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) 2013 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
29 #include <sys/types.h>
30 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
35 #include <sys/mntent.h>
36 #include <sys/mkdev.h>
37 #include <sys/u8_textprep.h>
38 #include <sys/dsl_dataset.h>
40 #include <sys/vfs_opreg.h>
41 #include <sys/vnode.h>
44 #include <sys/errno.h>
45 #include <sys/unistd.h>
47 #include <sys/atomic.h>
49 #include "fs/fs_subr.h"
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_acl.h>
52 #include <sys/zfs_ioctl.h>
53 #include <sys/zfs_rlock.h>
54 #include <sys/zfs_fuid.h>
55 #include <sys/zfs_vnops.h>
56 #include <sys/zfs_ctldir.h>
57 #include <sys/dnode.h>
58 #include <sys/fs/zfs.h>
59 #include <sys/kidmap.h>
64 #include <sys/dmu_objset.h>
65 #include <sys/refcount.h>
68 #include <sys/zfs_znode.h>
70 #include <sys/zfs_sa.h>
71 #include <sys/zfs_stat.h>
74 #include "zfs_comutil.h"
77 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
78 * turned on when DEBUG is also defined.
85 #define ZNODE_STAT_ADD(stat) ((stat)++)
87 #define ZNODE_STAT_ADD(stat) /* nothing */
88 #endif /* ZNODE_STATS */
91 * Functions needed for userland (ie: libzpool) are not put under
92 * #ifdef_KERNEL; the rest of the functions have dependencies
93 * (such as VFS logic) that will not compile easily in userland.
97 static kmem_cache_t
*znode_cache
= NULL
;
98 static kmem_cache_t
*znode_hold_cache
= NULL
;
99 unsigned int zfs_object_mutex_size
= ZFS_OBJ_MTX_SZ
;
103 zfs_znode_cache_constructor(void *buf
, void *arg
, int kmflags
)
107 inode_init_once(ZTOI(zp
));
108 list_link_init(&zp
->z_link_node
);
110 mutex_init(&zp
->z_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
111 rw_init(&zp
->z_parent_lock
, NULL
, RW_DEFAULT
, NULL
);
112 rw_init(&zp
->z_name_lock
, NULL
, RW_DEFAULT
, NULL
);
113 mutex_init(&zp
->z_acl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
114 rw_init(&zp
->z_xattr_lock
, NULL
, RW_DEFAULT
, NULL
);
116 zfs_rlock_init(&zp
->z_range_lock
);
118 zp
->z_dirlocks
= NULL
;
119 zp
->z_acl_cached
= NULL
;
120 zp
->z_xattr_cached
= NULL
;
127 zfs_znode_cache_destructor(void *buf
, void *arg
)
131 ASSERT(!list_link_active(&zp
->z_link_node
));
132 mutex_destroy(&zp
->z_lock
);
133 rw_destroy(&zp
->z_parent_lock
);
134 rw_destroy(&zp
->z_name_lock
);
135 mutex_destroy(&zp
->z_acl_lock
);
136 rw_destroy(&zp
->z_xattr_lock
);
137 zfs_rlock_destroy(&zp
->z_range_lock
);
139 ASSERT(zp
->z_dirlocks
== NULL
);
140 ASSERT(zp
->z_acl_cached
== NULL
);
141 ASSERT(zp
->z_xattr_cached
== NULL
);
145 zfs_znode_hold_cache_constructor(void *buf
, void *arg
, int kmflags
)
147 znode_hold_t
*zh
= buf
;
149 mutex_init(&zh
->zh_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
150 refcount_create(&zh
->zh_refcount
);
151 zh
->zh_obj
= ZFS_NO_OBJECT
;
157 zfs_znode_hold_cache_destructor(void *buf
, void *arg
)
159 znode_hold_t
*zh
= buf
;
161 mutex_destroy(&zh
->zh_lock
);
162 refcount_destroy(&zh
->zh_refcount
);
169 * Initialize zcache. The KMC_SLAB hint is used in order that it be
170 * backed by kmalloc() when on the Linux slab in order that any
171 * wait_on_bit() operations on the related inode operate properly.
173 ASSERT(znode_cache
== NULL
);
174 znode_cache
= kmem_cache_create("zfs_znode_cache",
175 sizeof (znode_t
), 0, zfs_znode_cache_constructor
,
176 zfs_znode_cache_destructor
, NULL
, NULL
, NULL
, KMC_SLAB
);
178 ASSERT(znode_hold_cache
== NULL
);
179 znode_hold_cache
= kmem_cache_create("zfs_znode_hold_cache",
180 sizeof (znode_hold_t
), 0, zfs_znode_hold_cache_constructor
,
181 zfs_znode_hold_cache_destructor
, NULL
, NULL
, NULL
, 0);
191 kmem_cache_destroy(znode_cache
);
194 if (znode_hold_cache
)
195 kmem_cache_destroy(znode_hold_cache
);
196 znode_hold_cache
= NULL
;
200 * The zfs_znode_hold_enter() / zfs_znode_hold_exit() functions are used to
201 * serialize access to a znode and its SA buffer while the object is being
202 * created or destroyed. This kind of locking would normally reside in the
203 * znode itself but in this case that's impossible because the znode and SA
204 * buffer may not yet exist. Therefore the locking is handled externally
205 * with an array of mutexs and AVLs trees which contain per-object locks.
207 * In zfs_znode_hold_enter() a per-object lock is created as needed, inserted
208 * in to the correct AVL tree and finally the per-object lock is held. In
209 * zfs_znode_hold_exit() the process is reversed. The per-object lock is
210 * released, removed from the AVL tree and destroyed if there are no waiters.
212 * This scheme has two important properties:
214 * 1) No memory allocations are performed while holding one of the z_hold_locks.
215 * This ensures evict(), which can be called from direct memory reclaim, will
216 * never block waiting on a z_hold_locks which just happens to have hashed
219 * 2) All locks used to serialize access to an object are per-object and never
220 * shared. This minimizes lock contention without creating a large number
221 * of dedicated locks.
223 * On the downside it does require znode_lock_t structures to be frequently
224 * allocated and freed. However, because these are backed by a kmem cache
225 * and very short lived this cost is minimal.
