4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */
30 #include <sys/types.h>
31 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
36 #include <sys/mntent.h>
37 #include <sys/u8_textprep.h>
38 #include <sys/dsl_dataset.h>
40 #include <sys/vnode.h>
43 #include <sys/errno.h>
44 #include <sys/unistd.h>
45 #include <sys/atomic.h>
46 #include <sys/zfs_dir.h>
47 #include <sys/zfs_acl.h>
48 #include <sys/zfs_ioctl.h>
49 #include <sys/zfs_rlock.h>
50 #include <sys/zfs_fuid.h>
51 #include <sys/dnode.h>
52 #include <sys/fs/zfs.h>
53 #include <sys/kidmap.h>
57 #include <sys/refcount.h>
60 #include <sys/zfs_znode.h>
62 #include <sys/zfs_sa.h>
63 #include <sys/zfs_stat.h>
64 #include <sys/refcount.h>
67 #include "zfs_comutil.h"
69 /* Used by fstat(1). */
70 SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 0, sizeof(znode_t),
74 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
75 * turned on when DEBUG is also defined.
82 #define ZNODE_STAT_ADD(stat) ((stat)++)
84 #define ZNODE_STAT_ADD(stat) /* nothing */
85 #endif /* ZNODE_STATS */
88 * Functions needed for userland (ie: libzpool) are not put under
89 * #ifdef_KERNEL; the rest of the functions have dependencies
90 * (such as VFS logic) that will not compile easily in userland.
94 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to
95 * be freed before it can be safely accessed.
97 krwlock_t zfsvfs_lock;
99 static kmem_cache_t *znode_cache = NULL;
103 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
106 * We should never drop all dbuf refs without first clearing
107 * the eviction callback.
109 panic("evicting znode %p\n", user_ptr);
112 extern struct vop_vector zfs_vnodeops;
113 extern struct vop_vector zfs_fifoops;
114 extern struct vop_vector zfs_shareops;
117 * XXX: We cannot use this function as a cache constructor, because
118 * there is one global cache for all file systems and we need
119 * to pass vfsp here, which is not possible, because argument
120 * 'cdrarg' is defined at kmem_cache_create() time.
124 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
131 POINTER_INVALIDATE(&zp->z_zfsvfs);
132 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
135 error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp);
136 if (error != 0 && (kmflags & KM_NOSLEEP))
139 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
141 vp->v_data = (caddr_t)zp;
148 list_link_init(&zp->z_link_node);
150 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
151 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
152 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
153 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
155 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
156 avl_create(&zp->z_range_avl, zfs_range_compare,
157 sizeof (rl_t), offsetof(rl_t, r_node));
159 zp->z_dirlocks = NULL;
160 zp->z_acl_cached = NULL;
167 zfs_znode_cache_destructor(void *buf, void *arg)
171 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
172 ASSERT(ZTOV(zp) == NULL);
174 ASSERT(!list_link_active(&zp->z_link_node));
175 mutex_destroy(&zp->z_lock);
176 rw_destroy(&zp->z_parent_lock);
177 rw_destroy(&zp->z_name_lock);
178 mutex_destroy(&zp->z_acl_lock);
179 avl_destroy(&zp->z_range_avl);
180 mutex_destroy(&zp->z_range_lock);
182 ASSERT(zp->z_dirlocks == NULL);
183 ASSERT(zp->z_acl_cached == NULL);
188 uint64_t zms_zfsvfs_invalid;
189 uint64_t zms_zfsvfs_recheck1;
190 uint64_t zms_zfsvfs_unmounted;
191 uint64_t zms_zfsvfs_recheck2;
192 uint64_t zms_obj_held;
193 uint64_t zms_vnode_locked;
194 uint64_t zms_not_only_dnlc;
196 #endif /* ZNODE_STATS */
200 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
205 nzp->z_zfsvfs = ozp->z_zfsvfs;
209 nzp->z_vnode = ozp->z_vnode;
210 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
211 ZTOV(ozp)->v_data = ozp;
212 ZTOV(nzp)->v_data = nzp;
214 nzp->z_id = ozp->z_id;
215 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
216 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0);
217 nzp->z_unlinked = ozp->z_unlinked;
218 nzp->z_atime_dirty = ozp->z_atime_dirty;
219 nzp->z_zn_prefetch = ozp->z_zn_prefetch;
220 nzp->z_blksz = ozp->z_blksz;
221 nzp->z_seq = ozp->z_seq;
222 nzp->z_mapcnt = ozp->z_mapcnt;
223 nzp->z_gen = ozp->z_gen;
224 nzp->z_sync_cnt = ozp->z_sync_cnt;
225 nzp->z_is_sa = ozp->z_is_sa;
226 nzp->z_sa_hdl = ozp->z_sa_hdl;
227 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
228 nzp->z_links = ozp->z_links;
229 nzp->z_size = ozp->z_size;
230 nzp->z_pflags = ozp->z_pflags;
231 nzp->z_uid = ozp->z_uid;
232 nzp->z_gid = ozp->z_gid;
233 nzp->z_mode = ozp->z_mode;
236 * Since this is just an idle znode and kmem is already dealing with
237 * memory pressure, release any cached ACL.
239 if (ozp->z_acl_cached) {
240 zfs_acl_free(ozp->z_acl_cached);
241 ozp->z_acl_cached = NULL;
244 sa_set_userp(nzp->z_sa_hdl, nzp);
247 * Invalidate the original znode by clearing fields that provide a
248 * pointer back to the znode. Set the low bit of the vfs pointer to
249 * ensure that zfs_znode_move() recognizes the znode as invalid in any
250 * subsequent callback.
