xref: /titanic_50/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision cfe63ebf3593a7e45618955bfd1a754f30fc5fcf)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012 by Delphix. All rights reserved.
24  */
25 
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2010 Robert Milkowski */
28 
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/time.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
35 #include <sys/vfs.h>
36 #include <sys/vfs_opreg.h>
37 #include <sys/vnode.h>
38 #include <sys/file.h>
39 #include <sys/stat.h>
40 #include <sys/kmem.h>
41 #include <sys/taskq.h>
42 #include <sys/uio.h>
43 #include <sys/vmsystm.h>
44 #include <sys/atomic.h>
45 #include <sys/vm.h>
46 #include <vm/seg_vn.h>
47 #include <vm/pvn.h>
48 #include <vm/as.h>
49 #include <vm/kpm.h>
50 #include <vm/seg_kpm.h>
51 #include <sys/mman.h>
52 #include <sys/pathname.h>
53 #include <sys/cmn_err.h>
54 #include <sys/errno.h>
55 #include <sys/unistd.h>
56 #include <sys/zfs_dir.h>
57 #include <sys/zfs_acl.h>
58 #include <sys/zfs_ioctl.h>
59 #include <sys/fs/zfs.h>
60 #include <sys/dmu.h>
61 #include <sys/dmu_objset.h>
62 #include <sys/spa.h>
63 #include <sys/txg.h>
64 #include <sys/dbuf.h>
65 #include <sys/zap.h>
66 #include <sys/sa.h>
67 #include <sys/dirent.h>
68 #include <sys/policy.h>
69 #include <sys/sunddi.h>
70 #include <sys/filio.h>
71 #include <sys/sid.h>
72 #include "fs/fs_subr.h"
73 #include <sys/zfs_ctldir.h>
74 #include <sys/zfs_fuid.h>
75 #include <sys/zfs_sa.h>
76 #include <sys/dnlc.h>
77 #include <sys/zfs_rlock.h>
78 #include <sys/extdirent.h>
79 #include <sys/kidmap.h>
80 #include <sys/cred.h>
81 #include <sys/attr.h>
82 
83 /*
84  * Programming rules.
85  *
86  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
87  * properly lock its in-core state, create a DMU transaction, do the work,
88  * record this work in the intent log (ZIL), commit the DMU transaction,
89  * and wait for the intent log to commit if it is a synchronous operation.
90  * Moreover, the vnode ops must work in both normal and log replay context.
91  * The ordering of events is important to avoid deadlocks and references
92  * to freed memory.  The example below illustrates the following Big Rules:
93  *
94  *  (1) A check must be made in each zfs thread for a mounted file system.
95  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
96  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
97  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
98  *      can return EIO from the calling function.
99  *
100  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
101  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
102  *	First, if it's the last reference, the vnode/znode
103  *	can be freed, so the zp may point to freed memory.  Second, the last
104  *	reference will call zfs_zinactive(), which may induce a lot of work --
105  *	pushing cached pages (which acquires range locks) and syncing out
106  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
107  *	which could deadlock the system if you were already holding one.
108  *	If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
109  *
110  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
111  *	as they can span dmu_tx_assign() calls.
112  *
113  *  (4)	Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
114  *	This is critical because we don't want to block while holding locks.
115  *	Note, in particular, that if a lock is sometimes acquired before
116  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
117  *	use a non-blocking assign can deadlock the system.  The scenario:
118  *
119  *	Thread A has grabbed a lock before calling dmu_tx_assign().
120  *	Thread B is in an already-assigned tx, and blocks for this lock.
121  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
122  *	forever, because the previous txg can't quiesce until B's tx commits.
123  *
124  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
125  *	then drop all locks, call dmu_tx_wait(), and try again.
126  *
127  *  (5)	If the operation succeeded, generate the intent log entry for it
128  *	before dropping locks.  This ensures that the ordering of events
129  *	in the intent log matches the order in which they actually occurred.
130  *      During ZIL replay the zfs_log_* functions will update the sequence
131  *	number to indicate the zil transaction has replayed.
132  *
133  *  (6)	At the end of each vnode op, the DMU tx must always commit,
134  *	regardless of whether there were any errors.
135  *
136  *  (7)	After dropping all locks, invoke zil_commit(zilog, foid)
137  *	to ensure that synchronous semantics are provided when necessary.
138  *
139  * In general, this is how things should be ordered in each vnode op:
140  *
141  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
142  * top:
143  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
144  *	rw_enter(...);			// grab any other locks you need
145  *	tx = dmu_tx_create(...);	// get DMU tx
146  *	dmu_tx_hold_*();		// hold each object you might modify
147  *	error = dmu_tx_assign(tx, TXG_NOWAIT);	// try to assign
148  *	if (error) {
149  *		rw_exit(...);		// drop locks
150  *		zfs_dirent_unlock(dl);	// unlock directory entry
151  *		VN_RELE(...);		// release held vnodes
152  *		if (error == ERESTART) {
153  *			dmu_tx_wait(tx);
154  *			dmu_tx_abort(tx);
155  *			goto top;
156  *		}
157  *		dmu_tx_abort(tx);	// abort DMU tx
158  *		ZFS_EXIT(zfsvfs);	// finished in zfs
159  *		return (error);		// really out of space
160  *	}
161  *	error = do_real_work();		// do whatever this VOP does
162  *	if (error == 0)
163  *		zfs_log_*(...);		// on success, make ZIL entry
164  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
165  *	rw_exit(...);			// drop locks
166  *	zfs_dirent_unlock(dl);		// unlock directory entry
167  *	VN_RELE(...);			// release held vnodes
168  *	zil_commit(zilog, foid);	// synchronous when necessary
169  *	ZFS_EXIT(zfsvfs);		// finished in zfs
170  *	return (error);			// done, report error
171  */
172 
173 /* ARGSUSED */
174 static int
175 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
176 {
177 	znode_t	*zp = VTOZ(*vpp);
178 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
179 
180 	ZFS_ENTER(zfsvfs);
181 	ZFS_VERIFY_ZP(zp);
182 
183 	if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
184 	    ((flag & FAPPEND) == 0)) {
185 		ZFS_EXIT(zfsvfs);
186 		return (EPERM);
187 	}
188 
189 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
190 	    ZTOV(zp)->v_type == VREG &&
191 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
192 		if (fs_vscan(*vpp, cr, 0) != 0) {
193 			ZFS_EXIT(zfsvfs);
194 			return (EACCES);
195 		}
196 	}
197 
198 	/* Keep a count of the synchronous opens in the znode */
199 	if (flag & (FSYNC | FDSYNC))
200 		atomic_inc_32(&zp->z_sync_cnt);
201 
202 	ZFS_EXIT(zfsvfs);
203 	return (0);
204 }
205 
206 /* ARGSUSED */
207 static int
208 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
209     caller_context_t *ct)
210 {
211 	znode_t	*zp = VTOZ(vp);
212 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
213 
214 	/*
215 	 * Clean up any locks held by this process on the vp.
216 	 */
217 	cleanlocks(vp, ddi_get_pid(), 0);
218 	cleanshares(vp, ddi_get_pid());
219 
220 	ZFS_ENTER(zfsvfs);
221 	ZFS_VERIFY_ZP(zp);
222 
223 	/* Decrement the synchronous opens in the znode */
224 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
225 		atomic_dec_32(&zp->z_sync_cnt);
226 
227 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
228 	    ZTOV(zp)->v_type == VREG &&
229 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
230 		VERIFY(fs_vscan(vp, cr, 1) == 0);
231 
232 	ZFS_EXIT(zfsvfs);
233 	return (0);
234 }
235 
236 /*
237  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
238  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
239  */
240 static int
241 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
242 {
243 	znode_t	*zp = VTOZ(vp);
244 	uint64_t noff = (uint64_t)*off; /* new offset */
245 	uint64_t file_sz;
246 	int error;
247 	boolean_t hole;
248 
249 	file_sz = zp->z_size;
250 	if (noff >= file_sz)  {
251 		return (ENXIO);
252 	}
253 
254 	if (cmd == _FIO_SEEK_HOLE)
255 		hole = B_TRUE;
256 	else
257 		hole = B_FALSE;
258 
259 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
260 
261 	/* end of file? */
262 	if ((error == ESRCH) || (noff > file_sz)) {
263 		/*
264 		 * Handle the virtual hole at the end of file.
265 		 */
266 		if (hole) {
267 			*off = file_sz;
268 			return (0);
269 		}
270 		return (ENXIO);
271 	}
272 
273 	if (noff < *off)
274 		return (error);
275 	*off = noff;
276 	return (error);
277 }
278 
279 /* ARGSUSED */
280 static int
281 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
282     int *rvalp, caller_context_t *ct)
283 {
284 	offset_t off;
285 	int error;
286 	zfsvfs_t *zfsvfs;
287 	znode_t *zp;
288 
289 	switch (com) {
290 	case _FIOFFS:
291 		return (zfs_sync(vp->v_vfsp, 0, cred));
292 
293 		/*
294 		 * The following two ioctls are used by bfu.  Faking out,
295 		 * necessary to avoid bfu errors.
296 		 */
297 	case _FIOGDIO:
298 	case _FIOSDIO:
299 		return (0);
300 
301 	case _FIO_SEEK_DATA:
302 	case _FIO_SEEK_HOLE:
303 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
304 			return (EFAULT);
305 
306 		zp = VTOZ(vp);
307 		zfsvfs = zp->z_zfsvfs;
308 		ZFS_ENTER(zfsvfs);
309 		ZFS_VERIFY_ZP(zp);
310 
311 		/* offset parameter is in/out */
312 		error = zfs_holey(vp, com, &off);
313 		ZFS_EXIT(zfsvfs);
314 		if (error)
315 			return (error);
316 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
317 			return (EFAULT);
318 		return (0);
319 	}
320 	return (ENOTTY);
321 }
322 
323 /*
324  * Utility functions to map and unmap a single physical page.  These
325  * are used to manage the mappable copies of ZFS file data, and therefore
326  * do not update ref/mod bits.
327  */
328 caddr_t
329 zfs_map_page(page_t *pp, enum seg_rw rw)
330 {
331 	if (kpm_enable)
332 		return (hat_kpm_mapin(pp, 0));
333 	ASSERT(rw == S_READ || rw == S_WRITE);
334 	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
335 	    (caddr_t)-1));
336 }
337 
338 void
339 zfs_unmap_page(page_t *pp, caddr_t addr)
340 {
341 	if (kpm_enable) {
342 		hat_kpm_mapout(pp, 0, addr);
343 	} else {
344 		ppmapout(addr);
345 	}
346 }
347 
348 /*
349  * When a file is memory mapped, we must keep the IO data synchronized
350  * between the DMU cache and the memory mapped pages.  What this means:
351  *
352  * On Write:	If we find a memory mapped page, we write to *both*
353  *		the page and the dmu buffer.
354  */
355 static void
356 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
357 {
358 	int64_t	off;
359 
360 	off = start & PAGEOFFSET;
361 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
362 		page_t *pp;
363 		uint64_t nbytes = MIN(PAGESIZE - off, len);
364 
365 		if (pp = page_lookup(vp, start, SE_SHARED)) {
366 			caddr_t va;
367 
368 			va = zfs_map_page(pp, S_WRITE);
369 			(void) dmu_read(os, oid, start+off, nbytes, va+off,
370 			    DMU_READ_PREFETCH);
371 			zfs_unmap_page(pp, va);
372 			page_unlock(pp);
373 		}
374 		len -= nbytes;
375 		off = 0;
376 	}
377 }
378 
379 /*
380  * When a file is memory mapped, we must keep the IO data synchronized
381  * between the DMU cache and the memory mapped pages.  What this means:
382  *
383  * On Read:	We "read" preferentially from memory mapped pages,
384  *		else we default from the dmu buffer.
385  *
386  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
387  *	the file is memory mapped.
388  */
389 static int
390 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
391 {
392 	znode_t *zp = VTOZ(vp);
393 	objset_t *os = zp->z_zfsvfs->z_os;
394 	int64_t	start, off;
395 	int len = nbytes;
396 	int error = 0;
397 
398 	start = uio->uio_loffset;
399 	off = start & PAGEOFFSET;
400 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
401 		page_t *pp;
402 		uint64_t bytes = MIN(PAGESIZE - off, len);
403 
404 		if (pp = page_lookup(vp, start, SE_SHARED)) {
405 			caddr_t va;
406 
407 			va = zfs_map_page(pp, S_READ);
408 			error = uiomove(va + off, bytes, UIO_READ, uio);
409 			zfs_unmap_page(pp, va);
410 			page_unlock(pp);
411 		} else {
412 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
413 		}
414 		len -= bytes;
415 		off = 0;
416 		if (error)
417 			break;
418 	}
419 	return (error);
420 }
421 
422 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
423 
424 /*
425  * Read bytes from specified file into supplied buffer.
426  *
427  *	IN:	vp	- vnode of file to be read from.
428  *		uio	- structure supplying read location, range info,
429  *			  and return buffer.
430  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
431  *		cr	- credentials of caller.
432  *		ct	- caller context
433  *
434  *	OUT:	uio	- updated offset and range, buffer filled.
435  *
436  *	RETURN:	0 if success
437  *		error code if failure
438  *
439  * Side Effects:
440  *	vp - atime updated if byte count > 0
441  */
442 /* ARGSUSED */
443 static int
444 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
445 {
446 	znode_t		*zp = VTOZ(vp);
447 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
448 	objset_t	*os;
449 	ssize_t		n, nbytes;
450 	int		error = 0;
451 	rl_t		*rl;
452 	xuio_t		*xuio = NULL;
453 
454 	ZFS_ENTER(zfsvfs);
455 	ZFS_VERIFY_ZP(zp);
456 	os = zfsvfs->z_os;
457 
458 	if (zp->z_pflags & ZFS_AV_QUARANTINED) {
459 		ZFS_EXIT(zfsvfs);
460 		return (EACCES);
461 	}
462 
463 	/*
464 	 * Validate file offset
465 	 */
466 	if (uio->uio_loffset < (offset_t)0) {
467 		ZFS_EXIT(zfsvfs);
468 		return (EINVAL);
469 	}
470 
471 	/*
472 	 * Fasttrack empty reads
473 	 */
474 	if (uio->uio_resid == 0) {
475 		ZFS_EXIT(zfsvfs);
476 		return (0);
477 	}
478 
479 	/*
480 	 * Check for mandatory locks
481 	 */
482 	if (MANDMODE(zp->z_mode)) {
483 		if (error = chklock(vp, FREAD,
484 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
485 			ZFS_EXIT(zfsvfs);
486 			return (error);
487 		}
488 	}
489 
490 	/*
491 	 * If we're in FRSYNC mode, sync out this znode before reading it.
492 	 */
493 	if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
494 		zil_commit(zfsvfs->z_log, zp->z_id);
495 
496 	/*
497 	 * Lock the range against changes.
498 	 */
499 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
500 
501 	/*
502 	 * If we are reading past end-of-file we can skip
503 	 * to the end; but we might still need to set atime.
504 	 */
505 	if (uio->uio_loffset >= zp->z_size) {
506 		error = 0;
507 		goto out;
508 	}
509 
510 	ASSERT(uio->uio_loffset < zp->z_size);
511 	n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
512 
513 	if ((uio->uio_extflg == UIO_XUIO) &&
514 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
515 		int nblk;
516 		int blksz = zp->z_blksz;
517 		uint64_t offset = uio->uio_loffset;
518 
519 		xuio = (xuio_t *)uio;
520 		if ((ISP2(blksz))) {
521 			nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
522 			    blksz)) / blksz;
523 		} else {
524 			ASSERT(offset + n <= blksz);
525 			nblk = 1;
526 		}
527 		(void) dmu_xuio_init(xuio, nblk);
528 
529 		if (vn_has_cached_data(vp)) {
530 			/*
531 			 * For simplicity, we always allocate a full buffer
532 			 * even if we only expect to read a portion of a block.
533 			 */
534 			while (--nblk >= 0) {
535 				(void) dmu_xuio_add(xuio,
536 				    dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
537 				    blksz), 0, blksz);
538 			}
539 		}
540 	}
541 
542 	while (n > 0) {
543 		nbytes = MIN(n, zfs_read_chunk_size -
544 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
545 
546 		if (vn_has_cached_data(vp))
547 			error = mappedread(vp, nbytes, uio);
548 		else
549 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
550 		if (error) {
551 			/* convert checksum errors into IO errors */
552 			if (error == ECKSUM)
553 				error = EIO;
554 			break;
555 		}
556 
557 		n -= nbytes;
558 	}
559 out:
560 	zfs_range_unlock(rl);
561 
562 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
563 	ZFS_EXIT(zfsvfs);
564 	return (error);
565 }
566 
567 /*
568  * Write the bytes to a file.
569  *
570  *	IN:	vp	- vnode of file to be written to.
571  *		uio	- structure supplying write location, range info,
572  *			  and data buffer.
573  *		ioflag	- FAPPEND flag set if in append mode.
574  *		cr	- credentials of caller.
575  *		ct	- caller context (NFS/CIFS fem monitor only)
576  *
577  *	OUT:	uio	- updated offset and range.
578  *
579  *	RETURN:	0 if success
580  *		error code if failure
581  *
582  * Timestamps:
583  *	vp - ctime|mtime updated if byte count > 0
584  */
585 
586 /* ARGSUSED */
587 static int
588 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
589 {
590 	znode_t		*zp = VTOZ(vp);
591 	rlim64_t	limit = uio->uio_llimit;
592 	ssize_t		start_resid = uio->uio_resid;
593 	ssize_t		tx_bytes;
594 	uint64_t	end_size;
595 	dmu_tx_t	*tx;
596 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
597 	zilog_t		*zilog;
598 	offset_t	woff;
599 	ssize_t		n, nbytes;
600 	rl_t		*rl;
601 	int		max_blksz = zfsvfs->z_max_blksz;
602 	int		error = 0;
603 	arc_buf_t	*abuf;
604 	iovec_t		*aiov = NULL;
605 	xuio_t		*xuio = NULL;
606 	int		i_iov = 0;
607 	int		iovcnt = uio->uio_iovcnt;
608 	iovec_t		*iovp = uio->uio_iov;
609 	int		write_eof;
610 	int		count = 0;
611 	sa_bulk_attr_t	bulk[4];
612 	uint64_t	mtime[2], ctime[2];
613 
614 	/*
615 	 * Fasttrack empty write
616 	 */
617 	n = start_resid;
618 	if (n == 0)
619 		return (0);
620 
621 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
622 		limit = MAXOFFSET_T;
623 
624 	ZFS_ENTER(zfsvfs);
625 	ZFS_VERIFY_ZP(zp);
626 
627 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
628 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
629 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
630 	    &zp->z_size, 8);
631 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
632 	    &zp->z_pflags, 8);
633 
634 	/*
635 	 * If immutable or not appending then return EPERM
636 	 */
637 	if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
638 	    ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
639 	    (uio->uio_loffset < zp->z_size))) {
640 		ZFS_EXIT(zfsvfs);
641 		return (EPERM);
642 	}
643 
644 	zilog = zfsvfs->z_log;
645 
646 	/*
647 	 * Validate file offset
648 	 */
649 	woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
650 	if (woff < 0) {
651 		ZFS_EXIT(zfsvfs);
652 		return (EINVAL);
653 	}
654 
655 	/*
656 	 * Check for mandatory locks before calling zfs_range_lock()
657 	 * in order to prevent a deadlock with locks set via fcntl().
658 	 */
659 	if (MANDMODE((mode_t)zp->z_mode) &&
660 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
661 		ZFS_EXIT(zfsvfs);
662 		return (error);
663 	}
664 
665 	/*
666 	 * Pre-fault the pages to ensure slow (eg NFS) pages
667 	 * don't hold up txg.
668 	 * Skip this if uio contains loaned arc_buf.
669 	 */
670 	if ((uio->uio_extflg == UIO_XUIO) &&
671 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
672 		xuio = (xuio_t *)uio;
673 	else
674 		uio_prefaultpages(MIN(n, max_blksz), uio);
675 
676 	/*
677 	 * If in append mode, set the io offset pointer to eof.
