xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 79315247)
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, 2017 by Delphix. All rights reserved.
25  * Copyright (c) 2014 Integros [integros.com]
26  * Copyright 2015 Joyent, Inc.
27  * Copyright 2017 Nexenta Systems, Inc.
28  */
29 
30 /* Portions Copyright 2007 Jeremy Teo */
31 /* Portions Copyright 2010 Robert Milkowski */
32 
33 #include <sys/types.h>
34 #include <sys/param.h>
35 #include <sys/time.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/resource.h>
39 #include <sys/vfs.h>
40 #include <sys/vfs_opreg.h>
41 #include <sys/vnode.h>
42 #include <sys/file.h>
43 #include <sys/stat.h>
44 #include <sys/kmem.h>
45 #include <sys/taskq.h>
46 #include <sys/uio.h>
47 #include <sys/vmsystm.h>
48 #include <sys/atomic.h>
49 #include <sys/vm.h>
50 #include <vm/seg_vn.h>
51 #include <vm/pvn.h>
52 #include <vm/as.h>
53 #include <vm/kpm.h>
54 #include <vm/seg_kpm.h>
55 #include <sys/mman.h>
56 #include <sys/pathname.h>
57 #include <sys/cmn_err.h>
58 #include <sys/errno.h>
59 #include <sys/unistd.h>
60 #include <sys/zfs_dir.h>
61 #include <sys/zfs_acl.h>
62 #include <sys/zfs_ioctl.h>
63 #include <sys/fs/zfs.h>
64 #include <sys/dmu.h>
65 #include <sys/dmu_objset.h>
66 #include <sys/spa.h>
67 #include <sys/txg.h>
68 #include <sys/dbuf.h>
69 #include <sys/zap.h>
70 #include <sys/sa.h>
71 #include <sys/dirent.h>
72 #include <sys/policy.h>
73 #include <sys/sunddi.h>
74 #include <sys/filio.h>
75 #include <sys/sid.h>
76 #include "fs/fs_subr.h"
77 #include <sys/zfs_ctldir.h>
78 #include <sys/zfs_fuid.h>
79 #include <sys/zfs_sa.h>
80 #include <sys/dnlc.h>
81 #include <sys/zfs_rlock.h>
82 #include <sys/extdirent.h>
83 #include <sys/kidmap.h>
84 #include <sys/cred.h>
85 #include <sys/attr.h>
86 #include <sys/zil.h>
87 
88 /*
89  * Programming rules.
90  *
91  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
92  * properly lock its in-core state, create a DMU transaction, do the work,
93  * record this work in the intent log (ZIL), commit the DMU transaction,
94  * and wait for the intent log to commit if it is a synchronous operation.
95  * Moreover, the vnode ops must work in both normal and log replay context.
96  * The ordering of events is important to avoid deadlocks and references
97  * to freed memory.  The example below illustrates the following Big Rules:
98  *
99  *  (1)	A check must be made in each zfs thread for a mounted file system.
100  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
101  *	A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
102  *	must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
103  *	can return EIO from the calling function.
104  *
105  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
106  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
107  *	First, if it's the last reference, the vnode/znode
108  *	can be freed, so the zp may point to freed memory.  Second, the last
109  *	reference will call zfs_zinactive(), which may induce a lot of work --
110  *	pushing cached pages (which acquires range locks) and syncing out
111  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
112  *	which could deadlock the system if you were already holding one.
113  *	If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
114  *
115  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
116  *	as they can span dmu_tx_assign() calls.
117  *
118  *  (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
119  *      dmu_tx_assign().  This is critical because we don't want to block
120  *      while holding locks.
121  *
122  *	If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT.  This
123  *	reduces lock contention and CPU usage when we must wait (note that if
124  *	throughput is constrained by the storage, nearly every transaction
125  *	must wait).
126  *
127  *      Note, in particular, that if a lock is sometimes acquired before
128  *      the tx assigns, and sometimes after (e.g. z_lock), then failing
129  *      to use a non-blocking assign can deadlock the system.  The scenario:
130  *
131  *	Thread A has grabbed a lock before calling dmu_tx_assign().
132  *	Thread B is in an already-assigned tx, and blocks for this lock.
133  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
134  *	forever, because the previous txg can't quiesce until B's tx commits.
135  *
136  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
137  *	then drop all locks, call dmu_tx_wait(), and try again.  On subsequent
138  *	calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT,
139  *	to indicate that this operation has already called dmu_tx_wait().
140  *	This will ensure that we don't retry forever, waiting a short bit
141  *	each time.
142  *
143  *  (5)	If the operation succeeded, generate the intent log entry for it
144  *	before dropping locks.  This ensures that the ordering of events
145  *	in the intent log matches the order in which they actually occurred.
146  *	During ZIL replay the zfs_log_* functions will update the sequence
147  *	number to indicate the zil transaction has replayed.
148  *
149  *  (6)	At the end of each vnode op, the DMU tx must always commit,
150  *	regardless of whether there were any errors.
151  *
152  *  (7)	After dropping all locks, invoke zil_commit(zilog, foid)
153  *	to ensure that synchronous semantics are provided when necessary.
154  *
155  * In general, this is how things should be ordered in each vnode op:
156  *
157  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
158  * top:
159  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
160  *	rw_enter(...);			// grab any other locks you need
161  *	tx = dmu_tx_create(...);	// get DMU tx
162  *	dmu_tx_hold_*();		// hold each object you might modify
163  *	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
164  *	if (error) {
165  *		rw_exit(...);		// drop locks
166  *		zfs_dirent_unlock(dl);	// unlock directory entry
167  *		VN_RELE(...);		// release held vnodes
168  *		if (error == ERESTART) {
169  *			waited = B_TRUE;
170  *			dmu_tx_wait(tx);
171  *			dmu_tx_abort(tx);
172  *			goto top;
173  *		}
174  *		dmu_tx_abort(tx);	// abort DMU tx
175  *		ZFS_EXIT(zfsvfs);	// finished in zfs
176  *		return (error);		// really out of space
177  *	}
178  *	error = do_real_work();		// do whatever this VOP does
179  *	if (error == 0)
180  *		zfs_log_*(...);		// on success, make ZIL entry
181  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
182  *	rw_exit(...);			// drop locks
183  *	zfs_dirent_unlock(dl);		// unlock directory entry
184  *	VN_RELE(...);			// release held vnodes
185  *	zil_commit(zilog, foid);	// synchronous when necessary
186  *	ZFS_EXIT(zfsvfs);		// finished in zfs
187  *	return (error);			// done, report error
188  */
189 
190 /* ARGSUSED */
191 static int
192 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
193 {
194 	znode_t	*zp = VTOZ(*vpp);
195 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
196 
197 	ZFS_ENTER(zfsvfs);
198 	ZFS_VERIFY_ZP(zp);
199 
200 	if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
201 	    ((flag & FAPPEND) == 0)) {
202 		ZFS_EXIT(zfsvfs);
203 		return (SET_ERROR(EPERM));
204 	}
205 
206 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
207 	    ZTOV(zp)->v_type == VREG &&
208 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
209 		if (fs_vscan(*vpp, cr, 0) != 0) {
210 			ZFS_EXIT(zfsvfs);
211 			return (SET_ERROR(EACCES));
212 		}
213 	}
214 
215 	/* Keep a count of the synchronous opens in the znode */
216 	if (flag & (FSYNC | FDSYNC))
217 		atomic_inc_32(&zp->z_sync_cnt);
218 
219 	ZFS_EXIT(zfsvfs);
220 	return (0);
221 }
222 
223 /* ARGSUSED */
224 static int
225 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
226     caller_context_t *ct)
227 {
228 	znode_t	*zp = VTOZ(vp);
229 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
230 
231 	/*
232 	 * Clean up any locks held by this process on the vp.
233 	 */
234 	cleanlocks(vp, ddi_get_pid(), 0);
235 	cleanshares(vp, ddi_get_pid());
236 
237 	ZFS_ENTER(zfsvfs);
238 	ZFS_VERIFY_ZP(zp);
239 
240 	/* Decrement the synchronous opens in the znode */
241 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
242 		atomic_dec_32(&zp->z_sync_cnt);
243 
244 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
245 	    ZTOV(zp)->v_type == VREG &&
246 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
247 		VERIFY(fs_vscan(vp, cr, 1) == 0);
248 
249 	ZFS_EXIT(zfsvfs);
250 	return (0);
251 }
252 
253 /*
254  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
255  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
256  */
257 static int
258 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
259 {
260 	znode_t	*zp = VTOZ(vp);
261 	uint64_t noff = (uint64_t)*off; /* new offset */
262 	uint64_t file_sz;
263 	int error;
264 	boolean_t hole;
265 
266 	file_sz = zp->z_size;
267 	if (noff >= file_sz)  {
268 		return (SET_ERROR(ENXIO));
269 	}
270 
271 	if (cmd == _FIO_SEEK_HOLE)
272 		hole = B_TRUE;
273 	else
274 		hole = B_FALSE;
275 
276 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
277 
278 	if (error == ESRCH)
279 		return (SET_ERROR(ENXIO));
280 
281 	/*
282 	 * We could find a hole that begins after the logical end-of-file,
283 	 * because dmu_offset_next() only works on whole blocks.  If the
284 	 * EOF falls mid-block, then indicate that the "virtual hole"
285 	 * at the end of the file begins at the logical EOF, rather than
286 	 * at the end of the last block.
287 	 */
288 	if (noff > file_sz) {
289 		ASSERT(hole);
290 		noff = file_sz;
291 	}
292 
293 	if (noff < *off)
294 		return (error);
295 	*off = noff;
296 	return (error);
297 }
298 
299 /* ARGSUSED */
300 static int
301 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
302     int *rvalp, caller_context_t *ct)
303 {
304 	offset_t off;
305 	offset_t ndata;
306 	dmu_object_info_t doi;
307 	int error;
308 	zfsvfs_t *zfsvfs;
309 	znode_t *zp;
310 
311 	switch (com) {
312 	case _FIOFFS:
313 	{
314 		return (zfs_sync(vp->v_vfsp, 0, cred));
315 
316 		/*
317 		 * The following two ioctls are used by bfu.  Faking out,
318 		 * necessary to avoid bfu errors.
319 		 */
320 	}
321 	case _FIOGDIO:
322 	case _FIOSDIO:
323 	{
324 		return (0);
325 	}
326 
327 	case _FIO_SEEK_DATA:
328 	case _FIO_SEEK_HOLE:
329 	{
330 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
331 			return (SET_ERROR(EFAULT));
332 
333 		zp = VTOZ(vp);
334 		zfsvfs = zp->z_zfsvfs;
335 		ZFS_ENTER(zfsvfs);
336 		ZFS_VERIFY_ZP(zp);
337 
338 		/* offset parameter is in/out */
339 		error = zfs_holey(vp, com, &off);
340 		ZFS_EXIT(zfsvfs);
341 		if (error)
342 			return (error);
343 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
344 			return (SET_ERROR(EFAULT));
345 		return (0);
346 	}
347 	case _FIO_COUNT_FILLED:
348 	{
349 		/*
350 		 * _FIO_COUNT_FILLED adds a new ioctl command which
351 		 * exposes the number of filled blocks in a
352 		 * ZFS object.
353 		 */
354 		zp = VTOZ(vp);
355 		zfsvfs = zp->z_zfsvfs;
356 		ZFS_ENTER(zfsvfs);
357 		ZFS_VERIFY_ZP(zp);
358 
359 		/*
360 		 * Wait for all dirty blocks for this object
361 		 * to get synced out to disk, and the DMU info
362 		 * updated.
363 		 */
364 		error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id);
365 		if (error) {
366 			ZFS_EXIT(zfsvfs);
367 			return (error);
368 		}
369 
370 		/*
371 		 * Retrieve fill count from DMU object.
372 		 */
373 		error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi);
374 		if (error) {
375 			ZFS_EXIT(zfsvfs);
376 			return (error);
377 		}
378 
379 		ndata = doi.doi_fill_count;
380 
381 		ZFS_EXIT(zfsvfs);
382 		if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag))
383 			return (SET_ERROR(EFAULT));
384 		return (0);
385 	}
386 	}
387 	return (SET_ERROR(ENOTTY));
388 }
389 
390 /*
391  * Utility functions to map and unmap a single physical page.  These
392  * are used to manage the mappable copies of ZFS file data, and therefore
393  * do not update ref/mod bits.
394  */
395 caddr_t
396 zfs_map_page(page_t *pp, enum seg_rw rw)
397 {
398 	if (kpm_enable)
399 		return (hat_kpm_mapin(pp, 0));
400 	ASSERT(rw == S_READ || rw == S_WRITE);
401 	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
402 	    (caddr_t)-1));
403 }
404 
405 void
406 zfs_unmap_page(page_t *pp, caddr_t addr)
407 {
408 	if (kpm_enable) {
409 		hat_kpm_mapout(pp, 0, addr);
410 	} else {
411 		ppmapout(addr);
412 	}
413 }
414 
415 /*
416  * When a file is memory mapped, we must keep the IO data synchronized
417  * between the DMU cache and the memory mapped pages.  What this means:
418  *
419  * On Write:	If we find a memory mapped page, we write to *both*
420  *		the page and the dmu buffer.
