xref: /titanic_44/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 8339b41da2395f0525c46ceedfeb01961893ec44)
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, seq, 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, seq, 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_last_itx, 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(n, 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 		 */
870 		mutex_enter(&zp->z_acl_lock);
871 		if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
872 		    (S_IXUSR >> 6))) != 0 &&
873 		    (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
874 		    secpolicy_vnode_setid_retain(cr,
875 		    (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
876 			uint64_t newmode;
877 			zp->z_mode &= ~(S_ISUID | S_ISGID);
878 			newmode = zp->z_mode;
879 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
880 			    (void *)&newmode, sizeof (uint64_t), tx);
881 		}
882 		mutex_exit(&zp->z_acl_lock);
883 
884 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
885 		    B_TRUE);
886 
887 		/*
888 		 * Update the file size (zp_size) if it has changed;
889 		 * account for possible concurrent updates.
890 		 */
891 		while ((end_size = zp->z_size) < uio->uio_loffset) {
892 			(void) atomic_cas_64(&zp->z_size, end_size,
893 			    uio->uio_loffset);
894 			ASSERT(error == 0);
895 		}
896 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
897 
898 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
899 		dmu_tx_commit(tx);
900 
901 		if (error != 0)
902 			break;
903 		ASSERT(tx_bytes == nbytes);
904 		n -= nbytes;
905 	}
906 
907 	zfs_range_unlock(rl);
908 
909 	/*
910 	 * If we're in replay mode, or we made no progress, return error.
911 	 * Otherwise, it's at least a partial write, so it's successful.
912 	 */
913 	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
914 		ZFS_EXIT(zfsvfs);
915 		return (error);
916 	}
917 
918 	if (ioflag & (FSYNC | FDSYNC) ||
919 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
920 		zil_commit(zilog, zp->z_last_itx, zp->z_id);
921 
922 	ZFS_EXIT(zfsvfs);
923 	return (0);
924 }
925 
926 void
927 zfs_get_done(zgd_t *zgd, int error)
928 {
929 	znode_t *zp = zgd->zgd_private;
930 	objset_t *os = zp->z_zfsvfs->z_os;
931 
932 	if (zgd->zgd_db)
933 		dmu_buf_rele(zgd->zgd_db, zgd);
934 
935 	zfs_range_unlock(zgd->zgd_rl);
936 
937 	/*
938 	 * Release the vnode asynchronously as we currently have the
939 	 * txg stopped from syncing.
940 	 */
941 	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
942 
943 	if (error == 0 && zgd->zgd_bp)
944 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
945 
946 	kmem_free(zgd, sizeof (zgd_t));
947 }
948 
949 #ifdef DEBUG
950 static int zil_fault_io = 0;
951 #endif
952 
953 /*
954  * Get data to generate a TX_WRITE intent log record.
955  */
956 int
957 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
958 {
959 	zfsvfs_t *zfsvfs = arg;
960 	objset_t *os = zfsvfs->z_os;
961 	znode_t *zp;
962 	uint64_t object = lr->lr_foid;
963 	uint64_t offset = lr->lr_offset;
964 	uint64_t size = lr->lr_length;
965 	blkptr_t *bp = &lr->lr_blkptr;
966 	dmu_buf_t *db;
967 	zgd_t *zgd;
968 	int error = 0;
969 
970 	ASSERT(zio != NULL);
971 	ASSERT(size != 0);
972 
973 	/*
974 	 * Nothing to do if the file has been removed
975 	 */
976 	if (zfs_zget(zfsvfs, object, &zp) != 0)
977 		return (ENOENT);
978 	if (zp->z_unlinked) {
979 		/*
980 		 * Release the vnode asynchronously as we currently have the
981 		 * txg stopped from syncing.
982 		 */
983 		VN_RELE_ASYNC(ZTOV(zp),
984 		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
985 		return (ENOENT);
986 	}
987 
988 	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
989 	zgd->zgd_zilog = zfsvfs->z_log;
990 	zgd->zgd_private = zp;
991 
992 	/*
993 	 * Write records come in two flavors: immediate and indirect.
994 	 * For small writes it's cheaper to store the data with the
995 	 * log record (immediate); for large writes it's cheaper to
996 	 * sync the data and get a pointer to it (indirect) so that
997 	 * we don't have to write the data twice.
998 	 */
999 	if (buf != NULL) { /* immediate write */
1000 		zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1001 		/* test for truncation needs to be done while range locked */
1002 		if (offset >= zp->z_size) {
1003 			error = ENOENT;
1004 		} else {
1005 			error = dmu_read(os, object, offset, size, buf,
1006 			    DMU_READ_NO_PREFETCH);
1007 		}
1008 		ASSERT(error == 0 || error == ENOENT);
1009 	} else { /* indirect write */
1010 		/*
1011 		 * Have to lock the whole block to ensure when it's
1012 		 * written out and it's checksum is being calculated
1013 		 * that no one can change the data. We need to re-check
1014 		 * blocksize after we get the lock in case it's changed!
1015 		 */
1016 		for (;;) {
1017 			uint64_t blkoff;
1018 			size = zp->z_blksz;
1019 			blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1020 			offset -= blkoff;
1021 			zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1022 			    RL_READER);
1023 			if (zp->z_blksz == size)
1024 				break;
1025 			offset += blkoff;
1026 			zfs_range_unlock(zgd->zgd_rl);
1027 		}
1028 		/* test for truncation needs to be done while range locked */
1029 		if (lr->lr_offset >= zp->z_size)
1030 			error = ENOENT;
1031 #ifdef DEBUG
1032 		if (zil_fault_io) {
1033 			error = EIO;
1034 			zil_fault_io = 0;
1035 		}
1036 #endif
1037 		if (error == 0)
1038 			error = dmu_buf_hold(os, object, offset, zgd, &db,
1039 			    DMU_READ_NO_PREFETCH);
1040 
1041 		if (error == 0) {
1042 			zgd->zgd_db = db;
1043 			zgd->zgd_bp = bp;
1044 
1045 			ASSERT(db->db_offset == offset);
1046 			ASSERT(db->db_size == size);
1047 
1048 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1049 			    zfs_get_done, zgd);
1050 			ASSERT(error || lr->lr_length <= zp->z_blksz);
1051 
1052 			/*
1053 			 * On success, we need to wait for the write I/O
1054 			 * initiated by dmu_sync() to complete before we can
1055 			 * release this dbuf.  We will finish everything up
1056 			 * in the zfs_get_done() callback.
1057 			 */
1058 			if (error == 0)
1059 				return (0);
1060 
1061 			if (error == EALREADY) {
1062 				lr->lr_common.lrc_txtype = TX_WRITE2;
1063 				error = 0;
1064 			}
1065 		}
1066 	}
1067 
1068 	zfs_get_done(zgd, error);
1069 
1070 	return (error);
1071 }
1072 
1073 /*ARGSUSED*/
1074 static int
1075 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1076     caller_context_t *ct)
1077 {
1078 	znode_t *zp = VTOZ(vp);
1079 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1080 	int error;
1081 
1082 	ZFS_ENTER(zfsvfs);
1083 	ZFS_VERIFY_ZP(zp);
1084 
1085 	if (flag & V_ACE_MASK)
1086 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1087 	else
1088 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
1089 
1090 	ZFS_EXIT(zfsvfs);
1091 	return (error);
1092 }
1093 
1094 /*
1095  * If vnode is for a device return a specfs vnode instead.
1096  */
1097 static int
1098 specvp_check(vnode_t **vpp, cred_t *cr)
1099 {
1100 	int error = 0;
1101 
1102 	if (IS_DEVVP(*vpp)) {
1103 		struct vnode *svp;
1104 
1105 		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1106 		VN_RELE(*vpp);
1107 		if (svp == NULL)
1108 			error = ENOSYS;
1109 		*vpp = svp;
1110 	}
1111 	return (error);
1112 }
1113 
1114 
1115 /*
1116  * Lookup an entry in a directory, or an extended attribute directory.
1117  * If it exists, return a held vnode reference for it.
1118  *
1119  *	IN:	dvp	- vnode of directory to search.
1120  *		nm	- name of entry to lookup.
1121  *		pnp	- full pathname to lookup [UNUSED].
1122  *		flags	- LOOKUP_XATTR set if looking for an attribute.
1123  *		rdir	- root directory vnode [UNUSED].
1124  *		cr	- credentials of caller.
1125  *		ct	- caller context
1126  *		direntflags - directory lookup flags
1127  *		realpnp - returned pathname.
1128  *
1129  *	OUT:	vpp	- vnode of located entry, NULL if not found.
1130  *
1131  *	RETURN:	0 if success
1132  *		error code if failure
1133  *
1134  * Timestamps:
1135  *	NA
1136  */
1137 /* ARGSUSED */
1138 static int
1139 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1140     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1141     int *direntflags, pathname_t *realpnp)
1142 {
1143 	znode_t *zdp = VTOZ(dvp);
1144 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1145 	int	error = 0;
1146 
1147 	/* fast path */
1148 	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1149 
1150 		if (dvp->v_type != VDIR) {
1151 			return (ENOTDIR);
1152 		} else if (zdp->z_sa_hdl == NULL) {
1153 			return (EIO);
1154 		}
1155 
1156 		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1157 			error = zfs_fastaccesschk_execute(zdp, cr);
1158 			if (!error) {
1159 				*vpp = dvp;
1160 				VN_HOLD(*vpp);
1161 				return (0);
1162 			}
1163 			return (error);
1164 		} else {
1165 			vnode_t *tvp = dnlc_lookup(dvp, nm);
1166 
1167 			if (tvp) {
1168 				error = zfs_fastaccesschk_execute(zdp, cr);
1169 				if (error) {
1170 					VN_RELE(tvp);
1171 					return (error);
1172 				}
1173 				if (tvp == DNLC_NO_VNODE) {
1174 					VN_RELE(tvp);
1175 					return (ENOENT);
1176 				} else {
1177 					*vpp = tvp;
1178 					return (specvp_check(vpp, cr));
1179 				}
1180 			}
1181 		}
1182 	}
1183 
1184 	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1185 
1186 	ZFS_ENTER(zfsvfs);
1187 	ZFS_VERIFY_ZP(zdp);
1188 
1189 	*vpp = NULL;
1190 
1191 	if (flags & LOOKUP_XATTR) {
1192 		/*
1193 		 * If the xattr property is off, refuse the lookup request.
1194 		 */
1195 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1196 			ZFS_EXIT(zfsvfs);
1197 			return (EINVAL);
1198 		}
1199 
1200 		/*
1201 		 * We don't allow recursive attributes..
1202 		 * Maybe someday we will.
1203 		 */
1204 		if (zdp->z_pflags & ZFS_XATTR) {
1205 			ZFS_EXIT(zfsvfs);
1206 			return (EINVAL);
1207 		}
1208 
1209 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1210 			ZFS_EXIT(zfsvfs);
1211 			return (error);
1212 		}
1213 
1214 		/*
1215 		 * Do we have permission to get into attribute directory?
1216 		 */
1217 
1218 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1219 		    B_FALSE, cr)) {
1220 			VN_RELE(*vpp);
1221 			*vpp = NULL;
1222 		}
1223 
1224 		ZFS_EXIT(zfsvfs);
1225 		return (error);
1226 	}
1227 
1228 	if (dvp->v_type != VDIR) {
1229 		ZFS_EXIT(zfsvfs);
1230 		return (ENOTDIR);
1231 	}
1232 
1233 	/*
1234 	 * Check accessibility of directory.
1235 	 */
1236 
1237 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1238 		ZFS_EXIT(zfsvfs);
1239 		return (error);
1240 	}
1241 
1242 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1243 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1244 		ZFS_EXIT(zfsvfs);
1245 		return (EILSEQ);
1246 	}
1247 
1248 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1249 	if (error == 0)
1250 		error = specvp_check(vpp, cr);
1251 
1252 	ZFS_EXIT(zfsvfs);
1253 	return (error);
1254 }
1255 
1256 /*
1257  * Attempt to create a new entry in a directory.  If the entry
1258  * already exists, truncate the file if permissible, else return
1259  * an error.  Return the vp of the created or trunc'd file.
1260  *
1261  *	IN:	dvp	- vnode of directory to put new file entry in.
1262  *		name	- name of new file entry.
1263  *		vap	- attributes of new file.
1264  *		excl	- flag indicating exclusive or non-exclusive mode.
1265  *		mode	- mode to open file with.
1266  *		cr	- credentials of caller.
1267  *		flag	- large file flag [UNUSED].
1268  *		ct	- caller context
1269  *		vsecp 	- ACL to be set
1270  *
1271  *	OUT:	vpp	- vnode of created or trunc'd entry.
1272  *
1273  *	RETURN:	0 if success
1274  *		error code if failure
1275  *
1276  * Timestamps:
1277  *	dvp - ctime|mtime updated if new entry created
1278  *	 vp - ctime|mtime always, atime if new
1279  */
1280 
1281 /* ARGSUSED */
1282 static int
1283 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1284     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1285     vsecattr_t *vsecp)
1286 {
1287 	znode_t		*zp, *dzp = VTOZ(dvp);
1288 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1289 	zilog_t		*zilog;
1290 	objset_t	*os;
1291 	zfs_dirlock_t	*dl;
1292 	dmu_tx_t	*tx;
1293 	int		error;
1294 	ksid_t		*ksid;
1295 	uid_t		uid;
1296 	gid_t		gid = crgetgid(cr);
1297 	zfs_acl_ids_t   acl_ids;
1298 	boolean_t	fuid_dirtied;
1299 	boolean_t	have_acl = B_FALSE;
1300 
1301 	/*
1302 	 * If we have an ephemeral id, ACL, or XVATTR then
1303 	 * make sure file system is at proper version
1304 	 */
1305 
1306 	ksid = crgetsid(cr, KSID_OWNER);
1307 	if (ksid)
1308 		uid = ksid_getid(ksid);
1309 	else
1310 		uid = crgetuid(cr);
1311 
1312 	if (zfsvfs->z_use_fuids == B_FALSE &&
1313 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1314 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1315 		return (EINVAL);
1316 
1317 	ZFS_ENTER(zfsvfs);
1318 	ZFS_VERIFY_ZP(dzp);
1319 	os = zfsvfs->z_os;
1320 	zilog = zfsvfs->z_log;
1321 
1322 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1323 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1324 		ZFS_EXIT(zfsvfs);
1325 		return (EILSEQ);
1326 	}
1327 
1328 	if (vap->va_mask & AT_XVATTR) {
1329 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1330 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1331 			ZFS_EXIT(zfsvfs);
1332 			return (error);
1333 		}
1334 	}
1335 top:
1336 	*vpp = NULL;
1337 
1338 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1339 		vap->va_mode &= ~VSVTX;
1340 
1341 	if (*name == '\0') {
1342 		/*
1343 		 * Null component name refers to the directory itself.
1344 		 */
1345 		VN_HOLD(dvp);
1346 		zp = dzp;
1347 		dl = NULL;
1348 		error = 0;
1349 	} else {
1350 		/* possible VN_HOLD(zp) */
1351 		int zflg = 0;
1352 
1353 		if (flag & FIGNORECASE)
1354 			zflg |= ZCILOOK;
1355 
1356 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1357 		    NULL, NULL);
1358 		if (error) {
1359 			if (strcmp(name, "..") == 0)
1360 				error = EISDIR;
1361 			ZFS_EXIT(zfsvfs);
1362 			return (error);
1363 		}
1364 	}
1365 
1366 	if (zp == NULL) {
1367 		uint64_t txtype;
1368 
1369 		/*
1370 		 * Create a new file object and update the directory
1371 		 * to reference it.
1372 		 */
1373 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1374 			goto out;
1375 		}
1376 
1377 		/*
1378 		 * We only support the creation of regular files in
1379 		 * extended attribute directories.
