xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 42b14111)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
25  * Copyright (c) 2014 Integros [integros.com]
26  * Copyright 2015 Joyent, Inc.
27  * Copyright 2017 Nexenta Systems, Inc.
28  */
29 
30 /* Portions Copyright 2007 Jeremy Teo */
31 /* Portions Copyright 2010 Robert Milkowski */
32 
33 #include <sys/types.h>
34 #include <sys/param.h>
35 #include <sys/time.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/resource.h>
39 #include <sys/vfs.h>
40 #include <sys/vfs_opreg.h>
41 #include <sys/vnode.h>
42 #include <sys/file.h>
43 #include <sys/stat.h>
44 #include <sys/kmem.h>
45 #include <sys/taskq.h>
46 #include <sys/uio.h>
47 #include <sys/vmsystm.h>
48 #include <sys/atomic.h>
49 #include <sys/vm.h>
50 #include <vm/seg_vn.h>
51 #include <vm/pvn.h>
52 #include <vm/as.h>
53 #include <vm/kpm.h>
54 #include <vm/seg_kpm.h>
55 #include <sys/mman.h>
56 #include <sys/pathname.h>
57 #include <sys/cmn_err.h>
58 #include <sys/errno.h>
59 #include <sys/unistd.h>
60 #include <sys/zfs_dir.h>
61 #include <sys/zfs_acl.h>
62 #include <sys/zfs_ioctl.h>
63 #include <sys/fs/zfs.h>
64 #include <sys/dmu.h>
65 #include <sys/dmu_objset.h>
66 #include <sys/spa.h>
67 #include <sys/txg.h>
68 #include <sys/dbuf.h>
69 #include <sys/zap.h>
70 #include <sys/sa.h>
71 #include <sys/dirent.h>
72 #include <sys/policy.h>
73 #include <sys/sunddi.h>
74 #include <sys/filio.h>
75 #include <sys/sid.h>
76 #include "fs/fs_subr.h"
77 #include <sys/zfs_ctldir.h>
78 #include <sys/zfs_fuid.h>
79 #include <sys/zfs_sa.h>
80 #include <sys/dnlc.h>
81 #include <sys/zfs_rlock.h>
82 #include <sys/extdirent.h>
83 #include <sys/kidmap.h>
84 #include <sys/cred.h>
85 #include <sys/attr.h>
86 #include <sys/zil.h>
87 
88 /*
89  * Programming rules.
90  *
91  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
92  * properly lock its in-core state, create a DMU transaction, do the work,
93  * record this work in the intent log (ZIL), commit the DMU transaction,
94  * and wait for the intent log to commit if it is a synchronous operation.
95  * Moreover, the vnode ops must work in both normal and log replay context.
96  * The ordering of events is important to avoid deadlocks and references
97  * to freed memory.  The example below illustrates the following Big Rules:
98  *
99  *  (1)	A check must be made in each zfs thread for a mounted file system.
100  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
101  *	A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
102  *	must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
103  *	can return EIO from the calling function.
104  *
105  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
106  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
107  *	First, if it's the last reference, the vnode/znode
108  *	can be freed, so the zp may point to freed memory.  Second, the last
109  *	reference will call zfs_zinactive(), which may induce a lot of work --
110  *	pushing cached pages (which acquires range locks) and syncing out
111  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
112  *	which could deadlock the system if you were already holding one.
113  *	If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
114  *
115  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
116  *	as they can span dmu_tx_assign() calls.
117  *
118  *  (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
119  *      dmu_tx_assign().  This is critical because we don't want to block
120  *      while holding locks.
121  *
122  *	If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT.  This
123  *	reduces lock contention and CPU usage when we must wait (note that if
124  *	throughput is constrained by the storage, nearly every transaction
125  *	must wait).
126  *
127  *      Note, in particular, that if a lock is sometimes acquired before
128  *      the tx assigns, and sometimes after (e.g. z_lock), then failing
129  *      to use a non-blocking assign can deadlock the system.  The scenario:
130  *
131  *	Thread A has grabbed a lock before calling dmu_tx_assign().
132  *	Thread B is in an already-assigned tx, and blocks for this lock.
133  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
134  *	forever, because the previous txg can't quiesce until B's tx commits.
135  *
136  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
137  *	then drop all locks, call dmu_tx_wait(), and try again.  On subsequent
138  *	calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
139  *	to indicate that this operation has already called dmu_tx_wait().
140  *	This will ensure that we don't retry forever, waiting a short bit
141  *	each time.
142  *
143  *  (5)	If the operation succeeded, generate the intent log entry for it
144  *	before dropping locks.  This ensures that the ordering of events
145  *	in the intent log matches the order in which they actually occurred.
146  *	During ZIL replay the zfs_log_* functions will update the sequence
147  *	number to indicate the zil transaction has replayed.
148  *
149  *  (6)	At the end of each vnode op, the DMU tx must always commit,
150  *	regardless of whether there were any errors.
151  *
152  *  (7)	After dropping all locks, invoke zil_commit(zilog, foid)
153  *	to ensure that synchronous semantics are provided when necessary.
154  *
155  * In general, this is how things should be ordered in each vnode op:
156  *
157  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
158  * top:
159  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
160  *	rw_enter(...);			// grab any other locks you need
161  *	tx = dmu_tx_create(...);	// get DMU tx
162  *	dmu_tx_hold_*();		// hold each object you might modify
163  *	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
164  *	if (error) {
165  *		rw_exit(...);		// drop locks
166  *		zfs_dirent_unlock(dl);	// unlock directory entry
167  *		VN_RELE(...);		// release held vnodes
168  *		if (error == ERESTART) {
169  *			waited = B_TRUE;
170  *			dmu_tx_wait(tx);
171  *			dmu_tx_abort(tx);
172  *			goto top;
173  *		}
174  *		dmu_tx_abort(tx);	// abort DMU tx
175  *		ZFS_EXIT(zfsvfs);	// finished in zfs
176  *		return (error);		// really out of space
177  *	}
178  *	error = do_real_work();		// do whatever this VOP does
179  *	if (error == 0)
180  *		zfs_log_*(...);		// on success, make ZIL entry
181  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
182  *	rw_exit(...);			// drop locks
183  *	zfs_dirent_unlock(dl);		// unlock directory entry
184  *	VN_RELE(...);			// release held vnodes
185  *	zil_commit(zilog, foid);	// synchronous when necessary
186  *	ZFS_EXIT(zfsvfs);		// finished in zfs
187  *	return (error);			// done, report error
188  */
189 
190 /* ARGSUSED */
191 static int
192 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
193 {
194 	znode_t	*zp = VTOZ(*vpp);
195 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
196 
197 	ZFS_ENTER(zfsvfs);
198 	ZFS_VERIFY_ZP(zp);
199 
200 	if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
201 	    ((flag & FAPPEND) == 0)) {
202 		ZFS_EXIT(zfsvfs);
203 		return (SET_ERROR(EPERM));
204 	}
205 
206 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
207 	    ZTOV(zp)->v_type == VREG &&
208 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
209 		if (fs_vscan(*vpp, cr, 0) != 0) {
210 			ZFS_EXIT(zfsvfs);
211 			return (SET_ERROR(EACCES));
212 		}
213 	}
214 
215 	/* Keep a count of the synchronous opens in the znode */
216 	if (flag & (FSYNC | FDSYNC))
217 		atomic_inc_32(&zp->z_sync_cnt);
218 
219 	ZFS_EXIT(zfsvfs);
220 	return (0);
221 }
222 
223 /* ARGSUSED */
224 static int
225 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
226     caller_context_t *ct)
227 {
228 	znode_t	*zp = VTOZ(vp);
229 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
230 
231 	/*
232 	 * Clean up any locks held by this process on the vp.
233 	 */
234 	cleanlocks(vp, ddi_get_pid(), 0);
235 	cleanshares(vp, ddi_get_pid());
236 
237 	ZFS_ENTER(zfsvfs);
238 	ZFS_VERIFY_ZP(zp);
239 
240 	/* Decrement the synchronous opens in the znode */
241 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
242 		atomic_dec_32(&zp->z_sync_cnt);
243 
244 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
245 	    ZTOV(zp)->v_type == VREG &&
246 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
247 		VERIFY(fs_vscan(vp, cr, 1) == 0);
248 
249 	ZFS_EXIT(zfsvfs);
250 	return (0);
251 }
252 
253 /*
254  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
255  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
256  */
257 static int
258 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
259 {
260 	znode_t	*zp = VTOZ(vp);
261 	uint64_t noff = (uint64_t)*off; /* new offset */
262 	uint64_t file_sz;
263 	int error;
264 	boolean_t hole;
265 
266 	file_sz = zp->z_size;
267 	if (noff >= file_sz)  {
268 		return (SET_ERROR(ENXIO));
269 	}
270 
271 	if (cmd == _FIO_SEEK_HOLE)
272 		hole = B_TRUE;
273 	else
274 		hole = B_FALSE;
275 
276 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
277 
278 	if (error == ESRCH)
279 		return (SET_ERROR(ENXIO));
280 
281 	/*
282 	 * We could find a hole that begins after the logical end-of-file,
283 	 * because dmu_offset_next() only works on whole blocks.  If the
284 	 * EOF falls mid-block, then indicate that the "virtual hole"
285 	 * at the end of the file begins at the logical EOF, rather than
286 	 * at the end of the last block.
287 	 */
288 	if (noff > file_sz) {
289 		ASSERT(hole);
290 		noff = file_sz;
291 	}
292 
293 	if (noff < *off)
294 		return (error);
295 	*off = noff;
296 	return (error);
297 }
298 
299 /* ARGSUSED */
300 static int
301 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
302     int *rvalp, caller_context_t *ct)
303 {
304 	offset_t off;
305 	offset_t ndata;
306 	dmu_object_info_t doi;
307 	int error;
308 	zfsvfs_t *zfsvfs;
309 	znode_t *zp;
310 
311 	switch (com) {
312 	case _FIOFFS:
313 	{
314 		return (zfs_sync(vp->v_vfsp, 0, cred));
315 
316 		/*
317 		 * The following two ioctls are used by bfu.  Faking out,
318 		 * necessary to avoid bfu errors.
319 		 */
320 	}
321 	case _FIOGDIO:
322 	case _FIOSDIO:
323 	{
324 		return (0);
325 	}
326 
327 	case _FIO_SEEK_DATA:
328 	case _FIO_SEEK_HOLE:
329 	{
330 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
331 			return (SET_ERROR(EFAULT));
332 
333 		zp = VTOZ(vp);
334 		zfsvfs = zp->z_zfsvfs;
335 		ZFS_ENTER(zfsvfs);
336 		ZFS_VERIFY_ZP(zp);
337 
338 		/* offset parameter is in/out */
339 		error = zfs_holey(vp, com, &off);
340 		ZFS_EXIT(zfsvfs);
341 		if (error)
342 			return (error);
343 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
344 			return (SET_ERROR(EFAULT));
345 		return (0);
346 	}
347 	case _FIO_COUNT_FILLED:
348 	{
349 		/*
350 		 * _FIO_COUNT_FILLED adds a new ioctl command which
351 		 * exposes the number of filled blocks in a
352 		 * ZFS object.
353 		 */
354 		zp = VTOZ(vp);
355 		zfsvfs = zp->z_zfsvfs;
356 		ZFS_ENTER(zfsvfs);
357 		ZFS_VERIFY_ZP(zp);
358 
359 		/*
360 		 * Wait for all dirty blocks for this object
361 		 * to get synced out to disk, and the DMU info
362 		 * updated.
363 		 */
364 		error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id);
365 		if (error) {
366 			ZFS_EXIT(zfsvfs);
367 			return (error);
368 		}
369 
370 		/*
371 		 * Retrieve fill count from DMU object.
372 		 */
373 		error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi);
374 		if (error) {
375 			ZFS_EXIT(zfsvfs);
376 			return (error);
377 		}
378 
379 		ndata = doi.doi_fill_count;
380 
381 		ZFS_EXIT(zfsvfs);
382 		if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag))
383 			return (SET_ERROR(EFAULT));
384 		return (0);
385 	}
386 	}
387 	return (SET_ERROR(ENOTTY));
388 }
389 
390 /*
391  * Utility functions to map and unmap a single physical page.  These
392  * are used to manage the mappable copies of ZFS file data, and therefore
393  * do not update ref/mod bits.
