xref: /titanic_51/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 89c0ae934a35c2ab26de63e1d9fba6bcb04c6306)
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 <vm/seg_vn.h>
44 #include <vm/pvn.h>
45 #include <vm/as.h>
46 #include <sys/mman.h>
47 #include <sys/pathname.h>
48 #include <sys/cmn_err.h>
49 #include <sys/errno.h>
50 #include <sys/unistd.h>
51 #include <sys/zfs_vfsops.h>
52 #include <sys/zfs_dir.h>
53 #include <sys/zfs_acl.h>
54 #include <sys/zfs_ioctl.h>
55 #include <sys/fs/zfs.h>
56 #include <sys/dmu.h>
57 #include <sys/spa.h>
58 #include <sys/txg.h>
59 #include <sys/refcount.h>  /* temporary for debugging purposes */
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 
70 /*
71  * Programming rules.
72  *
73  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
74  * properly lock its in-core state, create a DMU transaction, do the work,
75  * record this work in the intent log (ZIL), commit the DMU transaction,
76  * and wait the the intent log to commit if it's is a synchronous operation.
77  * Morover, the vnode ops must work in both normal and log replay context.
78  * The ordering of events is important to avoid deadlocks and references
79  * to freed memory.  The example below illustrates the following Big Rules:
80  *
81  *  (1) A check must be made in each zfs thread for a mounted file system.
82  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
83  *	A ZFS_EXIT(zfsvfs) is needed before all returns.
84  *
85  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
86  *	and ZFS_EXIT(). This is for 3 reasons:
87  *	First, if it's the last reference, the vnode/znode
88  *	can be freed, so the zp may point to freed memory.  Second, the last
89  *	reference will call zfs_zinactive(), which may induce a lot of work --
90  *	pushing cached pages (which requires z_grow_lock) and syncing out
91  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
92  *	which could deadlock the system if you were already holding one.
93  *
94  *  (3)	Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign().
95  *	In normal operation, this will be TXG_NOWAIT.  During ZIL replay,
96  *	it will be a specific txg.  Either way, dmu_tx_assign() never blocks.
97  *	This is critical because we don't want to block while holding locks.
98  *	Note, in particular, that if a lock is sometimes acquired before
99  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
100  *	use a non-blocking assign can deadlock the system.  The scenario:
101  *
102  *	Thread A has grabbed a lock before calling dmu_tx_assign().
103  *	Thread B is in an already-assigned tx, and blocks for this lock.
104  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
105  *	forever, because the previous txg can't quiesce until B's tx commits.
106  *
107  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
108  *	then drop all locks, call txg_wait_open(), and try again.
109  *
110  *  (4)	If the operation succeeded, generate the intent log entry for it
111  *	before dropping locks.  This ensures that the ordering of events
112  *	in the intent log matches the order in which they actually occurred.
113  *
114  *  (5)	At the end of each vnode op, the DMU tx must always commit,
115  *	regardless of whether there were any errors.
116  *
117  *  (6)	After dropping all locks, invoke zil_commit(zilog, seq, ioflag)
118  *	to ensure that synchronous semantics are provided when necessary.
119  *
120  * In general, this is how things should be ordered in each vnode op:
121  *
122  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
123  * top:
124  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
125  *	rw_enter(...);			// grab any other locks you need
126  *	tx = dmu_tx_create(...);	// get DMU tx
127  *	dmu_tx_hold_*();		// hold each object you might modify
128  *	error = dmu_tx_assign(tx, zfsvfs->z_assign);	// try to assign
129  *	if (error) {
130  *		dmu_tx_abort(tx);	// abort DMU tx
131  *		rw_exit(...);		// drop locks
132  *		zfs_dirent_unlock(dl);	// unlock directory entry
133  *		VN_RELE(...);		// release held vnodes
134  *		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
135  *			txg_wait_open(dmu_objset_pool(os), 0);
136  *			goto top;
137  *		}
138  *		ZFS_EXIT(zfsvfs);	// finished in zfs
139  *		return (error);		// really out of space
140  *	}
141  *	error = do_real_work();		// do whatever this VOP does
142  *	if (error == 0)
143  *		seq = zfs_log_*(...);	// on success, make ZIL entry
144  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
145  *	rw_exit(...);			// drop locks
146  *	zfs_dirent_unlock(dl);		// unlock directory entry
147  *	VN_RELE(...);			// release held vnodes
148  *	zil_commit(zilog, seq, ioflag);	// synchronous when necessary
149  *	ZFS_EXIT(zfsvfs);		// finished in zfs
150  *	return (error);			// done, report error
151  */
152 
153 /* ARGSUSED */
154 static int
155 zfs_open(vnode_t **vpp, int flag, cred_t *cr)
156 {
157 	return (0);
158 }
159 
160 /* ARGSUSED */
161 static int
162 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr)
163 {
164 	/*
165 	 * Clean up any locks held by this process on the vp.
166 	 */
167 	cleanlocks(vp, ddi_get_pid(), 0);
168 	cleanshares(vp, ddi_get_pid());
169 
170 	return (0);
171 }
172 
173 /*
174  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
175  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
176  */
177 static int
178 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
179 {
180 	znode_t	*zp = VTOZ(vp);
181 	uint64_t noff = (uint64_t)*off; /* new offset */
182 	uint64_t file_sz;
183 	int error;
184 	boolean_t hole;
185 
186 	rw_enter(&zp->z_grow_lock, RW_READER);
187 	file_sz = zp->z_phys->zp_size;
188 	if (noff >= file_sz)  {
189 		rw_exit(&zp->z_grow_lock);
190 		return (ENXIO);
191 	}
192 
193 	if (cmd == _FIO_SEEK_HOLE)
194 		hole = B_TRUE;
195 	else
196 		hole = B_FALSE;
197 
198 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
199 	rw_exit(&zp->z_grow_lock);
200 
201 	/* end of file? */
202 	if ((error == ESRCH) || (noff > file_sz)) {
203 		/*
204 		 * Handle the virtual hole at the end of file.
205 		 */
206 		if (hole) {
207 			*off = file_sz;
208 			return (0);
209 		}
210 		return (ENXIO);
211 	}
212 
213 	if (noff < *off)
214 		return (error);
215 	*off = noff;
216 	return (error);
217 }
218 
219 /* ARGSUSED */
220 static int
221 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
222     int *rvalp)
223 {
224 	offset_t off;
225 	int error;
226 	zfsvfs_t *zfsvfs;
227 
228 	switch (com) {
229 	    case _FIOFFS:
230 		return (zfs_sync(vp->v_vfsp, 0, cred));
231 
232 	    case _FIO_SEEK_DATA:
233 	    case _FIO_SEEK_HOLE:
234 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
235 			return (EFAULT);
236 
237 		zfsvfs = VTOZ(vp)->z_zfsvfs;
238 		ZFS_ENTER(zfsvfs);
239 
240 		/* offset parameter is in/out */
241 		error = zfs_holey(vp, com, &off);
242 		ZFS_EXIT(zfsvfs);
243 		if (error)
244 			return (error);
245 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
246 			return (EFAULT);
247 		return (0);
248 	}
249 	return (ENOTTY);
250 }
251 
252 /*
253  * When a file is memory mapped, we must keep the IO data synchronized
254  * between the DMU cache and the memory mapped pages.  What this means:
255  *
256  * On Write:	If we find a memory mapped page, we write to *both*
257  *		the page and the dmu buffer.
258  *
259  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
260  *	the file is memory mapped.
261  */
262 static int
263 mappedwrite(vnode_t *vp, uint64_t woff, int nbytes, uio_t *uio, dmu_tx_t *tx)
264 {
265 	znode_t	*zp = VTOZ(vp);
266 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
267 	int64_t	start, off;
268 	int len = nbytes;
269 	int error = 0;
270 
271 	start = uio->uio_loffset;
272 	off = start & PAGEOFFSET;
273 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
274 		page_t *pp;
275 		uint64_t bytes = MIN(PAGESIZE - off, len);
276 
277 		/*
278 		 * We don't want a new page to "appear" in the middle of
279 		 * the file update (because it may not get the write
280 		 * update data), so we grab a lock to block
281 		 * zfs_getpage().
282 		 */
283 		rw_enter(&zp->z_map_lock, RW_WRITER);
284 		if (pp = page_lookup(vp, start, SE_SHARED)) {
285 			caddr_t va;
286 
287 			rw_exit(&zp->z_map_lock);
288 			va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L);
289 			error = uiomove(va+off, bytes, UIO_WRITE, uio);
290 			if (error == 0) {
291 				dmu_write(zfsvfs->z_os, zp->z_id,
292 				    woff, bytes, va+off, tx);
293 			}
294 			ppmapout(va);
295 			page_unlock(pp);
296 		} else {
297 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
298 			    woff, bytes, uio, tx);
299 			rw_exit(&zp->z_map_lock);
300 		}
301 		len -= bytes;
302 		woff += bytes;
303 		off = 0;
304 		if (error)
305 			break;
306 	}
307 	return (error);
308 }
309 
310 /*
311  * When a file is memory mapped, we must keep the IO data synchronized
312  * between the DMU cache and the memory mapped pages.  What this means:
313  *
314  * On Read:	We "read" preferentially from memory mapped pages,
315  *		else we default from the dmu buffer.
316  *
317  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
318  *	the file is memory mapped.
319  */
320 static int
321 mappedread(vnode_t *vp, char *addr, int nbytes, uio_t *uio)
322 {
323 	int64_t	start, off, bytes;
324 	int len = nbytes;
325 	int error = 0;
326 
327 	start = uio->uio_loffset;
328 	off = start & PAGEOFFSET;
329 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
330 		page_t *pp;
331 
332 		bytes = MIN(PAGESIZE - off, len);
333 		if (pp = page_lookup(vp, start, SE_SHARED)) {
334 			caddr_t va;
335 
336 			va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L);
337 			error = uiomove(va + off, bytes, UIO_READ, uio);
338 			ppmapout(va);
339 			page_unlock(pp);
340 		} else {
341 			/* XXX use dmu_read here? */
342 			error = uiomove(addr, bytes, UIO_READ, uio);
343 		}
344 		len -= bytes;
345 		addr += bytes;
346 		off = 0;
347 		if (error)
348 			break;
349 	}
350 	return (error);
351 }
352 
353 uint_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
354 
355 /*
356  * Read bytes from specified file into supplied buffer.
357  *
358  *	IN:	vp	- vnode of file to be read from.
359  *		uio	- structure supplying read location, range info,
360  *			  and return buffer.
361  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
362  *		cr	- credentials of caller.
363  *
364  *	OUT:	uio	- updated offset and range, buffer filled.
365  *
366  *	RETURN:	0 if success
367  *		error code if failure
368  *
369  * Side Effects:
370  *	vp - atime updated if byte count > 0
371  */
372 /* ARGSUSED */
373 static int
374 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
375 {
376 	znode_t		*zp = VTOZ(vp);
377 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
378 	uint64_t	delta;
379 	ssize_t		n, size, cnt, ndone;
380 	int		error, i, numbufs;
381 	dmu_buf_t	*dbp, **dbpp;
382 
383 	ZFS_ENTER(zfsvfs);
384 
385 	/*
386 	 * Validate file offset
387 	 */
388 	if (uio->uio_loffset < (offset_t)0) {
389 		ZFS_EXIT(zfsvfs);
390 		return (EINVAL);
391 	}
392 
393 	/*
394 	 * Fasttrack empty reads
395 	 */
396 	if (uio->uio_resid == 0) {
397 		ZFS_EXIT(zfsvfs);
398 		return (0);
399 	}
400 
401 	/*
402 	 * Check for region locks
403 	 */
404 	if (MANDMODE((mode_t)zp->z_phys->zp_mode)) {
405 		if (error = chklock(vp, FREAD,
406 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
407 			ZFS_EXIT(zfsvfs);
408 			return (error);
409 		}
410 	}
411 
412 	/*
413 	 * If we're in FRSYNC mode, sync out this znode before reading it.
414 	 */
415 	zil_commit(zfsvfs->z_log, zp->z_last_itx, ioflag & FRSYNC);
416 
417 	/*
418 	 * Make sure nobody restructures the file (changes block size)
419 	 * in the middle of the read.
420 	 */
421 	rw_enter(&zp->z_grow_lock, RW_READER);
422 	/*
423 	 * If we are reading past end-of-file we can skip
424 	 * to the end; but we might still need to set atime.
425 	 */
426 	if (uio->uio_loffset >= zp->z_phys->zp_size) {
427 		cnt = 0;
428 		error = 0;
429 		goto out;
430 	}
431 
432 	cnt = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset);
433 
434 	for (ndone = 0; ndone < cnt; ndone += zfs_read_chunk_size) {
435 		ASSERT(uio->uio_loffset < zp->z_phys->zp_size);
436 		n = MIN(zfs_read_chunk_size,
437 		    zp->z_phys->zp_size - uio->uio_loffset);
438 		n = MIN(n, cnt);
439 		dbpp = dmu_buf_hold_array(zfsvfs->z_os, zp->z_id,
440 		    uio->uio_loffset, n, &numbufs);
441 		if (error = dmu_buf_read_array_canfail(dbpp, numbufs)) {
442 			dmu_buf_rele_array(dbpp, numbufs);
443 			goto out;
444 		}
445 		/*
446 		 * Compute the adjustment to align the dmu buffers
447 		 * with the uio buffer.
448 		 */
449 		delta = uio->uio_loffset - dbpp[0]->db_offset;
450 
451 		for (i = 0; i < numbufs; i++) {
452 			if (n < 0)
453 				break;
454 			dbp = dbpp[i];
455 			size = dbp->db_size - delta;
456 			/*
457 			 * XXX -- this is correct, but may be suboptimal.
458 			 * If the pages are all clean, we don't need to
459 			 * go through mappedread().  Maybe the VMODSORT
460 			 * stuff can help us here.
461 			 */
462 			if (vn_has_cached_data(vp)) {
463 				error = mappedread(vp, (caddr_t)dbp->db_data +
464 				    delta, (n < size ? n : size), uio);
465 			} else {
466 				error = uiomove((caddr_t)dbp->db_data + delta,
467 					(n < size ? n : size), UIO_READ, uio);
468 			}
469 			if (error) {
470 				dmu_buf_rele_array(dbpp, numbufs);
471 				goto out;
472 			}
473 			n -= dbp->db_size;
474 			if (delta) {
475 				n += delta;
476 				delta = 0;
477 			}
478 		}
479 		dmu_buf_rele_array(dbpp, numbufs);
480 	}
481 out:
482 	rw_exit(&zp->z_grow_lock);
483 
484 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
485 	ZFS_EXIT(zfsvfs);
486 	return (error);
487 }
488 
489 /*
490  * Fault in the pages of the first n bytes specified by the uio structure.
491  * 1 byte in each page is touched and the uio struct is unmodified.
492  * Any error will exit this routine as this is only a best
493  * attempt to get the pages resident. This is a copy of ufs_trans_touch().
494  */
495 static void
496 zfs_prefault_write(ssize_t n, struct uio *uio)
497 {
498 	struct iovec *iov;
499 	ulong_t cnt, incr;
500 	caddr_t p;
501 	uint8_t tmp;
502 
503 	iov = uio->uio_iov;
504 
505 	while (n) {
506 		cnt = MIN(iov->iov_len, n);
507 		if (cnt == 0) {
508 			/* empty iov entry */
509 			iov++;
510 			continue;
511 		}
512 		n -= cnt;
513 		/*
514 		 * touch each page in this segment.
