xref: /freebsd-src/sys/contrib/openzfs/module/os/linux/zfs/zfs_vnops_os.c (revision c14e17a49cdd72c8c05d12fbec631c409b14b9cb)
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 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
25  * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
26  * Copyright 2017 Nexenta Systems, Inc.
27  */
28 
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
31 
32 
33 #include <sys/types.h>
34 #include <sys/param.h>
35 #include <sys/time.h>
36 #include <sys/sysmacros.h>
37 #include <sys/vfs.h>
38 #include <sys/file.h>
39 #include <sys/stat.h>
40 #include <sys/kmem.h>
41 #include <sys/taskq.h>
42 #include <sys/uio.h>
43 #include <sys/vmsystm.h>
44 #include <sys/atomic.h>
45 #include <sys/pathname.h>
46 #include <sys/cmn_err.h>
47 #include <sys/errno.h>
48 #include <sys/zfs_dir.h>
49 #include <sys/zfs_acl.h>
50 #include <sys/zfs_ioctl.h>
51 #include <sys/fs/zfs.h>
52 #include <sys/dmu.h>
53 #include <sys/dmu_objset.h>
54 #include <sys/spa.h>
55 #include <sys/txg.h>
56 #include <sys/dbuf.h>
57 #include <sys/zap.h>
58 #include <sys/sa.h>
59 #include <sys/policy.h>
60 #include <sys/sunddi.h>
61 #include <sys/sid.h>
62 #include <sys/zfs_ctldir.h>
63 #include <sys/zfs_fuid.h>
64 #include <sys/zfs_quota.h>
65 #include <sys/zfs_sa.h>
66 #include <sys/zfs_vnops.h>
67 #include <sys/zfs_rlock.h>
68 #include <sys/cred.h>
69 #include <sys/zpl.h>
70 #include <sys/zil.h>
71 #include <sys/sa_impl.h>
72 
73 /*
74  * Programming rules.
75  *
76  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
77  * properly lock its in-core state, create a DMU transaction, do the work,
78  * record this work in the intent log (ZIL), commit the DMU transaction,
79  * and wait for the intent log to commit if it is a synchronous operation.
80  * Moreover, the vnode ops must work in both normal and log replay context.
81  * The ordering of events is important to avoid deadlocks and references
82  * to freed memory.  The example below illustrates the following Big Rules:
83  *
84  *  (1) A check must be made in each zfs thread for a mounted file system.
85  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
86  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
87  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
88  *      can return EIO from the calling function.
89  *
90  *  (2)	zrele() should always be the last thing except for zil_commit()
91  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
92  *	First, if it's the last reference, the vnode/znode
93  *	can be freed, so the zp may point to freed memory.  Second, the last
94  *	reference will call zfs_zinactive(), which may induce a lot of work --
95  *	pushing cached pages (which acquires range locks) and syncing out
96  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
97  *	which could deadlock the system if you were already holding one.
98  *	If you must call zrele() within a tx then use zfs_zrele_async().
99  *
100  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
101  *	as they can span dmu_tx_assign() calls.
102  *
103  *  (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
104  *      dmu_tx_assign().  This is critical because we don't want to block
105  *      while holding locks.
106  *
107  *	If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT.  This
108  *	reduces lock contention and CPU usage when we must wait (note that if
109  *	throughput is constrained by the storage, nearly every transaction
110  *	must wait).
111  *
112  *      Note, in particular, that if a lock is sometimes acquired before
113  *      the tx assigns, and sometimes after (e.g. z_lock), then failing
114  *      to use a non-blocking assign can deadlock the system.  The scenario:
115  *
116  *	Thread A has grabbed a lock before calling dmu_tx_assign().
117  *	Thread B is in an already-assigned tx, and blocks for this lock.
118  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
119  *	forever, because the previous txg can't quiesce until B's tx commits.
120  *
121  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
122  *	then drop all locks, call dmu_tx_wait(), and try again.  On subsequent
123  *	calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT,
124  *	to indicate that this operation has already called dmu_tx_wait().
125  *	This will ensure that we don't retry forever, waiting a short bit
126  *	each time.
127  *
128  *  (5)	If the operation succeeded, generate the intent log entry for it
129  *	before dropping locks.  This ensures that the ordering of events
130  *	in the intent log matches the order in which they actually occurred.
131  *	During ZIL replay the zfs_log_* functions will update the sequence
132  *	number to indicate the zil transaction has replayed.
133  *
134  *  (6)	At the end of each vnode op, the DMU tx must always commit,
135  *	regardless of whether there were any errors.
136  *
137  *  (7)	After dropping all locks, invoke zil_commit(zilog, foid)
138  *	to ensure that synchronous semantics are provided when necessary.
139  *
140  * In general, this is how things should be ordered in each vnode op:
141  *
142  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
143  * top:
144  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may igrab())
145  *	rw_enter(...);			// grab any other locks you need
146  *	tx = dmu_tx_create(...);	// get DMU tx
147  *	dmu_tx_hold_*();		// hold each object you might modify
148  *	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
149  *	if (error) {
150  *		rw_exit(...);		// drop locks
151  *		zfs_dirent_unlock(dl);	// unlock directory entry
152  *		zrele(...);		// release held znodes
153  *		if (error == ERESTART) {
154  *			waited = B_TRUE;
155  *			dmu_tx_wait(tx);
156  *			dmu_tx_abort(tx);
157  *			goto top;
158  *		}
159  *		dmu_tx_abort(tx);	// abort DMU tx
160  *		ZFS_EXIT(zfsvfs);	// finished in zfs
161  *		return (error);		// really out of space
162  *	}
163  *	error = do_real_work();		// do whatever this VOP does
164  *	if (error == 0)
165  *		zfs_log_*(...);		// on success, make ZIL entry
166  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
167  *	rw_exit(...);			// drop locks
168  *	zfs_dirent_unlock(dl);		// unlock directory entry
169  *	zrele(...);			// release held znodes
170  *	zil_commit(zilog, foid);	// synchronous when necessary
171  *	ZFS_EXIT(zfsvfs);		// finished in zfs
172  *	return (error);			// done, report error
173  */
174 
175 /*
176  * Virus scanning is unsupported.  It would be possible to add a hook
177  * here to performance the required virus scan.  This could be done
178  * entirely in the kernel or potentially as an update to invoke a
179  * scanning utility.
180  */
181 static int
182 zfs_vscan(struct inode *ip, cred_t *cr, int async)
183 {
184 	return (0);
185 }
186 
187 /* ARGSUSED */
188 int
189 zfs_open(struct inode *ip, int mode, int flag, cred_t *cr)
190 {
191 	znode_t	*zp = ITOZ(ip);
192 	zfsvfs_t *zfsvfs = ITOZSB(ip);
193 
194 	ZFS_ENTER(zfsvfs);
195 	ZFS_VERIFY_ZP(zp);
196 
197 	/* Honor ZFS_APPENDONLY file attribute */
198 	if ((mode & FMODE_WRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
199 	    ((flag & O_APPEND) == 0)) {
200 		ZFS_EXIT(zfsvfs);
201 		return (SET_ERROR(EPERM));
202 	}
203 
204 	/* Virus scan eligible files on open */
205 	if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
206 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
207 		if (zfs_vscan(ip, cr, 0) != 0) {
208 			ZFS_EXIT(zfsvfs);
209 			return (SET_ERROR(EACCES));
210 		}
211 	}
212 
213 	/* Keep a count of the synchronous opens in the znode */
214 	if (flag & O_SYNC)
215 		atomic_inc_32(&zp->z_sync_cnt);
216 
217 	ZFS_EXIT(zfsvfs);
218 	return (0);
219 }
220 
221 /* ARGSUSED */
222 int
223 zfs_close(struct inode *ip, int flag, cred_t *cr)
224 {
225 	znode_t	*zp = ITOZ(ip);
226 	zfsvfs_t *zfsvfs = ITOZSB(ip);
227 
228 	ZFS_ENTER(zfsvfs);
229 	ZFS_VERIFY_ZP(zp);
230 
231 	/* Decrement the synchronous opens in the znode */
232 	if (flag & O_SYNC)
233 		atomic_dec_32(&zp->z_sync_cnt);
234 
235 	if (!zfs_has_ctldir(zp) && zfsvfs->z_vscan && S_ISREG(ip->i_mode) &&
236 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
237 		VERIFY(zfs_vscan(ip, cr, 1) == 0);
238 
239 	ZFS_EXIT(zfsvfs);
240 	return (0);
241 }
242 
243 #if defined(_KERNEL)
244 /*
245  * When a file is memory mapped, we must keep the IO data synchronized
246  * between the DMU cache and the memory mapped pages.  What this means:
247  *
248  * On Write:	If we find a memory mapped page, we write to *both*
249  *		the page and the dmu buffer.
250  */
251 void
252 update_pages(znode_t *zp, int64_t start, int len, objset_t *os)
253 {
254 	struct inode *ip = ZTOI(zp);
255 	struct address_space *mp = ip->i_mapping;
256 	struct page *pp;
257 	uint64_t nbytes;
258 	int64_t	off;
259 	void *pb;
260 
261 	off = start & (PAGE_SIZE-1);
262 	for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
263 		nbytes = MIN(PAGE_SIZE - off, len);
264 
265 		pp = find_lock_page(mp, start >> PAGE_SHIFT);
266 		if (pp) {
267 			if (mapping_writably_mapped(mp))
268 				flush_dcache_page(pp);
269 
270 			pb = kmap(pp);
271 			(void) dmu_read(os, zp->z_id, start + off, nbytes,
272 			    pb + off, DMU_READ_PREFETCH);
273 			kunmap(pp);
274 
275 			if (mapping_writably_mapped(mp))
276 				flush_dcache_page(pp);
277 
278 			mark_page_accessed(pp);
279 			SetPageUptodate(pp);
280 			ClearPageError(pp);
281 			unlock_page(pp);
282 			put_page(pp);
283 		}
284 
285 		len -= nbytes;
286 		off = 0;
287 	}
288 }
289 
290 /*
291  * When a file is memory mapped, we must keep the IO data synchronized
292  * between the DMU cache and the memory mapped pages.  What this means:
293  *
294  * On Read:	We "read" preferentially from memory mapped pages,
295  *		else we default from the dmu buffer.
296  *
297  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
298  *	 the file is memory mapped.
299  */
300 int
301 mappedread(znode_t *zp, int nbytes, uio_t *uio)
302 {
303 	struct inode *ip = ZTOI(zp);
304 	struct address_space *mp = ip->i_mapping;
305 	struct page *pp;
306 	int64_t	start, off;
307 	uint64_t bytes;
308 	int len = nbytes;
309 	int error = 0;
310 	void *pb;
311 
312 	start = uio->uio_loffset;
313 	off = start & (PAGE_SIZE-1);
314 	for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) {
315 		bytes = MIN(PAGE_SIZE - off, len);
316 
317 		pp = find_lock_page(mp, start >> PAGE_SHIFT);
318 		if (pp) {
319 			ASSERT(PageUptodate(pp));
320 			unlock_page(pp);
321 
322 			pb = kmap(pp);
323 			error = uiomove(pb + off, bytes, UIO_READ, uio);
324 			kunmap(pp);
325 
326 			if (mapping_writably_mapped(mp))
327 				flush_dcache_page(pp);
328 
329 			mark_page_accessed(pp);
330 			put_page(pp);
331 		} else {
332 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
333 			    uio, bytes);
334 		}
335 
336 		len -= bytes;
337 		off = 0;
338 		if (error)
339 			break;
340 	}
341 	return (error);
342 }
343 #endif /* _KERNEL */
344 
345 unsigned long zfs_delete_blocks = DMU_MAX_DELETEBLKCNT;
346 
347 /*
348  * Write the bytes to a file.
349  *
350  *	IN:	zp	- znode of file to be written to
351  *		data	- bytes to write
352  *		len	- number of bytes to write
353  *		pos	- offset to start writing at
354  *
355  *	OUT:	resid	- remaining bytes to write
356  *
357  *	RETURN:	0 if success
358  *		positive error code if failure.  EIO is	returned
359  *		for a short write when residp isn't provided.
360  *
361  * Timestamps:
362  *	zp - ctime|mtime updated if byte count > 0
363  */
364 int
365 zfs_write_simple(znode_t *zp, const void *data, size_t len,
366     loff_t pos, size_t *residp)
367 {
368 	fstrans_cookie_t cookie;
369 	int error;
370 
371 	struct iovec iov;
372 	iov.iov_base = (void *)data;
373 	iov.iov_len = len;
374 
375 	uio_t uio;
376 	uio_iovec_init(&uio, &iov, 1, pos, UIO_SYSSPACE, len, 0);
377 
378 	cookie = spl_fstrans_mark();
379 	error = zfs_write(zp, &uio, 0, kcred);
380 	spl_fstrans_unmark(cookie);
381 
382 	if (error == 0) {
383 		if (residp != NULL)
384 			*residp = uio_resid(&uio);
385 		else if (uio_resid(&uio) != 0)
386 			error = SET_ERROR(EIO);
387 	}
388 
389 	return (error);
390 }
391 
392 void
393 zfs_zrele_async(znode_t *zp)
394 {
395 	struct inode *ip = ZTOI(zp);
396 	objset_t *os = ITOZSB(ip)->z_os;
397 
398 	ASSERT(atomic_read(&ip->i_count) > 0);
399 	ASSERT(os != NULL);
400 
401 	if (atomic_read(&ip->i_count) == 1)
402 		VERIFY(taskq_dispatch(dsl_pool_zrele_taskq(dmu_objset_pool(os)),
403 		    (task_func_t *)iput, ip, TQ_SLEEP) != TASKQID_INVALID);
404 	else
405 		zrele(zp);
406 }
407 
408 
409 /*
410  * Lookup an entry in a directory, or an extended attribute directory.
411  * If it exists, return a held inode reference for it.
412  *
413  *	IN:	zdp	- znode of directory to search.
414  *		nm	- name of entry to lookup.
415  *		flags	- LOOKUP_XATTR set if looking for an attribute.
416  *		cr	- credentials of caller.
417  *		direntflags - directory lookup flags
418  *		realpnp - returned pathname.
419  *
420  *	OUT:	zpp	- znode of located entry, NULL if not found.
421  *
422  *	RETURN:	0 on success, error code on failure.
423  *
424  * Timestamps:
425  *	NA
426  */
427 /* ARGSUSED */
428 int
429 zfs_lookup(znode_t *zdp, char *nm, znode_t **zpp, int flags, cred_t *cr,
430     int *direntflags, pathname_t *realpnp)
431 {
432 	zfsvfs_t *zfsvfs = ZTOZSB(zdp);
433 	int error = 0;
434 
435 	/*
436 	 * Fast path lookup, however we must skip DNLC lookup
437 	 * for case folding or normalizing lookups because the
438 	 * DNLC code only stores the passed in name.  This means
439 	 * creating 'a' and removing 'A' on a case insensitive
440 	 * file system would work, but DNLC still thinks 'a'
441 	 * exists and won't let you create it again on the next
442 	 * pass through fast path.
443 	 */
444 	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
445 
446 		if (!S_ISDIR(ZTOI(zdp)->i_mode)) {
447 			return (SET_ERROR(ENOTDIR));
448 		} else if (zdp->z_sa_hdl == NULL) {
449 			return (SET_ERROR(EIO));
450 		}
451 
452 		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
453 			error = zfs_fastaccesschk_execute(zdp, cr);
454 			if (!error) {
455 				*zpp = zdp;
456 				zhold(*zpp);
457 				return (0);
458 			}
459 			return (error);
460 		}
461 	}
462 
463 	ZFS_ENTER(zfsvfs);
464 	ZFS_VERIFY_ZP(zdp);
465 
466 	*zpp = NULL;
467 
468 	if (flags & LOOKUP_XATTR) {
469 		/*
470 		 * We don't allow recursive attributes..
471 		 * Maybe someday we will.
472 		 */
473 		if (zdp->z_pflags & ZFS_XATTR) {
474 			ZFS_EXIT(zfsvfs);
475 			return (SET_ERROR(EINVAL));
476 		}
477 
478 		if ((error = zfs_get_xattrdir(zdp, zpp, cr, flags))) {
479 			ZFS_EXIT(zfsvfs);
480 			return (error);
481 		}
482 
483 		/*
484 		 * Do we have permission to get into attribute directory?
485 		 */
486 
487 		if ((error = zfs_zaccess(*zpp, ACE_EXECUTE, 0,
488 		    B_FALSE, cr))) {
489 			zrele(*zpp);
490 			*zpp = NULL;
491 		}
492 
493 		ZFS_EXIT(zfsvfs);
494 		return (error);
495 	}
496 
497 	if (!S_ISDIR(ZTOI(zdp)->i_mode)) {
498 		ZFS_EXIT(zfsvfs);
499 		return (SET_ERROR(ENOTDIR));
500 	}
501 
502 	/*
503 	 * Check accessibility of directory.
