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, 2018 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Integros [integros.com]
25 */
26
27/* Portions Copyright 2007 Jeremy Teo */
28
29#ifdef _KERNEL
30#include <sys/types.h>
31#include <sys/param.h>
32#include <sys/time.h>
33#include <sys/systm.h>
34#include <sys/sysmacros.h>
35#include <sys/resource.h>
36#include <sys/mntent.h>
37#include <sys/mkdev.h>
38#include <sys/u8_textprep.h>
39#include <sys/dsl_dataset.h>
40#include <sys/vfs.h>
41#include <sys/vfs_opreg.h>
42#include <sys/vnode.h>
43#include <sys/file.h>
44#include <sys/kmem.h>
45#include <sys/errno.h>
46#include <sys/unistd.h>
47#include <sys/mode.h>
48#include <sys/atomic.h>
49#include <vm/pvn.h>
50#include "fs/fs_subr.h"
51#include <sys/zfs_dir.h>
52#include <sys/zfs_acl.h>
53#include <sys/zfs_ioctl.h>
54#include <sys/zfs_rlock.h>
55#include <sys/zfs_fuid.h>
56#include <sys/dnode.h>
57#include <sys/fs/zfs.h>
58#include <sys/kidmap.h>
59#endif /* _KERNEL */
60
61#include <sys/dmu.h>
62#include <sys/dmu_objset.h>
63#include <sys/dmu_tx.h>
64#include <sys/refcount.h>
65#include <sys/stat.h>
66#include <sys/zap.h>
67#include <sys/zfs_znode.h>
68#include <sys/sa.h>
69#include <sys/zfs_sa.h>
70#include <sys/zfs_stat.h>
71
72#include "zfs_prop.h"
73#include "zfs_comutil.h"
74
75/*
76 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only
77 * turned on when DEBUG is also defined.
78 */
79#ifdef	DEBUG
80#define	ZNODE_STATS
81#endif	/* DEBUG */
82
83#ifdef	ZNODE_STATS
84#define	ZNODE_STAT_ADD(stat)			((stat)++)
85#else
86#define	ZNODE_STAT_ADD(stat)			/* nothing */
87#endif	/* ZNODE_STATS */
88
89/*
90 * Functions needed for userland (ie: libzpool) are not put under
91 * #ifdef_KERNEL; the rest of the functions have dependencies
92 * (such as VFS logic) that will not compile easily in userland.
93 */
94#ifdef _KERNEL
95/*
96 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to
97 * be freed before it can be safely accessed.
98 */
99krwlock_t zfsvfs_lock;
100
101static kmem_cache_t *znode_cache = NULL;
102
103/*ARGSUSED*/
104static void
105znode_evict_error(dmu_buf_t *dbuf, void *user_ptr)
106{
107	/*
108	 * We should never drop all dbuf refs without first clearing
109	 * the eviction callback.
110	 */
111	panic("evicting znode %p\n", user_ptr);
112}
113
114/*
115 * This callback is invoked when acquiring a RL_WRITER or RL_APPEND lock on
116 * z_rangelock. It will modify the offset and length of the lock to reflect
117 * znode-specific information, and convert RL_APPEND to RL_WRITER.  This is
118 * called with the rangelock_t's rl_lock held, which avoids races.
119 */
120static void
121zfs_rangelock_cb(locked_range_t *new, void *arg)
122{
123	znode_t *zp = arg;
124
125	/*
126	 * If in append mode, convert to writer and lock starting at the
127	 * current end of file.
128	 */
129	if (new->lr_type == RL_APPEND) {
130		new->lr_offset = zp->z_size;
131		new->lr_type = RL_WRITER;
132	}
133
134	/*
135	 * If we need to grow the block size then lock the whole file range.
136	 */
137	uint64_t end_size = MAX(zp->z_size, new->lr_offset + new->lr_length);
138	if (end_size > zp->z_blksz && (!ISP2(zp->z_blksz) ||
139	    zp->z_blksz < zp->z_zfsvfs->z_max_blksz)) {
140		new->lr_offset = 0;
141		new->lr_length = UINT64_MAX;
142	}
143}
144
145/*ARGSUSED*/
146static int
147zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
148{
149	znode_t *zp = buf;
150
151	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
152
153	zp->z_vnode = vn_alloc(kmflags);
154	if (zp->z_vnode == NULL) {
155		return (-1);
156	}
157	ZTOV(zp)->v_data = zp;
158
159	list_link_init(&zp->z_link_node);
160
161	mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL);
162	rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL);
163	rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL);
164	mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL);
165
166	rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp);
167
168	zp->z_dirlocks = NULL;
169	zp->z_acl_cached = NULL;
170	zp->z_moved = 0;
171	return (0);
172}
173
174/*ARGSUSED*/
175static void
176zfs_znode_cache_destructor(void *buf, void *arg)
177{
178	znode_t *zp = buf;
179
180	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
181	ASSERT(ZTOV(zp)->v_data == zp);
182	vn_free(ZTOV(zp));
183	ASSERT(!list_link_active(&zp->z_link_node));
184	mutex_destroy(&zp->z_lock);
185	rw_destroy(&zp->z_parent_lock);
186	rw_destroy(&zp->z_name_lock);
187	mutex_destroy(&zp->z_acl_lock);
188	rangelock_fini(&zp->z_rangelock);
189
190	ASSERT(zp->z_dirlocks == NULL);
191	ASSERT(zp->z_acl_cached == NULL);
192}
193
194#ifdef	ZNODE_STATS
195static struct {
196	uint64_t zms_zfsvfs_invalid;
197	uint64_t zms_zfsvfs_recheck1;
198	uint64_t zms_zfsvfs_unmounted;
199	uint64_t zms_zfsvfs_recheck2;
200	uint64_t zms_obj_held;
201	uint64_t zms_vnode_locked;
202	uint64_t zms_not_only_dnlc;
203} znode_move_stats;
204#endif	/* ZNODE_STATS */
205
206static void
207zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
208{
209	vnode_t *vp;
210
211	/* Copy fields. */
212	nzp->z_zfsvfs = ozp->z_zfsvfs;
213
214	/* Swap vnodes. */
215	vp = nzp->z_vnode;
216	nzp->z_vnode = ozp->z_vnode;
217	ozp->z_vnode = vp; /* let destructor free the overwritten vnode */
218	ZTOV(ozp)->v_data = ozp;
219	ZTOV(nzp)->v_data = nzp;
220
221	nzp->z_id = ozp->z_id;
222	ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */
223	nzp->z_unlinked = ozp->z_unlinked;
224	nzp->z_atime_dirty = ozp->z_atime_dirty;
225	nzp->z_zn_prefetch = ozp->z_zn_prefetch;
226	nzp->z_blksz = ozp->z_blksz;
227	nzp->z_seq = ozp->z_seq;
228	nzp->z_mapcnt = ozp->z_mapcnt;
229	nzp->z_gen = ozp->z_gen;
230	nzp->z_sync_cnt = ozp->z_sync_cnt;
231	nzp->z_is_sa = ozp->z_is_sa;
232	nzp->z_sa_hdl = ozp->z_sa_hdl;
233	bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2);
234	nzp->z_links = ozp->z_links;
235	nzp->z_size = ozp->z_size;
236	nzp->z_pflags = ozp->z_pflags;
237	nzp->z_uid = ozp->z_uid;
238	nzp->z_gid = ozp->z_gid;
239	nzp->z_mode = ozp->z_mode;
240
241	/*
242	 * Since this is just an idle znode and kmem is already dealing with
243	 * memory pressure, release any cached ACL.
244	 */
245	if (ozp->z_acl_cached) {
246		zfs_acl_free(ozp->z_acl_cached);
247		ozp->z_acl_cached = NULL;
248	}
249
250	sa_set_userp(nzp->z_sa_hdl, nzp);
251
252	/*
253	 * Invalidate the original znode by clearing fields that provide a
254	 * pointer back to the znode. Set the low bit of the vfs pointer to
255	 * ensure that zfs_znode_move() recognizes the znode as invalid in any
256	 * subsequent callback.
257	 */
258	ozp->z_sa_hdl = NULL;
259	POINTER_INVALIDATE(&ozp->z_zfsvfs);
260
261	/*
262	 * Mark the znode.
263	 */
264	nzp->z_moved = 1;
265	ozp->z_moved = (uint8_t)-1;
266}
267
268/*ARGSUSED*/
269static kmem_cbrc_t
270zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg)
271{
272	znode_t *ozp = buf, *nzp = newbuf;
273	zfsvfs_t *zfsvfs;
274	vnode_t *vp;
275
276	/*
277	 * The znode is on the file system's list of known znodes if the vfs
278	 * pointer is valid. We set the low bit of the vfs pointer when freeing
279	 * the znode to invalidate it, and the memory patterns written by kmem
280	 * (baddcafe and deadbeef) set at least one of the two low bits. A newly
281	 * created znode sets the vfs pointer last of all to indicate that the
282	 * znode is known and in a valid state to be moved by this function.
