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, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Integros [integros.com]
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */
31 #include <sys/types.h>
32 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/sysmacros.h>
36 #include <sys/resource.h>
37 #include <sys/mntent.h>
38 #include <sys/u8_textprep.h>
39 #include <sys/dsl_dataset.h>
41 #include <sys/vnode.h>
44 #include <sys/errno.h>
45 #include <sys/unistd.h>
46 #include <sys/atomic.h>
47 #include <sys/zfs_dir.h>
48 #include <sys/zfs_acl.h>
49 #include <sys/zfs_ioctl.h>
50 #include <sys/zfs_rlock.h>
51 #include <sys/zfs_fuid.h>
52 #include <sys/dnode.h>
53 #include <sys/fs/zfs.h>
54 #include <sys/kidmap.h>
58 #include <sys/dmu_objset.h>
59 #include <sys/refcount.h>
62 #include <sys/zfs_znode.h>
64 #include <sys/zfs_sa.h>
65 #include <sys/zfs_stat.h>
66 #include <sys/refcount.h>
69 #include "zfs_comutil.h"
71 /* Used by fstat(1). */
72 SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD,
73 SYSCTL_NULL_INT_PTR, sizeof(znode_t), "sizeof(znode_t)");
76 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
77 * turned on when DEBUG is also defined.
84 #define ZNODE_STAT_ADD(stat) ((stat)++)
86 #define ZNODE_STAT_ADD(stat) /* nothing */
87 #endif /* ZNODE_STATS */
90 * Functions needed for userland (ie: libzpool) are not put under
91 * #ifdef_KERNEL; the rest of the functions have dependencies
92 * (such as VFS logic) that will not compile easily in userland.
96 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to
97 * be freed before it can be safely accessed.
99 krwlock_t zfsvfs_lock;
101 static kmem_cache_t *znode_cache = NULL;
105 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
108 * We should never drop all dbuf refs without first clearing
109 * the eviction callback.
111 panic("evicting znode %p\n", user_ptr);
114 extern struct vop_vector zfs_vnodeops;
115 extern struct vop_vector zfs_fifoops;
116 extern struct vop_vector zfs_shareops;
119 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
123 POINTER_INVALIDATE(&zp->z_zfsvfs);
125 list_link_init(&zp->z_link_node);
127 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
128 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
129 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
130 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
132 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
133 avl_create(&zp->z_range_avl, zfs_range_compare,
134 sizeof (rl_t), offsetof(rl_t, r_node));
136 zp->z_dirlocks = NULL;
137 zp->z_acl_cached = NULL;
145 zfs_znode_cache_destructor(void *buf, void *arg)
149 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
150 ASSERT(ZTOV(zp) == NULL);
152 ASSERT(!list_link_active(&zp->z_link_node));
153 mutex_destroy(&zp->z_lock);
154 rw_destroy(&zp->z_parent_lock);
155 rw_destroy(&zp->z_name_lock);
156 mutex_destroy(&zp->z_acl_lock);
157 avl_destroy(&zp->z_range_avl);
158 mutex_destroy(&zp->z_range_lock);
160 ASSERT(zp->z_dirlocks == NULL);
161 ASSERT(zp->z_acl_cached == NULL);
166 uint64_t zms_zfsvfs_invalid;
167 uint64_t zms_zfsvfs_recheck1;
168 uint64_t zms_zfsvfs_unmounted;
169 uint64_t zms_zfsvfs_recheck2;
170 uint64_t zms_obj_held;
171 uint64_t zms_vnode_locked;
172 uint64_t zms_not_only_dnlc;
174 #endif /* ZNODE_STATS */
178 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
183 nzp->z_zfsvfs = ozp->z_zfsvfs;
187 nzp->z_vnode = ozp->z_vnode;
188 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
189 ZTOV(ozp)->v_data = ozp;
190 ZTOV(nzp)->v_data = nzp;
192 nzp->z_id = ozp->z_id;
193 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
194 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0);
195 nzp->z_unlinked = ozp->z_unlinked;
196 nzp->z_atime_dirty = ozp->z_atime_dirty;
197 nzp->z_zn_prefetch = ozp->z_zn_prefetch;
198 nzp->z_blksz = ozp->z_blksz;
199 nzp->z_seq = ozp->z_seq;
200 nzp->z_mapcnt = ozp->z_mapcnt;
201 nzp->z_gen = ozp->z_gen;
202 nzp->z_sync_cnt = ozp->z_sync_cnt;
203 nzp->z_is_sa = ozp->z_is_sa;
204 nzp->z_sa_hdl = ozp->z_sa_hdl;
205 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
206 nzp->z_links = ozp->z_links;
207 nzp->z_size = ozp->z_size;
208 nzp->z_pflags = ozp->z_pflags;
209 nzp->z_uid = ozp->z_uid;
210 nzp->z_gid = ozp->z_gid;
211 nzp->z_mode = ozp->z_mode;
214 * Since this is just an idle znode and kmem is already dealing with
215 * memory pressure, release any cached ACL.
217 if (ozp->z_acl_cached) {
218 zfs_acl_free(ozp->z_acl_cached);
219 ozp->z_acl_cached = NULL;
222 sa_set_userp(nzp->z_sa_hdl, nzp);
225 * Invalidate the original znode by clearing fields that provide a
226 * pointer back to the znode. Set the low bit of the vfs pointer to
227 * ensure that zfs_znode_move() recognizes the znode as invalid in any
228 * subsequent callback.
