xref: /titanic_51/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 52ccf843e173e2a4a657360b0a22853fd413905f)
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 2008 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 #include <sys/attr.h>
77 
78 /*
79  * Programming rules.
80  *
81  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
82  * properly lock its in-core state, create a DMU transaction, do the work,
83  * record this work in the intent log (ZIL), commit the DMU transaction,
84  * and wait for the intent log to commit if it is a synchronous operation.
85  * Moreover, the vnode ops must work in both normal and log replay context.
86  * The ordering of events is important to avoid deadlocks and references
87  * to freed memory.  The example below illustrates the following Big Rules:
88  *
89  *  (1) A check must be made in each zfs thread for a mounted file system.
90  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
91  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
92  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
93  *      can return EIO from the calling function.
94  *
95  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
96  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
97  *	First, if it's the last reference, the vnode/znode
98  *	can be freed, so the zp may point to freed memory.  Second, the last
99  *	reference will call zfs_zinactive(), which may induce a lot of work --
100  *	pushing cached pages (which acquires range locks) and syncing out
101  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
102  *	which could deadlock the system if you were already holding one.
103  *
104  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
105  *	as they can span dmu_tx_assign() calls.
106  *
107  *  (4)	Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign().
108  *	In normal operation, this will be TXG_NOWAIT.  During ZIL replay,
109  *	it will be a specific txg.  Either way, dmu_tx_assign() never blocks.
110  *	This is critical because we don't want to block while holding locks.
111  *	Note, in particular, that if a lock is sometimes acquired before
112  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
113  *	use a non-blocking assign can deadlock the system.  The scenario:
114  *
115  *	Thread A has grabbed a lock before calling dmu_tx_assign().
116  *	Thread B is in an already-assigned tx, and blocks for this lock.
117  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
118  *	forever, because the previous txg can't quiesce until B's tx commits.
119  *
120  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
121  *	then drop all locks, call dmu_tx_wait(), and try again.
122  *
123  *  (5)	If the operation succeeded, generate the intent log entry for it
124  *	before dropping locks.  This ensures that the ordering of events
125  *	in the intent log matches the order in which they actually occurred.
126  *
127  *  (6)	At the end of each vnode op, the DMU tx must always commit,
128  *	regardless of whether there were any errors.
129  *
130  *  (7)	After dropping all locks, invoke zil_commit(zilog, seq, foid)
131  *	to ensure that synchronous semantics are provided when necessary.
132  *
133  * In general, this is how things should be ordered in each vnode op:
134  *
135  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
136  * top:
137  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
138  *	rw_enter(...);			// grab any other locks you need
139  *	tx = dmu_tx_create(...);	// get DMU tx
140  *	dmu_tx_hold_*();		// hold each object you might modify
141  *	error = dmu_tx_assign(tx, zfsvfs->z_assign);	// try to assign
142  *	if (error) {
143  *		rw_exit(...);		// drop locks
144  *		zfs_dirent_unlock(dl);	// unlock directory entry
145  *		VN_RELE(...);		// release held vnodes
146  *		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
147  *			dmu_tx_wait(tx);
148  *			dmu_tx_abort(tx);
149  *			goto top;
150  *		}
151  *		dmu_tx_abort(tx);	// abort DMU tx
152  *		ZFS_EXIT(zfsvfs);	// finished in zfs
153  *		return (error);		// really out of space
154  *	}
155  *	error = do_real_work();		// do whatever this VOP does
156  *	if (error == 0)
157  *		zfs_log_*(...);		// on success, make ZIL entry
158  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
159  *	rw_exit(...);			// drop locks
160  *	zfs_dirent_unlock(dl);		// unlock directory entry
161  *	VN_RELE(...);			// release held vnodes
162  *	zil_commit(zilog, seq, foid);	// synchronous when necessary
163  *	ZFS_EXIT(zfsvfs);		// finished in zfs
164  *	return (error);			// done, report error
165  */
166 
167 /* ARGSUSED */
168 static int
169 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
170 {
171 	znode_t	*zp = VTOZ(*vpp);
172 
173 	if ((flag & FWRITE) && (zp->z_phys->zp_flags & ZFS_APPENDONLY) &&
174 	    ((flag & FAPPEND) == 0)) {
175 		return (EPERM);
176 	}
177 
178 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
179 	    ZTOV(zp)->v_type == VREG &&
180 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
181 	    zp->z_phys->zp_size > 0)
182 		if (fs_vscan(*vpp, cr, 0) != 0)
183 			return (EACCES);
184 
185 	/* Keep a count of the synchronous opens in the znode */
186 	if (flag & (FSYNC | FDSYNC))
187 		atomic_inc_32(&zp->z_sync_cnt);
188 
189 	return (0);
190 }
191 
192 /* ARGSUSED */
193 static int
194 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
195     caller_context_t *ct)
196 {
197 	znode_t	*zp = VTOZ(vp);
198 
199 	/* Decrement the synchronous opens in the znode */
200 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
201 		atomic_dec_32(&zp->z_sync_cnt);
202 
203 	/*
204 	 * Clean up any locks held by this process on the vp.
205 	 */
206 	cleanlocks(vp, ddi_get_pid(), 0);
207 	cleanshares(vp, ddi_get_pid());
208 
209 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
210 	    ZTOV(zp)->v_type == VREG &&
211 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
212 	    zp->z_phys->zp_size > 0)
213 		VERIFY(fs_vscan(vp, cr, 1) == 0);
214 
215 	return (0);
216 }
217 
218 /*
219  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
220  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
221  */
222 static int
223 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
224 {
225 	znode_t	*zp = VTOZ(vp);
226 	uint64_t noff = (uint64_t)*off; /* new offset */
227 	uint64_t file_sz;
228 	int error;
229 	boolean_t hole;
230 
231 	file_sz = zp->z_phys->zp_size;
232 	if (noff >= file_sz)  {
233 		return (ENXIO);
234 	}
235 
236 	if (cmd == _FIO_SEEK_HOLE)
237 		hole = B_TRUE;
238 	else
239 		hole = B_FALSE;
240 
241 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
242 
243 	/* end of file? */
244 	if ((error == ESRCH) || (noff > file_sz)) {
245 		/*
246 		 * Handle the virtual hole at the end of file.
247 		 */
248 		if (hole) {
249 			*off = file_sz;
250 			return (0);
251 		}
252 		return (ENXIO);
253 	}
254 
255 	if (noff < *off)
256 		return (error);
257 	*off = noff;
258 	return (error);
259 }
260 
261 /* ARGSUSED */
262 static int
263 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
264     int *rvalp, caller_context_t *ct)
265 {
266 	offset_t off;
267 	int error;
268 	zfsvfs_t *zfsvfs;
269 	znode_t *zp;
270 
271 	switch (com) {
272 	case _FIOFFS:
273 		return (zfs_sync(vp->v_vfsp, 0, cred));
274 
275 		/*
276 		 * The following two ioctls are used by bfu.  Faking out,
277 		 * necessary to avoid bfu errors.
278 		 */
279 	case _FIOGDIO:
280 	case _FIOSDIO:
281 		return (0);
282 
283 	case _FIO_SEEK_DATA:
284 	case _FIO_SEEK_HOLE:
285 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
286 			return (EFAULT);
287 
288 		zp = VTOZ(vp);
289 		zfsvfs = zp->z_zfsvfs;
290 		ZFS_ENTER(zfsvfs);
291 		ZFS_VERIFY_ZP(zp);
292 
293 		/* offset parameter is in/out */
294 		error = zfs_holey(vp, com, &off);
295 		ZFS_EXIT(zfsvfs);
296 		if (error)
297 			return (error);
298 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
299 			return (EFAULT);
300 		return (0);
301 	}
302 	return (ENOTTY);
303 }
304 
305 /*
306  * When a file is memory mapped, we must keep the IO data synchronized
307  * between the DMU cache and the memory mapped pages.  What this means:
308  *
309  * On Write:	If we find a memory mapped page, we write to *both*
310  *		the page and the dmu buffer.
311  *
312  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
313  *	the file is memory mapped.
314  */
315 static int
316 mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx)
317 {
318 	znode_t	*zp = VTOZ(vp);
319 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
320 	int64_t	start, off;
321 	int len = nbytes;
322 	int error = 0;
323 
324 	start = uio->uio_loffset;
325 	off = start & PAGEOFFSET;
326 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
327 		page_t *pp;
328 		uint64_t bytes = MIN(PAGESIZE - off, len);
329 		uint64_t woff = uio->uio_loffset;
330 
331 		/*
332 		 * We don't want a new page to "appear" in the middle of
333 		 * the file update (because it may not get the write
334 		 * update data), so we grab a lock to block
335 		 * zfs_getpage().
336 		 */
337 		rw_enter(&zp->z_map_lock, RW_WRITER);
338 		if (pp = page_lookup(vp, start, SE_SHARED)) {
339 			caddr_t va;
340 
341 			rw_exit(&zp->z_map_lock);
342 			va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L);
343 			error = uiomove(va+off, bytes, UIO_WRITE, uio);
344 			if (error == 0) {
345 				dmu_write(zfsvfs->z_os, zp->z_id,
346 				    woff, bytes, va+off, tx);
347 			}
348 			ppmapout(va);
349 			page_unlock(pp);
350 		} else {
351 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
352 			    uio, bytes, tx);
353 			rw_exit(&zp->z_map_lock);
354 		}
355 		len -= bytes;
356 		off = 0;
357 		if (error)
358 			break;
359 	}
360 	return (error);
361 }
362 
363 /*
364  * When a file is memory mapped, we must keep the IO data synchronized
365  * between the DMU cache and the memory mapped pages.  What this means:
366  *
367  * On Read:	We "read" preferentially from memory mapped pages,
368  *		else we default from the dmu buffer.
369  *
370  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
371  *	the file is memory mapped.
372  */
373 static int
374 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
375 {
376 	znode_t *zp = VTOZ(vp);
377 	objset_t *os = zp->z_zfsvfs->z_os;
378 	int64_t	start, off;
379 	int len = nbytes;
380 	int error = 0;
381 
382 	start = uio->uio_loffset;
383 	off = start & PAGEOFFSET;
384 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
385 		page_t *pp;
386 		uint64_t bytes = MIN(PAGESIZE - off, len);
387 
388 		if (pp = page_lookup(vp, start, SE_SHARED)) {
389 			caddr_t va;
390 
391 			va = ppmapin(pp, PROT_READ, (caddr_t)-1L);
392 			error = uiomove(va + off, bytes, UIO_READ, uio);
393 			ppmapout(va);
394 			page_unlock(pp);
395 		} else {
396 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
397 		}
398 		len -= bytes;
399 		off = 0;
400 		if (error)
401 			break;
402 	}
403 	return (error);
404 }
405 
406 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
407 
408 /*
409  * Read bytes from specified file into supplied buffer.
410  *
411  *	IN:	vp	- vnode of file to be read from.
412  *		uio	- structure supplying read location, range info,
413  *			  and return buffer.
414  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
415  *		cr	- credentials of caller.
416  *		ct	- caller context
417  *
418  *	OUT:	uio	- updated offset and range, buffer filled.
419  *
420  *	RETURN:	0 if success
421  *		error code if failure
422  *
423  * Side Effects:
424  *	vp - atime updated if byte count > 0
425  */
426 /* ARGSUSED */
427 static int
428 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
429 {
430 	znode_t		*zp = VTOZ(vp);
431 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
432 	objset_t	*os;
433 	ssize_t		n, nbytes;
434 	int		error;
435 	rl_t		*rl;
436 
437 	ZFS_ENTER(zfsvfs);
438 	ZFS_VERIFY_ZP(zp);
439 	os = zfsvfs->z_os;
440 
441 	if (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) {
442 		ZFS_EXIT(zfsvfs);
443 		return (EACCES);
444 	}
445 
446 	/*
447 	 * Validate file offset
448 	 */
449 	if (uio->uio_loffset < (offset_t)0) {
450 		ZFS_EXIT(zfsvfs);
451 		return (EINVAL);
452 	}
453 
454 	/*
455 	 * Fasttrack empty reads
456 	 */
457 	if (uio->uio_resid == 0) {
458 		ZFS_EXIT(zfsvfs);
459 		return (0);
460 	}
461 
462 	/*
463 	 * Check for mandatory locks
464 	 */
465 	if (MANDMODE((mode_t)zp->z_phys->zp_mode)) {
466 		if (error = chklock(vp, FREAD,
467 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
468 			ZFS_EXIT(zfsvfs);
469 			return (error);
470 		}
471 	}
472 
473 	/*
474 	 * If we're in FRSYNC mode, sync out this znode before reading it.
475 	 */
476 	if (ioflag & FRSYNC)
477 		zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
478 
479 	/*
480 	 * Lock the range against changes.
481 	 */
482 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
483 
484 	/*
485 	 * If we are reading past end-of-file we can skip
486 	 * to the end; but we might still need to set atime.
487 	 */
488 	if (uio->uio_loffset >= zp->z_phys->zp_size) {
489 		error = 0;
490 		goto out;
491 	}
492 
493 	ASSERT(uio->uio_loffset < zp->z_phys->zp_size);
494 	n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset);
495 
496 	while (n > 0) {
497 		nbytes = MIN(n, zfs_read_chunk_size -
498 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
499 
500 		if (vn_has_cached_data(vp))
501 			error = mappedread(vp, nbytes, uio);
502 		else
503 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
504 		if (error)
505 			break;
506 
507 		n -= nbytes;
508 	}
509 
510 out:
511 	zfs_range_unlock(rl);
512 
513 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
514 	ZFS_EXIT(zfsvfs);
515 	return (error);
516 }
517 
518 /*
519  * Fault in the pages of the first n bytes specified by the uio structure.
520  * 1 byte in each page is touched and the uio struct is unmodified.
521  * Any error will exit this routine as this is only a best
522  * attempt to get the pages resident. This is a copy of ufs_trans_touch().
523  */
524 static void
525 zfs_prefault_write(ssize_t n, struct uio *uio)
526 {
527 	struct iovec *iov;
528 	ulong_t cnt, incr;
529 	caddr_t p;
530 	uint8_t tmp;
531 
532 	iov = uio->uio_iov;
533 
534 	while (n) {
535 		cnt = MIN(iov->iov_len, n);
536 		if (cnt == 0) {
537 			/* empty iov entry */
538 			iov++;
539 			continue;
540 		}
541 		n -= cnt;
542 		/*
543 		 * touch each page in this segment.
544 		 */
545 		p = iov->iov_base;
546 		while (cnt) {
547 			switch (uio->uio_segflg) {
548 			case UIO_USERSPACE:
549 			case UIO_USERISPACE:
550 				if (fuword8(p, &tmp))
551 					return;
552 				break;
553 			case UIO_SYSSPACE:
554 				if (kcopy(p, &tmp, 1))
555 					return;
556 				break;
557 			}
558 			incr = MIN(cnt, PAGESIZE);
559 			p += incr;
560 			cnt -= incr;
561 		}
562 		/*
563 		 * touch the last byte in case it straddles a page.
564 		 */
565 		p--;
566 		switch (uio->uio_segflg) {
567 		case UIO_USERSPACE:
568 		case UIO_USERISPACE:
569 			if (fuword8(p, &tmp))
570 				return;
571 			break;
572 		case UIO_SYSSPACE:
573 			if (kcopy(p, &tmp, 1))
574 				return;
575 			break;
576 		}
577 		iov++;
578 	}
579 }
580 
581 /*
582  * Write the bytes to a file.
583  *
584  *	IN:	vp	- vnode of file to be written to.
585  *		uio	- structure supplying write location, range info,
586  *			  and data buffer.
587  *		ioflag	- FAPPEND flag set if in append mode.
588  *		cr	- credentials of caller.
589  *		ct	- caller context (NFS/CIFS fem monitor only)
590  *
591  *	OUT:	uio	- updated offset and range.
592  *
593  *	RETURN:	0 if success
594  *		error code if failure
595  *
596  * Timestamps:
597  *	vp - ctime|mtime updated if byte count > 0
598  */
599 /* ARGSUSED */
600 static int
601 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
602 {
603 	znode_t		*zp = VTOZ(vp);
604 	rlim64_t	limit = uio->uio_llimit;
605 	ssize_t		start_resid = uio->uio_resid;
606 	ssize_t		tx_bytes;
607 	uint64_t	end_size;
608 	dmu_tx_t	*tx;
609 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
610 	zilog_t		*zilog;
611 	offset_t	woff;
612 	ssize_t		n, nbytes;
613 	rl_t		*rl;
614 	int		max_blksz = zfsvfs->z_max_blksz;
615 	uint64_t	pflags;
616 	int		error;
617 
618 	/*
619 	 * Fasttrack empty write
620 	 */
621 	n = start_resid;
622 	if (n == 0)
623 		return (0);
624 
625 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
626 		limit = MAXOFFSET_T;
627 
628 	ZFS_ENTER(zfsvfs);
629 	ZFS_VERIFY_ZP(zp);
630 
631 	/*
632 	 * If immutable or not appending then return EPERM
633 	 */
634 	pflags = zp->z_phys->zp_flags;
635 	if ((pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
636 	    ((pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
637 	    (uio->uio_loffset < zp->z_phys->zp_size))) {
638 		ZFS_EXIT(zfsvfs);
639 		return (EPERM);
640 	}
641 
642 	zilog = zfsvfs->z_log;
643 
644 	/*
645 	 * Pre-fault the pages to ensure slow (eg NFS) pages
646 	 * don't hold up txg.
