xref: /titanic_41/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 0e2b17abcde4314f4a65c36af6387d0cb3d8f5fb)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2013 by Delphix. All rights reserved.
24  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
25  */
26 
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2010 Robert Milkowski */
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 <vm/kpm.h>
51 #include <vm/seg_kpm.h>
52 #include <sys/mman.h>
53 #include <sys/pathname.h>
54 #include <sys/cmn_err.h>
55 #include <sys/errno.h>
56 #include <sys/unistd.h>
57 #include <sys/zfs_dir.h>
58 #include <sys/zfs_acl.h>
59 #include <sys/zfs_ioctl.h>
60 #include <sys/fs/zfs.h>
61 #include <sys/dmu.h>
62 #include <sys/dmu_objset.h>
63 #include <sys/spa.h>
64 #include <sys/txg.h>
65 #include <sys/dbuf.h>
66 #include <sys/zap.h>
67 #include <sys/sa.h>
68 #include <sys/dirent.h>
69 #include <sys/policy.h>
70 #include <sys/sunddi.h>
71 #include <sys/filio.h>
72 #include <sys/sid.h>
73 #include "fs/fs_subr.h"
74 #include <sys/zfs_ctldir.h>
75 #include <sys/zfs_fuid.h>
76 #include <sys/zfs_sa.h>
77 #include <sys/dnlc.h>
78 #include <sys/zfs_rlock.h>
79 #include <sys/extdirent.h>
80 #include <sys/kidmap.h>
81 #include <sys/cred.h>
82 #include <sys/attr.h>
83 #include <sys/zfs_events.h>
84 #include <sys/fs/zev.h>
85 
86 /*
87  * Programming rules.
88  *
89  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
90  * properly lock its in-core state, create a DMU transaction, do the work,
91  * record this work in the intent log (ZIL), commit the DMU transaction,
92  * and wait for the intent log to commit if it is a synchronous operation.
93  * Moreover, the vnode ops must work in both normal and log replay context.
94  * The ordering of events is important to avoid deadlocks and references
95  * to freed memory.  The example below illustrates the following Big Rules:
96  *
97  *  (1)	A check must be made in each zfs thread for a mounted file system.
98  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
99  *	A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
100  *	must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
101  *	can return EIO from the calling function.
102  *
103  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
104  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
105  *	First, if it's the last reference, the vnode/znode
106  *	can be freed, so the zp may point to freed memory.  Second, the last
107  *	reference will call zfs_zinactive(), which may induce a lot of work --
108  *	pushing cached pages (which acquires range locks) and syncing out
109  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
110  *	which could deadlock the system if you were already holding one.
111  *	If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
112  *
113  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
114  *	as they can span dmu_tx_assign() calls.
115  *
116  *  (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
117  *      dmu_tx_assign().  This is critical because we don't want to block
118  *      while holding locks.
119  *
120  *	If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT.  This
121  *	reduces lock contention and CPU usage when we must wait (note that if
122  *	throughput is constrained by the storage, nearly every transaction
123  *	must wait).
124  *
125  *      Note, in particular, that if a lock is sometimes acquired before
126  *      the tx assigns, and sometimes after (e.g. z_lock), then failing
127  *      to use a non-blocking assign can deadlock the system.  The scenario:
128  *
129  *	Thread A has grabbed a lock before calling dmu_tx_assign().
130  *	Thread B is in an already-assigned tx, and blocks for this lock.
131  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
132  *	forever, because the previous txg can't quiesce until B's tx commits.
133  *
134  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
135  *	then drop all locks, call dmu_tx_wait(), and try again.  On subsequent
136  *	calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
137  *	to indicate that this operation has already called dmu_tx_wait().
138  *	This will ensure that we don't retry forever, waiting a short bit
139  *	each time.
140  *
141  *  (5)	If the operation succeeded, generate the intent log entry for it
142  *	before dropping locks.  This ensures that the ordering of events
143  *	in the intent log matches the order in which they actually occurred.
144  *	During ZIL replay the zfs_log_* functions will update the sequence
145  *	number to indicate the zil transaction has replayed.
146  *
147  *  (6)	At the end of each vnode op, the DMU tx must always commit,
148  *	regardless of whether there were any errors.
149  *
150  *  (7)	After dropping all locks, invoke zil_commit(zilog, foid)
151  *	to ensure that synchronous semantics are provided when necessary.
152  *
153  * In general, this is how things should be ordered in each vnode op:
154  *
155  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
156  * top:
157  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
158  *	rw_enter(...);			// grab any other locks you need
159  *	tx = dmu_tx_create(...);	// get DMU tx
160  *	dmu_tx_hold_*();		// hold each object you might modify
161  *	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
162  *	if (error) {
163  *		rw_exit(...);		// drop locks
164  *		zfs_dirent_unlock(dl);	// unlock directory entry
165  *		VN_RELE(...);		// release held vnodes
166  *		if (error == ERESTART) {
167  *			waited = B_TRUE;
168  *			dmu_tx_wait(tx);
169  *			dmu_tx_abort(tx);
170  *			goto top;
171  *		}
172  *		dmu_tx_abort(tx);	// abort DMU tx
173  *		ZFS_EXIT(zfsvfs);	// finished in zfs
174  *		return (error);		// really out of space
175  *	}
176  *	error = do_real_work();		// do whatever this VOP does
177  *	if (error == 0)
178  *		zfs_log_*(...);		// on success, make ZIL entry
179  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
180  *	rw_exit(...);			// drop locks
181  *	zfs_dirent_unlock(dl);		// unlock directory entry
182  *	VN_RELE(...);			// release held vnodes
183  *	zil_commit(zilog, foid);	// synchronous when necessary
184  *	ZFS_EXIT(zfsvfs);		// finished in zfs
185  *	return (error);			// done, report error
186  */
187 
188 /* ARGSUSED */
189 static int
190 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
191 {
192 	znode_t	*zp = VTOZ(*vpp);
193 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
194 
195 	ZFS_ENTER(zfsvfs);
196 	ZFS_VERIFY_ZP(zp);
197 
198 	if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
199 	    ((flag & FAPPEND) == 0)) {
200 		ZFS_EXIT(zfsvfs);
201 		return (SET_ERROR(EPERM));
202 	}
203 
204 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
205 	    ZTOV(zp)->v_type == VREG &&
206 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
207 		if (fs_vscan(*vpp, cr, 0) != 0) {
208 			ZFS_EXIT(zfsvfs);
209 			return (SET_ERROR(EACCES));
210 		}
211 	}
212 
213 	/* Keep a count of the synchronous opens in the znode */
214 	if (flag & (FSYNC | FDSYNC))
215 		atomic_inc_32(&zp->z_sync_cnt);
216 
217 	ZFS_EXIT(zfsvfs);
218 	return (0);
219 }
220 
221 /* ARGSUSED */
222 static int
223 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
224     caller_context_t *ct)
225 {
226 	znode_t	*zp = VTOZ(vp);
227 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
228 
229 	/*
230 	 * Clean up any locks held by this process on the vp.
231 	 */
232 	cleanlocks(vp, ddi_get_pid(), 0);
233 	cleanshares(vp, ddi_get_pid());
234 
235 	ZFS_ENTER(zfsvfs);
236 	ZFS_VERIFY_ZP(zp);
237 
238 	/* Decrement the synchronous opens in the znode */
239 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
240 		atomic_dec_32(&zp->z_sync_cnt);
241 
242 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
243 	    ZTOV(zp)->v_type == VREG &&
244 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
245 		VERIFY(fs_vscan(vp, cr, 1) == 0);
246 
247 	if (ZTOV(zp)->v_type == VREG && zp->z_new_content) {
248 		zp->z_new_content = 0;
249 		rw_enter(&rz_zev_rwlock, RW_READER);
250 		if (rz_zev_callbacks &&
251 		    rz_zev_callbacks->rz_zev_znode_close_after_update)
252 			rz_zev_callbacks->rz_zev_znode_close_after_update(zp);
253 		rw_exit(&rz_zev_rwlock);
254 	}
255 
256 	ZFS_EXIT(zfsvfs);
257 	return (0);
258 }
259 
260 /*
261  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
262  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
263  */
264 static int
265 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
266 {
267 	znode_t	*zp = VTOZ(vp);
268 	uint64_t noff = (uint64_t)*off; /* new offset */
269 	uint64_t file_sz;
270 	int error;
271 	boolean_t hole;
272 
273 	file_sz = zp->z_size;
274 	if (noff >= file_sz)  {
275 		return (SET_ERROR(ENXIO));
276 	}
277 
278 	if (cmd == _FIO_SEEK_HOLE)
279 		hole = B_TRUE;
280 	else
281 		hole = B_FALSE;
282 
283 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
284 
285 	/* end of file? */
286 	if ((error == ESRCH) || (noff > file_sz)) {
287 		/*
288 		 * Handle the virtual hole at the end of file.
289 		 */
290 		if (hole) {
291 			*off = file_sz;
292 			return (0);
293 		}
294 		return (SET_ERROR(ENXIO));
295 	}
296 
297 	if (noff < *off)
298 		return (error);
299 	*off = noff;
300 	return (error);
301 }
302 
303 /* ARGSUSED */
304 static int
305 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
306     int *rvalp, caller_context_t *ct)
307 {
308 	offset_t off;
309 	int error;
310 	zfsvfs_t *zfsvfs;
311 	znode_t *zp;
312 
313 	switch (com) {
314 	case _FIOFFS:
315 		return (zfs_sync(vp->v_vfsp, 0, cred));
316 
317 		/*
318 		 * The following two ioctls are used by bfu.  Faking out,
319 		 * necessary to avoid bfu errors.
320 		 */
321 	case _FIOGDIO:
322 	case _FIOSDIO:
323 		return (0);
324 
325 	case _FIO_SEEK_DATA:
326 	case _FIO_SEEK_HOLE:
327 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
328 			return (SET_ERROR(EFAULT));
329 
330 		zp = VTOZ(vp);
331 		zfsvfs = zp->z_zfsvfs;
332 		ZFS_ENTER(zfsvfs);
333 		ZFS_VERIFY_ZP(zp);
334 
335 		/* offset parameter is in/out */
336 		error = zfs_holey(vp, com, &off);
337 		ZFS_EXIT(zfsvfs);
338 		if (error)
339 			return (error);
340 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
341 			return (SET_ERROR(EFAULT));
342 		return (0);
343 	}
344 	return (SET_ERROR(ENOTTY));
345 }
346 
347 /*
348  * Utility functions to map and unmap a single physical page.  These
349  * are used to manage the mappable copies of ZFS file data, and therefore
350  * do not update ref/mod bits.
351  */
352 caddr_t
353 zfs_map_page(page_t *pp, enum seg_rw rw)
354 {
355 	if (kpm_enable)
356 		return (hat_kpm_mapin(pp, 0));
357 	ASSERT(rw == S_READ || rw == S_WRITE);
358 	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
359 	    (caddr_t)-1));
360 }
361 
362 void
363 zfs_unmap_page(page_t *pp, caddr_t addr)
364 {
365 	if (kpm_enable) {
366 		hat_kpm_mapout(pp, 0, addr);
367 	} else {
368 		ppmapout(addr);
369 	}
370 }
371 
372 /*
373  * When a file is memory mapped, we must keep the IO data synchronized
374  * between the DMU cache and the memory mapped pages.  What this means:
375  *
376  * On Write:	If we find a memory mapped page, we write to *both*
377  *		the page and the dmu buffer.
378  */
379 static void
380 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
381 {
382 	int64_t	off;
383 
384 	off = start & PAGEOFFSET;
385 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
386 		page_t *pp;
387 		uint64_t nbytes = MIN(PAGESIZE - off, len);
388 
389 		if (pp = page_lookup(vp, start, SE_SHARED)) {
390 			caddr_t va;
391 
392 			va = zfs_map_page(pp, S_WRITE);
393 			(void) dmu_read(os, oid, start+off, nbytes, va+off,
394 			    DMU_READ_PREFETCH);
395 			zfs_unmap_page(pp, va);
396 			page_unlock(pp);
397 		}
398 		len -= nbytes;
399 		off = 0;
400 	}
401 }
402 
403 /*
404  * When a file is memory mapped, we must keep the IO data synchronized
405  * between the DMU cache and the memory mapped pages.  What this means:
406  *
407  * On Read:	We "read" preferentially from memory mapped pages,
408  *		else we default from the dmu buffer.
409  *
410  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
411  *	 the file is memory mapped.
412  */
413 static int
414 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
415 {
416 	znode_t *zp = VTOZ(vp);
417 	objset_t *os = zp->z_zfsvfs->z_os;
418 	int64_t	start, off;
419 	int len = nbytes;
420 	int error = 0;
421 
422 	start = uio->uio_loffset;
423 	off = start & PAGEOFFSET;
424 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
425 		page_t *pp;
426 		uint64_t bytes = MIN(PAGESIZE - off, len);
427 
428 		if (pp = page_lookup(vp, start, SE_SHARED)) {
429 			caddr_t va;
430 
431 			va = zfs_map_page(pp, S_READ);
432 			error = uiomove(va + off, bytes, UIO_READ, uio);
433 			zfs_unmap_page(pp, va);
434 			page_unlock(pp);
435 		} else {
436 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
437 		}
438 		len -= bytes;
439 		off = 0;
440 		if (error)
441 			break;
442 	}
443 	return (error);
444 }
445 
446 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
447 
448 /*
449  * Read bytes from specified file into supplied buffer.
450  *
451  *	IN:	vp	- vnode of file to be read from.
452  *		uio	- structure supplying read location, range info,
453  *			  and return buffer.
454  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
455  *		cr	- credentials of caller.
456  *		ct	- caller context
457  *
458  *	OUT:	uio	- updated offset and range, buffer filled.
459  *
460  *	RETURN:	0 on success, error code on failure.
461  *
462  * Side Effects:
463  *	vp - atime updated if byte count > 0
464  */
465 /* ARGSUSED */
466 static int
467 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
468 {
469 	znode_t		*zp = VTOZ(vp);
470 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
471 	objset_t	*os;
472 	ssize_t		n, nbytes;
473 	int		error = 0;
474 	rl_t		*rl;
475 	xuio_t		*xuio = NULL;
476 
477 	ZFS_ENTER(zfsvfs);
478 	ZFS_VERIFY_ZP(zp);
479 	os = zfsvfs->z_os;
480 
481 	if (zp->z_pflags & ZFS_AV_QUARANTINED) {
482 		ZFS_EXIT(zfsvfs);
483 		return (SET_ERROR(EACCES));
484 	}
485 
486 	/*
487 	 * Validate file offset
488 	 */
489 	if (uio->uio_loffset < (offset_t)0) {
490 		ZFS_EXIT(zfsvfs);
491 		return (SET_ERROR(EINVAL));
492 	}
493 
494 	/*
495 	 * Fasttrack empty reads
496 	 */
497 	if (uio->uio_resid == 0) {
498 		ZFS_EXIT(zfsvfs);
499 		return (0);
500 	}
501 
502 	/*
503 	 * Check for mandatory locks
504 	 */
505 	if (MANDMODE(zp->z_mode)) {
506 		if (error = chklock(vp, FREAD,
507 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
508 			ZFS_EXIT(zfsvfs);
509 			return (error);
510 		}
511 	}
512 
513 	/*
514 	 * If we're in FRSYNC mode, sync out this znode before reading it.
515 	 */
516 	if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
517 		zil_commit(zfsvfs->z_log, zp->z_id);
518 
519 	/*
520 	 * Lock the range against changes.
521 	 */
522 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
523 
524 	/*
525 	 * If we are reading past end-of-file we can skip
526 	 * to the end; but we might still need to set atime.
527 	 */
528 	if (uio->uio_loffset >= zp->z_size) {
529 		error = 0;
530 		goto out;
531 	}
532 
533 	ASSERT(uio->uio_loffset < zp->z_size);
534 	n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
535 
536 	if ((uio->uio_extflg == UIO_XUIO) &&
537 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
538 		int nblk;
539 		int blksz = zp->z_blksz;
540 		uint64_t offset = uio->uio_loffset;
541 
542 		xuio = (xuio_t *)uio;
543 		if ((ISP2(blksz))) {
544 			nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
545 			    blksz)) / blksz;
546 		} else {
547 			ASSERT(offset + n <= blksz);
548 			nblk = 1;
549 		}
550 		(void) dmu_xuio_init(xuio, nblk);
551 
552 		if (vn_has_cached_data(vp)) {
553 			/*
554 			 * For simplicity, we always allocate a full buffer
555 			 * even if we only expect to read a portion of a block.
556 			 */
557 			while (--nblk >= 0) {
558 				(void) dmu_xuio_add(xuio,
559 				    dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
560 				    blksz), 0, blksz);
561 			}
562 		}
563 	}
564 
565 	while (n > 0) {
566 		nbytes = MIN(n, zfs_read_chunk_size -
567 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
568 
569 		if (vn_has_cached_data(vp))
570 			error = mappedread(vp, nbytes, uio);
571 		else
572 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
573 		if (error) {
574 			/* convert checksum errors into IO errors */
575 			if (error == ECKSUM)
576 				error = SET_ERROR(EIO);
577 			break;
578 		}
579 
580 		n -= nbytes;
581 	}
582 out:
583 	zfs_range_unlock(rl);
584 
585 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
586 	ZFS_EXIT(zfsvfs);
587 	return (error);
588 }
589 
590 /*
591  * Write the bytes to a file.
592  *
593  *	IN:	vp	- vnode of file to be written to.
594  *		uio	- structure supplying write location, range info,
595  *			  and data buffer.
596  *		ioflag	- FAPPEND, FSYNC, and/or FDSYNC.  FAPPEND is
597  *			  set if in append mode.
598  *		cr	- credentials of caller.
599  *		ct	- caller context (NFS/CIFS fem monitor only)
600  *
601  *	OUT:	uio	- updated offset and range.
602  *
603  *	RETURN:	0 on success, error code on failure.
604  *
605  * Timestamps:
606  *	vp - ctime|mtime updated if byte count > 0
607  */
608 
609 /* ARGSUSED */
610 static int
611 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
612 {
613 	znode_t		*zp = VTOZ(vp);
614 	rlim64_t	limit = uio->uio_llimit;
615 	ssize_t		start_resid = uio->uio_resid;
616 	ssize_t		tx_bytes;
617 	uint64_t	end_size;
618 	dmu_tx_t	*tx;
619 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
620 	zilog_t		*zilog;
621 	offset_t	woff;
622 	ssize_t		n, nbytes;
623 	rl_t		*rl;
624 	int		max_blksz = zfsvfs->z_max_blksz;
625 	int		error = 0;
626 	arc_buf_t	*abuf;
627 	iovec_t		*aiov = NULL;
628 	xuio_t		*xuio = NULL;
629 	int		i_iov = 0;
630 	int		iovcnt = uio->uio_iovcnt;
631 	iovec_t		*iovp = uio->uio_iov;
632 	int		write_eof;
633 	int		count = 0;
634 	sa_bulk_attr_t	bulk[4];
635 	uint64_t	mtime[2], ctime[2];
636 	ssize_t		lock_off, lock_len;
637 
638 	/*
639 	 * Fasttrack empty write
640 	 */
641 	n = start_resid;
642 	if (n == 0)
643 		return (0);
644 
645 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
646 		limit = MAXOFFSET_T;
647 
648 	ZFS_ENTER(zfsvfs);
649 	ZFS_VERIFY_ZP(zp);
650 
651 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
652 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
653 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
654 	    &zp->z_size, 8);
655 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
656 	    &zp->z_pflags, 8);
657 
658 	/*
659 	 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
660 	 * callers might not be able to detect properly that we are read-only,
661 	 * so check it explicitly here.
662 	 */
663 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
664 		ZFS_EXIT(zfsvfs);
665 		return (SET_ERROR(EROFS));
666 	}
667 
668 	/*
669 	 * If immutable or not appending then return EPERM
670 	 */
671 	if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
672 	    ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
673 	    (uio->uio_loffset < zp->z_size))) {
674 		ZFS_EXIT(zfsvfs);
675 		return (SET_ERROR(EPERM));
676 	}
677 
678 	zilog = zfsvfs->z_log;
679 
680 	/*
681 	 * Validate file offset
682 	 */
683 	woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
684 	if (woff < 0) {
685 		ZFS_EXIT(zfsvfs);
686 		return (SET_ERROR(EINVAL));
687 	}
688 
689 	/*
690 	 * Check for mandatory locks before calling zfs_range_lock()
691 	 * in order to prevent a deadlock with locks set via fcntl().
