xref: /titanic_50/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 7bc22e45a20f905cdd06bb98c98a5c8be7fd25c0)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /* Portions Copyright 2007 Jeremy Teo */
27 
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/time.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
33 #include <sys/resource.h>
34 #include <sys/vfs.h>
35 #include <sys/vfs_opreg.h>
36 #include <sys/vnode.h>
37 #include <sys/file.h>
38 #include <sys/stat.h>
39 #include <sys/kmem.h>
40 #include <sys/taskq.h>
41 #include <sys/uio.h>
42 #include <sys/vmsystm.h>
43 #include <sys/atomic.h>
44 #include <sys/vm.h>
45 #include <vm/seg_vn.h>
46 #include <vm/pvn.h>
47 #include <vm/as.h>
48 #include <vm/kpm.h>
49 #include <vm/seg_kpm.h>
50 #include <sys/mman.h>
51 #include <sys/pathname.h>
52 #include <sys/cmn_err.h>
53 #include <sys/errno.h>
54 #include <sys/unistd.h>
55 #include <sys/zfs_dir.h>
56 #include <sys/zfs_acl.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/fs/zfs.h>
59 #include <sys/dmu.h>
60 #include <sys/spa.h>
61 #include <sys/txg.h>
62 #include <sys/dbuf.h>
63 #include <sys/zap.h>
64 #include <sys/dirent.h>
65 #include <sys/policy.h>
66 #include <sys/sunddi.h>
67 #include <sys/filio.h>
68 #include <sys/sid.h>
69 #include "fs/fs_subr.h"
70 #include <sys/zfs_ctldir.h>
71 #include <sys/zfs_fuid.h>
72 #include <sys/dnlc.h>
73 #include <sys/zfs_rlock.h>
74 #include <sys/extdirent.h>
75 #include <sys/kidmap.h>
76 #include <sys/cred_impl.h>
77 #include <sys/attr.h>
78 
79 /*
80  * Programming rules.
81  *
82  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
83  * properly lock its in-core state, create a DMU transaction, do the work,
84  * record this work in the intent log (ZIL), commit the DMU transaction,
85  * and wait for the intent log to commit if it is a synchronous operation.
86  * Moreover, the vnode ops must work in both normal and log replay context.
87  * The ordering of events is important to avoid deadlocks and references
88  * to freed memory.  The example below illustrates the following Big Rules:
89  *
90  *  (1) A check must be made in each zfs thread for a mounted file system.
91  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
92  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
93  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
94  *      can return EIO from the calling function.
95  *
96  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
97  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
98  *	First, if it's the last reference, the vnode/znode
99  *	can be freed, so the zp may point to freed memory.  Second, the last
100  *	reference will call zfs_zinactive(), which may induce a lot of work --
101  *	pushing cached pages (which acquires range locks) and syncing out
102  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
103  *	which could deadlock the system if you were already holding one.
104  *
105  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
106  *	as they can span dmu_tx_assign() calls.
107  *
108  *  (4)	Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign().
109  *	In normal operation, this will be TXG_NOWAIT.  During ZIL replay,
110  *	it will be a specific txg.  Either way, dmu_tx_assign() never blocks.
111  *	This is critical because we don't want to block while holding locks.
112  *	Note, in particular, that if a lock is sometimes acquired before
113  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
114  *	use a non-blocking assign can deadlock the system.  The scenario:
115  *
116  *	Thread A has grabbed a lock before calling dmu_tx_assign().
117  *	Thread B is in an already-assigned tx, and blocks for this lock.
118  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
119  *	forever, because the previous txg can't quiesce until B's tx commits.
120  *
121  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
122  *	then drop all locks, call dmu_tx_wait(), and try again.
123  *
124  *  (5)	If the operation succeeded, generate the intent log entry for it
125  *	before dropping locks.  This ensures that the ordering of events
126  *	in the intent log matches the order in which they actually occurred.
127  *
128  *  (6)	At the end of each vnode op, the DMU tx must always commit,
129  *	regardless of whether there were any errors.
130  *
131  *  (7)	After dropping all locks, invoke zil_commit(zilog, seq, foid)
132  *	to ensure that synchronous semantics are provided when necessary.
133  *
134  * In general, this is how things should be ordered in each vnode op:
135  *
136  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
137  * top:
138  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
139  *	rw_enter(...);			// grab any other locks you need
140  *	tx = dmu_tx_create(...);	// get DMU tx
141  *	dmu_tx_hold_*();		// hold each object you might modify
142  *	error = dmu_tx_assign(tx, zfsvfs->z_assign);	// try to assign
143  *	if (error) {
144  *		rw_exit(...);		// drop locks
145  *		zfs_dirent_unlock(dl);	// unlock directory entry
146  *		VN_RELE(...);		// release held vnodes
147  *		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
148  *			dmu_tx_wait(tx);
149  *			dmu_tx_abort(tx);
150  *			goto top;
151  *		}
152  *		dmu_tx_abort(tx);	// abort DMU tx
153  *		ZFS_EXIT(zfsvfs);	// finished in zfs
154  *		return (error);		// really out of space
155  *	}
156  *	error = do_real_work();		// do whatever this VOP does
157  *	if (error == 0)
158  *		zfs_log_*(...);		// on success, make ZIL entry
159  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
160  *	rw_exit(...);			// drop locks
161  *	zfs_dirent_unlock(dl);		// unlock directory entry
162  *	VN_RELE(...);			// release held vnodes
163  *	zil_commit(zilog, seq, foid);	// synchronous when necessary
164  *	ZFS_EXIT(zfsvfs);		// finished in zfs
165  *	return (error);			// done, report error
166  */
167 
168 /* ARGSUSED */
169 static int
170 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
171 {
172 	znode_t	*zp = VTOZ(*vpp);
173 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
174 
175 	ZFS_ENTER(zfsvfs);
176 	ZFS_VERIFY_ZP(zp);
177 
178 	if ((flag & FWRITE) && (zp->z_phys->zp_flags & ZFS_APPENDONLY) &&
179 	    ((flag & FAPPEND) == 0)) {
180 		ZFS_EXIT(zfsvfs);
181 		return (EPERM);
182 	}
183 
184 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
185 	    ZTOV(zp)->v_type == VREG &&
186 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
187 	    zp->z_phys->zp_size > 0) {
188 		if (fs_vscan(*vpp, cr, 0) != 0) {
189 			ZFS_EXIT(zfsvfs);
190 			return (EACCES);
191 		}
192 	}
193 
194 	/* Keep a count of the synchronous opens in the znode */
195 	if (flag & (FSYNC | FDSYNC))
196 		atomic_inc_32(&zp->z_sync_cnt);
197 
198 	ZFS_EXIT(zfsvfs);
199 	return (0);
200 }
201 
202 /* ARGSUSED */
203 static int
204 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
205     caller_context_t *ct)
206 {
207 	znode_t	*zp = VTOZ(vp);
208 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
209 
210 	ZFS_ENTER(zfsvfs);
211 	ZFS_VERIFY_ZP(zp);
212 
213 	/* Decrement the synchronous opens in the znode */
214 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
215 		atomic_dec_32(&zp->z_sync_cnt);
216 
217 	/*
218 	 * Clean up any locks held by this process on the vp.
219 	 */
220 	cleanlocks(vp, ddi_get_pid(), 0);
221 	cleanshares(vp, ddi_get_pid());
222 
223 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
224 	    ZTOV(zp)->v_type == VREG &&
225 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
226 	    zp->z_phys->zp_size > 0)
227 		VERIFY(fs_vscan(vp, cr, 1) == 0);
228 
229 	ZFS_EXIT(zfsvfs);
230 	return (0);
231 }
232 
233 /*
234  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
235  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
236  */
237 static int
238 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
239 {
240 	znode_t	*zp = VTOZ(vp);
241 	uint64_t noff = (uint64_t)*off; /* new offset */
242 	uint64_t file_sz;
243 	int error;
244 	boolean_t hole;
245 
246 	file_sz = zp->z_phys->zp_size;
247 	if (noff >= file_sz)  {
248 		return (ENXIO);
249 	}
250 
251 	if (cmd == _FIO_SEEK_HOLE)
252 		hole = B_TRUE;
253 	else
254 		hole = B_FALSE;
255 
256 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
257 
258 	/* end of file? */
259 	if ((error == ESRCH) || (noff > file_sz)) {
260 		/*
261 		 * Handle the virtual hole at the end of file.
262 		 */
263 		if (hole) {
264 			*off = file_sz;
265 			return (0);
266 		}
267 		return (ENXIO);
268 	}
269 
270 	if (noff < *off)
271 		return (error);
272 	*off = noff;
273 	return (error);
274 }
275 
276 /* ARGSUSED */
277 static int
278 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
279     int *rvalp, caller_context_t *ct)
280 {
281 	offset_t off;
282 	int error;
283 	zfsvfs_t *zfsvfs;
284 	znode_t *zp;
285 
286 	switch (com) {
287 	case _FIOFFS:
288 		return (zfs_sync(vp->v_vfsp, 0, cred));
289 
290 		/*
291 		 * The following two ioctls are used by bfu.  Faking out,
292 		 * necessary to avoid bfu errors.
293 		 */
294 	case _FIOGDIO:
295 	case _FIOSDIO:
296 		return (0);
297 
298 	case _FIO_SEEK_DATA:
299 	case _FIO_SEEK_HOLE:
300 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
301 			return (EFAULT);
302 
303 		zp = VTOZ(vp);
304 		zfsvfs = zp->z_zfsvfs;
305 		ZFS_ENTER(zfsvfs);
306 		ZFS_VERIFY_ZP(zp);
307 
308 		/* offset parameter is in/out */
309 		error = zfs_holey(vp, com, &off);
310 		ZFS_EXIT(zfsvfs);
311 		if (error)
312 			return (error);
313 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
314 			return (EFAULT);
315 		return (0);
316 	}
317 	return (ENOTTY);
318 }
319 
320 /*
321  * Utility functions to map and unmap a single physical page.  These
322  * are used to manage the mappable copies of ZFS file data, and therefore
323  * do not update ref/mod bits.
324  */
325 caddr_t
326 zfs_map_page(page_t *pp, enum seg_rw rw)
327 {
328 	if (kpm_enable)
329 		return (hat_kpm_mapin(pp, 0));
330 	ASSERT(rw == S_READ || rw == S_WRITE);
331 	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
332 	    (caddr_t)-1));
333 }
334 
335 void
336 zfs_unmap_page(page_t *pp, caddr_t addr)
337 {
338 	if (kpm_enable) {
339 		hat_kpm_mapout(pp, 0, addr);
340 	} else {
341 		ppmapout(addr);
342 	}
343 }
344 
345 /*
346  * When a file is memory mapped, we must keep the IO data synchronized
347  * between the DMU cache and the memory mapped pages.  What this means:
348  *
349  * On Write:	If we find a memory mapped page, we write to *both*
350  *		the page and the dmu buffer.
351  *
352  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
353  *	the file is memory mapped.
354  */
355 static int
356 mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx)
357 {
358 	znode_t	*zp = VTOZ(vp);
359 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
360 	int64_t	start, off;
361 	int len = nbytes;
362 	int error = 0;
363 
364 	start = uio->uio_loffset;
365 	off = start & PAGEOFFSET;
366 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
367 		page_t *pp;
368 		uint64_t bytes = MIN(PAGESIZE - off, len);
369 		uint64_t woff = uio->uio_loffset;
370 
371 		/*
372 		 * We don't want a new page to "appear" in the middle of
373 		 * the file update (because it may not get the write
374 		 * update data), so we grab a lock to block
375 		 * zfs_getpage().
376 		 */
377 		rw_enter(&zp->z_map_lock, RW_WRITER);
378 		if (pp = page_lookup(vp, start, SE_SHARED)) {
379 			caddr_t va;
380 
381 			rw_exit(&zp->z_map_lock);
382 			va = zfs_map_page(pp, S_WRITE);
383 			error = uiomove(va+off, bytes, UIO_WRITE, uio);
384 			if (error == 0) {
385 				dmu_write(zfsvfs->z_os, zp->z_id,
386 				    woff, bytes, va+off, tx);
387 			}
388 			zfs_unmap_page(pp, va);
389 			page_unlock(pp);
390 		} else {
391 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
392 			    uio, bytes, tx);
393 			rw_exit(&zp->z_map_lock);
394 		}
395 		len -= bytes;
396 		off = 0;
397 		if (error)
398 			break;
399 	}
400 	return (error);
401 }
402 
403 /*
404  * When a file is memory mapped, we must keep the IO data synchronized
405  * between the DMU cache and the memory mapped pages.  What this means:
406  *
407  * On Read:	We "read" preferentially from memory mapped pages,
408  *		else we default from the dmu buffer.
409  *
410  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
411  *	the file is memory mapped.
412  */
413 static int
414 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
415 {
416 	znode_t *zp = VTOZ(vp);
417 	objset_t *os = zp->z_zfsvfs->z_os;
418 	int64_t	start, off;
419 	int len = nbytes;
420 	int error = 0;
421 
422 	start = uio->uio_loffset;
423 	off = start & PAGEOFFSET;
424 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
425 		page_t *pp;
426 		uint64_t bytes = MIN(PAGESIZE - off, len);
427 
428 		if (pp = page_lookup(vp, start, SE_SHARED)) {
429 			caddr_t va;
430 
431 			va = zfs_map_page(pp, S_READ);
432 			error = uiomove(va + off, bytes, UIO_READ, uio);
433 			zfs_unmap_page(pp, va);
434 			page_unlock(pp);
435 		} else {
436 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
437 		}
438 		len -= bytes;
439 		off = 0;
440 		if (error)
441 			break;
442 	}
443 	return (error);
444 }
445 
446 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
447 
448 /*
449  * Read bytes from specified file into supplied buffer.
450  *
451  *	IN:	vp	- vnode of file to be read from.
452  *		uio	- structure supplying read location, range info,
453  *			  and return buffer.
454  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
455  *		cr	- credentials of caller.
456  *		ct	- caller context
457  *
458  *	OUT:	uio	- updated offset and range, buffer filled.
459  *
460  *	RETURN:	0 if success
461  *		error code if failure
462  *
463  * Side Effects:
464  *	vp - atime updated if byte count > 0
465  */
466 /* ARGSUSED */
467 static int
468 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
469 {
470 	znode_t		*zp = VTOZ(vp);
471 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
472 	objset_t	*os;
473 	ssize_t		n, nbytes;
474 	int		error;
475 	rl_t		*rl;
476 
477 	ZFS_ENTER(zfsvfs);
478 	ZFS_VERIFY_ZP(zp);
479 	os = zfsvfs->z_os;
480 
481 	if (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) {
482 		ZFS_EXIT(zfsvfs);
483 		return (EACCES);
484 	}
485 
486 	/*
487 	 * Validate file offset
488 	 */
489 	if (uio->uio_loffset < (offset_t)0) {
490 		ZFS_EXIT(zfsvfs);
491 		return (EINVAL);
492 	}
493 
494 	/*
495 	 * Fasttrack empty reads
496 	 */
497 	if (uio->uio_resid == 0) {
498 		ZFS_EXIT(zfsvfs);
499 		return (0);
500 	}
501 
502 	/*
503 	 * Check for mandatory locks
504 	 */
505 	if (MANDMODE((mode_t)zp->z_phys->zp_mode)) {
506 		if (error = chklock(vp, FREAD,
507 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
508 			ZFS_EXIT(zfsvfs);
509 			return (error);
510 		}
511 	}
512 
513 	/*
514 	 * If we're in FRSYNC mode, sync out this znode before reading it.
515 	 */
516 	if (ioflag & FRSYNC)
517 		zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
518 
519 	/*
520 	 * Lock the range against changes.
521 	 */
522 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
523 
524 	/*
525 	 * If we are reading past end-of-file we can skip
526 	 * to the end; but we might still need to set atime.
527 	 */
528 	if (uio->uio_loffset >= zp->z_phys->zp_size) {
529 		error = 0;
530 		goto out;
531 	}
532 
533 	ASSERT(uio->uio_loffset < zp->z_phys->zp_size);
534 	n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset);
535 
536 	while (n > 0) {
537 		nbytes = MIN(n, zfs_read_chunk_size -
538 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
539 
540 		if (vn_has_cached_data(vp))
541 			error = mappedread(vp, nbytes, uio);
542 		else
543 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
544 		if (error) {
545 			/* convert checksum errors into IO errors */
546 			if (error == ECKSUM)
547 				error = EIO;
548 			break;
549 		}
550 
551 		n -= nbytes;
552 	}
553 
554 out:
555 	zfs_range_unlock(rl);
556 
557 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
558 	ZFS_EXIT(zfsvfs);
559 	return (error);
560 }
561 
562 /*
563  * Write the bytes to a file.
564  *
565  *	IN:	vp	- vnode of file to be written to.
566  *		uio	- structure supplying write location, range info,
567  *			  and data buffer.
568  *		ioflag	- FAPPEND flag set if in append mode.
569  *		cr	- credentials of caller.
570  *		ct	- caller context (NFS/CIFS fem monitor only)
571  *
572  *	OUT:	uio	- updated offset and range.
