755 DMU_OT_PLAIN_FILE_CONTENTS
, 0,
756 obj_type
, bonuslen
, dnodesize
, tx
));
758 obj
= dmu_object_alloc_dnsize(zfsvfs
->z_os
,
759 DMU_OT_PLAIN_FILE_CONTENTS
, 0,
760 obj_type
, bonuslen
, dnodesize
, tx
);
764 zh
= zfs_znode_hold_enter(zfsvfs
, obj
);
765 VERIFY0(sa_buf_hold(zfsvfs
->z_os
, obj
, NULL
, &db
));
768 * If this is the root, fix up the half-initialized parent pointer
769 * to reference the just-allocated physical data area.
771 if (flag
& IS_ROOT_NODE
) {
776 * If parent is an xattr, so am I.
778 if (dzp
->z_pflags
& ZFS_XATTR
) {
782 if (zfsvfs
->z_use_fuids
)
783 pflags
= ZFS_ARCHIVE
| ZFS_AV_MODIFIED
;
787 if (S_ISDIR(vap
->va_mode
)) {
788 size
= 2; /* contents ("." and "..") */
792 links
= (flag
& IS_TMPFILE
) ? 0 : 1;
795 if (S_ISBLK(vap
->va_mode
) || S_ISCHR(vap
->va_mode
))
799 mode
= acl_ids
->z_mode
;
803 if (S_ISREG(vap
->va_mode
) || S_ISDIR(vap
->va_mode
)) {
805 * With ZFS_PROJID flag, we can easily know whether there is
806 * project ID stored on disk or not. See zfs_space_delta_cb().
808 if (obj_type
!= DMU_OT_ZNODE
&&
809 dmu_objset_projectquota_enabled(zfsvfs
->z_os
))
810 pflags
|= ZFS_PROJID
;
813 * Inherit project ID from parent if required.
815 projid
= zfs_inherit_projid(dzp
);
816 if (dzp
->z_pflags
& ZFS_PROJINHERIT
)
817 pflags
|= ZFS_PROJINHERIT
;
821 * No execs denied will be determined when zfs_mode_compute() is called.
823 pflags
|= acl_ids
->z_aclp
->z_hints
&
824 (ZFS_ACL_TRIVIAL
|ZFS_INHERIT_ACE
|ZFS_ACL_AUTO_INHERIT
|
825 ZFS_ACL_DEFAULTED
|ZFS_ACL_PROTECTED
);
827 ZFS_TIME_ENCODE(&now
, crtime
);
828 ZFS_TIME_ENCODE(&now
, ctime
);
830 if (vap
->va_mask
& ATTR_ATIME
) {
831 ZFS_TIME_ENCODE(&vap
->va_atime
, atime
);
833 ZFS_TIME_ENCODE(&now
, atime
);
836 if (vap
->va_mask
& ATTR_MTIME
) {
837 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
839 ZFS_TIME_ENCODE(&now
, mtime
);
842 /* Now add in all of the "SA" attributes */
843 VERIFY(0 == sa_handle_get_from_db(zfsvfs
->z_os
, db
, NULL
, SA_HDL_SHARED
,
847 * Setup the array of attributes to be replaced/set on the new file
849 * order for DMU_OT_ZNODE is critical since it needs to be constructed
850 * in the old znode_phys_t format. Don't change this ordering
852 sa_attrs
= kmem_alloc(sizeof (sa_bulk_attr_t
) * ZPL_END
, KM_SLEEP
);
854 if (obj_type
== DMU_OT_ZNODE
) {
855 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_ATIME(zfsvfs
),
857 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_MTIME(zfsvfs
),
859 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_CTIME(zfsvfs
),
861 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_CRTIME(zfsvfs
),
863 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_GEN(zfsvfs
),
865 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_MODE(zfsvfs
),
867 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_SIZE(zfsvfs
),
869 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_PARENT(zfsvfs
),
872 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_MODE(zfsvfs
),
874 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_SIZE(zfsvfs
),
876 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_GEN(zfsvfs
),
878 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_UID(zfsvfs
),
879 NULL
, &acl_ids
->z_fuid
, 8);
880 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_GID(zfsvfs
),
881 NULL
, &acl_ids
->z_fgid
, 8);
882 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_PARENT(zfsvfs
),
884 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_FLAGS(zfsvfs
),
886 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_ATIME(zfsvfs
),
888 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_MTIME(zfsvfs
),
890 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_CTIME(zfsvfs
),
892 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_CRTIME(zfsvfs
),
896 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_LINKS(zfsvfs
), NULL
, &links
, 8);
898 if (obj_type
== DMU_OT_ZNODE
) {
899 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_XATTR(zfsvfs
), NULL
,
901 } else if (dmu_objset_projectquota_enabled(zfsvfs
->z_os
) &&
902 pflags
& ZFS_PROJID
) {
903 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_PROJID(zfsvfs
),
906 if (obj_type
== DMU_OT_ZNODE
||
907 (S_ISBLK(vap
->va_mode
) || S_ISCHR(vap
->va_mode
))) {
908 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_RDEV(zfsvfs
),
911 if (obj_type
== DMU_OT_ZNODE
) {
912 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_FLAGS(zfsvfs
),
914 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_UID(zfsvfs
), NULL
,
915 &acl_ids
->z_fuid
, 8);
916 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_GID(zfsvfs
), NULL
,
917 &acl_ids
->z_fgid
, 8);
918 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_PAD(zfsvfs
), NULL
, pad
,
919 sizeof (uint64_t) * 4);
920 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_ZNODE_ACL(zfsvfs
), NULL
,
921 &acl_phys
, sizeof (zfs_acl_phys_t
));
922 } else if (acl_ids
->z_aclp
->z_version
>= ZFS_ACL_VERSION_FUID
) {
923 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_DACL_COUNT(zfsvfs
), NULL
,
924 &acl_ids
->z_aclp
->z_acl_count
, 8);
925 locate
.cb_aclp
= acl_ids
->z_aclp
;
926 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_DACL_ACES(zfsvfs
),
927 zfs_acl_data_locator
, &locate
,
928 acl_ids
->z_aclp
->z_acl_bytes
);
929 mode
= zfs_mode_compute(mode
, acl_ids
->z_aclp
, &pflags
,
930 acl_ids
->z_fuid
, acl_ids
->z_fgid
);
933 VERIFY(sa_replace_all_by_template(sa_hdl
, sa_attrs
, cnt
, tx
) == 0);
935 if (!(flag
& IS_ROOT_NODE
)) {
937 * The call to zfs_znode_alloc() may fail if memory is low
938 * via the call path: alloc_inode() -> inode_init_always() ->
939 * security_inode_alloc() -> inode_alloc_security(). Since
940 * the existing code is written such that zfs_mknode() can
941 * not fail retry until sufficient memory has been reclaimed.