692 	 */
693 	if (MANDMODE((mode_t)zp->z_mode) &&
694 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
695 		ZFS_EXIT(zfsvfs);
696 		return (error);
697 	}
698 
699 	/*
700 	 * Pre-fault the pages to ensure slow (eg NFS) pages
701 	 * don't hold up txg.
702 	 * Skip this if uio contains loaned arc_buf.
703 	 */
704 	if ((uio->uio_extflg == UIO_XUIO) &&
705 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
706 		xuio = (xuio_t *)uio;
707 	else
708 		uio_prefaultpages(MIN(n, max_blksz), uio);
709 
710 	/*
711 	 * If in append mode, set the io offset pointer to eof.
712 	 */
713 	if (ioflag & FAPPEND) {
714 		/*
715 		 * Obtain an appending range lock to guarantee file append
716 		 * semantics.  We reset the write offset once we have the lock.
717 		 */
718 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
719 		woff = rl->r_off;
720 		if (rl->r_len == UINT64_MAX) {
721 			/*
722 			 * We overlocked the file because this write will cause
723 			 * the file block size to increase.
724 			 * Note that zp_size cannot change with this lock held.
725 			 */
726 			woff = zp->z_size;
727 		}
728 		uio->uio_loffset = woff;
729 	} else {
730 		/*
731 		 * Note that if the file block size will change as a result of
732 		 * this write, then this range lock will lock the entire file
733 		 * so that we can re-write the block safely.
734 		 */
735 
736 		/*
737 		 * If in zev mode, lock offsets are quantized to 1MB chunks
738 		 * so that we can calculate level 1 checksums later on.
739 		 */
740 		if (rz_zev_active()) {
741 			/* start of this megabyte */
742 			lock_off = P2ALIGN(woff, ZEV_L1_SIZE);
743 			/* full megabytes */
744 			lock_len = n + (woff - lock_off);
745 			lock_len = P2ROUNDUP(lock_len, ZEV_L1_SIZE);
746 		} else {
747 			lock_off = woff;
748 			lock_len = n;
749 		}
750 
751 		rl = zfs_range_lock(zp, lock_off, lock_len, RL_WRITER);
752 	}
753 
754 	if (woff >= limit) {
755 		zfs_range_unlock(rl);
756 		ZFS_EXIT(zfsvfs);
757 		return (SET_ERROR(EFBIG));
758 	}
759 
760 	if ((woff + n) > limit || woff > (limit - n))
761 		n = limit - woff;
762 
763 	/* Will this write extend the file length? */
764 	write_eof = (woff + n > zp->z_size);
765 
766 	end_size = MAX(zp->z_size, woff + n);
767 
768 	/*
769 	 * Write the file in reasonable size chunks.  Each chunk is written
770 	 * in a separate transaction; this keeps the intent log records small
771 	 * and allows us to do more fine-grained space accounting.
772 	 */
773 	while (n > 0) {
774 		abuf = NULL;
775 		woff = uio->uio_loffset;
776 		if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
777 		    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
778 			if (abuf != NULL)
779 				dmu_return_arcbuf(abuf);
780 			error = SET_ERROR(EDQUOT);
781 			break;
782 		}
783 
784 		if (xuio && abuf == NULL) {
785 			ASSERT(i_iov < iovcnt);
786 			aiov = &iovp[i_iov];
787 			abuf = dmu_xuio_arcbuf(xuio, i_iov);
788 			dmu_xuio_clear(xuio, i_iov);
789 			DTRACE_PROBE3(zfs_cp_write, int, i_iov,
790 			    iovec_t *, aiov, arc_buf_t *, abuf);
791 			ASSERT((aiov->iov_base == abuf->b_data) ||
792 			    ((char *)aiov->iov_base - (char *)abuf->b_data +
793 			    aiov->iov_len == arc_buf_size(abuf)));
794 			i_iov++;
795 		} else if (abuf == NULL && n >= max_blksz &&
796 		    woff >= zp->z_size &&
797 		    P2PHASE(woff, max_blksz) == 0 &&
798 		    zp->z_blksz == max_blksz) {
799 			/*
800 			 * This write covers a full block.  "Borrow" a buffer
801 			 * from the dmu so that we can fill it before we enter
802 			 * a transaction.  This avoids the possibility of
803 			 * holding up the transaction if the data copy hangs
804 			 * up on a pagefault (e.g., from an NFS server mapping).
805 			 */
806 			size_t cbytes;
807 
808 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
809 			    max_blksz);
810 			ASSERT(abuf != NULL);
811 			ASSERT(arc_buf_size(abuf) == max_blksz);
812 			if (error = uiocopy(abuf->b_data, max_blksz,
813 			    UIO_WRITE, uio, &cbytes)) {
814 				dmu_return_arcbuf(abuf);
815 				break;
816 			}
817 			ASSERT(cbytes == max_blksz);
818 		}
819 
820 		/*
821 		 * Start a transaction.
822 		 */
823 		tx = dmu_tx_create(zfsvfs->z_os);
824 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
825 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
826 		zfs_sa_upgrade_txholds(tx, zp);
827 		error = dmu_tx_assign(tx, TXG_WAIT);
828 		if (error) {
829 			dmu_tx_abort(tx);
830 			if (abuf != NULL)
831 				dmu_return_arcbuf(abuf);
832 			break;
833 		}
834 
835 		/*
836 		 * If zfs_range_lock() over-locked we grow the blocksize
837 		 * and then reduce the lock range.  This will only happen
838 		 * on the first iteration since zfs_range_reduce() will
839 		 * shrink down r_len to the appropriate size.
840 		 */
841 		if (rl->r_len == UINT64_MAX) {
842 			uint64_t new_blksz;
843 
844 			if (zp->z_blksz > max_blksz) {
845 				ASSERT(!ISP2(zp->z_blksz));
846 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
847 			} else {
848 				new_blksz = MIN(end_size, max_blksz);
849 			}
850 			zfs_grow_blocksize(zp, new_blksz, tx);
851 			zfs_range_reduce(rl, woff, n);
852 		}
853 
854 		/*
855 		 * XXX - should we really limit each write to z_max_blksz?
856 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
857 		 */
858 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
859 
860 		if (abuf == NULL) {
861 			tx_bytes = uio->uio_resid;
862 			error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
863 			    uio, nbytes, tx);
864 			tx_bytes -= uio->uio_resid;
865 		} else {
866 			tx_bytes = nbytes;
867 			ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
868 			/*
869 			 * If this is not a full block write, but we are
870 			 * extending the file past EOF and this data starts
871 			 * block-aligned, use assign_arcbuf().  Otherwise,
872 			 * write via dmu_write().
873 			 */
874 			if (tx_bytes < max_blksz && (!write_eof ||
875 			    aiov->iov_base != abuf->b_data)) {
876 				ASSERT(xuio);
877 				dmu_write(zfsvfs->z_os, zp->z_id, woff,
878 				    aiov->iov_len, aiov->iov_base, tx);
879 				dmu_return_arcbuf(abuf);
880 				xuio_stat_wbuf_copied();
881 			} else {
882 				ASSERT(xuio || tx_bytes == max_blksz);
883 				dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
884 				    woff, abuf, tx);
885 			}
886 			ASSERT(tx_bytes <= uio->uio_resid);
887 			uioskip(uio, tx_bytes);
888 		}
889 		if (tx_bytes && vn_has_cached_data(vp)) {
890 			update_pages(vp, woff,
891 			    tx_bytes, zfsvfs->z_os, zp->z_id);
892 		}
893 
894 		/*
895 		 * If we made no progress, we're done.  If we made even
896 		 * partial progress, update the znode and ZIL accordingly.
897 		 */
898 		if (tx_bytes == 0) {
899 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
900 			    (void *)&zp->z_size, sizeof (uint64_t), tx);
901 			dmu_tx_commit(tx);
902 			ASSERT(error != 0);
903 			break;
904 		}
905 
906 		/*
907 		 * Clear Set-UID/Set-GID bits on successful write if not
908 		 * privileged and at least one of the excute bits is set.
909 		 *
910 		 * It would be nice to to this after all writes have
911 		 * been done, but that would still expose the ISUID/ISGID
912 		 * to another app after the partial write is committed.
913 		 *
914 		 * Note: we don't call zfs_fuid_map_id() here because
915 		 * user 0 is not an ephemeral uid.
916 		 */
917 		mutex_enter(&zp->z_acl_lock);
918 		if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
919 		    (S_IXUSR >> 6))) != 0 &&
920 		    (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
921 		    secpolicy_vnode_setid_retain(cr,
922 		    (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
923 			uint64_t newmode;
924 			zp->z_mode &= ~(S_ISUID | S_ISGID);
925 			newmode = zp->z_mode;
926 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
927 			    (void *)&newmode, sizeof (uint64_t), tx);
928 		}
929 		mutex_exit(&zp->z_acl_lock);
930 
931 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
932 		    B_TRUE);
933 
934 		/*
935 		 * Update the file size (zp_size) if it has changed;
936 		 * account for possible concurrent updates.
937 		 */
938 		while ((end_size = zp->z_size) < uio->uio_loffset) {
939 			(void) atomic_cas_64(&zp->z_size, end_size,
940 			    uio->uio_loffset);
941 			ASSERT(error == 0);
942 		}
943 		/*
944 		 * If we are replaying and eof is non zero then force
945 		 * the file size to the specified eof. Note, there's no
946 		 * concurrency during replay.
947 		 */
948 		if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
949 			zp->z_size = zfsvfs->z_replay_eof;
950 
951 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
952 
953 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
954 		dmu_tx_commit(tx);
955 
956 		if (error != 0)
957 			break;
958 		ASSERT(tx_bytes == nbytes);
959 		n -= nbytes;
960 
961 		if (!xuio && n > 0)
962 			uio_prefaultpages(MIN(n, max_blksz), uio);
963 	}
964 
965 	zfs_range_unlock(rl);
966 
967 	/*
968 	 * If we're in replay mode, or we made no progress, return error.
969 	 * Otherwise, it's at least a partial write, so it's successful.
970 	 */
971 	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
972 		ZFS_EXIT(zfsvfs);
973 		return (error);
974 	}
975 
976 	if (ioflag & (FSYNC | FDSYNC) ||
977 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
978 		zil_commit(zilog, zp->z_id);
979 
980 	ZFS_EXIT(zfsvfs);
981 	return (0);
982 }
983 
984 void
985 zfs_get_done(zgd_t *zgd, int error)
986 {
987 	znode_t *zp = zgd->zgd_private;
988 	objset_t *os = zp->z_zfsvfs->z_os;
989 
990 	if (zgd->zgd_db)
991 		dmu_buf_rele(zgd->zgd_db, zgd);
992 
993 	zfs_range_unlock(zgd->zgd_rl);
994 
995 	/*
996 	 * Release the vnode asynchronously as we currently have the
997 	 * txg stopped from syncing.
998 	 */
999 	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1000 
1001 	if (error == 0 && zgd->zgd_bp)
1002 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1003 
1004 	kmem_free(zgd, sizeof (zgd_t));
1005 }
1006 
1007 #ifdef DEBUG
1008 static int zil_fault_io = 0;
1009 #endif
1010 
1011 /*
1012  * Get data to generate a TX_WRITE intent log record.
1013  */
1014 int
1015 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1016 {
1017 	zfsvfs_t *zfsvfs = arg;
1018 	objset_t *os = zfsvfs->z_os;
1019 	znode_t *zp;
1020 	uint64_t object = lr->lr_foid;
1021 	uint64_t offset = lr->lr_offset;
1022 	uint64_t size = lr->lr_length;
1023 	blkptr_t *bp = &lr->lr_blkptr;
1024 	dmu_buf_t *db;
1025 	zgd_t *zgd;
1026 	int error = 0;
1027 
1028 	ASSERT(zio != NULL);
1029 	ASSERT(size != 0);
1030 
1031 	/*
1032 	 * Nothing to do if the file has been removed
1033 	 */
1034 	if (zfs_zget(zfsvfs, object, &zp) != 0)
1035 		return (SET_ERROR(ENOENT));
1036 	if (zp->z_unlinked) {
1037 		/*
1038 		 * Release the vnode asynchronously as we currently have the
1039 		 * txg stopped from syncing.
1040 		 */
1041 		VN_RELE_ASYNC(ZTOV(zp),
1042 		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1043 		return (SET_ERROR(ENOENT));
1044 	}
1045 
1046 	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1047 	zgd->zgd_zilog = zfsvfs->z_log;
1048 	zgd->zgd_private = zp;
1049 
1050 	/*
1051 	 * Write records come in two flavors: immediate and indirect.
1052 	 * For small writes it's cheaper to store the data with the
1053 	 * log record (immediate); for large writes it's cheaper to
1054 	 * sync the data and get a pointer to it (indirect) so that
1055 	 * we don't have to write the data twice.
1056 	 */
1057 	if (buf != NULL) { /* immediate write */
1058 		zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1059 		/* test for truncation needs to be done while range locked */
1060 		if (offset >= zp->z_size) {
1061 			error = SET_ERROR(ENOENT);
1062 		} else {
1063 			error = dmu_read(os, object, offset, size, buf,
1064 			    DMU_READ_NO_PREFETCH);
1065 		}
1066 		ASSERT(error == 0 || error == ENOENT);
1067 	} else { /* indirect write */
1068 		/*
1069 		 * Have to lock the whole block to ensure when it's
1070 		 * written out and it's checksum is being calculated
1071 		 * that no one can change the data. We need to re-check
1072 		 * blocksize after we get the lock in case it's changed!
1073 		 */
1074 		for (;;) {
1075 			uint64_t blkoff;
1076 			size = zp->z_blksz;
1077 			blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1078 			offset -= blkoff;
1079 			zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1080 			    RL_READER);
1081 			if (zp->z_blksz == size)
1082 				break;
1083 			offset += blkoff;
1084 			zfs_range_unlock(zgd->zgd_rl);
1085 		}
1086 		/* test for truncation needs to be done while range locked */
1087 		if (lr->lr_offset >= zp->z_size)
1088 			error = SET_ERROR(ENOENT);
1089 #ifdef DEBUG
1090 		if (zil_fault_io) {
1091 			error = SET_ERROR(EIO);
1092 			zil_fault_io = 0;
1093 		}
1094 #endif
1095 		if (error == 0)
1096 			error = dmu_buf_hold(os, object, offset, zgd, &db,
1097 			    DMU_READ_NO_PREFETCH);
1098 
1099 		if (error == 0) {
1100 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1101 			if (obp) {
1102 				ASSERT(BP_IS_HOLE(bp));
1103 				*bp = *obp;
1104 			}
1105 
1106 			zgd->zgd_db = db;
1107 			zgd->zgd_bp = bp;
1108 
1109 			ASSERT(db->db_offset == offset);
1110 			ASSERT(db->db_size == size);
1111 
1112 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1113 			    zfs_get_done, zgd);
1114 			ASSERT(error || lr->lr_length <= zp->z_blksz);
1115 
1116 			/*
1117 			 * On success, we need to wait for the write I/O
1118 			 * initiated by dmu_sync() to complete before we can
1119 			 * release this dbuf.  We will finish everything up
1120 			 * in the zfs_get_done() callback.
1121 			 */
1122 			if (error == 0)
1123 				return (0);
1124 
1125 			if (error == EALREADY) {
1126 				lr->lr_common.lrc_txtype = TX_WRITE2;
1127 				error = 0;
1128 			}
1129 		}
1130 	}
1131 
1132 	zfs_get_done(zgd, error);
1133 
1134 	return (error);
1135 }
1136 
1137 /*ARGSUSED*/
1138 static int
1139 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1140     caller_context_t *ct)
1141 {
1142 	znode_t *zp = VTOZ(vp);
1143 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1144 	int error;
1145 
1146 	ZFS_ENTER(zfsvfs);
1147 	ZFS_VERIFY_ZP(zp);
1148 
1149 	if (flag & V_ACE_MASK)
1150 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1151 	else
1152 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
1153 
1154 	ZFS_EXIT(zfsvfs);
1155 	return (error);
1156 }
1157 
1158 /*
1159  * If vnode is for a device return a specfs vnode instead.
1160  */
1161 static int
1162 specvp_check(vnode_t **vpp, cred_t *cr)
1163 {
1164 	int error = 0;
1165 
1166 	if (IS_DEVVP(*vpp)) {
1167 		struct vnode *svp;
1168 
1169 		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1170 		VN_RELE(*vpp);
1171 		if (svp == NULL)
1172 			error = SET_ERROR(ENOSYS);
1173 		*vpp = svp;
1174 	}
1175 	return (error);
1176 }
1177 
1178 
1179 /*
1180  * Lookup an entry in a directory, or an extended attribute directory.
1181  * If it exists, return a held vnode reference for it.
1182  *
1183  *	IN:	dvp	- vnode of directory to search.
1184  *		nm	- name of entry to lookup.
1185  *		pnp	- full pathname to lookup [UNUSED].
1186  *		flags	- LOOKUP_XATTR set if looking for an attribute.
1187  *		rdir	- root directory vnode [UNUSED].
1188  *		cr	- credentials of caller.
1189  *		ct	- caller context
1190  *		direntflags - directory lookup flags
1191  *		realpnp - returned pathname.
1192  *
1193  *	OUT:	vpp	- vnode of located entry, NULL if not found.
1194  *
1195  *	RETURN:	0 on success, error code on failure.
1196  *
1197  * Timestamps:
1198  *	NA
1199  */
1200 /* ARGSUSED */
1201 static int
1202 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1203     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1204     int *direntflags, pathname_t *realpnp)
1205 {
1206 	znode_t *zdp = VTOZ(dvp);
1207 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1208 	int	error = 0;
1209 
1210 	/* fast path */
1211 	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1212 
1213 		if (dvp->v_type != VDIR) {
1214 			return (SET_ERROR(ENOTDIR));
1215 		} else if (zdp->z_sa_hdl == NULL) {
1216 			return (SET_ERROR(EIO));
1217 		}
1218 
1219 		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1220 			error = zfs_fastaccesschk_execute(zdp, cr);
1221 			if (!error) {
1222 				*vpp = dvp;
1223 				VN_HOLD(*vpp);
1224 				return (0);
1225 			}
1226 			return (error);
1227 		} else {
1228 			vnode_t *tvp = dnlc_lookup(dvp, nm);
1229 
1230 			if (tvp) {
1231 				error = zfs_fastaccesschk_execute(zdp, cr);
1232 				if (error) {
1233 					VN_RELE(tvp);
1234 					return (error);
1235 				}
1236 				if (tvp == DNLC_NO_VNODE) {
1237 					VN_RELE(tvp);
1238 					return (SET_ERROR(ENOENT));
1239 				} else {
1240 					*vpp = tvp;
1241 					return (specvp_check(vpp, cr));
1242 				}
1243 			}
1244 		}
1245 	}
1246 
1247 	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1248 
1249 	ZFS_ENTER(zfsvfs);
1250 	ZFS_VERIFY_ZP(zdp);
1251 
1252 	*vpp = NULL;
1253 
1254 	if (flags & LOOKUP_XATTR) {
1255 		/*
1256 		 * If the xattr property is off, refuse the lookup request.
1257 		 */
1258 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1259 			ZFS_EXIT(zfsvfs);
1260 			return (SET_ERROR(EINVAL));
1261 		}
1262 
1263 		/*
1264 		 * We don't allow recursive attributes..
1265 		 * Maybe someday we will.
1266 		 */
1267 		if (zdp->z_pflags & ZFS_XATTR) {
1268 			ZFS_EXIT(zfsvfs);
1269 			return (SET_ERROR(EINVAL));
1270 		}
1271 
1272 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1273 			ZFS_EXIT(zfsvfs);
1274 			return (error);
1275 		}
1276 
1277 		/*
1278 		 * Do we have permission to get into attribute directory?
1279 		 */
1280 
1281 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1282 		    B_FALSE, cr)) {
1283 			VN_RELE(*vpp);
1284 			*vpp = NULL;
1285 		}
1286 
1287 		ZFS_EXIT(zfsvfs);
1288 		return (error);
1289 	}
1290 
1291 	if (dvp->v_type != VDIR) {
1292 		ZFS_EXIT(zfsvfs);
1293 		return (SET_ERROR(ENOTDIR));
1294 	}
1295 
1296 	/*
1297 	 * Check accessibility of directory.
1298 	 */
1299 
1300 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1301 		ZFS_EXIT(zfsvfs);
1302 		return (error);
1303 	}
1304 
1305 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1306 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1307 		ZFS_EXIT(zfsvfs);
1308 		return (SET_ERROR(EILSEQ));
1309 	}
1310 
1311 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1312 	if (error == 0)
1313 		error = specvp_check(vpp, cr);
1314 
1315 	ZFS_EXIT(zfsvfs);
1316 	return (error);
1317 }
1318 
1319 /*
1320  * Attempt to create a new entry in a directory.  If the entry
1321  * already exists, truncate the file if permissible, else return
1322  * an error.  Return the vp of the created or trunc'd file.