678 	 */
679 	if (ioflag & FAPPEND) {
680 		/*
681 		 * Obtain an appending range lock to guarantee file append
682 		 * semantics.  We reset the write offset once we have the lock.
683 		 */
684 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
685 		woff = rl->r_off;
686 		if (rl->r_len == UINT64_MAX) {
687 			/*
688 			 * We overlocked the file because this write will cause
689 			 * the file block size to increase.
690 			 * Note that zp_size cannot change with this lock held.
691 			 */
692 			woff = zp->z_size;
693 		}
694 		uio->uio_loffset = woff;
695 	} else {
696 		/*
697 		 * Note that if the file block size will change as a result of
698 		 * this write, then this range lock will lock the entire file
699 		 * so that we can re-write the block safely.
700 		 */
701 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
702 	}
703 
704 	if (woff >= limit) {
705 		zfs_range_unlock(rl);
706 		ZFS_EXIT(zfsvfs);
707 		return (EFBIG);
708 	}
709 
710 	if ((woff + n) > limit || woff > (limit - n))
711 		n = limit - woff;
712 
713 	/* Will this write extend the file length? */
714 	write_eof = (woff + n > zp->z_size);
715 
716 	end_size = MAX(zp->z_size, woff + n);
717 
718 	/*
719 	 * Write the file in reasonable size chunks.  Each chunk is written
720 	 * in a separate transaction; this keeps the intent log records small
721 	 * and allows us to do more fine-grained space accounting.
722 	 */
723 	while (n > 0) {
724 		abuf = NULL;
725 		woff = uio->uio_loffset;
726 again:
727 		if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
728 		    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
729 			if (abuf != NULL)
730 				dmu_return_arcbuf(abuf);
731 			error = EDQUOT;
732 			break;
733 		}
734 
735 		if (xuio && abuf == NULL) {
736 			ASSERT(i_iov < iovcnt);
737 			aiov = &iovp[i_iov];
738 			abuf = dmu_xuio_arcbuf(xuio, i_iov);
739 			dmu_xuio_clear(xuio, i_iov);
740 			DTRACE_PROBE3(zfs_cp_write, int, i_iov,
741 			    iovec_t *, aiov, arc_buf_t *, abuf);
742 			ASSERT((aiov->iov_base == abuf->b_data) ||
743 			    ((char *)aiov->iov_base - (char *)abuf->b_data +
744 			    aiov->iov_len == arc_buf_size(abuf)));
745 			i_iov++;
746 		} else if (abuf == NULL && n >= max_blksz &&
747 		    woff >= zp->z_size &&
748 		    P2PHASE(woff, max_blksz) == 0 &&
749 		    zp->z_blksz == max_blksz) {
750 			/*
751 			 * This write covers a full block.  "Borrow" a buffer
752 			 * from the dmu so that we can fill it before we enter
753 			 * a transaction.  This avoids the possibility of
754 			 * holding up the transaction if the data copy hangs
755 			 * up on a pagefault (e.g., from an NFS server mapping).
756 			 */
757 			size_t cbytes;
758 
759 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
760 			    max_blksz);
761 			ASSERT(abuf != NULL);
762 			ASSERT(arc_buf_size(abuf) == max_blksz);
763 			if (error = uiocopy(abuf->b_data, max_blksz,
764 			    UIO_WRITE, uio, &cbytes)) {
765 				dmu_return_arcbuf(abuf);
766 				break;
767 			}
768 			ASSERT(cbytes == max_blksz);
769 		}
770 
771 		/*
772 		 * Start a transaction.
773 		 */
774 		tx = dmu_tx_create(zfsvfs->z_os);
775 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
776 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
777 		zfs_sa_upgrade_txholds(tx, zp);
778 		error = dmu_tx_assign(tx, TXG_NOWAIT);
779 		if (error) {
780 			if (error == ERESTART) {
781 				dmu_tx_wait(tx);
782 				dmu_tx_abort(tx);
783 				goto again;
784 			}
785 			dmu_tx_abort(tx);
786 			if (abuf != NULL)
787 				dmu_return_arcbuf(abuf);
788 			break;
789 		}
790 
791 		/*
792 		 * If zfs_range_lock() over-locked we grow the blocksize
793 		 * and then reduce the lock range.  This will only happen
794 		 * on the first iteration since zfs_range_reduce() will
795 		 * shrink down r_len to the appropriate size.
796 		 */
797 		if (rl->r_len == UINT64_MAX) {
798 			uint64_t new_blksz;
799 
800 			if (zp->z_blksz > max_blksz) {
801 				ASSERT(!ISP2(zp->z_blksz));
802 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
803 			} else {
804 				new_blksz = MIN(end_size, max_blksz);
805 			}
806 			zfs_grow_blocksize(zp, new_blksz, tx);
807 			zfs_range_reduce(rl, woff, n);
808 		}
809 
810 		/*
811 		 * XXX - should we really limit each write to z_max_blksz?
812 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
813 		 */
814 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
815 
816 		if (abuf == NULL) {
817 			tx_bytes = uio->uio_resid;
818 			error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
819 			    uio, nbytes, tx);
820 			tx_bytes -= uio->uio_resid;
821 		} else {
822 			tx_bytes = nbytes;
823 			ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
824 			/*
825 			 * If this is not a full block write, but we are
826 			 * extending the file past EOF and this data starts
827 			 * block-aligned, use assign_arcbuf().  Otherwise,
828 			 * write via dmu_write().
829 			 */
830 			if (tx_bytes < max_blksz && (!write_eof ||
831 			    aiov->iov_base != abuf->b_data)) {
832 				ASSERT(xuio);
833 				dmu_write(zfsvfs->z_os, zp->z_id, woff,
834 				    aiov->iov_len, aiov->iov_base, tx);
835 				dmu_return_arcbuf(abuf);
836 				xuio_stat_wbuf_copied();
837 			} else {
838 				ASSERT(xuio || tx_bytes == max_blksz);
839 				dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
840 				    woff, abuf, tx);
841 			}
842 			ASSERT(tx_bytes <= uio->uio_resid);
843 			uioskip(uio, tx_bytes);
844 		}
845 		if (tx_bytes && vn_has_cached_data(vp)) {
846 			update_pages(vp, woff,
847 			    tx_bytes, zfsvfs->z_os, zp->z_id);
848 		}
849 
850 		/*
851 		 * If we made no progress, we're done.  If we made even
852 		 * partial progress, update the znode and ZIL accordingly.
853 		 */
854 		if (tx_bytes == 0) {
855 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
856 			    (void *)&zp->z_size, sizeof (uint64_t), tx);
857 			dmu_tx_commit(tx);
858 			ASSERT(error != 0);
859 			break;
860 		}
861 
862 		/*
863 		 * Clear Set-UID/Set-GID bits on successful write if not
864 		 * privileged and at least one of the excute bits is set.
865 		 *
866 		 * It would be nice to to this after all writes have
867 		 * been done, but that would still expose the ISUID/ISGID
868 		 * to another app after the partial write is committed.
869 		 *
870 		 * Note: we don't call zfs_fuid_map_id() here because
871 		 * user 0 is not an ephemeral uid.
872 		 */
873 		mutex_enter(&zp->z_acl_lock);
874 		if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
875 		    (S_IXUSR >> 6))) != 0 &&
876 		    (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
877 		    secpolicy_vnode_setid_retain(cr,
878 		    (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
879 			uint64_t newmode;
880 			zp->z_mode &= ~(S_ISUID | S_ISGID);
881 			newmode = zp->z_mode;
882 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
883 			    (void *)&newmode, sizeof (uint64_t), tx);
884 		}
885 		mutex_exit(&zp->z_acl_lock);
886 
887 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
888 		    B_TRUE);
889 
890 		/*
891 		 * Update the file size (zp_size) if it has changed;
892 		 * account for possible concurrent updates.
893 		 */
894 		while ((end_size = zp->z_size) < uio->uio_loffset) {
895 			(void) atomic_cas_64(&zp->z_size, end_size,
896 			    uio->uio_loffset);
897 			ASSERT(error == 0);
898 		}
899 		/*
900 		 * If we are replaying and eof is non zero then force
901 		 * the file size to the specified eof. Note, there's no
902 		 * concurrency during replay.
903 		 */
904 		if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
905 			zp->z_size = zfsvfs->z_replay_eof;
906 
907 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
908 
909 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
910 		dmu_tx_commit(tx);
911 
912 		if (error != 0)
913 			break;
914 		ASSERT(tx_bytes == nbytes);
915 		n -= nbytes;
916 
917 		if (!xuio && n > 0)
918 			uio_prefaultpages(MIN(n, max_blksz), uio);
919 	}
920 
921 	zfs_range_unlock(rl);
922 
923 	/*
924 	 * If we're in replay mode, or we made no progress, return error.
925 	 * Otherwise, it's at least a partial write, so it's successful.
926 	 */
927 	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
928 		ZFS_EXIT(zfsvfs);
929 		return (error);
930 	}
931 
932 	if (ioflag & (FSYNC | FDSYNC) ||
933 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
934 		zil_commit(zilog, zp->z_id);
935 
936 	ZFS_EXIT(zfsvfs);
937 	return (0);
938 }
939 
940 void
941 zfs_get_done(zgd_t *zgd, int error)
942 {
943 	znode_t *zp = zgd->zgd_private;
944 	objset_t *os = zp->z_zfsvfs->z_os;
945 
946 	if (zgd->zgd_db)
947 		dmu_buf_rele(zgd->zgd_db, zgd);
948 
949 	zfs_range_unlock(zgd->zgd_rl);
950 
951 	/*
952 	 * Release the vnode asynchronously as we currently have the
953 	 * txg stopped from syncing.
954 	 */
955 	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
956 
957 	if (error == 0 && zgd->zgd_bp)
958 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
959 
960 	kmem_free(zgd, sizeof (zgd_t));
961 }
962 
963 #ifdef DEBUG
964 static int zil_fault_io = 0;
965 #endif
966 
967 /*
968  * Get data to generate a TX_WRITE intent log record.
969  */
970 int
971 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
972 {
973 	zfsvfs_t *zfsvfs = arg;
974 	objset_t *os = zfsvfs->z_os;
975 	znode_t *zp;
976 	uint64_t object = lr->lr_foid;
977 	uint64_t offset = lr->lr_offset;
978 	uint64_t size = lr->lr_length;
979 	blkptr_t *bp = &lr->lr_blkptr;
980 	dmu_buf_t *db;
981 	zgd_t *zgd;
982 	int error = 0;
983 
984 	ASSERT(zio != NULL);
985 	ASSERT(size != 0);
986 
987 	/*
988 	 * Nothing to do if the file has been removed
989 	 */
990 	if (zfs_zget(zfsvfs, object, &zp) != 0)
991 		return (ENOENT);
992 	if (zp->z_unlinked) {
993 		/*
994 		 * Release the vnode asynchronously as we currently have the
995 		 * txg stopped from syncing.
996 		 */
997 		VN_RELE_ASYNC(ZTOV(zp),
998 		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
999 		return (ENOENT);
1000 	}
1001 
1002 	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1003 	zgd->zgd_zilog = zfsvfs->z_log;
1004 	zgd->zgd_private = zp;
1005 
1006 	/*
1007 	 * Write records come in two flavors: immediate and indirect.
1008 	 * For small writes it's cheaper to store the data with the
1009 	 * log record (immediate); for large writes it's cheaper to
1010 	 * sync the data and get a pointer to it (indirect) so that
1011 	 * we don't have to write the data twice.
1012 	 */
1013 	if (buf != NULL) { /* immediate write */
1014 		zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1015 		/* test for truncation needs to be done while range locked */
1016 		if (offset >= zp->z_size) {
1017 			error = ENOENT;
1018 		} else {
1019 			error = dmu_read(os, object, offset, size, buf,
1020 			    DMU_READ_NO_PREFETCH);
1021 		}
1022 		ASSERT(error == 0 || error == ENOENT);
1023 	} else { /* indirect write */
1024 		/*
1025 		 * Have to lock the whole block to ensure when it's
1026 		 * written out and it's checksum is being calculated
1027 		 * that no one can change the data. We need to re-check
1028 		 * blocksize after we get the lock in case it's changed!
1029 		 */
1030 		for (;;) {
1031 			uint64_t blkoff;
1032 			size = zp->z_blksz;
1033 			blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1034 			offset -= blkoff;
1035 			zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1036 			    RL_READER);
1037 			if (zp->z_blksz == size)
1038 				break;
1039 			offset += blkoff;
1040 			zfs_range_unlock(zgd->zgd_rl);
1041 		}
1042 		/* test for truncation needs to be done while range locked */
1043 		if (lr->lr_offset >= zp->z_size)
1044 			error = ENOENT;
1045 #ifdef DEBUG
1046 		if (zil_fault_io) {
1047 			error = EIO;
1048 			zil_fault_io = 0;
1049 		}
1050 #endif
1051 		if (error == 0)
1052 			error = dmu_buf_hold(os, object, offset, zgd, &db,
1053 			    DMU_READ_NO_PREFETCH);
1054 
1055 		if (error == 0) {
1056 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1057 			if (obp) {
1058 				ASSERT(BP_IS_HOLE(bp));
1059 				*bp = *obp;
1060 			}
1061 
1062 			zgd->zgd_db = db;
1063 			zgd->zgd_bp = bp;
1064 
1065 			ASSERT(db->db_offset == offset);
1066 			ASSERT(db->db_size == size);
1067 
1068 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1069 			    zfs_get_done, zgd);
1070 			ASSERT(error || lr->lr_length <= zp->z_blksz);
1071 
1072 			/*
1073 			 * On success, we need to wait for the write I/O
1074 			 * initiated by dmu_sync() to complete before we can
1075 			 * release this dbuf.  We will finish everything up
1076 			 * in the zfs_get_done() callback.
1077 			 */
1078 			if (error == 0)
1079 				return (0);
1080 
1081 			if (error == EALREADY) {
1082 				lr->lr_common.lrc_txtype = TX_WRITE2;
1083 				error = 0;
1084 			}
1085 		}
1086 	}
1087 
1088 	zfs_get_done(zgd, error);
1089 
1090 	return (error);
1091 }
1092 
1093 /*ARGSUSED*/
1094 static int
1095 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1096     caller_context_t *ct)
1097 {
1098 	znode_t *zp = VTOZ(vp);
1099 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1100 	int error;
1101 
1102 	ZFS_ENTER(zfsvfs);
1103 	ZFS_VERIFY_ZP(zp);
1104 
1105 	if (flag & V_ACE_MASK)
1106 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1107 	else
1108 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
1109 
1110 	ZFS_EXIT(zfsvfs);
1111 	return (error);
1112 }
1113 
1114 /*
1115  * If vnode is for a device return a specfs vnode instead.
1116  */
1117 static int
1118 specvp_check(vnode_t **vpp, cred_t *cr)
1119 {
1120 	int error = 0;
1121 
1122 	if (IS_DEVVP(*vpp)) {
1123 		struct vnode *svp;
1124 
1125 		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1126 		VN_RELE(*vpp);
1127 		if (svp == NULL)
1128 			error = ENOSYS;
1129 		*vpp = svp;
1130 	}
1131 	return (error);
1132 }
1133 
1134 
1135 /*
1136  * Lookup an entry in a directory, or an extended attribute directory.
1137  * If it exists, return a held vnode reference for it.
1138  *
1139  *	IN:	dvp	- vnode of directory to search.
1140  *		nm	- name of entry to lookup.
1141  *		pnp	- full pathname to lookup [UNUSED].
1142  *		flags	- LOOKUP_XATTR set if looking for an attribute.
1143  *		rdir	- root directory vnode [UNUSED].
1144  *		cr	- credentials of caller.
1145  *		ct	- caller context
1146  *		direntflags - directory lookup flags
1147  *		realpnp - returned pathname.
1148  *
1149  *	OUT:	vpp	- vnode of located entry, NULL if not found.
1150  *
1151  *	RETURN:	0 if success
1152  *		error code if failure
1153  *
1154  * Timestamps:
1155  *	NA
1156  */
1157 /* ARGSUSED */
1158 static int
1159 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1160     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1161     int *direntflags, pathname_t *realpnp)
1162 {
1163 	znode_t *zdp = VTOZ(dvp);
1164 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1165 	int	error = 0;
1166 
1167 	/* fast path */
1168 	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1169 
1170 		if (dvp->v_type != VDIR) {
1171 			return (ENOTDIR);
1172 		} else if (zdp->z_sa_hdl == NULL) {
1173 			return (EIO);
1174 		}
1175 
1176 		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1177 			error = zfs_fastaccesschk_execute(zdp, cr);
1178 			if (!error) {
1179 				*vpp = dvp;
1180 				VN_HOLD(*vpp);
1181 				return (0);
1182 			}
1183 			return (error);
1184 		} else {
1185 			vnode_t *tvp = dnlc_lookup(dvp, nm);
1186 
1187 			if (tvp) {
1188 				error = zfs_fastaccesschk_execute(zdp, cr);
1189 				if (error) {
1190 					VN_RELE(tvp);
1191 					return (error);
1192 				}
1193 				if (tvp == DNLC_NO_VNODE) {
1194 					VN_RELE(tvp);
1195 					return (ENOENT);
1196 				} else {
1197 					*vpp = tvp;
1198 					return (specvp_check(vpp, cr));
1199 				}
1200 			}
1201 		}
1202 	}
1203 
1204 	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1205 
1206 	ZFS_ENTER(zfsvfs);
1207 	ZFS_VERIFY_ZP(zdp);
1208 
1209 	*vpp = NULL;
1210 
1211 	if (flags & LOOKUP_XATTR) {
1212 		/*
1213 		 * If the xattr property is off, refuse the lookup request.
1214 		 */
1215 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1216 			ZFS_EXIT(zfsvfs);
1217 			return (EINVAL);
1218 		}
1219 
1220 		/*
1221 		 * We don't allow recursive attributes..
1222 		 * Maybe someday we will.
1223 		 */
1224 		if (zdp->z_pflags & ZFS_XATTR) {
1225 			ZFS_EXIT(zfsvfs);
1226 			return (EINVAL);
1227 		}
1228 
1229 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1230 			ZFS_EXIT(zfsvfs);
1231 			return (error);
1232 		}
1233 
1234 		/*
1235 		 * Do we have permission to get into attribute directory?
1236 		 */
1237 
1238 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1239 		    B_FALSE, cr)) {
1240 			VN_RELE(*vpp);
1241 			*vpp = NULL;
1242 		}
1243 
1244 		ZFS_EXIT(zfsvfs);
1245 		return (error);
1246 	}
1247 
1248 	if (dvp->v_type != VDIR) {
1249 		ZFS_EXIT(zfsvfs);
1250 		return (ENOTDIR);
1251 	}
1252 
1253 	/*
1254 	 * Check accessibility of directory.
1255 	 */
1256 
1257 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1258 		ZFS_EXIT(zfsvfs);
1259 		return (error);
1260 	}
1261 
1262 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1263 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1264 		ZFS_EXIT(zfsvfs);
1265 		return (EILSEQ);
1266 	}
1267 
1268 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1269 	if (error == 0)
1270 		error = specvp_check(vpp, cr);
1271 
1272 	ZFS_EXIT(zfsvfs);
1273 	return (error);
1274 }
1275 
1276 /*
1277  * Attempt to create a new entry in a directory.  If the entry
1278  * already exists, truncate the file if permissible, else return
1279  * an error.  Return the vp of the created or trunc'd file.
1280  *
1281  *	IN:	dvp	- vnode of directory to put new file entry in.
1282  *		name	- name of new file entry.
1283  *		vap	- attributes of new file.
1284  *		excl	- flag indicating exclusive or non-exclusive mode.
1285  *		mode	- mode to open file with.
1286  *		cr	- credentials of caller.
1287  *		flag	- large file flag [UNUSED].