647 	 */
648 	zfs_prefault_write(n, uio);
649 
650 	/*
651 	 * If in append mode, set the io offset pointer to eof.
652 	 */
653 	if (ioflag & FAPPEND) {
654 		/*
655 		 * Range lock for a file append:
656 		 * The value for the start of range will be determined by
657 		 * zfs_range_lock() (to guarantee append semantics).
658 		 * If this write will cause the block size to increase,
659 		 * zfs_range_lock() will lock the entire file, so we must
660 		 * later reduce the range after we grow the block size.
661 		 */
662 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
663 		if (rl->r_len == UINT64_MAX) {
664 			/* overlocked, zp_size can't change */
665 			woff = uio->uio_loffset = zp->z_phys->zp_size;
666 		} else {
667 			woff = uio->uio_loffset = rl->r_off;
668 		}
669 	} else {
670 		woff = uio->uio_loffset;
671 		/*
672 		 * Validate file offset
673 		 */
674 		if (woff < 0) {
675 			ZFS_EXIT(zfsvfs);
676 			return (EINVAL);
677 		}
678 
679 		/*
680 		 * If we need to grow the block size then zfs_range_lock()
681 		 * will lock a wider range than we request here.
682 		 * Later after growing the block size we reduce the range.
683 		 */
684 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
685 	}
686 
687 	if (woff >= limit) {
688 		zfs_range_unlock(rl);
689 		ZFS_EXIT(zfsvfs);
690 		return (EFBIG);
691 	}
692 
693 	if ((woff + n) > limit || woff > (limit - n))
694 		n = limit - woff;
695 
696 	/*
697 	 * Check for mandatory locks
698 	 */
699 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) &&
700 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
701 		zfs_range_unlock(rl);
702 		ZFS_EXIT(zfsvfs);
703 		return (error);
704 	}
705 	end_size = MAX(zp->z_phys->zp_size, woff + n);
706 
707 	/*
708 	 * Write the file in reasonable size chunks.  Each chunk is written
709 	 * in a separate transaction; this keeps the intent log records small
710 	 * and allows us to do more fine-grained space accounting.
711 	 */
712 	while (n > 0) {
713 		/*
714 		 * Start a transaction.
715 		 */
716 		woff = uio->uio_loffset;
717 		tx = dmu_tx_create(zfsvfs->z_os);
718 		dmu_tx_hold_bonus(tx, zp->z_id);
719 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
720 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
721 		if (error) {
722 			if (error == ERESTART &&
723 			    zfsvfs->z_assign == TXG_NOWAIT) {
724 				dmu_tx_wait(tx);
725 				dmu_tx_abort(tx);
726 				continue;
727 			}
728 			dmu_tx_abort(tx);
729 			break;
730 		}
731 
732 		/*
733 		 * If zfs_range_lock() over-locked we grow the blocksize
734 		 * and then reduce the lock range.  This will only happen
735 		 * on the first iteration since zfs_range_reduce() will
736 		 * shrink down r_len to the appropriate size.
737 		 */
738 		if (rl->r_len == UINT64_MAX) {
739 			uint64_t new_blksz;
740 
741 			if (zp->z_blksz > max_blksz) {
742 				ASSERT(!ISP2(zp->z_blksz));
743 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
744 			} else {
745 				new_blksz = MIN(end_size, max_blksz);
746 			}
747 			zfs_grow_blocksize(zp, new_blksz, tx);
748 			zfs_range_reduce(rl, woff, n);
749 		}
750 
751 		/*
752 		 * XXX - should we really limit each write to z_max_blksz?
753 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
754 		 */
755 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
756 		rw_enter(&zp->z_map_lock, RW_READER);
757 
758 		tx_bytes = uio->uio_resid;
759 		if (vn_has_cached_data(vp)) {
760 			rw_exit(&zp->z_map_lock);
761 			error = mappedwrite(vp, nbytes, uio, tx);
762 		} else {
763 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
764 			    uio, nbytes, tx);
765 			rw_exit(&zp->z_map_lock);
766 		}
767 		tx_bytes -= uio->uio_resid;
768 
769 		/*
770 		 * If we made no progress, we're done.  If we made even
771 		 * partial progress, update the znode and ZIL accordingly.
772 		 */
773 		if (tx_bytes == 0) {
774 			dmu_tx_commit(tx);
775 			ASSERT(error != 0);
776 			break;
777 		}
778 
779 		/*
780 		 * Clear Set-UID/Set-GID bits on successful write if not
781 		 * privileged and at least one of the excute bits is set.
782 		 *
783 		 * It would be nice to to this after all writes have
784 		 * been done, but that would still expose the ISUID/ISGID
785 		 * to another app after the partial write is committed.
786 		 *
787 		 * Note: we don't call zfs_fuid_map_id() here because
788 		 * user 0 is not an ephemeral uid.
789 		 */
790 		mutex_enter(&zp->z_acl_lock);
791 		if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) |
792 		    (S_IXUSR >> 6))) != 0 &&
793 		    (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 &&
794 		    secpolicy_vnode_setid_retain(cr,
795 		    (zp->z_phys->zp_mode & S_ISUID) != 0 &&
796 		    zp->z_phys->zp_uid == 0) != 0) {
797 			zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID);
798 		}
799 		mutex_exit(&zp->z_acl_lock);
800 
801 		/*
802 		 * Update time stamp.  NOTE: This marks the bonus buffer as
803 		 * dirty, so we don't have to do it again for zp_size.
804 		 */
805 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
806 
807 		/*
808 		 * Update the file size (zp_size) if it has changed;
809 		 * account for possible concurrent updates.
810 		 */
811 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset)
812 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
813 			    uio->uio_loffset);
814 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
815 		dmu_tx_commit(tx);
816 
817 		if (error != 0)
818 			break;
819 		ASSERT(tx_bytes == nbytes);
820 		n -= nbytes;
821 	}
822 
823 	zfs_range_unlock(rl);
824 
825 	/*
826 	 * If we're in replay mode, or we made no progress, return error.
827 	 * Otherwise, it's at least a partial write, so it's successful.
828 	 */
829 	if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) {
830 		ZFS_EXIT(zfsvfs);
831 		return (error);
832 	}
833 
834 	if (ioflag & (FSYNC | FDSYNC))
835 		zil_commit(zilog, zp->z_last_itx, zp->z_id);
836 
837 	ZFS_EXIT(zfsvfs);
838 	return (0);
839 }
840 
841 void
842 zfs_get_done(dmu_buf_t *db, void *vzgd)
843 {
844 	zgd_t *zgd = (zgd_t *)vzgd;
845 	rl_t *rl = zgd->zgd_rl;
846 	vnode_t *vp = ZTOV(rl->r_zp);
847 
848 	dmu_buf_rele(db, vzgd);
849 	zfs_range_unlock(rl);
850 	VN_RELE(vp);
851 	zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
852 	kmem_free(zgd, sizeof (zgd_t));
853 }
854 
855 /*
856  * Get data to generate a TX_WRITE intent log record.
857  */
858 int
859 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
860 {
861 	zfsvfs_t *zfsvfs = arg;
862 	objset_t *os = zfsvfs->z_os;
863 	znode_t *zp;
864 	uint64_t off = lr->lr_offset;
865 	dmu_buf_t *db;
866 	rl_t *rl;
867 	zgd_t *zgd;
868 	int dlen = lr->lr_length;		/* length of user data */
869 	int error = 0;
870 
871 	ASSERT(zio);
872 	ASSERT(dlen != 0);
873 
874 	/*
875 	 * Nothing to do if the file has been removed
876 	 */
877 	if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0)
878 		return (ENOENT);
879 	if (zp->z_unlinked) {
880 		VN_RELE(ZTOV(zp));
881 		return (ENOENT);
882 	}
883 
884 	/*
885 	 * Write records come in two flavors: immediate and indirect.
886 	 * For small writes it's cheaper to store the data with the
887 	 * log record (immediate); for large writes it's cheaper to
888 	 * sync the data and get a pointer to it (indirect) so that
889 	 * we don't have to write the data twice.
890 	 */
891 	if (buf != NULL) { /* immediate write */
892 		rl = zfs_range_lock(zp, off, dlen, RL_READER);
893 		/* test for truncation needs to be done while range locked */
894 		if (off >= zp->z_phys->zp_size) {
895 			error = ENOENT;
896 			goto out;
897 		}
898 		VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf));
899 	} else { /* indirect write */
900 		uint64_t boff; /* block starting offset */
901 
902 		/*
903 		 * Have to lock the whole block to ensure when it's
904 		 * written out and it's checksum is being calculated
905 		 * that no one can change the data. We need to re-check
906 		 * blocksize after we get the lock in case it's changed!
907 		 */
908 		for (;;) {
909 			if (ISP2(zp->z_blksz)) {
910 				boff = P2ALIGN_TYPED(off, zp->z_blksz,
911 				    uint64_t);
912 			} else {
913 				boff = 0;
914 			}
915 			dlen = zp->z_blksz;
916 			rl = zfs_range_lock(zp, boff, dlen, RL_READER);
917 			if (zp->z_blksz == dlen)
918 				break;
919 			zfs_range_unlock(rl);
920 		}
921 		/* test for truncation needs to be done while range locked */
922 		if (off >= zp->z_phys->zp_size) {
923 			error = ENOENT;
924 			goto out;
925 		}
926 		zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
927 		zgd->zgd_rl = rl;
928 		zgd->zgd_zilog = zfsvfs->z_log;
929 		zgd->zgd_bp = &lr->lr_blkptr;
930 		VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db));
931 		ASSERT(boff == db->db_offset);
932 		lr->lr_blkoff = off - boff;
933 		error = dmu_sync(zio, db, &lr->lr_blkptr,
934 		    lr->lr_common.lrc_txg, zfs_get_done, zgd);
935 		ASSERT((error && error != EINPROGRESS) ||
936 		    lr->lr_length <= zp->z_blksz);
937 		if (error == 0)
938 			zil_add_block(zfsvfs->z_log, &lr->lr_blkptr);
939 		/*
940 		 * If we get EINPROGRESS, then we need to wait for a
941 		 * write IO initiated by dmu_sync() to complete before
942 		 * we can release this dbuf.  We will finish everything
943 		 * up in the zfs_get_done() callback.
944 		 */
945 		if (error == EINPROGRESS)
946 			return (0);
947 		dmu_buf_rele(db, zgd);
948 		kmem_free(zgd, sizeof (zgd_t));
949 	}
950 out:
951 	zfs_range_unlock(rl);
952 	VN_RELE(ZTOV(zp));
953 	return (error);
954 }
955 
956 /*ARGSUSED*/
957 static int
958 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
959     caller_context_t *ct)
960 {
961 	znode_t *zp = VTOZ(vp);
962 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
963 	int error;
964 
965 	ZFS_ENTER(zfsvfs);
966 	ZFS_VERIFY_ZP(zp);
967 
968 	if (flag & V_ACE_MASK)
969 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
970 	else
971 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
972 
973 	ZFS_EXIT(zfsvfs);
974 	return (error);
975 }
976 
977 /*
978  * Lookup an entry in a directory, or an extended attribute directory.
979  * If it exists, return a held vnode reference for it.
980  *
981  *	IN:	dvp	- vnode of directory to search.
982  *		nm	- name of entry to lookup.
983  *		pnp	- full pathname to lookup [UNUSED].
984  *		flags	- LOOKUP_XATTR set if looking for an attribute.
985  *		rdir	- root directory vnode [UNUSED].
986  *		cr	- credentials of caller.
987  *		ct	- caller context
988  *		direntflags - directory lookup flags
989  *		realpnp - returned pathname.
990  *
991  *	OUT:	vpp	- vnode of located entry, NULL if not found.
992  *
993  *	RETURN:	0 if success
994  *		error code if failure
995  *
996  * Timestamps:
997  *	NA
998  */
999 /* ARGSUSED */
1000 static int
1001 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1002     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1003     int *direntflags, pathname_t *realpnp)
1004 {
1005 	znode_t *zdp = VTOZ(dvp);
1006 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1007 	int	error;
1008 
1009 	ZFS_ENTER(zfsvfs);
1010 	ZFS_VERIFY_ZP(zdp);
1011 
1012 	*vpp = NULL;
1013 
1014 	if (flags & LOOKUP_XATTR) {
1015 		/*
1016 		 * If the xattr property is off, refuse the lookup request.
1017 		 */
1018 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1019 			ZFS_EXIT(zfsvfs);
1020 			return (EINVAL);
1021 		}
1022 
1023 		/*
1024 		 * We don't allow recursive attributes..
1025 		 * Maybe someday we will.
1026 		 */
1027 		if (zdp->z_phys->zp_flags & ZFS_XATTR) {
1028 			ZFS_EXIT(zfsvfs);
1029 			return (EINVAL);
1030 		}
1031 
1032 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1033 			ZFS_EXIT(zfsvfs);
1034 			return (error);
1035 		}
1036 
1037 		/*
1038 		 * Do we have permission to get into attribute directory?
1039 		 */
1040 
1041 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1042 		    B_FALSE, cr)) {
1043 			VN_RELE(*vpp);
1044 			*vpp = NULL;
1045 		}
1046 
1047 		ZFS_EXIT(zfsvfs);
1048 		return (error);
1049 	}
1050 
1051 	if (dvp->v_type != VDIR) {
1052 		ZFS_EXIT(zfsvfs);
1053 		return (ENOTDIR);
1054 	}
1055 
1056 	/*
1057 	 * Check accessibility of directory.
1058 	 */
1059 
1060 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1061 		ZFS_EXIT(zfsvfs);
1062 		return (error);
1063 	}
1064 
1065 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1066 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1067 		ZFS_EXIT(zfsvfs);
1068 		return (EILSEQ);
1069 	}
1070 
1071 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1072 	if (error == 0) {
1073 		/*
1074 		 * Convert device special files
1075 		 */
1076 		if (IS_DEVVP(*vpp)) {
1077 			vnode_t	*svp;
1078 
1079 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1080 			VN_RELE(*vpp);
1081 			if (svp == NULL)
1082 				error = ENOSYS;
1083 			else
1084 				*vpp = svp;
1085 		}
1086 	}
1087 
1088 	ZFS_EXIT(zfsvfs);
1089 	return (error);
1090 }
1091 
1092 /*
1093  * Attempt to create a new entry in a directory.  If the entry
1094  * already exists, truncate the file if permissible, else return
1095  * an error.  Return the vp of the created or trunc'd file.
1096  *
1097  *	IN:	dvp	- vnode of directory to put new file entry in.
1098  *		name	- name of new file entry.
1099  *		vap	- attributes of new file.
1100  *		excl	- flag indicating exclusive or non-exclusive mode.
1101  *		mode	- mode to open file with.
1102  *		cr	- credentials of caller.
1103  *		flag	- large file flag [UNUSED].
1104  *		ct	- caller context
1105  *		vsecp 	- ACL to be set
1106  *
1107  *	OUT:	vpp	- vnode of created or trunc'd entry.
1108  *
1109  *	RETURN:	0 if success
1110  *		error code if failure
1111  *
1112  * Timestamps:
1113  *	dvp - ctime|mtime updated if new entry created
1114  *	 vp - ctime|mtime always, atime if new
1115  */
1116 
1117 /* ARGSUSED */
1118 static int
1119 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1120     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1121     vsecattr_t *vsecp)
1122 {
1123 	znode_t		*zp, *dzp = VTOZ(dvp);
1124 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1125 	zilog_t		*zilog;
1126 	objset_t	*os;
1127 	zfs_dirlock_t	*dl;
1128 	dmu_tx_t	*tx;
1129 	int		error;
1130 	zfs_acl_t	*aclp = NULL;
1131 	zfs_fuid_info_t *fuidp = NULL;
1132 
1133 	/*
1134 	 * If we have an ephemeral id, ACL, or XVATTR then
1135 	 * make sure file system is at proper version
1136 	 */
1137 
1138 	if (zfsvfs->z_use_fuids == B_FALSE &&
1139 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1140 	    IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))))
1141 		return (EINVAL);
1142 
1143 	ZFS_ENTER(zfsvfs);
1144 	ZFS_VERIFY_ZP(dzp);
1145 	os = zfsvfs->z_os;
1146 	zilog = zfsvfs->z_log;
1147 
1148 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1149 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1150 		ZFS_EXIT(zfsvfs);
1151 		return (EILSEQ);
1152 	}
1153 
1154 	if (vap->va_mask & AT_XVATTR) {
1155 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1156 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1157 			ZFS_EXIT(zfsvfs);
1158 			return (error);
1159 		}
1160 	}
1161 top:
1162 	*vpp = NULL;
1163 
1164 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1165 		vap->va_mode &= ~VSVTX;
1166 
1167 	if (*name == '\0') {
1168 		/*
1169 		 * Null component name refers to the directory itself.
