xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision d840867f3a8b0ba209ef90762b3f9c72a5f92cc5)
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 static int
1792 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr)
1793 {
1794 	znode_t	*zp = VTOZ(vp);
1795 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1796 
1797 	/*
1798 	 * Regardless of whether this is required for standards conformance,
1799 	 * this is the logical behavior when fsync() is called on a file with
1800 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
1801 	 * going to be pushed out as part of the zil_commit().
1802 	 */
1803 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
1804 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
1805 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr);
1806 
1807 	ZFS_ENTER(zfsvfs);
1808 	zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
1809 	ZFS_EXIT(zfsvfs);
1810 	return (0);
1811 }
1812 
1813 /*
1814  * Get the requested file attributes and place them in the provided
1815  * vattr structure.
1816  *
1817  *	IN:	vp	- vnode of file.
1818  *		vap	- va_mask identifies requested attributes.
1819  *		flags	- [UNUSED]
1820  *		cr	- credentials of caller.
1821  *
1822  *	OUT:	vap	- attribute values.
1823  *
1824  *	RETURN:	0 (always succeeds)
1825  */
1826 /* ARGSUSED */
1827 static int
1828 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr)
1829 {
1830 	znode_t *zp = VTOZ(vp);
1831 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1832 	znode_phys_t *pzp = zp->z_phys;
1833 	int	error;
1834 	uint64_t links;
1835 
1836 	ZFS_ENTER(zfsvfs);
1837 
1838 	/*
1839 	 * Return all attributes.  It's cheaper to provide the answer
1840 	 * than to determine whether we were asked the question.
1841 	 */
1842 	mutex_enter(&zp->z_lock);
1843 
1844 	vap->va_type = vp->v_type;
1845 	vap->va_mode = pzp->zp_mode & MODEMASK;
1846 	vap->va_uid = zp->z_phys->zp_uid;
1847 	vap->va_gid = zp->z_phys->zp_gid;
1848 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
1849 	vap->va_nodeid = zp->z_id;
1850 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
1851 		links = pzp->zp_links + 1;
1852 	else
1853 		links = pzp->zp_links;
1854 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
1855 	vap->va_size = pzp->zp_size;
1856 	vap->va_rdev = vp->v_rdev;
1857 	vap->va_seq = zp->z_seq;
1858 
1859 	ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime);
1860 	ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime);
1861 	ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime);
1862 
1863 	/*
1864 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
1865 	 * Also, if we are the owner don't bother, since owner should
1866 	 * always be allowed to read basic attributes of file.
1867 	 */
1868 	if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) &&
1869 	    (zp->z_phys->zp_uid != crgetuid(cr))) {
1870 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) {
1871 			mutex_exit(&zp->z_lock);
1872 			ZFS_EXIT(zfsvfs);
1873 			return (error);
1874 		}
1875 	}
1876 
1877 	mutex_exit(&zp->z_lock);
1878 
1879 	dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks);
1880 
1881 	if (zp->z_blksz == 0) {
1882 		/*
1883 		 * Block size hasn't been set; suggest maximal I/O transfers.
1884 		 */
1885 		vap->va_blksize = zfsvfs->z_max_blksz;
1886 	}
1887 
1888 	ZFS_EXIT(zfsvfs);
1889 	return (0);
1890 }
1891 
1892 /*
1893  * Set the file attributes to the values contained in the
1894  * vattr structure.
1895  *
1896  *	IN:	vp	- vnode of file to be modified.
1897  *		vap	- new attribute values.
1898  *		flags	- ATTR_UTIME set if non-default time values provided.
1899  *		cr	- credentials of caller.
1900  *
1901  *	RETURN:	0 if success
1902  *		error code if failure
1903  *
1904  * Timestamps:
1905  *	vp - ctime updated, mtime updated if size changed.
1906  */
1907 /* ARGSUSED */
1908 static int
1909 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1910 	caller_context_t *ct)
1911 {
1912 	struct znode	*zp = VTOZ(vp);
1913 	znode_phys_t	*pzp = zp->z_phys;
1914 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1915 	zilog_t		*zilog = zfsvfs->z_log;
1916 	dmu_tx_t	*tx;
1917 	vattr_t		oldva;
1918 	uint_t		mask = vap->va_mask;
1919 	uint_t		saved_mask;
1920 	int		trim_mask = 0;
1921 	uint64_t	new_mode;
1922 	znode_t		*attrzp;
1923 	int		need_policy = FALSE;
1924 	int		err;
1925 
1926 	if (mask == 0)
1927 		return (0);
1928 
1929 	if (mask & AT_NOSET)
1930 		return (EINVAL);
1931 
1932 	if (mask & AT_SIZE && vp->v_type == VDIR)
1933 		return (EISDIR);
1934 
1935 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO)
1936 		return (EINVAL);
1937 
1938 	ZFS_ENTER(zfsvfs);
1939 
1940 top:
1941 	attrzp = NULL;
1942 
1943 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
1944 		ZFS_EXIT(zfsvfs);
1945 		return (EROFS);
1946 	}
1947 
1948 	/*
1949 	 * First validate permissions
1950 	 */
1951 
1952 	if (mask & AT_SIZE) {
1953 		err = zfs_zaccess(zp, ACE_WRITE_DATA, cr);
1954 		if (err) {
1955 			ZFS_EXIT(zfsvfs);
1956 			return (err);
1957 		}
1958 		/*
1959 		 * XXX - Note, we are not providing any open
1960 		 * mode flags here (like FNDELAY), so we may
1961 		 * block if there are locks present... this
1962 		 * should be addressed in openat().
1963 		 */
1964 		do {
1965 			err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
1966 			/* NB: we already did dmu_tx_wait() if necessary */
1967 		} while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
1968 		if (err) {
1969 			ZFS_EXIT(zfsvfs);
1970 			return (err);
1971 		}
1972 	}
1973 
1974 	if (mask & (AT_ATIME|AT_MTIME))
1975 		need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr);
1976 
1977 	if (mask & (AT_UID|AT_GID)) {
1978 		int	idmask = (mask & (AT_UID|AT_GID));
1979 		int	take_owner;
1980 		int	take_group;
1981 
1982 		/*
1983 		 * NOTE: even if a new mode is being set,
1984 		 * we may clear S_ISUID/S_ISGID bits.
1985 		 */
1986 
1987 		if (!(mask & AT_MODE))
1988 			vap->va_mode = pzp->zp_mode;
1989 
1990 		/*
1991 		 * Take ownership or chgrp to group we are a member of
1992 		 */
1993 
1994 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
1995 		take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr);
1996 
1997 		/*
1998 		 * If both AT_UID and AT_GID are set then take_owner and
1999 		 * take_group must both be set in order to allow taking
2000 		 * ownership.
2001 		 *
2002 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2003 		 *
2004 		 */
2005 
2006 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2007 		    ((idmask == AT_UID) && take_owner) ||
2008 		    ((idmask == AT_GID) && take_group)) {
2009 			if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) {
2010 				/*
2011 				 * Remove setuid/setgid for non-privileged users
2012 				 */
2013 				secpolicy_setid_clear(vap, cr);
2014 				trim_mask = (mask & (AT_UID|AT_GID));
2015 			} else {
2016 				need_policy =  TRUE;
2017 			}
2018 		} else {
2019 			need_policy =  TRUE;
2020 		}
2021 	}
2022 
2023 	mutex_enter(&zp->z_lock);
2024 	oldva.va_mode = pzp->zp_mode;
2025 	oldva.va_uid = zp->z_phys->zp_uid;
2026 	oldva.va_gid = zp->z_phys->zp_gid;
2027 	mutex_exit(&zp->z_lock);
2028 
2029 	if (mask & AT_MODE) {
2030 		if (zfs_zaccess_v4_perm(zp, ACE_WRITE_ACL, cr) == 0) {
2031 			err = secpolicy_setid_setsticky_clear(vp, vap,
2032 			    &oldva, cr);
2033 			if (err) {
2034 				ZFS_EXIT(zfsvfs);
2035 				return (err);
2036 			}
2037 			trim_mask |= AT_MODE;
2038 		} else {
2039 			need_policy = TRUE;
2040 		}
2041 	}
2042 
2043 	if (need_policy) {
2044 		/*
2045 		 * If trim_mask is set then take ownership
2046 		 * has been granted or write_acl is present and user
2047 		 * has the ability to modify mode.  In that case remove
2048 		 * UID|GID and or MODE from mask so that
2049 		 * secpolicy_vnode_setattr() doesn't revoke it.