944 *zpp
= zfs_znode_alloc(zfsvfs
, db
, 0, obj_type
, sa_hdl
);
945 } while (*zpp
== NULL
);
947 VERIFY(*zpp
!= NULL
);
951 * If we are creating the root node, the "parent" we
952 * passed in is the znode for the root.
956 (*zpp
)->z_sa_hdl
= sa_hdl
;
959 (*zpp
)->z_pflags
= pflags
;
960 (*zpp
)->z_mode
= ZTOI(*zpp
)->i_mode
= mode
;
961 (*zpp
)->z_dnodesize
= dnodesize
;
962 (*zpp
)->z_projid
= projid
;
964 if (obj_type
== DMU_OT_ZNODE
||
965 acl_ids
->z_aclp
->z_version
< ZFS_ACL_VERSION_FUID
) {
966 VERIFY0(zfs_aclset_common(*zpp
, acl_ids
->z_aclp
, cr
, tx
));
968 kmem_free(sa_attrs
, sizeof (sa_bulk_attr_t
) * ZPL_END
);
969 zfs_znode_hold_exit(zfsvfs
, zh
);
973 * Update in-core attributes. It is assumed the caller will be doing an
974 * sa_bulk_update to push the changes out.
977 zfs_xvattr_set(znode_t
*zp
, xvattr_t
*xvap
, dmu_tx_t
*tx
)
980 boolean_t update_inode
= B_FALSE
;
982 xoap
= xva_getxoptattr(xvap
);
985 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
987 ZFS_TIME_ENCODE(&xoap
->xoa_createtime
, times
);
988 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_CRTIME(ZTOZSB(zp
)),
989 ×
, sizeof (times
), tx
);
990 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
992 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
993 ZFS_ATTR_SET(zp
, ZFS_READONLY
, xoap
->xoa_readonly
,
995 XVA_SET_RTN(xvap
, XAT_READONLY
);
997 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
998 ZFS_ATTR_SET(zp
, ZFS_HIDDEN
, xoap
->xoa_hidden
,
1000 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
1002 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
1003 ZFS_ATTR_SET(zp
, ZFS_SYSTEM
, xoap
->xoa_system
,
1005 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
1007 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
1008 ZFS_ATTR_SET(zp
, ZFS_ARCHIVE
, xoap
->xoa_archive
,
1010 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
1012 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
1013 ZFS_ATTR_SET(zp
, ZFS_IMMUTABLE
, xoap
->xoa_immutable
,
1015 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
1017 update_inode
= B_TRUE
;
1019 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
1020 ZFS_ATTR_SET(zp
, ZFS_NOUNLINK
, xoap
->xoa_nounlink
,
1022 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
1024 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
1025 ZFS_ATTR_SET(zp
, ZFS_APPENDONLY
, xoap
->xoa_appendonly
,
1027 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
1029 update_inode
= B_TRUE
;
1031 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
1032 ZFS_ATTR_SET(zp
, ZFS_NODUMP
, xoap
->xoa_nodump
,
1034 XVA_SET_RTN(xvap
, XAT_NODUMP
);
1036 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
1037 ZFS_ATTR_SET(zp
, ZFS_OPAQUE
, xoap
->xoa_opaque
,
1039 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
1041 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
1042 ZFS_ATTR_SET(zp
, ZFS_AV_QUARANTINED
,
1043 xoap
->xoa_av_quarantined
, zp
->z_pflags
, tx
);
1044 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
1046 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
1047 ZFS_ATTR_SET(zp
, ZFS_AV_MODIFIED
, xoap
->xoa_av_modified
,
1049 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
1051 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
)) {
1052 zfs_sa_set_scanstamp(zp
, xvap
, tx
);
1053 XVA_SET_RTN(xvap
, XAT_AV_SCANSTAMP
);
1055 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
1056 ZFS_ATTR_SET(zp
, ZFS_REPARSE
, xoap
->xoa_reparse
,
1058 XVA_SET_RTN(xvap
, XAT_REPARSE
);
1060 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
1061 ZFS_ATTR_SET(zp
, ZFS_OFFLINE
, xoap
->xoa_offline
,
1063 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
1065 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
1066 ZFS_ATTR_SET(zp
, ZFS_SPARSE
, xoap
->xoa_sparse
,
1068 XVA_SET_RTN(xvap
, XAT_SPARSE
);
1070 if (XVA_ISSET_REQ(xvap
, XAT_PROJINHERIT
)) {
1071 ZFS_ATTR_SET(zp
, ZFS_PROJINHERIT
, xoap
->xoa_projinherit
,
1073 XVA_SET_RTN(xvap
, XAT_PROJINHERIT
);
1077 zfs_set_inode_flags(zp
, ZTOI(zp
));
1081 zfs_zget(zfsvfs_t
*zfsvfs
, uint64_t obj_num
, znode_t
**zpp
)
1083 dmu_object_info_t doi
;
1093 zh
= zfs_znode_hold_enter(zfsvfs
, obj_num
);
1095 err
= sa_buf_hold(zfsvfs
->z_os
, obj_num
, NULL
, &db
);
1097 zfs_znode_hold_exit(zfsvfs
, zh
);
1101 dmu_object_info_from_db(db
, &doi
);
1102 if (doi
.doi_bonus_type
!= DMU_OT_SA
&&
1103 (doi
.doi_bonus_type
!= DMU_OT_ZNODE
||
1104 (doi
.doi_bonus_type
== DMU_OT_ZNODE
&&
1105 doi
.doi_bonus_size
< sizeof (znode_phys_t
)))) {
1106 sa_buf_rele(db
, NULL
);
1107 zfs_znode_hold_exit(zfsvfs
, zh
);
1108 return (SET_ERROR(EINVAL
));
1111 hdl
= dmu_buf_get_user(db
);
1113 zp
= sa_get_userdata(hdl
);
1117 * Since "SA" does immediate eviction we
1118 * should never find a sa handle that doesn't
1119 * know about the znode.
1122 ASSERT3P(zp
, !=, NULL
);
1124 mutex_enter(&zp
->z_lock
);
1125 ASSERT3U(zp
->z_id
, ==, obj_num
);
1127 * If zp->z_unlinked is set, the znode is already marked
1128 * for deletion and should not be discovered. Check this
1129 * after checking igrab() due to fsetxattr() & O_TMPFILE.
