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.
25 /* Portions Copyright 2007 Jeremy Teo */
28 #include <sys/types.h>
29 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
33 #include <sys/resource.h>
34 #include <sys/mntent.h>
35 #include <sys/u8_textprep.h>
36 #include <sys/dsl_dataset.h>
38 #include <sys/vnode.h>
41 #include <sys/errno.h>
42 #include <sys/unistd.h>
43 #include <sys/atomic.h>
44 #include <sys/zfs_dir.h>
45 #include <sys/zfs_acl.h>
46 #include <sys/zfs_ioctl.h>
47 #include <sys/zfs_rlock.h>
48 #include <sys/zfs_fuid.h>
49 #include <sys/dnode.h>
50 #include <sys/fs/zfs.h>
51 #include <sys/kidmap.h>
55 #include <sys/refcount.h>
58 #include <sys/zfs_znode.h>
60 #include <sys/zfs_sa.h>
61 #include <sys/zfs_stat.h>
62 #include <sys/refcount.h>
65 #include "zfs_comutil.h"
67 /* Used by fstat(1). */
68 SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 0, sizeof(znode_t),
72 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
73 * turned on when DEBUG is also defined.
80 #define ZNODE_STAT_ADD(stat) ((stat)++)
82 #define ZNODE_STAT_ADD(stat) /* nothing */
83 #endif /* ZNODE_STATS */
86 * Functions needed for userland (ie: libzpool) are not put under
87 * #ifdef_KERNEL; the rest of the functions have dependencies
88 * (such as VFS logic) that will not compile easily in userland.
92 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to
93 * be freed before it can be safely accessed.
95 krwlock_t zfsvfs_lock;
97 static kmem_cache_t *znode_cache = NULL;
101 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
104 * We should never drop all dbuf refs without first clearing
105 * the eviction callback.
107 panic("evicting znode %p\n", user_ptr);
110 extern struct vop_vector zfs_vnodeops;
111 extern struct vop_vector zfs_fifoops;
112 extern struct vop_vector zfs_shareops;
115 * XXX: We cannot use this function as a cache constructor, because
116 * there is one global cache for all file systems and we need
117 * to pass vfsp here, which is not possible, because argument
118 * 'cdrarg' is defined at kmem_cache_create() time.
122 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
129 POINTER_INVALIDATE(&zp->z_zfsvfs);
130 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
133 error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp);
134 if (error != 0 && (kmflags & KM_NOSLEEP))
137 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
139 vp->v_data = (caddr_t)zp;
146 list_link_init(&zp->z_link_node);
148 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
149 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
150 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
151 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
153 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
154 avl_create(&zp->z_range_avl, zfs_range_compare,
155 sizeof (rl_t), offsetof(rl_t, r_node));
157 zp->z_dirlocks = NULL;
158 zp->z_acl_cached = NULL;
165 zfs_znode_cache_destructor(void *buf, void *arg)
169 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
170 ASSERT(ZTOV(zp) == NULL);
172 ASSERT(!list_link_active(&zp->z_link_node));
173 mutex_destroy(&zp->z_lock);
174 rw_destroy(&zp->z_parent_lock);
175 rw_destroy(&zp->z_name_lock);
176 mutex_destroy(&zp->z_acl_lock);
177 avl_destroy(&zp->z_range_avl);
178 mutex_destroy(&zp->z_range_lock);
180 ASSERT(zp->z_dirlocks == NULL);
181 ASSERT(zp->z_acl_cached == NULL);
186 uint64_t zms_zfsvfs_invalid;
187 uint64_t zms_zfsvfs_recheck1;
188 uint64_t zms_zfsvfs_unmounted;
189 uint64_t zms_zfsvfs_recheck2;
190 uint64_t zms_obj_held;
191 uint64_t zms_vnode_locked;
192 uint64_t zms_not_only_dnlc;
194 #endif /* ZNODE_STATS */
198 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
203 nzp->z_zfsvfs = ozp->z_zfsvfs;
207 nzp->z_vnode = ozp->z_vnode;
208 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
209 ZTOV(ozp)->v_data = ozp;
210 ZTOV(nzp)->v_data = nzp;
212 nzp->z_id = ozp->z_id;
213 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
214 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0);
215 nzp->z_unlinked = ozp->z_unlinked;
216 nzp->z_atime_dirty = ozp->z_atime_dirty;
217 nzp->z_zn_prefetch = ozp->z_zn_prefetch;
218 nzp->z_blksz = ozp->z_blksz;
219 nzp->z_seq = ozp->z_seq;
220 nzp->z_mapcnt = ozp->z_mapcnt;
221 nzp->z_gen = ozp->z_gen;
222 nzp->z_sync_cnt = ozp->z_sync_cnt;
223 nzp->z_is_sa = ozp->z_is_sa;
224 nzp->z_sa_hdl = ozp->z_sa_hdl;
225 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
226 nzp->z_links = ozp->z_links;
227 nzp->z_size = ozp->z_size;
228 nzp->z_pflags = ozp->z_pflags;
229 nzp->z_uid = ozp->z_uid;
230 nzp->z_gid = ozp->z_gid;
231 nzp->z_mode = ozp->z_mode;
234 * Since this is just an idle znode and kmem is already dealing with
235 * memory pressure, release any cached ACL.
237 if (ozp->z_acl_cached) {
238 zfs_acl_free(ozp->z_acl_cached);
239 ozp->z_acl_cached = NULL;
242 sa_set_userp(nzp->z_sa_hdl, nzp);
245 * Invalidate the original znode by clearing fields that provide a
246 * pointer back to the znode. Set the low bit of the vfs pointer to
247 * ensure that zfs_znode_move() recognizes the znode as invalid in any
248 * subsequent callback.
