xref: /titanic_52/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 8682d1ef2a0960ed5a9f05b9448eaa3e68ac931f)
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 /* Portions Copyright 2007 Jeremy Teo */
26 /* Portions Copyright 2010 Robert Milkowski */
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/dmu_objset.h>
61 #include <sys/spa.h>
62 #include <sys/txg.h>
63 #include <sys/dbuf.h>
64 #include <sys/zap.h>
65 #include <sys/sa.h>
66 #include <sys/dirent.h>
67 #include <sys/policy.h>
68 #include <sys/sunddi.h>
69 #include <sys/filio.h>
70 #include <sys/sid.h>
71 #include "fs/fs_subr.h"
72 #include <sys/zfs_ctldir.h>
73 #include <sys/zfs_fuid.h>
74 #include <sys/zfs_sa.h>
75 #include <sys/dnlc.h>
76 #include <sys/zfs_rlock.h>
77 #include <sys/extdirent.h>
78 #include <sys/kidmap.h>
79 #include <sys/cred.h>
80 #include <sys/attr.h>
81 
82 /*
83  * Programming rules.
84  *
85  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
86  * properly lock its in-core state, create a DMU transaction, do the work,
87  * record this work in the intent log (ZIL), commit the DMU transaction,
88  * and wait for the intent log to commit if it is a synchronous operation.
89  * Moreover, the vnode ops must work in both normal and log replay context.
90  * The ordering of events is important to avoid deadlocks and references
91  * to freed memory.  The example below illustrates the following Big Rules:
92  *
93  *  (1) A check must be made in each zfs thread for a mounted file system.
94  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
95  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
96  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
97  *      can return EIO from the calling function.
98  *
99  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
100  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
101  *	First, if it's the last reference, the vnode/znode
102  *	can be freed, so the zp may point to freed memory.  Second, the last
103  *	reference will call zfs_zinactive(), which may induce a lot of work --
104  *	pushing cached pages (which acquires range locks) and syncing out
105  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
106  *	which could deadlock the system if you were already holding one.
107  *	If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
108  *
109  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
110  *	as they can span dmu_tx_assign() calls.
111  *
112  *  (4)	Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
113  *	This is critical because we don't want to block while holding locks.
114  *	Note, in particular, that if a lock is sometimes acquired before
115  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
116  *	use a non-blocking assign can deadlock the system.  The scenario:
117  *
118  *	Thread A has grabbed a lock before calling dmu_tx_assign().
119  *	Thread B is in an already-assigned tx, and blocks for this lock.
120  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
121  *	forever, because the previous txg can't quiesce until B's tx commits.
122  *
123  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
124  *	then drop all locks, call dmu_tx_wait(), and try again.
125  *
126  *  (5)	If the operation succeeded, generate the intent log entry for it
127  *	before dropping locks.  This ensures that the ordering of events
128  *	in the intent log matches the order in which they actually occurred.
129  *      During ZIL replay the zfs_log_* functions will update the sequence
130  *	number to indicate the zil transaction has replayed.
131  *
132  *  (6)	At the end of each vnode op, the DMU tx must always commit,
133  *	regardless of whether there were any errors.
134  *
135  *  (7)	After dropping all locks, invoke zil_commit(zilog, foid)
136  *	to ensure that synchronous semantics are provided when necessary.
137  *
138  * In general, this is how things should be ordered in each vnode op:
139  *
140  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
141  * top:
142  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
143  *	rw_enter(...);			// grab any other locks you need
144  *	tx = dmu_tx_create(...);	// get DMU tx
145  *	dmu_tx_hold_*();		// hold each object you might modify
146  *	error = dmu_tx_assign(tx, TXG_NOWAIT);	// try to assign
147  *	if (error) {
148  *		rw_exit(...);		// drop locks
149  *		zfs_dirent_unlock(dl);	// unlock directory entry
150  *		VN_RELE(...);		// release held vnodes
151  *		if (error == ERESTART) {
152  *			dmu_tx_wait(tx);
153  *			dmu_tx_abort(tx);
154  *			goto top;
155  *		}
156  *		dmu_tx_abort(tx);	// abort DMU tx
157  *		ZFS_EXIT(zfsvfs);	// finished in zfs
158  *		return (error);		// really out of space
159  *	}
160  *	error = do_real_work();		// do whatever this VOP does
161  *	if (error == 0)
162  *		zfs_log_*(...);		// on success, make ZIL entry
163  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
164  *	rw_exit(...);			// drop locks
165  *	zfs_dirent_unlock(dl);		// unlock directory entry
166  *	VN_RELE(...);			// release held vnodes
167  *	zil_commit(zilog, foid);	// synchronous when necessary
168  *	ZFS_EXIT(zfsvfs);		// finished in zfs
169  *	return (error);			// done, report error
170  */
171 
172 /* ARGSUSED */
173 static int
174 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
175 {
176 	znode_t	*zp = VTOZ(*vpp);
177 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
178 
179 	ZFS_ENTER(zfsvfs);
180 	ZFS_VERIFY_ZP(zp);
181 
182 	if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
183 	    ((flag & FAPPEND) == 0)) {
184 		ZFS_EXIT(zfsvfs);
185 		return (EPERM);
186 	}
187 
188 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
189 	    ZTOV(zp)->v_type == VREG &&
190 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
191 		if (fs_vscan(*vpp, cr, 0) != 0) {
192 			ZFS_EXIT(zfsvfs);
193 			return (EACCES);
194 		}
195 	}
196 
197 	/* Keep a count of the synchronous opens in the znode */
198 	if (flag & (FSYNC | FDSYNC))
199 		atomic_inc_32(&zp->z_sync_cnt);
200 
201 	ZFS_EXIT(zfsvfs);
202 	return (0);
203 }
204 
205 /* ARGSUSED */
206 static int
207 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
208     caller_context_t *ct)
209 {
210 	znode_t	*zp = VTOZ(vp);
211 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
212 
213 	/*
214 	 * Clean up any locks held by this process on the vp.
215 	 */
216 	cleanlocks(vp, ddi_get_pid(), 0);
217 	cleanshares(vp, ddi_get_pid());
218 
219 	ZFS_ENTER(zfsvfs);
220 	ZFS_VERIFY_ZP(zp);
221 
222 	/* Decrement the synchronous opens in the znode */
223 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
224 		atomic_dec_32(&zp->z_sync_cnt);
225 
226 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
227 	    ZTOV(zp)->v_type == VREG &&
228 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
229 		VERIFY(fs_vscan(vp, cr, 1) == 0);
230 
231 	ZFS_EXIT(zfsvfs);
232 	return (0);
233 }
234 
235 /*
236  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
237  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
238  */
239 static int
240 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
241 {
242 	znode_t	*zp = VTOZ(vp);
243 	uint64_t noff = (uint64_t)*off; /* new offset */
244 	uint64_t file_sz;
245 	int error;
246 	boolean_t hole;
247 
248 	file_sz = zp->z_size;
249 	if (noff >= file_sz)  {
250 		return (ENXIO);
251 	}
252 
253 	if (cmd == _FIO_SEEK_HOLE)
254 		hole = B_TRUE;
255 	else
256 		hole = B_FALSE;
257 
258 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
259 
260 	/* end of file? */
261 	if ((error == ESRCH) || (noff > file_sz)) {
262 		/*
263 		 * Handle the virtual hole at the end of file.
264 		 */
265 		if (hole) {
266 			*off = file_sz;
267 			return (0);
268 		}
269 		return (ENXIO);
270 	}
271 
272 	if (noff < *off)
273 		return (error);
274 	*off = noff;
275 	return (error);
276 }
277 
278 /* ARGSUSED */
279 static int
280 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
281     int *rvalp, caller_context_t *ct)
282 {
283 	offset_t off;
284 	int error;
285 	zfsvfs_t *zfsvfs;
286 	znode_t *zp;
287 
288 	switch (com) {
289 	case _FIOFFS:
290 		return (zfs_sync(vp->v_vfsp, 0, cred));
291 
292 		/*
293 		 * The following two ioctls are used by bfu.  Faking out,
294 		 * necessary to avoid bfu errors.
295 		 */
296 	case _FIOGDIO:
297 	case _FIOSDIO:
298 		return (0);
299 
300 	case _FIO_SEEK_DATA:
301 	case _FIO_SEEK_HOLE:
302 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
303 			return (EFAULT);
304 
305 		zp = VTOZ(vp);
306 		zfsvfs = zp->z_zfsvfs;
307 		ZFS_ENTER(zfsvfs);
308 		ZFS_VERIFY_ZP(zp);
309 
310 		/* offset parameter is in/out */
311 		error = zfs_holey(vp, com, &off);
312 		ZFS_EXIT(zfsvfs);
313 		if (error)
314 			return (error);
315 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
316 			return (EFAULT);
317 		return (0);
318 	}
319 	return (ENOTTY);
320 }
321 
322 /*
323  * Utility functions to map and unmap a single physical page.  These
324  * are used to manage the mappable copies of ZFS file data, and therefore
325  * do not update ref/mod bits.
326  */
327 caddr_t
328 zfs_map_page(page_t *pp, enum seg_rw rw)
329 {
330 	if (kpm_enable)
331 		return (hat_kpm_mapin(pp, 0));
332 	ASSERT(rw == S_READ || rw == S_WRITE);
333 	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
334 	    (caddr_t)-1));
335 }
336 
337 void
338 zfs_unmap_page(page_t *pp, caddr_t addr)
339 {
340 	if (kpm_enable) {
341 		hat_kpm_mapout(pp, 0, addr);
342 	} else {
343 		ppmapout(addr);
344 	}
345 }
346 
347 /*
348  * When a file is memory mapped, we must keep the IO data synchronized
349  * between the DMU cache and the memory mapped pages.  What this means:
350  *
351  * On Write:	If we find a memory mapped page, we write to *both*
352  *		the page and the dmu buffer.
353  */
354 static void
355 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
356 {
357 	int64_t	off;
358 
359 	off = start & PAGEOFFSET;
360 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
361 		page_t *pp;
362 		uint64_t nbytes = MIN(PAGESIZE - off, len);
363 
364 		if (pp = page_lookup(vp, start, SE_SHARED)) {
365 			caddr_t va;
366 
367 			va = zfs_map_page(pp, S_WRITE);
368 			(void) dmu_read(os, oid, start+off, nbytes, va+off,
369 			    DMU_READ_PREFETCH);
370 			zfs_unmap_page(pp, va);
371 			page_unlock(pp);
372 		}
373 		len -= nbytes;
374 		off = 0;
375 	}
376 }
377 
378 /*
379  * When a file is memory mapped, we must keep the IO data synchronized
380  * between the DMU cache and the memory mapped pages.  What this means:
381  *
382  * On Read:	We "read" preferentially from memory mapped pages,
383  *		else we default from the dmu buffer.
384  *
385  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
386  *	the file is memory mapped.
387  */
388 static int
389 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
390 {
391 	znode_t *zp = VTOZ(vp);
392 	objset_t *os = zp->z_zfsvfs->z_os;
393 	int64_t	start, off;
394 	int len = nbytes;
395 	int error = 0;
396 
397 	start = uio->uio_loffset;
398 	off = start & PAGEOFFSET;
399 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
400 		page_t *pp;
401 		uint64_t bytes = MIN(PAGESIZE - off, len);
402 
403 		if (pp = page_lookup(vp, start, SE_SHARED)) {
404 			caddr_t va;
405 
406 			va = zfs_map_page(pp, S_READ);
407 			error = uiomove(va + off, bytes, UIO_READ, uio);
408 			zfs_unmap_page(pp, va);
409 			page_unlock(pp);
410 		} else {
411 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
412 		}
413 		len -= bytes;
414 		off = 0;
415 		if (error)
416 			break;
417 	}
418 	return (error);
419 }
420 
421 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
422 
423 /*
424  * Read bytes from specified file into supplied buffer.
425  *
426  *	IN:	vp	- vnode of file to be read from.
427  *		uio	- structure supplying read location, range info,
428  *			  and return buffer.
429  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
430  *		cr	- credentials of caller.
431  *		ct	- caller context
432  *
433  *	OUT:	uio	- updated offset and range, buffer filled.
434  *
435  *	RETURN:	0 if success
436  *		error code if failure
437  *
438  * Side Effects:
439  *	vp - atime updated if byte count > 0
440  */
441 /* ARGSUSED */
442 static int
443 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
444 {
445 	znode_t		*zp = VTOZ(vp);
446 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
447 	objset_t	*os;
448 	ssize_t		n, nbytes;
449 	int		error;
450 	rl_t		*rl;
451 	xuio_t		*xuio = NULL;
452 
453 	ZFS_ENTER(zfsvfs);
454 	ZFS_VERIFY_ZP(zp);
455 	os = zfsvfs->z_os;
456 
457 	if (zp->z_pflags & ZFS_AV_QUARANTINED) {
458 		ZFS_EXIT(zfsvfs);
459 		return (EACCES);
460 	}
461 
462 	/*
463 	 * Validate file offset
464 	 */
465 	if (uio->uio_loffset < (offset_t)0) {
466 		ZFS_EXIT(zfsvfs);
467 		return (EINVAL);
468 	}
469 
470 	/*
471 	 * Fasttrack empty reads
472 	 */
473 	if (uio->uio_resid == 0) {
474 		ZFS_EXIT(zfsvfs);
475 		return (0);
476 	}
477 
478 	/*
479 	 * Check for mandatory locks
480 	 */
481 	if (MANDMODE(zp->z_mode)) {
482 		if (error = chklock(vp, FREAD,
483 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
484 			ZFS_EXIT(zfsvfs);
485 			return (error);
486 		}
487 	}
488 
489 	/*
490 	 * If we're in FRSYNC mode, sync out this znode before reading it.
491 	 */
492 	if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
493 		zil_commit(zfsvfs->z_log, zp->z_id);
494 
495 	/*
496 	 * Lock the range against changes.
497 	 */
498 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
499 
500 	/*
501 	 * If we are reading past end-of-file we can skip
502 	 * to the end; but we might still need to set atime.
503 	 */
504 	if (uio->uio_loffset >= zp->z_size) {
505 		error = 0;
506 		goto out;
507 	}
508 
509 	ASSERT(uio->uio_loffset < zp->z_size);
510 	n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
511 
512 	if ((uio->uio_extflg == UIO_XUIO) &&
513 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
514 		int nblk;
515 		int blksz = zp->z_blksz;
516 		uint64_t offset = uio->uio_loffset;
517 
518 		xuio = (xuio_t *)uio;
519 		if ((ISP2(blksz))) {
520 			nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
521 			    blksz)) / blksz;
522 		} else {
523 			ASSERT(offset + n <= blksz);
524 			nblk = 1;
525 		}
526 		(void) dmu_xuio_init(xuio, nblk);
527 
528 		if (vn_has_cached_data(vp)) {
529 			/*
530 			 * For simplicity, we always allocate a full buffer
531 			 * even if we only expect to read a portion of a block.
532 			 */
533 			while (--nblk >= 0) {
534 				(void) dmu_xuio_add(xuio,
535 				    dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
536 				    blksz), 0, blksz);
537 			}
538 		}
539 	}
540 
541 	while (n > 0) {
542 		nbytes = MIN(n, zfs_read_chunk_size -
543 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
544 
545 		if (vn_has_cached_data(vp))
546 			error = mappedread(vp, nbytes, uio);
547 		else
548 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
549 		if (error) {
550 			/* convert checksum errors into IO errors */
551 			if (error == ECKSUM)
552 				error = EIO;
553 			break;
554 		}
555 
556 		n -= nbytes;
557 	}
558 out:
559 	zfs_range_unlock(rl);
560 
561 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
562 	ZFS_EXIT(zfsvfs);
563 	return (error);
564 }
565 
566 /*
567  * Write the bytes to a file.
568  *
569  *	IN:	vp	- vnode of file to be written to.
570  *		uio	- structure supplying write location, range info,
571  *			  and data buffer.
572  *		ioflag	- FAPPEND flag set if in append mode.
573  *		cr	- credentials of caller.
574  *		ct	- caller context (NFS/CIFS fem monitor only)
575  *
576  *	OUT:	uio	- updated offset and range.
577  *
578  *	RETURN:	0 if success
579  *		error code if failure
580  *
581  * Timestamps:
582  *	vp - ctime|mtime updated if byte count > 0
583  */
584 
585 /* ARGSUSED */
586 static int
587 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
588 {
589 	znode_t		*zp = VTOZ(vp);
590 	rlim64_t	limit = uio->uio_llimit;
591 	ssize_t		start_resid = uio->uio_resid;
592 	ssize_t		tx_bytes;
593 	uint64_t	end_size;
594 	dmu_tx_t	*tx;
595 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
596 	zilog_t		*zilog;
597 	offset_t	woff;
598 	ssize_t		n, nbytes;
599 	rl_t		*rl;
600 	int		max_blksz = zfsvfs->z_max_blksz;
601 	int		error;
602 	arc_buf_t	*abuf;
603 	iovec_t		*aiov;
604 	xuio_t		*xuio = NULL;
605 	int		i_iov = 0;
606 	int		iovcnt = uio->uio_iovcnt;
607 	iovec_t		*iovp = uio->uio_iov;
608 	int		write_eof;
609 	int		count = 0;
610 	sa_bulk_attr_t	bulk[4];
611 	uint64_t	mtime[2], ctime[2];
612 
613 	/*
614 	 * Fasttrack empty write
615 	 */
616 	n = start_resid;
617 	if (n == 0)
618 		return (0);
619 
620 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
621 		limit = MAXOFFSET_T;
622 
623 	ZFS_ENTER(zfsvfs);
624 	ZFS_VERIFY_ZP(zp);
625 
626 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
627 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
628 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
629 	    &zp->z_size, 8);
630 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
631 	    &zp->z_pflags, 8);
632 
633 	/*
634 	 * If immutable or not appending then return EPERM
635 	 */
636 	if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
637 	    ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
638 	    (uio->uio_loffset < zp->z_size))) {
639 		ZFS_EXIT(zfsvfs);
640 		return (EPERM);
641 	}
642 
643 	zilog = zfsvfs->z_log;
644 
645 	/*
646 	 * Validate file offset
647 	 */
648 	woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
649 	if (woff < 0) {
650 		ZFS_EXIT(zfsvfs);
651 		return (EINVAL);
652 	}
653 
654 	/*
655 	 * Check for mandatory locks before calling zfs_range_lock()
656 	 * in order to prevent a deadlock with locks set via fcntl().
657 	 */
658 	if (MANDMODE((mode_t)zp->z_mode) &&
659 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
660 		ZFS_EXIT(zfsvfs);
661 		return (error);
662 	}
663 
664 	/*
665 	 * Pre-fault the pages to ensure slow (eg NFS) pages
666 	 * don't hold up txg.
667 	 * Skip this if uio contains loaned arc_buf.
668 	 */
669 	if ((uio->uio_extflg == UIO_XUIO) &&
670 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
671 		xuio = (xuio_t *)uio;
672 	else
673 		uio_prefaultpages(MIN(n, max_blksz), uio);
674 
675 	/*
676 	 * If in append mode, set the io offset pointer to eof.
677 	 */
678 	if (ioflag & FAPPEND) {
679 		/*
680 		 * Obtain an appending range lock to guarantee file append
681 		 * semantics.  We reset the write offset once we have the lock.
682 		 */
683 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
684 		woff = rl->r_off;
685 		if (rl->r_len == UINT64_MAX) {
686 			/*
687 			 * We overlocked the file because this write will cause
688 			 * the file block size to increase.
689 			 * Note that zp_size cannot change with this lock held.
690 			 */
691 			woff = zp->z_size;
692 		}
693 		uio->uio_loffset = woff;
694 	} else {
695 		/*
696 		 * Note that if the file block size will change as a result of
697 		 * this write, then this range lock will lock the entire file
698 		 * so that we can re-write the block safely.