228 zfs_znode_hold_compare(const void *a
, const void *b
)
230 const znode_hold_t
*zh_a
= a
;
231 const znode_hold_t
*zh_b
= b
;
233 if (zh_a
->zh_obj
< zh_b
->zh_obj
)
235 else if (zh_a
->zh_obj
> zh_b
->zh_obj
)
242 zfs_znode_held(zfs_sb_t
*zsb
, uint64_t obj
)
244 znode_hold_t
*zh
, search
;
245 int i
= ZFS_OBJ_HASH(zsb
, obj
);
250 mutex_enter(&zsb
->z_hold_locks
[i
]);
251 zh
= avl_find(&zsb
->z_hold_trees
[i
], &search
, NULL
);
252 held
= (zh
&& MUTEX_HELD(&zh
->zh_lock
)) ? B_TRUE
: B_FALSE
;
253 mutex_exit(&zsb
->z_hold_locks
[i
]);
258 static znode_hold_t
*
259 zfs_znode_hold_enter(zfs_sb_t
*zsb
, uint64_t obj
)
261 znode_hold_t
*zh
, *zh_new
, search
;
262 int i
= ZFS_OBJ_HASH(zsb
, obj
);
263 boolean_t found
= B_FALSE
;
265 zh_new
= kmem_cache_alloc(znode_hold_cache
, KM_SLEEP
);
266 zh_new
->zh_obj
= obj
;
269 mutex_enter(&zsb
->z_hold_locks
[i
]);
270 zh
= avl_find(&zsb
->z_hold_trees
[i
], &search
, NULL
);
271 if (likely(zh
== NULL
)) {
273 avl_add(&zsb
->z_hold_trees
[i
], zh
);
275 ASSERT3U(zh
->zh_obj
, ==, obj
);
278 refcount_add(&zh
->zh_refcount
, NULL
);
279 mutex_exit(&zsb
->z_hold_locks
[i
]);
282 kmem_cache_free(znode_hold_cache
, zh_new
);
284 ASSERT(MUTEX_NOT_HELD(&zh
->zh_lock
));
285 ASSERT3S(refcount_count(&zh
->zh_refcount
), >, 0);
286 mutex_enter(&zh
->zh_lock
);
292 zfs_znode_hold_exit(zfs_sb_t
*zsb
, znode_hold_t
*zh
)
294 int i
= ZFS_OBJ_HASH(zsb
, zh
->zh_obj
);
295 boolean_t remove
= B_FALSE
;
297 ASSERT(zfs_znode_held(zsb
, zh
->zh_obj
));
298 ASSERT3S(refcount_count(&zh
->zh_refcount
), >, 0);
299 mutex_exit(&zh
->zh_lock
);
301 mutex_enter(&zsb
->z_hold_locks
[i
]);
302 if (refcount_remove(&zh
->zh_refcount
, NULL
) == 0) {
303 avl_remove(&zsb
->z_hold_trees
[i
], zh
);
306 mutex_exit(&zsb
->z_hold_locks
[i
]);
308 if (remove
== B_TRUE
)
309 kmem_cache_free(znode_hold_cache
, zh
);
313 zfs_create_share_dir(zfs_sb_t
*zsb
, dmu_tx_t
*tx
)
315 #ifdef HAVE_SMB_SHARE
316 zfs_acl_ids_t acl_ids
;
323 vattr
.va_mask
= AT_MODE
|AT_UID
|AT_GID
|AT_TYPE
;
324 vattr
.va_mode
= S_IFDIR
| 0555;
325 vattr
.va_uid
= crgetuid(kcred
);
326 vattr
.va_gid
= crgetgid(kcred
);
328 sharezp
= kmem_cache_alloc(znode_cache
, KM_SLEEP
);
329 sharezp
->z_moved
= 0;
330 sharezp
->z_unlinked
= 0;
331 sharezp
->z_atime_dirty
= 0;
332 sharezp
->z_zfsvfs
= zfsvfs
;
333 sharezp
->z_is_sa
= zfsvfs
->z_use_sa
;
339 VERIFY(0 == zfs_acl_ids_create(sharezp
, IS_ROOT_NODE
, &vattr
,
340 kcred
, NULL
, &acl_ids
));
341 zfs_mknode(sharezp
, &vattr
, tx
, kcred
, IS_ROOT_NODE
, &zp
, &acl_ids
);
342 ASSERT3P(zp
, ==, sharezp
);
343 ASSERT(!vn_in_dnlc(ZTOV(sharezp
))); /* not valid to move */
344 POINTER_INVALIDATE(&sharezp
->z_zfsvfs
);
345 error
= zap_add(zfsvfs
->z_os
, MASTER_NODE_OBJ
,
346 ZFS_SHARES_DIR
, 8, 1, &sharezp
->z_id
, tx
);
347 zfsvfs
->z_shares_dir
= sharezp
->z_id
;
349 zfs_acl_ids_free(&acl_ids
);
350 // ZTOV(sharezp)->v_count = 0;
351 sa_handle_destroy(sharezp
->z_sa_hdl
);
352 kmem_cache_free(znode_cache
, sharezp
);
357 #endif /* HAVE_SMB_SHARE */
361 zfs_znode_sa_init(zfs_sb_t
*zsb
, znode_t
*zp
,
362 dmu_buf_t
*db
, dmu_object_type_t obj_type
, sa_handle_t
*sa_hdl
)
364 ASSERT(zfs_znode_held(zsb
, zp
->z_id
));
366 mutex_enter(&zp
->z_lock
);
368 ASSERT(zp
->z_sa_hdl
== NULL
);
369 ASSERT(zp
->z_acl_cached
== NULL
);
370 if (sa_hdl
== NULL
) {
371 VERIFY(0 == sa_handle_get_from_db(zsb
->z_os
, db
, zp
,
372 SA_HDL_SHARED
, &zp
->z_sa_hdl
));
374 zp
->z_sa_hdl
= sa_hdl
;
375 sa_set_userp(sa_hdl
, zp
);
378 zp
->z_is_sa
= (obj_type
== DMU_OT_SA
) ? B_TRUE
: B_FALSE
;
380 mutex_exit(&zp
->z_lock
);
384 zfs_znode_dmu_fini(znode_t
*zp
)
386 ASSERT(zfs_znode_held(ZTOZSB(zp
), zp
->z_id
) || zp
->z_unlinked
||
387 RW_WRITE_HELD(&ZTOZSB(zp
)->z_teardown_inactive_lock
));
389 sa_handle_destroy(zp
->z_sa_hdl
);
394 * Called by new_inode() to allocate a new inode.
397 zfs_inode_alloc(struct super_block
*sb
, struct inode
**ip
)
401 zp
= kmem_cache_alloc(znode_cache
, KM_SLEEP
);
408 * Called in multiple places when an inode should be destroyed.
411 zfs_inode_destroy(struct inode
*ip
)
413 znode_t
*zp
= ITOZ(ip
);
414 zfs_sb_t
*zsb
= ZTOZSB(zp
);
416 mutex_enter(&zsb
->z_znodes_lock
);
417 if (list_link_active(&zp
->z_link_node
)) {
418 list_remove(&zsb
->z_all_znodes
, zp
);
421 mutex_exit(&zsb
->z_znodes_lock
);
423 if (zp
->z_acl_cached
) {
424 zfs_acl_free(zp
->z_acl_cached
);
425 zp
->z_acl_cached
= NULL
;
428 if (zp
->z_xattr_cached
) {
429 nvlist_free(zp
->z_xattr_cached
);
430 zp
->z_xattr_cached
= NULL
;
433 kmem_cache_free(znode_cache
, zp
);
437 zfs_inode_set_ops(zfs_sb_t
*zsb
, struct inode
*ip
)
441 switch (ip
->i_mode
& S_IFMT
) {
443 ip
->i_op
= &zpl_inode_operations
;
444 ip
->i_fop
= &zpl_file_operations
;
445 ip
->i_mapping
->a_ops
= &zpl_address_space_operations
;
449 ip
->i_op
= &zpl_dir_inode_operations
;
450 ip
->i_fop
= &zpl_dir_file_operations
;
451 ITOZ(ip
)->z_zn_prefetch
= B_TRUE
;
455 ip
->i_op
= &zpl_symlink_inode_operations
;
459 * rdev is only stored in a SA only for device files.
463 sa_lookup(ITOZ(ip
)->z_sa_hdl
, SA_ZPL_RDEV(zsb
), &rdev
,
468 init_special_inode(ip
, ip
->i_mode
, rdev
);
469 ip
->i_op
= &zpl_special_inode_operations
;
473 zfs_panic_recover("inode %llu has invalid mode: 0x%x\n",
474 (u_longlong_t
)ip
->i_ino
, ip
->i_mode
);
476 /* Assume the inode is a file and attempt to continue */
477 ip
->i_mode
= S_IFREG
| 0644;
478 ip
->i_op
= &zpl_inode_operations
;
479 ip
->i_fop
= &zpl_file_operations
;
480 ip
->i_mapping
->a_ops
= &zpl_address_space_operations
;
486 zfs_set_inode_flags(znode_t
*zp
, struct inode
*ip
)
489 * Linux and Solaris have different sets of file attributes, so we
490 * restrict this conversion to the intersection of the two.
493 if (zp
->z_pflags
& ZFS_IMMUTABLE
)
494 ip
->i_flags
|= S_IMMUTABLE
;
496 ip
->i_flags
&= ~S_IMMUTABLE
;
498 if (zp
->z_pflags
& ZFS_APPENDONLY
)
499 ip
->i_flags
|= S_APPEND
;
501 ip
->i_flags
&= ~S_APPEND
;
505 * Update the embedded inode given the znode. We should work toward
506 * eliminating this function as soon as possible by removing values
507 * which are duplicated between the znode and inode. If the generic
508 * inode has the correct field it should be used, and the ZFS code
509 * updated to access the inode. This can be done incrementally.
