xref: /titanic_50/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 63602c90af2802f2ab666ab39bb4aea7d02500c3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /* Portions Copyright 2007 Jeremy Teo */
27 
28 #pragma ident	"%Z%%M%	%I%	%E% SMI"
29 
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/time.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
36 #include <sys/vfs.h>
37 #include <sys/vfs_opreg.h>
38 #include <sys/vnode.h>
39 #include <sys/file.h>
40 #include <sys/stat.h>
41 #include <sys/kmem.h>
42 #include <sys/taskq.h>
43 #include <sys/uio.h>
44 #include <sys/vmsystm.h>
45 #include <sys/atomic.h>
46 #include <sys/vm.h>
47 #include <vm/seg_vn.h>
48 #include <vm/pvn.h>
49 #include <vm/as.h>
50 #include <sys/mman.h>
51 #include <sys/pathname.h>
52 #include <sys/cmn_err.h>
53 #include <sys/errno.h>
54 #include <sys/unistd.h>
55 #include <sys/zfs_vfsops.h>
56 #include <sys/zfs_dir.h>
57 #include <sys/zfs_acl.h>
58 #include <sys/zfs_ioctl.h>
59 #include <sys/fs/zfs.h>
60 #include <sys/dmu.h>
61 #include <sys/spa.h>
62 #include <sys/txg.h>
63 #include <sys/dbuf.h>
64 #include <sys/zap.h>
65 #include <sys/dirent.h>
66 #include <sys/policy.h>
67 #include <sys/sunddi.h>
68 #include <sys/filio.h>
69 #include "fs/fs_subr.h"
70 #include <sys/zfs_ctldir.h>
71 #include <sys/dnlc.h>
72 #include <sys/zfs_rlock.h>
73 
74 /*
75  * Programming rules.
76  *
77  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
78  * properly lock its in-core state, create a DMU transaction, do the work,
79  * record this work in the intent log (ZIL), commit the DMU transaction,
80  * and wait the the intent log to commit if it's is a synchronous operation.
81  * Morover, the vnode ops must work in both normal and log replay context.
82  * The ordering of events is important to avoid deadlocks and references
83  * to freed memory.  The example below illustrates the following Big Rules:
84  *
85  *  (1) A check must be made in each zfs thread for a mounted file system.
86  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
87  *	A ZFS_EXIT(zfsvfs) is needed before all returns.
88  *
89  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
90  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
91  *	First, if it's the last reference, the vnode/znode
92  *	can be freed, so the zp may point to freed memory.  Second, the last
93  *	reference will call zfs_zinactive(), which may induce a lot of work --
94  *	pushing cached pages (which acquires range locks) and syncing out
95  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
96  *	which could deadlock the system if you were already holding one.
97  *
98  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
99  *	as they can span dmu_tx_assign() calls.
100  *
101  *  (4)	Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign().
102  *	In normal operation, this will be TXG_NOWAIT.  During ZIL replay,
103  *	it will be a specific txg.  Either way, dmu_tx_assign() never blocks.
104  *	This is critical because we don't want to block while holding locks.
105  *	Note, in particular, that if a lock is sometimes acquired before
106  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
107  *	use a non-blocking assign can deadlock the system.  The scenario:
108  *
109  *	Thread A has grabbed a lock before calling dmu_tx_assign().
110  *	Thread B is in an already-assigned tx, and blocks for this lock.
111  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
112  *	forever, because the previous txg can't quiesce until B's tx commits.
113  *
114  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
115  *	then drop all locks, call dmu_tx_wait(), and try again.
116  *
117  *  (5)	If the operation succeeded, generate the intent log entry for it
118  *	before dropping locks.  This ensures that the ordering of events
119  *	in the intent log matches the order in which they actually occurred.
120  *
121  *  (6)	At the end of each vnode op, the DMU tx must always commit,
122  *	regardless of whether there were any errors.
123  *
124  *  (7)	After dropping all locks, invoke zil_commit(zilog, seq, foid)
125  *	to ensure that synchronous semantics are provided when necessary.
126  *
127  * In general, this is how things should be ordered in each vnode op:
128  *
129  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
130  * top:
131  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
132  *	rw_enter(...);			// grab any other locks you need
133  *	tx = dmu_tx_create(...);	// get DMU tx
134  *	dmu_tx_hold_*();		// hold each object you might modify
135  *	error = dmu_tx_assign(tx, zfsvfs->z_assign);	// try to assign
136  *	if (error) {
137  *		rw_exit(...);		// drop locks
138  *		zfs_dirent_unlock(dl);	// unlock directory entry
139  *		VN_RELE(...);		// release held vnodes
140  *		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
141  *			dmu_tx_wait(tx);
142  *			dmu_tx_abort(tx);
143  *			goto top;
144  *		}
145  *		dmu_tx_abort(tx);	// abort DMU tx
146  *		ZFS_EXIT(zfsvfs);	// finished in zfs
147  *		return (error);		// really out of space
148  *	}
149  *	error = do_real_work();		// do whatever this VOP does
150  *	if (error == 0)
151  *		zfs_log_*(...);		// on success, make ZIL entry
152  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
153  *	rw_exit(...);			// drop locks
154  *	zfs_dirent_unlock(dl);		// unlock directory entry
155  *	VN_RELE(...);			// release held vnodes
156  *	zil_commit(zilog, seq, foid);	// synchronous when necessary
157  *	ZFS_EXIT(zfsvfs);		// finished in zfs
158  *	return (error);			// done, report error
159  */
160 /* ARGSUSED */
161 static int
162 zfs_open(vnode_t **vpp, int flag, cred_t *cr)
163 {
164 	znode_t	*zp = VTOZ(*vpp);
165 
166 	/* Keep a count of the synchronous opens in the znode */
167 	if (flag & (FSYNC | FDSYNC))
168 		atomic_inc_32(&zp->z_sync_cnt);
169 	return (0);
170 }
171 
172 /* ARGSUSED */
173 static int
174 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr)
175 {
176 	znode_t	*zp = VTOZ(vp);
177 
178 	/* Decrement the synchronous opens in the znode */
179 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
180 		atomic_dec_32(&zp->z_sync_cnt);
181 
182 	/*
183 	 * Clean up any locks held by this process on the vp.
184 	 */
185 	cleanlocks(vp, ddi_get_pid(), 0);
186 	cleanshares(vp, ddi_get_pid());
187 
188 	return (0);
189 }
190 
191 /*
192  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
193  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
194  */
195 static int
196 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
197 {
198 	znode_t	*zp = VTOZ(vp);
199 	uint64_t noff = (uint64_t)*off; /* new offset */
200 	uint64_t file_sz;
201 	int error;
202 	boolean_t hole;
203 
204 	file_sz = zp->z_phys->zp_size;
205 	if (noff >= file_sz)  {
206 		return (ENXIO);
207 	}
208 
209 	if (cmd == _FIO_SEEK_HOLE)
210 		hole = B_TRUE;
211 	else
212 		hole = B_FALSE;
213 
214 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
215 
216 	/* end of file? */
217 	if ((error == ESRCH) || (noff > file_sz)) {
218 		/*
219 		 * Handle the virtual hole at the end of file.
220 		 */
221 		if (hole) {
222 			*off = file_sz;
223 			return (0);
224 		}
225 		return (ENXIO);
226 	}
227 
228 	if (noff < *off)
229 		return (error);
230 	*off = noff;
231 	return (error);
232 }
233 
234 /* ARGSUSED */
235 static int
236 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
237     int *rvalp)
238 {
239 	offset_t off;
240 	int error;
241 	zfsvfs_t *zfsvfs;
242 
243 	switch (com) {
244 	case _FIOFFS:
245 		return (zfs_sync(vp->v_vfsp, 0, cred));
246 
247 		/*
248 		 * The following two ioctls are used by bfu.  Faking out,
249 		 * necessary to avoid bfu errors.
250 		 */
251 	case _FIOGDIO:
252 	case _FIOSDIO:
253 		return (0);
254 
255 	case _FIO_SEEK_DATA:
256 	case _FIO_SEEK_HOLE:
257 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
258 			return (EFAULT);
259 
260 		zfsvfs = VTOZ(vp)->z_zfsvfs;
261 		ZFS_ENTER(zfsvfs);
262 
263 		/* offset parameter is in/out */
264 		error = zfs_holey(vp, com, &off);
265 		ZFS_EXIT(zfsvfs);
266 		if (error)
267 			return (error);
268 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
269 			return (EFAULT);
270 		return (0);
271 	}
272 	return (ENOTTY);
273 }
274 
275 /*
276  * When a file is memory mapped, we must keep the IO data synchronized
277  * between the DMU cache and the memory mapped pages.  What this means:
278  *
279  * On Write:	If we find a memory mapped page, we write to *both*
280  *		the page and the dmu buffer.
281  *
282  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
283  *	the file is memory mapped.
284  */
285 static int
286 mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx)
287 {
288 	znode_t	*zp = VTOZ(vp);
289 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
290 	int64_t	start, off;
291 	int len = nbytes;
292 	int error = 0;
293 
294 	start = uio->uio_loffset;
295 	off = start & PAGEOFFSET;
296 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
297 		page_t *pp;
298 		uint64_t bytes = MIN(PAGESIZE - off, len);
299 		uint64_t woff = uio->uio_loffset;
300 
301 		/*
302 		 * We don't want a new page to "appear" in the middle of
303 		 * the file update (because it may not get the write
304 		 * update data), so we grab a lock to block
305 		 * zfs_getpage().
306 		 */
307 		rw_enter(&zp->z_map_lock, RW_WRITER);
308 		if (pp = page_lookup(vp, start, SE_SHARED)) {
309 			caddr_t va;
310 
311 			rw_exit(&zp->z_map_lock);
312 			va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L);
313 			error = uiomove(va+off, bytes, UIO_WRITE, uio);
314 			if (error == 0) {
315 				dmu_write(zfsvfs->z_os, zp->z_id,
316 				    woff, bytes, va+off, tx);
317 			}
318 			ppmapout(va);
319 			page_unlock(pp);
320 		} else {
321 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
322 			    uio, bytes, tx);
323 			rw_exit(&zp->z_map_lock);
324 		}
325 		len -= bytes;
326 		off = 0;
327 		if (error)
328 			break;
329 	}
330 	return (error);
331 }
332 
333 /*
334  * When a file is memory mapped, we must keep the IO data synchronized
335  * between the DMU cache and the memory mapped pages.  What this means:
336  *
337  * On Read:	We "read" preferentially from memory mapped pages,
338  *		else we default from the dmu buffer.
339  *
340  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
341  *	the file is memory mapped.
342  */
343 static int
344 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
345 {
346 	znode_t *zp = VTOZ(vp);
347 	objset_t *os = zp->z_zfsvfs->z_os;
348 	int64_t	start, off;
349 	int len = nbytes;
350 	int error = 0;
351 
352 	start = uio->uio_loffset;
353 	off = start & PAGEOFFSET;
354 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
355 		page_t *pp;
356 		uint64_t bytes = MIN(PAGESIZE - off, len);
357 
358 		if (pp = page_lookup(vp, start, SE_SHARED)) {
359 			caddr_t va;
360 
361 			va = ppmapin(pp, PROT_READ, (caddr_t)-1L);
362 			error = uiomove(va + off, bytes, UIO_READ, uio);
363 			ppmapout(va);
364 			page_unlock(pp);
365 		} else {
366 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
367 		}
368 		len -= bytes;
369 		off = 0;
370 		if (error)
371 			break;
372 	}
373 	return (error);
374 }
375 
376 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
377 
378 /*
379  * Read bytes from specified file into supplied buffer.
380  *
381  *	IN:	vp	- vnode of file to be read from.
382  *		uio	- structure supplying read location, range info,
383  *			  and return buffer.
384  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
385  *		cr	- credentials of caller.
386  *
387  *	OUT:	uio	- updated offset and range, buffer filled.
388  *
389  *	RETURN:	0 if success
390  *		error code if failure
391  *
392  * Side Effects:
393  *	vp - atime updated if byte count > 0
394  */
395 /* ARGSUSED */
396 static int
397 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
398 {
399 	znode_t		*zp = VTOZ(vp);
400 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
401 	objset_t	*os = zfsvfs->z_os;
402 	ssize_t		n, nbytes;
403 	int		error;
404 	rl_t		*rl;
405 
406 	ZFS_ENTER(zfsvfs);
407 
408 	/*
409 	 * Validate file offset
410 	 */
411 	if (uio->uio_loffset < (offset_t)0) {
412 		ZFS_EXIT(zfsvfs);
413 		return (EINVAL);
414 	}
415 
416 	/*
417 	 * Fasttrack empty reads
418 	 */
419 	if (uio->uio_resid == 0) {
420 		ZFS_EXIT(zfsvfs);
421 		return (0);
422 	}
423 
424 	/*
425 	 * Check for mandatory locks
426 	 */
427 	if (MANDMODE((mode_t)zp->z_phys->zp_mode)) {
428 		if (error = chklock(vp, FREAD,
429 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
430 			ZFS_EXIT(zfsvfs);
431 			return (error);
432 		}
433 	}
434 
435 	/*
436 	 * If we're in FRSYNC mode, sync out this znode before reading it.
437 	 */
438 	if (ioflag & FRSYNC)
439 		zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
440 
441 	/*
442 	 * Lock the range against changes.
443 	 */
444 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
445 
446 	/*
447 	 * If we are reading past end-of-file we can skip
448 	 * to the end; but we might still need to set atime.
449 	 */
450 	if (uio->uio_loffset >= zp->z_phys->zp_size) {
451 		error = 0;
452 		goto out;
453 	}
454 
455 	ASSERT(uio->uio_loffset < zp->z_phys->zp_size);
456 	n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset);
457 
458 	while (n > 0) {
459 		nbytes = MIN(n, zfs_read_chunk_size -
460 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
461 
462 		if (vn_has_cached_data(vp))
463 			error = mappedread(vp, nbytes, uio);
464 		else
465 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
466 		if (error)
467 			break;
468 
469 		n -= nbytes;
470 	}
471 
472 out:
473 	zfs_range_unlock(rl);
474 
475 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
476 	ZFS_EXIT(zfsvfs);
477 	return (error);
478 }
479 
480 /*
481  * Fault in the pages of the first n bytes specified by the uio structure.
482  * 1 byte in each page is touched and the uio struct is unmodified.
483  * Any error will exit this routine as this is only a best
484  * attempt to get the pages resident. This is a copy of ufs_trans_touch().