394  */
395 caddr_t
396 zfs_map_page(page_t *pp, enum seg_rw rw)
397 {
398 	if (kpm_enable)
399 		return (hat_kpm_mapin(pp, 0));
400 	ASSERT(rw == S_READ || rw == S_WRITE);
401 	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
402 	    (caddr_t)-1));
403 }
404 
405 void
406 zfs_unmap_page(page_t *pp, caddr_t addr)
407 {
408 	if (kpm_enable) {
409 		hat_kpm_mapout(pp, 0, addr);
410 	} else {
411 		ppmapout(addr);
412 	}
413 }
414 
415 /*
416  * When a file is memory mapped, we must keep the IO data synchronized
417  * between the DMU cache and the memory mapped pages.  What this means:
418  *
419  * On Write:	If we find a memory mapped page, we write to *both*
420  *		the page and the dmu buffer.
421  */
422 static void
423 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
424 {
425 	int64_t	off;
426 
427 	off = start & PAGEOFFSET;
428 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
429 		page_t *pp;
430 		uint64_t nbytes = MIN(PAGESIZE - off, len);
431 
432 		if (pp = page_lookup(vp, start, SE_SHARED)) {
433 			caddr_t va;
434 
435 			va = zfs_map_page(pp, S_WRITE);
436 			(void) dmu_read(os, oid, start+off, nbytes, va+off,
437 			    DMU_READ_PREFETCH);
438 			zfs_unmap_page(pp, va);
439 			page_unlock(pp);
440 		}
441 		len -= nbytes;
442 		off = 0;
443 	}
444 }
445 
446 /*
447  * When a file is memory mapped, we must keep the IO data synchronized
448  * between the DMU cache and the memory mapped pages.  What this means:
449  *
450  * On Read:	We "read" preferentially from memory mapped pages,
451  *		else we default from the dmu buffer.
452  *
453  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
454  *	 the file is memory mapped.
455  */
456 static int
457 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
458 {
459 	znode_t *zp = VTOZ(vp);
460 	int64_t	start, off;
461 	int len = nbytes;
462 	int error = 0;
463 
464 	start = uio->uio_loffset;
465 	off = start & PAGEOFFSET;
466 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
467 		page_t *pp;
468 		uint64_t bytes = MIN(PAGESIZE - off, len);
469 
470 		if (pp = page_lookup(vp, start, SE_SHARED)) {
471 			caddr_t va;
472 
473 			va = zfs_map_page(pp, S_READ);
474 			error = uiomove(va + off, bytes, UIO_READ, uio);
475 			zfs_unmap_page(pp, va);
476 			page_unlock(pp);
477 		} else {
478 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
479 			    uio, bytes);
480 		}
481 		len -= bytes;
482 		off = 0;
483 		if (error)
484 			break;
485 	}
486 	return (error);
487 }
488 
489 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
490 
491 /*
492  * Read bytes from specified file into supplied buffer.
493  *
494  *	IN:	vp	- vnode of file to be read from.
495  *		uio	- structure supplying read location, range info,
496  *			  and return buffer.
497  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
498  *		cr	- credentials of caller.
499  *		ct	- caller context
500  *
501  *	OUT:	uio	- updated offset and range, buffer filled.
502  *
503  *	RETURN:	0 on success, error code on failure.
504  *
505  * Side Effects:
506  *	vp - atime updated if byte count > 0
507  */
508 /* ARGSUSED */
509 static int
510 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
511 {
512 	znode_t		*zp = VTOZ(vp);
513 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
514 	ssize_t		n, nbytes;
515 	int		error = 0;
516 	rl_t		*rl;
517 	xuio_t		*xuio = NULL;
518 
519 	ZFS_ENTER(zfsvfs);
520 	ZFS_VERIFY_ZP(zp);
521 
522 	if (zp->z_pflags & ZFS_AV_QUARANTINED) {
523 		ZFS_EXIT(zfsvfs);
524 		return (SET_ERROR(EACCES));
525 	}
526 
527 	/*
528 	 * Validate file offset
529 	 */
530 	if (uio->uio_loffset < (offset_t)0) {
531 		ZFS_EXIT(zfsvfs);
532 		return (SET_ERROR(EINVAL));
533 	}
534 
535 	/*
536 	 * Fasttrack empty reads
537 	 */
538 	if (uio->uio_resid == 0) {
539 		ZFS_EXIT(zfsvfs);
540 		return (0);
541 	}
542 
543 	/*
544 	 * Check for mandatory locks
545 	 */
546 	if (MANDMODE(zp->z_mode)) {
547 		if (error = chklock(vp, FREAD,
548 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
549 			ZFS_EXIT(zfsvfs);
550 			return (error);
551 		}
552 	}
553 
554 	/*
555 	 * If we're in FRSYNC mode, sync out this znode before reading it.
556 	 */
557 	if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
558 		zil_commit(zfsvfs->z_log, zp->z_id);
559 
560 	/*
561 	 * Lock the range against changes.
562 	 */
563 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
564 
565 	/*
566 	 * If we are reading past end-of-file we can skip
567 	 * to the end; but we might still need to set atime.
568 	 */
569 	if (uio->uio_loffset >= zp->z_size) {
570 		error = 0;
571 		goto out;
572 	}
573 
574 	ASSERT(uio->uio_loffset < zp->z_size);
575 	n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
576 
577 	if ((uio->uio_extflg == UIO_XUIO) &&
578 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
579 		int nblk;
580 		int blksz = zp->z_blksz;
581 		uint64_t offset = uio->uio_loffset;
582 
583 		xuio = (xuio_t *)uio;
584 		if ((ISP2(blksz))) {
585 			nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
586 			    blksz)) / blksz;
587 		} else {
588 			ASSERT(offset + n <= blksz);
589 			nblk = 1;
590 		}
591 		(void) dmu_xuio_init(xuio, nblk);
592 
593 		if (vn_has_cached_data(vp)) {
594 			/*
595 			 * For simplicity, we always allocate a full buffer
596 			 * even if we only expect to read a portion of a block.
597 			 */
598 			while (--nblk >= 0) {
599 				(void) dmu_xuio_add(xuio,
600 				    dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
601 				    blksz), 0, blksz);
602 			}
603 		}
604 	}
605 
606 	while (n > 0) {
607 		nbytes = MIN(n, zfs_read_chunk_size -
608 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
609 
610 		if (vn_has_cached_data(vp)) {
611 			error = mappedread(vp, nbytes, uio);
612 		} else {
613 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
614 			    uio, nbytes);
615 		}
616 		if (error) {
617 			/* convert checksum errors into IO errors */
618 			if (error == ECKSUM)
619 				error = SET_ERROR(EIO);
620 			break;
621 		}
622 
623 		n -= nbytes;
624 	}
625 out:
626 	zfs_range_unlock(rl);
627 
628 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
629 	ZFS_EXIT(zfsvfs);
630 	return (error);
631 }
632 
633 /*
634  * Write the bytes to a file.
635  *
636  *	IN:	vp	- vnode of file to be written to.
637  *		uio	- structure supplying write location, range info,
638  *			  and data buffer.
639  *		ioflag	- FAPPEND, FSYNC, and/or FDSYNC.  FAPPEND is
640  *			  set if in append mode.
641  *		cr	- credentials of caller.
642  *		ct	- caller context (NFS/CIFS fem monitor only)
643  *
644  *	OUT:	uio	- updated offset and range.
645  *
646  *	RETURN:	0 on success, error code on failure.
647  *
648  * Timestamps:
649  *	vp - ctime|mtime updated if byte count > 0
650  */
651 
652 /* ARGSUSED */
653 static int
654 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
655 {
656 	znode_t		*zp = VTOZ(vp);
657 	rlim64_t	limit = uio->uio_llimit;
658 	ssize_t		start_resid = uio->uio_resid;
659 	ssize_t		tx_bytes;
660 	uint64_t	end_size;
661 	dmu_tx_t	*tx;
662 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
663 	zilog_t		*zilog;
664 	offset_t	woff;
665 	ssize_t		n, nbytes;
666 	rl_t		*rl;
667 	int		max_blksz = zfsvfs->z_max_blksz;
668 	int		error = 0;
669 	arc_buf_t	*abuf;
670 	iovec_t		*aiov = NULL;
671 	xuio_t		*xuio = NULL;
672 	int		i_iov = 0;
673 	int		iovcnt = uio->uio_iovcnt;
674 	iovec_t		*iovp = uio->uio_iov;
675 	int		write_eof;
676 	int		count = 0;
677 	sa_bulk_attr_t	bulk[4];
678 	uint64_t	mtime[2], ctime[2];
679 
680 	/*
681 	 * Fasttrack empty write
682 	 */
683 	n = start_resid;
684 	if (n == 0)
685 		return (0);
686 
687 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
688 		limit = MAXOFFSET_T;
689 
690 	ZFS_ENTER(zfsvfs);
691 	ZFS_VERIFY_ZP(zp);
692 
693 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
694 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
695 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
696 	    &zp->z_size, 8);
697 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
698 	    &zp->z_pflags, 8);
699 
700 	/*
701 	 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
702 	 * callers might not be able to detect properly that we are read-only,
703 	 * so check it explicitly here.
704 	 */
705 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
706 		ZFS_EXIT(zfsvfs);
707 		return (SET_ERROR(EROFS));
708 	}
709 
710 	/*
711 	 * If immutable or not appending then return EPERM.
712 	 * Intentionally allow ZFS_READONLY through here.
713 	 * See zfs_zaccess_common()
714 	 */
715 	if ((zp->z_pflags & ZFS_IMMUTABLE) ||
716 	    ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
717 	    (uio->uio_loffset < zp->z_size))) {
718 		ZFS_EXIT(zfsvfs);
719 		return (SET_ERROR(EPERM));
720 	}
721 
722 	zilog = zfsvfs->z_log;
723 
724 	/*
725 	 * Validate file offset
726 	 */
727 	woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
728 	if (woff < 0) {
729 		ZFS_EXIT(zfsvfs);
730 		return (SET_ERROR(EINVAL));
731 	}
732 
733 	/*
734 	 * Check for mandatory locks before calling zfs_range_lock()
735 	 * in order to prevent a deadlock with locks set via fcntl().
736 	 */
737 	if (MANDMODE((mode_t)zp->z_mode) &&
738 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
739 		ZFS_EXIT(zfsvfs);
740 		return (error);
741 	}
742 
743 	/*
744 	 * Pre-fault the pages to ensure slow (eg NFS) pages
745 	 * don't hold up txg.
746 	 * Skip this if uio contains loaned arc_buf.
747 	 */
748 	if ((uio->uio_extflg == UIO_XUIO) &&
749 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
750 		xuio = (xuio_t *)uio;
751 	else
752 		uio_prefaultpages(MIN(n, max_blksz), uio);
753 
754 	/*
755 	 * If in append mode, set the io offset pointer to eof.
756 	 */
757 	if (ioflag & FAPPEND) {
758 		/*
759 		 * Obtain an appending range lock to guarantee file append
760 		 * semantics.  We reset the write offset once we have the lock.
761 		 */
762 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
763 		woff = rl->r_off;
764 		if (rl->r_len == UINT64_MAX) {
765 			/*
766 			 * We overlocked the file because this write will cause
767 			 * the file block size to increase.
768 			 * Note that zp_size cannot change with this lock held.
769 			 */
770 			woff = zp->z_size;
771 		}
772 		uio->uio_loffset = woff;
773 	} else {
774 		/*
775 		 * Note that if the file block size will change as a result of
776 		 * this write, then this range lock will lock the entire file
777 		 * so that we can re-write the block safely.
778 		 */
779 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
780 	}
781 
782 	if (woff >= limit) {
783 		zfs_range_unlock(rl);
784 		ZFS_EXIT(zfsvfs);
785 		return (SET_ERROR(EFBIG));
786 	}
787 
788 	if ((woff + n) > limit || woff > (limit - n))
789 		n = limit - woff;
790 
791 	/* Will this write extend the file length? */
792 	write_eof = (woff + n > zp->z_size);
793 
794 	end_size = MAX(zp->z_size, woff + n);
795 
796 	/*
797 	 * Write the file in reasonable size chunks.  Each chunk is written
798 	 * in a separate transaction; this keeps the intent log records small
799 	 * and allows us to do more fine-grained space accounting.