512 zfs_inode_update_impl(znode_t
*zp
, boolean_t
new)
517 u_longlong_t i_blocks
;
518 uint64_t atime
[2], mtime
[2], ctime
[2];
524 /* Skip .zfs control nodes which do not exist on disk. */
525 if (zfsctl_is_node(ip
))
528 sa_lookup(zp
->z_sa_hdl
, SA_ZPL_ATIME(zsb
), &atime
, 16);
529 sa_lookup(zp
->z_sa_hdl
, SA_ZPL_MTIME(zsb
), &mtime
, 16);
530 sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CTIME(zsb
), &ctime
, 16);
532 dmu_object_size_from_db(sa_get_db(zp
->z_sa_hdl
), &blksize
, &i_blocks
);
534 spin_lock(&ip
->i_lock
);
535 ip
->i_generation
= zp
->z_gen
;
536 ip
->i_uid
= SUID_TO_KUID(zp
->z_uid
);
537 ip
->i_gid
= SGID_TO_KGID(zp
->z_gid
);
538 set_nlink(ip
, zp
->z_links
);
539 ip
->i_mode
= zp
->z_mode
;
540 zfs_set_inode_flags(zp
, ip
);
541 ip
->i_blkbits
= SPA_MINBLOCKSHIFT
;
542 ip
->i_blocks
= i_blocks
;
545 * Only read atime from SA if we are newly created inode (or rezget),
546 * otherwise i_atime might be dirty.
549 ZFS_TIME_DECODE(&ip
->i_atime
, atime
);
550 ZFS_TIME_DECODE(&ip
->i_mtime
, mtime
);
551 ZFS_TIME_DECODE(&ip
->i_ctime
, ctime
);
553 i_size_write(ip
, zp
->z_size
);
554 spin_unlock(&ip
->i_lock
);
558 zfs_inode_update_new(znode_t
*zp
)
560 zfs_inode_update_impl(zp
, B_TRUE
);
564 zfs_inode_update(znode_t
*zp
)
566 zfs_inode_update_impl(zp
, B_FALSE
);
570 * Construct a znode+inode and initialize.
572 * This does not do a call to dmu_set_user() that is
573 * up to the caller to do, in case you don't want to
577 zfs_znode_alloc(zfs_sb_t
*zsb
, dmu_buf_t
*db
, int blksz
,
578 dmu_object_type_t obj_type
, uint64_t obj
, sa_handle_t
*hdl
)
584 sa_bulk_attr_t bulk
[8];
589 ip
= new_inode(zsb
->z_sb
);
594 ASSERT(zp
->z_dirlocks
== NULL
);
595 ASSERT3P(zp
->z_acl_cached
, ==, NULL
);
596 ASSERT3P(zp
->z_xattr_cached
, ==, NULL
);
600 zp
->z_atime_dirty
= 0;
602 zp
->z_id
= db
->db_object
;
604 zp
->z_seq
= 0x7A4653;
606 zp
->z_is_mapped
= B_FALSE
;
607 zp
->z_is_ctldir
= B_FALSE
;
608 zp
->z_is_stale
= B_FALSE
;
609 zp
->z_range_lock
.zr_size
= &zp
->z_size
;
610 zp
->z_range_lock
.zr_blksz
= &zp
->z_blksz
;
611 zp
->z_range_lock
.zr_max_blksz
= &ZTOZSB(zp
)->z_max_blksz
;
613 zfs_znode_sa_init(zsb
, zp
, db
, obj_type
, hdl
);
615 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
), NULL
, &mode
, 8);
616 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GEN(zsb
), NULL
, &zp
->z_gen
, 8);
617 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zsb
), NULL
, &zp
->z_size
, 8);
618 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zsb
), NULL
, &zp
->z_links
, 8);
619 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
621 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_PARENT(zsb
), NULL
,
623 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zsb
), NULL
, &zp
->z_uid
, 8);
624 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zsb
), NULL
, &zp
->z_gid
, 8);
626 if (sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
) != 0 || zp
->z_gen
== 0) {
628 sa_handle_destroy(zp
->z_sa_hdl
);
636 zfs_inode_update_new(zp
);
637 zfs_inode_set_ops(zsb
, ip
);
640 * The only way insert_inode_locked() can fail is if the ip->i_ino
641 * number is already hashed for this super block. This can never
642 * happen because the inode numbers map 1:1 with the object numbers.
644 * The one exception is rolling back a mounted file system, but in
645 * this case all the active inode are unhashed during the rollback.
647 VERIFY3S(insert_inode_locked(ip
), ==, 0);
649 mutex_enter(&zsb
->z_znodes_lock
);
650 list_insert_tail(&zsb
->z_all_znodes
, zp
);
653 mutex_exit(&zsb
->z_znodes_lock
);
655 unlock_new_inode(ip
);
664 * Safely mark an inode dirty. Inodes which are part of a read-only
665 * file system or snapshot may not be dirtied.
668 zfs_mark_inode_dirty(struct inode
*ip
)
670 zfs_sb_t
*zsb
= ITOZSB(ip
);
672 if (zfs_is_readonly(zsb
) || dmu_objset_is_snapshot(zsb
->z_os
))
675 mark_inode_dirty(ip
);
678 static uint64_t empty_xattr
;
679 static uint64_t pad
[4];
680 static zfs_acl_phys_t acl_phys
;
682 * Create a new DMU object to hold a zfs znode.
684 * IN: dzp - parent directory for new znode
685 * vap - file attributes for new znode
686 * tx - dmu transaction id for zap operations
687 * cr - credentials of caller
689 * IS_ROOT_NODE - new object will be root
690 * IS_XATTR - new object is an attribute
691 * bonuslen - length of bonus buffer
692 * setaclp - File/Dir initial ACL
693 * fuidp - Tracks fuid allocation.
695 * OUT: zpp - allocated znode
699 zfs_mknode(znode_t
*dzp
, vattr_t
*vap
, dmu_tx_t
*tx
, cred_t
*cr
,
700 uint_t flag
, znode_t
**zpp
, zfs_acl_ids_t
*acl_ids
)
702 uint64_t crtime
[2], atime
[2], mtime
[2], ctime
[2];
703 uint64_t mode
, size
, links
, parent
, pflags
;
704 uint64_t dzp_pflags
= 0;
706 zfs_sb_t
*zsb
= ZTOZSB(dzp
);
712 dmu_object_type_t obj_type
;
713 sa_bulk_attr_t
*sa_attrs
;
715 zfs_acl_locator_cb_t locate
= { 0 };
719 obj
= vap
->va_nodeid
;
720 now
= vap
->va_ctime
; /* see zfs_replay_create() */
721 gen
= vap
->va_nblocks
; /* ditto */
725 gen
= dmu_tx_get_txg(tx
);
728 obj_type
= zsb
->z_use_sa
? DMU_OT_SA
: DMU_OT_ZNODE
;
729 bonuslen
= (obj_type
== DMU_OT_SA
) ?
730 DN_MAX_BONUSLEN
: ZFS_OLD_ZNODE_PHYS_SIZE
;
733 * Create a new DMU object.
736 * There's currently no mechanism for pre-reading the blocks that will
737 * be needed to allocate a new object, so we accept the small chance
738 * that there will be an i/o error and we will fail one of the
741 if (S_ISDIR(vap
->va_mode
)) {
743 VERIFY0(zap_create_claim_norm(zsb
->z_os
, obj
,
744 zsb
->z_norm
, DMU_OT_DIRECTORY_CONTENTS
,
745 obj_type
, bonuslen
, tx
));
747 obj
= zap_create_norm(zsb
->z_os
,
748 zsb
->z_norm
, DMU_OT_DIRECTORY_CONTENTS
,
749 obj_type
, bonuslen
, tx
);
753 VERIFY0(dmu_object_claim(zsb
->z_os
, obj
,
754 DMU_OT_PLAIN_FILE_CONTENTS
, 0,
755 obj_type
, bonuslen
, tx
));
757 obj
= dmu_object_alloc(zsb
->z_os
,
758 DMU_OT_PLAIN_FILE_CONTENTS
, 0,
759 obj_type
, bonuslen
, tx
);
763 zh
= zfs_znode_hold_enter(zsb
, obj
);
764 VERIFY(0 == sa_buf_hold(zsb
->z_os
, obj
, NULL
, &db
));
767 * If this is the root, fix up the half-initialized parent pointer
768 * to reference the just-allocated physical data area.
770 if (flag
& IS_ROOT_NODE
) {
773 dzp_pflags
= dzp
->z_pflags
;
777 * If parent is an xattr, so am I.