515 		 */
516 		p = iov->iov_base;
517 		while (cnt) {
518 			switch (uio->uio_segflg) {
519 			case UIO_USERSPACE:
520 			case UIO_USERISPACE:
521 				if (fuword8(p, &tmp))
522 					return;
523 				break;
524 			case UIO_SYSSPACE:
525 				if (kcopy(p, &tmp, 1))
526 					return;
527 				break;
528 			}
529 			incr = MIN(cnt, PAGESIZE);
530 			p += incr;
531 			cnt -= incr;
532 		}
533 		/*
534 		 * touch the last byte in case it straddles a page.
535 		 */
536 		p--;
537 		switch (uio->uio_segflg) {
538 		case UIO_USERSPACE:
539 		case UIO_USERISPACE:
540 			if (fuword8(p, &tmp))
541 				return;
542 			break;
543 		case UIO_SYSSPACE:
544 			if (kcopy(p, &tmp, 1))
545 				return;
546 			break;
547 		}
548 		iov++;
549 	}
550 }
551 
552 /*
553  * Write the bytes to a file.
554  *
555  *	IN:	vp	- vnode of file to be written to.
556  *		uio	- structure supplying write location, range info,
557  *			  and data buffer.
558  *		ioflag	- FAPPEND flag set if in append mode.
559  *		cr	- credentials of caller.
560  *
561  *	OUT:	uio	- updated offset and range.
562  *
563  *	RETURN:	0 if success
564  *		error code if failure
565  *
566  * Timestamps:
567  *	vp - ctime|mtime updated if byte count > 0
568  *
569  * Note: zfs_write() holds z_append_lock across calls to txg_wait_open().
570  * It has to because of the semantics of FAPPEND.  The implication is that
571  * we must never grab z_append_lock while in an assigned tx.
572  */
573 /* ARGSUSED */
574 static int
575 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
576 {
577 	znode_t		*zp = VTOZ(vp);
578 	rlim64_t	limit = uio->uio_llimit;
579 	ssize_t		start_resid = uio->uio_resid;
580 	ssize_t		tx_bytes;
581 	uint64_t	end_size;
582 	dmu_tx_t	*tx;
583 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
584 	zilog_t		*zilog = zfsvfs->z_log;
585 	uint64_t	seq = 0;
586 	offset_t	woff;
587 	ssize_t		n, nbytes;
588 	int		max_blksz = zfsvfs->z_max_blksz;
589 	int		need_append_lock, error;
590 	krw_t		grow_rw = RW_READER;
591 
592 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
593 		limit = MAXOFFSET_T;
594 
595 	n = start_resid;
596 
597 	/*
598 	 * Fasttrack empty write
599 	 */
600 	if (n == 0)
601 		return (0);
602 
603 	ZFS_ENTER(zfsvfs);
604 
605 	/*
606 	 * Pre-fault the pages to ensure slow (eg NFS) pages don't hold up txg
607 	 */
608 	zfs_prefault_write(MIN(start_resid, SPA_MAXBLOCKSIZE), uio);
609 
610 	/*
611 	 * If in append mode, set the io offset pointer to eof.
612 	 */
613 	need_append_lock = ioflag & FAPPEND;
614 	if (need_append_lock) {
615 		rw_enter(&zp->z_append_lock, RW_WRITER);
616 		woff = uio->uio_loffset = zp->z_phys->zp_size;
617 	} else {
618 		woff = uio->uio_loffset;
619 		/*
620 		 * Validate file offset
621 		 */
622 		if (woff < 0) {
623 			ZFS_EXIT(zfsvfs);
624 			return (EINVAL);
625 		}
626 
627 		/*
628 		 * If this write could change the file length,
629 		 * we need to synchronize with "appenders".
630 		 */
631 		if (woff < limit - n && woff + n > zp->z_phys->zp_size) {
632 			need_append_lock = TRUE;
633 			rw_enter(&zp->z_append_lock, RW_READER);
634 		}
635 	}
636 
637 	if (woff >= limit) {
638 		error = EFBIG;
639 		goto no_tx_done;
640 	}
641 
642 	if ((woff + n) > limit || woff > (limit - n))
643 		n = limit - woff;
644 
645 	/*
646 	 * Check for region locks
647 	 */
648 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) &&
649 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0)
650 		goto no_tx_done;
651 top:
652 	/*
653 	 * Make sure nobody restructures the file (changes block size)
654 	 * in the middle of the write.
655 	 */
656 	rw_enter(&zp->z_grow_lock, grow_rw);
657 
658 	end_size = MAX(zp->z_phys->zp_size, woff + n);
659 	tx = dmu_tx_create(zfsvfs->z_os);
660 	dmu_tx_hold_bonus(tx, zp->z_id);
661 	dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
662 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
663 	if (error) {
664 		dmu_tx_abort(tx);
665 		rw_exit(&zp->z_grow_lock);
666 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
667 			txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0);
668 			goto top;
669 		}
670 		goto no_tx_done;
671 	}
672 
673 	if (end_size > zp->z_blksz &&
674 	    (!ISP2(zp->z_blksz) || zp->z_blksz < max_blksz)) {
675 		uint64_t new_blksz;
676 		/*
677 		 * This write will increase the file size beyond
678 		 * the current block size so increase the block size.
679 		 */
680 		if (grow_rw == RW_READER && !rw_tryupgrade(&zp->z_grow_lock)) {
681 			dmu_tx_commit(tx);
682 			rw_exit(&zp->z_grow_lock);
683 			grow_rw = RW_WRITER;
684 			goto top;
685 		}
686 		if (zp->z_blksz > max_blksz) {
687 			ASSERT(!ISP2(zp->z_blksz));
688 			new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
689 		} else {
690 			new_blksz = MIN(end_size, max_blksz);
691 		}
692 		error = zfs_grow_blocksize(zp, new_blksz, tx);
693 		if (error) {
694 			tx_bytes = 0;
695 			goto tx_done;
696 		}
697 	}
698 
699 	if (grow_rw == RW_WRITER) {
700 		rw_downgrade(&zp->z_grow_lock);
701 		grow_rw = RW_READER;
702 	}
703 
704 	/*
705 	 * The file data does not fit in the znode "cache", so we
706 	 * will be writing to the file block data buffers.
707 	 * Each buffer will be written in a separate transaction;
708 	 * this keeps the intent log records small and allows us
709 	 * to do more fine-grained space accounting.
710 	 */
711 	while (n > 0) {
712 		/*
713 		 * XXX - should we really limit each write to z_max_blksz?
714 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
715 		 */
716 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
717 		rw_enter(&zp->z_map_lock, RW_READER);
718 
719 		tx_bytes = uio->uio_resid;
720 		if (vn_has_cached_data(vp)) {
721 			rw_exit(&zp->z_map_lock);
722 			error = mappedwrite(vp, woff, nbytes, uio, tx);
723 		} else {
724 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
725 			    woff, nbytes, uio, tx);
726 			rw_exit(&zp->z_map_lock);
727 		}
728 		tx_bytes -= uio->uio_resid;
729 
730 		if (error) {
731 			/* XXX - do we need to "clean up" the dmu buffer? */
732 			break;
733 		}
734 
735 		ASSERT(tx_bytes == nbytes);
736 
737 		n -= nbytes;
738 		if (n <= 0)
739 			break;
740 
741 		/*
742 		 * We have more work ahead of us, so wrap up this transaction
743 		 * and start another.  Exact same logic as tx_done below.
744 		 */
745 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) {
746 			dmu_buf_will_dirty(zp->z_dbuf, tx);
747 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
748 			    uio->uio_loffset);
749 		}
750 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
751 		seq = zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes,
752 		    ioflag, uio);
753 		dmu_tx_commit(tx);
754 
755 		/* Pre-fault the next set of pages */
756 		zfs_prefault_write(MIN(n, SPA_MAXBLOCKSIZE), uio);
757 
758 		/*
759 		 * Start another transaction.
760 		 */
761 		woff = uio->uio_loffset;
762 		tx = dmu_tx_create(zfsvfs->z_os);
763 		dmu_tx_hold_bonus(tx, zp->z_id);
764 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
765 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
766 		if (error) {
767 			dmu_tx_abort(tx);
768 			rw_exit(&zp->z_grow_lock);
769 			if (error == ERESTART &&
770 			    zfsvfs->z_assign == TXG_NOWAIT) {
771 				txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0);
772 				goto top;
773 			}
774 			goto no_tx_done;
775 		}
776 	}
777 
778 tx_done:
779 
780 	if (tx_bytes != 0) {
781 		/*
782 		 * Update the file size if it has changed; account
783 		 * for possible concurrent updates.
784 		 */
785 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) {
786 			dmu_buf_will_dirty(zp->z_dbuf, tx);
787 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
788 			    uio->uio_loffset);
789 		}
790 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
791 		seq = zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes,
792 		    ioflag, uio);
793 	}
794 	dmu_tx_commit(tx);
795 
796 	rw_exit(&zp->z_grow_lock);
797 
798 no_tx_done:
799 
800 	if (need_append_lock)
801 		rw_exit(&zp->z_append_lock);
802 
803 	/*
804 	 * If we're in replay mode, or we made no progress, return error.
805 	 * Otherwise, it's at least a partial write, so it's successful.
806 	 */
807 	if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) {
808 		ZFS_EXIT(zfsvfs);
809 		return (error);
810 	}
811 
812 	zil_commit(zilog, seq, ioflag & (FSYNC | FDSYNC));
813 
814 	ZFS_EXIT(zfsvfs);
815 	return (0);
816 }
817 
818 /*
819  * Get data to generate a TX_WRITE intent log record.
820  */
821 int
822 zfs_get_data(void *arg, lr_write_t *lr)
823 {
824 	zfsvfs_t *zfsvfs = arg;
825 	objset_t *os = zfsvfs->z_os;
826 	znode_t *zp;
827 	uint64_t off = lr->lr_offset;
828 	int dlen = lr->lr_length;  		/* length of user data */
829 	int reclen = lr->lr_common.lrc_reclen;
830 	int error = 0;
831 
832 	ASSERT(dlen != 0);
833 
834 	/*
835 	 * Nothing to do if the file has been removed or truncated.
836 	 */
837 	if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0)
838 		return (ENOENT);
839 	if (off >= zp->z_phys->zp_size || zp->z_reap) {
840 		VN_RELE(ZTOV(zp));
841 		return (ENOENT);
842 	}
843 
844 	/*
845 	 * Write records come in two flavors: immediate and indirect.
846 	 * For small writes it's cheaper to store the data with the
847 	 * log record (immediate); for large writes it's cheaper to
848 	 * sync the data and get a pointer to it (indirect) so that
849 	 * we don't have to write the data twice.
850 	 */
851 	if (sizeof (lr_write_t) + dlen <= reclen) { /* immediate write */
852 		rw_enter(&zp->z_grow_lock, RW_READER);
853 		dmu_buf_t *db = dmu_buf_hold(os, lr->lr_foid, off);
854 		dmu_buf_read(db);
855 		bcopy((char *)db->db_data + off - db->db_offset, lr + 1, dlen);
856 		dmu_buf_rele(db);
857 		rw_exit(&zp->z_grow_lock);
858 	} else {
859 		/*
860 		 * We have to grab z_grow_lock as RW_WRITER because
861 		 * dmu_sync() can't handle concurrent dbuf_dirty() (6313856).
862 		 * z_grow_lock will be replaced with a range lock soon,
863 		 * which will eliminate the concurrency hit, but dmu_sync()
864 		 * really needs more thought.  It shouldn't have to rely on
865 		 * the caller to provide MT safety.
866 		 */
867 		rw_enter(&zp->z_grow_lock, RW_WRITER);
868 		txg_suspend(dmu_objset_pool(os));
869 		error = dmu_sync(os, lr->lr_foid, off, &lr->lr_blkoff,
870 		    &lr->lr_blkptr, lr->lr_common.lrc_txg);
871 		txg_resume(dmu_objset_pool(os));
872 		rw_exit(&zp->z_grow_lock);
873 	}
874 	VN_RELE(ZTOV(zp));
875 	return (error);
876 }
877 
878 /*ARGSUSED*/
879 static int
880 zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr)
881 {
882 	znode_t *zp = VTOZ(vp);
883 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
884 	int error;
885 
886 	ZFS_ENTER(zfsvfs);
887 	error = zfs_zaccess_rwx(zp, mode, cr);
888 	ZFS_EXIT(zfsvfs);
889 	return (error);
890 }
891 
892 /*
893  * Lookup an entry in a directory, or an extended attribute directory.
894  * If it exists, return a held vnode reference for it.
895  *
896  *	IN:	dvp	- vnode of directory to search.
897  *		nm	- name of entry to lookup.
898  *		pnp	- full pathname to lookup [UNUSED].
899  *		flags	- LOOKUP_XATTR set if looking for an attribute.
900  *		rdir	- root directory vnode [UNUSED].
901  *		cr	- credentials of caller.
902  *
903  *	OUT:	vpp	- vnode of located entry, NULL if not found.
904  *
905  *	RETURN:	0 if success
906  *		error code if failure
907  *
908  * Timestamps:
909  *	NA
910  */
911 /* ARGSUSED */
912 static int
913 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
914     int flags, vnode_t *rdir, cred_t *cr)
915 {
916 
917 	znode_t *zdp = VTOZ(dvp);
918 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
919 	int	error;
920 
921 	ZFS_ENTER(zfsvfs);
922 
923 	*vpp = NULL;
924 
925 	if (flags & LOOKUP_XATTR) {
926 		/*
927 		 * We don't allow recursive attributes..
928 		 * Maybe someday we will.
929 		 */
930 		if (zdp->z_phys->zp_flags & ZFS_XATTR) {
931 			ZFS_EXIT(zfsvfs);
932 			return (EINVAL);
933 		}
934 
935 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr)) {
936 			ZFS_EXIT(zfsvfs);
937 			return (error);
938 		}
939 
940 		/*
941 		 * Do we have permission to get into attribute directory?
942 		 */
943 
944 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) {
945 			VN_RELE(*vpp);
946 		}
947 
948 		ZFS_EXIT(zfsvfs);
949 		return (error);
950 	}
951 
952 	if (dvp->v_type != VDIR) {
953 		ZFS_EXIT(zfsvfs);
954 		return (ENOTDIR);
955 	}
956 
957 	/*
958 	 * Check accessibility of directory.
959 	 */
960 
961 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) {
962 		ZFS_EXIT(zfsvfs);
963 		return (error);
964 	}
965 
966 	if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) {
967 
968 		/*
969 		 * Convert device special files
970 		 */
971 		if (IS_DEVVP(*vpp)) {
972 			vnode_t	*svp;
973 
974 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
975 			VN_RELE(*vpp);
976 			if (svp == NULL)
977 				error = ENOSYS;
978 			else
979 				*vpp = svp;
980 		}
981 	}
982 
983 	ZFS_EXIT(zfsvfs);
984 	return (error);
985 }
986 
987 /*
988  * Attempt to create a new entry in a directory.  If the entry
989  * already exists, truncate the file if permissible, else return
990  * an error.  Return the vp of the created or trunc'd file.