485  */
486 static void
487 zfs_prefault_write(ssize_t n, struct uio *uio)
488 {
489 	struct iovec *iov;
490 	ulong_t cnt, incr;
491 	caddr_t p;
492 	uint8_t tmp;
493 
494 	iov = uio->uio_iov;
495 
496 	while (n) {
497 		cnt = MIN(iov->iov_len, n);
498 		if (cnt == 0) {
499 			/* empty iov entry */
500 			iov++;
501 			continue;
502 		}
503 		n -= cnt;
504 		/*
505 		 * touch each page in this segment.
506 		 */
507 		p = iov->iov_base;
508 		while (cnt) {
509 			switch (uio->uio_segflg) {
510 			case UIO_USERSPACE:
511 			case UIO_USERISPACE:
512 				if (fuword8(p, &tmp))
513 					return;
514 				break;
515 			case UIO_SYSSPACE:
516 				if (kcopy(p, &tmp, 1))
517 					return;
518 				break;
519 			}
520 			incr = MIN(cnt, PAGESIZE);
521 			p += incr;
522 			cnt -= incr;
523 		}
524 		/*
525 		 * touch the last byte in case it straddles a page.
526 		 */
527 		p--;
528 		switch (uio->uio_segflg) {
529 		case UIO_USERSPACE:
530 		case UIO_USERISPACE:
531 			if (fuword8(p, &tmp))
532 				return;
533 			break;
534 		case UIO_SYSSPACE:
535 			if (kcopy(p, &tmp, 1))
536 				return;
537 			break;
538 		}
539 		iov++;
540 	}
541 }
542 
543 /*
544  * Write the bytes to a file.
545  *
546  *	IN:	vp	- vnode of file to be written to.
547  *		uio	- structure supplying write location, range info,
548  *			  and data buffer.
549  *		ioflag	- FAPPEND flag set if in append mode.
550  *		cr	- credentials of caller.
551  *
552  *	OUT:	uio	- updated offset and range.
553  *
554  *	RETURN:	0 if success
555  *		error code if failure
556  *
557  * Timestamps:
558  *	vp - ctime|mtime updated if byte count > 0
559  */
560 /* ARGSUSED */
561 static int
562 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
563 {
564 	znode_t		*zp = VTOZ(vp);
565 	rlim64_t	limit = uio->uio_llimit;
566 	ssize_t		start_resid = uio->uio_resid;
567 	ssize_t		tx_bytes;
568 	uint64_t	end_size;
569 	dmu_tx_t	*tx;
570 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
571 	zilog_t		*zilog = zfsvfs->z_log;
572 	offset_t	woff;
573 	ssize_t		n, nbytes;
574 	rl_t		*rl;
575 	int		max_blksz = zfsvfs->z_max_blksz;
576 	int		error;
577 
578 	/*
579 	 * Fasttrack empty write
580 	 */
581 	n = start_resid;
582 	if (n == 0)
583 		return (0);
584 
585 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
586 		limit = MAXOFFSET_T;
587 
588 	ZFS_ENTER(zfsvfs);
589 
590 	/*
591 	 * Pre-fault the pages to ensure slow (eg NFS) pages
592 	 * don't hold up txg.
593 	 */
594 	zfs_prefault_write(n, uio);
595 
596 	/*
597 	 * If in append mode, set the io offset pointer to eof.
598 	 */
599 	if (ioflag & FAPPEND) {
600 		/*
601 		 * Range lock for a file append:
602 		 * The value for the start of range will be determined by
603 		 * zfs_range_lock() (to guarantee append semantics).
604 		 * If this write will cause the block size to increase,
605 		 * zfs_range_lock() will lock the entire file, so we must
606 		 * later reduce the range after we grow the block size.
607 		 */
608 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
609 		if (rl->r_len == UINT64_MAX) {
610 			/* overlocked, zp_size can't change */
611 			woff = uio->uio_loffset = zp->z_phys->zp_size;
612 		} else {
613 			woff = uio->uio_loffset = rl->r_off;
614 		}
615 	} else {
616 		woff = uio->uio_loffset;
617 		/*
618 		 * Validate file offset
619 		 */
620 		if (woff < 0) {
621 			ZFS_EXIT(zfsvfs);
622 			return (EINVAL);
623 		}
624 
625 		/*
626 		 * If we need to grow the block size then zfs_range_lock()
627 		 * will lock a wider range than we request here.
628 		 * Later after growing the block size we reduce the range.
629 		 */
630 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
631 	}
632 
633 	if (woff >= limit) {
634 		zfs_range_unlock(rl);
635 		ZFS_EXIT(zfsvfs);
636 		return (EFBIG);
637 	}
638 
639 	if ((woff + n) > limit || woff > (limit - n))
640 		n = limit - woff;
641 
642 	/*
643 	 * Check for mandatory locks
644 	 */
645 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) &&
646 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
647 		zfs_range_unlock(rl);
648 		ZFS_EXIT(zfsvfs);
649 		return (error);
650 	}
651 	end_size = MAX(zp->z_phys->zp_size, woff + n);
652 
653 	/*
654 	 * Write the file in reasonable size chunks.  Each chunk is written
655 	 * in a separate transaction; this keeps the intent log records small
656 	 * and allows us to do more fine-grained space accounting.
657 	 */
658 	while (n > 0) {
659 		/*
660 		 * Start a transaction.
661 		 */
662 		woff = uio->uio_loffset;
663 		tx = dmu_tx_create(zfsvfs->z_os);
664 		dmu_tx_hold_bonus(tx, zp->z_id);
665 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
666 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
667 		if (error) {
668 			if (error == ERESTART &&
669 			    zfsvfs->z_assign == TXG_NOWAIT) {
670 				dmu_tx_wait(tx);
671 				dmu_tx_abort(tx);
672 				continue;
673 			}
674 			dmu_tx_abort(tx);
675 			break;
676 		}
677 
678 		/*
679 		 * If zfs_range_lock() over-locked we grow the blocksize
680 		 * and then reduce the lock range.  This will only happen
681 		 * on the first iteration since zfs_range_reduce() will
682 		 * shrink down r_len to the appropriate size.
683 		 */
684 		if (rl->r_len == UINT64_MAX) {
685 			uint64_t new_blksz;
686 
687 			if (zp->z_blksz > max_blksz) {
688 				ASSERT(!ISP2(zp->z_blksz));
689 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
690 			} else {
691 				new_blksz = MIN(end_size, max_blksz);
692 			}
693 			zfs_grow_blocksize(zp, new_blksz, tx);
694 			zfs_range_reduce(rl, woff, n);
695 		}
696 
697 		/*
698 		 * XXX - should we really limit each write to z_max_blksz?
699 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
700 		 */
701 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
702 		rw_enter(&zp->z_map_lock, RW_READER);
703 
704 		tx_bytes = uio->uio_resid;
705 		if (vn_has_cached_data(vp)) {
706 			rw_exit(&zp->z_map_lock);
707 			error = mappedwrite(vp, nbytes, uio, tx);
708 		} else {
709 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
710 			    uio, nbytes, tx);
711 			rw_exit(&zp->z_map_lock);
712 		}
713 		tx_bytes -= uio->uio_resid;
714 
715 		/*
716 		 * If we made no progress, we're done.  If we made even
717 		 * partial progress, update the znode and ZIL accordingly.
718 		 */
719 		if (tx_bytes == 0) {
720 			dmu_tx_commit(tx);
721 			ASSERT(error != 0);
722 			break;
723 		}
724 
725 		/*
726 		 * Clear Set-UID/Set-GID bits on successful write if not
727 		 * privileged and at least one of the excute bits is set.
728 		 *
729 		 * It would be nice to to this after all writes have
730 		 * been done, but that would still expose the ISUID/ISGID
731 		 * to another app after the partial write is committed.
732 		 */
733 		mutex_enter(&zp->z_acl_lock);
734 		if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) |
735 		    (S_IXUSR >> 6))) != 0 &&
736 		    (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 &&
737 		    secpolicy_vnode_setid_retain(cr,
738 		    (zp->z_phys->zp_mode & S_ISUID) != 0 &&
739 		    zp->z_phys->zp_uid == 0) != 0) {
740 			zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID);
741 		}
742 		mutex_exit(&zp->z_acl_lock);
743 
744 		/*
745 		 * Update time stamp.  NOTE: This marks the bonus buffer as
746 		 * dirty, so we don't have to do it again for zp_size.
747 		 */
748 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
749 
750 		/*
751 		 * Update the file size (zp_size) if it has changed;
752 		 * account for possible concurrent updates.
753 		 */
754 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset)
755 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
756 			    uio->uio_loffset);
757 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
758 		dmu_tx_commit(tx);
759 
760 		if (error != 0)
761 			break;
762 		ASSERT(tx_bytes == nbytes);
763 		n -= nbytes;
764 	}
765 
766 	zfs_range_unlock(rl);
767 
768 	/*
769 	 * If we're in replay mode, or we made no progress, return error.
770 	 * Otherwise, it's at least a partial write, so it's successful.
771 	 */
772 	if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) {
773 		ZFS_EXIT(zfsvfs);
774 		return (error);
775 	}
776 
777 	if (ioflag & (FSYNC | FDSYNC))
778 		zil_commit(zilog, zp->z_last_itx, zp->z_id);
779 
780 	ZFS_EXIT(zfsvfs);
781 	return (0);
782 }
783 
784 void
785 zfs_get_done(dmu_buf_t *db, void *vzgd)
786 {
787 	zgd_t *zgd = (zgd_t *)vzgd;
788 	rl_t *rl = zgd->zgd_rl;
789 	vnode_t *vp = ZTOV(rl->r_zp);
790 
791 	dmu_buf_rele(db, vzgd);
792 	zfs_range_unlock(rl);
793 	VN_RELE(vp);
794 	zil_add_vdev(zgd->zgd_zilog, DVA_GET_VDEV(BP_IDENTITY(zgd->zgd_bp)));
795 	kmem_free(zgd, sizeof (zgd_t));
796 }
797 
798 /*
799  * Get data to generate a TX_WRITE intent log record.
800  */
801 int
802 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
803 {
804 	zfsvfs_t *zfsvfs = arg;
805 	objset_t *os = zfsvfs->z_os;
806 	znode_t *zp;
807 	uint64_t off = lr->lr_offset;
808 	dmu_buf_t *db;
809 	rl_t *rl;
810 	zgd_t *zgd;
811 	int dlen = lr->lr_length;		/* length of user data */
812 	int error = 0;
813 
814 	ASSERT(zio);
815 	ASSERT(dlen != 0);
816 
817 	/*
818 	 * Nothing to do if the file has been removed
819 	 */
820 	if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0)
821 		return (ENOENT);
822 	if (zp->z_unlinked) {
823 		VN_RELE(ZTOV(zp));
824 		return (ENOENT);
825 	}
826 
827 	/*
828 	 * Write records come in two flavors: immediate and indirect.
829 	 * For small writes it's cheaper to store the data with the
830 	 * log record (immediate); for large writes it's cheaper to
831 	 * sync the data and get a pointer to it (indirect) so that
832 	 * we don't have to write the data twice.
833 	 */
834 	if (buf != NULL) { /* immediate write */
835 		rl = zfs_range_lock(zp, off, dlen, RL_READER);
836 		/* test for truncation needs to be done while range locked */
837 		if (off >= zp->z_phys->zp_size) {
838 			error = ENOENT;
839 			goto out;
840 		}
841 		VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf));
842 	} else { /* indirect write */
843 		uint64_t boff; /* block starting offset */
844 
845 		/*
846 		 * Have to lock the whole block to ensure when it's
847 		 * written out and it's checksum is being calculated
848 		 * that no one can change the data. We need to re-check
849 		 * blocksize after we get the lock in case it's changed!
850 		 */
851 		for (;;) {
852 			if (ISP2(zp->z_blksz)) {
853 				boff = P2ALIGN_TYPED(off, zp->z_blksz,
854 				    uint64_t);
855 			} else {
856 				boff = 0;
857 			}
858 			dlen = zp->z_blksz;
859 			rl = zfs_range_lock(zp, boff, dlen, RL_READER);
860 			if (zp->z_blksz == dlen)
861 				break;
862 			zfs_range_unlock(rl);
863 		}
864 		/* test for truncation needs to be done while range locked */
865 		if (off >= zp->z_phys->zp_size) {
866 			error = ENOENT;
867 			goto out;
868 		}
869 		zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
870 		zgd->zgd_rl = rl;
871 		zgd->zgd_zilog = zfsvfs->z_log;
872 		zgd->zgd_bp = &lr->lr_blkptr;
873 		VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db));
874 		ASSERT(boff == db->db_offset);
875 		lr->lr_blkoff = off - boff;
876 		error = dmu_sync(zio, db, &lr->lr_blkptr,
877 		    lr->lr_common.lrc_txg, zfs_get_done, zgd);
878 		ASSERT((error && error != EINPROGRESS) ||
879 		    lr->lr_length <= zp->z_blksz);
880 		if (error == 0) {
881 			zil_add_vdev(zfsvfs->z_log,
882 			    DVA_GET_VDEV(BP_IDENTITY(&lr->lr_blkptr)));
883 		}
884 		/*
885 		 * If we get EINPROGRESS, then we need to wait for a
886 		 * write IO initiated by dmu_sync() to complete before
887 		 * we can release this dbuf.  We will finish everything
888 		 * up in the zfs_get_done() callback.
889 		 */
890 		if (error == EINPROGRESS)
891 			return (0);
892 		dmu_buf_rele(db, zgd);
893 		kmem_free(zgd, sizeof (zgd_t));
894 	}
895 out:
896 	zfs_range_unlock(rl);
897 	VN_RELE(ZTOV(zp));
898 	return (error);
899 }
900 
901 /*ARGSUSED*/
902 static int
903 zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr)
904 {
905 	znode_t *zp = VTOZ(vp);
906 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
907 	int error;
908 
909 	ZFS_ENTER(zfsvfs);
910 	error = zfs_zaccess_rwx(zp, mode, cr);
911 	ZFS_EXIT(zfsvfs);
912 	return (error);
913 }
914 
915 /*
916  * Lookup an entry in a directory, or an extended attribute directory.
917  * If it exists, return a held vnode reference for it.
918  *
919  *	IN:	dvp	- vnode of directory to search.
920  *		nm	- name of entry to lookup.
921  *		pnp	- full pathname to lookup [UNUSED].
922  *		flags	- LOOKUP_XATTR set if looking for an attribute.
923  *		rdir	- root directory vnode [UNUSED].
924  *		cr	- credentials of caller.
925  *
926  *	OUT:	vpp	- vnode of located entry, NULL if not found.
927  *
928  *	RETURN:	0 if success
929  *		error code if failure
930  *
931  * Timestamps:
932  *	NA
933  */
934 /* ARGSUSED */
935 static int
936 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
937     int flags, vnode_t *rdir, cred_t *cr)
938 {
939 
940 	znode_t *zdp = VTOZ(dvp);
941 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
942 	int	error;
943 
944 	ZFS_ENTER(zfsvfs);
945 
946 	*vpp = NULL;
947 
948 	if (flags & LOOKUP_XATTR) {
949 		/*
950 		 * If the xattr property is off, refuse the lookup request.
