xref: /titanic_44/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 903a11ebdc8df157c4700150f41f1f262f4a8ae8)
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 			/* we can't hold any locks when calling zfs_freesp() */
1308 			zfs_dirent_unlock(dl);
1309 			dl = NULL;
1310 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1311 			if (error == 0) {
1312 				vnevent_create(ZTOV(zp), ct);
1313 			}
1314 		}
1315 	}
1316 out:
1317 
1318 	if (dl)
1319 		zfs_dirent_unlock(dl);
1320 
1321 	if (error) {
1322 		if (zp)
1323 			VN_RELE(ZTOV(zp));
1324 	} else {
1325 		*vpp = ZTOV(zp);
1326 		/*
1327 		 * If vnode is for a device return a specfs vnode instead.
1328 		 */
1329 		if (IS_DEVVP(*vpp)) {
1330 			struct vnode *svp;
1331 
1332 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1333 			VN_RELE(*vpp);
1334 			if (svp == NULL) {
1335 				error = ENOSYS;
1336 			}
1337 			*vpp = svp;
1338 		}
1339 	}
1340 	if (aclp)
1341 		zfs_acl_free(aclp);
1342 
1343 	ZFS_EXIT(zfsvfs);
1344 	return (error);
1345 }
1346 
1347 /*
1348  * Remove an entry from a directory.
1349  *
1350  *	IN:	dvp	- vnode of directory to remove entry from.
1351  *		name	- name of entry to remove.
1352  *		cr	- credentials of caller.
1353  *		ct	- caller context
1354  *		flags	- case flags
1355  *
1356  *	RETURN:	0 if success
1357  *		error code if failure
1358  *
1359  * Timestamps:
1360  *	dvp - ctime|mtime
1361  *	 vp - ctime (if nlink > 0)
1362  */
1363 /*ARGSUSED*/
1364 static int
1365 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1366     int flags)
1367 {
1368 	znode_t		*zp, *dzp = VTOZ(dvp);
1369 	znode_t		*xzp = NULL;
1370 	vnode_t		*vp;
1371 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1372 	zilog_t		*zilog;
1373 	uint64_t	acl_obj, xattr_obj;
1374 	zfs_dirlock_t	*dl;
1375 	dmu_tx_t	*tx;
1376 	boolean_t	may_delete_now, delete_now = FALSE;
1377 	boolean_t	unlinked, toobig = FALSE;
1378 	uint64_t	txtype;
1379 	pathname_t	*realnmp = NULL;
1380 	pathname_t	realnm;
1381 	int		error;
1382 	int		zflg = ZEXISTS;
1383 
1384 	ZFS_ENTER(zfsvfs);
1385 	ZFS_VERIFY_ZP(dzp);
1386 	zilog = zfsvfs->z_log;
1387 
1388 	if (flags & FIGNORECASE) {
1389 		zflg |= ZCILOOK;
1390 		pn_alloc(&realnm);
1391 		realnmp = &realnm;
1392 	}
1393 
1394 top:
1395 	/*
1396 	 * Attempt to lock directory; fail if entry doesn't exist.
1397 	 */
1398 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1399 	    NULL, realnmp)) {
1400 		if (realnmp)
1401 			pn_free(realnmp);
1402 		ZFS_EXIT(zfsvfs);
1403 		return (error);
1404 	}
1405 
1406 	vp = ZTOV(zp);
1407 
1408 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1409 		goto out;
1410 	}
1411 
1412 	/*
1413 	 * Need to use rmdir for removing directories.
1414 	 */
1415 	if (vp->v_type == VDIR) {
1416 		error = EPERM;
1417 		goto out;
1418 	}
1419 
1420 	vnevent_remove(vp, dvp, name, ct);
1421 
1422 	if (realnmp)
1423 		dnlc_remove(dvp, realnmp->pn_buf);
1424 	else
1425 		dnlc_remove(dvp, name);
1426 
1427 	mutex_enter(&vp->v_lock);
1428 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1429 	mutex_exit(&vp->v_lock);
1430 
1431 	/*
1432 	 * We may delete the znode now, or we may put it in the unlinked set;
1433 	 * it depends on whether we're the last link, and on whether there are
1434 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1435 	 * allow for either case.
1436 	 */
1437 	tx = dmu_tx_create(zfsvfs->z_os);
1438 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1439 	dmu_tx_hold_bonus(tx, zp->z_id);
1440 	if (may_delete_now) {
1441 		toobig =
1442 		    zp->z_phys->zp_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1443 		/* if the file is too big, only hold_free a token amount */
1444 		dmu_tx_hold_free(tx, zp->z_id, 0,
1445 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1446 	}
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 && !toobig &&
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 is 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 		/* XXX - would it be OK to generate a log record here? */
2455 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2456 		if (err) {
2457 			ZFS_EXIT(zfsvfs);
2458 			return (err);
2459 		}
2460 	}
2461 
2462 	if (mask & (AT_ATIME|AT_MTIME) ||
2463 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2464 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2465 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2466 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2467 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM))))
2468 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2469 		    skipaclchk, cr);
2470 
2471 	if (mask & (AT_UID|AT_GID)) {
2472 		int	idmask = (mask & (AT_UID|AT_GID));
2473 		int	take_owner;
2474 		int	take_group;
2475 
2476 		/*
2477 		 * NOTE: even if a new mode is being set,
2478 		 * we may clear S_ISUID/S_ISGID bits.
2479 		 */
2480 
2481 		if (!(mask & AT_MODE))
2482 			vap->va_mode = pzp->zp_mode;
2483 
2484 		/*
2485 		 * Take ownership or chgrp to group we are a member of
2486 		 */
2487 
2488 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2489 		take_group = (mask & AT_GID) &&
2490 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2491 
2492 		/*
2493 		 * If both AT_UID and AT_GID are set then take_owner and
2494 		 * take_group must both be set in order to allow taking
2495 		 * ownership.
2496 		 *
2497 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2498 		 *
2499 		 */
2500 
2501 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2502 		    ((idmask == AT_UID) && take_owner) ||
2503 		    ((idmask == AT_GID) && take_group)) {
2504 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2505 			    skipaclchk, cr) == 0) {
2506 				/*
2507 				 * Remove setuid/setgid for non-privileged users
2508 				 */
2509 				secpolicy_setid_clear(vap, cr);
2510 				trim_mask = (mask & (AT_UID|AT_GID));
2511 			} else {
2512 				need_policy =  TRUE;
2513 			}
2514 		} else {
2515 			need_policy =  TRUE;
2516 		}
2517 	}
2518 
2519 	mutex_enter(&zp->z_lock);
2520 	oldva.va_mode = pzp->zp_mode;
2521 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2522 	if (mask & AT_XVATTR) {
2523 		if ((need_policy == FALSE) &&
2524 		    (XVA_ISSET_REQ(xvap, XAT_APPENDONLY) &&
2525 		    xoap->xoa_appendonly !=
2526 		    ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) ||
2527 		    (XVA_ISSET_REQ(xvap, XAT_NOUNLINK) &&
2528 		    xoap->xoa_nounlink !=
2529 		    ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) ||
2530 		    (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE) &&
2531 		    xoap->xoa_immutable !=
2532 		    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) ||
2533 		    (XVA_ISSET_REQ(xvap, XAT_NODUMP) &&
2534 		    xoap->xoa_nodump !=
2535 		    ((pzp->zp_flags & ZFS_NODUMP) != 0)) ||
2536 		    (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED) &&
2537 		    xoap->xoa_av_modified !=
2538 		    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) ||
2539 		    ((XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED) &&
2540 		    ((vp->v_type != VREG && xoap->xoa_av_quarantined) ||
2541 		    xoap->xoa_av_quarantined !=
2542 		    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)))) ||
2543 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
2544 		    (XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2545 			need_policy = TRUE;
2546 		}
2547 	}
2548 
2549 	mutex_exit(&zp->z_lock);
2550 
2551 	if (mask & AT_MODE) {
2552 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2553 			err = secpolicy_setid_setsticky_clear(vp, vap,
2554 			    &oldva, cr);
2555 			if (err) {
2556 				ZFS_EXIT(zfsvfs);
2557 				return (err);
2558 			}
2559 			trim_mask |= AT_MODE;
2560 		} else {
2561 			need_policy = TRUE;
2562 		}
2563 	}
2564 
2565 	if (need_policy) {
2566 		/*
2567 		 * If trim_mask is set then take ownership
2568 		 * has been granted or write_acl is present and user
2569 		 * has the ability to modify mode.  In that case remove
2570 		 * UID|GID and or MODE from mask so that
2571 		 * secpolicy_vnode_setattr() doesn't revoke it.
