xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_znode.c (revision c701fde6911c957e71b37aac4daf672bd828f4d7)
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  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24  */
25 
26 /* Portions Copyright 2007 Jeremy Teo */
27 
28 #ifdef _KERNEL
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/time.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
35 #include <sys/mntent.h>
36 #include <sys/mkdev.h>
37 #include <sys/u8_textprep.h>
38 #include <sys/dsl_dataset.h>
39 #include <sys/vfs.h>
40 #include <sys/vfs_opreg.h>
41 #include <sys/vnode.h>
42 #include <sys/file.h>
43 #include <sys/kmem.h>
44 #include <sys/errno.h>
45 #include <sys/unistd.h>
46 #include <sys/mode.h>
47 #include <sys/atomic.h>
48 #include <vm/pvn.h>
49 #include "fs/fs_subr.h"
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_acl.h>
52 #include <sys/zfs_ioctl.h>
53 #include <sys/zfs_rlock.h>
54 #include <sys/zfs_fuid.h>
55 #include <sys/dnode.h>
56 #include <sys/fs/zfs.h>
57 #include <sys/kidmap.h>
58 #endif /* _KERNEL */
59 
60 #include <sys/dmu.h>
61 #include <sys/dmu_objset.h>
62 #include <sys/refcount.h>
63 #include <sys/stat.h>
64 #include <sys/zap.h>
65 #include <sys/zfs_znode.h>
66 #include <sys/sa.h>
67 #include <sys/zfs_sa.h>
68 #include <sys/zfs_stat.h>
69 
70 #include "zfs_prop.h"
71 #include "zfs_comutil.h"
72 
73 /*
74  * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
75  * turned on when DEBUG is also defined.
76  */
77 #ifdef	DEBUG
78 #define	ZNODE_STATS
79 #endif	/* DEBUG */
80 
81 #ifdef	ZNODE_STATS
82 #define	ZNODE_STAT_ADD(stat)			((stat)++)
83 #else
84 #define	ZNODE_STAT_ADD(stat)			/* nothing */
85 #endif	/* ZNODE_STATS */
86 
87 /*
88  * Functions needed for userland (ie: libzpool) are not put under
89  * #ifdef_KERNEL; the rest of the functions have dependencies
90  * (such as VFS logic) that will not compile easily in userland.
91  */
92 #ifdef _KERNEL
93 /*
94  * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to
95  * be freed before it can be safely accessed.
96  */
97 krwlock_t zfsvfs_lock;
98 
99 static kmem_cache_t *znode_cache = NULL;
100 
101 /*ARGSUSED*/
102 static void
103 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
104 {
105 	/*
106 	 * We should never drop all dbuf refs without first clearing
107 	 * the eviction callback.
108 	 */
109 	panic("evicting znode %p\n", user_ptr);
110 }
111 
112 /*ARGSUSED*/
113 static int
114 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
115 {
116 	znode_t *zp = buf;
117 
118 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
119 
120 	zp->z_vnode = vn_alloc(kmflags);
121 	if (zp->z_vnode == NULL) {
122 		return (-1);
123 	}
124 	ZTOV(zp)->v_data = zp;
125 
126 	list_link_init(&zp->z_link_node);
127 
128 	mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
129 	rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
130 	rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
131 	mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
132 
133 	mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL);
134 	avl_create(&zp->z_range_avl, zfs_range_compare,
135 	    sizeof (rl_t), offsetof(rl_t, r_node));
136 
137 	zp->z_dirlocks = NULL;
138 	zp->z_acl_cached = NULL;
139 	zp->z_moved = 0;
140 	return (0);
141 }
142 
143 /*ARGSUSED*/
144 static void
145 zfs_znode_cache_destructor(void *buf, void *arg)
146 {
147 	znode_t *zp = buf;
148 
149 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
150 	ASSERT(ZTOV(zp)->v_data == zp);
151 	vn_free(ZTOV(zp));
152 	ASSERT(!list_link_active(&zp->z_link_node));
153 	mutex_destroy(&zp->z_lock);
154 	rw_destroy(&zp->z_parent_lock);
155 	rw_destroy(&zp->z_name_lock);
156 	mutex_destroy(&zp->z_acl_lock);
157 	avl_destroy(&zp->z_range_avl);
158 	mutex_destroy(&zp->z_range_lock);
159 
160 	ASSERT(zp->z_dirlocks == NULL);
161 	ASSERT(zp->z_acl_cached == NULL);
162 }
163 
164 #ifdef	ZNODE_STATS
165 static struct {
166 	uint64_t zms_zfsvfs_invalid;
167 	uint64_t zms_zfsvfs_recheck1;
168 	uint64_t zms_zfsvfs_unmounted;
169 	uint64_t zms_zfsvfs_recheck2;
170 	uint64_t zms_obj_held;
171 	uint64_t zms_vnode_locked;
172 	uint64_t zms_not_only_dnlc;
173 } znode_move_stats;
174 #endif	/* ZNODE_STATS */
175 
176 static void
177 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
178 {
179 	vnode_t *vp;
180 
181 	/* Copy fields. */
182 	nzp->z_zfsvfs = ozp->z_zfsvfs;
183 
184 	/* Swap vnodes. */
185 	vp = nzp->z_vnode;
186 	nzp->z_vnode = ozp->z_vnode;
187 	ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
188 	ZTOV(ozp)->v_data = ozp;
189 	ZTOV(nzp)->v_data = nzp;
190 
191 	nzp->z_id = ozp->z_id;
192 	ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
193 	ASSERT(avl_numnodes(&ozp->z_range_avl) == 0);
194 	nzp->z_unlinked = ozp->z_unlinked;
195 	nzp->z_atime_dirty = ozp->z_atime_dirty;
196 	nzp->z_zn_prefetch = ozp->z_zn_prefetch;
197 	nzp->z_blksz = ozp->z_blksz;
198 	nzp->z_seq = ozp->z_seq;
199 	nzp->z_mapcnt = ozp->z_mapcnt;
200 	nzp->z_gen = ozp->z_gen;
201 	nzp->z_sync_cnt = ozp->z_sync_cnt;
202 	nzp->z_is_sa = ozp->z_is_sa;
203 	nzp->z_sa_hdl = ozp->z_sa_hdl;
204 	bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
205 	nzp->z_links = ozp->z_links;
206 	nzp->z_size = ozp->z_size;
207 	nzp->z_pflags = ozp->z_pflags;
208 	nzp->z_uid = ozp->z_uid;
209 	nzp->z_gid = ozp->z_gid;
210 	nzp->z_mode = ozp->z_mode;
211 
212 	/*
213 	 * Since this is just an idle znode and kmem is already dealing with
214 	 * memory pressure, release any cached ACL.
215 	 */
216 	if (ozp->z_acl_cached) {
217 		zfs_acl_free(ozp->z_acl_cached);
218 		ozp->z_acl_cached = NULL;
219 	}
220 
221 	sa_set_userp(nzp->z_sa_hdl, nzp);
222 
223 	/*
224 	 * Invalidate the original znode by clearing fields that provide a
225 	 * pointer back to the znode. Set the low bit of the vfs pointer to
226 	 * ensure that zfs_znode_move() recognizes the znode as invalid in any
227 	 * subsequent callback.
228 	 */
229 	ozp->z_sa_hdl = NULL;
230 	POINTER_INVALIDATE(&ozp->z_zfsvfs);
231 
232 	/*
233 	 * Mark the znode.