951 		 */
952 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
953 			ZFS_EXIT(zfsvfs);
954 			return (EINVAL);
955 		}
956 
957 		/*
958 		 * We don't allow recursive attributes..
959 		 * Maybe someday we will.
960 		 */
961 		if (zdp->z_phys->zp_flags & ZFS_XATTR) {
962 			ZFS_EXIT(zfsvfs);
963 			return (EINVAL);
964 		}
965 
966 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
967 			ZFS_EXIT(zfsvfs);
968 			return (error);
969 		}
970 
971 		/*
972 		 * Do we have permission to get into attribute directory?
973 		 */
974 
975 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) {
976 			VN_RELE(*vpp);
977 		}
978 
979 		ZFS_EXIT(zfsvfs);
980 		return (error);
981 	}
982 
983 	if (dvp->v_type != VDIR) {
984 		ZFS_EXIT(zfsvfs);
985 		return (ENOTDIR);
986 	}
987 
988 	/*
989 	 * Check accessibility of directory.
990 	 */
991 
992 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) {
993 		ZFS_EXIT(zfsvfs);
994 		return (error);
995 	}
996 
997 	if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) {
998 
999 		/*
1000 		 * Convert device special files
1001 		 */
1002 		if (IS_DEVVP(*vpp)) {
1003 			vnode_t	*svp;
1004 
1005 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1006 			VN_RELE(*vpp);
1007 			if (svp == NULL)
1008 				error = ENOSYS;
1009 			else
1010 				*vpp = svp;
1011 		}
1012 	}
1013 
1014 	ZFS_EXIT(zfsvfs);
1015 	return (error);
1016 }
1017 
1018 /*
1019  * Attempt to create a new entry in a directory.  If the entry
1020  * already exists, truncate the file if permissible, else return
1021  * an error.  Return the vp of the created or trunc'd file.
1022  *
1023  *	IN:	dvp	- vnode of directory to put new file entry in.
1024  *		name	- name of new file entry.
1025  *		vap	- attributes of new file.
1026  *		excl	- flag indicating exclusive or non-exclusive mode.
1027  *		mode	- mode to open file with.
1028  *		cr	- credentials of caller.
1029  *		flag	- large file flag [UNUSED].
1030  *
1031  *	OUT:	vpp	- vnode of created or trunc'd entry.
1032  *
1033  *	RETURN:	0 if success
1034  *		error code if failure
1035  *
1036  * Timestamps:
1037  *	dvp - ctime|mtime updated if new entry created
1038  *	 vp - ctime|mtime always, atime if new
1039  */
1040 /* ARGSUSED */
1041 static int
1042 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1043     int mode, vnode_t **vpp, cred_t *cr, int flag)
1044 {
1045 	znode_t		*zp, *dzp = VTOZ(dvp);
1046 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1047 	zilog_t		*zilog = zfsvfs->z_log;
1048 	objset_t	*os = zfsvfs->z_os;
1049 	zfs_dirlock_t	*dl;
1050 	dmu_tx_t	*tx;
1051 	int		error;
1052 	uint64_t	zoid;
1053 
1054 	ZFS_ENTER(zfsvfs);
1055 
1056 top:
1057 	*vpp = NULL;
1058 
1059 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1060 		vap->va_mode &= ~VSVTX;
1061 
1062 	if (*name == '\0') {
1063 		/*
1064 		 * Null component name refers to the directory itself.
1065 		 */
1066 		VN_HOLD(dvp);
1067 		zp = dzp;
1068 		dl = NULL;
1069 		error = 0;
1070 	} else {
1071 		/* possible VN_HOLD(zp) */
1072 		if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) {
1073 			if (strcmp(name, "..") == 0)
1074 				error = EISDIR;
1075 			ZFS_EXIT(zfsvfs);
1076 			return (error);
1077 		}
1078 	}
1079 
1080 	zoid = zp ? zp->z_id : -1ULL;
1081 
1082 	if (zp == NULL) {
1083 		/*
1084 		 * Create a new file object and update the directory
1085 		 * to reference it.
1086 		 */
1087 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
1088 			goto out;
1089 		}
1090 
1091 		/*
1092 		 * We only support the creation of regular files in
1093 		 * extended attribute directories.
1094 		 */
1095 		if ((dzp->z_phys->zp_flags & ZFS_XATTR) &&
1096 		    (vap->va_type != VREG)) {
1097 			error = EINVAL;
1098 			goto out;
1099 		}
1100 
1101 		tx = dmu_tx_create(os);
1102 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1103 		dmu_tx_hold_bonus(tx, dzp->z_id);
1104 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1105 		if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
1106 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1107 			    0, SPA_MAXBLOCKSIZE);
1108 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
1109 		if (error) {
1110 			zfs_dirent_unlock(dl);
1111 			if (error == ERESTART &&
1112 			    zfsvfs->z_assign == TXG_NOWAIT) {
1113 				dmu_tx_wait(tx);
1114 				dmu_tx_abort(tx);
1115 				goto top;
1116 			}
1117 			dmu_tx_abort(tx);
1118 			ZFS_EXIT(zfsvfs);
1119 			return (error);
1120 		}
1121 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
1122 		ASSERT(zp->z_id == zoid);
1123 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1124 		zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name);
1125 		dmu_tx_commit(tx);
1126 	} else {
1127 		/*
1128 		 * A directory entry already exists for this name.
1129 		 */
1130 		/*
1131 		 * Can't truncate an existing file if in exclusive mode.
1132 		 */
1133 		if (excl == EXCL) {
1134 			error = EEXIST;
1135 			goto out;
1136 		}
1137 		/*
1138 		 * Can't open a directory for writing.
1139 		 */
1140 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1141 			error = EISDIR;
1142 			goto out;
1143 		}
1144 		/*
1145 		 * Verify requested access to file.
1146 		 */
1147 		if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) {
1148 			goto out;
1149 		}
1150 
1151 		mutex_enter(&dzp->z_lock);
1152 		dzp->z_seq++;
1153 		mutex_exit(&dzp->z_lock);
1154 
1155 		/*
1156 		 * Truncate regular files if requested.
1157 		 */
1158 		if ((ZTOV(zp)->v_type == VREG) &&
1159 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1160 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1161 			if (error == ERESTART &&
1162 			    zfsvfs->z_assign == TXG_NOWAIT) {
1163 				/* NB: we already did dmu_tx_wait() */
1164 				zfs_dirent_unlock(dl);
1165 				VN_RELE(ZTOV(zp));
1166 				goto top;
1167 			}
1168 		}
1169 	}
1170 out:
1171 
1172 	if (dl)
1173 		zfs_dirent_unlock(dl);
1174 
1175 	if (error) {
1176 		if (zp)
1177 			VN_RELE(ZTOV(zp));
1178 	} else {
1179 		*vpp = ZTOV(zp);
1180 		/*
1181 		 * If vnode is for a device return a specfs vnode instead.
1182 		 */
1183 		if (IS_DEVVP(*vpp)) {
1184 			struct vnode *svp;
1185 
1186 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1187 			VN_RELE(*vpp);
1188 			if (svp == NULL) {
1189 				error = ENOSYS;
1190 			}
1191 			*vpp = svp;
1192 		}
1193 	}
1194 
1195 	ZFS_EXIT(zfsvfs);
1196 	return (error);
1197 }
1198 
1199 /*
1200  * Remove an entry from a directory.
1201  *
1202  *	IN:	dvp	- vnode of directory to remove entry from.
1203  *		name	- name of entry to remove.
1204  *		cr	- credentials of caller.
1205  *
1206  *	RETURN:	0 if success
1207  *		error code if failure
1208  *
1209  * Timestamps:
1210  *	dvp - ctime|mtime
1211  *	 vp - ctime (if nlink > 0)
1212  */
1213 static int
1214 zfs_remove(vnode_t *dvp, char *name, cred_t *cr)
1215 {
1216 	znode_t		*zp, *dzp = VTOZ(dvp);
1217 	znode_t		*xzp = NULL;
1218 	vnode_t		*vp;
1219 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1220 	zilog_t		*zilog = zfsvfs->z_log;
1221 	uint64_t	acl_obj, xattr_obj;
1222 	zfs_dirlock_t	*dl;
1223 	dmu_tx_t	*tx;
1224 	boolean_t	may_delete_now, delete_now = FALSE;
1225 	boolean_t	unlinked;
1226 	int		error;
1227 
1228 	ZFS_ENTER(zfsvfs);
1229 
1230 top:
1231 	/*
1232 	 * Attempt to lock directory; fail if entry doesn't exist.
1233 	 */
1234 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) {
1235 		ZFS_EXIT(zfsvfs);
1236 		return (error);
1237 	}
1238 
1239 	vp = ZTOV(zp);
1240 
1241 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1242 		goto out;
1243 	}
1244 
1245 	/*
1246 	 * Need to use rmdir for removing directories.
1247 	 */
1248 	if (vp->v_type == VDIR) {
1249 		error = EPERM;
1250 		goto out;
1251 	}
1252 
1253 	vnevent_remove(vp);
1254 
1255 	dnlc_remove(dvp, name);
1256 
1257 	mutex_enter(&vp->v_lock);
1258 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1259 	mutex_exit(&vp->v_lock);
1260 
1261 	/*
1262 	 * We may delete the znode now, or we may put it in the unlinked set;
1263 	 * it depends on whether we're the last link, and on whether there are
1264 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1265 	 * allow for either case.
1266 	 */
1267 	tx = dmu_tx_create(zfsvfs->z_os);
1268 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1269 	dmu_tx_hold_bonus(tx, zp->z_id);
1270 	if (may_delete_now)
1271 		dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
1272 
1273 	/* are there any extended attributes? */
1274 	if ((xattr_obj = zp->z_phys->zp_xattr) != 0) {
1275 		/* XXX - do we need this if we are deleting? */
1276 		dmu_tx_hold_bonus(tx, xattr_obj);
1277 	}
1278 
1279 	/* are there any additional acls */
1280 	if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 &&
1281 	    may_delete_now)
1282 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1283 
1284 	/* charge as an update -- would be nice not to charge at all */
1285 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1286 
1287 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1288 	if (error) {
1289 		zfs_dirent_unlock(dl);
1290 		VN_RELE(vp);
1291 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1292 			dmu_tx_wait(tx);
1293 			dmu_tx_abort(tx);
1294 			goto top;
1295 		}
1296 		dmu_tx_abort(tx);
1297 		ZFS_EXIT(zfsvfs);
1298 		return (error);
1299 	}
1300 
1301 	/*
1302 	 * Remove the directory entry.
1303 	 */
1304 	error = zfs_link_destroy(dl, zp, tx, 0, &unlinked);
1305 
1306 	if (error) {
1307 		dmu_tx_commit(tx);
1308 		goto out;
1309 	}
1310 
1311 	if (unlinked) {
1312 		mutex_enter(&vp->v_lock);
1313 		delete_now = may_delete_now &&
1314 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1315 		    zp->z_phys->zp_xattr == xattr_obj &&
1316 		    zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj;
1317 		mutex_exit(&vp->v_lock);
1318 	}
1319 
1320 	if (delete_now) {
1321 		if (zp->z_phys->zp_xattr) {
1322 			error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
1323 			ASSERT3U(error, ==, 0);
1324 			ASSERT3U(xzp->z_phys->zp_links, ==, 2);
1325 			dmu_buf_will_dirty(xzp->z_dbuf, tx);
1326 			mutex_enter(&xzp->z_lock);
1327 			xzp->z_unlinked = 1;
1328 			xzp->z_phys->zp_links = 0;
1329 			mutex_exit(&xzp->z_lock);
1330 			zfs_unlinked_add(xzp, tx);
1331 			zp->z_phys->zp_xattr = 0; /* probably unnecessary */
1332 		}
1333 		mutex_enter(&zp->z_lock);
1334 		mutex_enter(&vp->v_lock);
1335 		vp->v_count--;
1336 		ASSERT3U(vp->v_count, ==, 0);
1337 		mutex_exit(&vp->v_lock);
1338 		mutex_exit(&zp->z_lock);
1339 		zfs_znode_delete(zp, tx);
1340 		VFS_RELE(zfsvfs->z_vfs);
1341 	} else if (unlinked) {
1342 		zfs_unlinked_add(zp, tx);
1343 	}
1344 
1345 	zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name);
1346 
1347 	dmu_tx_commit(tx);
1348 out:
1349 	zfs_dirent_unlock(dl);
1350 
1351 	if (!delete_now) {
1352 		VN_RELE(vp);
1353 	} else if (xzp) {
1354 		/* this rele delayed to prevent nesting transactions */
1355 		VN_RELE(ZTOV(xzp));
1356 	}
1357 
1358 	ZFS_EXIT(zfsvfs);
1359 	return (error);
1360 }
1361 
1362 /*
1363  * Create a new directory and insert it into dvp using the name
1364  * provided.  Return a pointer to the inserted directory.
1365  *
1366  *	IN:	dvp	- vnode of directory to add subdir to.
1367  *		dirname	- name of new directory.
1368  *		vap	- attributes of new directory.
1369  *		cr	- credentials of caller.
1370  *
1371  *	OUT:	vpp	- vnode of created directory.
1372  *
1373  *	RETURN:	0 if success
1374  *		error code if failure
1375  *
1376  * Timestamps:
1377  *	dvp - ctime|mtime updated
1378  *	 vp - ctime|mtime|atime updated
1379  */
1380 static int
1381 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr)
1382 {
1383 	znode_t		*zp, *dzp = VTOZ(dvp);
1384 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1385 	zilog_t		*zilog = zfsvfs->z_log;
1386 	zfs_dirlock_t	*dl;
1387 	uint64_t	zoid = 0;
1388 	dmu_tx_t	*tx;
1389 	int		error;
1390 
1391 	ASSERT(vap->va_type == VDIR);
1392 
1393 	ZFS_ENTER(zfsvfs);
1394 
1395 	if (dzp->z_phys->zp_flags & ZFS_XATTR) {
1396 		ZFS_EXIT(zfsvfs);
1397 		return (EINVAL);
1398 	}
1399 top:
1400 	*vpp = NULL;
1401 
1402 	/*
1403 	 * First make sure the new directory doesn't exist.
1404 	 */
1405 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) {
1406 		ZFS_EXIT(zfsvfs);
1407 		return (error);
1408 	}
1409 
1410 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) {
1411 		zfs_dirent_unlock(dl);
1412 		ZFS_EXIT(zfsvfs);
1413 		return (error);
1414 	}
1415 
1416 	/*
1417 	 * Add a new entry to the directory.