283	 */
284	zfsvfs = ozp->z_zfsvfs;
285	if (!POINTER_IS_VALID(zfsvfs)) {
286		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid);
287		return (KMEM_CBRC_DONT_KNOW);
288	}
289
290	/*
291	 * Close a small window in which it's possible that the filesystem could
292	 * be unmounted and freed, and zfsvfs, though valid in the previous
293	 * statement, could point to unrelated memory by the time we try to
294	 * prevent the filesystem from being unmounted.
295	 */
296	rw_enter(&zfsvfs_lock, RW_WRITER);
297	if (zfsvfs != ozp->z_zfsvfs) {
298		rw_exit(&zfsvfs_lock);
299		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1);
300		return (KMEM_CBRC_DONT_KNOW);
301	}
302
303	/*
304	 * If the znode is still valid, then so is the file system. We know that
305	 * no valid file system can be freed while we hold zfsvfs_lock, so we
306	 * can safely ensure that the filesystem is not and will not be
307	 * unmounted. The next statement is equivalent to ZFS_ENTER().
308	 */
309	rrm_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG);
310	if (zfsvfs->z_unmounted) {
311		ZFS_EXIT(zfsvfs);
312		rw_exit(&zfsvfs_lock);
313		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted);
314		return (KMEM_CBRC_DONT_KNOW);
315	}
316	rw_exit(&zfsvfs_lock);
317
318	mutex_enter(&zfsvfs->z_znodes_lock);
319	/*
320	 * Recheck the vfs pointer in case the znode was removed just before
321	 * acquiring the lock.
322	 */
323	if (zfsvfs != ozp->z_zfsvfs) {
324		mutex_exit(&zfsvfs->z_znodes_lock);
325		ZFS_EXIT(zfsvfs);
326		ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2);
327		return (KMEM_CBRC_DONT_KNOW);
328	}
329
330	/*
331	 * At this point we know that as long as we hold z_znodes_lock, the
332	 * znode cannot be freed and fields within the znode can be safely
333	 * accessed. Now, prevent a race with zfs_zget().
334	 */
335	if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) {
336		mutex_exit(&zfsvfs->z_znodes_lock);
337		ZFS_EXIT(zfsvfs);
338		ZNODE_STAT_ADD(znode_move_stats.zms_obj_held);
339		return (KMEM_CBRC_LATER);
340	}
341
342	vp = ZTOV(ozp);
343	if (mutex_tryenter(&vp->v_lock) == 0) {
344		ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
345		mutex_exit(&zfsvfs->z_znodes_lock);
346		ZFS_EXIT(zfsvfs);
347		ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked);
348		return (KMEM_CBRC_LATER);
349	}
350
351	/* Only move znodes that are referenced _only_ by the DNLC. */
352	if (vp->v_count != 1 || !vn_in_dnlc(vp)) {
353		mutex_exit(&vp->v_lock);
354		ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
355		mutex_exit(&zfsvfs->z_znodes_lock);
356		ZFS_EXIT(zfsvfs);
357		ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc);
358		return (KMEM_CBRC_LATER);
359	}
360
361	/*
362	 * The znode is known and in a valid state to move. We're holding the
363	 * locks needed to execute the critical section.
364	 */
365	zfs_znode_move_impl(ozp, nzp);
366	mutex_exit(&vp->v_lock);
367	ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id);
368
369	list_link_replace(&ozp->z_link_node, &nzp->z_link_node);
370	mutex_exit(&zfsvfs->z_znodes_lock);
371	ZFS_EXIT(zfsvfs);
372
373	return (KMEM_CBRC_YES);
374}
375
376void
377zfs_znode_init(void)
378{
379	/*
380	 * Initialize zcache
381	 */
382	rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL);
383	ASSERT(znode_cache == NULL);
384	znode_cache = kmem_cache_create("zfs_znode_cache",
385	    sizeof (znode_t), 0, zfs_znode_cache_constructor,
386	    zfs_znode_cache_destructor, NULL, NULL, NULL, 0);
387	kmem_cache_set_move(znode_cache, zfs_znode_move);
388}
389
390void
391zfs_znode_fini(void)
392{
393	/*
394	 * Cleanup vfs & vnode ops
395	 */
396	zfs_remove_op_tables();
397
398	/*
399	 * Cleanup zcache
400	 */
401	if (znode_cache)
402		kmem_cache_destroy(znode_cache);
403	znode_cache = NULL;
404	rw_destroy(&zfsvfs_lock);
405}
406
407struct vnodeops *zfs_dvnodeops;
408struct vnodeops *zfs_fvnodeops;
409struct vnodeops *zfs_symvnodeops;
410struct vnodeops *zfs_xdvnodeops;
411struct vnodeops *zfs_evnodeops;
412struct vnodeops *zfs_sharevnodeops;
413
414void
415zfs_remove_op_tables()
416{
417	/*
418	 * Remove vfs ops
419	 */
420	ASSERT(zfsfstype);
421	(void) vfs_freevfsops_by_type(zfsfstype);
422	zfsfstype = 0;
423
424	/*
425	 * Remove vnode ops
426	 */
427	if (zfs_dvnodeops)
428		vn_freevnodeops(zfs_dvnodeops);
429	if (zfs_fvnodeops)
430		vn_freevnodeops(zfs_fvnodeops);
431	if (zfs_symvnodeops)
432		vn_freevnodeops(zfs_symvnodeops);
433	if (zfs_xdvnodeops)
434		vn_freevnodeops(zfs_xdvnodeops);
435	if (zfs_evnodeops)
436		vn_freevnodeops(zfs_evnodeops);
437	if (zfs_sharevnodeops)
438		vn_freevnodeops(zfs_sharevnodeops);
439
440	zfs_dvnodeops = NULL;
441	zfs_fvnodeops = NULL;
442	zfs_symvnodeops = NULL;
443	zfs_xdvnodeops = NULL;
444	zfs_evnodeops = NULL;
445	zfs_sharevnodeops = NULL;
446}
447
448extern const fs_operation_def_t zfs_dvnodeops_template[];
449extern const fs_operation_def_t zfs_fvnodeops_template[];
450extern const fs_operation_def_t zfs_xdvnodeops_template[];
451extern const fs_operation_def_t zfs_symvnodeops_template[];
452extern const fs_operation_def_t zfs_evnodeops_template[];
453extern const fs_operation_def_t zfs_sharevnodeops_template[];
454
455int
456zfs_create_op_tables()
457{
458	int error;
459
460	/*
461	 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs()
462	 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv).
463	 * In this case we just return as the ops vectors are already set up.
464	 */
465	if (zfs_dvnodeops)
466		return (0);
467
468	error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template,
469	    &zfs_dvnodeops);
470	if (error)
471		return (error);
472
473	error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template,
474	    &zfs_fvnodeops);
475	if (error)
476		return (error);
477
478	error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template,
479	    &zfs_symvnodeops);
480	if (error)
481		return (error);
482
483	error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template,
484	    &zfs_xdvnodeops);
485	if (error)
486		return (error);
487
488	error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template,
489	    &zfs_evnodeops);
490	if (error)
491		return (error);
492
493	error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template,
494	    &zfs_sharevnodeops);
495
496	return (error);
497}
498
499int
500zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx)
501{
502	zfs_acl_ids_t acl_ids;
503	vattr_t vattr;
504	znode_t *sharezp;
505	vnode_t *vp;
506	znode_t *zp;
507	int error;
508
509	vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
510	vattr.va_type = VDIR;
511	vattr.va_mode = S_IFDIR|0555;
512	vattr.va_uid = crgetuid(kcred);
513	vattr.va_gid = crgetgid(kcred);
514
515	sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP);
516	ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs));
517	sharezp->z_moved = 0;
518	sharezp->z_unlinked = 0;
519	sharezp->z_atime_dirty = 0;
520	sharezp->z_zfsvfs = zfsvfs;
521	sharezp->z_is_sa = zfsvfs->z_use_sa;
522	sharezp->z_pflags = 0;
523
524	vp = ZTOV(sharezp);
525	vn_reinit(vp);
526	vp->v_type = VDIR;
527
528	VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr,
529	    kcred, NULL, &acl_ids));
530	zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids);
531	ASSERT3P(zp, ==, sharezp);
532	ASSERT(!vn_in_dnlc(ZTOV(sharezp))); /* not valid to move */
533	POINTER_INVALIDATE(&sharezp->z_zfsvfs);
534	error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
535	    ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx);
536	zfsvfs->z_shares_dir = sharezp->z_id;
537
538	zfs_acl_ids_free(&acl_ids);
539	ZTOV(sharezp)->v_count = 0;
540	sa_handle_destroy(sharezp->z_sa_hdl);
541	kmem_cache_free(znode_cache, sharezp);
542
543	return (error);
544}
545
546/*
547 * define a couple of values we need available
548 * for both 64 and 32 bit environments.