504 	 */
505 
506 	if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr))) {
507 		ZFS_EXIT(zfsvfs);
508 		return (error);
509 	}
510 
511 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
512 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
513 		ZFS_EXIT(zfsvfs);
514 		return (SET_ERROR(EILSEQ));
515 	}
516 
517 	error = zfs_dirlook(zdp, nm, zpp, flags, direntflags, realpnp);
518 	if ((error == 0) && (*zpp))
519 		zfs_inode_update(*zpp);
520 
521 	ZFS_EXIT(zfsvfs);
522 	return (error);
523 }
524 
525 /*
526  * Attempt to create a new entry in a directory.  If the entry
527  * already exists, truncate the file if permissible, else return
528  * an error.  Return the ip of the created or trunc'd file.
529  *
530  *	IN:	dzp	- znode of directory to put new file entry in.
531  *		name	- name of new file entry.
532  *		vap	- attributes of new file.
533  *		excl	- flag indicating exclusive or non-exclusive mode.
534  *		mode	- mode to open file with.
535  *		cr	- credentials of caller.
536  *		flag	- file flag.
537  *		vsecp	- ACL to be set
538  *
539  *	OUT:	zpp	- znode of created or trunc'd entry.
540  *
541  *	RETURN:	0 on success, error code on failure.
542  *
543  * Timestamps:
544  *	dzp - ctime|mtime updated if new entry created
545  *	 zp - ctime|mtime always, atime if new
546  */
547 
548 /* ARGSUSED */
549 int
550 zfs_create(znode_t *dzp, char *name, vattr_t *vap, int excl,
551     int mode, znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp)
552 {
553 	znode_t		*zp;
554 	zfsvfs_t	*zfsvfs = ZTOZSB(dzp);
555 	zilog_t		*zilog;
556 	objset_t	*os;
557 	zfs_dirlock_t	*dl;
558 	dmu_tx_t	*tx;
559 	int		error;
560 	uid_t		uid;
561 	gid_t		gid;
562 	zfs_acl_ids_t   acl_ids;
563 	boolean_t	fuid_dirtied;
564 	boolean_t	have_acl = B_FALSE;
565 	boolean_t	waited = B_FALSE;
566 
567 	/*
568 	 * If we have an ephemeral id, ACL, or XVATTR then
569 	 * make sure file system is at proper version
570 	 */
571 
572 	gid = crgetgid(cr);
573 	uid = crgetuid(cr);
574 
575 	if (zfsvfs->z_use_fuids == B_FALSE &&
576 	    (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
577 		return (SET_ERROR(EINVAL));
578 
579 	if (name == NULL)
580 		return (SET_ERROR(EINVAL));
581 
582 	ZFS_ENTER(zfsvfs);
583 	ZFS_VERIFY_ZP(dzp);
584 	os = zfsvfs->z_os;
585 	zilog = zfsvfs->z_log;
586 
587 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
588 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
589 		ZFS_EXIT(zfsvfs);
590 		return (SET_ERROR(EILSEQ));
591 	}
592 
593 	if (vap->va_mask & ATTR_XVATTR) {
594 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
595 		    crgetuid(cr), cr, vap->va_mode)) != 0) {
596 			ZFS_EXIT(zfsvfs);
597 			return (error);
598 		}
599 	}
600 
601 top:
602 	*zpp = NULL;
603 	if (*name == '\0') {
604 		/*
605 		 * Null component name refers to the directory itself.
606 		 */
607 		zhold(dzp);
608 		zp = dzp;
609 		dl = NULL;
610 		error = 0;
611 	} else {
612 		/* possible igrab(zp) */
613 		int zflg = 0;
614 
615 		if (flag & FIGNORECASE)
616 			zflg |= ZCILOOK;
617 
618 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
619 		    NULL, NULL);
620 		if (error) {
621 			if (have_acl)
622 				zfs_acl_ids_free(&acl_ids);
623 			if (strcmp(name, "..") == 0)
624 				error = SET_ERROR(EISDIR);
625 			ZFS_EXIT(zfsvfs);
626 			return (error);
627 		}
628 	}
629 
630 	if (zp == NULL) {
631 		uint64_t txtype;
632 		uint64_t projid = ZFS_DEFAULT_PROJID;
633 
634 		/*
635 		 * Create a new file object and update the directory
636 		 * to reference it.
637 		 */
638 		if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
639 			if (have_acl)
640 				zfs_acl_ids_free(&acl_ids);
641 			goto out;
642 		}
643 
644 		/*
645 		 * We only support the creation of regular files in
646 		 * extended attribute directories.
647 		 */
648 
649 		if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) {
650 			if (have_acl)
651 				zfs_acl_ids_free(&acl_ids);
652 			error = SET_ERROR(EINVAL);
653 			goto out;
654 		}
655 
656 		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
657 		    cr, vsecp, &acl_ids)) != 0)
658 			goto out;
659 		have_acl = B_TRUE;
660 
661 		if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode))
662 			projid = zfs_inherit_projid(dzp);
663 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) {
664 			zfs_acl_ids_free(&acl_ids);
665 			error = SET_ERROR(EDQUOT);
666 			goto out;
667 		}
668 
669 		tx = dmu_tx_create(os);
670 
671 		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
672 		    ZFS_SA_BASE_ATTR_SIZE);
673 
674 		fuid_dirtied = zfsvfs->z_fuid_dirty;
675 		if (fuid_dirtied)
676 			zfs_fuid_txhold(zfsvfs, tx);
677 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
678 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
679 		if (!zfsvfs->z_use_sa &&
680 		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
681 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
682 			    0, acl_ids.z_aclp->z_acl_bytes);
683 		}
684 
685 		error = dmu_tx_assign(tx,
686 		    (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
687 		if (error) {
688 			zfs_dirent_unlock(dl);
689 			if (error == ERESTART) {
690 				waited = B_TRUE;
691 				dmu_tx_wait(tx);
692 				dmu_tx_abort(tx);
693 				goto top;
694 			}
695 			zfs_acl_ids_free(&acl_ids);
696 			dmu_tx_abort(tx);
697 			ZFS_EXIT(zfsvfs);
698 			return (error);
699 		}
700 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
701 
702 		error = zfs_link_create(dl, zp, tx, ZNEW);
703 		if (error != 0) {
704 			/*
705 			 * Since, we failed to add the directory entry for it,
706 			 * delete the newly created dnode.
707 			 */
708 			zfs_znode_delete(zp, tx);
709 			remove_inode_hash(ZTOI(zp));
710 			zfs_acl_ids_free(&acl_ids);
711 			dmu_tx_commit(tx);
712 			goto out;
713 		}
714 
715 		if (fuid_dirtied)
716 			zfs_fuid_sync(zfsvfs, tx);
717 
718 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
719 		if (flag & FIGNORECASE)
720 			txtype |= TX_CI;
721 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
722 		    vsecp, acl_ids.z_fuidp, vap);
723 		zfs_acl_ids_free(&acl_ids);
724 		dmu_tx_commit(tx);
725 	} else {
726 		int aflags = (flag & O_APPEND) ? V_APPEND : 0;
727 
728 		if (have_acl)
729 			zfs_acl_ids_free(&acl_ids);
730 		have_acl = B_FALSE;
731 
732 		/*
733 		 * A directory entry already exists for this name.
734 		 */
735 		/*
736 		 * Can't truncate an existing file if in exclusive mode.
737 		 */
738 		if (excl) {
739 			error = SET_ERROR(EEXIST);
740 			goto out;
741 		}
742 		/*
743 		 * Can't open a directory for writing.
744 		 */
745 		if (S_ISDIR(ZTOI(zp)->i_mode)) {
746 			error = SET_ERROR(EISDIR);
747 			goto out;
748 		}
749 		/*
750 		 * Verify requested access to file.
751 		 */
752 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
753 			goto out;
754 		}
755 
756 		mutex_enter(&dzp->z_lock);
757 		dzp->z_seq++;
758 		mutex_exit(&dzp->z_lock);
759 
760 		/*
761 		 * Truncate regular files if requested.
762 		 */
763 		if (S_ISREG(ZTOI(zp)->i_mode) &&
764 		    (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) {
765 			/* we can't hold any locks when calling zfs_freesp() */
766 			if (dl) {
767 				zfs_dirent_unlock(dl);
768 				dl = NULL;
769 			}
770 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
771 		}
772 	}
773 out:
774 
775 	if (dl)
776 		zfs_dirent_unlock(dl);
777 
778 	if (error) {
779 		if (zp)
780 			zrele(zp);
781 	} else {
782 		zfs_inode_update(dzp);
783 		zfs_inode_update(zp);
784 		*zpp = zp;
785 	}
786 
787 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
788 		zil_commit(zilog, 0);
789 
790 	ZFS_EXIT(zfsvfs);
791 	return (error);
792 }
793 
794 /* ARGSUSED */
795 int
796 zfs_tmpfile(struct inode *dip, vattr_t *vap, int excl,
797     int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp)
798 {
799 	znode_t		*zp = NULL, *dzp = ITOZ(dip);
800 	zfsvfs_t	*zfsvfs = ITOZSB(dip);
801 	objset_t	*os;
802 	dmu_tx_t	*tx;
803 	int		error;
804 	uid_t		uid;
805 	gid_t		gid;
806 	zfs_acl_ids_t   acl_ids;
807 	uint64_t	projid = ZFS_DEFAULT_PROJID;
808 	boolean_t	fuid_dirtied;
809 	boolean_t	have_acl = B_FALSE;
810 	boolean_t	waited = B_FALSE;
811 
812 	/*
813 	 * If we have an ephemeral id, ACL, or XVATTR then
814 	 * make sure file system is at proper version
815 	 */
816 
817 	gid = crgetgid(cr);
818 	uid = crgetuid(cr);
819 
820 	if (zfsvfs->z_use_fuids == B_FALSE &&
821 	    (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
822 		return (SET_ERROR(EINVAL));
823 
824 	ZFS_ENTER(zfsvfs);
825 	ZFS_VERIFY_ZP(dzp);
826 	os = zfsvfs->z_os;
827 
828 	if (vap->va_mask & ATTR_XVATTR) {
829 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
830 		    crgetuid(cr), cr, vap->va_mode)) != 0) {
831 			ZFS_EXIT(zfsvfs);
832 			return (error);
833 		}
834 	}
835 
836 top:
837 	*ipp = NULL;
838 
839 	/*
840 	 * Create a new file object and update the directory
841 	 * to reference it.
842 	 */
843 	if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
844 		if (have_acl)
845 			zfs_acl_ids_free(&acl_ids);
846 		goto out;
847 	}
848 
849 	if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
850 	    cr, vsecp, &acl_ids)) != 0)
851 		goto out;
852 	have_acl = B_TRUE;
853 
854 	if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode))
855 		projid = zfs_inherit_projid(dzp);
856 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) {
857 		zfs_acl_ids_free(&acl_ids);
858 		error = SET_ERROR(EDQUOT);
859 		goto out;
860 	}
861 
862 	tx = dmu_tx_create(os);
863 
864 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
865 	    ZFS_SA_BASE_ATTR_SIZE);
866 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
867 
868 	fuid_dirtied = zfsvfs->z_fuid_dirty;
869 	if (fuid_dirtied)
870 		zfs_fuid_txhold(zfsvfs, tx);
871 	if (!zfsvfs->z_use_sa &&
872 	    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
873 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
874 		    0, acl_ids.z_aclp->z_acl_bytes);
875 	}
876 	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
877 	if (error) {
878 		if (error == ERESTART) {
879 			waited = B_TRUE;
880 			dmu_tx_wait(tx);
881 			dmu_tx_abort(tx);
882 			goto top;
883 		}
884 		zfs_acl_ids_free(&acl_ids);
885 		dmu_tx_abort(tx);
886 		ZFS_EXIT(zfsvfs);
887 		return (error);
888 	}
889 	zfs_mknode(dzp, vap, tx, cr, IS_TMPFILE, &zp, &acl_ids);
890 
891 	if (fuid_dirtied)
892 		zfs_fuid_sync(zfsvfs, tx);
893 
894 	/* Add to unlinked set */
895 	zp->z_unlinked = B_TRUE;
896 	zfs_unlinked_add(zp, tx);
897 	zfs_acl_ids_free(&acl_ids);
898 	dmu_tx_commit(tx);
899 out:
900 
901 	if (error) {
902 		if (zp)
903 			zrele(zp);
904 	} else {
905 		zfs_inode_update(dzp);
906 		zfs_inode_update(zp);
907 		*ipp = ZTOI(zp);
908 	}
909 
910 	ZFS_EXIT(zfsvfs);
911 	return (error);
912 }
913 
914 /*
915  * Remove an entry from a directory.
916  *
917  *	IN:	dzp	- znode of directory to remove entry from.
918  *		name	- name of entry to remove.
919  *		cr	- credentials of caller.
920  *		flags	- case flags.
921  *
922  *	RETURN:	0 if success
923  *		error code if failure
924  *
925  * Timestamps:
926  *	dzp - ctime|mtime
927  *	 ip - ctime (if nlink > 0)
928  */
929 
930 uint64_t null_xattr = 0;
931 
932 /*ARGSUSED*/
933 int
934 zfs_remove(znode_t *dzp, char *name, cred_t *cr, int flags)
935 {
936 	znode_t		*zp;
937 	znode_t		*xzp;
938 	zfsvfs_t	*zfsvfs = ZTOZSB(dzp);
939 	zilog_t		*zilog;
940 	uint64_t	acl_obj, xattr_obj;
941 	uint64_t	xattr_obj_unlinked = 0;
942 	uint64_t	obj = 0;
943 	uint64_t	links;
944 	zfs_dirlock_t	*dl;
945 	dmu_tx_t	*tx;
946 	boolean_t	may_delete_now, delete_now = FALSE;
947 	boolean_t	unlinked, toobig = FALSE;
948 	uint64_t	txtype;
949 	pathname_t	*realnmp = NULL;
950 	pathname_t	realnm;
951 	int		error;
952 	int		zflg = ZEXISTS;
953 	boolean_t	waited = B_FALSE;
954 
955 	if (name == NULL)
956 		return (SET_ERROR(EINVAL));
957 
958 	ZFS_ENTER(zfsvfs);
959 	ZFS_VERIFY_ZP(dzp);
960 	zilog = zfsvfs->z_log;
961 
962 	if (flags & FIGNORECASE) {
963 		zflg |= ZCILOOK;
964 		pn_alloc(&realnm);
965 		realnmp = &realnm;
966 	}
967 
968 top:
969 	xattr_obj = 0;
970 	xzp = NULL;
971 	/*
972 	 * Attempt to lock directory; fail if entry doesn't exist.
973 	 */
974 	if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
975 	    NULL, realnmp))) {
976 		if (realnmp)
977 			pn_free(realnmp);
978 		ZFS_EXIT(zfsvfs);
979 		return (error);
980 	}
981 
982 	if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
983 		goto out;
984 	}
985 
986 	/*
987 	 * Need to use rmdir for removing directories.
988 	 */
989 	if (S_ISDIR(ZTOI(zp)->i_mode)) {
990 		error = SET_ERROR(EPERM);
991 		goto out;
992 	}
993 
994 	mutex_enter(&zp->z_lock);
995 	may_delete_now = atomic_read(&ZTOI(zp)->i_count) == 1 &&
996 	    !(zp->z_is_mapped);
997 	mutex_exit(&zp->z_lock);
998 
999 	/*
1000 	 * We may delete the znode now, or we may put it in the unlinked set;
1001 	 * it depends on whether we're the last link, and on whether there are
1002 	 * other holds on the inode.  So we dmu_tx_hold() the right things to
1003 	 * allow for either case.
1004 	 */
1005 	obj = zp->z_id;
1006 	tx = dmu_tx_create(zfsvfs->z_os);
1007 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1008 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1009 	zfs_sa_upgrade_txholds(tx, zp);
1010 	zfs_sa_upgrade_txholds(tx, dzp);
1011 	if (may_delete_now) {
1012 		toobig = zp->z_size > zp->z_blksz * zfs_delete_blocks;
1013 		/* if the file is too big, only hold_free a token amount */
1014 		dmu_tx_hold_free(tx, zp->z_id, 0,
1015 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1016 	}
1017 
1018 	/* are there any extended attributes? */
1019 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1020 	    &xattr_obj, sizeof (xattr_obj));
1021 	if (error == 0 && xattr_obj) {
1022 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1023 		ASSERT0(error);
1024 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1025 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1026 	}
1027 
1028 	mutex_enter(&zp->z_lock);
1029 	if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1030 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1031 	mutex_exit(&zp->z_lock);
1032 
1033 	/* charge as an update -- would be nice not to charge at all */
1034 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1035 
1036 	/*
1037 	 * Mark this transaction as typically resulting in a net free of space
1038 	 */
1039 	dmu_tx_mark_netfree(tx);
1040 
1041 	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1042 	if (error) {
1043 		zfs_dirent_unlock(dl);
1044 		if (error == ERESTART) {
1045 			waited = B_TRUE;
1046 			dmu_tx_wait(tx);
1047 			dmu_tx_abort(tx);
1048 			zrele(zp);
1049 			if (xzp)
1050 				zrele(xzp);
1051 			goto top;
1052 		}
1053 		if (realnmp)
1054 			pn_free(realnmp);
1055 		dmu_tx_abort(tx);
1056 		zrele(zp);
1057 		if (xzp)
1058 			zrele(xzp);
1059 		ZFS_EXIT(zfsvfs);
1060 		return (error);
1061 	}
1062 
1063 	/*
1064 	 * Remove the directory entry.
