xref: /titanic_44/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 0ba2cbe97e0678a691742f98d2532caed0a2c4aa)
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 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/time.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
33 #include <sys/resource.h>
34 #include <sys/vfs.h>
35 #include <sys/vnode.h>
36 #include <sys/file.h>
37 #include <sys/stat.h>
38 #include <sys/kmem.h>
39 #include <sys/taskq.h>
40 #include <sys/uio.h>
41 #include <sys/vmsystm.h>
42 #include <sys/atomic.h>
43 #include <sys/vm.h>
44 #include <vm/seg_vn.h>
45 #include <vm/pvn.h>
46 #include <vm/as.h>
47 #include <sys/mman.h>
48 #include <sys/pathname.h>
49 #include <sys/cmn_err.h>
50 #include <sys/errno.h>
51 #include <sys/unistd.h>
52 #include <sys/zfs_vfsops.h>
53 #include <sys/zfs_dir.h>
54 #include <sys/zfs_acl.h>
55 #include <sys/zfs_ioctl.h>
56 #include <sys/fs/zfs.h>
57 #include <sys/dmu.h>
58 #include <sys/spa.h>
59 #include <sys/txg.h>
60 #include <sys/dbuf.h>
61 #include <sys/zap.h>
62 #include <sys/dirent.h>
63 #include <sys/policy.h>
64 #include <sys/sunddi.h>
65 #include <sys/filio.h>
66 #include "fs/fs_subr.h"
67 #include <sys/zfs_ctldir.h>
68 #include <sys/dnlc.h>
69 #include <sys/zfs_rlock.h>
70 
71 /*
72  * Programming rules.
73  *
74  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
75  * properly lock its in-core state, create a DMU transaction, do the work,
76  * record this work in the intent log (ZIL), commit the DMU transaction,
77  * and wait the the intent log to commit if it's is a synchronous operation.
78  * Morover, the vnode ops must work in both normal and log replay context.
79  * The ordering of events is important to avoid deadlocks and references
80  * to freed memory.  The example below illustrates the following Big Rules:
81  *
82  *  (1) A check must be made in each zfs thread for a mounted file system.
83  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
84  *	A ZFS_EXIT(zfsvfs) is needed before all returns.
85  *
86  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
87  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
88  *	First, if it's the last reference, the vnode/znode
89  *	can be freed, so the zp may point to freed memory.  Second, the last
90  *	reference will call zfs_zinactive(), which may induce a lot of work --
91  *	pushing cached pages (which acquires range locks) and syncing out
92  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
93  *	which could deadlock the system if you were already holding one.
94  *
95  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
96  *	as they can span dmu_tx_assign() calls.
97  *
98  *  (4)	Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign().
99  *	In normal operation, this will be TXG_NOWAIT.  During ZIL replay,
100  *	it will be a specific txg.  Either way, dmu_tx_assign() never blocks.
101  *	This is critical because we don't want to block while holding locks.
102  *	Note, in particular, that if a lock is sometimes acquired before
103  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
104  *	use a non-blocking assign can deadlock the system.  The scenario:
105  *
106  *	Thread A has grabbed a lock before calling dmu_tx_assign().
107  *	Thread B is in an already-assigned tx, and blocks for this lock.
108  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
109  *	forever, because the previous txg can't quiesce until B's tx commits.
110  *
111  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
112  *	then drop all locks, call dmu_tx_wait(), and try again.
113  *
114  *  (5)	If the operation succeeded, generate the intent log entry for it
115  *	before dropping locks.  This ensures that the ordering of events
116  *	in the intent log matches the order in which they actually occurred.
117  *
118  *  (6)	At the end of each vnode op, the DMU tx must always commit,
119  *	regardless of whether there were any errors.
120  *
121  *  (7)	After dropping all locks, invoke zil_commit(zilog, seq, foid)
122  *	to ensure that synchronous semantics are provided when necessary.
123  *
124  * In general, this is how things should be ordered in each vnode op:
125  *
126  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
127  * top:
128  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
129  *	rw_enter(...);			// grab any other locks you need
130  *	tx = dmu_tx_create(...);	// get DMU tx
131  *	dmu_tx_hold_*();		// hold each object you might modify
132  *	error = dmu_tx_assign(tx, zfsvfs->z_assign);	// try to assign
133  *	if (error) {
134  *		rw_exit(...);		// drop locks
135  *		zfs_dirent_unlock(dl);	// unlock directory entry
136  *		VN_RELE(...);		// release held vnodes
137  *		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
138  *			dmu_tx_wait(tx);
139  *			dmu_tx_abort(tx);
140  *			goto top;
141  *		}
142  *		dmu_tx_abort(tx);	// abort DMU tx
143  *		ZFS_EXIT(zfsvfs);	// finished in zfs
144  *		return (error);		// really out of space
145  *	}
146  *	error = do_real_work();		// do whatever this VOP does
147  *	if (error == 0)
148  *		zfs_log_*(...);		// on success, make ZIL entry
149  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
150  *	rw_exit(...);			// drop locks
151  *	zfs_dirent_unlock(dl);		// unlock directory entry
152  *	VN_RELE(...);			// release held vnodes
153  *	zil_commit(zilog, seq, foid);	// synchronous when necessary
154  *	ZFS_EXIT(zfsvfs);		// finished in zfs
155  *	return (error);			// done, report error
156  */
157 /* ARGSUSED */
158 static int
159 zfs_open(vnode_t **vpp, int flag, cred_t *cr)
160 {
161 	znode_t	*zp = VTOZ(*vpp);
162 
163 	/* Keep a count of the synchronous opens in the znode */
164 	if (flag & (FSYNC | FDSYNC))
165 		atomic_inc_32(&zp->z_sync_cnt);
166 	return (0);
167 }
168 
169 /* ARGSUSED */
170 static int
171 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr)
172 {
173 	znode_t	*zp = VTOZ(vp);
174 
175 	/* Decrement the synchronous opens in the znode */
176 	if (flag & (FSYNC | FDSYNC))
177 		atomic_dec_32(&zp->z_sync_cnt);
178 
179 	/*
180 	 * Clean up any locks held by this process on the vp.
181 	 */
182 	cleanlocks(vp, ddi_get_pid(), 0);
183 	cleanshares(vp, ddi_get_pid());
184 
185 	return (0);
186 }
187 
188 /*
189  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
190  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
191  */
192 static int
193 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
194 {
195 	znode_t	*zp = VTOZ(vp);
196 	uint64_t noff = (uint64_t)*off; /* new offset */
197 	uint64_t file_sz;
198 	int error;
199 	boolean_t hole;
200 
201 	file_sz = zp->z_phys->zp_size;
202 	if (noff >= file_sz)  {
203 		return (ENXIO);
204 	}
205 
206 	if (cmd == _FIO_SEEK_HOLE)
207 		hole = B_TRUE;
208 	else
209 		hole = B_FALSE;
210 
211 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
212 
213 	/* end of file? */
214 	if ((error == ESRCH) || (noff > file_sz)) {
215 		/*
216 		 * Handle the virtual hole at the end of file.
217 		 */
218 		if (hole) {
219 			*off = file_sz;
220 			return (0);
221 		}
222 		return (ENXIO);
223 	}
224 
225 	if (noff < *off)
226 		return (error);
227 	*off = noff;
228 	return (error);
229 }
230 
231 /* ARGSUSED */
232 static int
233 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
234     int *rvalp)
235 {
236 	offset_t off;
237 	int error;
238 	zfsvfs_t *zfsvfs;
239 
240 	switch (com) {
241 	    case _FIOFFS:
242 		return (zfs_sync(vp->v_vfsp, 0, cred));
243 
244 		/*
245 		 * The following two ioctls are used by bfu.  Faking out,
246 		 * necessary to avoid bfu errors.
247 		 */
248 	    case _FIOGDIO:
249 	    case _FIOSDIO:
250 		return (0);
251 
252 	    case _FIO_SEEK_DATA:
253 	    case _FIO_SEEK_HOLE:
254 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
255 			return (EFAULT);
256 
257 		zfsvfs = VTOZ(vp)->z_zfsvfs;
258 		ZFS_ENTER(zfsvfs);
259 
260 		/* offset parameter is in/out */
261 		error = zfs_holey(vp, com, &off);
262 		ZFS_EXIT(zfsvfs);
263 		if (error)
264 			return (error);
265 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
266 			return (EFAULT);
267 		return (0);
268 	}
269 	return (ENOTTY);
270 }
271 
272 /*
273  * When a file is memory mapped, we must keep the IO data synchronized
274  * between the DMU cache and the memory mapped pages.  What this means:
275  *
276  * On Write:	If we find a memory mapped page, we write to *both*
277  *		the page and the dmu buffer.
278  *
279  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
280  *	the file is memory mapped.
281  */
282 static int
283 mappedwrite(vnode_t *vp, uint64_t woff, int nbytes, uio_t *uio, dmu_tx_t *tx)
284 {
285 	znode_t	*zp = VTOZ(vp);
286 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
287 	int64_t	start, off;
288 	int len = nbytes;
289 	int error = 0;
290 
291 	start = uio->uio_loffset;
292 	off = start & PAGEOFFSET;
293 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
294 		page_t *pp;
295 		uint64_t bytes = MIN(PAGESIZE - off, len);
296 
297 		/*
298 		 * We don't want a new page to "appear" in the middle of
299 		 * the file update (because it may not get the write
300 		 * update data), so we grab a lock to block
301 		 * zfs_getpage().
302 		 */
303 		rw_enter(&zp->z_map_lock, RW_WRITER);
304 		if (pp = page_lookup(vp, start, SE_SHARED)) {
305 			caddr_t va;
306 
307 			rw_exit(&zp->z_map_lock);
308 			va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L);
309 			error = uiomove(va+off, bytes, UIO_WRITE, uio);
310 			if (error == 0) {
311 				dmu_write(zfsvfs->z_os, zp->z_id,
312 				    woff, bytes, va+off, tx);
313 			}
314 			ppmapout(va);
315 			page_unlock(pp);
316 		} else {
317 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
318 			    woff, bytes, uio, tx);
319 			rw_exit(&zp->z_map_lock);
320 		}
321 		len -= bytes;
322 		woff += bytes;
323 		off = 0;
324 		if (error)
325 			break;
326 	}
327 	return (error);
328 }
329 
330 /*
331  * When a file is memory mapped, we must keep the IO data synchronized
332  * between the DMU cache and the memory mapped pages.  What this means:
333  *
334  * On Read:	We "read" preferentially from memory mapped pages,
335  *		else we default from the dmu buffer.
336  *
337  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
338  *	the file is memory mapped.
339  */
340 static int
341 mappedread(vnode_t *vp, char *addr, int nbytes, uio_t *uio)
342 {
343 	int64_t	start, off, bytes;
344 	int len = nbytes;
345 	int error = 0;
346 
347 	start = uio->uio_loffset;
348 	off = start & PAGEOFFSET;
349 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
350 		page_t *pp;
351 
352 		bytes = MIN(PAGESIZE - off, len);
353 		if (pp = page_lookup(vp, start, SE_SHARED)) {
354 			caddr_t va;
355 
356 			va = ppmapin(pp, PROT_READ, (caddr_t)-1L);
357 			error = uiomove(va + off, bytes, UIO_READ, uio);
358 			ppmapout(va);
359 			page_unlock(pp);
360 		} else {
361 			/* XXX use dmu_read here? */
362 			error = uiomove(addr, bytes, UIO_READ, uio);
363 		}
364 		len -= bytes;
365 		addr += bytes;
366 		off = 0;
367 		if (error)
368 			break;
369 	}
370 	return (error);
371 }
372 
373 uint_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
374 
375 /*
376  * Read bytes from specified file into supplied buffer.
377  *
378  *	IN:	vp	- vnode of file to be read from.
379  *		uio	- structure supplying read location, range info,
380  *			  and return buffer.
381  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
382  *		cr	- credentials of caller.
383  *
384  *	OUT:	uio	- updated offset and range, buffer filled.
385  *
386  *	RETURN:	0 if success
387  *		error code if failure
388  *
389  * Side Effects:
390  *	vp - atime updated if byte count > 0
391  */
392 /* ARGSUSED */
393 static int
394 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
395 {
396 	znode_t		*zp = VTOZ(vp);
397 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
398 	uint64_t	delta;
399 	ssize_t		n, size, cnt, ndone;
400 	int		error, i, numbufs;
401 	dmu_buf_t	*dbp, **dbpp;
402 	rl_t		*rl;
403 
404 	ZFS_ENTER(zfsvfs);
405 
406 	/*
407 	 * Validate file offset
408 	 */
409 	if (uio->uio_loffset < (offset_t)0) {
410 		ZFS_EXIT(zfsvfs);
411 		return (EINVAL);
412 	}
413 
414 	/*
415 	 * Fasttrack empty reads
416 	 */
417 	if (uio->uio_resid == 0) {
418 		ZFS_EXIT(zfsvfs);
419 		return (0);
420 	}
421 
422 	/*
423 	 * Check for mandatory locks
424 	 */
425 	if (MANDMODE((mode_t)zp->z_phys->zp_mode)) {
426 		if (error = chklock(vp, FREAD,
427 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
428 			ZFS_EXIT(zfsvfs);
429 			return (error);
430 		}
431 	}
432 
433 	/*
434 	 * If we're in FRSYNC mode, sync out this znode before reading it.
435 	 */
436 	if (ioflag & FRSYNC)
437 		zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
438 
439 	/*
440 	 * Lock the range against changes.
441 	 */
442 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
443 
444 	/*
445 	 * If we are reading past end-of-file we can skip
446 	 * to the end; but we might still need to set atime.
447 	 */
448 	if (uio->uio_loffset >= zp->z_phys->zp_size) {
449 		cnt = 0;
450 		error = 0;
451 		goto out;
452 	}
453 
454 	cnt = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset);
455 
456 	for (ndone = 0; ndone < cnt; ndone += zfs_read_chunk_size) {
457 		ASSERT(uio->uio_loffset < zp->z_phys->zp_size);
458 		n = MIN(zfs_read_chunk_size,
459 		    zp->z_phys->zp_size - uio->uio_loffset);
460 		n = MIN(n, cnt);
461 		error = dmu_buf_hold_array_by_bonus(zp->z_dbuf,
462 		    uio->uio_loffset, n, TRUE, FTAG, &numbufs, &dbpp);
463 		if (error)
464 			goto out;
465 		/*
466 		 * Compute the adjustment to align the dmu buffers
467 		 * with the uio buffer.
468 		 */
469 		delta = uio->uio_loffset - dbpp[0]->db_offset;
470 
471 		for (i = 0; i < numbufs; i++) {
472 			if (n < 0)
473 				break;
474 			dbp = dbpp[i];
475 			size = dbp->db_size - delta;
476 			/*
477 			 * XXX -- this is correct, but may be suboptimal.
478 			 * If the pages are all clean, we don't need to
479 			 * go through mappedread().  Maybe the VMODSORT
480 			 * stuff can help us here.