699 		 */
700 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
701 	}
702 
703 	if (woff >= limit) {
704 		zfs_range_unlock(rl);
705 		ZFS_EXIT(zfsvfs);
706 		return (EFBIG);
707 	}
708 
709 	if ((woff + n) > limit || woff > (limit - n))
710 		n = limit - woff;
711 
712 	/* Will this write extend the file length? */
713 	write_eof = (woff + n > zp->z_size);
714 
715 	end_size = MAX(zp->z_size, woff + n);
716 
717 	/*
718 	 * Write the file in reasonable size chunks.  Each chunk is written
719 	 * in a separate transaction; this keeps the intent log records small
720 	 * and allows us to do more fine-grained space accounting.
721 	 */
722 	while (n > 0) {
723 		abuf = NULL;
724 		woff = uio->uio_loffset;
725 again:
726 		if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
727 		    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
728 			if (abuf != NULL)
729 				dmu_return_arcbuf(abuf);
730 			error = EDQUOT;
731 			break;
732 		}
733 
734 		if (xuio && abuf == NULL) {
735 			ASSERT(i_iov < iovcnt);
736 			aiov = &iovp[i_iov];
737 			abuf = dmu_xuio_arcbuf(xuio, i_iov);
738 			dmu_xuio_clear(xuio, i_iov);
739 			DTRACE_PROBE3(zfs_cp_write, int, i_iov,
740 			    iovec_t *, aiov, arc_buf_t *, abuf);
741 			ASSERT((aiov->iov_base == abuf->b_data) ||
742 			    ((char *)aiov->iov_base - (char *)abuf->b_data +
743 			    aiov->iov_len == arc_buf_size(abuf)));
744 			i_iov++;
745 		} else if (abuf == NULL && n >= max_blksz &&
746 		    woff >= zp->z_size &&
747 		    P2PHASE(woff, max_blksz) == 0 &&
748 		    zp->z_blksz == max_blksz) {
749 			/*
750 			 * This write covers a full block.  "Borrow" a buffer
751 			 * from the dmu so that we can fill it before we enter
752 			 * a transaction.  This avoids the possibility of
753 			 * holding up the transaction if the data copy hangs
754 			 * up on a pagefault (e.g., from an NFS server mapping).
755 			 */
756 			size_t cbytes;
757 
758 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
759 			    max_blksz);
760 			ASSERT(abuf != NULL);
761 			ASSERT(arc_buf_size(abuf) == max_blksz);
762 			if (error = uiocopy(abuf->b_data, max_blksz,
763 			    UIO_WRITE, uio, &cbytes)) {
764 				dmu_return_arcbuf(abuf);
765 				break;
766 			}
767 			ASSERT(cbytes == max_blksz);
768 		}
769 
770 		/*
771 		 * Start a transaction.
772 		 */
773 		tx = dmu_tx_create(zfsvfs->z_os);
774 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
775 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
776 		zfs_sa_upgrade_txholds(tx, zp);
777 		error = dmu_tx_assign(tx, TXG_NOWAIT);
778 		if (error) {
779 			if (error == ERESTART) {
780 				dmu_tx_wait(tx);
781 				dmu_tx_abort(tx);
782 				goto again;
783 			}
784 			dmu_tx_abort(tx);
785 			if (abuf != NULL)
786 				dmu_return_arcbuf(abuf);
787 			break;
788 		}
789 
790 		/*
791 		 * If zfs_range_lock() over-locked we grow the blocksize
792 		 * and then reduce the lock range.  This will only happen
793 		 * on the first iteration since zfs_range_reduce() will
794 		 * shrink down r_len to the appropriate size.
795 		 */
796 		if (rl->r_len == UINT64_MAX) {
797 			uint64_t new_blksz;
798 
799 			if (zp->z_blksz > max_blksz) {
800 				ASSERT(!ISP2(zp->z_blksz));
801 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
802 			} else {
803 				new_blksz = MIN(end_size, max_blksz);
804 			}
805 			zfs_grow_blocksize(zp, new_blksz, tx);
806 			zfs_range_reduce(rl, woff, n);
807 		}
808 
809 		/*
810 		 * XXX - should we really limit each write to z_max_blksz?
811 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
812 		 */
813 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
814 
815 		if (abuf == NULL) {
816 			tx_bytes = uio->uio_resid;
817 			error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
818 			    uio, nbytes, tx);
819 			tx_bytes -= uio->uio_resid;
820 		} else {
821 			tx_bytes = nbytes;
822 			ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
823 			/*
824 			 * If this is not a full block write, but we are
825 			 * extending the file past EOF and this data starts
826 			 * block-aligned, use assign_arcbuf().  Otherwise,
827 			 * write via dmu_write().
828 			 */
829 			if (tx_bytes < max_blksz && (!write_eof ||
830 			    aiov->iov_base != abuf->b_data)) {
831 				ASSERT(xuio);
832 				dmu_write(zfsvfs->z_os, zp->z_id, woff,
833 				    aiov->iov_len, aiov->iov_base, tx);
834 				dmu_return_arcbuf(abuf);
835 				xuio_stat_wbuf_copied();
836 			} else {
837 				ASSERT(xuio || tx_bytes == max_blksz);
838 				dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
839 				    woff, abuf, tx);
840 			}
841 			ASSERT(tx_bytes <= uio->uio_resid);
842 			uioskip(uio, tx_bytes);
843 		}
844 		if (tx_bytes && vn_has_cached_data(vp)) {
845 			update_pages(vp, woff,
846 			    tx_bytes, zfsvfs->z_os, zp->z_id);
847 		}
848 
849 		/*
850 		 * If we made no progress, we're done.  If we made even
851 		 * partial progress, update the znode and ZIL accordingly.
852 		 */
853 		if (tx_bytes == 0) {
854 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
855 			    (void *)&zp->z_size, sizeof (uint64_t), tx);
856 			dmu_tx_commit(tx);
857 			ASSERT(error != 0);
858 			break;
859 		}
860 
861 		/*
862 		 * Clear Set-UID/Set-GID bits on successful write if not
863 		 * privileged and at least one of the excute bits is set.
864 		 *
865 		 * It would be nice to to this after all writes have
866 		 * been done, but that would still expose the ISUID/ISGID
867 		 * to another app after the partial write is committed.
868 		 *
869 		 * Note: we don't call zfs_fuid_map_id() here because
870 		 * user 0 is not an ephemeral uid.
871 		 */
872 		mutex_enter(&zp->z_acl_lock);
873 		if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
874 		    (S_IXUSR >> 6))) != 0 &&
875 		    (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
876 		    secpolicy_vnode_setid_retain(cr,
877 		    (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
878 			uint64_t newmode;
879 			zp->z_mode &= ~(S_ISUID | S_ISGID);
880 			newmode = zp->z_mode;
881 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
882 			    (void *)&newmode, sizeof (uint64_t), tx);
883 		}
884 		mutex_exit(&zp->z_acl_lock);
885 
886 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
887 		    B_TRUE);
888 
889 		/*
890 		 * Update the file size (zp_size) if it has changed;
891 		 * account for possible concurrent updates.
892 		 */
893 		while ((end_size = zp->z_size) < uio->uio_loffset) {
894 			(void) atomic_cas_64(&zp->z_size, end_size,
895 			    uio->uio_loffset);
896 			ASSERT(error == 0);
897 		}
898 		/*
899 		 * If we are replaying and eof is non zero then force
900 		 * the file size to the specified eof. Note, there's no
901 		 * concurrency during replay.
902 		 */
903 		if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
904 			zp->z_size = zfsvfs->z_replay_eof;
905 
906 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
907 
908 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
909 		dmu_tx_commit(tx);
910 
911 		if (error != 0)
912 			break;
913 		ASSERT(tx_bytes == nbytes);
914 		n -= nbytes;
915 
916 		if (!xuio && n > 0)
917 			uio_prefaultpages(MIN(n, max_blksz), uio);
918 	}
919 
920 	zfs_range_unlock(rl);
921 
922 	/*
923 	 * If we're in replay mode, or we made no progress, return error.
924 	 * Otherwise, it's at least a partial write, so it's successful.
925 	 */
926 	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
927 		ZFS_EXIT(zfsvfs);
928 		return (error);
929 	}
930 
931 	if (ioflag & (FSYNC | FDSYNC) ||
932 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
933 		zil_commit(zilog, zp->z_id);
934 
935 	ZFS_EXIT(zfsvfs);
936 	return (0);
937 }
938 
939 void
940 zfs_get_done(zgd_t *zgd, int error)
941 {
942 	znode_t *zp = zgd->zgd_private;
943 	objset_t *os = zp->z_zfsvfs->z_os;
944 
945 	if (zgd->zgd_db)
946 		dmu_buf_rele(zgd->zgd_db, zgd);
947 
948 	zfs_range_unlock(zgd->zgd_rl);
949 
950 	/*
951 	 * Release the vnode asynchronously as we currently have the
952 	 * txg stopped from syncing.
953 	 */
954 	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
955 
956 	if (error == 0 && zgd->zgd_bp)
957 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
958 
959 	kmem_free(zgd, sizeof (zgd_t));
960 }
961 
962 #ifdef DEBUG
963 static int zil_fault_io = 0;
964 #endif
965 
966 /*
967  * Get data to generate a TX_WRITE intent log record.
968  */
969 int
970 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
971 {
972 	zfsvfs_t *zfsvfs = arg;
973 	objset_t *os = zfsvfs->z_os;
974 	znode_t *zp;
975 	uint64_t object = lr->lr_foid;
976 	uint64_t offset = lr->lr_offset;
977 	uint64_t size = lr->lr_length;
978 	blkptr_t *bp = &lr->lr_blkptr;
979 	dmu_buf_t *db;
980 	zgd_t *zgd;
981 	int error = 0;
982 
983 	ASSERT(zio != NULL);
984 	ASSERT(size != 0);
985 
986 	/*
987 	 * Nothing to do if the file has been removed
988 	 */
989 	if (zfs_zget(zfsvfs, object, &zp) != 0)
990 		return (ENOENT);
991 	if (zp->z_unlinked) {
992 		/*
993 		 * Release the vnode asynchronously as we currently have the
994 		 * txg stopped from syncing.
995 		 */
996 		VN_RELE_ASYNC(ZTOV(zp),
997 		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
998 		return (ENOENT);
999 	}
1000 
1001 	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1002 	zgd->zgd_zilog = zfsvfs->z_log;
1003 	zgd->zgd_private = zp;
1004 
1005 	/*
1006 	 * Write records come in two flavors: immediate and indirect.
1007 	 * For small writes it's cheaper to store the data with the
1008 	 * log record (immediate); for large writes it's cheaper to
1009 	 * sync the data and get a pointer to it (indirect) so that
1010 	 * we don't have to write the data twice.
1011 	 */
1012 	if (buf != NULL) { /* immediate write */
1013 		zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1014 		/* test for truncation needs to be done while range locked */
1015 		if (offset >= zp->z_size) {
1016 			error = ENOENT;
1017 		} else {
1018 			error = dmu_read(os, object, offset, size, buf,
1019 			    DMU_READ_NO_PREFETCH);
1020 		}
1021 		ASSERT(error == 0 || error == ENOENT);
1022 	} else { /* indirect write */
1023 		/*
1024 		 * Have to lock the whole block to ensure when it's
1025 		 * written out and it's checksum is being calculated
1026 		 * that no one can change the data. We need to re-check
1027 		 * blocksize after we get the lock in case it's changed!
1028 		 */
1029 		for (;;) {
1030 			uint64_t blkoff;
1031 			size = zp->z_blksz;
1032 			blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1033 			offset -= blkoff;
1034 			zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1035 			    RL_READER);
1036 			if (zp->z_blksz == size)
1037 				break;
1038 			offset += blkoff;
1039 			zfs_range_unlock(zgd->zgd_rl);
1040 		}
1041 		/* test for truncation needs to be done while range locked */
1042 		if (lr->lr_offset >= zp->z_size)
1043 			error = ENOENT;
1044 #ifdef DEBUG
1045 		if (zil_fault_io) {
1046 			error = EIO;
1047 			zil_fault_io = 0;
1048 		}
1049 #endif
1050 		if (error == 0)
1051 			error = dmu_buf_hold(os, object, offset, zgd, &db,
1052 			    DMU_READ_NO_PREFETCH);
1053 
1054 		if (error == 0) {
1055 			zgd->zgd_db = db;
1056 			zgd->zgd_bp = bp;
1057 
1058 			ASSERT(db->db_offset == offset);
1059 			ASSERT(db->db_size == size);
1060 
1061 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1062 			    zfs_get_done, zgd);
1063 			ASSERT(error || lr->lr_length <= zp->z_blksz);
1064 
1065 			/*
1066 			 * On success, we need to wait for the write I/O
1067 			 * initiated by dmu_sync() to complete before we can
1068 			 * release this dbuf.  We will finish everything up
1069 			 * in the zfs_get_done() callback.
1070 			 */
1071 			if (error == 0)
1072 				return (0);
1073 
1074 			if (error == EALREADY) {
1075 				lr->lr_common.lrc_txtype = TX_WRITE2;
1076 				error = 0;
1077 			}
1078 		}
1079 	}
1080 
1081 	zfs_get_done(zgd, error);
1082 
1083 	return (error);
1084 }
1085 
1086 /*ARGSUSED*/
1087 static int
1088 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1089     caller_context_t *ct)
1090 {
1091 	znode_t *zp = VTOZ(vp);
1092 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1093 	int error;
1094 
1095 	ZFS_ENTER(zfsvfs);
1096 	ZFS_VERIFY_ZP(zp);
1097 
1098 	if (flag & V_ACE_MASK)
1099 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1100 	else
1101 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
1102 
1103 	ZFS_EXIT(zfsvfs);
1104 	return (error);
1105 }
1106 
1107 /*
1108  * If vnode is for a device return a specfs vnode instead.
1109  */
1110 static int
1111 specvp_check(vnode_t **vpp, cred_t *cr)
1112 {
1113 	int error = 0;
1114 
1115 	if (IS_DEVVP(*vpp)) {
1116 		struct vnode *svp;
1117 
1118 		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1119 		VN_RELE(*vpp);
1120 		if (svp == NULL)
1121 			error = ENOSYS;
1122 		*vpp = svp;
1123 	}
1124 	return (error);
1125 }
1126 
1127 
1128 /*
1129  * Lookup an entry in a directory, or an extended attribute directory.
1130  * If it exists, return a held vnode reference for it.
1131  *
1132  *	IN:	dvp	- vnode of directory to search.
1133  *		nm	- name of entry to lookup.
1134  *		pnp	- full pathname to lookup [UNUSED].
1135  *		flags	- LOOKUP_XATTR set if looking for an attribute.
1136  *		rdir	- root directory vnode [UNUSED].
1137  *		cr	- credentials of caller.
1138  *		ct	- caller context
1139  *		direntflags - directory lookup flags
1140  *		realpnp - returned pathname.
1141  *
1142  *	OUT:	vpp	- vnode of located entry, NULL if not found.
1143  *
1144  *	RETURN:	0 if success
1145  *		error code if failure
1146  *
1147  * Timestamps:
1148  *	NA
1149  */
1150 /* ARGSUSED */
1151 static int
1152 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1153     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1154     int *direntflags, pathname_t *realpnp)
1155 {
1156 	znode_t *zdp = VTOZ(dvp);
1157 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1158 	int	error = 0;
1159 
1160 	/* fast path */
1161 	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1162 
1163 		if (dvp->v_type != VDIR) {
1164 			return (ENOTDIR);
1165 		} else if (zdp->z_sa_hdl == NULL) {
1166 			return (EIO);
1167 		}
1168 
1169 		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1170 			error = zfs_fastaccesschk_execute(zdp, cr);
1171 			if (!error) {
1172 				*vpp = dvp;
1173 				VN_HOLD(*vpp);
1174 				return (0);
1175 			}
1176 			return (error);
1177 		} else {
1178 			vnode_t *tvp = dnlc_lookup(dvp, nm);
1179 
1180 			if (tvp) {
1181 				error = zfs_fastaccesschk_execute(zdp, cr);
1182 				if (error) {
1183 					VN_RELE(tvp);
1184 					return (error);
1185 				}
1186 				if (tvp == DNLC_NO_VNODE) {
1187 					VN_RELE(tvp);
1188 					return (ENOENT);
1189 				} else {
1190 					*vpp = tvp;
1191 					return (specvp_check(vpp, cr));
1192 				}
1193 			}
1194 		}
1195 	}
1196 
1197 	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1198 
1199 	ZFS_ENTER(zfsvfs);
1200 	ZFS_VERIFY_ZP(zdp);
1201 
1202 	*vpp = NULL;
1203 
1204 	if (flags & LOOKUP_XATTR) {
1205 		/*
1206 		 * If the xattr property is off, refuse the lookup request.
1207 		 */
1208 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1209 			ZFS_EXIT(zfsvfs);
1210 			return (EINVAL);
1211 		}
1212 
1213 		/*
1214 		 * We don't allow recursive attributes..
1215 		 * Maybe someday we will.
1216 		 */
1217 		if (zdp->z_pflags & ZFS_XATTR) {
1218 			ZFS_EXIT(zfsvfs);
1219 			return (EINVAL);
1220 		}
1221 
1222 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1223 			ZFS_EXIT(zfsvfs);
1224 			return (error);
1225 		}
1226 
1227 		/*
1228 		 * Do we have permission to get into attribute directory?
1229 		 */
1230 
1231 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1232 		    B_FALSE, cr)) {
1233 			VN_RELE(*vpp);
1234 			*vpp = NULL;
1235 		}
1236 
1237 		ZFS_EXIT(zfsvfs);
1238 		return (error);
1239 	}
1240 
1241 	if (dvp->v_type != VDIR) {
1242 		ZFS_EXIT(zfsvfs);
1243 		return (ENOTDIR);
1244 	}
1245 
1246 	/*
1247 	 * Check accessibility of directory.
1248 	 */
1249 
1250 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1251 		ZFS_EXIT(zfsvfs);
1252 		return (error);
1253 	}
1254 
1255 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1256 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1257 		ZFS_EXIT(zfsvfs);
1258 		return (EILSEQ);
1259 	}
1260 
1261 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1262 	if (error == 0)
1263 		error = specvp_check(vpp, cr);
1264 
1265 	ZFS_EXIT(zfsvfs);
1266 	return (error);
1267 }
1268 
1269 /*
1270  * Attempt to create a new entry in a directory.  If the entry
1271  * already exists, truncate the file if permissible, else return
1272  * an error.  Return the vp of the created or trunc'd file.
1273  *
1274  *	IN:	dvp	- vnode of directory to put new file entry in.
1275  *		name	- name of new file entry.
1276  *		vap	- attributes of new file.
1277  *		excl	- flag indicating exclusive or non-exclusive mode.
1278  *		mode	- mode to open file with.
1279  *		cr	- credentials of caller.
1280  *		flag	- large file flag [UNUSED].
1281  *		ct	- caller context
1282  *		vsecp 	- ACL to be set
1283  *
1284  *	OUT:	vpp	- vnode of created or trunc'd entry.