2050 		 */
2051 
2052 		if (trim_mask) {
2053 			saved_mask = vap->va_mask;
2054 			vap->va_mask &= ~trim_mask;
2055 
2056 		}
2057 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2058 		    (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp);
2059 		if (err) {
2060 			ZFS_EXIT(zfsvfs);
2061 			return (err);
2062 		}
2063 
2064 		if (trim_mask)
2065 			vap->va_mask |= saved_mask;
2066 	}
2067 
2068 	/*
2069 	 * secpolicy_vnode_setattr, or take ownership may have
2070 	 * changed va_mask
2071 	 */
2072 	mask = vap->va_mask;
2073 
2074 	tx = dmu_tx_create(zfsvfs->z_os);
2075 	dmu_tx_hold_bonus(tx, zp->z_id);
2076 
2077 	if (mask & AT_MODE) {
2078 		uint64_t pmode = pzp->zp_mode;
2079 
2080 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2081 
2082 		if (zp->z_phys->zp_acl.z_acl_extern_obj)
2083 			dmu_tx_hold_write(tx,
2084 			    pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE);
2085 		else
2086 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2087 			    0, ZFS_ACL_SIZE(MAX_ACL_SIZE));
2088 	}
2089 
2090 	if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) {
2091 		err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp);
2092 		if (err) {
2093 			dmu_tx_abort(tx);
2094 			ZFS_EXIT(zfsvfs);
2095 			return (err);
2096 		}
2097 		dmu_tx_hold_bonus(tx, attrzp->z_id);
2098 	}
2099 
2100 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2101 	if (err) {
2102 		if (attrzp)
2103 			VN_RELE(ZTOV(attrzp));
2104 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2105 			dmu_tx_wait(tx);
2106 			dmu_tx_abort(tx);
2107 			goto top;
2108 		}
2109 		dmu_tx_abort(tx);
2110 		ZFS_EXIT(zfsvfs);
2111 		return (err);
2112 	}
2113 
2114 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2115 
2116 	/*
2117 	 * Set each attribute requested.
2118 	 * We group settings according to the locks they need to acquire.
2119 	 *
2120 	 * Note: you cannot set ctime directly, although it will be
2121 	 * updated as a side-effect of calling this function.
2122 	 */
2123 
2124 	mutex_enter(&zp->z_lock);
2125 
2126 	if (mask & AT_MODE) {
2127 		err = zfs_acl_chmod_setattr(zp, new_mode, tx);
2128 		ASSERT3U(err, ==, 0);
2129 	}
2130 
2131 	if (attrzp)
2132 		mutex_enter(&attrzp->z_lock);
2133 
2134 	if (mask & AT_UID) {
2135 		zp->z_phys->zp_uid = (uint64_t)vap->va_uid;
2136 		if (attrzp) {
2137 			attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid;
2138 		}
2139 	}
2140 
2141 	if (mask & AT_GID) {
2142 		zp->z_phys->zp_gid = (uint64_t)vap->va_gid;
2143 		if (attrzp)
2144 			attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid;
2145 	}
2146 
2147 	if (attrzp)
2148 		mutex_exit(&attrzp->z_lock);
2149 
2150 	if (mask & AT_ATIME)
2151 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2152 
2153 	if (mask & AT_MTIME)
2154 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2155 
2156 	if (mask & AT_SIZE)
2157 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2158 	else if (mask != 0)
2159 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2160 
2161 	if (mask != 0)
2162 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask);
2163 
2164 	mutex_exit(&zp->z_lock);
2165 
2166 	if (attrzp)
2167 		VN_RELE(ZTOV(attrzp));
2168 
2169 	dmu_tx_commit(tx);
2170 
2171 	ZFS_EXIT(zfsvfs);
2172 	return (err);
2173 }
2174 
2175 typedef struct zfs_zlock {
2176 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2177 	znode_t		*zl_znode;	/* znode we held */
2178 	struct zfs_zlock *zl_next;	/* next in list */
2179 } zfs_zlock_t;
2180 
2181 /*
2182  * Drop locks and release vnodes that were held by zfs_rename_lock().
2183  */
2184 static void
2185 zfs_rename_unlock(zfs_zlock_t **zlpp)
2186 {
2187 	zfs_zlock_t *zl;
2188 
2189 	while ((zl = *zlpp) != NULL) {
2190 		if (zl->zl_znode != NULL)
2191 			VN_RELE(ZTOV(zl->zl_znode));
2192 		rw_exit(zl->zl_rwlock);
2193 		*zlpp = zl->zl_next;
2194 		kmem_free(zl, sizeof (*zl));
2195 	}
2196 }
2197 
2198 /*
2199  * Search back through the directory tree, using the ".." entries.
2200  * Lock each directory in the chain to prevent concurrent renames.
2201  * Fail any attempt to move a directory into one of its own descendants.
2202  * XXX - z_parent_lock can overlap with map or grow locks
2203  */
2204 static int
2205 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2206 {
2207 	zfs_zlock_t	*zl;
2208 	znode_t		*zp = tdzp;
2209 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2210 	uint64_t	*oidp = &zp->z_id;
2211 	krwlock_t	*rwlp = &szp->z_parent_lock;
2212 	krw_t		rw = RW_WRITER;
2213 
2214 	/*
2215 	 * First pass write-locks szp and compares to zp->z_id.
2216 	 * Later passes read-lock zp and compare to zp->z_parent.
2217 	 */
2218 	do {
2219 		if (!rw_tryenter(rwlp, rw)) {
2220 			/*
2221 			 * Another thread is renaming in this path.
2222 			 * Note that if we are a WRITER, we don't have any
2223 			 * parent_locks held yet.
2224 			 */
2225 			if (rw == RW_READER && zp->z_id > szp->z_id) {
2226 				/*
2227 				 * Drop our locks and restart
2228 				 */
2229 				zfs_rename_unlock(&zl);
2230 				*zlpp = NULL;
2231 				zp = tdzp;
2232 				oidp = &zp->z_id;
2233 				rwlp = &szp->z_parent_lock;
2234 				rw = RW_WRITER;
2235 				continue;
2236 			} else {
2237 				/*
2238 				 * Wait for other thread to drop its locks
2239 				 */
2240 				rw_enter(rwlp, rw);
2241 			}
2242 		}
2243 
2244 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2245 		zl->zl_rwlock = rwlp;
2246 		zl->zl_znode = NULL;
2247 		zl->zl_next = *zlpp;
2248 		*zlpp = zl;
2249 
2250 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2251 			return (EINVAL);
2252 
2253 		if (*oidp == rootid)		/* We've hit the top */
2254 			return (0);
2255 
2256 		if (rw == RW_READER) {		/* i.e. not the first pass */
2257 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2258 			if (error)
2259 				return (error);
2260 			zl->zl_znode = zp;
2261 		}
2262 		oidp = &zp->z_phys->zp_parent;
2263 		rwlp = &zp->z_parent_lock;
2264 		rw = RW_READER;
2265 
2266 	} while (zp->z_id != sdzp->z_id);
2267 
2268 	return (0);
2269 }
2270 
2271 /*
2272  * Move an entry from the provided source directory to the target
2273  * directory.  Change the entry name as indicated.