421  */
422 static void
423 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
424 {
425 	int64_t	off;
426 
427 	off = start & PAGEOFFSET;
428 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
429 		page_t *pp;
430 		uint64_t nbytes = MIN(PAGESIZE - off, len);
431 
432 		if (pp = page_lookup(vp, start, SE_SHARED)) {
433 			caddr_t va;
434 
435 			va = zfs_map_page(pp, S_WRITE);
436 			(void) dmu_read(os, oid, start+off, nbytes, va+off,
437 			    DMU_READ_PREFETCH);
438 			zfs_unmap_page(pp, va);
439 			page_unlock(pp);
440 		}
441 		len -= nbytes;
442 		off = 0;
443 	}
444 }
445 
446 /*
447  * When a file is memory mapped, we must keep the IO data synchronized
448  * between the DMU cache and the memory mapped pages.  What this means:
449  *
450  * On Read:	We "read" preferentially from memory mapped pages,
451  *		else we default from the dmu buffer.
452  *
453  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
454  *	 the file is memory mapped.
455  */
456 static int
457 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
458 {
459 	znode_t *zp = VTOZ(vp);
460 	int64_t	start, off;
461 	int len = nbytes;
462 	int error = 0;
463 
464 	start = uio->uio_loffset;
465 	off = start & PAGEOFFSET;
466 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
467 		page_t *pp;
468 		uint64_t bytes = MIN(PAGESIZE - off, len);
469 
470 		if (pp = page_lookup(vp, start, SE_SHARED)) {
471 			caddr_t va;
472 
473 			va = zfs_map_page(pp, S_READ);
474 			error = uiomove(va + off, bytes, UIO_READ, uio);
475 			zfs_unmap_page(pp, va);
476 			page_unlock(pp);
477 		} else {
478 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
479 			    uio, bytes);
480 		}
481 		len -= bytes;
482 		off = 0;
483 		if (error)
484 			break;
485 	}
486 	return (error);
487 }
488 
489 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
490 
491 /*
492  * Read bytes from specified file into supplied buffer.
493  *
494  *	IN:	vp	- vnode of file to be read from.
495  *		uio	- structure supplying read location, range info,
496  *			  and return buffer.
497  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
498  *		cr	- credentials of caller.
499  *		ct	- caller context
500  *
501  *	OUT:	uio	- updated offset and range, buffer filled.
502  *
503  *	RETURN:	0 on success, error code on failure.
504  *
505  * Side Effects:
506  *	vp - atime updated if byte count > 0
507  */
508 /* ARGSUSED */
509 static int
510 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
511 {
512 	znode_t		*zp = VTOZ(vp);
513 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
514 	ssize_t		n, nbytes;
515 	int		error = 0;
516 	xuio_t		*xuio = NULL;
517 
518 	ZFS_ENTER(zfsvfs);
519 	ZFS_VERIFY_ZP(zp);
520 
521 	if (zp->z_pflags & ZFS_AV_QUARANTINED) {
522 		ZFS_EXIT(zfsvfs);
523 		return (SET_ERROR(EACCES));
524 	}
525 
526 	/*
527 	 * Validate file offset
528 	 */
529 	if (uio->uio_loffset < (offset_t)0) {
530 		ZFS_EXIT(zfsvfs);
531 		return (SET_ERROR(EINVAL));
532 	}
533 
534 	/*
535 	 * Fasttrack empty reads
536 	 */
537 	if (uio->uio_resid == 0) {
538 		ZFS_EXIT(zfsvfs);
539 		return (0);
540 	}
541 
542 	/*
543 	 * Check for mandatory locks
544 	 */
545 	if (MANDMODE(zp->z_mode)) {
546 		if (error = chklock(vp, FREAD,
547 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
548 			ZFS_EXIT(zfsvfs);
549 			return (error);
550 		}
551 	}
552 
553 	/*
554 	 * If we're in FRSYNC mode, sync out this znode before reading it.
555 	 */
556 	if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
557 		zil_commit(zfsvfs->z_log, zp->z_id);
558 
559 	/*
560 	 * Lock the range against changes.
561 	 */
562 	locked_range_t *lr = rangelock_enter(&zp->z_rangelock,
563 	    uio->uio_loffset, uio->uio_resid, RL_READER);
564 
565 	/*
566 	 * If we are reading past end-of-file we can skip
567 	 * to the end; but we might still need to set atime.
568 	 */
569 	if (uio->uio_loffset >= zp->z_size) {
570 		error = 0;
571 		goto out;
572 	}
573 
574 	ASSERT(uio->uio_loffset < zp->z_size);
575 	n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
576 
577 	if ((uio->uio_extflg == UIO_XUIO) &&
578 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
579 		int nblk;
580 		int blksz = zp->z_blksz;
581 		uint64_t offset = uio->uio_loffset;
582 
583 		xuio = (xuio_t *)uio;
584 		if ((ISP2(blksz))) {
585 			nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
586 			    blksz)) / blksz;
587 		} else {
588 			ASSERT(offset + n <= blksz);
589 			nblk = 1;
590 		}
591 		(void) dmu_xuio_init(xuio, nblk);
592 
593 		if (vn_has_cached_data(vp)) {
594 			/*
595 			 * For simplicity, we always allocate a full buffer
596 			 * even if we only expect to read a portion of a block.
597 			 */
598 			while (--nblk >= 0) {
599 				(void) dmu_xuio_add(xuio,
600 				    dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
601 				    blksz), 0, blksz);
602 			}
603 		}
604 	}
605 
606 	while (n > 0) {
607 		nbytes = MIN(n, zfs_read_chunk_size -
608 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
609 
610 		if (vn_has_cached_data(vp)) {
611 			error = mappedread(vp, nbytes, uio);
612 		} else {
613 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
614 			    uio, nbytes);
615 		}
616 		if (error) {
617 			/* convert checksum errors into IO errors */
618 			if (error == ECKSUM)
619 				error = SET_ERROR(EIO);
620 			break;
621 		}
622 
623 		n -= nbytes;
624 	}
625 out:
626 	rangelock_exit(lr);
627 
628 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
629 	ZFS_EXIT(zfsvfs);
630 	return (error);
631 }
632 
633 /*
634  * Write the bytes to a file.
635  *
636  *	IN:	vp	- vnode of file to be written to.
637  *		uio	- structure supplying write location, range info,
638  *			  and data buffer.
639  *		ioflag	- FAPPEND, FSYNC, and/or FDSYNC.  FAPPEND is
640  *			  set if in append mode.
641  *		cr	- credentials of caller.
642  *		ct	- caller context (NFS/CIFS fem monitor only)
643  *
644  *	OUT:	uio	- updated offset and range.
645  *
646  *	RETURN:	0 on success, error code on failure.
647  *
648  * Timestamps:
649  *	vp - ctime|mtime updated if byte count > 0
650  */
651 
652 /* ARGSUSED */
653 static int
654 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
655 {
656 	znode_t		*zp = VTOZ(vp);
657 	rlim64_t	limit = uio->uio_llimit;
658 	ssize_t		start_resid = uio->uio_resid;
659 	ssize_t		tx_bytes;
660 	uint64_t	end_size;
661 	dmu_tx_t	*tx;
662 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
663 	zilog_t		*zilog;
664 	offset_t	woff;
665 	ssize_t		n, nbytes;
666 	int		max_blksz = zfsvfs->z_max_blksz;
667 	int		error = 0;
668 	arc_buf_t	*abuf;
669 	iovec_t		*aiov = NULL;
670 	xuio_t		*xuio = NULL;
671 	int		i_iov = 0;
672 	int		iovcnt = uio->uio_iovcnt;
673 	iovec_t		*iovp = uio->uio_iov;
674 	int		write_eof;
675 	int		count = 0;
676 	sa_bulk_attr_t	bulk[4];
677 	uint64_t	mtime[2], ctime[2];
678 
679 	/*
680 	 * Fasttrack empty write
681 	 */
682 	n = start_resid;
683 	if (n == 0)
684 		return (0);
685 
686 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
687 		limit = MAXOFFSET_T;
688 
689 	ZFS_ENTER(zfsvfs);
690 	ZFS_VERIFY_ZP(zp);
691 
692 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
693 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
694 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
695 	    &zp->z_size, 8);
696 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
697 	    &zp->z_pflags, 8);
698 
699 	/*
700 	 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
701 	 * callers might not be able to detect properly that we are read-only,
702 	 * so check it explicitly here.
703 	 */
704 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
705 		ZFS_EXIT(zfsvfs);
706 		return (SET_ERROR(EROFS));
707 	}
708 
709 	/*
710 	 * If immutable or not appending then return EPERM.
711 	 * Intentionally allow ZFS_READONLY through here.
712 	 * See zfs_zaccess_common()
713 	 */
714 	if ((zp->z_pflags & ZFS_IMMUTABLE) ||
715 	    ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
716 	    (uio->uio_loffset < zp->z_size))) {
717 		ZFS_EXIT(zfsvfs);
718 		return (SET_ERROR(EPERM));
719 	}
720 
721 	zilog = zfsvfs->z_log;
722 
723 	/*
724 	 * Validate file offset
725 	 */
726 	woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
727 	if (woff < 0) {
728 		ZFS_EXIT(zfsvfs);
729 		return (SET_ERROR(EINVAL));
730 	}
731 
732 	/*
733 	 * Check for mandatory locks before calling rangelock_enter()
734 	 * in order to prevent a deadlock with locks set via fcntl().
735 	 */
736 	if (MANDMODE((mode_t)zp->z_mode) &&
737 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
738 		ZFS_EXIT(zfsvfs);
739 		return (error);
740 	}
741 
742 	/*
743 	 * Pre-fault the pages to ensure slow (eg NFS) pages
744 	 * don't hold up txg.
745 	 * Skip this if uio contains loaned arc_buf.
746 	 */
747 	if ((uio->uio_extflg == UIO_XUIO) &&
748 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
749 		xuio = (xuio_t *)uio;
750 	else
751 		uio_prefaultpages(MIN(n, max_blksz), uio);
752 
753 	/*
754 	 * If in append mode, set the io offset pointer to eof.
755 	 */
756 	locked_range_t *lr;
757 	if (ioflag & FAPPEND) {
758 		/*
759 		 * Obtain an appending range lock to guarantee file append
760 		 * semantics.  We reset the write offset once we have the lock.
761 		 */
762 		lr = rangelock_enter(&zp->z_rangelock, 0, n, RL_APPEND);
763 		woff = lr->lr_offset;
764 		if (lr->lr_length == UINT64_MAX) {
765 			/*
766 			 * We overlocked the file because this write will cause
767 			 * the file block size to increase.
768 			 * Note that zp_size cannot change with this lock held.
769 			 */
770 			woff = zp->z_size;
771 		}
772 		uio->uio_loffset = woff;
773 	} else {
774 		/*
775 		 * Note that if the file block size will change as a result of
776 		 * this write, then this range lock will lock the entire file
777 		 * so that we can re-write the block safely.
778 		 */
779 		lr = rangelock_enter(&zp->z_rangelock, woff, n, RL_WRITER);
780 	}
781 
782 	if (woff >= limit) {
783 		rangelock_exit(lr);
784 		ZFS_EXIT(zfsvfs);
785 		return (SET_ERROR(EFBIG));
786 	}
787 
788 	if ((woff + n) > limit || woff > (limit - n))
789 		n = limit - woff;
790 
791 	/* Will this write extend the file length? */
792 	write_eof = (woff + n > zp->z_size);
793 
794 	end_size = MAX(zp->z_size, woff + n);
795 
796 	/*
797 	 * Write the file in reasonable size chunks.  Each chunk is written
798 	 * in a separate transaction; this keeps the intent log records small
799 	 * and allows us to do more fine-grained space accounting.
800 	 */
801 	while (n > 0) {
802 		abuf = NULL;
803 		woff = uio->uio_loffset;
804 		if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
805 		    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
806 			if (abuf != NULL)
807 				dmu_return_arcbuf(abuf);
808 			error = SET_ERROR(EDQUOT);
809 			break;
810 		}
811 
812 		if (xuio && abuf == NULL) {
813 			ASSERT(i_iov < iovcnt);
814 			aiov = &iovp[i_iov];
815 			abuf = dmu_xuio_arcbuf(xuio, i_iov);
816 			dmu_xuio_clear(xuio, i_iov);
817 			DTRACE_PROBE3(zfs_cp_write, int, i_iov,
818 			    iovec_t *, aiov, arc_buf_t *, abuf);
819 			ASSERT((aiov->iov_base == abuf->b_data) ||
820 			    ((char *)aiov->iov_base - (char *)abuf->b_data +
821 			    aiov->iov_len == arc_buf_size(abuf)));
822 			i_iov++;
823 		} else if (abuf == NULL && n >= max_blksz &&
824 		    woff >= zp->z_size &&
825 		    P2PHASE(woff, max_blksz) == 0 &&
826 		    zp->z_blksz == max_blksz) {
827 			/*
828 			 * This write covers a full block.  "Borrow" a buffer
829 			 * from the dmu so that we can fill it before we enter
830 			 * a transaction.  This avoids the possibility of
831 			 * holding up the transaction if the data copy hangs
832 			 * up on a pagefault (e.g., from an NFS server mapping).