230 ozp->z_sa_hdl = NULL;
231 POINTER_INVALIDATE(&ozp->z_zfsvfs);
237 ozp->z_moved = (uint8_t)-1;
242 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
244 znode_t *ozp = buf, *nzp = newbuf;
249 * The znode is on the file system's list of known znodes if the vfs
250 * pointer is valid. We set the low bit of the vfs pointer when freeing
251 * the znode to invalidate it, and the memory patterns written by kmem
252 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
253 * created znode sets the vfs pointer last of all to indicate that the
254 * znode is known and in a valid state to be moved by this function.
256 zfsvfs = ozp->z_zfsvfs;
257 if (!POINTER_IS_VALID(zfsvfs)) {
258 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
259 return (KMEM_CBRC_DONT_KNOW);
263 * Close a small window in which it's possible that the filesystem could
264 * be unmounted and freed, and zfsvfs, though valid in the previous
265 * statement, could point to unrelated memory by the time we try to
266 * prevent the filesystem from being unmounted.
268 rw_enter(&zfsvfs_lock, RW_WRITER);
269 if (zfsvfs != ozp->z_zfsvfs) {
270 rw_exit(&zfsvfs_lock);
271 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
272 return (KMEM_CBRC_DONT_KNOW);
276 * If the znode is still valid, then so is the file system. We know that
277 * no valid file system can be freed while we hold zfsvfs_lock, so we
278 * can safely ensure that the filesystem is not and will not be
279 * unmounted. The next statement is equivalent to ZFS_ENTER().
281 rrm_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
282 if (zfsvfs->z_unmounted) {
284 rw_exit(&zfsvfs_lock);
285 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
286 return (KMEM_CBRC_DONT_KNOW);
288 rw_exit(&zfsvfs_lock);
290 mutex_enter(&zfsvfs->z_znodes_lock);
292 * Recheck the vfs pointer in case the znode was removed just before
293 * acquiring the lock.
295 if (zfsvfs != ozp->z_zfsvfs) {
296 mutex_exit(&zfsvfs->z_znodes_lock);
298 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
299 return (KMEM_CBRC_DONT_KNOW);
303 * At this point we know that as long as we hold z_znodes_lock, the
304 * znode cannot be freed and fields within the znode can be safely
305 * accessed. Now, prevent a race with zfs_zget().
307 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
308 mutex_exit(&zfsvfs->z_znodes_lock);
310 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
311 return (KMEM_CBRC_LATER);
315 if (mutex_tryenter(&vp->v_lock) == 0) {
316 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
317 mutex_exit(&zfsvfs->z_znodes_lock);
319 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
320 return (KMEM_CBRC_LATER);
323 /* Only move znodes that are referenced _only_ by the DNLC. */
324 if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
325 mutex_exit(&vp->v_lock);
326 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
327 mutex_exit(&zfsvfs->z_znodes_lock);
329 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
330 return (KMEM_CBRC_LATER);
334 * The znode is known and in a valid state to move. We're holding the
335 * locks needed to execute the critical section.
337 zfs_znode_move_impl(ozp, nzp);
338 mutex_exit(&vp->v_lock);
339 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
341 list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
342 mutex_exit(&zfsvfs->z_znodes_lock);
345 return (KMEM_CBRC_YES);
355 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
356 ASSERT(znode_cache == NULL);
357 znode_cache = kmem_cache_create("zfs_znode_cache",
358 sizeof (znode_t), 0, zfs_znode_cache_constructor,
359 zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
360 kmem_cache_set_move(znode_cache, zfs_znode_move);
368 * Cleanup vfs & vnode ops
370 zfs_remove_op_tables();
377 kmem_cache_destroy(znode_cache);
379 rw_destroy(&zfsvfs_lock);
383 struct vnodeops *zfs_dvnodeops;
384 struct vnodeops *zfs_fvnodeops;
385 struct vnodeops *zfs_symvnodeops;
386 struct vnodeops *zfs_xdvnodeops;
387 struct vnodeops *zfs_evnodeops;
388 struct vnodeops *zfs_sharevnodeops;
391 zfs_remove_op_tables()
397 (void) vfs_freevfsops_by_type(zfsfstype);
404 vn_freevnodeops(zfs_dvnodeops);
406 vn_freevnodeops(zfs_fvnodeops);
408 vn_freevnodeops(zfs_symvnodeops);
410 vn_freevnodeops(zfs_xdvnodeops);
412 vn_freevnodeops(zfs_evnodeops);
413 if (zfs_sharevnodeops)
414 vn_freevnodeops(zfs_sharevnodeops);
416 zfs_dvnodeops = NULL;
417 zfs_fvnodeops = NULL;
418 zfs_symvnodeops = NULL;
419 zfs_xdvnodeops = NULL;
420 zfs_evnodeops = NULL;
421 zfs_sharevnodeops = NULL;
424 extern const fs_operation_def_t zfs_dvnodeops_template[];
425 extern const fs_operation_def_t zfs_fvnodeops_template[];
426 extern const fs_operation_def_t zfs_xdvnodeops_template[];
427 extern const fs_operation_def_t zfs_symvnodeops_template[];
428 extern const fs_operation_def_t zfs_evnodeops_template[];
429 extern const fs_operation_def_t zfs_sharevnodeops_template[];
432 zfs_create_op_tables()
437 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
438 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
439 * In this case we just return as the ops vectors are already set up.