573  *
574  *	RETURN:	0 if success
575  *		error code if failure
576  *
577  * Timestamps:
578  *	vp - ctime|mtime updated if byte count > 0
579  */
580 /* ARGSUSED */
581 static int
582 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
583 {
584 	znode_t		*zp = VTOZ(vp);
585 	rlim64_t	limit = uio->uio_llimit;
586 	ssize_t		start_resid = uio->uio_resid;
587 	ssize_t		tx_bytes;
588 	uint64_t	end_size;
589 	dmu_tx_t	*tx;
590 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
591 	zilog_t		*zilog;
592 	offset_t	woff;
593 	ssize_t		n, nbytes;
594 	rl_t		*rl;
595 	int		max_blksz = zfsvfs->z_max_blksz;
596 	uint64_t	pflags;
597 	int		error;
598 
599 	/*
600 	 * Fasttrack empty write
601 	 */
602 	n = start_resid;
603 	if (n == 0)
604 		return (0);
605 
606 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
607 		limit = MAXOFFSET_T;
608 
609 	ZFS_ENTER(zfsvfs);
610 	ZFS_VERIFY_ZP(zp);
611 
612 	/*
613 	 * If immutable or not appending then return EPERM
614 	 */
615 	pflags = zp->z_phys->zp_flags;
616 	if ((pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
617 	    ((pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
618 	    (uio->uio_loffset < zp->z_phys->zp_size))) {
619 		ZFS_EXIT(zfsvfs);
620 		return (EPERM);
621 	}
622 
623 	zilog = zfsvfs->z_log;
624 
625 	/*
626 	 * Pre-fault the pages to ensure slow (eg NFS) pages
627 	 * don't hold up txg.
628 	 */
629 	uio_prefaultpages(n, uio);
630 
631 	/*
632 	 * If in append mode, set the io offset pointer to eof.
633 	 */
634 	if (ioflag & FAPPEND) {
635 		/*
636 		 * Range lock for a file append:
637 		 * The value for the start of range will be determined by
638 		 * zfs_range_lock() (to guarantee append semantics).
639 		 * If this write will cause the block size to increase,
640 		 * zfs_range_lock() will lock the entire file, so we must
641 		 * later reduce the range after we grow the block size.
642 		 */
643 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
644 		if (rl->r_len == UINT64_MAX) {
645 			/* overlocked, zp_size can't change */
646 			woff = uio->uio_loffset = zp->z_phys->zp_size;
647 		} else {
648 			woff = uio->uio_loffset = rl->r_off;
649 		}
650 	} else {
651 		woff = uio->uio_loffset;
652 		/*
653 		 * Validate file offset
654 		 */
655 		if (woff < 0) {
656 			ZFS_EXIT(zfsvfs);
657 			return (EINVAL);
658 		}
659 
660 		/*
661 		 * If we need to grow the block size then zfs_range_lock()
662 		 * will lock a wider range than we request here.
663 		 * Later after growing the block size we reduce the range.
664 		 */
665 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
666 	}
667 
668 	if (woff >= limit) {
669 		zfs_range_unlock(rl);
670 		ZFS_EXIT(zfsvfs);
671 		return (EFBIG);
672 	}
673 
674 	if ((woff + n) > limit || woff > (limit - n))
675 		n = limit - woff;
676 
677 	/*
678 	 * Check for mandatory locks
679 	 */
680 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) &&
681 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
682 		zfs_range_unlock(rl);
683 		ZFS_EXIT(zfsvfs);
684 		return (error);
685 	}
686 	end_size = MAX(zp->z_phys->zp_size, woff + n);
687 
688 	/*
689 	 * Write the file in reasonable size chunks.  Each chunk is written
690 	 * in a separate transaction; this keeps the intent log records small
691 	 * and allows us to do more fine-grained space accounting.
692 	 */
693 	while (n > 0) {
694 		/*
695 		 * Start a transaction.
696 		 */
697 		woff = uio->uio_loffset;
698 		tx = dmu_tx_create(zfsvfs->z_os);
699 		dmu_tx_hold_bonus(tx, zp->z_id);
700 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
701 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
702 		if (error) {
703 			if (error == ERESTART &&
704 			    zfsvfs->z_assign == TXG_NOWAIT) {
705 				dmu_tx_wait(tx);
706 				dmu_tx_abort(tx);
707 				continue;
708 			}
709 			dmu_tx_abort(tx);
710 			break;
711 		}
712 
713 		/*
714 		 * If zfs_range_lock() over-locked we grow the blocksize
715 		 * and then reduce the lock range.  This will only happen
716 		 * on the first iteration since zfs_range_reduce() will
717 		 * shrink down r_len to the appropriate size.
718 		 */
719 		if (rl->r_len == UINT64_MAX) {
720 			uint64_t new_blksz;
721 
722 			if (zp->z_blksz > max_blksz) {
723 				ASSERT(!ISP2(zp->z_blksz));
724 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
725 			} else {
726 				new_blksz = MIN(end_size, max_blksz);
727 			}
728 			zfs_grow_blocksize(zp, new_blksz, tx);
729 			zfs_range_reduce(rl, woff, n);
730 		}
731 
732 		/*
733 		 * XXX - should we really limit each write to z_max_blksz?
734 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
735 		 */
736 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
737 		rw_enter(&zp->z_map_lock, RW_READER);
738 
739 		tx_bytes = uio->uio_resid;
740 		if (vn_has_cached_data(vp)) {
741 			rw_exit(&zp->z_map_lock);
742 			error = mappedwrite(vp, nbytes, uio, tx);
743 		} else {
744 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
745 			    uio, nbytes, tx);
746 			rw_exit(&zp->z_map_lock);
747 		}
748 		tx_bytes -= uio->uio_resid;
749 
750 		/*
751 		 * If we made no progress, we're done.  If we made even
752 		 * partial progress, update the znode and ZIL accordingly.
753 		 */
754 		if (tx_bytes == 0) {
755 			dmu_tx_commit(tx);
756 			ASSERT(error != 0);
757 			break;
758 		}
759 
760 		/*
761 		 * Clear Set-UID/Set-GID bits on successful write if not
762 		 * privileged and at least one of the excute bits is set.
763 		 *
764 		 * It would be nice to to this after all writes have
765 		 * been done, but that would still expose the ISUID/ISGID
766 		 * to another app after the partial write is committed.
767 		 *
768 		 * Note: we don't call zfs_fuid_map_id() here because
769 		 * user 0 is not an ephemeral uid.
770 		 */
771 		mutex_enter(&zp->z_acl_lock);
772 		if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) |
773 		    (S_IXUSR >> 6))) != 0 &&
774 		    (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 &&
775 		    secpolicy_vnode_setid_retain(cr,
776 		    (zp->z_phys->zp_mode & S_ISUID) != 0 &&
777 		    zp->z_phys->zp_uid == 0) != 0) {
778 			zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID);
779 		}
780 		mutex_exit(&zp->z_acl_lock);
781 
782 		/*
783 		 * Update time stamp.  NOTE: This marks the bonus buffer as
784 		 * dirty, so we don't have to do it again for zp_size.
785 		 */
786 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
787 
788 		/*
789 		 * Update the file size (zp_size) if it has changed;
790 		 * account for possible concurrent updates.
791 		 */
792 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset)
793 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
794 			    uio->uio_loffset);
795 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
796 		dmu_tx_commit(tx);
797 
798 		if (error != 0)
799 			break;
800 		ASSERT(tx_bytes == nbytes);
801 		n -= nbytes;
802 	}
803 
804 	zfs_range_unlock(rl);
805 
806 	/*
807 	 * If we're in replay mode, or we made no progress, return error.
808 	 * Otherwise, it's at least a partial write, so it's successful.
809 	 */
810 	if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) {
811 		ZFS_EXIT(zfsvfs);
812 		return (error);
813 	}
814 
815 	if (ioflag & (FSYNC | FDSYNC))
816 		zil_commit(zilog, zp->z_last_itx, zp->z_id);
817 
818 	ZFS_EXIT(zfsvfs);
819 	return (0);
820 }
821 
822 void
823 zfs_get_done(dmu_buf_t *db, void *vzgd)
824 {
825 	zgd_t *zgd = (zgd_t *)vzgd;
826 	rl_t *rl = zgd->zgd_rl;
827 	vnode_t *vp = ZTOV(rl->r_zp);
828 
829 	dmu_buf_rele(db, vzgd);
830 	zfs_range_unlock(rl);
831 	VN_RELE(vp);
832 	zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
833 	kmem_free(zgd, sizeof (zgd_t));
834 }
835 
836 /*
837  * Get data to generate a TX_WRITE intent log record.
838  */
839 int
840 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
841 {
842 	zfsvfs_t *zfsvfs = arg;
843 	objset_t *os = zfsvfs->z_os;
844 	znode_t *zp;
845 	uint64_t off = lr->lr_offset;
846 	dmu_buf_t *db;
847 	rl_t *rl;
848 	zgd_t *zgd;
849 	int dlen = lr->lr_length;		/* length of user data */
850 	int error = 0;
851 
852 	ASSERT(zio);
853 	ASSERT(dlen != 0);
854 
855 	/*
856 	 * Nothing to do if the file has been removed
857 	 */
858 	if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0)
859 		return (ENOENT);
860 	if (zp->z_unlinked) {
861 		VN_RELE(ZTOV(zp));
862 		return (ENOENT);
863 	}
864 
865 	/*
866 	 * Write records come in two flavors: immediate and indirect.
867 	 * For small writes it's cheaper to store the data with the
868 	 * log record (immediate); for large writes it's cheaper to
869 	 * sync the data and get a pointer to it (indirect) so that
870 	 * we don't have to write the data twice.
871 	 */
872 	if (buf != NULL) { /* immediate write */
873 		rl = zfs_range_lock(zp, off, dlen, RL_READER);
874 		/* test for truncation needs to be done while range locked */
875 		if (off >= zp->z_phys->zp_size) {
876 			error = ENOENT;
877 			goto out;
878 		}
879 		VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf));
880 	} else { /* indirect write */
881 		uint64_t boff; /* block starting offset */
882 
883 		/*
884 		 * Have to lock the whole block to ensure when it's
885 		 * written out and it's checksum is being calculated
886 		 * that no one can change the data. We need to re-check
887 		 * blocksize after we get the lock in case it's changed!
888 		 */
889 		for (;;) {
890 			if (ISP2(zp->z_blksz)) {
891 				boff = P2ALIGN_TYPED(off, zp->z_blksz,
892 				    uint64_t);
893 			} else {
894 				boff = 0;
895 			}
896 			dlen = zp->z_blksz;
897 			rl = zfs_range_lock(zp, boff, dlen, RL_READER);
898 			if (zp->z_blksz == dlen)
899 				break;
900 			zfs_range_unlock(rl);
901 		}
902 		/* test for truncation needs to be done while range locked */
903 		if (off >= zp->z_phys->zp_size) {
904 			error = ENOENT;
905 			goto out;
906 		}
907 		zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
908 		zgd->zgd_rl = rl;
909 		zgd->zgd_zilog = zfsvfs->z_log;
910 		zgd->zgd_bp = &lr->lr_blkptr;
911 		VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db));
912 		ASSERT(boff == db->db_offset);
913 		lr->lr_blkoff = off - boff;
914 		error = dmu_sync(zio, db, &lr->lr_blkptr,
915 		    lr->lr_common.lrc_txg, zfs_get_done, zgd);
916 		ASSERT((error && error != EINPROGRESS) ||
917 		    lr->lr_length <= zp->z_blksz);
918 		if (error == 0)
919 			zil_add_block(zfsvfs->z_log, &lr->lr_blkptr);
920 		/*
921 		 * If we get EINPROGRESS, then we need to wait for a
922 		 * write IO initiated by dmu_sync() to complete before
923 		 * we can release this dbuf.  We will finish everything
924 		 * up in the zfs_get_done() callback.
925 		 */
926 		if (error == EINPROGRESS)
927 			return (0);
928 		dmu_buf_rele(db, zgd);
929 		kmem_free(zgd, sizeof (zgd_t));
930 	}
931 out:
932 	zfs_range_unlock(rl);
933 	VN_RELE(ZTOV(zp));
934 	return (error);
935 }
936 
937 /*ARGSUSED*/
938 static int
939 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
940     caller_context_t *ct)
941 {
942 	znode_t *zp = VTOZ(vp);
943 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
944 	int error;
945 
946 	ZFS_ENTER(zfsvfs);
947 	ZFS_VERIFY_ZP(zp);
948 
949 	if (flag & V_ACE_MASK)
950 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
951 	else
952 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
953 
954 	ZFS_EXIT(zfsvfs);
955 	return (error);
956 }
957 
958 /*
959  * Lookup an entry in a directory, or an extended attribute directory.
960  * If it exists, return a held vnode reference for it.
961  *
962  *	IN:	dvp	- vnode of directory to search.
963  *		nm	- name of entry to lookup.
964  *		pnp	- full pathname to lookup [UNUSED].
965  *		flags	- LOOKUP_XATTR set if looking for an attribute.
966  *		rdir	- root directory vnode [UNUSED].
967  *		cr	- credentials of caller.
968  *		ct	- caller context
969  *		direntflags - directory lookup flags
970  *		realpnp - returned pathname.
971  *
972  *	OUT:	vpp	- vnode of located entry, NULL if not found.
973  *
974  *	RETURN:	0 if success
975  *		error code if failure
976  *
977  * Timestamps:
978  *	NA
979  */
980 /* ARGSUSED */
981 static int
982 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
983     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
984     int *direntflags, pathname_t *realpnp)
985 {
986 	znode_t *zdp = VTOZ(dvp);
987 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
988 	int	error;
989 
990 	ZFS_ENTER(zfsvfs);
991 	ZFS_VERIFY_ZP(zdp);
992 
993 	*vpp = NULL;
994 
995 	if (flags & LOOKUP_XATTR) {
996 		/*
997 		 * If the xattr property is off, refuse the lookup request.
998 		 */
999 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1000 			ZFS_EXIT(zfsvfs);
1001 			return (EINVAL);
1002 		}
1003 
1004 		/*
1005 		 * We don't allow recursive attributes..
1006 		 * Maybe someday we will.
1007 		 */
1008 		if (zdp->z_phys->zp_flags & ZFS_XATTR) {
1009 			ZFS_EXIT(zfsvfs);
1010 			return (EINVAL);
1011 		}
1012 
1013 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1014 			ZFS_EXIT(zfsvfs);
1015 			return (error);
1016 		}
1017 
1018 		/*
1019 		 * Do we have permission to get into attribute directory?
1020 		 */
1021 
1022 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1023 		    B_FALSE, cr)) {
1024 			VN_RELE(*vpp);
1025 			*vpp = NULL;
1026 		}
1027 
1028 		ZFS_EXIT(zfsvfs);
1029 		return (error);
1030 	}
1031 
1032 	if (dvp->v_type != VDIR) {
1033 		ZFS_EXIT(zfsvfs);
1034 		return (ENOTDIR);
1035 	}
1036 
1037 	/*
1038 	 * Check accessibility of directory.
1039 	 */
1040 
1041 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1042 		ZFS_EXIT(zfsvfs);
1043 		return (error);
1044 	}
1045 
1046 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1047 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1048 		ZFS_EXIT(zfsvfs);
1049 		return (EILSEQ);
1050 	}
1051 
1052 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1053 	if (error == 0) {
1054 		/*
1055 		 * Convert device special files
1056 		 */
1057 		if (IS_DEVVP(*vpp)) {
1058 			vnode_t	*svp;
1059 
1060 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1061 			VN_RELE(*vpp);
1062 			if (svp == NULL)
1063 				error = ENOSYS;
1064 			else
1065 				*vpp = svp;
1066 		}
1067 	}
1068 
1069 	ZFS_EXIT(zfsvfs);
1070 	return (error);
1071 }
1072 
1073 /*
1074  * Attempt to create a new entry in a directory.  If the entry
1075  * already exists, truncate the file if permissible, else return
1076  * an error.  Return the vp of the created or trunc'd file.
1077  *
1078  *	IN:	dvp	- vnode of directory to put new file entry in.
1079  *		name	- name of new file entry.
1080  *		vap	- attributes of new file.
1081  *		excl	- flag indicating exclusive or non-exclusive mode.
1082  *		mode	- mode to open file with.
1083  *		cr	- credentials of caller.
1084  *		flag	- large file flag [UNUSED].
1085  *		ct	- caller context
1086  *		vsecp 	- ACL to be set
1087  *
1088  *	OUT:	vpp	- vnode of created or trunc'd entry.
1089  *
1090  *	RETURN:	0 if success
1091  *		error code if failure
1092  *
1093  * Timestamps:
1094  *	dvp - ctime|mtime updated if new entry created
1095  *	 vp - ctime|mtime always, atime if new
1096  */
1097 
1098 /* ARGSUSED */
1099 static int
1100 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1101     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1102     vsecattr_t *vsecp)
1103 {
1104 	znode_t		*zp, *dzp = VTOZ(dvp);
1105 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1106 	zilog_t		*zilog;
1107 	objset_t	*os;
1108 	zfs_dirlock_t	*dl;
1109 	dmu_tx_t	*tx;
1110 	int		error;
1111 	zfs_acl_t	*aclp = NULL;
1112 	zfs_fuid_info_t *fuidp = NULL;
1113 	ksid_t		*ksid;
1114 	uid_t		uid;
1115 	gid_t		gid = crgetgid(cr);
1116 
1117 	/*
1118 	 * If we have an ephemeral id, ACL, or XVATTR then
1119 	 * make sure file system is at proper version
1120 	 */
1121 
1122 	ksid = crgetsid(cr, KSID_OWNER);
1123 	if (ksid)
1124 		uid = ksid_getid(ksid);
1125 	else
1126 		uid = crgetuid(cr);
1127 
1128 	if (zfsvfs->z_use_fuids == B_FALSE &&
1129 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1130 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1131 		return (EINVAL);
1132 
1133 	ZFS_ENTER(zfsvfs);
1134 	ZFS_VERIFY_ZP(dzp);
1135 	os = zfsvfs->z_os;
1136 	zilog = zfsvfs->z_log;
1137 
1138 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1139 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1140 		ZFS_EXIT(zfsvfs);
1141 		return (EILSEQ);
1142 	}
1143 
1144 	if (vap->va_mask & AT_XVATTR) {
1145 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1146 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1147 			ZFS_EXIT(zfsvfs);
1148 			return (error);
1149 		}
1150 	}
1151 top:
1152 	*vpp = NULL;
1153 
1154 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1155 		vap->va_mode &= ~VSVTX;
1156 
1157 	if (*name == '\0') {
1158 		/*
1159 		 * Null component name refers to the directory itself.
