xref: /titanic_52/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 1ce1951135b81c803c8dcf2f3c756009b1b0170a)
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 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /* Portions Copyright 2007 Jeremy Teo */
27 
28 #pragma ident	"%Z%%M%	%I%	%E% SMI"
29 
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/time.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
36 #include <sys/vfs.h>
37 #include <sys/vfs_opreg.h>
38 #include <sys/vnode.h>
39 #include <sys/file.h>
40 #include <sys/stat.h>
41 #include <sys/kmem.h>
42 #include <sys/taskq.h>
43 #include <sys/uio.h>
44 #include <sys/vmsystm.h>
45 #include <sys/atomic.h>
46 #include <sys/vm.h>
47 #include <vm/seg_vn.h>
48 #include <vm/pvn.h>
49 #include <vm/as.h>
50 #include <sys/mman.h>
51 #include <sys/pathname.h>
52 #include <sys/cmn_err.h>
53 #include <sys/errno.h>
54 #include <sys/unistd.h>
55 #include <sys/zfs_dir.h>
56 #include <sys/zfs_acl.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/fs/zfs.h>
59 #include <sys/dmu.h>
60 #include <sys/spa.h>
61 #include <sys/txg.h>
62 #include <sys/dbuf.h>
63 #include <sys/zap.h>
64 #include <sys/dirent.h>
65 #include <sys/policy.h>
66 #include <sys/sunddi.h>
67 #include <sys/filio.h>
68 #include "fs/fs_subr.h"
69 #include <sys/zfs_ctldir.h>
70 #include <sys/zfs_fuid.h>
71 #include <sys/dnlc.h>
72 #include <sys/zfs_rlock.h>
73 #include <sys/extdirent.h>
74 #include <sys/kidmap.h>
75 #include <sys/cred_impl.h>
76 
77 /*
78  * Programming rules.
79  *
80  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
81  * properly lock its in-core state, create a DMU transaction, do the work,
82  * record this work in the intent log (ZIL), commit the DMU transaction,
83  * and wait for the intent log to commit if it is a synchronous operation.
84  * Moreover, the vnode ops must work in both normal and log replay context.
85  * The ordering of events is important to avoid deadlocks and references
86  * to freed memory.  The example below illustrates the following Big Rules:
87  *
88  *  (1) A check must be made in each zfs thread for a mounted file system.
89  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
90  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
91  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
92  *      can return EIO from the calling function.
93  *
94  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
95  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
96  *	First, if it's the last reference, the vnode/znode
97  *	can be freed, so the zp may point to freed memory.  Second, the last
98  *	reference will call zfs_zinactive(), which may induce a lot of work --
99  *	pushing cached pages (which acquires range locks) and syncing out
100  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
101  *	which could deadlock the system if you were already holding one.
102  *
103  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
104  *	as they can span dmu_tx_assign() calls.
105  *
106  *  (4)	Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign().
107  *	In normal operation, this will be TXG_NOWAIT.  During ZIL replay,
108  *	it will be a specific txg.  Either way, dmu_tx_assign() never blocks.
109  *	This is critical because we don't want to block while holding locks.
110  *	Note, in particular, that if a lock is sometimes acquired before
111  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
112  *	use a non-blocking assign can deadlock the system.  The scenario:
113  *
114  *	Thread A has grabbed a lock before calling dmu_tx_assign().
115  *	Thread B is in an already-assigned tx, and blocks for this lock.
116  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
117  *	forever, because the previous txg can't quiesce until B's tx commits.
118  *
119  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
120  *	then drop all locks, call dmu_tx_wait(), and try again.
121  *
122  *  (5)	If the operation succeeded, generate the intent log entry for it
123  *	before dropping locks.  This ensures that the ordering of events
124  *	in the intent log matches the order in which they actually occurred.
125  *
126  *  (6)	At the end of each vnode op, the DMU tx must always commit,
127  *	regardless of whether there were any errors.
128  *
129  *  (7)	After dropping all locks, invoke zil_commit(zilog, seq, foid)
130  *	to ensure that synchronous semantics are provided when necessary.
131  *
132  * In general, this is how things should be ordered in each vnode op:
133  *
134  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
135  * top:
136  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
137  *	rw_enter(...);			// grab any other locks you need
138  *	tx = dmu_tx_create(...);	// get DMU tx
139  *	dmu_tx_hold_*();		// hold each object you might modify
140  *	error = dmu_tx_assign(tx, zfsvfs->z_assign);	// try to assign
141  *	if (error) {
142  *		rw_exit(...);		// drop locks
143  *		zfs_dirent_unlock(dl);	// unlock directory entry
144  *		VN_RELE(...);		// release held vnodes
145  *		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
146  *			dmu_tx_wait(tx);
147  *			dmu_tx_abort(tx);
148  *			goto top;
149  *		}
150  *		dmu_tx_abort(tx);	// abort DMU tx
151  *		ZFS_EXIT(zfsvfs);	// finished in zfs
152  *		return (error);		// really out of space
153  *	}
154  *	error = do_real_work();		// do whatever this VOP does
155  *	if (error == 0)
156  *		zfs_log_*(...);		// on success, make ZIL entry
157  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
158  *	rw_exit(...);			// drop locks
159  *	zfs_dirent_unlock(dl);		// unlock directory entry
160  *	VN_RELE(...);			// release held vnodes
161  *	zil_commit(zilog, seq, foid);	// synchronous when necessary
162  *	ZFS_EXIT(zfsvfs);		// finished in zfs
163  *	return (error);			// done, report error
164  */
165 
166 /* ARGSUSED */
167 static int
168 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
169 {
170 	znode_t	*zp = VTOZ(*vpp);
171 
172 	if ((flag & FWRITE) && (zp->z_phys->zp_flags & ZFS_APPENDONLY) &&
173 	    ((flag & FAPPEND) == 0)) {
174 		return (EPERM);
175 	}
176 
177 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
178 	    ZTOV(zp)->v_type == VREG &&
179 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
180 	    zp->z_phys->zp_size > 0)
181 		if (fs_vscan(*vpp, cr, 0) != 0)
182 			return (EACCES);
183 
184 	/* Keep a count of the synchronous opens in the znode */
185 	if (flag & (FSYNC | FDSYNC))
186 		atomic_inc_32(&zp->z_sync_cnt);
187 
188 	return (0);
189 }
190 
191 /* ARGSUSED */
192 static int
193 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
194     caller_context_t *ct)
195 {
196 	znode_t	*zp = VTOZ(vp);
197 
198 	/* Decrement the synchronous opens in the znode */
199 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
200 		atomic_dec_32(&zp->z_sync_cnt);
201 
202 	/*
203 	 * Clean up any locks held by this process on the vp.
204 	 */
205 	cleanlocks(vp, ddi_get_pid(), 0);
206 	cleanshares(vp, ddi_get_pid());
207 
208 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
209 	    ZTOV(zp)->v_type == VREG &&
210 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
211 	    zp->z_phys->zp_size > 0)
212 		VERIFY(fs_vscan(vp, cr, 1) == 0);
213 
214 	return (0);
215 }
216 
217 /*
218  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
219  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
220  */
221 static int
222 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
223 {
224 	znode_t	*zp = VTOZ(vp);
225 	uint64_t noff = (uint64_t)*off; /* new offset */
226 	uint64_t file_sz;
227 	int error;
228 	boolean_t hole;
229 
230 	file_sz = zp->z_phys->zp_size;
231 	if (noff >= file_sz)  {
232 		return (ENXIO);
233 	}
234 
235 	if (cmd == _FIO_SEEK_HOLE)
236 		hole = B_TRUE;
237 	else
238 		hole = B_FALSE;
239 
240 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
241 
242 	/* end of file? */
243 	if ((error == ESRCH) || (noff > file_sz)) {
244 		/*
245 		 * Handle the virtual hole at the end of file.
246 		 */
247 		if (hole) {
248 			*off = file_sz;
249 			return (0);
250 		}
251 		return (ENXIO);
252 	}
253 
254 	if (noff < *off)
255 		return (error);
256 	*off = noff;
257 	return (error);
258 }
259 
260 /* ARGSUSED */
261 static int
262 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
263     int *rvalp, caller_context_t *ct)
264 {
265 	offset_t off;
266 	int error;
267 	zfsvfs_t *zfsvfs;
268 	znode_t *zp;
269 
270 	switch (com) {
271 	case _FIOFFS:
272 		return (zfs_sync(vp->v_vfsp, 0, cred));
273 
274 		/*
275 		 * The following two ioctls are used by bfu.  Faking out,
276 		 * necessary to avoid bfu errors.
277 		 */
278 	case _FIOGDIO:
279 	case _FIOSDIO:
280 		return (0);
281 
282 	case _FIO_SEEK_DATA:
283 	case _FIO_SEEK_HOLE:
284 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
285 			return (EFAULT);
286 
287 		zp = VTOZ(vp);
288 		zfsvfs = zp->z_zfsvfs;
289 		ZFS_ENTER(zfsvfs);
290 		ZFS_VERIFY_ZP(zp);
291 
292 		/* offset parameter is in/out */
293 		error = zfs_holey(vp, com, &off);
294 		ZFS_EXIT(zfsvfs);
295 		if (error)
296 			return (error);
297 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
298 			return (EFAULT);
299 		return (0);
300 	}
301 	return (ENOTTY);
302 }
303 
304 /*
305  * When a file is memory mapped, we must keep the IO data synchronized
306  * between the DMU cache and the memory mapped pages.  What this means:
307  *
308  * On Write:	If we find a memory mapped page, we write to *both*
309  *		the page and the dmu buffer.
310  *
311  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
312  *	the file is memory mapped.
313  */
314 static int
315 mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx)
316 {
317 	znode_t	*zp = VTOZ(vp);
318 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
319 	int64_t	start, off;
320 	int len = nbytes;
321 	int error = 0;
322 
323 	start = uio->uio_loffset;
324 	off = start & PAGEOFFSET;
325 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
326 		page_t *pp;
327 		uint64_t bytes = MIN(PAGESIZE - off, len);
328 		uint64_t woff = uio->uio_loffset;
329 
330 		/*
331 		 * We don't want a new page to "appear" in the middle of
332 		 * the file update (because it may not get the write
333 		 * update data), so we grab a lock to block
334 		 * zfs_getpage().
335 		 */
336 		rw_enter(&zp->z_map_lock, RW_WRITER);
337 		if (pp = page_lookup(vp, start, SE_SHARED)) {
338 			caddr_t va;
339 
340 			rw_exit(&zp->z_map_lock);
341 			va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L);
342 			error = uiomove(va+off, bytes, UIO_WRITE, uio);
343 			if (error == 0) {
344 				dmu_write(zfsvfs->z_os, zp->z_id,
345 				    woff, bytes, va+off, tx);
346 			}
347 			ppmapout(va);
348 			page_unlock(pp);
349 		} else {
350 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
351 			    uio, bytes, tx);
352 			rw_exit(&zp->z_map_lock);
353 		}
354 		len -= bytes;
355 		off = 0;
356 		if (error)
357 			break;
358 	}
359 	return (error);
360 }
361 
362 /*
363  * When a file is memory mapped, we must keep the IO data synchronized
364  * between the DMU cache and the memory mapped pages.  What this means:
365  *
366  * On Read:	We "read" preferentially from memory mapped pages,
367  *		else we default from the dmu buffer.
368  *
369  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
370  *	the file is memory mapped.
371  */
372 static int
373 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
374 {
375 	znode_t *zp = VTOZ(vp);
376 	objset_t *os = zp->z_zfsvfs->z_os;
377 	int64_t	start, off;
378 	int len = nbytes;
379 	int error = 0;
380 
381 	start = uio->uio_loffset;
382 	off = start & PAGEOFFSET;
383 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
384 		page_t *pp;
385 		uint64_t bytes = MIN(PAGESIZE - off, len);
386 
387 		if (pp = page_lookup(vp, start, SE_SHARED)) {
388 			caddr_t va;
389 
390 			va = ppmapin(pp, PROT_READ, (caddr_t)-1L);
391 			error = uiomove(va + off, bytes, UIO_READ, uio);
392 			ppmapout(va);
393 			page_unlock(pp);
394 		} else {
395 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
396 		}
397 		len -= bytes;
398 		off = 0;
399 		if (error)
400 			break;
401 	}
402 	return (error);
403 }
404 
405 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
406 
407 /*
408  * Read bytes from specified file into supplied buffer.
409  *
410  *	IN:	vp	- vnode of file to be read from.
411  *		uio	- structure supplying read location, range info,
412  *			  and return buffer.
413  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
414  *		cr	- credentials of caller.
415  *		ct	- caller context
416  *
417  *	OUT:	uio	- updated offset and range, buffer filled.
418  *
419  *	RETURN:	0 if success
420  *		error code if failure
421  *
422  * Side Effects:
423  *	vp - atime updated if byte count > 0
424  */
425 /* ARGSUSED */
426 static int
427 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
428 {
429 	znode_t		*zp = VTOZ(vp);
430 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
431 	objset_t	*os;
432 	ssize_t		n, nbytes;
433 	int		error;
434 	rl_t		*rl;
435 
436 	ZFS_ENTER(zfsvfs);
437 	ZFS_VERIFY_ZP(zp);
438 	os = zfsvfs->z_os;
439 
440 	/*
441 	 * Validate file offset
442 	 */
443 	if (uio->uio_loffset < (offset_t)0) {
444 		ZFS_EXIT(zfsvfs);
445 		return (EINVAL);
446 	}
447 
448 	/*
449 	 * Fasttrack empty reads
450 	 */
451 	if (uio->uio_resid == 0) {
452 		ZFS_EXIT(zfsvfs);
453 		return (0);
454 	}
455 
456 	/*
457 	 * Check for mandatory locks
458 	 */
459 	if (MANDMODE((mode_t)zp->z_phys->zp_mode)) {
460 		if (error = chklock(vp, FREAD,
461 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
462 			ZFS_EXIT(zfsvfs);
463 			return (error);
464 		}
465 	}
466 
467 	/*
468 	 * If we're in FRSYNC mode, sync out this znode before reading it.
469 	 */
470 	if (ioflag & FRSYNC)
471 		zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
472 
473 	/*
474 	 * Lock the range against changes.
475 	 */
476 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
477 
478 	/*
479 	 * If we are reading past end-of-file we can skip
480 	 * to the end; but we might still need to set atime.
481 	 */
482 	if (uio->uio_loffset >= zp->z_phys->zp_size) {
483 		error = 0;
484 		goto out;
485 	}
486 
487 	ASSERT(uio->uio_loffset < zp->z_phys->zp_size);
488 	n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset);
489 
490 	while (n > 0) {
491 		nbytes = MIN(n, zfs_read_chunk_size -
492 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
493 
494 		if (vn_has_cached_data(vp))
495 			error = mappedread(vp, nbytes, uio);
496 		else
497 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
498 		if (error)
499 			break;
500 
501 		n -= nbytes;
502 	}
503 
504 out:
505 	zfs_range_unlock(rl);
506 
507 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
508 	ZFS_EXIT(zfsvfs);
509 	return (error);
510 }
511 
512 /*
513  * Fault in the pages of the first n bytes specified by the uio structure.
514  * 1 byte in each page is touched and the uio struct is unmodified.
515  * Any error will exit this routine as this is only a best
516  * attempt to get the pages resident. This is a copy of ufs_trans_touch().
517  */
518 static void
519 zfs_prefault_write(ssize_t n, struct uio *uio)
520 {
521 	struct iovec *iov;
522 	ulong_t cnt, incr;
523 	caddr_t p;
524 	uint8_t tmp;
525 
526 	iov = uio->uio_iov;
527 
528 	while (n) {
529 		cnt = MIN(iov->iov_len, n);
530 		if (cnt == 0) {
531 			/* empty iov entry */
532 			iov++;
533 			continue;
534 		}
535 		n -= cnt;
536 		/*
537 		 * touch each page in this segment.
538 		 */
539 		p = iov->iov_base;
540 		while (cnt) {
541 			switch (uio->uio_segflg) {
542 			case UIO_USERSPACE:
543 			case UIO_USERISPACE:
544 				if (fuword8(p, &tmp))
545 					return;
546 				break;
547 			case UIO_SYSSPACE:
548 				if (kcopy(p, &tmp, 1))
549 					return;
550 				break;
551 			}
552 			incr = MIN(cnt, PAGESIZE);
553 			p += incr;
554 			cnt -= incr;
555 		}
556 		/*
557 		 * touch the last byte in case it straddles a page.
558 		 */
559 		p--;
560 		switch (uio->uio_segflg) {
561 		case UIO_USERSPACE:
562 		case UIO_USERISPACE:
563 			if (fuword8(p, &tmp))
564 				return;
565 			break;
566 		case UIO_SYSSPACE:
567 			if (kcopy(p, &tmp, 1))
568 				return;
569 			break;
570 		}
571 		iov++;
572 	}
573 }
574 
575 /*
576  * Write the bytes to a file.
577  *
578  *	IN:	vp	- vnode of file to be written to.
579  *		uio	- structure supplying write location, range info,
580  *			  and data buffer.
581  *		ioflag	- FAPPEND flag set if in append mode.
582  *		cr	- credentials of caller.
583  *		ct	- caller context (NFS/CIFS fem monitor only)
584  *
585  *	OUT:	uio	- updated offset and range.