800 	 */
801 	while (n > 0) {
802 		abuf = NULL;
803 		woff = uio->uio_loffset;
804 		if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
805 		    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
806 			if (abuf != NULL)
807 				dmu_return_arcbuf(abuf);
808 			error = SET_ERROR(EDQUOT);
809 			break;
810 		}
811 
812 		if (xuio && abuf == NULL) {
813 			ASSERT(i_iov < iovcnt);
814 			aiov = &iovp[i_iov];
815 			abuf = dmu_xuio_arcbuf(xuio, i_iov);
816 			dmu_xuio_clear(xuio, i_iov);
817 			DTRACE_PROBE3(zfs_cp_write, int, i_iov,
818 			    iovec_t *, aiov, arc_buf_t *, abuf);
819 			ASSERT((aiov->iov_base == abuf->b_data) ||
820 			    ((char *)aiov->iov_base - (char *)abuf->b_data +
821 			    aiov->iov_len == arc_buf_size(abuf)));
822 			i_iov++;
823 		} else if (abuf == NULL && n >= max_blksz &&
824 		    woff >= zp->z_size &&
825 		    P2PHASE(woff, max_blksz) == 0 &&
826 		    zp->z_blksz == max_blksz) {
827 			/*
828 			 * This write covers a full block.  "Borrow" a buffer
829 			 * from the dmu so that we can fill it before we enter
830 			 * a transaction.  This avoids the possibility of
831 			 * holding up the transaction if the data copy hangs
832 			 * up on a pagefault (e.g., from an NFS server mapping).
833 			 */
834 			size_t cbytes;
835 
836 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
837 			    max_blksz);
838 			ASSERT(abuf != NULL);
839 			ASSERT(arc_buf_size(abuf) == max_blksz);
840 			if (error = uiocopy(abuf->b_data, max_blksz,
841 			    UIO_WRITE, uio, &cbytes)) {
842 				dmu_return_arcbuf(abuf);
843 				break;
844 			}
845 			ASSERT(cbytes == max_blksz);
846 		}
847 
848 		/*
849 		 * Start a transaction.
850 		 */
851 		tx = dmu_tx_create(zfsvfs->z_os);
852 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
853 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
854 		zfs_sa_upgrade_txholds(tx, zp);
855 		error = dmu_tx_assign(tx, TXG_WAIT);
856 		if (error) {
857 			dmu_tx_abort(tx);
858 			if (abuf != NULL)
859 				dmu_return_arcbuf(abuf);
860 			break;
861 		}
862 
863 		/*
864 		 * If zfs_range_lock() over-locked we grow the blocksize
865 		 * and then reduce the lock range.  This will only happen
866 		 * on the first iteration since zfs_range_reduce() will
867 		 * shrink down r_len to the appropriate size.
868 		 */
869 		if (rl->r_len == UINT64_MAX) {
870 			uint64_t new_blksz;
871 
872 			if (zp->z_blksz > max_blksz) {
873 				/*
874 				 * File's blocksize is already larger than the
875 				 * "recordsize" property.  Only let it grow to
876 				 * the next power of 2.
877 				 */
878 				ASSERT(!ISP2(zp->z_blksz));
879 				new_blksz = MIN(end_size,
880 				    1 << highbit64(zp->z_blksz));
881 			} else {
882 				new_blksz = MIN(end_size, max_blksz);
883 			}
884 			zfs_grow_blocksize(zp, new_blksz, tx);
885 			zfs_range_reduce(rl, woff, n);
886 		}
887 
888 		/*
889 		 * XXX - should we really limit each write to z_max_blksz?
890 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
891 		 */
892 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
893 
894 		if (abuf == NULL) {
895 			tx_bytes = uio->uio_resid;
896 			error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
897 			    uio, nbytes, tx);
898 			tx_bytes -= uio->uio_resid;
899 		} else {
900 			tx_bytes = nbytes;
901 			ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
902 			/*
903 			 * If this is not a full block write, but we are
904 			 * extending the file past EOF and this data starts
905 			 * block-aligned, use assign_arcbuf().  Otherwise,
906 			 * write via dmu_write().
907 			 */
908 			if (tx_bytes < max_blksz && (!write_eof ||
909 			    aiov->iov_base != abuf->b_data)) {
910 				ASSERT(xuio);
911 				dmu_write(zfsvfs->z_os, zp->z_id, woff,
912 				    aiov->iov_len, aiov->iov_base, tx);
913 				dmu_return_arcbuf(abuf);
914 				xuio_stat_wbuf_copied();
915 			} else {
916 				ASSERT(xuio || tx_bytes == max_blksz);
917 				dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
918 				    woff, abuf, tx);
919 			}
920 			ASSERT(tx_bytes <= uio->uio_resid);
921 			uioskip(uio, tx_bytes);
922 		}
923 		if (tx_bytes && vn_has_cached_data(vp)) {
924 			update_pages(vp, woff,
925 			    tx_bytes, zfsvfs->z_os, zp->z_id);
926 		}
927 
928 		/*
929 		 * If we made no progress, we're done.  If we made even
930 		 * partial progress, update the znode and ZIL accordingly.
931 		 */
932 		if (tx_bytes == 0) {
933 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
934 			    (void *)&zp->z_size, sizeof (uint64_t), tx);
935 			dmu_tx_commit(tx);
936 			ASSERT(error != 0);
937 			break;
938 		}
939 
940 		/*
941 		 * Clear Set-UID/Set-GID bits on successful write if not
942 		 * privileged and at least one of the excute bits is set.
943 		 *
944 		 * It would be nice to to this after all writes have
945 		 * been done, but that would still expose the ISUID/ISGID
946 		 * to another app after the partial write is committed.
947 		 *
948 		 * Note: we don't call zfs_fuid_map_id() here because
949 		 * user 0 is not an ephemeral uid.
950 		 */
951 		mutex_enter(&zp->z_acl_lock);
952 		if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
953 		    (S_IXUSR >> 6))) != 0 &&
954 		    (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
955 		    secpolicy_vnode_setid_retain(cr,
956 		    (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
957 			uint64_t newmode;
958 			zp->z_mode &= ~(S_ISUID | S_ISGID);
959 			newmode = zp->z_mode;
960 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
961 			    (void *)&newmode, sizeof (uint64_t), tx);
962 		}
963 		mutex_exit(&zp->z_acl_lock);
964 
965 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
966 		    B_TRUE);
967 
968 		/*
969 		 * Update the file size (zp_size) if it has changed;
970 		 * account for possible concurrent updates.
971 		 */
972 		while ((end_size = zp->z_size) < uio->uio_loffset) {
973 			(void) atomic_cas_64(&zp->z_size, end_size,
974 			    uio->uio_loffset);
975 			ASSERT(error == 0);
976 		}
977 		/*
978 		 * If we are replaying and eof is non zero then force
979 		 * the file size to the specified eof. Note, there's no
980 		 * concurrency during replay.
981 		 */
982 		if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
983 			zp->z_size = zfsvfs->z_replay_eof;
984 
985 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
986 
987 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
988 		dmu_tx_commit(tx);
989 
990 		if (error != 0)
991 			break;
992 		ASSERT(tx_bytes == nbytes);
993 		n -= nbytes;
994 
995 		if (!xuio && n > 0)
996 			uio_prefaultpages(MIN(n, max_blksz), uio);
997 	}
998 
999 	zfs_range_unlock(rl);
1000 
1001 	/*
1002 	 * If we're in replay mode, or we made no progress, return error.
1003 	 * Otherwise, it's at least a partial write, so it's successful.
1004 	 */
1005 	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1006 		ZFS_EXIT(zfsvfs);
1007 		return (error);
1008 	}
1009 
1010 	if (ioflag & (FSYNC | FDSYNC) ||
1011 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1012 		zil_commit(zilog, zp->z_id);
1013 
1014 	ZFS_EXIT(zfsvfs);
1015 	return (0);
1016 }
1017 
1018 void
1019 zfs_get_done(zgd_t *zgd, int error)
1020 {
1021 	znode_t *zp = zgd->zgd_private;
1022 	objset_t *os = zp->z_zfsvfs->z_os;
1023 
1024 	if (zgd->zgd_db)
1025 		dmu_buf_rele(zgd->zgd_db, zgd);
1026 
1027 	zfs_range_unlock(zgd->zgd_rl);
1028 
1029 	/*
1030 	 * Release the vnode asynchronously as we currently have the
1031 	 * txg stopped from syncing.
1032 	 */
1033 	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1034 
1035 	if (error == 0 && zgd->zgd_bp)
1036 		zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
1037 
1038 	kmem_free(zgd, sizeof (zgd_t));
1039 }
1040 
1041 #ifdef DEBUG
1042 static int zil_fault_io = 0;
1043 #endif
1044 
1045 /*
1046  * Get data to generate a TX_WRITE intent log record.
1047  */
1048 int
1049 zfs_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
1050 {
1051 	zfsvfs_t *zfsvfs = arg;
1052 	objset_t *os = zfsvfs->z_os;
1053 	znode_t *zp;
1054 	uint64_t object = lr->lr_foid;
1055 	uint64_t offset = lr->lr_offset;
1056 	uint64_t size = lr->lr_length;
1057 	dmu_buf_t *db;
1058 	zgd_t *zgd;
1059 	int error = 0;
1060 
1061 	ASSERT3P(lwb, !=, NULL);
1062 	ASSERT3P(zio, !=, NULL);
1063 	ASSERT3U(size, !=, 0);
1064 
1065 	/*
1066 	 * Nothing to do if the file has been removed
1067 	 */
1068 	if (zfs_zget(zfsvfs, object, &zp) != 0)
1069 		return (SET_ERROR(ENOENT));
1070 	if (zp->z_unlinked) {
1071 		/*
1072 		 * Release the vnode asynchronously as we currently have the
1073 		 * txg stopped from syncing.
1074 		 */
1075 		VN_RELE_ASYNC(ZTOV(zp),
1076 		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1077 		return (SET_ERROR(ENOENT));
1078 	}
1079 
1080 	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1081 	zgd->zgd_lwb = lwb;
1082 	zgd->zgd_private = zp;
1083 
1084 	/*
1085 	 * Write records come in two flavors: immediate and indirect.
1086 	 * For small writes it's cheaper to store the data with the
1087 	 * log record (immediate); for large writes it's cheaper to
1088 	 * sync the data and get a pointer to it (indirect) so that
1089 	 * we don't have to write the data twice.
1090 	 */
1091 	if (buf != NULL) { /* immediate write */
1092 		zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1093 		/* test for truncation needs to be done while range locked */
1094 		if (offset >= zp->z_size) {
1095 			error = SET_ERROR(ENOENT);
1096 		} else {
1097 			error = dmu_read(os, object, offset, size, buf,
1098 			    DMU_READ_NO_PREFETCH);
1099 		}
1100 		ASSERT(error == 0 || error == ENOENT);
1101 	} else { /* indirect write */
1102 		/*
1103 		 * Have to lock the whole block to ensure when it's
1104 		 * written out and its checksum is being calculated
1105 		 * that no one can change the data. We need to re-check
1106 		 * blocksize after we get the lock in case it's changed!
1107 		 */
1108 		for (;;) {
1109 			uint64_t blkoff;
1110 			size = zp->z_blksz;
1111 			blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1112 			offset -= blkoff;
1113 			zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1114 			    RL_READER);
1115 			if (zp->z_blksz == size)
1116 				break;
1117 			offset += blkoff;
1118 			zfs_range_unlock(zgd->zgd_rl);
1119 		}
1120 		/* test for truncation needs to be done while range locked */
1121 		if (lr->lr_offset >= zp->z_size)
1122 			error = SET_ERROR(ENOENT);
1123 #ifdef DEBUG
1124 		if (zil_fault_io) {
1125 			error = SET_ERROR(EIO);
1126 			zil_fault_io = 0;
1127 		}
1128 #endif
1129 		if (error == 0)
1130 			error = dmu_buf_hold(os, object, offset, zgd, &db,
1131 			    DMU_READ_NO_PREFETCH);
1132 
1133 		if (error == 0) {
1134 			blkptr_t *bp = &lr->lr_blkptr;
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 	/*
1241 	 * Fast path lookup, however we must skip DNLC lookup
1242 	 * for case folding or normalizing lookups because the
1243 	 * DNLC code only stores the passed in name.  This means
1244 	 * creating 'a' and removing 'A' on a case insensitive
1245 	 * file system would work, but DNLC still thinks 'a'
1246 	 * exists and won't let you create it again on the next
1247 	 * pass through fast path.
1248 	 */
1249 	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1250 
1251 		if (dvp->v_type != VDIR) {
1252 			return (SET_ERROR(ENOTDIR));
1253 		} else if (zdp->z_sa_hdl == NULL) {
1254 			return (SET_ERROR(EIO));
1255 		}
1256 
1257 		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1258 			error = zfs_fastaccesschk_execute(zdp, cr);
1259 			if (!error) {
1260 				*vpp = dvp;
1261 				VN_HOLD(*vpp);
1262 				return (0);
1263 			}
1264 			return (error);
1265 		} else if (!zdp->z_zfsvfs->z_norm &&
1266 		    (zdp->z_zfsvfs->z_case == ZFS_CASE_SENSITIVE)) {
1267 
1268 			vnode_t *tvp = dnlc_lookup(dvp, nm);
1269 
1270 			if (tvp) {
1271 				error = zfs_fastaccesschk_execute(zdp, cr);
1272 				if (error) {
1273 					VN_RELE(tvp);
1274 					return (error);
1275 				}
1276 				if (tvp == DNLC_NO_VNODE) {
1277 					VN_RELE(tvp);
1278 					return (SET_ERROR(ENOENT));
1279 				} else {
1280 					*vpp = tvp;
1281 					return (specvp_check(vpp, cr));
1282 				}
1283 			}
1284 		}
1285 	}
1286 
1287 	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1288 
1289 	ZFS_ENTER(zfsvfs);
1290 	ZFS_VERIFY_ZP(zdp);
1291 
1292 	*vpp = NULL;
1293 
1294 	if (flags & LOOKUP_XATTR) {
1295 		/*
1296 		 * If the xattr property is off, refuse the lookup request.
