xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 0c94e1af)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2014 Integros [integros.com]
26  * Copyright 2015 Joyent, Inc.
27  */
28 
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
31 
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/time.h>
35 #include <sys/systm.h>
36 #include <sys/sysmacros.h>
37 #include <sys/resource.h>
38 #include <sys/vfs.h>
39 #include <sys/vfs_opreg.h>
40 #include <sys/vnode.h>
41 #include <sys/file.h>
42 #include <sys/stat.h>
43 #include <sys/kmem.h>
44 #include <sys/taskq.h>
45 #include <sys/uio.h>
46 #include <sys/vmsystm.h>
47 #include <sys/atomic.h>
48 #include <sys/vm.h>
49 #include <vm/seg_vn.h>
50 #include <vm/pvn.h>
51 #include <vm/as.h>
52 #include <vm/kpm.h>
53 #include <vm/seg_kpm.h>
54 #include <sys/mman.h>
55 #include <sys/pathname.h>
56 #include <sys/cmn_err.h>
57 #include <sys/errno.h>
58 #include <sys/unistd.h>
59 #include <sys/zfs_dir.h>
60 #include <sys/zfs_acl.h>
61 #include <sys/zfs_ioctl.h>
62 #include <sys/fs/zfs.h>
63 #include <sys/dmu.h>
64 #include <sys/dmu_objset.h>
65 #include <sys/spa.h>
66 #include <sys/txg.h>
67 #include <sys/dbuf.h>
68 #include <sys/zap.h>
69 #include <sys/sa.h>
70 #include <sys/dirent.h>
71 #include <sys/policy.h>
72 #include <sys/sunddi.h>
73 #include <sys/filio.h>
74 #include <sys/sid.h>
75 #include "fs/fs_subr.h"
76 #include <sys/zfs_ctldir.h>
77 #include <sys/zfs_fuid.h>
78 #include <sys/zfs_sa.h>
79 #include <sys/dnlc.h>
80 #include <sys/zfs_rlock.h>
81 #include <sys/extdirent.h>
82 #include <sys/kidmap.h>
83 #include <sys/cred.h>
84 #include <sys/attr.h>
85 
86 /*
87  * Programming rules.
88  *
89  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
90  * properly lock its in-core state, create a DMU transaction, do the work,
91  * record this work in the intent log (ZIL), commit the DMU transaction,
92  * and wait for the intent log to commit if it is a synchronous operation.
93  * Moreover, the vnode ops must work in both normal and log replay context.
94  * The ordering of events is important to avoid deadlocks and references
95  * to freed memory.  The example below illustrates the following Big Rules:
96  *
97  *  (1)	A check must be made in each zfs thread for a mounted file system.
98  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
99  *	A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
100  *	must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
101  *	can return EIO from the calling function.
102  *
103  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
104  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
105  *	First, if it's the last reference, the vnode/znode
106  *	can be freed, so the zp may point to freed memory.  Second, the last
107  *	reference will call zfs_zinactive(), which may induce a lot of work --
108  *	pushing cached pages (which acquires range locks) and syncing out
109  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
110  *	which could deadlock the system if you were already holding one.
111  *	If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
112  *
113  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
114  *	as they can span dmu_tx_assign() calls.
115  *
116  *  (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
117  *      dmu_tx_assign().  This is critical because we don't want to block
118  *      while holding locks.
119  *
120  *	If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT.  This
121  *	reduces lock contention and CPU usage when we must wait (note that if
122  *	throughput is constrained by the storage, nearly every transaction
123  *	must wait).
124  *
125  *      Note, in particular, that if a lock is sometimes acquired before
126  *      the tx assigns, and sometimes after (e.g. z_lock), then failing
127  *      to use a non-blocking assign can deadlock the system.  The scenario:
128  *
129  *	Thread A has grabbed a lock before calling dmu_tx_assign().
130  *	Thread B is in an already-assigned tx, and blocks for this lock.
131  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
132  *	forever, because the previous txg can't quiesce until B's tx commits.
133  *
134  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
135  *	then drop all locks, call dmu_tx_wait(), and try again.  On subsequent
136  *	calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
137  *	to indicate that this operation has already called dmu_tx_wait().
138  *	This will ensure that we don't retry forever, waiting a short bit
139  *	each time.
140  *
141  *  (5)	If the operation succeeded, generate the intent log entry for it
142  *	before dropping locks.  This ensures that the ordering of events
143  *	in the intent log matches the order in which they actually occurred.
144  *	During ZIL replay the zfs_log_* functions will update the sequence
145  *	number to indicate the zil transaction has replayed.
146  *
147  *  (6)	At the end of each vnode op, the DMU tx must always commit,
148  *	regardless of whether there were any errors.
149  *
150  *  (7)	After dropping all locks, invoke zil_commit(zilog, foid)
151  *	to ensure that synchronous semantics are provided when necessary.
152  *
153  * In general, this is how things should be ordered in each vnode op:
154  *
155  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
156  * top:
157  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
158  *	rw_enter(...);			// grab any other locks you need
159  *	tx = dmu_tx_create(...);	// get DMU tx
160  *	dmu_tx_hold_*();		// hold each object you might modify
161  *	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
162  *	if (error) {
163  *		rw_exit(...);		// drop locks
164  *		zfs_dirent_unlock(dl);	// unlock directory entry
165  *		VN_RELE(...);		// release held vnodes
166  *		if (error == ERESTART) {
167  *			waited = B_TRUE;
168  *			dmu_tx_wait(tx);
169  *			dmu_tx_abort(tx);
170  *			goto top;
171  *		}
172  *		dmu_tx_abort(tx);	// abort DMU tx
173  *		ZFS_EXIT(zfsvfs);	// finished in zfs
174  *		return (error);		// really out of space
175  *	}
176  *	error = do_real_work();		// do whatever this VOP does
177  *	if (error == 0)
178  *		zfs_log_*(...);		// on success, make ZIL entry
179  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
180  *	rw_exit(...);			// drop locks
181  *	zfs_dirent_unlock(dl);		// unlock directory entry
182  *	VN_RELE(...);			// release held vnodes
183  *	zil_commit(zilog, foid);	// synchronous when necessary
184  *	ZFS_EXIT(zfsvfs);		// finished in zfs
185  *	return (error);			// done, report error
186  */
187 
188 /* ARGSUSED */
189 static int
190 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
191 {
192 	znode_t	*zp = VTOZ(*vpp);
193 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
194 
195 	ZFS_ENTER(zfsvfs);
196 	ZFS_VERIFY_ZP(zp);
197 
198 	if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
199 	    ((flag & FAPPEND) == 0)) {
200 		ZFS_EXIT(zfsvfs);
201 		return (SET_ERROR(EPERM));
202 	}
203 
204 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
205 	    ZTOV(zp)->v_type == VREG &&
206 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
207 		if (fs_vscan(*vpp, cr, 0) != 0) {
208 			ZFS_EXIT(zfsvfs);
209 			return (SET_ERROR(EACCES));
210 		}
211 	}
212 
213 	/* Keep a count of the synchronous opens in the znode */
214 	if (flag & (FSYNC | FDSYNC))
215 		atomic_inc_32(&zp->z_sync_cnt);
216 
217 	ZFS_EXIT(zfsvfs);
218 	return (0);
219 }
220 
221 /* ARGSUSED */
222 static int
223 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
224     caller_context_t *ct)
225 {
226 	znode_t	*zp = VTOZ(vp);
227 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
228 
229 	/*
230 	 * Clean up any locks held by this process on the vp.
231 	 */
232 	cleanlocks(vp, ddi_get_pid(), 0);
233 	cleanshares(vp, ddi_get_pid());
234 
235 	ZFS_ENTER(zfsvfs);
236 	ZFS_VERIFY_ZP(zp);
237 
238 	/* Decrement the synchronous opens in the znode */
239 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
240 		atomic_dec_32(&zp->z_sync_cnt);
241 
242 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
243 	    ZTOV(zp)->v_type == VREG &&
244 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
245 		VERIFY(fs_vscan(vp, cr, 1) == 0);
246 
247 	ZFS_EXIT(zfsvfs);
248 	return (0);
249 }
250 
251 /*
252  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
253  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
254  */
255 static int
256 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
257 {
258 	znode_t	*zp = VTOZ(vp);
259 	uint64_t noff = (uint64_t)*off; /* new offset */
260 	uint64_t file_sz;
261 	int error;
262 	boolean_t hole;
263 
264 	file_sz = zp->z_size;
265 	if (noff >= file_sz)  {
266 		return (SET_ERROR(ENXIO));
267 	}
268 
269 	if (cmd == _FIO_SEEK_HOLE)
270 		hole = B_TRUE;
271 	else
272 		hole = B_FALSE;
273 
274 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
275 
276 	if (error == ESRCH)
277 		return (SET_ERROR(ENXIO));
278 
279 	/*
280 	 * We could find a hole that begins after the logical end-of-file,
281 	 * because dmu_offset_next() only works on whole blocks.  If the
282 	 * EOF falls mid-block, then indicate that the "virtual hole"
283 	 * at the end of the file begins at the logical EOF, rather than
284 	 * at the end of the last block.
285 	 */
286 	if (noff > file_sz) {
287 		ASSERT(hole);
288 		noff = file_sz;
289 	}
290 
291 	if (noff < *off)
292 		return (error);
293 	*off = noff;
294 	return (error);
295 }
296 
297 /* ARGSUSED */
298 static int
299 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
300     int *rvalp, caller_context_t *ct)
301 {
302 	offset_t off;
303 	offset_t ndata;
304 	dmu_object_info_t doi;
305 	int error;
306 	zfsvfs_t *zfsvfs;
307 	znode_t *zp;
308 
309 	switch (com) {
310 	case _FIOFFS:
311 	{
312 		return (zfs_sync(vp->v_vfsp, 0, cred));
313 
314 		/*
315 		 * The following two ioctls are used by bfu.  Faking out,
316 		 * necessary to avoid bfu errors.
317 		 */
318 	}
319 	case _FIOGDIO:
320 	case _FIOSDIO:
321 	{
322 		return (0);
323 	}
324 
325 	case _FIO_SEEK_DATA:
326 	case _FIO_SEEK_HOLE:
327 	{
328 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
329 			return (SET_ERROR(EFAULT));
330 
331 		zp = VTOZ(vp);
332 		zfsvfs = zp->z_zfsvfs;
333 		ZFS_ENTER(zfsvfs);
334 		ZFS_VERIFY_ZP(zp);
335 
336 		/* offset parameter is in/out */
337 		error = zfs_holey(vp, com, &off);
338 		ZFS_EXIT(zfsvfs);
339 		if (error)
340 			return (error);
341 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
342 			return (SET_ERROR(EFAULT));
343 		return (0);
344 	}
345 	case _FIO_COUNT_FILLED:
346 	{
347 		/*
348 		 * _FIO_COUNT_FILLED adds a new ioctl command which
349 		 * exposes the number of filled blocks in a
350 		 * ZFS object.
351 		 */
352 		zp = VTOZ(vp);
353 		zfsvfs = zp->z_zfsvfs;
354 		ZFS_ENTER(zfsvfs);
355 		ZFS_VERIFY_ZP(zp);
356 
357 		/*
358 		 * Wait for all dirty blocks for this object
359 		 * to get synced out to disk, and the DMU info
360 		 * updated.
361 		 */
362 		error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id);
363 		if (error) {
364 			ZFS_EXIT(zfsvfs);
365 			return (error);
366 		}
367 
368 		/*
369 		 * Retrieve fill count from DMU object.
370 		 */
371 		error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi);
372 		if (error) {
373 			ZFS_EXIT(zfsvfs);
374 			return (error);
375 		}
376 
377 		ndata = doi.doi_fill_count;
378 
379 		ZFS_EXIT(zfsvfs);
380 		if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag))
381 			return (SET_ERROR(EFAULT));
382 		return (0);
383 	}
384 	}
385 	return (SET_ERROR(ENOTTY));
386 }
387 
388 /*
389  * Utility functions to map and unmap a single physical page.  These
390  * are used to manage the mappable copies of ZFS file data, and therefore
391  * do not update ref/mod bits.
392  */
393 caddr_t
394 zfs_map_page(page_t *pp, enum seg_rw rw)
395 {
396 	if (kpm_enable)
397 		return (hat_kpm_mapin(pp, 0));
398 	ASSERT(rw == S_READ || rw == S_WRITE);
399 	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
400 	    (caddr_t)-1));
401 }
402 
403 void
404 zfs_unmap_page(page_t *pp, caddr_t addr)
405 {
406 	if (kpm_enable) {
407 		hat_kpm_mapout(pp, 0, addr);
408 	} else {
409 		ppmapout(addr);
410 	}
411 }
412 
413 /*
414  * When a file is memory mapped, we must keep the IO data synchronized
415  * between the DMU cache and the memory mapped pages.  What this means:
416  *
417  * On Write:	If we find a memory mapped page, we write to *both*
418  *		the page and the dmu buffer.
