xref: /titanic_50/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 2017c9656f884256b400be40fa25d96d630bf02a)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /* Portions Copyright 2007 Jeremy Teo */
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/vfs_opreg.h>
36 #include <sys/vnode.h>
37 #include <sys/file.h>
38 #include <sys/stat.h>
39 #include <sys/kmem.h>
40 #include <sys/taskq.h>
41 #include <sys/uio.h>
42 #include <sys/vmsystm.h>
43 #include <sys/atomic.h>
44 #include <sys/vm.h>
45 #include <vm/seg_vn.h>
46 #include <vm/pvn.h>
47 #include <vm/as.h>
48 #include <vm/kpm.h>
49 #include <vm/seg_kpm.h>
50 #include <sys/mman.h>
51 #include <sys/pathname.h>
52 #include <sys/cmn_err.h>
53 #include <sys/errno.h>
54 #include <sys/unistd.h>
55 #include <sys/zfs_dir.h>
56 #include <sys/zfs_acl.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/fs/zfs.h>
59 #include <sys/dmu.h>
60 #include <sys/spa.h>
61 #include <sys/txg.h>
62 #include <sys/dbuf.h>
63 #include <sys/zap.h>
64 #include <sys/dirent.h>
65 #include <sys/policy.h>
66 #include <sys/sunddi.h>
67 #include <sys/filio.h>
68 #include <sys/sid.h>
69 #include "fs/fs_subr.h"
70 #include <sys/zfs_ctldir.h>
71 #include <sys/zfs_fuid.h>
72 #include <sys/dnlc.h>
73 #include <sys/zfs_rlock.h>
74 #include <sys/extdirent.h>
75 #include <sys/kidmap.h>
76 #include <sys/cred_impl.h>
77 #include <sys/attr.h>
78 
79 /*
80  * Programming rules.
81  *
82  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
83  * properly lock its in-core state, create a DMU transaction, do the work,
84  * record this work in the intent log (ZIL), commit the DMU transaction,
85  * and wait for the intent log to commit if it is a synchronous operation.
86  * Moreover, the vnode ops must work in both normal and log replay context.
87  * The ordering of events is important to avoid deadlocks and references
88  * to freed memory.  The example below illustrates the following Big Rules:
89  *
90  *  (1) A check must be made in each zfs thread for a mounted file system.
91  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
92  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
93  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
94  *      can return EIO from the calling function.
95  *
96  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
97  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
98  *	First, if it's the last reference, the vnode/znode
99  *	can be freed, so the zp may point to freed memory.  Second, the last
100  *	reference will call zfs_zinactive(), which may induce a lot of work --
101  *	pushing cached pages (which acquires range locks) and syncing out
102  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
103  *	which could deadlock the system if you were already holding one.
104  *	If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
105  *
106  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
107  *	as they can span dmu_tx_assign() calls.
108  *
109  *  (4)	Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
110  *	This is critical because we don't want to block while holding locks.
111  *	Note, in particular, that if a lock is sometimes acquired before
112  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
113  *	use a non-blocking assign can deadlock the system.  The scenario:
114  *
115  *	Thread A has grabbed a lock before calling dmu_tx_assign().
116  *	Thread B is in an already-assigned tx, and blocks for this lock.
117  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
118  *	forever, because the previous txg can't quiesce until B's tx commits.
119  *
120  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
121  *	then drop all locks, call dmu_tx_wait(), and try again.
122  *
123  *  (5)	If the operation succeeded, generate the intent log entry for it
124  *	before dropping locks.  This ensures that the ordering of events
125  *	in the intent log matches the order in which they actually occurred.
126  *      During ZIL replay the zfs_log_* functions will update the sequence
127  *	number to indicate the zil transaction has replayed.
128  *
129  *  (6)	At the end of each vnode op, the DMU tx must always commit,
130  *	regardless of whether there were any errors.
131  *
132  *  (7)	After dropping all locks, invoke zil_commit(zilog, seq, foid)
133  *	to ensure that synchronous semantics are provided when necessary.
134  *
135  * In general, this is how things should be ordered in each vnode op:
136  *
137  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
138  * top:
139  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
140  *	rw_enter(...);			// grab any other locks you need
141  *	tx = dmu_tx_create(...);	// get DMU tx
142  *	dmu_tx_hold_*();		// hold each object you might modify
143  *	error = dmu_tx_assign(tx, TXG_NOWAIT);	// try to assign
144  *	if (error) {
145  *		rw_exit(...);		// drop locks
146  *		zfs_dirent_unlock(dl);	// unlock directory entry
147  *		VN_RELE(...);		// release held vnodes
148  *		if (error == ERESTART) {
149  *			dmu_tx_wait(tx);
150  *			dmu_tx_abort(tx);
151  *			goto top;
152  *		}
153  *		dmu_tx_abort(tx);	// abort DMU tx
154  *		ZFS_EXIT(zfsvfs);	// finished in zfs
155  *		return (error);		// really out of space
156  *	}
157  *	error = do_real_work();		// do whatever this VOP does
158  *	if (error == 0)
159  *		zfs_log_*(...);		// on success, make ZIL entry
160  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
161  *	rw_exit(...);			// drop locks
162  *	zfs_dirent_unlock(dl);		// unlock directory entry
163  *	VN_RELE(...);			// release held vnodes
164  *	zil_commit(zilog, seq, foid);	// synchronous when necessary
165  *	ZFS_EXIT(zfsvfs);		// finished in zfs
166  *	return (error);			// done, report error
167  */
168 
169 /* ARGSUSED */
170 static int
171 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
172 {
173 	znode_t	*zp = VTOZ(*vpp);
174 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
175 
176 	ZFS_ENTER(zfsvfs);
177 	ZFS_VERIFY_ZP(zp);
178 
179 	if ((flag & FWRITE) && (zp->z_phys->zp_flags & ZFS_APPENDONLY) &&
180 	    ((flag & FAPPEND) == 0)) {
181 		ZFS_EXIT(zfsvfs);
182 		return (EPERM);
183 	}
184 
185 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
186 	    ZTOV(zp)->v_type == VREG &&
187 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
188 	    zp->z_phys->zp_size > 0) {
189 		if (fs_vscan(*vpp, cr, 0) != 0) {
190 			ZFS_EXIT(zfsvfs);
191 			return (EACCES);
192 		}
193 	}
194 
195 	/* Keep a count of the synchronous opens in the znode */
196 	if (flag & (FSYNC | FDSYNC))
197 		atomic_inc_32(&zp->z_sync_cnt);
198 
199 	ZFS_EXIT(zfsvfs);
200 	return (0);
201 }
202 
203 /* ARGSUSED */
204 static int
205 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
206     caller_context_t *ct)
207 {
208 	znode_t	*zp = VTOZ(vp);
209 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
210 
211 	ZFS_ENTER(zfsvfs);
212 	ZFS_VERIFY_ZP(zp);
213 
214 	/* Decrement the synchronous opens in the znode */
215 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
216 		atomic_dec_32(&zp->z_sync_cnt);
217 
218 	/*
219 	 * Clean up any locks held by this process on the vp.
220 	 */
221 	cleanlocks(vp, ddi_get_pid(), 0);
222 	cleanshares(vp, ddi_get_pid());
223 
224 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
225 	    ZTOV(zp)->v_type == VREG &&
226 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
227 	    zp->z_phys->zp_size > 0)
228 		VERIFY(fs_vscan(vp, cr, 1) == 0);
229 
230 	ZFS_EXIT(zfsvfs);
231 	return (0);
232 }
233 
234 /*
235  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
236  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
237  */
238 static int
239 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
240 {
241 	znode_t	*zp = VTOZ(vp);
242 	uint64_t noff = (uint64_t)*off; /* new offset */
243 	uint64_t file_sz;
244 	int error;
245 	boolean_t hole;
246 
247 	file_sz = zp->z_phys->zp_size;
248 	if (noff >= file_sz)  {
249 		return (ENXIO);
250 	}
251 
252 	if (cmd == _FIO_SEEK_HOLE)
253 		hole = B_TRUE;
254 	else
255 		hole = B_FALSE;
256 
257 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
258 
259 	/* end of file? */
260 	if ((error == ESRCH) || (noff > file_sz)) {
261 		/*
262 		 * Handle the virtual hole at the end of file.
263 		 */
264 		if (hole) {
265 			*off = file_sz;
266 			return (0);
267 		}
268 		return (ENXIO);
269 	}
270 
271 	if (noff < *off)
272 		return (error);
273 	*off = noff;
274 	return (error);
275 }
276 
277 /* ARGSUSED */
278 static int
279 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
280     int *rvalp, caller_context_t *ct)
281 {
282 	offset_t off;
283 	int error;
284 	zfsvfs_t *zfsvfs;
285 	znode_t *zp;
286 
287 	switch (com) {
288 	case _FIOFFS:
289 		return (zfs_sync(vp->v_vfsp, 0, cred));
290 
291 		/*
292 		 * The following two ioctls are used by bfu.  Faking out,
293 		 * necessary to avoid bfu errors.
294 		 */
295 	case _FIOGDIO:
296 	case _FIOSDIO:
297 		return (0);
298 
299 	case _FIO_SEEK_DATA:
300 	case _FIO_SEEK_HOLE:
301 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
302 			return (EFAULT);
303 
304 		zp = VTOZ(vp);
305 		zfsvfs = zp->z_zfsvfs;
306 		ZFS_ENTER(zfsvfs);
307 		ZFS_VERIFY_ZP(zp);
308 
309 		/* offset parameter is in/out */
310 		error = zfs_holey(vp, com, &off);
311 		ZFS_EXIT(zfsvfs);
312 		if (error)
313 			return (error);
314 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
315 			return (EFAULT);
316 		return (0);
317 	}
318 	return (ENOTTY);
319 }
320 
321 /*
322  * Utility functions to map and unmap a single physical page.  These
323  * are used to manage the mappable copies of ZFS file data, and therefore
324  * do not update ref/mod bits.
325  */
326 caddr_t
327 zfs_map_page(page_t *pp, enum seg_rw rw)
328 {
329 	if (kpm_enable)
330 		return (hat_kpm_mapin(pp, 0));
331 	ASSERT(rw == S_READ || rw == S_WRITE);
332 	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
333 	    (caddr_t)-1));
334 }
335 
336 void
337 zfs_unmap_page(page_t *pp, caddr_t addr)
338 {
339 	if (kpm_enable) {
340 		hat_kpm_mapout(pp, 0, addr);
341 	} else {
342 		ppmapout(addr);
343 	}
344 }
345 
346 /*
347  * When a file is memory mapped, we must keep the IO data synchronized
348  * between the DMU cache and the memory mapped pages.  What this means:
349  *
350  * On Write:	If we find a memory mapped page, we write to *both*
351  *		the page and the dmu buffer.
352  */
353 static void
354 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
355 {
356 	int64_t	off;
357 
358 	off = start & PAGEOFFSET;
359 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
360 		page_t *pp;
361 		uint64_t nbytes = MIN(PAGESIZE - off, len);
362 
363 		if (pp = page_lookup(vp, start, SE_SHARED)) {
364 			caddr_t va;
365 
366 			va = zfs_map_page(pp, S_WRITE);
367 			(void) dmu_read(os, oid, start+off, nbytes, va+off,
368 			    DMU_READ_PREFETCH);
369 			zfs_unmap_page(pp, va);
370 			page_unlock(pp);
371 		}
372 		len -= nbytes;
373 		off = 0;
374 	}
375 }
376 
377 /*
378  * When a file is memory mapped, we must keep the IO data synchronized
379  * between the DMU cache and the memory mapped pages.  What this means:
380  *
381  * On Read:	We "read" preferentially from memory mapped pages,
382  *		else we default from the dmu buffer.
383  *
384  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
385  *	the file is memory mapped.
386  */
387 static int
388 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
389 {
390 	znode_t *zp = VTOZ(vp);
391 	objset_t *os = zp->z_zfsvfs->z_os;
392 	int64_t	start, off;
393 	int len = nbytes;
394 	int error = 0;
395 
396 	start = uio->uio_loffset;
397 	off = start & PAGEOFFSET;
398 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
399 		page_t *pp;
400 		uint64_t bytes = MIN(PAGESIZE - off, len);
401 
402 		if (pp = page_lookup(vp, start, SE_SHARED)) {
403 			caddr_t va;
404 
405 			va = zfs_map_page(pp, S_READ);
406 			error = uiomove(va + off, bytes, UIO_READ, uio);
407 			zfs_unmap_page(pp, va);
408 			page_unlock(pp);
409 		} else {
410 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
411 		}
412 		len -= bytes;
413 		off = 0;
414 		if (error)
415 			break;
416 	}
417 	return (error);
418 }
419 
420 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
421 
422 /*
423  * Read bytes from specified file into supplied buffer.
424  *
425  *	IN:	vp	- vnode of file to be read from.
426  *		uio	- structure supplying read location, range info,
427  *			  and return buffer.
428  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
429  *		cr	- credentials of caller.
430  *		ct	- caller context
431  *
432  *	OUT:	uio	- updated offset and range, buffer filled.
433  *
434  *	RETURN:	0 if success
435  *		error code if failure
436  *
437  * Side Effects:
438  *	vp - atime updated if byte count > 0
439  */
440 /* ARGSUSED */
441 static int
442 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
443 {
444 	znode_t		*zp = VTOZ(vp);
445 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
446 	objset_t	*os;
447 	ssize_t		n, nbytes;
448 	int		error;
449 	rl_t		*rl;
450 
451 	ZFS_ENTER(zfsvfs);
452 	ZFS_VERIFY_ZP(zp);
453 	os = zfsvfs->z_os;
454 
455 	if (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) {
456 		ZFS_EXIT(zfsvfs);
457 		return (EACCES);
458 	}
459 
460 	/*
461 	 * Validate file offset
462 	 */
463 	if (uio->uio_loffset < (offset_t)0) {
464 		ZFS_EXIT(zfsvfs);
465 		return (EINVAL);
466 	}
467 
468 	/*
469 	 * Fasttrack empty reads
470 	 */
471 	if (uio->uio_resid == 0) {
472 		ZFS_EXIT(zfsvfs);
473 		return (0);
474 	}
475 
476 	/*
477 	 * Check for mandatory locks
478 	 */
479 	if (MANDMODE((mode_t)zp->z_phys->zp_mode)) {
480 		if (error = chklock(vp, FREAD,
481 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
482 			ZFS_EXIT(zfsvfs);
483 			return (error);
484 		}
485 	}
486 
487 	/*
488 	 * If we're in FRSYNC mode, sync out this znode before reading it.
489 	 */
490 	if (ioflag & FRSYNC)
491 		zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
492 
493 	/*
494 	 * Lock the range against changes.
495 	 */
496 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
497 
498 	/*
499 	 * If we are reading past end-of-file we can skip
500 	 * to the end; but we might still need to set atime.
501 	 */
502 	if (uio->uio_loffset >= zp->z_phys->zp_size) {
503 		error = 0;
504 		goto out;
505 	}
506 
507 	ASSERT(uio->uio_loffset < zp->z_phys->zp_size);
508 	n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset);
509 
510 	while (n > 0) {
511 		nbytes = MIN(n, zfs_read_chunk_size -
512 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
513 
514 		if (vn_has_cached_data(vp))
515 			error = mappedread(vp, nbytes, uio);
516 		else
517 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
518 		if (error) {
519 			/* convert checksum errors into IO errors */
520 			if (error == ECKSUM)
521 				error = EIO;
522 			break;
523 		}
524 
525 		n -= nbytes;
526 	}
527 
528 out:
529 	zfs_range_unlock(rl);
530 
531 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
532 	ZFS_EXIT(zfsvfs);
533 	return (error);
534 }
535 
536 /*
537  * Write the bytes to a file.
538  *
539  *	IN:	vp	- vnode of file to be written to.
540  *		uio	- structure supplying write location, range info,
541  *			  and data buffer.
542  *		ioflag	- FAPPEND flag set if in append mode.
543  *		cr	- credentials of caller.
544  *		ct	- caller context (NFS/CIFS fem monitor only)
545  *
546  *	OUT:	uio	- updated offset and range.
547  *
548  *	RETURN:	0 if success
549  *		error code if failure
550  *
551  * Timestamps:
552  *	vp - ctime|mtime updated if byte count > 0
553  */
554 /* ARGSUSED */
555 static int
556 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
557 {
558 	znode_t		*zp = VTOZ(vp);
559 	rlim64_t	limit = uio->uio_llimit;
560 	ssize_t		start_resid = uio->uio_resid;
561 	ssize_t		tx_bytes;
562 	uint64_t	end_size;
563 	dmu_tx_t	*tx;
564 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
565 	zilog_t		*zilog;
566 	offset_t	woff;
567 	ssize_t		n, nbytes;
568 	rl_t		*rl;
569 	int		max_blksz = zfsvfs->z_max_blksz;
570 	uint64_t	pflags;
571 	int		error;
572 	arc_buf_t	*abuf;
573 
574 	/*
575 	 * Fasttrack empty write
576 	 */
577 	n = start_resid;
578 	if (n == 0)
579 		return (0);
580 
581 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
582 		limit = MAXOFFSET_T;
583 
584 	ZFS_ENTER(zfsvfs);
585 	ZFS_VERIFY_ZP(zp);
586 
587 	/*
588 	 * If immutable or not appending then return EPERM
589 	 */
590 	pflags = zp->z_phys->zp_flags;
591 	if ((pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
592 	    ((pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
593 	    (uio->uio_loffset < zp->z_phys->zp_size))) {
594 		ZFS_EXIT(zfsvfs);
595 		return (EPERM);
596 	}
597 
598 	zilog = zfsvfs->z_log;
599 
600 	/*
601 	 * Pre-fault the pages to ensure slow (eg NFS) pages
602 	 * don't hold up txg.