1160 		 */
1161 		VN_HOLD(dvp);
1162 		zp = dzp;
1163 		dl = NULL;
1164 		error = 0;
1165 	} else {
1166 		/* possible VN_HOLD(zp) */
1167 		int zflg = 0;
1168 
1169 		if (flag & FIGNORECASE)
1170 			zflg |= ZCILOOK;
1171 
1172 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1173 		    NULL, NULL);
1174 		if (error) {
1175 			if (strcmp(name, "..") == 0)
1176 				error = EISDIR;
1177 			ZFS_EXIT(zfsvfs);
1178 			if (aclp)
1179 				zfs_acl_free(aclp);
1180 			return (error);
1181 		}
1182 	}
1183 	if (vsecp && aclp == NULL) {
1184 		error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp);
1185 		if (error) {
1186 			ZFS_EXIT(zfsvfs);
1187 			if (dl)
1188 				zfs_dirent_unlock(dl);
1189 			return (error);
1190 		}
1191 	}
1192 
1193 	if (zp == NULL) {
1194 		uint64_t txtype;
1195 
1196 		/*
1197 		 * Create a new file object and update the directory
1198 		 * to reference it.
1199 		 */
1200 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1201 			goto out;
1202 		}
1203 
1204 		/*
1205 		 * We only support the creation of regular files in
1206 		 * extended attribute directories.
1207 		 */
1208 		if ((dzp->z_phys->zp_flags & ZFS_XATTR) &&
1209 		    (vap->va_type != VREG)) {
1210 			error = EINVAL;
1211 			goto out;
1212 		}
1213 
1214 		tx = dmu_tx_create(os);
1215 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1216 		if ((aclp && aclp->z_has_fuids) || IS_EPHEMERAL(uid) ||
1217 		    IS_EPHEMERAL(gid)) {
1218 			if (zfsvfs->z_fuid_obj == 0) {
1219 				dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1220 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1221 				    FUID_SIZE_ESTIMATE(zfsvfs));
1222 				dmu_tx_hold_zap(tx, MASTER_NODE_OBJ,
1223 				    FALSE, NULL);
1224 			} else {
1225 				dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
1226 				dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
1227 				    FUID_SIZE_ESTIMATE(zfsvfs));
1228 			}
1229 		}
1230 		dmu_tx_hold_bonus(tx, dzp->z_id);
1231 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1232 		if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) {
1233 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1234 			    0, SPA_MAXBLOCKSIZE);
1235 		}
1236 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
1237 		if (error) {
1238 			zfs_dirent_unlock(dl);
1239 			if (error == ERESTART &&
1240 			    zfsvfs->z_assign == TXG_NOWAIT) {
1241 				dmu_tx_wait(tx);
1242 				dmu_tx_abort(tx);
1243 				goto top;
1244 			}
1245 			dmu_tx_abort(tx);
1246 			ZFS_EXIT(zfsvfs);
1247 			if (aclp)
1248 				zfs_acl_free(aclp);
1249 			return (error);
1250 		}
1251 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp);
1252 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1253 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1254 		if (flag & FIGNORECASE)
1255 			txtype |= TX_CI;
1256 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1257 		    vsecp, fuidp, vap);
1258 		if (fuidp)
1259 			zfs_fuid_info_free(fuidp);
1260 		dmu_tx_commit(tx);
1261 	} else {
1262 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1263 
1264 		/*
1265 		 * A directory entry already exists for this name.
1266 		 */
1267 		/*
1268 		 * Can't truncate an existing file if in exclusive mode.
1269 		 */
1270 		if (excl == EXCL) {
1271 			error = EEXIST;
1272 			goto out;
1273 		}
1274 		/*
1275 		 * Can't open a directory for writing.
1276 		 */
1277 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1278 			error = EISDIR;
1279 			goto out;
1280 		}
1281 		/*
1282 		 * Verify requested access to file.
1283 		 */
1284 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1285 			goto out;
1286 		}
1287 
1288 		mutex_enter(&dzp->z_lock);
1289 		dzp->z_seq++;
1290 		mutex_exit(&dzp->z_lock);
1291 
1292 		/*
1293 		 * Truncate regular files if requested.
1294 		 */
1295 		if ((ZTOV(zp)->v_type == VREG) &&
1296 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1297 			/* we can't hold any locks when calling zfs_freesp() */
1298 			zfs_dirent_unlock(dl);
1299 			dl = NULL;
1300 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1301 			if (error == 0) {
1302 				vnevent_create(ZTOV(zp), ct);
1303 			}
1304 		}
1305 	}
1306 out:
1307 
1308 	if (dl)
1309 		zfs_dirent_unlock(dl);
1310 
1311 	if (error) {
1312 		if (zp)
1313 			VN_RELE(ZTOV(zp));
1314 	} else {
1315 		*vpp = ZTOV(zp);
1316 		/*
1317 		 * If vnode is for a device return a specfs vnode instead.
1318 		 */
1319 		if (IS_DEVVP(*vpp)) {
1320 			struct vnode *svp;
1321 
1322 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1323 			VN_RELE(*vpp);
1324 			if (svp == NULL) {
1325 				error = ENOSYS;
1326 			}
1327 			*vpp = svp;
1328 		}
1329 	}
1330 	if (aclp)
1331 		zfs_acl_free(aclp);
1332 
1333 	ZFS_EXIT(zfsvfs);
1334 	return (error);
1335 }
1336 
1337 /*
1338  * Remove an entry from a directory.
1339  *
1340  *	IN:	dvp	- vnode of directory to remove entry from.
1341  *		name	- name of entry to remove.
1342  *		cr	- credentials of caller.
1343  *		ct	- caller context
1344  *		flags	- case flags
1345  *
1346  *	RETURN:	0 if success
1347  *		error code if failure
1348  *
1349  * Timestamps:
1350  *	dvp - ctime|mtime
1351  *	 vp - ctime (if nlink > 0)
1352  */
1353 /*ARGSUSED*/
1354 static int
1355 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1356     int flags)
1357 {
1358 	znode_t		*zp, *dzp = VTOZ(dvp);
1359 	znode_t		*xzp = NULL;
1360 	vnode_t		*vp;
1361 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1362 	zilog_t		*zilog;
1363 	uint64_t	acl_obj, xattr_obj;
1364 	zfs_dirlock_t	*dl;
1365 	dmu_tx_t	*tx;
1366 	boolean_t	may_delete_now, delete_now = FALSE;
1367 	boolean_t	unlinked, toobig = FALSE;
1368 	uint64_t	txtype;
1369 	pathname_t	*realnmp = NULL;
1370 	pathname_t	realnm;
1371 	int		error;
1372 	int		zflg = ZEXISTS;
1373 
1374 	ZFS_ENTER(zfsvfs);
1375 	ZFS_VERIFY_ZP(dzp);
1376 	zilog = zfsvfs->z_log;
1377 
1378 	if (flags & FIGNORECASE) {
1379 		zflg |= ZCILOOK;
1380 		pn_alloc(&realnm);
1381 		realnmp = &realnm;
1382 	}
1383 
1384 top:
1385 	/*
1386 	 * Attempt to lock directory; fail if entry doesn't exist.
1387 	 */
1388 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1389 	    NULL, realnmp)) {
1390 		if (realnmp)
1391 			pn_free(realnmp);
1392 		ZFS_EXIT(zfsvfs);
1393 		return (error);
1394 	}
1395 
1396 	vp = ZTOV(zp);
1397 
1398 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1399 		goto out;
1400 	}
1401 
1402 	/*
1403 	 * Need to use rmdir for removing directories.
1404 	 */
1405 	if (vp->v_type == VDIR) {
1406 		error = EPERM;
1407 		goto out;
1408 	}
1409 
1410 	vnevent_remove(vp, dvp, name, ct);
1411 
1412 	if (realnmp)
1413 		dnlc_remove(dvp, realnmp->pn_buf);
1414 	else
1415 		dnlc_remove(dvp, name);
1416 
1417 	mutex_enter(&vp->v_lock);
1418 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1419 	mutex_exit(&vp->v_lock);
1420 
1421 	/*
1422 	 * We may delete the znode now, or we may put it in the unlinked set;
1423 	 * it depends on whether we're the last link, and on whether there are
1424 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1425 	 * allow for either case.
1426 	 */
1427 	tx = dmu_tx_create(zfsvfs->z_os);
1428 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1429 	dmu_tx_hold_bonus(tx, zp->z_id);
1430 	if (may_delete_now) {
1431 		toobig =
1432 		    zp->z_phys->zp_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1433 		/* if the file is too big, only hold_free a token amount */
1434 		dmu_tx_hold_free(tx, zp->z_id, 0,
1435 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1436 	}
1437 
1438 	/* are there any extended attributes? */
1439 	if ((xattr_obj = zp->z_phys->zp_xattr) != 0) {
1440 		/* XXX - do we need this if we are deleting? */
1441 		dmu_tx_hold_bonus(tx, xattr_obj);
1442 	}
1443 
1444 	/* are there any additional acls */
1445 	if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 &&
1446 	    may_delete_now)
1447 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1448 
1449 	/* charge as an update -- would be nice not to charge at all */
1450 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1451 
1452 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1453 	if (error) {
1454 		zfs_dirent_unlock(dl);
1455 		VN_RELE(vp);
1456 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1457 			dmu_tx_wait(tx);
1458 			dmu_tx_abort(tx);
1459 			goto top;
1460 		}
1461 		if (realnmp)
1462 			pn_free(realnmp);
1463 		dmu_tx_abort(tx);
1464 		ZFS_EXIT(zfsvfs);
1465 		return (error);
1466 	}
1467 
1468 	/*
1469 	 * Remove the directory entry.
1470 	 */
1471 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1472 
1473 	if (error) {
1474 		dmu_tx_commit(tx);
1475 		goto out;
1476 	}
1477 
1478 	if (unlinked) {
1479 		mutex_enter(&vp->v_lock);
1480 		delete_now = may_delete_now && !toobig &&
1481 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1482 		    zp->z_phys->zp_xattr == xattr_obj &&
1483 		    zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj;
1484 		mutex_exit(&vp->v_lock);
1485 	}
1486 
1487 	if (delete_now) {
1488 		if (zp->z_phys->zp_xattr) {
1489 			error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
1490 			ASSERT3U(error, ==, 0);
1491 			ASSERT3U(xzp->z_phys->zp_links, ==, 2);
1492 			dmu_buf_will_dirty(xzp->z_dbuf, tx);
1493 			mutex_enter(&xzp->z_lock);
1494 			xzp->z_unlinked = 1;
1495 			xzp->z_phys->zp_links = 0;
1496 			mutex_exit(&xzp->z_lock);
1497 			zfs_unlinked_add(xzp, tx);
1498 			zp->z_phys->zp_xattr = 0; /* probably unnecessary */
1499 		}
1500 		mutex_enter(&zp->z_lock);
1501 		mutex_enter(&vp->v_lock);
1502 		vp->v_count--;
1503 		ASSERT3U(vp->v_count, ==, 0);
1504 		mutex_exit(&vp->v_lock);
1505 		mutex_exit(&zp->z_lock);
1506 		zfs_znode_delete(zp, tx);
1507 	} else if (unlinked) {
1508 		zfs_unlinked_add(zp, tx);
1509 	}
1510 
1511 	txtype = TX_REMOVE;
1512 	if (flags & FIGNORECASE)
1513 		txtype |= TX_CI;
1514 	zfs_log_remove(zilog, tx, txtype, dzp, name);
1515 
1516 	dmu_tx_commit(tx);
1517 out:
1518 	if (realnmp)
1519 		pn_free(realnmp);
1520 
1521 	zfs_dirent_unlock(dl);
1522 
1523 	if (!delete_now) {
1524 		VN_RELE(vp);
1525 	} else if (xzp) {
1526 		/* this rele is delayed to prevent nesting transactions */
1527 		VN_RELE(ZTOV(xzp));
1528 	}
1529 
1530 	ZFS_EXIT(zfsvfs);
1531 	return (error);
1532 }
1533 
1534 /*
1535  * Create a new directory and insert it into dvp using the name
1536  * provided.  Return a pointer to the inserted directory.
1537  *
1538  *	IN:	dvp	- vnode of directory to add subdir to.
1539  *		dirname	- name of new directory.
1540  *		vap	- attributes of new directory.
1541  *		cr	- credentials of caller.
1542  *		ct	- caller context
1543  *		vsecp	- ACL to be set
1544  *
1545  *	OUT:	vpp	- vnode of created directory.
1546  *
1547  *	RETURN:	0 if success
1548  *		error code if failure
1549  *
1550  * Timestamps:
1551  *	dvp - ctime|mtime updated
1552  *	 vp - ctime|mtime|atime updated
1553  */
1554 /*ARGSUSED*/
1555 static int
1556 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1557     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1558 {
1559 	znode_t		*zp, *dzp = VTOZ(dvp);
1560 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1561 	zilog_t		*zilog;
1562 	zfs_dirlock_t	*dl;
1563 	uint64_t	txtype;
1564 	dmu_tx_t	*tx;
1565 	int		error;
1566 	zfs_acl_t	*aclp = NULL;
1567 	zfs_fuid_info_t	*fuidp = NULL;
1568 	int		zf = ZNEW;
1569 	ksid_t		*ksid;
1570 	uid_t		uid;
1571 	gid_t		gid = crgetgid(cr);
1572 
1573 	ASSERT(vap->va_type == VDIR);
1574 
1575 	/*
1576 	 * If we have an ephemeral id, ACL, or XVATTR then
1577 	 * make sure file system is at proper version
1578 	 */
1579 
1580 	ksid = crgetsid(cr, KSID_OWNER);
1581 	if (ksid)
1582 		uid = ksid_getid(ksid);
1583 	else
1584 		uid = crgetuid(cr);
1585 	if (zfsvfs->z_use_fuids == B_FALSE &&
1586 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1587 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1588 		return (EINVAL);
1589 
1590 	ZFS_ENTER(zfsvfs);
1591 	ZFS_VERIFY_ZP(dzp);
1592 	zilog = zfsvfs->z_log;
1593 
1594 	if (dzp->z_phys->zp_flags & ZFS_XATTR) {
1595 		ZFS_EXIT(zfsvfs);
1596 		return (EINVAL);
1597 	}
1598 
1599 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1600 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1601 		ZFS_EXIT(zfsvfs);
1602 		return (EILSEQ);
1603 	}
1604 	if (flags & FIGNORECASE)
1605 		zf |= ZCILOOK;
1606 
1607 	if (vap->va_mask & AT_XVATTR)
1608 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1609 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1610 			ZFS_EXIT(zfsvfs);
1611 			return (error);
1612 		}
1613 
1614 	/*
1615 	 * First make sure the new directory doesn't exist.
1616 	 */
1617 top:
1618 	*vpp = NULL;
1619 
1620 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1621 	    NULL, NULL)) {
1622 		ZFS_EXIT(zfsvfs);
1623 		return (error);
1624 	}
1625 
1626 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1627 		zfs_dirent_unlock(dl);
1628 		ZFS_EXIT(zfsvfs);
1629 		return (error);
1630 	}
1631 
1632 	if (vsecp && aclp == NULL) {
1633 		error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp);
1634 		if (error) {
1635 			zfs_dirent_unlock(dl);
1636 			ZFS_EXIT(zfsvfs);
1637 			return (error);
1638 		}
1639 	}
1640 	/*
1641 	 * Add a new entry to the directory.
1642 	 */
1643 	tx = dmu_tx_create(zfsvfs->z_os);
1644 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1645 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1646 	if ((aclp && aclp->z_has_fuids) || IS_EPHEMERAL(uid) ||
1647 	    IS_EPHEMERAL(gid)) {
1648 		if (zfsvfs->z_fuid_obj == 0) {
1649 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1650 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1651 			    FUID_SIZE_ESTIMATE(zfsvfs));
1652 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
1653 		} else {
1654 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
1655 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
1656 			    FUID_SIZE_ESTIMATE(zfsvfs));
1657 		}
1658 	}
1659 	if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp)
1660 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1661 		    0, SPA_MAXBLOCKSIZE);
1662 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1663 	if (error) {
1664 		zfs_dirent_unlock(dl);
1665 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1666 			dmu_tx_wait(tx);
1667 			dmu_tx_abort(tx);
1668 			goto top;
1669 		}
1670 		dmu_tx_abort(tx);
1671 		ZFS_EXIT(zfsvfs);
1672 		if (aclp)
1673 			zfs_acl_free(aclp);
1674 		return (error);
1675 	}
1676 
1677 	/*
1678 	 * Create new node.
1679 	 */
1680 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp);
1681 
1682 	if (aclp)
1683 		zfs_acl_free(aclp);
1684 
1685 	/*
1686 	 * Now put new name in parent dir.
1687 	 */
1688 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1689 
1690 	*vpp = ZTOV(zp);
1691 
1692 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1693 	if (flags & FIGNORECASE)
1694 		txtype |= TX_CI;
1695 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, fuidp, vap);
1696 
1697 	if (fuidp)
1698 		zfs_fuid_info_free(fuidp);
1699 	dmu_tx_commit(tx);
1700 
1701 	zfs_dirent_unlock(dl);
1702 
1703 	ZFS_EXIT(zfsvfs);
1704 	return (0);
1705 }
1706 
1707 /*
1708  * Remove a directory subdir entry.  If the current working
1709  * directory is the same as the subdir to be removed, the
1710  * remove will fail.
1711  *
1712  *	IN:	dvp	- vnode of directory to remove from.
1713  *		name	- name of directory to be removed.
1714  *		cwd	- vnode of current working directory.