833 			 */
834 			size_t cbytes;
835 
836 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
837 			    max_blksz);
838 			ASSERT(abuf != NULL);
839 			ASSERT(arc_buf_size(abuf) == max_blksz);
840 			if (error = uiocopy(abuf->b_data, max_blksz,
841 			    UIO_WRITE, uio, &cbytes)) {
842 				dmu_return_arcbuf(abuf);
843 				break;
844 			}
845 			ASSERT(cbytes == max_blksz);
846 		}
847 
848 		/*
849 		 * Start a transaction.
850 		 */
851 		tx = dmu_tx_create(zfsvfs->z_os);
852 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
853 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
854 		zfs_sa_upgrade_txholds(tx, zp);
855 		error = dmu_tx_assign(tx, TXG_WAIT);
856 		if (error) {
857 			dmu_tx_abort(tx);
858 			if (abuf != NULL)
859 				dmu_return_arcbuf(abuf);
860 			break;
861 		}
862 
863 		/*
864 		 * If rangelock_enter() over-locked we grow the blocksize
865 		 * and then reduce the lock range.  This will only happen
866 		 * on the first iteration since rangelock_reduce() will
867 		 * shrink down lr_length to the appropriate size.
868 		 */
869 		if (lr->lr_length == UINT64_MAX) {
870 			uint64_t new_blksz;
871 
872 			if (zp->z_blksz > max_blksz) {
873 				/*
874 				 * File's blocksize is already larger than the
875 				 * "recordsize" property.  Only let it grow to
876 				 * the next power of 2.
877 				 */
878 				ASSERT(!ISP2(zp->z_blksz));
879 				new_blksz = MIN(end_size,
880 				    1 << highbit64(zp->z_blksz));
881 			} else {
882 				new_blksz = MIN(end_size, max_blksz);
883 			}
884 			zfs_grow_blocksize(zp, new_blksz, tx);
885 			rangelock_reduce(lr, woff, n);
886 		}
887 
888 		/*
889 		 * XXX - should we really limit each write to z_max_blksz?
890 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
891 		 */
892 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
893 
894 		if (abuf == NULL) {
895 			tx_bytes = uio->uio_resid;
896 			error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
897 			    uio, nbytes, tx);
898 			tx_bytes -= uio->uio_resid;
899 		} else {
900 			tx_bytes = nbytes;
901 			ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
902 			/*
903 			 * If this is not a full block write, but we are
904 			 * extending the file past EOF and this data starts
905 			 * block-aligned, use assign_arcbuf().  Otherwise,
906 			 * write via dmu_write().
907 			 */
908 			if (tx_bytes < max_blksz && (!write_eof ||
909 			    aiov->iov_base != abuf->b_data)) {
910 				ASSERT(xuio);
911 				dmu_write(zfsvfs->z_os, zp->z_id, woff,
912 				    aiov->iov_len, aiov->iov_base, tx);
913 				dmu_return_arcbuf(abuf);
914 				xuio_stat_wbuf_copied();
915 			} else {
916 				ASSERT(xuio || tx_bytes == max_blksz);
917 				dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
918 				    woff, abuf, tx);
919 			}
920 			ASSERT(tx_bytes <= uio->uio_resid);
921 			uioskip(uio, tx_bytes);
922 		}
923 		if (tx_bytes && vn_has_cached_data(vp)) {
924 			update_pages(vp, woff,
925 			    tx_bytes, zfsvfs->z_os, zp->z_id);
926 		}
927 
928 		/*
929 		 * If we made no progress, we're done.  If we made even
930 		 * partial progress, update the znode and ZIL accordingly.
931 		 */
932 		if (tx_bytes == 0) {
933 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
934 			    (void *)&zp->z_size, sizeof (uint64_t), tx);
935 			dmu_tx_commit(tx);
936 			ASSERT(error != 0);
937 			break;
938 		}
939 
940 		/*
941 		 * Clear Set-UID/Set-GID bits on successful write if not
942 		 * privileged and at least one of the excute bits is set.
943 		 *
944 		 * It would be nice to to this after all writes have
945 		 * been done, but that would still expose the ISUID/ISGID
946 		 * to another app after the partial write is committed.
947 		 *
948 		 * Note: we don't call zfs_fuid_map_id() here because
949 		 * user 0 is not an ephemeral uid.
950 		 */
951 		mutex_enter(&zp->z_acl_lock);
952 		if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
953 		    (S_IXUSR >> 6))) != 0 &&
954 		    (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
955 		    secpolicy_vnode_setid_retain(cr,
956 		    (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
957 			uint64_t newmode;
958 			zp->z_mode &= ~(S_ISUID | S_ISGID);
959 			newmode = zp->z_mode;
960 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
961 			    (void *)&newmode, sizeof (uint64_t), tx);
962 		}
963 		mutex_exit(&zp->z_acl_lock);
964 
965 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
966 		    B_TRUE);
967 
968 		/*
969 		 * Update the file size (zp_size) if it has changed;
970 		 * account for possible concurrent updates.
971 		 */
972 		while ((end_size = zp->z_size) < uio->uio_loffset) {
973 			(void) atomic_cas_64(&zp->z_size, end_size,
974 			    uio->uio_loffset);
975 			ASSERT(error == 0);
976 		}
977 		/*
978 		 * If we are replaying and eof is non zero then force
979 		 * the file size to the specified eof. Note, there's no
980 		 * concurrency during replay.
981 		 */
982 		if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
983 			zp->z_size = zfsvfs->z_replay_eof;
984 
985 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
986 
987 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
988 		dmu_tx_commit(tx);
989 
990 		if (error != 0)
991 			break;
992 		ASSERT(tx_bytes == nbytes);
993 		n -= nbytes;
994 
995 		if (!xuio && n > 0)
996 			uio_prefaultpages(MIN(n, max_blksz), uio);
997 	}
998 
999 	rangelock_exit(lr);
1000 
1001 	/*
1002 	 * If we're in replay mode, or we made no progress, return error.
1003 	 * Otherwise, it's at least a partial write, so it's successful.
1004 	 */
1005 	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1006 		ZFS_EXIT(zfsvfs);
1007 		return (error);
1008 	}
1009 
1010 	if (ioflag & (FSYNC | FDSYNC) ||
1011 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1012 		zil_commit(zilog, zp->z_id);
1013 
1014 	ZFS_EXIT(zfsvfs);
1015 	return (0);
1016 }
1017 
1018 /* ARGSUSED */
1019 void
1020 zfs_get_done(zgd_t *zgd, int error)
1021 {
1022 	znode_t *zp = zgd->zgd_private;
1023 	objset_t *os = zp->z_zfsvfs->z_os;
1024 
1025 	if (zgd->zgd_db)
1026 		dmu_buf_rele(zgd->zgd_db, zgd);
1027 
1028 	rangelock_exit(zgd->zgd_lr);
1029 
1030 	/*
1031 	 * Release the vnode asynchronously as we currently have the
1032 	 * txg stopped from syncing.
1033 	 */
1034 	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1035 
1036 	kmem_free(zgd, sizeof (zgd_t));
1037 }
1038 
1039 #ifdef DEBUG
1040 static int zil_fault_io = 0;
1041 #endif
1042 
1043 /*
1044  * Get data to generate a TX_WRITE intent log record.
1045  */
1046 int
1047 zfs_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
1048 {
1049 	zfsvfs_t *zfsvfs = arg;
1050 	objset_t *os = zfsvfs->z_os;
1051 	znode_t *zp;
1052 	uint64_t object = lr->lr_foid;
1053 	uint64_t offset = lr->lr_offset;
1054 	uint64_t size = lr->lr_length;
1055 	dmu_buf_t *db;
1056 	zgd_t *zgd;
1057 	int error = 0;
1058 
1059 	ASSERT3P(lwb, !=, NULL);
1060 	ASSERT3P(zio, !=, NULL);
1061 	ASSERT3U(size, !=, 0);
1062 
1063 	/*
1064 	 * Nothing to do if the file has been removed
1065 	 */
1066 	if (zfs_zget(zfsvfs, object, &zp) != 0)
1067 		return (SET_ERROR(ENOENT));
1068 	if (zp->z_unlinked) {
1069 		/*
1070 		 * Release the vnode asynchronously as we currently have the
1071 		 * txg stopped from syncing.
1072 		 */
1073 		VN_RELE_ASYNC(ZTOV(zp),
1074 		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1075 		return (SET_ERROR(ENOENT));
1076 	}
1077 
1078 	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1079 	zgd->zgd_lwb = lwb;
1080 	zgd->zgd_private = zp;
1081 
1082 	/*
1083 	 * Write records come in two flavors: immediate and indirect.
1084 	 * For small writes it's cheaper to store the data with the
1085 	 * log record (immediate); for large writes it's cheaper to
1086 	 * sync the data and get a pointer to it (indirect) so that
1087 	 * we don't have to write the data twice.
1088 	 */
1089 	if (buf != NULL) { /* immediate write */
1090 		zgd->zgd_lr = rangelock_enter(&zp->z_rangelock,
1091 		    offset, size, RL_READER);
1092 		/* test for truncation needs to be done while range locked */
1093 		if (offset >= zp->z_size) {
1094 			error = SET_ERROR(ENOENT);
1095 		} else {
1096 			error = dmu_read(os, object, offset, size, buf,
1097 			    DMU_READ_NO_PREFETCH);
1098 		}
1099 		ASSERT(error == 0 || error == ENOENT);
1100 	} else { /* indirect write */
1101 		/*
1102 		 * Have to lock the whole block to ensure when it's
1103 		 * written out and its checksum is being calculated
1104 		 * that no one can change the data. We need to re-check
1105 		 * blocksize after we get the lock in case it's changed!
1106 		 */
1107 		for (;;) {
1108 			uint64_t blkoff;
1109 			size = zp->z_blksz;
1110 			blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1111 			offset -= blkoff;
1112 			zgd->zgd_lr = rangelock_enter(&zp->z_rangelock,
1113 			    offset, size, RL_READER);
1114 			if (zp->z_blksz == size)
1115 				break;
1116 			offset += blkoff;
1117 			rangelock_exit(zgd->zgd_lr);
1118 		}
1119 		/* test for truncation needs to be done while range locked */
1120 		if (lr->lr_offset >= zp->z_size)
1121 			error = SET_ERROR(ENOENT);
1122 #ifdef DEBUG
1123 		if (zil_fault_io) {
1124 			error = SET_ERROR(EIO);
1125 			zil_fault_io = 0;
1126 		}
1127 #endif
1128 		if (error == 0)
1129 			error = dmu_buf_hold(os, object, offset, zgd, &db,
1130 			    DMU_READ_NO_PREFETCH);
1131 
1132 		if (error == 0) {
1133 			blkptr_t *bp = &lr->lr_blkptr;
1134 
1135 			zgd->zgd_db = db;
1136 			zgd->zgd_bp = bp;
1137 
1138 			ASSERT(db->db_offset == offset);
1139 			ASSERT(db->db_size == size);
1140 
1141 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1142 			    zfs_get_done, zgd);
1143 			ASSERT(error || lr->lr_length <= size);
1144 
1145 			/*
1146 			 * On success, we need to wait for the write I/O
1147 			 * initiated by dmu_sync() to complete before we can
1148 			 * release this dbuf.  We will finish everything up
1149 			 * in the zfs_get_done() callback.
1150 			 */
1151 			if (error == 0)
1152 				return (0);
1153 
1154 			if (error == EALREADY) {
1155 				lr->lr_common.lrc_txtype = TX_WRITE2;
1156 				/*
1157 				 * TX_WRITE2 relies on the data previously
1158 				 * written by the TX_WRITE that caused
1159 				 * EALREADY.  We zero out the BP because
1160 				 * it is the old, currently-on-disk BP.
1161 				 */
1162 				zgd->zgd_bp = NULL;
1163 				BP_ZERO(bp);
1164 				error = 0;
1165 			}
1166 		}
1167 	}
1168 
1169 	zfs_get_done(zgd, error);
1170 
1171 	return (error);
1172 }
1173 
1174 /*ARGSUSED*/
1175 static int
1176 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1177     caller_context_t *ct)
1178 {
1179 	znode_t *zp = VTOZ(vp);
1180 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1181 	int error;
1182 
1183 	ZFS_ENTER(zfsvfs);
1184 	ZFS_VERIFY_ZP(zp);
1185 
1186 	if (flag & V_ACE_MASK)
1187 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1188 	else
1189 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
1190 
1191 	ZFS_EXIT(zfsvfs);
1192 	return (error);
1193 }
1194 
1195 /*
1196  * If vnode is for a device return a specfs vnode instead.
1197  */
1198 static int
1199 specvp_check(vnode_t **vpp, cred_t *cr)
1200 {
1201 	int error = 0;
1202 
1203 	if (IS_DEVVP(*vpp)) {
1204 		struct vnode *svp;
1205 
1206 		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1207 		VN_RELE(*vpp);
1208 		if (svp == NULL)
1209 			error = SET_ERROR(ENOSYS);
1210 		*vpp = svp;
1211 	}
1212 	return (error);
1213 }
1214 
1215 
1216 /*
1217  * Lookup an entry in a directory, or an extended attribute directory.