1297 		 */
1298 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1299 			ZFS_EXIT(zfsvfs);
1300 			return (SET_ERROR(EINVAL));
1301 		}
1302 
1303 		/*
1304 		 * We don't allow recursive attributes..
1305 		 * Maybe someday we will.
1306 		 */
1307 		if (zdp->z_pflags & ZFS_XATTR) {
1308 			ZFS_EXIT(zfsvfs);
1309 			return (SET_ERROR(EINVAL));
1310 		}
1311 
1312 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1313 			ZFS_EXIT(zfsvfs);
1314 			return (error);
1315 		}
1316 
1317 		/*
1318 		 * Do we have permission to get into attribute directory?
1319 		 */
1320 
1321 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1322 		    B_FALSE, cr)) {
1323 			VN_RELE(*vpp);
1324 			*vpp = NULL;
1325 		}
1326 
1327 		ZFS_EXIT(zfsvfs);
1328 		return (error);
1329 	}
1330 
1331 	if (dvp->v_type != VDIR) {
1332 		ZFS_EXIT(zfsvfs);
1333 		return (SET_ERROR(ENOTDIR));
1334 	}
1335 
1336 	/*
1337 	 * Check accessibility of directory.
1338 	 */
1339 
1340 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1341 		ZFS_EXIT(zfsvfs);
1342 		return (error);
1343 	}
1344 
1345 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1346 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1347 		ZFS_EXIT(zfsvfs);
1348 		return (SET_ERROR(EILSEQ));
1349 	}
1350 
1351 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1352 	if (error == 0)
1353 		error = specvp_check(vpp, cr);
1354 
1355 	ZFS_EXIT(zfsvfs);
1356 	return (error);
1357 }
1358 
1359 /*
1360  * Attempt to create a new entry in a directory.  If the entry
1361  * already exists, truncate the file if permissible, else return
1362  * an error.  Return the vp of the created or trunc'd file.
1363  *
1364  *	IN:	dvp	- vnode of directory to put new file entry in.
1365  *		name	- name of new file entry.
1366  *		vap	- attributes of new file.
1367  *		excl	- flag indicating exclusive or non-exclusive mode.
1368  *		mode	- mode to open file with.
1369  *		cr	- credentials of caller.
1370  *		flag	- large file flag [UNUSED].
1371  *		ct	- caller context
1372  *		vsecp	- ACL to be set
1373  *
1374  *	OUT:	vpp	- vnode of created or trunc'd entry.
1375  *
1376  *	RETURN:	0 on success, error code on failure.
1377  *
1378  * Timestamps:
1379  *	dvp - ctime|mtime updated if new entry created
1380  *	 vp - ctime|mtime always, atime if new
1381  */
1382 
1383 /* ARGSUSED */
1384 static int
1385 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1386     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1387     vsecattr_t *vsecp)
1388 {
1389 	znode_t		*zp, *dzp = VTOZ(dvp);
1390 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1391 	zilog_t		*zilog;
1392 	objset_t	*os;
1393 	zfs_dirlock_t	*dl;
1394 	dmu_tx_t	*tx;
1395 	int		error;
1396 	ksid_t		*ksid;
1397 	uid_t		uid;
1398 	gid_t		gid = crgetgid(cr);
1399 	zfs_acl_ids_t   acl_ids;
1400 	boolean_t	fuid_dirtied;
1401 	boolean_t	have_acl = B_FALSE;
1402 	boolean_t	waited = B_FALSE;
1403 
1404 	/*
1405 	 * If we have an ephemeral id, ACL, or XVATTR then
1406 	 * make sure file system is at proper version
1407 	 */
1408 
1409 	ksid = crgetsid(cr, KSID_OWNER);
1410 	if (ksid)
1411 		uid = ksid_getid(ksid);
1412 	else
1413 		uid = crgetuid(cr);
1414 
1415 	if (zfsvfs->z_use_fuids == B_FALSE &&
1416 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1417 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1418 		return (SET_ERROR(EINVAL));
1419 
1420 	ZFS_ENTER(zfsvfs);
1421 	ZFS_VERIFY_ZP(dzp);
1422 	os = zfsvfs->z_os;
1423 	zilog = zfsvfs->z_log;
1424 
1425 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1426 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1427 		ZFS_EXIT(zfsvfs);
1428 		return (SET_ERROR(EILSEQ));
1429 	}
1430 
1431 	if (vap->va_mask & AT_XVATTR) {
1432 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1433 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1434 			ZFS_EXIT(zfsvfs);
1435 			return (error);
1436 		}
1437 	}
1438 top:
1439 	*vpp = NULL;
1440 
1441 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1442 		vap->va_mode &= ~VSVTX;
1443 
1444 	if (*name == '\0') {
1445 		/*
1446 		 * Null component name refers to the directory itself.
1447 		 */
1448 		VN_HOLD(dvp);
1449 		zp = dzp;
1450 		dl = NULL;
1451 		error = 0;
1452 	} else {
1453 		/* possible VN_HOLD(zp) */
1454 		int zflg = 0;
1455 
1456 		if (flag & FIGNORECASE)
1457 			zflg |= ZCILOOK;
1458 
1459 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1460 		    NULL, NULL);
1461 		if (error) {
1462 			if (have_acl)
1463 				zfs_acl_ids_free(&acl_ids);
1464 			if (strcmp(name, "..") == 0)
1465 				error = SET_ERROR(EISDIR);
1466 			ZFS_EXIT(zfsvfs);
1467 			return (error);
1468 		}
1469 	}
1470 
1471 	if (zp == NULL) {
1472 		uint64_t txtype;
1473 
1474 		/*
1475 		 * Create a new file object and update the directory
1476 		 * to reference it.
1477 		 */
1478 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1479 			if (have_acl)
1480 				zfs_acl_ids_free(&acl_ids);
1481 			goto out;
1482 		}
1483 
1484 		/*
1485 		 * We only support the creation of regular files in
1486 		 * extended attribute directories.
1487 		 */
1488 
1489 		if ((dzp->z_pflags & ZFS_XATTR) &&
1490 		    (vap->va_type != VREG)) {
1491 			if (have_acl)
1492 				zfs_acl_ids_free(&acl_ids);
1493 			error = SET_ERROR(EINVAL);
1494 			goto out;
1495 		}
1496 
1497 		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1498 		    cr, vsecp, &acl_ids)) != 0)
1499 			goto out;
1500 		have_acl = B_TRUE;
1501 
1502 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1503 			zfs_acl_ids_free(&acl_ids);
1504 			error = SET_ERROR(EDQUOT);
1505 			goto out;
1506 		}
1507 
1508 		tx = dmu_tx_create(os);
1509 
1510 		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1511 		    ZFS_SA_BASE_ATTR_SIZE);
1512 
1513 		fuid_dirtied = zfsvfs->z_fuid_dirty;
1514 		if (fuid_dirtied)
1515 			zfs_fuid_txhold(zfsvfs, tx);
1516 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1517 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1518 		if (!zfsvfs->z_use_sa &&
1519 		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1520 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1521 			    0, acl_ids.z_aclp->z_acl_bytes);
1522 		}
1523 		error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1524 		if (error) {
1525 			zfs_dirent_unlock(dl);
1526 			if (error == ERESTART) {
1527 				waited = B_TRUE;
1528 				dmu_tx_wait(tx);
1529 				dmu_tx_abort(tx);
1530 				goto top;
1531 			}
1532 			zfs_acl_ids_free(&acl_ids);
1533 			dmu_tx_abort(tx);
1534 			ZFS_EXIT(zfsvfs);
1535 			return (error);
1536 		}
1537 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1538 
1539 		if (fuid_dirtied)
1540 			zfs_fuid_sync(zfsvfs, tx);
1541 
1542 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1543 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1544 		if (flag & FIGNORECASE)
1545 			txtype |= TX_CI;
1546 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1547 		    vsecp, acl_ids.z_fuidp, vap);
1548 		zfs_acl_ids_free(&acl_ids);
1549 		dmu_tx_commit(tx);
1550 	} else {
1551 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1552 
1553 		if (have_acl)
1554 			zfs_acl_ids_free(&acl_ids);
1555 		have_acl = B_FALSE;
1556 
1557 		/*
1558 		 * A directory entry already exists for this name.
1559 		 */
1560 		/*
1561 		 * Can't truncate an existing file if in exclusive mode.
1562 		 */
1563 		if (excl == EXCL) {
1564 			error = SET_ERROR(EEXIST);
1565 			goto out;
1566 		}
1567 		/*
1568 		 * Can't open a directory for writing.
1569 		 */
1570 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1571 			error = SET_ERROR(EISDIR);
1572 			goto out;
1573 		}
1574 		/*
1575 		 * Verify requested access to file.
1576 		 */
1577 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1578 			goto out;
1579 		}
1580 
1581 		mutex_enter(&dzp->z_lock);
1582 		dzp->z_seq++;
1583 		mutex_exit(&dzp->z_lock);
1584 
1585 		/*
1586 		 * Truncate regular files if requested.
1587 		 */
1588 		if ((ZTOV(zp)->v_type == VREG) &&
1589 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1590 			/* we can't hold any locks when calling zfs_freesp() */
1591 			zfs_dirent_unlock(dl);
1592 			dl = NULL;
1593 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1594 			if (error == 0) {
1595 				vnevent_create(ZTOV(zp), ct);
1596 			}
1597 		}
1598 	}
1599 out:
1600 
1601 	if (dl)
1602 		zfs_dirent_unlock(dl);
1603 
1604 	if (error) {
1605 		if (zp)
1606 			VN_RELE(ZTOV(zp));
1607 	} else {
1608 		*vpp = ZTOV(zp);
1609 		error = specvp_check(vpp, cr);
1610 	}
1611 
1612 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1613 		zil_commit(zilog, 0);
1614 
1615 	ZFS_EXIT(zfsvfs);
1616 	return (error);
1617 }
1618 
1619 /*
1620  * Remove an entry from a directory.
1621  *
1622  *	IN:	dvp	- vnode of directory to remove entry from.
1623  *		name	- name of entry to remove.
1624  *		cr	- credentials of caller.
1625  *		ct	- caller context
1626  *		flags	- case flags
1627  *
1628  *	RETURN:	0 on success, error code on failure.
1629  *
1630  * Timestamps:
1631  *	dvp - ctime|mtime
1632  *	 vp - ctime (if nlink > 0)
1633  */
1634 
1635 uint64_t null_xattr = 0;
1636 
1637 /*ARGSUSED*/
1638 static int
1639 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1640     int flags)
1641 {
1642 	znode_t		*zp, *dzp = VTOZ(dvp);
1643 	znode_t		*xzp;
1644 	vnode_t		*vp;
1645 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1646 	zilog_t		*zilog;
1647 	uint64_t	acl_obj, xattr_obj;
1648 	uint64_t	xattr_obj_unlinked = 0;
1649 	uint64_t	obj = 0;
1650 	zfs_dirlock_t	*dl;
1651 	dmu_tx_t	*tx;
1652 	boolean_t	may_delete_now, delete_now = FALSE;
1653 	boolean_t	unlinked, toobig = FALSE;
1654 	uint64_t	txtype;
1655 	pathname_t	*realnmp = NULL;
1656 	pathname_t	realnm;
1657 	int		error;
1658 	int		zflg = ZEXISTS;
1659 	boolean_t	waited = B_FALSE;
1660 
1661 	ZFS_ENTER(zfsvfs);
1662 	ZFS_VERIFY_ZP(dzp);
1663 	zilog = zfsvfs->z_log;
1664 
1665 	if (flags & FIGNORECASE) {
1666 		zflg |= ZCILOOK;
1667 		pn_alloc(&realnm);
1668 		realnmp = &realnm;
1669 	}
1670 
1671 top:
1672 	xattr_obj = 0;
1673 	xzp = NULL;
1674 	/*
1675 	 * Attempt to lock directory; fail if entry doesn't exist.