586  *
587  *	RETURN:	0 if success
588  *		error code if failure
589  *
590  * Timestamps:
591  *	vp - ctime|mtime updated if byte count > 0
592  */
593 /* ARGSUSED */
594 static int
595 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
596 {
597 	znode_t		*zp = VTOZ(vp);
598 	rlim64_t	limit = uio->uio_llimit;
599 	ssize_t		start_resid = uio->uio_resid;
600 	ssize_t		tx_bytes;
601 	uint64_t	end_size;
602 	dmu_tx_t	*tx;
603 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
604 	zilog_t		*zilog;
605 	offset_t	woff;
606 	ssize_t		n, nbytes;
607 	rl_t		*rl;
608 	int		max_blksz = zfsvfs->z_max_blksz;
609 	uint64_t	pflags = zp->z_phys->zp_flags;
610 	int		error;
611 
612 	/*
613 	 * If immutable or not appending then return EPERM
614 	 */
615 	if ((pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
616 	    ((pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
617 	    (uio->uio_loffset < zp->z_phys->zp_size)))
618 		return (EPERM);
619 
620 	/*
621 	 * Fasttrack empty write
622 	 */
623 	n = start_resid;
624 	if (n == 0)
625 		return (0);
626 
627 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
628 		limit = MAXOFFSET_T;
629 
630 	ZFS_ENTER(zfsvfs);
631 	ZFS_VERIFY_ZP(zp);
632 	zilog = zfsvfs->z_log;
633 
634 	/*
635 	 * Pre-fault the pages to ensure slow (eg NFS) pages
636 	 * don't hold up txg.
637 	 */
638 	zfs_prefault_write(n, uio);
639 
640 	/*
641 	 * If in append mode, set the io offset pointer to eof.
642 	 */
643 	if (ioflag & FAPPEND) {
644 		/*
645 		 * Range lock for a file append:
646 		 * The value for the start of range will be determined by
647 		 * zfs_range_lock() (to guarantee append semantics).
648 		 * If this write will cause the block size to increase,
649 		 * zfs_range_lock() will lock the entire file, so we must
650 		 * later reduce the range after we grow the block size.
651 		 */
652 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
653 		if (rl->r_len == UINT64_MAX) {
654 			/* overlocked, zp_size can't change */
655 			woff = uio->uio_loffset = zp->z_phys->zp_size;
656 		} else {
657 			woff = uio->uio_loffset = rl->r_off;
658 		}
659 	} else {
660 		woff = uio->uio_loffset;
661 		/*
662 		 * Validate file offset
663 		 */
664 		if (woff < 0) {
665 			ZFS_EXIT(zfsvfs);
666 			return (EINVAL);
667 		}
668 
669 		/*
670 		 * If we need to grow the block size then zfs_range_lock()
671 		 * will lock a wider range than we request here.
672 		 * Later after growing the block size we reduce the range.
673 		 */
674 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
675 	}
676 
677 	if (woff >= limit) {
678 		zfs_range_unlock(rl);
679 		ZFS_EXIT(zfsvfs);
680 		return (EFBIG);
681 	}
682 
683 	if ((woff + n) > limit || woff > (limit - n))
684 		n = limit - woff;
685 
686 	/*
687 	 * Check for mandatory locks
688 	 */
689 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) &&
690 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
691 		zfs_range_unlock(rl);
692 		ZFS_EXIT(zfsvfs);
693 		return (error);
694 	}
695 	end_size = MAX(zp->z_phys->zp_size, woff + n);
696 
697 	/*
698 	 * Write the file in reasonable size chunks.  Each chunk is written
699 	 * in a separate transaction; this keeps the intent log records small
700 	 * and allows us to do more fine-grained space accounting.
701 	 */
702 	while (n > 0) {
703 		/*
704 		 * Start a transaction.
705 		 */
706 		woff = uio->uio_loffset;
707 		tx = dmu_tx_create(zfsvfs->z_os);
708 		dmu_tx_hold_bonus(tx, zp->z_id);
709 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
710 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
711 		if (error) {
712 			if (error == ERESTART &&
713 			    zfsvfs->z_assign == TXG_NOWAIT) {
714 				dmu_tx_wait(tx);
715 				dmu_tx_abort(tx);
716 				continue;
717 			}
718 			dmu_tx_abort(tx);
719 			break;
720 		}
721 
722 		/*
723 		 * If zfs_range_lock() over-locked we grow the blocksize
724 		 * and then reduce the lock range.  This will only happen
725 		 * on the first iteration since zfs_range_reduce() will
726 		 * shrink down r_len to the appropriate size.
727 		 */
728 		if (rl->r_len == UINT64_MAX) {
729 			uint64_t new_blksz;
730 
731 			if (zp->z_blksz > max_blksz) {
732 				ASSERT(!ISP2(zp->z_blksz));
733 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
734 			} else {
735 				new_blksz = MIN(end_size, max_blksz);
736 			}
737 			zfs_grow_blocksize(zp, new_blksz, tx);
738 			zfs_range_reduce(rl, woff, n);
739 		}
740 
741 		/*
742 		 * XXX - should we really limit each write to z_max_blksz?
743 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
744 		 */
745 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
746 		rw_enter(&zp->z_map_lock, RW_READER);
747 
748 		tx_bytes = uio->uio_resid;
749 		if (vn_has_cached_data(vp)) {
750 			rw_exit(&zp->z_map_lock);
751 			error = mappedwrite(vp, nbytes, uio, tx);
752 		} else {
753 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
754 			    uio, nbytes, tx);
755 			rw_exit(&zp->z_map_lock);
756 		}
757 		tx_bytes -= uio->uio_resid;
758 
759 		/*
760 		 * If we made no progress, we're done.  If we made even
761 		 * partial progress, update the znode and ZIL accordingly.
762 		 */
763 		if (tx_bytes == 0) {
764 			dmu_tx_commit(tx);
765 			ASSERT(error != 0);
766 			break;
767 		}
768 
769 		/*
770 		 * Clear Set-UID/Set-GID bits on successful write if not
771 		 * privileged and at least one of the excute bits is set.
772 		 *
773 		 * It would be nice to to this after all writes have
774 		 * been done, but that would still expose the ISUID/ISGID
775 		 * to another app after the partial write is committed.
776 		 *
777 		 * Note: we don't call zfs_fuid_map_id() here because
778 		 * user 0 is not an ephemeral uid.
779 		 */
780 		mutex_enter(&zp->z_acl_lock);
781 		if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) |
782 		    (S_IXUSR >> 6))) != 0 &&
783 		    (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 &&
784 		    secpolicy_vnode_setid_retain(cr,
785 		    (zp->z_phys->zp_mode & S_ISUID) != 0 &&
786 		    zp->z_phys->zp_uid == 0) != 0) {
787 			zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID);
788 		}
789 		mutex_exit(&zp->z_acl_lock);
790 
791 		/*
792 		 * Update time stamp.  NOTE: This marks the bonus buffer as
793 		 * dirty, so we don't have to do it again for zp_size.
794 		 */
795 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
796 
797 		/*
798 		 * Update the file size (zp_size) if it has changed;
799 		 * account for possible concurrent updates.
800 		 */
801 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset)
802 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
803 			    uio->uio_loffset);
804 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
805 		dmu_tx_commit(tx);
806 
807 		if (error != 0)
808 			break;
809 		ASSERT(tx_bytes == nbytes);
810 		n -= nbytes;
811 	}
812 
813 	zfs_range_unlock(rl);
814 
815 	/*
816 	 * If we're in replay mode, or we made no progress, return error.
817 	 * Otherwise, it's at least a partial write, so it's successful.
818 	 */
819 	if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) {
820 		ZFS_EXIT(zfsvfs);
821 		return (error);
822 	}
823 
824 	if (ioflag & (FSYNC | FDSYNC))
825 		zil_commit(zilog, zp->z_last_itx, zp->z_id);
826 
827 	ZFS_EXIT(zfsvfs);
828 	return (0);
829 }
830 
831 void
832 zfs_get_done(dmu_buf_t *db, void *vzgd)
833 {
834 	zgd_t *zgd = (zgd_t *)vzgd;
835 	rl_t *rl = zgd->zgd_rl;
836 	vnode_t *vp = ZTOV(rl->r_zp);
837 
838 	dmu_buf_rele(db, vzgd);
839 	zfs_range_unlock(rl);
840 	VN_RELE(vp);
841 	zil_add_vdev(zgd->zgd_zilog, DVA_GET_VDEV(BP_IDENTITY(zgd->zgd_bp)));
842 	kmem_free(zgd, sizeof (zgd_t));
843 }
844 
845 /*
846  * Get data to generate a TX_WRITE intent log record.
847  */
848 int
849 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
850 {
851 	zfsvfs_t *zfsvfs = arg;
852 	objset_t *os = zfsvfs->z_os;
853 	znode_t *zp;
854 	uint64_t off = lr->lr_offset;
855 	dmu_buf_t *db;
856 	rl_t *rl;
857 	zgd_t *zgd;
858 	int dlen = lr->lr_length;		/* length of user data */
859 	int error = 0;
860 
861 	ASSERT(zio);
862 	ASSERT(dlen != 0);
863 
864 	/*
865 	 * Nothing to do if the file has been removed
866 	 */
867 	if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0)
868 		return (ENOENT);
869 	if (zp->z_unlinked) {
870 		VN_RELE(ZTOV(zp));
871 		return (ENOENT);
872 	}
873 
874 	/*
875 	 * Write records come in two flavors: immediate and indirect.
876 	 * For small writes it's cheaper to store the data with the
877 	 * log record (immediate); for large writes it's cheaper to
878 	 * sync the data and get a pointer to it (indirect) so that
879 	 * we don't have to write the data twice.
880 	 */
881 	if (buf != NULL) { /* immediate write */
882 		rl = zfs_range_lock(zp, off, dlen, RL_READER);
883 		/* test for truncation needs to be done while range locked */
884 		if (off >= zp->z_phys->zp_size) {
885 			error = ENOENT;
886 			goto out;
887 		}
888 		VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf));
889 	} else { /* indirect write */
890 		uint64_t boff; /* block starting offset */
891 
892 		/*
893 		 * Have to lock the whole block to ensure when it's
894 		 * written out and it's checksum is being calculated
895 		 * that no one can change the data. We need to re-check
896 		 * blocksize after we get the lock in case it's changed!
897 		 */
898 		for (;;) {
899 			if (ISP2(zp->z_blksz)) {
900 				boff = P2ALIGN_TYPED(off, zp->z_blksz,
901 				    uint64_t);
902 			} else {
903 				boff = 0;
904 			}
905 			dlen = zp->z_blksz;
906 			rl = zfs_range_lock(zp, boff, dlen, RL_READER);
907 			if (zp->z_blksz == dlen)
908 				break;
909 			zfs_range_unlock(rl);
910 		}
911 		/* test for truncation needs to be done while range locked */
912 		if (off >= zp->z_phys->zp_size) {
913 			error = ENOENT;
914 			goto out;
915 		}
916 		zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
917 		zgd->zgd_rl = rl;
918 		zgd->zgd_zilog = zfsvfs->z_log;
919 		zgd->zgd_bp = &lr->lr_blkptr;
920 		VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db));
921 		ASSERT(boff == db->db_offset);
922 		lr->lr_blkoff = off - boff;
923 		error = dmu_sync(zio, db, &lr->lr_blkptr,
924 		    lr->lr_common.lrc_txg, zfs_get_done, zgd);
925 		ASSERT((error && error != EINPROGRESS) ||
926 		    lr->lr_length <= zp->z_blksz);
927 		if (error == 0) {
928 			zil_add_vdev(zfsvfs->z_log,
929 			    DVA_GET_VDEV(BP_IDENTITY(&lr->lr_blkptr)));
930 		}
931 		/*
932 		 * If we get EINPROGRESS, then we need to wait for a
933 		 * write IO initiated by dmu_sync() to complete before
934 		 * we can release this dbuf.  We will finish everything
935 		 * up in the zfs_get_done() callback.
936 		 */
937 		if (error == EINPROGRESS)
938 			return (0);
939 		dmu_buf_rele(db, zgd);
940 		kmem_free(zgd, sizeof (zgd_t));
941 	}
942 out:
943 	zfs_range_unlock(rl);
944 	VN_RELE(ZTOV(zp));
945 	return (error);
946 }
947 
948 /*ARGSUSED*/
949 static int
950 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
951     caller_context_t *ct)
952 {
953 	znode_t *zp = VTOZ(vp);
954 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
955 	int error;
956 
957 	ZFS_ENTER(zfsvfs);
958 	ZFS_VERIFY_ZP(zp);
959 
960 	if (flag & V_ACE_MASK)
961 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
962 	else
963 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
964 
965 	ZFS_EXIT(zfsvfs);
966 	return (error);
967 }
968 
969 /*
970  * Lookup an entry in a directory, or an extended attribute directory.
971  * If it exists, return a held vnode reference for it.
972  *
973  *	IN:	dvp	- vnode of directory to search.
974  *		nm	- name of entry to lookup.
975  *		pnp	- full pathname to lookup [UNUSED].
976  *		flags	- LOOKUP_XATTR set if looking for an attribute.
977  *		rdir	- root directory vnode [UNUSED].
978  *		cr	- credentials of caller.
979  *		ct	- caller context
980  *		direntflags - directory lookup flags
981  *		realpnp - returned pathname.
982  *
983  *	OUT:	vpp	- vnode of located entry, NULL if not found.
984  *
985  *	RETURN:	0 if success
986  *		error code if failure
987  *
988  * Timestamps:
989  *	NA
990  */
991 /* ARGSUSED */
992 static int
993 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
994     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
995     int *direntflags, pathname_t *realpnp)
996 {
997 	znode_t *zdp = VTOZ(dvp);
998 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
999 	int	error;
1000 
1001 	ZFS_ENTER(zfsvfs);
1002 	ZFS_VERIFY_ZP(zdp);
1003 
1004 	*vpp = NULL;
1005 
1006 	if (flags & LOOKUP_XATTR) {
1007 		/*
1008 		 * If the xattr property is off, refuse the lookup request.
1009 		 */
1010 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1011 			ZFS_EXIT(zfsvfs);
1012 			return (EINVAL);
1013 		}
1014 
1015 		/*
1016 		 * We don't allow recursive attributes..
1017 		 * Maybe someday we will.
1018 		 */
1019 		if (zdp->z_phys->zp_flags & ZFS_XATTR) {
1020 			ZFS_EXIT(zfsvfs);
1021 			return (EINVAL);
1022 		}
1023 
1024 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1025 			ZFS_EXIT(zfsvfs);
1026 			return (error);
1027 		}
1028 
1029 		/*
1030 		 * Do we have permission to get into attribute directory?
1031 		 */
1032 
1033 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1034 		    B_FALSE, cr)) {
1035 			VN_RELE(*vpp);
1036 			*vpp = NULL;
1037 		}
1038 
1039 		ZFS_EXIT(zfsvfs);
1040 		return (error);
1041 	}
1042 
1043 	if (dvp->v_type != VDIR) {
1044 		ZFS_EXIT(zfsvfs);
1045 		return (ENOTDIR);
1046 	}
1047 
1048 	/*
1049 	 * Check accessibility of directory.
1050 	 */
1051 
1052 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1053 		ZFS_EXIT(zfsvfs);
1054 		return (error);
1055 	}
1056 
1057 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1058 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1059 		ZFS_EXIT(zfsvfs);
1060 		return (EILSEQ);
1061 	}
1062 
1063 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1064 	if (error == 0) {
1065 		/*
1066 		 * Convert device special files
1067 		 */
1068 		if (IS_DEVVP(*vpp)) {
1069 			vnode_t	*svp;
1070 
1071 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1072 			VN_RELE(*vpp);
1073 			if (svp == NULL)
1074 				error = ENOSYS;
1075 			else
1076 				*vpp = svp;
1077 		}
1078 	}
1079 
1080 	ZFS_EXIT(zfsvfs);
1081 	return (error);
1082 }
1083 
1084 /*
1085  * Attempt to create a new entry in a directory.  If the entry
1086  * already exists, truncate the file if permissible, else return
1087  * an error.  Return the vp of the created or trunc'd file.
1088  *
1089  *	IN:	dvp	- vnode of directory to put new file entry in.
1090  *		name	- name of new file entry.
1091  *		vap	- attributes of new file.
1092  *		excl	- flag indicating exclusive or non-exclusive mode.
1093  *		mode	- mode to open file with.
1094  *		cr	- credentials of caller.
1095  *		flag	- large file flag [UNUSED].
1096  *		ct	- caller context
1097  *		vsecp 	- ACL to be set
1098  *
1099  *	OUT:	vpp	- vnode of created or trunc'd entry.