1131 * If igrab() returns NULL the VFS has independently
1132 * determined the inode should be evicted and has
1133 * called iput_final() to start the eviction process.
1134 * The SA handle is still valid but because the VFS
1135 * requires that the eviction succeed we must drop
1136 * our locks and references to allow the eviction to
1137 * complete. The zfs_zget() may then be retried.
1139 * This unlikely case could be optimized by registering
1140 * a sops->drop_inode() callback. The callback would
1141 * need to detect the active SA hold thereby informing
1142 * the VFS that this inode should not be evicted.
1144 if (igrab(ZTOI(zp
)) == NULL
) {
1146 err
= SET_ERROR(ENOENT
);
1148 err
= SET_ERROR(EAGAIN
);
1154 mutex_exit(&zp
->z_lock
);
1155 sa_buf_rele(db
, NULL
);
1156 zfs_znode_hold_exit(zfsvfs
, zh
);
1158 if (err
== EAGAIN
) {
1159 /* inode might need this to finish evict */
1167 * Not found create new znode/vnode but only if file exists.
1169 * There is a small window where zfs_vget() could
1170 * find this object while a file create is still in
1171 * progress. This is checked for in zfs_znode_alloc()
1173 * if zfs_znode_alloc() fails it will drop the hold on the
1176 zp
= zfs_znode_alloc(zfsvfs
, db
, doi
.doi_data_block_size
,
1177 doi
.doi_bonus_type
, NULL
);
1179 err
= SET_ERROR(ENOENT
);
1183 zfs_znode_hold_exit(zfsvfs
, zh
);
1188 zfs_rezget(znode_t
*zp
)
1190 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
1191 dmu_object_info_t doi
;
1193 uint64_t obj_num
= zp
->z_id
;
1196 sa_bulk_attr_t bulk
[11];
1200 uint64_t z_uid
, z_gid
;
1201 uint64_t atime
[2], mtime
[2], ctime
[2], btime
[2];
1202 inode_timespec_t tmp_ts
;
1203 uint64_t projid
= ZFS_DEFAULT_PROJID
;
1207 * skip ctldir, otherwise they will always get invalidated. This will
1208 * cause funny behaviour for the mounted snapdirs. Especially for
1209 * Linux >= 3.18, d_invalidate will detach the mountpoint and prevent
1210 * anyone automount it again as long as someone is still using the
1213 if (zp
->z_is_ctldir
)
1216 zh
= zfs_znode_hold_enter(zfsvfs
, obj_num
);
1218 mutex_enter(&zp
->z_acl_lock
);
1219 if (zp
->z_acl_cached
) {
1220 zfs_acl_free(zp
->z_acl_cached
);
1221 zp
->z_acl_cached
= NULL
;
1223 mutex_exit(&zp
->z_acl_lock
);
1225 rw_enter(&zp
->z_xattr_lock
, RW_WRITER
);
1226 if (zp
->z_xattr_cached
) {
1227 nvlist_free(zp
->z_xattr_cached
);
1228 zp
->z_xattr_cached
= NULL
;
1230 rw_exit(&zp
->z_xattr_lock
);
1232 ASSERT(zp
->z_sa_hdl
== NULL
);
1233 err
= sa_buf_hold(zfsvfs
->z_os
, obj_num
, NULL
, &db
);
1235 zfs_znode_hold_exit(zfsvfs
, zh
);
1239 dmu_object_info_from_db(db
, &doi
);
1240 if (doi
.doi_bonus_type
!= DMU_OT_SA
&&
1241 (doi
.doi_bonus_type
!= DMU_OT_ZNODE
||
1242 (doi
.doi_bonus_type
== DMU_OT_ZNODE
&&
1243 doi
.doi_bonus_size
< sizeof (znode_phys_t
)))) {
1244 sa_buf_rele(db
, NULL
);
1245 zfs_znode_hold_exit(zfsvfs
, zh
);
1246 return (SET_ERROR(EINVAL
));
1249 zfs_znode_sa_init(zfsvfs
, zp
, db
, doi
.doi_bonus_type
, NULL
);
1251 /* reload cached values */
1252 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GEN(zfsvfs
), NULL
,
1253 &gen
, sizeof (gen
));
1254 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
), NULL
,
1255 &zp
->z_size
, sizeof (zp
->z_size
));
1256 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zfsvfs
), NULL
,
1257 &links
, sizeof (links
));
1258 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
1259 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
1260 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zfsvfs
), NULL
,
1261 &z_uid
, sizeof (z_uid
));
1262 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zfsvfs
), NULL
,
1263 &z_gid
, sizeof (z_gid
));
1264 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
), NULL
,
1265 &mode
, sizeof (mode
));
1266 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
,
1268 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
1270 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
1272 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CRTIME(zfsvfs
), NULL
, &btime
, 16);
1274 if (sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) {
1275 zfs_znode_dmu_fini(zp
);
1276 zfs_znode_hold_exit(zfsvfs
, zh
);
1277 return (SET_ERROR(EIO
));
1280 if (dmu_objset_projectquota_enabled(zfsvfs
->z_os
)) {
1281 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PROJID(zfsvfs
),
1283 if (err
!= 0 && err
!= ENOENT
) {
1284 zfs_znode_dmu_fini(zp
);
1285 zfs_znode_hold_exit(zfsvfs
, zh
);
1286 return (SET_ERROR(err
));
1290 zp
->z_projid
= projid
;
1291 zp
->z_mode
= ZTOI(zp
)->i_mode
= mode
;
1292 zfs_uid_write(ZTOI(zp
), z_uid
);
1293 zfs_gid_write(ZTOI(zp
), z_gid
);
1295 ZFS_TIME_DECODE(&tmp_ts
, atime
);
1296 zpl_inode_set_atime_to_ts(ZTOI(zp
), tmp_ts
);
1297 ZFS_TIME_DECODE(&tmp_ts
, mtime
);
1298 zpl_inode_set_mtime_to_ts(ZTOI(zp
), tmp_ts
);
1299 ZFS_TIME_DECODE(&tmp_ts
, ctime
);
1300 zpl_inode_set_ctime_to_ts(ZTOI(zp
), tmp_ts
);
1301 ZFS_TIME_DECODE(&zp
->z_btime
, btime
);
1303 if ((uint32_t)gen
!= ZTOI(zp
)->i_generation
) {
1304 zfs_znode_dmu_fini(zp
);
1305 zfs_znode_hold_exit(zfsvfs
, zh
);
1306 return (SET_ERROR(EIO
));
1309 set_nlink(ZTOI(zp
), (uint32_t)links
);
1310 zfs_set_inode_flags(zp
, ZTOI(zp
));
1312 zp
->z_blksz
= doi
.doi_data_block_size
;
1313 zp
->z_atime_dirty
= B_FALSE
;
1314 zfs_znode_update_vfs(zp
);
1317 * If the file has zero links, then it has been unlinked on the send
1318 * side and it must be in the received unlinked set.