1285  *
1286  *	RETURN:	0 if success
1287  *		error code if failure
1288  *
1289  * Timestamps:
1290  *	dvp - ctime|mtime updated if new entry created
1291  *	 vp - ctime|mtime always, atime if new
1292  */
1293 
1294 /* ARGSUSED */
1295 static int
1296 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1297     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1298     vsecattr_t *vsecp)
1299 {
1300 	znode_t		*zp, *dzp = VTOZ(dvp);
1301 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1302 	zilog_t		*zilog;
1303 	objset_t	*os;
1304 	zfs_dirlock_t	*dl;
1305 	dmu_tx_t	*tx;
1306 	int		error;
1307 	ksid_t		*ksid;
1308 	uid_t		uid;
1309 	gid_t		gid = crgetgid(cr);
1310 	zfs_acl_ids_t   acl_ids;
1311 	boolean_t	fuid_dirtied;
1312 	boolean_t	have_acl = B_FALSE;
1313 
1314 	/*
1315 	 * If we have an ephemeral id, ACL, or XVATTR then
1316 	 * make sure file system is at proper version
1317 	 */
1318 
1319 	ksid = crgetsid(cr, KSID_OWNER);
1320 	if (ksid)
1321 		uid = ksid_getid(ksid);
1322 	else
1323 		uid = crgetuid(cr);
1324 
1325 	if (zfsvfs->z_use_fuids == B_FALSE &&
1326 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1327 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1328 		return (EINVAL);
1329 
1330 	ZFS_ENTER(zfsvfs);
1331 	ZFS_VERIFY_ZP(dzp);
1332 	os = zfsvfs->z_os;
1333 	zilog = zfsvfs->z_log;
1334 
1335 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1336 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1337 		ZFS_EXIT(zfsvfs);
1338 		return (EILSEQ);
1339 	}
1340 
1341 	if (vap->va_mask & AT_XVATTR) {
1342 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1343 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1344 			ZFS_EXIT(zfsvfs);
1345 			return (error);
1346 		}
1347 	}
1348 top:
1349 	*vpp = NULL;
1350 
1351 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1352 		vap->va_mode &= ~VSVTX;
1353 
1354 	if (*name == '\0') {
1355 		/*
1356 		 * Null component name refers to the directory itself.
1357 		 */
1358 		VN_HOLD(dvp);
1359 		zp = dzp;
1360 		dl = NULL;
1361 		error = 0;
1362 	} else {
1363 		/* possible VN_HOLD(zp) */
1364 		int zflg = 0;
1365 
1366 		if (flag & FIGNORECASE)
1367 			zflg |= ZCILOOK;
1368 
1369 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1370 		    NULL, NULL);
1371 		if (error) {
1372 			if (have_acl)
1373 				zfs_acl_ids_free(&acl_ids);
1374 			if (strcmp(name, "..") == 0)
1375 				error = EISDIR;
1376 			ZFS_EXIT(zfsvfs);
1377 			return (error);
1378 		}
1379 	}
1380 
1381 	if (zp == NULL) {
1382 		uint64_t txtype;
1383 
1384 		/*
1385 		 * Create a new file object and update the directory
1386 		 * to reference it.
1387 		 */
1388 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1389 			if (have_acl)
1390 				zfs_acl_ids_free(&acl_ids);
1391 			goto out;
1392 		}
1393 
1394 		/*
1395 		 * We only support the creation of regular files in
1396 		 * extended attribute directories.
1397 		 */
1398 
1399 		if ((dzp->z_pflags & ZFS_XATTR) &&
1400 		    (vap->va_type != VREG)) {
1401 			if (have_acl)
1402 				zfs_acl_ids_free(&acl_ids);
1403 			error = EINVAL;
1404 			goto out;
1405 		}
1406 
1407 		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1408 		    cr, vsecp, &acl_ids)) != 0)
1409 			goto out;
1410 		have_acl = B_TRUE;
1411 
1412 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1413 			zfs_acl_ids_free(&acl_ids);
1414 			error = EDQUOT;
1415 			goto out;
1416 		}
1417 
1418 		tx = dmu_tx_create(os);
1419 
1420 		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1421 		    ZFS_SA_BASE_ATTR_SIZE);
1422 
1423 		fuid_dirtied = zfsvfs->z_fuid_dirty;
1424 		if (fuid_dirtied)
1425 			zfs_fuid_txhold(zfsvfs, tx);
1426 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1427 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1428 		if (!zfsvfs->z_use_sa &&
1429 		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1430 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1431 			    0, acl_ids.z_aclp->z_acl_bytes);
1432 		}
1433 		error = dmu_tx_assign(tx, TXG_NOWAIT);
1434 		if (error) {
1435 			zfs_dirent_unlock(dl);
1436 			if (error == ERESTART) {
1437 				dmu_tx_wait(tx);
1438 				dmu_tx_abort(tx);
1439 				goto top;
1440 			}
1441 			zfs_acl_ids_free(&acl_ids);
1442 			dmu_tx_abort(tx);
1443 			ZFS_EXIT(zfsvfs);
1444 			return (error);
1445 		}
1446 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1447 
1448 		if (fuid_dirtied)
1449 			zfs_fuid_sync(zfsvfs, tx);
1450 
1451 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1452 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1453 		if (flag & FIGNORECASE)
1454 			txtype |= TX_CI;
1455 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1456 		    vsecp, acl_ids.z_fuidp, vap);
1457 		zfs_acl_ids_free(&acl_ids);
1458 		dmu_tx_commit(tx);
1459 	} else {
1460 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1461 
1462 		if (have_acl)
1463 			zfs_acl_ids_free(&acl_ids);
1464 		have_acl = B_FALSE;
1465 
1466 		/*
1467 		 * A directory entry already exists for this name.
1468 		 */
1469 		/*
1470 		 * Can't truncate an existing file if in exclusive mode.
1471 		 */
1472 		if (excl == EXCL) {
1473 			error = EEXIST;
1474 			goto out;
1475 		}
1476 		/*
1477 		 * Can't open a directory for writing.
1478 		 */
1479 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1480 			error = EISDIR;
1481 			goto out;
1482 		}
1483 		/*
1484 		 * Verify requested access to file.
1485 		 */
1486 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1487 			goto out;
1488 		}
1489 
1490 		mutex_enter(&dzp->z_lock);
1491 		dzp->z_seq++;
1492 		mutex_exit(&dzp->z_lock);
1493 
1494 		/*
1495 		 * Truncate regular files if requested.
1496 		 */
1497 		if ((ZTOV(zp)->v_type == VREG) &&
1498 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1499 			/* we can't hold any locks when calling zfs_freesp() */
1500 			zfs_dirent_unlock(dl);
1501 			dl = NULL;
1502 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1503 			if (error == 0) {
1504 				vnevent_create(ZTOV(zp), ct);
1505 			}
1506 		}
1507 	}
1508 out:
1509 
1510 	if (dl)
1511 		zfs_dirent_unlock(dl);
1512 
1513 	if (error) {
1514 		if (zp)
1515 			VN_RELE(ZTOV(zp));
1516 	} else {
1517 		*vpp = ZTOV(zp);
1518 		error = specvp_check(vpp, cr);
1519 	}
1520 
1521 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1522 		zil_commit(zilog, 0);
1523 
1524 	ZFS_EXIT(zfsvfs);
1525 	return (error);
1526 }
1527 
1528 /*
1529  * Remove an entry from a directory.
1530  *
1531  *	IN:	dvp	- vnode of directory to remove entry from.
1532  *		name	- name of entry to remove.
1533  *		cr	- credentials of caller.
1534  *		ct	- caller context
1535  *		flags	- case flags
1536  *
1537  *	RETURN:	0 if success
1538  *		error code if failure
1539  *
1540  * Timestamps:
1541  *	dvp - ctime|mtime
1542  *	 vp - ctime (if nlink > 0)
1543  */
1544 
1545 uint64_t null_xattr = 0;
1546 
1547 /*ARGSUSED*/
1548 static int
1549 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1550     int flags)
1551 {
1552 	znode_t		*zp, *dzp = VTOZ(dvp);
1553 	znode_t		*xzp;
1554 	vnode_t		*vp;
1555 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1556 	zilog_t		*zilog;
1557 	uint64_t	acl_obj, xattr_obj;
1558 	uint64_t 	xattr_obj_unlinked = 0;
1559 	uint64_t	obj = 0;
1560 	zfs_dirlock_t	*dl;
1561 	dmu_tx_t	*tx;
1562 	boolean_t	may_delete_now, delete_now = FALSE;
1563 	boolean_t	unlinked, toobig = FALSE;
1564 	uint64_t	txtype;
1565 	pathname_t	*realnmp = NULL;
1566 	pathname_t	realnm;
1567 	int		error;
1568 	int		zflg = ZEXISTS;
1569 
1570 	ZFS_ENTER(zfsvfs);
1571 	ZFS_VERIFY_ZP(dzp);
1572 	zilog = zfsvfs->z_log;
1573 
1574 	if (flags & FIGNORECASE) {
1575 		zflg |= ZCILOOK;
1576 		pn_alloc(&realnm);
1577 		realnmp = &realnm;
1578 	}
1579 
1580 top:
1581 	xattr_obj = 0;
1582 	xzp = NULL;
1583 	/*
1584 	 * Attempt to lock directory; fail if entry doesn't exist.
1585 	 */
1586 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1587 	    NULL, realnmp)) {
1588 		if (realnmp)
1589 			pn_free(realnmp);
1590 		ZFS_EXIT(zfsvfs);
1591 		return (error);
1592 	}
1593 
1594 	vp = ZTOV(zp);
1595 
1596 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1597 		goto out;
1598 	}
1599 
1600 	/*
1601 	 * Need to use rmdir for removing directories.
1602 	 */
1603 	if (vp->v_type == VDIR) {
1604 		error = EPERM;
1605 		goto out;
1606 	}
1607 
1608 	vnevent_remove(vp, dvp, name, ct);
1609 
1610 	if (realnmp)
1611 		dnlc_remove(dvp, realnmp->pn_buf);
1612 	else
1613 		dnlc_remove(dvp, name);
1614 
1615 	mutex_enter(&vp->v_lock);
1616 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1617 	mutex_exit(&vp->v_lock);
1618 
1619 	/*
1620 	 * We may delete the znode now, or we may put it in the unlinked set;
1621 	 * it depends on whether we're the last link, and on whether there are
1622 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1623 	 * allow for either case.
1624 	 */
1625 	obj = zp->z_id;
1626 	tx = dmu_tx_create(zfsvfs->z_os);
1627 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1628 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1629 	zfs_sa_upgrade_txholds(tx, zp);
1630 	zfs_sa_upgrade_txholds(tx, dzp);
1631 	if (may_delete_now) {
1632 		toobig =
1633 		    zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1634 		/* if the file is too big, only hold_free a token amount */
1635 		dmu_tx_hold_free(tx, zp->z_id, 0,
1636 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1637 	}
1638 
1639 	/* are there any extended attributes? */
1640 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1641 	    &xattr_obj, sizeof (xattr_obj));
1642 	if (error == 0 && xattr_obj) {
1643 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1644 		ASSERT3U(error, ==, 0);
1645 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1646 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1647 	}
1648 
1649 	mutex_enter(&zp->z_lock);
1650 	if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1651 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1652 	mutex_exit(&zp->z_lock);
1653 
1654 	/* charge as an update -- would be nice not to charge at all */
1655 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1656 
1657 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1658 	if (error) {
1659 		zfs_dirent_unlock(dl);
1660 		VN_RELE(vp);
1661 		if (xzp)
1662 			VN_RELE(ZTOV(xzp));
1663 		if (error == ERESTART) {
1664 			dmu_tx_wait(tx);
1665 			dmu_tx_abort(tx);
1666 			goto top;
1667 		}
1668 		if (realnmp)
1669 			pn_free(realnmp);
1670 		dmu_tx_abort(tx);
1671 		ZFS_EXIT(zfsvfs);
1672 		return (error);
1673 	}
1674 
1675 	/*
1676 	 * Remove the directory entry.
1677 	 */
1678 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1679 
1680 	if (error) {
1681 		dmu_tx_commit(tx);
1682 		goto out;
1683 	}
1684 
1685 	if (unlinked) {
1686 
1687 		/*
1688 		 * Hold z_lock so that we can make sure that the ACL obj
1689 		 * hasn't changed.  Could have been deleted due to
1690 		 * zfs_sa_upgrade().
1691 		 */
1692 		mutex_enter(&zp->z_lock);
1693 		mutex_enter(&vp->v_lock);
1694 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1695 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1696 		delete_now = may_delete_now && !toobig &&
1697 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1698 		    xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1699 		    acl_obj;
1700 		mutex_exit(&vp->v_lock);
1701 	}
1702 
1703 	if (delete_now) {
1704 		if (xattr_obj_unlinked) {
1705 			ASSERT3U(xzp->z_links, ==, 2);
1706 			mutex_enter(&xzp->z_lock);
1707 			xzp->z_unlinked = 1;
1708 			xzp->z_links = 0;
1709 			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1710 			    &xzp->z_links, sizeof (xzp->z_links), tx);
1711 			ASSERT3U(error,  ==,  0);
1712 			mutex_exit(&xzp->z_lock);
1713 			zfs_unlinked_add(xzp, tx);
1714 
1715 			if (zp->z_is_sa)
1716 				error = sa_remove(zp->z_sa_hdl,
1717 				    SA_ZPL_XATTR(zfsvfs), tx);
1718 			else
1719 				error = sa_update(zp->z_sa_hdl,
1720 				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
1721 				    sizeof (uint64_t), tx);
1722 			ASSERT3U(error, ==, 0);
1723 		}
1724 		mutex_enter(&vp->v_lock);
1725 		vp->v_count--;
1726 		ASSERT3U(vp->v_count, ==, 0);
1727 		mutex_exit(&vp->v_lock);
1728 		mutex_exit(&zp->z_lock);
1729 		zfs_znode_delete(zp, tx);
1730 	} else if (unlinked) {
1731 		mutex_exit(&zp->z_lock);
1732 		zfs_unlinked_add(zp, tx);
1733 	}
1734 
1735 	txtype = TX_REMOVE;
1736 	if (flags & FIGNORECASE)
1737 		txtype |= TX_CI;
1738 	zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1739 
1740 	dmu_tx_commit(tx);
1741 out:
1742 	if (realnmp)
1743 		pn_free(realnmp);
1744 
1745 	zfs_dirent_unlock(dl);
1746 
1747 	if (!delete_now)
1748 		VN_RELE(vp);
1749 	if (xzp)
1750 		VN_RELE(ZTOV(xzp));
1751 
1752 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1753 		zil_commit(zilog, 0);
1754 
1755 	ZFS_EXIT(zfsvfs);
1756 	return (error);
1757 }
1758 
1759 /*
1760  * Create a new directory and insert it into dvp using the name
1761  * provided.  Return a pointer to the inserted directory.
1762  *
1763  *	IN:	dvp	- vnode of directory to add subdir to.
1764  *		dirname	- name of new directory.
1765  *		vap	- attributes of new directory.
1766  *		cr	- credentials of caller.
1767  *		ct	- caller context
1768  *		vsecp	- ACL to be set
1769  *
1770  *	OUT:	vpp	- vnode of created directory.
1771  *
1772  *	RETURN:	0 if success
1773  *		error code if failure
1774  *
1775  * Timestamps:
1776  *	dvp - ctime|mtime updated
1777  *	 vp - ctime|mtime|atime updated
1778  */
1779 /*ARGSUSED*/
1780 static int
1781 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1782     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1783 {
1784 	znode_t		*zp, *dzp = VTOZ(dvp);
1785 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1786 	zilog_t		*zilog;
1787 	zfs_dirlock_t	*dl;
1788 	uint64_t	txtype;
1789 	dmu_tx_t	*tx;
1790 	int		error;
1791 	int		zf = ZNEW;
1792 	ksid_t		*ksid;
1793 	uid_t		uid;
1794 	gid_t		gid = crgetgid(cr);
1795 	zfs_acl_ids_t   acl_ids;
1796 	boolean_t	fuid_dirtied;
1797 
1798 	ASSERT(vap->va_type == VDIR);
1799 
1800 	/*
1801 	 * If we have an ephemeral id, ACL, or XVATTR then
1802 	 * make sure file system is at proper version
1803 	 */
1804 
1805 	ksid = crgetsid(cr, KSID_OWNER);
1806 	if (ksid)
1807 		uid = ksid_getid(ksid);
1808 	else
1809 		uid = crgetuid(cr);
1810 	if (zfsvfs->z_use_fuids == B_FALSE &&
1811 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1812 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1813 		return (EINVAL);
1814 
1815 	ZFS_ENTER(zfsvfs);
1816 	ZFS_VERIFY_ZP(dzp);
1817 	zilog = zfsvfs->z_log;
1818 
1819 	if (dzp->z_pflags & ZFS_XATTR) {
1820 		ZFS_EXIT(zfsvfs);
1821 		return (EINVAL);
1822 	}
1823 
1824 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1825 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1826 		ZFS_EXIT(zfsvfs);
1827 		return (EILSEQ);
1828 	}
1829 	if (flags & FIGNORECASE)
1830 		zf |= ZCILOOK;
1831 
1832 	if (vap->va_mask & AT_XVATTR) {
1833 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1834 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1835 			ZFS_EXIT(zfsvfs);
1836 			return (error);
1837 		}
1838 	}
1839 
1840 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1841 	    vsecp, &acl_ids)) != 0) {
1842 		ZFS_EXIT(zfsvfs);
1843 		return (error);
1844 	}
1845 	/*
1846 	 * First make sure the new directory doesn't exist.
1847 	 *
1848 	 * Existence is checked first to make sure we don't return
1849 	 * EACCES instead of EEXIST which can cause some applications
1850 	 * to fail.
1851 	 */
1852 top:
1853 	*vpp = NULL;
1854 
1855 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1856 	    NULL, NULL)) {
1857 		zfs_acl_ids_free(&acl_ids);
1858 		ZFS_EXIT(zfsvfs);
1859 		return (error);
1860 	}
1861 
1862 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1863 		zfs_acl_ids_free(&acl_ids);
1864 		zfs_dirent_unlock(dl);
1865 		ZFS_EXIT(zfsvfs);
1866 		return (error);
1867 	}
1868 
1869 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1870 		zfs_acl_ids_free(&acl_ids);
1871 		zfs_dirent_unlock(dl);
1872 		ZFS_EXIT(zfsvfs);
1873 		return (EDQUOT);
1874 	}
1875 
1876 	/*
1877 	 * Add a new entry to the directory.
1878 	 */
1879 	tx = dmu_tx_create(zfsvfs->z_os);
1880 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1881 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1882 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1883 	if (fuid_dirtied)
1884 		zfs_fuid_txhold(zfsvfs, tx);
1885 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1886 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1887 		    acl_ids.z_aclp->z_acl_bytes);
1888 	}
1889 
1890 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1891 	    ZFS_SA_BASE_ATTR_SIZE);
1892 
1893 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1894 	if (error) {
1895 		zfs_dirent_unlock(dl);
1896 		if (error == ERESTART) {
1897 			dmu_tx_wait(tx);
1898 			dmu_tx_abort(tx);
1899 			goto top;
1900 		}
1901 		zfs_acl_ids_free(&acl_ids);
1902 		dmu_tx_abort(tx);
1903 		ZFS_EXIT(zfsvfs);
1904 		return (error);
1905 	}
1906 
1907 	/*
1908 	 * Create new node.
1909 	 */
1910 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1911 
1912 	if (fuid_dirtied)
1913 		zfs_fuid_sync(zfsvfs, tx);
1914 
1915 	/*
1916 	 * Now put new name in parent dir.
1917 	 */
1918 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1919 
1920 	*vpp = ZTOV(zp);
1921 
1922 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1923 	if (flags & FIGNORECASE)
1924 		txtype |= TX_CI;
1925 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1926 	    acl_ids.z_fuidp, vap);
1927 
1928 	zfs_acl_ids_free(&acl_ids);
1929 
1930 	dmu_tx_commit(tx);
1931 
1932 	zfs_dirent_unlock(dl);
1933 
1934 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1935 		zil_commit(zilog, 0);
1936 
1937 	ZFS_EXIT(zfsvfs);
1938 	return (0);
1939 }
1940 
1941 /*
1942  * Remove a directory subdir entry.  If the current working
1943  * directory is the same as the subdir to be removed, the
1944  * remove will fail.
1945  *
1946  *	IN:	dvp	- vnode of directory to remove from.
1947  *		name	- name of directory to be removed.
