xref: /titanic_41/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 15d9d0b528387242011cdcc6190c9e598cfe3a07)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /* Portions Copyright 2007 Jeremy Teo */
27 
28 #pragma ident	"%Z%%M%	%I%	%E% SMI"
29 
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/time.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
36 #include <sys/vfs.h>
37 #include <sys/vfs_opreg.h>
38 #include <sys/vnode.h>
39 #include <sys/file.h>
40 #include <sys/stat.h>
41 #include <sys/kmem.h>
42 #include <sys/taskq.h>
43 #include <sys/uio.h>
44 #include <sys/vmsystm.h>
45 #include <sys/atomic.h>
46 #include <sys/vm.h>
47 #include <vm/seg_vn.h>
48 #include <vm/pvn.h>
49 #include <vm/as.h>
50 #include <sys/mman.h>
51 #include <sys/pathname.h>
52 #include <sys/cmn_err.h>
53 #include <sys/errno.h>
54 #include <sys/unistd.h>
55 #include <sys/zfs_dir.h>
56 #include <sys/zfs_acl.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/fs/zfs.h>
59 #include <sys/dmu.h>
60 #include <sys/spa.h>
61 #include <sys/txg.h>
62 #include <sys/dbuf.h>
63 #include <sys/zap.h>
64 #include <sys/dirent.h>
65 #include <sys/policy.h>
66 #include <sys/sunddi.h>
67 #include <sys/filio.h>
68 #include "fs/fs_subr.h"
69 #include <sys/zfs_ctldir.h>
70 #include <sys/zfs_fuid.h>
71 #include <sys/dnlc.h>
72 #include <sys/zfs_rlock.h>
73 #include <sys/extdirent.h>
74 #include <sys/kidmap.h>
75 #include <sys/cred_impl.h>
76 #include <sys/attr.h>
77 
78 /*
79  * Programming rules.
80  *
81  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
82  * properly lock its in-core state, create a DMU transaction, do the work,
83  * record this work in the intent log (ZIL), commit the DMU transaction,
84  * and wait for the intent log to commit if it is a synchronous operation.
85  * Moreover, the vnode ops must work in both normal and log replay context.
86  * The ordering of events is important to avoid deadlocks and references
87  * to freed memory.  The example below illustrates the following Big Rules:
88  *
89  *  (1) A check must be made in each zfs thread for a mounted file system.
90  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
91  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
92  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
93  *      can return EIO from the calling function.
94  *
95  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
96  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
97  *	First, if it's the last reference, the vnode/znode
98  *	can be freed, so the zp may point to freed memory.  Second, the last
99  *	reference will call zfs_zinactive(), which may induce a lot of work --
100  *	pushing cached pages (which acquires range locks) and syncing out
101  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
102  *	which could deadlock the system if you were already holding one.
103  *
104  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
105  *	as they can span dmu_tx_assign() calls.
106  *
107  *  (4)	Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign().
108  *	In normal operation, this will be TXG_NOWAIT.  During ZIL replay,
109  *	it will be a specific txg.  Either way, dmu_tx_assign() never blocks.
110  *	This is critical because we don't want to block while holding locks.
111  *	Note, in particular, that if a lock is sometimes acquired before
112  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
113  *	use a non-blocking assign can deadlock the system.  The scenario:
114  *
115  *	Thread A has grabbed a lock before calling dmu_tx_assign().
116  *	Thread B is in an already-assigned tx, and blocks for this lock.
117  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
118  *	forever, because the previous txg can't quiesce until B's tx commits.
119  *
120  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
121  *	then drop all locks, call dmu_tx_wait(), and try again.
122  *
123  *  (5)	If the operation succeeded, generate the intent log entry for it
124  *	before dropping locks.  This ensures that the ordering of events
125  *	in the intent log matches the order in which they actually occurred.
126  *
127  *  (6)	At the end of each vnode op, the DMU tx must always commit,
128  *	regardless of whether there were any errors.
129  *
130  *  (7)	After dropping all locks, invoke zil_commit(zilog, seq, foid)
131  *	to ensure that synchronous semantics are provided when necessary.
132  *
133  * In general, this is how things should be ordered in each vnode op:
134  *
135  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
136  * top:
137  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
138  *	rw_enter(...);			// grab any other locks you need
139  *	tx = dmu_tx_create(...);	// get DMU tx
140  *	dmu_tx_hold_*();		// hold each object you might modify
141  *	error = dmu_tx_assign(tx, zfsvfs->z_assign);	// try to assign
142  *	if (error) {
143  *		rw_exit(...);		// drop locks
144  *		zfs_dirent_unlock(dl);	// unlock directory entry
145  *		VN_RELE(...);		// release held vnodes
146  *		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
147  *			dmu_tx_wait(tx);
148  *			dmu_tx_abort(tx);
149  *			goto top;
150  *		}
151  *		dmu_tx_abort(tx);	// abort DMU tx
152  *		ZFS_EXIT(zfsvfs);	// finished in zfs
153  *		return (error);		// really out of space
154  *	}
155  *	error = do_real_work();		// do whatever this VOP does
156  *	if (error == 0)
157  *		zfs_log_*(...);		// on success, make ZIL entry
158  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
159  *	rw_exit(...);			// drop locks
160  *	zfs_dirent_unlock(dl);		// unlock directory entry
161  *	VN_RELE(...);			// release held vnodes
162  *	zil_commit(zilog, seq, foid);	// synchronous when necessary
163  *	ZFS_EXIT(zfsvfs);		// finished in zfs
164  *	return (error);			// done, report error
165  */
166 
167 /* ARGSUSED */
168 static int
169 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
170 {
171 	znode_t	*zp = VTOZ(*vpp);
172 
173 	if ((flag & FWRITE) && (zp->z_phys->zp_flags & ZFS_APPENDONLY) &&
174 	    ((flag & FAPPEND) == 0)) {
175 		return (EPERM);
176 	}
177 
178 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
179 	    ZTOV(zp)->v_type == VREG &&
180 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
181 	    zp->z_phys->zp_size > 0)
182 		if (fs_vscan(*vpp, cr, 0) != 0)
183 			return (EACCES);
184 
185 	/* Keep a count of the synchronous opens in the znode */
186 	if (flag & (FSYNC | FDSYNC))
187 		atomic_inc_32(&zp->z_sync_cnt);
188 
189 	return (0);
190 }
191 
192 /* ARGSUSED */
193 static int
194 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
195     caller_context_t *ct)
196 {
197 	znode_t	*zp = VTOZ(vp);
198 
199 	/* Decrement the synchronous opens in the znode */
200 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
201 		atomic_dec_32(&zp->z_sync_cnt);
202 
203 	/*
204 	 * Clean up any locks held by this process on the vp.
205 	 */
206 	cleanlocks(vp, ddi_get_pid(), 0);
207 	cleanshares(vp, ddi_get_pid());
208 
209 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
210 	    ZTOV(zp)->v_type == VREG &&
211 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
212 	    zp->z_phys->zp_size > 0)
213 		VERIFY(fs_vscan(vp, cr, 1) == 0);
214 
215 	return (0);
216 }
217 
218 /*
219  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
220  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
221  */
222 static int
223 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
224 {
225 	znode_t	*zp = VTOZ(vp);
226 	uint64_t noff = (uint64_t)*off; /* new offset */
227 	uint64_t file_sz;
228 	int error;
229 	boolean_t hole;
230 
231 	file_sz = zp->z_phys->zp_size;
232 	if (noff >= file_sz)  {
233 		return (ENXIO);
234 	}
235 
236 	if (cmd == _FIO_SEEK_HOLE)
237 		hole = B_TRUE;
238 	else
239 		hole = B_FALSE;
240 
241 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
242 
243 	/* end of file? */
244 	if ((error == ESRCH) || (noff > file_sz)) {
245 		/*
246 		 * Handle the virtual hole at the end of file.
247 		 */
248 		if (hole) {
249 			*off = file_sz;
250 			return (0);
251 		}
252 		return (ENXIO);
253 	}
254 
255 	if (noff < *off)
256 		return (error);
257 	*off = noff;
258 	return (error);
259 }
260 
261 /* ARGSUSED */
262 static int
263 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
264     int *rvalp, caller_context_t *ct)
265 {
266 	offset_t off;
267 	int error;
268 	zfsvfs_t *zfsvfs;
269 	znode_t *zp;
270 
271 	switch (com) {
272 	case _FIOFFS:
273 		return (zfs_sync(vp->v_vfsp, 0, cred));
274 
275 		/*
276 		 * The following two ioctls are used by bfu.  Faking out,
277 		 * necessary to avoid bfu errors.
278 		 */
279 	case _FIOGDIO:
280 	case _FIOSDIO:
281 		return (0);
282 
283 	case _FIO_SEEK_DATA:
284 	case _FIO_SEEK_HOLE:
285 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
286 			return (EFAULT);
287 
288 		zp = VTOZ(vp);
289 		zfsvfs = zp->z_zfsvfs;
290 		ZFS_ENTER(zfsvfs);
291 		ZFS_VERIFY_ZP(zp);
292 
293 		/* offset parameter is in/out */
294 		error = zfs_holey(vp, com, &off);
295 		ZFS_EXIT(zfsvfs);
296 		if (error)
297 			return (error);
298 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
299 			return (EFAULT);
300 		return (0);
301 	}
302 	return (ENOTTY);
303 }
304 
305 /*
306  * When a file is memory mapped, we must keep the IO data synchronized
307  * between the DMU cache and the memory mapped pages.  What this means:
308  *
309  * On Write:	If we find a memory mapped page, we write to *both*
310  *		the page and the dmu buffer.
311  *
312  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
313  *	the file is memory mapped.
314  */
315 static int
316 mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx)
317 {
318 	znode_t	*zp = VTOZ(vp);
319 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
320 	int64_t	start, off;
321 	int len = nbytes;
322 	int error = 0;
323 
324 	start = uio->uio_loffset;
325 	off = start & PAGEOFFSET;
326 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
327 		page_t *pp;
328 		uint64_t bytes = MIN(PAGESIZE - off, len);
329 		uint64_t woff = uio->uio_loffset;
330 
331 		/*
332 		 * We don't want a new page to "appear" in the middle of
333 		 * the file update (because it may not get the write
334 		 * update data), so we grab a lock to block
335 		 * zfs_getpage().
336 		 */
337 		rw_enter(&zp->z_map_lock, RW_WRITER);
338 		if (pp = page_lookup(vp, start, SE_SHARED)) {
339 			caddr_t va;
340 
341 			rw_exit(&zp->z_map_lock);
342 			va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L);
343 			error = uiomove(va+off, bytes, UIO_WRITE, uio);
344 			if (error == 0) {
345 				dmu_write(zfsvfs->z_os, zp->z_id,
346 				    woff, bytes, va+off, tx);
347 			}
348 			ppmapout(va);
349 			page_unlock(pp);
350 		} else {
351 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
352 			    uio, bytes, tx);
353 			rw_exit(&zp->z_map_lock);
354 		}
355 		len -= bytes;
356 		off = 0;
357 		if (error)
358 			break;
359 	}
360 	return (error);
361 }
362 
363 /*
364  * When a file is memory mapped, we must keep the IO data synchronized
365  * between the DMU cache and the memory mapped pages.  What this means:
366  *
367  * On Read:	We "read" preferentially from memory mapped pages,
368  *		else we default from the dmu buffer.
369  *
370  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
371  *	the file is memory mapped.
372  */
373 static int
374 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
375 {
376 	znode_t *zp = VTOZ(vp);
377 	objset_t *os = zp->z_zfsvfs->z_os;
378 	int64_t	start, off;
379 	int len = nbytes;
380 	int error = 0;
381 
382 	start = uio->uio_loffset;
383 	off = start & PAGEOFFSET;
384 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
385 		page_t *pp;
386 		uint64_t bytes = MIN(PAGESIZE - off, len);
387 
388 		if (pp = page_lookup(vp, start, SE_SHARED)) {
389 			caddr_t va;
390 
391 			va = ppmapin(pp, PROT_READ, (caddr_t)-1L);
392 			error = uiomove(va + off, bytes, UIO_READ, uio);
393 			ppmapout(va);
394 			page_unlock(pp);
395 		} else {
396 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
397 		}
398 		len -= bytes;
399 		off = 0;
400 		if (error)
401 			break;
402 	}
403 	return (error);
404 }
405 
406 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
407 
408 /*
409  * Read bytes from specified file into supplied buffer.
410  *
411  *	IN:	vp	- vnode of file to be read from.
412  *		uio	- structure supplying read location, range info,
413  *			  and return buffer.
414  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
415  *		cr	- credentials of caller.
416  *		ct	- caller context
417  *
418  *	OUT:	uio	- updated offset and range, buffer filled.
419  *
420  *	RETURN:	0 if success
421  *		error code if failure
422  *
423  * Side Effects:
424  *	vp - atime updated if byte count > 0
425  */
426 /* ARGSUSED */
427 static int
428 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
429 {
430 	znode_t		*zp = VTOZ(vp);
431 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
432 	objset_t	*os;
433 	ssize_t		n, nbytes;
434 	int		error;
435 	rl_t		*rl;
436 
437 	ZFS_ENTER(zfsvfs);
438 	ZFS_VERIFY_ZP(zp);
439 	os = zfsvfs->z_os;
440 
441 	if (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) {
442 		ZFS_EXIT(zfsvfs);
443 		return (EACCES);
444 	}
445 
446 	/*
447 	 * Validate file offset
448 	 */
449 	if (uio->uio_loffset < (offset_t)0) {
450 		ZFS_EXIT(zfsvfs);
451 		return (EINVAL);
452 	}
453 
454 	/*
455 	 * Fasttrack empty reads
456 	 */
457 	if (uio->uio_resid == 0) {
458 		ZFS_EXIT(zfsvfs);
459 		return (0);
460 	}
461 
462 	/*
463 	 * Check for mandatory locks
464 	 */
465 	if (MANDMODE((mode_t)zp->z_phys->zp_mode)) {
466 		if (error = chklock(vp, FREAD,
467 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
468 			ZFS_EXIT(zfsvfs);
469 			return (error);
470 		}
471 	}
472 
473 	/*
474 	 * If we're in FRSYNC mode, sync out this znode before reading it.
475 	 */
476 	if (ioflag & FRSYNC)
477 		zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
478 
479 	/*
480 	 * Lock the range against changes.
481 	 */
482 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
483 
484 	/*
485 	 * If we are reading past end-of-file we can skip
486 	 * to the end; but we might still need to set atime.
487 	 */
488 	if (uio->uio_loffset >= zp->z_phys->zp_size) {
489 		error = 0;
490 		goto out;
491 	}
492 
493 	ASSERT(uio->uio_loffset < zp->z_phys->zp_size);
494 	n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset);
495 
496 	while (n > 0) {
497 		nbytes = MIN(n, zfs_read_chunk_size -
498 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
499 
500 		if (vn_has_cached_data(vp))
501 			error = mappedread(vp, nbytes, uio);
502 		else
503 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
504 		if (error)
505 			break;
506 
507 		n -= nbytes;
508 	}
509 
510 out:
511 	zfs_range_unlock(rl);
512 
513 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
514 	ZFS_EXIT(zfsvfs);
515 	return (error);
516 }
517 
518 /*
519  * Fault in the pages of the first n bytes specified by the uio structure.
520  * 1 byte in each page is touched and the uio struct is unmodified.
521  * Any error will exit this routine as this is only a best
522  * attempt to get the pages resident. This is a copy of ufs_trans_touch().
523  */
524 static void
525 zfs_prefault_write(ssize_t n, struct uio *uio)
526 {
527 	struct iovec *iov;
528 	ulong_t cnt, incr;
529 	caddr_t p;
530 	uint8_t tmp;
531 
532 	iov = uio->uio_iov;
533 
534 	while (n) {
535 		cnt = MIN(iov->iov_len, n);
536 		if (cnt == 0) {
537 			/* empty iov entry */
538 			iov++;
539 			continue;
540 		}
541 		n -= cnt;
542 		/*
543 		 * touch each page in this segment.
544 		 */
545 		p = iov->iov_base;
546 		while (cnt) {
547 			switch (uio->uio_segflg) {
548 			case UIO_USERSPACE:
549 			case UIO_USERISPACE:
550 				if (fuword8(p, &tmp))
551 					return;
552 				break;
553 			case UIO_SYSSPACE:
554 				if (kcopy(p, &tmp, 1))
555 					return;
556 				break;
557 			}
558 			incr = MIN(cnt, PAGESIZE);
559 			p += incr;
560 			cnt -= incr;
561 		}
562 		/*
563 		 * touch the last byte in case it straddles a page.
564 		 */
565 		p--;
566 		switch (uio->uio_segflg) {
567 		case UIO_USERSPACE:
568 		case UIO_USERISPACE:
569 			if (fuword8(p, &tmp))
570 				return;
571 			break;
572 		case UIO_SYSSPACE:
573 			if (kcopy(p, &tmp, 1))
574 				return;
575 			break;
576 		}
577 		iov++;
578 	}
579 }
580 
581 /*
582  * Write the bytes to a file.
583  *
584  *	IN:	vp	- vnode of file to be written to.
585  *		uio	- structure supplying write location, range info,
586  *			  and data buffer.
587  *		ioflag	- FAPPEND flag set if in append mode.
588  *		cr	- credentials of caller.
589  *		ct	- caller context (NFS/CIFS fem monitor only)
590  *
591  *	OUT:	uio	- updated offset and range.