779 if (dzp_pflags
& ZFS_XATTR
) {
783 if (zsb
->z_use_fuids
)
784 pflags
= ZFS_ARCHIVE
| ZFS_AV_MODIFIED
;
788 if (S_ISDIR(vap
->va_mode
)) {
789 size
= 2; /* contents ("." and "..") */
790 links
= (flag
& (IS_ROOT_NODE
| IS_XATTR
)) ? 2 : 1;
795 if (S_ISBLK(vap
->va_mode
) || S_ISCHR(vap
->va_mode
))
799 mode
= acl_ids
->z_mode
;
804 * No execs denied will be deterimed when zfs_mode_compute() is called.
806 pflags
|= acl_ids
->z_aclp
->z_hints
&
807 (ZFS_ACL_TRIVIAL
|ZFS_INHERIT_ACE
|ZFS_ACL_AUTO_INHERIT
|
808 ZFS_ACL_DEFAULTED
|ZFS_ACL_PROTECTED
);
810 ZFS_TIME_ENCODE(&now
, crtime
);
811 ZFS_TIME_ENCODE(&now
, ctime
);
813 if (vap
->va_mask
& ATTR_ATIME
) {
814 ZFS_TIME_ENCODE(&vap
->va_atime
, atime
);
816 ZFS_TIME_ENCODE(&now
, atime
);
819 if (vap
->va_mask
& ATTR_MTIME
) {
820 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
822 ZFS_TIME_ENCODE(&now
, mtime
);
825 /* Now add in all of the "SA" attributes */
826 VERIFY(0 == sa_handle_get_from_db(zsb
->z_os
, db
, NULL
, SA_HDL_SHARED
,
830 * Setup the array of attributes to be replaced/set on the new file
832 * order for DMU_OT_ZNODE is critical since it needs to be constructed
833 * in the old znode_phys_t format. Don't change this ordering
835 sa_attrs
= kmem_alloc(sizeof (sa_bulk_attr_t
) * ZPL_END
, KM_SLEEP
);
837 if (obj_type
== DMU_OT_ZNODE
) {
838 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_ATIME(zsb
),
840 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_MTIME(zsb
),
842 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_CTIME(zsb
),
844 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_CRTIME(zsb
),
846 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_GEN(zsb
),
848 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_MODE(zsb
),
850 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_SIZE(zsb
),
852 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_PARENT(zsb
),
855 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_MODE(zsb
),
857 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_SIZE(zsb
),
859 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_GEN(zsb
),
861 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_UID(zsb
),
862 NULL
, &acl_ids
->z_fuid
, 8);
863 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_GID(zsb
),
864 NULL
, &acl_ids
->z_fgid
, 8);
865 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_PARENT(zsb
),
867 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_FLAGS(zsb
),
869 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_ATIME(zsb
),
871 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_MTIME(zsb
),
873 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_CTIME(zsb
),
875 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_CRTIME(zsb
),
879 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_LINKS(zsb
), NULL
, &links
, 8);
881 if (obj_type
== DMU_OT_ZNODE
) {
882 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_XATTR(zsb
), NULL
,
885 if (obj_type
== DMU_OT_ZNODE
||
886 (S_ISBLK(vap
->va_mode
) || S_ISCHR(vap
->va_mode
))) {
887 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_RDEV(zsb
),
890 if (obj_type
== DMU_OT_ZNODE
) {
891 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_FLAGS(zsb
),
893 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_UID(zsb
), NULL
,
894 &acl_ids
->z_fuid
, 8);
895 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_GID(zsb
), NULL
,
896 &acl_ids
->z_fgid
, 8);
897 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_PAD(zsb
), NULL
, pad
,
898 sizeof (uint64_t) * 4);
899 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_ZNODE_ACL(zsb
), NULL
,
900 &acl_phys
, sizeof (zfs_acl_phys_t
));
901 } else if (acl_ids
->z_aclp
->z_version
>= ZFS_ACL_VERSION_FUID
) {
902 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_DACL_COUNT(zsb
), NULL
,
903 &acl_ids
->z_aclp
->z_acl_count
, 8);
904 locate
.cb_aclp
= acl_ids
->z_aclp
;
905 SA_ADD_BULK_ATTR(sa_attrs
, cnt
, SA_ZPL_DACL_ACES(zsb
),
906 zfs_acl_data_locator
, &locate
,
907 acl_ids
->z_aclp
->z_acl_bytes
);
908 mode
= zfs_mode_compute(mode
, acl_ids
->z_aclp
, &pflags
,
909 acl_ids
->z_fuid
, acl_ids
->z_fgid
);
912 VERIFY(sa_replace_all_by_template(sa_hdl
, sa_attrs
, cnt
, tx
) == 0);
914 if (!(flag
& IS_ROOT_NODE
)) {
915 *zpp
= zfs_znode_alloc(zsb
, db
, 0, obj_type
, obj
, sa_hdl
);
916 VERIFY(*zpp
!= NULL
);
920 * If we are creating the root node, the "parent" we
921 * passed in is the znode for the root.
925 (*zpp
)->z_sa_hdl
= sa_hdl
;
928 (*zpp
)->z_pflags
= pflags
;
929 (*zpp
)->z_mode
= mode
;
931 if (obj_type
== DMU_OT_ZNODE
||
932 acl_ids
->z_aclp
->z_version
< ZFS_ACL_VERSION_FUID
) {
933 VERIFY0(zfs_aclset_common(*zpp
, acl_ids
->z_aclp
, cr
, tx
));
935 kmem_free(sa_attrs
, sizeof (sa_bulk_attr_t
) * ZPL_END
);
936 zfs_znode_hold_exit(zsb
, zh
);
940 * Update in-core attributes. It is assumed the caller will be doing an
941 * sa_bulk_update to push the changes out.
944 zfs_xvattr_set(znode_t
*zp
, xvattr_t
*xvap
, dmu_tx_t
*tx
)
948 xoap
= xva_getxoptattr(xvap
);
951 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
953 ZFS_TIME_ENCODE(&xoap
->xoa_createtime
, times
);
954 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_CRTIME(ZTOZSB(zp
)),
955 ×
, sizeof (times
), tx
);
956 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
958 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
959 ZFS_ATTR_SET(zp
, ZFS_READONLY
, xoap
->xoa_readonly
,
961 XVA_SET_RTN(xvap
, XAT_READONLY
);
963 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
964 ZFS_ATTR_SET(zp
, ZFS_HIDDEN
, xoap
->xoa_hidden
,
966 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
968 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
969 ZFS_ATTR_SET(zp
, ZFS_SYSTEM
, xoap
->xoa_system
,
971 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
973 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
974 ZFS_ATTR_SET(zp
, ZFS_ARCHIVE
, xoap
->xoa_archive
,
976 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
978 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
979 ZFS_ATTR_SET(zp
, ZFS_IMMUTABLE
, xoap
->xoa_immutable
,
981 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
983 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
984 ZFS_ATTR_SET(zp
, ZFS_NOUNLINK
, xoap
->xoa_nounlink
,
986 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
988 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
989 ZFS_ATTR_SET(zp
, ZFS_APPENDONLY
, xoap
->xoa_appendonly
,
991 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
993 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
994 ZFS_ATTR_SET(zp
, ZFS_NODUMP
, xoap
->xoa_nodump
,
996 XVA_SET_RTN(xvap
, XAT_NODUMP
);
998 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
999 ZFS_ATTR_SET(zp
, ZFS_OPAQUE
, xoap
->xoa_opaque
,
1001 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
1003 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
1004 ZFS_ATTR_SET(zp
, ZFS_AV_QUARANTINED
,
1005 xoap
->xoa_av_quarantined
, zp
->z_pflags
, tx
);
1006 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
1008 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
1009 ZFS_ATTR_SET(zp
, ZFS_AV_MODIFIED
, xoap
->xoa_av_modified
,
1011 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
1013 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
)) {
1014 zfs_sa_set_scanstamp(zp
, xvap
, tx
);
1015 XVA_SET_RTN(xvap
, XAT_AV_SCANSTAMP
);
1017 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
1018 ZFS_ATTR_SET(zp
, ZFS_REPARSE
, xoap
->xoa_reparse
,
1020 XVA_SET_RTN(xvap
, XAT_REPARSE
);
1022 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
1023 ZFS_ATTR_SET(zp
, ZFS_OFFLINE
, xoap
->xoa_offline
,
1025 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
1027 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
1028 ZFS_ATTR_SET(zp
, ZFS_SPARSE
, xoap
->xoa_sparse
,
1030 XVA_SET_RTN(xvap
, XAT_SPARSE
);
1035 zfs_zget(zfs_sb_t
*zsb
, uint64_t obj_num
, znode_t
**zpp
)
1037 dmu_object_info_t doi
;
1047 zh
= zfs_znode_hold_enter(zsb
, obj_num
);
1049 err
= sa_buf_hold(zsb
->z_os
, obj_num
, NULL
, &db
);
1051 zfs_znode_hold_exit(zsb
, zh
);
1055 dmu_object_info_from_db(db
, &doi
);
1056 if (doi
.doi_bonus_type
!= DMU_OT_SA
&&
1057 (doi
.doi_bonus_type
!= DMU_OT_ZNODE
||
1058 (doi
.doi_bonus_type
== DMU_OT_ZNODE
&&
1059 doi
.doi_bonus_size
< sizeof (znode_phys_t
)))) {
1060 sa_buf_rele(db
, NULL
);
1061 zfs_znode_hold_exit(zsb
, zh
);
1062 return (SET_ERROR(EINVAL
));
1065 hdl
= dmu_buf_get_user(db
);
1067 zp
= sa_get_userdata(hdl
);
1071 * Since "SA" does immediate eviction we
1072 * should never find a sa handle that doesn't
1073 * know about the znode.
