xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_znode.c (revision bf7fda8965eb0f1d22b8e7bf1684b99227cd2b64)
1 /*
2  * CDDL HEADER START
3  *
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.
7  *
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.
12  *
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]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 /* Portions Copyright 2007 Jeremy Teo */
26 
27 #ifdef _KERNEL
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/time.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
33 #include <sys/resource.h>
34 #include <sys/mntent.h>
35 #include <sys/mkdev.h>
36 #include <sys/u8_textprep.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/vfs.h>
39 #include <sys/vfs_opreg.h>
40 #include <sys/vnode.h>
41 #include <sys/file.h>
42 #include <sys/kmem.h>
43 #include <sys/errno.h>
44 #include <sys/unistd.h>
45 #include <sys/mode.h>
46 #include <sys/atomic.h>
47 #include <vm/pvn.h>
48 #include "fs/fs_subr.h"
49 #include <sys/zfs_dir.h>
50 #include <sys/zfs_acl.h>
51 #include <sys/zfs_ioctl.h>
52 #include <sys/zfs_rlock.h>
53 #include <sys/zfs_fuid.h>
54 #include <sys/dnode.h>
55 #include <sys/fs/zfs.h>
56 #include <sys/kidmap.h>
57 #endif /* _KERNEL */
58 
59 #include <sys/dmu.h>
60 #include <sys/refcount.h>
61 #include <sys/stat.h>
62 #include <sys/zap.h>
63 #include <sys/zfs_znode.h>
64 #include <sys/sa.h>
65 #include <sys/zfs_sa.h>
66 
67 #include "zfs_prop.h"
68 #include "zfs_comutil.h"
69 
70 /*
71  * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
72  * turned on when DEBUG is also defined.
73  */
74 #ifdef	DEBUG
75 #define	ZNODE_STATS
76 #endif	/* DEBUG */
77 
78 #ifdef	ZNODE_STATS
79 #define	ZNODE_STAT_ADD(stat)			((stat)++)
80 #else
81 #define	ZNODE_STAT_ADD(stat)			/* nothing */
82 #endif	/* ZNODE_STATS */
83 
84 /*
85  * Functions needed for userland (ie: libzpool) are not put under
86  * #ifdef_KERNEL; the rest of the functions have dependencies
87  * (such as VFS logic) that will not compile easily in userland.
88  */
89 #ifdef _KERNEL
90 /*
91  * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to
92  * be freed before it can be safely accessed.
93  */
94 krwlock_t zfsvfs_lock;
95 
96 static kmem_cache_t *znode_cache = NULL;
97 
98 /*ARGSUSED*/
99 static void
100 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
101 {
102 	/*
103 	 * We should never drop all dbuf refs without first clearing
104 	 * the eviction callback.
105 	 */
106 	panic("evicting znode %p\n", user_ptr);
107 }
108 
109 /*ARGSUSED*/
110 static int
111 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
112 {
113 	znode_t *zp = buf;
114 
115 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
116 
117 	zp->z_vnode = vn_alloc(kmflags);
118 	if (zp->z_vnode == NULL) {
119 		return (-1);
120 	}
121 	ZTOV(zp)->v_data = zp;
122 
123 	list_link_init(&zp->z_link_node);
124 
125 	mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
126 	rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
127 	rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
128 	mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
129 
130 	mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
131 	avl_create(&zp->z_range_avl, zfs_range_compare,
132 	    sizeof (rl_t), offsetof(rl_t, r_node));
133 
134 	zp->z_dirlocks = NULL;
135 	zp->z_acl_cached = NULL;
136 	zp->z_moved = 0;
137 	return (0);
138 }
139 
140 /*ARGSUSED*/
141 static void
142 zfs_znode_cache_destructor(void *buf, void *arg)
143 {
144 	znode_t *zp = buf;
145 
146 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
147 	ASSERT(ZTOV(zp)->v_data == zp);
148 	vn_free(ZTOV(zp));
149 	ASSERT(!list_link_active(&zp->z_link_node));
150 	mutex_destroy(&zp->z_lock);
151 	rw_destroy(&zp->z_parent_lock);
152 	rw_destroy(&zp->z_name_lock);
153 	mutex_destroy(&zp->z_acl_lock);
154 	avl_destroy(&zp->z_range_avl);
155 	mutex_destroy(&zp->z_range_lock);
156 
157 	ASSERT(zp->z_dirlocks == NULL);
158 	ASSERT(zp->z_acl_cached == NULL);
159 }
160 
161 #ifdef	ZNODE_STATS
162 static struct {
163 	uint64_t zms_zfsvfs_invalid;
164 	uint64_t zms_zfsvfs_recheck1;
165 	uint64_t zms_zfsvfs_unmounted;
166 	uint64_t zms_zfsvfs_recheck2;
167 	uint64_t zms_obj_held;
168 	uint64_t zms_vnode_locked;
169 	uint64_t zms_not_only_dnlc;
170 } znode_move_stats;
171 #endif	/* ZNODE_STATS */
172 
173 static void
174 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
175 {
176 	vnode_t *vp;
177 
178 	/* Copy fields. */
179 	nzp->z_zfsvfs = ozp->z_zfsvfs;
180 
181 	/* Swap vnodes. */
182 	vp = nzp->z_vnode;
183 	nzp->z_vnode = ozp->z_vnode;
184 	ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
185 	ZTOV(ozp)->v_data = ozp;
186 	ZTOV(nzp)->v_data = nzp;
187 
188 	nzp->z_id = ozp->z_id;
189 	ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
190 	ASSERT(avl_numnodes(&ozp->z_range_avl) == 0);
191 	nzp->z_unlinked = ozp->z_unlinked;
192 	nzp->z_atime_dirty = ozp->z_atime_dirty;
193 	nzp->z_zn_prefetch = ozp->z_zn_prefetch;
194 	nzp->z_blksz = ozp->z_blksz;
195 	nzp->z_seq = ozp->z_seq;
196 	nzp->z_mapcnt = ozp->z_mapcnt;
197 	nzp->z_gen = ozp->z_gen;
198 	nzp->z_sync_cnt = ozp->z_sync_cnt;
199 	nzp->z_is_sa = ozp->z_is_sa;
200 	nzp->z_sa_hdl = ozp->z_sa_hdl;
201 	bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
202 	nzp->z_links = ozp->z_links;
203 	nzp->z_size = ozp->z_size;
204 	nzp->z_pflags = ozp->z_pflags;
205 	nzp->z_uid = ozp->z_uid;
206 	nzp->z_gid = ozp->z_gid;
207 	nzp->z_mode = ozp->z_mode;
208 
209 	/*
210 	 * Since this is just an idle znode and kmem is already dealing with
211 	 * memory pressure, release any cached ACL.
212 	 */
213 	if (ozp->z_acl_cached) {
214 		zfs_acl_free(ozp->z_acl_cached);
215 		ozp->z_acl_cached = NULL;
216 	}
217 
218 	sa_set_userp(nzp->z_sa_hdl, nzp);
219 
220 	/*
221 	 * Invalidate the original znode by clearing fields that provide a
222 	 * pointer back to the znode. Set the low bit of the vfs pointer to
223 	 * ensure that zfs_znode_move() recognizes the znode as invalid in any
224 	 * subsequent callback.
225 	 */
226 	ozp->z_sa_hdl = NULL;
227 	POINTER_INVALIDATE(&ozp->z_zfsvfs);
228 
229 	/*
230 	 * Mark the znode.
231 	 */
232 	nzp->z_moved = 1;
233 	ozp->z_moved = (uint8_t)-1;
234 }
235 
236 /*ARGSUSED*/
237 static kmem_cbrc_t
238 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
239 {
240 	znode_t *ozp = buf, *nzp = newbuf;
241 	zfsvfs_t *zfsvfs;
242 	vnode_t *vp;
243 
244 	/*
245 	 * The znode is on the file system's list of known znodes if the vfs
246 	 * pointer is valid. We set the low bit of the vfs pointer when freeing
247 	 * the znode to invalidate it, and the memory patterns written by kmem
248 	 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
249 	 * created znode sets the vfs pointer last of all to indicate that the
250 	 * znode is known and in a valid state to be moved by this function.