444 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
449 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
454 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
459 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
464 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
469 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
477 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
479 zfs_acl_ids_t acl_ids;
485 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
486 vattr.va_type = VDIR;
487 vattr.va_mode = S_IFDIR|0555;
488 vattr.va_uid = crgetuid(kcred);
489 vattr.va_gid = crgetgid(kcred);
491 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
492 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
493 sharezp->z_moved = 0;
494 sharezp->z_unlinked = 0;
495 sharezp->z_atime_dirty = 0;
496 sharezp->z_zfsvfs = zfsvfs;
497 sharezp->z_is_sa = zfsvfs->z_use_sa;
499 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
500 kcred, NULL, &acl_ids));
501 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
502 ASSERT3P(zp, ==, sharezp);
503 POINTER_INVALIDATE(&sharezp->z_zfsvfs);
504 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
505 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
506 zfsvfs->z_shares_dir = sharezp->z_id;
508 zfs_acl_ids_free(&acl_ids);
509 sa_handle_destroy(sharezp->z_sa_hdl);
510 kmem_cache_free(znode_cache, sharezp);
516 * define a couple of values we need available
517 * for both 64 and 32 bit environments.
520 #define NBITSMINOR64 32
523 #define MAXMAJ64 0xffffffffUL
526 #define MAXMIN64 0xffffffffUL
530 * Create special expldev for ZFS private use.
531 * Can't use standard expldev since it doesn't do
532 * what we want. The standard expldev() takes a
533 * dev32_t in LP64 and expands it to a long dev_t.
534 * We need an interface that takes a dev32_t in ILP32
535 * and expands it to a long dev_t.
538 zfs_expldev(dev_t dev)
540 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev));
543 * Special cmpldev for ZFS private use.
544 * Can't use standard cmpldev since it takes
545 * a long dev_t and compresses it to dev32_t in
546 * LP64. We need to do a compaction of a long dev_t
547 * to a dev32_t in ILP32.
550 zfs_cmpldev(uint64_t dev)
552 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64)));
556 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
557 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
559 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
560 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
562 mutex_enter(&zp->z_lock);
564 ASSERT(zp->z_sa_hdl == NULL);
565 ASSERT(zp->z_acl_cached == NULL);
566 if (sa_hdl == NULL) {
567 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
568 SA_HDL_SHARED, &zp->z_sa_hdl));
570 zp->z_sa_hdl = sa_hdl;
571 sa_set_userp(sa_hdl, zp);
574 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
577 * Slap on VROOT if we are the root znode
579 if (zp->z_id == zfsvfs->z_root)
580 ZTOV(zp)->v_flag |= VROOT;
582 mutex_exit(&zp->z_lock);
587 zfs_znode_dmu_fini(znode_t *zp)
589 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
591 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
593 sa_handle_destroy(zp->z_sa_hdl);
598 zfs_vnode_forget(vnode_t *vp)
601 /* copied from insmntque_stddtr */
603 vp->v_op = &dead_vnodeops;
609 * Construct a new znode/vnode and intialize.
611 * This does not do a call to dmu_set_user() that is
612 * up to the caller to do, in case you don't want to
616 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
617 dmu_object_type_t obj_type, sa_handle_t *hdl)
623 sa_bulk_attr_t bulk[9];
627 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
629 KASSERT(curthread->td_vp_reserv > 0,
630 ("zfs_znode_alloc: getnewvnode without any vnodes reserved"));
631 error = getnewvnode("zfs", zfsvfs->z_parent->z_vfs, &zfs_vnodeops, &vp);
633 kmem_cache_free(znode_cache, zp);
639 ASSERT(zp->z_dirlocks == NULL);
640 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
644 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
645 * the zfs_znode_move() callback.
649 zp->z_atime_dirty = 0;
651 zp->z_id = db->db_object;
653 zp->z_seq = 0x7A4653;
658 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
660 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
661 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
662 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
664 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
666 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
668 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
669 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
671 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
673 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
676 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) {
678 sa_handle_destroy(zp->z_sa_hdl);
679 zfs_vnode_forget(vp);
681 kmem_cache_free(znode_cache, zp);
687 vp->v_type = IFTOVT((mode_t)mode);
689 switch (vp->v_type) {
691 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
698 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
699 &rdev, sizeof (rdev)) == 0);
701 vp->v_rdev = zfs_cmpldev(rdev);
710 vp->v_op = &zfs_fifoops;
713 if (parent == zfsvfs->z_shares_dir) {
714 ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
715 vp->v_op = &zfs_shareops;
720 vn_setops(vp, zfs_symvnodeops);
723 vn_setops(vp, zfs_evnodeops);
728 mutex_enter(&zfsvfs->z_znodes_lock);
729 list_insert_tail(&zfsvfs->z_all_znodes, zp);
732 * Everything else must be valid before assigning z_zfsvfs makes the
733 * znode eligible for zfs_znode_move().
735 zp->z_zfsvfs = zfsvfs;
736 mutex_exit(&zfsvfs->z_znodes_lock);
739 * Acquire vnode lock before making it available to the world.
741 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
743 if (vp->v_type != VFIFO)
746 VFS_HOLD(zfsvfs->z_vfs);
750 static uint64_t empty_xattr;
751 static uint64_t pad[4];
752 static zfs_acl_phys_t acl_phys;
754 * Create a new DMU object to hold a zfs znode.