991  *
992  *	IN:	dvp	- vnode of directory to put new file entry in.
993  *		name	- name of new file entry.
994  *		vap	- attributes of new file.
995  *		excl	- flag indicating exclusive or non-exclusive mode.
996  *		mode	- mode to open file with.
997  *		cr	- credentials of caller.
998  *		flag	- large file flag [UNUSED].
999  *
1000  *	OUT:	vpp	- vnode of created or trunc'd entry.
1001  *
1002  *	RETURN:	0 if success
1003  *		error code if failure
1004  *
1005  * Timestamps:
1006  *	dvp - ctime|mtime updated if new entry created
1007  *	 vp - ctime|mtime always, atime if new
1008  */
1009 /* ARGSUSED */
1010 static int
1011 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1012     int mode, vnode_t **vpp, cred_t *cr, int flag)
1013 {
1014 	znode_t		*zp, *dzp = VTOZ(dvp);
1015 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1016 	zilog_t		*zilog = zfsvfs->z_log;
1017 	uint64_t	seq = 0;
1018 	objset_t	*os = zfsvfs->z_os;
1019 	zfs_dirlock_t	*dl;
1020 	dmu_tx_t	*tx;
1021 	int		error;
1022 	uint64_t	zoid;
1023 
1024 	ZFS_ENTER(zfsvfs);
1025 
1026 top:
1027 	*vpp = NULL;
1028 
1029 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1030 		vap->va_mode &= ~VSVTX;
1031 
1032 	if (*name == '\0') {
1033 		/*
1034 		 * Null component name refers to the directory itself.
1035 		 */
1036 		VN_HOLD(dvp);
1037 		zp = dzp;
1038 		dl = NULL;
1039 		error = 0;
1040 	} else {
1041 		/* possible VN_HOLD(zp) */
1042 		if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) {
1043 			if (strcmp(name, "..") == 0)
1044 				error = EISDIR;
1045 			ZFS_EXIT(zfsvfs);
1046 			return (error);
1047 		}
1048 	}
1049 
1050 	zoid = zp ? zp->z_id : -1ULL;
1051 
1052 	if (zp == NULL) {
1053 		/*
1054 		 * Create a new file object and update the directory
1055 		 * to reference it.
1056 		 */
1057 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
1058 			goto out;
1059 		}
1060 
1061 		/*
1062 		 * We only support the creation of regular files in
1063 		 * extended attribute directories.
1064 		 */
1065 		if ((dzp->z_phys->zp_flags & ZFS_XATTR) &&
1066 		    (vap->va_type != VREG)) {
1067 			error = EINVAL;
1068 			goto out;
1069 		}
1070 
1071 		tx = dmu_tx_create(os);
1072 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1073 		dmu_tx_hold_bonus(tx, dzp->z_id);
1074 		dmu_tx_hold_zap(tx, dzp->z_id, 1);
1075 		if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
1076 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1077 			    0, SPA_MAXBLOCKSIZE);
1078 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
1079 		if (error) {
1080 			dmu_tx_abort(tx);
1081 			zfs_dirent_unlock(dl);
1082 			if (error == ERESTART &&
1083 			    zfsvfs->z_assign == TXG_NOWAIT) {
1084 				txg_wait_open(dmu_objset_pool(os), 0);
1085 				goto top;
1086 			}
1087 			ZFS_EXIT(zfsvfs);
1088 			return (error);
1089 		}
1090 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
1091 		ASSERT(zp->z_id == zoid);
1092 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1093 		seq = zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name);
1094 		dmu_tx_commit(tx);
1095 	} else {
1096 		/*
1097 		 * A directory entry already exists for this name.
1098 		 */
1099 		/*
1100 		 * Can't truncate an existing file if in exclusive mode.
1101 		 */
1102 		if (excl == EXCL) {
1103 			error = EEXIST;
1104 			goto out;
1105 		}
1106 		/*
1107 		 * Can't open a directory for writing.
1108 		 */
1109 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1110 			error = EISDIR;
1111 			goto out;
1112 		}
1113 		/*
1114 		 * Verify requested access to file.
1115 		 */
1116 		if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) {
1117 			goto out;
1118 		}
1119 		/*
1120 		 * Truncate regular files if requested.
1121 		 */
1122 
1123 		/*
1124 		 * Need to update dzp->z_seq?
1125 		 */
1126 
1127 		mutex_enter(&dzp->z_lock);
1128 		dzp->z_seq++;
1129 		mutex_exit(&dzp->z_lock);
1130 
1131 		if ((ZTOV(zp)->v_type == VREG) && (zp->z_phys->zp_size != 0) &&
1132 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1133 			/*
1134 			 * Truncate the file.
1135 			 */
1136 			tx = dmu_tx_create(os);
1137 			dmu_tx_hold_bonus(tx, zoid);
1138 			dmu_tx_hold_free(tx, zoid, 0, DMU_OBJECT_END);
1139 			error = dmu_tx_assign(tx, zfsvfs->z_assign);
1140 			if (error) {
1141 				dmu_tx_abort(tx);
1142 				if (dl)
1143 					zfs_dirent_unlock(dl);
1144 				VN_RELE(ZTOV(zp));
1145 				if (error == ERESTART &&
1146 				    zfsvfs->z_assign == TXG_NOWAIT) {
1147 					txg_wait_open(dmu_objset_pool(os), 0);
1148 					goto top;
1149 				}
1150 				ZFS_EXIT(zfsvfs);
1151 				return (error);
1152 			}
1153 			/*
1154 			 * Grab the grow_lock to serialize this change with
1155 			 * respect to other file manipulations.
1156 			 */
1157 			rw_enter(&zp->z_grow_lock, RW_WRITER);
1158 			error = zfs_freesp(zp, 0, 0, mode, tx, cr);
1159 			if (error == 0) {
1160 				zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
1161 				seq = zfs_log_truncate(zilog, tx,
1162 				    TX_TRUNCATE, zp, 0, 0);
1163 			}
1164 			rw_exit(&zp->z_grow_lock);
1165 			dmu_tx_commit(tx);
1166 		}
1167 	}
1168 out:
1169 
1170 	if (dl)
1171 		zfs_dirent_unlock(dl);
1172 
1173 	if (error) {
1174 		if (zp)
1175 			VN_RELE(ZTOV(zp));
1176 	} else {
1177 		*vpp = ZTOV(zp);
1178 		/*
1179 		 * If vnode is for a device return a specfs vnode instead.
1180 		 */
1181 		if (IS_DEVVP(*vpp)) {
1182 			struct vnode *svp;
1183 
1184 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1185 			VN_RELE(*vpp);
1186 			if (svp == NULL) {
1187 				error = ENOSYS;
1188 			}
1189 			*vpp = svp;
1190 		}
1191 	}
1192 
1193 	zil_commit(zilog, seq, 0);
1194 
1195 	ZFS_EXIT(zfsvfs);
1196 	return (error);
1197 }
1198 
1199 /*
1200  * Remove an entry from a directory.
1201  *
1202  *	IN:	dvp	- vnode of directory to remove entry from.
1203  *		name	- name of entry to remove.
1204  *		cr	- credentials of caller.
1205  *
1206  *	RETURN:	0 if success
1207  *		error code if failure
1208  *
1209  * Timestamps:
1210  *	dvp - ctime|mtime
1211  *	 vp - ctime (if nlink > 0)
1212  */
1213 static int
1214 zfs_remove(vnode_t *dvp, char *name, cred_t *cr)
1215 {
1216 	znode_t		*zp, *dzp = VTOZ(dvp);
1217 	znode_t		*xzp = NULL;
1218 	vnode_t		*vp;
1219 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1220 	zilog_t		*zilog = zfsvfs->z_log;
1221 	uint64_t	seq = 0;
1222 	uint64_t	acl_obj, xattr_obj;
1223 	zfs_dirlock_t	*dl;
1224 	dmu_tx_t	*tx;
1225 	int		may_delete_now, delete_now = FALSE;
1226 	int		reaped;
1227 	int		error;
1228 
1229 	ZFS_ENTER(zfsvfs);
1230 
1231 top:
1232 	/*
1233 	 * Attempt to lock directory; fail if entry doesn't exist.
1234 	 */
1235 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) {
1236 		ZFS_EXIT(zfsvfs);
1237 		return (error);
1238 	}
1239 
1240 	vp = ZTOV(zp);
1241 
1242 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1243 		goto out;
1244 	}
1245 
1246 	/*
1247 	 * Need to use rmdir for removing directories.
1248 	 */
1249 	if (vp->v_type == VDIR) {
1250 		error = EPERM;
1251 		goto out;
1252 	}
1253 
1254 	vnevent_remove(vp);
1255 
1256 	dnlc_remove(dvp, name);
1257 
1258 	mutex_enter(&vp->v_lock);
1259 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1260 	mutex_exit(&vp->v_lock);
1261 
1262 	/*
1263 	 * We may delete the znode now, or we may put it on the delete queue;
1264 	 * it depends on whether we're the last link, and on whether there are
1265 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1266 	 * allow for either case.
1267 	 */
1268 	tx = dmu_tx_create(zfsvfs->z_os);
1269 	dmu_tx_hold_zap(tx, dzp->z_id, -1);
1270 	dmu_tx_hold_bonus(tx, zp->z_id);
1271 	if (may_delete_now)
1272 		dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
1273 
1274 	/* are there any extended attributes? */
1275 	if ((xattr_obj = zp->z_phys->zp_xattr) != 0) {
1276 		/*
1277 		 * XXX - There is a possibility that the delete
1278 		 * of the parent file could succeed, but then we get
1279 		 * an ENOSPC when we try to delete the xattrs...
1280 		 * so we would need to re-try the deletes periodically
1281 		 */
1282 		/* XXX - do we need this if we are deleting? */
1283 		dmu_tx_hold_bonus(tx, xattr_obj);
1284 	}
1285 
1286 	/* are there any additional acls */
1287 	if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 &&
1288 	    may_delete_now)
1289 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1290 
1291 	/* charge as an update -- would be nice not to charge at all */
1292 	dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, -1);
1293 
1294 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1295 	if (error) {
1296 		dmu_tx_abort(tx);
1297 		zfs_dirent_unlock(dl);
1298 		VN_RELE(vp);
1299 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1300 			txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0);
1301 			goto top;
1302 		}
1303 		ZFS_EXIT(zfsvfs);
1304 		return (error);
1305 	}
1306 
1307 	/*
1308 	 * Remove the directory entry.
1309 	 */
1310 	error = zfs_link_destroy(dl, zp, tx, 0, &reaped);
1311 
1312 	if (error) {
1313 		dmu_tx_commit(tx);
1314 		goto out;
1315 	}
1316 
1317 	if (reaped) {
1318 		mutex_enter(&vp->v_lock);
1319 		delete_now = may_delete_now &&
1320 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1321 		    zp->z_phys->zp_xattr == xattr_obj &&
1322 		    zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj;
1323 		mutex_exit(&vp->v_lock);
1324 	}
1325 
1326 	if (delete_now) {
1327 		if (zp->z_phys->zp_xattr) {
1328 			error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
1329 			ASSERT3U(error, ==, 0);
1330 			ASSERT3U(xzp->z_phys->zp_links, ==, 2);
1331 			dmu_buf_will_dirty(xzp->z_dbuf, tx);
1332 			mutex_enter(&xzp->z_lock);
1333 			xzp->z_reap = 1;
1334 			xzp->z_phys->zp_links = 0;
1335 			mutex_exit(&xzp->z_lock);
1336 			zfs_dq_add(xzp, tx);
1337 			zp->z_phys->zp_xattr = 0; /* probably unnecessary */
1338 		}
1339 		mutex_enter(&zp->z_lock);
1340 		mutex_enter(&vp->v_lock);
1341 		vp->v_count--;
1342 		ASSERT3U(vp->v_count, ==, 0);
1343 		mutex_exit(&vp->v_lock);
1344 		zp->z_active = 0;
1345 		mutex_exit(&zp->z_lock);
1346 		zfs_znode_delete(zp, tx);
1347 		VFS_RELE(zfsvfs->z_vfs);
1348 	} else if (reaped) {
1349 		zfs_dq_add(zp, tx);
1350 	}
1351 
1352 	seq = zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name);
1353 
1354 	dmu_tx_commit(tx);
1355 out:
1356 	zfs_dirent_unlock(dl);
1357 
1358 	if (!delete_now) {
1359 		VN_RELE(vp);
1360 	} else if (xzp) {
1361 		/* this rele delayed to prevent nesting transactions */
1362 		VN_RELE(ZTOV(xzp));
1363 	}
1364 
1365 	zil_commit(zilog, seq, 0);
1366 
1367 	ZFS_EXIT(zfsvfs);
1368 	return (error);
1369 }
1370 
1371 /*
1372  * Create a new directory and insert it into dvp using the name
1373  * provided.  Return a pointer to the inserted directory.
1374  *
1375  *	IN:	dvp	- vnode of directory to add subdir to.
1376  *		dirname	- name of new directory.
1377  *		vap	- attributes of new directory.
1378  *		cr	- credentials of caller.
1379  *
1380  *	OUT:	vpp	- vnode of created directory.
1381  *
1382  *	RETURN:	0 if success
1383  *		error code if failure
1384  *
1385  * Timestamps:
1386  *	dvp - ctime|mtime updated
1387  *	 vp - ctime|mtime|atime updated
1388  */
1389 static int
1390 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr)
1391 {
1392 	znode_t		*zp, *dzp = VTOZ(dvp);
1393 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1394 	zilog_t		*zilog = zfsvfs->z_log;
1395 	uint64_t	seq = 0;
1396 	zfs_dirlock_t	*dl;
1397 	uint64_t	zoid = 0;
1398 	dmu_tx_t	*tx;
1399 	int		error;
1400 
1401 	ASSERT(vap->va_type == VDIR);
1402 
1403 	ZFS_ENTER(zfsvfs);
1404 
1405 	if (dzp->z_phys->zp_flags & ZFS_XATTR) {
1406 		ZFS_EXIT(zfsvfs);
1407 		return (EINVAL);
1408 	}
1409 top:
1410 	*vpp = NULL;
1411 
1412 	/*
1413 	 * First make sure the new directory doesn't exist.
1414 	 */
1415 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) {
1416 		ZFS_EXIT(zfsvfs);
1417 		return (error);
1418 	}
1419 
1420 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) {
1421 		zfs_dirent_unlock(dl);
1422 		ZFS_EXIT(zfsvfs);
1423 		return (error);
1424 	}
1425 
1426 	/*
1427 	 * Add a new entry to the directory.
1428 	 */
1429 	tx = dmu_tx_create(zfsvfs->z_os);
1430 	dmu_tx_hold_zap(tx, dzp->z_id, 1);
1431 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 0);
1432 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
1433 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1434 		    0, SPA_MAXBLOCKSIZE);
1435 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1436 	if (error) {
1437 		dmu_tx_abort(tx);
1438 		zfs_dirent_unlock(dl);
1439 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1440 			txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0);
1441 			goto top;
1442 		}
1443 		ZFS_EXIT(zfsvfs);
1444 		return (error);
1445 	}
1446 
1447 	/*
1448 	 * Create new node.