251 	 */
252 	zfsvfs = ozp->z_zfsvfs;
253 	if (!POINTER_IS_VALID(zfsvfs)) {
254 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
255 		return (KMEM_CBRC_DONT_KNOW);
256 	}
257 
258 	/*
259 	 * Close a small window in which it's possible that the filesystem could
260 	 * be unmounted and freed, and zfsvfs, though valid in the previous
261 	 * statement, could point to unrelated memory by the time we try to
262 	 * prevent the filesystem from being unmounted.
263 	 */
264 	rw_enter(&zfsvfs_lock, RW_WRITER);
265 	if (zfsvfs != ozp->z_zfsvfs) {
266 		rw_exit(&zfsvfs_lock);
267 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
268 		return (KMEM_CBRC_DONT_KNOW);
269 	}
270 
271 	/*
272 	 * If the znode is still valid, then so is the file system. We know that
273 	 * no valid file system can be freed while we hold zfsvfs_lock, so we
274 	 * can safely ensure that the filesystem is not and will not be
275 	 * unmounted. The next statement is equivalent to ZFS_ENTER().
276 	 */
277 	rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
278 	if (zfsvfs->z_unmounted) {
279 		ZFS_EXIT(zfsvfs);
280 		rw_exit(&zfsvfs_lock);
281 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
282 		return (KMEM_CBRC_DONT_KNOW);
283 	}
284 	rw_exit(&zfsvfs_lock);
285 
286 	mutex_enter(&zfsvfs->z_znodes_lock);
287 	/*
288 	 * Recheck the vfs pointer in case the znode was removed just before
289 	 * acquiring the lock.
290 	 */
291 	if (zfsvfs != ozp->z_zfsvfs) {
292 		mutex_exit(&zfsvfs->z_znodes_lock);
293 		ZFS_EXIT(zfsvfs);
294 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
295 		return (KMEM_CBRC_DONT_KNOW);
296 	}
297 
298 	/*
299 	 * At this point we know that as long as we hold z_znodes_lock, the
300 	 * znode cannot be freed and fields within the znode can be safely
301 	 * accessed. Now, prevent a race with zfs_zget().
302 	 */
303 	if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
304 		mutex_exit(&zfsvfs->z_znodes_lock);
305 		ZFS_EXIT(zfsvfs);
306 		ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
307 		return (KMEM_CBRC_LATER);
308 	}
309 
310 	vp = ZTOV(ozp);
311 	if (mutex_tryenter(&vp->v_lock) == 0) {
312 		ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
313 		mutex_exit(&zfsvfs->z_znodes_lock);
314 		ZFS_EXIT(zfsvfs);
315 		ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
316 		return (KMEM_CBRC_LATER);
317 	}
318 
319 	/* Only move znodes that are referenced _only_ by the DNLC. */
320 	if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
321 		mutex_exit(&vp->v_lock);
322 		ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
323 		mutex_exit(&zfsvfs->z_znodes_lock);
324 		ZFS_EXIT(zfsvfs);
325 		ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
326 		return (KMEM_CBRC_LATER);
327 	}
328 
329 	/*
330 	 * The znode is known and in a valid state to move. We're holding the
331 	 * locks needed to execute the critical section.
332 	 */
333 	zfs_znode_move_impl(ozp, nzp);
334 	mutex_exit(&vp->v_lock);
335 	ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
336 
337 	list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
338 	mutex_exit(&zfsvfs->z_znodes_lock);
339 	ZFS_EXIT(zfsvfs);
340 
341 	return (KMEM_CBRC_YES);
342 }
343 
344 void
345 zfs_znode_init(void)
346 {
347 	/*
348 	 * Initialize zcache
349 	 */
350 	rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
351 	ASSERT(znode_cache == NULL);
352 	znode_cache = kmem_cache_create("zfs_znode_cache",
353 	    sizeof (znode_t), 0, zfs_znode_cache_constructor,
354 	    zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
355 	kmem_cache_set_move(znode_cache, zfs_znode_move);
356 }
357 
358 void
359 zfs_znode_fini(void)
360 {
361 	/*
362 	 * Cleanup vfs & vnode ops
363 	 */
364 	zfs_remove_op_tables();
365 
366 	/*
367 	 * Cleanup zcache
368 	 */
369 	if (znode_cache)
370 		kmem_cache_destroy(znode_cache);
371 	znode_cache = NULL;
372 	rw_destroy(&zfsvfs_lock);
373 }
374 
375 struct vnodeops *zfs_dvnodeops;
376 struct vnodeops *zfs_fvnodeops;
377 struct vnodeops *zfs_symvnodeops;
378 struct vnodeops *zfs_xdvnodeops;
379 struct vnodeops *zfs_evnodeops;
380 struct vnodeops *zfs_sharevnodeops;
381 
382 void
383 zfs_remove_op_tables()
384 {
385 	/*
386 	 * Remove vfs ops
387 	 */
388 	ASSERT(zfsfstype);
389 	(void) vfs_freevfsops_by_type(zfsfstype);
390 	zfsfstype = 0;
391 
392 	/*
393 	 * Remove vnode ops
394 	 */
395 	if (zfs_dvnodeops)
396 		vn_freevnodeops(zfs_dvnodeops);
397 	if (zfs_fvnodeops)
398 		vn_freevnodeops(zfs_fvnodeops);
399 	if (zfs_symvnodeops)
400 		vn_freevnodeops(zfs_symvnodeops);
401 	if (zfs_xdvnodeops)
402 		vn_freevnodeops(zfs_xdvnodeops);
403 	if (zfs_evnodeops)
404 		vn_freevnodeops(zfs_evnodeops);
405 	if (zfs_sharevnodeops)
406 		vn_freevnodeops(zfs_sharevnodeops);
407 
408 	zfs_dvnodeops = NULL;
409 	zfs_fvnodeops = NULL;
410 	zfs_symvnodeops = NULL;
411 	zfs_xdvnodeops = NULL;
412 	zfs_evnodeops = NULL;
413 	zfs_sharevnodeops = NULL;
414 }
415 
416 extern const fs_operation_def_t zfs_dvnodeops_template[];
417 extern const fs_operation_def_t zfs_fvnodeops_template[];
418 extern const fs_operation_def_t zfs_xdvnodeops_template[];
419 extern const fs_operation_def_t zfs_symvnodeops_template[];
420 extern const fs_operation_def_t zfs_evnodeops_template[];
421 extern const fs_operation_def_t zfs_sharevnodeops_template[];
422 
423 int
424 zfs_create_op_tables()
425 {
426 	int error;
427 
428 	/*
429 	 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
430 	 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
431 	 * In this case we just return as the ops vectors are already set up.