592  *
593  *	RETURN:	0 if success
594  *		error code if failure
595  *
596  * Timestamps:
597  *	vp - ctime|mtime updated if byte count > 0
598  */
599 /* ARGSUSED */
600 static int
601 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
602 {
603 	znode_t		*zp = VTOZ(vp);
604 	rlim64_t	limit = uio->uio_llimit;
605 	ssize_t		start_resid = uio->uio_resid;
606 	ssize_t		tx_bytes;
607 	uint64_t	end_size;
608 	dmu_tx_t	*tx;
609 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
610 	zilog_t		*zilog;
611 	offset_t	woff;
612 	ssize_t		n, nbytes;
613 	rl_t		*rl;
614 	int		max_blksz = zfsvfs->z_max_blksz;
615 	uint64_t	pflags = zp->z_phys->zp_flags;
616 	int		error;
617 
618 	/*
619 	 * If immutable or not appending then return EPERM
620 	 */
621 	if ((pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
622 	    ((pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
623 	    (uio->uio_loffset < zp->z_phys->zp_size)))
624 		return (EPERM);
625 
626 	/*
627 	 * Fasttrack empty write
628 	 */
629 	n = start_resid;
630 	if (n == 0)
631 		return (0);
632 
633 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
634 		limit = MAXOFFSET_T;
635 
636 	ZFS_ENTER(zfsvfs);
637 	ZFS_VERIFY_ZP(zp);
638 	zilog = zfsvfs->z_log;
639 
640 	/*
641 	 * Pre-fault the pages to ensure slow (eg NFS) pages
642 	 * don't hold up txg.
643 	 */
644 	zfs_prefault_write(n, uio);
645 
646 	/*
647 	 * If in append mode, set the io offset pointer to eof.
648 	 */
649 	if (ioflag & FAPPEND) {
650 		/*
651 		 * Range lock for a file append:
652 		 * The value for the start of range will be determined by
653 		 * zfs_range_lock() (to guarantee append semantics).
654 		 * If this write will cause the block size to increase,
655 		 * zfs_range_lock() will lock the entire file, so we must
656 		 * later reduce the range after we grow the block size.
657 		 */
658 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
659 		if (rl->r_len == UINT64_MAX) {
660 			/* overlocked, zp_size can't change */
661 			woff = uio->uio_loffset = zp->z_phys->zp_size;
662 		} else {
663 			woff = uio->uio_loffset = rl->r_off;
664 		}
665 	} else {
666 		woff = uio->uio_loffset;
667 		/*
668 		 * Validate file offset
669 		 */
670 		if (woff < 0) {
671 			ZFS_EXIT(zfsvfs);
672 			return (EINVAL);
673 		}
674 
675 		/*
676 		 * If we need to grow the block size then zfs_range_lock()
677 		 * will lock a wider range than we request here.
678 		 * Later after growing the block size we reduce the range.
679 		 */
680 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
681 	}
682 
683 	if (woff >= limit) {
684 		zfs_range_unlock(rl);
685 		ZFS_EXIT(zfsvfs);
686 		return (EFBIG);
687 	}
688 
689 	if ((woff + n) > limit || woff > (limit - n))
690 		n = limit - woff;
691 
692 	/*
693 	 * Check for mandatory locks
694 	 */
695 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) &&
696 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
697 		zfs_range_unlock(rl);
698 		ZFS_EXIT(zfsvfs);
699 		return (error);
700 	}
701 	end_size = MAX(zp->z_phys->zp_size, woff + n);
702 
703 	/*
704 	 * Write the file in reasonable size chunks.  Each chunk is written
705 	 * in a separate transaction; this keeps the intent log records small
706 	 * and allows us to do more fine-grained space accounting.
707 	 */
708 	while (n > 0) {
709 		/*
710 		 * Start a transaction.
711 		 */
712 		woff = uio->uio_loffset;
713 		tx = dmu_tx_create(zfsvfs->z_os);
714 		dmu_tx_hold_bonus(tx, zp->z_id);
715 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
716 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
717 		if (error) {
718 			if (error == ERESTART &&
719 			    zfsvfs->z_assign == TXG_NOWAIT) {
720 				dmu_tx_wait(tx);
721 				dmu_tx_abort(tx);
722 				continue;
723 			}
724 			dmu_tx_abort(tx);
725 			break;
726 		}
727 
728 		/*
729 		 * If zfs_range_lock() over-locked we grow the blocksize
730 		 * and then reduce the lock range.  This will only happen
731 		 * on the first iteration since zfs_range_reduce() will
732 		 * shrink down r_len to the appropriate size.
733 		 */
734 		if (rl->r_len == UINT64_MAX) {
735 			uint64_t new_blksz;
736 
737 			if (zp->z_blksz > max_blksz) {
738 				ASSERT(!ISP2(zp->z_blksz));
739 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
740 			} else {
741 				new_blksz = MIN(end_size, max_blksz);
742 			}
743 			zfs_grow_blocksize(zp, new_blksz, tx);
744 			zfs_range_reduce(rl, woff, n);
745 		}
746 
747 		/*
748 		 * XXX - should we really limit each write to z_max_blksz?
749 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
750 		 */
751 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
752 		rw_enter(&zp->z_map_lock, RW_READER);
753 
754 		tx_bytes = uio->uio_resid;
755 		if (vn_has_cached_data(vp)) {
756 			rw_exit(&zp->z_map_lock);
757 			error = mappedwrite(vp, nbytes, uio, tx);
758 		} else {
759 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
760 			    uio, nbytes, tx);
761 			rw_exit(&zp->z_map_lock);
762 		}
763 		tx_bytes -= uio->uio_resid;
764 
765 		/*
766 		 * If we made no progress, we're done.  If we made even
767 		 * partial progress, update the znode and ZIL accordingly.
768 		 */
769 		if (tx_bytes == 0) {
770 			dmu_tx_commit(tx);
771 			ASSERT(error != 0);
772 			break;
773 		}
774 
775 		/*
776 		 * Clear Set-UID/Set-GID bits on successful write if not
777 		 * privileged and at least one of the excute bits is set.
778 		 *
779 		 * It would be nice to to this after all writes have
780 		 * been done, but that would still expose the ISUID/ISGID
781 		 * to another app after the partial write is committed.
782 		 *
783 		 * Note: we don't call zfs_fuid_map_id() here because
784 		 * user 0 is not an ephemeral uid.
785 		 */
786 		mutex_enter(&zp->z_acl_lock);
787 		if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) |
788 		    (S_IXUSR >> 6))) != 0 &&
789 		    (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 &&
790 		    secpolicy_vnode_setid_retain(cr,
791 		    (zp->z_phys->zp_mode & S_ISUID) != 0 &&
792 		    zp->z_phys->zp_uid == 0) != 0) {
793 			zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID);
794 		}
795 		mutex_exit(&zp->z_acl_lock);
796 
797 		/*
798 		 * Update time stamp.  NOTE: This marks the bonus buffer as
799 		 * dirty, so we don't have to do it again for zp_size.
800 		 */
801 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
802 
803 		/*
804 		 * Update the file size (zp_size) if it has changed;
805 		 * account for possible concurrent updates.
806 		 */
807 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset)
808 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
809 			    uio->uio_loffset);
810 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
811 		dmu_tx_commit(tx);
812 
813 		if (error != 0)
814 			break;
815 		ASSERT(tx_bytes == nbytes);
816 		n -= nbytes;
817 	}
818 
819 	zfs_range_unlock(rl);
820 
821 	/*
822 	 * If we're in replay mode, or we made no progress, return error.
823 	 * Otherwise, it's at least a partial write, so it's successful.
824 	 */
825 	if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) {
826 		ZFS_EXIT(zfsvfs);
827 		return (error);
828 	}
829 
830 	if (ioflag & (FSYNC | FDSYNC))
831 		zil_commit(zilog, zp->z_last_itx, zp->z_id);
832 
833 	ZFS_EXIT(zfsvfs);
834 	return (0);
835 }
836 
837 void
838 zfs_get_done(dmu_buf_t *db, void *vzgd)
839 {
840 	zgd_t *zgd = (zgd_t *)vzgd;
841 	rl_t *rl = zgd->zgd_rl;
842 	vnode_t *vp = ZTOV(rl->r_zp);
843 
844 	dmu_buf_rele(db, vzgd);
845 	zfs_range_unlock(rl);
846 	VN_RELE(vp);
847 	zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
848 	kmem_free(zgd, sizeof (zgd_t));
849 }
850 
851 /*
852  * Get data to generate a TX_WRITE intent log record.
853  */
854 int
855 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
856 {
857 	zfsvfs_t *zfsvfs = arg;
858 	objset_t *os = zfsvfs->z_os;
859 	znode_t *zp;
860 	uint64_t off = lr->lr_offset;
861 	dmu_buf_t *db;
862 	rl_t *rl;
863 	zgd_t *zgd;
864 	int dlen = lr->lr_length;		/* length of user data */
865 	int error = 0;
866 
867 	ASSERT(zio);
868 	ASSERT(dlen != 0);
869 
870 	/*
871 	 * Nothing to do if the file has been removed
872 	 */
873 	if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0)
874 		return (ENOENT);
875 	if (zp->z_unlinked) {
876 		VN_RELE(ZTOV(zp));
877 		return (ENOENT);
878 	}
879 
880 	/*
881 	 * Write records come in two flavors: immediate and indirect.
882 	 * For small writes it's cheaper to store the data with the
883 	 * log record (immediate); for large writes it's cheaper to
884 	 * sync the data and get a pointer to it (indirect) so that
885 	 * we don't have to write the data twice.
886 	 */
887 	if (buf != NULL) { /* immediate write */
888 		rl = zfs_range_lock(zp, off, dlen, RL_READER);
889 		/* test for truncation needs to be done while range locked */
890 		if (off >= zp->z_phys->zp_size) {
891 			error = ENOENT;
892 			goto out;
893 		}
894 		VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf));
895 	} else { /* indirect write */
896 		uint64_t boff; /* block starting offset */
897 
898 		/*
899 		 * Have to lock the whole block to ensure when it's
900 		 * written out and it's checksum is being calculated
901 		 * that no one can change the data. We need to re-check
902 		 * blocksize after we get the lock in case it's changed!
903 		 */
904 		for (;;) {
905 			if (ISP2(zp->z_blksz)) {
906 				boff = P2ALIGN_TYPED(off, zp->z_blksz,
907 				    uint64_t);
908 			} else {
909 				boff = 0;
910 			}
911 			dlen = zp->z_blksz;
912 			rl = zfs_range_lock(zp, boff, dlen, RL_READER);
913 			if (zp->z_blksz == dlen)
914 				break;
915 			zfs_range_unlock(rl);
916 		}
917 		/* test for truncation needs to be done while range locked */
918 		if (off >= zp->z_phys->zp_size) {
919 			error = ENOENT;
920 			goto out;
921 		}
922 		zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
923 		zgd->zgd_rl = rl;
924 		zgd->zgd_zilog = zfsvfs->z_log;
925 		zgd->zgd_bp = &lr->lr_blkptr;
926 		VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db));
927 		ASSERT(boff == db->db_offset);
928 		lr->lr_blkoff = off - boff;
929 		error = dmu_sync(zio, db, &lr->lr_blkptr,
930 		    lr->lr_common.lrc_txg, zfs_get_done, zgd);
931 		ASSERT((error && error != EINPROGRESS) ||
932 		    lr->lr_length <= zp->z_blksz);
933 		if (error == 0)
934 			zil_add_block(zfsvfs->z_log, &lr->lr_blkptr);
935 		/*
936 		 * If we get EINPROGRESS, then we need to wait for a
937 		 * write IO initiated by dmu_sync() to complete before
938 		 * we can release this dbuf.  We will finish everything
939 		 * up in the zfs_get_done() callback.
940 		 */
941 		if (error == EINPROGRESS)
942 			return (0);
943 		dmu_buf_rele(db, zgd);
944 		kmem_free(zgd, sizeof (zgd_t));
945 	}
946 out:
947 	zfs_range_unlock(rl);
948 	VN_RELE(ZTOV(zp));
949 	return (error);
950 }
951 
952 /*ARGSUSED*/
953 static int
954 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
955     caller_context_t *ct)
956 {
957 	znode_t *zp = VTOZ(vp);
958 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
959 	int error;
960 
961 	ZFS_ENTER(zfsvfs);
962 	ZFS_VERIFY_ZP(zp);
963 
964 	if (flag & V_ACE_MASK)
965 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
966 	else
967 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
968 
969 	ZFS_EXIT(zfsvfs);
970 	return (error);
971 }
972 
973 /*
974  * Lookup an entry in a directory, or an extended attribute directory.
975  * If it exists, return a held vnode reference for it.
976  *
977  *	IN:	dvp	- vnode of directory to search.
978  *		nm	- name of entry to lookup.
979  *		pnp	- full pathname to lookup [UNUSED].
980  *		flags	- LOOKUP_XATTR set if looking for an attribute.
981  *		rdir	- root directory vnode [UNUSED].
982  *		cr	- credentials of caller.
983  *		ct	- caller context
984  *		direntflags - directory lookup flags
985  *		realpnp - returned pathname.
986  *
987  *	OUT:	vpp	- vnode of located entry, NULL if not found.
988  *
989  *	RETURN:	0 if success
990  *		error code if failure
991  *
992  * Timestamps:
993  *	NA
994  */
995 /* ARGSUSED */
996 static int
997 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
998     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
999     int *direntflags, pathname_t *realpnp)
1000 {
1001 	znode_t *zdp = VTOZ(dvp);
1002 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1003 	int	error;
1004 
1005 	ZFS_ENTER(zfsvfs);
1006 	ZFS_VERIFY_ZP(zdp);
1007 
1008 	*vpp = NULL;
1009 
1010 	if (flags & LOOKUP_XATTR) {
1011 		/*
1012 		 * If the xattr property is off, refuse the lookup request.
1013 		 */
1014 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1015 			ZFS_EXIT(zfsvfs);
1016 			return (EINVAL);
1017 		}
1018 
1019 		/*
1020 		 * We don't allow recursive attributes..
1021 		 * Maybe someday we will.
1022 		 */
1023 		if (zdp->z_phys->zp_flags & ZFS_XATTR) {
1024 			ZFS_EXIT(zfsvfs);
1025 			return (EINVAL);
1026 		}
1027 
1028 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1029 			ZFS_EXIT(zfsvfs);
1030 			return (error);
1031 		}
1032 
1033 		/*
1034 		 * Do we have permission to get into attribute directory?
1035 		 */
1036 
1037 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1038 		    B_FALSE, cr)) {
1039 			VN_RELE(*vpp);
1040 			*vpp = NULL;
1041 		}
1042 
1043 		ZFS_EXIT(zfsvfs);
1044 		return (error);
1045 	}
1046 
1047 	if (dvp->v_type != VDIR) {
1048 		ZFS_EXIT(zfsvfs);
1049 		return (ENOTDIR);
1050 	}
1051 
1052 	/*
1053 	 * Check accessibility of directory.
1054 	 */
1055 
1056 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1057 		ZFS_EXIT(zfsvfs);
1058 		return (error);
1059 	}
1060 
1061 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1062 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1063 		ZFS_EXIT(zfsvfs);
1064 		return (EILSEQ);
1065 	}
1066 
1067 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1068 	if (error == 0) {
1069 		/*
1070 		 * Convert device special files
1071 		 */
1072 		if (IS_DEVVP(*vpp)) {
1073 			vnode_t	*svp;
1074 
1075 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1076 			VN_RELE(*vpp);
1077 			if (svp == NULL)
1078 				error = ENOSYS;
1079 			else
1080 				*vpp = svp;
1081 		}
1082 	}
1083 
1084 	ZFS_EXIT(zfsvfs);
1085 	return (error);
1086 }
1087 
1088 /*
1089  * Attempt to create a new entry in a directory.  If the entry
1090  * already exists, truncate the file if permissible, else return
1091  * an error.  Return the vp of the created or trunc'd file.
1092  *
1093  *	IN:	dvp	- vnode of directory to put new file entry in.
1094  *		name	- name of new file entry.
1095  *		vap	- attributes of new file.
1096  *		excl	- flag indicating exclusive or non-exclusive mode.
1097  *		mode	- mode to open file with.
1098  *		cr	- credentials of caller.
1099  *		flag	- large file flag [UNUSED].
1100  *		ct	- caller context
1101  *		vsecp 	- ACL to be set
1102  *
1103  *	OUT:	vpp	- vnode of created or trunc'd entry.
1104  *
1105  *	RETURN:	0 if success
1106  *		error code if failure
1107  *
1108  * Timestamps:
1109  *	dvp - ctime|mtime updated if new entry created
1110  *	 vp - ctime|mtime always, atime if new
1111  */
1112 
1113 /* ARGSUSED */
1114 static int
1115 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1116     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1117     vsecattr_t *vsecp)
1118 {
1119 	znode_t		*zp, *dzp = VTOZ(dvp);
1120 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1121 	zilog_t		*zilog;
1122 	objset_t	*os;
1123 	zfs_dirlock_t	*dl;
1124 	dmu_tx_t	*tx;
1125 	int		error;
1126 	zfs_acl_t	*aclp = NULL;
1127 	zfs_fuid_info_t *fuidp = NULL;
1128 
1129 	/*
1130 	 * If we have an ephemeral id, ACL, or XVATTR then
1131 	 * make sure file system is at proper version
1132 	 */
1133 
1134 	if (zfsvfs->z_use_fuids == B_FALSE &&
1135 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1136 	    IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))))
1137 		return (EINVAL);
1138 
1139 	ZFS_ENTER(zfsvfs);
1140 	ZFS_VERIFY_ZP(dzp);
1141 	os = zfsvfs->z_os;
1142 	zilog = zfsvfs->z_log;
1143 
1144 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1145 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1146 		ZFS_EXIT(zfsvfs);
1147 		return (EILSEQ);
1148 	}
1149 
1150 	if (vap->va_mask & AT_XVATTR) {
1151 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1152 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1153 			ZFS_EXIT(zfsvfs);
1154 			return (error);
1155 		}
1156 	}
1157 top:
1158 	*vpp = NULL;
1159 
1160 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1161 		vap->va_mode &= ~VSVTX;
1162 
1163 	if (*name == '\0') {
1164 		/*
1165 		 * Null component name refers to the directory itself.