756 * IN: dzp - parent directory for new znode
757 * vap - file attributes for new znode
758 * tx - dmu transaction id for zap operations
759 * cr - credentials of caller
761 * IS_ROOT_NODE - new object will be root
762 * IS_XATTR - new object is an attribute
763 * bonuslen - length of bonus buffer
764 * setaclp - File/Dir initial ACL
765 * fuidp - Tracks fuid allocation.
767 * OUT: zpp - allocated znode
771 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
772 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
774 uint64_t crtime[2], atime[2], mtime[2], ctime[2];
775 uint64_t mode, size, links, parent, pflags;
776 uint64_t dzp_pflags = 0;
778 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
785 dmu_object_type_t obj_type;
786 sa_bulk_attr_t sa_attrs[ZPL_END];
788 zfs_acl_locator_cb_t locate = { 0 };
790 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
792 if (zfsvfs->z_replay) {
793 obj = vap->va_nodeid;
794 now = vap->va_ctime; /* see zfs_replay_create() */
795 gen = vap->va_nblocks; /* ditto */
799 gen = dmu_tx_get_txg(tx);
802 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
803 bonuslen = (obj_type == DMU_OT_SA) ?
804 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
807 * Create a new DMU object.
810 * There's currently no mechanism for pre-reading the blocks that will
811 * be needed to allocate a new object, so we accept the small chance
812 * that there will be an i/o error and we will fail one of the
815 if (vap->va_type == VDIR) {
816 if (zfsvfs->z_replay) {
817 VERIFY0(zap_create_claim_norm(zfsvfs->z_os, obj,
818 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
819 obj_type, bonuslen, tx));
821 obj = zap_create_norm(zfsvfs->z_os,
822 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
823 obj_type, bonuslen, tx);
826 if (zfsvfs->z_replay) {
827 VERIFY0(dmu_object_claim(zfsvfs->z_os, obj,
828 DMU_OT_PLAIN_FILE_CONTENTS, 0,
829 obj_type, bonuslen, tx));
831 obj = dmu_object_alloc(zfsvfs->z_os,
832 DMU_OT_PLAIN_FILE_CONTENTS, 0,
833 obj_type, bonuslen, tx);
837 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
838 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
841 * If this is the root, fix up the half-initialized parent pointer
842 * to reference the just-allocated physical data area.
844 if (flag & IS_ROOT_NODE) {
847 dzp_pflags = dzp->z_pflags;
851 * If parent is an xattr, so am I.
853 if (dzp_pflags & ZFS_XATTR) {
857 if (zfsvfs->z_use_fuids)
858 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
862 if (vap->va_type == VDIR) {
863 size = 2; /* contents ("." and "..") */
864 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
869 if (vap->va_type == VBLK || vap->va_type == VCHR) {
870 rdev = zfs_expldev(vap->va_rdev);
874 mode = acl_ids->z_mode;
879 * No execs denied will be deterimed when zfs_mode_compute() is called.
881 pflags |= acl_ids->z_aclp->z_hints &
882 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
883 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
885 ZFS_TIME_ENCODE(&now, crtime);
886 ZFS_TIME_ENCODE(&now, ctime);
888 if (vap->va_mask & AT_ATIME) {
889 ZFS_TIME_ENCODE(&vap->va_atime, atime);
891 ZFS_TIME_ENCODE(&now, atime);
894 if (vap->va_mask & AT_MTIME) {
895 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
897 ZFS_TIME_ENCODE(&now, mtime);
900 /* Now add in all of the "SA" attributes */
901 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
905 * Setup the array of attributes to be replaced/set on the new file
907 * order for DMU_OT_ZNODE is critical since it needs to be constructed
908 * in the old znode_phys_t format. Don't change this ordering
911 if (obj_type == DMU_OT_ZNODE) {
912 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
914 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
916 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
918 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
920 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
922 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
924 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
926 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
929 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
931 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
933 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
935 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
936 &acl_ids->z_fuid, 8);
937 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
938 &acl_ids->z_fgid, 8);
939 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
941 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
943 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
945 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
947 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
949 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
953 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
955 if (obj_type == DMU_OT_ZNODE) {
956 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
959 if (obj_type == DMU_OT_ZNODE ||
960 (vap->va_type == VBLK || vap->va_type == VCHR)) {
961 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
965 if (obj_type == DMU_OT_ZNODE) {
966 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
968 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
969 &acl_ids->z_fuid, 8);
970 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
971 &acl_ids->z_fgid, 8);
972 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
973 sizeof (uint64_t) * 4);
974 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
975 &acl_phys, sizeof (zfs_acl_phys_t));
976 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
977 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
978 &acl_ids->z_aclp->z_acl_count, 8);
979 locate.cb_aclp = acl_ids->z_aclp;
980 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
981 zfs_acl_data_locator, &locate,
982 acl_ids->z_aclp->z_acl_bytes);
983 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
984 acl_ids->z_fuid, acl_ids->z_fgid);
987 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
989 if (!(flag & IS_ROOT_NODE)) {
990 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
991 ASSERT(*zpp != NULL);
994 * If we are creating the root node, the "parent" we
995 * passed in is the znode for the root.