1323  *
1324  *	IN:	dvp	- vnode of directory to put new file entry in.
1325  *		name	- name of new file entry.
1326  *		vap	- attributes of new file.
1327  *		excl	- flag indicating exclusive or non-exclusive mode.
1328  *		mode	- mode to open file with.
1329  *		cr	- credentials of caller.
1330  *		flag	- large file flag [UNUSED].
1331  *		ct	- caller context
1332  *		vsecp 	- ACL to be set
1333  *
1334  *	OUT:	vpp	- vnode of created or trunc'd entry.
1335  *
1336  *	RETURN:	0 on success, error code on failure.
1337  *
1338  * Timestamps:
1339  *	dvp - ctime|mtime updated if new entry created
1340  *	 vp - ctime|mtime always, atime if new
1341  */
1342 
1343 /* ARGSUSED */
1344 static int
1345 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1346     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1347     vsecattr_t *vsecp)
1348 {
1349 	znode_t		*zp, *dzp = VTOZ(dvp);
1350 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1351 	zilog_t		*zilog;
1352 	objset_t	*os;
1353 	zfs_dirlock_t	*dl;
1354 	dmu_tx_t	*tx;
1355 	int		error;
1356 	ksid_t		*ksid;
1357 	uid_t		uid;
1358 	gid_t		gid = crgetgid(cr);
1359 	zfs_acl_ids_t   acl_ids;
1360 	boolean_t	fuid_dirtied;
1361 	boolean_t	have_acl = B_FALSE;
1362 	boolean_t	waited = B_FALSE;
1363 
1364 	/*
1365 	 * If we have an ephemeral id, ACL, or XVATTR then
1366 	 * make sure file system is at proper version
1367 	 */
1368 
1369 	ksid = crgetsid(cr, KSID_OWNER);
1370 	if (ksid)
1371 		uid = ksid_getid(ksid);
1372 	else
1373 		uid = crgetuid(cr);
1374 
1375 	if (zfsvfs->z_use_fuids == B_FALSE &&
1376 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1377 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1378 		return (SET_ERROR(EINVAL));
1379 
1380 	ZFS_ENTER(zfsvfs);
1381 	ZFS_VERIFY_ZP(dzp);
1382 	os = zfsvfs->z_os;
1383 	zilog = zfsvfs->z_log;
1384 
1385 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1386 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1387 		ZFS_EXIT(zfsvfs);
1388 		return (SET_ERROR(EILSEQ));
1389 	}
1390 
1391 	if (vap->va_mask & AT_XVATTR) {
1392 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1393 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1394 			ZFS_EXIT(zfsvfs);
1395 			return (error);
1396 		}
1397 	}
1398 top:
1399 	*vpp = NULL;
1400 
1401 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1402 		vap->va_mode &= ~VSVTX;
1403 
1404 	if (*name == '\0') {
1405 		/*
1406 		 * Null component name refers to the directory itself.
1407 		 */
1408 		VN_HOLD(dvp);
1409 		zp = dzp;
1410 		dl = NULL;
1411 		error = 0;
1412 	} else {
1413 		/* possible VN_HOLD(zp) */
1414 		int zflg = 0;
1415 
1416 		if (flag & FIGNORECASE)
1417 			zflg |= ZCILOOK;
1418 
1419 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1420 		    NULL, NULL);
1421 		if (error) {
1422 			if (have_acl)
1423 				zfs_acl_ids_free(&acl_ids);
1424 			if (strcmp(name, "..") == 0)
1425 				error = SET_ERROR(EISDIR);
1426 			ZFS_EXIT(zfsvfs);
1427 			return (error);
1428 		}
1429 	}
1430 
1431 	if (zp == NULL) {
1432 		uint64_t txtype;
1433 
1434 		/*
1435 		 * Create a new file object and update the directory
1436 		 * to reference it.
1437 		 */
1438 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1439 			if (have_acl)
1440 				zfs_acl_ids_free(&acl_ids);
1441 			goto out;
1442 		}
1443 
1444 		/*
1445 		 * We only support the creation of regular files in
1446 		 * extended attribute directories.
1447 		 */
1448 
1449 		if ((dzp->z_pflags & ZFS_XATTR) &&
1450 		    (vap->va_type != VREG)) {
1451 			if (have_acl)
1452 				zfs_acl_ids_free(&acl_ids);
1453 			error = SET_ERROR(EINVAL);
1454 			goto out;
1455 		}
1456 
1457 		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1458 		    cr, vsecp, &acl_ids)) != 0)
1459 			goto out;
1460 		have_acl = B_TRUE;
1461 
1462 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1463 			zfs_acl_ids_free(&acl_ids);
1464 			error = SET_ERROR(EDQUOT);
1465 			goto out;
1466 		}
1467 
1468 		tx = dmu_tx_create(os);
1469 
1470 		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1471 		    ZFS_SA_BASE_ATTR_SIZE);
1472 
1473 		fuid_dirtied = zfsvfs->z_fuid_dirty;
1474 		if (fuid_dirtied)
1475 			zfs_fuid_txhold(zfsvfs, tx);
1476 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1477 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1478 		if (!zfsvfs->z_use_sa &&
1479 		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1480 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1481 			    0, acl_ids.z_aclp->z_acl_bytes);
1482 		}
1483 		error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1484 		if (error) {
1485 			zfs_dirent_unlock(dl);
1486 			if (error == ERESTART) {
1487 				waited = B_TRUE;
1488 				dmu_tx_wait(tx);
1489 				dmu_tx_abort(tx);
1490 				goto top;
1491 			}
1492 			zfs_acl_ids_free(&acl_ids);
1493 			dmu_tx_abort(tx);
1494 			ZFS_EXIT(zfsvfs);
1495 			return (error);
1496 		}
1497 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1498 
1499 		if (fuid_dirtied)
1500 			zfs_fuid_sync(zfsvfs, tx);
1501 
1502 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1503 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1504 		if (flag & FIGNORECASE)
1505 			txtype |= TX_CI;
1506 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1507 		    vsecp, acl_ids.z_fuidp, vap);
1508 		zfs_acl_ids_free(&acl_ids);
1509 		dmu_tx_commit(tx);
1510 	} else {
1511 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1512 
1513 		if (have_acl)
1514 			zfs_acl_ids_free(&acl_ids);
1515 		have_acl = B_FALSE;
1516 
1517 		/*
1518 		 * A directory entry already exists for this name.
1519 		 */
1520 		/*
1521 		 * Can't truncate an existing file if in exclusive mode.
1522 		 */
1523 		if (excl == EXCL) {
1524 			error = SET_ERROR(EEXIST);
1525 			goto out;
1526 		}
1527 		/*
1528 		 * Can't open a directory for writing.
1529 		 */
1530 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1531 			error = SET_ERROR(EISDIR);
1532 			goto out;
1533 		}
1534 		/*
1535 		 * Verify requested access to file.
1536 		 */
1537 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1538 			goto out;
1539 		}
1540 
1541 		mutex_enter(&dzp->z_lock);
1542 		dzp->z_seq++;
1543 		mutex_exit(&dzp->z_lock);
1544 
1545 		/*
1546 		 * Truncate regular files if requested.
1547 		 */
1548 		if ((ZTOV(zp)->v_type == VREG) &&
1549 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1550 			/* we can't hold any locks when calling zfs_freesp() */
1551 			zfs_dirent_unlock(dl);
1552 			dl = NULL;
1553 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1554 			if (error == 0) {
1555 				vnevent_create(ZTOV(zp), ct);
1556 			}
1557 		}
1558 	}
1559 out:
1560 
1561 	if (dl)
1562 		zfs_dirent_unlock(dl);
1563 
1564 	if (error) {
1565 		if (zp)
1566 			VN_RELE(ZTOV(zp));
1567 	} else {
1568 		*vpp = ZTOV(zp);
1569 		error = specvp_check(vpp, cr);
1570 	}
1571 
1572 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1573 		zil_commit(zilog, 0);
1574 
1575 	ZFS_EXIT(zfsvfs);
1576 	return (error);
1577 }
1578 
1579 /*
1580  * Remove an entry from a directory.
1581  *
1582  *	IN:	dvp	- vnode of directory to remove entry from.
1583  *		name	- name of entry to remove.
1584  *		cr	- credentials of caller.
1585  *		ct	- caller context
1586  *		flags	- case flags
1587  *
1588  *	RETURN:	0 on success, error code on failure.
1589  *
1590  * Timestamps:
1591  *	dvp - ctime|mtime
1592  *	 vp - ctime (if nlink > 0)
1593  */
1594 
1595 uint64_t null_xattr = 0;
1596 
1597 /*ARGSUSED*/
1598 static int
1599 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1600     int flags)
1601 {
1602 	znode_t		*zp, *dzp = VTOZ(dvp);
1603 	znode_t		*xzp;
1604 	vnode_t		*vp;
1605 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1606 	zilog_t		*zilog;
1607 	uint64_t	acl_obj, xattr_obj;
1608 	uint64_t 	xattr_obj_unlinked = 0;
1609 	uint64_t	obj = 0;
1610 	zfs_dirlock_t	*dl;
1611 	dmu_tx_t	*tx;
1612 	boolean_t	may_delete_now, delete_now = FALSE;
1613 	boolean_t	unlinked, toobig = FALSE;
1614 	uint64_t	txtype;
1615 	pathname_t	*realnmp = NULL;
1616 	pathname_t	realnm;
1617 	int		error;
1618 	int		zflg = ZEXISTS;
1619 	boolean_t	waited = B_FALSE;
1620 
1621 	ZFS_ENTER(zfsvfs);
1622 	ZFS_VERIFY_ZP(dzp);
1623 	zilog = zfsvfs->z_log;
1624 
1625 	if (flags & FIGNORECASE) {
1626 		zflg |= ZCILOOK;
1627 		pn_alloc(&realnm);
1628 		realnmp = &realnm;
1629 	}
1630 
1631 top:
1632 	xattr_obj = 0;
1633 	xzp = NULL;
1634 	/*
1635 	 * Attempt to lock directory; fail if entry doesn't exist.
1636 	 */
1637 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1638 	    NULL, realnmp)) {
1639 		if (realnmp)
1640 			pn_free(realnmp);
1641 		ZFS_EXIT(zfsvfs);
1642 		return (error);
1643 	}
1644 
1645 	vp = ZTOV(zp);
1646 
1647 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1648 		goto out;
1649 	}
1650 
1651 	/*
1652 	 * Need to use rmdir for removing directories.
1653 	 */
1654 	if (vp->v_type == VDIR) {
1655 		error = SET_ERROR(EPERM);
1656 		goto out;
1657 	}
1658 
1659 	vnevent_remove(vp, dvp, name, ct);
1660 
1661 	if (realnmp)
1662 		dnlc_remove(dvp, realnmp->pn_buf);
1663 	else
1664 		dnlc_remove(dvp, name);
1665 
1666 	mutex_enter(&vp->v_lock);
1667 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1668 	mutex_exit(&vp->v_lock);
1669 
1670 	/*
1671 	 * We may delete the znode now, or we may put it in the unlinked set;
1672 	 * it depends on whether we're the last link, and on whether there are
1673 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1674 	 * allow for either case.
1675 	 */
1676 	obj = zp->z_id;
1677 	tx = dmu_tx_create(zfsvfs->z_os);
1678 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1679 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1680 	zfs_sa_upgrade_txholds(tx, zp);
1681 	zfs_sa_upgrade_txholds(tx, dzp);
1682 	if (may_delete_now) {
1683 		toobig =
1684 		    zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1685 		/* if the file is too big, only hold_free a token amount */
1686 		dmu_tx_hold_free(tx, zp->z_id, 0,
1687 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1688 	}
1689 
1690 	/* are there any extended attributes? */
1691 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1692 	    &xattr_obj, sizeof (xattr_obj));
1693 	if (error == 0 && xattr_obj) {
1694 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1695 		ASSERT0(error);
1696 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1697 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1698 	}
1699 
1700 	mutex_enter(&zp->z_lock);
1701 	if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1702 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1703 	mutex_exit(&zp->z_lock);
1704 
1705 	/* charge as an update -- would be nice not to charge at all */
1706 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1707 
1708 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1709 	if (error) {
1710 		zfs_dirent_unlock(dl);
1711 		VN_RELE(vp);
1712 		if (xzp)
1713 			VN_RELE(ZTOV(xzp));
1714 		if (error == ERESTART) {
1715 			waited = B_TRUE;
1716 			dmu_tx_wait(tx);
1717 			dmu_tx_abort(tx);
1718 			goto top;
1719 		}
1720 		if (realnmp)
1721 			pn_free(realnmp);
1722 		dmu_tx_abort(tx);
1723 		ZFS_EXIT(zfsvfs);
1724 		return (error);
1725 	}
1726 
1727 	/*
1728 	 * Remove the directory entry.
1729 	 */
1730 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1731 
1732 	if (error) {
1733 		dmu_tx_commit(tx);
1734 		goto out;
1735 	}
1736 
1737 	if (unlinked) {
1738 
1739 		/*
1740 		 * Hold z_lock so that we can make sure that the ACL obj
1741 		 * hasn't changed.  Could have been deleted due to
1742 		 * zfs_sa_upgrade().
1743 		 */
1744 		mutex_enter(&zp->z_lock);
1745 		mutex_enter(&vp->v_lock);
1746 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1747 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1748 		delete_now = may_delete_now && !toobig &&
1749 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1750 		    xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1751 		    acl_obj;
1752 		mutex_exit(&vp->v_lock);
1753 	}
1754 
1755 	txtype = TX_REMOVE;
1756 	if (flags & FIGNORECASE)
1757 		txtype |= TX_CI;
1758 	rw_enter(&rz_zev_rwlock, RW_READER);
1759 	if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_znode_remove)
1760 		rz_zev_callbacks->rz_zev_znode_remove(dzp, zp, tx,
1761 		                                      name, txtype);
1762 	rw_exit(&rz_zev_rwlock);
1763 
1764 	if (delete_now) {
1765 		if (xattr_obj_unlinked) {
1766 			ASSERT3U(xzp->z_links, ==, 2);
1767 			mutex_enter(&xzp->z_lock);
1768 			xzp->z_unlinked = 1;
1769 			xzp->z_links = 0;
1770 			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1771 			    &xzp->z_links, sizeof (xzp->z_links), tx);
1772 			ASSERT3U(error,  ==,  0);
1773 			mutex_exit(&xzp->z_lock);
1774 			zfs_unlinked_add(xzp, tx);
1775 
1776 			if (zp->z_is_sa)
1777 				error = sa_remove(zp->z_sa_hdl,
1778 				    SA_ZPL_XATTR(zfsvfs), tx);
1779 			else
1780 				error = sa_update(zp->z_sa_hdl,
1781 				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
1782 				    sizeof (uint64_t), tx);
1783 			ASSERT0(error);
1784 		}
1785 		mutex_enter(&vp->v_lock);
1786 		vp->v_count--;
1787 		ASSERT0(vp->v_count);
1788 		mutex_exit(&vp->v_lock);
1789 		mutex_exit(&zp->z_lock);
1790 		zfs_znode_delete(zp, tx);
1791 	} else if (unlinked) {
1792 		mutex_exit(&zp->z_lock);
1793 		zfs_unlinked_add(zp, tx);
1794 	}
1795 
1796 	txtype = TX_REMOVE;
1797 	if (flags & FIGNORECASE)
1798 		txtype |= TX_CI;
1799 	zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1800 
1801 	dmu_tx_commit(tx);
1802 out:
1803 	if (realnmp)
1804 		pn_free(realnmp);
1805 
1806 	zfs_dirent_unlock(dl);
1807 
1808 	if (!delete_now)
1809 		VN_RELE(vp);
1810 	if (xzp)
1811 		VN_RELE(ZTOV(xzp));
1812 
1813 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1814 		zil_commit(zilog, 0);
1815 
1816 	ZFS_EXIT(zfsvfs);
1817 	return (error);
1818 }
1819 
1820 /*
1821  * Create a new directory and insert it into dvp using the name
1822  * provided.  Return a pointer to the inserted directory.
1823  *
1824  *	IN:	dvp	- vnode of directory to add subdir to.
1825  *		dirname	- name of new directory.
1826  *		vap	- attributes of new directory.
1827  *		cr	- credentials of caller.
1828  *		ct	- caller context
1829  *		flags	- case flags
1830  *		vsecp	- ACL to be set
1831  *
1832  *	OUT:	vpp	- vnode of created directory.
1833  *
1834  *	RETURN:	0 on success, error code on failure.
1835  *
1836  * Timestamps:
1837  *	dvp - ctime|mtime updated
1838  *	 vp - ctime|mtime|atime updated
1839  */
1840 /*ARGSUSED*/
1841 static int
1842 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1843     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1844 {
1845 	znode_t		*zp, *dzp = VTOZ(dvp);
1846 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1847 	zilog_t		*zilog;
1848 	zfs_dirlock_t	*dl;
1849 	uint64_t	txtype;
1850 	dmu_tx_t	*tx;
1851 	int		error;
1852 	int		zf = ZNEW;
1853 	ksid_t		*ksid;
1854 	uid_t		uid;
1855 	gid_t		gid = crgetgid(cr);
1856 	zfs_acl_ids_t   acl_ids;
1857 	boolean_t	fuid_dirtied;
1858 	boolean_t	waited = B_FALSE;
1859 
1860 	ASSERT(vap->va_type == VDIR);
1861 
1862 	/*
1863 	 * If we have an ephemeral id, ACL, or XVATTR then
1864 	 * make sure file system is at proper version
1865 	 */
1866 
1867 	ksid = crgetsid(cr, KSID_OWNER);
1868 	if (ksid)
1869 		uid = ksid_getid(ksid);
1870 	else
1871 		uid = crgetuid(cr);
1872 	if (zfsvfs->z_use_fuids == B_FALSE &&
1873 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1874 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1875 		return (SET_ERROR(EINVAL));
1876 
1877 	ZFS_ENTER(zfsvfs);
1878 	ZFS_VERIFY_ZP(dzp);
1879 	zilog = zfsvfs->z_log;
1880 
1881 	if (dzp->z_pflags & ZFS_XATTR) {
1882 		ZFS_EXIT(zfsvfs);
1883 		return (SET_ERROR(EINVAL));
1884 	}
1885 
1886 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1887 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1888 		ZFS_EXIT(zfsvfs);
1889 		return (SET_ERROR(EILSEQ));
1890 	}
1891 	if (flags & FIGNORECASE)
1892 		zf |= ZCILOOK;
1893 
1894 	if (vap->va_mask & AT_XVATTR) {
1895 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1896 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1897 			ZFS_EXIT(zfsvfs);
1898 			return (error);
1899 		}
1900 	}
1901 
1902 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1903 	    vsecp, &acl_ids)) != 0) {
1904 		ZFS_EXIT(zfsvfs);
1905 		return (error);
1906 	}
1907 	/*
1908 	 * First make sure the new directory doesn't exist.
1909 	 *
1910 	 * Existence is checked first to make sure we don't return
1911 	 * EACCES instead of EEXIST which can cause some applications
1912 	 * to fail.
1913 	 */
1914 top:
1915 	*vpp = NULL;
1916 
1917 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1918 	    NULL, NULL)) {
1919 		zfs_acl_ids_free(&acl_ids);
1920 		ZFS_EXIT(zfsvfs);
1921 		return (error);
1922 	}
1923 
1924 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1925 		zfs_acl_ids_free(&acl_ids);
1926 		zfs_dirent_unlock(dl);
1927 		ZFS_EXIT(zfsvfs);
1928 		return (error);
1929 	}
1930 
1931 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1932 		zfs_acl_ids_free(&acl_ids);
1933 		zfs_dirent_unlock(dl);
1934 		ZFS_EXIT(zfsvfs);
1935 		return (SET_ERROR(EDQUOT));
1936 	}
1937 
1938 	/*
1939 	 * Add a new entry to the directory.