1166 		 */
1167 		VN_HOLD(dvp);
1168 		zp = dzp;
1169 		dl = NULL;
1170 		error = 0;
1171 	} else {
1172 		/* possible VN_HOLD(zp) */
1173 		int zflg = 0;
1174 
1175 		if (flag & FIGNORECASE)
1176 			zflg |= ZCILOOK;
1177 
1178 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1179 		    NULL, NULL);
1180 		if (error) {
1181 			if (strcmp(name, "..") == 0)
1182 				error = EISDIR;
1183 			ZFS_EXIT(zfsvfs);
1184 			if (aclp)
1185 				zfs_acl_free(aclp);
1186 			return (error);
1187 		}
1188 	}
1189 	if (vsecp && aclp == NULL) {
1190 		error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp);
1191 		if (error) {
1192 			ZFS_EXIT(zfsvfs);
1193 			if (dl)
1194 				zfs_dirent_unlock(dl);
1195 			return (error);
1196 		}
1197 	}
1198 
1199 	if (zp == NULL) {
1200 		uint64_t txtype;
1201 
1202 		/*
1203 		 * Create a new file object and update the directory
1204 		 * to reference it.
1205 		 */
1206 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1207 			goto out;
1208 		}
1209 
1210 		/*
1211 		 * We only support the creation of regular files in
1212 		 * extended attribute directories.
1213 		 */
1214 		if ((dzp->z_phys->zp_flags & ZFS_XATTR) &&
1215 		    (vap->va_type != VREG)) {
1216 			error = EINVAL;
1217 			goto out;
1218 		}
1219 
1220 		tx = dmu_tx_create(os);
1221 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1222 		if ((aclp && aclp->z_has_fuids) || IS_EPHEMERAL(crgetuid(cr)) ||
1223 		    IS_EPHEMERAL(crgetgid(cr))) {
1224 			if (zfsvfs->z_fuid_obj == 0) {
1225 				dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1226 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1227 				    FUID_SIZE_ESTIMATE(zfsvfs));
1228 				dmu_tx_hold_zap(tx, MASTER_NODE_OBJ,
1229 				    FALSE, NULL);
1230 			} else {
1231 				dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
1232 				dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
1233 				    FUID_SIZE_ESTIMATE(zfsvfs));
1234 			}
1235 		}
1236 		dmu_tx_hold_bonus(tx, dzp->z_id);
1237 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1238 		if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) {
1239 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1240 			    0, SPA_MAXBLOCKSIZE);
1241 		}
1242 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
1243 		if (error) {
1244 			zfs_dirent_unlock(dl);
1245 			if (error == ERESTART &&
1246 			    zfsvfs->z_assign == TXG_NOWAIT) {
1247 				dmu_tx_wait(tx);
1248 				dmu_tx_abort(tx);
1249 				goto top;
1250 			}
1251 			dmu_tx_abort(tx);
1252 			ZFS_EXIT(zfsvfs);
1253 			if (aclp)
1254 				zfs_acl_free(aclp);
1255 			return (error);
1256 		}
1257 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp);
1258 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1259 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1260 		if (flag & FIGNORECASE)
1261 			txtype |= TX_CI;
1262 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1263 		    vsecp, fuidp, vap);
1264 		if (fuidp)
1265 			zfs_fuid_info_free(fuidp);
1266 		dmu_tx_commit(tx);
1267 	} else {
1268 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1269 
1270 		/*
1271 		 * A directory entry already exists for this name.
1272 		 */
1273 		/*
1274 		 * Can't truncate an existing file if in exclusive mode.
1275 		 */
1276 		if (excl == EXCL) {
1277 			error = EEXIST;
1278 			goto out;
1279 		}
1280 		/*
1281 		 * Can't open a directory for writing.
1282 		 */
1283 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1284 			error = EISDIR;
1285 			goto out;
1286 		}
1287 		/*
1288 		 * Verify requested access to file.
1289 		 */
1290 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1291 			goto out;
1292 		}
1293 
1294 		mutex_enter(&dzp->z_lock);
1295 		dzp->z_seq++;
1296 		mutex_exit(&dzp->z_lock);
1297 
1298 		/*
1299 		 * Truncate regular files if requested.
1300 		 */
1301 		if ((ZTOV(zp)->v_type == VREG) &&
1302 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1303 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1304 			if (error == ERESTART &&
1305 			    zfsvfs->z_assign == TXG_NOWAIT) {
1306 				/* NB: we already did dmu_tx_wait() */
1307 				zfs_dirent_unlock(dl);
1308 				VN_RELE(ZTOV(zp));
1309 				goto top;
1310 			}
1311 
1312 			if (error == 0) {
1313 				vnevent_create(ZTOV(zp), ct);
1314 			}
1315 		}
1316 	}
1317 out:
1318 
1319 	if (dl)
1320 		zfs_dirent_unlock(dl);
1321 
1322 	if (error) {
1323 		if (zp)
1324 			VN_RELE(ZTOV(zp));
1325 	} else {
1326 		*vpp = ZTOV(zp);
1327 		/*
1328 		 * If vnode is for a device return a specfs vnode instead.
1329 		 */
1330 		if (IS_DEVVP(*vpp)) {
1331 			struct vnode *svp;
1332 
1333 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1334 			VN_RELE(*vpp);
1335 			if (svp == NULL) {
1336 				error = ENOSYS;
1337 			}
1338 			*vpp = svp;
1339 		}
1340 	}
1341 	if (aclp)
1342 		zfs_acl_free(aclp);
1343 
1344 	ZFS_EXIT(zfsvfs);
1345 	return (error);
1346 }
1347 
1348 /*
1349  * Remove an entry from a directory.
1350  *
1351  *	IN:	dvp	- vnode of directory to remove entry from.
1352  *		name	- name of entry to remove.
1353  *		cr	- credentials of caller.
1354  *		ct	- caller context
1355  *		flags	- case flags
1356  *
1357  *	RETURN:	0 if success
1358  *		error code if failure
1359  *
1360  * Timestamps:
1361  *	dvp - ctime|mtime
1362  *	 vp - ctime (if nlink > 0)
1363  */
1364 /*ARGSUSED*/
1365 static int
1366 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1367     int flags)
1368 {
1369 	znode_t		*zp, *dzp = VTOZ(dvp);
1370 	znode_t		*xzp = NULL;
1371 	vnode_t		*vp;
1372 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1373 	zilog_t		*zilog;
1374 	uint64_t	acl_obj, xattr_obj;
1375 	zfs_dirlock_t	*dl;
1376 	dmu_tx_t	*tx;
1377 	boolean_t	may_delete_now, delete_now = FALSE;
1378 	boolean_t	unlinked;
1379 	uint64_t	txtype;
1380 	pathname_t	*realnmp = NULL;
1381 	pathname_t	realnm;
1382 	int		error;
1383 	int		zflg = ZEXISTS;
1384 
1385 	ZFS_ENTER(zfsvfs);
1386 	ZFS_VERIFY_ZP(dzp);
1387 	zilog = zfsvfs->z_log;
1388 
1389 	if (flags & FIGNORECASE) {
1390 		zflg |= ZCILOOK;
1391 		pn_alloc(&realnm);
1392 		realnmp = &realnm;
1393 	}
1394 
1395 top:
1396 	/*
1397 	 * Attempt to lock directory; fail if entry doesn't exist.
1398 	 */
1399 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1400 	    NULL, realnmp)) {
1401 		if (realnmp)
1402 			pn_free(realnmp);
1403 		ZFS_EXIT(zfsvfs);
1404 		return (error);
1405 	}
1406 
1407 	vp = ZTOV(zp);
1408 
1409 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1410 		goto out;
1411 	}
1412 
1413 	/*
1414 	 * Need to use rmdir for removing directories.
1415 	 */
1416 	if (vp->v_type == VDIR) {
1417 		error = EPERM;
1418 		goto out;
1419 	}
1420 
1421 	vnevent_remove(vp, dvp, name, ct);
1422 
1423 	if (realnmp)
1424 		dnlc_remove(dvp, realnmp->pn_path);
1425 	else
1426 		dnlc_remove(dvp, name);
1427 
1428 	mutex_enter(&vp->v_lock);
1429 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1430 	mutex_exit(&vp->v_lock);
1431 
1432 	/*
1433 	 * We may delete the znode now, or we may put it in the unlinked set;
1434 	 * it depends on whether we're the last link, and on whether there are
1435 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1436 	 * allow for either case.
1437 	 */
1438 	tx = dmu_tx_create(zfsvfs->z_os);
1439 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1440 	dmu_tx_hold_bonus(tx, zp->z_id);
1441 	if (may_delete_now)
1442 		dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
1443 
1444 	/* are there any extended attributes? */
1445 	if ((xattr_obj = zp->z_phys->zp_xattr) != 0) {
1446 		/* XXX - do we need this if we are deleting? */
1447 		dmu_tx_hold_bonus(tx, xattr_obj);
1448 	}
1449 
1450 	/* are there any additional acls */
1451 	if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 &&
1452 	    may_delete_now)
1453 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1454 
1455 	/* charge as an update -- would be nice not to charge at all */
1456 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1457 
1458 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1459 	if (error) {
1460 		zfs_dirent_unlock(dl);
1461 		VN_RELE(vp);
1462 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1463 			dmu_tx_wait(tx);
1464 			dmu_tx_abort(tx);
1465 			goto top;
1466 		}
1467 		if (realnmp)
1468 			pn_free(realnmp);
1469 		dmu_tx_abort(tx);
1470 		ZFS_EXIT(zfsvfs);
1471 		return (error);
1472 	}
1473 
1474 	/*
1475 	 * Remove the directory entry.
1476 	 */
1477 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1478 
1479 	if (error) {
1480 		dmu_tx_commit(tx);
1481 		goto out;
1482 	}
1483 
1484 	if (unlinked) {
1485 		mutex_enter(&vp->v_lock);
1486 		delete_now = may_delete_now &&
1487 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1488 		    zp->z_phys->zp_xattr == xattr_obj &&
1489 		    zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj;
1490 		mutex_exit(&vp->v_lock);
1491 	}
1492 
1493 	if (delete_now) {
1494 		if (zp->z_phys->zp_xattr) {
1495 			error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
1496 			ASSERT3U(error, ==, 0);
1497 			ASSERT3U(xzp->z_phys->zp_links, ==, 2);
1498 			dmu_buf_will_dirty(xzp->z_dbuf, tx);
1499 			mutex_enter(&xzp->z_lock);
1500 			xzp->z_unlinked = 1;
1501 			xzp->z_phys->zp_links = 0;
1502 			mutex_exit(&xzp->z_lock);
1503 			zfs_unlinked_add(xzp, tx);
1504 			zp->z_phys->zp_xattr = 0; /* probably unnecessary */
1505 		}
1506 		mutex_enter(&zp->z_lock);
1507 		mutex_enter(&vp->v_lock);
1508 		vp->v_count--;
1509 		ASSERT3U(vp->v_count, ==, 0);
1510 		mutex_exit(&vp->v_lock);
1511 		mutex_exit(&zp->z_lock);
1512 		zfs_znode_delete(zp, tx);
1513 	} else if (unlinked) {
1514 		zfs_unlinked_add(zp, tx);
1515 	}
1516 
1517 	txtype = TX_REMOVE;
1518 	if (flags & FIGNORECASE)
1519 		txtype |= TX_CI;
1520 	zfs_log_remove(zilog, tx, txtype, dzp, name);
1521 
1522 	dmu_tx_commit(tx);
1523 out:
1524 	if (realnmp)
1525 		pn_free(realnmp);
1526 
1527 	zfs_dirent_unlock(dl);
1528 
1529 	if (!delete_now) {
1530 		VN_RELE(vp);
1531 	} else if (xzp) {
1532 		/* this rele delayed to prevent nesting transactions */
1533 		VN_RELE(ZTOV(xzp));
1534 	}
1535 
1536 	ZFS_EXIT(zfsvfs);
1537 	return (error);
1538 }
1539 
1540 /*
1541  * Create a new directory and insert it into dvp using the name
1542  * provided.  Return a pointer to the inserted directory.
1543  *
1544  *	IN:	dvp	- vnode of directory to add subdir to.
1545  *		dirname	- name of new directory.
1546  *		vap	- attributes of new directory.
1547  *		cr	- credentials of caller.
1548  *		ct	- caller context
1549  *		vsecp	- ACL to be set
1550  *
1551  *	OUT:	vpp	- vnode of created directory.
1552  *
1553  *	RETURN:	0 if success
1554  *		error code if failure
1555  *
1556  * Timestamps:
1557  *	dvp - ctime|mtime updated
1558  *	 vp - ctime|mtime|atime updated
1559  */
1560 /*ARGSUSED*/
1561 static int
1562 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1563     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1564 {
1565 	znode_t		*zp, *dzp = VTOZ(dvp);
1566 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1567 	zilog_t		*zilog;
1568 	zfs_dirlock_t	*dl;
1569 	uint64_t	txtype;
1570 	dmu_tx_t	*tx;
1571 	int		error;
1572 	zfs_acl_t	*aclp = NULL;
1573 	zfs_fuid_info_t	*fuidp = NULL;
1574 	int		zf = ZNEW;
1575 
1576 	ASSERT(vap->va_type == VDIR);
1577 
1578 	/*
1579 	 * If we have an ephemeral id, ACL, or XVATTR then
1580 	 * make sure file system is at proper version
1581 	 */
1582 
1583 	if (zfsvfs->z_use_fuids == B_FALSE &&
1584 	    (vsecp || (vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(crgetuid(cr))||
1585 	    IS_EPHEMERAL(crgetgid(cr))))
1586 		return (EINVAL);
1587 
1588 	ZFS_ENTER(zfsvfs);
1589 	ZFS_VERIFY_ZP(dzp);
1590 	zilog = zfsvfs->z_log;
1591 
1592 	if (dzp->z_phys->zp_flags & ZFS_XATTR) {
1593 		ZFS_EXIT(zfsvfs);
1594 		return (EINVAL);
1595 	}
1596 
1597 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1598 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1599 		ZFS_EXIT(zfsvfs);
1600 		return (EILSEQ);
1601 	}
1602 	if (flags & FIGNORECASE)
1603 		zf |= ZCILOOK;
1604 
1605 	if (vap->va_mask & AT_XVATTR)
1606 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1607 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1608 			ZFS_EXIT(zfsvfs);
1609 			return (error);
1610 		}
1611 
1612 	/*
1613 	 * First make sure the new directory doesn't exist.
1614 	 */
1615 top:
1616 	*vpp = NULL;
1617 
1618 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1619 	    NULL, NULL)) {
1620 		ZFS_EXIT(zfsvfs);
1621 		return (error);
1622 	}
1623 
1624 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1625 		zfs_dirent_unlock(dl);
1626 		ZFS_EXIT(zfsvfs);
1627 		return (error);
1628 	}
1629 
1630 	if (vsecp && aclp == NULL) {
1631 		error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp);
1632 		if (error) {
1633 			zfs_dirent_unlock(dl);
1634 			ZFS_EXIT(zfsvfs);
1635 			return (error);
1636 		}
1637 	}
1638 	/*
1639 	 * Add a new entry to the directory.
1640 	 */
1641 	tx = dmu_tx_create(zfsvfs->z_os);
1642 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1643 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1644 	if ((aclp && aclp->z_has_fuids) || IS_EPHEMERAL(crgetuid(cr)) ||
1645 	    IS_EPHEMERAL(crgetgid(cr))) {
1646 		if (zfsvfs->z_fuid_obj == 0) {
1647 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1648 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1649 			    FUID_SIZE_ESTIMATE(zfsvfs));
1650 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
1651 		} else {
1652 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
1653 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
1654 			    FUID_SIZE_ESTIMATE(zfsvfs));
1655 		}
1656 	}
1657 	if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp)
1658 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1659 		    0, SPA_MAXBLOCKSIZE);
1660 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1661 	if (error) {
1662 		zfs_dirent_unlock(dl);
1663 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1664 			dmu_tx_wait(tx);
1665 			dmu_tx_abort(tx);
1666 			goto top;
1667 		}
1668 		dmu_tx_abort(tx);
1669 		ZFS_EXIT(zfsvfs);
1670 		if (aclp)
1671 			zfs_acl_free(aclp);
1672 		return (error);
1673 	}
1674 
1675 	/*
1676 	 * Create new node.
1677 	 */
1678 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp);
1679 
1680 	if (aclp)
1681 		zfs_acl_free(aclp);
1682 
1683 	/*
1684 	 * Now put new name in parent dir.
1685 	 */
1686 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1687 
1688 	*vpp = ZTOV(zp);
1689 
1690 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1691 	if (flags & FIGNORECASE)
1692 		txtype |= TX_CI;
1693 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, fuidp, vap);
1694 
1695 	if (fuidp)
1696 		zfs_fuid_info_free(fuidp);
1697 	dmu_tx_commit(tx);
1698 
1699 	zfs_dirent_unlock(dl);
1700 
1701 	ZFS_EXIT(zfsvfs);
1702 	return (0);
1703 }
1704 
1705 /*
1706  * Remove a directory subdir entry.  If the current working
1707  * directory is the same as the subdir to be removed, the
1708  * remove will fail.