1288  *		ct	- caller context
1289  *		vsecp 	- ACL to be set
1290  *
1291  *	OUT:	vpp	- vnode of created or trunc'd entry.
1292  *
1293  *	RETURN:	0 if success
1294  *		error code if failure
1295  *
1296  * Timestamps:
1297  *	dvp - ctime|mtime updated if new entry created
1298  *	 vp - ctime|mtime always, atime if new
1299  */
1300 
1301 /* ARGSUSED */
1302 static int
1303 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1304     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1305     vsecattr_t *vsecp)
1306 {
1307 	znode_t		*zp, *dzp = VTOZ(dvp);
1308 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1309 	zilog_t		*zilog;
1310 	objset_t	*os;
1311 	zfs_dirlock_t	*dl;
1312 	dmu_tx_t	*tx;
1313 	int		error;
1314 	ksid_t		*ksid;
1315 	uid_t		uid;
1316 	gid_t		gid = crgetgid(cr);
1317 	zfs_acl_ids_t   acl_ids;
1318 	boolean_t	fuid_dirtied;
1319 	boolean_t	have_acl = B_FALSE;
1320 
1321 	/*
1322 	 * If we have an ephemeral id, ACL, or XVATTR then
1323 	 * make sure file system is at proper version
1324 	 */
1325 
1326 	ksid = crgetsid(cr, KSID_OWNER);
1327 	if (ksid)
1328 		uid = ksid_getid(ksid);
1329 	else
1330 		uid = crgetuid(cr);
1331 
1332 	if (zfsvfs->z_use_fuids == B_FALSE &&
1333 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1334 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1335 		return (EINVAL);
1336 
1337 	ZFS_ENTER(zfsvfs);
1338 	ZFS_VERIFY_ZP(dzp);
1339 	os = zfsvfs->z_os;
1340 	zilog = zfsvfs->z_log;
1341 
1342 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1343 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1344 		ZFS_EXIT(zfsvfs);
1345 		return (EILSEQ);
1346 	}
1347 
1348 	if (vap->va_mask & AT_XVATTR) {
1349 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1350 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1351 			ZFS_EXIT(zfsvfs);
1352 			return (error);
1353 		}
1354 	}
1355 top:
1356 	*vpp = NULL;
1357 
1358 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1359 		vap->va_mode &= ~VSVTX;
1360 
1361 	if (*name == '\0') {
1362 		/*
1363 		 * Null component name refers to the directory itself.
1364 		 */
1365 		VN_HOLD(dvp);
1366 		zp = dzp;
1367 		dl = NULL;
1368 		error = 0;
1369 	} else {
1370 		/* possible VN_HOLD(zp) */
1371 		int zflg = 0;
1372 
1373 		if (flag & FIGNORECASE)
1374 			zflg |= ZCILOOK;
1375 
1376 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1377 		    NULL, NULL);
1378 		if (error) {
1379 			if (have_acl)
1380 				zfs_acl_ids_free(&acl_ids);
1381 			if (strcmp(name, "..") == 0)
1382 				error = EISDIR;
1383 			ZFS_EXIT(zfsvfs);
1384 			return (error);
1385 		}
1386 	}
1387 
1388 	if (zp == NULL) {
1389 		uint64_t txtype;
1390 
1391 		/*
1392 		 * Create a new file object and update the directory
1393 		 * to reference it.
1394 		 */
1395 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1396 			if (have_acl)
1397 				zfs_acl_ids_free(&acl_ids);
1398 			goto out;
1399 		}
1400 
1401 		/*
1402 		 * We only support the creation of regular files in
1403 		 * extended attribute directories.
1404 		 */
1405 
1406 		if ((dzp->z_pflags & ZFS_XATTR) &&
1407 		    (vap->va_type != VREG)) {
1408 			if (have_acl)
1409 				zfs_acl_ids_free(&acl_ids);
1410 			error = EINVAL;
1411 			goto out;
1412 		}
1413 
1414 		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1415 		    cr, vsecp, &acl_ids)) != 0)
1416 			goto out;
1417 		have_acl = B_TRUE;
1418 
1419 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1420 			zfs_acl_ids_free(&acl_ids);
1421 			error = EDQUOT;
1422 			goto out;
1423 		}
1424 
1425 		tx = dmu_tx_create(os);
1426 
1427 		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1428 		    ZFS_SA_BASE_ATTR_SIZE);
1429 
1430 		fuid_dirtied = zfsvfs->z_fuid_dirty;
1431 		if (fuid_dirtied)
1432 			zfs_fuid_txhold(zfsvfs, tx);
1433 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1434 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1435 		if (!zfsvfs->z_use_sa &&
1436 		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1437 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1438 			    0, acl_ids.z_aclp->z_acl_bytes);
1439 		}
1440 		error = dmu_tx_assign(tx, TXG_NOWAIT);
1441 		if (error) {
1442 			zfs_dirent_unlock(dl);
1443 			if (error == ERESTART) {
1444 				dmu_tx_wait(tx);
1445 				dmu_tx_abort(tx);
1446 				goto top;
1447 			}
1448 			zfs_acl_ids_free(&acl_ids);
1449 			dmu_tx_abort(tx);
1450 			ZFS_EXIT(zfsvfs);
1451 			return (error);
1452 		}
1453 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1454 
1455 		if (fuid_dirtied)
1456 			zfs_fuid_sync(zfsvfs, tx);
1457 
1458 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1459 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1460 		if (flag & FIGNORECASE)
1461 			txtype |= TX_CI;
1462 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1463 		    vsecp, acl_ids.z_fuidp, vap);
1464 		zfs_acl_ids_free(&acl_ids);
1465 		dmu_tx_commit(tx);
1466 	} else {
1467 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1468 
1469 		if (have_acl)
1470 			zfs_acl_ids_free(&acl_ids);
1471 		have_acl = B_FALSE;
1472 
1473 		/*
1474 		 * A directory entry already exists for this name.
1475 		 */
1476 		/*
1477 		 * Can't truncate an existing file if in exclusive mode.
1478 		 */
1479 		if (excl == EXCL) {
1480 			error = EEXIST;
1481 			goto out;
1482 		}
1483 		/*
1484 		 * Can't open a directory for writing.
1485 		 */
1486 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1487 			error = EISDIR;
1488 			goto out;
1489 		}
1490 		/*
1491 		 * Verify requested access to file.
1492 		 */
1493 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1494 			goto out;
1495 		}
1496 
1497 		mutex_enter(&dzp->z_lock);
1498 		dzp->z_seq++;
1499 		mutex_exit(&dzp->z_lock);
1500 
1501 		/*
1502 		 * Truncate regular files if requested.
1503 		 */
1504 		if ((ZTOV(zp)->v_type == VREG) &&
1505 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1506 			/* we can't hold any locks when calling zfs_freesp() */
1507 			zfs_dirent_unlock(dl);
1508 			dl = NULL;
1509 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1510 			if (error == 0) {
1511 				vnevent_create(ZTOV(zp), ct);
1512 			}
1513 		}
1514 	}
1515 out:
1516 
1517 	if (dl)
1518 		zfs_dirent_unlock(dl);
1519 
1520 	if (error) {
1521 		if (zp)
1522 			VN_RELE(ZTOV(zp));
1523 	} else {
1524 		*vpp = ZTOV(zp);
1525 		error = specvp_check(vpp, cr);
1526 	}
1527 
1528 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1529 		zil_commit(zilog, 0);
1530 
1531 	ZFS_EXIT(zfsvfs);
1532 	return (error);
1533 }
1534 
1535 /*
1536  * Remove an entry from a directory.
1537  *
1538  *	IN:	dvp	- vnode of directory to remove entry from.
1539  *		name	- name of entry to remove.
1540  *		cr	- credentials of caller.
1541  *		ct	- caller context
1542  *		flags	- case flags
1543  *
1544  *	RETURN:	0 if success
1545  *		error code if failure
1546  *
1547  * Timestamps:
1548  *	dvp - ctime|mtime
1549  *	 vp - ctime (if nlink > 0)
1550  */
1551 
1552 uint64_t null_xattr = 0;
1553 
1554 /*ARGSUSED*/
1555 static int
1556 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1557     int flags)
1558 {
1559 	znode_t		*zp, *dzp = VTOZ(dvp);
1560 	znode_t		*xzp;
1561 	vnode_t		*vp;
1562 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1563 	zilog_t		*zilog;
1564 	uint64_t	acl_obj, xattr_obj;
1565 	uint64_t 	xattr_obj_unlinked = 0;
1566 	uint64_t	obj = 0;
1567 	zfs_dirlock_t	*dl;
1568 	dmu_tx_t	*tx;
1569 	boolean_t	may_delete_now, delete_now = FALSE;
1570 	boolean_t	unlinked, toobig = FALSE;
1571 	uint64_t	txtype;
1572 	pathname_t	*realnmp = NULL;
1573 	pathname_t	realnm;
1574 	int		error;
1575 	int		zflg = ZEXISTS;
1576 
1577 	ZFS_ENTER(zfsvfs);
1578 	ZFS_VERIFY_ZP(dzp);
1579 	zilog = zfsvfs->z_log;
1580 
1581 	if (flags & FIGNORECASE) {
1582 		zflg |= ZCILOOK;
1583 		pn_alloc(&realnm);
1584 		realnmp = &realnm;
1585 	}
1586 
1587 top:
1588 	xattr_obj = 0;
1589 	xzp = NULL;
1590 	/*
1591 	 * Attempt to lock directory; fail if entry doesn't exist.
1592 	 */
1593 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1594 	    NULL, realnmp)) {
1595 		if (realnmp)
1596 			pn_free(realnmp);
1597 		ZFS_EXIT(zfsvfs);
1598 		return (error);
1599 	}
1600 
1601 	vp = ZTOV(zp);
1602 
1603 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1604 		goto out;
1605 	}
1606 
1607 	/*
1608 	 * Need to use rmdir for removing directories.
1609 	 */
1610 	if (vp->v_type == VDIR) {
1611 		error = EPERM;
1612 		goto out;
1613 	}
1614 
1615 	vnevent_remove(vp, dvp, name, ct);
1616 
1617 	if (realnmp)
1618 		dnlc_remove(dvp, realnmp->pn_buf);
1619 	else
1620 		dnlc_remove(dvp, name);
1621 
1622 	mutex_enter(&vp->v_lock);
1623 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1624 	mutex_exit(&vp->v_lock);
1625 
1626 	/*
1627 	 * We may delete the znode now, or we may put it in the unlinked set;
1628 	 * it depends on whether we're the last link, and on whether there are
1629 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1630 	 * allow for either case.
1631 	 */
1632 	obj = zp->z_id;
1633 	tx = dmu_tx_create(zfsvfs->z_os);
1634 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1635 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1636 	zfs_sa_upgrade_txholds(tx, zp);
1637 	zfs_sa_upgrade_txholds(tx, dzp);
1638 	if (may_delete_now) {
1639 		toobig =
1640 		    zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1641 		/* if the file is too big, only hold_free a token amount */
1642 		dmu_tx_hold_free(tx, zp->z_id, 0,
1643 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1644 	}
1645 
1646 	/* are there any extended attributes? */
1647 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1648 	    &xattr_obj, sizeof (xattr_obj));
1649 	if (error == 0 && xattr_obj) {
1650 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1651 		ASSERT0(error);
1652 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1653 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1654 	}
1655 
1656 	mutex_enter(&zp->z_lock);
1657 	if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1658 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1659 	mutex_exit(&zp->z_lock);
1660 
1661 	/* charge as an update -- would be nice not to charge at all */
1662 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1663 
1664 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1665 	if (error) {
1666 		zfs_dirent_unlock(dl);
1667 		VN_RELE(vp);
1668 		if (xzp)
1669 			VN_RELE(ZTOV(xzp));
1670 		if (error == ERESTART) {
1671 			dmu_tx_wait(tx);
1672 			dmu_tx_abort(tx);
1673 			goto top;
1674 		}
1675 		if (realnmp)
1676 			pn_free(realnmp);
1677 		dmu_tx_abort(tx);
1678 		ZFS_EXIT(zfsvfs);
1679 		return (error);
1680 	}
1681 
1682 	/*
1683 	 * Remove the directory entry.
1684 	 */
1685 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1686 
1687 	if (error) {
1688 		dmu_tx_commit(tx);
1689 		goto out;
1690 	}
1691 
1692 	if (unlinked) {
1693 
1694 		/*
1695 		 * Hold z_lock so that we can make sure that the ACL obj
1696 		 * hasn't changed.  Could have been deleted due to
1697 		 * zfs_sa_upgrade().
1698 		 */
1699 		mutex_enter(&zp->z_lock);
1700 		mutex_enter(&vp->v_lock);
1701 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1702 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1703 		delete_now = may_delete_now && !toobig &&
1704 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1705 		    xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1706 		    acl_obj;
1707 		mutex_exit(&vp->v_lock);
1708 	}
1709 
1710 	if (delete_now) {
1711 		if (xattr_obj_unlinked) {
1712 			ASSERT3U(xzp->z_links, ==, 2);
1713 			mutex_enter(&xzp->z_lock);
1714 			xzp->z_unlinked = 1;
1715 			xzp->z_links = 0;
1716 			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1717 			    &xzp->z_links, sizeof (xzp->z_links), tx);
1718 			ASSERT3U(error,  ==,  0);
1719 			mutex_exit(&xzp->z_lock);
1720 			zfs_unlinked_add(xzp, tx);
1721 
1722 			if (zp->z_is_sa)
1723 				error = sa_remove(zp->z_sa_hdl,
1724 				    SA_ZPL_XATTR(zfsvfs), tx);
1725 			else
1726 				error = sa_update(zp->z_sa_hdl,
1727 				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
1728 				    sizeof (uint64_t), tx);
1729 			ASSERT0(error);
1730 		}
1731 		mutex_enter(&vp->v_lock);
1732 		vp->v_count--;
1733 		ASSERT0(vp->v_count);
1734 		mutex_exit(&vp->v_lock);
1735 		mutex_exit(&zp->z_lock);
1736 		zfs_znode_delete(zp, tx);
1737 	} else if (unlinked) {
1738 		mutex_exit(&zp->z_lock);
1739 		zfs_unlinked_add(zp, tx);
1740 	}
1741 
1742 	txtype = TX_REMOVE;
1743 	if (flags & FIGNORECASE)
1744 		txtype |= TX_CI;
1745 	zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1746 
1747 	dmu_tx_commit(tx);
1748 out:
1749 	if (realnmp)
1750 		pn_free(realnmp);
1751 
1752 	zfs_dirent_unlock(dl);
1753 
1754 	if (!delete_now)
1755 		VN_RELE(vp);
1756 	if (xzp)
1757 		VN_RELE(ZTOV(xzp));
1758 
1759 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1760 		zil_commit(zilog, 0);
1761 
1762 	ZFS_EXIT(zfsvfs);
1763 	return (error);
1764 }
1765 
1766 /*
1767  * Create a new directory and insert it into dvp using the name
1768  * provided.  Return a pointer to the inserted directory.
1769  *
1770  *	IN:	dvp	- vnode of directory to add subdir to.
1771  *		dirname	- name of new directory.
1772  *		vap	- attributes of new directory.
1773  *		cr	- credentials of caller.
1774  *		ct	- caller context
1775  *		vsecp	- ACL to be set
1776  *
1777  *	OUT:	vpp	- vnode of created directory.
1778  *
1779  *	RETURN:	0 if success
1780  *		error code if failure
1781  *
1782  * Timestamps:
1783  *	dvp - ctime|mtime updated
1784  *	 vp - ctime|mtime|atime updated
1785  */
1786 /*ARGSUSED*/
1787 static int
1788 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1789     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1790 {
1791 	znode_t		*zp, *dzp = VTOZ(dvp);
1792 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1793 	zilog_t		*zilog;
1794 	zfs_dirlock_t	*dl;
1795 	uint64_t	txtype;
1796 	dmu_tx_t	*tx;
1797 	int		error;
1798 	int		zf = ZNEW;
1799 	ksid_t		*ksid;
1800 	uid_t		uid;
1801 	gid_t		gid = crgetgid(cr);
1802 	zfs_acl_ids_t   acl_ids;
1803 	boolean_t	fuid_dirtied;
1804 
1805 	ASSERT(vap->va_type == VDIR);
1806 
1807 	/*
1808 	 * If we have an ephemeral id, ACL, or XVATTR then
1809 	 * make sure file system is at proper version
1810 	 */
1811 
1812 	ksid = crgetsid(cr, KSID_OWNER);
1813 	if (ksid)
1814 		uid = ksid_getid(ksid);
1815 	else
1816 		uid = crgetuid(cr);
1817 	if (zfsvfs->z_use_fuids == B_FALSE &&
1818 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1819 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1820 		return (EINVAL);
1821 
1822 	ZFS_ENTER(zfsvfs);
1823 	ZFS_VERIFY_ZP(dzp);
1824 	zilog = zfsvfs->z_log;
1825 
1826 	if (dzp->z_pflags & ZFS_XATTR) {
1827 		ZFS_EXIT(zfsvfs);
1828 		return (EINVAL);
1829 	}
1830 
1831 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1832 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1833 		ZFS_EXIT(zfsvfs);
1834 		return (EILSEQ);
1835 	}
1836 	if (flags & FIGNORECASE)
1837 		zf |= ZCILOOK;
1838 
1839 	if (vap->va_mask & AT_XVATTR) {
1840 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1841 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1842 			ZFS_EXIT(zfsvfs);
1843 			return (error);
1844 		}
1845 	}
1846 
1847 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1848 	    vsecp, &acl_ids)) != 0) {
1849 		ZFS_EXIT(zfsvfs);
1850 		return (error);
1851 	}
1852 	/*
1853 	 * First make sure the new directory doesn't exist.
1854 	 *
1855 	 * Existence is checked first to make sure we don't return
1856 	 * EACCES instead of EEXIST which can cause some applications
1857 	 * to fail.
1858 	 */
1859 top:
1860 	*vpp = NULL;
1861 
1862 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1863 	    NULL, NULL)) {
1864 		zfs_acl_ids_free(&acl_ids);
1865 		ZFS_EXIT(zfsvfs);
1866 		return (error);
1867 	}
1868 
1869 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1870 		zfs_acl_ids_free(&acl_ids);
1871 		zfs_dirent_unlock(dl);
1872 		ZFS_EXIT(zfsvfs);
1873 		return (error);
1874 	}
1875 
1876 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1877 		zfs_acl_ids_free(&acl_ids);
1878 		zfs_dirent_unlock(dl);
1879 		ZFS_EXIT(zfsvfs);
1880 		return (EDQUOT);
1881 	}
1882 
1883 	/*
1884 	 * Add a new entry to the directory.
1885 	 */
1886 	tx = dmu_tx_create(zfsvfs->z_os);
1887 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1888 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1889 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1890 	if (fuid_dirtied)
1891 		zfs_fuid_txhold(zfsvfs, tx);
1892 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1893 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1894 		    acl_ids.z_aclp->z_acl_bytes);
1895 	}
1896 
1897 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1898 	    ZFS_SA_BASE_ATTR_SIZE);
1899 
1900 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1901 	if (error) {
1902 		zfs_dirent_unlock(dl);
1903 		if (error == ERESTART) {
1904 			dmu_tx_wait(tx);
1905 			dmu_tx_abort(tx);
1906 			goto top;
1907 		}
1908 		zfs_acl_ids_free(&acl_ids);
1909 		dmu_tx_abort(tx);
1910 		ZFS_EXIT(zfsvfs);
1911 		return (error);
1912 	}
1913 
1914 	/*
1915 	 * Create new node.
1916 	 */
1917 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1918 
1919 	if (fuid_dirtied)
1920 		zfs_fuid_sync(zfsvfs, tx);
1921 
1922 	/*
1923 	 * Now put new name in parent dir.