999 (*zpp)->z_sa_hdl = sa_hdl;
1002 (*zpp)->z_pflags = pflags;
1003 (*zpp)->z_mode = mode;
1005 if (vap->va_mask & AT_XVATTR)
1006 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
1008 if (obj_type == DMU_OT_ZNODE ||
1009 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1010 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
1012 if (!(flag & IS_ROOT_NODE)) {
1016 vp->v_vflag |= VV_FORCEINSMQ;
1017 err = insmntque(vp, zfsvfs->z_vfs);
1018 vp->v_vflag &= ~VV_FORCEINSMQ;
1019 KASSERT(err == 0, ("insmntque() failed: error %d", err));
1021 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1025 * Update in-core attributes. It is assumed the caller will be doing an
1026 * sa_bulk_update to push the changes out.
1029 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1033 xoap = xva_getxoptattr(xvap);
1036 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1038 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1039 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1040 ×, sizeof (times), tx);
1041 XVA_SET_RTN(xvap, XAT_CREATETIME);
1043 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1044 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1046 XVA_SET_RTN(xvap, XAT_READONLY);
1048 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1049 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1051 XVA_SET_RTN(xvap, XAT_HIDDEN);
1053 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1054 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1056 XVA_SET_RTN(xvap, XAT_SYSTEM);
1058 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1059 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1061 XVA_SET_RTN(xvap, XAT_ARCHIVE);
1063 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1064 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1066 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1068 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1069 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1071 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1073 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1074 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1076 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1078 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1079 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1081 XVA_SET_RTN(xvap, XAT_NODUMP);
1083 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1084 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1086 XVA_SET_RTN(xvap, XAT_OPAQUE);
1088 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1089 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1090 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1091 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1093 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1094 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1096 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1098 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1099 zfs_sa_set_scanstamp(zp, xvap, tx);
1100 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1102 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1103 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1105 XVA_SET_RTN(xvap, XAT_REPARSE);
1107 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1108 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1110 XVA_SET_RTN(xvap, XAT_OFFLINE);
1112 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1113 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1115 XVA_SET_RTN(xvap, XAT_SPARSE);
1120 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1122 dmu_object_info_t doi;
1132 getnewvnode_reserve(1);
1135 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1137 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1139 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1140 getnewvnode_drop_reserve();
1144 dmu_object_info_from_db(db, &doi);
1145 if (doi.doi_bonus_type != DMU_OT_SA &&
1146 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1147 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1148 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1149 sa_buf_rele(db, NULL);
1150 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1152 getnewvnode_drop_reserve();
1154 return (SET_ERROR(EINVAL));
1157 hdl = dmu_buf_get_user(db);
1159 zp = sa_get_userdata(hdl);
1163 * Since "SA" does immediate eviction we
1164 * should never find a sa handle that doesn't
1165 * know about the znode.
1168 ASSERT3P(zp, !=, NULL);
1170 mutex_enter(&zp->z_lock);
1171 ASSERT3U(zp->z_id, ==, obj_num);
1172 if (zp->z_unlinked) {
1173 err = SET_ERROR(ENOENT);
1180 /* Don't let the vnode disappear after ZFS_OBJ_HOLD_EXIT. */
1184 mutex_exit(&zp->z_lock);
1185 sa_buf_rele(db, NULL);
1186 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1189 locked = VOP_ISLOCKED(vp);
1191 if ((vp->v_iflag & VI_DOOMED) != 0 &&
1192 locked != LK_EXCLUSIVE) {
1194 * The vnode is doomed and this thread doesn't
1195 * hold the exclusive lock on it, so the vnode
1196 * must be being reclaimed by another thread.
1197 * Otherwise the doomed vnode is being reclaimed
1198 * by this thread and zfs_zget is called from
1207 getnewvnode_drop_reserve();
1212 * Not found create new znode/vnode
1213 * but only if file exists.
1215 * There is a small window where zfs_vget() could
1216 * find this object while a file create is still in
1217 * progress. This is checked for in zfs_znode_alloc()
1219 * if zfs_znode_alloc() fails it will drop the hold on the
1222 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1223 doi.doi_bonus_type, NULL);
1225 err = SET_ERROR(ENOENT);
1230 vnode_t *vp = ZTOV(zp);
1232 err = insmntque(vp, zfsvfs->z_vfs);
1234 vp->v_hash = obj_num;
1238 zfs_znode_dmu_fini(zp);
1243 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1244 getnewvnode_drop_reserve();
1249 zfs_rezget(znode_t *zp)
1251 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1252 dmu_object_info_t doi;
1255 uint64_t obj_num = zp->z_id;
1256 uint64_t mode, size;
1257 sa_bulk_attr_t bulk[8];
1262 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1264 mutex_enter(&zp->z_acl_lock);
1265 if (zp->z_acl_cached) {
1266 zfs_acl_free(zp->z_acl_cached);
1267 zp->z_acl_cached = NULL;
1270 mutex_exit(&zp->z_acl_lock);
1271 ASSERT(zp->z_sa_hdl == NULL);
1272 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1274 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1278 dmu_object_info_from_db(db, &doi);
1279 if (doi.doi_bonus_type != DMU_OT_SA &&
1280 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1281 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1282 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1283 sa_buf_rele(db, NULL);
1284 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1285 return (SET_ERROR(EINVAL));
1288 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1291 /* reload cached values */
1292 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1293 &gen, sizeof (gen));
1294 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1295 &zp->z_size, sizeof (zp->z_size));
1296 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1297 &zp->z_links, sizeof (zp->z_links));
1298 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1299 &zp->z_pflags, sizeof (zp->z_pflags));
1300 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1301 &zp->z_atime, sizeof (zp->z_atime));
1302 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1303 &zp->z_uid, sizeof (zp->z_uid));
1304 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1305 &zp->z_gid, sizeof (zp->z_gid));
1306 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1307 &mode, sizeof (mode));
1309 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1310 zfs_znode_dmu_fini(zp);
1311 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1312 return (SET_ERROR(EIO));
1317 if (gen != zp->z_gen) {
1318 zfs_znode_dmu_fini(zp);
1319 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1320 return (SET_ERROR(EIO));
1324 * It is highly improbable but still quite possible that two
1325 * objects in different datasets are created with the same
1326 * object numbers and in transaction groups with the same
1327 * numbers. znodes corresponding to those objects would
1328 * have the same z_id and z_gen, but their other attributes
1330 * zfs recv -F may replace one of such objects with the other.