1218  * If it exists, return a held vnode reference for it.
1219  *
1220  *	IN:	dvp	- vnode of directory to search.
1221  *		nm	- name of entry to lookup.
1222  *		pnp	- full pathname to lookup [UNUSED].
1223  *		flags	- LOOKUP_XATTR set if looking for an attribute.
1224  *		rdir	- root directory vnode [UNUSED].
1225  *		cr	- credentials of caller.
1226  *		ct	- caller context
1227  *		direntflags - directory lookup flags
1228  *		realpnp - returned pathname.
1229  *
1230  *	OUT:	vpp	- vnode of located entry, NULL if not found.
1231  *
1232  *	RETURN:	0 on success, error code on failure.
1233  *
1234  * Timestamps:
1235  *	NA
1236  */
1237 /* ARGSUSED */
1238 static int
1239 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1240     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1241     int *direntflags, pathname_t *realpnp)
1242 {
1243 	znode_t *zdp = VTOZ(dvp);
1244 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1245 	int	error = 0;
1246 
1247 	/*
1248 	 * Fast path lookup, however we must skip DNLC lookup
1249 	 * for case folding or normalizing lookups because the
1250 	 * DNLC code only stores the passed in name.  This means
1251 	 * creating 'a' and removing 'A' on a case insensitive
1252 	 * file system would work, but DNLC still thinks 'a'
1253 	 * exists and won't let you create it again on the next
1254 	 * pass through fast path.
1255 	 */
1256 	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1257 
1258 		if (dvp->v_type != VDIR) {
1259 			return (SET_ERROR(ENOTDIR));
1260 		} else if (zdp->z_sa_hdl == NULL) {
1261 			return (SET_ERROR(EIO));
1262 		}
1263 
1264 		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1265 			error = zfs_fastaccesschk_execute(zdp, cr);
1266 			if (!error) {
1267 				*vpp = dvp;
1268 				VN_HOLD(*vpp);
1269 				return (0);
1270 			}
1271 			return (error);
1272 		} else if (!zdp->z_zfsvfs->z_norm &&
1273 		    (zdp->z_zfsvfs->z_case == ZFS_CASE_SENSITIVE)) {
1274 
1275 			vnode_t *tvp = dnlc_lookup(dvp, nm);
1276 
1277 			if (tvp) {
1278 				error = zfs_fastaccesschk_execute(zdp, cr);
1279 				if (error) {
1280 					VN_RELE(tvp);
1281 					return (error);
1282 				}
1283 				if (tvp == DNLC_NO_VNODE) {
1284 					VN_RELE(tvp);
1285 					return (SET_ERROR(ENOENT));
1286 				} else {
1287 					*vpp = tvp;
1288 					return (specvp_check(vpp, cr));
1289 				}
1290 			}
1291 		}
1292 	}
1293 
1294 	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1295 
1296 	ZFS_ENTER(zfsvfs);
1297 	ZFS_VERIFY_ZP(zdp);
1298 
1299 	*vpp = NULL;
1300 
1301 	if (flags & LOOKUP_XATTR) {
1302 		/*
1303 		 * If the xattr property is off, refuse the lookup request.
1304 		 */
1305 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1306 			ZFS_EXIT(zfsvfs);
1307 			return (SET_ERROR(EINVAL));
1308 		}
1309 
1310 		/*
1311 		 * We don't allow recursive attributes..
1312 		 * Maybe someday we will.
1313 		 */
1314 		if (zdp->z_pflags & ZFS_XATTR) {
1315 			ZFS_EXIT(zfsvfs);
1316 			return (SET_ERROR(EINVAL));
1317 		}
1318 
1319 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1320 			ZFS_EXIT(zfsvfs);
1321 			return (error);
1322 		}
1323 
1324 		/*
1325 		 * Do we have permission to get into attribute directory?
1326 		 */
1327 
1328 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1329 		    B_FALSE, cr)) {
1330 			VN_RELE(*vpp);
1331 			*vpp = NULL;
1332 		}
1333 
1334 		ZFS_EXIT(zfsvfs);
1335 		return (error);
1336 	}
1337 
1338 	if (dvp->v_type != VDIR) {
1339 		ZFS_EXIT(zfsvfs);
1340 		return (SET_ERROR(ENOTDIR));
1341 	}
1342 
1343 	/*
1344 	 * Check accessibility of directory.
1345 	 */
1346 
1347 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1348 		ZFS_EXIT(zfsvfs);
1349 		return (error);
1350 	}
1351 
1352 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1353 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1354 		ZFS_EXIT(zfsvfs);
1355 		return (SET_ERROR(EILSEQ));
1356 	}
1357 
1358 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1359 	if (error == 0)
1360 		error = specvp_check(vpp, cr);
1361 
1362 	ZFS_EXIT(zfsvfs);
1363 	return (error);
1364 }
1365 
1366 /*
1367  * Attempt to create a new entry in a directory.  If the entry
1368  * already exists, truncate the file if permissible, else return
1369  * an error.  Return the vp of the created or trunc'd file.
1370  *
1371  *	IN:	dvp	- vnode of directory to put new file entry in.
1372  *		name	- name of new file entry.
1373  *		vap	- attributes of new file.
1374  *		excl	- flag indicating exclusive or non-exclusive mode.
1375  *		mode	- mode to open file with.
1376  *		cr	- credentials of caller.
1377  *		flag	- large file flag [UNUSED].
1378  *		ct	- caller context
1379  *		vsecp	- ACL to be set
1380  *
1381  *	OUT:	vpp	- vnode of created or trunc'd entry.
1382  *
1383  *	RETURN:	0 on success, error code on failure.
1384  *
1385  * Timestamps:
1386  *	dvp - ctime|mtime updated if new entry created
1387  *	 vp - ctime|mtime always, atime if new
1388  */
1389 
1390 /* ARGSUSED */
1391 static int
1392 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1393     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1394     vsecattr_t *vsecp)
1395 {
1396 	znode_t		*zp, *dzp = VTOZ(dvp);
1397 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1398 	zilog_t		*zilog;
1399 	objset_t	*os;
1400 	zfs_dirlock_t	*dl;
1401 	dmu_tx_t	*tx;
1402 	int		error;
1403 	ksid_t		*ksid;
1404 	uid_t		uid;
1405 	gid_t		gid = crgetgid(cr);
1406 	zfs_acl_ids_t   acl_ids;
1407 	boolean_t	fuid_dirtied;
1408 	boolean_t	have_acl = B_FALSE;
1409 	boolean_t	waited = B_FALSE;
1410 
1411 	/*
1412 	 * If we have an ephemeral id, ACL, or XVATTR then
1413 	 * make sure file system is at proper version
1414 	 */
1415 
1416 	ksid = crgetsid(cr, KSID_OWNER);
1417 	if (ksid)
1418 		uid = ksid_getid(ksid);
1419 	else
1420 		uid = crgetuid(cr);
1421 
1422 	if (zfsvfs->z_use_fuids == B_FALSE &&
1423 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1424 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1425 		return (SET_ERROR(EINVAL));
1426 
1427 	ZFS_ENTER(zfsvfs);
1428 	ZFS_VERIFY_ZP(dzp);
1429 	os = zfsvfs->z_os;
1430 	zilog = zfsvfs->z_log;
1431 
1432 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1433 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1434 		ZFS_EXIT(zfsvfs);
1435 		return (SET_ERROR(EILSEQ));
1436 	}
1437 
1438 	if (vap->va_mask & AT_XVATTR) {
1439 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1440 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1441 			ZFS_EXIT(zfsvfs);
1442 			return (error);
1443 		}
1444 	}
1445 top:
1446 	*vpp = NULL;
1447 
1448 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1449 		vap->va_mode &= ~VSVTX;
1450 
1451 	if (*name == '\0') {
1452 		/*
1453 		 * Null component name refers to the directory itself.
1454 		 */
1455 		VN_HOLD(dvp);
1456 		zp = dzp;
1457 		dl = NULL;
1458 		error = 0;
1459 	} else {
1460 		/* possible VN_HOLD(zp) */
1461 		int zflg = 0;
1462 
1463 		if (flag & FIGNORECASE)
1464 			zflg |= ZCILOOK;
1465 
1466 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1467 		    NULL, NULL);
1468 		if (error) {
1469 			if (have_acl)
1470 				zfs_acl_ids_free(&acl_ids);
1471 			if (strcmp(name, "..") == 0)
1472 				error = SET_ERROR(EISDIR);
1473 			ZFS_EXIT(zfsvfs);
1474 			return (error);
1475 		}
1476 	}
1477 
1478 	if (zp == NULL) {
1479 		uint64_t txtype;
1480 
1481 		/*
1482 		 * Create a new file object and update the directory
1483 		 * to reference it.
1484 		 */
1485 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1486 			if (have_acl)
1487 				zfs_acl_ids_free(&acl_ids);
1488 			goto out;
1489 		}
1490 
1491 		/*
1492 		 * We only support the creation of regular files in
1493 		 * extended attribute directories.
1494 		 */
1495 
1496 		if ((dzp->z_pflags & ZFS_XATTR) &&
1497 		    (vap->va_type != VREG)) {
1498 			if (have_acl)
1499 				zfs_acl_ids_free(&acl_ids);
1500 			error = SET_ERROR(EINVAL);
1501 			goto out;
1502 		}
1503 
1504 		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1505 		    cr, vsecp, &acl_ids)) != 0)
1506 			goto out;
1507 		have_acl = B_TRUE;
1508 
1509 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1510 			zfs_acl_ids_free(&acl_ids);
1511 			error = SET_ERROR(EDQUOT);
1512 			goto out;
1513 		}
1514 
1515 		tx = dmu_tx_create(os);
1516 
1517 		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1518 		    ZFS_SA_BASE_ATTR_SIZE);
1519 
1520 		fuid_dirtied = zfsvfs->z_fuid_dirty;
1521 		if (fuid_dirtied)
1522 			zfs_fuid_txhold(zfsvfs, tx);
1523 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1524 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1525 		if (!zfsvfs->z_use_sa &&
1526 		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1527 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1528 			    0, acl_ids.z_aclp->z_acl_bytes);
1529 		}
1530 		error = dmu_tx_assign(tx,
1531 		    (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1532 		if (error) {
1533 			zfs_dirent_unlock(dl);
1534 			if (error == ERESTART) {
1535 				waited = B_TRUE;
1536 				dmu_tx_wait(tx);
1537 				dmu_tx_abort(tx);
1538 				goto top;
1539 			}
1540 			zfs_acl_ids_free(&acl_ids);
1541 			dmu_tx_abort(tx);
1542 			ZFS_EXIT(zfsvfs);
1543 			return (error);
1544 		}
1545 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1546 
1547 		if (fuid_dirtied)
1548 			zfs_fuid_sync(zfsvfs, tx);
1549 
1550 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1551 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1552 		if (flag & FIGNORECASE)
1553 			txtype |= TX_CI;
1554 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1555 		    vsecp, acl_ids.z_fuidp, vap);
1556 		zfs_acl_ids_free(&acl_ids);
1557 		dmu_tx_commit(tx);
1558 	} else {
1559 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1560 
1561 		if (have_acl)
1562 			zfs_acl_ids_free(&acl_ids);
1563 		have_acl = B_FALSE;
1564 
1565 		/*
1566 		 * A directory entry already exists for this name.
1567 		 */
1568 		/*
1569 		 * Can't truncate an existing file if in exclusive mode.
1570 		 */
1571 		if (excl == EXCL) {
1572 			error = SET_ERROR(EEXIST);
1573 			goto out;
1574 		}
1575 		/*
1576 		 * Can't open a directory for writing.
1577 		 */
1578 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1579 			error = SET_ERROR(EISDIR);
1580 			goto out;
1581 		}
1582 		/*
1583 		 * Verify requested access to file.
1584 		 */
1585 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1586 			goto out;
1587 		}
1588 
1589 		mutex_enter(&dzp->z_lock);
1590 		dzp->z_seq++;
1591 		mutex_exit(&dzp->z_lock);
1592 
1593 		/*
1594 		 * Truncate regular files if requested.
1595 		 */
1596 		if ((ZTOV(zp)->v_type == VREG) &&
1597 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1598 			/* we can't hold any locks when calling zfs_freesp() */
1599 			zfs_dirent_unlock(dl);
1600 			dl = NULL;
1601 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1602 			if (error == 0) {
1603 				vnevent_create(ZTOV(zp), ct);
1604 			}
1605 		}
1606 	}
1607 out:
1608 
1609 	if (dl)
1610 		zfs_dirent_unlock(dl);
1611 
1612 	if (error) {
1613 		if (zp)
1614 			VN_RELE(ZTOV(zp));
1615 	} else {
1616 		*vpp = ZTOV(zp);
1617 		error = specvp_check(vpp, cr);
1618 	}
1619 
1620 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1621 		zil_commit(zilog, 0);
1622 
1623 	ZFS_EXIT(zfsvfs);
1624 	return (error);
1625 }
1626 
1627 /*
1628  * Remove an entry from a directory.