1715  *		cr	- credentials of caller.
1716  *		ct	- caller context
1717  *		flags	- case flags
1718  *
1719  *	RETURN:	0 if success
1720  *		error code if failure
1721  *
1722  * Timestamps:
1723  *	dvp - ctime|mtime updated
1724  */
1725 /*ARGSUSED*/
1726 static int
1727 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1728     caller_context_t *ct, int flags)
1729 {
1730 	znode_t		*dzp = VTOZ(dvp);
1731 	znode_t		*zp;
1732 	vnode_t		*vp;
1733 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1734 	zilog_t		*zilog;
1735 	zfs_dirlock_t	*dl;
1736 	dmu_tx_t	*tx;
1737 	int		error;
1738 	int		zflg = ZEXISTS;
1739 
1740 	ZFS_ENTER(zfsvfs);
1741 	ZFS_VERIFY_ZP(dzp);
1742 	zilog = zfsvfs->z_log;
1743 
1744 	if (flags & FIGNORECASE)
1745 		zflg |= ZCILOOK;
1746 top:
1747 	zp = NULL;
1748 
1749 	/*
1750 	 * Attempt to lock directory; fail if entry doesn't exist.
1751 	 */
1752 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1753 	    NULL, NULL)) {
1754 		ZFS_EXIT(zfsvfs);
1755 		return (error);
1756 	}
1757 
1758 	vp = ZTOV(zp);
1759 
1760 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1761 		goto out;
1762 	}
1763 
1764 	if (vp->v_type != VDIR) {
1765 		error = ENOTDIR;
1766 		goto out;
1767 	}
1768 
1769 	if (vp == cwd) {
1770 		error = EINVAL;
1771 		goto out;
1772 	}
1773 
1774 	vnevent_rmdir(vp, dvp, name, ct);
1775 
1776 	/*
1777 	 * Grab a lock on the directory to make sure that noone is
1778 	 * trying to add (or lookup) entries while we are removing it.
1779 	 */
1780 	rw_enter(&zp->z_name_lock, RW_WRITER);
1781 
1782 	/*
1783 	 * Grab a lock on the parent pointer to make sure we play well
1784 	 * with the treewalk and directory rename code.
1785 	 */
1786 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1787 
1788 	tx = dmu_tx_create(zfsvfs->z_os);
1789 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1790 	dmu_tx_hold_bonus(tx, zp->z_id);
1791 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1792 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1793 	if (error) {
1794 		rw_exit(&zp->z_parent_lock);
1795 		rw_exit(&zp->z_name_lock);
1796 		zfs_dirent_unlock(dl);
1797 		VN_RELE(vp);
1798 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1799 			dmu_tx_wait(tx);
1800 			dmu_tx_abort(tx);
1801 			goto top;
1802 		}
1803 		dmu_tx_abort(tx);
1804 		ZFS_EXIT(zfsvfs);
1805 		return (error);
1806 	}
1807 
1808 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
1809 
1810 	if (error == 0) {
1811 		uint64_t txtype = TX_RMDIR;
1812 		if (flags & FIGNORECASE)
1813 			txtype |= TX_CI;
1814 		zfs_log_remove(zilog, tx, txtype, dzp, name);
1815 	}
1816 
1817 	dmu_tx_commit(tx);
1818 
1819 	rw_exit(&zp->z_parent_lock);
1820 	rw_exit(&zp->z_name_lock);
1821 out:
1822 	zfs_dirent_unlock(dl);
1823 
1824 	VN_RELE(vp);
1825 
1826 	ZFS_EXIT(zfsvfs);
1827 	return (error);
1828 }
1829 
1830 /*
1831  * Read as many directory entries as will fit into the provided
1832  * buffer from the given directory cursor position (specified in
1833  * the uio structure.
1834  *
1835  *	IN:	vp	- vnode of directory to read.
1836  *		uio	- structure supplying read location, range info,
1837  *			  and return buffer.
1838  *		cr	- credentials of caller.
1839  *		ct	- caller context
1840  *		flags	- case flags
1841  *
1842  *	OUT:	uio	- updated offset and range, buffer filled.
1843  *		eofp	- set to true if end-of-file detected.
1844  *
1845  *	RETURN:	0 if success
1846  *		error code if failure
1847  *
1848  * Timestamps:
1849  *	vp - atime updated
1850  *
1851  * Note that the low 4 bits of the cookie returned by zap is always zero.
1852  * This allows us to use the low range for "special" directory entries:
1853  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1854  * we use the offset 2 for the '.zfs' directory.
1855  */
1856 /* ARGSUSED */
1857 static int
1858 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
1859     caller_context_t *ct, int flags)
1860 {
1861 	znode_t		*zp = VTOZ(vp);
1862 	iovec_t		*iovp;
1863 	edirent_t	*eodp;
1864 	dirent64_t	*odp;
1865 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1866 	objset_t	*os;
1867 	caddr_t		outbuf;
1868 	size_t		bufsize;
1869 	zap_cursor_t	zc;
1870 	zap_attribute_t	zap;
1871 	uint_t		bytes_wanted;
1872 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1873 	int		local_eof;
1874 	int		outcount;
1875 	int		error;
1876 	uint8_t		prefetch;
1877 	boolean_t	check_sysattrs;
1878 
1879 	ZFS_ENTER(zfsvfs);
1880 	ZFS_VERIFY_ZP(zp);
1881 
1882 	/*
1883 	 * If we are not given an eof variable,
1884 	 * use a local one.
1885 	 */
1886 	if (eofp == NULL)
1887 		eofp = &local_eof;
1888 
1889 	/*
1890 	 * Check for valid iov_len.
1891 	 */
1892 	if (uio->uio_iov->iov_len <= 0) {
1893 		ZFS_EXIT(zfsvfs);
1894 		return (EINVAL);
1895 	}
1896 
1897 	/*
1898 	 * Quit if directory has been removed (posix)
1899 	 */
1900 	if ((*eofp = zp->z_unlinked) != 0) {
1901 		ZFS_EXIT(zfsvfs);
1902 		return (0);
1903 	}
1904 
1905 	error = 0;
1906 	os = zfsvfs->z_os;
1907 	offset = uio->uio_loffset;
1908 	prefetch = zp->z_zn_prefetch;
1909 
1910 	/*
1911 	 * Initialize the iterator cursor.
1912 	 */
1913 	if (offset <= 3) {
1914 		/*
1915 		 * Start iteration from the beginning of the directory.
1916 		 */
1917 		zap_cursor_init(&zc, os, zp->z_id);
1918 	} else {
1919 		/*
1920 		 * The offset is a serialized cursor.
1921 		 */
1922 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1923 	}
1924 
1925 	/*
1926 	 * Get space to change directory entries into fs independent format.
1927 	 */
1928 	iovp = uio->uio_iov;
1929 	bytes_wanted = iovp->iov_len;
1930 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
1931 		bufsize = bytes_wanted;
1932 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
1933 		odp = (struct dirent64 *)outbuf;
1934 	} else {
1935 		bufsize = bytes_wanted;
1936 		odp = (struct dirent64 *)iovp->iov_base;
1937 	}
1938 	eodp = (struct edirent *)odp;
1939 
1940 	/*
1941 	 * If this VFS supports the system attribute view interface; and
1942 	 * we're looking at an extended attribute directory; and we care
1943 	 * about normalization conflicts on this vfs; then we must check
1944 	 * for normalization conflicts with the sysattr name space.
1945 	 */
1946 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
1947 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
1948 	    (flags & V_RDDIR_ENTFLAGS);
1949 
1950 	/*
1951 	 * Transform to file-system independent format
1952 	 */
1953 	outcount = 0;
1954 	while (outcount < bytes_wanted) {
1955 		ino64_t objnum;
1956 		ushort_t reclen;
1957 		off64_t *next;
1958 
1959 		/*
1960 		 * Special case `.', `..', and `.zfs'.
1961 		 */
1962 		if (offset == 0) {
1963 			(void) strcpy(zap.za_name, ".");
1964 			zap.za_normalization_conflict = 0;
1965 			objnum = zp->z_id;
1966 		} else if (offset == 1) {
1967 			(void) strcpy(zap.za_name, "..");
1968 			zap.za_normalization_conflict = 0;
1969 			objnum = zp->z_phys->zp_parent;
1970 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
1971 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1972 			zap.za_normalization_conflict = 0;
1973 			objnum = ZFSCTL_INO_ROOT;
1974 		} else {
1975 			/*
1976 			 * Grab next entry.
1977 			 */
1978 			if (error = zap_cursor_retrieve(&zc, &zap)) {
1979 				if ((*eofp = (error == ENOENT)) != 0)
1980 					break;
1981 				else
1982 					goto update;
1983 			}
1984 
1985 			if (zap.za_integer_length != 8 ||
1986 			    zap.za_num_integers != 1) {
1987 				cmn_err(CE_WARN, "zap_readdir: bad directory "
1988 				    "entry, obj = %lld, offset = %lld\n",
1989 				    (u_longlong_t)zp->z_id,
1990 				    (u_longlong_t)offset);
1991 				error = ENXIO;
1992 				goto update;
1993 			}
1994 
1995 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
1996 			/*
1997 			 * MacOS X can extract the object type here such as:
1998 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
1999 			 */
2000 
2001 			if (check_sysattrs && !zap.za_normalization_conflict) {
2002 				zap.za_normalization_conflict =
2003 				    xattr_sysattr_casechk(zap.za_name);
2004 			}
2005 		}
2006 
2007 		if (flags & V_RDDIR_ENTFLAGS)
2008 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2009 		else
2010 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2011 
2012 		/*
2013 		 * Will this entry fit in the buffer?
2014 		 */
2015 		if (outcount + reclen > bufsize) {
2016 			/*
2017 			 * Did we manage to fit anything in the buffer?
2018 			 */
2019 			if (!outcount) {
2020 				error = EINVAL;
2021 				goto update;
2022 			}
2023 			break;
2024 		}
2025 		if (flags & V_RDDIR_ENTFLAGS) {
2026 			/*
2027 			 * Add extended flag entry:
2028 			 */
2029 			eodp->ed_ino = objnum;
2030 			eodp->ed_reclen = reclen;
2031 			/* NOTE: ed_off is the offset for the *next* entry */
2032 			next = &(eodp->ed_off);
2033 			eodp->ed_eflags = zap.za_normalization_conflict ?
2034 			    ED_CASE_CONFLICT : 0;
2035 			(void) strncpy(eodp->ed_name, zap.za_name,
2036 			    EDIRENT_NAMELEN(reclen));
2037 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2038 		} else {
2039 			/*
2040 			 * Add normal entry:
2041 			 */
2042 			odp->d_ino = objnum;
2043 			odp->d_reclen = reclen;
2044 			/* NOTE: d_off is the offset for the *next* entry */
2045 			next = &(odp->d_off);
2046 			(void) strncpy(odp->d_name, zap.za_name,
2047 			    DIRENT64_NAMELEN(reclen));
2048 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2049 		}
2050 		outcount += reclen;
2051 
2052 		ASSERT(outcount <= bufsize);
2053 
2054 		/* Prefetch znode */
2055 		if (prefetch)
2056 			dmu_prefetch(os, objnum, 0, 0);
2057 
2058 		/*
2059 		 * Move to the next entry, fill in the previous offset.
2060 		 */
2061 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2062 			zap_cursor_advance(&zc);
2063 			offset = zap_cursor_serialize(&zc);
2064 		} else {
2065 			offset += 1;
2066 		}
2067 		*next = offset;
2068 	}
2069 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2070 
2071 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2072 		iovp->iov_base += outcount;
2073 		iovp->iov_len -= outcount;
2074 		uio->uio_resid -= outcount;
2075 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2076 		/*
2077 		 * Reset the pointer.
2078 		 */
2079 		offset = uio->uio_loffset;
2080 	}
2081 
2082 update:
2083 	zap_cursor_fini(&zc);
2084 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2085 		kmem_free(outbuf, bufsize);
2086 
2087 	if (error == ENOENT)
2088 		error = 0;
2089 
2090 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2091 
2092 	uio->uio_loffset = offset;
2093 	ZFS_EXIT(zfsvfs);
2094 	return (error);
2095 }
2096 
2097 ulong_t zfs_fsync_sync_cnt = 4;
2098 
2099 static int
2100 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2101 {
2102 	znode_t	*zp = VTOZ(vp);
2103 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2104 
2105 	/*
2106 	 * Regardless of whether this is required for standards conformance,
2107 	 * this is the logical behavior when fsync() is called on a file with
2108 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2109 	 * going to be pushed out as part of the zil_commit().
2110 	 */
2111 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2112 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2113 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2114 
2115 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2116 
2117 	ZFS_ENTER(zfsvfs);
2118 	ZFS_VERIFY_ZP(zp);
2119 	zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
2120 	ZFS_EXIT(zfsvfs);
2121 	return (0);
2122 }
2123 
2124 
2125 /*
2126  * Get the requested file attributes and place them in the provided
2127  * vattr structure.
2128  *
2129  *	IN:	vp	- vnode of file.
2130  *		vap	- va_mask identifies requested attributes.
2131  *			  If AT_XVATTR set, then optional attrs are requested
2132  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2133  *		cr	- credentials of caller.
2134  *		ct	- caller context
2135  *
2136  *	OUT:	vap	- attribute values.
2137  *
2138  *	RETURN:	0 (always succeeds)
2139  */
2140 /* ARGSUSED */
2141 static int
2142 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2143     caller_context_t *ct)
2144 {
2145 	znode_t *zp = VTOZ(vp);
2146 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2147 	znode_phys_t *pzp;
2148 	int	error = 0;
2149 	uint64_t links;
2150 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2151 	xoptattr_t *xoap = NULL;
2152 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2153 
2154 	ZFS_ENTER(zfsvfs);
2155 	ZFS_VERIFY_ZP(zp);
2156 	pzp = zp->z_phys;
2157 
2158 	mutex_enter(&zp->z_lock);
2159 
2160 	/*
2161 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2162 	 * Also, if we are the owner don't bother, since owner should
2163 	 * always be allowed to read basic attributes of file.
2164 	 */
2165 	if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL) &&
2166 	    (pzp->zp_uid != crgetuid(cr))) {
2167 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2168 		    skipaclchk, cr)) {
2169 			mutex_exit(&zp->z_lock);
2170 			ZFS_EXIT(zfsvfs);
2171 			return (error);
2172 		}
2173 	}
2174 
2175 	/*
2176 	 * Return all attributes.  It's cheaper to provide the answer
2177 	 * than to determine whether we were asked the question.
2178 	 */
2179 
2180 	vap->va_type = vp->v_type;
2181 	vap->va_mode = pzp->zp_mode & MODEMASK;
2182 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2183 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2184 	vap->va_nodeid = zp->z_id;
2185 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2186 		links = pzp->zp_links + 1;
2187 	else
2188 		links = pzp->zp_links;
2189 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2190 	vap->va_size = pzp->zp_size;
2191 	vap->va_rdev = vp->v_rdev;
2192 	vap->va_seq = zp->z_seq;
2193 
2194 	/*
2195 	 * Add in any requested optional attributes and the create time.
2196 	 * Also set the corresponding bits in the returned attribute bitmap.
2197 	 */
2198 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2199 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2200 			xoap->xoa_archive =
2201 			    ((pzp->zp_flags & ZFS_ARCHIVE) != 0);
2202 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2203 		}
2204 
2205 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2206 			xoap->xoa_readonly =
2207 			    ((pzp->zp_flags & ZFS_READONLY) != 0);
2208 			XVA_SET_RTN(xvap, XAT_READONLY);
2209 		}
2210 
2211 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2212 			xoap->xoa_system =
2213 			    ((pzp->zp_flags & ZFS_SYSTEM) != 0);
2214 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2215 		}
2216 
2217 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2218 			xoap->xoa_hidden =
2219 			    ((pzp->zp_flags & ZFS_HIDDEN) != 0);
2220 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2221 		}
2222 
2223 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2224 			xoap->xoa_nounlink =
2225 			    ((pzp->zp_flags & ZFS_NOUNLINK) != 0);
2226 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2227 		}
2228 
2229 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2230 			xoap->xoa_immutable =
2231 			    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0);
2232 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2233 		}
2234 
2235 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2236 			xoap->xoa_appendonly =
2237 			    ((pzp->zp_flags & ZFS_APPENDONLY) != 0);
2238 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2239 		}
2240 
2241 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2242 			xoap->xoa_nodump =
2243 			    ((pzp->zp_flags & ZFS_NODUMP) != 0);
2244 			XVA_SET_RTN(xvap, XAT_NODUMP);
2245 		}
2246 
2247 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2248 			xoap->xoa_opaque =
2249 			    ((pzp->zp_flags & ZFS_OPAQUE) != 0);
2250 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2251 		}
2252 
2253 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2254 			xoap->xoa_av_quarantined =
2255 			    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0);
2256 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2257 		}
2258 
2259 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2260 			xoap->xoa_av_modified =
2261 			    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0);
2262 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2263 		}
2264 
2265 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2266 		    vp->v_type == VREG &&
2267 		    (pzp->zp_flags & ZFS_BONUS_SCANSTAMP)) {
2268 			size_t len;
2269 			dmu_object_info_t doi;
2270 
2271 			/*
2272 			 * Only VREG files have anti-virus scanstamps, so we
2273 			 * won't conflict with symlinks in the bonus buffer.
2274 			 */
2275 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2276 			len = sizeof (xoap->xoa_av_scanstamp) +
2277 			    sizeof (znode_phys_t);
2278 			if (len <= doi.doi_bonus_size) {
2279 				/*
2280 				 * pzp points to the start of the
2281 				 * znode_phys_t. pzp + 1 points to the
2282 				 * first byte after the znode_phys_t.