603 	 */
604 	uio_prefaultpages(n, uio);
605 
606 	/*
607 	 * If in append mode, set the io offset pointer to eof.
608 	 */
609 	if (ioflag & FAPPEND) {
610 		/*
611 		 * Range lock for a file append:
612 		 * The value for the start of range will be determined by
613 		 * zfs_range_lock() (to guarantee append semantics).
614 		 * If this write will cause the block size to increase,
615 		 * zfs_range_lock() will lock the entire file, so we must
616 		 * later reduce the range after we grow the block size.
617 		 */
618 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
619 		if (rl->r_len == UINT64_MAX) {
620 			/* overlocked, zp_size can't change */
621 			woff = uio->uio_loffset = zp->z_phys->zp_size;
622 		} else {
623 			woff = uio->uio_loffset = rl->r_off;
624 		}
625 	} else {
626 		woff = uio->uio_loffset;
627 		/*
628 		 * Validate file offset
629 		 */
630 		if (woff < 0) {
631 			ZFS_EXIT(zfsvfs);
632 			return (EINVAL);
633 		}
634 
635 		/*
636 		 * If we need to grow the block size then zfs_range_lock()
637 		 * will lock a wider range than we request here.
638 		 * Later after growing the block size we reduce the range.
639 		 */
640 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
641 	}
642 
643 	if (woff >= limit) {
644 		zfs_range_unlock(rl);
645 		ZFS_EXIT(zfsvfs);
646 		return (EFBIG);
647 	}
648 
649 	if ((woff + n) > limit || woff > (limit - n))
650 		n = limit - woff;
651 
652 	/*
653 	 * Check for mandatory locks
654 	 */
655 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) &&
656 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
657 		zfs_range_unlock(rl);
658 		ZFS_EXIT(zfsvfs);
659 		return (error);
660 	}
661 	end_size = MAX(zp->z_phys->zp_size, woff + n);
662 
663 	/*
664 	 * Write the file in reasonable size chunks.  Each chunk is written
665 	 * in a separate transaction; this keeps the intent log records small
666 	 * and allows us to do more fine-grained space accounting.
667 	 */
668 	while (n > 0) {
669 		abuf = NULL;
670 		woff = uio->uio_loffset;
671 
672 again:
673 		if (zfs_usergroup_overquota(zfsvfs,
674 		    B_FALSE, zp->z_phys->zp_uid) ||
675 		    zfs_usergroup_overquota(zfsvfs,
676 		    B_TRUE, zp->z_phys->zp_gid)) {
677 			if (abuf != NULL)
678 				dmu_return_arcbuf(abuf);
679 			error = EDQUOT;
680 			break;
681 		}
682 
683 		/*
684 		 * If dmu_assign_arcbuf() is expected to execute with minimum
685 		 * overhead loan an arc buffer and copy user data to it before
686 		 * we enter a txg.  This avoids holding a txg forever while we
687 		 * pagefault on a hanging NFS server mapping.
688 		 */
689 		if (abuf == NULL && n >= max_blksz &&
690 		    woff >= zp->z_phys->zp_size &&
691 		    P2PHASE(woff, max_blksz) == 0 &&
692 		    zp->z_blksz == max_blksz) {
693 			size_t cbytes;
694 
695 			abuf = dmu_request_arcbuf(zp->z_dbuf, max_blksz);
696 			ASSERT(abuf != NULL);
697 			ASSERT(arc_buf_size(abuf) == max_blksz);
698 			if (error = uiocopy(abuf->b_data, max_blksz,
699 			    UIO_WRITE, uio, &cbytes)) {
700 				dmu_return_arcbuf(abuf);
701 				break;
702 			}
703 			ASSERT(cbytes == max_blksz);
704 		}
705 
706 		/*
707 		 * Start a transaction.
708 		 */
709 		tx = dmu_tx_create(zfsvfs->z_os);
710 		dmu_tx_hold_bonus(tx, zp->z_id);
711 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
712 		error = dmu_tx_assign(tx, TXG_NOWAIT);
713 		if (error) {
714 			if (error == ERESTART) {
715 				dmu_tx_wait(tx);
716 				dmu_tx_abort(tx);
717 				goto again;
718 			}
719 			dmu_tx_abort(tx);
720 			if (abuf != NULL)
721 				dmu_return_arcbuf(abuf);
722 			break;
723 		}
724 
725 		/*
726 		 * If zfs_range_lock() over-locked we grow the blocksize
727 		 * and then reduce the lock range.  This will only happen
728 		 * on the first iteration since zfs_range_reduce() will
729 		 * shrink down r_len to the appropriate size.
730 		 */
731 		if (rl->r_len == UINT64_MAX) {
732 			uint64_t new_blksz;
733 
734 			if (zp->z_blksz > max_blksz) {
735 				ASSERT(!ISP2(zp->z_blksz));
736 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
737 			} else {
738 				new_blksz = MIN(end_size, max_blksz);
739 			}
740 			zfs_grow_blocksize(zp, new_blksz, tx);
741 			zfs_range_reduce(rl, woff, n);
742 		}
743 
744 		/*
745 		 * XXX - should we really limit each write to z_max_blksz?
746 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
747 		 */
748 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
749 
750 		if (abuf == NULL) {
751 			tx_bytes = uio->uio_resid;
752 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id, uio,
753 			    nbytes, tx);
754 			tx_bytes -= uio->uio_resid;
755 		} else {
756 			tx_bytes = nbytes;
757 			ASSERT(tx_bytes == max_blksz);
758 			dmu_assign_arcbuf(zp->z_dbuf, woff, abuf, tx);
759 			ASSERT(tx_bytes <= uio->uio_resid);
760 			uioskip(uio, tx_bytes);
761 		}
762 		if (tx_bytes && vn_has_cached_data(vp)) {
763 			update_pages(vp, woff,
764 			    tx_bytes, zfsvfs->z_os, zp->z_id);
765 		}
766 
767 		/*
768 		 * If we made no progress, we're done.  If we made even
769 		 * partial progress, update the znode and ZIL accordingly.
770 		 */
771 		if (tx_bytes == 0) {
772 			dmu_tx_commit(tx);
773 			ASSERT(error != 0);
774 			break;
775 		}
776 
777 		/*
778 		 * Clear Set-UID/Set-GID bits on successful write if not
779 		 * privileged and at least one of the excute bits is set.
780 		 *
781 		 * It would be nice to to this after all writes have
782 		 * been done, but that would still expose the ISUID/ISGID
783 		 * to another app after the partial write is committed.
784 		 *
785 		 * Note: we don't call zfs_fuid_map_id() here because
786 		 * user 0 is not an ephemeral uid.
787 		 */
788 		mutex_enter(&zp->z_acl_lock);
789 		if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) |
790 		    (S_IXUSR >> 6))) != 0 &&
791 		    (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 &&
792 		    secpolicy_vnode_setid_retain(cr,
793 		    (zp->z_phys->zp_mode & S_ISUID) != 0 &&
794 		    zp->z_phys->zp_uid == 0) != 0) {
795 			zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID);
796 		}
797 		mutex_exit(&zp->z_acl_lock);
798 
799 		/*
800 		 * Update time stamp.  NOTE: This marks the bonus buffer as
801 		 * dirty, so we don't have to do it again for zp_size.
802 		 */
803 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
804 
805 		/*
806 		 * Update the file size (zp_size) if it has changed;
807 		 * account for possible concurrent updates.
808 		 */
809 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset)
810 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
811 			    uio->uio_loffset);
812 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
813 		dmu_tx_commit(tx);
814 
815 		if (error != 0)
816 			break;
817 		ASSERT(tx_bytes == nbytes);
818 		n -= nbytes;
819 	}
820 
821 	zfs_range_unlock(rl);
822 
823 	/*
824 	 * If we're in replay mode, or we made no progress, return error.
825 	 * Otherwise, it's at least a partial write, so it's successful.
826 	 */
827 	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
828 		ZFS_EXIT(zfsvfs);
829 		return (error);
830 	}
831 
832 	if (ioflag & (FSYNC | FDSYNC))
833 		zil_commit(zilog, zp->z_last_itx, zp->z_id);
834 
835 	ZFS_EXIT(zfsvfs);
836 	return (0);
837 }
838 
839 void
840 zfs_get_done(dmu_buf_t *db, void *vzgd)
841 {
842 	zgd_t *zgd = (zgd_t *)vzgd;
843 	rl_t *rl = zgd->zgd_rl;
844 	vnode_t *vp = ZTOV(rl->r_zp);
845 	objset_t *os = rl->r_zp->z_zfsvfs->z_os;
846 
847 	dmu_buf_rele(db, vzgd);
848 	zfs_range_unlock(rl);
849 	/*
850 	 * Release the vnode asynchronously as we currently have the
851 	 * txg stopped from syncing.
852 	 */
853 	VN_RELE_ASYNC(vp, dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
854 	zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
855 	kmem_free(zgd, sizeof (zgd_t));
856 }
857 
858 /*
859  * Get data to generate a TX_WRITE intent log record.
860  */
861 int
862 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
863 {
864 	zfsvfs_t *zfsvfs = arg;
865 	objset_t *os = zfsvfs->z_os;
866 	znode_t *zp;
867 	uint64_t off = lr->lr_offset;
868 	dmu_buf_t *db;
869 	rl_t *rl;
870 	zgd_t *zgd;
871 	int dlen = lr->lr_length;		/* length of user data */
872 	int error = 0;
873 
874 	ASSERT(zio);
875 	ASSERT(dlen != 0);
876 
877 	/*
878 	 * Nothing to do if the file has been removed
879 	 */
880 	if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0)
881 		return (ENOENT);
882 	if (zp->z_unlinked) {
883 		/*
884 		 * Release the vnode asynchronously as we currently have the
885 		 * txg stopped from syncing.
886 		 */
887 		VN_RELE_ASYNC(ZTOV(zp),
888 		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
889 		return (ENOENT);
890 	}
891 
892 	/*
893 	 * Write records come in two flavors: immediate and indirect.
894 	 * For small writes it's cheaper to store the data with the
895 	 * log record (immediate); for large writes it's cheaper to
896 	 * sync the data and get a pointer to it (indirect) so that
897 	 * we don't have to write the data twice.
898 	 */
899 	if (buf != NULL) { /* immediate write */
900 		rl = zfs_range_lock(zp, off, dlen, RL_READER);
901 		/* test for truncation needs to be done while range locked */
902 		if (off >= zp->z_phys->zp_size) {
903 			error = ENOENT;
904 			goto out;
905 		}
906 		VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf,
907 		    DMU_READ_NO_PREFETCH));
908 	} else { /* indirect write */
909 		uint64_t boff; /* block starting offset */
910 
911 		/*
912 		 * Have to lock the whole block to ensure when it's
913 		 * written out and it's checksum is being calculated
914 		 * that no one can change the data. We need to re-check
915 		 * blocksize after we get the lock in case it's changed!
916 		 */
917 		for (;;) {
918 			if (ISP2(zp->z_blksz)) {
919 				boff = P2ALIGN_TYPED(off, zp->z_blksz,
920 				    uint64_t);
921 			} else {
922 				boff = 0;
923 			}
924 			dlen = zp->z_blksz;
925 			rl = zfs_range_lock(zp, boff, dlen, RL_READER);
926 			if (zp->z_blksz == dlen)
927 				break;
928 			zfs_range_unlock(rl);
929 		}
930 		/* test for truncation needs to be done while range locked */
931 		if (off >= zp->z_phys->zp_size) {
932 			error = ENOENT;
933 			goto out;
934 		}
935 		zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
936 		zgd->zgd_rl = rl;
937 		zgd->zgd_zilog = zfsvfs->z_log;
938 		zgd->zgd_bp = &lr->lr_blkptr;
939 		VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db));
940 		ASSERT(boff == db->db_offset);
941 		lr->lr_blkoff = off - boff;
942 		error = dmu_sync(zio, db, &lr->lr_blkptr,
943 		    lr->lr_common.lrc_txg, zfs_get_done, zgd);
944 		ASSERT((error && error != EINPROGRESS) ||
945 		    lr->lr_length <= zp->z_blksz);
946 		if (error == 0)
947 			zil_add_block(zfsvfs->z_log, &lr->lr_blkptr);
948 		/*
949 		 * If we get EINPROGRESS, then we need to wait for a
950 		 * write IO initiated by dmu_sync() to complete before
951 		 * we can release this dbuf.  We will finish everything
952 		 * up in the zfs_get_done() callback.
953 		 */
954 		if (error == EINPROGRESS)
955 			return (0);
956 		dmu_buf_rele(db, zgd);
957 		kmem_free(zgd, sizeof (zgd_t));
958 	}
959 out:
960 	zfs_range_unlock(rl);
961 	/*
962 	 * Release the vnode asynchronously as we currently have the
963 	 * txg stopped from syncing.
964 	 */
965 	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
966 	return (error);
967 }
968 
969 /*ARGSUSED*/
970 static int
971 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
972     caller_context_t *ct)
973 {
974 	znode_t *zp = VTOZ(vp);
975 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
976 	int error;
977 
978 	ZFS_ENTER(zfsvfs);
979 	ZFS_VERIFY_ZP(zp);
980 
981 	if (flag & V_ACE_MASK)
982 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
983 	else
984 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
985 
986 	ZFS_EXIT(zfsvfs);
987 	return (error);
988 }
989 
990 /*
991  * Lookup an entry in a directory, or an extended attribute directory.
992  * If it exists, return a held vnode reference for it.
993  *
994  *	IN:	dvp	- vnode of directory to search.
995  *		nm	- name of entry to lookup.
996  *		pnp	- full pathname to lookup [UNUSED].
997  *		flags	- LOOKUP_XATTR set if looking for an attribute.
998  *		rdir	- root directory vnode [UNUSED].
999  *		cr	- credentials of caller.
1000  *		ct	- caller context
1001  *		direntflags - directory lookup flags
1002  *		realpnp - returned pathname.
1003  *
1004  *	OUT:	vpp	- vnode of located entry, NULL if not found.
1005  *
1006  *	RETURN:	0 if success
1007  *		error code if failure
1008  *
1009  * Timestamps:
1010  *	NA
1011  */
1012 /* ARGSUSED */
1013 static int
1014 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1015     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1016     int *direntflags, pathname_t *realpnp)
1017 {
1018 	znode_t *zdp = VTOZ(dvp);
1019 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1020 	int	error;
1021 
1022 	ZFS_ENTER(zfsvfs);
1023 	ZFS_VERIFY_ZP(zdp);
1024 
1025 	*vpp = NULL;
1026 
1027 	if (flags & LOOKUP_XATTR) {
1028 		/*
1029 		 * If the xattr property is off, refuse the lookup request.
1030 		 */
1031 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1032 			ZFS_EXIT(zfsvfs);
1033 			return (EINVAL);
1034 		}
1035 
1036 		/*
1037 		 * We don't allow recursive attributes..
1038 		 * Maybe someday we will.
1039 		 */
1040 		if (zdp->z_phys->zp_flags & ZFS_XATTR) {
1041 			ZFS_EXIT(zfsvfs);
1042 			return (EINVAL);
1043 		}
1044 
1045 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1046 			ZFS_EXIT(zfsvfs);
1047 			return (error);
1048 		}
1049 
1050 		/*
1051 		 * Do we have permission to get into attribute directory?
1052 		 */
1053 
1054 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1055 		    B_FALSE, cr)) {
1056 			VN_RELE(*vpp);
1057 			*vpp = NULL;
1058 		}
1059 
1060 		ZFS_EXIT(zfsvfs);
1061 		return (error);
1062 	}
1063 
1064 	if (dvp->v_type != VDIR) {
1065 		ZFS_EXIT(zfsvfs);
1066 		return (ENOTDIR);
1067 	}
1068 
1069 	/*
1070 	 * Check accessibility of directory.
1071 	 */
1072 
1073 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1074 		ZFS_EXIT(zfsvfs);
1075 		return (error);
1076 	}
1077 
1078 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1079 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1080 		ZFS_EXIT(zfsvfs);
1081 		return (EILSEQ);
1082 	}
1083 
1084 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1085 	if (error == 0) {
1086 		/*
1087 		 * Convert device special files
1088 		 */
1089 		if (IS_DEVVP(*vpp)) {
1090 			vnode_t	*svp;
1091 
1092 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1093 			VN_RELE(*vpp);
1094 			if (svp == NULL)
1095 				error = ENOSYS;
1096 			else
1097 				*vpp = svp;
1098 		}
1099 	}
1100 
1101 	ZFS_EXIT(zfsvfs);
1102 	return (error);
1103 }
1104 
1105 /*
1106  * Attempt to create a new entry in a directory.  If the entry
1107  * already exists, truncate the file if permissible, else return
1108  * an error.  Return the vp of the created or trunc'd file.
1109  *
1110  *	IN:	dvp	- vnode of directory to put new file entry in.
1111  *		name	- name of new file entry.
1112  *		vap	- attributes of new file.
1113  *		excl	- flag indicating exclusive or non-exclusive mode.
1114  *		mode	- mode to open file with.
1115  *		cr	- credentials of caller.
1116  *		flag	- large file flag [UNUSED].
1117  *		ct	- caller context
1118  *		vsecp 	- ACL to be set
1119  *
1120  *	OUT:	vpp	- vnode of created or trunc'd entry.