1924 	 */
1925 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1926 
1927 	*vpp = ZTOV(zp);
1928 
1929 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1930 	if (flags & FIGNORECASE)
1931 		txtype |= TX_CI;
1932 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1933 	    acl_ids.z_fuidp, vap);
1934 
1935 	zfs_acl_ids_free(&acl_ids);
1936 
1937 	dmu_tx_commit(tx);
1938 
1939 	zfs_dirent_unlock(dl);
1940 
1941 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1942 		zil_commit(zilog, 0);
1943 
1944 	ZFS_EXIT(zfsvfs);
1945 	return (0);
1946 }
1947 
1948 /*
1949  * Remove a directory subdir entry.  If the current working
1950  * directory is the same as the subdir to be removed, the
1951  * remove will fail.
1952  *
1953  *	IN:	dvp	- vnode of directory to remove from.
1954  *		name	- name of directory to be removed.
1955  *		cwd	- vnode of current working directory.
1956  *		cr	- credentials of caller.
1957  *		ct	- caller context
1958  *		flags	- case flags
1959  *
1960  *	RETURN:	0 if success
1961  *		error code if failure
1962  *
1963  * Timestamps:
1964  *	dvp - ctime|mtime updated
1965  */
1966 /*ARGSUSED*/
1967 static int
1968 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1969     caller_context_t *ct, int flags)
1970 {
1971 	znode_t		*dzp = VTOZ(dvp);
1972 	znode_t		*zp;
1973 	vnode_t		*vp;
1974 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1975 	zilog_t		*zilog;
1976 	zfs_dirlock_t	*dl;
1977 	dmu_tx_t	*tx;
1978 	int		error;
1979 	int		zflg = ZEXISTS;
1980 
1981 	ZFS_ENTER(zfsvfs);
1982 	ZFS_VERIFY_ZP(dzp);
1983 	zilog = zfsvfs->z_log;
1984 
1985 	if (flags & FIGNORECASE)
1986 		zflg |= ZCILOOK;
1987 top:
1988 	zp = NULL;
1989 
1990 	/*
1991 	 * Attempt to lock directory; fail if entry doesn't exist.
1992 	 */
1993 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1994 	    NULL, NULL)) {
1995 		ZFS_EXIT(zfsvfs);
1996 		return (error);
1997 	}
1998 
1999 	vp = ZTOV(zp);
2000 
2001 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2002 		goto out;
2003 	}
2004 
2005 	if (vp->v_type != VDIR) {
2006 		error = ENOTDIR;
2007 		goto out;
2008 	}
2009 
2010 	if (vp == cwd) {
2011 		error = EINVAL;
2012 		goto out;
2013 	}
2014 
2015 	vnevent_rmdir(vp, dvp, name, ct);
2016 
2017 	/*
2018 	 * Grab a lock on the directory to make sure that noone is
2019 	 * trying to add (or lookup) entries while we are removing it.
2020 	 */
2021 	rw_enter(&zp->z_name_lock, RW_WRITER);
2022 
2023 	/*
2024 	 * Grab a lock on the parent pointer to make sure we play well
2025 	 * with the treewalk and directory rename code.
2026 	 */
2027 	rw_enter(&zp->z_parent_lock, RW_WRITER);
2028 
2029 	tx = dmu_tx_create(zfsvfs->z_os);
2030 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2031 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2032 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2033 	zfs_sa_upgrade_txholds(tx, zp);
2034 	zfs_sa_upgrade_txholds(tx, dzp);
2035 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2036 	if (error) {
2037 		rw_exit(&zp->z_parent_lock);
2038 		rw_exit(&zp->z_name_lock);
2039 		zfs_dirent_unlock(dl);
2040 		VN_RELE(vp);
2041 		if (error == ERESTART) {
2042 			dmu_tx_wait(tx);
2043 			dmu_tx_abort(tx);
2044 			goto top;
2045 		}
2046 		dmu_tx_abort(tx);
2047 		ZFS_EXIT(zfsvfs);
2048 		return (error);
2049 	}
2050 
2051 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2052 
2053 	if (error == 0) {
2054 		uint64_t txtype = TX_RMDIR;
2055 		if (flags & FIGNORECASE)
2056 			txtype |= TX_CI;
2057 		zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2058 	}
2059 
2060 	dmu_tx_commit(tx);
2061 
2062 	rw_exit(&zp->z_parent_lock);
2063 	rw_exit(&zp->z_name_lock);
2064 out:
2065 	zfs_dirent_unlock(dl);
2066 
2067 	VN_RELE(vp);
2068 
2069 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2070 		zil_commit(zilog, 0);
2071 
2072 	ZFS_EXIT(zfsvfs);
2073 	return (error);
2074 }
2075 
2076 /*
2077  * Read as many directory entries as will fit into the provided
2078  * buffer from the given directory cursor position (specified in
2079  * the uio structure.
2080  *
2081  *	IN:	vp	- vnode of directory to read.
2082  *		uio	- structure supplying read location, range info,
2083  *			  and return buffer.
2084  *		cr	- credentials of caller.
2085  *		ct	- caller context
2086  *		flags	- case flags
2087  *
2088  *	OUT:	uio	- updated offset and range, buffer filled.
2089  *		eofp	- set to true if end-of-file detected.
2090  *
2091  *	RETURN:	0 if success
2092  *		error code if failure
2093  *
2094  * Timestamps:
2095  *	vp - atime updated
2096  *
2097  * Note that the low 4 bits of the cookie returned by zap is always zero.
2098  * This allows us to use the low range for "special" directory entries:
2099  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2100  * we use the offset 2 for the '.zfs' directory.
2101  */
2102 /* ARGSUSED */
2103 static int
2104 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2105     caller_context_t *ct, int flags)
2106 {
2107 	znode_t		*zp = VTOZ(vp);
2108 	iovec_t		*iovp;
2109 	edirent_t	*eodp;
2110 	dirent64_t	*odp;
2111 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2112 	objset_t	*os;
2113 	caddr_t		outbuf;
2114 	size_t		bufsize;
2115 	zap_cursor_t	zc;
2116 	zap_attribute_t	zap;
2117 	uint_t		bytes_wanted;
2118 	uint64_t	offset; /* must be unsigned; checks for < 1 */
2119 	uint64_t	parent;
2120 	int		local_eof;
2121 	int		outcount;
2122 	int		error;
2123 	uint8_t		prefetch;
2124 	boolean_t	check_sysattrs;
2125 
2126 	ZFS_ENTER(zfsvfs);
2127 	ZFS_VERIFY_ZP(zp);
2128 
2129 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2130 	    &parent, sizeof (parent))) != 0) {
2131 		ZFS_EXIT(zfsvfs);
2132 		return (error);
2133 	}
2134 
2135 	/*
2136 	 * If we are not given an eof variable,
2137 	 * use a local one.
2138 	 */
2139 	if (eofp == NULL)
2140 		eofp = &local_eof;
2141 
2142 	/*
2143 	 * Check for valid iov_len.
2144 	 */
2145 	if (uio->uio_iov->iov_len <= 0) {
2146 		ZFS_EXIT(zfsvfs);
2147 		return (EINVAL);
2148 	}
2149 
2150 	/*
2151 	 * Quit if directory has been removed (posix)
2152 	 */
2153 	if ((*eofp = zp->z_unlinked) != 0) {
2154 		ZFS_EXIT(zfsvfs);
2155 		return (0);
2156 	}
2157 
2158 	error = 0;
2159 	os = zfsvfs->z_os;
2160 	offset = uio->uio_loffset;
2161 	prefetch = zp->z_zn_prefetch;
2162 
2163 	/*
2164 	 * Initialize the iterator cursor.
2165 	 */
2166 	if (offset <= 3) {
2167 		/*
2168 		 * Start iteration from the beginning of the directory.
2169 		 */
2170 		zap_cursor_init(&zc, os, zp->z_id);
2171 	} else {
2172 		/*
2173 		 * The offset is a serialized cursor.
2174 		 */
2175 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2176 	}
2177 
2178 	/*
2179 	 * Get space to change directory entries into fs independent format.
2180 	 */
2181 	iovp = uio->uio_iov;
2182 	bytes_wanted = iovp->iov_len;
2183 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2184 		bufsize = bytes_wanted;
2185 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
2186 		odp = (struct dirent64 *)outbuf;
2187 	} else {
2188 		bufsize = bytes_wanted;
2189 		outbuf = NULL;
2190 		odp = (struct dirent64 *)iovp->iov_base;
2191 	}
2192 	eodp = (struct edirent *)odp;
2193 
2194 	/*
2195 	 * If this VFS supports the system attribute view interface; and
2196 	 * we're looking at an extended attribute directory; and we care
2197 	 * about normalization conflicts on this vfs; then we must check
2198 	 * for normalization conflicts with the sysattr name space.
2199 	 */
2200 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2201 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2202 	    (flags & V_RDDIR_ENTFLAGS);
2203 
2204 	/*
2205 	 * Transform to file-system independent format
2206 	 */
2207 	outcount = 0;
2208 	while (outcount < bytes_wanted) {
2209 		ino64_t objnum;
2210 		ushort_t reclen;
2211 		off64_t *next = NULL;
2212 
2213 		/*
2214 		 * Special case `.', `..', and `.zfs'.
2215 		 */
2216 		if (offset == 0) {
2217 			(void) strcpy(zap.za_name, ".");
2218 			zap.za_normalization_conflict = 0;
2219 			objnum = zp->z_id;
2220 		} else if (offset == 1) {
2221 			(void) strcpy(zap.za_name, "..");
2222 			zap.za_normalization_conflict = 0;
2223 			objnum = parent;
2224 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2225 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2226 			zap.za_normalization_conflict = 0;
2227 			objnum = ZFSCTL_INO_ROOT;
2228 		} else {
2229 			/*
2230 			 * Grab next entry.
2231 			 */
2232 			if (error = zap_cursor_retrieve(&zc, &zap)) {
2233 				if ((*eofp = (error == ENOENT)) != 0)
2234 					break;
2235 				else
2236 					goto update;
2237 			}
2238 
2239 			if (zap.za_integer_length != 8 ||
2240 			    zap.za_num_integers != 1) {
2241 				cmn_err(CE_WARN, "zap_readdir: bad directory "
2242 				    "entry, obj = %lld, offset = %lld\n",
2243 				    (u_longlong_t)zp->z_id,
2244 				    (u_longlong_t)offset);
2245 				error = ENXIO;
2246 				goto update;
2247 			}
2248 
2249 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2250 			/*
2251 			 * MacOS X can extract the object type here such as:
2252 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2253 			 */
2254 
2255 			if (check_sysattrs && !zap.za_normalization_conflict) {
2256 				zap.za_normalization_conflict =
2257 				    xattr_sysattr_casechk(zap.za_name);
2258 			}
2259 		}
2260 
2261 		if (flags & V_RDDIR_ACCFILTER) {
2262 			/*
2263 			 * If we have no access at all, don't include
2264 			 * this entry in the returned information
2265 			 */
2266 			znode_t	*ezp;
2267 			if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2268 				goto skip_entry;
2269 			if (!zfs_has_access(ezp, cr)) {
2270 				VN_RELE(ZTOV(ezp));
2271 				goto skip_entry;
2272 			}
2273 			VN_RELE(ZTOV(ezp));
2274 		}
2275 
2276 		if (flags & V_RDDIR_ENTFLAGS)
2277 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2278 		else
2279 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2280 
2281 		/*
2282 		 * Will this entry fit in the buffer?
2283 		 */
2284 		if (outcount + reclen > bufsize) {
2285 			/*
2286 			 * Did we manage to fit anything in the buffer?
2287 			 */
2288 			if (!outcount) {
2289 				error = EINVAL;
2290 				goto update;
2291 			}
2292 			break;
2293 		}
2294 		if (flags & V_RDDIR_ENTFLAGS) {
2295 			/*
2296 			 * Add extended flag entry:
2297 			 */
2298 			eodp->ed_ino = objnum;
2299 			eodp->ed_reclen = reclen;
2300 			/* NOTE: ed_off is the offset for the *next* entry */
2301 			next = &(eodp->ed_off);
2302 			eodp->ed_eflags = zap.za_normalization_conflict ?
2303 			    ED_CASE_CONFLICT : 0;
2304 			(void) strncpy(eodp->ed_name, zap.za_name,
2305 			    EDIRENT_NAMELEN(reclen));
2306 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2307 		} else {
2308 			/*
2309 			 * Add normal entry:
2310 			 */
2311 			odp->d_ino = objnum;
2312 			odp->d_reclen = reclen;
2313 			/* NOTE: d_off is the offset for the *next* entry */
2314 			next = &(odp->d_off);
2315 			(void) strncpy(odp->d_name, zap.za_name,
2316 			    DIRENT64_NAMELEN(reclen));
2317 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2318 		}
2319 		outcount += reclen;
2320 
2321 		ASSERT(outcount <= bufsize);
2322 
2323 		/* Prefetch znode */
2324 		if (prefetch)
2325 			dmu_prefetch(os, objnum, 0, 0);
2326 
2327 	skip_entry:
2328 		/*
2329 		 * Move to the next entry, fill in the previous offset.
2330 		 */
2331 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2332 			zap_cursor_advance(&zc);
2333 			offset = zap_cursor_serialize(&zc);
2334 		} else {
2335 			offset += 1;
2336 		}
2337 		if (next)
2338 			*next = offset;
2339 	}
2340 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2341 
2342 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2343 		iovp->iov_base += outcount;
2344 		iovp->iov_len -= outcount;
2345 		uio->uio_resid -= outcount;
2346 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2347 		/*
2348 		 * Reset the pointer.
2349 		 */
2350 		offset = uio->uio_loffset;
2351 	}
2352 
2353 update:
2354 	zap_cursor_fini(&zc);
2355 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2356 		kmem_free(outbuf, bufsize);
2357 
2358 	if (error == ENOENT)
2359 		error = 0;
2360 
2361 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2362 
2363 	uio->uio_loffset = offset;
2364 	ZFS_EXIT(zfsvfs);
2365 	return (error);
2366 }
2367 
2368 ulong_t zfs_fsync_sync_cnt = 4;
2369 
2370 static int
2371 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2372 {
2373 	znode_t	*zp = VTOZ(vp);
2374 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2375 
2376 	/*
2377 	 * Regardless of whether this is required for standards conformance,
2378 	 * this is the logical behavior when fsync() is called on a file with
2379 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2380 	 * going to be pushed out as part of the zil_commit().
2381 	 */
2382 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2383 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2384 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2385 
2386 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2387 
2388 	if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2389 		ZFS_ENTER(zfsvfs);
2390 		ZFS_VERIFY_ZP(zp);
2391 		zil_commit(zfsvfs->z_log, zp->z_id);
2392 		ZFS_EXIT(zfsvfs);
2393 	}
2394 	return (0);
2395 }
2396 
2397 
2398 /*
2399  * Get the requested file attributes and place them in the provided
2400  * vattr structure.
2401  *
2402  *	IN:	vp	- vnode of file.
2403  *		vap	- va_mask identifies requested attributes.
2404  *			  If AT_XVATTR set, then optional attrs are requested
2405  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2406  *		cr	- credentials of caller.
2407  *		ct	- caller context
2408  *
2409  *	OUT:	vap	- attribute values.
2410  *
2411  *	RETURN:	0 (always succeeds)
2412  */
2413 /* ARGSUSED */
2414 static int
2415 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2416     caller_context_t *ct)
2417 {
2418 	znode_t *zp = VTOZ(vp);
2419 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2420 	int	error = 0;
2421 	uint64_t links;
2422 	uint64_t mtime[2], ctime[2];
2423 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2424 	xoptattr_t *xoap = NULL;
2425 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2426 	sa_bulk_attr_t bulk[2];
2427 	int count = 0;
2428 
2429 	ZFS_ENTER(zfsvfs);
2430 	ZFS_VERIFY_ZP(zp);
2431 
2432 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2433 
2434 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2435 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2436 
2437 	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2438 		ZFS_EXIT(zfsvfs);
2439 		return (error);
2440 	}
2441 
2442 	/*
2443 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2444 	 * Also, if we are the owner don't bother, since owner should
2445 	 * always be allowed to read basic attributes of file.
2446 	 */
2447 	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2448 	    (vap->va_uid != crgetuid(cr))) {
2449 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2450 		    skipaclchk, cr)) {
2451 			ZFS_EXIT(zfsvfs);
2452 			return (error);
2453 		}
2454 	}
2455 
2456 	/*
2457 	 * Return all attributes.  It's cheaper to provide the answer
2458 	 * than to determine whether we were asked the question.
2459 	 */
2460 
2461 	mutex_enter(&zp->z_lock);
2462 	vap->va_type = vp->v_type;
2463 	vap->va_mode = zp->z_mode & MODEMASK;
2464 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2465 	vap->va_nodeid = zp->z_id;
2466 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2467 		links = zp->z_links + 1;
2468 	else
2469 		links = zp->z_links;
2470 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2471 	vap->va_size = zp->z_size;
2472 	vap->va_rdev = vp->v_rdev;
2473 	vap->va_seq = zp->z_seq;
2474 
2475 	/*
2476 	 * Add in any requested optional attributes and the create time.
2477 	 * Also set the corresponding bits in the returned attribute bitmap.
2478 	 */
2479 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2480 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2481 			xoap->xoa_archive =
2482 			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2483 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2484 		}
2485 
2486 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2487 			xoap->xoa_readonly =
2488 			    ((zp->z_pflags & ZFS_READONLY) != 0);
2489 			XVA_SET_RTN(xvap, XAT_READONLY);
2490 		}
2491 
2492 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2493 			xoap->xoa_system =
2494 			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2495 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2496 		}
2497 
2498 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2499 			xoap->xoa_hidden =
2500 			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2501 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2502 		}
2503 
2504 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2505 			xoap->xoa_nounlink =
2506 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2507 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2508 		}
2509 
2510 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2511 			xoap->xoa_immutable =
2512 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2513 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2514 		}
2515 
2516 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2517 			xoap->xoa_appendonly =
2518 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2519 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2520 		}
2521 
2522 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2523 			xoap->xoa_nodump =
2524 			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2525 			XVA_SET_RTN(xvap, XAT_NODUMP);
2526 		}
2527 
2528 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2529 			xoap->xoa_opaque =
2530 			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2531 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2532 		}
2533 
2534 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2535 			xoap->xoa_av_quarantined =
2536 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2537 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2538 		}
2539 
2540 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2541 			xoap->xoa_av_modified =
2542 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2543 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2544 		}
2545 
2546 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2547 		    vp->v_type == VREG) {
2548 			zfs_sa_get_scanstamp(zp, xvap);
2549 		}
2550 
2551 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2552 			uint64_t times[2];
2553 
2554 			(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2555 			    times, sizeof (times));
2556 			ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2557 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2558 		}
2559 
2560 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2561 			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2562 			XVA_SET_RTN(xvap, XAT_REPARSE);
2563 		}
2564 		if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2565 			xoap->xoa_generation = zp->z_gen;
2566 			XVA_SET_RTN(xvap, XAT_GEN);
2567 		}
2568 
2569 		if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2570 			xoap->xoa_offline =
2571 			    ((zp->z_pflags & ZFS_OFFLINE) != 0);
2572 			XVA_SET_RTN(xvap, XAT_OFFLINE);
2573 		}
2574 
2575 		if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2576 			xoap->xoa_sparse =
2577 			    ((zp->z_pflags & ZFS_SPARSE) != 0);
2578 			XVA_SET_RTN(xvap, XAT_SPARSE);
2579 		}
2580 	}
2581 
2582 	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2583 	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2584 	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2585 
2586 	mutex_exit(&zp->z_lock);
2587 
2588 	sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2589 
2590 	if (zp->z_blksz == 0) {
2591 		/*
2592 		 * Block size hasn't been set; suggest maximal I/O transfers.
2593 		 */
2594 		vap->va_blksize = zfsvfs->z_max_blksz;
2595 	}
2596 
2597 	ZFS_EXIT(zfsvfs);
2598 	return (0);
2599 }
2600 
2601 /*
2602  * Set the file attributes to the values contained in the
2603  * vattr structure.
2604  *
2605  *	IN:	vp	- vnode of file to be modified.