481 			 */
482 			if (vn_has_cached_data(vp)) {
483 				error = mappedread(vp, (caddr_t)dbp->db_data +
484 				    delta, (n < size ? n : size), uio);
485 			} else {
486 				error = uiomove((caddr_t)dbp->db_data + delta,
487 					(n < size ? n : size), UIO_READ, uio);
488 			}
489 			if (error) {
490 				dmu_buf_rele_array(dbpp, numbufs, FTAG);
491 				goto out;
492 			}
493 			n -= dbp->db_size;
494 			if (delta) {
495 				n += delta;
496 				delta = 0;
497 			}
498 		}
499 		dmu_buf_rele_array(dbpp, numbufs, FTAG);
500 	}
501 out:
502 	zfs_range_unlock(rl);
503 
504 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
505 	ZFS_EXIT(zfsvfs);
506 	return (error);
507 }
508 
509 /*
510  * Fault in the pages of the first n bytes specified by the uio structure.
511  * 1 byte in each page is touched and the uio struct is unmodified.
512  * Any error will exit this routine as this is only a best
513  * attempt to get the pages resident. This is a copy of ufs_trans_touch().
514  */
515 static void
516 zfs_prefault_write(ssize_t n, struct uio *uio)
517 {
518 	struct iovec *iov;
519 	ulong_t cnt, incr;
520 	caddr_t p;
521 	uint8_t tmp;
522 
523 	iov = uio->uio_iov;
524 
525 	while (n) {
526 		cnt = MIN(iov->iov_len, n);
527 		if (cnt == 0) {
528 			/* empty iov entry */
529 			iov++;
530 			continue;
531 		}
532 		n -= cnt;
533 		/*
534 		 * touch each page in this segment.
535 		 */
536 		p = iov->iov_base;
537 		while (cnt) {
538 			switch (uio->uio_segflg) {
539 			case UIO_USERSPACE:
540 			case UIO_USERISPACE:
541 				if (fuword8(p, &tmp))
542 					return;
543 				break;
544 			case UIO_SYSSPACE:
545 				if (kcopy(p, &tmp, 1))
546 					return;
547 				break;
548 			}
549 			incr = MIN(cnt, PAGESIZE);
550 			p += incr;
551 			cnt -= incr;
552 		}
553 		/*
554 		 * touch the last byte in case it straddles a page.
555 		 */
556 		p--;
557 		switch (uio->uio_segflg) {
558 		case UIO_USERSPACE:
559 		case UIO_USERISPACE:
560 			if (fuword8(p, &tmp))
561 				return;
562 			break;
563 		case UIO_SYSSPACE:
564 			if (kcopy(p, &tmp, 1))
565 				return;
566 			break;
567 		}
568 		iov++;
569 	}
570 }
571 
572 /*
573  * Write the bytes to a file.
574  *
575  *	IN:	vp	- vnode of file to be written to.
576  *		uio	- structure supplying write location, range info,
577  *			  and data buffer.
578  *		ioflag	- FAPPEND flag set if in append mode.
579  *		cr	- credentials of caller.
580  *
581  *	OUT:	uio	- updated offset and range.
582  *
583  *	RETURN:	0 if success
584  *		error code if failure
585  *
586  * Timestamps:
587  *	vp - ctime|mtime updated if byte count > 0
588  */
589 /* ARGSUSED */
590 static int
591 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
592 {
593 	znode_t		*zp = VTOZ(vp);
594 	rlim64_t	limit = uio->uio_llimit;
595 	ssize_t		start_resid = uio->uio_resid;
596 	ssize_t		tx_bytes;
597 	uint64_t	end_size;
598 	dmu_tx_t	*tx;
599 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
600 	zilog_t		*zilog = zfsvfs->z_log;
601 	offset_t	woff;
602 	ssize_t		n, nbytes;
603 	rl_t		*rl;
604 	int		max_blksz = zfsvfs->z_max_blksz;
605 	int		error;
606 
607 	/*
608 	 * Fasttrack empty write
609 	 */
610 	n = start_resid;
611 	if (n == 0)
612 		return (0);
613 
614 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
615 		limit = MAXOFFSET_T;
616 
617 	ZFS_ENTER(zfsvfs);
618 
619 	/*
620 	 * Pre-fault the pages to ensure slow (eg NFS) pages
621 	 * don't hold up txg.
622 	 */
623 	zfs_prefault_write(n, uio);
624 
625 	/*
626 	 * If in append mode, set the io offset pointer to eof.
627 	 */
628 	if (ioflag & FAPPEND) {
629 		/*
630 		 * Range lock for a file append:
631 		 * The value for the start of range will be determined by
632 		 * zfs_range_lock() (to guarantee append semantics).
633 		 * If this write will cause the block size to increase,
634 		 * zfs_range_lock() will lock the entire file, so we must
635 		 * later reduce the range after we grow the block size.
636 		 */
637 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
638 		if (rl->r_len == UINT64_MAX) {
639 			/* overlocked, zp_size can't change */
640 			woff = uio->uio_loffset = zp->z_phys->zp_size;
641 		} else {
642 			woff = uio->uio_loffset = rl->r_off;
643 		}
644 	} else {
645 		woff = uio->uio_loffset;
646 		/*
647 		 * Validate file offset
648 		 */
649 		if (woff < 0) {
650 			ZFS_EXIT(zfsvfs);
651 			return (EINVAL);
652 		}
653 
654 		/*
655 		 * If we need to grow the block size then zfs_range_lock()
656 		 * will lock a wider range than we request here.
657 		 * Later after growing the block size we reduce the range.
658 		 */
659 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
660 	}
661 
662 	if (woff >= limit) {
663 		error = EFBIG;
664 		goto no_tx_done;
665 	}
666 
667 	if ((woff + n) > limit || woff > (limit - n))
668 		n = limit - woff;
669 
670 	/*
671 	 * Check for mandatory locks
672 	 */
673 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) &&
674 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0)
675 		goto no_tx_done;
676 	end_size = MAX(zp->z_phys->zp_size, woff + n);
677 top:
678 	tx = dmu_tx_create(zfsvfs->z_os);
679 	dmu_tx_hold_bonus(tx, zp->z_id);
680 	dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
681 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
682 	if (error) {
683 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
684 			dmu_tx_wait(tx);
685 			dmu_tx_abort(tx);
686 			goto top;
687 		}
688 		dmu_tx_abort(tx);
689 		goto no_tx_done;
690 	}
691 
692 	/*
693 	 * If zfs_range_lock() over-locked we grow the blocksize
694 	 * and then reduce the lock range.
695 	 */
696 	if (rl->r_len == UINT64_MAX) {
697 		uint64_t new_blksz;
698 
699 		if (zp->z_blksz > max_blksz) {
700 			ASSERT(!ISP2(zp->z_blksz));
701 			new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
702 		} else {
703 			new_blksz = MIN(end_size, max_blksz);
704 		}
705 		zfs_grow_blocksize(zp, new_blksz, tx);
706 		zfs_range_reduce(rl, woff, n);
707 	}
708 
709 	/*
710 	 * The file data does not fit in the znode "cache", so we
711 	 * will be writing to the file block data buffers.
712 	 * Each buffer will be written in a separate transaction;
713 	 * this keeps the intent log records small and allows us
714 	 * to do more fine-grained space accounting.
715 	 */
716 	while (n > 0) {
717 		/*
718 		 * XXX - should we really limit each write to z_max_blksz?
719 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
720 		 */
721 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
722 		rw_enter(&zp->z_map_lock, RW_READER);
723 
724 		tx_bytes = uio->uio_resid;
725 		if (vn_has_cached_data(vp)) {
726 			rw_exit(&zp->z_map_lock);
727 			error = mappedwrite(vp, woff, nbytes, uio, tx);
728 		} else {
729 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
730 			    woff, nbytes, uio, tx);
731 			rw_exit(&zp->z_map_lock);
732 		}
733 		tx_bytes -= uio->uio_resid;
734 
735 		if (error) {
736 			/* XXX - do we need to "clean up" the dmu buffer? */
737 			break;
738 		}
739 
740 		ASSERT(tx_bytes == nbytes);
741 
742 		/*
743 		 * Clear Set-UID/Set-GID bits on successful write if not
744 		 * privileged and at least one of the excute bits is set.
745 		 *
746 		 * It would be nice to to this after all writes have
747 		 * been done, but that would still expose the ISUID/ISGID
748 		 * to another app after the partial write is committed.
749 		 */
750 
751 		mutex_enter(&zp->z_acl_lock);
752 		if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) |
753 		    (S_IXUSR >> 6))) != 0 &&
754 		    (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 &&
755 		    secpolicy_vnode_setid_retain(cr,
756 		    (zp->z_phys->zp_mode & S_ISUID) != 0 &&
757 		    zp->z_phys->zp_uid == 0) != 0) {
758 			    zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID);
759 		}
760 		mutex_exit(&zp->z_acl_lock);
761 
762 		n -= nbytes;
763 		if (n <= 0)
764 			break;
765 
766 		/*
767 		 * We have more work ahead of us, so wrap up this transaction
768 		 * and start another.  Exact same logic as tx_done below.
769 		 */
770 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) {
771 			dmu_buf_will_dirty(zp->z_dbuf, tx);
772 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
773 			    uio->uio_loffset);
774 		}
775 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
776 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes,
777 		    ioflag, uio);
778 		dmu_tx_commit(tx);
779 
780 		/*
781 		 * Start another transaction.
782 		 */
783 		woff = uio->uio_loffset;
784 		tx = dmu_tx_create(zfsvfs->z_os);
785 		dmu_tx_hold_bonus(tx, zp->z_id);
786 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
787 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
788 		if (error) {
789 			if (error == ERESTART &&
790 			    zfsvfs->z_assign == TXG_NOWAIT) {
791 				dmu_tx_wait(tx);
792 				dmu_tx_abort(tx);
793 				goto top;
794 			}
795 			dmu_tx_abort(tx);
796 			goto no_tx_done;
797 		}
798 	}
799 
800 tx_done:
801 
802 	if (tx_bytes != 0) {
803 		/*
804 		 * Update the file size if it has changed; account
805 		 * for possible concurrent updates.
806 		 */
807 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) {
808 			dmu_buf_will_dirty(zp->z_dbuf, tx);
809 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
810 			    uio->uio_loffset);
811 		}
812 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
813 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes,
814 		    ioflag, uio);
815 	}
816 	dmu_tx_commit(tx);
817 
818 
819 no_tx_done:
820 
821 	zfs_range_unlock(rl);
822 
823 	/*
824 	 * If we're in replay mode, or we made no progress, return error.
825 	 * Otherwise, it's at least a partial write, so it's successful.
826 	 */
827 	if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) {
828 		ZFS_EXIT(zfsvfs);
829 		return (error);
830 	}
831 
832 	if (ioflag & (FSYNC | FDSYNC))
833 		zil_commit(zilog, zp->z_last_itx, zp->z_id);
834 
835 	ZFS_EXIT(zfsvfs);
836 	return (0);
837 }
838 
839 void
840 zfs_get_done(dmu_buf_t *db, void *vzgd)
841 {
842 	zgd_t *zgd = (zgd_t *)vzgd;
843 	rl_t *rl = zgd->zgd_rl;
844 	vnode_t *vp = ZTOV(rl->r_zp);
845 
846 	dmu_buf_rele(db, vzgd);
847 	zfs_range_unlock(rl);
848 	VN_RELE(vp);
849 	zil_add_vdev(zgd->zgd_zilog, DVA_GET_VDEV(BP_IDENTITY(zgd->zgd_bp)));
850 	kmem_free(zgd, sizeof (zgd_t));
851 }
852 
853 /*
854  * Get data to generate a TX_WRITE intent log record.
855  */
856 int
857 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
858 {
859 	zfsvfs_t *zfsvfs = arg;
860 	objset_t *os = zfsvfs->z_os;
861 	znode_t *zp;
862 	uint64_t off = lr->lr_offset;
863 	dmu_buf_t *db;
864 	rl_t *rl;
865 	zgd_t *zgd;
866 	int dlen = lr->lr_length;  		/* length of user data */
867 	int error = 0;
868 
869 	ASSERT(zio);
870 	ASSERT(dlen != 0);
871 
872 	/*
873 	 * Nothing to do if the file has been removed
874 	 */
875 	if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0)
876 		return (ENOENT);
877 	if (zp->z_reap) {
878 		VN_RELE(ZTOV(zp));
879 		return (ENOENT);
880 	}
881 
882 	/*
883 	 * Write records come in two flavors: immediate and indirect.
884 	 * For small writes it's cheaper to store the data with the
885 	 * log record (immediate); for large writes it's cheaper to
886 	 * sync the data and get a pointer to it (indirect) so that
887 	 * we don't have to write the data twice.
888 	 */
889 	if (buf != NULL) { /* immediate write */
890 		rl = zfs_range_lock(zp, off, dlen, RL_READER);
891 		/* test for truncation needs to be done while range locked */
892 		if (off >= zp->z_phys->zp_size) {
893 			error = ENOENT;
894 			goto out;
895 		}
896 		VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf));
897 	} else { /* indirect write */
898 		uint64_t boff; /* block starting offset */
899 
900 		ASSERT3U(dlen, <=, zp->z_blksz);
901 		/*
902 		 * Have to lock the whole block to ensure when it's
903 		 * written out and it's checksum is being calculated
904 		 * that no one can change the data. We need to re-check
905 		 * blocksize after we get the lock in case it's changed!
906 		 */
907 		for (;;) {
908 			if (ISP2(zp->z_blksz)) {
909 				boff = P2ALIGN_TYPED(off, zp->z_blksz,
910 				    uint64_t);
911 			} else {
912 				boff = 0;
913 			}
914 			dlen = zp->z_blksz;
915 			rl = zfs_range_lock(zp, boff, dlen, RL_READER);
916 			if (zp->z_blksz == dlen)
917 				break;
918 			zfs_range_unlock(rl);
919 		}
920 		/* test for truncation needs to be done while range locked */
921 		if (off >= zp->z_phys->zp_size) {
922 			error = ENOENT;
923 			goto out;
924 		}
925 		zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
926 		zgd->zgd_rl = rl;
927 		zgd->zgd_zilog = zfsvfs->z_log;
928 		zgd->zgd_bp = &lr->lr_blkptr;
929 		VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db));
930 		ASSERT(boff == db->db_offset);
931 		lr->lr_blkoff = off - boff;
932 		error = dmu_sync(zio, db, &lr->lr_blkptr,
933 		    lr->lr_common.lrc_txg, zfs_get_done, zgd);
934 		if (error == 0) {
935 			zil_add_vdev(zfsvfs->z_log,
936 			    DVA_GET_VDEV(BP_IDENTITY(&lr->lr_blkptr)));
937 		}
938 		/*
939 		 * If we get EINPROGRESS, then we need to wait for a
940 		 * write IO initiated by dmu_sync() to complete before
941 		 * we can release this dbuf.  We will finish everything
942 		 * up in the zfs_get_done() callback.
943 		 */
944 		if (error == EINPROGRESS)
945 			return (0);
946 		dmu_buf_rele(db, zgd);
947 		kmem_free(zgd, sizeof (zgd_t));
948 	}
949 out:
950 	zfs_range_unlock(rl);
951 	VN_RELE(ZTOV(zp));
952 	return (error);
953 }
954 
955 /*ARGSUSED*/
956 static int
957 zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr)
958 {
959 	znode_t *zp = VTOZ(vp);
960 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
961 	int error;
962 
963 	ZFS_ENTER(zfsvfs);
964 	error = zfs_zaccess_rwx(zp, mode, cr);
965 	ZFS_EXIT(zfsvfs);
966 	return (error);
967 }
968 
969 /*
970  * Lookup an entry in a directory, or an extended attribute directory.
