xref: /titanic_41/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision ab5b81dc12eecf427a851f7fae335f2667b08d7b)
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) 2012, 2014 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 	if (error == ESRCH)
286 		return (SET_ERROR(ENXIO));
287 
288 	/*
289 	 * We could find a hole that begins after the logical end-of-file,
290 	 * because dmu_offset_next() only works on whole blocks.  If the
291 	 * EOF falls mid-block, then indicate that the "virtual hole"
292 	 * at the end of the file begins at the logical EOF, rather than
293 	 * at the end of the last block.
294 	 */
295 	if (noff > file_sz) {
296 		ASSERT(hole);
297 		noff = file_sz;
298 	}
299 
300 	if (noff < *off)
301 		return (error);
302 	*off = noff;
303 	return (error);
304 }
305 
306 /* ARGSUSED */
307 static int
308 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
309     int *rvalp, caller_context_t *ct)
310 {
311 	offset_t off;
312 	int error;
313 	zfsvfs_t *zfsvfs;
314 	znode_t *zp;
315 
316 	switch (com) {
317 	case _FIOFFS:
318 		return (zfs_sync(vp->v_vfsp, 0, cred));
319 
320 		/*
321 		 * The following two ioctls are used by bfu.  Faking out,
322 		 * necessary to avoid bfu errors.
323 		 */
324 	case _FIOGDIO:
325 	case _FIOSDIO:
326 		return (0);
327 
328 	case _FIO_SEEK_DATA:
329 	case _FIO_SEEK_HOLE:
330 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
331 			return (SET_ERROR(EFAULT));
332 
333 		zp = VTOZ(vp);
334 		zfsvfs = zp->z_zfsvfs;
335 		ZFS_ENTER(zfsvfs);
336 		ZFS_VERIFY_ZP(zp);
337 
338 		/* offset parameter is in/out */
339 		error = zfs_holey(vp, com, &off);
340 		ZFS_EXIT(zfsvfs);
341 		if (error)
342 			return (error);
343 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
344 			return (SET_ERROR(EFAULT));
345 		return (0);
346 	}
347 	return (SET_ERROR(ENOTTY));
348 }
349 
350 /*
351  * Utility functions to map and unmap a single physical page.  These
352  * are used to manage the mappable copies of ZFS file data, and therefore
353  * do not update ref/mod bits.
354  */
355 caddr_t
356 zfs_map_page(page_t *pp, enum seg_rw rw)
357 {
358 	if (kpm_enable)
359 		return (hat_kpm_mapin(pp, 0));
360 	ASSERT(rw == S_READ || rw == S_WRITE);
361 	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
362 	    (caddr_t)-1));
363 }
364 
365 void
366 zfs_unmap_page(page_t *pp, caddr_t addr)
367 {
368 	if (kpm_enable) {
369 		hat_kpm_mapout(pp, 0, addr);
370 	} else {
371 		ppmapout(addr);
372 	}
373 }
374 
375 /*
376  * When a file is memory mapped, we must keep the IO data synchronized
377  * between the DMU cache and the memory mapped pages.  What this means:
378  *
379  * On Write:	If we find a memory mapped page, we write to *both*
380  *		the page and the dmu buffer.
381  */
382 static void
383 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
384 {
385 	int64_t	off;
386 
387 	off = start & PAGEOFFSET;
388 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
389 		page_t *pp;
390 		uint64_t nbytes = MIN(PAGESIZE - off, len);
391 
392 		if (pp = page_lookup(vp, start, SE_SHARED)) {
393 			caddr_t va;
394 
395 			va = zfs_map_page(pp, S_WRITE);
396 			(void) dmu_read(os, oid, start+off, nbytes, va+off,
397 			    DMU_READ_PREFETCH);
398 			zfs_unmap_page(pp, va);
399 			page_unlock(pp);
400 		}
401 		len -= nbytes;
402 		off = 0;
403 	}
404 }
405 
406 /*
407  * When a file is memory mapped, we must keep the IO data synchronized
408  * between the DMU cache and the memory mapped pages.  What this means:
409  *
410  * On Read:	We "read" preferentially from memory mapped pages,
411  *		else we default from the dmu buffer.
412  *
413  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
414  *	 the file is memory mapped.
415  */
416 static int
417 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
418 {
419 	znode_t *zp = VTOZ(vp);
420 	int64_t	start, off;
421 	int len = nbytes;
422 	int error = 0;
423 
424 	start = uio->uio_loffset;
425 	off = start & PAGEOFFSET;
426 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
427 		page_t *pp;
428 		uint64_t bytes = MIN(PAGESIZE - off, len);
429 
430 		if (pp = page_lookup(vp, start, SE_SHARED)) {
431 			caddr_t va;
432 
433 			va = zfs_map_page(pp, S_READ);
434 			error = uiomove(va + off, bytes, UIO_READ, uio);
435 			zfs_unmap_page(pp, va);
436 			page_unlock(pp);
437 		} else {
438 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
439 			    uio, bytes);
440 		}
441 		len -= bytes;
442 		off = 0;
443 		if (error)
444 			break;
445 	}
446 	return (error);
447 }
448 
449 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
450 
451 /*
452  * Read bytes from specified file into supplied buffer.
453  *
454  *	IN:	vp	- vnode of file to be read from.
455  *		uio	- structure supplying read location, range info,
456  *			  and return buffer.
457  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
458  *		cr	- credentials of caller.
459  *		ct	- caller context
460  *
461  *	OUT:	uio	- updated offset and range, buffer filled.
462  *
463  *	RETURN:	0 on success, error code on failure.
464  *
465  * Side Effects:
466  *	vp - atime updated if byte count > 0
467  */
468 /* ARGSUSED */
469 static int
470 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
471 {
472 	znode_t		*zp = VTOZ(vp);
473 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
474 	ssize_t		n, nbytes;
475 	int		error = 0;
476 	rl_t		*rl;
477 	xuio_t		*xuio = NULL;
478 
479 	ZFS_ENTER(zfsvfs);
480 	ZFS_VERIFY_ZP(zp);
481 
482 	if (zp->z_pflags & ZFS_AV_QUARANTINED) {
483 		ZFS_EXIT(zfsvfs);
484 		return (SET_ERROR(EACCES));
485 	}
486 
487 	/*
488 	 * Validate file offset
489 	 */
490 	if (uio->uio_loffset < (offset_t)0) {
491 		ZFS_EXIT(zfsvfs);
492 		return (SET_ERROR(EINVAL));
493 	}
494 
495 	/*
496 	 * Fasttrack empty reads
497 	 */
498 	if (uio->uio_resid == 0) {
499 		ZFS_EXIT(zfsvfs);
500 		return (0);
501 	}
502 
503 	/*
504 	 * Check for mandatory locks
505 	 */
506 	if (MANDMODE(zp->z_mode)) {
507 		if (error = chklock(vp, FREAD,
508 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
509 			ZFS_EXIT(zfsvfs);
510 			return (error);
511 		}
512 	}
513 
514 	/*
515 	 * If we're in FRSYNC mode, sync out this znode before reading it.
516 	 */
517 	if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
518 		zil_commit(zfsvfs->z_log, zp->z_id);
519 
520 	/*
521 	 * Lock the range against changes.
522 	 */
523 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
524 
525 	/*
526 	 * If we are reading past end-of-file we can skip
527 	 * to the end; but we might still need to set atime.
528 	 */
529 	if (uio->uio_loffset >= zp->z_size) {
530 		error = 0;
531 		goto out;
532 	}
533 
534 	ASSERT(uio->uio_loffset < zp->z_size);
535 	n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
536 
537 	if ((uio->uio_extflg == UIO_XUIO) &&
538 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
539 		int nblk;
540 		int blksz = zp->z_blksz;
541 		uint64_t offset = uio->uio_loffset;
542 
543 		xuio = (xuio_t *)uio;
544 		if ((ISP2(blksz))) {
545 			nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
546 			    blksz)) / blksz;
547 		} else {
548 			ASSERT(offset + n <= blksz);
549 			nblk = 1;
550 		}
551 		(void) dmu_xuio_init(xuio, nblk);
552 
553 		if (vn_has_cached_data(vp)) {
554 			/*
555 			 * For simplicity, we always allocate a full buffer
556 			 * even if we only expect to read a portion of a block.
557 			 */
558 			while (--nblk >= 0) {
559 				(void) dmu_xuio_add(xuio,
560 				    dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
561 				    blksz), 0, blksz);
562 			}
563 		}
564 	}
565 
566 	while (n > 0) {
567 		nbytes = MIN(n, zfs_read_chunk_size -
568 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
569 
570 		if (vn_has_cached_data(vp)) {
571 			error = mappedread(vp, nbytes, uio);
572 		} else {
573 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
574 			    uio, nbytes);
575 		}
576 		if (error) {
577 			/* convert checksum errors into IO errors */
578 			if (error == ECKSUM)
579 				error = SET_ERROR(EIO);
580 			break;
581 		}
582 
583 		n -= nbytes;
584 	}
585 out:
586 	zfs_range_unlock(rl);
587 
588 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
589 	ZFS_EXIT(zfsvfs);
590 	return (error);
591 }
592 
593 /*
594  * Write the bytes to a file.
595  *
596  *	IN:	vp	- vnode of file to be written to.
597  *		uio	- structure supplying write location, range info,
598  *			  and data buffer.
599  *		ioflag	- FAPPEND, FSYNC, and/or FDSYNC.  FAPPEND is
600  *			  set if in append mode.
601  *		cr	- credentials of caller.
602  *		ct	- caller context (NFS/CIFS fem monitor only)
603  *
604  *	OUT:	uio	- updated offset and range.
605  *
606  *	RETURN:	0 on success, error code on failure.
607  *
608  * Timestamps:
609  *	vp - ctime|mtime updated if byte count > 0
610  */
611 
612 /* ARGSUSED */
613 static int
614 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
615 {
616 	znode_t		*zp = VTOZ(vp);
617 	rlim64_t	limit = uio->uio_llimit;
618 	ssize_t		start_resid = uio->uio_resid;
619 	ssize_t		tx_bytes;
620 	uint64_t	end_size;
621 	dmu_tx_t	*tx;
622 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
623 	zilog_t		*zilog;
624 	offset_t	woff;
625 	ssize_t		n, nbytes;
626 	rl_t		*rl;
627 	int		max_blksz = zfsvfs->z_max_blksz;
628 	int		error = 0;
629 	arc_buf_t	*abuf;
630 	iovec_t		*aiov = NULL;
631 	xuio_t		*xuio = NULL;
632 	int		i_iov = 0;
633 	int		iovcnt = uio->uio_iovcnt;
634 	iovec_t		*iovp = uio->uio_iov;
635 	int		write_eof;
636 	int		count = 0;
637 	sa_bulk_attr_t	bulk[4];
638 	uint64_t	mtime[2], ctime[2];
639 	ssize_t		lock_off, lock_len;
640 
641 	/*
642 	 * Fasttrack empty write
643 	 */
644 	n = start_resid;
645 	if (n == 0)
646 		return (0);
647 
648 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
649 		limit = MAXOFFSET_T;
650 
651 	ZFS_ENTER(zfsvfs);
652 	ZFS_VERIFY_ZP(zp);
653 
654 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
655 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
656 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
657 	    &zp->z_size, 8);
658 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
659 	    &zp->z_pflags, 8);
660 
661 	/*
662 	 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
663 	 * callers might not be able to detect properly that we are read-only,
664 	 * so check it explicitly here.
665 	 */
666 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
667 		ZFS_EXIT(zfsvfs);
668 		return (SET_ERROR(EROFS));
669 	}
670 
671 	/*
672 	 * If immutable or not appending then return EPERM
673 	 */
674 	if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
675 	    ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
676 	    (uio->uio_loffset < zp->z_size))) {
677 		ZFS_EXIT(zfsvfs);
678 		return (SET_ERROR(EPERM));
679 	}
680 
681 	zilog = zfsvfs->z_log;
682 
683 	/*
684 	 * Validate file offset
685 	 */
686 	woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
687 	if (woff < 0) {
688 		ZFS_EXIT(zfsvfs);
689 		return (SET_ERROR(EINVAL));
690 	}
691 
692 	/*
693 	 * Check for mandatory locks before calling zfs_range_lock()
694 	 * in order to prevent a deadlock with locks set via fcntl().
695 	 */
696 	if (MANDMODE((mode_t)zp->z_mode) &&
697 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
698 		ZFS_EXIT(zfsvfs);
699 		return (error);
700 	}
701 
702 	/*
703 	 * Pre-fault the pages to ensure slow (eg NFS) pages
704 	 * don't hold up txg.
705 	 * Skip this if uio contains loaned arc_buf.
706 	 */
707 	if ((uio->uio_extflg == UIO_XUIO) &&
708 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
709 		xuio = (xuio_t *)uio;
710 	else
711 		uio_prefaultpages(MIN(n, max_blksz), uio);
712 
713 	/*
714 	 * If in append mode, set the io offset pointer to eof.
715 	 */
716 	if (ioflag & FAPPEND) {
717 		/*
718 		 * Obtain an appending range lock to guarantee file append
719 		 * semantics.  We reset the write offset once we have the lock.
720 		 */
721 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
722 		woff = rl->r_off;
723 		if (rl->r_len == UINT64_MAX) {
724 			/*
725 			 * We overlocked the file because this write will cause
726 			 * the file block size to increase.
727 			 * Note that zp_size cannot change with this lock held.
728 			 */
729 			woff = zp->z_size;
730 		}
731 		uio->uio_loffset = woff;
732 	} else {
733 		/*
734 		 * Note that if the file block size will change as a result of
735 		 * this write, then this range lock will lock the entire file
736 		 * so that we can re-write the block safely.
737 		 */
738 
739 		/*
740 		 * If in zev mode, lock offsets are quantized to 1MB chunks
741 		 * so that we can calculate level 1 checksums later on.
742 		 */
743 		if (rz_zev_active()) {
744 			/* start of this megabyte */
745 			lock_off = P2ALIGN(woff, ZEV_L1_SIZE);
746 			/* full megabytes */
747 			lock_len = n + (woff - lock_off);
748 			lock_len = P2ROUNDUP(lock_len, ZEV_L1_SIZE);
749 		} else {
750 			lock_off = woff;
751 			lock_len = n;
752 		}
753 
754 		rl = zfs_range_lock(zp, lock_off, lock_len, RL_WRITER);
755 	}
756 
757 	if (woff >= limit) {
758 		zfs_range_unlock(rl);
759 		ZFS_EXIT(zfsvfs);
760 		return (SET_ERROR(EFBIG));
761 	}
762 
763 	if ((woff + n) > limit || woff > (limit - n))
764 		n = limit - woff;
765 
766 	/* Will this write extend the file length? */
767 	write_eof = (woff + n > zp->z_size);
768 
769 	end_size = MAX(zp->z_size, woff + n);
770 
771 	/*
772 	 * Write the file in reasonable size chunks.  Each chunk is written
773 	 * in a separate transaction; this keeps the intent log records small
774 	 * and allows us to do more fine-grained space accounting.
775 	 */
776 	while (n > 0) {
777 		abuf = NULL;
778 		woff = uio->uio_loffset;
779 		if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
780 		    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
781 			if (abuf != NULL)
782 				dmu_return_arcbuf(abuf);
783 			error = SET_ERROR(EDQUOT);
784 			break;
785 		}
786 
787 		if (xuio && abuf == NULL) {
788 			ASSERT(i_iov < iovcnt);
789 			aiov = &iovp[i_iov];
790 			abuf = dmu_xuio_arcbuf(xuio, i_iov);
791 			dmu_xuio_clear(xuio, i_iov);
792 			DTRACE_PROBE3(zfs_cp_write, int, i_iov,
793 			    iovec_t *, aiov, arc_buf_t *, abuf);
794 			ASSERT((aiov->iov_base == abuf->b_data) ||
795 			    ((char *)aiov->iov_base - (char *)abuf->b_data +
796 			    aiov->iov_len == arc_buf_size(abuf)));
797 			i_iov++;
798 		} else if (abuf == NULL && n >= max_blksz &&
799 		    woff >= zp->z_size &&
800 		    P2PHASE(woff, max_blksz) == 0 &&
801 		    zp->z_blksz == max_blksz) {
802 			/*
803 			 * This write covers a full block.  "Borrow" a buffer
804 			 * from the dmu so that we can fill it before we enter
805 			 * a transaction.  This avoids the possibility of
806 			 * holding up the transaction if the data copy hangs
807 			 * up on a pagefault (e.g., from an NFS server mapping).
808 			 */
809 			size_t cbytes;
810 
811 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
812 			    max_blksz);
813 			ASSERT(abuf != NULL);
814 			ASSERT(arc_buf_size(abuf) == max_blksz);
815 			if (error = uiocopy(abuf->b_data, max_blksz,
816 			    UIO_WRITE, uio, &cbytes)) {
817 				dmu_return_arcbuf(abuf);
818 				break;
819 			}
820 			ASSERT(cbytes == max_blksz);
821 		}
822 
823 		/*
824 		 * Start a transaction.
825 		 */
826 		tx = dmu_tx_create(zfsvfs->z_os);
827 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
828 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
829 		zfs_sa_upgrade_txholds(tx, zp);
830 		error = dmu_tx_assign(tx, TXG_WAIT);
831 		if (error) {
832 			dmu_tx_abort(tx);
833 			if (abuf != NULL)
834 				dmu_return_arcbuf(abuf);
835 			break;
836 		}
837 
838 		/*
839 		 * If zfs_range_lock() over-locked we grow the blocksize
840 		 * and then reduce the lock range.  This will only happen
841 		 * on the first iteration since zfs_range_reduce() will
842 		 * shrink down r_len to the appropriate size.
843 		 */
844 		if (rl->r_len == UINT64_MAX) {
845 			uint64_t new_blksz;
846 
847 			if (zp->z_blksz > max_blksz) {
848 				ASSERT(!ISP2(zp->z_blksz));
849 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
850 			} else {
851 				new_blksz = MIN(end_size, max_blksz);
852 			}
853 			zfs_grow_blocksize(zp, new_blksz, tx);
854 			zfs_range_reduce(rl, woff, n);
855 		}
856 
857 		/*
858 		 * XXX - should we really limit each write to z_max_blksz?
859 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
860 		 */
861 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
862 
863 		if (abuf == NULL) {
864 			tx_bytes = uio->uio_resid;
865 			error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
866 			    uio, nbytes, tx);
867 			tx_bytes -= uio->uio_resid;
868 		} else {
869 			tx_bytes = nbytes;
870 			ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
871 			/*
872 			 * If this is not a full block write, but we are
873 			 * extending the file past EOF and this data starts
874 			 * block-aligned, use assign_arcbuf().  Otherwise,
875 			 * write via dmu_write().
876 			 */
877 			if (tx_bytes < max_blksz && (!write_eof ||
878 			    aiov->iov_base != abuf->b_data)) {
879 				ASSERT(xuio);
880 				dmu_write(zfsvfs->z_os, zp->z_id, woff,
881 				    aiov->iov_len, aiov->iov_base, tx);
882 				dmu_return_arcbuf(abuf);
883 				xuio_stat_wbuf_copied();
884 			} else {
885 				ASSERT(xuio || tx_bytes == max_blksz);
886 				dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
887 				    woff, abuf, tx);
888 			}
889 			ASSERT(tx_bytes <= uio->uio_resid);
890 			uioskip(uio, tx_bytes);
891 		}
892 		if (tx_bytes && vn_has_cached_data(vp)) {
893 			update_pages(vp, woff,
894 			    tx_bytes, zfsvfs->z_os, zp->z_id);
895 		}
896 
897 		/*
898 		 * If we made no progress, we're done.  If we made even
899 		 * partial progress, update the znode and ZIL accordingly.