234 	 */
235 	nzp->z_moved = 1;
236 	ozp->z_moved = (uint8_t)-1;
237 }
238 
239 /*ARGSUSED*/
240 static kmem_cbrc_t
241 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
242 {
243 	znode_t *ozp = buf, *nzp = newbuf;
244 	zfsvfs_t *zfsvfs;
245 	vnode_t *vp;
246 
247 	/*
248 	 * The znode is on the file system's list of known znodes if the vfs
249 	 * pointer is valid. We set the low bit of the vfs pointer when freeing
250 	 * the znode to invalidate it, and the memory patterns written by kmem
251 	 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
252 	 * created znode sets the vfs pointer last of all to indicate that the
253 	 * znode is known and in a valid state to be moved by this function.
254 	 */
255 	zfsvfs = ozp->z_zfsvfs;
256 	if (!POINTER_IS_VALID(zfsvfs)) {
257 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
258 		return (KMEM_CBRC_DONT_KNOW);
259 	}
260 
261 	/*
262 	 * Close a small window in which it's possible that the filesystem could
263 	 * be unmounted and freed, and zfsvfs, though valid in the previous
264 	 * statement, could point to unrelated memory by the time we try to
265 	 * prevent the filesystem from being unmounted.
266 	 */
267 	rw_enter(&zfsvfs_lock, RW_WRITER);
268 	if (zfsvfs != ozp->z_zfsvfs) {
269 		rw_exit(&zfsvfs_lock);
270 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
271 		return (KMEM_CBRC_DONT_KNOW);
272 	}
273 
274 	/*
275 	 * If the znode is still valid, then so is the file system. We know that
276 	 * no valid file system can be freed while we hold zfsvfs_lock, so we
277 	 * can safely ensure that the filesystem is not and will not be
278 	 * unmounted. The next statement is equivalent to ZFS_ENTER().
279 	 */
280 	rrm_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
281 	if (zfsvfs->z_unmounted) {
282 		ZFS_EXIT(zfsvfs);
283 		rw_exit(&zfsvfs_lock);
284 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
285 		return (KMEM_CBRC_DONT_KNOW);
286 	}
287 	rw_exit(&zfsvfs_lock);
288 
289 	mutex_enter(&zfsvfs->z_znodes_lock);
290 	/*
291 	 * Recheck the vfs pointer in case the znode was removed just before
292 	 * acquiring the lock.
293 	 */
294 	if (zfsvfs != ozp->z_zfsvfs) {
295 		mutex_exit(&zfsvfs->z_znodes_lock);
296 		ZFS_EXIT(zfsvfs);
297 		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
298 		return (KMEM_CBRC_DONT_KNOW);
299 	}
300 
301 	/*
302 	 * At this point we know that as long as we hold z_znodes_lock, the
303 	 * znode cannot be freed and fields within the znode can be safely
304 	 * accessed. Now, prevent a race with zfs_zget().
305 	 */
306 	if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
307 		mutex_exit(&zfsvfs->z_znodes_lock);
308 		ZFS_EXIT(zfsvfs);
309 		ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
310 		return (KMEM_CBRC_LATER);
311 	}
312 
313 	vp = ZTOV(ozp);
314 	if (mutex_tryenter(&vp->v_lock) == 0) {
315 		ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
316 		mutex_exit(&zfsvfs->z_znodes_lock);
317 		ZFS_EXIT(zfsvfs);
318 		ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
319 		return (KMEM_CBRC_LATER);
320 	}
321 
322 	/* Only move znodes that are referenced _only_ by the DNLC. */
323 	if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
324 		mutex_exit(&vp->v_lock);
325 		ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
326 		mutex_exit(&zfsvfs->z_znodes_lock);
327 		ZFS_EXIT(zfsvfs);
328 		ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
329 		return (KMEM_CBRC_LATER);
330 	}
331 
332 	/*
333 	 * The znode is known and in a valid state to move. We're holding the
334 	 * locks needed to execute the critical section.
335 	 */
336 	zfs_znode_move_impl(ozp, nzp);
337 	mutex_exit(&vp->v_lock);
338 	ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
339 
340 	list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
341 	mutex_exit(&zfsvfs->z_znodes_lock);
342 	ZFS_EXIT(zfsvfs);
343 
344 	return (KMEM_CBRC_YES);
345 }
346 
347 void
348 zfs_znode_init(void)
349 {
350 	/*
351 	 * Initialize zcache
352 	 */
353 	rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
354 	ASSERT(znode_cache == NULL);
355 	znode_cache = kmem_cache_create("zfs_znode_cache",
356 	    sizeof (znode_t), 0, zfs_znode_cache_constructor,
357 	    zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
358 	kmem_cache_set_move(znode_cache, zfs_znode_move);
359 }
360 
361 void
362 zfs_znode_fini(void)
363 {
364 	/*
365 	 * Cleanup vfs & vnode ops
366 	 */
367 	zfs_remove_op_tables();
368 
369 	/*
370 	 * Cleanup zcache
371 	 */
372 	if (znode_cache)
373 		kmem_cache_destroy(znode_cache);
374 	znode_cache = NULL;
375 	rw_destroy(&zfsvfs_lock);
376 }
377 
378 struct vnodeops *zfs_dvnodeops;
379 struct vnodeops *zfs_fvnodeops;
380 struct vnodeops *zfs_symvnodeops;
381 struct vnodeops *zfs_xdvnodeops;
382 struct vnodeops *zfs_evnodeops;
383 struct vnodeops *zfs_sharevnodeops;
384 
385 void
386 zfs_remove_op_tables()
387 {
388 	/*
389 	 * Remove vfs ops
390 	 */
391 	ASSERT(zfsfstype);
392 	(void) vfs_freevfsops_by_type(zfsfstype);
393 	zfsfstype = 0;
394 
395 	/*
396 	 * Remove vnode ops
397 	 */
398 	if (zfs_dvnodeops)
399 		vn_freevnodeops(zfs_dvnodeops);
400 	if (zfs_fvnodeops)
401 		vn_freevnodeops(zfs_fvnodeops);
402 	if (zfs_symvnodeops)
403 		vn_freevnodeops(zfs_symvnodeops);
404 	if (zfs_xdvnodeops)
405 		vn_freevnodeops(zfs_xdvnodeops);
406 	if (zfs_evnodeops)
407 		vn_freevnodeops(zfs_evnodeops);
408 	if (zfs_sharevnodeops)
409 		vn_freevnodeops(zfs_sharevnodeops);
410 
411 	zfs_dvnodeops = NULL;
412 	zfs_fvnodeops = NULL;
413 	zfs_symvnodeops = NULL;
414 	zfs_xdvnodeops = NULL;
415 	zfs_evnodeops = NULL;
416 	zfs_sharevnodeops = NULL;
417 }
418 
419 extern const fs_operation_def_t zfs_dvnodeops_template[];
420 extern const fs_operation_def_t zfs_fvnodeops_template[];
421 extern const fs_operation_def_t zfs_xdvnodeops_template[];
422 extern const fs_operation_def_t zfs_symvnodeops_template[];
423 extern const fs_operation_def_t zfs_evnodeops_template[];
424 extern const fs_operation_def_t zfs_sharevnodeops_template[];
425 
426 int
427 zfs_create_op_tables()
428 {
429 	int error;
430 
431 	/*
432 	 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
433 	 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
434 	 * In this case we just return as the ops vectors are already set up.