971  * If it exists, return a held vnode reference for it.
972  *
973  *	IN:	dvp	- vnode of directory to search.
974  *		nm	- name of entry to lookup.
975  *		pnp	- full pathname to lookup [UNUSED].
976  *		flags	- LOOKUP_XATTR set if looking for an attribute.
977  *		rdir	- root directory vnode [UNUSED].
978  *		cr	- credentials of caller.
979  *
980  *	OUT:	vpp	- vnode of located entry, NULL if not found.
981  *
982  *	RETURN:	0 if success
983  *		error code if failure
984  *
985  * Timestamps:
986  *	NA
987  */
988 /* ARGSUSED */
989 static int
990 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
991     int flags, vnode_t *rdir, cred_t *cr)
992 {
993 
994 	znode_t *zdp = VTOZ(dvp);
995 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
996 	int	error;
997 
998 	ZFS_ENTER(zfsvfs);
999 
1000 	*vpp = NULL;
1001 
1002 	if (flags & LOOKUP_XATTR) {
1003 		/*
1004 		 * We don't allow recursive attributes..
1005 		 * Maybe someday we will.
1006 		 */
1007 		if (zdp->z_phys->zp_flags & ZFS_XATTR) {
1008 			ZFS_EXIT(zfsvfs);
1009 			return (EINVAL);
1010 		}
1011 
1012 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr)) {
1013 			ZFS_EXIT(zfsvfs);
1014 			return (error);
1015 		}
1016 
1017 		/*
1018 		 * Do we have permission to get into attribute directory?
1019 		 */
1020 
1021 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) {
1022 			VN_RELE(*vpp);
1023 		}
1024 
1025 		ZFS_EXIT(zfsvfs);
1026 		return (error);
1027 	}
1028 
1029 	if (dvp->v_type != VDIR) {
1030 		ZFS_EXIT(zfsvfs);
1031 		return (ENOTDIR);
1032 	}
1033 
1034 	/*
1035 	 * Check accessibility of directory.
1036 	 */
1037 
1038 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) {
1039 		ZFS_EXIT(zfsvfs);
1040 		return (error);
1041 	}
1042 
1043 	if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) {
1044 
1045 		/*
1046 		 * Convert device special files
1047 		 */
1048 		if (IS_DEVVP(*vpp)) {
1049 			vnode_t	*svp;
1050 
1051 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1052 			VN_RELE(*vpp);
1053 			if (svp == NULL)
1054 				error = ENOSYS;
1055 			else
1056 				*vpp = svp;
1057 		}
1058 	}
1059 
1060 	ZFS_EXIT(zfsvfs);
1061 	return (error);
1062 }
1063 
1064 /*
1065  * Attempt to create a new entry in a directory.  If the entry
1066  * already exists, truncate the file if permissible, else return
1067  * an error.  Return the vp of the created or trunc'd file.
1068  *
1069  *	IN:	dvp	- vnode of directory to put new file entry in.
1070  *		name	- name of new file entry.
1071  *		vap	- attributes of new file.
1072  *		excl	- flag indicating exclusive or non-exclusive mode.
1073  *		mode	- mode to open file with.
1074  *		cr	- credentials of caller.
1075  *		flag	- large file flag [UNUSED].
1076  *
1077  *	OUT:	vpp	- vnode of created or trunc'd entry.
1078  *
1079  *	RETURN:	0 if success
1080  *		error code if failure
1081  *
1082  * Timestamps:
1083  *	dvp - ctime|mtime updated if new entry created
1084  *	 vp - ctime|mtime always, atime if new
1085  */
1086 /* ARGSUSED */
1087 static int
1088 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1089     int mode, vnode_t **vpp, cred_t *cr, int flag)
1090 {
1091 	znode_t		*zp, *dzp = VTOZ(dvp);
1092 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1093 	zilog_t		*zilog = zfsvfs->z_log;
1094 	objset_t	*os = zfsvfs->z_os;
1095 	zfs_dirlock_t	*dl;
1096 	dmu_tx_t	*tx;
1097 	int		error;
1098 	uint64_t	zoid;
1099 
1100 	ZFS_ENTER(zfsvfs);
1101 
1102 top:
1103 	*vpp = NULL;
1104 
1105 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1106 		vap->va_mode &= ~VSVTX;
1107 
1108 	if (*name == '\0') {
1109 		/*
1110 		 * Null component name refers to the directory itself.
1111 		 */
1112 		VN_HOLD(dvp);
1113 		zp = dzp;
1114 		dl = NULL;
1115 		error = 0;
1116 	} else {
1117 		/* possible VN_HOLD(zp) */
1118 		if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) {
1119 			if (strcmp(name, "..") == 0)
1120 				error = EISDIR;
1121 			ZFS_EXIT(zfsvfs);
1122 			return (error);
1123 		}
1124 	}
1125 
1126 	zoid = zp ? zp->z_id : -1ULL;
1127 
1128 	if (zp == NULL) {
1129 		/*
1130 		 * Create a new file object and update the directory
1131 		 * to reference it.
1132 		 */
1133 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
1134 			goto out;
1135 		}
1136 
1137 		/*
1138 		 * We only support the creation of regular files in
1139 		 * extended attribute directories.
1140 		 */
1141 		if ((dzp->z_phys->zp_flags & ZFS_XATTR) &&
1142 		    (vap->va_type != VREG)) {
1143 			error = EINVAL;
1144 			goto out;
1145 		}
1146 
1147 		tx = dmu_tx_create(os);
1148 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1149 		dmu_tx_hold_bonus(tx, dzp->z_id);
1150 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1151 		if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
1152 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1153 			    0, SPA_MAXBLOCKSIZE);
1154 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
1155 		if (error) {
1156 			zfs_dirent_unlock(dl);
1157 			if (error == ERESTART &&
1158 			    zfsvfs->z_assign == TXG_NOWAIT) {
1159 				dmu_tx_wait(tx);
1160 				dmu_tx_abort(tx);
1161 				goto top;
1162 			}
1163 			dmu_tx_abort(tx);
1164 			ZFS_EXIT(zfsvfs);
1165 			return (error);
1166 		}
1167 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
1168 		ASSERT(zp->z_id == zoid);
1169 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1170 		zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name);
1171 		dmu_tx_commit(tx);
1172 	} else {
1173 		/*
1174 		 * A directory entry already exists for this name.
1175 		 */
1176 		/*
1177 		 * Can't truncate an existing file if in exclusive mode.
1178 		 */
1179 		if (excl == EXCL) {
1180 			error = EEXIST;
1181 			goto out;
1182 		}
1183 		/*
1184 		 * Can't open a directory for writing.
1185 		 */
1186 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1187 			error = EISDIR;
1188 			goto out;
1189 		}
1190 		/*
1191 		 * Verify requested access to file.
1192 		 */
1193 		if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) {
1194 			goto out;
1195 		}
1196 
1197 		mutex_enter(&dzp->z_lock);
1198 		dzp->z_seq++;
1199 		mutex_exit(&dzp->z_lock);
1200 
1201 		/*
1202 		 * Truncate regular files if requested.
1203 		 */
1204 		if ((ZTOV(zp)->v_type == VREG) &&
1205 		    (zp->z_phys->zp_size != 0) &&
1206 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1207 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1208 			if (error == ERESTART &&
1209 			    zfsvfs->z_assign == TXG_NOWAIT) {
1210 				/* NB: we already did dmu_tx_wait() */
1211 				zfs_dirent_unlock(dl);
1212 				VN_RELE(ZTOV(zp));
1213 				goto top;
1214 			}
1215 		}
1216 	}
1217 out:
1218 
1219 	if (dl)
1220 		zfs_dirent_unlock(dl);
1221 
1222 	if (error) {
1223 		if (zp)
1224 			VN_RELE(ZTOV(zp));
1225 	} else {
1226 		*vpp = ZTOV(zp);
1227 		/*
1228 		 * If vnode is for a device return a specfs vnode instead.
1229 		 */
1230 		if (IS_DEVVP(*vpp)) {
1231 			struct vnode *svp;
1232 
1233 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1234 			VN_RELE(*vpp);
1235 			if (svp == NULL) {
1236 				error = ENOSYS;
1237 			}
1238 			*vpp = svp;
1239 		}
1240 	}
1241 
1242 	ZFS_EXIT(zfsvfs);
1243 	return (error);
1244 }
1245 
1246 /*
1247  * Remove an entry from a directory.
1248  *
1249  *	IN:	dvp	- vnode of directory to remove entry from.
1250  *		name	- name of entry to remove.
1251  *		cr	- credentials of caller.
1252  *
1253  *	RETURN:	0 if success
1254  *		error code if failure
1255  *
1256  * Timestamps:
1257  *	dvp - ctime|mtime
1258  *	 vp - ctime (if nlink > 0)
1259  */
1260 static int
1261 zfs_remove(vnode_t *dvp, char *name, cred_t *cr)
1262 {
1263 	znode_t		*zp, *dzp = VTOZ(dvp);
1264 	znode_t		*xzp = NULL;
1265 	vnode_t		*vp;
1266 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1267 	zilog_t		*zilog = zfsvfs->z_log;
1268 	uint64_t	acl_obj, xattr_obj;
1269 	zfs_dirlock_t	*dl;
1270 	dmu_tx_t	*tx;
1271 	int		may_delete_now, delete_now = FALSE;
1272 	int		reaped;
1273 	int		error;
1274 
1275 	ZFS_ENTER(zfsvfs);
1276 
1277 top:
1278 	/*
1279 	 * Attempt to lock directory; fail if entry doesn't exist.
1280 	 */
1281 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) {
1282 		ZFS_EXIT(zfsvfs);
1283 		return (error);
1284 	}
1285 
1286 	vp = ZTOV(zp);
1287 
1288 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1289 		goto out;
1290 	}
1291 
1292 	/*
1293 	 * Need to use rmdir for removing directories.
1294 	 */
1295 	if (vp->v_type == VDIR) {
1296 		error = EPERM;
1297 		goto out;
1298 	}
1299 
1300 	vnevent_remove(vp);
1301 
1302 	dnlc_remove(dvp, name);
1303 
1304 	mutex_enter(&vp->v_lock);
1305 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1306 	mutex_exit(&vp->v_lock);
1307 
1308 	/*
1309 	 * We may delete the znode now, or we may put it on the delete queue;
1310 	 * it depends on whether we're the last link, and on whether there are
1311 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1312 	 * allow for either case.
1313 	 */
1314 	tx = dmu_tx_create(zfsvfs->z_os);
1315 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1316 	dmu_tx_hold_bonus(tx, zp->z_id);
1317 	if (may_delete_now)
1318 		dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
1319 
1320 	/* are there any extended attributes? */
1321 	if ((xattr_obj = zp->z_phys->zp_xattr) != 0) {
1322 		/*
1323 		 * XXX - There is a possibility that the delete
1324 		 * of the parent file could succeed, but then we get
1325 		 * an ENOSPC when we try to delete the xattrs...
1326 		 * so we would need to re-try the deletes periodically
1327 		 */
1328 		/* XXX - do we need this if we are deleting? */
1329 		dmu_tx_hold_bonus(tx, xattr_obj);
1330 	}
1331 
1332 	/* are there any additional acls */
1333 	if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 &&
1334 	    may_delete_now)
1335 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1336 
1337 	/* charge as an update -- would be nice not to charge at all */
1338 	dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL);
1339 
1340 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1341 	if (error) {
1342 		zfs_dirent_unlock(dl);
1343 		VN_RELE(vp);
1344 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1345 			dmu_tx_wait(tx);
1346 			dmu_tx_abort(tx);
1347 			goto top;
1348 		}
1349 		dmu_tx_abort(tx);
1350 		ZFS_EXIT(zfsvfs);
1351 		return (error);
1352 	}
1353 
1354 	/*
1355 	 * Remove the directory entry.
1356 	 */
1357 	error = zfs_link_destroy(dl, zp, tx, 0, &reaped);
1358 
1359 	if (error) {
1360 		dmu_tx_commit(tx);
1361 		goto out;
1362 	}
1363 
1364 	if (reaped) {
1365 		mutex_enter(&vp->v_lock);
1366 		delete_now = may_delete_now &&
1367 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1368 		    zp->z_phys->zp_xattr == xattr_obj &&
1369 		    zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj;
1370 		mutex_exit(&vp->v_lock);
1371 	}
1372 
1373 	if (delete_now) {
1374 		if (zp->z_phys->zp_xattr) {
1375 			error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
1376 			ASSERT3U(error, ==, 0);
1377 			ASSERT3U(xzp->z_phys->zp_links, ==, 2);
1378 			dmu_buf_will_dirty(xzp->z_dbuf, tx);
1379 			mutex_enter(&xzp->z_lock);
1380 			xzp->z_reap = 1;
1381 			xzp->z_phys->zp_links = 0;
1382 			mutex_exit(&xzp->z_lock);
1383 			zfs_dq_add(xzp, tx);
1384 			zp->z_phys->zp_xattr = 0; /* probably unnecessary */
1385 		}
1386 		mutex_enter(&zp->z_lock);
1387 		mutex_enter(&vp->v_lock);
1388 		vp->v_count--;
1389 		ASSERT3U(vp->v_count, ==, 0);
1390 		mutex_exit(&vp->v_lock);
1391 		mutex_exit(&zp->z_lock);
1392 		zfs_znode_delete(zp, tx);
1393 		VFS_RELE(zfsvfs->z_vfs);
1394 	} else if (reaped) {
1395 		zfs_dq_add(zp, tx);
1396 	}
1397 
1398 	zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name);
1399 
1400 	dmu_tx_commit(tx);
1401 out:
1402 	zfs_dirent_unlock(dl);
1403 
1404 	if (!delete_now) {
1405 		VN_RELE(vp);
1406 	} else if (xzp) {
1407 		/* this rele delayed to prevent nesting transactions */
1408 		VN_RELE(ZTOV(xzp));
1409 	}
1410 
1411 	ZFS_EXIT(zfsvfs);
1412 	return (error);
1413 }
1414 
1415 /*
1416  * Create a new directory and insert it into dvp using the name
1417  * provided.  Return a pointer to the inserted directory.
1418  *
1419  *	IN:	dvp	- vnode of directory to add subdir to.
1420  *		dirname	- name of new directory.
1421  *		vap	- attributes of new directory.
1422  *		cr	- credentials of caller.
1423  *
1424  *	OUT:	vpp	- vnode of created directory.
1425  *
1426  *	RETURN:	0 if success
1427  *		error code if failure
1428  *
1429  * Timestamps:
1430  *	dvp - ctime|mtime updated
1431  *	 vp - ctime|mtime|atime updated
1432  */
1433 static int
1434 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr)
1435 {
1436 	znode_t		*zp, *dzp = VTOZ(dvp);
1437 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1438 	zilog_t		*zilog = zfsvfs->z_log;
1439 	zfs_dirlock_t	*dl;
1440 	uint64_t	zoid = 0;
1441 	dmu_tx_t	*tx;
1442 	int		error;
1443 
1444 	ASSERT(vap->va_type == VDIR);
1445 
1446 	ZFS_ENTER(zfsvfs);
1447 
1448 	if (dzp->z_phys->zp_flags & ZFS_XATTR) {
1449 		ZFS_EXIT(zfsvfs);
1450 		return (EINVAL);
1451 	}
1452 top:
1453 	*vpp = NULL;
1454 
1455 	/*
1456 	 * First make sure the new directory doesn't exist.