900 		 */
901 		if (tx_bytes == 0) {
902 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
903 			    (void *)&zp->z_size, sizeof (uint64_t), tx);
904 			dmu_tx_commit(tx);
905 			ASSERT(error != 0);
906 			break;
907 		}
908 
909 		/*
910 		 * Clear Set-UID/Set-GID bits on successful write if not
911 		 * privileged and at least one of the excute bits is set.
912 		 *
913 		 * It would be nice to to this after all writes have
914 		 * been done, but that would still expose the ISUID/ISGID
915 		 * to another app after the partial write is committed.
916 		 *
917 		 * Note: we don't call zfs_fuid_map_id() here because
918 		 * user 0 is not an ephemeral uid.
919 		 */
920 		mutex_enter(&zp->z_acl_lock);
921 		if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
922 		    (S_IXUSR >> 6))) != 0 &&
923 		    (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
924 		    secpolicy_vnode_setid_retain(cr,
925 		    (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
926 			uint64_t newmode;
927 			zp->z_mode &= ~(S_ISUID | S_ISGID);
928 			newmode = zp->z_mode;
929 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
930 			    (void *)&newmode, sizeof (uint64_t), tx);
931 		}
932 		mutex_exit(&zp->z_acl_lock);
933 
934 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
935 		    B_TRUE);
936 
937 		/*
938 		 * Update the file size (zp_size) if it has changed;
939 		 * account for possible concurrent updates.
940 		 */
941 		while ((end_size = zp->z_size) < uio->uio_loffset) {
942 			(void) atomic_cas_64(&zp->z_size, end_size,
943 			    uio->uio_loffset);
944 			ASSERT(error == 0);
945 		}
946 		/*
947 		 * If we are replaying and eof is non zero then force
948 		 * the file size to the specified eof. Note, there's no
949 		 * concurrency during replay.
950 		 */
951 		if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
952 			zp->z_size = zfsvfs->z_replay_eof;
953 
954 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
955 
956 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
957 		dmu_tx_commit(tx);
958 
959 		if (error != 0)
960 			break;
961 		ASSERT(tx_bytes == nbytes);
962 		n -= nbytes;
963 
964 		if (!xuio && n > 0)
965 			uio_prefaultpages(MIN(n, max_blksz), uio);
966 	}
967 
968 	zfs_range_unlock(rl);
969 
970 	/*
971 	 * If we're in replay mode, or we made no progress, return error.
972 	 * Otherwise, it's at least a partial write, so it's successful.
973 	 */
974 	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
975 		ZFS_EXIT(zfsvfs);
976 		return (error);
977 	}
978 
979 	if (ioflag & (FSYNC | FDSYNC) ||
980 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
981 		zil_commit(zilog, zp->z_id);
982 
983 	ZFS_EXIT(zfsvfs);
984 	return (0);
985 }
986 
987 void
988 zfs_get_done(zgd_t *zgd, int error)
989 {
990 	znode_t *zp = zgd->zgd_private;
991 	objset_t *os = zp->z_zfsvfs->z_os;
992 
993 	if (zgd->zgd_db)
994 		dmu_buf_rele(zgd->zgd_db, zgd);
995 
996 	zfs_range_unlock(zgd->zgd_rl);
997 
998 	/*
999 	 * Release the vnode asynchronously as we currently have the
1000 	 * txg stopped from syncing.
1001 	 */
1002 	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1003 
1004 	if (error == 0 && zgd->zgd_bp)
1005 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1006 
1007 	kmem_free(zgd, sizeof (zgd_t));
1008 }
1009 
1010 #ifdef DEBUG
1011 static int zil_fault_io = 0;
1012 #endif
1013 
1014 /*
1015  * Get data to generate a TX_WRITE intent log record.
1016  */
1017 int
1018 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1019 {
1020 	zfsvfs_t *zfsvfs = arg;
1021 	objset_t *os = zfsvfs->z_os;
1022 	znode_t *zp;
1023 	uint64_t object = lr->lr_foid;
1024 	uint64_t offset = lr->lr_offset;
1025 	uint64_t size = lr->lr_length;
1026 	blkptr_t *bp = &lr->lr_blkptr;
1027 	dmu_buf_t *db;
1028 	zgd_t *zgd;
1029 	int error = 0;
1030 
1031 	ASSERT(zio != NULL);
1032 	ASSERT(size != 0);
1033 
1034 	/*
1035 	 * Nothing to do if the file has been removed
1036 	 */
1037 	if (zfs_zget(zfsvfs, object, &zp) != 0)
1038 		return (SET_ERROR(ENOENT));
1039 	if (zp->z_unlinked) {
1040 		/*
1041 		 * Release the vnode asynchronously as we currently have the
1042 		 * txg stopped from syncing.
1043 		 */
1044 		VN_RELE_ASYNC(ZTOV(zp),
1045 		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1046 		return (SET_ERROR(ENOENT));
1047 	}
1048 
1049 	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1050 	zgd->zgd_zilog = zfsvfs->z_log;
1051 	zgd->zgd_private = zp;
1052 
1053 	/*
1054 	 * Write records come in two flavors: immediate and indirect.
1055 	 * For small writes it's cheaper to store the data with the
1056 	 * log record (immediate); for large writes it's cheaper to
1057 	 * sync the data and get a pointer to it (indirect) so that
1058 	 * we don't have to write the data twice.
1059 	 */
1060 	if (buf != NULL) { /* immediate write */
1061 		zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1062 		/* test for truncation needs to be done while range locked */
1063 		if (offset >= zp->z_size) {
1064 			error = SET_ERROR(ENOENT);
1065 		} else {
1066 			error = dmu_read(os, object, offset, size, buf,
1067 			    DMU_READ_NO_PREFETCH);
1068 		}
1069 		ASSERT(error == 0 || error == ENOENT);
1070 	} else { /* indirect write */
1071 		/*
1072 		 * Have to lock the whole block to ensure when it's
1073 		 * written out and it's checksum is being calculated
1074 		 * that no one can change the data. We need to re-check
1075 		 * blocksize after we get the lock in case it's changed!
1076 		 */
1077 		for (;;) {
1078 			uint64_t blkoff;
1079 			size = zp->z_blksz;
1080 			blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1081 			offset -= blkoff;
1082 			zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1083 			    RL_READER);
1084 			if (zp->z_blksz == size)
1085 				break;
1086 			offset += blkoff;
1087 			zfs_range_unlock(zgd->zgd_rl);
1088 		}
1089 		/* test for truncation needs to be done while range locked */
1090 		if (lr->lr_offset >= zp->z_size)
1091 			error = SET_ERROR(ENOENT);
1092 #ifdef DEBUG
1093 		if (zil_fault_io) {
1094 			error = SET_ERROR(EIO);
1095 			zil_fault_io = 0;
1096 		}
1097 #endif
1098 		if (error == 0)
1099 			error = dmu_buf_hold(os, object, offset, zgd, &db,
1100 			    DMU_READ_NO_PREFETCH);
1101 
1102 		if (error == 0) {
1103 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1104 			if (obp) {
1105 				ASSERT(BP_IS_HOLE(bp));
1106 				*bp = *obp;
1107 			}
1108 
1109 			zgd->zgd_db = db;
1110 			zgd->zgd_bp = bp;
1111 
1112 			ASSERT(db->db_offset == offset);
1113 			ASSERT(db->db_size == size);
1114 
1115 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1116 			    zfs_get_done, zgd);
1117 			ASSERT(error || lr->lr_length <= zp->z_blksz);
1118 
1119 			/*
1120 			 * On success, we need to wait for the write I/O
1121 			 * initiated by dmu_sync() to complete before we can
1122 			 * release this dbuf.  We will finish everything up
1123 			 * in the zfs_get_done() callback.
1124 			 */
1125 			if (error == 0)
1126 				return (0);
1127 
1128 			if (error == EALREADY) {
1129 				lr->lr_common.lrc_txtype = TX_WRITE2;
1130 				error = 0;
1131 			}
1132 		}
1133 	}
1134 
1135 	zfs_get_done(zgd, error);
1136 
1137 	return (error);
1138 }
1139 
1140 /*ARGSUSED*/
1141 static int
1142 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1143     caller_context_t *ct)
1144 {
1145 	znode_t *zp = VTOZ(vp);
1146 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1147 	int error;
1148 
1149 	ZFS_ENTER(zfsvfs);
1150 	ZFS_VERIFY_ZP(zp);
1151 
1152 	if (flag & V_ACE_MASK)
1153 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1154 	else
1155 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
1156 
1157 	ZFS_EXIT(zfsvfs);
1158 	return (error);
1159 }
1160 
1161 /*
1162  * If vnode is for a device return a specfs vnode instead.
1163  */
1164 static int
1165 specvp_check(vnode_t **vpp, cred_t *cr)
1166 {
1167 	int error = 0;
1168 
1169 	if (IS_DEVVP(*vpp)) {
1170 		struct vnode *svp;
1171 
1172 		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1173 		VN_RELE(*vpp);
1174 		if (svp == NULL)
1175 			error = SET_ERROR(ENOSYS);
1176 		*vpp = svp;
1177 	}
1178 	return (error);
1179 }
1180 
1181 
1182 /*
1183  * Lookup an entry in a directory, or an extended attribute directory.
1184  * If it exists, return a held vnode reference for it.
1185  *
1186  *	IN:	dvp	- vnode of directory to search.
1187  *		nm	- name of entry to lookup.
1188  *		pnp	- full pathname to lookup [UNUSED].
1189  *		flags	- LOOKUP_XATTR set if looking for an attribute.
1190  *		rdir	- root directory vnode [UNUSED].
1191  *		cr	- credentials of caller.
1192  *		ct	- caller context
1193  *		direntflags - directory lookup flags
1194  *		realpnp - returned pathname.
1195  *
1196  *	OUT:	vpp	- vnode of located entry, NULL if not found.
1197  *
1198  *	RETURN:	0 on success, error code on failure.
1199  *
1200  * Timestamps:
1201  *	NA
1202  */
1203 /* ARGSUSED */
1204 static int
1205 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1206     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1207     int *direntflags, pathname_t *realpnp)
1208 {
1209 	znode_t *zdp = VTOZ(dvp);
1210 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1211 	int	error = 0;
1212 
1213 	/* fast path */
1214 	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1215 
1216 		if (dvp->v_type != VDIR) {
1217 			return (SET_ERROR(ENOTDIR));
1218 		} else if (zdp->z_sa_hdl == NULL) {
1219 			return (SET_ERROR(EIO));
1220 		}
1221 
1222 		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1223 			error = zfs_fastaccesschk_execute(zdp, cr);
1224 			if (!error) {
1225 				*vpp = dvp;
1226 				VN_HOLD(*vpp);
1227 				return (0);
1228 			}
1229 			return (error);
1230 		} else {
1231 			vnode_t *tvp = dnlc_lookup(dvp, nm);
1232 
1233 			if (tvp) {
1234 				error = zfs_fastaccesschk_execute(zdp, cr);
1235 				if (error) {
1236 					VN_RELE(tvp);
1237 					return (error);
1238 				}
1239 				if (tvp == DNLC_NO_VNODE) {
1240 					VN_RELE(tvp);
1241 					return (SET_ERROR(ENOENT));
1242 				} else {
1243 					*vpp = tvp;
1244 					return (specvp_check(vpp, cr));
1245 				}
1246 			}
1247 		}
1248 	}
1249 
1250 	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1251 
1252 	ZFS_ENTER(zfsvfs);
1253 	ZFS_VERIFY_ZP(zdp);
1254 
1255 	*vpp = NULL;
1256 
1257 	if (flags & LOOKUP_XATTR) {
1258 		/*
1259 		 * If the xattr property is off, refuse the lookup request.
1260 		 */
1261 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1262 			ZFS_EXIT(zfsvfs);
1263 			return (SET_ERROR(EINVAL));
1264 		}
1265 
1266 		/*
1267 		 * We don't allow recursive attributes..
1268 		 * Maybe someday we will.
1269 		 */
1270 		if (zdp->z_pflags & ZFS_XATTR) {
1271 			ZFS_EXIT(zfsvfs);
1272 			return (SET_ERROR(EINVAL));
1273 		}
1274 
1275 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1276 			ZFS_EXIT(zfsvfs);
1277 			return (error);
1278 		}
1279 
1280 		/*
1281 		 * Do we have permission to get into attribute directory?
1282 		 */
1283 
1284 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1285 		    B_FALSE, cr)) {
1286 			VN_RELE(*vpp);
1287 			*vpp = NULL;
1288 		}
1289 
1290 		ZFS_EXIT(zfsvfs);
1291 		return (error);
1292 	}
1293 
1294 	if (dvp->v_type != VDIR) {
1295 		ZFS_EXIT(zfsvfs);
1296 		return (SET_ERROR(ENOTDIR));
1297 	}
1298 
1299 	/*
1300 	 * Check accessibility of directory.
1301 	 */
1302 
1303 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1304 		ZFS_EXIT(zfsvfs);
1305 		return (error);
1306 	}
1307 
1308 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1309 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1310 		ZFS_EXIT(zfsvfs);
1311 		return (SET_ERROR(EILSEQ));
1312 	}
1313 
1314 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1315 	if (error == 0)
1316 		error = specvp_check(vpp, cr);
1317 
1318 	ZFS_EXIT(zfsvfs);
1319 	return (error);
1320 }
1321 
1322 /*
1323  * Attempt to create a new entry in a directory.  If the entry
1324  * already exists, truncate the file if permissible, else return
1325  * an error.  Return the vp of the created or trunc'd file.
1326  *
1327  *	IN:	dvp	- vnode of directory to put new file entry in.
1328  *		name	- name of new file entry.
1329  *		vap	- attributes of new file.
1330  *		excl	- flag indicating exclusive or non-exclusive mode.
1331  *		mode	- mode to open file with.
1332  *		cr	- credentials of caller.
1333  *		flag	- large file flag [UNUSED].
1334  *		ct	- caller context
1335  *		vsecp	- ACL to be set
1336  *
1337  *	OUT:	vpp	- vnode of created or trunc'd entry.
1338  *
1339  *	RETURN:	0 on success, error code on failure.
1340  *
1341  * Timestamps:
1342  *	dvp - ctime|mtime updated if new entry created
1343  *	 vp - ctime|mtime always, atime if new
1344  */
1345 
1346 /* ARGSUSED */
1347 static int
1348 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1349     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1350     vsecattr_t *vsecp)
1351 {
1352 	znode_t		*zp, *dzp = VTOZ(dvp);
1353 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1354 	zilog_t		*zilog;
1355 	objset_t	*os;
1356 	zfs_dirlock_t	*dl;
1357 	dmu_tx_t	*tx;
1358 	int		error;
1359 	ksid_t		*ksid;
1360 	uid_t		uid;
1361 	gid_t		gid = crgetgid(cr);
1362 	zfs_acl_ids_t   acl_ids;
1363 	boolean_t	fuid_dirtied;
1364 	boolean_t	have_acl = B_FALSE;
1365 	boolean_t	waited = B_FALSE;
1366 
1367 	/*
1368 	 * If we have an ephemeral id, ACL, or XVATTR then
1369 	 * make sure file system is at proper version
1370 	 */
1371 
1372 	ksid = crgetsid(cr, KSID_OWNER);
1373 	if (ksid)
1374 		uid = ksid_getid(ksid);
1375 	else
1376 		uid = crgetuid(cr);
1377 
1378 	if (zfsvfs->z_use_fuids == B_FALSE &&
1379 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1380 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1381 		return (SET_ERROR(EINVAL));
1382 
1383 	ZFS_ENTER(zfsvfs);
1384 	ZFS_VERIFY_ZP(dzp);
1385 	os = zfsvfs->z_os;
1386 	zilog = zfsvfs->z_log;
1387 
1388 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1389 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1390 		ZFS_EXIT(zfsvfs);
1391 		return (SET_ERROR(EILSEQ));
1392 	}
1393 
1394 	if (vap->va_mask & AT_XVATTR) {
1395 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1396 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1397 			ZFS_EXIT(zfsvfs);
1398 			return (error);
1399 		}
1400 	}
1401 top:
1402 	*vpp = NULL;
1403 
1404 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1405 		vap->va_mode &= ~VSVTX;
1406 
1407 	if (*name == '\0') {
1408 		/*
1409 		 * Null component name refers to the directory itself.
1410 		 */
1411 		VN_HOLD(dvp);
1412 		zp = dzp;
1413 		dl = NULL;
1414 		error = 0;
1415 	} else {
1416 		/* possible VN_HOLD(zp) */
1417 		int zflg = 0;
1418 
1419 		if (flag & FIGNORECASE)
1420 			zflg |= ZCILOOK;
1421 
1422 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1423 		    NULL, NULL);
1424 		if (error) {
1425 			if (have_acl)
1426 				zfs_acl_ids_free(&acl_ids);
1427 			if (strcmp(name, "..") == 0)
1428 				error = SET_ERROR(EISDIR);
1429 			ZFS_EXIT(zfsvfs);
1430 			return (error);
1431 		}
1432 	}
1433 
1434 	if (zp == NULL) {
1435 		uint64_t txtype;
1436 
1437 		/*
1438 		 * Create a new file object and update the directory
1439 		 * to reference it.
1440 		 */
1441 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1442 			if (have_acl)
1443 				zfs_acl_ids_free(&acl_ids);
1444 			goto out;
1445 		}
1446 
1447 		/*
1448 		 * We only support the creation of regular files in
1449 		 * extended attribute directories.
1450 		 */
1451 
1452 		if ((dzp->z_pflags & ZFS_XATTR) &&
1453 		    (vap->va_type != VREG)) {
1454 			if (have_acl)
1455 				zfs_acl_ids_free(&acl_ids);
1456 			error = SET_ERROR(EINVAL);
1457 			goto out;
1458 		}
1459 
1460 		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1461 		    cr, vsecp, &acl_ids)) != 0)
1462 			goto out;
1463 		have_acl = B_TRUE;
1464 
1465 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1466 			zfs_acl_ids_free(&acl_ids);
1467 			error = SET_ERROR(EDQUOT);
1468 			goto out;
1469 		}
1470 
1471 		tx = dmu_tx_create(os);
1472 
1473 		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1474 		    ZFS_SA_BASE_ATTR_SIZE);
1475 
1476 		fuid_dirtied = zfsvfs->z_fuid_dirty;
1477 		if (fuid_dirtied)
1478 			zfs_fuid_txhold(zfsvfs, tx);
1479 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1480 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1481 		if (!zfsvfs->z_use_sa &&
1482 		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1483 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1484 			    0, acl_ids.z_aclp->z_acl_bytes);
1485 		}
1486 		error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1487 		if (error) {
1488 			zfs_dirent_unlock(dl);
1489 			if (error == ERESTART) {
1490 				waited = B_TRUE;
1491 				dmu_tx_wait(tx);
1492 				dmu_tx_abort(tx);
1493 				goto top;
1494 			}
1495 			zfs_acl_ids_free(&acl_ids);
1496 			dmu_tx_abort(tx);
1497 			ZFS_EXIT(zfsvfs);
1498 			return (error);
1499 		}
1500 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1501 
1502 		if (fuid_dirtied)
1503 			zfs_fuid_sync(zfsvfs, tx);
1504 
1505 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1506 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1507 		if (flag & FIGNORECASE)
1508 			txtype |= TX_CI;
1509 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1510 		    vsecp, acl_ids.z_fuidp, vap);
1511 		zfs_acl_ids_free(&acl_ids);
1512 		dmu_tx_commit(tx);
1513 	} else {
1514 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1515 
1516 		if (have_acl)
1517 			zfs_acl_ids_free(&acl_ids);
1518 		have_acl = B_FALSE;
1519 
1520 		/*
1521 		 * A directory entry already exists for this name.