1948  *		cwd	- vnode of current working directory.
1949  *		cr	- credentials of caller.
1950  *		ct	- caller context
1951  *		flags	- case flags
1952  *
1953  *	RETURN:	0 if success
1954  *		error code if failure
1955  *
1956  * Timestamps:
1957  *	dvp - ctime|mtime updated
1958  */
1959 /*ARGSUSED*/
1960 static int
1961 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1962     caller_context_t *ct, int flags)
1963 {
1964 	znode_t		*dzp = VTOZ(dvp);
1965 	znode_t		*zp;
1966 	vnode_t		*vp;
1967 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1968 	zilog_t		*zilog;
1969 	zfs_dirlock_t	*dl;
1970 	dmu_tx_t	*tx;
1971 	int		error;
1972 	int		zflg = ZEXISTS;
1973 
1974 	ZFS_ENTER(zfsvfs);
1975 	ZFS_VERIFY_ZP(dzp);
1976 	zilog = zfsvfs->z_log;
1977 
1978 	if (flags & FIGNORECASE)
1979 		zflg |= ZCILOOK;
1980 top:
1981 	zp = NULL;
1982 
1983 	/*
1984 	 * Attempt to lock directory; fail if entry doesn't exist.
1985 	 */
1986 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1987 	    NULL, NULL)) {
1988 		ZFS_EXIT(zfsvfs);
1989 		return (error);
1990 	}
1991 
1992 	vp = ZTOV(zp);
1993 
1994 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1995 		goto out;
1996 	}
1997 
1998 	if (vp->v_type != VDIR) {
1999 		error = ENOTDIR;
2000 		goto out;
2001 	}
2002 
2003 	if (vp == cwd) {
2004 		error = EINVAL;
2005 		goto out;
2006 	}
2007 
2008 	vnevent_rmdir(vp, dvp, name, ct);
2009 
2010 	/*
2011 	 * Grab a lock on the directory to make sure that noone is
2012 	 * trying to add (or lookup) entries while we are removing it.
2013 	 */
2014 	rw_enter(&zp->z_name_lock, RW_WRITER);
2015 
2016 	/*
2017 	 * Grab a lock on the parent pointer to make sure we play well
2018 	 * with the treewalk and directory rename code.
2019 	 */
2020 	rw_enter(&zp->z_parent_lock, RW_WRITER);
2021 
2022 	tx = dmu_tx_create(zfsvfs->z_os);
2023 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2024 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2025 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2026 	zfs_sa_upgrade_txholds(tx, zp);
2027 	zfs_sa_upgrade_txholds(tx, dzp);
2028 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2029 	if (error) {
2030 		rw_exit(&zp->z_parent_lock);
2031 		rw_exit(&zp->z_name_lock);
2032 		zfs_dirent_unlock(dl);
2033 		VN_RELE(vp);
2034 		if (error == ERESTART) {
2035 			dmu_tx_wait(tx);
2036 			dmu_tx_abort(tx);
2037 			goto top;
2038 		}
2039 		dmu_tx_abort(tx);
2040 		ZFS_EXIT(zfsvfs);
2041 		return (error);
2042 	}
2043 
2044 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2045 
2046 	if (error == 0) {
2047 		uint64_t txtype = TX_RMDIR;
2048 		if (flags & FIGNORECASE)
2049 			txtype |= TX_CI;
2050 		zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2051 	}
2052 
2053 	dmu_tx_commit(tx);
2054 
2055 	rw_exit(&zp->z_parent_lock);
2056 	rw_exit(&zp->z_name_lock);
2057 out:
2058 	zfs_dirent_unlock(dl);
2059 
2060 	VN_RELE(vp);
2061 
2062 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2063 		zil_commit(zilog, 0);
2064 
2065 	ZFS_EXIT(zfsvfs);
2066 	return (error);
2067 }
2068 
2069 /*
2070  * Read as many directory entries as will fit into the provided
2071  * buffer from the given directory cursor position (specified in
2072  * the uio structure.
2073  *
2074  *	IN:	vp	- vnode of directory to read.
2075  *		uio	- structure supplying read location, range info,
2076  *			  and return buffer.
2077  *		cr	- credentials of caller.
2078  *		ct	- caller context
2079  *		flags	- case flags
2080  *
2081  *	OUT:	uio	- updated offset and range, buffer filled.
2082  *		eofp	- set to true if end-of-file detected.
2083  *
2084  *	RETURN:	0 if success
2085  *		error code if failure
2086  *
2087  * Timestamps:
2088  *	vp - atime updated
2089  *
2090  * Note that the low 4 bits of the cookie returned by zap is always zero.
2091  * This allows us to use the low range for "special" directory entries:
2092  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2093  * we use the offset 2 for the '.zfs' directory.
2094  */
2095 /* ARGSUSED */
2096 static int
2097 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2098     caller_context_t *ct, int flags)
2099 {
2100 	znode_t		*zp = VTOZ(vp);
2101 	iovec_t		*iovp;
2102 	edirent_t	*eodp;
2103 	dirent64_t	*odp;
2104 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2105 	objset_t	*os;
2106 	caddr_t		outbuf;
2107 	size_t		bufsize;
2108 	zap_cursor_t	zc;
2109 	zap_attribute_t	zap;
2110 	uint_t		bytes_wanted;
2111 	uint64_t	offset; /* must be unsigned; checks for < 1 */
2112 	uint64_t	parent;
2113 	int		local_eof;
2114 	int		outcount;
2115 	int		error;
2116 	uint8_t		prefetch;
2117 	boolean_t	check_sysattrs;
2118 
2119 	ZFS_ENTER(zfsvfs);
2120 	ZFS_VERIFY_ZP(zp);
2121 
2122 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2123 	    &parent, sizeof (parent))) != 0) {
2124 		ZFS_EXIT(zfsvfs);
2125 		return (error);
2126 	}
2127 
2128 	/*
2129 	 * If we are not given an eof variable,
2130 	 * use a local one.
2131 	 */
2132 	if (eofp == NULL)
2133 		eofp = &local_eof;
2134 
2135 	/*
2136 	 * Check for valid iov_len.
2137 	 */
2138 	if (uio->uio_iov->iov_len <= 0) {
2139 		ZFS_EXIT(zfsvfs);
2140 		return (EINVAL);
2141 	}
2142 
2143 	/*
2144 	 * Quit if directory has been removed (posix)
2145 	 */
2146 	if ((*eofp = zp->z_unlinked) != 0) {
2147 		ZFS_EXIT(zfsvfs);
2148 		return (0);
2149 	}
2150 
2151 	error = 0;
2152 	os = zfsvfs->z_os;
2153 	offset = uio->uio_loffset;
2154 	prefetch = zp->z_zn_prefetch;
2155 
2156 	/*
2157 	 * Initialize the iterator cursor.
2158 	 */
2159 	if (offset <= 3) {
2160 		/*
2161 		 * Start iteration from the beginning of the directory.
2162 		 */
2163 		zap_cursor_init(&zc, os, zp->z_id);
2164 	} else {
2165 		/*
2166 		 * The offset is a serialized cursor.
2167 		 */
2168 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2169 	}
2170 
2171 	/*
2172 	 * Get space to change directory entries into fs independent format.
2173 	 */
2174 	iovp = uio->uio_iov;
2175 	bytes_wanted = iovp->iov_len;
2176 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2177 		bufsize = bytes_wanted;
2178 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
2179 		odp = (struct dirent64 *)outbuf;
2180 	} else {
2181 		bufsize = bytes_wanted;
2182 		odp = (struct dirent64 *)iovp->iov_base;
2183 	}
2184 	eodp = (struct edirent *)odp;
2185 
2186 	/*
2187 	 * If this VFS supports the system attribute view interface; and
2188 	 * we're looking at an extended attribute directory; and we care
2189 	 * about normalization conflicts on this vfs; then we must check
2190 	 * for normalization conflicts with the sysattr name space.
2191 	 */
2192 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2193 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2194 	    (flags & V_RDDIR_ENTFLAGS);
2195 
2196 	/*
2197 	 * Transform to file-system independent format
2198 	 */
2199 	outcount = 0;
2200 	while (outcount < bytes_wanted) {
2201 		ino64_t objnum;
2202 		ushort_t reclen;
2203 		off64_t *next = NULL;
2204 
2205 		/*
2206 		 * Special case `.', `..', and `.zfs'.
2207 		 */
2208 		if (offset == 0) {
2209 			(void) strcpy(zap.za_name, ".");
2210 			zap.za_normalization_conflict = 0;
2211 			objnum = zp->z_id;
2212 		} else if (offset == 1) {
2213 			(void) strcpy(zap.za_name, "..");
2214 			zap.za_normalization_conflict = 0;
2215 			objnum = parent;
2216 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2217 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2218 			zap.za_normalization_conflict = 0;
2219 			objnum = ZFSCTL_INO_ROOT;
2220 		} else {
2221 			/*
2222 			 * Grab next entry.
2223 			 */
2224 			if (error = zap_cursor_retrieve(&zc, &zap)) {
2225 				if ((*eofp = (error == ENOENT)) != 0)
2226 					break;
2227 				else
2228 					goto update;
2229 			}
2230 
2231 			if (zap.za_integer_length != 8 ||
2232 			    zap.za_num_integers != 1) {
2233 				cmn_err(CE_WARN, "zap_readdir: bad directory "
2234 				    "entry, obj = %lld, offset = %lld\n",
2235 				    (u_longlong_t)zp->z_id,
2236 				    (u_longlong_t)offset);
2237 				error = ENXIO;
2238 				goto update;
2239 			}
2240 
2241 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2242 			/*
2243 			 * MacOS X can extract the object type here such as:
2244 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2245 			 */
2246 
2247 			if (check_sysattrs && !zap.za_normalization_conflict) {
2248 				zap.za_normalization_conflict =
2249 				    xattr_sysattr_casechk(zap.za_name);
2250 			}
2251 		}
2252 
2253 		if (flags & V_RDDIR_ACCFILTER) {
2254 			/*
2255 			 * If we have no access at all, don't include
2256 			 * this entry in the returned information
2257 			 */
2258 			znode_t	*ezp;
2259 			if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2260 				goto skip_entry;
2261 			if (!zfs_has_access(ezp, cr)) {
2262 				VN_RELE(ZTOV(ezp));
2263 				goto skip_entry;
2264 			}
2265 			VN_RELE(ZTOV(ezp));
2266 		}
2267 
2268 		if (flags & V_RDDIR_ENTFLAGS)
2269 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2270 		else
2271 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2272 
2273 		/*
2274 		 * Will this entry fit in the buffer?
2275 		 */
2276 		if (outcount + reclen > bufsize) {
2277 			/*
2278 			 * Did we manage to fit anything in the buffer?
2279 			 */
2280 			if (!outcount) {
2281 				error = EINVAL;
2282 				goto update;
2283 			}
2284 			break;
2285 		}
2286 		if (flags & V_RDDIR_ENTFLAGS) {
2287 			/*
2288 			 * Add extended flag entry:
2289 			 */
2290 			eodp->ed_ino = objnum;
2291 			eodp->ed_reclen = reclen;
2292 			/* NOTE: ed_off is the offset for the *next* entry */
2293 			next = &(eodp->ed_off);
2294 			eodp->ed_eflags = zap.za_normalization_conflict ?
2295 			    ED_CASE_CONFLICT : 0;
2296 			(void) strncpy(eodp->ed_name, zap.za_name,
2297 			    EDIRENT_NAMELEN(reclen));
2298 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2299 		} else {
2300 			/*
2301 			 * Add normal entry:
2302 			 */
2303 			odp->d_ino = objnum;
2304 			odp->d_reclen = reclen;
2305 			/* NOTE: d_off is the offset for the *next* entry */
2306 			next = &(odp->d_off);
2307 			(void) strncpy(odp->d_name, zap.za_name,
2308 			    DIRENT64_NAMELEN(reclen));
2309 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2310 		}
2311 		outcount += reclen;
2312 
2313 		ASSERT(outcount <= bufsize);
2314 
2315 		/* Prefetch znode */
2316 		if (prefetch)
2317 			dmu_prefetch(os, objnum, 0, 0);
2318 
2319 	skip_entry:
2320 		/*
2321 		 * Move to the next entry, fill in the previous offset.
2322 		 */
2323 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2324 			zap_cursor_advance(&zc);
2325 			offset = zap_cursor_serialize(&zc);
2326 		} else {
2327 			offset += 1;
2328 		}
2329 		if (next)
2330 			*next = offset;
2331 	}
2332 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2333 
2334 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2335 		iovp->iov_base += outcount;
2336 		iovp->iov_len -= outcount;
2337 		uio->uio_resid -= outcount;
2338 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2339 		/*
2340 		 * Reset the pointer.
2341 		 */
2342 		offset = uio->uio_loffset;
2343 	}
2344 
2345 update:
2346 	zap_cursor_fini(&zc);
2347 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2348 		kmem_free(outbuf, bufsize);
2349 
2350 	if (error == ENOENT)
2351 		error = 0;
2352 
2353 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2354 
2355 	uio->uio_loffset = offset;
2356 	ZFS_EXIT(zfsvfs);
2357 	return (error);
2358 }
2359 
2360 ulong_t zfs_fsync_sync_cnt = 4;
2361 
2362 static int
2363 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2364 {
2365 	znode_t	*zp = VTOZ(vp);
2366 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2367 
2368 	/*
2369 	 * Regardless of whether this is required for standards conformance,
2370 	 * this is the logical behavior when fsync() is called on a file with
2371 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2372 	 * going to be pushed out as part of the zil_commit().
2373 	 */
2374 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2375 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2376 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2377 
2378 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2379 
2380 	if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2381 		ZFS_ENTER(zfsvfs);
2382 		ZFS_VERIFY_ZP(zp);
2383 		zil_commit(zfsvfs->z_log, zp->z_id);
2384 		ZFS_EXIT(zfsvfs);
2385 	}
2386 	return (0);
2387 }
2388 
2389 
2390 /*
2391  * Get the requested file attributes and place them in the provided
2392  * vattr structure.
2393  *
2394  *	IN:	vp	- vnode of file.
2395  *		vap	- va_mask identifies requested attributes.
2396  *			  If AT_XVATTR set, then optional attrs are requested
2397  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2398  *		cr	- credentials of caller.
2399  *		ct	- caller context
2400  *
2401  *	OUT:	vap	- attribute values.
2402  *
2403  *	RETURN:	0 (always succeeds)
2404  */
2405 /* ARGSUSED */
2406 static int
2407 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2408     caller_context_t *ct)
2409 {
2410 	znode_t *zp = VTOZ(vp);
2411 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2412 	int	error = 0;
2413 	uint64_t links;
2414 	uint64_t mtime[2], ctime[2];
2415 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2416 	xoptattr_t *xoap = NULL;
2417 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2418 	sa_bulk_attr_t bulk[2];
2419 	int count = 0;
2420 
2421 	ZFS_ENTER(zfsvfs);
2422 	ZFS_VERIFY_ZP(zp);
2423 
2424 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2425 
2426 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2427 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2428 
2429 	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2430 		ZFS_EXIT(zfsvfs);
2431 		return (error);
2432 	}
2433 
2434 	/*
2435 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2436 	 * Also, if we are the owner don't bother, since owner should
2437 	 * always be allowed to read basic attributes of file.
2438 	 */
2439 	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2440 	    (vap->va_uid != crgetuid(cr))) {
2441 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2442 		    skipaclchk, cr)) {
2443 			ZFS_EXIT(zfsvfs);
2444 			return (error);
2445 		}
2446 	}
2447 
2448 	/*
2449 	 * Return all attributes.  It's cheaper to provide the answer
2450 	 * than to determine whether we were asked the question.
2451 	 */
2452 
2453 	mutex_enter(&zp->z_lock);
2454 	vap->va_type = vp->v_type;
2455 	vap->va_mode = zp->z_mode & MODEMASK;
2456 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2457 	vap->va_nodeid = zp->z_id;
2458 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2459 		links = zp->z_links + 1;
2460 	else
2461 		links = zp->z_links;
2462 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2463 	vap->va_size = zp->z_size;
2464 	vap->va_rdev = vp->v_rdev;
2465 	vap->va_seq = zp->z_seq;
2466 
2467 	/*
2468 	 * Add in any requested optional attributes and the create time.
2469 	 * Also set the corresponding bits in the returned attribute bitmap.
2470 	 */
2471 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2472 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2473 			xoap->xoa_archive =
2474 			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2475 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2476 		}
2477 
2478 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2479 			xoap->xoa_readonly =
2480 			    ((zp->z_pflags & ZFS_READONLY) != 0);
2481 			XVA_SET_RTN(xvap, XAT_READONLY);
2482 		}
2483 
2484 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2485 			xoap->xoa_system =
2486 			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2487 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2488 		}
2489 
2490 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2491 			xoap->xoa_hidden =
2492 			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2493 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2494 		}
2495 
2496 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2497 			xoap->xoa_nounlink =
2498 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2499 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2500 		}
2501 
2502 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2503 			xoap->xoa_immutable =
2504 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2505 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2506 		}
2507 
2508 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2509 			xoap->xoa_appendonly =
2510 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2511 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2512 		}
2513 
2514 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2515 			xoap->xoa_nodump =
2516 			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2517 			XVA_SET_RTN(xvap, XAT_NODUMP);
2518 		}
2519 
2520 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2521 			xoap->xoa_opaque =
2522 			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2523 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2524 		}
2525 
2526 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2527 			xoap->xoa_av_quarantined =
2528 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2529 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2530 		}
2531 
2532 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2533 			xoap->xoa_av_modified =
2534 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2535 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2536 		}
2537 
2538 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2539 		    vp->v_type == VREG) {
2540 			zfs_sa_get_scanstamp(zp, xvap);
2541 		}
2542 
2543 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2544 			uint64_t times[2];
2545 
2546 			(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2547 			    times, sizeof (times));
2548 			ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2549 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2550 		}
2551 
2552 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2553 			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2554 			XVA_SET_RTN(xvap, XAT_REPARSE);
2555 		}
2556 		if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2557 			xoap->xoa_generation = zp->z_gen;
2558 			XVA_SET_RTN(xvap, XAT_GEN);
2559 		}
2560 
2561 		if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2562 			xoap->xoa_offline =
2563 			    ((zp->z_pflags & ZFS_OFFLINE) != 0);
2564 			XVA_SET_RTN(xvap, XAT_OFFLINE);
2565 		}
2566 
2567 		if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2568 			xoap->xoa_sparse =
2569 			    ((zp->z_pflags & ZFS_SPARSE) != 0);
2570 			XVA_SET_RTN(xvap, XAT_SPARSE);
2571 		}
2572 	}
2573 
2574 	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2575 	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2576 	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2577 
2578 	mutex_exit(&zp->z_lock);
2579 
2580 	sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2581 
2582 	if (zp->z_blksz == 0) {
2583 		/*
2584 		 * Block size hasn't been set; suggest maximal I/O transfers.
2585 		 */
2586 		vap->va_blksize = zfsvfs->z_max_blksz;
2587 	}
2588 
2589 	ZFS_EXIT(zfsvfs);
2590 	return (0);
2591 }
2592 
2593 /*
2594  * Set the file attributes to the values contained in the
2595  * vattr structure.
2596  *
2597  *	IN:	vp	- vnode of file to be modified.
2598  *		vap	- new attribute values.
2599  *			  If AT_XVATTR set, then optional attrs are being set
2600  *		flags	- ATTR_UTIME set if non-default time values provided.
2601  *			- ATTR_NOACLCHECK (CIFS context only).
2602  *		cr	- credentials of caller.
2603  *		ct	- caller context
2604  *
2605  *	RETURN:	0 if success
2606  *		error code if failure
2607  *
2608  * Timestamps:
2609  *	vp - ctime updated, mtime updated if size changed.