1449 	 */
1450 	zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
1451 
1452 	/*
1453 	 * Now put new name in parent dir.
1454 	 */
1455 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1456 
1457 	*vpp = ZTOV(zp);
1458 
1459 	seq = zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname);
1460 	dmu_tx_commit(tx);
1461 
1462 	zfs_dirent_unlock(dl);
1463 
1464 	zil_commit(zilog, seq, 0);
1465 
1466 	ZFS_EXIT(zfsvfs);
1467 	return (0);
1468 }
1469 
1470 /*
1471  * Remove a directory subdir entry.  If the current working
1472  * directory is the same as the subdir to be removed, the
1473  * remove will fail.
1474  *
1475  *	IN:	dvp	- vnode of directory to remove from.
1476  *		name	- name of directory to be removed.
1477  *		cwd	- vnode of current working directory.
1478  *		cr	- credentials of caller.
1479  *
1480  *	RETURN:	0 if success
1481  *		error code if failure
1482  *
1483  * Timestamps:
1484  *	dvp - ctime|mtime updated
1485  */
1486 static int
1487 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr)
1488 {
1489 	znode_t		*dzp = VTOZ(dvp);
1490 	znode_t		*zp;
1491 	vnode_t		*vp;
1492 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1493 	zilog_t		*zilog = zfsvfs->z_log;
1494 	uint64_t	seq = 0;
1495 	zfs_dirlock_t	*dl;
1496 	dmu_tx_t	*tx;
1497 	int		error;
1498 
1499 	ZFS_ENTER(zfsvfs);
1500 
1501 top:
1502 	zp = NULL;
1503 
1504 	/*
1505 	 * Attempt to lock directory; fail if entry doesn't exist.
1506 	 */
1507 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) {
1508 		ZFS_EXIT(zfsvfs);
1509 		return (error);
1510 	}
1511 
1512 	vp = ZTOV(zp);
1513 
1514 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1515 		goto out;
1516 	}
1517 
1518 	if (vp->v_type != VDIR) {
1519 		error = ENOTDIR;
1520 		goto out;
1521 	}
1522 
1523 	if (vp == cwd) {
1524 		error = EINVAL;
1525 		goto out;
1526 	}
1527 
1528 	vnevent_rmdir(vp);
1529 
1530 	/*
1531 	 * Grab a lock on the parent pointer make sure we play well
1532 	 * with the treewalk and directory rename code.
1533 	 */
1534 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1535 
1536 	tx = dmu_tx_create(zfsvfs->z_os);
1537 	dmu_tx_hold_zap(tx, dzp->z_id, 1);
1538 	dmu_tx_hold_bonus(tx, zp->z_id);
1539 	dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, 1);
1540 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1541 	if (error) {
1542 		dmu_tx_abort(tx);
1543 		rw_exit(&zp->z_parent_lock);
1544 		zfs_dirent_unlock(dl);
1545 		VN_RELE(vp);
1546 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1547 			txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0);
1548 			goto top;
1549 		}
1550 		ZFS_EXIT(zfsvfs);
1551 		return (error);
1552 	}
1553 
1554 	error = zfs_link_destroy(dl, zp, tx, 0, NULL);
1555 
1556 	if (error == 0)
1557 		seq = zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name);
1558 
1559 	dmu_tx_commit(tx);
1560 
1561 	rw_exit(&zp->z_parent_lock);
1562 out:
1563 	zfs_dirent_unlock(dl);
1564 
1565 	VN_RELE(vp);
1566 
1567 	zil_commit(zilog, seq, 0);
1568 
1569 	ZFS_EXIT(zfsvfs);
1570 	return (error);
1571 }
1572 
1573 /*
1574  * Read as many directory entries as will fit into the provided
1575  * buffer from the given directory cursor position (specified in
1576  * the uio structure.
1577  *
1578  *	IN:	vp	- vnode of directory to read.
1579  *		uio	- structure supplying read location, range info,
1580  *			  and return buffer.
1581  *		cr	- credentials of caller.
1582  *
1583  *	OUT:	uio	- updated offset and range, buffer filled.
1584  *		eofp	- set to true if end-of-file detected.
1585  *
1586  *	RETURN:	0 if success
1587  *		error code if failure
1588  *
1589  * Timestamps:
1590  *	vp - atime updated
1591  *
1592  * Note that the low 4 bits of the cookie returned by zap is always zero.
1593  * This allows us to use the low range for "special" directory entries:
1594  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1595  * we use the offset 2 for the '.zfs' directory.
1596  */
1597 /* ARGSUSED */
1598 static int
1599 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp)
1600 {
1601 	znode_t		*zp = VTOZ(vp);
1602 	iovec_t		*iovp;
1603 	dirent64_t	*odp;
1604 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1605 	objset_t	*os;
1606 	caddr_t		outbuf;
1607 	size_t		bufsize;
1608 	zap_cursor_t	zc;
1609 	zap_attribute_t	zap;
1610 	uint_t		bytes_wanted;
1611 	ushort_t	this_reclen;
1612 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1613 	off64_t		*next;
1614 	int		local_eof;
1615 	int		outcount;
1616 	int		error;
1617 	uint8_t		prefetch;
1618 
1619 	ZFS_ENTER(zfsvfs);
1620 
1621 	/*
1622 	 * If we are not given an eof variable,
1623 	 * use a local one.
1624 	 */
1625 	if (eofp == NULL)
1626 		eofp = &local_eof;
1627 
1628 	/*
1629 	 * Check for valid iov_len.
1630 	 */
1631 	if (uio->uio_iov->iov_len <= 0) {
1632 		ZFS_EXIT(zfsvfs);
1633 		return (EINVAL);
1634 	}
1635 
1636 	/*
1637 	 * Quit if directory has been removed (posix)
1638 	 */
1639 	if ((*eofp = zp->z_reap) != 0) {
1640 		ZFS_EXIT(zfsvfs);
1641 		return (0);
1642 	}
1643 
1644 	error = 0;
1645 	os = zfsvfs->z_os;
1646 	offset = uio->uio_loffset;
1647 	prefetch = zp->z_zn_prefetch;
1648 
1649 	/*
1650 	 * Initialize the iterator cursor.
1651 	 */
1652 	if (offset <= 3) {
1653 		/*
1654 		 * Start iteration from the beginning of the directory.
1655 		 */
1656 		zap_cursor_init(&zc, os, zp->z_id);
1657 	} else {
1658 		/*
1659 		 * The offset is a serialized cursor.
1660 		 */
1661 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1662 	}
1663 
1664 	/*
1665 	 * Get space to change directory entries into fs independent format.
1666 	 */
1667 	iovp = uio->uio_iov;
1668 	bytes_wanted = iovp->iov_len;
1669 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
1670 		bufsize = bytes_wanted;
1671 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
1672 		odp = (struct dirent64 *)outbuf;
1673 	} else {
1674 		bufsize = bytes_wanted;
1675 		odp = (struct dirent64 *)iovp->iov_base;
1676 	}
1677 
1678 	/*
1679 	 * Transform to file-system independent format
1680 	 */
1681 	outcount = 0;
1682 	while (outcount < bytes_wanted) {
1683 		/*
1684 		 * Special case `.', `..', and `.zfs'.
1685 		 */
1686 		if (offset == 0) {
1687 			(void) strcpy(zap.za_name, ".");
1688 			zap.za_first_integer = zp->z_id;
1689 			this_reclen = DIRENT64_RECLEN(1);
1690 		} else if (offset == 1) {
1691 			(void) strcpy(zap.za_name, "..");
1692 			zap.za_first_integer = zp->z_phys->zp_parent;
1693 			this_reclen = DIRENT64_RECLEN(2);
1694 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
1695 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1696 			zap.za_first_integer = ZFSCTL_INO_ROOT;
1697 			this_reclen =
1698 			    DIRENT64_RECLEN(sizeof (ZFS_CTLDIR_NAME) - 1);
1699 		} else {
1700 			/*
1701 			 * Grab next entry.
1702 			 */
1703 			if (error = zap_cursor_retrieve(&zc, &zap)) {
1704 				if ((*eofp = (error == ENOENT)) != 0)
1705 					break;
1706 				else
1707 					goto update;
1708 			}
1709 
1710 			if (zap.za_integer_length != 8 ||
1711 			    zap.za_num_integers != 1) {
1712 				cmn_err(CE_WARN, "zap_readdir: bad directory "
1713 				    "entry, obj = %lld, offset = %lld\n",
1714 				    (u_longlong_t)zp->z_id,
1715 				    (u_longlong_t)offset);
1716 				error = ENXIO;
1717 				goto update;
1718 			}
1719 			this_reclen = DIRENT64_RECLEN(strlen(zap.za_name));
1720 		}
1721 
1722 		/*
1723 		 * Will this entry fit in the buffer?
1724 		 */
1725 		if (outcount + this_reclen > bufsize) {
1726 			/*
1727 			 * Did we manage to fit anything in the buffer?
1728 			 */
1729 			if (!outcount) {
1730 				error = EINVAL;
1731 				goto update;
1732 			}
1733 			break;
1734 		}
1735 		/*
1736 		 * Add this entry:
1737 		 */
1738 		odp->d_ino = (ino64_t)zap.za_first_integer;
1739 		odp->d_reclen = (ushort_t)this_reclen;
1740 		/* NOTE: d_off is the offset for the *next* entry */
1741 		next = &(odp->d_off);
1742 		(void) strncpy(odp->d_name, zap.za_name,
1743 		    DIRENT64_NAMELEN(this_reclen));
1744 		outcount += this_reclen;
1745 		odp = (dirent64_t *)((intptr_t)odp + this_reclen);
1746 
1747 		ASSERT(outcount <= bufsize);
1748 
1749 		/* Prefetch znode */
1750 		if (prefetch)
1751 			dmu_prefetch(os, zap.za_first_integer, 0, 0);
1752 
1753 		/*
1754 		 * Move to the next entry, fill in the previous offset.
1755 		 */
1756 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
1757 			zap_cursor_advance(&zc);
1758 			offset = zap_cursor_serialize(&zc);
1759 		} else {
1760 			offset += 1;
1761 		}
1762 		*next = offset;
1763 	}
1764 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
1765 
1766 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
1767 		iovp->iov_base += outcount;
1768 		iovp->iov_len -= outcount;
1769 		uio->uio_resid -= outcount;
1770 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
1771 		/*
1772 		 * Reset the pointer.
1773 		 */
1774 		offset = uio->uio_loffset;
1775 	}
1776 
1777 update:
1778 	zap_cursor_fini(&zc);
1779 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
1780 		kmem_free(outbuf, bufsize);
1781 
1782 	if (error == ENOENT)
1783 		error = 0;
1784 
1785 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
1786 
1787 	uio->uio_loffset = offset;
1788 	ZFS_EXIT(zfsvfs);
1789 	return (error);
1790 }
1791 
1792 /* ARGSUSED */
1793 static int
1794 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr)
1795 {
1796 	znode_t	*zp = VTOZ(vp);
1797 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1798 
1799 	ZFS_ENTER(zfsvfs);
1800 	zil_commit(zfsvfs->z_log, zp->z_last_itx, FSYNC);
1801 	ZFS_EXIT(zfsvfs);
1802 	return (0);
1803 }
1804 
1805 /*
1806  * Get the requested file attributes and place them in the provided
1807  * vattr structure.
1808  *
1809  *	IN:	vp	- vnode of file.
1810  *		vap	- va_mask identifies requested attributes.
1811  *		flags	- [UNUSED]
1812  *		cr	- credentials of caller.
1813  *
1814  *	OUT:	vap	- attribute values.
1815  *
1816  *	RETURN:	0 (always succeeds)
1817  */
1818 /* ARGSUSED */
1819 static int
1820 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr)
1821 {
1822 	znode_t *zp = VTOZ(vp);
1823 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1824 	znode_phys_t *pzp = zp->z_phys;
1825 	int	error;
1826 
1827 	ZFS_ENTER(zfsvfs);
1828 
1829 	/*
1830 	 * Return all attributes.  It's cheaper to provide the answer
1831 	 * than to determine whether we were asked the question.
1832 	 */
1833 	mutex_enter(&zp->z_lock);
1834 
1835 	vap->va_type = vp->v_type;
1836 	vap->va_mode = pzp->zp_mode & MODEMASK;
1837 	vap->va_uid = zp->z_phys->zp_uid;
1838 	vap->va_gid = zp->z_phys->zp_gid;
1839 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
1840 	vap->va_nodeid = zp->z_id;
1841 	vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX);	/* nlink_t limit! */
1842 	vap->va_size = pzp->zp_size;
1843 	vap->va_rdev = pzp->zp_rdev;
1844 	vap->va_seq = zp->z_seq;
1845 
1846 	ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime);
1847 	ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime);
1848 	ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime);
1849 
1850 	/*
1851 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
1852 	 * Also, if we are the owner don't bother, since owner should
1853 	 * always be allowed to read basic attributes of file.
1854 	 */
1855 	if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) &&
1856 	    (zp->z_phys->zp_uid != crgetuid(cr))) {
1857 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) {
1858 			mutex_exit(&zp->z_lock);
1859 			ZFS_EXIT(zfsvfs);
1860 			return (error);
1861 		}
1862 	}
1863 
1864 	mutex_exit(&zp->z_lock);
1865 
1866 	dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks);
1867 
1868 	if (zp->z_blksz == 0) {
1869 		/*
1870 		 * Block size hasn't been set; suggest maximal I/O transfers.
1871 		 */
1872 		vap->va_blksize = zfsvfs->z_max_blksz;
1873 	}
1874 
1875 	ZFS_EXIT(zfsvfs);
1876 	return (0);
1877 }
1878 
1879 /*
1880  * Set the file attributes to the values contained in the
1881  * vattr structure.
1882  *
1883  *	IN:	vp	- vnode of file to be modified.
1884  *		vap	- new attribute values.
1885  *		flags	- ATTR_UTIME set if non-default time values provided.
1886  *		cr	- credentials of caller.
1887  *
1888  *	RETURN:	0 if success
1889  *		error code if failure
1890  *
1891  * Timestamps:
1892  *	vp - ctime updated, mtime updated if size changed.
1893  */
1894 /* ARGSUSED */
1895 static int
1896 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1897 	caller_context_t *ct)
1898 {
1899 	struct znode	*zp = VTOZ(vp);
1900 	znode_phys_t	*pzp = zp->z_phys;
1901 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1902 	zilog_t		*zilog = zfsvfs->z_log;
1903 	uint64_t	seq = 0;
1904 	dmu_tx_t	*tx;
1905 	uint_t		mask = vap->va_mask;
1906 	uint_t		mask_applied = 0;
1907 	vattr_t		oldva;
1908 	int		trim_mask = FALSE;
1909 	int		saved_mask;
1910 	uint64_t	new_mode;
1911 	znode_t		*attrzp;
1912 	int		have_grow_lock;
1913 	int		need_policy = FALSE;
1914 	int		err;
1915 
1916 	if (mask == 0)
1917 		return (0);
1918 
1919 	if (mask & AT_NOSET)
1920 		return (EINVAL);
1921 
1922 	if (mask & AT_SIZE && vp->v_type == VDIR)
1923 		return (EISDIR);
1924 
1925 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO)
1926 		return (EINVAL);
1927 
1928 	ZFS_ENTER(zfsvfs);
1929 
1930 top:
1931 	have_grow_lock = FALSE;
1932 	attrzp = NULL;
1933 
1934 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
1935 		ZFS_EXIT(zfsvfs);
1936 		return (EROFS);
1937 	}
1938 
1939 	/*
1940 	 * First validate permissions
1941 	 */
1942 
1943 	if (mask & AT_SIZE) {
1944 		err = zfs_zaccess(zp, ACE_WRITE_DATA, cr);
1945 		if (err) {
1946 			ZFS_EXIT(zfsvfs);
1947 			return (err);
1948 		}
1949 	}
1950 
1951 	if (mask & (AT_ATIME|AT_MTIME))
1952 		need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr);
1953 
1954 	if (mask & (AT_UID|AT_GID)) {
1955 		int	idmask = (mask & (AT_UID|AT_GID));
1956 		int	take_owner;
1957 		int	take_group;
1958 
1959 		/*
1960 		 * NOTE: even if a new mode is being set,
1961 		 * we may clear S_ISUID/S_ISGID bits.