2283 				 */
2284 				(void) memcpy(xoap->xoa_av_scanstamp,
2285 				    pzp + 1,
2286 				    sizeof (xoap->xoa_av_scanstamp));
2287 				XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
2288 			}
2289 		}
2290 
2291 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2292 			ZFS_TIME_DECODE(&xoap->xoa_createtime, pzp->zp_crtime);
2293 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2294 		}
2295 	}
2296 
2297 	ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime);
2298 	ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime);
2299 	ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime);
2300 
2301 	mutex_exit(&zp->z_lock);
2302 
2303 	dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks);
2304 
2305 	if (zp->z_blksz == 0) {
2306 		/*
2307 		 * Block size hasn't been set; suggest maximal I/O transfers.
2308 		 */
2309 		vap->va_blksize = zfsvfs->z_max_blksz;
2310 	}
2311 
2312 	ZFS_EXIT(zfsvfs);
2313 	return (0);
2314 }
2315 
2316 /*
2317  * Set the file attributes to the values contained in the
2318  * vattr structure.
2319  *
2320  *	IN:	vp	- vnode of file to be modified.
2321  *		vap	- new attribute values.
2322  *			  If AT_XVATTR set, then optional attrs are being set
2323  *		flags	- ATTR_UTIME set if non-default time values provided.
2324  *			- ATTR_NOACLCHECK (CIFS context only).
2325  *		cr	- credentials of caller.
2326  *		ct	- caller context
2327  *
2328  *	RETURN:	0 if success
2329  *		error code if failure
2330  *
2331  * Timestamps:
2332  *	vp - ctime updated, mtime updated if size changed.
2333  */
2334 /* ARGSUSED */
2335 static int
2336 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2337 	caller_context_t *ct)
2338 {
2339 	znode_t		*zp = VTOZ(vp);
2340 	znode_phys_t	*pzp;
2341 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2342 	zilog_t		*zilog;
2343 	dmu_tx_t	*tx;
2344 	vattr_t		oldva;
2345 	xvattr_t	tmpxvattr;
2346 	uint_t		mask = vap->va_mask;
2347 	uint_t		saved_mask;
2348 	int		trim_mask = 0;
2349 	uint64_t	new_mode;
2350 	znode_t		*attrzp;
2351 	int		need_policy = FALSE;
2352 	int		err;
2353 	zfs_fuid_info_t *fuidp = NULL;
2354 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2355 	xoptattr_t	*xoap;
2356 	zfs_acl_t	*aclp = NULL;
2357 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2358 
2359 	if (mask == 0)
2360 		return (0);
2361 
2362 	if (mask & AT_NOSET)
2363 		return (EINVAL);
2364 
2365 	ZFS_ENTER(zfsvfs);
2366 	ZFS_VERIFY_ZP(zp);
2367 
2368 	pzp = zp->z_phys;
2369 	zilog = zfsvfs->z_log;
2370 
2371 	/*
2372 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2373 	 * that file system is at proper version level
2374 	 */
2375 
2376 	if (zfsvfs->z_use_fuids == B_FALSE &&
2377 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2378 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2379 	    (mask & AT_XVATTR))) {
2380 		ZFS_EXIT(zfsvfs);
2381 		return (EINVAL);
2382 	}
2383 
2384 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2385 		ZFS_EXIT(zfsvfs);
2386 		return (EISDIR);
2387 	}
2388 
2389 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2390 		ZFS_EXIT(zfsvfs);
2391 		return (EINVAL);
2392 	}
2393 
2394 	/*
2395 	 * If this is an xvattr_t, then get a pointer to the structure of
2396 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2397 	 */
2398 	xoap = xva_getxoptattr(xvap);
2399 
2400 	xva_init(&tmpxvattr);
2401 
2402 	/*
2403 	 * Immutable files can only alter immutable bit and atime
2404 	 */
2405 	if ((pzp->zp_flags & ZFS_IMMUTABLE) &&
2406 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2407 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2408 		ZFS_EXIT(zfsvfs);
2409 		return (EPERM);
2410 	}
2411 
2412 	if ((mask & AT_SIZE) && (pzp->zp_flags & ZFS_READONLY)) {
2413 		ZFS_EXIT(zfsvfs);
2414 		return (EPERM);
2415 	}
2416 
2417 	/*
2418 	 * Verify timestamps doesn't overflow 32 bits.
2419 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2420 	 * handle times greater than 2039.  This check should be removed
2421 	 * once large timestamps are fully supported.
2422 	 */
2423 	if (mask & (AT_ATIME | AT_MTIME)) {
2424 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2425 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2426 			ZFS_EXIT(zfsvfs);
2427 			return (EOVERFLOW);
2428 		}
2429 	}
2430 
2431 top:
2432 	attrzp = NULL;
2433 
2434 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2435 		ZFS_EXIT(zfsvfs);
2436 		return (EROFS);
2437 	}
2438 
2439 	/*
2440 	 * First validate permissions
2441 	 */
2442 
2443 	if (mask & AT_SIZE) {
2444 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2445 		if (err) {
2446 			ZFS_EXIT(zfsvfs);
2447 			return (err);
2448 		}
2449 		/*
2450 		 * XXX - Note, we are not providing any open
2451 		 * mode flags here (like FNDELAY), so we may
2452 		 * block if there are locks present... this
2453 		 * should be addressed in openat().
2454 		 */
2455 		/* XXX - would it be OK to generate a log record here? */
2456 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2457 		if (err) {
2458 			ZFS_EXIT(zfsvfs);
2459 			return (err);
2460 		}
2461 	}
2462 
2463 	if (mask & (AT_ATIME|AT_MTIME) ||
2464 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2465 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2466 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2467 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2468 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM))))
2469 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2470 		    skipaclchk, cr);
2471 
2472 	if (mask & (AT_UID|AT_GID)) {
2473 		int	idmask = (mask & (AT_UID|AT_GID));
2474 		int	take_owner;
2475 		int	take_group;
2476 
2477 		/*
2478 		 * NOTE: even if a new mode is being set,
2479 		 * we may clear S_ISUID/S_ISGID bits.
2480 		 */
2481 
2482 		if (!(mask & AT_MODE))
2483 			vap->va_mode = pzp->zp_mode;
2484 
2485 		/*
2486 		 * Take ownership or chgrp to group we are a member of
2487 		 */
2488 
2489 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2490 		take_group = (mask & AT_GID) &&
2491 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2492 
2493 		/*
2494 		 * If both AT_UID and AT_GID are set then take_owner and
2495 		 * take_group must both be set in order to allow taking
2496 		 * ownership.
2497 		 *
2498 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2499 		 *
2500 		 */
2501 
2502 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2503 		    ((idmask == AT_UID) && take_owner) ||
2504 		    ((idmask == AT_GID) && take_group)) {
2505 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2506 			    skipaclchk, cr) == 0) {
2507 				/*
2508 				 * Remove setuid/setgid for non-privileged users
2509 				 */
2510 				secpolicy_setid_clear(vap, cr);
2511 				trim_mask = (mask & (AT_UID|AT_GID));
2512 			} else {
2513 				need_policy =  TRUE;
2514 			}
2515 		} else {
2516 			need_policy =  TRUE;
2517 		}
2518 	}
2519 
2520 	mutex_enter(&zp->z_lock);
2521 	oldva.va_mode = pzp->zp_mode;
2522 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2523 	if (mask & AT_XVATTR) {
2524 		/*
2525 		 * Update xvattr mask to include only those attributes
2526 		 * that are actually changing.
2527 		 *
2528 		 * the bits will be restored prior to actually setting
2529 		 * the attributes so the caller thinks they were set.
2530 		 */
2531 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2532 			if (xoap->xoa_appendonly !=
2533 			    ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) {
2534 				need_policy = TRUE;
2535 			} else {
2536 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2537 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2538 			}
2539 		}
2540 
2541 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2542 			if (xoap->xoa_nounlink !=
2543 			    ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) {
2544 				need_policy = TRUE;
2545 			} else {
2546 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2547 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2548 			}
2549 		}
2550 
2551 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2552 			if (xoap->xoa_immutable !=
2553 			    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) {
2554 				need_policy = TRUE;
2555 			} else {
2556 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2557 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2558 			}
2559 		}
2560 
2561 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2562 			if (xoap->xoa_nodump !=
2563 			    ((pzp->zp_flags & ZFS_NODUMP) != 0)) {
2564 				need_policy = TRUE;
2565 			} else {
2566 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2567 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2568 			}
2569 		}
2570 
2571 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2572 			if (xoap->xoa_av_modified !=
2573 			    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) {
2574 				need_policy = TRUE;
2575 			} else {
2576 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2577 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2578 			}
2579 		}
2580 
2581 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2582 			if ((vp->v_type != VREG &&
2583 			    xoap->xoa_av_quarantined) ||
2584 			    xoap->xoa_av_quarantined !=
2585 			    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)) {
2586 				need_policy = TRUE;
2587 			} else {
2588 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2589 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2590 			}
2591 		}
2592 
2593 		if (need_policy == FALSE &&
2594 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2595 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2596 			need_policy = TRUE;
2597 		}
2598 	}
2599 
2600 	mutex_exit(&zp->z_lock);
2601 
2602 	if (mask & AT_MODE) {
2603 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2604 			err = secpolicy_setid_setsticky_clear(vp, vap,
2605 			    &oldva, cr);
2606 			if (err) {
2607 				ZFS_EXIT(zfsvfs);
2608 				return (err);
2609 			}
2610 			trim_mask |= AT_MODE;
2611 		} else {
2612 			need_policy = TRUE;
2613 		}
2614 	}
2615 
2616 	if (need_policy) {
2617 		/*
2618 		 * If trim_mask is set then take ownership
2619 		 * has been granted or write_acl is present and user
2620 		 * has the ability to modify mode.  In that case remove
2621 		 * UID|GID and or MODE from mask so that
2622 		 * secpolicy_vnode_setattr() doesn't revoke it.
2623 		 */
2624 
2625 		if (trim_mask) {
2626 			saved_mask = vap->va_mask;
2627 			vap->va_mask &= ~trim_mask;
2628 		}
2629 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2630 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2631 		if (err) {
2632 			ZFS_EXIT(zfsvfs);
2633 			return (err);
2634 		}
2635 
2636 		if (trim_mask)
2637 			vap->va_mask |= saved_mask;
2638 	}
2639 
2640 	/*
2641 	 * secpolicy_vnode_setattr, or take ownership may have
2642 	 * changed va_mask
2643 	 */
2644 	mask = vap->va_mask;
2645 
2646 	tx = dmu_tx_create(zfsvfs->z_os);
2647 	dmu_tx_hold_bonus(tx, zp->z_id);
2648 	if (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2649 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid))) {
2650 		if (zfsvfs->z_fuid_obj == 0) {
2651 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
2652 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2653 			    FUID_SIZE_ESTIMATE(zfsvfs));
2654 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
2655 		} else {
2656 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
2657 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
2658 			    FUID_SIZE_ESTIMATE(zfsvfs));
2659 		}
2660 	}
2661 
2662 	if (mask & AT_MODE) {
2663 		uint64_t pmode = pzp->zp_mode;
2664 
2665 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2666 
2667 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) {
2668 			dmu_tx_abort(tx);
2669 			ZFS_EXIT(zfsvfs);
2670 			return (err);
2671 		}
2672 		if (pzp->zp_acl.z_acl_extern_obj) {
2673 			/* Are we upgrading ACL from old V0 format to new V1 */
2674 			if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2675 			    pzp->zp_acl.z_acl_version ==
2676 			    ZFS_ACL_VERSION_INITIAL) {
2677 				dmu_tx_hold_free(tx,
2678 				    pzp->zp_acl.z_acl_extern_obj, 0,
2679 				    DMU_OBJECT_END);
2680 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2681 				    0, aclp->z_acl_bytes);
2682 			} else {
2683 				dmu_tx_hold_write(tx,
2684 				    pzp->zp_acl.z_acl_extern_obj, 0,
2685 				    aclp->z_acl_bytes);
2686 			}
2687 		} else if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2688 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2689 			    0, aclp->z_acl_bytes);
2690 		}
2691 	}
2692 
2693 	if ((mask & (AT_UID | AT_GID)) && pzp->zp_xattr != 0) {
2694 		err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp);
2695 		if (err) {
2696 			dmu_tx_abort(tx);
2697 			ZFS_EXIT(zfsvfs);
2698 			if (aclp)
2699 				zfs_acl_free(aclp);
2700 			return (err);
2701 		}
2702 		dmu_tx_hold_bonus(tx, attrzp->z_id);
2703 	}
2704 
2705 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2706 	if (err) {
2707 		if (attrzp)
2708 			VN_RELE(ZTOV(attrzp));
2709 
2710 		if (aclp) {
2711 			zfs_acl_free(aclp);
2712 			aclp = NULL;
2713 		}
2714 
2715 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2716 			dmu_tx_wait(tx);
2717 			dmu_tx_abort(tx);
2718 			goto top;
2719 		}
2720 		dmu_tx_abort(tx);
2721 		ZFS_EXIT(zfsvfs);
2722 		return (err);
2723 	}
2724 
2725 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2726 
2727 	/*
2728 	 * Set each attribute requested.
2729 	 * We group settings according to the locks they need to acquire.
2730 	 *
2731 	 * Note: you cannot set ctime directly, although it will be
2732 	 * updated as a side-effect of calling this function.
2733 	 */
2734 
2735 	mutex_enter(&zp->z_lock);
2736 
2737 	if (mask & AT_MODE) {
2738 		mutex_enter(&zp->z_acl_lock);
2739 		zp->z_phys->zp_mode = new_mode;
2740 		err = zfs_aclset_common(zp, aclp, cr, &fuidp, tx);
2741 		ASSERT3U(err, ==, 0);
2742 		mutex_exit(&zp->z_acl_lock);
2743 	}
2744 
2745 	if (attrzp)
2746 		mutex_enter(&attrzp->z_lock);
2747 
2748 	if (mask & AT_UID) {
2749 		pzp->zp_uid = zfs_fuid_create(zfsvfs,
2750 		    vap->va_uid, cr, ZFS_OWNER, tx, &fuidp);
2751 		if (attrzp) {
2752 			attrzp->z_phys->zp_uid = zfs_fuid_create(zfsvfs,
2753 			    vap->va_uid,  cr, ZFS_OWNER, tx, &fuidp);
2754 		}
2755 	}
2756 
2757 	if (mask & AT_GID) {
2758 		pzp->zp_gid = zfs_fuid_create(zfsvfs, vap->va_gid,
2759 		    cr, ZFS_GROUP, tx, &fuidp);
2760 		if (attrzp)
2761 			attrzp->z_phys->zp_gid = zfs_fuid_create(zfsvfs,
2762 			    vap->va_gid, cr, ZFS_GROUP, tx, &fuidp);
2763 	}
2764 
2765 	if (aclp)
2766 		zfs_acl_free(aclp);
2767 
2768 	if (attrzp)
2769 		mutex_exit(&attrzp->z_lock);
2770 
2771 	if (mask & AT_ATIME)
2772 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2773 
2774 	if (mask & AT_MTIME)
2775 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2776 
2777 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2778 	if (mask & AT_SIZE)
2779 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2780 	else if (mask != 0)
2781 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2782 	/*
2783 	 * Do this after setting timestamps to prevent timestamp
2784 	 * update from toggling bit
2785 	 */
2786 
2787 	if (xoap && (mask & AT_XVATTR)) {
2788 
2789 		/*
2790 		 * restore trimmed off masks
2791 		 * so that return masks can be set for caller.
2792 		 */
2793 
2794 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
2795 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
2796 		}
2797 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
2798 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
2799 		}
2800 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
2801 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
2802 		}
2803 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
2804 			XVA_SET_REQ(xvap, XAT_NODUMP);
2805 		}
2806 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
2807 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
2808 		}
2809 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
2810 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
2811 		}
2812 
2813 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
2814 			size_t len;
2815 			dmu_object_info_t doi;
2816 
2817 			ASSERT(vp->v_type == VREG);
2818 
2819 			/* Grow the bonus buffer if necessary. */
2820 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2821 			len = sizeof (xoap->xoa_av_scanstamp) +
2822 			    sizeof (znode_phys_t);
2823 			if (len > doi.doi_bonus_size)
2824 				VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0);
2825 		}
2826 		zfs_xvattr_set(zp, xvap);
2827 	}
2828 
2829 	if (mask != 0)
2830 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2831 
2832 	if (fuidp)
2833 		zfs_fuid_info_free(fuidp);
2834 	mutex_exit(&zp->z_lock);
2835 
2836 	if (attrzp)
2837 		VN_RELE(ZTOV(attrzp));
2838 
2839 	dmu_tx_commit(tx);
2840 
2841 	ZFS_EXIT(zfsvfs);
2842 	return (err);
2843 }
2844 
2845 typedef struct zfs_zlock {
2846 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2847 	znode_t		*zl_znode;	/* znode we held */
2848 	struct zfs_zlock *zl_next;	/* next in list */
2849 } zfs_zlock_t;
2850 
2851 /*
2852  * Drop locks and release vnodes that were held by zfs_rename_lock().
2853  */
2854 static void
2855 zfs_rename_unlock(zfs_zlock_t **zlpp)
2856 {
2857 	zfs_zlock_t *zl;
2858 
2859 	while ((zl = *zlpp) != NULL) {
2860 		if (zl->zl_znode != NULL)
2861 			VN_RELE(ZTOV(zl->zl_znode));
2862 		rw_exit(zl->zl_rwlock);
2863 		*zlpp = zl->zl_next;
2864 		kmem_free(zl, sizeof (*zl));
2865 	}
2866 }
2867 
2868 /*
2869  * Search back through the directory tree, using the ".." entries.
2870  * Lock each directory in the chain to prevent concurrent renames.
2871  * Fail any attempt to move a directory into one of its own descendants.