432 	 */
433 	if (zfs_dvnodeops)
434 		return (0);
435 
436 	error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
437 	    &zfs_dvnodeops);
438 	if (error)
439 		return (error);
440 
441 	error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
442 	    &zfs_fvnodeops);
443 	if (error)
444 		return (error);
445 
446 	error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
447 	    &zfs_symvnodeops);
448 	if (error)
449 		return (error);
450 
451 	error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
452 	    &zfs_xdvnodeops);
453 	if (error)
454 		return (error);
455 
456 	error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
457 	    &zfs_evnodeops);
458 	if (error)
459 		return (error);
460 
461 	error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
462 	    &zfs_sharevnodeops);
463 
464 	return (error);
465 }
466 
467 int
468 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
469 {
470 	zfs_acl_ids_t acl_ids;
471 	vattr_t vattr;
472 	znode_t *sharezp;
473 	vnode_t *vp;
474 	znode_t *zp;
475 	int error;
476 
477 	vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
478 	vattr.va_type = VDIR;
479 	vattr.va_mode = S_IFDIR|0555;
480 	vattr.va_uid = crgetuid(kcred);
481 	vattr.va_gid = crgetgid(kcred);
482 
483 	sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
484 	ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
485 	sharezp->z_moved = 0;
486 	sharezp->z_unlinked = 0;
487 	sharezp->z_atime_dirty = 0;
488 	sharezp->z_zfsvfs = zfsvfs;
489 	sharezp->z_is_sa = zfsvfs->z_use_sa;
490 
491 	vp = ZTOV(sharezp);
492 	vn_reinit(vp);
493 	vp->v_type = VDIR;
494 
495 	VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
496 	    kcred, NULL, &acl_ids));
497 	zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
498 	ASSERT3P(zp, ==, sharezp);
499 	ASSERT(!vn_in_dnlc(ZTOV(sharezp))); /* not valid to move */
500 	POINTER_INVALIDATE(&sharezp->z_zfsvfs);
501 	error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
502 	    ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
503 	zfsvfs->z_shares_dir = sharezp->z_id;
504 
505 	zfs_acl_ids_free(&acl_ids);
506 	ZTOV(sharezp)->v_count = 0;
507 	sa_handle_destroy(sharezp->z_sa_hdl);
508 	kmem_cache_free(znode_cache, sharezp);
509 
510 	return (error);
511 }
512 
513 /*
514  * define a couple of values we need available
515  * for both 64 and 32 bit environments.
516  */
517 #ifndef NBITSMINOR64
518 #define	NBITSMINOR64	32
519 #endif
520 #ifndef MAXMAJ64
521 #define	MAXMAJ64	0xffffffffUL
522 #endif
523 #ifndef	MAXMIN64
524 #define	MAXMIN64	0xffffffffUL
525 #endif
526 
527 /*
528  * Create special expldev for ZFS private use.
529  * Can't use standard expldev since it doesn't do
530  * what we want.  The standard expldev() takes a
531  * dev32_t in LP64 and expands it to a long dev_t.
532  * We need an interface that takes a dev32_t in ILP32
533  * and expands it to a long dev_t.
534  */
535 static uint64_t
536 zfs_expldev(dev_t dev)
537 {
538 #ifndef _LP64
539 	major_t major = (major_t)dev >> NBITSMINOR32 & MAXMAJ32;
540 	return (((uint64_t)major << NBITSMINOR64) |
541 	    ((minor_t)dev & MAXMIN32));
542 #else
543 	return (dev);
544 #endif
545 }
546 
547 /*
548  * Special cmpldev for ZFS private use.
549  * Can't use standard cmpldev since it takes
550  * a long dev_t and compresses it to dev32_t in
551  * LP64.  We need to do a compaction of a long dev_t
552  * to a dev32_t in ILP32.
553  */
554 dev_t
555 zfs_cmpldev(uint64_t dev)
556 {
557 #ifndef _LP64
558 	minor_t minor = (minor_t)dev & MAXMIN64;
559 	major_t major = (major_t)(dev >> NBITSMINOR64) & MAXMAJ64;
560 
561 	if (major > MAXMAJ32 || minor > MAXMIN32)
562 		return (NODEV32);
563 
564 	return (((dev32_t)major << NBITSMINOR32) | minor);
565 #else
566 	return (dev);
567 #endif
568 }
569 
570 static void
571 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
572     dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
573 {
574 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
575 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
576 
577 	mutex_enter(&zp->z_lock);
578 
579 	ASSERT(zp->z_sa_hdl == NULL);
580 	ASSERT(zp->z_acl_cached == NULL);
581 	if (sa_hdl == NULL) {
582 		VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
583 		    SA_HDL_SHARED, &zp->z_sa_hdl));
584 	} else {
585 		zp->z_sa_hdl = sa_hdl;
586 		sa_set_userp(sa_hdl, zp);
587 	}
588 
589 	zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
590 
591 	/*
592 	 * Slap on VROOT if we are the root znode
593 	 */
594 	if (zp->z_id == zfsvfs->z_root)
595 		ZTOV(zp)->v_flag |= VROOT;
596 
597 	mutex_exit(&zp->z_lock);
598 	vn_exists(ZTOV(zp));
599 }
600 
601 void
602 zfs_znode_dmu_fini(znode_t *zp)
603 {
604 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
605 	    zp->z_unlinked ||
606 	    RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
607 
608 	sa_handle_destroy(zp->z_sa_hdl);
609 	zp->z_sa_hdl = NULL;
610 }
611 
612 /*
613  * Construct a new znode/vnode and intialize.
614  *
615  * This does not do a call to dmu_set_user() that is
616  * up to the caller to do, in case you don't want to
617  * return the znode
618  */
619 static znode_t *
620 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
621     dmu_object_type_t obj_type, sa_handle_t *hdl)
622 {
623 	znode_t	*zp;
624 	vnode_t *vp;
625 	uint64_t mode;
626 	uint64_t parent;
627 	uint64_t uid, gid;
628 	sa_bulk_attr_t bulk[9];
629 	int count = 0;
630 
631 	zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
632 
633 	ASSERT(zp->z_dirlocks == NULL);
634 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
635 	zp->z_moved = 0;
636 
637 	/*
638 	 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
639 	 * the zfs_znode_move() callback.
640 	 */
641 	zp->z_sa_hdl = NULL;
642 	zp->z_unlinked = 0;
643 	zp->z_atime_dirty = 0;
644 	zp->z_mapcnt = 0;
645 	zp->z_id = db->db_object;
646 	zp->z_blksz = blksz;
647 	zp->z_seq = 0x7A4653;
648 	zp->z_sync_cnt = 0;
649 
650 	vp = ZTOV(zp);
651 	vn_reinit(vp);
652 
653 	zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
654 
655 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
656 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
657 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
658 	    &zp->z_size, 8);
659 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
660 	    &zp->z_links, 8);
661 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
662 	    &zp->z_pflags, 8);
663 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
664 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
665 	    &zp->z_atime, 16);
666 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
667 	    &uid, 8);
668 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
669 	    &gid, 8);
670 
671 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) {
672 		if (hdl == NULL)
673 			sa_handle_destroy(zp->z_sa_hdl);
674 		kmem_cache_free(znode_cache, zp);
675 		return (NULL);
676 	}
677 
678 	zp->z_uid = zfs_fuid_map_id(zfsvfs, uid, CRED(), ZFS_OWNER);
679 	zp->z_gid = zfs_fuid_map_id(zfsvfs, gid, CRED(), ZFS_GROUP);
680 	zp->z_mode = mode;
681 	vp->v_vfsp = zfsvfs->z_parent->z_vfs;
682 
683 	vp->v_type = IFTOVT((mode_t)mode);
684 
685 	switch (vp->v_type) {
686 	case VDIR:
687 		if (zp->z_pflags & ZFS_XATTR) {
688 			vn_setops(vp, zfs_xdvnodeops);
689 			vp->v_flag |= V_XATTRDIR;
690 		} else {
691 			vn_setops(vp, zfs_dvnodeops);
692 		}
693 		zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
694 		break;
695 	case VBLK:
696 	case VCHR:
697 		{
698 			uint64_t rdev;
699 			VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
700 			    &rdev, sizeof (rdev)) == 0);
701 
702 			vp->v_rdev = zfs_cmpldev(rdev);
703 		}
704 		/*FALLTHROUGH*/
705 	case VFIFO:
706 	case VSOCK:
707 	case VDOOR:
708 		vn_setops(vp, zfs_fvnodeops);
709 		break;
710 	case VREG:
711 		vp->v_flag |= VMODSORT;
712 		if (parent == zfsvfs->z_shares_dir) {
713 			ASSERT(uid == 0 && gid == 0);
714 			vn_setops(vp, zfs_sharevnodeops);
715 		} else {
716 			vn_setops(vp, zfs_fvnodeops);
717 		}
718 		break;
719 	case VLNK:
720 		vn_setops(vp, zfs_symvnodeops);
721 		break;
722 	default:
723 		vn_setops(vp, zfs_evnodeops);
724 		break;
725 	}
726 
727 	mutex_enter(&zfsvfs->z_znodes_lock);
728 	list_insert_tail(&zfsvfs->z_all_znodes, zp);
729 	membar_producer();
730 	/*
731 	 * Everything else must be valid before assigning z_zfsvfs makes the
732 	 * znode eligible for zfs_znode_move().