2572 		 */
2573 
2574 		if (trim_mask) {
2575 			saved_mask = vap->va_mask;
2576 			vap->va_mask &= ~trim_mask;
2577 		}
2578 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2579 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2580 		if (err) {
2581 			ZFS_EXIT(zfsvfs);
2582 			return (err);
2583 		}
2584 
2585 		if (trim_mask)
2586 			vap->va_mask |= saved_mask;
2587 	}
2588 
2589 	/*
2590 	 * secpolicy_vnode_setattr, or take ownership may have
2591 	 * changed va_mask
2592 	 */
2593 	mask = vap->va_mask;
2594 
2595 	tx = dmu_tx_create(zfsvfs->z_os);
2596 	dmu_tx_hold_bonus(tx, zp->z_id);
2597 	if (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2598 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid))) {
2599 		if (zfsvfs->z_fuid_obj == 0) {
2600 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
2601 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2602 			    FUID_SIZE_ESTIMATE(zfsvfs));
2603 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
2604 		} else {
2605 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
2606 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
2607 			    FUID_SIZE_ESTIMATE(zfsvfs));
2608 		}
2609 	}
2610 
2611 	if (mask & AT_MODE) {
2612 		uint64_t pmode = pzp->zp_mode;
2613 
2614 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2615 
2616 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) {
2617 			dmu_tx_abort(tx);
2618 			ZFS_EXIT(zfsvfs);
2619 			return (err);
2620 		}
2621 		if (pzp->zp_acl.z_acl_extern_obj) {
2622 			/* Are we upgrading ACL from old V0 format to new V1 */
2623 			if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2624 			    pzp->zp_acl.z_acl_version ==
2625 			    ZFS_ACL_VERSION_INITIAL) {
2626 				dmu_tx_hold_free(tx,
2627 				    pzp->zp_acl.z_acl_extern_obj, 0,
2628 				    DMU_OBJECT_END);
2629 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2630 				    0, aclp->z_acl_bytes);
2631 			} else {
2632 				dmu_tx_hold_write(tx,
2633 				    pzp->zp_acl.z_acl_extern_obj, 0,
2634 				    aclp->z_acl_bytes);
2635 			}
2636 		} else if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2637 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2638 			    0, aclp->z_acl_bytes);
2639 		}
2640 	}
2641 
2642 	if ((mask & (AT_UID | AT_GID)) && pzp->zp_xattr != 0) {
2643 		err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp);
2644 		if (err) {
2645 			dmu_tx_abort(tx);
2646 			ZFS_EXIT(zfsvfs);
2647 			if (aclp)
2648 				zfs_acl_free(aclp);
2649 			return (err);
2650 		}
2651 		dmu_tx_hold_bonus(tx, attrzp->z_id);
2652 	}
2653 
2654 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2655 	if (err) {
2656 		if (attrzp)
2657 			VN_RELE(ZTOV(attrzp));
2658 
2659 		if (aclp) {
2660 			zfs_acl_free(aclp);
2661 			aclp = NULL;
2662 		}
2663 
2664 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2665 			dmu_tx_wait(tx);
2666 			dmu_tx_abort(tx);
2667 			goto top;
2668 		}
2669 		dmu_tx_abort(tx);
2670 		ZFS_EXIT(zfsvfs);
2671 		return (err);
2672 	}
2673 
2674 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2675 
2676 	/*
2677 	 * Set each attribute requested.
2678 	 * We group settings according to the locks they need to acquire.
2679 	 *
2680 	 * Note: you cannot set ctime directly, although it will be
2681 	 * updated as a side-effect of calling this function.
2682 	 */
2683 
2684 	mutex_enter(&zp->z_lock);
2685 
2686 	if (mask & AT_MODE) {
2687 		mutex_enter(&zp->z_acl_lock);
2688 		zp->z_phys->zp_mode = new_mode;
2689 		err = zfs_aclset_common(zp, aclp, cr, &fuidp, tx);
2690 		ASSERT3U(err, ==, 0);
2691 		mutex_exit(&zp->z_acl_lock);
2692 	}
2693 
2694 	if (attrzp)
2695 		mutex_enter(&attrzp->z_lock);
2696 
2697 	if (mask & AT_UID) {
2698 		pzp->zp_uid = zfs_fuid_create(zfsvfs,
2699 		    vap->va_uid, cr, ZFS_OWNER, tx, &fuidp);
2700 		if (attrzp) {
2701 			attrzp->z_phys->zp_uid = zfs_fuid_create(zfsvfs,
2702 			    vap->va_uid,  cr, ZFS_OWNER, tx, &fuidp);
2703 		}
2704 	}
2705 
2706 	if (mask & AT_GID) {
2707 		pzp->zp_gid = zfs_fuid_create(zfsvfs, vap->va_gid,
2708 		    cr, ZFS_GROUP, tx, &fuidp);
2709 		if (attrzp)
2710 			attrzp->z_phys->zp_gid = zfs_fuid_create(zfsvfs,
2711 			    vap->va_gid, cr, ZFS_GROUP, tx, &fuidp);
2712 	}
2713 
2714 	if (aclp)
2715 		zfs_acl_free(aclp);
2716 
2717 	if (attrzp)
2718 		mutex_exit(&attrzp->z_lock);
2719 
2720 	if (mask & AT_ATIME)
2721 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2722 
2723 	if (mask & AT_MTIME)
2724 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2725 
2726 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2727 	if (mask & AT_SIZE)
2728 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2729 	else if (mask != 0)
2730 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2731 	/*
2732 	 * Do this after setting timestamps to prevent timestamp
2733 	 * update from toggling bit
2734 	 */
2735 
2736 	if (xoap && (mask & AT_XVATTR)) {
2737 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
2738 			size_t len;
2739 			dmu_object_info_t doi;
2740 
2741 			ASSERT(vp->v_type == VREG);
2742 
2743 			/* Grow the bonus buffer if necessary. */
2744 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2745 			len = sizeof (xoap->xoa_av_scanstamp) +
2746 			    sizeof (znode_phys_t);
2747 			if (len > doi.doi_bonus_size)
2748 				VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0);
2749 		}
2750 		zfs_xvattr_set(zp, xvap);
2751 	}
2752 
2753 	if (mask != 0)
2754 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2755 
2756 	if (fuidp)
2757 		zfs_fuid_info_free(fuidp);
2758 	mutex_exit(&zp->z_lock);
2759 
2760 	if (attrzp)
2761 		VN_RELE(ZTOV(attrzp));
2762 
2763 	dmu_tx_commit(tx);
2764 
2765 	ZFS_EXIT(zfsvfs);
2766 	return (err);
2767 }
2768 
2769 typedef struct zfs_zlock {
2770 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2771 	znode_t		*zl_znode;	/* znode we held */
2772 	struct zfs_zlock *zl_next;	/* next in list */
2773 } zfs_zlock_t;
2774 
2775 /*
2776  * Drop locks and release vnodes that were held by zfs_rename_lock().
2777  */
2778 static void
2779 zfs_rename_unlock(zfs_zlock_t **zlpp)
2780 {
2781 	zfs_zlock_t *zl;
2782 
2783 	while ((zl = *zlpp) != NULL) {
2784 		if (zl->zl_znode != NULL)
2785 			VN_RELE(ZTOV(zl->zl_znode));
2786 		rw_exit(zl->zl_rwlock);
2787 		*zlpp = zl->zl_next;
2788 		kmem_free(zl, sizeof (*zl));
2789 	}
2790 }
2791 
2792 /*
2793  * Search back through the directory tree, using the ".." entries.
2794  * Lock each directory in the chain to prevent concurrent renames.
2795  * Fail any attempt to move a directory into one of its own descendants.
2796  * XXX - z_parent_lock can overlap with map or grow locks
2797  */
2798 static int
2799 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2800 {
2801 	zfs_zlock_t	*zl;
2802 	znode_t		*zp = tdzp;
2803 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2804 	uint64_t	*oidp = &zp->z_id;
2805 	krwlock_t	*rwlp = &szp->z_parent_lock;
2806 	krw_t		rw = RW_WRITER;
2807 
2808 	/*
2809 	 * First pass write-locks szp and compares to zp->z_id.
2810 	 * Later passes read-lock zp and compare to zp->z_parent.
2811 	 */
2812 	do {
2813 		if (!rw_tryenter(rwlp, rw)) {
2814 			/*
2815 			 * Another thread is renaming in this path.
2816 			 * Note that if we are a WRITER, we don't have any
2817 			 * parent_locks held yet.