1065 	 */
1066 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1067 
1068 	if (error) {
1069 		dmu_tx_commit(tx);
1070 		goto out;
1071 	}
1072 
1073 	if (unlinked) {
1074 		/*
1075 		 * Hold z_lock so that we can make sure that the ACL obj
1076 		 * hasn't changed.  Could have been deleted due to
1077 		 * zfs_sa_upgrade().
1078 		 */
1079 		mutex_enter(&zp->z_lock);
1080 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1081 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1082 		delete_now = may_delete_now && !toobig &&
1083 		    atomic_read(&ZTOI(zp)->i_count) == 1 &&
1084 		    !(zp->z_is_mapped) && xattr_obj == xattr_obj_unlinked &&
1085 		    zfs_external_acl(zp) == acl_obj;
1086 	}
1087 
1088 	if (delete_now) {
1089 		if (xattr_obj_unlinked) {
1090 			ASSERT3U(ZTOI(xzp)->i_nlink, ==, 2);
1091 			mutex_enter(&xzp->z_lock);
1092 			xzp->z_unlinked = B_TRUE;
1093 			clear_nlink(ZTOI(xzp));
1094 			links = 0;
1095 			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1096 			    &links, sizeof (links), tx);
1097 			ASSERT3U(error,  ==,  0);
1098 			mutex_exit(&xzp->z_lock);
1099 			zfs_unlinked_add(xzp, tx);
1100 
1101 			if (zp->z_is_sa)
1102 				error = sa_remove(zp->z_sa_hdl,
1103 				    SA_ZPL_XATTR(zfsvfs), tx);
1104 			else
1105 				error = sa_update(zp->z_sa_hdl,
1106 				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
1107 				    sizeof (uint64_t), tx);
1108 			ASSERT0(error);
1109 		}
1110 		/*
1111 		 * Add to the unlinked set because a new reference could be
1112 		 * taken concurrently resulting in a deferred destruction.
1113 		 */
1114 		zfs_unlinked_add(zp, tx);
1115 		mutex_exit(&zp->z_lock);
1116 	} else if (unlinked) {
1117 		mutex_exit(&zp->z_lock);
1118 		zfs_unlinked_add(zp, tx);
1119 	}
1120 
1121 	txtype = TX_REMOVE;
1122 	if (flags & FIGNORECASE)
1123 		txtype |= TX_CI;
1124 	zfs_log_remove(zilog, tx, txtype, dzp, name, obj, unlinked);
1125 
1126 	dmu_tx_commit(tx);
1127 out:
1128 	if (realnmp)
1129 		pn_free(realnmp);
1130 
1131 	zfs_dirent_unlock(dl);
1132 	zfs_inode_update(dzp);
1133 	zfs_inode_update(zp);
1134 
1135 	if (delete_now)
1136 		zrele(zp);
1137 	else
1138 		zfs_zrele_async(zp);
1139 
1140 	if (xzp) {
1141 		zfs_inode_update(xzp);
1142 		zfs_zrele_async(xzp);
1143 	}
1144 
1145 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1146 		zil_commit(zilog, 0);
1147 
1148 	ZFS_EXIT(zfsvfs);
1149 	return (error);
1150 }
1151 
1152 /*
1153  * Create a new directory and insert it into dzp using the name
1154  * provided.  Return a pointer to the inserted directory.
1155  *
1156  *	IN:	dzp	- znode of directory to add subdir to.
1157  *		dirname	- name of new directory.
1158  *		vap	- attributes of new directory.
1159  *		cr	- credentials of caller.
1160  *		flags	- case flags.
1161  *		vsecp	- ACL to be set
1162  *
1163  *	OUT:	zpp	- znode of created directory.
1164  *
1165  *	RETURN:	0 if success
1166  *		error code if failure
1167  *
1168  * Timestamps:
1169  *	dzp - ctime|mtime updated
1170  *	zpp - ctime|mtime|atime updated
1171  */
1172 /*ARGSUSED*/
1173 int
1174 zfs_mkdir(znode_t *dzp, char *dirname, vattr_t *vap, znode_t **zpp,
1175     cred_t *cr, int flags, vsecattr_t *vsecp)
1176 {
1177 	znode_t		*zp;
1178 	zfsvfs_t	*zfsvfs = ZTOZSB(dzp);
1179 	zilog_t		*zilog;
1180 	zfs_dirlock_t	*dl;
1181 	uint64_t	txtype;
1182 	dmu_tx_t	*tx;
1183 	int		error;
1184 	int		zf = ZNEW;
1185 	uid_t		uid;
1186 	gid_t		gid = crgetgid(cr);
1187 	zfs_acl_ids_t   acl_ids;
1188 	boolean_t	fuid_dirtied;
1189 	boolean_t	waited = B_FALSE;
1190 
1191 	ASSERT(S_ISDIR(vap->va_mode));
1192 
1193 	/*
1194 	 * If we have an ephemeral id, ACL, or XVATTR then
1195 	 * make sure file system is at proper version
1196 	 */
1197 
1198 	uid = crgetuid(cr);
1199 	if (zfsvfs->z_use_fuids == B_FALSE &&
1200 	    (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1201 		return (SET_ERROR(EINVAL));
1202 
1203 	if (dirname == NULL)
1204 		return (SET_ERROR(EINVAL));
1205 
1206 	ZFS_ENTER(zfsvfs);
1207 	ZFS_VERIFY_ZP(dzp);
1208 	zilog = zfsvfs->z_log;
1209 
1210 	if (dzp->z_pflags & ZFS_XATTR) {
1211 		ZFS_EXIT(zfsvfs);
1212 		return (SET_ERROR(EINVAL));
1213 	}
1214 
1215 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1216 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1217 		ZFS_EXIT(zfsvfs);
1218 		return (SET_ERROR(EILSEQ));
1219 	}
1220 	if (flags & FIGNORECASE)
1221 		zf |= ZCILOOK;
1222 
1223 	if (vap->va_mask & ATTR_XVATTR) {
1224 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1225 		    crgetuid(cr), cr, vap->va_mode)) != 0) {
1226 			ZFS_EXIT(zfsvfs);
1227 			return (error);
1228 		}
1229 	}
1230 
1231 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1232 	    vsecp, &acl_ids)) != 0) {
1233 		ZFS_EXIT(zfsvfs);
1234 		return (error);
1235 	}
1236 	/*
1237 	 * First make sure the new directory doesn't exist.
1238 	 *
1239 	 * Existence is checked first to make sure we don't return
1240 	 * EACCES instead of EEXIST which can cause some applications
1241 	 * to fail.
1242 	 */
1243 top:
1244 	*zpp = NULL;
1245 
1246 	if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1247 	    NULL, NULL))) {
1248 		zfs_acl_ids_free(&acl_ids);
1249 		ZFS_EXIT(zfsvfs);
1250 		return (error);
1251 	}
1252 
1253 	if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr))) {
1254 		zfs_acl_ids_free(&acl_ids);
1255 		zfs_dirent_unlock(dl);
1256 		ZFS_EXIT(zfsvfs);
1257 		return (error);
1258 	}
1259 
1260 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) {
1261 		zfs_acl_ids_free(&acl_ids);
1262 		zfs_dirent_unlock(dl);
1263 		ZFS_EXIT(zfsvfs);
1264 		return (SET_ERROR(EDQUOT));
1265 	}
1266 
1267 	/*
1268 	 * Add a new entry to the directory.
1269 	 */
1270 	tx = dmu_tx_create(zfsvfs->z_os);
1271 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1272 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1273 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1274 	if (fuid_dirtied)
1275 		zfs_fuid_txhold(zfsvfs, tx);
1276 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1277 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1278 		    acl_ids.z_aclp->z_acl_bytes);
1279 	}
1280 
1281 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1282 	    ZFS_SA_BASE_ATTR_SIZE);
1283 
1284 	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1285 	if (error) {
1286 		zfs_dirent_unlock(dl);
1287 		if (error == ERESTART) {
1288 			waited = B_TRUE;
1289 			dmu_tx_wait(tx);
1290 			dmu_tx_abort(tx);
1291 			goto top;
1292 		}
1293 		zfs_acl_ids_free(&acl_ids);
1294 		dmu_tx_abort(tx);
1295 		ZFS_EXIT(zfsvfs);
1296 		return (error);
1297 	}
1298 
1299 	/*
1300 	 * Create new node.
1301 	 */
1302 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1303 
1304 	/*
1305 	 * Now put new name in parent dir.
1306 	 */
1307 	error = zfs_link_create(dl, zp, tx, ZNEW);
1308 	if (error != 0) {
1309 		zfs_znode_delete(zp, tx);
1310 		remove_inode_hash(ZTOI(zp));
1311 		goto out;
1312 	}
1313 
1314 	if (fuid_dirtied)
1315 		zfs_fuid_sync(zfsvfs, tx);
1316 
1317 	*zpp = zp;
1318 
1319 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1320 	if (flags & FIGNORECASE)
1321 		txtype |= TX_CI;
1322 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1323 	    acl_ids.z_fuidp, vap);
1324 
1325 out:
1326 	zfs_acl_ids_free(&acl_ids);
1327 
1328 	dmu_tx_commit(tx);
1329 
1330 	zfs_dirent_unlock(dl);
1331 
1332 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1333 		zil_commit(zilog, 0);
1334 
1335 	if (error != 0) {
1336 		zrele(zp);
1337 	} else {
1338 		zfs_inode_update(dzp);
1339 		zfs_inode_update(zp);
1340 	}
1341 	ZFS_EXIT(zfsvfs);
1342 	return (error);
1343 }
1344 
1345 /*
1346  * Remove a directory subdir entry.  If the current working
1347  * directory is the same as the subdir to be removed, the
1348  * remove will fail.
1349  *
1350  *	IN:	dzp	- znode of directory to remove from.
1351  *		name	- name of directory to be removed.
1352  *		cwd	- inode of current working directory.
1353  *		cr	- credentials of caller.
1354  *		flags	- case flags
1355  *
1356  *	RETURN:	0 on success, error code on failure.
1357  *
1358  * Timestamps:
1359  *	dzp - ctime|mtime updated
1360  */
1361 /*ARGSUSED*/
1362 int
1363 zfs_rmdir(znode_t *dzp, char *name, znode_t *cwd, cred_t *cr,
1364     int flags)
1365 {
1366 	znode_t		*zp;
1367 	zfsvfs_t	*zfsvfs = ZTOZSB(dzp);
1368 	zilog_t		*zilog;
1369 	zfs_dirlock_t	*dl;
1370 	dmu_tx_t	*tx;
1371 	int		error;
1372 	int		zflg = ZEXISTS;
1373 	boolean_t	waited = B_FALSE;
1374 
1375 	if (name == NULL)
1376 		return (SET_ERROR(EINVAL));
1377 
1378 	ZFS_ENTER(zfsvfs);
1379 	ZFS_VERIFY_ZP(dzp);
1380 	zilog = zfsvfs->z_log;
1381 
1382 	if (flags & FIGNORECASE)
1383 		zflg |= ZCILOOK;
1384 top:
1385 	zp = NULL;
1386 
1387 	/*
1388 	 * Attempt to lock directory; fail if entry doesn't exist.
1389 	 */
1390 	if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1391 	    NULL, NULL))) {
1392 		ZFS_EXIT(zfsvfs);
1393 		return (error);
1394 	}
1395 
1396 	if ((error = zfs_zaccess_delete(dzp, zp, cr))) {
1397 		goto out;
1398 	}
1399 
1400 	if (!S_ISDIR(ZTOI(zp)->i_mode)) {
1401 		error = SET_ERROR(ENOTDIR);
1402 		goto out;
1403 	}
1404 
1405 	if (zp == cwd) {
1406 		error = SET_ERROR(EINVAL);
1407 		goto out;
1408 	}
1409 
1410 	/*
1411 	 * Grab a lock on the directory to make sure that no one is
1412 	 * trying to add (or lookup) entries while we are removing it.
1413 	 */
1414 	rw_enter(&zp->z_name_lock, RW_WRITER);
1415 
1416 	/*
1417 	 * Grab a lock on the parent pointer to make sure we play well
1418 	 * with the treewalk and directory rename code.
1419 	 */
1420 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1421 
1422 	tx = dmu_tx_create(zfsvfs->z_os);
1423 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1424 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1425 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1426 	zfs_sa_upgrade_txholds(tx, zp);
1427 	zfs_sa_upgrade_txholds(tx, dzp);
1428 	dmu_tx_mark_netfree(tx);
1429 	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1430 	if (error) {
1431 		rw_exit(&zp->z_parent_lock);
1432 		rw_exit(&zp->z_name_lock);
1433 		zfs_dirent_unlock(dl);
1434 		if (error == ERESTART) {
1435 			waited = B_TRUE;
1436 			dmu_tx_wait(tx);
1437 			dmu_tx_abort(tx);
1438 			zrele(zp);
1439 			goto top;
1440 		}
1441 		dmu_tx_abort(tx);
1442 		zrele(zp);
1443 		ZFS_EXIT(zfsvfs);
1444 		return (error);
1445 	}
1446 
1447 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
1448 
1449 	if (error == 0) {
1450 		uint64_t txtype = TX_RMDIR;
1451 		if (flags & FIGNORECASE)
1452 			txtype |= TX_CI;
1453 		zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT,
1454 		    B_FALSE);
1455 	}
1456 
1457 	dmu_tx_commit(tx);
1458 
1459 	rw_exit(&zp->z_parent_lock);
1460 	rw_exit(&zp->z_name_lock);
1461 out:
1462 	zfs_dirent_unlock(dl);
1463 
1464 	zfs_inode_update(dzp);
1465 	zfs_inode_update(zp);
1466 	zrele(zp);
1467 
1468 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1469 		zil_commit(zilog, 0);
1470 
1471 	ZFS_EXIT(zfsvfs);
1472 	return (error);
1473 }
1474 
1475 /*
1476  * Read directory entries from the given directory cursor position and emit
1477  * name and position for each entry.
1478  *
1479  *	IN:	ip	- inode of directory to read.
1480  *		ctx	- directory entry context.
1481  *		cr	- credentials of caller.
1482  *
1483  *	RETURN:	0 if success
1484  *		error code if failure
1485  *
1486  * Timestamps:
1487  *	ip - atime updated
1488  *
1489  * Note that the low 4 bits of the cookie returned by zap is always zero.
1490  * This allows us to use the low range for "special" directory entries:
1491  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1492  * we use the offset 2 for the '.zfs' directory.
1493  */
1494 /* ARGSUSED */
1495 int
1496 zfs_readdir(struct inode *ip, zpl_dir_context_t *ctx, cred_t *cr)
1497 {
1498 	znode_t		*zp = ITOZ(ip);
1499 	zfsvfs_t	*zfsvfs = ITOZSB(ip);
1500 	objset_t	*os;
1501 	zap_cursor_t	zc;
1502 	zap_attribute_t	zap;
1503 	int		error;
1504 	uint8_t		prefetch;
1505 	uint8_t		type;
1506 	int		done = 0;
1507 	uint64_t	parent;
1508 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1509 
1510 	ZFS_ENTER(zfsvfs);
1511 	ZFS_VERIFY_ZP(zp);
1512 
1513 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
1514 	    &parent, sizeof (parent))) != 0)
1515 		goto out;
1516 
1517 	/*
1518 	 * Quit if directory has been removed (posix)
1519 	 */
1520 	if (zp->z_unlinked)
1521 		goto out;
1522 
1523 	error = 0;
1524 	os = zfsvfs->z_os;
1525 	offset = ctx->pos;
1526 	prefetch = zp->z_zn_prefetch;
1527 
1528 	/*
1529 	 * Initialize the iterator cursor.
1530 	 */
1531 	if (offset <= 3) {
1532 		/*
1533 		 * Start iteration from the beginning of the directory.
1534 		 */
1535 		zap_cursor_init(&zc, os, zp->z_id);
1536 	} else {
1537 		/*
1538 		 * The offset is a serialized cursor.
1539 		 */
1540 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1541 	}
1542 
1543 	/*
1544 	 * Transform to file-system independent format
1545 	 */
1546 	while (!done) {
1547 		uint64_t objnum;
1548 		/*
1549 		 * Special case `.', `..', and `.zfs'.
1550 		 */
1551 		if (offset == 0) {
1552 			(void) strcpy(zap.za_name, ".");
1553 			zap.za_normalization_conflict = 0;
1554 			objnum = zp->z_id;
1555 			type = DT_DIR;
1556 		} else if (offset == 1) {
1557 			(void) strcpy(zap.za_name, "..");
1558 			zap.za_normalization_conflict = 0;
1559 			objnum = parent;
1560 			type = DT_DIR;
1561 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
1562 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1563 			zap.za_normalization_conflict = 0;
1564 			objnum = ZFSCTL_INO_ROOT;
1565 			type = DT_DIR;
1566 		} else {
1567 			/*
1568 			 * Grab next entry.