1709  *
1710  *	IN:	dvp	- vnode of directory to remove from.
1711  *		name	- name of directory to be removed.
1712  *		cwd	- vnode of current working directory.
1713  *		cr	- credentials of caller.
1714  *		ct	- caller context
1715  *		flags	- case flags
1716  *
1717  *	RETURN:	0 if success
1718  *		error code if failure
1719  *
1720  * Timestamps:
1721  *	dvp - ctime|mtime updated
1722  */
1723 /*ARGSUSED*/
1724 static int
1725 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1726     caller_context_t *ct, int flags)
1727 {
1728 	znode_t		*dzp = VTOZ(dvp);
1729 	znode_t		*zp;
1730 	vnode_t		*vp;
1731 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1732 	zilog_t		*zilog;
1733 	zfs_dirlock_t	*dl;
1734 	dmu_tx_t	*tx;
1735 	int		error;
1736 	int		zflg = ZEXISTS;
1737 
1738 	ZFS_ENTER(zfsvfs);
1739 	ZFS_VERIFY_ZP(dzp);
1740 	zilog = zfsvfs->z_log;
1741 
1742 	if (flags & FIGNORECASE)
1743 		zflg |= ZCILOOK;
1744 top:
1745 	zp = NULL;
1746 
1747 	/*
1748 	 * Attempt to lock directory; fail if entry doesn't exist.
1749 	 */
1750 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1751 	    NULL, NULL)) {
1752 		ZFS_EXIT(zfsvfs);
1753 		return (error);
1754 	}
1755 
1756 	vp = ZTOV(zp);
1757 
1758 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1759 		goto out;
1760 	}
1761 
1762 	if (vp->v_type != VDIR) {
1763 		error = ENOTDIR;
1764 		goto out;
1765 	}
1766 
1767 	if (vp == cwd) {
1768 		error = EINVAL;
1769 		goto out;
1770 	}
1771 
1772 	vnevent_rmdir(vp, dvp, name, ct);
1773 
1774 	/*
1775 	 * Grab a lock on the directory to make sure that noone is
1776 	 * trying to add (or lookup) entries while we are removing it.
1777 	 */
1778 	rw_enter(&zp->z_name_lock, RW_WRITER);
1779 
1780 	/*
1781 	 * Grab a lock on the parent pointer to make sure we play well
1782 	 * with the treewalk and directory rename code.
1783 	 */
1784 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1785 
1786 	tx = dmu_tx_create(zfsvfs->z_os);
1787 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1788 	dmu_tx_hold_bonus(tx, zp->z_id);
1789 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1790 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1791 	if (error) {
1792 		rw_exit(&zp->z_parent_lock);
1793 		rw_exit(&zp->z_name_lock);
1794 		zfs_dirent_unlock(dl);
1795 		VN_RELE(vp);
1796 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1797 			dmu_tx_wait(tx);
1798 			dmu_tx_abort(tx);
1799 			goto top;
1800 		}
1801 		dmu_tx_abort(tx);
1802 		ZFS_EXIT(zfsvfs);
1803 		return (error);
1804 	}
1805 
1806 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
1807 
1808 	if (error == 0) {
1809 		uint64_t txtype = TX_RMDIR;
1810 		if (flags & FIGNORECASE)
1811 			txtype |= TX_CI;
1812 		zfs_log_remove(zilog, tx, txtype, dzp, name);
1813 	}
1814 
1815 	dmu_tx_commit(tx);
1816 
1817 	rw_exit(&zp->z_parent_lock);
1818 	rw_exit(&zp->z_name_lock);
1819 out:
1820 	zfs_dirent_unlock(dl);
1821 
1822 	VN_RELE(vp);
1823 
1824 	ZFS_EXIT(zfsvfs);
1825 	return (error);
1826 }
1827 
1828 /*
1829  * Read as many directory entries as will fit into the provided
1830  * buffer from the given directory cursor position (specified in
1831  * the uio structure.
1832  *
1833  *	IN:	vp	- vnode of directory to read.
1834  *		uio	- structure supplying read location, range info,
1835  *			  and return buffer.
1836  *		cr	- credentials of caller.
1837  *		ct	- caller context
1838  *		flags	- case flags
1839  *
1840  *	OUT:	uio	- updated offset and range, buffer filled.
1841  *		eofp	- set to true if end-of-file detected.
1842  *
1843  *	RETURN:	0 if success
1844  *		error code if failure
1845  *
1846  * Timestamps:
1847  *	vp - atime updated
1848  *
1849  * Note that the low 4 bits of the cookie returned by zap is always zero.
1850  * This allows us to use the low range for "special" directory entries:
1851  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1852  * we use the offset 2 for the '.zfs' directory.
1853  */
1854 /* ARGSUSED */
1855 static int
1856 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
1857     caller_context_t *ct, int flags)
1858 {
1859 	znode_t		*zp = VTOZ(vp);
1860 	iovec_t		*iovp;
1861 	edirent_t	*eodp;
1862 	dirent64_t	*odp;
1863 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1864 	objset_t	*os;
1865 	caddr_t		outbuf;
1866 	size_t		bufsize;
1867 	zap_cursor_t	zc;
1868 	zap_attribute_t	zap;
1869 	uint_t		bytes_wanted;
1870 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1871 	int		local_eof;
1872 	int		outcount;
1873 	int		error;
1874 	uint8_t		prefetch;
1875 	boolean_t	check_sysattrs;
1876 
1877 	ZFS_ENTER(zfsvfs);
1878 	ZFS_VERIFY_ZP(zp);
1879 
1880 	/*
1881 	 * If we are not given an eof variable,
1882 	 * use a local one.
1883 	 */
1884 	if (eofp == NULL)
1885 		eofp = &local_eof;
1886 
1887 	/*
1888 	 * Check for valid iov_len.
1889 	 */
1890 	if (uio->uio_iov->iov_len <= 0) {
1891 		ZFS_EXIT(zfsvfs);
1892 		return (EINVAL);
1893 	}
1894 
1895 	/*
1896 	 * Quit if directory has been removed (posix)
1897 	 */
1898 	if ((*eofp = zp->z_unlinked) != 0) {
1899 		ZFS_EXIT(zfsvfs);
1900 		return (0);
1901 	}
1902 
1903 	error = 0;
1904 	os = zfsvfs->z_os;
1905 	offset = uio->uio_loffset;
1906 	prefetch = zp->z_zn_prefetch;
1907 
1908 	/*
1909 	 * Initialize the iterator cursor.
1910 	 */
1911 	if (offset <= 3) {
1912 		/*
1913 		 * Start iteration from the beginning of the directory.
1914 		 */
1915 		zap_cursor_init(&zc, os, zp->z_id);
1916 	} else {
1917 		/*
1918 		 * The offset is a serialized cursor.
1919 		 */
1920 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1921 	}
1922 
1923 	/*
1924 	 * Get space to change directory entries into fs independent format.
1925 	 */
1926 	iovp = uio->uio_iov;
1927 	bytes_wanted = iovp->iov_len;
1928 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
1929 		bufsize = bytes_wanted;
1930 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
1931 		odp = (struct dirent64 *)outbuf;
1932 	} else {
1933 		bufsize = bytes_wanted;
1934 		odp = (struct dirent64 *)iovp->iov_base;
1935 	}
1936 	eodp = (struct edirent *)odp;
1937 
1938 	/*
1939 	 * If this VFS supports system attributes; and we're looking at an
1940 	 * extended attribute directory; and we care about normalization
1941 	 * conflicts on this vfs; then we must check for normalization
1942 	 * conflicts with the sysattr name space.
1943 	 */
1944 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) &&
1945 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
1946 	    (flags & V_RDDIR_ENTFLAGS);
1947 
1948 	/*
1949 	 * Transform to file-system independent format
1950 	 */
1951 	outcount = 0;
1952 	while (outcount < bytes_wanted) {
1953 		ino64_t objnum;
1954 		ushort_t reclen;
1955 		off64_t *next;
1956 
1957 		/*
1958 		 * Special case `.', `..', and `.zfs'.
1959 		 */
1960 		if (offset == 0) {
1961 			(void) strcpy(zap.za_name, ".");
1962 			zap.za_normalization_conflict = 0;
1963 			objnum = zp->z_id;
1964 		} else if (offset == 1) {
1965 			(void) strcpy(zap.za_name, "..");
1966 			zap.za_normalization_conflict = 0;
1967 			objnum = zp->z_phys->zp_parent;
1968 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
1969 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1970 			zap.za_normalization_conflict = 0;
1971 			objnum = ZFSCTL_INO_ROOT;
1972 		} else {
1973 			/*
1974 			 * Grab next entry.
1975 			 */
1976 			if (error = zap_cursor_retrieve(&zc, &zap)) {
1977 				if ((*eofp = (error == ENOENT)) != 0)
1978 					break;
1979 				else
1980 					goto update;
1981 			}
1982 
1983 			if (zap.za_integer_length != 8 ||
1984 			    zap.za_num_integers != 1) {
1985 				cmn_err(CE_WARN, "zap_readdir: bad directory "
1986 				    "entry, obj = %lld, offset = %lld\n",
1987 				    (u_longlong_t)zp->z_id,
1988 				    (u_longlong_t)offset);
1989 				error = ENXIO;
1990 				goto update;
1991 			}
1992 
1993 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
1994 			/*
1995 			 * MacOS X can extract the object type here such as:
1996 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
1997 			 */
1998 
1999 			if (check_sysattrs && !zap.za_normalization_conflict) {
2000 				zap.za_normalization_conflict =
2001 				    xattr_sysattr_casechk(zap.za_name);
2002 			}
2003 		}
2004 
2005 		if (flags & V_RDDIR_ENTFLAGS)
2006 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2007 		else
2008 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2009 
2010 		/*
2011 		 * Will this entry fit in the buffer?
2012 		 */
2013 		if (outcount + reclen > bufsize) {
2014 			/*
2015 			 * Did we manage to fit anything in the buffer?
2016 			 */
2017 			if (!outcount) {
2018 				error = EINVAL;
2019 				goto update;
2020 			}
2021 			break;
2022 		}
2023 		if (flags & V_RDDIR_ENTFLAGS) {
2024 			/*
2025 			 * Add extended flag entry:
2026 			 */
2027 			eodp->ed_ino = objnum;
2028 			eodp->ed_reclen = reclen;
2029 			/* NOTE: ed_off is the offset for the *next* entry */
2030 			next = &(eodp->ed_off);
2031 			eodp->ed_eflags = zap.za_normalization_conflict ?
2032 			    ED_CASE_CONFLICT : 0;
2033 			(void) strncpy(eodp->ed_name, zap.za_name,
2034 			    EDIRENT_NAMELEN(reclen));
2035 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2036 		} else {
2037 			/*
2038 			 * Add normal entry:
2039 			 */
2040 			odp->d_ino = objnum;
2041 			odp->d_reclen = reclen;
2042 			/* NOTE: d_off is the offset for the *next* entry */
2043 			next = &(odp->d_off);
2044 			(void) strncpy(odp->d_name, zap.za_name,
2045 			    DIRENT64_NAMELEN(reclen));
2046 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2047 		}
2048 		outcount += reclen;
2049 
2050 		ASSERT(outcount <= bufsize);
2051 
2052 		/* Prefetch znode */
2053 		if (prefetch)
2054 			dmu_prefetch(os, objnum, 0, 0);
2055 
2056 		/*
2057 		 * Move to the next entry, fill in the previous offset.
2058 		 */
2059 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2060 			zap_cursor_advance(&zc);
2061 			offset = zap_cursor_serialize(&zc);
2062 		} else {
2063 			offset += 1;
2064 		}
2065 		*next = offset;
2066 	}
2067 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2068 
2069 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2070 		iovp->iov_base += outcount;
2071 		iovp->iov_len -= outcount;
2072 		uio->uio_resid -= outcount;
2073 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2074 		/*
2075 		 * Reset the pointer.
2076 		 */
2077 		offset = uio->uio_loffset;
2078 	}
2079 
2080 update:
2081 	zap_cursor_fini(&zc);
2082 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2083 		kmem_free(outbuf, bufsize);
2084 
2085 	if (error == ENOENT)
2086 		error = 0;
2087 
2088 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2089 
2090 	uio->uio_loffset = offset;
2091 	ZFS_EXIT(zfsvfs);
2092 	return (error);
2093 }
2094 
2095 ulong_t zfs_fsync_sync_cnt = 4;
2096 
2097 static int
2098 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2099 {
2100 	znode_t	*zp = VTOZ(vp);
2101 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2102 
2103 	/*
2104 	 * Regardless of whether this is required for standards conformance,
2105 	 * this is the logical behavior when fsync() is called on a file with
2106 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2107 	 * going to be pushed out as part of the zil_commit().
2108 	 */
2109 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2110 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2111 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2112 
2113 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2114 
2115 	ZFS_ENTER(zfsvfs);
2116 	ZFS_VERIFY_ZP(zp);
2117 	zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
2118 	ZFS_EXIT(zfsvfs);
2119 	return (0);
2120 }
2121 
2122 
2123 /*
2124  * Get the requested file attributes and place them in the provided
2125  * vattr structure.
2126  *
2127  *	IN:	vp	- vnode of file.
2128  *		vap	- va_mask identifies requested attributes.
2129  *			  If AT_XVATTR set, then optional attrs are requested
2130  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2131  *		cr	- credentials of caller.
2132  *		ct	- caller context
2133  *
2134  *	OUT:	vap	- attribute values.
2135  *
2136  *	RETURN:	0 (always succeeds)
2137  */
2138 /* ARGSUSED */
2139 static int
2140 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2141     caller_context_t *ct)
2142 {
2143 	znode_t *zp = VTOZ(vp);
2144 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2145 	znode_phys_t *pzp;
2146 	int	error = 0;
2147 	uint64_t links;
2148 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2149 	xoptattr_t *xoap = NULL;
2150 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2151 
2152 	ZFS_ENTER(zfsvfs);
2153 	ZFS_VERIFY_ZP(zp);
2154 	pzp = zp->z_phys;
2155 
2156 	mutex_enter(&zp->z_lock);
2157 
2158 	/*
2159 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2160 	 * Also, if we are the owner don't bother, since owner should
2161 	 * always be allowed to read basic attributes of file.
2162 	 */
2163 	if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL) &&
2164 	    (pzp->zp_uid != crgetuid(cr))) {
2165 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2166 		    skipaclchk, cr)) {
2167 			mutex_exit(&zp->z_lock);
2168 			ZFS_EXIT(zfsvfs);
2169 			return (error);
2170 		}
2171 	}
2172 
2173 	/*
2174 	 * Return all attributes.  It's cheaper to provide the answer
2175 	 * than to determine whether we were asked the question.
2176 	 */
2177 
2178 	vap->va_type = vp->v_type;
2179 	vap->va_mode = pzp->zp_mode & MODEMASK;
2180 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2181 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2182 	vap->va_nodeid = zp->z_id;
2183 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2184 		links = pzp->zp_links + 1;
2185 	else
2186 		links = pzp->zp_links;
2187 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2188 	vap->va_size = pzp->zp_size;
2189 	vap->va_rdev = vp->v_rdev;
2190 	vap->va_seq = zp->z_seq;
2191 
2192 	/*
2193 	 * Add in any requested optional attributes and the create time.
2194 	 * Also set the corresponding bits in the returned attribute bitmap.
2195 	 */
2196 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2197 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2198 			xoap->xoa_archive =
2199 			    ((pzp->zp_flags & ZFS_ARCHIVE) != 0);
2200 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2201 		}
2202 
2203 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2204 			xoap->xoa_readonly =
2205 			    ((pzp->zp_flags & ZFS_READONLY) != 0);
2206 			XVA_SET_RTN(xvap, XAT_READONLY);
2207 		}
2208 
2209 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2210 			xoap->xoa_system =
2211 			    ((pzp->zp_flags & ZFS_SYSTEM) != 0);
2212 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2213 		}
2214 
2215 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2216 			xoap->xoa_hidden =
2217 			    ((pzp->zp_flags & ZFS_HIDDEN) != 0);
2218 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2219 		}
2220 
2221 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2222 			xoap->xoa_nounlink =
2223 			    ((pzp->zp_flags & ZFS_NOUNLINK) != 0);
2224 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2225 		}
2226 
2227 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2228 			xoap->xoa_immutable =
2229 			    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0);
2230 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2231 		}
2232 
2233 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2234 			xoap->xoa_appendonly =
2235 			    ((pzp->zp_flags & ZFS_APPENDONLY) != 0);
2236 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2237 		}
2238 
2239 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2240 			xoap->xoa_nodump =
2241 			    ((pzp->zp_flags & ZFS_NODUMP) != 0);
2242 			XVA_SET_RTN(xvap, XAT_NODUMP);
2243 		}
2244 
2245 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2246 			xoap->xoa_opaque =
2247 			    ((pzp->zp_flags & ZFS_OPAQUE) != 0);
2248 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2249 		}
2250 
2251 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2252 			xoap->xoa_av_quarantined =
2253 			    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0);
2254 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2255 		}
2256 
2257 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2258 			xoap->xoa_av_modified =
2259 			    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0);
2260 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2261 		}
2262 
2263 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2264 		    vp->v_type == VREG &&
2265 		    (pzp->zp_flags & ZFS_BONUS_SCANSTAMP)) {
2266 			size_t len;
2267 			dmu_object_info_t doi;
2268 
2269 			/*
2270 			 * Only VREG files have anti-virus scanstamps, so we
2271 			 * won't conflict with symlinks in the bonus buffer.