1121  *
1122  *	RETURN:	0 if success
1123  *		error code if failure
1124  *
1125  * Timestamps:
1126  *	dvp - ctime|mtime updated if new entry created
1127  *	 vp - ctime|mtime always, atime if new
1128  */
1129 
1130 /* ARGSUSED */
1131 static int
1132 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1133     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1134     vsecattr_t *vsecp)
1135 {
1136 	znode_t		*zp, *dzp = VTOZ(dvp);
1137 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1138 	zilog_t		*zilog;
1139 	objset_t	*os;
1140 	zfs_dirlock_t	*dl;
1141 	dmu_tx_t	*tx;
1142 	int		error;
1143 	ksid_t		*ksid;
1144 	uid_t		uid;
1145 	gid_t		gid = crgetgid(cr);
1146 	zfs_acl_ids_t	acl_ids;
1147 	boolean_t	fuid_dirtied;
1148 
1149 	/*
1150 	 * If we have an ephemeral id, ACL, or XVATTR then
1151 	 * make sure file system is at proper version
1152 	 */
1153 
1154 	ksid = crgetsid(cr, KSID_OWNER);
1155 	if (ksid)
1156 		uid = ksid_getid(ksid);
1157 	else
1158 		uid = crgetuid(cr);
1159 
1160 	if (zfsvfs->z_use_fuids == B_FALSE &&
1161 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1162 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1163 		return (EINVAL);
1164 
1165 	ZFS_ENTER(zfsvfs);
1166 	ZFS_VERIFY_ZP(dzp);
1167 	os = zfsvfs->z_os;
1168 	zilog = zfsvfs->z_log;
1169 
1170 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1171 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1172 		ZFS_EXIT(zfsvfs);
1173 		return (EILSEQ);
1174 	}
1175 
1176 	if (vap->va_mask & AT_XVATTR) {
1177 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1178 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1179 			ZFS_EXIT(zfsvfs);
1180 			return (error);
1181 		}
1182 	}
1183 top:
1184 	*vpp = NULL;
1185 
1186 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1187 		vap->va_mode &= ~VSVTX;
1188 
1189 	if (*name == '\0') {
1190 		/*
1191 		 * Null component name refers to the directory itself.
1192 		 */
1193 		VN_HOLD(dvp);
1194 		zp = dzp;
1195 		dl = NULL;
1196 		error = 0;
1197 	} else {
1198 		/* possible VN_HOLD(zp) */
1199 		int zflg = 0;
1200 
1201 		if (flag & FIGNORECASE)
1202 			zflg |= ZCILOOK;
1203 
1204 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1205 		    NULL, NULL);
1206 		if (error) {
1207 			if (strcmp(name, "..") == 0)
1208 				error = EISDIR;
1209 			ZFS_EXIT(zfsvfs);
1210 			return (error);
1211 		}
1212 	}
1213 	if (zp == NULL) {
1214 		uint64_t txtype;
1215 
1216 		/*
1217 		 * Create a new file object and update the directory
1218 		 * to reference it.
1219 		 */
1220 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1221 			goto out;
1222 		}
1223 
1224 		/*
1225 		 * We only support the creation of regular files in
1226 		 * extended attribute directories.
1227 		 */
1228 		if ((dzp->z_phys->zp_flags & ZFS_XATTR) &&
1229 		    (vap->va_type != VREG)) {
1230 			error = EINVAL;
1231 			goto out;
1232 		}
1233 
1234 		if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, vsecp,
1235 		    &acl_ids)) != 0)
1236 			goto out;
1237 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1238 			error = EDQUOT;
1239 			goto out;
1240 		}
1241 
1242 		tx = dmu_tx_create(os);
1243 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1244 		fuid_dirtied = zfsvfs->z_fuid_dirty;
1245 		if (fuid_dirtied)
1246 			zfs_fuid_txhold(zfsvfs, tx);
1247 		dmu_tx_hold_bonus(tx, dzp->z_id);
1248 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1249 		if (acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1250 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1251 			    0, SPA_MAXBLOCKSIZE);
1252 		}
1253 		error = dmu_tx_assign(tx, TXG_NOWAIT);
1254 		if (error) {
1255 			zfs_acl_ids_free(&acl_ids);
1256 			zfs_dirent_unlock(dl);
1257 			if (error == ERESTART) {
1258 				dmu_tx_wait(tx);
1259 				dmu_tx_abort(tx);
1260 				goto top;
1261 			}
1262 			dmu_tx_abort(tx);
1263 			ZFS_EXIT(zfsvfs);
1264 			return (error);
1265 		}
1266 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, &acl_ids);
1267 
1268 		if (fuid_dirtied)
1269 			zfs_fuid_sync(zfsvfs, tx);
1270 
1271 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1272 
1273 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1274 		if (flag & FIGNORECASE)
1275 			txtype |= TX_CI;
1276 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1277 		    vsecp, acl_ids.z_fuidp, vap);
1278 		zfs_acl_ids_free(&acl_ids);
1279 		dmu_tx_commit(tx);
1280 	} else {
1281 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1282 
1283 		/*
1284 		 * A directory entry already exists for this name.
1285 		 */
1286 		/*
1287 		 * Can't truncate an existing file if in exclusive mode.
1288 		 */
1289 		if (excl == EXCL) {
1290 			error = EEXIST;
1291 			goto out;
1292 		}
1293 		/*
1294 		 * Can't open a directory for writing.
1295 		 */
1296 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1297 			error = EISDIR;
1298 			goto out;
1299 		}
1300 		/*
1301 		 * Verify requested access to file.
1302 		 */
1303 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1304 			goto out;
1305 		}
1306 
1307 		mutex_enter(&dzp->z_lock);
1308 		dzp->z_seq++;
1309 		mutex_exit(&dzp->z_lock);
1310 
1311 		/*
1312 		 * Truncate regular files if requested.
1313 		 */
1314 		if ((ZTOV(zp)->v_type == VREG) &&
1315 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1316 			/* we can't hold any locks when calling zfs_freesp() */
1317 			zfs_dirent_unlock(dl);
1318 			dl = NULL;
1319 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1320 			if (error == 0) {
1321 				vnevent_create(ZTOV(zp), ct);
1322 			}
1323 		}
1324 	}
1325 out:
1326 
1327 	if (dl)
1328 		zfs_dirent_unlock(dl);
1329 
1330 	if (error) {
1331 		if (zp)
1332 			VN_RELE(ZTOV(zp));
1333 	} else {
1334 		*vpp = ZTOV(zp);
1335 		/*
1336 		 * If vnode is for a device return a specfs vnode instead.
1337 		 */
1338 		if (IS_DEVVP(*vpp)) {
1339 			struct vnode *svp;
1340 
1341 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1342 			VN_RELE(*vpp);
1343 			if (svp == NULL) {
1344 				error = ENOSYS;
1345 			}
1346 			*vpp = svp;
1347 		}
1348 	}
1349 
1350 	ZFS_EXIT(zfsvfs);
1351 	return (error);
1352 }
1353 
1354 /*
1355  * Remove an entry from a directory.
1356  *
1357  *	IN:	dvp	- vnode of directory to remove entry from.
1358  *		name	- name of entry to remove.
1359  *		cr	- credentials of caller.
1360  *		ct	- caller context
1361  *		flags	- case flags
1362  *
1363  *	RETURN:	0 if success
1364  *		error code if failure
1365  *
1366  * Timestamps:
1367  *	dvp - ctime|mtime
1368  *	 vp - ctime (if nlink > 0)
1369  */
1370 /*ARGSUSED*/
1371 static int
1372 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1373     int flags)
1374 {
1375 	znode_t		*zp, *dzp = VTOZ(dvp);
1376 	znode_t		*xzp = NULL;
1377 	vnode_t		*vp;
1378 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1379 	zilog_t		*zilog;
1380 	uint64_t	acl_obj, xattr_obj;
1381 	zfs_dirlock_t	*dl;
1382 	dmu_tx_t	*tx;
1383 	boolean_t	may_delete_now, delete_now = FALSE;
1384 	boolean_t	unlinked, toobig = FALSE;
1385 	uint64_t	txtype;
1386 	pathname_t	*realnmp = NULL;
1387 	pathname_t	realnm;
1388 	int		error;
1389 	int		zflg = ZEXISTS;
1390 
1391 	ZFS_ENTER(zfsvfs);
1392 	ZFS_VERIFY_ZP(dzp);
1393 	zilog = zfsvfs->z_log;
1394 
1395 	if (flags & FIGNORECASE) {
1396 		zflg |= ZCILOOK;
1397 		pn_alloc(&realnm);
1398 		realnmp = &realnm;
1399 	}
1400 
1401 top:
1402 	/*
1403 	 * Attempt to lock directory; fail if entry doesn't exist.
1404 	 */
1405 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1406 	    NULL, realnmp)) {
1407 		if (realnmp)
1408 			pn_free(realnmp);
1409 		ZFS_EXIT(zfsvfs);
1410 		return (error);
1411 	}
1412 
1413 	vp = ZTOV(zp);
1414 
1415 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1416 		goto out;
1417 	}
1418 
1419 	/*
1420 	 * Need to use rmdir for removing directories.
1421 	 */
1422 	if (vp->v_type == VDIR) {
1423 		error = EPERM;
1424 		goto out;
1425 	}
1426 
1427 	vnevent_remove(vp, dvp, name, ct);
1428 
1429 	if (realnmp)
1430 		dnlc_remove(dvp, realnmp->pn_buf);
1431 	else
1432 		dnlc_remove(dvp, name);
1433 
1434 	mutex_enter(&vp->v_lock);
1435 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1436 	mutex_exit(&vp->v_lock);
1437 
1438 	/*
1439 	 * We may delete the znode now, or we may put it in the unlinked set;
1440 	 * it depends on whether we're the last link, and on whether there are
1441 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1442 	 * allow for either case.
1443 	 */
1444 	tx = dmu_tx_create(zfsvfs->z_os);
1445 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1446 	dmu_tx_hold_bonus(tx, zp->z_id);
1447 	if (may_delete_now) {
1448 		toobig =
1449 		    zp->z_phys->zp_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1450 		/* if the file is too big, only hold_free a token amount */
1451 		dmu_tx_hold_free(tx, zp->z_id, 0,
1452 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1453 	}
1454 
1455 	/* are there any extended attributes? */
1456 	if ((xattr_obj = zp->z_phys->zp_xattr) != 0) {
1457 		/* XXX - do we need this if we are deleting? */
1458 		dmu_tx_hold_bonus(tx, xattr_obj);
1459 	}
1460 
1461 	/* are there any additional acls */
1462 	if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 &&
1463 	    may_delete_now)
1464 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1465 
1466 	/* charge as an update -- would be nice not to charge at all */
1467 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1468 
1469 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1470 	if (error) {
1471 		zfs_dirent_unlock(dl);
1472 		VN_RELE(vp);
1473 		if (error == ERESTART) {
1474 			dmu_tx_wait(tx);
1475 			dmu_tx_abort(tx);
1476 			goto top;
1477 		}
1478 		if (realnmp)
1479 			pn_free(realnmp);
1480 		dmu_tx_abort(tx);
1481 		ZFS_EXIT(zfsvfs);
1482 		return (error);
1483 	}
1484 
1485 	/*
1486 	 * Remove the directory entry.
1487 	 */
1488 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1489 
1490 	if (error) {
1491 		dmu_tx_commit(tx);
1492 		goto out;
1493 	}
1494 
1495 	if (unlinked) {
1496 		mutex_enter(&vp->v_lock);
1497 		delete_now = may_delete_now && !toobig &&
1498 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1499 		    zp->z_phys->zp_xattr == xattr_obj &&
1500 		    zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj;
1501 		mutex_exit(&vp->v_lock);
1502 	}
1503 
1504 	if (delete_now) {
1505 		if (zp->z_phys->zp_xattr) {
1506 			error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
1507 			ASSERT3U(error, ==, 0);
1508 			ASSERT3U(xzp->z_phys->zp_links, ==, 2);
1509 			dmu_buf_will_dirty(xzp->z_dbuf, tx);
1510 			mutex_enter(&xzp->z_lock);
1511 			xzp->z_unlinked = 1;
1512 			xzp->z_phys->zp_links = 0;
1513 			mutex_exit(&xzp->z_lock);
1514 			zfs_unlinked_add(xzp, tx);
1515 			zp->z_phys->zp_xattr = 0; /* probably unnecessary */
1516 		}
1517 		mutex_enter(&zp->z_lock);
1518 		mutex_enter(&vp->v_lock);
1519 		vp->v_count--;
1520 		ASSERT3U(vp->v_count, ==, 0);
1521 		mutex_exit(&vp->v_lock);
1522 		mutex_exit(&zp->z_lock);
1523 		zfs_znode_delete(zp, tx);
1524 	} else if (unlinked) {
1525 		zfs_unlinked_add(zp, tx);
1526 	}
1527 
1528 	txtype = TX_REMOVE;
1529 	if (flags & FIGNORECASE)
1530 		txtype |= TX_CI;
1531 	zfs_log_remove(zilog, tx, txtype, dzp, name);
1532 
1533 	dmu_tx_commit(tx);
1534 out:
1535 	if (realnmp)
1536 		pn_free(realnmp);
1537 
1538 	zfs_dirent_unlock(dl);
1539 
1540 	if (!delete_now) {
1541 		VN_RELE(vp);
1542 	} else if (xzp) {
1543 		/* this rele is delayed to prevent nesting transactions */
1544 		VN_RELE(ZTOV(xzp));
1545 	}
1546 
1547 	ZFS_EXIT(zfsvfs);
1548 	return (error);
1549 }
1550 
1551 /*
1552  * Create a new directory and insert it into dvp using the name
1553  * provided.  Return a pointer to the inserted directory.
1554  *
1555  *	IN:	dvp	- vnode of directory to add subdir to.
1556  *		dirname	- name of new directory.
1557  *		vap	- attributes of new directory.
1558  *		cr	- credentials of caller.
1559  *		ct	- caller context
1560  *		vsecp	- ACL to be set
1561  *
1562  *	OUT:	vpp	- vnode of created directory.
1563  *
1564  *	RETURN:	0 if success
1565  *		error code if failure
1566  *
1567  * Timestamps:
1568  *	dvp - ctime|mtime updated
1569  *	 vp - ctime|mtime|atime updated
1570  */
1571 /*ARGSUSED*/
1572 static int
1573 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1574     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1575 {
1576 	znode_t		*zp, *dzp = VTOZ(dvp);
1577 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1578 	zilog_t		*zilog;
1579 	zfs_dirlock_t	*dl;
1580 	uint64_t	txtype;
1581 	dmu_tx_t	*tx;
1582 	int		error;
1583 	int		zf = ZNEW;
1584 	ksid_t		*ksid;
1585 	uid_t		uid;
1586 	gid_t		gid = crgetgid(cr);
1587 	zfs_acl_ids_t	acl_ids;
1588 	boolean_t	fuid_dirtied;
1589 
1590 	ASSERT(vap->va_type == VDIR);
1591 
1592 	/*
1593 	 * If we have an ephemeral id, ACL, or XVATTR then
1594 	 * make sure file system is at proper version
1595 	 */
1596 
1597 	ksid = crgetsid(cr, KSID_OWNER);
1598 	if (ksid)
1599 		uid = ksid_getid(ksid);
1600 	else
1601 		uid = crgetuid(cr);
1602 	if (zfsvfs->z_use_fuids == B_FALSE &&
1603 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1604 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1605 		return (EINVAL);
1606 
1607 	ZFS_ENTER(zfsvfs);
1608 	ZFS_VERIFY_ZP(dzp);
1609 	zilog = zfsvfs->z_log;
1610 
1611 	if (dzp->z_phys->zp_flags & ZFS_XATTR) {
1612 		ZFS_EXIT(zfsvfs);
1613 		return (EINVAL);
1614 	}
1615 
1616 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1617 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1618 		ZFS_EXIT(zfsvfs);
1619 		return (EILSEQ);
1620 	}
1621 	if (flags & FIGNORECASE)
1622 		zf |= ZCILOOK;
1623 
1624 	if (vap->va_mask & AT_XVATTR)
1625 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1626 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1627 			ZFS_EXIT(zfsvfs);
1628 			return (error);
1629 		}
1630 
1631 	/*
1632 	 * First make sure the new directory doesn't exist.
1633 	 */
1634 top:
1635 	*vpp = NULL;
1636 
1637 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1638 	    NULL, NULL)) {
1639 		ZFS_EXIT(zfsvfs);
1640 		return (error);
1641 	}
1642 
1643 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1644 		zfs_dirent_unlock(dl);
1645 		ZFS_EXIT(zfsvfs);
1646 		return (error);
1647 	}
1648 
1649 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, vsecp,
1650 	    &acl_ids)) != 0) {
1651 		zfs_dirent_unlock(dl);
1652 		ZFS_EXIT(zfsvfs);
1653 		return (error);
1654 	}
1655 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1656 		zfs_dirent_unlock(dl);
1657 		ZFS_EXIT(zfsvfs);
1658 		return (EDQUOT);
1659 	}
1660 
1661 	/*
1662 	 * Add a new entry to the directory.
1663 	 */
1664 	tx = dmu_tx_create(zfsvfs->z_os);
1665 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1666 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1667 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1668 	if (fuid_dirtied)
1669 		zfs_fuid_txhold(zfsvfs, tx);
1670 	if (acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE)
1671 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1672 		    0, SPA_MAXBLOCKSIZE);
1673 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1674 	if (error) {
1675 		zfs_acl_ids_free(&acl_ids);
1676 		zfs_dirent_unlock(dl);
1677 		if (error == ERESTART) {
1678 			dmu_tx_wait(tx);
1679 			dmu_tx_abort(tx);
1680 			goto top;
1681 		}
1682 		dmu_tx_abort(tx);
1683 		ZFS_EXIT(zfsvfs);
1684 		return (error);
1685 	}
1686 
1687 	/*
1688 	 * Create new node.
1689 	 */
1690 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, &acl_ids);
1691 
1692 	if (fuid_dirtied)
1693 		zfs_fuid_sync(zfsvfs, tx);
1694 	/*
1695 	 * Now put new name in parent dir.