1170 		 */
1171 		VN_HOLD(dvp);
1172 		zp = dzp;
1173 		dl = NULL;
1174 		error = 0;
1175 	} else {
1176 		/* possible VN_HOLD(zp) */
1177 		int zflg = 0;
1178 
1179 		if (flag & FIGNORECASE)
1180 			zflg |= ZCILOOK;
1181 
1182 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1183 		    NULL, NULL);
1184 		if (error) {
1185 			if (strcmp(name, "..") == 0)
1186 				error = EISDIR;
1187 			ZFS_EXIT(zfsvfs);
1188 			if (aclp)
1189 				zfs_acl_free(aclp);
1190 			return (error);
1191 		}
1192 	}
1193 	if (vsecp && aclp == NULL) {
1194 		error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp);
1195 		if (error) {
1196 			ZFS_EXIT(zfsvfs);
1197 			if (dl)
1198 				zfs_dirent_unlock(dl);
1199 			return (error);
1200 		}
1201 	}
1202 
1203 	if (zp == NULL) {
1204 		uint64_t txtype;
1205 
1206 		/*
1207 		 * Create a new file object and update the directory
1208 		 * to reference it.
1209 		 */
1210 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1211 			goto out;
1212 		}
1213 
1214 		/*
1215 		 * We only support the creation of regular files in
1216 		 * extended attribute directories.
1217 		 */
1218 		if ((dzp->z_phys->zp_flags & ZFS_XATTR) &&
1219 		    (vap->va_type != VREG)) {
1220 			error = EINVAL;
1221 			goto out;
1222 		}
1223 
1224 		tx = dmu_tx_create(os);
1225 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1226 		if ((aclp && aclp->z_has_fuids) || IS_EPHEMERAL(crgetuid(cr)) ||
1227 		    IS_EPHEMERAL(crgetgid(cr))) {
1228 			if (zfsvfs->z_fuid_obj == 0) {
1229 				dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1230 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1231 				    FUID_SIZE_ESTIMATE(zfsvfs));
1232 				dmu_tx_hold_zap(tx, MASTER_NODE_OBJ,
1233 				    FALSE, NULL);
1234 			} else {
1235 				dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
1236 				dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
1237 				    FUID_SIZE_ESTIMATE(zfsvfs));
1238 			}
1239 		}
1240 		dmu_tx_hold_bonus(tx, dzp->z_id);
1241 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1242 		if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) {
1243 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1244 			    0, SPA_MAXBLOCKSIZE);
1245 		}
1246 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
1247 		if (error) {
1248 			zfs_dirent_unlock(dl);
1249 			if (error == ERESTART &&
1250 			    zfsvfs->z_assign == TXG_NOWAIT) {
1251 				dmu_tx_wait(tx);
1252 				dmu_tx_abort(tx);
1253 				goto top;
1254 			}
1255 			dmu_tx_abort(tx);
1256 			ZFS_EXIT(zfsvfs);
1257 			if (aclp)
1258 				zfs_acl_free(aclp);
1259 			return (error);
1260 		}
1261 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp);
1262 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1263 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1264 		if (flag & FIGNORECASE)
1265 			txtype |= TX_CI;
1266 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1267 		    vsecp, fuidp, vap);
1268 		if (fuidp)
1269 			zfs_fuid_info_free(fuidp);
1270 		dmu_tx_commit(tx);
1271 	} else {
1272 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1273 
1274 		/*
1275 		 * A directory entry already exists for this name.
1276 		 */
1277 		/*
1278 		 * Can't truncate an existing file if in exclusive mode.
1279 		 */
1280 		if (excl == EXCL) {
1281 			error = EEXIST;
1282 			goto out;
1283 		}
1284 		/*
1285 		 * Can't open a directory for writing.
1286 		 */
1287 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1288 			error = EISDIR;
1289 			goto out;
1290 		}
1291 		/*
1292 		 * Verify requested access to file.
1293 		 */
1294 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1295 			goto out;
1296 		}
1297 
1298 		mutex_enter(&dzp->z_lock);
1299 		dzp->z_seq++;
1300 		mutex_exit(&dzp->z_lock);
1301 
1302 		/*
1303 		 * Truncate regular files if requested.
1304 		 */
1305 		if ((ZTOV(zp)->v_type == VREG) &&
1306 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1307 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1308 			if (error == ERESTART &&
1309 			    zfsvfs->z_assign == TXG_NOWAIT) {
1310 				/* NB: we already did dmu_tx_wait() */
1311 				zfs_dirent_unlock(dl);
1312 				VN_RELE(ZTOV(zp));
1313 				goto top;
1314 			}
1315 
1316 			if (error == 0) {
1317 				vnevent_create(ZTOV(zp), ct);
1318 			}
1319 		}
1320 	}
1321 out:
1322 
1323 	if (dl)
1324 		zfs_dirent_unlock(dl);
1325 
1326 	if (error) {
1327 		if (zp)
1328 			VN_RELE(ZTOV(zp));
1329 	} else {
1330 		*vpp = ZTOV(zp);
1331 		/*
1332 		 * If vnode is for a device return a specfs vnode instead.
1333 		 */
1334 		if (IS_DEVVP(*vpp)) {
1335 			struct vnode *svp;
1336 
1337 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1338 			VN_RELE(*vpp);
1339 			if (svp == NULL) {
1340 				error = ENOSYS;
1341 			}
1342 			*vpp = svp;
1343 		}
1344 	}
1345 	if (aclp)
1346 		zfs_acl_free(aclp);
1347 
1348 	ZFS_EXIT(zfsvfs);
1349 	return (error);
1350 }
1351 
1352 /*
1353  * Remove an entry from a directory.
1354  *
1355  *	IN:	dvp	- vnode of directory to remove entry from.
1356  *		name	- name of entry to remove.
1357  *		cr	- credentials of caller.
1358  *		ct	- caller context
1359  *		flags	- case flags
1360  *
1361  *	RETURN:	0 if success
1362  *		error code if failure
1363  *
1364  * Timestamps:
1365  *	dvp - ctime|mtime
1366  *	 vp - ctime (if nlink > 0)
1367  */
1368 /*ARGSUSED*/
1369 static int
1370 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1371     int flags)
1372 {
1373 	znode_t		*zp, *dzp = VTOZ(dvp);
1374 	znode_t		*xzp = NULL;
1375 	vnode_t		*vp;
1376 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1377 	zilog_t		*zilog;
1378 	uint64_t	acl_obj, xattr_obj;
1379 	zfs_dirlock_t	*dl;
1380 	dmu_tx_t	*tx;
1381 	boolean_t	may_delete_now, delete_now = FALSE;
1382 	boolean_t	unlinked;
1383 	uint64_t	txtype;
1384 	pathname_t	*realnmp = NULL;
1385 	pathname_t	realnm;
1386 	int		error;
1387 	int		zflg = ZEXISTS;
1388 
1389 	ZFS_ENTER(zfsvfs);
1390 	ZFS_VERIFY_ZP(dzp);
1391 	zilog = zfsvfs->z_log;
1392 
1393 	if (flags & FIGNORECASE) {
1394 		zflg |= ZCILOOK;
1395 		pn_alloc(&realnm);
1396 		realnmp = &realnm;
1397 	}
1398 
1399 top:
1400 	/*
1401 	 * Attempt to lock directory; fail if entry doesn't exist.
1402 	 */
1403 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1404 	    NULL, realnmp)) {
1405 		if (realnmp)
1406 			pn_free(realnmp);
1407 		ZFS_EXIT(zfsvfs);
1408 		return (error);
1409 	}
1410 
1411 	vp = ZTOV(zp);
1412 
1413 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1414 		goto out;
1415 	}
1416 
1417 	/*
1418 	 * Need to use rmdir for removing directories.
1419 	 */
1420 	if (vp->v_type == VDIR) {
1421 		error = EPERM;
1422 		goto out;
1423 	}
1424 
1425 	vnevent_remove(vp, dvp, name, ct);
1426 
1427 	if (realnmp)
1428 		dnlc_remove(dvp, realnmp->pn_buf);
1429 	else
1430 		dnlc_remove(dvp, name);
1431 
1432 	mutex_enter(&vp->v_lock);
1433 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1434 	mutex_exit(&vp->v_lock);
1435 
1436 	/*
1437 	 * We may delete the znode now, or we may put it in the unlinked set;
1438 	 * it depends on whether we're the last link, and on whether there are
1439 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1440 	 * allow for either case.
1441 	 */
1442 	tx = dmu_tx_create(zfsvfs->z_os);
1443 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1444 	dmu_tx_hold_bonus(tx, zp->z_id);
1445 	if (may_delete_now)
1446 		dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
1447 
1448 	/* are there any extended attributes? */
1449 	if ((xattr_obj = zp->z_phys->zp_xattr) != 0) {
1450 		/* XXX - do we need this if we are deleting? */
1451 		dmu_tx_hold_bonus(tx, xattr_obj);
1452 	}
1453 
1454 	/* are there any additional acls */
1455 	if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 &&
1456 	    may_delete_now)
1457 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1458 
1459 	/* charge as an update -- would be nice not to charge at all */
1460 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1461 
1462 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1463 	if (error) {
1464 		zfs_dirent_unlock(dl);
1465 		VN_RELE(vp);
1466 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1467 			dmu_tx_wait(tx);
1468 			dmu_tx_abort(tx);
1469 			goto top;
1470 		}
1471 		if (realnmp)
1472 			pn_free(realnmp);
1473 		dmu_tx_abort(tx);
1474 		ZFS_EXIT(zfsvfs);
1475 		return (error);
1476 	}
1477 
1478 	/*
1479 	 * Remove the directory entry.
1480 	 */
1481 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1482 
1483 	if (error) {
1484 		dmu_tx_commit(tx);
1485 		goto out;
1486 	}
1487 
1488 	if (unlinked) {
1489 		mutex_enter(&vp->v_lock);
1490 		delete_now = may_delete_now &&
1491 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1492 		    zp->z_phys->zp_xattr == xattr_obj &&
1493 		    zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj;
1494 		mutex_exit(&vp->v_lock);
1495 	}
1496 
1497 	if (delete_now) {
1498 		if (zp->z_phys->zp_xattr) {
1499 			error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
1500 			ASSERT3U(error, ==, 0);
1501 			ASSERT3U(xzp->z_phys->zp_links, ==, 2);
1502 			dmu_buf_will_dirty(xzp->z_dbuf, tx);
1503 			mutex_enter(&xzp->z_lock);
1504 			xzp->z_unlinked = 1;
1505 			xzp->z_phys->zp_links = 0;
1506 			mutex_exit(&xzp->z_lock);
1507 			zfs_unlinked_add(xzp, tx);
1508 			zp->z_phys->zp_xattr = 0; /* probably unnecessary */
1509 		}
1510 		mutex_enter(&zp->z_lock);
1511 		mutex_enter(&vp->v_lock);
1512 		vp->v_count--;
1513 		ASSERT3U(vp->v_count, ==, 0);
1514 		mutex_exit(&vp->v_lock);
1515 		mutex_exit(&zp->z_lock);
1516 		zfs_znode_delete(zp, tx);
1517 	} else if (unlinked) {
1518 		zfs_unlinked_add(zp, tx);
1519 	}
1520 
1521 	txtype = TX_REMOVE;
1522 	if (flags & FIGNORECASE)
1523 		txtype |= TX_CI;
1524 	zfs_log_remove(zilog, tx, txtype, dzp, name);
1525 
1526 	dmu_tx_commit(tx);
1527 out:
1528 	if (realnmp)
1529 		pn_free(realnmp);
1530 
1531 	zfs_dirent_unlock(dl);
1532 
1533 	if (!delete_now) {
1534 		VN_RELE(vp);
1535 	} else if (xzp) {
1536 		/* this rele delayed to prevent nesting transactions */
1537 		VN_RELE(ZTOV(xzp));
1538 	}
1539 
1540 	ZFS_EXIT(zfsvfs);
1541 	return (error);
1542 }
1543 
1544 /*
1545  * Create a new directory and insert it into dvp using the name
1546  * provided.  Return a pointer to the inserted directory.
1547  *
1548  *	IN:	dvp	- vnode of directory to add subdir to.
1549  *		dirname	- name of new directory.
1550  *		vap	- attributes of new directory.
1551  *		cr	- credentials of caller.
1552  *		ct	- caller context
1553  *		vsecp	- ACL to be set
1554  *
1555  *	OUT:	vpp	- vnode of created directory.
1556  *
1557  *	RETURN:	0 if success
1558  *		error code if failure
1559  *
1560  * Timestamps:
1561  *	dvp - ctime|mtime updated
1562  *	 vp - ctime|mtime|atime updated
1563  */
1564 /*ARGSUSED*/
1565 static int
1566 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1567     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1568 {
1569 	znode_t		*zp, *dzp = VTOZ(dvp);
1570 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1571 	zilog_t		*zilog;
1572 	zfs_dirlock_t	*dl;
1573 	uint64_t	txtype;
1574 	dmu_tx_t	*tx;
1575 	int		error;
1576 	zfs_acl_t	*aclp = NULL;
1577 	zfs_fuid_info_t	*fuidp = NULL;
1578 	int		zf = ZNEW;
1579 
1580 	ASSERT(vap->va_type == VDIR);
1581 
1582 	/*
1583 	 * If we have an ephemeral id, ACL, or XVATTR then
1584 	 * make sure file system is at proper version
1585 	 */
1586 
1587 	if (zfsvfs->z_use_fuids == B_FALSE &&
1588 	    (vsecp || (vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(crgetuid(cr))||
1589 	    IS_EPHEMERAL(crgetgid(cr))))
1590 		return (EINVAL);
1591 
1592 	ZFS_ENTER(zfsvfs);
1593 	ZFS_VERIFY_ZP(dzp);
1594 	zilog = zfsvfs->z_log;
1595 
1596 	if (dzp->z_phys->zp_flags & ZFS_XATTR) {
1597 		ZFS_EXIT(zfsvfs);
1598 		return (EINVAL);
1599 	}
1600 
1601 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1602 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1603 		ZFS_EXIT(zfsvfs);
1604 		return (EILSEQ);
1605 	}
1606 	if (flags & FIGNORECASE)
1607 		zf |= ZCILOOK;
1608 
1609 	if (vap->va_mask & AT_XVATTR)
1610 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1611 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1612 			ZFS_EXIT(zfsvfs);
1613 			return (error);
1614 		}
1615 
1616 	/*
1617 	 * First make sure the new directory doesn't exist.
1618 	 */
1619 top:
1620 	*vpp = NULL;
1621 
1622 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1623 	    NULL, NULL)) {
1624 		ZFS_EXIT(zfsvfs);
1625 		return (error);
1626 	}
1627 
1628 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1629 		zfs_dirent_unlock(dl);
1630 		ZFS_EXIT(zfsvfs);
1631 		return (error);
1632 	}
1633 
1634 	if (vsecp && aclp == NULL) {
1635 		error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp);
1636 		if (error) {
1637 			zfs_dirent_unlock(dl);
1638 			ZFS_EXIT(zfsvfs);
1639 			return (error);
1640 		}
1641 	}
1642 	/*
1643 	 * Add a new entry to the directory.
1644 	 */
1645 	tx = dmu_tx_create(zfsvfs->z_os);
1646 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1647 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1648 	if ((aclp && aclp->z_has_fuids) || IS_EPHEMERAL(crgetuid(cr)) ||
1649 	    IS_EPHEMERAL(crgetgid(cr))) {
1650 		if (zfsvfs->z_fuid_obj == 0) {
1651 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1652 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1653 			    FUID_SIZE_ESTIMATE(zfsvfs));
1654 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
1655 		} else {
1656 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
1657 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
1658 			    FUID_SIZE_ESTIMATE(zfsvfs));
1659 		}
1660 	}
1661 	if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp)
1662 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1663 		    0, SPA_MAXBLOCKSIZE);
1664 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1665 	if (error) {
1666 		zfs_dirent_unlock(dl);
1667 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1668 			dmu_tx_wait(tx);
1669 			dmu_tx_abort(tx);
1670 			goto top;
1671 		}
1672 		dmu_tx_abort(tx);
1673 		ZFS_EXIT(zfsvfs);
1674 		if (aclp)
1675 			zfs_acl_free(aclp);
1676 		return (error);
1677 	}
1678 
1679 	/*
1680 	 * Create new node.
1681 	 */
1682 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp);
1683 
1684 	if (aclp)
1685 		zfs_acl_free(aclp);
1686 
1687 	/*
1688 	 * Now put new name in parent dir.
1689 	 */
1690 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1691 
1692 	*vpp = ZTOV(zp);
1693 
1694 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1695 	if (flags & FIGNORECASE)
1696 		txtype |= TX_CI;
1697 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, fuidp, vap);
1698 
1699 	if (fuidp)
1700 		zfs_fuid_info_free(fuidp);
1701 	dmu_tx_commit(tx);
1702 
1703 	zfs_dirent_unlock(dl);
1704 
1705 	ZFS_EXIT(zfsvfs);
1706 	return (0);
1707 }
1708 
1709 /*
1710  * Remove a directory subdir entry.  If the current working
1711  * directory is the same as the subdir to be removed, the
1712  * remove will fail.
1713  *
1714  *	IN:	dvp	- vnode of directory to remove from.
1715  *		name	- name of directory to be removed.
1716  *		cwd	- vnode of current working directory.
1717  *		cr	- credentials of caller.