549 */
550#ifndef NBITSMINOR64
551#define	NBITSMINOR64	32
552#endif
553#ifndef MAXMAJ64
554#define	MAXMAJ64	0xffffffffUL
555#endif
556#ifndef	MAXMIN64
557#define	MAXMIN64	0xffffffffUL
558#endif
559
560/*
561 * Create special expldev for ZFS private use.
562 * Can't use standard expldev since it doesn't do
563 * what we want.  The standard expldev() takes a
564 * dev32_t in LP64 and expands it to a long dev_t.
565 * We need an interface that takes a dev32_t in ILP32
566 * and expands it to a long dev_t.
567 */
568static uint64_t
569zfs_expldev(dev_t dev)
570{
571#ifndef _LP64
572	major_t major = (major_t)dev >> NBITSMINOR32 & MAXMAJ32;
573	return (((uint64_t)major << NBITSMINOR64) |
574	    ((minor_t)dev & MAXMIN32));
575#else
576	return (dev);
577#endif
578}
579
580/*
581 * Special cmpldev for ZFS private use.
582 * Can't use standard cmpldev since it takes
583 * a long dev_t and compresses it to dev32_t in
584 * LP64.  We need to do a compaction of a long dev_t
585 * to a dev32_t in ILP32.
586 */
587dev_t
588zfs_cmpldev(uint64_t dev)
589{
590#ifndef _LP64
591	minor_t minor = (minor_t)dev & MAXMIN64;
592	major_t major = (major_t)(dev >> NBITSMINOR64) & MAXMAJ64;
593
594	if (major > MAXMAJ32 || minor > MAXMIN32)
595		return (NODEV32);
596
597	return (((dev32_t)major << NBITSMINOR32) | minor);
598#else
599	return (dev);
600#endif
601}
602
603static void
604zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp,
605    dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl)
606{
607	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs));
608	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id)));
609
610	mutex_enter(&zp->z_lock);
611
612	ASSERT(zp->z_sa_hdl == NULL);
613	ASSERT(zp->z_acl_cached == NULL);
614	if (sa_hdl == NULL) {
615		VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp,
616		    SA_HDL_SHARED, &zp->z_sa_hdl));
617	} else {
618		zp->z_sa_hdl = sa_hdl;
619		sa_set_userp(sa_hdl, zp);
620	}
621
622	zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE;
623
624	/*
625	 * Slap on VROOT if we are the root znode
626	 */
627	if (zp->z_id == zfsvfs->z_root)
628		ZTOV(zp)->v_flag |= VROOT;
629
630	mutex_exit(&zp->z_lock);
631	vn_exists(ZTOV(zp));
632}
633
634void
635zfs_znode_dmu_fini(znode_t *zp)
636{
637	ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) ||
638	    zp->z_unlinked ||
639	    RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock));
640
641	sa_handle_destroy(zp->z_sa_hdl);
642	zp->z_sa_hdl = NULL;
643}
644
645/*
646 * Construct a new znode/vnode and intialize.
647 *
648 * This does not do a call to dmu_set_user() that is
649 * up to the caller to do, in case you don't want to
650 * return the znode
651 */
652static znode_t *
653zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
654    dmu_object_type_t obj_type, sa_handle_t *hdl)
655{
656	znode_t	*zp;
657	vnode_t *vp;
658	uint64_t mode;
659	uint64_t parent;
660	uint64_t projid = ZFS_DEFAULT_PROJID;
661	sa_bulk_attr_t bulk[11];
662	int count = 0;
663
664	zp = kmem_cache_alloc(znode_cache, KM_SLEEP);
665
666	ASSERT(zp->z_dirlocks == NULL);
667	ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs));
668	zp->z_moved = 0;
669
670	/*
671	 * Defer setting z_zfsvfs until the znode is ready to be a candidate for
672	 * the zfs_znode_move() callback.
673	 */
674	zp->z_sa_hdl = NULL;
675	zp->z_unlinked = 0;
676	zp->z_atime_dirty = 0;
677	zp->z_mapcnt = 0;
678	zp->z_id = db->db_object;
679	zp->z_blksz = blksz;
680	zp->z_seq = 0x7A4653;
681	zp->z_sync_cnt = 0;
682
683	vp = ZTOV(zp);
684	vn_reinit(vp);
685
686	zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl);
687
688	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
689	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8);
690	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
691	    &zp->z_size, 8);
692	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
693	    &zp->z_links, 8);
694	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
695	    &zp->z_pflags, 8);
696	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8);
697	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
698	    &zp->z_atime, 16);
699	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
700	    &zp->z_uid, 8);
701	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
702	    &zp->z_gid, 8);
703
704	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0 ||
705	    (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
706	    (zp->z_pflags & ZFS_PROJID) &&
707	    sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) {
708		if (hdl == NULL)
709			sa_handle_destroy(zp->z_sa_hdl);
710		kmem_cache_free(znode_cache, zp);
711		return (NULL);
712	}
713
714	zp->z_projid = projid;
715	zp->z_mode = mode;
716	vp->v_vfsp = zfsvfs->z_parent->z_vfs;
717
718	vp->v_type = IFTOVT((mode_t)mode);
719
720	switch (vp->v_type) {
721	case VDIR:
722		if (zp->z_pflags & ZFS_XATTR) {
723			vn_setops(vp, zfs_xdvnodeops);
724			vp->v_flag |= V_XATTRDIR;
725		} else {
726			vn_setops(vp, zfs_dvnodeops);
727		}
728		zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */
729		break;
730	case VBLK:
731	case VCHR:
732		{
733			uint64_t rdev;
734			VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs),
735			    &rdev, sizeof (rdev)) == 0);
736
737			vp->v_rdev = zfs_cmpldev(rdev);
738		}
739		/*FALLTHROUGH*/
740	case VFIFO:
741	case VSOCK:
742	case VDOOR:
743		vn_setops(vp, zfs_fvnodeops);
744		break;
745	case VREG:
746		vp->v_flag |= VMODSORT;
747		if (parent == zfsvfs->z_shares_dir) {
748			ASSERT(zp->z_uid == 0 && zp->z_gid == 0);
749			vn_setops(vp, zfs_sharevnodeops);
750		} else {
751			vn_setops(vp, zfs_fvnodeops);
752		}
753		break;
754	case VLNK:
755		vn_setops(vp, zfs_symvnodeops);
756		break;
757	default:
758		vn_setops(vp, zfs_evnodeops);
759		break;
760	}
761
762	mutex_enter(&zfsvfs->z_znodes_lock);
763	list_insert_tail(&zfsvfs->z_all_znodes, zp);
764	membar_producer();
765	/*
766	 * Everything else must be valid before assigning z_zfsvfs makes the
767	 * znode eligible for zfs_znode_move().
768	 */
769	zp->z_zfsvfs = zfsvfs;
770	mutex_exit(&zfsvfs->z_znodes_lock);
771
772	VFS_HOLD(zfsvfs->z_vfs);
773	return (zp);
774}
775
776static uint64_t empty_xattr;
777static uint64_t pad[4];
778static zfs_acl_phys_t acl_phys;
779/*
780 * Create a new DMU object to hold a zfs znode.
781 *
782 *	IN:	dzp	- parent directory for new znode
783 *		vap	- file attributes for new znode
784 *		tx	- dmu transaction id for zap operations
785 *		cr	- credentials of caller
786 *		flag	- flags:
787 *			  IS_ROOT_NODE	- new object will be root
788 *			  IS_XATTR	- new object is an attribute
789 *		bonuslen - length of bonus buffer
790 *		setaclp  - File/Dir initial ACL
791 *		fuidp	 - Tracks fuid allocation.