1319 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1320 * stale data and to prevent automatic removal of the file in
1321 * zfs_zinactive(). The file will be removed either when it is removed
1322 * on the send side and the next incremental stream is received or
1323 * when the unlinked set gets processed.
1325 zp
->z_unlinked
= (ZTOI(zp
)->i_nlink
== 0);
1327 zfs_znode_dmu_fini(zp
);
1329 zfs_znode_hold_exit(zfsvfs
, zh
);
1335 zfs_znode_delete(znode_t
*zp
, dmu_tx_t
*tx
)
1337 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
1338 objset_t
*os
= zfsvfs
->z_os
;
1339 uint64_t obj
= zp
->z_id
;
1340 uint64_t acl_obj
= zfs_external_acl(zp
);
1343 zh
= zfs_znode_hold_enter(zfsvfs
, obj
);
1345 VERIFY(!zp
->z_is_sa
);
1346 VERIFY(0 == dmu_object_free(os
, acl_obj
, tx
));
1348 VERIFY(0 == dmu_object_free(os
, obj
, tx
));
1349 zfs_znode_dmu_fini(zp
);
1350 zfs_znode_hold_exit(zfsvfs
, zh
);
1354 zfs_zinactive(znode_t
*zp
)
1356 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
1357 uint64_t z_id
= zp
->z_id
;
1360 ASSERT(zp
->z_sa_hdl
);
1363 * Don't allow a zfs_zget() while were trying to release this znode.
1365 zh
= zfs_znode_hold_enter(zfsvfs
, z_id
);
1367 mutex_enter(&zp
->z_lock
);
1370 * If this was the last reference to a file with no links, remove
1371 * the file from the file system unless the file system is mounted
1372 * read-only. That can happen, for example, if the file system was
1373 * originally read-write, the file was opened, then unlinked and
1374 * the file system was made read-only before the file was finally
1375 * closed. The file will remain in the unlinked set.
1377 if (zp
->z_unlinked
) {
1378 ASSERT(!zfsvfs
->z_issnap
);
1379 if (!zfs_is_readonly(zfsvfs
) && !zfs_unlink_suspend_progress
) {
1380 mutex_exit(&zp
->z_lock
);
1381 zfs_znode_hold_exit(zfsvfs
, zh
);
1387 mutex_exit(&zp
->z_lock
);
1388 zfs_znode_dmu_fini(zp
);
1390 zfs_znode_hold_exit(zfsvfs
, zh
);
1393 #if defined(HAVE_INODE_TIMESPEC64_TIMES)
1394 #define zfs_compare_timespec timespec64_compare
1396 #define zfs_compare_timespec timespec_compare
1400 * Determine whether the znode's atime must be updated. The logic mostly
1401 * duplicates the Linux kernel's relatime_need_update() functionality.
1402 * This function is only called if the underlying filesystem actually has
1403 * atime updates enabled.
1406 zfs_relatime_need_update(const struct inode
*ip
)
1408 inode_timespec_t now
, tmp_atime
, tmp_ts
;
1411 tmp_atime
= zpl_inode_get_atime(ip
);
1413 * In relatime mode, only update the atime if the previous atime
1414 * is earlier than either the ctime or mtime or if at least a day
1415 * has passed since the last update of atime.
1417 tmp_ts
= zpl_inode_get_mtime(ip
);
1418 if (zfs_compare_timespec(&tmp_ts
, &tmp_atime
) >= 0)
1421 tmp_ts
= zpl_inode_get_ctime(ip
);
1422 if (zfs_compare_timespec(&tmp_ts
, &tmp_atime
) >= 0)
1425 if ((hrtime_t
)now
.tv_sec
- (hrtime_t
)tmp_atime
.tv_sec
>= 24*60*60)
1432 * Prepare to update znode time stamps.
1434 * IN: zp - znode requiring timestamp update
1435 * flag - ATTR_MTIME, ATTR_CTIME flags
1441 * Note: We don't update atime here, because we rely on Linux VFS to do
1445 zfs_tstamp_update_setup(znode_t
*zp
, uint_t flag
, uint64_t mtime
[2],
1448 inode_timespec_t now
, tmp_ts
;
1454 if (flag
& ATTR_MTIME
) {
1455 ZFS_TIME_ENCODE(&now
, mtime
);
1456 ZFS_TIME_DECODE(&tmp_ts
, mtime
);
1457 zpl_inode_set_mtime_to_ts(ZTOI(zp
), tmp_ts
);
1458 if (ZTOZSB(zp
)->z_use_fuids
) {
1459 zp
->z_pflags
|= (ZFS_ARCHIVE
|
1464 if (flag
& ATTR_CTIME
) {
1465 ZFS_TIME_ENCODE(&now
, ctime
);
1466 ZFS_TIME_DECODE(&tmp_ts
, ctime
);
1467 zpl_inode_set_ctime_to_ts(ZTOI(zp
), tmp_ts
);
1468 if (ZTOZSB(zp
)->z_use_fuids
)
1469 zp
->z_pflags
|= ZFS_ARCHIVE
;
1474 * Grow the block size for a file.
1476 * IN: zp - znode of file to free data in.
1477 * size - requested block size
1478 * tx - open transaction.
1480 * NOTE: this function assumes that the znode is write locked.
1483 zfs_grow_blocksize(znode_t
*zp
, uint64_t size
, dmu_tx_t
*tx
)
1488 if (size
<= zp
->z_blksz
)
1491 * If the file size is already greater than the current blocksize,
1492 * we will not grow. If there is more than one block in a file,
1493 * the blocksize cannot change.
1495 if (zp
->z_blksz
&& zp
->z_size
> zp
->z_blksz
)
1498 error
= dmu_object_set_blocksize(ZTOZSB(zp
)->z_os
, zp
->z_id
,
1501 if (error
== ENOTSUP
)
1505 /* What blocksize did we actually get? */
1506 dmu_object_size_from_db(sa_get_db(zp
->z_sa_hdl
), &zp
->z_blksz
, &dummy
);
1510 * Increase the file length
1512 * IN: zp - znode of file to free data in.