1457 	 */
1458 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) {
1459 		ZFS_EXIT(zfsvfs);
1460 		return (error);
1461 	}
1462 
1463 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) {
1464 		zfs_dirent_unlock(dl);
1465 		ZFS_EXIT(zfsvfs);
1466 		return (error);
1467 	}
1468 
1469 	/*
1470 	 * Add a new entry to the directory.
1471 	 */
1472 	tx = dmu_tx_create(zfsvfs->z_os);
1473 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1474 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1475 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
1476 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1477 		    0, SPA_MAXBLOCKSIZE);
1478 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1479 	if (error) {
1480 		zfs_dirent_unlock(dl);
1481 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1482 			dmu_tx_wait(tx);
1483 			dmu_tx_abort(tx);
1484 			goto top;
1485 		}
1486 		dmu_tx_abort(tx);
1487 		ZFS_EXIT(zfsvfs);
1488 		return (error);
1489 	}
1490 
1491 	/*
1492 	 * Create new node.
1493 	 */
1494 	zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
1495 
1496 	/*
1497 	 * Now put new name in parent dir.
1498 	 */
1499 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1500 
1501 	*vpp = ZTOV(zp);
1502 
1503 	zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname);
1504 	dmu_tx_commit(tx);
1505 
1506 	zfs_dirent_unlock(dl);
1507 
1508 	ZFS_EXIT(zfsvfs);
1509 	return (0);
1510 }
1511 
1512 /*
1513  * Remove a directory subdir entry.  If the current working
1514  * directory is the same as the subdir to be removed, the
1515  * remove will fail.
1516  *
1517  *	IN:	dvp	- vnode of directory to remove from.
1518  *		name	- name of directory to be removed.
1519  *		cwd	- vnode of current working directory.
1520  *		cr	- credentials of caller.
1521  *
1522  *	RETURN:	0 if success
1523  *		error code if failure
1524  *
1525  * Timestamps:
1526  *	dvp - ctime|mtime updated
1527  */
1528 static int
1529 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr)
1530 {
1531 	znode_t		*dzp = VTOZ(dvp);
1532 	znode_t		*zp;
1533 	vnode_t		*vp;
1534 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1535 	zilog_t		*zilog = zfsvfs->z_log;
1536 	zfs_dirlock_t	*dl;
1537 	dmu_tx_t	*tx;
1538 	int		error;
1539 
1540 	ZFS_ENTER(zfsvfs);
1541 
1542 top:
1543 	zp = NULL;
1544 
1545 	/*
1546 	 * Attempt to lock directory; fail if entry doesn't exist.
1547 	 */
1548 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) {
1549 		ZFS_EXIT(zfsvfs);
1550 		return (error);
1551 	}
1552 
1553 	vp = ZTOV(zp);
1554 
1555 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1556 		goto out;
1557 	}
1558 
1559 	if (vp->v_type != VDIR) {
1560 		error = ENOTDIR;
1561 		goto out;
1562 	}
1563 
1564 	if (vp == cwd) {
1565 		error = EINVAL;
1566 		goto out;
1567 	}
1568 
1569 	vnevent_rmdir(vp);
1570 
1571 	/*
1572 	 * Grab a lock on the parent pointer make sure we play well
1573 	 * with the treewalk and directory rename code.
1574 	 */
1575 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1576 
1577 	tx = dmu_tx_create(zfsvfs->z_os);
1578 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1579 	dmu_tx_hold_bonus(tx, zp->z_id);
1580 	dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL);
1581 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1582 	if (error) {
1583 		rw_exit(&zp->z_parent_lock);
1584 		zfs_dirent_unlock(dl);
1585 		VN_RELE(vp);
1586 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1587 			dmu_tx_wait(tx);
1588 			dmu_tx_abort(tx);
1589 			goto top;
1590 		}
1591 		dmu_tx_abort(tx);
1592 		ZFS_EXIT(zfsvfs);
1593 		return (error);
1594 	}
1595 
1596 	error = zfs_link_destroy(dl, zp, tx, 0, NULL);
1597 
1598 	if (error == 0)
1599 		zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name);
1600 
1601 	dmu_tx_commit(tx);
1602 
1603 	rw_exit(&zp->z_parent_lock);
1604 out:
1605 	zfs_dirent_unlock(dl);
1606 
1607 	VN_RELE(vp);
1608 
1609 	ZFS_EXIT(zfsvfs);
1610 	return (error);
1611 }
1612 
1613 /*
1614  * Read as many directory entries as will fit into the provided
1615  * buffer from the given directory cursor position (specified in
1616  * the uio structure.
1617  *
1618  *	IN:	vp	- vnode of directory to read.
1619  *		uio	- structure supplying read location, range info,
1620  *			  and return buffer.
1621  *		cr	- credentials of caller.
1622  *
1623  *	OUT:	uio	- updated offset and range, buffer filled.
1624  *		eofp	- set to true if end-of-file detected.
1625  *
1626  *	RETURN:	0 if success
1627  *		error code if failure
1628  *
1629  * Timestamps:
1630  *	vp - atime updated
1631  *
1632  * Note that the low 4 bits of the cookie returned by zap is always zero.
1633  * This allows us to use the low range for "special" directory entries:
1634  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1635  * we use the offset 2 for the '.zfs' directory.
1636  */
1637 /* ARGSUSED */
1638 static int
1639 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp)
1640 {
1641 	znode_t		*zp = VTOZ(vp);
1642 	iovec_t		*iovp;
1643 	dirent64_t	*odp;
1644 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1645 	objset_t	*os;
1646 	caddr_t		outbuf;
1647 	size_t		bufsize;
1648 	zap_cursor_t	zc;
1649 	zap_attribute_t	zap;
1650 	uint_t		bytes_wanted;
1651 	ushort_t	this_reclen;
1652 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1653 	off64_t		*next;
1654 	int		local_eof;
1655 	int		outcount;
1656 	int		error;
1657 	uint8_t		prefetch;
1658 
1659 	ZFS_ENTER(zfsvfs);
1660 
1661 	/*
1662 	 * If we are not given an eof variable,
1663 	 * use a local one.
1664 	 */
1665 	if (eofp == NULL)
1666 		eofp = &local_eof;
1667 
1668 	/*
1669 	 * Check for valid iov_len.
1670 	 */
1671 	if (uio->uio_iov->iov_len <= 0) {
1672 		ZFS_EXIT(zfsvfs);
1673 		return (EINVAL);
1674 	}
1675 
1676 	/*
1677 	 * Quit if directory has been removed (posix)
1678 	 */
1679 	if ((*eofp = zp->z_reap) != 0) {
1680 		ZFS_EXIT(zfsvfs);
1681 		return (0);
1682 	}
1683 
1684 	error = 0;
1685 	os = zfsvfs->z_os;
1686 	offset = uio->uio_loffset;
1687 	prefetch = zp->z_zn_prefetch;
1688 
1689 	/*
1690 	 * Initialize the iterator cursor.
1691 	 */
1692 	if (offset <= 3) {
1693 		/*
1694 		 * Start iteration from the beginning of the directory.
1695 		 */
1696 		zap_cursor_init(&zc, os, zp->z_id);
1697 	} else {
1698 		/*
1699 		 * The offset is a serialized cursor.
1700 		 */
1701 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1702 	}
1703 
1704 	/*
1705 	 * Get space to change directory entries into fs independent format.
1706 	 */
1707 	iovp = uio->uio_iov;
1708 	bytes_wanted = iovp->iov_len;
1709 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
1710 		bufsize = bytes_wanted;
1711 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
1712 		odp = (struct dirent64 *)outbuf;
1713 	} else {
1714 		bufsize = bytes_wanted;
1715 		odp = (struct dirent64 *)iovp->iov_base;
1716 	}
1717 
1718 	/*
1719 	 * Transform to file-system independent format
1720 	 */
1721 	outcount = 0;
1722 	while (outcount < bytes_wanted) {
1723 		/*
1724 		 * Special case `.', `..', and `.zfs'.
1725 		 */
1726 		if (offset == 0) {
1727 			(void) strcpy(zap.za_name, ".");
1728 			zap.za_first_integer = zp->z_id;
1729 			this_reclen = DIRENT64_RECLEN(1);
1730 		} else if (offset == 1) {
1731 			(void) strcpy(zap.za_name, "..");
1732 			zap.za_first_integer = zp->z_phys->zp_parent;
1733 			this_reclen = DIRENT64_RECLEN(2);
1734 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
1735 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1736 			zap.za_first_integer = ZFSCTL_INO_ROOT;
1737 			this_reclen =
1738 			    DIRENT64_RECLEN(sizeof (ZFS_CTLDIR_NAME) - 1);
1739 		} else {
1740 			/*
1741 			 * Grab next entry.
1742 			 */
1743 			if (error = zap_cursor_retrieve(&zc, &zap)) {
1744 				if ((*eofp = (error == ENOENT)) != 0)
1745 					break;
1746 				else
1747 					goto update;
1748 			}
1749 
1750 			if (zap.za_integer_length != 8 ||
1751 			    zap.za_num_integers != 1) {
1752 				cmn_err(CE_WARN, "zap_readdir: bad directory "
1753 				    "entry, obj = %lld, offset = %lld\n",
1754 				    (u_longlong_t)zp->z_id,
1755 				    (u_longlong_t)offset);
1756 				error = ENXIO;
1757 				goto update;
1758 			}
1759 			this_reclen = DIRENT64_RECLEN(strlen(zap.za_name));
1760 		}
1761 
1762 		/*
1763 		 * Will this entry fit in the buffer?
1764 		 */
1765 		if (outcount + this_reclen > bufsize) {
1766 			/*
1767 			 * Did we manage to fit anything in the buffer?
1768 			 */
1769 			if (!outcount) {
1770 				error = EINVAL;
1771 				goto update;
1772 			}
1773 			break;
1774 		}
1775 		/*
1776 		 * Add this entry:
1777 		 */
1778 		odp->d_ino = (ino64_t)zap.za_first_integer;
1779 		odp->d_reclen = (ushort_t)this_reclen;
1780 		/* NOTE: d_off is the offset for the *next* entry */
1781 		next = &(odp->d_off);
1782 		(void) strncpy(odp->d_name, zap.za_name,
1783 		    DIRENT64_NAMELEN(this_reclen));
1784 		outcount += this_reclen;
1785 		odp = (dirent64_t *)((intptr_t)odp + this_reclen);
1786 
1787 		ASSERT(outcount <= bufsize);
1788 
1789 		/* Prefetch znode */
1790 		if (prefetch)
1791 			dmu_prefetch(os, zap.za_first_integer, 0, 0);
1792 
1793 		/*
1794 		 * Move to the next entry, fill in the previous offset.
1795 		 */
1796 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
1797 			zap_cursor_advance(&zc);
1798 			offset = zap_cursor_serialize(&zc);
1799 		} else {
1800 			offset += 1;
1801 		}
1802 		*next = offset;
1803 	}
1804 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
1805 
1806 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
1807 		iovp->iov_base += outcount;
1808 		iovp->iov_len -= outcount;
1809 		uio->uio_resid -= outcount;
1810 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
1811 		/*
1812 		 * Reset the pointer.
1813 		 */
1814 		offset = uio->uio_loffset;
1815 	}
1816 
1817 update:
1818 	zap_cursor_fini(&zc);
1819 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
1820 		kmem_free(outbuf, bufsize);
1821 
1822 	if (error == ENOENT)
1823 		error = 0;
1824 
1825 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
1826 
1827 	uio->uio_loffset = offset;
1828 	ZFS_EXIT(zfsvfs);
1829 	return (error);
1830 }
1831 
1832 static int
1833 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr)
1834 {
1835 	znode_t	*zp = VTOZ(vp);
1836 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1837 
1838 	/*
1839 	 * Regardless of whether this is required for standards conformance,
1840 	 * this is the logical behavior when fsync() is called on a file with
1841 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
1842 	 * going to be pushed out as part of the zil_commit().
1843 	 */
1844 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
1845 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
1846 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr);
1847 
1848 	ZFS_ENTER(zfsvfs);
1849 	zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
1850 	ZFS_EXIT(zfsvfs);
1851 	return (0);
1852 }
1853 
1854 /*
1855  * Get the requested file attributes and place them in the provided
1856  * vattr structure.
1857  *
1858  *	IN:	vp	- vnode of file.
1859  *		vap	- va_mask identifies requested attributes.
1860  *		flags	- [UNUSED]
1861  *		cr	- credentials of caller.
1862  *
1863  *	OUT:	vap	- attribute values.
1864  *
1865  *	RETURN:	0 (always succeeds)
1866  */
1867 /* ARGSUSED */
1868 static int
1869 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr)
1870 {
1871 	znode_t *zp = VTOZ(vp);
1872 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1873 	znode_phys_t *pzp = zp->z_phys;
1874 	int	error;
1875 
1876 	ZFS_ENTER(zfsvfs);
1877 
1878 	/*
1879 	 * Return all attributes.  It's cheaper to provide the answer
1880 	 * than to determine whether we were asked the question.
1881 	 */
1882 	mutex_enter(&zp->z_lock);
1883 
1884 	vap->va_type = vp->v_type;
1885 	vap->va_mode = pzp->zp_mode & MODEMASK;
1886 	vap->va_uid = zp->z_phys->zp_uid;
1887 	vap->va_gid = zp->z_phys->zp_gid;
1888 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
1889 	vap->va_nodeid = zp->z_id;
1890 	vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX);	/* nlink_t limit! */
1891 	vap->va_size = pzp->zp_size;
1892 	vap->va_rdev = vp->v_rdev;
1893 	vap->va_seq = zp->z_seq;
1894 
1895 	ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime);
1896 	ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime);
1897 	ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime);
1898 
1899 	/*
1900 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
1901 	 * Also, if we are the owner don't bother, since owner should
1902 	 * always be allowed to read basic attributes of file.
1903 	 */
1904 	if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) &&
1905 	    (zp->z_phys->zp_uid != crgetuid(cr))) {
1906 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) {
1907 			mutex_exit(&zp->z_lock);
1908 			ZFS_EXIT(zfsvfs);
1909 			return (error);
1910 		}
1911 	}
1912 
1913 	mutex_exit(&zp->z_lock);
1914 
1915 	dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks);
1916 
1917 	if (zp->z_blksz == 0) {
1918 		/*
1919 		 * Block size hasn't been set; suggest maximal I/O transfers.
1920 		 */
1921 		vap->va_blksize = zfsvfs->z_max_blksz;
1922 	}
1923 
1924 	ZFS_EXIT(zfsvfs);
1925 	return (0);
1926 }
1927 
1928 /*
1929  * Set the file attributes to the values contained in the
1930  * vattr structure.
1931  *
1932  *	IN:	vp	- vnode of file to be modified.
1933  *		vap	- new attribute values.
1934  *		flags	- ATTR_UTIME set if non-default time values provided.
1935  *		cr	- credentials of caller.
1936  *
1937  *	RETURN:	0 if success
1938  *		error code if failure
1939  *
1940  * Timestamps:
1941  *	vp - ctime updated, mtime updated if size changed.