1569 			 */
1570 			if ((error = zap_cursor_retrieve(&zc, &zap))) {
1571 				if (error == ENOENT)
1572 					break;
1573 				else
1574 					goto update;
1575 			}
1576 
1577 			/*
1578 			 * Allow multiple entries provided the first entry is
1579 			 * the object id.  Non-zpl consumers may safely make
1580 			 * use of the additional space.
1581 			 *
1582 			 * XXX: This should be a feature flag for compatibility
1583 			 */
1584 			if (zap.za_integer_length != 8 ||
1585 			    zap.za_num_integers == 0) {
1586 				cmn_err(CE_WARN, "zap_readdir: bad directory "
1587 				    "entry, obj = %lld, offset = %lld, "
1588 				    "length = %d, num = %lld\n",
1589 				    (u_longlong_t)zp->z_id,
1590 				    (u_longlong_t)offset,
1591 				    zap.za_integer_length,
1592 				    (u_longlong_t)zap.za_num_integers);
1593 				error = SET_ERROR(ENXIO);
1594 				goto update;
1595 			}
1596 
1597 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
1598 			type = ZFS_DIRENT_TYPE(zap.za_first_integer);
1599 		}
1600 
1601 		done = !zpl_dir_emit(ctx, zap.za_name, strlen(zap.za_name),
1602 		    objnum, type);
1603 		if (done)
1604 			break;
1605 
1606 		/* Prefetch znode */
1607 		if (prefetch) {
1608 			dmu_prefetch(os, objnum, 0, 0, 0,
1609 			    ZIO_PRIORITY_SYNC_READ);
1610 		}
1611 
1612 		/*
1613 		 * Move to the next entry, fill in the previous offset.
1614 		 */
1615 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
1616 			zap_cursor_advance(&zc);
1617 			offset = zap_cursor_serialize(&zc);
1618 		} else {
1619 			offset += 1;
1620 		}
1621 		ctx->pos = offset;
1622 	}
1623 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
1624 
1625 update:
1626 	zap_cursor_fini(&zc);
1627 	if (error == ENOENT)
1628 		error = 0;
1629 out:
1630 	ZFS_EXIT(zfsvfs);
1631 
1632 	return (error);
1633 }
1634 
1635 /*
1636  * Get the basic file attributes and place them in the provided kstat
1637  * structure.  The inode is assumed to be the authoritative source
1638  * for most of the attributes.  However, the znode currently has the
1639  * authoritative atime, blksize, and block count.
1640  *
1641  *	IN:	ip	- inode of file.
1642  *
1643  *	OUT:	sp	- kstat values.
1644  *
1645  *	RETURN:	0 (always succeeds)
1646  */
1647 /* ARGSUSED */
1648 int
1649 zfs_getattr_fast(struct inode *ip, struct kstat *sp)
1650 {
1651 	znode_t *zp = ITOZ(ip);
1652 	zfsvfs_t *zfsvfs = ITOZSB(ip);
1653 	uint32_t blksize;
1654 	u_longlong_t nblocks;
1655 
1656 	ZFS_ENTER(zfsvfs);
1657 	ZFS_VERIFY_ZP(zp);
1658 
1659 	mutex_enter(&zp->z_lock);
1660 
1661 	generic_fillattr(ip, sp);
1662 	/*
1663 	 * +1 link count for root inode with visible '.zfs' directory.
1664 	 */
1665 	if ((zp->z_id == zfsvfs->z_root) && zfs_show_ctldir(zp))
1666 		if (sp->nlink < ZFS_LINK_MAX)
1667 			sp->nlink++;
1668 
1669 	sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
1670 	sp->blksize = blksize;
1671 	sp->blocks = nblocks;
1672 
1673 	if (unlikely(zp->z_blksz == 0)) {
1674 		/*
1675 		 * Block size hasn't been set; suggest maximal I/O transfers.
1676 		 */
1677 		sp->blksize = zfsvfs->z_max_blksz;
1678 	}
1679 
1680 	mutex_exit(&zp->z_lock);
1681 
1682 	/*
1683 	 * Required to prevent NFS client from detecting different inode
1684 	 * numbers of snapshot root dentry before and after snapshot mount.
1685 	 */
1686 	if (zfsvfs->z_issnap) {
1687 		if (ip->i_sb->s_root->d_inode == ip)
1688 			sp->ino = ZFSCTL_INO_SNAPDIRS -
1689 			    dmu_objset_id(zfsvfs->z_os);
1690 	}
1691 
1692 	ZFS_EXIT(zfsvfs);
1693 
1694 	return (0);
1695 }
1696 
1697 /*
1698  * For the operation of changing file's user/group/project, we need to
1699  * handle not only the main object that is assigned to the file directly,
1700  * but also the ones that are used by the file via hidden xattr directory.
1701  *
1702  * Because the xattr directory may contains many EA entries, as to it may
1703  * be impossible to change all of them via the transaction of changing the
1704  * main object's user/group/project attributes. Then we have to change them
1705  * via other multiple independent transactions one by one. It may be not good
1706  * solution, but we have no better idea yet.
1707  */
1708 static int
1709 zfs_setattr_dir(znode_t *dzp)
1710 {
1711 	struct inode	*dxip = ZTOI(dzp);
1712 	struct inode	*xip = NULL;
1713 	zfsvfs_t	*zfsvfs = ZTOZSB(dzp);
1714 	objset_t	*os = zfsvfs->z_os;
1715 	zap_cursor_t	zc;
1716 	zap_attribute_t	zap;
1717 	zfs_dirlock_t	*dl;
1718 	znode_t		*zp = NULL;
1719 	dmu_tx_t	*tx = NULL;
1720 	uint64_t	uid, gid;
1721 	sa_bulk_attr_t	bulk[4];
1722 	int		count;
1723 	int		err;
1724 
1725 	zap_cursor_init(&zc, os, dzp->z_id);
1726 	while ((err = zap_cursor_retrieve(&zc, &zap)) == 0) {
1727 		count = 0;
1728 		if (zap.za_integer_length != 8 || zap.za_num_integers != 1) {
1729 			err = ENXIO;
1730 			break;
1731 		}
1732 
1733 		err = zfs_dirent_lock(&dl, dzp, (char *)zap.za_name, &zp,
1734 		    ZEXISTS, NULL, NULL);
1735 		if (err == ENOENT)
1736 			goto next;
1737 		if (err)
1738 			break;
1739 
1740 		xip = ZTOI(zp);
1741 		if (KUID_TO_SUID(xip->i_uid) == KUID_TO_SUID(dxip->i_uid) &&
1742 		    KGID_TO_SGID(xip->i_gid) == KGID_TO_SGID(dxip->i_gid) &&
1743 		    zp->z_projid == dzp->z_projid)
1744 			goto next;
1745 
1746 		tx = dmu_tx_create(os);
1747 		if (!(zp->z_pflags & ZFS_PROJID))
1748 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1749 		else
1750 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1751 
1752 		err = dmu_tx_assign(tx, TXG_WAIT);
1753 		if (err)
1754 			break;
1755 
1756 		mutex_enter(&dzp->z_lock);
1757 
1758 		if (KUID_TO_SUID(xip->i_uid) != KUID_TO_SUID(dxip->i_uid)) {
1759 			xip->i_uid = dxip->i_uid;
1760 			uid = zfs_uid_read(dxip);
1761 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
1762 			    &uid, sizeof (uid));
1763 		}
1764 
1765 		if (KGID_TO_SGID(xip->i_gid) != KGID_TO_SGID(dxip->i_gid)) {
1766 			xip->i_gid = dxip->i_gid;
1767 			gid = zfs_gid_read(dxip);
1768 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL,
1769 			    &gid, sizeof (gid));
1770 		}
1771 
1772 		if (zp->z_projid != dzp->z_projid) {
1773 			if (!(zp->z_pflags & ZFS_PROJID)) {
1774 				zp->z_pflags |= ZFS_PROJID;
1775 				SA_ADD_BULK_ATTR(bulk, count,
1776 				    SA_ZPL_FLAGS(zfsvfs), NULL, &zp->z_pflags,
1777 				    sizeof (zp->z_pflags));
1778 			}
1779 
1780 			zp->z_projid = dzp->z_projid;
1781 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PROJID(zfsvfs),
1782 			    NULL, &zp->z_projid, sizeof (zp->z_projid));
1783 		}
1784 
1785 		mutex_exit(&dzp->z_lock);
1786 
1787 		if (likely(count > 0)) {
1788 			err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
1789 			dmu_tx_commit(tx);
1790 		} else {
1791 			dmu_tx_abort(tx);
1792 		}
1793 		tx = NULL;
1794 		if (err != 0 && err != ENOENT)
1795 			break;
1796 
1797 next:
1798 		if (zp) {
1799 			zrele(zp);
1800 			zp = NULL;
1801 			zfs_dirent_unlock(dl);
1802 		}
1803 		zap_cursor_advance(&zc);
1804 	}
1805 
1806 	if (tx)
1807 		dmu_tx_abort(tx);
1808 	if (zp) {
1809 		zrele(zp);
1810 		zfs_dirent_unlock(dl);
1811 	}
1812 	zap_cursor_fini(&zc);
1813 
1814 	return (err == ENOENT ? 0 : err);
1815 }
1816 
1817 /*
1818  * Set the file attributes to the values contained in the
1819  * vattr structure.
1820  *
1821  *	IN:	zp	- znode of file to be modified.
1822  *		vap	- new attribute values.
1823  *			  If ATTR_XVATTR set, then optional attrs are being set
1824  *		flags	- ATTR_UTIME set if non-default time values provided.
1825  *			- ATTR_NOACLCHECK (CIFS context only).
1826  *		cr	- credentials of caller.
1827  *
1828  *	RETURN:	0 if success
1829  *		error code if failure
1830  *
1831  * Timestamps:
1832  *	ip - ctime updated, mtime updated if size changed.
1833  */
1834 /* ARGSUSED */
1835 int
1836 zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr)
1837 {
1838 	struct inode	*ip;
1839 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
1840 	objset_t	*os = zfsvfs->z_os;
1841 	zilog_t		*zilog;
1842 	dmu_tx_t	*tx;
1843 	vattr_t		oldva;
1844 	xvattr_t	*tmpxvattr;
1845 	uint_t		mask = vap->va_mask;
1846 	uint_t		saved_mask = 0;
1847 	int		trim_mask = 0;
1848 	uint64_t	new_mode;
1849 	uint64_t	new_kuid = 0, new_kgid = 0, new_uid, new_gid;
1850 	uint64_t	xattr_obj;
1851 	uint64_t	mtime[2], ctime[2], atime[2];
1852 	uint64_t	projid = ZFS_INVALID_PROJID;
1853 	znode_t		*attrzp;
1854 	int		need_policy = FALSE;
1855 	int		err, err2 = 0;
1856 	zfs_fuid_info_t *fuidp = NULL;
1857 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
1858 	xoptattr_t	*xoap;
1859 	zfs_acl_t	*aclp;
1860 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
1861 	boolean_t	fuid_dirtied = B_FALSE;
1862 	boolean_t	handle_eadir = B_FALSE;
1863 	sa_bulk_attr_t	*bulk, *xattr_bulk;
1864 	int		count = 0, xattr_count = 0, bulks = 8;
1865 
1866 	if (mask == 0)
1867 		return (0);
1868 
1869 	ZFS_ENTER(zfsvfs);
1870 	ZFS_VERIFY_ZP(zp);
1871 	ip = ZTOI(zp);
1872 
1873 	/*
1874 	 * If this is a xvattr_t, then get a pointer to the structure of
1875 	 * optional attributes.  If this is NULL, then we have a vattr_t.
1876 	 */
1877 	xoap = xva_getxoptattr(xvap);
1878 	if (xoap != NULL && (mask & ATTR_XVATTR)) {
1879 		if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
1880 			if (!dmu_objset_projectquota_enabled(os) ||
1881 			    (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode))) {
1882 				ZFS_EXIT(zfsvfs);
1883 				return (SET_ERROR(ENOTSUP));
1884 			}
1885 
1886 			projid = xoap->xoa_projid;
1887 			if (unlikely(projid == ZFS_INVALID_PROJID)) {
1888 				ZFS_EXIT(zfsvfs);
1889 				return (SET_ERROR(EINVAL));
1890 			}
1891 
1892 			if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID)
1893 				projid = ZFS_INVALID_PROJID;
1894 			else
1895 				need_policy = TRUE;
1896 		}
1897 
1898 		if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) &&
1899 		    (xoap->xoa_projinherit !=
1900 		    ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) &&
1901 		    (!dmu_objset_projectquota_enabled(os) ||
1902 		    (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode)))) {
1903 			ZFS_EXIT(zfsvfs);
1904 			return (SET_ERROR(ENOTSUP));
1905 		}
1906 	}
1907 
1908 	zilog = zfsvfs->z_log;
1909 
1910 	/*
1911 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
1912 	 * that file system is at proper version level
1913 	 */
1914 
1915 	if (zfsvfs->z_use_fuids == B_FALSE &&
1916 	    (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) ||
1917 	    ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) ||
1918 	    (mask & ATTR_XVATTR))) {
1919 		ZFS_EXIT(zfsvfs);
1920 		return (SET_ERROR(EINVAL));
1921 	}
1922 
1923 	if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) {
1924 		ZFS_EXIT(zfsvfs);
1925 		return (SET_ERROR(EISDIR));
1926 	}
1927 
1928 	if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) {
1929 		ZFS_EXIT(zfsvfs);
1930 		return (SET_ERROR(EINVAL));
1931 	}
1932 
1933 	tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP);
1934 	xva_init(tmpxvattr);
1935 
1936 	bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP);
1937 	xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP);
1938 
1939 	/*
1940 	 * Immutable files can only alter immutable bit and atime
1941 	 */
1942 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
1943 	    ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) ||
1944 	    ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
1945 		err = SET_ERROR(EPERM);
1946 		goto out3;
1947 	}
1948 
1949 	if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
1950 		err = SET_ERROR(EPERM);
1951 		goto out3;
1952 	}
1953 
1954 	/*
1955 	 * Verify timestamps doesn't overflow 32 bits.
1956 	 * ZFS can handle large timestamps, but 32bit syscalls can't
1957 	 * handle times greater than 2039.  This check should be removed
1958 	 * once large timestamps are fully supported.
1959 	 */
1960 	if (mask & (ATTR_ATIME | ATTR_MTIME)) {
1961 		if (((mask & ATTR_ATIME) &&
1962 		    TIMESPEC_OVERFLOW(&vap->va_atime)) ||
1963 		    ((mask & ATTR_MTIME) &&
1964 		    TIMESPEC_OVERFLOW(&vap->va_mtime))) {
1965 			err = SET_ERROR(EOVERFLOW);
1966 			goto out3;
1967 		}
1968 	}
1969 
1970 top:
1971 	attrzp = NULL;
1972 	aclp = NULL;
1973 
1974 	/* Can this be moved to before the top label? */
1975 	if (zfs_is_readonly(zfsvfs)) {
1976 		err = SET_ERROR(EROFS);
1977 		goto out3;
1978 	}
1979 
1980 	/*
1981 	 * First validate permissions
1982 	 */
1983 
1984 	if (mask & ATTR_SIZE) {
1985 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
1986 		if (err)
1987 			goto out3;
1988 
1989 		/*
1990 		 * XXX - Note, we are not providing any open
1991 		 * mode flags here (like FNDELAY), so we may
1992 		 * block if there are locks present... this
1993 		 * should be addressed in openat().
1994 		 */
1995 		/* XXX - would it be OK to generate a log record here? */
1996 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
1997 		if (err)
1998 			goto out3;
1999 	}
2000 
2001 	if (mask & (ATTR_ATIME|ATTR_MTIME) ||
2002 	    ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2003 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2004 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2005 	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2006 	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2007 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2008 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2009 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2010 		    skipaclchk, cr);
2011 	}
2012 
2013 	if (mask & (ATTR_UID|ATTR_GID)) {
2014 		int	idmask = (mask & (ATTR_UID|ATTR_GID));
2015 		int	take_owner;
2016 		int	take_group;
2017 
2018 		/*
2019 		 * NOTE: even if a new mode is being set,
2020 		 * we may clear S_ISUID/S_ISGID bits.
2021 		 */
2022 
2023 		if (!(mask & ATTR_MODE))
2024 			vap->va_mode = zp->z_mode;
2025 
2026 		/*
2027 		 * Take ownership or chgrp to group we are a member of
2028 		 */
2029 
2030 		take_owner = (mask & ATTR_UID) && (vap->va_uid == crgetuid(cr));
2031 		take_group = (mask & ATTR_GID) &&
2032 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2033 
2034 		/*
2035 		 * If both ATTR_UID and ATTR_GID are set then take_owner and
2036 		 * take_group must both be set in order to allow taking
2037 		 * ownership.