2610  */
2611 /* ARGSUSED */
2612 static int
2613 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2614 	caller_context_t *ct)
2615 {
2616 	znode_t		*zp = VTOZ(vp);
2617 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2618 	zilog_t		*zilog;
2619 	dmu_tx_t	*tx;
2620 	vattr_t		oldva;
2621 	xvattr_t	tmpxvattr;
2622 	uint_t		mask = vap->va_mask;
2623 	uint_t		saved_mask;
2624 	int		trim_mask = 0;
2625 	uint64_t	new_mode;
2626 	uint64_t	new_uid, new_gid;
2627 	uint64_t	xattr_obj;
2628 	uint64_t	mtime[2], ctime[2];
2629 	znode_t		*attrzp;
2630 	int		need_policy = FALSE;
2631 	int		err, err2;
2632 	zfs_fuid_info_t *fuidp = NULL;
2633 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2634 	xoptattr_t	*xoap;
2635 	zfs_acl_t	*aclp;
2636 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2637 	boolean_t	fuid_dirtied = B_FALSE;
2638 	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
2639 	int		count = 0, xattr_count = 0;
2640 
2641 	if (mask == 0)
2642 		return (0);
2643 
2644 	if (mask & AT_NOSET)
2645 		return (EINVAL);
2646 
2647 	ZFS_ENTER(zfsvfs);
2648 	ZFS_VERIFY_ZP(zp);
2649 
2650 	zilog = zfsvfs->z_log;
2651 
2652 	/*
2653 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2654 	 * that file system is at proper version level
2655 	 */
2656 
2657 	if (zfsvfs->z_use_fuids == B_FALSE &&
2658 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2659 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2660 	    (mask & AT_XVATTR))) {
2661 		ZFS_EXIT(zfsvfs);
2662 		return (EINVAL);
2663 	}
2664 
2665 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2666 		ZFS_EXIT(zfsvfs);
2667 		return (EISDIR);
2668 	}
2669 
2670 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2671 		ZFS_EXIT(zfsvfs);
2672 		return (EINVAL);
2673 	}
2674 
2675 	/*
2676 	 * If this is an xvattr_t, then get a pointer to the structure of
2677 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2678 	 */
2679 	xoap = xva_getxoptattr(xvap);
2680 
2681 	xva_init(&tmpxvattr);
2682 
2683 	/*
2684 	 * Immutable files can only alter immutable bit and atime
2685 	 */
2686 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2687 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2688 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2689 		ZFS_EXIT(zfsvfs);
2690 		return (EPERM);
2691 	}
2692 
2693 	if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2694 		ZFS_EXIT(zfsvfs);
2695 		return (EPERM);
2696 	}
2697 
2698 	/*
2699 	 * Verify timestamps doesn't overflow 32 bits.
2700 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2701 	 * handle times greater than 2039.  This check should be removed
2702 	 * once large timestamps are fully supported.
2703 	 */
2704 	if (mask & (AT_ATIME | AT_MTIME)) {
2705 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2706 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2707 			ZFS_EXIT(zfsvfs);
2708 			return (EOVERFLOW);
2709 		}
2710 	}
2711 
2712 top:
2713 	attrzp = NULL;
2714 	aclp = NULL;
2715 
2716 	/* Can this be moved to before the top label? */
2717 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2718 		ZFS_EXIT(zfsvfs);
2719 		return (EROFS);
2720 	}
2721 
2722 	/*
2723 	 * First validate permissions
2724 	 */
2725 
2726 	if (mask & AT_SIZE) {
2727 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2728 		if (err) {
2729 			ZFS_EXIT(zfsvfs);
2730 			return (err);
2731 		}
2732 		/*
2733 		 * XXX - Note, we are not providing any open
2734 		 * mode flags here (like FNDELAY), so we may
2735 		 * block if there are locks present... this
2736 		 * should be addressed in openat().
2737 		 */
2738 		/* XXX - would it be OK to generate a log record here? */
2739 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2740 		if (err) {
2741 			ZFS_EXIT(zfsvfs);
2742 			return (err);
2743 		}
2744 	}
2745 
2746 	if (mask & (AT_ATIME|AT_MTIME) ||
2747 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2748 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2749 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2750 	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2751 	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2752 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2753 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2754 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2755 		    skipaclchk, cr);
2756 	}
2757 
2758 	if (mask & (AT_UID|AT_GID)) {
2759 		int	idmask = (mask & (AT_UID|AT_GID));
2760 		int	take_owner;
2761 		int	take_group;
2762 
2763 		/*
2764 		 * NOTE: even if a new mode is being set,
2765 		 * we may clear S_ISUID/S_ISGID bits.
2766 		 */
2767 
2768 		if (!(mask & AT_MODE))
2769 			vap->va_mode = zp->z_mode;
2770 
2771 		/*
2772 		 * Take ownership or chgrp to group we are a member of
2773 		 */
2774 
2775 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2776 		take_group = (mask & AT_GID) &&
2777 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2778 
2779 		/*
2780 		 * If both AT_UID and AT_GID are set then take_owner and
2781 		 * take_group must both be set in order to allow taking
2782 		 * ownership.
2783 		 *
2784 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2785 		 *
2786 		 */
2787 
2788 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2789 		    ((idmask == AT_UID) && take_owner) ||
2790 		    ((idmask == AT_GID) && take_group)) {
2791 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2792 			    skipaclchk, cr) == 0) {
2793 				/*
2794 				 * Remove setuid/setgid for non-privileged users
2795 				 */
2796 				secpolicy_setid_clear(vap, cr);
2797 				trim_mask = (mask & (AT_UID|AT_GID));
2798 			} else {
2799 				need_policy =  TRUE;
2800 			}
2801 		} else {
2802 			need_policy =  TRUE;
2803 		}
2804 	}
2805 
2806 	mutex_enter(&zp->z_lock);
2807 	oldva.va_mode = zp->z_mode;
2808 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2809 	if (mask & AT_XVATTR) {
2810 		/*
2811 		 * Update xvattr mask to include only those attributes
2812 		 * that are actually changing.
2813 		 *
2814 		 * the bits will be restored prior to actually setting
2815 		 * the attributes so the caller thinks they were set.
2816 		 */
2817 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2818 			if (xoap->xoa_appendonly !=
2819 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2820 				need_policy = TRUE;
2821 			} else {
2822 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2823 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2824 			}
2825 		}
2826 
2827 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2828 			if (xoap->xoa_nounlink !=
2829 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2830 				need_policy = TRUE;
2831 			} else {
2832 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2833 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2834 			}
2835 		}
2836 
2837 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2838 			if (xoap->xoa_immutable !=
2839 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2840 				need_policy = TRUE;
2841 			} else {
2842 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2843 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2844 			}
2845 		}
2846 
2847 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2848 			if (xoap->xoa_nodump !=
2849 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2850 				need_policy = TRUE;
2851 			} else {
2852 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2853 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2854 			}
2855 		}
2856 
2857 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2858 			if (xoap->xoa_av_modified !=
2859 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2860 				need_policy = TRUE;
2861 			} else {
2862 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2863 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2864 			}
2865 		}
2866 
2867 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2868 			if ((vp->v_type != VREG &&
2869 			    xoap->xoa_av_quarantined) ||
2870 			    xoap->xoa_av_quarantined !=
2871 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2872 				need_policy = TRUE;
2873 			} else {
2874 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2875 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2876 			}
2877 		}
2878 
2879 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2880 			mutex_exit(&zp->z_lock);
2881 			ZFS_EXIT(zfsvfs);
2882 			return (EPERM);
2883 		}
2884 
2885 		if (need_policy == FALSE &&
2886 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2887 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2888 			need_policy = TRUE;
2889 		}
2890 	}
2891 
2892 	mutex_exit(&zp->z_lock);
2893 
2894 	if (mask & AT_MODE) {
2895 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2896 			err = secpolicy_setid_setsticky_clear(vp, vap,
2897 			    &oldva, cr);
2898 			if (err) {
2899 				ZFS_EXIT(zfsvfs);
2900 				return (err);
2901 			}
2902 			trim_mask |= AT_MODE;
2903 		} else {
2904 			need_policy = TRUE;
2905 		}
2906 	}
2907 
2908 	if (need_policy) {
2909 		/*
2910 		 * If trim_mask is set then take ownership
2911 		 * has been granted or write_acl is present and user
2912 		 * has the ability to modify mode.  In that case remove
2913 		 * UID|GID and or MODE from mask so that
2914 		 * secpolicy_vnode_setattr() doesn't revoke it.
2915 		 */
2916 
2917 		if (trim_mask) {
2918 			saved_mask = vap->va_mask;
2919 			vap->va_mask &= ~trim_mask;
2920 		}
2921 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2922 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2923 		if (err) {
2924 			ZFS_EXIT(zfsvfs);
2925 			return (err);
2926 		}
2927 
2928 		if (trim_mask)
2929 			vap->va_mask |= saved_mask;
2930 	}
2931 
2932 	/*
2933 	 * secpolicy_vnode_setattr, or take ownership may have
2934 	 * changed va_mask
2935 	 */
2936 	mask = vap->va_mask;
2937 
2938 	if ((mask & (AT_UID | AT_GID))) {
2939 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2940 		    &xattr_obj, sizeof (xattr_obj));
2941 
2942 		if (err == 0 && xattr_obj) {
2943 			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
2944 			if (err)
2945 				goto out2;
2946 		}
2947 		if (mask & AT_UID) {
2948 			new_uid = zfs_fuid_create(zfsvfs,
2949 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2950 			if (new_uid != zp->z_uid &&
2951 			    zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
2952 				if (attrzp)
2953 					VN_RELE(ZTOV(attrzp));
2954 				err = EDQUOT;
2955 				goto out2;
2956 			}
2957 		}
2958 
2959 		if (mask & AT_GID) {
2960 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
2961 			    cr, ZFS_GROUP, &fuidp);
2962 			if (new_gid != zp->z_gid &&
2963 			    zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
2964 				if (attrzp)
2965 					VN_RELE(ZTOV(attrzp));
2966 				err = EDQUOT;
2967 				goto out2;
2968 			}
2969 		}
2970 	}
2971 	tx = dmu_tx_create(zfsvfs->z_os);
2972 
2973 	if (mask & AT_MODE) {
2974 		uint64_t pmode = zp->z_mode;
2975 		uint64_t acl_obj;
2976 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2977 
2978 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
2979 			goto out;
2980 
2981 		mutex_enter(&zp->z_lock);
2982 		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2983 			/*
2984 			 * Are we upgrading ACL from old V0 format
2985 			 * to V1 format?
2986 			 */
2987 			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2988 			    zfs_znode_acl_version(zp) ==
2989 			    ZFS_ACL_VERSION_INITIAL) {
2990 				dmu_tx_hold_free(tx, acl_obj, 0,
2991 				    DMU_OBJECT_END);
2992 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2993 				    0, aclp->z_acl_bytes);
2994 			} else {
2995 				dmu_tx_hold_write(tx, acl_obj, 0,
2996 				    aclp->z_acl_bytes);
2997 			}
2998 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2999 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3000 			    0, aclp->z_acl_bytes);
3001 		}
3002 		mutex_exit(&zp->z_lock);
3003 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3004 	} else {
3005 		if ((mask & AT_XVATTR) &&
3006 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3007 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3008 		else
3009 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3010 	}
3011 
3012 	if (attrzp) {
3013 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3014 	}
3015 
3016 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3017 	if (fuid_dirtied)
3018 		zfs_fuid_txhold(zfsvfs, tx);
3019 
3020 	zfs_sa_upgrade_txholds(tx, zp);
3021 
3022 	err = dmu_tx_assign(tx, TXG_NOWAIT);
3023 	if (err) {
3024 		if (err == ERESTART)
3025 			dmu_tx_wait(tx);
3026 		goto out;
3027 	}
3028 
3029 	count = 0;
3030 	/*
3031 	 * Set each attribute requested.
3032 	 * We group settings according to the locks they need to acquire.
3033 	 *
3034 	 * Note: you cannot set ctime directly, although it will be
3035 	 * updated as a side-effect of calling this function.
3036 	 */
3037 
3038 
3039 	if (mask & (AT_UID|AT_GID|AT_MODE))
3040 		mutex_enter(&zp->z_acl_lock);
3041 	mutex_enter(&zp->z_lock);
3042 
3043 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3044 	    &zp->z_pflags, sizeof (zp->z_pflags));
3045 
3046 	if (attrzp) {
3047 		if (mask & (AT_UID|AT_GID|AT_MODE))
3048 			mutex_enter(&attrzp->z_acl_lock);
3049 		mutex_enter(&attrzp->z_lock);
3050 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3051 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3052 		    sizeof (attrzp->z_pflags));
3053 	}
3054 
3055 	if (mask & (AT_UID|AT_GID)) {
3056 
3057 		if (mask & AT_UID) {
3058 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3059 			    &new_uid, sizeof (new_uid));
3060 			zp->z_uid = new_uid;
3061 			if (attrzp) {
3062 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3063 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3064 				    sizeof (new_uid));
3065 				attrzp->z_uid = new_uid;
3066 			}
3067 		}
3068 
3069 		if (mask & AT_GID) {
3070 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3071 			    NULL, &new_gid, sizeof (new_gid));
3072 			zp->z_gid = new_gid;
3073 			if (attrzp) {
3074 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3075 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3076 				    sizeof (new_gid));
3077 				attrzp->z_gid = new_gid;
3078 			}
3079 		}
3080 		if (!(mask & AT_MODE)) {
3081 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3082 			    NULL, &new_mode, sizeof (new_mode));
3083 			new_mode = zp->z_mode;
3084 		}
3085 		err = zfs_acl_chown_setattr(zp);
3086 		ASSERT(err == 0);
3087 		if (attrzp) {
3088 			err = zfs_acl_chown_setattr(attrzp);
3089 			ASSERT(err == 0);
3090 		}
3091 	}
3092 
3093 	if (mask & AT_MODE) {
3094 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3095 		    &new_mode, sizeof (new_mode));
3096 		zp->z_mode = new_mode;
3097 		ASSERT3U((uintptr_t)aclp, !=, NULL);
3098 		err = zfs_aclset_common(zp, aclp, cr, tx);
3099 		ASSERT3U(err, ==, 0);
3100 		if (zp->z_acl_cached)
3101 			zfs_acl_free(zp->z_acl_cached);
3102 		zp->z_acl_cached = aclp;
3103 		aclp = NULL;
3104 	}
3105 
3106 
3107 	if (mask & AT_ATIME) {
3108 		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3109 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3110 		    &zp->z_atime, sizeof (zp->z_atime));
3111 	}
3112 
3113 	if (mask & AT_MTIME) {
3114 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3115 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3116 		    mtime, sizeof (mtime));
3117 	}
3118 
3119 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3120 	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3121 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3122 		    NULL, mtime, sizeof (mtime));
3123 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3124 		    &ctime, sizeof (ctime));
3125 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3126 		    B_TRUE);
3127 	} else if (mask != 0) {
3128 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3129 		    &ctime, sizeof (ctime));
3130 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3131 		    B_TRUE);
3132 		if (attrzp) {
3133 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3134 			    SA_ZPL_CTIME(zfsvfs), NULL,
3135 			    &ctime, sizeof (ctime));
3136 			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3137 			    mtime, ctime, B_TRUE);
3138 		}
3139 	}
3140 	/*
3141 	 * Do this after setting timestamps to prevent timestamp
3142 	 * update from toggling bit
3143 	 */
3144 
3145 	if (xoap && (mask & AT_XVATTR)) {
3146 
3147 		/*
3148 		 * restore trimmed off masks
3149 		 * so that return masks can be set for caller.
3150 		 */
3151 
3152 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3153 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
3154 		}
3155 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3156 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
3157 		}
3158 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3159 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3160 		}
3161 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3162 			XVA_SET_REQ(xvap, XAT_NODUMP);
3163 		}
3164 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3165 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3166 		}
3167 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3168 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3169 		}
3170 
3171 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3172 			ASSERT(vp->v_type == VREG);
3173 
3174 		zfs_xvattr_set(zp, xvap, tx);
3175 	}
3176 
3177 	if (fuid_dirtied)
3178 		zfs_fuid_sync(zfsvfs, tx);
3179 
3180 	if (mask != 0)
3181 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3182 
3183 	mutex_exit(&zp->z_lock);
3184 	if (mask & (AT_UID|AT_GID|AT_MODE))
3185 		mutex_exit(&zp->z_acl_lock);
3186 
3187 	if (attrzp) {
3188 		if (mask & (AT_UID|AT_GID|AT_MODE))
3189 			mutex_exit(&attrzp->z_acl_lock);
3190 		mutex_exit(&attrzp->z_lock);
3191 	}
3192 out:
3193 	if (err == 0 && attrzp) {
3194 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3195 		    xattr_count, tx);
3196 		ASSERT(err2 == 0);
3197 	}
3198 
3199 	if (attrzp)
3200 		VN_RELE(ZTOV(attrzp));
3201 	if (aclp)
3202 		zfs_acl_free(aclp);
3203 
3204 	if (fuidp) {
3205 		zfs_fuid_info_free(fuidp);
3206 		fuidp = NULL;
3207 	}
3208 
3209 	if (err) {
3210 		dmu_tx_abort(tx);
3211 		if (err == ERESTART)
3212 			goto top;
3213 	} else {
3214 		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3215 		dmu_tx_commit(tx);
3216 	}
3217 
3218 out2:
3219 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3220 		zil_commit(zilog, 0);
3221 
3222 	ZFS_EXIT(zfsvfs);
3223 	return (err);
3224 }
3225 
3226 typedef struct zfs_zlock {
3227 	krwlock_t	*zl_rwlock;	/* lock we acquired */
3228 	znode_t		*zl_znode;	/* znode we held */
3229 	struct zfs_zlock *zl_next;	/* next in list */
3230 } zfs_zlock_t;
3231 
3232 /*
3233  * Drop locks and release vnodes that were held by zfs_rename_lock().
3234  */
3235 static void
3236 zfs_rename_unlock(zfs_zlock_t **zlpp)
3237 {
3238 	zfs_zlock_t *zl;
3239 
3240 	while ((zl = *zlpp) != NULL) {
3241 		if (zl->zl_znode != NULL)
3242 			VN_RELE(ZTOV(zl->zl_znode));
3243 		rw_exit(zl->zl_rwlock);
3244 		*zlpp = zl->zl_next;
3245 		kmem_free(zl, sizeof (*zl));
3246 	}
3247 }
3248 
3249 /*
3250  * Search back through the directory tree, using the ".." entries.
3251  * Lock each directory in the chain to prevent concurrent renames.
3252  * Fail any attempt to move a directory into one of its own descendants.
3253  * XXX - z_parent_lock can overlap with map or grow locks
3254  */
3255 static int
3256 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3257 {
3258 	zfs_zlock_t	*zl;
3259 	znode_t		*zp = tdzp;
3260 	uint64_t	rootid = zp->z_zfsvfs->z_root;
3261 	uint64_t	oidp = zp->z_id;
3262 	krwlock_t	*rwlp = &szp->z_parent_lock;
3263 	krw_t		rw = RW_WRITER;
3264 
3265 	/*
3266 	 * First pass write-locks szp and compares to zp->z_id.
3267 	 * Later passes read-lock zp and compare to zp->z_parent.
3268 	 */
3269 	do {
3270 		if (!rw_tryenter(rwlp, rw)) {
3271 			/*
3272 			 * Another thread is renaming in this path.
3273 			 * Note that if we are a WRITER, we don't have any
3274 			 * parent_locks held yet.