1100  *
1101  *	RETURN:	0 if success
1102  *		error code if failure
1103  *
1104  * Timestamps:
1105  *	dvp - ctime|mtime updated if new entry created
1106  *	 vp - ctime|mtime always, atime if new
1107  */
1108 
1109 /* ARGSUSED */
1110 static int
1111 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1112     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1113     vsecattr_t *vsecp)
1114 {
1115 	znode_t		*zp, *dzp = VTOZ(dvp);
1116 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1117 	zilog_t		*zilog;
1118 	objset_t	*os;
1119 	zfs_dirlock_t	*dl;
1120 	dmu_tx_t	*tx;
1121 	int		error;
1122 	zfs_acl_t	*aclp = NULL;
1123 	zfs_fuid_info_t *fuidp = NULL;
1124 
1125 	/*
1126 	 * If we have an ephemeral id, ACL, or XVATTR then
1127 	 * make sure file system is at proper version
1128 	 */
1129 
1130 	if (zfsvfs->z_use_fuids == B_FALSE &&
1131 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1132 	    IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))))
1133 		return (EINVAL);
1134 
1135 	ZFS_ENTER(zfsvfs);
1136 	ZFS_VERIFY_ZP(dzp);
1137 	os = zfsvfs->z_os;
1138 	zilog = zfsvfs->z_log;
1139 
1140 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1141 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1142 		ZFS_EXIT(zfsvfs);
1143 		return (EILSEQ);
1144 	}
1145 
1146 	if (vap->va_mask & AT_XVATTR) {
1147 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1148 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1149 			ZFS_EXIT(zfsvfs);
1150 			return (error);
1151 		}
1152 	}
1153 top:
1154 	*vpp = NULL;
1155 
1156 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1157 		vap->va_mode &= ~VSVTX;
1158 
1159 	if (*name == '\0') {
1160 		/*
1161 		 * Null component name refers to the directory itself.
1162 		 */
1163 		VN_HOLD(dvp);
1164 		zp = dzp;
1165 		dl = NULL;
1166 		error = 0;
1167 	} else {
1168 		/* possible VN_HOLD(zp) */
1169 		int zflg = 0;
1170 
1171 		if (flag & FIGNORECASE)
1172 			zflg |= ZCILOOK;
1173 
1174 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1175 		    NULL, NULL);
1176 		if (error) {
1177 			if (strcmp(name, "..") == 0)
1178 				error = EISDIR;
1179 			ZFS_EXIT(zfsvfs);
1180 			if (aclp)
1181 				zfs_acl_free(aclp);
1182 			return (error);
1183 		}
1184 	}
1185 	if (vsecp && aclp == NULL) {
1186 		error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp);
1187 		if (error) {
1188 			ZFS_EXIT(zfsvfs);
1189 			if (dl)
1190 				zfs_dirent_unlock(dl);
1191 			return (error);
1192 		}
1193 	}
1194 
1195 	if (zp == NULL) {
1196 		uint64_t txtype;
1197 
1198 		/*
1199 		 * Create a new file object and update the directory
1200 		 * to reference it.
1201 		 */
1202 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1203 			goto out;
1204 		}
1205 
1206 		/*
1207 		 * We only support the creation of regular files in
1208 		 * extended attribute directories.
1209 		 */
1210 		if ((dzp->z_phys->zp_flags & ZFS_XATTR) &&
1211 		    (vap->va_type != VREG)) {
1212 			error = EINVAL;
1213 			goto out;
1214 		}
1215 
1216 		tx = dmu_tx_create(os);
1217 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1218 		if (zfsvfs->z_fuid_obj == 0) {
1219 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1220 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1221 			    SPA_MAXBLOCKSIZE);
1222 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
1223 		} else {
1224 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
1225 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
1226 			    SPA_MAXBLOCKSIZE);
1227 		}
1228 		dmu_tx_hold_bonus(tx, dzp->z_id);
1229 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1230 		if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) {
1231 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1232 			    0, SPA_MAXBLOCKSIZE);
1233 		}
1234 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
1235 		if (error) {
1236 			zfs_dirent_unlock(dl);
1237 			if (error == ERESTART &&
1238 			    zfsvfs->z_assign == TXG_NOWAIT) {
1239 				dmu_tx_wait(tx);
1240 				dmu_tx_abort(tx);
1241 				goto top;
1242 			}
1243 			dmu_tx_abort(tx);
1244 			ZFS_EXIT(zfsvfs);
1245 			if (aclp)
1246 				zfs_acl_free(aclp);
1247 			return (error);
1248 		}
1249 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp);
1250 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1251 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1252 		if (flag & FIGNORECASE)
1253 			txtype |= TX_CI;
1254 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1255 		    vsecp, fuidp, vap);
1256 		if (fuidp)
1257 			zfs_fuid_info_free(fuidp);
1258 		dmu_tx_commit(tx);
1259 	} else {
1260 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1261 
1262 		/*
1263 		 * A directory entry already exists for this name.
1264 		 */
1265 		/*
1266 		 * Can't truncate an existing file if in exclusive mode.
1267 		 */
1268 		if (excl == EXCL) {
1269 			error = EEXIST;
1270 			goto out;
1271 		}
1272 		/*
1273 		 * Can't open a directory for writing.
1274 		 */
1275 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1276 			error = EISDIR;
1277 			goto out;
1278 		}
1279 		/*
1280 		 * Verify requested access to file.
1281 		 */
1282 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1283 			goto out;
1284 		}
1285 
1286 		mutex_enter(&dzp->z_lock);
1287 		dzp->z_seq++;
1288 		mutex_exit(&dzp->z_lock);
1289 
1290 		/*
1291 		 * Truncate regular files if requested.
1292 		 */
1293 		if ((ZTOV(zp)->v_type == VREG) &&
1294 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1295 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1296 			if (error == ERESTART &&
1297 			    zfsvfs->z_assign == TXG_NOWAIT) {
1298 				/* NB: we already did dmu_tx_wait() */
1299 				zfs_dirent_unlock(dl);
1300 				VN_RELE(ZTOV(zp));
1301 				goto top;
1302 			}
1303 
1304 			if (error == 0) {
1305 				vnevent_create(ZTOV(zp), ct);
1306 			}
1307 		}
1308 	}
1309 out:
1310 
1311 	if (dl)
1312 		zfs_dirent_unlock(dl);
1313 
1314 	if (error) {
1315 		if (zp)
1316 			VN_RELE(ZTOV(zp));
1317 	} else {
1318 		*vpp = ZTOV(zp);
1319 		/*
1320 		 * If vnode is for a device return a specfs vnode instead.
1321 		 */
1322 		if (IS_DEVVP(*vpp)) {
1323 			struct vnode *svp;
1324 
1325 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1326 			VN_RELE(*vpp);
1327 			if (svp == NULL) {
1328 				error = ENOSYS;
1329 			}
1330 			*vpp = svp;
1331 		}
1332 	}
1333 	if (aclp)
1334 		zfs_acl_free(aclp);
1335 
1336 	ZFS_EXIT(zfsvfs);
1337 	return (error);
1338 }
1339 
1340 /*
1341  * Remove an entry from a directory.
1342  *
1343  *	IN:	dvp	- vnode of directory to remove entry from.
1344  *		name	- name of entry to remove.
1345  *		cr	- credentials of caller.
1346  *		ct	- caller context
1347  *		flags	- case flags
1348  *
1349  *	RETURN:	0 if success
1350  *		error code if failure
1351  *
1352  * Timestamps:
1353  *	dvp - ctime|mtime
1354  *	 vp - ctime (if nlink > 0)
1355  */
1356 /*ARGSUSED*/
1357 static int
1358 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1359     int flags)
1360 {
1361 	znode_t		*zp, *dzp = VTOZ(dvp);
1362 	znode_t		*xzp = NULL;
1363 	vnode_t		*vp;
1364 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1365 	zilog_t		*zilog;
1366 	uint64_t	acl_obj, xattr_obj;
1367 	zfs_dirlock_t	*dl;
1368 	dmu_tx_t	*tx;
1369 	boolean_t	may_delete_now, delete_now = FALSE;
1370 	boolean_t	unlinked;
1371 	uint64_t	txtype;
1372 	pathname_t	*realnmp = NULL;
1373 	pathname_t	realnm;
1374 	int		error;
1375 	int		zflg = ZEXISTS;
1376 
1377 	ZFS_ENTER(zfsvfs);
1378 	ZFS_VERIFY_ZP(dzp);
1379 	zilog = zfsvfs->z_log;
1380 
1381 	if (flags & FIGNORECASE) {
1382 		zflg |= ZCILOOK;
1383 		pn_alloc(&realnm);
1384 		realnmp = &realnm;
1385 	}
1386 
1387 top:
1388 	/*
1389 	 * Attempt to lock directory; fail if entry doesn't exist.
1390 	 */
1391 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1392 	    NULL, realnmp)) {
1393 		if (realnmp)
1394 			pn_free(realnmp);
1395 		ZFS_EXIT(zfsvfs);
1396 		return (error);
1397 	}
1398 
1399 	vp = ZTOV(zp);
1400 
1401 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1402 		goto out;
1403 	}
1404 
1405 	/*
1406 	 * Need to use rmdir for removing directories.
1407 	 */
1408 	if (vp->v_type == VDIR) {
1409 		error = EPERM;
1410 		goto out;
1411 	}
1412 
1413 	vnevent_remove(vp, dvp, name, ct);
1414 
1415 	if (realnmp)
1416 		dnlc_remove(dvp, realnmp->pn_path);
1417 	else
1418 		dnlc_remove(dvp, name);
1419 
1420 	mutex_enter(&vp->v_lock);
1421 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1422 	mutex_exit(&vp->v_lock);
1423 
1424 	/*
1425 	 * We may delete the znode now, or we may put it in the unlinked set;
1426 	 * it depends on whether we're the last link, and on whether there are
1427 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1428 	 * allow for either case.
1429 	 */
1430 	tx = dmu_tx_create(zfsvfs->z_os);
1431 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1432 	dmu_tx_hold_bonus(tx, zp->z_id);
1433 	if (may_delete_now)
1434 		dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
1435 
1436 	/* are there any extended attributes? */
1437 	if ((xattr_obj = zp->z_phys->zp_xattr) != 0) {
1438 		/* XXX - do we need this if we are deleting? */
1439 		dmu_tx_hold_bonus(tx, xattr_obj);
1440 	}
1441 
1442 	/* are there any additional acls */
1443 	if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 &&
1444 	    may_delete_now)
1445 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1446 
1447 	/* charge as an update -- would be nice not to charge at all */
1448 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1449 
1450 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1451 	if (error) {
1452 		zfs_dirent_unlock(dl);
1453 		VN_RELE(vp);
1454 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1455 			dmu_tx_wait(tx);
1456 			dmu_tx_abort(tx);
1457 			goto top;
1458 		}
1459 		if (realnmp)
1460 			pn_free(realnmp);
1461 		dmu_tx_abort(tx);
1462 		ZFS_EXIT(zfsvfs);
1463 		return (error);
1464 	}
1465 
1466 	/*
1467 	 * Remove the directory entry.
1468 	 */
1469 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1470 
1471 	if (error) {
1472 		dmu_tx_commit(tx);
1473 		goto out;
1474 	}
1475 
1476 	if (unlinked) {
1477 		mutex_enter(&vp->v_lock);
1478 		delete_now = may_delete_now &&
1479 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1480 		    zp->z_phys->zp_xattr == xattr_obj &&
1481 		    zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj;
1482 		mutex_exit(&vp->v_lock);
1483 	}
1484 
1485 	if (delete_now) {
1486 		if (zp->z_phys->zp_xattr) {
1487 			error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
1488 			ASSERT3U(error, ==, 0);
1489 			ASSERT3U(xzp->z_phys->zp_links, ==, 2);
1490 			dmu_buf_will_dirty(xzp->z_dbuf, tx);
1491 			mutex_enter(&xzp->z_lock);
1492 			xzp->z_unlinked = 1;
1493 			xzp->z_phys->zp_links = 0;
1494 			mutex_exit(&xzp->z_lock);
1495 			zfs_unlinked_add(xzp, tx);
1496 			zp->z_phys->zp_xattr = 0; /* probably unnecessary */
1497 		}
1498 		mutex_enter(&zp->z_lock);
1499 		mutex_enter(&vp->v_lock);
1500 		vp->v_count--;
1501 		ASSERT3U(vp->v_count, ==, 0);
1502 		mutex_exit(&vp->v_lock);
1503 		mutex_exit(&zp->z_lock);
1504 		zfs_znode_delete(zp, tx);
1505 	} else if (unlinked) {
1506 		zfs_unlinked_add(zp, tx);
1507 	}
1508 
1509 	txtype = TX_REMOVE;
1510 	if (flags & FIGNORECASE)
1511 		txtype |= TX_CI;
1512 	zfs_log_remove(zilog, tx, txtype, dzp, name);
1513 
1514 	dmu_tx_commit(tx);
1515 out:
1516 	if (realnmp)
1517 		pn_free(realnmp);
1518 
1519 	zfs_dirent_unlock(dl);
1520 
1521 	if (!delete_now) {
1522 		VN_RELE(vp);
1523 	} else if (xzp) {
1524 		/* this rele delayed to prevent nesting transactions */
1525 		VN_RELE(ZTOV(xzp));
1526 	}
1527 
1528 	ZFS_EXIT(zfsvfs);
1529 	return (error);
1530 }
1531 
1532 /*
1533  * Create a new directory and insert it into dvp using the name
1534  * provided.  Return a pointer to the inserted directory.
1535  *
1536  *	IN:	dvp	- vnode of directory to add subdir to.
1537  *		dirname	- name of new directory.
1538  *		vap	- attributes of new directory.
1539  *		cr	- credentials of caller.
1540  *		ct	- caller context
1541  *		vsecp	- ACL to be set
1542  *
1543  *	OUT:	vpp	- vnode of created directory.
1544  *
1545  *	RETURN:	0 if success
1546  *		error code if failure
1547  *
1548  * Timestamps:
1549  *	dvp - ctime|mtime updated
1550  *	 vp - ctime|mtime|atime updated
1551  */
1552 /*ARGSUSED*/
1553 static int
1554 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1555     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1556 {
1557 	znode_t		*zp, *dzp = VTOZ(dvp);
1558 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1559 	zilog_t		*zilog;
1560 	zfs_dirlock_t	*dl;
1561 	uint64_t	txtype;
1562 	dmu_tx_t	*tx;
1563 	int		error;
1564 	zfs_acl_t	*aclp = NULL;
1565 	zfs_fuid_info_t	*fuidp = NULL;
1566 	int		zf = ZNEW;
1567 
1568 	ASSERT(vap->va_type == VDIR);
1569 
1570 	/*
1571 	 * If we have an ephemeral id, ACL, or XVATTR then
1572 	 * make sure file system is at proper version
1573 	 */
1574 
1575 	if (zfsvfs->z_use_fuids == B_FALSE &&
1576 	    (vsecp || (vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(crgetuid(cr))||
1577 	    IS_EPHEMERAL(crgetgid(cr))))
1578 		return (EINVAL);
1579 
1580 	ZFS_ENTER(zfsvfs);
1581 	ZFS_VERIFY_ZP(dzp);
1582 	zilog = zfsvfs->z_log;
1583 
1584 	if (dzp->z_phys->zp_flags & ZFS_XATTR) {
1585 		ZFS_EXIT(zfsvfs);
1586 		return (EINVAL);
1587 	}
1588 
1589 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1590 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1591 		ZFS_EXIT(zfsvfs);
1592 		return (EILSEQ);
1593 	}
1594 	if (flags & FIGNORECASE)
1595 		zf |= ZCILOOK;
1596 
1597 	if (vap->va_mask & AT_XVATTR)
1598 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1599 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1600 			ZFS_EXIT(zfsvfs);
1601 			return (error);
1602 		}
1603 
1604 	/*
1605 	 * First make sure the new directory doesn't exist.
1606 	 */
1607 top:
1608 	*vpp = NULL;
1609 
1610 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1611 	    NULL, NULL)) {
1612 		ZFS_EXIT(zfsvfs);
1613 		return (error);
1614 	}
1615 
1616 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1617 		zfs_dirent_unlock(dl);
1618 		ZFS_EXIT(zfsvfs);
1619 		return (error);
1620 	}
1621 
1622 	if (vsecp && aclp == NULL) {
1623 		error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp);
1624 		if (error) {
1625 			zfs_dirent_unlock(dl);
1626 			ZFS_EXIT(zfsvfs);
1627 			return (error);
1628 		}
1629 	}
1630 	/*
1631 	 * Add a new entry to the directory.
1632 	 */
1633 	tx = dmu_tx_create(zfsvfs->z_os);
1634 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1635 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1636 	if (zfsvfs->z_fuid_obj == 0) {
1637 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1638 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1639 		    SPA_MAXBLOCKSIZE);
1640 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
1641 	} else {
1642 		dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
1643 		dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
1644 		    SPA_MAXBLOCKSIZE);
1645 	}
1646 	if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp)
1647 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1648 		    0, SPA_MAXBLOCKSIZE);
1649 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1650 	if (error) {
1651 		zfs_dirent_unlock(dl);
1652 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1653 			dmu_tx_wait(tx);
1654 			dmu_tx_abort(tx);
1655 			goto top;
1656 		}
1657 		dmu_tx_abort(tx);
1658 		ZFS_EXIT(zfsvfs);
1659 		if (aclp)
1660 			zfs_acl_free(aclp);
1661 		return (error);
1662 	}
1663 
1664 	/*
1665 	 * Create new node.
1666 	 */
1667 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp);
1668 
1669 	if (aclp)
1670 		zfs_acl_free(aclp);
1671 
1672 	/*
1673 	 * Now put new name in parent dir.