435 	 */
436 	if (zfs_dvnodeops)
437 		return (0);
438 
439 	error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
440 	    &zfs_dvnodeops);
441 	if (error)
442 		return (error);
443 
444 	error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
445 	    &zfs_fvnodeops);
446 	if (error)
447 		return (error);
448 
449 	error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
450 	    &zfs_symvnodeops);
451 	if (error)
452 		return (error);
453 
454 	error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
455 	    &zfs_xdvnodeops);
456 	if (error)
457 		return (error);
458 
459 	error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
460 	    &zfs_evnodeops);
461 	if (error)
462 		return (error);
463 
464 	error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
465 	    &zfs_sharevnodeops);
466 
467 	return (error);
468 }
469 
470 int
471 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
472 {
473 	zfs_acl_ids_t acl_ids;
474 	vattr_t vattr;
475 	znode_t *sharezp;
476 	vnode_t *vp;
477 	znode_t *zp;
478 	int error;
479 
480 	vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
481 	vattr.va_type = VDIR;
482 	vattr.va_mode = S_IFDIR|0555;
483 	vattr.va_uid = crgetuid(kcred);
484 	vattr.va_gid = crgetgid(kcred);
485 
486 	sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
487 	ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
488 	sharezp->z_moved = 0;
489 	sharezp->z_unlinked = 0;
490 	sharezp->z_atime_dirty = 0;
491 	sharezp->z_zfsvfs = zfsvfs;
492 	sharezp->z_is_sa = zfsvfs->z_use_sa;
493 
494 	vp = ZTOV(sharezp);
495 	vn_reinit(vp);
496 	vp->v_type = VDIR;
497 
498 	VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
499 	    kcred, NULL, &acl_ids));
500 	zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
501 	ASSERT3P(zp, ==, sharezp);
502 	ASSERT(!vn_in_dnlc(ZTOV(sharezp))); /* not valid to move */
503 	POINTER_INVALIDATE(&sharezp->z_zfsvfs);
504 	error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
505 	    ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
506 	zfsvfs->z_shares_dir = sharezp->z_id;
507 
508 	zfs_acl_ids_free(&acl_ids);
509 	ZTOV(sharezp)->v_count = 0;
510 	sa_handle_destroy(sharezp->z_sa_hdl);
511 	kmem_cache_free(znode_cache, sharezp);
512 
513 	return (error);
514 }
515 
516 /*
517  * define a couple of values we need available
518  * for both 64 and 32 bit environments.
519  */
520 #ifndef NBITSMINOR64
521 #define	NBITSMINOR64	32
522 #endif
523 #ifndef MAXMAJ64
524 #define	MAXMAJ64	0xffffffffUL
525 #endif
526 #ifndef	MAXMIN64
527 #define	MAXMIN64	0xffffffffUL
528 #endif
529 
530 /*
531  * Create special expldev for ZFS private use.
532  * Can't use standard expldev since it doesn't do
533  * what we want.  The standard expldev() takes a
534  * dev32_t in LP64 and expands it to a long dev_t.
535  * We need an interface that takes a dev32_t in ILP32
536  * and expands it to a long dev_t.
537  */
538 static uint64_t
539 zfs_expldev(dev_t dev)
540 {
541 #ifndef _LP64
542 	major_t major = (major_t)dev >> NBITSMINOR32 & MAXMAJ32;
543 	return (((uint64_t)major << NBITSMINOR64) |
544 	    ((minor_t)dev & MAXMIN32));
545 #else
546 	return (dev);
547 #endif
548 }
549 
550 /*
551  * Special cmpldev for ZFS private use.
552  * Can't use standard cmpldev since it takes
553  * a long dev_t and compresses it to dev32_t in
554  * LP64.  We need to do a compaction of a long dev_t
555  * to a dev32_t in ILP32.
556  */
557 dev_t
558 zfs_cmpldev(uint64_t dev)
559 {
560 #ifndef _LP64
561 	minor_t minor = (minor_t)dev & MAXMIN64;
562 	major_t major = (major_t)(dev >> NBITSMINOR64) & MAXMAJ64;
563 
564 	if (major > MAXMAJ32 || minor > MAXMIN32)
565 		return (NODEV32);
566 
567 	return (((dev32_t)major << NBITSMINOR32) | minor);
568 #else
569 	return (dev);
570 #endif
571 }
572 
573 static void
574 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
575     dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
576 {
577 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
578 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
579 
580 	mutex_enter(&zp->z_lock);
581 
582 	ASSERT(zp->z_sa_hdl == NULL);
583 	ASSERT(zp->z_acl_cached == NULL);
584 	if (sa_hdl == NULL) {
585 		VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
586 		    SA_HDL_SHARED, &zp->z_sa_hdl));
587 	} else {
588 		zp->z_sa_hdl = sa_hdl;
589 		sa_set_userp(sa_hdl, zp);
590 	}
591 
592 	zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
593 
594 	/*
595 	 * Slap on VROOT if we are the root znode
596 	 */
597 	if (zp->z_id == zfsvfs->z_root)
598 		ZTOV(zp)->v_flag |= VROOT;
599 
600 	mutex_exit(&zp->z_lock);
601 	vn_exists(ZTOV(zp));
602 }
603 
604 void
605 zfs_znode_dmu_fini(znode_t *zp)
606 {
607 	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
608 	    zp->z_unlinked ||
609 	    RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
610 
611 	sa_handle_destroy(zp->z_sa_hdl);
612 	zp->z_sa_hdl = NULL;
613 }
614 
615 /*
616  * Construct a new znode/vnode and intialize.
617  *
618  * This does not do a call to dmu_set_user() that is
619  * up to the caller to do, in case you don't want to
620  * return the znode
621  */
622 static znode_t *
623 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
624     dmu_object_type_t obj_type, sa_handle_t *hdl)
625 {
626 	znode_t	*zp;
627 	vnode_t *vp;
628 	uint64_t mode;
629 	uint64_t parent;
630 	sa_bulk_attr_t bulk[9];
631 	int count = 0;
632 
633 	zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
634 
635 	ASSERT(zp->z_dirlocks == NULL);
636 	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
637 	zp->z_moved = 0;
638 
639 	/*
640 	 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
641 	 * the zfs_znode_move() callback.
642 	 */
643 	zp->z_sa_hdl = NULL;
644 	zp->z_unlinked = 0;
645 	zp->z_atime_dirty = 0;
646 	zp->z_mapcnt = 0;
647 	zp->z_id = db->db_object;
648 	zp->z_blksz = blksz;
649 	zp->z_seq = 0x7A4653;
650 	zp->z_sync_cnt = 0;
651 
652 	vp = ZTOV(zp);
653 	vn_reinit(vp);
654 
655 	zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
656 
657 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
658 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
659 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
660 	    &zp->z_size, 8);
661 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
662 	    &zp->z_links, 8);
663 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
664 	    &zp->z_pflags, 8);
665 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
666 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
667 	    &zp->z_atime, 16);
668 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
669 	    &zp->z_uid, 8);
670 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
671 	    &zp->z_gid, 8);
672 
673 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) {
674 		if (hdl == NULL)
675 			sa_handle_destroy(zp->z_sa_hdl);
676 		kmem_cache_free(znode_cache, zp);
677 		return (NULL);
678 	}
679 
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(zp->z_uid == 0 && zp->z_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		bonuslen;
774 	sa_handle_t	*sa_hdl;
775 	dmu_object_type_t obj_type;
776 	sa_bulk_attr_t	sa_attrs[ZPL_END];
777 	int		cnt = 0;
778 	zfs_acl_locator_cb_t locate = { 0 };
779 
780 	ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
781 
782 	if (zfsvfs->z_replay) {
783 		obj = vap->va_nodeid;
784 		now = vap->va_ctime;		/* see zfs_replay_create() */
785 		gen = vap->va_nblocks;		/* ditto */
786 	} else {
787 		obj = 0;
788 		gethrestime(&now);
789 		gen = dmu_tx_get_txg(tx);
790 	}
791 
792 	obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
793 	bonuslen = (obj_type == DMU_OT_SA) ?