2818 			 */
2819 			if (rw == RW_READER && zp->z_id > szp->z_id) {
2820 				/*
2821 				 * Drop our locks and restart
2822 				 */
2823 				zfs_rename_unlock(&zl);
2824 				*zlpp = NULL;
2825 				zp = tdzp;
2826 				oidp = &zp->z_id;
2827 				rwlp = &szp->z_parent_lock;
2828 				rw = RW_WRITER;
2829 				continue;
2830 			} else {
2831 				/*
2832 				 * Wait for other thread to drop its locks
2833 				 */
2834 				rw_enter(rwlp, rw);
2835 			}
2836 		}
2837 
2838 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2839 		zl->zl_rwlock = rwlp;
2840 		zl->zl_znode = NULL;
2841 		zl->zl_next = *zlpp;
2842 		*zlpp = zl;
2843 
2844 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2845 			return (EINVAL);
2846 
2847 		if (*oidp == rootid)		/* We've hit the top */
2848 			return (0);
2849 
2850 		if (rw == RW_READER) {		/* i.e. not the first pass */
2851 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2852 			if (error)
2853 				return (error);
2854 			zl->zl_znode = zp;
2855 		}
2856 		oidp = &zp->z_phys->zp_parent;
2857 		rwlp = &zp->z_parent_lock;
2858 		rw = RW_READER;
2859 
2860 	} while (zp->z_id != sdzp->z_id);
2861 
2862 	return (0);
2863 }
2864 
2865 /*
2866  * Move an entry from the provided source directory to the target
2867  * directory.  Change the entry name as indicated.
2868  *
2869  *	IN:	sdvp	- Source directory containing the "old entry".
2870  *		snm	- Old entry name.
2871  *		tdvp	- Target directory to contain the "new entry".
2872  *		tnm	- New entry name.
2873  *		cr	- credentials of caller.
2874  *		ct	- caller context
2875  *		flags	- case flags
2876  *
2877  *	RETURN:	0 if success
2878  *		error code if failure
2879  *
2880  * Timestamps:
2881  *	sdvp,tdvp - ctime|mtime updated
2882  */
2883 /*ARGSUSED*/
2884 static int
2885 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
2886     caller_context_t *ct, int flags)
2887 {
2888 	znode_t		*tdzp, *szp, *tzp;
2889 	znode_t		*sdzp = VTOZ(sdvp);
2890 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2891 	zilog_t		*zilog;
2892 	vnode_t		*realvp;
2893 	zfs_dirlock_t	*sdl, *tdl;
2894 	dmu_tx_t	*tx;
2895 	zfs_zlock_t	*zl;
2896 	int		cmp, serr, terr;
2897 	int		error = 0;
2898 	int		zflg = 0;
2899 
2900 	ZFS_ENTER(zfsvfs);
2901 	ZFS_VERIFY_ZP(sdzp);
2902 	zilog = zfsvfs->z_log;
2903 
2904 	/*
2905 	 * Make sure we have the real vp for the target directory.
2906 	 */
2907 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
2908 		tdvp = realvp;
2909 
2910 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
2911 		ZFS_EXIT(zfsvfs);
2912 		return (EXDEV);
2913 	}
2914 
2915 	tdzp = VTOZ(tdvp);
2916 	ZFS_VERIFY_ZP(tdzp);
2917 	if (zfsvfs->z_utf8 && u8_validate(tnm,
2918 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2919 		ZFS_EXIT(zfsvfs);
2920 		return (EILSEQ);
2921 	}
2922 
2923 	if (flags & FIGNORECASE)
2924 		zflg |= ZCILOOK;
2925 
2926 top:
2927 	szp = NULL;
2928 	tzp = NULL;
2929 	zl = NULL;
2930 
2931 	/*
2932 	 * This is to prevent the creation of links into attribute space
2933 	 * by renaming a linked file into/outof an attribute directory.
2934 	 * See the comment in zfs_link() for why this is considered bad.
2935 	 */
2936 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
2937 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
2938 		ZFS_EXIT(zfsvfs);
2939 		return (EINVAL);
2940 	}
2941 
2942 	/*
2943 	 * Lock source and target directory entries.  To prevent deadlock,
2944 	 * a lock ordering must be defined.  We lock the directory with
2945 	 * the smallest object id first, or if it's a tie, the one with
2946 	 * the lexically first name.
2947 	 */
2948 	if (sdzp->z_id < tdzp->z_id) {
2949 		cmp = -1;
2950 	} else if (sdzp->z_id > tdzp->z_id) {
2951 		cmp = 1;
2952 	} else {
2953 		/*
2954 		 * First compare the two name arguments without
2955 		 * considering any case folding.
2956 		 */
2957 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
2958 
2959 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
2960 		ASSERT(error == 0 || !zfsvfs->z_utf8);
2961 		if (cmp == 0) {
2962 			/*
2963 			 * POSIX: "If the old argument and the new argument
2964 			 * both refer to links to the same existing file,
2965 			 * the rename() function shall return successfully
2966 			 * and perform no other action."
2967 			 */
2968 			ZFS_EXIT(zfsvfs);
2969 			return (0);
2970 		}
2971 		/*
2972 		 * If the file system is case-folding, then we may
2973 		 * have some more checking to do.  A case-folding file
2974 		 * system is either supporting mixed case sensitivity
2975 		 * access or is completely case-insensitive.  Note
2976 		 * that the file system is always case preserving.
2977 		 *
2978 		 * In mixed sensitivity mode case sensitive behavior
2979 		 * is the default.  FIGNORECASE must be used to
2980 		 * explicitly request case insensitive behavior.
2981 		 *
2982 		 * If the source and target names provided differ only
2983 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
2984 		 * we will treat this as a special case in the
2985 		 * case-insensitive mode: as long as the source name
2986 		 * is an exact match, we will allow this to proceed as
2987 		 * a name-change request.
2988 		 */
2989 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
2990 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
2991 		    flags & FIGNORECASE)) &&
2992 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
2993 		    &error) == 0) {
2994 			/*
2995 			 * case preserving rename request, require exact
2996 			 * name matches
2997 			 */
2998 			zflg |= ZCIEXACT;
2999 			zflg &= ~ZCILOOK;
3000 		}
3001 	}
3002 
3003 	if (cmp < 0) {
3004 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3005 		    ZEXISTS | zflg, NULL, NULL);
3006 		terr = zfs_dirent_lock(&tdl,
3007 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3008 	} else {
3009 		terr = zfs_dirent_lock(&tdl,
3010 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3011 		serr = zfs_dirent_lock(&sdl,
3012 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3013 		    NULL, NULL);
3014 	}
3015 
3016 	if (serr) {
3017 		/*
3018 		 * Source entry invalid or not there.
3019 		 */
3020 		if (!terr) {
3021 			zfs_dirent_unlock(tdl);
3022 			if (tzp)
3023 				VN_RELE(ZTOV(tzp));
3024 		}
3025 		if (strcmp(snm, "..") == 0)
3026 			serr = EINVAL;
3027 		ZFS_EXIT(zfsvfs);
3028 		return (serr);
3029 	}
3030 	if (terr) {
3031 		zfs_dirent_unlock(sdl);
3032 		VN_RELE(ZTOV(szp));
3033 		if (strcmp(tnm, "..") == 0)
3034 			terr = EINVAL;
3035 		ZFS_EXIT(zfsvfs);
3036 		return (terr);
3037 	}
3038 
3039 	/*
3040 	 * Must have write access at the source to remove the old entry
3041 	 * and write access at the target to create the new entry.
3042 	 * Note that if target and source are the same, this can be
3043 	 * done in a single check.
3044 	 */
3045 
3046 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3047 		goto out;
3048 
3049 	if (ZTOV(szp)->v_type == VDIR) {
3050 		/*
3051 		 * Check to make sure rename is valid.
3052 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3053 		 */
3054 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3055 			goto out;
3056 	}
3057 
3058 	/*
3059 	 * Does target exist?
3060 	 */
3061 	if (tzp) {
3062 		/*
3063 		 * Source and target must be the same type.
3064 		 */
3065 		if (ZTOV(szp)->v_type == VDIR) {
3066 			if (ZTOV(tzp)->v_type != VDIR) {
3067 				error = ENOTDIR;
3068 				goto out;
3069 			}
3070 		} else {
3071 			if (ZTOV(tzp)->v_type == VDIR) {
3072 				error = EISDIR;
3073 				goto out;
3074 			}
3075 		}
3076 		/*
3077 		 * POSIX dictates that when the source and target
3078 		 * entries refer to the same file object, rename
3079 		 * must do nothing and exit without error.
3080 		 */
3081 		if (szp->z_id == tzp->z_id) {
3082 			error = 0;
3083 			goto out;
3084 		}
3085 	}
3086 
3087 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3088 	if (tzp)
3089 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3090 
3091 	/*
3092 	 * notify the target directory if it is not the same
3093 	 * as source directory.