1718  *		ct	- caller context
1719  *		flags	- case flags
1720  *
1721  *	RETURN:	0 if success
1722  *		error code if failure
1723  *
1724  * Timestamps:
1725  *	dvp - ctime|mtime updated
1726  */
1727 /*ARGSUSED*/
1728 static int
1729 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1730     caller_context_t *ct, int flags)
1731 {
1732 	znode_t		*dzp = VTOZ(dvp);
1733 	znode_t		*zp;
1734 	vnode_t		*vp;
1735 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1736 	zilog_t		*zilog;
1737 	zfs_dirlock_t	*dl;
1738 	dmu_tx_t	*tx;
1739 	int		error;
1740 	int		zflg = ZEXISTS;
1741 
1742 	ZFS_ENTER(zfsvfs);
1743 	ZFS_VERIFY_ZP(dzp);
1744 	zilog = zfsvfs->z_log;
1745 
1746 	if (flags & FIGNORECASE)
1747 		zflg |= ZCILOOK;
1748 top:
1749 	zp = NULL;
1750 
1751 	/*
1752 	 * Attempt to lock directory; fail if entry doesn't exist.
1753 	 */
1754 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1755 	    NULL, NULL)) {
1756 		ZFS_EXIT(zfsvfs);
1757 		return (error);
1758 	}
1759 
1760 	vp = ZTOV(zp);
1761 
1762 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1763 		goto out;
1764 	}
1765 
1766 	if (vp->v_type != VDIR) {
1767 		error = ENOTDIR;
1768 		goto out;
1769 	}
1770 
1771 	if (vp == cwd) {
1772 		error = EINVAL;
1773 		goto out;
1774 	}
1775 
1776 	vnevent_rmdir(vp, dvp, name, ct);
1777 
1778 	/*
1779 	 * Grab a lock on the directory to make sure that noone is
1780 	 * trying to add (or lookup) entries while we are removing it.
1781 	 */
1782 	rw_enter(&zp->z_name_lock, RW_WRITER);
1783 
1784 	/*
1785 	 * Grab a lock on the parent pointer to make sure we play well
1786 	 * with the treewalk and directory rename code.
1787 	 */
1788 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1789 
1790 	tx = dmu_tx_create(zfsvfs->z_os);
1791 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1792 	dmu_tx_hold_bonus(tx, zp->z_id);
1793 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1794 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1795 	if (error) {
1796 		rw_exit(&zp->z_parent_lock);
1797 		rw_exit(&zp->z_name_lock);
1798 		zfs_dirent_unlock(dl);
1799 		VN_RELE(vp);
1800 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1801 			dmu_tx_wait(tx);
1802 			dmu_tx_abort(tx);
1803 			goto top;
1804 		}
1805 		dmu_tx_abort(tx);
1806 		ZFS_EXIT(zfsvfs);
1807 		return (error);
1808 	}
1809 
1810 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
1811 
1812 	if (error == 0) {
1813 		uint64_t txtype = TX_RMDIR;
1814 		if (flags & FIGNORECASE)
1815 			txtype |= TX_CI;
1816 		zfs_log_remove(zilog, tx, txtype, dzp, name);
1817 	}
1818 
1819 	dmu_tx_commit(tx);
1820 
1821 	rw_exit(&zp->z_parent_lock);
1822 	rw_exit(&zp->z_name_lock);
1823 out:
1824 	zfs_dirent_unlock(dl);
1825 
1826 	VN_RELE(vp);
1827 
1828 	ZFS_EXIT(zfsvfs);
1829 	return (error);
1830 }
1831 
1832 /*
1833  * Read as many directory entries as will fit into the provided
1834  * buffer from the given directory cursor position (specified in
1835  * the uio structure.
1836  *
1837  *	IN:	vp	- vnode of directory to read.
1838  *		uio	- structure supplying read location, range info,
1839  *			  and return buffer.
1840  *		cr	- credentials of caller.
1841  *		ct	- caller context
1842  *		flags	- case flags
1843  *
1844  *	OUT:	uio	- updated offset and range, buffer filled.
1845  *		eofp	- set to true if end-of-file detected.
1846  *
1847  *	RETURN:	0 if success
1848  *		error code if failure
1849  *
1850  * Timestamps:
1851  *	vp - atime updated
1852  *
1853  * Note that the low 4 bits of the cookie returned by zap is always zero.
1854  * This allows us to use the low range for "special" directory entries:
1855  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1856  * we use the offset 2 for the '.zfs' directory.
1857  */
1858 /* ARGSUSED */
1859 static int
1860 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
1861     caller_context_t *ct, int flags)
1862 {
1863 	znode_t		*zp = VTOZ(vp);
1864 	iovec_t		*iovp;
1865 	edirent_t	*eodp;
1866 	dirent64_t	*odp;
1867 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1868 	objset_t	*os;
1869 	caddr_t		outbuf;
1870 	size_t		bufsize;
1871 	zap_cursor_t	zc;
1872 	zap_attribute_t	zap;
1873 	uint_t		bytes_wanted;
1874 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1875 	int		local_eof;
1876 	int		outcount;
1877 	int		error;
1878 	uint8_t		prefetch;
1879 	boolean_t	check_sysattrs;
1880 
1881 	ZFS_ENTER(zfsvfs);
1882 	ZFS_VERIFY_ZP(zp);
1883 
1884 	/*
1885 	 * If we are not given an eof variable,
1886 	 * use a local one.
1887 	 */
1888 	if (eofp == NULL)
1889 		eofp = &local_eof;
1890 
1891 	/*
1892 	 * Check for valid iov_len.
1893 	 */
1894 	if (uio->uio_iov->iov_len <= 0) {
1895 		ZFS_EXIT(zfsvfs);
1896 		return (EINVAL);
1897 	}
1898 
1899 	/*
1900 	 * Quit if directory has been removed (posix)
1901 	 */
1902 	if ((*eofp = zp->z_unlinked) != 0) {
1903 		ZFS_EXIT(zfsvfs);
1904 		return (0);
1905 	}
1906 
1907 	error = 0;
1908 	os = zfsvfs->z_os;
1909 	offset = uio->uio_loffset;
1910 	prefetch = zp->z_zn_prefetch;
1911 
1912 	/*
1913 	 * Initialize the iterator cursor.
1914 	 */
1915 	if (offset <= 3) {
1916 		/*
1917 		 * Start iteration from the beginning of the directory.
1918 		 */
1919 		zap_cursor_init(&zc, os, zp->z_id);
1920 	} else {
1921 		/*
1922 		 * The offset is a serialized cursor.
1923 		 */
1924 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1925 	}
1926 
1927 	/*
1928 	 * Get space to change directory entries into fs independent format.
1929 	 */
1930 	iovp = uio->uio_iov;
1931 	bytes_wanted = iovp->iov_len;
1932 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
1933 		bufsize = bytes_wanted;
1934 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
1935 		odp = (struct dirent64 *)outbuf;
1936 	} else {
1937 		bufsize = bytes_wanted;
1938 		odp = (struct dirent64 *)iovp->iov_base;
1939 	}
1940 	eodp = (struct edirent *)odp;
1941 
1942 	/*
1943 	 * If this VFS supports system attributes; and we're looking at an
1944 	 * extended attribute directory; and we care about normalization
1945 	 * conflicts on this vfs; then we must check for normalization
1946 	 * conflicts with the sysattr name space.
1947 	 */
1948 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) &&
1949 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
1950 	    (flags & V_RDDIR_ENTFLAGS);
1951 
1952 	/*
1953 	 * Transform to file-system independent format
1954 	 */
1955 	outcount = 0;
1956 	while (outcount < bytes_wanted) {
1957 		ino64_t objnum;
1958 		ushort_t reclen;
1959 		off64_t *next;
1960 
1961 		/*
1962 		 * Special case `.', `..', and `.zfs'.
1963 		 */
1964 		if (offset == 0) {
1965 			(void) strcpy(zap.za_name, ".");
1966 			zap.za_normalization_conflict = 0;
1967 			objnum = zp->z_id;
1968 		} else if (offset == 1) {
1969 			(void) strcpy(zap.za_name, "..");
1970 			zap.za_normalization_conflict = 0;
1971 			objnum = zp->z_phys->zp_parent;
1972 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
1973 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1974 			zap.za_normalization_conflict = 0;
1975 			objnum = ZFSCTL_INO_ROOT;
1976 		} else {
1977 			/*
1978 			 * Grab next entry.
1979 			 */
1980 			if (error = zap_cursor_retrieve(&zc, &zap)) {
1981 				if ((*eofp = (error == ENOENT)) != 0)
1982 					break;
1983 				else
1984 					goto update;
1985 			}
1986 
1987 			if (zap.za_integer_length != 8 ||
1988 			    zap.za_num_integers != 1) {
1989 				cmn_err(CE_WARN, "zap_readdir: bad directory "
1990 				    "entry, obj = %lld, offset = %lld\n",
1991 				    (u_longlong_t)zp->z_id,
1992 				    (u_longlong_t)offset);
1993 				error = ENXIO;
1994 				goto update;
1995 			}
1996 
1997 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
1998 			/*
1999 			 * MacOS X can extract the object type here such as:
2000 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2001 			 */
2002 
2003 			if (check_sysattrs && !zap.za_normalization_conflict) {
2004 				zap.za_normalization_conflict =
2005 				    xattr_sysattr_casechk(zap.za_name);
2006 			}
2007 		}
2008 
2009 		if (flags & V_RDDIR_ENTFLAGS)
2010 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2011 		else
2012 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2013 
2014 		/*
2015 		 * Will this entry fit in the buffer?
2016 		 */
2017 		if (outcount + reclen > bufsize) {
2018 			/*
2019 			 * Did we manage to fit anything in the buffer?
2020 			 */
2021 			if (!outcount) {
2022 				error = EINVAL;
2023 				goto update;
2024 			}
2025 			break;
2026 		}
2027 		if (flags & V_RDDIR_ENTFLAGS) {
2028 			/*
2029 			 * Add extended flag entry:
2030 			 */
2031 			eodp->ed_ino = objnum;
2032 			eodp->ed_reclen = reclen;
2033 			/* NOTE: ed_off is the offset for the *next* entry */
2034 			next = &(eodp->ed_off);
2035 			eodp->ed_eflags = zap.za_normalization_conflict ?
2036 			    ED_CASE_CONFLICT : 0;
2037 			(void) strncpy(eodp->ed_name, zap.za_name,
2038 			    EDIRENT_NAMELEN(reclen));
2039 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2040 		} else {
2041 			/*
2042 			 * Add normal entry:
2043 			 */
2044 			odp->d_ino = objnum;
2045 			odp->d_reclen = reclen;
2046 			/* NOTE: d_off is the offset for the *next* entry */
2047 			next = &(odp->d_off);
2048 			(void) strncpy(odp->d_name, zap.za_name,
2049 			    DIRENT64_NAMELEN(reclen));
2050 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2051 		}
2052 		outcount += reclen;
2053 
2054 		ASSERT(outcount <= bufsize);
2055 
2056 		/* Prefetch znode */
2057 		if (prefetch)
2058 			dmu_prefetch(os, objnum, 0, 0);
2059 
2060 		/*
2061 		 * Move to the next entry, fill in the previous offset.
2062 		 */
2063 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2064 			zap_cursor_advance(&zc);
2065 			offset = zap_cursor_serialize(&zc);
2066 		} else {
2067 			offset += 1;
2068 		}
2069 		*next = offset;
2070 	}
2071 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2072 
2073 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2074 		iovp->iov_base += outcount;
2075 		iovp->iov_len -= outcount;
2076 		uio->uio_resid -= outcount;
2077 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2078 		/*
2079 		 * Reset the pointer.
2080 		 */
2081 		offset = uio->uio_loffset;
2082 	}
2083 
2084 update:
2085 	zap_cursor_fini(&zc);
2086 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2087 		kmem_free(outbuf, bufsize);
2088 
2089 	if (error == ENOENT)
2090 		error = 0;
2091 
2092 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2093 
2094 	uio->uio_loffset = offset;
2095 	ZFS_EXIT(zfsvfs);
2096 	return (error);
2097 }
2098 
2099 ulong_t zfs_fsync_sync_cnt = 4;
2100 
2101 static int
2102 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2103 {
2104 	znode_t	*zp = VTOZ(vp);
2105 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2106 
2107 	/*
2108 	 * Regardless of whether this is required for standards conformance,
2109 	 * this is the logical behavior when fsync() is called on a file with
2110 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2111 	 * going to be pushed out as part of the zil_commit().
2112 	 */
2113 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2114 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2115 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2116 
2117 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2118 
2119 	ZFS_ENTER(zfsvfs);
2120 	ZFS_VERIFY_ZP(zp);
2121 	zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
2122 	ZFS_EXIT(zfsvfs);
2123 	return (0);
2124 }
2125 
2126 
2127 /*
2128  * Get the requested file attributes and place them in the provided
2129  * vattr structure.
2130  *
2131  *	IN:	vp	- vnode of file.
2132  *		vap	- va_mask identifies requested attributes.
2133  *			  If AT_XVATTR set, then optional attrs are requested
2134  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2135  *		cr	- credentials of caller.
2136  *		ct	- caller context
2137  *
2138  *	OUT:	vap	- attribute values.
2139  *
2140  *	RETURN:	0 (always succeeds)
2141  */
2142 /* ARGSUSED */
2143 static int
2144 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2145     caller_context_t *ct)
2146 {
2147 	znode_t *zp = VTOZ(vp);
2148 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2149 	znode_phys_t *pzp;
2150 	int	error = 0;
2151 	uint64_t links;
2152 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2153 	xoptattr_t *xoap = NULL;
2154 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2155 
2156 	ZFS_ENTER(zfsvfs);
2157 	ZFS_VERIFY_ZP(zp);
2158 	pzp = zp->z_phys;
2159 
2160 	mutex_enter(&zp->z_lock);
2161 
2162 	/*
2163 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2164 	 * Also, if we are the owner don't bother, since owner should
2165 	 * always be allowed to read basic attributes of file.
2166 	 */
2167 	if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL) &&
2168 	    (pzp->zp_uid != crgetuid(cr))) {
2169 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2170 		    skipaclchk, cr)) {
2171 			mutex_exit(&zp->z_lock);
2172 			ZFS_EXIT(zfsvfs);
2173 			return (error);
2174 		}
2175 	}
2176 
2177 	/*
2178 	 * Return all attributes.  It's cheaper to provide the answer
2179 	 * than to determine whether we were asked the question.
2180 	 */
2181 
2182 	vap->va_type = vp->v_type;
2183 	vap->va_mode = pzp->zp_mode & MODEMASK;
2184 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2185 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2186 	vap->va_nodeid = zp->z_id;
2187 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2188 		links = pzp->zp_links + 1;
2189 	else
2190 		links = pzp->zp_links;
2191 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2192 	vap->va_size = pzp->zp_size;
2193 	vap->va_rdev = vp->v_rdev;
2194 	vap->va_seq = zp->z_seq;
2195 
2196 	/*
2197 	 * Add in any requested optional attributes and the create time.
2198 	 * Also set the corresponding bits in the returned attribute bitmap.
2199 	 */
2200 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2201 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2202 			xoap->xoa_archive =
2203 			    ((pzp->zp_flags & ZFS_ARCHIVE) != 0);
2204 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2205 		}
2206 
2207 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2208 			xoap->xoa_readonly =
2209 			    ((pzp->zp_flags & ZFS_READONLY) != 0);
2210 			XVA_SET_RTN(xvap, XAT_READONLY);
2211 		}
2212 
2213 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2214 			xoap->xoa_system =
2215 			    ((pzp->zp_flags & ZFS_SYSTEM) != 0);
2216 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2217 		}
2218 
2219 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2220 			xoap->xoa_hidden =
2221 			    ((pzp->zp_flags & ZFS_HIDDEN) != 0);
2222 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2223 		}
2224 
2225 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2226 			xoap->xoa_nounlink =
2227 			    ((pzp->zp_flags & ZFS_NOUNLINK) != 0);
2228 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2229 		}
2230 
2231 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2232 			xoap->xoa_immutable =
2233 			    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0);
2234 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2235 		}
2236 
2237 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2238 			xoap->xoa_appendonly =
2239 			    ((pzp->zp_flags & ZFS_APPENDONLY) != 0);
2240 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2241 		}
2242 
2243 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2244 			xoap->xoa_nodump =
2245 			    ((pzp->zp_flags & ZFS_NODUMP) != 0);
2246 			XVA_SET_RTN(xvap, XAT_NODUMP);
2247 		}
2248 
2249 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2250 			xoap->xoa_opaque =
2251 			    ((pzp->zp_flags & ZFS_OPAQUE) != 0);
2252 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2253 		}
2254 
2255 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2256 			xoap->xoa_av_quarantined =
2257 			    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0);
2258 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2259 		}
2260 
2261 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2262 			xoap->xoa_av_modified =
2263 			    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0);
2264 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2265 		}
2266 
2267 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2268 		    vp->v_type == VREG &&
2269 		    (pzp->zp_flags & ZFS_BONUS_SCANSTAMP)) {
2270 			size_t len;
2271 			dmu_object_info_t doi;
2272 
2273 			/*
2274 			 * Only VREG files have anti-virus scanstamps, so we
2275 			 * won't conflict with symlinks in the bonus buffer.
2276 			 */
2277 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2278 			len = sizeof (xoap->xoa_av_scanstamp) +
2279 			    sizeof (znode_phys_t);
2280 			if (len <= doi.doi_bonus_size) {
2281 				/*
2282 				 * pzp points to the start of the
2283 				 * znode_phys_t. pzp + 1 points to the
2284 				 * first byte after the znode_phys_t.