250 ozp->z_sa_hdl = NULL;
251 POINTER_INVALIDATE(&ozp->z_zfsvfs);
257 ozp->z_moved = (uint8_t)-1;
262 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
264 znode_t *ozp = buf, *nzp = newbuf;
269 * The znode is on the file system's list of known znodes if the vfs
270 * pointer is valid. We set the low bit of the vfs pointer when freeing
271 * the znode to invalidate it, and the memory patterns written by kmem
272 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
273 * created znode sets the vfs pointer last of all to indicate that the
274 * znode is known and in a valid state to be moved by this function.
276 zfsvfs = ozp->z_zfsvfs;
277 if (!POINTER_IS_VALID(zfsvfs)) {
278 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
279 return (KMEM_CBRC_DONT_KNOW);
283 * Close a small window in which it's possible that the filesystem could
284 * be unmounted and freed, and zfsvfs, though valid in the previous
285 * statement, could point to unrelated memory by the time we try to
286 * prevent the filesystem from being unmounted.
288 rw_enter(&zfsvfs_lock, RW_WRITER);
289 if (zfsvfs != ozp->z_zfsvfs) {
290 rw_exit(&zfsvfs_lock);
291 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
292 return (KMEM_CBRC_DONT_KNOW);
296 * If the znode is still valid, then so is the file system. We know that
297 * no valid file system can be freed while we hold zfsvfs_lock, so we
298 * can safely ensure that the filesystem is not and will not be
299 * unmounted. The next statement is equivalent to ZFS_ENTER().
301 rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
302 if (zfsvfs->z_unmounted) {
304 rw_exit(&zfsvfs_lock);
305 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
306 return (KMEM_CBRC_DONT_KNOW);
308 rw_exit(&zfsvfs_lock);
310 mutex_enter(&zfsvfs->z_znodes_lock);
312 * Recheck the vfs pointer in case the znode was removed just before
313 * acquiring the lock.
315 if (zfsvfs != ozp->z_zfsvfs) {
316 mutex_exit(&zfsvfs->z_znodes_lock);
318 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
319 return (KMEM_CBRC_DONT_KNOW);
323 * At this point we know that as long as we hold z_znodes_lock, the
324 * znode cannot be freed and fields within the znode can be safely
325 * accessed. Now, prevent a race with zfs_zget().
327 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
328 mutex_exit(&zfsvfs->z_znodes_lock);
330 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
331 return (KMEM_CBRC_LATER);
335 if (mutex_tryenter(&vp->v_lock) == 0) {
336 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
337 mutex_exit(&zfsvfs->z_znodes_lock);
339 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
340 return (KMEM_CBRC_LATER);
343 /* Only move znodes that are referenced _only_ by the DNLC. */
344 if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
345 mutex_exit(&vp->v_lock);
346 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
347 mutex_exit(&zfsvfs->z_znodes_lock);
349 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
350 return (KMEM_CBRC_LATER);
354 * The znode is known and in a valid state to move. We're holding the
355 * locks needed to execute the critical section.
357 zfs_znode_move_impl(ozp, nzp);
358 mutex_exit(&vp->v_lock);
359 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
361 list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
362 mutex_exit(&zfsvfs->z_znodes_lock);
365 return (KMEM_CBRC_YES);
375 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
376 ASSERT(znode_cache == NULL);
377 znode_cache = kmem_cache_create("zfs_znode_cache",
378 sizeof (znode_t), 0, /* zfs_znode_cache_constructor */ NULL,
379 zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
380 kmem_cache_set_move(znode_cache, zfs_znode_move);
388 * Cleanup vfs & vnode ops
390 zfs_remove_op_tables();
397 kmem_cache_destroy(znode_cache);
399 rw_destroy(&zfsvfs_lock);
403 struct vnodeops *zfs_dvnodeops;
404 struct vnodeops *zfs_fvnodeops;
405 struct vnodeops *zfs_symvnodeops;
406 struct vnodeops *zfs_xdvnodeops;
407 struct vnodeops *zfs_evnodeops;
408 struct vnodeops *zfs_sharevnodeops;
411 zfs_remove_op_tables()
417 (void) vfs_freevfsops_by_type(zfsfstype);
424 vn_freevnodeops(zfs_dvnodeops);
426 vn_freevnodeops(zfs_fvnodeops);
428 vn_freevnodeops(zfs_symvnodeops);
430 vn_freevnodeops(zfs_xdvnodeops);
432 vn_freevnodeops(zfs_evnodeops);
433 if (zfs_sharevnodeops)
434 vn_freevnodeops(zfs_sharevnodeops);
436 zfs_dvnodeops = NULL;
437 zfs_fvnodeops = NULL;
438 zfs_symvnodeops = NULL;
439 zfs_xdvnodeops = NULL;
440 zfs_evnodeops = NULL;
441 zfs_sharevnodeops = NULL;
444 extern const fs_operation_def_t zfs_dvnodeops_template[];
445 extern const fs_operation_def_t zfs_fvnodeops_template[];
446 extern const fs_operation_def_t zfs_xdvnodeops_template[];
447 extern const fs_operation_def_t zfs_symvnodeops_template[];
448 extern const fs_operation_def_t zfs_evnodeops_template[];
449 extern const fs_operation_def_t zfs_sharevnodeops_template[];
452 zfs_create_op_tables()
457 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
458 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
459 * In this case we just return as the ops vectors are already set up.