3275 			 */
3276 			if (rw == RW_READER && zp->z_id > szp->z_id) {
3277 				/*
3278 				 * Drop our locks and restart
3279 				 */
3280 				zfs_rename_unlock(&zl);
3281 				*zlpp = NULL;
3282 				zp = tdzp;
3283 				oidp = zp->z_id;
3284 				rwlp = &szp->z_parent_lock;
3285 				rw = RW_WRITER;
3286 				continue;
3287 			} else {
3288 				/*
3289 				 * Wait for other thread to drop its locks
3290 				 */
3291 				rw_enter(rwlp, rw);
3292 			}
3293 		}
3294 
3295 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3296 		zl->zl_rwlock = rwlp;
3297 		zl->zl_znode = NULL;
3298 		zl->zl_next = *zlpp;
3299 		*zlpp = zl;
3300 
3301 		if (oidp == szp->z_id)		/* We're a descendant of szp */
3302 			return (EINVAL);
3303 
3304 		if (oidp == rootid)		/* We've hit the top */
3305 			return (0);
3306 
3307 		if (rw == RW_READER) {		/* i.e. not the first pass */
3308 			int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3309 			if (error)
3310 				return (error);
3311 			zl->zl_znode = zp;
3312 		}
3313 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3314 		    &oidp, sizeof (oidp));
3315 		rwlp = &zp->z_parent_lock;
3316 		rw = RW_READER;
3317 
3318 	} while (zp->z_id != sdzp->z_id);
3319 
3320 	return (0);
3321 }
3322 
3323 /*
3324  * Move an entry from the provided source directory to the target
3325  * directory.  Change the entry name as indicated.
3326  *
3327  *	IN:	sdvp	- Source directory containing the "old entry".
3328  *		snm	- Old entry name.
3329  *		tdvp	- Target directory to contain the "new entry".
3330  *		tnm	- New entry name.
3331  *		cr	- credentials of caller.
3332  *		ct	- caller context
3333  *		flags	- case flags
3334  *
3335  *	RETURN:	0 if success
3336  *		error code if failure
3337  *
3338  * Timestamps:
3339  *	sdvp,tdvp - ctime|mtime updated
3340  */
3341 /*ARGSUSED*/
3342 static int
3343 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3344     caller_context_t *ct, int flags)
3345 {
3346 	znode_t		*tdzp, *szp, *tzp;
3347 	znode_t		*sdzp = VTOZ(sdvp);
3348 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
3349 	zilog_t		*zilog;
3350 	vnode_t		*realvp;
3351 	zfs_dirlock_t	*sdl, *tdl;
3352 	dmu_tx_t	*tx;
3353 	zfs_zlock_t	*zl;
3354 	int		cmp, serr, terr;
3355 	int		error = 0;
3356 	int		zflg = 0;
3357 
3358 	ZFS_ENTER(zfsvfs);
3359 	ZFS_VERIFY_ZP(sdzp);
3360 	zilog = zfsvfs->z_log;
3361 
3362 	/*
3363 	 * Make sure we have the real vp for the target directory.
3364 	 */
3365 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3366 		tdvp = realvp;
3367 
3368 	if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3369 		ZFS_EXIT(zfsvfs);
3370 		return (EXDEV);
3371 	}
3372 
3373 	tdzp = VTOZ(tdvp);
3374 	ZFS_VERIFY_ZP(tdzp);
3375 	if (zfsvfs->z_utf8 && u8_validate(tnm,
3376 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3377 		ZFS_EXIT(zfsvfs);
3378 		return (EILSEQ);
3379 	}
3380 
3381 	if (flags & FIGNORECASE)
3382 		zflg |= ZCILOOK;
3383 
3384 top:
3385 	szp = NULL;
3386 	tzp = NULL;
3387 	zl = NULL;
3388 
3389 	/*
3390 	 * This is to prevent the creation of links into attribute space
3391 	 * by renaming a linked file into/outof an attribute directory.
3392 	 * See the comment in zfs_link() for why this is considered bad.
3393 	 */
3394 	if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3395 		ZFS_EXIT(zfsvfs);
3396 		return (EINVAL);
3397 	}
3398 
3399 	/*
3400 	 * Lock source and target directory entries.  To prevent deadlock,
3401 	 * a lock ordering must be defined.  We lock the directory with
3402 	 * the smallest object id first, or if it's a tie, the one with
3403 	 * the lexically first name.
3404 	 */
3405 	if (sdzp->z_id < tdzp->z_id) {
3406 		cmp = -1;
3407 	} else if (sdzp->z_id > tdzp->z_id) {
3408 		cmp = 1;
3409 	} else {
3410 		/*
3411 		 * First compare the two name arguments without
3412 		 * considering any case folding.
3413 		 */
3414 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3415 
3416 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3417 		ASSERT(error == 0 || !zfsvfs->z_utf8);
3418 		if (cmp == 0) {
3419 			/*
3420 			 * POSIX: "If the old argument and the new argument
3421 			 * both refer to links to the same existing file,
3422 			 * the rename() function shall return successfully
3423 			 * and perform no other action."
3424 			 */
3425 			ZFS_EXIT(zfsvfs);
3426 			return (0);
3427 		}
3428 		/*
3429 		 * If the file system is case-folding, then we may
3430 		 * have some more checking to do.  A case-folding file
3431 		 * system is either supporting mixed case sensitivity
3432 		 * access or is completely case-insensitive.  Note
3433 		 * that the file system is always case preserving.
3434 		 *
3435 		 * In mixed sensitivity mode case sensitive behavior
3436 		 * is the default.  FIGNORECASE must be used to
3437 		 * explicitly request case insensitive behavior.
3438 		 *
3439 		 * If the source and target names provided differ only
3440 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
3441 		 * we will treat this as a special case in the
3442 		 * case-insensitive mode: as long as the source name
3443 		 * is an exact match, we will allow this to proceed as
3444 		 * a name-change request.
3445 		 */
3446 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3447 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
3448 		    flags & FIGNORECASE)) &&
3449 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3450 		    &error) == 0) {
3451 			/*
3452 			 * case preserving rename request, require exact
3453 			 * name matches
3454 			 */
3455 			zflg |= ZCIEXACT;
3456 			zflg &= ~ZCILOOK;
3457 		}
3458 	}
3459 
3460 	/*
3461 	 * If the source and destination directories are the same, we should
3462 	 * grab the z_name_lock of that directory only once.
3463 	 */
3464 	if (sdzp == tdzp) {
3465 		zflg |= ZHAVELOCK;
3466 		rw_enter(&sdzp->z_name_lock, RW_READER);
3467 	}
3468 
3469 	if (cmp < 0) {
3470 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3471 		    ZEXISTS | zflg, NULL, NULL);
3472 		terr = zfs_dirent_lock(&tdl,
3473 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3474 	} else {
3475 		terr = zfs_dirent_lock(&tdl,
3476 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3477 		serr = zfs_dirent_lock(&sdl,
3478 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3479 		    NULL, NULL);
3480 	}
3481 
3482 	if (serr) {
3483 		/*
3484 		 * Source entry invalid or not there.
3485 		 */
3486 		if (!terr) {
3487 			zfs_dirent_unlock(tdl);
3488 			if (tzp)
3489 				VN_RELE(ZTOV(tzp));
3490 		}
3491 
3492 		if (sdzp == tdzp)
3493 			rw_exit(&sdzp->z_name_lock);
3494 
3495 		if (strcmp(snm, "..") == 0)
3496 			serr = EINVAL;
3497 		ZFS_EXIT(zfsvfs);
3498 		return (serr);
3499 	}
3500 	if (terr) {
3501 		zfs_dirent_unlock(sdl);
3502 		VN_RELE(ZTOV(szp));
3503 
3504 		if (sdzp == tdzp)
3505 			rw_exit(&sdzp->z_name_lock);
3506 
3507 		if (strcmp(tnm, "..") == 0)
3508 			terr = EINVAL;
3509 		ZFS_EXIT(zfsvfs);
3510 		return (terr);
3511 	}
3512 
3513 	/*
3514 	 * Must have write access at the source to remove the old entry
3515 	 * and write access at the target to create the new entry.
3516 	 * Note that if target and source are the same, this can be
3517 	 * done in a single check.
3518 	 */
3519 
3520 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3521 		goto out;
3522 
3523 	if (ZTOV(szp)->v_type == VDIR) {
3524 		/*
3525 		 * Check to make sure rename is valid.
3526 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3527 		 */
3528 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3529 			goto out;
3530 	}
3531 
3532 	/*
3533 	 * Does target exist?
3534 	 */
3535 	if (tzp) {
3536 		/*
3537 		 * Source and target must be the same type.
3538 		 */
3539 		if (ZTOV(szp)->v_type == VDIR) {
3540 			if (ZTOV(tzp)->v_type != VDIR) {
3541 				error = ENOTDIR;
3542 				goto out;
3543 			}
3544 		} else {
3545 			if (ZTOV(tzp)->v_type == VDIR) {
3546 				error = EISDIR;
3547 				goto out;
3548 			}
3549 		}
3550 		/*
3551 		 * POSIX dictates that when the source and target
3552 		 * entries refer to the same file object, rename
3553 		 * must do nothing and exit without error.
3554 		 */
3555 		if (szp->z_id == tzp->z_id) {
3556 			error = 0;
3557 			goto out;
3558 		}
3559 	}
3560 
3561 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3562 	if (tzp)
3563 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3564 
3565 	/*
3566 	 * notify the target directory if it is not the same
3567 	 * as source directory.
3568 	 */
3569 	if (tdvp != sdvp) {
3570 		vnevent_rename_dest_dir(tdvp, ct);
3571 	}
3572 
3573 	tx = dmu_tx_create(zfsvfs->z_os);
3574 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3575 	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3576 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3577 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3578 	if (sdzp != tdzp) {
3579 		dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3580 		zfs_sa_upgrade_txholds(tx, tdzp);
3581 	}
3582 	if (tzp) {
3583 		dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3584 		zfs_sa_upgrade_txholds(tx, tzp);
3585 	}
3586 
3587 	zfs_sa_upgrade_txholds(tx, szp);
3588 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3589 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3590 	if (error) {
3591 		if (zl != NULL)
3592 			zfs_rename_unlock(&zl);
3593 		zfs_dirent_unlock(sdl);
3594 		zfs_dirent_unlock(tdl);
3595 
3596 		if (sdzp == tdzp)
3597 			rw_exit(&sdzp->z_name_lock);
3598 
3599 		VN_RELE(ZTOV(szp));
3600 		if (tzp)
3601 			VN_RELE(ZTOV(tzp));
3602 		if (error == ERESTART) {
3603 			dmu_tx_wait(tx);
3604 			dmu_tx_abort(tx);
3605 			goto top;
3606 		}
3607 		dmu_tx_abort(tx);
3608 		ZFS_EXIT(zfsvfs);
3609 		return (error);
3610 	}
3611 
3612 	if (tzp)	/* Attempt to remove the existing target */
3613 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3614 
3615 	if (error == 0) {
3616 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3617 		if (error == 0) {
3618 			szp->z_pflags |= ZFS_AV_MODIFIED;
3619 
3620 			error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3621 			    (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3622 			ASSERT3U(error, ==, 0);
3623 
3624 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3625 			if (error == 0) {
3626 				zfs_log_rename(zilog, tx, TX_RENAME |
3627 				    (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3628 				    sdl->dl_name, tdzp, tdl->dl_name, szp);
3629 
3630 				/*
3631 				 * Update path information for the target vnode
3632 				 */
3633 				vn_renamepath(tdvp, ZTOV(szp), tnm,
3634 				    strlen(tnm));
3635 			} else {
3636 				/*
3637 				 * At this point, we have successfully created
3638 				 * the target name, but have failed to remove
3639 				 * the source name.  Since the create was done
3640 				 * with the ZRENAMING flag, there are
3641 				 * complications; for one, the link count is
3642 				 * wrong.  The easiest way to deal with this
3643 				 * is to remove the newly created target, and
3644 				 * return the original error.  This must
3645 				 * succeed; fortunately, it is very unlikely to
3646 				 * fail, since we just created it.
3647 				 */
3648 				VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3649 				    ZRENAMING, NULL), ==, 0);
3650 			}
3651 		}
3652 	}
3653 
3654 	dmu_tx_commit(tx);
3655 out:
3656 	if (zl != NULL)
3657 		zfs_rename_unlock(&zl);
3658 
3659 	zfs_dirent_unlock(sdl);
3660 	zfs_dirent_unlock(tdl);
3661 
3662 	if (sdzp == tdzp)
3663 		rw_exit(&sdzp->z_name_lock);
3664 
3665 
3666 	VN_RELE(ZTOV(szp));
3667 	if (tzp)
3668 		VN_RELE(ZTOV(tzp));
3669 
3670 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3671 		zil_commit(zilog, 0);
3672 
3673 	ZFS_EXIT(zfsvfs);
3674 	return (error);
3675 }
3676 
3677 /*
3678  * Insert the indicated symbolic reference entry into the directory.
3679  *
3680  *	IN:	dvp	- Directory to contain new symbolic link.
3681  *		link	- Name for new symlink entry.
3682  *		vap	- Attributes of new entry.
3683  *		target	- Target path of new symlink.
3684  *		cr	- credentials of caller.
3685  *		ct	- caller context
3686  *		flags	- case flags
3687  *
3688  *	RETURN:	0 if success
3689  *		error code if failure
3690  *
3691  * Timestamps:
3692  *	dvp - ctime|mtime updated
3693  */
3694 /*ARGSUSED*/
3695 static int
3696 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3697     caller_context_t *ct, int flags)
3698 {
3699 	znode_t		*zp, *dzp = VTOZ(dvp);
3700 	zfs_dirlock_t	*dl;
3701 	dmu_tx_t	*tx;
3702 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3703 	zilog_t		*zilog;
3704 	uint64_t	len = strlen(link);
3705 	int		error;
3706 	int		zflg = ZNEW;
3707 	zfs_acl_ids_t	acl_ids;
3708 	boolean_t	fuid_dirtied;
3709 	uint64_t	txtype = TX_SYMLINK;
3710 
3711 	ASSERT(vap->va_type == VLNK);
3712 
3713 	ZFS_ENTER(zfsvfs);
3714 	ZFS_VERIFY_ZP(dzp);
3715 	zilog = zfsvfs->z_log;
3716 
3717 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3718 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3719 		ZFS_EXIT(zfsvfs);
3720 		return (EILSEQ);
3721 	}
3722 	if (flags & FIGNORECASE)
3723 		zflg |= ZCILOOK;
3724 
3725 	if (len > MAXPATHLEN) {
3726 		ZFS_EXIT(zfsvfs);
3727 		return (ENAMETOOLONG);
3728 	}
3729 
3730 	if ((error = zfs_acl_ids_create(dzp, 0,
3731 	    vap, cr, NULL, &acl_ids)) != 0) {
3732 		ZFS_EXIT(zfsvfs);
3733 		return (error);
3734 	}
3735 top:
3736 	/*
3737 	 * Attempt to lock directory; fail if entry already exists.
3738 	 */
3739 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3740 	if (error) {
3741 		zfs_acl_ids_free(&acl_ids);
3742 		ZFS_EXIT(zfsvfs);
3743 		return (error);
3744 	}
3745 
3746 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3747 		zfs_acl_ids_free(&acl_ids);
3748 		zfs_dirent_unlock(dl);
3749 		ZFS_EXIT(zfsvfs);
3750 		return (error);
3751 	}
3752 
3753 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3754 		zfs_acl_ids_free(&acl_ids);
3755 		zfs_dirent_unlock(dl);
3756 		ZFS_EXIT(zfsvfs);
3757 		return (EDQUOT);
3758 	}
3759 	tx = dmu_tx_create(zfsvfs->z_os);
3760 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3761 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3762 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3763 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3764 	    ZFS_SA_BASE_ATTR_SIZE + len);
3765 	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3766 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3767 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3768 		    acl_ids.z_aclp->z_acl_bytes);
3769 	}
3770 	if (fuid_dirtied)
3771 		zfs_fuid_txhold(zfsvfs, tx);
3772 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3773 	if (error) {
3774 		zfs_dirent_unlock(dl);
3775 		if (error == ERESTART) {
3776 			dmu_tx_wait(tx);
3777 			dmu_tx_abort(tx);
3778 			goto top;
3779 		}
3780 		zfs_acl_ids_free(&acl_ids);
3781 		dmu_tx_abort(tx);
3782 		ZFS_EXIT(zfsvfs);
3783 		return (error);
3784 	}
3785 
3786 	/*
3787 	 * Create a new object for the symlink.
3788 	 * for version 4 ZPL datsets the symlink will be an SA attribute
3789 	 */
3790 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3791 
3792 	if (fuid_dirtied)
3793 		zfs_fuid_sync(zfsvfs, tx);
3794 
3795 	mutex_enter(&zp->z_lock);
3796 	if (zp->z_is_sa)
3797 		error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3798 		    link, len, tx);
3799 	else
3800 		zfs_sa_symlink(zp, link, len, tx);
3801 	mutex_exit(&zp->z_lock);
3802 
3803 	zp->z_size = len;
3804 	(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3805 	    &zp->z_size, sizeof (zp->z_size), tx);
3806 	/*
3807 	 * Insert the new object into the directory.
3808 	 */
3809 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3810 
3811 	if (flags & FIGNORECASE)
3812 		txtype |= TX_CI;
3813 	zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3814 
3815 	zfs_acl_ids_free(&acl_ids);
3816 
3817 	dmu_tx_commit(tx);
3818 
3819 	zfs_dirent_unlock(dl);
3820 
3821 	VN_RELE(ZTOV(zp));
3822 
3823 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3824 		zil_commit(zilog, 0);
3825 
3826 	ZFS_EXIT(zfsvfs);
3827 	return (error);
3828 }
3829 
3830 /*
3831  * Return, in the buffer contained in the provided uio structure,
3832  * the symbolic path referred to by vp.
3833  *
3834  *	IN:	vp	- vnode of symbolic link.
3835  *		uoip	- structure to contain the link path.
3836  *		cr	- credentials of caller.
3837  *		ct	- caller context
3838  *
3839  *	OUT:	uio	- structure to contain the link path.
3840  *
3841  *	RETURN:	0 if success
3842  *		error code if failure
3843  *
3844  * Timestamps:
3845  *	vp - atime updated
3846  */
3847 /* ARGSUSED */
3848 static int
3849 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3850 {
3851 	znode_t		*zp = VTOZ(vp);
3852 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3853 	int		error;
3854 
3855 	ZFS_ENTER(zfsvfs);
3856 	ZFS_VERIFY_ZP(zp);
3857 
3858 	mutex_enter(&zp->z_lock);
3859 	if (zp->z_is_sa)
3860 		error = sa_lookup_uio(zp->z_sa_hdl,
3861 		    SA_ZPL_SYMLINK(zfsvfs), uio);
3862 	else
3863 		error = zfs_sa_readlink(zp, uio);
3864 	mutex_exit(&zp->z_lock);
3865 
3866 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3867 
3868 	ZFS_EXIT(zfsvfs);
3869 	return (error);
3870 }
3871 
3872 /*
3873  * Insert a new entry into directory tdvp referencing svp.
3874  *
3875  *	IN:	tdvp	- Directory to contain new entry.
3876  *		svp	- vnode of new entry.
3877  *		name	- name of new entry.
3878  *		cr	- credentials of caller.
3879  *		ct	- caller context
3880  *
3881  *	RETURN:	0 if success
3882  *		error code if failure
3883  *
3884  * Timestamps:
3885  *	tdvp - ctime|mtime updated
3886  *	 svp - ctime updated
3887  */
3888 /* ARGSUSED */
3889 static int
3890 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3891     caller_context_t *ct, int flags)
3892 {
3893 	znode_t		*dzp = VTOZ(tdvp);
3894 	znode_t		*tzp, *szp;
3895 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3896 	zilog_t		*zilog;
3897 	zfs_dirlock_t	*dl;
3898 	dmu_tx_t	*tx;
3899 	vnode_t		*realvp;
3900 	int		error;
3901 	int		zf = ZNEW;
3902 	uint64_t	parent;
3903 	uid_t		owner;
3904 
3905 	ASSERT(tdvp->v_type == VDIR);
3906 
3907 	ZFS_ENTER(zfsvfs);
3908 	ZFS_VERIFY_ZP(dzp);
3909 	zilog = zfsvfs->z_log;
3910 
3911 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3912 		svp = realvp;
3913 
3914 	/*
3915 	 * POSIX dictates that we return EPERM here.