419  */
420 static void
421 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
422 {
423 	int64_t	off;
424 
425 	off = start & PAGEOFFSET;
426 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
427 		page_t *pp;
428 		uint64_t nbytes = MIN(PAGESIZE - off, len);
429 
430 		if (pp = page_lookup(vp, start, SE_SHARED)) {
431 			caddr_t va;
432 
433 			va = zfs_map_page(pp, S_WRITE);
434 			(void) dmu_read(os, oid, start+off, nbytes, va+off,
435 			    DMU_READ_PREFETCH);
436 			zfs_unmap_page(pp, va);
437 			page_unlock(pp);
438 		}
439 		len -= nbytes;
440 		off = 0;
441 	}
442 }
443 
444 /*
445  * When a file is memory mapped, we must keep the IO data synchronized
446  * between the DMU cache and the memory mapped pages.  What this means:
447  *
448  * On Read:	We "read" preferentially from memory mapped pages,
449  *		else we default from the dmu buffer.
450  *
451  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
452  *	 the file is memory mapped.
453  */
454 static int
455 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
456 {
457 	znode_t *zp = VTOZ(vp);
458 	int64_t	start, off;
459 	int len = nbytes;
460 	int error = 0;
461 
462 	start = uio->uio_loffset;
463 	off = start & PAGEOFFSET;
464 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
465 		page_t *pp;
466 		uint64_t bytes = MIN(PAGESIZE - off, len);
467 
468 		if (pp = page_lookup(vp, start, SE_SHARED)) {
469 			caddr_t va;
470 
471 			va = zfs_map_page(pp, S_READ);
472 			error = uiomove(va + off, bytes, UIO_READ, uio);
473 			zfs_unmap_page(pp, va);
474 			page_unlock(pp);
475 		} else {
476 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
477 			    uio, bytes);
478 		}
479 		len -= bytes;
480 		off = 0;
481 		if (error)
482 			break;
483 	}
484 	return (error);
485 }
486 
487 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
488 
489 /*
490  * Read bytes from specified file into supplied buffer.
491  *
492  *	IN:	vp	- vnode of file to be read from.
493  *		uio	- structure supplying read location, range info,
494  *			  and return buffer.
495  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
496  *		cr	- credentials of caller.
497  *		ct	- caller context
498  *
499  *	OUT:	uio	- updated offset and range, buffer filled.
500  *
501  *	RETURN:	0 on success, error code on failure.
502  *
503  * Side Effects:
504  *	vp - atime updated if byte count > 0
505  */
506 /* ARGSUSED */
507 static int
508 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
509 {
510 	znode_t		*zp = VTOZ(vp);
511 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
512 	ssize_t		n, nbytes;
513 	int		error = 0;
514 	rl_t		*rl;
515 	xuio_t		*xuio = NULL;
516 
517 	ZFS_ENTER(zfsvfs);
518 	ZFS_VERIFY_ZP(zp);
519 
520 	if (zp->z_pflags & ZFS_AV_QUARANTINED) {
521 		ZFS_EXIT(zfsvfs);
522 		return (SET_ERROR(EACCES));
523 	}
524 
525 	/*
526 	 * Validate file offset
527 	 */
528 	if (uio->uio_loffset < (offset_t)0) {
529 		ZFS_EXIT(zfsvfs);
530 		return (SET_ERROR(EINVAL));
531 	}
532 
533 	/*
534 	 * Fasttrack empty reads
535 	 */
536 	if (uio->uio_resid == 0) {
537 		ZFS_EXIT(zfsvfs);
538 		return (0);
539 	}
540 
541 	/*
542 	 * Check for mandatory locks
543 	 */
544 	if (MANDMODE(zp->z_mode)) {
545 		if (error = chklock(vp, FREAD,
546 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
547 			ZFS_EXIT(zfsvfs);
548 			return (error);
549 		}
550 	}
551 
552 	/*
553 	 * If we're in FRSYNC mode, sync out this znode before reading it.
554 	 */
555 	if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
556 		zil_commit(zfsvfs->z_log, zp->z_id);
557 
558 	/*
559 	 * Lock the range against changes.
560 	 */
561 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
562 
563 	/*
564 	 * If we are reading past end-of-file we can skip
565 	 * to the end; but we might still need to set atime.
566 	 */
567 	if (uio->uio_loffset >= zp->z_size) {
568 		error = 0;
569 		goto out;
570 	}
571 
572 	ASSERT(uio->uio_loffset < zp->z_size);
573 	n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
574 
575 	if ((uio->uio_extflg == UIO_XUIO) &&
576 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
577 		int nblk;
578 		int blksz = zp->z_blksz;
579 		uint64_t offset = uio->uio_loffset;
580 
581 		xuio = (xuio_t *)uio;
582 		if ((ISP2(blksz))) {
583 			nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
584 			    blksz)) / blksz;
585 		} else {
586 			ASSERT(offset + n <= blksz);
587 			nblk = 1;
588 		}
589 		(void) dmu_xuio_init(xuio, nblk);
590 
591 		if (vn_has_cached_data(vp)) {
592 			/*
593 			 * For simplicity, we always allocate a full buffer
594 			 * even if we only expect to read a portion of a block.
595 			 */
596 			while (--nblk >= 0) {
597 				(void) dmu_xuio_add(xuio,
598 				    dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
599 				    blksz), 0, blksz);
600 			}
601 		}
602 	}
603 
604 	while (n > 0) {
605 		nbytes = MIN(n, zfs_read_chunk_size -
606 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
607 
608 		if (vn_has_cached_data(vp)) {
609 			error = mappedread(vp, nbytes, uio);
610 		} else {
611 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
612 			    uio, nbytes);
613 		}
614 		if (error) {
615 			/* convert checksum errors into IO errors */
616 			if (error == ECKSUM)
617 				error = SET_ERROR(EIO);
618 			break;
619 		}
620 
621 		n -= nbytes;
622 	}
623 out:
624 	zfs_range_unlock(rl);
625 
626 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
627 	ZFS_EXIT(zfsvfs);
628 	return (error);
629 }
630 
631 /*
632  * Write the bytes to a file.
633  *
634  *	IN:	vp	- vnode of file to be written to.
635  *		uio	- structure supplying write location, range info,
636  *			  and data buffer.
637  *		ioflag	- FAPPEND, FSYNC, and/or FDSYNC.  FAPPEND is
638  *			  set if in append mode.
639  *		cr	- credentials of caller.
640  *		ct	- caller context (NFS/CIFS fem monitor only)
641  *
642  *	OUT:	uio	- updated offset and range.
643  *
644  *	RETURN:	0 on success, error code on failure.
645  *
646  * Timestamps:
647  *	vp - ctime|mtime updated if byte count > 0
648  */
649 
650 /* ARGSUSED */
651 static int
652 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
653 {
654 	znode_t		*zp = VTOZ(vp);
655 	rlim64_t	limit = uio->uio_llimit;
656 	ssize_t		start_resid = uio->uio_resid;
657 	ssize_t		tx_bytes;
658 	uint64_t	end_size;
659 	dmu_tx_t	*tx;
660 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
661 	zilog_t		*zilog;
662 	offset_t	woff;
663 	ssize_t		n, nbytes;
664 	rl_t		*rl;
665 	int		max_blksz = zfsvfs->z_max_blksz;
666 	int		error = 0;
667 	arc_buf_t	*abuf;
668 	iovec_t		*aiov = NULL;
669 	xuio_t		*xuio = NULL;
670 	int		i_iov = 0;
671 	int		iovcnt = uio->uio_iovcnt;
672 	iovec_t		*iovp = uio->uio_iov;
673 	int		write_eof;
674 	int		count = 0;
675 	sa_bulk_attr_t	bulk[4];
676 	uint64_t	mtime[2], ctime[2];
677 
678 	/*
679 	 * Fasttrack empty write
680 	 */
681 	n = start_resid;
682 	if (n == 0)
683 		return (0);
684 
685 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
686 		limit = MAXOFFSET_T;
687 
688 	ZFS_ENTER(zfsvfs);
689 	ZFS_VERIFY_ZP(zp);
690 
691 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
692 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
693 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
694 	    &zp->z_size, 8);
695 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
696 	    &zp->z_pflags, 8);
697 
698 	/*
699 	 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
700 	 * callers might not be able to detect properly that we are read-only,
701 	 * so check it explicitly here.
702 	 */
703 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
704 		ZFS_EXIT(zfsvfs);
705 		return (SET_ERROR(EROFS));
706 	}
707 
708 	/*
709 	 * If immutable or not appending then return EPERM
710 	 */
711 	if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
712 	    ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
713 	    (uio->uio_loffset < zp->z_size))) {
714 		ZFS_EXIT(zfsvfs);
715 		return (SET_ERROR(EPERM));
716 	}
717 
718 	zilog = zfsvfs->z_log;
719 
720 	/*
721 	 * Validate file offset
722 	 */
723 	woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
724 	if (woff < 0) {
725 		ZFS_EXIT(zfsvfs);
726 		return (SET_ERROR(EINVAL));
727 	}
728 
729 	/*
730 	 * Check for mandatory locks before calling zfs_range_lock()
731 	 * in order to prevent a deadlock with locks set via fcntl().
732 	 */
733 	if (MANDMODE((mode_t)zp->z_mode) &&
734 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
735 		ZFS_EXIT(zfsvfs);
736 		return (error);
737 	}
738 
739 	/*
740 	 * Pre-fault the pages to ensure slow (eg NFS) pages
741 	 * don't hold up txg.
742 	 * Skip this if uio contains loaned arc_buf.
743 	 */
744 	if ((uio->uio_extflg == UIO_XUIO) &&
745 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
746 		xuio = (xuio_t *)uio;
747 	else
748 		uio_prefaultpages(MIN(n, max_blksz), uio);
749 
750 	/*
751 	 * If in append mode, set the io offset pointer to eof.
752 	 */
753 	if (ioflag & FAPPEND) {
754 		/*
755 		 * Obtain an appending range lock to guarantee file append
756 		 * semantics.  We reset the write offset once we have the lock.
757 		 */
758 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
759 		woff = rl->r_off;
760 		if (rl->r_len == UINT64_MAX) {
761 			/*
762 			 * We overlocked the file because this write will cause
763 			 * the file block size to increase.
764 			 * Note that zp_size cannot change with this lock held.
765 			 */
766 			woff = zp->z_size;
767 		}
768 		uio->uio_loffset = woff;
769 	} else {
770 		/*
771 		 * Note that if the file block size will change as a result of
772 		 * this write, then this range lock will lock the entire file
773 		 * so that we can re-write the block safely.
774 		 */
775 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
776 	}
777 
778 	if (woff >= limit) {
779 		zfs_range_unlock(rl);
780 		ZFS_EXIT(zfsvfs);
781 		return (SET_ERROR(EFBIG));
782 	}
783 
784 	if ((woff + n) > limit || woff > (limit - n))
785 		n = limit - woff;
786 
787 	/* Will this write extend the file length? */
788 	write_eof = (woff + n > zp->z_size);
789 
790 	end_size = MAX(zp->z_size, woff + n);
791 
792 	/*
793 	 * Write the file in reasonable size chunks.  Each chunk is written
794 	 * in a separate transaction; this keeps the intent log records small
795 	 * and allows us to do more fine-grained space accounting.
796 	 */
797 	while (n > 0) {
798 		abuf = NULL;
799 		woff = uio->uio_loffset;
800 		if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
801 		    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
802 			if (abuf != NULL)
803 				dmu_return_arcbuf(abuf);
804 			error = SET_ERROR(EDQUOT);
805 			break;
806 		}
807 
808 		if (xuio && abuf == NULL) {
809 			ASSERT(i_iov < iovcnt);
810 			aiov = &iovp[i_iov];
811 			abuf = dmu_xuio_arcbuf(xuio, i_iov);
812 			dmu_xuio_clear(xuio, i_iov);
813 			DTRACE_PROBE3(zfs_cp_write, int, i_iov,
814 			    iovec_t *, aiov, arc_buf_t *, abuf);
815 			ASSERT((aiov->iov_base == abuf->b_data) ||
816 			    ((char *)aiov->iov_base - (char *)abuf->b_data +
817 			    aiov->iov_len == arc_buf_size(abuf)));
818 			i_iov++;
819 		} else if (abuf == NULL && n >= max_blksz &&
820 		    woff >= zp->z_size &&
821 		    P2PHASE(woff, max_blksz) == 0 &&
822 		    zp->z_blksz == max_blksz) {
823 			/*
824 			 * This write covers a full block.  "Borrow" a buffer
825 			 * from the dmu so that we can fill it before we enter
826 			 * a transaction.  This avoids the possibility of
827 			 * holding up the transaction if the data copy hangs
828 			 * up on a pagefault (e.g., from an NFS server mapping).