464 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
469 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
474 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
479 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
484 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
489 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
497 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
499 zfs_acl_ids_t acl_ids;
506 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
507 vattr.va_type = VDIR;
508 vattr.va_mode = S_IFDIR|0555;
509 vattr.va_uid = crgetuid(kcred);
510 vattr.va_gid = crgetgid(kcred);
512 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
513 zfs_znode_cache_constructor(sharezp, zfsvfs->z_parent->z_vfs, 0);
514 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
515 sharezp->z_moved = 0;
516 sharezp->z_unlinked = 0;
517 sharezp->z_atime_dirty = 0;
518 sharezp->z_zfsvfs = zfsvfs;
519 sharezp->z_is_sa = zfsvfs->z_use_sa;
521 sharezp->z_vnode = &vnode;
522 vnode.v_data = sharezp;
527 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
528 kcred, NULL, &acl_ids));
529 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
530 ASSERT3P(zp, ==, sharezp);
531 POINTER_INVALIDATE(&sharezp->z_zfsvfs);
532 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
533 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
534 zfsvfs->z_shares_dir = sharezp->z_id;
536 zfs_acl_ids_free(&acl_ids);
537 ZTOV(sharezp)->v_data = NULL;
538 ZTOV(sharezp)->v_count = 0;
539 ZTOV(sharezp)->v_holdcnt = 0;
541 sa_handle_destroy(sharezp->z_sa_hdl);
542 sharezp->z_vnode = NULL;
543 kmem_cache_free(znode_cache, sharezp);
549 * define a couple of values we need available
550 * for both 64 and 32 bit environments.
553 #define NBITSMINOR64 32
556 #define MAXMAJ64 0xffffffffUL
559 #define MAXMIN64 0xffffffffUL
563 * Create special expldev for ZFS private use.
564 * Can't use standard expldev since it doesn't do
565 * what we want. The standard expldev() takes a
566 * dev32_t in LP64 and expands it to a long dev_t.
567 * We need an interface that takes a dev32_t in ILP32
568 * and expands it to a long dev_t.
571 zfs_expldev(dev_t dev)
573 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev));
576 * Special cmpldev for ZFS private use.
577 * Can't use standard cmpldev since it takes
578 * a long dev_t and compresses it to dev32_t in
579 * LP64. We need to do a compaction of a long dev_t
580 * to a dev32_t in ILP32.
583 zfs_cmpldev(uint64_t dev)
585 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64)));
589 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
590 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
592 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
593 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
595 mutex_enter(&zp->z_lock);
597 ASSERT(zp->z_sa_hdl == NULL);
598 ASSERT(zp->z_acl_cached == NULL);
599 if (sa_hdl == NULL) {
600 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
601 SA_HDL_SHARED, &zp->z_sa_hdl));
603 zp->z_sa_hdl = sa_hdl;
604 sa_set_userp(sa_hdl, zp);
607 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
610 * Slap on VROOT if we are the root znode
612 if (zp->z_id == zfsvfs->z_root)
613 ZTOV(zp)->v_flag |= VROOT;
615 mutex_exit(&zp->z_lock);
620 zfs_znode_dmu_fini(znode_t *zp)
622 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
624 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
626 sa_handle_destroy(zp->z_sa_hdl);
631 zfs_vnode_forget(vnode_t *vp)
637 vp->v_iflag |= VI_DOOMED;
643 * Construct a new znode/vnode and intialize.
645 * This does not do a call to dmu_set_user() that is
646 * up to the caller to do, in case you don't want to
650 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
651 dmu_object_type_t obj_type, sa_handle_t *hdl)
657 sa_bulk_attr_t bulk[9];
660 zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
661 zfs_znode_cache_constructor(zp, zfsvfs->z_parent->z_vfs, 0);
663 ASSERT(zp->z_dirlocks == NULL);
664 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
668 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
669 * the zfs_znode_move() callback.
673 zp->z_atime_dirty = 0;
675 zp->z_id = db->db_object;
677 zp->z_seq = 0x7A4653;
682 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
684 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
685 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
686 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
688 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
690 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
692 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
693 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
695 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
697 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
700 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) {
702 sa_handle_destroy(zp->z_sa_hdl);
703 zfs_vnode_forget(vp);
705 kmem_cache_free(znode_cache, zp);
711 vp->v_type = IFTOVT((mode_t)mode);
713 switch (vp->v_type) {
715 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
722 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
723 &rdev, sizeof (rdev)) == 0);
725 vp->v_rdev = zfs_cmpldev(rdev);
734 vp->v_op = &zfs_fifoops;
737 if (parent == zfsvfs->z_shares_dir) {
738 ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
739 vp->v_op = &zfs_shareops;
744 vn_setops(vp, zfs_symvnodeops);
747 vn_setops(vp, zfs_evnodeops);
751 if (vp->v_type != VFIFO)
754 mutex_enter(&zfsvfs->z_znodes_lock);
755 list_insert_tail(&zfsvfs->z_all_znodes, zp);
758 * Everything else must be valid before assigning z_zfsvfs makes the
759 * znode eligible for zfs_znode_move().
761 zp->z_zfsvfs = zfsvfs;
762 mutex_exit(&zfsvfs->z_znodes_lock);
764 VFS_HOLD(zfsvfs->z_vfs);
768 static uint64_t empty_xattr;
769 static uint64_t pad[4];
770 static zfs_acl_phys_t acl_phys;
772 * Create a new DMU object to hold a zfs znode.