2606  *		vap	- new attribute values.
2607  *			  If AT_XVATTR set, then optional attrs are being set
2608  *		flags	- ATTR_UTIME set if non-default time values provided.
2609  *			- ATTR_NOACLCHECK (CIFS context only).
2610  *		cr	- credentials of caller.
2611  *		ct	- caller context
2612  *
2613  *	RETURN:	0 if success
2614  *		error code if failure
2615  *
2616  * Timestamps:
2617  *	vp - ctime updated, mtime updated if size changed.
2618  */
2619 /* ARGSUSED */
2620 static int
2621 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2622 	caller_context_t *ct)
2623 {
2624 	znode_t		*zp = VTOZ(vp);
2625 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2626 	zilog_t		*zilog;
2627 	dmu_tx_t	*tx;
2628 	vattr_t		oldva;
2629 	xvattr_t	tmpxvattr;
2630 	uint_t		mask = vap->va_mask;
2631 	uint_t		saved_mask = 0;
2632 	int		trim_mask = 0;
2633 	uint64_t	new_mode;
2634 	uint64_t	new_uid, new_gid;
2635 	uint64_t	xattr_obj;
2636 	uint64_t	mtime[2], ctime[2];
2637 	znode_t		*attrzp;
2638 	int		need_policy = FALSE;
2639 	int		err, err2;
2640 	zfs_fuid_info_t *fuidp = NULL;
2641 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2642 	xoptattr_t	*xoap;
2643 	zfs_acl_t	*aclp;
2644 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2645 	boolean_t	fuid_dirtied = B_FALSE;
2646 	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
2647 	int		count = 0, xattr_count = 0;
2648 
2649 	if (mask == 0)
2650 		return (0);
2651 
2652 	if (mask & AT_NOSET)
2653 		return (EINVAL);
2654 
2655 	ZFS_ENTER(zfsvfs);
2656 	ZFS_VERIFY_ZP(zp);
2657 
2658 	zilog = zfsvfs->z_log;
2659 
2660 	/*
2661 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2662 	 * that file system is at proper version level
2663 	 */
2664 
2665 	if (zfsvfs->z_use_fuids == B_FALSE &&
2666 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2667 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2668 	    (mask & AT_XVATTR))) {
2669 		ZFS_EXIT(zfsvfs);
2670 		return (EINVAL);
2671 	}
2672 
2673 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2674 		ZFS_EXIT(zfsvfs);
2675 		return (EISDIR);
2676 	}
2677 
2678 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2679 		ZFS_EXIT(zfsvfs);
2680 		return (EINVAL);
2681 	}
2682 
2683 	/*
2684 	 * If this is an xvattr_t, then get a pointer to the structure of
2685 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2686 	 */
2687 	xoap = xva_getxoptattr(xvap);
2688 
2689 	xva_init(&tmpxvattr);
2690 
2691 	/*
2692 	 * Immutable files can only alter immutable bit and atime
2693 	 */
2694 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2695 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2696 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2697 		ZFS_EXIT(zfsvfs);
2698 		return (EPERM);
2699 	}
2700 
2701 	if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2702 		ZFS_EXIT(zfsvfs);
2703 		return (EPERM);
2704 	}
2705 
2706 	/*
2707 	 * Verify timestamps doesn't overflow 32 bits.
2708 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2709 	 * handle times greater than 2039.  This check should be removed
2710 	 * once large timestamps are fully supported.
2711 	 */
2712 	if (mask & (AT_ATIME | AT_MTIME)) {
2713 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2714 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2715 			ZFS_EXIT(zfsvfs);
2716 			return (EOVERFLOW);
2717 		}
2718 	}
2719 
2720 top:
2721 	attrzp = NULL;
2722 	aclp = NULL;
2723 
2724 	/* Can this be moved to before the top label? */
2725 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2726 		ZFS_EXIT(zfsvfs);
2727 		return (EROFS);
2728 	}
2729 
2730 	/*
2731 	 * First validate permissions
2732 	 */
2733 
2734 	if (mask & AT_SIZE) {
2735 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2736 		if (err) {
2737 			ZFS_EXIT(zfsvfs);
2738 			return (err);
2739 		}
2740 		/*
2741 		 * XXX - Note, we are not providing any open
2742 		 * mode flags here (like FNDELAY), so we may
2743 		 * block if there are locks present... this
2744 		 * should be addressed in openat().
2745 		 */
2746 		/* XXX - would it be OK to generate a log record here? */
2747 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2748 		if (err) {
2749 			ZFS_EXIT(zfsvfs);
2750 			return (err);
2751 		}
2752 	}
2753 
2754 	if (mask & (AT_ATIME|AT_MTIME) ||
2755 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2756 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2757 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2758 	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2759 	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2760 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2761 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2762 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2763 		    skipaclchk, cr);
2764 	}
2765 
2766 	if (mask & (AT_UID|AT_GID)) {
2767 		int	idmask = (mask & (AT_UID|AT_GID));
2768 		int	take_owner;
2769 		int	take_group;
2770 
2771 		/*
2772 		 * NOTE: even if a new mode is being set,
2773 		 * we may clear S_ISUID/S_ISGID bits.
2774 		 */
2775 
2776 		if (!(mask & AT_MODE))
2777 			vap->va_mode = zp->z_mode;
2778 
2779 		/*
2780 		 * Take ownership or chgrp to group we are a member of
2781 		 */
2782 
2783 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2784 		take_group = (mask & AT_GID) &&
2785 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2786 
2787 		/*
2788 		 * If both AT_UID and AT_GID are set then take_owner and
2789 		 * take_group must both be set in order to allow taking
2790 		 * ownership.
2791 		 *
2792 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2793 		 *
2794 		 */
2795 
2796 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2797 		    ((idmask == AT_UID) && take_owner) ||
2798 		    ((idmask == AT_GID) && take_group)) {
2799 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2800 			    skipaclchk, cr) == 0) {
2801 				/*
2802 				 * Remove setuid/setgid for non-privileged users
2803 				 */
2804 				secpolicy_setid_clear(vap, cr);
2805 				trim_mask = (mask & (AT_UID|AT_GID));
2806 			} else {
2807 				need_policy =  TRUE;
2808 			}
2809 		} else {
2810 			need_policy =  TRUE;
2811 		}
2812 	}
2813 
2814 	mutex_enter(&zp->z_lock);
2815 	oldva.va_mode = zp->z_mode;
2816 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2817 	if (mask & AT_XVATTR) {
2818 		/*
2819 		 * Update xvattr mask to include only those attributes
2820 		 * that are actually changing.
2821 		 *
2822 		 * the bits will be restored prior to actually setting
2823 		 * the attributes so the caller thinks they were set.
2824 		 */
2825 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2826 			if (xoap->xoa_appendonly !=
2827 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2828 				need_policy = TRUE;
2829 			} else {
2830 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2831 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2832 			}
2833 		}
2834 
2835 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2836 			if (xoap->xoa_nounlink !=
2837 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2838 				need_policy = TRUE;
2839 			} else {
2840 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2841 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2842 			}
2843 		}
2844 
2845 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2846 			if (xoap->xoa_immutable !=
2847 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2848 				need_policy = TRUE;
2849 			} else {
2850 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2851 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2852 			}
2853 		}
2854 
2855 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2856 			if (xoap->xoa_nodump !=
2857 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2858 				need_policy = TRUE;
2859 			} else {
2860 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2861 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2862 			}
2863 		}
2864 
2865 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2866 			if (xoap->xoa_av_modified !=
2867 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2868 				need_policy = TRUE;
2869 			} else {
2870 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2871 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2872 			}
2873 		}
2874 
2875 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2876 			if ((vp->v_type != VREG &&
2877 			    xoap->xoa_av_quarantined) ||
2878 			    xoap->xoa_av_quarantined !=
2879 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2880 				need_policy = TRUE;
2881 			} else {
2882 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2883 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2884 			}
2885 		}
2886 
2887 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2888 			mutex_exit(&zp->z_lock);
2889 			ZFS_EXIT(zfsvfs);
2890 			return (EPERM);
2891 		}
2892 
2893 		if (need_policy == FALSE &&
2894 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2895 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2896 			need_policy = TRUE;
2897 		}
2898 	}
2899 
2900 	mutex_exit(&zp->z_lock);
2901 
2902 	if (mask & AT_MODE) {
2903 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2904 			err = secpolicy_setid_setsticky_clear(vp, vap,
2905 			    &oldva, cr);
2906 			if (err) {
2907 				ZFS_EXIT(zfsvfs);
2908 				return (err);
2909 			}
2910 			trim_mask |= AT_MODE;
2911 		} else {
2912 			need_policy = TRUE;
2913 		}
2914 	}
2915 
2916 	if (need_policy) {
2917 		/*
2918 		 * If trim_mask is set then take ownership
2919 		 * has been granted or write_acl is present and user
2920 		 * has the ability to modify mode.  In that case remove
2921 		 * UID|GID and or MODE from mask so that
2922 		 * secpolicy_vnode_setattr() doesn't revoke it.
2923 		 */
2924 
2925 		if (trim_mask) {
2926 			saved_mask = vap->va_mask;
2927 			vap->va_mask &= ~trim_mask;
2928 		}
2929 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2930 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2931 		if (err) {
2932 			ZFS_EXIT(zfsvfs);
2933 			return (err);
2934 		}
2935 
2936 		if (trim_mask)
2937 			vap->va_mask |= saved_mask;
2938 	}
2939 
2940 	/*
2941 	 * secpolicy_vnode_setattr, or take ownership may have
2942 	 * changed va_mask
2943 	 */
2944 	mask = vap->va_mask;
2945 
2946 	if ((mask & (AT_UID | AT_GID))) {
2947 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2948 		    &xattr_obj, sizeof (xattr_obj));
2949 
2950 		if (err == 0 && xattr_obj) {
2951 			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
2952 			if (err)
2953 				goto out2;
2954 		}
2955 		if (mask & AT_UID) {
2956 			new_uid = zfs_fuid_create(zfsvfs,
2957 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2958 			if (new_uid != zp->z_uid &&
2959 			    zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
2960 				if (attrzp)
2961 					VN_RELE(ZTOV(attrzp));
2962 				err = EDQUOT;
2963 				goto out2;
2964 			}
2965 		}
2966 
2967 		if (mask & AT_GID) {
2968 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
2969 			    cr, ZFS_GROUP, &fuidp);
2970 			if (new_gid != zp->z_gid &&
2971 			    zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
2972 				if (attrzp)
2973 					VN_RELE(ZTOV(attrzp));
2974 				err = EDQUOT;
2975 				goto out2;
2976 			}
2977 		}
2978 	}
2979 	tx = dmu_tx_create(zfsvfs->z_os);
2980 
2981 	if (mask & AT_MODE) {
2982 		uint64_t pmode = zp->z_mode;
2983 		uint64_t acl_obj;
2984 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2985 
2986 		if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
2987 		    !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
2988 			err = EPERM;
2989 			goto out;
2990 		}
2991 
2992 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
2993 			goto out;
2994 
2995 		mutex_enter(&zp->z_lock);
2996 		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2997 			/*
2998 			 * Are we upgrading ACL from old V0 format
2999 			 * to V1 format?
3000 			 */
3001 			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3002 			    zfs_znode_acl_version(zp) ==
3003 			    ZFS_ACL_VERSION_INITIAL) {
3004 				dmu_tx_hold_free(tx, acl_obj, 0,
3005 				    DMU_OBJECT_END);
3006 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3007 				    0, aclp->z_acl_bytes);
3008 			} else {
3009 				dmu_tx_hold_write(tx, acl_obj, 0,
3010 				    aclp->z_acl_bytes);
3011 			}
3012 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3013 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3014 			    0, aclp->z_acl_bytes);
3015 		}
3016 		mutex_exit(&zp->z_lock);
3017 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3018 	} else {
3019 		if ((mask & AT_XVATTR) &&
3020 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3021 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3022 		else
3023 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3024 	}
3025 
3026 	if (attrzp) {
3027 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3028 	}
3029 
3030 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3031 	if (fuid_dirtied)
3032 		zfs_fuid_txhold(zfsvfs, tx);
3033 
3034 	zfs_sa_upgrade_txholds(tx, zp);
3035 
3036 	err = dmu_tx_assign(tx, TXG_NOWAIT);
3037 	if (err) {
3038 		if (err == ERESTART)
3039 			dmu_tx_wait(tx);
3040 		goto out;
3041 	}
3042 
3043 	count = 0;
3044 	/*
3045 	 * Set each attribute requested.
3046 	 * We group settings according to the locks they need to acquire.
3047 	 *
3048 	 * Note: you cannot set ctime directly, although it will be
3049 	 * updated as a side-effect of calling this function.
3050 	 */
3051 
3052 
3053 	if (mask & (AT_UID|AT_GID|AT_MODE))
3054 		mutex_enter(&zp->z_acl_lock);
3055 	mutex_enter(&zp->z_lock);
3056 
3057 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3058 	    &zp->z_pflags, sizeof (zp->z_pflags));
3059 
3060 	if (attrzp) {
3061 		if (mask & (AT_UID|AT_GID|AT_MODE))
3062 			mutex_enter(&attrzp->z_acl_lock);
3063 		mutex_enter(&attrzp->z_lock);
3064 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3065 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3066 		    sizeof (attrzp->z_pflags));
3067 	}
3068 
3069 	if (mask & (AT_UID|AT_GID)) {
3070 
3071 		if (mask & AT_UID) {
3072 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3073 			    &new_uid, sizeof (new_uid));
3074 			zp->z_uid = new_uid;
3075 			if (attrzp) {
3076 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3077 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3078 				    sizeof (new_uid));
3079 				attrzp->z_uid = new_uid;
3080 			}
3081 		}
3082 
3083 		if (mask & AT_GID) {
3084 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3085 			    NULL, &new_gid, sizeof (new_gid));
3086 			zp->z_gid = new_gid;
3087 			if (attrzp) {
3088 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3089 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3090 				    sizeof (new_gid));
3091 				attrzp->z_gid = new_gid;
3092 			}
3093 		}
3094 		if (!(mask & AT_MODE)) {
3095 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3096 			    NULL, &new_mode, sizeof (new_mode));
3097 			new_mode = zp->z_mode;
3098 		}
3099 		err = zfs_acl_chown_setattr(zp);
3100 		ASSERT(err == 0);
3101 		if (attrzp) {
3102 			err = zfs_acl_chown_setattr(attrzp);
3103 			ASSERT(err == 0);
3104 		}
3105 	}
3106 
3107 	if (mask & AT_MODE) {
3108 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3109 		    &new_mode, sizeof (new_mode));
3110 		zp->z_mode = new_mode;
3111 		ASSERT3U((uintptr_t)aclp, !=, NULL);
3112 		err = zfs_aclset_common(zp, aclp, cr, tx);
3113 		ASSERT0(err);
3114 		if (zp->z_acl_cached)
3115 			zfs_acl_free(zp->z_acl_cached);
3116 		zp->z_acl_cached = aclp;
3117 		aclp = NULL;
3118 	}
3119 
3120 
3121 	if (mask & AT_ATIME) {
3122 		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3123 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3124 		    &zp->z_atime, sizeof (zp->z_atime));
3125 	}
3126 
3127 	if (mask & AT_MTIME) {
3128 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3129 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3130 		    mtime, sizeof (mtime));
3131 	}
3132 
3133 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3134 	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3135 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3136 		    NULL, mtime, sizeof (mtime));
3137 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3138 		    &ctime, sizeof (ctime));
3139 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3140 		    B_TRUE);
3141 	} else if (mask != 0) {
3142 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3143 		    &ctime, sizeof (ctime));
3144 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3145 		    B_TRUE);
3146 		if (attrzp) {
3147 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3148 			    SA_ZPL_CTIME(zfsvfs), NULL,
3149 			    &ctime, sizeof (ctime));
3150 			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3151 			    mtime, ctime, B_TRUE);
3152 		}
3153 	}
3154 	/*
3155 	 * Do this after setting timestamps to prevent timestamp
3156 	 * update from toggling bit
3157 	 */
3158 
3159 	if (xoap && (mask & AT_XVATTR)) {
3160 
3161 		/*
3162 		 * restore trimmed off masks
3163 		 * so that return masks can be set for caller.
3164 		 */
3165 
3166 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3167 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
3168 		}
3169 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3170 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
3171 		}
3172 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3173 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3174 		}
3175 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3176 			XVA_SET_REQ(xvap, XAT_NODUMP);
3177 		}
3178 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3179 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3180 		}
3181 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3182 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3183 		}
3184 
3185 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3186 			ASSERT(vp->v_type == VREG);
3187 
3188 		zfs_xvattr_set(zp, xvap, tx);
3189 	}
3190 
3191 	if (fuid_dirtied)
3192 		zfs_fuid_sync(zfsvfs, tx);
3193 
3194 	if (mask != 0)
3195 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3196 
3197 	mutex_exit(&zp->z_lock);
3198 	if (mask & (AT_UID|AT_GID|AT_MODE))
3199 		mutex_exit(&zp->z_acl_lock);
3200 
3201 	if (attrzp) {
3202 		if (mask & (AT_UID|AT_GID|AT_MODE))
3203 			mutex_exit(&attrzp->z_acl_lock);
3204 		mutex_exit(&attrzp->z_lock);
3205 	}
3206 out:
3207 	if (err == 0 && attrzp) {
3208 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3209 		    xattr_count, tx);
3210 		ASSERT(err2 == 0);
3211 	}
3212 
3213 	if (attrzp)
3214 		VN_RELE(ZTOV(attrzp));
3215 	if (aclp)
3216 		zfs_acl_free(aclp);
3217 
3218 	if (fuidp) {
3219 		zfs_fuid_info_free(fuidp);
3220 		fuidp = NULL;
3221 	}
3222 
3223 	if (err) {
3224 		dmu_tx_abort(tx);
3225 		if (err == ERESTART)
3226 			goto top;
3227 	} else {
3228 		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3229 		dmu_tx_commit(tx);
3230 	}
3231 
3232 out2:
3233 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3234 		zil_commit(zilog, 0);
3235 
3236 	ZFS_EXIT(zfsvfs);
3237 	return (err);
3238 }
3239 
3240 typedef struct zfs_zlock {
3241 	krwlock_t	*zl_rwlock;	/* lock we acquired */
3242 	znode_t		*zl_znode;	/* znode we held */
3243 	struct zfs_zlock *zl_next;	/* next in list */
3244 } zfs_zlock_t;
3245 
3246 /*
3247  * Drop locks and release vnodes that were held by zfs_rename_lock().
3248  */
3249 static void
3250 zfs_rename_unlock(zfs_zlock_t **zlpp)
3251 {
3252 	zfs_zlock_t *zl;
3253 
3254 	while ((zl = *zlpp) != NULL) {
3255 		if (zl->zl_znode != NULL)
3256 			VN_RELE(ZTOV(zl->zl_znode));
3257 		rw_exit(zl->zl_rwlock);
3258 		*zlpp = zl->zl_next;
3259 		kmem_free(zl, sizeof (*zl));
3260 	}
3261 }
3262 
3263 /*
3264  * Search back through the directory tree, using the ".." entries.
3265  * Lock each directory in the chain to prevent concurrent renames.
3266  * Fail any attempt to move a directory into one of its own descendants.
3267  * XXX - z_parent_lock can overlap with map or grow locks
3268  */
3269 static int
3270 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3271 {
3272 	zfs_zlock_t	*zl;
3273 	znode_t		*zp = tdzp;
3274 	uint64_t	rootid = zp->z_zfsvfs->z_root;
3275 	uint64_t	oidp = zp->z_id;
3276 	krwlock_t	*rwlp = &szp->z_parent_lock;
3277 	krw_t		rw = RW_WRITER;
3278 
3279 	/*
3280 	 * First pass write-locks szp and compares to zp->z_id.
3281 	 * Later passes read-lock zp and compare to zp->z_parent.
3282 	 */
3283 	do {
3284 		if (!rw_tryenter(rwlp, rw)) {
3285 			/*
3286 			 * Another thread is renaming in this path.