252 ozp->z_sa_hdl = NULL;
253 POINTER_INVALIDATE(&ozp->z_zfsvfs);
259 ozp->z_moved = (uint8_t)-1;
264 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
266 znode_t *ozp = buf, *nzp = newbuf;
271 * The znode is on the file system's list of known znodes if the vfs
272 * pointer is valid. We set the low bit of the vfs pointer when freeing
273 * the znode to invalidate it, and the memory patterns written by kmem
274 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
275 * created znode sets the vfs pointer last of all to indicate that the
276 * znode is known and in a valid state to be moved by this function.
278 zfsvfs = ozp->z_zfsvfs;
279 if (!POINTER_IS_VALID(zfsvfs)) {
280 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
281 return (KMEM_CBRC_DONT_KNOW);
285 * Close a small window in which it's possible that the filesystem could
286 * be unmounted and freed, and zfsvfs, though valid in the previous
287 * statement, could point to unrelated memory by the time we try to
288 * prevent the filesystem from being unmounted.
290 rw_enter(&zfsvfs_lock, RW_WRITER);
291 if (zfsvfs != ozp->z_zfsvfs) {
292 rw_exit(&zfsvfs_lock);
293 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
294 return (KMEM_CBRC_DONT_KNOW);
298 * If the znode is still valid, then so is the file system. We know that
299 * no valid file system can be freed while we hold zfsvfs_lock, so we
300 * can safely ensure that the filesystem is not and will not be
301 * unmounted. The next statement is equivalent to ZFS_ENTER().
303 rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
304 if (zfsvfs->z_unmounted) {
306 rw_exit(&zfsvfs_lock);
307 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
308 return (KMEM_CBRC_DONT_KNOW);
310 rw_exit(&zfsvfs_lock);
312 mutex_enter(&zfsvfs->z_znodes_lock);
314 * Recheck the vfs pointer in case the znode was removed just before
315 * acquiring the lock.
317 if (zfsvfs != ozp->z_zfsvfs) {
318 mutex_exit(&zfsvfs->z_znodes_lock);
320 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
321 return (KMEM_CBRC_DONT_KNOW);
325 * At this point we know that as long as we hold z_znodes_lock, the
326 * znode cannot be freed and fields within the znode can be safely
327 * accessed. Now, prevent a race with zfs_zget().
329 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
330 mutex_exit(&zfsvfs->z_znodes_lock);
332 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
333 return (KMEM_CBRC_LATER);
337 if (mutex_tryenter(&vp->v_lock) == 0) {
338 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
339 mutex_exit(&zfsvfs->z_znodes_lock);
341 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
342 return (KMEM_CBRC_LATER);
345 /* Only move znodes that are referenced _only_ by the DNLC. */
346 if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
347 mutex_exit(&vp->v_lock);
348 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
349 mutex_exit(&zfsvfs->z_znodes_lock);
351 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
352 return (KMEM_CBRC_LATER);
356 * The znode is known and in a valid state to move. We're holding the
357 * locks needed to execute the critical section.
359 zfs_znode_move_impl(ozp, nzp);
360 mutex_exit(&vp->v_lock);
361 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
363 list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
364 mutex_exit(&zfsvfs->z_znodes_lock);
367 return (KMEM_CBRC_YES);
377 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
378 ASSERT(znode_cache == NULL);
379 znode_cache = kmem_cache_create("zfs_znode_cache",
380 sizeof (znode_t), 0, /* zfs_znode_cache_constructor */ NULL,
381 zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
382 kmem_cache_set_move(znode_cache, zfs_znode_move);
390 * Cleanup vfs & vnode ops
392 zfs_remove_op_tables();
399 kmem_cache_destroy(znode_cache);
401 rw_destroy(&zfsvfs_lock);
405 struct vnodeops *zfs_dvnodeops;
406 struct vnodeops *zfs_fvnodeops;
407 struct vnodeops *zfs_symvnodeops;
408 struct vnodeops *zfs_xdvnodeops;
409 struct vnodeops *zfs_evnodeops;
410 struct vnodeops *zfs_sharevnodeops;
413 zfs_remove_op_tables()
419 (void) vfs_freevfsops_by_type(zfsfstype);
426 vn_freevnodeops(zfs_dvnodeops);
428 vn_freevnodeops(zfs_fvnodeops);
430 vn_freevnodeops(zfs_symvnodeops);
432 vn_freevnodeops(zfs_xdvnodeops);
434 vn_freevnodeops(zfs_evnodeops);
435 if (zfs_sharevnodeops)
436 vn_freevnodeops(zfs_sharevnodeops);
438 zfs_dvnodeops = NULL;
439 zfs_fvnodeops = NULL;
440 zfs_symvnodeops = NULL;
441 zfs_xdvnodeops = NULL;
442 zfs_evnodeops = NULL;
443 zfs_sharevnodeops = NULL;
446 extern const fs_operation_def_t zfs_dvnodeops_template[];
447 extern const fs_operation_def_t zfs_fvnodeops_template[];
448 extern const fs_operation_def_t zfs_xdvnodeops_template[];
449 extern const fs_operation_def_t zfs_symvnodeops_template[];
450 extern const fs_operation_def_t zfs_evnodeops_template[];
451 extern const fs_operation_def_t zfs_sharevnodeops_template[];
454 zfs_create_op_tables()
459 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
460 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
461 * In this case we just return as the ops vectors are already set up.