1942  */
1943 /* ARGSUSED */
1944 static int
1945 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1946 	caller_context_t *ct)
1947 {
1948 	struct znode	*zp = VTOZ(vp);
1949 	znode_phys_t	*pzp = zp->z_phys;
1950 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1951 	zilog_t		*zilog = zfsvfs->z_log;
1952 	dmu_tx_t	*tx;
1953 	vattr_t		oldva;
1954 	uint_t		mask = vap->va_mask;
1955 	uint_t		saved_mask;
1956 	int		trim_mask = 0;
1957 	uint64_t	new_mode;
1958 	znode_t		*attrzp;
1959 	int		need_policy = FALSE;
1960 	int		err;
1961 
1962 	if (mask == 0)
1963 		return (0);
1964 
1965 	if (mask & AT_NOSET)
1966 		return (EINVAL);
1967 
1968 	if (mask & AT_SIZE && vp->v_type == VDIR)
1969 		return (EISDIR);
1970 
1971 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO)
1972 		return (EINVAL);
1973 
1974 	ZFS_ENTER(zfsvfs);
1975 
1976 top:
1977 	attrzp = NULL;
1978 
1979 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
1980 		ZFS_EXIT(zfsvfs);
1981 		return (EROFS);
1982 	}
1983 
1984 	/*
1985 	 * First validate permissions
1986 	 */
1987 
1988 	if (mask & AT_SIZE) {
1989 		err = zfs_zaccess(zp, ACE_WRITE_DATA, cr);
1990 		if (err) {
1991 			ZFS_EXIT(zfsvfs);
1992 			return (err);
1993 		}
1994 		/*
1995 		 * XXX - Note, we are not providing any open
1996 		 * mode flags here (like FNDELAY), so we may
1997 		 * block if there are locks present... this
1998 		 * should be addressed in openat().
1999 		 */
2000 		do {
2001 			err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2002 			/* NB: we already did dmu_tx_wait() if necessary */
2003 		} while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
2004 		if (err) {
2005 			ZFS_EXIT(zfsvfs);
2006 			return (err);
2007 		}
2008 	}
2009 
2010 	if (mask & (AT_ATIME|AT_MTIME))
2011 		need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr);
2012 
2013 	if (mask & (AT_UID|AT_GID)) {
2014 		int	idmask = (mask & (AT_UID|AT_GID));
2015 		int	take_owner;
2016 		int	take_group;
2017 
2018 		/*
2019 		 * NOTE: even if a new mode is being set,
2020 		 * we may clear S_ISUID/S_ISGID bits.
2021 		 */
2022 
2023 		if (!(mask & AT_MODE))
2024 			vap->va_mode = pzp->zp_mode;
2025 
2026 		/*
2027 		 * Take ownership or chgrp to group we are a member of
2028 		 */
2029 
2030 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2031 		take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr);
2032 
2033 		/*
2034 		 * If both AT_UID and AT_GID are set then take_owner and
2035 		 * take_group must both be set in order to allow taking
2036 		 * ownership.
2037 		 *
2038 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2039 		 *
2040 		 */
2041 
2042 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2043 		    ((idmask == AT_UID) && take_owner) ||
2044 		    ((idmask == AT_GID) && take_group)) {
2045 			if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) {
2046 				/*
2047 				 * Remove setuid/setgid for non-privileged users
2048 				 */
2049 				secpolicy_setid_clear(vap, cr);
2050 				trim_mask = (mask & (AT_UID|AT_GID));
2051 			} else {
2052 				need_policy =  TRUE;
2053 			}
2054 		} else {
2055 			need_policy =  TRUE;
2056 		}
2057 	}
2058 
2059 	mutex_enter(&zp->z_lock);
2060 	oldva.va_mode = pzp->zp_mode;
2061 	oldva.va_uid = zp->z_phys->zp_uid;
2062 	oldva.va_gid = zp->z_phys->zp_gid;
2063 	mutex_exit(&zp->z_lock);
2064 
2065 	if (mask & AT_MODE) {
2066 		if (zfs_zaccess_v4_perm(zp, ACE_WRITE_ACL, cr) == 0) {
2067 			err = secpolicy_setid_setsticky_clear(vp, vap,
2068 			    &oldva, cr);
2069 			if (err) {
2070 				ZFS_EXIT(zfsvfs);
2071 				return (err);
2072 			}
2073 			trim_mask |= AT_MODE;
2074 		} else {
2075 			need_policy = TRUE;
2076 		}
2077 	}
2078 
2079 	if (need_policy) {
2080 		/*
2081 		 * If trim_mask is set then take ownership
2082 		 * has been granted or write_acl is present and user
2083 		 * has the ability to modify mode.  In that case remove
2084 		 * UID|GID and or MODE from mask so that
2085 		 * secpolicy_vnode_setattr() doesn't revoke it.
2086 		 */
2087 
2088 		if (trim_mask) {
2089 			saved_mask = vap->va_mask;
2090 			vap->va_mask &= ~trim_mask;
2091 
2092 		}
2093 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2094 		    (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp);
2095 		if (err) {
2096 			ZFS_EXIT(zfsvfs);
2097 			return (err);
2098 		}
2099 
2100 		if (trim_mask)
2101 			vap->va_mask |= saved_mask;
2102 	}
2103 
2104 	/*
2105 	 * secpolicy_vnode_setattr, or take ownership may have
2106 	 * changed va_mask
2107 	 */
2108 	mask = vap->va_mask;
2109 
2110 	tx = dmu_tx_create(zfsvfs->z_os);
2111 	dmu_tx_hold_bonus(tx, zp->z_id);
2112 
2113 	if (mask & AT_MODE) {
2114 		uint64_t pmode = pzp->zp_mode;
2115 
2116 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2117 
2118 		if (zp->z_phys->zp_acl.z_acl_extern_obj)
2119 			dmu_tx_hold_write(tx,
2120 			    pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE);
2121 		else
2122 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2123 			    0, ZFS_ACL_SIZE(MAX_ACL_SIZE));
2124 	}
2125 
2126 	if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) {
2127 		err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp);
2128 		if (err) {
2129 			dmu_tx_abort(tx);
2130 			ZFS_EXIT(zfsvfs);
2131 			return (err);
2132 		}
2133 		dmu_tx_hold_bonus(tx, attrzp->z_id);
2134 	}
2135 
2136 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2137 	if (err) {
2138 		if (attrzp)
2139 			VN_RELE(ZTOV(attrzp));
2140 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2141 			dmu_tx_wait(tx);
2142 			dmu_tx_abort(tx);
2143 			goto top;
2144 		}
2145 		dmu_tx_abort(tx);
2146 		ZFS_EXIT(zfsvfs);
2147 		return (err);
2148 	}
2149 
2150 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2151 
2152 	/*
2153 	 * Set each attribute requested.
2154 	 * We group settings according to the locks they need to acquire.
2155 	 *
2156 	 * Note: you cannot set ctime directly, although it will be
2157 	 * updated as a side-effect of calling this function.
2158 	 */
2159 
2160 	mutex_enter(&zp->z_lock);
2161 
2162 	if (mask & AT_MODE) {
2163 		err = zfs_acl_chmod_setattr(zp, new_mode, tx);
2164 		ASSERT3U(err, ==, 0);
2165 	}
2166 
2167 	if (attrzp)
2168 		mutex_enter(&attrzp->z_lock);
2169 
2170 	if (mask & AT_UID) {
2171 		zp->z_phys->zp_uid = (uint64_t)vap->va_uid;
2172 		if (attrzp) {
2173 			attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid;
2174 		}
2175 	}
2176 
2177 	if (mask & AT_GID) {
2178 		zp->z_phys->zp_gid = (uint64_t)vap->va_gid;
2179 		if (attrzp)
2180 			attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid;
2181 	}
2182 
2183 	if (attrzp)
2184 		mutex_exit(&attrzp->z_lock);
2185 
2186 	if (mask & AT_ATIME)
2187 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2188 
2189 	if (mask & AT_MTIME)
2190 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2191 
2192 	if (mask & AT_SIZE)
2193 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2194 	else if (mask != 0)
2195 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2196 
2197 	if (mask != 0)
2198 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask);
2199 
2200 	mutex_exit(&zp->z_lock);
2201 
2202 	if (attrzp)
2203 		VN_RELE(ZTOV(attrzp));
2204 
2205 	dmu_tx_commit(tx);
2206 
2207 	ZFS_EXIT(zfsvfs);
2208 	return (err);
2209 }
2210 
2211 /*
2212  * Search back through the directory tree, using the ".." entries.
2213  * Lock each directory in the chain to prevent concurrent renames.
2214  * Fail any attempt to move a directory into one of its own descendants.
2215  * XXX - z_parent_lock can overlap with map or grow locks
2216  */
2217 typedef struct zfs_zlock {
2218 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2219 	znode_t		*zl_znode;	/* znode we held */
2220 	struct zfs_zlock *zl_next;	/* next in list */
2221 } zfs_zlock_t;
2222 
2223 static int
2224 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2225 {
2226 	zfs_zlock_t	*zl;
2227 	znode_t 	*zp = tdzp;
2228 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2229 	uint64_t	*oidp = &zp->z_id;
2230 	krwlock_t	*rwlp = &szp->z_parent_lock;
2231 	krw_t		rw = RW_WRITER;
2232 
2233 	/*
2234 	 * First pass write-locks szp and compares to zp->z_id.
2235 	 * Later passes read-lock zp and compare to zp->z_parent.
2236 	 */
2237 	do {
2238 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2239 		zl->zl_rwlock = rwlp;
2240 		zl->zl_znode = NULL;
2241 		zl->zl_next = *zlpp;
2242 		*zlpp = zl;
2243 
2244 		rw_enter(rwlp, rw);
2245 
2246 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2247 			return (EINVAL);
2248 
2249 		if (*oidp == rootid)		/* We've hit the top */
2250 			return (0);
2251 
2252 		if (rw == RW_READER) {		/* i.e. not the first pass */
2253 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2254 			if (error)
2255 				return (error);
2256 			zl->zl_znode = zp;
2257 		}
2258 		oidp = &zp->z_phys->zp_parent;
2259 		rwlp = &zp->z_parent_lock;
2260 		rw = RW_READER;
2261 
2262 	} while (zp->z_id != sdzp->z_id);
2263 
2264 	return (0);
2265 }
2266 
2267 /*
2268  * Drop locks and release vnodes that were held by zfs_rename_lock().
2269  */
2270 static void
2271 zfs_rename_unlock(zfs_zlock_t **zlpp)
2272 {
2273 	zfs_zlock_t *zl;
2274 
2275 	while ((zl = *zlpp) != NULL) {
2276 		if (zl->zl_znode != NULL)
2277 			VN_RELE(ZTOV(zl->zl_znode));
2278 		rw_exit(zl->zl_rwlock);
2279 		*zlpp = zl->zl_next;
2280 		kmem_free(zl, sizeof (*zl));
2281 	}
2282 }
2283 
2284 /*
2285  * Move an entry from the provided source directory to the target
2286  * directory.  Change the entry name as indicated.
2287  *
2288  *	IN:	sdvp	- Source directory containing the "old entry".
2289  *		snm	- Old entry name.
2290  *		tdvp	- Target directory to contain the "new entry".
2291  *		tnm	- New entry name.
2292  *		cr	- credentials of caller.
2293  *
2294  *	RETURN:	0 if success
2295  *		error code if failure
2296  *
2297  * Timestamps:
2298  *	sdvp,tdvp - ctime|mtime updated
2299  */
2300 static int
2301 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr)
2302 {
2303 	znode_t		*tdzp, *szp, *tzp;
2304 	znode_t		*sdzp = VTOZ(sdvp);
2305 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2306 	zilog_t		*zilog = zfsvfs->z_log;
2307 	vnode_t		*realvp;
2308 	zfs_dirlock_t	*sdl, *tdl;
2309 	dmu_tx_t	*tx;
2310 	zfs_zlock_t	*zl;
2311 	int		cmp, serr, terr, error;
2312 
2313 	ZFS_ENTER(zfsvfs);
2314 
2315 	/*
2316 	 * Make sure we have the real vp for the target directory.
2317 	 */
2318 	if (VOP_REALVP(tdvp, &realvp) == 0)
2319 		tdvp = realvp;
2320 
2321 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
2322 		ZFS_EXIT(zfsvfs);
2323 		return (EXDEV);
2324 	}
2325 
2326 	tdzp = VTOZ(tdvp);
2327 top:
2328 	szp = NULL;
2329 	tzp = NULL;
2330 	zl = NULL;
2331 
2332 	/*
2333 	 * This is to prevent the creation of links into attribute space
2334 	 * by renaming a linked file into/outof an attribute directory.
2335 	 * See the comment in zfs_link() for why this is considered bad.
2336 	 */
2337 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
2338 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
2339 		ZFS_EXIT(zfsvfs);
2340 		return (EINVAL);
2341 	}
2342 
2343 	/*
2344 	 * Lock source and target directory entries.  To prevent deadlock,
2345 	 * a lock ordering must be defined.  We lock the directory with
2346 	 * the smallest object id first, or if it's a tie, the one with
2347 	 * the lexically first name.
2348 	 */
2349 	if (sdzp->z_id < tdzp->z_id) {
2350 		cmp = -1;
2351 	} else if (sdzp->z_id > tdzp->z_id) {
2352 		cmp = 1;
2353 	} else {
2354 		cmp = strcmp(snm, tnm);
2355 		if (cmp == 0) {
2356 			/*
2357 			 * POSIX: "If the old argument and the new argument
2358 			 * both refer to links to the same existing file,
2359 			 * the rename() function shall return successfully
2360 			 * and perform no other action."
2361 			 */
2362 			ZFS_EXIT(zfsvfs);
2363 			return (0);
2364 		}
2365 	}
2366 	if (cmp < 0) {
2367 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS);
2368 		terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0);
2369 	} else {
2370 		terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0);
2371 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS);
2372 	}
2373 
2374 	if (serr) {
2375 		/*
2376 		 * Source entry invalid or not there.
2377 		 */
2378 		if (!terr) {
2379 			zfs_dirent_unlock(tdl);
2380 			if (tzp)
2381 				VN_RELE(ZTOV(tzp));
2382 		}
2383 		if (strcmp(snm, "..") == 0)
2384 			serr = EINVAL;
2385 		ZFS_EXIT(zfsvfs);
2386 		return (serr);
2387 	}
2388 	if (terr) {
2389 		zfs_dirent_unlock(sdl);
2390 		VN_RELE(ZTOV(szp));
2391 		if (strcmp(tnm, "..") == 0)
2392 			terr = EINVAL;
2393 		ZFS_EXIT(zfsvfs);
2394 		return (terr);
2395 	}
2396 
2397 	/*
2398 	 * Must have write access at the source to remove the old entry
2399 	 * and write access at the target to create the new entry.
2400 	 * Note that if target and source are the same, this can be
2401 	 * done in a single check.
2402 	 */
2403 
2404 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
2405 		goto out;
2406 
2407 	if (ZTOV(szp)->v_type == VDIR) {
2408 		/*
2409 		 * Check to make sure rename is valid.
2410 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
2411 		 */
2412 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
2413 			goto out;
2414 	}
2415 
2416 	/*
2417 	 * Does target exist?
2418 	 */
2419 	if (tzp) {
2420 		/*
2421 		 * Source and target must be the same type.