1076 ASSERT3P(zp
, !=, NULL
);
1078 mutex_enter(&zp
->z_lock
);
1079 ASSERT3U(zp
->z_id
, ==, obj_num
);
1080 if (zp
->z_unlinked
) {
1081 err
= SET_ERROR(ENOENT
);
1084 * If igrab() returns NULL the VFS has independently
1085 * determined the inode should be evicted and has
1086 * called iput_final() to start the eviction process.
1087 * The SA handle is still valid but because the VFS
1088 * requires that the eviction succeed we must drop
1089 * our locks and references to allow the eviction to
1090 * complete. The zfs_zget() may then be retried.
1092 * This unlikely case could be optimized by registering
1093 * a sops->drop_inode() callback. The callback would
1094 * need to detect the active SA hold thereby informing
1095 * the VFS that this inode should not be evicted.
1097 if (igrab(ZTOI(zp
)) == NULL
) {
1098 mutex_exit(&zp
->z_lock
);
1099 sa_buf_rele(db
, NULL
);
1100 zfs_znode_hold_exit(zsb
, zh
);
1101 /* inode might need this to finish evict */
1108 mutex_exit(&zp
->z_lock
);
1109 sa_buf_rele(db
, NULL
);
1110 zfs_znode_hold_exit(zsb
, zh
);
1115 * Not found create new znode/vnode but only if file exists.
1117 * There is a small window where zfs_vget() could
1118 * find this object while a file create is still in
1119 * progress. This is checked for in zfs_znode_alloc()
1121 * if zfs_znode_alloc() fails it will drop the hold on the
1124 zp
= zfs_znode_alloc(zsb
, db
, doi
.doi_data_block_size
,
1125 doi
.doi_bonus_type
, obj_num
, NULL
);
1127 err
= SET_ERROR(ENOENT
);
1131 zfs_znode_hold_exit(zsb
, zh
);
1136 zfs_rezget(znode_t
*zp
)
1138 zfs_sb_t
*zsb
= ZTOZSB(zp
);
1139 dmu_object_info_t doi
;
1141 uint64_t obj_num
= zp
->z_id
;
1143 sa_bulk_attr_t bulk
[7];
1150 * skip ctldir, otherwise they will always get invalidated. This will
1151 * cause funny behaviour for the mounted snapdirs. Especially for
1152 * Linux >= 3.18, d_invalidate will detach the mountpoint and prevent
1153 * anyone automount it again as long as someone is still using the
1156 if (zp
->z_is_ctldir
)
1159 zh
= zfs_znode_hold_enter(zsb
, obj_num
);
1161 mutex_enter(&zp
->z_acl_lock
);
1162 if (zp
->z_acl_cached
) {
1163 zfs_acl_free(zp
->z_acl_cached
);
1164 zp
->z_acl_cached
= NULL
;
1166 mutex_exit(&zp
->z_acl_lock
);
1168 rw_enter(&zp
->z_xattr_lock
, RW_WRITER
);
1169 if (zp
->z_xattr_cached
) {
1170 nvlist_free(zp
->z_xattr_cached
);
1171 zp
->z_xattr_cached
= NULL
;
1173 rw_exit(&zp
->z_xattr_lock
);
1175 ASSERT(zp
->z_sa_hdl
== NULL
);
1176 err
= sa_buf_hold(zsb
->z_os
, obj_num
, NULL
, &db
);
1178 zfs_znode_hold_exit(zsb
, zh
);
1182 dmu_object_info_from_db(db
, &doi
);
1183 if (doi
.doi_bonus_type
!= DMU_OT_SA
&&
1184 (doi
.doi_bonus_type
!= DMU_OT_ZNODE
||
1185 (doi
.doi_bonus_type
== DMU_OT_ZNODE
&&
1186 doi
.doi_bonus_size
< sizeof (znode_phys_t
)))) {
1187 sa_buf_rele(db
, NULL
);
1188 zfs_znode_hold_exit(zsb
, zh
);
1189 return (SET_ERROR(EINVAL
));
1192 zfs_znode_sa_init(zsb
, zp
, db
, doi
.doi_bonus_type
, NULL
);
1194 /* reload cached values */
1195 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GEN(zsb
), NULL
,
1196 &gen
, sizeof (gen
));
1197 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zsb
), NULL
,
1198 &zp
->z_size
, sizeof (zp
->z_size
));
1199 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_LINKS(zsb
), NULL
,
1200 &zp
->z_links
, sizeof (zp
->z_links
));
1201 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
1202 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
1203 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zsb
), NULL
,
1204 &zp
->z_uid
, sizeof (zp
->z_uid
));
1205 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zsb
), NULL
,
1206 &zp
->z_gid
, sizeof (zp
->z_gid
));
1207 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
), NULL
,
1208 &mode
, sizeof (mode
));
1210 if (sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) {
1211 zfs_znode_dmu_fini(zp
);
1212 zfs_znode_hold_exit(zsb
, zh
);
1213 return (SET_ERROR(EIO
));
1218 if (gen
!= zp
->z_gen
) {
1219 zfs_znode_dmu_fini(zp
);
1220 zfs_znode_hold_exit(zsb
, zh
);
1221 return (SET_ERROR(EIO
));
1224 zp
->z_unlinked
= (zp
->z_links
== 0);
1225 zp
->z_blksz
= doi
.doi_data_block_size
;
1226 zp
->z_atime_dirty
= 0;
1227 zfs_inode_update_new(zp
);
1229 zfs_znode_hold_exit(zsb
, zh
);
1235 zfs_znode_delete(znode_t
*zp
, dmu_tx_t
*tx
)
1237 zfs_sb_t
*zsb
= ZTOZSB(zp
);
1238 objset_t
*os
= zsb
->z_os
;
1239 uint64_t obj
= zp
->z_id
;
1240 uint64_t acl_obj
= zfs_external_acl(zp
);
1243 zh
= zfs_znode_hold_enter(zsb
, obj
);
1245 VERIFY(!zp
->z_is_sa
);
1246 VERIFY(0 == dmu_object_free(os
, acl_obj
, tx
));
1248 VERIFY(0 == dmu_object_free(os
, obj
, tx
));
1249 zfs_znode_dmu_fini(zp
);
1250 zfs_znode_hold_exit(zsb
, zh
);
1254 zfs_zinactive(znode_t
*zp
)
1256 zfs_sb_t
*zsb
= ZTOZSB(zp
);
1257 uint64_t z_id
= zp
->z_id
;
1260 ASSERT(zp
->z_sa_hdl
);
1263 * Don't allow a zfs_zget() while were trying to release this znode.