829 			 */
830 			size_t cbytes;
831 
832 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
833 			    max_blksz);
834 			ASSERT(abuf != NULL);
835 			ASSERT(arc_buf_size(abuf) == max_blksz);
836 			if (error = uiocopy(abuf->b_data, max_blksz,
837 			    UIO_WRITE, uio, &cbytes)) {
838 				dmu_return_arcbuf(abuf);
839 				break;
840 			}
841 			ASSERT(cbytes == max_blksz);
842 		}
843 
844 		/*
845 		 * Start a transaction.
846 		 */
847 		tx = dmu_tx_create(zfsvfs->z_os);
848 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
849 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
850 		zfs_sa_upgrade_txholds(tx, zp);
851 		error = dmu_tx_assign(tx, TXG_WAIT);
852 		if (error) {
853 			dmu_tx_abort(tx);
854 			if (abuf != NULL)
855 				dmu_return_arcbuf(abuf);
856 			break;
857 		}
858 
859 		/*
860 		 * If zfs_range_lock() over-locked we grow the blocksize
861 		 * and then reduce the lock range.  This will only happen
862 		 * on the first iteration since zfs_range_reduce() will
863 		 * shrink down r_len to the appropriate size.
864 		 */
865 		if (rl->r_len == UINT64_MAX) {
866 			uint64_t new_blksz;
867 
868 			if (zp->z_blksz > max_blksz) {
869 				/*
870 				 * File's blocksize is already larger than the
871 				 * "recordsize" property.  Only let it grow to
872 				 * the next power of 2.
873 				 */
874 				ASSERT(!ISP2(zp->z_blksz));
875 				new_blksz = MIN(end_size,
876 				    1 << highbit64(zp->z_blksz));
877 			} else {
878 				new_blksz = MIN(end_size, max_blksz);
879 			}
880 			zfs_grow_blocksize(zp, new_blksz, tx);
881 			zfs_range_reduce(rl, woff, n);
882 		}
883 
884 		/*
885 		 * XXX - should we really limit each write to z_max_blksz?
886 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
887 		 */
888 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
889 
890 		if (abuf == NULL) {
891 			tx_bytes = uio->uio_resid;
892 			error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
893 			    uio, nbytes, tx);
894 			tx_bytes -= uio->uio_resid;
895 		} else {
896 			tx_bytes = nbytes;
897 			ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
898 			/*
899 			 * If this is not a full block write, but we are
900 			 * extending the file past EOF and this data starts
901 			 * block-aligned, use assign_arcbuf().  Otherwise,
902 			 * write via dmu_write().
903 			 */
904 			if (tx_bytes < max_blksz && (!write_eof ||
905 			    aiov->iov_base != abuf->b_data)) {
906 				ASSERT(xuio);
907 				dmu_write(zfsvfs->z_os, zp->z_id, woff,
908 				    aiov->iov_len, aiov->iov_base, tx);
909 				dmu_return_arcbuf(abuf);
910 				xuio_stat_wbuf_copied();
911 			} else {
912 				ASSERT(xuio || tx_bytes == max_blksz);
913 				dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
914 				    woff, abuf, tx);
915 			}
916 			ASSERT(tx_bytes <= uio->uio_resid);
917 			uioskip(uio, tx_bytes);
918 		}
919 		if (tx_bytes && vn_has_cached_data(vp)) {
920 			update_pages(vp, woff,
921 			    tx_bytes, zfsvfs->z_os, zp->z_id);
922 		}
923 
924 		/*
925 		 * If we made no progress, we're done.  If we made even
926 		 * partial progress, update the znode and ZIL accordingly.
927 		 */
928 		if (tx_bytes == 0) {
929 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
930 			    (void *)&zp->z_size, sizeof (uint64_t), tx);
931 			dmu_tx_commit(tx);
932 			ASSERT(error != 0);
933 			break;
934 		}
935 
936 		/*
937 		 * Clear Set-UID/Set-GID bits on successful write if not
938 		 * privileged and at least one of the excute bits is set.
939 		 *
940 		 * It would be nice to to this after all writes have
941 		 * been done, but that would still expose the ISUID/ISGID
942 		 * to another app after the partial write is committed.
943 		 *
944 		 * Note: we don't call zfs_fuid_map_id() here because
945 		 * user 0 is not an ephemeral uid.
946 		 */
947 		mutex_enter(&zp->z_acl_lock);
948 		if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
949 		    (S_IXUSR >> 6))) != 0 &&
950 		    (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
951 		    secpolicy_vnode_setid_retain(cr,
952 		    (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
953 			uint64_t newmode;
954 			zp->z_mode &= ~(S_ISUID | S_ISGID);
955 			newmode = zp->z_mode;
956 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
957 			    (void *)&newmode, sizeof (uint64_t), tx);
958 		}
959 		mutex_exit(&zp->z_acl_lock);
960 
961 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
962 		    B_TRUE);
963 
964 		/*
965 		 * Update the file size (zp_size) if it has changed;
966 		 * account for possible concurrent updates.
967 		 */
968 		while ((end_size = zp->z_size) < uio->uio_loffset) {
969 			(void) atomic_cas_64(&zp->z_size, end_size,
970 			    uio->uio_loffset);
971 			ASSERT(error == 0);
972 		}
973 		/*
974 		 * If we are replaying and eof is non zero then force
975 		 * the file size to the specified eof. Note, there's no
976 		 * concurrency during replay.
977 		 */
978 		if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
979 			zp->z_size = zfsvfs->z_replay_eof;
980 
981 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
982 
983 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
984 		dmu_tx_commit(tx);
985 
986 		if (error != 0)
987 			break;
988 		ASSERT(tx_bytes == nbytes);
989 		n -= nbytes;
990 
991 		if (!xuio && n > 0)
992 			uio_prefaultpages(MIN(n, max_blksz), uio);
993 	}
994 
995 	zfs_range_unlock(rl);
996 
997 	/*
998 	 * If we're in replay mode, or we made no progress, return error.
999 	 * Otherwise, it's at least a partial write, so it's successful.
1000 	 */
1001 	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1002 		ZFS_EXIT(zfsvfs);
1003 		return (error);
1004 	}
1005 
1006 	if (ioflag & (FSYNC | FDSYNC) ||
1007 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1008 		zil_commit(zilog, zp->z_id);
1009 
1010 	ZFS_EXIT(zfsvfs);
1011 	return (0);
1012 }
1013 
1014 void
1015 zfs_get_done(zgd_t *zgd, int error)
1016 {
1017 	znode_t *zp = zgd->zgd_private;
1018 	objset_t *os = zp->z_zfsvfs->z_os;
1019 
1020 	if (zgd->zgd_db)
1021 		dmu_buf_rele(zgd->zgd_db, zgd);
1022 
1023 	zfs_range_unlock(zgd->zgd_rl);
1024 
1025 	/*
1026 	 * Release the vnode asynchronously as we currently have the
1027 	 * txg stopped from syncing.
1028 	 */
1029 	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1030 
1031 	if (error == 0 && zgd->zgd_bp)
1032 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1033 
1034 	kmem_free(zgd, sizeof (zgd_t));
1035 }
1036 
1037 #ifdef DEBUG
1038 static int zil_fault_io = 0;
1039 #endif
1040 
1041 /*
1042  * Get data to generate a TX_WRITE intent log record.
1043  */
1044 int
1045 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1046 {
1047 	zfsvfs_t *zfsvfs = arg;
1048 	objset_t *os = zfsvfs->z_os;
1049 	znode_t *zp;
1050 	uint64_t object = lr->lr_foid;
1051 	uint64_t offset = lr->lr_offset;
1052 	uint64_t size = lr->lr_length;
1053 	blkptr_t *bp = &lr->lr_blkptr;
1054 	dmu_buf_t *db;
1055 	zgd_t *zgd;
1056 	int error = 0;
1057 
1058 	ASSERT(zio != NULL);
1059 	ASSERT(size != 0);
1060 
1061 	/*
1062 	 * Nothing to do if the file has been removed
1063 	 */
1064 	if (zfs_zget(zfsvfs, object, &zp) != 0)
1065 		return (SET_ERROR(ENOENT));
1066 	if (zp->z_unlinked) {
1067 		/*
1068 		 * Release the vnode asynchronously as we currently have the
1069 		 * txg stopped from syncing.
1070 		 */
1071 		VN_RELE_ASYNC(ZTOV(zp),
1072 		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1073 		return (SET_ERROR(ENOENT));
1074 	}
1075 
1076 	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1077 	zgd->zgd_zilog = zfsvfs->z_log;
1078 	zgd->zgd_private = zp;
1079 
1080 	/*
1081 	 * Write records come in two flavors: immediate and indirect.
1082 	 * For small writes it's cheaper to store the data with the
1083 	 * log record (immediate); for large writes it's cheaper to
1084 	 * sync the data and get a pointer to it (indirect) so that
1085 	 * we don't have to write the data twice.
1086 	 */
1087 	if (buf != NULL) { /* immediate write */
1088 		zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1089 		/* test for truncation needs to be done while range locked */
1090 		if (offset >= zp->z_size) {
1091 			error = SET_ERROR(ENOENT);
1092 		} else {
1093 			error = dmu_read(os, object, offset, size, buf,
1094 			    DMU_READ_NO_PREFETCH);
1095 		}
1096 		ASSERT(error == 0 || error == ENOENT);
1097 	} else { /* indirect write */
1098 		/*
1099 		 * Have to lock the whole block to ensure when it's
1100 		 * written out and it's checksum is being calculated
1101 		 * that no one can change the data. We need to re-check
1102 		 * blocksize after we get the lock in case it's changed!
1103 		 */
1104 		for (;;) {
1105 			uint64_t blkoff;
1106 			size = zp->z_blksz;
1107 			blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1108 			offset -= blkoff;
1109 			zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1110 			    RL_READER);
1111 			if (zp->z_blksz == size)
1112 				break;
1113 			offset += blkoff;
1114 			zfs_range_unlock(zgd->zgd_rl);
1115 		}
1116 		/* test for truncation needs to be done while range locked */
1117 		if (lr->lr_offset >= zp->z_size)
1118 			error = SET_ERROR(ENOENT);
1119 #ifdef DEBUG
1120 		if (zil_fault_io) {
1121 			error = SET_ERROR(EIO);
1122 			zil_fault_io = 0;
1123 		}
1124 #endif
1125 		if (error == 0)
1126 			error = dmu_buf_hold(os, object, offset, zgd, &db,
1127 			    DMU_READ_NO_PREFETCH);
1128 
1129 		if (error == 0) {
1130 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1131 			if (obp) {
1132 				ASSERT(BP_IS_HOLE(bp));
1133 				*bp = *obp;
1134 			}
1135 
1136 			zgd->zgd_db = db;
1137 			zgd->zgd_bp = bp;
1138 
1139 			ASSERT(db->db_offset == offset);
1140 			ASSERT(db->db_size == size);
1141 
1142 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1143 			    zfs_get_done, zgd);
1144 			ASSERT(error || lr->lr_length <= size);
1145 
1146 			/*
1147 			 * On success, we need to wait for the write I/O
1148 			 * initiated by dmu_sync() to complete before we can
1149 			 * release this dbuf.  We will finish everything up
1150 			 * in the zfs_get_done() callback.
1151 			 */
1152 			if (error == 0)
1153 				return (0);
1154 
1155 			if (error == EALREADY) {
1156 				lr->lr_common.lrc_txtype = TX_WRITE2;
1157 				error = 0;
1158 			}
1159 		}
1160 	}
1161 
1162 	zfs_get_done(zgd, error);
1163 
1164 	return (error);
1165 }
1166 
1167 /*ARGSUSED*/
1168 static int
1169 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1170     caller_context_t *ct)
1171 {
1172 	znode_t *zp = VTOZ(vp);
1173 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1174 	int error;
1175 
1176 	ZFS_ENTER(zfsvfs);
1177 	ZFS_VERIFY_ZP(zp);
1178 
1179 	if (flag & V_ACE_MASK)
1180 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1181 	else
1182 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
1183 
1184 	ZFS_EXIT(zfsvfs);
1185 	return (error);
1186 }
1187 
1188 /*
1189  * If vnode is for a device return a specfs vnode instead.
1190  */
1191 static int
1192 specvp_check(vnode_t **vpp, cred_t *cr)
1193 {
1194 	int error = 0;
1195 
1196 	if (IS_DEVVP(*vpp)) {
1197 		struct vnode *svp;
1198 
1199 		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1200 		VN_RELE(*vpp);
1201 		if (svp == NULL)
1202 			error = SET_ERROR(ENOSYS);
1203 		*vpp = svp;
1204 	}
1205 	return (error);
1206 }
1207 
1208 
1209 /*
1210  * Lookup an entry in a directory, or an extended attribute directory.
1211  * If it exists, return a held vnode reference for it.
1212  *
1213  *	IN:	dvp	- vnode of directory to search.
1214  *		nm	- name of entry to lookup.
1215  *		pnp	- full pathname to lookup [UNUSED].
1216  *		flags	- LOOKUP_XATTR set if looking for an attribute.
1217  *		rdir	- root directory vnode [UNUSED].
1218  *		cr	- credentials of caller.