1696 	 */
1697 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1698 
1699 	*vpp = ZTOV(zp);
1700 
1701 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1702 	if (flags & FIGNORECASE)
1703 		txtype |= TX_CI;
1704 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1705 	    acl_ids.z_fuidp, vap);
1706 
1707 	zfs_acl_ids_free(&acl_ids);
1708 	dmu_tx_commit(tx);
1709 
1710 	zfs_dirent_unlock(dl);
1711 
1712 	ZFS_EXIT(zfsvfs);
1713 	return (0);
1714 }
1715 
1716 /*
1717  * Remove a directory subdir entry.  If the current working
1718  * directory is the same as the subdir to be removed, the
1719  * remove will fail.
1720  *
1721  *	IN:	dvp	- vnode of directory to remove from.
1722  *		name	- name of directory to be removed.
1723  *		cwd	- vnode of current working directory.
1724  *		cr	- credentials of caller.
1725  *		ct	- caller context
1726  *		flags	- case flags
1727  *
1728  *	RETURN:	0 if success
1729  *		error code if failure
1730  *
1731  * Timestamps:
1732  *	dvp - ctime|mtime updated
1733  */
1734 /*ARGSUSED*/
1735 static int
1736 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1737     caller_context_t *ct, int flags)
1738 {
1739 	znode_t		*dzp = VTOZ(dvp);
1740 	znode_t		*zp;
1741 	vnode_t		*vp;
1742 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1743 	zilog_t		*zilog;
1744 	zfs_dirlock_t	*dl;
1745 	dmu_tx_t	*tx;
1746 	int		error;
1747 	int		zflg = ZEXISTS;
1748 
1749 	ZFS_ENTER(zfsvfs);
1750 	ZFS_VERIFY_ZP(dzp);
1751 	zilog = zfsvfs->z_log;
1752 
1753 	if (flags & FIGNORECASE)
1754 		zflg |= ZCILOOK;
1755 top:
1756 	zp = NULL;
1757 
1758 	/*
1759 	 * Attempt to lock directory; fail if entry doesn't exist.
1760 	 */
1761 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1762 	    NULL, NULL)) {
1763 		ZFS_EXIT(zfsvfs);
1764 		return (error);
1765 	}
1766 
1767 	vp = ZTOV(zp);
1768 
1769 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1770 		goto out;
1771 	}
1772 
1773 	if (vp->v_type != VDIR) {
1774 		error = ENOTDIR;
1775 		goto out;
1776 	}
1777 
1778 	if (vp == cwd) {
1779 		error = EINVAL;
1780 		goto out;
1781 	}
1782 
1783 	vnevent_rmdir(vp, dvp, name, ct);
1784 
1785 	/*
1786 	 * Grab a lock on the directory to make sure that noone is
1787 	 * trying to add (or lookup) entries while we are removing it.
1788 	 */
1789 	rw_enter(&zp->z_name_lock, RW_WRITER);
1790 
1791 	/*
1792 	 * Grab a lock on the parent pointer to make sure we play well
1793 	 * with the treewalk and directory rename code.
1794 	 */
1795 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1796 
1797 	tx = dmu_tx_create(zfsvfs->z_os);
1798 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1799 	dmu_tx_hold_bonus(tx, zp->z_id);
1800 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1801 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1802 	if (error) {
1803 		rw_exit(&zp->z_parent_lock);
1804 		rw_exit(&zp->z_name_lock);
1805 		zfs_dirent_unlock(dl);
1806 		VN_RELE(vp);
1807 		if (error == ERESTART) {
1808 			dmu_tx_wait(tx);
1809 			dmu_tx_abort(tx);
1810 			goto top;
1811 		}
1812 		dmu_tx_abort(tx);
1813 		ZFS_EXIT(zfsvfs);
1814 		return (error);
1815 	}
1816 
1817 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
1818 
1819 	if (error == 0) {
1820 		uint64_t txtype = TX_RMDIR;
1821 		if (flags & FIGNORECASE)
1822 			txtype |= TX_CI;
1823 		zfs_log_remove(zilog, tx, txtype, dzp, name);
1824 	}
1825 
1826 	dmu_tx_commit(tx);
1827 
1828 	rw_exit(&zp->z_parent_lock);
1829 	rw_exit(&zp->z_name_lock);
1830 out:
1831 	zfs_dirent_unlock(dl);
1832 
1833 	VN_RELE(vp);
1834 
1835 	ZFS_EXIT(zfsvfs);
1836 	return (error);
1837 }
1838 
1839 /*
1840  * Read as many directory entries as will fit into the provided
1841  * buffer from the given directory cursor position (specified in
1842  * the uio structure.
1843  *
1844  *	IN:	vp	- vnode of directory to read.
1845  *		uio	- structure supplying read location, range info,
1846  *			  and return buffer.
1847  *		cr	- credentials of caller.
1848  *		ct	- caller context
1849  *		flags	- case flags
1850  *
1851  *	OUT:	uio	- updated offset and range, buffer filled.
1852  *		eofp	- set to true if end-of-file detected.
1853  *
1854  *	RETURN:	0 if success
1855  *		error code if failure
1856  *
1857  * Timestamps:
1858  *	vp - atime updated
1859  *
1860  * Note that the low 4 bits of the cookie returned by zap is always zero.
1861  * This allows us to use the low range for "special" directory entries:
1862  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1863  * we use the offset 2 for the '.zfs' directory.
1864  */
1865 /* ARGSUSED */
1866 static int
1867 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
1868     caller_context_t *ct, int flags)
1869 {
1870 	znode_t		*zp = VTOZ(vp);
1871 	iovec_t		*iovp;
1872 	edirent_t	*eodp;
1873 	dirent64_t	*odp;
1874 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1875 	objset_t	*os;
1876 	caddr_t		outbuf;
1877 	size_t		bufsize;
1878 	zap_cursor_t	zc;
1879 	zap_attribute_t	zap;
1880 	uint_t		bytes_wanted;
1881 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1882 	int		local_eof;
1883 	int		outcount;
1884 	int		error;
1885 	uint8_t		prefetch;
1886 	boolean_t	check_sysattrs;
1887 
1888 	ZFS_ENTER(zfsvfs);
1889 	ZFS_VERIFY_ZP(zp);
1890 
1891 	/*
1892 	 * If we are not given an eof variable,
1893 	 * use a local one.
1894 	 */
1895 	if (eofp == NULL)
1896 		eofp = &local_eof;
1897 
1898 	/*
1899 	 * Check for valid iov_len.
1900 	 */
1901 	if (uio->uio_iov->iov_len <= 0) {
1902 		ZFS_EXIT(zfsvfs);
1903 		return (EINVAL);
1904 	}
1905 
1906 	/*
1907 	 * Quit if directory has been removed (posix)
1908 	 */
1909 	if ((*eofp = zp->z_unlinked) != 0) {
1910 		ZFS_EXIT(zfsvfs);
1911 		return (0);
1912 	}
1913 
1914 	error = 0;
1915 	os = zfsvfs->z_os;
1916 	offset = uio->uio_loffset;
1917 	prefetch = zp->z_zn_prefetch;
1918 
1919 	/*
1920 	 * Initialize the iterator cursor.
1921 	 */
1922 	if (offset <= 3) {
1923 		/*
1924 		 * Start iteration from the beginning of the directory.
1925 		 */
1926 		zap_cursor_init(&zc, os, zp->z_id);
1927 	} else {
1928 		/*
1929 		 * The offset is a serialized cursor.
1930 		 */
1931 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1932 	}
1933 
1934 	/*
1935 	 * Get space to change directory entries into fs independent format.
1936 	 */
1937 	iovp = uio->uio_iov;
1938 	bytes_wanted = iovp->iov_len;
1939 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
1940 		bufsize = bytes_wanted;
1941 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
1942 		odp = (struct dirent64 *)outbuf;
1943 	} else {
1944 		bufsize = bytes_wanted;
1945 		odp = (struct dirent64 *)iovp->iov_base;
1946 	}
1947 	eodp = (struct edirent *)odp;
1948 
1949 	/*
1950 	 * If this VFS supports the system attribute view interface; and
1951 	 * we're looking at an extended attribute directory; and we care
1952 	 * about normalization conflicts on this vfs; then we must check
1953 	 * for normalization conflicts with the sysattr name space.
1954 	 */
1955 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
1956 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
1957 	    (flags & V_RDDIR_ENTFLAGS);
1958 
1959 	/*
1960 	 * Transform to file-system independent format
1961 	 */
1962 	outcount = 0;
1963 	while (outcount < bytes_wanted) {
1964 		ino64_t objnum;
1965 		ushort_t reclen;
1966 		off64_t *next;
1967 
1968 		/*
1969 		 * Special case `.', `..', and `.zfs'.
1970 		 */
1971 		if (offset == 0) {
1972 			(void) strcpy(zap.za_name, ".");
1973 			zap.za_normalization_conflict = 0;
1974 			objnum = zp->z_id;
1975 		} else if (offset == 1) {
1976 			(void) strcpy(zap.za_name, "..");
1977 			zap.za_normalization_conflict = 0;
1978 			objnum = zp->z_phys->zp_parent;
1979 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
1980 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1981 			zap.za_normalization_conflict = 0;
1982 			objnum = ZFSCTL_INO_ROOT;
1983 		} else {
1984 			/*
1985 			 * Grab next entry.
1986 			 */
1987 			if (error = zap_cursor_retrieve(&zc, &zap)) {
1988 				if ((*eofp = (error == ENOENT)) != 0)
1989 					break;
1990 				else
1991 					goto update;
1992 			}
1993 
1994 			if (zap.za_integer_length != 8 ||
1995 			    zap.za_num_integers != 1) {
1996 				cmn_err(CE_WARN, "zap_readdir: bad directory "
1997 				    "entry, obj = %lld, offset = %lld\n",
1998 				    (u_longlong_t)zp->z_id,
1999 				    (u_longlong_t)offset);
2000 				error = ENXIO;
2001 				goto update;
2002 			}
2003 
2004 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2005 			/*
2006 			 * MacOS X can extract the object type here such as:
2007 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2008 			 */
2009 
2010 			if (check_sysattrs && !zap.za_normalization_conflict) {
2011 				zap.za_normalization_conflict =
2012 				    xattr_sysattr_casechk(zap.za_name);
2013 			}
2014 		}
2015 
2016 		if (flags & V_RDDIR_ENTFLAGS)
2017 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2018 		else
2019 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2020 
2021 		/*
2022 		 * Will this entry fit in the buffer?
2023 		 */
2024 		if (outcount + reclen > bufsize) {
2025 			/*
2026 			 * Did we manage to fit anything in the buffer?
2027 			 */
2028 			if (!outcount) {
2029 				error = EINVAL;
2030 				goto update;
2031 			}
2032 			break;
2033 		}
2034 		if (flags & V_RDDIR_ENTFLAGS) {
2035 			/*
2036 			 * Add extended flag entry:
2037 			 */
2038 			eodp->ed_ino = objnum;
2039 			eodp->ed_reclen = reclen;
2040 			/* NOTE: ed_off is the offset for the *next* entry */
2041 			next = &(eodp->ed_off);
2042 			eodp->ed_eflags = zap.za_normalization_conflict ?
2043 			    ED_CASE_CONFLICT : 0;
2044 			(void) strncpy(eodp->ed_name, zap.za_name,
2045 			    EDIRENT_NAMELEN(reclen));
2046 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2047 		} else {
2048 			/*
2049 			 * Add normal entry:
2050 			 */
2051 			odp->d_ino = objnum;
2052 			odp->d_reclen = reclen;
2053 			/* NOTE: d_off is the offset for the *next* entry */
2054 			next = &(odp->d_off);
2055 			(void) strncpy(odp->d_name, zap.za_name,
2056 			    DIRENT64_NAMELEN(reclen));
2057 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2058 		}
2059 		outcount += reclen;
2060 
2061 		ASSERT(outcount <= bufsize);
2062 
2063 		/* Prefetch znode */
2064 		if (prefetch)
2065 			dmu_prefetch(os, objnum, 0, 0);
2066 
2067 		/*
2068 		 * Move to the next entry, fill in the previous offset.
2069 		 */
2070 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2071 			zap_cursor_advance(&zc);
2072 			offset = zap_cursor_serialize(&zc);
2073 		} else {
2074 			offset += 1;
2075 		}
2076 		*next = offset;
2077 	}
2078 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2079 
2080 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2081 		iovp->iov_base += outcount;
2082 		iovp->iov_len -= outcount;
2083 		uio->uio_resid -= outcount;
2084 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2085 		/*
2086 		 * Reset the pointer.
2087 		 */
2088 		offset = uio->uio_loffset;
2089 	}
2090 
2091 update:
2092 	zap_cursor_fini(&zc);
2093 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2094 		kmem_free(outbuf, bufsize);
2095 
2096 	if (error == ENOENT)
2097 		error = 0;
2098 
2099 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2100 
2101 	uio->uio_loffset = offset;
2102 	ZFS_EXIT(zfsvfs);
2103 	return (error);
2104 }
2105 
2106 ulong_t zfs_fsync_sync_cnt = 4;
2107 
2108 static int
2109 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2110 {
2111 	znode_t	*zp = VTOZ(vp);
2112 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2113 
2114 	/*
2115 	 * Regardless of whether this is required for standards conformance,
2116 	 * this is the logical behavior when fsync() is called on a file with
2117 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2118 	 * going to be pushed out as part of the zil_commit().
2119 	 */
2120 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2121 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2122 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2123 
2124 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2125 
2126 	ZFS_ENTER(zfsvfs);
2127 	ZFS_VERIFY_ZP(zp);
2128 	zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
2129 	ZFS_EXIT(zfsvfs);
2130 	return (0);
2131 }
2132 
2133 
2134 /*
2135  * Get the requested file attributes and place them in the provided
2136  * vattr structure.
2137  *
2138  *	IN:	vp	- vnode of file.
2139  *		vap	- va_mask identifies requested attributes.
2140  *			  If AT_XVATTR set, then optional attrs are requested
2141  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2142  *		cr	- credentials of caller.
2143  *		ct	- caller context
2144  *
2145  *	OUT:	vap	- attribute values.
2146  *
2147  *	RETURN:	0 (always succeeds)
2148  */
2149 /* ARGSUSED */
2150 static int
2151 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2152     caller_context_t *ct)
2153 {
2154 	znode_t *zp = VTOZ(vp);
2155 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2156 	znode_phys_t *pzp;
2157 	int	error = 0;
2158 	uint64_t links;
2159 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2160 	xoptattr_t *xoap = NULL;
2161 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2162 
2163 	ZFS_ENTER(zfsvfs);
2164 	ZFS_VERIFY_ZP(zp);
2165 	pzp = zp->z_phys;
2166 
2167 	mutex_enter(&zp->z_lock);
2168 
2169 	/*
2170 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2171 	 * Also, if we are the owner don't bother, since owner should
2172 	 * always be allowed to read basic attributes of file.
2173 	 */
2174 	if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL) &&
2175 	    (pzp->zp_uid != crgetuid(cr))) {
2176 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2177 		    skipaclchk, cr)) {
2178 			mutex_exit(&zp->z_lock);
2179 			ZFS_EXIT(zfsvfs);
2180 			return (error);
2181 		}
2182 	}
2183 
2184 	/*
2185 	 * Return all attributes.  It's cheaper to provide the answer
2186 	 * than to determine whether we were asked the question.
2187 	 */
2188 
2189 	vap->va_type = vp->v_type;
2190 	vap->va_mode = pzp->zp_mode & MODEMASK;
2191 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2192 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2193 	vap->va_nodeid = zp->z_id;
2194 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2195 		links = pzp->zp_links + 1;
2196 	else
2197 		links = pzp->zp_links;
2198 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2199 	vap->va_size = pzp->zp_size;
2200 	vap->va_rdev = vp->v_rdev;
2201 	vap->va_seq = zp->z_seq;
2202 
2203 	/*
2204 	 * Add in any requested optional attributes and the create time.
2205 	 * Also set the corresponding bits in the returned attribute bitmap.
2206 	 */
2207 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2208 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2209 			xoap->xoa_archive =
2210 			    ((pzp->zp_flags & ZFS_ARCHIVE) != 0);
2211 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2212 		}
2213 
2214 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2215 			xoap->xoa_readonly =
2216 			    ((pzp->zp_flags & ZFS_READONLY) != 0);
2217 			XVA_SET_RTN(xvap, XAT_READONLY);
2218 		}
2219 
2220 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2221 			xoap->xoa_system =
2222 			    ((pzp->zp_flags & ZFS_SYSTEM) != 0);
2223 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2224 		}
2225 
2226 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2227 			xoap->xoa_hidden =
2228 			    ((pzp->zp_flags & ZFS_HIDDEN) != 0);
2229 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2230 		}
2231 
2232 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2233 			xoap->xoa_nounlink =
2234 			    ((pzp->zp_flags & ZFS_NOUNLINK) != 0);
2235 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2236 		}
2237 
2238 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2239 			xoap->xoa_immutable =
2240 			    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0);
2241 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2242 		}
2243 
2244 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2245 			xoap->xoa_appendonly =
2246 			    ((pzp->zp_flags & ZFS_APPENDONLY) != 0);
2247 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2248 		}
2249 
2250 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2251 			xoap->xoa_nodump =
2252 			    ((pzp->zp_flags & ZFS_NODUMP) != 0);
2253 			XVA_SET_RTN(xvap, XAT_NODUMP);
2254 		}
2255 
2256 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2257 			xoap->xoa_opaque =
2258 			    ((pzp->zp_flags & ZFS_OPAQUE) != 0);
2259 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2260 		}
2261 
2262 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2263 			xoap->xoa_av_quarantined =
2264 			    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0);
2265 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2266 		}
2267 
2268 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2269 			xoap->xoa_av_modified =
2270 			    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0);
2271 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2272 		}
2273 
2274 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2275 		    vp->v_type == VREG &&
2276 		    (pzp->zp_flags & ZFS_BONUS_SCANSTAMP)) {
2277 			size_t len;
2278 			dmu_object_info_t doi;
2279 
2280 			/*
2281 			 * Only VREG files have anti-virus scanstamps, so we
2282 			 * won't conflict with symlinks in the bonus buffer.