733 	 */
734 	zp->z_zfsvfs = zfsvfs;
735 	mutex_exit(&zfsvfs->z_znodes_lock);
736 
737 	VFS_HOLD(zfsvfs->z_vfs);
738 	return (zp);
739 }
740 
741 static uint64_t empty_xattr;
742 static uint64_t pad[4];
743 static zfs_acl_phys_t acl_phys;
744 /*
745  * Create a new DMU object to hold a zfs znode.
746  *
747  *	IN:	dzp	- parent directory for new znode
748  *		vap	- file attributes for new znode
749  *		tx	- dmu transaction id for zap operations
750  *		cr	- credentials of caller
751  *		flag	- flags:
752  *			  IS_ROOT_NODE	- new object will be root
753  *			  IS_XATTR	- new object is an attribute
754  *		bonuslen - length of bonus buffer
755  *		setaclp  - File/Dir initial ACL
756  *		fuidp	 - Tracks fuid allocation.
757  *
758  *	OUT:	zpp	- allocated znode
759  *
760  */
761 void
762 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
763     uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
764 {
765 	uint64_t	crtime[2], atime[2], mtime[2], ctime[2];
766 	uint64_t	mode, size, links, parent, pflags;
767 	uint64_t	dzp_pflags = 0;
768 	uint64_t	rdev = 0;
769 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
770 	dmu_buf_t	*db;
771 	timestruc_t	now;
772 	uint64_t	gen, obj;
773 	int		err;
774 	int		bonuslen;
775 	sa_handle_t	*sa_hdl;
776 	dmu_object_type_t obj_type;
777 	sa_bulk_attr_t	sa_attrs[ZPL_END];
778 	int		cnt = 0;
779 	zfs_acl_locator_cb_t locate = { 0 };
780 
781 	ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
782 
783 	if (zfsvfs->z_replay) {
784 		obj = vap->va_nodeid;
785 		now = vap->va_ctime;		/* see zfs_replay_create() */
786 		gen = vap->va_nblocks;		/* ditto */
787 	} else {
788 		obj = 0;
789 		gethrestime(&now);
790 		gen = dmu_tx_get_txg(tx);
791 	}
792 
793 	obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
794 	bonuslen = (obj_type == DMU_OT_SA) ?
795 	    DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
796 
797 	/*
798 	 * Create a new DMU object.
799 	 */
800 	/*
801 	 * There's currently no mechanism for pre-reading the blocks that will
802 	 * be needed to allocate a new object, so we accept the small chance
803 	 * that there will be an i/o error and we will fail one of the
804 	 * assertions below.
805 	 */
806 	if (vap->va_type == VDIR) {
807 		if (zfsvfs->z_replay) {
808 			err = zap_create_claim_norm(zfsvfs->z_os, obj,
809 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
810 			    obj_type, bonuslen, tx);
811 			ASSERT3U(err, ==, 0);
812 		} else {
813 			obj = zap_create_norm(zfsvfs->z_os,
814 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
815 			    obj_type, bonuslen, tx);
816 		}
817 	} else {
818 		if (zfsvfs->z_replay) {
819 			err = dmu_object_claim(zfsvfs->z_os, obj,
820 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
821 			    obj_type, bonuslen, tx);
822 			ASSERT3U(err, ==, 0);
823 		} else {
824 			obj = dmu_object_alloc(zfsvfs->z_os,
825 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
826 			    obj_type, bonuslen, tx);
827 		}
828 	}
829 
830 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
831 	VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
832 
833 	/*
834 	 * If this is the root, fix up the half-initialized parent pointer
835 	 * to reference the just-allocated physical data area.
836 	 */
837 	if (flag & IS_ROOT_NODE) {
838 		dzp->z_id = obj;
839 	} else {
840 		dzp_pflags = dzp->z_pflags;
841 	}
842 
843 	/*
844 	 * If parent is an xattr, so am I.
845 	 */
846 	if (dzp_pflags & ZFS_XATTR) {
847 		flag |= IS_XATTR;
848 	}
849 
850 	if (zfsvfs->z_use_fuids)
851 		pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
852 	else
853 		pflags = 0;
854 
855 	if (vap->va_type == VDIR) {
856 		size = 2;		/* contents ("." and "..") */
857 		links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
858 	} else {
859 		size = links = 0;
860 	}
861 
862 	if (vap->va_type == VBLK || vap->va_type == VCHR) {
863 		rdev = zfs_expldev(vap->va_rdev);
864 	}
865 
866 	parent = dzp->z_id;
867 	mode = acl_ids->z_mode;
868 	if (flag & IS_XATTR)
869 		pflags |= ZFS_XATTR;
870 
871 	/*
872 	 * No execs denied will be deterimed when zfs_mode_compute() is called.
873 	 */
874 	pflags |= acl_ids->z_aclp->z_hints &
875 	    (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
876 	    ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
877 
878 	ZFS_TIME_ENCODE(&now, crtime);
879 	ZFS_TIME_ENCODE(&now, ctime);
880 
881 	if (vap->va_mask & AT_ATIME) {
882 		ZFS_TIME_ENCODE(&vap->va_atime, atime);
883 	} else {
884 		ZFS_TIME_ENCODE(&now, atime);
885 	}
886 
887 	if (vap->va_mask & AT_MTIME) {
888 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
889 	} else {
890 		ZFS_TIME_ENCODE(&now, mtime);
891 	}
892 
893 	/* Now add in all of the "SA" attributes */
894 	VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
895 	    &sa_hdl));
896 
897 	/*
898 	 * Setup the array of attributes to be replaced/set on the new file
899 	 *
900 	 * order for  DMU_OT_ZNODE is critical since it needs to be constructed
901 	 * in the old znode_phys_t format.  Don't change this ordering
902 	 */
903 
904 	if (obj_type == DMU_OT_ZNODE) {
905 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
906 		    NULL, &atime, 16);
907 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
908 		    NULL, &mtime, 16);
909 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
910 		    NULL, &ctime, 16);
911 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
912 		    NULL, &crtime, 16);
913 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
914 		    NULL, &gen, 8);
915 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
916 		    NULL, &mode, 8);
917 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
918 		    NULL, &size, 8);
919 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
920 		    NULL, &parent, 8);
921 	} else {
922 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
923 		    NULL, &mode, 8);
924 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
925 		    NULL, &size, 8);
926 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
927 		    NULL, &gen, 8);
928 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
929 		    &acl_ids->z_fuid, 8);
930 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
931 		    &acl_ids->z_fgid, 8);
932 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
933 		    NULL, &parent, 8);
934 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
935 		    NULL, &pflags, 8);
936 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
937 		    NULL, &atime, 16);
938 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
939 		    NULL, &mtime, 16);
940 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
941 		    NULL, &ctime, 16);
942 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
943 		    NULL, &crtime, 16);
944 	}
945 
946 	SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
947 
948 	if (obj_type == DMU_OT_ZNODE) {
949 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
950 		    &empty_xattr, 8);
951 	}
952 	if (obj_type == DMU_OT_ZNODE ||
953 	    (vap->va_type == VBLK || vap->va_type == VCHR)) {
954 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
955 		    NULL, &rdev, 8);
956 
957 	}
958 	if (obj_type == DMU_OT_ZNODE) {
959 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
960 		    NULL, &pflags, 8);
961 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
962 		    &acl_ids->z_fuid, 8);
963 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
964 		    &acl_ids->z_fgid, 8);
965 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
966 		    sizeof (uint64_t) * 4);
967 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
968 		    &acl_phys, sizeof (zfs_acl_phys_t));
969 	} else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
970 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
971 		    &acl_ids->z_aclp->z_acl_count, 8);
972 		locate.cb_aclp = acl_ids->z_aclp;
973 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
974 		    zfs_acl_data_locator, &locate,
975 		    acl_ids->z_aclp->z_acl_bytes);
976 		mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
977 		    acl_ids->z_fuid, acl_ids->z_fgid);
978 	}
979 
980 	VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
981 
982 	if (!(flag & IS_ROOT_NODE)) {
983 		*zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
984 		ASSERT(*zpp != NULL);
985 	} else {
986 		/*
987 		 * If we are creating the root node, the "parent" we
988 		 * passed in is the znode for the root.