1674 	 */
1675 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1676 
1677 	*vpp = ZTOV(zp);
1678 
1679 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1680 	if (flags & FIGNORECASE)
1681 		txtype |= TX_CI;
1682 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, fuidp, vap);
1683 
1684 	if (fuidp)
1685 		zfs_fuid_info_free(fuidp);
1686 	dmu_tx_commit(tx);
1687 
1688 	zfs_dirent_unlock(dl);
1689 
1690 	ZFS_EXIT(zfsvfs);
1691 	return (0);
1692 }
1693 
1694 /*
1695  * Remove a directory subdir entry.  If the current working
1696  * directory is the same as the subdir to be removed, the
1697  * remove will fail.
1698  *
1699  *	IN:	dvp	- vnode of directory to remove from.
1700  *		name	- name of directory to be removed.
1701  *		cwd	- vnode of current working directory.
1702  *		cr	- credentials of caller.
1703  *		ct	- caller context
1704  *		flags	- case flags
1705  *
1706  *	RETURN:	0 if success
1707  *		error code if failure
1708  *
1709  * Timestamps:
1710  *	dvp - ctime|mtime updated
1711  */
1712 /*ARGSUSED*/
1713 static int
1714 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1715     caller_context_t *ct, int flags)
1716 {
1717 	znode_t		*dzp = VTOZ(dvp);
1718 	znode_t		*zp;
1719 	vnode_t		*vp;
1720 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1721 	zilog_t		*zilog;
1722 	zfs_dirlock_t	*dl;
1723 	dmu_tx_t	*tx;
1724 	int		error;
1725 	int		zflg = ZEXISTS;
1726 
1727 	ZFS_ENTER(zfsvfs);
1728 	ZFS_VERIFY_ZP(dzp);
1729 	zilog = zfsvfs->z_log;
1730 
1731 	if (flags & FIGNORECASE)
1732 		zflg |= ZCILOOK;
1733 top:
1734 	zp = NULL;
1735 
1736 	/*
1737 	 * Attempt to lock directory; fail if entry doesn't exist.
1738 	 */
1739 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1740 	    NULL, NULL)) {
1741 		ZFS_EXIT(zfsvfs);
1742 		return (error);
1743 	}
1744 
1745 	vp = ZTOV(zp);
1746 
1747 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1748 		goto out;
1749 	}
1750 
1751 	if (vp->v_type != VDIR) {
1752 		error = ENOTDIR;
1753 		goto out;
1754 	}
1755 
1756 	if (vp == cwd) {
1757 		error = EINVAL;
1758 		goto out;
1759 	}
1760 
1761 	vnevent_rmdir(vp, dvp, name, ct);
1762 
1763 	/*
1764 	 * Grab a lock on the directory to make sure that noone is
1765 	 * trying to add (or lookup) entries while we are removing it.
1766 	 */
1767 	rw_enter(&zp->z_name_lock, RW_WRITER);
1768 
1769 	/*
1770 	 * Grab a lock on the parent pointer to make sure we play well
1771 	 * with the treewalk and directory rename code.
1772 	 */
1773 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1774 
1775 	tx = dmu_tx_create(zfsvfs->z_os);
1776 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1777 	dmu_tx_hold_bonus(tx, zp->z_id);
1778 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1779 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1780 	if (error) {
1781 		rw_exit(&zp->z_parent_lock);
1782 		rw_exit(&zp->z_name_lock);
1783 		zfs_dirent_unlock(dl);
1784 		VN_RELE(vp);
1785 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1786 			dmu_tx_wait(tx);
1787 			dmu_tx_abort(tx);
1788 			goto top;
1789 		}
1790 		dmu_tx_abort(tx);
1791 		ZFS_EXIT(zfsvfs);
1792 		return (error);
1793 	}
1794 
1795 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
1796 
1797 	if (error == 0) {
1798 		uint64_t txtype = TX_RMDIR;
1799 		if (flags & FIGNORECASE)
1800 			txtype |= TX_CI;
1801 		zfs_log_remove(zilog, tx, txtype, dzp, name);
1802 	}
1803 
1804 	dmu_tx_commit(tx);
1805 
1806 	rw_exit(&zp->z_parent_lock);
1807 	rw_exit(&zp->z_name_lock);
1808 out:
1809 	zfs_dirent_unlock(dl);
1810 
1811 	VN_RELE(vp);
1812 
1813 	ZFS_EXIT(zfsvfs);
1814 	return (error);
1815 }
1816 
1817 /*
1818  * Read as many directory entries as will fit into the provided
1819  * buffer from the given directory cursor position (specified in
1820  * the uio structure.
1821  *
1822  *	IN:	vp	- vnode of directory to read.
1823  *		uio	- structure supplying read location, range info,
1824  *			  and return buffer.
1825  *		cr	- credentials of caller.
1826  *		ct	- caller context
1827  *		flags	- case flags
1828  *
1829  *	OUT:	uio	- updated offset and range, buffer filled.
1830  *		eofp	- set to true if end-of-file detected.
1831  *
1832  *	RETURN:	0 if success
1833  *		error code if failure
1834  *
1835  * Timestamps:
1836  *	vp - atime updated
1837  *
1838  * Note that the low 4 bits of the cookie returned by zap is always zero.
1839  * This allows us to use the low range for "special" directory entries:
1840  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1841  * we use the offset 2 for the '.zfs' directory.
1842  */
1843 /* ARGSUSED */
1844 static int
1845 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
1846     caller_context_t *ct, int flags)
1847 {
1848 	znode_t		*zp = VTOZ(vp);
1849 	iovec_t		*iovp;
1850 	edirent_t	*eodp;
1851 	dirent64_t	*odp;
1852 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1853 	objset_t	*os;
1854 	caddr_t		outbuf;
1855 	size_t		bufsize;
1856 	zap_cursor_t	zc;
1857 	zap_attribute_t	zap;
1858 	uint_t		bytes_wanted;
1859 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1860 	int		local_eof;
1861 	int		outcount;
1862 	int		error;
1863 	uint8_t		prefetch;
1864 
1865 	ZFS_ENTER(zfsvfs);
1866 	ZFS_VERIFY_ZP(zp);
1867 
1868 	/*
1869 	 * If we are not given an eof variable,
1870 	 * use a local one.
1871 	 */
1872 	if (eofp == NULL)
1873 		eofp = &local_eof;
1874 
1875 	/*
1876 	 * Check for valid iov_len.
1877 	 */
1878 	if (uio->uio_iov->iov_len <= 0) {
1879 		ZFS_EXIT(zfsvfs);
1880 		return (EINVAL);
1881 	}
1882 
1883 	/*
1884 	 * Quit if directory has been removed (posix)
1885 	 */
1886 	if ((*eofp = zp->z_unlinked) != 0) {
1887 		ZFS_EXIT(zfsvfs);
1888 		return (0);
1889 	}
1890 
1891 	error = 0;
1892 	os = zfsvfs->z_os;
1893 	offset = uio->uio_loffset;
1894 	prefetch = zp->z_zn_prefetch;
1895 
1896 	/*
1897 	 * Initialize the iterator cursor.
1898 	 */
1899 	if (offset <= 3) {
1900 		/*
1901 		 * Start iteration from the beginning of the directory.
1902 		 */
1903 		zap_cursor_init(&zc, os, zp->z_id);
1904 	} else {
1905 		/*
1906 		 * The offset is a serialized cursor.
1907 		 */
1908 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1909 	}
1910 
1911 	/*
1912 	 * Get space to change directory entries into fs independent format.
1913 	 */
1914 	iovp = uio->uio_iov;
1915 	bytes_wanted = iovp->iov_len;
1916 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
1917 		bufsize = bytes_wanted;
1918 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
1919 		odp = (struct dirent64 *)outbuf;
1920 	} else {
1921 		bufsize = bytes_wanted;
1922 		odp = (struct dirent64 *)iovp->iov_base;
1923 	}
1924 	eodp = (struct edirent *)odp;
1925 
1926 	/*
1927 	 * Transform to file-system independent format
1928 	 */
1929 	outcount = 0;
1930 	while (outcount < bytes_wanted) {
1931 		ino64_t objnum;
1932 		ushort_t reclen;
1933 		off64_t *next;
1934 
1935 		/*
1936 		 * Special case `.', `..', and `.zfs'.
1937 		 */
1938 		if (offset == 0) {
1939 			(void) strcpy(zap.za_name, ".");
1940 			zap.za_normalization_conflict = 0;
1941 			objnum = zp->z_id;
1942 		} else if (offset == 1) {
1943 			(void) strcpy(zap.za_name, "..");
1944 			zap.za_normalization_conflict = 0;
1945 			objnum = zp->z_phys->zp_parent;
1946 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
1947 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1948 			zap.za_normalization_conflict = 0;
1949 			objnum = ZFSCTL_INO_ROOT;
1950 		} else {
1951 			/*
1952 			 * Grab next entry.
1953 			 */
1954 			if (error = zap_cursor_retrieve(&zc, &zap)) {
1955 				if ((*eofp = (error == ENOENT)) != 0)
1956 					break;
1957 				else
1958 					goto update;
1959 			}
1960 
1961 			if (zap.za_integer_length != 8 ||
1962 			    zap.za_num_integers != 1) {
1963 				cmn_err(CE_WARN, "zap_readdir: bad directory "
1964 				    "entry, obj = %lld, offset = %lld\n",
1965 				    (u_longlong_t)zp->z_id,
1966 				    (u_longlong_t)offset);
1967 				error = ENXIO;
1968 				goto update;
1969 			}
1970 
1971 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
1972 			/*
1973 			 * MacOS X can extract the object type here such as:
1974 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
1975 			 */
1976 		}
1977 
1978 		if (flags & V_RDDIR_ENTFLAGS)
1979 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
1980 		else
1981 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
1982 
1983 		/*
1984 		 * Will this entry fit in the buffer?
1985 		 */
1986 		if (outcount + reclen > bufsize) {
1987 			/*
1988 			 * Did we manage to fit anything in the buffer?
1989 			 */
1990 			if (!outcount) {
1991 				error = EINVAL;
1992 				goto update;
1993 			}
1994 			break;
1995 		}
1996 		if (flags & V_RDDIR_ENTFLAGS) {
1997 			/*
1998 			 * Add extended flag entry:
1999 			 */
2000 			eodp->ed_ino = objnum;
2001 			eodp->ed_reclen = reclen;
2002 			/* NOTE: ed_off is the offset for the *next* entry */
2003 			next = &(eodp->ed_off);
2004 			eodp->ed_eflags = zap.za_normalization_conflict ?
2005 			    ED_CASE_CONFLICT : 0;
2006 			(void) strncpy(eodp->ed_name, zap.za_name,
2007 			    EDIRENT_NAMELEN(reclen));
2008 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2009 		} else {
2010 			/*
2011 			 * Add normal entry:
2012 			 */
2013 			odp->d_ino = objnum;
2014 			odp->d_reclen = reclen;
2015 			/* NOTE: d_off is the offset for the *next* entry */
2016 			next = &(odp->d_off);
2017 			(void) strncpy(odp->d_name, zap.za_name,
2018 			    DIRENT64_NAMELEN(reclen));
2019 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2020 		}
2021 		outcount += reclen;
2022 
2023 		ASSERT(outcount <= bufsize);
2024 
2025 		/* Prefetch znode */
2026 		if (prefetch)
2027 			dmu_prefetch(os, objnum, 0, 0);
2028 
2029 		/*
2030 		 * Move to the next entry, fill in the previous offset.
2031 		 */
2032 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2033 			zap_cursor_advance(&zc);
2034 			offset = zap_cursor_serialize(&zc);
2035 		} else {
2036 			offset += 1;
2037 		}
2038 		*next = offset;
2039 	}
2040 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2041 
2042 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2043 		iovp->iov_base += outcount;
2044 		iovp->iov_len -= outcount;
2045 		uio->uio_resid -= outcount;
2046 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2047 		/*
2048 		 * Reset the pointer.
2049 		 */
2050 		offset = uio->uio_loffset;
2051 	}
2052 
2053 update:
2054 	zap_cursor_fini(&zc);
2055 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2056 		kmem_free(outbuf, bufsize);
2057 
2058 	if (error == ENOENT)
2059 		error = 0;
2060 
2061 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2062 
2063 	uio->uio_loffset = offset;
2064 	ZFS_EXIT(zfsvfs);
2065 	return (error);
2066 }
2067 
2068 ulong_t zfs_fsync_sync_cnt = 4;
2069 
2070 static int
2071 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2072 {
2073 	znode_t	*zp = VTOZ(vp);
2074 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2075 
2076 	/*
2077 	 * Regardless of whether this is required for standards conformance,
2078 	 * this is the logical behavior when fsync() is called on a file with
2079 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2080 	 * going to be pushed out as part of the zil_commit().
2081 	 */
2082 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2083 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2084 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2085 
2086 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2087 
2088 	ZFS_ENTER(zfsvfs);
2089 	ZFS_VERIFY_ZP(zp);
2090 	zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
2091 	ZFS_EXIT(zfsvfs);
2092 	return (0);
2093 }
2094 
2095 
2096 /*
2097  * Get the requested file attributes and place them in the provided
2098  * vattr structure.
2099  *
2100  *	IN:	vp	- vnode of file.
2101  *		vap	- va_mask identifies requested attributes.
2102  *			  If AT_XVATTR set, then optional attrs are requested
2103  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2104  *		cr	- credentials of caller.
2105  *		ct	- caller context
2106  *
2107  *	OUT:	vap	- attribute values.
2108  *
2109  *	RETURN:	0 (always succeeds)
2110  */
2111 /* ARGSUSED */
2112 static int
2113 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2114     caller_context_t *ct)
2115 {
2116 	znode_t *zp = VTOZ(vp);
2117 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2118 	znode_phys_t *pzp;
2119 	int	error = 0;
2120 	uint64_t links;
2121 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2122 	xoptattr_t *xoap = NULL;
2123 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2124 
2125 	ZFS_ENTER(zfsvfs);
2126 	ZFS_VERIFY_ZP(zp);
2127 	pzp = zp->z_phys;
2128 
2129 	mutex_enter(&zp->z_lock);
2130 
2131 	/*
2132 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2133 	 * Also, if we are the owner don't bother, since owner should
2134 	 * always be allowed to read basic attributes of file.
2135 	 */
2136 	if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL) &&
2137 	    (pzp->zp_uid != crgetuid(cr))) {
2138 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2139 		    skipaclchk, cr)) {
2140 			mutex_exit(&zp->z_lock);
2141 			ZFS_EXIT(zfsvfs);
2142 			return (error);
2143 		}
2144 	}
2145 
2146 	/*
2147 	 * Return all attributes.  It's cheaper to provide the answer
2148 	 * than to determine whether we were asked the question.
2149 	 */
2150 
2151 	vap->va_type = vp->v_type;
2152 	vap->va_mode = pzp->zp_mode & MODEMASK;
2153 	zfs_fuid_map_ids(zp, &vap->va_uid, &vap->va_gid);
2154 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2155 	vap->va_nodeid = zp->z_id;
2156 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2157 		links = pzp->zp_links + 1;
2158 	else
2159 		links = pzp->zp_links;
2160 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2161 	vap->va_size = pzp->zp_size;
2162 	vap->va_rdev = vp->v_rdev;
2163 	vap->va_seq = zp->z_seq;
2164 
2165 	/*
2166 	 * Add in any requested optional attributes and the create time.
2167 	 * Also set the corresponding bits in the returned attribute bitmap.
2168 	 */
2169 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2170 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2171 			xoap->xoa_archive =
2172 			    ((pzp->zp_flags & ZFS_ARCHIVE) != 0);
2173 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2174 		}
2175 
2176 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2177 			xoap->xoa_readonly =
2178 			    ((pzp->zp_flags & ZFS_READONLY) != 0);
2179 			XVA_SET_RTN(xvap, XAT_READONLY);
2180 		}
2181 
2182 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2183 			xoap->xoa_system =
2184 			    ((pzp->zp_flags & ZFS_SYSTEM) != 0);
2185 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2186 		}
2187 
2188 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2189 			xoap->xoa_hidden =
2190 			    ((pzp->zp_flags & ZFS_HIDDEN) != 0);
2191 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2192 		}
2193 
2194 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2195 			xoap->xoa_nounlink =
2196 			    ((pzp->zp_flags & ZFS_NOUNLINK) != 0);
2197 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2198 		}
2199 
2200 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2201 			xoap->xoa_immutable =
2202 			    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0);
2203 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2204 		}
2205 
2206 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2207 			xoap->xoa_appendonly =
2208 			    ((pzp->zp_flags & ZFS_APPENDONLY) != 0);
2209 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2210 		}
2211 
2212 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2213 			xoap->xoa_nodump =
2214 			    ((pzp->zp_flags & ZFS_NODUMP) != 0);
2215 			XVA_SET_RTN(xvap, XAT_NODUMP);
2216 		}
2217 
2218 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2219 			xoap->xoa_opaque =
2220 			    ((pzp->zp_flags & ZFS_OPAQUE) != 0);
2221 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2222 		}
2223 
2224 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2225 			xoap->xoa_av_quarantined =
2226 			    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0);
2227 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2228 		}
2229 
2230 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2231 			xoap->xoa_av_modified =
2232 			    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0);
2233 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2234 		}
2235 
2236 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2237 		    vp->v_type == VREG &&
2238 		    (pzp->zp_flags & ZFS_BONUS_SCANSTAMP)) {
2239 			size_t len;
2240 			dmu_object_info_t doi;
2241 
2242 			/*
2243 			 * Only VREG files have anti-virus scanstamps, so we
2244 			 * won't conflict with symlinks in the bonus buffer.