794 	    DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE;
795 
796 	/*
797 	 * Create a new DMU object.
798 	 */
799 	/*
800 	 * There's currently no mechanism for pre-reading the blocks that will
801 	 * be needed to allocate a new object, so we accept the small chance
802 	 * that there will be an i/o error and we will fail one of the
803 	 * assertions below.
804 	 */
805 	if (vap->va_type == VDIR) {
806 		if (zfsvfs->z_replay) {
807 			VERIFY0(zap_create_claim_norm(zfsvfs->z_os, obj,
808 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
809 			    obj_type, bonuslen, tx));
810 		} else {
811 			obj = zap_create_norm(zfsvfs->z_os,
812 			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
813 			    obj_type, bonuslen, tx);
814 		}
815 	} else {
816 		if (zfsvfs->z_replay) {
817 			VERIFY0(dmu_object_claim(zfsvfs->z_os, obj,
818 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
819 			    obj_type, bonuslen, tx));
820 		} else {
821 			obj = dmu_object_alloc(zfsvfs->z_os,
822 			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
823 			    obj_type, bonuslen, tx);
824 		}
825 	}
826 
827 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
828 	VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
829 
830 	/*
831 	 * If this is the root, fix up the half-initialized parent pointer
832 	 * to reference the just-allocated physical data area.
833 	 */
834 	if (flag & IS_ROOT_NODE) {
835 		dzp->z_id = obj;
836 	} else {
837 		dzp_pflags = dzp->z_pflags;
838 	}
839 
840 	/*
841 	 * If parent is an xattr, so am I.
842 	 */
843 	if (dzp_pflags & ZFS_XATTR) {
844 		flag |= IS_XATTR;
845 	}
846 
847 	if (zfsvfs->z_use_fuids)
848 		pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
849 	else
850 		pflags = 0;
851 
852 	if (vap->va_type == VDIR) {
853 		size = 2;		/* contents ("." and "..") */
854 		links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
855 	} else {
856 		size = links = 0;
857 	}
858 
859 	if (vap->va_type == VBLK || vap->va_type == VCHR) {
860 		rdev = zfs_expldev(vap->va_rdev);
861 	}
862 
863 	parent = dzp->z_id;
864 	mode = acl_ids->z_mode;
865 	if (flag & IS_XATTR)
866 		pflags |= ZFS_XATTR;
867 
868 	/*
869 	 * No execs denied will be deterimed when zfs_mode_compute() is called.
870 	 */
871 	pflags |= acl_ids->z_aclp->z_hints &
872 	    (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
873 	    ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
874 
875 	ZFS_TIME_ENCODE(&now, crtime);
876 	ZFS_TIME_ENCODE(&now, ctime);
877 
878 	if (vap->va_mask & AT_ATIME) {
879 		ZFS_TIME_ENCODE(&vap->va_atime, atime);
880 	} else {
881 		ZFS_TIME_ENCODE(&now, atime);
882 	}
883 
884 	if (vap->va_mask & AT_MTIME) {
885 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
886 	} else {
887 		ZFS_TIME_ENCODE(&now, mtime);
888 	}
889 
890 	/* Now add in all of the "SA" attributes */
891 	VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
892 	    &sa_hdl));
893 
894 	/*
895 	 * Setup the array of attributes to be replaced/set on the new file
896 	 *
897 	 * order for  DMU_OT_ZNODE is critical since it needs to be constructed
898 	 * in the old znode_phys_t format.  Don't change this ordering
899 	 */
900 
901 	if (obj_type == DMU_OT_ZNODE) {
902 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
903 		    NULL, &atime, 16);
904 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
905 		    NULL, &mtime, 16);
906 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
907 		    NULL, &ctime, 16);
908 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
909 		    NULL, &crtime, 16);
910 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
911 		    NULL, &gen, 8);
912 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
913 		    NULL, &mode, 8);
914 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
915 		    NULL, &size, 8);
916 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
917 		    NULL, &parent, 8);
918 	} else {
919 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
920 		    NULL, &mode, 8);
921 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
922 		    NULL, &size, 8);
923 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
924 		    NULL, &gen, 8);
925 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
926 		    &acl_ids->z_fuid, 8);
927 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
928 		    &acl_ids->z_fgid, 8);
929 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
930 		    NULL, &parent, 8);
931 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
932 		    NULL, &pflags, 8);
933 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
934 		    NULL, &atime, 16);
935 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
936 		    NULL, &mtime, 16);
937 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
938 		    NULL, &ctime, 16);
939 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
940 		    NULL, &crtime, 16);
941 	}
942 
943 	SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
944 
945 	if (obj_type == DMU_OT_ZNODE) {
946 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
947 		    &empty_xattr, 8);
948 	}
949 	if (obj_type == DMU_OT_ZNODE ||
950 	    (vap->va_type == VBLK || vap->va_type == VCHR)) {
951 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
952 		    NULL, &rdev, 8);
953 
954 	}
955 	if (obj_type == DMU_OT_ZNODE) {
956 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
957 		    NULL, &pflags, 8);
958 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
959 		    &acl_ids->z_fuid, 8);
960 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
961 		    &acl_ids->z_fgid, 8);
962 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
963 		    sizeof (uint64_t) * 4);
964 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
965 		    &acl_phys, sizeof (zfs_acl_phys_t));
966 	} else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
967 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
968 		    &acl_ids->z_aclp->z_acl_count, 8);
969 		locate.cb_aclp = acl_ids->z_aclp;
970 		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
971 		    zfs_acl_data_locator, &locate,
972 		    acl_ids->z_aclp->z_acl_bytes);
973 		mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
974 		    acl_ids->z_fuid, acl_ids->z_fgid);
975 	}
976 
977 	VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
978 
979 	if (!(flag & IS_ROOT_NODE)) {
980 		*zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
981 		ASSERT(*zpp != NULL);
982 	} else {
983 		/*
984 		 * If we are creating the root node, the "parent" we
985 		 * passed in is the znode for the root.
986 		 */
987 		*zpp = dzp;
988 
989 		(*zpp)->z_sa_hdl = sa_hdl;
990 	}
991 
992 	(*zpp)->z_pflags = pflags;
993 	(*zpp)->z_mode = mode;
994 
995 	if (vap->va_mask & AT_XVATTR)
996 		zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
997 
998 	if (obj_type == DMU_OT_ZNODE ||
999 	    acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1000 		VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
1001 	}
1002 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1003 }
1004 
1005 /*
1006  * Update in-core attributes.  It is assumed the caller will be doing an
1007  * sa_bulk_update to push the changes out.