1676 	 */
1677 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1678 	    NULL, realnmp)) {
1679 		if (realnmp)
1680 			pn_free(realnmp);
1681 		ZFS_EXIT(zfsvfs);
1682 		return (error);
1683 	}
1684 
1685 	vp = ZTOV(zp);
1686 
1687 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1688 		goto out;
1689 	}
1690 
1691 	/*
1692 	 * Need to use rmdir for removing directories.
1693 	 */
1694 	if (vp->v_type == VDIR) {
1695 		error = SET_ERROR(EPERM);
1696 		goto out;
1697 	}
1698 
1699 	vnevent_remove(vp, dvp, name, ct);
1700 
1701 	if (realnmp)
1702 		dnlc_remove(dvp, realnmp->pn_buf);
1703 	else
1704 		dnlc_remove(dvp, name);
1705 
1706 	mutex_enter(&vp->v_lock);
1707 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1708 	mutex_exit(&vp->v_lock);
1709 
1710 	/*
1711 	 * We may delete the znode now, or we may put it in the unlinked set;
1712 	 * it depends on whether we're the last link, and on whether there are
1713 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1714 	 * allow for either case.
1715 	 */
1716 	obj = zp->z_id;
1717 	tx = dmu_tx_create(zfsvfs->z_os);
1718 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1719 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1720 	zfs_sa_upgrade_txholds(tx, zp);
1721 	zfs_sa_upgrade_txholds(tx, dzp);
1722 	if (may_delete_now) {
1723 		toobig =
1724 		    zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1725 		/* if the file is too big, only hold_free a token amount */
1726 		dmu_tx_hold_free(tx, zp->z_id, 0,
1727 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1728 	}
1729 
1730 	/* are there any extended attributes? */
1731 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1732 	    &xattr_obj, sizeof (xattr_obj));
1733 	if (error == 0 && xattr_obj) {
1734 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1735 		ASSERT0(error);
1736 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1737 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1738 	}
1739 
1740 	mutex_enter(&zp->z_lock);
1741 	if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1742 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1743 	mutex_exit(&zp->z_lock);
1744 
1745 	/* charge as an update -- would be nice not to charge at all */
1746 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1747 
1748 	/*
1749 	 * Mark this transaction as typically resulting in a net free of space
1750 	 */
1751 	dmu_tx_mark_netfree(tx);
1752 
1753 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1754 	if (error) {
1755 		zfs_dirent_unlock(dl);
1756 		VN_RELE(vp);
1757 		if (xzp)
1758 			VN_RELE(ZTOV(xzp));
1759 		if (error == ERESTART) {
1760 			waited = B_TRUE;
1761 			dmu_tx_wait(tx);
1762 			dmu_tx_abort(tx);
1763 			goto top;
1764 		}
1765 		if (realnmp)
1766 			pn_free(realnmp);
1767 		dmu_tx_abort(tx);
1768 		ZFS_EXIT(zfsvfs);
1769 		return (error);
1770 	}
1771 
1772 	/*
1773 	 * Remove the directory entry.
1774 	 */
1775 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1776 
1777 	if (error) {
1778 		dmu_tx_commit(tx);
1779 		goto out;
1780 	}
1781 
1782 	if (unlinked) {
1783 		/*
1784 		 * Hold z_lock so that we can make sure that the ACL obj
1785 		 * hasn't changed.  Could have been deleted due to
1786 		 * zfs_sa_upgrade().
1787 		 */
1788 		mutex_enter(&zp->z_lock);
1789 		mutex_enter(&vp->v_lock);
1790 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1791 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1792 		delete_now = may_delete_now && !toobig &&
1793 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1794 		    xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1795 		    acl_obj;
1796 		mutex_exit(&vp->v_lock);
1797 	}
1798 
1799 	if (delete_now) {
1800 		if (xattr_obj_unlinked) {
1801 			ASSERT3U(xzp->z_links, ==, 2);
1802 			mutex_enter(&xzp->z_lock);
1803 			xzp->z_unlinked = 1;
1804 			xzp->z_links = 0;
1805 			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1806 			    &xzp->z_links, sizeof (xzp->z_links), tx);
1807 			ASSERT3U(error,  ==,  0);
1808 			mutex_exit(&xzp->z_lock);
1809 			zfs_unlinked_add(xzp, tx);
1810 
1811 			if (zp->z_is_sa)
1812 				error = sa_remove(zp->z_sa_hdl,
1813 				    SA_ZPL_XATTR(zfsvfs), tx);
1814 			else
1815 				error = sa_update(zp->z_sa_hdl,
1816 				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
1817 				    sizeof (uint64_t), tx);
1818 			ASSERT0(error);
1819 		}
1820 		mutex_enter(&vp->v_lock);
1821 		VN_RELE_LOCKED(vp);
1822 		ASSERT0(vp->v_count);
1823 		mutex_exit(&vp->v_lock);
1824 		mutex_exit(&zp->z_lock);
1825 		zfs_znode_delete(zp, tx);
1826 	} else if (unlinked) {
1827 		mutex_exit(&zp->z_lock);
1828 		zfs_unlinked_add(zp, tx);
1829 	}
1830 
1831 	txtype = TX_REMOVE;
1832 	if (flags & FIGNORECASE)
1833 		txtype |= TX_CI;
1834 	zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1835 
1836 	dmu_tx_commit(tx);
1837 out:
1838 	if (realnmp)
1839 		pn_free(realnmp);
1840 
1841 	zfs_dirent_unlock(dl);
1842 
1843 	if (!delete_now)
1844 		VN_RELE(vp);
1845 	if (xzp)
1846 		VN_RELE(ZTOV(xzp));
1847 
1848 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1849 		zil_commit(zilog, 0);
1850 
1851 	ZFS_EXIT(zfsvfs);
1852 	return (error);
1853 }
1854 
1855 /*
1856  * Create a new directory and insert it into dvp using the name
1857  * provided.  Return a pointer to the inserted directory.
1858  *
1859  *	IN:	dvp	- vnode of directory to add subdir to.
1860  *		dirname	- name of new directory.
1861  *		vap	- attributes of new directory.
1862  *		cr	- credentials of caller.
1863  *		ct	- caller context
1864  *		flags	- case flags
1865  *		vsecp	- ACL to be set
1866  *
1867  *	OUT:	vpp	- vnode of created directory.
1868  *
1869  *	RETURN:	0 on success, error code on failure.
1870  *
1871  * Timestamps:
1872  *	dvp - ctime|mtime updated
1873  *	 vp - ctime|mtime|atime updated
1874  */
1875 /*ARGSUSED*/
1876 static int
1877 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1878     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1879 {
1880 	znode_t		*zp, *dzp = VTOZ(dvp);
1881 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1882 	zilog_t		*zilog;
1883 	zfs_dirlock_t	*dl;
1884 	uint64_t	txtype;
1885 	dmu_tx_t	*tx;
1886 	int		error;
1887 	int		zf = ZNEW;
1888 	ksid_t		*ksid;
1889 	uid_t		uid;
1890 	gid_t		gid = crgetgid(cr);
1891 	zfs_acl_ids_t   acl_ids;
1892 	boolean_t	fuid_dirtied;
1893 	boolean_t	waited = B_FALSE;
1894 
1895 	ASSERT(vap->va_type == VDIR);
1896 
1897 	/*
1898 	 * If we have an ephemeral id, ACL, or XVATTR then
1899 	 * make sure file system is at proper version
1900 	 */
1901 
1902 	ksid = crgetsid(cr, KSID_OWNER);
1903 	if (ksid)
1904 		uid = ksid_getid(ksid);
1905 	else
1906 		uid = crgetuid(cr);
1907 	if (zfsvfs->z_use_fuids == B_FALSE &&
1908 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1909 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1910 		return (SET_ERROR(EINVAL));
1911 
1912 	ZFS_ENTER(zfsvfs);
1913 	ZFS_VERIFY_ZP(dzp);
1914 	zilog = zfsvfs->z_log;
1915 
1916 	if (dzp->z_pflags & ZFS_XATTR) {
1917 		ZFS_EXIT(zfsvfs);
1918 		return (SET_ERROR(EINVAL));
1919 	}
1920 
1921 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1922 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1923 		ZFS_EXIT(zfsvfs);
1924 		return (SET_ERROR(EILSEQ));
1925 	}
1926 	if (flags & FIGNORECASE)
1927 		zf |= ZCILOOK;
1928 
1929 	if (vap->va_mask & AT_XVATTR) {
1930 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1931 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1932 			ZFS_EXIT(zfsvfs);
1933 			return (error);
1934 		}
1935 	}
1936 
1937 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1938 	    vsecp, &acl_ids)) != 0) {
1939 		ZFS_EXIT(zfsvfs);
1940 		return (error);
1941 	}
1942 	/*
1943 	 * First make sure the new directory doesn't exist.
1944 	 *
1945 	 * Existence is checked first to make sure we don't return
1946 	 * EACCES instead of EEXIST which can cause some applications
1947 	 * to fail.
1948 	 */
1949 top:
1950 	*vpp = NULL;
1951 
1952 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1953 	    NULL, NULL)) {
1954 		zfs_acl_ids_free(&acl_ids);
1955 		ZFS_EXIT(zfsvfs);
1956 		return (error);
1957 	}
1958 
1959 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1960 		zfs_acl_ids_free(&acl_ids);
1961 		zfs_dirent_unlock(dl);
1962 		ZFS_EXIT(zfsvfs);
1963 		return (error);
1964 	}
1965 
1966 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1967 		zfs_acl_ids_free(&acl_ids);
1968 		zfs_dirent_unlock(dl);
1969 		ZFS_EXIT(zfsvfs);
1970 		return (SET_ERROR(EDQUOT));
1971 	}
1972 
1973 	/*
1974 	 * Add a new entry to the directory.
1975 	 */
1976 	tx = dmu_tx_create(zfsvfs->z_os);
1977 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1978 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1979 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1980 	if (fuid_dirtied)
1981 		zfs_fuid_txhold(zfsvfs, tx);
1982 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1983 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1984 		    acl_ids.z_aclp->z_acl_bytes);
1985 	}
1986 
1987 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1988 	    ZFS_SA_BASE_ATTR_SIZE);
1989 
1990 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1991 	if (error) {
1992 		zfs_dirent_unlock(dl);
1993 		if (error == ERESTART) {
1994 			waited = B_TRUE;
1995 			dmu_tx_wait(tx);
1996 			dmu_tx_abort(tx);
1997 			goto top;
1998 		}
1999 		zfs_acl_ids_free(&acl_ids);
2000 		dmu_tx_abort(tx);
2001 		ZFS_EXIT(zfsvfs);
2002 		return (error);
2003 	}
2004 
2005 	/*
2006 	 * Create new node.
2007 	 */
2008 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2009 
2010 	if (fuid_dirtied)
2011 		zfs_fuid_sync(zfsvfs, tx);
2012 
2013 	/*
2014 	 * Now put new name in parent dir.
2015 	 */
2016 	(void) zfs_link_create(dl, zp, tx, ZNEW);
2017 
2018 	*vpp = ZTOV(zp);
2019 
2020 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2021 	if (flags & FIGNORECASE)
2022 		txtype |= TX_CI;
2023 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2024 	    acl_ids.z_fuidp, vap);
2025 
2026 	zfs_acl_ids_free(&acl_ids);
2027 
2028 	dmu_tx_commit(tx);
2029 
2030 	zfs_dirent_unlock(dl);
2031 
2032 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2033 		zil_commit(zilog, 0);
2034 
2035 	ZFS_EXIT(zfsvfs);
2036 	return (0);
2037 }
2038 
2039 /*
2040  * Remove a directory subdir entry.  If the current working
2041  * directory is the same as the subdir to be removed, the
2042  * remove will fail.
2043  *
2044  *	IN:	dvp	- vnode of directory to remove from.
2045  *		name	- name of directory to be removed.
2046  *		cwd	- vnode of current working directory.
2047  *		cr	- credentials of caller.
2048  *		ct	- caller context
2049  *		flags	- case flags
2050  *
2051  *	RETURN:	0 on success, error code on failure.
2052  *
2053  * Timestamps:
2054  *	dvp - ctime|mtime updated
2055  */
2056 /*ARGSUSED*/
2057 static int
2058 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2059     caller_context_t *ct, int flags)
2060 {
2061 	znode_t		*dzp = VTOZ(dvp);
2062 	znode_t		*zp;
2063 	vnode_t		*vp;
2064 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2065 	zilog_t		*zilog;
2066 	zfs_dirlock_t	*dl;
2067 	dmu_tx_t	*tx;
2068 	int		error;
2069 	int		zflg = ZEXISTS;
2070 	boolean_t	waited = B_FALSE;
2071 
2072 	ZFS_ENTER(zfsvfs);
2073 	ZFS_VERIFY_ZP(dzp);
2074 	zilog = zfsvfs->z_log;
2075 
2076 	if (flags & FIGNORECASE)
2077 		zflg |= ZCILOOK;
2078 top:
2079 	zp = NULL;
2080 
2081 	/*
2082 	 * Attempt to lock directory; fail if entry doesn't exist.