1380 		 */
1381 
1382 		if ((dzp->z_pflags & ZFS_XATTR) &&
1383 		    (vap->va_type != VREG)) {
1384 			error = EINVAL;
1385 			goto out;
1386 		}
1387 
1388 		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1389 		    cr, vsecp, &acl_ids)) != 0)
1390 			goto out;
1391 		have_acl = B_TRUE;
1392 
1393 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1394 			zfs_acl_ids_free(&acl_ids);
1395 			error = EDQUOT;
1396 			goto out;
1397 		}
1398 
1399 		tx = dmu_tx_create(os);
1400 
1401 		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1402 		    ZFS_SA_BASE_ATTR_SIZE);
1403 
1404 		fuid_dirtied = zfsvfs->z_fuid_dirty;
1405 		if (fuid_dirtied)
1406 			zfs_fuid_txhold(zfsvfs, tx);
1407 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1408 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1409 		if (!zfsvfs->z_use_sa &&
1410 		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1411 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1412 			    0, acl_ids.z_aclp->z_acl_bytes);
1413 		}
1414 		error = dmu_tx_assign(tx, TXG_NOWAIT);
1415 		if (error) {
1416 			zfs_dirent_unlock(dl);
1417 			if (error == ERESTART) {
1418 				dmu_tx_wait(tx);
1419 				dmu_tx_abort(tx);
1420 				goto top;
1421 			}
1422 			zfs_acl_ids_free(&acl_ids);
1423 			dmu_tx_abort(tx);
1424 			ZFS_EXIT(zfsvfs);
1425 			return (error);
1426 		}
1427 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1428 
1429 		if (fuid_dirtied)
1430 			zfs_fuid_sync(zfsvfs, tx);
1431 
1432 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1433 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1434 		if (flag & FIGNORECASE)
1435 			txtype |= TX_CI;
1436 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1437 		    vsecp, acl_ids.z_fuidp, vap);
1438 		zfs_acl_ids_free(&acl_ids);
1439 		dmu_tx_commit(tx);
1440 	} else {
1441 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1442 
1443 		/*
1444 		 * A directory entry already exists for this name.
1445 		 */
1446 		/*
1447 		 * Can't truncate an existing file if in exclusive mode.
1448 		 */
1449 		if (excl == EXCL) {
1450 			error = EEXIST;
1451 			goto out;
1452 		}
1453 		/*
1454 		 * Can't open a directory for writing.
1455 		 */
1456 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1457 			error = EISDIR;
1458 			goto out;
1459 		}
1460 		/*
1461 		 * Verify requested access to file.
1462 		 */
1463 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1464 			goto out;
1465 		}
1466 
1467 		mutex_enter(&dzp->z_lock);
1468 		dzp->z_seq++;
1469 		mutex_exit(&dzp->z_lock);
1470 
1471 		/*
1472 		 * Truncate regular files if requested.
1473 		 */
1474 		if ((ZTOV(zp)->v_type == VREG) &&
1475 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1476 			/* we can't hold any locks when calling zfs_freesp() */
1477 			zfs_dirent_unlock(dl);
1478 			dl = NULL;
1479 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1480 			if (error == 0) {
1481 				vnevent_create(ZTOV(zp), ct);
1482 			}
1483 		}
1484 	}
1485 out:
1486 
1487 	if (dl)
1488 		zfs_dirent_unlock(dl);
1489 
1490 	if (error) {
1491 		if (zp)
1492 			VN_RELE(ZTOV(zp));
1493 	} else {
1494 		*vpp = ZTOV(zp);
1495 		error = specvp_check(vpp, cr);
1496 	}
1497 
1498 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1499 		zil_commit(zilog, UINT64_MAX, 0);
1500 
1501 	ZFS_EXIT(zfsvfs);
1502 	return (error);
1503 }
1504 
1505 /*
1506  * Remove an entry from a directory.
1507  *
1508  *	IN:	dvp	- vnode of directory to remove entry from.
1509  *		name	- name of entry to remove.
1510  *		cr	- credentials of caller.
1511  *		ct	- caller context
1512  *		flags	- case flags
1513  *
1514  *	RETURN:	0 if success
1515  *		error code if failure
1516  *
1517  * Timestamps:
1518  *	dvp - ctime|mtime
1519  *	 vp - ctime (if nlink > 0)
1520  */
1521 
1522 uint64_t null_xattr = 0;
1523 
1524 /*ARGSUSED*/
1525 static int
1526 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1527     int flags)
1528 {
1529 	znode_t		*zp, *dzp = VTOZ(dvp);
1530 	znode_t		*xzp = NULL;
1531 	vnode_t		*vp;
1532 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1533 	zilog_t		*zilog;
1534 	uint64_t	acl_obj, xattr_obj = 0;
1535 	uint64_t 	xattr_obj_unlinked = 0;
1536 	zfs_dirlock_t	*dl;
1537 	dmu_tx_t	*tx;
1538 	boolean_t	may_delete_now, delete_now = FALSE;
1539 	boolean_t	unlinked, toobig = FALSE;
1540 	uint64_t	txtype;
1541 	pathname_t	*realnmp = NULL;
1542 	pathname_t	realnm;
1543 	int		error;
1544 	int		zflg = ZEXISTS;
1545 
1546 	ZFS_ENTER(zfsvfs);
1547 	ZFS_VERIFY_ZP(dzp);
1548 	zilog = zfsvfs->z_log;
1549 
1550 	if (flags & FIGNORECASE) {
1551 		zflg |= ZCILOOK;
1552 		pn_alloc(&realnm);
1553 		realnmp = &realnm;
1554 	}
1555 
1556 top:
1557 	/*
1558 	 * Attempt to lock directory; fail if entry doesn't exist.
1559 	 */
1560 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1561 	    NULL, realnmp)) {
1562 		if (realnmp)
1563 			pn_free(realnmp);
1564 		ZFS_EXIT(zfsvfs);
1565 		return (error);
1566 	}
1567 
1568 	vp = ZTOV(zp);
1569 
1570 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1571 		goto out;
1572 	}
1573 
1574 	/*
1575 	 * Need to use rmdir for removing directories.
1576 	 */
1577 	if (vp->v_type == VDIR) {
1578 		error = EPERM;
1579 		goto out;
1580 	}
1581 
1582 	vnevent_remove(vp, dvp, name, ct);
1583 
1584 	if (realnmp)
1585 		dnlc_remove(dvp, realnmp->pn_buf);
1586 	else
1587 		dnlc_remove(dvp, name);
1588 
1589 	mutex_enter(&vp->v_lock);
1590 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1591 	mutex_exit(&vp->v_lock);
1592 
1593 	/*
1594 	 * We may delete the znode now, or we may put it in the unlinked set;
1595 	 * it depends on whether we're the last link, and on whether there are
1596 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1597 	 * allow for either case.
1598 	 */
1599 	tx = dmu_tx_create(zfsvfs->z_os);
1600 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1601 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1602 	zfs_sa_upgrade_txholds(tx, zp);
1603 	zfs_sa_upgrade_txholds(tx, dzp);
1604 	if (may_delete_now) {
1605 		toobig =
1606 		    zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1607 		/* if the file is too big, only hold_free a token amount */
1608 		dmu_tx_hold_free(tx, zp->z_id, 0,
1609 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1610 	}
1611 
1612 	/* are there any extended attributes? */
1613 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1614 	    &xattr_obj, sizeof (xattr_obj));
1615 	if (xattr_obj) {
1616 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1617 		ASSERT3U(error, ==, 0);
1618 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1619 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1620 	}
1621 
1622 	/* are there any additional acls */
1623 	if ((acl_obj = ZFS_EXTERNAL_ACL(zp)) != 0 && may_delete_now)
1624 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1625 
1626 	/* charge as an update -- would be nice not to charge at all */
1627 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1628 
1629 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1630 	if (error) {
1631 		zfs_dirent_unlock(dl);
1632 		VN_RELE(vp);
1633 		if (error == ERESTART) {
1634 			dmu_tx_wait(tx);
1635 			dmu_tx_abort(tx);
1636 			goto top;
1637 		}
1638 		if (realnmp)
1639 			pn_free(realnmp);
1640 		dmu_tx_abort(tx);
1641 		ZFS_EXIT(zfsvfs);
1642 		return (error);
1643 	}
1644 
1645 	/*
1646 	 * Remove the directory entry.
1647 	 */
1648 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1649 
1650 	if (error) {
1651 		dmu_tx_commit(tx);
1652 		goto out;
1653 	}
1654 
1655 	if (unlinked) {
1656 
1657 		mutex_enter(&vp->v_lock);
1658 
1659 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1660 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1661 		delete_now = may_delete_now && !toobig &&
1662 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1663 		    xattr_obj == xattr_obj_unlinked && ZFS_EXTERNAL_ACL(zp) ==
1664 		    acl_obj;
1665 		mutex_exit(&vp->v_lock);
1666 	}
1667 
1668 	if (delete_now) {
1669 		if (xattr_obj_unlinked) {
1670 			ASSERT3U(xzp->z_links, ==, 2);
1671 			mutex_enter(&xzp->z_lock);
1672 			xzp->z_unlinked = 1;
1673 			xzp->z_links = 0;
1674 			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1675 			    &xzp->z_links, sizeof (xzp->z_links), tx);
1676 			ASSERT3U(error,  ==,  0);
1677 			mutex_exit(&xzp->z_lock);
1678 			zfs_unlinked_add(xzp, tx);
1679 			if (zp->z_is_sa)
1680 				error = sa_remove(zp->z_sa_hdl,
1681 				    SA_ZPL_XATTR(zfsvfs), tx);
1682 			else
1683 				error = sa_update(zp->z_sa_hdl,
1684 				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
1685 				    sizeof (uint64_t), tx);
1686 			ASSERT3U(error, ==, 0);
1687 		}
1688 		mutex_enter(&zp->z_lock);
1689 		mutex_enter(&vp->v_lock);
1690 		vp->v_count--;
1691 		ASSERT3U(vp->v_count, ==, 0);
1692 		mutex_exit(&vp->v_lock);
1693 		mutex_exit(&zp->z_lock);
1694 		zfs_znode_delete(zp, tx);
1695 	} else if (unlinked) {
1696 		zfs_unlinked_add(zp, tx);
1697 	}
1698 
1699 	txtype = TX_REMOVE;
1700 	if (flags & FIGNORECASE)
1701 		txtype |= TX_CI;
1702 	zfs_log_remove(zilog, tx, txtype, dzp, name);
1703 
1704 	dmu_tx_commit(tx);
1705 out:
1706 	if (realnmp)
1707 		pn_free(realnmp);
1708 
1709 	zfs_dirent_unlock(dl);
1710 
1711 	if (!delete_now)
1712 		VN_RELE(vp);
1713 	if (xzp)
1714 		VN_RELE(ZTOV(xzp));
1715 
1716 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1717 		zil_commit(zilog, UINT64_MAX, 0);
1718 
1719 	ZFS_EXIT(zfsvfs);
1720 	return (error);
1721 }
1722 
1723 /*
1724  * Create a new directory and insert it into dvp using the name
1725  * provided.  Return a pointer to the inserted directory.
1726  *
1727  *	IN:	dvp	- vnode of directory to add subdir to.
1728  *		dirname	- name of new directory.
1729  *		vap	- attributes of new directory.
1730  *		cr	- credentials of caller.
1731  *		ct	- caller context
1732  *		vsecp	- ACL to be set
1733  *
1734  *	OUT:	vpp	- vnode of created directory.
1735  *
1736  *	RETURN:	0 if success
1737  *		error code if failure
1738  *
1739  * Timestamps:
1740  *	dvp - ctime|mtime updated
1741  *	 vp - ctime|mtime|atime updated
1742  */
1743 /*ARGSUSED*/
1744 static int
1745 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1746     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1747 {
1748 	znode_t		*zp, *dzp = VTOZ(dvp);
1749 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1750 	zilog_t		*zilog;
1751 	zfs_dirlock_t	*dl;
1752 	uint64_t	txtype;
1753 	dmu_tx_t	*tx;
1754 	int		error;
1755 	int		zf = ZNEW;
1756 	ksid_t		*ksid;
1757 	uid_t		uid;
1758 	gid_t		gid = crgetgid(cr);
1759 	zfs_acl_ids_t   acl_ids;
1760 	boolean_t	fuid_dirtied;
1761 
1762 	ASSERT(vap->va_type == VDIR);
1763 
1764 	/*
1765 	 * If we have an ephemeral id, ACL, or XVATTR then
1766 	 * make sure file system is at proper version
1767 	 */
1768 
1769 	ksid = crgetsid(cr, KSID_OWNER);
1770 	if (ksid)
1771 		uid = ksid_getid(ksid);
1772 	else
1773 		uid = crgetuid(cr);
1774 	if (zfsvfs->z_use_fuids == B_FALSE &&
1775 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1776 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1777 		return (EINVAL);
1778 
1779 	ZFS_ENTER(zfsvfs);
1780 	ZFS_VERIFY_ZP(dzp);
1781 	zilog = zfsvfs->z_log;
1782 
1783 	if (dzp->z_pflags & ZFS_XATTR) {
1784 		ZFS_EXIT(zfsvfs);
1785 		return (EINVAL);
1786 	}
1787 
1788 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1789 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1790 		ZFS_EXIT(zfsvfs);
1791 		return (EILSEQ);
1792 	}
1793 	if (flags & FIGNORECASE)
1794 		zf |= ZCILOOK;
1795 
1796 	if (vap->va_mask & AT_XVATTR) {
1797 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1798 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1799 			ZFS_EXIT(zfsvfs);
1800 			return (error);
1801 		}
1802 	}
1803 
1804 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1805 	    vsecp, &acl_ids)) != 0) {
1806 		ZFS_EXIT(zfsvfs);
1807 		return (error);
1808 	}
1809 	/*
1810 	 * First make sure the new directory doesn't exist.
1811 	 *
1812 	 * Existence is checked first to make sure we don't return
1813 	 * EACCES instead of EEXIST which can cause some applications
1814 	 * to fail.
1815 	 */
1816 top:
1817 	*vpp = NULL;
1818 
1819 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1820 	    NULL, NULL)) {
1821 		zfs_acl_ids_free(&acl_ids);
1822 		ZFS_EXIT(zfsvfs);
1823 		return (error);
1824 	}
1825 
1826 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1827 		zfs_acl_ids_free(&acl_ids);
1828 		zfs_dirent_unlock(dl);
1829 		ZFS_EXIT(zfsvfs);
1830 		return (error);
1831 	}
1832 
1833 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1834 		zfs_acl_ids_free(&acl_ids);
1835 		zfs_dirent_unlock(dl);
1836 		ZFS_EXIT(zfsvfs);
1837 		return (EDQUOT);
1838 	}
1839 
1840 	/*
1841 	 * Add a new entry to the directory.
1842 	 */
1843 	tx = dmu_tx_create(zfsvfs->z_os);
1844 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1845 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1846 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1847 	if (fuid_dirtied)
1848 		zfs_fuid_txhold(zfsvfs, tx);
1849 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1850 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1851 		    acl_ids.z_aclp->z_acl_bytes);
1852 	}
1853 
1854 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1855 	    ZFS_SA_BASE_ATTR_SIZE);
1856 
1857 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1858 	if (error) {
1859 		zfs_dirent_unlock(dl);
1860 		if (error == ERESTART) {
1861 			dmu_tx_wait(tx);
1862 			dmu_tx_abort(tx);
1863 			goto top;
1864 		}
1865 		zfs_acl_ids_free(&acl_ids);
1866 		dmu_tx_abort(tx);
1867 		ZFS_EXIT(zfsvfs);
1868 		return (error);
1869 	}
1870 
1871 	/*
1872 	 * Create new node.
1873 	 */
1874 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1875 
1876 	if (fuid_dirtied)
1877 		zfs_fuid_sync(zfsvfs, tx);
1878 
1879 	/*
1880 	 * Now put new name in parent dir.
1881 	 */
1882 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1883 
1884 	*vpp = ZTOV(zp);
1885 
1886 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1887 	if (flags & FIGNORECASE)
1888 		txtype |= TX_CI;
1889 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1890 	    acl_ids.z_fuidp, vap);
1891 
1892 	zfs_acl_ids_free(&acl_ids);
1893 
1894 	dmu_tx_commit(tx);
1895 
1896 	zfs_dirent_unlock(dl);
1897 
1898 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1899 		zil_commit(zilog, UINT64_MAX, 0);
1900 
1901 	ZFS_EXIT(zfsvfs);
1902 	return (0);
1903 }
1904 
1905 /*
1906  * Remove a directory subdir entry.  If the current working
1907  * directory is the same as the subdir to be removed, the
1908  * remove will fail.