1962 		 */
1963 
1964 		if (!(mask & AT_MODE))
1965 			vap->va_mode = pzp->zp_mode;
1966 
1967 		/*
1968 		 * Take ownership or chgrp to group we are a member of
1969 		 */
1970 
1971 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
1972 		take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr);
1973 
1974 		/*
1975 		 * If both AT_UID and AT_GID are set then take_owner and
1976 		 * take_group must both be set in order to allow taking
1977 		 * ownership.
1978 		 *
1979 		 * Otherwise, send the check through secpolicy_vnode_setattr()
1980 		 *
1981 		 */
1982 
1983 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
1984 		    ((idmask == AT_UID) && take_owner) ||
1985 		    ((idmask == AT_GID) && take_group)) {
1986 			if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) {
1987 				/*
1988 				 * Remove setuid/setgid for non-privileged users
1989 				 */
1990 				secpolicy_setid_clear(vap, cr);
1991 				trim_mask = TRUE;
1992 				saved_mask = vap->va_mask;
1993 			} else {
1994 				need_policy =  TRUE;
1995 			}
1996 		} else {
1997 			need_policy =  TRUE;
1998 		}
1999 	}
2000 
2001 	if (mask & AT_MODE)
2002 		need_policy = TRUE;
2003 
2004 	if (need_policy) {
2005 		mutex_enter(&zp->z_lock);
2006 		oldva.va_mode = pzp->zp_mode;
2007 		oldva.va_uid = zp->z_phys->zp_uid;
2008 		oldva.va_gid = zp->z_phys->zp_gid;
2009 		mutex_exit(&zp->z_lock);
2010 
2011 		/*
2012 		 * If trim_mask is set then take ownership
2013 		 * has been granted.  In that case remove
2014 		 * UID|GID from mask so that
2015 		 * secpolicy_vnode_setattr() doesn't revoke it.
2016 		 */
2017 		if (trim_mask)
2018 			vap->va_mask &= ~(AT_UID|AT_GID);
2019 
2020 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2021 		    (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp);
2022 		if (err) {
2023 			ZFS_EXIT(zfsvfs);
2024 			return (err);
2025 		}
2026 
2027 		if (trim_mask)
2028 			vap->va_mask |= (saved_mask & (AT_UID|AT_GID));
2029 	}
2030 
2031 	/*
2032 	 * secpolicy_vnode_setattr, or take ownership may have
2033 	 * changed va_mask
2034 	 */
2035 	mask = vap->va_mask;
2036 
2037 	tx = dmu_tx_create(zfsvfs->z_os);
2038 	dmu_tx_hold_bonus(tx, zp->z_id);
2039 
2040 	if (mask & AT_MODE) {
2041 
2042 		new_mode = (pzp->zp_mode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2043 
2044 		if (zp->z_phys->zp_acl.z_acl_extern_obj)
2045 			dmu_tx_hold_write(tx,
2046 			    pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE);
2047 		else
2048 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2049 			    0, ZFS_ACL_SIZE(MAX_ACL_SIZE));
2050 	}
2051 
2052 	if (mask & AT_SIZE) {
2053 		uint64_t off = vap->va_size;
2054 		/*
2055 		 * Grab the grow_lock to serialize this change with
2056 		 * respect to other file manipulations.
2057 		 */
2058 		rw_enter(&zp->z_grow_lock, RW_WRITER);
2059 		have_grow_lock = TRUE;
2060 		if (off < zp->z_phys->zp_size)
2061 			dmu_tx_hold_free(tx, zp->z_id, off, DMU_OBJECT_END);
2062 		else if (zp->z_phys->zp_size &&
2063 		    zp->z_blksz < zfsvfs->z_max_blksz && off > zp->z_blksz)
2064 			/* we will rewrite this block if we grow */
2065 			dmu_tx_hold_write(tx, zp->z_id, 0, zp->z_phys->zp_size);
2066 	}
2067 
2068 	if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) {
2069 		err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp);
2070 		if (err) {
2071 			dmu_tx_abort(tx);
2072 			if (have_grow_lock)
2073 				rw_exit(&zp->z_grow_lock);
2074 			ZFS_EXIT(zfsvfs);
2075 			return (err);
2076 		}
2077 		dmu_tx_hold_bonus(tx, attrzp->z_id);
2078 	}
2079 
2080 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2081 	if (err) {
2082 		if (attrzp)
2083 			VN_RELE(ZTOV(attrzp));
2084 		dmu_tx_abort(tx);
2085 		if (have_grow_lock)
2086 			rw_exit(&zp->z_grow_lock);
2087 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2088 			txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0);
2089 			goto top;
2090 		}
2091 		ZFS_EXIT(zfsvfs);
2092 		return (err);
2093 	}
2094 
2095 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2096 
2097 	/*
2098 	 * Set each attribute requested.
2099 	 * We group settings according to the locks they need to acquire.
2100 	 *
2101 	 * Note: you cannot set ctime directly, although it will be
2102 	 * updated as a side-effect of calling this function.
2103 	 */
2104 	if (mask & AT_SIZE) {
2105 		/*
2106 		 * XXX - Note, we are not providing any open
2107 		 * mode flags here (like FNDELAY), so we may
2108 		 * block if there are locks present... this
2109 		 * should be addressed in openat().
2110 		 */
2111 		err = zfs_freesp(zp, vap->va_size, 0, 0, tx, cr);
2112 		if (err) {
2113 			mutex_enter(&zp->z_lock);
2114 			goto out;
2115 		}
2116 		mask_applied |= AT_SIZE;
2117 	}
2118 
2119 	mask_applied = mask;	/* no errors after this point */
2120 
2121 	mutex_enter(&zp->z_lock);
2122 
2123 	if (mask & AT_MODE) {
2124 		err = zfs_acl_chmod_setattr(zp, new_mode, tx);
2125 		ASSERT3U(err, ==, 0);
2126 	}
2127 
2128 	if (attrzp)
2129 		mutex_enter(&attrzp->z_lock);
2130 
2131 	if (mask & AT_UID) {
2132 		zp->z_phys->zp_uid = (uint64_t)vap->va_uid;
2133 		if (attrzp) {
2134 			attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid;
2135 		}
2136 	}
2137 
2138 	if (mask & AT_GID) {
2139 		zp->z_phys->zp_gid = (uint64_t)vap->va_gid;
2140 		if (attrzp)
2141 			attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid;
2142 	}
2143 
2144 	if (attrzp)
2145 		mutex_exit(&attrzp->z_lock);
2146 
2147 	if (mask & AT_ATIME)
2148 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2149 
2150 	if (mask & AT_MTIME)
2151 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2152 
2153 	if (mask_applied & AT_SIZE)
2154 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2155 	else if (mask_applied != 0)
2156 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2157 
2158 out:
2159 
2160 	if (mask_applied != 0)
2161 		seq = zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap,
2162 		    mask_applied);
2163 
2164 	mutex_exit(&zp->z_lock);
2165 
2166 	if (attrzp)
2167 		VN_RELE(ZTOV(attrzp));
2168 
2169 	if (have_grow_lock)
2170 		rw_exit(&zp->z_grow_lock);
2171 
2172 	dmu_tx_commit(tx);
2173 
2174 	zil_commit(zilog, seq, 0);
2175 
2176 	ZFS_EXIT(zfsvfs);
2177 	return (err);
2178 }
2179 
2180 /*
2181  * Search back through the directory tree, using the ".." entries.
2182  * Lock each directory in the chain to prevent concurrent renames.
2183  * Fail any attempt to move a directory into one of its own descendants.
2184  * XXX - z_parent_lock can overlap with map or grow locks
2185  */
2186 typedef struct zfs_zlock {
2187 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2188 	znode_t		*zl_znode;	/* znode we held */
2189 	struct zfs_zlock *zl_next;	/* next in list */
2190 } zfs_zlock_t;
2191 
2192 static int
2193 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2194 {
2195 	zfs_zlock_t	*zl;
2196 	znode_t 	*zp = tdzp;
2197 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2198 	uint64_t	*oidp = &zp->z_id;
2199 	krwlock_t	*rwlp = &szp->z_parent_lock;
2200 	krw_t		rw = RW_WRITER;
2201 
2202 	/*
2203 	 * First pass write-locks szp and compares to zp->z_id.
2204 	 * Later passes read-lock zp and compare to zp->z_parent.
2205 	 */
2206 	do {
2207 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2208 		zl->zl_rwlock = rwlp;
2209 		zl->zl_znode = NULL;
2210 		zl->zl_next = *zlpp;
2211 		*zlpp = zl;
2212 
2213 		rw_enter(rwlp, rw);
2214 
2215 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2216 			return (EINVAL);
2217 
2218 		if (*oidp == rootid)		/* We've hit the top */
2219 			return (0);
2220 
2221 		if (rw == RW_READER) {		/* i.e. not the first pass */
2222 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2223 			if (error)
2224 				return (error);
2225 			zl->zl_znode = zp;
2226 		}
2227 		oidp = &zp->z_phys->zp_parent;
2228 		rwlp = &zp->z_parent_lock;
2229 		rw = RW_READER;
2230 
2231 	} while (zp->z_id != sdzp->z_id);
2232 
2233 	return (0);
2234 }
2235 
2236 /*
2237  * Drop locks and release vnodes that were held by zfs_rename_lock().
2238  */
2239 static void
2240 zfs_rename_unlock(zfs_zlock_t **zlpp)
2241 {
2242 	zfs_zlock_t *zl;
2243 
2244 	while ((zl = *zlpp) != NULL) {
2245 		if (zl->zl_znode != NULL)
2246 			VN_RELE(ZTOV(zl->zl_znode));
2247 		rw_exit(zl->zl_rwlock);
2248 		*zlpp = zl->zl_next;
2249 		kmem_free(zl, sizeof (*zl));
2250 	}
2251 }
2252 
2253 /*
2254  * Move an entry from the provided source directory to the target
2255  * directory.  Change the entry name as indicated.
2256  *
2257  *	IN:	sdvp	- Source directory containing the "old entry".
2258  *		snm	- Old entry name.
2259  *		tdvp	- Target directory to contain the "new entry".
2260  *		tnm	- New entry name.
2261  *		cr	- credentials of caller.
2262  *
2263  *	RETURN:	0 if success
2264  *		error code if failure
2265  *
2266  * Timestamps:
2267  *	sdvp,tdvp - ctime|mtime updated
2268  */
2269 static int
2270 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr)
2271 {
2272 	znode_t		*tdzp, *szp, *tzp;
2273 	znode_t		*sdzp = VTOZ(sdvp);
2274 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2275 	zilog_t		*zilog = zfsvfs->z_log;
2276 	uint64_t	seq = 0;
2277 	vnode_t		*realvp;
2278 	zfs_dirlock_t	*sdl, *tdl;
2279 	dmu_tx_t	*tx;
2280 	zfs_zlock_t	*zl;
2281 	int		cmp, serr, terr, error;
2282 
2283 	ZFS_ENTER(zfsvfs);
2284 
2285 	/*
2286 	 * Make sure we have the real vp for the target directory.
2287 	 */
2288 	if (VOP_REALVP(tdvp, &realvp) == 0)
2289 		tdvp = realvp;
2290 
2291 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
2292 		ZFS_EXIT(zfsvfs);
2293 		return (EXDEV);
2294 	}
2295 
2296 	tdzp = VTOZ(tdvp);
2297 top:
2298 	szp = NULL;
2299 	tzp = NULL;
2300 	zl = NULL;
2301 
2302 	/*
2303 	 * This is to prevent the creation of links into attribute space
2304 	 * by renaming a linked file into/outof an attribute directory.
2305 	 * See the comment in zfs_link() for why this is considered bad.
2306 	 */
2307 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
2308 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
2309 		ZFS_EXIT(zfsvfs);
2310 		return (EINVAL);
2311 	}
2312 
2313 	/*
2314 	 * Lock source and target directory entries.  To prevent deadlock,
2315 	 * a lock ordering must be defined.  We lock the directory with
2316 	 * the smallest object id first, or if it's a tie, the one with
2317 	 * the lexically first name.
2318 	 */
2319 	if (sdzp->z_id < tdzp->z_id) {
2320 		cmp = -1;
2321 	} else if (sdzp->z_id > tdzp->z_id) {
2322 		cmp = 1;
2323 	} else {
2324 		cmp = strcmp(snm, tnm);
2325 		if (cmp == 0) {
2326 			/*
2327 			 * POSIX: "If the old argument and the new argument
2328 			 * both refer to links to the same existing file,
2329 			 * the rename() function shall return successfully
2330 			 * and perform no other action."
2331 			 */
2332 			ZFS_EXIT(zfsvfs);
2333 			return (0);
2334 		}
2335 	}
2336 	if (cmp < 0) {
2337 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS);
2338 		terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0);
2339 	} else {
2340 		terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0);
2341 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS);
2342 	}
2343 
2344 	if (serr) {
2345 		/*
2346 		 * Source entry invalid or not there.
2347 		 */
2348 		if (!terr) {
2349 			zfs_dirent_unlock(tdl);
2350 			if (tzp)
2351 				VN_RELE(ZTOV(tzp));
2352 		}
2353 		if (strcmp(snm, "..") == 0)
2354 			serr = EINVAL;
2355 		ZFS_EXIT(zfsvfs);
2356 		return (serr);
2357 	}
2358 	if (terr) {
2359 		zfs_dirent_unlock(sdl);
2360 		VN_RELE(ZTOV(szp));
2361 		if (strcmp(tnm, "..") == 0)
2362 			terr = EINVAL;
2363 		ZFS_EXIT(zfsvfs);
2364 		return (terr);
2365 	}
2366 
2367 	/*
2368 	 * Must have write access at the source to remove the old entry
2369 	 * and write access at the target to create the new entry.
2370 	 * Note that if target and source are the same, this can be
2371 	 * done in a single check.
2372 	 */
2373 
2374 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
2375 		goto out;
2376 
2377 	if (ZTOV(szp)->v_type == VDIR) {
2378 		/*
2379 		 * Check to make sure rename is valid.
2380 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
2381 		 */
2382 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
2383 			goto out;
2384 	}
2385 
2386 	/*
2387 	 * Does target exist?
2388 	 */
2389 	if (tzp) {
2390 		/*
2391 		 * Source and target must be the same type.