1522 		 */
1523 		/*
1524 		 * Can't truncate an existing file if in exclusive mode.
1525 		 */
1526 		if (excl == EXCL) {
1527 			error = SET_ERROR(EEXIST);
1528 			goto out;
1529 		}
1530 		/*
1531 		 * Can't open a directory for writing.
1532 		 */
1533 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1534 			error = SET_ERROR(EISDIR);
1535 			goto out;
1536 		}
1537 		/*
1538 		 * Verify requested access to file.
1539 		 */
1540 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1541 			goto out;
1542 		}
1543 
1544 		mutex_enter(&dzp->z_lock);
1545 		dzp->z_seq++;
1546 		mutex_exit(&dzp->z_lock);
1547 
1548 		/*
1549 		 * Truncate regular files if requested.
1550 		 */
1551 		if ((ZTOV(zp)->v_type == VREG) &&
1552 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1553 			/* we can't hold any locks when calling zfs_freesp() */
1554 			zfs_dirent_unlock(dl);
1555 			dl = NULL;
1556 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1557 			if (error == 0) {
1558 				vnevent_create(ZTOV(zp), ct);
1559 			}
1560 		}
1561 	}
1562 out:
1563 
1564 	if (dl)
1565 		zfs_dirent_unlock(dl);
1566 
1567 	if (error) {
1568 		if (zp)
1569 			VN_RELE(ZTOV(zp));
1570 	} else {
1571 		*vpp = ZTOV(zp);
1572 		error = specvp_check(vpp, cr);
1573 	}
1574 
1575 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1576 		zil_commit(zilog, 0);
1577 
1578 	ZFS_EXIT(zfsvfs);
1579 	return (error);
1580 }
1581 
1582 /*
1583  * Remove an entry from a directory.
1584  *
1585  *	IN:	dvp	- vnode of directory to remove entry from.
1586  *		name	- name of entry to remove.
1587  *		cr	- credentials of caller.
1588  *		ct	- caller context
1589  *		flags	- case flags
1590  *
1591  *	RETURN:	0 on success, error code on failure.
1592  *
1593  * Timestamps:
1594  *	dvp - ctime|mtime
1595  *	 vp - ctime (if nlink > 0)
1596  */
1597 
1598 uint64_t null_xattr = 0;
1599 
1600 /*ARGSUSED*/
1601 static int
1602 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1603     int flags)
1604 {
1605 	znode_t		*zp, *dzp = VTOZ(dvp);
1606 	znode_t		*xzp;
1607 	vnode_t		*vp;
1608 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1609 	zilog_t		*zilog;
1610 	uint64_t	acl_obj, xattr_obj;
1611 	uint64_t	xattr_obj_unlinked = 0;
1612 	uint64_t	obj = 0;
1613 	zfs_dirlock_t	*dl;
1614 	dmu_tx_t	*tx;
1615 	boolean_t	may_delete_now, delete_now = FALSE;
1616 	boolean_t	unlinked, toobig = FALSE;
1617 	uint64_t	txtype;
1618 	pathname_t	*realnmp = NULL;
1619 	pathname_t	realnm;
1620 	int		error;
1621 	int		zflg = ZEXISTS;
1622 	boolean_t	waited = B_FALSE;
1623 
1624 	ZFS_ENTER(zfsvfs);
1625 	ZFS_VERIFY_ZP(dzp);
1626 	zilog = zfsvfs->z_log;
1627 
1628 	if (flags & FIGNORECASE) {
1629 		zflg |= ZCILOOK;
1630 		pn_alloc(&realnm);
1631 		realnmp = &realnm;
1632 	}
1633 
1634 top:
1635 	xattr_obj = 0;
1636 	xzp = NULL;
1637 	/*
1638 	 * Attempt to lock directory; fail if entry doesn't exist.
1639 	 */
1640 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1641 	    NULL, realnmp)) {
1642 		if (realnmp)
1643 			pn_free(realnmp);
1644 		ZFS_EXIT(zfsvfs);
1645 		return (error);
1646 	}
1647 
1648 	vp = ZTOV(zp);
1649 
1650 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1651 		goto out;
1652 	}
1653 
1654 	/*
1655 	 * Need to use rmdir for removing directories.
1656 	 */
1657 	if (vp->v_type == VDIR) {
1658 		error = SET_ERROR(EPERM);
1659 		goto out;
1660 	}
1661 
1662 	vnevent_remove(vp, dvp, name, ct);
1663 
1664 	if (realnmp)
1665 		dnlc_remove(dvp, realnmp->pn_buf);
1666 	else
1667 		dnlc_remove(dvp, name);
1668 
1669 	mutex_enter(&vp->v_lock);
1670 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1671 	mutex_exit(&vp->v_lock);
1672 
1673 	/*
1674 	 * We may delete the znode now, or we may put it in the unlinked set;
1675 	 * it depends on whether we're the last link, and on whether there are
1676 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1677 	 * allow for either case.
1678 	 */
1679 	obj = zp->z_id;
1680 	tx = dmu_tx_create(zfsvfs->z_os);
1681 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1682 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1683 	zfs_sa_upgrade_txholds(tx, zp);
1684 	zfs_sa_upgrade_txholds(tx, dzp);
1685 	if (may_delete_now) {
1686 		toobig =
1687 		    zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1688 		/* if the file is too big, only hold_free a token amount */
1689 		dmu_tx_hold_free(tx, zp->z_id, 0,
1690 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1691 	}
1692 
1693 	/* are there any extended attributes? */
1694 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1695 	    &xattr_obj, sizeof (xattr_obj));
1696 	if (error == 0 && xattr_obj) {
1697 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1698 		ASSERT0(error);
1699 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1700 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1701 	}
1702 
1703 	mutex_enter(&zp->z_lock);
1704 	if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1705 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1706 	mutex_exit(&zp->z_lock);
1707 
1708 	/* charge as an update -- would be nice not to charge at all */
1709 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1710 
1711 	/*
1712 	 * Mark this transaction as typically resulting in a net free of
1713 	 * space, unless object removal will be delayed indefinitely
1714 	 * (due to active holds on the vnode due to the file being open).
1715 	 */
1716 	if (may_delete_now)
1717 		dmu_tx_mark_netfree(tx);
1718 
1719 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1720 	if (error) {
1721 		zfs_dirent_unlock(dl);
1722 		VN_RELE(vp);
1723 		if (xzp)
1724 			VN_RELE(ZTOV(xzp));
1725 		if (error == ERESTART) {
1726 			waited = B_TRUE;
1727 			dmu_tx_wait(tx);
1728 			dmu_tx_abort(tx);
1729 			goto top;
1730 		}
1731 		if (realnmp)
1732 			pn_free(realnmp);
1733 		dmu_tx_abort(tx);
1734 		ZFS_EXIT(zfsvfs);
1735 		return (error);
1736 	}
1737 
1738 	/*
1739 	 * Remove the directory entry.
1740 	 */
1741 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1742 
1743 	if (error) {
1744 		dmu_tx_commit(tx);
1745 		goto out;
1746 	}
1747 
1748 	if (unlinked) {
1749 		/*
1750 		 * Hold z_lock so that we can make sure that the ACL obj
1751 		 * hasn't changed.  Could have been deleted due to
1752 		 * zfs_sa_upgrade().
1753 		 */
1754 		mutex_enter(&zp->z_lock);
1755 		mutex_enter(&vp->v_lock);
1756 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1757 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1758 		delete_now = may_delete_now && !toobig &&
1759 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1760 		    xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1761 		    acl_obj;
1762 		mutex_exit(&vp->v_lock);
1763 	}
1764 
1765 	txtype = TX_REMOVE;
1766 	if (flags & FIGNORECASE)
1767 		txtype |= TX_CI;
1768 	rw_enter(&rz_zev_rwlock, RW_READER);
1769 	if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_znode_remove)
1770 		rz_zev_callbacks->rz_zev_znode_remove(dzp, zp, tx,
1771 		                                      name, txtype);
1772 	rw_exit(&rz_zev_rwlock);
1773 
1774 	if (delete_now) {
1775 		if (xattr_obj_unlinked) {
1776 			ASSERT3U(xzp->z_links, ==, 2);
1777 			mutex_enter(&xzp->z_lock);
1778 			xzp->z_unlinked = 1;
1779 			xzp->z_links = 0;
1780 			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1781 			    &xzp->z_links, sizeof (xzp->z_links), tx);
1782 			ASSERT3U(error,  ==,  0);
1783 			mutex_exit(&xzp->z_lock);
1784 			zfs_unlinked_add(xzp, tx);
1785 
1786 			if (zp->z_is_sa)
1787 				error = sa_remove(zp->z_sa_hdl,
1788 				    SA_ZPL_XATTR(zfsvfs), tx);
1789 			else
1790 				error = sa_update(zp->z_sa_hdl,
1791 				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
1792 				    sizeof (uint64_t), tx);
1793 			ASSERT0(error);
1794 		}
1795 		mutex_enter(&vp->v_lock);
1796 		vp->v_count--;
1797 		ASSERT0(vp->v_count);
1798 		mutex_exit(&vp->v_lock);
1799 		mutex_exit(&zp->z_lock);
1800 		zfs_znode_delete(zp, tx);
1801 	} else if (unlinked) {
1802 		mutex_exit(&zp->z_lock);
1803 		zfs_unlinked_add(zp, tx);
1804 	}
1805 
1806 	txtype = TX_REMOVE;
1807 	if (flags & FIGNORECASE)
1808 		txtype |= TX_CI;
1809 	zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1810 
1811 	dmu_tx_commit(tx);
1812 out:
1813 	if (realnmp)
1814 		pn_free(realnmp);
1815 
1816 	zfs_dirent_unlock(dl);
1817 
1818 	if (!delete_now)
1819 		VN_RELE(vp);
1820 	if (xzp)
1821 		VN_RELE(ZTOV(xzp));
1822 
1823 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1824 		zil_commit(zilog, 0);
1825 
1826 	ZFS_EXIT(zfsvfs);
1827 	return (error);
1828 }
1829 
1830 /*
1831  * Create a new directory and insert it into dvp using the name
1832  * provided.  Return a pointer to the inserted directory.
1833  *
1834  *	IN:	dvp	- vnode of directory to add subdir to.
1835  *		dirname	- name of new directory.
1836  *		vap	- attributes of new directory.
1837  *		cr	- credentials of caller.
1838  *		ct	- caller context
1839  *		flags	- case flags
1840  *		vsecp	- ACL to be set
1841  *
1842  *	OUT:	vpp	- vnode of created directory.
1843  *
1844  *	RETURN:	0 on success, error code on failure.
1845  *
1846  * Timestamps:
1847  *	dvp - ctime|mtime updated
1848  *	 vp - ctime|mtime|atime updated
1849  */
1850 /*ARGSUSED*/
1851 static int
1852 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1853     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1854 {
1855 	znode_t		*zp, *dzp = VTOZ(dvp);
1856 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1857 	zilog_t		*zilog;
1858 	zfs_dirlock_t	*dl;
1859 	uint64_t	txtype;
1860 	dmu_tx_t	*tx;
1861 	int		error;
1862 	int		zf = ZNEW;
1863 	ksid_t		*ksid;
1864 	uid_t		uid;
1865 	gid_t		gid = crgetgid(cr);
1866 	zfs_acl_ids_t   acl_ids;
1867 	boolean_t	fuid_dirtied;
1868 	boolean_t	waited = B_FALSE;
1869 
1870 	ASSERT(vap->va_type == VDIR);
1871 
1872 	/*
1873 	 * If we have an ephemeral id, ACL, or XVATTR then
1874 	 * make sure file system is at proper version
1875 	 */
1876 
1877 	ksid = crgetsid(cr, KSID_OWNER);
1878 	if (ksid)
1879 		uid = ksid_getid(ksid);
1880 	else
1881 		uid = crgetuid(cr);
1882 	if (zfsvfs->z_use_fuids == B_FALSE &&
1883 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1884 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1885 		return (SET_ERROR(EINVAL));
1886 
1887 	ZFS_ENTER(zfsvfs);
1888 	ZFS_VERIFY_ZP(dzp);
1889 	zilog = zfsvfs->z_log;
1890 
1891 	if (dzp->z_pflags & ZFS_XATTR) {
1892 		ZFS_EXIT(zfsvfs);
1893 		return (SET_ERROR(EINVAL));
1894 	}
1895 
1896 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1897 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1898 		ZFS_EXIT(zfsvfs);
1899 		return (SET_ERROR(EILSEQ));
1900 	}
1901 	if (flags & FIGNORECASE)
1902 		zf |= ZCILOOK;
1903 
1904 	if (vap->va_mask & AT_XVATTR) {
1905 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1906 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1907 			ZFS_EXIT(zfsvfs);
1908 			return (error);
1909 		}
1910 	}
1911 
1912 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1913 	    vsecp, &acl_ids)) != 0) {
1914 		ZFS_EXIT(zfsvfs);
1915 		return (error);
1916 	}
1917 	/*
1918 	 * First make sure the new directory doesn't exist.
1919 	 *
1920 	 * Existence is checked first to make sure we don't return
1921 	 * EACCES instead of EEXIST which can cause some applications
1922 	 * to fail.
1923 	 */
1924 top:
1925 	*vpp = NULL;
1926 
1927 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1928 	    NULL, NULL)) {
1929 		zfs_acl_ids_free(&acl_ids);
1930 		ZFS_EXIT(zfsvfs);
1931 		return (error);
1932 	}
1933 
1934 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1935 		zfs_acl_ids_free(&acl_ids);
1936 		zfs_dirent_unlock(dl);
1937 		ZFS_EXIT(zfsvfs);
1938 		return (error);
1939 	}
1940 
1941 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1942 		zfs_acl_ids_free(&acl_ids);
1943 		zfs_dirent_unlock(dl);
1944 		ZFS_EXIT(zfsvfs);
1945 		return (SET_ERROR(EDQUOT));
1946 	}
1947 
1948 	/*
1949 	 * Add a new entry to the directory.
1950 	 */
1951 	tx = dmu_tx_create(zfsvfs->z_os);
1952 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1953 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1954 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1955 	if (fuid_dirtied)
1956 		zfs_fuid_txhold(zfsvfs, tx);
1957 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1958 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1959 		    acl_ids.z_aclp->z_acl_bytes);
1960 	}
1961 
1962 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1963 	    ZFS_SA_BASE_ATTR_SIZE);
1964 
1965 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1966 	if (error) {
1967 		zfs_dirent_unlock(dl);
1968 		if (error == ERESTART) {
1969 			waited = B_TRUE;
1970 			dmu_tx_wait(tx);
1971 			dmu_tx_abort(tx);
1972 			goto top;
1973 		}
1974 		zfs_acl_ids_free(&acl_ids);
1975 		dmu_tx_abort(tx);
1976 		ZFS_EXIT(zfsvfs);
1977 		return (error);
1978 	}
1979 
1980 	/*
1981 	 * Create new node.
1982 	 */
1983 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1984 
1985 	if (fuid_dirtied)
1986 		zfs_fuid_sync(zfsvfs, tx);
1987 
1988 	/*
1989 	 * Now put new name in parent dir.
1990 	 */
1991 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1992 
1993 	*vpp = ZTOV(zp);
1994 
1995 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1996 	if (flags & FIGNORECASE)
1997 		txtype |= TX_CI;
1998 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1999 	    acl_ids.z_fuidp, vap);
2000 
2001 	zfs_acl_ids_free(&acl_ids);
2002 
2003 	dmu_tx_commit(tx);
2004 
2005 	zfs_dirent_unlock(dl);
2006 
2007 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2008 		zil_commit(zilog, 0);
2009 
2010 	ZFS_EXIT(zfsvfs);
2011 	return (0);
2012 }
2013 
2014 /*
2015  * Remove a directory subdir entry.  If the current working
2016  * directory is the same as the subdir to be removed, the
2017  * remove will fail.
2018  *
2019  *	IN:	dvp	- vnode of directory to remove from.
2020  *		name	- name of directory to be removed.
2021  *		cwd	- vnode of current working directory.
2022  *		cr	- credentials of caller.
2023  *		ct	- caller context
2024  *		flags	- case flags
2025  *
2026  *	RETURN:	0 on success, error code on failure.
2027  *
2028  * Timestamps:
2029  *	dvp - ctime|mtime updated
2030  */
2031 /*ARGSUSED*/
2032 static int
2033 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2034     caller_context_t *ct, int flags)
2035 {
2036 	znode_t		*dzp = VTOZ(dvp);
2037 	znode_t		*zp;
2038 	vnode_t		*vp;
2039 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2040 	zilog_t		*zilog;
2041 	zfs_dirlock_t	*dl;
2042 	dmu_tx_t	*tx;
2043 	int		error;
2044 	int		zflg = ZEXISTS;
2045 	boolean_t	waited = B_FALSE;
2046 
2047 	ZFS_ENTER(zfsvfs);
2048 	ZFS_VERIFY_ZP(dzp);
2049 	zilog = zfsvfs->z_log;
2050 
2051 	if (flags & FIGNORECASE)
2052 		zflg |= ZCILOOK;
2053 top:
2054 	zp = NULL;
2055 
2056 	/*
2057 	 * Attempt to lock directory; fail if entry doesn't exist.
2058 	 */
2059 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2060 	    NULL, NULL)) {
2061 		ZFS_EXIT(zfsvfs);
2062 		return (error);
2063 	}
2064 
2065 	vp = ZTOV(zp);
2066 
2067 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2068 		goto out;
2069 	}
2070 
2071 	if (vp->v_type != VDIR) {
2072 		error = SET_ERROR(ENOTDIR);
2073 		goto out;
2074 	}
2075 
2076 	if (vp == cwd) {
2077 		error = SET_ERROR(EINVAL);
2078 		goto out;
2079 	}
2080 
2081 	vnevent_rmdir(vp, dvp, name, ct);
2082 
2083 	/*
2084 	 * Grab a lock on the directory to make sure that noone is
2085 	 * trying to add (or lookup) entries while we are removing it.
2086 	 */
2087 	rw_enter(&zp->z_name_lock, RW_WRITER);
2088 
2089 	/*
2090 	 * Grab a lock on the parent pointer to make sure we play well
2091 	 * with the treewalk and directory rename code.
2092 	 */
2093 	rw_enter(&zp->z_parent_lock, RW_WRITER);
2094 
2095 	tx = dmu_tx_create(zfsvfs->z_os);
2096 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2097 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2098 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2099 	zfs_sa_upgrade_txholds(tx, zp);
2100 	zfs_sa_upgrade_txholds(tx, dzp);
2101 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2102 	if (error) {
2103 		rw_exit(&zp->z_parent_lock);
2104 		rw_exit(&zp->z_name_lock);
2105 		zfs_dirent_unlock(dl);
2106 		VN_RELE(vp);
2107 		if (error == ERESTART) {
2108 			waited = B_TRUE;
2109 			dmu_tx_wait(tx);
2110 			dmu_tx_abort(tx);
2111 			goto top;
2112 		}
2113 		dmu_tx_abort(tx);
2114 		ZFS_EXIT(zfsvfs);
2115 		return (error);
2116 	}
2117 
2118 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2119 
2120 	if (error == 0) {
2121 		uint64_t txtype = TX_RMDIR;
2122 		if (flags & FIGNORECASE)
2123 			txtype |= TX_CI;
2124 
2125 		rw_enter(&rz_zev_rwlock, RW_READER);
2126 		if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_znode_remove)
2127 			rz_zev_callbacks->rz_zev_znode_remove(dzp, zp, tx,
2128 			                                      name, txtype);
2129 		rw_exit(&rz_zev_rwlock);
2130 
2131 		zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2132 	}
2133 
2134 	dmu_tx_commit(tx);
2135 
2136 	rw_exit(&zp->z_parent_lock);
2137 	rw_exit(&zp->z_name_lock);
2138 out:
2139 	zfs_dirent_unlock(dl);
2140 
2141 	VN_RELE(vp);
2142 
2143 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2144 		zil_commit(zilog, 0);
2145 
2146 	ZFS_EXIT(zfsvfs);
2147 	return (error);
2148 }
2149 
2150 /*
2151  * Read as many directory entries as will fit into the provided
2152  * buffer from the given directory cursor position (specified in
2153  * the uio structure).