1909  *
1910  *	IN:	dvp	- vnode of directory to remove from.
1911  *		name	- name of directory to be removed.
1912  *		cwd	- vnode of current working directory.
1913  *		cr	- credentials of caller.
1914  *		ct	- caller context
1915  *		flags	- case flags
1916  *
1917  *	RETURN:	0 if success
1918  *		error code if failure
1919  *
1920  * Timestamps:
1921  *	dvp - ctime|mtime updated
1922  */
1923 /*ARGSUSED*/
1924 static int
1925 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1926     caller_context_t *ct, int flags)
1927 {
1928 	znode_t		*dzp = VTOZ(dvp);
1929 	znode_t		*zp;
1930 	vnode_t		*vp;
1931 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1932 	zilog_t		*zilog;
1933 	zfs_dirlock_t	*dl;
1934 	dmu_tx_t	*tx;
1935 	int		error;
1936 	int		zflg = ZEXISTS;
1937 
1938 	ZFS_ENTER(zfsvfs);
1939 	ZFS_VERIFY_ZP(dzp);
1940 	zilog = zfsvfs->z_log;
1941 
1942 	if (flags & FIGNORECASE)
1943 		zflg |= ZCILOOK;
1944 top:
1945 	zp = NULL;
1946 
1947 	/*
1948 	 * Attempt to lock directory; fail if entry doesn't exist.
1949 	 */
1950 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1951 	    NULL, NULL)) {
1952 		ZFS_EXIT(zfsvfs);
1953 		return (error);
1954 	}
1955 
1956 	vp = ZTOV(zp);
1957 
1958 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1959 		goto out;
1960 	}
1961 
1962 	if (vp->v_type != VDIR) {
1963 		error = ENOTDIR;
1964 		goto out;
1965 	}
1966 
1967 	if (vp == cwd) {
1968 		error = EINVAL;
1969 		goto out;
1970 	}
1971 
1972 	vnevent_rmdir(vp, dvp, name, ct);
1973 
1974 	/*
1975 	 * Grab a lock on the directory to make sure that noone is
1976 	 * trying to add (or lookup) entries while we are removing it.
1977 	 */
1978 	rw_enter(&zp->z_name_lock, RW_WRITER);
1979 
1980 	/*
1981 	 * Grab a lock on the parent pointer to make sure we play well
1982 	 * with the treewalk and directory rename code.
1983 	 */
1984 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1985 
1986 	tx = dmu_tx_create(zfsvfs->z_os);
1987 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1988 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1989 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1990 	zfs_sa_upgrade_txholds(tx, zp);
1991 	zfs_sa_upgrade_txholds(tx, dzp);
1992 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1993 	if (error) {
1994 		rw_exit(&zp->z_parent_lock);
1995 		rw_exit(&zp->z_name_lock);
1996 		zfs_dirent_unlock(dl);
1997 		VN_RELE(vp);
1998 		if (error == ERESTART) {
1999 			dmu_tx_wait(tx);
2000 			dmu_tx_abort(tx);
2001 			goto top;
2002 		}
2003 		dmu_tx_abort(tx);
2004 		ZFS_EXIT(zfsvfs);
2005 		return (error);
2006 	}
2007 
2008 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2009 
2010 	if (error == 0) {
2011 		uint64_t txtype = TX_RMDIR;
2012 		if (flags & FIGNORECASE)
2013 			txtype |= TX_CI;
2014 		zfs_log_remove(zilog, tx, txtype, dzp, name);
2015 	}
2016 
2017 	dmu_tx_commit(tx);
2018 
2019 	rw_exit(&zp->z_parent_lock);
2020 	rw_exit(&zp->z_name_lock);
2021 out:
2022 	zfs_dirent_unlock(dl);
2023 
2024 	VN_RELE(vp);
2025 
2026 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2027 		zil_commit(zilog, UINT64_MAX, 0);
2028 
2029 	ZFS_EXIT(zfsvfs);
2030 	return (error);
2031 }
2032 
2033 /*
2034  * Read as many directory entries as will fit into the provided
2035  * buffer from the given directory cursor position (specified in
2036  * the uio structure.
2037  *
2038  *	IN:	vp	- vnode of directory to read.
2039  *		uio	- structure supplying read location, range info,
2040  *			  and return buffer.
2041  *		cr	- credentials of caller.
2042  *		ct	- caller context
2043  *		flags	- case flags
2044  *
2045  *	OUT:	uio	- updated offset and range, buffer filled.
2046  *		eofp	- set to true if end-of-file detected.
2047  *
2048  *	RETURN:	0 if success
2049  *		error code if failure
2050  *
2051  * Timestamps:
2052  *	vp - atime updated
2053  *
2054  * Note that the low 4 bits of the cookie returned by zap is always zero.
2055  * This allows us to use the low range for "special" directory entries:
2056  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2057  * we use the offset 2 for the '.zfs' directory.
2058  */
2059 /* ARGSUSED */
2060 static int
2061 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2062     caller_context_t *ct, int flags)
2063 {
2064 	znode_t		*zp = VTOZ(vp);
2065 	iovec_t		*iovp;
2066 	edirent_t	*eodp;
2067 	dirent64_t	*odp;
2068 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2069 	objset_t	*os;
2070 	caddr_t		outbuf;
2071 	size_t		bufsize;
2072 	zap_cursor_t	zc;
2073 	zap_attribute_t	zap;
2074 	uint_t		bytes_wanted;
2075 	uint64_t	offset; /* must be unsigned; checks for < 1 */
2076 	uint64_t	parent;
2077 	int		local_eof;
2078 	int		outcount;
2079 	int		error;
2080 	uint8_t		prefetch;
2081 	boolean_t	check_sysattrs;
2082 
2083 	ZFS_ENTER(zfsvfs);
2084 	ZFS_VERIFY_ZP(zp);
2085 
2086 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2087 	    &parent, sizeof (parent))) != 0) {
2088 		ZFS_EXIT(zfsvfs);
2089 		return (error);
2090 	}
2091 
2092 	/*
2093 	 * If we are not given an eof variable,
2094 	 * use a local one.
2095 	 */
2096 	if (eofp == NULL)
2097 		eofp = &local_eof;
2098 
2099 	/*
2100 	 * Check for valid iov_len.
2101 	 */
2102 	if (uio->uio_iov->iov_len <= 0) {
2103 		ZFS_EXIT(zfsvfs);
2104 		return (EINVAL);
2105 	}
2106 
2107 	/*
2108 	 * Quit if directory has been removed (posix)
2109 	 */
2110 	if ((*eofp = zp->z_unlinked) != 0) {
2111 		ZFS_EXIT(zfsvfs);
2112 		return (0);
2113 	}
2114 
2115 	error = 0;
2116 	os = zfsvfs->z_os;
2117 	offset = uio->uio_loffset;
2118 	prefetch = zp->z_zn_prefetch;
2119 
2120 	/*
2121 	 * Initialize the iterator cursor.
2122 	 */
2123 	if (offset <= 3) {
2124 		/*
2125 		 * Start iteration from the beginning of the directory.
2126 		 */
2127 		zap_cursor_init(&zc, os, zp->z_id);
2128 	} else {
2129 		/*
2130 		 * The offset is a serialized cursor.
2131 		 */
2132 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2133 	}
2134 
2135 	/*
2136 	 * Get space to change directory entries into fs independent format.
2137 	 */
2138 	iovp = uio->uio_iov;
2139 	bytes_wanted = iovp->iov_len;
2140 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2141 		bufsize = bytes_wanted;
2142 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
2143 		odp = (struct dirent64 *)outbuf;
2144 	} else {
2145 		bufsize = bytes_wanted;
2146 		odp = (struct dirent64 *)iovp->iov_base;
2147 	}
2148 	eodp = (struct edirent *)odp;
2149 
2150 	/*
2151 	 * If this VFS supports the system attribute view interface; and
2152 	 * we're looking at an extended attribute directory; and we care
2153 	 * about normalization conflicts on this vfs; then we must check
2154 	 * for normalization conflicts with the sysattr name space.
2155 	 */
2156 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2157 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2158 	    (flags & V_RDDIR_ENTFLAGS);
2159 
2160 	/*
2161 	 * Transform to file-system independent format
2162 	 */
2163 	outcount = 0;
2164 	while (outcount < bytes_wanted) {
2165 		ino64_t objnum;
2166 		ushort_t reclen;
2167 		off64_t *next;
2168 
2169 		/*
2170 		 * Special case `.', `..', and `.zfs'.
2171 		 */
2172 		if (offset == 0) {
2173 			(void) strcpy(zap.za_name, ".");
2174 			zap.za_normalization_conflict = 0;
2175 			objnum = zp->z_id;
2176 		} else if (offset == 1) {
2177 			(void) strcpy(zap.za_name, "..");
2178 			zap.za_normalization_conflict = 0;
2179 			objnum = parent;
2180 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2181 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2182 			zap.za_normalization_conflict = 0;
2183 			objnum = ZFSCTL_INO_ROOT;
2184 		} else {
2185 			/*
2186 			 * Grab next entry.
2187 			 */
2188 			if (error = zap_cursor_retrieve(&zc, &zap)) {
2189 				if ((*eofp = (error == ENOENT)) != 0)
2190 					break;
2191 				else
2192 					goto update;
2193 			}
2194 
2195 			if (zap.za_integer_length != 8 ||
2196 			    zap.za_num_integers != 1) {
2197 				cmn_err(CE_WARN, "zap_readdir: bad directory "
2198 				    "entry, obj = %lld, offset = %lld\n",
2199 				    (u_longlong_t)zp->z_id,
2200 				    (u_longlong_t)offset);
2201 				error = ENXIO;
2202 				goto update;
2203 			}
2204 
2205 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2206 			/*
2207 			 * MacOS X can extract the object type here such as:
2208 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2209 			 */
2210 
2211 			if (check_sysattrs && !zap.za_normalization_conflict) {
2212 				zap.za_normalization_conflict =
2213 				    xattr_sysattr_casechk(zap.za_name);
2214 			}
2215 		}
2216 
2217 		if (flags & V_RDDIR_ACCFILTER) {
2218 			/*
2219 			 * If we have no access at all, don't include
2220 			 * this entry in the returned information
2221 			 */
2222 			znode_t	*ezp;
2223 			if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2224 				goto skip_entry;
2225 			if (!zfs_has_access(ezp, cr)) {
2226 				VN_RELE(ZTOV(ezp));
2227 				goto skip_entry;
2228 			}
2229 			VN_RELE(ZTOV(ezp));
2230 		}
2231 
2232 		if (flags & V_RDDIR_ENTFLAGS)
2233 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2234 		else
2235 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2236 
2237 		/*
2238 		 * Will this entry fit in the buffer?
2239 		 */
2240 		if (outcount + reclen > bufsize) {
2241 			/*
2242 			 * Did we manage to fit anything in the buffer?
2243 			 */
2244 			if (!outcount) {
2245 				error = EINVAL;
2246 				goto update;
2247 			}
2248 			break;
2249 		}
2250 		if (flags & V_RDDIR_ENTFLAGS) {
2251 			/*
2252 			 * Add extended flag entry:
2253 			 */
2254 			eodp->ed_ino = objnum;
2255 			eodp->ed_reclen = reclen;
2256 			/* NOTE: ed_off is the offset for the *next* entry */
2257 			next = &(eodp->ed_off);
2258 			eodp->ed_eflags = zap.za_normalization_conflict ?
2259 			    ED_CASE_CONFLICT : 0;
2260 			(void) strncpy(eodp->ed_name, zap.za_name,
2261 			    EDIRENT_NAMELEN(reclen));
2262 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2263 		} else {
2264 			/*
2265 			 * Add normal entry:
2266 			 */
2267 			odp->d_ino = objnum;
2268 			odp->d_reclen = reclen;
2269 			/* NOTE: d_off is the offset for the *next* entry */
2270 			next = &(odp->d_off);
2271 			(void) strncpy(odp->d_name, zap.za_name,
2272 			    DIRENT64_NAMELEN(reclen));
2273 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2274 		}
2275 		outcount += reclen;
2276 
2277 		ASSERT(outcount <= bufsize);
2278 
2279 		/* Prefetch znode */
2280 		if (prefetch)
2281 			dmu_prefetch(os, objnum, 0, 0);
2282 
2283 	skip_entry:
2284 		/*
2285 		 * Move to the next entry, fill in the previous offset.
2286 		 */
2287 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2288 			zap_cursor_advance(&zc);
2289 			offset = zap_cursor_serialize(&zc);
2290 		} else {
2291 			offset += 1;
2292 		}
2293 		*next = offset;
2294 	}
2295 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2296 
2297 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2298 		iovp->iov_base += outcount;
2299 		iovp->iov_len -= outcount;
2300 		uio->uio_resid -= outcount;
2301 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2302 		/*
2303 		 * Reset the pointer.
2304 		 */
2305 		offset = uio->uio_loffset;
2306 	}
2307 
2308 update:
2309 	zap_cursor_fini(&zc);
2310 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2311 		kmem_free(outbuf, bufsize);
2312 
2313 	if (error == ENOENT)
2314 		error = 0;
2315 
2316 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2317 
2318 	uio->uio_loffset = offset;
2319 	ZFS_EXIT(zfsvfs);
2320 	return (error);
2321 }
2322 
2323 ulong_t zfs_fsync_sync_cnt = 4;
2324 
2325 static int
2326 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2327 {
2328 	znode_t	*zp = VTOZ(vp);
2329 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2330 
2331 	/*
2332 	 * Regardless of whether this is required for standards conformance,
2333 	 * this is the logical behavior when fsync() is called on a file with
2334 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2335 	 * going to be pushed out as part of the zil_commit().
2336 	 */
2337 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2338 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2339 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2340 
2341 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2342 
2343 	if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2344 		ZFS_ENTER(zfsvfs);
2345 		ZFS_VERIFY_ZP(zp);
2346 		zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
2347 		ZFS_EXIT(zfsvfs);
2348 	}
2349 	return (0);
2350 }
2351 
2352 
2353 /*
2354  * Get the requested file attributes and place them in the provided
2355  * vattr structure.
2356  *
2357  *	IN:	vp	- vnode of file.
2358  *		vap	- va_mask identifies requested attributes.
2359  *			  If AT_XVATTR set, then optional attrs are requested
2360  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2361  *		cr	- credentials of caller.
2362  *		ct	- caller context
2363  *
2364  *	OUT:	vap	- attribute values.
2365  *
2366  *	RETURN:	0 (always succeeds)
2367  */
2368 /* ARGSUSED */
2369 static int
2370 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2371     caller_context_t *ct)
2372 {
2373 	znode_t *zp = VTOZ(vp);
2374 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2375 	int	error = 0;
2376 	uint64_t links;
2377 	uint64_t mtime[2], ctime[2];
2378 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2379 	xoptattr_t *xoap = NULL;
2380 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2381 	sa_bulk_attr_t bulk[2];
2382 	int count = 0;
2383 
2384 	ZFS_ENTER(zfsvfs);
2385 	ZFS_VERIFY_ZP(zp);
2386 
2387 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2388 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2389 
2390 	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2391 		ZFS_EXIT(zfsvfs);
2392 		return (error);
2393 	}
2394 
2395 	/*
2396 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2397 	 * Also, if we are the owner don't bother, since owner should
2398 	 * always be allowed to read basic attributes of file.
2399 	 */
2400 	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && (zp->z_uid != crgetuid(cr))) {
2401 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2402 		    skipaclchk, cr)) {
2403 			ZFS_EXIT(zfsvfs);
2404 			return (error);
2405 		}
2406 	}
2407 
2408 	/*
2409 	 * Return all attributes.  It's cheaper to provide the answer
2410 	 * than to determine whether we were asked the question.
2411 	 */
2412 
2413 	mutex_enter(&zp->z_lock);
2414 	vap->va_type = vp->v_type;
2415 	vap->va_mode = zp->z_mode & MODEMASK;
2416 	vap->va_uid = zp->z_uid;
2417 	vap->va_gid = zp->z_gid;
2418 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2419 	vap->va_nodeid = zp->z_id;
2420 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2421 		links = zp->z_links + 1;
2422 	else
2423 		links = zp->z_links;
2424 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2425 	vap->va_size = zp->z_size;
2426 	vap->va_rdev = vp->v_rdev;
2427 	vap->va_seq = zp->z_seq;
2428 
2429 	/*
2430 	 * Add in any requested optional attributes and the create time.