1331 * As a result file properties recorded in the replaced
1332 * object's vnode may no longer match the received object's
1333 * properties. At present the only cached property is the
1334 * files type recorded in v_type.
1335 * So, handle this case by leaving the old vnode and znode
1336 * disassociated from the actual object. A new vnode and a
1337 * znode will be created if the object is accessed
1338 * (e.g. via a look-up). The old vnode and znode will be
1339 * recycled when the last vnode reference is dropped.
1342 if (vp != NULL && vp->v_type != IFTOVT((mode_t)zp->z_mode)) {
1343 zfs_znode_dmu_fini(zp);
1344 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1348 zp->z_unlinked = (zp->z_links == 0);
1349 zp->z_blksz = doi.doi_data_block_size;
1351 vn_pages_remove(vp, 0, 0);
1352 if (zp->z_size != size)
1353 vnode_pager_setsize(vp, zp->z_size);
1356 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1362 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1364 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1365 objset_t *os = zfsvfs->z_os;
1366 uint64_t obj = zp->z_id;
1367 uint64_t acl_obj = zfs_external_acl(zp);
1369 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1371 VERIFY(!zp->z_is_sa);
1372 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1374 VERIFY(0 == dmu_object_free(os, obj, tx));
1375 zfs_znode_dmu_fini(zp);
1376 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1381 zfs_zinactive(znode_t *zp)
1383 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1384 uint64_t z_id = zp->z_id;
1386 ASSERT(zp->z_sa_hdl);
1389 * Don't allow a zfs_zget() while were trying to release this znode
1391 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1393 mutex_enter(&zp->z_lock);
1396 * If this was the last reference to a file with no links,
1397 * remove the file from the file system.
1399 if (zp->z_unlinked) {
1400 mutex_exit(&zp->z_lock);
1401 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1406 mutex_exit(&zp->z_lock);
1407 zfs_znode_dmu_fini(zp);
1408 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1413 zfs_znode_free(znode_t *zp)
1415 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1417 ASSERT(zp->z_sa_hdl == NULL);
1419 mutex_enter(&zfsvfs->z_znodes_lock);
1420 POINTER_INVALIDATE(&zp->z_zfsvfs);
1421 list_remove(&zfsvfs->z_all_znodes, zp);
1422 mutex_exit(&zfsvfs->z_znodes_lock);
1424 if (zp->z_acl_cached) {
1425 zfs_acl_free(zp->z_acl_cached);
1426 zp->z_acl_cached = NULL;
1429 kmem_cache_free(znode_cache, zp);
1431 VFS_RELE(zfsvfs->z_vfs);
1435 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1436 uint64_t ctime[2], boolean_t have_tx)
1442 if (have_tx) { /* will sa_bulk_update happen really soon? */
1443 zp->z_atime_dirty = 0;
1446 zp->z_atime_dirty = 1;
1449 if (flag & AT_ATIME) {
1450 ZFS_TIME_ENCODE(&now, zp->z_atime);
1453 if (flag & AT_MTIME) {
1454 ZFS_TIME_ENCODE(&now, mtime);
1455 if (zp->z_zfsvfs->z_use_fuids) {
1456 zp->z_pflags |= (ZFS_ARCHIVE |
1461 if (flag & AT_CTIME) {
1462 ZFS_TIME_ENCODE(&now, ctime);
1463 if (zp->z_zfsvfs->z_use_fuids)
1464 zp->z_pflags |= ZFS_ARCHIVE;
1469 * Grow the block size for a file.
1471 * IN: zp - znode of file to free data in.
1472 * size - requested block size
1473 * tx - open transaction.
1475 * NOTE: this function assumes that the znode is write locked.
1478 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1483 if (size <= zp->z_blksz)
1486 * If the file size is already greater than the current blocksize,
1487 * we will not grow. If there is more than one block in a file,
1488 * the blocksize cannot change.
1490 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1493 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1496 if (error == ENOTSUP)
1500 /* What blocksize did we actually get? */
1501 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1506 * This is a dummy interface used when pvn_vplist_dirty() should *not*
1507 * be calling back into the fs for a putpage(). E.g.: when truncating
1508 * a file, the pages being "thrown away* don't need to be written out.
1512 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1513 int flags, cred_t *cr)
1521 * Increase the file length
1523 * IN: zp - znode of file to free data in.
1524 * end - new end-of-file
1526 * RETURN: 0 on success, error code on failure
1529 zfs_extend(znode_t *zp, uint64_t end)
1531 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1538 * We will change zp_size, lock the whole file.
1540 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1543 * Nothing to do if file already at desired length.