3287 			 * Note that if we are a WRITER, we don't have any
3288 			 * parent_locks held yet.
3289 			 */
3290 			if (rw == RW_READER && zp->z_id > szp->z_id) {
3291 				/*
3292 				 * Drop our locks and restart
3293 				 */
3294 				zfs_rename_unlock(&zl);
3295 				*zlpp = NULL;
3296 				zp = tdzp;
3297 				oidp = zp->z_id;
3298 				rwlp = &szp->z_parent_lock;
3299 				rw = RW_WRITER;
3300 				continue;
3301 			} else {
3302 				/*
3303 				 * Wait for other thread to drop its locks
3304 				 */
3305 				rw_enter(rwlp, rw);
3306 			}
3307 		}
3308 
3309 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3310 		zl->zl_rwlock = rwlp;
3311 		zl->zl_znode = NULL;
3312 		zl->zl_next = *zlpp;
3313 		*zlpp = zl;
3314 
3315 		if (oidp == szp->z_id)		/* We're a descendant of szp */
3316 			return (EINVAL);
3317 
3318 		if (oidp == rootid)		/* We've hit the top */
3319 			return (0);
3320 
3321 		if (rw == RW_READER) {		/* i.e. not the first pass */
3322 			int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3323 			if (error)
3324 				return (error);
3325 			zl->zl_znode = zp;
3326 		}
3327 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3328 		    &oidp, sizeof (oidp));
3329 		rwlp = &zp->z_parent_lock;
3330 		rw = RW_READER;
3331 
3332 	} while (zp->z_id != sdzp->z_id);
3333 
3334 	return (0);
3335 }
3336 
3337 /*
3338  * Move an entry from the provided source directory to the target
3339  * directory.  Change the entry name as indicated.
3340  *
3341  *	IN:	sdvp	- Source directory containing the "old entry".
3342  *		snm	- Old entry name.
3343  *		tdvp	- Target directory to contain the "new entry".
3344  *		tnm	- New entry name.
3345  *		cr	- credentials of caller.
3346  *		ct	- caller context
3347  *		flags	- case flags
3348  *
3349  *	RETURN:	0 if success
3350  *		error code if failure
3351  *
3352  * Timestamps:
3353  *	sdvp,tdvp - ctime|mtime updated
3354  */
3355 /*ARGSUSED*/
3356 static int
3357 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3358     caller_context_t *ct, int flags)
3359 {
3360 	znode_t		*tdzp, *szp, *tzp;
3361 	znode_t		*sdzp = VTOZ(sdvp);
3362 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
3363 	zilog_t		*zilog;
3364 	vnode_t		*realvp;
3365 	zfs_dirlock_t	*sdl, *tdl;
3366 	dmu_tx_t	*tx;
3367 	zfs_zlock_t	*zl;
3368 	int		cmp, serr, terr;
3369 	int		error = 0;
3370 	int		zflg = 0;
3371 
3372 	ZFS_ENTER(zfsvfs);
3373 	ZFS_VERIFY_ZP(sdzp);
3374 	zilog = zfsvfs->z_log;
3375 
3376 	/*
3377 	 * Make sure we have the real vp for the target directory.
3378 	 */
3379 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3380 		tdvp = realvp;
3381 
3382 	if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3383 		ZFS_EXIT(zfsvfs);
3384 		return (EXDEV);
3385 	}
3386 
3387 	tdzp = VTOZ(tdvp);
3388 	ZFS_VERIFY_ZP(tdzp);
3389 	if (zfsvfs->z_utf8 && u8_validate(tnm,
3390 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3391 		ZFS_EXIT(zfsvfs);
3392 		return (EILSEQ);
3393 	}
3394 
3395 	if (flags & FIGNORECASE)
3396 		zflg |= ZCILOOK;
3397 
3398 top:
3399 	szp = NULL;
3400 	tzp = NULL;
3401 	zl = NULL;
3402 
3403 	/*
3404 	 * This is to prevent the creation of links into attribute space
3405 	 * by renaming a linked file into/outof an attribute directory.
3406 	 * See the comment in zfs_link() for why this is considered bad.
3407 	 */
3408 	if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3409 		ZFS_EXIT(zfsvfs);
3410 		return (EINVAL);
3411 	}
3412 
3413 	/*
3414 	 * Lock source and target directory entries.  To prevent deadlock,
3415 	 * a lock ordering must be defined.  We lock the directory with
3416 	 * the smallest object id first, or if it's a tie, the one with
3417 	 * the lexically first name.
3418 	 */
3419 	if (sdzp->z_id < tdzp->z_id) {
3420 		cmp = -1;
3421 	} else if (sdzp->z_id > tdzp->z_id) {
3422 		cmp = 1;
3423 	} else {
3424 		/*
3425 		 * First compare the two name arguments without
3426 		 * considering any case folding.
3427 		 */
3428 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3429 
3430 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3431 		ASSERT(error == 0 || !zfsvfs->z_utf8);
3432 		if (cmp == 0) {
3433 			/*
3434 			 * POSIX: "If the old argument and the new argument
3435 			 * both refer to links to the same existing file,
3436 			 * the rename() function shall return successfully
3437 			 * and perform no other action."
3438 			 */
3439 			ZFS_EXIT(zfsvfs);
3440 			return (0);
3441 		}
3442 		/*
3443 		 * If the file system is case-folding, then we may
3444 		 * have some more checking to do.  A case-folding file
3445 		 * system is either supporting mixed case sensitivity
3446 		 * access or is completely case-insensitive.  Note
3447 		 * that the file system is always case preserving.
3448 		 *
3449 		 * In mixed sensitivity mode case sensitive behavior
3450 		 * is the default.  FIGNORECASE must be used to
3451 		 * explicitly request case insensitive behavior.
3452 		 *
3453 		 * If the source and target names provided differ only
3454 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
3455 		 * we will treat this as a special case in the
3456 		 * case-insensitive mode: as long as the source name
3457 		 * is an exact match, we will allow this to proceed as
3458 		 * a name-change request.
3459 		 */
3460 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3461 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
3462 		    flags & FIGNORECASE)) &&
3463 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3464 		    &error) == 0) {
3465 			/*
3466 			 * case preserving rename request, require exact
3467 			 * name matches
3468 			 */
3469 			zflg |= ZCIEXACT;
3470 			zflg &= ~ZCILOOK;
3471 		}
3472 	}
3473 
3474 	/*
3475 	 * If the source and destination directories are the same, we should
3476 	 * grab the z_name_lock of that directory only once.
3477 	 */
3478 	if (sdzp == tdzp) {
3479 		zflg |= ZHAVELOCK;
3480 		rw_enter(&sdzp->z_name_lock, RW_READER);
3481 	}
3482 
3483 	if (cmp < 0) {
3484 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3485 		    ZEXISTS | zflg, NULL, NULL);
3486 		terr = zfs_dirent_lock(&tdl,
3487 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3488 	} else {
3489 		terr = zfs_dirent_lock(&tdl,
3490 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3491 		serr = zfs_dirent_lock(&sdl,
3492 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3493 		    NULL, NULL);
3494 	}
3495 
3496 	if (serr) {
3497 		/*
3498 		 * Source entry invalid or not there.
3499 		 */
3500 		if (!terr) {
3501 			zfs_dirent_unlock(tdl);
3502 			if (tzp)
3503 				VN_RELE(ZTOV(tzp));
3504 		}
3505 
3506 		if (sdzp == tdzp)
3507 			rw_exit(&sdzp->z_name_lock);
3508 
3509 		if (strcmp(snm, "..") == 0)
3510 			serr = EINVAL;
3511 		ZFS_EXIT(zfsvfs);
3512 		return (serr);
3513 	}
3514 	if (terr) {
3515 		zfs_dirent_unlock(sdl);
3516 		VN_RELE(ZTOV(szp));
3517 
3518 		if (sdzp == tdzp)
3519 			rw_exit(&sdzp->z_name_lock);
3520 
3521 		if (strcmp(tnm, "..") == 0)
3522 			terr = EINVAL;
3523 		ZFS_EXIT(zfsvfs);
3524 		return (terr);
3525 	}
3526 
3527 	/*
3528 	 * Must have write access at the source to remove the old entry
3529 	 * and write access at the target to create the new entry.
3530 	 * Note that if target and source are the same, this can be
3531 	 * done in a single check.
3532 	 */
3533 
3534 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3535 		goto out;
3536 
3537 	if (ZTOV(szp)->v_type == VDIR) {
3538 		/*
3539 		 * Check to make sure rename is valid.
3540 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3541 		 */
3542 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3543 			goto out;
3544 	}
3545 
3546 	/*
3547 	 * Does target exist?
3548 	 */
3549 	if (tzp) {
3550 		/*
3551 		 * Source and target must be the same type.
3552 		 */
3553 		if (ZTOV(szp)->v_type == VDIR) {
3554 			if (ZTOV(tzp)->v_type != VDIR) {
3555 				error = ENOTDIR;
3556 				goto out;
3557 			}
3558 		} else {
3559 			if (ZTOV(tzp)->v_type == VDIR) {
3560 				error = EISDIR;
3561 				goto out;
3562 			}
3563 		}
3564 		/*
3565 		 * POSIX dictates that when the source and target
3566 		 * entries refer to the same file object, rename
3567 		 * must do nothing and exit without error.
3568 		 */
3569 		if (szp->z_id == tzp->z_id) {
3570 			error = 0;
3571 			goto out;
3572 		}
3573 	}
3574 
3575 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3576 	if (tzp)
3577 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3578 
3579 	/*
3580 	 * notify the target directory if it is not the same
3581 	 * as source directory.
3582 	 */
3583 	if (tdvp != sdvp) {
3584 		vnevent_rename_dest_dir(tdvp, ct);
3585 	}
3586 
3587 	tx = dmu_tx_create(zfsvfs->z_os);
3588 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3589 	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3590 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3591 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3592 	if (sdzp != tdzp) {
3593 		dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3594 		zfs_sa_upgrade_txholds(tx, tdzp);
3595 	}
3596 	if (tzp) {
3597 		dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3598 		zfs_sa_upgrade_txholds(tx, tzp);
3599 	}
3600 
3601 	zfs_sa_upgrade_txholds(tx, szp);
3602 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3603 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3604 	if (error) {
3605 		if (zl != NULL)
3606 			zfs_rename_unlock(&zl);
3607 		zfs_dirent_unlock(sdl);
3608 		zfs_dirent_unlock(tdl);
3609 
3610 		if (sdzp == tdzp)
3611 			rw_exit(&sdzp->z_name_lock);
3612 
3613 		VN_RELE(ZTOV(szp));
3614 		if (tzp)
3615 			VN_RELE(ZTOV(tzp));
3616 		if (error == ERESTART) {
3617 			dmu_tx_wait(tx);
3618 			dmu_tx_abort(tx);
3619 			goto top;
3620 		}
3621 		dmu_tx_abort(tx);
3622 		ZFS_EXIT(zfsvfs);
3623 		return (error);
3624 	}
3625 
3626 	if (tzp)	/* Attempt to remove the existing target */
3627 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3628 
3629 	if (error == 0) {
3630 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3631 		if (error == 0) {
3632 			szp->z_pflags |= ZFS_AV_MODIFIED;
3633 
3634 			error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3635 			    (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3636 			ASSERT0(error);
3637 
3638 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3639 			if (error == 0) {
3640 				zfs_log_rename(zilog, tx, TX_RENAME |
3641 				    (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3642 				    sdl->dl_name, tdzp, tdl->dl_name, szp);
3643 
3644 				/*
3645 				 * Update path information for the target vnode
3646 				 */
3647 				vn_renamepath(tdvp, ZTOV(szp), tnm,
3648 				    strlen(tnm));
3649 			} else {
3650 				/*
3651 				 * At this point, we have successfully created
3652 				 * the target name, but have failed to remove
3653 				 * the source name.  Since the create was done
3654 				 * with the ZRENAMING flag, there are
3655 				 * complications; for one, the link count is
3656 				 * wrong.  The easiest way to deal with this
3657 				 * is to remove the newly created target, and
3658 				 * return the original error.  This must
3659 				 * succeed; fortunately, it is very unlikely to
3660 				 * fail, since we just created it.
3661 				 */
3662 				VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3663 				    ZRENAMING, NULL), ==, 0);
3664 			}
3665 		}
3666 	}
3667 
3668 	dmu_tx_commit(tx);
3669 out:
3670 	if (zl != NULL)
3671 		zfs_rename_unlock(&zl);
3672 
3673 	zfs_dirent_unlock(sdl);
3674 	zfs_dirent_unlock(tdl);
3675 
3676 	if (sdzp == tdzp)
3677 		rw_exit(&sdzp->z_name_lock);
3678 
3679 
3680 	VN_RELE(ZTOV(szp));
3681 	if (tzp)
3682 		VN_RELE(ZTOV(tzp));
3683 
3684 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3685 		zil_commit(zilog, 0);
3686 
3687 	ZFS_EXIT(zfsvfs);
3688 	return (error);
3689 }
3690 
3691 /*
3692  * Insert the indicated symbolic reference entry into the directory.
3693  *
3694  *	IN:	dvp	- Directory to contain new symbolic link.
3695  *		link	- Name for new symlink entry.
3696  *		vap	- Attributes of new entry.
3697  *		target	- Target path of new symlink.
3698  *		cr	- credentials of caller.
3699  *		ct	- caller context
3700  *		flags	- case flags
3701  *
3702  *	RETURN:	0 if success
3703  *		error code if failure
3704  *
3705  * Timestamps:
3706  *	dvp - ctime|mtime updated
3707  */
3708 /*ARGSUSED*/
3709 static int
3710 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3711     caller_context_t *ct, int flags)
3712 {
3713 	znode_t		*zp, *dzp = VTOZ(dvp);
3714 	zfs_dirlock_t	*dl;
3715 	dmu_tx_t	*tx;
3716 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3717 	zilog_t		*zilog;
3718 	uint64_t	len = strlen(link);
3719 	int		error;
3720 	int		zflg = ZNEW;
3721 	zfs_acl_ids_t	acl_ids;
3722 	boolean_t	fuid_dirtied;
3723 	uint64_t	txtype = TX_SYMLINK;
3724 
3725 	ASSERT(vap->va_type == VLNK);
3726 
3727 	ZFS_ENTER(zfsvfs);
3728 	ZFS_VERIFY_ZP(dzp);
3729 	zilog = zfsvfs->z_log;
3730 
3731 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3732 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3733 		ZFS_EXIT(zfsvfs);
3734 		return (EILSEQ);
3735 	}
3736 	if (flags & FIGNORECASE)
3737 		zflg |= ZCILOOK;
3738 
3739 	if (len > MAXPATHLEN) {
3740 		ZFS_EXIT(zfsvfs);
3741 		return (ENAMETOOLONG);
3742 	}
3743 
3744 	if ((error = zfs_acl_ids_create(dzp, 0,
3745 	    vap, cr, NULL, &acl_ids)) != 0) {
3746 		ZFS_EXIT(zfsvfs);
3747 		return (error);
3748 	}
3749 top:
3750 	/*
3751 	 * Attempt to lock directory; fail if entry already exists.
3752 	 */
3753 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3754 	if (error) {
3755 		zfs_acl_ids_free(&acl_ids);
3756 		ZFS_EXIT(zfsvfs);
3757 		return (error);
3758 	}
3759 
3760 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3761 		zfs_acl_ids_free(&acl_ids);
3762 		zfs_dirent_unlock(dl);
3763 		ZFS_EXIT(zfsvfs);
3764 		return (error);
3765 	}
3766 
3767 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3768 		zfs_acl_ids_free(&acl_ids);
3769 		zfs_dirent_unlock(dl);
3770 		ZFS_EXIT(zfsvfs);
3771 		return (EDQUOT);
3772 	}
3773 	tx = dmu_tx_create(zfsvfs->z_os);
3774 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3775 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3776 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3777 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3778 	    ZFS_SA_BASE_ATTR_SIZE + len);
3779 	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3780 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3781 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3782 		    acl_ids.z_aclp->z_acl_bytes);
3783 	}
3784 	if (fuid_dirtied)
3785 		zfs_fuid_txhold(zfsvfs, tx);
3786 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3787 	if (error) {
3788 		zfs_dirent_unlock(dl);
3789 		if (error == ERESTART) {
3790 			dmu_tx_wait(tx);
3791 			dmu_tx_abort(tx);
3792 			goto top;
3793 		}
3794 		zfs_acl_ids_free(&acl_ids);
3795 		dmu_tx_abort(tx);
3796 		ZFS_EXIT(zfsvfs);
3797 		return (error);
3798 	}
3799 
3800 	/*
3801 	 * Create a new object for the symlink.
3802 	 * for version 4 ZPL datsets the symlink will be an SA attribute
3803 	 */
3804 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3805 
3806 	if (fuid_dirtied)
3807 		zfs_fuid_sync(zfsvfs, tx);
3808 
3809 	mutex_enter(&zp->z_lock);
3810 	if (zp->z_is_sa)
3811 		error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3812 		    link, len, tx);
3813 	else
3814 		zfs_sa_symlink(zp, link, len, tx);
3815 	mutex_exit(&zp->z_lock);
3816 
3817 	zp->z_size = len;
3818 	(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3819 	    &zp->z_size, sizeof (zp->z_size), tx);
3820 	/*
3821 	 * Insert the new object into the directory.
3822 	 */
3823 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3824 
3825 	if (flags & FIGNORECASE)
3826 		txtype |= TX_CI;
3827 	zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3828 
3829 	zfs_acl_ids_free(&acl_ids);
3830 
3831 	dmu_tx_commit(tx);
3832 
3833 	zfs_dirent_unlock(dl);
3834 
3835 	VN_RELE(ZTOV(zp));
3836 
3837 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3838 		zil_commit(zilog, 0);
3839 
3840 	ZFS_EXIT(zfsvfs);
3841 	return (error);
3842 }
3843 
3844 /*
3845  * Return, in the buffer contained in the provided uio structure,
3846  * the symbolic path referred to by vp.
3847  *
3848  *	IN:	vp	- vnode of symbolic link.
3849  *		uoip	- structure to contain the link path.
3850  *		cr	- credentials of caller.
3851  *		ct	- caller context
3852  *
3853  *	OUT:	uio	- structure to contain the link path.
3854  *
3855  *	RETURN:	0 if success
3856  *		error code if failure
3857  *
3858  * Timestamps:
3859  *	vp - atime updated
3860  */
3861 /* ARGSUSED */
3862 static int
3863 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3864 {
3865 	znode_t		*zp = VTOZ(vp);
3866 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3867 	int		error;
3868 
3869 	ZFS_ENTER(zfsvfs);
3870 	ZFS_VERIFY_ZP(zp);
3871 
3872 	mutex_enter(&zp->z_lock);
3873 	if (zp->z_is_sa)
3874 		error = sa_lookup_uio(zp->z_sa_hdl,
3875 		    SA_ZPL_SYMLINK(zfsvfs), uio);
3876 	else
3877 		error = zfs_sa_readlink(zp, uio);
3878 	mutex_exit(&zp->z_lock);
3879 
3880 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3881 
3882 	ZFS_EXIT(zfsvfs);
3883 	return (error);
3884 }
3885 
3886 /*
3887  * Insert a new entry into directory tdvp referencing svp.
3888  *
3889  *	IN:	tdvp	- Directory to contain new entry.
3890  *		svp	- vnode of new entry.
3891  *		name	- name of new entry.
3892  *		cr	- credentials of caller.