989 		 */
990 		*zpp = dzp;
991 
992 		(*zpp)->z_sa_hdl = sa_hdl;
993 	}
994 
995 	(*zpp)->z_pflags = pflags;
996 	(*zpp)->z_mode = mode;
997 
998 	if (vap->va_mask & AT_XVATTR)
999 		zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
1000 
1001 	if (obj_type == DMU_OT_ZNODE ||
1002 	    acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1003 		err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx);
1004 		ASSERT3P(err, ==, 0);
1005 	}
1006 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1007 }
1008 
1009 /*
1010  * zfs_xvattr_set only updates the in-core attributes
1011  * it is assumed the caller will be doing an sa_bulk_update
1012  * to push the changes out
1013  */
1014 void
1015 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1016 {
1017 	xoptattr_t *xoap;
1018 
1019 	xoap = xva_getxoptattr(xvap);
1020 	ASSERT(xoap);
1021 
1022 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1023 		uint64_t times[2];
1024 		ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1025 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1026 		    &times, sizeof (times), tx);
1027 		XVA_SET_RTN(xvap, XAT_CREATETIME);
1028 	}
1029 	if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1030 		ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1031 		    zp->z_pflags, tx);
1032 		XVA_SET_RTN(xvap, XAT_READONLY);
1033 	}
1034 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1035 		ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1036 		    zp->z_pflags, tx);
1037 		XVA_SET_RTN(xvap, XAT_HIDDEN);
1038 	}
1039 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1040 		ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1041 		    zp->z_pflags, tx);
1042 		XVA_SET_RTN(xvap, XAT_SYSTEM);
1043 	}
1044 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1045 		ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1046 		    zp->z_pflags, tx);
1047 		XVA_SET_RTN(xvap, XAT_ARCHIVE);
1048 	}
1049 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1050 		ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1051 		    zp->z_pflags, tx);
1052 		XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1053 	}
1054 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1055 		ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1056 		    zp->z_pflags, tx);
1057 		XVA_SET_RTN(xvap, XAT_NOUNLINK);
1058 	}
1059 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1060 		ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1061 		    zp->z_pflags, tx);
1062 		XVA_SET_RTN(xvap, XAT_APPENDONLY);
1063 	}
1064 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1065 		ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1066 		    zp->z_pflags, tx);
1067 		XVA_SET_RTN(xvap, XAT_NODUMP);
1068 	}
1069 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1070 		ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1071 		    zp->z_pflags, tx);
1072 		XVA_SET_RTN(xvap, XAT_OPAQUE);
1073 	}
1074 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1075 		ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1076 		    xoap->xoa_av_quarantined, zp->z_pflags, tx);
1077 		XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1078 	}
1079 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1080 		ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1081 		    zp->z_pflags, tx);
1082 		XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1083 	}
1084 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1085 		zfs_sa_set_scanstamp(zp, xvap, tx);
1086 		XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1087 	}
1088 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1089 		ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1090 		    zp->z_pflags, tx);
1091 		XVA_SET_RTN(xvap, XAT_REPARSE);
1092 	}
1093 }
1094 
1095 int
1096 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1097 {
1098 	dmu_object_info_t doi;
1099 	dmu_buf_t	*db;
1100 	znode_t		*zp;
1101 	int err;
1102 	sa_handle_t	*hdl;
1103 
1104 	*zpp = NULL;
1105 
1106 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1107 
1108 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1109 	if (err) {
1110 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1111 		return (err);
1112 	}
1113 
1114 	dmu_object_info_from_db(db, &doi);
1115 	if (doi.doi_bonus_type != DMU_OT_SA &&
1116 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1117 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1118 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1119 		sa_buf_rele(db, NULL);
1120 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1121 		return (EINVAL);
1122 	}
1123 
1124 	hdl = dmu_buf_get_user(db);
1125 	if (hdl != NULL) {
1126 		zp  = sa_get_userdata(hdl);
1127 
1128 
1129 		/*
1130 		 * Since "SA" does immediate eviction we
1131 		 * should never find a sa handle that doesn't
1132 		 * know about the znode.
1133 		 */
1134 
1135 		ASSERT3P(zp, !=, NULL);
1136 
1137 		mutex_enter(&zp->z_lock);
1138 		ASSERT3U(zp->z_id, ==, obj_num);
1139 		if (zp->z_unlinked) {
1140 			err = ENOENT;
1141 		} else {
1142 			VN_HOLD(ZTOV(zp));
1143 			*zpp = zp;
1144 			err = 0;
1145 		}
1146 		sa_buf_rele(db, NULL);
1147 		mutex_exit(&zp->z_lock);
1148 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1149 		return (err);
1150 	}
1151 
1152 	/*
1153 	 * Not found create new znode/vnode
1154 	 * but only if file exists.
1155 	 *
1156 	 * There is a small window where zfs_vget() could
1157 	 * find this object while a file create is still in
1158 	 * progress.  This is checked for in zfs_znode_alloc()
1159 	 *
1160 	 * if zfs_znode_alloc() fails it will drop the hold on the
1161 	 * bonus buffer.
1162 	 */
1163 	zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1164 	    doi.doi_bonus_type, NULL);
1165 	if (zp == NULL) {
1166 		err = ENOENT;
1167 	} else {
1168 		*zpp = zp;
1169 	}
1170 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1171 	return (err);
1172 }
1173 
1174 int
1175 zfs_rezget(znode_t *zp)
1176 {
1177 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1178 	dmu_object_info_t doi;
1179 	dmu_buf_t *db;
1180 	uint64_t obj_num = zp->z_id;
1181 	uint64_t mode;
1182 	uint64_t uid, gid;
1183 	sa_bulk_attr_t bulk[8];
1184 	int err;
1185 	int count = 0;
1186 	uint64_t gen;
1187 
1188 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1189 
1190 	mutex_enter(&zp->z_acl_lock);
1191 	if (zp->z_acl_cached) {
1192 		zfs_acl_free(zp->z_acl_cached);
1193 		zp->z_acl_cached = NULL;
1194 	}
1195 
1196 	mutex_exit(&zp->z_acl_lock);
1197 	ASSERT(zp->z_sa_hdl == NULL);
1198 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1199 	if (err) {
1200 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1201 		return (err);
1202 	}
1203 
1204 	dmu_object_info_from_db(db, &doi);
1205 	if (doi.doi_bonus_type != DMU_OT_SA &&
1206 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1207 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1208 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1209 		sa_buf_rele(db, NULL);
1210 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1211 		return (EINVAL);
1212 	}
1213 
1214 	zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1215 
1216 	/* reload cached values */
1217 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1218 	    &gen, sizeof (gen));
1219 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1220 	    &zp->z_size, sizeof (zp->z_size));
1221 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1222 	    &zp->z_links, sizeof (zp->z_links));
1223 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1224 	    &zp->z_pflags, sizeof (zp->z_pflags));
1225 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1226 	    &zp->z_atime, sizeof (zp->z_atime));
1227 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1228 	    &uid, sizeof (uid));
1229 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1230 	    &gid, sizeof (gid));
1231 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1232 	    &mode, sizeof (mode));
1233 
1234 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1235 		zfs_znode_dmu_fini(zp);
1236 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1237 		return (EIO);
1238 	}
1239 
1240 	zp->z_mode = mode;
1241 
1242 	if (gen != zp->z_gen) {
1243 		zfs_znode_dmu_fini(zp);
1244 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1245 		return (EIO);
1246 	}
1247 
1248 	zp->z_uid = zfs_fuid_map_id(zfsvfs, uid, CRED(), ZFS_OWNER);
1249 	zp->z_gid = zfs_fuid_map_id(zfsvfs, gid, CRED(), ZFS_GROUP);
1250 	zp->z_unlinked = (zp->z_links == 0);
1251 	zp->z_blksz = doi.doi_data_block_size;
1252 
1253 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1254 
1255 	return (0);
1256 }
1257 
1258 void
1259 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1260 {
1261 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1262 	objset_t *os = zfsvfs->z_os;
1263 	uint64_t obj = zp->z_id;
1264 	uint64_t acl_obj = zfs_external_acl(zp);
1265 
1266 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1267 	if (acl_obj) {
1268 		VERIFY(!zp->z_is_sa);
1269 		VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1270 	}
1271 	VERIFY(0 == dmu_object_free(os, obj, tx));
1272 	zfs_znode_dmu_fini(zp);
1273 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1274 	zfs_znode_free(zp);
1275 }
1276 
1277 void
1278 zfs_zinactive(znode_t *zp)
1279 {
1280 	vnode_t	*vp = ZTOV(zp);
1281 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1282 	uint64_t z_id = zp->z_id;
1283 
1284 	ASSERT(zp->z_sa_hdl);
1285 
1286 	/*
1287 	 * Don't allow a zfs_zget() while were trying to release this znode
1288 	 */
1289 	ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1290 
1291 	mutex_enter(&zp->z_lock);
1292 	mutex_enter(&vp->v_lock);
1293 	vp->v_count--;
1294 	if (vp->v_count > 0 || vn_has_cached_data(vp)) {
1295 		/*
1296 		 * If the hold count is greater than zero, somebody has
1297 		 * obtained a new reference on this znode while we were
1298 		 * processing it here, so we are done.  If we still have
1299 		 * mapped pages then we are also done, since we don't
1300 		 * want to inactivate the znode until the pages get pushed.