1545 if (end <= zp->z_size) {
1546 zfs_range_unlock(rl);
1549 tx = dmu_tx_create(zfsvfs->z_os);
1550 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1551 zfs_sa_upgrade_txholds(tx, zp);
1552 if (end > zp->z_blksz &&
1553 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1555 * We are growing the file past the current block size.
1557 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1559 * File's blocksize is already larger than the
1560 * "recordsize" property. Only let it grow to
1561 * the next power of 2.
1563 ASSERT(!ISP2(zp->z_blksz));
1564 newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1566 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1568 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1573 error = dmu_tx_assign(tx, TXG_WAIT);
1576 zfs_range_unlock(rl);
1581 zfs_grow_blocksize(zp, newblksz, tx);
1585 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1586 &zp->z_size, sizeof (zp->z_size), tx));
1588 vnode_pager_setsize(ZTOV(zp), end);
1590 zfs_range_unlock(rl);
1598 * Free space in a file.
1600 * IN: zp - znode of file to free data in.
1601 * off - start of section to free.
1602 * len - length of section to free.
1604 * RETURN: 0 on success, error code on failure
1607 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1609 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1614 * Lock the range being freed.
1616 rl = zfs_range_lock(zp, off, len, RL_WRITER);
1619 * Nothing to do if file already at desired length.
1621 if (off >= zp->z_size) {
1622 zfs_range_unlock(rl);
1626 if (off + len > zp->z_size)
1627 len = zp->z_size - off;
1629 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1633 * In FreeBSD we cannot free block in the middle of a file,
1634 * but only at the end of a file, so this code path should
1637 vnode_pager_setsize(ZTOV(zp), off);
1640 zfs_range_unlock(rl);
1648 * IN: zp - znode of file to free data in.
1649 * end - new end-of-file.
1651 * RETURN: 0 on success, error code on failure
1654 zfs_trunc(znode_t *zp, uint64_t end)
1656 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1657 vnode_t *vp = ZTOV(zp);
1661 sa_bulk_attr_t bulk[2];
1665 * We will change zp_size, lock the whole file.
1667 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1670 * Nothing to do if file already at desired length.
1672 if (end >= zp->z_size) {
1673 zfs_range_unlock(rl);
1677 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1);
1679 zfs_range_unlock(rl);
1682 tx = dmu_tx_create(zfsvfs->z_os);
1683 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1684 zfs_sa_upgrade_txholds(tx, zp);
1685 dmu_tx_mark_netfree(tx);
1686 error = dmu_tx_assign(tx, TXG_WAIT);
1689 zfs_range_unlock(rl);
1694 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1695 NULL, &zp->z_size, sizeof (zp->z_size));
1698 zp->z_pflags &= ~ZFS_SPARSE;
1699 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1700 NULL, &zp->z_pflags, 8);
1702 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1707 * Clear any mapped pages in the truncated region. This has to
1708 * happen outside of the transaction to avoid the possibility of
1709 * a deadlock with someone trying to push a page that we are
1710 * about to invalidate.
1712 vnode_pager_setsize(vp, end);
1714 zfs_range_unlock(rl);
1720 * Free space in a file
1722 * IN: zp - znode of file to free data in.
1723 * off - start of range
1724 * len - end of range (0 => EOF)
1725 * flag - current file open mode flags.
1726 * log - TRUE if this action should be logged
1728 * RETURN: 0 on success, error code on failure
1731 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1733 vnode_t *vp = ZTOV(zp);
1735 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1736 zilog_t *zilog = zfsvfs->z_log;
1738 uint64_t mtime[2], ctime[2];
1739 sa_bulk_attr_t bulk[3];
1743 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1744 sizeof (mode))) != 0)
1747 if (off > zp->z_size) {
1748 error = zfs_extend(zp, off+len);
1749 if (error == 0 && log)
1756 * Check for any locks in the region to be freed.
1759 if (MANDLOCK(vp, (mode_t)mode)) {
1760 uint64_t length = (len ? len : zp->z_size - off);
1761 if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1766 error = zfs_trunc(zp, off);
1768 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1769 off + len > zp->z_size)
1770 error = zfs_extend(zp, off+len);
1775 tx = dmu_tx_create(zfsvfs->z_os);
1776 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1777 zfs_sa_upgrade_txholds(tx, zp);
1778 error = dmu_tx_assign(tx, TXG_WAIT);
1784 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1785 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1786 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1787 NULL, &zp->z_pflags, 8);
1788 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1789 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1792 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1799 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1801 uint64_t moid, obj, sa_obj, version;
1802 uint64_t sense = ZFS_CASE_SENSITIVE;
1807 znode_t *rootzp = NULL;
1811 zfs_acl_ids_t acl_ids;
1814 * First attempt to create master node.
1817 * In an empty objset, there are no blocks to read and thus
1818 * there can be no i/o errors (which we assert below).
1820 moid = MASTER_NODE_OBJ;
1821 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1822 DMU_OT_NONE, 0, tx);
1826 * Set starting attributes.
1828 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1830 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1831 /* For the moment we expect all zpl props to be uint64_ts */
1835 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1836 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1837 name = nvpair_name(elem);
1838 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1842 error = zap_update(os, moid, name, 8, 1, &val, tx);
1845 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1847 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1850 ASSERT(version != 0);
1851 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1854 * Create zap object used for SA attribute registration
1857 if (version >= ZPL_VERSION_SA) {
1858 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1859 DMU_OT_NONE, 0, tx);
1860 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1866 * Create a delete queue.