774 * IN: dzp - parent directory for new znode
775 * vap - file attributes for new znode
776 * tx - dmu transaction id for zap operations
777 * cr - credentials of caller
779 * IS_ROOT_NODE - new object will be root
780 * IS_XATTR - new object is an attribute
781 * bonuslen - length of bonus buffer
782 * setaclp - File/Dir initial ACL
783 * fuidp - Tracks fuid allocation.
785 * OUT: zpp - allocated znode
789 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
790 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
792 uint64_t crtime[2], atime[2], mtime[2], ctime[2];
793 uint64_t mode, size, links, parent, pflags;
794 uint64_t dzp_pflags = 0;
796 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
803 dmu_object_type_t obj_type;
804 sa_bulk_attr_t sa_attrs[ZPL_END];
806 zfs_acl_locator_cb_t locate = { 0 };
808 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
810 if (zfsvfs->z_replay) {
811 obj = vap->va_nodeid;
812 now = vap->va_ctime; /* see zfs_replay_create() */
813 gen = vap->va_nblocks; /* ditto */
817 gen = dmu_tx_get_txg(tx);
820 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
821 bonuslen = (obj_type == DMU_OT_SA) ?
822 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
825 * Create a new DMU object.
828 * There's currently no mechanism for pre-reading the blocks that will
829 * be needed to allocate a new object, so we accept the small chance
830 * that there will be an i/o error and we will fail one of the
833 if (vap->va_type == VDIR) {
834 if (zfsvfs->z_replay) {
835 err = zap_create_claim_norm(zfsvfs->z_os, obj,
836 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
837 obj_type, bonuslen, tx);
838 ASSERT3U(err, ==, 0);
840 obj = zap_create_norm(zfsvfs->z_os,
841 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
842 obj_type, bonuslen, tx);
845 if (zfsvfs->z_replay) {
846 err = dmu_object_claim(zfsvfs->z_os, obj,
847 DMU_OT_PLAIN_FILE_CONTENTS, 0,
848 obj_type, bonuslen, tx);
849 ASSERT3U(err, ==, 0);
851 obj = dmu_object_alloc(zfsvfs->z_os,
852 DMU_OT_PLAIN_FILE_CONTENTS, 0,
853 obj_type, bonuslen, tx);
857 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
858 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
861 * If this is the root, fix up the half-initialized parent pointer
862 * to reference the just-allocated physical data area.
864 if (flag & IS_ROOT_NODE) {
867 dzp_pflags = dzp->z_pflags;
871 * If parent is an xattr, so am I.
873 if (dzp_pflags & ZFS_XATTR) {
877 if (zfsvfs->z_use_fuids)
878 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
882 if (vap->va_type == VDIR) {
883 size = 2; /* contents ("." and "..") */
884 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
889 if (vap->va_type == VBLK || vap->va_type == VCHR) {
890 rdev = zfs_expldev(vap->va_rdev);
894 mode = acl_ids->z_mode;
899 * No execs denied will be deterimed when zfs_mode_compute() is called.
901 pflags |= acl_ids->z_aclp->z_hints &
902 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
903 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
905 ZFS_TIME_ENCODE(&now, crtime);
906 ZFS_TIME_ENCODE(&now, ctime);
908 if (vap->va_mask & AT_ATIME) {
909 ZFS_TIME_ENCODE(&vap->va_atime, atime);
911 ZFS_TIME_ENCODE(&now, atime);
914 if (vap->va_mask & AT_MTIME) {
915 ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
917 ZFS_TIME_ENCODE(&now, mtime);
920 /* Now add in all of the "SA" attributes */
921 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
925 * Setup the array of attributes to be replaced/set on the new file
927 * order for DMU_OT_ZNODE is critical since it needs to be constructed
928 * in the old znode_phys_t format. Don't change this ordering
931 if (obj_type == DMU_OT_ZNODE) {
932 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
934 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
936 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
938 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
940 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
942 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
944 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
946 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
949 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
951 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
953 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
955 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
956 &acl_ids->z_fuid, 8);
957 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
958 &acl_ids->z_fgid, 8);
959 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
961 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
963 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
965 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
967 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
969 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
973 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
975 if (obj_type == DMU_OT_ZNODE) {
976 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
979 if (obj_type == DMU_OT_ZNODE ||
980 (vap->va_type == VBLK || vap->va_type == VCHR)) {
981 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
985 if (obj_type == DMU_OT_ZNODE) {
986 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
988 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
989 &acl_ids->z_fuid, 8);
990 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
991 &acl_ids->z_fgid, 8);
992 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
993 sizeof (uint64_t) * 4);
994 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
995 &acl_phys, sizeof (zfs_acl_phys_t));
996 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
997 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
998 &acl_ids->z_aclp->z_acl_count, 8);
999 locate.cb_aclp = acl_ids->z_aclp;
1000 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
1001 zfs_acl_data_locator, &locate,
1002 acl_ids->z_aclp->z_acl_bytes);
1003 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
1004 acl_ids->z_fuid, acl_ids->z_fgid);
1007 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
1009 if (!(flag & IS_ROOT_NODE)) {
1010 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
1011 ASSERT(*zpp != NULL);
1014 * If we are creating the root node, the "parent" we
1015 * passed in is the znode for the root.