2431 	 * Also set the corresponding bits in the returned attribute bitmap.
2432 	 */
2433 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2434 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2435 			xoap->xoa_archive =
2436 			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2437 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2438 		}
2439 
2440 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2441 			xoap->xoa_readonly =
2442 			    ((zp->z_pflags & ZFS_READONLY) != 0);
2443 			XVA_SET_RTN(xvap, XAT_READONLY);
2444 		}
2445 
2446 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2447 			xoap->xoa_system =
2448 			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2449 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2450 		}
2451 
2452 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2453 			xoap->xoa_hidden =
2454 			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2455 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2456 		}
2457 
2458 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2459 			xoap->xoa_nounlink =
2460 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2461 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2462 		}
2463 
2464 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2465 			xoap->xoa_immutable =
2466 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2467 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2468 		}
2469 
2470 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2471 			xoap->xoa_appendonly =
2472 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2473 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2474 		}
2475 
2476 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2477 			xoap->xoa_nodump =
2478 			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2479 			XVA_SET_RTN(xvap, XAT_NODUMP);
2480 		}
2481 
2482 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2483 			xoap->xoa_opaque =
2484 			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2485 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2486 		}
2487 
2488 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2489 			xoap->xoa_av_quarantined =
2490 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2491 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2492 		}
2493 
2494 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2495 			xoap->xoa_av_modified =
2496 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2497 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2498 		}
2499 
2500 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2501 		    vp->v_type == VREG) {
2502 			zfs_sa_get_scanstamp(zp, xvap);
2503 		}
2504 
2505 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2506 			uint64_t times[2];
2507 
2508 			(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2509 			    times, sizeof (times));
2510 			ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2511 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2512 		}
2513 
2514 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2515 			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2516 			XVA_SET_RTN(xvap, XAT_REPARSE);
2517 		}
2518 	}
2519 
2520 	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2521 	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2522 	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2523 
2524 	mutex_exit(&zp->z_lock);
2525 
2526 	sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2527 
2528 	if (zp->z_blksz == 0) {
2529 		/*
2530 		 * Block size hasn't been set; suggest maximal I/O transfers.
2531 		 */
2532 		vap->va_blksize = zfsvfs->z_max_blksz;
2533 	}
2534 
2535 	ZFS_EXIT(zfsvfs);
2536 	return (0);
2537 }
2538 
2539 /*
2540  * Set the file attributes to the values contained in the
2541  * vattr structure.
2542  *
2543  *	IN:	vp	- vnode of file to be modified.
2544  *		vap	- new attribute values.
2545  *			  If AT_XVATTR set, then optional attrs are being set
2546  *		flags	- ATTR_UTIME set if non-default time values provided.
2547  *			- ATTR_NOACLCHECK (CIFS context only).
2548  *		cr	- credentials of caller.
2549  *		ct	- caller context
2550  *
2551  *	RETURN:	0 if success
2552  *		error code if failure
2553  *
2554  * Timestamps:
2555  *	vp - ctime updated, mtime updated if size changed.
2556  */
2557 /* ARGSUSED */
2558 static int
2559 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2560 	caller_context_t *ct)
2561 {
2562 	znode_t		*zp = VTOZ(vp);
2563 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2564 	zilog_t		*zilog;
2565 	dmu_tx_t	*tx;
2566 	vattr_t		oldva;
2567 	xvattr_t	tmpxvattr;
2568 	uint_t		mask = vap->va_mask;
2569 	uint_t		saved_mask;
2570 	int		trim_mask = 0;
2571 	uint64_t	new_mode;
2572 	uint64_t	new_uid, new_gid;
2573 	uint64_t	xattr_obj = 0;
2574 	uint64_t	mtime[2], ctime[2];
2575 	znode_t		*attrzp;
2576 	int		need_policy = FALSE;
2577 	int		err, err2;
2578 	zfs_fuid_info_t *fuidp = NULL;
2579 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2580 	xoptattr_t	*xoap;
2581 	zfs_acl_t	*aclp = NULL;
2582 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2583 	boolean_t	fuid_dirtied = B_FALSE;
2584 	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
2585 	int		count = 0, xattr_count = 0;
2586 
2587 	if (mask == 0)
2588 		return (0);
2589 
2590 	if (mask & AT_NOSET)
2591 		return (EINVAL);
2592 
2593 	ZFS_ENTER(zfsvfs);
2594 	ZFS_VERIFY_ZP(zp);
2595 
2596 	zilog = zfsvfs->z_log;
2597 
2598 	/*
2599 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2600 	 * that file system is at proper version level
2601 	 */
2602 
2603 	if (zfsvfs->z_use_fuids == B_FALSE &&
2604 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2605 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2606 	    (mask & AT_XVATTR))) {
2607 		ZFS_EXIT(zfsvfs);
2608 		return (EINVAL);
2609 	}
2610 
2611 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2612 		ZFS_EXIT(zfsvfs);
2613 		return (EISDIR);
2614 	}
2615 
2616 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2617 		ZFS_EXIT(zfsvfs);
2618 		return (EINVAL);
2619 	}
2620 
2621 	/*
2622 	 * If this is an xvattr_t, then get a pointer to the structure of
2623 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2624 	 */
2625 	xoap = xva_getxoptattr(xvap);
2626 
2627 	xva_init(&tmpxvattr);
2628 
2629 	/*
2630 	 * Immutable files can only alter immutable bit and atime
2631 	 */
2632 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2633 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2634 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2635 		ZFS_EXIT(zfsvfs);
2636 		return (EPERM);
2637 	}
2638 
2639 	if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2640 		ZFS_EXIT(zfsvfs);
2641 		return (EPERM);
2642 	}
2643 
2644 	/*
2645 	 * Verify timestamps doesn't overflow 32 bits.
2646 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2647 	 * handle times greater than 2039.  This check should be removed
2648 	 * once large timestamps are fully supported.
2649 	 */
2650 	if (mask & (AT_ATIME | AT_MTIME)) {
2651 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2652 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2653 			ZFS_EXIT(zfsvfs);
2654 			return (EOVERFLOW);
2655 		}
2656 	}
2657 
2658 top:
2659 	attrzp = NULL;
2660 
2661 	/* Can this be moved to before the top label? */
2662 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2663 		ZFS_EXIT(zfsvfs);
2664 		return (EROFS);
2665 	}
2666 
2667 	/*
2668 	 * First validate permissions
2669 	 */
2670 
2671 	if (mask & AT_SIZE) {
2672 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2673 		if (err) {
2674 			ZFS_EXIT(zfsvfs);
2675 			return (err);
2676 		}
2677 		/*
2678 		 * XXX - Note, we are not providing any open
2679 		 * mode flags here (like FNDELAY), so we may
2680 		 * block if there are locks present... this
2681 		 * should be addressed in openat().
2682 		 */
2683 		/* XXX - would it be OK to generate a log record here? */
2684 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2685 		if (err) {
2686 			ZFS_EXIT(zfsvfs);
2687 			return (err);
2688 		}
2689 	}
2690 
2691 	if (mask & (AT_ATIME|AT_MTIME) ||
2692 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2693 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2694 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2695 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2696 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2697 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2698 		    skipaclchk, cr);
2699 	}
2700 
2701 	if (mask & (AT_UID|AT_GID)) {
2702 		int	idmask = (mask & (AT_UID|AT_GID));
2703 		int	take_owner;
2704 		int	take_group;
2705 
2706 		/*
2707 		 * NOTE: even if a new mode is being set,
2708 		 * we may clear S_ISUID/S_ISGID bits.
2709 		 */
2710 
2711 		if (!(mask & AT_MODE))
2712 			vap->va_mode = zp->z_mode;
2713 
2714 		/*
2715 		 * Take ownership or chgrp to group we are a member of
2716 		 */
2717 
2718 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2719 		take_group = (mask & AT_GID) &&
2720 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2721 
2722 		/*
2723 		 * If both AT_UID and AT_GID are set then take_owner and
2724 		 * take_group must both be set in order to allow taking
2725 		 * ownership.
2726 		 *
2727 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2728 		 *
2729 		 */
2730 
2731 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2732 		    ((idmask == AT_UID) && take_owner) ||
2733 		    ((idmask == AT_GID) && take_group)) {
2734 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2735 			    skipaclchk, cr) == 0) {
2736 				/*
2737 				 * Remove setuid/setgid for non-privileged users
2738 				 */
2739 				secpolicy_setid_clear(vap, cr);
2740 				trim_mask = (mask & (AT_UID|AT_GID));
2741 			} else {
2742 				need_policy =  TRUE;
2743 			}
2744 		} else {
2745 			need_policy =  TRUE;
2746 		}
2747 	}
2748 
2749 	mutex_enter(&zp->z_lock);
2750 	oldva.va_mode = zp->z_mode;
2751 	oldva.va_uid = zp->z_uid;
2752 	oldva.va_gid = zp->z_gid;
2753 	if (mask & AT_XVATTR) {
2754 		/*
2755 		 * Update xvattr mask to include only those attributes
2756 		 * that are actually changing.
2757 		 *
2758 		 * the bits will be restored prior to actually setting
2759 		 * the attributes so the caller thinks they were set.
2760 		 */
2761 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2762 			if (xoap->xoa_appendonly !=
2763 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2764 				need_policy = TRUE;
2765 			} else {
2766 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2767 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2768 			}
2769 		}
2770 
2771 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2772 			if (xoap->xoa_nounlink !=
2773 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2774 				need_policy = TRUE;
2775 			} else {
2776 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2777 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2778 			}
2779 		}
2780 
2781 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2782 			if (xoap->xoa_immutable !=
2783 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2784 				need_policy = TRUE;
2785 			} else {
2786 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2787 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2788 			}
2789 		}
2790 
2791 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2792 			if (xoap->xoa_nodump !=
2793 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2794 				need_policy = TRUE;
2795 			} else {
2796 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2797 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2798 			}
2799 		}
2800 
2801 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2802 			if (xoap->xoa_av_modified !=
2803 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2804 				need_policy = TRUE;
2805 			} else {
2806 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2807 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2808 			}
2809 		}
2810 
2811 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2812 			if ((vp->v_type != VREG &&
2813 			    xoap->xoa_av_quarantined) ||
2814 			    xoap->xoa_av_quarantined !=
2815 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2816 				need_policy = TRUE;
2817 			} else {
2818 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2819 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2820 			}
2821 		}
2822 
2823 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2824 			mutex_exit(&zp->z_lock);
2825 			ZFS_EXIT(zfsvfs);
2826 			return (EPERM);
2827 		}
2828 
2829 		if (need_policy == FALSE &&
2830 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2831 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2832 			need_policy = TRUE;
2833 		}
2834 	}
2835 
2836 	mutex_exit(&zp->z_lock);
2837 
2838 	if (mask & AT_MODE) {
2839 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2840 			err = secpolicy_setid_setsticky_clear(vp, vap,
2841 			    &oldva, cr);
2842 			if (err) {
2843 				ZFS_EXIT(zfsvfs);
2844 				return (err);
2845 			}
2846 			trim_mask |= AT_MODE;
2847 		} else {
2848 			need_policy = TRUE;
2849 		}
2850 	}
2851 
2852 	if (need_policy) {
2853 		/*
2854 		 * If trim_mask is set then take ownership
2855 		 * has been granted or write_acl is present and user
2856 		 * has the ability to modify mode.  In that case remove
2857 		 * UID|GID and or MODE from mask so that
2858 		 * secpolicy_vnode_setattr() doesn't revoke it.
2859 		 */
2860 
2861 		if (trim_mask) {
2862 			saved_mask = vap->va_mask;
2863 			vap->va_mask &= ~trim_mask;
2864 		}
2865 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2866 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2867 		if (err) {
2868 			ZFS_EXIT(zfsvfs);
2869 			return (err);
2870 		}
2871 
2872 		if (trim_mask)
2873 			vap->va_mask |= saved_mask;
2874 	}
2875 
2876 	/*
2877 	 * secpolicy_vnode_setattr, or take ownership may have
2878 	 * changed va_mask
2879 	 */
2880 	mask = vap->va_mask;
2881 
2882 	if ((mask & (AT_UID | AT_GID))) {
2883 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xattr_obj,
2884 		    sizeof (xattr_obj));
2885 
2886 		if (xattr_obj) {
2887 			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
2888 			if (err)
2889 				goto out2;
2890 		}
2891 		if (mask & AT_UID) {
2892 			new_uid = zfs_fuid_create(zfsvfs,
2893 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2894 			if (vap->va_uid != zp->z_uid &&
2895 			    zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
2896 				err = EDQUOT;
2897 				goto out2;
2898 			}
2899 		}
2900 
2901 		if (mask & AT_GID) {
2902 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
2903 			    cr, ZFS_GROUP, &fuidp);
2904 			if (new_gid != zp->z_gid &&
2905 			    zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
2906 				err = EDQUOT;
2907 				goto out2;
2908 			}
2909 		}
2910 	}
2911 	tx = dmu_tx_create(zfsvfs->z_os);
2912 
2913 	if (mask & AT_MODE) {
2914 		uint64_t pmode = zp->z_mode;
2915 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2916 
2917 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
2918 			goto out;
2919 
2920 		if (!zp->z_is_sa && ZFS_EXTERNAL_ACL(zp)) {
2921 			/*
2922 			 * Are we upgrading ACL from old V0 format
2923 			 * to V1 format?
2924 			 */
2925 			if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2926 			    ZNODE_ACL_VERSION(zp) ==
2927 			    ZFS_ACL_VERSION_INITIAL) {
2928 				dmu_tx_hold_free(tx,
2929 				    ZFS_EXTERNAL_ACL(zp), 0,
2930 				    DMU_OBJECT_END);
2931 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2932 				    0, aclp->z_acl_bytes);
2933 			} else {
2934 				dmu_tx_hold_write(tx, ZFS_EXTERNAL_ACL(zp), 0,
2935 				    aclp->z_acl_bytes);
2936 			}
2937 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2938 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2939 			    0, aclp->z_acl_bytes);
2940 		}
2941 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2942 	} else {
2943 		if ((mask & AT_XVATTR) &&
2944 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
2945 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2946 		else
2947 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2948 	}
2949 
2950 	if (attrzp) {
2951 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
2952 	}
2953 
2954 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2955 	if (fuid_dirtied)
2956 		zfs_fuid_txhold(zfsvfs, tx);
2957 
2958 	zfs_sa_upgrade_txholds(tx, zp);
2959 
2960 	err = dmu_tx_assign(tx, TXG_NOWAIT);
2961 	if (err) {
2962 		if (err == ERESTART)
2963 			dmu_tx_wait(tx);
2964 		goto out;
2965 	}
2966 
2967 	count = 0;
2968 	/*
2969 	 * Set each attribute requested.
2970 	 * We group settings according to the locks they need to acquire.
2971 	 *
2972 	 * Note: you cannot set ctime directly, although it will be
2973 	 * updated as a side-effect of calling this function.