792 *
793 *	OUT:	zpp	- allocated znode
794 *
795 */
796void
797zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
798    uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids)
799{
800	uint64_t	crtime[2], atime[2], mtime[2], ctime[2];
801	uint64_t	mode, size, links, parent, pflags;
802	uint64_t	dzp_pflags = 0;
803	uint64_t	projid = ZFS_DEFAULT_PROJID;
804	uint64_t	rdev = 0;
805	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
806	dmu_buf_t	*db;
807	timestruc_t	now;
808	uint64_t	gen, obj;
809	int		bonuslen;
810	int		dnodesize;
811	sa_handle_t	*sa_hdl;
812	dmu_object_type_t obj_type;
813	sa_bulk_attr_t	*sa_attrs;
814	int		cnt = 0;
815	zfs_acl_locator_cb_t locate = { 0 };
816
817	ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
818
819	if (zfsvfs->z_replay) {
820		obj = vap->va_nodeid;
821		now = vap->va_ctime;		/* see zfs_replay_create() */
822		gen = vap->va_nblocks;		/* ditto */
823		dnodesize = vap->va_fsid;	/* ditto */
824	} else {
825		obj = 0;
826		gethrestime(&now);
827		gen = dmu_tx_get_txg(tx);
828		dnodesize = dmu_objset_dnodesize(zfsvfs->z_os);
829	}
830
831	if (dnodesize == 0)
832		dnodesize = DNODE_MIN_SIZE;
833
834	obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE;
835	bonuslen = (obj_type == DMU_OT_SA) ?
836	    DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE;
837
838	/*
839	 * Create a new DMU object.
840	 */
841	/*
842	 * There's currently no mechanism for pre-reading the blocks that will
843	 * be needed to allocate a new object, so we accept the small chance
844	 * that there will be an i/o error and we will fail one of the
845	 * assertions below.
846	 */
847	if (vap->va_type == VDIR) {
848		if (zfsvfs->z_replay) {
849			VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj,
850			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
851			    obj_type, bonuslen, dnodesize, tx));
852		} else {
853			obj = zap_create_norm_dnsize(zfsvfs->z_os,
854			    zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS,
855			    obj_type, bonuslen, dnodesize, tx);
856		}
857	} else {
858		if (zfsvfs->z_replay) {
859			VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj,
860			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
861			    obj_type, bonuslen, dnodesize, tx));
862		} else {
863			obj = dmu_object_alloc_dnsize(zfsvfs->z_os,
864			    DMU_OT_PLAIN_FILE_CONTENTS, 0,
865			    obj_type, bonuslen, dnodesize, tx);
866		}
867	}
868
869	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
870	VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db));
871
872	/*
873	 * If this is the root, fix up the half-initialized parent pointer
874	 * to reference the just-allocated physical data area.
875	 */
876	if (flag & IS_ROOT_NODE) {
877		dzp->z_id = obj;
878	}
879
880	/*
881	 * If parent is an xattr, so am I.
882	 */
883	if (dzp->z_pflags & ZFS_XATTR) {
884		flag |= IS_XATTR;
885	}
886
887	if (zfsvfs->z_use_fuids)
888		pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED;
889	else
890		pflags = 0;
891
892	if (vap->va_type == VDIR) {
893		size = 2;		/* contents ("." and "..") */
894		links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1;
895	} else {
896		size = links = 0;
897	}
898
899	if (vap->va_type == VBLK || vap->va_type == VCHR) {
900		rdev = zfs_expldev(vap->va_rdev);
901	}
902
903	parent = dzp->z_id;
904	mode = acl_ids->z_mode;
905	if (flag & IS_XATTR)
906		pflags |= ZFS_XATTR;
907
908	if (vap->va_type == VREG || vap->va_type == VDIR) {
909		/*
910		 * With ZFS_PROJID flag, we can easily know whether there is
911		 * project ID stored on disk or not. See zfs_space_delta_cb().
912		 */
913		if (obj_type != DMU_OT_ZNODE &&
914		    dmu_objset_projectquota_enabled(zfsvfs->z_os))
915			pflags |= ZFS_PROJID;
916
917		/*
918		 * Inherit project ID from parent if required.
919		 */
920		projid = zfs_inherit_projid(dzp);
921		if (dzp->z_pflags & ZFS_PROJINHERIT)
922			pflags |= ZFS_PROJINHERIT;
923	}
924
925	/*
926	 * No execs denied will be deterimed when zfs_mode_compute() is called.
927	 */
928	pflags |= acl_ids->z_aclp->z_hints &
929	    (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT|
930	    ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED);
931
932	ZFS_TIME_ENCODE(&now, crtime);
933	ZFS_TIME_ENCODE(&now, ctime);
934
935	if (vap->va_mask & AT_ATIME) {
936		ZFS_TIME_ENCODE(&vap->va_atime, atime);
937	} else {
938		ZFS_TIME_ENCODE(&now, atime);
939	}
940
941	if (vap->va_mask & AT_MTIME) {
942		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
943	} else {
944		ZFS_TIME_ENCODE(&now, mtime);
945	}
946
947	/* Now add in all of the "SA" attributes */
948	VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED,
949	    &sa_hdl));
950
951	/*
952	 * Setup the array of attributes to be replaced/set on the new file
953	 *
954	 * order for  DMU_OT_ZNODE is critical since it needs to be constructed
955	 * in the old znode_phys_t format.  Don't change this ordering
956	 */
957	sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP);
958
959	if (obj_type == DMU_OT_ZNODE) {
960		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
961		    NULL, &atime, 16);
962		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
963		    NULL, &mtime, 16);
964		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
965		    NULL, &ctime, 16);
966		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
967		    NULL, &crtime, 16);
968		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
969		    NULL, &gen, 8);
970		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
971		    NULL, &mode, 8);
972		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
973		    NULL, &size, 8);
974		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
975		    NULL, &parent, 8);
976	} else {
977		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs),
978		    NULL, &mode, 8);
979		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs),
980		    NULL, &size, 8);
981		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs),
982		    NULL, &gen, 8);
983		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs),
984		    NULL, &acl_ids->z_fuid, 8);
985		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs),
986		    NULL, &acl_ids->z_fgid, 8);
987		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs),
988		    NULL, &parent, 8);
989		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
990		    NULL, &pflags, 8);
991		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs),
992		    NULL, &atime, 16);
993		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs),
994		    NULL, &mtime, 16);
995		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs),
996		    NULL, &ctime, 16);
997		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs),
998		    NULL, &crtime, 16);
999	}
1000
1001	SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8);
1002
1003	if (obj_type == DMU_OT_ZNODE) {
1004		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL,
1005		    &empty_xattr, 8);
1006	} else if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
1007	    pflags & ZFS_PROJID) {
1008		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PROJID(zfsvfs),
1009		    NULL, &projid, 8);
1010	}
1011	if (obj_type == DMU_OT_ZNODE ||
1012	    (vap->va_type == VBLK || vap->va_type == VCHR)) {
1013		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs),
1014		    NULL, &rdev, 8);
1015
1016	}
1017	if (obj_type == DMU_OT_ZNODE) {
1018		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs),
1019		    NULL, &pflags, 8);
1020		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL,
1021		    &acl_ids->z_fuid, 8);
1022		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL,
1023		    &acl_ids->z_fgid, 8);
1024		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad,
1025		    sizeof (uint64_t) * 4);
1026		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1027		    &acl_phys, sizeof (zfs_acl_phys_t));
1028	} else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) {
1029		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL,
1030		    &acl_ids->z_aclp->z_acl_count, 8);
1031		locate.cb_aclp = acl_ids->z_aclp;
1032		SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs),
1033		    zfs_acl_data_locator, &locate,
1034		    acl_ids->z_aclp->z_acl_bytes);
1035		mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags,
1036		    acl_ids->z_fuid, acl_ids->z_fgid);
1037	}
1038
1039	VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0);
1040
1041	if (!(flag & IS_ROOT_NODE)) {
1042		*zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl);
1043		ASSERT(*zpp != NULL);
1044	} else {
1045		/*
1046		 * If we are creating the root node, the "parent" we
1047		 * passed in is the znode for the root.
1048		 */
1049		*zpp = dzp;
1050
1051		(*zpp)->z_sa_hdl = sa_hdl;
1052	}
1053
1054	(*zpp)->z_pflags = pflags;
1055	(*zpp)->z_mode = mode;
1056	(*zpp)->z_dnodesize = dnodesize;
1057	(*zpp)->z_projid = projid;
1058
1059	if (vap->va_mask & AT_XVATTR)
1060		zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx);
1061
1062	if (obj_type == DMU_OT_ZNODE ||
1063	    acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) {
1064		VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx));
1065	}
1066	kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END);
1067	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1068}
1069
1070/*
1071 * Update in-core attributes.  It is assumed the caller will be doing an
1072 * sa_bulk_update to push the changes out.