1019 (*zpp)->z_sa_hdl = sa_hdl;
1022 (*zpp)->z_pflags = pflags;
1023 (*zpp)->z_mode = mode;
1025 if (vap->va_mask & AT_XVATTR)
1026 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
1028 if (obj_type == DMU_OT_ZNODE ||
1029 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1030 err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx);
1031 ASSERT3P(err, ==, 0);
1033 if (!(flag & IS_ROOT_NODE)) {
1037 vp->v_vflag |= VV_FORCEINSMQ;
1038 err = insmntque(vp, zfsvfs->z_vfs);
1039 vp->v_vflag &= ~VV_FORCEINSMQ;
1040 KASSERT(err == 0, ("insmntque() failed: error %d", err));
1042 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1046 * zfs_xvattr_set only updates the in-core attributes
1047 * it is assumed the caller will be doing an sa_bulk_update
1048 * to push the changes out
1051 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1055 xoap = xva_getxoptattr(xvap);
1058 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1060 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1061 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1062 ×, sizeof (times), tx);
1063 XVA_SET_RTN(xvap, XAT_CREATETIME);
1065 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1066 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1068 XVA_SET_RTN(xvap, XAT_READONLY);
1070 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1071 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1073 XVA_SET_RTN(xvap, XAT_HIDDEN);
1075 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1076 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1078 XVA_SET_RTN(xvap, XAT_SYSTEM);
1080 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1081 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1083 XVA_SET_RTN(xvap, XAT_ARCHIVE);
1085 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1086 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1088 XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1090 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1091 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1093 XVA_SET_RTN(xvap, XAT_NOUNLINK);
1095 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1096 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1098 XVA_SET_RTN(xvap, XAT_APPENDONLY);
1100 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1101 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1103 XVA_SET_RTN(xvap, XAT_NODUMP);
1105 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1106 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1108 XVA_SET_RTN(xvap, XAT_OPAQUE);
1110 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1111 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1112 xoap->xoa_av_quarantined, zp->z_pflags, tx);
1113 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1115 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1116 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1118 XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1120 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1121 zfs_sa_set_scanstamp(zp, xvap, tx);
1122 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1124 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1125 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1127 XVA_SET_RTN(xvap, XAT_REPARSE);
1129 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1130 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1132 XVA_SET_RTN(xvap, XAT_OFFLINE);
1134 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1135 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1137 XVA_SET_RTN(xvap, XAT_SPARSE);
1142 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1144 dmu_object_info_t doi;
1154 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1156 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1158 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1162 dmu_object_info_from_db(db, &doi);
1163 if (doi.doi_bonus_type != DMU_OT_SA &&
1164 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1165 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1166 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1167 sa_buf_rele(db, NULL);
1168 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1172 hdl = dmu_buf_get_user(db);
1174 zp = sa_get_userdata(hdl);
1178 * Since "SA" does immediate eviction we
1179 * should never find a sa handle that doesn't
1180 * know about the znode.
1183 ASSERT3P(zp, !=, NULL);
1185 mutex_enter(&zp->z_lock);
1186 ASSERT3U(zp->z_id, ==, obj_num);
1187 if (zp->z_unlinked) {
1198 if ((vp->v_iflag & VI_DOOMED) != 0) {
1201 * Don't VN_RELE() vnode here, because
1202 * it can call vn_lock() which creates
1203 * LOR between vnode lock and znode
1204 * lock. We will VN_RELE() the vnode
1205 * after droping znode lock.
1211 ZFS_LOG(1, "dying znode detected (zp=%p)", zp);
1215 * znode is dying so we can't reuse it, we must
1216 * wait until destruction is completed.
1218 sa_buf_rele(db, NULL);
1219 mutex_exit(&zp->z_lock);
1220 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1223 tsleep(zp, 0, "zcollide", 1);
1229 sa_buf_rele(db, NULL);
1230 mutex_exit(&zp->z_lock);
1231 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1236 * Not found create new znode/vnode
1237 * but only if file exists.
1239 * There is a small window where zfs_vget() could
1240 * find this object while a file create is still in
1241 * progress. This is checked for in zfs_znode_alloc()
1243 * if zfs_znode_alloc() fails it will drop the hold on the
1246 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1247 doi.doi_bonus_type, NULL);
1254 vnode_t *vp = ZTOV(zp);
1256 err = insmntque(vp, zfsvfs->z_vfs);
1261 zfs_znode_dmu_fini(zp);
1266 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1271 zfs_rezget(znode_t *zp)
1273 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1274 dmu_object_info_t doi;
1277 uint64_t obj_num = zp->z_id;
1278 uint64_t mode, size;
1279 sa_bulk_attr_t bulk[8];
1284 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1286 mutex_enter(&zp->z_acl_lock);
1287 if (zp->z_acl_cached) {
1288 zfs_acl_free(zp->z_acl_cached);
1289 zp->z_acl_cached = NULL;
1292 mutex_exit(&zp->z_acl_lock);
1293 ASSERT(zp->z_sa_hdl == NULL);
1294 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1296 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1300 dmu_object_info_from_db(db, &doi);
1301 if (doi.doi_bonus_type != DMU_OT_SA &&
1302 (doi.doi_bonus_type != DMU_OT_ZNODE ||
1303 (doi.doi_bonus_type == DMU_OT_ZNODE &&
1304 doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1305 sa_buf_rele(db, NULL);
1306 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1310 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1313 /* reload cached values */
1314 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1315 &gen, sizeof (gen));
1316 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1317 &zp->z_size, sizeof (zp->z_size));
1318 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1319 &zp->z_links, sizeof (zp->z_links));
1320 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1321 &zp->z_pflags, sizeof (zp->z_pflags));
1322 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1323 &zp->z_atime, sizeof (zp->z_atime));
1324 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1325 &zp->z_uid, sizeof (zp->z_uid));
1326 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1327 &zp->z_gid, sizeof (zp->z_gid));
1328 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1329 &mode, sizeof (mode));
1331 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1332 zfs_znode_dmu_fini(zp);
1333 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1339 if (gen != zp->z_gen) {
1340 zfs_znode_dmu_fini(zp);
1341 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1346 * XXXPJD: Not sure how is that possible, but under heavy
1347 * zfs recv -F load it happens that z_gen is the same, but
1348 * vnode type is different than znode type. This would mean
1349 * that for example regular file was replaced with directory
1350 * which has the same object number.