3094 	 */
3095 	if (tdvp != sdvp) {
3096 		vnevent_rename_dest_dir(tdvp, ct);
3097 	}
3098 
3099 	tx = dmu_tx_create(zfsvfs->z_os);
3100 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
3101 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
3102 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3103 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3104 	if (sdzp != tdzp)
3105 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
3106 	if (tzp)
3107 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* parent changes */
3108 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3109 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3110 	if (error) {
3111 		if (zl != NULL)
3112 			zfs_rename_unlock(&zl);
3113 		zfs_dirent_unlock(sdl);
3114 		zfs_dirent_unlock(tdl);
3115 		VN_RELE(ZTOV(szp));
3116 		if (tzp)
3117 			VN_RELE(ZTOV(tzp));
3118 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3119 			dmu_tx_wait(tx);
3120 			dmu_tx_abort(tx);
3121 			goto top;
3122 		}
3123 		dmu_tx_abort(tx);
3124 		ZFS_EXIT(zfsvfs);
3125 		return (error);
3126 	}
3127 
3128 	if (tzp)	/* Attempt to remove the existing target */
3129 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3130 
3131 	if (error == 0) {
3132 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3133 		if (error == 0) {
3134 			szp->z_phys->zp_flags |= ZFS_AV_MODIFIED;
3135 
3136 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3137 			ASSERT(error == 0);
3138 
3139 			zfs_log_rename(zilog, tx,
3140 			    TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0),
3141 			    sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp);
3142 
3143 			/* Update path information for the target vnode */
3144 			vn_renamepath(tdvp, ZTOV(szp), tnm, strlen(tnm));
3145 		}
3146 	}
3147 
3148 	dmu_tx_commit(tx);
3149 out:
3150 	if (zl != NULL)
3151 		zfs_rename_unlock(&zl);
3152 
3153 	zfs_dirent_unlock(sdl);
3154 	zfs_dirent_unlock(tdl);
3155 
3156 	VN_RELE(ZTOV(szp));
3157 	if (tzp)
3158 		VN_RELE(ZTOV(tzp));
3159 
3160 	ZFS_EXIT(zfsvfs);
3161 	return (error);
3162 }
3163 
3164 /*
3165  * Insert the indicated symbolic reference entry into the directory.
3166  *
3167  *	IN:	dvp	- Directory to contain new symbolic link.
3168  *		link	- Name for new symlink entry.
3169  *		vap	- Attributes of new entry.
3170  *		target	- Target path of new symlink.
3171  *		cr	- credentials of caller.
3172  *		ct	- caller context
3173  *		flags	- case flags
3174  *
3175  *	RETURN:	0 if success
3176  *		error code if failure
3177  *
3178  * Timestamps:
3179  *	dvp - ctime|mtime updated
3180  */
3181 /*ARGSUSED*/
3182 static int
3183 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3184     caller_context_t *ct, int flags)
3185 {
3186 	znode_t		*zp, *dzp = VTOZ(dvp);
3187 	zfs_dirlock_t	*dl;
3188 	dmu_tx_t	*tx;
3189 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3190 	zilog_t		*zilog;
3191 	int		len = strlen(link);
3192 	int		error;
3193 	int		zflg = ZNEW;
3194 	zfs_fuid_info_t *fuidp = NULL;
3195 
3196 	ASSERT(vap->va_type == VLNK);
3197 
3198 	ZFS_ENTER(zfsvfs);
3199 	ZFS_VERIFY_ZP(dzp);
3200 	zilog = zfsvfs->z_log;
3201 
3202 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3203 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3204 		ZFS_EXIT(zfsvfs);
3205 		return (EILSEQ);
3206 	}
3207 	if (flags & FIGNORECASE)
3208 		zflg |= ZCILOOK;
3209 top:
3210 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3211 		ZFS_EXIT(zfsvfs);
3212 		return (error);
3213 	}
3214 
3215 	if (len > MAXPATHLEN) {
3216 		ZFS_EXIT(zfsvfs);
3217 		return (ENAMETOOLONG);
3218 	}
3219 
3220 	/*
3221 	 * Attempt to lock directory; fail if entry already exists.
3222 	 */
3223 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3224 	if (error) {
3225 		ZFS_EXIT(zfsvfs);
3226 		return (error);
3227 	}
3228 
3229 	tx = dmu_tx_create(zfsvfs->z_os);
3230 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3231 	dmu_tx_hold_bonus(tx, dzp->z_id);
3232 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3233 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
3234 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
3235 	if (IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))) {
3236 		if (zfsvfs->z_fuid_obj == 0) {
3237 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
3238 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3239 			    FUID_SIZE_ESTIMATE(zfsvfs));
3240 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
3241 		} else {
3242 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
3243 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
3244 			    FUID_SIZE_ESTIMATE(zfsvfs));
3245 		}
3246 	}
3247 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3248 	if (error) {
3249 		zfs_dirent_unlock(dl);
3250 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3251 			dmu_tx_wait(tx);
3252 			dmu_tx_abort(tx);
3253 			goto top;
3254 		}
3255 		dmu_tx_abort(tx);
3256 		ZFS_EXIT(zfsvfs);
3257 		return (error);
3258 	}
3259 
3260 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
3261 
3262 	/*
3263 	 * Create a new object for the symlink.
3264 	 * Put the link content into bonus buffer if it will fit;
3265 	 * otherwise, store it just like any other file data.
3266 	 */
3267 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
3268 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, len, NULL, &fuidp);
3269 		if (len != 0)
3270 			bcopy(link, zp->z_phys + 1, len);
3271 	} else {
3272 		dmu_buf_t *dbp;
3273 
3274 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, NULL, &fuidp);
3275 		/*
3276 		 * Nothing can access the znode yet so no locking needed
3277 		 * for growing the znode's blocksize.
3278 		 */
3279 		zfs_grow_blocksize(zp, len, tx);
3280 
3281 		VERIFY(0 == dmu_buf_hold(zfsvfs->z_os,
3282 		    zp->z_id, 0, FTAG, &dbp));
3283 		dmu_buf_will_dirty(dbp, tx);
3284 
3285 		ASSERT3U(len, <=, dbp->db_size);
3286 		bcopy(link, dbp->db_data, len);
3287 		dmu_buf_rele(dbp, FTAG);
3288 	}
3289 	zp->z_phys->zp_size = len;
3290 
3291 	/*
3292 	 * Insert the new object into the directory.
3293 	 */
3294 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3295 out:
3296 	if (error == 0) {
3297 		uint64_t txtype = TX_SYMLINK;
3298 		if (flags & FIGNORECASE)
3299 			txtype |= TX_CI;
3300 		zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3301 	}
3302 	if (fuidp)
3303 		zfs_fuid_info_free(fuidp);
3304 
3305 	dmu_tx_commit(tx);
3306 
3307 	zfs_dirent_unlock(dl);
3308 
3309 	VN_RELE(ZTOV(zp));
3310 
3311 	ZFS_EXIT(zfsvfs);
3312 	return (error);
3313 }
3314 
3315 /*
3316  * Return, in the buffer contained in the provided uio structure,
3317  * the symbolic path referred to by vp.
3318  *
3319  *	IN:	vp	- vnode of symbolic link.
3320  *		uoip	- structure to contain the link path.
3321  *		cr	- credentials of caller.
3322  *		ct	- caller context
3323  *
3324  *	OUT:	uio	- structure to contain the link path.
3325  *
3326  *	RETURN:	0 if success
3327  *		error code if failure
3328  *
3329  * Timestamps:
3330  *	vp - atime updated
3331  */
3332 /* ARGSUSED */
3333 static int
3334 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3335 {
3336 	znode_t		*zp = VTOZ(vp);
3337 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3338 	size_t		bufsz;
3339 	int		error;
3340 
3341 	ZFS_ENTER(zfsvfs);
3342 	ZFS_VERIFY_ZP(zp);
3343 
3344 	bufsz = (size_t)zp->z_phys->zp_size;
3345 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
3346 		error = uiomove(zp->z_phys + 1,
3347 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3348 	} else {
3349 		dmu_buf_t *dbp;
3350 		error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp);
3351 		if (error) {
3352 			ZFS_EXIT(zfsvfs);
3353 			return (error);
3354 		}
3355 		error = uiomove(dbp->db_data,
3356 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3357 		dmu_buf_rele(dbp, FTAG);
3358 	}
3359 
3360 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3361 	ZFS_EXIT(zfsvfs);
3362 	return (error);
3363 }
3364 
3365 /*
3366  * Insert a new entry into directory tdvp referencing svp.
3367  *
3368  *	IN:	tdvp	- Directory to contain new entry.
3369  *		svp	- vnode of new entry.
3370  *		name	- name of new entry.
3371  *		cr	- credentials of caller.