2154  *
2155  *	IN:	vp	- vnode of directory to read.
2156  *		uio	- structure supplying read location, range info,
2157  *			  and return buffer.
2158  *		cr	- credentials of caller.
2159  *		ct	- caller context
2160  *		flags	- case flags
2161  *
2162  *	OUT:	uio	- updated offset and range, buffer filled.
2163  *		eofp	- set to true if end-of-file detected.
2164  *
2165  *	RETURN:	0 on success, error code on failure.
2166  *
2167  * Timestamps:
2168  *	vp - atime updated
2169  *
2170  * Note that the low 4 bits of the cookie returned by zap is always zero.
2171  * This allows us to use the low range for "special" directory entries:
2172  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2173  * we use the offset 2 for the '.zfs' directory.
2174  */
2175 /* ARGSUSED */
2176 static int
2177 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2178     caller_context_t *ct, int flags)
2179 {
2180 	znode_t		*zp = VTOZ(vp);
2181 	iovec_t		*iovp;
2182 	edirent_t	*eodp;
2183 	dirent64_t	*odp;
2184 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2185 	objset_t	*os;
2186 	caddr_t		outbuf;
2187 	size_t		bufsize;
2188 	zap_cursor_t	zc;
2189 	zap_attribute_t	zap;
2190 	uint_t		bytes_wanted;
2191 	uint64_t	offset; /* must be unsigned; checks for < 1 */
2192 	uint64_t	parent;
2193 	int		local_eof;
2194 	int		outcount;
2195 	int		error;
2196 	uint8_t		prefetch;
2197 	boolean_t	check_sysattrs;
2198 
2199 	ZFS_ENTER(zfsvfs);
2200 	ZFS_VERIFY_ZP(zp);
2201 
2202 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2203 	    &parent, sizeof (parent))) != 0) {
2204 		ZFS_EXIT(zfsvfs);
2205 		return (error);
2206 	}
2207 
2208 	/*
2209 	 * If we are not given an eof variable,
2210 	 * use a local one.
2211 	 */
2212 	if (eofp == NULL)
2213 		eofp = &local_eof;
2214 
2215 	/*
2216 	 * Check for valid iov_len.
2217 	 */
2218 	if (uio->uio_iov->iov_len <= 0) {
2219 		ZFS_EXIT(zfsvfs);
2220 		return (SET_ERROR(EINVAL));
2221 	}
2222 
2223 	/*
2224 	 * Quit if directory has been removed (posix)
2225 	 */
2226 	if ((*eofp = zp->z_unlinked) != 0) {
2227 		ZFS_EXIT(zfsvfs);
2228 		return (0);
2229 	}
2230 
2231 	error = 0;
2232 	os = zfsvfs->z_os;
2233 	offset = uio->uio_loffset;
2234 	prefetch = zp->z_zn_prefetch;
2235 
2236 	/*
2237 	 * Initialize the iterator cursor.
2238 	 */
2239 	if (offset <= 3) {
2240 		/*
2241 		 * Start iteration from the beginning of the directory.
2242 		 */
2243 		zap_cursor_init(&zc, os, zp->z_id);
2244 	} else {
2245 		/*
2246 		 * The offset is a serialized cursor.
2247 		 */
2248 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2249 	}
2250 
2251 	/*
2252 	 * Get space to change directory entries into fs independent format.
2253 	 */
2254 	iovp = uio->uio_iov;
2255 	bytes_wanted = iovp->iov_len;
2256 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2257 		bufsize = bytes_wanted;
2258 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
2259 		odp = (struct dirent64 *)outbuf;
2260 	} else {
2261 		bufsize = bytes_wanted;
2262 		outbuf = NULL;
2263 		odp = (struct dirent64 *)iovp->iov_base;
2264 	}
2265 	eodp = (struct edirent *)odp;
2266 
2267 	/*
2268 	 * If this VFS supports the system attribute view interface; and
2269 	 * we're looking at an extended attribute directory; and we care
2270 	 * about normalization conflicts on this vfs; then we must check
2271 	 * for normalization conflicts with the sysattr name space.
2272 	 */
2273 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2274 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2275 	    (flags & V_RDDIR_ENTFLAGS);
2276 
2277 	/*
2278 	 * Transform to file-system independent format
2279 	 */
2280 	outcount = 0;
2281 	while (outcount < bytes_wanted) {
2282 		ino64_t objnum;
2283 		ushort_t reclen;
2284 		off64_t *next = NULL;
2285 
2286 		/*
2287 		 * Special case `.', `..', and `.zfs'.
2288 		 */
2289 		if (offset == 0) {
2290 			(void) strcpy(zap.za_name, ".");
2291 			zap.za_normalization_conflict = 0;
2292 			objnum = zp->z_id;
2293 		} else if (offset == 1) {
2294 			(void) strcpy(zap.za_name, "..");
2295 			zap.za_normalization_conflict = 0;
2296 			objnum = parent;
2297 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2298 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2299 			zap.za_normalization_conflict = 0;
2300 			objnum = ZFSCTL_INO_ROOT;
2301 		} else {
2302 			/*
2303 			 * Grab next entry.
2304 			 */
2305 			if (error = zap_cursor_retrieve(&zc, &zap)) {
2306 				if ((*eofp = (error == ENOENT)) != 0)
2307 					break;
2308 				else
2309 					goto update;
2310 			}
2311 
2312 			if (zap.za_integer_length != 8 ||
2313 			    zap.za_num_integers != 1) {
2314 				cmn_err(CE_WARN, "zap_readdir: bad directory "
2315 				    "entry, obj = %lld, offset = %lld\n",
2316 				    (u_longlong_t)zp->z_id,
2317 				    (u_longlong_t)offset);
2318 				error = SET_ERROR(ENXIO);
2319 				goto update;
2320 			}
2321 
2322 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2323 			/*
2324 			 * MacOS X can extract the object type here such as:
2325 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2326 			 */
2327 
2328 			if (check_sysattrs && !zap.za_normalization_conflict) {
2329 				zap.za_normalization_conflict =
2330 				    xattr_sysattr_casechk(zap.za_name);
2331 			}
2332 		}
2333 
2334 		if (flags & V_RDDIR_ACCFILTER) {
2335 			/*
2336 			 * If we have no access at all, don't include
2337 			 * this entry in the returned information
2338 			 */
2339 			znode_t	*ezp;
2340 			if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2341 				goto skip_entry;
2342 			if (!zfs_has_access(ezp, cr)) {
2343 				VN_RELE(ZTOV(ezp));
2344 				goto skip_entry;
2345 			}
2346 			VN_RELE(ZTOV(ezp));
2347 		}
2348 
2349 		if (flags & V_RDDIR_ENTFLAGS)
2350 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2351 		else
2352 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2353 
2354 		/*
2355 		 * Will this entry fit in the buffer?
2356 		 */
2357 		if (outcount + reclen > bufsize) {
2358 			/*
2359 			 * Did we manage to fit anything in the buffer?
2360 			 */
2361 			if (!outcount) {
2362 				error = SET_ERROR(EINVAL);
2363 				goto update;
2364 			}
2365 			break;
2366 		}
2367 		if (flags & V_RDDIR_ENTFLAGS) {
2368 			/*
2369 			 * Add extended flag entry:
2370 			 */
2371 			eodp->ed_ino = objnum;
2372 			eodp->ed_reclen = reclen;
2373 			/* NOTE: ed_off is the offset for the *next* entry */
2374 			next = &(eodp->ed_off);
2375 			eodp->ed_eflags = zap.za_normalization_conflict ?
2376 			    ED_CASE_CONFLICT : 0;
2377 			(void) strncpy(eodp->ed_name, zap.za_name,
2378 			    EDIRENT_NAMELEN(reclen));
2379 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2380 		} else {
2381 			/*
2382 			 * Add normal entry:
2383 			 */
2384 			odp->d_ino = objnum;
2385 			odp->d_reclen = reclen;
2386 			/* NOTE: d_off is the offset for the *next* entry */
2387 			next = &(odp->d_off);
2388 			(void) strncpy(odp->d_name, zap.za_name,
2389 			    DIRENT64_NAMELEN(reclen));
2390 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2391 		}
2392 		outcount += reclen;
2393 
2394 		ASSERT(outcount <= bufsize);
2395 
2396 		/* Prefetch znode */
2397 		if (prefetch)
2398 			dmu_prefetch(os, objnum, 0, 0);
2399 
2400 	skip_entry:
2401 		/*
2402 		 * Move to the next entry, fill in the previous offset.
2403 		 */
2404 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2405 			zap_cursor_advance(&zc);
2406 			offset = zap_cursor_serialize(&zc);
2407 		} else {
2408 			offset += 1;
2409 		}
2410 		if (next)
2411 			*next = offset;
2412 	}
2413 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2414 
2415 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2416 		iovp->iov_base += outcount;
2417 		iovp->iov_len -= outcount;
2418 		uio->uio_resid -= outcount;
2419 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2420 		/*
2421 		 * Reset the pointer.
2422 		 */
2423 		offset = uio->uio_loffset;
2424 	}
2425 
2426 update:
2427 	zap_cursor_fini(&zc);
2428 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2429 		kmem_free(outbuf, bufsize);
2430 
2431 	if (error == ENOENT)
2432 		error = 0;
2433 
2434 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2435 
2436 	uio->uio_loffset = offset;
2437 	ZFS_EXIT(zfsvfs);
2438 	return (error);
2439 }
2440 
2441 ulong_t zfs_fsync_sync_cnt = 4;
2442 
2443 static int
2444 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2445 {
2446 	znode_t	*zp = VTOZ(vp);
2447 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2448 
2449 	/*
2450 	 * Regardless of whether this is required for standards conformance,
2451 	 * this is the logical behavior when fsync() is called on a file with
2452 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2453 	 * going to be pushed out as part of the zil_commit().
2454 	 */
2455 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2456 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2457 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2458 
2459 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2460 
2461 	if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2462 		ZFS_ENTER(zfsvfs);
2463 		ZFS_VERIFY_ZP(zp);
2464 		zil_commit(zfsvfs->z_log, zp->z_id);
2465 		ZFS_EXIT(zfsvfs);
2466 	}
2467 	return (0);
2468 }
2469 
2470 
2471 /*
2472  * Get the requested file attributes and place them in the provided
2473  * vattr structure.
2474  *
2475  *	IN:	vp	- vnode of file.
2476  *		vap	- va_mask identifies requested attributes.
2477  *			  If AT_XVATTR set, then optional attrs are requested
2478  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2479  *		cr	- credentials of caller.
2480  *		ct	- caller context
2481  *
2482  *	OUT:	vap	- attribute values.
2483  *
2484  *	RETURN:	0 (always succeeds).
2485  */
2486 /* ARGSUSED */
2487 static int
2488 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2489     caller_context_t *ct)
2490 {
2491 	znode_t *zp = VTOZ(vp);
2492 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2493 	int	error = 0;
2494 	uint64_t links;
2495 	uint64_t mtime[2], ctime[2];
2496 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2497 	xoptattr_t *xoap = NULL;
2498 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2499 	sa_bulk_attr_t bulk[2];
2500 	int count = 0;
2501 
2502 	ZFS_ENTER(zfsvfs);
2503 	ZFS_VERIFY_ZP(zp);
2504 
2505 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2506 
2507 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2508 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2509 
2510 	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2511 		ZFS_EXIT(zfsvfs);
2512 		return (error);
2513 	}
2514 
2515 	/*
2516 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2517 	 * Also, if we are the owner don't bother, since owner should
2518 	 * always be allowed to read basic attributes of file.
2519 	 */
2520 	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2521 	    (vap->va_uid != crgetuid(cr))) {
2522 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2523 		    skipaclchk, cr)) {
2524 			ZFS_EXIT(zfsvfs);
2525 			return (error);
2526 		}
2527 	}
2528 
2529 	/*
2530 	 * Return all attributes.  It's cheaper to provide the answer
2531 	 * than to determine whether we were asked the question.
2532 	 */
2533 
2534 	mutex_enter(&zp->z_lock);
2535 	vap->va_type = vp->v_type;
2536 	vap->va_mode = zp->z_mode & MODEMASK;
2537 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2538 	vap->va_nodeid = zp->z_id;
2539 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2540 		links = zp->z_links + 1;
2541 	else
2542 		links = zp->z_links;
2543 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2544 	vap->va_size = zp->z_size;
2545 	vap->va_rdev = vp->v_rdev;
2546 	vap->va_seq = zp->z_seq;
2547 
2548 	/*
2549 	 * Add in any requested optional attributes and the create time.
2550 	 * Also set the corresponding bits in the returned attribute bitmap.
2551 	 */
2552 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2553 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2554 			xoap->xoa_archive =
2555 			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2556 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2557 		}
2558 
2559 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2560 			xoap->xoa_readonly =
2561 			    ((zp->z_pflags & ZFS_READONLY) != 0);
2562 			XVA_SET_RTN(xvap, XAT_READONLY);
2563 		}
2564 
2565 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2566 			xoap->xoa_system =
2567 			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2568 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2569 		}
2570 
2571 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2572 			xoap->xoa_hidden =
2573 			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2574 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2575 		}
2576 
2577 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2578 			xoap->xoa_nounlink =
2579 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2580 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2581 		}
2582 
2583 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2584 			xoap->xoa_immutable =
2585 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2586 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2587 		}
2588 
2589 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2590 			xoap->xoa_appendonly =
2591 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2592 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2593 		}
2594 
2595 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2596 			xoap->xoa_nodump =
2597 			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2598 			XVA_SET_RTN(xvap, XAT_NODUMP);
2599 		}
2600 
2601 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2602 			xoap->xoa_opaque =
2603 			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2604 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2605 		}
2606 
2607 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2608 			xoap->xoa_av_quarantined =
2609 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2610 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2611 		}
2612 
2613 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2614 			xoap->xoa_av_modified =
2615 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2616 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2617 		}
2618 
2619 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2620 		    vp->v_type == VREG) {
2621 			zfs_sa_get_scanstamp(zp, xvap);
2622 		}
2623 
2624 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2625 			uint64_t times[2];
2626 
2627 			(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2628 			    times, sizeof (times));
2629 			ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2630 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2631 		}
2632 
2633 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2634 			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2635 			XVA_SET_RTN(xvap, XAT_REPARSE);
2636 		}
2637 		if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2638 			xoap->xoa_generation = zp->z_gen;
2639 			XVA_SET_RTN(xvap, XAT_GEN);
2640 		}
2641 
2642 		if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2643 			xoap->xoa_offline =
2644 			    ((zp->z_pflags & ZFS_OFFLINE) != 0);
2645 			XVA_SET_RTN(xvap, XAT_OFFLINE);
2646 		}
2647 
2648 		if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2649 			xoap->xoa_sparse =
2650 			    ((zp->z_pflags & ZFS_SPARSE) != 0);
2651 			XVA_SET_RTN(xvap, XAT_SPARSE);
2652 		}
2653 	}
2654 
2655 	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2656 	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2657 	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2658 
2659 	mutex_exit(&zp->z_lock);
2660 
2661 	sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2662 
2663 	if (zp->z_blksz == 0) {
2664 		/*
2665 		 * Block size hasn't been set; suggest maximal I/O transfers.
2666 		 */
2667 		vap->va_blksize = zfsvfs->z_max_blksz;
2668 	}
2669 
2670 	ZFS_EXIT(zfsvfs);
2671 	return (0);
2672 }
2673 
2674 /*
2675  * Set the file attributes to the values contained in the
2676  * vattr structure.
2677  *
2678  *	IN:	vp	- vnode of file to be modified.
2679  *		vap	- new attribute values.
2680  *			  If AT_XVATTR set, then optional attrs are being set
2681  *		flags	- ATTR_UTIME set if non-default time values provided.
2682  *			- ATTR_NOACLCHECK (CIFS context only).
2683  *		cr	- credentials of caller.
2684  *		ct	- caller context
2685  *
2686  *	RETURN:	0 on success, error code on failure.
2687  *
2688  * Timestamps:
2689  *	vp - ctime updated, mtime updated if size changed.
2690  */
2691 /* ARGSUSED */
2692 static int
2693 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2694     caller_context_t *ct)
2695 {
2696 	znode_t		*zp = VTOZ(vp);
2697 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2698 	zilog_t		*zilog;
2699 	dmu_tx_t	*tx;
2700 	vattr_t		oldva;
2701 	xvattr_t	tmpxvattr;
2702 	uint_t		mask = vap->va_mask;
2703 	uint_t		saved_mask = 0;
2704 	int		trim_mask = 0;
2705 	uint64_t	new_mode;
2706 	uint64_t	new_uid, new_gid;
2707 	uint64_t	xattr_obj;
2708 	uint64_t	mtime[2], ctime[2];
2709 	znode_t		*attrzp;
2710 	int		need_policy = FALSE;
2711 	int		err, err2;
2712 	zfs_fuid_info_t *fuidp = NULL;
2713 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2714 	xoptattr_t	*xoap;
2715 	zfs_acl_t	*aclp;
2716 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2717 	boolean_t	fuid_dirtied = B_FALSE;
2718 	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
2719 	int		count = 0, xattr_count = 0;
2720 
2721 	if (mask == 0)
2722 		return (0);
2723 
2724 	if (mask & AT_NOSET)
2725 		return (SET_ERROR(EINVAL));
2726 
2727 	ZFS_ENTER(zfsvfs);
2728 	ZFS_VERIFY_ZP(zp);
2729 
2730 	zilog = zfsvfs->z_log;
2731 
2732 	/*
2733 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2734 	 * that file system is at proper version level
2735 	 */
2736 
2737 	if (zfsvfs->z_use_fuids == B_FALSE &&
2738 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2739 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2740 	    (mask & AT_XVATTR))) {
2741 		ZFS_EXIT(zfsvfs);
2742 		return (SET_ERROR(EINVAL));
2743 	}
2744 
2745 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2746 		ZFS_EXIT(zfsvfs);
2747 		return (SET_ERROR(EISDIR));
2748 	}
2749 
2750 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2751 		ZFS_EXIT(zfsvfs);
2752 		return (SET_ERROR(EINVAL));
2753 	}
2754 
2755 	/*
2756 	 * If this is an xvattr_t, then get a pointer to the structure of
2757 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2758 	 */
2759 	xoap = xva_getxoptattr(xvap);
2760 
2761 	xva_init(&tmpxvattr);
2762 
2763 	/*
2764 	 * Immutable files can only alter immutable bit and atime
2765 	 */
2766 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2767 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2768 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2769 		ZFS_EXIT(zfsvfs);
2770 		return (SET_ERROR(EPERM));
2771 	}
2772 
2773 	if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2774 		ZFS_EXIT(zfsvfs);
2775 		return (SET_ERROR(EPERM));
2776 	}
2777 
2778 	/*
2779 	 * Verify timestamps doesn't overflow 32 bits.