1629  *
1630  *	IN:	dvp	- vnode of directory to remove entry from.
1631  *		name	- name of entry to remove.
1632  *		cr	- credentials of caller.
1633  *		ct	- caller context
1634  *		flags	- case flags
1635  *
1636  *	RETURN:	0 on success, error code on failure.
1637  *
1638  * Timestamps:
1639  *	dvp - ctime|mtime
1640  *	 vp - ctime (if nlink > 0)
1641  */
1642 
1643 uint64_t null_xattr = 0;
1644 
1645 /*ARGSUSED*/
1646 static int
1647 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1648     int flags)
1649 {
1650 	znode_t		*zp, *dzp = VTOZ(dvp);
1651 	znode_t		*xzp;
1652 	vnode_t		*vp;
1653 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1654 	zilog_t		*zilog;
1655 	uint64_t	acl_obj, xattr_obj;
1656 	uint64_t	xattr_obj_unlinked = 0;
1657 	uint64_t	obj = 0;
1658 	zfs_dirlock_t	*dl;
1659 	dmu_tx_t	*tx;
1660 	boolean_t	may_delete_now, delete_now = FALSE;
1661 	boolean_t	unlinked, toobig = FALSE;
1662 	uint64_t	txtype;
1663 	pathname_t	*realnmp = NULL;
1664 	pathname_t	realnm;
1665 	int		error;
1666 	int		zflg = ZEXISTS;
1667 	boolean_t	waited = B_FALSE;
1668 
1669 	ZFS_ENTER(zfsvfs);
1670 	ZFS_VERIFY_ZP(dzp);
1671 	zilog = zfsvfs->z_log;
1672 
1673 	if (flags & FIGNORECASE) {
1674 		zflg |= ZCILOOK;
1675 		pn_alloc(&realnm);
1676 		realnmp = &realnm;
1677 	}
1678 
1679 top:
1680 	xattr_obj = 0;
1681 	xzp = NULL;
1682 	/*
1683 	 * Attempt to lock directory; fail if entry doesn't exist.
1684 	 */
1685 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1686 	    NULL, realnmp)) {
1687 		if (realnmp)
1688 			pn_free(realnmp);
1689 		ZFS_EXIT(zfsvfs);
1690 		return (error);
1691 	}
1692 
1693 	vp = ZTOV(zp);
1694 
1695 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1696 		goto out;
1697 	}
1698 
1699 	/*
1700 	 * Need to use rmdir for removing directories.
1701 	 */
1702 	if (vp->v_type == VDIR) {
1703 		error = SET_ERROR(EPERM);
1704 		goto out;
1705 	}
1706 
1707 	vnevent_remove(vp, dvp, name, ct);
1708 
1709 	if (realnmp)
1710 		dnlc_remove(dvp, realnmp->pn_buf);
1711 	else
1712 		dnlc_remove(dvp, name);
1713 
1714 	mutex_enter(&vp->v_lock);
1715 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1716 	mutex_exit(&vp->v_lock);
1717 
1718 	/*
1719 	 * We may delete the znode now, or we may put it in the unlinked set;
1720 	 * it depends on whether we're the last link, and on whether there are
1721 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1722 	 * allow for either case.
1723 	 */
1724 	obj = zp->z_id;
1725 	tx = dmu_tx_create(zfsvfs->z_os);
1726 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1727 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1728 	zfs_sa_upgrade_txholds(tx, zp);
1729 	zfs_sa_upgrade_txholds(tx, dzp);
1730 	if (may_delete_now) {
1731 		toobig =
1732 		    zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1733 		/* if the file is too big, only hold_free a token amount */
1734 		dmu_tx_hold_free(tx, zp->z_id, 0,
1735 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1736 	}
1737 
1738 	/* are there any extended attributes? */
1739 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1740 	    &xattr_obj, sizeof (xattr_obj));
1741 	if (error == 0 && xattr_obj) {
1742 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1743 		ASSERT0(error);
1744 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1745 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1746 	}
1747 
1748 	mutex_enter(&zp->z_lock);
1749 	if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1750 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1751 	mutex_exit(&zp->z_lock);
1752 
1753 	/* charge as an update -- would be nice not to charge at all */
1754 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1755 
1756 	/*
1757 	 * Mark this transaction as typically resulting in a net free of space
1758 	 */
1759 	dmu_tx_mark_netfree(tx);
1760 
1761 	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1762 	if (error) {
1763 		zfs_dirent_unlock(dl);
1764 		VN_RELE(vp);
1765 		if (xzp)
1766 			VN_RELE(ZTOV(xzp));
1767 		if (error == ERESTART) {
1768 			waited = B_TRUE;
1769 			dmu_tx_wait(tx);
1770 			dmu_tx_abort(tx);
1771 			goto top;
1772 		}
1773 		if (realnmp)
1774 			pn_free(realnmp);
1775 		dmu_tx_abort(tx);
1776 		ZFS_EXIT(zfsvfs);
1777 		return (error);
1778 	}
1779 
1780 	/*
1781 	 * Remove the directory entry.
1782 	 */
1783 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1784 
1785 	if (error) {
1786 		dmu_tx_commit(tx);
1787 		goto out;
1788 	}
1789 
1790 	if (unlinked) {
1791 		/*
1792 		 * Hold z_lock so that we can make sure that the ACL obj
1793 		 * hasn't changed.  Could have been deleted due to
1794 		 * zfs_sa_upgrade().
1795 		 */
1796 		mutex_enter(&zp->z_lock);
1797 		mutex_enter(&vp->v_lock);
1798 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1799 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1800 		delete_now = may_delete_now && !toobig &&
1801 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1802 		    xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1803 		    acl_obj;
1804 		mutex_exit(&vp->v_lock);
1805 	}
1806 
1807 	if (delete_now) {
1808 		if (xattr_obj_unlinked) {
1809 			ASSERT3U(xzp->z_links, ==, 2);
1810 			mutex_enter(&xzp->z_lock);
1811 			xzp->z_unlinked = 1;
1812 			xzp->z_links = 0;
1813 			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1814 			    &xzp->z_links, sizeof (xzp->z_links), tx);
1815 			ASSERT3U(error,  ==,  0);
1816 			mutex_exit(&xzp->z_lock);
1817 			zfs_unlinked_add(xzp, tx);
1818 
1819 			if (zp->z_is_sa)
1820 				error = sa_remove(zp->z_sa_hdl,
1821 				    SA_ZPL_XATTR(zfsvfs), tx);
1822 			else
1823 				error = sa_update(zp->z_sa_hdl,
1824 				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
1825 				    sizeof (uint64_t), tx);
1826 			ASSERT0(error);
1827 		}
1828 		mutex_enter(&vp->v_lock);
1829 		VN_RELE_LOCKED(vp);
1830 		ASSERT0(vp->v_count);
1831 		mutex_exit(&vp->v_lock);
1832 		mutex_exit(&zp->z_lock);
1833 		zfs_znode_delete(zp, tx);
1834 	} else if (unlinked) {
1835 		mutex_exit(&zp->z_lock);
1836 		zfs_unlinked_add(zp, tx);
1837 	}
1838 
1839 	txtype = TX_REMOVE;
1840 	if (flags & FIGNORECASE)
1841 		txtype |= TX_CI;
1842 	zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1843 
1844 	dmu_tx_commit(tx);
1845 out:
1846 	if (realnmp)
1847 		pn_free(realnmp);
1848 
1849 	zfs_dirent_unlock(dl);
1850 
1851 	if (!delete_now)
1852 		VN_RELE(vp);
1853 	if (xzp)
1854 		VN_RELE(ZTOV(xzp));
1855 
1856 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1857 		zil_commit(zilog, 0);
1858 
1859 	ZFS_EXIT(zfsvfs);
1860 	return (error);
1861 }
1862 
1863 /*
1864  * Create a new directory and insert it into dvp using the name
1865  * provided.  Return a pointer to the inserted directory.
1866  *
1867  *	IN:	dvp	- vnode of directory to add subdir to.
1868  *		dirname	- name of new directory.
1869  *		vap	- attributes of new directory.
1870  *		cr	- credentials of caller.
1871  *		ct	- caller context
1872  *		flags	- case flags
1873  *		vsecp	- ACL to be set
1874  *
1875  *	OUT:	vpp	- vnode of created directory.
1876  *
1877  *	RETURN:	0 on success, error code on failure.
1878  *
1879  * Timestamps:
1880  *	dvp - ctime|mtime updated
1881  *	 vp - ctime|mtime|atime updated
1882  */
1883 /*ARGSUSED*/
1884 static int
1885 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1886     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1887 {
1888 	znode_t		*zp, *dzp = VTOZ(dvp);
1889 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1890 	zilog_t		*zilog;
1891 	zfs_dirlock_t	*dl;
1892 	uint64_t	txtype;
1893 	dmu_tx_t	*tx;
1894 	int		error;
1895 	int		zf = ZNEW;
1896 	ksid_t		*ksid;
1897 	uid_t		uid;
1898 	gid_t		gid = crgetgid(cr);
1899 	zfs_acl_ids_t   acl_ids;
1900 	boolean_t	fuid_dirtied;
1901 	boolean_t	waited = B_FALSE;
1902 
1903 	ASSERT(vap->va_type == VDIR);
1904 
1905 	/*
1906 	 * If we have an ephemeral id, ACL, or XVATTR then
1907 	 * make sure file system is at proper version
1908 	 */
1909 
1910 	ksid = crgetsid(cr, KSID_OWNER);
1911 	if (ksid)
1912 		uid = ksid_getid(ksid);
1913 	else
1914 		uid = crgetuid(cr);
1915 	if (zfsvfs->z_use_fuids == B_FALSE &&
1916 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1917 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1918 		return (SET_ERROR(EINVAL));
1919 
1920 	ZFS_ENTER(zfsvfs);
1921 	ZFS_VERIFY_ZP(dzp);
1922 	zilog = zfsvfs->z_log;
1923 
1924 	if (dzp->z_pflags & ZFS_XATTR) {
1925 		ZFS_EXIT(zfsvfs);
1926 		return (SET_ERROR(EINVAL));
1927 	}
1928 
1929 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1930 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1931 		ZFS_EXIT(zfsvfs);
1932 		return (SET_ERROR(EILSEQ));
1933 	}
1934 	if (flags & FIGNORECASE)
1935 		zf |= ZCILOOK;
1936 
1937 	if (vap->va_mask & AT_XVATTR) {
1938 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1939 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1940 			ZFS_EXIT(zfsvfs);
1941 			return (error);
1942 		}
1943 	}
1944 
1945 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1946 	    vsecp, &acl_ids)) != 0) {
1947 		ZFS_EXIT(zfsvfs);
1948 		return (error);
1949 	}
1950 	/*
1951 	 * First make sure the new directory doesn't exist.
1952 	 *
1953 	 * Existence is checked first to make sure we don't return
1954 	 * EACCES instead of EEXIST which can cause some applications
1955 	 * to fail.
1956 	 */
1957 top:
1958 	*vpp = NULL;
1959 
1960 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1961 	    NULL, NULL)) {
1962 		zfs_acl_ids_free(&acl_ids);
1963 		ZFS_EXIT(zfsvfs);
1964 		return (error);
1965 	}
1966 
1967 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1968 		zfs_acl_ids_free(&acl_ids);
1969 		zfs_dirent_unlock(dl);
1970 		ZFS_EXIT(zfsvfs);
1971 		return (error);
1972 	}
1973 
1974 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1975 		zfs_acl_ids_free(&acl_ids);
1976 		zfs_dirent_unlock(dl);
1977 		ZFS_EXIT(zfsvfs);
1978 		return (SET_ERROR(EDQUOT));
1979 	}
1980 
1981 	/*
1982 	 * Add a new entry to the directory.
1983 	 */
1984 	tx = dmu_tx_create(zfsvfs->z_os);
1985 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1986 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1987 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1988 	if (fuid_dirtied)
1989 		zfs_fuid_txhold(zfsvfs, tx);
1990 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1991 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1992 		    acl_ids.z_aclp->z_acl_bytes);
1993 	}
1994 
1995 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1996 	    ZFS_SA_BASE_ATTR_SIZE);
1997 
1998 	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
1999 	if (error) {
2000 		zfs_dirent_unlock(dl);
2001 		if (error == ERESTART) {
2002 			waited = B_TRUE;
2003 			dmu_tx_wait(tx);
2004 			dmu_tx_abort(tx);
2005 			goto top;
2006 		}
2007 		zfs_acl_ids_free(&acl_ids);
2008 		dmu_tx_abort(tx);
2009 		ZFS_EXIT(zfsvfs);
2010 		return (error);
2011 	}
2012 
2013 	/*
2014 	 * Create new node.
2015 	 */
2016 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2017 
2018 	if (fuid_dirtied)
2019 		zfs_fuid_sync(zfsvfs, tx);
2020 
2021 	/*
2022 	 * Now put new name in parent dir.