2285 				 */
2286 				(void) memcpy(xoap->xoa_av_scanstamp,
2287 				    pzp + 1,
2288 				    sizeof (xoap->xoa_av_scanstamp));
2289 				XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
2290 			}
2291 		}
2292 
2293 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2294 			ZFS_TIME_DECODE(&xoap->xoa_createtime, pzp->zp_crtime);
2295 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2296 		}
2297 	}
2298 
2299 	ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime);
2300 	ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime);
2301 	ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime);
2302 
2303 	mutex_exit(&zp->z_lock);
2304 
2305 	dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks);
2306 
2307 	if (zp->z_blksz == 0) {
2308 		/*
2309 		 * Block size hasn't been set; suggest maximal I/O transfers.
2310 		 */
2311 		vap->va_blksize = zfsvfs->z_max_blksz;
2312 	}
2313 
2314 	ZFS_EXIT(zfsvfs);
2315 	return (0);
2316 }
2317 
2318 /*
2319  * Set the file attributes to the values contained in the
2320  * vattr structure.
2321  *
2322  *	IN:	vp	- vnode of file to be modified.
2323  *		vap	- new attribute values.
2324  *			  If AT_XVATTR set, then optional attrs are being set
2325  *		flags	- ATTR_UTIME set if non-default time values provided.
2326  *			- ATTR_NOACLCHECK (CIFS context only).
2327  *		cr	- credentials of caller.
2328  *		ct	- caller context
2329  *
2330  *	RETURN:	0 if success
2331  *		error code if failure
2332  *
2333  * Timestamps:
2334  *	vp - ctime updated, mtime updated if size changed.
2335  */
2336 /* ARGSUSED */
2337 static int
2338 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2339 	caller_context_t *ct)
2340 {
2341 	znode_t		*zp = VTOZ(vp);
2342 	znode_phys_t	*pzp;
2343 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2344 	zilog_t		*zilog;
2345 	dmu_tx_t	*tx;
2346 	vattr_t		oldva;
2347 	uint_t		mask = vap->va_mask;
2348 	uint_t		saved_mask;
2349 	int		trim_mask = 0;
2350 	uint64_t	new_mode;
2351 	znode_t		*attrzp;
2352 	int		need_policy = FALSE;
2353 	int		err;
2354 	zfs_fuid_info_t *fuidp = NULL;
2355 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2356 	xoptattr_t	*xoap;
2357 	zfs_acl_t	*aclp = NULL;
2358 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2359 
2360 	if (mask == 0)
2361 		return (0);
2362 
2363 	if (mask & AT_NOSET)
2364 		return (EINVAL);
2365 
2366 	ZFS_ENTER(zfsvfs);
2367 	ZFS_VERIFY_ZP(zp);
2368 
2369 	pzp = zp->z_phys;
2370 	zilog = zfsvfs->z_log;
2371 
2372 	/*
2373 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2374 	 * that file system is at proper version level
2375 	 */
2376 
2377 	if (zfsvfs->z_use_fuids == B_FALSE &&
2378 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2379 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2380 	    (mask & AT_XVATTR))) {
2381 		ZFS_EXIT(zfsvfs);
2382 		return (EINVAL);
2383 	}
2384 
2385 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2386 		ZFS_EXIT(zfsvfs);
2387 		return (EISDIR);
2388 	}
2389 
2390 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2391 		ZFS_EXIT(zfsvfs);
2392 		return (EINVAL);
2393 	}
2394 
2395 	/*
2396 	 * If this is an xvattr_t, then get a pointer to the structure of
2397 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2398 	 */
2399 	xoap = xva_getxoptattr(xvap);
2400 
2401 	/*
2402 	 * Immutable files can only alter immutable bit and atime
2403 	 */
2404 	if ((pzp->zp_flags & ZFS_IMMUTABLE) &&
2405 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2406 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2407 		ZFS_EXIT(zfsvfs);
2408 		return (EPERM);
2409 	}
2410 
2411 	if ((mask & AT_SIZE) && (pzp->zp_flags & ZFS_READONLY)) {
2412 		ZFS_EXIT(zfsvfs);
2413 		return (EPERM);
2414 	}
2415 
2416 	/*
2417 	 * Verify timestamps doesn't overflow 32 bits.
2418 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2419 	 * handle times greater than 2039.  This check should be removed
2420 	 * once large timestamps are fully supported.
2421 	 */
2422 	if (mask & (AT_ATIME | AT_MTIME)) {
2423 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2424 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2425 			ZFS_EXIT(zfsvfs);
2426 			return (EOVERFLOW);
2427 		}
2428 	}
2429 
2430 top:
2431 	attrzp = NULL;
2432 
2433 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2434 		ZFS_EXIT(zfsvfs);
2435 		return (EROFS);
2436 	}
2437 
2438 	/*
2439 	 * First validate permissions
2440 	 */
2441 
2442 	if (mask & AT_SIZE) {
2443 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2444 		if (err) {
2445 			ZFS_EXIT(zfsvfs);
2446 			return (err);
2447 		}
2448 		/*
2449 		 * XXX - Note, we are not providing any open
2450 		 * mode flags here (like FNDELAY), so we may
2451 		 * block if there are locks present... this
2452 		 * should be addressed in openat().
2453 		 */
2454 		do {
2455 			err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2456 			/* NB: we already did dmu_tx_wait() if necessary */
2457 		} while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
2458 		if (err) {
2459 			ZFS_EXIT(zfsvfs);
2460 			return (err);
2461 		}
2462 	}
2463 
2464 	if (mask & (AT_ATIME|AT_MTIME) ||
2465 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2466 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2467 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2468 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2469 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM))))
2470 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2471 		    skipaclchk, cr);
2472 
2473 	if (mask & (AT_UID|AT_GID)) {
2474 		int	idmask = (mask & (AT_UID|AT_GID));
2475 		int	take_owner;
2476 		int	take_group;
2477 
2478 		/*
2479 		 * NOTE: even if a new mode is being set,
2480 		 * we may clear S_ISUID/S_ISGID bits.
2481 		 */
2482 
2483 		if (!(mask & AT_MODE))
2484 			vap->va_mode = pzp->zp_mode;
2485 
2486 		/*
2487 		 * Take ownership or chgrp to group we are a member of
2488 		 */
2489 
2490 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2491 		take_group = (mask & AT_GID) &&
2492 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2493 
2494 		/*
2495 		 * If both AT_UID and AT_GID are set then take_owner and
2496 		 * take_group must both be set in order to allow taking
2497 		 * ownership.
2498 		 *
2499 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2500 		 *
2501 		 */
2502 
2503 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2504 		    ((idmask == AT_UID) && take_owner) ||
2505 		    ((idmask == AT_GID) && take_group)) {
2506 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2507 			    skipaclchk, cr) == 0) {
2508 				/*
2509 				 * Remove setuid/setgid for non-privileged users
2510 				 */
2511 				secpolicy_setid_clear(vap, cr);
2512 				trim_mask = (mask & (AT_UID|AT_GID));
2513 			} else {
2514 				need_policy =  TRUE;
2515 			}
2516 		} else {
2517 			need_policy =  TRUE;
2518 		}
2519 	}
2520 
2521 	mutex_enter(&zp->z_lock);
2522 	oldva.va_mode = pzp->zp_mode;
2523 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2524 	if (mask & AT_XVATTR) {
2525 		if ((need_policy == FALSE) &&
2526 		    (XVA_ISSET_REQ(xvap, XAT_APPENDONLY) &&
2527 		    xoap->xoa_appendonly !=
2528 		    ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) ||
2529 		    (XVA_ISSET_REQ(xvap, XAT_NOUNLINK) &&
2530 		    xoap->xoa_nounlink !=
2531 		    ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) ||
2532 		    (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE) &&
2533 		    xoap->xoa_immutable !=
2534 		    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) ||
2535 		    (XVA_ISSET_REQ(xvap, XAT_NODUMP) &&
2536 		    xoap->xoa_nodump !=
2537 		    ((pzp->zp_flags & ZFS_NODUMP) != 0)) ||
2538 		    (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED) &&
2539 		    xoap->xoa_av_modified !=
2540 		    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) ||
2541 		    ((XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED) &&
2542 		    ((vp->v_type != VREG && xoap->xoa_av_quarantined) ||
2543 		    xoap->xoa_av_quarantined !=
2544 		    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)))) ||
2545 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
2546 		    (XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2547 			need_policy = TRUE;
2548 		}
2549 	}
2550 
2551 	mutex_exit(&zp->z_lock);
2552 
2553 	if (mask & AT_MODE) {
2554 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2555 			err = secpolicy_setid_setsticky_clear(vp, vap,
2556 			    &oldva, cr);
2557 			if (err) {
2558 				ZFS_EXIT(zfsvfs);
2559 				return (err);
2560 			}
2561 			trim_mask |= AT_MODE;
2562 		} else {
2563 			need_policy = TRUE;
2564 		}
2565 	}
2566 
2567 	if (need_policy) {
2568 		/*
2569 		 * If trim_mask is set then take ownership
2570 		 * has been granted or write_acl is present and user
2571 		 * has the ability to modify mode.  In that case remove
2572 		 * UID|GID and or MODE from mask so that
2573 		 * secpolicy_vnode_setattr() doesn't revoke it.
2574 		 */
2575 
2576 		if (trim_mask) {
2577 			saved_mask = vap->va_mask;
2578 			vap->va_mask &= ~trim_mask;
2579 		}
2580 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2581 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2582 		if (err) {
2583 			ZFS_EXIT(zfsvfs);
2584 			return (err);
2585 		}
2586 
2587 		if (trim_mask)
2588 			vap->va_mask |= saved_mask;
2589 	}
2590 
2591 	/*
2592 	 * secpolicy_vnode_setattr, or take ownership may have
2593 	 * changed va_mask
2594 	 */
2595 	mask = vap->va_mask;
2596 
2597 	tx = dmu_tx_create(zfsvfs->z_os);
2598 	dmu_tx_hold_bonus(tx, zp->z_id);
2599 	if (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2600 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid))) {
2601 		if (zfsvfs->z_fuid_obj == 0) {
2602 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
2603 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2604 			    FUID_SIZE_ESTIMATE(zfsvfs));
2605 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
2606 		} else {
2607 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
2608 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
2609 			    FUID_SIZE_ESTIMATE(zfsvfs));
2610 		}
2611 	}
2612 
2613 	if (mask & AT_MODE) {
2614 		uint64_t pmode = pzp->zp_mode;
2615 
2616 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2617 
2618 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) {
2619 			dmu_tx_abort(tx);
2620 			ZFS_EXIT(zfsvfs);
2621 			return (err);
2622 		}
2623 		if (pzp->zp_acl.z_acl_extern_obj) {
2624 			/* Are we upgrading ACL from old V0 format to new V1 */
2625 			if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2626 			    pzp->zp_acl.z_acl_version ==
2627 			    ZFS_ACL_VERSION_INITIAL) {
2628 				dmu_tx_hold_free(tx,
2629 				    pzp->zp_acl.z_acl_extern_obj, 0,
2630 				    DMU_OBJECT_END);
2631 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2632 				    0, aclp->z_acl_bytes);
2633 			} else {
2634 				dmu_tx_hold_write(tx,
2635 				    pzp->zp_acl.z_acl_extern_obj, 0,
2636 				    aclp->z_acl_bytes);
2637 			}
2638 		} else if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2639 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2640 			    0, aclp->z_acl_bytes);
2641 		}
2642 	}
2643 
2644 	if ((mask & (AT_UID | AT_GID)) && pzp->zp_xattr != 0) {
2645 		err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp);
2646 		if (err) {
2647 			dmu_tx_abort(tx);
2648 			ZFS_EXIT(zfsvfs);
2649 			if (aclp)
2650 				zfs_acl_free(aclp);
2651 			return (err);
2652 		}
2653 		dmu_tx_hold_bonus(tx, attrzp->z_id);
2654 	}
2655 
2656 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2657 	if (err) {
2658 		if (attrzp)
2659 			VN_RELE(ZTOV(attrzp));
2660 
2661 		if (aclp) {
2662 			zfs_acl_free(aclp);
2663 			aclp = NULL;
2664 		}
2665 
2666 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2667 			dmu_tx_wait(tx);
2668 			dmu_tx_abort(tx);
2669 			goto top;
2670 		}
2671 		dmu_tx_abort(tx);
2672 		ZFS_EXIT(zfsvfs);
2673 		return (err);
2674 	}
2675 
2676 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2677 
2678 	/*
2679 	 * Set each attribute requested.
2680 	 * We group settings according to the locks they need to acquire.
2681 	 *
2682 	 * Note: you cannot set ctime directly, although it will be
2683 	 * updated as a side-effect of calling this function.
2684 	 */
2685 
2686 	mutex_enter(&zp->z_lock);
2687 
2688 	if (mask & AT_MODE) {
2689 		mutex_enter(&zp->z_acl_lock);
2690 		zp->z_phys->zp_mode = new_mode;
2691 		err = zfs_aclset_common(zp, aclp, cr, &fuidp, tx);
2692 		ASSERT3U(err, ==, 0);
2693 		mutex_exit(&zp->z_acl_lock);
2694 	}
2695 
2696 	if (attrzp)
2697 		mutex_enter(&attrzp->z_lock);
2698 
2699 	if (mask & AT_UID) {
2700 		pzp->zp_uid = zfs_fuid_create(zfsvfs,
2701 		    vap->va_uid, cr, ZFS_OWNER, tx, &fuidp);
2702 		if (attrzp) {
2703 			attrzp->z_phys->zp_uid = zfs_fuid_create(zfsvfs,
2704 			    vap->va_uid,  cr, ZFS_OWNER, tx, &fuidp);
2705 		}
2706 	}
2707 
2708 	if (mask & AT_GID) {
2709 		pzp->zp_gid = zfs_fuid_create(zfsvfs, vap->va_gid,
2710 		    cr, ZFS_GROUP, tx, &fuidp);
2711 		if (attrzp)
2712 			attrzp->z_phys->zp_gid = zfs_fuid_create(zfsvfs,
2713 			    vap->va_gid, cr, ZFS_GROUP, tx, &fuidp);
2714 	}
2715 
2716 	if (aclp)
2717 		zfs_acl_free(aclp);
2718 
2719 	if (attrzp)
2720 		mutex_exit(&attrzp->z_lock);
2721 
2722 	if (mask & AT_ATIME)
2723 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2724 
2725 	if (mask & AT_MTIME)
2726 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2727 
2728 	if (mask & AT_SIZE)
2729 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2730 	else if (mask != 0)
2731 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2732 	/*
2733 	 * Do this after setting timestamps to prevent timestamp
2734 	 * update from toggling bit
2735 	 */
2736 
2737 	if (xoap && (mask & AT_XVATTR)) {
2738 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
2739 			size_t len;
2740 			dmu_object_info_t doi;
2741 
2742 			ASSERT(vp->v_type == VREG);
2743 
2744 			/* Grow the bonus buffer if necessary. */
2745 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2746 			len = sizeof (xoap->xoa_av_scanstamp) +
2747 			    sizeof (znode_phys_t);
2748 			if (len > doi.doi_bonus_size)
2749 				VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0);
2750 		}
2751 		zfs_xvattr_set(zp, xvap);
2752 	}
2753 
2754 	if (mask != 0)
2755 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2756 
2757 	if (fuidp)
2758 		zfs_fuid_info_free(fuidp);
2759 	mutex_exit(&zp->z_lock);
2760 
2761 	if (attrzp)
2762 		VN_RELE(ZTOV(attrzp));
2763 
2764 	dmu_tx_commit(tx);
2765 
2766 	ZFS_EXIT(zfsvfs);
2767 	return (err);
2768 }
2769 
2770 typedef struct zfs_zlock {
2771 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2772 	znode_t		*zl_znode;	/* znode we held */
2773 	struct zfs_zlock *zl_next;	/* next in list */
2774 } zfs_zlock_t;
2775 
2776 /*
2777  * Drop locks and release vnodes that were held by zfs_rename_lock().
2778  */
2779 static void
2780 zfs_rename_unlock(zfs_zlock_t **zlpp)
2781 {
2782 	zfs_zlock_t *zl;
2783 
2784 	while ((zl = *zlpp) != NULL) {
2785 		if (zl->zl_znode != NULL)
2786 			VN_RELE(ZTOV(zl->zl_znode));
2787 		rw_exit(zl->zl_rwlock);
2788 		*zlpp = zl->zl_next;
2789 		kmem_free(zl, sizeof (*zl));
2790 	}
2791 }
2792 
2793 /*
2794  * Search back through the directory tree, using the ".." entries.
2795  * Lock each directory in the chain to prevent concurrent renames.
2796  * Fail any attempt to move a directory into one of its own descendants.
2797  * XXX - z_parent_lock can overlap with map or grow locks
2798  */
2799 static int
2800 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2801 {
2802 	zfs_zlock_t	*zl;
2803 	znode_t		*zp = tdzp;
2804 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2805 	uint64_t	*oidp = &zp->z_id;
2806 	krwlock_t	*rwlp = &szp->z_parent_lock;
2807 	krw_t		rw = RW_WRITER;
2808 
2809 	/*
2810 	 * First pass write-locks szp and compares to zp->z_id.
2811 	 * Later passes read-lock zp and compare to zp->z_parent.
2812 	 */
2813 	do {
2814 		if (!rw_tryenter(rwlp, rw)) {
2815 			/*
2816 			 * Another thread is renaming in this path.
2817 			 * Note that if we are a WRITER, we don't have any
2818 			 * parent_locks held yet.