1940 	 */
1941 	tx = dmu_tx_create(zfsvfs->z_os);
1942 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1943 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1944 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1945 	if (fuid_dirtied)
1946 		zfs_fuid_txhold(zfsvfs, tx);
1947 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1948 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1949 		    acl_ids.z_aclp->z_acl_bytes);
1950 	}
1951 
1952 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1953 	    ZFS_SA_BASE_ATTR_SIZE);
1954 
1955 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1956 	if (error) {
1957 		zfs_dirent_unlock(dl);
1958 		if (error == ERESTART) {
1959 			waited = B_TRUE;
1960 			dmu_tx_wait(tx);
1961 			dmu_tx_abort(tx);
1962 			goto top;
1963 		}
1964 		zfs_acl_ids_free(&acl_ids);
1965 		dmu_tx_abort(tx);
1966 		ZFS_EXIT(zfsvfs);
1967 		return (error);
1968 	}
1969 
1970 	/*
1971 	 * Create new node.
1972 	 */
1973 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1974 
1975 	if (fuid_dirtied)
1976 		zfs_fuid_sync(zfsvfs, tx);
1977 
1978 	/*
1979 	 * Now put new name in parent dir.
1980 	 */
1981 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1982 
1983 	*vpp = ZTOV(zp);
1984 
1985 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1986 	if (flags & FIGNORECASE)
1987 		txtype |= TX_CI;
1988 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1989 	    acl_ids.z_fuidp, vap);
1990 
1991 	zfs_acl_ids_free(&acl_ids);
1992 
1993 	dmu_tx_commit(tx);
1994 
1995 	zfs_dirent_unlock(dl);
1996 
1997 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1998 		zil_commit(zilog, 0);
1999 
2000 	ZFS_EXIT(zfsvfs);
2001 	return (0);
2002 }
2003 
2004 /*
2005  * Remove a directory subdir entry.  If the current working
2006  * directory is the same as the subdir to be removed, the
2007  * remove will fail.
2008  *
2009  *	IN:	dvp	- vnode of directory to remove from.
2010  *		name	- name of directory to be removed.
2011  *		cwd	- vnode of current working directory.
2012  *		cr	- credentials of caller.
2013  *		ct	- caller context
2014  *		flags	- case flags
2015  *
2016  *	RETURN:	0 on success, error code on failure.
2017  *
2018  * Timestamps:
2019  *	dvp - ctime|mtime updated
2020  */
2021 /*ARGSUSED*/
2022 static int
2023 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2024     caller_context_t *ct, int flags)
2025 {
2026 	znode_t		*dzp = VTOZ(dvp);
2027 	znode_t		*zp;
2028 	vnode_t		*vp;
2029 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2030 	zilog_t		*zilog;
2031 	zfs_dirlock_t	*dl;
2032 	dmu_tx_t	*tx;
2033 	int		error;
2034 	int		zflg = ZEXISTS;
2035 	boolean_t	waited = B_FALSE;
2036 
2037 	ZFS_ENTER(zfsvfs);
2038 	ZFS_VERIFY_ZP(dzp);
2039 	zilog = zfsvfs->z_log;
2040 
2041 	if (flags & FIGNORECASE)
2042 		zflg |= ZCILOOK;
2043 top:
2044 	zp = NULL;
2045 
2046 	/*
2047 	 * Attempt to lock directory; fail if entry doesn't exist.
2048 	 */
2049 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2050 	    NULL, NULL)) {
2051 		ZFS_EXIT(zfsvfs);
2052 		return (error);
2053 	}
2054 
2055 	vp = ZTOV(zp);
2056 
2057 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2058 		goto out;
2059 	}
2060 
2061 	if (vp->v_type != VDIR) {
2062 		error = SET_ERROR(ENOTDIR);
2063 		goto out;
2064 	}
2065 
2066 	if (vp == cwd) {
2067 		error = SET_ERROR(EINVAL);
2068 		goto out;
2069 	}
2070 
2071 	vnevent_rmdir(vp, dvp, name, ct);
2072 
2073 	/*
2074 	 * Grab a lock on the directory to make sure that noone is
2075 	 * trying to add (or lookup) entries while we are removing it.
2076 	 */
2077 	rw_enter(&zp->z_name_lock, RW_WRITER);
2078 
2079 	/*
2080 	 * Grab a lock on the parent pointer to make sure we play well
2081 	 * with the treewalk and directory rename code.
2082 	 */
2083 	rw_enter(&zp->z_parent_lock, RW_WRITER);
2084 
2085 	tx = dmu_tx_create(zfsvfs->z_os);
2086 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2087 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2088 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2089 	zfs_sa_upgrade_txholds(tx, zp);
2090 	zfs_sa_upgrade_txholds(tx, dzp);
2091 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2092 	if (error) {
2093 		rw_exit(&zp->z_parent_lock);
2094 		rw_exit(&zp->z_name_lock);
2095 		zfs_dirent_unlock(dl);
2096 		VN_RELE(vp);
2097 		if (error == ERESTART) {
2098 			waited = B_TRUE;
2099 			dmu_tx_wait(tx);
2100 			dmu_tx_abort(tx);
2101 			goto top;
2102 		}
2103 		dmu_tx_abort(tx);
2104 		ZFS_EXIT(zfsvfs);
2105 		return (error);
2106 	}
2107 
2108 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2109 
2110 	if (error == 0) {
2111 		uint64_t txtype = TX_RMDIR;
2112 		if (flags & FIGNORECASE)
2113 			txtype |= TX_CI;
2114 
2115 		rw_enter(&rz_zev_rwlock, RW_READER);
2116 		if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_znode_remove)
2117 			rz_zev_callbacks->rz_zev_znode_remove(dzp, zp, tx,
2118 			                                      name, txtype);
2119 		rw_exit(&rz_zev_rwlock);
2120 
2121 		zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2122 	}
2123 
2124 	dmu_tx_commit(tx);
2125 
2126 	rw_exit(&zp->z_parent_lock);
2127 	rw_exit(&zp->z_name_lock);
2128 out:
2129 	zfs_dirent_unlock(dl);
2130 
2131 	VN_RELE(vp);
2132 
2133 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2134 		zil_commit(zilog, 0);
2135 
2136 	ZFS_EXIT(zfsvfs);
2137 	return (error);
2138 }
2139 
2140 /*
2141  * Read as many directory entries as will fit into the provided
2142  * buffer from the given directory cursor position (specified in
2143  * the uio structure).
2144  *
2145  *	IN:	vp	- vnode of directory to read.
2146  *		uio	- structure supplying read location, range info,
2147  *			  and return buffer.
2148  *		cr	- credentials of caller.
2149  *		ct	- caller context
2150  *		flags	- case flags
2151  *
2152  *	OUT:	uio	- updated offset and range, buffer filled.
2153  *		eofp	- set to true if end-of-file detected.
2154  *
2155  *	RETURN:	0 on success, error code on failure.
2156  *
2157  * Timestamps:
2158  *	vp - atime updated
2159  *
2160  * Note that the low 4 bits of the cookie returned by zap is always zero.
2161  * This allows us to use the low range for "special" directory entries:
2162  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2163  * we use the offset 2 for the '.zfs' directory.
2164  */
2165 /* ARGSUSED */
2166 static int
2167 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2168     caller_context_t *ct, int flags)
2169 {
2170 	znode_t		*zp = VTOZ(vp);
2171 	iovec_t		*iovp;
2172 	edirent_t	*eodp;
2173 	dirent64_t	*odp;
2174 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2175 	objset_t	*os;
2176 	caddr_t		outbuf;
2177 	size_t		bufsize;
2178 	zap_cursor_t	zc;
2179 	zap_attribute_t	zap;
2180 	uint_t		bytes_wanted;
2181 	uint64_t	offset; /* must be unsigned; checks for < 1 */
2182 	uint64_t	parent;
2183 	int		local_eof;
2184 	int		outcount;
2185 	int		error;
2186 	uint8_t		prefetch;
2187 	boolean_t	check_sysattrs;
2188 
2189 	ZFS_ENTER(zfsvfs);
2190 	ZFS_VERIFY_ZP(zp);
2191 
2192 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2193 	    &parent, sizeof (parent))) != 0) {
2194 		ZFS_EXIT(zfsvfs);
2195 		return (error);
2196 	}
2197 
2198 	/*
2199 	 * If we are not given an eof variable,
2200 	 * use a local one.
2201 	 */
2202 	if (eofp == NULL)
2203 		eofp = &local_eof;
2204 
2205 	/*
2206 	 * Check for valid iov_len.
2207 	 */
2208 	if (uio->uio_iov->iov_len <= 0) {
2209 		ZFS_EXIT(zfsvfs);
2210 		return (SET_ERROR(EINVAL));
2211 	}
2212 
2213 	/*
2214 	 * Quit if directory has been removed (posix)
2215 	 */
2216 	if ((*eofp = zp->z_unlinked) != 0) {
2217 		ZFS_EXIT(zfsvfs);
2218 		return (0);
2219 	}
2220 
2221 	error = 0;
2222 	os = zfsvfs->z_os;
2223 	offset = uio->uio_loffset;
2224 	prefetch = zp->z_zn_prefetch;
2225 
2226 	/*
2227 	 * Initialize the iterator cursor.
2228 	 */
2229 	if (offset <= 3) {
2230 		/*
2231 		 * Start iteration from the beginning of the directory.
2232 		 */
2233 		zap_cursor_init(&zc, os, zp->z_id);
2234 	} else {
2235 		/*
2236 		 * The offset is a serialized cursor.
2237 		 */
2238 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2239 	}
2240 
2241 	/*
2242 	 * Get space to change directory entries into fs independent format.
2243 	 */
2244 	iovp = uio->uio_iov;
2245 	bytes_wanted = iovp->iov_len;
2246 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2247 		bufsize = bytes_wanted;
2248 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
2249 		odp = (struct dirent64 *)outbuf;
2250 	} else {
2251 		bufsize = bytes_wanted;
2252 		outbuf = NULL;
2253 		odp = (struct dirent64 *)iovp->iov_base;
2254 	}
2255 	eodp = (struct edirent *)odp;
2256 
2257 	/*
2258 	 * If this VFS supports the system attribute view interface; and
2259 	 * we're looking at an extended attribute directory; and we care
2260 	 * about normalization conflicts on this vfs; then we must check
2261 	 * for normalization conflicts with the sysattr name space.
2262 	 */
2263 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2264 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2265 	    (flags & V_RDDIR_ENTFLAGS);
2266 
2267 	/*
2268 	 * Transform to file-system independent format
2269 	 */
2270 	outcount = 0;
2271 	while (outcount < bytes_wanted) {
2272 		ino64_t objnum;
2273 		ushort_t reclen;
2274 		off64_t *next = NULL;
2275 
2276 		/*
2277 		 * Special case `.', `..', and `.zfs'.
2278 		 */
2279 		if (offset == 0) {
2280 			(void) strcpy(zap.za_name, ".");
2281 			zap.za_normalization_conflict = 0;
2282 			objnum = zp->z_id;
2283 		} else if (offset == 1) {
2284 			(void) strcpy(zap.za_name, "..");
2285 			zap.za_normalization_conflict = 0;
2286 			objnum = parent;
2287 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2288 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2289 			zap.za_normalization_conflict = 0;
2290 			objnum = ZFSCTL_INO_ROOT;
2291 		} else {
2292 			/*
2293 			 * Grab next entry.
2294 			 */
2295 			if (error = zap_cursor_retrieve(&zc, &zap)) {
2296 				if ((*eofp = (error == ENOENT)) != 0)
2297 					break;
2298 				else
2299 					goto update;
2300 			}
2301 
2302 			if (zap.za_integer_length != 8 ||
2303 			    zap.za_num_integers != 1) {
2304 				cmn_err(CE_WARN, "zap_readdir: bad directory "
2305 				    "entry, obj = %lld, offset = %lld\n",
2306 				    (u_longlong_t)zp->z_id,
2307 				    (u_longlong_t)offset);
2308 				error = SET_ERROR(ENXIO);
2309 				goto update;
2310 			}
2311 
2312 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2313 			/*
2314 			 * MacOS X can extract the object type here such as:
2315 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2316 			 */
2317 
2318 			if (check_sysattrs && !zap.za_normalization_conflict) {
2319 				zap.za_normalization_conflict =
2320 				    xattr_sysattr_casechk(zap.za_name);
2321 			}
2322 		}
2323 
2324 		if (flags & V_RDDIR_ACCFILTER) {
2325 			/*
2326 			 * If we have no access at all, don't include
2327 			 * this entry in the returned information
2328 			 */
2329 			znode_t	*ezp;
2330 			if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2331 				goto skip_entry;
2332 			if (!zfs_has_access(ezp, cr)) {
2333 				VN_RELE(ZTOV(ezp));
2334 				goto skip_entry;
2335 			}
2336 			VN_RELE(ZTOV(ezp));
2337 		}
2338 
2339 		if (flags & V_RDDIR_ENTFLAGS)
2340 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2341 		else
2342 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2343 
2344 		/*
2345 		 * Will this entry fit in the buffer?
2346 		 */
2347 		if (outcount + reclen > bufsize) {
2348 			/*
2349 			 * Did we manage to fit anything in the buffer?
2350 			 */
2351 			if (!outcount) {
2352 				error = SET_ERROR(EINVAL);
2353 				goto update;
2354 			}
2355 			break;
2356 		}
2357 		if (flags & V_RDDIR_ENTFLAGS) {
2358 			/*
2359 			 * Add extended flag entry:
2360 			 */
2361 			eodp->ed_ino = objnum;
2362 			eodp->ed_reclen = reclen;
2363 			/* NOTE: ed_off is the offset for the *next* entry */
2364 			next = &(eodp->ed_off);
2365 			eodp->ed_eflags = zap.za_normalization_conflict ?
2366 			    ED_CASE_CONFLICT : 0;
2367 			(void) strncpy(eodp->ed_name, zap.za_name,
2368 			    EDIRENT_NAMELEN(reclen));
2369 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2370 		} else {
2371 			/*
2372 			 * Add normal entry:
2373 			 */
2374 			odp->d_ino = objnum;
2375 			odp->d_reclen = reclen;
2376 			/* NOTE: d_off is the offset for the *next* entry */
2377 			next = &(odp->d_off);
2378 			(void) strncpy(odp->d_name, zap.za_name,
2379 			    DIRENT64_NAMELEN(reclen));
2380 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2381 		}
2382 		outcount += reclen;
2383 
2384 		ASSERT(outcount <= bufsize);
2385 
2386 		/* Prefetch znode */
2387 		if (prefetch)
2388 			dmu_prefetch(os, objnum, 0, 0);
2389 
2390 	skip_entry:
2391 		/*
2392 		 * Move to the next entry, fill in the previous offset.
2393 		 */
2394 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2395 			zap_cursor_advance(&zc);
2396 			offset = zap_cursor_serialize(&zc);
2397 		} else {
2398 			offset += 1;
2399 		}
2400 		if (next)
2401 			*next = offset;
2402 	}
2403 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2404 
2405 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2406 		iovp->iov_base += outcount;
2407 		iovp->iov_len -= outcount;
2408 		uio->uio_resid -= outcount;
2409 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2410 		/*
2411 		 * Reset the pointer.
2412 		 */
2413 		offset = uio->uio_loffset;
2414 	}
2415 
2416 update:
2417 	zap_cursor_fini(&zc);
2418 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2419 		kmem_free(outbuf, bufsize);
2420 
2421 	if (error == ENOENT)
2422 		error = 0;
2423 
2424 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2425 
2426 	uio->uio_loffset = offset;
2427 	ZFS_EXIT(zfsvfs);
2428 	return (error);
2429 }
2430 
2431 ulong_t zfs_fsync_sync_cnt = 4;
2432 
2433 static int
2434 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2435 {
2436 	znode_t	*zp = VTOZ(vp);
2437 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2438 
2439 	/*
2440 	 * Regardless of whether this is required for standards conformance,
2441 	 * this is the logical behavior when fsync() is called on a file with
2442 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2443 	 * going to be pushed out as part of the zil_commit().
2444 	 */
2445 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2446 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2447 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2448 
2449 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2450 
2451 	if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2452 		ZFS_ENTER(zfsvfs);
2453 		ZFS_VERIFY_ZP(zp);
2454 		zil_commit(zfsvfs->z_log, zp->z_id);
2455 		ZFS_EXIT(zfsvfs);
2456 	}
2457 	return (0);
2458 }
2459 
2460 
2461 /*
2462  * Get the requested file attributes and place them in the provided
2463  * vattr structure.
2464  *
2465  *	IN:	vp	- vnode of file.
2466  *		vap	- va_mask identifies requested attributes.
2467  *			  If AT_XVATTR set, then optional attrs are requested
2468  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2469  *		cr	- credentials of caller.
2470  *		ct	- caller context
2471  *
2472  *	OUT:	vap	- attribute values.
2473  *
2474  *	RETURN:	0 (always succeeds).
2475  */
2476 /* ARGSUSED */
2477 static int
2478 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2479     caller_context_t *ct)
2480 {
2481 	znode_t *zp = VTOZ(vp);
2482 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2483 	int	error = 0;
2484 	uint64_t links;
2485 	uint64_t mtime[2], ctime[2];
2486 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2487 	xoptattr_t *xoap = NULL;
2488 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2489 	sa_bulk_attr_t bulk[2];
2490 	int count = 0;
2491 
2492 	ZFS_ENTER(zfsvfs);
2493 	ZFS_VERIFY_ZP(zp);
2494 
2495 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2496 
2497 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2498 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2499 
2500 	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2501 		ZFS_EXIT(zfsvfs);
2502 		return (error);
2503 	}
2504 
2505 	/*
2506 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2507 	 * Also, if we are the owner don't bother, since owner should
2508 	 * always be allowed to read basic attributes of file.
2509 	 */
2510 	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2511 	    (vap->va_uid != crgetuid(cr))) {
2512 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2513 		    skipaclchk, cr)) {
2514 			ZFS_EXIT(zfsvfs);
2515 			return (error);
2516 		}
2517 	}
2518 
2519 	/*
2520 	 * Return all attributes.  It's cheaper to provide the answer
2521 	 * than to determine whether we were asked the question.
2522 	 */
2523 
2524 	mutex_enter(&zp->z_lock);
2525 	vap->va_type = vp->v_type;
2526 	vap->va_mode = zp->z_mode & MODEMASK;
2527 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2528 	vap->va_nodeid = zp->z_id;
2529 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2530 		links = zp->z_links + 1;
2531 	else
2532 		links = zp->z_links;
2533 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2534 	vap->va_size = zp->z_size;
2535 	vap->va_rdev = vp->v_rdev;
2536 	vap->va_seq = zp->z_seq;
2537 
2538 	/*
2539 	 * Add in any requested optional attributes and the create time.
2540 	 * Also set the corresponding bits in the returned attribute bitmap.
2541 	 */
2542 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2543 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2544 			xoap->xoa_archive =
2545 			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2546 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2547 		}
2548 
2549 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2550 			xoap->xoa_readonly =
2551 			    ((zp->z_pflags & ZFS_READONLY) != 0);
2552 			XVA_SET_RTN(xvap, XAT_READONLY);
2553 		}
2554 
2555 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2556 			xoap->xoa_system =
2557 			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2558 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2559 		}
2560 
2561 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2562 			xoap->xoa_hidden =
2563 			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2564 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2565 		}
2566 
2567 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2568 			xoap->xoa_nounlink =
2569 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2570 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2571 		}
2572 
2573 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2574 			xoap->xoa_immutable =
2575 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2576 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2577 		}
2578 
2579 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2580 			xoap->xoa_appendonly =
2581 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2582 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2583 		}
2584 
2585 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2586 			xoap->xoa_nodump =
2587 			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2588 			XVA_SET_RTN(xvap, XAT_NODUMP);
2589 		}
2590 
2591 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2592 			xoap->xoa_opaque =
2593 			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2594 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2595 		}
2596 
2597 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2598 			xoap->xoa_av_quarantined =
2599 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2600 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2601 		}
2602 
2603 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2604 			xoap->xoa_av_modified =
2605 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2606 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2607 		}
2608 
2609 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2610 		    vp->v_type == VREG) {
2611 			zfs_sa_get_scanstamp(zp, xvap);
2612 		}
2613 
2614 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2615 			uint64_t times[2];
2616 
2617 			(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2618 			    times, sizeof (times));
2619 			ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2620 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2621 		}
2622 
2623 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2624 			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2625 			XVA_SET_RTN(xvap, XAT_REPARSE);
2626 		}
2627 		if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2628 			xoap->xoa_generation = zp->z_gen;
2629 			XVA_SET_RTN(xvap, XAT_GEN);
2630 		}
2631 
2632 		if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2633 			xoap->xoa_offline =
2634 			    ((zp->z_pflags & ZFS_OFFLINE) != 0);
2635 			XVA_SET_RTN(xvap, XAT_OFFLINE);
2636 		}
2637 
2638 		if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2639 			xoap->xoa_sparse =
2640 			    ((zp->z_pflags & ZFS_SPARSE) != 0);
2641 			XVA_SET_RTN(xvap, XAT_SPARSE);
2642 		}
2643 	}
2644 
2645 	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2646 	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2647 	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2648 
2649 	mutex_exit(&zp->z_lock);
2650 
2651 	sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2652 
2653 	if (zp->z_blksz == 0) {
2654 		/*
2655 		 * Block size hasn't been set; suggest maximal I/O transfers.