1073 */
1074void
1075zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx)
1076{
1077	xoptattr_t *xoap;
1078
1079	xoap = xva_getxoptattr(xvap);
1080	ASSERT(xoap);
1081
1082	if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
1083		uint64_t times[2];
1084		ZFS_TIME_ENCODE(&xoap->xoa_createtime, times);
1085		(void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs),
1086		    &times, sizeof (times), tx);
1087		XVA_SET_RTN(xvap, XAT_CREATETIME);
1088	}
1089	if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
1090		ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly,
1091		    zp->z_pflags, tx);
1092		XVA_SET_RTN(xvap, XAT_READONLY);
1093	}
1094	if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
1095		ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden,
1096		    zp->z_pflags, tx);
1097		XVA_SET_RTN(xvap, XAT_HIDDEN);
1098	}
1099	if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
1100		ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system,
1101		    zp->z_pflags, tx);
1102		XVA_SET_RTN(xvap, XAT_SYSTEM);
1103	}
1104	if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
1105		ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive,
1106		    zp->z_pflags, tx);
1107		XVA_SET_RTN(xvap, XAT_ARCHIVE);
1108	}
1109	if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
1110		ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable,
1111		    zp->z_pflags, tx);
1112		XVA_SET_RTN(xvap, XAT_IMMUTABLE);
1113	}
1114	if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
1115		ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink,
1116		    zp->z_pflags, tx);
1117		XVA_SET_RTN(xvap, XAT_NOUNLINK);
1118	}
1119	if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
1120		ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly,
1121		    zp->z_pflags, tx);
1122		XVA_SET_RTN(xvap, XAT_APPENDONLY);
1123	}
1124	if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
1125		ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump,
1126		    zp->z_pflags, tx);
1127		XVA_SET_RTN(xvap, XAT_NODUMP);
1128	}
1129	if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
1130		ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque,
1131		    zp->z_pflags, tx);
1132		XVA_SET_RTN(xvap, XAT_OPAQUE);
1133	}
1134	if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
1135		ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED,
1136		    xoap->xoa_av_quarantined, zp->z_pflags, tx);
1137		XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
1138	}
1139	if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
1140		ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified,
1141		    zp->z_pflags, tx);
1142		XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
1143	}
1144	if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
1145		zfs_sa_set_scanstamp(zp, xvap, tx);
1146		XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
1147	}
1148	if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
1149		ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse,
1150		    zp->z_pflags, tx);
1151		XVA_SET_RTN(xvap, XAT_REPARSE);
1152	}
1153	if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
1154		ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline,
1155		    zp->z_pflags, tx);
1156		XVA_SET_RTN(xvap, XAT_OFFLINE);
1157	}
1158	if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
1159		ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse,
1160		    zp->z_pflags, tx);
1161		XVA_SET_RTN(xvap, XAT_SPARSE);
1162	}
1163	if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
1164		ZFS_ATTR_SET(zp, ZFS_PROJINHERIT, xoap->xoa_projinherit,
1165		    zp->z_pflags, tx);
1166		XVA_SET_RTN(xvap, XAT_PROJINHERIT);
1167	}
1168}
1169
1170int
1171zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp)
1172{
1173	dmu_object_info_t doi;
1174	dmu_buf_t	*db;
1175	znode_t		*zp;
1176	int err;
1177	sa_handle_t	*hdl;
1178
1179	*zpp = NULL;
1180
1181	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1182
1183	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1184	if (err) {
1185		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1186		return (err);
1187	}
1188
1189	dmu_object_info_from_db(db, &doi);
1190	if (doi.doi_bonus_type != DMU_OT_SA &&
1191	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1192	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1193	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1194		sa_buf_rele(db, NULL);
1195		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1196		return (SET_ERROR(EINVAL));
1197	}
1198
1199	hdl = dmu_buf_get_user(db);
1200	if (hdl != NULL) {
1201		zp  = sa_get_userdata(hdl);
1202
1203
1204		/*
1205		 * Since "SA" does immediate eviction we
1206		 * should never find a sa handle that doesn't
1207		 * know about the znode.
1208		 */
1209
1210		ASSERT3P(zp, !=, NULL);
1211
1212		mutex_enter(&zp->z_lock);
1213		ASSERT3U(zp->z_id, ==, obj_num);
1214		if (zp->z_unlinked) {
1215			err = SET_ERROR(ENOENT);
1216		} else {
1217			VN_HOLD(ZTOV(zp));
1218			*zpp = zp;
1219			err = 0;
1220		}
1221		mutex_exit(&zp->z_lock);
1222		sa_buf_rele(db, NULL);
1223		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1224		return (err);
1225	}
1226
1227	/*
1228	 * Not found create new znode/vnode
1229	 * but only if file exists.
1230	 *
1231	 * There is a small window where zfs_vget() could
1232	 * find this object while a file create is still in
1233	 * progress.  This is checked for in zfs_znode_alloc()
1234	 *
1235	 * if zfs_znode_alloc() fails it will drop the hold on the
1236	 * bonus buffer.
1237	 */
1238	zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size,
1239	    doi.doi_bonus_type, NULL);
1240	if (zp == NULL) {
1241		err = SET_ERROR(ENOENT);
1242	} else {
1243		*zpp = zp;
1244	}
1245	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1246	return (err);
1247}
1248
1249int
1250zfs_rezget(znode_t *zp)
1251{
1252	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1253	dmu_object_info_t doi;
1254	dmu_buf_t *db;
1255	uint64_t obj_num = zp->z_id;
1256	uint64_t mode;
1257	sa_bulk_attr_t bulk[10];
1258	int err;
1259	int count = 0;
1260	uint64_t gen;
1261	uint64_t projid = ZFS_DEFAULT_PROJID;
1262
1263	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num);
1264
1265	mutex_enter(&zp->z_acl_lock);
1266	if (zp->z_acl_cached) {
1267		zfs_acl_free(zp->z_acl_cached);
1268		zp->z_acl_cached = NULL;
1269	}
1270
1271	mutex_exit(&zp->z_acl_lock);
1272	ASSERT(zp->z_sa_hdl == NULL);
1273	err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db);
1274	if (err) {
1275		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1276		return (err);
1277	}
1278
1279	dmu_object_info_from_db(db, &doi);
1280	if (doi.doi_bonus_type != DMU_OT_SA &&
1281	    (doi.doi_bonus_type != DMU_OT_ZNODE ||
1282	    (doi.doi_bonus_type == DMU_OT_ZNODE &&
1283	    doi.doi_bonus_size < sizeof (znode_phys_t)))) {
1284		sa_buf_rele(db, NULL);
1285		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1286		return (SET_ERROR(EINVAL));
1287	}
1288
1289	zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL);
1290
1291	/* reload cached values */
1292	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL,
1293	    &gen, sizeof (gen));
1294	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
1295	    &zp->z_size, sizeof (zp->z_size));
1296	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
1297	    &zp->z_links, sizeof (zp->z_links));
1298	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1299	    &zp->z_pflags, sizeof (zp->z_pflags));
1300	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
1301	    &zp->z_atime, sizeof (zp->z_atime));
1302	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1303	    &zp->z_uid, sizeof (zp->z_uid));
1304	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1305	    &zp->z_gid, sizeof (zp->z_gid));
1306	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1307	    &mode, sizeof (mode));
1308
1309	if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) {
1310		zfs_znode_dmu_fini(zp);
1311		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1312		return (SET_ERROR(EIO));
1313	}
1314
1315	if (dmu_objset_projectquota_enabled(zfsvfs->z_os)) {
1316		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs),
1317		    &projid, 8);
1318		if (err != 0 && err != ENOENT) {
1319			zfs_znode_dmu_fini(zp);
1320			ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1321			return (SET_ERROR(err));
1322		}
1323	}
1324
1325	zp->z_projid = projid;
1326	zp->z_mode = mode;
1327
1328	if (gen != zp->z_gen) {
1329		zfs_znode_dmu_fini(zp);
1330		ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1331		return (SET_ERROR(EIO));
1332	}
1333
1334	zp->z_blksz = doi.doi_data_block_size;
1335
1336	/*
1337	 * If the file has zero links, then it has been unlinked on the send
1338	 * side and it must be in the received unlinked set.
1339	 * We call zfs_znode_dmu_fini() now to prevent any accesses to the
1340	 * stale data and to prevent automatical removal of the file in
1341	 * zfs_zinactive().  The file will be removed either when it is removed
1342	 * on the send side and the next incremental stream is received or
1343	 * when the unlinked set gets processed.