2245 			 */
2246 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2247 			len = sizeof (xoap->xoa_av_scanstamp) +
2248 			    sizeof (znode_phys_t);
2249 			if (len <= doi.doi_bonus_size) {
2250 				/*
2251 				 * pzp points to the start of the
2252 				 * znode_phys_t. pzp + 1 points to the
2253 				 * first byte after the znode_phys_t.
2254 				 */
2255 				(void) memcpy(xoap->xoa_av_scanstamp,
2256 				    pzp + 1,
2257 				    sizeof (xoap->xoa_av_scanstamp));
2258 				XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
2259 			}
2260 		}
2261 
2262 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2263 			ZFS_TIME_DECODE(&xoap->xoa_createtime, pzp->zp_crtime);
2264 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2265 		}
2266 	}
2267 
2268 	ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime);
2269 	ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime);
2270 	ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime);
2271 
2272 	mutex_exit(&zp->z_lock);
2273 
2274 	dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks);
2275 
2276 	if (zp->z_blksz == 0) {
2277 		/*
2278 		 * Block size hasn't been set; suggest maximal I/O transfers.
2279 		 */
2280 		vap->va_blksize = zfsvfs->z_max_blksz;
2281 	}
2282 
2283 	ZFS_EXIT(zfsvfs);
2284 	return (0);
2285 }
2286 
2287 /*
2288  * Set the file attributes to the values contained in the
2289  * vattr structure.
2290  *
2291  *	IN:	vp	- vnode of file to be modified.
2292  *		vap	- new attribute values.
2293  *			  If AT_XVATTR set, then optional attrs are being set
2294  *		flags	- ATTR_UTIME set if non-default time values provided.
2295  *			- ATTR_NOACLCHECK (CIFS context only).
2296  *		cr	- credentials of caller.
2297  *		ct	- caller context
2298  *
2299  *	RETURN:	0 if success
2300  *		error code if failure
2301  *
2302  * Timestamps:
2303  *	vp - ctime updated, mtime updated if size changed.
2304  */
2305 /* ARGSUSED */
2306 static int
2307 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2308 	caller_context_t *ct)
2309 {
2310 	znode_t		*zp = VTOZ(vp);
2311 	znode_phys_t	*pzp;
2312 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2313 	zilog_t		*zilog;
2314 	dmu_tx_t	*tx;
2315 	vattr_t		oldva;
2316 	uint_t		mask = vap->va_mask;
2317 	uint_t		saved_mask;
2318 	int		trim_mask = 0;
2319 	uint64_t	new_mode;
2320 	znode_t		*attrzp;
2321 	int		need_policy = FALSE;
2322 	int		err;
2323 	zfs_fuid_info_t *fuidp = NULL;
2324 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2325 	xoptattr_t	*xoap;
2326 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2327 
2328 	if (mask == 0)
2329 		return (0);
2330 
2331 	if (mask & AT_NOSET)
2332 		return (EINVAL);
2333 
2334 	ZFS_ENTER(zfsvfs);
2335 	ZFS_VERIFY_ZP(zp);
2336 
2337 	pzp = zp->z_phys;
2338 	zilog = zfsvfs->z_log;
2339 
2340 	/*
2341 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2342 	 * that file system is at proper version level
2343 	 */
2344 
2345 	if (zfsvfs->z_use_fuids == B_FALSE &&
2346 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2347 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2348 	    (mask & AT_XVATTR))) {
2349 		ZFS_EXIT(zfsvfs);
2350 		return (EINVAL);
2351 	}
2352 
2353 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2354 		ZFS_EXIT(zfsvfs);
2355 		return (EISDIR);
2356 	}
2357 
2358 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2359 		ZFS_EXIT(zfsvfs);
2360 		return (EINVAL);
2361 	}
2362 
2363 	/*
2364 	 * If this is an xvattr_t, then get a pointer to the structure of
2365 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2366 	 */
2367 	xoap = xva_getxoptattr(xvap);
2368 
2369 	/*
2370 	 * Immutable files can only alter immutable bit and atime
2371 	 */
2372 	if ((pzp->zp_flags & ZFS_IMMUTABLE) &&
2373 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2374 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2375 		ZFS_EXIT(zfsvfs);
2376 		return (EPERM);
2377 	}
2378 
2379 	if ((mask & AT_SIZE) && (pzp->zp_flags & ZFS_READONLY)) {
2380 		ZFS_EXIT(zfsvfs);
2381 		return (EPERM);
2382 	}
2383 
2384 top:
2385 	attrzp = NULL;
2386 
2387 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2388 		ZFS_EXIT(zfsvfs);
2389 		return (EROFS);
2390 	}
2391 
2392 	/*
2393 	 * First validate permissions
2394 	 */
2395 
2396 	if (mask & AT_SIZE) {
2397 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2398 		if (err) {
2399 			ZFS_EXIT(zfsvfs);
2400 			return (err);
2401 		}
2402 		/*
2403 		 * XXX - Note, we are not providing any open
2404 		 * mode flags here (like FNDELAY), so we may
2405 		 * block if there are locks present... this
2406 		 * should be addressed in openat().
2407 		 */
2408 		do {
2409 			err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2410 			/* NB: we already did dmu_tx_wait() if necessary */
2411 		} while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
2412 		if (err) {
2413 			ZFS_EXIT(zfsvfs);
2414 			return (err);
2415 		}
2416 	}
2417 
2418 	if (mask & (AT_ATIME|AT_MTIME) ||
2419 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2420 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2421 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2422 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2423 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM))))
2424 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2425 		    skipaclchk, cr);
2426 
2427 	if (mask & (AT_UID|AT_GID)) {
2428 		int	idmask = (mask & (AT_UID|AT_GID));
2429 		int	take_owner;
2430 		int	take_group;
2431 
2432 		/*
2433 		 * NOTE: even if a new mode is being set,
2434 		 * we may clear S_ISUID/S_ISGID bits.
2435 		 */
2436 
2437 		if (!(mask & AT_MODE))
2438 			vap->va_mode = pzp->zp_mode;
2439 
2440 		/*
2441 		 * Take ownership or chgrp to group we are a member of
2442 		 */
2443 
2444 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2445 		take_group = (mask & AT_GID) &&
2446 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2447 
2448 		/*
2449 		 * If both AT_UID and AT_GID are set then take_owner and
2450 		 * take_group must both be set in order to allow taking
2451 		 * ownership.
2452 		 *
2453 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2454 		 *
2455 		 */
2456 
2457 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2458 		    ((idmask == AT_UID) && take_owner) ||
2459 		    ((idmask == AT_GID) && take_group)) {
2460 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2461 			    skipaclchk, cr) == 0) {
2462 				/*
2463 				 * Remove setuid/setgid for non-privileged users
2464 				 */
2465 				secpolicy_setid_clear(vap, cr);
2466 				trim_mask = (mask & (AT_UID|AT_GID));
2467 			} else {
2468 				need_policy =  TRUE;
2469 			}
2470 		} else {
2471 			need_policy =  TRUE;
2472 		}
2473 	}
2474 
2475 	mutex_enter(&zp->z_lock);
2476 	oldva.va_mode = pzp->zp_mode;
2477 	zfs_fuid_map_ids(zp, &oldva.va_uid, &oldva.va_gid);
2478 	if (mask & AT_XVATTR) {
2479 		if ((need_policy == FALSE) &&
2480 		    (XVA_ISSET_REQ(xvap, XAT_APPENDONLY) &&
2481 		    xoap->xoa_appendonly !=
2482 		    ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) ||
2483 		    (XVA_ISSET_REQ(xvap, XAT_NOUNLINK) &&
2484 		    xoap->xoa_nounlink !=
2485 		    ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) ||
2486 		    (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE) &&
2487 		    xoap->xoa_immutable !=
2488 		    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) ||
2489 		    (XVA_ISSET_REQ(xvap, XAT_NODUMP) &&
2490 		    xoap->xoa_nodump !=
2491 		    ((pzp->zp_flags & ZFS_NODUMP) != 0)) ||
2492 		    (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED) &&
2493 		    xoap->xoa_av_modified !=
2494 		    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) ||
2495 		    ((XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED) &&
2496 		    ((vp->v_type != VREG && xoap->xoa_av_quarantined) ||
2497 		    xoap->xoa_av_quarantined !=
2498 		    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)))) ||
2499 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
2500 		    (XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2501 			need_policy = TRUE;
2502 		}
2503 	}
2504 
2505 	mutex_exit(&zp->z_lock);
2506 
2507 	if (mask & AT_MODE) {
2508 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2509 			err = secpolicy_setid_setsticky_clear(vp, vap,
2510 			    &oldva, cr);
2511 			if (err) {
2512 				ZFS_EXIT(zfsvfs);
2513 				return (err);
2514 			}
2515 			trim_mask |= AT_MODE;
2516 		} else {
2517 			need_policy = TRUE;
2518 		}
2519 	}
2520 
2521 	if (need_policy) {
2522 		/*
2523 		 * If trim_mask is set then take ownership
2524 		 * has been granted or write_acl is present and user
2525 		 * has the ability to modify mode.  In that case remove
2526 		 * UID|GID and or MODE from mask so that
2527 		 * secpolicy_vnode_setattr() doesn't revoke it.
2528 		 */
2529 
2530 		if (trim_mask) {
2531 			saved_mask = vap->va_mask;
2532 			vap->va_mask &= ~trim_mask;
2533 		}
2534 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2535 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2536 		if (err) {
2537 			ZFS_EXIT(zfsvfs);
2538 			return (err);
2539 		}
2540 
2541 		if (trim_mask)
2542 			vap->va_mask |= saved_mask;
2543 	}
2544 
2545 	/*
2546 	 * secpolicy_vnode_setattr, or take ownership may have
2547 	 * changed va_mask
2548 	 */
2549 	mask = vap->va_mask;
2550 
2551 	tx = dmu_tx_create(zfsvfs->z_os);
2552 	dmu_tx_hold_bonus(tx, zp->z_id);
2553 	if (zfsvfs->z_fuid_obj == 0) {
2554 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
2555 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2556 		    SPA_MAXBLOCKSIZE);
2557 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
2558 	} else {
2559 		dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
2560 		dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
2561 		    SPA_MAXBLOCKSIZE);
2562 	}
2563 
2564 	if (mask & AT_MODE) {
2565 		uint64_t pmode = pzp->zp_mode;
2566 
2567 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2568 
2569 		if (pzp->zp_acl.z_acl_extern_obj) {
2570 			/* Are we upgrading ACL from old V0 format to new V1 */
2571 			if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2572 			    pzp->zp_acl.z_acl_version ==
2573 			    ZFS_ACL_VERSION_INITIAL) {
2574 				dmu_tx_hold_free(tx,
2575 				    pzp->zp_acl.z_acl_extern_obj, 0,
2576 				    DMU_OBJECT_END);
2577 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2578 				    0, sizeof (zfs_object_ace_t) * 2048 + 6);
2579 			} else {
2580 				dmu_tx_hold_write(tx,
2581 				    pzp->zp_acl.z_acl_extern_obj, 0,
2582 				    SPA_MAXBLOCKSIZE);
2583 			}
2584 		} else {
2585 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2586 			    0, sizeof (zfs_object_ace_t) * 2048 + 6);
2587 		}
2588 	}
2589 
2590 	if ((mask & (AT_UID | AT_GID)) && pzp->zp_xattr != 0) {
2591 		err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp);
2592 		if (err) {
2593 			dmu_tx_abort(tx);
2594 			ZFS_EXIT(zfsvfs);
2595 			return (err);
2596 		}
2597 		dmu_tx_hold_bonus(tx, attrzp->z_id);
2598 	}
2599 
2600 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2601 	if (err) {
2602 		if (attrzp)
2603 			VN_RELE(ZTOV(attrzp));
2604 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2605 			dmu_tx_wait(tx);
2606 			dmu_tx_abort(tx);
2607 			goto top;
2608 		}
2609 		dmu_tx_abort(tx);
2610 		ZFS_EXIT(zfsvfs);
2611 		return (err);
2612 	}
2613 
2614 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2615 
2616 	/*
2617 	 * Set each attribute requested.
2618 	 * We group settings according to the locks they need to acquire.
2619 	 *
2620 	 * Note: you cannot set ctime directly, although it will be
2621 	 * updated as a side-effect of calling this function.
2622 	 */
2623 
2624 	mutex_enter(&zp->z_lock);
2625 
2626 	if (mask & AT_MODE) {
2627 		err = zfs_acl_chmod_setattr(zp, new_mode, tx);
2628 		ASSERT3U(err, ==, 0);
2629 	}
2630 
2631 	if (attrzp)
2632 		mutex_enter(&attrzp->z_lock);
2633 
2634 	if (mask & AT_UID) {
2635 		pzp->zp_uid = zfs_fuid_create(zfsvfs,
2636 		    vap->va_uid, ZFS_OWNER, tx, &fuidp);
2637 		if (attrzp) {
2638 			attrzp->z_phys->zp_uid = zfs_fuid_create(zfsvfs,
2639 			    vap->va_uid,  ZFS_OWNER, tx, &fuidp);
2640 		}
2641 	}
2642 
2643 	if (mask & AT_GID) {
2644 		pzp->zp_gid = zfs_fuid_create(zfsvfs, vap->va_gid,
2645 		    ZFS_GROUP, tx, &fuidp);
2646 		if (attrzp)
2647 			attrzp->z_phys->zp_gid = zfs_fuid_create(zfsvfs,
2648 			    vap->va_gid, ZFS_GROUP, tx, &fuidp);
2649 	}
2650 
2651 	if (attrzp)
2652 		mutex_exit(&attrzp->z_lock);
2653 
2654 	if (mask & AT_ATIME)
2655 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2656 
2657 	if (mask & AT_MTIME)
2658 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2659 
2660 	if (mask & AT_SIZE)
2661 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2662 	else if (mask != 0)
2663 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2664 	/*
2665 	 * Do this after setting timestamps to prevent timestamp
2666 	 * update from toggling bit
2667 	 */
2668 
2669 	if (xoap && (mask & AT_XVATTR)) {
2670 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
2671 			size_t len;
2672 			dmu_object_info_t doi;
2673 
2674 			ASSERT(vp->v_type == VREG);
2675 
2676 			/* Grow the bonus buffer if necessary. */
2677 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2678 			len = sizeof (xoap->xoa_av_scanstamp) +
2679 			    sizeof (znode_phys_t);
2680 			if (len > doi.doi_bonus_size)
2681 				VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0);
2682 		}
2683 		zfs_xvattr_set(zp, xvap);
2684 	}
2685 
2686 	if (mask != 0)
2687 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2688 
2689 	if (fuidp)
2690 		zfs_fuid_info_free(fuidp);
2691 	mutex_exit(&zp->z_lock);
2692 
2693 	if (attrzp)
2694 		VN_RELE(ZTOV(attrzp));
2695 
2696 	dmu_tx_commit(tx);
2697 
2698 	ZFS_EXIT(zfsvfs);
2699 	return (err);
2700 }
2701 
2702 typedef struct zfs_zlock {
2703 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2704 	znode_t		*zl_znode;	/* znode we held */
2705 	struct zfs_zlock *zl_next;	/* next in list */
2706 } zfs_zlock_t;
2707 
2708 /*
2709  * Drop locks and release vnodes that were held by zfs_rename_lock().
2710  */
2711 static void
2712 zfs_rename_unlock(zfs_zlock_t **zlpp)
2713 {
2714 	zfs_zlock_t *zl;
2715 
2716 	while ((zl = *zlpp) != NULL) {
2717 		if (zl->zl_znode != NULL)
2718 			VN_RELE(ZTOV(zl->zl_znode));
2719 		rw_exit(zl->zl_rwlock);
2720 		*zlpp = zl->zl_next;
2721 		kmem_free(zl, sizeof (*zl));
2722 	}
2723 }
2724 
2725 /*
2726  * Search back through the directory tree, using the ".." entries.
2727  * Lock each directory in the chain to prevent concurrent renames.
2728  * Fail any attempt to move a directory into one of its own descendants.
2729  * XXX - z_parent_lock can overlap with map or grow locks
2730  */
2731 static int
2732 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2733 {
2734 	zfs_zlock_t	*zl;
2735 	znode_t		*zp = tdzp;
2736 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2737 	uint64_t	*oidp = &zp->z_id;
2738 	krwlock_t	*rwlp = &szp->z_parent_lock;
2739 	krw_t		rw = RW_WRITER;
2740 
2741 	/*
2742 	 * First pass write-locks szp and compares to zp->z_id.
2743 	 * Later passes read-lock zp and compare to zp->z_parent.
2744 	 */
2745 	do {
2746 		if (!rw_tryenter(rwlp, rw)) {
2747 			/*
2748 			 * Another thread is renaming in this path.
2749 			 * Note that if we are a WRITER, we don't have any
2750 			 * parent_locks held yet.