2274  *
2275  *	IN:	sdvp	- Source directory containing the "old entry".
2276  *		snm	- Old entry name.
2277  *		tdvp	- Target directory to contain the "new entry".
2278  *		tnm	- New entry name.
2279  *		cr	- credentials of caller.
2280  *
2281  *	RETURN:	0 if success
2282  *		error code if failure
2283  *
2284  * Timestamps:
2285  *	sdvp,tdvp - ctime|mtime updated
2286  */
2287 static int
2288 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr)
2289 {
2290 	znode_t		*tdzp, *szp, *tzp;
2291 	znode_t		*sdzp = VTOZ(sdvp);
2292 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2293 	zilog_t		*zilog = zfsvfs->z_log;
2294 	vnode_t		*realvp;
2295 	zfs_dirlock_t	*sdl, *tdl;
2296 	dmu_tx_t	*tx;
2297 	zfs_zlock_t	*zl;
2298 	int		cmp, serr, terr, error;
2299 
2300 	ZFS_ENTER(zfsvfs);
2301 
2302 	/*
2303 	 * Make sure we have the real vp for the target directory.
2304 	 */
2305 	if (VOP_REALVP(tdvp, &realvp) == 0)
2306 		tdvp = realvp;
2307 
2308 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
2309 		ZFS_EXIT(zfsvfs);
2310 		return (EXDEV);
2311 	}
2312 
2313 	tdzp = VTOZ(tdvp);
2314 top:
2315 	szp = NULL;
2316 	tzp = NULL;
2317 	zl = NULL;
2318 
2319 	/*
2320 	 * This is to prevent the creation of links into attribute space
2321 	 * by renaming a linked file into/outof an attribute directory.
2322 	 * See the comment in zfs_link() for why this is considered bad.
2323 	 */
2324 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
2325 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
2326 		ZFS_EXIT(zfsvfs);
2327 		return (EINVAL);
2328 	}
2329 
2330 	/*
2331 	 * Lock source and target directory entries.  To prevent deadlock,
2332 	 * a lock ordering must be defined.  We lock the directory with
2333 	 * the smallest object id first, or if it's a tie, the one with
2334 	 * the lexically first name.
2335 	 */
2336 	if (sdzp->z_id < tdzp->z_id) {
2337 		cmp = -1;
2338 	} else if (sdzp->z_id > tdzp->z_id) {
2339 		cmp = 1;
2340 	} else {
2341 		cmp = strcmp(snm, tnm);
2342 		if (cmp == 0) {
2343 			/*
2344 			 * POSIX: "If the old argument and the new argument
2345 			 * both refer to links to the same existing file,
2346 			 * the rename() function shall return successfully
2347 			 * and perform no other action."
2348 			 */
2349 			ZFS_EXIT(zfsvfs);
2350 			return (0);
2351 		}
2352 	}
2353 	if (cmp < 0) {
2354 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS);
2355 		terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0);
2356 	} else {
2357 		terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0);
2358 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS);
2359 	}
2360 
2361 	if (serr) {
2362 		/*
2363 		 * Source entry invalid or not there.
2364 		 */
2365 		if (!terr) {
2366 			zfs_dirent_unlock(tdl);
2367 			if (tzp)
2368 				VN_RELE(ZTOV(tzp));
2369 		}
2370 		if (strcmp(snm, "..") == 0)
2371 			serr = EINVAL;
2372 		ZFS_EXIT(zfsvfs);
2373 		return (serr);
2374 	}
2375 	if (terr) {
2376 		zfs_dirent_unlock(sdl);
2377 		VN_RELE(ZTOV(szp));
2378 		if (strcmp(tnm, "..") == 0)
2379 			terr = EINVAL;
2380 		ZFS_EXIT(zfsvfs);
2381 		return (terr);
2382 	}
2383 
2384 	/*
2385 	 * Must have write access at the source to remove the old entry
2386 	 * and write access at the target to create the new entry.
2387 	 * Note that if target and source are the same, this can be
2388 	 * done in a single check.
2389 	 */
2390 
2391 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
2392 		goto out;
2393 
2394 	if (ZTOV(szp)->v_type == VDIR) {
2395 		/*
2396 		 * Check to make sure rename is valid.
2397 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
2398 		 */
2399 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
2400 			goto out;
2401 	}
2402 
2403 	/*
2404 	 * Does target exist?
2405 	 */
2406 	if (tzp) {
2407 		/*
2408 		 * Source and target must be the same type.
2409 		 */
2410 		if (ZTOV(szp)->v_type == VDIR) {
2411 			if (ZTOV(tzp)->v_type != VDIR) {
2412 				error = ENOTDIR;
2413 				goto out;
2414 			}
2415 		} else {
2416 			if (ZTOV(tzp)->v_type == VDIR) {
2417 				error = EISDIR;
2418 				goto out;
2419 			}
2420 		}
2421 		/*
2422 		 * POSIX dictates that when the source and target
2423 		 * entries refer to the same file object, rename
2424 		 * must do nothing and exit without error.
2425 		 */
2426 		if (szp->z_id == tzp->z_id) {
2427 			error = 0;
2428 			goto out;
2429 		}
2430 	}
2431 
2432 	vnevent_rename_src(ZTOV(szp));
2433 	if (tzp)
2434 		vnevent_rename_dest(ZTOV(tzp));
2435 
2436 	tx = dmu_tx_create(zfsvfs->z_os);
2437 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
2438 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
2439 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
2440 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
2441 	if (sdzp != tdzp)
2442 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
2443 	if (tzp)
2444 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* parent changes */
2445 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2446 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2447 	if (error) {
2448 		if (zl != NULL)
2449 			zfs_rename_unlock(&zl);
2450 		zfs_dirent_unlock(sdl);
2451 		zfs_dirent_unlock(tdl);
2452 		VN_RELE(ZTOV(szp));
2453 		if (tzp)
2454 			VN_RELE(ZTOV(tzp));
2455 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2456 			dmu_tx_wait(tx);
2457 			dmu_tx_abort(tx);
2458 			goto top;
2459 		}
2460 		dmu_tx_abort(tx);
2461 		ZFS_EXIT(zfsvfs);
2462 		return (error);
2463 	}
2464 
2465 	if (tzp)	/* Attempt to remove the existing target */
2466 		error = zfs_link_destroy(tdl, tzp, tx, 0, NULL);
2467 
2468 	if (error == 0) {
2469 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
2470 		if (error == 0) {
2471 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
2472 			ASSERT(error == 0);
2473 			zfs_log_rename(zilog, tx, TX_RENAME, sdzp,
2474 			    sdl->dl_name, tdzp, tdl->dl_name, szp);
2475 		}
2476 	}
2477 
2478 	dmu_tx_commit(tx);
2479 out:
2480 	if (zl != NULL)
2481 		zfs_rename_unlock(&zl);
2482 
2483 	zfs_dirent_unlock(sdl);
2484 	zfs_dirent_unlock(tdl);
2485 
2486 	VN_RELE(ZTOV(szp));
2487 	if (tzp)
2488 		VN_RELE(ZTOV(tzp));
2489 
2490 	ZFS_EXIT(zfsvfs);
2491 	return (error);
2492 }
2493 
2494 /*
2495  * Insert the indicated symbolic reference entry into the directory.
2496  *
2497  *	IN:	dvp	- Directory to contain new symbolic link.
2498  *		link	- Name for new symlink entry.
2499  *		vap	- Attributes of new entry.
2500  *		target	- Target path of new symlink.
2501  *		cr	- credentials of caller.