2038 		 *
2039 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2040 		 *
2041 		 */
2042 
2043 		if (((idmask == (ATTR_UID|ATTR_GID)) &&
2044 		    take_owner && take_group) ||
2045 		    ((idmask == ATTR_UID) && take_owner) ||
2046 		    ((idmask == ATTR_GID) && take_group)) {
2047 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2048 			    skipaclchk, cr) == 0) {
2049 				/*
2050 				 * Remove setuid/setgid for non-privileged users
2051 				 */
2052 				(void) secpolicy_setid_clear(vap, cr);
2053 				trim_mask = (mask & (ATTR_UID|ATTR_GID));
2054 			} else {
2055 				need_policy =  TRUE;
2056 			}
2057 		} else {
2058 			need_policy =  TRUE;
2059 		}
2060 	}
2061 
2062 	mutex_enter(&zp->z_lock);
2063 	oldva.va_mode = zp->z_mode;
2064 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2065 	if (mask & ATTR_XVATTR) {
2066 		/*
2067 		 * Update xvattr mask to include only those attributes
2068 		 * that are actually changing.
2069 		 *
2070 		 * the bits will be restored prior to actually setting
2071 		 * the attributes so the caller thinks they were set.
2072 		 */
2073 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2074 			if (xoap->xoa_appendonly !=
2075 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2076 				need_policy = TRUE;
2077 			} else {
2078 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2079 				XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY);
2080 			}
2081 		}
2082 
2083 		if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) {
2084 			if (xoap->xoa_projinherit !=
2085 			    ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) {
2086 				need_policy = TRUE;
2087 			} else {
2088 				XVA_CLR_REQ(xvap, XAT_PROJINHERIT);
2089 				XVA_SET_REQ(tmpxvattr, XAT_PROJINHERIT);
2090 			}
2091 		}
2092 
2093 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2094 			if (xoap->xoa_nounlink !=
2095 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2096 				need_policy = TRUE;
2097 			} else {
2098 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2099 				XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK);
2100 			}
2101 		}
2102 
2103 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2104 			if (xoap->xoa_immutable !=
2105 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2106 				need_policy = TRUE;
2107 			} else {
2108 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2109 				XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE);
2110 			}
2111 		}
2112 
2113 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2114 			if (xoap->xoa_nodump !=
2115 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2116 				need_policy = TRUE;
2117 			} else {
2118 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2119 				XVA_SET_REQ(tmpxvattr, XAT_NODUMP);
2120 			}
2121 		}
2122 
2123 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2124 			if (xoap->xoa_av_modified !=
2125 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2126 				need_policy = TRUE;
2127 			} else {
2128 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2129 				XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED);
2130 			}
2131 		}
2132 
2133 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2134 			if ((!S_ISREG(ip->i_mode) &&
2135 			    xoap->xoa_av_quarantined) ||
2136 			    xoap->xoa_av_quarantined !=
2137 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2138 				need_policy = TRUE;
2139 			} else {
2140 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2141 				XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED);
2142 			}
2143 		}
2144 
2145 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2146 			mutex_exit(&zp->z_lock);
2147 			err = SET_ERROR(EPERM);
2148 			goto out3;
2149 		}
2150 
2151 		if (need_policy == FALSE &&
2152 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2153 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2154 			need_policy = TRUE;
2155 		}
2156 	}
2157 
2158 	mutex_exit(&zp->z_lock);
2159 
2160 	if (mask & ATTR_MODE) {
2161 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2162 			err = secpolicy_setid_setsticky_clear(ip, vap,
2163 			    &oldva, cr);
2164 			if (err)
2165 				goto out3;
2166 
2167 			trim_mask |= ATTR_MODE;
2168 		} else {
2169 			need_policy = TRUE;
2170 		}
2171 	}
2172 
2173 	if (need_policy) {
2174 		/*
2175 		 * If trim_mask is set then take ownership
2176 		 * has been granted or write_acl is present and user
2177 		 * has the ability to modify mode.  In that case remove
2178 		 * UID|GID and or MODE from mask so that
2179 		 * secpolicy_vnode_setattr() doesn't revoke it.
2180 		 */
2181 
2182 		if (trim_mask) {
2183 			saved_mask = vap->va_mask;
2184 			vap->va_mask &= ~trim_mask;
2185 		}
2186 		err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags,
2187 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2188 		if (err)
2189 			goto out3;
2190 
2191 		if (trim_mask)
2192 			vap->va_mask |= saved_mask;
2193 	}
2194 
2195 	/*
2196 	 * secpolicy_vnode_setattr, or take ownership may have
2197 	 * changed va_mask
2198 	 */
2199 	mask = vap->va_mask;
2200 
2201 	if ((mask & (ATTR_UID | ATTR_GID)) || projid != ZFS_INVALID_PROJID) {
2202 		handle_eadir = B_TRUE;
2203 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2204 		    &xattr_obj, sizeof (xattr_obj));
2205 
2206 		if (err == 0 && xattr_obj) {
2207 			err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp);
2208 			if (err)
2209 				goto out2;
2210 		}
2211 		if (mask & ATTR_UID) {
2212 			new_kuid = zfs_fuid_create(zfsvfs,
2213 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2214 			if (new_kuid != KUID_TO_SUID(ZTOI(zp)->i_uid) &&
2215 			    zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT,
2216 			    new_kuid)) {
2217 				if (attrzp)
2218 					zrele(attrzp);
2219 				err = SET_ERROR(EDQUOT);
2220 				goto out2;
2221 			}
2222 		}
2223 
2224 		if (mask & ATTR_GID) {
2225 			new_kgid = zfs_fuid_create(zfsvfs,
2226 			    (uint64_t)vap->va_gid, cr, ZFS_GROUP, &fuidp);
2227 			if (new_kgid != KGID_TO_SGID(ZTOI(zp)->i_gid) &&
2228 			    zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT,
2229 			    new_kgid)) {
2230 				if (attrzp)
2231 					zrele(attrzp);
2232 				err = SET_ERROR(EDQUOT);
2233 				goto out2;
2234 			}
2235 		}
2236 
2237 		if (projid != ZFS_INVALID_PROJID &&
2238 		    zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) {
2239 			if (attrzp)
2240 				zrele(attrzp);
2241 			err = EDQUOT;
2242 			goto out2;
2243 		}
2244 	}
2245 	tx = dmu_tx_create(os);
2246 
2247 	if (mask & ATTR_MODE) {
2248 		uint64_t pmode = zp->z_mode;
2249 		uint64_t acl_obj;
2250 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2251 
2252 		if (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_RESTRICTED &&
2253 		    !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
2254 			err = EPERM;
2255 			goto out;
2256 		}
2257 
2258 		if ((err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)))
2259 			goto out;
2260 
2261 		mutex_enter(&zp->z_lock);
2262 		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2263 			/*
2264 			 * Are we upgrading ACL from old V0 format
2265 			 * to V1 format?
2266 			 */
2267 			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2268 			    zfs_znode_acl_version(zp) ==
2269 			    ZFS_ACL_VERSION_INITIAL) {
2270 				dmu_tx_hold_free(tx, acl_obj, 0,
2271 				    DMU_OBJECT_END);
2272 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2273 				    0, aclp->z_acl_bytes);
2274 			} else {
2275 				dmu_tx_hold_write(tx, acl_obj, 0,
2276 				    aclp->z_acl_bytes);
2277 			}
2278 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2279 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2280 			    0, aclp->z_acl_bytes);
2281 		}
2282 		mutex_exit(&zp->z_lock);
2283 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2284 	} else {
2285 		if (((mask & ATTR_XVATTR) &&
2286 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
2287 		    (projid != ZFS_INVALID_PROJID &&
2288 		    !(zp->z_pflags & ZFS_PROJID)))
2289 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2290 		else
2291 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2292 	}
2293 
2294 	if (attrzp) {
2295 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2296 	}
2297 
2298 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2299 	if (fuid_dirtied)
2300 		zfs_fuid_txhold(zfsvfs, tx);
2301 
2302 	zfs_sa_upgrade_txholds(tx, zp);
2303 
2304 	err = dmu_tx_assign(tx, TXG_WAIT);
2305 	if (err)
2306 		goto out;
2307 
2308 	count = 0;
2309 	/*
2310 	 * Set each attribute requested.
2311 	 * We group settings according to the locks they need to acquire.
2312 	 *
2313 	 * Note: you cannot set ctime directly, although it will be
2314 	 * updated as a side-effect of calling this function.
2315 	 */
2316 
2317 	if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) {
2318 		/*
2319 		 * For the existed object that is upgraded from old system,
2320 		 * its on-disk layout has no slot for the project ID attribute.
2321 		 * But quota accounting logic needs to access related slots by
2322 		 * offset directly. So we need to adjust old objects' layout
2323 		 * to make the project ID to some unified and fixed offset.
2324 		 */
2325 		if (attrzp)
2326 			err = sa_add_projid(attrzp->z_sa_hdl, tx, projid);
2327 		if (err == 0)
2328 			err = sa_add_projid(zp->z_sa_hdl, tx, projid);
2329 
2330 		if (unlikely(err == EEXIST))
2331 			err = 0;
2332 		else if (err != 0)
2333 			goto out;
2334 		else
2335 			projid = ZFS_INVALID_PROJID;
2336 	}
2337 
2338 	if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2339 		mutex_enter(&zp->z_acl_lock);
2340 	mutex_enter(&zp->z_lock);
2341 
2342 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
2343 	    &zp->z_pflags, sizeof (zp->z_pflags));
2344 
2345 	if (attrzp) {
2346 		if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2347 			mutex_enter(&attrzp->z_acl_lock);
2348 		mutex_enter(&attrzp->z_lock);
2349 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2350 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
2351 		    sizeof (attrzp->z_pflags));
2352 		if (projid != ZFS_INVALID_PROJID) {
2353 			attrzp->z_projid = projid;
2354 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2355 			    SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid,
2356 			    sizeof (attrzp->z_projid));
2357 		}
2358 	}
2359 
2360 	if (mask & (ATTR_UID|ATTR_GID)) {
2361 
2362 		if (mask & ATTR_UID) {
2363 			ZTOI(zp)->i_uid = SUID_TO_KUID(new_kuid);
2364 			new_uid = zfs_uid_read(ZTOI(zp));
2365 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
2366 			    &new_uid, sizeof (new_uid));
2367 			if (attrzp) {
2368 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2369 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
2370 				    sizeof (new_uid));
2371 				ZTOI(attrzp)->i_uid = SUID_TO_KUID(new_uid);
2372 			}
2373 		}
2374 
2375 		if (mask & ATTR_GID) {
2376 			ZTOI(zp)->i_gid = SGID_TO_KGID(new_kgid);
2377 			new_gid = zfs_gid_read(ZTOI(zp));
2378 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
2379 			    NULL, &new_gid, sizeof (new_gid));
2380 			if (attrzp) {
2381 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2382 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
2383 				    sizeof (new_gid));
2384 				ZTOI(attrzp)->i_gid = SGID_TO_KGID(new_kgid);
2385 			}
2386 		}
2387 		if (!(mask & ATTR_MODE)) {
2388 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
2389 			    NULL, &new_mode, sizeof (new_mode));
2390 			new_mode = zp->z_mode;
2391 		}
2392 		err = zfs_acl_chown_setattr(zp);
2393 		ASSERT(err == 0);
2394 		if (attrzp) {
2395 			err = zfs_acl_chown_setattr(attrzp);
2396 			ASSERT(err == 0);
2397 		}
2398 	}
2399 
2400 	if (mask & ATTR_MODE) {
2401 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
2402 		    &new_mode, sizeof (new_mode));
2403 		zp->z_mode = ZTOI(zp)->i_mode = new_mode;
2404 		ASSERT3P(aclp, !=, NULL);
2405 		err = zfs_aclset_common(zp, aclp, cr, tx);
2406 		ASSERT0(err);
2407 		if (zp->z_acl_cached)
2408 			zfs_acl_free(zp->z_acl_cached);
2409 		zp->z_acl_cached = aclp;
2410 		aclp = NULL;
2411 	}
2412 
2413 	if ((mask & ATTR_ATIME) || zp->z_atime_dirty) {
2414 		zp->z_atime_dirty = B_FALSE;
2415 		ZFS_TIME_ENCODE(&ip->i_atime, atime);
2416 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
2417 		    &atime, sizeof (atime));
2418 	}
2419 
2420 	if (mask & (ATTR_MTIME | ATTR_SIZE)) {
2421 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
2422 		ZTOI(zp)->i_mtime = zpl_inode_timestamp_truncate(
2423 		    vap->va_mtime, ZTOI(zp));
2424 
2425 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
2426 		    mtime, sizeof (mtime));
2427 	}
2428 
2429 	if (mask & (ATTR_CTIME | ATTR_SIZE)) {
2430 		ZFS_TIME_ENCODE(&vap->va_ctime, ctime);
2431 		ZTOI(zp)->i_ctime = zpl_inode_timestamp_truncate(vap->va_ctime,
2432 		    ZTOI(zp));
2433 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
2434 		    ctime, sizeof (ctime));
2435 	}
2436 
2437 	if (projid != ZFS_INVALID_PROJID) {
2438 		zp->z_projid = projid;
2439 		SA_ADD_BULK_ATTR(bulk, count,
2440 		    SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid,
2441 		    sizeof (zp->z_projid));
2442 	}
2443 
2444 	if (attrzp && mask) {
2445 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2446 		    SA_ZPL_CTIME(zfsvfs), NULL, &ctime,
2447 		    sizeof (ctime));
2448 	}
2449 
2450 	/*
2451 	 * Do this after setting timestamps to prevent timestamp
2452 	 * update from toggling bit
2453 	 */
2454 
2455 	if (xoap && (mask & ATTR_XVATTR)) {
2456 
2457 		/*
2458 		 * restore trimmed off masks
2459 		 * so that return masks can be set for caller.
2460 		 */
2461 
2462 		if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) {
2463 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
2464 		}
2465 		if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) {
2466 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
2467 		}
2468 		if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) {
2469 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
2470 		}
2471 		if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) {
2472 			XVA_SET_REQ(xvap, XAT_NODUMP);
2473 		}
2474 		if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) {
2475 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
2476 		}
2477 		if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) {
2478 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
2479 		}
2480 		if (XVA_ISSET_REQ(tmpxvattr, XAT_PROJINHERIT)) {
2481 			XVA_SET_REQ(xvap, XAT_PROJINHERIT);
2482 		}
2483 
2484 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2485 			ASSERT(S_ISREG(ip->i_mode));
2486 
2487 		zfs_xvattr_set(zp, xvap, tx);
2488 	}
2489 
2490 	if (fuid_dirtied)
2491 		zfs_fuid_sync(zfsvfs, tx);
2492 
2493 	if (mask != 0)
2494 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2495 
2496 	mutex_exit(&zp->z_lock);
2497 	if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2498 		mutex_exit(&zp->z_acl_lock);
2499 
2500 	if (attrzp) {
2501 		if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE))
2502 			mutex_exit(&attrzp->z_acl_lock);
2503 		mutex_exit(&attrzp->z_lock);
2504 	}
2505 out:
2506 	if (err == 0 && xattr_count > 0) {
2507 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
2508 		    xattr_count, tx);
2509 		ASSERT(err2 == 0);
2510 	}
2511 
2512 	if (aclp)
2513 		zfs_acl_free(aclp);
2514 
2515 	if (fuidp) {
2516 		zfs_fuid_info_free(fuidp);
2517 		fuidp = NULL;
2518 	}
2519 
2520 	if (err) {
2521 		dmu_tx_abort(tx);
2522 		if (attrzp)
2523 			zrele(attrzp);
2524 		if (err == ERESTART)
2525 			goto top;
2526 	} else {
2527 		if (count > 0)
2528 			err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
2529 		dmu_tx_commit(tx);
2530 		if (attrzp) {
2531 			if (err2 == 0 && handle_eadir)
2532 				err2 = zfs_setattr_dir(attrzp);
2533 			zrele(attrzp);
2534 		}
2535 		zfs_inode_update(zp);
2536 	}
2537 
2538 out2:
2539 	if (os->os_sync == ZFS_SYNC_ALWAYS)
2540 		zil_commit(zilog, 0);
2541 
2542 out3:
2543 	kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * bulks);
2544 	kmem_free(bulk, sizeof (sa_bulk_attr_t) * bulks);
2545 	kmem_free(tmpxvattr, sizeof (xvattr_t));
2546 	ZFS_EXIT(zfsvfs);
2547 	return (err);
2548 }
2549 
2550 typedef struct zfs_zlock {
2551 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2552 	znode_t		*zl_znode;	/* znode we held */
2553 	struct zfs_zlock *zl_next;	/* next in list */
2554 } zfs_zlock_t;
2555 
2556 /*
2557  * Drop locks and release vnodes that were held by zfs_rename_lock().
2558  */
2559 static void
2560 zfs_rename_unlock(zfs_zlock_t **zlpp)
2561 {
2562 	zfs_zlock_t *zl;
2563 
2564 	while ((zl = *zlpp) != NULL) {
2565 		if (zl->zl_znode != NULL)
2566 			zfs_zrele_async(zl->zl_znode);
2567 		rw_exit(zl->zl_rwlock);
2568 		*zlpp = zl->zl_next;
2569 		kmem_free(zl, sizeof (*zl));
2570 	}
2571 }
2572 
2573 /*
2574  * Search back through the directory tree, using the ".." entries.
2575  * Lock each directory in the chain to prevent concurrent renames.