2023 	 */
2024 	(void) zfs_link_create(dl, zp, tx, ZNEW);
2025 
2026 	*vpp = ZTOV(zp);
2027 
2028 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2029 	if (flags & FIGNORECASE)
2030 		txtype |= TX_CI;
2031 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2032 	    acl_ids.z_fuidp, vap);
2033 
2034 	zfs_acl_ids_free(&acl_ids);
2035 
2036 	dmu_tx_commit(tx);
2037 
2038 	zfs_dirent_unlock(dl);
2039 
2040 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2041 		zil_commit(zilog, 0);
2042 
2043 	ZFS_EXIT(zfsvfs);
2044 	return (0);
2045 }
2046 
2047 /*
2048  * Remove a directory subdir entry.  If the current working
2049  * directory is the same as the subdir to be removed, the
2050  * remove will fail.
2051  *
2052  *	IN:	dvp	- vnode of directory to remove from.
2053  *		name	- name of directory to be removed.
2054  *		cwd	- vnode of current working directory.
2055  *		cr	- credentials of caller.
2056  *		ct	- caller context
2057  *		flags	- case flags
2058  *
2059  *	RETURN:	0 on success, error code on failure.
2060  *
2061  * Timestamps:
2062  *	dvp - ctime|mtime updated
2063  */
2064 /*ARGSUSED*/
2065 static int
2066 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2067     caller_context_t *ct, int flags)
2068 {
2069 	znode_t		*dzp = VTOZ(dvp);
2070 	znode_t		*zp;
2071 	vnode_t		*vp;
2072 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2073 	zilog_t		*zilog;
2074 	zfs_dirlock_t	*dl;
2075 	dmu_tx_t	*tx;
2076 	int		error;
2077 	int		zflg = ZEXISTS;
2078 	boolean_t	waited = B_FALSE;
2079 
2080 	ZFS_ENTER(zfsvfs);
2081 	ZFS_VERIFY_ZP(dzp);
2082 	zilog = zfsvfs->z_log;
2083 
2084 	if (flags & FIGNORECASE)
2085 		zflg |= ZCILOOK;
2086 top:
2087 	zp = NULL;
2088 
2089 	/*
2090 	 * Attempt to lock directory; fail if entry doesn't exist.
2091 	 */
2092 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2093 	    NULL, NULL)) {
2094 		ZFS_EXIT(zfsvfs);
2095 		return (error);
2096 	}
2097 
2098 	vp = ZTOV(zp);
2099 
2100 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2101 		goto out;
2102 	}
2103 
2104 	if (vp->v_type != VDIR) {
2105 		error = SET_ERROR(ENOTDIR);
2106 		goto out;
2107 	}
2108 
2109 	if (vp == cwd) {
2110 		error = SET_ERROR(EINVAL);
2111 		goto out;
2112 	}
2113 
2114 	vnevent_rmdir(vp, dvp, name, ct);
2115 
2116 	/*
2117 	 * Grab a lock on the directory to make sure that noone is
2118 	 * trying to add (or lookup) entries while we are removing it.
2119 	 */
2120 	rw_enter(&zp->z_name_lock, RW_WRITER);
2121 
2122 	/*
2123 	 * Grab a lock on the parent pointer to make sure we play well
2124 	 * with the treewalk and directory rename code.
2125 	 */
2126 	rw_enter(&zp->z_parent_lock, RW_WRITER);
2127 
2128 	tx = dmu_tx_create(zfsvfs->z_os);
2129 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2130 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2131 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2132 	zfs_sa_upgrade_txholds(tx, zp);
2133 	zfs_sa_upgrade_txholds(tx, dzp);
2134 	dmu_tx_mark_netfree(tx);
2135 	error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
2136 	if (error) {
2137 		rw_exit(&zp->z_parent_lock);
2138 		rw_exit(&zp->z_name_lock);
2139 		zfs_dirent_unlock(dl);
2140 		VN_RELE(vp);
2141 		if (error == ERESTART) {
2142 			waited = B_TRUE;
2143 			dmu_tx_wait(tx);
2144 			dmu_tx_abort(tx);
2145 			goto top;
2146 		}
2147 		dmu_tx_abort(tx);
2148 		ZFS_EXIT(zfsvfs);
2149 		return (error);
2150 	}
2151 
2152 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2153 
2154 	if (error == 0) {
2155 		uint64_t txtype = TX_RMDIR;
2156 		if (flags & FIGNORECASE)
2157 			txtype |= TX_CI;
2158 		zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2159 	}
2160 
2161 	dmu_tx_commit(tx);
2162 
2163 	rw_exit(&zp->z_parent_lock);
2164 	rw_exit(&zp->z_name_lock);
2165 out:
2166 	zfs_dirent_unlock(dl);
2167 
2168 	VN_RELE(vp);
2169 
2170 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2171 		zil_commit(zilog, 0);
2172 
2173 	ZFS_EXIT(zfsvfs);
2174 	return (error);
2175 }
2176 
2177 /*
2178  * Read as many directory entries as will fit into the provided
2179  * buffer from the given directory cursor position (specified in
2180  * the uio structure).
2181  *
2182  *	IN:	vp	- vnode of directory to read.
2183  *		uio	- structure supplying read location, range info,
2184  *			  and return buffer.
2185  *		cr	- credentials of caller.
2186  *		ct	- caller context
2187  *		flags	- case flags
2188  *
2189  *	OUT:	uio	- updated offset and range, buffer filled.
2190  *		eofp	- set to true if end-of-file detected.
2191  *
2192  *	RETURN:	0 on success, error code on failure.
2193  *
2194  * Timestamps:
2195  *	vp - atime updated
2196  *
2197  * Note that the low 4 bits of the cookie returned by zap is always zero.
2198  * This allows us to use the low range for "special" directory entries:
2199  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2200  * we use the offset 2 for the '.zfs' directory.
2201  */
2202 /* ARGSUSED */
2203 static int
2204 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2205     caller_context_t *ct, int flags)
2206 {
2207 	znode_t		*zp = VTOZ(vp);
2208 	iovec_t		*iovp;
2209 	edirent_t	*eodp;
2210 	dirent64_t	*odp;
2211 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2212 	objset_t	*os;
2213 	caddr_t		outbuf;
2214 	size_t		bufsize;
2215 	zap_cursor_t	zc;
2216 	zap_attribute_t	zap;
2217 	uint_t		bytes_wanted;
2218 	uint64_t	offset; /* must be unsigned; checks for < 1 */
2219 	uint64_t	parent;
2220 	int		local_eof;
2221 	int		outcount;
2222 	int		error;
2223 	uint8_t		prefetch;
2224 	boolean_t	check_sysattrs;
2225 
2226 	ZFS_ENTER(zfsvfs);
2227 	ZFS_VERIFY_ZP(zp);
2228 
2229 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2230 	    &parent, sizeof (parent))) != 0) {
2231 		ZFS_EXIT(zfsvfs);
2232 		return (error);
2233 	}
2234 
2235 	/*
2236 	 * If we are not given an eof variable,
2237 	 * use a local one.
2238 	 */
2239 	if (eofp == NULL)
2240 		eofp = &local_eof;
2241 
2242 	/*
2243 	 * Check for valid iov_len.
2244 	 */
2245 	if (uio->uio_iov->iov_len <= 0) {
2246 		ZFS_EXIT(zfsvfs);
2247 		return (SET_ERROR(EINVAL));
2248 	}
2249 
2250 	/*
2251 	 * Quit if directory has been removed (posix)
2252 	 */
2253 	if ((*eofp = zp->z_unlinked) != 0) {
2254 		ZFS_EXIT(zfsvfs);
2255 		return (0);
2256 	}
2257 
2258 	error = 0;
2259 	os = zfsvfs->z_os;
2260 	offset = uio->uio_loffset;
2261 	prefetch = zp->z_zn_prefetch;
2262 
2263 	/*
2264 	 * Initialize the iterator cursor.
2265 	 */
2266 	if (offset <= 3) {
2267 		/*
2268 		 * Start iteration from the beginning of the directory.
2269 		 */
2270 		zap_cursor_init(&zc, os, zp->z_id);
2271 	} else {
2272 		/*
2273 		 * The offset is a serialized cursor.
2274 		 */
2275 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2276 	}
2277 
2278 	/*
2279 	 * Get space to change directory entries into fs independent format.
2280 	 */
2281 	iovp = uio->uio_iov;
2282 	bytes_wanted = iovp->iov_len;
2283 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2284 		bufsize = bytes_wanted;
2285 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
2286 		odp = (struct dirent64 *)outbuf;
2287 	} else {
2288 		bufsize = bytes_wanted;
2289 		outbuf = NULL;
2290 		odp = (struct dirent64 *)iovp->iov_base;
2291 	}
2292 	eodp = (struct edirent *)odp;
2293 
2294 	/*
2295 	 * If this VFS supports the system attribute view interface; and
2296 	 * we're looking at an extended attribute directory; and we care
2297 	 * about normalization conflicts on this vfs; then we must check
2298 	 * for normalization conflicts with the sysattr name space.
2299 	 */
2300 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2301 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2302 	    (flags & V_RDDIR_ENTFLAGS);
2303 
2304 	/*
2305 	 * Transform to file-system independent format
2306 	 */
2307 	outcount = 0;
2308 	while (outcount < bytes_wanted) {
2309 		ino64_t objnum;
2310 		ushort_t reclen;
2311 		off64_t *next = NULL;
2312 
2313 		/*
2314 		 * Special case `.', `..', and `.zfs'.
2315 		 */
2316 		if (offset == 0) {
2317 			(void) strcpy(zap.za_name, ".");
2318 			zap.za_normalization_conflict = 0;
2319 			objnum = zp->z_id;
2320 		} else if (offset == 1) {
2321 			(void) strcpy(zap.za_name, "..");
2322 			zap.za_normalization_conflict = 0;
2323 			objnum = parent;
2324 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2325 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2326 			zap.za_normalization_conflict = 0;
2327 			objnum = ZFSCTL_INO_ROOT;
2328 		} else {
2329 			/*
2330 			 * Grab next entry.
2331 			 */
2332 			if (error = zap_cursor_retrieve(&zc, &zap)) {
2333 				if ((*eofp = (error == ENOENT)) != 0)
2334 					break;
2335 				else
2336 					goto update;
2337 			}
2338 
2339 			if (zap.za_integer_length != 8 ||
2340 			    zap.za_num_integers != 1) {
2341 				cmn_err(CE_WARN, "zap_readdir: bad directory "
2342 				    "entry, obj = %lld, offset = %lld\n",
2343 				    (u_longlong_t)zp->z_id,
2344 				    (u_longlong_t)offset);
2345 				error = SET_ERROR(ENXIO);
2346 				goto update;
2347 			}
2348 
2349 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2350 			/*
2351 			 * MacOS X can extract the object type here such as:
2352 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2353 			 */
2354 
2355 			if (check_sysattrs && !zap.za_normalization_conflict) {
2356 				zap.za_normalization_conflict =
2357 				    xattr_sysattr_casechk(zap.za_name);
2358 			}
2359 		}
2360 
2361 		if (flags & V_RDDIR_ACCFILTER) {
2362 			/*
2363 			 * If we have no access at all, don't include
2364 			 * this entry in the returned information
2365 			 */
2366 			znode_t	*ezp;
2367 			if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2368 				goto skip_entry;
2369 			if (!zfs_has_access(ezp, cr)) {
2370 				VN_RELE(ZTOV(ezp));
2371 				goto skip_entry;
2372 			}
2373 			VN_RELE(ZTOV(ezp));
2374 		}
2375 
2376 		if (flags & V_RDDIR_ENTFLAGS)
2377 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2378 		else
2379 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2380 
2381 		/*
2382 		 * Will this entry fit in the buffer?
2383 		 */
2384 		if (outcount + reclen > bufsize) {
2385 			/*
2386 			 * Did we manage to fit anything in the buffer?
2387 			 */
2388 			if (!outcount) {
2389 				error = SET_ERROR(EINVAL);
2390 				goto update;
2391 			}
2392 			break;
2393 		}
2394 		if (flags & V_RDDIR_ENTFLAGS) {
2395 			/*
2396 			 * Add extended flag entry:
2397 			 */
2398 			eodp->ed_ino = objnum;
2399 			eodp->ed_reclen = reclen;
2400 			/* NOTE: ed_off is the offset for the *next* entry */
2401 			next = &(eodp->ed_off);
2402 			eodp->ed_eflags = zap.za_normalization_conflict ?
2403 			    ED_CASE_CONFLICT : 0;
2404 			(void) strncpy(eodp->ed_name, zap.za_name,
2405 			    EDIRENT_NAMELEN(reclen));
2406 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2407 		} else {
2408 			/*
2409 			 * Add normal entry:
2410 			 */
2411 			odp->d_ino = objnum;
2412 			odp->d_reclen = reclen;
2413 			/* NOTE: d_off is the offset for the *next* entry */
2414 			next = &(odp->d_off);
2415 			(void) strncpy(odp->d_name, zap.za_name,
2416 			    DIRENT64_NAMELEN(reclen));
2417 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2418 		}
2419 		outcount += reclen;
2420 
2421 		ASSERT(outcount <= bufsize);
2422 
2423 		/* Prefetch znode */
2424 		if (prefetch)
2425 			dmu_prefetch(os, objnum, 0, 0, 0,
2426 			    ZIO_PRIORITY_SYNC_READ);
2427 
2428 	skip_entry:
2429 		/*
2430 		 * Move to the next entry, fill in the previous offset.