3372  *		ct	- caller context
3373  *
3374  *	RETURN:	0 if success
3375  *		error code if failure
3376  *
3377  * Timestamps:
3378  *	tdvp - ctime|mtime updated
3379  *	 svp - ctime updated
3380  */
3381 /* ARGSUSED */
3382 static int
3383 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3384     caller_context_t *ct, int flags)
3385 {
3386 	znode_t		*dzp = VTOZ(tdvp);
3387 	znode_t		*tzp, *szp;
3388 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3389 	zilog_t		*zilog;
3390 	zfs_dirlock_t	*dl;
3391 	dmu_tx_t	*tx;
3392 	vnode_t		*realvp;
3393 	int		error;
3394 	int		zf = ZNEW;
3395 	uid_t		owner;
3396 
3397 	ASSERT(tdvp->v_type == VDIR);
3398 
3399 	ZFS_ENTER(zfsvfs);
3400 	ZFS_VERIFY_ZP(dzp);
3401 	zilog = zfsvfs->z_log;
3402 
3403 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3404 		svp = realvp;
3405 
3406 	if (svp->v_vfsp != tdvp->v_vfsp) {
3407 		ZFS_EXIT(zfsvfs);
3408 		return (EXDEV);
3409 	}
3410 	szp = VTOZ(svp);
3411 	ZFS_VERIFY_ZP(szp);
3412 
3413 	if (zfsvfs->z_utf8 && u8_validate(name,
3414 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3415 		ZFS_EXIT(zfsvfs);
3416 		return (EILSEQ);
3417 	}
3418 	if (flags & FIGNORECASE)
3419 		zf |= ZCILOOK;
3420 
3421 top:
3422 	/*
3423 	 * We do not support links between attributes and non-attributes
3424 	 * because of the potential security risk of creating links
3425 	 * into "normal" file space in order to circumvent restrictions
3426 	 * imposed in attribute space.
3427 	 */
3428 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
3429 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
3430 		ZFS_EXIT(zfsvfs);
3431 		return (EINVAL);
3432 	}
3433 
3434 	/*
3435 	 * POSIX dictates that we return EPERM here.
3436 	 * Better choices include ENOTSUP or EISDIR.
3437 	 */
3438 	if (svp->v_type == VDIR) {
3439 		ZFS_EXIT(zfsvfs);
3440 		return (EPERM);
3441 	}
3442 
3443 	owner = zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, cr, ZFS_OWNER);
3444 	if (owner != crgetuid(cr) &&
3445 	    secpolicy_basic_link(cr) != 0) {
3446 		ZFS_EXIT(zfsvfs);
3447 		return (EPERM);
3448 	}
3449 
3450 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3451 		ZFS_EXIT(zfsvfs);
3452 		return (error);
3453 	}
3454 
3455 	/*
3456 	 * Attempt to lock directory; fail if entry already exists.
3457 	 */
3458 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3459 	if (error) {
3460 		ZFS_EXIT(zfsvfs);
3461 		return (error);
3462 	}
3463 
3464 	tx = dmu_tx_create(zfsvfs->z_os);
3465 	dmu_tx_hold_bonus(tx, szp->z_id);
3466 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3467 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3468 	if (error) {
3469 		zfs_dirent_unlock(dl);
3470 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3471 			dmu_tx_wait(tx);
3472 			dmu_tx_abort(tx);
3473 			goto top;
3474 		}
3475 		dmu_tx_abort(tx);
3476 		ZFS_EXIT(zfsvfs);
3477 		return (error);
3478 	}
3479 
3480 	error = zfs_link_create(dl, szp, tx, 0);
3481 
3482 	if (error == 0) {
3483 		uint64_t txtype = TX_LINK;
3484 		if (flags & FIGNORECASE)
3485 			txtype |= TX_CI;
3486 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3487 	}
3488 
3489 	dmu_tx_commit(tx);
3490 
3491 	zfs_dirent_unlock(dl);
3492 
3493 	if (error == 0) {
3494 		vnevent_link(svp, ct);
3495 	}
3496 
3497 	ZFS_EXIT(zfsvfs);
3498 	return (error);
3499 }
3500 
3501 /*
3502  * zfs_null_putapage() is used when the file system has been force
3503  * unmounted. It just drops the pages.
3504  */
3505 /* ARGSUSED */
3506 static int
3507 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3508 		size_t *lenp, int flags, cred_t *cr)
3509 {
3510 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
3511 	return (0);
3512 }
3513 
3514 /*
3515  * Push a page out to disk, klustering if possible.
3516  *
3517  *	IN:	vp	- file to push page to.
3518  *		pp	- page to push.
3519  *		flags	- additional flags.
3520  *		cr	- credentials of caller.
3521  *
3522  *	OUT:	offp	- start of range pushed.
3523  *		lenp	- len of range pushed.
3524  *
3525  *	RETURN:	0 if success
3526  *		error code if failure
3527  *
3528  * NOTE: callers must have locked the page to be pushed.  On
3529  * exit, the page (and all other pages in the kluster) must be
3530  * unlocked.
3531  */
3532 /* ARGSUSED */
3533 static int
3534 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3535 		size_t *lenp, int flags, cred_t *cr)
3536 {
3537 	znode_t		*zp = VTOZ(vp);
3538 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3539 	zilog_t		*zilog = zfsvfs->z_log;
3540 	dmu_tx_t	*tx;
3541 	rl_t		*rl;
3542 	u_offset_t	off, koff;
3543 	size_t		len, klen;
3544 	uint64_t	filesz;
3545 	int		err;
3546 
3547 	filesz = zp->z_phys->zp_size;
3548 	off = pp->p_offset;
3549 	len = PAGESIZE;
3550 	/*
3551 	 * If our blocksize is bigger than the page size, try to kluster
3552 	 * muiltiple pages so that we write a full block (thus avoiding
3553 	 * a read-modify-write).
3554 	 */
3555 	if (off < filesz && zp->z_blksz > PAGESIZE) {
3556 		if (!ISP2(zp->z_blksz)) {
3557 			/* Only one block in the file. */
3558 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3559 			koff = 0;
3560 		} else {
3561 			klen = zp->z_blksz;
3562 			koff = P2ALIGN(off, (u_offset_t)klen);
3563 		}
3564 		ASSERT(koff <= filesz);
3565 		if (koff + klen > filesz)
3566 			klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE);
3567 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
3568 	}
3569 	ASSERT3U(btop(len), ==, btopr(len));
3570 top:
3571 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
3572 	/*
3573 	 * Can't push pages past end-of-file.
3574 	 */
3575 	filesz = zp->z_phys->zp_size;
3576 	if (off >= filesz) {
3577 		/* ignore all pages */
3578 		err = 0;
3579 		goto out;
3580 	} else if (off + len > filesz) {
3581 		int npages = btopr(filesz - off);
3582 		page_t *trunc;
3583 
3584 		page_list_break(&pp, &trunc, npages);
3585 		/* ignore pages past end of file */
3586 		if (trunc)
3587 			pvn_write_done(trunc, flags);
3588 		len = filesz - off;
3589 	}
3590 
3591 	tx = dmu_tx_create(zfsvfs->z_os);
3592 	dmu_tx_hold_write(tx, zp->z_id, off, len);
3593 	dmu_tx_hold_bonus(tx, zp->z_id);
3594 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
3595 	if (err != 0) {
3596 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3597 			zfs_range_unlock(rl);
3598 			dmu_tx_wait(tx);
3599 			dmu_tx_abort(tx);
3600 			err = 0;
3601 			goto top;
3602 		}
3603 		dmu_tx_abort(tx);
3604 		goto out;
3605 	}
3606 
3607 	if (zp->z_blksz <= PAGESIZE) {
3608 		caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1);
3609 		ASSERT3U(len, <=, PAGESIZE);
3610 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
3611 		ppmapout(va);
3612 	} else {
3613 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
3614 	}
3615 
3616 	if (err == 0) {
3617 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
3618 		zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0);
3619 		dmu_tx_commit(tx);
3620 	}
3621 
3622 out:
3623 	zfs_range_unlock(rl);
3624 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
3625 	if (offp)
3626 		*offp = off;
3627 	if (lenp)
3628 		*lenp = len;
3629 
3630 	return (err);
3631 }
3632 
3633 /*
3634  * Copy the portion of the file indicated from pages into the file.
3635  * The pages are stored in a page list attached to the files vnode.
3636  *
3637  *	IN:	vp	- vnode of file to push page data to.
3638  *		off	- position in file to put data.
3639  *		len	- amount of data to write.
3640  *		flags	- flags to control the operation.
3641  *		cr	- credentials of caller.
3642  *		ct	- caller context.