1513 * end - new end-of-file
1515 * RETURN: 0 on success, error code on failure
1518 zfs_extend(znode_t
*zp
, uint64_t end
)
1520 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
1522 zfs_locked_range_t
*lr
;
1527 * We will change zp_size, lock the whole file.
1529 lr
= zfs_rangelock_enter(&zp
->z_rangelock
, 0, UINT64_MAX
, RL_WRITER
);
1532 * Nothing to do if file already at desired length.
1534 if (end
<= zp
->z_size
) {
1535 zfs_rangelock_exit(lr
);
1538 tx
= dmu_tx_create(zfsvfs
->z_os
);
1539 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1540 zfs_sa_upgrade_txholds(tx
, zp
);
1541 if (end
> zp
->z_blksz
&&
1542 (!ISP2(zp
->z_blksz
) || zp
->z_blksz
< zfsvfs
->z_max_blksz
)) {
1544 * We are growing the file past the current block size.
1546 if (zp
->z_blksz
> ZTOZSB(zp
)->z_max_blksz
) {
1548 * File's blocksize is already larger than the
1549 * "recordsize" property. Only let it grow to
1550 * the next power of 2.
1552 ASSERT(!ISP2(zp
->z_blksz
));
1553 newblksz
= MIN(end
, 1 << highbit64(zp
->z_blksz
));
1555 newblksz
= MIN(end
, ZTOZSB(zp
)->z_max_blksz
);
1557 dmu_tx_hold_write(tx
, zp
->z_id
, 0, newblksz
);
1562 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1565 zfs_rangelock_exit(lr
);
1570 zfs_grow_blocksize(zp
, newblksz
, tx
);
1574 VERIFY(0 == sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(ZTOZSB(zp
)),
1575 &zp
->z_size
, sizeof (zp
->z_size
), tx
));
1577 zfs_rangelock_exit(lr
);
1585 * zfs_zero_partial_page - Modeled after update_pages() but
1586 * with different arguments and semantics for use by zfs_freesp().
1588 * Zeroes a piece of a single page cache entry for zp at offset
1589 * start and length len.
1591 * Caller must acquire a range lock on the file for the region
1592 * being zeroed in order that the ARC and page cache stay in sync.
1595 zfs_zero_partial_page(znode_t
*zp
, uint64_t start
, uint64_t len
)
1597 struct address_space
*mp
= ZTOI(zp
)->i_mapping
;
1602 ASSERT((start
& PAGE_MASK
) == ((start
+ len
- 1) & PAGE_MASK
));
1604 off
= start
& (PAGE_SIZE
- 1);
1607 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
1609 if (mapping_writably_mapped(mp
))
1610 flush_dcache_page(pp
);
1613 memset(pb
+ off
, 0, len
);
1616 if (mapping_writably_mapped(mp
))
1617 flush_dcache_page(pp
);
1619 mark_page_accessed(pp
);
1620 SetPageUptodate(pp
);
1628 * Free space in a file.
1630 * IN: zp - znode of file to free data in.
1631 * off - start of section to free.
1632 * len - length of section to free.
1634 * RETURN: 0 on success, error code on failure
1637 zfs_free_range(znode_t
*zp
, uint64_t off
, uint64_t len
)
1639 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
1640 zfs_locked_range_t
*lr
;
1644 * Lock the range being freed.
1646 lr
= zfs_rangelock_enter(&zp
->z_rangelock
, off
, len
, RL_WRITER
);
1649 * Nothing to do if file already at desired length.
1651 if (off
>= zp
->z_size
) {
1652 zfs_rangelock_exit(lr
);
1656 if (off
+ len
> zp
->z_size
)
1657 len
= zp
->z_size
- off
;
1659 error
= dmu_free_long_range(zfsvfs
->z_os
, zp
->z_id
, off
, len
);
1662 * Zero partial page cache entries. This must be done under a
1663 * range lock in order to keep the ARC and page cache in sync.
1665 if (zn_has_cached_data(zp
, off
, off
+ len
- 1)) {
1666 loff_t first_page
, last_page
, page_len
;
1667 loff_t first_page_offset
, last_page_offset
;
1669 /* first possible full page in hole */
1670 first_page
= (off
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1671 /* last page of hole */
1672 last_page
= (off
+ len
) >> PAGE_SHIFT
;
1674 /* offset of first_page */
1675 first_page_offset
= first_page
<< PAGE_SHIFT
;
1676 /* offset of last_page */
1677 last_page_offset
= last_page
<< PAGE_SHIFT
;
1679 /* truncate whole pages */
1680 if (last_page_offset
> first_page_offset
) {
1681 truncate_inode_pages_range(ZTOI(zp
)->i_mapping
,
1682 first_page_offset
, last_page_offset
- 1);
1685 /* truncate sub-page ranges */
1686 if (first_page
> last_page
) {
1687 /* entire punched area within a single page */
1688 zfs_zero_partial_page(zp
, off
, len
);
1690 /* beginning of punched area at the end of a page */
1691 page_len
= first_page_offset
- off
;
1693 zfs_zero_partial_page(zp
, off
, page_len
);
1695 /* end of punched area at the beginning of a page */
1696 page_len
= off
+ len
- last_page_offset
;
1698 zfs_zero_partial_page(zp
, last_page_offset
,
1702 zfs_rangelock_exit(lr
);
1710 * IN: zp - znode of file to free data in.
1711 * end - new end-of-file.
1713 * RETURN: 0 on success, error code on failure
1716 zfs_trunc(znode_t
*zp
, uint64_t end
)
1718 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
1720 zfs_locked_range_t
*lr
;
1722 sa_bulk_attr_t bulk
[2];
1726 * We will change zp_size, lock the whole file.
1728 lr
= zfs_rangelock_enter(&zp
->z_rangelock
, 0, UINT64_MAX
, RL_WRITER
);
1731 * Nothing to do if file already at desired length.
1733 if (end
>= zp
->z_size
) {
1734 zfs_rangelock_exit(lr
);
1738 error
= dmu_free_long_range(zfsvfs
->z_os
, zp
->z_id
, end
,
1741 zfs_rangelock_exit(lr
);
1744 tx
= dmu_tx_create(zfsvfs
->z_os
);
1745 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1746 zfs_sa_upgrade_txholds(tx
, zp
);
1747 dmu_tx_mark_netfree(tx
);
1748 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1751 zfs_rangelock_exit(lr
);
1756 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
),
1757 NULL
, &zp
->z_size
, sizeof (zp
->z_size
));
1760 zp
->z_pflags
&= ~ZFS_SPARSE
;
1761 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
),
1762 NULL
, &zp
->z_pflags
, 8);
1764 VERIFY(sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
) == 0);
1767 zfs_rangelock_exit(lr
);
1773 * Free space in a file
1775 * IN: zp - znode of file to free data in.