1008  */
1009 void
1010 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1011 {
1012 	xoptattr_t *xoap;
1013 
1014 	xoap = xva_getxoptattr(xvap);
1015 	ASSERT(xoap);
1016 
1017 	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1018 		uint64_t times[2];
1019 		ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1020 		(void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1021 		    &times, sizeof (times), tx);
1022 		XVA_SET_RTN(xvap, XAT_CREATETIME);
1023 	}
1024 	if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1025 		ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1026 		    zp->z_pflags, tx);
1027 		XVA_SET_RTN(xvap, XAT_READONLY);
1028 	}
1029 	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1030 		ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1031 		    zp->z_pflags, tx);
1032 		XVA_SET_RTN(xvap, XAT_HIDDEN);
1033 	}
1034 	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1035 		ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1036 		    zp->z_pflags, tx);
1037 		XVA_SET_RTN(xvap, XAT_SYSTEM);
1038 	}
1039 	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1040 		ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1041 		    zp->z_pflags, tx);
1042 		XVA_SET_RTN(xvap, XAT_ARCHIVE);
1043 	}
1044 	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1045 		ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1046 		    zp->z_pflags, tx);
1047 		XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1048 	}
1049 	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1050 		ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1051 		    zp->z_pflags, tx);
1052 		XVA_SET_RTN(xvap, XAT_NOUNLINK);
1053 	}
1054 	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1055 		ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1056 		    zp->z_pflags, tx);
1057 		XVA_SET_RTN(xvap, XAT_APPENDONLY);
1058 	}
1059 	if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1060 		ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1061 		    zp->z_pflags, tx);
1062 		XVA_SET_RTN(xvap, XAT_NODUMP);
1063 	}
1064 	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1065 		ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1066 		    zp->z_pflags, tx);
1067 		XVA_SET_RTN(xvap, XAT_OPAQUE);
1068 	}
1069 	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1070 		ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1071 		    xoap->xoa_av_quarantined, zp->z_pflags, tx);
1072 		XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1073 	}
1074 	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1075 		ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1076 		    zp->z_pflags, tx);
1077 		XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1078 	}
1079 	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1080 		zfs_sa_set_scanstamp(zp, xvap, tx);
1081 		XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1082 	}
1083 	if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1084 		ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1085 		    zp->z_pflags, tx);
1086 		XVA_SET_RTN(xvap, XAT_REPARSE);
1087 	}
1088 	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1089 		ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1090 		    zp->z_pflags, tx);
1091 		XVA_SET_RTN(xvap, XAT_OFFLINE);
1092 	}
1093 	if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1094 		ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1095 		    zp->z_pflags, tx);
1096 		XVA_SET_RTN(xvap, XAT_SPARSE);
1097 	}
1098 }
1099 
1100 int
1101 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1102 {
1103 	dmu_object_info_t doi;
1104 	dmu_buf_t	*db;
1105 	znode_t		*zp;
1106 	int err;
1107 	sa_handle_t	*hdl;
1108 
1109 	*zpp = NULL;
1110 
1111 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1112 
1113 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1114 	if (err) {
1115 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1116 		return (err);
1117 	}
1118 
1119 	dmu_object_info_from_db(db, &doi);
1120 	if (doi.doi_bonus_type != DMU_OT_SA &&
1121 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1122 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1123 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1124 		sa_buf_rele(db, NULL);
1125 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1126 		return (SET_ERROR(EINVAL));
1127 	}
1128 
1129 	hdl = dmu_buf_get_user(db);
1130 	if (hdl != NULL) {
1131 		zp  = sa_get_userdata(hdl);
1132 
1133 
1134 		/*
1135 		 * Since "SA" does immediate eviction we
1136 		 * should never find a sa handle that doesn't
1137 		 * know about the znode.
1138 		 */
1139 
1140 		ASSERT3P(zp, !=, NULL);
1141 
1142 		mutex_enter(&zp->z_lock);
1143 		ASSERT3U(zp->z_id, ==, obj_num);
1144 		if (zp->z_unlinked) {
1145 			err = SET_ERROR(ENOENT);
1146 		} else {
1147 			VN_HOLD(ZTOV(zp));
1148 			*zpp = zp;
1149 			err = 0;
1150 		}
1151 		sa_buf_rele(db, NULL);
1152 		mutex_exit(&zp->z_lock);
1153 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1154 		return (err);
1155 	}
1156 
1157 	/*
1158 	 * Not found create new znode/vnode
1159 	 * but only if file exists.
1160 	 *
1161 	 * There is a small window where zfs_vget() could
1162 	 * find this object while a file create is still in
1163 	 * progress.  This is checked for in zfs_znode_alloc()
1164 	 *
1165 	 * if zfs_znode_alloc() fails it will drop the hold on the
1166 	 * bonus buffer.
1167 	 */
1168 	zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1169 	    doi.doi_bonus_type, NULL);
1170 	if (zp == NULL) {
1171 		err = SET_ERROR(ENOENT);
1172 	} else {
1173 		*zpp = zp;
1174 	}
1175 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1176 	return (err);
1177 }
1178 
1179 int
1180 zfs_rezget(znode_t *zp)
1181 {
1182 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1183 	dmu_object_info_t doi;
1184 	dmu_buf_t *db;
1185 	uint64_t obj_num = zp->z_id;
1186 	uint64_t mode;
1187 	sa_bulk_attr_t bulk[8];
1188 	int err;
1189 	int count = 0;
1190 	uint64_t gen;
1191 
1192 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1193 
1194 	mutex_enter(&zp->z_acl_lock);
1195 	if (zp->z_acl_cached) {
1196 		zfs_acl_free(zp->z_acl_cached);
1197 		zp->z_acl_cached = NULL;
1198 	}
1199 
1200 	mutex_exit(&zp->z_acl_lock);
1201 	ASSERT(zp->z_sa_hdl == NULL);
1202 	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1203 	if (err) {
1204 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1205 		return (err);
1206 	}
1207 
1208 	dmu_object_info_from_db(db, &doi);
1209 	if (doi.doi_bonus_type != DMU_OT_SA &&
1210 	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1211 	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1212 	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1213 		sa_buf_rele(db, NULL);
1214 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1215 		return (SET_ERROR(EINVAL));
1216 	}
1217 
1218 	zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1219 
1220 	/* reload cached values */
1221 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1222 	    &gen, sizeof (gen));
1223 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1224 	    &zp->z_size, sizeof (zp->z_size));
1225 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1226 	    &zp->z_links, sizeof (zp->z_links));
1227 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1228 	    &zp->z_pflags, sizeof (zp->z_pflags));
1229 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1230 	    &zp->z_atime, sizeof (zp->z_atime));
1231 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1232 	    &zp->z_uid, sizeof (zp->z_uid));
1233 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1234 	    &zp->z_gid, sizeof (zp->z_gid));
1235 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1236 	    &mode, sizeof (mode));
1237 
1238 	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1239 		zfs_znode_dmu_fini(zp);
1240 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1241 		return (SET_ERROR(EIO));
1242 	}
1243 
1244 	zp->z_mode = mode;
1245 
1246 	if (gen != zp->z_gen) {
1247 		zfs_znode_dmu_fini(zp);
1248 		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1249 		return (SET_ERROR(EIO));
1250 	}
1251 
1252 	zp->z_unlinked = (zp->z_links == 0);
1253 	zp->z_blksz = doi.doi_data_block_size;
1254 
1255 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1256 
1257 	return (0);
1258 }
1259 
1260 void
1261 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1262 {
1263 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1264 	objset_t *os = zfsvfs->z_os;
1265 	uint64_t obj = zp->z_id;
1266 	uint64_t acl_obj = zfs_external_acl(zp);
1267 
1268 	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1269 	if (acl_obj) {
1270 		VERIFY(!zp->z_is_sa);
1271 		VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1272 	}
1273 	VERIFY(0 == dmu_object_free(os, obj, tx));
1274 	zfs_znode_dmu_fini(zp);
1275 	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1276 	zfs_znode_free(zp);
1277 }
1278 
1279 void
1280 zfs_zinactive(znode_t *zp)
1281 {
1282 	vnode_t	*vp = ZTOV(zp);
1283 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1284 	uint64_t z_id = zp->z_id;
1285 
1286 	ASSERT(zp->z_sa_hdl);
1287 
1288 	/*
1289 	 * Don't allow a zfs_zget() while were trying to release this znode
1290 	 */
1291 	ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1292 
1293 	mutex_enter(&zp->z_lock);
1294 	mutex_enter(&vp->v_lock);
1295 	vp->v_count--;
1296 	if (vp->v_count > 0 || vn_has_cached_data(vp)) {
1297 		/*
1298 		 * If the hold count is greater than zero, somebody has
1299 		 * obtained a new reference on this znode while we were
1300 		 * processing it here, so we are done.  If we still have
1301 		 * mapped pages then we are also done, since we don't
1302 		 * want to inactivate the znode until the pages get pushed.