1418 	 */
1419 	tx = dmu_tx_create(zfsvfs->z_os);
1420 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1421 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1422 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
1423 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1424 		    0, SPA_MAXBLOCKSIZE);
1425 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1426 	if (error) {
1427 		zfs_dirent_unlock(dl);
1428 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1429 			dmu_tx_wait(tx);
1430 			dmu_tx_abort(tx);
1431 			goto top;
1432 		}
1433 		dmu_tx_abort(tx);
1434 		ZFS_EXIT(zfsvfs);
1435 		return (error);
1436 	}
1437 
1438 	/*
1439 	 * Create new node.
1440 	 */
1441 	zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
1442 
1443 	/*
1444 	 * Now put new name in parent dir.
1445 	 */
1446 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1447 
1448 	*vpp = ZTOV(zp);
1449 
1450 	zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname);
1451 	dmu_tx_commit(tx);
1452 
1453 	zfs_dirent_unlock(dl);
1454 
1455 	ZFS_EXIT(zfsvfs);
1456 	return (0);
1457 }
1458 
1459 /*
1460  * Remove a directory subdir entry.  If the current working
1461  * directory is the same as the subdir to be removed, the
1462  * remove will fail.
1463  *
1464  *	IN:	dvp	- vnode of directory to remove from.
1465  *		name	- name of directory to be removed.
1466  *		cwd	- vnode of current working directory.
1467  *		cr	- credentials of caller.
1468  *
1469  *	RETURN:	0 if success
1470  *		error code if failure
1471  *
1472  * Timestamps:
1473  *	dvp - ctime|mtime updated
1474  */
1475 static int
1476 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr)
1477 {
1478 	znode_t		*dzp = VTOZ(dvp);
1479 	znode_t		*zp;
1480 	vnode_t		*vp;
1481 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1482 	zilog_t		*zilog = zfsvfs->z_log;
1483 	zfs_dirlock_t	*dl;
1484 	dmu_tx_t	*tx;
1485 	int		error;
1486 
1487 	ZFS_ENTER(zfsvfs);
1488 
1489 top:
1490 	zp = NULL;
1491 
1492 	/*
1493 	 * Attempt to lock directory; fail if entry doesn't exist.
1494 	 */
1495 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) {
1496 		ZFS_EXIT(zfsvfs);
1497 		return (error);
1498 	}
1499 
1500 	vp = ZTOV(zp);
1501 
1502 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1503 		goto out;
1504 	}
1505 
1506 	if (vp->v_type != VDIR) {
1507 		error = ENOTDIR;
1508 		goto out;
1509 	}
1510 
1511 	if (vp == cwd) {
1512 		error = EINVAL;
1513 		goto out;
1514 	}
1515 
1516 	vnevent_rmdir(vp);
1517 
1518 	/*
1519 	 * Grab a lock on the directory to make sure that noone is
1520 	 * trying to add (or lookup) entries while we are removing it.
1521 	 */
1522 	rw_enter(&zp->z_name_lock, RW_WRITER);
1523 
1524 	/*
1525 	 * Grab a lock on the parent pointer to make sure we play well
1526 	 * with the treewalk and directory rename code.
1527 	 */
1528 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1529 
1530 	tx = dmu_tx_create(zfsvfs->z_os);
1531 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1532 	dmu_tx_hold_bonus(tx, zp->z_id);
1533 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1534 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1535 	if (error) {
1536 		rw_exit(&zp->z_parent_lock);
1537 		rw_exit(&zp->z_name_lock);
1538 		zfs_dirent_unlock(dl);
1539 		VN_RELE(vp);
1540 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1541 			dmu_tx_wait(tx);
1542 			dmu_tx_abort(tx);
1543 			goto top;
1544 		}
1545 		dmu_tx_abort(tx);
1546 		ZFS_EXIT(zfsvfs);
1547 		return (error);
1548 	}
1549 
1550 	error = zfs_link_destroy(dl, zp, tx, 0, NULL);
1551 
1552 	if (error == 0)
1553 		zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name);
1554 
1555 	dmu_tx_commit(tx);
1556 
1557 	rw_exit(&zp->z_parent_lock);
1558 	rw_exit(&zp->z_name_lock);
1559 out:
1560 	zfs_dirent_unlock(dl);
1561 
1562 	VN_RELE(vp);
1563 
1564 	ZFS_EXIT(zfsvfs);
1565 	return (error);
1566 }
1567 
1568 /*
1569  * Read as many directory entries as will fit into the provided
1570  * buffer from the given directory cursor position (specified in
1571  * the uio structure.
1572  *
1573  *	IN:	vp	- vnode of directory to read.
1574  *		uio	- structure supplying read location, range info,
1575  *			  and return buffer.
1576  *		cr	- credentials of caller.
1577  *
1578  *	OUT:	uio	- updated offset and range, buffer filled.
1579  *		eofp	- set to true if end-of-file detected.
1580  *
1581  *	RETURN:	0 if success
1582  *		error code if failure
1583  *
1584  * Timestamps:
1585  *	vp - atime updated
1586  *
1587  * Note that the low 4 bits of the cookie returned by zap is always zero.
1588  * This allows us to use the low range for "special" directory entries:
1589  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1590  * we use the offset 2 for the '.zfs' directory.
1591  */
1592 /* ARGSUSED */
1593 static int
1594 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp)
1595 {
1596 	znode_t		*zp = VTOZ(vp);
1597 	iovec_t		*iovp;
1598 	dirent64_t	*odp;
1599 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1600 	objset_t	*os;
1601 	caddr_t		outbuf;
1602 	size_t		bufsize;
1603 	zap_cursor_t	zc;
1604 	zap_attribute_t	zap;
1605 	uint_t		bytes_wanted;
1606 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1607 	int		local_eof;
1608 	int		outcount;
1609 	int		error;
1610 	uint8_t		prefetch;
1611 
1612 	ZFS_ENTER(zfsvfs);
1613 
1614 	/*
1615 	 * If we are not given an eof variable,
1616 	 * use a local one.
1617 	 */
1618 	if (eofp == NULL)
1619 		eofp = &local_eof;
1620 
1621 	/*
1622 	 * Check for valid iov_len.
1623 	 */
1624 	if (uio->uio_iov->iov_len <= 0) {
1625 		ZFS_EXIT(zfsvfs);
1626 		return (EINVAL);
1627 	}
1628 
1629 	/*
1630 	 * Quit if directory has been removed (posix)
1631 	 */
1632 	if ((*eofp = zp->z_unlinked) != 0) {
1633 		ZFS_EXIT(zfsvfs);
1634 		return (0);
1635 	}
1636 
1637 	error = 0;
1638 	os = zfsvfs->z_os;
1639 	offset = uio->uio_loffset;
1640 	prefetch = zp->z_zn_prefetch;
1641 
1642 	/*
1643 	 * Initialize the iterator cursor.
1644 	 */
1645 	if (offset <= 3) {
1646 		/*
1647 		 * Start iteration from the beginning of the directory.
1648 		 */
1649 		zap_cursor_init(&zc, os, zp->z_id);
1650 	} else {
1651 		/*
1652 		 * The offset is a serialized cursor.
1653 		 */
1654 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1655 	}
1656 
1657 	/*
1658 	 * Get space to change directory entries into fs independent format.
1659 	 */
1660 	iovp = uio->uio_iov;
1661 	bytes_wanted = iovp->iov_len;
1662 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
1663 		bufsize = bytes_wanted;
1664 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
1665 		odp = (struct dirent64 *)outbuf;
1666 	} else {
1667 		bufsize = bytes_wanted;
1668 		odp = (struct dirent64 *)iovp->iov_base;
1669 	}
1670 
1671 	/*
1672 	 * Transform to file-system independent format
1673 	 */
1674 	outcount = 0;
1675 	while (outcount < bytes_wanted) {
1676 		ino64_t objnum;
1677 		ushort_t reclen;
1678 		off64_t *next;
1679 
1680 		/*
1681 		 * Special case `.', `..', and `.zfs'.
1682 		 */
1683 		if (offset == 0) {
1684 			(void) strcpy(zap.za_name, ".");
1685 			objnum = zp->z_id;
1686 		} else if (offset == 1) {
1687 			(void) strcpy(zap.za_name, "..");
1688 			objnum = zp->z_phys->zp_parent;
1689 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
1690 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1691 			objnum = ZFSCTL_INO_ROOT;
1692 		} else {
1693 			/*
1694 			 * Grab next entry.
1695 			 */
1696 			if (error = zap_cursor_retrieve(&zc, &zap)) {
1697 				if ((*eofp = (error == ENOENT)) != 0)
1698 					break;
1699 				else
1700 					goto update;
1701 			}
1702 
1703 			if (zap.za_integer_length != 8 ||
1704 			    zap.za_num_integers != 1) {
1705 				cmn_err(CE_WARN, "zap_readdir: bad directory "
1706 				    "entry, obj = %lld, offset = %lld\n",
1707 				    (u_longlong_t)zp->z_id,
1708 				    (u_longlong_t)offset);
1709 				error = ENXIO;
1710 				goto update;
1711 			}
1712 
1713 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
1714 			/*
1715 			 * MacOS X can extract the object type here such as:
1716 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
1717 			 */
1718 		}
1719 		reclen = DIRENT64_RECLEN(strlen(zap.za_name));
1720 
1721 		/*
1722 		 * Will this entry fit in the buffer?
1723 		 */
1724 		if (outcount + reclen > bufsize) {
1725 			/*
1726 			 * Did we manage to fit anything in the buffer?
1727 			 */
1728 			if (!outcount) {
1729 				error = EINVAL;
1730 				goto update;
1731 			}
1732 			break;
1733 		}
1734 		/*
1735 		 * Add this entry:
1736 		 */
1737 		odp->d_ino = objnum;
1738 		odp->d_reclen = reclen;
1739 		/* NOTE: d_off is the offset for the *next* entry */
1740 		next = &(odp->d_off);
1741 		(void) strncpy(odp->d_name, zap.za_name,
1742 		    DIRENT64_NAMELEN(reclen));
1743 		outcount += reclen;
1744 		odp = (dirent64_t *)((intptr_t)odp + reclen);
1745 
1746 		ASSERT(outcount <= bufsize);
1747 
1748 		/* Prefetch znode */
1749 		if (prefetch)
1750 			dmu_prefetch(os, objnum, 0, 0);
1751 
1752 		/*
1753 		 * Move to the next entry, fill in the previous offset.
1754 		 */
1755 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
1756 			zap_cursor_advance(&zc);
1757 			offset = zap_cursor_serialize(&zc);
1758 		} else {
1759 			offset += 1;
1760 		}
1761 		*next = offset;
1762 	}
1763 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
1764 
1765 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
1766 		iovp->iov_base += outcount;
1767 		iovp->iov_len -= outcount;
1768 		uio->uio_resid -= outcount;
1769 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
1770 		/*
1771 		 * Reset the pointer.
1772 		 */
1773 		offset = uio->uio_loffset;
1774 	}
1775 
1776 update:
1777 	zap_cursor_fini(&zc);
1778 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
1779 		kmem_free(outbuf, bufsize);
1780 
1781 	if (error == ENOENT)
1782 		error = 0;
1783 
1784 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
1785 
1786 	uio->uio_loffset = offset;
1787 	ZFS_EXIT(zfsvfs);
1788 	return (error);
1789 }
1790 
1791 ulong_t zfs_fsync_sync_cnt = 4;
1792 
1793 static int
1794 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr)
1795 {
1796 	znode_t	*zp = VTOZ(vp);
1797 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1798 
1799 	/*
1800 	 * Regardless of whether this is required for standards conformance,
1801 	 * this is the logical behavior when fsync() is called on a file with
1802 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
1803 	 * going to be pushed out as part of the zil_commit().
1804 	 */
1805 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
1806 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
1807 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr);
1808 
1809 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
1810 
1811 	ZFS_ENTER(zfsvfs);
1812 	zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
1813 	ZFS_EXIT(zfsvfs);
1814 	return (0);
1815 }
1816 
1817 /*
1818  * Get the requested file attributes and place them in the provided
1819  * vattr structure.
1820  *
1821  *	IN:	vp	- vnode of file.
1822  *		vap	- va_mask identifies requested attributes.
1823  *		flags	- [UNUSED]
1824  *		cr	- credentials of caller.
1825  *
1826  *	OUT:	vap	- attribute values.
1827  *
1828  *	RETURN:	0 (always succeeds)
1829  */
1830 /* ARGSUSED */
1831 static int
1832 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr)
1833 {
1834 	znode_t *zp = VTOZ(vp);
1835 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1836 	znode_phys_t *pzp = zp->z_phys;
1837 	int	error;
1838 	uint64_t links;
1839 
1840 	ZFS_ENTER(zfsvfs);
1841 
1842 	/*
1843 	 * Return all attributes.  It's cheaper to provide the answer
1844 	 * than to determine whether we were asked the question.
1845 	 */
1846 	mutex_enter(&zp->z_lock);
1847 
1848 	vap->va_type = vp->v_type;
1849 	vap->va_mode = pzp->zp_mode & MODEMASK;
1850 	vap->va_uid = zp->z_phys->zp_uid;
1851 	vap->va_gid = zp->z_phys->zp_gid;
1852 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
1853 	vap->va_nodeid = zp->z_id;
1854 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
1855 		links = pzp->zp_links + 1;
1856 	else
1857 		links = pzp->zp_links;
1858 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
1859 	vap->va_size = pzp->zp_size;
1860 	vap->va_rdev = vp->v_rdev;
1861 	vap->va_seq = zp->z_seq;
1862 
1863 	ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime);
1864 	ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime);
1865 	ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime);
1866 
1867 	/*
1868 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
1869 	 * Also, if we are the owner don't bother, since owner should
1870 	 * always be allowed to read basic attributes of file.
1871 	 */
1872 	if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) &&
1873 	    (zp->z_phys->zp_uid != crgetuid(cr))) {
1874 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) {
1875 			mutex_exit(&zp->z_lock);
1876 			ZFS_EXIT(zfsvfs);
1877 			return (error);
1878 		}
1879 	}
1880 
1881 	mutex_exit(&zp->z_lock);
1882 
1883 	dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks);
1884 
1885 	if (zp->z_blksz == 0) {
1886 		/*
1887 		 * Block size hasn't been set; suggest maximal I/O transfers.
1888 		 */
1889 		vap->va_blksize = zfsvfs->z_max_blksz;
1890 	}
1891 
1892 	ZFS_EXIT(zfsvfs);
1893 	return (0);
1894 }
1895 
1896 /*
1897  * Set the file attributes to the values contained in the
1898  * vattr structure.
1899  *
1900  *	IN:	vp	- vnode of file to be modified.