2780 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2781 	 * handle times greater than 2039.  This check should be removed
2782 	 * once large timestamps are fully supported.
2783 	 */
2784 	if (mask & (AT_ATIME | AT_MTIME)) {
2785 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2786 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2787 			ZFS_EXIT(zfsvfs);
2788 			return (SET_ERROR(EOVERFLOW));
2789 		}
2790 	}
2791 
2792 top:
2793 	attrzp = NULL;
2794 	aclp = NULL;
2795 
2796 	/* Can this be moved to before the top label? */
2797 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2798 		ZFS_EXIT(zfsvfs);
2799 		return (SET_ERROR(EROFS));
2800 	}
2801 
2802 	/*
2803 	 * First validate permissions
2804 	 */
2805 
2806 	if (mask & AT_SIZE) {
2807 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2808 		if (err) {
2809 			ZFS_EXIT(zfsvfs);
2810 			return (err);
2811 		}
2812 		/*
2813 		 * XXX - Note, we are not providing any open
2814 		 * mode flags here (like FNDELAY), so we may
2815 		 * block if there are locks present... this
2816 		 * should be addressed in openat().
2817 		 */
2818 		/* XXX - would it be OK to generate a log record here? */
2819 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2820 		if (err) {
2821 			ZFS_EXIT(zfsvfs);
2822 			return (err);
2823 		}
2824 
2825 		if (vap->va_size == 0)
2826 			vnevent_truncate(ZTOV(zp), ct);
2827 	}
2828 
2829 	if (mask & (AT_ATIME|AT_MTIME) ||
2830 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2831 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2832 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2833 	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2834 	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2835 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2836 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2837 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2838 		    skipaclchk, cr);
2839 	}
2840 
2841 	if (mask & (AT_UID|AT_GID)) {
2842 		int	idmask = (mask & (AT_UID|AT_GID));
2843 		int	take_owner;
2844 		int	take_group;
2845 
2846 		/*
2847 		 * NOTE: even if a new mode is being set,
2848 		 * we may clear S_ISUID/S_ISGID bits.
2849 		 */
2850 
2851 		if (!(mask & AT_MODE))
2852 			vap->va_mode = zp->z_mode;
2853 
2854 		/*
2855 		 * Take ownership or chgrp to group we are a member of
2856 		 */
2857 
2858 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2859 		take_group = (mask & AT_GID) &&
2860 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2861 
2862 		/*
2863 		 * If both AT_UID and AT_GID are set then take_owner and
2864 		 * take_group must both be set in order to allow taking
2865 		 * ownership.
2866 		 *
2867 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2868 		 *
2869 		 */
2870 
2871 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2872 		    ((idmask == AT_UID) && take_owner) ||
2873 		    ((idmask == AT_GID) && take_group)) {
2874 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2875 			    skipaclchk, cr) == 0) {
2876 				/*
2877 				 * Remove setuid/setgid for non-privileged users
2878 				 */
2879 				secpolicy_setid_clear(vap, cr);
2880 				trim_mask = (mask & (AT_UID|AT_GID));
2881 			} else {
2882 				need_policy =  TRUE;
2883 			}
2884 		} else {
2885 			need_policy =  TRUE;
2886 		}
2887 	}
2888 
2889 	mutex_enter(&zp->z_lock);
2890 	oldva.va_mode = zp->z_mode;
2891 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2892 	if (mask & AT_XVATTR) {
2893 		/*
2894 		 * Update xvattr mask to include only those attributes
2895 		 * that are actually changing.
2896 		 *
2897 		 * the bits will be restored prior to actually setting
2898 		 * the attributes so the caller thinks they were set.
2899 		 */
2900 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2901 			if (xoap->xoa_appendonly !=
2902 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2903 				need_policy = TRUE;
2904 			} else {
2905 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2906 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2907 			}
2908 		}
2909 
2910 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2911 			if (xoap->xoa_nounlink !=
2912 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2913 				need_policy = TRUE;
2914 			} else {
2915 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2916 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2917 			}
2918 		}
2919 
2920 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2921 			if (xoap->xoa_immutable !=
2922 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2923 				need_policy = TRUE;
2924 			} else {
2925 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2926 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2927 			}
2928 		}
2929 
2930 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2931 			if (xoap->xoa_nodump !=
2932 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2933 				need_policy = TRUE;
2934 			} else {
2935 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2936 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2937 			}
2938 		}
2939 
2940 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2941 			if (xoap->xoa_av_modified !=
2942 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2943 				need_policy = TRUE;
2944 			} else {
2945 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2946 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2947 			}
2948 		}
2949 
2950 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2951 			if ((vp->v_type != VREG &&
2952 			    xoap->xoa_av_quarantined) ||
2953 			    xoap->xoa_av_quarantined !=
2954 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2955 				need_policy = TRUE;
2956 			} else {
2957 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2958 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2959 			}
2960 		}
2961 
2962 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2963 			mutex_exit(&zp->z_lock);
2964 			ZFS_EXIT(zfsvfs);
2965 			return (SET_ERROR(EPERM));
2966 		}
2967 
2968 		if (need_policy == FALSE &&
2969 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2970 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2971 			need_policy = TRUE;
2972 		}
2973 	}
2974 
2975 	mutex_exit(&zp->z_lock);
2976 
2977 	if (mask & AT_MODE) {
2978 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2979 			err = secpolicy_setid_setsticky_clear(vp, vap,
2980 			    &oldva, cr);
2981 			if (err) {
2982 				ZFS_EXIT(zfsvfs);
2983 				return (err);
2984 			}
2985 			trim_mask |= AT_MODE;
2986 		} else {
2987 			need_policy = TRUE;
2988 		}
2989 	}
2990 
2991 	if (need_policy) {
2992 		/*
2993 		 * If trim_mask is set then take ownership
2994 		 * has been granted or write_acl is present and user
2995 		 * has the ability to modify mode.  In that case remove
2996 		 * UID|GID and or MODE from mask so that
2997 		 * secpolicy_vnode_setattr() doesn't revoke it.
2998 		 */
2999 
3000 		if (trim_mask) {
3001 			saved_mask = vap->va_mask;
3002 			vap->va_mask &= ~trim_mask;
3003 		}
3004 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3005 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3006 		if (err) {
3007 			ZFS_EXIT(zfsvfs);
3008 			return (err);
3009 		}
3010 
3011 		if (trim_mask)
3012 			vap->va_mask |= saved_mask;
3013 	}
3014 
3015 	/*
3016 	 * secpolicy_vnode_setattr, or take ownership may have
3017 	 * changed va_mask
3018 	 */
3019 	mask = vap->va_mask;
3020 
3021 	if ((mask & (AT_UID | AT_GID))) {
3022 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3023 		    &xattr_obj, sizeof (xattr_obj));
3024 
3025 		if (err == 0 && xattr_obj) {
3026 			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3027 			if (err)
3028 				goto out2;
3029 		}
3030 		if (mask & AT_UID) {
3031 			new_uid = zfs_fuid_create(zfsvfs,
3032 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3033 			if (new_uid != zp->z_uid &&
3034 			    zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3035 				if (attrzp)
3036 					VN_RELE(ZTOV(attrzp));
3037 				err = SET_ERROR(EDQUOT);
3038 				goto out2;
3039 			}
3040 		}
3041 
3042 		if (mask & AT_GID) {
3043 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3044 			    cr, ZFS_GROUP, &fuidp);
3045 			if (new_gid != zp->z_gid &&
3046 			    zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3047 				if (attrzp)
3048 					VN_RELE(ZTOV(attrzp));
3049 				err = SET_ERROR(EDQUOT);
3050 				goto out2;
3051 			}
3052 		}
3053 	}
3054 	tx = dmu_tx_create(zfsvfs->z_os);
3055 
3056 	if (mask & AT_MODE) {
3057 		uint64_t pmode = zp->z_mode;
3058 		uint64_t acl_obj;
3059 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3060 
3061 		if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3062 		    !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3063 			err = SET_ERROR(EPERM);
3064 			goto out;
3065 		}
3066 
3067 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3068 			goto out;
3069 
3070 		mutex_enter(&zp->z_lock);
3071 		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3072 			/*
3073 			 * Are we upgrading ACL from old V0 format
3074 			 * to V1 format?
3075 			 */
3076 			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3077 			    zfs_znode_acl_version(zp) ==
3078 			    ZFS_ACL_VERSION_INITIAL) {
3079 				dmu_tx_hold_free(tx, acl_obj, 0,
3080 				    DMU_OBJECT_END);
3081 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3082 				    0, aclp->z_acl_bytes);
3083 			} else {
3084 				dmu_tx_hold_write(tx, acl_obj, 0,
3085 				    aclp->z_acl_bytes);
3086 			}
3087 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3088 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3089 			    0, aclp->z_acl_bytes);
3090 		}
3091 		mutex_exit(&zp->z_lock);
3092 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3093 	} else {
3094 		if ((mask & AT_XVATTR) &&
3095 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3096 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3097 		else
3098 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3099 	}
3100 
3101 	if (attrzp) {
3102 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3103 	}
3104 
3105 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3106 	if (fuid_dirtied)
3107 		zfs_fuid_txhold(zfsvfs, tx);
3108 
3109 	zfs_sa_upgrade_txholds(tx, zp);
3110 
3111 	err = dmu_tx_assign(tx, TXG_WAIT);
3112 	if (err)
3113 		goto out;
3114 
3115 	count = 0;
3116 	/*
3117 	 * Set each attribute requested.
3118 	 * We group settings according to the locks they need to acquire.
3119 	 *
3120 	 * Note: you cannot set ctime directly, although it will be
3121 	 * updated as a side-effect of calling this function.
3122 	 */
3123 
3124 
3125 	if (mask & (AT_UID|AT_GID|AT_MODE))
3126 		mutex_enter(&zp->z_acl_lock);
3127 	mutex_enter(&zp->z_lock);
3128 
3129 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3130 	    &zp->z_pflags, sizeof (zp->z_pflags));
3131 
3132 	if (attrzp) {
3133 		if (mask & (AT_UID|AT_GID|AT_MODE))
3134 			mutex_enter(&attrzp->z_acl_lock);
3135 		mutex_enter(&attrzp->z_lock);
3136 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3137 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3138 		    sizeof (attrzp->z_pflags));
3139 	}
3140 
3141 	if (mask & (AT_UID|AT_GID)) {
3142 
3143 		if (mask & AT_UID) {
3144 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3145 			    &new_uid, sizeof (new_uid));
3146 			zp->z_uid = new_uid;
3147 			if (attrzp) {
3148 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3149 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3150 				    sizeof (new_uid));
3151 				attrzp->z_uid = new_uid;
3152 			}
3153 		}
3154 
3155 		if (mask & AT_GID) {
3156 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3157 			    NULL, &new_gid, sizeof (new_gid));
3158 			zp->z_gid = new_gid;
3159 			if (attrzp) {
3160 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3161 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3162 				    sizeof (new_gid));
3163 				attrzp->z_gid = new_gid;
3164 			}
3165 		}
3166 		if (!(mask & AT_MODE)) {
3167 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3168 			    NULL, &new_mode, sizeof (new_mode));
3169 			new_mode = zp->z_mode;
3170 		}
3171 		err = zfs_acl_chown_setattr(zp);
3172 		ASSERT(err == 0);
3173 		if (attrzp) {
3174 			err = zfs_acl_chown_setattr(attrzp);
3175 			ASSERT(err == 0);
3176 		}
3177 	}
3178 
3179 	if (mask & AT_MODE) {
3180 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3181 		    &new_mode, sizeof (new_mode));
3182 		zp->z_mode = new_mode;
3183 		ASSERT3U((uintptr_t)aclp, !=, NULL);
3184 		err = zfs_aclset_common(zp, aclp, cr, tx);
3185 		ASSERT0(err);
3186 		if (zp->z_acl_cached)
3187 			zfs_acl_free(zp->z_acl_cached);
3188 		zp->z_acl_cached = aclp;
3189 		aclp = NULL;
3190 	}
3191 
3192 
3193 	if (mask & AT_ATIME) {
3194 		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3195 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3196 		    &zp->z_atime, sizeof (zp->z_atime));
3197 	}
3198 
3199 	if (mask & AT_MTIME) {
3200 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3201 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3202 		    mtime, sizeof (mtime));
3203 	}
3204 
3205 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3206 	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3207 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3208 		    NULL, mtime, sizeof (mtime));
3209 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3210 		    &ctime, sizeof (ctime));
3211 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3212 		    B_TRUE);
3213 	} else if (mask != 0) {
3214 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3215 		    &ctime, sizeof (ctime));
3216 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3217 		    B_TRUE);
3218 		if (attrzp) {
3219 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3220 			    SA_ZPL_CTIME(zfsvfs), NULL,
3221 			    &ctime, sizeof (ctime));
3222 			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3223 			    mtime, ctime, B_TRUE);
3224 		}
3225 	}
3226 	/*
3227 	 * Do this after setting timestamps to prevent timestamp
3228 	 * update from toggling bit
3229 	 */
3230 
3231 	if (xoap && (mask & AT_XVATTR)) {
3232 
3233 		/*
3234 		 * restore trimmed off masks
3235 		 * so that return masks can be set for caller.
3236 		 */
3237 
3238 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3239 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
3240 		}
3241 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3242 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
3243 		}
3244 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3245 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3246 		}
3247 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3248 			XVA_SET_REQ(xvap, XAT_NODUMP);
3249 		}
3250 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3251 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3252 		}
3253 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3254 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3255 		}
3256 
3257 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3258 			ASSERT(vp->v_type == VREG);
3259 
3260 		zfs_xvattr_set(zp, xvap, tx);
3261 	}
3262 
3263 	if (fuid_dirtied)
3264 		zfs_fuid_sync(zfsvfs, tx);
3265 
3266 	if (mask != 0)
3267 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3268 
3269 	mutex_exit(&zp->z_lock);
3270 	if (mask & (AT_UID|AT_GID|AT_MODE))
3271 		mutex_exit(&zp->z_acl_lock);
3272 
3273 	if (attrzp) {
3274 		if (mask & (AT_UID|AT_GID|AT_MODE))
3275 			mutex_exit(&attrzp->z_acl_lock);
3276 		mutex_exit(&attrzp->z_lock);
3277 	}
3278 out:
3279 	if (err == 0 && attrzp) {
3280 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3281 		    xattr_count, tx);
3282 		ASSERT(err2 == 0);
3283 	}
3284 
3285 	if (attrzp)
3286 		VN_RELE(ZTOV(attrzp));
3287 
3288 	if (aclp)
3289 		zfs_acl_free(aclp);
3290 
3291 	if (fuidp) {
3292 		zfs_fuid_info_free(fuidp);
3293 		fuidp = NULL;
3294 	}
3295 
3296 	if (err) {
3297 		dmu_tx_abort(tx);
3298 		if (err == ERESTART)
3299 			goto top;
3300 	} else {
3301 		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3302 		rw_enter(&rz_zev_rwlock, RW_READER);
3303 		if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_znode_setattr)
3304 			rz_zev_callbacks->rz_zev_znode_setattr(zp, tx);
3305 		rw_exit(&rz_zev_rwlock);
3306 		dmu_tx_commit(tx);
3307 	}
3308 
3309 out2:
3310 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3311 		zil_commit(zilog, 0);
3312 
3313 	ZFS_EXIT(zfsvfs);
3314 	return (err);
3315 }
3316 
3317 typedef struct zfs_zlock {
3318 	krwlock_t	*zl_rwlock;	/* lock we acquired */
3319 	znode_t		*zl_znode;	/* znode we held */
3320 	struct zfs_zlock *zl_next;	/* next in list */
3321 } zfs_zlock_t;
3322 
3323 /*
3324  * Drop locks and release vnodes that were held by zfs_rename_lock().
3325  */
3326 static void
3327 zfs_rename_unlock(zfs_zlock_t **zlpp)
3328 {
3329 	zfs_zlock_t *zl;
3330 
3331 	while ((zl = *zlpp) != NULL) {
3332 		if (zl->zl_znode != NULL)
3333 			VN_RELE(ZTOV(zl->zl_znode));
3334 		rw_exit(zl->zl_rwlock);
3335 		*zlpp = zl->zl_next;
3336 		kmem_free(zl, sizeof (*zl));
3337 	}
3338 }
3339 
3340 /*
3341  * Search back through the directory tree, using the ".." entries.
3342  * Lock each directory in the chain to prevent concurrent renames.
3343  * Fail any attempt to move a directory into one of its own descendants.
3344  * XXX - z_parent_lock can overlap with map or grow locks
3345  */
3346 static int
3347 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3348 {
3349 	zfs_zlock_t	*zl;
3350 	znode_t		*zp = tdzp;
3351 	uint64_t	rootid = zp->z_zfsvfs->z_root;
3352 	uint64_t	oidp = zp->z_id;
3353 	krwlock_t	*rwlp = &szp->z_parent_lock;
3354 	krw_t		rw = RW_WRITER;
3355 
3356 	/*
3357 	 * First pass write-locks szp and compares to zp->z_id.
3358 	 * Later passes read-lock zp and compare to zp->z_parent.
3359 	 */
3360 	do {
3361 		if (!rw_tryenter(rwlp, rw)) {
3362 			/*
3363 			 * Another thread is renaming in this path.
3364 			 * Note that if we are a WRITER, we don't have any
3365 			 * parent_locks held yet.
3366 			 */
3367 			if (rw == RW_READER && zp->z_id > szp->z_id) {
3368 				/*
3369 				 * Drop our locks and restart
3370 				 */
3371 				zfs_rename_unlock(&zl);
3372 				*zlpp = NULL;
3373 				zp = tdzp;
3374 				oidp = zp->z_id;
3375 				rwlp = &szp->z_parent_lock;
3376 				rw = RW_WRITER;
3377 				continue;
3378 			} else {
3379 				/*
3380 				 * Wait for other thread to drop its locks
3381 				 */
3382 				rw_enter(rwlp, rw);
3383 			}
3384 		}
3385 
3386 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3387 		zl->zl_rwlock = rwlp;
3388 		zl->zl_znode = NULL;
3389 		zl->zl_next = *zlpp;
3390 		*zlpp = zl;
3391 
3392 		if (oidp == szp->z_id)		/* We're a descendant of szp */
3393 			return (SET_ERROR(EINVAL));
3394 
3395 		if (oidp == rootid)		/* We've hit the top */
3396 			return (0);
3397 
3398 		if (rw == RW_READER) {		/* i.e. not the first pass */
3399 			int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3400 			if (error)
3401 				return (error);
3402 			zl->zl_znode = zp;
3403 		}
3404 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3405 		    &oidp, sizeof (oidp));
3406 		rwlp = &zp->z_parent_lock;
3407 		rw = RW_READER;
3408 
3409 	} while (zp->z_id != sdzp->z_id);
3410 
3411 	return (0);
3412 }
3413 
3414 /*
3415  * Move an entry from the provided source directory to the target
3416  * directory.  Change the entry name as indicated.
3417  *
3418  *	IN:	sdvp	- Source directory containing the "old entry".
3419  *		snm	- Old entry name.
3420  *		tdvp	- Target directory to contain the "new entry".
3421  *		tnm	- New entry name.
3422  *		cr	- credentials of caller.
3423  *		ct	- caller context
3424  *		flags	- case flags
3425  *
3426  *	RETURN:	0 on success, error code on failure.