466 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
471 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
476 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
481 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
486 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
491 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
499 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
501 zfs_acl_ids_t acl_ids;
508 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
509 vattr.va_type = VDIR;
510 vattr.va_mode = S_IFDIR|0555;
511 vattr.va_uid = crgetuid(kcred);
512 vattr.va_gid = crgetgid(kcred);
514 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
515 zfs_znode_cache_constructor(sharezp, zfsvfs->z_parent->z_vfs, 0);
516 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
517 sharezp->z_moved = 0;
518 sharezp->z_unlinked = 0;
519 sharezp->z_atime_dirty = 0;
520 sharezp->z_zfsvfs = zfsvfs;
521 sharezp->z_is_sa = zfsvfs->z_use_sa;
523 sharezp->z_vnode = &vnode;
524 vnode.v_data = sharezp;
529 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
530 kcred, NULL, &acl_ids));
531 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
532 ASSERT3P(zp, ==, sharezp);
533 POINTER_INVALIDATE(&sharezp->z_zfsvfs);
534 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
535 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
536 zfsvfs->z_shares_dir = sharezp->z_id;
538 zfs_acl_ids_free(&acl_ids);
539 ZTOV(sharezp)->v_data = NULL;
540 ZTOV(sharezp)->v_count = 0;
541 ZTOV(sharezp)->v_holdcnt = 0;
543 sa_handle_destroy(sharezp->z_sa_hdl);
544 sharezp->z_vnode = NULL;
545 kmem_cache_free(znode_cache, sharezp);
551 * define a couple of values we need available
552 * for both 64 and 32 bit environments.
555 #define NBITSMINOR64 32
558 #define MAXMAJ64 0xffffffffUL
561 #define MAXMIN64 0xffffffffUL
565 * Create special expldev for ZFS private use.
566 * Can't use standard expldev since it doesn't do
567 * what we want. The standard expldev() takes a
568 * dev32_t in LP64 and expands it to a long dev_t.
569 * We need an interface that takes a dev32_t in ILP32
570 * and expands it to a long dev_t.
573 zfs_expldev(dev_t dev)
575 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev));
578 * Special cmpldev for ZFS private use.
579 * Can't use standard cmpldev since it takes
580 * a long dev_t and compresses it to dev32_t in
581 * LP64. We need to do a compaction of a long dev_t
582 * to a dev32_t in ILP32.
585 zfs_cmpldev(uint64_t dev)
587 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64)));
591 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
592 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
594 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
595 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
597 mutex_enter(&zp->z_lock);
599 ASSERT(zp->z_sa_hdl == NULL);
600 ASSERT(zp->z_acl_cached == NULL);
601 if (sa_hdl == NULL) {
602 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
603 SA_HDL_SHARED, &zp->z_sa_hdl));
605 zp->z_sa_hdl = sa_hdl;
606 sa_set_userp(sa_hdl, zp);
609 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
612 * Slap on VROOT if we are the root znode
614 if (zp->z_id == zfsvfs->z_root)
615 ZTOV(zp)->v_flag |= VROOT;
617 mutex_exit(&zp->z_lock);
622 zfs_znode_dmu_fini(znode_t *zp)
624 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
626 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
628 sa_handle_destroy(zp->z_sa_hdl);
633 zfs_vnode_forget(vnode_t *vp)
636 /* copied from insmntque_stddtr */
638 vp->v_op = &dead_vnodeops;
644 * Construct a new znode/vnode and intialize.
646 * This does not do a call to dmu_set_user() that is
647 * up to the caller to do, in case you don't want to
651 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
652 dmu_object_type_t obj_type, sa_handle_t *hdl)
658 sa_bulk_attr_t bulk[9];
661 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
662 zfs_znode_cache_constructor(zp, zfsvfs->z_parent->z_vfs, 0);
664 ASSERT(zp->z_dirlocks == NULL);
665 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
669 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
670 * the zfs_znode_move() callback.
674 zp->z_atime_dirty = 0;
676 zp->z_id = db->db_object;
678 zp->z_seq = 0x7A4653;
683 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
685 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
686 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
687 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
689 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
691 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
693 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
694 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
696 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
698 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
701 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) {
703 sa_handle_destroy(zp->z_sa_hdl);
704 zfs_vnode_forget(vp);
706 kmem_cache_free(znode_cache, zp);
712 vp->v_type = IFTOVT((mode_t)mode);
714 switch (vp->v_type) {
716 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
723 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
724 &rdev, sizeof (rdev)) == 0);
726 vp->v_rdev = zfs_cmpldev(rdev);
735 vp->v_op = &zfs_fifoops;
738 if (parent == zfsvfs->z_shares_dir) {
739 ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
740 vp->v_op = &zfs_shareops;
745 vn_setops(vp, zfs_symvnodeops);
748 vn_setops(vp, zfs_evnodeops);
752 if (vp->v_type != VFIFO)
755 mutex_enter(&zfsvfs->z_znodes_lock);
756 list_insert_tail(&zfsvfs->z_all_znodes, zp);
759 * Everything else must be valid before assigning z_zfsvfs makes the
760 * znode eligible for zfs_znode_move().
762 zp->z_zfsvfs = zfsvfs;
763 mutex_exit(&zfsvfs->z_znodes_lock);
765 VFS_HOLD(zfsvfs->z_vfs);
769 static uint64_t empty_xattr;
770 static uint64_t pad[4];
771 static zfs_acl_phys_t acl_phys;
773 * Create a new DMU object to hold a zfs znode.