3916 	 * Better choices include ENOTSUP or EISDIR.
3917 	 */
3918 	if (svp->v_type == VDIR) {
3919 		ZFS_EXIT(zfsvfs);
3920 		return (EPERM);
3921 	}
3922 
3923 	if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
3924 		ZFS_EXIT(zfsvfs);
3925 		return (EXDEV);
3926 	}
3927 
3928 	szp = VTOZ(svp);
3929 	ZFS_VERIFY_ZP(szp);
3930 
3931 	/* Prevent links to .zfs/shares files */
3932 
3933 	if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3934 	    &parent, sizeof (uint64_t))) != 0) {
3935 		ZFS_EXIT(zfsvfs);
3936 		return (error);
3937 	}
3938 	if (parent == zfsvfs->z_shares_dir) {
3939 		ZFS_EXIT(zfsvfs);
3940 		return (EPERM);
3941 	}
3942 
3943 	if (zfsvfs->z_utf8 && u8_validate(name,
3944 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3945 		ZFS_EXIT(zfsvfs);
3946 		return (EILSEQ);
3947 	}
3948 	if (flags & FIGNORECASE)
3949 		zf |= ZCILOOK;
3950 
3951 	/*
3952 	 * We do not support links between attributes and non-attributes
3953 	 * because of the potential security risk of creating links
3954 	 * into "normal" file space in order to circumvent restrictions
3955 	 * imposed in attribute space.
3956 	 */
3957 	if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3958 		ZFS_EXIT(zfsvfs);
3959 		return (EINVAL);
3960 	}
3961 
3962 
3963 	owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
3964 	if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3965 		ZFS_EXIT(zfsvfs);
3966 		return (EPERM);
3967 	}
3968 
3969 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3970 		ZFS_EXIT(zfsvfs);
3971 		return (error);
3972 	}
3973 
3974 top:
3975 	/*
3976 	 * Attempt to lock directory; fail if entry already exists.
3977 	 */
3978 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3979 	if (error) {
3980 		ZFS_EXIT(zfsvfs);
3981 		return (error);
3982 	}
3983 
3984 	tx = dmu_tx_create(zfsvfs->z_os);
3985 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3986 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3987 	zfs_sa_upgrade_txholds(tx, szp);
3988 	zfs_sa_upgrade_txholds(tx, dzp);
3989 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3990 	if (error) {
3991 		zfs_dirent_unlock(dl);
3992 		if (error == ERESTART) {
3993 			dmu_tx_wait(tx);
3994 			dmu_tx_abort(tx);
3995 			goto top;
3996 		}
3997 		dmu_tx_abort(tx);
3998 		ZFS_EXIT(zfsvfs);
3999 		return (error);
4000 	}
4001 
4002 	error = zfs_link_create(dl, szp, tx, 0);
4003 
4004 	if (error == 0) {
4005 		uint64_t txtype = TX_LINK;
4006 		if (flags & FIGNORECASE)
4007 			txtype |= TX_CI;
4008 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4009 	}
4010 
4011 	dmu_tx_commit(tx);
4012 
4013 	zfs_dirent_unlock(dl);
4014 
4015 	if (error == 0) {
4016 		vnevent_link(svp, ct);
4017 	}
4018 
4019 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4020 		zil_commit(zilog, 0);
4021 
4022 	ZFS_EXIT(zfsvfs);
4023 	return (error);
4024 }
4025 
4026 /*
4027  * zfs_null_putapage() is used when the file system has been force
4028  * unmounted. It just drops the pages.
4029  */
4030 /* ARGSUSED */
4031 static int
4032 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4033 		size_t *lenp, int flags, cred_t *cr)
4034 {
4035 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4036 	return (0);
4037 }
4038 
4039 /*
4040  * Push a page out to disk, klustering if possible.
4041  *
4042  *	IN:	vp	- file to push page to.
4043  *		pp	- page to push.
4044  *		flags	- additional flags.
4045  *		cr	- credentials of caller.
4046  *
4047  *	OUT:	offp	- start of range pushed.
4048  *		lenp	- len of range pushed.
4049  *
4050  *	RETURN:	0 if success
4051  *		error code if failure
4052  *
4053  * NOTE: callers must have locked the page to be pushed.  On
4054  * exit, the page (and all other pages in the kluster) must be
4055  * unlocked.
4056  */
4057 /* ARGSUSED */
4058 static int
4059 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4060 		size_t *lenp, int flags, cred_t *cr)
4061 {
4062 	znode_t		*zp = VTOZ(vp);
4063 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4064 	dmu_tx_t	*tx;
4065 	u_offset_t	off, koff;
4066 	size_t		len, klen;
4067 	int		err;
4068 
4069 	off = pp->p_offset;
4070 	len = PAGESIZE;
4071 	/*
4072 	 * If our blocksize is bigger than the page size, try to kluster
4073 	 * multiple pages so that we write a full block (thus avoiding
4074 	 * a read-modify-write).
4075 	 */
4076 	if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4077 		klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4078 		koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4079 		ASSERT(koff <= zp->z_size);
4080 		if (koff + klen > zp->z_size)
4081 			klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4082 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4083 	}
4084 	ASSERT3U(btop(len), ==, btopr(len));
4085 
4086 	/*
4087 	 * Can't push pages past end-of-file.
4088 	 */
4089 	if (off >= zp->z_size) {
4090 		/* ignore all pages */
4091 		err = 0;
4092 		goto out;
4093 	} else if (off + len > zp->z_size) {
4094 		int npages = btopr(zp->z_size - off);
4095 		page_t *trunc;
4096 
4097 		page_list_break(&pp, &trunc, npages);
4098 		/* ignore pages past end of file */
4099 		if (trunc)
4100 			pvn_write_done(trunc, flags);
4101 		len = zp->z_size - off;
4102 	}
4103 
4104 	if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4105 	    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4106 		err = EDQUOT;
4107 		goto out;
4108 	}
4109 top:
4110 	tx = dmu_tx_create(zfsvfs->z_os);
4111 	dmu_tx_hold_write(tx, zp->z_id, off, len);
4112 
4113 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4114 	zfs_sa_upgrade_txholds(tx, zp);
4115 	err = dmu_tx_assign(tx, TXG_NOWAIT);
4116 	if (err != 0) {
4117 		if (err == ERESTART) {
4118 			dmu_tx_wait(tx);
4119 			dmu_tx_abort(tx);
4120 			goto top;
4121 		}
4122 		dmu_tx_abort(tx);
4123 		goto out;
4124 	}
4125 
4126 	if (zp->z_blksz <= PAGESIZE) {
4127 		caddr_t va = zfs_map_page(pp, S_READ);
4128 		ASSERT3U(len, <=, PAGESIZE);
4129 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4130 		zfs_unmap_page(pp, va);
4131 	} else {
4132 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4133 	}
4134 
4135 	if (err == 0) {
4136 		uint64_t mtime[2], ctime[2];
4137 		sa_bulk_attr_t bulk[3];
4138 		int count = 0;
4139 
4140 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4141 		    &mtime, 16);
4142 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4143 		    &ctime, 16);
4144 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4145 		    &zp->z_pflags, 8);
4146 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4147 		    B_TRUE);
4148 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4149 	}
4150 	dmu_tx_commit(tx);
4151 
4152 out:
4153 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4154 	if (offp)
4155 		*offp = off;
4156 	if (lenp)
4157 		*lenp = len;
4158 
4159 	return (err);
4160 }
4161 
4162 /*
4163  * Copy the portion of the file indicated from pages into the file.
4164  * The pages are stored in a page list attached to the files vnode.
4165  *
4166  *	IN:	vp	- vnode of file to push page data to.
4167  *		off	- position in file to put data.
4168  *		len	- amount of data to write.
4169  *		flags	- flags to control the operation.
4170  *		cr	- credentials of caller.
4171  *		ct	- caller context.
4172  *
4173  *	RETURN:	0 if success
4174  *		error code if failure
4175  *
4176  * Timestamps:
4177  *	vp - ctime|mtime updated
4178  */
4179 /*ARGSUSED*/
4180 static int
4181 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4182     caller_context_t *ct)
4183 {
4184 	znode_t		*zp = VTOZ(vp);
4185 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4186 	page_t		*pp;
4187 	size_t		io_len;
4188 	u_offset_t	io_off;
4189 	uint_t		blksz;
4190 	rl_t		*rl;
4191 	int		error = 0;
4192 
4193 	ZFS_ENTER(zfsvfs);
4194 	ZFS_VERIFY_ZP(zp);
4195 
4196 	/*
4197 	 * There's nothing to do if no data is cached.
4198 	 */
4199 	if (!vn_has_cached_data(vp)) {
4200 		ASSERT(len == 0);
4201 		ZFS_EXIT(zfsvfs);
4202 		return (0);
4203 	}
4204 
4205 	/*
4206 	 * Align this request to the file block size in case we kluster.
4207 	 * XXX - this can result in pretty aggresive locking, which can
4208 	 * impact simultanious read/write access.  One option might be
4209 	 * to break up long requests (len == 0) into block-by-block
4210 	 * operations to get narrower locking.
4211 	 */
4212 	blksz = zp->z_blksz;
4213 	if (ISP2(blksz))
4214 		io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4215 	else
4216 		io_off = 0;
4217 	if (len > 0 && ISP2(blksz))
4218 		io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4219 	else
4220 		io_len = 0;
4221 
4222 	if (io_len == 0) {
4223 		/*
4224 		 * Search the entire vp list for pages >= io_off.
4225 		 */
4226 		rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4227 		error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4228 		goto out;
4229 	}
4230 	rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4231 
4232 	if (off > zp->z_size) {
4233 		/* past end of file */
4234 		zfs_range_unlock(rl);
4235 		ZFS_EXIT(zfsvfs);
4236 		return (0);
4237 	}
4238 
4239 	len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4240 
4241 	for (off = io_off; io_off < off + len; io_off += io_len) {
4242 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4243 			pp = page_lookup(vp, io_off,
4244 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4245 		} else {
4246 			pp = page_lookup_nowait(vp, io_off,
4247 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4248 		}
4249 
4250 		if (pp != NULL && pvn_getdirty(pp, flags)) {
4251 			int err;
4252 
4253 			/*
4254 			 * Found a dirty page to push
4255 			 */
4256 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4257 			if (err)
4258 				error = err;
4259 		} else {
4260 			io_len = PAGESIZE;
4261 		}
4262 	}
4263 out:
4264 	zfs_range_unlock(rl);
4265 	if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4266 		zil_commit(zfsvfs->z_log, zp->z_id);
4267 	ZFS_EXIT(zfsvfs);
4268 	return (error);
4269 }
4270 
4271 /*ARGSUSED*/
4272 void
4273 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4274 {
4275 	znode_t	*zp = VTOZ(vp);
4276 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4277 	int error;
4278 
4279 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4280 	if (zp->z_sa_hdl == NULL) {
4281 		/*
4282 		 * The fs has been unmounted, or we did a
4283 		 * suspend/resume and this file no longer exists.
4284 		 */
4285 		if (vn_has_cached_data(vp)) {
4286 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4287 			    B_INVAL, cr);
4288 		}
4289 
4290 		mutex_enter(&zp->z_lock);
4291 		mutex_enter(&vp->v_lock);
4292 		ASSERT(vp->v_count == 1);
4293 		vp->v_count = 0;
4294 		mutex_exit(&vp->v_lock);
4295 		mutex_exit(&zp->z_lock);
4296 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
4297 		zfs_znode_free(zp);
4298 		return;
4299 	}
4300 
4301 	/*
4302 	 * Attempt to push any data in the page cache.  If this fails
4303 	 * we will get kicked out later in zfs_zinactive().
4304 	 */
4305 	if (vn_has_cached_data(vp)) {
4306 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4307 		    cr);
4308 	}
4309 
4310 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4311 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4312 
4313 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4314 		zfs_sa_upgrade_txholds(tx, zp);
4315 		error = dmu_tx_assign(tx, TXG_WAIT);
4316 		if (error) {
4317 			dmu_tx_abort(tx);
4318 		} else {
4319 			mutex_enter(&zp->z_lock);
4320 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4321 			    (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4322 			zp->z_atime_dirty = 0;
4323 			mutex_exit(&zp->z_lock);
4324 			dmu_tx_commit(tx);
4325 		}
4326 	}
4327 
4328 	zfs_zinactive(zp);
4329 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
4330 }
4331 
4332 /*
4333  * Bounds-check the seek operation.
4334  *
4335  *	IN:	vp	- vnode seeking within
4336  *		ooff	- old file offset
4337  *		noffp	- pointer to new file offset
4338  *		ct	- caller context
4339  *
4340  *	RETURN:	0 if success
4341  *		EINVAL if new offset invalid
4342  */
4343 /* ARGSUSED */
4344 static int
4345 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4346     caller_context_t *ct)
4347 {
4348 	if (vp->v_type == VDIR)
4349 		return (0);
4350 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4351 }
4352 
4353 /*
4354  * Pre-filter the generic locking function to trap attempts to place
4355  * a mandatory lock on a memory mapped file.
4356  */
4357 static int
4358 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4359     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4360 {
4361 	znode_t *zp = VTOZ(vp);
4362 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4363 
4364 	ZFS_ENTER(zfsvfs);
4365 	ZFS_VERIFY_ZP(zp);
4366 
4367 	/*
4368 	 * We are following the UFS semantics with respect to mapcnt
4369 	 * here: If we see that the file is mapped already, then we will
4370 	 * return an error, but we don't worry about races between this
4371 	 * function and zfs_map().
4372 	 */
4373 	if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4374 		ZFS_EXIT(zfsvfs);
4375 		return (EAGAIN);
4376 	}
4377 	ZFS_EXIT(zfsvfs);
4378 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4379 }
4380 
4381 /*
4382  * If we can't find a page in the cache, we will create a new page
4383  * and fill it with file data.  For efficiency, we may try to fill
4384  * multiple pages at once (klustering) to fill up the supplied page
4385  * list.  Note that the pages to be filled are held with an exclusive
4386  * lock to prevent access by other threads while they are being filled.
4387  */
4388 static int
4389 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4390     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4391 {
4392 	znode_t *zp = VTOZ(vp);
4393 	page_t *pp, *cur_pp;
4394 	objset_t *os = zp->z_zfsvfs->z_os;
4395 	u_offset_t io_off, total;
4396 	size_t io_len;
4397 	int err;
4398 
4399 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4400 		/*
4401 		 * We only have a single page, don't bother klustering
4402 		 */
4403 		io_off = off;
4404 		io_len = PAGESIZE;
4405 		pp = page_create_va(vp, io_off, io_len,
4406 		    PG_EXCL | PG_WAIT, seg, addr);
4407 	} else {
4408 		/*
4409 		 * Try to find enough pages to fill the page list
4410 		 */
4411 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4412 		    &io_len, off, plsz, 0);
4413 	}
4414 	if (pp == NULL) {
4415 		/*
4416 		 * The page already exists, nothing to do here.
4417 		 */
4418 		*pl = NULL;
4419 		return (0);
4420 	}
4421 
4422 	/*
4423 	 * Fill the pages in the kluster.
4424 	 */
4425 	cur_pp = pp;
4426 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4427 		caddr_t va;
4428 
4429 		ASSERT3U(io_off, ==, cur_pp->p_offset);
4430 		va = zfs_map_page(cur_pp, S_WRITE);
4431 		err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4432 		    DMU_READ_PREFETCH);
4433 		zfs_unmap_page(cur_pp, va);
4434 		if (err) {
4435 			/* On error, toss the entire kluster */
4436 			pvn_read_done(pp, B_ERROR);
4437 			/* convert checksum errors into IO errors */
4438 			if (err == ECKSUM)
4439 				err = EIO;
4440 			return (err);
4441 		}
4442 		cur_pp = cur_pp->p_next;
4443 	}
4444 
4445 	/*
4446 	 * Fill in the page list array from the kluster starting
4447 	 * from the desired offset `off'.
4448 	 * NOTE: the page list will always be null terminated.
4449 	 */
4450 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4451 	ASSERT(pl == NULL || (*pl)->p_offset == off);
4452 
4453 	return (0);
4454 }
4455 
4456 /*
4457  * Return pointers to the pages for the file region [off, off + len]
4458  * in the pl array.  If plsz is greater than len, this function may
4459  * also return page pointers from after the specified region
4460  * (i.e. the region [off, off + plsz]).  These additional pages are
4461  * only returned if they are already in the cache, or were created as
4462  * part of a klustered read.
4463  *
4464  *	IN:	vp	- vnode of file to get data from.
4465  *		off	- position in file to get data from.
4466  *		len	- amount of data to retrieve.
4467  *		plsz	- length of provided page list.
4468  *		seg	- segment to obtain pages for.
4469  *		addr	- virtual address of fault.
4470  *		rw	- mode of created pages.
4471  *		cr	- credentials of caller.
4472  *		ct	- caller context.
4473  *
4474  *	OUT:	protp	- protection mode of created pages.
4475  *		pl	- list of pages created.
4476  *
4477  *	RETURN:	0 if success
4478  *		error code if failure
4479  *
4480  * Timestamps:
4481  *	vp - atime updated
4482  */
4483 /* ARGSUSED */
4484 static int
4485 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4486 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4487 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4488 {
4489 	znode_t		*zp = VTOZ(vp);
4490 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4491 	page_t		**pl0 = pl;
4492 	int		err = 0;
4493 
4494 	/* we do our own caching, faultahead is unnecessary */
4495 	if (pl == NULL)
4496 		return (0);
4497 	else if (len > plsz)
4498 		len = plsz;
4499 	else
4500 		len = P2ROUNDUP(len, PAGESIZE);
4501 	ASSERT(plsz >= len);
4502 
4503 	ZFS_ENTER(zfsvfs);
4504 	ZFS_VERIFY_ZP(zp);
4505 
4506 	if (protp)
4507 		*protp = PROT_ALL;
4508 
4509 	/*
4510 	 * Loop through the requested range [off, off + len) looking
4511 	 * for pages.  If we don't find a page, we will need to create
4512 	 * a new page and fill it with data from the file.
4513 	 */
4514 	while (len > 0) {
4515 		if (*pl = page_lookup(vp, off, SE_SHARED))
4516 			*(pl+1) = NULL;
4517 		else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4518 			goto out;
4519 		while (*pl) {
4520 			ASSERT3U((*pl)->p_offset, ==, off);
4521 			off += PAGESIZE;
4522 			addr += PAGESIZE;
4523 			if (len > 0) {
4524 				ASSERT3U(len, >=, PAGESIZE);
4525 				len -= PAGESIZE;
4526 			}
4527 			ASSERT3U(plsz, >=, PAGESIZE);
4528 			plsz -= PAGESIZE;
4529 			pl++;
4530 		}
4531 	}
4532 
4533 	/*
4534 	 * Fill out the page array with any pages already in the cache.
4535 	 */
4536 	while (plsz > 0 &&
4537 	    (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4538 			off += PAGESIZE;
4539 			plsz -= PAGESIZE;
4540 	}
4541 out:
4542 	if (err) {
4543 		/*
4544 		 * Release any pages we have previously locked.