2751 			 */
2752 			if (rw == RW_READER && zp->z_id > szp->z_id) {
2753 				/*
2754 				 * Drop our locks and restart
2755 				 */
2756 				zfs_rename_unlock(&zl);
2757 				*zlpp = NULL;
2758 				zp = tdzp;
2759 				oidp = &zp->z_id;
2760 				rwlp = &szp->z_parent_lock;
2761 				rw = RW_WRITER;
2762 				continue;
2763 			} else {
2764 				/*
2765 				 * Wait for other thread to drop its locks
2766 				 */
2767 				rw_enter(rwlp, rw);
2768 			}
2769 		}
2770 
2771 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2772 		zl->zl_rwlock = rwlp;
2773 		zl->zl_znode = NULL;
2774 		zl->zl_next = *zlpp;
2775 		*zlpp = zl;
2776 
2777 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2778 			return (EINVAL);
2779 
2780 		if (*oidp == rootid)		/* We've hit the top */
2781 			return (0);
2782 
2783 		if (rw == RW_READER) {		/* i.e. not the first pass */
2784 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2785 			if (error)
2786 				return (error);
2787 			zl->zl_znode = zp;
2788 		}
2789 		oidp = &zp->z_phys->zp_parent;
2790 		rwlp = &zp->z_parent_lock;
2791 		rw = RW_READER;
2792 
2793 	} while (zp->z_id != sdzp->z_id);
2794 
2795 	return (0);
2796 }
2797 
2798 /*
2799  * Move an entry from the provided source directory to the target
2800  * directory.  Change the entry name as indicated.
2801  *
2802  *	IN:	sdvp	- Source directory containing the "old entry".
2803  *		snm	- Old entry name.
2804  *		tdvp	- Target directory to contain the "new entry".
2805  *		tnm	- New entry name.
2806  *		cr	- credentials of caller.
2807  *		ct	- caller context
2808  *		flags	- case flags
2809  *
2810  *	RETURN:	0 if success
2811  *		error code if failure
2812  *
2813  * Timestamps:
2814  *	sdvp,tdvp - ctime|mtime updated
2815  */
2816 /*ARGSUSED*/
2817 static int
2818 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
2819     caller_context_t *ct, int flags)
2820 {
2821 	znode_t		*tdzp, *szp, *tzp;
2822 	znode_t		*sdzp = VTOZ(sdvp);
2823 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2824 	zilog_t		*zilog;
2825 	vnode_t		*realvp;
2826 	zfs_dirlock_t	*sdl, *tdl;
2827 	dmu_tx_t	*tx;
2828 	zfs_zlock_t	*zl;
2829 	int		cmp, serr, terr;
2830 	int		error = 0;
2831 	int		zflg = 0;
2832 
2833 	ZFS_ENTER(zfsvfs);
2834 	ZFS_VERIFY_ZP(sdzp);
2835 	zilog = zfsvfs->z_log;
2836 
2837 	/*
2838 	 * Make sure we have the real vp for the target directory.
2839 	 */
2840 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
2841 		tdvp = realvp;
2842 
2843 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
2844 		ZFS_EXIT(zfsvfs);
2845 		return (EXDEV);
2846 	}
2847 
2848 	tdzp = VTOZ(tdvp);
2849 	ZFS_VERIFY_ZP(tdzp);
2850 	if (zfsvfs->z_utf8 && u8_validate(tnm,
2851 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2852 		ZFS_EXIT(zfsvfs);
2853 		return (EILSEQ);
2854 	}
2855 
2856 	if (flags & FIGNORECASE)
2857 		zflg |= ZCILOOK;
2858 
2859 top:
2860 	szp = NULL;
2861 	tzp = NULL;
2862 	zl = NULL;
2863 
2864 	/*
2865 	 * This is to prevent the creation of links into attribute space
2866 	 * by renaming a linked file into/outof an attribute directory.
2867 	 * See the comment in zfs_link() for why this is considered bad.
2868 	 */
2869 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
2870 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
2871 		ZFS_EXIT(zfsvfs);
2872 		return (EINVAL);
2873 	}
2874 
2875 	/*
2876 	 * Lock source and target directory entries.  To prevent deadlock,
2877 	 * a lock ordering must be defined.  We lock the directory with
2878 	 * the smallest object id first, or if it's a tie, the one with
2879 	 * the lexically first name.
2880 	 */
2881 	if (sdzp->z_id < tdzp->z_id) {
2882 		cmp = -1;
2883 	} else if (sdzp->z_id > tdzp->z_id) {
2884 		cmp = 1;
2885 	} else {
2886 		/*
2887 		 * First compare the two name arguments without
2888 		 * considering any case folding.
2889 		 */
2890 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
2891 
2892 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
2893 		ASSERT(error == 0 || !zfsvfs->z_utf8);
2894 		if (cmp == 0) {
2895 			/*
2896 			 * POSIX: "If the old argument and the new argument
2897 			 * both refer to links to the same existing file,
2898 			 * the rename() function shall return successfully
2899 			 * and perform no other action."
2900 			 */
2901 			ZFS_EXIT(zfsvfs);
2902 			return (0);
2903 		}
2904 		/*
2905 		 * If the file system is case-folding, then we may
2906 		 * have some more checking to do.  A case-folding file
2907 		 * system is either supporting mixed case sensitivity
2908 		 * access or is completely case-insensitive.  Note
2909 		 * that the file system is always case preserving.
2910 		 *
2911 		 * In mixed sensitivity mode case sensitive behavior
2912 		 * is the default.  FIGNORECASE must be used to
2913 		 * explicitly request case insensitive behavior.
2914 		 *
2915 		 * If the source and target names provided differ only
2916 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
2917 		 * we will treat this as a special case in the
2918 		 * case-insensitive mode: as long as the source name
2919 		 * is an exact match, we will allow this to proceed as
2920 		 * a name-change request.
2921 		 */
2922 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
2923 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
2924 		    flags & FIGNORECASE)) &&
2925 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
2926 		    &error) == 0) {
2927 			/*
2928 			 * case preserving rename request, require exact
2929 			 * name matches
2930 			 */
2931 			zflg |= ZCIEXACT;
2932 			zflg &= ~ZCILOOK;
2933 		}
2934 	}
2935 
2936 	if (cmp < 0) {
2937 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
2938 		    ZEXISTS | zflg, NULL, NULL);
2939 		terr = zfs_dirent_lock(&tdl,
2940 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
2941 	} else {
2942 		terr = zfs_dirent_lock(&tdl,
2943 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
2944 		serr = zfs_dirent_lock(&sdl,
2945 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
2946 		    NULL, NULL);
2947 	}
2948 
2949 	if (serr) {
2950 		/*
2951 		 * Source entry invalid or not there.
2952 		 */
2953 		if (!terr) {
2954 			zfs_dirent_unlock(tdl);
2955 			if (tzp)
2956 				VN_RELE(ZTOV(tzp));
2957 		}
2958 		if (strcmp(snm, "..") == 0)
2959 			serr = EINVAL;
2960 		ZFS_EXIT(zfsvfs);
2961 		return (serr);
2962 	}
2963 	if (terr) {
2964 		zfs_dirent_unlock(sdl);
2965 		VN_RELE(ZTOV(szp));
2966 		if (strcmp(tnm, "..") == 0)
2967 			terr = EINVAL;
2968 		ZFS_EXIT(zfsvfs);
2969 		return (terr);
2970 	}
2971 
2972 	/*
2973 	 * Must have write access at the source to remove the old entry
2974 	 * and write access at the target to create the new entry.
2975 	 * Note that if target and source are the same, this can be
2976 	 * done in a single check.
2977 	 */
2978 
2979 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
2980 		goto out;
2981 
2982 	if (ZTOV(szp)->v_type == VDIR) {
2983 		/*
2984 		 * Check to make sure rename is valid.
2985 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
2986 		 */
2987 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
2988 			goto out;
2989 	}
2990 
2991 	/*
2992 	 * Does target exist?
2993 	 */
2994 	if (tzp) {
2995 		/*
2996 		 * Source and target must be the same type.
2997 		 */
2998 		if (ZTOV(szp)->v_type == VDIR) {
2999 			if (ZTOV(tzp)->v_type != VDIR) {
3000 				error = ENOTDIR;
3001 				goto out;
3002 			}
3003 		} else {
3004 			if (ZTOV(tzp)->v_type == VDIR) {
3005 				error = EISDIR;
3006 				goto out;
3007 			}
3008 		}
3009 		/*
3010 		 * POSIX dictates that when the source and target
3011 		 * entries refer to the same file object, rename
3012 		 * must do nothing and exit without error.
3013 		 */
3014 		if (szp->z_id == tzp->z_id) {
3015 			error = 0;
3016 			goto out;
3017 		}
3018 	}
3019 
3020 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3021 	if (tzp)
3022 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3023 
3024 	/*
3025 	 * notify the target directory if it is not the same
3026 	 * as source directory.
3027 	 */
3028 	if (tdvp != sdvp) {
3029 		vnevent_rename_dest_dir(tdvp, ct);
3030 	}
3031 
3032 	tx = dmu_tx_create(zfsvfs->z_os);
3033 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
3034 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
3035 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3036 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3037 	if (sdzp != tdzp)
3038 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
3039 	if (tzp)
3040 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* parent changes */
3041 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3042 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3043 	if (error) {
3044 		if (zl != NULL)
3045 			zfs_rename_unlock(&zl);
3046 		zfs_dirent_unlock(sdl);
3047 		zfs_dirent_unlock(tdl);
3048 		VN_RELE(ZTOV(szp));
3049 		if (tzp)
3050 			VN_RELE(ZTOV(tzp));
3051 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3052 			dmu_tx_wait(tx);
3053 			dmu_tx_abort(tx);
3054 			goto top;
3055 		}
3056 		dmu_tx_abort(tx);
3057 		ZFS_EXIT(zfsvfs);
3058 		return (error);
3059 	}
3060 
3061 	if (tzp)	/* Attempt to remove the existing target */
3062 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3063 
3064 	if (error == 0) {
3065 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3066 		if (error == 0) {
3067 			szp->z_phys->zp_flags |= ZFS_AV_MODIFIED;
3068 
3069 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3070 			ASSERT(error == 0);
3071 
3072 			zfs_log_rename(zilog, tx,
3073 			    TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0),
3074 			    sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp);
3075 		}
3076 	}
3077 
3078 	dmu_tx_commit(tx);
3079 out:
3080 	if (zl != NULL)
3081 		zfs_rename_unlock(&zl);
3082 
3083 	zfs_dirent_unlock(sdl);
3084 	zfs_dirent_unlock(tdl);
3085 
3086 	VN_RELE(ZTOV(szp));
3087 	if (tzp)
3088 		VN_RELE(ZTOV(tzp));
3089 
3090 	ZFS_EXIT(zfsvfs);
3091 	return (error);
3092 }
3093 
3094 /*
3095  * Insert the indicated symbolic reference entry into the directory.
3096  *
3097  *	IN:	dvp	- Directory to contain new symbolic link.
3098  *		link	- Name for new symlink entry.
3099  *		vap	- Attributes of new entry.
3100  *		target	- Target path of new symlink.
3101  *		cr	- credentials of caller.
3102  *		ct	- caller context
3103  *		flags	- case flags
3104  *
3105  *	RETURN:	0 if success
3106  *		error code if failure
3107  *
3108  * Timestamps:
3109  *	dvp - ctime|mtime updated
3110  */
3111 /*ARGSUSED*/
3112 static int
3113 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3114     caller_context_t *ct, int flags)
3115 {
3116 	znode_t		*zp, *dzp = VTOZ(dvp);
3117 	zfs_dirlock_t	*dl;
3118 	dmu_tx_t	*tx;
3119 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3120 	zilog_t		*zilog;
3121 	int		len = strlen(link);
3122 	int		error;
3123 	int		zflg = ZNEW;
3124 	zfs_fuid_info_t *fuidp = NULL;
3125 
3126 	ASSERT(vap->va_type == VLNK);
3127 
3128 	ZFS_ENTER(zfsvfs);
3129 	ZFS_VERIFY_ZP(dzp);
3130 	zilog = zfsvfs->z_log;
3131 
3132 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3133 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3134 		ZFS_EXIT(zfsvfs);
3135 		return (EILSEQ);
3136 	}
3137 	if (flags & FIGNORECASE)
3138 		zflg |= ZCILOOK;
3139 top:
3140 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3141 		ZFS_EXIT(zfsvfs);
3142 		return (error);
3143 	}
3144 
3145 	if (len > MAXPATHLEN) {
3146 		ZFS_EXIT(zfsvfs);
3147 		return (ENAMETOOLONG);
3148 	}
3149 
3150 	/*
3151 	 * Attempt to lock directory; fail if entry already exists.
3152 	 */
3153 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3154 	if (error) {
3155 		ZFS_EXIT(zfsvfs);
3156 		return (error);
3157 	}
3158 
3159 	tx = dmu_tx_create(zfsvfs->z_os);
3160 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3161 	dmu_tx_hold_bonus(tx, dzp->z_id);
3162 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3163 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
3164 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
3165 	if (zfsvfs->z_fuid_obj == 0) {
3166 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
3167 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3168 		    SPA_MAXBLOCKSIZE);
3169 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
3170 	} else {
3171 		dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
3172 		dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
3173 		    SPA_MAXBLOCKSIZE);
3174 	}
3175 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3176 	if (error) {
3177 		zfs_dirent_unlock(dl);
3178 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3179 			dmu_tx_wait(tx);
3180 			dmu_tx_abort(tx);
3181 			goto top;
3182 		}
3183 		dmu_tx_abort(tx);
3184 		ZFS_EXIT(zfsvfs);
3185 		return (error);
3186 	}
3187 
3188 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
3189 
3190 	/*
3191 	 * Create a new object for the symlink.
3192 	 * Put the link content into bonus buffer if it will fit;
3193 	 * otherwise, store it just like any other file data.
3194 	 */
3195 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
3196 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, len, NULL, &fuidp);
3197 		if (len != 0)
3198 			bcopy(link, zp->z_phys + 1, len);
3199 	} else {
3200 		dmu_buf_t *dbp;
3201 
3202 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, NULL, &fuidp);
3203 		/*
3204 		 * Nothing can access the znode yet so no locking needed
3205 		 * for growing the znode's blocksize.
3206 		 */
3207 		zfs_grow_blocksize(zp, len, tx);
3208 
3209 		VERIFY(0 == dmu_buf_hold(zfsvfs->z_os,
3210 		    zp->z_id, 0, FTAG, &dbp));
3211 		dmu_buf_will_dirty(dbp, tx);
3212 
3213 		ASSERT3U(len, <=, dbp->db_size);
3214 		bcopy(link, dbp->db_data, len);
3215 		dmu_buf_rele(dbp, FTAG);
3216 	}
3217 	zp->z_phys->zp_size = len;
3218 
3219 	/*
3220 	 * Insert the new object into the directory.
3221 	 */
3222 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3223 out:
3224 	if (error == 0) {
3225 		uint64_t txtype = TX_SYMLINK;
3226 		if (flags & FIGNORECASE)
3227 			txtype |= TX_CI;
3228 		zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3229 	}
3230 	if (fuidp)
3231 		zfs_fuid_info_free(fuidp);
3232 
3233 	dmu_tx_commit(tx);
3234 
3235 	zfs_dirent_unlock(dl);
3236 
3237 	VN_RELE(ZTOV(zp));
3238 
3239 	ZFS_EXIT(zfsvfs);
3240 	return (error);
3241 }
3242 
3243 /*
3244  * Return, in the buffer contained in the provided uio structure,
3245  * the symbolic path referred to by vp.
3246  *
3247  *	IN:	vp	- vnode of symbolic link.
3248  *		uoip	- structure to contain the link path.
3249  *		cr	- credentials of caller.
3250  *		ct	- caller context
3251  *
3252  *	OUT:	uio	- structure to contain the link path.
3253  *
3254  *	RETURN:	0 if success
3255  *		error code if failure
3256  *
3257  * Timestamps:
3258  *	vp - atime updated
3259  */
3260 /* ARGSUSED */
3261 static int
3262 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3263 {
3264 	znode_t		*zp = VTOZ(vp);
3265 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3266 	size_t		bufsz;
3267 	int		error;
3268 
3269 	ZFS_ENTER(zfsvfs);
3270 	ZFS_VERIFY_ZP(zp);
3271 
3272 	bufsz = (size_t)zp->z_phys->zp_size;
3273 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
3274 		error = uiomove(zp->z_phys + 1,
3275 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3276 	} else {
3277 		dmu_buf_t *dbp;
3278 		error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp);
3279 		if (error) {
3280 			ZFS_EXIT(zfsvfs);
3281 			return (error);
3282 		}
3283 		error = uiomove(dbp->db_data,
3284 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3285 		dmu_buf_rele(dbp, FTAG);
3286 	}
3287 
3288 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3289 	ZFS_EXIT(zfsvfs);
3290 	return (error);
3291 }
3292 
3293 /*
3294  * Insert a new entry into directory tdvp referencing svp.
3295  *
3296  *	IN:	tdvp	- Directory to contain new entry.
3297  *		svp	- vnode of new entry.
3298  *		name	- name of new entry.
3299  *		cr	- credentials of caller.