2502  *
2503  *	RETURN:	0 if success
2504  *		error code if failure
2505  *
2506  * Timestamps:
2507  *	dvp - ctime|mtime updated
2508  */
2509 static int
2510 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr)
2511 {
2512 	znode_t		*zp, *dzp = VTOZ(dvp);
2513 	zfs_dirlock_t	*dl;
2514 	dmu_tx_t	*tx;
2515 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2516 	zilog_t		*zilog = zfsvfs->z_log;
2517 	uint64_t	zoid;
2518 	int		len = strlen(link);
2519 	int		error;
2520 
2521 	ASSERT(vap->va_type == VLNK);
2522 
2523 	ZFS_ENTER(zfsvfs);
2524 top:
2525 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
2526 		ZFS_EXIT(zfsvfs);
2527 		return (error);
2528 	}
2529 
2530 	if (len > MAXPATHLEN) {
2531 		ZFS_EXIT(zfsvfs);
2532 		return (ENAMETOOLONG);
2533 	}
2534 
2535 	/*
2536 	 * Attempt to lock directory; fail if entry already exists.
2537 	 */
2538 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) {
2539 		ZFS_EXIT(zfsvfs);
2540 		return (error);
2541 	}
2542 
2543 	tx = dmu_tx_create(zfsvfs->z_os);
2544 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
2545 	dmu_tx_hold_bonus(tx, dzp->z_id);
2546 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
2547 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
2548 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
2549 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2550 	if (error) {
2551 		zfs_dirent_unlock(dl);
2552 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2553 			dmu_tx_wait(tx);
2554 			dmu_tx_abort(tx);
2555 			goto top;
2556 		}
2557 		dmu_tx_abort(tx);
2558 		ZFS_EXIT(zfsvfs);
2559 		return (error);
2560 	}
2561 
2562 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
2563 
2564 	/*
2565 	 * Create a new object for the symlink.
2566 	 * Put the link content into bonus buffer if it will fit;
2567 	 * otherwise, store it just like any other file data.
2568 	 */
2569 	zoid = 0;
2570 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
2571 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len);
2572 		if (len != 0)
2573 			bcopy(link, zp->z_phys + 1, len);
2574 	} else {
2575 		dmu_buf_t *dbp;
2576 
2577 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
2578 
2579 		/*
2580 		 * Nothing can access the znode yet so no locking needed
2581 		 * for growing the znode's blocksize.
2582 		 */
2583 		zfs_grow_blocksize(zp, len, tx);
2584 
2585 		VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp));
2586 		dmu_buf_will_dirty(dbp, tx);
2587 
2588 		ASSERT3U(len, <=, dbp->db_size);
2589 		bcopy(link, dbp->db_data, len);
2590 		dmu_buf_rele(dbp, FTAG);
2591 	}
2592 	zp->z_phys->zp_size = len;
2593 
2594 	/*
2595 	 * Insert the new object into the directory.
2596 	 */
2597 	(void) zfs_link_create(dl, zp, tx, ZNEW);
2598 out:
2599 	if (error == 0)
2600 		zfs_log_symlink(zilog, tx, TX_SYMLINK, dzp, zp, name, link);
2601 
2602 	dmu_tx_commit(tx);
2603 
2604 	zfs_dirent_unlock(dl);
2605 
2606 	VN_RELE(ZTOV(zp));
2607 
2608 	ZFS_EXIT(zfsvfs);
2609 	return (error);
2610 }
2611 
2612 /*
2613  * Return, in the buffer contained in the provided uio structure,
2614  * the symbolic path referred to by vp.
2615  *
2616  *	IN:	vp	- vnode of symbolic link.
2617  *		uoip	- structure to contain the link path.
2618  *		cr	- credentials of caller.
2619  *
2620  *	OUT:	uio	- structure to contain the link path.
2621  *
2622  *	RETURN:	0 if success
2623  *		error code if failure
2624  *
2625  * Timestamps:
2626  *	vp - atime updated
2627  */
2628 /* ARGSUSED */
2629 static int
2630 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr)
2631 {
2632 	znode_t		*zp = VTOZ(vp);
2633 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2634 	size_t		bufsz;
2635 	int		error;
2636 
2637 	ZFS_ENTER(zfsvfs);
2638 
2639 	bufsz = (size_t)zp->z_phys->zp_size;
2640 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
2641 		error = uiomove(zp->z_phys + 1,
2642 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
2643 	} else {
2644 		dmu_buf_t *dbp;
2645 		error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp);
2646 		if (error) {
2647 			ZFS_EXIT(zfsvfs);
2648 			return (error);
2649 		}
2650 		error = uiomove(dbp->db_data,
2651 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
2652 		dmu_buf_rele(dbp, FTAG);
2653 	}
2654 
2655 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2656 	ZFS_EXIT(zfsvfs);
2657 	return (error);
2658 }
2659 
2660 /*
2661  * Insert a new entry into directory tdvp referencing svp.
2662  *
2663  *	IN:	tdvp	- Directory to contain new entry.
2664  *		svp	- vnode of new entry.
2665  *		name	- name of new entry.
2666  *		cr	- credentials of caller.
2667  *
2668  *	RETURN:	0 if success
2669  *		error code if failure
2670  *
2671  * Timestamps:
2672  *	tdvp - ctime|mtime updated
2673  *	 svp - ctime updated
2674  */
2675 /* ARGSUSED */
2676 static int
2677 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr)
2678 {
2679 	znode_t		*dzp = VTOZ(tdvp);
2680 	znode_t		*tzp, *szp;
2681 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2682 	zilog_t		*zilog = zfsvfs->z_log;
2683 	zfs_dirlock_t	*dl;
2684 	dmu_tx_t	*tx;
2685 	vnode_t		*realvp;
2686 	int		error;
2687 
2688 	ASSERT(tdvp->v_type == VDIR);
2689 
2690 	ZFS_ENTER(zfsvfs);
2691 
2692 	if (VOP_REALVP(svp, &realvp) == 0)
2693 		svp = realvp;
2694 
2695 	if (svp->v_vfsp != tdvp->v_vfsp) {
2696 		ZFS_EXIT(zfsvfs);
2697 		return (EXDEV);
2698 	}
2699 
2700 	szp = VTOZ(svp);
2701 top:
2702 	/*
2703 	 * We do not support links between attributes and non-attributes
2704 	 * because of the potential security risk of creating links
2705 	 * into "normal" file space in order to circumvent restrictions
2706 	 * imposed in attribute space.
2707 	 */
2708 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
2709 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
2710 		ZFS_EXIT(zfsvfs);
2711 		return (EINVAL);
2712 	}
2713 
2714 	/*
2715 	 * POSIX dictates that we return EPERM here.
2716 	 * Better choices include ENOTSUP or EISDIR.
2717 	 */
2718 	if (svp->v_type == VDIR) {
2719 		ZFS_EXIT(zfsvfs);
2720 		return (EPERM);
2721 	}
2722 
2723 	if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) &&
2724 	    secpolicy_basic_link(cr) != 0) {
2725 		ZFS_EXIT(zfsvfs);
2726 		return (EPERM);
2727 	}
2728 
2729 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
2730 		ZFS_EXIT(zfsvfs);
2731 		return (error);
2732 	}
2733 
2734 	/*
2735 	 * Attempt to lock directory; fail if entry already exists.
2736 	 */
2737 	if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) {
2738 		ZFS_EXIT(zfsvfs);
2739 		return (error);
2740 	}
2741 
2742 	tx = dmu_tx_create(zfsvfs->z_os);
2743 	dmu_tx_hold_bonus(tx, szp->z_id);
2744 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
2745 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2746 	if (error) {
2747 		zfs_dirent_unlock(dl);
2748 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2749 			dmu_tx_wait(tx);
2750 			dmu_tx_abort(tx);
2751 			goto top;
2752 		}
2753 		dmu_tx_abort(tx);
2754 		ZFS_EXIT(zfsvfs);
2755 		return (error);
2756 	}
2757 
2758 	error = zfs_link_create(dl, szp, tx, 0);
2759 
2760 	if (error == 0)
2761 		zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name);
2762 
2763 	dmu_tx_commit(tx);
2764 
2765 	zfs_dirent_unlock(dl);
2766 
2767 	ZFS_EXIT(zfsvfs);
2768 	return (error);
2769 }
2770 
2771 /*
2772  * zfs_null_putapage() is used when the file system has been force
2773  * unmounted. It just drops the pages.