1354 vp->v_type != IFTOVT((mode_t)zp->z_mode)) {
1355 zfs_znode_dmu_fini(zp);
1356 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1360 zp->z_unlinked = (zp->z_links == 0);
1361 zp->z_blksz = doi.doi_data_block_size;
1363 vn_pages_remove(vp, 0, 0);
1364 if (zp->z_size != size)
1365 vnode_pager_setsize(vp, zp->z_size);
1368 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1374 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1376 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1377 objset_t *os = zfsvfs->z_os;
1378 uint64_t obj = zp->z_id;
1379 uint64_t acl_obj = zfs_external_acl(zp);
1381 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1383 VERIFY(!zp->z_is_sa);
1384 VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1386 VERIFY(0 == dmu_object_free(os, obj, tx));
1387 zfs_znode_dmu_fini(zp);
1388 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1393 zfs_zinactive(znode_t *zp)
1395 vnode_t *vp = ZTOV(zp);
1396 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1397 uint64_t z_id = zp->z_id;
1400 ASSERT(zp->z_sa_hdl);
1403 * Don't allow a zfs_zget() while were trying to release this znode
1405 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1407 mutex_enter(&zp->z_lock);
1409 if (vp->v_count > 0) {
1411 * If the hold count is greater than zero, somebody has
1412 * obtained a new reference on this znode while we were
1413 * processing it here, so we are done.
1416 mutex_exit(&zp->z_lock);
1417 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1423 * If this was the last reference to a file with no links,
1424 * remove the file from the file system.
1426 if (zp->z_unlinked) {
1427 mutex_exit(&zp->z_lock);
1428 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1429 ASSERT(vp->v_count == 0);
1430 vrecycle(vp, curthread);
1431 vfslocked = VFS_LOCK_GIANT(zfsvfs->z_vfs);
1433 VFS_UNLOCK_GIANT(vfslocked);
1437 mutex_exit(&zp->z_lock);
1438 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1442 zfs_znode_free(znode_t *zp)
1444 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1446 ASSERT(ZTOV(zp) == NULL);
1447 ASSERT(zp->z_sa_hdl == NULL);
1448 mutex_enter(&zfsvfs->z_znodes_lock);
1449 POINTER_INVALIDATE(&zp->z_zfsvfs);
1450 list_remove(&zfsvfs->z_all_znodes, zp);
1451 mutex_exit(&zfsvfs->z_znodes_lock);
1453 if (zp->z_acl_cached) {
1454 zfs_acl_free(zp->z_acl_cached);
1455 zp->z_acl_cached = NULL;
1458 kmem_cache_free(znode_cache, zp);
1460 VFS_RELE(zfsvfs->z_vfs);
1464 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1465 uint64_t ctime[2], boolean_t have_tx)
1471 if (have_tx) { /* will sa_bulk_update happen really soon? */
1472 zp->z_atime_dirty = 0;
1475 zp->z_atime_dirty = 1;
1478 if (flag & AT_ATIME) {
1479 ZFS_TIME_ENCODE(&now, zp->z_atime);
1482 if (flag & AT_MTIME) {
1483 ZFS_TIME_ENCODE(&now, mtime);
1484 if (zp->z_zfsvfs->z_use_fuids) {
1485 zp->z_pflags |= (ZFS_ARCHIVE |
1490 if (flag & AT_CTIME) {
1491 ZFS_TIME_ENCODE(&now, ctime);
1492 if (zp->z_zfsvfs->z_use_fuids)
1493 zp->z_pflags |= ZFS_ARCHIVE;
1498 * Grow the block size for a file.
1500 * IN: zp - znode of file to free data in.
1501 * size - requested block size
1502 * tx - open transaction.
1504 * NOTE: this function assumes that the znode is write locked.
1507 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1512 if (size <= zp->z_blksz)
1515 * If the file size is already greater than the current blocksize,
1516 * we will not grow. If there is more than one block in a file,
1517 * the blocksize cannot change.
1519 if (zp->z_blksz && zp->z_size > zp->z_blksz)
1522 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1525 if (error == ENOTSUP)
1527 ASSERT3U(error, ==, 0);
1529 /* What blocksize did we actually get? */
1530 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1535 * This is a dummy interface used when pvn_vplist_dirty() should *not*
1536 * be calling back into the fs for a putpage(). E.g.: when truncating
1537 * a file, the pages being "thrown away* don't need to be written out.
1541 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1542 int flags, cred_t *cr)
1550 * Increase the file length
1552 * IN: zp - znode of file to free data in.
1553 * end - new end-of-file
1555 * RETURN: 0 if success
1556 * error code if failure
1559 zfs_extend(znode_t *zp, uint64_t end)
1561 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1568 * We will change zp_size, lock the whole file.
1570 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1573 * Nothing to do if file already at desired length.