3643  *
3644  *	RETURN:	0 if success
3645  *		error code if failure
3646  *
3647  * Timestamps:
3648  *	vp - ctime|mtime updated
3649  */
3650 /*ARGSUSED*/
3651 static int
3652 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
3653     caller_context_t *ct)
3654 {
3655 	znode_t		*zp = VTOZ(vp);
3656 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3657 	page_t		*pp;
3658 	size_t		io_len;
3659 	u_offset_t	io_off;
3660 	uint64_t	filesz;
3661 	int		error = 0;
3662 
3663 	ZFS_ENTER(zfsvfs);
3664 	ZFS_VERIFY_ZP(zp);
3665 
3666 	if (len == 0) {
3667 		/*
3668 		 * Search the entire vp list for pages >= off.
3669 		 */
3670 		error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage,
3671 		    flags, cr);
3672 		goto out;
3673 	}
3674 
3675 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3676 	if (off > filesz) {
3677 		/* past end of file */
3678 		ZFS_EXIT(zfsvfs);
3679 		return (0);
3680 	}
3681 
3682 	len = MIN(len, filesz - off);
3683 
3684 	for (io_off = off; io_off < off + len; io_off += io_len) {
3685 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
3686 			pp = page_lookup(vp, io_off,
3687 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
3688 		} else {
3689 			pp = page_lookup_nowait(vp, io_off,
3690 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
3691 		}
3692 
3693 		if (pp != NULL && pvn_getdirty(pp, flags)) {
3694 			int err;
3695 
3696 			/*
3697 			 * Found a dirty page to push
3698 			 */
3699 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
3700 			if (err)
3701 				error = err;
3702 		} else {
3703 			io_len = PAGESIZE;
3704 		}
3705 	}
3706 out:
3707 	if ((flags & B_ASYNC) == 0)
3708 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
3709 	ZFS_EXIT(zfsvfs);
3710 	return (error);
3711 }
3712 
3713 /*ARGSUSED*/
3714 void
3715 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
3716 {
3717 	znode_t	*zp = VTOZ(vp);
3718 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3719 	int error;
3720 
3721 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
3722 	if (zp->z_dbuf == NULL) {
3723 		/*
3724 		 * The fs has been unmounted, or we did a
3725 		 * suspend/resume and this file no longer exists.
3726 		 */
3727 		if (vn_has_cached_data(vp)) {
3728 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
3729 			    B_INVAL, cr);
3730 		}
3731 
3732 		mutex_enter(&zp->z_lock);
3733 		vp->v_count = 0; /* count arrives as 1 */
3734 		mutex_exit(&zp->z_lock);
3735 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
3736 		zfs_znode_free(zp);
3737 		return;
3738 	}
3739 
3740 	/*
3741 	 * Attempt to push any data in the page cache.  If this fails
3742 	 * we will get kicked out later in zfs_zinactive().
3743 	 */
3744 	if (vn_has_cached_data(vp)) {
3745 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
3746 		    cr);
3747 	}
3748 
3749 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3750 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3751 
3752 		dmu_tx_hold_bonus(tx, zp->z_id);
3753 		error = dmu_tx_assign(tx, TXG_WAIT);
3754 		if (error) {
3755 			dmu_tx_abort(tx);
3756 		} else {
3757 			dmu_buf_will_dirty(zp->z_dbuf, tx);
3758 			mutex_enter(&zp->z_lock);
3759 			zp->z_atime_dirty = 0;
3760 			mutex_exit(&zp->z_lock);
3761 			dmu_tx_commit(tx);
3762 		}
3763 	}
3764 
3765 	zfs_zinactive(zp);
3766 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
3767 }
3768 
3769 /*
3770  * Bounds-check the seek operation.
3771  *
3772  *	IN:	vp	- vnode seeking within
3773  *		ooff	- old file offset
3774  *		noffp	- pointer to new file offset
3775  *		ct	- caller context
3776  *
3777  *	RETURN:	0 if success
3778  *		EINVAL if new offset invalid
3779  */
3780 /* ARGSUSED */
3781 static int
3782 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
3783     caller_context_t *ct)
3784 {
3785 	if (vp->v_type == VDIR)
3786 		return (0);
3787 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3788 }
3789 
3790 /*
3791  * Pre-filter the generic locking function to trap attempts to place
3792  * a mandatory lock on a memory mapped file.
3793  */
3794 static int
3795 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3796     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
3797 {
3798 	znode_t *zp = VTOZ(vp);
3799 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3800 	int error;
3801 
3802 	ZFS_ENTER(zfsvfs);
3803 	ZFS_VERIFY_ZP(zp);
3804 
3805 	/*
3806 	 * We are following the UFS semantics with respect to mapcnt
3807 	 * here: If we see that the file is mapped already, then we will
3808 	 * return an error, but we don't worry about races between this
3809 	 * function and zfs_map().
3810 	 */
3811 	if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3812 		ZFS_EXIT(zfsvfs);
3813 		return (EAGAIN);
3814 	}
3815 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct);
3816 	ZFS_EXIT(zfsvfs);
3817 	return (error);
3818 }
3819 
3820 /*
3821  * If we can't find a page in the cache, we will create a new page
3822  * and fill it with file data.  For efficiency, we may try to fill
3823  * multiple pages at once (klustering).
3824  */
3825 static int
3826 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3827     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3828 {
3829 	znode_t *zp = VTOZ(vp);
3830 	page_t *pp, *cur_pp;
3831 	objset_t *os = zp->z_zfsvfs->z_os;
3832 	caddr_t va;
3833 	u_offset_t io_off, total;
3834 	uint64_t oid = zp->z_id;
3835 	size_t io_len;
3836 	uint64_t filesz;
3837 	int err;
3838 
3839 	/*
3840 	 * If we are only asking for a single page don't bother klustering.
3841 	 */
3842 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3843 	if (off >= filesz)
3844 		return (EFAULT);
3845 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3846 		io_off = off;
3847 		io_len = PAGESIZE;
3848 		pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr);
3849 	} else {
3850 		/*
3851 		 * Try to fill a kluster of pages (a blocks worth).
3852 		 */
3853 		size_t klen;
3854 		u_offset_t koff;
3855 
3856 		if (!ISP2(zp->z_blksz)) {
3857 			/* Only one block in the file. */
3858 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3859 			koff = 0;
3860 		} else {
3861 			/*
3862 			 * It would be ideal to align our offset to the
3863 			 * blocksize but doing so has resulted in some
3864 			 * strange application crashes. For now, we
3865 			 * leave the offset as is and only adjust the
3866 			 * length if we are off the end of the file.
3867 			 */
3868 			koff = off;
3869 			klen = plsz;
3870 		}
3871 		ASSERT(koff <= filesz);
3872 		if (koff + klen > filesz)
3873 			klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff;
3874 		ASSERT3U(off, >=, koff);
3875 		ASSERT3U(off, <, koff + klen);
3876 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3877 		    &io_len, koff, klen, 0);
3878 	}
3879 	if (pp == NULL) {
3880 		/*
3881 		 * Some other thread entered the page before us.
3882 		 * Return to zfs_getpage to retry the lookup.
3883 		 */
3884 		*pl = NULL;
3885 		return (0);
3886 	}
3887 
3888 	/*
3889 	 * Fill the pages in the kluster.
3890 	 */
3891 	cur_pp = pp;
3892 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3893 		ASSERT3U(io_off, ==, cur_pp->p_offset);
3894 		va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1);
3895 		err = dmu_read(os, oid, io_off, PAGESIZE, va);
3896 		ppmapout(va);
3897 		if (err) {
3898 			/* On error, toss the entire kluster */
3899 			pvn_read_done(pp, B_ERROR);
3900 			return (err);
3901 		}
3902 		cur_pp = cur_pp->p_next;
3903 	}
3904 out:
3905 	/*
3906 	 * Fill in the page list array from the kluster.  If
3907 	 * there are too many pages in the kluster, return
3908 	 * as many pages as possible starting from the desired
3909 	 * offset `off'.
3910 	 * NOTE: the page list will always be null terminated.
3911 	 */
3912 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3913 
3914 	return (0);
3915 }
3916 
3917 /*
3918  * Return pointers to the pages for the file region [off, off + len]
3919  * in the pl array.  If plsz is greater than len, this function may
3920  * also return page pointers from before or after the specified
3921  * region (i.e. some region [off', off' + plsz]).  These additional
3922  * pages are only returned if they are already in the cache, or were
3923  * created as part of a klustered read.
3924  *
3925  *	IN:	vp	- vnode of file to get data from.
3926  *		off	- position in file to get data from.
3927  *		len	- amount of data to retrieve.
3928  *		plsz	- length of provided page list.
3929  *		seg	- segment to obtain pages for.
3930  *		addr	- virtual address of fault.
3931  *		rw	- mode of created pages.
3932  *		cr	- credentials of caller.
3933  *		ct	- caller context.
3934  *
3935  *	OUT:	protp	- protection mode of created pages.
3936  *		pl	- list of pages created.