2431 		 */
2432 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2433 			zap_cursor_advance(&zc);
2434 			offset = zap_cursor_serialize(&zc);
2435 		} else {
2436 			offset += 1;
2437 		}
2438 		if (next)
2439 			*next = offset;
2440 	}
2441 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2442 
2443 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2444 		iovp->iov_base += outcount;
2445 		iovp->iov_len -= outcount;
2446 		uio->uio_resid -= outcount;
2447 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2448 		/*
2449 		 * Reset the pointer.
2450 		 */
2451 		offset = uio->uio_loffset;
2452 	}
2453 
2454 update:
2455 	zap_cursor_fini(&zc);
2456 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2457 		kmem_free(outbuf, bufsize);
2458 
2459 	if (error == ENOENT)
2460 		error = 0;
2461 
2462 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2463 
2464 	uio->uio_loffset = offset;
2465 	ZFS_EXIT(zfsvfs);
2466 	return (error);
2467 }
2468 
2469 ulong_t zfs_fsync_sync_cnt = 4;
2470 
2471 static int
2472 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2473 {
2474 	znode_t	*zp = VTOZ(vp);
2475 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2476 
2477 	/*
2478 	 * Regardless of whether this is required for standards conformance,
2479 	 * this is the logical behavior when fsync() is called on a file with
2480 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2481 	 * going to be pushed out as part of the zil_commit().
2482 	 */
2483 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2484 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2485 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2486 
2487 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2488 
2489 	if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2490 		ZFS_ENTER(zfsvfs);
2491 		ZFS_VERIFY_ZP(zp);
2492 		zil_commit(zfsvfs->z_log, zp->z_id);
2493 		ZFS_EXIT(zfsvfs);
2494 	}
2495 	return (0);
2496 }
2497 
2498 
2499 /*
2500  * Get the requested file attributes and place them in the provided
2501  * vattr structure.
2502  *
2503  *	IN:	vp	- vnode of file.
2504  *		vap	- va_mask identifies requested attributes.
2505  *			  If AT_XVATTR set, then optional attrs are requested
2506  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2507  *		cr	- credentials of caller.
2508  *		ct	- caller context
2509  *
2510  *	OUT:	vap	- attribute values.
2511  *
2512  *	RETURN:	0 (always succeeds).
2513  */
2514 /* ARGSUSED */
2515 static int
2516 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2517     caller_context_t *ct)
2518 {
2519 	znode_t *zp = VTOZ(vp);
2520 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2521 	int	error = 0;
2522 	uint64_t links;
2523 	uint64_t mtime[2], ctime[2];
2524 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2525 	xoptattr_t *xoap = NULL;
2526 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2527 	sa_bulk_attr_t bulk[2];
2528 	int count = 0;
2529 
2530 	ZFS_ENTER(zfsvfs);
2531 	ZFS_VERIFY_ZP(zp);
2532 
2533 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2534 
2535 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2536 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2537 
2538 	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2539 		ZFS_EXIT(zfsvfs);
2540 		return (error);
2541 	}
2542 
2543 	/*
2544 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2545 	 * Also, if we are the owner don't bother, since owner should
2546 	 * always be allowed to read basic attributes of file.
2547 	 */
2548 	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2549 	    (vap->va_uid != crgetuid(cr))) {
2550 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2551 		    skipaclchk, cr)) {
2552 			ZFS_EXIT(zfsvfs);
2553 			return (error);
2554 		}
2555 	}
2556 
2557 	/*
2558 	 * Return all attributes.  It's cheaper to provide the answer
2559 	 * than to determine whether we were asked the question.
2560 	 */
2561 
2562 	mutex_enter(&zp->z_lock);
2563 	vap->va_type = vp->v_type;
2564 	vap->va_mode = zp->z_mode & MODEMASK;
2565 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2566 	vap->va_nodeid = zp->z_id;
2567 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2568 		links = zp->z_links + 1;
2569 	else
2570 		links = zp->z_links;
2571 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2572 	vap->va_size = zp->z_size;
2573 	vap->va_rdev = vp->v_rdev;
2574 	vap->va_seq = zp->z_seq;
2575 
2576 	/*
2577 	 * Add in any requested optional attributes and the create time.
2578 	 * Also set the corresponding bits in the returned attribute bitmap.
2579 	 */
2580 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2581 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2582 			xoap->xoa_archive =
2583 			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2584 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2585 		}
2586 
2587 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2588 			xoap->xoa_readonly =
2589 			    ((zp->z_pflags & ZFS_READONLY) != 0);
2590 			XVA_SET_RTN(xvap, XAT_READONLY);
2591 		}
2592 
2593 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2594 			xoap->xoa_system =
2595 			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2596 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2597 		}
2598 
2599 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2600 			xoap->xoa_hidden =
2601 			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2602 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2603 		}
2604 
2605 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2606 			xoap->xoa_nounlink =
2607 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2608 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2609 		}
2610 
2611 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2612 			xoap->xoa_immutable =
2613 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2614 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2615 		}
2616 
2617 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2618 			xoap->xoa_appendonly =
2619 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2620 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2621 		}
2622 
2623 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2624 			xoap->xoa_nodump =
2625 			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2626 			XVA_SET_RTN(xvap, XAT_NODUMP);
2627 		}
2628 
2629 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2630 			xoap->xoa_opaque =
2631 			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2632 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2633 		}
2634 
2635 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2636 			xoap->xoa_av_quarantined =
2637 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2638 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2639 		}
2640 
2641 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2642 			xoap->xoa_av_modified =
2643 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2644 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2645 		}
2646 
2647 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2648 		    vp->v_type == VREG) {
2649 			zfs_sa_get_scanstamp(zp, xvap);
2650 		}
2651 
2652 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2653 			uint64_t times[2];
2654 
2655 			(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2656 			    times, sizeof (times));
2657 			ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2658 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2659 		}
2660 
2661 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2662 			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2663 			XVA_SET_RTN(xvap, XAT_REPARSE);
2664 		}
2665 		if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2666 			xoap->xoa_generation = zp->z_gen;
2667 			XVA_SET_RTN(xvap, XAT_GEN);
2668 		}
2669 
2670 		if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2671 			xoap->xoa_offline =
2672 			    ((zp->z_pflags & ZFS_OFFLINE) != 0);
2673 			XVA_SET_RTN(xvap, XAT_OFFLINE);
2674 		}
2675 
2676 		if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2677 			xoap->xoa_sparse =
2678 			    ((zp->z_pflags & ZFS_SPARSE) != 0);
2679 			XVA_SET_RTN(xvap, XAT_SPARSE);
2680 		}
2681 	}
2682 
2683 	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2684 	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2685 	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2686 
2687 	mutex_exit(&zp->z_lock);
2688 
2689 	sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2690 
2691 	if (zp->z_blksz == 0) {
2692 		/*
2693 		 * Block size hasn't been set; suggest maximal I/O transfers.
2694 		 */
2695 		vap->va_blksize = zfsvfs->z_max_blksz;
2696 	}
2697 
2698 	ZFS_EXIT(zfsvfs);
2699 	return (0);
2700 }
2701 
2702 /*
2703  * Set the file attributes to the values contained in the
2704  * vattr structure.
2705  *
2706  *	IN:	vp	- vnode of file to be modified.
2707  *		vap	- new attribute values.
2708  *			  If AT_XVATTR set, then optional attrs are being set
2709  *		flags	- ATTR_UTIME set if non-default time values provided.
2710  *			- ATTR_NOACLCHECK (CIFS context only).
2711  *		cr	- credentials of caller.
2712  *		ct	- caller context
2713  *
2714  *	RETURN:	0 on success, error code on failure.
2715  *
2716  * Timestamps:
2717  *	vp - ctime updated, mtime updated if size changed.
2718  */
2719 /* ARGSUSED */
2720 static int
2721 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2722     caller_context_t *ct)
2723 {
2724 	znode_t		*zp = VTOZ(vp);
2725 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2726 	zilog_t		*zilog;
2727 	dmu_tx_t	*tx;
2728 	vattr_t		oldva;
2729 	xvattr_t	tmpxvattr;
2730 	uint_t		mask = vap->va_mask;
2731 	uint_t		saved_mask = 0;
2732 	int		trim_mask = 0;
2733 	uint64_t	new_mode;
2734 	uint64_t	new_uid, new_gid;
2735 	uint64_t	xattr_obj;
2736 	uint64_t	mtime[2], ctime[2];
2737 	znode_t		*attrzp;
2738 	int		need_policy = FALSE;
2739 	int		err, err2;
2740 	zfs_fuid_info_t *fuidp = NULL;
2741 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2742 	xoptattr_t	*xoap;
2743 	zfs_acl_t	*aclp;
2744 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2745 	boolean_t	fuid_dirtied = B_FALSE;
2746 	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
2747 	int		count = 0, xattr_count = 0;
2748 
2749 	if (mask == 0)
2750 		return (0);
2751 
2752 	if (mask & AT_NOSET)
2753 		return (SET_ERROR(EINVAL));
2754 
2755 	ZFS_ENTER(zfsvfs);
2756 	ZFS_VERIFY_ZP(zp);
2757 
2758 	zilog = zfsvfs->z_log;
2759 
2760 	/*
2761 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2762 	 * that file system is at proper version level
2763 	 */
2764 
2765 	if (zfsvfs->z_use_fuids == B_FALSE &&
2766 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2767 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2768 	    (mask & AT_XVATTR))) {
2769 		ZFS_EXIT(zfsvfs);
2770 		return (SET_ERROR(EINVAL));
2771 	}
2772 
2773 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2774 		ZFS_EXIT(zfsvfs);
2775 		return (SET_ERROR(EISDIR));
2776 	}
2777 
2778 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2779 		ZFS_EXIT(zfsvfs);
2780 		return (SET_ERROR(EINVAL));
2781 	}
2782 
2783 	/*
2784 	 * If this is an xvattr_t, then get a pointer to the structure of
2785 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2786 	 */
2787 	xoap = xva_getxoptattr(xvap);
2788 
2789 	xva_init(&tmpxvattr);
2790 
2791 	/*
2792 	 * Immutable files can only alter immutable bit and atime
2793 	 */
2794 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2795 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2796 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2797 		ZFS_EXIT(zfsvfs);
2798 		return (SET_ERROR(EPERM));
2799 	}
2800 
2801 	/*
2802 	 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
2803 	 */
2804 
2805 	/*
2806 	 * Verify timestamps doesn't overflow 32 bits.
2807 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2808 	 * handle times greater than 2039.  This check should be removed
2809 	 * once large timestamps are fully supported.
2810 	 */
2811 	if (mask & (AT_ATIME | AT_MTIME)) {
2812 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2813 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2814 			ZFS_EXIT(zfsvfs);
2815 			return (SET_ERROR(EOVERFLOW));
2816 		}
2817 	}
2818 
2819 top:
2820 	attrzp = NULL;
2821 	aclp = NULL;
2822 
2823 	/* Can this be moved to before the top label? */
2824 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2825 		ZFS_EXIT(zfsvfs);
2826 		return (SET_ERROR(EROFS));
2827 	}
2828 
2829 	/*
2830 	 * First validate permissions
2831 	 */
2832 
2833 	if (mask & AT_SIZE) {
2834 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2835 		if (err) {
2836 			ZFS_EXIT(zfsvfs);
2837 			return (err);
2838 		}
2839 		/*
2840 		 * XXX - Note, we are not providing any open
2841 		 * mode flags here (like FNDELAY), so we may
2842 		 * block if there are locks present... this
2843 		 * should be addressed in openat().
2844 		 */
2845 		/* XXX - would it be OK to generate a log record here? */
2846 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2847 		if (err) {
2848 			ZFS_EXIT(zfsvfs);
2849 			return (err);
2850 		}
2851 
2852 		if (vap->va_size == 0)
2853 			vnevent_truncate(ZTOV(zp), ct);
2854 	}
2855 
2856 	if (mask & (AT_ATIME|AT_MTIME) ||
2857 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2858 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2859 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2860 	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2861 	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2862 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2863 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2864 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2865 		    skipaclchk, cr);
2866 	}
2867 
2868 	if (mask & (AT_UID|AT_GID)) {
2869 		int	idmask = (mask & (AT_UID|AT_GID));
2870 		int	take_owner;
2871 		int	take_group;
2872 
2873 		/*
2874 		 * NOTE: even if a new mode is being set,
2875 		 * we may clear S_ISUID/S_ISGID bits.
2876 		 */
2877 
2878 		if (!(mask & AT_MODE))
2879 			vap->va_mode = zp->z_mode;
2880 
2881 		/*
2882 		 * Take ownership or chgrp to group we are a member of
2883 		 */
2884 
2885 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2886 		take_group = (mask & AT_GID) &&
2887 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2888 
2889 		/*
2890 		 * If both AT_UID and AT_GID are set then take_owner and
2891 		 * take_group must both be set in order to allow taking
2892 		 * ownership.