3427  *
3428  * Timestamps:
3429  *	sdvp,tdvp - ctime|mtime updated
3430  */
3431 /*ARGSUSED*/
3432 static int
3433 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3434     caller_context_t *ct, int flags)
3435 {
3436 	znode_t		*tdzp, *szp, *tzp;
3437 	znode_t		*sdzp = VTOZ(sdvp);
3438 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
3439 	zilog_t		*zilog;
3440 	vnode_t		*realvp;
3441 	zfs_dirlock_t	*sdl, *tdl;
3442 	dmu_tx_t	*tx;
3443 	zfs_zlock_t	*zl;
3444 	int		cmp, serr, terr;
3445 	int		error = 0;
3446 	int		zflg = 0;
3447 	boolean_t	waited = B_FALSE;
3448 
3449 	ZFS_ENTER(zfsvfs);
3450 	ZFS_VERIFY_ZP(sdzp);
3451 	zilog = zfsvfs->z_log;
3452 
3453 	/*
3454 	 * Make sure we have the real vp for the target directory.
3455 	 */
3456 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3457 		tdvp = realvp;
3458 
3459 	tdzp = VTOZ(tdvp);
3460 	ZFS_VERIFY_ZP(tdzp);
3461 
3462 	/*
3463 	 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3464 	 * ctldir appear to have the same v_vfsp.
3465 	 */
3466 	if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3467 		ZFS_EXIT(zfsvfs);
3468 		return (SET_ERROR(EXDEV));
3469 	}
3470 
3471 	if (zfsvfs->z_utf8 && u8_validate(tnm,
3472 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3473 		ZFS_EXIT(zfsvfs);
3474 		return (SET_ERROR(EILSEQ));
3475 	}
3476 
3477 	if (flags & FIGNORECASE)
3478 		zflg |= ZCILOOK;
3479 
3480 top:
3481 	szp = NULL;
3482 	tzp = NULL;
3483 	zl = NULL;
3484 
3485 	/*
3486 	 * This is to prevent the creation of links into attribute space
3487 	 * by renaming a linked file into/outof an attribute directory.
3488 	 * See the comment in zfs_link() for why this is considered bad.
3489 	 */
3490 	if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3491 		ZFS_EXIT(zfsvfs);
3492 		return (SET_ERROR(EINVAL));
3493 	}
3494 
3495 	/*
3496 	 * Lock source and target directory entries.  To prevent deadlock,
3497 	 * a lock ordering must be defined.  We lock the directory with
3498 	 * the smallest object id first, or if it's a tie, the one with
3499 	 * the lexically first name.
3500 	 */
3501 	if (sdzp->z_id < tdzp->z_id) {
3502 		cmp = -1;
3503 	} else if (sdzp->z_id > tdzp->z_id) {
3504 		cmp = 1;
3505 	} else {
3506 		/*
3507 		 * First compare the two name arguments without
3508 		 * considering any case folding.
3509 		 */
3510 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3511 
3512 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3513 		ASSERT(error == 0 || !zfsvfs->z_utf8);
3514 		if (cmp == 0) {
3515 			/*
3516 			 * POSIX: "If the old argument and the new argument
3517 			 * both refer to links to the same existing file,
3518 			 * the rename() function shall return successfully
3519 			 * and perform no other action."
3520 			 */
3521 			ZFS_EXIT(zfsvfs);
3522 			return (0);
3523 		}
3524 		/*
3525 		 * If the file system is case-folding, then we may
3526 		 * have some more checking to do.  A case-folding file
3527 		 * system is either supporting mixed case sensitivity
3528 		 * access or is completely case-insensitive.  Note
3529 		 * that the file system is always case preserving.
3530 		 *
3531 		 * In mixed sensitivity mode case sensitive behavior
3532 		 * is the default.  FIGNORECASE must be used to
3533 		 * explicitly request case insensitive behavior.
3534 		 *
3535 		 * If the source and target names provided differ only
3536 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
3537 		 * we will treat this as a special case in the
3538 		 * case-insensitive mode: as long as the source name
3539 		 * is an exact match, we will allow this to proceed as
3540 		 * a name-change request.
3541 		 */
3542 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3543 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
3544 		    flags & FIGNORECASE)) &&
3545 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3546 		    &error) == 0) {
3547 			/*
3548 			 * case preserving rename request, require exact
3549 			 * name matches
3550 			 */
3551 			zflg |= ZCIEXACT;
3552 			zflg &= ~ZCILOOK;
3553 		}
3554 	}
3555 
3556 	/*
3557 	 * If the source and destination directories are the same, we should
3558 	 * grab the z_name_lock of that directory only once.
3559 	 */
3560 	if (sdzp == tdzp) {
3561 		zflg |= ZHAVELOCK;
3562 		rw_enter(&sdzp->z_name_lock, RW_READER);
3563 	}
3564 
3565 	if (cmp < 0) {
3566 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3567 		    ZEXISTS | zflg, NULL, NULL);
3568 		terr = zfs_dirent_lock(&tdl,
3569 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3570 	} else {
3571 		terr = zfs_dirent_lock(&tdl,
3572 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3573 		serr = zfs_dirent_lock(&sdl,
3574 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3575 		    NULL, NULL);
3576 	}
3577 
3578 	if (serr) {
3579 		/*
3580 		 * Source entry invalid or not there.
3581 		 */
3582 		if (!terr) {
3583 			zfs_dirent_unlock(tdl);
3584 			if (tzp)
3585 				VN_RELE(ZTOV(tzp));
3586 		}
3587 
3588 		if (sdzp == tdzp)
3589 			rw_exit(&sdzp->z_name_lock);
3590 
3591 		if (strcmp(snm, "..") == 0)
3592 			serr = SET_ERROR(EINVAL);
3593 		ZFS_EXIT(zfsvfs);
3594 		return (serr);
3595 	}
3596 	if (terr) {
3597 		zfs_dirent_unlock(sdl);
3598 		VN_RELE(ZTOV(szp));
3599 
3600 		if (sdzp == tdzp)
3601 			rw_exit(&sdzp->z_name_lock);
3602 
3603 		if (strcmp(tnm, "..") == 0)
3604 			terr = SET_ERROR(EINVAL);
3605 		ZFS_EXIT(zfsvfs);
3606 		return (terr);
3607 	}
3608 
3609 	/*
3610 	 * Must have write access at the source to remove the old entry
3611 	 * and write access at the target to create the new entry.
3612 	 * Note that if target and source are the same, this can be
3613 	 * done in a single check.
3614 	 */
3615 
3616 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3617 		goto out;
3618 
3619 	if (ZTOV(szp)->v_type == VDIR) {
3620 		/*
3621 		 * Check to make sure rename is valid.
3622 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3623 		 */
3624 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3625 			goto out;
3626 	}
3627 
3628 	/*
3629 	 * Does target exist?
3630 	 */
3631 	if (tzp) {
3632 		/*
3633 		 * Source and target must be the same type.
3634 		 */
3635 		if (ZTOV(szp)->v_type == VDIR) {
3636 			if (ZTOV(tzp)->v_type != VDIR) {
3637 				error = SET_ERROR(ENOTDIR);
3638 				goto out;
3639 			}
3640 		} else {
3641 			if (ZTOV(tzp)->v_type == VDIR) {
3642 				error = SET_ERROR(EISDIR);
3643 				goto out;
3644 			}
3645 		}
3646 		/*
3647 		 * POSIX dictates that when the source and target
3648 		 * entries refer to the same file object, rename
3649 		 * must do nothing and exit without error.
3650 		 */
3651 		if (szp->z_id == tzp->z_id) {
3652 			error = 0;
3653 			goto out;
3654 		}
3655 	}
3656 
3657 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3658 	if (tzp)
3659 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3660 
3661 	/*
3662 	 * notify the target directory if it is not the same
3663 	 * as source directory.
3664 	 */
3665 	if (tdvp != sdvp) {
3666 		vnevent_rename_dest_dir(tdvp, ct);
3667 	}
3668 
3669 	tx = dmu_tx_create(zfsvfs->z_os);
3670 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3671 	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3672 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3673 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3674 	if (sdzp != tdzp) {
3675 		dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3676 		zfs_sa_upgrade_txholds(tx, tdzp);
3677 	}
3678 	if (tzp) {
3679 		dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3680 		zfs_sa_upgrade_txholds(tx, tzp);
3681 	}
3682 
3683 	zfs_sa_upgrade_txholds(tx, szp);
3684 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3685 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3686 	if (error) {
3687 		if (zl != NULL)
3688 			zfs_rename_unlock(&zl);
3689 		zfs_dirent_unlock(sdl);
3690 		zfs_dirent_unlock(tdl);
3691 
3692 		if (sdzp == tdzp)
3693 			rw_exit(&sdzp->z_name_lock);
3694 
3695 		VN_RELE(ZTOV(szp));
3696 		if (tzp)
3697 			VN_RELE(ZTOV(tzp));
3698 		if (error == ERESTART) {
3699 			waited = B_TRUE;
3700 			dmu_tx_wait(tx);
3701 			dmu_tx_abort(tx);
3702 			goto top;
3703 		}
3704 		dmu_tx_abort(tx);
3705 		ZFS_EXIT(zfsvfs);
3706 		return (error);
3707 	}
3708 
3709 	if (tzp)	/* Attempt to remove the existing target */
3710 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3711 
3712 	if (error == 0) {
3713 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3714 		if (error == 0) {
3715 			szp->z_pflags |= ZFS_AV_MODIFIED;
3716 
3717 			error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3718 			    (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3719 			ASSERT0(error);
3720 
3721 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3722 			if (error == 0) {
3723 				zfs_log_rename(zilog, tx, TX_RENAME |
3724 				    (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3725 				    sdl->dl_name, tdzp, tdl->dl_name, szp, tzp);
3726 
3727 				/*
3728 				 * Update path information for the target vnode
3729 				 */
3730 				vn_renamepath(tdvp, ZTOV(szp), tnm,
3731 				    strlen(tnm));
3732 			} else {
3733 				/*
3734 				 * At this point, we have successfully created
3735 				 * the target name, but have failed to remove
3736 				 * the source name.  Since the create was done
3737 				 * with the ZRENAMING flag, there are
3738 				 * complications; for one, the link count is
3739 				 * wrong.  The easiest way to deal with this
3740 				 * is to remove the newly created target, and
3741 				 * return the original error.  This must
3742 				 * succeed; fortunately, it is very unlikely to
3743 				 * fail, since we just created it.
3744 				 */
3745 				VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3746 				    ZRENAMING, NULL), ==, 0);
3747 			}
3748 		}
3749 	}
3750 
3751 	dmu_tx_commit(tx);
3752 out:
3753 	if (zl != NULL)
3754 		zfs_rename_unlock(&zl);
3755 
3756 	zfs_dirent_unlock(sdl);
3757 	zfs_dirent_unlock(tdl);
3758 
3759 	if (sdzp == tdzp)
3760 		rw_exit(&sdzp->z_name_lock);
3761 
3762 
3763 	VN_RELE(ZTOV(szp));
3764 	if (tzp)
3765 		VN_RELE(ZTOV(tzp));
3766 
3767 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3768 		zil_commit(zilog, 0);
3769 
3770 	ZFS_EXIT(zfsvfs);
3771 	return (error);
3772 }
3773 
3774 /*
3775  * Insert the indicated symbolic reference entry into the directory.
3776  *
3777  *	IN:	dvp	- Directory to contain new symbolic link.
3778  *		link	- Name for new symlink entry.
3779  *		vap	- Attributes of new entry.
3780  *		cr	- credentials of caller.
3781  *		ct	- caller context
3782  *		flags	- case flags
3783  *
3784  *	RETURN:	0 on success, error code on failure.
3785  *
3786  * Timestamps:
3787  *	dvp - ctime|mtime updated
3788  */
3789 /*ARGSUSED*/
3790 static int
3791 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3792     caller_context_t *ct, int flags)
3793 {
3794 	znode_t		*zp, *dzp = VTOZ(dvp);
3795 	zfs_dirlock_t	*dl;
3796 	dmu_tx_t	*tx;
3797 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3798 	zilog_t		*zilog;
3799 	uint64_t	len = strlen(link);
3800 	int		error;
3801 	int		zflg = ZNEW;
3802 	zfs_acl_ids_t	acl_ids;
3803 	boolean_t	fuid_dirtied;
3804 	uint64_t	txtype = TX_SYMLINK;
3805 	boolean_t	waited = B_FALSE;
3806 
3807 	ASSERT(vap->va_type == VLNK);
3808 
3809 	ZFS_ENTER(zfsvfs);
3810 	ZFS_VERIFY_ZP(dzp);
3811 	zilog = zfsvfs->z_log;
3812 
3813 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3814 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3815 		ZFS_EXIT(zfsvfs);
3816 		return (SET_ERROR(EILSEQ));
3817 	}
3818 	if (flags & FIGNORECASE)
3819 		zflg |= ZCILOOK;
3820 
3821 	if (len > MAXPATHLEN) {
3822 		ZFS_EXIT(zfsvfs);
3823 		return (SET_ERROR(ENAMETOOLONG));
3824 	}
3825 
3826 	if ((error = zfs_acl_ids_create(dzp, 0,
3827 	    vap, cr, NULL, &acl_ids)) != 0) {
3828 		ZFS_EXIT(zfsvfs);
3829 		return (error);
3830 	}
3831 top:
3832 	/*
3833 	 * Attempt to lock directory; fail if entry already exists.
3834 	 */
3835 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3836 	if (error) {
3837 		zfs_acl_ids_free(&acl_ids);
3838 		ZFS_EXIT(zfsvfs);
3839 		return (error);
3840 	}
3841 
3842 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3843 		zfs_acl_ids_free(&acl_ids);
3844 		zfs_dirent_unlock(dl);
3845 		ZFS_EXIT(zfsvfs);
3846 		return (error);
3847 	}
3848 
3849 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3850 		zfs_acl_ids_free(&acl_ids);
3851 		zfs_dirent_unlock(dl);
3852 		ZFS_EXIT(zfsvfs);
3853 		return (SET_ERROR(EDQUOT));
3854 	}
3855 	tx = dmu_tx_create(zfsvfs->z_os);
3856 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3857 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3858 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3859 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3860 	    ZFS_SA_BASE_ATTR_SIZE + len);
3861 	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3862 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3863 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3864 		    acl_ids.z_aclp->z_acl_bytes);
3865 	}
3866 	if (fuid_dirtied)
3867 		zfs_fuid_txhold(zfsvfs, tx);
3868 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3869 	if (error) {
3870 		zfs_dirent_unlock(dl);
3871 		if (error == ERESTART) {
3872 			waited = B_TRUE;
3873 			dmu_tx_wait(tx);
3874 			dmu_tx_abort(tx);
3875 			goto top;
3876 		}
3877 		zfs_acl_ids_free(&acl_ids);
3878 		dmu_tx_abort(tx);
3879 		ZFS_EXIT(zfsvfs);
3880 		return (error);
3881 	}
3882 
3883 	/*
3884 	 * Create a new object for the symlink.
3885 	 * for version 4 ZPL datsets the symlink will be an SA attribute
3886 	 */
3887 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3888 
3889 	if (fuid_dirtied)
3890 		zfs_fuid_sync(zfsvfs, tx);
3891 
3892 	mutex_enter(&zp->z_lock);
3893 	if (zp->z_is_sa)
3894 		error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3895 		    link, len, tx);
3896 	else
3897 		zfs_sa_symlink(zp, link, len, tx);
3898 	mutex_exit(&zp->z_lock);
3899 
3900 	zp->z_size = len;
3901 	(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3902 	    &zp->z_size, sizeof (zp->z_size), tx);
3903 	/*
3904 	 * Insert the new object into the directory.
3905 	 */
3906 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3907 
3908 	if (flags & FIGNORECASE)
3909 		txtype |= TX_CI;
3910 	zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3911 
3912 	zfs_acl_ids_free(&acl_ids);
3913 
3914 	dmu_tx_commit(tx);
3915 
3916 	zfs_dirent_unlock(dl);
3917 
3918 	VN_RELE(ZTOV(zp));
3919 
3920 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3921 		zil_commit(zilog, 0);
3922 
3923 	ZFS_EXIT(zfsvfs);
3924 	return (error);
3925 }
3926 
3927 /*
3928  * Return, in the buffer contained in the provided uio structure,
3929  * the symbolic path referred to by vp.
3930  *
3931  *	IN:	vp	- vnode of symbolic link.
3932  *		uio	- structure to contain the link path.
3933  *		cr	- credentials of caller.
3934  *		ct	- caller context
3935  *
3936  *	OUT:	uio	- structure containing the link path.
3937  *
3938  *	RETURN:	0 on success, error code on failure.
3939  *
3940  * Timestamps:
3941  *	vp - atime updated
3942  */
3943 /* ARGSUSED */
3944 static int
3945 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3946 {
3947 	znode_t		*zp = VTOZ(vp);
3948 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3949 	int		error;
3950 
3951 	ZFS_ENTER(zfsvfs);
3952 	ZFS_VERIFY_ZP(zp);
3953 
3954 	mutex_enter(&zp->z_lock);
3955 	if (zp->z_is_sa)
3956 		error = sa_lookup_uio(zp->z_sa_hdl,
3957 		    SA_ZPL_SYMLINK(zfsvfs), uio);
3958 	else
3959 		error = zfs_sa_readlink(zp, uio);
3960 	mutex_exit(&zp->z_lock);
3961 
3962 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3963 
3964 	ZFS_EXIT(zfsvfs);
3965 	return (error);
3966 }
3967 
3968 /*
3969  * Insert a new entry into directory tdvp referencing svp.
3970  *
3971  *	IN:	tdvp	- Directory to contain new entry.
3972  *		svp	- vnode of new entry.
3973  *		name	- name of new entry.
3974  *		cr	- credentials of caller.
3975  *		ct	- caller context
3976  *
3977  *	RETURN:	0 on success, error code on failure.
3978  *
3979  * Timestamps:
3980  *	tdvp - ctime|mtime updated
3981  *	 svp - ctime updated
3982  */
3983 /* ARGSUSED */
3984 static int
3985 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3986     caller_context_t *ct, int flags)
3987 {
3988 	znode_t		*dzp = VTOZ(tdvp);
3989 	znode_t		*tzp, *szp;
3990 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3991 	zilog_t		*zilog;
3992 	zfs_dirlock_t	*dl;
3993 	dmu_tx_t	*tx;
3994 	vnode_t		*realvp;
3995 	int		error;
3996 	int		zf = ZNEW;
3997 	uint64_t	parent;
3998 	uid_t		owner;
3999 	boolean_t	waited = B_FALSE;
4000 
4001 	ASSERT(tdvp->v_type == VDIR);
4002 
4003 	ZFS_ENTER(zfsvfs);
4004 	ZFS_VERIFY_ZP(dzp);
4005 	zilog = zfsvfs->z_log;
4006 
4007 	if (VOP_REALVP(svp, &realvp, ct) == 0)
4008 		svp = realvp;
4009 
4010 	/*
4011 	 * POSIX dictates that we return EPERM here.
4012 	 * Better choices include ENOTSUP or EISDIR.
4013 	 */
4014 	if (svp->v_type == VDIR) {
4015 		ZFS_EXIT(zfsvfs);
4016 		return (SET_ERROR(EPERM));
4017 	}
4018 
4019 	szp = VTOZ(svp);
4020 	ZFS_VERIFY_ZP(szp);
4021 
4022 	/*
4023 	 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4024 	 * ctldir appear to have the same v_vfsp.