1303 		 *
1304 		 * XXX - if vn_has_cached_data(vp) is true, but count == 0,
1305 		 * this seems like it would leave the znode hanging with
1306 		 * no chance to go inactive...
1307 		 */
1308 		mutex_exit(&vp->v_lock);
1309 		mutex_exit(&zp->z_lock);
1310 		ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1311 		return;
1312 	}
1313 	mutex_exit(&vp->v_lock);
1314 
1315 	/*
1316 	 * If this was the last reference to a file with no links,
1317 	 * remove the file from the file system.
1318 	 */
1319 	if (zp->z_unlinked) {
1320 		mutex_exit(&zp->z_lock);
1321 		ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1322 		zfs_rmnode(zp);
1323 		return;
1324 	}
1325 
1326 	mutex_exit(&zp->z_lock);
1327 	zfs_znode_dmu_fini(zp);
1328 	ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1329 	zfs_znode_free(zp);
1330 }
1331 
1332 void
1333 zfs_znode_free(znode_t *zp)
1334 {
1335 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1336 
1337 	vn_invalid(ZTOV(zp));
1338 
1339 	ASSERT(ZTOV(zp)->v_count == 0);
1340 
1341 	mutex_enter(&zfsvfs->z_znodes_lock);
1342 	POINTER_INVALIDATE(&zp->z_zfsvfs);
1343 	list_remove(&zfsvfs->z_all_znodes, zp);
1344 	mutex_exit(&zfsvfs->z_znodes_lock);
1345 
1346 	if (zp->z_acl_cached) {
1347 		zfs_acl_free(zp->z_acl_cached);
1348 		zp->z_acl_cached = NULL;
1349 	}
1350 
1351 	kmem_cache_free(znode_cache, zp);
1352 
1353 	VFS_RELE(zfsvfs->z_vfs);
1354 }
1355 
1356 void
1357 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1358     uint64_t ctime[2], boolean_t have_tx)
1359 {
1360 	timestruc_t	now;
1361 
1362 	gethrestime(&now);
1363 
1364 	if (have_tx) {	/* will sa_bulk_update happen really soon? */
1365 		zp->z_atime_dirty = 0;
1366 		zp->z_seq++;
1367 	} else {
1368 		zp->z_atime_dirty = 1;
1369 	}
1370 
1371 	if (flag & AT_ATIME) {
1372 		ZFS_TIME_ENCODE(&now, zp->z_atime);
1373 	}
1374 
1375 	if (flag & AT_MTIME) {
1376 		ZFS_TIME_ENCODE(&now, mtime);
1377 		if (zp->z_zfsvfs->z_use_fuids) {
1378 			zp->z_pflags |= (ZFS_ARCHIVE |
1379 			    ZFS_AV_MODIFIED);
1380 		}
1381 	}
1382 
1383 	if (flag & AT_CTIME) {
1384 		ZFS_TIME_ENCODE(&now, ctime);
1385 		if (zp->z_zfsvfs->z_use_fuids)
1386 			zp->z_pflags |= ZFS_ARCHIVE;
1387 	}
1388 }
1389 
1390 /*
1391  * Grow the block size for a file.
1392  *
1393  *	IN:	zp	- znode of file to free data in.
1394  *		size	- requested block size
1395  *		tx	- open transaction.
1396  *
1397  * NOTE: this function assumes that the znode is write locked.
1398  */
1399 void
1400 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1401 {
1402 	int		error;
1403 	u_longlong_t	dummy;
1404 
1405 	if (size <= zp->z_blksz)
1406 		return;
1407 	/*
1408 	 * If the file size is already greater than the current blocksize,
1409 	 * we will not grow.  If there is more than one block in a file,
1410 	 * the blocksize cannot change.
1411 	 */
1412 	if (zp->z_blksz && zp->z_size > zp->z_blksz)
1413 		return;
1414 
1415 	error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1416 	    size, 0, tx);
1417 
1418 	if (error == ENOTSUP)
1419 		return;
1420 	ASSERT0(error);
1421 
1422 	/* What blocksize did we actually get? */
1423 	dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1424 }
1425 
1426 /*
1427  * This is a dummy interface used when pvn_vplist_dirty() should *not*
1428  * be calling back into the fs for a putpage().  E.g.: when truncating
1429  * a file, the pages being "thrown away* don't need to be written out.
1430  */
1431 /* ARGSUSED */
1432 static int
1433 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1434     int flags, cred_t *cr)
1435 {
1436 	ASSERT(0);
1437 	return (0);
1438 }
1439 
1440 /*
1441  * Increase the file length
1442  *
1443  *	IN:	zp	- znode of file to free data in.
1444  *		end	- new end-of-file
1445  *
1446  *	RETURN:	0 on success, error code on failure
1447  */
1448 static int
1449 zfs_extend(znode_t *zp, uint64_t end)
1450 {
1451 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1452 	dmu_tx_t *tx;
1453 	rl_t *rl;
1454 	uint64_t newblksz;
1455 	int error;
1456 
1457 	/*
1458 	 * We will change zp_size, lock the whole file.
1459 	 */
1460 	rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER);
1461 
1462 	/*
1463 	 * Nothing to do if file already at desired length.
1464 	 */
1465 	if (end <= zp->z_size) {
1466 		zfs_range_unlock(rl);
1467 		return (0);
1468 	}
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 			/*
1479 			 * File's blocksize is already larger than the
1480 			 * "recordsize" property.  Only let it grow to
1481 			 * the next power of 2.
1482 			 */
1483 			ASSERT(!ISP2(zp->z_blksz));
1484 			newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1485 		} else {
1486 			newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1487 		}
1488 		dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1489 	} else {
1490 		newblksz = 0;
1491 	}
1492 
1493 	error = dmu_tx_assign(tx, TXG_WAIT);
1494 	if (error) {
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 on success, error code on failure
1523  */
1524 static int
1525 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1526 {
1527 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1528 	rl_t *rl;
1529 	int error;
1530 
1531 	/*
1532 	 * Lock the range being freed.
1533 	 */
1534 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
1535 
1536 	/*
1537 	 * Nothing to do if file already at desired length.
1538 	 */
1539 	if (off >= zp->z_size) {
1540 		zfs_range_unlock(rl);
1541 		return (0);
1542 	}
1543 
1544 	if (off + len > zp->z_size)
1545 		len = zp->z_size - off;
1546 
1547 	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1548 
1549 	zfs_range_unlock(rl);
1550 
1551 	return (error);
1552 }
1553 
1554 /*
1555  * Truncate a file
1556  *
1557  *	IN:	zp	- znode of file to free data in.
1558  *		end	- new end-of-file.