1301 		 *
1302 		 * XXX - if vn_has_cached_data(vp) is true, but count == 0,
1303 		 * this seems like it would leave the znode hanging with
1304 		 * no chance to go inactive...
1305 		 */
1306 		mutex_exit(&vp->v_lock);
1307 		mutex_exit(&zp->z_lock);
1308 		ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1309 		return;
1310 	}
1311 	mutex_exit(&vp->v_lock);
1312 
1313 	/*
1314 	 * If this was the last reference to a file with no links,
1315 	 * remove the file from the file system.
1316 	 */
1317 	if (zp->z_unlinked) {
1318 		mutex_exit(&zp->z_lock);
1319 		ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1320 		zfs_rmnode(zp);
1321 		return;
1322 	}
1323 
1324 	mutex_exit(&zp->z_lock);
1325 	zfs_znode_dmu_fini(zp);
1326 	ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1327 	zfs_znode_free(zp);
1328 }
1329 
1330 void
1331 zfs_znode_free(znode_t *zp)
1332 {
1333 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1334 
1335 	vn_invalid(ZTOV(zp));
1336 
1337 	ASSERT(ZTOV(zp)->v_count == 0);
1338 
1339 	mutex_enter(&zfsvfs->z_znodes_lock);
1340 	POINTER_INVALIDATE(&zp->z_zfsvfs);
1341 	list_remove(&zfsvfs->z_all_znodes, zp);
1342 	mutex_exit(&zfsvfs->z_znodes_lock);
1343 
1344 	if (zp->z_acl_cached) {
1345 		zfs_acl_free(zp->z_acl_cached);
1346 		zp->z_acl_cached = NULL;
1347 	}
1348 
1349 	kmem_cache_free(znode_cache, zp);
1350 
1351 	VFS_RELE(zfsvfs->z_vfs);
1352 }
1353 
1354 void
1355 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1356     uint64_t ctime[2], boolean_t have_tx)
1357 {
1358 	timestruc_t	now;
1359 
1360 	gethrestime(&now);
1361 
1362 	if (have_tx) {	/* will sa_bulk_update happen really soon? */
1363 		zp->z_atime_dirty = 0;
1364 		zp->z_seq++;
1365 	} else {
1366 		zp->z_atime_dirty = 1;
1367 	}
1368 
1369 	if (flag & AT_ATIME) {
1370 		ZFS_TIME_ENCODE(&now, zp->z_atime);
1371 	}
1372 
1373 	if (flag & AT_MTIME) {
1374 		ZFS_TIME_ENCODE(&now, mtime);
1375 		if (zp->z_zfsvfs->z_use_fuids) {
1376 			zp->z_pflags |= (ZFS_ARCHIVE |
1377 			    ZFS_AV_MODIFIED);
1378 		}
1379 	}
1380 
1381 	if (flag & AT_CTIME) {
1382 		ZFS_TIME_ENCODE(&now, ctime);
1383 		if (zp->z_zfsvfs->z_use_fuids)
1384 			zp->z_pflags |= ZFS_ARCHIVE;
1385 	}
1386 }
1387 
1388 /*
1389  * Grow the block size for a file.
1390  *
1391  *	IN:	zp	- znode of file to free data in.
1392  *		size	- requested block size
1393  *		tx	- open transaction.
1394  *
1395  * NOTE: this function assumes that the znode is write locked.
1396  */
1397 void
1398 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1399 {
1400 	int		error;
1401 	u_longlong_t	dummy;
1402 
1403 	if (size <= zp->z_blksz)
1404 		return;
1405 	/*
1406 	 * If the file size is already greater than the current blocksize,
1407 	 * we will not grow.  If there is more than one block in a file,
1408 	 * the blocksize cannot change.
1409 	 */
1410 	if (zp->z_blksz && zp->z_size > zp->z_blksz)
1411 		return;
1412 
1413 	error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1414 	    size, 0, tx);
1415 
1416 	if (error == ENOTSUP)
1417 		return;
1418 	ASSERT3U(error, ==, 0);
1419 
1420 	/* What blocksize did we actually get? */
1421 	dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1422 }
1423 
1424 /*
1425  * This is a dummy interface used when pvn_vplist_dirty() should *not*
1426  * be calling back into the fs for a putpage().  E.g.: when truncating
1427  * a file, the pages being "thrown away* don't need to be written out.
1428  */
1429 /* ARGSUSED */
1430 static int
1431 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1432     int flags, cred_t *cr)
1433 {
1434 	ASSERT(0);
1435 	return (0);
1436 }
1437 
1438 /*
1439  * Increase the file length
1440  *
1441  *	IN:	zp	- znode of file to free data in.
1442  *		end	- new end-of-file
1443  *
1444  * 	RETURN:	0 if success
1445  *		error code if failure
1446  */
1447 static int
1448 zfs_extend(znode_t *zp, uint64_t end)
1449 {
1450 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1451 	dmu_tx_t *tx;
1452 	rl_t *rl;
1453 	uint64_t newblksz;
1454 	int error;
1455 
1456 	/*
1457 	 * We will change zp_size, lock the whole file.
1458 	 */
1459 	rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1460 
1461 	/*
1462 	 * Nothing to do if file already at desired length.
1463 	 */
1464 	if (end <= zp->z_size) {
1465 		zfs_range_unlock(rl);
1466 		return (0);
1467 	}
1468 top:
1469 	tx = dmu_tx_create(zfsvfs->z_os);
1470 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1471 	zfs_sa_upgrade_txholds(tx, zp);
1472 	if (end > zp->z_blksz &&
1473 	    (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1474 		/*
1475 		 * We are growing the file past the current block size.