1901  *		vap	- new attribute values.
1902  *		flags	- ATTR_UTIME set if non-default time values provided.
1903  *		cr	- credentials of caller.
1904  *
1905  *	RETURN:	0 if success
1906  *		error code if failure
1907  *
1908  * Timestamps:
1909  *	vp - ctime updated, mtime updated if size changed.
1910  */
1911 /* ARGSUSED */
1912 static int
1913 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1914 	caller_context_t *ct)
1915 {
1916 	struct znode	*zp = VTOZ(vp);
1917 	znode_phys_t	*pzp = zp->z_phys;
1918 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1919 	zilog_t		*zilog = zfsvfs->z_log;
1920 	dmu_tx_t	*tx;
1921 	vattr_t		oldva;
1922 	uint_t		mask = vap->va_mask;
1923 	uint_t		saved_mask;
1924 	int		trim_mask = 0;
1925 	uint64_t	new_mode;
1926 	znode_t		*attrzp;
1927 	int		need_policy = FALSE;
1928 	int		err;
1929 
1930 	if (mask == 0)
1931 		return (0);
1932 
1933 	if (mask & AT_NOSET)
1934 		return (EINVAL);
1935 
1936 	if (mask & AT_SIZE && vp->v_type == VDIR)
1937 		return (EISDIR);
1938 
1939 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO)
1940 		return (EINVAL);
1941 
1942 	ZFS_ENTER(zfsvfs);
1943 
1944 top:
1945 	attrzp = NULL;
1946 
1947 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
1948 		ZFS_EXIT(zfsvfs);
1949 		return (EROFS);
1950 	}
1951 
1952 	/*
1953 	 * First validate permissions
1954 	 */
1955 
1956 	if (mask & AT_SIZE) {
1957 		err = zfs_zaccess(zp, ACE_WRITE_DATA, cr);
1958 		if (err) {
1959 			ZFS_EXIT(zfsvfs);
1960 			return (err);
1961 		}
1962 		/*
1963 		 * XXX - Note, we are not providing any open
1964 		 * mode flags here (like FNDELAY), so we may
1965 		 * block if there are locks present... this
1966 		 * should be addressed in openat().
1967 		 */
1968 		do {
1969 			err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
1970 			/* NB: we already did dmu_tx_wait() if necessary */
1971 		} while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
1972 		if (err) {
1973 			ZFS_EXIT(zfsvfs);
1974 			return (err);
1975 		}
1976 	}
1977 
1978 	if (mask & (AT_ATIME|AT_MTIME))
1979 		need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr);
1980 
1981 	if (mask & (AT_UID|AT_GID)) {
1982 		int	idmask = (mask & (AT_UID|AT_GID));
1983 		int	take_owner;
1984 		int	take_group;
1985 
1986 		/*
1987 		 * NOTE: even if a new mode is being set,
1988 		 * we may clear S_ISUID/S_ISGID bits.
1989 		 */
1990 
1991 		if (!(mask & AT_MODE))
1992 			vap->va_mode = pzp->zp_mode;
1993 
1994 		/*
1995 		 * Take ownership or chgrp to group we are a member of
1996 		 */
1997 
1998 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
1999 		take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr);
2000 
2001 		/*
2002 		 * If both AT_UID and AT_GID are set then take_owner and
2003 		 * take_group must both be set in order to allow taking
2004 		 * ownership.
2005 		 *
2006 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2007 		 *
2008 		 */
2009 
2010 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2011 		    ((idmask == AT_UID) && take_owner) ||
2012 		    ((idmask == AT_GID) && take_group)) {
2013 			if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) {
2014 				/*
2015 				 * Remove setuid/setgid for non-privileged users
2016 				 */
2017 				secpolicy_setid_clear(vap, cr);
2018 				trim_mask = (mask & (AT_UID|AT_GID));
2019 			} else {
2020 				need_policy =  TRUE;
2021 			}
2022 		} else {
2023 			need_policy =  TRUE;
2024 		}
2025 	}
2026 
2027 	mutex_enter(&zp->z_lock);
2028 	oldva.va_mode = pzp->zp_mode;
2029 	oldva.va_uid = zp->z_phys->zp_uid;
2030 	oldva.va_gid = zp->z_phys->zp_gid;
2031 	mutex_exit(&zp->z_lock);
2032 
2033 	if (mask & AT_MODE) {
2034 		if (zfs_zaccess_v4_perm(zp, ACE_WRITE_ACL, cr) == 0) {
2035 			err = secpolicy_setid_setsticky_clear(vp, vap,
2036 			    &oldva, cr);
2037 			if (err) {
2038 				ZFS_EXIT(zfsvfs);
2039 				return (err);
2040 			}
2041 			trim_mask |= AT_MODE;
2042 		} else {
2043 			need_policy = TRUE;
2044 		}
2045 	}
2046 
2047 	if (need_policy) {
2048 		/*
2049 		 * If trim_mask is set then take ownership
2050 		 * has been granted or write_acl is present and user
2051 		 * has the ability to modify mode.  In that case remove
2052 		 * UID|GID and or MODE from mask so that
2053 		 * secpolicy_vnode_setattr() doesn't revoke it.
2054 		 */
2055 
2056 		if (trim_mask) {
2057 			saved_mask = vap->va_mask;
2058 			vap->va_mask &= ~trim_mask;
2059 
2060 		}
2061 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2062 		    (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp);
2063 		if (err) {
2064 			ZFS_EXIT(zfsvfs);
2065 			return (err);
2066 		}
2067 
2068 		if (trim_mask)
2069 			vap->va_mask |= saved_mask;
2070 	}
2071 
2072 	/*
2073 	 * secpolicy_vnode_setattr, or take ownership may have
2074 	 * changed va_mask
2075 	 */
2076 	mask = vap->va_mask;
2077 
2078 	tx = dmu_tx_create(zfsvfs->z_os);
2079 	dmu_tx_hold_bonus(tx, zp->z_id);
2080 
2081 	if (mask & AT_MODE) {
2082 		uint64_t pmode = pzp->zp_mode;
2083 
2084 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2085 
2086 		if (zp->z_phys->zp_acl.z_acl_extern_obj)
2087 			dmu_tx_hold_write(tx,
2088 			    pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE);
2089 		else
2090 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2091 			    0, ZFS_ACL_SIZE(MAX_ACL_SIZE));
2092 	}
2093 
2094 	if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) {
2095 		err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp);
2096 		if (err) {
2097 			dmu_tx_abort(tx);
2098 			ZFS_EXIT(zfsvfs);
2099 			return (err);
2100 		}
2101 		dmu_tx_hold_bonus(tx, attrzp->z_id);
2102 	}
2103 
2104 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2105 	if (err) {
2106 		if (attrzp)
2107 			VN_RELE(ZTOV(attrzp));
2108 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2109 			dmu_tx_wait(tx);
2110 			dmu_tx_abort(tx);
2111 			goto top;
2112 		}
2113 		dmu_tx_abort(tx);
2114 		ZFS_EXIT(zfsvfs);
2115 		return (err);
2116 	}
2117 
2118 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2119 
2120 	/*
2121 	 * Set each attribute requested.
2122 	 * We group settings according to the locks they need to acquire.
2123 	 *
2124 	 * Note: you cannot set ctime directly, although it will be
2125 	 * updated as a side-effect of calling this function.
2126 	 */
2127 
2128 	mutex_enter(&zp->z_lock);
2129 
2130 	if (mask & AT_MODE) {
2131 		err = zfs_acl_chmod_setattr(zp, new_mode, tx);
2132 		ASSERT3U(err, ==, 0);
2133 	}
2134 
2135 	if (attrzp)
2136 		mutex_enter(&attrzp->z_lock);
2137 
2138 	if (mask & AT_UID) {
2139 		zp->z_phys->zp_uid = (uint64_t)vap->va_uid;
2140 		if (attrzp) {
2141 			attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid;
2142 		}
2143 	}
2144 
2145 	if (mask & AT_GID) {
2146 		zp->z_phys->zp_gid = (uint64_t)vap->va_gid;
2147 		if (attrzp)
2148 			attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid;
2149 	}
2150 
2151 	if (attrzp)
2152 		mutex_exit(&attrzp->z_lock);
2153 
2154 	if (mask & AT_ATIME)
2155 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2156 
2157 	if (mask & AT_MTIME)
2158 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2159 
2160 	if (mask & AT_SIZE)
2161 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2162 	else if (mask != 0)
2163 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2164 
2165 	if (mask != 0)
2166 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask);
2167 
2168 	mutex_exit(&zp->z_lock);
2169 
2170 	if (attrzp)
2171 		VN_RELE(ZTOV(attrzp));
2172 
2173 	dmu_tx_commit(tx);
2174 
2175 	ZFS_EXIT(zfsvfs);
2176 	return (err);
2177 }
2178 
2179 typedef struct zfs_zlock {
2180 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2181 	znode_t		*zl_znode;	/* znode we held */
2182 	struct zfs_zlock *zl_next;	/* next in list */
2183 } zfs_zlock_t;
2184 
2185 /*
2186  * Drop locks and release vnodes that were held by zfs_rename_lock().
2187  */
2188 static void
2189 zfs_rename_unlock(zfs_zlock_t **zlpp)
2190 {
2191 	zfs_zlock_t *zl;
2192 
2193 	while ((zl = *zlpp) != NULL) {
2194 		if (zl->zl_znode != NULL)
2195 			VN_RELE(ZTOV(zl->zl_znode));
2196 		rw_exit(zl->zl_rwlock);
2197 		*zlpp = zl->zl_next;
2198 		kmem_free(zl, sizeof (*zl));
2199 	}
2200 }
2201 
2202 /*
2203  * Search back through the directory tree, using the ".." entries.
2204  * Lock each directory in the chain to prevent concurrent renames.
2205  * Fail any attempt to move a directory into one of its own descendants.
2206  * XXX - z_parent_lock can overlap with map or grow locks
2207  */
2208 static int
2209 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2210 {
2211 	zfs_zlock_t	*zl;
2212 	znode_t		*zp = tdzp;
2213 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2214 	uint64_t	*oidp = &zp->z_id;
2215 	krwlock_t	*rwlp = &szp->z_parent_lock;
2216 	krw_t		rw = RW_WRITER;
2217 
2218 	/*
2219 	 * First pass write-locks szp and compares to zp->z_id.
2220 	 * Later passes read-lock zp and compare to zp->z_parent.
2221 	 */
2222 	do {
2223 		if (!rw_tryenter(rwlp, rw)) {
2224 			/*
2225 			 * Another thread is renaming in this path.
2226 			 * Note that if we are a WRITER, we don't have any
2227 			 * parent_locks held yet.
2228 			 */
2229 			if (rw == RW_READER && zp->z_id > szp->z_id) {
2230 				/*
2231 				 * Drop our locks and restart
2232 				 */
2233 				zfs_rename_unlock(&zl);
2234 				*zlpp = NULL;
2235 				zp = tdzp;
2236 				oidp = &zp->z_id;
2237 				rwlp = &szp->z_parent_lock;
2238 				rw = RW_WRITER;
2239 				continue;
2240 			} else {
2241 				/*
2242 				 * Wait for other thread to drop its locks
2243 				 */
2244 				rw_enter(rwlp, rw);
2245 			}
2246 		}
2247 
2248 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2249 		zl->zl_rwlock = rwlp;
2250 		zl->zl_znode = NULL;
2251 		zl->zl_next = *zlpp;
2252 		*zlpp = zl;
2253 
2254 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2255 			return (EINVAL);
2256 
2257 		if (*oidp == rootid)		/* We've hit the top */
2258 			return (0);
2259 
2260 		if (rw == RW_READER) {		/* i.e. not the first pass */
2261 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2262 			if (error)
2263 				return (error);
2264 			zl->zl_znode = zp;
2265 		}
2266 		oidp = &zp->z_phys->zp_parent;
2267 		rwlp = &zp->z_parent_lock;
2268 		rw = RW_READER;
2269 
2270 	} while (zp->z_id != sdzp->z_id);
2271 
2272 	return (0);
2273 }
2274 
2275 /*
2276  * Move an entry from the provided source directory to the target
2277  * directory.  Change the entry name as indicated.
2278  *
2279  *	IN:	sdvp	- Source directory containing the "old entry".
2280  *		snm	- Old entry name.
2281  *		tdvp	- Target directory to contain the "new entry".
2282  *		tnm	- New entry name.
2283  *		cr	- credentials of caller.
2284  *
2285  *	RETURN:	0 if success
2286  *		error code if failure
2287  *
2288  * Timestamps:
2289  *	sdvp,tdvp - ctime|mtime updated
2290  */
2291 static int
2292 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr)
2293 {
2294 	znode_t		*tdzp, *szp, *tzp;
2295 	znode_t		*sdzp = VTOZ(sdvp);
2296 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2297 	zilog_t		*zilog = zfsvfs->z_log;
2298 	vnode_t		*realvp;
2299 	zfs_dirlock_t	*sdl, *tdl;
2300 	dmu_tx_t	*tx;
2301 	zfs_zlock_t	*zl;
2302 	int		cmp, serr, terr, error;
2303 
2304 	ZFS_ENTER(zfsvfs);
2305 
2306 	/*
2307 	 * Make sure we have the real vp for the target directory.
2308 	 */
2309 	if (VOP_REALVP(tdvp, &realvp) == 0)
2310 		tdvp = realvp;
2311 
2312 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
2313 		ZFS_EXIT(zfsvfs);
2314 		return (EXDEV);
2315 	}
2316 
2317 	tdzp = VTOZ(tdvp);
2318 top:
2319 	szp = NULL;
2320 	tzp = NULL;
2321 	zl = NULL;
2322 
2323 	/*
2324 	 * This is to prevent the creation of links into attribute space
2325 	 * by renaming a linked file into/outof an attribute directory.
2326 	 * See the comment in zfs_link() for why this is considered bad.
2327 	 */
2328 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
2329 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
2330 		ZFS_EXIT(zfsvfs);
2331 		return (EINVAL);
2332 	}
2333 
2334 	/*
2335 	 * Lock source and target directory entries.  To prevent deadlock,
2336 	 * a lock ordering must be defined.  We lock the directory with
2337 	 * the smallest object id first, or if it's a tie, the one with
2338 	 * the lexically first name.
2339 	 */
2340 	if (sdzp->z_id < tdzp->z_id) {
2341 		cmp = -1;
2342 	} else if (sdzp->z_id > tdzp->z_id) {
2343 		cmp = 1;
2344 	} else {
2345 		cmp = strcmp(snm, tnm);
2346 		if (cmp == 0) {
2347 			/*
2348 			 * POSIX: "If the old argument and the new argument
2349 			 * both refer to links to the same existing file,
2350 			 * the rename() function shall return successfully
2351 			 * and perform no other action."