1344	 */
1345	zp->z_unlinked = (zp->z_links == 0);
1346	if (zp->z_unlinked)
1347		zfs_znode_dmu_fini(zp);
1348
1349	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num);
1350
1351	return (0);
1352}
1353
1354void
1355zfs_znode_delete(znode_t *zp, dmu_tx_t *tx)
1356{
1357	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1358	objset_t *os = zfsvfs->z_os;
1359	uint64_t obj = zp->z_id;
1360	uint64_t acl_obj = zfs_external_acl(zp);
1361
1362	ZFS_OBJ_HOLD_ENTER(zfsvfs, obj);
1363	if (acl_obj) {
1364		VERIFY(!zp->z_is_sa);
1365		VERIFY(0 == dmu_object_free(os, acl_obj, tx));
1366	}
1367	VERIFY(0 == dmu_object_free(os, obj, tx));
1368	zfs_znode_dmu_fini(zp);
1369	ZFS_OBJ_HOLD_EXIT(zfsvfs, obj);
1370	zfs_znode_free(zp);
1371}
1372
1373void
1374zfs_zinactive(znode_t *zp)
1375{
1376	vnode_t	*vp = ZTOV(zp);
1377	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1378	uint64_t z_id = zp->z_id;
1379
1380	ASSERT(zp->z_sa_hdl);
1381
1382	/*
1383	 * Don't allow a zfs_zget() while were trying to release this znode
1384	 */
1385	ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id);
1386
1387	mutex_enter(&zp->z_lock);
1388	mutex_enter(&vp->v_lock);
1389	VN_RELE_LOCKED(vp);
1390	if (vp->v_count > 0 || vn_has_cached_data(vp)) {
1391		/*
1392		 * If the hold count is greater than zero, somebody has
1393		 * obtained a new reference on this znode while we were
1394		 * processing it here, so we are done.  If we still have
1395		 * mapped pages then we are also done, since we don't
1396		 * want to inactivate the znode until the pages get pushed.
1397		 *
1398		 * XXX - if vn_has_cached_data(vp) is true, but count == 0,
1399		 * this seems like it would leave the znode hanging with
1400		 * no chance to go inactive...
1401		 */
1402		mutex_exit(&vp->v_lock);
1403		mutex_exit(&zp->z_lock);
1404		ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1405		return;
1406	}
1407	mutex_exit(&vp->v_lock);
1408
1409	/*
1410	 * If this was the last reference to a file with no links, remove
1411	 * the file from the file system unless the file system is mounted
1412	 * read-only.  That can happen, for example, if the file system was
1413	 * originally read-write, the file was opened, then unlinked and
1414	 * the file system was made read-only before the file was finally
1415	 * closed.  The file will remain in the unlinked set.
1416	 */
1417	if (zp->z_unlinked) {
1418		ASSERT(!zfsvfs->z_issnap);
1419		if ((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) {
1420			mutex_exit(&zp->z_lock);
1421			ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1422			zfs_rmnode(zp);
1423			return;
1424		}
1425	}
1426
1427	mutex_exit(&zp->z_lock);
1428	zfs_znode_dmu_fini(zp);
1429	ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id);
1430	zfs_znode_free(zp);
1431}
1432
1433void
1434zfs_znode_free(znode_t *zp)
1435{
1436	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1437
1438	vn_invalid(ZTOV(zp));
1439
1440	ASSERT(ZTOV(zp)->v_count == 0);
1441
1442	mutex_enter(&zfsvfs->z_znodes_lock);
1443	POINTER_INVALIDATE(&zp->z_zfsvfs);
1444	list_remove(&zfsvfs->z_all_znodes, zp);
1445	mutex_exit(&zfsvfs->z_znodes_lock);
1446
1447	if (zp->z_acl_cached) {
1448		zfs_acl_free(zp->z_acl_cached);
1449		zp->z_acl_cached = NULL;
1450	}
1451
1452	kmem_cache_free(znode_cache, zp);
1453
1454	VFS_RELE(zfsvfs->z_vfs);
1455}
1456
1457void
1458zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2],
1459    uint64_t ctime[2], boolean_t have_tx)
1460{
1461	timestruc_t	now;
1462
1463	gethrestime(&now);
1464
1465	if (have_tx) {	/* will sa_bulk_update happen really soon? */
1466		zp->z_atime_dirty = 0;
1467		zp->z_seq++;
1468	} else {
1469		zp->z_atime_dirty = 1;
1470	}
1471
1472	if (flag & AT_ATIME) {
1473		ZFS_TIME_ENCODE(&now, zp->z_atime);
1474	}
1475
1476	if (flag & AT_MTIME) {
1477		ZFS_TIME_ENCODE(&now, mtime);
1478		if (zp->z_zfsvfs->z_use_fuids) {
1479			zp->z_pflags |= (ZFS_ARCHIVE |
1480			    ZFS_AV_MODIFIED);
1481		}
1482	}
1483
1484	if (flag & AT_CTIME) {
1485		ZFS_TIME_ENCODE(&now, ctime);
1486		if (zp->z_zfsvfs->z_use_fuids)
1487			zp->z_pflags |= ZFS_ARCHIVE;
1488	}
1489}
1490
1491/*
1492 * Grow the block size for a file.
1493 *
1494 *	IN:	zp	- znode of file to free data in.
1495 *		size	- requested block size
1496 *		tx	- open transaction.
1497 *
1498 * NOTE: this function assumes that the znode is write locked.
1499 */
1500void
1501zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
1502{
1503	int		error;
1504	u_longlong_t	dummy;
1505
1506	if (size <= zp->z_blksz)
1507		return;
1508	/*
1509	 * If the file size is already greater than the current blocksize,
1510	 * we will not grow.  If there is more than one block in a file,
1511	 * the blocksize cannot change.
1512	 */
1513	if (zp->z_blksz && zp->z_size > zp->z_blksz)
1514		return;
1515
1516	error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id,
1517	    size, 0, tx);
1518
1519	if (error == ENOTSUP)
1520		return;
1521	ASSERT0(error);
1522
1523	/* What blocksize did we actually get? */
1524	dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy);
1525}
1526
1527/*
1528 * This is a dummy interface used when pvn_vplist_dirty() should *not*
1529 * be calling back into the fs for a putpage().  E.g.: when truncating
1530 * a file, the pages being "thrown away* don't need to be written out.
1531 */
1532/* ARGSUSED */
1533static int
1534zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
1535    int flags, cred_t *cr)
1536{
1537	ASSERT(0);
1538	return (0);
1539}
1540
1541/*
1542 * Increase the file length
1543 *
1544 *	IN:	zp	- znode of file to free data in.
1545 *		end	- new end-of-file
1546 *
1547 *	RETURN:	0 on success, error code on failure
1548 */
1549static int
1550zfs_extend(znode_t *zp, uint64_t end)
1551{
1552	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1553	dmu_tx_t *tx;
1554	locked_range_t *lr;
1555	uint64_t newblksz;
1556	int error;
1557
1558	/*
1559	 * We will change zp_size, lock the whole file.
1560	 */
1561	lr = rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1562
1563	/*
1564	 * Nothing to do if file already at desired length.
1565	 */
1566	if (end <= zp->z_size) {
1567		rangelock_exit(lr);
1568		return (0);
1569	}
1570	tx = dmu_tx_create(zfsvfs->z_os);
1571	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1572	zfs_sa_upgrade_txholds(tx, zp);
1573	if (end > zp->z_blksz &&
1574	    (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) {
1575		/*
1576		 * We are growing the file past the current block size.
1577		 */
1578		if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) {
1579			/*
1580			 * File's blocksize is already larger than the
1581			 * "recordsize" property.  Only let it grow to
1582			 * the next power of 2.
1583			 */
1584			ASSERT(!ISP2(zp->z_blksz));
1585			newblksz = MIN(end, 1 << highbit64(zp->z_blksz));
1586		} else {
1587			newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz);
1588		}
1589		dmu_tx_hold_write(tx, zp->z_id, 0, newblksz);
1590	} else {
1591		newblksz = 0;
1592	}
1593
1594	error = dmu_tx_assign(tx, TXG_WAIT);
1595	if (error) {
1596		dmu_tx_abort(tx);
1597		rangelock_exit(lr);
1598		return (error);
1599	}
1600
1601	if (newblksz)
1602		zfs_grow_blocksize(zp, newblksz, tx);
1603
1604	zp->z_size = end;
1605
1606	VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs),
1607	    &zp->z_size, sizeof (zp->z_size), tx));
1608
1609	rangelock_exit(lr);
1610
1611	dmu_tx_commit(tx);
1612
1613	return (0);
1614}
1615
1616/*
1617 * Free space in a file.
1618 *
1619 *	IN:	zp	- znode of file to free data in.
1620 *		off	- start of section to free.
1621 *		len	- length of section to free.
1622 *
1623 *	RETURN:	0 on success, error code on failure
1624 */
1625static int
1626zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
1627{
1628	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1629	locked_range_t *lr;
1630	int error;
1631
1632	/*
1633	 * Lock the range being freed.
1634	 */
1635	lr = rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER);
1636
1637	/*
1638	 * Nothing to do if file already at desired length.