2576  * Fail any attempt to move a directory into one of its own descendants.
2577  * XXX - z_parent_lock can overlap with map or grow locks
2578  */
2579 static int
2580 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2581 {
2582 	zfs_zlock_t	*zl;
2583 	znode_t		*zp = tdzp;
2584 	uint64_t	rootid = ZTOZSB(zp)->z_root;
2585 	uint64_t	oidp = zp->z_id;
2586 	krwlock_t	*rwlp = &szp->z_parent_lock;
2587 	krw_t		rw = RW_WRITER;
2588 
2589 	/*
2590 	 * First pass write-locks szp and compares to zp->z_id.
2591 	 * Later passes read-lock zp and compare to zp->z_parent.
2592 	 */
2593 	do {
2594 		if (!rw_tryenter(rwlp, rw)) {
2595 			/*
2596 			 * Another thread is renaming in this path.
2597 			 * Note that if we are a WRITER, we don't have any
2598 			 * parent_locks held yet.
2599 			 */
2600 			if (rw == RW_READER && zp->z_id > szp->z_id) {
2601 				/*
2602 				 * Drop our locks and restart
2603 				 */
2604 				zfs_rename_unlock(&zl);
2605 				*zlpp = NULL;
2606 				zp = tdzp;
2607 				oidp = zp->z_id;
2608 				rwlp = &szp->z_parent_lock;
2609 				rw = RW_WRITER;
2610 				continue;
2611 			} else {
2612 				/*
2613 				 * Wait for other thread to drop its locks
2614 				 */
2615 				rw_enter(rwlp, rw);
2616 			}
2617 		}
2618 
2619 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2620 		zl->zl_rwlock = rwlp;
2621 		zl->zl_znode = NULL;
2622 		zl->zl_next = *zlpp;
2623 		*zlpp = zl;
2624 
2625 		if (oidp == szp->z_id)		/* We're a descendant of szp */
2626 			return (SET_ERROR(EINVAL));
2627 
2628 		if (oidp == rootid)		/* We've hit the top */
2629 			return (0);
2630 
2631 		if (rw == RW_READER) {		/* i.e. not the first pass */
2632 			int error = zfs_zget(ZTOZSB(zp), oidp, &zp);
2633 			if (error)
2634 				return (error);
2635 			zl->zl_znode = zp;
2636 		}
2637 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)),
2638 		    &oidp, sizeof (oidp));
2639 		rwlp = &zp->z_parent_lock;
2640 		rw = RW_READER;
2641 
2642 	} while (zp->z_id != sdzp->z_id);
2643 
2644 	return (0);
2645 }
2646 
2647 /*
2648  * Move an entry from the provided source directory to the target
2649  * directory.  Change the entry name as indicated.
2650  *
2651  *	IN:	sdzp	- Source directory containing the "old entry".
2652  *		snm	- Old entry name.
2653  *		tdzp	- Target directory to contain the "new entry".
2654  *		tnm	- New entry name.
2655  *		cr	- credentials of caller.
2656  *		flags	- case flags
2657  *
2658  *	RETURN:	0 on success, error code on failure.
2659  *
2660  * Timestamps:
2661  *	sdzp,tdzp - ctime|mtime updated
2662  */
2663 /*ARGSUSED*/
2664 int
2665 zfs_rename(znode_t *sdzp, char *snm, znode_t *tdzp, char *tnm,
2666     cred_t *cr, int flags)
2667 {
2668 	znode_t		*szp, *tzp;
2669 	zfsvfs_t	*zfsvfs = ZTOZSB(sdzp);
2670 	zilog_t		*zilog;
2671 	zfs_dirlock_t	*sdl, *tdl;
2672 	dmu_tx_t	*tx;
2673 	zfs_zlock_t	*zl;
2674 	int		cmp, serr, terr;
2675 	int		error = 0;
2676 	int		zflg = 0;
2677 	boolean_t	waited = B_FALSE;
2678 
2679 	if (snm == NULL || tnm == NULL)
2680 		return (SET_ERROR(EINVAL));
2681 
2682 	ZFS_ENTER(zfsvfs);
2683 	ZFS_VERIFY_ZP(sdzp);
2684 	zilog = zfsvfs->z_log;
2685 
2686 	ZFS_VERIFY_ZP(tdzp);
2687 
2688 	/*
2689 	 * We check i_sb because snapshots and the ctldir must have different
2690 	 * super blocks.
2691 	 */
2692 	if (ZTOI(tdzp)->i_sb != ZTOI(sdzp)->i_sb ||
2693 	    zfsctl_is_node(ZTOI(tdzp))) {
2694 		ZFS_EXIT(zfsvfs);
2695 		return (SET_ERROR(EXDEV));
2696 	}
2697 
2698 	if (zfsvfs->z_utf8 && u8_validate(tnm,
2699 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2700 		ZFS_EXIT(zfsvfs);
2701 		return (SET_ERROR(EILSEQ));
2702 	}
2703 
2704 	if (flags & FIGNORECASE)
2705 		zflg |= ZCILOOK;
2706 
2707 top:
2708 	szp = NULL;
2709 	tzp = NULL;
2710 	zl = NULL;
2711 
2712 	/*
2713 	 * This is to prevent the creation of links into attribute space
2714 	 * by renaming a linked file into/outof an attribute directory.
2715 	 * See the comment in zfs_link() for why this is considered bad.
2716 	 */
2717 	if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
2718 		ZFS_EXIT(zfsvfs);
2719 		return (SET_ERROR(EINVAL));
2720 	}
2721 
2722 	/*
2723 	 * Lock source and target directory entries.  To prevent deadlock,
2724 	 * a lock ordering must be defined.  We lock the directory with
2725 	 * the smallest object id first, or if it's a tie, the one with
2726 	 * the lexically first name.
2727 	 */
2728 	if (sdzp->z_id < tdzp->z_id) {
2729 		cmp = -1;
2730 	} else if (sdzp->z_id > tdzp->z_id) {
2731 		cmp = 1;
2732 	} else {
2733 		/*
2734 		 * First compare the two name arguments without
2735 		 * considering any case folding.
2736 		 */
2737 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
2738 
2739 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
2740 		ASSERT(error == 0 || !zfsvfs->z_utf8);
2741 		if (cmp == 0) {
2742 			/*
2743 			 * POSIX: "If the old argument and the new argument
2744 			 * both refer to links to the same existing file,
2745 			 * the rename() function shall return successfully
2746 			 * and perform no other action."
2747 			 */
2748 			ZFS_EXIT(zfsvfs);
2749 			return (0);
2750 		}
2751 		/*
2752 		 * If the file system is case-folding, then we may
2753 		 * have some more checking to do.  A case-folding file
2754 		 * system is either supporting mixed case sensitivity
2755 		 * access or is completely case-insensitive.  Note
2756 		 * that the file system is always case preserving.
2757 		 *
2758 		 * In mixed sensitivity mode case sensitive behavior
2759 		 * is the default.  FIGNORECASE must be used to
2760 		 * explicitly request case insensitive behavior.
2761 		 *
2762 		 * If the source and target names provided differ only
2763 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
2764 		 * we will treat this as a special case in the
2765 		 * case-insensitive mode: as long as the source name
2766 		 * is an exact match, we will allow this to proceed as
2767 		 * a name-change request.
2768 		 */
2769 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
2770 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
2771 		    flags & FIGNORECASE)) &&
2772 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
2773 		    &error) == 0) {
2774 			/*
2775 			 * case preserving rename request, require exact
2776 			 * name matches
2777 			 */
2778 			zflg |= ZCIEXACT;
2779 			zflg &= ~ZCILOOK;
2780 		}
2781 	}
2782 
2783 	/*
2784 	 * If the source and destination directories are the same, we should
2785 	 * grab the z_name_lock of that directory only once.
2786 	 */
2787 	if (sdzp == tdzp) {
2788 		zflg |= ZHAVELOCK;
2789 		rw_enter(&sdzp->z_name_lock, RW_READER);
2790 	}
2791 
2792 	if (cmp < 0) {
2793 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
2794 		    ZEXISTS | zflg, NULL, NULL);
2795 		terr = zfs_dirent_lock(&tdl,
2796 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
2797 	} else {
2798 		terr = zfs_dirent_lock(&tdl,
2799 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
2800 		serr = zfs_dirent_lock(&sdl,
2801 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
2802 		    NULL, NULL);
2803 	}
2804 
2805 	if (serr) {
2806 		/*
2807 		 * Source entry invalid or not there.
2808 		 */
2809 		if (!terr) {
2810 			zfs_dirent_unlock(tdl);
2811 			if (tzp)
2812 				zrele(tzp);
2813 		}
2814 
2815 		if (sdzp == tdzp)
2816 			rw_exit(&sdzp->z_name_lock);
2817 
2818 		if (strcmp(snm, "..") == 0)
2819 			serr = EINVAL;
2820 		ZFS_EXIT(zfsvfs);
2821 		return (serr);
2822 	}
2823 	if (terr) {
2824 		zfs_dirent_unlock(sdl);
2825 		zrele(szp);
2826 
2827 		if (sdzp == tdzp)
2828 			rw_exit(&sdzp->z_name_lock);
2829 
2830 		if (strcmp(tnm, "..") == 0)
2831 			terr = EINVAL;
2832 		ZFS_EXIT(zfsvfs);
2833 		return (terr);
2834 	}
2835 
2836 	/*
2837 	 * If we are using project inheritance, means if the directory has
2838 	 * ZFS_PROJINHERIT set, then its descendant directories will inherit
2839 	 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
2840 	 * such case, we only allow renames into our tree when the project
2841 	 * IDs are the same.
2842 	 */
2843 	if (tdzp->z_pflags & ZFS_PROJINHERIT &&
2844 	    tdzp->z_projid != szp->z_projid) {
2845 		error = SET_ERROR(EXDEV);
2846 		goto out;
2847 	}
2848 
2849 	/*
2850 	 * Must have write access at the source to remove the old entry
2851 	 * and write access at the target to create the new entry.
2852 	 * Note that if target and source are the same, this can be
2853 	 * done in a single check.
2854 	 */
2855 
2856 	if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)))
2857 		goto out;
2858 
2859 	if (S_ISDIR(ZTOI(szp)->i_mode)) {
2860 		/*
2861 		 * Check to make sure rename is valid.
2862 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
2863 		 */
2864 		if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl)))
2865 			goto out;
2866 	}
2867 
2868 	/*
2869 	 * Does target exist?
2870 	 */
2871 	if (tzp) {
2872 		/*
2873 		 * Source and target must be the same type.
2874 		 */
2875 		if (S_ISDIR(ZTOI(szp)->i_mode)) {
2876 			if (!S_ISDIR(ZTOI(tzp)->i_mode)) {
2877 				error = SET_ERROR(ENOTDIR);
2878 				goto out;
2879 			}
2880 		} else {
2881 			if (S_ISDIR(ZTOI(tzp)->i_mode)) {
2882 				error = SET_ERROR(EISDIR);
2883 				goto out;
2884 			}
2885 		}
2886 		/*
2887 		 * POSIX dictates that when the source and target
2888 		 * entries refer to the same file object, rename
2889 		 * must do nothing and exit without error.
2890 		 */
2891 		if (szp->z_id == tzp->z_id) {
2892 			error = 0;
2893 			goto out;
2894 		}
2895 	}
2896 
2897 	tx = dmu_tx_create(zfsvfs->z_os);
2898 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
2899 	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
2900 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
2901 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
2902 	if (sdzp != tdzp) {
2903 		dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
2904 		zfs_sa_upgrade_txholds(tx, tdzp);
2905 	}
2906 	if (tzp) {
2907 		dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
2908 		zfs_sa_upgrade_txholds(tx, tzp);
2909 	}
2910 
2911 	zfs_sa_upgrade_txholds(tx, szp);
2912 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2913 	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
2914 	if (error) {
2915 		if (zl != NULL)
2916 			zfs_rename_unlock(&zl);
2917 		zfs_dirent_unlock(sdl);
2918 		zfs_dirent_unlock(tdl);
2919 
2920 		if (sdzp == tdzp)
2921 			rw_exit(&sdzp->z_name_lock);
2922 
2923 		if (error == ERESTART) {
2924 			waited = B_TRUE;
2925 			dmu_tx_wait(tx);
2926 			dmu_tx_abort(tx);
2927 			zrele(szp);
2928 			if (tzp)
2929 				zrele(tzp);
2930 			goto top;
2931 		}
2932 		dmu_tx_abort(tx);
2933 		zrele(szp);
2934 		if (tzp)
2935 			zrele(tzp);
2936 		ZFS_EXIT(zfsvfs);
2937 		return (error);
2938 	}
2939 
2940 	if (tzp)	/* Attempt to remove the existing target */
2941 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
2942 
2943 	if (error == 0) {
2944 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
2945 		if (error == 0) {
2946 			szp->z_pflags |= ZFS_AV_MODIFIED;
2947 			if (tdzp->z_pflags & ZFS_PROJINHERIT)
2948 				szp->z_pflags |= ZFS_PROJINHERIT;
2949 
2950 			error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
2951 			    (void *)&szp->z_pflags, sizeof (uint64_t), tx);
2952 			ASSERT0(error);
2953 
2954 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
2955 			if (error == 0) {
2956 				zfs_log_rename(zilog, tx, TX_RENAME |
2957 				    (flags & FIGNORECASE ? TX_CI : 0), sdzp,
2958 				    sdl->dl_name, tdzp, tdl->dl_name, szp);
2959 			} else {
2960 				/*
2961 				 * At this point, we have successfully created
2962 				 * the target name, but have failed to remove
2963 				 * the source name.  Since the create was done
2964 				 * with the ZRENAMING flag, there are
2965 				 * complications; for one, the link count is
2966 				 * wrong.  The easiest way to deal with this
2967 				 * is to remove the newly created target, and
2968 				 * return the original error.  This must
2969 				 * succeed; fortunately, it is very unlikely to
2970 				 * fail, since we just created it.
2971 				 */
2972 				VERIFY3U(zfs_link_destroy(tdl, szp, tx,
2973 				    ZRENAMING, NULL), ==, 0);
2974 			}
2975 		} else {
2976 			/*
2977 			 * If we had removed the existing target, subsequent
2978 			 * call to zfs_link_create() to add back the same entry
2979 			 * but, the new dnode (szp) should not fail.
2980 			 */
2981 			ASSERT(tzp == NULL);
2982 		}
2983 	}
2984 
2985 	dmu_tx_commit(tx);
2986 out:
2987 	if (zl != NULL)
2988 		zfs_rename_unlock(&zl);
2989 
2990 	zfs_dirent_unlock(sdl);
2991 	zfs_dirent_unlock(tdl);
2992 
2993 	zfs_inode_update(sdzp);
2994 	if (sdzp == tdzp)
2995 		rw_exit(&sdzp->z_name_lock);
2996 
2997 	if (sdzp != tdzp)
2998 		zfs_inode_update(tdzp);
2999 
3000 	zfs_inode_update(szp);
3001 	zrele(szp);
3002 	if (tzp) {
3003 		zfs_inode_update(tzp);
3004 		zrele(tzp);
3005 	}
3006 
3007 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3008 		zil_commit(zilog, 0);
3009 
3010 	ZFS_EXIT(zfsvfs);
3011 	return (error);
3012 }
3013 
3014 /*
3015  * Insert the indicated symbolic reference entry into the directory.
3016  *
3017  *	IN:	dzp	- Directory to contain new symbolic link.
3018  *		name	- Name of directory entry in dip.
3019  *		vap	- Attributes of new entry.
3020  *		link	- Name for new symlink entry.
3021  *		cr	- credentials of caller.
3022  *		flags	- case flags
3023  *
3024  *	OUT:	zpp	- Znode for new symbolic link.
3025  *
3026  *	RETURN:	0 on success, error code on failure.
3027  *
3028  * Timestamps:
3029  *	dip - ctime|mtime updated
3030  */
3031 /*ARGSUSED*/
3032 int
3033 zfs_symlink(znode_t *dzp, char *name, vattr_t *vap, char *link,
3034     znode_t **zpp, cred_t *cr, int flags)
3035 {
3036 	znode_t		*zp;
3037 	zfs_dirlock_t	*dl;
3038 	dmu_tx_t	*tx;
3039 	zfsvfs_t	*zfsvfs = ZTOZSB(dzp);
3040 	zilog_t		*zilog;
3041 	uint64_t	len = strlen(link);
3042 	int		error;
3043 	int		zflg = ZNEW;
3044 	zfs_acl_ids_t	acl_ids;
3045 	boolean_t	fuid_dirtied;
3046 	uint64_t	txtype = TX_SYMLINK;
3047 	boolean_t	waited = B_FALSE;
3048 
3049 	ASSERT(S_ISLNK(vap->va_mode));
3050 
3051 	if (name == NULL)
3052 		return (SET_ERROR(EINVAL));
3053 
3054 	ZFS_ENTER(zfsvfs);
3055 	ZFS_VERIFY_ZP(dzp);
3056 	zilog = zfsvfs->z_log;
3057 
3058 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3059 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3060 		ZFS_EXIT(zfsvfs);
3061 		return (SET_ERROR(EILSEQ));
3062 	}
3063 	if (flags & FIGNORECASE)
3064 		zflg |= ZCILOOK;
3065 
3066 	if (len > MAXPATHLEN) {
3067 		ZFS_EXIT(zfsvfs);
3068 		return (SET_ERROR(ENAMETOOLONG));
3069 	}
3070 
3071 	if ((error = zfs_acl_ids_create(dzp, 0,
3072 	    vap, cr, NULL, &acl_ids)) != 0) {
3073 		ZFS_EXIT(zfsvfs);
3074 		return (error);
3075 	}
3076 top:
3077 	*zpp = NULL;
3078 
3079 	/*
3080 	 * Attempt to lock directory; fail if entry already exists.