1575 if (end <= zp->z_size) {
1576 zfs_range_unlock(rl);
1580 tx = dmu_tx_create(zfsvfs->z_os);
1581 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1582 zfs_sa_upgrade_txholds(tx, zp);
1583 if (end > zp->z_blksz &&
1584 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1586 * We are growing the file past the current block size.
1588 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1589 ASSERT(!ISP2(zp->z_blksz));
1590 newblksz = MIN(end, SPA_MAXBLOCKSIZE);
1592 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1594 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1599 error = dmu_tx_assign(tx, TXG_NOWAIT);
1601 if (error == ERESTART) {
1607 zfs_range_unlock(rl);
1612 zfs_grow_blocksize(zp, newblksz, tx);
1616 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1617 &zp->z_size, sizeof (zp->z_size), tx));
1619 vnode_pager_setsize(ZTOV(zp), end);
1621 zfs_range_unlock(rl);
1629 * Free space in a file.
1631 * IN: zp - znode of file to free data in.
1632 * off - start of section to free.
1633 * len - length of section to free.
1635 * RETURN: 0 if success
1636 * error code if failure
1639 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1641 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1646 * Lock the range being freed.
1648 rl = zfs_range_lock(zp, off, len, RL_WRITER);
1651 * Nothing to do if file already at desired length.
1653 if (off >= zp->z_size) {
1654 zfs_range_unlock(rl);
1658 if (off + len > zp->z_size)
1659 len = zp->z_size - off;
1661 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1665 * In FreeBSD we cannot free block in the middle of a file,
1666 * but only at the end of a file, so this code path should
1669 vnode_pager_setsize(ZTOV(zp), off);
1672 zfs_range_unlock(rl);
1680 * IN: zp - znode of file to free data in.
1681 * end - new end-of-file.
1683 * RETURN: 0 if success
1684 * error code if failure
1687 zfs_trunc(znode_t *zp, uint64_t end)
1689 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1690 vnode_t *vp = ZTOV(zp);
1694 sa_bulk_attr_t bulk[2];
1698 * We will change zp_size, lock the whole file.
1700 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1703 * Nothing to do if file already at desired length.
1705 if (end >= zp->z_size) {
1706 zfs_range_unlock(rl);
1710 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1);
1712 zfs_range_unlock(rl);
1716 tx = dmu_tx_create(zfsvfs->z_os);
1717 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1718 zfs_sa_upgrade_txholds(tx, zp);
1719 error = dmu_tx_assign(tx, TXG_NOWAIT);
1721 if (error == ERESTART) {
1727 zfs_range_unlock(rl);
1732 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1733 NULL, &zp->z_size, sizeof (zp->z_size));
1736 zp->z_pflags &= ~ZFS_SPARSE;
1737 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1738 NULL, &zp->z_pflags, 8);
1740 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1745 * Clear any mapped pages in the truncated region. This has to
1746 * happen outside of the transaction to avoid the possibility of
1747 * a deadlock with someone trying to push a page that we are
1748 * about to invalidate.
1750 vnode_pager_setsize(vp, end);
1752 zfs_range_unlock(rl);
1758 * Free space in a file
1760 * IN: zp - znode of file to free data in.
1761 * off - start of range
1762 * len - end of range (0 => EOF)
1763 * flag - current file open mode flags.
1764 * log - TRUE if this action should be logged
1766 * RETURN: 0 if success
1767 * error code if failure
1770 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1772 vnode_t *vp = ZTOV(zp);
1774 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1775 zilog_t *zilog = zfsvfs->z_log;
1777 uint64_t mtime[2], ctime[2];
1778 sa_bulk_attr_t bulk[3];
1782 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1783 sizeof (mode))) != 0)
1786 if (off > zp->z_size) {
1787 error = zfs_extend(zp, off+len);
1788 if (error == 0 && log)
1795 * Check for any locks in the region to be freed.
1798 if (MANDLOCK(vp, (mode_t)mode)) {
1799 uint64_t length = (len ? len : zp->z_size - off);
1800 if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1805 error = zfs_trunc(zp, off);
1807 if ((error = zfs_free_range(zp, off, len)) == 0 &&
1808 off + len > zp->z_size)
1809 error = zfs_extend(zp, off+len);
1814 tx = dmu_tx_create(zfsvfs->z_os);
1815 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1816 zfs_sa_upgrade_txholds(tx, zp);
1817 error = dmu_tx_assign(tx, TXG_NOWAIT);
1819 if (error == ERESTART) {
1828 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1829 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1830 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1831 NULL, &zp->z_pflags, 8);
1832 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1833 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1836 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1843 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1846 uint64_t moid, obj, sa_obj, version;
1847 uint64_t sense = ZFS_CASE_SENSITIVE;
1852 znode_t *rootzp = NULL;
1856 zfs_acl_ids_t acl_ids;
1859 * First attempt to create master node.
1862 * In an empty objset, there are no blocks to read and thus
1863 * there can be no i/o errors (which we assert below).
1865 moid = MASTER_NODE_OBJ;
1866 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1867 DMU_OT_NONE, 0, tx);
1871 * Set starting attributes.
1873 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1875 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1876 /* For the moment we expect all zpl props to be uint64_ts */
1880 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1881 VERIFY(nvpair_value_uint64(elem, &val) == 0);
1882 name = nvpair_name(elem);
1883 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1887 error = zap_update(os, moid, name, 8, 1, &val, tx);
1890 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1892 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1895 ASSERT(version != 0);
1896 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1899 * Create zap object used for SA attribute registration
1902 if (version >= ZPL_VERSION_SA) {
1903 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1904 DMU_OT_NONE, 0, tx);
1905 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1911 * Create a delete queue.