1639	 */
1640	if (off >= zp->z_size) {
1641		rangelock_exit(lr);
1642		return (0);
1643	}
1644
1645	if (off + len > zp->z_size)
1646		len = zp->z_size - off;
1647
1648	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len);
1649
1650	rangelock_exit(lr);
1651
1652	return (error);
1653}
1654
1655/*
1656 * Truncate a file
1657 *
1658 *	IN:	zp	- znode of file to free data in.
1659 *		end	- new end-of-file.
1660 *
1661 *	RETURN:	0 on success, error code on failure
1662 */
1663static int
1664zfs_trunc(znode_t *zp, uint64_t end)
1665{
1666	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1667	vnode_t *vp = ZTOV(zp);
1668	dmu_tx_t *tx;
1669	locked_range_t *lr;
1670	int error;
1671	sa_bulk_attr_t bulk[2];
1672	int count = 0;
1673
1674	/*
1675	 * We will change zp_size, lock the whole file.
1676	 */
1677	lr = rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER);
1678
1679	/*
1680	 * Nothing to do if file already at desired length.
1681	 */
1682	if (end >= zp->z_size) {
1683		rangelock_exit(lr);
1684		return (0);
1685	}
1686
1687	error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end,
1688	    DMU_OBJECT_END);
1689	if (error) {
1690		rangelock_exit(lr);
1691		return (error);
1692	}
1693	tx = dmu_tx_create(zfsvfs->z_os);
1694	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1695	zfs_sa_upgrade_txholds(tx, zp);
1696	dmu_tx_mark_netfree(tx);
1697	error = dmu_tx_assign(tx, TXG_WAIT);
1698	if (error) {
1699		dmu_tx_abort(tx);
1700		rangelock_exit(lr);
1701		return (error);
1702	}
1703
1704	zp->z_size = end;
1705	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
1706	    NULL, &zp->z_size, sizeof (zp->z_size));
1707
1708	if (end == 0) {
1709		zp->z_pflags &= ~ZFS_SPARSE;
1710		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1711		    NULL, &zp->z_pflags, 8);
1712	}
1713	VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0);
1714
1715	dmu_tx_commit(tx);
1716
1717	/*
1718	 * Clear any mapped pages in the truncated region.  This has to
1719	 * happen outside of the transaction to avoid the possibility of
1720	 * a deadlock with someone trying to push a page that we are
1721	 * about to invalidate.
1722	 */
1723	if (vn_has_cached_data(vp)) {
1724		page_t *pp;
1725		uint64_t start = end & PAGEMASK;
1726		int poff = end & PAGEOFFSET;
1727
1728		if (poff != 0 && (pp = page_lookup(vp, start, SE_SHARED))) {
1729			/*
1730			 * We need to zero a partial page.
1731			 */
1732			pagezero(pp, poff, PAGESIZE - poff);
1733			start += PAGESIZE;
1734			page_unlock(pp);
1735		}
1736		error = pvn_vplist_dirty(vp, start, zfs_no_putpage,
1737		    B_INVAL | B_TRUNC, NULL);
1738		ASSERT(error == 0);
1739	}
1740
1741	rangelock_exit(lr);
1742
1743	return (0);
1744}
1745
1746/*
1747 * Free space in a file
1748 *
1749 *	IN:	zp	- znode of file to free data in.
1750 *		off	- start of range
1751 *		len	- end of range (0 => EOF)
1752 *		flag	- current file open mode flags.
1753 *		log	- TRUE if this action should be logged
1754 *
1755 *	RETURN:	0 on success, error code on failure
1756 */
1757int
1758zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
1759{
1760	vnode_t *vp = ZTOV(zp);
1761	dmu_tx_t *tx;
1762	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1763	zilog_t *zilog = zfsvfs->z_log;
1764	uint64_t mode;
1765	uint64_t mtime[2], ctime[2];
1766	sa_bulk_attr_t bulk[3];
1767	int count = 0;
1768	int error;
1769
1770	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode,
1771	    sizeof (mode))) != 0)
1772		return (error);
1773
1774	if (off > zp->z_size) {
1775		error =  zfs_extend(zp, off+len);
1776		if (error == 0 && log)
1777			goto log;
1778		else
1779			return (error);
1780	}
1781
1782	/*
1783	 * Check for any locks in the region to be freed.
1784	 */
1785
1786	if (MANDLOCK(vp, (mode_t)mode)) {
1787		uint64_t length = (len ? len : zp->z_size - off);
1788		if (error = chklock(vp, FWRITE, off, length, flag, NULL))
1789			return (error);
1790	}
1791
1792	if (len == 0) {
1793		error = zfs_trunc(zp, off);
1794	} else {
1795		if ((error = zfs_free_range(zp, off, len)) == 0 &&
1796		    off + len > zp->z_size)
1797			error = zfs_extend(zp, off+len);
1798	}
1799	if (error || !log)
1800		return (error);
1801log:
1802	tx = dmu_tx_create(zfsvfs->z_os);
1803	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1804	zfs_sa_upgrade_txholds(tx, zp);
1805	error = dmu_tx_assign(tx, TXG_WAIT);
1806	if (error) {
1807		dmu_tx_abort(tx);
1808		return (error);
1809	}
1810
1811	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16);
1812	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16);
1813	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
1814	    NULL, &zp->z_pflags, 8);
1815	zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
1816	error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1817	ASSERT(error == 0);
1818
1819	zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
1820
1821	dmu_tx_commit(tx);
1822	return (0);
1823}
1824
1825void
1826zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx)
1827{
1828	uint64_t	moid, obj, sa_obj, version;
1829	uint64_t	sense = ZFS_CASE_SENSITIVE;
1830	uint64_t	norm = 0;
1831	nvpair_t	*elem;
1832	int		error;
1833	int		i;
1834	znode_t		*rootzp = NULL;
1835	zfsvfs_t	*zfsvfs;
1836	vnode_t		*vp;
1837	vattr_t		vattr;
1838	znode_t		*zp;
1839	zfs_acl_ids_t	acl_ids;
1840
1841	/*
1842	 * First attempt to create master node.
1843	 */
1844	/*
1845	 * In an empty objset, there are no blocks to read and thus
1846	 * there can be no i/o errors (which we assert below).
1847	 */
1848	moid = MASTER_NODE_OBJ;
1849	error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE,
1850	    DMU_OT_NONE, 0, tx);
1851	ASSERT(error == 0);
1852
1853	/*
1854	 * Set starting attributes.
1855	 */
1856	version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os)));
1857	elem = NULL;
1858	while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) {
1859		/* For the moment we expect all zpl props to be uint64_ts */
1860		uint64_t val;
1861		char *name;
1862
1863		ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64);
1864		VERIFY(nvpair_value_uint64(elem, &val) == 0);
1865		name = nvpair_name(elem);
1866		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) {
1867			if (val < version)
1868				version = val;
1869		} else {
1870			error = zap_update(os, moid, name, 8, 1, &val, tx);
1871		}
1872		ASSERT(error == 0);
1873		if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0)
1874			norm = val;
1875		else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0)
1876			sense = val;
1877	}
1878	ASSERT(version != 0);
1879	error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx);
1880
1881	/*
1882	 * Create zap object used for SA attribute registration
1883	 */
1884
1885	if (version >= ZPL_VERSION_SA) {
1886		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1887		    DMU_OT_NONE, 0, tx);
1888		error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1889		ASSERT(error == 0);
1890	} else {
1891		sa_obj = 0;
1892	}
1893	/*
1894	 * Create a delete queue.
1895	 */
1896	obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx);
1897
1898	error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx);
1899	ASSERT(error == 0);
1900
1901	/*
1902	 * Create root znode.  Create minimal znode/vnode/zfsvfs
1903	 * to allow zfs_mknode to work.
1904	 */
1905	vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE;
1906	vattr.va_type = VDIR;
1907	vattr.va_mode = S_IFDIR|0755;
1908	vattr.va_uid = crgetuid(cr);
1909	vattr.va_gid = crgetgid(cr);
1910
1911	rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP);
1912	ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs));
1913	rootzp->z_moved = 0;
1914	rootzp->z_unlinked = 0;
1915	rootzp->z_atime_dirty = 0;
1916	rootzp->z_is_sa = USE_SA(version, os);
1917	rootzp->z_pflags = 0;
1918
1919	vp = ZTOV(rootzp);
1920	vn_reinit(vp);
1921	vp->v_type = VDIR;
1922
1923	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1924	zfsvfs->z_os = os;
1925	zfsvfs->z_parent = zfsvfs;
1926	zfsvfs->z_version = version;
1927	zfsvfs->z_use_fuids = USE_FUIDS(version, os);
1928	zfsvfs->z_use_sa = USE_SA(version, os);
1929	zfsvfs->z_norm = norm;
1930
1931	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1932	    &zfsvfs->z_attr_table);
1933
1934	ASSERT(error == 0);
1935
1936	/*
1937	 * Fold case on file systems that are always or sometimes case
1938	 * insensitive.