2422 		 */
2423 		if (ZTOV(szp)->v_type == VDIR) {
2424 			if (ZTOV(tzp)->v_type != VDIR) {
2425 				error = ENOTDIR;
2426 				goto out;
2427 			}
2428 		} else {
2429 			if (ZTOV(tzp)->v_type == VDIR) {
2430 				error = EISDIR;
2431 				goto out;
2432 			}
2433 		}
2434 		/*
2435 		 * POSIX dictates that when the source and target
2436 		 * entries refer to the same file object, rename
2437 		 * must do nothing and exit without error.
2438 		 */
2439 		if (szp->z_id == tzp->z_id) {
2440 			error = 0;
2441 			goto out;
2442 		}
2443 	}
2444 
2445 	vnevent_rename_src(ZTOV(szp));
2446 	if (tzp)
2447 		vnevent_rename_dest(ZTOV(tzp));
2448 
2449 	tx = dmu_tx_create(zfsvfs->z_os);
2450 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
2451 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
2452 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
2453 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
2454 	if (sdzp != tdzp)
2455 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
2456 	if (tzp)
2457 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* parent changes */
2458 	dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL);
2459 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2460 	if (error) {
2461 		if (zl != NULL)
2462 			zfs_rename_unlock(&zl);
2463 		zfs_dirent_unlock(sdl);
2464 		zfs_dirent_unlock(tdl);
2465 		VN_RELE(ZTOV(szp));
2466 		if (tzp)
2467 			VN_RELE(ZTOV(tzp));
2468 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2469 			dmu_tx_wait(tx);
2470 			dmu_tx_abort(tx);
2471 			goto top;
2472 		}
2473 		dmu_tx_abort(tx);
2474 		ZFS_EXIT(zfsvfs);
2475 		return (error);
2476 	}
2477 
2478 	if (tzp)	/* Attempt to remove the existing target */
2479 		error = zfs_link_destroy(tdl, tzp, tx, 0, NULL);
2480 
2481 	if (error == 0) {
2482 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
2483 		if (error == 0) {
2484 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
2485 			ASSERT(error == 0);
2486 			zfs_log_rename(zilog, tx, TX_RENAME, sdzp,
2487 			    sdl->dl_name, tdzp, tdl->dl_name, szp);
2488 		}
2489 	}
2490 
2491 	dmu_tx_commit(tx);
2492 out:
2493 	if (zl != NULL)
2494 		zfs_rename_unlock(&zl);
2495 
2496 	zfs_dirent_unlock(sdl);
2497 	zfs_dirent_unlock(tdl);
2498 
2499 	VN_RELE(ZTOV(szp));
2500 	if (tzp)
2501 		VN_RELE(ZTOV(tzp));
2502 
2503 	ZFS_EXIT(zfsvfs);
2504 	return (error);
2505 }
2506 
2507 /*
2508  * Insert the indicated symbolic reference entry into the directory.
2509  *
2510  *	IN:	dvp	- Directory to contain new symbolic link.
2511  *		link	- Name for new symlink entry.
2512  *		vap	- Attributes of new entry.
2513  *		target	- Target path of new symlink.
2514  *		cr	- credentials of caller.
2515  *
2516  *	RETURN:	0 if success
2517  *		error code if failure
2518  *
2519  * Timestamps:
2520  *	dvp - ctime|mtime updated
2521  */
2522 static int
2523 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr)
2524 {
2525 	znode_t		*zp, *dzp = VTOZ(dvp);
2526 	zfs_dirlock_t	*dl;
2527 	dmu_tx_t	*tx;
2528 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2529 	zilog_t		*zilog = zfsvfs->z_log;
2530 	uint64_t	zoid;
2531 	int		len = strlen(link);
2532 	int		error;
2533 
2534 	ASSERT(vap->va_type == VLNK);
2535 
2536 	ZFS_ENTER(zfsvfs);
2537 top:
2538 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
2539 		ZFS_EXIT(zfsvfs);
2540 		return (error);
2541 	}
2542 
2543 	if (len > MAXPATHLEN) {
2544 		ZFS_EXIT(zfsvfs);
2545 		return (ENAMETOOLONG);
2546 	}
2547 
2548 	/*
2549 	 * Attempt to lock directory; fail if entry already exists.
2550 	 */
2551 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) {
2552 		ZFS_EXIT(zfsvfs);
2553 		return (error);
2554 	}
2555 
2556 	tx = dmu_tx_create(zfsvfs->z_os);
2557 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
2558 	dmu_tx_hold_bonus(tx, dzp->z_id);
2559 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
2560 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
2561 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
2562 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2563 	if (error) {
2564 		zfs_dirent_unlock(dl);
2565 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2566 			dmu_tx_wait(tx);
2567 			dmu_tx_abort(tx);
2568 			goto top;
2569 		}
2570 		dmu_tx_abort(tx);
2571 		ZFS_EXIT(zfsvfs);
2572 		return (error);
2573 	}
2574 
2575 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
2576 
2577 	/*
2578 	 * Create a new object for the symlink.
2579 	 * Put the link content into bonus buffer if it will fit;
2580 	 * otherwise, store it just like any other file data.
2581 	 */
2582 	zoid = 0;
2583 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
2584 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len);
2585 		if (len != 0)
2586 			bcopy(link, zp->z_phys + 1, len);
2587 	} else {
2588 		dmu_buf_t *dbp;
2589 
2590 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
2591 
2592 		/*
2593 		 * Nothing can access the znode yet so no locking needed
2594 		 * for growing the znode's blocksize.
2595 		 */
2596 		zfs_grow_blocksize(zp, len, tx);
2597 
2598 		VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp));
2599 		dmu_buf_will_dirty(dbp, tx);
2600 
2601 		ASSERT3U(len, <=, dbp->db_size);
2602 		bcopy(link, dbp->db_data, len);
2603 		dmu_buf_rele(dbp, FTAG);
2604 	}
2605 	zp->z_phys->zp_size = len;
2606 
2607 	/*
2608 	 * Insert the new object into the directory.
2609 	 */
2610 	(void) zfs_link_create(dl, zp, tx, ZNEW);
2611 out:
2612 	if (error == 0)
2613 		zfs_log_symlink(zilog, tx, TX_SYMLINK, dzp, zp, name, link);
2614 
2615 	dmu_tx_commit(tx);
2616 
2617 	zfs_dirent_unlock(dl);
2618 
2619 	VN_RELE(ZTOV(zp));
2620 
2621 	ZFS_EXIT(zfsvfs);
2622 	return (error);
2623 }
2624 
2625 /*
2626  * Return, in the buffer contained in the provided uio structure,
2627  * the symbolic path referred to by vp.
2628  *
2629  *	IN:	vp	- vnode of symbolic link.
2630  *		uoip	- structure to contain the link path.
2631  *		cr	- credentials of caller.
2632  *
2633  *	OUT:	uio	- structure to contain the link path.
2634  *
2635  *	RETURN:	0 if success
2636  *		error code if failure
2637  *
2638  * Timestamps:
2639  *	vp - atime updated
2640  */
2641 /* ARGSUSED */
2642 static int
2643 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr)
2644 {
2645 	znode_t		*zp = VTOZ(vp);
2646 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2647 	size_t		bufsz;
2648 	int		error;
2649 
2650 	ZFS_ENTER(zfsvfs);
2651 
2652 	bufsz = (size_t)zp->z_phys->zp_size;
2653 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
2654 		error = uiomove(zp->z_phys + 1,
2655 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
2656 	} else {
2657 		dmu_buf_t *dbp;
2658 		error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp);
2659 		if (error) {
2660 			ZFS_EXIT(zfsvfs);
2661 			return (error);
2662 		}
2663 		error = uiomove(dbp->db_data,
2664 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
2665 		dmu_buf_rele(dbp, FTAG);
2666 	}
2667 
2668 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2669 	ZFS_EXIT(zfsvfs);
2670 	return (error);
2671 }
2672 
2673 /*
2674  * Insert a new entry into directory tdvp referencing svp.
2675  *
2676  *	IN:	tdvp	- Directory to contain new entry.
2677  *		svp	- vnode of new entry.
2678  *		name	- name of new entry.
2679  *		cr	- credentials of caller.
2680  *
2681  *	RETURN:	0 if success
2682  *		error code if failure
2683  *
2684  * Timestamps:
2685  *	tdvp - ctime|mtime updated
2686  *	 svp - ctime updated
2687  */
2688 /* ARGSUSED */
2689 static int
2690 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr)
2691 {
2692 	znode_t		*dzp = VTOZ(tdvp);
2693 	znode_t		*tzp, *szp;
2694 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2695 	zilog_t		*zilog = zfsvfs->z_log;
2696 	zfs_dirlock_t	*dl;
2697 	dmu_tx_t	*tx;
2698 	vnode_t		*realvp;
2699 	int		error;
2700 
2701 	ASSERT(tdvp->v_type == VDIR);
2702 
2703 	ZFS_ENTER(zfsvfs);
2704 
2705 	if (VOP_REALVP(svp, &realvp) == 0)
2706 		svp = realvp;
2707 
2708 	if (svp->v_vfsp != tdvp->v_vfsp) {
2709 		ZFS_EXIT(zfsvfs);
2710 		return (EXDEV);
2711 	}
2712 
2713 	szp = VTOZ(svp);
2714 top:
2715 	/*
2716 	 * We do not support links between attributes and non-attributes
2717 	 * because of the potential security risk of creating links
2718 	 * into "normal" file space in order to circumvent restrictions
2719 	 * imposed in attribute space.
2720 	 */
2721 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
2722 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
2723 		ZFS_EXIT(zfsvfs);
2724 		return (EINVAL);
2725 	}
2726 
2727 	/*
2728 	 * POSIX dictates that we return EPERM here.
2729 	 * Better choices include ENOTSUP or EISDIR.
2730 	 */
2731 	if (svp->v_type == VDIR) {
2732 		ZFS_EXIT(zfsvfs);
2733 		return (EPERM);
2734 	}
2735 
2736 	if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) &&
2737 	    secpolicy_basic_link(cr) != 0) {
2738 		ZFS_EXIT(zfsvfs);
2739 		return (EPERM);
2740 	}
2741 
2742 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
2743 		ZFS_EXIT(zfsvfs);
2744 		return (error);
2745 	}
2746 
2747 	/*
2748 	 * Attempt to lock directory; fail if entry already exists.
2749 	 */
2750 	if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) {
2751 		ZFS_EXIT(zfsvfs);
2752 		return (error);
2753 	}
2754 
2755 	tx = dmu_tx_create(zfsvfs->z_os);
2756 	dmu_tx_hold_bonus(tx, szp->z_id);
2757 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
2758 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2759 	if (error) {
2760 		zfs_dirent_unlock(dl);
2761 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2762 			dmu_tx_wait(tx);
2763 			dmu_tx_abort(tx);
2764 			goto top;
2765 		}
2766 		dmu_tx_abort(tx);
2767 		ZFS_EXIT(zfsvfs);
2768 		return (error);
2769 	}
2770 
2771 	error = zfs_link_create(dl, szp, tx, 0);
2772 
2773 	if (error == 0)
2774 		zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name);
2775 
2776 	dmu_tx_commit(tx);
2777 
2778 	zfs_dirent_unlock(dl);
2779 
2780 	ZFS_EXIT(zfsvfs);
2781 	return (error);
2782 }
2783 
2784 /*
2785  * zfs_null_putapage() is used when the file system has been force
2786  * unmounted. It just drops the pages.
2787  */
2788 /* ARGSUSED */
2789 static int
2790 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
2791 		size_t *lenp, int flags, cred_t *cr)
2792 {
2793 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
2794 	return (0);
2795 }
2796 
2797 /*
2798  * Push a page out to disk, klustering if possible.
2799  *
2800  *	IN:	vp	- file to push page to.
2801  *		pp	- page to push.
2802  *		flags	- additional flags.
2803  *		cr	- credentials of caller.
2804  *
2805  *	OUT:	offp	- start of range pushed.
2806  *		lenp	- len of range pushed.
2807  *
2808  *	RETURN:	0 if success
2809  *		error code if failure
2810  *
2811  * NOTE: callers must have locked the page to be pushed.  On
2812  * exit, the page (and all other pages in the kluster) must be
2813  * unlocked.
2814  */
2815 /* ARGSUSED */
2816 static int
2817 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
2818 		size_t *lenp, int flags, cred_t *cr)
2819 {
2820 	znode_t		*zp = VTOZ(vp);
2821 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2822 	zilog_t		*zilog = zfsvfs->z_log;
2823 	dmu_tx_t	*tx;
2824 	rl_t		*rl;
2825 	u_offset_t	off, koff;
2826 	size_t		len, klen;
2827 	int		err;
2828 
2829 	off = pp->p_offset;
2830 	len = PAGESIZE;
2831 	/*
2832 	 * If our blocksize is bigger than the page size, try to kluster
2833 	 * muiltiple pages so that we write a full block (thus avoiding
2834 	 * a read-modify-write).
2835 	 */
2836 	if (zp->z_blksz > PAGESIZE) {
2837 		uint64_t filesz = zp->z_phys->zp_size;
2838 
2839 		if (!ISP2(zp->z_blksz)) {
2840 			/* Only one block in the file. */
2841 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
2842 			koff = 0;
2843 		} else {
2844 			klen = zp->z_blksz;
2845 			koff = P2ALIGN(off, (u_offset_t)klen);
2846 		}
2847 		ASSERT(koff <= filesz);
2848 		if (koff + klen > filesz)
2849 			klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE);
2850 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
2851 	}
2852 	ASSERT3U(btop(len), ==, btopr(len));
2853 top:
2854 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
2855 	/*
2856 	 * Can't push pages past end-of-file.
2857 	 */
2858 	if (off >= zp->z_phys->zp_size) {
2859 		/* discard all pages */
2860 		flags |= B_INVAL;
2861 		err = 0;
2862 		goto out;
2863 	} else if (off + len > zp->z_phys->zp_size) {
2864 		int npages = btopr(zp->z_phys->zp_size - off);
2865 		page_t *trunc;
2866 
2867 		page_list_break(&pp, &trunc, npages);
2868 		/* discard pages past end of file */
2869 		if (trunc)
2870 			pvn_write_done(trunc, B_INVAL | flags);
2871 		len = zp->z_phys->zp_size - off;
2872 	}
2873 
2874 	tx = dmu_tx_create(zfsvfs->z_os);
2875 	dmu_tx_hold_write(tx, zp->z_id, off, len);
2876 	dmu_tx_hold_bonus(tx, zp->z_id);
2877 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2878 	if (err != 0) {
2879 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2880 			zfs_range_unlock(rl);
2881 			dmu_tx_wait(tx);
2882 			dmu_tx_abort(tx);
2883 			err = 0;
2884 			goto top;
2885 		}
2886 		dmu_tx_abort(tx);
2887 		goto out;
2888 	}
2889 
2890 	if (zp->z_blksz <= PAGESIZE) {
2891 		caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1);
2892 		ASSERT3U(len, <=, PAGESIZE);
2893 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
2894 		ppmapout(va);
2895 	} else {
2896 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
2897 	}
2898 
2899 	if (err == 0) {
2900 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
2901 		(void) zfs_log_write(
2902 		    zilog, tx, TX_WRITE, zp, off, len, 0, NULL);
2903 		dmu_tx_commit(tx);
2904 	}
2905 
2906 out:
2907 	zfs_range_unlock(rl);
2908 	pvn_write_done(pp, (err ? B_ERROR : 0) | B_WRITE | flags);
2909 	if (offp)
2910 		*offp = off;
2911 	if (lenp)
2912 		*lenp = len;
2913 
2914 	return (err);
2915 }
2916 
2917 /*
2918  * Copy the portion of the file indicated from pages into the file.