4025 	 */
4026 	if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4027 		ZFS_EXIT(zfsvfs);
4028 		return (SET_ERROR(EXDEV));
4029 	}
4030 
4031 	/* Prevent links to .zfs/shares files */
4032 
4033 	if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4034 	    &parent, sizeof (uint64_t))) != 0) {
4035 		ZFS_EXIT(zfsvfs);
4036 		return (error);
4037 	}
4038 	if (parent == zfsvfs->z_shares_dir) {
4039 		ZFS_EXIT(zfsvfs);
4040 		return (SET_ERROR(EPERM));
4041 	}
4042 
4043 	if (zfsvfs->z_utf8 && u8_validate(name,
4044 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4045 		ZFS_EXIT(zfsvfs);
4046 		return (SET_ERROR(EILSEQ));
4047 	}
4048 	if (flags & FIGNORECASE)
4049 		zf |= ZCILOOK;
4050 
4051 	/*
4052 	 * We do not support links between attributes and non-attributes
4053 	 * because of the potential security risk of creating links
4054 	 * into "normal" file space in order to circumvent restrictions
4055 	 * imposed in attribute space.
4056 	 */
4057 	if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4058 		ZFS_EXIT(zfsvfs);
4059 		return (SET_ERROR(EINVAL));
4060 	}
4061 
4062 
4063 	owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4064 	if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4065 		ZFS_EXIT(zfsvfs);
4066 		return (SET_ERROR(EPERM));
4067 	}
4068 
4069 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4070 		ZFS_EXIT(zfsvfs);
4071 		return (error);
4072 	}
4073 
4074 top:
4075 	/*
4076 	 * Attempt to lock directory; fail if entry already exists.
4077 	 */
4078 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4079 	if (error) {
4080 		ZFS_EXIT(zfsvfs);
4081 		return (error);
4082 	}
4083 
4084 	tx = dmu_tx_create(zfsvfs->z_os);
4085 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4086 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4087 	zfs_sa_upgrade_txholds(tx, szp);
4088 	zfs_sa_upgrade_txholds(tx, dzp);
4089 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4090 	if (error) {
4091 		zfs_dirent_unlock(dl);
4092 		if (error == ERESTART) {
4093 			waited = B_TRUE;
4094 			dmu_tx_wait(tx);
4095 			dmu_tx_abort(tx);
4096 			goto top;
4097 		}
4098 		dmu_tx_abort(tx);
4099 		ZFS_EXIT(zfsvfs);
4100 		return (error);
4101 	}
4102 
4103 	error = zfs_link_create(dl, szp, tx, 0);
4104 
4105 	if (error == 0) {
4106 		uint64_t txtype = TX_LINK;
4107 		if (flags & FIGNORECASE)
4108 			txtype |= TX_CI;
4109 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4110 	}
4111 
4112 	dmu_tx_commit(tx);
4113 
4114 	zfs_dirent_unlock(dl);
4115 
4116 	if (error == 0) {
4117 		vnevent_link(svp, ct);
4118 	}
4119 
4120 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4121 		zil_commit(zilog, 0);
4122 
4123 	ZFS_EXIT(zfsvfs);
4124 	return (error);
4125 }
4126 
4127 /*
4128  * zfs_null_putapage() is used when the file system has been force
4129  * unmounted. It just drops the pages.
4130  */
4131 /* ARGSUSED */
4132 static int
4133 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4134 		size_t *lenp, int flags, cred_t *cr)
4135 {
4136 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4137 	return (0);
4138 }
4139 
4140 /*
4141  * Push a page out to disk, klustering if possible.
4142  *
4143  *	IN:	vp	- file to push page to.
4144  *		pp	- page to push.
4145  *		flags	- additional flags.
4146  *		cr	- credentials of caller.
4147  *
4148  *	OUT:	offp	- start of range pushed.
4149  *		lenp	- len of range pushed.
4150  *
4151  *	RETURN:	0 on success, error code on failure.
4152  *
4153  * NOTE: callers must have locked the page to be pushed.  On
4154  * exit, the page (and all other pages in the kluster) must be
4155  * unlocked.
4156  */
4157 /* ARGSUSED */
4158 static int
4159 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4160 		size_t *lenp, int flags, cred_t *cr)
4161 {
4162 	znode_t		*zp = VTOZ(vp);
4163 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4164 	dmu_tx_t	*tx;
4165 	u_offset_t	off, koff;
4166 	size_t		len, klen;
4167 	int		err;
4168 
4169 	off = pp->p_offset;
4170 	len = PAGESIZE;
4171 	/*
4172 	 * If our blocksize is bigger than the page size, try to kluster
4173 	 * multiple pages so that we write a full block (thus avoiding
4174 	 * a read-modify-write).
4175 	 */
4176 	if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4177 		klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4178 		koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4179 		ASSERT(koff <= zp->z_size);
4180 		if (koff + klen > zp->z_size)
4181 			klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4182 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4183 	}
4184 	ASSERT3U(btop(len), ==, btopr(len));
4185 
4186 	/*
4187 	 * Can't push pages past end-of-file.
4188 	 */
4189 	if (off >= zp->z_size) {
4190 		/* ignore all pages */
4191 		err = 0;
4192 		goto out;
4193 	} else if (off + len > zp->z_size) {
4194 		int npages = btopr(zp->z_size - off);
4195 		page_t *trunc;
4196 
4197 		page_list_break(&pp, &trunc, npages);
4198 		/* ignore pages past end of file */
4199 		if (trunc)
4200 			pvn_write_done(trunc, flags);
4201 		len = zp->z_size - off;
4202 	}
4203 
4204 	if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4205 	    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4206 		err = SET_ERROR(EDQUOT);
4207 		goto out;
4208 	}
4209 	tx = dmu_tx_create(zfsvfs->z_os);
4210 	dmu_tx_hold_write(tx, zp->z_id, off, len);
4211 
4212 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4213 	zfs_sa_upgrade_txholds(tx, zp);
4214 	err = dmu_tx_assign(tx, TXG_WAIT);
4215 	if (err != 0) {
4216 		dmu_tx_abort(tx);
4217 		goto out;
4218 	}
4219 
4220 	if (zp->z_blksz <= PAGESIZE) {
4221 		caddr_t va = zfs_map_page(pp, S_READ);
4222 		ASSERT3U(len, <=, PAGESIZE);
4223 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4224 		zfs_unmap_page(pp, va);
4225 	} else {
4226 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4227 	}
4228 
4229 	if (err == 0) {
4230 		uint64_t mtime[2], ctime[2];
4231 		sa_bulk_attr_t bulk[3];
4232 		int count = 0;
4233 
4234 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4235 		    &mtime, 16);
4236 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4237 		    &ctime, 16);
4238 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4239 		    &zp->z_pflags, 8);
4240 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4241 		    B_TRUE);
4242 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4243 	}
4244 	dmu_tx_commit(tx);
4245 
4246 out:
4247 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4248 	if (offp)
4249 		*offp = off;
4250 	if (lenp)
4251 		*lenp = len;
4252 
4253 	return (err);
4254 }
4255 
4256 /*
4257  * Copy the portion of the file indicated from pages into the file.
4258  * The pages are stored in a page list attached to the files vnode.
4259  *
4260  *	IN:	vp	- vnode of file to push page data to.
4261  *		off	- position in file to put data.
4262  *		len	- amount of data to write.
4263  *		flags	- flags to control the operation.
4264  *		cr	- credentials of caller.
4265  *		ct	- caller context.
4266  *
4267  *	RETURN:	0 on success, error code on failure.
4268  *
4269  * Timestamps:
4270  *	vp - ctime|mtime updated
4271  */
4272 /*ARGSUSED*/
4273 static int
4274 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4275     caller_context_t *ct)
4276 {
4277 	znode_t		*zp = VTOZ(vp);
4278 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4279 	page_t		*pp;
4280 	size_t		io_len;
4281 	u_offset_t	io_off;
4282 	uint_t		blksz;
4283 	rl_t		*rl;
4284 	int		error = 0;
4285 
4286 	ZFS_ENTER(zfsvfs);
4287 	ZFS_VERIFY_ZP(zp);
4288 
4289 	/*
4290 	 * There's nothing to do if no data is cached.
4291 	 */
4292 	if (!vn_has_cached_data(vp)) {
4293 		ZFS_EXIT(zfsvfs);
4294 		return (0);
4295 	}
4296 
4297 	/*
4298 	 * Align this request to the file block size in case we kluster.
4299 	 * XXX - this can result in pretty aggresive locking, which can
4300 	 * impact simultanious read/write access.  One option might be
4301 	 * to break up long requests (len == 0) into block-by-block
4302 	 * operations to get narrower locking.
4303 	 */
4304 	blksz = zp->z_blksz;
4305 	if (ISP2(blksz))
4306 		io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4307 	else
4308 		io_off = 0;
4309 	if (len > 0 && ISP2(blksz))
4310 		io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4311 	else
4312 		io_len = 0;
4313 
4314 	if (io_len == 0) {
4315 		/*
4316 		 * Search the entire vp list for pages >= io_off.
4317 		 */
4318 		rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4319 		error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4320 		goto out;
4321 	}
4322 	rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4323 
4324 	if (off > zp->z_size) {
4325 		/* past end of file */
4326 		zfs_range_unlock(rl);
4327 		ZFS_EXIT(zfsvfs);
4328 		return (0);
4329 	}
4330 
4331 	len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4332 
4333 	for (off = io_off; io_off < off + len; io_off += io_len) {
4334 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4335 			pp = page_lookup(vp, io_off,
4336 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4337 		} else {
4338 			pp = page_lookup_nowait(vp, io_off,
4339 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4340 		}
4341 
4342 		if (pp != NULL && pvn_getdirty(pp, flags)) {
4343 			int err;
4344 
4345 			/*
4346 			 * Found a dirty page to push
4347 			 */
4348 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4349 			if (err)
4350 				error = err;
4351 		} else {
4352 			io_len = PAGESIZE;
4353 		}
4354 	}
4355 out:
4356 	zfs_range_unlock(rl);
4357 	if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4358 		zil_commit(zfsvfs->z_log, zp->z_id);
4359 	ZFS_EXIT(zfsvfs);
4360 	return (error);
4361 }
4362 
4363 /*ARGSUSED*/
4364 void
4365 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4366 {
4367 	znode_t	*zp = VTOZ(vp);
4368 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4369 	int error;
4370 
4371 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4372 	if (zp->z_sa_hdl == NULL) {
4373 		/*
4374 		 * The fs has been unmounted, or we did a
4375 		 * suspend/resume and this file no longer exists.
4376 		 */
4377 		if (vn_has_cached_data(vp)) {
4378 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4379 			    B_INVAL, cr);
4380 		}
4381 
4382 		mutex_enter(&zp->z_lock);
4383 		mutex_enter(&vp->v_lock);
4384 		ASSERT(vp->v_count == 1);
4385 		vp->v_count = 0;
4386 		mutex_exit(&vp->v_lock);
4387 		mutex_exit(&zp->z_lock);
4388 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
4389 		zfs_znode_free(zp);
4390 		return;
4391 	}
4392 
4393 	/*
4394 	 * Attempt to push any data in the page cache.  If this fails
4395 	 * we will get kicked out later in zfs_zinactive().
4396 	 */
4397 	if (vn_has_cached_data(vp)) {
4398 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4399 		    cr);
4400 	}
4401 
4402 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4403 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4404 
4405 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4406 		zfs_sa_upgrade_txholds(tx, zp);
4407 		error = dmu_tx_assign(tx, TXG_WAIT);
4408 		if (error) {
4409 			dmu_tx_abort(tx);
4410 		} else {
4411 			mutex_enter(&zp->z_lock);
4412 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4413 			    (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4414 			zp->z_atime_dirty = 0;
4415 			mutex_exit(&zp->z_lock);
4416 			dmu_tx_commit(tx);
4417 		}
4418 	}
4419 
4420 	zfs_zinactive(zp);
4421 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
4422 }
4423 
4424 /*
4425  * Bounds-check the seek operation.
4426  *
4427  *	IN:	vp	- vnode seeking within
4428  *		ooff	- old file offset
4429  *		noffp	- pointer to new file offset
4430  *		ct	- caller context
4431  *
4432  *	RETURN:	0 on success, EINVAL if new offset invalid.
4433  */
4434 /* ARGSUSED */
4435 static int
4436 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4437     caller_context_t *ct)
4438 {
4439 	if (vp->v_type == VDIR)
4440 		return (0);
4441 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4442 }
4443 
4444 /*
4445  * Pre-filter the generic locking function to trap attempts to place
4446  * a mandatory lock on a memory mapped file.
4447  */
4448 static int
4449 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4450     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4451 {
4452 	znode_t *zp = VTOZ(vp);
4453 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4454 
4455 	ZFS_ENTER(zfsvfs);
4456 	ZFS_VERIFY_ZP(zp);
4457 
4458 	/*
4459 	 * We are following the UFS semantics with respect to mapcnt
4460 	 * here: If we see that the file is mapped already, then we will
4461 	 * return an error, but we don't worry about races between this
4462 	 * function and zfs_map().
4463 	 */
4464 	if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4465 		ZFS_EXIT(zfsvfs);
4466 		return (SET_ERROR(EAGAIN));
4467 	}
4468 	ZFS_EXIT(zfsvfs);
4469 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4470 }
4471 
4472 /*
4473  * If we can't find a page in the cache, we will create a new page
4474  * and fill it with file data.  For efficiency, we may try to fill
4475  * multiple pages at once (klustering) to fill up the supplied page
4476  * list.  Note that the pages to be filled are held with an exclusive
4477  * lock to prevent access by other threads while they are being filled.
4478  */
4479 static int
4480 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4481     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4482 {
4483 	znode_t *zp = VTOZ(vp);
4484 	page_t *pp, *cur_pp;
4485 	objset_t *os = zp->z_zfsvfs->z_os;
4486 	u_offset_t io_off, total;
4487 	size_t io_len;
4488 	int err;
4489 
4490 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4491 		/*
4492 		 * We only have a single page, don't bother klustering
4493 		 */
4494 		io_off = off;
4495 		io_len = PAGESIZE;
4496 		pp = page_create_va(vp, io_off, io_len,
4497 		    PG_EXCL | PG_WAIT, seg, addr);
4498 	} else {
4499 		/*
4500 		 * Try to find enough pages to fill the page list
4501 		 */
4502 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4503 		    &io_len, off, plsz, 0);
4504 	}
4505 	if (pp == NULL) {
4506 		/*
4507 		 * The page already exists, nothing to do here.
4508 		 */
4509 		*pl = NULL;
4510 		return (0);
4511 	}
4512 
4513 	/*
4514 	 * Fill the pages in the kluster.
4515 	 */
4516 	cur_pp = pp;
4517 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4518 		caddr_t va;
4519 
4520 		ASSERT3U(io_off, ==, cur_pp->p_offset);
4521 		va = zfs_map_page(cur_pp, S_WRITE);
4522 		err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4523 		    DMU_READ_PREFETCH);
4524 		zfs_unmap_page(cur_pp, va);
4525 		if (err) {
4526 			/* On error, toss the entire kluster */
4527 			pvn_read_done(pp, B_ERROR);
4528 			/* convert checksum errors into IO errors */
4529 			if (err == ECKSUM)
4530 				err = SET_ERROR(EIO);
4531 			return (err);
4532 		}
4533 		cur_pp = cur_pp->p_next;
4534 	}
4535 
4536 	/*
4537 	 * Fill in the page list array from the kluster starting
4538 	 * from the desired offset `off'.
4539 	 * NOTE: the page list will always be null terminated.
4540 	 */
4541 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4542 	ASSERT(pl == NULL || (*pl)->p_offset == off);
4543 
4544 	return (0);
4545 }
4546 
4547 /*
4548  * Return pointers to the pages for the file region [off, off + len]
4549  * in the pl array.  If plsz is greater than len, this function may
4550  * also return page pointers from after the specified region
4551  * (i.e. the region [off, off + plsz]).  These additional pages are
4552  * only returned if they are already in the cache, or were created as
4553  * part of a klustered read.
4554  *
4555  *	IN:	vp	- vnode of file to get data from.
4556  *		off	- position in file to get data from.
4557  *		len	- amount of data to retrieve.
4558  *		plsz	- length of provided page list.
4559  *		seg	- segment to obtain pages for.
4560  *		addr	- virtual address of fault.
4561  *		rw	- mode of created pages.
4562  *		cr	- credentials of caller.
4563  *		ct	- caller context.
4564  *
4565  *	OUT:	protp	- protection mode of created pages.
4566  *		pl	- list of pages created.
4567  *
4568  *	RETURN:	0 on success, error code on failure.
4569  *
4570  * Timestamps:
4571  *	vp - atime updated
4572  */
4573 /* ARGSUSED */
4574 static int
4575 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4576     page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4577     enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4578 {
4579 	znode_t		*zp = VTOZ(vp);
4580 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4581 	page_t		**pl0 = pl;
4582 	int		err = 0;
4583 
4584 	/* we do our own caching, faultahead is unnecessary */
4585 	if (pl == NULL)
4586 		return (0);
4587 	else if (len > plsz)
4588 		len = plsz;
4589 	else
4590 		len = P2ROUNDUP(len, PAGESIZE);
4591 	ASSERT(plsz >= len);
4592 
4593 	ZFS_ENTER(zfsvfs);
4594 	ZFS_VERIFY_ZP(zp);
4595 
4596 	if (protp)
4597 		*protp = PROT_ALL;
4598 
4599 	/*
4600 	 * Loop through the requested range [off, off + len) looking
4601 	 * for pages.  If we don't find a page, we will need to create
4602 	 * a new page and fill it with data from the file.
4603 	 */
4604 	while (len > 0) {
4605 		if (*pl = page_lookup(vp, off, SE_SHARED))
4606 			*(pl+1) = NULL;
4607 		else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4608 			goto out;
4609 		while (*pl) {
4610 			ASSERT3U((*pl)->p_offset, ==, off);
4611 			off += PAGESIZE;
4612 			addr += PAGESIZE;
4613 			if (len > 0) {
4614 				ASSERT3U(len, >=, PAGESIZE);
4615 				len -= PAGESIZE;
4616 			}
4617 			ASSERT3U(plsz, >=, PAGESIZE);
4618 			plsz -= PAGESIZE;
4619 			pl++;
4620 		}
4621 	}
4622 
4623 	/*
4624 	 * Fill out the page array with any pages already in the cache.
4625 	 */
4626 	while (plsz > 0 &&
4627 	    (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4628 			off += PAGESIZE;
4629 			plsz -= PAGESIZE;
4630 	}
4631 out:
4632 	if (err) {
4633 		/*
4634 		 * Release any pages we have previously locked.
4635 		 */
4636 		while (pl > pl0)
4637 			page_unlock(*--pl);
4638 	} else {
4639 		ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4640 	}
4641 
4642 	*pl = NULL;
4643 
4644 	ZFS_EXIT(zfsvfs);
4645 	return (err);
4646 }
4647 
4648 /*
4649  * Request a memory map for a section of a file.  This code interacts
4650  * with common code and the VM system as follows:
4651  *
4652  * - common code calls mmap(), which ends up in smmap_common()
4653  * - this calls VOP_MAP(), which takes you into (say) zfs
4654  * - zfs_map() calls as_map(), passing segvn_create() as the callback
4655  * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4656  * - zfs_addmap() updates z_mapcnt
4657  */
4658 /*ARGSUSED*/
4659 static int
4660 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4661     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4662     caller_context_t *ct)
4663 {
4664 	znode_t *zp = VTOZ(vp);
4665 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4666 	segvn_crargs_t	vn_a;
4667 	int		error;
4668 
4669 	ZFS_ENTER(zfsvfs);
4670 	ZFS_VERIFY_ZP(zp);
4671 
4672 	if ((prot & PROT_WRITE) && (zp->z_pflags &
4673 	    (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4674 		ZFS_EXIT(zfsvfs);
4675 		return (SET_ERROR(EPERM));
4676 	}
4677 
4678 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4679 	    (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4680 		ZFS_EXIT(zfsvfs);
4681 		return (SET_ERROR(EACCES));
4682 	}
4683 
4684 	if (vp->v_flag & VNOMAP) {
4685 		ZFS_EXIT(zfsvfs);
4686 		return (SET_ERROR(ENOSYS));
4687 	}
4688 
4689 	if (off < 0 || len > MAXOFFSET_T - off) {
4690 		ZFS_EXIT(zfsvfs);
4691 		return (SET_ERROR(ENXIO));
4692 	}
4693 
4694 	if (vp->v_type != VREG) {
4695 		ZFS_EXIT(zfsvfs);
4696 		return (SET_ERROR(ENODEV));
4697 	}
4698 
4699 	/*
4700 	 * If file is locked, disallow mapping.