2283 			 */
2284 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2285 			len = sizeof (xoap->xoa_av_scanstamp) +
2286 			    sizeof (znode_phys_t);
2287 			if (len <= doi.doi_bonus_size) {
2288 				/*
2289 				 * pzp points to the start of the
2290 				 * znode_phys_t. pzp + 1 points to the
2291 				 * first byte after the znode_phys_t.
2292 				 */
2293 				(void) memcpy(xoap->xoa_av_scanstamp,
2294 				    pzp + 1,
2295 				    sizeof (xoap->xoa_av_scanstamp));
2296 				XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
2297 			}
2298 		}
2299 
2300 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2301 			ZFS_TIME_DECODE(&xoap->xoa_createtime, pzp->zp_crtime);
2302 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2303 		}
2304 	}
2305 
2306 	ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime);
2307 	ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime);
2308 	ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime);
2309 
2310 	mutex_exit(&zp->z_lock);
2311 
2312 	dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks);
2313 
2314 	if (zp->z_blksz == 0) {
2315 		/*
2316 		 * Block size hasn't been set; suggest maximal I/O transfers.
2317 		 */
2318 		vap->va_blksize = zfsvfs->z_max_blksz;
2319 	}
2320 
2321 	ZFS_EXIT(zfsvfs);
2322 	return (0);
2323 }
2324 
2325 /*
2326  * Set the file attributes to the values contained in the
2327  * vattr structure.
2328  *
2329  *	IN:	vp	- vnode of file to be modified.
2330  *		vap	- new attribute values.
2331  *			  If AT_XVATTR set, then optional attrs are being set
2332  *		flags	- ATTR_UTIME set if non-default time values provided.
2333  *			- ATTR_NOACLCHECK (CIFS context only).
2334  *		cr	- credentials of caller.
2335  *		ct	- caller context
2336  *
2337  *	RETURN:	0 if success
2338  *		error code if failure
2339  *
2340  * Timestamps:
2341  *	vp - ctime updated, mtime updated if size changed.
2342  */
2343 /* ARGSUSED */
2344 static int
2345 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2346 	caller_context_t *ct)
2347 {
2348 	znode_t		*zp = VTOZ(vp);
2349 	znode_phys_t	*pzp;
2350 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2351 	zilog_t		*zilog;
2352 	dmu_tx_t	*tx;
2353 	vattr_t		oldva;
2354 	xvattr_t	tmpxvattr;
2355 	uint_t		mask = vap->va_mask;
2356 	uint_t		saved_mask;
2357 	int		trim_mask = 0;
2358 	uint64_t	new_mode;
2359 	uint64_t	new_uid, new_gid;
2360 	znode_t		*attrzp;
2361 	int		need_policy = FALSE;
2362 	int		err;
2363 	zfs_fuid_info_t *fuidp = NULL;
2364 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2365 	xoptattr_t	*xoap;
2366 	zfs_acl_t	*aclp = NULL;
2367 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2368 	boolean_t fuid_dirtied = B_FALSE;
2369 
2370 	if (mask == 0)
2371 		return (0);
2372 
2373 	if (mask & AT_NOSET)
2374 		return (EINVAL);
2375 
2376 	ZFS_ENTER(zfsvfs);
2377 	ZFS_VERIFY_ZP(zp);
2378 
2379 	pzp = zp->z_phys;
2380 	zilog = zfsvfs->z_log;
2381 
2382 	/*
2383 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2384 	 * that file system is at proper version level
2385 	 */
2386 
2387 	if (zfsvfs->z_use_fuids == B_FALSE &&
2388 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2389 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2390 	    (mask & AT_XVATTR))) {
2391 		ZFS_EXIT(zfsvfs);
2392 		return (EINVAL);
2393 	}
2394 
2395 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2396 		ZFS_EXIT(zfsvfs);
2397 		return (EISDIR);
2398 	}
2399 
2400 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2401 		ZFS_EXIT(zfsvfs);
2402 		return (EINVAL);
2403 	}
2404 
2405 	/*
2406 	 * If this is an xvattr_t, then get a pointer to the structure of
2407 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2408 	 */
2409 	xoap = xva_getxoptattr(xvap);
2410 
2411 	xva_init(&tmpxvattr);
2412 
2413 	/*
2414 	 * Immutable files can only alter immutable bit and atime
2415 	 */
2416 	if ((pzp->zp_flags & ZFS_IMMUTABLE) &&
2417 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2418 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2419 		ZFS_EXIT(zfsvfs);
2420 		return (EPERM);
2421 	}
2422 
2423 	if ((mask & AT_SIZE) && (pzp->zp_flags & ZFS_READONLY)) {
2424 		ZFS_EXIT(zfsvfs);
2425 		return (EPERM);
2426 	}
2427 
2428 	/*
2429 	 * Verify timestamps doesn't overflow 32 bits.
2430 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2431 	 * handle times greater than 2039.  This check should be removed
2432 	 * once large timestamps are fully supported.
2433 	 */
2434 	if (mask & (AT_ATIME | AT_MTIME)) {
2435 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2436 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2437 			ZFS_EXIT(zfsvfs);
2438 			return (EOVERFLOW);
2439 		}
2440 	}
2441 
2442 top:
2443 	attrzp = NULL;
2444 
2445 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2446 		ZFS_EXIT(zfsvfs);
2447 		return (EROFS);
2448 	}
2449 
2450 	/*
2451 	 * First validate permissions
2452 	 */
2453 
2454 	if (mask & AT_SIZE) {
2455 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2456 		if (err) {
2457 			ZFS_EXIT(zfsvfs);
2458 			return (err);
2459 		}
2460 		/*
2461 		 * XXX - Note, we are not providing any open
2462 		 * mode flags here (like FNDELAY), so we may
2463 		 * block if there are locks present... this
2464 		 * should be addressed in openat().
2465 		 */
2466 		/* XXX - would it be OK to generate a log record here? */
2467 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2468 		if (err) {
2469 			ZFS_EXIT(zfsvfs);
2470 			return (err);
2471 		}
2472 	}
2473 
2474 	if (mask & (AT_ATIME|AT_MTIME) ||
2475 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2476 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2477 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2478 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2479 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM))))
2480 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2481 		    skipaclchk, cr);
2482 
2483 	if (mask & (AT_UID|AT_GID)) {
2484 		int	idmask = (mask & (AT_UID|AT_GID));
2485 		int	take_owner;
2486 		int	take_group;
2487 
2488 		/*
2489 		 * NOTE: even if a new mode is being set,
2490 		 * we may clear S_ISUID/S_ISGID bits.
2491 		 */
2492 
2493 		if (!(mask & AT_MODE))
2494 			vap->va_mode = pzp->zp_mode;
2495 
2496 		/*
2497 		 * Take ownership or chgrp to group we are a member of
2498 		 */
2499 
2500 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2501 		take_group = (mask & AT_GID) &&
2502 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2503 
2504 		/*
2505 		 * If both AT_UID and AT_GID are set then take_owner and
2506 		 * take_group must both be set in order to allow taking
2507 		 * ownership.
2508 		 *
2509 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2510 		 *
2511 		 */
2512 
2513 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2514 		    ((idmask == AT_UID) && take_owner) ||
2515 		    ((idmask == AT_GID) && take_group)) {
2516 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2517 			    skipaclchk, cr) == 0) {
2518 				/*
2519 				 * Remove setuid/setgid for non-privileged users
2520 				 */
2521 				secpolicy_setid_clear(vap, cr);
2522 				trim_mask = (mask & (AT_UID|AT_GID));
2523 			} else {
2524 				need_policy =  TRUE;
2525 			}
2526 		} else {
2527 			need_policy =  TRUE;
2528 		}
2529 	}
2530 
2531 	mutex_enter(&zp->z_lock);
2532 	oldva.va_mode = pzp->zp_mode;
2533 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2534 	if (mask & AT_XVATTR) {
2535 		/*
2536 		 * Update xvattr mask to include only those attributes
2537 		 * that are actually changing.
2538 		 *
2539 		 * the bits will be restored prior to actually setting
2540 		 * the attributes so the caller thinks they were set.
2541 		 */
2542 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2543 			if (xoap->xoa_appendonly !=
2544 			    ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) {
2545 				need_policy = TRUE;
2546 			} else {
2547 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2548 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2549 			}
2550 		}
2551 
2552 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2553 			if (xoap->xoa_nounlink !=
2554 			    ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) {
2555 				need_policy = TRUE;
2556 			} else {
2557 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2558 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2559 			}
2560 		}
2561 
2562 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2563 			if (xoap->xoa_immutable !=
2564 			    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) {
2565 				need_policy = TRUE;
2566 			} else {
2567 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2568 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2569 			}
2570 		}
2571 
2572 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2573 			if (xoap->xoa_nodump !=
2574 			    ((pzp->zp_flags & ZFS_NODUMP) != 0)) {
2575 				need_policy = TRUE;
2576 			} else {
2577 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2578 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2579 			}
2580 		}
2581 
2582 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2583 			if (xoap->xoa_av_modified !=
2584 			    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) {
2585 				need_policy = TRUE;
2586 			} else {
2587 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2588 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2589 			}
2590 		}
2591 
2592 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2593 			if ((vp->v_type != VREG &&
2594 			    xoap->xoa_av_quarantined) ||
2595 			    xoap->xoa_av_quarantined !=
2596 			    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)) {
2597 				need_policy = TRUE;
2598 			} else {
2599 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2600 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2601 			}
2602 		}
2603 
2604 		if (need_policy == FALSE &&
2605 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2606 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2607 			need_policy = TRUE;
2608 		}
2609 	}
2610 
2611 	mutex_exit(&zp->z_lock);
2612 
2613 	if (mask & AT_MODE) {
2614 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2615 			err = secpolicy_setid_setsticky_clear(vp, vap,
2616 			    &oldva, cr);
2617 			if (err) {
2618 				ZFS_EXIT(zfsvfs);
2619 				return (err);
2620 			}
2621 			trim_mask |= AT_MODE;
2622 		} else {
2623 			need_policy = TRUE;
2624 		}
2625 	}
2626 
2627 	if (need_policy) {
2628 		/*
2629 		 * If trim_mask is set then take ownership
2630 		 * has been granted or write_acl is present and user
2631 		 * has the ability to modify mode.  In that case remove
2632 		 * UID|GID and or MODE from mask so that
2633 		 * secpolicy_vnode_setattr() doesn't revoke it.
2634 		 */
2635 
2636 		if (trim_mask) {
2637 			saved_mask = vap->va_mask;
2638 			vap->va_mask &= ~trim_mask;
2639 		}
2640 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2641 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2642 		if (err) {
2643 			ZFS_EXIT(zfsvfs);
2644 			return (err);
2645 		}
2646 
2647 		if (trim_mask)
2648 			vap->va_mask |= saved_mask;
2649 	}
2650 
2651 	/*
2652 	 * secpolicy_vnode_setattr, or take ownership may have
2653 	 * changed va_mask
2654 	 */
2655 	mask = vap->va_mask;
2656 
2657 	tx = dmu_tx_create(zfsvfs->z_os);
2658 	dmu_tx_hold_bonus(tx, zp->z_id);
2659 
2660 	if (mask & AT_MODE) {
2661 		uint64_t pmode = pzp->zp_mode;
2662 
2663 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2664 
2665 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
2666 			goto out;
2667 		if (pzp->zp_acl.z_acl_extern_obj) {
2668 			/* Are we upgrading ACL from old V0 format to new V1 */
2669 			if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2670 			    pzp->zp_acl.z_acl_version ==
2671 			    ZFS_ACL_VERSION_INITIAL) {
2672 				dmu_tx_hold_free(tx,
2673 				    pzp->zp_acl.z_acl_extern_obj, 0,
2674 				    DMU_OBJECT_END);
2675 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2676 				    0, aclp->z_acl_bytes);
2677 			} else {
2678 				dmu_tx_hold_write(tx,
2679 				    pzp->zp_acl.z_acl_extern_obj, 0,
2680 				    aclp->z_acl_bytes);
2681 			}
2682 		} else if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2683 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2684 			    0, aclp->z_acl_bytes);
2685 		}
2686 	}
2687 
2688 	if (mask & (AT_UID | AT_GID)) {
2689 		if (pzp->zp_xattr) {
2690 			err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp);
2691 			if (err)
2692 				goto out;
2693 			dmu_tx_hold_bonus(tx, attrzp->z_id);
2694 		}
2695 		if (mask & AT_UID) {
2696 			new_uid = zfs_fuid_create(zfsvfs,
2697 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2698 			if (new_uid != pzp->zp_uid &&
2699 			    zfs_usergroup_overquota(zfsvfs, B_FALSE, new_uid)) {
2700 				err = EDQUOT;
2701 				goto out;
2702 			}
2703 		}
2704 
2705 		if (mask & AT_GID) {
2706 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
2707 			    cr, ZFS_GROUP, &fuidp);
2708 			if (new_gid != pzp->zp_gid &&
2709 			    zfs_usergroup_overquota(zfsvfs, B_TRUE, new_gid)) {
2710 				err = EDQUOT;
2711 				goto out;
2712 			}
2713 		}
2714 		fuid_dirtied = zfsvfs->z_fuid_dirty;
2715 		if (fuid_dirtied) {
2716 			if (zfsvfs->z_fuid_obj == 0) {
2717 				dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
2718 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2719 				    FUID_SIZE_ESTIMATE(zfsvfs));
2720 				dmu_tx_hold_zap(tx, MASTER_NODE_OBJ,
2721 				    FALSE, NULL);
2722 			} else {
2723 				dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
2724 				dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
2725 				    FUID_SIZE_ESTIMATE(zfsvfs));
2726 			}
2727 		}
2728 	}
2729 
2730 	err = dmu_tx_assign(tx, TXG_NOWAIT);
2731 	if (err) {
2732 		if (err == ERESTART)
2733 			dmu_tx_wait(tx);
2734 		goto out;
2735 	}
2736 
2737 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2738 
2739 	/*
2740 	 * Set each attribute requested.
2741 	 * We group settings according to the locks they need to acquire.
2742 	 *
2743 	 * Note: you cannot set ctime directly, although it will be
2744 	 * updated as a side-effect of calling this function.
2745 	 */
2746 
2747 	mutex_enter(&zp->z_lock);
2748 
2749 	if (mask & AT_MODE) {
2750 		mutex_enter(&zp->z_acl_lock);
2751 		zp->z_phys->zp_mode = new_mode;
2752 		err = zfs_aclset_common(zp, aclp, cr, tx);
2753 		ASSERT3U(err, ==, 0);
2754 		mutex_exit(&zp->z_acl_lock);
2755 	}
2756 
2757 	if (attrzp)
2758 		mutex_enter(&attrzp->z_lock);
2759 
2760 	if (mask & AT_UID) {
2761 		pzp->zp_uid = new_uid;
2762 		if (attrzp)
2763 			attrzp->z_phys->zp_uid = new_uid;
2764 	}
2765 
2766 	if (mask & AT_GID) {
2767 		pzp->zp_gid = new_gid;
2768 		if (attrzp)
2769 			attrzp->z_phys->zp_gid = new_gid;
2770 	}
2771 
2772 	if (attrzp)
2773 		mutex_exit(&attrzp->z_lock);
2774 
2775 	if (mask & AT_ATIME)
2776 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2777 
2778 	if (mask & AT_MTIME)
2779 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2780 
2781 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2782 	if (mask & AT_SIZE)
2783 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2784 	else if (mask != 0)
2785 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2786 	/*
2787 	 * Do this after setting timestamps to prevent timestamp
2788 	 * update from toggling bit
2789 	 */
2790 
2791 	if (xoap && (mask & AT_XVATTR)) {
2792 
2793 		/*
2794 		 * restore trimmed off masks
2795 		 * so that return masks can be set for caller.
2796 		 */
2797 
2798 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
2799 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
2800 		}
2801 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
2802 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
2803 		}
2804 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
2805 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
2806 		}
2807 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
2808 			XVA_SET_REQ(xvap, XAT_NODUMP);
2809 		}
2810 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
2811 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
2812 		}
2813 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
2814 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
2815 		}
2816 
2817 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
2818 			size_t len;
2819 			dmu_object_info_t doi;
2820 
2821 			ASSERT(vp->v_type == VREG);
2822 
2823 			/* Grow the bonus buffer if necessary. */
2824 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2825 			len = sizeof (xoap->xoa_av_scanstamp) +
2826 			    sizeof (znode_phys_t);
2827 			if (len > doi.doi_bonus_size)
2828 				VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0);
2829 		}
2830 		zfs_xvattr_set(zp, xvap);
2831 	}
2832 
2833 	if (fuid_dirtied)
2834 		zfs_fuid_sync(zfsvfs, tx);
2835 
2836 	if (mask != 0)
2837 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2838 
2839 	mutex_exit(&zp->z_lock);
2840 
2841 out:
2842 	if (attrzp)
2843 		VN_RELE(ZTOV(attrzp));
2844 
2845 	if (aclp) {
2846 		zfs_acl_free(aclp);
2847 		aclp = NULL;
2848 	}
2849 
2850 	if (fuidp) {
2851 		zfs_fuid_info_free(fuidp);
2852 		fuidp = NULL;
2853 	}
2854 
2855 	if (err)
2856 		dmu_tx_abort(tx);
2857 	else
2858 		dmu_tx_commit(tx);
2859 
2860 	if (err == ERESTART)
2861 		goto top;
2862 
2863 	ZFS_EXIT(zfsvfs);
2864 	return (err);
2865 }
2866 
2867 typedef struct zfs_zlock {
2868 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2869 	znode_t		*zl_znode;	/* znode we held */
2870 	struct zfs_zlock *zl_next;	/* next in list */
2871 } zfs_zlock_t;
2872 
2873 /*
2874  * Drop locks and release vnodes that were held by zfs_rename_lock().