3937  *
3938  *	RETURN:	0 if success
3939  *		error code if failure
3940  *
3941  * Timestamps:
3942  *	vp - atime updated
3943  */
3944 /* ARGSUSED */
3945 static int
3946 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3947 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3948 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
3949 {
3950 	znode_t		*zp = VTOZ(vp);
3951 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3952 	page_t		*pp, **pl0 = pl;
3953 	int		need_unlock = 0, err = 0;
3954 	offset_t	orig_off;
3955 
3956 	ZFS_ENTER(zfsvfs);
3957 	ZFS_VERIFY_ZP(zp);
3958 
3959 	if (protp)
3960 		*protp = PROT_ALL;
3961 
3962 	/* no faultahead (for now) */
3963 	if (pl == NULL) {
3964 		ZFS_EXIT(zfsvfs);
3965 		return (0);
3966 	}
3967 
3968 	/* can't fault past EOF */
3969 	if (off >= zp->z_phys->zp_size) {
3970 		ZFS_EXIT(zfsvfs);
3971 		return (EFAULT);
3972 	}
3973 	orig_off = off;
3974 
3975 	/*
3976 	 * If we already own the lock, then we must be page faulting
3977 	 * in the middle of a write to this file (i.e., we are writing
3978 	 * to this file using data from a mapped region of the file).
3979 	 */
3980 	if (rw_owner(&zp->z_map_lock) != curthread) {
3981 		rw_enter(&zp->z_map_lock, RW_WRITER);
3982 		need_unlock = TRUE;
3983 	}
3984 
3985 	/*
3986 	 * Loop through the requested range [off, off + len] looking
3987 	 * for pages.  If we don't find a page, we will need to create
3988 	 * a new page and fill it with data from the file.
3989 	 */
3990 	while (len > 0) {
3991 		if (plsz < PAGESIZE)
3992 			break;
3993 		if (pp = page_lookup(vp, off, SE_SHARED)) {
3994 			*pl++ = pp;
3995 			off += PAGESIZE;
3996 			addr += PAGESIZE;
3997 			len -= PAGESIZE;
3998 			plsz -= PAGESIZE;
3999 		} else {
4000 			err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw);
4001 			if (err)
4002 				goto out;
4003 			/*
4004 			 * klustering may have changed our region
4005 			 * to be block aligned.
4006 			 */
4007 			if (((pp = *pl) != 0) && (off != pp->p_offset)) {
4008 				int delta = off - pp->p_offset;
4009 				len += delta;
4010 				off -= delta;
4011 				addr -= delta;
4012 			}
4013 			while (*pl) {
4014 				pl++;
4015 				off += PAGESIZE;
4016 				addr += PAGESIZE;
4017 				plsz -= PAGESIZE;
4018 				if (len > PAGESIZE)
4019 					len -= PAGESIZE;
4020 				else
4021 					len = 0;
4022 			}
4023 		}
4024 	}
4025 
4026 	/*
4027 	 * Fill out the page array with any pages already in the cache.
4028 	 */
4029 	while (plsz > 0) {
4030 		pp = page_lookup_nowait(vp, off, SE_SHARED);
4031 		if (pp == NULL)
4032 			break;
4033 		*pl++ = pp;
4034 		off += PAGESIZE;
4035 		plsz -= PAGESIZE;
4036 	}
4037 
4038 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4039 out:
4040 	/*
4041 	 * We can't grab the range lock for the page as reader which would
4042 	 * stop truncation as this leads to deadlock. So we need to recheck
4043 	 * the file size.
4044 	 */
4045 	if (orig_off >= zp->z_phys->zp_size)
4046 		err = EFAULT;
4047 	if (err) {
4048 		/*
4049 		 * Release any pages we have previously locked.
4050 		 */
4051 		while (pl > pl0)
4052 			page_unlock(*--pl);
4053 	}
4054 
4055 	*pl = NULL;
4056 
4057 	if (need_unlock)
4058 		rw_exit(&zp->z_map_lock);
4059 
4060 	ZFS_EXIT(zfsvfs);
4061 	return (err);
4062 }
4063 
4064 /*
4065  * Request a memory map for a section of a file.  This code interacts
4066  * with common code and the VM system as follows:
4067  *
4068  *	common code calls mmap(), which ends up in smmap_common()
4069  *
4070  *	this calls VOP_MAP(), which takes you into (say) zfs
4071  *
4072  *	zfs_map() calls as_map(), passing segvn_create() as the callback
4073  *
4074  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
4075  *
4076  *	zfs_addmap() updates z_mapcnt
4077  */
4078 /*ARGSUSED*/
4079 static int
4080 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4081     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4082     caller_context_t *ct)
4083 {
4084 	znode_t *zp = VTOZ(vp);
4085 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4086 	segvn_crargs_t	vn_a;
4087 	int		error;
4088 
4089 	ZFS_ENTER(zfsvfs);
4090 	ZFS_VERIFY_ZP(zp);
4091 
4092 	if ((prot & PROT_WRITE) &&
4093 	    (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY |
4094 	    ZFS_APPENDONLY))) {
4095 		ZFS_EXIT(zfsvfs);
4096 		return (EPERM);
4097 	}
4098 
4099 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4100 	    (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED)) {
4101 		ZFS_EXIT(zfsvfs);
4102 		return (EACCES);
4103 	}
4104 
4105 	if (vp->v_flag & VNOMAP) {
4106 		ZFS_EXIT(zfsvfs);
4107 		return (ENOSYS);
4108 	}
4109 
4110 	if (off < 0 || len > MAXOFFSET_T - off) {
4111 		ZFS_EXIT(zfsvfs);
4112 		return (ENXIO);
4113 	}
4114 
4115 	if (vp->v_type != VREG) {
4116 		ZFS_EXIT(zfsvfs);
4117 		return (ENODEV);
4118 	}
4119 
4120 	/*
4121 	 * If file is locked, disallow mapping.
4122 	 */
4123 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) {
4124 		ZFS_EXIT(zfsvfs);
4125 		return (EAGAIN);
4126 	}
4127 
4128 	as_rangelock(as);
4129 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4130 	if (error != 0) {
4131 		as_rangeunlock(as);
4132 		ZFS_EXIT(zfsvfs);
4133 		return (error);
4134 	}
4135 
4136 	vn_a.vp = vp;
4137 	vn_a.offset = (u_offset_t)off;
4138 	vn_a.type = flags & MAP_TYPE;
4139 	vn_a.prot = prot;
4140 	vn_a.maxprot = maxprot;
4141 	vn_a.cred = cr;
4142 	vn_a.amp = NULL;
4143 	vn_a.flags = flags & ~MAP_TYPE;
4144 	vn_a.szc = 0;
4145 	vn_a.lgrp_mem_policy_flags = 0;
4146 
4147 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4148 
4149 	as_rangeunlock(as);
4150 	ZFS_EXIT(zfsvfs);
4151 	return (error);
4152 }
4153 
4154 /* ARGSUSED */
4155 static int
4156 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4157     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4158     caller_context_t *ct)
4159 {
4160 	uint64_t pages = btopr(len);
4161 
4162 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4163 	return (0);
4164 }
4165 
4166 /*
4167  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4168  * more accurate mtime for the associated file.  Since we don't have a way of
4169  * detecting when the data was actually modified, we have to resort to
4170  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4171  * last page is pushed.  The problem occurs when the msync() call is omitted,
4172  * which by far the most common case:
4173  *
4174  * 	open()
4175  * 	mmap()
4176  * 	<modify memory>
4177  * 	munmap()
4178  * 	close()
4179  * 	<time lapse>
4180  * 	putpage() via fsflush
4181  *
4182  * If we wait until fsflush to come along, we can have a modification time that
4183  * is some arbitrary point in the future.  In order to prevent this in the
4184  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4185  * torn down.
4186  */
4187 /* ARGSUSED */
4188 static int
4189 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4190     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4191     caller_context_t *ct)
4192 {
4193 	uint64_t pages = btopr(len);
4194 
4195 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4196 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4197 
4198 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4199 	    vn_has_cached_data(vp))
4200 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4201 
4202 	return (0);
4203 }
4204 
4205 /*
4206  * Free or allocate space in a file.  Currently, this function only
4207  * supports the `F_FREESP' command.  However, this command is somewhat
4208  * misnamed, as its functionality includes the ability to allocate as
4209  * well as free space.
4210  *
4211  *	IN:	vp	- vnode of file to free data in.
4212  *		cmd	- action to take (only F_FREESP supported).
4213  *		bfp	- section of file to free/alloc.
4214  *		flag	- current file open mode flags.
4215  *		offset	- current file offset.
4216  *		cr	- credentials of caller [UNUSED].
4217  *		ct	- caller context.