1868 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1870 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1874 * Create root znode. Create minimal znode/vnode/zfsvfs
1875 * to allow zfs_mknode to work.
1878 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1879 vattr.va_type = VDIR;
1880 vattr.va_mode = S_IFDIR|0755;
1881 vattr.va_uid = crgetuid(cr);
1882 vattr.va_gid = crgetgid(cr);
1884 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1886 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1887 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1888 rootzp->z_moved = 0;
1889 rootzp->z_unlinked = 0;
1890 rootzp->z_atime_dirty = 0;
1891 rootzp->z_is_sa = USE_SA(version, os);
1894 zfsvfs->z_parent = zfsvfs;
1895 zfsvfs->z_version = version;
1896 zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1897 zfsvfs->z_use_sa = USE_SA(version, os);
1898 zfsvfs->z_norm = norm;
1900 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1901 &zfsvfs->z_attr_table);
1906 * Fold case on file systems that are always or sometimes case
1909 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1910 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1912 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1913 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1914 offsetof(znode_t, z_link_node));
1916 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1917 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1919 rootzp->z_zfsvfs = zfsvfs;
1920 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1921 cr, NULL, &acl_ids));
1922 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1923 ASSERT3P(zp, ==, rootzp);
1924 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1926 zfs_acl_ids_free(&acl_ids);
1927 POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1929 sa_handle_destroy(rootzp->z_sa_hdl);
1930 kmem_cache_free(znode_cache, rootzp);
1933 * Create shares directory
1936 error = zfs_create_share_dir(zfsvfs, tx);
1940 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1941 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1942 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1945 #endif /* _KERNEL */
1948 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1950 uint64_t sa_obj = 0;
1953 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1954 if (error != 0 && error != ENOENT)
1957 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1962 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1963 dmu_buf_t **db, void *tag)
1965 dmu_object_info_t doi;
1968 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
1971 dmu_object_info_from_db(*db, &doi);
1972 if ((doi.doi_bonus_type != DMU_OT_SA &&
1973 doi.doi_bonus_type != DMU_OT_ZNODE) ||
1974 doi.doi_bonus_type == DMU_OT_ZNODE &&
1975 doi.doi_bonus_size < sizeof (znode_phys_t)) {
1976 sa_buf_rele(*db, tag);
1977 return (SET_ERROR(ENOTSUP));
1980 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
1982 sa_buf_rele(*db, tag);
1990 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
1992 sa_handle_destroy(hdl);
1993 sa_buf_rele(db, tag);
1997 * Given an object number, return its parent object number and whether
1998 * or not the object is an extended attribute directory.
2001 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
2002 uint64_t *pobjp, int *is_xattrdir)
2007 uint64_t parent_mode;
2008 sa_bulk_attr_t bulk[3];
2009 sa_handle_t *sa_hdl;
2014 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
2015 &parent, sizeof (parent));
2016 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
2017 &pflags, sizeof (pflags));
2018 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2019 &mode, sizeof (mode));
2021 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
2025 * When a link is removed its parent pointer is not changed and will
2026 * be invalid. There are two cases where a link is removed but the
2027 * file stays around, when it goes to the delete queue and when there
2028 * are additional links.
2030 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
2034 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
2035 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2039 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
2042 * Extended attributes can be applied to files, directories, etc.
2043 * Otherwise the parent must be a directory.
2045 if (!*is_xattrdir && !S_ISDIR(parent_mode))
2046 return (SET_ERROR(EINVAL));
2054 * Given an object number, return some zpl level statistics
2057 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2060 sa_bulk_attr_t bulk[4];
2063 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2064 &sb->zs_mode, sizeof (sb->zs_mode));
2065 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2066 &sb->zs_gen, sizeof (sb->zs_gen));
2067 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2068 &sb->zs_links, sizeof (sb->zs_links));
2069 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2070 &sb->zs_ctime, sizeof (sb->zs_ctime));
2072 return (sa_bulk_lookup(hdl, bulk, count));
2076 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2077 sa_attr_type_t *sa_table, char *buf, int len)
2079 sa_handle_t *sa_hdl;
2080 sa_handle_t *prevhdl = NULL;
2081 dmu_buf_t *prevdb = NULL;
2082 dmu_buf_t *sa_db = NULL;
2083 char *path = buf + len - 1;
2091 char component[MAXNAMELEN + 2];
2096 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2098 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2099 &is_xattrdir)) != 0)
2110 (void) sprintf(component + 1, "<xattrdir>");
2112 error = zap_value_search(osp, pobj, obj,
2113 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2118 complen = strlen(component);
2120 ASSERT(path >= buf);
2121 bcopy(component, path, complen);
2124 if (sa_hdl != hdl) {
2128 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2136 if (sa_hdl != NULL && sa_hdl != hdl) {
2137 ASSERT(sa_db != NULL);
2138 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2142 (void) memmove(buf, path, buf + len - path);
2148 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2150 sa_attr_type_t *sa_table;
2155 error = zfs_sa_setup(osp, &sa_table);
2159 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2163 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2165 zfs_release_sa_handle(hdl, db, FTAG);
2170 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2173 char *path = buf + len - 1;
2174 sa_attr_type_t *sa_table;
2181 error = zfs_sa_setup(osp, &sa_table);
2185 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2189 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2191 zfs_release_sa_handle(hdl, db, FTAG);
2195 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2197 zfs_release_sa_handle(hdl, db, FTAG);