1776 * off - start of range
1777 * len - end of range (0 => EOF)
1778 * flag - current file open mode flags.
1779 * log - TRUE if this action should be logged
1781 * RETURN: 0 on success, error code on failure
1784 zfs_freesp(znode_t
*zp
, uint64_t off
, uint64_t len
, int flag
, boolean_t log
)
1787 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
1788 zilog_t
*zilog
= zfsvfs
->z_log
;
1790 uint64_t mtime
[2], ctime
[2];
1791 sa_bulk_attr_t bulk
[3];
1795 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_MODE(zfsvfs
), &mode
,
1796 sizeof (mode
))) != 0)
1799 if (off
> zp
->z_size
) {
1800 error
= zfs_extend(zp
, off
+len
);
1801 if (error
== 0 && log
)
1807 error
= zfs_trunc(zp
, off
);
1809 if ((error
= zfs_free_range(zp
, off
, len
)) == 0 &&
1810 off
+ len
> zp
->z_size
)
1811 error
= zfs_extend(zp
, off
+len
);
1816 tx
= dmu_tx_create(zfsvfs
->z_os
);
1817 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1818 zfs_sa_upgrade_txholds(tx
, zp
);
1819 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1825 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, mtime
, 16);
1826 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, ctime
, 16);
1827 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
),
1828 NULL
, &zp
->z_pflags
, 8);
1829 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
);
1830 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
1833 zfs_log_truncate(zilog
, tx
, TX_TRUNCATE
, zp
, off
, len
);
1837 zfs_znode_update_vfs(zp
);
1842 * Truncate the page cache - for file truncate operations, use
1843 * the purpose-built API for truncations. For punching operations,
1844 * the truncation is handled under a range lock in zfs_free_range.
1847 truncate_setsize(ZTOI(zp
), off
);
1852 zfs_create_fs(objset_t
*os
, cred_t
*cr
, nvlist_t
*zplprops
, dmu_tx_t
*tx
)
1854 struct super_block
*sb
;
1856 uint64_t moid
, obj
, sa_obj
, version
;
1857 uint64_t sense
= ZFS_CASE_SENSITIVE
;
1863 znode_t
*rootzp
= NULL
;
1866 zfs_acl_ids_t acl_ids
;
1869 * First attempt to create master node.
1872 * In an empty objset, there are no blocks to read and thus
1873 * there can be no i/o errors (which we assert below).
1875 moid
= MASTER_NODE_OBJ
;
1876 error
= zap_create_claim(os
, moid
, DMU_OT_MASTER_NODE
,
1877 DMU_OT_NONE
, 0, tx
);
1881 * Set starting attributes.
1883 version
= zfs_zpl_version_map(spa_version(dmu_objset_spa(os
)));
1885 while ((elem
= nvlist_next_nvpair(zplprops
, elem
)) != NULL
) {
1886 /* For the moment we expect all zpl props to be uint64_ts */
1890 ASSERT(nvpair_type(elem
) == DATA_TYPE_UINT64
);
1891 VERIFY(nvpair_value_uint64(elem
, &val
) == 0);
1892 name
= nvpair_name(elem
);
1893 if (strcmp(name
, zfs_prop_to_name(ZFS_PROP_VERSION
)) == 0) {
1897 error
= zap_update(os
, moid
, name
, 8, 1, &val
, tx
);
1900 if (strcmp(name
, zfs_prop_to_name(ZFS_PROP_NORMALIZE
)) == 0)
1902 else if (strcmp(name
, zfs_prop_to_name(ZFS_PROP_CASE
)) == 0)
1905 ASSERT(version
!= 0);
1906 error
= zap_update(os
, moid
, ZPL_VERSION_STR
, 8, 1, &version
, tx
);
1910 * Create zap object used for SA attribute registration
1913 if (version
>= ZPL_VERSION_SA
) {
1914 sa_obj
= zap_create(os
, DMU_OT_SA_MASTER_NODE
,
1915 DMU_OT_NONE
, 0, tx
);
1916 error
= zap_add(os
, moid
, ZFS_SA_ATTRS
, 8, 1, &sa_obj
, tx
);
1922 * Create a delete queue.
1924 obj
= zap_create(os
, DMU_OT_UNLINKED_SET
, DMU_OT_NONE
, 0, tx
);
1926 error
= zap_add(os
, moid
, ZFS_UNLINKED_SET
, 8, 1, &obj
, tx
);
1930 * Create root znode. Create minimal znode/inode/zfsvfs/sb
1931 * to allow zfs_mknode to work.