3893  *		ct	- caller context
3894  *
3895  *	RETURN:	0 if success
3896  *		error code if failure
3897  *
3898  * Timestamps:
3899  *	tdvp - ctime|mtime updated
3900  *	 svp - ctime updated
3901  */
3902 /* ARGSUSED */
3903 static int
3904 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3905     caller_context_t *ct, int flags)
3906 {
3907 	znode_t		*dzp = VTOZ(tdvp);
3908 	znode_t		*tzp, *szp;
3909 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3910 	zilog_t		*zilog;
3911 	zfs_dirlock_t	*dl;
3912 	dmu_tx_t	*tx;
3913 	vnode_t		*realvp;
3914 	int		error;
3915 	int		zf = ZNEW;
3916 	uint64_t	parent;
3917 	uid_t		owner;
3918 
3919 	ASSERT(tdvp->v_type == VDIR);
3920 
3921 	ZFS_ENTER(zfsvfs);
3922 	ZFS_VERIFY_ZP(dzp);
3923 	zilog = zfsvfs->z_log;
3924 
3925 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3926 		svp = realvp;
3927 
3928 	/*
3929 	 * POSIX dictates that we return EPERM here.
3930 	 * Better choices include ENOTSUP or EISDIR.
3931 	 */
3932 	if (svp->v_type == VDIR) {
3933 		ZFS_EXIT(zfsvfs);
3934 		return (EPERM);
3935 	}
3936 
3937 	if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
3938 		ZFS_EXIT(zfsvfs);
3939 		return (EXDEV);
3940 	}
3941 
3942 	szp = VTOZ(svp);
3943 	ZFS_VERIFY_ZP(szp);
3944 
3945 	/* Prevent links to .zfs/shares files */
3946 
3947 	if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3948 	    &parent, sizeof (uint64_t))) != 0) {
3949 		ZFS_EXIT(zfsvfs);
3950 		return (error);
3951 	}
3952 	if (parent == zfsvfs->z_shares_dir) {
3953 		ZFS_EXIT(zfsvfs);
3954 		return (EPERM);
3955 	}
3956 
3957 	if (zfsvfs->z_utf8 && u8_validate(name,
3958 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3959 		ZFS_EXIT(zfsvfs);
3960 		return (EILSEQ);
3961 	}
3962 	if (flags & FIGNORECASE)
3963 		zf |= ZCILOOK;
3964 
3965 	/*
3966 	 * We do not support links between attributes and non-attributes
3967 	 * because of the potential security risk of creating links
3968 	 * into "normal" file space in order to circumvent restrictions
3969 	 * imposed in attribute space.
3970 	 */
3971 	if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3972 		ZFS_EXIT(zfsvfs);
3973 		return (EINVAL);
3974 	}
3975 
3976 
3977 	owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
3978 	if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3979 		ZFS_EXIT(zfsvfs);
3980 		return (EPERM);
3981 	}
3982 
3983 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3984 		ZFS_EXIT(zfsvfs);
3985 		return (error);
3986 	}
3987 
3988 top:
3989 	/*
3990 	 * Attempt to lock directory; fail if entry already exists.
3991 	 */
3992 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3993 	if (error) {
3994 		ZFS_EXIT(zfsvfs);
3995 		return (error);
3996 	}
3997 
3998 	tx = dmu_tx_create(zfsvfs->z_os);
3999 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4000 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4001 	zfs_sa_upgrade_txholds(tx, szp);
4002 	zfs_sa_upgrade_txholds(tx, dzp);
4003 	error = dmu_tx_assign(tx, TXG_NOWAIT);
4004 	if (error) {
4005 		zfs_dirent_unlock(dl);
4006 		if (error == ERESTART) {
4007 			dmu_tx_wait(tx);
4008 			dmu_tx_abort(tx);
4009 			goto top;
4010 		}
4011 		dmu_tx_abort(tx);
4012 		ZFS_EXIT(zfsvfs);
4013 		return (error);
4014 	}
4015 
4016 	error = zfs_link_create(dl, szp, tx, 0);
4017 
4018 	if (error == 0) {
4019 		uint64_t txtype = TX_LINK;
4020 		if (flags & FIGNORECASE)
4021 			txtype |= TX_CI;
4022 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4023 	}
4024 
4025 	dmu_tx_commit(tx);
4026 
4027 	zfs_dirent_unlock(dl);
4028 
4029 	if (error == 0) {
4030 		vnevent_link(svp, ct);
4031 	}
4032 
4033 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4034 		zil_commit(zilog, 0);
4035 
4036 	ZFS_EXIT(zfsvfs);
4037 	return (error);
4038 }
4039 
4040 /*
4041  * zfs_null_putapage() is used when the file system has been force
4042  * unmounted. It just drops the pages.
4043  */
4044 /* ARGSUSED */
4045 static int
4046 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4047 		size_t *lenp, int flags, cred_t *cr)
4048 {
4049 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4050 	return (0);
4051 }
4052 
4053 /*
4054  * Push a page out to disk, klustering if possible.
4055  *
4056  *	IN:	vp	- file to push page to.
4057  *		pp	- page to push.
4058  *		flags	- additional flags.
4059  *		cr	- credentials of caller.
4060  *
4061  *	OUT:	offp	- start of range pushed.
4062  *		lenp	- len of range pushed.
4063  *
4064  *	RETURN:	0 if success
4065  *		error code if failure
4066  *
4067  * NOTE: callers must have locked the page to be pushed.  On
4068  * exit, the page (and all other pages in the kluster) must be
4069  * unlocked.
4070  */
4071 /* ARGSUSED */
4072 static int
4073 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4074 		size_t *lenp, int flags, cred_t *cr)
4075 {
4076 	znode_t		*zp = VTOZ(vp);
4077 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4078 	dmu_tx_t	*tx;
4079 	u_offset_t	off, koff;
4080 	size_t		len, klen;
4081 	int		err;
4082 
4083 	off = pp->p_offset;
4084 	len = PAGESIZE;
4085 	/*
4086 	 * If our blocksize is bigger than the page size, try to kluster
4087 	 * multiple pages so that we write a full block (thus avoiding
4088 	 * a read-modify-write).
4089 	 */
4090 	if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4091 		klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4092 		koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4093 		ASSERT(koff <= zp->z_size);
4094 		if (koff + klen > zp->z_size)
4095 			klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4096 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4097 	}
4098 	ASSERT3U(btop(len), ==, btopr(len));
4099 
4100 	/*
4101 	 * Can't push pages past end-of-file.
4102 	 */
4103 	if (off >= zp->z_size) {
4104 		/* ignore all pages */
4105 		err = 0;
4106 		goto out;
4107 	} else if (off + len > zp->z_size) {
4108 		int npages = btopr(zp->z_size - off);
4109 		page_t *trunc;
4110 
4111 		page_list_break(&pp, &trunc, npages);
4112 		/* ignore pages past end of file */
4113 		if (trunc)
4114 			pvn_write_done(trunc, flags);
4115 		len = zp->z_size - off;
4116 	}
4117 
4118 	if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4119 	    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4120 		err = EDQUOT;
4121 		goto out;
4122 	}
4123 top:
4124 	tx = dmu_tx_create(zfsvfs->z_os);
4125 	dmu_tx_hold_write(tx, zp->z_id, off, len);
4126 
4127 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4128 	zfs_sa_upgrade_txholds(tx, zp);
4129 	err = dmu_tx_assign(tx, TXG_NOWAIT);
4130 	if (err != 0) {
4131 		if (err == ERESTART) {
4132 			dmu_tx_wait(tx);
4133 			dmu_tx_abort(tx);
4134 			goto top;
4135 		}
4136 		dmu_tx_abort(tx);
4137 		goto out;
4138 	}
4139 
4140 	if (zp->z_blksz <= PAGESIZE) {
4141 		caddr_t va = zfs_map_page(pp, S_READ);
4142 		ASSERT3U(len, <=, PAGESIZE);
4143 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4144 		zfs_unmap_page(pp, va);
4145 	} else {
4146 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4147 	}
4148 
4149 	if (err == 0) {
4150 		uint64_t mtime[2], ctime[2];
4151 		sa_bulk_attr_t bulk[3];
4152 		int count = 0;
4153 
4154 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4155 		    &mtime, 16);
4156 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4157 		    &ctime, 16);
4158 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4159 		    &zp->z_pflags, 8);
4160 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4161 		    B_TRUE);
4162 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4163 	}
4164 	dmu_tx_commit(tx);
4165 
4166 out:
4167 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4168 	if (offp)
4169 		*offp = off;
4170 	if (lenp)
4171 		*lenp = len;
4172 
4173 	return (err);
4174 }
4175 
4176 /*
4177  * Copy the portion of the file indicated from pages into the file.
4178  * The pages are stored in a page list attached to the files vnode.
4179  *
4180  *	IN:	vp	- vnode of file to push page data to.
4181  *		off	- position in file to put data.
4182  *		len	- amount of data to write.
4183  *		flags	- flags to control the operation.
4184  *		cr	- credentials of caller.
4185  *		ct	- caller context.
4186  *
4187  *	RETURN:	0 if success
4188  *		error code if failure
4189  *
4190  * Timestamps:
4191  *	vp - ctime|mtime updated
4192  */
4193 /*ARGSUSED*/
4194 static int
4195 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4196     caller_context_t *ct)
4197 {
4198 	znode_t		*zp = VTOZ(vp);
4199 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4200 	page_t		*pp;
4201 	size_t		io_len;
4202 	u_offset_t	io_off;
4203 	uint_t		blksz;
4204 	rl_t		*rl;
4205 	int		error = 0;
4206 
4207 	ZFS_ENTER(zfsvfs);
4208 	ZFS_VERIFY_ZP(zp);
4209 
4210 	/*
4211 	 * There's nothing to do if no data is cached.
4212 	 */
4213 	if (!vn_has_cached_data(vp)) {
4214 		ZFS_EXIT(zfsvfs);
4215 		return (0);
4216 	}
4217 
4218 	/*
4219 	 * Align this request to the file block size in case we kluster.
4220 	 * XXX - this can result in pretty aggresive locking, which can
4221 	 * impact simultanious read/write access.  One option might be
4222 	 * to break up long requests (len == 0) into block-by-block
4223 	 * operations to get narrower locking.
4224 	 */
4225 	blksz = zp->z_blksz;
4226 	if (ISP2(blksz))
4227 		io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4228 	else
4229 		io_off = 0;
4230 	if (len > 0 && ISP2(blksz))
4231 		io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4232 	else
4233 		io_len = 0;
4234 
4235 	if (io_len == 0) {
4236 		/*
4237 		 * Search the entire vp list for pages >= io_off.
4238 		 */
4239 		rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4240 		error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4241 		goto out;
4242 	}
4243 	rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4244 
4245 	if (off > zp->z_size) {
4246 		/* past end of file */
4247 		zfs_range_unlock(rl);
4248 		ZFS_EXIT(zfsvfs);
4249 		return (0);
4250 	}
4251 
4252 	len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4253 
4254 	for (off = io_off; io_off < off + len; io_off += io_len) {
4255 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4256 			pp = page_lookup(vp, io_off,
4257 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4258 		} else {
4259 			pp = page_lookup_nowait(vp, io_off,
4260 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4261 		}
4262 
4263 		if (pp != NULL && pvn_getdirty(pp, flags)) {
4264 			int err;
4265 
4266 			/*
4267 			 * Found a dirty page to push
4268 			 */
4269 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4270 			if (err)
4271 				error = err;
4272 		} else {
4273 			io_len = PAGESIZE;
4274 		}
4275 	}
4276 out:
4277 	zfs_range_unlock(rl);
4278 	if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4279 		zil_commit(zfsvfs->z_log, zp->z_id);
4280 	ZFS_EXIT(zfsvfs);
4281 	return (error);
4282 }
4283 
4284 /*ARGSUSED*/
4285 void
4286 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4287 {
4288 	znode_t	*zp = VTOZ(vp);
4289 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4290 	int error;
4291 
4292 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4293 	if (zp->z_sa_hdl == NULL) {
4294 		/*
4295 		 * The fs has been unmounted, or we did a
4296 		 * suspend/resume and this file no longer exists.
4297 		 */
4298 		if (vn_has_cached_data(vp)) {
4299 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4300 			    B_INVAL, cr);
4301 		}
4302 
4303 		mutex_enter(&zp->z_lock);
4304 		mutex_enter(&vp->v_lock);
4305 		ASSERT(vp->v_count == 1);
4306 		vp->v_count = 0;
4307 		mutex_exit(&vp->v_lock);
4308 		mutex_exit(&zp->z_lock);
4309 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
4310 		zfs_znode_free(zp);
4311 		return;
4312 	}
4313 
4314 	/*
4315 	 * Attempt to push any data in the page cache.  If this fails
4316 	 * we will get kicked out later in zfs_zinactive().
4317 	 */
4318 	if (vn_has_cached_data(vp)) {
4319 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4320 		    cr);
4321 	}
4322 
4323 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4324 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4325 
4326 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4327 		zfs_sa_upgrade_txholds(tx, zp);
4328 		error = dmu_tx_assign(tx, TXG_WAIT);
4329 		if (error) {
4330 			dmu_tx_abort(tx);
4331 		} else {
4332 			mutex_enter(&zp->z_lock);
4333 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4334 			    (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4335 			zp->z_atime_dirty = 0;
4336 			mutex_exit(&zp->z_lock);
4337 			dmu_tx_commit(tx);
4338 		}
4339 	}
4340 
4341 	zfs_zinactive(zp);
4342 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
4343 }
4344 
4345 /*
4346  * Bounds-check the seek operation.
4347  *
4348  *	IN:	vp	- vnode seeking within
4349  *		ooff	- old file offset
4350  *		noffp	- pointer to new file offset
4351  *		ct	- caller context
4352  *
4353  *	RETURN:	0 if success
4354  *		EINVAL if new offset invalid
4355  */
4356 /* ARGSUSED */
4357 static int
4358 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4359     caller_context_t *ct)
4360 {
4361 	if (vp->v_type == VDIR)
4362 		return (0);
4363 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4364 }
4365 
4366 /*
4367  * Pre-filter the generic locking function to trap attempts to place
4368  * a mandatory lock on a memory mapped file.
4369  */
4370 static int
4371 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4372     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4373 {
4374 	znode_t *zp = VTOZ(vp);
4375 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4376 
4377 	ZFS_ENTER(zfsvfs);
4378 	ZFS_VERIFY_ZP(zp);
4379 
4380 	/*
4381 	 * We are following the UFS semantics with respect to mapcnt
4382 	 * here: If we see that the file is mapped already, then we will
4383 	 * return an error, but we don't worry about races between this
4384 	 * function and zfs_map().
4385 	 */
4386 	if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4387 		ZFS_EXIT(zfsvfs);
4388 		return (EAGAIN);
4389 	}
4390 	ZFS_EXIT(zfsvfs);
4391 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4392 }
4393 
4394 /*
4395  * If we can't find a page in the cache, we will create a new page
4396  * and fill it with file data.  For efficiency, we may try to fill
4397  * multiple pages at once (klustering) to fill up the supplied page
4398  * list.  Note that the pages to be filled are held with an exclusive
4399  * lock to prevent access by other threads while they are being filled.
4400  */
4401 static int
4402 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4403     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4404 {
4405 	znode_t *zp = VTOZ(vp);
4406 	page_t *pp, *cur_pp;
4407 	objset_t *os = zp->z_zfsvfs->z_os;
4408 	u_offset_t io_off, total;
4409 	size_t io_len;
4410 	int err;
4411 
4412 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4413 		/*
4414 		 * We only have a single page, don't bother klustering
4415 		 */
4416 		io_off = off;
4417 		io_len = PAGESIZE;
4418 		pp = page_create_va(vp, io_off, io_len,
4419 		    PG_EXCL | PG_WAIT, seg, addr);
4420 	} else {
4421 		/*
4422 		 * Try to find enough pages to fill the page list
4423 		 */
4424 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4425 		    &io_len, off, plsz, 0);
4426 	}
4427 	if (pp == NULL) {
4428 		/*
4429 		 * The page already exists, nothing to do here.
4430 		 */
4431 		*pl = NULL;
4432 		return (0);
4433 	}
4434 
4435 	/*
4436 	 * Fill the pages in the kluster.
4437 	 */
4438 	cur_pp = pp;
4439 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4440 		caddr_t va;
4441 
4442 		ASSERT3U(io_off, ==, cur_pp->p_offset);
4443 		va = zfs_map_page(cur_pp, S_WRITE);
4444 		err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4445 		    DMU_READ_PREFETCH);
4446 		zfs_unmap_page(cur_pp, va);
4447 		if (err) {
4448 			/* On error, toss the entire kluster */
4449 			pvn_read_done(pp, B_ERROR);
4450 			/* convert checksum errors into IO errors */
4451 			if (err == ECKSUM)
4452 				err = EIO;
4453 			return (err);
4454 		}
4455 		cur_pp = cur_pp->p_next;
4456 	}
4457 
4458 	/*
4459 	 * Fill in the page list array from the kluster starting
4460 	 * from the desired offset `off'.
4461 	 * NOTE: the page list will always be null terminated.
4462 	 */
4463 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4464 	ASSERT(pl == NULL || (*pl)->p_offset == off);
4465 
4466 	return (0);
4467 }
4468 
4469 /*
4470  * Return pointers to the pages for the file region [off, off + len]
4471  * in the pl array.  If plsz is greater than len, this function may
4472  * also return page pointers from after the specified region
4473  * (i.e. the region [off, off + plsz]).  These additional pages are
4474  * only returned if they are already in the cache, or were created as
4475  * part of a klustered read.
4476  *
4477  *	IN:	vp	- vnode of file to get data from.
4478  *		off	- position in file to get data from.
4479  *		len	- amount of data to retrieve.
4480  *		plsz	- length of provided page list.
4481  *		seg	- segment to obtain pages for.
4482  *		addr	- virtual address of fault.
4483  *		rw	- mode of created pages.
4484  *		cr	- credentials of caller.
4485  *		ct	- caller context.
4486  *
4487  *	OUT:	protp	- protection mode of created pages.
4488  *		pl	- list of pages created.
4489  *
4490  *	RETURN:	0 if success
4491  *		error code if failure
4492  *
4493  * Timestamps:
4494  *	vp - atime updated
4495  */
4496 /* ARGSUSED */
4497 static int
4498 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4499 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4500 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4501 {
4502 	znode_t		*zp = VTOZ(vp);
4503 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4504 	page_t		**pl0 = pl;
4505 	int		err = 0;
4506 
4507 	/* we do our own caching, faultahead is unnecessary */
4508 	if (pl == NULL)
4509 		return (0);
4510 	else if (len > plsz)
4511 		len = plsz;
4512 	else
4513 		len = P2ROUNDUP(len, PAGESIZE);
4514 	ASSERT(plsz >= len);
4515 
4516 	ZFS_ENTER(zfsvfs);
4517 	ZFS_VERIFY_ZP(zp);
4518 
4519 	if (protp)
4520 		*protp = PROT_ALL;
4521 
4522 	/*
4523 	 * Loop through the requested range [off, off + len) looking
4524 	 * for pages.  If we don't find a page, we will need to create
4525 	 * a new page and fill it with data from the file.
4526 	 */
4527 	while (len > 0) {
4528 		if (*pl = page_lookup(vp, off, SE_SHARED))
4529 			*(pl+1) = NULL;
4530 		else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4531 			goto out;
4532 		while (*pl) {
4533 			ASSERT3U((*pl)->p_offset, ==, off);
4534 			off += PAGESIZE;
4535 			addr += PAGESIZE;
4536 			if (len > 0) {
4537 				ASSERT3U(len, >=, PAGESIZE);
4538 				len -= PAGESIZE;
4539 			}
4540 			ASSERT3U(plsz, >=, PAGESIZE);
4541 			plsz -= PAGESIZE;
4542 			pl++;
4543 		}
4544 	}
4545 
4546 	/*
4547 	 * Fill out the page array with any pages already in the cache.
4548 	 */
4549 	while (plsz > 0 &&
4550 	    (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4551 			off += PAGESIZE;
4552 			plsz -= PAGESIZE;
4553 	}
4554 out:
4555 	if (err) {
4556 		/*
4557 		 * Release any pages we have previously locked.