1476 		 */
1477 		if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1478 			ASSERT(!ISP2(zp->z_blksz));
1479 			newblksz = MIN(end, SPA_MAXBLOCKSIZE);
1480 		} else {
1481 			newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1482 		}
1483 		dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1484 	} else {
1485 		newblksz = 0;
1486 	}
1487 
1488 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1489 	if (error) {
1490 		if (error == ERESTART) {
1491 			dmu_tx_wait(tx);
1492 			dmu_tx_abort(tx);
1493 			goto top;
1494 		}
1495 		dmu_tx_abort(tx);
1496 		zfs_range_unlock(rl);
1497 		return (error);
1498 	}
1499 
1500 	if (newblksz)
1501 		zfs_grow_blocksize(zp, newblksz, tx);
1502 
1503 	zp->z_size = end;
1504 
1505 	VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1506 	    &zp->z_size, sizeof (zp->z_size), tx));
1507 
1508 	zfs_range_unlock(rl);
1509 
1510 	dmu_tx_commit(tx);
1511 
1512 	return (0);
1513 }
1514 
1515 /*
1516  * Free space in a file.
1517  *
1518  *	IN:	zp	- znode of file to free data in.
1519  *		off	- start of section to free.
1520  *		len	- length of section to free.
1521  *
1522  * 	RETURN:	0 if success
1523  *		error code if failure
1524  */
1525 static int
1526 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1527 {
1528 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1529 	rl_t *rl;
1530 	int error;
1531 
1532 	/*
1533 	 * Lock the range being freed.
1534 	 */
1535 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
1536 
1537 	/*
1538 	 * Nothing to do if file already at desired length.
1539 	 */
1540 	if (off >= zp->z_size) {
1541 		zfs_range_unlock(rl);
1542 		return (0);
1543 	}
1544 
1545 	if (off + len > zp->z_size)
1546 		len = zp->z_size - off;
1547 
1548 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1549 
1550 	zfs_range_unlock(rl);
1551 
1552 	return (error);
1553 }
1554 
1555 /*
1556  * Truncate a file
1557  *
1558  *	IN:	zp	- znode of file to free data in.
1559  *		end	- new end-of-file.
1560  *
1561  * 	RETURN:	0 if success
1562  *		error code if failure
1563  */
1564 static int
1565 zfs_trunc(znode_t *zp, uint64_t end)
1566 {
1567 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1568 	vnode_t *vp = ZTOV(zp);
1569 	dmu_tx_t *tx;
1570 	rl_t *rl;
1571 	int error;
1572 
1573 	/*
1574 	 * We will change zp_size, lock the whole file.
1575 	 */
1576 	rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1577 
1578 	/*
1579 	 * Nothing to do if file already at desired length.
1580 	 */
1581 	if (end >= zp->z_size) {
1582 		zfs_range_unlock(rl);
1583 		return (0);
1584 	}
1585 
1586 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,  -1);
1587 	if (error) {
1588 		zfs_range_unlock(rl);
1589 		return (error);
1590 	}
1591 top:
1592 	tx = dmu_tx_create(zfsvfs->z_os);
1593 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1594 	zfs_sa_upgrade_txholds(tx, zp);
1595 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1596 	if (error) {
1597 		if (error == ERESTART) {
1598 			dmu_tx_wait(tx);
1599 			dmu_tx_abort(tx);
1600 			goto top;
1601 		}
1602 		dmu_tx_abort(tx);
1603 		zfs_range_unlock(rl);
1604 		return (error);
1605 	}
1606 
1607 	zp->z_size = end;
1608 
1609 	VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1610 	    &zp->z_size, sizeof (zp->z_size), tx));
1611 
1612 	dmu_tx_commit(tx);
1613 
1614 	/*
1615 	 * Clear any mapped pages in the truncated region.  This has to
1616 	 * happen outside of the transaction to avoid the possibility of
1617 	 * a deadlock with someone trying to push a page that we are
1618 	 * about to invalidate.
1619 	 */
1620 	if (vn_has_cached_data(vp)) {
1621 		page_t *pp;
1622 		uint64_t start = end & PAGEMASK;
1623 		int poff = end & PAGEOFFSET;
1624 
1625 		if (poff != 0 && (pp = page_lookup(vp, start, SE_SHARED))) {
1626 			/*
1627 			 * We need to zero a partial page.
1628 			 */
1629 			pagezero(pp, poff, PAGESIZE - poff);
1630 			start += PAGESIZE;
1631 			page_unlock(pp);
1632 		}
1633 		error = pvn_vplist_dirty(vp, start, zfs_no_putpage,
1634 		    B_INVAL | B_TRUNC, NULL);
1635 		ASSERT(error == 0);
1636 	}
1637 
1638 	zfs_range_unlock(rl);
1639 
1640 	return (0);
1641 }
1642 
1643 /*
1644  * Free space in a file
1645  *
1646  *	IN:	zp	- znode of file to free data in.
1647  *		off	- start of range
1648  *		len	- end of range (0 => EOF)
1649  *		flag	- current file open mode flags.
1650  *		log	- TRUE if this action should be logged
1651  *
1652  * 	RETURN:	0 if success
1653  *		error code if failure
1654  */
1655 int
1656 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1657 {
1658 	vnode_t *vp = ZTOV(zp);
1659 	dmu_tx_t *tx;
1660 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1661 	zilog_t *zilog = zfsvfs->z_log;
1662 	uint64_t mode;
1663 	uint64_t mtime[2], ctime[2];
1664 	sa_bulk_attr_t bulk[3];
1665 	int count = 0;
1666 	int error;
1667 
1668 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1669 	    sizeof (mode))) != 0)
1670 		return (error);
1671 
1672 	if (off > zp->z_size) {
1673 		error =  zfs_extend(zp, off+len);
1674 		if (error == 0 && log)
1675 			goto log;
1676 		else
1677 			return (error);
1678 	}
1679 
1680 	/*
1681 	 * Check for any locks in the region to be freed.
1682 	 */
1683 
1684 	if (MANDLOCK(vp, (mode_t)mode)) {
1685 		uint64_t length = (len ? len : zp->z_size - off);
1686 		if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1687 			return (error);
1688 	}
1689 
1690 	if (len == 0) {
1691 		error = zfs_trunc(zp, off);
1692 	} else {
1693 		if ((error = zfs_free_range(zp, off, len)) == 0 &&
1694 		    off + len > zp->z_size)
1695 			error = zfs_extend(zp, off+len);
1696 	}
1697 	if (error || !log)
1698 		return (error);
1699 log:
1700 	tx = dmu_tx_create(zfsvfs->z_os);
1701 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1702 	zfs_sa_upgrade_txholds(tx, zp);
1703 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1704 	if (error) {
1705 		if (error == ERESTART) {
1706 			dmu_tx_wait(tx);
1707 			dmu_tx_abort(tx);
1708 			goto log;
1709 		}
1710 		dmu_tx_abort(tx);
1711 		return (error);
1712 	}
1713 
1714 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1715 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1716 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1717 	    NULL, &zp->z_pflags, 8);
1718 	zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1719 	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1720 	ASSERT(error == 0);
1721 
1722 	zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1723 
1724 	dmu_tx_commit(tx);
1725 	return (0);
1726 }
1727 
1728 void
1729 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1730 {
1731 	zfsvfs_t	zfsvfs;
1732 	uint64_t	moid, obj, sa_obj, version;
1733 	uint64_t	sense = ZFS_CASE_SENSITIVE;
1734 	uint64_t	norm = 0;
1735 	nvpair_t	*elem;
1736 	int		error;
1737 	int		i;
1738 	znode_t		*rootzp = NULL;
1739 	vnode_t		*vp;
1740 	vattr_t		vattr;
1741 	znode_t		*zp;
1742 	zfs_acl_ids_t	acl_ids;
1743 
1744 	/*
1745 	 * First attempt to create master node.
1746 	 */
1747 	/*
1748 	 * In an empty objset, there are no blocks to read and thus
1749 	 * there can be no i/o errors (which we assert below).