1933 vattr
.va_mask
= ATTR_MODE
|ATTR_UID
|ATTR_GID
;
1934 vattr
.va_mode
= S_IFDIR
|0755;
1935 vattr
.va_uid
= crgetuid(cr
);
1936 vattr
.va_gid
= crgetgid(cr
);
1938 rootzp
= kmem_cache_alloc(znode_cache
, KM_SLEEP
);
1939 rootzp
->z_unlinked
= B_FALSE
;
1940 rootzp
->z_atime_dirty
= B_FALSE
;
1941 rootzp
->z_is_sa
= USE_SA(version
, os
);
1942 rootzp
->z_pflags
= 0;
1944 zfsvfs
= kmem_zalloc(sizeof (zfsvfs_t
), KM_SLEEP
);
1946 zfsvfs
->z_parent
= zfsvfs
;
1947 zfsvfs
->z_version
= version
;
1948 zfsvfs
->z_use_fuids
= USE_FUIDS(version
, os
);
1949 zfsvfs
->z_use_sa
= USE_SA(version
, os
);
1950 zfsvfs
->z_norm
= norm
;
1952 sb
= kmem_zalloc(sizeof (struct super_block
), KM_SLEEP
);
1953 sb
->s_fs_info
= zfsvfs
;
1955 ZTOI(rootzp
)->i_sb
= sb
;
1957 error
= sa_setup(os
, sa_obj
, zfs_attr_table
, ZPL_END
,
1958 &zfsvfs
->z_attr_table
);
1963 * Fold case on file systems that are always or sometimes case
1966 if (sense
== ZFS_CASE_INSENSITIVE
|| sense
== ZFS_CASE_MIXED
)
1967 zfsvfs
->z_norm
|= U8_TEXTPREP_TOUPPER
;
1969 mutex_init(&zfsvfs
->z_znodes_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1970 list_create(&zfsvfs
->z_all_znodes
, sizeof (znode_t
),
1971 offsetof(znode_t
, z_link_node
));
1973 size
= MIN(1 << (highbit64(zfs_object_mutex_size
)-1), ZFS_OBJ_MTX_MAX
);
1974 zfsvfs
->z_hold_size
= size
;
1975 zfsvfs
->z_hold_trees
= vmem_zalloc(sizeof (avl_tree_t
) * size
,
1977 zfsvfs
->z_hold_locks
= vmem_zalloc(sizeof (kmutex_t
) * size
, KM_SLEEP
);
1978 for (i
= 0; i
!= size
; i
++) {
1979 avl_create(&zfsvfs
->z_hold_trees
[i
], zfs_znode_hold_compare
,
1980 sizeof (znode_hold_t
), offsetof(znode_hold_t
, zh_node
));
1981 mutex_init(&zfsvfs
->z_hold_locks
[i
], NULL
, MUTEX_DEFAULT
, NULL
);
1984 VERIFY(0 == zfs_acl_ids_create(rootzp
, IS_ROOT_NODE
, &vattr
,
1985 cr
, NULL
, &acl_ids
, zfs_init_idmap
));
1986 zfs_mknode(rootzp
, &vattr
, tx
, cr
, IS_ROOT_NODE
, &zp
, &acl_ids
);
1987 ASSERT3P(zp
, ==, rootzp
);
1988 error
= zap_add(os
, moid
, ZFS_ROOT_OBJ
, 8, 1, &rootzp
->z_id
, tx
);
1990 zfs_acl_ids_free(&acl_ids
);
1992 atomic_set(&ZTOI(rootzp
)->i_count
, 0);
1993 sa_handle_destroy(rootzp
->z_sa_hdl
);
1994 kmem_cache_free(znode_cache
, rootzp
);
1996 for (i
= 0; i
!= size
; i
++) {
1997 avl_destroy(&zfsvfs
->z_hold_trees
[i
]);
1998 mutex_destroy(&zfsvfs
->z_hold_locks
[i
]);
2001 mutex_destroy(&zfsvfs
->z_znodes_lock
);
2003 vmem_free(zfsvfs
->z_hold_trees
, sizeof (avl_tree_t
) * size
);
2004 vmem_free(zfsvfs
->z_hold_locks
, sizeof (kmutex_t
) * size
);
2005 kmem_free(sb
, sizeof (struct super_block
));
2006 kmem_free(zfsvfs
, sizeof (zfsvfs_t
));
2008 #endif /* _KERNEL */
2011 zfs_sa_setup(objset_t
*osp
, sa_attr_type_t
**sa_table
)
2013 uint64_t sa_obj
= 0;
2016 error
= zap_lookup(osp
, MASTER_NODE_OBJ
, ZFS_SA_ATTRS
, 8, 1, &sa_obj
);
2017 if (error
!= 0 && error
!= ENOENT
)
2020 error
= sa_setup(osp
, sa_obj
, zfs_attr_table
, ZPL_END
, sa_table
);
2025 zfs_grab_sa_handle(objset_t
*osp
, uint64_t obj
, sa_handle_t
**hdlp
,
2026 dmu_buf_t
**db
, const void *tag
)
2028 dmu_object_info_t doi
;
2031 if ((error
= sa_buf_hold(osp
, obj
, tag
, db
)) != 0)
2034 dmu_object_info_from_db(*db
, &doi
);
2035 if ((doi
.doi_bonus_type
!= DMU_OT_SA
&&
2036 doi
.doi_bonus_type
!= DMU_OT_ZNODE
) ||
2037 (doi
.doi_bonus_type
== DMU_OT_ZNODE
&&
2038 doi
.doi_bonus_size
< sizeof (znode_phys_t
))) {
2039 sa_buf_rele(*db
, tag
);
2040 return (SET_ERROR(ENOTSUP
));
2043 error
= sa_handle_get(osp
, obj
, NULL
, SA_HDL_PRIVATE
, hdlp
);
2045 sa_buf_rele(*db
, tag
);
2053 zfs_release_sa_handle(sa_handle_t
*hdl
, dmu_buf_t
*db
, const void *tag
)
2055 sa_handle_destroy(hdl
);
2056 sa_buf_rele(db
, tag
);
2060 * Given an object number, return its parent object number and whether
2061 * or not the object is an extended attribute directory.
2064 zfs_obj_to_pobj(objset_t
*osp
, sa_handle_t
*hdl
, sa_attr_type_t
*sa_table
,
2065 uint64_t *pobjp
, int *is_xattrdir
)
2070 uint64_t parent_mode
;
2071 sa_bulk_attr_t bulk
[3];
2072 sa_handle_t
*sa_hdl
;
2077 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_PARENT
], NULL
,
2078 &parent
, sizeof (parent
));
2079 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_FLAGS
], NULL
,
2080 &pflags
, sizeof (pflags
));
2081 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_MODE
], NULL
,
2082 &mode
, sizeof (mode
));
2084 if ((error
= sa_bulk_lookup(hdl
, bulk
, count
)) != 0)
2088 * When a link is removed its parent pointer is not changed and will
2089 * be invalid. There are two cases where a link is removed but the
2090 * file stays around, when it goes to the delete queue and when there
2091 * are additional links.
2093 error
= zfs_grab_sa_handle(osp
, parent
, &sa_hdl
, &sa_db
, FTAG
);
2097 error
= sa_lookup(sa_hdl
, ZPL_MODE
, &parent_mode
, sizeof (parent_mode
));
2098 zfs_release_sa_handle(sa_hdl
, sa_db
, FTAG
);
2102 *is_xattrdir
= ((pflags
& ZFS_XATTR
) != 0) && S_ISDIR(mode
);
2105 * Extended attributes can be applied to files, directories, etc.
2106 * Otherwise the parent must be a directory.