2774  */
2775 /* ARGSUSED */
2776 static int
2777 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
2778 		size_t *lenp, int flags, cred_t *cr)
2779 {
2780 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
2781 	return (0);
2782 }
2783 
2784 /*
2785  * Push a page out to disk, klustering if possible.
2786  *
2787  *	IN:	vp	- file to push page to.
2788  *		pp	- page to push.
2789  *		flags	- additional flags.
2790  *		cr	- credentials of caller.
2791  *
2792  *	OUT:	offp	- start of range pushed.
2793  *		lenp	- len of range pushed.
2794  *
2795  *	RETURN:	0 if success
2796  *		error code if failure
2797  *
2798  * NOTE: callers must have locked the page to be pushed.  On
2799  * exit, the page (and all other pages in the kluster) must be
2800  * unlocked.
2801  */
2802 /* ARGSUSED */
2803 static int
2804 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
2805 		size_t *lenp, int flags, cred_t *cr)
2806 {
2807 	znode_t		*zp = VTOZ(vp);
2808 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2809 	zilog_t		*zilog = zfsvfs->z_log;
2810 	dmu_tx_t	*tx;
2811 	rl_t		*rl;
2812 	u_offset_t	off, koff;
2813 	size_t		len, klen;
2814 	uint64_t	filesz;
2815 	int		err;
2816 
2817 	filesz = zp->z_phys->zp_size;
2818 	off = pp->p_offset;
2819 	len = PAGESIZE;
2820 	/*
2821 	 * If our blocksize is bigger than the page size, try to kluster
2822 	 * muiltiple pages so that we write a full block (thus avoiding
2823 	 * a read-modify-write).
2824 	 */
2825 	if (off < filesz && zp->z_blksz > PAGESIZE) {
2826 		if (!ISP2(zp->z_blksz)) {
2827 			/* Only one block in the file. */
2828 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
2829 			koff = 0;
2830 		} else {
2831 			klen = zp->z_blksz;
2832 			koff = P2ALIGN(off, (u_offset_t)klen);
2833 		}
2834 		ASSERT(koff <= filesz);
2835 		if (koff + klen > filesz)
2836 			klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE);
2837 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
2838 	}
2839 	ASSERT3U(btop(len), ==, btopr(len));
2840 top:
2841 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
2842 	/*
2843 	 * Can't push pages past end-of-file.
2844 	 */
2845 	filesz = zp->z_phys->zp_size;
2846 	if (off >= filesz) {
2847 		/* ignore all pages */
2848 		err = 0;
2849 		goto out;
2850 	} else if (off + len > filesz) {
2851 		int npages = btopr(filesz - off);
2852 		page_t *trunc;
2853 
2854 		page_list_break(&pp, &trunc, npages);
2855 		/* ignore pages past end of file */
2856 		if (trunc)
2857 			pvn_write_done(trunc, flags);
2858 		len = filesz - off;
2859 	}
2860 
2861 	tx = dmu_tx_create(zfsvfs->z_os);
2862 	dmu_tx_hold_write(tx, zp->z_id, off, len);
2863 	dmu_tx_hold_bonus(tx, zp->z_id);
2864 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2865 	if (err != 0) {
2866 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2867 			zfs_range_unlock(rl);
2868 			dmu_tx_wait(tx);
2869 			dmu_tx_abort(tx);
2870 			err = 0;
2871 			goto top;
2872 		}
2873 		dmu_tx_abort(tx);
2874 		goto out;
2875 	}
2876 
2877 	if (zp->z_blksz <= PAGESIZE) {
2878 		caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1);
2879 		ASSERT3U(len, <=, PAGESIZE);
2880 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
2881 		ppmapout(va);
2882 	} else {
2883 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
2884 	}
2885 
2886 	if (err == 0) {
2887 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
2888 		zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0);
2889 		dmu_tx_commit(tx);
2890 	}
2891 
2892 out:
2893 	zfs_range_unlock(rl);
2894 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
2895 	if (offp)
2896 		*offp = off;
2897 	if (lenp)
2898 		*lenp = len;
2899 
2900 	return (err);
2901 }
2902 
2903 /*
2904  * Copy the portion of the file indicated from pages into the file.
2905  * The pages are stored in a page list attached to the files vnode.
2906  *
2907  *	IN:	vp	- vnode of file to push page data to.
2908  *		off	- position in file to put data.
2909  *		len	- amount of data to write.
2910  *		flags	- flags to control the operation.
2911  *		cr	- credentials of caller.
2912  *
2913  *	RETURN:	0 if success
2914  *		error code if failure
2915  *
2916  * Timestamps:
2917  *	vp - ctime|mtime updated
2918  */
2919 static int
2920 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr)
2921 {
2922 	znode_t		*zp = VTOZ(vp);
2923 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2924 	page_t		*pp;
2925 	size_t		io_len;
2926 	u_offset_t	io_off;
2927 	uint64_t	filesz;
2928 	int		error = 0;
2929 
2930 	ZFS_ENTER(zfsvfs);
2931 
2932 	ASSERT(zp->z_dbuf_held && zp->z_phys);
2933 
2934 	if (len == 0) {
2935 		/*
2936 		 * Search the entire vp list for pages >= off.
2937 		 */
2938 		error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage,
2939 		    flags, cr);
2940 		goto out;
2941 	}
2942 
2943 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
2944 	if (off > filesz) {
2945 		/* past end of file */
2946 		ZFS_EXIT(zfsvfs);
2947 		return (0);
2948 	}
2949 
2950 	len = MIN(len, filesz - off);
2951 
2952 	for (io_off = off; io_off < off + len; io_off += io_len) {
2953 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
2954 			pp = page_lookup(vp, io_off,
2955 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
2956 		} else {
2957 			pp = page_lookup_nowait(vp, io_off,
2958 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
2959 		}
2960 
2961 		if (pp != NULL && pvn_getdirty(pp, flags)) {
2962 			int err;
2963 
2964 			/*
2965 			 * Found a dirty page to push
2966 			 */
2967 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
2968 			if (err)
2969 				error = err;
2970 		} else {
2971 			io_len = PAGESIZE;
2972 		}
2973 	}
2974 out:
2975 	if ((flags & B_ASYNC) == 0)
2976 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
2977 	ZFS_EXIT(zfsvfs);
2978 	return (error);
2979 }
2980 
2981 void
2982 zfs_inactive(vnode_t *vp, cred_t *cr)
2983 {
2984 	znode_t	*zp = VTOZ(vp);
2985 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2986 	int error;
2987 
2988 	rw_enter(&zfsvfs->z_um_lock, RW_READER);
2989 	if (zfsvfs->z_unmounted2) {
2990 		ASSERT(zp->z_dbuf_held == 0);
2991 
2992 		if (vn_has_cached_data(vp)) {
2993 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
2994 			    B_INVAL, cr);
2995 		}
2996 
2997 		mutex_enter(&zp->z_lock);
2998 		vp->v_count = 0; /* count arrives as 1 */
2999 		if (zp->z_dbuf == NULL) {
3000 			mutex_exit(&zp->z_lock);
3001 			zfs_znode_free(zp);
3002 		} else {
3003 			mutex_exit(&zp->z_lock);
3004 		}
3005 		rw_exit(&zfsvfs->z_um_lock);
3006 		VFS_RELE(zfsvfs->z_vfs);
3007 		return;
3008 	}
3009 
3010 	/*
3011 	 * Attempt to push any data in the page cache.  If this fails
3012 	 * we will get kicked out later in zfs_zinactive().