1939	 */
1940	if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED)
1941		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1942
1943	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1944	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1945	    offsetof(znode_t, z_link_node));
1946
1947	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1948		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1949
1950	rootzp->z_zfsvfs = zfsvfs;
1951	VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr,
1952	    cr, NULL, &acl_ids));
1953	zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids);
1954	ASSERT3P(zp, ==, rootzp);
1955	ASSERT(!vn_in_dnlc(ZTOV(rootzp))); /* not valid to move */
1956	error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx);
1957	ASSERT(error == 0);
1958	zfs_acl_ids_free(&acl_ids);
1959	POINTER_INVALIDATE(&rootzp->z_zfsvfs);
1960
1961	ZTOV(rootzp)->v_count = 0;
1962	sa_handle_destroy(rootzp->z_sa_hdl);
1963	kmem_cache_free(znode_cache, rootzp);
1964
1965	/*
1966	 * Create shares directory
1967	 */
1968
1969	error = zfs_create_share_dir(zfsvfs, tx);
1970
1971	ASSERT(error == 0);
1972
1973	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1974		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1975	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1976}
1977
1978#endif /* _KERNEL */
1979
1980static int
1981zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table)
1982{
1983	uint64_t sa_obj = 0;
1984	int error;
1985
1986	error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj);
1987	if (error != 0 && error != ENOENT)
1988		return (error);
1989
1990	error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table);
1991	return (error);
1992}
1993
1994static int
1995zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp,
1996    dmu_buf_t **db, void *tag)
1997{
1998	dmu_object_info_t doi;
1999	int error;
2000
2001	if ((error = sa_buf_hold(osp, obj, tag, db)) != 0)
2002		return (error);
2003
2004	dmu_object_info_from_db(*db, &doi);
2005	if ((doi.doi_bonus_type != DMU_OT_SA &&
2006	    doi.doi_bonus_type != DMU_OT_ZNODE) ||
2007	    doi.doi_bonus_type == DMU_OT_ZNODE &&
2008	    doi.doi_bonus_size < sizeof (znode_phys_t)) {
2009		sa_buf_rele(*db, tag);
2010		return (SET_ERROR(ENOTSUP));
2011	}
2012
2013	error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp);
2014	if (error != 0) {
2015		sa_buf_rele(*db, tag);
2016		return (error);
2017	}
2018
2019	return (0);
2020}
2021
2022void
2023zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag)
2024{
2025	sa_handle_destroy(hdl);
2026	sa_buf_rele(db, tag);
2027}
2028
2029/*
2030 * Given an object number, return its parent object number and whether
2031 * or not the object is an extended attribute directory.
2032 */
2033static int
2034zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table,
2035    uint64_t *pobjp, int *is_xattrdir)
2036{
2037	uint64_t parent;
2038	uint64_t pflags;
2039	uint64_t mode;
2040	uint64_t parent_mode;
2041	sa_bulk_attr_t bulk[3];
2042	sa_handle_t *sa_hdl;
2043	dmu_buf_t *sa_db;
2044	int count = 0;
2045	int error;
2046
2047	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL,
2048	    &parent, sizeof (parent));
2049	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL,
2050	    &pflags, sizeof (pflags));
2051	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2052	    &mode, sizeof (mode));
2053
2054	if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0)
2055		return (error);
2056
2057	/*
2058	 * When a link is removed its parent pointer is not changed and will
2059	 * be invalid.  There are two cases where a link is removed but the
2060	 * file stays around, when it goes to the delete queue and when there
2061	 * are additional links.
2062	 */
2063	error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG);
2064	if (error != 0)
2065		return (error);
2066
2067	error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode));
2068	zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2069	if (error != 0)
2070		return (error);
2071
2072	*is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode);
2073
2074	/*
2075	 * Extended attributes can be applied to files, directories, etc.
2076	 * Otherwise the parent must be a directory.
2077	 */
2078	if (!*is_xattrdir && !S_ISDIR(parent_mode))
2079		return (SET_ERROR(EINVAL));
2080
2081	*pobjp = parent;
2082
2083	return (0);
2084}
2085
2086/*
2087 * Given an object number, return some zpl level statistics
2088 */
2089static int
2090zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table,
2091    zfs_stat_t *sb)
2092{
2093	sa_bulk_attr_t bulk[4];
2094	int count = 0;
2095
2096	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL,
2097	    &sb->zs_mode, sizeof (sb->zs_mode));
2098	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL,
2099	    &sb->zs_gen, sizeof (sb->zs_gen));
2100	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL,
2101	    &sb->zs_links, sizeof (sb->zs_links));
2102	SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL,
2103	    &sb->zs_ctime, sizeof (sb->zs_ctime));
2104
2105	return (sa_bulk_lookup(hdl, bulk, count));
2106}
2107
2108static int
2109zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl,
2110    sa_attr_type_t *sa_table, char *buf, int len)
2111{
2112	sa_handle_t *sa_hdl;
2113	sa_handle_t *prevhdl = NULL;
2114	dmu_buf_t *prevdb = NULL;
2115	dmu_buf_t *sa_db = NULL;
2116	char *path = buf + len - 1;
2117	int error;
2118
2119	*path = '\0';
2120	sa_hdl = hdl;
2121
2122	uint64_t deleteq_obj;
2123	VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ,
2124	    ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj));
2125	error = zap_lookup_int(osp, deleteq_obj, obj);
2126	if (error == 0) {
2127		return (ESTALE);
2128	} else if (error != ENOENT) {
2129		return (error);
2130	}
2131	error = 0;
2132
2133	for (;;) {
2134		uint64_t pobj;
2135		char component[MAXNAMELEN + 2];
2136		size_t complen;
2137		int is_xattrdir;
2138
2139		if (prevdb)
2140			zfs_release_sa_handle(prevhdl, prevdb, FTAG);
2141
2142		if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj,
2143		    &is_xattrdir)) != 0)
2144			break;
2145
2146		if (pobj == obj) {
2147			if (path[0] != '/')
2148				*--path = '/';
2149			break;
2150		}
2151
2152		component[0] = '/';
2153		if (is_xattrdir) {
2154			(void) sprintf(component + 1, "<xattrdir>");
2155		} else {
2156			error = zap_value_search(osp, pobj, obj,
2157			    ZFS_DIRENT_OBJ(-1ULL), component + 1);
2158			if (error != 0)
2159				break;
2160		}
2161
2162		complen = strlen(component);
2163		path -= complen;
2164		ASSERT(path >= buf);
2165		bcopy(component, path, complen);
2166		obj = pobj;
2167
2168		if (sa_hdl != hdl) {
2169			prevhdl = sa_hdl;
2170			prevdb = sa_db;
2171		}
2172		error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG);
2173		if (error != 0) {
2174			sa_hdl = prevhdl;
2175			sa_db = prevdb;
2176			break;
2177		}
2178	}
2179
2180	if (sa_hdl != NULL && sa_hdl != hdl) {
2181		ASSERT(sa_db != NULL);
2182		zfs_release_sa_handle(sa_hdl, sa_db, FTAG);
2183	}
2184
2185	if (error == 0)
2186		(void) memmove(buf, path, buf + len - path);
2187
2188	return (error);
2189}
2190
2191int
2192zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len)
2193{
2194	sa_attr_type_t *sa_table;
2195	sa_handle_t *hdl;
2196	dmu_buf_t *db;
2197	int error;
2198
2199	error = zfs_sa_setup(osp, &sa_table);
2200	if (error != 0)
2201		return (error);
2202
2203	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2204	if (error != 0)
2205		return (error);
2206
2207	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2208
2209	zfs_release_sa_handle(hdl, db, FTAG);
2210	return (error);
2211}
2212
2213int
2214zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb,
2215    char *buf, int len)
2216{
2217	char *path = buf + len - 1;
2218	sa_attr_type_t *sa_table;
2219	sa_handle_t *hdl;
2220	dmu_buf_t *db;
2221	int error;
2222
2223	*path = '\0';
2224
2225	error = zfs_sa_setup(osp, &sa_table);
2226	if (error != 0)
2227		return (error);
2228
2229	error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG);
2230	if (error != 0)
2231		return (error);
2232
2233	error = zfs_obj_to_stats_impl(hdl, sa_table, sb);
2234	if (error != 0) {
2235		zfs_release_sa_handle(hdl, db, FTAG);
2236		return (error);
2237	}
2238
2239	error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len);
2240
2241	zfs_release_sa_handle(hdl, db, FTAG);
2242	return (error);
2243}
2244