2819 			 */
2820 			if (rw == RW_READER && zp->z_id > szp->z_id) {
2821 				/*
2822 				 * Drop our locks and restart
2823 				 */
2824 				zfs_rename_unlock(&zl);
2825 				*zlpp = NULL;
2826 				zp = tdzp;
2827 				oidp = &zp->z_id;
2828 				rwlp = &szp->z_parent_lock;
2829 				rw = RW_WRITER;
2830 				continue;
2831 			} else {
2832 				/*
2833 				 * Wait for other thread to drop its locks
2834 				 */
2835 				rw_enter(rwlp, rw);
2836 			}
2837 		}
2838 
2839 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2840 		zl->zl_rwlock = rwlp;
2841 		zl->zl_znode = NULL;
2842 		zl->zl_next = *zlpp;
2843 		*zlpp = zl;
2844 
2845 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2846 			return (EINVAL);
2847 
2848 		if (*oidp == rootid)		/* We've hit the top */
2849 			return (0);
2850 
2851 		if (rw == RW_READER) {		/* i.e. not the first pass */
2852 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2853 			if (error)
2854 				return (error);
2855 			zl->zl_znode = zp;
2856 		}
2857 		oidp = &zp->z_phys->zp_parent;
2858 		rwlp = &zp->z_parent_lock;
2859 		rw = RW_READER;
2860 
2861 	} while (zp->z_id != sdzp->z_id);
2862 
2863 	return (0);
2864 }
2865 
2866 /*
2867  * Move an entry from the provided source directory to the target
2868  * directory.  Change the entry name as indicated.
2869  *
2870  *	IN:	sdvp	- Source directory containing the "old entry".
2871  *		snm	- Old entry name.
2872  *		tdvp	- Target directory to contain the "new entry".
2873  *		tnm	- New entry name.
2874  *		cr	- credentials of caller.
2875  *		ct	- caller context
2876  *		flags	- case flags
2877  *
2878  *	RETURN:	0 if success
2879  *		error code if failure
2880  *
2881  * Timestamps:
2882  *	sdvp,tdvp - ctime|mtime updated
2883  */
2884 /*ARGSUSED*/
2885 static int
2886 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
2887     caller_context_t *ct, int flags)
2888 {
2889 	znode_t		*tdzp, *szp, *tzp;
2890 	znode_t		*sdzp = VTOZ(sdvp);
2891 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2892 	zilog_t		*zilog;
2893 	vnode_t		*realvp;
2894 	zfs_dirlock_t	*sdl, *tdl;
2895 	dmu_tx_t	*tx;
2896 	zfs_zlock_t	*zl;
2897 	int		cmp, serr, terr;
2898 	int		error = 0;
2899 	int		zflg = 0;
2900 
2901 	ZFS_ENTER(zfsvfs);
2902 	ZFS_VERIFY_ZP(sdzp);
2903 	zilog = zfsvfs->z_log;
2904 
2905 	/*
2906 	 * Make sure we have the real vp for the target directory.
2907 	 */
2908 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
2909 		tdvp = realvp;
2910 
2911 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
2912 		ZFS_EXIT(zfsvfs);
2913 		return (EXDEV);
2914 	}
2915 
2916 	tdzp = VTOZ(tdvp);
2917 	ZFS_VERIFY_ZP(tdzp);
2918 	if (zfsvfs->z_utf8 && u8_validate(tnm,
2919 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2920 		ZFS_EXIT(zfsvfs);
2921 		return (EILSEQ);
2922 	}
2923 
2924 	if (flags & FIGNORECASE)
2925 		zflg |= ZCILOOK;
2926 
2927 top:
2928 	szp = NULL;
2929 	tzp = NULL;
2930 	zl = NULL;
2931 
2932 	/*
2933 	 * This is to prevent the creation of links into attribute space
2934 	 * by renaming a linked file into/outof an attribute directory.
2935 	 * See the comment in zfs_link() for why this is considered bad.
2936 	 */
2937 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
2938 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
2939 		ZFS_EXIT(zfsvfs);
2940 		return (EINVAL);
2941 	}
2942 
2943 	/*
2944 	 * Lock source and target directory entries.  To prevent deadlock,
2945 	 * a lock ordering must be defined.  We lock the directory with
2946 	 * the smallest object id first, or if it's a tie, the one with
2947 	 * the lexically first name.
2948 	 */
2949 	if (sdzp->z_id < tdzp->z_id) {
2950 		cmp = -1;
2951 	} else if (sdzp->z_id > tdzp->z_id) {
2952 		cmp = 1;
2953 	} else {
2954 		/*
2955 		 * First compare the two name arguments without
2956 		 * considering any case folding.
2957 		 */
2958 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
2959 
2960 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
2961 		ASSERT(error == 0 || !zfsvfs->z_utf8);
2962 		if (cmp == 0) {
2963 			/*
2964 			 * POSIX: "If the old argument and the new argument
2965 			 * both refer to links to the same existing file,
2966 			 * the rename() function shall return successfully
2967 			 * and perform no other action."
2968 			 */
2969 			ZFS_EXIT(zfsvfs);
2970 			return (0);
2971 		}
2972 		/*
2973 		 * If the file system is case-folding, then we may
2974 		 * have some more checking to do.  A case-folding file
2975 		 * system is either supporting mixed case sensitivity
2976 		 * access or is completely case-insensitive.  Note
2977 		 * that the file system is always case preserving.
2978 		 *
2979 		 * In mixed sensitivity mode case sensitive behavior
2980 		 * is the default.  FIGNORECASE must be used to
2981 		 * explicitly request case insensitive behavior.
2982 		 *
2983 		 * If the source and target names provided differ only
2984 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
2985 		 * we will treat this as a special case in the
2986 		 * case-insensitive mode: as long as the source name
2987 		 * is an exact match, we will allow this to proceed as
2988 		 * a name-change request.
2989 		 */
2990 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
2991 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
2992 		    flags & FIGNORECASE)) &&
2993 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
2994 		    &error) == 0) {
2995 			/*
2996 			 * case preserving rename request, require exact
2997 			 * name matches
2998 			 */
2999 			zflg |= ZCIEXACT;
3000 			zflg &= ~ZCILOOK;
3001 		}
3002 	}
3003 
3004 	if (cmp < 0) {
3005 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3006 		    ZEXISTS | zflg, NULL, NULL);
3007 		terr = zfs_dirent_lock(&tdl,
3008 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3009 	} else {
3010 		terr = zfs_dirent_lock(&tdl,
3011 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3012 		serr = zfs_dirent_lock(&sdl,
3013 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3014 		    NULL, NULL);
3015 	}
3016 
3017 	if (serr) {
3018 		/*
3019 		 * Source entry invalid or not there.
3020 		 */
3021 		if (!terr) {
3022 			zfs_dirent_unlock(tdl);
3023 			if (tzp)
3024 				VN_RELE(ZTOV(tzp));
3025 		}
3026 		if (strcmp(snm, "..") == 0)
3027 			serr = EINVAL;
3028 		ZFS_EXIT(zfsvfs);
3029 		return (serr);
3030 	}
3031 	if (terr) {
3032 		zfs_dirent_unlock(sdl);
3033 		VN_RELE(ZTOV(szp));
3034 		if (strcmp(tnm, "..") == 0)
3035 			terr = EINVAL;
3036 		ZFS_EXIT(zfsvfs);
3037 		return (terr);
3038 	}
3039 
3040 	/*
3041 	 * Must have write access at the source to remove the old entry
3042 	 * and write access at the target to create the new entry.
3043 	 * Note that if target and source are the same, this can be
3044 	 * done in a single check.
3045 	 */
3046 
3047 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3048 		goto out;
3049 
3050 	if (ZTOV(szp)->v_type == VDIR) {
3051 		/*
3052 		 * Check to make sure rename is valid.
3053 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3054 		 */
3055 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3056 			goto out;
3057 	}
3058 
3059 	/*
3060 	 * Does target exist?
3061 	 */
3062 	if (tzp) {
3063 		/*
3064 		 * Source and target must be the same type.
3065 		 */
3066 		if (ZTOV(szp)->v_type == VDIR) {
3067 			if (ZTOV(tzp)->v_type != VDIR) {
3068 				error = ENOTDIR;
3069 				goto out;
3070 			}
3071 		} else {
3072 			if (ZTOV(tzp)->v_type == VDIR) {
3073 				error = EISDIR;
3074 				goto out;
3075 			}
3076 		}
3077 		/*
3078 		 * POSIX dictates that when the source and target
3079 		 * entries refer to the same file object, rename
3080 		 * must do nothing and exit without error.
3081 		 */
3082 		if (szp->z_id == tzp->z_id) {
3083 			error = 0;
3084 			goto out;
3085 		}
3086 	}
3087 
3088 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3089 	if (tzp)
3090 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3091 
3092 	/*
3093 	 * notify the target directory if it is not the same
3094 	 * as source directory.
3095 	 */
3096 	if (tdvp != sdvp) {
3097 		vnevent_rename_dest_dir(tdvp, ct);
3098 	}
3099 
3100 	tx = dmu_tx_create(zfsvfs->z_os);
3101 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
3102 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
3103 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3104 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3105 	if (sdzp != tdzp)
3106 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
3107 	if (tzp)
3108 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* parent changes */
3109 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3110 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3111 	if (error) {
3112 		if (zl != NULL)
3113 			zfs_rename_unlock(&zl);
3114 		zfs_dirent_unlock(sdl);
3115 		zfs_dirent_unlock(tdl);
3116 		VN_RELE(ZTOV(szp));
3117 		if (tzp)
3118 			VN_RELE(ZTOV(tzp));
3119 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3120 			dmu_tx_wait(tx);
3121 			dmu_tx_abort(tx);
3122 			goto top;
3123 		}
3124 		dmu_tx_abort(tx);
3125 		ZFS_EXIT(zfsvfs);
3126 		return (error);
3127 	}
3128 
3129 	if (tzp)	/* Attempt to remove the existing target */
3130 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3131 
3132 	if (error == 0) {
3133 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3134 		if (error == 0) {
3135 			szp->z_phys->zp_flags |= ZFS_AV_MODIFIED;
3136 
3137 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3138 			ASSERT(error == 0);
3139 
3140 			zfs_log_rename(zilog, tx,
3141 			    TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0),
3142 			    sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp);
3143 		}
3144 	}
3145 
3146 	dmu_tx_commit(tx);
3147 out:
3148 	if (zl != NULL)
3149 		zfs_rename_unlock(&zl);
3150 
3151 	zfs_dirent_unlock(sdl);
3152 	zfs_dirent_unlock(tdl);
3153 
3154 	VN_RELE(ZTOV(szp));
3155 	if (tzp)
3156 		VN_RELE(ZTOV(tzp));
3157 
3158 	ZFS_EXIT(zfsvfs);
3159 	return (error);
3160 }
3161 
3162 /*
3163  * Insert the indicated symbolic reference entry into the directory.
3164  *
3165  *	IN:	dvp	- Directory to contain new symbolic link.
3166  *		link	- Name for new symlink entry.
3167  *		vap	- Attributes of new entry.
3168  *		target	- Target path of new symlink.
3169  *		cr	- credentials of caller.
3170  *		ct	- caller context
3171  *		flags	- case flags
3172  *
3173  *	RETURN:	0 if success
3174  *		error code if failure
3175  *
3176  * Timestamps:
3177  *	dvp - ctime|mtime updated
3178  */
3179 /*ARGSUSED*/
3180 static int
3181 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3182     caller_context_t *ct, int flags)
3183 {
3184 	znode_t		*zp, *dzp = VTOZ(dvp);
3185 	zfs_dirlock_t	*dl;
3186 	dmu_tx_t	*tx;
3187 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3188 	zilog_t		*zilog;
3189 	int		len = strlen(link);
3190 	int		error;
3191 	int		zflg = ZNEW;
3192 	zfs_fuid_info_t *fuidp = NULL;
3193 
3194 	ASSERT(vap->va_type == VLNK);
3195 
3196 	ZFS_ENTER(zfsvfs);
3197 	ZFS_VERIFY_ZP(dzp);
3198 	zilog = zfsvfs->z_log;
3199 
3200 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3201 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3202 		ZFS_EXIT(zfsvfs);
3203 		return (EILSEQ);
3204 	}
3205 	if (flags & FIGNORECASE)
3206 		zflg |= ZCILOOK;
3207 top:
3208 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3209 		ZFS_EXIT(zfsvfs);
3210 		return (error);
3211 	}
3212 
3213 	if (len > MAXPATHLEN) {
3214 		ZFS_EXIT(zfsvfs);
3215 		return (ENAMETOOLONG);
3216 	}
3217 
3218 	/*
3219 	 * Attempt to lock directory; fail if entry already exists.
3220 	 */
3221 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3222 	if (error) {
3223 		ZFS_EXIT(zfsvfs);
3224 		return (error);
3225 	}
3226 
3227 	tx = dmu_tx_create(zfsvfs->z_os);
3228 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3229 	dmu_tx_hold_bonus(tx, dzp->z_id);
3230 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3231 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
3232 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
3233 	if (IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))) {
3234 		if (zfsvfs->z_fuid_obj == 0) {
3235 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
3236 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3237 			    FUID_SIZE_ESTIMATE(zfsvfs));
3238 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
3239 		} else {
3240 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
3241 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
3242 			    FUID_SIZE_ESTIMATE(zfsvfs));
3243 		}
3244 	}
3245 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3246 	if (error) {
3247 		zfs_dirent_unlock(dl);
3248 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3249 			dmu_tx_wait(tx);
3250 			dmu_tx_abort(tx);
3251 			goto top;
3252 		}
3253 		dmu_tx_abort(tx);
3254 		ZFS_EXIT(zfsvfs);
3255 		return (error);
3256 	}
3257 
3258 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
3259 
3260 	/*
3261 	 * Create a new object for the symlink.
3262 	 * Put the link content into bonus buffer if it will fit;
3263 	 * otherwise, store it just like any other file data.
3264 	 */
3265 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
3266 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, len, NULL, &fuidp);
3267 		if (len != 0)
3268 			bcopy(link, zp->z_phys + 1, len);
3269 	} else {
3270 		dmu_buf_t *dbp;
3271 
3272 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, NULL, &fuidp);
3273 		/*
3274 		 * Nothing can access the znode yet so no locking needed
3275 		 * for growing the znode's blocksize.
3276 		 */
3277 		zfs_grow_blocksize(zp, len, tx);
3278 
3279 		VERIFY(0 == dmu_buf_hold(zfsvfs->z_os,
3280 		    zp->z_id, 0, FTAG, &dbp));
3281 		dmu_buf_will_dirty(dbp, tx);
3282 
3283 		ASSERT3U(len, <=, dbp->db_size);
3284 		bcopy(link, dbp->db_data, len);
3285 		dmu_buf_rele(dbp, FTAG);
3286 	}
3287 	zp->z_phys->zp_size = len;
3288 
3289 	/*
3290 	 * Insert the new object into the directory.
3291 	 */
3292 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3293 out:
3294 	if (error == 0) {
3295 		uint64_t txtype = TX_SYMLINK;
3296 		if (flags & FIGNORECASE)
3297 			txtype |= TX_CI;
3298 		zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3299 	}
3300 	if (fuidp)
3301 		zfs_fuid_info_free(fuidp);
3302 
3303 	dmu_tx_commit(tx);
3304 
3305 	zfs_dirent_unlock(dl);
3306 
3307 	VN_RELE(ZTOV(zp));
3308 
3309 	ZFS_EXIT(zfsvfs);
3310 	return (error);
3311 }
3312 
3313 /*
3314  * Return, in the buffer contained in the provided uio structure,
3315  * the symbolic path referred to by vp.
3316  *
3317  *	IN:	vp	- vnode of symbolic link.
3318  *		uoip	- structure to contain the link path.
3319  *		cr	- credentials of caller.
3320  *		ct	- caller context
3321  *
3322  *	OUT:	uio	- structure to contain the link path.
3323  *
3324  *	RETURN:	0 if success
3325  *		error code if failure
3326  *
3327  * Timestamps:
3328  *	vp - atime updated
3329  */
3330 /* ARGSUSED */
3331 static int
3332 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3333 {
3334 	znode_t		*zp = VTOZ(vp);
3335 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3336 	size_t		bufsz;
3337 	int		error;
3338 
3339 	ZFS_ENTER(zfsvfs);
3340 	ZFS_VERIFY_ZP(zp);
3341 
3342 	bufsz = (size_t)zp->z_phys->zp_size;
3343 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
3344 		error = uiomove(zp->z_phys + 1,
3345 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3346 	} else {
3347 		dmu_buf_t *dbp;
3348 		error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp);
3349 		if (error) {
3350 			ZFS_EXIT(zfsvfs);
3351 			return (error);
3352 		}
3353 		error = uiomove(dbp->db_data,
3354 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3355 		dmu_buf_rele(dbp, FTAG);
3356 	}
3357 
3358 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3359 	ZFS_EXIT(zfsvfs);
3360 	return (error);
3361 }
3362 
3363 /*
3364  * Insert a new entry into directory tdvp referencing svp.
3365  *
3366  *	IN:	tdvp	- Directory to contain new entry.
3367  *		svp	- vnode of new entry.
3368  *		name	- name of new entry.
3369  *		cr	- credentials of caller.