775 * IN: dzp - parent directory for new znode
776 * vap - file attributes for new znode
777 * tx - dmu transaction id for zap operations
778 * cr - credentials of caller
780 * IS_ROOT_NODE - new object will be root
781 * IS_XATTR - new object is an attribute
782 * bonuslen - length of bonus buffer
783 * setaclp - File/Dir initial ACL
784 * fuidp - Tracks fuid allocation.
786 * OUT: zpp - allocated znode
790 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
791 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
793 uint64_t crtime[2], atime[2], mtime[2], ctime[2];
794 uint64_t mode, size, links, parent, pflags;
795 uint64_t dzp_pflags = 0;
797 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
804 dmu_object_type_t obj_type;
805 sa_bulk_attr_t sa_attrs[ZPL_END];
807 zfs_acl_locator_cb_t locate = { 0 };
809 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
811 if (zfsvfs->z_replay) {
812 obj = vap->va_nodeid;
813 now = vap->va_ctime; /* see zfs_replay_create() */
814 gen = vap->va_nblocks; /* ditto */
818 gen = dmu_tx_get_txg(tx);
821 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
822 bonuslen = (obj_type == DMU_OT_SA) ?
823 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
826 * Create a new DMU object.
829 * There's currently no mechanism for pre-reading the blocks that will
830 * be needed to allocate a new object, so we accept the small chance
831 * that there will be an i/o error and we will fail one of the
834 if (vap->va_type == VDIR) {
835 if (zfsvfs->z_replay) {
836 err = zap_create_claim_norm(zfsvfs->z_os, obj,
837 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
838 obj_type, bonuslen, tx);
841 obj = zap_create_norm(zfsvfs->z_os,
842 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
843 obj_type, bonuslen, tx);
846 if (zfsvfs->z_replay) {
847 err = dmu_object_claim(zfsvfs->z_os, obj,
848 DMU_OT_PLAIN_FILE_CONTENTS, 0,
849 obj_type, bonuslen, tx);
852 obj = dmu_object_alloc(zfsvfs->z_os,
853 DMU_OT_PLAIN_FILE_CONTENTS, 0,
854 obj_type, bonuslen, tx);
858 getnewvnode_reserve(1);
859 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
860 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
863 * If this is the root, fix up the half-initialized parent pointer
864 * to reference the just-allocated physical data area.
866 if (flag & IS_ROOT_NODE) {
869 dzp_pflags = dzp->z_pflags;
873 * If parent is an xattr, so am I.
875 if (dzp_pflags & ZFS_XATTR) {
879 if (zfsvfs->z_use_fuids)
880 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
884 if (vap->va_type == VDIR) {
885 size = 2; /* contents ("." and "..") */
886 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
891 if (vap->va_type == VBLK || vap->va_type == VCHR) {
892 rdev = zfs_expldev(vap->va_rdev);
896 mode = acl_ids->z_mode;
901 * No execs denied will be deterimed when zfs_mode_compute() is called.
903 pflags |= acl_ids->z_aclp->z_hints &
904 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
905 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
907 ZFS_TIME_ENCODE(&now, crtime);
908 ZFS_TIME_ENCODE(&now, ctime);
910 if (vap->va_mask & AT_ATIME) {
911 ZFS_TIME_ENCODE(&vap->va_atime, atime);
913 ZFS_TIME_ENCODE(&now, atime);
916 if (vap->va_mask & AT_MTIME) {
917 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
919 ZFS_TIME_ENCODE(&now, mtime);
922 /* Now add in all of the "SA" attributes */
923 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
927 * Setup the array of attributes to be replaced/set on the new file
929 * order for DMU_OT_ZNODE is critical since it needs to be constructed
930 * in the old znode_phys_t format. Don't change this ordering
933 if (obj_type == DMU_OT_ZNODE) {
934 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
936 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
938 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
940 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
942 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
944 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
946 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
948 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
951 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
953 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
955 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
957 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
958 &acl_ids->z_fuid, 8);
959 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
960 &acl_ids->z_fgid, 8);
961 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
963 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
965 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
967 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
969 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
971 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
975 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
977 if (obj_type == DMU_OT_ZNODE) {
978 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
981 if (obj_type == DMU_OT_ZNODE ||
982 (vap->va_type == VBLK || vap->va_type == VCHR)) {
983 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
987 if (obj_type == DMU_OT_ZNODE) {
988 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
990 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
991 &acl_ids->z_fuid, 8);
992 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
993 &acl_ids->z_fgid, 8);
994 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
995 sizeof (uint64_t) * 4);
996 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
997 &acl_phys, sizeof (zfs_acl_phys_t));
998 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
999 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
1000 &acl_ids->z_aclp->z_acl_count, 8);
1001 locate.cb_aclp = acl_ids->z_aclp;
1002 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
1003 zfs_acl_data_locator, &locate,
1004 acl_ids->z_aclp->z_acl_bytes);
1005 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
1006 acl_ids->z_fuid, acl_ids->z_fgid);
1009 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
1011 if (!(flag & IS_ROOT_NODE)) {
1012 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
1013 ASSERT(*zpp != NULL);
1016 * If we are creating the root node, the "parent" we
1017 * passed in is the znode for the root.