2352 			 */
2353 			ZFS_EXIT(zfsvfs);
2354 			return (0);
2355 		}
2356 	}
2357 	if (cmp < 0) {
2358 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS);
2359 		terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0);
2360 	} else {
2361 		terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0);
2362 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS);
2363 	}
2364 
2365 	if (serr) {
2366 		/*
2367 		 * Source entry invalid or not there.
2368 		 */
2369 		if (!terr) {
2370 			zfs_dirent_unlock(tdl);
2371 			if (tzp)
2372 				VN_RELE(ZTOV(tzp));
2373 		}
2374 		if (strcmp(snm, "..") == 0)
2375 			serr = EINVAL;
2376 		ZFS_EXIT(zfsvfs);
2377 		return (serr);
2378 	}
2379 	if (terr) {
2380 		zfs_dirent_unlock(sdl);
2381 		VN_RELE(ZTOV(szp));
2382 		if (strcmp(tnm, "..") == 0)
2383 			terr = EINVAL;
2384 		ZFS_EXIT(zfsvfs);
2385 		return (terr);
2386 	}
2387 
2388 	/*
2389 	 * Must have write access at the source to remove the old entry
2390 	 * and write access at the target to create the new entry.
2391 	 * Note that if target and source are the same, this can be
2392 	 * done in a single check.
2393 	 */
2394 
2395 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
2396 		goto out;
2397 
2398 	if (ZTOV(szp)->v_type == VDIR) {
2399 		/*
2400 		 * Check to make sure rename is valid.
2401 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
2402 		 */
2403 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
2404 			goto out;
2405 	}
2406 
2407 	/*
2408 	 * Does target exist?
2409 	 */
2410 	if (tzp) {
2411 		/*
2412 		 * Source and target must be the same type.
2413 		 */
2414 		if (ZTOV(szp)->v_type == VDIR) {
2415 			if (ZTOV(tzp)->v_type != VDIR) {
2416 				error = ENOTDIR;
2417 				goto out;
2418 			}
2419 		} else {
2420 			if (ZTOV(tzp)->v_type == VDIR) {
2421 				error = EISDIR;
2422 				goto out;
2423 			}
2424 		}
2425 		/*
2426 		 * POSIX dictates that when the source and target
2427 		 * entries refer to the same file object, rename
2428 		 * must do nothing and exit without error.
2429 		 */
2430 		if (szp->z_id == tzp->z_id) {
2431 			error = 0;
2432 			goto out;
2433 		}
2434 	}
2435 
2436 	vnevent_rename_src(ZTOV(szp));
2437 	if (tzp)
2438 		vnevent_rename_dest(ZTOV(tzp));
2439 
2440 	tx = dmu_tx_create(zfsvfs->z_os);
2441 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
2442 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
2443 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
2444 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
2445 	if (sdzp != tdzp)
2446 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
2447 	if (tzp)
2448 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* parent changes */
2449 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2450 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2451 	if (error) {
2452 		if (zl != NULL)
2453 			zfs_rename_unlock(&zl);
2454 		zfs_dirent_unlock(sdl);
2455 		zfs_dirent_unlock(tdl);
2456 		VN_RELE(ZTOV(szp));
2457 		if (tzp)
2458 			VN_RELE(ZTOV(tzp));
2459 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2460 			dmu_tx_wait(tx);
2461 			dmu_tx_abort(tx);
2462 			goto top;
2463 		}
2464 		dmu_tx_abort(tx);
2465 		ZFS_EXIT(zfsvfs);
2466 		return (error);
2467 	}
2468 
2469 	if (tzp)	/* Attempt to remove the existing target */
2470 		error = zfs_link_destroy(tdl, tzp, tx, 0, NULL);
2471 
2472 	if (error == 0) {
2473 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
2474 		if (error == 0) {
2475 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
2476 			ASSERT(error == 0);
2477 			zfs_log_rename(zilog, tx, TX_RENAME, sdzp,
2478 			    sdl->dl_name, tdzp, tdl->dl_name, szp);
2479 		}
2480 	}
2481 
2482 	dmu_tx_commit(tx);
2483 out:
2484 	if (zl != NULL)
2485 		zfs_rename_unlock(&zl);
2486 
2487 	zfs_dirent_unlock(sdl);
2488 	zfs_dirent_unlock(tdl);
2489 
2490 	VN_RELE(ZTOV(szp));
2491 	if (tzp)
2492 		VN_RELE(ZTOV(tzp));
2493 
2494 	ZFS_EXIT(zfsvfs);
2495 	return (error);
2496 }
2497 
2498 /*
2499  * Insert the indicated symbolic reference entry into the directory.
2500  *
2501  *	IN:	dvp	- Directory to contain new symbolic link.
2502  *		link	- Name for new symlink entry.
2503  *		vap	- Attributes of new entry.
2504  *		target	- Target path of new symlink.
2505  *		cr	- credentials of caller.
2506  *
2507  *	RETURN:	0 if success
2508  *		error code if failure
2509  *
2510  * Timestamps:
2511  *	dvp - ctime|mtime updated
2512  */
2513 static int
2514 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr)
2515 {
2516 	znode_t		*zp, *dzp = VTOZ(dvp);
2517 	zfs_dirlock_t	*dl;
2518 	dmu_tx_t	*tx;
2519 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2520 	zilog_t		*zilog = zfsvfs->z_log;
2521 	uint64_t	zoid;
2522 	int		len = strlen(link);
2523 	int		error;
2524 
2525 	ASSERT(vap->va_type == VLNK);
2526 
2527 	ZFS_ENTER(zfsvfs);
2528 top:
2529 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
2530 		ZFS_EXIT(zfsvfs);
2531 		return (error);
2532 	}
2533 
2534 	if (len > MAXPATHLEN) {
2535 		ZFS_EXIT(zfsvfs);
2536 		return (ENAMETOOLONG);
2537 	}
2538 
2539 	/*
2540 	 * Attempt to lock directory; fail if entry already exists.
2541 	 */
2542 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) {
2543 		ZFS_EXIT(zfsvfs);
2544 		return (error);
2545 	}
2546 
2547 	tx = dmu_tx_create(zfsvfs->z_os);
2548 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
2549 	dmu_tx_hold_bonus(tx, dzp->z_id);
2550 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
2551 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
2552 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
2553 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2554 	if (error) {
2555 		zfs_dirent_unlock(dl);
2556 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2557 			dmu_tx_wait(tx);
2558 			dmu_tx_abort(tx);
2559 			goto top;
2560 		}
2561 		dmu_tx_abort(tx);
2562 		ZFS_EXIT(zfsvfs);
2563 		return (error);
2564 	}
2565 
2566 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
2567 
2568 	/*
2569 	 * Create a new object for the symlink.
2570 	 * Put the link content into bonus buffer if it will fit;
2571 	 * otherwise, store it just like any other file data.
2572 	 */
2573 	zoid = 0;
2574 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
2575 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len);
2576 		if (len != 0)
2577 			bcopy(link, zp->z_phys + 1, len);
2578 	} else {
2579 		dmu_buf_t *dbp;
2580 
2581 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
2582 
2583 		/*
2584 		 * Nothing can access the znode yet so no locking needed
2585 		 * for growing the znode's blocksize.
2586 		 */
2587 		zfs_grow_blocksize(zp, len, tx);
2588 
2589 		VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp));
2590 		dmu_buf_will_dirty(dbp, tx);
2591 
2592 		ASSERT3U(len, <=, dbp->db_size);
2593 		bcopy(link, dbp->db_data, len);
2594 		dmu_buf_rele(dbp, FTAG);
2595 	}
2596 	zp->z_phys->zp_size = len;
2597 
2598 	/*
2599 	 * Insert the new object into the directory.
2600 	 */
2601 	(void) zfs_link_create(dl, zp, tx, ZNEW);
2602 out:
2603 	if (error == 0)
2604 		zfs_log_symlink(zilog, tx, TX_SYMLINK, dzp, zp, name, link);
2605 
2606 	dmu_tx_commit(tx);
2607 
2608 	zfs_dirent_unlock(dl);
2609 
2610 	VN_RELE(ZTOV(zp));
2611 
2612 	ZFS_EXIT(zfsvfs);
2613 	return (error);
2614 }
2615 
2616 /*
2617  * Return, in the buffer contained in the provided uio structure,
2618  * the symbolic path referred to by vp.
2619  *
2620  *	IN:	vp	- vnode of symbolic link.
2621  *		uoip	- structure to contain the link path.
2622  *		cr	- credentials of caller.
2623  *
2624  *	OUT:	uio	- structure to contain the link path.
2625  *
2626  *	RETURN:	0 if success
2627  *		error code if failure
2628  *
2629  * Timestamps:
2630  *	vp - atime updated
2631  */
2632 /* ARGSUSED */
2633 static int
2634 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr)
2635 {
2636 	znode_t		*zp = VTOZ(vp);
2637 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2638 	size_t		bufsz;
2639 	int		error;
2640 
2641 	ZFS_ENTER(zfsvfs);
2642 
2643 	bufsz = (size_t)zp->z_phys->zp_size;
2644 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
2645 		error = uiomove(zp->z_phys + 1,
2646 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
2647 	} else {
2648 		dmu_buf_t *dbp;
2649 		error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp);
2650 		if (error) {
2651 			ZFS_EXIT(zfsvfs);
2652 			return (error);
2653 		}
2654 		error = uiomove(dbp->db_data,
2655 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
2656 		dmu_buf_rele(dbp, FTAG);
2657 	}
2658 
2659 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2660 	ZFS_EXIT(zfsvfs);
2661 	return (error);
2662 }
2663 
2664 /*
2665  * Insert a new entry into directory tdvp referencing svp.
2666  *
2667  *	IN:	tdvp	- Directory to contain new entry.
2668  *		svp	- vnode of new entry.
2669  *		name	- name of new entry.
2670  *		cr	- credentials of caller.
2671  *
2672  *	RETURN:	0 if success
2673  *		error code if failure
2674  *
2675  * Timestamps:
2676  *	tdvp - ctime|mtime updated
2677  *	 svp - ctime updated
2678  */
2679 /* ARGSUSED */
2680 static int
2681 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr)
2682 {
2683 	znode_t		*dzp = VTOZ(tdvp);
2684 	znode_t		*tzp, *szp;
2685 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2686 	zilog_t		*zilog = zfsvfs->z_log;
2687 	zfs_dirlock_t	*dl;
2688 	dmu_tx_t	*tx;
2689 	vnode_t		*realvp;
2690 	int		error;
2691 
2692 	ASSERT(tdvp->v_type == VDIR);
2693 
2694 	ZFS_ENTER(zfsvfs);
2695 
2696 	if (VOP_REALVP(svp, &realvp) == 0)
2697 		svp = realvp;
2698 
2699 	if (svp->v_vfsp != tdvp->v_vfsp) {
2700 		ZFS_EXIT(zfsvfs);
2701 		return (EXDEV);
2702 	}
2703 
2704 	szp = VTOZ(svp);
2705 top:
2706 	/*
2707 	 * We do not support links between attributes and non-attributes
2708 	 * because of the potential security risk of creating links
2709 	 * into "normal" file space in order to circumvent restrictions
2710 	 * imposed in attribute space.
2711 	 */
2712 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
2713 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
2714 		ZFS_EXIT(zfsvfs);
2715 		return (EINVAL);
2716 	}
2717 
2718 	/*
2719 	 * POSIX dictates that we return EPERM here.
2720 	 * Better choices include ENOTSUP or EISDIR.
2721 	 */
2722 	if (svp->v_type == VDIR) {
2723 		ZFS_EXIT(zfsvfs);
2724 		return (EPERM);
2725 	}
2726 
2727 	if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) &&
2728 	    secpolicy_basic_link(cr) != 0) {
2729 		ZFS_EXIT(zfsvfs);
2730 		return (EPERM);
2731 	}
2732 
2733 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
2734 		ZFS_EXIT(zfsvfs);
2735 		return (error);
2736 	}
2737 
2738 	/*
2739 	 * Attempt to lock directory; fail if entry already exists.
2740 	 */
2741 	if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) {
2742 		ZFS_EXIT(zfsvfs);
2743 		return (error);
2744 	}
2745 
2746 	tx = dmu_tx_create(zfsvfs->z_os);
2747 	dmu_tx_hold_bonus(tx, szp->z_id);
2748 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
2749 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2750 	if (error) {
2751 		zfs_dirent_unlock(dl);
2752 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2753 			dmu_tx_wait(tx);
2754 			dmu_tx_abort(tx);
2755 			goto top;
2756 		}
2757 		dmu_tx_abort(tx);
2758 		ZFS_EXIT(zfsvfs);
2759 		return (error);
2760 	}
2761 
2762 	error = zfs_link_create(dl, szp, tx, 0);
2763 
2764 	if (error == 0)
2765 		zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name);
2766 
2767 	dmu_tx_commit(tx);
2768 
2769 	zfs_dirent_unlock(dl);
2770 
2771 	ZFS_EXIT(zfsvfs);
2772 	return (error);
2773 }
2774 
2775 /*
2776  * zfs_null_putapage() is used when the file system has been force
2777  * unmounted. It just drops the pages.
2778  */
2779 /* ARGSUSED */
2780 static int
2781 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
2782 		size_t *lenp, int flags, cred_t *cr)
2783 {
2784 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
2785 	return (0);
2786 }
2787 
2788 /*
2789  * Push a page out to disk, klustering if possible.
2790  *
2791  *	IN:	vp	- file to push page to.
2792  *		pp	- page to push.
2793  *		flags	- additional flags.
2794  *		cr	- credentials of caller.
2795  *
2796  *	OUT:	offp	- start of range pushed.
2797  *		lenp	- len of range pushed.
2798  *
2799  *	RETURN:	0 if success
2800  *		error code if failure
2801  *
2802  * NOTE: callers must have locked the page to be pushed.  On
2803  * exit, the page (and all other pages in the kluster) must be
2804  * unlocked.
2805  */
2806 /* ARGSUSED */
2807 static int
2808 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
2809 		size_t *lenp, int flags, cred_t *cr)
2810 {
2811 	znode_t		*zp = VTOZ(vp);
2812 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2813 	zilog_t		*zilog = zfsvfs->z_log;
2814 	dmu_tx_t	*tx;
2815 	rl_t		*rl;
2816 	u_offset_t	off, koff;
2817 	size_t		len, klen;
2818 	uint64_t	filesz;
2819 	int		err;
2820 
2821 	filesz = zp->z_phys->zp_size;
2822 	off = pp->p_offset;
2823 	len = PAGESIZE;
2824 	/*
2825 	 * If our blocksize is bigger than the page size, try to kluster
2826 	 * muiltiple pages so that we write a full block (thus avoiding
2827 	 * a read-modify-write).