1265 zh
= zfs_znode_hold_enter(zsb
, z_id
);
1267 mutex_enter(&zp
->z_lock
);
1270 * If this was the last reference to a file with no links,
1271 * remove the file from the file system.
1273 if (zp
->z_unlinked
) {
1274 mutex_exit(&zp
->z_lock
);
1275 zfs_znode_hold_exit(zsb
, zh
);
1280 mutex_exit(&zp
->z_lock
);
1281 zfs_znode_dmu_fini(zp
);
1283 zfs_znode_hold_exit(zsb
, zh
);
1287 zfs_compare_timespec(struct timespec
*t1
, struct timespec
*t2
)
1289 if (t1
->tv_sec
< t2
->tv_sec
)
1292 if (t1
->tv_sec
> t2
->tv_sec
)
1295 return (t1
->tv_nsec
- t2
->tv_nsec
);
1299 * Prepare to update znode time stamps.
1301 * IN: zp - znode requiring timestamp update
1302 * flag - ATTR_MTIME, ATTR_CTIME flags
1308 * Note: We don't update atime here, because we rely on Linux VFS to do
1312 zfs_tstamp_update_setup(znode_t
*zp
, uint_t flag
, uint64_t mtime
[2],
1321 if (flag
& ATTR_MTIME
) {
1322 ZFS_TIME_ENCODE(&now
, mtime
);
1323 if (ZTOZSB(zp
)->z_use_fuids
) {
1324 zp
->z_pflags
|= (ZFS_ARCHIVE
|
1329 if (flag
& ATTR_CTIME
) {
1330 ZFS_TIME_ENCODE(&now
, ctime
);
1331 if (ZTOZSB(zp
)->z_use_fuids
)
1332 zp
->z_pflags
|= ZFS_ARCHIVE
;
1337 * Grow the block size for a file.
1339 * IN: zp - znode of file to free data in.
1340 * size - requested block size
1341 * tx - open transaction.
1343 * NOTE: this function assumes that the znode is write locked.
1346 zfs_grow_blocksize(znode_t
*zp
, uint64_t size
, dmu_tx_t
*tx
)
1351 if (size
<= zp
->z_blksz
)
1354 * If the file size is already greater than the current blocksize,
1355 * we will not grow. If there is more than one block in a file,
1356 * the blocksize cannot change.
1358 if (zp
->z_blksz
&& zp
->z_size
> zp
->z_blksz
)
1361 error
= dmu_object_set_blocksize(ZTOZSB(zp
)->z_os
, zp
->z_id
,
1364 if (error
== ENOTSUP
)
1368 /* What blocksize did we actually get? */
1369 dmu_object_size_from_db(sa_get_db(zp
->z_sa_hdl
), &zp
->z_blksz
, &dummy
);
1373 * Increase the file length
1375 * IN: zp - znode of file to free data in.
1376 * end - new end-of-file
1378 * RETURN: 0 on success, error code on failure
1381 zfs_extend(znode_t
*zp
, uint64_t end
)
1383 zfs_sb_t
*zsb
= ZTOZSB(zp
);
1390 * We will change zp_size, lock the whole file.
1392 rl
= zfs_range_lock(&zp
->z_range_lock
, 0, UINT64_MAX
, RL_WRITER
);
1395 * Nothing to do if file already at desired length.
1397 if (end
<= zp
->z_size
) {
1398 zfs_range_unlock(rl
);
1401 tx
= dmu_tx_create(zsb
->z_os
);
1402 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1403 zfs_sa_upgrade_txholds(tx
, zp
);
1404 if (end
> zp
->z_blksz
&&
1405 (!ISP2(zp
->z_blksz
) || zp
->z_blksz
< zsb
->z_max_blksz
)) {
1407 * We are growing the file past the current block size.
1409 if (zp
->z_blksz
> ZTOZSB(zp
)->z_max_blksz
) {
1411 * File's blocksize is already larger than the
1412 * "recordsize" property. Only let it grow to
1413 * the next power of 2.
1415 ASSERT(!ISP2(zp
->z_blksz
));
1416 newblksz
= MIN(end
, 1 << highbit64(zp
->z_blksz
));
1418 newblksz
= MIN(end
, ZTOZSB(zp
)->z_max_blksz
);
1420 dmu_tx_hold_write(tx
, zp
->z_id
, 0, newblksz
);
1425 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1428 zfs_range_unlock(rl
);
1433 zfs_grow_blocksize(zp
, newblksz
, tx
);
1437 VERIFY(0 == sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(ZTOZSB(zp
)),
1438 &zp
->z_size
, sizeof (zp
->z_size
), tx
));
1440 zfs_range_unlock(rl
);
1448 * zfs_zero_partial_page - Modeled after update_pages() but
1449 * with different arguments and semantics for use by zfs_freesp().
1451 * Zeroes a piece of a single page cache entry for zp at offset
1452 * start and length len.
1454 * Caller must acquire a range lock on the file for the region
1455 * being zeroed in order that the ARC and page cache stay in sync.
1458 zfs_zero_partial_page(znode_t
*zp
, uint64_t start
, uint64_t len
)
1460 struct address_space
*mp
= ZTOI(zp
)->i_mapping
;
1465 ASSERT((start
& PAGE_MASK
) == ((start
+ len
- 1) & PAGE_MASK
));
1467 off
= start
& (PAGE_SIZE
- 1);
1470 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
1472 if (mapping_writably_mapped(mp
))
1473 flush_dcache_page(pp
);
1476 bzero(pb
+ off
, len
);
1479 if (mapping_writably_mapped(mp
))
1480 flush_dcache_page(pp
);
1482 mark_page_accessed(pp
);
1483 SetPageUptodate(pp
);
1491 * Free space in a file.
1493 * IN: zp - znode of file to free data in.
1494 * off - start of section to free.
1495 * len - length of section to free.
1497 * RETURN: 0 on success, error code on failure
1500 zfs_free_range(znode_t
*zp
, uint64_t off
, uint64_t len
)
1502 zfs_sb_t
*zsb
= ZTOZSB(zp
);
1507 * Lock the range being freed.
1509 rl
= zfs_range_lock(&zp
->z_range_lock
, off
, len
, RL_WRITER
);
1512 * Nothing to do if file already at desired length.
1514 if (off
>= zp
->z_size
) {
1515 zfs_range_unlock(rl
);
1519 if (off
+ len
> zp
->z_size
)
1520 len
= zp
->z_size
- off
;
1522 error
= dmu_free_long_range(zsb
->z_os
, zp
->z_id
, off
, len
);
1525 * Zero partial page cache entries. This must be done under a
1526 * range lock in order to keep the ARC and page cache in sync.
1528 if (zp
->z_is_mapped
) {
1529 loff_t first_page
, last_page
, page_len
;
1530 loff_t first_page_offset
, last_page_offset
;
1532 /* first possible full page in hole */
1533 first_page
= (off
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1534 /* last page of hole */
1535 last_page
= (off
+ len
) >> PAGE_SHIFT
;
1537 /* offset of first_page */
1538 first_page_offset
= first_page
<< PAGE_SHIFT
;
1539 /* offset of last_page */
1540 last_page_offset
= last_page
<< PAGE_SHIFT
;
1542 /* truncate whole pages */
1543 if (last_page_offset
> first_page_offset
) {
1544 truncate_inode_pages_range(ZTOI(zp
)->i_mapping
,
1545 first_page_offset
, last_page_offset
- 1);
1548 /* truncate sub-page ranges */
1549 if (first_page
> last_page
) {
1550 /* entire punched area within a single page */
1551 zfs_zero_partial_page(zp
, off
, len
);
1553 /* beginning of punched area at the end of a page */
1554 page_len
= first_page_offset
- off
;
1556 zfs_zero_partial_page(zp
, off
, page_len
);
1558 /* end of punched area at the beginning of a page */
1559 page_len
= off
+ len
- last_page_offset
;
1561 zfs_zero_partial_page(zp
, last_page_offset
,
1565 zfs_range_unlock(rl
);
1573 * IN: zp - znode of file to free data in.
1574 * end - new end-of-file.
1576 * RETURN: 0 on success, error code on failure
1579 zfs_trunc(znode_t
*zp
, uint64_t end
)
1581 zfs_sb_t
*zsb
= ZTOZSB(zp
);
1585 sa_bulk_attr_t bulk
[2];
1589 * We will change zp_size, lock the whole file.