1913 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1915 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1919 * Create root znode. Create minimal znode/vnode/zfsvfs
1920 * to allow zfs_mknode to work.
1923 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1924 vattr.va_type = VDIR;
1925 vattr.va_mode = S_IFDIR|0755;
1926 vattr.va_uid = crgetuid(cr);
1927 vattr.va_gid = crgetgid(cr);
1929 bzero(&zfsvfs, sizeof (zfsvfs_t));
1931 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1932 zfs_znode_cache_constructor(rootzp, NULL, 0);
1933 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1934 rootzp->z_moved = 0;
1935 rootzp->z_unlinked = 0;
1936 rootzp->z_atime_dirty = 0;
1937 rootzp->z_is_sa = USE_SA(version, os);
1939 vnode.v_type = VDIR;
1940 vnode.v_data = rootzp;
1941 rootzp->z_vnode = &vnode;
1944 zfsvfs.z_parent = &zfsvfs;
1945 zfsvfs.z_version = version;
1946 zfsvfs.z_use_fuids = USE_FUIDS(version, os);
1947 zfsvfs.z_use_sa = USE_SA(version, os);
1948 zfsvfs.z_norm = norm;
1950 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1951 &zfsvfs.z_attr_table);
1956 * Fold case on file systems that are always or sometimes case
1959 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1960 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER;
1962 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1963 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t),
1964 offsetof(znode_t, z_link_node));
1966 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1967 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1969 rootzp->z_zfsvfs = &zfsvfs;
1970 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1971 cr, NULL, &acl_ids));
1972 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1973 ASSERT3P(zp, ==, rootzp);
1974 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1976 zfs_acl_ids_free(&acl_ids);
1977 POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1979 sa_handle_destroy(rootzp->z_sa_hdl);
1980 rootzp->z_vnode = NULL;
1981 kmem_cache_free(znode_cache, rootzp);
1984 * Create shares directory
1987 error = zfs_create_share_dir(&zfsvfs, tx);
1991 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1992 mutex_destroy(&zfsvfs.z_hold_mtx[i]);
1995 #endif /* _KERNEL */
1998 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
2000 uint64_t sa_obj = 0;
2003 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
2004 if (error != 0 && error != ENOENT)
2007 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
2012 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
2013 dmu_buf_t **db, void *tag)
2015 dmu_object_info_t doi;
2018 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
2021 dmu_object_info_from_db(*db, &doi);
2022 if ((doi.doi_bonus_type != DMU_OT_SA &&
2023 doi.doi_bonus_type != DMU_OT_ZNODE) ||
2024 doi.doi_bonus_type == DMU_OT_ZNODE &&
2025 doi.doi_bonus_size < sizeof (znode_phys_t)) {
2026 sa_buf_rele(*db, tag);
2030 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
2032 sa_buf_rele(*db, tag);
2040 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
2042 sa_handle_destroy(hdl);
2043 sa_buf_rele(db, tag);
2047 * Given an object number, return its parent object number and whether
2048 * or not the object is an extended attribute directory.
2051 zfs_obj_to_pobj(sa_handle_t *hdl, sa_attr_type_t *sa_table, uint64_t *pobjp,
2057 sa_bulk_attr_t bulk[3];
2061 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
2062 &parent, sizeof (parent));
2063 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
2064 &pflags, sizeof (pflags));
2065 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2066 &mode, sizeof (mode));
2068 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
2072 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
2078 * Given an object number, return some zpl level statistics
2081 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2084 sa_bulk_attr_t bulk[4];
2087 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2088 &sb->zs_mode, sizeof (sb->zs_mode));
2089 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2090 &sb->zs_gen, sizeof (sb->zs_gen));
2091 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2092 &sb->zs_links, sizeof (sb->zs_links));
2093 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2094 &sb->zs_ctime, sizeof (sb->zs_ctime));
2096 return (sa_bulk_lookup(hdl, bulk, count));
2100 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2101 sa_attr_type_t *sa_table, char *buf, int len)
2103 sa_handle_t *sa_hdl;
2104 sa_handle_t *prevhdl = NULL;
2105 dmu_buf_t *prevdb = NULL;
2106 dmu_buf_t *sa_db = NULL;
2107 char *path = buf + len - 1;
2115 char component[MAXNAMELEN + 2];
2120 zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2122 if ((error = zfs_obj_to_pobj(sa_hdl, sa_table, &pobj,
2123 &is_xattrdir)) != 0)
2134 (void) sprintf(component + 1, "<xattrdir>");
2136 error = zap_value_search(osp, pobj, obj,
2137 ZFS_DIRENT_OBJ(-1ULL), component + 1);
2142 complen = strlen(component);
2144 ASSERT(path >= buf);
2145 bcopy(component, path, complen);
2148 if (sa_hdl != hdl) {
2152 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2160 if (sa_hdl != NULL && sa_hdl != hdl) {
2161 ASSERT(sa_db != NULL);
2162 zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2166 (void) memmove(buf, path, buf + len - path);
2172 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2174 sa_attr_type_t *sa_table;
2179 error = zfs_sa_setup(osp, &sa_table);
2183 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2187 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2189 zfs_release_sa_handle(hdl, db, FTAG);
2194 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2197 char *path = buf + len - 1;
2198 sa_attr_type_t *sa_table;
2205 error = zfs_sa_setup(osp, &sa_table);
2209 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2213 error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2215 zfs_release_sa_handle(hdl, db, FTAG);
2219 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2221 zfs_release_sa_handle(hdl, db, FTAG);