2919  * The pages are stored in a page list attached to the files vnode.
2920  *
2921  *	IN:	vp	- vnode of file to push page data to.
2922  *		off	- position in file to put data.
2923  *		len	- amount of data to write.
2924  *		flags	- flags to control the operation.
2925  *		cr	- credentials of caller.
2926  *
2927  *	RETURN:	0 if success
2928  *		error code if failure
2929  *
2930  * Timestamps:
2931  *	vp - ctime|mtime updated
2932  */
2933 static int
2934 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr)
2935 {
2936 	znode_t		*zp = VTOZ(vp);
2937 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2938 	page_t		*pp;
2939 	size_t		io_len;
2940 	u_offset_t	io_off;
2941 	uint64_t	filesz;
2942 	int		error = 0;
2943 
2944 	ZFS_ENTER(zfsvfs);
2945 
2946 	ASSERT(zp->z_dbuf_held && zp->z_phys);
2947 
2948 	if (len == 0) {
2949 		/*
2950 		 * Search the entire vp list for pages >= off.
2951 		 */
2952 		error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage,
2953 		    flags, cr);
2954 		goto out;
2955 	}
2956 
2957 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
2958 	if (off > filesz) {
2959 		/* past end of file */
2960 		ZFS_EXIT(zfsvfs);
2961 		return (0);
2962 	}
2963 
2964 	len = MIN(len, filesz - off);
2965 
2966 	for (io_off = off; io_off < off + len; io_off += io_len) {
2967 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
2968 			pp = page_lookup(vp, io_off,
2969 				(flags & (B_INVAL | B_FREE)) ?
2970 					SE_EXCL : SE_SHARED);
2971 		} else {
2972 			pp = page_lookup_nowait(vp, io_off,
2973 				(flags & B_FREE) ? SE_EXCL : SE_SHARED);
2974 		}
2975 
2976 		if (pp != NULL && pvn_getdirty(pp, flags)) {
2977 			int err;
2978 
2979 			/*
2980 			 * Found a dirty page to push
2981 			 */
2982 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
2983 			if (err)
2984 				error = err;
2985 		} else {
2986 			io_len = PAGESIZE;
2987 		}
2988 	}
2989 out:
2990 	if ((flags & B_ASYNC) == 0)
2991 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
2992 	ZFS_EXIT(zfsvfs);
2993 	return (error);
2994 }
2995 
2996 void
2997 zfs_inactive(vnode_t *vp, cred_t *cr)
2998 {
2999 	znode_t	*zp = VTOZ(vp);
3000 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3001 	int error;
3002 
3003 	rw_enter(&zfsvfs->z_um_lock, RW_READER);
3004 	if (zfsvfs->z_unmounted2) {
3005 		ASSERT(zp->z_dbuf_held == 0);
3006 
3007 		if (vn_has_cached_data(vp)) {
3008 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
3009 			    B_INVAL, cr);
3010 		}
3011 
3012 		mutex_enter(&zp->z_lock);
3013 		vp->v_count = 0; /* count arrives as 1 */
3014 		if (zp->z_dbuf == NULL) {
3015 			mutex_exit(&zp->z_lock);
3016 			zfs_znode_free(zp);
3017 		} else {
3018 			mutex_exit(&zp->z_lock);
3019 		}
3020 		rw_exit(&zfsvfs->z_um_lock);
3021 		VFS_RELE(zfsvfs->z_vfs);
3022 		return;
3023 	}
3024 
3025 	/*
3026 	 * Attempt to push any data in the page cache.  If this fails
3027 	 * we will get kicked out later in zfs_zinactive().
3028 	 */
3029 	if (vn_has_cached_data(vp)) {
3030 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
3031 		    cr);
3032 	}
3033 
3034 	if (zp->z_atime_dirty && zp->z_reap == 0) {
3035 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3036 
3037 		dmu_tx_hold_bonus(tx, zp->z_id);
3038 		error = dmu_tx_assign(tx, TXG_WAIT);
3039 		if (error) {
3040 			dmu_tx_abort(tx);
3041 		} else {
3042 			dmu_buf_will_dirty(zp->z_dbuf, tx);
3043 			mutex_enter(&zp->z_lock);
3044 			zp->z_atime_dirty = 0;
3045 			mutex_exit(&zp->z_lock);
3046 			dmu_tx_commit(tx);
3047 		}
3048 	}
3049 
3050 	zfs_zinactive(zp);
3051 	rw_exit(&zfsvfs->z_um_lock);
3052 }
3053 
3054 /*
3055  * Bounds-check the seek operation.
3056  *
3057  *	IN:	vp	- vnode seeking within
3058  *		ooff	- old file offset
3059  *		noffp	- pointer to new file offset
3060  *
3061  *	RETURN:	0 if success
3062  *		EINVAL if new offset invalid
3063  */
3064 /* ARGSUSED */
3065 static int
3066 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp)
3067 {
3068 	if (vp->v_type == VDIR)
3069 		return (0);
3070 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3071 }
3072 
3073 /*
3074  * Pre-filter the generic locking function to trap attempts to place
3075  * a mandatory lock on a memory mapped file.
3076  */
3077 static int
3078 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3079     flk_callback_t *flk_cbp, cred_t *cr)
3080 {
3081 	znode_t *zp = VTOZ(vp);
3082 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3083 	int error;
3084 
3085 	ZFS_ENTER(zfsvfs);
3086 
3087 	/*
3088 	 * We are following the UFS semantics with respect to mapcnt
3089 	 * here: If we see that the file is mapped already, then we will
3090 	 * return an error, but we don't worry about races between this
3091 	 * function and zfs_map().
3092 	 */
3093 	if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3094 		ZFS_EXIT(zfsvfs);
3095 		return (EAGAIN);
3096 	}
3097 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr);
3098 	ZFS_EXIT(zfsvfs);
3099 	return (error);
3100 }
3101 
3102 /*
3103  * If we can't find a page in the cache, we will create a new page
3104  * and fill it with file data.  For efficiency, we may try to fill
3105  * multiple pages at once (klustering).
3106  */
3107 static int
3108 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3109     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3110 {
3111 	znode_t *zp = VTOZ(vp);
3112 	page_t *pp, *cur_pp;
3113 	objset_t *os = zp->z_zfsvfs->z_os;
3114 	caddr_t va;
3115 	u_offset_t io_off, total;
3116 	uint64_t oid = zp->z_id;
3117 	size_t io_len;
3118 	uint64_t filesz;
3119 	int err;
3120 
3121 	/*
3122 	 * If we are only asking for a single page don't bother klustering.
3123 	 */
3124 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3125 	if (off >= filesz)
3126 		return (EFAULT);
3127 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3128 		io_off = off;
3129 		io_len = PAGESIZE;
3130 		pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr);
3131 	} else {
3132 		/*
3133 		 * Try to fill a kluster of pages (a blocks worth).
3134 		 */
3135 		size_t klen;
3136 		u_offset_t koff;
3137 
3138 		if (!ISP2(zp->z_blksz)) {
3139 			/* Only one block in the file. */
3140 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3141 			koff = 0;
3142 		} else {
3143 			/*
3144 			 * It would be ideal to align our offset to the
3145 			 * blocksize but doing so has resulted in some
3146 			 * strange application crashes. For now, we
3147 			 * leave the offset as is and only adjust the
3148 			 * length if we are off the end of the file.
3149 			 */
3150 			koff = off;
3151 			klen = plsz;
3152 		}
3153 		ASSERT(koff <= filesz);
3154 		if (koff + klen > filesz)
3155 			klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff;
3156 		ASSERT3U(off, >=, koff);
3157 		ASSERT3U(off, <, koff + klen);
3158 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3159 			    &io_len, koff, klen, 0);
3160 	}
3161 	if (pp == NULL) {
3162 		/*
3163 		 * Some other thread entered the page before us.
3164 		 * Return to zfs_getpage to retry the lookup.
3165 		 */
3166 		*pl = NULL;
3167 		return (0);
3168 	}
3169 
3170 	/*
3171 	 * Fill the pages in the kluster.
3172 	 */
3173 	cur_pp = pp;
3174 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3175 		ASSERT3U(io_off, ==, cur_pp->p_offset);
3176 		va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1);
3177 		err = dmu_read(os, oid, io_off, PAGESIZE, va);
3178 		ppmapout(va);
3179 		if (err) {
3180 			/* On error, toss the entire kluster */
3181 			pvn_read_done(pp, B_ERROR);
3182 			return (err);
3183 		}
3184 		cur_pp = cur_pp->p_next;
3185 	}
3186 out:
3187 	/*
3188 	 * Fill in the page list array from the kluster.  If
3189 	 * there are too many pages in the kluster, return
3190 	 * as many pages as possible starting from the desired
3191 	 * offset `off'.
3192 	 * NOTE: the page list will always be null terminated.
3193 	 */
3194 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3195 
3196 	return (0);
3197 }
3198 
3199 /*
3200  * Return pointers to the pages for the file region [off, off + len]
3201  * in the pl array.  If plsz is greater than len, this function may
3202  * also return page pointers from before or after the specified
3203  * region (i.e. some region [off', off' + plsz]).  These additional
3204  * pages are only returned if they are already in the cache, or were
3205  * created as part of a klustered read.
3206  *
3207  *	IN:	vp	- vnode of file to get data from.
3208  *		off	- position in file to get data from.
3209  *		len	- amount of data to retrieve.
3210  *		plsz	- length of provided page list.
3211  *		seg	- segment to obtain pages for.
3212  *		addr	- virtual address of fault.
3213  *		rw	- mode of created pages.
3214  *		cr	- credentials of caller.
3215  *
3216  *	OUT:	protp	- protection mode of created pages.
3217  *		pl	- list of pages created.
3218  *
3219  *	RETURN:	0 if success
3220  *		error code if failure
3221  *
3222  * Timestamps:
3223  *	vp - atime updated
3224  */
3225 /* ARGSUSED */
3226 static int
3227 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3228 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3229 	enum seg_rw rw, cred_t *cr)
3230 {
3231 	znode_t		*zp = VTOZ(vp);
3232 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3233 	page_t		*pp, **pl0 = pl;
3234 	int		need_unlock = 0, err = 0;
3235 	offset_t	orig_off;
3236 
3237 	ZFS_ENTER(zfsvfs);
3238 
3239 	if (protp)
3240 		*protp = PROT_ALL;
3241 
3242 	ASSERT(zp->z_dbuf_held && zp->z_phys);
3243 
3244 	/* no faultahead (for now) */
3245 	if (pl == NULL) {
3246 		ZFS_EXIT(zfsvfs);
3247 		return (0);
3248 	}
3249 
3250 	/* can't fault past EOF */
3251 	if (off >= zp->z_phys->zp_size) {
3252 		ZFS_EXIT(zfsvfs);
3253 		return (EFAULT);
3254 	}
3255 	orig_off = off;
3256 
3257 	/*
3258 	 * If we already own the lock, then we must be page faulting
3259 	 * in the middle of a write to this file (i.e., we are writing
3260 	 * to this file using data from a mapped region of the file).
3261 	 */
3262 	if (rw_owner(&zp->z_map_lock) != curthread) {
3263 		rw_enter(&zp->z_map_lock, RW_WRITER);
3264 		need_unlock = TRUE;
3265 	}
3266 
3267 	/*
3268 	 * Loop through the requested range [off, off + len] looking
3269 	 * for pages.  If we don't find a page, we will need to create
3270 	 * a new page and fill it with data from the file.
3271 	 */
3272 	while (len > 0) {
3273 		if (plsz < PAGESIZE)
3274 			break;
3275 		if (pp = page_lookup(vp, off, SE_SHARED)) {
3276 			*pl++ = pp;
3277 			off += PAGESIZE;
3278 			addr += PAGESIZE;
3279 			len -= PAGESIZE;
3280 			plsz -= PAGESIZE;
3281 		} else {
3282 			err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw);
3283 			if (err)
3284 				goto out;
3285 			/*
3286 			 * klustering may have changed our region
3287 			 * to be block aligned.
3288 			 */
3289 			if (((pp = *pl) != 0) && (off != pp->p_offset)) {
3290 				int delta = off - pp->p_offset;
3291 				len += delta;
3292 				off -= delta;
3293 				addr -= delta;
3294 			}
3295 			while (*pl) {
3296 				pl++;
3297 				off += PAGESIZE;
3298 				addr += PAGESIZE;
3299 				plsz -= PAGESIZE;
3300 				if (len > PAGESIZE)
3301 					len -= PAGESIZE;
3302 				else
3303 					len = 0;
3304 			}
3305 		}
3306 	}
3307 
3308 	/*
3309 	 * Fill out the page array with any pages already in the cache.
3310 	 */
3311 	while (plsz > 0) {
3312 		pp = page_lookup_nowait(vp, off, SE_SHARED);
3313 		if (pp == NULL)
3314 			break;
3315 		*pl++ = pp;
3316 		off += PAGESIZE;
3317 		plsz -= PAGESIZE;
3318 	}
3319 
3320 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3321 out:
3322 	/*
3323 	 * We can't grab the range lock for the page as reader which would
3324 	 * stop truncation as this leads to deadlock. So we need to recheck
3325 	 * the file size.
3326 	 */
3327 	if (orig_off >= zp->z_phys->zp_size)
3328 		err = EFAULT;
3329 	if (err) {
3330 		/*
3331 		 * Release any pages we have previously locked.
3332 		 */
3333 		while (pl > pl0)
3334 			page_unlock(*--pl);
3335 	}
3336 
3337 	*pl = NULL;
3338 
3339 	if (need_unlock)
3340 		rw_exit(&zp->z_map_lock);
3341 
3342 	ZFS_EXIT(zfsvfs);
3343 	return (err);
3344 }
3345 
3346 /*
3347  * Request a memory map for a section of a file.  This code interacts
3348  * with common code and the VM system as follows:
3349  *
3350  *	common code calls mmap(), which ends up in smmap_common()
3351  *
3352  *	this calls VOP_MAP(), which takes you into (say) zfs
3353  *
3354  *	zfs_map() calls as_map(), passing segvn_create() as the callback
3355  *
3356  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
3357  *
3358  *	zfs_addmap() updates z_mapcnt
3359  */
3360 static int
3361 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
3362     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3363 {
3364 	znode_t *zp = VTOZ(vp);
3365 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3366 	segvn_crargs_t	vn_a;
3367 	int		error;
3368 
3369 	ZFS_ENTER(zfsvfs);
3370 
3371 	if (vp->v_flag & VNOMAP) {
3372 		ZFS_EXIT(zfsvfs);
3373 		return (ENOSYS);
3374 	}
3375 
3376 	if (off < 0 || len > MAXOFFSET_T - off) {
3377 		ZFS_EXIT(zfsvfs);
3378 		return (ENXIO);
3379 	}
3380 
3381 	if (vp->v_type != VREG) {
3382 		ZFS_EXIT(zfsvfs);
3383 		return (ENODEV);
3384 	}
3385 
3386 	/*
3387 	 * If file is locked, disallow mapping.