2272 			 */
2273 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2274 			len = sizeof (xoap->xoa_av_scanstamp) +
2275 			    sizeof (znode_phys_t);
2276 			if (len <= doi.doi_bonus_size) {
2277 				/*
2278 				 * pzp points to the start of the
2279 				 * znode_phys_t. pzp + 1 points to the
2280 				 * first byte after the znode_phys_t.
2281 				 */
2282 				(void) memcpy(xoap->xoa_av_scanstamp,
2283 				    pzp + 1,
2284 				    sizeof (xoap->xoa_av_scanstamp));
2285 				XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
2286 			}
2287 		}
2288 
2289 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2290 			ZFS_TIME_DECODE(&xoap->xoa_createtime, pzp->zp_crtime);
2291 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2292 		}
2293 	}
2294 
2295 	ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime);
2296 	ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime);
2297 	ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime);
2298 
2299 	mutex_exit(&zp->z_lock);
2300 
2301 	dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks);
2302 
2303 	if (zp->z_blksz == 0) {
2304 		/*
2305 		 * Block size hasn't been set; suggest maximal I/O transfers.
2306 		 */
2307 		vap->va_blksize = zfsvfs->z_max_blksz;
2308 	}
2309 
2310 	ZFS_EXIT(zfsvfs);
2311 	return (0);
2312 }
2313 
2314 /*
2315  * Set the file attributes to the values contained in the
2316  * vattr structure.
2317  *
2318  *	IN:	vp	- vnode of file to be modified.
2319  *		vap	- new attribute values.
2320  *			  If AT_XVATTR set, then optional attrs are being set
2321  *		flags	- ATTR_UTIME set if non-default time values provided.
2322  *			- ATTR_NOACLCHECK (CIFS context only).
2323  *		cr	- credentials of caller.
2324  *		ct	- caller context
2325  *
2326  *	RETURN:	0 if success
2327  *		error code if failure
2328  *
2329  * Timestamps:
2330  *	vp - ctime updated, mtime updated if size changed.
2331  */
2332 /* ARGSUSED */
2333 static int
2334 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2335 	caller_context_t *ct)
2336 {
2337 	znode_t		*zp = VTOZ(vp);
2338 	znode_phys_t	*pzp;
2339 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2340 	zilog_t		*zilog;
2341 	dmu_tx_t	*tx;
2342 	vattr_t		oldva;
2343 	uint_t		mask = vap->va_mask;
2344 	uint_t		saved_mask;
2345 	int		trim_mask = 0;
2346 	uint64_t	new_mode;
2347 	znode_t		*attrzp;
2348 	int		need_policy = FALSE;
2349 	int		err;
2350 	zfs_fuid_info_t *fuidp = NULL;
2351 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2352 	xoptattr_t	*xoap;
2353 	zfs_acl_t	*aclp = NULL;
2354 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2355 
2356 	if (mask == 0)
2357 		return (0);
2358 
2359 	if (mask & AT_NOSET)
2360 		return (EINVAL);
2361 
2362 	ZFS_ENTER(zfsvfs);
2363 	ZFS_VERIFY_ZP(zp);
2364 
2365 	pzp = zp->z_phys;
2366 	zilog = zfsvfs->z_log;
2367 
2368 	/*
2369 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2370 	 * that file system is at proper version level
2371 	 */
2372 
2373 	if (zfsvfs->z_use_fuids == B_FALSE &&
2374 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2375 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2376 	    (mask & AT_XVATTR))) {
2377 		ZFS_EXIT(zfsvfs);
2378 		return (EINVAL);
2379 	}
2380 
2381 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2382 		ZFS_EXIT(zfsvfs);
2383 		return (EISDIR);
2384 	}
2385 
2386 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2387 		ZFS_EXIT(zfsvfs);
2388 		return (EINVAL);
2389 	}
2390 
2391 	/*
2392 	 * If this is an xvattr_t, then get a pointer to the structure of
2393 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2394 	 */
2395 	xoap = xva_getxoptattr(xvap);
2396 
2397 	/*
2398 	 * Immutable files can only alter immutable bit and atime
2399 	 */
2400 	if ((pzp->zp_flags & ZFS_IMMUTABLE) &&
2401 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2402 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2403 		ZFS_EXIT(zfsvfs);
2404 		return (EPERM);
2405 	}
2406 
2407 	if ((mask & AT_SIZE) && (pzp->zp_flags & ZFS_READONLY)) {
2408 		ZFS_EXIT(zfsvfs);
2409 		return (EPERM);
2410 	}
2411 
2412 top:
2413 	attrzp = NULL;
2414 
2415 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2416 		ZFS_EXIT(zfsvfs);
2417 		return (EROFS);
2418 	}
2419 
2420 	/*
2421 	 * First validate permissions
2422 	 */
2423 
2424 	if (mask & AT_SIZE) {
2425 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2426 		if (err) {
2427 			ZFS_EXIT(zfsvfs);
2428 			return (err);
2429 		}
2430 		/*
2431 		 * XXX - Note, we are not providing any open
2432 		 * mode flags here (like FNDELAY), so we may
2433 		 * block if there are locks present... this
2434 		 * should be addressed in openat().
2435 		 */
2436 		do {
2437 			err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2438 			/* NB: we already did dmu_tx_wait() if necessary */
2439 		} while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
2440 		if (err) {
2441 			ZFS_EXIT(zfsvfs);
2442 			return (err);
2443 		}
2444 	}
2445 
2446 	if (mask & (AT_ATIME|AT_MTIME) ||
2447 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2448 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2449 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2450 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2451 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM))))
2452 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2453 		    skipaclchk, cr);
2454 
2455 	if (mask & (AT_UID|AT_GID)) {
2456 		int	idmask = (mask & (AT_UID|AT_GID));
2457 		int	take_owner;
2458 		int	take_group;
2459 
2460 		/*
2461 		 * NOTE: even if a new mode is being set,
2462 		 * we may clear S_ISUID/S_ISGID bits.
2463 		 */
2464 
2465 		if (!(mask & AT_MODE))
2466 			vap->va_mode = pzp->zp_mode;
2467 
2468 		/*
2469 		 * Take ownership or chgrp to group we are a member of
2470 		 */
2471 
2472 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2473 		take_group = (mask & AT_GID) &&
2474 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2475 
2476 		/*
2477 		 * If both AT_UID and AT_GID are set then take_owner and
2478 		 * take_group must both be set in order to allow taking
2479 		 * ownership.
2480 		 *
2481 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2482 		 *
2483 		 */
2484 
2485 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2486 		    ((idmask == AT_UID) && take_owner) ||
2487 		    ((idmask == AT_GID) && take_group)) {
2488 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2489 			    skipaclchk, cr) == 0) {
2490 				/*
2491 				 * Remove setuid/setgid for non-privileged users
2492 				 */
2493 				secpolicy_setid_clear(vap, cr);
2494 				trim_mask = (mask & (AT_UID|AT_GID));
2495 			} else {
2496 				need_policy =  TRUE;
2497 			}
2498 		} else {
2499 			need_policy =  TRUE;
2500 		}
2501 	}
2502 
2503 	mutex_enter(&zp->z_lock);
2504 	oldva.va_mode = pzp->zp_mode;
2505 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2506 	if (mask & AT_XVATTR) {
2507 		if ((need_policy == FALSE) &&
2508 		    (XVA_ISSET_REQ(xvap, XAT_APPENDONLY) &&
2509 		    xoap->xoa_appendonly !=
2510 		    ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) ||
2511 		    (XVA_ISSET_REQ(xvap, XAT_NOUNLINK) &&
2512 		    xoap->xoa_nounlink !=
2513 		    ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) ||
2514 		    (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE) &&
2515 		    xoap->xoa_immutable !=
2516 		    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) ||
2517 		    (XVA_ISSET_REQ(xvap, XAT_NODUMP) &&
2518 		    xoap->xoa_nodump !=
2519 		    ((pzp->zp_flags & ZFS_NODUMP) != 0)) ||
2520 		    (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED) &&
2521 		    xoap->xoa_av_modified !=
2522 		    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) ||
2523 		    ((XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED) &&
2524 		    ((vp->v_type != VREG && xoap->xoa_av_quarantined) ||
2525 		    xoap->xoa_av_quarantined !=
2526 		    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)))) ||
2527 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
2528 		    (XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2529 			need_policy = TRUE;
2530 		}
2531 	}
2532 
2533 	mutex_exit(&zp->z_lock);
2534 
2535 	if (mask & AT_MODE) {
2536 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2537 			err = secpolicy_setid_setsticky_clear(vp, vap,
2538 			    &oldva, cr);
2539 			if (err) {
2540 				ZFS_EXIT(zfsvfs);
2541 				return (err);
2542 			}
2543 			trim_mask |= AT_MODE;
2544 		} else {
2545 			need_policy = TRUE;
2546 		}
2547 	}
2548 
2549 	if (need_policy) {
2550 		/*
2551 		 * If trim_mask is set then take ownership
2552 		 * has been granted or write_acl is present and user
2553 		 * has the ability to modify mode.  In that case remove
2554 		 * UID|GID and or MODE from mask so that
2555 		 * secpolicy_vnode_setattr() doesn't revoke it.
2556 		 */
2557 
2558 		if (trim_mask) {
2559 			saved_mask = vap->va_mask;
2560 			vap->va_mask &= ~trim_mask;
2561 		}
2562 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2563 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2564 		if (err) {
2565 			ZFS_EXIT(zfsvfs);
2566 			return (err);
2567 		}
2568 
2569 		if (trim_mask)
2570 			vap->va_mask |= saved_mask;
2571 	}
2572 
2573 	/*
2574 	 * secpolicy_vnode_setattr, or take ownership may have
2575 	 * changed va_mask
2576 	 */
2577 	mask = vap->va_mask;
2578 
2579 	tx = dmu_tx_create(zfsvfs->z_os);
2580 	dmu_tx_hold_bonus(tx, zp->z_id);
2581 	if (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2582 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid))) {
2583 		if (zfsvfs->z_fuid_obj == 0) {
2584 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
2585 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2586 			    FUID_SIZE_ESTIMATE(zfsvfs));
2587 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
2588 		} else {
2589 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
2590 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
2591 			    FUID_SIZE_ESTIMATE(zfsvfs));
2592 		}
2593 	}
2594 
2595 	if (mask & AT_MODE) {
2596 		uint64_t pmode = pzp->zp_mode;
2597 
2598 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2599 
2600 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) {
2601 			dmu_tx_abort(tx);
2602 			ZFS_EXIT(zfsvfs);
2603 			return (err);
2604 		}
2605 		if (pzp->zp_acl.z_acl_extern_obj) {
2606 			/* Are we upgrading ACL from old V0 format to new V1 */
2607 			if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2608 			    pzp->zp_acl.z_acl_version ==
2609 			    ZFS_ACL_VERSION_INITIAL) {
2610 				dmu_tx_hold_free(tx,
2611 				    pzp->zp_acl.z_acl_extern_obj, 0,
2612 				    DMU_OBJECT_END);
2613 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2614 				    0, aclp->z_acl_bytes);
2615 			} else {
2616 				dmu_tx_hold_write(tx,
2617 				    pzp->zp_acl.z_acl_extern_obj, 0,
2618 				    aclp->z_acl_bytes);
2619 			}
2620 		} else if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL)) {
2621 			if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2622 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2623 				    0, aclp->z_acl_bytes);
2624 			}
2625 		}
2626 	}
2627 
2628 	if ((mask & (AT_UID | AT_GID)) && pzp->zp_xattr != 0) {
2629 		err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp);
2630 		if (err) {
2631 			dmu_tx_abort(tx);
2632 			ZFS_EXIT(zfsvfs);
2633 			if (aclp)
2634 				zfs_acl_free(aclp);
2635 			return (err);
2636 		}
2637 		dmu_tx_hold_bonus(tx, attrzp->z_id);
2638 	}
2639 
2640 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2641 	if (err) {
2642 		if (attrzp)
2643 			VN_RELE(ZTOV(attrzp));
2644 
2645 		if (aclp) {
2646 			zfs_acl_free(aclp);
2647 			aclp = NULL;
2648 		}
2649 
2650 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2651 			dmu_tx_wait(tx);
2652 			dmu_tx_abort(tx);
2653 			goto top;
2654 		}
2655 		dmu_tx_abort(tx);
2656 		ZFS_EXIT(zfsvfs);
2657 		return (err);
2658 	}
2659 
2660 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2661 
2662 	/*
2663 	 * Set each attribute requested.
2664 	 * We group settings according to the locks they need to acquire.
2665 	 *
2666 	 * Note: you cannot set ctime directly, although it will be
2667 	 * updated as a side-effect of calling this function.
2668 	 */
2669 
2670 	mutex_enter(&zp->z_lock);
2671 
2672 	if (mask & AT_MODE) {
2673 		mutex_enter(&zp->z_acl_lock);
2674 		zp->z_phys->zp_mode = new_mode;
2675 		err = zfs_aclset_common(zp, aclp, cr, &fuidp, tx);
2676 		ASSERT3U(err, ==, 0);
2677 		mutex_exit(&zp->z_acl_lock);
2678 	}
2679 
2680 	if (attrzp)
2681 		mutex_enter(&attrzp->z_lock);
2682 
2683 	if (mask & AT_UID) {
2684 		pzp->zp_uid = zfs_fuid_create(zfsvfs,
2685 		    vap->va_uid, cr, ZFS_OWNER, tx, &fuidp);
2686 		if (attrzp) {
2687 			attrzp->z_phys->zp_uid = zfs_fuid_create(zfsvfs,
2688 			    vap->va_uid,  cr, ZFS_OWNER, tx, &fuidp);
2689 		}
2690 	}
2691 
2692 	if (mask & AT_GID) {
2693 		pzp->zp_gid = zfs_fuid_create(zfsvfs, vap->va_gid,
2694 		    cr, ZFS_GROUP, tx, &fuidp);
2695 		if (attrzp)
2696 			attrzp->z_phys->zp_gid = zfs_fuid_create(zfsvfs,
2697 			    vap->va_gid, cr, ZFS_GROUP, tx, &fuidp);
2698 	}
2699 
2700 	if (aclp)
2701 		zfs_acl_free(aclp);
2702 
2703 	if (attrzp)
2704 		mutex_exit(&attrzp->z_lock);
2705 
2706 	if (mask & AT_ATIME)
2707 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2708 
2709 	if (mask & AT_MTIME)
2710 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2711 
2712 	if (mask & AT_SIZE)
2713 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2714 	else if (mask != 0)
2715 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2716 	/*
2717 	 * Do this after setting timestamps to prevent timestamp
2718 	 * update from toggling bit
2719 	 */
2720 
2721 	if (xoap && (mask & AT_XVATTR)) {
2722 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
2723 			size_t len;
2724 			dmu_object_info_t doi;
2725 
2726 			ASSERT(vp->v_type == VREG);
2727 
2728 			/* Grow the bonus buffer if necessary. */
2729 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2730 			len = sizeof (xoap->xoa_av_scanstamp) +
2731 			    sizeof (znode_phys_t);
2732 			if (len > doi.doi_bonus_size)
2733 				VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0);
2734 		}
2735 		zfs_xvattr_set(zp, xvap);
2736 	}
2737 
2738 	if (mask != 0)
2739 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2740 
2741 	if (fuidp)
2742 		zfs_fuid_info_free(fuidp);
2743 	mutex_exit(&zp->z_lock);
2744 
2745 	if (attrzp)
2746 		VN_RELE(ZTOV(attrzp));
2747 
2748 	dmu_tx_commit(tx);
2749 
2750 	ZFS_EXIT(zfsvfs);
2751 	return (err);
2752 }
2753 
2754 typedef struct zfs_zlock {
2755 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2756 	znode_t		*zl_znode;	/* znode we held */
2757 	struct zfs_zlock *zl_next;	/* next in list */
2758 } zfs_zlock_t;
2759 
2760 /*
2761  * Drop locks and release vnodes that were held by zfs_rename_lock().
2762  */
2763 static void
2764 zfs_rename_unlock(zfs_zlock_t **zlpp)
2765 {
2766 	zfs_zlock_t *zl;
2767 
2768 	while ((zl = *zlpp) != NULL) {
2769 		if (zl->zl_znode != NULL)
2770 			VN_RELE(ZTOV(zl->zl_znode));
2771 		rw_exit(zl->zl_rwlock);
2772 		*zlpp = zl->zl_next;
2773 		kmem_free(zl, sizeof (*zl));
2774 	}
2775 }
2776 
2777 /*
2778  * Search back through the directory tree, using the ".." entries.
2779  * Lock each directory in the chain to prevent concurrent renames.
2780  * Fail any attempt to move a directory into one of its own descendants.
2781  * XXX - z_parent_lock can overlap with map or grow locks
2782  */
2783 static int
2784 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2785 {
2786 	zfs_zlock_t	*zl;
2787 	znode_t		*zp = tdzp;
2788 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2789 	uint64_t	*oidp = &zp->z_id;
2790 	krwlock_t	*rwlp = &szp->z_parent_lock;
2791 	krw_t		rw = RW_WRITER;
2792 
2793 	/*
2794 	 * First pass write-locks szp and compares to zp->z_id.
2795 	 * Later passes read-lock zp and compare to zp->z_parent.
2796 	 */
2797 	do {
2798 		if (!rw_tryenter(rwlp, rw)) {
2799 			/*
2800 			 * Another thread is renaming in this path.