3370  *		ct	- caller context
3371  *
3372  *	RETURN:	0 if success
3373  *		error code if failure
3374  *
3375  * Timestamps:
3376  *	tdvp - ctime|mtime updated
3377  *	 svp - ctime updated
3378  */
3379 /* ARGSUSED */
3380 static int
3381 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3382     caller_context_t *ct, int flags)
3383 {
3384 	znode_t		*dzp = VTOZ(tdvp);
3385 	znode_t		*tzp, *szp;
3386 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3387 	zilog_t		*zilog;
3388 	zfs_dirlock_t	*dl;
3389 	dmu_tx_t	*tx;
3390 	vnode_t		*realvp;
3391 	int		error;
3392 	int		zf = ZNEW;
3393 	uid_t		owner;
3394 
3395 	ASSERT(tdvp->v_type == VDIR);
3396 
3397 	ZFS_ENTER(zfsvfs);
3398 	ZFS_VERIFY_ZP(dzp);
3399 	zilog = zfsvfs->z_log;
3400 
3401 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3402 		svp = realvp;
3403 
3404 	if (svp->v_vfsp != tdvp->v_vfsp) {
3405 		ZFS_EXIT(zfsvfs);
3406 		return (EXDEV);
3407 	}
3408 	szp = VTOZ(svp);
3409 	ZFS_VERIFY_ZP(szp);
3410 
3411 	if (zfsvfs->z_utf8 && u8_validate(name,
3412 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3413 		ZFS_EXIT(zfsvfs);
3414 		return (EILSEQ);
3415 	}
3416 	if (flags & FIGNORECASE)
3417 		zf |= ZCILOOK;
3418 
3419 top:
3420 	/*
3421 	 * We do not support links between attributes and non-attributes
3422 	 * because of the potential security risk of creating links
3423 	 * into "normal" file space in order to circumvent restrictions
3424 	 * imposed in attribute space.
3425 	 */
3426 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
3427 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
3428 		ZFS_EXIT(zfsvfs);
3429 		return (EINVAL);
3430 	}
3431 
3432 	/*
3433 	 * POSIX dictates that we return EPERM here.
3434 	 * Better choices include ENOTSUP or EISDIR.
3435 	 */
3436 	if (svp->v_type == VDIR) {
3437 		ZFS_EXIT(zfsvfs);
3438 		return (EPERM);
3439 	}
3440 
3441 	owner = zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, cr, ZFS_OWNER);
3442 	if (owner != crgetuid(cr) &&
3443 	    secpolicy_basic_link(cr) != 0) {
3444 		ZFS_EXIT(zfsvfs);
3445 		return (EPERM);
3446 	}
3447 
3448 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3449 		ZFS_EXIT(zfsvfs);
3450 		return (error);
3451 	}
3452 
3453 	/*
3454 	 * Attempt to lock directory; fail if entry already exists.
3455 	 */
3456 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3457 	if (error) {
3458 		ZFS_EXIT(zfsvfs);
3459 		return (error);
3460 	}
3461 
3462 	tx = dmu_tx_create(zfsvfs->z_os);
3463 	dmu_tx_hold_bonus(tx, szp->z_id);
3464 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3465 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3466 	if (error) {
3467 		zfs_dirent_unlock(dl);
3468 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3469 			dmu_tx_wait(tx);
3470 			dmu_tx_abort(tx);
3471 			goto top;
3472 		}
3473 		dmu_tx_abort(tx);
3474 		ZFS_EXIT(zfsvfs);
3475 		return (error);
3476 	}
3477 
3478 	error = zfs_link_create(dl, szp, tx, 0);
3479 
3480 	if (error == 0) {
3481 		uint64_t txtype = TX_LINK;
3482 		if (flags & FIGNORECASE)
3483 			txtype |= TX_CI;
3484 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3485 	}
3486 
3487 	dmu_tx_commit(tx);
3488 
3489 	zfs_dirent_unlock(dl);
3490 
3491 	if (error == 0) {
3492 		vnevent_link(svp, ct);
3493 	}
3494 
3495 	ZFS_EXIT(zfsvfs);
3496 	return (error);
3497 }
3498 
3499 /*
3500  * zfs_null_putapage() is used when the file system has been force
3501  * unmounted. It just drops the pages.
3502  */
3503 /* ARGSUSED */
3504 static int
3505 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3506 		size_t *lenp, int flags, cred_t *cr)
3507 {
3508 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
3509 	return (0);
3510 }
3511 
3512 /*
3513  * Push a page out to disk, klustering if possible.
3514  *
3515  *	IN:	vp	- file to push page to.
3516  *		pp	- page to push.
3517  *		flags	- additional flags.
3518  *		cr	- credentials of caller.
3519  *
3520  *	OUT:	offp	- start of range pushed.
3521  *		lenp	- len of range pushed.
3522  *
3523  *	RETURN:	0 if success
3524  *		error code if failure
3525  *
3526  * NOTE: callers must have locked the page to be pushed.  On
3527  * exit, the page (and all other pages in the kluster) must be
3528  * unlocked.
3529  */
3530 /* ARGSUSED */
3531 static int
3532 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3533 		size_t *lenp, int flags, cred_t *cr)
3534 {
3535 	znode_t		*zp = VTOZ(vp);
3536 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3537 	zilog_t		*zilog = zfsvfs->z_log;
3538 	dmu_tx_t	*tx;
3539 	rl_t		*rl;
3540 	u_offset_t	off, koff;
3541 	size_t		len, klen;
3542 	uint64_t	filesz;
3543 	int		err;
3544 
3545 	filesz = zp->z_phys->zp_size;
3546 	off = pp->p_offset;
3547 	len = PAGESIZE;
3548 	/*
3549 	 * If our blocksize is bigger than the page size, try to kluster
3550 	 * muiltiple pages so that we write a full block (thus avoiding
3551 	 * a read-modify-write).
3552 	 */
3553 	if (off < filesz && zp->z_blksz > PAGESIZE) {
3554 		if (!ISP2(zp->z_blksz)) {
3555 			/* Only one block in the file. */
3556 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3557 			koff = 0;
3558 		} else {
3559 			klen = zp->z_blksz;
3560 			koff = P2ALIGN(off, (u_offset_t)klen);
3561 		}
3562 		ASSERT(koff <= filesz);
3563 		if (koff + klen > filesz)
3564 			klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE);
3565 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
3566 	}
3567 	ASSERT3U(btop(len), ==, btopr(len));
3568 top:
3569 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
3570 	/*
3571 	 * Can't push pages past end-of-file.
3572 	 */
3573 	filesz = zp->z_phys->zp_size;
3574 	if (off >= filesz) {
3575 		/* ignore all pages */
3576 		err = 0;
3577 		goto out;
3578 	} else if (off + len > filesz) {
3579 		int npages = btopr(filesz - off);
3580 		page_t *trunc;
3581 
3582 		page_list_break(&pp, &trunc, npages);
3583 		/* ignore pages past end of file */
3584 		if (trunc)
3585 			pvn_write_done(trunc, flags);
3586 		len = filesz - off;
3587 	}
3588 
3589 	tx = dmu_tx_create(zfsvfs->z_os);
3590 	dmu_tx_hold_write(tx, zp->z_id, off, len);
3591 	dmu_tx_hold_bonus(tx, zp->z_id);
3592 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
3593 	if (err != 0) {
3594 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3595 			zfs_range_unlock(rl);
3596 			dmu_tx_wait(tx);
3597 			dmu_tx_abort(tx);
3598 			err = 0;
3599 			goto top;
3600 		}
3601 		dmu_tx_abort(tx);
3602 		goto out;
3603 	}
3604 
3605 	if (zp->z_blksz <= PAGESIZE) {
3606 		caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1);
3607 		ASSERT3U(len, <=, PAGESIZE);
3608 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
3609 		ppmapout(va);
3610 	} else {
3611 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
3612 	}
3613 
3614 	if (err == 0) {
3615 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
3616 		zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0);
3617 		dmu_tx_commit(tx);
3618 	}
3619 
3620 out:
3621 	zfs_range_unlock(rl);
3622 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
3623 	if (offp)
3624 		*offp = off;
3625 	if (lenp)
3626 		*lenp = len;
3627 
3628 	return (err);
3629 }
3630 
3631 /*
3632  * Copy the portion of the file indicated from pages into the file.
3633  * The pages are stored in a page list attached to the files vnode.
3634  *
3635  *	IN:	vp	- vnode of file to push page data to.
3636  *		off	- position in file to put data.
3637  *		len	- amount of data to write.
3638  *		flags	- flags to control the operation.
3639  *		cr	- credentials of caller.
3640  *		ct	- caller context.
3641  *
3642  *	RETURN:	0 if success
3643  *		error code if failure
3644  *
3645  * Timestamps:
3646  *	vp - ctime|mtime updated
3647  */
3648 /*ARGSUSED*/
3649 static int
3650 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
3651     caller_context_t *ct)
3652 {
3653 	znode_t		*zp = VTOZ(vp);
3654 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3655 	page_t		*pp;
3656 	size_t		io_len;
3657 	u_offset_t	io_off;
3658 	uint64_t	filesz;
3659 	int		error = 0;
3660 
3661 	ZFS_ENTER(zfsvfs);
3662 	ZFS_VERIFY_ZP(zp);
3663 
3664 	if (len == 0) {
3665 		/*
3666 		 * Search the entire vp list for pages >= off.
3667 		 */
3668 		error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage,
3669 		    flags, cr);
3670 		goto out;
3671 	}
3672 
3673 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3674 	if (off > filesz) {
3675 		/* past end of file */
3676 		ZFS_EXIT(zfsvfs);
3677 		return (0);
3678 	}
3679 
3680 	len = MIN(len, filesz - off);
3681 
3682 	for (io_off = off; io_off < off + len; io_off += io_len) {
3683 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
3684 			pp = page_lookup(vp, io_off,
3685 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
3686 		} else {
3687 			pp = page_lookup_nowait(vp, io_off,
3688 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
3689 		}
3690 
3691 		if (pp != NULL && pvn_getdirty(pp, flags)) {
3692 			int err;
3693 
3694 			/*
3695 			 * Found a dirty page to push
3696 			 */
3697 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
3698 			if (err)
3699 				error = err;
3700 		} else {
3701 			io_len = PAGESIZE;
3702 		}
3703 	}
3704 out:
3705 	if ((flags & B_ASYNC) == 0)
3706 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
3707 	ZFS_EXIT(zfsvfs);
3708 	return (error);
3709 }
3710 
3711 /*ARGSUSED*/
3712 void
3713 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
3714 {
3715 	znode_t	*zp = VTOZ(vp);
3716 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3717 	int error;
3718 
3719 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
3720 	if (zp->z_dbuf == NULL) {
3721 		/*
3722 		 * The fs has been unmounted, or we did a
3723 		 * suspend/resume and this file no longer exists.
3724 		 */
3725 		if (vn_has_cached_data(vp)) {
3726 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
3727 			    B_INVAL, cr);
3728 		}
3729 
3730 		mutex_enter(&zp->z_lock);
3731 		vp->v_count = 0; /* count arrives as 1 */
3732 		mutex_exit(&zp->z_lock);
3733 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
3734 		zfs_znode_free(zp);
3735 		return;
3736 	}
3737 
3738 	/*
3739 	 * Attempt to push any data in the page cache.  If this fails
3740 	 * we will get kicked out later in zfs_zinactive().
3741 	 */
3742 	if (vn_has_cached_data(vp)) {
3743 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
3744 		    cr);
3745 	}
3746 
3747 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3748 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3749 
3750 		dmu_tx_hold_bonus(tx, zp->z_id);
3751 		error = dmu_tx_assign(tx, TXG_WAIT);
3752 		if (error) {
3753 			dmu_tx_abort(tx);
3754 		} else {
3755 			dmu_buf_will_dirty(zp->z_dbuf, tx);
3756 			mutex_enter(&zp->z_lock);
3757 			zp->z_atime_dirty = 0;
3758 			mutex_exit(&zp->z_lock);
3759 			dmu_tx_commit(tx);
3760 		}
3761 	}
3762 
3763 	zfs_zinactive(zp);
3764 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
3765 }
3766 
3767 /*
3768  * Bounds-check the seek operation.
3769  *
3770  *	IN:	vp	- vnode seeking within
3771  *		ooff	- old file offset
3772  *		noffp	- pointer to new file offset
3773  *		ct	- caller context
3774  *
3775  *	RETURN:	0 if success
3776  *		EINVAL if new offset invalid
3777  */
3778 /* ARGSUSED */
3779 static int
3780 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
3781     caller_context_t *ct)
3782 {
3783 	if (vp->v_type == VDIR)
3784 		return (0);
3785 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3786 }
3787 
3788 /*
3789  * Pre-filter the generic locking function to trap attempts to place
3790  * a mandatory lock on a memory mapped file.
3791  */
3792 static int
3793 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3794     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
3795 {
3796 	znode_t *zp = VTOZ(vp);
3797 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3798 	int error;
3799 
3800 	ZFS_ENTER(zfsvfs);
3801 	ZFS_VERIFY_ZP(zp);
3802 
3803 	/*
3804 	 * We are following the UFS semantics with respect to mapcnt
3805 	 * here: If we see that the file is mapped already, then we will
3806 	 * return an error, but we don't worry about races between this
3807 	 * function and zfs_map().
3808 	 */
3809 	if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3810 		ZFS_EXIT(zfsvfs);
3811 		return (EAGAIN);
3812 	}
3813 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct);
3814 	ZFS_EXIT(zfsvfs);
3815 	return (error);
3816 }
3817 
3818 /*
3819  * If we can't find a page in the cache, we will create a new page
3820  * and fill it with file data.  For efficiency, we may try to fill
3821  * multiple pages at once (klustering).
3822  */
3823 static int
3824 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3825     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3826 {
3827 	znode_t *zp = VTOZ(vp);
3828 	page_t *pp, *cur_pp;
3829 	objset_t *os = zp->z_zfsvfs->z_os;
3830 	caddr_t va;
3831 	u_offset_t io_off, total;
3832 	uint64_t oid = zp->z_id;
3833 	size_t io_len;
3834 	uint64_t filesz;
3835 	int err;
3836 
3837 	/*
3838 	 * If we are only asking for a single page don't bother klustering.
3839 	 */
3840 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3841 	if (off >= filesz)
3842 		return (EFAULT);
3843 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3844 		io_off = off;
3845 		io_len = PAGESIZE;
3846 		pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr);
3847 	} else {
3848 		/*
3849 		 * Try to fill a kluster of pages (a blocks worth).
3850 		 */
3851 		size_t klen;
3852 		u_offset_t koff;
3853 
3854 		if (!ISP2(zp->z_blksz)) {
3855 			/* Only one block in the file. */
3856 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3857 			koff = 0;
3858 		} else {
3859 			/*
3860 			 * It would be ideal to align our offset to the
3861 			 * blocksize but doing so has resulted in some
3862 			 * strange application crashes. For now, we
3863 			 * leave the offset as is and only adjust the
3864 			 * length if we are off the end of the file.
3865 			 */
3866 			koff = off;
3867 			klen = plsz;
3868 		}
3869 		ASSERT(koff <= filesz);
3870 		if (koff + klen > filesz)
3871 			klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff;
3872 		ASSERT3U(off, >=, koff);
3873 		ASSERT3U(off, <, koff + klen);
3874 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3875 		    &io_len, koff, klen, 0);
3876 	}
3877 	if (pp == NULL) {
3878 		/*
3879 		 * Some other thread entered the page before us.
3880 		 * Return to zfs_getpage to retry the lookup.
3881 		 */
3882 		*pl = NULL;
3883 		return (0);
3884 	}
3885 
3886 	/*
3887 	 * Fill the pages in the kluster.
3888 	 */
3889 	cur_pp = pp;
3890 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3891 		ASSERT3U(io_off, ==, cur_pp->p_offset);
3892 		va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1);
3893 		err = dmu_read(os, oid, io_off, PAGESIZE, va);
3894 		ppmapout(va);
3895 		if (err) {
3896 			/* On error, toss the entire kluster */
3897 			pvn_read_done(pp, B_ERROR);
3898 			return (err);
3899 		}
3900 		cur_pp = cur_pp->p_next;
3901 	}
3902 out:
3903 	/*
3904 	 * Fill in the page list array from the kluster.  If
3905 	 * there are too many pages in the kluster, return
3906 	 * as many pages as possible starting from the desired
3907 	 * offset `off'.
3908 	 * NOTE: the page list will always be null terminated.
3909 	 */
3910 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3911 
3912 	return (0);
3913 }
3914 
3915 /*
3916  * Return pointers to the pages for the file region [off, off + len]
3917  * in the pl array.  If plsz is greater than len, this function may
3918  * also return page pointers from before or after the specified
3919  * region (i.e. some region [off', off' + plsz]).  These additional
3920  * pages are only returned if they are already in the cache, or were
3921  * created as part of a klustered read.
3922  *
3923  *	IN:	vp	- vnode of file to get data from.
3924  *		off	- position in file to get data from.
3925  *		len	- amount of data to retrieve.
3926  *		plsz	- length of provided page list.
3927  *		seg	- segment to obtain pages for.
3928  *		addr	- virtual address of fault.
3929  *		rw	- mode of created pages.
3930  *		cr	- credentials of caller.
3931  *		ct	- caller context.
3932  *
3933  *	OUT:	protp	- protection mode of created pages.
3934  *		pl	- list of pages created.