2974 	 */
2975 
2976 	mutex_enter(&zp->z_lock);
2977 
2978 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
2979 	    &zp->z_pflags, sizeof (zp->z_pflags));
2980 
2981 	if (attrzp) {
2982 		mutex_enter(&attrzp->z_lock);
2983 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2984 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
2985 		    sizeof (attrzp->z_pflags));
2986 	}
2987 
2988 	if (mask & (AT_UID|AT_GID)) {
2989 
2990 		if (mask & AT_UID) {
2991 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
2992 			    &new_uid, sizeof (new_uid));
2993 			zp->z_uid = zfs_fuid_map_id(zfsvfs, new_uid,
2994 			    cr, ZFS_OWNER);
2995 			if (attrzp) {
2996 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
2997 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
2998 				    sizeof (new_uid));
2999 				attrzp->z_uid = zp->z_uid;
3000 			}
3001 		}
3002 
3003 		if (mask & AT_GID) {
3004 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3005 			    NULL, &new_gid, sizeof (new_gid));
3006 			zp->z_gid = zfs_fuid_map_id(zfsvfs, new_gid, cr,
3007 			    ZFS_GROUP);
3008 			if (attrzp) {
3009 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3010 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3011 				    sizeof (new_gid));
3012 				attrzp->z_gid = zp->z_gid;
3013 			}
3014 		}
3015 		if (!(mask & AT_MODE)) {
3016 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3017 			    NULL, &new_mode, sizeof (new_mode));
3018 			new_mode = zp->z_mode;
3019 		}
3020 		err = zfs_acl_chown_setattr(zp);
3021 		ASSERT(err == 0);
3022 		if (attrzp) {
3023 			err = zfs_acl_chown_setattr(attrzp);
3024 			ASSERT(err == 0);
3025 		}
3026 	}
3027 
3028 	if (mask & AT_MODE) {
3029 		mutex_enter(&zp->z_acl_lock);
3030 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3031 		    &new_mode, sizeof (new_mode));
3032 		zp->z_mode = new_mode;
3033 		ASSERT3U((uintptr_t)aclp, !=, NULL);
3034 		err = zfs_aclset_common(zp, aclp, cr, tx);
3035 		ASSERT3U(err, ==, 0);
3036 		zp->z_acl_cached = aclp;
3037 		aclp = NULL;
3038 		mutex_exit(&zp->z_acl_lock);
3039 	}
3040 
3041 	if (attrzp)
3042 		mutex_exit(&attrzp->z_lock);
3043 
3044 	if (mask & AT_ATIME) {
3045 		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3046 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3047 		    &zp->z_atime, sizeof (zp->z_atime));
3048 	}
3049 
3050 	if (mask & AT_MTIME) {
3051 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3052 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3053 		    mtime, sizeof (mtime));
3054 	}
3055 
3056 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3057 	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3058 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3059 		    NULL, mtime, sizeof (mtime));
3060 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3061 		    &ctime, sizeof (ctime));
3062 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3063 		    B_TRUE);
3064 	} else if (mask != 0) {
3065 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3066 		    &ctime, sizeof (ctime));
3067 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3068 		    B_TRUE);
3069 		if (attrzp) {
3070 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3071 			    SA_ZPL_CTIME(zfsvfs), NULL,
3072 			    &ctime, sizeof (ctime));
3073 			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3074 			    mtime, ctime, B_TRUE);
3075 		}
3076 	}
3077 	/*
3078 	 * Do this after setting timestamps to prevent timestamp
3079 	 * update from toggling bit
3080 	 */
3081 
3082 	if (xoap && (mask & AT_XVATTR)) {
3083 
3084 		/*
3085 		 * restore trimmed off masks
3086 		 * so that return masks can be set for caller.
3087 		 */
3088 
3089 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3090 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
3091 		}
3092 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3093 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
3094 		}
3095 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3096 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3097 		}
3098 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3099 			XVA_SET_REQ(xvap, XAT_NODUMP);
3100 		}
3101 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3102 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3103 		}
3104 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3105 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3106 		}
3107 
3108 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3109 			ASSERT(vp->v_type == VREG);
3110 
3111 		zfs_xvattr_set(zp, xvap, tx);
3112 	}
3113 
3114 	if (fuid_dirtied)
3115 		zfs_fuid_sync(zfsvfs, tx);
3116 
3117 	if (mask != 0)
3118 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3119 
3120 	mutex_exit(&zp->z_lock);
3121 
3122 out:
3123 	if (err == 0 && attrzp) {
3124 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3125 		    xattr_count, tx);
3126 		ASSERT(err2 == 0);
3127 	}
3128 
3129 	if (attrzp)
3130 		VN_RELE(ZTOV(attrzp));
3131 	if (aclp)
3132 		zfs_acl_free(aclp);
3133 
3134 	if (fuidp) {
3135 		zfs_fuid_info_free(fuidp);
3136 		fuidp = NULL;
3137 	}
3138 
3139 	if (err) {
3140 		dmu_tx_abort(tx);
3141 		if (err == ERESTART)
3142 			goto top;
3143 	} else {
3144 		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3145 		dmu_tx_commit(tx);
3146 	}
3147 
3148 
3149 out2:
3150 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3151 		zil_commit(zilog, UINT64_MAX, 0);
3152 
3153 	ZFS_EXIT(zfsvfs);
3154 	return (err);
3155 }
3156 
3157 typedef struct zfs_zlock {
3158 	krwlock_t	*zl_rwlock;	/* lock we acquired */
3159 	znode_t		*zl_znode;	/* znode we held */
3160 	struct zfs_zlock *zl_next;	/* next in list */
3161 } zfs_zlock_t;
3162 
3163 /*
3164  * Drop locks and release vnodes that were held by zfs_rename_lock().
3165  */
3166 static void
3167 zfs_rename_unlock(zfs_zlock_t **zlpp)
3168 {
3169 	zfs_zlock_t *zl;
3170 
3171 	while ((zl = *zlpp) != NULL) {
3172 		if (zl->zl_znode != NULL)
3173 			VN_RELE(ZTOV(zl->zl_znode));
3174 		rw_exit(zl->zl_rwlock);
3175 		*zlpp = zl->zl_next;
3176 		kmem_free(zl, sizeof (*zl));
3177 	}
3178 }
3179 
3180 /*
3181  * Search back through the directory tree, using the ".." entries.
3182  * Lock each directory in the chain to prevent concurrent renames.
3183  * Fail any attempt to move a directory into one of its own descendants.
3184  * XXX - z_parent_lock can overlap with map or grow locks
3185  */
3186 static int
3187 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3188 {
3189 	zfs_zlock_t	*zl;
3190 	znode_t		*zp = tdzp;
3191 	uint64_t	rootid = zp->z_zfsvfs->z_root;
3192 	uint64_t	oidp = zp->z_id;
3193 	krwlock_t	*rwlp = &szp->z_parent_lock;
3194 	krw_t		rw = RW_WRITER;
3195 
3196 	/*
3197 	 * First pass write-locks szp and compares to zp->z_id.
3198 	 * Later passes read-lock zp and compare to zp->z_parent.
3199 	 */
3200 	do {
3201 		if (!rw_tryenter(rwlp, rw)) {
3202 			/*
3203 			 * Another thread is renaming in this path.
3204 			 * Note that if we are a WRITER, we don't have any
3205 			 * parent_locks held yet.
3206 			 */
3207 			if (rw == RW_READER && zp->z_id > szp->z_id) {
3208 				/*
3209 				 * Drop our locks and restart
3210 				 */
3211 				zfs_rename_unlock(&zl);
3212 				*zlpp = NULL;
3213 				zp = tdzp;
3214 				oidp = zp->z_id;
3215 				rwlp = &szp->z_parent_lock;
3216 				rw = RW_WRITER;
3217 				continue;
3218 			} else {
3219 				/*
3220 				 * Wait for other thread to drop its locks
3221 				 */
3222 				rw_enter(rwlp, rw);
3223 			}
3224 		}
3225 
3226 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3227 		zl->zl_rwlock = rwlp;
3228 		zl->zl_znode = NULL;
3229 		zl->zl_next = *zlpp;
3230 		*zlpp = zl;
3231 
3232 		if (oidp == szp->z_id)		/* We're a descendant of szp */
3233 			return (EINVAL);
3234 
3235 		if (oidp == rootid)		/* We've hit the top */
3236 			return (0);
3237 
3238 		if (rw == RW_READER) {		/* i.e. not the first pass */
3239 			int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3240 			if (error)
3241 				return (error);
3242 			zl->zl_znode = zp;
3243 		}
3244 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3245 		    &oidp, sizeof (oidp));
3246 		rwlp = &zp->z_parent_lock;
3247 		rw = RW_READER;
3248 
3249 	} while (zp->z_id != sdzp->z_id);
3250 
3251 	return (0);
3252 }
3253 
3254 /*
3255  * Move an entry from the provided source directory to the target
3256  * directory.  Change the entry name as indicated.
3257  *
3258  *	IN:	sdvp	- Source directory containing the "old entry".
3259  *		snm	- Old entry name.
3260  *		tdvp	- Target directory to contain the "new entry".
3261  *		tnm	- New entry name.
3262  *		cr	- credentials of caller.
3263  *		ct	- caller context
3264  *		flags	- case flags
3265  *
3266  *	RETURN:	0 if success
3267  *		error code if failure
3268  *
3269  * Timestamps:
3270  *	sdvp,tdvp - ctime|mtime updated
3271  */
3272 /*ARGSUSED*/
3273 static int
3274 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3275     caller_context_t *ct, int flags)
3276 {
3277 	znode_t		*tdzp, *szp, *tzp;
3278 	znode_t		*sdzp = VTOZ(sdvp);
3279 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
3280 	zilog_t		*zilog;
3281 	vnode_t		*realvp;
3282 	zfs_dirlock_t	*sdl, *tdl;
3283 	dmu_tx_t	*tx;
3284 	zfs_zlock_t	*zl;
3285 	int		cmp, serr, terr;
3286 	int		error = 0;
3287 	int		zflg = 0;
3288 
3289 	ZFS_ENTER(zfsvfs);
3290 	ZFS_VERIFY_ZP(sdzp);
3291 	zilog = zfsvfs->z_log;
3292 
3293 	/*
3294 	 * Make sure we have the real vp for the target directory.
3295 	 */
3296 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3297 		tdvp = realvp;
3298 
3299 	if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3300 		ZFS_EXIT(zfsvfs);
3301 		return (EXDEV);
3302 	}
3303 
3304 	tdzp = VTOZ(tdvp);
3305 	ZFS_VERIFY_ZP(tdzp);
3306 	if (zfsvfs->z_utf8 && u8_validate(tnm,
3307 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3308 		ZFS_EXIT(zfsvfs);
3309 		return (EILSEQ);
3310 	}
3311 
3312 	if (flags & FIGNORECASE)
3313 		zflg |= ZCILOOK;
3314 
3315 top:
3316 	szp = NULL;
3317 	tzp = NULL;
3318 	zl = NULL;
3319 
3320 	/*
3321 	 * This is to prevent the creation of links into attribute space
3322 	 * by renaming a linked file into/outof an attribute directory.
3323 	 * See the comment in zfs_link() for why this is considered bad.
3324 	 */
3325 	if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3326 		ZFS_EXIT(zfsvfs);
3327 		return (EINVAL);
3328 	}
3329 
3330 	/*
3331 	 * Lock source and target directory entries.  To prevent deadlock,
3332 	 * a lock ordering must be defined.  We lock the directory with
3333 	 * the smallest object id first, or if it's a tie, the one with
3334 	 * the lexically first name.
3335 	 */
3336 	if (sdzp->z_id < tdzp->z_id) {
3337 		cmp = -1;
3338 	} else if (sdzp->z_id > tdzp->z_id) {
3339 		cmp = 1;
3340 	} else {
3341 		/*
3342 		 * First compare the two name arguments without
3343 		 * considering any case folding.
3344 		 */
3345 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3346 
3347 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3348 		ASSERT(error == 0 || !zfsvfs->z_utf8);
3349 		if (cmp == 0) {
3350 			/*
3351 			 * POSIX: "If the old argument and the new argument
3352 			 * both refer to links to the same existing file,
3353 			 * the rename() function shall return successfully
3354 			 * and perform no other action."
3355 			 */
3356 			ZFS_EXIT(zfsvfs);
3357 			return (0);
3358 		}
3359 		/*
3360 		 * If the file system is case-folding, then we may
3361 		 * have some more checking to do.  A case-folding file
3362 		 * system is either supporting mixed case sensitivity
3363 		 * access or is completely case-insensitive.  Note
3364 		 * that the file system is always case preserving.
3365 		 *
3366 		 * In mixed sensitivity mode case sensitive behavior
3367 		 * is the default.  FIGNORECASE must be used to
3368 		 * explicitly request case insensitive behavior.
3369 		 *
3370 		 * If the source and target names provided differ only
3371 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
3372 		 * we will treat this as a special case in the
3373 		 * case-insensitive mode: as long as the source name
3374 		 * is an exact match, we will allow this to proceed as
3375 		 * a name-change request.
3376 		 */
3377 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3378 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
3379 		    flags & FIGNORECASE)) &&
3380 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3381 		    &error) == 0) {
3382 			/*
3383 			 * case preserving rename request, require exact
3384 			 * name matches
3385 			 */
3386 			zflg |= ZCIEXACT;
3387 			zflg &= ~ZCILOOK;
3388 		}
3389 	}
3390 
3391 	/*
3392 	 * If the source and destination directories are the same, we should
3393 	 * grab the z_name_lock of that directory only once.
3394 	 */
3395 	if (sdzp == tdzp) {
3396 		zflg |= ZHAVELOCK;
3397 		rw_enter(&sdzp->z_name_lock, RW_READER);
3398 	}
3399 
3400 	if (cmp < 0) {
3401 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3402 		    ZEXISTS | zflg, NULL, NULL);
3403 		terr = zfs_dirent_lock(&tdl,
3404 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3405 	} else {
3406 		terr = zfs_dirent_lock(&tdl,
3407 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3408 		serr = zfs_dirent_lock(&sdl,
3409 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3410 		    NULL, NULL);
3411 	}
3412 
3413 	if (serr) {
3414 		/*
3415 		 * Source entry invalid or not there.
3416 		 */
3417 		if (!terr) {
3418 			zfs_dirent_unlock(tdl);
3419 			if (tzp)
3420 				VN_RELE(ZTOV(tzp));
3421 		}
3422 
3423 		if (sdzp == tdzp)
3424 			rw_exit(&sdzp->z_name_lock);
3425 
3426 		if (strcmp(snm, "..") == 0)
3427 			serr = EINVAL;
3428 		ZFS_EXIT(zfsvfs);
3429 		return (serr);
3430 	}
3431 	if (terr) {
3432 		zfs_dirent_unlock(sdl);
3433 		VN_RELE(ZTOV(szp));
3434 
3435 		if (sdzp == tdzp)
3436 			rw_exit(&sdzp->z_name_lock);
3437 
3438 		if (strcmp(tnm, "..") == 0)
3439 			terr = EINVAL;
3440 		ZFS_EXIT(zfsvfs);
3441 		return (terr);
3442 	}
3443 
3444 	/*
3445 	 * Must have write access at the source to remove the old entry
3446 	 * and write access at the target to create the new entry.
3447 	 * Note that if target and source are the same, this can be
3448 	 * done in a single check.
3449 	 */
3450 
3451 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3452 		goto out;
3453 
3454 	if (ZTOV(szp)->v_type == VDIR) {
3455 		/*
3456 		 * Check to make sure rename is valid.
3457 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3458 		 */
3459 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3460 			goto out;
3461 	}
3462 
3463 	/*
3464 	 * Does target exist?
3465 	 */
3466 	if (tzp) {
3467 		/*
3468 		 * Source and target must be the same type.
3469 		 */
3470 		if (ZTOV(szp)->v_type == VDIR) {
3471 			if (ZTOV(tzp)->v_type != VDIR) {
3472 				error = ENOTDIR;
3473 				goto out;
3474 			}
3475 		} else {
3476 			if (ZTOV(tzp)->v_type == VDIR) {
3477 				error = EISDIR;
3478 				goto out;
3479 			}
3480 		}
3481 		/*
3482 		 * POSIX dictates that when the source and target
3483 		 * entries refer to the same file object, rename
3484 		 * must do nothing and exit without error.
3485 		 */
3486 		if (szp->z_id == tzp->z_id) {
3487 			error = 0;
3488 			goto out;
3489 		}
3490 	}
3491 
3492 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3493 	if (tzp)
3494 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3495 
3496 	/*
3497 	 * notify the target directory if it is not the same
3498 	 * as source directory.