2893 		 *
2894 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2895 		 *
2896 		 */
2897 
2898 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2899 		    ((idmask == AT_UID) && take_owner) ||
2900 		    ((idmask == AT_GID) && take_group)) {
2901 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2902 			    skipaclchk, cr) == 0) {
2903 				/*
2904 				 * Remove setuid/setgid for non-privileged users
2905 				 */
2906 				secpolicy_setid_clear(vap, cr);
2907 				trim_mask = (mask & (AT_UID|AT_GID));
2908 			} else {
2909 				need_policy =  TRUE;
2910 			}
2911 		} else {
2912 			need_policy =  TRUE;
2913 		}
2914 	}
2915 
2916 	mutex_enter(&zp->z_lock);
2917 	oldva.va_mode = zp->z_mode;
2918 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2919 	if (mask & AT_XVATTR) {
2920 		/*
2921 		 * Update xvattr mask to include only those attributes
2922 		 * that are actually changing.
2923 		 *
2924 		 * the bits will be restored prior to actually setting
2925 		 * the attributes so the caller thinks they were set.
2926 		 */
2927 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2928 			if (xoap->xoa_appendonly !=
2929 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2930 				need_policy = TRUE;
2931 			} else {
2932 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2933 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2934 			}
2935 		}
2936 
2937 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2938 			if (xoap->xoa_nounlink !=
2939 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2940 				need_policy = TRUE;
2941 			} else {
2942 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2943 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2944 			}
2945 		}
2946 
2947 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2948 			if (xoap->xoa_immutable !=
2949 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2950 				need_policy = TRUE;
2951 			} else {
2952 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2953 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2954 			}
2955 		}
2956 
2957 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2958 			if (xoap->xoa_nodump !=
2959 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2960 				need_policy = TRUE;
2961 			} else {
2962 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2963 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2964 			}
2965 		}
2966 
2967 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2968 			if (xoap->xoa_av_modified !=
2969 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2970 				need_policy = TRUE;
2971 			} else {
2972 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2973 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2974 			}
2975 		}
2976 
2977 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2978 			if ((vp->v_type != VREG &&
2979 			    xoap->xoa_av_quarantined) ||
2980 			    xoap->xoa_av_quarantined !=
2981 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2982 				need_policy = TRUE;
2983 			} else {
2984 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2985 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2986 			}
2987 		}
2988 
2989 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2990 			mutex_exit(&zp->z_lock);
2991 			ZFS_EXIT(zfsvfs);
2992 			return (SET_ERROR(EPERM));
2993 		}
2994 
2995 		if (need_policy == FALSE &&
2996 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2997 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2998 			need_policy = TRUE;
2999 		}
3000 	}
3001 
3002 	mutex_exit(&zp->z_lock);
3003 
3004 	if (mask & AT_MODE) {
3005 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
3006 			err = secpolicy_setid_setsticky_clear(vp, vap,
3007 			    &oldva, cr);
3008 			if (err) {
3009 				ZFS_EXIT(zfsvfs);
3010 				return (err);
3011 			}
3012 			trim_mask |= AT_MODE;
3013 		} else {
3014 			need_policy = TRUE;
3015 		}
3016 	}
3017 
3018 	if (need_policy) {
3019 		/*
3020 		 * If trim_mask is set then take ownership
3021 		 * has been granted or write_acl is present and user
3022 		 * has the ability to modify mode.  In that case remove
3023 		 * UID|GID and or MODE from mask so that
3024 		 * secpolicy_vnode_setattr() doesn't revoke it.
3025 		 */
3026 
3027 		if (trim_mask) {
3028 			saved_mask = vap->va_mask;
3029 			vap->va_mask &= ~trim_mask;
3030 		}
3031 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3032 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3033 		if (err) {
3034 			ZFS_EXIT(zfsvfs);
3035 			return (err);
3036 		}
3037 
3038 		if (trim_mask)
3039 			vap->va_mask |= saved_mask;
3040 	}
3041 
3042 	/*
3043 	 * secpolicy_vnode_setattr, or take ownership may have
3044 	 * changed va_mask
3045 	 */
3046 	mask = vap->va_mask;
3047 
3048 	if ((mask & (AT_UID | AT_GID))) {
3049 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3050 		    &xattr_obj, sizeof (xattr_obj));
3051 
3052 		if (err == 0 && xattr_obj) {
3053 			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3054 			if (err)
3055 				goto out2;
3056 		}
3057 		if (mask & AT_UID) {
3058 			new_uid = zfs_fuid_create(zfsvfs,
3059 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3060 			if (new_uid != zp->z_uid &&
3061 			    zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3062 				if (attrzp)
3063 					VN_RELE(ZTOV(attrzp));
3064 				err = SET_ERROR(EDQUOT);
3065 				goto out2;
3066 			}
3067 		}
3068 
3069 		if (mask & AT_GID) {
3070 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3071 			    cr, ZFS_GROUP, &fuidp);
3072 			if (new_gid != zp->z_gid &&
3073 			    zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3074 				if (attrzp)
3075 					VN_RELE(ZTOV(attrzp));
3076 				err = SET_ERROR(EDQUOT);
3077 				goto out2;
3078 			}
3079 		}
3080 	}
3081 	tx = dmu_tx_create(zfsvfs->z_os);
3082 
3083 	if (mask & AT_MODE) {
3084 		uint64_t pmode = zp->z_mode;
3085 		uint64_t acl_obj;
3086 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3087 
3088 		if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3089 		    !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3090 			err = SET_ERROR(EPERM);
3091 			goto out;
3092 		}
3093 
3094 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3095 			goto out;
3096 
3097 		mutex_enter(&zp->z_lock);
3098 		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3099 			/*
3100 			 * Are we upgrading ACL from old V0 format
3101 			 * to V1 format?
3102 			 */
3103 			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3104 			    zfs_znode_acl_version(zp) ==
3105 			    ZFS_ACL_VERSION_INITIAL) {
3106 				dmu_tx_hold_free(tx, acl_obj, 0,
3107 				    DMU_OBJECT_END);
3108 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3109 				    0, aclp->z_acl_bytes);
3110 			} else {
3111 				dmu_tx_hold_write(tx, acl_obj, 0,
3112 				    aclp->z_acl_bytes);
3113 			}
3114 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3115 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3116 			    0, aclp->z_acl_bytes);
3117 		}
3118 		mutex_exit(&zp->z_lock);
3119 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3120 	} else {
3121 		if ((mask & AT_XVATTR) &&
3122 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3123 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3124 		else
3125 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3126 	}
3127 
3128 	if (attrzp) {
3129 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3130 	}
3131 
3132 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3133 	if (fuid_dirtied)
3134 		zfs_fuid_txhold(zfsvfs, tx);
3135 
3136 	zfs_sa_upgrade_txholds(tx, zp);
3137 
3138 	err = dmu_tx_assign(tx, TXG_WAIT);
3139 	if (err)
3140 		goto out;
3141 
3142 	count = 0;
3143 	/*
3144 	 * Set each attribute requested.
3145 	 * We group settings according to the locks they need to acquire.
3146 	 *
3147 	 * Note: you cannot set ctime directly, although it will be
3148 	 * updated as a side-effect of calling this function.
3149 	 */
3150 
3151 
3152 	if (mask & (AT_UID|AT_GID|AT_MODE))
3153 		mutex_enter(&zp->z_acl_lock);
3154 	mutex_enter(&zp->z_lock);
3155 
3156 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3157 	    &zp->z_pflags, sizeof (zp->z_pflags));
3158 
3159 	if (attrzp) {
3160 		if (mask & (AT_UID|AT_GID|AT_MODE))
3161 			mutex_enter(&attrzp->z_acl_lock);
3162 		mutex_enter(&attrzp->z_lock);
3163 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3164 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3165 		    sizeof (attrzp->z_pflags));
3166 	}
3167 
3168 	if (mask & (AT_UID|AT_GID)) {
3169 
3170 		if (mask & AT_UID) {
3171 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3172 			    &new_uid, sizeof (new_uid));
3173 			zp->z_uid = new_uid;
3174 			if (attrzp) {
3175 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3176 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3177 				    sizeof (new_uid));
3178 				attrzp->z_uid = new_uid;
3179 			}
3180 		}
3181 
3182 		if (mask & AT_GID) {
3183 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3184 			    NULL, &new_gid, sizeof (new_gid));
3185 			zp->z_gid = new_gid;
3186 			if (attrzp) {
3187 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3188 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3189 				    sizeof (new_gid));
3190 				attrzp->z_gid = new_gid;
3191 			}
3192 		}
3193 		if (!(mask & AT_MODE)) {
3194 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3195 			    NULL, &new_mode, sizeof (new_mode));
3196 			new_mode = zp->z_mode;
3197 		}
3198 		err = zfs_acl_chown_setattr(zp);
3199 		ASSERT(err == 0);
3200 		if (attrzp) {
3201 			err = zfs_acl_chown_setattr(attrzp);
3202 			ASSERT(err == 0);
3203 		}
3204 	}
3205 
3206 	if (mask & AT_MODE) {
3207 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3208 		    &new_mode, sizeof (new_mode));
3209 		zp->z_mode = new_mode;
3210 		ASSERT3U((uintptr_t)aclp, !=, NULL);
3211 		err = zfs_aclset_common(zp, aclp, cr, tx);
3212 		ASSERT0(err);
3213 		if (zp->z_acl_cached)
3214 			zfs_acl_free(zp->z_acl_cached);
3215 		zp->z_acl_cached = aclp;
3216 		aclp = NULL;
3217 	}
3218 
3219 
3220 	if (mask & AT_ATIME) {
3221 		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3222 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3223 		    &zp->z_atime, sizeof (zp->z_atime));
3224 	}
3225 
3226 	if (mask & AT_MTIME) {
3227 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3228 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3229 		    mtime, sizeof (mtime));
3230 	}
3231 
3232 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3233 	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3234 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3235 		    NULL, mtime, sizeof (mtime));
3236 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3237 		    &ctime, sizeof (ctime));
3238 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3239 		    B_TRUE);
3240 	} else if (mask != 0) {
3241 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3242 		    &ctime, sizeof (ctime));
3243 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3244 		    B_TRUE);
3245 		if (attrzp) {
3246 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3247 			    SA_ZPL_CTIME(zfsvfs), NULL,
3248 			    &ctime, sizeof (ctime));
3249 			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3250 			    mtime, ctime, B_TRUE);
3251 		}
3252 	}
3253 	/*
3254 	 * Do this after setting timestamps to prevent timestamp
3255 	 * update from toggling bit
3256 	 */
3257 
3258 	if (xoap && (mask & AT_XVATTR)) {
3259 
3260 		/*
3261 		 * restore trimmed off masks
3262 		 * so that return masks can be set for caller.
3263 		 */
3264 
3265 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3266 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
3267 		}
3268 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3269 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
3270 		}
3271 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3272 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3273 		}
3274 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3275 			XVA_SET_REQ(xvap, XAT_NODUMP);
3276 		}
3277 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3278 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3279 		}
3280 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3281 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3282 		}
3283 
3284 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3285 			ASSERT(vp->v_type == VREG);
3286 
3287 		zfs_xvattr_set(zp, xvap, tx);
3288 	}
3289 
3290 	if (fuid_dirtied)
3291 		zfs_fuid_sync(zfsvfs, tx);
3292 
3293 	if (mask != 0)
3294 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3295 
3296 	mutex_exit(&zp->z_lock);
3297 	if (mask & (AT_UID|AT_GID|AT_MODE))
3298 		mutex_exit(&zp->z_acl_lock);
3299 
3300 	if (attrzp) {
3301 		if (mask & (AT_UID|AT_GID|AT_MODE))
3302 			mutex_exit(&attrzp->z_acl_lock);
3303 		mutex_exit(&attrzp->z_lock);
3304 	}
3305 out:
3306 	if (err == 0 && attrzp) {
3307 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3308 		    xattr_count, tx);
3309 		ASSERT(err2 == 0);
3310 	}
3311 
3312 	if (attrzp)
3313 		VN_RELE(ZTOV(attrzp));
3314 
3315 	if (aclp)
3316 		zfs_acl_free(aclp);
3317 
3318 	if (fuidp) {
3319 		zfs_fuid_info_free(fuidp);
3320 		fuidp = NULL;
3321 	}
3322 
3323 	if (err) {
3324 		dmu_tx_abort(tx);
3325 		if (err == ERESTART)
3326 			goto top;
3327 	} else {
3328 		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3329 		dmu_tx_commit(tx);
3330 	}
3331 
3332 out2:
3333 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3334 		zil_commit(zilog, 0);
3335 
3336 	ZFS_EXIT(zfsvfs);
3337 	return (err);
3338 }
3339 
3340 typedef struct zfs_zlock {
3341 	krwlock_t	*zl_rwlock;	/* lock we acquired */
3342 	znode_t		*zl_znode;	/* znode we held */
3343 	struct zfs_zlock *zl_next;	/* next in list */
3344 } zfs_zlock_t;
3345 
3346 /*
3347  * Drop locks and release vnodes that were held by zfs_rename_lock().
3348  */
3349 static void
3350 zfs_rename_unlock(zfs_zlock_t **zlpp)
3351 {
3352 	zfs_zlock_t *zl;
3353 
3354 	while ((zl = *zlpp) != NULL) {
3355 		if (