2108 if (!*is_xattrdir
&& !S_ISDIR(parent_mode
))
2109 return (SET_ERROR(EINVAL
));
2117 * Given an object number, return some zpl level statistics
2120 zfs_obj_to_stats_impl(sa_handle_t
*hdl
, sa_attr_type_t
*sa_table
,
2123 sa_bulk_attr_t bulk
[4];
2126 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_MODE
], NULL
,
2127 &sb
->zs_mode
, sizeof (sb
->zs_mode
));
2128 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_GEN
], NULL
,
2129 &sb
->zs_gen
, sizeof (sb
->zs_gen
));
2130 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_LINKS
], NULL
,
2131 &sb
->zs_links
, sizeof (sb
->zs_links
));
2132 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_CTIME
], NULL
,
2133 &sb
->zs_ctime
, sizeof (sb
->zs_ctime
));
2135 return (sa_bulk_lookup(hdl
, bulk
, count
));
2139 zfs_obj_to_path_impl(objset_t
*osp
, uint64_t obj
, sa_handle_t
*hdl
,
2140 sa_attr_type_t
*sa_table
, char *buf
, int len
)
2142 sa_handle_t
*sa_hdl
;
2143 sa_handle_t
*prevhdl
= NULL
;
2144 dmu_buf_t
*prevdb
= NULL
;
2145 dmu_buf_t
*sa_db
= NULL
;
2146 char *path
= buf
+ len
- 1;
2152 uint64_t deleteq_obj
;
2153 VERIFY0(zap_lookup(osp
, MASTER_NODE_OBJ
,
2154 ZFS_UNLINKED_SET
, sizeof (uint64_t), 1, &deleteq_obj
));
2155 error
= zap_lookup_int(osp
, deleteq_obj
, obj
);
2158 } else if (error
!= ENOENT
) {
2164 char component
[MAXNAMELEN
+ 2];
2166 int is_xattrdir
= 0;
2169 ASSERT(prevhdl
!= NULL
);
2170 zfs_release_sa_handle(prevhdl
, prevdb
, FTAG
);
2173 if ((error
= zfs_obj_to_pobj(osp
, sa_hdl
, sa_table
, &pobj
,
2174 &is_xattrdir
)) != 0)
2185 strcpy(component
+ 1, "<xattrdir>");
2187 error
= zap_value_search(osp
, pobj
, obj
,
2188 ZFS_DIRENT_OBJ(-1ULL), component
+ 1);
2193 complen
= strlen(component
);
2195 ASSERT(path
>= buf
);
2196 memcpy(path
, component
, complen
);
2199 if (sa_hdl
!= hdl
) {
2203 error
= zfs_grab_sa_handle(osp
, obj
, &sa_hdl
, &sa_db
, FTAG
);
2211 if (sa_hdl
!= NULL
&& sa_hdl
!= hdl
) {
2212 ASSERT(sa_db
!= NULL
);
2213 zfs_release_sa_handle(sa_hdl
, sa_db
, FTAG
);
2217 (void) memmove(buf
, path
, buf
+ len
- path
);
2223 zfs_obj_to_path(objset_t
*osp
, uint64_t obj
, char *buf
, int len
)
2225 sa_attr_type_t
*sa_table
;
2230 error
= zfs_sa_setup(osp
, &sa_table
);
2234 error
= zfs_grab_sa_handle(osp
, obj
, &hdl
, &db
, FTAG
);
2238 error
= zfs_obj_to_path_impl(osp
, obj
, hdl
, sa_table
, buf
, len
);
2240 zfs_release_sa_handle(hdl
, db
, FTAG
);
2245 zfs_obj_to_stats(objset_t
*osp
, uint64_t obj
, zfs_stat_t
*sb
,
2248 char *path
= buf
+ len
- 1;
2249 sa_attr_type_t
*sa_table
;
2256 error
= zfs_sa_setup(osp
, &sa_table
);
2260 error
= zfs_grab_sa_handle(osp
, obj
, &hdl
, &db
, FTAG
);
2264 error
= zfs_obj_to_stats_impl(hdl
, sa_table
, sb
);
2266 zfs_release_sa_handle(hdl
, db
, FTAG
);
2270 error
= zfs_obj_to_path_impl(osp
, obj
, hdl
, sa_table
, buf
, len
);
2272 zfs_release_sa_handle(hdl
, db
, FTAG
);
2277 * Read a property stored within the master node.
2280 zfs_get_zplprop(objset_t
*os
, zfs_prop_t prop
, uint64_t *value
)
2282 uint64_t *cached_copy
= NULL
;
2285 * Figure out where in the objset_t the cached copy would live, if it
2286 * is available for the requested property.
2290 case ZFS_PROP_VERSION
:
2291 cached_copy
= &os
->os_version
;
2293 case ZFS_PROP_NORMALIZE
:
2294 cached_copy
= &os
->os_normalization
;
2296 case ZFS_PROP_UTF8ONLY
:
2297 cached_copy
= &os
->os_utf8only
;
2300 cached_copy
= &os
->os_casesensitivity
;
2306 if (cached_copy
!= NULL
&& *cached_copy
!= OBJSET_PROP_UNINITIALIZED
) {
2307 *value
= *cached_copy
;
2312 * If the property wasn't cached, look up the file system's value for
2313 * the property. For the version property, we look up a slightly
2318 if (prop
== ZFS_PROP_VERSION
)
2319 pname
= ZPL_VERSION_STR
;
2321 pname
= zfs_prop_to_name(prop
);
2324 ASSERT3U(os
->os_phys
->os_type
, ==, DMU_OST_ZFS
);
2325 error
= zap_lookup(os
, MASTER_NODE_OBJ
, pname
, 8, 1, value
);
2328 if (error
== ENOENT
) {
2329 /* No value set, use the default value */
2331 case ZFS_PROP_VERSION
:
2332 *value
= ZPL_VERSION
;
2334 case ZFS_PROP_NORMALIZE
:
2335 case ZFS_PROP_UTF8ONLY
:
2339 *value
= ZFS_CASE_SENSITIVE
;
2341 case ZFS_PROP_ACLTYPE
:
2342 *value
= ZFS_ACLTYPE_OFF
;
2351 * If one of the methods for getting the property value above worked,
2352 * copy it into the objset_t's cache.
2354 if (error
== 0 && cached_copy
!= NULL
) {
2355 *cached_copy
= *value
;
2361 #if defined(_KERNEL)
2362 EXPORT_SYMBOL(zfs_create_fs
);
2363 EXPORT_SYMBOL(zfs_obj_to_path
);
2366 module_param(zfs_object_mutex_size
, uint
, 0644);
2367 MODULE_PARM_DESC(zfs_object_mutex_size
, "Size of znode hold array");
2368 module_param(zfs_unlink_suspend_progress
, int, 0644);
2369 MODULE_PARM_DESC(zfs_unlink_suspend_progress
, "Set to prevent async unlinks "
2370 "(debug - leaks space into the unlinked set)");