4545 		 */
4546 		while (pl > pl0)
4547 			page_unlock(*--pl);
4548 	} else {
4549 		ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4550 	}
4551 
4552 	*pl = NULL;
4553 
4554 	ZFS_EXIT(zfsvfs);
4555 	return (err);
4556 }
4557 
4558 /*
4559  * Request a memory map for a section of a file.  This code interacts
4560  * with common code and the VM system as follows:
4561  *
4562  *	common code calls mmap(), which ends up in smmap_common()
4563  *
4564  *	this calls VOP_MAP(), which takes you into (say) zfs
4565  *
4566  *	zfs_map() calls as_map(), passing segvn_create() as the callback
4567  *
4568  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
4569  *
4570  *	zfs_addmap() updates z_mapcnt
4571  */
4572 /*ARGSUSED*/
4573 static int
4574 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4575     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4576     caller_context_t *ct)
4577 {
4578 	znode_t *zp = VTOZ(vp);
4579 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4580 	segvn_crargs_t	vn_a;
4581 	int		error;
4582 
4583 	ZFS_ENTER(zfsvfs);
4584 	ZFS_VERIFY_ZP(zp);
4585 
4586 	if ((prot & PROT_WRITE) && (zp->z_pflags &
4587 	    (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4588 		ZFS_EXIT(zfsvfs);
4589 		return (EPERM);
4590 	}
4591 
4592 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4593 	    (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4594 		ZFS_EXIT(zfsvfs);
4595 		return (EACCES);
4596 	}
4597 
4598 	if (vp->v_flag & VNOMAP) {
4599 		ZFS_EXIT(zfsvfs);
4600 		return (ENOSYS);
4601 	}
4602 
4603 	if (off < 0 || len > MAXOFFSET_T - off) {
4604 		ZFS_EXIT(zfsvfs);
4605 		return (ENXIO);
4606 	}
4607 
4608 	if (vp->v_type != VREG) {
4609 		ZFS_EXIT(zfsvfs);
4610 		return (ENODEV);
4611 	}
4612 
4613 	/*
4614 	 * If file is locked, disallow mapping.
4615 	 */
4616 	if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4617 		ZFS_EXIT(zfsvfs);
4618 		return (EAGAIN);
4619 	}
4620 
4621 	as_rangelock(as);
4622 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4623 	if (error != 0) {
4624 		as_rangeunlock(as);
4625 		ZFS_EXIT(zfsvfs);
4626 		return (error);
4627 	}
4628 
4629 	vn_a.vp = vp;
4630 	vn_a.offset = (u_offset_t)off;
4631 	vn_a.type = flags & MAP_TYPE;
4632 	vn_a.prot = prot;
4633 	vn_a.maxprot = maxprot;
4634 	vn_a.cred = cr;
4635 	vn_a.amp = NULL;
4636 	vn_a.flags = flags & ~MAP_TYPE;
4637 	vn_a.szc = 0;
4638 	vn_a.lgrp_mem_policy_flags = 0;
4639 
4640 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4641 
4642 	as_rangeunlock(as);
4643 	ZFS_EXIT(zfsvfs);
4644 	return (error);
4645 }
4646 
4647 /* ARGSUSED */
4648 static int
4649 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4650     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4651     caller_context_t *ct)
4652 {
4653 	uint64_t pages = btopr(len);
4654 
4655 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4656 	return (0);
4657 }
4658 
4659 /*
4660  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4661  * more accurate mtime for the associated file.  Since we don't have a way of
4662  * detecting when the data was actually modified, we have to resort to
4663  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4664  * last page is pushed.  The problem occurs when the msync() call is omitted,
4665  * which by far the most common case:
4666  *
4667  * 	open()
4668  * 	mmap()
4669  * 	<modify memory>
4670  * 	munmap()
4671  * 	close()
4672  * 	<time lapse>
4673  * 	putpage() via fsflush
4674  *
4675  * If we wait until fsflush to come along, we can have a modification time that
4676  * is some arbitrary point in the future.  In order to prevent this in the
4677  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4678  * torn down.
4679  */
4680 /* ARGSUSED */
4681 static int
4682 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4683     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4684     caller_context_t *ct)
4685 {
4686 	uint64_t pages = btopr(len);
4687 
4688 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4689 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4690 
4691 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4692 	    vn_has_cached_data(vp))
4693 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4694 
4695 	return (0);
4696 }
4697 
4698 /*
4699  * Free or allocate space in a file.  Currently, this function only
4700  * supports the `F_FREESP' command.  However, this command is somewhat
4701  * misnamed, as its functionality includes the ability to allocate as
4702  * well as free space.
4703  *
4704  *	IN:	vp	- vnode of file to free data in.
4705  *		cmd	- action to take (only F_FREESP supported).
4706  *		bfp	- section of file to free/alloc.
4707  *		flag	- current file open mode flags.
4708  *		offset	- current file offset.
4709  *		cr	- credentials of caller [UNUSED].
4710  *		ct	- caller context.
4711  *
4712  *	RETURN:	0 if success
4713  *		error code if failure
4714  *
4715  * Timestamps:
4716  *	vp - ctime|mtime updated
4717  */
4718 /* ARGSUSED */
4719 static int
4720 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4721     offset_t offset, cred_t *cr, caller_context_t *ct)
4722 {
4723 	znode_t		*zp = VTOZ(vp);
4724 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4725 	uint64_t	off, len;
4726 	int		error;
4727 
4728 	ZFS_ENTER(zfsvfs);
4729 	ZFS_VERIFY_ZP(zp);
4730 
4731 	if (cmd != F_FREESP) {
4732 		ZFS_EXIT(zfsvfs);
4733 		return (EINVAL);
4734 	}
4735 
4736 	if (error = convoff(vp, bfp, 0, offset)) {
4737 		ZFS_EXIT(zfsvfs);
4738 		return (error);
4739 	}
4740 
4741 	if (bfp->l_len < 0) {
4742 		ZFS_EXIT(zfsvfs);
4743 		return (EINVAL);
4744 	}
4745 
4746 	off = bfp->l_start;
4747 	len = bfp->l_len; /* 0 means from off to end of file */
4748 
4749 	error = zfs_freesp(zp, off, len, flag, TRUE);
4750 
4751 	ZFS_EXIT(zfsvfs);
4752 	return (error);
4753 }
4754 
4755 /*ARGSUSED*/
4756 static int
4757 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4758 {
4759 	znode_t		*zp = VTOZ(vp);
4760 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4761 	uint32_t	gen;
4762 	uint64_t	gen64;
4763 	uint64_t	object = zp->z_id;
4764 	zfid_short_t	*zfid;
4765 	int		size, i, error;
4766 
4767 	ZFS_ENTER(zfsvfs);
4768 	ZFS_VERIFY_ZP(zp);
4769 
4770 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4771 	    &gen64, sizeof (uint64_t))) != 0) {
4772 		ZFS_EXIT(zfsvfs);
4773 		return (error);
4774 	}
4775 
4776 	gen = (uint32_t)gen64;
4777 
4778 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4779 	if (fidp->fid_len < size) {
4780 		fidp->fid_len = size;
4781 		ZFS_EXIT(zfsvfs);
4782 		return (ENOSPC);
4783 	}
4784 
4785 	zfid = (zfid_short_t *)fidp;
4786 
4787 	zfid->zf_len = size;
4788 
4789 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4790 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4791 
4792 	/* Must have a non-zero generation number to distinguish from .zfs */
4793 	if (gen == 0)
4794 		gen = 1;
4795 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4796 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4797 
4798 	if (size == LONG_FID_LEN) {
4799 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4800 		zfid_long_t	*zlfid;
4801 
4802 		zlfid = (zfid_long_t *)fidp;
4803 
4804 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4805 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4806 
4807 		/* XXX - this should be the generation number for the objset */
4808 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4809 			zlfid->zf_setgen[i] = 0;
4810 	}
4811 
4812 	ZFS_EXIT(zfsvfs);
4813 	return (0);
4814 }
4815 
4816 static int
4817 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4818     caller_context_t *ct)
4819 {
4820 	znode_t		*zp, *xzp;
4821 	zfsvfs_t	*zfsvfs;
4822 	zfs_dirlock_t	*dl;
4823 	int		error;
4824 
4825 	switch (cmd) {
4826 	case _PC_LINK_MAX:
4827 		*valp = ULONG_MAX;
4828 		return (0);
4829 
4830 	case _PC_FILESIZEBITS:
4831 		*valp = 64;
4832 		return (0);
4833 
4834 	case _PC_XATTR_EXISTS:
4835 		zp = VTOZ(vp);
4836 		zfsvfs = zp->z_zfsvfs;
4837 		ZFS_ENTER(zfsvfs);
4838 		ZFS_VERIFY_ZP(zp);
4839 		*valp = 0;
4840 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4841 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4842 		if (error == 0) {
4843 			zfs_dirent_unlock(dl);
4844 			if (!zfs_dirempty(xzp))
4845 				*valp = 1;
4846 			VN_RELE(ZTOV(xzp));
4847 		} else if (error == ENOENT) {
4848 			/*
4849 			 * If there aren't extended attributes, it's the
4850 			 * same as having zero of them.
4851 			 */
4852 			error = 0;
4853 		}
4854 		ZFS_EXIT(zfsvfs);
4855 		return (error);
4856 
4857 	case _PC_SATTR_ENABLED:
4858 	case _PC_SATTR_EXISTS:
4859 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4860 		    (vp->v_type == VREG || vp->v_type == VDIR);
4861 		return (0);
4862 
4863 	case _PC_ACCESS_FILTERING:
4864 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4865 		    vp->v_type == VDIR;
4866 		return (0);
4867 
4868 	case _PC_ACL_ENABLED:
4869 		*valp = _ACL_ACE_ENABLED;
4870 		return (0);
4871 
4872 	case _PC_MIN_HOLE_SIZE:
4873 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4874 		return (0);
4875 
4876 	case _PC_TIMESTAMP_RESOLUTION:
4877 		/* nanosecond timestamp resolution */
4878 		*valp = 1L;
4879 		return (0);
4880 
4881 	default:
4882 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4883 	}
4884 }
4885 
4886 /*ARGSUSED*/
4887 static int
4888 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4889     caller_context_t *ct)
4890 {
4891 	znode_t *zp = VTOZ(vp);
4892 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4893 	int error;
4894 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4895 
4896 	ZFS_ENTER(zfsvfs);
4897 	ZFS_VERIFY_ZP(zp);
4898 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4899 	ZFS_EXIT(zfsvfs);
4900 
4901 	return (error);
4902 }
4903 
4904 /*ARGSUSED*/
4905 static int
4906 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4907     caller_context_t *ct)
4908 {
4909 	znode_t *zp = VTOZ(vp);
4910 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4911 	int error;
4912 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4913 	zilog_t	*zilog = zfsvfs->z_log;
4914 
4915 	ZFS_ENTER(zfsvfs);
4916 	ZFS_VERIFY_ZP(zp);
4917 
4918 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4919 
4920 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4921 		zil_commit(zilog, 0);
4922 
4923 	ZFS_EXIT(zfsvfs);
4924 	return (error);
4925 }
4926 
4927 /*
4928  * Tunable, both must be a power of 2.
4929  *
4930  * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4931  * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4932  *                an arcbuf for a partial block read
4933  */
4934 int zcr_blksz_min = (1 << 10);	/* 1K */
4935 int zcr_blksz_max = (1 << 17);	/* 128K */
4936 
4937 /*ARGSUSED*/
4938 static int
4939 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
4940     caller_context_t *ct)
4941 {
4942 	znode_t	*zp = VTOZ(vp);
4943 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4944 	int max_blksz = zfsvfs->z_max_blksz;
4945 	uio_t *uio = &xuio->xu_uio;
4946 	ssize_t size = uio->uio_resid;
4947 	offset_t offset = uio->uio_loffset;
4948 	int blksz;
4949 	int fullblk, i;
4950 	arc_buf_t *abuf;
4951 	ssize_t maxsize;
4952 	int preamble, postamble;
4953 
4954 	if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4955 		return (EINVAL);
4956 
4957 	ZFS_ENTER(zfsvfs);
4958 	ZFS_VERIFY_ZP(zp);
4959 	switch (ioflag) {
4960 	case UIO_WRITE:
4961 		/*
4962 		 * Loan out an arc_buf for write if write size is bigger than
4963 		 * max_blksz, and the file's block size is also max_blksz.
4964 		 */
4965 		blksz = max_blksz;
4966 		if (size < blksz || zp->z_blksz != blksz) {
4967 			ZFS_EXIT(zfsvfs);
4968 			return (EINVAL);
4969 		}
4970 		/*
4971 		 * Caller requests buffers for write before knowing where the
4972 		 * write offset might be (e.g. NFS TCP write).
4973 		 */
4974 		if (offset == -1) {
4975 			preamble = 0;
4976 		} else {
4977 			preamble = P2PHASE(offset, blksz);
4978 			if (preamble) {
4979 				preamble = blksz - preamble;
4980 				size -= preamble;
4981 			}
4982 		}
4983 
4984 		postamble = P2PHASE(size, blksz);
4985 		size -= postamble;
4986 
4987 		fullblk = size / blksz;
4988 		(void) dmu_xuio_init(xuio,
4989 		    (preamble != 0) + fullblk + (postamble != 0));
4990 		DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
4991 		    int, postamble, int,
4992 		    (preamble != 0) + fullblk + (postamble != 0));
4993 
4994 		/*
4995 		 * Have to fix iov base/len for partial buffers.  They
4996 		 * currently represent full arc_buf's.
4997 		 */
4998 		if (preamble) {
4999 			/* data begins in the middle of the arc_buf */
5000 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5001 			    blksz);
5002 			ASSERT(abuf);
5003 			(void) dmu_xuio_add(xuio, abuf,
5004 			    blksz - preamble, preamble);
5005 		}
5006 
5007 		for (i = 0; i < fullblk; i++) {
5008 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5009 			    blksz);
5010 			ASSERT(abuf);
5011 			(void) dmu_xuio_add(xuio, abuf, 0, blksz);
5012 		}
5013 
5014 		if (postamble) {
5015 			/* data ends in the middle of the arc_buf */
5016 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5017 			    blksz);
5018 			ASSERT(abuf);
5019 			(void) dmu_xuio_add(xuio, abuf, 0, postamble);
5020 		}
5021 		break;
5022 	case UIO_READ:
5023 		/*
5024 		 * Loan out an arc_buf for read if the read size is larger than
5025 		 * the current file block size.  Block alignment is not
5026 		 * considered.  Partial arc_buf will be loaned out for read.
5027 		 */
5028 		blksz = zp->z_blksz;
5029 		if (blksz < zcr_blksz_min)
5030 			blksz = zcr_blksz_min;
5031 		if (blksz > zcr_blksz_max)
5032 			blksz = zcr_blksz_max;
5033 		/* avoid potential complexity of dealing with it */
5034 		if (blksz > max_blksz) {
5035 			ZFS_EXIT(zfsvfs);
5036 			return (EINVAL);
5037 		}
5038 
5039 		maxsize = zp->z_size - uio->uio_loffset;
5040 		if (size > maxsize)
5041 			size = maxsize;
5042 
5043 		if (size < blksz || vn_has_cached_data(vp)) {
5044 			ZFS_EXIT(zfsvfs);
5045 			return (EINVAL);
5046 		}
5047 		break;
5048 	default:
5049 		ZFS_EXIT(zfsvfs);
5050 		return (EINVAL);
5051 	}
5052 
5053 	uio->uio_extflg = UIO_XUIO;
5054 	XUIO_XUZC_RW(xuio) = ioflag;
5055 	ZFS_EXIT(zfsvfs);
5056 	return (0);
5057 }
5058 
5059 /*ARGSUSED*/
5060 static int
5061 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5062 {
5063 	int i;
5064 	arc_buf_t *abuf;
5065 	int ioflag = XUIO_XUZC_RW(xuio);
5066 
5067 	ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5068 
5069 	i = dmu_xuio_cnt(xuio);
5070 	while (i-- > 0) {
5071 		abuf = dmu_xuio_arcbuf(xuio, i);
5072 		/*
5073 		 * if abuf == NULL, it must be a write buffer
5074 		 * that has been returned in zfs_write().
5075 		 */
5076 		if (abuf)
5077 			dmu_return_arcbuf(abuf);
5078 		ASSERT(abuf || ioflag == UIO_WRITE);
5079 	}
5080 
5081 	dmu_xuio_fini(xuio);
5082 	return (0);
5083 }
5084 
5085 /*
5086  * Predeclare these here so that the compiler assumes that
5087  * this is an "old style" function declaration that does
5088  * not include arguments => we won't get type mismatch errors
5089  * in the initializations that follow.
5090  */
5091 static int zfs_inval();
5092 static int zfs_isdir();
5093 
5094 static int
5095 zfs_inval()
5096 {
5097 	return (EINVAL);
5098 }
5099 
5100 static int
5101 zfs_isdir()
5102 {
5103 	return (EISDIR);
5104 }
5105 /*
5106  * Directory vnode operations template
5107  */
5108 vnodeops_t *zfs_dvnodeops;
5109 const fs_operation_def_t zfs_dvnodeops_template[] = {
5110 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5111 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5112 	VOPNAME_READ,		{ .error = zfs_isdir },
5113 	VOPNAME_WRITE,		{ .error = zfs_isdir },
5114 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5115 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5116 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5117 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5118 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5119 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5120 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5121 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5122 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5123 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
5124 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5125 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5126 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
5127 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5128 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5129 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5130 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5131 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5132 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5133 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5134 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
5135 	NULL,			NULL
5136 };
5137 
5138 /*
5139  * Regular file vnode operations template
5140  */
5141 vnodeops_t *zfs_fvnodeops;
5142 const fs_operation_def_t zfs_fvnodeops_template[] = {
5143 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5144 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5145 	VOPNAME_READ,		{ .vop_read = zfs_read },
5146 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
5147 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5148 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5149 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5150 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5151 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5152 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5153 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5154 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5155 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5156 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5157 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
5158 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
5159 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
5160 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
5161 	VOPNAME_MAP,		{ .vop_map = zfs_map },
5162 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
5163 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
5164 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5165 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5166 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5167 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5168 	VOPNAME_REQZCBUF, 	{ .vop_reqzcbuf = zfs_reqzcbuf },
5169 	VOPNAME_RETZCBUF, 	{ .vop_retzcbuf = zfs_retzcbuf },
5170 	NULL,			NULL
5171 };
5172 
5173 /*
5174  * Symbolic link vnode operations template
5175  */
5176 vnodeops_t *zfs_symvnodeops;
5177 const fs_operation_def_t zfs_symvnodeops_template[] = {
5178 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5179 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5180 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5181 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5182 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
5183 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5184 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5185 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5186 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5187 	NULL,			NULL
5188 };
5189 
5190 /*
5191  * special share hidden files vnode operations template
5192  */
5193 vnodeops_t *zfs_sharevnodeops;
5194 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5195 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5196 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5197 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5198 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5199 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5200 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5201 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5202 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5203 	NULL,			NULL
5204 };
5205 
5206 /*
5207  * Extended attribute directory vnode operations template
5208  *	This template is identical to the directory vnodes
5209  *	operation template except for restricted operations:
5210  *		VOP_MKDIR()
5211  *		VOP_SYMLINK()
5212  * Note that there are other restrictions embedded in:
5213  *	zfs_create()	- restrict type to VREG
5214  *	zfs_link()	- no links into/out of attribute space
5215  *	zfs_rename()	- no moves into/out of attribute space
5216  */
5217 vnodeops_t *zfs_xdvnodeops;
5218 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5219 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5220 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5221 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5222 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5223 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5224 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5225 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5226 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5227 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5228 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5229 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5230 	VOPNAME_MKDIR,		{ .error = zfs_inval },
5231 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5232 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5233 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
5234 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5235 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5236 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5237 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5238 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5239 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5240 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5241 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5242 	NULL,			NULL
5243 };
5244 
5245 /*
5246  * Error vnode operations template
5247  */
5248 vnodeops_t *zfs_evnodeops;
5249 const fs_operation_def_t zfs_evnodeops_template[] = {
5250 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5251 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5252 	NULL,			NULL
5253 };
5254