2875  */
2876 static void
2877 zfs_rename_unlock(zfs_zlock_t **zlpp)
2878 {
2879 	zfs_zlock_t *zl;
2880 
2881 	while ((zl = *zlpp) != NULL) {
2882 		if (zl->zl_znode != NULL)
2883 			VN_RELE(ZTOV(zl->zl_znode));
2884 		rw_exit(zl->zl_rwlock);
2885 		*zlpp = zl->zl_next;
2886 		kmem_free(zl, sizeof (*zl));
2887 	}
2888 }
2889 
2890 /*
2891  * Search back through the directory tree, using the ".." entries.
2892  * Lock each directory in the chain to prevent concurrent renames.
2893  * Fail any attempt to move a directory into one of its own descendants.
2894  * XXX - z_parent_lock can overlap with map or grow locks
2895  */
2896 static int
2897 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2898 {
2899 	zfs_zlock_t	*zl;
2900 	znode_t		*zp = tdzp;
2901 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2902 	uint64_t	*oidp = &zp->z_id;
2903 	krwlock_t	*rwlp = &szp->z_parent_lock;
2904 	krw_t		rw = RW_WRITER;
2905 
2906 	/*
2907 	 * First pass write-locks szp and compares to zp->z_id.
2908 	 * Later passes read-lock zp and compare to zp->z_parent.
2909 	 */
2910 	do {
2911 		if (!rw_tryenter(rwlp, rw)) {
2912 			/*
2913 			 * Another thread is renaming in this path.
2914 			 * Note that if we are a WRITER, we don't have any
2915 			 * parent_locks held yet.
2916 			 */
2917 			if (rw == RW_READER && zp->z_id > szp->z_id) {
2918 				/*
2919 				 * Drop our locks and restart
2920 				 */
2921 				zfs_rename_unlock(&zl);
2922 				*zlpp = NULL;
2923 				zp = tdzp;
2924 				oidp = &zp->z_id;
2925 				rwlp = &szp->z_parent_lock;
2926 				rw = RW_WRITER;
2927 				continue;
2928 			} else {
2929 				/*
2930 				 * Wait for other thread to drop its locks
2931 				 */
2932 				rw_enter(rwlp, rw);
2933 			}
2934 		}
2935 
2936 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2937 		zl->zl_rwlock = rwlp;
2938 		zl->zl_znode = NULL;
2939 		zl->zl_next = *zlpp;
2940 		*zlpp = zl;
2941 
2942 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2943 			return (EINVAL);
2944 
2945 		if (*oidp == rootid)		/* We've hit the top */
2946 			return (0);
2947 
2948 		if (rw == RW_READER) {		/* i.e. not the first pass */
2949 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2950 			if (error)
2951 				return (error);
2952 			zl->zl_znode = zp;
2953 		}
2954 		oidp = &zp->z_phys->zp_parent;
2955 		rwlp = &zp->z_parent_lock;
2956 		rw = RW_READER;
2957 
2958 	} while (zp->z_id != sdzp->z_id);
2959 
2960 	return (0);
2961 }
2962 
2963 /*
2964  * Move an entry from the provided source directory to the target
2965  * directory.  Change the entry name as indicated.
2966  *
2967  *	IN:	sdvp	- Source directory containing the "old entry".
2968  *		snm	- Old entry name.
2969  *		tdvp	- Target directory to contain the "new entry".
2970  *		tnm	- New entry name.
2971  *		cr	- credentials of caller.
2972  *		ct	- caller context
2973  *		flags	- case flags
2974  *
2975  *	RETURN:	0 if success
2976  *		error code if failure
2977  *
2978  * Timestamps:
2979  *	sdvp,tdvp - ctime|mtime updated
2980  */
2981 /*ARGSUSED*/
2982 static int
2983 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
2984     caller_context_t *ct, int flags)
2985 {
2986 	znode_t		*tdzp, *szp, *tzp;
2987 	znode_t		*sdzp = VTOZ(sdvp);
2988 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2989 	zilog_t		*zilog;
2990 	vnode_t		*realvp;
2991 	zfs_dirlock_t	*sdl, *tdl;
2992 	dmu_tx_t	*tx;
2993 	zfs_zlock_t	*zl;
2994 	int		cmp, serr, terr;
2995 	int		error = 0;
2996 	int		zflg = 0;
2997 
2998 	ZFS_ENTER(zfsvfs);
2999 	ZFS_VERIFY_ZP(sdzp);
3000 	zilog = zfsvfs->z_log;
3001 
3002 	/*
3003 	 * Make sure we have the real vp for the target directory.
3004 	 */
3005 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3006 		tdvp = realvp;
3007 
3008 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
3009 		ZFS_EXIT(zfsvfs);
3010 		return (EXDEV);
3011 	}
3012 
3013 	tdzp = VTOZ(tdvp);
3014 	ZFS_VERIFY_ZP(tdzp);
3015 	if (zfsvfs->z_utf8 && u8_validate(tnm,
3016 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3017 		ZFS_EXIT(zfsvfs);
3018 		return (EILSEQ);
3019 	}
3020 
3021 	if (flags & FIGNORECASE)
3022 		zflg |= ZCILOOK;
3023 
3024 top:
3025 	szp = NULL;
3026 	tzp = NULL;
3027 	zl = NULL;
3028 
3029 	/*
3030 	 * This is to prevent the creation of links into attribute space
3031 	 * by renaming a linked file into/outof an attribute directory.
3032 	 * See the comment in zfs_link() for why this is considered bad.
3033 	 */
3034 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
3035 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
3036 		ZFS_EXIT(zfsvfs);
3037 		return (EINVAL);
3038 	}
3039 
3040 	/*
3041 	 * Lock source and target directory entries.  To prevent deadlock,
3042 	 * a lock ordering must be defined.  We lock the directory with
3043 	 * the smallest object id first, or if it's a tie, the one with
3044 	 * the lexically first name.
3045 	 */
3046 	if (sdzp->z_id < tdzp->z_id) {
3047 		cmp = -1;
3048 	} else if (sdzp->z_id > tdzp->z_id) {
3049 		cmp = 1;
3050 	} else {
3051 		/*
3052 		 * First compare the two name arguments without
3053 		 * considering any case folding.
3054 		 */
3055 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3056 
3057 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3058 		ASSERT(error == 0 || !zfsvfs->z_utf8);
3059 		if (cmp == 0) {
3060 			/*
3061 			 * POSIX: "If the old argument and the new argument
3062 			 * both refer to links to the same existing file,
3063 			 * the rename() function shall return successfully
3064 			 * and perform no other action."
3065 			 */
3066 			ZFS_EXIT(zfsvfs);
3067 			return (0);
3068 		}
3069 		/*
3070 		 * If the file system is case-folding, then we may
3071 		 * have some more checking to do.  A case-folding file
3072 		 * system is either supporting mixed case sensitivity
3073 		 * access or is completely case-insensitive.  Note
3074 		 * that the file system is always case preserving.
3075 		 *
3076 		 * In mixed sensitivity mode case sensitive behavior
3077 		 * is the default.  FIGNORECASE must be used to
3078 		 * explicitly request case insensitive behavior.
3079 		 *
3080 		 * If the source and target names provided differ only
3081 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
3082 		 * we will treat this as a special case in the
3083 		 * case-insensitive mode: as long as the source name
3084 		 * is an exact match, we will allow this to proceed as
3085 		 * a name-change request.
3086 		 */
3087 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3088 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
3089 		    flags & FIGNORECASE)) &&
3090 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3091 		    &error) == 0) {
3092 			/*
3093 			 * case preserving rename request, require exact
3094 			 * name matches
3095 			 */
3096 			zflg |= ZCIEXACT;
3097 			zflg &= ~ZCILOOK;
3098 		}
3099 	}
3100 
3101 	if (cmp < 0) {
3102 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3103 		    ZEXISTS | zflg, NULL, NULL);
3104 		terr = zfs_dirent_lock(&tdl,
3105 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3106 	} else {
3107 		terr = zfs_dirent_lock(&tdl,
3108 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3109 		serr = zfs_dirent_lock(&sdl,
3110 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3111 		    NULL, NULL);
3112 	}
3113 
3114 	if (serr) {
3115 		/*
3116 		 * Source entry invalid or not there.
3117 		 */
3118 		if (!terr) {
3119 			zfs_dirent_unlock(tdl);
3120 			if (tzp)
3121 				VN_RELE(ZTOV(tzp));
3122 		}
3123 		if (strcmp(snm, "..") == 0)
3124 			serr = EINVAL;
3125 		ZFS_EXIT(zfsvfs);
3126 		return (serr);
3127 	}
3128 	if (terr) {
3129 		zfs_dirent_unlock(sdl);
3130 		VN_RELE(ZTOV(szp));
3131 		if (strcmp(tnm, "..") == 0)
3132 			terr = EINVAL;
3133 		ZFS_EXIT(zfsvfs);
3134 		return (terr);
3135 	}
3136 
3137 	/*
3138 	 * Must have write access at the source to remove the old entry
3139 	 * and write access at the target to create the new entry.
3140 	 * Note that if target and source are the same, this can be
3141 	 * done in a single check.
3142 	 */
3143 
3144 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3145 		goto out;
3146 
3147 	if (ZTOV(szp)->v_type == VDIR) {
3148 		/*
3149 		 * Check to make sure rename is valid.
3150 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3151 		 */
3152 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3153 			goto out;
3154 	}
3155 
3156 	/*
3157 	 * Does target exist?
3158 	 */
3159 	if (tzp) {
3160 		/*
3161 		 * Source and target must be the same type.
3162 		 */
3163 		if (ZTOV(szp)->v_type == VDIR) {
3164 			if (ZTOV(tzp)->v_type != VDIR) {
3165 				error = ENOTDIR;
3166 				goto out;
3167 			}
3168 		} else {
3169 			if (ZTOV(tzp)->v_type == VDIR) {
3170 				error = EISDIR;
3171 				goto out;
3172 			}
3173 		}
3174 		/*
3175 		 * POSIX dictates that when the source and target
3176 		 * entries refer to the same file object, rename
3177 		 * must do nothing and exit without error.
3178 		 */
3179 		if (szp->z_id == tzp->z_id) {
3180 			error = 0;
3181 			goto out;
3182 		}
3183 	}
3184 
3185 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3186 	if (tzp)
3187 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3188 
3189 	/*
3190 	 * notify the target directory if it is not the same
3191 	 * as source directory.
3192 	 */
3193 	if (tdvp != sdvp) {
3194 		vnevent_rename_dest_dir(tdvp, ct);
3195 	}
3196 
3197 	tx = dmu_tx_create(zfsvfs->z_os);
3198 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
3199 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
3200 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3201 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3202 	if (sdzp != tdzp)
3203 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
3204 	if (tzp)
3205 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* parent changes */
3206 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3207 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3208 	if (error) {
3209 		if (zl != NULL)
3210 			zfs_rename_unlock(&zl);
3211 		zfs_dirent_unlock(sdl);
3212 		zfs_dirent_unlock(tdl);
3213 		VN_RELE(ZTOV(szp));
3214 		if (tzp)
3215 			VN_RELE(ZTOV(tzp));
3216 		if (error == ERESTART) {
3217 			dmu_tx_wait(tx);
3218 			dmu_tx_abort(tx);
3219 			goto top;
3220 		}
3221 		dmu_tx_abort(tx);
3222 		ZFS_EXIT(zfsvfs);
3223 		return (error);
3224 	}
3225 
3226 	if (tzp)	/* Attempt to remove the existing target */
3227 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3228 
3229 	if (error == 0) {
3230 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3231 		if (error == 0) {
3232 			szp->z_phys->zp_flags |= ZFS_AV_MODIFIED;
3233 
3234 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3235 			ASSERT(error == 0);
3236 
3237 			zfs_log_rename(zilog, tx,
3238 			    TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0),
3239 			    sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp);
3240 
3241 			/* Update path information for the target vnode */
3242 			vn_renamepath(tdvp, ZTOV(szp), tnm, strlen(tnm));
3243 		}
3244 	}
3245 
3246 	dmu_tx_commit(tx);
3247 out:
3248 	if (zl != NULL)
3249 		zfs_rename_unlock(&zl);
3250 
3251 	zfs_dirent_unlock(sdl);
3252 	zfs_dirent_unlock(tdl);
3253 
3254 	VN_RELE(ZTOV(szp));
3255 	if (tzp)
3256 		VN_RELE(ZTOV(tzp));
3257 
3258 	ZFS_EXIT(zfsvfs);
3259 	return (error);
3260 }
3261 
3262 /*
3263  * Insert the indicated symbolic reference entry into the directory.
3264  *
3265  *	IN:	dvp	- Directory to contain new symbolic link.
3266  *		link	- Name for new symlink entry.
3267  *		vap	- Attributes of new entry.
3268  *		target	- Target path of new symlink.
3269  *		cr	- credentials of caller.
3270  *		ct	- caller context
3271  *		flags	- case flags
3272  *
3273  *	RETURN:	0 if success
3274  *		error code if failure
3275  *
3276  * Timestamps:
3277  *	dvp - ctime|mtime updated
3278  */
3279 /*ARGSUSED*/
3280 static int
3281 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3282     caller_context_t *ct, int flags)
3283 {
3284 	znode_t		*zp, *dzp = VTOZ(dvp);
3285 	zfs_dirlock_t	*dl;
3286 	dmu_tx_t	*tx;
3287 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3288 	zilog_t		*zilog;
3289 	int		len = strlen(link);
3290 	int		error;
3291 	int		zflg = ZNEW;
3292 	zfs_acl_ids_t	acl_ids;
3293 	boolean_t	fuid_dirtied;
3294 
3295 	ASSERT(vap->va_type == VLNK);
3296 
3297 	ZFS_ENTER(zfsvfs);
3298 	ZFS_VERIFY_ZP(dzp);
3299 	zilog = zfsvfs->z_log;
3300 
3301 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3302 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3303 		ZFS_EXIT(zfsvfs);
3304 		return (EILSEQ);
3305 	}
3306 	if (flags & FIGNORECASE)
3307 		zflg |= ZCILOOK;
3308 top:
3309 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3310 		ZFS_EXIT(zfsvfs);
3311 		return (error);
3312 	}
3313 
3314 	if (len > MAXPATHLEN) {
3315 		ZFS_EXIT(zfsvfs);
3316 		return (ENAMETOOLONG);
3317 	}
3318 
3319 	/*
3320 	 * Attempt to lock directory; fail if entry already exists.
3321 	 */
3322 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3323 	if (error) {
3324 		ZFS_EXIT(zfsvfs);
3325 		return (error);
3326 	}
3327 
3328 	VERIFY(0 == zfs_acl_ids_create(dzp, 0, vap, cr, NULL, &acl_ids));
3329 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3330 		zfs_acl_ids_free(&acl_ids);
3331 		zfs_dirent_unlock(dl);
3332 		ZFS_EXIT(zfsvfs);
3333 		return (EDQUOT);
3334 	}
3335 	tx = dmu_tx_create(zfsvfs->z_os);
3336 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3337 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3338 	dmu_tx_hold_bonus(tx, dzp->z_id);
3339 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3340 	if (acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE)
3341 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
3342 	if (fuid_dirtied)
3343 		zfs_fuid_txhold(zfsvfs, tx);
3344 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3345 	if (error) {
3346 		zfs_acl_ids_free(&acl_ids);
3347 		zfs_dirent_unlock(dl);
3348 		if (error == ERESTART) {
3349 			dmu_tx_wait(tx);
3350 			dmu_tx_abort(tx);
3351 			goto top;
3352 		}
3353 		dmu_tx_abort(tx);
3354 		ZFS_EXIT(zfsvfs);
3355 		return (error);
3356 	}
3357 
3358 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
3359 
3360 	/*
3361 	 * Create a new object for the symlink.
3362 	 * Put the link content into bonus buffer if it will fit;
3363 	 * otherwise, store it just like any other file data.
3364 	 */
3365 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
3366 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, len, &acl_ids);
3367 		if (len != 0)
3368 			bcopy(link, zp->z_phys + 1, len);
3369 	} else {
3370 		dmu_buf_t *dbp;
3371 
3372 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, &acl_ids);
3373 
3374 		if (fuid_dirtied)
3375 			zfs_fuid_sync(zfsvfs, tx);
3376 		/*
3377 		 * Nothing can access the znode yet so no locking needed
3378 		 * for growing the znode's blocksize.
3379 		 */
3380 		zfs_grow_blocksize(zp, len, tx);
3381 
3382 		VERIFY(0 == dmu_buf_hold(zfsvfs->z_os,
3383 		    zp->z_id, 0, FTAG, &dbp));
3384 		dmu_buf_will_dirty(dbp, tx);
3385 
3386 		ASSERT3U(len, <=, dbp->db_size);
3387 		bcopy(link, dbp->db_data, len);
3388 		dmu_buf_rele(dbp, FTAG);
3389 	}
3390 	zp->z_phys->zp_size = len;
3391 
3392 	/*
3393 	 * Insert the new object into the directory.
3394 	 */
3395 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3396 	if (error == 0) {
3397 		uint64_t txtype = TX_SYMLINK;
3398 		if (flags & FIGNORECASE)
3399 			txtype |= TX_CI;
3400 		zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3401 	}
3402 
3403 	zfs_acl_ids_free(&acl_ids);
3404 
3405 	dmu_tx_commit(tx);
3406 
3407 	zfs_dirent_unlock(dl);
3408 
3409 	VN_RELE(ZTOV(zp));
3410 
3411 	ZFS_EXIT(zfsvfs);
3412 	return (error);
3413 }
3414 
3415 /*
3416  * Return, in the buffer contained in the provided uio structure,
3417  * the symbolic path referred to by vp.