1591 rl
= zfs_range_lock(&zp
->z_range_lock
, 0, UINT64_MAX
, RL_WRITER
);
1594 * Nothing to do if file already at desired length.
1596 if (end
>= zp
->z_size
) {
1597 zfs_range_unlock(rl
);
1601 error
= dmu_free_long_range(zsb
->z_os
, zp
->z_id
, end
, -1);
1603 zfs_range_unlock(rl
);
1606 tx
= dmu_tx_create(zsb
->z_os
);
1607 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1608 zfs_sa_upgrade_txholds(tx
, zp
);
1609 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1612 zfs_range_unlock(rl
);
1617 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zsb
),
1618 NULL
, &zp
->z_size
, sizeof (zp
->z_size
));
1621 zp
->z_pflags
&= ~ZFS_SPARSE
;
1622 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
),
1623 NULL
, &zp
->z_pflags
, 8);
1625 VERIFY(sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
) == 0);
1629 zfs_range_unlock(rl
);
1635 * Free space in a file
1637 * IN: zp - znode of file to free data in.
1638 * off - start of range
1639 * len - end of range (0 => EOF)
1640 * flag - current file open mode flags.
1641 * log - TRUE if this action should be logged
1643 * RETURN: 0 on success, error code on failure
1646 zfs_freesp(znode_t
*zp
, uint64_t off
, uint64_t len
, int flag
, boolean_t log
)
1649 zfs_sb_t
*zsb
= ZTOZSB(zp
);
1650 zilog_t
*zilog
= zsb
->z_log
;
1652 uint64_t mtime
[2], ctime
[2];
1653 sa_bulk_attr_t bulk
[3];
1657 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_MODE(zsb
), &mode
,
1658 sizeof (mode
))) != 0)
1661 if (off
> zp
->z_size
) {
1662 error
= zfs_extend(zp
, off
+len
);
1663 if (error
== 0 && log
)
1669 error
= zfs_trunc(zp
, off
);
1671 if ((error
= zfs_free_range(zp
, off
, len
)) == 0 &&
1672 off
+ len
> zp
->z_size
)
1673 error
= zfs_extend(zp
, off
+len
);
1678 tx
= dmu_tx_create(zsb
->z_os
);
1679 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1680 zfs_sa_upgrade_txholds(tx
, zp
);
1681 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1687 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, mtime
, 16);
1688 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, ctime
, 16);
1689 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
),
1690 NULL
, &zp
->z_pflags
, 8);
1691 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
);
1692 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
1695 zfs_log_truncate(zilog
, tx
, TX_TRUNCATE
, zp
, off
, len
);
1699 zfs_inode_update(zp
);
1704 * Truncate the page cache - for file truncate operations, use
1705 * the purpose-built API for truncations. For punching operations,
1706 * the truncation is handled under a range lock in zfs_free_range.
1709 truncate_setsize(ZTOI(zp
), off
);
1714 zfs_create_fs(objset_t
*os
, cred_t
*cr
, nvlist_t
*zplprops
, dmu_tx_t
*tx
)
1716 struct super_block
*sb
;
1718 uint64_t moid
, obj
, sa_obj
, version
;
1719 uint64_t sense
= ZFS_CASE_SENSITIVE
;
1725 znode_t
*rootzp
= NULL
;
1728 zfs_acl_ids_t acl_ids
;
1731 * First attempt to create master node.
1734 * In an empty objset, there are no blocks to read and thus
1735 * there can be no i/o errors (which we assert below).
1737 moid
= MASTER_NODE_OBJ
;
1738 error
= zap_create_claim(os
, moid
, DMU_OT_MASTER_NODE
,
1739 DMU_OT_NONE
, 0, tx
);
1743 * Set starting attributes.
1745 version
= zfs_zpl_version_map(spa_version(dmu_objset_spa(os
)));
1747 while ((elem
= nvlist_next_nvpair(zplprops
, elem
)) != NULL
) {
1748 /* For the moment we expect all zpl props to be uint64_ts */
1752 ASSERT(nvpair_type(elem
) == DATA_TYPE_UINT64
);
1753 VERIFY(nvpair_value_uint64(elem
, &val
) == 0);
1754 name
= nvpair_name(elem
);
1755 if (strcmp(name
, zfs_prop_to_name(ZFS_PROP_VERSION
)) == 0) {
1759 error
= zap_update(os
, moid
, name
, 8, 1, &val
, tx
);
1762 if (strcmp(name
, zfs_prop_to_name(ZFS_PROP_NORMALIZE
)) == 0)
1764 else if (strcmp(name
, zfs_prop_to_name(ZFS_PROP_CASE
)) == 0)
1767 ASSERT(version
!= 0);
1768 error
= zap_update(os
, moid
, ZPL_VERSION_STR
, 8, 1, &version
, tx
);
1771 * Create zap object used for SA attribute registration
1774 if (version
>= ZPL_VERSION_SA
) {
1775 sa_obj
= zap_create(os
, DMU_OT_SA_MASTER_NODE
,
1776 DMU_OT_NONE
, 0, tx
);
1777 error
= zap_add(os
, moid
, ZFS_SA_ATTRS
, 8, 1, &sa_obj
, tx
);
1783 * Create a delete queue.
1785 obj
= zap_create(os
, DMU_OT_UNLINKED_SET
, DMU_OT_NONE
, 0, tx
);
1787 error
= zap_add(os
, moid
, ZFS_UNLINKED_SET
, 8, 1, &obj
, tx
);
1791 * Create root znode. Create minimal znode/inode/zsb/sb
1792 * to allow zfs_mknode to work.
1794 vattr
.va_mask
= ATTR_MODE
|ATTR_UID
|ATTR_GID
;
1795 vattr
.va_mode
= S_IFDIR
|0755;
1796 vattr
.va_uid
= crgetuid(cr
);
1797 vattr
.va_gid
= crgetgid(cr
);
1799 rootzp
= kmem_cache_alloc(znode_cache
, KM_SLEEP
);
1800 rootzp
->z_moved
= 0;
1801 rootzp
->z_unlinked
= 0;
1802 rootzp
->z_atime_dirty
= 0;
1803 rootzp
->z_is_sa
= USE_SA(version
, os
);
1805 zsb
= kmem_zalloc(sizeof (zfs_sb_t
), KM_SLEEP
);
1807 zsb
->z_parent
= zsb
;
1808 zsb
->z_version
= version
;
1809 zsb
->z_use_fuids
= USE_FUIDS(version
, os
);
1810 zsb
->z_use_sa
= USE_SA(version
, os
);
1813 sb
= kmem_zalloc(sizeof (struct super_block
), KM_SLEEP
);
1814 sb
->s_fs_info
= zsb
;
1816 ZTOI(rootzp
)->i_sb
= sb
;
1818 error
= sa_setup(os
, sa_obj
, zfs_attr_table
, ZPL_END
,
1819 &zsb
->z_attr_table
);
1824 * Fold case on file systems that are always or sometimes case
1827 if (sense
== ZFS_CASE_INSENSITIVE
|| sense
== ZFS_CASE_MIXED
)
1828 zsb
->z_norm
|= U8_TEXTPREP_TOUPPER
;
1830 mutex_init(&zsb
->z_znodes_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1831 list_create(&zsb
->z_all_znodes
, sizeof (znode_t
),
1832 offsetof(znode_t
, z_link_node
));
1834 size
= MIN(1 << (highbit64(zfs_object_mutex_size
)-1), ZFS_OBJ_MTX_MAX
);
1835 zsb
->z_hold_size
= size
;
1836 zsb
->z_hold_trees
= vmem_zalloc(sizeof (avl_tree_t