3300  *		ct	- caller context
3301  *
3302  *	RETURN:	0 if success
3303  *		error code if failure
3304  *
3305  * Timestamps:
3306  *	tdvp - ctime|mtime updated
3307  *	 svp - ctime updated
3308  */
3309 /* ARGSUSED */
3310 static int
3311 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3312     caller_context_t *ct, int flags)
3313 {
3314 	znode_t		*dzp = VTOZ(tdvp);
3315 	znode_t		*tzp, *szp;
3316 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3317 	zilog_t		*zilog;
3318 	zfs_dirlock_t	*dl;
3319 	dmu_tx_t	*tx;
3320 	vnode_t		*realvp;
3321 	int		error;
3322 	int		zf = ZNEW;
3323 	uid_t		owner;
3324 
3325 	ASSERT(tdvp->v_type == VDIR);
3326 
3327 	ZFS_ENTER(zfsvfs);
3328 	ZFS_VERIFY_ZP(dzp);
3329 	zilog = zfsvfs->z_log;
3330 
3331 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3332 		svp = realvp;
3333 
3334 	if (svp->v_vfsp != tdvp->v_vfsp) {
3335 		ZFS_EXIT(zfsvfs);
3336 		return (EXDEV);
3337 	}
3338 	szp = VTOZ(svp);
3339 	ZFS_VERIFY_ZP(szp);
3340 
3341 	if (zfsvfs->z_utf8 && u8_validate(name,
3342 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3343 		ZFS_EXIT(zfsvfs);
3344 		return (EILSEQ);
3345 	}
3346 	if (flags & FIGNORECASE)
3347 		zf |= ZCILOOK;
3348 
3349 top:
3350 	/*
3351 	 * We do not support links between attributes and non-attributes
3352 	 * because of the potential security risk of creating links
3353 	 * into "normal" file space in order to circumvent restrictions
3354 	 * imposed in attribute space.
3355 	 */
3356 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
3357 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
3358 		ZFS_EXIT(zfsvfs);
3359 		return (EINVAL);
3360 	}
3361 
3362 	/*
3363 	 * POSIX dictates that we return EPERM here.
3364 	 * Better choices include ENOTSUP or EISDIR.
3365 	 */
3366 	if (svp->v_type == VDIR) {
3367 		ZFS_EXIT(zfsvfs);
3368 		return (EPERM);
3369 	}
3370 
3371 	zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, ZFS_OWNER, &owner);
3372 	if (owner != crgetuid(cr) &&
3373 	    secpolicy_basic_link(cr) != 0) {
3374 		ZFS_EXIT(zfsvfs);
3375 		return (EPERM);
3376 	}
3377 
3378 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3379 		ZFS_EXIT(zfsvfs);
3380 		return (error);
3381 	}
3382 
3383 	/*
3384 	 * Attempt to lock directory; fail if entry already exists.
3385 	 */
3386 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3387 	if (error) {
3388 		ZFS_EXIT(zfsvfs);
3389 		return (error);
3390 	}
3391 
3392 	tx = dmu_tx_create(zfsvfs->z_os);
3393 	dmu_tx_hold_bonus(tx, szp->z_id);
3394 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3395 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3396 	if (error) {
3397 		zfs_dirent_unlock(dl);
3398 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3399 			dmu_tx_wait(tx);
3400 			dmu_tx_abort(tx);
3401 			goto top;
3402 		}
3403 		dmu_tx_abort(tx);
3404 		ZFS_EXIT(zfsvfs);
3405 		return (error);
3406 	}
3407 
3408 	error = zfs_link_create(dl, szp, tx, 0);
3409 
3410 	if (error == 0) {
3411 		uint64_t txtype = TX_LINK;
3412 		if (flags & FIGNORECASE)
3413 			txtype |= TX_CI;
3414 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3415 	}
3416 
3417 	dmu_tx_commit(tx);
3418 
3419 	zfs_dirent_unlock(dl);
3420 
3421 	if (error == 0) {
3422 		vnevent_link(svp, ct);
3423 	}
3424 
3425 	ZFS_EXIT(zfsvfs);
3426 	return (error);
3427 }
3428 
3429 /*
3430  * zfs_null_putapage() is used when the file system has been force
3431  * unmounted. It just drops the pages.
3432  */
3433 /* ARGSUSED */
3434 static int
3435 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3436 		size_t *lenp, int flags, cred_t *cr)
3437 {
3438 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
3439 	return (0);
3440 }
3441 
3442 /*
3443  * Push a page out to disk, klustering if possible.
3444  *
3445  *	IN:	vp	- file to push page to.
3446  *		pp	- page to push.
3447  *		flags	- additional flags.
3448  *		cr	- credentials of caller.
3449  *
3450  *	OUT:	offp	- start of range pushed.
3451  *		lenp	- len of range pushed.
3452  *
3453  *	RETURN:	0 if success
3454  *		error code if failure
3455  *
3456  * NOTE: callers must have locked the page to be pushed.  On
3457  * exit, the page (and all other pages in the kluster) must be
3458  * unlocked.
3459  */
3460 /* ARGSUSED */
3461 static int
3462 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3463 		size_t *lenp, int flags, cred_t *cr)
3464 {
3465 	znode_t		*zp = VTOZ(vp);
3466 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3467 	zilog_t		*zilog = zfsvfs->z_log;
3468 	dmu_tx_t	*tx;
3469 	rl_t		*rl;
3470 	u_offset_t	off, koff;
3471 	size_t		len, klen;
3472 	uint64_t	filesz;
3473 	int		err;
3474 
3475 	filesz = zp->z_phys->zp_size;
3476 	off = pp->p_offset;
3477 	len = PAGESIZE;
3478 	/*
3479 	 * If our blocksize is bigger than the page size, try to kluster
3480 	 * muiltiple pages so that we write a full block (thus avoiding
3481 	 * a read-modify-write).
3482 	 */
3483 	if (off < filesz && zp->z_blksz > PAGESIZE) {
3484 		if (!ISP2(zp->z_blksz)) {
3485 			/* Only one block in the file. */
3486 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3487 			koff = 0;
3488 		} else {
3489 			klen = zp->z_blksz;
3490 			koff = P2ALIGN(off, (u_offset_t)klen);
3491 		}
3492 		ASSERT(koff <= filesz);
3493 		if (koff + klen > filesz)
3494 			klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE);
3495 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
3496 	}
3497 	ASSERT3U(btop(len), ==, btopr(len));
3498 top:
3499 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
3500 	/*
3501 	 * Can't push pages past end-of-file.
3502 	 */
3503 	filesz = zp->z_phys->zp_size;
3504 	if (off >= filesz) {
3505 		/* ignore all pages */
3506 		err = 0;
3507 		goto out;
3508 	} else if (off + len > filesz) {
3509 		int npages = btopr(filesz - off);
3510 		page_t *trunc;
3511 
3512 		page_list_break(&pp, &trunc, npages);
3513 		/* ignore pages past end of file */
3514 		if (trunc)
3515 			pvn_write_done(trunc, flags);
3516 		len = filesz - off;
3517 	}
3518 
3519 	tx = dmu_tx_create(zfsvfs->z_os);
3520 	dmu_tx_hold_write(tx, zp->z_id, off, len);
3521 	dmu_tx_hold_bonus(tx, zp->z_id);
3522 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
3523 	if (err != 0) {
3524 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3525 			zfs_range_unlock(rl);
3526 			dmu_tx_wait(tx);
3527 			dmu_tx_abort(tx);
3528 			err = 0;
3529 			goto top;
3530 		}
3531 		dmu_tx_abort(tx);
3532 		goto out;
3533 	}
3534 
3535 	if (zp->z_blksz <= PAGESIZE) {
3536 		caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1);
3537 		ASSERT3U(len, <=, PAGESIZE);
3538 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
3539 		ppmapout(va);
3540 	} else {
3541 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
3542 	}
3543 
3544 	if (err == 0) {
3545 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
3546 		zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0);
3547 		dmu_tx_commit(tx);
3548 	}
3549 
3550 out:
3551 	zfs_range_unlock(rl);
3552 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
3553 	if (offp)
3554 		*offp = off;
3555 	if (lenp)
3556 		*lenp = len;
3557 
3558 	return (err);
3559 }
3560 
3561 /*
3562  * Copy the portion of the file indicated from pages into the file.
3563  * The pages are stored in a page list attached to the files vnode.
3564  *
3565  *	IN:	vp	- vnode of file to push page data to.
3566  *		off	- position in file to put data.
3567  *		len	- amount of data to write.
3568  *		flags	- flags to control the operation.
3569  *		cr	- credentials of caller.
3570  *		ct	- caller context.
3571  *
3572  *	RETURN:	0 if success
3573  *		error code if failure
3574  *
3575  * Timestamps:
3576  *	vp - ctime|mtime updated
3577  */
3578 /*ARGSUSED*/
3579 static int
3580 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
3581     caller_context_t *ct)
3582 {
3583 	znode_t		*zp = VTOZ(vp);
3584 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3585 	page_t		*pp;
3586 	size_t		io_len;
3587 	u_offset_t	io_off;
3588 	uint64_t	filesz;
3589 	int		error = 0;
3590 
3591 	ZFS_ENTER(zfsvfs);
3592 	ZFS_VERIFY_ZP(zp);
3593 
3594 	if (len == 0) {
3595 		/*
3596 		 * Search the entire vp list for pages >= off.
3597 		 */
3598 		error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage,
3599 		    flags, cr);
3600 		goto out;
3601 	}
3602 
3603 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3604 	if (off > filesz) {
3605 		/* past end of file */
3606 		ZFS_EXIT(zfsvfs);
3607 		return (0);
3608 	}
3609 
3610 	len = MIN(len, filesz - off);
3611 
3612 	for (io_off = off; io_off < off + len; io_off += io_len) {
3613 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
3614 			pp = page_lookup(vp, io_off,
3615 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
3616 		} else {
3617 			pp = page_lookup_nowait(vp, io_off,
3618 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
3619 		}
3620 
3621 		if (pp != NULL && pvn_getdirty(pp, flags)) {
3622 			int err;
3623 
3624 			/*
3625 			 * Found a dirty page to push
3626 			 */
3627 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
3628 			if (err)
3629 				error = err;
3630 		} else {
3631 			io_len = PAGESIZE;
3632 		}
3633 	}
3634 out:
3635 	if ((flags & B_ASYNC) == 0)
3636 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
3637 	ZFS_EXIT(zfsvfs);
3638 	return (error);
3639 }
3640 
3641 /*ARGSUSED*/
3642 void
3643 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
3644 {
3645 	znode_t	*zp = VTOZ(vp);
3646 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3647 	int error;
3648 
3649 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
3650 	if (zp->z_dbuf == NULL) {
3651 		/*
3652 		 * The fs has been unmounted, or we did a
3653 		 * suspend/resume and this file no longer exists.
3654 		 */
3655 		if (vn_has_cached_data(vp)) {
3656 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
3657 			    B_INVAL, cr);
3658 		}
3659 
3660 		mutex_enter(&zp->z_lock);
3661 		vp->v_count = 0; /* count arrives as 1 */
3662 		mutex_exit(&zp->z_lock);
3663 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
3664 		zfs_znode_free(zp);
3665 		return;
3666 	}
3667 
3668 	/*
3669 	 * Attempt to push any data in the page cache.  If this fails
3670 	 * we will get kicked out later in zfs_zinactive().
3671 	 */
3672 	if (vn_has_cached_data(vp)) {
3673 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
3674 		    cr);
3675 	}
3676 
3677 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3678 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3679 
3680 		dmu_tx_hold_bonus(tx, zp->z_id);
3681 		error = dmu_tx_assign(tx, TXG_WAIT);
3682 		if (error) {
3683 			dmu_tx_abort(tx);
3684 		} else {
3685 			dmu_buf_will_dirty(zp->z_dbuf, tx);
3686 			mutex_enter(&zp->z_lock);
3687 			zp->z_atime_dirty = 0;
3688 			mutex_exit(&zp->z_lock);
3689 			dmu_tx_commit(tx);
3690 		}
3691 	}
3692 
3693 	zfs_zinactive(zp);
3694 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
3695 }
3696 
3697 /*
3698  * Bounds-check the seek operation.
3699  *
3700  *	IN:	vp	- vnode seeking within
3701  *		ooff	- old file offset
3702  *		noffp	- pointer to new file offset
3703  *		ct	- caller context
3704  *
3705  *	RETURN:	0 if success
3706  *		EINVAL if new offset invalid
3707  */
3708 /* ARGSUSED */
3709 static int
3710 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
3711     caller_context_t *ct)
3712 {
3713 	if (vp->v_type == VDIR)
3714 		return (0);
3715 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3716 }
3717 
3718 /*
3719  * Pre-filter the generic locking function to trap attempts to place
3720  * a mandatory lock on a memory mapped file.
3721  */
3722 static int
3723 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3724     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
3725 {
3726 	znode_t *zp = VTOZ(vp);
3727 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3728 	int error;
3729 
3730 	ZFS_ENTER(zfsvfs);
3731 	ZFS_VERIFY_ZP(zp);
3732 
3733 	/*
3734 	 * We are following the UFS semantics with respect to mapcnt
3735 	 * here: If we see that the file is mapped already, then we will
3736 	 * return an error, but we don't worry about races between this
3737 	 * function and zfs_map().
3738 	 */
3739 	if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3740 		ZFS_EXIT(zfsvfs);
3741 		return (EAGAIN);
3742 	}
3743 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct);
3744 	ZFS_EXIT(zfsvfs);
3745 	return (error);
3746 }
3747 
3748 /*
3749  * If we can't find a page in the cache, we will create a new page
3750  * and fill it with file data.  For efficiency, we may try to fill
3751  * multiple pages at once (klustering).
3752  */
3753 static int
3754 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3755     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3756 {
3757 	znode_t *zp = VTOZ(vp);
3758 	page_t *pp, *cur_pp;
3759 	objset_t *os = zp->z_zfsvfs->z_os;
3760 	caddr_t va;
3761 	u_offset_t io_off, total;
3762 	uint64_t oid = zp->z_id;
3763 	size_t io_len;
3764 	uint64_t filesz;
3765 	int err;
3766 
3767 	/*
3768 	 * If we are only asking for a single page don't bother klustering.
3769 	 */
3770 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3771 	if (off >= filesz)
3772 		return (EFAULT);
3773 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3774 		io_off = off;
3775 		io_len = PAGESIZE;
3776 		pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr);
3777 	} else {
3778 		/*
3779 		 * Try to fill a kluster of pages (a blocks worth).
3780 		 */
3781 		size_t klen;
3782 		u_offset_t koff;
3783 
3784 		if (!ISP2(zp->z_blksz)) {
3785 			/* Only one block in the file. */
3786 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3787 			koff = 0;
3788 		} else {
3789 			/*
3790 			 * It would be ideal to align our offset to the
3791 			 * blocksize but doing so has resulted in some
3792 			 * strange application crashes. For now, we
3793 			 * leave the offset as is and only adjust the
3794 			 * length if we are off the end of the file.
3795 			 */
3796 			koff = off;
3797 			klen = plsz;
3798 		}
3799 		ASSERT(koff <= filesz);
3800 		if (koff + klen > filesz)
3801 			klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff;
3802 		ASSERT3U(off, >=, koff);
3803 		ASSERT3U(off, <, koff + klen);
3804 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3805 		    &io_len, koff, klen, 0);
3806 	}
3807 	if (pp == NULL) {
3808 		/*
3809 		 * Some other thread entered the page before us.
3810 		 * Return to zfs_getpage to retry the lookup.
3811 		 */
3812 		*pl = NULL;
3813 		return (0);
3814 	}
3815 
3816 	/*
3817 	 * Fill the pages in the kluster.
3818 	 */
3819 	cur_pp = pp;
3820 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3821 		ASSERT3U(io_off, ==, cur_pp->p_offset);
3822 		va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1);
3823 		err = dmu_read(os, oid, io_off, PAGESIZE, va);
3824 		ppmapout(va);
3825 		if (err) {
3826 			/* On error, toss the entire kluster */
3827 			pvn_read_done(pp, B_ERROR);
3828 			return (err);
3829 		}
3830 		cur_pp = cur_pp->p_next;
3831 	}
3832 out:
3833 	/*
3834 	 * Fill in the page list array from the kluster.  If
3835 	 * there are too many pages in the kluster, return
3836 	 * as many pages as possible starting from the desired
3837 	 * offset `off'.
3838 	 * NOTE: the page list will always be null terminated.
3839 	 */
3840 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3841 
3842 	return (0);
3843 }
3844 
3845 /*
3846  * Return pointers to the pages for the file region [off, off + len]
3847  * in the pl array.  If plsz is greater than len, this function may
3848  * also return page pointers from before or after the specified
3849  * region (i.e. some region [off', off' + plsz]).  These additional
3850  * pages are only returned if they are already in the cache, or were
3851  * created as part of a klustered read.
3852  *
3853  *	IN:	vp	- vnode of file to get data from.
3854  *		off	- position in file to get data from.
3855  *		len	- amount of data to retrieve.
3856  *		plsz	- length of provided page list.
3857  *		seg	- segment to obtain pages for.
3858  *		addr	- virtual address of fault.
3859  *		rw	- mode of created pages.
3860  *		cr	- credentials of caller.
3861  *		ct	- caller context.
3862  *
3863  *	OUT:	protp	- protection mode of created pages.
3864  *		pl	- list of pages created.