1559  *
1560  *	RETURN:	0 on success, error code on failure
1561  */
1562 static int
1563 zfs_trunc(znode_t *zp, uint64_t end)
1564 {
1565 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1566 	vnode_t *vp = ZTOV(zp);
1567 	dmu_tx_t *tx;
1568 	rl_t *rl;
1569 	int error;
1570 	sa_bulk_attr_t bulk[2];
1571 	int count = 0;
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 	tx = dmu_tx_create(zfsvfs->z_os);
1592 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1593 	zfs_sa_upgrade_txholds(tx, zp);
1594 	dmu_tx_mark_netfree(tx);
1595 	error = dmu_tx_assign(tx, TXG_WAIT);
1596 	if (error) {
1597 		dmu_tx_abort(tx);
1598 		zfs_range_unlock(rl);
1599 		return (error);
1600 	}
1601 
1602 	zp->z_size = end;
1603 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1604 	    NULL, &zp->z_size, sizeof (zp->z_size));
1605 
1606 	if (end == 0) {
1607 		zp->z_pflags &= ~ZFS_SPARSE;
1608 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1609 		    NULL, &zp->z_pflags, 8);
1610 	}
1611 	VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1612 
1613 	dmu_tx_commit(tx);
1614 
1615 	/*
1616 	 * Clear any mapped pages in the truncated region.  This has to
1617 	 * happen outside of the transaction to avoid the possibility of
1618 	 * a deadlock with someone trying to push a page that we are
1619 	 * about to invalidate.
1620 	 */
1621 	if (vn_has_cached_data(vp)) {
1622 		page_t *pp;
1623 		uint64_t start = end & PAGEMASK;
1624 		int poff = end & PAGEOFFSET;
1625 
1626 		if (poff != 0 && (pp = page_lookup(vp, start, SE_SHARED))) {
1627 			/*
1628 			 * We need to zero a partial page.
1629 			 */
1630 			pagezero(pp, poff, PAGESIZE - poff);
1631 			start += PAGESIZE;
1632 			page_unlock(pp);
1633 		}
1634 		error = pvn_vplist_dirty(vp, start, zfs_no_putpage,
1635 		    B_INVAL | B_TRUNC, NULL);
1636 		ASSERT(error == 0);
1637 	}
1638 
1639 	zfs_range_unlock(rl);
1640 
1641 	return (0);
1642 }
1643 
1644 /*
1645  * Free space in a file
1646  *
1647  *	IN:	zp	- znode of file to free data in.
1648  *		off	- start of range
1649  *		len	- end of range (0 => EOF)
1650  *		flag	- current file open mode flags.
1651  *		log	- TRUE if this action should be logged
1652  *
1653  *	RETURN:	0 on success, error code on 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_WAIT);
1704 	if (error) {
1705 		dmu_tx_abort(tx);
1706 		return (error);
1707 	}
1708 
1709 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1710 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1711 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1712 	    NULL, &zp->z_pflags, 8);
1713 	zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1714 	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1715 	ASSERT(error == 0);
1716 
1717 	zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1718 
1719 	dmu_tx_commit(tx);
1720 	return (0);
1721 }
1722 
1723 void
1724 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1725 {
1726 	uint64_t	moid, obj, sa_obj, version;
1727 	uint64_t	sense = ZFS_CASE_SENSITIVE;
1728 	uint64_t	norm = 0;
1729 	nvpair_t	*elem;
1730 	int		error;
1731 	int		i;
1732 	znode_t		*rootzp = NULL;
1733 	zfsvfs_t	*zfsvfs;
1734 	vnode_t		*vp;
1735 	vattr_t		vattr;
1736 	znode_t		*zp;
1737 	zfs_acl_ids_t	acl_ids;
1738 
1739 	/*
1740 	 * First attempt to create master node.
1741 	 */
1742 	/*
1743 	 * In an empty objset, there are no blocks to read and thus
1744 	 * there can be no i/o errors (which we assert below).
1745 	 */
1746 	moid = MASTER_NODE_OBJ;
1747 	error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1748 	    DMU_OT_NONE, 0, tx);
1749 	ASSERT(error == 0);
1750 
1751 	/*
1752 	 * Set starting attributes.
1753 	 */
1754 	version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1755 	elem = NULL;
1756 	while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1757 		/* For the moment we expect all zpl props to be uint64_ts */
1758 		uint64_t val;
1759 		char *name;
1760 
1761 		ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1762 		VERIFY(nvpair_value_uint64(elem, &val) == 0);
1763 		name = nvpair_name(elem);
1764 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1765 			if (val < version)
1766 				version = val;
1767 		} else {
1768 			error = zap_update(os, moid, name, 8, 1, &val, tx);
1769 		}
1770 		ASSERT(error == 0);
1771 		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1772 			norm = val;
1773 		else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1774 			sense = val;
1775 	}
1776 	ASSERT(version != 0);
1777 	error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1778 
1779 	/*
1780 	 * Create zap object used for SA attribute registration
1781 	 */
1782 
1783 	if (version >= ZPL_VERSION_SA) {
1784 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1785 		    DMU_OT_NONE, 0, tx);
1786 		error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1787 		ASSERT(error == 0);
1788 	} else {
1789 		sa_obj = 0;
1790 	}
1791 	/*
1792 	 * Create a delete queue.
1793 	 */
1794 	obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1795 
1796 	error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1797 	ASSERT(error == 0);
1798 
1799 	/*
1800 	 * Create root znode.  Create minimal znode/vnode/zfsvfs
1801 	 * to allow zfs_mknode to work.
1802 	 */
1803 	vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1804 	vattr.va_type = VDIR;
1805 	vattr.va_mode = S_IFDIR|0755;
1806 	vattr.va_uid = crgetuid(cr);
1807 	vattr.va_gid = crgetgid(cr);
1808 
1809 	rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1810 	ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1811 	rootzp->z_moved = 0;
1812 	rootzp->z_unlinked = 0;
1813 	rootzp->z_atime_dirty = 0;
1814 	rootzp->z_is_sa = USE_SA(version, os);
1815 
1816 	vp = ZTOV(rootzp);
1817 	vn_reinit(vp);
1818 	vp->v_type = VDIR;
1819 
1820 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1821 	zfsvfs->z_os = os;
1822 	zfsvfs->z_parent = zfsvfs;
1823 	zfsvfs->z_version = version;
1824 	zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1825 	zfsvfs->z_use_sa = USE_SA(version, os);
1826 	zfsvfs->z_norm = norm;
1827 
1828 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1829 	    &zfsvfs->z_attr_table);
1830 
1831 	ASSERT(error == 0);
1832 
1833 	/*
1834 	 * Fold case on file systems that are always or sometimes case
1835 	 * insensitive.