1219  *		ct	- caller context
1220  *		direntflags - directory lookup flags
1221  *		realpnp - returned pathname.
1222  *
1223  *	OUT:	vpp	- vnode of located entry, NULL if not found.
1224  *
1225  *	RETURN:	0 on success, error code on failure.
1226  *
1227  * Timestamps:
1228  *	NA
1229  */
1230 /* ARGSUSED */
1231 static int
1232 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1233     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1234     int *direntflags, pathname_t *realpnp)
1235 {
1236 	znode_t *zdp = VTOZ(dvp);
1237 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1238 	int	error = 0;
1239 
1240 	/* fast path */
1241 	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1242 
1243 		if (dvp->v_type != VDIR) {
1244 			return (SET_ERROR(ENOTDIR));
1245 		} else if (zdp->z_sa_hdl == NULL) {
1246 			return (SET_ERROR(EIO));
1247 		}
1248 
1249 		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1250 			error = zfs_fastaccesschk_execute(zdp, cr);
1251 			if (!error) {
1252 				*vpp = dvp;
1253 				VN_HOLD(*vpp);
1254 				return (0);
1255 			}
1256 			return (error);
1257 		} else {
1258 			vnode_t *tvp = dnlc_lookup(dvp, nm);
1259 
1260 			if (tvp) {
1261 				error = zfs_fastaccesschk_execute(zdp, cr);
1262 				if (error) {
1263 					VN_RELE(tvp);
1264 					return (error);
1265 				}
1266 				if (tvp == DNLC_NO_VNODE) {
1267 					VN_RELE(tvp);
1268 					return (SET_ERROR(ENOENT));
1269 				} else {
1270 					*vpp = tvp;
1271 					return (specvp_check(vpp, cr));
1272 				}
1273 			}
1274 		}
1275 	}
1276 
1277 	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1278 
1279 	ZFS_ENTER(zfsvfs);
1280 	ZFS_VERIFY_ZP(zdp);
1281 
1282 	*vpp = NULL;
1283 
1284 	if (flags & LOOKUP_XATTR) {
1285 		/*
1286 		 * If the xattr property is off, refuse the lookup request.
1287 		 */
1288 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1289 			ZFS_EXIT(zfsvfs);
1290 			return (SET_ERROR(EINVAL));
1291 		}
1292 
1293 		/*
1294 		 * We don't allow recursive attributes..
1295 		 * Maybe someday we will.
1296 		 */
1297 		if (zdp->z_pflags & ZFS_XATTR) {
1298 			ZFS_EXIT(zfsvfs);
1299 			return (SET_ERROR(EINVAL));
1300 		}
1301 
1302 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1303 			ZFS_EXIT(zfsvfs);
1304 			return (error);
1305 		}
1306 
1307 		/*
1308 		 * Do we have permission to get into attribute directory?
1309 		 */
1310 
1311 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1312 		    B_FALSE, cr)) {
1313 			VN_RELE(*vpp);
1314 			*vpp = NULL;
1315 		}
1316 
1317 		ZFS_EXIT(zfsvfs);
1318 		return (error);
1319 	}
1320 
1321 	if (dvp->v_type != VDIR) {
1322 		ZFS_EXIT(zfsvfs);
1323 		return (SET_ERROR(ENOTDIR));
1324 	}
1325 
1326 	/*
1327 	 * Check accessibility of directory.
1328 	 */
1329 
1330 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1331 		ZFS_EXIT(zfsvfs);
1332 		return (error);
1333 	}
1334 
1335 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1336 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1337 		ZFS_EXIT(zfsvfs);
1338 		return (SET_ERROR(EILSEQ));
1339 	}
1340 
1341 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1342 	if (error == 0)
1343 		error = specvp_check(vpp, cr);
1344 
1345 	ZFS_EXIT(zfsvfs);
1346 	return (error);
1347 }
1348 
1349 /*
1350  * Attempt to create a new entry in a directory.  If the entry
1351  * already exists, truncate the file if permissible, else return
1352  * an error.  Return the vp of the created or trunc'd file.
1353  *
1354  *	IN:	dvp	- vnode of directory to put new file entry in.
1355  *		name	- name of new file entry.
1356  *		vap	- attributes of new file.
1357  *		excl	- flag indicating exclusive or non-exclusive mode.
1358  *		mode	- mode to open file with.
1359  *		cr	- credentials of caller.
1360  *		flag	- large file flag [UNUSED].
1361  *		ct	- caller context
1362  *		vsecp	- ACL to be set
1363  *
1364  *	OUT:	vpp	- vnode of created or trunc'd entry.
1365  *
1366  *	RETURN:	0 on success, error code on failure.
1367  *
1368  * Timestamps:
1369  *	dvp - ctime|mtime updated if new entry created
1370  *	 vp - ctime|mtime always, atime if new
1371  */
1372 
1373 /* ARGSUSED */
1374 static int
1375 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1376     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1377     vsecattr_t *vsecp)
1378 {
1379 	znode_t		*zp, *dzp = VTOZ(dvp);
1380 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1381 	zilog_t		*zilog;
1382 	objset_t	*os;
1383 	zfs_dirlock_t	*dl;
1384 	dmu_tx_t	*tx;
1385 	int		error;
1386 	ksid_t		*ksid;
1387 	uid_t		uid;
1388 	gid_t		gid = crgetgid(cr);
1389 	zfs_acl_ids_t   acl_ids;
1390 	boolean_t	fuid_dirtied;
1391 	boolean_t	have_acl = B_FALSE;
1392 	boolean_t	waited = B_FALSE;
1393 
1394 	/*
1395 	 * If we have an ephemeral id, ACL, or XVATTR then
1396 	 * make sure file system is at proper version
1397 	 */
1398 
1399 	ksid = crgetsid(cr, KSID_OWNER);
1400 	if (ksid)
1401 		uid = ksid_getid(ksid);
1402 	else
1403 		uid = crgetuid(cr);
1404 
1405 	if (zfsvfs->z_use_fuids == B_FALSE &&
1406 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1407 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1408 		return (SET_ERROR(EINVAL));
1409 
1410 	ZFS_ENTER(zfsvfs);
1411 	ZFS_VERIFY_ZP(dzp);
1412 	os = zfsvfs->z_os;
1413 	zilog = zfsvfs->z_log;
1414 
1415 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1416 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1417 		ZFS_EXIT(zfsvfs);
1418 		return (SET_ERROR(EILSEQ));
1419 	}
1420 
1421 	if (vap->va_mask & AT_XVATTR) {
1422 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1423 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1424 			ZFS_EXIT(zfsvfs);
1425 			return (error);
1426 		}
1427 	}
1428 top:
1429 	*vpp = NULL;
1430 
1431 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1432 		vap->va_mode &= ~VSVTX;
1433 
1434 	if (*name == '\0') {
1435 		/*
1436 		 * Null component name refers to the directory itself.
1437 		 */
1438 		VN_HOLD(dvp);
1439 		zp = dzp;
1440 		dl = NULL;
1441 		error = 0;
1442 	} else {
1443 		/* possible VN_HOLD(zp) */
1444 		int zflg = 0;
1445 
1446 		if (flag & FIGNORECASE)
1447 			zflg |= ZCILOOK;
1448 
1449 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1450 		    NULL, NULL);
1451 		if (error) {
1452 			if (have_acl)
1453 				zfs_acl_ids_free(&acl_ids);
1454 			if (strcmp(name, "..") == 0)
1455 				error = SET_ERROR(EISDIR);
1456 			ZFS_EXIT(zfsvfs);
1457 			return (error);
1458 		}
1459 	}
1460 
1461 	if (zp == NULL) {
1462 		uint64_t txtype;
1463 
1464 		/*
1465 		 * Create a new file object and update the directory
1466 		 * to reference it.
1467 		 */
1468 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1469 			if (have_acl)
1470 				zfs_acl_ids_free(&acl_ids);
1471 			goto out;
1472 		}
1473 
1474 		/*
1475 		 * We only support the creation of regular files in
1476 		 * extended attribute directories.
1477 		 */
1478 
1479 		if ((dzp->z_pflags & ZFS_XATTR) &&
1480 		    (vap->va_type != VREG)) {
1481 			if (have_acl)
1482 				zfs_acl_ids_free(&acl_ids);
1483 			error = SET_ERROR(EINVAL);
1484 			goto out;
1485 		}
1486 
1487 		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1488 		    cr, vsecp, &acl_ids)) != 0)
1489 			goto out;
1490 		have_acl = B_TRUE;
1491 
1492 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1493 			zfs_acl_ids_free(&acl_ids);
1494 			error = SET_ERROR(EDQUOT);
1495 			goto out;
1496 		}
1497 
1498 		tx = dmu_tx_create(os);
1499 
1500 		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1501 		    ZFS_SA_BASE_ATTR_SIZE);
1502 
1503 		fuid_dirtied = zfsvfs->z_fuid_dirty;
1504 		if (fuid_dirtied)
1505 			zfs_fuid_txhold(zfsvfs, tx);
1506 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1507 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1508 		if (!zfsvfs->z_use_sa &&
1509 		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1510 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1511 			    0, acl_ids.z_aclp->z_acl_bytes);
1512 		}
1513 		error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1514 		if (error) {
1515 			zfs_dirent_unlock(dl);
1516 			if (error == ERESTART) {
1517 				waited = B_TRUE;
1518 				dmu_tx_wait(tx);
1519 				dmu_tx_abort(tx);
1520 				goto top;
1521 			}
1522 			zfs_acl_ids_free(&acl_ids);
1523 			dmu_tx_abort(tx);
1524 			ZFS_EXIT(zfsvfs);
1525 			return (error);
1526 		}
1527 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1528 
1529 		if (fuid_dirtied)
1530 			zfs_fuid_sync(zfsvfs, tx);
1531 
1532 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1533 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1534 		if (flag & FIGNORECASE)
1535 			txtype |= TX_CI;
1536 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1537 		    vsecp, acl_ids.z_fuidp, vap);
1538 		zfs_acl_ids_free(&acl_ids);
1539 		dmu_tx_commit(tx);
1540 	} else {
1541 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1542 
1543 		if (have_acl)
1544 			zfs_acl_ids_free(&acl_ids);
1545 		have_acl = B_FALSE;
1546 
1547 		/*
1548 		 * A directory entry already exists for this name.
1549 		 */
1550 		/*
1551 		 * Can't truncate an existing file if in exclusive mode.
1552 		 */
1553 		if (excl == EXCL) {
1554 			error = SET_ERROR(EEXIST);
1555 			goto out;
1556 		}
1557 		/*
1558 		 * Can't open a directory for writing.
1559 		 */
1560 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1561 			error = SET_ERROR(EISDIR);
1562 			goto out;
1563 		}
1564 		/*
1565 		 * Verify requested access to file.
1566 		 */
1567 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1568 			goto out;
1569 		}
1570 
1571 		mutex_enter(&dzp->z_lock);
1572 		dzp->z_seq++;
1573 		mutex_exit(&dzp->z_lock);
1574 
1575 		/*
1576 		 * Truncate regular files if requested.
1577 		 */
1578 		if ((ZTOV(zp)->v_type == VREG) &&
1579 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1580 			/* we can't hold any locks when calling zfs_freesp() */
1581 			zfs_dirent_unlock(dl);
1582 			dl = NULL;
1583 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1584 			if (error == 0) {
1585 				vnevent_create(ZTOV(zp), ct);
1586 			}
1587 		}
1588 	}
1589 out:
1590 
1591 	if (dl)
1592 		zfs_dirent_unlock(dl);
1593 
1594 	if (error) {
1595 		if (zp)
1596 			VN_RELE(ZTOV(zp));
1597 	} else {
1598 		*vpp = ZTOV(zp);
1599 		error = specvp_check(vpp, cr);
1600 	}
1601 
1602 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1603 		zil_commit(zilog, 0);
1604 
1605 	ZFS_EXIT(zfsvfs);
1606 	return (error);
1607 }
1608 
1609 /*
1610  * Remove an entry from a directory.
1611  *
1612  *	IN:	dvp	- vnode of directory to remove entry from.
1613  *		name	- name of entry to remove.
1614  *		cr	- credentials of caller.
1615  *		ct	- caller context
1616  *		flags	- case flags
1617  *
1618  *	RETURN:	0 on success, error code on failure.