3418  *
3419  *	IN:	vp	- vnode of symbolic link.
3420  *		uoip	- structure to contain the link path.
3421  *		cr	- credentials of caller.
3422  *		ct	- caller context
3423  *
3424  *	OUT:	uio	- structure to contain the link path.
3425  *
3426  *	RETURN:	0 if success
3427  *		error code if failure
3428  *
3429  * Timestamps:
3430  *	vp - atime updated
3431  */
3432 /* ARGSUSED */
3433 static int
3434 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3435 {
3436 	znode_t		*zp = VTOZ(vp);
3437 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3438 	size_t		bufsz;
3439 	int		error;
3440 
3441 	ZFS_ENTER(zfsvfs);
3442 	ZFS_VERIFY_ZP(zp);
3443 
3444 	bufsz = (size_t)zp->z_phys->zp_size;
3445 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
3446 		error = uiomove(zp->z_phys + 1,
3447 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3448 	} else {
3449 		dmu_buf_t *dbp;
3450 		error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp);
3451 		if (error) {
3452 			ZFS_EXIT(zfsvfs);
3453 			return (error);
3454 		}
3455 		error = uiomove(dbp->db_data,
3456 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3457 		dmu_buf_rele(dbp, FTAG);
3458 	}
3459 
3460 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3461 	ZFS_EXIT(zfsvfs);
3462 	return (error);
3463 }
3464 
3465 /*
3466  * Insert a new entry into directory tdvp referencing svp.
3467  *
3468  *	IN:	tdvp	- Directory to contain new entry.
3469  *		svp	- vnode of new entry.
3470  *		name	- name of new entry.
3471  *		cr	- credentials of caller.
3472  *		ct	- caller context
3473  *
3474  *	RETURN:	0 if success
3475  *		error code if failure
3476  *
3477  * Timestamps:
3478  *	tdvp - ctime|mtime updated
3479  *	 svp - ctime updated
3480  */
3481 /* ARGSUSED */
3482 static int
3483 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3484     caller_context_t *ct, int flags)
3485 {
3486 	znode_t		*dzp = VTOZ(tdvp);
3487 	znode_t		*tzp, *szp;
3488 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3489 	zilog_t		*zilog;
3490 	zfs_dirlock_t	*dl;
3491 	dmu_tx_t	*tx;
3492 	vnode_t		*realvp;
3493 	int		error;
3494 	int		zf = ZNEW;
3495 	uid_t		owner;
3496 
3497 	ASSERT(tdvp->v_type == VDIR);
3498 
3499 	ZFS_ENTER(zfsvfs);
3500 	ZFS_VERIFY_ZP(dzp);
3501 	zilog = zfsvfs->z_log;
3502 
3503 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3504 		svp = realvp;
3505 
3506 	if (svp->v_vfsp != tdvp->v_vfsp) {
3507 		ZFS_EXIT(zfsvfs);
3508 		return (EXDEV);
3509 	}
3510 	szp = VTOZ(svp);
3511 	ZFS_VERIFY_ZP(szp);
3512 
3513 	if (zfsvfs->z_utf8 && u8_validate(name,
3514 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3515 		ZFS_EXIT(zfsvfs);
3516 		return (EILSEQ);
3517 	}
3518 	if (flags & FIGNORECASE)
3519 		zf |= ZCILOOK;
3520 
3521 top:
3522 	/*
3523 	 * We do not support links between attributes and non-attributes
3524 	 * because of the potential security risk of creating links
3525 	 * into "normal" file space in order to circumvent restrictions
3526 	 * imposed in attribute space.
3527 	 */
3528 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
3529 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
3530 		ZFS_EXIT(zfsvfs);
3531 		return (EINVAL);
3532 	}
3533 
3534 	/*
3535 	 * POSIX dictates that we return EPERM here.
3536 	 * Better choices include ENOTSUP or EISDIR.
3537 	 */
3538 	if (svp->v_type == VDIR) {
3539 		ZFS_EXIT(zfsvfs);
3540 		return (EPERM);
3541 	}
3542 
3543 	owner = zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, cr, ZFS_OWNER);
3544 	if (owner != crgetuid(cr) &&
3545 	    secpolicy_basic_link(cr) != 0) {
3546 		ZFS_EXIT(zfsvfs);
3547 		return (EPERM);
3548 	}
3549 
3550 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3551 		ZFS_EXIT(zfsvfs);
3552 		return (error);
3553 	}
3554 
3555 	/*
3556 	 * Attempt to lock directory; fail if entry already exists.
3557 	 */
3558 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3559 	if (error) {
3560 		ZFS_EXIT(zfsvfs);
3561 		return (error);
3562 	}
3563 
3564 	tx = dmu_tx_create(zfsvfs->z_os);
3565 	dmu_tx_hold_bonus(tx, szp->z_id);
3566 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3567 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3568 	if (error) {
3569 		zfs_dirent_unlock(dl);
3570 		if (error == ERESTART) {
3571 			dmu_tx_wait(tx);
3572 			dmu_tx_abort(tx);
3573 			goto top;
3574 		}
3575 		dmu_tx_abort(tx);
3576 		ZFS_EXIT(zfsvfs);
3577 		return (error);
3578 	}
3579 
3580 	error = zfs_link_create(dl, szp, tx, 0);
3581 
3582 	if (error == 0) {
3583 		uint64_t txtype = TX_LINK;
3584 		if (flags & FIGNORECASE)
3585 			txtype |= TX_CI;
3586 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3587 	}
3588 
3589 	dmu_tx_commit(tx);
3590 
3591 	zfs_dirent_unlock(dl);
3592 
3593 	if (error == 0) {
3594 		vnevent_link(svp, ct);
3595 	}
3596 
3597 	ZFS_EXIT(zfsvfs);
3598 	return (error);
3599 }
3600 
3601 /*
3602  * zfs_null_putapage() is used when the file system has been force
3603  * unmounted. It just drops the pages.
3604  */
3605 /* ARGSUSED */
3606 static int
3607 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3608 		size_t *lenp, int flags, cred_t *cr)
3609 {
3610 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
3611 	return (0);
3612 }
3613 
3614 /*
3615  * Push a page out to disk, klustering if possible.
3616  *
3617  *	IN:	vp	- file to push page to.
3618  *		pp	- page to push.
3619  *		flags	- additional flags.
3620  *		cr	- credentials of caller.
3621  *
3622  *	OUT:	offp	- start of range pushed.
3623  *		lenp	- len of range pushed.
3624  *
3625  *	RETURN:	0 if success
3626  *		error code if failure
3627  *
3628  * NOTE: callers must have locked the page to be pushed.  On
3629  * exit, the page (and all other pages in the kluster) must be
3630  * unlocked.
3631  */
3632 /* ARGSUSED */
3633 static int
3634 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3635 		size_t *lenp, int flags, cred_t *cr)
3636 {
3637 	znode_t		*zp = VTOZ(vp);
3638 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3639 	dmu_tx_t	*tx;
3640 	u_offset_t	off, koff;
3641 	size_t		len, klen;
3642 	uint64_t	filesz;
3643 	int		err;
3644 
3645 	filesz = zp->z_phys->zp_size;
3646 	off = pp->p_offset;
3647 	len = PAGESIZE;
3648 	/*
3649 	 * If our blocksize is bigger than the page size, try to kluster
3650 	 * multiple pages so that we write a full block (thus avoiding
3651 	 * a read-modify-write).
3652 	 */
3653 	if (off < filesz && zp->z_blksz > PAGESIZE) {
3654 		klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3655 		koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
3656 		ASSERT(koff <= filesz);
3657 		if (koff + klen > filesz)
3658 			klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE);
3659 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
3660 	}
3661 	ASSERT3U(btop(len), ==, btopr(len));
3662 
3663 	/*
3664 	 * Can't push pages past end-of-file.
3665 	 */
3666 	if (off >= filesz) {
3667 		/* ignore all pages */
3668 		err = 0;
3669 		goto out;
3670 	} else if (off + len > filesz) {
3671 		int npages = btopr(filesz - off);
3672 		page_t *trunc;
3673 
3674 		page_list_break(&pp, &trunc, npages);
3675 		/* ignore pages past end of file */
3676 		if (trunc)
3677 			pvn_write_done(trunc, flags);
3678 		len = filesz - off;
3679 	}
3680 
3681 	if (zfs_usergroup_overquota(zfsvfs, B_FALSE, zp->z_phys->zp_uid) ||
3682 	    zfs_usergroup_overquota(zfsvfs, B_TRUE, zp->z_phys->zp_gid)) {
3683 		err = EDQUOT;
3684 		goto out;
3685 	}
3686 top:
3687 	tx = dmu_tx_create(zfsvfs->z_os);
3688 	dmu_tx_hold_write(tx, zp->z_id, off, len);
3689 	dmu_tx_hold_bonus(tx, zp->z_id);
3690 	err = dmu_tx_assign(tx, TXG_NOWAIT);
3691 	if (err != 0) {
3692 		if (err == ERESTART) {
3693 			dmu_tx_wait(tx);
3694 			dmu_tx_abort(tx);
3695 			goto top;
3696 		}
3697 		dmu_tx_abort(tx);
3698 		goto out;
3699 	}
3700 
3701 	if (zp->z_blksz <= PAGESIZE) {
3702 		caddr_t va = zfs_map_page(pp, S_READ);
3703 		ASSERT3U(len, <=, PAGESIZE);
3704 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
3705 		zfs_unmap_page(pp, va);
3706 	} else {
3707 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
3708 	}
3709 
3710 	if (err == 0) {
3711 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
3712 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
3713 		dmu_tx_commit(tx);
3714 	}
3715 
3716 out:
3717 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
3718 	if (offp)
3719 		*offp = off;
3720 	if (lenp)
3721 		*lenp = len;
3722 
3723 	return (err);
3724 }
3725 
3726 /*
3727  * Copy the portion of the file indicated from pages into the file.
3728  * The pages are stored in a page list attached to the files vnode.
3729  *
3730  *	IN:	vp	- vnode of file to push page data to.
3731  *		off	- position in file to put data.
3732  *		len	- amount of data to write.
3733  *		flags	- flags to control the operation.
3734  *		cr	- credentials of caller.
3735  *		ct	- caller context.
3736  *
3737  *	RETURN:	0 if success
3738  *		error code if failure
3739  *
3740  * Timestamps:
3741  *	vp - ctime|mtime updated
3742  */
3743 /*ARGSUSED*/
3744 static int
3745 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
3746     caller_context_t *ct)
3747 {
3748 	znode_t		*zp = VTOZ(vp);
3749 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3750 	page_t		*pp;
3751 	size_t		io_len;
3752 	u_offset_t	io_off;
3753 	uint_t		blksz;
3754 	rl_t		*rl;
3755 	int		error = 0;
3756 
3757 	ZFS_ENTER(zfsvfs);
3758 	ZFS_VERIFY_ZP(zp);
3759 
3760 	/*
3761 	 * Align this request to the file block size in case we kluster.
3762 	 * XXX - this can result in pretty aggresive locking, which can
3763 	 * impact simultanious read/write access.  One option might be
3764 	 * to break up long requests (len == 0) into block-by-block
3765 	 * operations to get narrower locking.
3766 	 */
3767 	blksz = zp->z_blksz;
3768 	if (ISP2(blksz))
3769 		io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
3770 	else
3771 		io_off = 0;
3772 	if (len > 0 && ISP2(blksz))
3773 		io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
3774 	else
3775 		io_len = 0;
3776 
3777 	if (io_len == 0) {
3778 		/*
3779 		 * Search the entire vp list for pages >= io_off.
3780 		 */
3781 		rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
3782 		error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
3783 		goto out;
3784 	}
3785 	rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
3786 
3787 	if (off > zp->z_phys->zp_size) {
3788 		/* past end of file */
3789 		zfs_range_unlock(rl);
3790 		ZFS_EXIT(zfsvfs);
3791 		return (0);
3792 	}
3793 
3794 	len = MIN(io_len, P2ROUNDUP(zp->z_phys->zp_size, PAGESIZE) - io_off);
3795 
3796 	for (off = io_off; io_off < off + len; io_off += io_len) {
3797 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
3798 			pp = page_lookup(vp, io_off,
3799 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
3800 		} else {
3801 			pp = page_lookup_nowait(vp, io_off,
3802 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
3803 		}
3804 
3805 		if (pp != NULL && pvn_getdirty(pp, flags)) {
3806 			int err;
3807 
3808 			/*
3809 			 * Found a dirty page to push
3810 			 */
3811 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
3812 			if (err)
3813 				error = err;
3814 		} else {
3815 			io_len = PAGESIZE;
3816 		}
3817 	}
3818 out:
3819 	zfs_range_unlock(rl);
3820 	if ((flags & B_ASYNC) == 0)
3821 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
3822 	ZFS_EXIT(zfsvfs);
3823 	return (error);
3824 }
3825 
3826 /*ARGSUSED*/
3827 void
3828 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
3829 {
3830 	znode_t	*zp = VTOZ(vp);
3831 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3832 	int error;
3833 
3834 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
3835 	if (zp->z_dbuf == NULL) {
3836 		/*
3837 		 * The fs has been unmounted, or we did a
3838 		 * suspend/resume and this file no longer exists.
3839 		 */
3840 		if (vn_has_cached_data(vp)) {
3841 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
3842 			    B_INVAL, cr);
3843 		}
3844 
3845 		mutex_enter(&zp->z_lock);
3846 		vp->v_count = 0; /* count arrives as 1 */
3847 		mutex_exit(&zp->z_lock);
3848 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
3849 		zfs_znode_free(zp);
3850 		return;
3851 	}
3852 
3853 	/*
3854 	 * Attempt to push any data in the page cache.  If this fails
3855 	 * we will get kicked out later in zfs_zinactive().
3856 	 */
3857 	if (vn_has_cached_data(vp)) {
3858 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
3859 		    cr);
3860 	}
3861 
3862 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3863 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3864 
3865 		dmu_tx_hold_bonus(tx, zp->z_id);
3866 		error = dmu_tx_assign(tx, TXG_WAIT);
3867 		if (error) {
3868 			dmu_tx_abort(tx);
3869 		} else {
3870 			dmu_buf_will_dirty(zp->z_dbuf, tx);
3871 			mutex_enter(&zp->z_lock);
3872 			zp->z_atime_dirty = 0;
3873 			mutex_exit(&zp->z_lock);
3874 			dmu_tx_commit(tx);
3875 		}
3876 	}
3877 
3878 	zfs_zinactive(zp);
3879 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
3880 }
3881 
3882 /*
3883  * Bounds-check the seek operation.
3884  *
3885  *	IN:	vp	- vnode seeking within
3886  *		ooff	- old file offset
3887  *		noffp	- pointer to new file offset
3888  *		ct	- caller context
3889  *
3890  *	RETURN:	0 if success
3891  *		EINVAL if new offset invalid
3892  */
3893 /* ARGSUSED */
3894 static int
3895 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
3896     caller_context_t *ct)
3897 {
3898 	if (vp->v_type == VDIR)
3899 		return (0);
3900 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3901 }
3902 
3903 /*
3904  * Pre-filter the generic locking function to trap attempts to place
3905  * a mandatory lock on a memory mapped file.
3906  */
3907 static int
3908 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3909     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
3910 {
3911 	znode_t *zp = VTOZ(vp);
3912 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3913 	int error;
3914 
3915 	ZFS_ENTER(zfsvfs);
3916 	ZFS_VERIFY_ZP(zp);
3917 
3918 	/*
3919 	 * We are following the UFS semantics with respect to mapcnt
3920 	 * here: If we see that the file is mapped already, then we will
3921 	 * return an error, but we don't worry about races between this
3922 	 * function and zfs_map().
3923 	 */
3924 	if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3925 		ZFS_EXIT(zfsvfs);
3926 		return (EAGAIN);
3927 	}
3928 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct);
3929 	ZFS_EXIT(zfsvfs);
3930 	return (error);
3931 }
3932 
3933 /*
3934  * If we can't find a page in the cache, we will create a new page
3935  * and fill it with file data.  For efficiency, we may try to fill
3936  * multiple pages at once (klustering) to fill up the supplied page
3937  * list.  Note that the pages to be filled are held with an exclusive
3938  * lock to prevent access by other threads while they are being filled.
3939  */
3940 static int
3941 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3942     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3943 {
3944 	znode_t *zp = VTOZ(vp);
3945 	page_t *pp, *cur_pp;
3946 	objset_t *os = zp->z_zfsvfs->z_os;
3947 	u_offset_t io_off, total;
3948 	size_t io_len;
3949 	int err;
3950 
3951 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3952 		/*
3953 		 * We only have a single page, don't bother klustering
3954 		 */
3955 		io_off = off;
3956 		io_len = PAGESIZE;
3957 		pp = page_create_va(vp, io_off, io_len,
3958 		    PG_EXCL | PG_WAIT, seg, addr);
3959 	} else {
3960 		/*
3961 		 * Try to find enough pages to fill the page list
3962 		 */
3963 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3964 		    &io_len, off, plsz, 0);
3965 	}
3966 	if (pp == NULL) {
3967 		/*
3968 		 * The page already exists, nothing to do here.
3969 		 */
3970 		*pl = NULL;
3971 		return (0);
3972 	}
3973 
3974 	/*
3975 	 * Fill the pages in the kluster.