1836 	 */
1837 	if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1838 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1839 
1840 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1841 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1842 	    offsetof(znode_t, z_link_node));
1843 
1844 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1845 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1846 
1847 	rootzp->z_zfsvfs = zfsvfs;
1848 	VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1849 	    cr, NULL, &acl_ids));
1850 	zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1851 	ASSERT3P(zp, ==, rootzp);
1852 	ASSERT(!vn_in_dnlc(ZTOV(rootzp))); /* not valid to move */
1853 	error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1854 	ASSERT(error == 0);
1855 	zfs_acl_ids_free(&acl_ids);
1856 	POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1857 
1858 	ZTOV(rootzp)->v_count = 0;
1859 	sa_handle_destroy(rootzp->z_sa_hdl);
1860 	kmem_cache_free(znode_cache, rootzp);
1861 
1862 	/*
1863 	 * Create shares directory
1864 	 */
1865 
1866 	error = zfs_create_share_dir(zfsvfs, tx);
1867 
1868 	ASSERT(error == 0);
1869 
1870 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1871 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1872 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1873 }
1874 
1875 #endif /* _KERNEL */
1876 
1877 static int
1878 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1879 {
1880 	uint64_t sa_obj = 0;
1881 	int error;
1882 
1883 	error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1884 	if (error != 0 && error != ENOENT)
1885 		return (error);
1886 
1887 	error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1888 	return (error);
1889 }
1890 
1891 static int
1892 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1893     dmu_buf_t **db, void *tag)
1894 {
1895 	dmu_object_info_t doi;
1896 	int error;
1897 
1898 	if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
1899 		return (error);
1900 
1901 	dmu_object_info_from_db(*db, &doi);
1902 	if ((doi.doi_bonus_type != DMU_OT_SA &&
1903 	    doi.doi_bonus_type != DMU_OT_ZNODE) ||
1904 	    doi.doi_bonus_type == DMU_OT_ZNODE &&
1905 	    doi.doi_bonus_size < sizeof (znode_phys_t)) {
1906 		sa_buf_rele(*db, tag);
1907 		return (SET_ERROR(ENOTSUP));
1908 	}
1909 
1910 	error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
1911 	if (error != 0) {
1912 		sa_buf_rele(*db, tag);
1913 		return (error);
1914 	}
1915 
1916 	return (0);
1917 }
1918 
1919 void
1920 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
1921 {
1922 	sa_handle_destroy(hdl);
1923 	sa_buf_rele(db, tag);
1924 }
1925 
1926 /*
1927  * Given an object number, return its parent object number and whether
1928  * or not the object is an extended attribute directory.
1929  */
1930 static int
1931 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
1932     uint64_t *pobjp, int *is_xattrdir)
1933 {
1934 	uint64_t parent;
1935 	uint64_t pflags;
1936 	uint64_t mode;
1937 	uint64_t parent_mode;
1938 	sa_bulk_attr_t bulk[3];
1939 	sa_handle_t *sa_hdl;
1940 	dmu_buf_t *sa_db;
1941 	int count = 0;
1942 	int error;
1943 
1944 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
1945 	    &parent, sizeof (parent));
1946 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
1947 	    &pflags, sizeof (pflags));
1948 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1949 	    &mode, sizeof (mode));
1950 
1951 	if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
1952 		return (error);
1953 
1954 	/*
1955 	 * When a link is removed its parent pointer is not changed and will
1956 	 * be invalid.  There are two cases where a link is removed but the
1957 	 * file stays around, when it goes to the delete queue and when there
1958 	 * are additional links.
1959 	 */
1960 	error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
1961 	if (error != 0)
1962 		return (error);
1963 
1964 	error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
1965 	zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
1966 	if (error != 0)
1967 		return (error);
1968 
1969 	*is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
1970 
1971 	/*
1972 	 * Extended attributes can be applied to files, directories, etc.
1973 	 * Otherwise the parent must be a directory.
1974 	 */
1975 	if (!*is_xattrdir && !S_ISDIR(parent_mode))
1976 		return (SET_ERROR(EINVAL));
1977 
1978 	*pobjp = parent;
1979 
1980 	return (0);
1981 }
1982 
1983 /*
1984  * Given an object number, return some zpl level statistics
1985  */
1986 static int
1987 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
1988     zfs_stat_t *sb)
1989 {
1990 	sa_bulk_attr_t bulk[4];
1991 	int count = 0;
1992 
1993 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
1994 	    &sb->zs_mode, sizeof (sb->zs_mode));
1995 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
1996 	    &sb->zs_gen, sizeof (sb->zs_gen));
1997 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
1998 	    &sb->zs_links, sizeof (sb->zs_links));
1999 	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2000 	    &sb->zs_ctime, sizeof (sb->zs_ctime));
2001 
2002 	return (sa_bulk_lookup(hdl, bulk, count));
2003 }
2004 
2005 static int
2006 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2007     sa_attr_type_t *sa_table, char *buf, int len)
2008 {
2009 	sa_handle_t *sa_hdl;
2010 	sa_handle_t *prevhdl = NULL;
2011 	dmu_buf_t *prevdb = NULL;
2012 	dmu_buf_t *sa_db = NULL;
2013 	char *path = buf + len - 1;
2014 	int error;
2015 
2016 	*path = '\0';
2017 	sa_hdl = hdl;
2018 
2019 	for (;;) {
2020 		uint64_t pobj;
2021 		char component[MAXNAMELEN + 2];
2022 		size_t complen;
2023 		int is_xattrdir;
2024 
2025 		if (prevdb)
2026 			zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2027 
2028 		if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2029 		    &is_xattrdir)) != 0)
2030 			break;
2031 
2032 		if (pobj == obj) {
2033 			if (path[0] != '/')
2034 				*--path = '/';
2035 			break;
2036 		}
2037 
2038 		component[0] = '/';
2039 		if (is_xattrdir) {
2040 			(void) sprintf(component + 1, "<xattrdir>");
2041 		} else {
2042 			error = zap_value_search(osp, pobj, obj,
2043 			    ZFS_DIRENT_OBJ(-1ULL), component + 1);
2044 			if (error != 0)
2045 				break;
2046 		}
2047 
2048 		complen = strlen(component);
2049 		path -= complen;
2050 		ASSERT(path >= buf);
2051 		bcopy(component, path, complen);
2052 		obj = pobj;
2053 
2054 		if (sa_hdl != hdl) {
2055 			prevhdl = sa_hdl;
2056 			prevdb = sa_db;
2057 		}
2058 		error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2059 		if (error != 0) {
2060 			sa_hdl = prevhdl;
2061 			sa_db = prevdb;
2062 			break;
2063 		}
2064 	}
2065 
2066 	if (sa_hdl != NULL && sa_hdl != hdl) {
2067 		ASSERT(sa_db != NULL);
2068 		zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2069 	}
2070 
2071 	if (error == 0)
2072 		(void) memmove(buf, path, buf + len - path);
2073 
2074 	return (error);
2075 }
2076 
2077 int
2078 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2079 {
2080 	sa_attr_type_t *sa_table;
2081 	sa_handle_t *hdl;
2082 	dmu_buf_t *db;
2083 	int error;
2084 
2085 	error = zfs_sa_setup(osp, &sa_table);
2086 	if (error != 0)
2087 		return (error);
2088 
2089 	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2090 	if (error != 0)
2091 		return (error);
2092 
2093 	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2094 
2095 	zfs_release_sa_handle(hdl, db, FTAG);
2096 	return (error);
2097 }
2098 
2099 int
2100 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2101     char *buf, int len)
2102 {
2103 	char *path = buf + len - 1;
2104 	sa_attr_type_t *sa_table;
2105 	sa_handle_t *hdl;
2106 	dmu_buf_t *db;
2107 	int error;
2108 
2109 	*path = '\0';
2110 
2111 	error = zfs_sa_setup(osp, &sa_table);
2112 	if (error != 0)
2113 		return (error);
2114 
2115 	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2116 	if (error != 0)
2117 		return (error);
2118 
2119 	error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2120 	if (error != 0) {
2121 		zfs_release_sa_handle(hdl, db, FTAG);
2122 		return (error);
2123 	}
2124 
2125 	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2126 
2127 	zfs_release_sa_handle(hdl, db, FTAG);
2128 	return (error);
2129 }
2130