1619  *
1620  * Timestamps:
1621  *	dvp - ctime|mtime
1622  *	 vp - ctime (if nlink > 0)
1623  */
1624 
1625 uint64_t null_xattr = 0;
1626 
1627 /*ARGSUSED*/
1628 static int
1629 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1630     int flags)
1631 {
1632 	znode_t		*zp, *dzp = VTOZ(dvp);
1633 	znode_t		*xzp;
1634 	vnode_t		*vp;
1635 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1636 	zilog_t		*zilog;
1637 	uint64_t	acl_obj, xattr_obj;
1638 	uint64_t	xattr_obj_unlinked = 0;
1639 	uint64_t	obj = 0;
1640 	zfs_dirlock_t	*dl;
1641 	dmu_tx_t	*tx;
1642 	boolean_t	may_delete_now, delete_now = FALSE;
1643 	boolean_t	unlinked, toobig = FALSE;
1644 	uint64_t	txtype;
1645 	pathname_t	*realnmp = NULL;
1646 	pathname_t	realnm;
1647 	int		error;
1648 	int		zflg = ZEXISTS;
1649 	boolean_t	waited = B_FALSE;
1650 
1651 	ZFS_ENTER(zfsvfs);
1652 	ZFS_VERIFY_ZP(dzp);
1653 	zilog = zfsvfs->z_log;
1654 
1655 	if (flags & FIGNORECASE) {
1656 		zflg |= ZCILOOK;
1657 		pn_alloc(&realnm);
1658 		realnmp = &realnm;
1659 	}
1660 
1661 top:
1662 	xattr_obj = 0;
1663 	xzp = NULL;
1664 	/*
1665 	 * Attempt to lock directory; fail if entry doesn't exist.
1666 	 */
1667 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1668 	    NULL, realnmp)) {
1669 		if (realnmp)
1670 			pn_free(realnmp);
1671 		ZFS_EXIT(zfsvfs);
1672 		return (error);
1673 	}
1674 
1675 	vp = ZTOV(zp);
1676 
1677 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1678 		goto out;
1679 	}
1680 
1681 	/*
1682 	 * Need to use rmdir for removing directories.
1683 	 */
1684 	if (vp->v_type == VDIR) {
1685 		error = SET_ERROR(EPERM);
1686 		goto out;
1687 	}
1688 
1689 	vnevent_remove(vp, dvp, name, ct);
1690 
1691 	if (realnmp)
1692 		dnlc_remove(dvp, realnmp->pn_buf);
1693 	else
1694 		dnlc_remove(dvp, name);
1695 
1696 	mutex_enter(&vp->v_lock);
1697 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1698 	mutex_exit(&vp->v_lock);
1699 
1700 	/*
1701 	 * We may delete the znode now, or we may put it in the unlinked set;
1702 	 * it depends on whether we're the last link, and on whether there are
1703 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1704 	 * allow for either case.
1705 	 */
1706 	obj = zp->z_id;
1707 	tx = dmu_tx_create(zfsvfs->z_os);
1708 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1709 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1710 	zfs_sa_upgrade_txholds(tx, zp);
1711 	zfs_sa_upgrade_txholds(tx, dzp);
1712 	if (may_delete_now) {
1713 		toobig =
1714 		    zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1715 		/* if the file is too big, only hold_free a token amount */
1716 		dmu_tx_hold_free(tx, zp->z_id, 0,
1717 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1718 	}
1719 
1720 	/* are there any extended attributes? */
1721 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1722 	    &xattr_obj, sizeof (xattr_obj));
1723 	if (error == 0 && xattr_obj) {
1724 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1725 		ASSERT0(error);
1726 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1727 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1728 	}
1729 
1730 	mutex_enter(&zp->z_lock);
1731 	if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1732 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1733 	mutex_exit(&zp->z_lock);
1734 
1735 	/* charge as an update -- would be nice not to charge at all */
1736 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1737 
1738 	/*
1739 	 * Mark this transaction as typically resulting in a net free of space
1740 	 */
1741 	dmu_tx_mark_netfree(tx);
1742 
1743 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1744 	if (error) {
1745 		zfs_dirent_unlock(dl);
1746 		VN_RELE(vp);
1747 		if (xzp)
1748 			VN_RELE(ZTOV(xzp));
1749 		if (error == ERESTART) {
1750 			waited = B_TRUE;
1751 			dmu_tx_wait(tx);
1752 			dmu_tx_abort(tx);
1753 			goto top;
1754 		}
1755 		if (realnmp)
1756 			pn_free(realnmp);
1757 		dmu_tx_abort(tx);
1758 		ZFS_EXIT(zfsvfs);
1759 		return (error);
1760 	}
1761 
1762 	/*
1763 	 * Remove the directory entry.
1764 	 */
1765 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1766 
1767 	if (error) {
1768 		dmu_tx_commit(tx);
1769 		goto out;
1770 	}
1771 
1772 	if (unlinked) {
1773 		/*
1774 		 * Hold z_lock so that we can make sure that the ACL obj
1775 		 * hasn't changed.  Could have been deleted due to
1776 		 * zfs_sa_upgrade().
1777 		 */
1778 		mutex_enter(&zp->z_lock);
1779 		mutex_enter(&vp->v_lock);
1780 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1781 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1782 		delete_now = may_delete_now && !toobig &&
1783 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1784 		    xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1785 		    acl_obj;
1786 		mutex_exit(&vp->v_lock);
1787 	}
1788 
1789 	if (delete_now) {
1790 		if (xattr_obj_unlinked) {
1791 			ASSERT3U(xzp->z_links, ==, 2);
1792 			mutex_enter(&xzp->z_lock);
1793 			xzp->z_unlinked = 1;
1794 			xzp->z_links = 0;
1795 			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1796 			    &xzp->z_links, sizeof (xzp->z_links), tx);
1797 			ASSERT3U(error,  ==,  0);
1798 			mutex_exit(&xzp->z_lock);
1799 			zfs_unlinked_add(xzp, tx);
1800 
1801 			if (zp->z_is_sa)
1802 				error = sa_remove(zp->z_sa_hdl,
1803 				    SA_ZPL_XATTR(zfsvfs), tx);
1804 			else
1805 				error = sa_update(zp->z_sa_hdl,
1806 				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
1807 				    sizeof (uint64_t), tx);
1808 			ASSERT0(error);
1809 		}
1810 		mutex_enter(&vp->v_lock);
1811 		vp->v_count--;
1812 		ASSERT0(vp->v_count);
1813 		mutex_exit(&vp->v_lock);
1814 		mutex_exit(&zp->z_lock);
1815 		zfs_znode_delete(zp, tx);
1816 	} else if (unlinked) {
1817 		mutex_exit(&zp->z_lock);
1818 		zfs_unlinked_add(zp, tx);
1819 	}
1820 
1821 	txtype = TX_REMOVE;
1822 	if (flags & FIGNORECASE)
1823 		txtype |= TX_CI;
1824 	zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1825 
1826 	dmu_tx_commit(tx);
1827 out:
1828 	if (realnmp)
1829 		pn_free(realnmp);
1830 
1831 	zfs_dirent_unlock(dl);
1832 
1833 	if (!delete_now)
1834 		VN_RELE(vp);
1835 	if (xzp)
1836 		VN_RELE(ZTOV(xzp));
1837 
1838 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1839 		zil_commit(zilog, 0);
1840 
1841 	ZFS_EXIT(zfsvfs);
1842 	return (error);
1843 }
1844 
1845 /*
1846  * Create a new directory and insert it into dvp using the name
1847  * provided.  Return a pointer to the inserted directory.
1848  *
1849  *	IN:	dvp	- vnode of directory to add subdir to.
1850  *		dirname	- name of new directory.
1851  *		vap	- attributes of new directory.
1852  *		cr	- credentials of caller.
1853  *		ct	- caller context
1854  *		flags	- case flags
1855  *		vsecp	- ACL to be set
1856  *
1857  *	OUT:	vpp	- vnode of created directory.
1858  *
1859  *	RETURN:	0 on success, error code on failure.
1860  *
1861  * Timestamps:
1862  *	dvp - ctime|mtime updated
1863  *	 vp - ctime|mtime|atime updated
1864  */
1865 /*ARGSUSED*/
1866 static int
1867 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1868     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1869 {
1870 	znode_t		*zp, *dzp = VTOZ(dvp);
1871 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1872 	zilog_t		*zilog;
1873 	zfs_dirlock_t	*dl;
1874 	uint64_t	txtype;
1875 	dmu_tx_t	*tx;
1876 	int		error;
1877 	int		zf = ZNEW;
1878 	ksid_t		*ksid;
1879 	uid_t		uid;
1880 	gid_t		gid = crgetgid(cr);
1881 	zfs_acl_ids_t   acl_ids;
1882 	boolean_t	fuid_dirtied;
1883 	boolean_t	waited = B_FALSE;
1884 
1885 	ASSERT(vap->va_type == VDIR);
1886 
1887 	/*
1888 	 * If we have an ephemeral id, ACL, or XVATTR then
1889 	 * make sure file system is at proper version
1890 	 */
1891 
1892 	ksid = crgetsid(cr, KSID_OWNER);
1893 	if (ksid)
1894 		uid = ksid_getid(ksid);
1895 	else
1896 		uid = crgetuid(cr);
1897 	if (zfsvfs->z_use_fuids == B_FALSE &&
1898 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1899 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1900 		return (SET_ERROR(EINVAL));
1901 
1902 	ZFS_ENTER(zfsvfs);
1903 	ZFS_VERIFY_ZP(dzp);
1904 	zilog = zfsvfs->z_log;
1905 
1906 	if (dzp->z_pflags & ZFS_XATTR) {
1907 		ZFS_EXIT(zfsvfs);
1908 		return (SET_ERROR(EINVAL));
1909 	}
1910 
1911 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1912 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1913 		ZFS_EXIT(zfsvfs);
1914 		return (SET_ERROR(EILSEQ));
1915 	}
1916 	if (flags & FIGNORECASE)
1917 		zf |= ZCILOOK;
1918 
1919 	if (vap->va_mask & AT_XVATTR) {
1920 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1921 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1922 			ZFS_EXIT(zfsvfs);
1923 			return (error);
1924 		}
1925 	}
1926 
1927 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1928 	    vsecp, &acl_ids)) != 0) {
1929 		ZFS_EXIT(zfsvfs);
1930 		return (error);
1931 	}
1932 	/*
1933 	 * First make sure the new directory doesn't exist.
1934 	 *
1935 	 * Existence is checked first to make sure we don't return
1936 	 * EACCES instead of EEXIST which can cause some applications
1937 	 * to fail.
1938 	 */
1939 top:
1940 	*vpp = NULL;
1941 
1942 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1943 	    NULL, NULL)) {
1944 		zfs_acl_ids_free(&acl_ids);
1945 		ZFS_EXIT(zfsvfs);
1946 		return (error);
1947 	}
1948 
1949 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1950 		zfs_acl_ids_free(&acl_ids);
1951 		zfs_dirent_unlock(dl);
1952 		ZFS_EXIT(zfsvfs);
1953 		return (error);
1954 	}
1955 
1956 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1957 		zfs_acl_ids_free(&acl_ids);
1958 		zfs_dirent_unlock(dl);
1959 		ZFS_EXIT(zfsvfs);
1960 		return (SET_ERROR(EDQUOT));
1961 	}
1962 
1963 	/*
1964 	 * Add a new entry to the directory.
1965 	 */
1966 	tx = dmu_tx_create(zfsvfs->z_os);
1967 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1968 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1969 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1970 	if (fuid_dirtied)
1971 		zfs_fuid_txhold(zfsvfs, tx);
1972 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1973 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1974 		    acl_ids.z_aclp->z_acl_bytes);
1975 	}
1976 
1977 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1978 	    ZFS_SA_BASE_ATTR_SIZE);
1979 
1980 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1981 	if (error) {
1982 		zfs_dirent_unlock(dl);
1983 		if (error == ERESTART) {
1984 			waited = B_TRUE;
1985 			dmu_tx_wait(tx);
1986 			dmu_tx_abort(tx);
1987 			goto top;
1988 		}
1989 		zfs_acl_ids_free(&acl_ids);
1990 		dmu_tx_abort(tx);
1991 		ZFS_EXIT(zfsvfs);
1992 		return (error);
1993 	}
1994 
1995 	/*
1996 	 * Create new node.
1997 	 */
1998 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1999 
2000 	if (fuid_dirtied)
2001 		zfs_fuid_sync(zfsvfs, tx);
2002 
2003 	/*
2004 	 * Now put new name in parent dir.
2005 	 */
2006 	(void) zfs_link_create(dl, zp, tx, ZNEW);
2007 
2008 	*vpp = ZTOV(zp);
2009 
2010 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2011 	if (flags & FIGNORECASE)
2012 		txtype |= TX_CI;
2013 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2014 	    acl_ids.z_fuidp, vap);
2015 
2016 	zfs_acl_ids_free(&acl_ids);
2017 
2018 	dmu_tx_commit(tx);
2019 
2020 	zfs_dirent_unlock(dl);
2021 
2022 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2023 		zil_commit(zilog, 0);
2024 
2025 	ZFS_EXIT(zfsvfs);
2026 	return (0);
2027 }
2028 
2029 /*
2030  * Remove a directory subdir entry.  If the current working
2031  * directory is the same as the subdir to be removed, the
2032  * remove will fail.
2033  *
2034  *	IN:	dvp	- vnode of directory to remove from.
2035  *		name	- name of directory to be removed.
2036  *		cwd	- vnode of current working directory.
2037  *		cr	- credentials of caller.