3013 	 */
3014 	if (vn_has_cached_data(vp)) {
3015 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
3016 		    cr);
3017 	}
3018 
3019 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3020 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3021 
3022 		dmu_tx_hold_bonus(tx, zp->z_id);
3023 		error = dmu_tx_assign(tx, TXG_WAIT);
3024 		if (error) {
3025 			dmu_tx_abort(tx);
3026 		} else {
3027 			dmu_buf_will_dirty(zp->z_dbuf, tx);
3028 			mutex_enter(&zp->z_lock);
3029 			zp->z_atime_dirty = 0;
3030 			mutex_exit(&zp->z_lock);
3031 			dmu_tx_commit(tx);
3032 		}
3033 	}
3034 
3035 	zfs_zinactive(zp);
3036 	rw_exit(&zfsvfs->z_um_lock);
3037 }
3038 
3039 /*
3040  * Bounds-check the seek operation.
3041  *
3042  *	IN:	vp	- vnode seeking within
3043  *		ooff	- old file offset
3044  *		noffp	- pointer to new file offset
3045  *
3046  *	RETURN:	0 if success
3047  *		EINVAL if new offset invalid
3048  */
3049 /* ARGSUSED */
3050 static int
3051 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp)
3052 {
3053 	if (vp->v_type == VDIR)
3054 		return (0);
3055 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3056 }
3057 
3058 /*
3059  * Pre-filter the generic locking function to trap attempts to place
3060  * a mandatory lock on a memory mapped file.
3061  */
3062 static int
3063 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3064     flk_callback_t *flk_cbp, cred_t *cr)
3065 {
3066 	znode_t *zp = VTOZ(vp);
3067 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3068 	int error;
3069 
3070 	ZFS_ENTER(zfsvfs);
3071 
3072 	/*
3073 	 * We are following the UFS semantics with respect to mapcnt
3074 	 * here: If we see that the file is mapped already, then we will
3075 	 * return an error, but we don't worry about races between this
3076 	 * function and zfs_map().
3077 	 */
3078 	if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3079 		ZFS_EXIT(zfsvfs);
3080 		return (EAGAIN);
3081 	}
3082 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr);
3083 	ZFS_EXIT(zfsvfs);
3084 	return (error);
3085 }
3086 
3087 /*
3088  * If we can't find a page in the cache, we will create a new page
3089  * and fill it with file data.  For efficiency, we may try to fill
3090  * multiple pages at once (klustering).
3091  */
3092 static int
3093 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3094     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3095 {
3096 	znode_t *zp = VTOZ(vp);
3097 	page_t *pp, *cur_pp;
3098 	objset_t *os = zp->z_zfsvfs->z_os;
3099 	caddr_t va;
3100 	u_offset_t io_off, total;
3101 	uint64_t oid = zp->z_id;
3102 	size_t io_len;
3103 	uint64_t filesz;
3104 	int err;
3105 
3106 	/*
3107 	 * If we are only asking for a single page don't bother klustering.
3108 	 */
3109 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3110 	if (off >= filesz)
3111 		return (EFAULT);
3112 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3113 		io_off = off;
3114 		io_len = PAGESIZE;
3115 		pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr);
3116 	} else {
3117 		/*
3118 		 * Try to fill a kluster of pages (a blocks worth).
3119 		 */
3120 		size_t klen;
3121 		u_offset_t koff;
3122 
3123 		if (!ISP2(zp->z_blksz)) {
3124 			/* Only one block in the file. */
3125 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3126 			koff = 0;
3127 		} else {
3128 			/*
3129 			 * It would be ideal to align our offset to the
3130 			 * blocksize but doing so has resulted in some
3131 			 * strange application crashes. For now, we
3132 			 * leave the offset as is and only adjust the
3133 			 * length if we are off the end of the file.
3134 			 */
3135 			koff = off;
3136 			klen = plsz;
3137 		}
3138 		ASSERT(koff <= filesz);
3139 		if (koff + klen > filesz)
3140 			klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff;
3141 		ASSERT3U(off, >=, koff);
3142 		ASSERT3U(off, <, koff + klen);
3143 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3144 		    &io_len, koff, klen, 0);
3145 	}
3146 	if (pp == NULL) {
3147 		/*
3148 		 * Some other thread entered the page before us.
3149 		 * Return to zfs_getpage to retry the lookup.
3150 		 */
3151 		*pl = NULL;
3152 		return (0);
3153 	}
3154 
3155 	/*
3156 	 * Fill the pages in the kluster.
3157 	 */
3158 	cur_pp = pp;
3159 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3160 		ASSERT3U(io_off, ==, cur_pp->p_offset);
3161 		va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1);
3162 		err = dmu_read(os, oid, io_off, PAGESIZE, va);
3163 		ppmapout(va);
3164 		if (err) {
3165 			/* On error, toss the entire kluster */
3166 			pvn_read_done(pp, B_ERROR);
3167 			return (err);
3168 		}
3169 		cur_pp = cur_pp->p_next;
3170 	}
3171 out:
3172 	/*
3173 	 * Fill in the page list array from the kluster.  If
3174 	 * there are too many pages in the kluster, return
3175 	 * as many pages as possible starting from the desired
3176 	 * offset `off'.
3177 	 * NOTE: the page list will always be null terminated.
3178 	 */
3179 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3180 
3181 	return (0);
3182 }
3183 
3184 /*
3185  * Return pointers to the pages for the file region [off, off + len]
3186  * in the pl array.  If plsz is greater than len, this function may
3187  * also return page pointers from before or after the specified
3188  * region (i.e. some region [off', off' + plsz]).  These additional
3189  * pages are only returned if they are already in the cache, or were
3190  * created as part of a klustered read.
3191  *
3192  *	IN:	vp	- vnode of file to get data from.
3193  *		off	- position in file to get data from.
3194  *		len	- amount of data to retrieve.
3195  *		plsz	- length of provided page list.
3196  *		seg	- segment to obtain pages for.
3197  *		addr	- virtual address of fault.
3198  *		rw	- mode of created pages.
3199  *		cr	- credentials of caller.
3200  *
3201  *	OUT:	protp	- protection mode of created pages.
3202  *		pl	- list of pages created.
3203  *
3204  *	RETURN:	0 if success
3205  *		error code if failure
3206  *
3207  * Timestamps:
3208  *	vp - atime updated
3209  */
3210 /* ARGSUSED */
3211 static int
3212 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3213 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3214 	enum seg_rw rw, cred_t *cr)
3215 {
3216 	znode_t		*zp = VTOZ(vp);
3217 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3218 	page_t		*pp, **pl0 = pl;
3219 	int		need_unlock = 0, err = 0;
3220 	offset_t	orig_off;
3221 
3222 	ZFS_ENTER(zfsvfs);
3223 
3224 	if (protp)
3225 		*protp = PROT_ALL;
3226 
3227 	ASSERT(zp->z_dbuf_held && zp->z_phys);
3228 
3229 	/* no faultahead (for now) */
3230 	if (pl == NULL) {
3231 		ZFS_EXIT(zfsvfs);
3232 		return (0);
3233 	}
3234 
3235 	/* can't fault past EOF */
3236 	if (off >= zp->z_phys->zp_size) {
3237 		ZFS_EXIT(zfsvfs);
3238 		return (EFAULT);
3239 	}
3240 	orig_off = off;
3241 
3242 	/*
3243 	 * If we already own the lock, then we must be page faulting
3244 	 * in the middle of a write to this file (i.e., we are writing
3245 	 * to this file using data from a mapped region of the file).
3246 	 */
3247 	if (rw_owner(&zp->z_map_lock) != curthread) {
3248 		rw_enter(&zp->z_map_lock, RW_WRITER);
3249 		need_unlock = TRUE;
3250 	}
3251 
3252 	/*
3253 	 * Loop through the requested range [off, off + len] looking
3254 	 * for pages.  If we don't find a page, we will need to create
3255 	 * a new page and fill it with data from the file.