2801 			 * Note that if we are a WRITER, we don't have any
2802 			 * parent_locks held yet.
2803 			 */
2804 			if (rw == RW_READER && zp->z_id > szp->z_id) {
2805 				/*
2806 				 * Drop our locks and restart
2807 				 */
2808 				zfs_rename_unlock(&zl);
2809 				*zlpp = NULL;
2810 				zp = tdzp;
2811 				oidp = &zp->z_id;
2812 				rwlp = &szp->z_parent_lock;
2813 				rw = RW_WRITER;
2814 				continue;
2815 			} else {
2816 				/*
2817 				 * Wait for other thread to drop its locks
2818 				 */
2819 				rw_enter(rwlp, rw);
2820 			}
2821 		}
2822 
2823 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2824 		zl->zl_rwlock = rwlp;
2825 		zl->zl_znode = NULL;
2826 		zl->zl_next = *zlpp;
2827 		*zlpp = zl;
2828 
2829 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2830 			return (EINVAL);
2831 
2832 		if (*oidp == rootid)		/* We've hit the top */
2833 			return (0);
2834 
2835 		if (rw == RW_READER) {		/* i.e. not the first pass */
2836 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2837 			if (error)
2838 				return (error);
2839 			zl->zl_znode = zp;
2840 		}
2841 		oidp = &zp->z_phys->zp_parent;
2842 		rwlp = &zp->z_parent_lock;
2843 		rw = RW_READER;
2844 
2845 	} while (zp->z_id != sdzp->z_id);
2846 
2847 	return (0);
2848 }
2849 
2850 /*
2851  * Move an entry from the provided source directory to the target
2852  * directory.  Change the entry name as indicated.
2853  *
2854  *	IN:	sdvp	- Source directory containing the "old entry".
2855  *		snm	- Old entry name.
2856  *		tdvp	- Target directory to contain the "new entry".
2857  *		tnm	- New entry name.
2858  *		cr	- credentials of caller.
2859  *		ct	- caller context
2860  *		flags	- case flags
2861  *
2862  *	RETURN:	0 if success
2863  *		error code if failure
2864  *
2865  * Timestamps:
2866  *	sdvp,tdvp - ctime|mtime updated
2867  */
2868 /*ARGSUSED*/
2869 static int
2870 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
2871     caller_context_t *ct, int flags)
2872 {
2873 	znode_t		*tdzp, *szp, *tzp;
2874 	znode_t		*sdzp = VTOZ(sdvp);
2875 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2876 	zilog_t		*zilog;
2877 	vnode_t		*realvp;
2878 	zfs_dirlock_t	*sdl, *tdl;
2879 	dmu_tx_t	*tx;
2880 	zfs_zlock_t	*zl;
2881 	int		cmp, serr, terr;
2882 	int		error = 0;
2883 	int		zflg = 0;
2884 
2885 	ZFS_ENTER(zfsvfs);
2886 	ZFS_VERIFY_ZP(sdzp);
2887 	zilog = zfsvfs->z_log;
2888 
2889 	/*
2890 	 * Make sure we have the real vp for the target directory.
2891 	 */
2892 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
2893 		tdvp = realvp;
2894 
2895 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
2896 		ZFS_EXIT(zfsvfs);
2897 		return (EXDEV);
2898 	}
2899 
2900 	tdzp = VTOZ(tdvp);
2901 	ZFS_VERIFY_ZP(tdzp);
2902 	if (zfsvfs->z_utf8 && u8_validate(tnm,
2903 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2904 		ZFS_EXIT(zfsvfs);
2905 		return (EILSEQ);
2906 	}
2907 
2908 	if (flags & FIGNORECASE)
2909 		zflg |= ZCILOOK;
2910 
2911 top:
2912 	szp = NULL;
2913 	tzp = NULL;
2914 	zl = NULL;
2915 
2916 	/*
2917 	 * This is to prevent the creation of links into attribute space
2918 	 * by renaming a linked file into/outof an attribute directory.
2919 	 * See the comment in zfs_link() for why this is considered bad.
2920 	 */
2921 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
2922 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
2923 		ZFS_EXIT(zfsvfs);
2924 		return (EINVAL);
2925 	}
2926 
2927 	/*
2928 	 * Lock source and target directory entries.  To prevent deadlock,
2929 	 * a lock ordering must be defined.  We lock the directory with
2930 	 * the smallest object id first, or if it's a tie, the one with
2931 	 * the lexically first name.
2932 	 */
2933 	if (sdzp->z_id < tdzp->z_id) {
2934 		cmp = -1;
2935 	} else if (sdzp->z_id > tdzp->z_id) {
2936 		cmp = 1;
2937 	} else {
2938 		/*
2939 		 * First compare the two name arguments without
2940 		 * considering any case folding.
2941 		 */
2942 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
2943 
2944 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
2945 		ASSERT(error == 0 || !zfsvfs->z_utf8);
2946 		if (cmp == 0) {
2947 			/*
2948 			 * POSIX: "If the old argument and the new argument
2949 			 * both refer to links to the same existing file,
2950 			 * the rename() function shall return successfully
2951 			 * and perform no other action."
2952 			 */
2953 			ZFS_EXIT(zfsvfs);
2954 			return (0);
2955 		}
2956 		/*
2957 		 * If the file system is case-folding, then we may
2958 		 * have some more checking to do.  A case-folding file
2959 		 * system is either supporting mixed case sensitivity
2960 		 * access or is completely case-insensitive.  Note
2961 		 * that the file system is always case preserving.
2962 		 *
2963 		 * In mixed sensitivity mode case sensitive behavior
2964 		 * is the default.  FIGNORECASE must be used to
2965 		 * explicitly request case insensitive behavior.
2966 		 *
2967 		 * If the source and target names provided differ only
2968 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
2969 		 * we will treat this as a special case in the
2970 		 * case-insensitive mode: as long as the source name
2971 		 * is an exact match, we will allow this to proceed as
2972 		 * a name-change request.
2973 		 */
2974 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
2975 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
2976 		    flags & FIGNORECASE)) &&
2977 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
2978 		    &error) == 0) {
2979 			/*
2980 			 * case preserving rename request, require exact
2981 			 * name matches
2982 			 */
2983 			zflg |= ZCIEXACT;
2984 			zflg &= ~ZCILOOK;
2985 		}
2986 	}
2987 
2988 	if (cmp < 0) {
2989 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
2990 		    ZEXISTS | zflg, NULL, NULL);
2991 		terr = zfs_dirent_lock(&tdl,
2992 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
2993 	} else {
2994 		terr = zfs_dirent_lock(&tdl,
2995 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
2996 		serr = zfs_dirent_lock(&sdl,
2997 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
2998 		    NULL, NULL);
2999 	}
3000 
3001 	if (serr) {
3002 		/*
3003 		 * Source entry invalid or not there.
3004 		 */
3005 		if (!terr) {
3006 			zfs_dirent_unlock(tdl);
3007 			if (tzp)
3008 				VN_RELE(ZTOV(tzp));
3009 		}
3010 		if (strcmp(snm, "..") == 0)
3011 			serr = EINVAL;
3012 		ZFS_EXIT(zfsvfs);
3013 		return (serr);
3014 	}
3015 	if (terr) {
3016 		zfs_dirent_unlock(sdl);
3017 		VN_RELE(ZTOV(szp));
3018 		if (strcmp(tnm, "..") == 0)
3019 			terr = EINVAL;
3020 		ZFS_EXIT(zfsvfs);
3021 		return (terr);
3022 	}
3023 
3024 	/*
3025 	 * Must have write access at the source to remove the old entry
3026 	 * and write access at the target to create the new entry.
3027 	 * Note that if target and source are the same, this can be
3028 	 * done in a single check.
3029 	 */
3030 
3031 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3032 		goto out;
3033 
3034 	if (ZTOV(szp)->v_type == VDIR) {
3035 		/*
3036 		 * Check to make sure rename is valid.
3037 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3038 		 */
3039 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3040 			goto out;
3041 	}
3042 
3043 	/*
3044 	 * Does target exist?
3045 	 */
3046 	if (tzp) {
3047 		/*
3048 		 * Source and target must be the same type.
3049 		 */
3050 		if (ZTOV(szp)->v_type == VDIR) {
3051 			if (ZTOV(tzp)->v_type != VDIR) {
3052 				error = ENOTDIR;
3053 				goto out;
3054 			}
3055 		} else {
3056 			if (ZTOV(tzp)->v_type == VDIR) {
3057 				error = EISDIR;
3058 				goto out;
3059 			}
3060 		}
3061 		/*
3062 		 * POSIX dictates that when the source and target
3063 		 * entries refer to the same file object, rename
3064 		 * must do nothing and exit without error.
3065 		 */
3066 		if (szp->z_id == tzp->z_id) {
3067 			error = 0;
3068 			goto out;
3069 		}
3070 	}
3071 
3072 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3073 	if (tzp)
3074 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3075 
3076 	/*
3077 	 * notify the target directory if it is not the same
3078 	 * as source directory.
3079 	 */
3080 	if (tdvp != sdvp) {
3081 		vnevent_rename_dest_dir(tdvp, ct);
3082 	}
3083 
3084 	tx = dmu_tx_create(zfsvfs->z_os);
3085 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
3086 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
3087 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3088 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3089 	if (sdzp != tdzp)
3090 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
3091 	if (tzp)
3092 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* parent changes */
3093 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3094 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3095 	if (error) {
3096 		if (zl != NULL)
3097 			zfs_rename_unlock(&zl);
3098 		zfs_dirent_unlock(sdl);
3099 		zfs_dirent_unlock(tdl);
3100 		VN_RELE(ZTOV(szp));
3101 		if (tzp)
3102 			VN_RELE(ZTOV(tzp));
3103 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3104 			dmu_tx_wait(tx);
3105 			dmu_tx_abort(tx);
3106 			goto top;
3107 		}
3108 		dmu_tx_abort(tx);
3109 		ZFS_EXIT(zfsvfs);
3110 		return (error);
3111 	}
3112 
3113 	if (tzp)	/* Attempt to remove the existing target */
3114 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3115 
3116 	if (error == 0) {
3117 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3118 		if (error == 0) {
3119 			szp->z_phys->zp_flags |= ZFS_AV_MODIFIED;
3120 
3121 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3122 			ASSERT(error == 0);
3123 
3124 			zfs_log_rename(zilog, tx,
3125 			    TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0),
3126 			    sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp);
3127 		}
3128 	}
3129 
3130 	dmu_tx_commit(tx);
3131 out:
3132 	if (zl != NULL)
3133 		zfs_rename_unlock(&zl);
3134 
3135 	zfs_dirent_unlock(sdl);
3136 	zfs_dirent_unlock(tdl);
3137 
3138 	VN_RELE(ZTOV(szp));
3139 	if (tzp)
3140 		VN_RELE(ZTOV(tzp));
3141 
3142 	ZFS_EXIT(zfsvfs);
3143 	return (error);
3144 }
3145 
3146 /*
3147  * Insert the indicated symbolic reference entry into the directory.
3148  *
3149  *	IN:	dvp	- Directory to contain new symbolic link.
3150  *		link	- Name for new symlink entry.
3151  *		vap	- Attributes of new entry.
3152  *		target	- Target path of new symlink.
3153  *		cr	- credentials of caller.
3154  *		ct	- caller context
3155  *		flags	- case flags
3156  *
3157  *	RETURN:	0 if success
3158  *		error code if failure
3159  *
3160  * Timestamps:
3161  *	dvp - ctime|mtime updated
3162  */
3163 /*ARGSUSED*/
3164 static int
3165 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3166     caller_context_t *ct, int flags)
3167 {
3168 	znode_t		*zp, *dzp = VTOZ(dvp);
3169 	zfs_dirlock_t	*dl;
3170 	dmu_tx_t	*tx;
3171 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3172 	zilog_t		*zilog;
3173 	int		len = strlen(link);
3174 	int		error;
3175 	int		zflg = ZNEW;
3176 	zfs_fuid_info_t *fuidp = NULL;
3177 
3178 	ASSERT(vap->va_type == VLNK);
3179 
3180 	ZFS_ENTER(zfsvfs);
3181 	ZFS_VERIFY_ZP(dzp);
3182 	zilog = zfsvfs->z_log;
3183 
3184 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3185 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3186 		ZFS_EXIT(zfsvfs);
3187 		return (EILSEQ);
3188 	}
3189 	if (flags & FIGNORECASE)
3190 		zflg |= ZCILOOK;
3191 top:
3192 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3193 		ZFS_EXIT(zfsvfs);
3194 		return (error);
3195 	}
3196 
3197 	if (len > MAXPATHLEN) {
3198 		ZFS_EXIT(zfsvfs);
3199 		return (ENAMETOOLONG);
3200 	}
3201 
3202 	/*
3203 	 * Attempt to lock directory; fail if entry already exists.
3204 	 */
3205 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3206 	if (error) {
3207 		ZFS_EXIT(zfsvfs);
3208 		return (error);
3209 	}
3210 
3211 	tx = dmu_tx_create(zfsvfs->z_os);
3212 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3213 	dmu_tx_hold_bonus(tx, dzp->z_id);
3214 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3215 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
3216 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
3217 	if (IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))) {
3218 		if (zfsvfs->z_fuid_obj == 0) {
3219 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
3220 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3221 			    FUID_SIZE_ESTIMATE(zfsvfs));
3222 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
3223 		} else {
3224 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
3225 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
3226 			    FUID_SIZE_ESTIMATE(zfsvfs));
3227 		}
3228 	}
3229 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3230 	if (error) {
3231 		zfs_dirent_unlock(dl);
3232 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3233 			dmu_tx_wait(tx);
3234 			dmu_tx_abort(tx);
3235 			goto top;
3236 		}
3237 		dmu_tx_abort(tx);
3238 		ZFS_EXIT(zfsvfs);
3239 		return (error);
3240 	}
3241 
3242 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
3243 
3244 	/*
3245 	 * Create a new object for the symlink.
3246 	 * Put the link content into bonus buffer if it will fit;
3247 	 * otherwise, store it just like any other file data.
3248 	 */
3249 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
3250 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, len, NULL, &fuidp);
3251 		if (len != 0)
3252 			bcopy(link, zp->z_phys + 1, len);
3253 	} else {
3254 		dmu_buf_t *dbp;
3255 
3256 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, NULL, &fuidp);
3257 		/*
3258 		 * Nothing can access the znode yet so no locking needed
3259 		 * for growing the znode's blocksize.
3260 		 */
3261 		zfs_grow_blocksize(zp, len, tx);
3262 
3263 		VERIFY(0 == dmu_buf_hold(zfsvfs->z_os,
3264 		    zp->z_id, 0, FTAG, &dbp));
3265 		dmu_buf_will_dirty(dbp, tx);
3266 
3267 		ASSERT3U(len, <=, dbp->db_size);
3268 		bcopy(link, dbp->db_data, len);
3269 		dmu_buf_rele(dbp, FTAG);
3270 	}
3271 	zp->z_phys->zp_size = len;
3272 
3273 	/*
3274 	 * Insert the new object into the directory.
3275 	 */
3276 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3277 out:
3278 	if (error == 0) {
3279 		uint64_t txtype = TX_SYMLINK;
3280 		if (flags & FIGNORECASE)
3281 			txtype |= TX_CI;
3282 		zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3283 	}
3284 	if (fuidp)
3285 		zfs_fuid_info_free(fuidp);
3286 
3287 	dmu_tx_commit(tx);
3288 
3289 	zfs_dirent_unlock(dl);
3290 
3291 	VN_RELE(ZTOV(zp));
3292 
3293 	ZFS_EXIT(zfsvfs);
3294 	return (error);
3295 }
3296 
3297 /*
3298  * Return, in the buffer contained in the provided uio structure,
3299  * the symbolic path referred to by vp.
3300  *
3301  *	IN:	vp	- vnode of symbolic link.
3302  *		uoip	- structure to contain the link path.
3303  *		cr	- credentials of caller.
3304  *		ct	- caller context
3305  *
3306  *	OUT:	uio	- structure to contain the link path.
3307  *
3308  *	RETURN:	0 if success
3309  *		error code if failure
3310  *
3311  * Timestamps:
3312  *	vp - atime updated
3313  */
3314 /* ARGSUSED */
3315 static int
3316 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3317 {
3318 	znode_t		*zp = VTOZ(vp);
3319 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3320 	size_t		bufsz;
3321 	int		error;
3322 
3323 	ZFS_ENTER(zfsvfs);
3324 	ZFS_VERIFY_ZP(zp);
3325 
3326 	bufsz = (size_t)zp->z_phys->zp_size;
3327 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
3328 		error = uiomove(zp->z_phys + 1,
3329 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3330 	} else {
3331 		dmu_buf_t *dbp;
3332 		error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp);
3333 		if (error) {
3334 			ZFS_EXIT(zfsvfs);
3335 			return (error);
3336 		}
3337 		error = uiomove(dbp->db_data,
3338 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3339 		dmu_buf_rele(dbp, FTAG);
3340 	}
3341 
3342 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3343 	ZFS_EXIT(zfsvfs);
3344 	return (error);
3345 }
3346 
3347 /*
3348  * Insert a new entry into directory tdvp referencing svp.
3349  *
3350  *	IN:	tdvp	- Directory to contain new entry.
3351  *		svp	- vnode of new entry.
3352  *		name	- name of new entry.
3353  *		cr	- credentials of caller.