2872  * XXX - z_parent_lock can overlap with map or grow locks
2873  */
2874 static int
2875 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2876 {
2877 	zfs_zlock_t	*zl;
2878 	znode_t		*zp = tdzp;
2879 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2880 	uint64_t	*oidp = &zp->z_id;
2881 	krwlock_t	*rwlp = &szp->z_parent_lock;
2882 	krw_t		rw = RW_WRITER;
2883 
2884 	/*
2885 	 * First pass write-locks szp and compares to zp->z_id.
2886 	 * Later passes read-lock zp and compare to zp->z_parent.
2887 	 */
2888 	do {
2889 		if (!rw_tryenter(rwlp, rw)) {
2890 			/*
2891 			 * Another thread is renaming in this path.
2892 			 * Note that if we are a WRITER, we don't have any
2893 			 * parent_locks held yet.
2894 			 */
2895 			if (rw == RW_READER && zp->z_id > szp->z_id) {
2896 				/*
2897 				 * Drop our locks and restart
2898 				 */
2899 				zfs_rename_unlock(&zl);
2900 				*zlpp = NULL;
2901 				zp = tdzp;
2902 				oidp = &zp->z_id;
2903 				rwlp = &szp->z_parent_lock;
2904 				rw = RW_WRITER;
2905 				continue;
2906 			} else {
2907 				/*
2908 				 * Wait for other thread to drop its locks
2909 				 */
2910 				rw_enter(rwlp, rw);
2911 			}
2912 		}
2913 
2914 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2915 		zl->zl_rwlock = rwlp;
2916 		zl->zl_znode = NULL;
2917 		zl->zl_next = *zlpp;
2918 		*zlpp = zl;
2919 
2920 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2921 			return (EINVAL);
2922 
2923 		if (*oidp == rootid)		/* We've hit the top */
2924 			return (0);
2925 
2926 		if (rw == RW_READER) {		/* i.e. not the first pass */
2927 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2928 			if (error)
2929 				return (error);
2930 			zl->zl_znode = zp;
2931 		}
2932 		oidp = &zp->z_phys->zp_parent;
2933 		rwlp = &zp->z_parent_lock;
2934 		rw = RW_READER;
2935 
2936 	} while (zp->z_id != sdzp->z_id);
2937 
2938 	return (0);
2939 }
2940 
2941 /*
2942  * Move an entry from the provided source directory to the target
2943  * directory.  Change the entry name as indicated.
2944  *
2945  *	IN:	sdvp	- Source directory containing the "old entry".
2946  *		snm	- Old entry name.
2947  *		tdvp	- Target directory to contain the "new entry".
2948  *		tnm	- New entry name.
2949  *		cr	- credentials of caller.
2950  *		ct	- caller context
2951  *		flags	- case flags
2952  *
2953  *	RETURN:	0 if success
2954  *		error code if failure
2955  *
2956  * Timestamps:
2957  *	sdvp,tdvp - ctime|mtime updated
2958  */
2959 /*ARGSUSED*/
2960 static int
2961 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
2962     caller_context_t *ct, int flags)
2963 {
2964 	znode_t		*tdzp, *szp, *tzp;
2965 	znode_t		*sdzp = VTOZ(sdvp);
2966 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2967 	zilog_t		*zilog;
2968 	vnode_t		*realvp;
2969 	zfs_dirlock_t	*sdl, *tdl;
2970 	dmu_tx_t	*tx;
2971 	zfs_zlock_t	*zl;
2972 	int		cmp, serr, terr;
2973 	int		error = 0;
2974 	int		zflg = 0;
2975 
2976 	ZFS_ENTER(zfsvfs);
2977 	ZFS_VERIFY_ZP(sdzp);
2978 	zilog = zfsvfs->z_log;
2979 
2980 	/*
2981 	 * Make sure we have the real vp for the target directory.
2982 	 */
2983 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
2984 		tdvp = realvp;
2985 
2986 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
2987 		ZFS_EXIT(zfsvfs);
2988 		return (EXDEV);
2989 	}
2990 
2991 	tdzp = VTOZ(tdvp);
2992 	ZFS_VERIFY_ZP(tdzp);
2993 	if (zfsvfs->z_utf8 && u8_validate(tnm,
2994 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2995 		ZFS_EXIT(zfsvfs);
2996 		return (EILSEQ);
2997 	}
2998 
2999 	if (flags & FIGNORECASE)
3000 		zflg |= ZCILOOK;
3001 
3002 top:
3003 	szp = NULL;
3004 	tzp = NULL;
3005 	zl = NULL;
3006 
3007 	/*
3008 	 * This is to prevent the creation of links into attribute space
3009 	 * by renaming a linked file into/outof an attribute directory.
3010 	 * See the comment in zfs_link() for why this is considered bad.
3011 	 */
3012 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
3013 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
3014 		ZFS_EXIT(zfsvfs);
3015 		return (EINVAL);
3016 	}
3017 
3018 	/*
3019 	 * Lock source and target directory entries.  To prevent deadlock,
3020 	 * a lock ordering must be defined.  We lock the directory with
3021 	 * the smallest object id first, or if it's a tie, the one with
3022 	 * the lexically first name.
3023 	 */
3024 	if (sdzp->z_id < tdzp->z_id) {
3025 		cmp = -1;
3026 	} else if (sdzp->z_id > tdzp->z_id) {
3027 		cmp = 1;
3028 	} else {
3029 		/*
3030 		 * First compare the two name arguments without
3031 		 * considering any case folding.
3032 		 */
3033 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3034 
3035 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3036 		ASSERT(error == 0 || !zfsvfs->z_utf8);
3037 		if (cmp == 0) {
3038 			/*
3039 			 * POSIX: "If the old argument and the new argument
3040 			 * both refer to links to the same existing file,
3041 			 * the rename() function shall return successfully
3042 			 * and perform no other action."
3043 			 */
3044 			ZFS_EXIT(zfsvfs);
3045 			return (0);
3046 		}
3047 		/*
3048 		 * If the file system is case-folding, then we may
3049 		 * have some more checking to do.  A case-folding file
3050 		 * system is either supporting mixed case sensitivity
3051 		 * access or is completely case-insensitive.  Note
3052 		 * that the file system is always case preserving.
3053 		 *
3054 		 * In mixed sensitivity mode case sensitive behavior
3055 		 * is the default.  FIGNORECASE must be used to
3056 		 * explicitly request case insensitive behavior.
3057 		 *
3058 		 * If the source and target names provided differ only
3059 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
3060 		 * we will treat this as a special case in the
3061 		 * case-insensitive mode: as long as the source name
3062 		 * is an exact match, we will allow this to proceed as
3063 		 * a name-change request.
3064 		 */
3065 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3066 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
3067 		    flags & FIGNORECASE)) &&
3068 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3069 		    &error) == 0) {
3070 			/*
3071 			 * case preserving rename request, require exact
3072 			 * name matches
3073 			 */
3074 			zflg |= ZCIEXACT;
3075 			zflg &= ~ZCILOOK;
3076 		}
3077 	}
3078 
3079 	if (cmp < 0) {
3080 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3081 		    ZEXISTS | zflg, NULL, NULL);
3082 		terr = zfs_dirent_lock(&tdl,
3083 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3084 	} else {
3085 		terr = zfs_dirent_lock(&tdl,
3086 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3087 		serr = zfs_dirent_lock(&sdl,
3088 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3089 		    NULL, NULL);
3090 	}
3091 
3092 	if (serr) {
3093 		/*
3094 		 * Source entry invalid or not there.
3095 		 */
3096 		if (!terr) {
3097 			zfs_dirent_unlock(tdl);
3098 			if (tzp)
3099 				VN_RELE(ZTOV(tzp));
3100 		}
3101 		if (strcmp(snm, "..") == 0)
3102 			serr = EINVAL;
3103 		ZFS_EXIT(zfsvfs);
3104 		return (serr);
3105 	}
3106 	if (terr) {
3107 		zfs_dirent_unlock(sdl);
3108 		VN_RELE(ZTOV(szp));
3109 		if (strcmp(tnm, "..") == 0)
3110 			terr = EINVAL;
3111 		ZFS_EXIT(zfsvfs);
3112 		return (terr);
3113 	}
3114 
3115 	/*
3116 	 * Must have write access at the source to remove the old entry
3117 	 * and write access at the target to create the new entry.
3118 	 * Note that if target and source are the same, this can be
3119 	 * done in a single check.
3120 	 */
3121 
3122 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3123 		goto out;
3124 
3125 	if (ZTOV(szp)->v_type == VDIR) {
3126 		/*
3127 		 * Check to make sure rename is valid.
3128 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3129 		 */
3130 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3131 			goto out;
3132 	}
3133 
3134 	/*
3135 	 * Does target exist?
3136 	 */
3137 	if (tzp) {
3138 		/*
3139 		 * Source and target must be the same type.
3140 		 */
3141 		if (ZTOV(szp)->v_type == VDIR) {
3142 			if (ZTOV(tzp)->v_type != VDIR) {
3143 				error = ENOTDIR;
3144 				goto out;
3145 			}
3146 		} else {
3147 			if (ZTOV(tzp)->v_type == VDIR) {
3148 				error = EISDIR;
3149 				goto out;
3150 			}
3151 		}
3152 		/*
3153 		 * POSIX dictates that when the source and target
3154 		 * entries refer to the same file object, rename
3155 		 * must do nothing and exit without error.
3156 		 */
3157 		if (szp->z_id == tzp->z_id) {
3158 			error = 0;
3159 			goto out;
3160 		}
3161 	}
3162 
3163 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3164 	if (tzp)
3165 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3166 
3167 	/*
3168 	 * notify the target directory if it is not the same
3169 	 * as source directory.
3170 	 */
3171 	if (tdvp != sdvp) {
3172 		vnevent_rename_dest_dir(tdvp, ct);
3173 	}
3174 
3175 	tx = dmu_tx_create(zfsvfs->z_os);
3176 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
3177 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
3178 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3179 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3180 	if (sdzp != tdzp)
3181 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
3182 	if (tzp)
3183 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* parent changes */
3184 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3185 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3186 	if (error) {
3187 		if (zl != NULL)
3188 			zfs_rename_unlock(&zl);
3189 		zfs_dirent_unlock(sdl);
3190 		zfs_dirent_unlock(tdl);
3191 		VN_RELE(ZTOV(szp));
3192 		if (tzp)
3193 			VN_RELE(ZTOV(tzp));
3194 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3195 			dmu_tx_wait(tx);
3196 			dmu_tx_abort(tx);
3197 			goto top;
3198 		}
3199 		dmu_tx_abort(tx);
3200 		ZFS_EXIT(zfsvfs);
3201 		return (error);
3202 	}
3203 
3204 	if (tzp)	/* Attempt to remove the existing target */
3205 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3206 
3207 	if (error == 0) {
3208 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3209 		if (error == 0) {
3210 			szp->z_phys->zp_flags |= ZFS_AV_MODIFIED;
3211 
3212 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3213 			ASSERT(error == 0);
3214 
3215 			zfs_log_rename(zilog, tx,
3216 			    TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0),
3217 			    sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp);
3218 
3219 			/* Update path information for the target vnode */
3220 			vn_renamepath(tdvp, ZTOV(szp), tnm, strlen(tnm));
3221 		}
3222 	}
3223 
3224 	dmu_tx_commit(tx);
3225 out:
3226 	if (zl != NULL)
3227 		zfs_rename_unlock(&zl);
3228 
3229 	zfs_dirent_unlock(sdl);
3230 	zfs_dirent_unlock(tdl);
3231 
3232 	VN_RELE(ZTOV(szp));
3233 	if (tzp)
3234 		VN_RELE(ZTOV(tzp));
3235 
3236 	ZFS_EXIT(zfsvfs);
3237 	return (error);
3238 }
3239 
3240 /*
3241  * Insert the indicated symbolic reference entry into the directory.
3242  *
3243  *	IN:	dvp	- Directory to contain new symbolic link.
3244  *		link	- Name for new symlink entry.
3245  *		vap	- Attributes of new entry.
3246  *		target	- Target path of new symlink.
3247  *		cr	- credentials of caller.
3248  *		ct	- caller context
3249  *		flags	- case flags
3250  *
3251  *	RETURN:	0 if success
3252  *		error code if failure
3253  *
3254  * Timestamps:
3255  *	dvp - ctime|mtime updated
3256  */
3257 /*ARGSUSED*/
3258 static int
3259 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3260     caller_context_t *ct, int flags)
3261 {
3262 	znode_t		*zp, *dzp = VTOZ(dvp);
3263 	zfs_dirlock_t	*dl;
3264 	dmu_tx_t	*tx;
3265 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3266 	zilog_t		*zilog;
3267 	int		len = strlen(link);
3268 	int		error;
3269 	int		zflg = ZNEW;
3270 	zfs_fuid_info_t *fuidp = NULL;
3271 
3272 	ASSERT(vap->va_type == VLNK);
3273 
3274 	ZFS_ENTER(zfsvfs);
3275 	ZFS_VERIFY_ZP(dzp);
3276 	zilog = zfsvfs->z_log;
3277 
3278 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3279 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3280 		ZFS_EXIT(zfsvfs);
3281 		return (EILSEQ);
3282 	}
3283 	if (flags & FIGNORECASE)
3284 		zflg |= ZCILOOK;
3285 top:
3286 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3287 		ZFS_EXIT(zfsvfs);
3288 		return (error);
3289 	}
3290 
3291 	if (len > MAXPATHLEN) {
3292 		ZFS_EXIT(zfsvfs);
3293 		return (ENAMETOOLONG);
3294 	}
3295 
3296 	/*
3297 	 * Attempt to lock directory; fail if entry already exists.
3298 	 */
3299 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3300 	if (error) {
3301 		ZFS_EXIT(zfsvfs);
3302 		return (error);
3303 	}
3304 
3305 	tx = dmu_tx_create(zfsvfs->z_os);
3306 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3307 	dmu_tx_hold_bonus(tx, dzp->z_id);
3308 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3309 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
3310 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
3311 	if (IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))) {
3312 		if (zfsvfs->z_fuid_obj == 0) {
3313 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
3314 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3315 			    FUID_SIZE_ESTIMATE(zfsvfs));
3316 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
3317 		} else {
3318 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
3319 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
3320 			    FUID_SIZE_ESTIMATE(zfsvfs));
3321 		}
3322 	}
3323 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3324 	if (error) {
3325 		zfs_dirent_unlock(dl);
3326 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3327 			dmu_tx_wait(tx);
3328 			dmu_tx_abort(tx);
3329 			goto top;
3330 		}
3331 		dmu_tx_abort(tx);
3332 		ZFS_EXIT(zfsvfs);
3333 		return (error);
3334 	}
3335 
3336 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
3337 
3338 	/*
3339 	 * Create a new object for the symlink.
3340 	 * Put the link content into bonus buffer if it will fit;
3341 	 * otherwise, store it just like any other file data.
3342 	 */
3343 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
3344 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, len, NULL, &fuidp);
3345 		if (len != 0)
3346 			bcopy(link, zp->z_phys + 1, len);
3347 	} else {
3348 		dmu_buf_t *dbp;
3349 
3350 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, NULL, &fuidp);
3351 		/*
3352 		 * Nothing can access the znode yet so no locking needed
3353 		 * for growing the znode's blocksize.
3354 		 */
3355 		zfs_grow_blocksize(zp, len, tx);
3356 
3357 		VERIFY(0 == dmu_buf_hold(zfsvfs->z_os,
3358 		    zp->z_id, 0, FTAG, &dbp));
3359 		dmu_buf_will_dirty(dbp, tx);
3360 
3361 		ASSERT3U(len, <=, dbp->db_size);
3362 		bcopy(link, dbp->db_data, len);
3363 		dmu_buf_rele(dbp, FTAG);
3364 	}
3365 	zp->z_phys->zp_size = len;
3366 
3367 	/*
3368 	 * Insert the new object into the directory.
3369 	 */
3370 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3371 out:
3372 	if (error == 0) {
3373 		uint64_t txtype = TX_SYMLINK;
3374 		if (flags & FIGNORECASE)
3375 			txtype |= TX_CI;
3376 		zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3377 	}
3378 	if (fuidp)
3379 		zfs_fuid_info_free(fuidp);
3380 
3381 	dmu_tx_commit(tx);
3382 
3383 	zfs_dirent_unlock(dl);
3384 
3385 	VN_RELE(ZTOV(zp));
3386 
3387 	ZFS_EXIT(zfsvfs);
3388 	return (error);
3389 }
3390 
3391 /*
3392  * Return, in the buffer contained in the provided uio structure,
3393  * the symbolic path referred to by vp.
3394  *
3395  *	IN:	vp	- vnode of symbolic link.
3396  *		uoip	- structure to contain the link path.
3397  *		cr	- credentials of caller.
3398  *		ct	- caller context
3399  *
3400  *	OUT:	uio	- structure to contain the link path.
3401  *
3402  *	RETURN:	0 if success
3403  *		error code if failure
3404  *
3405  * Timestamps:
3406  *	vp - atime updated
3407  */
3408 /* ARGSUSED */
3409 static int
3410 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3411 {
3412 	znode_t		*zp = VTOZ(vp);
3413 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3414 	size_t		bufsz;
3415 	int		error;
3416 
3417 	ZFS_ENTER(zfsvfs);
3418 	ZFS_VERIFY_ZP(zp);
3419 
3420 	bufsz = (size_t)zp->z_phys->zp_size;
3421 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
3422 		error = uiomove(zp->z_phys + 1,
3423 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3424 	} else {
3425 		dmu_buf_t *dbp;
3426 		error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp);
3427 		if (error) {
3428 			ZFS_EXIT(zfsvfs);
3429 			return (error);
3430 		}
3431 		error = uiomove(dbp->db_data,
3432 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3433 		dmu_buf_rele(dbp, FTAG);
3434 	}
3435 
3436 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3437 	ZFS_EXIT(zfsvfs);
3438 	return (error);
3439 }
3440 
3441 /*
3442  * Insert a new entry into directory tdvp referencing svp.