2392 		 */
2393 		if (ZTOV(szp)->v_type == VDIR) {
2394 			if (ZTOV(tzp)->v_type != VDIR) {
2395 				error = ENOTDIR;
2396 				goto out;
2397 			}
2398 		} else {
2399 			if (ZTOV(tzp)->v_type == VDIR) {
2400 				error = EISDIR;
2401 				goto out;
2402 			}
2403 		}
2404 		/*
2405 		 * POSIX dictates that when the source and target
2406 		 * entries refer to the same file object, rename
2407 		 * must do nothing and exit without error.
2408 		 */
2409 		if (szp->z_id == tzp->z_id) {
2410 			error = 0;
2411 			goto out;
2412 		}
2413 	}
2414 
2415 	vnevent_rename_src(ZTOV(szp));
2416 	if (tzp)
2417 		vnevent_rename_dest(ZTOV(tzp));
2418 
2419 	tx = dmu_tx_create(zfsvfs->z_os);
2420 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
2421 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
2422 	if (sdzp != tdzp) {
2423 		dmu_tx_hold_zap(tx, sdzp->z_id, 1);
2424 		dmu_tx_hold_zap(tx, tdzp->z_id, 1);
2425 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
2426 	} else {
2427 		dmu_tx_hold_zap(tx, sdzp->z_id, 2);
2428 	}
2429 	if (tzp) {
2430 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* nlink changes */
2431 	}
2432 	dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, 1);
2433 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2434 	if (error) {
2435 		dmu_tx_abort(tx);
2436 		if (zl != NULL)
2437 			zfs_rename_unlock(&zl);
2438 		zfs_dirent_unlock(sdl);
2439 		zfs_dirent_unlock(tdl);
2440 		VN_RELE(ZTOV(szp));
2441 		if (tzp)
2442 			VN_RELE(ZTOV(tzp));
2443 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2444 			txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0);
2445 			goto top;
2446 		}
2447 		ZFS_EXIT(zfsvfs);
2448 		return (error);
2449 	}
2450 
2451 	if (tzp)	/* Attempt to remove the existing target */
2452 		error = zfs_link_destroy(tdl, tzp, tx, 0, NULL);
2453 
2454 	if (error == 0) {
2455 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
2456 		if (error == 0) {
2457 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
2458 			ASSERT(error == 0);
2459 			seq = zfs_log_rename(zilog, tx, TX_RENAME,
2460 			    sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp);
2461 		}
2462 	}
2463 
2464 	dmu_tx_commit(tx);
2465 out:
2466 	if (zl != NULL)
2467 		zfs_rename_unlock(&zl);
2468 
2469 	zfs_dirent_unlock(sdl);
2470 	zfs_dirent_unlock(tdl);
2471 
2472 	VN_RELE(ZTOV(szp));
2473 	if (tzp)
2474 		VN_RELE(ZTOV(tzp));
2475 
2476 	zil_commit(zilog, seq, 0);
2477 
2478 	ZFS_EXIT(zfsvfs);
2479 	return (error);
2480 }
2481 
2482 /*
2483  * Insert the indicated symbolic reference entry into the directory.
2484  *
2485  *	IN:	dvp	- Directory to contain new symbolic link.
2486  *		link	- Name for new symlink entry.
2487  *		vap	- Attributes of new entry.
2488  *		target	- Target path of new symlink.
2489  *		cr	- credentials of caller.
2490  *
2491  *	RETURN:	0 if success
2492  *		error code if failure
2493  *
2494  * Timestamps:
2495  *	dvp - ctime|mtime updated
2496  */
2497 static int
2498 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr)
2499 {
2500 	znode_t		*zp, *dzp = VTOZ(dvp);
2501 	zfs_dirlock_t	*dl;
2502 	dmu_tx_t	*tx;
2503 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2504 	zilog_t		*zilog = zfsvfs->z_log;
2505 	uint64_t	seq = 0;
2506 	uint64_t	zoid;
2507 	int		len = strlen(link);
2508 	int		error;
2509 
2510 	ASSERT(vap->va_type == VLNK);
2511 
2512 	ZFS_ENTER(zfsvfs);
2513 top:
2514 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
2515 		ZFS_EXIT(zfsvfs);
2516 		return (error);
2517 	}
2518 
2519 	if (len > MAXPATHLEN) {
2520 		ZFS_EXIT(zfsvfs);
2521 		return (ENAMETOOLONG);
2522 	}
2523 
2524 	/*
2525 	 * Attempt to lock directory; fail if entry already exists.
2526 	 */
2527 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) {
2528 		ZFS_EXIT(zfsvfs);
2529 		return (error);
2530 	}
2531 
2532 	tx = dmu_tx_create(zfsvfs->z_os);
2533 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
2534 	dmu_tx_hold_bonus(tx, dzp->z_id);
2535 	dmu_tx_hold_zap(tx, dzp->z_id, 1);
2536 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
2537 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
2538 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2539 	if (error) {
2540 		dmu_tx_abort(tx);
2541 		zfs_dirent_unlock(dl);
2542 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2543 			txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0);
2544 			goto top;
2545 		}
2546 		ZFS_EXIT(zfsvfs);
2547 		return (error);
2548 	}
2549 
2550 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
2551 
2552 	/*
2553 	 * Create a new object for the symlink.
2554 	 * Put the link content into bonus buffer if it will fit;
2555 	 * otherwise, store it just like any other file data.
2556 	 */
2557 	zoid = 0;
2558 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
2559 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len);
2560 		if (len != 0)
2561 			bcopy(link, zp->z_phys + 1, len);
2562 	} else {
2563 		dmu_buf_t *dbp;
2564 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
2565 
2566 		rw_enter(&zp->z_grow_lock, RW_WRITER);
2567 		error = zfs_grow_blocksize(zp, len, tx);
2568 		rw_exit(&zp->z_grow_lock);
2569 		if (error)
2570 			goto out;
2571 
2572 		dbp = dmu_buf_hold(zfsvfs->z_os, zoid, 0);
2573 		dmu_buf_will_dirty(dbp, tx);
2574 
2575 		ASSERT3U(len, <=, dbp->db_size);
2576 		bcopy(link, dbp->db_data, len);
2577 		dmu_buf_rele(dbp);
2578 	}
2579 	zp->z_phys->zp_size = len;
2580 
2581 	/*
2582 	 * Insert the new object into the directory.
2583 	 */
2584 	(void) zfs_link_create(dl, zp, tx, ZNEW);
2585 out:
2586 	if (error == 0)
2587 		seq = zfs_log_symlink(zilog, tx, TX_SYMLINK,
2588 		    dzp, zp, name, link);
2589 
2590 	dmu_tx_commit(tx);
2591 
2592 	zfs_dirent_unlock(dl);
2593 
2594 	VN_RELE(ZTOV(zp));
2595 
2596 	zil_commit(zilog, seq, 0);
2597 
2598 	ZFS_EXIT(zfsvfs);
2599 	return (error);
2600 }
2601 
2602 /*
2603  * Return, in the buffer contained in the provided uio structure,
2604  * the symbolic path referred to by vp.
2605  *
2606  *	IN:	vp	- vnode of symbolic link.
2607  *		uoip	- structure to contain the link path.
2608  *		cr	- credentials of caller.
2609  *
2610  *	OUT:	uio	- structure to contain the link path.
2611  *
2612  *	RETURN:	0 if success
2613  *		error code if failure
2614  *
2615  * Timestamps:
2616  *	vp - atime updated
2617  */
2618 /* ARGSUSED */
2619 static int
2620 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr)
2621 {
2622 	znode_t		*zp = VTOZ(vp);
2623 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2624 	size_t		bufsz;
2625 	int		error;
2626 
2627 	ZFS_ENTER(zfsvfs);
2628 
2629 	bufsz = (size_t)zp->z_phys->zp_size;
2630 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
2631 		error = uiomove(zp->z_phys + 1,
2632 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
2633 	} else {
2634 		dmu_buf_t *dbp = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0);
2635 		if ((error = dmu_buf_read_canfail(dbp)) != 0) {
2636 			dmu_buf_rele(dbp);
2637 			ZFS_EXIT(zfsvfs);
2638 			return (error);
2639 		}
2640 		error = uiomove(dbp->db_data,
2641 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
2642 		dmu_buf_rele(dbp);
2643 	}
2644 
2645 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2646 	ZFS_EXIT(zfsvfs);
2647 	return (error);
2648 }
2649 
2650 /*
2651  * Insert a new entry into directory tdvp referencing svp.
2652  *
2653  *	IN:	tdvp	- Directory to contain new entry.
2654  *		svp	- vnode of new entry.
2655  *		name	- name of new entry.
2656  *		cr	- credentials of caller.
2657  *
2658  *	RETURN:	0 if success
2659  *		error code if failure
2660  *
2661  * Timestamps:
2662  *	tdvp - ctime|mtime updated
2663  *	 svp - ctime updated
2664  */
2665 /* ARGSUSED */
2666 static int
2667 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr)
2668 {
2669 	znode_t		*dzp = VTOZ(tdvp);
2670 	znode_t		*tzp, *szp;
2671 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2672 	zilog_t		*zilog = zfsvfs->z_log;
2673 	uint64_t	seq = 0;
2674 	zfs_dirlock_t	*dl;
2675 	dmu_tx_t	*tx;
2676 	vnode_t		*realvp;
2677 	int		error;
2678 
2679 	ASSERT(tdvp->v_type == VDIR);
2680 
2681 	ZFS_ENTER(zfsvfs);
2682 
2683 	if (VOP_REALVP(svp, &realvp) == 0)
2684 		svp = realvp;
2685 
2686 	if (svp->v_vfsp != tdvp->v_vfsp) {
2687 		ZFS_EXIT(zfsvfs);
2688 		return (EXDEV);
2689 	}
2690 
2691 	szp = VTOZ(svp);
2692 top:
2693 	/*
2694 	 * We do not support links between attributes and non-attributes
2695 	 * because of the potential security risk of creating links
2696 	 * into "normal" file space in order to circumvent restrictions
2697 	 * imposed in attribute space.
2698 	 */
2699 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
2700 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
2701 		ZFS_EXIT(zfsvfs);
2702 		return (EINVAL);
2703 	}
2704 
2705 	/*
2706 	 * POSIX dictates that we return EPERM here.
2707 	 * Better choices include ENOTSUP or EISDIR.
2708 	 */
2709 	if (svp->v_type == VDIR) {
2710 		ZFS_EXIT(zfsvfs);
2711 		return (EPERM);
2712 	}
2713 
2714 	if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) &&
2715 	    secpolicy_basic_link(cr) != 0) {
2716 		ZFS_EXIT(zfsvfs);
2717 		return (EPERM);
2718 	}
2719 
2720 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
2721 		ZFS_EXIT(zfsvfs);
2722 		return (error);
2723 	}
2724 
2725 	/*
2726 	 * Attempt to lock directory; fail if entry already exists.
2727 	 */
2728 	if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) {
2729 		ZFS_EXIT(zfsvfs);
2730 		return (error);
2731 	}
2732 
2733 	tx = dmu_tx_create(zfsvfs->z_os);
2734 	dmu_tx_hold_bonus(tx, szp->z_id);
2735 	dmu_tx_hold_zap(tx, dzp->z_id, 1);
2736 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2737 	if (error) {
2738 		dmu_tx_abort(tx);
2739 		zfs_dirent_unlock(dl);
2740 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2741 			txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0);
2742 			goto top;
2743 		}
2744 		ZFS_EXIT(zfsvfs);
2745 		return (error);
2746 	}
2747 
2748 	error = zfs_link_create(dl, szp, tx, 0);
2749 
2750 	if (error == 0)
2751 		seq = zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name);
2752 
2753 	dmu_tx_commit(tx);
2754 
2755 	zfs_dirent_unlock(dl);
2756 
2757 	zil_commit(zilog, seq, 0);
2758 
2759 	ZFS_EXIT(zfsvfs);
2760 	return (error);
2761 }
2762 
2763 /*
2764  * zfs_null_putapage() is used when the file system has been force
2765  * unmounted. It just drops the pages.
2766  */
2767 /* ARGSUSED */
2768 static int
2769 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
2770 		size_t *lenp, int flags, cred_t *cr)
2771 {
2772 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
2773 	return (0);
2774 }
2775 
2776 /* ARGSUSED */
2777 static int
2778 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
2779 		size_t *lenp, int flags, cred_t *cr)
2780 {
2781 	znode_t		*zp = VTOZ(vp);
2782 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2783 	zilog_t		*zilog = zfsvfs->z_log;
2784 	dmu_tx_t	*tx;
2785 	u_offset_t	off;
2786 	ssize_t		len;
2787 	caddr_t		va;
2788 	int		err;
2789 
2790 top:
2791 	rw_enter(&zp->z_grow_lock, RW_READER);
2792 
2793 	off = pp->p_offset;
2794 	len = MIN(PAGESIZE, zp->z_phys->zp_size - off);
2795 
2796 	tx = dmu_tx_create(zfsvfs->z_os);
2797 	dmu_tx_hold_write(tx, zp->z_id, off, len);
2798 	dmu_tx_hold_bonus(tx, zp->z_id);
2799 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2800 	if (err != 0) {
2801 		dmu_tx_abort(tx);
2802 		rw_exit(&zp->z_grow_lock);
2803 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2804 			txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0);
2805 			goto top;
2806 		}
2807 		goto out;
2808 	}
2809 
2810 	va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1);
2811 
2812 	dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
2813 
2814 	ppmapout(va);
2815 
2816 	zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
2817 	(void) zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0, NULL);
2818 	dmu_tx_commit(tx);
2819 
2820 	rw_exit(&zp->z_grow_lock);
2821 
2822 	pvn_write_done(pp, B_WRITE | flags);
2823 	if (offp)
2824 		*offp = off;
2825 	if (lenp)
2826 		*lenp = len;
2827 
2828 out:
2829 	return (err);
2830 }
2831 
2832 /*
2833  * Copy the portion of the file indicated from pages into the file.
2834  * The pages are stored in a page list attached to the files vnode.
2835  *
2836  *	IN:	vp	- vnode of file to push page data to.
2837  *		off	- position in file to put data.
2838  *		len	- amount of data to write.
2839  *		flags	- flags to control the operation.
2840  *		cr	- credentials of caller.
2841  *
2842  *	RETURN:	0 if success
2843  *		error code if failure
2844  *
2845  * Timestamps:
2846  *	vp - ctime|mtime updated
2847  */
2848 static int
2849 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr)
2850 {
2851 	znode_t		*zp = VTOZ(vp);
2852 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2853 	page_t		*pp;
2854 	size_t		io_len;
2855 	u_offset_t	io_off;
2856 	int		error = 0;
2857 
2858 	ZFS_ENTER(zfsvfs);
2859 
2860 	ASSERT(zp->z_dbuf_held && zp->z_phys);
2861 
2862 	if (len == 0) {
2863 		/*
2864 		 * Search the entire vp list for pages >= off.