) * size
, KM_SLEEP
);
1837 zsb
->z_hold_locks
= vmem_zalloc(sizeof (kmutex_t
) * size
, KM_SLEEP
);
1838 for (i
= 0; i
!= size
; i
++) {
1839 avl_create(&zsb
->z_hold_trees
[i
], zfs_znode_hold_compare
,
1840 sizeof (znode_hold_t
), offsetof(znode_hold_t
, zh_node
));
1841 mutex_init(&zsb
->z_hold_locks
[i
], NULL
, MUTEX_DEFAULT
, NULL
);
1844 VERIFY(0 == zfs_acl_ids_create(rootzp
, IS_ROOT_NODE
, &vattr
,
1845 cr
, NULL
, &acl_ids
));
1846 zfs_mknode(rootzp
, &vattr
, tx
, cr
, IS_ROOT_NODE
, &zp
, &acl_ids
);
1847 ASSERT3P(zp
, ==, rootzp
);
1848 error
= zap_add(os
, moid
, ZFS_ROOT_OBJ
, 8, 1, &rootzp
->z_id
, tx
);
1850 zfs_acl_ids_free(&acl_ids
);
1852 atomic_set(&ZTOI(rootzp
)->i_count
, 0);
1853 sa_handle_destroy(rootzp
->z_sa_hdl
);
1854 kmem_cache_free(znode_cache
, rootzp
);
1857 * Create shares directory
1859 error
= zfs_create_share_dir(zsb
, tx
);
1862 for (i
= 0; i
!= size
; i
++) {
1863 avl_destroy(&zsb
->z_hold_trees
[i
]);
1864 mutex_destroy(&zsb
->z_hold_locks
[i
]);
1867 vmem_free(zsb
->z_hold_trees
, sizeof (avl_tree_t
) * size
);
1868 vmem_free(zsb
->z_hold_locks
, sizeof (kmutex_t
) * size
);
1869 kmem_free(sb
, sizeof (struct super_block
));
1870 kmem_free(zsb
, sizeof (zfs_sb_t
));
1872 #endif /* _KERNEL */
1875 zfs_sa_setup(objset_t
*osp
, sa_attr_type_t
**sa_table
)
1877 uint64_t sa_obj
= 0;
1880 error
= zap_lookup(osp
, MASTER_NODE_OBJ
, ZFS_SA_ATTRS
, 8, 1, &sa_obj
);
1881 if (error
!= 0 && error
!= ENOENT
)
1884 error
= sa_setup(osp
, sa_obj
, zfs_attr_table
, ZPL_END
, sa_table
);
1889 zfs_grab_sa_handle(objset_t
*osp
, uint64_t obj
, sa_handle_t
**hdlp
,
1890 dmu_buf_t
**db
, void *tag
)
1892 dmu_object_info_t doi
;
1895 if ((error
= sa_buf_hold(osp
, obj
, tag
, db
)) != 0)
1898 dmu_object_info_from_db(*db
, &doi
);
1899 if ((doi
.doi_bonus_type
!= DMU_OT_SA
&&
1900 doi
.doi_bonus_type
!= DMU_OT_ZNODE
) ||
1901 (doi
.doi_bonus_type
== DMU_OT_ZNODE
&&
1902 doi
.doi_bonus_size
< sizeof (znode_phys_t
))) {
1903 sa_buf_rele(*db
, tag
);
1904 return (SET_ERROR(ENOTSUP
));
1907 error
= sa_handle_get(osp
, obj
, NULL
, SA_HDL_PRIVATE
, hdlp
);
1909 sa_buf_rele(*db
, tag
);
1917 zfs_release_sa_handle(sa_handle_t
*hdl
, dmu_buf_t
*db
, void *tag
)
1919 sa_handle_destroy(hdl
);
1920 sa_buf_rele(db
, tag
);
1924 * Given an object number, return its parent object number and whether
1925 * or not the object is an extended attribute directory.
1928 zfs_obj_to_pobj(sa_handle_t
*hdl
, sa_attr_type_t
*sa_table
, uint64_t *pobjp
,
1934 sa_bulk_attr_t bulk
[3];
1938 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_PARENT
], NULL
,
1939 &parent
, sizeof (parent
));
1940 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_FLAGS
], NULL
,
1941 &pflags
, sizeof (pflags
));
1942 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_MODE
], NULL
,
1943 &mode
, sizeof (mode
));
1945 if ((error
= sa_bulk_lookup(hdl
, bulk
, count
)) != 0)
1949 *is_xattrdir
= ((pflags
& ZFS_XATTR
) != 0) && S_ISDIR(mode
);
1955 * Given an object number, return some zpl level statistics
1958 zfs_obj_to_stats_impl(sa_handle_t
*hdl
, sa_attr_type_t
*sa_table
,
1961 sa_bulk_attr_t bulk
[4];
1964 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_MODE
], NULL
,
1965 &sb
->zs_mode
, sizeof (sb
->zs_mode
));
1966 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_GEN
], NULL
,
1967 &sb
->zs_gen
, sizeof (sb
->zs_gen
));
1968 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_LINKS
], NULL
,
1969 &sb
->zs_links
, sizeof (sb
->zs_links
));
1970 SA_ADD_BULK_ATTR(bulk
, count
, sa_table
[ZPL_CTIME
], NULL
,
1971 &sb
->zs_ctime
, sizeof (sb
->zs_ctime
));
1973 return (sa_bulk_lookup(hdl
, bulk
, count
));
1977 zfs_obj_to_path_impl(objset_t
*osp
, uint64_t obj
, sa_handle_t
*hdl
,
1978 sa_attr_type_t
*sa_table
, char *buf
, int len
)
1980 sa_handle_t
*sa_hdl
;
1981 sa_handle_t
*prevhdl
= NULL
;
1982 dmu_buf_t
*prevdb
= NULL
;
1983 dmu_buf_t
*sa_db
= NULL
;
1984 char *path
= buf
+ len
- 1;
1992 char component
[MAXNAMELEN
+ 2];
1994 int is_xattrdir
= 0;
1997 zfs_release_sa_handle(prevhdl
, prevdb
, FTAG
);
1999 if ((error
= zfs_obj_to_pobj(sa_hdl
, sa_table
, &pobj
,
2000 &is_xattrdir
)) != 0)
2011 (void) sprintf(component
+ 1, "<xattrdir>");
2013 error
= zap_value_search(osp
, pobj
, obj
,
2014 ZFS_DIRENT_OBJ(-1ULL), component
+ 1);
2019 complen
= strlen(component
);
2021 ASSERT(path
>= buf
);
2022 bcopy(component
, path
, complen
);
2025 if (sa_hdl
!= hdl
) {
2029 error
= zfs_grab_sa_handle(osp
, obj
, &sa_hdl
, &sa_db
, FTAG
);
2037 if (sa_hdl
!= NULL
&& sa_hdl
!= hdl
) {
2038 ASSERT(sa_db
!= NULL
);
2039 zfs_release_sa_handle(sa_hdl
, sa_db
, FTAG
);
2043 (void) memmove(buf
, path
, buf
+ len
- path
);
2049 zfs_obj_to_path(objset_t
*osp
, uint64_t obj
, char *buf
, int len
)
2051 sa_attr_type_t
*sa_table
;
2056 error
= zfs_sa_setup(osp
, &sa_table
);
2060 error
= zfs_grab_sa_handle(osp
, obj
, &hdl
, &db
, FTAG
);
2064 error
= zfs_obj_to_path_impl(osp
, obj
, hdl
, sa_table
, buf
, len
);
2066 zfs_release_sa_handle(hdl
, db
, FTAG
);
2071 zfs_obj_to_stats(objset_t
*osp
, uint64_t obj
, zfs_stat_t
*sb
,
2074 char *path
= buf
+ len
- 1;
2075 sa_attr_type_t
*sa_table
;
2082 error
= zfs_sa_setup(osp
, &sa_table
);
2086 error
= zfs_grab_sa_handle(osp
, obj
, &hdl
, &db
, FTAG
);
2090 error
= zfs_obj_to_stats_impl(hdl
, sa_table
, sb
);
2092 zfs_release_sa_handle(hdl
, db
, FTAG
);
2096 error
= zfs_obj_to_path_impl(osp
, obj
, hdl
, sa_table
, buf
, len
);
2098 zfs_release_sa_handle(hdl
, db
, FTAG
);
2102 #if defined(_KERNEL) && defined(HAVE_SPL)
2103 EXPORT_SYMBOL(zfs_create_fs
);
2104 EXPORT_SYMBOL(zfs_obj_to_path
);
2106 module_param(zfs_object_mutex_size
, uint
, 0644);
2107 MODULE_PARM_DESC(zfs_object_mutex_size
, "Size of znode hold array");