2083 	 */
2084 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2085 	    NULL, NULL)) {
2086 		ZFS_EXIT(zfsvfs);
2087 		return (error);
2088 	}
2089 
2090 	vp = ZTOV(zp);
2091 
2092 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2093 		goto out;
2094 	}
2095 
2096 	if (vp->v_type != VDIR) {
2097 		error = SET_ERROR(ENOTDIR);
2098 		goto out;
2099 	}
2100 
2101 	if (vp == cwd) {
2102 		error = SET_ERROR(EINVAL);
2103 		goto out;
2104 	}
2105 
2106 	vnevent_rmdir(vp, dvp, name, ct);
2107 
2108 	/*
2109 	 * Grab a lock on the directory to make sure that noone is
2110 	 * trying to add (or lookup) entries while we are removing it.
2111 	 */
2112 	rw_enter(&zp->z_name_lock, RW_WRITER);
2113 
2114 	/*
2115 	 * Grab a lock on the parent pointer to make sure we play well
2116 	 * with the treewalk and directory rename code.
2117 	 */
2118 	rw_enter(&zp->z_parent_lock, RW_WRITER);
2119 
2120 	tx = dmu_tx_create(zfsvfs->z_os);
2121 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2122 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2123 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2124 	zfs_sa_upgrade_txholds(tx, zp);
2125 	zfs_sa_upgrade_txholds(tx, dzp);
2126 	dmu_tx_mark_netfree(tx);
2127 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2128 	if (error) {
2129 		rw_exit(&zp->z_parent_lock);
2130 		rw_exit(&zp->z_name_lock);
2131 		zfs_dirent_unlock(dl);
2132 		VN_RELE(vp);
2133 		if (error == ERESTART) {
2134 			waited = B_TRUE;
2135 			dmu_tx_wait(tx);
2136 			dmu_tx_abort(tx);
2137 			goto top;
2138 		}
2139 		dmu_tx_abort(tx);
2140 		ZFS_EXIT(zfsvfs);
2141 		return (error);
2142 	}
2143 
2144 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2145 
2146 	if (error == 0) {
2147 		uint64_t txtype = TX_RMDIR;
2148 		if (flags & FIGNORECASE)
2149 			txtype |= TX_CI;
2150 		zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2151 	}
2152 
2153 	dmu_tx_commit(tx);
2154 
2155 	rw_exit(&zp->z_parent_lock);
2156 	rw_exit(&zp->z_name_lock);
2157 out:
2158 	zfs_dirent_unlock(dl);
2159 
2160 	VN_RELE(vp);
2161 
2162 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2163 		zil_commit(zilog, 0);
2164 
2165 	ZFS_EXIT(zfsvfs);
2166 	return (error);
2167 }
2168 
2169 /*
2170  * Read as many directory entries as will fit into the provided
2171  * buffer from the given directory cursor position (specified in
2172  * the uio structure).
2173  *
2174  *	IN:	vp	- vnode of directory to read.
2175  *		uio	- structure supplying read location, range info,
2176  *			  and return buffer.
2177  *		cr	- credentials of caller.
2178  *		ct	- caller context
2179  *		flags	- case flags
2180  *
2181  *	OUT:	uio	- updated offset and range, buffer filled.
2182  *		eofp	- set to true if end-of-file detected.
2183  *
2184  *	RETURN:	0 on success, error code on failure.
2185  *
2186  * Timestamps:
2187  *	vp - atime updated
2188  *
2189  * Note that the low 4 bits of the cookie returned by zap is always zero.
2190  * This allows us to use the low range for "special" directory entries:
2191  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2192  * we use the offset 2 for the '.zfs' directory.
2193  */
2194 /* ARGSUSED */
2195 static int
2196 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2197     caller_context_t *ct, int flags)
2198 {
2199 	znode_t		*zp = VTOZ(vp);
2200 	iovec_t		*iovp;
2201 	edirent_t	*eodp;
2202 	dirent64_t	*odp;
2203 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2204 	objset_t	*os;
2205 	caddr_t		outbuf;
2206 	size_t		bufsize;
2207 	zap_cursor_t	zc;
2208 	zap_attribute_t	zap;
2209 	uint_t		bytes_wanted;
2210 	uint64_t	offset; /* must be unsigned; checks for < 1 */
2211 	uint64_t	parent;
2212 	int		local_eof;
2213 	int		outcount;
2214 	int		error;
2215 	uint8_t		prefetch;
2216 	boolean_t	check_sysattrs;
2217 
2218 	ZFS_ENTER(zfsvfs);
2219 	ZFS_VERIFY_ZP(zp);
2220 
2221 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2222 	    &parent, sizeof (parent))) != 0) {
2223 		ZFS_EXIT(zfsvfs);
2224 		return (error);
2225 	}
2226 
2227 	/*
2228 	 * If we are not given an eof variable,
2229 	 * use a local one.
2230 	 */
2231 	if (eofp == NULL)
2232 		eofp = &local_eof;
2233 
2234 	/*
2235 	 * Check for valid iov_len.
2236 	 */
2237 	if (uio->uio_iov->iov_len <= 0) {
2238 		ZFS_EXIT(zfsvfs);
2239 		return (SET_ERROR(EINVAL));
2240 	}
2241 
2242 	/*
2243 	 * Quit if directory has been removed (posix)
2244 	 */
2245 	if ((*eofp = zp->z_unlinked) != 0) {
2246 		ZFS_EXIT(zfsvfs);
2247 		return (0);
2248 	}
2249 
2250 	error = 0;
2251 	os = zfsvfs->z_os;
2252 	offset = uio->uio_loffset;
2253 	prefetch = zp->z_zn_prefetch;
2254 
2255 	/*
2256 	 * Initialize the iterator cursor.
2257 	 */
2258 	if (offset <= 3) {
2259 		/*
2260 		 * Start iteration from the beginning of the directory.
2261 		 */
2262 		zap_cursor_init(&zc, os, zp->z_id);
2263 	} else {
2264 		/*
2265 		 * The offset is a serialized cursor.
2266 		 */
2267 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2268 	}
2269 
2270 	/*
2271 	 * Get space to change directory entries into fs independent format.
2272 	 */
2273 	iovp = uio->uio_iov;
2274 	bytes_wanted = iovp->iov_len;
2275 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2276 		bufsize = bytes_wanted;
2277 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
2278 		odp = (struct dirent64 *)outbuf;
2279 	} else {
2280 		bufsize = bytes_wanted;
2281 		outbuf = NULL;
2282 		odp = (struct dirent64 *)iovp->iov_base;
2283 	}
2284 	eodp = (struct edirent *)odp;
2285 
2286 	/*
2287 	 * If this VFS supports the system attribute view interface; and
2288 	 * we're looking at an extended attribute directory; and we care
2289 	 * about normalization conflicts on this vfs; then we must check
2290 	 * for normalization conflicts with the sysattr name space.
2291 	 */
2292 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2293 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2294 	    (flags & V_RDDIR_ENTFLAGS);
2295 
2296 	/*
2297 	 * Transform to file-system independent format
2298 	 */
2299 	outcount = 0;
2300 	while (outcount < bytes_wanted) {
2301 		ino64_t objnum;
2302 		ushort_t reclen;
2303 		off64_t *next = NULL;
2304 
2305 		/*
2306 		 * Special case `.', `..', and `.zfs'.
2307 		 */
2308 		if (offset == 0) {
2309 			(void) strcpy(zap.za_name, ".");
2310 			zap.za_normalization_conflict = 0;
2311 			objnum = zp->z_id;
2312 		} else if (offset == 1) {
2313 			(void) strcpy(zap.za_name, "..");
2314 			zap.za_normalization_conflict = 0;
2315 			objnum = parent;
2316 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2317 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2318 			zap.za_normalization_conflict = 0;
2319 			objnum = ZFSCTL_INO_ROOT;
2320 		} else {
2321 			/*
2322 			 * Grab next entry.
2323 			 */
2324 			if (error = zap_cursor_retrieve(&zc, &zap)) {
2325 				if ((*eofp = (error == ENOENT)) != 0)
2326 					break;
2327 				else
2328 					goto update;
2329 			}
2330 
2331 			if (zap.za_integer_length != 8 ||
2332 			    zap.za_num_integers != 1) {
2333 				cmn_err(CE_WARN, "zap_readdir: bad directory "
2334 				    "entry, obj = %lld, offset = %lld\n",
2335 				    (u_longlong_t)zp->z_id,
2336 				    (u_longlong_t)offset);
2337 				error = SET_ERROR(ENXIO);
2338 				goto update;
2339 			}
2340 
2341 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2342 			/*
2343 			 * MacOS X can extract the object type here such as:
2344 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2345 			 */
2346 
2347 			if (check_sysattrs && !zap.za_normalization_conflict) {
2348 				zap.za_normalization_conflict =
2349 				    xattr_sysattr_casechk(zap.za_name);
2350 			}
2351 		}
2352 
2353 		if (flags & V_RDDIR_ACCFILTER) {
2354 			/*
2355 			 * If we have no access at all, don't include
2356 			 * this entry in the returned information
2357 			 */
2358 			znode_t	*ezp;
2359 			if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2360 				goto skip_entry;
2361 			if (!zfs_has_access(ezp, cr)) {
2362 				VN_RELE(ZTOV(ezp));
2363 				goto skip_entry;
2364 			}
2365 			VN_RELE(ZTOV(ezp));
2366 		}
2367 
2368 		if (flags & V_RDDIR_ENTFLAGS)
2369 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2370 		else
2371 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2372 
2373 		/*
2374 		 * Will this entry fit in the buffer?
2375 		 */
2376 		if (outcount + reclen > bufsize) {
2377 			/*
2378 			 * Did we manage to fit anything in the buffer?
2379 			 */
2380 			if (!outcount) {
2381 				error = SET_ERROR(EINVAL);
2382 				goto update;
2383 			}
2384 			break;
2385 		}
2386 		if (flags & V_RDDIR_ENTFLAGS) {
2387 			/*
2388 			 * Add extended flag entry:
2389 			 */
2390 			eodp->ed_ino = objnum;
2391 			eodp->ed_reclen = reclen;
2392 			/* NOTE: ed_off is the offset for the *next* entry */
2393 			next = &(eodp->ed_off);
2394 			eodp->ed_eflags = zap.za_normalization_conflict ?
2395 			    ED_CASE_CONFLICT : 0;
2396 			(void) strncpy(eodp->ed_name, zap.za_name,
2397 			    EDIRENT_NAMELEN(reclen));
2398 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2399 		} else {
2400 			/*
2401 			 * Add normal entry:
2402 			 */
2403 			odp->d_ino = objnum;
2404 			odp->d_reclen = reclen;
2405 			/* NOTE: d_off is the offset for the *next* entry */
2406 			next = &(odp->d_off);
2407 			(void) strncpy(odp->d_name, zap.za_name,
2408 			    DIRENT64_NAMELEN(reclen));
2409 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2410 		}
2411 		outcount += reclen;
2412 
2413 		ASSERT(outcount <= bufsize);
2414 
2415 		/* Prefetch znode */
2416 		if (prefetch)
2417 			dmu_prefetch(os, objnum, 0, 0, 0,
2418 			    ZIO_PRIORITY_SYNC_READ);
2419 
2420 	skip_entry:
2421 		/*
2422 		 * Move to the next entry, fill in the previous offset.
2423 		 */
2424 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2425 			zap_cursor_advance(&zc);
2426 			offset = zap_cursor_serialize(&zc);
2427 		} else {
2428 			offset += 1;
2429 		}
2430 		if (next)
2431 			*next = offset;
2432 	}
2433 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2434 
2435 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2436 		iovp->iov_base += outcount;
2437 		iovp->iov_len -= outcount;
2438 		uio->uio_resid -= outcount;
2439 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2440 		/*
2441 		 * Reset the pointer.
2442 		 */
2443 		offset = uio->uio_loffset;
2444 	}
2445 
2446 update:
2447 	zap_cursor_fini(&zc);
2448 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2449 		kmem_free(outbuf, bufsize);
2450 
2451 	if (error == ENOENT)
2452 		error = 0;
2453 
2454 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2455 
2456 	uio->uio_loffset = offset;
2457 	ZFS_EXIT(zfsvfs);
2458 	return (error);
2459 }
2460 
2461 ulong_t zfs_fsync_sync_cnt = 4;
2462 
2463 static int
2464 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2465 {
2466 	znode_t	*zp = VTOZ(vp);
2467 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2468 
2469 	/*
2470 	 * Regardless of whether this is required for standards conformance,
2471 	 * this is the logical behavior when fsync() is called on a file with
2472 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2473 	 * going to be pushed out as part of the zil_commit().