3081 	 */
3082 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3083 	if (error) {
3084 		zfs_acl_ids_free(&acl_ids);
3085 		ZFS_EXIT(zfsvfs);
3086 		return (error);
3087 	}
3088 
3089 	if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3090 		zfs_acl_ids_free(&acl_ids);
3091 		zfs_dirent_unlock(dl);
3092 		ZFS_EXIT(zfsvfs);
3093 		return (error);
3094 	}
3095 
3096 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, ZFS_DEFAULT_PROJID)) {
3097 		zfs_acl_ids_free(&acl_ids);
3098 		zfs_dirent_unlock(dl);
3099 		ZFS_EXIT(zfsvfs);
3100 		return (SET_ERROR(EDQUOT));
3101 	}
3102 	tx = dmu_tx_create(zfsvfs->z_os);
3103 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3104 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3105 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3106 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3107 	    ZFS_SA_BASE_ATTR_SIZE + len);
3108 	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3109 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3110 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3111 		    acl_ids.z_aclp->z_acl_bytes);
3112 	}
3113 	if (fuid_dirtied)
3114 		zfs_fuid_txhold(zfsvfs, tx);
3115 	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
3116 	if (error) {
3117 		zfs_dirent_unlock(dl);
3118 		if (error == ERESTART) {
3119 			waited = B_TRUE;
3120 			dmu_tx_wait(tx);
3121 			dmu_tx_abort(tx);
3122 			goto top;
3123 		}
3124 		zfs_acl_ids_free(&acl_ids);
3125 		dmu_tx_abort(tx);
3126 		ZFS_EXIT(zfsvfs);
3127 		return (error);
3128 	}
3129 
3130 	/*
3131 	 * Create a new object for the symlink.
3132 	 * for version 4 ZPL datsets the symlink will be an SA attribute
3133 	 */
3134 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3135 
3136 	if (fuid_dirtied)
3137 		zfs_fuid_sync(zfsvfs, tx);
3138 
3139 	mutex_enter(&zp->z_lock);
3140 	if (zp->z_is_sa)
3141 		error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3142 		    link, len, tx);
3143 	else
3144 		zfs_sa_symlink(zp, link, len, tx);
3145 	mutex_exit(&zp->z_lock);
3146 
3147 	zp->z_size = len;
3148 	(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3149 	    &zp->z_size, sizeof (zp->z_size), tx);
3150 	/*
3151 	 * Insert the new object into the directory.
3152 	 */
3153 	error = zfs_link_create(dl, zp, tx, ZNEW);
3154 	if (error != 0) {
3155 		zfs_znode_delete(zp, tx);
3156 		remove_inode_hash(ZTOI(zp));
3157 	} else {
3158 		if (flags & FIGNORECASE)
3159 			txtype |= TX_CI;
3160 		zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3161 
3162 		zfs_inode_update(dzp);
3163 		zfs_inode_update(zp);
3164 	}
3165 
3166 	zfs_acl_ids_free(&acl_ids);
3167 
3168 	dmu_tx_commit(tx);
3169 
3170 	zfs_dirent_unlock(dl);
3171 
3172 	if (error == 0) {
3173 		*zpp = zp;
3174 
3175 		if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3176 			zil_commit(zilog, 0);
3177 	} else {
3178 		zrele(zp);
3179 	}
3180 
3181 	ZFS_EXIT(zfsvfs);
3182 	return (error);
3183 }
3184 
3185 /*
3186  * Return, in the buffer contained in the provided uio structure,
3187  * the symbolic path referred to by ip.
3188  *
3189  *	IN:	ip	- inode of symbolic link
3190  *		uio	- structure to contain the link path.
3191  *		cr	- credentials of caller.
3192  *
3193  *	RETURN:	0 if success
3194  *		error code if failure
3195  *
3196  * Timestamps:
3197  *	ip - atime updated
3198  */
3199 /* ARGSUSED */
3200 int
3201 zfs_readlink(struct inode *ip, uio_t *uio, cred_t *cr)
3202 {
3203 	znode_t		*zp = ITOZ(ip);
3204 	zfsvfs_t	*zfsvfs = ITOZSB(ip);
3205 	int		error;
3206 
3207 	ZFS_ENTER(zfsvfs);
3208 	ZFS_VERIFY_ZP(zp);
3209 
3210 	mutex_enter(&zp->z_lock);
3211 	if (zp->z_is_sa)
3212 		error = sa_lookup_uio(zp->z_sa_hdl,
3213 		    SA_ZPL_SYMLINK(zfsvfs), uio);
3214 	else
3215 		error = zfs_sa_readlink(zp, uio);
3216 	mutex_exit(&zp->z_lock);
3217 
3218 	ZFS_EXIT(zfsvfs);
3219 	return (error);
3220 }
3221 
3222 /*
3223  * Insert a new entry into directory tdzp referencing szp.
3224  *
3225  *	IN:	tdzp	- Directory to contain new entry.
3226  *		szp	- znode of new entry.
3227  *		name	- name of new entry.
3228  *		cr	- credentials of caller.
3229  *		flags	- case flags.
3230  *
3231  *	RETURN:	0 if success
3232  *		error code if failure
3233  *
3234  * Timestamps:
3235  *	tdzp - ctime|mtime updated
3236  *	 szp - ctime updated
3237  */
3238 /* ARGSUSED */
3239 int
3240 zfs_link(znode_t *tdzp, znode_t *szp, char *name, cred_t *cr,
3241     int flags)
3242 {
3243 	struct inode *sip = ZTOI(szp);
3244 	znode_t		*tzp;
3245 	zfsvfs_t	*zfsvfs = ZTOZSB(tdzp);
3246 	zilog_t		*zilog;
3247 	zfs_dirlock_t	*dl;
3248 	dmu_tx_t	*tx;
3249 	int		error;
3250 	int		zf = ZNEW;
3251 	uint64_t	parent;
3252 	uid_t		owner;
3253 	boolean_t	waited = B_FALSE;
3254 	boolean_t	is_tmpfile = 0;
3255 	uint64_t	txg;
3256 #ifdef HAVE_TMPFILE
3257 	is_tmpfile = (sip->i_nlink == 0 && (sip->i_state & I_LINKABLE));
3258 #endif
3259 	ASSERT(S_ISDIR(ZTOI(tdzp)->i_mode));
3260 
3261 	if (name == NULL)
3262 		return (SET_ERROR(EINVAL));
3263 
3264 	ZFS_ENTER(zfsvfs);
3265 	ZFS_VERIFY_ZP(tdzp);
3266 	zilog = zfsvfs->z_log;
3267 
3268 	/*
3269 	 * POSIX dictates that we return EPERM here.
3270 	 * Better choices include ENOTSUP or EISDIR.
3271 	 */
3272 	if (S_ISDIR(sip->i_mode)) {
3273 		ZFS_EXIT(zfsvfs);
3274 		return (SET_ERROR(EPERM));
3275 	}
3276 
3277 	ZFS_VERIFY_ZP(szp);
3278 
3279 	/*
3280 	 * If we are using project inheritance, means if the directory has
3281 	 * ZFS_PROJINHERIT set, then its descendant directories will inherit
3282 	 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
3283 	 * such case, we only allow hard link creation in our tree when the
3284 	 * project IDs are the same.
3285 	 */
3286 	if (tdzp->z_pflags & ZFS_PROJINHERIT &&
3287 	    tdzp->z_projid != szp->z_projid) {
3288 		ZFS_EXIT(zfsvfs);
3289 		return (SET_ERROR(EXDEV));
3290 	}
3291 
3292 	/*
3293 	 * We check i_sb because snapshots and the ctldir must have different
3294 	 * super blocks.
3295 	 */
3296 	if (sip->i_sb != ZTOI(tdzp)->i_sb || zfsctl_is_node(sip)) {
3297 		ZFS_EXIT(zfsvfs);
3298 		return (SET_ERROR(EXDEV));
3299 	}
3300 
3301 	/* Prevent links to .zfs/shares files */
3302 
3303 	if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3304 	    &parent, sizeof (uint64_t))) != 0) {
3305 		ZFS_EXIT(zfsvfs);
3306 		return (error);
3307 	}
3308 	if (parent == zfsvfs->z_shares_dir) {
3309 		ZFS_EXIT(zfsvfs);
3310 		return (SET_ERROR(EPERM));
3311 	}
3312 
3313 	if (zfsvfs->z_utf8 && u8_validate(name,
3314 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3315 		ZFS_EXIT(zfsvfs);
3316 		return (SET_ERROR(EILSEQ));
3317 	}
3318 	if (flags & FIGNORECASE)
3319 		zf |= ZCILOOK;
3320 
3321 	/*
3322 	 * We do not support links between attributes and non-attributes
3323 	 * because of the potential security risk of creating links
3324 	 * into "normal" file space in order to circumvent restrictions
3325 	 * imposed in attribute space.
3326 	 */
3327 	if ((szp->z_pflags & ZFS_XATTR) != (tdzp->z_pflags & ZFS_XATTR)) {
3328 		ZFS_EXIT(zfsvfs);
3329 		return (SET_ERROR(EINVAL));
3330 	}
3331 
3332 	owner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(sip->i_uid),
3333 	    cr, ZFS_OWNER);
3334 	if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3335 		ZFS_EXIT(zfsvfs);
3336 		return (SET_ERROR(EPERM));
3337 	}
3338 
3339 	if ((error = zfs_zaccess(tdzp, ACE_ADD_FILE, 0, B_FALSE, cr))) {
3340 		ZFS_EXIT(zfsvfs);
3341 		return (error);
3342 	}
3343 
3344 top:
3345 	/*
3346 	 * Attempt to lock directory; fail if entry already exists.
3347 	 */
3348 	error = zfs_dirent_lock(&dl, tdzp, name, &tzp, zf, NULL, NULL);
3349 	if (error) {
3350 		ZFS_EXIT(zfsvfs);
3351 		return (error);
3352 	}
3353 
3354 	tx = dmu_tx_create(zfsvfs->z_os);
3355 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3356 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, name);
3357 	if (is_tmpfile)
3358 		dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3359 
3360 	zfs_sa_upgrade_txholds(tx, szp);
3361 	zfs_sa_upgrade_txholds(tx, tdzp);
3362 	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
3363 	if (error) {
3364 		zfs_dirent_unlock(dl);
3365 		if (error == ERESTART) {
3366 			waited = B_TRUE;
3367 			dmu_tx_wait(tx);
3368 			dmu_tx_abort(tx);
3369 			goto top;
3370 		}
3371 		dmu_tx_abort(tx);
3372 		ZFS_EXIT(zfsvfs);
3373 		return (error);
3374 	}
3375 	/* unmark z_unlinked so zfs_link_create will not reject */
3376 	if (is_tmpfile)
3377 		szp->z_unlinked = B_FALSE;
3378 	error = zfs_link_create(dl, szp, tx, 0);
3379 
3380 	if (error == 0) {
3381 		uint64_t txtype = TX_LINK;
3382 		/*
3383 		 * tmpfile is created to be in z_unlinkedobj, so remove it.
3384 		 * Also, we don't log in ZIL, because all previous file
3385 		 * operation on the tmpfile are ignored by ZIL. Instead we
3386 		 * always wait for txg to sync to make sure all previous
3387 		 * operation are sync safe.
3388 		 */
3389 		if (is_tmpfile) {
3390 			VERIFY(zap_remove_int(zfsvfs->z_os,
3391 			    zfsvfs->z_unlinkedobj, szp->z_id, tx) == 0);
3392 		} else {
3393 			if (flags & FIGNORECASE)
3394 				txtype |= TX_CI;
3395 			zfs_log_link(zilog, tx, txtype, tdzp, szp, name);
3396 		}
3397 	} else if (is_tmpfile) {
3398 		/* restore z_unlinked since when linking failed */
3399 		szp->z_unlinked = B_TRUE;
3400 	}
3401 	txg = dmu_tx_get_txg(tx);
3402 	dmu_tx_commit(tx);
3403 
3404 	zfs_dirent_unlock(dl);
3405 
3406 	if (!is_tmpfile && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3407 		zil_commit(zilog, 0);
3408 
3409 	if (is_tmpfile && zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED)
3410 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), txg);
3411 
3412 	zfs_inode_update(tdzp);
3413 	zfs_inode_update(szp);
3414 	ZFS_EXIT(zfsvfs);
3415 	return (error);
3416 }
3417 
3418 static void
3419 zfs_putpage_commit_cb(void *arg)
3420 {
3421 	struct page *pp = arg;
3422 
3423 	ClearPageError(pp);
3424 	end_page_writeback(pp);
3425 }
3426 
3427 /*
3428  * Push a page out to disk, once the page is on stable storage the
3429  * registered commit callback will be run as notification of completion.
3430  *
3431  *	IN:	ip	- page mapped for inode.
3432  *		pp	- page to push (page is locked)
3433  *		wbc	- writeback control data
3434  *
3435  *	RETURN:	0 if success
3436  *		error code if failure
3437  *
3438  * Timestamps:
3439  *	ip - ctime|mtime updated
3440  */
3441 /* ARGSUSED */
3442 int
3443 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
3444 {
3445 	znode_t		*zp = ITOZ(ip);
3446 	zfsvfs_t	*zfsvfs = ITOZSB(ip);
3447 	loff_t		offset;
3448 	loff_t		pgoff;
3449 	unsigned int	pglen;
3450 	dmu_tx_t	*tx;
3451 	caddr_t		va;
3452 	int		err = 0;
3453 	uint64_t	mtime[2], ctime[2];
3454 	sa_bulk_attr_t	bulk[3];
3455 	int		cnt = 0;
3456 	struct address_space *mapping;
3457 
3458 	ZFS_ENTER(zfsvfs);
3459 	ZFS_VERIFY_ZP(zp);
3460 
3461 	ASSERT(PageLocked(pp));
3462 
3463 	pgoff = page_offset(pp);	/* Page byte-offset in file */
3464 	offset = i_size_read(ip);	/* File length in bytes */
3465 	pglen = MIN(PAGE_SIZE,		/* Page length in bytes */
3466 	    P2ROUNDUP(offset, PAGE_SIZE)-pgoff);
3467 
3468 	/* Page is beyond end of file */
3469 	if (pgoff >= offset) {
3470 		unlock_page(pp);
3471 		ZFS_EXIT(zfsvfs);
3472 		return (0);
3473 	}
3474 
3475 	/* Truncate page length to end of file */
3476 	if (pgoff + pglen > offset)
3477 		pglen = offset - pgoff;
3478 
3479 #if 0
3480 	/*
3481 	 * FIXME: Allow mmap writes past its quota.  The correct fix
3482 	 * is to register a page_mkwrite() handler to count the page
3483 	 * against its quota when it is about to be dirtied.
3484 	 */
3485 	if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT,
3486 	    KUID_TO_SUID(ip->i_uid)) ||
3487 	    zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT,
3488 	    KGID_TO_SGID(ip->i_gid)) ||
3489 	    (zp->z_projid != ZFS_DEFAULT_PROJID &&
3490 	    zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT,
3491 	    zp->z_projid))) {
3492 		err = EDQUOT;
3493 	}
3494 #endif
3495 
3496 	/*
3497 	 * The ordering here is critical and must adhere to the following
3498 	 * rules in order to avoid deadlocking in either zfs_read() or
3499 	 * zfs_free_range() due to a lock inversion.
3500 	 *
3501 	 * 1) The page must be unlocked prior to acquiring the range lock.
3502 	 *    This is critical because zfs_read() calls find_lock_page()
3503 	 *    which may block on the page lock while holding the range lock.
3504 	 *
3505 	 * 2) Before setting or clearing write back on a page the range lock
3506 	 *    must be held in order to prevent a lock inversion with the
3507 	 *    zfs_free_range() function.
3508 	 *
3509 	 * This presents a problem because upon entering this function the
3510 	 * page lock is already held.  To safely acquire the range lock the
3511 	 * page lock must be dropped.  This creates a window where another
3512 	 * process could truncate, invalidate, dirty, or write out the page.
3513 	 *
3514 	 * Therefore, after successfully reacquiring the range and page locks
3515 	 * the current page state is checked.  In the common case everything
3516 	 * will be as is expected and it can be written out.  However, if
3517 	 * the page state has changed it must be handled accordingly.