4701 	 */
4702 	if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4703 		ZFS_EXIT(zfsvfs);
4704 		return (SET_ERROR(EAGAIN));
4705 	}
4706 
4707 	as_rangelock(as);
4708 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4709 	if (error != 0) {
4710 		as_rangeunlock(as);
4711 		ZFS_EXIT(zfsvfs);
4712 		return (error);
4713 	}
4714 
4715 	vn_a.vp = vp;
4716 	vn_a.offset = (u_offset_t)off;
4717 	vn_a.type = flags & MAP_TYPE;
4718 	vn_a.prot = prot;
4719 	vn_a.maxprot = maxprot;
4720 	vn_a.cred = cr;
4721 	vn_a.amp = NULL;
4722 	vn_a.flags = flags & ~MAP_TYPE;
4723 	vn_a.szc = 0;
4724 	vn_a.lgrp_mem_policy_flags = 0;
4725 
4726 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4727 
4728 	as_rangeunlock(as);
4729 	ZFS_EXIT(zfsvfs);
4730 	return (error);
4731 }
4732 
4733 /* ARGSUSED */
4734 static int
4735 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4736     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4737     caller_context_t *ct)
4738 {
4739 	uint64_t pages = btopr(len);
4740 
4741 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4742 	return (0);
4743 }
4744 
4745 /*
4746  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4747  * more accurate mtime for the associated file.  Since we don't have a way of
4748  * detecting when the data was actually modified, we have to resort to
4749  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4750  * last page is pushed.  The problem occurs when the msync() call is omitted,
4751  * which by far the most common case:
4752  *
4753  *	open()
4754  *	mmap()
4755  *	<modify memory>
4756  *	munmap()
4757  *	close()
4758  *	<time lapse>
4759  *	putpage() via fsflush
4760  *
4761  * If we wait until fsflush to come along, we can have a modification time that
4762  * is some arbitrary point in the future.  In order to prevent this in the
4763  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4764  * torn down.
4765  */
4766 /* ARGSUSED */
4767 static int
4768 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4769     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4770     caller_context_t *ct)
4771 {
4772 	uint64_t pages = btopr(len);
4773 
4774 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4775 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4776 
4777 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4778 	    vn_has_cached_data(vp))
4779 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4780 
4781 	return (0);
4782 }
4783 
4784 /*
4785  * Free or allocate space in a file.  Currently, this function only
4786  * supports the `F_FREESP' command.  However, this command is somewhat
4787  * misnamed, as its functionality includes the ability to allocate as
4788  * well as free space.
4789  *
4790  *	IN:	vp	- vnode of file to free data in.
4791  *		cmd	- action to take (only F_FREESP supported).
4792  *		bfp	- section of file to free/alloc.
4793  *		flag	- current file open mode flags.
4794  *		offset	- current file offset.
4795  *		cr	- credentials of caller [UNUSED].
4796  *		ct	- caller context.
4797  *
4798  *	RETURN:	0 on success, error code on failure.
4799  *
4800  * Timestamps:
4801  *	vp - ctime|mtime updated
4802  */
4803 /* ARGSUSED */
4804 static int
4805 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4806     offset_t offset, cred_t *cr, caller_context_t *ct)
4807 {
4808 	znode_t		*zp = VTOZ(vp);
4809 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4810 	uint64_t	off, len;
4811 	int		error;
4812 
4813 	ZFS_ENTER(zfsvfs);
4814 	ZFS_VERIFY_ZP(zp);
4815 
4816 	if (cmd != F_FREESP) {
4817 		ZFS_EXIT(zfsvfs);
4818 		return (SET_ERROR(EINVAL));
4819 	}
4820 
4821 	/*
4822 	 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
4823 	 * callers might not be able to detect properly that we are read-only,
4824 	 * so check it explicitly here.
4825 	 */
4826 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
4827 		ZFS_EXIT(zfsvfs);
4828 		return (SET_ERROR(EROFS));
4829 	}
4830 
4831 	if (error = convoff(vp, bfp, 0, offset)) {
4832 		ZFS_EXIT(zfsvfs);
4833 		return (error);
4834 	}
4835 
4836 	if (bfp->l_len < 0) {
4837 		ZFS_EXIT(zfsvfs);
4838 		return (SET_ERROR(EINVAL));
4839 	}
4840 
4841 	off = bfp->l_start;
4842 	len = bfp->l_len; /* 0 means from off to end of file */
4843 
4844 	error = zfs_freesp(zp, off, len, flag, TRUE);
4845 
4846 	if (error == 0 && off == 0 && len == 0)
4847 		vnevent_truncate(ZTOV(zp), ct);
4848 
4849 	ZFS_EXIT(zfsvfs);
4850 	return (error);
4851 }
4852 
4853 /*ARGSUSED*/
4854 static int
4855 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4856 {
4857 	znode_t		*zp = VTOZ(vp);
4858 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4859 	uint32_t	gen;
4860 	uint64_t	gen64;
4861 	uint64_t	object = zp->z_id;
4862 	zfid_short_t	*zfid;
4863 	int		size, i, error;
4864 
4865 	ZFS_ENTER(zfsvfs);
4866 	ZFS_VERIFY_ZP(zp);
4867 
4868 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4869 	    &gen64, sizeof (uint64_t))) != 0) {
4870 		ZFS_EXIT(zfsvfs);
4871 		return (error);
4872 	}
4873 
4874 	gen = (uint32_t)gen64;
4875 
4876 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4877 	if (fidp->fid_len < size) {
4878 		fidp->fid_len = size;
4879 		ZFS_EXIT(zfsvfs);
4880 		return (SET_ERROR(ENOSPC));
4881 	}
4882 
4883 	zfid = (zfid_short_t *)fidp;
4884 
4885 	zfid->zf_len = size;
4886 
4887 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4888 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4889 
4890 	/* Must have a non-zero generation number to distinguish from .zfs */
4891 	if (gen == 0)
4892 		gen = 1;
4893 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4894 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4895 
4896 	if (size == LONG_FID_LEN) {
4897 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4898 		zfid_long_t	*zlfid;
4899 
4900 		zlfid = (zfid_long_t *)fidp;
4901 
4902 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4903 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4904 
4905 		/* XXX - this should be the generation number for the objset */
4906 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4907 			zlfid->zf_setgen[i] = 0;
4908 	}
4909 
4910 	ZFS_EXIT(zfsvfs);
4911 	return (0);
4912 }
4913 
4914 static int
4915 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4916     caller_context_t *ct)
4917 {
4918 	znode_t		*zp, *xzp;
4919 	zfsvfs_t	*zfsvfs;
4920 	zfs_dirlock_t	*dl;
4921 	int		error;
4922 
4923 	switch (cmd) {
4924 	case _PC_LINK_MAX:
4925 		*valp = ULONG_MAX;
4926 		return (0);
4927 
4928 	case _PC_FILESIZEBITS:
4929 		*valp = 64;
4930 		return (0);
4931 
4932 	case _PC_XATTR_EXISTS:
4933 		zp = VTOZ(vp);
4934 		zfsvfs = zp->z_zfsvfs;
4935 		ZFS_ENTER(zfsvfs);
4936 		ZFS_VERIFY_ZP(zp);
4937 		*valp = 0;
4938 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4939 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4940 		if (error == 0) {
4941 			zfs_dirent_unlock(dl);
4942 			if (!zfs_dirempty(xzp))
4943 				*valp = 1;
4944 			VN_RELE(ZTOV(xzp));
4945 		} else if (error == ENOENT) {
4946 			/*
4947 			 * If there aren't extended attributes, it's the
4948 			 * same as having zero of them.
4949 			 */
4950 			error = 0;
4951 		}
4952 		ZFS_EXIT(zfsvfs);
4953 		return (error);
4954 
4955 	case _PC_SATTR_ENABLED:
4956 	case _PC_SATTR_EXISTS:
4957 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4958 		    (vp->v_type == VREG || vp->v_type == VDIR);
4959 		return (0);
4960 
4961 	case _PC_ACCESS_FILTERING:
4962 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4963 		    vp->v_type == VDIR;
4964 		return (0);
4965 
4966 	case _PC_ACL_ENABLED:
4967 		*valp = _ACL_ACE_ENABLED;
4968 		return (0);
4969 
4970 	case _PC_MIN_HOLE_SIZE:
4971 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4972 		return (0);
4973 
4974 	case _PC_TIMESTAMP_RESOLUTION:
4975 		/* nanosecond timestamp resolution */
4976 		*valp = 1L;
4977 		return (0);
4978 
4979 	default:
4980 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4981 	}
4982 }
4983 
4984 /*ARGSUSED*/
4985 static int
4986 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4987     caller_context_t *ct)
4988 {
4989 	znode_t *zp = VTOZ(vp);
4990 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4991 	int error;
4992 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4993 
4994 	ZFS_ENTER(zfsvfs);
4995 	ZFS_VERIFY_ZP(zp);
4996 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4997 	ZFS_EXIT(zfsvfs);
4998 
4999 	return (error);
5000 }
5001 
5002 /*ARGSUSED*/
5003 static int
5004 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5005     caller_context_t *ct)
5006 {
5007 	znode_t *zp = VTOZ(vp);
5008 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5009 	int error;
5010 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5011 	zilog_t	*zilog = zfsvfs->z_log;
5012 
5013 	ZFS_ENTER(zfsvfs);
5014 	ZFS_VERIFY_ZP(zp);
5015 
5016 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5017 
5018 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5019 		zil_commit(zilog, 0);
5020 
5021 	ZFS_EXIT(zfsvfs);
5022 	return (error);
5023 }
5024 
5025 /*
5026  * The smallest read we may consider to loan out an arcbuf.
5027  * This must be a power of 2.
5028  */
5029 int zcr_blksz_min = (1 << 10);	/* 1K */
5030 /*
5031  * If set to less than the file block size, allow loaning out of an
5032  * arcbuf for a partial block read.  This must be a power of 2.
5033  */
5034 int zcr_blksz_max = (1 << 17);	/* 128K */
5035 
5036 /*ARGSUSED*/
5037 static int
5038 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5039     caller_context_t *ct)
5040 {
5041 	znode_t	*zp = VTOZ(vp);
5042 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5043 	int max_blksz = zfsvfs->z_max_blksz;
5044 	uio_t *uio = &xuio->xu_uio;
5045 	ssize_t size = uio->uio_resid;
5046 	offset_t offset = uio->uio_loffset;
5047 	int blksz;
5048 	int fullblk, i;
5049 	arc_buf_t *abuf;
5050 	ssize_t maxsize;
5051 	int preamble, postamble;
5052 
5053 	if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5054 		return (SET_ERROR(EINVAL));
5055 
5056 	ZFS_ENTER(zfsvfs);
5057 	ZFS_VERIFY_ZP(zp);
5058 	switch (ioflag) {
5059 	case UIO_WRITE:
5060 		/*
5061 		 * Loan out an arc_buf for write if write size is bigger than
5062 		 * max_blksz, and the file's block size is also max_blksz.
5063 		 */
5064 		blksz = max_blksz;
5065 		if (size < blksz || zp->z_blksz != blksz) {
5066 			ZFS_EXIT(zfsvfs);
5067 			return (SET_ERROR(EINVAL));
5068 		}
5069 		/*
5070 		 * Caller requests buffers for write before knowing where the
5071 		 * write offset might be (e.g. NFS TCP write).
5072 		 */
5073 		if (offset == -1) {
5074 			preamble = 0;
5075 		} else {
5076 			preamble = P2PHASE(offset, blksz);
5077 			if (preamble) {
5078 				preamble = blksz - preamble;
5079 				size -= preamble;
5080 			}
5081 		}
5082 
5083 		postamble = P2PHASE(size, blksz);
5084 		size -= postamble;
5085 
5086 		fullblk = size / blksz;
5087 		(void) dmu_xuio_init(xuio,
5088 		    (preamble != 0) + fullblk + (postamble != 0));
5089 		DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5090 		    int, postamble, int,
5091 		    (preamble != 0) + fullblk + (postamble != 0));
5092 
5093 		/*
5094 		 * Have to fix iov base/len for partial buffers.  They
5095 		 * currently represent full arc_buf's.
5096 		 */
5097 		if (preamble) {
5098 			/* data begins in the middle of the arc_buf */
5099 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5100 			    blksz);
5101 			ASSERT(abuf);
5102 			(void) dmu_xuio_add(xuio, abuf,
5103 			    blksz - preamble, preamble);
5104 		}
5105 
5106 		for (i = 0; i < fullblk; i++) {
5107 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5108 			    blksz);
5109 			ASSERT(abuf);
5110 			(void) dmu_xuio_add(xuio, abuf, 0, blksz);
5111 		}
5112 
5113 		if (postamble) {
5114 			/* data ends in the middle of the arc_buf */
5115 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5116 			    blksz);
5117 			ASSERT(abuf);
5118 			(void) dmu_xuio_add(xuio, abuf, 0, postamble);
5119 		}
5120 		break;
5121 	case UIO_READ:
5122 		/*
5123 		 * Loan out an arc_buf for read if the read size is larger than
5124 		 * the current file block size.  Block alignment is not
5125 		 * considered.  Partial arc_buf will be loaned out for read.
5126 		 */
5127 		blksz = zp->z_blksz;
5128 		if (blksz < zcr_blksz_min)
5129 			blksz = zcr_blksz_min;
5130 		if (blksz > zcr_blksz_max)
5131 			blksz = zcr_blksz_max;
5132 		/* avoid potential complexity of dealing with it */
5133 		if (blksz > max_blksz) {
5134 			ZFS_EXIT(zfsvfs);
5135 			return (SET_ERROR(EINVAL));
5136 		}
5137 
5138 		maxsize = zp->z_size - uio->uio_loffset;
5139 		if (size > maxsize)
5140 			size = maxsize;
5141 
5142 		if (size < blksz || vn_has_cached_data(vp)) {
5143 			ZFS_EXIT(zfsvfs);
5144 			return (SET_ERROR(EINVAL));
5145 		}
5146 		break;
5147 	default:
5148 		ZFS_EXIT(zfsvfs);
5149 		return (SET_ERROR(EINVAL));
5150 	}
5151 
5152 	uio->uio_extflg = UIO_XUIO;
5153 	XUIO_XUZC_RW(xuio) = ioflag;
5154 	ZFS_EXIT(zfsvfs);
5155 	return (0);
5156 }
5157 
5158 /*ARGSUSED*/
5159 static int
5160 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5161 {
5162 	int i;
5163 	arc_buf_t *abuf;
5164 	int ioflag = XUIO_XUZC_RW(xuio);
5165 
5166 	ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5167 
5168 	i = dmu_xuio_cnt(xuio);
5169 	while (i-- > 0) {
5170 		abuf = dmu_xuio_arcbuf(xuio, i);
5171 		/*
5172 		 * if abuf == NULL, it must be a write buffer
5173 		 * that has been returned in zfs_write().
5174 		 */
5175 		if (abuf)
5176 			dmu_return_arcbuf(abuf);
5177 		ASSERT(abuf || ioflag == UIO_WRITE);
5178 	}
5179 
5180 	dmu_xuio_fini(xuio);
5181 	return (0);
5182 }
5183 
5184 /*
5185  * Predeclare these here so that the compiler assumes that
5186  * this is an "old style" function declaration that does
5187  * not include arguments => we won't get type mismatch errors
5188  * in the initializations that follow.
5189  */
5190 static int zfs_inval();
5191 static int zfs_isdir();
5192 
5193 static int
5194 zfs_inval()
5195 {
5196 	return (SET_ERROR(EINVAL));
5197 }
5198 
5199 static int
5200 zfs_isdir()
5201 {
5202 	return (SET_ERROR(EISDIR));
5203 }
5204 /*
5205  * Directory vnode operations template
5206  */
5207 vnodeops_t *zfs_dvnodeops;
5208 const fs_operation_def_t zfs_dvnodeops_template[] = {
5209 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5210 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5211 	VOPNAME_READ,		{ .error = zfs_isdir },
5212 	VOPNAME_WRITE,		{ .error = zfs_isdir },
5213 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5214 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5215 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5216 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5217 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5218 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5219 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5220 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5221 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5222 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
5223 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5224 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5225 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
5226 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5227 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5228 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5229 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5230 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5231 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5232 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5233 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5234 	NULL,			NULL
5235 };
5236 
5237 /*
5238  * Regular file vnode operations template
5239  */
5240 vnodeops_t *zfs_fvnodeops;
5241 const fs_operation_def_t zfs_fvnodeops_template[] = {
5242 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5243 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5244 	VOPNAME_READ,		{ .vop_read = zfs_read },
5245 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
5246 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5247 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5248 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5249 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5250 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5251 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5252 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5253 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5254 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5255 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5256 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
5257 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
5258 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
5259 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
5260 	VOPNAME_MAP,		{ .vop_map = zfs_map },
5261 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
5262 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
5263 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5264 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5265 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5266 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5267 	VOPNAME_REQZCBUF,	{ .vop_reqzcbuf = zfs_reqzcbuf },
5268 	VOPNAME_RETZCBUF,	{ .vop_retzcbuf = zfs_retzcbuf },
5269 	NULL,			NULL
5270 };
5271 
5272 /*
5273  * Symbolic link vnode operations template
5274  */
5275 vnodeops_t *zfs_symvnodeops;
5276 const fs_operation_def_t zfs_symvnodeops_template[] = {
5277 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5278 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5279 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5280 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5281 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
5282 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5283 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5284 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5285 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5286 	NULL,			NULL
5287 };
5288 
5289 /*
5290  * special share hidden files vnode operations template
5291  */
5292 vnodeops_t *zfs_sharevnodeops;
5293 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5294 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5295 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5296 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5297 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5298 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5299 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5300 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5301 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5302 	NULL,			NULL
5303 };
5304 
5305 /*
5306  * Extended attribute directory vnode operations template
5307  *
5308  * This template is identical to the directory vnodes
5309  * operation template except for restricted operations:
5310  *	VOP_MKDIR()
5311  *	VOP_SYMLINK()
5312  *
5313  * Note that there are other restrictions embedded in:
5314  *	zfs_create()	- restrict type to VREG
5315  *	zfs_link()	- no links into/out of attribute space
5316  *	zfs_rename()	- no moves into/out of attribute space
5317  */
5318 vnodeops_t *zfs_xdvnodeops;
5319 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5320 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5321 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5322 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5323 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5324 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5325 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5326 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5327 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5328 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5329 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5330 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5331 	VOPNAME_MKDIR,		{ .error = zfs_inval },
5332 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5333 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5334 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
5335 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5336 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5337 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5338 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5339 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5340 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5341 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5342 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5343 	NULL,			NULL
5344 };
5345 
5346 /*
5347  * Error vnode operations template
5348  */
5349 vnodeops_t *zfs_evnodeops;
5350 const fs_operation_def_t zfs_evnodeops_template[] = {
5351 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5352 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5353 	NULL,			NULL
5354 };
5355