1021 (*zpp)->z_sa_hdl = sa_hdl;
1024 (*zpp)->z_pflags = pflags;
1025 (*zpp)->z_mode = mode;
1027 if (vap->va_mask & AT_XVATTR)
1028 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
1030 if (obj_type == DMU_OT_ZNODE ||
1031 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1032 err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx);
1035 if (!(flag & IS_ROOT_NODE)) {
1039 vp->v_vflag |= VV_FORCEINSMQ;
1040 err = insmntque(vp, zfsvfs->z_vfs);
1041 vp->v_vflag &= ~VV_FORCEINSMQ;
1042 KASSERT(err == 0, ("insmntque() failed: error %d", err));
1044 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1045 getnewvnode_drop_reserve();
1049 * zfs_xvattr_set only updates the in-core attributes
1050 * it is assumed the caller will be doing an sa_bulk_update
1051 * to push the changes out
1054 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1058 xoap = xva_getxoptattr(xvap);
1061 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1063 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1064 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1065 ×, sizeof (times), tx);
1066 XVA_SET_RTN(xvap, XAT_CREATETIME);
1068 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1069 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1071 XVA_SET_RTN(xvap, XAT_READONLY);
1073 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1074 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1076 XVA_SET_RTN(xvap, XAT_HIDDEN);
1078 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1079 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1081 XVA_SET_RTN(xvap, XAT_SYSTEM);
1083 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1084 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1086 XVA_SET_RTN(xvap, XAT_ARCHIVE);
1088 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1089 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1091 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1093 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1094 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1096 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1098 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1099 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1101 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1103 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1104 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1106 XVA_SET_RTN(xvap, XAT_NODUMP);
1108 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1109 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1111 XVA_SET_RTN(xvap, XAT_OPAQUE);
1113 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1114 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1115 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1116 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1118 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1119 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1121 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1123 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1124 zfs_sa_set_scanstamp(zp, xvap, tx);
1125 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1127 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1128 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1130 XVA_SET_RTN(xvap, XAT_REPARSE);
1132 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1133 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1135 XVA_SET_RTN(xvap, XAT_OFFLINE);
1137 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1138 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1140 XVA_SET_RTN(xvap, XAT_SPARSE);
1145 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1147 dmu_object_info_t doi;
1156 getnewvnode_reserve(1);
1158 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1160 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1162 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1163 getnewvnode_drop_reserve();
1167 dmu_object_info_from_db(db, &doi);
1168 if (doi.doi_bonus_type != DMU_OT_SA &&
1169 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1170 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1171 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1172 sa_buf_rele(db, NULL);
1173 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1174 getnewvnode_drop_reserve();
1178 hdl = dmu_buf_get_user(db);
1180 zp = sa_get_userdata(hdl);
1184 * Since "SA" does immediate eviction we
1185 * should never find a sa handle that doesn't
1186 * know about the znode.
1189 ASSERT3P(zp, !=, NULL);
1191 mutex_enter(&zp->z_lock);
1192 ASSERT3U(zp->z_id, ==, obj_num);
1193 if (zp->z_unlinked) {
1204 if ((vp->v_iflag & VI_DOOMED) != 0) {
1207 * Don't VN_RELE() vnode here, because
1208 * it can call vn_lock() which creates
1209 * LOR between vnode lock and znode
1210 * lock. We will VN_RELE() the vnode
1211 * after droping znode lock.
1217 ZFS_LOG(1, "dying znode detected (zp=%p)", zp);
1221 * znode is dying so we can't reuse it, we must
1222 * wait until destruction is completed.
1224 sa_buf_rele(db, NULL);
1225 mutex_exit(&zp->z_lock);
1226 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1229 tsleep(zp, 0, "zcollide", 1);
1235 sa_buf_rele(db, NULL);
1236 mutex_exit(&zp->z_lock);
1237 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1238 getnewvnode_drop_reserve();
1243 * Not found create new znode/vnode
1244 * but only if file exists.
1246 * There is a small window where zfs_vget() could
1247 * find this object while a file create is still in
1248 * progress. This is checked for in zfs_znode_alloc()
1250 * if zfs_znode_alloc() fails it will drop the hold on the
1253 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1254 doi.doi_bonus_type, NULL);
1261 vnode_t *vp = ZTOV(zp);
1263 err = insmntque(vp, zfsvfs->z_vfs);
1268 zfs_znode_dmu_fini(zp);
1273 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1274 getnewvnode_drop_reserve();
1279 zfs_rezget(znode_t *zp)
1281 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1282 dmu_object_info_t doi;
1285 uint64_t obj_num = zp->z_id;
1286 uint64_t mode, size;
1287 sa_bulk_attr_t bulk[8];
1292 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1294 mutex_enter(&zp->z_acl_lock);
1295 if (zp->z_acl_cached) {
1296 zfs_acl_free(zp->z_acl_cached);
1297 zp->z_acl_cached = NULL;
1300 mutex_exit(&zp->z_acl_lock);
1301 ASSERT(zp->z_sa_hdl == NULL);
1302 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1304 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1308 dmu_object_info_from_db(db, &doi);
1309 if (doi.doi_bonus_type != DMU_OT_SA &&
1310 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1311 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1312 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1313 sa_buf_rele(db, NULL);
1314 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1318 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1321 /* reload cached values */
1322 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1323 &gen, sizeof (gen));
1324 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1325 &zp->z_size, sizeof (zp->z_size));
1326 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1327 &zp->z_links, sizeof (zp->z_links));
1328 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1329 &zp->z_pflags, sizeof (zp->z_pflags));
1330 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1331 &zp->z_atime, sizeof (zp->z_atime));
1332 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1333 &zp->z_uid, sizeof (zp->z_uid));
1334 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1335 &zp->z_gid, sizeof (zp->z_gid));
1336 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1337 &mode, sizeof (mode));
1339 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1340 zfs_znode_dmu_fini(zp);
1341 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1347 if (gen != zp->z_gen) {
1348 zfs_znode_dmu_fini(zp);
1349 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1354 * XXXPJD: Not sure how is that possible, but under heavy
1355 * zfs recv -F load it happens that z_gen is the same, but
1356 * vnode type is different than znode type. This would mean
1357 * that for example regular file was replaced with directory
1358 * which has the same object number.