2656 		 */
2657 		vap->va_blksize = zfsvfs->z_max_blksz;
2658 	}
2659 
2660 	ZFS_EXIT(zfsvfs);
2661 	return (0);
2662 }
2663 
2664 /*
2665  * Set the file attributes to the values contained in the
2666  * vattr structure.
2667  *
2668  *	IN:	vp	- vnode of file to be modified.
2669  *		vap	- new attribute values.
2670  *			  If AT_XVATTR set, then optional attrs are being set
2671  *		flags	- ATTR_UTIME set if non-default time values provided.
2672  *			- ATTR_NOACLCHECK (CIFS context only).
2673  *		cr	- credentials of caller.
2674  *		ct	- caller context
2675  *
2676  *	RETURN:	0 on success, error code on failure.
2677  *
2678  * Timestamps:
2679  *	vp - ctime updated, mtime updated if size changed.
2680  */
2681 /* ARGSUSED */
2682 static int
2683 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2684     caller_context_t *ct)
2685 {
2686 	znode_t		*zp = VTOZ(vp);
2687 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2688 	zilog_t		*zilog;
2689 	dmu_tx_t	*tx;
2690 	vattr_t		oldva;
2691 	xvattr_t	tmpxvattr;
2692 	uint_t		mask = vap->va_mask;
2693 	uint_t		saved_mask = 0;
2694 	int		trim_mask = 0;
2695 	uint64_t	new_mode;
2696 	uint64_t	new_uid, new_gid;
2697 	uint64_t	xattr_obj;
2698 	uint64_t	mtime[2], ctime[2];
2699 	znode_t		*attrzp;
2700 	int		need_policy = FALSE;
2701 	int		err, err2;
2702 	zfs_fuid_info_t *fuidp = NULL;
2703 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2704 	xoptattr_t	*xoap;
2705 	zfs_acl_t	*aclp;
2706 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2707 	boolean_t	fuid_dirtied = B_FALSE;
2708 	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
2709 	int		count = 0, xattr_count = 0;
2710 
2711 	if (mask == 0)
2712 		return (0);
2713 
2714 	if (mask & AT_NOSET)
2715 		return (SET_ERROR(EINVAL));
2716 
2717 	ZFS_ENTER(zfsvfs);
2718 	ZFS_VERIFY_ZP(zp);
2719 
2720 	zilog = zfsvfs->z_log;
2721 
2722 	/*
2723 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2724 	 * that file system is at proper version level
2725 	 */
2726 
2727 	if (zfsvfs->z_use_fuids == B_FALSE &&
2728 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2729 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2730 	    (mask & AT_XVATTR))) {
2731 		ZFS_EXIT(zfsvfs);
2732 		return (SET_ERROR(EINVAL));
2733 	}
2734 
2735 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2736 		ZFS_EXIT(zfsvfs);
2737 		return (SET_ERROR(EISDIR));
2738 	}
2739 
2740 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2741 		ZFS_EXIT(zfsvfs);
2742 		return (SET_ERROR(EINVAL));
2743 	}
2744 
2745 	/*
2746 	 * If this is an xvattr_t, then get a pointer to the structure of
2747 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2748 	 */
2749 	xoap = xva_getxoptattr(xvap);
2750 
2751 	xva_init(&tmpxvattr);
2752 
2753 	/*
2754 	 * Immutable files can only alter immutable bit and atime
2755 	 */
2756 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2757 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2758 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2759 		ZFS_EXIT(zfsvfs);
2760 		return (SET_ERROR(EPERM));
2761 	}
2762 
2763 	if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2764 		ZFS_EXIT(zfsvfs);
2765 		return (SET_ERROR(EPERM));
2766 	}
2767 
2768 	/*
2769 	 * Verify timestamps doesn't overflow 32 bits.
2770 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2771 	 * handle times greater than 2039.  This check should be removed
2772 	 * once large timestamps are fully supported.
2773 	 */
2774 	if (mask & (AT_ATIME | AT_MTIME)) {
2775 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2776 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2777 			ZFS_EXIT(zfsvfs);
2778 			return (SET_ERROR(EOVERFLOW));
2779 		}
2780 	}
2781 
2782 top:
2783 	attrzp = NULL;
2784 	aclp = NULL;
2785 
2786 	/* Can this be moved to before the top label? */
2787 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2788 		ZFS_EXIT(zfsvfs);
2789 		return (SET_ERROR(EROFS));
2790 	}
2791 
2792 	/*
2793 	 * First validate permissions
2794 	 */
2795 
2796 	if (mask & AT_SIZE) {
2797 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2798 		if (err) {
2799 			ZFS_EXIT(zfsvfs);
2800 			return (err);
2801 		}
2802 		/*
2803 		 * XXX - Note, we are not providing any open
2804 		 * mode flags here (like FNDELAY), so we may
2805 		 * block if there are locks present... this
2806 		 * should be addressed in openat().
2807 		 */
2808 		/* XXX - would it be OK to generate a log record here? */
2809 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2810 		if (err) {
2811 			ZFS_EXIT(zfsvfs);
2812 			return (err);
2813 		}
2814 
2815 		if (vap->va_size == 0)
2816 			vnevent_truncate(ZTOV(zp), ct);
2817 	}
2818 
2819 	if (mask & (AT_ATIME|AT_MTIME) ||
2820 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2821 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2822 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2823 	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2824 	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2825 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2826 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2827 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2828 		    skipaclchk, cr);
2829 	}
2830 
2831 	if (mask & (AT_UID|AT_GID)) {
2832 		int	idmask = (mask & (AT_UID|AT_GID));
2833 		int	take_owner;
2834 		int	take_group;
2835 
2836 		/*
2837 		 * NOTE: even if a new mode is being set,
2838 		 * we may clear S_ISUID/S_ISGID bits.
2839 		 */
2840 
2841 		if (!(mask & AT_MODE))
2842 			vap->va_mode = zp->z_mode;
2843 
2844 		/*
2845 		 * Take ownership or chgrp to group we are a member of
2846 		 */
2847 
2848 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2849 		take_group = (mask & AT_GID) &&
2850 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2851 
2852 		/*
2853 		 * If both AT_UID and AT_GID are set then take_owner and
2854 		 * take_group must both be set in order to allow taking
2855 		 * ownership.
2856 		 *
2857 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2858 		 *
2859 		 */
2860 
2861 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2862 		    ((idmask == AT_UID) && take_owner) ||
2863 		    ((idmask == AT_GID) && take_group)) {
2864 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2865 			    skipaclchk, cr) == 0) {
2866 				/*
2867 				 * Remove setuid/setgid for non-privileged users
2868 				 */
2869 				secpolicy_setid_clear(vap, cr);
2870 				trim_mask = (mask & (AT_UID|AT_GID));
2871 			} else {
2872 				need_policy =  TRUE;
2873 			}
2874 		} else {
2875 			need_policy =  TRUE;
2876 		}
2877 	}
2878 
2879 	mutex_enter(&zp->z_lock);
2880 	oldva.va_mode = zp->z_mode;
2881 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2882 	if (mask & AT_XVATTR) {
2883 		/*
2884 		 * Update xvattr mask to include only those attributes
2885 		 * that are actually changing.
2886 		 *
2887 		 * the bits will be restored prior to actually setting
2888 		 * the attributes so the caller thinks they were set.
2889 		 */
2890 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2891 			if (xoap->xoa_appendonly !=
2892 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2893 				need_policy = TRUE;
2894 			} else {
2895 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2896 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2897 			}
2898 		}
2899 
2900 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2901 			if (xoap->xoa_nounlink !=
2902 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2903 				need_policy = TRUE;
2904 			} else {
2905 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2906 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2907 			}
2908 		}
2909 
2910 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2911 			if (xoap->xoa_immutable !=
2912 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2913 				need_policy = TRUE;
2914 			} else {
2915 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2916 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2917 			}
2918 		}
2919 
2920 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2921 			if (xoap->xoa_nodump !=
2922 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2923 				need_policy = TRUE;
2924 			} else {
2925 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2926 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2927 			}
2928 		}
2929 
2930 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2931 			if (xoap->xoa_av_modified !=
2932 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2933 				need_policy = TRUE;
2934 			} else {
2935 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2936 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2937 			}
2938 		}
2939 
2940 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2941 			if ((vp->v_type != VREG &&
2942 			    xoap->xoa_av_quarantined) ||
2943 			    xoap->xoa_av_quarantined !=
2944 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2945 				need_policy = TRUE;
2946 			} else {
2947 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2948 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2949 			}
2950 		}
2951 
2952 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2953 			mutex_exit(&zp->z_lock);
2954 			ZFS_EXIT(zfsvfs);
2955 			return (SET_ERROR(EPERM));
2956 		}
2957 
2958 		if (need_policy == FALSE &&
2959 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2960 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2961 			need_policy = TRUE;
2962 		}
2963 	}
2964 
2965 	mutex_exit(&zp->z_lock);
2966 
2967 	if (mask & AT_MODE) {
2968 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2969 			err = secpolicy_setid_setsticky_clear(vp, vap,
2970 			    &oldva, cr);
2971 			if (err) {
2972 				ZFS_EXIT(zfsvfs);
2973 				return (err);
2974 			}
2975 			trim_mask |= AT_MODE;
2976 		} else {
2977 			need_policy = TRUE;
2978 		}
2979 	}
2980 
2981 	if (need_policy) {
2982 		/*
2983 		 * If trim_mask is set then take ownership
2984 		 * has been granted or write_acl is present and user
2985 		 * has the ability to modify mode.  In that case remove
2986 		 * UID|GID and or MODE from mask so that
2987 		 * secpolicy_vnode_setattr() doesn't revoke it.
2988 		 */
2989 
2990 		if (trim_mask) {
2991 			saved_mask = vap->va_mask;
2992 			vap->va_mask &= ~trim_mask;
2993 		}
2994 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2995 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2996 		if (err) {
2997 			ZFS_EXIT(zfsvfs);
2998 			return (err);
2999 		}
3000 
3001 		if (trim_mask)
3002 			vap->va_mask |= saved_mask;
3003 	}
3004 
3005 	/*
3006 	 * secpolicy_vnode_setattr, or take ownership may have
3007 	 * changed va_mask
3008 	 */
3009 	mask = vap->va_mask;
3010 
3011 	if ((mask & (AT_UID | AT_GID))) {
3012 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3013 		    &xattr_obj, sizeof (xattr_obj));
3014 
3015 		if (err == 0 && xattr_obj) {
3016 			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3017 			if (err)
3018 				goto out2;
3019 		}
3020 		if (mask & AT_UID) {
3021 			new_uid = zfs_fuid_create(zfsvfs,
3022 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3023 			if (new_uid != zp->z_uid &&
3024 			    zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3025 				if (attrzp)
3026 					VN_RELE(ZTOV(attrzp));
3027 				err = SET_ERROR(EDQUOT);
3028 				goto out2;
3029 			}
3030 		}
3031 
3032 		if (mask & AT_GID) {
3033 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3034 			    cr, ZFS_GROUP, &fuidp);
3035 			if (new_gid != zp->z_gid &&
3036 			    zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3037 				if (attrzp)
3038 					VN_RELE(ZTOV(attrzp));
3039 				err = SET_ERROR(EDQUOT);
3040 				goto out2;
3041 			}
3042 		}
3043 	}
3044 	tx = dmu_tx_create(zfsvfs->z_os);
3045 
3046 	if (mask & AT_MODE) {
3047 		uint64_t pmode = zp->z_mode;
3048 		uint64_t acl_obj;
3049 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3050 
3051 		if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3052 		    !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3053 			err = SET_ERROR(EPERM);
3054 			goto out;
3055 		}
3056 
3057 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3058 			goto out;
3059 
3060 		mutex_enter(&zp->z_lock);
3061 		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3062 			/*
3063 			 * Are we upgrading ACL from old V0 format
3064 			 * to V1 format?
3065 			 */
3066 			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3067 			    zfs_znode_acl_version(zp) ==
3068 			    ZFS_ACL_VERSION_INITIAL) {
3069 				dmu_tx_hold_free(tx, acl_obj, 0,
3070 				    DMU_OBJECT_END);
3071 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3072 				    0, aclp->z_acl_bytes);
3073 			} else {
3074 				dmu_tx_hold_write(tx, acl_obj, 0,
3075 				    aclp->z_acl_bytes);
3076 			}
3077 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3078 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3079 			    0, aclp->z_acl_bytes);
3080 		}
3081 		mutex_exit(&zp->z_lock);
3082 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3083 	} else {
3084 		if ((mask & AT_XVATTR) &&
3085 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3086 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3087 		else
3088 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3089 	}
3090 
3091 	if (attrzp) {
3092 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3093 	}
3094 
3095 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3096 	if (fuid_dirtied)
3097 		zfs_fuid_txhold(zfsvfs, tx);
3098 
3099 	zfs_sa_upgrade_txholds(tx, zp);
3100 
3101 	err = dmu_tx_assign(tx, TXG_WAIT);
3102 	if (err)
3103 		goto out;
3104 
3105 	count = 0;
3106 	/*
3107 	 * Set each attribute requested.
3108 	 * We group settings according to the locks they need to acquire.
3109 	 *
3110 	 * Note: you cannot set ctime directly, although it will be
3111 	 * updated as a side-effect of calling this function.
3112 	 */
3113 
3114 
3115 	if (mask & (AT_UID|AT_GID|AT_MODE))
3116 		mutex_enter(&zp->z_acl_lock);
3117 	mutex_enter(&zp->z_lock);
3118 
3119 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3120 	    &zp->z_pflags, sizeof (zp->z_pflags));
3121 
3122 	if (attrzp) {
3123 		if (mask & (AT_UID|AT_GID|AT_MODE))
3124 			mutex_enter(&attrzp->z_acl_lock);
3125 		mutex_enter(&attrzp->z_lock);
3126 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3127 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3128 		    sizeof (attrzp->z_pflags));
3129 	}
3130 
3131 	if (mask & (AT_UID|AT_GID)) {
3132 
3133 		if (mask & AT_UID) {
3134 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3135 			    &new_uid, sizeof (new_uid));
3136 			zp->z_uid = new_uid;
3137 			if (attrzp) {
3138 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3139 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3140 				    sizeof (new_uid));
3141 				attrzp->z_uid = new_uid;
3142 			}
3143 		}
3144 
3145 		if (mask & AT_GID) {
3146 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3147 			    NULL, &new_gid, sizeof (new_gid));
3148 			zp->z_gid = new_gid;
3149 			if (attrzp) {
3150 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3151 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3152 				    sizeof (new_gid));
3153 				attrzp->z_gid = new_gid;
3154 			}
3155 		}
3156 		if (!(mask & AT_MODE)) {
3157 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3158 			    NULL, &new_mode, sizeof (new_mode));
3159 			new_mode = zp->z_mode;
3160 		}
3161 		err = zfs_acl_chown_setattr(zp);
3162 		ASSERT(err == 0);
3163 		if (attrzp) {
3164 			err = zfs_acl_chown_setattr(attrzp);
3165 			ASSERT(err == 0);
3166 		}
3167 	}
3168 
3169 	if (mask & AT_MODE) {
3170 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3171 		    &new_mode, sizeof (new_mode));
3172 		zp->z_mode = new_mode;
3173 		ASSERT3U((uintptr_t)aclp, !=, NULL);
3174 		err = zfs_aclset_common(zp, aclp, cr, tx);
3175 		ASSERT0(err);
3176 		if (zp->z_acl_cached)
3177 			zfs_acl_free(zp->z_acl_cached);
3178 		zp->z_acl_cached = aclp;
3179 		aclp = NULL;
3180 	}
3181 
3182 
3183 	if (mask & AT_ATIME) {
3184 		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3185 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3186 		    &zp->z_atime, sizeof (zp->z_atime));
3187 	}
3188 
3189 	if (mask & AT_MTIME) {
3190 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3191 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3192 		    mtime, sizeof (mtime));
3193 	}
3194 
3195 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3196 	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3197 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3198 		    NULL, mtime, sizeof (mtime));
3199 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3200 		    &ctime, sizeof (ctime));
3201 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3202 		    B_TRUE);
3203 	} else if (mask != 0) {
3204 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3205 		    &ctime, sizeof (ctime));
3206 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3207 		    B_TRUE);
3208 		if (attrzp) {
3209 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3210 			    SA_ZPL_CTIME(zfsvfs), NULL,
3211 			    &ctime, sizeof (ctime));
3212 			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3213 			    mtime, ctime, B_TRUE);
3214 		}
3215 	}
3216 	/*
3217 	 * Do this after setting timestamps to prevent timestamp
3218 	 * update from toggling bit
3219 	 */
3220 
3221 	if (xoap && (mask & AT_XVATTR)) {
3222 
3223 		/*
3224 		 * restore trimmed off masks
3225 		 * so that return masks can be set for caller.
3226 		 */
3227 
3228 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3229 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
3230 		}
3231 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3232 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
3233 		}
3234 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3235 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3236 		}
3237 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3238 			XVA_SET_REQ(xvap, XAT_NODUMP);
3239 		}
3240 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3241 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3242 		}
3243 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3244 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3245 		}
3246 
3247 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3248 			ASSERT(vp->v_type == VREG);
3249 
3250 		zfs_xvattr_set(zp, xvap, tx);
3251 	}
3252 
3253 	if (fuid_dirtied)
3254 		zfs_fuid_sync(zfsvfs, tx);
3255 
3256 	if (mask != 0)
3257 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3258 
3259 	mutex_exit(&zp->z_lock);
3260 	if (mask & (AT_UID|AT_GID|AT_MODE))
3261 		mutex_exit(&zp->z_acl_lock);
3262 
3263 	if (attrzp) {
3264 		if (mask & (AT_UID|AT_GID|AT_MODE))
3265 			mutex_exit(&attrzp->z_acl_lock);
3266 		mutex_exit(&attrzp->z_lock);
3267 	}
3268 out:
3269 	if (err == 0 && attrzp) {
3270 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3271 		    xattr_count, tx);
3272 		ASSERT(err2 == 0);
3273 	}
3274 
3275 	if (attrzp)
3276 		VN_RELE(ZTOV(attrzp));
3277 
3278 	if (aclp)
3279 		zfs_acl_free(aclp);
3280 
3281 	if (fuidp) {
3282 		zfs_fuid_info_free(fuidp);
3283 		fuidp = NULL;
3284 	}
3285 
3286 	if (err) {
3287 		dmu_tx_abort(tx);
3288 		if (err == ERESTART)
3289 			goto top;
3290 	} else {
3291 		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3292 		rw_enter(&rz_zev_rwlock, RW_READER);
3293 		if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_znode_setattr)
3294 			rz_zev_callbacks->rz_zev_znode_setattr(zp, tx);
3295 		rw_exit(&rz_zev_rwlock);
3296 		dmu_tx_commit(tx);
3297 	}
3298 
3299 out2:
3300 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3301 		zil_commit(zilog, 0);
3302 
3303 	ZFS_EXIT(zfsvfs);
3304 	return (err);
3305 }
3306 
3307 typedef struct zfs_zlock {
3308 	krwlock_t	*zl_rwlock;	/* lock we acquired */
3309 	znode_t		*zl_znode;	/* znode we held */
3310 	struct zfs_zlock *zl_next;	/* next in list */
3311 } zfs_zlock_t;
3312 
3313 /*
3314  * Drop locks and release vnodes that were held by zfs_rename_lock().
3315  */
3316 static void
3317 zfs_rename_unlock(zfs_zlock_t **zlpp)
3318 {
3319 	zfs_zlock_t *zl;
3320 
3321 	while ((zl = *zlpp) != NULL) {
3322 		if (zl->zl_znode != NULL)
3323 			VN_RELE(ZTOV(zl->zl_znode));
3324 		rw_exit(zl->zl_rwlock);
3325 		*zlpp = zl->zl_next;
3326 		kmem_free(zl, sizeof (*zl));
3327 	}
3328 }
3329 
3330 /*
3331  * Search back through the directory tree, using the ".." entries.
3332  * Lock each directory in the chain to prevent concurrent renames.
3333  * Fail any attempt to move a directory into one of its own descendants.