2038  *		ct	- caller context
2039  *		flags	- case flags
2040  *
2041  *	RETURN:	0 on success, error code on failure.
2042  *
2043  * Timestamps:
2044  *	dvp - ctime|mtime updated
2045  */
2046 /*ARGSUSED*/
2047 static int
2048 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2049     caller_context_t *ct, int flags)
2050 {
2051 	znode_t		*dzp = VTOZ(dvp);
2052 	znode_t		*zp;
2053 	vnode_t		*vp;
2054 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2055 	zilog_t		*zilog;
2056 	zfs_dirlock_t	*dl;
2057 	dmu_tx_t	*tx;
2058 	int		error;
2059 	int		zflg = ZEXISTS;
2060 	boolean_t	waited = B_FALSE;
2061 
2062 	ZFS_ENTER(zfsvfs);
2063 	ZFS_VERIFY_ZP(dzp);
2064 	zilog = zfsvfs->z_log;
2065 
2066 	if (flags & FIGNORECASE)
2067 		zflg |= ZCILOOK;
2068 top:
2069 	zp = NULL;
2070 
2071 	/*
2072 	 * Attempt to lock directory; fail if entry doesn't exist.
2073 	 */
2074 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2075 	    NULL, NULL)) {
2076 		ZFS_EXIT(zfsvfs);
2077 		return (error);
2078 	}
2079 
2080 	vp = ZTOV(zp);
2081 
2082 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2083 		goto out;
2084 	}
2085 
2086 	if (vp->v_type != VDIR) {
2087 		error = SET_ERROR(ENOTDIR);
2088 		goto out;
2089 	}
2090 
2091 	if (vp == cwd) {
2092 		error = SET_ERROR(EINVAL);
2093 		goto out;
2094 	}
2095 
2096 	vnevent_rmdir(vp, dvp, name, ct);
2097 
2098 	/*
2099 	 * Grab a lock on the directory to make sure that noone is
2100 	 * trying to add (or lookup) entries while we are removing it.
2101 	 */
2102 	rw_enter(&zp->z_name_lock, RW_WRITER);
2103 
2104 	/*
2105 	 * Grab a lock on the parent pointer to make sure we play well
2106 	 * with the treewalk and directory rename code.
2107 	 */
2108 	rw_enter(&zp->z_parent_lock, RW_WRITER);
2109 
2110 	tx = dmu_tx_create(zfsvfs->z_os);
2111 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2112 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2113 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2114 	zfs_sa_upgrade_txholds(tx, zp);
2115 	zfs_sa_upgrade_txholds(tx, dzp);
2116 	dmu_tx_mark_netfree(tx);
2117 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2118 	if (error) {
2119 		rw_exit(&zp->z_parent_lock);
2120 		rw_exit(&zp->z_name_lock);
2121 		zfs_dirent_unlock(dl);
2122 		VN_RELE(vp);
2123 		if (error == ERESTART) {
2124 			waited = B_TRUE;
2125 			dmu_tx_wait(tx);
2126 			dmu_tx_abort(tx);
2127 			goto top;
2128 		}
2129 		dmu_tx_abort(tx);
2130 		ZFS_EXIT(zfsvfs);
2131 		return (error);
2132 	}
2133 
2134 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2135 
2136 	if (error == 0) {
2137 		uint64_t txtype = TX_RMDIR;
2138 		if (flags & FIGNORECASE)
2139 			txtype |= TX_CI;
2140 		zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2141 	}
2142 
2143 	dmu_tx_commit(tx);
2144 
2145 	rw_exit(&zp->z_parent_lock);
2146 	rw_exit(&zp->z_name_lock);
2147 out:
2148 	zfs_dirent_unlock(dl);
2149 
2150 	VN_RELE(vp);
2151 
2152 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2153 		zil_commit(zilog, 0);
2154 
2155 	ZFS_EXIT(zfsvfs);
2156 	return (error);
2157 }
2158 
2159 /*
2160  * Read as many directory entries as will fit into the provided
2161  * buffer from the given directory cursor position (specified in
2162  * the uio structure).
2163  *
2164  *	IN:	vp	- vnode of directory to read.
2165  *		uio	- structure supplying read location, range info,
2166  *			  and return buffer.
2167  *		cr	- credentials of caller.
2168  *		ct	- caller context
2169  *		flags	- case flags
2170  *
2171  *	OUT:	uio	- updated offset and range, buffer filled.
2172  *		eofp	- set to true if end-of-file detected.
2173  *
2174  *	RETURN:	0 on success, error code on failure.
2175  *
2176  * Timestamps:
2177  *	vp - atime updated
2178  *
2179  * Note that the low 4 bits of the cookie returned by zap is always zero.
2180  * This allows us to use the low range for "special" directory entries:
2181  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2182  * we use the offset 2 for the '.zfs' directory.
2183  */
2184 /* ARGSUSED */
2185 static int
2186 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2187     caller_context_t *ct, int flags)
2188 {
2189 	znode_t		*zp = VTOZ(vp);
2190 	iovec_t		*iovp;
2191 	edirent_t	*eodp;
2192 	dirent64_t	*odp;
2193 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2194 	objset_t	*os;
2195 	caddr_t		outbuf;
2196 	size_t		bufsize;
2197 	zap_cursor_t	zc;
2198 	zap_attribute_t	zap;
2199 	uint_t		bytes_wanted;
2200 	uint64_t	offset; /* must be unsigned; checks for < 1 */
2201 	uint64_t	parent;
2202 	int		local_eof;
2203 	int		outcount;
2204 	int		error;
2205 	uint8_t		prefetch;
2206 	boolean_t	check_sysattrs;
2207 
2208 	ZFS_ENTER(zfsvfs);
2209 	ZFS_VERIFY_ZP(zp);
2210 
2211 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2212 	    &parent, sizeof (parent))) != 0) {
2213 		ZFS_EXIT(zfsvfs);
2214 		return (error);
2215 	}
2216 
2217 	/*
2218 	 * If we are not given an eof variable,
2219 	 * use a local one.
2220 	 */
2221 	if (eofp == NULL)
2222 		eofp = &local_eof;
2223 
2224 	/*
2225 	 * Check for valid iov_len.
2226 	 */
2227 	if (uio->uio_iov->iov_len <= 0) {
2228 		ZFS_EXIT(zfsvfs);
2229 		return (SET_ERROR(EINVAL));
2230 	}
2231 
2232 	/*
2233 	 * Quit if directory has been removed (posix)
2234 	 */
2235 	if ((*eofp = zp->z_unlinked) != 0) {
2236 		ZFS_EXIT(zfsvfs);
2237 		return (0);
2238 	}
2239 
2240 	error = 0;
2241 	os = zfsvfs->z_os;
2242 	offset = uio->uio_loffset;
2243 	prefetch = zp->z_zn_prefetch;
2244 
2245 	/*
2246 	 * Initialize the iterator cursor.
2247 	 */
2248 	if (offset <= 3) {
2249 		/*
2250 		 * Start iteration from the beginning of the directory.
2251 		 */
2252 		zap_cursor_init(&zc, os, zp->z_id);
2253 	} else {
2254 		/*
2255 		 * The offset is a serialized cursor.
2256 		 */
2257 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2258 	}
2259 
2260 	/*
2261 	 * Get space to change directory entries into fs independent format.
2262 	 */
2263 	iovp = uio->uio_iov;
2264 	bytes_wanted = iovp->iov_len;
2265 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2266 		bufsize = bytes_wanted;
2267 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
2268 		odp = (struct dirent64 *)outbuf;
2269 	} else {
2270 		bufsize = bytes_wanted;
2271 		outbuf = NULL;
2272 		odp = (struct dirent64 *)iovp->iov_base;
2273 	}
2274 	eodp = (struct edirent *)odp;
2275 
2276 	/*
2277 	 * If this VFS supports the system attribute view interface; and
2278 	 * we're looking at an extended attribute directory; and we care
2279 	 * about normalization conflicts on this vfs; then we must check
2280 	 * for normalization conflicts with the sysattr name space.
2281 	 */
2282 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2283 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2284 	    (flags & V_RDDIR_ENTFLAGS);
2285 
2286 	/*
2287 	 * Transform to file-system independent format
2288 	 */
2289 	outcount = 0;
2290 	while (outcount < bytes_wanted) {
2291 		ino64_t objnum;
2292 		ushort_t reclen;
2293 		off64_t *next = NULL;
2294 
2295 		/*
2296 		 * Special case `.', `..', and `.zfs'.
2297 		 */
2298 		if (offset == 0) {
2299 			(void) strcpy(zap.za_name, ".");
2300 			zap.za_normalization_conflict = 0;
2301 			objnum = zp->z_id;
2302 		} else if (offset == 1) {
2303 			(void) strcpy(zap.za_name, "..");
2304 			zap.za_normalization_conflict = 0;
2305 			objnum = parent;
2306 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2307 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2308 			zap.za_normalization_conflict = 0;
2309 			objnum = ZFSCTL_INO_ROOT;
2310 		} else {
2311 			/*
2312 			 * Grab next entry.
2313 			 */
2314 			if (error = zap_cursor_retrieve(&zc, &zap)) {
2315 				if ((*eofp = (error == ENOENT)) != 0)
2316 					break;
2317 				else
2318 					goto update;
2319 			}
2320 
2321 			if (zap.za_integer_length != 8 ||
2322 			    zap.za_num_integers != 1) {
2323 				cmn_err(CE_WARN, "zap_readdir: bad directory "
2324 				    "entry, obj = %lld, offset = %lld\n",
2325 				    (u_longlong_t)zp->z_id,
2326 				    (u_longlong_t)offset);
2327 				error = SET_ERROR(ENXIO);
2328 				goto update;
2329 			}
2330 
2331 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2332 			/*
2333 			 * MacOS X can extract the object type here such as:
2334 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2335 			 */
2336 
2337 			if (check_sysattrs && !zap.za_normalization_conflict) {
2338 				zap.za_normalization_conflict =
2339 				    xattr_sysattr_casechk(zap.za_name);
2340 			}
2341 		}
2342 
2343 		if (flags & V_RDDIR_ACCFILTER) {
2344 			/*
2345 			 * If we have no access at all, don't include
2346 			 * this entry in the returned information
2347 			 */
2348 			znode_t	*ezp;
2349 			if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2350 				goto skip_entry;
2351 			if (!zfs_has_access(ezp, cr)) {
2352 				VN_RELE(ZTOV(ezp));
2353 				goto skip_entry;
2354 			}
2355 			VN_RELE(ZTOV(ezp));
2356 		}
2357 
2358 		if (flags & V_RDDIR_ENTFLAGS)
2359 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2360 		else
2361 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2362 
2363 		/*
2364 		 * Will this entry fit in the buffer?
2365 		 */
2366 		if (outcount + reclen > bufsize) {
2367 			/*
2368 			 * Did we manage to fit anything in the buffer?
2369 			 */
2370 			if (!outcount) {
2371 				error = SET_ERROR(EINVAL);
2372 				goto update;
2373 			}
2374 			break;
2375 		}
2376 		if (flags & V_RDDIR_ENTFLAGS) {
2377 			/*
2378 			 * Add extended flag entry:
2379 			 */
2380 			eodp->ed_ino = objnum;
2381 			eodp->ed_reclen = reclen;
2382 			/* NOTE: ed_off is the offset for the *next* entry */
2383 			next = &(eodp->ed_off);
2384 			eodp->ed_eflags = zap.za_normalization_conflict ?
2385 			    ED_CASE_CONFLICT : 0;
2386 			(void) strncpy(eodp->ed_name, zap.za_name,
2387 			    EDIRENT_NAMELEN(reclen));
2388 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2389 		} else {
2390 			/*
2391 			 * Add normal entry:
2392 			 */
2393 			odp->d_ino = objnum;
2394 			odp->d_reclen = reclen;
2395 			/* NOTE: d_off is the offset for the *next* entry */
2396 			next = &(odp->d_off);
2397 			(void) strncpy(odp->d_name, zap.za_name,
2398 			    DIRENT64_NAMELEN(reclen));
2399 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2400 		}
2401 		outcount += reclen;
2402 
2403 		ASSERT(outcount <= bufsize);
2404 
2405 		/* Prefetch znode */
2406 		if (prefetch)
2407 			dmu_prefetch(os, objnum, 0, 0, 0,
2408 			    ZIO_PRIORITY_SYNC_READ);
2409 
2410 	skip_entry:
2411 		/*
2412 		 * Move to the next entry, fill in the previous offset.
2413 		 */
2414 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2415 			zap_cursor_advance(&zc);
2416 			offset = zap_cursor_serialize(&zc);
2417 		} else {
2418 			offset += 1;
2419 		}
2420 		if (next)
2421 			*next = offset;
2422 	}
2423 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2424 
2425 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2426 		iovp->iov_base += outcount;
2427 		iovp->iov_len -= outcount;
2428 		uio->uio_resid -= outcount;
2429 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2430 		/*
2431 		 * Reset the pointer.