3354  *		ct	- caller context
3355  *
3356  *	RETURN:	0 if success
3357  *		error code if failure
3358  *
3359  * Timestamps:
3360  *	tdvp - ctime|mtime updated
3361  *	 svp - ctime updated
3362  */
3363 /* ARGSUSED */
3364 static int
3365 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3366     caller_context_t *ct, int flags)
3367 {
3368 	znode_t		*dzp = VTOZ(tdvp);
3369 	znode_t		*tzp, *szp;
3370 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3371 	zilog_t		*zilog;
3372 	zfs_dirlock_t	*dl;
3373 	dmu_tx_t	*tx;
3374 	vnode_t		*realvp;
3375 	int		error;
3376 	int		zf = ZNEW;
3377 	uid_t		owner;
3378 
3379 	ASSERT(tdvp->v_type == VDIR);
3380 
3381 	ZFS_ENTER(zfsvfs);
3382 	ZFS_VERIFY_ZP(dzp);
3383 	zilog = zfsvfs->z_log;
3384 
3385 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3386 		svp = realvp;
3387 
3388 	if (svp->v_vfsp != tdvp->v_vfsp) {
3389 		ZFS_EXIT(zfsvfs);
3390 		return (EXDEV);
3391 	}
3392 	szp = VTOZ(svp);
3393 	ZFS_VERIFY_ZP(szp);
3394 
3395 	if (zfsvfs->z_utf8 && u8_validate(name,
3396 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3397 		ZFS_EXIT(zfsvfs);
3398 		return (EILSEQ);
3399 	}
3400 	if (flags & FIGNORECASE)
3401 		zf |= ZCILOOK;
3402 
3403 top:
3404 	/*
3405 	 * We do not support links between attributes and non-attributes
3406 	 * because of the potential security risk of creating links
3407 	 * into "normal" file space in order to circumvent restrictions
3408 	 * imposed in attribute space.
3409 	 */
3410 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
3411 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
3412 		ZFS_EXIT(zfsvfs);
3413 		return (EINVAL);
3414 	}
3415 
3416 	/*
3417 	 * POSIX dictates that we return EPERM here.
3418 	 * Better choices include ENOTSUP or EISDIR.
3419 	 */
3420 	if (svp->v_type == VDIR) {
3421 		ZFS_EXIT(zfsvfs);
3422 		return (EPERM);
3423 	}
3424 
3425 	owner = zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, cr, ZFS_OWNER);
3426 	if (owner != crgetuid(cr) &&
3427 	    secpolicy_basic_link(cr) != 0) {
3428 		ZFS_EXIT(zfsvfs);
3429 		return (EPERM);
3430 	}
3431 
3432 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3433 		ZFS_EXIT(zfsvfs);
3434 		return (error);
3435 	}
3436 
3437 	/*
3438 	 * Attempt to lock directory; fail if entry already exists.
3439 	 */
3440 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3441 	if (error) {
3442 		ZFS_EXIT(zfsvfs);
3443 		return (error);
3444 	}
3445 
3446 	tx = dmu_tx_create(zfsvfs->z_os);
3447 	dmu_tx_hold_bonus(tx, szp->z_id);
3448 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3449 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3450 	if (error) {
3451 		zfs_dirent_unlock(dl);
3452 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3453 			dmu_tx_wait(tx);
3454 			dmu_tx_abort(tx);
3455 			goto top;
3456 		}
3457 		dmu_tx_abort(tx);
3458 		ZFS_EXIT(zfsvfs);
3459 		return (error);
3460 	}
3461 
3462 	error = zfs_link_create(dl, szp, tx, 0);
3463 
3464 	if (error == 0) {
3465 		uint64_t txtype = TX_LINK;
3466 		if (flags & FIGNORECASE)
3467 			txtype |= TX_CI;
3468 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3469 	}
3470 
3471 	dmu_tx_commit(tx);
3472 
3473 	zfs_dirent_unlock(dl);
3474 
3475 	if (error == 0) {
3476 		vnevent_link(svp, ct);
3477 	}
3478 
3479 	ZFS_EXIT(zfsvfs);
3480 	return (error);
3481 }
3482 
3483 /*
3484  * zfs_null_putapage() is used when the file system has been force
3485  * unmounted. It just drops the pages.
3486  */
3487 /* ARGSUSED */
3488 static int
3489 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3490 		size_t *lenp, int flags, cred_t *cr)
3491 {
3492 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
3493 	return (0);
3494 }
3495 
3496 /*
3497  * Push a page out to disk, klustering if possible.
3498  *
3499  *	IN:	vp	- file to push page to.
3500  *		pp	- page to push.
3501  *		flags	- additional flags.
3502  *		cr	- credentials of caller.
3503  *
3504  *	OUT:	offp	- start of range pushed.
3505  *		lenp	- len of range pushed.
3506  *
3507  *	RETURN:	0 if success
3508  *		error code if failure
3509  *
3510  * NOTE: callers must have locked the page to be pushed.  On
3511  * exit, the page (and all other pages in the kluster) must be
3512  * unlocked.
3513  */
3514 /* ARGSUSED */
3515 static int
3516 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3517 		size_t *lenp, int flags, cred_t *cr)
3518 {
3519 	znode_t		*zp = VTOZ(vp);
3520 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3521 	zilog_t		*zilog = zfsvfs->z_log;
3522 	dmu_tx_t	*tx;
3523 	rl_t		*rl;
3524 	u_offset_t	off, koff;
3525 	size_t		len, klen;
3526 	uint64_t	filesz;
3527 	int		err;
3528 
3529 	filesz = zp->z_phys->zp_size;
3530 	off = pp->p_offset;
3531 	len = PAGESIZE;
3532 	/*
3533 	 * If our blocksize is bigger than the page size, try to kluster
3534 	 * muiltiple pages so that we write a full block (thus avoiding
3535 	 * a read-modify-write).
3536 	 */
3537 	if (off < filesz && zp->z_blksz > PAGESIZE) {
3538 		if (!ISP2(zp->z_blksz)) {
3539 			/* Only one block in the file. */
3540 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3541 			koff = 0;
3542 		} else {
3543 			klen = zp->z_blksz;
3544 			koff = P2ALIGN(off, (u_offset_t)klen);
3545 		}
3546 		ASSERT(koff <= filesz);
3547 		if (koff + klen > filesz)
3548 			klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE);
3549 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
3550 	}
3551 	ASSERT3U(btop(len), ==, btopr(len));
3552 top:
3553 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
3554 	/*
3555 	 * Can't push pages past end-of-file.
3556 	 */
3557 	filesz = zp->z_phys->zp_size;
3558 	if (off >= filesz) {
3559 		/* ignore all pages */
3560 		err = 0;
3561 		goto out;
3562 	} else if (off + len > filesz) {
3563 		int npages = btopr(filesz - off);
3564 		page_t *trunc;
3565 
3566 		page_list_break(&pp, &trunc, npages);
3567 		/* ignore pages past end of file */
3568 		if (trunc)
3569 			pvn_write_done(trunc, flags);
3570 		len = filesz - off;
3571 	}
3572 
3573 	tx = dmu_tx_create(zfsvfs->z_os);
3574 	dmu_tx_hold_write(tx, zp->z_id, off, len);
3575 	dmu_tx_hold_bonus(tx, zp->z_id);
3576 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
3577 	if (err != 0) {
3578 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3579 			zfs_range_unlock(rl);
3580 			dmu_tx_wait(tx);
3581 			dmu_tx_abort(tx);
3582 			err = 0;
3583 			goto top;
3584 		}
3585 		dmu_tx_abort(tx);
3586 		goto out;
3587 	}
3588 
3589 	if (zp->z_blksz <= PAGESIZE) {
3590 		caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1);
3591 		ASSERT3U(len, <=, PAGESIZE);
3592 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
3593 		ppmapout(va);
3594 	} else {
3595 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
3596 	}
3597 
3598 	if (err == 0) {
3599 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
3600 		zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0);
3601 		dmu_tx_commit(tx);
3602 	}
3603 
3604 out:
3605 	zfs_range_unlock(rl);
3606 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
3607 	if (offp)
3608 		*offp = off;
3609 	if (lenp)
3610 		*lenp = len;
3611 
3612 	return (err);
3613 }
3614 
3615 /*
3616  * Copy the portion of the file indicated from pages into the file.
3617  * The pages are stored in a page list attached to the files vnode.
3618  *
3619  *	IN:	vp	- vnode of file to push page data to.
3620  *		off	- position in file to put data.
3621  *		len	- amount of data to write.
3622  *		flags	- flags to control the operation.
3623  *		cr	- credentials of caller.
3624  *		ct	- caller context.
3625  *
3626  *	RETURN:	0 if success
3627  *		error code if failure
3628  *
3629  * Timestamps:
3630  *	vp - ctime|mtime updated
3631  */
3632 /*ARGSUSED*/
3633 static int
3634 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
3635     caller_context_t *ct)
3636 {
3637 	znode_t		*zp = VTOZ(vp);
3638 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3639 	page_t		*pp;
3640 	size_t		io_len;
3641 	u_offset_t	io_off;
3642 	uint64_t	filesz;
3643 	int		error = 0;
3644 
3645 	ZFS_ENTER(zfsvfs);
3646 	ZFS_VERIFY_ZP(zp);
3647 
3648 	if (len == 0) {
3649 		/*
3650 		 * Search the entire vp list for pages >= off.
3651 		 */
3652 		error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage,
3653 		    flags, cr);
3654 		goto out;
3655 	}
3656 
3657 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3658 	if (off > filesz) {
3659 		/* past end of file */
3660 		ZFS_EXIT(zfsvfs);
3661 		return (0);
3662 	}
3663 
3664 	len = MIN(len, filesz - off);
3665 
3666 	for (io_off = off; io_off < off + len; io_off += io_len) {
3667 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
3668 			pp = page_lookup(vp, io_off,
3669 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
3670 		} else {
3671 			pp = page_lookup_nowait(vp, io_off,
3672 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
3673 		}
3674 
3675 		if (pp != NULL && pvn_getdirty(pp, flags)) {
3676 			int err;
3677 
3678 			/*
3679 			 * Found a dirty page to push
3680 			 */
3681 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
3682 			if (err)
3683 				error = err;
3684 		} else {
3685 			io_len = PAGESIZE;
3686 		}
3687 	}
3688 out:
3689 	if ((flags & B_ASYNC) == 0)
3690 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
3691 	ZFS_EXIT(zfsvfs);
3692 	return (error);
3693 }
3694 
3695 /*ARGSUSED*/
3696 void
3697 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
3698 {
3699 	znode_t	*zp = VTOZ(vp);
3700 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3701 	int error;
3702 
3703 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
3704 	if (zp->z_dbuf == NULL) {
3705 		/*
3706 		 * The fs has been unmounted, or we did a
3707 		 * suspend/resume and this file no longer exists.
3708 		 */
3709 		if (vn_has_cached_data(vp)) {
3710 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
3711 			    B_INVAL, cr);
3712 		}
3713 
3714 		mutex_enter(&zp->z_lock);
3715 		vp->v_count = 0; /* count arrives as 1 */
3716 		mutex_exit(&zp->z_lock);
3717 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
3718 		zfs_znode_free(zp);
3719 		return;
3720 	}
3721 
3722 	/*
3723 	 * Attempt to push any data in the page cache.  If this fails
3724 	 * we will get kicked out later in zfs_zinactive().
3725 	 */
3726 	if (vn_has_cached_data(vp)) {
3727 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
3728 		    cr);
3729 	}
3730 
3731 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3732 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3733 
3734 		dmu_tx_hold_bonus(tx, zp->z_id);
3735 		error = dmu_tx_assign(tx, TXG_WAIT);
3736 		if (error) {
3737 			dmu_tx_abort(tx);
3738 		} else {
3739 			dmu_buf_will_dirty(zp->z_dbuf, tx);
3740 			mutex_enter(&zp->z_lock);
3741 			zp->z_atime_dirty = 0;
3742 			mutex_exit(&zp->z_lock);
3743 			dmu_tx_commit(tx);
3744 		}
3745 	}
3746 
3747 	zfs_zinactive(zp);
3748 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
3749 }
3750 
3751 /*
3752  * Bounds-check the seek operation.
3753  *
3754  *	IN:	vp	- vnode seeking within
3755  *		ooff	- old file offset
3756  *		noffp	- pointer to new file offset
3757  *		ct	- caller context
3758  *
3759  *	RETURN:	0 if success
3760  *		EINVAL if new offset invalid
3761  */
3762 /* ARGSUSED */
3763 static int
3764 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
3765     caller_context_t *ct)
3766 {
3767 	if (vp->v_type == VDIR)
3768 		return (0);
3769 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3770 }
3771 
3772 /*
3773  * Pre-filter the generic locking function to trap attempts to place
3774  * a mandatory lock on a memory mapped file.
3775  */
3776 static int
3777 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3778     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
3779 {
3780 	znode_t *zp = VTOZ(vp);
3781 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3782 	int error;
3783 
3784 	ZFS_ENTER(zfsvfs);
3785 	ZFS_VERIFY_ZP(zp);
3786 
3787 	/*
3788 	 * We are following the UFS semantics with respect to mapcnt
3789 	 * here: If we see that the file is mapped already, then we will
3790 	 * return an error, but we don't worry about races between this
3791 	 * function and zfs_map().
3792 	 */
3793 	if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3794 		ZFS_EXIT(zfsvfs);
3795 		return (EAGAIN);
3796 	}
3797 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct);
3798 	ZFS_EXIT(zfsvfs);
3799 	return (error);
3800 }
3801 
3802 /*
3803  * If we can't find a page in the cache, we will create a new page
3804  * and fill it with file data.  For efficiency, we may try to fill
3805  * multiple pages at once (klustering).
3806  */
3807 static int
3808 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3809     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3810 {
3811 	znode_t *zp = VTOZ(vp);
3812 	page_t *pp, *cur_pp;
3813 	objset_t *os = zp->z_zfsvfs->z_os;
3814 	caddr_t va;
3815 	u_offset_t io_off, total;
3816 	uint64_t oid = zp->z_id;
3817 	size_t io_len;
3818 	uint64_t filesz;
3819 	int err;
3820 
3821 	/*
3822 	 * If we are only asking for a single page don't bother klustering.
3823 	 */
3824 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3825 	if (off >= filesz)
3826 		return (EFAULT);
3827 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3828 		io_off = off;
3829 		io_len = PAGESIZE;
3830 		pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr);
3831 	} else {
3832 		/*
3833 		 * Try to fill a kluster of pages (a blocks worth).
3834 		 */
3835 		size_t klen;
3836 		u_offset_t koff;
3837 
3838 		if (!ISP2(zp->z_blksz)) {
3839 			/* Only one block in the file. */
3840 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3841 			koff = 0;
3842 		} else {
3843 			/*
3844 			 * It would be ideal to align our offset to the
3845 			 * blocksize but doing so has resulted in some
3846 			 * strange application crashes. For now, we
3847 			 * leave the offset as is and only adjust the
3848 			 * length if we are off the end of the file.
3849 			 */
3850 			koff = off;
3851 			klen = plsz;
3852 		}
3853 		ASSERT(koff <= filesz);
3854 		if (koff + klen > filesz)
3855 			klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff;
3856 		ASSERT3U(off, >=, koff);
3857 		ASSERT3U(off, <, koff + klen);
3858 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3859 		    &io_len, koff, klen, 0);
3860 	}
3861 	if (pp == NULL) {
3862 		/*
3863 		 * Some other thread entered the page before us.
3864 		 * Return to zfs_getpage to retry the lookup.
3865 		 */
3866 		*pl = NULL;
3867 		return (0);
3868 	}
3869 
3870 	/*
3871 	 * Fill the pages in the kluster.
3872 	 */
3873 	cur_pp = pp;
3874 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3875 		ASSERT3U(io_off, ==, cur_pp->p_offset);
3876 		va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1);
3877 		err = dmu_read(os, oid, io_off, PAGESIZE, va);
3878 		ppmapout(va);
3879 		if (err) {
3880 			/* On error, toss the entire kluster */
3881 			pvn_read_done(pp, B_ERROR);
3882 			return (err);
3883 		}
3884 		cur_pp = cur_pp->p_next;
3885 	}
3886 out:
3887 	/*
3888 	 * Fill in the page list array from the kluster.  If
3889 	 * there are too many pages in the kluster, return
3890 	 * as many pages as possible starting from the desired
3891 	 * offset `off'.
3892 	 * NOTE: the page list will always be null terminated.
3893 	 */
3894 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3895 
3896 	return (0);
3897 }
3898 
3899 /*
3900  * Return pointers to the pages for the file region [off, off + len]
3901  * in the pl array.  If plsz is greater than len, this function may
3902  * also return page pointers from before or after the specified
3903  * region (i.e. some region [off', off' + plsz]).  These additional
3904  * pages are only returned if they are already in the cache, or were
3905  * created as part of a klustered read.
3906  *
3907  *	IN:	vp	- vnode of file to get data from.
3908  *		off	- position in file to get data from.
3909  *		len	- amount of data to retrieve.
3910  *		plsz	- length of provided page list.
3911  *		seg	- segment to obtain pages for.
3912  *		addr	- virtual address of fault.
3913  *		rw	- mode of created pages.
3914  *		cr	- credentials of caller.
3915  *		ct	- caller context.
3916  *
3917  *	OUT:	protp	- protection mode of created pages.
3918  *		pl	- list of pages created.