3334  * XXX - z_parent_lock can overlap with map or grow locks
3335  */
3336 static int
3337 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3338 {
3339 	zfs_zlock_t	*zl;
3340 	znode_t		*zp = tdzp;
3341 	uint64_t	rootid = zp->z_zfsvfs->z_root;
3342 	uint64_t	oidp = zp->z_id;
3343 	krwlock_t	*rwlp = &szp->z_parent_lock;
3344 	krw_t		rw = RW_WRITER;
3345 
3346 	/*
3347 	 * First pass write-locks szp and compares to zp->z_id.
3348 	 * Later passes read-lock zp and compare to zp->z_parent.
3349 	 */
3350 	do {
3351 		if (!rw_tryenter(rwlp, rw)) {
3352 			/*
3353 			 * Another thread is renaming in this path.
3354 			 * Note that if we are a WRITER, we don't have any
3355 			 * parent_locks held yet.
3356 			 */
3357 			if (rw == RW_READER && zp->z_id > szp->z_id) {
3358 				/*
3359 				 * Drop our locks and restart
3360 				 */
3361 				zfs_rename_unlock(&zl);
3362 				*zlpp = NULL;
3363 				zp = tdzp;
3364 				oidp = zp->z_id;
3365 				rwlp = &szp->z_parent_lock;
3366 				rw = RW_WRITER;
3367 				continue;
3368 			} else {
3369 				/*
3370 				 * Wait for other thread to drop its locks
3371 				 */
3372 				rw_enter(rwlp, rw);
3373 			}
3374 		}
3375 
3376 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3377 		zl->zl_rwlock = rwlp;
3378 		zl->zl_znode = NULL;
3379 		zl->zl_next = *zlpp;
3380 		*zlpp = zl;
3381 
3382 		if (oidp == szp->z_id)		/* We're a descendant of szp */
3383 			return (SET_ERROR(EINVAL));
3384 
3385 		if (oidp == rootid)		/* We've hit the top */
3386 			return (0);
3387 
3388 		if (rw == RW_READER) {		/* i.e. not the first pass */
3389 			int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3390 			if (error)
3391 				return (error);
3392 			zl->zl_znode = zp;
3393 		}
3394 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3395 		    &oidp, sizeof (oidp));
3396 		rwlp = &zp->z_parent_lock;
3397 		rw = RW_READER;
3398 
3399 	} while (zp->z_id != sdzp->z_id);
3400 
3401 	return (0);
3402 }
3403 
3404 /*
3405  * Move an entry from the provided source directory to the target
3406  * directory.  Change the entry name as indicated.
3407  *
3408  *	IN:	sdvp	- Source directory containing the "old entry".
3409  *		snm	- Old entry name.
3410  *		tdvp	- Target directory to contain the "new entry".
3411  *		tnm	- New entry name.
3412  *		cr	- credentials of caller.
3413  *		ct	- caller context
3414  *		flags	- case flags
3415  *
3416  *	RETURN:	0 on success, error code on failure.
3417  *
3418  * Timestamps:
3419  *	sdvp,tdvp - ctime|mtime updated
3420  */
3421 /*ARGSUSED*/
3422 static int
3423 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3424     caller_context_t *ct, int flags)
3425 {
3426 	znode_t		*tdzp, *szp, *tzp;
3427 	znode_t		*sdzp = VTOZ(sdvp);
3428 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
3429 	zilog_t		*zilog;
3430 	vnode_t		*realvp;
3431 	zfs_dirlock_t	*sdl, *tdl;
3432 	dmu_tx_t	*tx;
3433 	zfs_zlock_t	*zl;
3434 	int		cmp, serr, terr;
3435 	int		error = 0;
3436 	int		zflg = 0;
3437 	boolean_t	waited = B_FALSE;
3438 
3439 	ZFS_ENTER(zfsvfs);
3440 	ZFS_VERIFY_ZP(sdzp);
3441 	zilog = zfsvfs->z_log;
3442 
3443 	/*
3444 	 * Make sure we have the real vp for the target directory.
3445 	 */
3446 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3447 		tdvp = realvp;
3448 
3449 	tdzp = VTOZ(tdvp);
3450 	ZFS_VERIFY_ZP(tdzp);
3451 
3452 	/*
3453 	 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3454 	 * ctldir appear to have the same v_vfsp.
3455 	 */
3456 	if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3457 		ZFS_EXIT(zfsvfs);
3458 		return (SET_ERROR(EXDEV));
3459 	}
3460 
3461 	if (zfsvfs->z_utf8 && u8_validate(tnm,
3462 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3463 		ZFS_EXIT(zfsvfs);
3464 		return (SET_ERROR(EILSEQ));
3465 	}
3466 
3467 	if (flags & FIGNORECASE)
3468 		zflg |= ZCILOOK;
3469 
3470 top:
3471 	szp = NULL;
3472 	tzp = NULL;
3473 	zl = NULL;
3474 
3475 	/*
3476 	 * This is to prevent the creation of links into attribute space
3477 	 * by renaming a linked file into/outof an attribute directory.
3478 	 * See the comment in zfs_link() for why this is considered bad.
3479 	 */
3480 	if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3481 		ZFS_EXIT(zfsvfs);
3482 		return (SET_ERROR(EINVAL));
3483 	}
3484 
3485 	/*
3486 	 * Lock source and target directory entries.  To prevent deadlock,
3487 	 * a lock ordering must be defined.  We lock the directory with
3488 	 * the smallest object id first, or if it's a tie, the one with
3489 	 * the lexically first name.
3490 	 */
3491 	if (sdzp->z_id < tdzp->z_id) {
3492 		cmp = -1;
3493 	} else if (sdzp->z_id > tdzp->z_id) {
3494 		cmp = 1;
3495 	} else {
3496 		/*
3497 		 * First compare the two name arguments without
3498 		 * considering any case folding.
3499 		 */
3500 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3501 
3502 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3503 		ASSERT(error == 0 || !zfsvfs->z_utf8);
3504 		if (cmp == 0) {
3505 			/*
3506 			 * POSIX: "If the old argument and the new argument
3507 			 * both refer to links to the same existing file,
3508 			 * the rename() function shall return successfully
3509 			 * and perform no other action."
3510 			 */
3511 			ZFS_EXIT(zfsvfs);
3512 			return (0);
3513 		}
3514 		/*
3515 		 * If the file system is case-folding, then we may
3516 		 * have some more checking to do.  A case-folding file
3517 		 * system is either supporting mixed case sensitivity
3518 		 * access or is completely case-insensitive.  Note
3519 		 * that the file system is always case preserving.
3520 		 *
3521 		 * In mixed sensitivity mode case sensitive behavior
3522 		 * is the default.  FIGNORECASE must be used to
3523 		 * explicitly request case insensitive behavior.
3524 		 *
3525 		 * If the source and target names provided differ only
3526 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
3527 		 * we will treat this as a special case in the
3528 		 * case-insensitive mode: as long as the source name
3529 		 * is an exact match, we will allow this to proceed as
3530 		 * a name-change request.
3531 		 */
3532 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3533 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
3534 		    flags & FIGNORECASE)) &&
3535 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3536 		    &error) == 0) {
3537 			/*
3538 			 * case preserving rename request, require exact
3539 			 * name matches
3540 			 */
3541 			zflg |= ZCIEXACT;
3542 			zflg &= ~ZCILOOK;
3543 		}
3544 	}
3545 
3546 	/*
3547 	 * If the source and destination directories are the same, we should
3548 	 * grab the z_name_lock of that directory only once.
3549 	 */
3550 	if (sdzp == tdzp) {
3551 		zflg |= ZHAVELOCK;
3552 		rw_enter(&sdzp->z_name_lock, RW_READER);
3553 	}
3554 
3555 	if (cmp < 0) {
3556 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3557 		    ZEXISTS | zflg, NULL, NULL);
3558 		terr = zfs_dirent_lock(&tdl,
3559 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3560 	} else {
3561 		terr = zfs_dirent_lock(&tdl,
3562 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3563 		serr = zfs_dirent_lock(&sdl,
3564 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3565 		    NULL, NULL);
3566 	}
3567 
3568 	if (serr) {
3569 		/*
3570 		 * Source entry invalid or not there.
3571 		 */
3572 		if (!terr) {
3573 			zfs_dirent_unlock(tdl);
3574 			if (tzp)
3575 				VN_RELE(ZTOV(tzp));
3576 		}
3577 
3578 		if (sdzp == tdzp)
3579 			rw_exit(&sdzp->z_name_lock);
3580 
3581 		if (strcmp(snm, "..") == 0)
3582 			serr = SET_ERROR(EINVAL);
3583 		ZFS_EXIT(zfsvfs);
3584 		return (serr);
3585 	}
3586 	if (terr) {
3587 		zfs_dirent_unlock(sdl);
3588 		VN_RELE(ZTOV(szp));
3589 
3590 		if (sdzp == tdzp)
3591 			rw_exit(&sdzp->z_name_lock);
3592 
3593 		if (strcmp(tnm, "..") == 0)
3594 			terr = SET_ERROR(EINVAL);
3595 		ZFS_EXIT(zfsvfs);
3596 		return (terr);
3597 	}
3598 
3599 	/*
3600 	 * Must have write access at the source to remove the old entry
3601 	 * and write access at the target to create the new entry.
3602 	 * Note that if target and source are the same, this can be
3603 	 * done in a single check.
3604 	 */
3605 
3606 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3607 		goto out;
3608 
3609 	if (ZTOV(szp)->v_type == VDIR) {
3610 		/*
3611 		 * Check to make sure rename is valid.
3612 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3613 		 */
3614 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3615 			goto out;
3616 	}
3617 
3618 	/*
3619 	 * Does target exist?
3620 	 */
3621 	if (tzp) {
3622 		/*
3623 		 * Source and target must be the same type.
3624 		 */
3625 		if (ZTOV(szp)->v_type == VDIR) {
3626 			if (ZTOV(tzp)->v_type != VDIR) {
3627 				error = SET_ERROR(ENOTDIR);
3628 				goto out;
3629 			}
3630 		} else {
3631 			if (ZTOV(tzp)->v_type == VDIR) {
3632 				error = SET_ERROR(EISDIR);
3633 				goto out;
3634 			}
3635 		}
3636 		/*
3637 		 * POSIX dictates that when the source and target
3638 		 * entries refer to the same file object, rename
3639 		 * must do nothing and exit without error.
3640 		 */
3641 		if (szp->z_id == tzp->z_id) {
3642 			error = 0;
3643 			goto out;
3644 		}
3645 	}
3646 
3647 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3648 	if (tzp)
3649 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3650 
3651 	/*
3652 	 * notify the target directory if it is not the same
3653 	 * as source directory.
3654 	 */
3655 	if (tdvp != sdvp) {
3656 		vnevent_rename_dest_dir(tdvp, ct);
3657 	}
3658 
3659 	tx = dmu_tx_create(zfsvfs->z_os);
3660 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3661 	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3662 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3663 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3664 	if (sdzp != tdzp) {
3665 		dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3666 		zfs_sa_upgrade_txholds(tx, tdzp);
3667 	}
3668 	if (tzp) {
3669 		dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3670 		zfs_sa_upgrade_txholds(tx, tzp);
3671 	}
3672 
3673 	zfs_sa_upgrade_txholds(tx, szp);
3674 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3675 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3676 	if (error) {
3677 		if (zl != NULL)
3678 			zfs_rename_unlock(&zl);
3679 		zfs_dirent_unlock(sdl);
3680 		zfs_dirent_unlock(tdl);
3681 
3682 		if (sdzp == tdzp)
3683 			rw_exit(&sdzp->z_name_lock);
3684 
3685 		VN_RELE(ZTOV(szp));
3686 		if (tzp)
3687 			VN_RELE(ZTOV(tzp));
3688 		if (error == ERESTART) {
3689 			waited = B_TRUE;
3690 			dmu_tx_wait(tx);
3691 			dmu_tx_abort(tx);
3692 			goto top;
3693 		}
3694 		dmu_tx_abort(tx);
3695 		ZFS_EXIT(zfsvfs);
3696 		return (error);
3697 	}
3698 
3699 	if (tzp)	/* Attempt to remove the existing target */
3700 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3701 
3702 	if (error == 0) {
3703 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3704 		if (error == 0) {
3705 			szp->z_pflags |= ZFS_AV_MODIFIED;
3706 
3707 			error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3708 			    (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3709 			ASSERT0(error);
3710 
3711 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3712 			if (error == 0) {
3713 				zfs_log_rename(zilog, tx, TX_RENAME |
3714 				    (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3715 				    sdl->dl_name, tdzp, tdl->dl_name, szp);
3716 
3717 				/*
3718 				 * Update path information for the target vnode
3719 				 */
3720 				vn_renamepath(tdvp, ZTOV(szp), tnm,
3721 				    strlen(tnm));
3722 			} else {
3723 				/*
3724 				 * At this point, we have successfully created
3725 				 * the target name, but have failed to remove
3726 				 * the source name.  Since the create was done
3727 				 * with the ZRENAMING flag, there are
3728 				 * complications; for one, the link count is
3729 				 * wrong.  The easiest way to deal with this
3730 				 * is to remove the newly created target, and
3731 				 * return the original error.  This must
3732 				 * succeed; fortunately, it is very unlikely to
3733 				 * fail, since we just created it.
3734 				 */
3735 				VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3736 				    ZRENAMING, NULL), ==, 0);
3737 			}
3738 		}
3739 	}
3740 
3741 	dmu_tx_commit(tx);
3742 out:
3743 	if (zl != NULL)
3744 		zfs_rename_unlock(&zl);
3745 
3746 	zfs_dirent_unlock(sdl);
3747 	zfs_dirent_unlock(tdl);
3748 
3749 	if (sdzp == tdzp)
3750 		rw_exit(&sdzp->z_name_lock);
3751 
3752 
3753 	VN_RELE(ZTOV(szp));
3754 	if (tzp)
3755 		VN_RELE(ZTOV(tzp));
3756 
3757 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3758 		zil_commit(zilog, 0);
3759 
3760 	ZFS_EXIT(zfsvfs);
3761 	return (error);
3762 }
3763 
3764 /*
3765  * Insert the indicated symbolic reference entry into the directory.
3766  *
3767  *	IN:	dvp	- Directory to contain new symbolic link.
3768  *		link	- Name for new symlink entry.
3769  *		vap	- Attributes of new entry.
3770  *		cr	- credentials of caller.
3771  *		ct	- caller context
3772  *		flags	- case flags
3773  *
3774  *	RETURN:	0 on success, error code on failure.
3775  *
3776  * Timestamps:
3777  *	dvp - ctime|mtime updated
3778  */
3779 /*ARGSUSED*/
3780 static int
3781 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3782     caller_context_t *ct, int flags)
3783 {
3784 	znode_t		*zp, *dzp = VTOZ(dvp);
3785 	zfs_dirlock_t	*dl;
3786 	dmu_tx_t	*tx;
3787 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3788 	zilog_t		*zilog;
3789 	uint64_t	len = strlen(link);
3790 	int		error;
3791 	int		zflg = ZNEW;
3792 	zfs_acl_ids_t	acl_ids;
3793 	boolean_t	fuid_dirtied;
3794 	uint64_t	txtype = TX_SYMLINK;
3795 	boolean_t	waited = B_FALSE;
3796 
3797 	ASSERT(vap->va_type == VLNK);
3798 
3799 	ZFS_ENTER(zfsvfs);
3800 	ZFS_VERIFY_ZP(dzp);
3801 	zilog = zfsvfs->z_log;
3802 
3803 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3804 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3805 		ZFS_EXIT(zfsvfs);
3806 		return (SET_ERROR(EILSEQ));
3807 	}
3808 	if (flags & FIGNORECASE)
3809 		zflg |= ZCILOOK;
3810 
3811 	if (len > MAXPATHLEN) {
3812 		ZFS_EXIT(zfsvfs);
3813 		return (SET_ERROR(ENAMETOOLONG));
3814 	}
3815 
3816 	if ((error = zfs_acl_ids_create(dzp, 0,
3817 	    vap, cr, NULL, &acl_ids)) != 0) {
3818 		ZFS_EXIT(zfsvfs);
3819 		return (error);
3820 	}
3821 top:
3822 	/*
3823 	 * Attempt to lock directory; fail if entry already exists.
3824 	 */
3825 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3826 	if (error) {
3827 		zfs_acl_ids_free(&acl_ids);
3828 		ZFS_EXIT(zfsvfs);
3829 		return (error);
3830 	}
3831 
3832 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3833 		zfs_acl_ids_free(&acl_ids);
3834 		zfs_dirent_unlock(dl);
3835 		ZFS_EXIT(zfsvfs);
3836 		return (error);
3837 	}
3838 
3839 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3840 		zfs_acl_ids_free(&acl_ids);
3841 		zfs_dirent_unlock(dl);
3842 		ZFS_EXIT(zfsvfs);
3843 		return (SET_ERROR(EDQUOT));
3844 	}
3845 	tx = dmu_tx_create(zfsvfs->z_os);
3846 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3847 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3848 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3849 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3850 	    ZFS_SA_BASE_ATTR_SIZE + len);
3851 	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3852 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3853 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3854 		    acl_ids.z_aclp->z_acl_bytes);
3855 	}
3856 	if (fuid_dirtied)
3857 		zfs_fuid_txhold(zfsvfs, tx);
3858 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3859 	if (error) {
3860 		zfs_dirent_unlock(dl);
3861 		if (error == ERESTART) {
3862 			waited = B_TRUE;
3863 			dmu_tx_wait(tx);
3864 			dmu_tx_abort(tx);
3865 			goto top;
3866 		}
3867 		zfs_acl_ids_free(&acl_ids);
3868 		dmu_tx_abort(tx);
3869 		ZFS_EXIT(zfsvfs);
3870 		return (error);
3871 	}
3872 
3873 	/*
3874 	 * Create a new object for the symlink.
3875 	 * for version 4 ZPL datsets the symlink will be an SA attribute
3876 	 */
3877 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3878 
3879 	if (fuid_dirtied)
3880 		zfs_fuid_sync(zfsvfs, tx);
3881 
3882 	mutex_enter(&zp->z_lock);
3883 	if (zp->z_is_sa)
3884 		error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3885 		    link, len, tx);
3886 	else
3887 		zfs_sa_symlink(zp, link, len, tx);
3888 	mutex_exit(&zp->z_lock);
3889 
3890 	zp->z_size = len;
3891 	(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3892 	    &zp->z_size, sizeof (zp->z_size), tx);
3893 	/*
3894 	 * Insert the new object into the directory.
3895 	 */
3896 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3897 
3898 	if (flags & FIGNORECASE)
3899 		txtype |= TX_CI;
3900 	zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3901 
3902 	zfs_acl_ids_free(&acl_ids);
3903 
3904 	dmu_tx_commit(tx);
3905 
3906 	zfs_dirent_unlock(dl);
3907 
3908 	VN_RELE(ZTOV(zp));
3909 
3910 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3911 		zil_commit(zilog, 0);
3912 
3913 	ZFS_EXIT(zfsvfs);
3914 	return (error);
3915 }
3916 
3917 /*
3918  * Return, in the buffer contained in the provided uio structure,
3919  * the symbolic path referred to by vp.
3920  *
3921  *	IN:	vp	- vnode of symbolic link.
3922  *		uio	- structure to contain the link path.
3923  *		cr	- credentials of caller.
3924  *		ct	- caller context
3925  *
3926  *	OUT:	uio	- structure containing the link path.
3927  *
3928  *	RETURN:	0 on success, error code on failure.
3929  *
3930  * Timestamps:
3931  *	vp - atime updated
3932  */
3933 /* ARGSUSED */
3934 static int
3935 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3936 {
3937 	znode_t		*zp = VTOZ(vp);
3938 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3939 	int		error;
3940 
3941 	ZFS_ENTER(zfsvfs);
3942 	ZFS_VERIFY_ZP(zp);
3943 
3944 	mutex_enter(&zp->z_lock);
3945 	if (zp->z_is_sa)
3946 		error = sa_lookup_uio(zp->z_sa_hdl,
3947 		    SA_ZPL_SYMLINK(zfsvfs), uio);
3948 	else
3949 		error = zfs_sa_readlink(zp, uio);
3950 	mutex_exit(&zp->z_lock);
3951 
3952 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3953 
3954 	ZFS_EXIT(zfsvfs);
3955 	return (error);
3956 }
3957 
3958 /*
3959  * Insert a new entry into directory tdvp referencing svp.
3960  *
3961  *	IN:	tdvp	- Directory to contain new entry.
3962  *		svp	- vnode of new entry.
3963  *		name	- name of new entry.
3964  *		cr	- credentials of caller.
3965  *		ct	- caller context
3966  *
3967  *	RETURN:	0 on success, error code on failure.