3499 	 */
3500 	if (tdvp != sdvp) {
3501 		vnevent_rename_dest_dir(tdvp, ct);
3502 	}
3503 
3504 	tx = dmu_tx_create(zfsvfs->z_os);
3505 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3506 	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3507 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3508 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3509 	if (sdzp != tdzp) {
3510 		dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3511 		zfs_sa_upgrade_txholds(tx, tdzp);
3512 	}
3513 	if (tzp) {
3514 		dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3515 		zfs_sa_upgrade_txholds(tx, tzp);
3516 	}
3517 
3518 	zfs_sa_upgrade_txholds(tx, szp);
3519 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3520 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3521 	if (error) {
3522 		if (zl != NULL)
3523 			zfs_rename_unlock(&zl);
3524 		zfs_dirent_unlock(sdl);
3525 		zfs_dirent_unlock(tdl);
3526 
3527 		if (sdzp == tdzp)
3528 			rw_exit(&sdzp->z_name_lock);
3529 
3530 		VN_RELE(ZTOV(szp));
3531 		if (tzp)
3532 			VN_RELE(ZTOV(tzp));
3533 		if (error == ERESTART) {
3534 			dmu_tx_wait(tx);
3535 			dmu_tx_abort(tx);
3536 			goto top;
3537 		}
3538 		dmu_tx_abort(tx);
3539 		ZFS_EXIT(zfsvfs);
3540 		return (error);
3541 	}
3542 
3543 	if (tzp)	/* Attempt to remove the existing target */
3544 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3545 
3546 	if (error == 0) {
3547 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3548 		if (error == 0) {
3549 			szp->z_pflags |= ZFS_AV_MODIFIED;
3550 
3551 			error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3552 			    (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3553 			ASSERT3U(error, ==, 0);
3554 
3555 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3556 			ASSERT3U(error, ==, 0);
3557 
3558 			zfs_log_rename(zilog, tx,
3559 			    TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0),
3560 			    sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp);
3561 
3562 			/* Update path information for the target vnode */
3563 			vn_renamepath(tdvp, ZTOV(szp), tnm, strlen(tnm));
3564 		}
3565 	}
3566 
3567 	dmu_tx_commit(tx);
3568 out:
3569 	if (zl != NULL)
3570 		zfs_rename_unlock(&zl);
3571 
3572 	zfs_dirent_unlock(sdl);
3573 	zfs_dirent_unlock(tdl);
3574 
3575 	if (sdzp == tdzp)
3576 		rw_exit(&sdzp->z_name_lock);
3577 
3578 
3579 	VN_RELE(ZTOV(szp));
3580 	if (tzp)
3581 		VN_RELE(ZTOV(tzp));
3582 
3583 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3584 		zil_commit(zilog, UINT64_MAX, 0);
3585 
3586 	ZFS_EXIT(zfsvfs);
3587 	return (error);
3588 }
3589 
3590 /*
3591  * Insert the indicated symbolic reference entry into the directory.
3592  *
3593  *	IN:	dvp	- Directory to contain new symbolic link.
3594  *		link	- Name for new symlink entry.
3595  *		vap	- Attributes of new entry.
3596  *		target	- Target path of new symlink.
3597  *		cr	- credentials of caller.
3598  *		ct	- caller context
3599  *		flags	- case flags
3600  *
3601  *	RETURN:	0 if success
3602  *		error code if failure
3603  *
3604  * Timestamps:
3605  *	dvp - ctime|mtime updated
3606  */
3607 /*ARGSUSED*/
3608 static int
3609 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3610     caller_context_t *ct, int flags)
3611 {
3612 	znode_t		*zp, *dzp = VTOZ(dvp);
3613 	zfs_dirlock_t	*dl;
3614 	dmu_tx_t	*tx;
3615 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3616 	zilog_t		*zilog;
3617 	uint64_t	len = strlen(link);
3618 	int		error;
3619 	int		zflg = ZNEW;
3620 	zfs_acl_ids_t	acl_ids;
3621 	boolean_t	fuid_dirtied;
3622 	uint64_t	txtype = TX_SYMLINK;
3623 
3624 	ASSERT(vap->va_type == VLNK);
3625 
3626 	ZFS_ENTER(zfsvfs);
3627 	ZFS_VERIFY_ZP(dzp);
3628 	zilog = zfsvfs->z_log;
3629 
3630 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3631 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3632 		ZFS_EXIT(zfsvfs);
3633 		return (EILSEQ);
3634 	}
3635 	if (flags & FIGNORECASE)
3636 		zflg |= ZCILOOK;
3637 
3638 	if (len > MAXPATHLEN) {
3639 		ZFS_EXIT(zfsvfs);
3640 		return (ENAMETOOLONG);
3641 	}
3642 
3643 	if ((error = zfs_acl_ids_create(dzp, 0,
3644 	    vap, cr, NULL, &acl_ids)) != 0) {
3645 		ZFS_EXIT(zfsvfs);
3646 		return (error);
3647 	}
3648 top:
3649 	/*
3650 	 * Attempt to lock directory; fail if entry already exists.
3651 	 */
3652 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3653 	if (error) {
3654 		zfs_acl_ids_free(&acl_ids);
3655 		ZFS_EXIT(zfsvfs);
3656 		return (error);
3657 	}
3658 
3659 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3660 		zfs_acl_ids_free(&acl_ids);
3661 		ZFS_EXIT(zfsvfs);
3662 		return (error);
3663 	}
3664 
3665 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3666 		zfs_acl_ids_free(&acl_ids);
3667 		zfs_dirent_unlock(dl);
3668 		ZFS_EXIT(zfsvfs);
3669 		return (EDQUOT);
3670 	}
3671 	tx = dmu_tx_create(zfsvfs->z_os);
3672 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3673 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3674 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3675 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3676 	    ZFS_SA_BASE_ATTR_SIZE + len);
3677 	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3678 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3679 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3680 		    acl_ids.z_aclp->z_acl_bytes);
3681 	}
3682 	if (fuid_dirtied)
3683 		zfs_fuid_txhold(zfsvfs, tx);
3684 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3685 	if (error) {
3686 		zfs_dirent_unlock(dl);
3687 		if (error == ERESTART) {
3688 			dmu_tx_wait(tx);
3689 			dmu_tx_abort(tx);
3690 			goto top;
3691 		}
3692 		zfs_acl_ids_free(&acl_ids);
3693 		dmu_tx_abort(tx);
3694 		ZFS_EXIT(zfsvfs);
3695 		return (error);
3696 	}
3697 
3698 	/*
3699 	 * Create a new object for the symlink.
3700 	 * for version 4 ZPL datsets the symlink will be an SA attribute
3701 	 */
3702 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3703 
3704 	if (fuid_dirtied)
3705 		zfs_fuid_sync(zfsvfs, tx);
3706 
3707 	if (zp->z_is_sa)
3708 		error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3709 		    link, len, tx);
3710 	else
3711 		zfs_sa_symlink(zp, link, len, tx);
3712 
3713 	zp->z_size = len;
3714 	(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3715 	    &zp->z_size, sizeof (zp->z_size), tx);
3716 	/*
3717 	 * Insert the new object into the directory.
3718 	 */
3719 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3720 
3721 	if (flags & FIGNORECASE)
3722 		txtype |= TX_CI;
3723 	zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3724 
3725 	zfs_acl_ids_free(&acl_ids);
3726 
3727 	dmu_tx_commit(tx);
3728 
3729 	zfs_dirent_unlock(dl);
3730 
3731 	VN_RELE(ZTOV(zp));
3732 
3733 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3734 		zil_commit(zilog, UINT64_MAX, 0);
3735 
3736 	ZFS_EXIT(zfsvfs);
3737 	return (error);
3738 }
3739 
3740 /*
3741  * Return, in the buffer contained in the provided uio structure,
3742  * the symbolic path referred to by vp.
3743  *
3744  *	IN:	vp	- vnode of symbolic link.
3745  *		uoip	- structure to contain the link path.
3746  *		cr	- credentials of caller.
3747  *		ct	- caller context
3748  *
3749  *	OUT:	uio	- structure to contain the link path.
3750  *
3751  *	RETURN:	0 if success
3752  *		error code if failure
3753  *
3754  * Timestamps:
3755  *	vp - atime updated
3756  */
3757 /* ARGSUSED */
3758 static int
3759 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3760 {
3761 	znode_t		*zp = VTOZ(vp);
3762 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3763 	int		error;
3764 
3765 	ZFS_ENTER(zfsvfs);
3766 	ZFS_VERIFY_ZP(zp);
3767 
3768 	if (zp->z_is_sa)
3769 		error = sa_lookup_uio(zp->z_sa_hdl,
3770 		    SA_ZPL_SYMLINK(zfsvfs), uio);
3771 	else
3772 		error = zfs_sa_readlink(zp, uio);
3773 
3774 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3775 
3776 	ZFS_EXIT(zfsvfs);
3777 	return (error);
3778 }
3779 
3780 /*
3781  * Insert a new entry into directory tdvp referencing svp.
3782  *
3783  *	IN:	tdvp	- Directory to contain new entry.
3784  *		svp	- vnode of new entry.
3785  *		name	- name of new entry.
3786  *		cr	- credentials of caller.
3787  *		ct	- caller context
3788  *
3789  *	RETURN:	0 if success
3790  *		error code if failure
3791  *
3792  * Timestamps:
3793  *	tdvp - ctime|mtime updated
3794  *	 svp - ctime updated
3795  */
3796 /* ARGSUSED */
3797 static int
3798 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3799     caller_context_t *ct, int flags)
3800 {
3801 	znode_t		*dzp = VTOZ(tdvp);
3802 	znode_t		*tzp, *szp;
3803 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3804 	zilog_t		*zilog;
3805 	zfs_dirlock_t	*dl;
3806 	dmu_tx_t	*tx;
3807 	vnode_t		*realvp;
3808 	int		error;
3809 	int		zf = ZNEW;
3810 	uint64_t	parent;
3811 
3812 	ASSERT(tdvp->v_type == VDIR);
3813 
3814 	ZFS_ENTER(zfsvfs);
3815 	ZFS_VERIFY_ZP(dzp);
3816 	zilog = zfsvfs->z_log;
3817 
3818 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3819 		svp = realvp;
3820 
3821 	/*
3822 	 * POSIX dictates that we return EPERM here.
3823 	 * Better choices include ENOTSUP or EISDIR.
3824 	 */
3825 	if (svp->v_type == VDIR) {
3826 		ZFS_EXIT(zfsvfs);
3827 		return (EPERM);
3828 	}
3829 
3830 	if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
3831 		ZFS_EXIT(zfsvfs);
3832 		return (EXDEV);
3833 	}
3834 
3835 	szp = VTOZ(svp);
3836 	ZFS_VERIFY_ZP(szp);
3837 
3838 	/* Prevent links to .zfs/shares files */
3839 
3840 	if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3841 	    &parent, sizeof (uint64_t))) != 0) {
3842 		ZFS_EXIT(zfsvfs);
3843 		return (error);
3844 	}
3845 	if (parent == zfsvfs->z_shares_dir) {
3846 		ZFS_EXIT(zfsvfs);
3847 		return (EPERM);
3848 	}
3849 
3850 	if (zfsvfs->z_utf8 && u8_validate(name,
3851 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3852 		ZFS_EXIT(zfsvfs);
3853 		return (EILSEQ);
3854 	}
3855 	if (flags & FIGNORECASE)
3856 		zf |= ZCILOOK;
3857 
3858 	/*
3859 	 * We do not support links between attributes and non-attributes
3860 	 * because of the potential security risk of creating links
3861 	 * into "normal" file space in order to circumvent restrictions
3862 	 * imposed in attribute space.
3863 	 */
3864 	if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3865 		ZFS_EXIT(zfsvfs);
3866 		return (EINVAL);
3867 	}
3868 
3869 
3870 	if (szp->z_uid != crgetuid(cr) &&
3871 	    secpolicy_basic_link(cr) != 0) {
3872 		ZFS_EXIT(zfsvfs);
3873 		return (EPERM);
3874 	}
3875 
3876 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3877 		ZFS_EXIT(zfsvfs);
3878 		return (error);
3879 	}
3880 
3881 top:
3882 	/*
3883 	 * Attempt to lock directory; fail if entry already exists.
3884 	 */
3885 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3886 	if (error) {
3887 		ZFS_EXIT(zfsvfs);
3888 		return (error);
3889 	}
3890 
3891 	tx = dmu_tx_create(zfsvfs->z_os);
3892 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3893 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3894 	zfs_sa_upgrade_txholds(tx, szp);
3895 	zfs_sa_upgrade_txholds(tx, dzp);
3896 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3897 	if (error) {
3898 		zfs_dirent_unlock(dl);
3899 		if (error == ERESTART) {
3900 			dmu_tx_wait(tx);
3901 			dmu_tx_abort(tx);
3902 			goto top;
3903 		}
3904 		dmu_tx_abort(tx);
3905 		ZFS_EXIT(zfsvfs);
3906 		return (error);
3907 	}
3908 
3909 	error = zfs_link_create(dl, szp, tx, 0);
3910 
3911 	if (error == 0) {
3912 		uint64_t txtype = TX_LINK;
3913 		if (flags & FIGNORECASE)
3914 			txtype |= TX_CI;
3915 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3916 	}
3917 
3918 	dmu_tx_commit(tx);
3919 
3920 	zfs_dirent_unlock(dl);
3921 
3922 	if (error == 0) {
3923 		vnevent_link(svp, ct);
3924 	}
3925 
3926 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3927 		zil_commit(zilog, UINT64_MAX, 0);
3928 
3929 	ZFS_EXIT(zfsvfs);
3930 	return (error);
3931 }
3932 
3933 /*
3934  * zfs_null_putapage() is used when the file system has been force
3935  * unmounted. It just drops the pages.
3936  */
3937 /* ARGSUSED */
3938 static int
3939 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3940 		size_t *lenp, int flags, cred_t *cr)
3941 {
3942 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
3943 	return (0);
3944 }
3945 
3946 /*
3947  * Push a page out to disk, klustering if possible.
3948  *
3949  *	IN:	vp	- file to push page to.
3950  *		pp	- page to push.
3951  *		flags	- additional flags.
3952  *		cr	- credentials of caller.
3953  *
3954  *	OUT:	offp	- start of range pushed.
3955  *		lenp	- len of range pushed.
3956  *
3957  *	RETURN:	0 if success
3958  *		error code if failure
3959  *
3960  * NOTE: callers must have locked the page to be pushed.  On
3961  * exit, the page (and all other pages in the kluster) must be
3962  * unlocked.
3963  */
3964 /* ARGSUSED */
3965 static int
3966 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3967 		size_t *lenp, int flags, cred_t *cr)
3968 {
3969 	znode_t		*zp = VTOZ(vp);
3970 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3971 	dmu_tx_t	*tx;
3972 	u_offset_t	off, koff;
3973 	size_t		len, klen;
3974 	int		err;
3975 
3976 	off = pp->p_offset;
3977 	len = PAGESIZE;
3978 	/*
3979 	 * If our blocksize is bigger than the page size, try to kluster
3980 	 * multiple pages so that we write a full block (thus avoiding
3981 	 * a read-modify-write).
3982 	 */
3983 	if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
3984 		klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3985 		koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
3986 		ASSERT(koff <= zp->z_size);
3987 		if (koff + klen > zp->z_size)
3988 			klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
3989 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
3990 	}
3991 	ASSERT3U(btop(len), ==, btopr(len));
3992 
3993 	/*
3994 	 * Can't push pages past end-of-file.
3995 	 */
3996 	if (off >= zp->z_size) {
3997 		/* ignore all pages */
3998 		err = 0;
3999 		goto out;
4000 	} else if (off + len > zp->z_size) {
4001 		int npages = btopr(zp->z_size - off);
4002 		page_t *trunc;
4003 
4004 		page_list_break(&pp, &trunc, npages);
4005 		/* ignore pages past end of file */
4006 		if (trunc)
4007 			pvn_write_done(trunc, flags);
4008 		len = zp->z_size - off;
4009 	}
4010 
4011 	if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4012 	    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4013 		err = EDQUOT;
4014 		goto out;
4015 	}
4016 top:
4017 	tx = dmu_tx_create(zfsvfs->z_os);
4018 	dmu_tx_hold_write(tx, zp->z_id, off, len);
4019 
4020 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4021 	zfs_sa_upgrade_txholds(tx, zp);
4022 	err = dmu_tx_assign(tx, TXG_NOWAIT);
4023 	if (err != 0) {
4024 		if (err == ERESTART) {
4025 			dmu_tx_wait(tx);
4026 			dmu_tx_abort(tx);
4027 			goto top;
4028 		}
4029 		dmu_tx_abort(tx);
4030 		goto out;
4031 	}
4032 
4033 	if (zp->z_blksz <= PAGESIZE) {
4034 		caddr_t va = zfs_map_page(pp, S_READ);
4035 		ASSERT3U(len, <=, PAGESIZE);
4036 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4037 		zfs_unmap_page(pp, va);
4038 	} else {
4039 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4040 	}
4041 
4042 	if (err == 0) {
4043 		uint64_t mtime[2], ctime[2];
4044 		sa_bulk_attr_t bulk[3];
4045 		int count = 0;
4046 
4047 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4048 		    &mtime, 16);
4049 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4050 		    &ctime, 16);
4051 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4052 		    &zp->z_pflags, 8);
4053 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4054 		    B_TRUE);
4055 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4056 	}
4057 	dmu_tx_commit(tx);
4058 
4059 out:
4060 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4061 	if (offp)
4062 		*offp = off;
4063 	if (lenp)
4064 		*lenp = len;
4065 
4066 	return (err);
4067 }
4068 
4069 /*
4070  * Copy the portion of the file indicated from pages into the file.