3865  *
3866  *	RETURN:	0 if success
3867  *		error code if failure
3868  *
3869  * Timestamps:
3870  *	vp - atime updated
3871  */
3872 /* ARGSUSED */
3873 static int
3874 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3875 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3876 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
3877 {
3878 	znode_t		*zp = VTOZ(vp);
3879 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3880 	page_t		*pp, **pl0 = pl;
3881 	int		need_unlock = 0, err = 0;
3882 	offset_t	orig_off;
3883 
3884 	ZFS_ENTER(zfsvfs);
3885 	ZFS_VERIFY_ZP(zp);
3886 
3887 	if (protp)
3888 		*protp = PROT_ALL;
3889 
3890 	/* no faultahead (for now) */
3891 	if (pl == NULL) {
3892 		ZFS_EXIT(zfsvfs);
3893 		return (0);
3894 	}
3895 
3896 	/* can't fault past EOF */
3897 	if (off >= zp->z_phys->zp_size) {
3898 		ZFS_EXIT(zfsvfs);
3899 		return (EFAULT);
3900 	}
3901 	orig_off = off;
3902 
3903 	/*
3904 	 * If we already own the lock, then we must be page faulting
3905 	 * in the middle of a write to this file (i.e., we are writing
3906 	 * to this file using data from a mapped region of the file).
3907 	 */
3908 	if (rw_owner(&zp->z_map_lock) != curthread) {
3909 		rw_enter(&zp->z_map_lock, RW_WRITER);
3910 		need_unlock = TRUE;
3911 	}
3912 
3913 	/*
3914 	 * Loop through the requested range [off, off + len] looking
3915 	 * for pages.  If we don't find a page, we will need to create
3916 	 * a new page and fill it with data from the file.
3917 	 */
3918 	while (len > 0) {
3919 		if (plsz < PAGESIZE)
3920 			break;
3921 		if (pp = page_lookup(vp, off, SE_SHARED)) {
3922 			*pl++ = pp;
3923 			off += PAGESIZE;
3924 			addr += PAGESIZE;
3925 			len -= PAGESIZE;
3926 			plsz -= PAGESIZE;
3927 		} else {
3928 			err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw);
3929 			if (err)
3930 				goto out;
3931 			/*
3932 			 * klustering may have changed our region
3933 			 * to be block aligned.
3934 			 */
3935 			if (((pp = *pl) != 0) && (off != pp->p_offset)) {
3936 				int delta = off - pp->p_offset;
3937 				len += delta;
3938 				off -= delta;
3939 				addr -= delta;
3940 			}
3941 			while (*pl) {
3942 				pl++;
3943 				off += PAGESIZE;
3944 				addr += PAGESIZE;
3945 				plsz -= PAGESIZE;
3946 				if (len > PAGESIZE)
3947 					len -= PAGESIZE;
3948 				else
3949 					len = 0;
3950 			}
3951 		}
3952 	}
3953 
3954 	/*
3955 	 * Fill out the page array with any pages already in the cache.
3956 	 */
3957 	while (plsz > 0) {
3958 		pp = page_lookup_nowait(vp, off, SE_SHARED);
3959 		if (pp == NULL)
3960 			break;
3961 		*pl++ = pp;
3962 		off += PAGESIZE;
3963 		plsz -= PAGESIZE;
3964 	}
3965 
3966 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3967 out:
3968 	/*
3969 	 * We can't grab the range lock for the page as reader which would
3970 	 * stop truncation as this leads to deadlock. So we need to recheck
3971 	 * the file size.
3972 	 */
3973 	if (orig_off >= zp->z_phys->zp_size)
3974 		err = EFAULT;
3975 	if (err) {
3976 		/*
3977 		 * Release any pages we have previously locked.
3978 		 */
3979 		while (pl > pl0)
3980 			page_unlock(*--pl);
3981 	}
3982 
3983 	*pl = NULL;
3984 
3985 	if (need_unlock)
3986 		rw_exit(&zp->z_map_lock);
3987 
3988 	ZFS_EXIT(zfsvfs);
3989 	return (err);
3990 }
3991 
3992 /*
3993  * Request a memory map for a section of a file.  This code interacts
3994  * with common code and the VM system as follows:
3995  *
3996  *	common code calls mmap(), which ends up in smmap_common()
3997  *
3998  *	this calls VOP_MAP(), which takes you into (say) zfs
3999  *
4000  *	zfs_map() calls as_map(), passing segvn_create() as the callback
4001  *
4002  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
4003  *
4004  *	zfs_addmap() updates z_mapcnt
4005  */
4006 /*ARGSUSED*/
4007 static int
4008 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4009     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4010     caller_context_t *ct)
4011 {
4012 	znode_t *zp = VTOZ(vp);
4013 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4014 	segvn_crargs_t	vn_a;
4015 	int		error;
4016 
4017 	if ((prot & PROT_WRITE) &&
4018 	    (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY |
4019 	    ZFS_APPENDONLY)))
4020 		return (EPERM);
4021 
4022 	ZFS_ENTER(zfsvfs);
4023 	ZFS_VERIFY_ZP(zp);
4024 
4025 	if (vp->v_flag & VNOMAP) {
4026 		ZFS_EXIT(zfsvfs);
4027 		return (ENOSYS);
4028 	}
4029 
4030 	if (off < 0 || len > MAXOFFSET_T - off) {
4031 		ZFS_EXIT(zfsvfs);
4032 		return (ENXIO);
4033 	}
4034 
4035 	if (vp->v_type != VREG) {
4036 		ZFS_EXIT(zfsvfs);
4037 		return (ENODEV);
4038 	}
4039 
4040 	/*
4041 	 * If file is locked, disallow mapping.
4042 	 */
4043 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) {
4044 		ZFS_EXIT(zfsvfs);
4045 		return (EAGAIN);
4046 	}
4047 
4048 	as_rangelock(as);
4049 	if ((flags & MAP_FIXED) == 0) {
4050 		map_addr(addrp, len, off, 1, flags);
4051 		if (*addrp == NULL) {
4052 			as_rangeunlock(as);
4053 			ZFS_EXIT(zfsvfs);
4054 			return (ENOMEM);
4055 		}
4056 	} else {
4057 		/*
4058 		 * User specified address - blow away any previous mappings
4059 		 */
4060 		(void) as_unmap(as, *addrp, len);
4061 	}
4062 
4063 	vn_a.vp = vp;
4064 	vn_a.offset = (u_offset_t)off;
4065 	vn_a.type = flags & MAP_TYPE;
4066 	vn_a.prot = prot;
4067 	vn_a.maxprot = maxprot;
4068 	vn_a.cred = cr;
4069 	vn_a.amp = NULL;
4070 	vn_a.flags = flags & ~MAP_TYPE;
4071 	vn_a.szc = 0;
4072 	vn_a.lgrp_mem_policy_flags = 0;
4073 
4074 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4075 
4076 	as_rangeunlock(as);
4077 	ZFS_EXIT(zfsvfs);
4078 	return (error);
4079 }
4080 
4081 /* ARGSUSED */
4082 static int
4083 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4084     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4085     caller_context_t *ct)
4086 {
4087 	uint64_t pages = btopr(len);
4088 
4089 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4090 	return (0);
4091 }
4092 
4093 /*
4094  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4095  * more accurate mtime for the associated file.  Since we don't have a way of
4096  * detecting when the data was actually modified, we have to resort to
4097  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4098  * last page is pushed.  The problem occurs when the msync() call is omitted,
4099  * which by far the most common case:
4100  *
4101  * 	open()
4102  * 	mmap()
4103  * 	<modify memory>
4104  * 	munmap()
4105  * 	close()
4106  * 	<time lapse>
4107  * 	putpage() via fsflush
4108  *
4109  * If we wait until fsflush to come along, we can have a modification time that
4110  * is some arbitrary point in the future.  In order to prevent this in the
4111  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4112  * torn down.
4113  */
4114 /* ARGSUSED */
4115 static int
4116 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4117     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4118     caller_context_t *ct)
4119 {
4120 	uint64_t pages = btopr(len);
4121 
4122 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4123 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4124 
4125 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4126 	    vn_has_cached_data(vp))
4127 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4128 
4129 	return (0);
4130 }
4131 
4132 /*
4133  * Free or allocate space in a file.  Currently, this function only
4134  * supports the `F_FREESP' command.  However, this command is somewhat
4135  * misnamed, as its functionality includes the ability to allocate as
4136  * well as free space.
4137  *
4138  *	IN:	vp	- vnode of file to free data in.
4139  *		cmd	- action to take (only F_FREESP supported).
4140  *		bfp	- section of file to free/alloc.
4141  *		flag	- current file open mode flags.
4142  *		offset	- current file offset.
4143  *		cr	- credentials of caller [UNUSED].
4144  *		ct	- caller context.
4145  *
4146  *	RETURN:	0 if success
4147  *		error code if failure
4148  *
4149  * Timestamps:
4150  *	vp - ctime|mtime updated
4151  */
4152 /* ARGSUSED */
4153 static int
4154 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4155     offset_t offset, cred_t *cr, caller_context_t *ct)
4156 {
4157 	znode_t		*zp = VTOZ(vp);
4158 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4159 	uint64_t	off, len;
4160 	int		error;
4161 
4162 	ZFS_ENTER(zfsvfs);
4163 	ZFS_VERIFY_ZP(zp);
4164 
4165 top:
4166 	if (cmd != F_FREESP) {
4167 		ZFS_EXIT(zfsvfs);
4168 		return (EINVAL);
4169 	}
4170 
4171 	if (error = convoff(vp, bfp, 0, offset)) {
4172 		ZFS_EXIT(zfsvfs);
4173 		return (error);
4174 	}
4175 
4176 	if (bfp->l_len < 0) {
4177 		ZFS_EXIT(zfsvfs);
4178 		return (EINVAL);
4179 	}
4180 
4181 	off = bfp->l_start;
4182 	len = bfp->l_len; /* 0 means from off to end of file */
4183 
4184 	do {
4185 		error = zfs_freesp(zp, off, len, flag, TRUE);
4186 		/* NB: we already did dmu_tx_wait() if necessary */
4187 	} while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
4188 
4189 	ZFS_EXIT(zfsvfs);
4190 	return (error);
4191 }
4192 
4193 /*ARGSUSED*/
4194 static int
4195 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4196 {
4197 	znode_t		*zp = VTOZ(vp);
4198 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4199 	uint32_t	gen;
4200 	uint64_t	object = zp->z_id;
4201 	zfid_short_t	*zfid;
4202 	int		size, i;
4203 
4204 	ZFS_ENTER(zfsvfs);
4205 	ZFS_VERIFY_ZP(zp);
4206 	gen = (uint32_t)zp->z_gen;
4207 
4208 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4209 	if (fidp->fid_len < size) {
4210 		fidp->fid_len = size;
4211 		ZFS_EXIT(zfsvfs);
4212 		return (ENOSPC);
4213 	}
4214 
4215 	zfid = (zfid_short_t *)fidp;
4216 
4217 	zfid->zf_len = size;
4218 
4219 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4220 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4221 
4222 	/* Must have a non-zero generation number to distinguish from .zfs */
4223 	if (gen == 0)
4224 		gen = 1;
4225 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4226 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4227 
4228 	if (size == LONG_FID_LEN) {
4229 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4230 		zfid_long_t	*zlfid;
4231 
4232 		zlfid = (zfid_long_t *)fidp;
4233 
4234 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4235 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4236 
4237 		/* XXX - this should be the generation number for the objset */
4238 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4239 			zlfid->zf_setgen[i] = 0;
4240 	}
4241 
4242 	ZFS_EXIT(zfsvfs);
4243 	return (0);
4244 }
4245 
4246 static int
4247 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4248     caller_context_t *ct)
4249 {
4250 	znode_t		*zp, *xzp;
4251 	zfsvfs_t	*zfsvfs;
4252 	zfs_dirlock_t	*dl;
4253 	int		error;
4254 
4255 	switch (cmd) {
4256 	case _PC_LINK_MAX:
4257 		*valp = ULONG_MAX;
4258 		return (0);
4259 
4260 	case _PC_FILESIZEBITS:
4261 		*valp = 64;
4262 		return (0);
4263 
4264 	case _PC_XATTR_EXISTS:
4265 		zp = VTOZ(vp);
4266 		zfsvfs = zp->z_zfsvfs;
4267 		ZFS_ENTER(zfsvfs);
4268 		ZFS_VERIFY_ZP(zp);
4269 		*valp = 0;
4270 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4271 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4272 		if (error == 0) {
4273 			zfs_dirent_unlock(dl);
4274 			if (!zfs_dirempty(xzp))
4275 				*valp = 1;
4276 			VN_RELE(ZTOV(xzp));
4277 		} else if (error == ENOENT) {
4278 			/*
4279 			 * If there aren't extended attributes, it's the
4280 			 * same as having zero of them.
4281 			 */
4282 			error = 0;
4283 		}
4284 		ZFS_EXIT(zfsvfs);
4285 		return (error);
4286 
4287 	case _PC_SATTR_ENABLED:
4288 	case _PC_SATTR_EXISTS:
4289 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) &&
4290 		    (vp->v_type == VREG || vp->v_type == VDIR);
4291 		return (0);
4292 
4293 	case _PC_ACL_ENABLED:
4294 		*valp = _ACL_ACE_ENABLED;
4295 		return (0);
4296 
4297 	case _PC_MIN_HOLE_SIZE:
4298 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4299 		return (0);
4300 
4301 	default:
4302 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4303 	}
4304 }
4305 
4306 /*ARGSUSED*/
4307 static int
4308 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4309     caller_context_t *ct)
4310 {
4311 	znode_t *zp = VTOZ(vp);
4312 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4313 	int error;
4314 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4315 
4316 	ZFS_ENTER(zfsvfs);
4317 	ZFS_VERIFY_ZP(zp);
4318 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4319 	ZFS_EXIT(zfsvfs);
4320 
4321 	return (error);
4322 }
4323 
4324 /*ARGSUSED*/
4325 static int
4326 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4327     caller_context_t *ct)
4328 {
4329 	znode_t *zp = VTOZ(vp);
4330 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4331 	int error;
4332 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4333 
4334 	ZFS_ENTER(zfsvfs);
4335 	ZFS_VERIFY_ZP(zp);
4336 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4337 	ZFS_EXIT(zfsvfs);
4338 	return (error);
4339 }
4340 
4341 /*
4342  * Predeclare these here so that the compiler assumes that
4343  * this is an "old style" function declaration that does
4344  * not include arguments => we won't get type mismatch errors
4345  * in the initializations that follow.
4346  */
4347 static int zfs_inval();
4348 static int zfs_isdir();
4349 
4350 static int
4351 zfs_inval()
4352 {
4353 	return (EINVAL);
4354 }
4355 
4356 static int
4357 zfs_isdir()
4358 {
4359 	return (EISDIR);
4360 }
4361 /*
4362  * Directory vnode operations template
4363  */
4364 vnodeops_t *zfs_dvnodeops;
4365 const fs_operation_def_t zfs_dvnodeops_template[] = {
4366 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4367 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4368 	VOPNAME_READ,		{ .error = zfs_isdir },
4369 	VOPNAME_WRITE,		{ .error = zfs_isdir },
4370 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4371 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4372 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4373 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4374 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4375 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4376 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4377 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4378 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4379 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
4380 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4381 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4382 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
4383 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4384 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4385 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4386 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4387 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4388 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4389 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4390 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
4391 	NULL,			NULL
4392 };
4393 
4394 /*
4395  * Regular file vnode operations template
4396  */
4397 vnodeops_t *zfs_fvnodeops;
4398 const fs_operation_def_t zfs_fvnodeops_template[] = {
4399 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4400 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4401 	VOPNAME_READ,		{ .vop_read = zfs_read },
4402 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
4403 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4404 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4405 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4406 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4407 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4408 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4409 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4410 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4411 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4412 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4413 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
4414 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
4415 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
4416 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
4417 	VOPNAME_MAP,		{ .vop_map = zfs_map },
4418 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
4419 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
4420 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4421 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4422 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4423 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4424 	NULL,			NULL
4425 };
4426 
4427 /*
4428  * Symbolic link vnode operations template
4429  */
4430 vnodeops_t *zfs_symvnodeops;
4431 const fs_operation_def_t zfs_symvnodeops_template[] = {
4432 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4433 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4434 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4435 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4436 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
4437 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4438 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4439 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4440 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4441 	NULL,			NULL
4442 };
4443 
4444 /*
4445  * Extended attribute directory vnode operations template
4446  *	This template is identical to the directory vnodes
4447  *	operation template except for restricted operations:
4448  *		VOP_MKDIR()
4449  *		VOP_SYMLINK()
4450  * Note that there are other restrictions embedded in:
4451  *	zfs_create()	- restrict type to VREG
4452  *	zfs_link()	- no links into/out of attribute space
4453  *	zfs_rename()	- no moves into/out of attribute space
4454  */
4455 vnodeops_t *zfs_xdvnodeops;
4456 const fs_operation_def_t zfs_xdvnodeops_template[] = {
4457 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4458 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4459 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4460 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4461 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4462 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4463 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4464 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4465 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4466 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4467 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4468 	VOPNAME_MKDIR,		{ .error = zfs_inval },
4469 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4470 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4471 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
4472 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4473 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4474 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4475 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4476 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4477 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4478 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4479 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4480 	NULL,			NULL
4481 };
4482 
4483 /*
4484  * Error vnode operations template
4485  */
4486 vnodeops_t *zfs_evnodeops;
4487 const fs_operation_def_t zfs_evnodeops_template[] = {
4488 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4489 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4490 	NULL,			NULL
4491 };
4492