3919  *
3920  *	RETURN:	0 if success
3921  *		error code if failure
3922  *
3923  * Timestamps:
3924  *	vp - atime updated
3925  */
3926 /* ARGSUSED */
3927 static int
3928 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3929 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3930 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
3931 {
3932 	znode_t		*zp = VTOZ(vp);
3933 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3934 	page_t		*pp, **pl0 = pl;
3935 	int		need_unlock = 0, err = 0;
3936 	offset_t	orig_off;
3937 
3938 	ZFS_ENTER(zfsvfs);
3939 	ZFS_VERIFY_ZP(zp);
3940 
3941 	if (protp)
3942 		*protp = PROT_ALL;
3943 
3944 	/* no faultahead (for now) */
3945 	if (pl == NULL) {
3946 		ZFS_EXIT(zfsvfs);
3947 		return (0);
3948 	}
3949 
3950 	/* can't fault past EOF */
3951 	if (off >= zp->z_phys->zp_size) {
3952 		ZFS_EXIT(zfsvfs);
3953 		return (EFAULT);
3954 	}
3955 	orig_off = off;
3956 
3957 	/*
3958 	 * If we already own the lock, then we must be page faulting
3959 	 * in the middle of a write to this file (i.e., we are writing
3960 	 * to this file using data from a mapped region of the file).
3961 	 */
3962 	if (rw_owner(&zp->z_map_lock) != curthread) {
3963 		rw_enter(&zp->z_map_lock, RW_WRITER);
3964 		need_unlock = TRUE;
3965 	}
3966 
3967 	/*
3968 	 * Loop through the requested range [off, off + len] looking
3969 	 * for pages.  If we don't find a page, we will need to create
3970 	 * a new page and fill it with data from the file.
3971 	 */
3972 	while (len > 0) {
3973 		if (plsz < PAGESIZE)
3974 			break;
3975 		if (pp = page_lookup(vp, off, SE_SHARED)) {
3976 			*pl++ = pp;
3977 			off += PAGESIZE;
3978 			addr += PAGESIZE;
3979 			len -= PAGESIZE;
3980 			plsz -= PAGESIZE;
3981 		} else {
3982 			err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw);
3983 			if (err)
3984 				goto out;
3985 			/*
3986 			 * klustering may have changed our region
3987 			 * to be block aligned.
3988 			 */
3989 			if (((pp = *pl) != 0) && (off != pp->p_offset)) {
3990 				int delta = off - pp->p_offset;
3991 				len += delta;
3992 				off -= delta;
3993 				addr -= delta;
3994 			}
3995 			while (*pl) {
3996 				pl++;
3997 				off += PAGESIZE;
3998 				addr += PAGESIZE;
3999 				plsz -= PAGESIZE;
4000 				if (len > PAGESIZE)
4001 					len -= PAGESIZE;
4002 				else
4003 					len = 0;
4004 			}
4005 		}
4006 	}
4007 
4008 	/*
4009 	 * Fill out the page array with any pages already in the cache.
4010 	 */
4011 	while (plsz > 0) {
4012 		pp = page_lookup_nowait(vp, off, SE_SHARED);
4013 		if (pp == NULL)
4014 			break;
4015 		*pl++ = pp;
4016 		off += PAGESIZE;
4017 		plsz -= PAGESIZE;
4018 	}
4019 
4020 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4021 out:
4022 	/*
4023 	 * We can't grab the range lock for the page as reader which would
4024 	 * stop truncation as this leads to deadlock. So we need to recheck
4025 	 * the file size.
4026 	 */
4027 	if (orig_off >= zp->z_phys->zp_size)
4028 		err = EFAULT;
4029 	if (err) {
4030 		/*
4031 		 * Release any pages we have previously locked.
4032 		 */
4033 		while (pl > pl0)
4034 			page_unlock(*--pl);
4035 	}
4036 
4037 	*pl = NULL;
4038 
4039 	if (need_unlock)
4040 		rw_exit(&zp->z_map_lock);
4041 
4042 	ZFS_EXIT(zfsvfs);
4043 	return (err);
4044 }
4045 
4046 /*
4047  * Request a memory map for a section of a file.  This code interacts
4048  * with common code and the VM system as follows:
4049  *
4050  *	common code calls mmap(), which ends up in smmap_common()
4051  *
4052  *	this calls VOP_MAP(), which takes you into (say) zfs
4053  *
4054  *	zfs_map() calls as_map(), passing segvn_create() as the callback
4055  *
4056  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
4057  *
4058  *	zfs_addmap() updates z_mapcnt
4059  */
4060 /*ARGSUSED*/
4061 static int
4062 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4063     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4064     caller_context_t *ct)
4065 {
4066 	znode_t *zp = VTOZ(vp);
4067 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4068 	segvn_crargs_t	vn_a;
4069 	int		error;
4070 
4071 	ZFS_ENTER(zfsvfs);
4072 	ZFS_VERIFY_ZP(zp);
4073 
4074 	if ((prot & PROT_WRITE) &&
4075 	    (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY |
4076 	    ZFS_APPENDONLY))) {
4077 		ZFS_EXIT(zfsvfs);
4078 		return (EPERM);
4079 	}
4080 
4081 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4082 	    (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED)) {
4083 		ZFS_EXIT(zfsvfs);
4084 		return (EACCES);
4085 	}
4086 
4087 	if (vp->v_flag & VNOMAP) {
4088 		ZFS_EXIT(zfsvfs);
4089 		return (ENOSYS);
4090 	}
4091 
4092 	if (off < 0 || len > MAXOFFSET_T - off) {
4093 		ZFS_EXIT(zfsvfs);
4094 		return (ENXIO);
4095 	}
4096 
4097 	if (vp->v_type != VREG) {
4098 		ZFS_EXIT(zfsvfs);
4099 		return (ENODEV);
4100 	}
4101 
4102 	/*
4103 	 * If file is locked, disallow mapping.
4104 	 */
4105 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) {
4106 		ZFS_EXIT(zfsvfs);
4107 		return (EAGAIN);
4108 	}
4109 
4110 	as_rangelock(as);
4111 	if ((flags & MAP_FIXED) == 0) {
4112 		map_addr(addrp, len, off, 1, flags);
4113 		if (*addrp == NULL) {
4114 			as_rangeunlock(as);
4115 			ZFS_EXIT(zfsvfs);
4116 			return (ENOMEM);
4117 		}
4118 	} else {
4119 		/*
4120 		 * User specified address - blow away any previous mappings
4121 		 */
4122 		(void) as_unmap(as, *addrp, len);
4123 	}
4124 
4125 	vn_a.vp = vp;
4126 	vn_a.offset = (u_offset_t)off;
4127 	vn_a.type = flags & MAP_TYPE;
4128 	vn_a.prot = prot;
4129 	vn_a.maxprot = maxprot;
4130 	vn_a.cred = cr;
4131 	vn_a.amp = NULL;
4132 	vn_a.flags = flags & ~MAP_TYPE;
4133 	vn_a.szc = 0;
4134 	vn_a.lgrp_mem_policy_flags = 0;
4135 
4136 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4137 
4138 	as_rangeunlock(as);
4139 	ZFS_EXIT(zfsvfs);
4140 	return (error);
4141 }
4142 
4143 /* ARGSUSED */
4144 static int
4145 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4146     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4147     caller_context_t *ct)
4148 {
4149 	uint64_t pages = btopr(len);
4150 
4151 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4152 	return (0);
4153 }
4154 
4155 /*
4156  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4157  * more accurate mtime for the associated file.  Since we don't have a way of
4158  * detecting when the data was actually modified, we have to resort to
4159  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4160  * last page is pushed.  The problem occurs when the msync() call is omitted,
4161  * which by far the most common case:
4162  *
4163  * 	open()
4164  * 	mmap()
4165  * 	<modify memory>
4166  * 	munmap()
4167  * 	close()
4168  * 	<time lapse>
4169  * 	putpage() via fsflush
4170  *
4171  * If we wait until fsflush to come along, we can have a modification time that
4172  * is some arbitrary point in the future.  In order to prevent this in the
4173  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4174  * torn down.
4175  */
4176 /* ARGSUSED */
4177 static int
4178 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4179     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4180     caller_context_t *ct)
4181 {
4182 	uint64_t pages = btopr(len);
4183 
4184 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4185 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4186 
4187 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4188 	    vn_has_cached_data(vp))
4189 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4190 
4191 	return (0);
4192 }
4193 
4194 /*
4195  * Free or allocate space in a file.  Currently, this function only
4196  * supports the `F_FREESP' command.  However, this command is somewhat
4197  * misnamed, as its functionality includes the ability to allocate as
4198  * well as free space.
4199  *
4200  *	IN:	vp	- vnode of file to free data in.
4201  *		cmd	- action to take (only F_FREESP supported).
4202  *		bfp	- section of file to free/alloc.
4203  *		flag	- current file open mode flags.
4204  *		offset	- current file offset.
4205  *		cr	- credentials of caller [UNUSED].
4206  *		ct	- caller context.
4207  *
4208  *	RETURN:	0 if success
4209  *		error code if failure
4210  *
4211  * Timestamps:
4212  *	vp - ctime|mtime updated
4213  */
4214 /* ARGSUSED */
4215 static int
4216 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4217     offset_t offset, cred_t *cr, caller_context_t *ct)
4218 {
4219 	znode_t		*zp = VTOZ(vp);
4220 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4221 	uint64_t	off, len;
4222 	int		error;
4223 
4224 	ZFS_ENTER(zfsvfs);
4225 	ZFS_VERIFY_ZP(zp);
4226 
4227 top:
4228 	if (cmd != F_FREESP) {
4229 		ZFS_EXIT(zfsvfs);
4230 		return (EINVAL);
4231 	}
4232 
4233 	if (error = convoff(vp, bfp, 0, offset)) {
4234 		ZFS_EXIT(zfsvfs);
4235 		return (error);
4236 	}
4237 
4238 	if (bfp->l_len < 0) {
4239 		ZFS_EXIT(zfsvfs);
4240 		return (EINVAL);
4241 	}
4242 
4243 	off = bfp->l_start;
4244 	len = bfp->l_len; /* 0 means from off to end of file */
4245 
4246 	do {
4247 		error = zfs_freesp(zp, off, len, flag, TRUE);
4248 		/* NB: we already did dmu_tx_wait() if necessary */
4249 	} while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
4250 
4251 	ZFS_EXIT(zfsvfs);
4252 	return (error);
4253 }
4254 
4255 /*ARGSUSED*/
4256 static int
4257 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4258 {
4259 	znode_t		*zp = VTOZ(vp);
4260 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4261 	uint32_t	gen;
4262 	uint64_t	object = zp->z_id;
4263 	zfid_short_t	*zfid;
4264 	int		size, i;
4265 
4266 	ZFS_ENTER(zfsvfs);
4267 	ZFS_VERIFY_ZP(zp);
4268 	gen = (uint32_t)zp->z_gen;
4269 
4270 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4271 	if (fidp->fid_len < size) {
4272 		fidp->fid_len = size;
4273 		ZFS_EXIT(zfsvfs);
4274 		return (ENOSPC);
4275 	}
4276 
4277 	zfid = (zfid_short_t *)fidp;
4278 
4279 	zfid->zf_len = size;
4280 
4281 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4282 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4283 
4284 	/* Must have a non-zero generation number to distinguish from .zfs */
4285 	if (gen == 0)
4286 		gen = 1;
4287 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4288 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4289 
4290 	if (size == LONG_FID_LEN) {
4291 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4292 		zfid_long_t	*zlfid;
4293 
4294 		zlfid = (zfid_long_t *)fidp;
4295 
4296 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4297 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4298 
4299 		/* XXX - this should be the generation number for the objset */
4300 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4301 			zlfid->zf_setgen[i] = 0;
4302 	}
4303 
4304 	ZFS_EXIT(zfsvfs);
4305 	return (0);
4306 }
4307 
4308 static int
4309 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4310     caller_context_t *ct)
4311 {
4312 	znode_t		*zp, *xzp;
4313 	zfsvfs_t	*zfsvfs;
4314 	zfs_dirlock_t	*dl;
4315 	int		error;
4316 
4317 	switch (cmd) {
4318 	case _PC_LINK_MAX:
4319 		*valp = ULONG_MAX;
4320 		return (0);
4321 
4322 	case _PC_FILESIZEBITS:
4323 		*valp = 64;
4324 		return (0);
4325 
4326 	case _PC_XATTR_EXISTS:
4327 		zp = VTOZ(vp);
4328 		zfsvfs = zp->z_zfsvfs;
4329 		ZFS_ENTER(zfsvfs);
4330 		ZFS_VERIFY_ZP(zp);
4331 		*valp = 0;
4332 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4333 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4334 		if (error == 0) {
4335 			zfs_dirent_unlock(dl);
4336 			if (!zfs_dirempty(xzp))
4337 				*valp = 1;
4338 			VN_RELE(ZTOV(xzp));
4339 		} else if (error == ENOENT) {
4340 			/*
4341 			 * If there aren't extended attributes, it's the
4342 			 * same as having zero of them.
4343 			 */
4344 			error = 0;
4345 		}
4346 		ZFS_EXIT(zfsvfs);
4347 		return (error);
4348 
4349 	case _PC_SATTR_ENABLED:
4350 	case _PC_SATTR_EXISTS:
4351 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_XVATTR) &&
4352 		    (vp->v_type == VREG || vp->v_type == VDIR);
4353 		return (0);
4354 
4355 	case _PC_ACL_ENABLED:
4356 		*valp = _ACL_ACE_ENABLED;
4357 		return (0);
4358 
4359 	case _PC_MIN_HOLE_SIZE:
4360 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4361 		return (0);
4362 
4363 	default:
4364 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4365 	}
4366 }
4367 
4368 /*ARGSUSED*/
4369 static int
4370 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4371     caller_context_t *ct)
4372 {
4373 	znode_t *zp = VTOZ(vp);
4374 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4375 	int error;
4376 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4377 
4378 	ZFS_ENTER(zfsvfs);
4379 	ZFS_VERIFY_ZP(zp);
4380 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4381 	ZFS_EXIT(zfsvfs);
4382 
4383 	return (error);
4384 }
4385 
4386 /*ARGSUSED*/
4387 static int
4388 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4389     caller_context_t *ct)
4390 {
4391 	znode_t *zp = VTOZ(vp);
4392 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4393 	int error;
4394 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4395 
4396 	ZFS_ENTER(zfsvfs);
4397 	ZFS_VERIFY_ZP(zp);
4398 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4399 	ZFS_EXIT(zfsvfs);
4400 	return (error);
4401 }
4402 
4403 /*
4404  * Predeclare these here so that the compiler assumes that
4405  * this is an "old style" function declaration that does
4406  * not include arguments => we won't get type mismatch errors
4407  * in the initializations that follow.
4408  */
4409 static int zfs_inval();
4410 static int zfs_isdir();
4411 
4412 static int
4413 zfs_inval()
4414 {
4415 	return (EINVAL);
4416 }
4417 
4418 static int
4419 zfs_isdir()
4420 {
4421 	return (EISDIR);
4422 }
4423 /*
4424  * Directory vnode operations template
4425  */
4426 vnodeops_t *zfs_dvnodeops;
4427 const fs_operation_def_t zfs_dvnodeops_template[] = {
4428 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4429 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4430 	VOPNAME_READ,		{ .error = zfs_isdir },
4431 	VOPNAME_WRITE,		{ .error = zfs_isdir },
4432 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4433 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4434 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4435 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4436 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4437 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4438 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4439 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4440 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4441 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
4442 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4443 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4444 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
4445 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4446 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4447 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4448 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4449 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4450 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4451 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4452 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
4453 	NULL,			NULL
4454 };
4455 
4456 /*
4457  * Regular file vnode operations template
4458  */
4459 vnodeops_t *zfs_fvnodeops;
4460 const fs_operation_def_t zfs_fvnodeops_template[] = {
4461 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4462 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4463 	VOPNAME_READ,		{ .vop_read = zfs_read },
4464 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
4465 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4466 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4467 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4468 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4469 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4470 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4471 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4472 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4473 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4474 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4475 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
4476 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
4477 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
4478 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
4479 	VOPNAME_MAP,		{ .vop_map = zfs_map },
4480 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
4481 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
4482 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4483 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4484 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4485 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4486 	NULL,			NULL
4487 };
4488 
4489 /*
4490  * Symbolic link vnode operations template
4491  */
4492 vnodeops_t *zfs_symvnodeops;
4493 const fs_operation_def_t zfs_symvnodeops_template[] = {
4494 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4495 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4496 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4497 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4498 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
4499 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4500 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4501 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4502 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4503 	NULL,			NULL
4504 };
4505 
4506 /*
4507  * Extended attribute directory vnode operations template
4508  *	This template is identical to the directory vnodes
4509  *	operation template except for restricted operations:
4510  *		VOP_MKDIR()
4511  *		VOP_SYMLINK()
4512  * Note that there are other restrictions embedded in:
4513  *	zfs_create()	- restrict type to VREG
4514  *	zfs_link()	- no links into/out of attribute space
4515  *	zfs_rename()	- no moves into/out of attribute space
4516  */
4517 vnodeops_t *zfs_xdvnodeops;
4518 const fs_operation_def_t zfs_xdvnodeops_template[] = {
4519 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4520 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4521 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4522 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4523 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4524 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4525 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4526 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4527 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4528 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4529 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4530 	VOPNAME_MKDIR,		{ .error = zfs_inval },
4531 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4532 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4533 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
4534 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4535 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4536 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4537 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4538 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4539 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4540 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4541 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4542 	NULL,			NULL
4543 };
4544 
4545 /*
4546  * Error vnode operations template
4547  */
4548 vnodeops_t *zfs_evnodeops;
4549 const fs_operation_def_t zfs_evnodeops_template[] = {
4550 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4551 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4552 	NULL,			NULL
4553 };
4554