3968  *
3969  * Timestamps:
3970  *	tdvp - ctime|mtime updated
3971  *	 svp - ctime updated
3972  */
3973 /* ARGSUSED */
3974 static int
3975 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3976     caller_context_t *ct, int flags)
3977 {
3978 	znode_t		*dzp = VTOZ(tdvp);
3979 	znode_t		*tzp, *szp;
3980 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3981 	zilog_t		*zilog;
3982 	zfs_dirlock_t	*dl;
3983 	dmu_tx_t	*tx;
3984 	vnode_t		*realvp;
3985 	int		error;
3986 	int		zf = ZNEW;
3987 	uint64_t	parent;
3988 	uid_t		owner;
3989 	boolean_t	waited = B_FALSE;
3990 
3991 	ASSERT(tdvp->v_type == VDIR);
3992 
3993 	ZFS_ENTER(zfsvfs);
3994 	ZFS_VERIFY_ZP(dzp);
3995 	zilog = zfsvfs->z_log;
3996 
3997 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3998 		svp = realvp;
3999 
4000 	/*
4001 	 * POSIX dictates that we return EPERM here.
4002 	 * Better choices include ENOTSUP or EISDIR.
4003 	 */
4004 	if (svp->v_type == VDIR) {
4005 		ZFS_EXIT(zfsvfs);
4006 		return (SET_ERROR(EPERM));
4007 	}
4008 
4009 	szp = VTOZ(svp);
4010 	ZFS_VERIFY_ZP(szp);
4011 
4012 	/*
4013 	 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4014 	 * ctldir appear to have the same v_vfsp.
4015 	 */
4016 	if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4017 		ZFS_EXIT(zfsvfs);
4018 		return (SET_ERROR(EXDEV));
4019 	}
4020 
4021 	/* Prevent links to .zfs/shares files */
4022 
4023 	if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4024 	    &parent, sizeof (uint64_t))) != 0) {
4025 		ZFS_EXIT(zfsvfs);
4026 		return (error);
4027 	}
4028 	if (parent == zfsvfs->z_shares_dir) {
4029 		ZFS_EXIT(zfsvfs);
4030 		return (SET_ERROR(EPERM));
4031 	}
4032 
4033 	if (zfsvfs->z_utf8 && u8_validate(name,
4034 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4035 		ZFS_EXIT(zfsvfs);
4036 		return (SET_ERROR(EILSEQ));
4037 	}
4038 	if (flags & FIGNORECASE)
4039 		zf |= ZCILOOK;
4040 
4041 	/*
4042 	 * We do not support links between attributes and non-attributes
4043 	 * because of the potential security risk of creating links
4044 	 * into "normal" file space in order to circumvent restrictions
4045 	 * imposed in attribute space.
4046 	 */
4047 	if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4048 		ZFS_EXIT(zfsvfs);
4049 		return (SET_ERROR(EINVAL));
4050 	}
4051 
4052 
4053 	owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4054 	if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4055 		ZFS_EXIT(zfsvfs);
4056 		return (SET_ERROR(EPERM));
4057 	}
4058 
4059 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4060 		ZFS_EXIT(zfsvfs);
4061 		return (error);
4062 	}
4063 
4064 top:
4065 	/*
4066 	 * Attempt to lock directory; fail if entry already exists.
4067 	 */
4068 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4069 	if (error) {
4070 		ZFS_EXIT(zfsvfs);
4071 		return (error);
4072 	}
4073 
4074 	tx = dmu_tx_create(zfsvfs->z_os);
4075 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4076 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4077 	zfs_sa_upgrade_txholds(tx, szp);
4078 	zfs_sa_upgrade_txholds(tx, dzp);
4079 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4080 	if (error) {
4081 		zfs_dirent_unlock(dl);
4082 		if (error == ERESTART) {
4083 			waited = B_TRUE;
4084 			dmu_tx_wait(tx);
4085 			dmu_tx_abort(tx);
4086 			goto top;
4087 		}
4088 		dmu_tx_abort(tx);
4089 		ZFS_EXIT(zfsvfs);
4090 		return (error);
4091 	}
4092 
4093 	error = zfs_link_create(dl, szp, tx, 0);
4094 
4095 	if (error == 0) {
4096 		uint64_t txtype = TX_LINK;
4097 		if (flags & FIGNORECASE)
4098 			txtype |= TX_CI;
4099 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4100 	}
4101 
4102 	dmu_tx_commit(tx);
4103 
4104 	zfs_dirent_unlock(dl);
4105 
4106 	if (error == 0) {
4107 		vnevent_link(svp, ct);
4108 	}
4109 
4110 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4111 		zil_commit(zilog, 0);
4112 
4113 	ZFS_EXIT(zfsvfs);
4114 	return (error);
4115 }
4116 
4117 /*
4118  * zfs_null_putapage() is used when the file system has been force
4119  * unmounted. It just drops the pages.
4120  */
4121 /* ARGSUSED */
4122 static int
4123 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4124 		size_t *lenp, int flags, cred_t *cr)
4125 {
4126 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4127 	return (0);
4128 }
4129 
4130 /*
4131  * Push a page out to disk, klustering if possible.
4132  *
4133  *	IN:	vp	- file to push page to.
4134  *		pp	- page to push.
4135  *		flags	- additional flags.
4136  *		cr	- credentials of caller.
4137  *
4138  *	OUT:	offp	- start of range pushed.
4139  *		lenp	- len of range pushed.
4140  *
4141  *	RETURN:	0 on success, error code on failure.
4142  *
4143  * NOTE: callers must have locked the page to be pushed.  On
4144  * exit, the page (and all other pages in the kluster) must be
4145  * unlocked.
4146  */
4147 /* ARGSUSED */
4148 static int
4149 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4150 		size_t *lenp, int flags, cred_t *cr)
4151 {
4152 	znode_t		*zp = VTOZ(vp);
4153 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4154 	dmu_tx_t	*tx;
4155 	u_offset_t	off, koff;
4156 	size_t		len, klen;
4157 	int		err;
4158 
4159 	off = pp->p_offset;
4160 	len = PAGESIZE;
4161 	/*
4162 	 * If our blocksize is bigger than the page size, try to kluster
4163 	 * multiple pages so that we write a full block (thus avoiding
4164 	 * a read-modify-write).
4165 	 */
4166 	if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4167 		klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4168 		koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4169 		ASSERT(koff <= zp->z_size);
4170 		if (koff + klen > zp->z_size)
4171 			klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4172 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4173 	}
4174 	ASSERT3U(btop(len), ==, btopr(len));
4175 
4176 	/*
4177 	 * Can't push pages past end-of-file.
4178 	 */
4179 	if (off >= zp->z_size) {
4180 		/* ignore all pages */
4181 		err = 0;
4182 		goto out;
4183 	} else if (off + len > zp->z_size) {
4184 		int npages = btopr(zp->z_size - off);
4185 		page_t *trunc;
4186 
4187 		page_list_break(&pp, &trunc, npages);
4188 		/* ignore pages past end of file */
4189 		if (trunc)
4190 			pvn_write_done(trunc, flags);
4191 		len = zp->z_size - off;
4192 	}
4193 
4194 	if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4195 	    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4196 		err = SET_ERROR(EDQUOT);
4197 		goto out;
4198 	}
4199 	tx = dmu_tx_create(zfsvfs->z_os);
4200 	dmu_tx_hold_write(tx, zp->z_id, off, len);
4201 
4202 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4203 	zfs_sa_upgrade_txholds(tx, zp);
4204 	err = dmu_tx_assign(tx, TXG_WAIT);
4205 	if (err != 0) {
4206 		dmu_tx_abort(tx);
4207 		goto out;
4208 	}
4209 
4210 	if (zp->z_blksz <= PAGESIZE) {
4211 		caddr_t va = zfs_map_page(pp, S_READ);
4212 		ASSERT3U(len, <=, PAGESIZE);
4213 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4214 		zfs_unmap_page(pp, va);
4215 	} else {
4216 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4217 	}
4218 
4219 	if (err == 0) {
4220 		uint64_t mtime[2], ctime[2];
4221 		sa_bulk_attr_t bulk[3];
4222 		int count = 0;
4223 
4224 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4225 		    &mtime, 16);
4226 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4227 		    &ctime, 16);
4228 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4229 		    &zp->z_pflags, 8);
4230 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4231 		    B_TRUE);
4232 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4233 	}
4234 	dmu_tx_commit(tx);
4235 
4236 out:
4237 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4238 	if (offp)
4239 		*offp = off;
4240 	if (lenp)
4241 		*lenp = len;
4242 
4243 	return (err);
4244 }
4245 
4246 /*
4247  * Copy the portion of the file indicated from pages into the file.
4248  * The pages are stored in a page list attached to the files vnode.
4249  *
4250  *	IN:	vp	- vnode of file to push page data to.
4251  *		off	- position in file to put data.
4252  *		len	- amount of data to write.
4253  *		flags	- flags to control the operation.
4254  *		cr	- credentials of caller.
4255  *		ct	- caller context.
4256  *
4257  *	RETURN:	0 on success, error code on failure.
4258  *
4259  * Timestamps:
4260  *	vp - ctime|mtime updated
4261  */
4262 /*ARGSUSED*/
4263 static int
4264 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4265     caller_context_t *ct)
4266 {
4267 	znode_t		*zp = VTOZ(vp);
4268 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4269 	page_t		*pp;
4270 	size_t		io_len;
4271 	u_offset_t	io_off;
4272 	uint_t		blksz;
4273 	rl_t		*rl;
4274 	int		error = 0;
4275 
4276 	ZFS_ENTER(zfsvfs);
4277 	ZFS_VERIFY_ZP(zp);
4278 
4279 	/*
4280 	 * There's nothing to do if no data is cached.
4281 	 */
4282 	if (!vn_has_cached_data(vp)) {
4283 		ZFS_EXIT(zfsvfs);
4284 		return (0);
4285 	}
4286 
4287 	/*
4288 	 * Align this request to the file block size in case we kluster.
4289 	 * XXX - this can result in pretty aggresive locking, which can
4290 	 * impact simultanious read/write access.  One option might be
4291 	 * to break up long requests (len == 0) into block-by-block
4292 	 * operations to get narrower locking.
4293 	 */
4294 	blksz = zp->z_blksz;
4295 	if (ISP2(blksz))
4296 		io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4297 	else
4298 		io_off = 0;
4299 	if (len > 0 && ISP2(blksz))
4300 		io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4301 	else
4302 		io_len = 0;
4303 
4304 	if (io_len == 0) {
4305 		/*
4306 		 * Search the entire vp list for pages >= io_off.
4307 		 */
4308 		rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4309 		error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4310 		goto out;
4311 	}
4312 	rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4313 
4314 	if (off > zp->z_size) {
4315 		/* past end of file */
4316 		zfs_range_unlock(rl);
4317 		ZFS_EXIT(zfsvfs);
4318 		return (0);
4319 	}
4320 
4321 	len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4322 
4323 	for (off = io_off; io_off < off + len; io_off += io_len) {
4324 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4325 			pp = page_lookup(vp, io_off,
4326 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4327 		} else {
4328 			pp = page_lookup_nowait(vp, io_off,
4329 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4330 		}
4331 
4332 		if (pp != NULL && pvn_getdirty(pp, flags)) {
4333 			int err;
4334 
4335 			/*
4336 			 * Found a dirty page to push
4337 			 */
4338 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4339 			if (err)
4340 				error = err;
4341 		} else {
4342 			io_len = PAGESIZE;
4343 		}
4344 	}
4345 out:
4346 	zfs_range_unlock(rl);
4347 	if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4348 		zil_commit(zfsvfs->z_log, zp->z_id);
4349 	ZFS_EXIT(zfsvfs);
4350 	return (error);
4351 }
4352 
4353 /*ARGSUSED*/
4354 void
4355 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4356 {
4357 	znode_t	*zp = VTOZ(vp);
4358 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4359 	int error;
4360 
4361 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4362 	if (zp->z_sa_hdl == NULL) {
4363 		/*
4364 		 * The fs has been unmounted, or we did a
4365 		 * suspend/resume and this file no longer exists.
4366 		 */
4367 		if (vn_has_cached_data(vp)) {
4368 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4369 			    B_INVAL, cr);
4370 		}
4371 
4372 		mutex_enter(&zp->z_lock);
4373 		mutex_enter(&vp->v_lock);
4374 		ASSERT(vp->v_count == 1);
4375 		vp->v_count = 0;
4376 		mutex_exit(&vp->v_lock);
4377 		mutex_exit(&zp->z_lock);
4378 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
4379 		zfs_znode_free(zp);
4380 		return;
4381 	}
4382 
4383 	/*
4384 	 * Attempt to push any data in the page cache.  If this fails
4385 	 * we will get kicked out later in zfs_zinactive().
4386 	 */
4387 	if (vn_has_cached_data(vp)) {
4388 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4389 		    cr);
4390 	}
4391 
4392 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4393 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4394 
4395 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4396 		zfs_sa_upgrade_txholds(tx, zp);
4397 		error = dmu_tx_assign(tx, TXG_WAIT);
4398 		if (error) {
4399 			dmu_tx_abort(tx);
4400 		} else {
4401 			mutex_enter(&zp->z_lock);
4402 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4403 			    (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4404 			zp->z_atime_dirty = 0;
4405 			mutex_exit(&zp->z_lock);
4406 			dmu_tx_commit(tx);
4407 		}
4408 	}
4409 
4410 	zfs_zinactive(zp);
4411 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
4412 }
4413 
4414 /*
4415  * Bounds-check the seek operation.
4416  *
4417  *	IN:	vp	- vnode seeking within
4418  *		ooff	- old file offset
4419  *		noffp	- pointer to new file offset
4420  *		ct	- caller context
4421  *
4422  *	RETURN:	0 on success, EINVAL if new offset invalid.
4423  */
4424 /* ARGSUSED */
4425 static int
4426 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4427     caller_context_t *ct)
4428 {
4429 	if (vp->v_type == VDIR)
4430 		return (0);
4431 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4432 }
4433 
4434 /*
4435  * Pre-filter the generic locking function to trap attempts to place
4436  * a mandatory lock on a memory mapped file.
4437  */
4438 static int
4439 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4440     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4441 {
4442 	znode_t *zp = VTOZ(vp);
4443 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4444 
4445 	ZFS_ENTER(zfsvfs);
4446 	ZFS_VERIFY_ZP(zp);
4447 
4448 	/*
4449 	 * We are following the UFS semantics with respect to mapcnt
4450 	 * here: If we see that the file is mapped already, then we will
4451 	 * return an error, but we don't worry about races between this
4452 	 * function and zfs_map().
4453 	 */
4454 	if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4455 		ZFS_EXIT(zfsvfs);
4456 		return (SET_ERROR(EAGAIN));
4457 	}
4458 	ZFS_EXIT(zfsvfs);
4459 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4460 }
4461 
4462 /*
4463  * If we can't find a page in the cache, we will create a new page
4464  * and fill it with file data.  For efficiency, we may try to fill
4465  * multiple pages at once (klustering) to fill up the supplied page
4466  * list.  Note that the pages to be filled are held with an exclusive
4467  * lock to prevent access by other threads while they are being filled.
4468  */
4469 static int
4470 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4471     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4472 {
4473 	znode_t *zp = VTOZ(vp);
4474 	page_t *pp, *cur_pp;
4475 	objset_t *os = zp->z_zfsvfs->z_os;
4476 	u_offset_t io_off, total;
4477 	size_t io_len;
4478 	int err;
4479 
4480 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4481 		/*
4482 		 * We only have a single page, don't bother klustering
4483 		 */
4484 		io_off = off;
4485 		io_len = PAGESIZE;
4486 		pp = page_create_va(vp, io_off, io_len,
4487 		    PG_EXCL | PG_WAIT, seg, addr);
4488 	} else {
4489 		/*
4490 		 * Try to find enough pages to fill the page list
4491 		 */
4492 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4493 		    &io_len, off, plsz, 0);
4494 	}
4495 	if (pp == NULL) {
4496 		/*
4497 		 * The page already exists, nothing to do here.
4498 		 */
4499 		*pl = NULL;
4500 		return (0);
4501 	}
4502 
4503 	/*
4504 	 * Fill the pages in the kluster.
4505 	 */
4506 	cur_pp = pp;
4507 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4508 		caddr_t va;
4509 
4510 		ASSERT3U(io_off, ==, cur_pp->p_offset);
4511 		va = zfs_map_page(cur_pp, S_WRITE);
4512 		err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4513 		    DMU_READ_PREFETCH);
4514 		zfs_unmap_page(cur_pp, va);
4515 		if (err) {
4516 			/* On error, toss the entire kluster */
4517 			pvn_read_done(pp, B_ERROR);
4518 			/* convert checksum errors into IO errors */
4519 			if (err == ECKSUM)
4520 				err = SET_ERROR(EIO);
4521 			return (err);
4522 		}
4523 		cur_pp = cur_pp->p_next;
4524 	}
4525 
4526 	/*
4527 	 * Fill in the page list array from the kluster starting
4528 	 * from the desired offset `off'.
4529 	 * NOTE: the page list will always be null terminated.
4530 	 */
4531 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4532 	ASSERT(pl == NULL || (*pl)->p_offset == off);
4533 
4534 	return (0);
4535 }
4536 
4537 /*
4538  * Return pointers to the pages for the file region [off, off + len]
4539  * in the pl array.  If plsz is greater than len, this function may
4540  * also return page pointers from after the specified region
4541  * (i.e. the region [off, off + plsz]).  These additional pages are
4542  * only returned if they are already in the cache, or were created as
4543  * part of a klustered read.
4544  *
4545  *	IN:	vp	- vnode of file to get data from.
4546  *		off	- position in file to get data from.
4547  *		len	- amount of data to retrieve.
4548  *		plsz	- length of provided page list.
4549  *		seg	- segment to obtain pages for.
4550  *		addr	- virtual address of fault.
4551  *		rw	- mode of created pages.
4552  *		cr	- credentials of caller.
4553  *		ct	- caller context.
4554  *
4555  *	OUT:	protp	- protection mode of created pages.
4556  *		pl	- list of pages created.
4557  *
4558  *	RETURN:	0 on success, error code on failure.
4559  *
4560  * Timestamps:
4561  *	vp - atime updated
4562  */
4563 /* ARGSUSED */
4564 static int
4565 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4566     page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4567     enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4568 {
4569 	znode_t		*zp = VTOZ(vp);
4570 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4571 	page_t		**pl0 = pl;
4572 	int		err = 0;
4573 
4574 	/* we do our own caching, faultahead is unnecessary */
4575 	if (pl == NULL)
4576 		return (0);
4577 	else if (len > plsz)
4578 		len = plsz;
4579 	else
4580 		len = P2ROUNDUP(len, PAGESIZE);
4581 	ASSERT(plsz >= len);
4582 
4583 	ZFS_ENTER(zfsvfs);
4584 	ZFS_VERIFY_ZP(zp);
4585 
4586 	if (protp)
4587 		*protp = PROT_ALL;
4588 
4589 	/*
4590 	 * Loop through the requested range [off, off + len) looking
4591 	 * for pages.  If we don't find a page, we will need to create
4592 	 * a new page and fill it with data from the file.
4593 	 */
4594 	while (len > 0) {
4595 		if (*pl = page_lookup(vp, off, SE_SHARED))
4596 			*(pl+1) = NULL;
4597 		else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4598 			goto out;
4599 		while (*pl) {
4600 			ASSERT3U((*pl)->p_offset, ==, off);
4601 			off += PAGESIZE;
4602 			addr += PAGESIZE;
4603 			if (len > 0) {
4604 				ASSERT3U(len, >=, PAGESIZE);
4605 				len -= PAGESIZE;
4606 			}
4607 			ASSERT3U(plsz, >=, PAGESIZE);
4608 			plsz -= PAGESIZE;
4609 			pl++;
4610 		}
4611 	}
4612 
4613 	/*
4614 	 * Fill out the page array with any pages already in the cache.
4615 	 */
4616 	while (plsz > 0 &&
4617 	    (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4618 			off += PAGESIZE;
4619 			plsz -= PAGESIZE;
4620 	}
4621 out:
4622 	if (err) {
4623 		/*
4624 		 * Release any pages we have previously locked.