3443  *
3444  *	IN:	tdvp	- Directory to contain new entry.
3445  *		svp	- vnode of new entry.
3446  *		name	- name of new entry.
3447  *		cr	- credentials of caller.
3448  *		ct	- caller context
3449  *
3450  *	RETURN:	0 if success
3451  *		error code if failure
3452  *
3453  * Timestamps:
3454  *	tdvp - ctime|mtime updated
3455  *	 svp - ctime updated
3456  */
3457 /* ARGSUSED */
3458 static int
3459 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3460     caller_context_t *ct, int flags)
3461 {
3462 	znode_t		*dzp = VTOZ(tdvp);
3463 	znode_t		*tzp, *szp;
3464 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3465 	zilog_t		*zilog;
3466 	zfs_dirlock_t	*dl;
3467 	dmu_tx_t	*tx;
3468 	vnode_t		*realvp;
3469 	int		error;
3470 	int		zf = ZNEW;
3471 	uid_t		owner;
3472 
3473 	ASSERT(tdvp->v_type == VDIR);
3474 
3475 	ZFS_ENTER(zfsvfs);
3476 	ZFS_VERIFY_ZP(dzp);
3477 	zilog = zfsvfs->z_log;
3478 
3479 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3480 		svp = realvp;
3481 
3482 	if (svp->v_vfsp != tdvp->v_vfsp) {
3483 		ZFS_EXIT(zfsvfs);
3484 		return (EXDEV);
3485 	}
3486 	szp = VTOZ(svp);
3487 	ZFS_VERIFY_ZP(szp);
3488 
3489 	if (zfsvfs->z_utf8 && u8_validate(name,
3490 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3491 		ZFS_EXIT(zfsvfs);
3492 		return (EILSEQ);
3493 	}
3494 	if (flags & FIGNORECASE)
3495 		zf |= ZCILOOK;
3496 
3497 top:
3498 	/*
3499 	 * We do not support links between attributes and non-attributes
3500 	 * because of the potential security risk of creating links
3501 	 * into "normal" file space in order to circumvent restrictions
3502 	 * imposed in attribute space.
3503 	 */
3504 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
3505 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
3506 		ZFS_EXIT(zfsvfs);
3507 		return (EINVAL);
3508 	}
3509 
3510 	/*
3511 	 * POSIX dictates that we return EPERM here.
3512 	 * Better choices include ENOTSUP or EISDIR.
3513 	 */
3514 	if (svp->v_type == VDIR) {
3515 		ZFS_EXIT(zfsvfs);
3516 		return (EPERM);
3517 	}
3518 
3519 	owner = zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, cr, ZFS_OWNER);
3520 	if (owner != crgetuid(cr) &&
3521 	    secpolicy_basic_link(cr) != 0) {
3522 		ZFS_EXIT(zfsvfs);
3523 		return (EPERM);
3524 	}
3525 
3526 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3527 		ZFS_EXIT(zfsvfs);
3528 		return (error);
3529 	}
3530 
3531 	/*
3532 	 * Attempt to lock directory; fail if entry already exists.
3533 	 */
3534 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3535 	if (error) {
3536 		ZFS_EXIT(zfsvfs);
3537 		return (error);
3538 	}
3539 
3540 	tx = dmu_tx_create(zfsvfs->z_os);
3541 	dmu_tx_hold_bonus(tx, szp->z_id);
3542 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3543 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3544 	if (error) {
3545 		zfs_dirent_unlock(dl);
3546 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3547 			dmu_tx_wait(tx);
3548 			dmu_tx_abort(tx);
3549 			goto top;
3550 		}
3551 		dmu_tx_abort(tx);
3552 		ZFS_EXIT(zfsvfs);
3553 		return (error);
3554 	}
3555 
3556 	error = zfs_link_create(dl, szp, tx, 0);
3557 
3558 	if (error == 0) {
3559 		uint64_t txtype = TX_LINK;
3560 		if (flags & FIGNORECASE)
3561 			txtype |= TX_CI;
3562 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3563 	}
3564 
3565 	dmu_tx_commit(tx);
3566 
3567 	zfs_dirent_unlock(dl);
3568 
3569 	if (error == 0) {
3570 		vnevent_link(svp, ct);
3571 	}
3572 
3573 	ZFS_EXIT(zfsvfs);
3574 	return (error);
3575 }
3576 
3577 /*
3578  * zfs_null_putapage() is used when the file system has been force
3579  * unmounted. It just drops the pages.
3580  */
3581 /* ARGSUSED */
3582 static int
3583 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3584 		size_t *lenp, int flags, cred_t *cr)
3585 {
3586 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
3587 	return (0);
3588 }
3589 
3590 /*
3591  * Push a page out to disk, klustering if possible.
3592  *
3593  *	IN:	vp	- file to push page to.
3594  *		pp	- page to push.
3595  *		flags	- additional flags.
3596  *		cr	- credentials of caller.
3597  *
3598  *	OUT:	offp	- start of range pushed.
3599  *		lenp	- len of range pushed.
3600  *
3601  *	RETURN:	0 if success
3602  *		error code if failure
3603  *
3604  * NOTE: callers must have locked the page to be pushed.  On
3605  * exit, the page (and all other pages in the kluster) must be
3606  * unlocked.
3607  */
3608 /* ARGSUSED */
3609 static int
3610 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3611 		size_t *lenp, int flags, cred_t *cr)
3612 {
3613 	znode_t		*zp = VTOZ(vp);
3614 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3615 	zilog_t		*zilog = zfsvfs->z_log;
3616 	dmu_tx_t	*tx;
3617 	rl_t		*rl;
3618 	u_offset_t	off, koff;
3619 	size_t		len, klen;
3620 	uint64_t	filesz;
3621 	int		err;
3622 
3623 	filesz = zp->z_phys->zp_size;
3624 	off = pp->p_offset;
3625 	len = PAGESIZE;
3626 	/*
3627 	 * If our blocksize is bigger than the page size, try to kluster
3628 	 * muiltiple pages so that we write a full block (thus avoiding
3629 	 * a read-modify-write).
3630 	 */
3631 	if (off < filesz && zp->z_blksz > PAGESIZE) {
3632 		if (!ISP2(zp->z_blksz)) {
3633 			/* Only one block in the file. */
3634 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3635 			koff = 0;
3636 		} else {
3637 			klen = zp->z_blksz;
3638 			koff = P2ALIGN(off, (u_offset_t)klen);
3639 		}
3640 		ASSERT(koff <= filesz);
3641 		if (koff + klen > filesz)
3642 			klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE);
3643 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
3644 	}
3645 	ASSERT3U(btop(len), ==, btopr(len));
3646 top:
3647 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
3648 	/*
3649 	 * Can't push pages past end-of-file.
3650 	 */
3651 	filesz = zp->z_phys->zp_size;
3652 	if (off >= filesz) {
3653 		/* ignore all pages */
3654 		err = 0;
3655 		goto out;
3656 	} else if (off + len > filesz) {
3657 		int npages = btopr(filesz - off);
3658 		page_t *trunc;
3659 
3660 		page_list_break(&pp, &trunc, npages);
3661 		/* ignore pages past end of file */
3662 		if (trunc)
3663 			pvn_write_done(trunc, flags);
3664 		len = filesz - off;
3665 	}
3666 
3667 	tx = dmu_tx_create(zfsvfs->z_os);
3668 	dmu_tx_hold_write(tx, zp->z_id, off, len);
3669 	dmu_tx_hold_bonus(tx, zp->z_id);
3670 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
3671 	if (err != 0) {
3672 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3673 			zfs_range_unlock(rl);
3674 			dmu_tx_wait(tx);
3675 			dmu_tx_abort(tx);
3676 			err = 0;
3677 			goto top;
3678 		}
3679 		dmu_tx_abort(tx);
3680 		goto out;
3681 	}
3682 
3683 	if (zp->z_blksz <= PAGESIZE) {
3684 		caddr_t va = zfs_map_page(pp, S_READ);
3685 		ASSERT3U(len, <=, PAGESIZE);
3686 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
3687 		zfs_unmap_page(pp, va);
3688 	} else {
3689 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
3690 	}
3691 
3692 	if (err == 0) {
3693 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
3694 		zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0);
3695 		dmu_tx_commit(tx);
3696 	}
3697 
3698 out:
3699 	zfs_range_unlock(rl);
3700 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
3701 	if (offp)
3702 		*offp = off;
3703 	if (lenp)
3704 		*lenp = len;
3705 
3706 	return (err);
3707 }
3708 
3709 /*
3710  * Copy the portion of the file indicated from pages into the file.
3711  * The pages are stored in a page list attached to the files vnode.
3712  *
3713  *	IN:	vp	- vnode of file to push page data to.
3714  *		off	- position in file to put data.
3715  *		len	- amount of data to write.
3716  *		flags	- flags to control the operation.
3717  *		cr	- credentials of caller.
3718  *		ct	- caller context.
3719  *
3720  *	RETURN:	0 if success
3721  *		error code if failure
3722  *
3723  * Timestamps:
3724  *	vp - ctime|mtime updated
3725  */
3726 /*ARGSUSED*/
3727 static int
3728 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
3729     caller_context_t *ct)
3730 {
3731 	znode_t		*zp = VTOZ(vp);
3732 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3733 	page_t		*pp;
3734 	size_t		io_len;
3735 	u_offset_t	io_off;
3736 	uint64_t	filesz;
3737 	int		error = 0;
3738 
3739 	ZFS_ENTER(zfsvfs);
3740 	ZFS_VERIFY_ZP(zp);
3741 
3742 	if (len == 0) {
3743 		/*
3744 		 * Search the entire vp list for pages >= off.
3745 		 */
3746 		error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage,
3747 		    flags, cr);
3748 		goto out;
3749 	}
3750 
3751 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3752 	if (off > filesz) {
3753 		/* past end of file */
3754 		ZFS_EXIT(zfsvfs);
3755 		return (0);
3756 	}
3757 
3758 	len = MIN(len, filesz - off);
3759 
3760 	for (io_off = off; io_off < off + len; io_off += io_len) {
3761 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
3762 			pp = page_lookup(vp, io_off,
3763 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
3764 		} else {
3765 			pp = page_lookup_nowait(vp, io_off,
3766 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
3767 		}
3768 
3769 		if (pp != NULL && pvn_getdirty(pp, flags)) {
3770 			int err;
3771 
3772 			/*
3773 			 * Found a dirty page to push
3774 			 */
3775 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
3776 			if (err)
3777 				error = err;
3778 		} else {
3779 			io_len = PAGESIZE;
3780 		}
3781 	}
3782 out:
3783 	if ((flags & B_ASYNC) == 0)
3784 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
3785 	ZFS_EXIT(zfsvfs);
3786 	return (error);
3787 }
3788 
3789 /*ARGSUSED*/
3790 void
3791 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
3792 {
3793 	znode_t	*zp = VTOZ(vp);
3794 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3795 	int error;
3796 
3797 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
3798 	if (zp->z_dbuf == NULL) {
3799 		/*
3800 		 * The fs has been unmounted, or we did a
3801 		 * suspend/resume and this file no longer exists.
3802 		 */
3803 		if (vn_has_cached_data(vp)) {
3804 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
3805 			    B_INVAL, cr);
3806 		}
3807 
3808 		mutex_enter(&zp->z_lock);
3809 		vp->v_count = 0; /* count arrives as 1 */
3810 		mutex_exit(&zp->z_lock);
3811 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
3812 		zfs_znode_free(zp);
3813 		return;
3814 	}
3815 
3816 	/*
3817 	 * Attempt to push any data in the page cache.  If this fails
3818 	 * we will get kicked out later in zfs_zinactive().
3819 	 */
3820 	if (vn_has_cached_data(vp)) {
3821 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
3822 		    cr);
3823 	}
3824 
3825 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3826 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3827 
3828 		dmu_tx_hold_bonus(tx, zp->z_id);
3829 		error = dmu_tx_assign(tx, TXG_WAIT);
3830 		if (error) {
3831 			dmu_tx_abort(tx);
3832 		} else {
3833 			dmu_buf_will_dirty(zp->z_dbuf, tx);
3834 			mutex_enter(&zp->z_lock);
3835 			zp->z_atime_dirty = 0;
3836 			mutex_exit(&zp->z_lock);
3837 			dmu_tx_commit(tx);
3838 		}
3839 	}
3840 
3841 	zfs_zinactive(zp);
3842 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
3843 }
3844 
3845 /*
3846  * Bounds-check the seek operation.
3847  *
3848  *	IN:	vp	- vnode seeking within
3849  *		ooff	- old file offset
3850  *		noffp	- pointer to new file offset
3851  *		ct	- caller context
3852  *
3853  *	RETURN:	0 if success
3854  *		EINVAL if new offset invalid
3855  */
3856 /* ARGSUSED */
3857 static int
3858 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
3859     caller_context_t *ct)
3860 {
3861 	if (vp->v_type == VDIR)
3862 		return (0);
3863 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3864 }
3865 
3866 /*
3867  * Pre-filter the generic locking function to trap attempts to place
3868  * a mandatory lock on a memory mapped file.
3869  */
3870 static int
3871 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3872     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
3873 {
3874 	znode_t *zp = VTOZ(vp);
3875 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3876 	int error;
3877 
3878 	ZFS_ENTER(zfsvfs);
3879 	ZFS_VERIFY_ZP(zp);
3880 
3881 	/*
3882 	 * We are following the UFS semantics with respect to mapcnt
3883 	 * here: If we see that the file is mapped already, then we will
3884 	 * return an error, but we don't worry about races between this
3885 	 * function and zfs_map().
3886 	 */
3887 	if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3888 		ZFS_EXIT(zfsvfs);
3889 		return (EAGAIN);
3890 	}
3891 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct);
3892 	ZFS_EXIT(zfsvfs);
3893 	return (error);
3894 }
3895 
3896 /*
3897  * If we can't find a page in the cache, we will create a new page
3898  * and fill it with file data.  For efficiency, we may try to fill
3899  * multiple pages at once (klustering).
3900  */
3901 static int
3902 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3903     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3904 {
3905 	znode_t *zp = VTOZ(vp);
3906 	page_t *pp, *cur_pp;
3907 	objset_t *os = zp->z_zfsvfs->z_os;
3908 	caddr_t va;
3909 	u_offset_t io_off, total;
3910 	uint64_t oid = zp->z_id;
3911 	size_t io_len;
3912 	uint64_t filesz;
3913 	int err;
3914 
3915 	/*
3916 	 * If we are only asking for a single page don't bother klustering.
3917 	 */
3918 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3919 	if (off >= filesz)
3920 		return (EFAULT);
3921 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3922 		io_off = off;
3923 		io_len = PAGESIZE;
3924 		pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr);
3925 	} else {
3926 		/*
3927 		 * Try to fill a kluster of pages (a blocks worth).
3928 		 */
3929 		size_t klen;
3930 		u_offset_t koff;
3931 
3932 		if (!ISP2(zp->z_blksz)) {
3933 			/* Only one block in the file. */
3934 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3935 			koff = 0;
3936 		} else {
3937 			/*
3938 			 * It would be ideal to align our offset to the
3939 			 * blocksize but doing so has resulted in some
3940 			 * strange application crashes. For now, we
3941 			 * leave the offset as is and only adjust the
3942 			 * length if we are off the end of the file.
3943 			 */
3944 			koff = off;
3945 			klen = plsz;
3946 		}
3947 		ASSERT(koff <= filesz);
3948 		if (koff + klen > filesz)
3949 			klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff;
3950 		ASSERT3U(off, >=, koff);
3951 		ASSERT3U(off, <, koff + klen);
3952 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3953 		    &io_len, koff, klen, 0);
3954 	}
3955 	if (pp == NULL) {
3956 		/*
3957 		 * Some other thread entered the page before us.
3958 		 * Return to zfs_getpage to retry the lookup.
3959 		 */
3960 		*pl = NULL;
3961 		return (0);
3962 	}
3963 
3964 	/*
3965 	 * Fill the pages in the kluster.
3966 	 */
3967 	cur_pp = pp;
3968 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3969 		ASSERT3U(io_off, ==, cur_pp->p_offset);
3970 		va = zfs_map_page(cur_pp, S_WRITE);
3971 		err = dmu_read(os, oid, io_off, PAGESIZE, va);
3972 		zfs_unmap_page(cur_pp, va);
3973 		if (err) {
3974 			/* On error, toss the entire kluster */
3975 			pvn_read_done(pp, B_ERROR);
3976 			/* convert checksum errors into IO errors */
3977 			if (err == ECKSUM)
3978 				err = EIO;
3979 			return (err);
3980 		}
3981 		cur_pp = cur_pp->p_next;
3982 	}
3983 out:
3984 	/*
3985 	 * Fill in the page list array from the kluster.  If
3986 	 * there are too many pages in the kluster, return
3987 	 * as many pages as possible starting from the desired
3988 	 * offset `off'.
3989 	 * NOTE: the page list will always be null terminated.
3990 	 */
3991 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3992 
3993 	return (0);
3994 }
3995 
3996 /*
3997  * Return pointers to the pages for the file region [off, off + len]
3998  * in the pl array.  If plsz is greater than len, this function may
3999  * also return page pointers from before or after the specified
4000  * region (i.e. some region [off', off' + plsz]).  These additional
4001  * pages are only returned if they are already in the cache, or were
4002  * created as part of a klustered read.
4003  *
4004  *	IN:	vp	- vnode of file to get data from.
4005  *		off	- position in file to get data from.
4006  *		len	- amount of data to retrieve.