3256 	 */
3257 	while (len > 0) {
3258 		if (plsz < PAGESIZE)
3259 			break;
3260 		if (pp = page_lookup(vp, off, SE_SHARED)) {
3261 			*pl++ = pp;
3262 			off += PAGESIZE;
3263 			addr += PAGESIZE;
3264 			len -= PAGESIZE;
3265 			plsz -= PAGESIZE;
3266 		} else {
3267 			err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw);
3268 			if (err)
3269 				goto out;
3270 			/*
3271 			 * klustering may have changed our region
3272 			 * to be block aligned.
3273 			 */
3274 			if (((pp = *pl) != 0) && (off != pp->p_offset)) {
3275 				int delta = off - pp->p_offset;
3276 				len += delta;
3277 				off -= delta;
3278 				addr -= delta;
3279 			}
3280 			while (*pl) {
3281 				pl++;
3282 				off += PAGESIZE;
3283 				addr += PAGESIZE;
3284 				plsz -= PAGESIZE;
3285 				if (len > PAGESIZE)
3286 					len -= PAGESIZE;
3287 				else
3288 					len = 0;
3289 			}
3290 		}
3291 	}
3292 
3293 	/*
3294 	 * Fill out the page array with any pages already in the cache.
3295 	 */
3296 	while (plsz > 0) {
3297 		pp = page_lookup_nowait(vp, off, SE_SHARED);
3298 		if (pp == NULL)
3299 			break;
3300 		*pl++ = pp;
3301 		off += PAGESIZE;
3302 		plsz -= PAGESIZE;
3303 	}
3304 
3305 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3306 out:
3307 	/*
3308 	 * We can't grab the range lock for the page as reader which would
3309 	 * stop truncation as this leads to deadlock. So we need to recheck
3310 	 * the file size.
3311 	 */
3312 	if (orig_off >= zp->z_phys->zp_size)
3313 		err = EFAULT;
3314 	if (err) {
3315 		/*
3316 		 * Release any pages we have previously locked.
3317 		 */
3318 		while (pl > pl0)
3319 			page_unlock(*--pl);
3320 	}
3321 
3322 	*pl = NULL;
3323 
3324 	if (need_unlock)
3325 		rw_exit(&zp->z_map_lock);
3326 
3327 	ZFS_EXIT(zfsvfs);
3328 	return (err);
3329 }
3330 
3331 /*
3332  * Request a memory map for a section of a file.  This code interacts
3333  * with common code and the VM system as follows:
3334  *
3335  *	common code calls mmap(), which ends up in smmap_common()
3336  *
3337  *	this calls VOP_MAP(), which takes you into (say) zfs
3338  *
3339  *	zfs_map() calls as_map(), passing segvn_create() as the callback
3340  *
3341  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
3342  *
3343  *	zfs_addmap() updates z_mapcnt
3344  */
3345 static int
3346 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
3347     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3348 {
3349 	znode_t *zp = VTOZ(vp);
3350 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3351 	segvn_crargs_t	vn_a;
3352 	int		error;
3353 
3354 	ZFS_ENTER(zfsvfs);
3355 
3356 	if (vp->v_flag & VNOMAP) {
3357 		ZFS_EXIT(zfsvfs);
3358 		return (ENOSYS);
3359 	}
3360 
3361 	if (off < 0 || len > MAXOFFSET_T - off) {
3362 		ZFS_EXIT(zfsvfs);
3363 		return (ENXIO);
3364 	}
3365 
3366 	if (vp->v_type != VREG) {
3367 		ZFS_EXIT(zfsvfs);
3368 		return (ENODEV);
3369 	}
3370 
3371 	/*
3372 	 * If file is locked, disallow mapping.
3373 	 */
3374 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) {
3375 		ZFS_EXIT(zfsvfs);
3376 		return (EAGAIN);
3377 	}
3378 
3379 	as_rangelock(as);
3380 	if ((flags & MAP_FIXED) == 0) {
3381 		map_addr(addrp, len, off, 1, flags);
3382 		if (*addrp == NULL) {
3383 			as_rangeunlock(as);
3384 			ZFS_EXIT(zfsvfs);
3385 			return (ENOMEM);
3386 		}
3387 	} else {
3388 		/*
3389 		 * User specified address - blow away any previous mappings
3390 		 */
3391 		(void) as_unmap(as, *addrp, len);
3392 	}
3393 
3394 	vn_a.vp = vp;
3395 	vn_a.offset = (u_offset_t)off;
3396 	vn_a.type = flags & MAP_TYPE;
3397 	vn_a.prot = prot;
3398 	vn_a.maxprot = maxprot;
3399 	vn_a.cred = cr;
3400 	vn_a.amp = NULL;
3401 	vn_a.flags = flags & ~MAP_TYPE;
3402 	vn_a.szc = 0;
3403 	vn_a.lgrp_mem_policy_flags = 0;
3404 
3405 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
3406 
3407 	as_rangeunlock(as);
3408 	ZFS_EXIT(zfsvfs);
3409 	return (error);
3410 }
3411 
3412 /* ARGSUSED */
3413 static int
3414 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3415     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3416 {
3417 	uint64_t pages = btopr(len);
3418 
3419 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
3420 	return (0);
3421 }
3422 
3423 /*
3424  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
3425  * more accurate mtime for the associated file.  Since we don't have a way of
3426  * detecting when the data was actually modified, we have to resort to
3427  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
3428  * last page is pushed.  The problem occurs when the msync() call is omitted,
3429  * which by far the most common case:
3430  *
3431  * 	open()
3432  * 	mmap()
3433  * 	<modify memory>
3434  * 	munmap()
3435  * 	close()
3436  * 	<time lapse>
3437  * 	putpage() via fsflush
3438  *
3439  * If we wait until fsflush to come along, we can have a modification time that
3440  * is some arbitrary point in the future.  In order to prevent this in the
3441  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
3442  * torn down.
3443  */
3444 /* ARGSUSED */
3445 static int
3446 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3447     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr)
3448 {
3449 	uint64_t pages = btopr(len);
3450 
3451 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
3452 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
3453 
3454 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
3455 	    vn_has_cached_data(vp))
3456 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr);
3457 
3458 	return (0);
3459 }
3460 
3461 /*
3462  * Free or allocate space in a file.  Currently, this function only
3463  * supports the `F_FREESP' command.  However, this command is somewhat
3464  * misnamed, as its functionality includes the ability to allocate as
3465  * well as free space.
3466  *
3467  *	IN:	vp	- vnode of file to free data in.
3468  *		cmd	- action to take (only F_FREESP supported).
3469  *		bfp	- section of file to free/alloc.
3470  *		flag	- current file open mode flags.
3471  *		offset	- current file offset.
3472  *		cr	- credentials of caller [UNUSED].