2432 		 */
2433 		offset = uio->uio_loffset;
2434 	}
2435 
2436 update:
2437 	zap_cursor_fini(&zc);
2438 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2439 		kmem_free(outbuf, bufsize);
2440 
2441 	if (error == ENOENT)
2442 		error = 0;
2443 
2444 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2445 
2446 	uio->uio_loffset = offset;
2447 	ZFS_EXIT(zfsvfs);
2448 	return (error);
2449 }
2450 
2451 ulong_t zfs_fsync_sync_cnt = 4;
2452 
2453 static int
2454 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2455 {
2456 	znode_t	*zp = VTOZ(vp);
2457 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2458 
2459 	/*
2460 	 * Regardless of whether this is required for standards conformance,
2461 	 * this is the logical behavior when fsync() is called on a file with
2462 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2463 	 * going to be pushed out as part of the zil_commit().
2464 	 */
2465 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2466 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2467 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2468 
2469 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2470 
2471 	if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2472 		ZFS_ENTER(zfsvfs);
2473 		ZFS_VERIFY_ZP(zp);
2474 		zil_commit(zfsvfs->z_log, zp->z_id);
2475 		ZFS_EXIT(zfsvfs);
2476 	}
2477 	return (0);
2478 }
2479 
2480 
2481 /*
2482  * Get the requested file attributes and place them in the provided
2483  * vattr structure.
2484  *
2485  *	IN:	vp	- vnode of file.
2486  *		vap	- va_mask identifies requested attributes.
2487  *			  If AT_XVATTR set, then optional attrs are requested
2488  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2489  *		cr	- credentials of caller.
2490  *		ct	- caller context
2491  *
2492  *	OUT:	vap	- attribute values.
2493  *
2494  *	RETURN:	0 (always succeeds).
2495  */
2496 /* ARGSUSED */
2497 static int
2498 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2499     caller_context_t *ct)
2500 {
2501 	znode_t *zp = VTOZ(vp);
2502 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2503 	int	error = 0;
2504 	uint64_t links;
2505 	uint64_t mtime[2], ctime[2];
2506 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2507 	xoptattr_t *xoap = NULL;
2508 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2509 	sa_bulk_attr_t bulk[2];
2510 	int count = 0;
2511 
2512 	ZFS_ENTER(zfsvfs);
2513 	ZFS_VERIFY_ZP(zp);
2514 
2515 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2516 
2517 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2518 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2519 
2520 	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2521 		ZFS_EXIT(zfsvfs);
2522 		return (error);
2523 	}
2524 
2525 	/*
2526 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2527 	 * Also, if we are the owner don't bother, since owner should
2528 	 * always be allowed to read basic attributes of file.
2529 	 */
2530 	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2531 	    (vap->va_uid != crgetuid(cr))) {
2532 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2533 		    skipaclchk, cr)) {
2534 			ZFS_EXIT(zfsvfs);
2535 			return (error);
2536 		}
2537 	}
2538 
2539 	/*
2540 	 * Return all attributes.  It's cheaper to provide the answer
2541 	 * than to determine whether we were asked the question.
2542 	 */
2543 
2544 	mutex_enter(&zp->z_lock);
2545 	vap->va_type = vp->v_type;
2546 	vap->va_mode = zp->z_mode & MODEMASK;
2547 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2548 	vap->va_nodeid = zp->z_id;
2549 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2550 		links = zp->z_links + 1;
2551 	else
2552 		links = zp->z_links;
2553 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2554 	vap->va_size = zp->z_size;
2555 	vap->va_rdev = vp->v_rdev;
2556 	vap->va_seq = zp->z_seq;
2557 
2558 	/*
2559 	 * Add in any requested optional attributes and the create time.
2560 	 * Also set the corresponding bits in the returned attribute bitmap.
2561 	 */
2562 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2563 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2564 			xoap->xoa_archive =
2565 			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2566 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2567 		}
2568 
2569 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2570 			xoap->xoa_readonly =
2571 			    ((zp->z_pflags & ZFS_READONLY) != 0);
2572 			XVA_SET_RTN(xvap, XAT_READONLY);
2573 		}
2574 
2575 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2576 			xoap->xoa_system =
2577 			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2578 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2579 		}
2580 
2581 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2582 			xoap->xoa_hidden =
2583 			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2584 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2585 		}
2586 
2587 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2588 			xoap->xoa_nounlink =
2589 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2590 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2591 		}
2592 
2593 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2594 			xoap->xoa_immutable =
2595 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2596 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2597 		}
2598 
2599 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2600 			xoap->xoa_appendonly =
2601 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2602 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2603 		}
2604 
2605 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2606 			xoap->xoa_nodump =
2607 			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2608 			XVA_SET_RTN(xvap, XAT_NODUMP);
2609 		}
2610 
2611 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2612 			xoap->xoa_opaque =
2613 			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2614 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2615 		}
2616 
2617 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2618 			xoap->xoa_av_quarantined =
2619 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2620 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2621 		}
2622 
2623 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2624 			xoap->xoa_av_modified =
2625 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2626 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2627 		}
2628 
2629 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2630 		    vp->v_type == VREG) {
2631 			zfs_sa_get_scanstamp(zp, xvap);
2632 		}
2633 
2634 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2635 			uint64_t times[2];
2636 
2637 			(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2638 			    times, sizeof (times));
2639 			ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2640 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2641 		}
2642 
2643 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2644 			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2645 			XVA_SET_RTN(xvap, XAT_REPARSE);
2646 		}
2647 		if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2648 			xoap->xoa_generation = zp->z_gen;
2649 			XVA_SET_RTN(xvap, XAT_GEN);
2650 		}
2651 
2652 		if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2653 			xoap->xoa_offline =
2654 			    ((zp->z_pflags & ZFS_OFFLINE) != 0);
2655 			XVA_SET_RTN(xvap, XAT_OFFLINE);
2656 		}
2657 
2658 		if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2659 			xoap->xoa_sparse =
2660 			    ((zp->z_pflags & ZFS_SPARSE) != 0);
2661 			XVA_SET_RTN(xvap, XAT_SPARSE);
2662 		}
2663 	}
2664 
2665 	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2666 	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2667 	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2668 
2669 	mutex_exit(&zp->z_lock);
2670 
2671 	sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2672 
2673 	if (zp->z_blksz == 0) {
2674 		/*
2675 		 * Block size hasn't been set; suggest maximal I/O transfers.
2676 		 */
2677 		vap->va_blksize = zfsvfs->z_max_blksz;
2678 	}
2679 
2680 	ZFS_EXIT(zfsvfs);
2681 	return (0);
2682 }
2683 
2684 /*
2685  * Set the file attributes to the values contained in the
2686  * vattr structure.
2687  *
2688  *	IN:	vp	- vnode of file to be modified.
2689  *		vap	- new attribute values.
2690  *			  If AT_XVATTR set, then optional attrs are being set
2691  *		flags	- ATTR_UTIME set if non-default time values provided.
2692  *			- ATTR_NOACLCHECK (CIFS context only).
2693  *		cr	- credentials of caller.
2694  *		ct	- caller context
2695  *
2696  *	RETURN:	0 on success, error code on failure.
2697  *
2698  * Timestamps:
2699  *	vp - ctime updated, mtime updated if size changed.
2700  */
2701 /* ARGSUSED */
2702 static int
2703 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2704     caller_context_t *ct)
2705 {
2706 	znode_t		*zp = VTOZ(vp);
2707 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2708 	zilog_t		*zilog;
2709 	dmu_tx_t	*tx;
2710 	vattr_t		oldva;
2711 	xvattr_t	tmpxvattr;
2712 	uint_t		mask = vap->va_mask;
2713 	uint_t		saved_mask = 0;
2714 	int		trim_mask = 0;
2715 	uint64_t	new_mode;
2716 	uint64_t	new_uid, new_gid;
2717 	uint64_t	xattr_obj;
2718 	uint64_t	mtime[2], ctime[2];
2719 	znode_t		*attrzp;
2720 	int		need_policy = FALSE;
2721 	int		err, err2;
2722 	zfs_fuid_info_t *fuidp = NULL;
2723 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2724 	xoptattr_t	*xoap;
2725 	zfs_acl_t	*aclp;
2726 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2727 	boolean_t	fuid_dirtied = B_FALSE;
2728 	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
2729 	int		count = 0, xattr_count = 0;
2730 
2731 	if (mask == 0)
2732 		return (0);
2733 
2734 	if (mask & AT_NOSET)
2735 		return (SET_ERROR(EINVAL));
2736 
2737 	ZFS_ENTER(zfsvfs);
2738 	ZFS_VERIFY_ZP(zp);
2739 
2740 	zilog = zfsvfs->z_log;
2741 
2742 	/*
2743 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2744 	 * that file system is at proper version level
2745 	 */
2746 
2747 	if (zfsvfs->z_use_fuids == B_FALSE &&
2748 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2749 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2750 	    (mask & AT_XVATTR))) {
2751 		ZFS_EXIT(zfsvfs);
2752 		return (SET_ERROR(EINVAL));
2753 	}
2754 
2755 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2756 		ZFS_EXIT(zfsvfs);
2757 		return (SET_ERROR(EISDIR));
2758 	}
2759 
2760 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2761 		ZFS_EXIT(zfsvfs);
2762 		return (SET_ERROR(EINVAL));
2763 	}
2764 
2765 	/*
2766 	 * If this is an xvattr_t, then get a pointer to the structure of
2767 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2768 	 */
2769 	xoap = xva_getxoptattr(xvap);
2770 
2771 	xva_init(&tmpxvattr);
2772 
2773 	/*
2774 	 * Immutable files can only alter immutable bit and atime
2775 	 */
2776 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2777 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2778 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2779 		ZFS_EXIT(zfsvfs);
2780 		return (SET_ERROR(EPERM));
2781 	}
2782 
2783 	if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2784 		ZFS_EXIT(zfsvfs);
2785 		return (SET_ERROR(EPERM));
2786 	}
2787 
2788 	/*
2789 	 * Verify timestamps doesn't overflow 32 bits.
2790 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2791 	 * handle times greater than 2039.  This check should be removed
2792 	 * once large timestamps are fully supported.
2793 	 */
2794 	if (mask & (AT_ATIME | AT_MTIME)) {
2795 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2796 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2797 			ZFS_EXIT(zfsvfs);
2798 			return (SET_ERROR(EOVERFLOW));
2799 		}
2800 	}
2801 
2802 top:
2803 	attrzp = NULL;
2804 	aclp = NULL;
2805 
2806 	/* Can this be moved to before the top label? */
2807 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2808 		ZFS_EXIT(zfsvfs);
2809 		return (SET_ERROR(EROFS));
2810 	}
2811 
2812 	/*
2813 	 * First validate permissions
2814 	 */
2815 
2816 	if (mask & AT_SIZE) {
2817 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2818 		if (err) {
2819 			ZFS_EXIT(zfsvfs);
2820 			return (err);
2821 		}
2822 		/*
2823 		 * XXX - Note, we are not providing any open
2824 		 * mode flags here (like FNDELAY), so we may
2825 		 * block if there are locks present... this
2826 		 * should be addressed in openat().
2827 		 */
2828 		/* XXX - would it be OK to generate a log record here? */
2829 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2830 		if (err) {
2831 			ZFS_EXIT(zfsvfs);
2832 			return (err);
2833 		}
2834 
2835 		if (vap->va_size == 0)
2836 			vnevent_truncate(ZTOV(zp), ct);
2837 	}
2838 
2839 	if (mask & (AT_ATIME|AT_MTIME) ||
2840 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2841 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2842 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2843 	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2844 	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2845 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2846 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2847 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2848 		    skipaclchk, cr);
2849 	}
2850 
2851 	if (mask & (AT_UID|AT_GID)) {
2852 		int	idmask = (mask & (AT_UID|AT_GID));
2853 		int	take_owner;
2854 		int	take_group;
2855 
2856 		/*
2857 		 * NOTE: even if a new mode is being set,
2858 		 * we may clear S_ISUID/S_ISGID bits.
2859 		 */
2860 
2861 		if (!(mask & AT_MODE))
2862 			vap->va_mode = zp->z_mode;
2863 
2864 		/*
2865 		 * Take ownership or chgrp to group we are a member of
2866 		 */
2867 
2868 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2869 		take_group = (mask & AT_GID) &&
2870 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2871 
2872 		/*
2873 		 * If both AT_UID and AT_GID are set then take_owner and
2874 		 * take_group must both be set in order to allow taking
2875 		 * ownership.
2876 		 *
2877 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2878 		 *
2879 		 */
2880 
2881 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2882 		    ((idmask == AT_UID) && take_owner) ||
2883 		    ((idmask == AT_GID) && take_group)) {
2884 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2885 			    skipaclchk, cr) == 0) {
2886 				/*
2887 				 * Remove setuid/setgid for non-privileged users
2888 				 */
2889 				secpolicy_setid_clear(vap, cr);
2890 				trim_mask = (mask & (AT_UID|AT_GID));
2891 			} else {
2892 				need_policy =  TRUE;
2893 			}
2894 		} else {
2895 			need_policy =  TRUE;
2896 		}
2897 	}
2898 
2899 	mutex_enter(&zp->z_lock);
2900 	oldva.va_mode = zp->z_mode;
2901 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2902 	if (mask & AT_XVATTR) {
2903 		/*
2904 		 * Update xvattr mask to include only those attributes
2905 		 * that are actually changing.