3976 	 */
3977 	cur_pp = pp;
3978 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3979 		caddr_t va;
3980 
3981 		ASSERT3U(io_off, ==, cur_pp->p_offset);
3982 		va = zfs_map_page(cur_pp, S_WRITE);
3983 		err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
3984 		    DMU_READ_PREFETCH);
3985 		zfs_unmap_page(cur_pp, va);
3986 		if (err) {
3987 			/* On error, toss the entire kluster */
3988 			pvn_read_done(pp, B_ERROR);
3989 			/* convert checksum errors into IO errors */
3990 			if (err == ECKSUM)
3991 				err = EIO;
3992 			return (err);
3993 		}
3994 		cur_pp = cur_pp->p_next;
3995 	}
3996 
3997 	/*
3998 	 * Fill in the page list array from the kluster starting
3999 	 * from the desired offset `off'.
4000 	 * NOTE: the page list will always be null terminated.
4001 	 */
4002 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4003 	ASSERT(pl == NULL || (*pl)->p_offset == off);
4004 
4005 	return (0);
4006 }
4007 
4008 /*
4009  * Return pointers to the pages for the file region [off, off + len]
4010  * in the pl array.  If plsz is greater than len, this function may
4011  * also return page pointers from after the specified region
4012  * (i.e. the region [off, off + plsz]).  These additional pages are
4013  * only returned if they are already in the cache, or were created as
4014  * part of a klustered read.
4015  *
4016  *	IN:	vp	- vnode of file to get data from.
4017  *		off	- position in file to get data from.
4018  *		len	- amount of data to retrieve.
4019  *		plsz	- length of provided page list.
4020  *		seg	- segment to obtain pages for.
4021  *		addr	- virtual address of fault.
4022  *		rw	- mode of created pages.
4023  *		cr	- credentials of caller.
4024  *		ct	- caller context.
4025  *
4026  *	OUT:	protp	- protection mode of created pages.
4027  *		pl	- list of pages created.
4028  *
4029  *	RETURN:	0 if success
4030  *		error code if failure
4031  *
4032  * Timestamps:
4033  *	vp - atime updated
4034  */
4035 /* ARGSUSED */
4036 static int
4037 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4038 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4039 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4040 {
4041 	znode_t		*zp = VTOZ(vp);
4042 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4043 	page_t		**pl0 = pl;
4044 	int		err = 0;
4045 
4046 	/* we do our own caching, faultahead is unnecessary */
4047 	if (pl == NULL)
4048 		return (0);
4049 	else if (len > plsz)
4050 		len = plsz;
4051 	else
4052 		len = P2ROUNDUP(len, PAGESIZE);
4053 	ASSERT(plsz >= len);
4054 
4055 	ZFS_ENTER(zfsvfs);
4056 	ZFS_VERIFY_ZP(zp);
4057 
4058 	if (protp)
4059 		*protp = PROT_ALL;
4060 
4061 	/*
4062 	 * Loop through the requested range [off, off + len) looking
4063 	 * for pages.  If we don't find a page, we will need to create
4064 	 * a new page and fill it with data from the file.
4065 	 */
4066 	while (len > 0) {
4067 		if (*pl = page_lookup(vp, off, SE_SHARED))
4068 			*(pl+1) = NULL;
4069 		else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4070 			goto out;
4071 		while (*pl) {
4072 			ASSERT3U((*pl)->p_offset, ==, off);
4073 			off += PAGESIZE;
4074 			addr += PAGESIZE;
4075 			if (len > 0) {
4076 				ASSERT3U(len, >=, PAGESIZE);
4077 				len -= PAGESIZE;
4078 			}
4079 			ASSERT3U(plsz, >=, PAGESIZE);
4080 			plsz -= PAGESIZE;
4081 			pl++;
4082 		}
4083 	}
4084 
4085 	/*
4086 	 * Fill out the page array with any pages already in the cache.
4087 	 */
4088 	while (plsz > 0 &&
4089 	    (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4090 			off += PAGESIZE;
4091 			plsz -= PAGESIZE;
4092 	}
4093 out:
4094 	if (err) {
4095 		/*
4096 		 * Release any pages we have previously locked.
4097 		 */
4098 		while (pl > pl0)
4099 			page_unlock(*--pl);
4100 	} else {
4101 		ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4102 	}
4103 
4104 	*pl = NULL;
4105 
4106 	ZFS_EXIT(zfsvfs);
4107 	return (err);
4108 }
4109 
4110 /*
4111  * Request a memory map for a section of a file.  This code interacts
4112  * with common code and the VM system as follows:
4113  *
4114  *	common code calls mmap(), which ends up in smmap_common()
4115  *
4116  *	this calls VOP_MAP(), which takes you into (say) zfs
4117  *
4118  *	zfs_map() calls as_map(), passing segvn_create() as the callback
4119  *
4120  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
4121  *
4122  *	zfs_addmap() updates z_mapcnt
4123  */
4124 /*ARGSUSED*/
4125 static int
4126 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4127     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4128     caller_context_t *ct)
4129 {
4130 	znode_t *zp = VTOZ(vp);
4131 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4132 	segvn_crargs_t	vn_a;
4133 	int		error;
4134 
4135 	ZFS_ENTER(zfsvfs);
4136 	ZFS_VERIFY_ZP(zp);
4137 
4138 	if ((prot & PROT_WRITE) &&
4139 	    (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY |
4140 	    ZFS_APPENDONLY))) {
4141 		ZFS_EXIT(zfsvfs);
4142 		return (EPERM);
4143 	}
4144 
4145 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4146 	    (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED)) {
4147 		ZFS_EXIT(zfsvfs);
4148 		return (EACCES);
4149 	}
4150 
4151 	if (vp->v_flag & VNOMAP) {
4152 		ZFS_EXIT(zfsvfs);
4153 		return (ENOSYS);
4154 	}
4155 
4156 	if (off < 0 || len > MAXOFFSET_T - off) {
4157 		ZFS_EXIT(zfsvfs);
4158 		return (ENXIO);
4159 	}
4160 
4161 	if (vp->v_type != VREG) {
4162 		ZFS_EXIT(zfsvfs);
4163 		return (ENODEV);
4164 	}
4165 
4166 	/*
4167 	 * If file is locked, disallow mapping.
4168 	 */
4169 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) {
4170 		ZFS_EXIT(zfsvfs);
4171 		return (EAGAIN);
4172 	}
4173 
4174 	as_rangelock(as);
4175 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4176 	if (error != 0) {
4177 		as_rangeunlock(as);
4178 		ZFS_EXIT(zfsvfs);
4179 		return (error);
4180 	}
4181 
4182 	vn_a.vp = vp;
4183 	vn_a.offset = (u_offset_t)off;
4184 	vn_a.type = flags & MAP_TYPE;
4185 	vn_a.prot = prot;
4186 	vn_a.maxprot = maxprot;
4187 	vn_a.cred = cr;
4188 	vn_a.amp = NULL;
4189 	vn_a.flags = flags & ~MAP_TYPE;
4190 	vn_a.szc = 0;
4191 	vn_a.lgrp_mem_policy_flags = 0;
4192 
4193 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4194 
4195 	as_rangeunlock(as);
4196 	ZFS_EXIT(zfsvfs);
4197 	return (error);
4198 }
4199 
4200 /* ARGSUSED */
4201 static int
4202 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4203     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4204     caller_context_t *ct)
4205 {
4206 	uint64_t pages = btopr(len);
4207 
4208 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4209 	return (0);
4210 }
4211 
4212 /*
4213  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4214  * more accurate mtime for the associated file.  Since we don't have a way of
4215  * detecting when the data was actually modified, we have to resort to
4216  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4217  * last page is pushed.  The problem occurs when the msync() call is omitted,
4218  * which by far the most common case:
4219  *
4220  * 	open()
4221  * 	mmap()
4222  * 	<modify memory>
4223  * 	munmap()
4224  * 	close()
4225  * 	<time lapse>
4226  * 	putpage() via fsflush
4227  *
4228  * If we wait until fsflush to come along, we can have a modification time that
4229  * is some arbitrary point in the future.  In order to prevent this in the
4230  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4231  * torn down.
4232  */
4233 /* ARGSUSED */
4234 static int
4235 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4236     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4237     caller_context_t *ct)
4238 {
4239 	uint64_t pages = btopr(len);
4240 
4241 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4242 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4243 
4244 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4245 	    vn_has_cached_data(vp))
4246 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4247 
4248 	return (0);
4249 }
4250 
4251 /*
4252  * Free or allocate space in a file.  Currently, this function only
4253  * supports the `F_FREESP' command.  However, this command is somewhat
4254  * misnamed, as its functionality includes the ability to allocate as
4255  * well as free space.
4256  *
4257  *	IN:	vp	- vnode of file to free data in.
4258  *		cmd	- action to take (only F_FREESP supported).
4259  *		bfp	- section of file to free/alloc.
4260  *		flag	- current file open mode flags.
4261  *		offset	- current file offset.
4262  *		cr	- credentials of caller [UNUSED].
4263  *		ct	- caller context.
4264  *
4265  *	RETURN:	0 if success
4266  *		error code if failure
4267  *
4268  * Timestamps:
4269  *	vp - ctime|mtime updated
4270  */
4271 /* ARGSUSED */
4272 static int
4273 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4274     offset_t offset, cred_t *cr, caller_context_t *ct)
4275 {
4276 	znode_t		*zp = VTOZ(vp);
4277 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4278 	uint64_t	off, len;
4279 	int		error;
4280 
4281 	ZFS_ENTER(zfsvfs);
4282 	ZFS_VERIFY_ZP(zp);
4283 
4284 	if (cmd != F_FREESP) {
4285 		ZFS_EXIT(zfsvfs);
4286 		return (EINVAL);
4287 	}
4288 
4289 	if (error = convoff(vp, bfp, 0, offset)) {
4290 		ZFS_EXIT(zfsvfs);
4291 		return (error);
4292 	}
4293 
4294 	if (bfp->l_len < 0) {
4295 		ZFS_EXIT(zfsvfs);
4296 		return (EINVAL);
4297 	}
4298 
4299 	off = bfp->l_start;
4300 	len = bfp->l_len; /* 0 means from off to end of file */
4301 
4302 	error = zfs_freesp(zp, off, len, flag, TRUE);
4303 
4304 	ZFS_EXIT(zfsvfs);
4305 	return (error);
4306 }
4307 
4308 /*ARGSUSED*/
4309 static int
4310 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4311 {
4312 	znode_t		*zp = VTOZ(vp);
4313 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4314 	uint32_t	gen;
4315 	uint64_t	object = zp->z_id;
4316 	zfid_short_t	*zfid;
4317 	int		size, i;
4318 
4319 	ZFS_ENTER(zfsvfs);
4320 	ZFS_VERIFY_ZP(zp);
4321 	gen = (uint32_t)zp->z_gen;
4322 
4323 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4324 	if (fidp->fid_len < size) {
4325 		fidp->fid_len = size;
4326 		ZFS_EXIT(zfsvfs);
4327 		return (ENOSPC);
4328 	}
4329 
4330 	zfid = (zfid_short_t *)fidp;
4331 
4332 	zfid->zf_len = size;
4333 
4334 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4335 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4336 
4337 	/* Must have a non-zero generation number to distinguish from .zfs */
4338 	if (gen == 0)
4339 		gen = 1;
4340 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4341 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4342 
4343 	if (size == LONG_FID_LEN) {
4344 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4345 		zfid_long_t	*zlfid;
4346 
4347 		zlfid = (zfid_long_t *)fidp;
4348 
4349 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4350 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4351 
4352 		/* XXX - this should be the generation number for the objset */
4353 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4354 			zlfid->zf_setgen[i] = 0;
4355 	}
4356 
4357 	ZFS_EXIT(zfsvfs);
4358 	return (0);
4359 }
4360 
4361 static int
4362 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4363     caller_context_t *ct)
4364 {
4365 	znode_t		*zp, *xzp;
4366 	zfsvfs_t	*zfsvfs;
4367 	zfs_dirlock_t	*dl;
4368 	int		error;
4369 
4370 	switch (cmd) {
4371 	case _PC_LINK_MAX:
4372 		*valp = ULONG_MAX;
4373 		return (0);
4374 
4375 	case _PC_FILESIZEBITS:
4376 		*valp = 64;
4377 		return (0);
4378 
4379 	case _PC_XATTR_EXISTS:
4380 		zp = VTOZ(vp);
4381 		zfsvfs = zp->z_zfsvfs;
4382 		ZFS_ENTER(zfsvfs);
4383 		ZFS_VERIFY_ZP(zp);
4384 		*valp = 0;
4385 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4386 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4387 		if (error == 0) {
4388 			zfs_dirent_unlock(dl);
4389 			if (!zfs_dirempty(xzp))
4390 				*valp = 1;
4391 			VN_RELE(ZTOV(xzp));
4392 		} else if (error == ENOENT) {
4393 			/*
4394 			 * If there aren't extended attributes, it's the
4395 			 * same as having zero of them.
4396 			 */
4397 			error = 0;
4398 		}
4399 		ZFS_EXIT(zfsvfs);
4400 		return (error);
4401 
4402 	case _PC_SATTR_ENABLED:
4403 	case _PC_SATTR_EXISTS:
4404 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4405 		    (vp->v_type == VREG || vp->v_type == VDIR);
4406 		return (0);
4407 
4408 	case _PC_ACL_ENABLED:
4409 		*valp = _ACL_ACE_ENABLED;
4410 		return (0);
4411 
4412 	case _PC_MIN_HOLE_SIZE:
4413 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4414 		return (0);
4415 
4416 	default:
4417 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4418 	}
4419 }
4420 
4421 /*ARGSUSED*/
4422 static int
4423 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4424     caller_context_t *ct)
4425 {
4426 	znode_t *zp = VTOZ(vp);
4427 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4428 	int error;
4429 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4430 
4431 	ZFS_ENTER(zfsvfs);
4432 	ZFS_VERIFY_ZP(zp);
4433 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4434 	ZFS_EXIT(zfsvfs);
4435 
4436 	return (error);
4437 }
4438 
4439 /*ARGSUSED*/
4440 static int
4441 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4442     caller_context_t *ct)
4443 {
4444 	znode_t *zp = VTOZ(vp);
4445 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4446 	int error;
4447 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4448 
4449 	ZFS_ENTER(zfsvfs);
4450 	ZFS_VERIFY_ZP(zp);
4451 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4452 	ZFS_EXIT(zfsvfs);
4453 	return (error);
4454 }
4455 
4456 /*
4457  * Predeclare these here so that the compiler assumes that
4458  * this is an "old style" function declaration that does
4459  * not include arguments => we won't get type mismatch errors
4460  * in the initializations that follow.
4461  */
4462 static int zfs_inval();
4463 static int zfs_isdir();
4464 
4465 static int
4466 zfs_inval()
4467 {
4468 	return (EINVAL);
4469 }
4470 
4471 static int
4472 zfs_isdir()
4473 {
4474 	return (EISDIR);
4475 }
4476 /*
4477  * Directory vnode operations template
4478  */
4479 vnodeops_t *zfs_dvnodeops;
4480 const fs_operation_def_t zfs_dvnodeops_template[] = {
4481 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4482 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4483 	VOPNAME_READ,		{ .error = zfs_isdir },
4484 	VOPNAME_WRITE,		{ .error = zfs_isdir },
4485 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4486 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4487 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4488 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4489 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4490 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4491 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4492 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4493 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4494 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
4495 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4496 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4497 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
4498 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4499 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4500 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4501 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4502 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4503 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4504 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4505 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
4506 	NULL,			NULL
4507 };
4508 
4509 /*
4510  * Regular file vnode operations template
4511  */
4512 vnodeops_t *zfs_fvnodeops;
4513 const fs_operation_def_t zfs_fvnodeops_template[] = {
4514 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4515 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4516 	VOPNAME_READ,		{ .vop_read = zfs_read },
4517 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
4518 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4519 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4520 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4521 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4522 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4523 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4524 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4525 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4526 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4527 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4528 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
4529 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
4530 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
4531 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
4532 	VOPNAME_MAP,		{ .vop_map = zfs_map },
4533 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
4534 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
4535 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4536 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4537 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4538 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4539 	NULL,			NULL
4540 };
4541 
4542 /*
4543  * Symbolic link vnode operations template
4544  */
4545 vnodeops_t *zfs_symvnodeops;
4546 const fs_operation_def_t zfs_symvnodeops_template[] = {
4547 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4548 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4549 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4550 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4551 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
4552 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4553 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4554 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4555 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4556 	NULL,			NULL
4557 };
4558 
4559 /*
4560  * special share hidden files vnode operations template
4561  */
4562 vnodeops_t *zfs_sharevnodeops;
4563 const fs_operation_def_t zfs_sharevnodeops_template[] = {
4564 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4565 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4566 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4567 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4568 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4569 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4570 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4571 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4572 	NULL,			NULL
4573 };
4574 
4575 /*
4576  * Extended attribute directory vnode operations template
4577  *	This template is identical to the directory vnodes
4578  *	operation template except for restricted operations:
4579  *		VOP_MKDIR()
4580  *		VOP_SYMLINK()
4581  * Note that there are other restrictions embedded in:
4582  *	zfs_create()	- restrict type to VREG
4583  *	zfs_link()	- no links into/out of attribute space
4584  *	zfs_rename()	- no moves into/out of attribute space
4585  */
4586 vnodeops_t *zfs_xdvnodeops;
4587 const fs_operation_def_t zfs_xdvnodeops_template[] = {
4588 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4589 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4590 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4591 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4592 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4593 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4594 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4595 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4596 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4597 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4598 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4599 	VOPNAME_MKDIR,		{ .error = zfs_inval },
4600 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4601 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4602 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
4603 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4604 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4605 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4606 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4607 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4608 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4609 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4610 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4611 	NULL,			NULL
4612 };
4613 
4614 /*
4615  * Error vnode operations template
4616  */
4617 vnodeops_t *zfs_evnodeops;
4618 const fs_operation_def_t zfs_evnodeops_template[] = {
4619 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4620 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4621 	NULL,			NULL
4622 };
4623