2474 	 */
2475 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2476 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2477 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2478 
2479 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2480 
2481 	if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2482 		ZFS_ENTER(zfsvfs);
2483 		ZFS_VERIFY_ZP(zp);
2484 		zil_commit(zfsvfs->z_log, zp->z_id);
2485 		ZFS_EXIT(zfsvfs);
2486 	}
2487 	return (0);
2488 }
2489 
2490 
2491 /*
2492  * Get the requested file attributes and place them in the provided
2493  * vattr structure.
2494  *
2495  *	IN:	vp	- vnode of file.
2496  *		vap	- va_mask identifies requested attributes.
2497  *			  If AT_XVATTR set, then optional attrs are requested
2498  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2499  *		cr	- credentials of caller.
2500  *		ct	- caller context
2501  *
2502  *	OUT:	vap	- attribute values.
2503  *
2504  *	RETURN:	0 (always succeeds).
2505  */
2506 /* ARGSUSED */
2507 static int
2508 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2509     caller_context_t *ct)
2510 {
2511 	znode_t *zp = VTOZ(vp);
2512 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2513 	int	error = 0;
2514 	uint64_t links;
2515 	uint64_t mtime[2], ctime[2];
2516 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2517 	xoptattr_t *xoap = NULL;
2518 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2519 	sa_bulk_attr_t bulk[2];
2520 	int count = 0;
2521 
2522 	ZFS_ENTER(zfsvfs);
2523 	ZFS_VERIFY_ZP(zp);
2524 
2525 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2526 
2527 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2528 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2529 
2530 	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2531 		ZFS_EXIT(zfsvfs);
2532 		return (error);
2533 	}
2534 
2535 	/*
2536 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2537 	 * Also, if we are the owner don't bother, since owner should
2538 	 * always be allowed to read basic attributes of file.
2539 	 */
2540 	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2541 	    (vap->va_uid != crgetuid(cr))) {
2542 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2543 		    skipaclchk, cr)) {
2544 			ZFS_EXIT(zfsvfs);
2545 			return (error);
2546 		}
2547 	}
2548 
2549 	/*
2550 	 * Return all attributes.  It's cheaper to provide the answer
2551 	 * than to determine whether we were asked the question.
2552 	 */
2553 
2554 	mutex_enter(&zp->z_lock);
2555 	vap->va_type = vp->v_type;
2556 	vap->va_mode = zp->z_mode & MODEMASK;
2557 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2558 	vap->va_nodeid = zp->z_id;
2559 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2560 		links = zp->z_links + 1;
2561 	else
2562 		links = zp->z_links;
2563 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2564 	vap->va_size = zp->z_size;
2565 	vap->va_rdev = vp->v_rdev;
2566 	vap->va_seq = zp->z_seq;
2567 
2568 	/*
2569 	 * Add in any requested optional attributes and the create time.
2570 	 * Also set the corresponding bits in the returned attribute bitmap.
2571 	 */
2572 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2573 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2574 			xoap->xoa_archive =
2575 			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2576 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2577 		}
2578 
2579 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2580 			xoap->xoa_readonly =
2581 			    ((zp->z_pflags & ZFS_READONLY) != 0);
2582 			XVA_SET_RTN(xvap, XAT_READONLY);
2583 		}
2584 
2585 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2586 			xoap->xoa_system =
2587 			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2588 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2589 		}
2590 
2591 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2592 			xoap->xoa_hidden =
2593 			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2594 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2595 		}
2596 
2597 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2598 			xoap->xoa_nounlink =
2599 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2600 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2601 		}
2602 
2603 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2604 			xoap->xoa_immutable =
2605 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2606 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2607 		}
2608 
2609 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2610 			xoap->xoa_appendonly =
2611 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2612 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2613 		}
2614 
2615 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2616 			xoap->xoa_nodump =
2617 			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2618 			XVA_SET_RTN(xvap, XAT_NODUMP);
2619 		}
2620 
2621 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2622 			xoap->xoa_opaque =
2623 			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2624 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2625 		}
2626 
2627 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2628 			xoap->xoa_av_quarantined =
2629 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2630 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2631 		}
2632 
2633 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2634 			xoap->xoa_av_modified =
2635 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2636 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2637 		}
2638 
2639 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2640 		    vp->v_type == VREG) {
2641 			zfs_sa_get_scanstamp(zp, xvap);
2642 		}
2643 
2644 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2645 			uint64_t times[2];
2646 
2647 			(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2648 			    times, sizeof (times));
2649 			ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2650 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2651 		}
2652 
2653 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2654 			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2655 			XVA_SET_RTN(xvap, XAT_REPARSE);
2656 		}
2657 		if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2658 			xoap->xoa_generation = zp->z_gen;
2659 			XVA_SET_RTN(xvap, XAT_GEN);
2660 		}
2661 
2662 		if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2663 			xoap->xoa_offline =
2664 			    ((zp->z_pflags & ZFS_OFFLINE) != 0);
2665 			XVA_SET_RTN(xvap, XAT_OFFLINE);
2666 		}
2667 
2668 		if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2669 			xoap->xoa_sparse =
2670 			    ((zp->z_pflags & ZFS_SPARSE) != 0);
2671 			XVA_SET_RTN(xvap, XAT_SPARSE);
2672 		}
2673 	}
2674 
2675 	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2676 	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2677 	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2678 
2679 	mutex_exit(&zp->z_lock);
2680 
2681 	sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2682 
2683 	if (zp->z_blksz == 0) {
2684 		/*
2685 		 * Block size hasn't been set; suggest maximal I/O transfers.
2686 		 */
2687 		vap->va_blksize = zfsvfs->z_max_blksz;
2688 	}
2689 
2690 	ZFS_EXIT(zfsvfs);
2691 	return (0);
2692 }
2693 
2694 /*
2695  * Set the file attributes to the values contained in the
2696  * vattr structure.
2697  *
2698  *	IN:	vp	- vnode of file to be modified.
2699  *		vap	- new attribute values.
2700  *			  If AT_XVATTR set, then optional attrs are being set
2701  *		flags	- ATTR_UTIME set if non-default time values provided.
2702  *			- ATTR_NOACLCHECK (CIFS context only).
2703  *		cr	- credentials of caller.
2704  *		ct	- caller context
2705  *
2706  *	RETURN:	0 on success, error code on failure.
2707  *
2708  * Timestamps:
2709  *	vp - ctime updated, mtime updated if size changed.
2710  */
2711 /* ARGSUSED */
2712 static int
2713 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2714     caller_context_t *ct)
2715 {
2716 	znode_t		*zp = VTOZ(vp);
2717 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2718 	zilog_t		*zilog;
2719 	dmu_tx_t	*tx;
2720 	vattr_t		oldva;
2721 	xvattr_t	tmpxvattr;
2722 	uint_t		mask = vap->va_mask;
2723 	uint_t		saved_mask = 0;
2724 	int		trim_mask = 0;
2725 	uint64_t	new_mode;
2726 	uint64_t	new_uid, new_gid;
2727 	uint64_t	xattr_obj;
2728 	uint64_t	mtime[2], ctime[2];
2729 	znode_t		*attrzp;
2730 	int		need_policy = FALSE;
2731 	int		err, err2;
2732 	zfs_fuid_info_t *fuidp = NULL;
2733 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2734 	xoptattr_t	*xoap;
2735 	zfs_acl_t	*aclp;
2736 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2737 	boolean_t	fuid_dirtied = B_FALSE;
2738 	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
2739 	int		count = 0, xattr_count = 0;
2740 
2741 	if (mask == 0)
2742 		return (0);
2743 
2744 	if (mask & AT_NOSET)
2745 		return (SET_ERROR(EINVAL));
2746 
2747 	ZFS_ENTER(zfsvfs);
2748 	ZFS_VERIFY_ZP(zp);
2749 
2750 	zilog = zfsvfs->z_log;
2751 
2752 	/*
2753 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2754 	 * that file system is at proper version level
2755 	 */
2756 
2757 	if (zfsvfs->z_use_fuids == B_FALSE &&
2758 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2759 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2760 	    (mask & AT_XVATTR))) {
2761 		ZFS_EXIT(zfsvfs);
2762 		return (SET_ERROR(EINVAL));
2763 	}
2764 
2765 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2766 		ZFS_EXIT(zfsvfs);
2767 		return (SET_ERROR(EISDIR));
2768 	}
2769 
2770 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2771 		ZFS_EXIT(zfsvfs);
2772 		return (SET_ERROR(EINVAL));
2773 	}
2774 
2775 	/*
2776 	 * If this is an xvattr_t, then get a pointer to the structure of
2777 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2778 	 */
2779 	xoap = xva_getxoptattr(xvap);
2780 
2781 	xva_init(&tmpxvattr);
2782 
2783 	/*
2784 	 * Immutable files can only alter immutable bit and atime
2785 	 */
2786 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2787 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2788 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2789 		ZFS_EXIT(zfsvfs);
2790 		return (SET_ERROR(EPERM));
2791 	}
2792 
2793 	/*
2794 	 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
2795 	 */
2796 
2797 	/*
2798 	 * Verify timestamps doesn't overflow 32 bits.
2799 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2800 	 * handle times greater than 2039.  This check should be removed
2801 	 * once large timestamps are fully supported.
2802 	 */
2803 	if (mask & (AT_ATIME | AT_MTIME)) {
2804 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2805 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2806 			ZFS_EXIT(zfsvfs);
2807 			return (SET_ERROR(EOVERFLOW));
2808 		}
2809 	}
2810 
2811 top:
2812 	attrzp = NULL;
2813 	aclp = NULL;
2814 
2815 	/* Can this be moved to before the top label? */
2816 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2817 		ZFS_EXIT(zfsvfs);
2818 		return (SET_ERROR(EROFS));
2819 	}
2820 
2821 	/*
2822 	 * First validate permissions
2823 	 */
2824 
2825 	if (mask & AT_SIZE) {
2826 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2827 		if (err) {
2828 			ZFS_EXIT(zfsvfs);
2829 			return (err);
2830 		}
2831 		/*
2832 		 * XXX - Note, we are not providing any open
2833 		 * mode flags here (like FNDELAY), so we may
2834 		 * block if there are locks present... this
2835 		 * should be addressed in openat().
2836 		 */
2837 		/* XXX - would it be OK to generate a log record here? */
2838 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2839 		if (err) {
2840 			ZFS_EXIT(zfsvfs);
2841 			return (err);
2842 		}
2843 
2844 		if (vap->va_size == 0)
2845 			vnevent_truncate(ZTOV(zp), ct);
2846 	}
2847 
2848 	if (mask & (AT_ATIME|AT_MTIME) ||
2849 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2850 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2851 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2852 	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2853 	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2854 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2855 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2856 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2857 		    skipaclchk, cr);
2858 	}
2859 
2860 	if (mask & (AT_UID|AT_GID)) {
2861 		int	idmask = (mask & (AT_UID|AT_GID));
2862 		int	take_owner;
2863 		int	take_group;
2864 
2865 		/*
2866 		 * NOTE: even if a new mode is being set,
2867 		 * we may clear S_ISUID/S_ISGID bits.
2868 		 */
2869 
2870 		if (!(mask & AT_MODE))
2871 			vap->va_mode = zp->z_mode;
2872 
2873 		/*
2874 		 * Take ownership or chgrp to group we are a member of
2875 		 */
2876 
2877 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2878 		take_group = (mask & AT_GID) &&
2879 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2880 
2881 		/*
2882 		 * If both AT_UID and AT_GID are set then take_owner and
2883 		 * take_group must both be set in order to allow taking
2884 		 * ownership.
2885 		 *
2886 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2887 		 *
2888 		 */
2889 
2890 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2891 		    ((idmask == AT_UID) && take_owner) ||
2892 		    ((idmask == AT_GID) && take_group)) {
2893 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2894 			    skipaclchk, cr) == 0) {
2895 				/*
2896 				 * Remove setuid/setgid for non-privileged users
2897 				 */
2898 				secpolicy_setid_clear(vap, cr);
2899 				trim_mask = (mask & (AT_UID|AT_GID));
2900 			} else {
2901 				need_policy =  TRUE;
2902 			}
2903 		} else {
2904 			need_policy =  TRUE;
2905 		}
2906 	}
2907 
2908 	mutex_enter(&zp->z_lock);
2909 	oldva.va_mode = zp->z_mode;
2910 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2911 	if (mask & AT_XVATTR) {
2912 		/*
2913 		 * Update xvattr mask to include only those attributes
2914 		 * that are actually changing.