4558 		 */
4559 		while (pl > pl0)
4560 			page_unlock(*--pl);
4561 	} else {
4562 		ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4563 	}
4564 
4565 	*pl = NULL;
4566 
4567 	ZFS_EXIT(zfsvfs);
4568 	return (err);
4569 }
4570 
4571 /*
4572  * Request a memory map for a section of a file.  This code interacts
4573  * with common code and the VM system as follows:
4574  *
4575  *	common code calls mmap(), which ends up in smmap_common()
4576  *
4577  *	this calls VOP_MAP(), which takes you into (say) zfs
4578  *
4579  *	zfs_map() calls as_map(), passing segvn_create() as the callback
4580  *
4581  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
4582  *
4583  *	zfs_addmap() updates z_mapcnt
4584  */
4585 /*ARGSUSED*/
4586 static int
4587 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4588     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4589     caller_context_t *ct)
4590 {
4591 	znode_t *zp = VTOZ(vp);
4592 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4593 	segvn_crargs_t	vn_a;
4594 	int		error;
4595 
4596 	ZFS_ENTER(zfsvfs);
4597 	ZFS_VERIFY_ZP(zp);
4598 
4599 	if ((prot & PROT_WRITE) && (zp->z_pflags &
4600 	    (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4601 		ZFS_EXIT(zfsvfs);
4602 		return (EPERM);
4603 	}
4604 
4605 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4606 	    (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4607 		ZFS_EXIT(zfsvfs);
4608 		return (EACCES);
4609 	}
4610 
4611 	if (vp->v_flag & VNOMAP) {
4612 		ZFS_EXIT(zfsvfs);
4613 		return (ENOSYS);
4614 	}
4615 
4616 	if (off < 0 || len > MAXOFFSET_T - off) {
4617 		ZFS_EXIT(zfsvfs);
4618 		return (ENXIO);
4619 	}
4620 
4621 	if (vp->v_type != VREG) {
4622 		ZFS_EXIT(zfsvfs);
4623 		return (ENODEV);
4624 	}
4625 
4626 	/*
4627 	 * If file is locked, disallow mapping.
4628 	 */
4629 	if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4630 		ZFS_EXIT(zfsvfs);
4631 		return (EAGAIN);
4632 	}
4633 
4634 	as_rangelock(as);
4635 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4636 	if (error != 0) {
4637 		as_rangeunlock(as);
4638 		ZFS_EXIT(zfsvfs);
4639 		return (error);
4640 	}
4641 
4642 	vn_a.vp = vp;
4643 	vn_a.offset = (u_offset_t)off;
4644 	vn_a.type = flags & MAP_TYPE;
4645 	vn_a.prot = prot;
4646 	vn_a.maxprot = maxprot;
4647 	vn_a.cred = cr;
4648 	vn_a.amp = NULL;
4649 	vn_a.flags = flags & ~MAP_TYPE;
4650 	vn_a.szc = 0;
4651 	vn_a.lgrp_mem_policy_flags = 0;
4652 
4653 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4654 
4655 	as_rangeunlock(as);
4656 	ZFS_EXIT(zfsvfs);
4657 	return (error);
4658 }
4659 
4660 /* ARGSUSED */
4661 static int
4662 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4663     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4664     caller_context_t *ct)
4665 {
4666 	uint64_t pages = btopr(len);
4667 
4668 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4669 	return (0);
4670 }
4671 
4672 /*
4673  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4674  * more accurate mtime for the associated file.  Since we don't have a way of
4675  * detecting when the data was actually modified, we have to resort to
4676  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4677  * last page is pushed.  The problem occurs when the msync() call is omitted,
4678  * which by far the most common case:
4679  *
4680  * 	open()
4681  * 	mmap()
4682  * 	<modify memory>
4683  * 	munmap()
4684  * 	close()
4685  * 	<time lapse>
4686  * 	putpage() via fsflush
4687  *
4688  * If we wait until fsflush to come along, we can have a modification time that
4689  * is some arbitrary point in the future.  In order to prevent this in the
4690  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4691  * torn down.
4692  */
4693 /* ARGSUSED */
4694 static int
4695 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4696     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4697     caller_context_t *ct)
4698 {
4699 	uint64_t pages = btopr(len);
4700 
4701 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4702 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4703 
4704 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4705 	    vn_has_cached_data(vp))
4706 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4707 
4708 	return (0);
4709 }
4710 
4711 /*
4712  * Free or allocate space in a file.  Currently, this function only
4713  * supports the `F_FREESP' command.  However, this command is somewhat
4714  * misnamed, as its functionality includes the ability to allocate as
4715  * well as free space.
4716  *
4717  *	IN:	vp	- vnode of file to free data in.
4718  *		cmd	- action to take (only F_FREESP supported).
4719  *		bfp	- section of file to free/alloc.
4720  *		flag	- current file open mode flags.
4721  *		offset	- current file offset.
4722  *		cr	- credentials of caller [UNUSED].
4723  *		ct	- caller context.
4724  *
4725  *	RETURN:	0 if success
4726  *		error code if failure
4727  *
4728  * Timestamps:
4729  *	vp - ctime|mtime updated
4730  */
4731 /* ARGSUSED */
4732 static int
4733 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4734     offset_t offset, cred_t *cr, caller_context_t *ct)
4735 {
4736 	znode_t		*zp = VTOZ(vp);
4737 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4738 	uint64_t	off, len;
4739 	int		error;
4740 
4741 	ZFS_ENTER(zfsvfs);
4742 	ZFS_VERIFY_ZP(zp);
4743 
4744 	if (cmd != F_FREESP) {
4745 		ZFS_EXIT(zfsvfs);
4746 		return (EINVAL);
4747 	}
4748 
4749 	if (error = convoff(vp, bfp, 0, offset)) {
4750 		ZFS_EXIT(zfsvfs);
4751 		return (error);
4752 	}
4753 
4754 	if (bfp->l_len < 0) {
4755 		ZFS_EXIT(zfsvfs);
4756 		return (EINVAL);
4757 	}
4758 
4759 	off = bfp->l_start;
4760 	len = bfp->l_len; /* 0 means from off to end of file */
4761 
4762 	error = zfs_freesp(zp, off, len, flag, TRUE);
4763 
4764 	ZFS_EXIT(zfsvfs);
4765 	return (error);
4766 }
4767 
4768 /*ARGSUSED*/
4769 static int
4770 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4771 {
4772 	znode_t		*zp = VTOZ(vp);
4773 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4774 	uint32_t	gen;
4775 	uint64_t	gen64;
4776 	uint64_t	object = zp->z_id;
4777 	zfid_short_t	*zfid;
4778 	int		size, i, error;
4779 
4780 	ZFS_ENTER(zfsvfs);
4781 	ZFS_VERIFY_ZP(zp);
4782 
4783 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4784 	    &gen64, sizeof (uint64_t))) != 0) {
4785 		ZFS_EXIT(zfsvfs);
4786 		return (error);
4787 	}
4788 
4789 	gen = (uint32_t)gen64;
4790 
4791 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4792 	if (fidp->fid_len < size) {
4793 		fidp->fid_len = size;
4794 		ZFS_EXIT(zfsvfs);
4795 		return (ENOSPC);
4796 	}
4797 
4798 	zfid = (zfid_short_t *)fidp;
4799 
4800 	zfid->zf_len = size;
4801 
4802 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4803 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4804 
4805 	/* Must have a non-zero generation number to distinguish from .zfs */
4806 	if (gen == 0)
4807 		gen = 1;
4808 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4809 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4810 
4811 	if (size == LONG_FID_LEN) {
4812 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4813 		zfid_long_t	*zlfid;
4814 
4815 		zlfid = (zfid_long_t *)fidp;
4816 
4817 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4818 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4819 
4820 		/* XXX - this should be the generation number for the objset */
4821 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4822 			zlfid->zf_setgen[i] = 0;
4823 	}
4824 
4825 	ZFS_EXIT(zfsvfs);
4826 	return (0);
4827 }
4828 
4829 static int
4830 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4831     caller_context_t *ct)
4832 {
4833 	znode_t		*zp, *xzp;
4834 	zfsvfs_t	*zfsvfs;
4835 	zfs_dirlock_t	*dl;
4836 	int		error;
4837 
4838 	switch (cmd) {
4839 	case _PC_LINK_MAX:
4840 		*valp = ULONG_MAX;
4841 		return (0);
4842 
4843 	case _PC_FILESIZEBITS:
4844 		*valp = 64;
4845 		return (0);
4846 
4847 	case _PC_XATTR_EXISTS:
4848 		zp = VTOZ(vp);
4849 		zfsvfs = zp->z_zfsvfs;
4850 		ZFS_ENTER(zfsvfs);
4851 		ZFS_VERIFY_ZP(zp);
4852 		*valp = 0;
4853 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4854 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4855 		if (error == 0) {
4856 			zfs_dirent_unlock(dl);
4857 			if (!zfs_dirempty(xzp))
4858 				*valp = 1;
4859 			VN_RELE(ZTOV(xzp));
4860 		} else if (error == ENOENT) {
4861 			/*
4862 			 * If there aren't extended attributes, it's the
4863 			 * same as having zero of them.
4864 			 */
4865 			error = 0;
4866 		}
4867 		ZFS_EXIT(zfsvfs);
4868 		return (error);
4869 
4870 	case _PC_SATTR_ENABLED:
4871 	case _PC_SATTR_EXISTS:
4872 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4873 		    (vp->v_type == VREG || vp->v_type == VDIR);
4874 		return (0);
4875 
4876 	case _PC_ACCESS_FILTERING:
4877 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4878 		    vp->v_type == VDIR;
4879 		return (0);
4880 
4881 	case _PC_ACL_ENABLED:
4882 		*valp = _ACL_ACE_ENABLED;
4883 		return (0);
4884 
4885 	case _PC_MIN_HOLE_SIZE:
4886 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4887 		return (0);
4888 
4889 	case _PC_TIMESTAMP_RESOLUTION:
4890 		/* nanosecond timestamp resolution */
4891 		*valp = 1L;
4892 		return (0);
4893 
4894 	default:
4895 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4896 	}
4897 }
4898 
4899 /*ARGSUSED*/
4900 static int
4901 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4902     caller_context_t *ct)
4903 {
4904 	znode_t *zp = VTOZ(vp);
4905 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4906 	int error;
4907 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4908 
4909 	ZFS_ENTER(zfsvfs);
4910 	ZFS_VERIFY_ZP(zp);
4911 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4912 	ZFS_EXIT(zfsvfs);
4913 
4914 	return (error);
4915 }
4916 
4917 /*ARGSUSED*/
4918 static int
4919 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4920     caller_context_t *ct)
4921 {
4922 	znode_t *zp = VTOZ(vp);
4923 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4924 	int error;
4925 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4926 	zilog_t	*zilog = zfsvfs->z_log;
4927 
4928 	ZFS_ENTER(zfsvfs);
4929 	ZFS_VERIFY_ZP(zp);
4930 
4931 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4932 
4933 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4934 		zil_commit(zilog, 0);
4935 
4936 	ZFS_EXIT(zfsvfs);
4937 	return (error);
4938 }
4939 
4940 /*
4941  * Tunable, both must be a power of 2.
4942  *
4943  * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4944  * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4945  *                an arcbuf for a partial block read
4946  */
4947 int zcr_blksz_min = (1 << 10);	/* 1K */
4948 int zcr_blksz_max = (1 << 17);	/* 128K */
4949 
4950 /*ARGSUSED*/
4951 static int
4952 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
4953     caller_context_t *ct)
4954 {
4955 	znode_t	*zp = VTOZ(vp);
4956 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4957 	int max_blksz = zfsvfs->z_max_blksz;
4958 	uio_t *uio = &xuio->xu_uio;
4959 	ssize_t size = uio->uio_resid;
4960 	offset_t offset = uio->uio_loffset;
4961 	int blksz;
4962 	int fullblk, i;
4963 	arc_buf_t *abuf;
4964 	ssize_t maxsize;
4965 	int preamble, postamble;
4966 
4967 	if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4968 		return (EINVAL);
4969 
4970 	ZFS_ENTER(zfsvfs);
4971 	ZFS_VERIFY_ZP(zp);
4972 	switch (ioflag) {
4973 	case UIO_WRITE:
4974 		/*
4975 		 * Loan out an arc_buf for write if write size is bigger than
4976 		 * max_blksz, and the file's block size is also max_blksz.
4977 		 */
4978 		blksz = max_blksz;
4979 		if (size < blksz || zp->z_blksz != blksz) {
4980 			ZFS_EXIT(zfsvfs);
4981 			return (EINVAL);
4982 		}
4983 		/*
4984 		 * Caller requests buffers for write before knowing where the
4985 		 * write offset might be (e.g. NFS TCP write).
4986 		 */
4987 		if (offset == -1) {
4988 			preamble = 0;
4989 		} else {
4990 			preamble = P2PHASE(offset, blksz);
4991 			if (preamble) {
4992 				preamble = blksz - preamble;
4993 				size -= preamble;
4994 			}
4995 		}
4996 
4997 		postamble = P2PHASE(size, blksz);
4998 		size -= postamble;
4999 
5000 		fullblk = size / blksz;
5001 		(void) dmu_xuio_init(xuio,
5002 		    (preamble != 0) + fullblk + (postamble != 0));
5003 		DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5004 		    int, postamble, int,
5005 		    (preamble != 0) + fullblk + (postamble != 0));
5006 
5007 		/*
5008 		 * Have to fix iov base/len for partial buffers.  They
5009 		 * currently represent full arc_buf's.
5010 		 */
5011 		if (preamble) {
5012 			/* data begins in the middle of the arc_buf */
5013 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5014 			    blksz);
5015 			ASSERT(abuf);
5016 			(void) dmu_xuio_add(xuio, abuf,
5017 			    blksz - preamble, preamble);
5018 		}
5019 
5020 		for (i = 0; i < fullblk; i++) {
5021 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5022 			    blksz);
5023 			ASSERT(abuf);
5024 			(void) dmu_xuio_add(xuio, abuf, 0, blksz);
5025 		}
5026 
5027 		if (postamble) {
5028 			/* data ends in the middle of the arc_buf */
5029 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5030 			    blksz);
5031 			ASSERT(abuf);
5032 			(void) dmu_xuio_add(xuio, abuf, 0, postamble);
5033 		}
5034 		break;
5035 	case UIO_READ:
5036 		/*
5037 		 * Loan out an arc_buf for read if the read size is larger than
5038 		 * the current file block size.  Block alignment is not
5039 		 * considered.  Partial arc_buf will be loaned out for read.
5040 		 */
5041 		blksz = zp->z_blksz;
5042 		if (blksz < zcr_blksz_min)
5043 			blksz = zcr_blksz_min;
5044 		if (blksz > zcr_blksz_max)
5045 			blksz = zcr_blksz_max;
5046 		/* avoid potential complexity of dealing with it */
5047 		if (blksz > max_blksz) {
5048 			ZFS_EXIT(zfsvfs);
5049 			return (EINVAL);
5050 		}
5051 
5052 		maxsize = zp->z_size - uio->uio_loffset;
5053 		if (size > maxsize)
5054 			size = maxsize;
5055 
5056 		if (size < blksz || vn_has_cached_data(vp)) {
5057 			ZFS_EXIT(zfsvfs);
5058 			return (EINVAL);
5059 		}
5060 		break;
5061 	default:
5062 		ZFS_EXIT(zfsvfs);
5063 		return (EINVAL);
5064 	}
5065 
5066 	uio->uio_extflg = UIO_XUIO;
5067 	XUIO_XUZC_RW(xuio) = ioflag;
5068 	ZFS_EXIT(zfsvfs);
5069 	return (0);
5070 }
5071 
5072 /*ARGSUSED*/
5073 static int
5074 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5075 {
5076 	int i;
5077 	arc_buf_t *abuf;
5078 	int ioflag = XUIO_XUZC_RW(xuio);
5079 
5080 	ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5081 
5082 	i = dmu_xuio_cnt(xuio);
5083 	while (i-- > 0) {
5084 		abuf = dmu_xuio_arcbuf(xuio, i);
5085 		/*
5086 		 * if abuf == NULL, it must be a write buffer
5087 		 * that has been returned in zfs_write().
5088 		 */
5089 		if (abuf)
5090 			dmu_return_arcbuf(abuf);
5091 		ASSERT(abuf || ioflag == UIO_WRITE);
5092 	}
5093 
5094 	dmu_xuio_fini(xuio);
5095 	return (0);
5096 }
5097 
5098 /*
5099  * Predeclare these here so that the compiler assumes that
5100  * this is an "old style" function declaration that does
5101  * not include arguments => we won't get type mismatch errors
5102  * in the initializations that follow.
5103  */
5104 static int zfs_inval();
5105 static int zfs_isdir();
5106 
5107 static int
5108 zfs_inval()
5109 {
5110 	return (EINVAL);
5111 }
5112 
5113 static int
5114 zfs_isdir()
5115 {
5116 	return (EISDIR);
5117 }
5118 /*
5119  * Directory vnode operations template
5120  */
5121 vnodeops_t *zfs_dvnodeops;
5122 const fs_operation_def_t zfs_dvnodeops_template[] = {
5123 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5124 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5125 	VOPNAME_READ,		{ .error = zfs_isdir },
5126 	VOPNAME_WRITE,		{ .error = zfs_isdir },
5127 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5128 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5129 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5130 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5131 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5132 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5133 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5134 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5135 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5136 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
5137 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5138 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5139 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
5140 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5141 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5142 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5143 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5144 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5145 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5146 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5147 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
5148 	NULL,			NULL
5149 };
5150 
5151 /*
5152  * Regular file vnode operations template
5153  */
5154 vnodeops_t *zfs_fvnodeops;
5155 const fs_operation_def_t zfs_fvnodeops_template[] = {
5156 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5157 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5158 	VOPNAME_READ,		{ .vop_read = zfs_read },
5159 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
5160 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5161 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5162 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5163 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5164 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5165 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5166 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5167 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5168 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5169 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5170 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
5171 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
5172 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
5173 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
5174 	VOPNAME_MAP,		{ .vop_map = zfs_map },
5175 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
5176 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
5177 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5178 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5179 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5180 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5181 	VOPNAME_REQZCBUF, 	{ .vop_reqzcbuf = zfs_reqzcbuf },
5182 	VOPNAME_RETZCBUF, 	{ .vop_retzcbuf = zfs_retzcbuf },
5183 	NULL,			NULL
5184 };
5185 
5186 /*
5187  * Symbolic link vnode operations template
5188  */
5189 vnodeops_t *zfs_symvnodeops;
5190 const fs_operation_def_t zfs_symvnodeops_template[] = {
5191 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5192 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5193 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5194 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5195 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
5196 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5197 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5198 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5199 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5200 	NULL,			NULL
5201 };
5202 
5203 /*
5204  * special share hidden files vnode operations template
5205  */
5206 vnodeops_t *zfs_sharevnodeops;
5207 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5208 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5209 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5210 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5211 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5212 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5213 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5214 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5215 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5216 	NULL,			NULL
5217 };
5218 
5219 /*
5220  * Extended attribute directory vnode operations template
5221  *	This template is identical to the directory vnodes
5222  *	operation template except for restricted operations:
5223  *		VOP_MKDIR()
5224  *		VOP_SYMLINK()
5225  * Note that there are other restrictions embedded in:
5226  *	zfs_create()	- restrict type to VREG
5227  *	zfs_link()	- no links into/out of attribute space
5228  *	zfs_rename()	- no moves into/out of attribute space
5229  */
5230 vnodeops_t *zfs_xdvnodeops;
5231 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5232 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5233 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5234 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5235 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5236 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5237 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5238 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5239 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5240 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5241 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5242 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5243 	VOPNAME_MKDIR,		{ .error = zfs_inval },
5244 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5245 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5246 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
5247 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5248 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5249 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5250 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5251 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5252 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5253 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5254 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5255 	NULL,			NULL
5256 };
5257 
5258 /*
5259  * Error vnode operations template
5260  */
5261 vnodeops_t *zfs_evnodeops;
5262 const fs_operation_def_t zfs_evnodeops_template[] = {
5263 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5264 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5265 	NULL,			NULL
5266 };
5267