4625 		 */
4626 		while (pl > pl0)
4627 			page_unlock(*--pl);
4628 	} else {
4629 		ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4630 	}
4631 
4632 	*pl = NULL;
4633 
4634 	ZFS_EXIT(zfsvfs);
4635 	return (err);
4636 }
4637 
4638 /*
4639  * Request a memory map for a section of a file.  This code interacts
4640  * with common code and the VM system as follows:
4641  *
4642  * - common code calls mmap(), which ends up in smmap_common()
4643  * - this calls VOP_MAP(), which takes you into (say) zfs
4644  * - zfs_map() calls as_map(), passing segvn_create() as the callback
4645  * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4646  * - zfs_addmap() updates z_mapcnt
4647  */
4648 /*ARGSUSED*/
4649 static int
4650 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4651     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4652     caller_context_t *ct)
4653 {
4654 	znode_t *zp = VTOZ(vp);
4655 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4656 	segvn_crargs_t	vn_a;
4657 	int		error;
4658 
4659 	ZFS_ENTER(zfsvfs);
4660 	ZFS_VERIFY_ZP(zp);
4661 
4662 	if ((prot & PROT_WRITE) && (zp->z_pflags &
4663 	    (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4664 		ZFS_EXIT(zfsvfs);
4665 		return (SET_ERROR(EPERM));
4666 	}
4667 
4668 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4669 	    (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4670 		ZFS_EXIT(zfsvfs);
4671 		return (SET_ERROR(EACCES));
4672 	}
4673 
4674 	if (vp->v_flag & VNOMAP) {
4675 		ZFS_EXIT(zfsvfs);
4676 		return (SET_ERROR(ENOSYS));
4677 	}
4678 
4679 	if (off < 0 || len > MAXOFFSET_T - off) {
4680 		ZFS_EXIT(zfsvfs);
4681 		return (SET_ERROR(ENXIO));
4682 	}
4683 
4684 	if (vp->v_type != VREG) {
4685 		ZFS_EXIT(zfsvfs);
4686 		return (SET_ERROR(ENODEV));
4687 	}
4688 
4689 	/*
4690 	 * If file is locked, disallow mapping.
4691 	 */
4692 	if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4693 		ZFS_EXIT(zfsvfs);
4694 		return (SET_ERROR(EAGAIN));
4695 	}
4696 
4697 	as_rangelock(as);
4698 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4699 	if (error != 0) {
4700 		as_rangeunlock(as);
4701 		ZFS_EXIT(zfsvfs);
4702 		return (error);
4703 	}
4704 
4705 	vn_a.vp = vp;
4706 	vn_a.offset = (u_offset_t)off;
4707 	vn_a.type = flags & MAP_TYPE;
4708 	vn_a.prot = prot;
4709 	vn_a.maxprot = maxprot;
4710 	vn_a.cred = cr;
4711 	vn_a.amp = NULL;
4712 	vn_a.flags = flags & ~MAP_TYPE;
4713 	vn_a.szc = 0;
4714 	vn_a.lgrp_mem_policy_flags = 0;
4715 
4716 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4717 
4718 	as_rangeunlock(as);
4719 	ZFS_EXIT(zfsvfs);
4720 	return (error);
4721 }
4722 
4723 /* ARGSUSED */
4724 static int
4725 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4726     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4727     caller_context_t *ct)
4728 {
4729 	uint64_t pages = btopr(len);
4730 
4731 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4732 	return (0);
4733 }
4734 
4735 /*
4736  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4737  * more accurate mtime for the associated file.  Since we don't have a way of
4738  * detecting when the data was actually modified, we have to resort to
4739  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4740  * last page is pushed.  The problem occurs when the msync() call is omitted,
4741  * which by far the most common case:
4742  *
4743  * 	open()
4744  * 	mmap()
4745  * 	<modify memory>
4746  * 	munmap()
4747  * 	close()
4748  * 	<time lapse>
4749  * 	putpage() via fsflush
4750  *
4751  * If we wait until fsflush to come along, we can have a modification time that
4752  * is some arbitrary point in the future.  In order to prevent this in the
4753  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4754  * torn down.
4755  */
4756 /* ARGSUSED */
4757 static int
4758 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4759     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4760     caller_context_t *ct)
4761 {
4762 	uint64_t pages = btopr(len);
4763 
4764 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4765 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4766 
4767 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4768 	    vn_has_cached_data(vp))
4769 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4770 
4771 	return (0);
4772 }
4773 
4774 /*
4775  * Free or allocate space in a file.  Currently, this function only
4776  * supports the `F_FREESP' command.  However, this command is somewhat
4777  * misnamed, as its functionality includes the ability to allocate as
4778  * well as free space.
4779  *
4780  *	IN:	vp	- vnode of file to free data in.
4781  *		cmd	- action to take (only F_FREESP supported).
4782  *		bfp	- section of file to free/alloc.
4783  *		flag	- current file open mode flags.
4784  *		offset	- current file offset.
4785  *		cr	- credentials of caller [UNUSED].
4786  *		ct	- caller context.
4787  *
4788  *	RETURN:	0 on success, error code on failure.
4789  *
4790  * Timestamps:
4791  *	vp - ctime|mtime updated
4792  */
4793 /* ARGSUSED */
4794 static int
4795 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4796     offset_t offset, cred_t *cr, caller_context_t *ct)
4797 {
4798 	znode_t		*zp = VTOZ(vp);
4799 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4800 	uint64_t	off, len;
4801 	int		error;
4802 
4803 	ZFS_ENTER(zfsvfs);
4804 	ZFS_VERIFY_ZP(zp);
4805 
4806 	if (cmd != F_FREESP) {
4807 		ZFS_EXIT(zfsvfs);
4808 		return (SET_ERROR(EINVAL));
4809 	}
4810 
4811 	/*
4812 	 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
4813 	 * callers might not be able to detect properly that we are read-only,
4814 	 * so check it explicitly here.
4815 	 */
4816 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
4817 		ZFS_EXIT(zfsvfs);
4818 		return (SET_ERROR(EROFS));
4819 	}
4820 
4821 	if (error = convoff(vp, bfp, 0, offset)) {
4822 		ZFS_EXIT(zfsvfs);
4823 		return (error);
4824 	}
4825 
4826 	if (bfp->l_len < 0) {
4827 		ZFS_EXIT(zfsvfs);
4828 		return (SET_ERROR(EINVAL));
4829 	}
4830 
4831 	off = bfp->l_start;
4832 	len = bfp->l_len; /* 0 means from off to end of file */
4833 
4834 	error = zfs_freesp(zp, off, len, flag, TRUE);
4835 
4836 	if (error == 0 && off == 0 && len == 0)
4837 		vnevent_truncate(ZTOV(zp), ct);
4838 
4839 	ZFS_EXIT(zfsvfs);
4840 	return (error);
4841 }
4842 
4843 /*ARGSUSED*/
4844 static int
4845 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4846 {
4847 	znode_t		*zp = VTOZ(vp);
4848 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4849 	uint32_t	gen;
4850 	uint64_t	gen64;
4851 	uint64_t	object = zp->z_id;
4852 	zfid_short_t	*zfid;
4853 	int		size, i, error;
4854 
4855 	ZFS_ENTER(zfsvfs);
4856 	ZFS_VERIFY_ZP(zp);
4857 
4858 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4859 	    &gen64, sizeof (uint64_t))) != 0) {
4860 		ZFS_EXIT(zfsvfs);
4861 		return (error);
4862 	}
4863 
4864 	gen = (uint32_t)gen64;
4865 
4866 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4867 	if (fidp->fid_len < size) {
4868 		fidp->fid_len = size;
4869 		ZFS_EXIT(zfsvfs);
4870 		return (SET_ERROR(ENOSPC));
4871 	}
4872 
4873 	zfid = (zfid_short_t *)fidp;
4874 
4875 	zfid->zf_len = size;
4876 
4877 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4878 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4879 
4880 	/* Must have a non-zero generation number to distinguish from .zfs */
4881 	if (gen == 0)
4882 		gen = 1;
4883 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4884 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4885 
4886 	if (size == LONG_FID_LEN) {
4887 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4888 		zfid_long_t	*zlfid;
4889 
4890 		zlfid = (zfid_long_t *)fidp;
4891 
4892 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4893 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4894 
4895 		/* XXX - this should be the generation number for the objset */
4896 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4897 			zlfid->zf_setgen[i] = 0;
4898 	}
4899 
4900 	ZFS_EXIT(zfsvfs);
4901 	return (0);
4902 }
4903 
4904 static int
4905 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4906     caller_context_t *ct)
4907 {
4908 	znode_t		*zp, *xzp;
4909 	zfsvfs_t	*zfsvfs;
4910 	zfs_dirlock_t	*dl;
4911 	int		error;
4912 
4913 	switch (cmd) {
4914 	case _PC_LINK_MAX:
4915 		*valp = ULONG_MAX;
4916 		return (0);
4917 
4918 	case _PC_FILESIZEBITS:
4919 		*valp = 64;
4920 		return (0);
4921 
4922 	case _PC_XATTR_EXISTS:
4923 		zp = VTOZ(vp);
4924 		zfsvfs = zp->z_zfsvfs;
4925 		ZFS_ENTER(zfsvfs);
4926 		ZFS_VERIFY_ZP(zp);
4927 		*valp = 0;
4928 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4929 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4930 		if (error == 0) {
4931 			zfs_dirent_unlock(dl);
4932 			if (!zfs_dirempty(xzp))
4933 				*valp = 1;
4934 			VN_RELE(ZTOV(xzp));
4935 		} else if (error == ENOENT) {
4936 			/*
4937 			 * If there aren't extended attributes, it's the
4938 			 * same as having zero of them.
4939 			 */
4940 			error = 0;
4941 		}
4942 		ZFS_EXIT(zfsvfs);
4943 		return (error);
4944 
4945 	case _PC_SATTR_ENABLED:
4946 	case _PC_SATTR_EXISTS:
4947 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4948 		    (vp->v_type == VREG || vp->v_type == VDIR);
4949 		return (0);
4950 
4951 	case _PC_ACCESS_FILTERING:
4952 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4953 		    vp->v_type == VDIR;
4954 		return (0);
4955 
4956 	case _PC_ACL_ENABLED:
4957 		*valp = _ACL_ACE_ENABLED;
4958 		return (0);
4959 
4960 	case _PC_MIN_HOLE_SIZE:
4961 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4962 		return (0);
4963 
4964 	case _PC_TIMESTAMP_RESOLUTION:
4965 		/* nanosecond timestamp resolution */
4966 		*valp = 1L;
4967 		return (0);
4968 
4969 	default:
4970 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4971 	}
4972 }
4973 
4974 /*ARGSUSED*/
4975 static int
4976 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4977     caller_context_t *ct)
4978 {
4979 	znode_t *zp = VTOZ(vp);
4980 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4981 	int error;
4982 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4983 
4984 	ZFS_ENTER(zfsvfs);
4985 	ZFS_VERIFY_ZP(zp);
4986 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4987 	ZFS_EXIT(zfsvfs);
4988 
4989 	return (error);
4990 }
4991 
4992 /*ARGSUSED*/
4993 static int
4994 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4995     caller_context_t *ct)
4996 {
4997 	znode_t *zp = VTOZ(vp);
4998 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4999 	int error;
5000 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5001 	zilog_t	*zilog = zfsvfs->z_log;
5002 
5003 	ZFS_ENTER(zfsvfs);
5004 	ZFS_VERIFY_ZP(zp);
5005 
5006 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5007 
5008 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5009 		zil_commit(zilog, 0);
5010 
5011 	ZFS_EXIT(zfsvfs);
5012 	return (error);
5013 }
5014 
5015 /*
5016  * The smallest read we may consider to loan out an arcbuf.
5017  * This must be a power of 2.
5018  */
5019 int zcr_blksz_min = (1 << 10);	/* 1K */
5020 /*
5021  * If set to less than the file block size, allow loaning out of an
5022  * arcbuf for a partial block read.  This must be a power of 2.
5023  */
5024 int zcr_blksz_max = (1 << 17);	/* 128K */
5025 
5026 /*ARGSUSED*/
5027 static int
5028 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5029     caller_context_t *ct)
5030 {
5031 	znode_t	*zp = VTOZ(vp);
5032 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5033 	int max_blksz = zfsvfs->z_max_blksz;
5034 	uio_t *uio = &xuio->xu_uio;
5035 	ssize_t size = uio->uio_resid;
5036 	offset_t offset = uio->uio_loffset;
5037 	int blksz;
5038 	int fullblk, i;
5039 	arc_buf_t *abuf;
5040 	ssize_t maxsize;
5041 	int preamble, postamble;
5042 
5043 	if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5044 		return (SET_ERROR(EINVAL));
5045 
5046 	ZFS_ENTER(zfsvfs);
5047 	ZFS_VERIFY_ZP(zp);
5048 	switch (ioflag) {
5049 	case UIO_WRITE:
5050 		/*
5051 		 * Loan out an arc_buf for write if write size is bigger than
5052 		 * max_blksz, and the file's block size is also max_blksz.
5053 		 */
5054 		blksz = max_blksz;
5055 		if (size < blksz || zp->z_blksz != blksz) {
5056 			ZFS_EXIT(zfsvfs);
5057 			return (SET_ERROR(EINVAL));
5058 		}
5059 		/*
5060 		 * Caller requests buffers for write before knowing where the
5061 		 * write offset might be (e.g. NFS TCP write).
5062 		 */
5063 		if (offset == -1) {
5064 			preamble = 0;
5065 		} else {
5066 			preamble = P2PHASE(offset, blksz);
5067 			if (preamble) {
5068 				preamble = blksz - preamble;
5069 				size -= preamble;
5070 			}
5071 		}
5072 
5073 		postamble = P2PHASE(size, blksz);
5074 		size -= postamble;
5075 
5076 		fullblk = size / blksz;
5077 		(void) dmu_xuio_init(xuio,
5078 		    (preamble != 0) + fullblk + (postamble != 0));
5079 		DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5080 		    int, postamble, int,
5081 		    (preamble != 0) + fullblk + (postamble != 0));
5082 
5083 		/*
5084 		 * Have to fix iov base/len for partial buffers.  They
5085 		 * currently represent full arc_buf's.
5086 		 */
5087 		if (preamble) {
5088 			/* data begins in the middle of the arc_buf */
5089 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5090 			    blksz);
5091 			ASSERT(abuf);
5092 			(void) dmu_xuio_add(xuio, abuf,
5093 			    blksz - preamble, preamble);
5094 		}
5095 
5096 		for (i = 0; i < fullblk; i++) {
5097 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5098 			    blksz);
5099 			ASSERT(abuf);
5100 			(void) dmu_xuio_add(xuio, abuf, 0, blksz);
5101 		}
5102 
5103 		if (postamble) {
5104 			/* data ends in the middle of the arc_buf */
5105 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5106 			    blksz);
5107 			ASSERT(abuf);
5108 			(void) dmu_xuio_add(xuio, abuf, 0, postamble);
5109 		}
5110 		break;
5111 	case UIO_READ:
5112 		/*
5113 		 * Loan out an arc_buf for read if the read size is larger than
5114 		 * the current file block size.  Block alignment is not
5115 		 * considered.  Partial arc_buf will be loaned out for read.
5116 		 */
5117 		blksz = zp->z_blksz;
5118 		if (blksz < zcr_blksz_min)
5119 			blksz = zcr_blksz_min;
5120 		if (blksz > zcr_blksz_max)
5121 			blksz = zcr_blksz_max;
5122 		/* avoid potential complexity of dealing with it */
5123 		if (blksz > max_blksz) {
5124 			ZFS_EXIT(zfsvfs);
5125 			return (SET_ERROR(EINVAL));
5126 		}
5127 
5128 		maxsize = zp->z_size - uio->uio_loffset;
5129 		if (size > maxsize)
5130 			size = maxsize;
5131 
5132 		if (size < blksz || vn_has_cached_data(vp)) {
5133 			ZFS_EXIT(zfsvfs);
5134 			return (SET_ERROR(EINVAL));
5135 		}
5136 		break;
5137 	default:
5138 		ZFS_EXIT(zfsvfs);
5139 		return (SET_ERROR(EINVAL));
5140 	}
5141 
5142 	uio->uio_extflg = UIO_XUIO;
5143 	XUIO_XUZC_RW(xuio) = ioflag;
5144 	ZFS_EXIT(zfsvfs);
5145 	return (0);
5146 }
5147 
5148 /*ARGSUSED*/
5149 static int
5150 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5151 {
5152 	int i;
5153 	arc_buf_t *abuf;
5154 	int ioflag = XUIO_XUZC_RW(xuio);
5155 
5156 	ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5157 
5158 	i = dmu_xuio_cnt(xuio);
5159 	while (i-- > 0) {
5160 		abuf = dmu_xuio_arcbuf(xuio, i);
5161 		/*
5162 		 * if abuf == NULL, it must be a write buffer
5163 		 * that has been returned in zfs_write().
5164 		 */
5165 		if (abuf)
5166 			dmu_return_arcbuf(abuf);
5167 		ASSERT(abuf || ioflag == UIO_WRITE);
5168 	}
5169 
5170 	dmu_xuio_fini(xuio);
5171 	return (0);
5172 }
5173 
5174 /*
5175  * Predeclare these here so that the compiler assumes that
5176  * this is an "old style" function declaration that does
5177  * not include arguments => we won't get type mismatch errors
5178  * in the initializations that follow.
5179  */
5180 static int zfs_inval();
5181 static int zfs_isdir();
5182 
5183 static int
5184 zfs_inval()
5185 {
5186 	return (SET_ERROR(EINVAL));
5187 }
5188 
5189 static int
5190 zfs_isdir()
5191 {
5192 	return (SET_ERROR(EISDIR));
5193 }
5194 /*
5195  * Directory vnode operations template
5196  */
5197 vnodeops_t *zfs_dvnodeops;
5198 const fs_operation_def_t zfs_dvnodeops_template[] = {
5199 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5200 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5201 	VOPNAME_READ,		{ .error = zfs_isdir },
5202 	VOPNAME_WRITE,		{ .error = zfs_isdir },
5203 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5204 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5205 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5206 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5207 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5208 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5209 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5210 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5211 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5212 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
5213 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5214 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5215 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
5216 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5217 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5218 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5219 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5220 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5221 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5222 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5223 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
5224 	NULL,			NULL
5225 };
5226 
5227 /*
5228  * Regular file vnode operations template
5229  */
5230 vnodeops_t *zfs_fvnodeops;
5231 const fs_operation_def_t zfs_fvnodeops_template[] = {
5232 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5233 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5234 	VOPNAME_READ,		{ .vop_read = zfs_read },
5235 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
5236 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5237 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5238 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5239 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5240 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5241 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5242 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5243 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5244 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5245 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5246 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
5247 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
5248 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
5249 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
5250 	VOPNAME_MAP,		{ .vop_map = zfs_map },
5251 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
5252 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
5253 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5254 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5255 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5256 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5257 	VOPNAME_REQZCBUF, 	{ .vop_reqzcbuf = zfs_reqzcbuf },
5258 	VOPNAME_RETZCBUF, 	{ .vop_retzcbuf = zfs_retzcbuf },
5259 	NULL,			NULL
5260 };
5261 
5262 /*
5263  * Symbolic link vnode operations template
5264  */
5265 vnodeops_t *zfs_symvnodeops;
5266 const fs_operation_def_t zfs_symvnodeops_template[] = {
5267 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5268 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5269 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5270 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5271 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
5272 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5273 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5274 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5275 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5276 	NULL,			NULL
5277 };
5278 
5279 /*
5280  * special share hidden files vnode operations template
5281  */
5282 vnodeops_t *zfs_sharevnodeops;
5283 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5284 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5285 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5286 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5287 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5288 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5289 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5290 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5291 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5292 	NULL,			NULL
5293 };
5294 
5295 /*
5296  * Extended attribute directory vnode operations template
5297  *
5298  * This template is identical to the directory vnodes
5299  * operation template except for restricted operations:
5300  *	VOP_MKDIR()
5301  *	VOP_SYMLINK()
5302  *
5303  * Note that there are other restrictions embedded in:
5304  *	zfs_create()	- restrict type to VREG
5305  *	zfs_link()	- no links into/out of attribute space
5306  *	zfs_rename()	- no moves into/out of attribute space
5307  */
5308 vnodeops_t *zfs_xdvnodeops;
5309 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5310 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5311 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5312 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5313 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5314 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5315 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5316 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5317 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5318 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5319 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5320 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5321 	VOPNAME_MKDIR,		{ .error = zfs_inval },
5322 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5323 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5324 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
5325 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5326 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5327 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5328 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5329 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5330 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5331 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5332 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5333 	NULL,			NULL
5334 };
5335 
5336 /*
5337  * Error vnode operations template
5338  */
5339 vnodeops_t *zfs_evnodeops;
5340 const fs_operation_def_t zfs_evnodeops_template[] = {
5341 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5342 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5343 	NULL,			NULL
5344 };
5345