1750 	 */
1751 	moid = MASTER_NODE_OBJ;
1752 	error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1753 	    DMU_OT_NONE, 0, tx);
1754 	ASSERT(error == 0);
1755 
1756 	/*
1757 	 * Set starting attributes.
1758 	 */
1759 	version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1760 	elem = NULL;
1761 	while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1762 		/* For the moment we expect all zpl props to be uint64_ts */
1763 		uint64_t val;
1764 		char *name;
1765 
1766 		ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1767 		VERIFY(nvpair_value_uint64(elem, &val) == 0);
1768 		name = nvpair_name(elem);
1769 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1770 			if (val < version)
1771 				version = val;
1772 		} else {
1773 			error = zap_update(os, moid, name, 8, 1, &val, tx);
1774 		}
1775 		ASSERT(error == 0);
1776 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1777 			norm = val;
1778 		else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1779 			sense = val;
1780 	}
1781 	ASSERT(version != 0);
1782 	error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1783 
1784 	/*
1785 	 * Create zap object used for SA attribute registration
1786 	 */
1787 
1788 	if (version >= ZPL_VERSION_SA) {
1789 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1790 		    DMU_OT_NONE, 0, tx);
1791 		error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1792 		ASSERT(error == 0);
1793 	} else {
1794 		sa_obj = 0;
1795 	}
1796 	/*
1797 	 * Create a delete queue.
1798 	 */
1799 	obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1800 
1801 	error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1802 	ASSERT(error == 0);
1803 
1804 	/*
1805 	 * Create root znode.  Create minimal znode/vnode/zfsvfs
1806 	 * to allow zfs_mknode to work.
1807 	 */
1808 	vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1809 	vattr.va_type = VDIR;
1810 	vattr.va_mode = S_IFDIR|0755;
1811 	vattr.va_uid = crgetuid(cr);
1812 	vattr.va_gid = crgetgid(cr);
1813 
1814 	rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1815 	ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1816 	rootzp->z_moved = 0;
1817 	rootzp->z_unlinked = 0;
1818 	rootzp->z_atime_dirty = 0;
1819 	rootzp->z_is_sa = USE_SA(version, os);
1820 
1821 	vp = ZTOV(rootzp);
1822 	vn_reinit(vp);
1823 	vp->v_type = VDIR;
1824 
1825 	bzero(&zfsvfs, sizeof (zfsvfs_t));
1826 
1827 	zfsvfs.z_os = os;
1828 	zfsvfs.z_parent = &zfsvfs;
1829 	zfsvfs.z_version = version;
1830 	zfsvfs.z_use_fuids = USE_FUIDS(version, os);
1831 	zfsvfs.z_use_sa = USE_SA(version, os);
1832 	zfsvfs.z_norm = norm;
1833 
1834 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1835 	    &zfsvfs.z_attr_table);
1836 
1837 	ASSERT(error == 0);
1838 
1839 	/*
1840 	 * Fold case on file systems that are always or sometimes case
1841 	 * insensitive.
1842 	 */
1843 	if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1844 		zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER;
1845 
1846 	mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1847 	list_create(&zfsvfs.z_all_znodes, sizeof (znode_t),
1848 	    offsetof(znode_t, z_link_node));
1849 
1850 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1851 		mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1852 
1853 	rootzp->z_zfsvfs = &zfsvfs;
1854 	VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1855 	    cr, NULL, &acl_ids));
1856 	zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1857 	ASSERT3P(zp, ==, rootzp);
1858 	ASSERT(!vn_in_dnlc(ZTOV(rootzp))); /* not valid to move */
1859 	error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1860 	ASSERT(error == 0);
1861 	zfs_acl_ids_free(&acl_ids);
1862 	POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1863 
1864 	ZTOV(rootzp)->v_count = 0;
1865 	sa_handle_destroy(rootzp->z_sa_hdl);
1866 	kmem_cache_free(znode_cache, rootzp);
1867 
1868 	/*
1869 	 * Create shares directory
1870 	 */
1871 
1872 	error = zfs_create_share_dir(&zfsvfs, tx);
1873 
1874 	ASSERT(error == 0);
1875 
1876 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1877 		mutex_destroy(&zfsvfs.z_hold_mtx[i]);
1878 }
1879 
1880 #endif /* _KERNEL */
1881 
1882 /*
1883  * Given an object number, return its parent object number and whether
1884  * or not the object is an extended attribute directory.
1885  */
1886 static int
1887 zfs_obj_to_pobj(objset_t *osp, uint64_t obj, uint64_t *pobjp, int *is_xattrdir,
1888     sa_attr_type_t *sa_table)
1889 {
1890 	dmu_buf_t *db;
1891 	dmu_object_info_t doi;
1892 	int error;
1893 	uint64_t parent;
1894 	uint64_t pflags;
1895 	uint64_t mode;
1896 	sa_bulk_attr_t bulk[3];
1897 	sa_handle_t *hdl;
1898 	int count = 0;
1899 
1900 	if ((error = sa_buf_hold(osp, obj, FTAG, &db)) != 0)
1901 		return (error);
1902 
1903 	dmu_object_info_from_db(db, &doi);
1904 	if ((doi.doi_bonus_type != DMU_OT_SA &&
1905 	    doi.doi_bonus_type != DMU_OT_ZNODE) ||
1906 	    doi.doi_bonus_type == DMU_OT_ZNODE &&
1907 	    doi.doi_bonus_size < sizeof (znode_phys_t)) {
1908 		sa_buf_rele(db, FTAG);
1909 		return (EINVAL);
1910 	}
1911 
1912 	if ((error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE,
1913 	    &hdl)) != 0) {
1914 		sa_buf_rele(db, FTAG);
1915 		return (error);
1916 	}
1917 
1918 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT],
1919 	    NULL, &parent, 8);
1920 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
1921 	    &pflags, 8);
1922 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1923 	    &mode, 8);
1924 
1925 	if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) {
1926 		sa_buf_rele(db, FTAG);
1927 		sa_handle_destroy(hdl);
1928 		return (error);
1929 	}
1930 	*pobjp = parent;
1931 	*is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
1932 	sa_handle_destroy(hdl);
1933 	sa_buf_rele(db, FTAG);
1934 
1935 	return (0);
1936 }
1937 
1938 int
1939 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
1940 {
1941 	char *path = buf + len - 1;
1942 	sa_attr_type_t *sa_table;
1943 	int error;
1944 	uint64_t sa_obj = 0;
1945 
1946 	*path = '\0';
1947 
1948 	error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1949 
1950 	if (error != 0 && error != ENOENT)
1951 		return (error);
1952 
1953 	if ((error = sa_setup(osp, sa_obj, zfs_attr_table,
1954 	    ZPL_END, &sa_table)) != 0)
1955 		return (error);
1956 
1957 	for (;;) {
1958 		uint64_t pobj;
1959 		char component[MAXNAMELEN + 2];
1960 		size_t complen;
1961 		int is_xattrdir;
1962 
1963 		if ((error = zfs_obj_to_pobj(osp, obj, &pobj,
1964 		    &is_xattrdir, sa_table)) != 0)
1965 			break;
1966 
1967 		if (pobj == obj) {
1968 			if (path[0] != '/')
1969 				*--path = '/';
1970 			break;
1971 		}
1972 
1973 		component[0] = '/';
1974 		if (is_xattrdir) {
1975 			(void) sprintf(component + 1, "<xattrdir>");
1976 		} else {
1977 			error = zap_value_search(osp, pobj, obj,
1978 			    ZFS_DIRENT_OBJ(-1ULL), component + 1);
1979 			if (error != 0)
1980 				break;
1981 		}
1982 
1983 		complen = strlen(component);
1984 		path -= complen;
1985 		ASSERT(path >= buf);
1986 		bcopy(component, path, complen);
1987 		obj = pobj;
1988 	}
1989 
1990 	if (error == 0)
1991 		(void) memmove(buf, path, buf + len - path);
1992 
1993 	return (error);
1994 }
1995