2906 		 *
2907 		 * the bits will be restored prior to actually setting
2908 		 * the attributes so the caller thinks they were set.
2909 		 */
2910 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2911 			if (xoap->xoa_appendonly !=
2912 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2913 				need_policy = TRUE;
2914 			} else {
2915 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2916 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2917 			}
2918 		}
2919 
2920 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2921 			if (xoap->xoa_nounlink !=
2922 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2923 				need_policy = TRUE;
2924 			} else {
2925 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2926 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2927 			}
2928 		}
2929 
2930 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2931 			if (xoap->xoa_immutable !=
2932 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2933 				need_policy = TRUE;
2934 			} else {
2935 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2936 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2937 			}
2938 		}
2939 
2940 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2941 			if (xoap->xoa_nodump !=
2942 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2943 				need_policy = TRUE;
2944 			} else {
2945 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2946 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2947 			}
2948 		}
2949 
2950 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2951 			if (xoap->xoa_av_modified !=
2952 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2953 				need_policy = TRUE;
2954 			} else {
2955 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2956 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2957 			}
2958 		}
2959 
2960 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2961 			if ((vp->v_type != VREG &&
2962 			    xoap->xoa_av_quarantined) ||
2963 			    xoap->xoa_av_quarantined !=
2964 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2965 				need_policy = TRUE;
2966 			} else {
2967 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2968 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2969 			}
2970 		}
2971 
2972 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2973 			mutex_exit(&zp->z_lock);
2974 			ZFS_EXIT(zfsvfs);
2975 			return (SET_ERROR(EPERM));
2976 		}
2977 
2978 		if (need_policy == FALSE &&
2979 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2980 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2981 			need_policy = TRUE;
2982 		}
2983 	}
2984 
2985 	mutex_exit(&zp->z_lock);
2986 
2987 	if (mask & AT_MODE) {
2988 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2989 			err = secpolicy_setid_setsticky_clear(vp, vap,
2990 			    &oldva, cr);
2991 			if (err) {
2992 				ZFS_EXIT(zfsvfs);
2993 				return (err);
2994 			}
2995 			trim_mask |= AT_MODE;
2996 		} else {
2997 			need_policy = TRUE;
2998 		}
2999 	}
3000 
3001 	if (need_policy) {
3002 		/*
3003 		 * If trim_mask is set then take ownership
3004 		 * has been granted or write_acl is present and user
3005 		 * has the ability to modify mode.  In that case remove
3006 		 * UID|GID and or MODE from mask so that
3007 		 * secpolicy_vnode_setattr() doesn't revoke it.
3008 		 */
3009 
3010 		if (trim_mask) {
3011 			saved_mask = vap->va_mask;
3012 			vap->va_mask &= ~trim_mask;
3013 		}
3014 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3015 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3016 		if (err) {
3017 			ZFS_EXIT(zfsvfs);
3018 			return (err);
3019 		}
3020 
3021 		if (trim_mask)
3022 			vap->va_mask |= saved_mask;
3023 	}
3024 
3025 	/*
3026 	 * secpolicy_vnode_setattr, or take ownership may have
3027 	 * changed va_mask
3028 	 */
3029 	mask = vap->va_mask;
3030 
3031 	if ((mask & (AT_UID | AT_GID))) {
3032 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3033 		    &xattr_obj, sizeof (xattr_obj));
3034 
3035 		if (err == 0 && xattr_obj) {
3036 			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3037 			if (err)
3038 				goto out2;
3039 		}
3040 		if (mask & AT_UID) {
3041 			new_uid = zfs_fuid_create(zfsvfs,
3042 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3043 			if (new_uid != zp->z_uid &&
3044 			    zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3045 				if (attrzp)
3046 					VN_RELE(ZTOV(attrzp));
3047 				err = SET_ERROR(EDQUOT);
3048 				goto out2;
3049 			}
3050 		}
3051 
3052 		if (mask & AT_GID) {
3053 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3054 			    cr, ZFS_GROUP, &fuidp);
3055 			if (new_gid != zp->z_gid &&
3056 			    zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3057 				if (attrzp)
3058 					VN_RELE(ZTOV(attrzp));
3059 				err = SET_ERROR(EDQUOT);
3060 				goto out2;
3061 			}
3062 		}
3063 	}
3064 	tx = dmu_tx_create(zfsvfs->z_os);
3065 
3066 	if (mask & AT_MODE) {
3067 		uint64_t pmode = zp->z_mode;
3068 		uint64_t acl_obj;
3069 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3070 
3071 		if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3072 		    !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3073 			err = SET_ERROR(EPERM);
3074 			goto out;
3075 		}
3076 
3077 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3078 			goto out;
3079 
3080 		mutex_enter(&zp->z_lock);
3081 		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3082 			/*
3083 			 * Are we upgrading ACL from old V0 format
3084 			 * to V1 format?
3085 			 */
3086 			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3087 			    zfs_znode_acl_version(zp) ==
3088 			    ZFS_ACL_VERSION_INITIAL) {
3089 				dmu_tx_hold_free(tx, acl_obj, 0,
3090 				    DMU_OBJECT_END);
3091 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3092 				    0, aclp->z_acl_bytes);
3093 			} else {
3094 				dmu_tx_hold_write(tx, acl_obj, 0,
3095 				    aclp->z_acl_bytes);
3096 			}
3097 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3098 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3099 			    0, aclp->z_acl_bytes);
3100 		}
3101 		mutex_exit(&zp->z_lock);
3102 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3103 	} else {
3104 		if ((mask & AT_XVATTR) &&
3105 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3106 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3107 		else
3108 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3109 	}
3110 
3111 	if (attrzp) {
3112 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3113 	}
3114 
3115 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3116 	if (fuid_dirtied)
3117 		zfs_fuid_txhold(zfsvfs, tx);
3118 
3119 	zfs_sa_upgrade_txholds(tx, zp);
3120 
3121 	err = dmu_tx_assign(tx, TXG_WAIT);
3122 	if (err)
3123 		goto out;
3124 
3125 	count = 0;
3126 	/*
3127 	 * Set each attribute requested.
3128 	 * We group settings according to the locks they need to acquire.
3129 	 *
3130 	 * Note: you cannot set ctime directly, although it will be
3131 	 * updated as a side-effect of calling this function.
3132 	 */
3133 
3134 
3135 	if (mask & (AT_UID|AT_GID|AT_MODE))
3136 		mutex_enter(&zp->z_acl_lock);
3137 	mutex_enter(&zp->z_lock);
3138 
3139 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3140 	    &zp->z_pflags, sizeof (zp->z_pflags));
3141 
3142 	if (attrzp) {
3143 		if (mask & (AT_UID|AT_GID|AT_MODE))
3144 			mutex_enter(&attrzp->z_acl_lock);
3145 		mutex_enter(&attrzp->z_lock);
3146 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3147 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3148 		    sizeof (attrzp->z_pflags));
3149 	}
3150 
3151 	if (mask & (AT_UID|AT_GID)) {
3152 
3153 		if (mask & AT_UID) {
3154 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3155 			    &new_uid, sizeof (new_uid));
3156 			zp->z_uid = new_uid;
3157 			if (attrzp) {
3158 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3159 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3160 				    sizeof (new_uid));
3161 				attrzp->z_uid = new_uid;
3162 			}
3163 		}
3164 
3165 		if (mask & AT_GID) {
3166 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3167 			    NULL, &new_gid, sizeof (new_gid));
3168 			zp->z_gid = new_gid;
3169 			if (attrzp) {
3170 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3171 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3172 				    sizeof (new_gid));
3173 				attrzp->z_gid = new_gid;
3174 			}
3175 		}
3176 		if (!(mask & AT_MODE)) {
3177 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3178 			    NULL, &new_mode, sizeof (new_mode));
3179 			new_mode = zp->z_mode;
3180 		}
3181 		err = zfs_acl_chown_setattr(zp);
3182 		ASSERT(err == 0);
3183 		if (attrzp) {
3184 			err = zfs_acl_chown_setattr(attrzp);
3185 			ASSERT(err == 0);
3186 		}
3187 	}
3188 
3189 	if (mask & AT_MODE) {
3190 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3191 		    &new_mode, sizeof (new_mode));
3192 		zp->z_mode = new_mode;
3193 		ASSERT3U((uintptr_t)aclp, !=, NULL);
3194 		err = zfs_aclset_common(zp, aclp, cr, tx);
3195 		ASSERT0(err);
3196 		if (zp->z_acl_cached)
3197 			zfs_acl_free(zp->z_acl_cached);
3198 		zp->z_acl_cached = aclp;
3199 		aclp = NULL;
3200 	}
3201 
3202 
3203 	if (mask & AT_ATIME) {
3204 		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3205 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3206 		    &zp->z_atime, sizeof (zp->z_atime));
3207 	}
3208 
3209 	if (mask & AT_MTIME) {
3210 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3211 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3212 		    mtime, sizeof (mtime));
3213 	}
3214 
3215 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3216 	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3217 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3218 		    NULL, mtime, sizeof (mtime));
3219 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3220 		    &ctime, sizeof (ctime));
3221 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3222 		    B_TRUE);
3223 	} else if (mask != 0) {
3224 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3225 		    &ctime, sizeof (ctime));
3226 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3227 		    B_TRUE);
3228 		if (attrzp) {
3229 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3230 			    SA_ZPL_CTIME(zfsvfs), NULL,
3231 			    &ctime, sizeof (ctime));
3232 			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3233 			    mtime, ctime, B_TRUE);
3234 		}
3235 	}
3236 	/*
3237 	 * Do this after setting timestamps to prevent timestamp
3238 	 * update from toggling bit
3239 	 */
3240 
3241 	if (xoap && (mask & AT_XVATTR)) {
3242 
3243 		/*
3244 		 * restore trimmed off masks
3245 		 * so that return masks can be set for caller.
3246 		 */
3247 
3248 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3249 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
3250 		}
3251 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3252 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
3253 		}
3254 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3255 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3256 		}
3257 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3258 			XVA_SET_REQ(xvap, XAT_NODUMP);
3259 		}
3260 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3261 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3262 		}
3263 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3264 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3265 		}
3266 
3267 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3268 			ASSERT(vp->v_type == VREG);
3269 
3270 		zfs_xvattr_set(zp, xvap, tx);
3271 	}
3272 
3273 	if (fuid_dirtied)
3274 		zfs_fuid_sync(zfsvfs, tx);
3275 
3276 	if (mask != 0)
3277 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3278 
3279 	mutex_exit(&zp->z_lock);
3280 	if (mask & (AT_UID|AT_GID|AT_MODE))
3281 		mutex_exit(&zp->z_acl_lock);
3282 
3283 	if (attrzp) {
3284 		if (mask & (AT_UID|AT_GID|AT_MODE))
3285 			mutex_exit(&attrzp->z_acl_lock);
3286 		mutex_exit(&attrzp->z_lock);
3287 	}
3288 out:
3289 	if (err == 0 && attrzp) {
3290 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3291 		    xattr_count, tx);
3292 		ASSERT(err2 == 0);
3293 	}
3294 
3295 	if (attrzp)
3296 		VN_RELE(ZTOV(attrzp));
3297 
3298 	if (aclp)
3299 		zfs_acl_free(aclp);
3300 
3301 	if (fuidp) {
3302 		zfs_fuid_info_free(fuidp);
3303 		fuidp = NULL;
3304 	}
3305 
3306 	if (err) {
3307 		dmu_tx_abort(tx);
3308 		if (err == ERESTART)
3309 			goto top;
3310 	} else {
3311 		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3312 		dmu_tx_commit(tx);
3313 	}
3314 
3315 out2:
3316 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3317 		zil_commit(zilog, 0);
3318 
3319 	ZFS_EXIT(zfsvfs);
3320 	return (err);
3321 }
3322 
3323 typedef struct zfs_zlock {
3324 	krwlock_t	*zl_rwlock;	/* lock we acquired */
3325 	znode_t		*zl_znode;	/* znode we held */
3326 	struct zfs_zlock *zl_next;	/* next in list */
3327 } zfs_zlock_t;
3328 
3329 /*
3330  * Drop locks and release vnodes that were held by zfs_rename_lock().
3331  */
3332 static void
3333 zfs_rename_unlock(zfs_zlock_t **zlpp)
3334 {
3335 	zfs_zlock_t *zl;
3336 
3337 	while ((zl = *zlpp) != NULL) {
3338 		if (zl->zl_znode != NULL)
3339 			VN_RELE(ZTOV(zl->zl_znode));
3340 		rw_exit(