3935  *
3936  *	RETURN:	0 if success
3937  *		error code if failure
3938  *
3939  * Timestamps:
3940  *	vp - atime updated
3941  */
3942 /* ARGSUSED */
3943 static int
3944 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3945 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3946 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
3947 {
3948 	znode_t		*zp = VTOZ(vp);
3949 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3950 	page_t		*pp, **pl0 = pl;
3951 	int		need_unlock = 0, err = 0;
3952 	offset_t	orig_off;
3953 
3954 	ZFS_ENTER(zfsvfs);
3955 	ZFS_VERIFY_ZP(zp);
3956 
3957 	if (protp)
3958 		*protp = PROT_ALL;
3959 
3960 	/* no faultahead (for now) */
3961 	if (pl == NULL) {
3962 		ZFS_EXIT(zfsvfs);
3963 		return (0);
3964 	}
3965 
3966 	/* can't fault past EOF */
3967 	if (off >= zp->z_phys->zp_size) {
3968 		ZFS_EXIT(zfsvfs);
3969 		return (EFAULT);
3970 	}
3971 	orig_off = off;
3972 
3973 	/*
3974 	 * If we already own the lock, then we must be page faulting
3975 	 * in the middle of a write to this file (i.e., we are writing
3976 	 * to this file using data from a mapped region of the file).
3977 	 */
3978 	if (rw_owner(&zp->z_map_lock) != curthread) {
3979 		rw_enter(&zp->z_map_lock, RW_WRITER);
3980 		need_unlock = TRUE;
3981 	}
3982 
3983 	/*
3984 	 * Loop through the requested range [off, off + len] looking
3985 	 * for pages.  If we don't find a page, we will need to create
3986 	 * a new page and fill it with data from the file.
3987 	 */
3988 	while (len > 0) {
3989 		if (plsz < PAGESIZE)
3990 			break;
3991 		if (pp = page_lookup(vp, off, SE_SHARED)) {
3992 			*pl++ = pp;
3993 			off += PAGESIZE;
3994 			addr += PAGESIZE;
3995 			len -= PAGESIZE;
3996 			plsz -= PAGESIZE;
3997 		} else {
3998 			err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw);
3999 			if (err)
4000 				goto out;
4001 			/*
4002 			 * klustering may have changed our region
4003 			 * to be block aligned.
4004 			 */
4005 			if (((pp = *pl) != 0) && (off != pp->p_offset)) {
4006 				int delta = off - pp->p_offset;
4007 				len += delta;
4008 				off -= delta;
4009 				addr -= delta;
4010 			}
4011 			while (*pl) {
4012 				pl++;
4013 				off += PAGESIZE;
4014 				addr += PAGESIZE;
4015 				plsz -= PAGESIZE;
4016 				if (len > PAGESIZE)
4017 					len -= PAGESIZE;
4018 				else
4019 					len = 0;
4020 			}
4021 		}
4022 	}
4023 
4024 	/*
4025 	 * Fill out the page array with any pages already in the cache.
4026 	 */
4027 	while (plsz > 0) {
4028 		pp = page_lookup_nowait(vp, off, SE_SHARED);
4029 		if (pp == NULL)
4030 			break;
4031 		*pl++ = pp;
4032 		off += PAGESIZE;
4033 		plsz -= PAGESIZE;
4034 	}
4035 
4036 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4037 out:
4038 	/*
4039 	 * We can't grab the range lock for the page as reader which would
4040 	 * stop truncation as this leads to deadlock. So we need to recheck
4041 	 * the file size.
4042 	 */
4043 	if (orig_off >= zp->z_phys->zp_size)
4044 		err = EFAULT;
4045 	if (err) {
4046 		/*
4047 		 * Release any pages we have previously locked.
4048 		 */
4049 		while (pl > pl0)
4050 			page_unlock(*--pl);
4051 	}
4052 
4053 	*pl = NULL;
4054 
4055 	if (need_unlock)
4056 		rw_exit(&zp->z_map_lock);
4057 
4058 	ZFS_EXIT(zfsvfs);
4059 	return (err);
4060 }
4061 
4062 /*
4063  * Request a memory map for a section of a file.  This code interacts
4064  * with common code and the VM system as follows:
4065  *
4066  *	common code calls mmap(), which ends up in smmap_common()
4067  *
4068  *	this calls VOP_MAP(), which takes you into (say) zfs
4069  *
4070  *	zfs_map() calls as_map(), passing segvn_create() as the callback
4071  *
4072  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
4073  *
4074  *	zfs_addmap() updates z_mapcnt
4075  */
4076 /*ARGSUSED*/
4077 static int
4078 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4079     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4080     caller_context_t *ct)
4081 {
4082 	znode_t *zp = VTOZ(vp);
4083 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4084 	segvn_crargs_t	vn_a;
4085 	int		error;
4086 
4087 	ZFS_ENTER(zfsvfs);
4088 	ZFS_VERIFY_ZP(zp);
4089 
4090 	if ((prot & PROT_WRITE) &&
4091 	    (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY |
4092 	    ZFS_APPENDONLY))) {
4093 		ZFS_EXIT(zfsvfs);
4094 		return (EPERM);
4095 	}
4096 
4097 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4098 	    (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED)) {
4099 		ZFS_EXIT(zfsvfs);
4100 		return (EACCES);
4101 	}
4102 
4103 	if (vp->v_flag & VNOMAP) {
4104 		ZFS_EXIT(zfsvfs);
4105 		return (ENOSYS);
4106 	}
4107 
4108 	if (off < 0 || len > MAXOFFSET_T - off) {
4109 		ZFS_EXIT(zfsvfs);
4110 		return (ENXIO);
4111 	}
4112 
4113 	if (vp->v_type != VREG) {
4114 		ZFS_EXIT(zfsvfs);
4115 		return (ENODEV);
4116 	}
4117 
4118 	/*
4119 	 * If file is locked, disallow mapping.
4120 	 */
4121 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) {
4122 		ZFS_EXIT(zfsvfs);
4123 		return (EAGAIN);
4124 	}
4125 
4126 	as_rangelock(as);
4127 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4128 	if (error != 0) {
4129 		as_rangeunlock(as);
4130 		ZFS_EXIT(zfsvfs);
4131 		return (error);
4132 	}
4133 
4134 	vn_a.vp = vp;
4135 	vn_a.offset = (u_offset_t)off;
4136 	vn_a.type = flags & MAP_TYPE;
4137 	vn_a.prot = prot;
4138 	vn_a.maxprot = maxprot;
4139 	vn_a.cred = cr;
4140 	vn_a.amp = NULL;
4141 	vn_a.flags = flags & ~MAP_TYPE;
4142 	vn_a.szc = 0;
4143 	vn_a.lgrp_mem_policy_flags = 0;
4144 
4145 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4146 
4147 	as_rangeunlock(as);
4148 	ZFS_EXIT(zfsvfs);
4149 	return (error);
4150 }
4151 
4152 /* ARGSUSED */
4153 static int
4154 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4155     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4156     caller_context_t *ct)
4157 {
4158 	uint64_t pages = btopr(len);
4159 
4160 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4161 	return (0);
4162 }
4163 
4164 /*
4165  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4166  * more accurate mtime for the associated file.  Since we don't have a way of
4167  * detecting when the data was actually modified, we have to resort to
4168  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4169  * last page is pushed.  The problem occurs when the msync() call is omitted,
4170  * which by far the most common case:
4171  *
4172  * 	open()
4173  * 	mmap()
4174  * 	<modify memory>
4175  * 	munmap()
4176  * 	close()
4177  * 	<time lapse>
4178  * 	putpage() via fsflush
4179  *
4180  * If we wait until fsflush to come along, we can have a modification time that
4181  * is some arbitrary point in the future.  In order to prevent this in the
4182  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4183  * torn down.
4184  */
4185 /* ARGSUSED */
4186 static int
4187 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4188     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4189     caller_context_t *ct)
4190 {
4191 	uint64_t pages = btopr(len);
4192 
4193 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4194 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4195 
4196 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4197 	    vn_has_cached_data(vp))
4198 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4199 
4200 	return (0);
4201 }
4202 
4203 /*
4204  * Free or allocate space in a file.  Currently, this function only
4205  * supports the `F_FREESP' command.  However, this command is somewhat
4206  * misnamed, as its functionality includes the ability to allocate as
4207  * well as free space.
4208  *
4209  *	IN:	vp	- vnode of file to free data in.
4210  *		cmd	- action to take (only F_FREESP supported).
4211  *		bfp	- section of file to free/alloc.
4212  *		flag	- current file open mode flags.
4213  *		offset	- current file offset.
4214  *		cr	- credentials of caller [UNUSED].
4215  *		ct	- caller context.
4216  *
4217  *	RETURN:	0 if success
4218  *		error code if failure
4219  *
4220  * Timestamps:
4221  *	vp - ctime|mtime updated
4222  */
4223 /* ARGSUSED */
4224 static int
4225 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4226     offset_t offset, cred_t *cr, caller_context_t *ct)
4227 {
4228 	znode_t		*zp = VTOZ(vp);
4229 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4230 	uint64_t	off, len;
4231 	int		error;
4232 
4233 	ZFS_ENTER(zfsvfs);
4234 	ZFS_VERIFY_ZP(zp);
4235 
4236 top:
4237 	if (cmd != F_FREESP) {
4238 		ZFS_EXIT(zfsvfs);
4239 		return (EINVAL);
4240 	}
4241 
4242 	if (error = convoff(vp, bfp, 0, offset)) {
4243 		ZFS_EXIT(zfsvfs);
4244 		return (error);
4245 	}
4246 
4247 	if (bfp->l_len < 0) {
4248 		ZFS_EXIT(zfsvfs);
4249 		return (EINVAL);
4250 	}
4251 
4252 	off = bfp->l_start;
4253 	len = bfp->l_len; /* 0 means from off to end of file */
4254 
4255 	do {
4256 		error = zfs_freesp(zp, off, len, flag, TRUE);
4257 		/* NB: we already did dmu_tx_wait() if necessary */
4258 	} while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
4259 
4260 	ZFS_EXIT(zfsvfs);
4261 	return (error);
4262 }
4263 
4264 /*ARGSUSED*/
4265 static int
4266 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4267 {
4268 	znode_t		*zp = VTOZ(vp);
4269 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4270 	uint32_t	gen;
4271 	uint64_t	object = zp->z_id;
4272 	zfid_short_t	*zfid;
4273 	int		size, i;
4274 
4275 	ZFS_ENTER(zfsvfs);
4276 	ZFS_VERIFY_ZP(zp);
4277 	gen = (uint32_t)zp->z_gen;
4278 
4279 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4280 	if (fidp->fid_len < size) {
4281 		fidp->fid_len = size;
4282 		ZFS_EXIT(zfsvfs);
4283 		return (ENOSPC);
4284 	}
4285 
4286 	zfid = (zfid_short_t *)fidp;
4287 
4288 	zfid->zf_len = size;
4289 
4290 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4291 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4292 
4293 	/* Must have a non-zero generation number to distinguish from .zfs */
4294 	if (gen == 0)
4295 		gen = 1;
4296 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4297 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4298 
4299 	if (size == LONG_FID_LEN) {
4300 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4301 		zfid_long_t	*zlfid;
4302 
4303 		zlfid = (zfid_long_t *)fidp;
4304 
4305 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4306 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4307 
4308 		/* XXX - this should be the generation number for the objset */
4309 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4310 			zlfid->zf_setgen[i] = 0;
4311 	}
4312 
4313 	ZFS_EXIT(zfsvfs);
4314 	return (0);
4315 }
4316 
4317 static int
4318 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4319     caller_context_t *ct)
4320 {
4321 	znode_t		*zp, *xzp;
4322 	zfsvfs_t	*zfsvfs;
4323 	zfs_dirlock_t	*dl;
4324 	int		error;
4325 
4326 	switch (cmd) {
4327 	case _PC_LINK_MAX:
4328 		*valp = ULONG_MAX;
4329 		return (0);
4330 
4331 	case _PC_FILESIZEBITS:
4332 		*valp = 64;
4333 		return (0);
4334 
4335 	case _PC_XATTR_EXISTS:
4336 		zp = VTOZ(vp);
4337 		zfsvfs = zp->z_zfsvfs;
4338 		ZFS_ENTER(zfsvfs);
4339 		ZFS_VERIFY_ZP(zp);
4340 		*valp = 0;
4341 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4342 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4343 		if (error == 0) {
4344 			zfs_dirent_unlock(dl);
4345 			if (!zfs_dirempty(xzp))
4346 				*valp = 1;
4347 			VN_RELE(ZTOV(xzp));
4348 		} else if (error == ENOENT) {
4349 			/*
4350 			 * If there aren't extended attributes, it's the
4351 			 * same as having zero of them.
4352 			 */
4353 			error = 0;
4354 		}
4355 		ZFS_EXIT(zfsvfs);
4356 		return (error);
4357 
4358 	case _PC_SATTR_ENABLED:
4359 	case _PC_SATTR_EXISTS:
4360 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) &&
4361 		    (vp->v_type == VREG || vp->v_type == VDIR);
4362 		return (0);
4363 
4364 	case _PC_ACL_ENABLED:
4365 		*valp = _ACL_ACE_ENABLED;
4366 		return (0);
4367 
4368 	case _PC_MIN_HOLE_SIZE:
4369 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4370 		return (0);
4371 
4372 	default:
4373 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4374 	}
4375 }
4376 
4377 /*ARGSUSED*/
4378 static int
4379 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4380     caller_context_t *ct)
4381 {
4382 	znode_t *zp = VTOZ(vp);
4383 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4384 	int error;
4385 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4386 
4387 	ZFS_ENTER(zfsvfs);
4388 	ZFS_VERIFY_ZP(zp);
4389 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4390 	ZFS_EXIT(zfsvfs);
4391 
4392 	return (error);
4393 }
4394 
4395 /*ARGSUSED*/
4396 static int
4397 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4398     caller_context_t *ct)
4399 {
4400 	znode_t *zp = VTOZ(vp);
4401 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4402 	int error;
4403 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4404 
4405 	ZFS_ENTER(zfsvfs);
4406 	ZFS_VERIFY_ZP(zp);
4407 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4408 	ZFS_EXIT(zfsvfs);
4409 	return (error);
4410 }
4411 
4412 /*
4413  * Predeclare these here so that the compiler assumes that
4414  * this is an "old style" function declaration that does
4415  * not include arguments => we won't get type mismatch errors
4416  * in the initializations that follow.
4417  */
4418 static int zfs_inval();
4419 static int zfs_isdir();
4420 
4421 static int
4422 zfs_inval()
4423 {
4424 	return (EINVAL);
4425 }
4426 
4427 static int
4428 zfs_isdir()
4429 {
4430 	return (EISDIR);
4431 }
4432 /*
4433  * Directory vnode operations template
4434  */
4435 vnodeops_t *zfs_dvnodeops;
4436 const fs_operation_def_t zfs_dvnodeops_template[] = {
4437 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4438 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4439 	VOPNAME_READ,		{ .error = zfs_isdir },
4440 	VOPNAME_WRITE,		{ .error = zfs_isdir },
4441 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4442 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4443 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4444 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4445 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4446 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4447 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4448 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4449 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4450 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
4451 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4452 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4453 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
4454 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4455 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4456 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4457 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4458 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4459 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4460 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4461 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
4462 	NULL,			NULL
4463 };
4464 
4465 /*
4466  * Regular file vnode operations template
4467  */
4468 vnodeops_t *zfs_fvnodeops;
4469 const fs_operation_def_t zfs_fvnodeops_template[] = {
4470 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4471 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4472 	VOPNAME_READ,		{ .vop_read = zfs_read },
4473 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
4474 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4475 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4476 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4477 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4478 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4479 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4480 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4481 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4482 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4483 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4484 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
4485 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
4486 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
4487 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
4488 	VOPNAME_MAP,		{ .vop_map = zfs_map },
4489 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
4490 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
4491 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4492 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4493 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4494 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4495 	NULL,			NULL
4496 };
4497 
4498 /*
4499  * Symbolic link vnode operations template
4500  */
4501 vnodeops_t *zfs_symvnodeops;
4502 const fs_operation_def_t zfs_symvnodeops_template[] = {
4503 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4504 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4505 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4506 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4507 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
4508 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4509 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4510 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4511 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4512 	NULL,			NULL
4513 };
4514 
4515 /*
4516  * Extended attribute directory vnode operations template
4517  *	This template is identical to the directory vnodes
4518  *	operation template except for restricted operations:
4519  *		VOP_MKDIR()
4520  *		VOP_SYMLINK()
4521  * Note that there are other restrictions embedded in:
4522  *	zfs_create()	- restrict type to VREG
4523  *	zfs_link()	- no links into/out of attribute space
4524  *	zfs_rename()	- no moves into/out of attribute space
4525  */
4526 vnodeops_t *zfs_xdvnodeops;
4527 const fs_operation_def_t zfs_xdvnodeops_template[] = {
4528 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4529 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4530 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4531 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4532 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4533 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4534 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4535 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4536 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4537 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4538 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4539 	VOPNAME_MKDIR,		{ .error = zfs_inval },
4540 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4541 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4542 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
4543 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4544 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4545 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4546 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4547 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4548 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4549 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4550 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4551 	NULL,			NULL
4552 };
4553 
4554 /*
4555  * Error vnode operations template
4556  */
4557 vnodeops_t *zfs_evnodeops;
4558 const fs_operation_def_t zfs_evnodeops_template[] = {
4559 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4560 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4561 	NULL,			NULL
4562 };
4563