4218  *
4219  *	RETURN:	0 if success
4220  *		error code if failure
4221  *
4222  * Timestamps:
4223  *	vp - ctime|mtime updated
4224  */
4225 /* ARGSUSED */
4226 static int
4227 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4228     offset_t offset, cred_t *cr, caller_context_t *ct)
4229 {
4230 	znode_t		*zp = VTOZ(vp);
4231 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4232 	uint64_t	off, len;
4233 	int		error;
4234 
4235 	ZFS_ENTER(zfsvfs);
4236 	ZFS_VERIFY_ZP(zp);
4237 
4238 	if (cmd != F_FREESP) {
4239 		ZFS_EXIT(zfsvfs);
4240 		return (EINVAL);
4241 	}
4242 
4243 	if (error = convoff(vp, bfp, 0, offset)) {
4244 		ZFS_EXIT(zfsvfs);
4245 		return (error);
4246 	}
4247 
4248 	if (bfp->l_len < 0) {
4249 		ZFS_EXIT(zfsvfs);
4250 		return (EINVAL);
4251 	}
4252 
4253 	off = bfp->l_start;
4254 	len = bfp->l_len; /* 0 means from off to end of file */
4255 
4256 	error = zfs_freesp(zp, off, len, flag, TRUE);
4257 
4258 	ZFS_EXIT(zfsvfs);
4259 	return (error);
4260 }
4261 
4262 /*ARGSUSED*/
4263 static int
4264 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4265 {
4266 	znode_t		*zp = VTOZ(vp);
4267 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4268 	uint32_t	gen;
4269 	uint64_t	object = zp->z_id;
4270 	zfid_short_t	*zfid;
4271 	int		size, i;
4272 
4273 	ZFS_ENTER(zfsvfs);
4274 	ZFS_VERIFY_ZP(zp);
4275 	gen = (uint32_t)zp->z_gen;
4276 
4277 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4278 	if (fidp->fid_len < size) {
4279 		fidp->fid_len = size;
4280 		ZFS_EXIT(zfsvfs);
4281 		return (ENOSPC);
4282 	}
4283 
4284 	zfid = (zfid_short_t *)fidp;
4285 
4286 	zfid->zf_len = size;
4287 
4288 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4289 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4290 
4291 	/* Must have a non-zero generation number to distinguish from .zfs */
4292 	if (gen == 0)
4293 		gen = 1;
4294 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4295 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4296 
4297 	if (size == LONG_FID_LEN) {
4298 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4299 		zfid_long_t	*zlfid;
4300 
4301 		zlfid = (zfid_long_t *)fidp;
4302 
4303 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4304 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4305 
4306 		/* XXX - this should be the generation number for the objset */
4307 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4308 			zlfid->zf_setgen[i] = 0;
4309 	}
4310 
4311 	ZFS_EXIT(zfsvfs);
4312 	return (0);
4313 }
4314 
4315 static int
4316 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4317     caller_context_t *ct)
4318 {
4319 	znode_t		*zp, *xzp;
4320 	zfsvfs_t	*zfsvfs;
4321 	zfs_dirlock_t	*dl;
4322 	int		error;
4323 
4324 	switch (cmd) {
4325 	case _PC_LINK_MAX:
4326 		*valp = ULONG_MAX;
4327 		return (0);
4328 
4329 	case _PC_FILESIZEBITS:
4330 		*valp = 64;
4331 		return (0);
4332 
4333 	case _PC_XATTR_EXISTS:
4334 		zp = VTOZ(vp);
4335 		zfsvfs = zp->z_zfsvfs;
4336 		ZFS_ENTER(zfsvfs);
4337 		ZFS_VERIFY_ZP(zp);
4338 		*valp = 0;
4339 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4340 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4341 		if (error == 0) {
4342 			zfs_dirent_unlock(dl);
4343 			if (!zfs_dirempty(xzp))
4344 				*valp = 1;
4345 			VN_RELE(ZTOV(xzp));
4346 		} else if (error == ENOENT) {
4347 			/*
4348 			 * If there aren't extended attributes, it's the
4349 			 * same as having zero of them.
4350 			 */
4351 			error = 0;
4352 		}
4353 		ZFS_EXIT(zfsvfs);
4354 		return (error);
4355 
4356 	case _PC_SATTR_ENABLED:
4357 	case _PC_SATTR_EXISTS:
4358 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) &&
4359 		    (vp->v_type == VREG || vp->v_type == VDIR);
4360 		return (0);
4361 
4362 	case _PC_ACL_ENABLED:
4363 		*valp = _ACL_ACE_ENABLED;
4364 		return (0);
4365 
4366 	case _PC_MIN_HOLE_SIZE:
4367 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4368 		return (0);
4369 
4370 	default:
4371 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4372 	}
4373 }
4374 
4375 /*ARGSUSED*/
4376 static int
4377 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4378     caller_context_t *ct)
4379 {
4380 	znode_t *zp = VTOZ(vp);
4381 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4382 	int error;
4383 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4384 
4385 	ZFS_ENTER(zfsvfs);
4386 	ZFS_VERIFY_ZP(zp);
4387 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4388 	ZFS_EXIT(zfsvfs);
4389 
4390 	return (error);
4391 }
4392 
4393 /*ARGSUSED*/
4394 static int
4395 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4396     caller_context_t *ct)
4397 {
4398 	znode_t *zp = VTOZ(vp);
4399 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4400 	int error;
4401 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4402 
4403 	ZFS_ENTER(zfsvfs);
4404 	ZFS_VERIFY_ZP(zp);
4405 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4406 	ZFS_EXIT(zfsvfs);
4407 	return (error);
4408 }
4409 
4410 /*
4411  * Predeclare these here so that the compiler assumes that
4412  * this is an "old style" function declaration that does
4413  * not include arguments => we won't get type mismatch errors
4414  * in the initializations that follow.
4415  */
4416 static int zfs_inval();
4417 static int zfs_isdir();
4418 
4419 static int
4420 zfs_inval()
4421 {
4422 	return (EINVAL);
4423 }
4424 
4425 static int
4426 zfs_isdir()
4427 {
4428 	return (EISDIR);
4429 }
4430 /*
4431  * Directory vnode operations template
4432  */
4433 vnodeops_t *zfs_dvnodeops;
4434 const fs_operation_def_t zfs_dvnodeops_template[] = {
4435 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4436 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4437 	VOPNAME_READ,		{ .error = zfs_isdir },
4438 	VOPNAME_WRITE,		{ .error = zfs_isdir },
4439 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4440 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4441 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4442 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4443 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4444 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4445 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4446 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4447 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4448 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
4449 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4450 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4451 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
4452 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4453 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4454 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4455 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4456 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4457 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4458 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4459 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
4460 	NULL,			NULL
4461 };
4462 
4463 /*
4464  * Regular file vnode operations template
4465  */
4466 vnodeops_t *zfs_fvnodeops;
4467 const fs_operation_def_t zfs_fvnodeops_template[] = {
4468 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4469 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4470 	VOPNAME_READ,		{ .vop_read = zfs_read },
4471 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
4472 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4473 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4474 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4475 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4476 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4477 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4478 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4479 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4480 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4481 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4482 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
4483 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
4484 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
4485 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
4486 	VOPNAME_MAP,		{ .vop_map = zfs_map },
4487 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
4488 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
4489 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4490 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4491 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4492 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4493 	NULL,			NULL
4494 };
4495 
4496 /*
4497  * Symbolic link vnode operations template
4498  */
4499 vnodeops_t *zfs_symvnodeops;
4500 const fs_operation_def_t zfs_symvnodeops_template[] = {
4501 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4502 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4503 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4504 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4505 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
4506 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4507 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4508 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4509 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4510 	NULL,			NULL
4511 };
4512 
4513 /*
4514  * Extended attribute directory vnode operations template
4515  *	This template is identical to the directory vnodes
4516  *	operation template except for restricted operations:
4517  *		VOP_MKDIR()
4518  *		VOP_SYMLINK()
4519  * Note that there are other restrictions embedded in:
4520  *	zfs_create()	- restrict type to VREG
4521  *	zfs_link()	- no links into/out of attribute space
4522  *	zfs_rename()	- no moves into/out of attribute space
4523  */
4524 vnodeops_t *zfs_xdvnodeops;
4525 const fs_operation_def_t zfs_xdvnodeops_template[] = {
4526 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4527 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4528 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4529 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4530 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4531 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4532 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4533 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4534 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4535 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4536 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4537 	VOPNAME_MKDIR,		{ .error = zfs_inval },
4538 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4539 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4540 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
4541 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4542 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4543 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4544 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4545 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4546 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4547 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4548 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4549 	NULL,			NULL
4550 };
4551 
4552 /*
4553  * Error vnode operations template
4554  */
4555 vnodeops_t *zfs_evnodeops;
4556 const fs_operation_def_t zfs_evnodeops_template[] = {
4557 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4558 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4559 	NULL,			NULL
4560 };
4561