4071  * The pages are stored in a page list attached to the files vnode.
4072  *
4073  *	IN:	vp	- vnode of file to push page data to.
4074  *		off	- position in file to put data.
4075  *		len	- amount of data to write.
4076  *		flags	- flags to control the operation.
4077  *		cr	- credentials of caller.
4078  *		ct	- caller context.
4079  *
4080  *	RETURN:	0 if success
4081  *		error code if failure
4082  *
4083  * Timestamps:
4084  *	vp - ctime|mtime updated
4085  */
4086 /*ARGSUSED*/
4087 static int
4088 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4089     caller_context_t *ct)
4090 {
4091 	znode_t		*zp = VTOZ(vp);
4092 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4093 	page_t		*pp;
4094 	size_t		io_len;
4095 	u_offset_t	io_off;
4096 	uint_t		blksz;
4097 	rl_t		*rl;
4098 	int		error = 0;
4099 
4100 	ZFS_ENTER(zfsvfs);
4101 	ZFS_VERIFY_ZP(zp);
4102 
4103 	/*
4104 	 * Align this request to the file block size in case we kluster.
4105 	 * XXX - this can result in pretty aggresive locking, which can
4106 	 * impact simultanious read/write access.  One option might be
4107 	 * to break up long requests (len == 0) into block-by-block
4108 	 * operations to get narrower locking.
4109 	 */
4110 	blksz = zp->z_blksz;
4111 	if (ISP2(blksz))
4112 		io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4113 	else
4114 		io_off = 0;
4115 	if (len > 0 && ISP2(blksz))
4116 		io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4117 	else
4118 		io_len = 0;
4119 
4120 	if (io_len == 0) {
4121 		/*
4122 		 * Search the entire vp list for pages >= io_off.
4123 		 */
4124 		rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4125 		error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4126 		goto out;
4127 	}
4128 	rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4129 
4130 	if (off > zp->z_size) {
4131 		/* past end of file */
4132 		zfs_range_unlock(rl);
4133 		ZFS_EXIT(zfsvfs);
4134 		return (0);
4135 	}
4136 
4137 	len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4138 
4139 	for (off = io_off; io_off < off + len; io_off += io_len) {
4140 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4141 			pp = page_lookup(vp, io_off,
4142 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4143 		} else {
4144 			pp = page_lookup_nowait(vp, io_off,
4145 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4146 		}
4147 
4148 		if (pp != NULL && pvn_getdirty(pp, flags)) {
4149 			int err;
4150 
4151 			/*
4152 			 * Found a dirty page to push
4153 			 */
4154 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4155 			if (err)
4156 				error = err;
4157 		} else {
4158 			io_len = PAGESIZE;
4159 		}
4160 	}
4161 out:
4162 	zfs_range_unlock(rl);
4163 	if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4164 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
4165 	ZFS_EXIT(zfsvfs);
4166 	return (error);
4167 }
4168 
4169 /*ARGSUSED*/
4170 void
4171 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4172 {
4173 	znode_t	*zp = VTOZ(vp);
4174 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4175 	int error;
4176 
4177 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4178 	if (zp->z_sa_hdl == NULL) {
4179 		/*
4180 		 * The fs has been unmounted, or we did a
4181 		 * suspend/resume and this file no longer exists.
4182 		 */
4183 		if (vn_has_cached_data(vp)) {
4184 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4185 			    B_INVAL, cr);
4186 		}
4187 
4188 		mutex_enter(&zp->z_lock);
4189 		mutex_enter(&vp->v_lock);
4190 		ASSERT(vp->v_count == 1);
4191 		vp->v_count = 0;
4192 		mutex_exit(&vp->v_lock);
4193 		mutex_exit(&zp->z_lock);
4194 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
4195 		zfs_znode_free(zp);
4196 		return;
4197 	}
4198 
4199 	/*
4200 	 * Attempt to push any data in the page cache.  If this fails
4201 	 * we will get kicked out later in zfs_zinactive().
4202 	 */
4203 	if (vn_has_cached_data(vp)) {
4204 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4205 		    cr);
4206 	}
4207 
4208 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4209 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4210 
4211 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4212 		zfs_sa_upgrade_txholds(tx, zp);
4213 		error = dmu_tx_assign(tx, TXG_WAIT);
4214 		if (error) {
4215 			dmu_tx_abort(tx);
4216 		} else {
4217 			mutex_enter(&zp->z_lock);
4218 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4219 			    (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4220 			zp->z_atime_dirty = 0;
4221 			mutex_exit(&zp->z_lock);
4222 			dmu_tx_commit(tx);
4223 		}
4224 	}
4225 
4226 	zfs_zinactive(zp);
4227 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
4228 }
4229 
4230 /*
4231  * Bounds-check the seek operation.
4232  *
4233  *	IN:	vp	- vnode seeking within
4234  *		ooff	- old file offset
4235  *		noffp	- pointer to new file offset
4236  *		ct	- caller context
4237  *
4238  *	RETURN:	0 if success
4239  *		EINVAL if new offset invalid
4240  */
4241 /* ARGSUSED */
4242 static int
4243 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4244     caller_context_t *ct)
4245 {
4246 	if (vp->v_type == VDIR)
4247 		return (0);
4248 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4249 }
4250 
4251 /*
4252  * Pre-filter the generic locking function to trap attempts to place
4253  * a mandatory lock on a memory mapped file.
4254  */
4255 static int
4256 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4257     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4258 {
4259 	znode_t *zp = VTOZ(vp);
4260 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4261 
4262 	ZFS_ENTER(zfsvfs);
4263 	ZFS_VERIFY_ZP(zp);
4264 
4265 	/*
4266 	 * We are following the UFS semantics with respect to mapcnt
4267 	 * here: If we see that the file is mapped already, then we will
4268 	 * return an error, but we don't worry about races between this
4269 	 * function and zfs_map().
4270 	 */
4271 	if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4272 		ZFS_EXIT(zfsvfs);
4273 		return (EAGAIN);
4274 	}
4275 	ZFS_EXIT(zfsvfs);
4276 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4277 }
4278 
4279 /*
4280  * If we can't find a page in the cache, we will create a new page
4281  * and fill it with file data.  For efficiency, we may try to fill
4282  * multiple pages at once (klustering) to fill up the supplied page
4283  * list.  Note that the pages to be filled are held with an exclusive
4284  * lock to prevent access by other threads while they are being filled.
4285  */
4286 static int
4287 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4288     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4289 {
4290 	znode_t *zp = VTOZ(vp);
4291 	page_t *pp, *cur_pp;
4292 	objset_t *os = zp->z_zfsvfs->z_os;
4293 	u_offset_t io_off, total;
4294 	size_t io_len;
4295 	int err;
4296 
4297 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4298 		/*
4299 		 * We only have a single page, don't bother klustering
4300 		 */
4301 		io_off = off;
4302 		io_len = PAGESIZE;
4303 		pp = page_create_va(vp, io_off, io_len,
4304 		    PG_EXCL | PG_WAIT, seg, addr);
4305 	} else {
4306 		/*
4307 		 * Try to find enough pages to fill the page list
4308 		 */
4309 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4310 		    &io_len, off, plsz, 0);
4311 	}
4312 	if (pp == NULL) {
4313 		/*
4314 		 * The page already exists, nothing to do here.
4315 		 */
4316 		*pl = NULL;
4317 		return (0);
4318 	}
4319 
4320 	/*
4321 	 * Fill the pages in the kluster.
4322 	 */
4323 	cur_pp = pp;
4324 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4325 		caddr_t va;
4326 
4327 		ASSERT3U(io_off, ==, cur_pp->p_offset);
4328 		va = zfs_map_page(cur_pp, S_WRITE);
4329 		err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4330 		    DMU_READ_PREFETCH);
4331 		zfs_unmap_page(cur_pp, va);
4332 		if (err) {
4333 			/* On error, toss the entire kluster */
4334 			pvn_read_done(pp, B_ERROR);
4335 			/* convert checksum errors into IO errors */
4336 			if (err == ECKSUM)
4337 				err = EIO;
4338 			return (err);
4339 		}
4340 		cur_pp = cur_pp->p_next;
4341 	}
4342 
4343 	/*
4344 	 * Fill in the page list array from the kluster starting
4345 	 * from the desired offset `off'.
4346 	 * NOTE: the page list will always be null terminated.
4347 	 */
4348 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4349 	ASSERT(pl == NULL || (*pl)->p_offset == off);
4350 
4351 	return (0);
4352 }
4353 
4354 /*
4355  * Return pointers to the pages for the file region [off, off + len]
4356  * in the pl array.  If plsz is greater than len, this function may
4357  * also return page pointers from after the specified region
4358  * (i.e. the region [off, off + plsz]).  These additional pages are
4359  * only returned if they are already in the cache, or were created as
4360  * part of a klustered read.
4361  *
4362  *	IN:	vp	- vnode of file to get data from.
4363  *		off	- position in file to get data from.
4364  *		len	- amount of data to retrieve.
4365  *		plsz	- length of provided page list.
4366  *		seg	- segment to obtain pages for.
4367  *		addr	- virtual address of fault.
4368  *		rw	- mode of created pages.
4369  *		cr	- credentials of caller.
4370  *		ct	- caller context.
4371  *
4372  *	OUT:	protp	- protection mode of created pages.
4373  *		pl	- list of pages created.
4374  *
4375  *	RETURN:	0 if success
4376  *		error code if failure
4377  *
4378  * Timestamps:
4379  *	vp - atime updated
4380  */
4381 /* ARGSUSED */
4382 static int
4383 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4384 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4385 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4386 {
4387 	znode_t		*zp = VTOZ(vp);
4388 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4389 	page_t		**pl0 = pl;
4390 	int		err = 0;
4391 
4392 	/* we do our own caching, faultahead is unnecessary */
4393 	if (pl == NULL)
4394 		return (0);
4395 	else if (len > plsz)
4396 		len = plsz;
4397 	else
4398 		len = P2ROUNDUP(len, PAGESIZE);
4399 	ASSERT(plsz >= len);
4400 
4401 	ZFS_ENTER(zfsvfs);
4402 	ZFS_VERIFY_ZP(zp);
4403 
4404 	if (protp)
4405 		*protp = PROT_ALL;
4406 
4407 	/*
4408 	 * Loop through the requested range [off, off + len) looking
4409 	 * for pages.  If we don't find a page, we will need to create
4410 	 * a new page and fill it with data from the file.
4411 	 */
4412 	while (len > 0) {
4413 		if (*pl = page_lookup(vp, off, SE_SHARED))
4414 			*(pl+1) = NULL;
4415 		else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4416 			goto out;
4417 		while (*pl) {
4418 			ASSERT3U((*pl)->p_offset, ==, off);
4419 			off += PAGESIZE;
4420 			addr += PAGESIZE;
4421 			if (len > 0) {
4422 				ASSERT3U(len, >=, PAGESIZE);
4423 				len -= PAGESIZE;
4424 			}
4425 			ASSERT3U(plsz, >=, PAGESIZE);
4426 			plsz -= PAGESIZE;
4427 			pl++;
4428 		}
4429 	}
4430 
4431 	/*
4432 	 * Fill out the page array with any pages already in the cache.
4433 	 */
4434 	while (plsz > 0 &&
4435 	    (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4436 			off += PAGESIZE;
4437 			plsz -= PAGESIZE;
4438 	}
4439 out:
4440 	if (err) {
4441 		/*
4442 		 * Release any pages we have previously locked.
4443 		 */
4444 		while (pl > pl0)
4445 			page_unlock(*--pl);
4446 	} else {
4447 		ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4448 	}
4449 
4450 	*pl = NULL;
4451 
4452 	ZFS_EXIT(zfsvfs);
4453 	return (err);
4454 }
4455 
4456 /*
4457  * Request a memory map for a section of a file.  This code interacts
4458  * with common code and the VM system as follows:
4459  *
4460  *	common code calls mmap(), which ends up in smmap_common()
4461  *
4462  *	this calls VOP_MAP(), which takes you into (say) zfs
4463  *
4464  *	zfs_map() calls as_map(), passing segvn_create() as the callback
4465  *
4466  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
4467  *
4468  *	zfs_addmap() updates z_mapcnt
4469  */
4470 /*ARGSUSED*/
4471 static int
4472 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4473     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4474     caller_context_t *ct)
4475 {
4476 	znode_t *zp = VTOZ(vp);
4477 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4478 	segvn_crargs_t	vn_a;
4479 	int		error;
4480 
4481 	ZFS_ENTER(zfsvfs);
4482 	ZFS_VERIFY_ZP(zp);
4483 
4484 	if ((prot & PROT_WRITE) && (zp->z_pflags &
4485 	    (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4486 		ZFS_EXIT(zfsvfs);
4487 		return (EPERM);
4488 	}
4489 
4490 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4491 	    (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4492 		ZFS_EXIT(zfsvfs);
4493 		return (EACCES);
4494 	}
4495 
4496 	if (vp->v_flag & VNOMAP) {
4497 		ZFS_EXIT(zfsvfs);
4498 		return (ENOSYS);
4499 	}
4500 
4501 	if (off < 0 || len > MAXOFFSET_T - off) {
4502 		ZFS_EXIT(zfsvfs);
4503 		return (ENXIO);
4504 	}
4505 
4506 	if (vp->v_type != VREG) {
4507 		ZFS_EXIT(zfsvfs);
4508 		return (ENODEV);
4509 	}
4510 
4511 	/*
4512 	 * If file is locked, disallow mapping.
4513 	 */
4514 	if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4515 		ZFS_EXIT(zfsvfs);
4516 		return (EAGAIN);
4517 	}
4518 
4519 	as_rangelock(as);
4520 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4521 	if (error != 0) {
4522 		as_rangeunlock(as);
4523 		ZFS_EXIT(zfsvfs);
4524 		return (error);
4525 	}
4526 
4527 	vn_a.vp = vp;
4528 	vn_a.offset = (u_offset_t)off;
4529 	vn_a.type = flags & MAP_TYPE;
4530 	vn_a.prot = prot;
4531 	vn_a.maxprot = maxprot;
4532 	vn_a.cred = cr;
4533 	vn_a.amp = NULL;
4534 	vn_a.flags = flags & ~MAP_TYPE;
4535 	vn_a.szc = 0;
4536 	vn_a.lgrp_mem_policy_flags = 0;
4537 
4538 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4539 
4540 	as_rangeunlock(as);
4541 	ZFS_EXIT(zfsvfs);
4542 	return (error);
4543 }
4544 
4545 /* ARGSUSED */
4546 static int
4547 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4548     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4549     caller_context_t *ct)
4550 {
4551 	uint64_t pages = btopr(len);
4552 
4553 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4554 	return (0);
4555 }
4556 
4557 /*
4558  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4559  * more accurate mtime for the associated file.  Since we don't have a way of
4560  * detecting when the data was actually modified, we have to resort to
4561  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4562  * last page is pushed.  The problem occurs when the msync() call is omitted,
4563  * which by far the most common case:
4564  *
4565  * 	open()
4566  * 	mmap()
4567  * 	<modify memory>
4568  * 	munmap()
4569  * 	close()
4570  * 	<time lapse>
4571  * 	putpage() via fsflush
4572  *
4573  * If we wait until fsflush to come along, we can have a modification time that
4574  * is some arbitrary point in the future.  In order to prevent this in the
4575  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4576  * torn down.