2828 	 */
2829 	if (off < filesz && zp->z_blksz > PAGESIZE) {
2830 		if (!ISP2(zp->z_blksz)) {
2831 			/* Only one block in the file. */
2832 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
2833 			koff = 0;
2834 		} else {
2835 			klen = zp->z_blksz;
2836 			koff = P2ALIGN(off, (u_offset_t)klen);
2837 		}
2838 		ASSERT(koff <= filesz);
2839 		if (koff + klen > filesz)
2840 			klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE);
2841 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
2842 	}
2843 	ASSERT3U(btop(len), ==, btopr(len));
2844 top:
2845 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
2846 	/*
2847 	 * Can't push pages past end-of-file.
2848 	 */
2849 	filesz = zp->z_phys->zp_size;
2850 	if (off >= filesz) {
2851 		/* ignore all pages */
2852 		err = 0;
2853 		goto out;
2854 	} else if (off + len > filesz) {
2855 		int npages = btopr(filesz - off);
2856 		page_t *trunc;
2857 
2858 		page_list_break(&pp, &trunc, npages);
2859 		/* ignore pages past end of file */
2860 		if (trunc)
2861 			pvn_write_done(trunc, flags);
2862 		len = filesz - off;
2863 	}
2864 
2865 	tx = dmu_tx_create(zfsvfs->z_os);
2866 	dmu_tx_hold_write(tx, zp->z_id, off, len);
2867 	dmu_tx_hold_bonus(tx, zp->z_id);
2868 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2869 	if (err != 0) {
2870 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2871 			zfs_range_unlock(rl);
2872 			dmu_tx_wait(tx);
2873 			dmu_tx_abort(tx);
2874 			err = 0;
2875 			goto top;
2876 		}
2877 		dmu_tx_abort(tx);
2878 		goto out;
2879 	}
2880 
2881 	if (zp->z_blksz <= PAGESIZE) {
2882 		caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1);
2883 		ASSERT3U(len, <=, PAGESIZE);
2884 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
2885 		ppmapout(va);
2886 	} else {
2887 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
2888 	}
2889 
2890 	if (err == 0) {
2891 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
2892 		zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0);
2893 		dmu_tx_commit(tx);
2894 	}
2895 
2896 out:
2897 	zfs_range_unlock(rl);
2898 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
2899 	if (offp)
2900 		*offp = off;
2901 	if (lenp)
2902 		*lenp = len;
2903 
2904 	return (err);
2905 }
2906 
2907 /*
2908  * Copy the portion of the file indicated from pages into the file.
2909  * The pages are stored in a page list attached to the files vnode.
2910  *
2911  *	IN:	vp	- vnode of file to push page data to.
2912  *		off	- position in file to put data.
2913  *		len	- amount of data to write.
2914  *		flags	- flags to control the operation.
2915  *		cr	- credentials of caller.
2916  *
2917  *	RETURN:	0 if success
2918  *		error code if failure
2919  *
2920  * Timestamps:
2921  *	vp - ctime|mtime updated
2922  */
2923 static int
2924 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr)
2925 {
2926 	znode_t		*zp = VTOZ(vp);
2927 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2928 	page_t		*pp;
2929 	size_t		io_len;
2930 	u_offset_t	io_off;
2931 	uint64_t	filesz;
2932 	int		error = 0;
2933 
2934 	ZFS_ENTER(zfsvfs);
2935 
2936 	ASSERT(zp->z_dbuf_held && zp->z_phys);
2937 
2938 	if (len == 0) {
2939 		/*
2940 		 * Search the entire vp list for pages >= off.
2941 		 */
2942 		error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage,
2943 		    flags, cr);
2944 		goto out;
2945 	}
2946 
2947 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
2948 	if (off > filesz) {
2949 		/* past end of file */
2950 		ZFS_EXIT(zfsvfs);
2951 		return (0);
2952 	}
2953 
2954 	len = MIN(len, filesz - off);
2955 
2956 	for (io_off = off; io_off < off + len; io_off += io_len) {
2957 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
2958 			pp = page_lookup(vp, io_off,
2959 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
2960 		} else {
2961 			pp = page_lookup_nowait(vp, io_off,
2962 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
2963 		}
2964 
2965 		if (pp != NULL && pvn_getdirty(pp, flags)) {
2966 			int err;
2967 
2968 			/*
2969 			 * Found a dirty page to push
2970 			 */
2971 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
2972 			if (err)
2973 				error = err;
2974 		} else {
2975 			io_len = PAGESIZE;
2976 		}
2977 	}
2978 out:
2979 	if ((flags & B_ASYNC) == 0)
2980 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
2981 	ZFS_EXIT(zfsvfs);
2982 	return (error);
2983 }
2984 
2985 void
2986 zfs_inactive(vnode_t *vp, cred_t *cr)
2987 {
2988 	znode_t	*zp = VTOZ(vp);
2989 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2990 	int error;
2991 
2992 	rw_enter(&zfsvfs->z_unmount_inactive_lock, RW_READER);
2993 	if (zfsvfs->z_unmounted) {
2994 		ASSERT(zp->z_dbuf_held == 0);
2995 
2996 		if (vn_has_cached_data(vp)) {
2997 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
2998 			    B_INVAL, cr);
2999 		}
3000 
3001 		mutex_enter(&zp->z_lock);
3002 		vp->v_count = 0; /* count arrives as 1 */
3003 		if (zp->z_dbuf == NULL) {
3004 			mutex_exit(&zp->z_lock);
3005 			zfs_znode_free(zp);
3006 		} else {
3007 			mutex_exit(&zp->z_lock);
3008 		}
3009 		rw_exit(&zfsvfs->z_unmount_inactive_lock);
3010 		VFS_RELE(zfsvfs->z_vfs);
3011 		return;
3012 	}
3013 
3014 	/*
3015 	 * Attempt to push any data in the page cache.  If this fails
3016 	 * we will get kicked out later in zfs_zinactive().
3017 	 */
3018 	if (vn_has_cached_data(vp)) {
3019 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
3020 		    cr);
3021 	}
3022 
3023 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3024 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3025 
3026 		dmu_tx_hold_bonus(tx, zp->z_id);
3027 		error = dmu_tx_assign(tx, TXG_WAIT);
3028 		if (error) {
3029 			dmu_tx_abort(tx);
3030 		} else {
3031 			dmu_buf_will_dirty(zp->z_dbuf, tx);
3032 			mutex_enter(&zp->z_lock);
3033 			zp->z_atime_dirty = 0;
3034 			mutex_exit(&zp->z_lock);
3035 			dmu_tx_commit(tx);
3036 		}
3037 	}
3038 
3039 	zfs_zinactive(zp);
3040 	rw_exit(&zfsvfs->z_unmount_inactive_lock);
3041 }
3042 
3043 /*
3044  * Bounds-check the seek operation.
3045  *
3046  *	IN:	vp	- vnode seeking within
3047  *		ooff	- old file offset
3048  *		noffp	- pointer to new file offset
3049  *
3050  *	RETURN:	0 if success
3051  *		EINVAL if new offset invalid
3052  */
3053 /* ARGSUSED */
3054 static int
3055 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp)
3056 {
3057 	if (vp->v_type == VDIR)
3058 		return (0);
3059 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3060 }
3061 
3062 /*
3063  * Pre-filter the generic locking function to trap attempts to place
3064  * a mandatory lock on a memory mapped file.
3065  */
3066 static int
3067 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3068     flk_callback_t *flk_cbp, cred_t *cr)
3069 {
3070 	znode_t *zp = VTOZ(vp);
3071 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3072 	int error;
3073 
3074 	ZFS_ENTER(zfsvfs);
3075 
3076 	/*
3077 	 * We are following the UFS semantics with respect to mapcnt
3078 	 * here: If we see that the file is mapped already, then we will
3079 	 * return an error, but we don't worry about races between this
3080 	 * function and zfs_map().
3081 	 */
3082 	if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3083 		ZFS_EXIT(zfsvfs);
3084 		return (EAGAIN);
3085 	}
3086 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr);
3087 	ZFS_EXIT(zfsvfs);
3088 	return (error);
3089 }
3090 
3091 /*
3092  * If we can't find a page in the cache, we will create a new page
3093  * and fill it with file data.  For efficiency, we may try to fill
3094  * multiple pages at once (klustering).
3095  */
3096 static int
3097 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3098     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3099 {
3100 	znode_t *zp = VTOZ(vp);
3101 	page_t *pp, *cur_pp;
3102 	objset_t *os = zp->z_zfsvfs->z_os;
3103 	caddr_t va;
3104 	u_offset_t io_off, total;
3105 	uint64_t oid = zp->z_id;
3106 	size_t io_len;
3107 	uint64_t filesz;
3108 	int err;
3109 
3110 	/*
3111 	 * If we are only asking for a single page don't bother klustering.
3112 	 */
3113 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3114 	if (off >= filesz)
3115 		return (EFAULT);
3116 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3117 		io_off = off;
3118 		io_len = PAGESIZE;
3119 		pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr);
3120 	} else {
3121 		/*
3122 		 * Try to fill a kluster of pages (a blocks worth).
3123 		 */
3124 		size_t klen;
3125 		u_offset_t koff;
3126 
3127 		if (!ISP2(zp->z_blksz)) {
3128 			/* Only one block in the file. */
3129 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3130 			koff = 0;
3131 		} else {
3132 			/*
3133 			 * It would be ideal to align our offset to the
3134 			 * blocksize but doing so has resulted in some
3135 			 * strange application crashes. For now, we
3136 			 * leave the offset as is and only adjust the
3137 			 * length if we are off the end of the file.
3138 			 */
3139 			koff = off;
3140 			klen = plsz;
3141 		}
3142 		ASSERT(koff <= filesz);
3143 		if (koff + klen > filesz)
3144 			klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff;
3145 		ASSERT3U(off, >=, koff);
3146 		ASSERT3U(off, <, koff + klen);
3147 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3148 		    &io_len, koff, klen, 0);
3149 	}
3150 	if (pp == NULL) {
3151 		/*
3152 		 * Some other thread entered the page before us.
3153 		 * Return to zfs_getpage to retry the lookup.
3154 		 */
3155 		*pl = NULL;
3156 		return (0);
3157 	}
3158 
3159 	/*
3160 	 * Fill the pages in the kluster.
3161 	 */
3162 	cur_pp = pp;
3163 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3164 		ASSERT3U(io_off, ==, cur_pp->p_offset);
3165 		va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1);
3166 		err = dmu_read(os, oid, io_off, PAGESIZE, va);
3167 		ppmapout(va);
3168 		if (err) {
3169 			/* On error, toss the entire kluster */
3170 			pvn_read_done(pp, B_ERROR);
3171 			return (err);
3172 		}
3173 		cur_pp = cur_pp->p_next;
3174 	}
3175 out:
3176 	/*
3177 	 * Fill in the page list array from the kluster.  If
3178 	 * there are too many pages in the kluster, return
3179 	 * as many pages as possible starting from the desired
3180 	 * offset `off'.
3181 	 * NOTE: the page list will always be null terminated.
3182 	 */
3183 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3184 
3185 	return (0);
3186 }
3187 
3188 /*
3189  * Return pointers to the pages for the file region [off, off + len]
3190  * in the pl array.  If plsz is greater than len, this function may
3191  * also return page pointers from before or after the specified
3192  * region (i.e. some region [off', off' + plsz]).  These additional
3193  * pages are only returned if they are already in the cache, or were
3194  * created as part of a klustered read.
3195  *
3196  *	IN:	vp	- vnode of file to get data from.
3197  *		off	- position in file to get data from.
3198  *		len	- amount of data to retrieve.
3199  *		plsz	- length of provided page list.
3200  *		seg	- segment to obtain pages for.
3201  *		addr	- virtual address of fault.
3202  *		rw	- mode of created pages.
3203  *		cr	- credentials of caller.
3204  *
3205  *	OUT:	protp	- protection mode of created pages.
3206  *		pl	- list of pages created.
3207  *
3208  *	RETURN:	0 if success
3209  *		error code if failure
3210  *
3211  * Timestamps:
3212  *	vp - atime updated
3213  */
3214 /* ARGSUSED */
3215 static int
3216 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3217 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3218 	enum seg_rw rw, cred_t *cr)
3219 {
3220 	znode_t		*zp = VTOZ(vp);
3221 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3222 	page_t		*pp, **pl0 = pl;
3223 	int		need_unlock = 0, err = 0;
3224 	offset_t	orig_off;
3225 
3226 	ZFS_ENTER(zfsvfs);
3227 
3228 	if (protp)
3229 		*protp = PROT_ALL;
3230 
3231 	ASSERT(zp->z_dbuf_held && zp->z_phys);
3232 
3233 	/* no faultahead (for now) */
3234 	if (pl == NULL) {
3235 		ZFS_EXIT(zfsvfs);
3236 		return (0);
3237 	}
3238 
3239 	/* can't fault past EOF */
3240 	if (off >= zp->z_phys->zp_size) {
3241 		ZFS_EXIT(zfsvfs);
3242 		return (EFAULT);
3243 	}
3244 	orig_off = off;
3245 
3246 	/*
3247 	 * If we already own the lock, then we must be page faulting
3248 	 * in the middle of a write to this file (i.e., we are writing
3249 	 * to this file using data from a mapped region of the file).
3250 	 */
3251 	if (rw_owner(&zp->z_map_lock) != curthread) {
3252 		rw_enter(&zp->z_map_lock, RW_WRITER);
3253 		need_unlock = TRUE;
3254 	}
3255 
3256 	/*
3257 	 * Loop through the requested range [off, off + len] looking
3258 	 * for pages.  If we don't find a page, we will need to create
3259 	 * a new page and fill it with data from the file.
3260 	 */
3261 	while (len > 0) {
3262 		if (plsz < PAGESIZE)
3263 			break;
3264 		if (pp = page_lookup(vp, off, SE_SHARED)) {
3265 			*pl++ = pp;
3266 			off += PAGESIZE;
3267 			addr += PAGESIZE;
3268 			len -= PAGESIZE;
3269 			plsz -= PAGESIZE;
3270 		} else {
3271 			err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw);
3272 			if (err)
3273 				goto out;
3274 			/*
3275 			 * klustering may have changed our region
3276 			 * to be block aligned.
3277 			 */
3278 			if (((pp = *pl) != 0) && (off != pp->p_offset)) {
3279 				int delta = off - pp->p_offset;
3280 				len += delta;
3281 				off -= delta;
3282 				addr -= delta;
3283 			}
3284 			while (*pl) {
3285 				pl++;
3286 				off += PAGESIZE;
3287 				addr += PAGESIZE;
3288 				plsz -= PAGESIZE;
3289 				if (len > PAGESIZE)
3290 					len -= PAGESIZE;
3291 				else
3292 					len = 0;
3293 			}
3294 		}
3295 	}
3296 
3297 	/*
3298 	 * Fill out the page array with any pages already in the cache.