3388 	 */
3389 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) {
3390 		ZFS_EXIT(zfsvfs);
3391 		return (EAGAIN);
3392 	}
3393 
3394 	as_rangelock(as);
3395 	if ((flags & MAP_FIXED) == 0) {
3396 		map_addr(addrp, len, off, 1, flags);
3397 		if (*addrp == NULL) {
3398 			as_rangeunlock(as);
3399 			ZFS_EXIT(zfsvfs);
3400 			return (ENOMEM);
3401 		}
3402 	} else {
3403 		/*
3404 		 * User specified address - blow away any previous mappings
3405 		 */
3406 		(void) as_unmap(as, *addrp, len);
3407 	}
3408 
3409 	vn_a.vp = vp;
3410 	vn_a.offset = (u_offset_t)off;
3411 	vn_a.type = flags & MAP_TYPE;
3412 	vn_a.prot = prot;
3413 	vn_a.maxprot = maxprot;
3414 	vn_a.cred = cr;
3415 	vn_a.amp = NULL;
3416 	vn_a.flags = flags & ~MAP_TYPE;
3417 	vn_a.szc = 0;
3418 	vn_a.lgrp_mem_policy_flags = 0;
3419 
3420 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
3421 
3422 	as_rangeunlock(as);
3423 	ZFS_EXIT(zfsvfs);
3424 	return (error);
3425 }
3426 
3427 /* ARGSUSED */
3428 static int
3429 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3430     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3431 {
3432 	uint64_t pages = btopr(len);
3433 
3434 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
3435 	return (0);
3436 }
3437 
3438 /*
3439  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
3440  * more accurate mtime for the associated file.  Since we don't have a way of
3441  * detecting when the data was actually modified, we have to resort to
3442  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
3443  * last page is pushed.  The problem occurs when the msync() call is omitted,
3444  * which by far the most common case:
3445  *
3446  * 	open()
3447  * 	mmap()
3448  * 	<modify memory>
3449  * 	munmap()
3450  * 	close()
3451  * 	<time lapse>
3452  * 	putpage() via fsflush
3453  *
3454  * If we wait until fsflush to come along, we can have a modification time that
3455  * is some arbitrary point in the future.  In order to prevent this in the
3456  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
3457  * torn down.
3458  */
3459 /* ARGSUSED */
3460 static int
3461 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3462     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr)
3463 {
3464 	uint64_t pages = btopr(len);
3465 
3466 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
3467 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
3468 
3469 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
3470 	    vn_has_cached_data(vp))
3471 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr);
3472 
3473 	return (0);
3474 }
3475 
3476 /*
3477  * Free or allocate space in a file.  Currently, this function only
3478  * supports the `F_FREESP' command.  However, this command is somewhat
3479  * misnamed, as its functionality includes the ability to allocate as
3480  * well as free space.
3481  *
3482  *	IN:	vp	- vnode of file to free data in.
3483  *		cmd	- action to take (only F_FREESP supported).
3484  *		bfp	- section of file to free/alloc.
3485  *		flag	- current file open mode flags.
3486  *		offset	- current file offset.
3487  *		cr	- credentials of caller [UNUSED].
3488  *
3489  *	RETURN:	0 if success
3490  *		error code if failure
3491  *
3492  * Timestamps:
3493  *	vp - ctime|mtime updated
3494  */
3495 /* ARGSUSED */
3496 static int
3497 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
3498     offset_t offset, cred_t *cr, caller_context_t *ct)
3499 {
3500 	znode_t		*zp = VTOZ(vp);
3501 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3502 	uint64_t	off, len;
3503 	int		error;
3504 
3505 	ZFS_ENTER(zfsvfs);
3506 
3507 top:
3508 	if (cmd != F_FREESP) {
3509 		ZFS_EXIT(zfsvfs);
3510 		return (EINVAL);
3511 	}
3512 
3513 	if (error = convoff(vp, bfp, 0, offset)) {
3514 		ZFS_EXIT(zfsvfs);
3515 		return (error);
3516 	}
3517 
3518 	if (bfp->l_len < 0) {
3519 		ZFS_EXIT(zfsvfs);
3520 		return (EINVAL);
3521 	}
3522 
3523 	off = bfp->l_start;
3524 	len = bfp->l_len; /* 0 means from off to end of file */
3525 
3526 	do {
3527 		error = zfs_freesp(zp, off, len, flag, TRUE);
3528 		/* NB: we already did dmu_tx_wait() if necessary */
3529 	} while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
3530 
3531 	ZFS_EXIT(zfsvfs);
3532 	return (error);
3533 }
3534 
3535 static int
3536 zfs_fid(vnode_t *vp, fid_t *fidp)
3537 {
3538 	znode_t		*zp = VTOZ(vp);
3539 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3540 	uint32_t	gen = (uint32_t)zp->z_phys->zp_gen;
3541 	uint64_t	object = zp->z_id;
3542 	zfid_short_t	*zfid;
3543 	int		size, i;
3544 
3545 	ZFS_ENTER(zfsvfs);
3546 
3547 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
3548 	if (fidp->fid_len < size) {
3549 		fidp->fid_len = size;
3550 		ZFS_EXIT(zfsvfs);
3551 		return (ENOSPC);
3552 	}
3553 
3554 	zfid = (zfid_short_t *)fidp;
3555 
3556 	zfid->zf_len = size;
3557 
3558 	for (i = 0; i < sizeof (zfid->zf_object); i++)
3559 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
3560 
3561 	/* Must have a non-zero generation number to distinguish from .zfs */
3562 	if (gen == 0)
3563 		gen = 1;
3564 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
3565 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
3566 
3567 	if (size == LONG_FID_LEN) {
3568 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
3569 		zfid_long_t	*zlfid;
3570 
3571 		zlfid = (zfid_long_t *)fidp;
3572 
3573 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
3574 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
3575 
3576 		/* XXX - this should be the generation number for the objset */
3577 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
3578 			zlfid->zf_setgen[i] = 0;
3579 	}
3580 
3581 	ZFS_EXIT(zfsvfs);
3582 	return (0);
3583 }
3584 
3585 static int
3586 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr)
3587 {
3588 	znode_t		*zp, *xzp;
3589 	zfsvfs_t	*zfsvfs;
3590 	zfs_dirlock_t	*dl;
3591 	int		error;
3592 
3593 	switch (cmd) {
3594 	case _PC_LINK_MAX:
3595 		*valp = ULONG_MAX;
3596 		return (0);
3597 
3598 	case _PC_FILESIZEBITS:
3599 		*valp = 64;
3600 		return (0);
3601 
3602 	case _PC_XATTR_EXISTS:
3603 		zp = VTOZ(vp);
3604 		zfsvfs = zp->z_zfsvfs;
3605 		ZFS_ENTER(zfsvfs);
3606 		*valp = 0;
3607 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
3608 		    ZXATTR | ZEXISTS | ZSHARED);
3609 		if (error == 0) {
3610 			zfs_dirent_unlock(dl);
3611 			if (!zfs_dirempty(xzp))
3612 				*valp = 1;
3613 			VN_RELE(ZTOV(xzp));
3614 		} else if (error == ENOENT) {
3615 			/*
3616 			 * If there aren't extended attributes, it's the
3617 			 * same as having zero of them.
3618 			 */
3619 			error = 0;
3620 		}
3621 		ZFS_EXIT(zfsvfs);
3622 		return (error);
3623 
3624 	case _PC_ACL_ENABLED:
3625 		*valp = _ACL_ACE_ENABLED;
3626 		return (0);
3627 
3628 	case _PC_MIN_HOLE_SIZE:
3629 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
3630 		return (0);
3631 
3632 	default:
3633 		return (fs_pathconf(vp, cmd, valp, cr));
3634 	}
3635 }
3636 
3637 /*ARGSUSED*/
3638 static int
3639 zfs_getsecattr(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_getacl(zp, vsecp, cr);
3647 	ZFS_EXIT(zfsvfs);
3648 
3649 	return (error);
3650 }
3651 
3652 /*ARGSUSED*/
3653 static int
3654 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr)
3655 {
3656 	znode_t *zp = VTOZ(vp);
3657 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3658 	int error;
3659 
3660 	ZFS_ENTER(zfsvfs);
3661 	error = zfs_setacl(zp, vsecp, cr);
3662 	ZFS_EXIT(zfsvfs);
3663 	return (error);
3664 }
3665 
3666 /*
3667  * Predeclare these here so that the compiler assumes that
3668  * this is an "old style" function declaration that does
3669  * not include arguments => we won't get type mismatch errors
3670  * in the initializations that follow.
3671  */
3672 static int zfs_inval();
3673 static int zfs_isdir();
3674 
3675 static int
3676 zfs_inval()
3677 {
3678 	return (EINVAL);
3679 }
3680 
3681 static int
3682 zfs_isdir()
3683 {
3684 	return (EISDIR);
3685 }
3686 /*
3687  * Directory vnode operations template
3688  */
3689 vnodeops_t *zfs_dvnodeops;
3690 const fs_operation_def_t zfs_dvnodeops_template[] = {
3691 	VOPNAME_OPEN, zfs_open,
3692 	VOPNAME_CLOSE, zfs_close,
3693 	VOPNAME_READ, zfs_isdir,
3694 	VOPNAME_WRITE, zfs_isdir,
3695 	VOPNAME_IOCTL, zfs_ioctl,
3696 	VOPNAME_GETATTR, zfs_getattr,
3697 	VOPNAME_SETATTR, zfs_setattr,
3698 	VOPNAME_ACCESS, zfs_access,
3699 	VOPNAME_LOOKUP, zfs_lookup,
3700 	VOPNAME_CREATE, zfs_create,
3701 	VOPNAME_REMOVE, zfs_remove,
3702 	VOPNAME_LINK, zfs_link,
3703 	VOPNAME_RENAME, zfs_rename,
3704 	VOPNAME_MKDIR, zfs_mkdir,
3705 	VOPNAME_RMDIR, zfs_rmdir,
3706 	VOPNAME_READDIR, zfs_readdir,
3707 	VOPNAME_SYMLINK, zfs_symlink,
3708 	VOPNAME_FSYNC, zfs_fsync,
3709 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3710 	VOPNAME_FID, zfs_fid,
3711 	VOPNAME_SEEK, zfs_seek,
3712 	VOPNAME_PATHCONF, zfs_pathconf,
3713 	VOPNAME_GETSECATTR, zfs_getsecattr,
3714 	VOPNAME_SETSECATTR, zfs_setsecattr,
3715 	NULL, NULL
3716 };
3717 
3718 /*
3719  * Regular file vnode operations template
3720  */
3721 vnodeops_t *zfs_fvnodeops;
3722 const fs_operation_def_t zfs_fvnodeops_template[] = {
3723 	VOPNAME_OPEN, zfs_open,
3724 	VOPNAME_CLOSE, zfs_close,
3725 	VOPNAME_READ, zfs_read,
3726 	VOPNAME_WRITE, zfs_write,
3727 	VOPNAME_IOCTL, zfs_ioctl,
3728 	VOPNAME_GETATTR, zfs_getattr,
3729 	VOPNAME_SETATTR, zfs_setattr,
3730 	VOPNAME_ACCESS, zfs_access,
3731 	VOPNAME_LOOKUP, zfs_lookup,
3732 	VOPNAME_RENAME, zfs_rename,
3733 	VOPNAME_FSYNC, zfs_fsync,
3734 	VOPNAME_INACTIVE, (fs_generic_func_p)zfs_inactive,
3735 	VOPNAME_FID, zfs_fid,
3736 	VOPNAME_SEEK, zfs_seek,
3737 	VOPNAME_FRLOCK, zfs_frlock,
3738 	VOPNAME_SPACE, zfs_space,
3739 	VOPNAME_GETPAGE, zfs_getpage,
3740 	VOPNAME_PUTPAGE, zfs_putpage,
3741 	VOPNAME_MAP, (fs_generic_func_p) zfs_map,
3742 	VOPNAME_ADDMAP, (fs_generic_func_p) zfs_addmap,
3743 	VOPNAME_DELMAP, zfs_delmap,
3744 	VOPNAME_PATHCONF, zfs_pathconf,
3745 	VOPNAME_GETSECATTR, zfs_getsecattr,
3746 	VOPNAME_SETSECATTR, zfs_setsecattr,
3747 	VOPNAME_VNEVENT, fs_vnevent_support,
3748 	NULL, NULL
3749 };
3750 
3751 /*
3752  * Symbolic link vnode operations template
3753  */
3754 vnodeops_t *zfs_symvnodeops;
3755 const fs_operation_def_t zfs_symvnodeops_template[] = {
3756 	VOPNAME_GETATTR, zfs_getattr,
3757 	VOPNAME_SETATTR, zfs_setattr,
3758 	VOPNAME_ACCESS, zfs_access,
3759 	VOPNAME_RENAME, zfs_rename,
3760 	VOPNAME_READLINK, zfs_readlink,
3761 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3762 	VOPNAME_FID, zfs_fid,
3763 	VOPNAME_PATHCONF, zfs_pathconf,
3764 	VOPNAME_VNEVENT, fs_vnevent_support,
3765 	NULL, NULL
3766 };
3767 
3768 /*
3769  * Extended attribute directory vnode operations template
3770  *	This template is identical to the directory vnodes
3771  *	operation template except for restricted operations:
3772  *		VOP_MKDIR()
3773  *		VOP_SYMLINK()
3774  * Note that there are other restrictions embedded in:
3775  *	zfs_create()	- restrict type to VREG
3776  *	zfs_link()	- no links into/out of attribute space
3777  *	zfs_rename()	- no moves into/out of attribute space
3778  */
3779 vnodeops_t *zfs_xdvnodeops;
3780 const fs_operation_def_t zfs_xdvnodeops_template[] = {
3781 	VOPNAME_OPEN, zfs_open,
3782 	VOPNAME_CLOSE, zfs_close,
3783 	VOPNAME_IOCTL, zfs_ioctl,
3784 	VOPNAME_GETATTR, zfs_getattr,
3785 	VOPNAME_SETATTR, zfs_setattr,
3786 	VOPNAME_ACCESS, zfs_access,
3787 	VOPNAME_LOOKUP, zfs_lookup,
3788 	VOPNAME_CREATE, zfs_create,
3789 	VOPNAME_REMOVE, zfs_remove,
3790 	VOPNAME_LINK, zfs_link,
3791 	VOPNAME_RENAME, zfs_rename,
3792 	VOPNAME_MKDIR, zfs_inval,
3793 	VOPNAME_RMDIR, zfs_rmdir,
3794 	VOPNAME_READDIR, zfs_readdir,
3795 	VOPNAME_SYMLINK, zfs_inval,
3796 	VOPNAME_FSYNC, zfs_fsync,
3797 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3798 	VOPNAME_FID, zfs_fid,
3799 	VOPNAME_SEEK, zfs_seek,
3800 	VOPNAME_PATHCONF, zfs_pathconf,
3801 	VOPNAME_GETSECATTR, zfs_getsecattr,
3802 	VOPNAME_SETSECATTR, zfs_setsecattr,
3803 	VOPNAME_VNEVENT, fs_vnevent_support,
3804 	NULL, NULL
3805 };
3806 
3807 /*
3808  * Error vnode operations template
3809  */
3810 vnodeops_t *zfs_evnodeops;
3811 const fs_operation_def_t zfs_evnodeops_template[] = {
3812 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3813 	VOPNAME_PATHCONF, zfs_pathconf,
3814 	NULL, NULL
3815 };
3816