4007  *		plsz	- length of provided page list.
4008  *		seg	- segment to obtain pages for.
4009  *		addr	- virtual address of fault.
4010  *		rw	- mode of created pages.
4011  *		cr	- credentials of caller.
4012  *		ct	- caller context.
4013  *
4014  *	OUT:	protp	- protection mode of created pages.
4015  *		pl	- list of pages created.
4016  *
4017  *	RETURN:	0 if success
4018  *		error code if failure
4019  *
4020  * Timestamps:
4021  *	vp - atime updated
4022  */
4023 /* ARGSUSED */
4024 static int
4025 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4026 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4027 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4028 {
4029 	znode_t		*zp = VTOZ(vp);
4030 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4031 	page_t		*pp, **pl0 = pl;
4032 	int		need_unlock = 0, err = 0;
4033 	offset_t	orig_off;
4034 
4035 	ZFS_ENTER(zfsvfs);
4036 	ZFS_VERIFY_ZP(zp);
4037 
4038 	if (protp)
4039 		*protp = PROT_ALL;
4040 
4041 	/* no faultahead (for now) */
4042 	if (pl == NULL) {
4043 		ZFS_EXIT(zfsvfs);
4044 		return (0);
4045 	}
4046 
4047 	/* can't fault past EOF */
4048 	if (off >= zp->z_phys->zp_size) {
4049 		ZFS_EXIT(zfsvfs);
4050 		return (EFAULT);
4051 	}
4052 	orig_off = off;
4053 
4054 	/*
4055 	 * If we already own the lock, then we must be page faulting
4056 	 * in the middle of a write to this file (i.e., we are writing
4057 	 * to this file using data from a mapped region of the file).
4058 	 */
4059 	if (rw_owner(&zp->z_map_lock) != curthread) {
4060 		rw_enter(&zp->z_map_lock, RW_WRITER);
4061 		need_unlock = TRUE;
4062 	}
4063 
4064 	/*
4065 	 * Loop through the requested range [off, off + len] looking
4066 	 * for pages.  If we don't find a page, we will need to create
4067 	 * a new page and fill it with data from the file.
4068 	 */
4069 	while (len > 0) {
4070 		if (plsz < PAGESIZE)
4071 			break;
4072 		if (pp = page_lookup(vp, off, SE_SHARED)) {
4073 			*pl++ = pp;
4074 			off += PAGESIZE;
4075 			addr += PAGESIZE;
4076 			len -= PAGESIZE;
4077 			plsz -= PAGESIZE;
4078 		} else {
4079 			err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw);
4080 			if (err)
4081 				goto out;
4082 			/*
4083 			 * klustering may have changed our region
4084 			 * to be block aligned.
4085 			 */
4086 			if (((pp = *pl) != 0) && (off != pp->p_offset)) {
4087 				int delta = off - pp->p_offset;
4088 				len += delta;
4089 				off -= delta;
4090 				addr -= delta;
4091 			}
4092 			while (*pl) {
4093 				pl++;
4094 				off += PAGESIZE;
4095 				addr += PAGESIZE;
4096 				plsz -= PAGESIZE;
4097 				if (len > PAGESIZE)
4098 					len -= PAGESIZE;
4099 				else
4100 					len = 0;
4101 			}
4102 		}
4103 	}
4104 
4105 	/*
4106 	 * Fill out the page array with any pages already in the cache.
4107 	 */
4108 	while (plsz > 0) {
4109 		pp = page_lookup_nowait(vp, off, SE_SHARED);
4110 		if (pp == NULL)
4111 			break;
4112 		*pl++ = pp;
4113 		off += PAGESIZE;
4114 		plsz -= PAGESIZE;
4115 	}
4116 
4117 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4118 out:
4119 	/*
4120 	 * We can't grab the range lock for the page as reader which would
4121 	 * stop truncation as this leads to deadlock. So we need to recheck
4122 	 * the file size.
4123 	 */
4124 	if (orig_off >= zp->z_phys->zp_size)
4125 		err = EFAULT;
4126 	if (err) {
4127 		/*
4128 		 * Release any pages we have previously locked.
4129 		 */
4130 		while (pl > pl0)
4131 			page_unlock(*--pl);
4132 	}
4133 
4134 	*pl = NULL;
4135 
4136 	if (need_unlock)
4137 		rw_exit(&zp->z_map_lock);
4138 
4139 	ZFS_EXIT(zfsvfs);
4140 	return (err);
4141 }
4142 
4143 /*
4144  * Request a memory map for a section of a file.  This code interacts
4145  * with common code and the VM system as follows:
4146  *
4147  *	common code calls mmap(), which ends up in smmap_common()
4148  *
4149  *	this calls VOP_MAP(), which takes you into (say) zfs
4150  *
4151  *	zfs_map() calls as_map(), passing segvn_create() as the callback
4152  *
4153  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
4154  *
4155  *	zfs_addmap() updates z_mapcnt
4156  */
4157 /*ARGSUSED*/
4158 static int
4159 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4160     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4161     caller_context_t *ct)
4162 {
4163 	znode_t *zp = VTOZ(vp);
4164 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4165 	segvn_crargs_t	vn_a;
4166 	int		error;
4167 
4168 	ZFS_ENTER(zfsvfs);
4169 	ZFS_VERIFY_ZP(zp);
4170 
4171 	if ((prot & PROT_WRITE) &&
4172 	    (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY |
4173 	    ZFS_APPENDONLY))) {
4174 		ZFS_EXIT(zfsvfs);
4175 		return (EPERM);
4176 	}
4177 
4178 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4179 	    (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED)) {
4180 		ZFS_EXIT(zfsvfs);
4181 		return (EACCES);
4182 	}
4183 
4184 	if (vp->v_flag & VNOMAP) {
4185 		ZFS_EXIT(zfsvfs);
4186 		return (ENOSYS);
4187 	}
4188 
4189 	if (off < 0 || len > MAXOFFSET_T - off) {
4190 		ZFS_EXIT(zfsvfs);
4191 		return (ENXIO);
4192 	}
4193 
4194 	if (vp->v_type != VREG) {
4195 		ZFS_EXIT(zfsvfs);
4196 		return (ENODEV);
4197 	}
4198 
4199 	/*
4200 	 * If file is locked, disallow mapping.
4201 	 */
4202 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) {
4203 		ZFS_EXIT(zfsvfs);
4204 		return (EAGAIN);
4205 	}
4206 
4207 	as_rangelock(as);
4208 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4209 	if (error != 0) {
4210 		as_rangeunlock(as);
4211 		ZFS_EXIT(zfsvfs);
4212 		return (error);
4213 	}
4214 
4215 	vn_a.vp = vp;
4216 	vn_a.offset = (u_offset_t)off;
4217 	vn_a.type = flags & MAP_TYPE;
4218 	vn_a.prot = prot;
4219 	vn_a.maxprot = maxprot;
4220 	vn_a.cred = cr;
4221 	vn_a.amp = NULL;
4222 	vn_a.flags = flags & ~MAP_TYPE;
4223 	vn_a.szc = 0;
4224 	vn_a.lgrp_mem_policy_flags = 0;
4225 
4226 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4227 
4228 	as_rangeunlock(as);
4229 	ZFS_EXIT(zfsvfs);
4230 	return (error);
4231 }
4232 
4233 /* ARGSUSED */
4234 static int
4235 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4236     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4237     caller_context_t *ct)
4238 {
4239 	uint64_t pages = btopr(len);
4240 
4241 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4242 	return (0);
4243 }
4244 
4245 /*
4246  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4247  * more accurate mtime for the associated file.  Since we don't have a way of
4248  * detecting when the data was actually modified, we have to resort to
4249  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4250  * last page is pushed.  The problem occurs when the msync() call is omitted,
4251  * which by far the most common case:
4252  *
4253  * 	open()
4254  * 	mmap()
4255  * 	<modify memory>
4256  * 	munmap()
4257  * 	close()
4258  * 	<time lapse>
4259  * 	putpage() via fsflush
4260  *
4261  * If we wait until fsflush to come along, we can have a modification time that
4262  * is some arbitrary point in the future.  In order to prevent this in the
4263  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4264  * torn down.
4265  */
4266 /* ARGSUSED */
4267 static int
4268 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4269     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4270     caller_context_t *ct)
4271 {
4272 	uint64_t pages = btopr(len);
4273 
4274 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4275 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4276 
4277 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4278 	    vn_has_cached_data(vp))
4279 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4280 
4281 	return (0);
4282 }
4283 
4284 /*
4285  * Free or allocate space in a file.  Currently, this function only
4286  * supports the `F_FREESP' command.  However, this command is somewhat
4287  * misnamed, as its functionality includes the ability to allocate as
4288  * well as free space.
4289  *
4290  *	IN:	vp	- vnode of file to free data in.
4291  *		cmd	- action to take (only F_FREESP supported).
4292  *		bfp	- section of file to free/alloc.
4293  *		flag	- current file open mode flags.
4294  *		offset	- current file offset.
4295  *		cr	- credentials of caller [UNUSED].
4296  *		ct	- caller context.
4297  *
4298  *	RETURN:	0 if success
4299  *		error code if failure
4300  *
4301  * Timestamps:
4302  *	vp - ctime|mtime updated
4303  */
4304 /* ARGSUSED */
4305 static int
4306 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4307     offset_t offset, cred_t *cr, caller_context_t *ct)
4308 {
4309 	znode_t		*zp = VTOZ(vp);
4310 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4311 	uint64_t	off, len;
4312 	int		error;
4313 
4314 	ZFS_ENTER(zfsvfs);
4315 	ZFS_VERIFY_ZP(zp);
4316 
4317 	if (cmd != F_FREESP) {
4318 		ZFS_EXIT(zfsvfs);
4319 		return (EINVAL);
4320 	}
4321 
4322 	if (error = convoff(vp, bfp, 0, offset)) {
4323 		ZFS_EXIT(zfsvfs);
4324 		return (error);
4325 	}
4326 
4327 	if (bfp->l_len < 0) {
4328 		ZFS_EXIT(zfsvfs);
4329 		return (EINVAL);
4330 	}
4331 
4332 	off = bfp->l_start;
4333 	len = bfp->l_len; /* 0 means from off to end of file */
4334 
4335 	error = zfs_freesp(zp, off, len, flag, TRUE);
4336 
4337 	ZFS_EXIT(zfsvfs);
4338 	return (error);
4339 }
4340 
4341 /*ARGSUSED*/
4342 static int
4343 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4344 {
4345 	znode_t		*zp = VTOZ(vp);
4346 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4347 	uint32_t	gen;
4348 	uint64_t	object = zp->z_id;
4349 	zfid_short_t	*zfid;
4350 	int		size, i;
4351 
4352 	ZFS_ENTER(zfsvfs);
4353 	ZFS_VERIFY_ZP(zp);
4354 	gen = (uint32_t)zp->z_gen;
4355 
4356 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4357 	if (fidp->fid_len < size) {
4358 		fidp->fid_len = size;
4359 		ZFS_EXIT(zfsvfs);
4360 		return (ENOSPC);
4361 	}
4362 
4363 	zfid = (zfid_short_t *)fidp;
4364 
4365 	zfid->zf_len = size;
4366 
4367 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4368 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4369 
4370 	/* Must have a non-zero generation number to distinguish from .zfs */
4371 	if (gen == 0)
4372 		gen = 1;
4373 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4374 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4375 
4376 	if (size == LONG_FID_LEN) {
4377 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4378 		zfid_long_t	*zlfid;
4379 
4380 		zlfid = (zfid_long_t *)fidp;
4381 
4382 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4383 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4384 
4385 		/* XXX - this should be the generation number for the objset */
4386 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4387 			zlfid->zf_setgen[i] = 0;
4388 	}
4389 
4390 	ZFS_EXIT(zfsvfs);
4391 	return (0);
4392 }
4393 
4394 static int
4395 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4396     caller_context_t *ct)
4397 {
4398 	znode_t		*zp, *xzp;
4399 	zfsvfs_t	*zfsvfs;
4400 	zfs_dirlock_t	*dl;
4401 	int		error;
4402 
4403 	switch (cmd) {
4404 	case _PC_LINK_MAX:
4405 		*valp = ULONG_MAX;
4406 		return (0);
4407 
4408 	case _PC_FILESIZEBITS:
4409 		*valp = 64;
4410 		return (0);
4411 
4412 	case _PC_XATTR_EXISTS:
4413 		zp = VTOZ(vp);
4414 		zfsvfs = zp->z_zfsvfs;
4415 		ZFS_ENTER(zfsvfs);
4416 		ZFS_VERIFY_ZP(zp);
4417 		*valp = 0;
4418 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4419 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4420 		if (error == 0) {
4421 			zfs_dirent_unlock(dl);
4422 			if (!zfs_dirempty(xzp))
4423 				*valp = 1;
4424 			VN_RELE(ZTOV(xzp));
4425 		} else if (error == ENOENT) {
4426 			/*
4427 			 * If there aren't extended attributes, it's the
4428 			 * same as having zero of them.
4429 			 */
4430 			error = 0;
4431 		}
4432 		ZFS_EXIT(zfsvfs);
4433 		return (error);
4434 
4435 	case _PC_SATTR_ENABLED:
4436 	case _PC_SATTR_EXISTS:
4437 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4438 		    (vp->v_type == VREG || vp->v_type == VDIR);
4439 		return (0);
4440 
4441 	case _PC_ACL_ENABLED:
4442 		*valp = _ACL_ACE_ENABLED;
4443 		return (0);
4444 
4445 	case _PC_MIN_HOLE_SIZE:
4446 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4447 		return (0);
4448 
4449 	default:
4450 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4451 	}
4452 }
4453 
4454 /*ARGSUSED*/
4455 static int
4456 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4457     caller_context_t *ct)
4458 {
4459 	znode_t *zp = VTOZ(vp);
4460 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4461 	int error;
4462 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4463 
4464 	ZFS_ENTER(zfsvfs);
4465 	ZFS_VERIFY_ZP(zp);
4466 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4467 	ZFS_EXIT(zfsvfs);
4468 
4469 	return (error);
4470 }
4471 
4472 /*ARGSUSED*/
4473 static int
4474 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4475     caller_context_t *ct)
4476 {
4477 	znode_t *zp = VTOZ(vp);
4478 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4479 	int error;
4480 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4481 
4482 	ZFS_ENTER(zfsvfs);
4483 	ZFS_VERIFY_ZP(zp);
4484 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4485 	ZFS_EXIT(zfsvfs);
4486 	return (error);
4487 }
4488 
4489 /*
4490  * Predeclare these here so that the compiler assumes that
4491  * this is an "old style" function declaration that does
4492  * not include arguments => we won't get type mismatch errors
4493  * in the initializations that follow.
4494  */
4495 static int zfs_inval();
4496 static int zfs_isdir();
4497 
4498 static int
4499 zfs_inval()
4500 {
4501 	return (EINVAL);
4502 }
4503 
4504 static int
4505 zfs_isdir()
4506 {
4507 	return (EISDIR);
4508 }
4509 /*
4510  * Directory vnode operations template
4511  */
4512 vnodeops_t *zfs_dvnodeops;
4513 const fs_operation_def_t zfs_dvnodeops_template[] = {
4514 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4515 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4516 	VOPNAME_READ,		{ .error = zfs_isdir },
4517 	VOPNAME_WRITE,		{ .error = zfs_isdir },
4518 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4519 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4520 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4521 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4522 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4523 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4524 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4525 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4526 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4527 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
4528 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4529 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4530 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
4531 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4532 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4533 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4534 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4535 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4536 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4537 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4538 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
4539 	NULL,			NULL
4540 };
4541 
4542 /*
4543  * Regular file vnode operations template
4544  */
4545 vnodeops_t *zfs_fvnodeops;
4546 const fs_operation_def_t zfs_fvnodeops_template[] = {
4547 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4548 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4549 	VOPNAME_READ,		{ .vop_read = zfs_read },
4550 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
4551 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4552 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4553 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4554 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4555 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4556 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4557 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4558 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4559 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4560 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4561 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
4562 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
4563 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
4564 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
4565 	VOPNAME_MAP,		{ .vop_map = zfs_map },
4566 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
4567 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
4568 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4569 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4570 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4571 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4572 	NULL,			NULL
4573 };
4574 
4575 /*
4576  * Symbolic link vnode operations template
4577  */
4578 vnodeops_t *zfs_symvnodeops;
4579 const fs_operation_def_t zfs_symvnodeops_template[] = {
4580 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4581 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4582 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4583 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4584 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
4585 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4586 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4587 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4588 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4589 	NULL,			NULL
4590 };
4591 
4592 /*
4593  * Extended attribute directory vnode operations template
4594  *	This template is identical to the directory vnodes
4595  *	operation template except for restricted operations:
4596  *		VOP_MKDIR()
4597  *		VOP_SYMLINK()
4598  * Note that there are other restrictions embedded in:
4599  *	zfs_create()	- restrict type to VREG
4600  *	zfs_link()	- no links into/out of attribute space
4601  *	zfs_rename()	- no moves into/out of attribute space
4602  */
4603 vnodeops_t *zfs_xdvnodeops;
4604 const fs_operation_def_t zfs_xdvnodeops_template[] = {
4605 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4606 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4607 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4608 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4609 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4610 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4611 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4612 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4613 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4614 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4615 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4616 	VOPNAME_MKDIR,		{ .error = zfs_inval },
4617 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4618 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4619 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
4620 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4621 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4622 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4623 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4624 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4625 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4626 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4627 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4628 	NULL,			NULL
4629 };
4630 
4631 /*
4632  * Error vnode operations template
4633  */
4634 vnodeops_t *zfs_evnodeops;
4635 const fs_operation_def_t zfs_evnodeops_template[] = {
4636 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4637 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4638 	NULL,			NULL
4639 };
4640