3518 	 */
3519 	mapping = pp->mapping;
3520 	redirty_page_for_writepage(wbc, pp);
3521 	unlock_page(pp);
3522 
3523 	zfs_locked_range_t *lr = zfs_rangelock_enter(&zp->z_rangelock,
3524 	    pgoff, pglen, RL_WRITER);
3525 	lock_page(pp);
3526 
3527 	/* Page mapping changed or it was no longer dirty, we're done */
3528 	if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) {
3529 		unlock_page(pp);
3530 		zfs_rangelock_exit(lr);
3531 		ZFS_EXIT(zfsvfs);
3532 		return (0);
3533 	}
3534 
3535 	/* Another process started write block if required */
3536 	if (PageWriteback(pp)) {
3537 		unlock_page(pp);
3538 		zfs_rangelock_exit(lr);
3539 
3540 		if (wbc->sync_mode != WB_SYNC_NONE) {
3541 			if (PageWriteback(pp))
3542 				wait_on_page_bit(pp, PG_writeback);
3543 		}
3544 
3545 		ZFS_EXIT(zfsvfs);
3546 		return (0);
3547 	}
3548 
3549 	/* Clear the dirty flag the required locks are held */
3550 	if (!clear_page_dirty_for_io(pp)) {
3551 		unlock_page(pp);
3552 		zfs_rangelock_exit(lr);
3553 		ZFS_EXIT(zfsvfs);
3554 		return (0);
3555 	}
3556 
3557 	/*
3558 	 * Counterpart for redirty_page_for_writepage() above.  This page
3559 	 * was in fact not skipped and should not be counted as if it were.
3560 	 */
3561 	wbc->pages_skipped--;
3562 	set_page_writeback(pp);
3563 	unlock_page(pp);
3564 
3565 	tx = dmu_tx_create(zfsvfs->z_os);
3566 	dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen);
3567 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3568 	zfs_sa_upgrade_txholds(tx, zp);
3569 
3570 	err = dmu_tx_assign(tx, TXG_NOWAIT);
3571 	if (err != 0) {
3572 		if (err == ERESTART)
3573 			dmu_tx_wait(tx);
3574 
3575 		dmu_tx_abort(tx);
3576 		__set_page_dirty_nobuffers(pp);
3577 		ClearPageError(pp);
3578 		end_page_writeback(pp);
3579 		zfs_rangelock_exit(lr);
3580 		ZFS_EXIT(zfsvfs);
3581 		return (err);
3582 	}
3583 
3584 	va = kmap(pp);
3585 	ASSERT3U(pglen, <=, PAGE_SIZE);
3586 	dmu_write(zfsvfs->z_os, zp->z_id, pgoff, pglen, va, tx);
3587 	kunmap(pp);
3588 
3589 	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
3590 	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
3591 	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zfsvfs), NULL,
3592 	    &zp->z_pflags, 8);
3593 
3594 	/* Preserve the mtime and ctime provided by the inode */
3595 	ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
3596 	ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
3597 	zp->z_atime_dirty = B_FALSE;
3598 	zp->z_seq++;
3599 
3600 	err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
3601 
3602 	zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0,
3603 	    zfs_putpage_commit_cb, pp);
3604 	dmu_tx_commit(tx);
3605 
3606 	zfs_rangelock_exit(lr);
3607 
3608 	if (wbc->sync_mode != WB_SYNC_NONE) {
3609 		/*
3610 		 * Note that this is rarely called under writepages(), because
3611 		 * writepages() normally handles the entire commit for
3612 		 * performance reasons.
3613 		 */
3614 		zil_commit(zfsvfs->z_log, zp->z_id);
3615 	}
3616 
3617 	ZFS_EXIT(zfsvfs);
3618 	return (err);
3619 }
3620 
3621 /*
3622  * Update the system attributes when the inode has been dirtied.  For the
3623  * moment we only update the mode, atime, mtime, and ctime.
3624  */
3625 int
3626 zfs_dirty_inode(struct inode *ip, int flags)
3627 {
3628 	znode_t		*zp = ITOZ(ip);
3629 	zfsvfs_t	*zfsvfs = ITOZSB(ip);
3630 	dmu_tx_t	*tx;
3631 	uint64_t	mode, atime[2], mtime[2], ctime[2];
3632 	sa_bulk_attr_t	bulk[4];
3633 	int		error = 0;
3634 	int		cnt = 0;
3635 
3636 	if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os))
3637 		return (0);
3638 
3639 	ZFS_ENTER(zfsvfs);
3640 	ZFS_VERIFY_ZP(zp);
3641 
3642 #ifdef I_DIRTY_TIME
3643 	/*
3644 	 * This is the lazytime semantic introduced in Linux 4.0
3645 	 * This flag will only be called from update_time when lazytime is set.
3646 	 * (Note, I_DIRTY_SYNC will also set if not lazytime)
3647 	 * Fortunately mtime and ctime are managed within ZFS itself, so we
3648 	 * only need to dirty atime.
3649 	 */
3650 	if (flags == I_DIRTY_TIME) {
3651 		zp->z_atime_dirty = B_TRUE;
3652 		goto out;
3653 	}
3654 #endif
3655 
3656 	tx = dmu_tx_create(zfsvfs->z_os);
3657 
3658 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3659 	zfs_sa_upgrade_txholds(tx, zp);
3660 
3661 	error = dmu_tx_assign(tx, TXG_WAIT);
3662 	if (error) {
3663 		dmu_tx_abort(tx);
3664 		goto out;
3665 	}
3666 
3667 	mutex_enter(&zp->z_lock);
3668 	zp->z_atime_dirty = B_FALSE;
3669 
3670 	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8);
3671 	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16);
3672 	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
3673 	SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
3674 
3675 	/* Preserve the mode, mtime and ctime provided by the inode */
3676 	ZFS_TIME_ENCODE(&ip->i_atime, atime);
3677 	ZFS_TIME_ENCODE(&ip->i_mtime, mtime);
3678 	ZFS_TIME_ENCODE(&ip->i_ctime, ctime);
3679 	mode = ip->i_mode;
3680 
3681 	zp->z_mode = mode;
3682 
3683 	error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx);
3684 	mutex_exit(&zp->z_lock);
3685 
3686 	dmu_tx_commit(tx);
3687 out:
3688 	ZFS_EXIT(zfsvfs);
3689 	return (error);
3690 }
3691 
3692 /*ARGSUSED*/
3693 void
3694 zfs_inactive(struct inode *ip)
3695 {
3696 	znode_t	*zp = ITOZ(ip);
3697 	zfsvfs_t *zfsvfs = ITOZSB(ip);
3698 	uint64_t atime[2];
3699 	int error;
3700 	int need_unlock = 0;
3701 
3702 	/* Only read lock if we haven't already write locked, e.g. rollback */
3703 	if (!RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)) {
3704 		need_unlock = 1;
3705 		rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
3706 	}
3707 	if (zp->z_sa_hdl == NULL) {
3708 		if (need_unlock)
3709 			rw_exit(&zfsvfs->z_teardown_inactive_lock);
3710 		return;
3711 	}
3712 
3713 	if (zp->z_atime_dirty && zp->z_unlinked == B_FALSE) {
3714 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3715 
3716 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3717 		zfs_sa_upgrade_txholds(tx, zp);
3718 		error = dmu_tx_assign(tx, TXG_WAIT);
3719 		if (error) {
3720 			dmu_tx_abort(tx);
3721 		} else {
3722 			ZFS_TIME_ENCODE(&ip->i_atime, atime);
3723 			mutex_enter(&zp->z_lock);
3724 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
3725 			    (void *)&atime, sizeof (atime), tx);
3726 			zp->z_atime_dirty = B_FALSE;
3727 			mutex_exit(&zp->z_lock);
3728 			dmu_tx_commit(tx);
3729 		}
3730 	}
3731 
3732 	zfs_zinactive(zp);
3733 	if (need_unlock)
3734 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
3735 }
3736 
3737 /*
3738  * Fill pages with data from the disk.
3739  */
3740 static int
3741 zfs_fillpage(struct inode *ip, struct page *pl[], int nr_pages)
3742 {
3743 	znode_t *zp = ITOZ(ip);
3744 	zfsvfs_t *zfsvfs = ITOZSB(ip);
3745 	objset_t *os;
3746 	struct page *cur_pp;
3747 	u_offset_t io_off, total;
3748 	size_t io_len;
3749 	loff_t i_size;
3750 	unsigned page_idx;
3751 	int err;
3752 
3753 	os = zfsvfs->z_os;
3754 	io_len = nr_pages << PAGE_SHIFT;
3755 	i_size = i_size_read(ip);
3756 	io_off = page_offset(pl[0]);
3757 
3758 	if (io_off + io_len > i_size)
3759 		io_len = i_size - io_off;
3760 
3761 	/*
3762 	 * Iterate over list of pages and read each page individually.
3763 	 */
3764 	page_idx = 0;
3765 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3766 		caddr_t va;
3767 
3768 		cur_pp = pl[page_idx++];
3769 		va = kmap(cur_pp);
3770 		err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
3771 		    DMU_READ_PREFETCH);
3772 		kunmap(cur_pp);
3773 		if (err) {
3774 			/* convert checksum errors into IO errors */
3775 			if (err == ECKSUM)
3776 				err = SET_ERROR(EIO);
3777 			return (err);
3778 		}
3779 	}
3780 
3781 	return (0);
3782 }
3783 
3784 /*
3785  * Uses zfs_fillpage to read data from the file and fill the pages.
3786  *
3787  *	IN:	ip	 - inode of file to get data from.
3788  *		pl	 - list of pages to read
3789  *		nr_pages - number of pages to read
3790  *
3791  *	RETURN:	0 on success, error code on failure.
3792  *
3793  * Timestamps:
3794  *	vp - atime updated
3795  */
3796 /* ARGSUSED */
3797 int
3798 zfs_getpage(struct inode *ip, struct page *pl[], int nr_pages)
3799 {
3800 	znode_t	 *zp  = ITOZ(ip);
3801 	zfsvfs_t *zfsvfs = ITOZSB(ip);
3802 	int	 err;
3803 
3804 	if (pl == NULL)
3805 		return (0);
3806 
3807 	ZFS_ENTER(zfsvfs);
3808 	ZFS_VERIFY_ZP(zp);
3809 
3810 	err = zfs_fillpage(ip, pl, nr_pages);
3811 
3812 	ZFS_EXIT(zfsvfs);
3813 	return (err);
3814 }
3815 
3816 /*
3817  * Check ZFS specific permissions to memory map a section of a file.
3818  *
3819  *	IN:	ip	- inode of the file to mmap
3820  *		off	- file offset
3821  *		addrp	- start address in memory region
3822  *		len	- length of memory region
3823  *		vm_flags- address flags
3824  *
3825  *	RETURN:	0 if success
3826  *		error code if failure
3827  */
3828 /*ARGSUSED*/
3829 int
3830 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len,
3831     unsigned long vm_flags)
3832 {
3833 	znode_t  *zp = ITOZ(ip);
3834 	zfsvfs_t *zfsvfs = ITOZSB(ip);
3835 
3836 	ZFS_ENTER(zfsvfs);
3837 	ZFS_VERIFY_ZP(zp);
3838 
3839 	if ((vm_flags & VM_WRITE) && (zp->z_pflags &
3840 	    (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
3841 		ZFS_EXIT(zfsvfs);
3842 		return (SET_ERROR(EPERM));
3843 	}
3844 
3845 	if ((vm_flags & (VM_READ | VM_EXEC)) &&
3846 	    (zp->z_pflags & ZFS_AV_QUARANTINED)) {
3847 		ZFS_EXIT(zfsvfs);
3848 		return (SET_ERROR(EACCES));
3849 	}
3850 
3851 	if (off < 0 || len > MAXOFFSET_T - off) {
3852 		ZFS_EXIT(zfsvfs);
3853 		return (SET_ERROR(ENXIO));
3854 	}
3855 
3856 	ZFS_EXIT(zfsvfs);
3857 	return (0);
3858 }
3859 
3860 /*
3861  * Free or allocate space in a file.  Currently, this function only
3862  * supports the `F_FREESP' command.  However, this command is somewhat
3863  * misnamed, as its functionality includes the ability to allocate as
3864  * well as free space.
3865  *
3866  *	IN:	zp	- znode of file to free data in.
3867  *		cmd	- action to take (only F_FREESP supported).
3868  *		bfp	- section of file to free/alloc.
3869  *		flag	- current file open mode flags.
3870  *		offset	- current file offset.
3871  *		cr	- credentials of caller.
3872  *
3873  *	RETURN:	0 on success, error code on failure.
3874  *
3875  * Timestamps:
3876  *	zp - ctime|mtime updated
3877  */
3878 /* ARGSUSED */
3879 int
3880 zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag,
3881     offset_t offset, cred_t *cr)
3882 {
3883 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
3884 	uint64_t	off, len;
3885 	int		error;
3886 
3887 	ZFS_ENTER(zfsvfs);
3888 	ZFS_VERIFY_ZP(zp);
3889 
3890 	if (cmd != F_FREESP) {
3891 		ZFS_EXIT(zfsvfs);
3892 		return (SET_ERROR(EINVAL));
3893 	}
3894 
3895 	/*
3896 	 * Callers might not be able to detect properly that we are read-only,
3897 	 * so check it explicitly here.
3898 	 */
3899 	if (zfs_is_readonly(zfsvfs)) {
3900 		ZFS_EXIT(zfsvfs);
3901 		return (SET_ERROR(EROFS));
3902 	}
3903 
3904 	if (bfp->l_len < 0) {
3905 		ZFS_EXIT(zfsvfs);
3906 		return (SET_ERROR(EINVAL));
3907 	}
3908 
3909 	/*
3910 	 * Permissions aren't checked on Solaris because on this OS
3911 	 * zfs_space() can only be called with an opened file handle.
3912 	 * On Linux we can get here through truncate_range() which
3913 	 * operates directly on inodes, so we need to check access rights.
3914 	 */
3915 	if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr))) {
3916 		ZFS_EXIT(zfsvfs);
3917 		return (error);
3918 	}
3919 
3920 	off = bfp->l_start;
3921 	len = bfp->l_len; /* 0 means from off to end of file */
3922 
3923 	error = zfs_freesp(zp, off, len, flag, TRUE);
3924 
3925 	ZFS_EXIT(zfsvfs);
3926 	return (error);
3927 }
3928 
3929 /*ARGSUSED*/
3930 int
3931 zfs_fid(struct inode *ip, fid_t *fidp)
3932 {
3933 	znode_t		*zp = ITOZ(ip);
3934 	zfsvfs_t	*zfsvfs = ITOZSB(ip);
3935 	uint32_t	gen;
3936 	uint64_t	gen64;
3937 	uint64_t	object = zp->z_id;
3938 	zfid_short_t	*zfid;
3939 	int		size, i, error;
3940 
3941 	ZFS_ENTER(zfsvfs);
3942 	ZFS_VERIFY_ZP(zp);
3943 
3944 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
3945 	    &gen64, sizeof (uint64_t))) != 0) {
3946 		ZFS_EXIT(zfsvfs);
3947 		return (error);
3948 	}
3949 
3950 	gen = (uint32_t)gen64;
3951 
3952 	size = SHORT_FID_LEN;
3953 
3954 	zfid = (zfid_short_t *)fidp;
3955 
3956 	zfid->zf_len = size;
3957 
3958 	for (i = 0; i < sizeof (zfid->zf_object); i++)
3959 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
3960 
3961 	/* Must have a non-zero generation number to distinguish from .zfs */
3962 	if (gen == 0)
3963 		gen = 1;
3964 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
3965 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
3966 
3967 	ZFS_EXIT(zfsvfs);
3968 	return (0);
3969 }
3970 
3971 #if defined(_KERNEL)
3972 EXPORT_SYMBOL(zfs_open);
3973 EXPORT_SYMBOL(zfs_close);
3974 EXPORT_SYMBOL(zfs_lookup);
3975 EXPORT_SYMBOL(zfs_create);
3976 EXPORT_SYMBOL(zfs_tmpfile);
3977 EXPORT_SYMBOL(zfs_remove);
3978 EXPORT_SYMBOL(zfs_mkdir);
3979 EXPORT_SYMBOL(zfs_rmdir);
3980 EXPORT_SYMBOL(zfs_readdir);
3981 EXPORT_SYMBOL(zfs_getattr_fast);
3982 EXPORT_SYMBOL(zfs_setattr);
3983 EXPORT_SYMBOL(zfs_rename);
3984 EXPORT_SYMBOL(zfs_symlink);
3985 EXPORT_SYMBOL(zfs_readlink);
3986 EXPORT_SYMBOL(zfs_link);
3987 EXPORT_SYMBOL(zfs_inactive);
3988 EXPORT_SYMBOL(zfs_space);
3989 EXPORT_SYMBOL(zfs_fid);
3990 EXPORT_SYMBOL(zfs_getpage);
3991 EXPORT_SYMBOL(zfs_putpage);
3992 EXPORT_SYMBOL(zfs_dirty_inode);
3993 EXPORT_SYMBOL(zfs_map);
3994 
3995 /* BEGIN CSTYLED */
3996 module_param(zfs_delete_blocks, ulong, 0644);
3997 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async");
3998 /* END CSTYLED */
3999 
4000 #endif
4001