3473  *
3474  *	RETURN:	0 if success
3475  *		error code if failure
3476  *
3477  * Timestamps:
3478  *	vp - ctime|mtime updated
3479  */
3480 /* ARGSUSED */
3481 static int
3482 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
3483     offset_t offset, cred_t *cr, caller_context_t *ct)
3484 {
3485 	znode_t		*zp = VTOZ(vp);
3486 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3487 	uint64_t	off, len;
3488 	int		error;
3489 
3490 	ZFS_ENTER(zfsvfs);
3491 
3492 top:
3493 	if (cmd != F_FREESP) {
3494 		ZFS_EXIT(zfsvfs);
3495 		return (EINVAL);
3496 	}
3497 
3498 	if (error = convoff(vp, bfp, 0, offset)) {
3499 		ZFS_EXIT(zfsvfs);
3500 		return (error);
3501 	}
3502 
3503 	if (bfp->l_len < 0) {
3504 		ZFS_EXIT(zfsvfs);
3505 		return (EINVAL);
3506 	}
3507 
3508 	off = bfp->l_start;
3509 	len = bfp->l_len; /* 0 means from off to end of file */
3510 
3511 	do {
3512 		error = zfs_freesp(zp, off, len, flag, TRUE);
3513 		/* NB: we already did dmu_tx_wait() if necessary */
3514 	} while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
3515 
3516 	ZFS_EXIT(zfsvfs);
3517 	return (error);
3518 }
3519 
3520 static int
3521 zfs_fid(vnode_t *vp, fid_t *fidp)
3522 {
3523 	znode_t		*zp = VTOZ(vp);
3524 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3525 	uint32_t	gen = (uint32_t)zp->z_phys->zp_gen;
3526 	uint64_t	object = zp->z_id;
3527 	zfid_short_t	*zfid;
3528 	int		size, i;
3529 
3530 	ZFS_ENTER(zfsvfs);
3531 
3532 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
3533 	if (fidp->fid_len < size) {
3534 		fidp->fid_len = size;
3535 		ZFS_EXIT(zfsvfs);
3536 		return (ENOSPC);
3537 	}
3538 
3539 	zfid = (zfid_short_t *)fidp;
3540 
3541 	zfid->zf_len = size;
3542 
3543 	for (i = 0; i < sizeof (zfid->zf_object); i++)
3544 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
3545 
3546 	/* Must have a non-zero generation number to distinguish from .zfs */
3547 	if (gen == 0)
3548 		gen = 1;
3549 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
3550 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
3551 
3552 	if (size == LONG_FID_LEN) {
3553 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
3554 		zfid_long_t	*zlfid;
3555 
3556 		zlfid = (zfid_long_t *)fidp;
3557 
3558 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
3559 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
3560 
3561 		/* XXX - this should be the generation number for the objset */
3562 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
3563 			zlfid->zf_setgen[i] = 0;
3564 	}
3565 
3566 	ZFS_EXIT(zfsvfs);
3567 	return (0);
3568 }
3569 
3570 static int
3571 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr)
3572 {
3573 	znode_t		*zp, *xzp;
3574 	zfsvfs_t	*zfsvfs;
3575 	zfs_dirlock_t	*dl;
3576 	int		error;
3577 
3578 	switch (cmd) {
3579 	case _PC_LINK_MAX:
3580 		*valp = ULONG_MAX;
3581 		return (0);
3582 
3583 	case _PC_FILESIZEBITS:
3584 		*valp = 64;
3585 		return (0);
3586 
3587 	case _PC_XATTR_EXISTS:
3588 		zp = VTOZ(vp);
3589 		zfsvfs = zp->z_zfsvfs;
3590 		ZFS_ENTER(zfsvfs);
3591 		*valp = 0;
3592 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
3593 		    ZXATTR | ZEXISTS | ZSHARED);
3594 		if (error == 0) {
3595 			zfs_dirent_unlock(dl);
3596 			if (!zfs_dirempty(xzp))
3597 				*valp = 1;
3598 			VN_RELE(ZTOV(xzp));
3599 		} else if (error == ENOENT) {
3600 			/*
3601 			 * If there aren't extended attributes, it's the
3602 			 * same as having zero of them.
3603 			 */
3604 			error = 0;
3605 		}
3606 		ZFS_EXIT(zfsvfs);
3607 		return (error);
3608 
3609 	case _PC_ACL_ENABLED:
3610 		*valp = _ACL_ACE_ENABLED;
3611 		return (0);
3612 
3613 	case _PC_MIN_HOLE_SIZE:
3614 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
3615 		return (0);
3616 
3617 	default:
3618 		return (fs_pathconf(vp, cmd, valp, cr));
3619 	}
3620 }
3621 
3622 /*ARGSUSED*/
3623 static int
3624 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr)
3625 {
3626 	znode_t *zp = VTOZ(vp);
3627 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3628 	int error;
3629 
3630 	ZFS_ENTER(zfsvfs);
3631 	error = zfs_getacl(zp, vsecp, cr);
3632 	ZFS_EXIT(zfsvfs);
3633 
3634 	return (error);
3635 }
3636 
3637 /*ARGSUSED*/
3638 static int
3639 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr)
3640 {
3641 	znode_t *zp = VTOZ(vp);
3642 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3643 	int error;
3644 
3645 	ZFS_ENTER(zfsvfs);
3646 	error = zfs_setacl(zp, vsecp, cr);
3647 	ZFS_EXIT(zfsvfs);
3648 	return (error);
3649 }
3650 
3651 /*
3652  * Predeclare these here so that the compiler assumes that
3653  * this is an "old style" function declaration that does
3654  * not include arguments => we won't get type mismatch errors
3655  * in the initializations that follow.
3656  */
3657 static int zfs_inval();
3658 static int zfs_isdir();
3659 
3660 static int
3661 zfs_inval()
3662 {
3663 	return (EINVAL);
3664 }
3665 
3666 static int
3667 zfs_isdir()
3668 {
3669 	return (EISDIR);
3670 }
3671 /*
3672  * Directory vnode operations template
3673  */
3674 vnodeops_t *zfs_dvnodeops;
3675 const fs_operation_def_t zfs_dvnodeops_template[] = {
3676 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
3677 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
3678 	VOPNAME_READ,		{ .error = zfs_isdir },
3679 	VOPNAME_WRITE,		{ .error = zfs_isdir },
3680 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
3681 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
3682 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
3683 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
3684 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
3685 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
3686 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
3687 	VOPNAME_LINK,		{ .vop_link = zfs_link },
3688 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
3689 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
3690 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
3691 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
3692 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
3693 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
3694 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
3695 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
3696 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
3697 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
3698 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
3699 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
3700 	NULL,			NULL
3701 };
3702 
3703 /*
3704  * Regular file vnode operations template
3705  */
3706 vnodeops_t *zfs_fvnodeops;
3707 const fs_operation_def_t zfs_fvnodeops_template[] = {
3708 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
3709 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
3710 	VOPNAME_READ,		{ .vop_read = zfs_read },
3711 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
3712 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
3713 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
3714 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
3715 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
3716 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
3717 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
3718 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
3719 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
3720 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
3721 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
3722 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
3723 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
3724 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
3725 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
3726 	VOPNAME_MAP,		{ .vop_map = zfs_map },
3727 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
3728 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
3729 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
3730 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
3731 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
3732 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
3733 	NULL,			NULL
3734 };
3735 
3736 /*
3737  * Symbolic link vnode operations template
3738  */
3739 vnodeops_t *zfs_symvnodeops;
3740 const fs_operation_def_t zfs_symvnodeops_template[] = {
3741 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
3742 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
3743 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
3744 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
3745 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
3746 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
3747 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
3748 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
3749 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
3750 	NULL,			NULL
3751 };
3752 
3753 /*
3754  * Extended attribute directory vnode operations template
3755  *	This template is identical to the directory vnodes
3756  *	operation template except for restricted operations:
3757  *		VOP_MKDIR()
3758  *		VOP_SYMLINK()
3759  * Note that there are other restrictions embedded in:
3760  *	zfs_create()	- restrict type to VREG
3761  *	zfs_link()	- no links into/out of attribute space
3762  *	zfs_rename()	- no moves into/out of attribute space
3763  */
3764 vnodeops_t *zfs_xdvnodeops;
3765 const fs_operation_def_t zfs_xdvnodeops_template[] = {
3766 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
3767 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
3768 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
3769 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
3770 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
3771 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
3772 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
3773 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
3774 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
3775 	VOPNAME_LINK,		{ .vop_link = zfs_link },
3776 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
3777 	VOPNAME_MKDIR,		{ .error = zfs_inval },
3778 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
3779 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
3780 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
3781 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
3782 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
3783 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
3784 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
3785 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
3786 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
3787 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
3788 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
3789 	NULL,			NULL
3790 };
3791 
3792 /*
3793  * Error vnode operations template
3794  */
3795 vnodeops_t *zfs_evnodeops;
3796 const fs_operation_def_t zfs_evnodeops_template[] = {
3797 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
3798 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
3799 	NULL,			NULL
3800 };
3801