1362 vp->v_type != IFTOVT((mode_t)zp->z_mode)) {
1363 zfs_znode_dmu_fini(zp);
1364 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1368 zp->z_unlinked = (zp->z_links == 0);
1369 zp->z_blksz = doi.doi_data_block_size;
1371 vn_pages_remove(vp, 0, 0);
1372 if (zp->z_size != size)
1373 vnode_pager_setsize(vp, zp->z_size);
1376 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1382 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1384 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1385 objset_t *os = zfsvfs->z_os;
1386 uint64_t obj = zp->z_id;
1387 uint64_t acl_obj = zfs_external_acl(zp);
1389 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1391 VERIFY(!zp->z_is_sa);
1392 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1394 VERIFY(0 == dmu_object_free(os, obj, tx));
1395 zfs_znode_dmu_fini(zp);
1396 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1401 zfs_zinactive(znode_t *zp)
1403 vnode_t *vp = ZTOV(zp);
1404 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1405 uint64_t z_id = zp->z_id;
1408 ASSERT(zp->z_sa_hdl);
1411 * Don't allow a zfs_zget() while were trying to release this znode
1413 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1415 mutex_enter(&zp->z_lock);
1417 if (vp->v_count > 0) {
1419 * If the hold count is greater than zero, somebody has
1420 * obtained a new reference on this znode while we were
1421 * processing it here, so we are done.
1424 mutex_exit(&zp->z_lock);
1425 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1431 * If this was the last reference to a file with no links,
1432 * remove the file from the file system.
1434 if (zp->z_unlinked) {
1435 mutex_exit(&zp->z_lock);
1436 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1437 ASSERT(vp->v_count == 0);
1438 vrecycle(vp, curthread);
1439 vfslocked = VFS_LOCK_GIANT(zfsvfs->z_vfs);
1441 VFS_UNLOCK_GIANT(vfslocked);
1445 mutex_exit(&zp->z_lock);
1446 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1450 zfs_znode_free(znode_t *zp)
1452 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1454 ASSERT(ZTOV(zp) == NULL);
1455 ASSERT(zp->z_sa_hdl == NULL);
1456 mutex_enter(&zfsvfs->z_znodes_lock);
1457 POINTER_INVALIDATE(&zp->z_zfsvfs);
1458 list_remove(&zfsvfs->z_all_znodes, zp);
1459 mutex_exit(&zfsvfs->z_znodes_lock);
1461 if (zp->z_acl_cached) {
1462 zfs_acl_free(zp->z_acl_cached);
1463 zp->z_acl_cached = NULL;
1466 kmem_cache_free(znode_cache, zp);
1468 VFS_RELE(zfsvfs->z_vfs);
1472 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1473 uint64_t ctime[2], boolean_t have_tx)
1479 if (have_tx) { /* will sa_bulk_update happen really soon? */
1480 zp->z_atime_dirty = 0;
1483 zp->z_atime_dirty = 1;
1486 if (flag & AT_ATIME) {
1487 ZFS_TIME_ENCODE(&now, zp->z_atime);
1490 if (flag & AT_MTIME) {
1491 ZFS_TIME_ENCODE(&now, mtime);
1492 if (zp->z_zfsvfs->z_use_fuids) {
1493 zp->z_pflags |= (ZFS_ARCHIVE |
1498 if (flag & AT_CTIME) {
1499 ZFS_TIME_ENCODE(&now, ctime);
1500 if (zp->z_zfsvfs->z_use_fuids)
1501 zp->z_pflags |= ZFS_ARCHIVE;
1506 * Grow the block size for a file.
1508 * IN: zp - znode of file to free data in.
1509 * size - requested block size
1510 * tx - open transaction.
1512 * NOTE: this function assumes that the znode is write locked.
1515 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1520 if (size <= zp->z_blksz)
1523 * If the file size is already greater than the current blocksize,
1524 * we will not grow. If there is more than one block in a file,
1525 * the blocksize cannot change.
1527 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1530 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1533 if (error == ENOTSUP)
1537 /* What blocksize did we actually get? */
1538 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1543 * This is a dummy interface used when pvn_vplist_dirty() should *not*
1544 * be calling back into the fs for a putpage(). E.g.: when truncating
1545 * a file, the pages being "thrown away* don't need to be written out.
1549 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1550 int flags, cred_t *cr)
1558 * Increase the file length
1560 * IN: zp - znode of file to free data in.
1561 * end - new end-of-file
1563 * RETURN: 0 if success
1564 * error code if failure
1567 zfs_extend(znode_t *zp, uint64_t end)
1569 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1576 * We will change zp_size, lock the whole file.
1578 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1581 * Nothing to do if file already at desired length.
1583 if (end <= zp->z_size) {
1584 zfs_range_unlock(rl);
1588 tx = dmu_tx_create(zfsvfs->z_os);
1589 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1590 zfs_sa_upgrade_txholds(tx, zp);
1591 if (end > zp->z_blksz &&
1592 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1594 * We are growing the file past the current block size.