2865 		 */
2866 		error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage,
2867 		    flags, cr);
2868 		goto out;
2869 	}
2870 
2871 	if (off > zp->z_phys->zp_size) {
2872 		/* past end of file */
2873 		ZFS_EXIT(zfsvfs);
2874 		return (0);
2875 	}
2876 
2877 	len = MIN(len, zp->z_phys->zp_size - off);
2878 
2879 	for (io_off = off; io_off < off + len; io_off += io_len) {
2880 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
2881 			pp  = page_lookup(vp, io_off,
2882 				(flags & (B_INVAL | B_FREE)) ?
2883 					SE_EXCL : SE_SHARED);
2884 		} else {
2885 			pp = page_lookup_nowait(vp, io_off,
2886 				(flags & B_FREE) ? SE_EXCL : SE_SHARED);
2887 		}
2888 
2889 		if (pp != NULL && pvn_getdirty(pp, flags)) {
2890 			int err;
2891 
2892 			/*
2893 			 * Found a dirty page to push
2894 			 */
2895 			if (err =
2896 			    zfs_putapage(vp, pp, &io_off, &io_len, flags, cr))
2897 				error = err;
2898 		} else {
2899 			io_len = PAGESIZE;
2900 		}
2901 	}
2902 out:
2903 	zil_commit(zfsvfs->z_log, UINT64_MAX, (flags & B_ASYNC) ? 0 : FDSYNC);
2904 	ZFS_EXIT(zfsvfs);
2905 	return (error);
2906 }
2907 
2908 void
2909 zfs_inactive(vnode_t *vp, cred_t *cr)
2910 {
2911 	znode_t	*zp = VTOZ(vp);
2912 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2913 	int error;
2914 
2915 	rw_enter(&zfsvfs->z_um_lock, RW_READER);
2916 	if (zfsvfs->z_unmounted2) {
2917 		ASSERT(zp->z_dbuf_held == 0);
2918 
2919 		if (vn_has_cached_data(vp)) {
2920 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
2921 			    B_INVAL, cr);
2922 		}
2923 
2924 		vp->v_count = 0; /* count arrives as 1 */
2925 		zfs_znode_free(zp);
2926 		rw_exit(&zfsvfs->z_um_lock);
2927 		VFS_RELE(zfsvfs->z_vfs);
2928 		return;
2929 	}
2930 
2931 	/*
2932 	 * Attempt to push any data in the page cache.  If this fails
2933 	 * we will get kicked out later in zfs_zinactive().
2934 	 */
2935 	if (vn_has_cached_data(vp)) {
2936 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
2937 		    cr);
2938 	}
2939 
2940 	if (zp->z_atime_dirty && zp->z_reap == 0) {
2941 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
2942 
2943 		dmu_tx_hold_bonus(tx, zp->z_id);
2944 		error = dmu_tx_assign(tx, TXG_WAIT);
2945 		if (error) {
2946 			dmu_tx_abort(tx);
2947 		} else {
2948 			dmu_buf_will_dirty(zp->z_dbuf, tx);
2949 			mutex_enter(&zp->z_lock);
2950 			zp->z_atime_dirty = 0;
2951 			mutex_exit(&zp->z_lock);
2952 			dmu_tx_commit(tx);
2953 		}
2954 	}
2955 
2956 	zfs_zinactive(zp);
2957 	rw_exit(&zfsvfs->z_um_lock);
2958 }
2959 
2960 /*
2961  * Bounds-check the seek operation.
2962  *
2963  *	IN:	vp	- vnode seeking within
2964  *		ooff	- old file offset
2965  *		noffp	- pointer to new file offset
2966  *
2967  *	RETURN:	0 if success
2968  *		EINVAL if new offset invalid
2969  */
2970 /* ARGSUSED */
2971 static int
2972 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp)
2973 {
2974 	if (vp->v_type == VDIR)
2975 		return (0);
2976 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
2977 }
2978 
2979 /*
2980  * Pre-filter the generic locking function to trap attempts to place
2981  * a mandatory lock on a memory mapped file.
2982  */
2983 static int
2984 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
2985     flk_callback_t *flk_cbp, cred_t *cr)
2986 {
2987 	znode_t *zp = VTOZ(vp);
2988 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2989 	uint_t cnt = 1;
2990 	int error;
2991 
2992 	ZFS_ENTER(zfsvfs);
2993 
2994 	/*
2995 	 * If file is being mapped, disallow frlock.  We set the mapcnt to
2996 	 * -1 here to signal that we are in the process of setting a lock.
2997 	 * This prevents a race with zfs_map().
2998 	 * XXX - well, sort of; since zfs_map() does not change z_mapcnt,
2999 	 * we could be in the middle of zfs_map() and still call fs_frlock().
3000 	 * Also, we are doing no checking in zfs_addmap() (where z_mapcnt
3001 	 * *is* manipulated).
3002 	 */
3003 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) &&
3004 	    (int)(cnt = atomic_cas_32(&zp->z_mapcnt, 0, -1)) > 0) {
3005 		ZFS_EXIT(zfsvfs);
3006 		return (EAGAIN);
3007 	}
3008 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr);
3009 	ASSERT((cnt != 0) || ((int)atomic_cas_32(&zp->z_mapcnt, -1, 0) == -1));
3010 	ZFS_EXIT(zfsvfs);
3011 	return (error);
3012 }
3013 
3014 /*
3015  * If we can't find a page in the cache, we will create a new page
3016  * and fill it with file data.  For efficiency, we may try to fill
3017  * multiple pages as once (klustering).
3018  */
3019 static int
3020 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3021     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3022 {
3023 	znode_t *zp = VTOZ(vp);
3024 	page_t *pp, *cur_pp;
3025 	objset_t *os = zp->z_zfsvfs->z_os;
3026 	caddr_t va;
3027 	u_offset_t io_off, total;
3028 	uint64_t oid = zp->z_id;
3029 	size_t io_len;
3030 	int err;
3031 
3032 	/*
3033 	 * If we are only asking for a single page don't bother klustering.
3034 	 */
3035 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE ||
3036 	    off > zp->z_phys->zp_size) {
3037 		io_off = off;
3038 		io_len = PAGESIZE;
3039 		pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr);
3040 	} else {
3041 		/*
3042 		 * Try to fill a kluster of pages (a blocks worth).
3043 		 */
3044 		size_t klen;
3045 		u_offset_t koff;
3046 
3047 		if (!ISP2(zp->z_blksz)) {
3048 			/* Only one block in the file. */
3049 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3050 			koff = 0;
3051 		} else {
3052 			klen = plsz;
3053 			koff = P2ALIGN(off, (u_offset_t)klen);
3054 		}
3055 		if (klen > zp->z_phys->zp_size)
3056 			klen = P2ROUNDUP(zp->z_phys->zp_size,
3057 			    (uint64_t)PAGESIZE);
3058 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3059 			    &io_len, koff, klen, 0);
3060 	}
3061 	if (pp == NULL) {
3062 		/*
3063 		 * Some other thread entered the page before us.
3064 		 * Return to zfs_getpage to retry the lookup.
3065 		 */
3066 		*pl = NULL;
3067 		return (0);
3068 	}
3069 
3070 	/*
3071 	 * Fill the pages in the kluster.
3072 	 */
3073 	cur_pp = pp;
3074 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3075 		ASSERT(io_off == cur_pp->p_offset);
3076 		va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1);
3077 		err = dmu_read_canfail(os, oid, io_off, PAGESIZE, va);
3078 		ppmapout(va);
3079 		if (err) {
3080 			/* On error, toss the entire kluster */
3081 			pvn_read_done(pp, B_ERROR);
3082 			return (err);
3083 		}
3084 		cur_pp = cur_pp->p_next;
3085 	}
3086 out:
3087 	/*
3088 	 * Fill in the page list array from the kluster.  If
3089 	 * there are too many pages in the kluster, return
3090 	 * as many pages as possible starting from the desired
3091 	 * offset `off'.
3092 	 * NOTE: the page list will always be null terminated.
3093 	 */
3094 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3095 
3096 	return (0);
3097 }
3098 
3099 /*
3100  * Return pointers to the pages for the file region [off, off + len]
3101  * in the pl array.  If plsz is greater than len, this function may
3102  * also return page pointers from before or after the specified
3103  * region (i.e. some region [off', off' + plsz]).  These additional
3104  * pages are only returned if they are already in the cache, or were
3105  * created as part of a klustered read.
3106  *
3107  *	IN:	vp	- vnode of file to get data from.
3108  *		off	- position in file to get data from.
3109  *		len	- amount of data to retrieve.
3110  *		plsz	- length of provided page list.
3111  *		seg	- segment to obtain pages for.
3112  *		addr	- virtual address of fault.
3113  *		rw	- mode of created pages.
3114  *		cr	- credentials of caller.
3115  *
3116  *	OUT:	protp	- protection mode of created pages.
3117  *		pl	- list of pages created.
3118  *
3119  *	RETURN:	0 if success
3120  *		error code if failure
3121  *
3122  * Timestamps:
3123  *	vp - atime updated
3124  */
3125 /* ARGSUSED */
3126 static int
3127 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3128 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3129 	enum seg_rw rw, cred_t *cr)
3130 {
3131 	znode_t		*zp = VTOZ(vp);
3132 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3133 	page_t		*pp, **pl0 = pl;
3134 	int		cnt = 0, need_unlock = 0, err = 0;
3135 
3136 	ZFS_ENTER(zfsvfs);
3137 
3138 	if (protp)
3139 		*protp = PROT_ALL;
3140 
3141 	ASSERT(zp->z_dbuf_held && zp->z_phys);
3142 
3143 	/* no faultahead (for now) */
3144 	if (pl == NULL) {
3145 		ZFS_EXIT(zfsvfs);
3146 		return (0);
3147 	}
3148 
3149 	/* can't fault past EOF */
3150 	if (off >= zp->z_phys->zp_size) {
3151 		ZFS_EXIT(zfsvfs);
3152 		return (EFAULT);
3153 	}
3154 
3155 	/*
3156 	 * Make sure nobody restructures the file (changes block size)
3157 	 * in the middle of the getpage.
3158 	 */
3159 	rw_enter(&zp->z_grow_lock, RW_READER);
3160 
3161 	/*
3162 	 * If we already own the lock, then we must be page faulting
3163 	 * in the middle of a write to this file (i.e., we are writing
3164 	 * to this file using data from a mapped region of the file).
3165 	 */
3166 	if (!rw_owner(&zp->z_map_lock)) {
3167 		rw_enter(&zp->z_map_lock, RW_WRITER);
3168 		need_unlock = TRUE;
3169 	}
3170 
3171 	/*
3172 	 * Loop through the requested range [off, off + len] looking
3173 	 * for pages.  If we don't find a page, we will need to create
3174 	 * a new page and fill it with data from the file.
3175 	 */
3176 	while (len > 0) {
3177 		if (plsz < PAGESIZE)
3178 			break;
3179 		if (pp = page_lookup(vp, off, SE_SHARED)) {
3180 			*pl++ = pp;
3181 			off += PAGESIZE;
3182 			addr += PAGESIZE;
3183 			len -= PAGESIZE;
3184 			plsz -= PAGESIZE;
3185 		} else {
3186 			err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw);
3187 			/*
3188 			 * klustering may have changed our region
3189 			 * to be block aligned.
3190 			 */
3191 			if (((pp = *pl) != 0) && (off != pp->p_offset)) {
3192 				int delta = off - pp->p_offset;
3193 				len += delta;
3194 				off -= delta;
3195 				addr -= delta;
3196 			}
3197 			while (*pl) {
3198 				pl++;
3199 				cnt++;
3200 				off += PAGESIZE;
3201 				addr += PAGESIZE;
3202 				plsz -= PAGESIZE;
3203 				if (len > PAGESIZE)
3204 					len -= PAGESIZE;
3205 				else
3206 					len = 0;
3207 			}
3208 		}
3209 		if (err)
3210 			goto out;
3211 	}
3212 
3213 	/*
3214 	 * Fill out the page array with any pages already in the cache.
3215 	 */
3216 	while (plsz > 0) {
3217 		pp = page_lookup_nowait(vp, off, SE_SHARED);
3218 		if (pp == NULL)
3219 			break;
3220 		*pl++ = pp;
3221 		off += PAGESIZE;
3222 		plsz -= PAGESIZE;
3223 	}
3224 
3225 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3226 out:
3227 	if (err) {
3228 		/*
3229 		 * Release any pages we have locked.
3230 		 */
3231 		while (pl > pl0)
3232 			page_unlock(*--pl);
3233 	}
3234 	*pl = NULL;
3235 
3236 	if (need_unlock)
3237 		rw_exit(&zp->z_map_lock);
3238 	rw_exit(&zp->z_grow_lock);
3239 
3240 	ZFS_EXIT(zfsvfs);
3241 	return (err);
3242 }
3243 
3244 static int
3245 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
3246     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3247 {
3248 	znode_t *zp = VTOZ(vp);
3249 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3250 	segvn_crargs_t	vn_a;
3251 	int		error;
3252 
3253 	ZFS_ENTER(zfsvfs);
3254 
3255 	if (vp->v_flag & VNOMAP) {
3256 		ZFS_EXIT(zfsvfs);
3257 		return (ENOSYS);
3258 	}
3259 
3260 	if (off < 0 || len > MAXOFFSET_T - off) {
3261 		ZFS_EXIT(zfsvfs);
3262 		return (ENXIO);
3263 	}
3264 
3265 	if (vp->v_type != VREG) {
3266 		ZFS_EXIT(zfsvfs);
3267 		return (ENODEV);
3268 	}
3269 
3270 	/*
3271 	 * If file is locked, disallow mapping.
3272 	 * XXX - since we don't modify z_mapcnt here, there is nothing
3273 	 * to stop a file lock being placed immediately after we complete
3274 	 * this check.
3275 	 */
3276 	if (MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3277 		if (vn_has_flocks(vp) || zp->z_mapcnt == -1) {
3278 			ZFS_EXIT(zfsvfs);
3279 			return (EAGAIN);
3280 		}
3281 	}
3282 
3283 	as_rangelock(as);
3284 	if ((flags & MAP_FIXED) == 0) {
3285 		map_addr(addrp, len, off, 1, flags);
3286 		if (*addrp == NULL) {
3287 			as_rangeunlock(as);
3288 			ZFS_EXIT(zfsvfs);
3289 			return (ENOMEM);
3290 		}
3291 	} else {
3292 		/*
3293 		 * User specified address - blow away any previous mappings
3294 		 */
3295 		(void) as_unmap(as, *addrp, len);
3296 	}
3297 
3298 	vn_a.vp = vp;
3299 	vn_a.offset = (u_offset_t)off;
3300 	vn_a.type = flags & MAP_TYPE;
3301 	vn_a.prot = prot;
3302 	vn_a.maxprot = maxprot;
3303 	vn_a.cred = cr;
3304 	vn_a.amp = NULL;
3305 	vn_a.flags = flags & ~MAP_TYPE;
3306 	vn_a.szc = 0;
3307 	vn_a.lgrp_mem_policy_flags = 0;
3308 
3309 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
3310 
3311 	as_rangeunlock(as);
3312 	ZFS_EXIT(zfsvfs);
3313 	return (error);
3314 }
3315 
3316 /* ARGSUSED */
3317 static int
3318 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3319     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3320 {
3321 	/*
3322 	 * XXX - shouldn't we be checking for file locks here?