3299 	 */
3300 	while (plsz > 0) {
3301 		pp = page_lookup_nowait(vp, off, SE_SHARED);
3302 		if (pp == NULL)
3303 			break;
3304 		*pl++ = pp;
3305 		off += PAGESIZE;
3306 		plsz -= PAGESIZE;
3307 	}
3308 
3309 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3310 out:
3311 	/*
3312 	 * We can't grab the range lock for the page as reader which would
3313 	 * stop truncation as this leads to deadlock. So we need to recheck
3314 	 * the file size.
3315 	 */
3316 	if (orig_off >= zp->z_phys->zp_size)
3317 		err = EFAULT;
3318 	if (err) {
3319 		/*
3320 		 * Release any pages we have previously locked.
3321 		 */
3322 		while (pl > pl0)
3323 			page_unlock(*--pl);
3324 	}
3325 
3326 	*pl = NULL;
3327 
3328 	if (need_unlock)
3329 		rw_exit(&zp->z_map_lock);
3330 
3331 	ZFS_EXIT(zfsvfs);
3332 	return (err);
3333 }
3334 
3335 /*
3336  * Request a memory map for a section of a file.  This code interacts
3337  * with common code and the VM system as follows:
3338  *
3339  *	common code calls mmap(), which ends up in smmap_common()
3340  *
3341  *	this calls VOP_MAP(), which takes you into (say) zfs
3342  *
3343  *	zfs_map() calls as_map(), passing segvn_create() as the callback
3344  *
3345  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
3346  *
3347  *	zfs_addmap() updates z_mapcnt
3348  */
3349 static int
3350 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
3351     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3352 {
3353 	znode_t *zp = VTOZ(vp);
3354 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3355 	segvn_crargs_t	vn_a;
3356 	int		error;
3357 
3358 	ZFS_ENTER(zfsvfs);
3359 
3360 	if (vp->v_flag & VNOMAP) {
3361 		ZFS_EXIT(zfsvfs);
3362 		return (ENOSYS);
3363 	}
3364 
3365 	if (off < 0 || len > MAXOFFSET_T - off) {
3366 		ZFS_EXIT(zfsvfs);
3367 		return (ENXIO);
3368 	}
3369 
3370 	if (vp->v_type != VREG) {
3371 		ZFS_EXIT(zfsvfs);
3372 		return (ENODEV);
3373 	}
3374 
3375 	/*
3376 	 * If file is locked, disallow mapping.
3377 	 */
3378 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) {
3379 		ZFS_EXIT(zfsvfs);
3380 		return (EAGAIN);
3381 	}
3382 
3383 	as_rangelock(as);
3384 	if ((flags & MAP_FIXED) == 0) {
3385 		map_addr(addrp, len, off, 1, flags);
3386 		if (*addrp == NULL) {
3387 			as_rangeunlock(as);
3388 			ZFS_EXIT(zfsvfs);
3389 			return (ENOMEM);
3390 		}
3391 	} else {
3392 		/*
3393 		 * User specified address - blow away any previous mappings
3394 		 */
3395 		(void) as_unmap(as, *addrp, len);
3396 	}
3397 
3398 	vn_a.vp = vp;
3399 	vn_a.offset = (u_offset_t)off;
3400 	vn_a.type = flags & MAP_TYPE;
3401 	vn_a.prot = prot;
3402 	vn_a.maxprot = maxprot;
3403 	vn_a.cred = cr;
3404 	vn_a.amp = NULL;
3405 	vn_a.flags = flags & ~MAP_TYPE;
3406 	vn_a.szc = 0;
3407 	vn_a.lgrp_mem_policy_flags = 0;
3408 
3409 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
3410 
3411 	as_rangeunlock(as);
3412 	ZFS_EXIT(zfsvfs);
3413 	return (error);
3414 }
3415 
3416 /* ARGSUSED */
3417 static int
3418 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3419     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3420 {
3421 	uint64_t pages = btopr(len);
3422 
3423 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
3424 	return (0);
3425 }
3426 
3427 /*
3428  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
3429  * more accurate mtime for the associated file.  Since we don't have a way of
3430  * detecting when the data was actually modified, we have to resort to
3431  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
3432  * last page is pushed.  The problem occurs when the msync() call is omitted,
3433  * which by far the most common case:
3434  *
3435  * 	open()
3436  * 	mmap()
3437  * 	<modify memory>
3438  * 	munmap()
3439  * 	close()
3440  * 	<time lapse>
3441  * 	putpage() via fsflush
3442  *
3443  * If we wait until fsflush to come along, we can have a modification time that
3444  * is some arbitrary point in the future.  In order to prevent this in the
3445  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
3446  * torn down.
3447  */
3448 /* ARGSUSED */
3449 static int
3450 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3451     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr)
3452 {
3453 	uint64_t pages = btopr(len);
3454 
3455 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
3456 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
3457 
3458 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
3459 	    vn_has_cached_data(vp))
3460 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr);
3461 
3462 	return (0);
3463 }
3464 
3465 /*
3466  * Free or allocate space in a file.  Currently, this function only
3467  * supports the `F_FREESP' command.  However, this command is somewhat
3468  * misnamed, as its functionality includes the ability to allocate as
3469  * well as free space.
3470  *
3471  *	IN:	vp	- vnode of file to free data in.
3472  *		cmd	- action to take (only F_FREESP supported).
3473  *		bfp	- section of file to free/alloc.
3474  *		flag	- current file open mode flags.
3475  *		offset	- current file offset.
3476  *		cr	- credentials of caller [UNUSED].
3477  *
3478  *	RETURN:	0 if success
3479  *		error code if failure
3480  *
3481  * Timestamps:
3482  *	vp - ctime|mtime updated
3483  */
3484 /* ARGSUSED */
3485 static int
3486 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
3487     offset_t offset, cred_t *cr, caller_context_t *ct)
3488 {
3489 	znode_t		*zp = VTOZ(vp);
3490 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3491 	uint64_t	off, len;
3492 	int		error;
3493 
3494 	ZFS_ENTER(zfsvfs);
3495 
3496 top:
3497 	if (cmd != F_FREESP) {
3498 		ZFS_EXIT(zfsvfs);
3499 		return (EINVAL);
3500 	}
3501 
3502 	if (error = convoff(vp, bfp, 0, offset)) {
3503 		ZFS_EXIT(zfsvfs);
3504 		return (error);
3505 	}
3506 
3507 	if (bfp->l_len < 0) {
3508 		ZFS_EXIT(zfsvfs);
3509 		return (EINVAL);
3510 	}
3511 
3512 	off = bfp->l_start;
3513 	len = bfp->l_len; /* 0 means from off to end of file */
3514 
3515 	do {
3516 		error = zfs_freesp(zp, off, len, flag, TRUE);
3517 		/* NB: we already did dmu_tx_wait() if necessary */
3518 	} while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
3519 
3520 	ZFS_EXIT(zfsvfs);
3521 	return (error);
3522 }
3523 
3524 static int
3525 zfs_fid(vnode_t *vp, fid_t *fidp)
3526 {
3527 	znode_t		*zp = VTOZ(vp);
3528 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3529 	uint32_t	gen = (uint32_t)zp->z_phys->zp_gen;
3530 	uint64_t	object = zp->z_id;
3531 	zfid_short_t	*zfid;
3532 	int		size, i;
3533 
3534 	ZFS_ENTER(zfsvfs);
3535 
3536 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
3537 	if (fidp->fid_len < size) {
3538 		fidp->fid_len = size;
3539 		ZFS_EXIT(zfsvfs);
3540 		return (ENOSPC);
3541 	}
3542 
3543 	zfid = (zfid_short_t *)fidp;
3544 
3545 	zfid->zf_len = size;
3546 
3547 	for (i = 0; i < sizeof (zfid->zf_object); i++)
3548 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
3549 
3550 	/* Must have a non-zero generation number to distinguish from .zfs */
3551 	if (gen == 0)
3552 		gen = 1;
3553 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
3554 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
3555 
3556 	if (size == LONG_FID_LEN) {
3557 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
3558 		zfid_long_t	*zlfid;
3559 
3560 		zlfid = (zfid_long_t *)fidp;
3561 
3562 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
3563 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
3564 
3565 		/* XXX - this should be the generation number for the objset */
3566 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
3567 			zlfid->zf_setgen[i] = 0;
3568 	}
3569 
3570 	ZFS_EXIT(zfsvfs);
3571 	return (0);
3572 }
3573 
3574 static int
3575 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr)
3576 {
3577 	znode_t		*zp, *xzp;
3578 	zfsvfs_t	*zfsvfs;
3579 	zfs_dirlock_t	*dl;
3580 	int		error;
3581 
3582 	switch (cmd) {
3583 	case _PC_LINK_MAX:
3584 		*valp = ULONG_MAX;
3585 		return (0);
3586 
3587 	case _PC_FILESIZEBITS:
3588 		*valp = 64;
3589 		return (0);
3590 
3591 	case _PC_XATTR_EXISTS:
3592 		zp = VTOZ(vp);
3593 		zfsvfs = zp->z_zfsvfs;
3594 		ZFS_ENTER(zfsvfs);
3595 		*valp = 0;
3596 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
3597 		    ZXATTR | ZEXISTS | ZSHARED);
3598 		if (error == 0) {
3599 			zfs_dirent_unlock(dl);
3600 			if (!zfs_dirempty(xzp))
3601 				*valp = 1;
3602 			VN_RELE(ZTOV(xzp));
3603 		} else if (error == ENOENT) {
3604 			/*
3605 			 * If there aren't extended attributes, it's the
3606 			 * same as having zero of them.
3607 			 */
3608 			error = 0;
3609 		}
3610 		ZFS_EXIT(zfsvfs);
3611 		return (error);
3612 
3613 	case _PC_ACL_ENABLED:
3614 		*valp = _ACL_ACE_ENABLED;
3615 		return (0);
3616 
3617 	case _PC_MIN_HOLE_SIZE:
3618 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
3619 		return (0);
3620 
3621 	default:
3622 		return (fs_pathconf(vp, cmd, valp, cr));
3623 	}
3624 }
3625 
3626 /*ARGSUSED*/
3627 static int
3628 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr)
3629 {
3630 	znode_t *zp = VTOZ(vp);
3631 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3632 	int error;
3633 
3634 	ZFS_ENTER(zfsvfs);
3635 	error = zfs_getacl(zp, vsecp, cr);
3636 	ZFS_EXIT(zfsvfs);
3637 
3638 	return (error);
3639 }
3640 
3641 /*ARGSUSED*/
3642 static int
3643 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr)
3644 {
3645 	znode_t *zp = VTOZ(vp);
3646 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3647 	int error;
3648 
3649 	ZFS_ENTER(zfsvfs);
3650 	error = zfs_setacl(zp, vsecp, cr);
3651 	ZFS_EXIT(zfsvfs);
3652 	return (error);
3653 }
3654 
3655 /*
3656  * Predeclare these here so that the compiler assumes that
3657  * this is an "old style" function declaration that does
3658  * not include arguments => we won't get type mismatch errors
3659  * in the initializations that follow.
3660  */
3661 static int zfs_inval();
3662 static int zfs_isdir();
3663 
3664 static int
3665 zfs_inval()
3666 {
3667 	return (EINVAL);
3668 }
3669 
3670 static int
3671 zfs_isdir()
3672 {
3673 	return (EISDIR);
3674 }
3675 /*
3676  * Directory vnode operations template
3677  */
3678 vnodeops_t *zfs_dvnodeops;
3679 const fs_operation_def_t zfs_dvnodeops_template[] = {
3680 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
3681 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
3682 	VOPNAME_READ,		{ .error = zfs_isdir },
3683 	VOPNAME_WRITE,		{ .error = zfs_isdir },
3684 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
3685 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
3686 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
3687 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
3688 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
3689 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
3690 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
3691 	VOPNAME_LINK,		{ .vop_link = zfs_link },
3692 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
3693 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
3694 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
3695 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
3696 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
3697 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
3698 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
3699 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
3700 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
3701 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
3702 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
3703 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
3704 	NULL,			NULL
3705 };
3706 
3707 /*
3708  * Regular file vnode operations template
3709  */
3710 vnodeops_t *zfs_fvnodeops;
3711 const fs_operation_def_t zfs_fvnodeops_template[] = {
3712 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
3713 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
3714 	VOPNAME_READ,		{ .vop_read = zfs_read },
3715 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
3716 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
3717 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
3718 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
3719 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
3720 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
3721 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
3722 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
3723 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
3724 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
3725 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
3726 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
3727 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
3728 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
3729 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
3730 	VOPNAME_MAP,		{ .vop_map = zfs_map },
3731 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
3732 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
3733 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
3734 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
3735 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
3736 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
3737 	NULL,			NULL
3738 };
3739 
3740 /*
3741  * Symbolic link vnode operations template
3742  */
3743 vnodeops_t *zfs_symvnodeops;
3744 const fs_operation_def_t zfs_symvnodeops_template[] = {
3745 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
3746 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
3747 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
3748 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
3749 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
3750 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
3751 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
3752 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
3753 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
3754 	NULL,			NULL
3755 };
3756 
3757 /*
3758  * Extended attribute directory vnode operations template
3759  *	This template is identical to the directory vnodes
3760  *	operation template except for restricted operations:
3761  *		VOP_MKDIR()
3762  *		VOP_SYMLINK()
3763  * Note that there are other restrictions embedded in:
3764  *	zfs_create()	- restrict type to VREG
3765  *	zfs_link()	- no links into/out of attribute space
3766  *	zfs_rename()	- no moves into/out of attribute space
3767  */
3768 vnodeops_t *zfs_xdvnodeops;
3769 const fs_operation_def_t zfs_xdvnodeops_template[] = {
3770 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
3771 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
3772 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
3773 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
3774 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
3775 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
3776 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
3777 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
3778 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
3779 	VOPNAME_LINK,		{ .vop_link = zfs_link },
3780 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
3781 	VOPNAME_MKDIR,		{ .error = zfs_inval },
3782 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
3783 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
3784 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
3785 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
3786 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
3787 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
3788 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
3789 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
3790 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
3791 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
3792 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
3793 	NULL,			NULL
3794 };
3795 
3796 /*
3797  * Error vnode operations template
3798  */
3799 vnodeops_t *zfs_evnodeops;
3800 const fs_operation_def_t zfs_evnodeops_template[] = {
3801 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
3802 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
3803 	NULL,			NULL
3804 };
3805