1596 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1597 ASSERT(!ISP2(zp->z_blksz));
1598 newblksz = MIN(end, SPA_MAXBLOCKSIZE);
1600 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1602 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1607 error = dmu_tx_assign(tx, TXG_NOWAIT);
1609 if (error == ERESTART) {
1615 zfs_range_unlock(rl);
1620 zfs_grow_blocksize(zp, newblksz, tx);
1624 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1625 &zp->z_size, sizeof (zp->z_size), tx));
1627 vnode_pager_setsize(ZTOV(zp), end);
1629 zfs_range_unlock(rl);
1637 * Free space in a file.
1639 * IN: zp - znode of file to free data in.
1640 * off - start of section to free.
1641 * len - length of section to free.
1643 * RETURN: 0 if success
1644 * error code if failure
1647 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1649 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1654 * Lock the range being freed.
1656 rl = zfs_range_lock(zp, off, len, RL_WRITER);
1659 * Nothing to do if file already at desired length.
1661 if (off >= zp->z_size) {
1662 zfs_range_unlock(rl);
1666 if (off + len > zp->z_size)
1667 len = zp->z_size - off;
1669 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1673 * In FreeBSD we cannot free block in the middle of a file,
1674 * but only at the end of a file, so this code path should
1677 vnode_pager_setsize(ZTOV(zp), off);
1680 zfs_range_unlock(rl);
1688 * IN: zp - znode of file to free data in.
1689 * end - new end-of-file.
1691 * RETURN: 0 if success
1692 * error code if failure
1695 zfs_trunc(znode_t *zp, uint64_t end)
1697 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1698 vnode_t *vp = ZTOV(zp);
1702 sa_bulk_attr_t bulk[2];
1706 * We will change zp_size, lock the whole file.
1708 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1711 * Nothing to do if file already at desired length.
1713 if (end >= zp->z_size) {
1714 zfs_range_unlock(rl);
1718 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1);
1720 zfs_range_unlock(rl);
1724 tx = dmu_tx_create(zfsvfs->z_os);
1725 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1726 zfs_sa_upgrade_txholds(tx, zp);
1727 error = dmu_tx_assign(tx, TXG_NOWAIT);
1729 if (error == ERESTART) {
1735 zfs_range_unlock(rl);
1740 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1741 NULL, &zp->z_size, sizeof (zp->z_size));
1744 zp->z_pflags &= ~ZFS_SPARSE;
1745 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1746 NULL, &zp->z_pflags, 8);
1748 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1753 * Clear any mapped pages in the truncated region. This has to
1754 * happen outside of the transaction to avoid the possibility of
1755 * a deadlock with someone trying to push a page that we are
1756 * about to invalidate.
1758 vnode_pager_setsize(vp, end);
1760 zfs_range_unlock(rl);
1766 * Free space in a file
1768 * IN: zp - znode of file to free data in.
1769 * off - start of range
1770 * len - end of range (0 => EOF)
1771 * flag - current file open mode flags.
1772 * log - TRUE if this action should be logged
1774 * RETURN: 0 if success
1775 * error code if failure
1778 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1780 vnode_t *vp = ZTOV(zp);
1782 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1783 zilog_t *zilog = zfsvfs->z_log;
1785 uint64_t mtime[2], ctime[2];
1786 sa_bulk_attr_t bulk[3];
1790 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1791 sizeof (mode))) != 0)
1794 if (off > zp->z_size) {
1795 error = zfs_extend(zp, off+len);
1796 if (error == 0 && log)
1803 * Check for any locks in the region to be freed.
1806 if (MANDLOCK(vp, (mode_t)mode)) {
1807 uint64_t length = (len ? len : zp->z_size - off);
1808 if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1813 error = zfs_trunc(zp, off);
1815 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1816 off + len > zp->z_size)
1817 error = zfs_extend(zp, off+len);
1822 tx = dmu_tx_create(zfsvfs->z_os);
1823 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1824 zfs_sa_upgrade_txholds(tx, zp);
1825 error = dmu_tx_assign(tx, TXG_NOWAIT);
1827 if (error == ERESTART) {
1836 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1837 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1838 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1839 NULL, &zp->z_pflags, 8);
1840 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1841 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1844 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1851 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1854 uint64_t moid, obj, sa_obj, version;
1855 uint64_t sense = ZFS_CASE_SENSITIVE;
1860 znode_t *rootzp = NULL;
1864 zfs_acl_ids_t acl_ids;
1867 * First attempt to create master node.
1870 * In an empty objset, there are no blocks to read and thus
1871 * there can be no i/o errors (which we assert below).
1873 moid = MASTER_NODE_OBJ;
1874 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1875 DMU_OT_NONE, 0, tx);
1879 * Set starting attributes.
1881 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1883 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1884 /* For the moment we expect all zpl props to be uint64_ts */
1888 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1889 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1890 name = nvpair_name(elem);
1891 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1895 error = zap_update(os, moid, name, 8, 1, &val, tx);
1898 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1900 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1903 ASSERT(version != 0);
1904 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1907 * Create zap object used for SA attribute registration
1910 if (version >= ZPL_VERSION_SA) {
1911 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1912 DMU_OT_NONE, 0, tx);
1913 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1919 * Create a delete queue.
1921 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1923 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1927 * Create root znode. Create minimal znode/vnode/zfsvfs
1928 * to allow zfs_mknode to work.