3323 	 */
3324 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, 0);
3325 	atomic_add_32(&VTOZ(vp)->z_mapcnt, btopr(len));
3326 	return (0);
3327 }
3328 
3329 /* ARGSUSED */
3330 static int
3331 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3332     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr)
3333 {
3334 	atomic_add_32(&VTOZ(vp)->z_mapcnt, -btopr(len));
3335 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, 0);
3336 	return (0);
3337 }
3338 
3339 /*
3340  * Free or allocate space in a file.  Currently, this function only
3341  * supports the `F_FREESP' command.  However, this command is somewhat
3342  * misnamed, as its functionality includes the ability to allocate as
3343  * well as free space.
3344  *
3345  *	IN:	vp	- vnode of file to free data in.
3346  *		cmd	- action to take (only F_FREESP supported).
3347  *		bfp	- section of file to free/alloc.
3348  *		flag	- current file open mode flags.
3349  *		offset	- current file offset.
3350  *		cr	- credentials of caller [UNUSED].
3351  *
3352  *	RETURN:	0 if success
3353  *		error code if failure
3354  *
3355  * Timestamps:
3356  *	vp - ctime|mtime updated
3357  *
3358  * NOTE: This function is limited in that it will only permit space to
3359  *   be freed at the end of a file.  In essence, this function simply
3360  *   allows one to set the file size.
3361  */
3362 /* ARGSUSED */
3363 static int
3364 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
3365     offset_t offset, cred_t *cr, caller_context_t *ct)
3366 {
3367 	dmu_tx_t	*tx;
3368 	znode_t		*zp = VTOZ(vp);
3369 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3370 	zilog_t		*zilog = zfsvfs->z_log;
3371 	uint64_t	seq = 0;
3372 	uint64_t	off, len;
3373 	int		error;
3374 
3375 	ZFS_ENTER(zfsvfs);
3376 
3377 top:
3378 	if (cmd != F_FREESP) {
3379 		ZFS_EXIT(zfsvfs);
3380 		return (EINVAL);
3381 	}
3382 
3383 	if (error = convoff(vp, bfp, 0, offset)) {
3384 		ZFS_EXIT(zfsvfs);
3385 		return (error);
3386 	}
3387 
3388 	if (bfp->l_len < 0) {
3389 		ZFS_EXIT(zfsvfs);
3390 		return (EINVAL);
3391 	}
3392 
3393 	off = bfp->l_start;
3394 	len = bfp->l_len;
3395 	tx = dmu_tx_create(zfsvfs->z_os);
3396 	/*
3397 	 * Grab the grow_lock to serialize this change with
3398 	 * respect to other file size changes.
3399 	 */
3400 	dmu_tx_hold_bonus(tx, zp->z_id);
3401 	rw_enter(&zp->z_grow_lock, RW_WRITER);
3402 	if (off + len > zp->z_blksz && zp->z_blksz < zfsvfs->z_max_blksz &&
3403 	    off >= zp->z_phys->zp_size) {
3404 		/*
3405 		 * We are increasing the length of the file,
3406 		 * and this may mean a block size increase.
3407 		 */
3408 		dmu_tx_hold_write(tx, zp->z_id, 0,
3409 		    MIN(off + len, zfsvfs->z_max_blksz));
3410 	} else if (off < zp->z_phys->zp_size) {
3411 		/*
3412 		 * If len == 0, we are truncating the file.
3413 		 */
3414 		dmu_tx_hold_free(tx, zp->z_id, off, len ? len : DMU_OBJECT_END);
3415 	}
3416 
3417 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3418 	if (error) {
3419 		dmu_tx_abort(tx);
3420 		rw_exit(&zp->z_grow_lock);
3421 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3422 			txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0);
3423 			goto top;
3424 		}
3425 		ZFS_EXIT(zfsvfs);
3426 		return (error);
3427 	}
3428 
3429 	error = zfs_freesp(zp, off, len, flag, tx, cr);
3430 
3431 	if (error == 0) {
3432 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
3433 		seq = zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
3434 	}
3435 
3436 	rw_exit(&zp->z_grow_lock);
3437 
3438 	dmu_tx_commit(tx);
3439 
3440 	zil_commit(zilog, seq, 0);
3441 
3442 	ZFS_EXIT(zfsvfs);
3443 	return (error);
3444 }
3445 
3446 static int
3447 zfs_fid(vnode_t *vp, fid_t *fidp)
3448 {
3449 	znode_t		*zp = VTOZ(vp);
3450 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3451 	uint32_t	gen = (uint32_t)zp->z_phys->zp_gen;
3452 	uint64_t	object = zp->z_id;
3453 	zfid_short_t	*zfid;
3454 	int		size, i;
3455 
3456 	ZFS_ENTER(zfsvfs);
3457 
3458 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
3459 	if (fidp->fid_len < size) {
3460 		fidp->fid_len = size;
3461 		ZFS_EXIT(zfsvfs);
3462 		return (ENOSPC);
3463 	}
3464 
3465 	zfid = (zfid_short_t *)fidp;
3466 
3467 	zfid->zf_len = size;
3468 
3469 	for (i = 0; i < sizeof (zfid->zf_object); i++)
3470 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
3471 
3472 	/* Must have a non-zero generation number to distinguish from .zfs */
3473 	if (gen == 0)
3474 		gen = 1;
3475 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
3476 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
3477 
3478 	if (size == LONG_FID_LEN) {
3479 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
3480 		zfid_long_t	*zlfid;
3481 
3482 		zlfid = (zfid_long_t *)fidp;
3483 
3484 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
3485 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
3486 
3487 		/* XXX - this should be the generation number for the objset */
3488 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
3489 			zlfid->zf_setgen[i] = 0;
3490 	}
3491 
3492 	ZFS_EXIT(zfsvfs);
3493 	return (0);
3494 }
3495 
3496 static int
3497 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr)
3498 {
3499 	znode_t		*zp, *xzp;
3500 	zfsvfs_t	*zfsvfs;
3501 	zfs_dirlock_t	*dl;
3502 	int		error;
3503 
3504 	switch (cmd) {
3505 	case _PC_LINK_MAX:
3506 		*valp = ULONG_MAX;
3507 		return (0);
3508 
3509 	case _PC_FILESIZEBITS:
3510 		*valp = 64;
3511 		return (0);
3512 
3513 	case _PC_XATTR_EXISTS:
3514 		zp = VTOZ(vp);
3515 		zfsvfs = zp->z_zfsvfs;
3516 		ZFS_ENTER(zfsvfs);
3517 		*valp = 0;
3518 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
3519 		    ZXATTR | ZEXISTS | ZSHARED);
3520 		if (error == 0) {
3521 			zfs_dirent_unlock(dl);
3522 			if (!zfs_dirempty(xzp))
3523 				*valp = 1;
3524 			VN_RELE(ZTOV(xzp));
3525 		} else if (error == ENOENT) {
3526 			/*
3527 			 * If there aren't extended attributes, it's the
3528 			 * same as having zero of them.
3529 			 */
3530 			error = 0;
3531 		}
3532 		ZFS_EXIT(zfsvfs);
3533 		return (error);
3534 
3535 	case _PC_ACL_ENABLED:
3536 		*valp = _ACL_ACE_ENABLED;
3537 		return (0);
3538 
3539 	case _PC_MIN_HOLE_SIZE:
3540 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
3541 		return (0);
3542 
3543 	default:
3544 		return (fs_pathconf(vp, cmd, valp, cr));
3545 	}
3546 }
3547 
3548 /*ARGSUSED*/
3549 static int
3550 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr)
3551 {
3552 	znode_t *zp = VTOZ(vp);
3553 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3554 	int error;
3555 
3556 	ZFS_ENTER(zfsvfs);
3557 	error = zfs_getacl(zp, vsecp, cr);
3558 	ZFS_EXIT(zfsvfs);
3559 
3560 	return (error);
3561 }
3562 
3563 /*ARGSUSED*/
3564 static int
3565 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr)
3566 {
3567 	znode_t *zp = VTOZ(vp);
3568 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3569 	int error;
3570 
3571 	ZFS_ENTER(zfsvfs);
3572 	error = zfs_setacl(zp, vsecp, cr);
3573 	ZFS_EXIT(zfsvfs);
3574 	return (error);
3575 }
3576 
3577 /*
3578  * Predeclare these here so that the compiler assumes that
3579  * this is an "old style" function declaration that does
3580  * not include arguments => we won't get type mismatch errors
3581  * in the initializations that follow.
3582  */
3583 static int zfs_inval();
3584 static int zfs_isdir();
3585 
3586 static int
3587 zfs_inval()
3588 {
3589 	return (EINVAL);
3590 }
3591 
3592 static int
3593 zfs_isdir()
3594 {
3595 	return (EISDIR);
3596 }
3597 /*
3598  * Directory vnode operations template
3599  */
3600 vnodeops_t *zfs_dvnodeops;
3601 const fs_operation_def_t zfs_dvnodeops_template[] = {
3602 	VOPNAME_OPEN, zfs_open,
3603 	VOPNAME_CLOSE, zfs_close,
3604 	VOPNAME_READ, zfs_isdir,
3605 	VOPNAME_WRITE, zfs_isdir,
3606 	VOPNAME_IOCTL, zfs_ioctl,
3607 	VOPNAME_GETATTR, zfs_getattr,
3608 	VOPNAME_SETATTR, zfs_setattr,
3609 	VOPNAME_ACCESS, zfs_access,
3610 	VOPNAME_LOOKUP, zfs_lookup,
3611 	VOPNAME_CREATE, zfs_create,
3612 	VOPNAME_REMOVE, zfs_remove,
3613 	VOPNAME_LINK, zfs_link,
3614 	VOPNAME_RENAME, zfs_rename,
3615 	VOPNAME_MKDIR, zfs_mkdir,
3616 	VOPNAME_RMDIR, zfs_rmdir,
3617 	VOPNAME_READDIR, zfs_readdir,
3618 	VOPNAME_SYMLINK, zfs_symlink,
3619 	VOPNAME_FSYNC, zfs_fsync,
3620 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3621 	VOPNAME_FID, zfs_fid,
3622 	VOPNAME_SEEK, zfs_seek,
3623 	VOPNAME_PATHCONF, zfs_pathconf,
3624 	VOPNAME_GETSECATTR, zfs_getsecattr,
3625 	VOPNAME_SETSECATTR, zfs_setsecattr,
3626 	NULL, NULL
3627 };
3628 
3629 /*
3630  * Regular file vnode operations template
3631  */
3632 vnodeops_t *zfs_fvnodeops;
3633 const fs_operation_def_t zfs_fvnodeops_template[] = {
3634 	VOPNAME_OPEN, zfs_open,
3635 	VOPNAME_CLOSE, zfs_close,
3636 	VOPNAME_READ, zfs_read,
3637 	VOPNAME_WRITE, zfs_write,
3638 	VOPNAME_IOCTL, zfs_ioctl,
3639 	VOPNAME_GETATTR, zfs_getattr,
3640 	VOPNAME_SETATTR, zfs_setattr,
3641 	VOPNAME_ACCESS, zfs_access,
3642 	VOPNAME_LOOKUP, zfs_lookup,
3643 	VOPNAME_RENAME, zfs_rename,
3644 	VOPNAME_FSYNC, zfs_fsync,
3645 	VOPNAME_INACTIVE, (fs_generic_func_p)zfs_inactive,
3646 	VOPNAME_FID, zfs_fid,
3647 	VOPNAME_SEEK, zfs_seek,
3648 	VOPNAME_FRLOCK, zfs_frlock,
3649 	VOPNAME_SPACE, zfs_space,
3650 	VOPNAME_GETPAGE, zfs_getpage,
3651 	VOPNAME_PUTPAGE, zfs_putpage,
3652 	VOPNAME_MAP, (fs_generic_func_p) zfs_map,
3653 	VOPNAME_ADDMAP, (fs_generic_func_p) zfs_addmap,
3654 	VOPNAME_DELMAP, zfs_delmap,
3655 	VOPNAME_PATHCONF, zfs_pathconf,
3656 	VOPNAME_GETSECATTR, zfs_getsecattr,
3657 	VOPNAME_SETSECATTR, zfs_setsecattr,
3658 	VOPNAME_VNEVENT, fs_vnevent_support,
3659 	NULL, NULL
3660 };
3661 
3662 /*
3663  * Symbolic link vnode operations template
3664  */
3665 vnodeops_t *zfs_symvnodeops;
3666 const fs_operation_def_t zfs_symvnodeops_template[] = {
3667 	VOPNAME_GETATTR, zfs_getattr,
3668 	VOPNAME_SETATTR, zfs_setattr,
3669 	VOPNAME_ACCESS, zfs_access,
3670 	VOPNAME_RENAME, zfs_rename,
3671 	VOPNAME_READLINK, zfs_readlink,
3672 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3673 	VOPNAME_FID, zfs_fid,
3674 	VOPNAME_PATHCONF, zfs_pathconf,
3675 	VOPNAME_VNEVENT, fs_vnevent_support,
3676 	NULL, NULL
3677 };
3678 
3679 /*
3680  * Extended attribute directory vnode operations template
3681  *	This template is identical to the directory vnodes
3682  *	operation template except for restricted operations:
3683  *		VOP_MKDIR()
3684  *		VOP_SYMLINK()
3685  * Note that there are other restrictions embedded in:
3686  *	zfs_create()	- restrict type to VREG
3687  *	zfs_link()	- no links into/out of attribute space
3688  *	zfs_rename()	- no moves into/out of attribute space
3689  */
3690 vnodeops_t *zfs_xdvnodeops;
3691 const fs_operation_def_t zfs_xdvnodeops_template[] = {
3692 	VOPNAME_OPEN, zfs_open,
3693 	VOPNAME_CLOSE, zfs_close,
3694 	VOPNAME_IOCTL, zfs_ioctl,
3695 	VOPNAME_GETATTR, zfs_getattr,
3696 	VOPNAME_SETATTR, zfs_setattr,
3697 	VOPNAME_ACCESS, zfs_access,
3698 	VOPNAME_LOOKUP, zfs_lookup,
3699 	VOPNAME_CREATE, zfs_create,
3700 	VOPNAME_REMOVE, zfs_remove,
3701 	VOPNAME_LINK, zfs_link,
3702 	VOPNAME_RENAME, zfs_rename,
3703 	VOPNAME_MKDIR, zfs_inval,
3704 	VOPNAME_RMDIR, zfs_rmdir,
3705 	VOPNAME_READDIR, zfs_readdir,
3706 	VOPNAME_SYMLINK, zfs_inval,
3707 	VOPNAME_FSYNC, zfs_fsync,
3708 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3709 	VOPNAME_FID, zfs_fid,
3710 	VOPNAME_SEEK, zfs_seek,
3711 	VOPNAME_PATHCONF, zfs_pathconf,
3712 	VOPNAME_GETSECATTR, zfs_getsecattr,
3713 	VOPNAME_SETSECATTR, zfs_setsecattr,
3714 	VOPNAME_VNEVENT, fs_vnevent_support,
3715 	NULL, NULL
3716 };
3717 
3718 /*
3719  * Error vnode operations template
3720  */
3721 vnodeops_t *zfs_evnodeops;
3722 const fs_operation_def_t zfs_evnodeops_template[] = {
3723 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3724 	VOPNAME_PATHCONF, zfs_pathconf,
3725 	NULL, NULL
3726 };
3727