4577  */
4578 /* ARGSUSED */
4579 static int
4580 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4581     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4582     caller_context_t *ct)
4583 {
4584 	uint64_t pages = btopr(len);
4585 
4586 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4587 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4588 
4589 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4590 	    vn_has_cached_data(vp))
4591 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4592 
4593 	return (0);
4594 }
4595 
4596 /*
4597  * Free or allocate space in a file.  Currently, this function only
4598  * supports the `F_FREESP' command.  However, this command is somewhat
4599  * misnamed, as its functionality includes the ability to allocate as
4600  * well as free space.
4601  *
4602  *	IN:	vp	- vnode of file to free data in.
4603  *		cmd	- action to take (only F_FREESP supported).
4604  *		bfp	- section of file to free/alloc.
4605  *		flag	- current file open mode flags.
4606  *		offset	- current file offset.
4607  *		cr	- credentials of caller [UNUSED].
4608  *		ct	- caller context.
4609  *
4610  *	RETURN:	0 if success
4611  *		error code if failure
4612  *
4613  * Timestamps:
4614  *	vp - ctime|mtime updated
4615  */
4616 /* ARGSUSED */
4617 static int
4618 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4619     offset_t offset, cred_t *cr, caller_context_t *ct)
4620 {
4621 	znode_t		*zp = VTOZ(vp);
4622 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4623 	uint64_t	off, len;
4624 	int		error;
4625 
4626 	ZFS_ENTER(zfsvfs);
4627 	ZFS_VERIFY_ZP(zp);
4628 
4629 	if (cmd != F_FREESP) {
4630 		ZFS_EXIT(zfsvfs);
4631 		return (EINVAL);
4632 	}
4633 
4634 	if (error = convoff(vp, bfp, 0, offset)) {
4635 		ZFS_EXIT(zfsvfs);
4636 		return (error);
4637 	}
4638 
4639 	if (bfp->l_len < 0) {
4640 		ZFS_EXIT(zfsvfs);
4641 		return (EINVAL);
4642 	}
4643 
4644 	off = bfp->l_start;
4645 	len = bfp->l_len; /* 0 means from off to end of file */
4646 
4647 	error = zfs_freesp(zp, off, len, flag, TRUE);
4648 
4649 	ZFS_EXIT(zfsvfs);
4650 	return (error);
4651 }
4652 
4653 /*ARGSUSED*/
4654 static int
4655 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4656 {
4657 	znode_t		*zp = VTOZ(vp);
4658 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4659 	uint32_t	gen;
4660 	uint64_t	gen64;
4661 	uint64_t	object = zp->z_id;
4662 	zfid_short_t	*zfid;
4663 	int		size, i, error;
4664 
4665 	ZFS_ENTER(zfsvfs);
4666 	ZFS_VERIFY_ZP(zp);
4667 
4668 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4669 	    &gen64, sizeof (uint64_t))) != 0) {
4670 		ZFS_EXIT(zfsvfs);
4671 		return (error);
4672 	}
4673 
4674 	gen = (uint32_t)gen64;
4675 
4676 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4677 	if (fidp->fid_len < size) {
4678 		fidp->fid_len = size;
4679 		ZFS_EXIT(zfsvfs);
4680 		return (ENOSPC);
4681 	}
4682 
4683 	zfid = (zfid_short_t *)fidp;
4684 
4685 	zfid->zf_len = size;
4686 
4687 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4688 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4689 
4690 	/* Must have a non-zero generation number to distinguish from .zfs */
4691 	if (gen == 0)
4692 		gen = 1;
4693 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4694 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4695 
4696 	if (size == LONG_FID_LEN) {
4697 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4698 		zfid_long_t	*zlfid;
4699 
4700 		zlfid = (zfid_long_t *)fidp;
4701 
4702 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4703 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4704 
4705 		/* XXX - this should be the generation number for the objset */
4706 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4707 			zlfid->zf_setgen[i] = 0;
4708 	}
4709 
4710 	ZFS_EXIT(zfsvfs);
4711 	return (0);
4712 }
4713 
4714 static int
4715 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4716     caller_context_t *ct)
4717 {
4718 	znode_t		*zp, *xzp;
4719 	zfsvfs_t	*zfsvfs;
4720 	zfs_dirlock_t	*dl;
4721 	int		error;
4722 
4723 	switch (cmd) {
4724 	case _PC_LINK_MAX:
4725 		*valp = ULONG_MAX;
4726 		return (0);
4727 
4728 	case _PC_FILESIZEBITS:
4729 		*valp = 64;
4730 		return (0);
4731 
4732 	case _PC_XATTR_EXISTS:
4733 		zp = VTOZ(vp);
4734 		zfsvfs = zp->z_zfsvfs;
4735 		ZFS_ENTER(zfsvfs);
4736 		ZFS_VERIFY_ZP(zp);
4737 		*valp = 0;
4738 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4739 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4740 		if (error == 0) {
4741 			zfs_dirent_unlock(dl);
4742 			if (!zfs_dirempty(xzp))
4743 				*valp = 1;
4744 			VN_RELE(ZTOV(xzp));
4745 		} else if (error == ENOENT) {
4746 			/*
4747 			 * If there aren't extended attributes, it's the
4748 			 * same as having zero of them.
4749 			 */
4750 			error = 0;
4751 		}
4752 		ZFS_EXIT(zfsvfs);
4753 		return (error);
4754 
4755 	case _PC_SATTR_ENABLED:
4756 	case _PC_SATTR_EXISTS:
4757 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4758 		    (vp->v_type == VREG || vp->v_type == VDIR);
4759 		return (0);
4760 
4761 	case _PC_ACCESS_FILTERING:
4762 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4763 		    vp->v_type == VDIR;
4764 		return (0);
4765 
4766 	case _PC_ACL_ENABLED:
4767 		*valp = _ACL_ACE_ENABLED;
4768 		return (0);
4769 
4770 	case _PC_MIN_HOLE_SIZE:
4771 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4772 		return (0);
4773 
4774 	case _PC_TIMESTAMP_RESOLUTION:
4775 		/* nanosecond timestamp resolution */
4776 		*valp = 1L;
4777 		return (0);
4778 
4779 	default:
4780 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4781 	}
4782 }
4783 
4784 /*ARGSUSED*/
4785 static int
4786 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4787     caller_context_t *ct)
4788 {
4789 	znode_t *zp = VTOZ(vp);
4790 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4791 	int error;
4792 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4793 
4794 	ZFS_ENTER(zfsvfs);
4795 	ZFS_VERIFY_ZP(zp);
4796 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4797 	ZFS_EXIT(zfsvfs);
4798 
4799 	return (error);
4800 }
4801 
4802 /*ARGSUSED*/
4803 static int
4804 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4805     caller_context_t *ct)
4806 {
4807 	znode_t *zp = VTOZ(vp);
4808 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4809 	int error;
4810 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4811 	zilog_t	*zilog = zfsvfs->z_log;
4812 
4813 	ZFS_ENTER(zfsvfs);
4814 	ZFS_VERIFY_ZP(zp);
4815 
4816 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4817 
4818 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4819 		zil_commit(zilog, UINT64_MAX, 0);
4820 
4821 	ZFS_EXIT(zfsvfs);
4822 	return (error);
4823 }
4824 
4825 /*
4826  * Tunable, both must be a power of 2.
4827  *
4828  * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4829  * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4830  *                an arcbuf for a partial block read
4831  */
4832 int zcr_blksz_min = (1 << 10);	/* 1K */
4833 int zcr_blksz_max = (1 << 17);	/* 128K */
4834 
4835 /*ARGSUSED*/
4836 static int
4837 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
4838     caller_context_t *ct)
4839 {
4840 	znode_t	*zp = VTOZ(vp);
4841 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4842 	int max_blksz = zfsvfs->z_max_blksz;
4843 	uio_t *uio = &xuio->xu_uio;
4844 	ssize_t size = uio->uio_resid;
4845 	offset_t offset = uio->uio_loffset;
4846 	int blksz;
4847 	int fullblk, i;
4848 	arc_buf_t *abuf;
4849 	ssize_t maxsize;
4850 	int preamble, postamble;
4851 
4852 	if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4853 		return (EINVAL);
4854 
4855 	ZFS_ENTER(zfsvfs);
4856 	ZFS_VERIFY_ZP(zp);
4857 	switch (ioflag) {
4858 	case UIO_WRITE:
4859 		/*
4860 		 * Loan out an arc_buf for write if write size is bigger than
4861 		 * max_blksz, and the file's block size is also max_blksz.
4862 		 */
4863 		blksz = max_blksz;
4864 		if (size < blksz || zp->z_blksz != blksz) {
4865 			ZFS_EXIT(zfsvfs);
4866 			return (EINVAL);
4867 		}
4868 		/*
4869 		 * Caller requests buffers for write before knowing where the
4870 		 * write offset might be (e.g. NFS TCP write).
4871 		 */
4872 		if (offset == -1) {
4873 			preamble = 0;
4874 		} else {
4875 			preamble = P2PHASE(offset, blksz);
4876 			if (preamble) {
4877 				preamble = blksz - preamble;
4878 				size -= preamble;
4879 			}
4880 		}
4881 
4882 		postamble = P2PHASE(size, blksz);
4883 		size -= postamble;
4884 
4885 		fullblk = size / blksz;
4886 		(void) dmu_xuio_init(xuio,
4887 		    (preamble != 0) + fullblk + (postamble != 0));
4888 		DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
4889 		    int, postamble, int,
4890 		    (preamble != 0) + fullblk + (postamble != 0));
4891 
4892 		/*
4893 		 * Have to fix iov base/len for partial buffers.  They
4894 		 * currently represent full arc_buf's.
4895 		 */
4896 		if (preamble) {
4897 			/* data begins in the middle of the arc_buf */
4898 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4899 			    blksz);
4900 			ASSERT(abuf);
4901 			(void) dmu_xuio_add(xuio, abuf,
4902 			    blksz - preamble, preamble);
4903 		}
4904 
4905 		for (i = 0; i < fullblk; i++) {
4906 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4907 			    blksz);
4908 			ASSERT(abuf);
4909 			(void) dmu_xuio_add(xuio, abuf, 0, blksz);
4910 		}
4911 
4912 		if (postamble) {
4913 			/* data ends in the middle of the arc_buf */
4914 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
4915 			    blksz);
4916 			ASSERT(abuf);
4917 			(void) dmu_xuio_add(xuio, abuf, 0, postamble);
4918 		}
4919 		break;
4920 	case UIO_READ:
4921 		/*
4922 		 * Loan out an arc_buf for read if the read size is larger than
4923 		 * the current file block size.  Block alignment is not
4924 		 * considered.  Partial arc_buf will be loaned out for read.
4925 		 */
4926 		blksz = zp->z_blksz;
4927 		if (blksz < zcr_blksz_min)
4928 			blksz = zcr_blksz_min;
4929 		if (blksz > zcr_blksz_max)
4930 			blksz = zcr_blksz_max;
4931 		/* avoid potential complexity of dealing with it */
4932 		if (blksz > max_blksz) {
4933 			ZFS_EXIT(zfsvfs);
4934 			return (EINVAL);
4935 		}
4936 
4937 		maxsize = zp->z_size - uio->uio_loffset;
4938 		if (size > maxsize)
4939 			size = maxsize;
4940 
4941 		if (size < blksz || vn_has_cached_data(vp)) {
4942 			ZFS_EXIT(zfsvfs);
4943 			return (EINVAL);
4944 		}
4945 		break;
4946 	default:
4947 		ZFS_EXIT(zfsvfs);
4948 		return (EINVAL);
4949 	}
4950 
4951 	uio->uio_extflg = UIO_XUIO;
4952 	XUIO_XUZC_RW(xuio) = ioflag;
4953 	ZFS_EXIT(zfsvfs);
4954 	return (0);
4955 }
4956 
4957 /*ARGSUSED*/
4958 static int
4959 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
4960 {
4961 	int i;
4962 	arc_buf_t *abuf;
4963 	int ioflag = XUIO_XUZC_RW(xuio);
4964 
4965 	ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
4966 
4967 	i = dmu_xuio_cnt(xuio);
4968 	while (i-- > 0) {
4969 		abuf = dmu_xuio_arcbuf(xuio, i);
4970 		/*
4971 		 * if abuf == NULL, it must be a write buffer
4972 		 * that has been returned in zfs_write().
4973 		 */
4974 		if (abuf)
4975 			dmu_return_arcbuf(abuf);
4976 		ASSERT(abuf || ioflag == UIO_WRITE);
4977 	}
4978 
4979 	dmu_xuio_fini(xuio);
4980 	return (0);
4981 }
4982 
4983 /*
4984  * Predeclare these here so that the compiler assumes that
4985  * this is an "old style" function declaration that does
4986  * not include arguments => we won't get type mismatch errors
4987  * in the initializations that follow.
4988  */
4989 static int zfs_inval();
4990 static int zfs_isdir();
4991 
4992 static int
4993 zfs_inval()
4994 {
4995 	return (EINVAL);
4996 }
4997 
4998 static int
4999 zfs_isdir()
5000 {
5001 	return (EISDIR);
5002 }
5003 /*
5004  * Directory vnode operations template
5005  */
5006 vnodeops_t *zfs_dvnodeops;
5007 const fs_operation_def_t zfs_dvnodeops_template[] = {
5008 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5009 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5010 	VOPNAME_READ,		{ .error = zfs_isdir },
5011 	VOPNAME_WRITE,		{ .error = zfs_isdir },
5012 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5013 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5014 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5015 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5016 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5017 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5018 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5019 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5020 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5021 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
5022 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5023 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5024 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
5025 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5026 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5027 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5028 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5029 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5030 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5031 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5032 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
5033 	NULL,			NULL
5034 };
5035 
5036 /*
5037  * Regular file vnode operations template
5038  */
5039 vnodeops_t *zfs_fvnodeops;
5040 const fs_operation_def_t zfs_fvnodeops_template[] = {
5041 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5042 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5043 	VOPNAME_READ,		{ .vop_read = zfs_read },
5044 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
5045 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5046 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5047 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5048 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5049 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5050 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5051 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5052 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5053 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5054 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5055 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
5056 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
5057 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
5058 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
5059 	VOPNAME_MAP,		{ .vop_map = zfs_map },
5060 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
5061 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
5062 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5063 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5064 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5065 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5066 	VOPNAME_REQZCBUF, 	{ .vop_reqzcbuf = zfs_reqzcbuf },
5067 	VOPNAME_RETZCBUF, 	{ .vop_retzcbuf = zfs_retzcbuf },
5068 	NULL,			NULL
5069 };
5070 
5071 /*
5072  * Symbolic link vnode operations template
5073  */
5074 vnodeops_t *zfs_symvnodeops;
5075 const fs_operation_def_t zfs_symvnodeops_template[] = {
5076 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5077 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5078 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5079 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5080 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
5081 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5082 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5083 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5084 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5085 	NULL,			NULL
5086 };
5087 
5088 /*
5089  * special share hidden files vnode operations template
5090  */
5091 vnodeops_t *zfs_sharevnodeops;
5092 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5093 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5094 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5095 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5096 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5097 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5098 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5099 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5100 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5101 	NULL,			NULL
5102 };
5103 
5104 /*
5105  * Extended attribute directory vnode operations template
5106  *	This template is identical to the directory vnodes
5107  *	operation template except for restricted operations:
5108  *		VOP_MKDIR()
5109  *		VOP_SYMLINK()
5110  * Note that there are other restrictions embedded in:
5111  *	zfs_create()	- restrict type to VREG
5112  *	zfs_link()	- no links into/out of attribute space
5113  *	zfs_rename()	- no moves into/out of attribute space
5114  */
5115 vnodeops_t *zfs_xdvnodeops;
5116 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5117 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5118 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5119 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5120 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5121 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5122 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5123 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5124 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5125 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5126 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5127 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5128 	VOPNAME_MKDIR,		{ .error = zfs_inval },
5129 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5130 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5131 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
5132 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5133 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5134 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5135 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5136 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5137 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5138 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5139 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5140 	NULL,			NULL
5141 };
5142 
5143 /*
5144  * Error vnode operations template
5145  */
5146 vnodeops_t *zfs_evnodeops;
5147 const fs_operation_def_t zfs_evnodeops_template[] = {
5148 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5149 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5150 	NULL,			NULL
5151 };
5152