1931 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1932 vattr.va_type = VDIR;
1933 vattr.va_mode = S_IFDIR|0755;
1934 vattr.va_uid = crgetuid(cr);
1935 vattr.va_gid = crgetgid(cr);
1937 bzero(&zfsvfs, sizeof (zfsvfs_t));
1939 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1940 zfs_znode_cache_constructor(rootzp, NULL, 0);
1941 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1942 rootzp->z_moved = 0;
1943 rootzp->z_unlinked = 0;
1944 rootzp->z_atime_dirty = 0;
1945 rootzp->z_is_sa = USE_SA(version, os);
1947 vnode.v_type = VDIR;
1948 vnode.v_data = rootzp;
1949 rootzp->z_vnode = &vnode;
1952 zfsvfs.z_parent = &zfsvfs;
1953 zfsvfs.z_version = version;
1954 zfsvfs.z_use_fuids = USE_FUIDS(version, os);
1955 zfsvfs.z_use_sa = USE_SA(version, os);
1956 zfsvfs.z_norm = norm;
1958 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1959 &zfsvfs.z_attr_table);
1964 * Fold case on file systems that are always or sometimes case
1967 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1968 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER;
1970 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1971 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t),
1972 offsetof(znode_t, z_link_node));
1974 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1975 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1977 rootzp->z_zfsvfs = &zfsvfs;
1978 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1979 cr, NULL, &acl_ids));
1980 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1981 ASSERT3P(zp, ==, rootzp);
1982 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1984 zfs_acl_ids_free(&acl_ids);
1985 POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1987 sa_handle_destroy(rootzp->z_sa_hdl);
1988 rootzp->z_vnode = NULL;
1989 kmem_cache_free(znode_cache, rootzp);
1992 * Create shares directory
1995 error = zfs_create_share_dir(&zfsvfs, tx);
1999 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
2000 mutex_destroy(&zfsvfs.z_hold_mtx[i]);
2003 #endif /* _KERNEL */
2006 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
2008 uint64_t sa_obj = 0;
2011 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
2012 if (error != 0 && error != ENOENT)
2015 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
2020 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
2021 dmu_buf_t **db, void *tag)
2023 dmu_object_info_t doi;
2026 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
2029 dmu_object_info_from_db(*db, &doi);
2030 if ((doi.doi_bonus_type != DMU_OT_SA &&
2031 doi.doi_bonus_type != DMU_OT_ZNODE) ||
2032 doi.doi_bonus_type == DMU_OT_ZNODE &&
2033 doi.doi_bonus_size < sizeof (znode_phys_t)) {
2034 sa_buf_rele(*db, tag);
2038 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
2040 sa_buf_rele(*db, tag);
2048 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
2050 sa_handle_destroy(hdl);
2051 sa_buf_rele(db, tag);
2055 * Given an object number, return its parent object number and whether
2056 * or not the object is an extended attribute directory.
2059 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
2060 uint64_t *pobjp, int *is_xattrdir)
2065 uint64_t parent_mode;
2066 sa_bulk_attr_t bulk[3];
2067 sa_handle_t *sa_hdl;
2072 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
2073 &parent, sizeof (parent));
2074 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
2075 &pflags, sizeof (pflags));
2076 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2077 &mode, sizeof (mode));
2079 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
2083 * When a link is removed its parent pointer is not changed and will
2084 * be invalid. There are two cases where a link is removed but the
2085 * file stays around, when it goes to the delete queue and when there
2086 * are additional links.
2088 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
2092 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
2093 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2097 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
2100 * Extended attributes can be applied to files, directories, etc.
2101 * Otherwise the parent must be a directory.
2103 if (!*is_xattrdir && !S_ISDIR(parent_mode))
2112 * Given an object number, return some zpl level statistics
2115 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2118 sa_bulk_attr_t bulk[4];
2121 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2122 &sb->zs_mode, sizeof (sb->zs_mode));
2123 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2124 &sb->zs_gen, sizeof (sb->zs_gen));
2125 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2126 &sb->zs_links, sizeof (sb->zs_links));
2127 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2128 &sb->zs_ctime, sizeof (sb->zs_ctime));
2130 return (sa_bulk_lookup(hdl, bulk, count));
2134 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2135 sa_attr_type_t *sa_table, char *buf, int len)
2137 sa_handle_t *sa_hdl;
2138 sa_handle_t *prevhdl = NULL;
2139 dmu_buf_t *prevdb = NULL;
2140 dmu_buf_t *sa_db = NULL;
2141 char *path = buf + len - 1;
2149 char component[MAXNAMELEN + 2];
2154 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2156 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2157 &is_xattrdir)) != 0)
2168 (void) sprintf(component + 1, "<xattrdir>");
2170 error = zap_value_search(osp, pobj, obj,
2171 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2176 complen = strlen(component);
2178 ASSERT(path >= buf);
2179 bcopy(component, path, complen);
2182 if (sa_hdl != hdl) {
2186 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2194 if (sa_hdl != NULL && sa_hdl != hdl) {
2195 ASSERT(sa_db != NULL);
2196 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2200 (void) memmove(buf, path, buf + len - path);
2206 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2208 sa_attr_type_t *sa_table;
2213 error = zfs_sa_setup(osp, &sa_table);
2217 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2221 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2223 zfs_release_sa_handle(hdl, db, FTAG);
2228 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2231 char *path = buf + len - 1;
2232 sa_attr_type_t *sa_table;
2239 error = zfs_sa_setup(osp, &sa_table);
2243 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2247 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2249 zfs_release_sa_handle(hdl, db, FTAG);
2253 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2255 zfs_release_sa_handle(hdl, db, FTAG);