2915 		 *
2916 		 * the bits will be restored prior to actually setting
2917 		 * the attributes so the caller thinks they were set.
2918 		 */
2919 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2920 			if (xoap->xoa_appendonly !=
2921 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2922 				need_policy = TRUE;
2923 			} else {
2924 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2925 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2926 			}
2927 		}
2928 
2929 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2930 			if (xoap->xoa_nounlink !=
2931 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2932 				need_policy = TRUE;
2933 			} else {
2934 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2935 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2936 			}
2937 		}
2938 
2939 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2940 			if (xoap->xoa_immutable !=
2941 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2942 				need_policy = TRUE;
2943 			} else {
2944 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2945 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2946 			}
2947 		}
2948 
2949 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2950 			if (xoap->xoa_nodump !=
2951 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2952 				need_policy = TRUE;
2953 			} else {
2954 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2955 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2956 			}
2957 		}
2958 
2959 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2960 			if (xoap->xoa_av_modified !=
2961 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2962 				need_policy = TRUE;
2963 			} else {
2964 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2965 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2966 			}
2967 		}
2968 
2969 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2970 			if ((vp->v_type != VREG &&
2971 			    xoap->xoa_av_quarantined) ||
2972 			    xoap->xoa_av_quarantined !=
2973 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2974 				need_policy = TRUE;
2975 			} else {
2976 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2977 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2978 			}
2979 		}
2980 
2981 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2982 			mutex_exit(&zp->z_lock);
2983 			ZFS_EXIT(zfsvfs);
2984 			return (SET_ERROR(EPERM));
2985 		}
2986 
2987 		if (need_policy == FALSE &&
2988 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2989 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2990 			need_policy = TRUE;
2991 		}
2992 	}
2993 
2994 	mutex_exit(&zp->z_lock);
2995 
2996 	if (mask & AT_MODE) {
2997 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2998 			err = secpolicy_setid_setsticky_clear(vp, vap,
2999 			    &oldva, cr);
3000 			if (err) {
3001 				ZFS_EXIT(zfsvfs);
3002 				return (err);
3003 			}
3004 			trim_mask |= AT_MODE;
3005 		} else {
3006 			need_policy = TRUE;
3007 		}
3008 	}
3009 
3010 	if (need_policy) {
3011 		/*
3012 		 * If trim_mask is set then take ownership
3013 		 * has been granted or write_acl is present and user
3014 		 * has the ability to modify mode.  In that case remove
3015 		 * UID|GID and or MODE from mask so that
3016 		 * secpolicy_vnode_setattr() doesn't revoke it.
3017 		 */
3018 
3019 		if (trim_mask) {
3020 			saved_mask = vap->va_mask;
3021 			vap->va_mask &= ~trim_mask;
3022 		}
3023 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3024 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3025 		if (err) {
3026 			ZFS_EXIT(zfsvfs);
3027 			return (err);
3028 		}
3029 
3030 		if (trim_mask)
3031 			vap->va_mask |= saved_mask;
3032 	}
3033 
3034 	/*
3035 	 * secpolicy_vnode_setattr, or take ownership may have
3036 	 * changed va_mask
3037 	 */
3038 	mask = vap->va_mask;
3039 
3040 	if ((mask & (AT_UID | AT_GID))) {
3041 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3042 		    &xattr_obj, sizeof (xattr_obj));
3043 
3044 		if (err == 0 && xattr_obj) {
3045 			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3046 			if (err)
3047 				goto out2;
3048 		}
3049 		if (mask & AT_UID) {
3050 			new_uid = zfs_fuid_create(zfsvfs,
3051 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3052 			if (new_uid != zp->z_uid &&
3053 			    zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3054 				if (attrzp)
3055 					VN_RELE(ZTOV(attrzp));
3056 				err = SET_ERROR(EDQUOT);
3057 				goto out2;
3058 			}
3059 		}
3060 
3061 		if (mask & AT_GID) {
3062 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3063 			    cr, ZFS_GROUP, &fuidp);
3064 			if (new_gid != zp->z_gid &&
3065 			    zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3066 				if (attrzp)
3067 					VN_RELE(ZTOV(attrzp));
3068 				err = SET_ERROR(EDQUOT);
3069 				goto out2;
3070 			}
3071 		}
3072 	}
3073 	tx = dmu_tx_create(zfsvfs->z_os);
3074 
3075 	if (mask & AT_MODE) {
3076 		uint64_t pmode = zp->z_mode;
3077 		uint64_t acl_obj;
3078 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3079 
3080 		if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3081 		    !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3082 			err = SET_ERROR(EPERM);
3083 			goto out;
3084 		}
3085 
3086 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3087 			goto out;
3088 
3089 		mutex_enter(&zp->z_lock);
3090 		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3091 			/*
3092 			 * Are we upgrading ACL from old V0 format
3093 			 * to V1 format?
3094 			 */
3095 			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3096 			    zfs_znode_acl_version(zp) ==
3097 			    ZFS_ACL_VERSION_INITIAL) {
3098 				dmu_tx_hold_free(tx, acl_obj, 0,
3099 				    DMU_OBJECT_END);
3100 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3101 				    0, aclp->z_acl_bytes);
3102 			} else {
3103 				dmu_tx_hold_write(tx, acl_obj, 0,
3104 				    aclp->z_acl_bytes);
3105 			}
3106 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3107 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3108 			    0, aclp->z_acl_bytes);
3109 		}
3110 		mutex_exit(&zp->z_lock);
3111 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3112 	} else {
3113 		if ((mask & AT_XVATTR) &&
3114 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3115 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3116 		else
3117 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3118 	}
3119 
3120 	if (attrzp) {
3121 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3122 	}
3123 
3124 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3125 	if (fuid_dirtied)
3126 		zfs_fuid_txhold(zfsvfs, tx);
3127 
3128 	zfs_sa_upgrade_txholds(tx, zp);
3129 
3130 	err = dmu_tx_assign(tx, TXG_WAIT);
3131 	if (err)
3132 		goto out;
3133 
3134 	count = 0;
3135 	/*
3136 	 * Set each attribute requested.
3137 	 * We group settings according to the locks they need to acquire.
3138 	 *
3139 	 * Note: you cannot set ctime directly, although it will be
3140 	 * updated as a side-effect of calling this function.
3141 	 */
3142 
3143 
3144 	if (mask & (AT_UID|AT_GID|AT_MODE))
3145 		mutex_enter(&zp->z_acl_lock);
3146 	mutex_enter(&zp->z_lock);
3147 
3148 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3149 	    &zp->z_pflags, sizeof (zp->z_pflags));
3150 
3151 	if (attrzp) {
3152 		if (mask & (AT_UID|AT_GID|AT_MODE))
3153 			mutex_enter(&attrzp->z_acl_lock);
3154 		mutex_enter(&attrzp->z_lock);
3155 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3156 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3157 		    sizeof (attrzp->z_pflags));
3158 	}
3159 
3160 	if (mask & (AT_UID|AT_GID)) {
3161 
3162 		if (mask & AT_UID) {
3163 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3164 			    &new_uid, sizeof (new_uid));
3165 			zp->z_uid = new_uid;
3166 			if (attrzp) {
3167 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3168 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3169 				    sizeof (new_uid));
3170 				attrzp->z_uid = new_uid;
3171 			}
3172 		}
3173 
3174 		if (mask & AT_GID) {
3175 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3176 			    NULL, &new_gid, sizeof (new_gid));
3177 			zp->z_gid = new_gid;
3178 			if (attrzp) {
3179 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3180 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3181 				    sizeof (new_gid));
3182 				attrzp->z_gid = new_gid;
3183 			}
3184 		}
3185 		if (!(mask & AT_MODE)) {
3186 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3187 			    NULL, &new_mode, sizeof (new_mode));
3188 			new_mode = zp->z_mode;
3189 		}
3190 		err = zfs_acl_chown_setattr(zp);
3191 		ASSERT(err == 0);
3192 		if (attrzp) {
3193 			err = zfs_acl_chown_setattr(attrzp);
3194 			ASSERT(err == 0);
3195 		}
3196 	}
3197 
3198 	if (mask & AT_MODE) {
3199 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3200 		    &new_mode, sizeof (new_mode));
3201 		zp->z_mode = new_mode;
3202 		ASSERT3U((uintptr_t)aclp, !=, NULL);
3203 		err = zfs_aclset_common(zp, aclp, cr, tx);
3204 		ASSERT0(err);
3205 		if (zp->z_acl_cached)
3206 			zfs_acl_free(zp->z_acl_cached);
3207 		zp->z_acl_cached = aclp;
3208 		aclp = NULL;
3209 	}
3210 
3211 
3212 	if (mask & AT_ATIME) {
3213 		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3214 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3215 		    &zp->z_atime, sizeof (zp->z_atime));
3216 	}
3217 
3218 	if (mask & AT_MTIME) {
3219 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3220 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3221 		    mtime, sizeof (mtime));
3222 	}
3223 
3224 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3225 	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3226 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3227 		    NULL, mtime, sizeof (mtime));
3228 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3229 		    &ctime, sizeof (ctime));
3230 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3231 		    B_TRUE);
3232 	} else if (mask != 0) {
3233 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3234 		    &ctime, sizeof (ctime));
3235 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3236 		    B_TRUE);
3237 		if (attrzp) {
3238 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3239 			    SA_ZPL_CTIME(zfsvfs), NULL,
3240 			    &ctime, sizeof (ctime));
3241 			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3242 			    mtime, ctime, B_TRUE);
3243 		}
3244 	}
3245 	/*
3246 	 * Do this after setting timestamps to prevent timestamp
3247 	 * update from toggling bit
3248 	 */
3249 
3250 	if (xoap && (mask & AT_XVATTR)) {
3251 
3252 		/*
3253 		 * restore trimmed off masks
3254 		 * so that return masks can be set for caller.
3255 		 */
3256 
3257 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3258 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
3259 		}
3260 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3261 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
3262 		}
3263 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3264 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3265 		}
3266 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3267 			XVA_SET_REQ(xvap, XAT_NODUMP);
3268 		}
3269 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3270 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3271 		}
3272 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3273 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3274 		}
3275 
3276 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3277 			ASSERT(vp->v_type == VREG);
3278 
3279 		zfs_xvattr_set(zp, xvap, tx);
3280 	}
3281 
3282 	if (fuid_dirtied)
3283 		zfs_fuid_sync(zfsvfs, tx);
3284 
3285 	if (mask != 0)
3286 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3287 
3288 	mutex_exit(&zp->z_lock);
3289 	if (mask & (AT_UID|AT_GID|AT_MODE))
3290 		mutex_exit(&zp->z_acl_lock);
3291 
3292 	if (attrzp) {
3293 		if (mask & (AT_UID|AT_GID|AT_MODE))
3294 			mutex_exit(&attrzp->z_acl_lock);
3295 		mutex_exit(&attrzp->z_lock);
3296 	}
3297 out:
3298 	if (err == 0 && attrzp) {
3299 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3300 		    xattr_count, tx);
3301 		ASSERT(err2 == 0);
3302 	}
3303 
3304 	if (attrzp)
3305 		VN_RELE(ZTOV(attrzp));
3306 
3307 	if (aclp)
3308 		zfs_acl_free(aclp);
3309 
3310 	if (fuidp) {
3311 		zfs_fuid_info_free(fuidp);
3312 		fuidp = NULL;
3313 	}
3314 
3315 	if (err) {
3316 		dmu_tx_abort(tx);
3317 		if (err == ERESTART)
3318 			goto top;
3319 	} else {
3320 		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3321 		dmu_tx_commit(tx);
3322 	}
3323 
3324 out2:
3325 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3326 		zil_commit(zilog, 0);
3327 
3328 	ZFS_EXIT(zfsvfs);
3329 	return (err);
3330 }
3331 
3332 typedef struct zfs_zlock {
3333 	krwlock_t	*zl_rwlock;	/* lock we acquired */
3334 	znode_t		*zl_znode;	/* znode we held */
3335 	struct zfs_zlock *zl_next;	/* next in list */
3336 } zfs_zlock_t;
3337 
3338 /*
3339  * Drop locks and release vnodes that were held by zfs_rename_lock().
3340  */
3341 static void
3342 zfs_rename_unlock(zfs_zlock_t **zlpp)
3343 {
3344 	zfs_zlock_t *zl;
3345 
3346 	while ((zl = *zlpp) != NULL) {
3347 		if (zl->zl_znode != NULL)
3348 			VN_RELE(ZTOV(zl->zl_znode));
3349 		rw