xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 69c9224907257f806b6bc9df1a730ae53582e95a)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012 by Delphix. All rights reserved.
24  */
25 
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2010 Robert Milkowski */
28 
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/time.h>
32 #include <sys/systm.h>
33 #include <sys/sysmacros.h>
34 #include <sys/resource.h>
35 #include <sys/vfs.h>
36 #include <sys/vfs_opreg.h>
37 #include <sys/vnode.h>
38 #include <sys/file.h>
39 #include <sys/stat.h>
40 #include <sys/kmem.h>
41 #include <sys/taskq.h>
42 #include <sys/uio.h>
43 #include <sys/vmsystm.h>
44 #include <sys/atomic.h>
45 #include <sys/vm.h>
46 #include <vm/seg_vn.h>
47 #include <vm/pvn.h>
48 #include <vm/as.h>
49 #include <vm/kpm.h>
50 #include <vm/seg_kpm.h>
51 #include <sys/mman.h>
52 #include <sys/pathname.h>
53 #include <sys/cmn_err.h>
54 #include <sys/errno.h>
55 #include <sys/unistd.h>
56 #include <sys/zfs_dir.h>
57 #include <sys/zfs_acl.h>
58 #include <sys/zfs_ioctl.h>
59 #include <sys/fs/zfs.h>
60 #include <sys/dmu.h>
61 #include <sys/dmu_objset.h>
62 #include <sys/spa.h>
63 #include <sys/txg.h>
64 #include <sys/dbuf.h>
65 #include <sys/zap.h>
66 #include <sys/sa.h>
67 #include <sys/dirent.h>
68 #include <sys/policy.h>
69 #include <sys/sunddi.h>
70 #include <sys/filio.h>
71 #include <sys/sid.h>
72 #include "fs/fs_subr.h"
73 #include <sys/zfs_ctldir.h>
74 #include <sys/zfs_fuid.h>
75 #include <sys/zfs_sa.h>
76 #include <sys/dnlc.h>
77 #include <sys/zfs_rlock.h>
78 #include <sys/extdirent.h>
79 #include <sys/kidmap.h>
80 #include <sys/cred.h>
81 #include <sys/attr.h>
82 
83 /*
84  * Programming rules.
85  *
86  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
87  * properly lock its in-core state, create a DMU transaction, do the work,
88  * record this work in the intent log (ZIL), commit the DMU transaction,
89  * and wait for the intent log to commit if it is a synchronous operation.
90  * Moreover, the vnode ops must work in both normal and log replay context.
91  * The ordering of events is important to avoid deadlocks and references
92  * to freed memory.  The example below illustrates the following Big Rules:
93  *
94  *  (1) A check must be made in each zfs thread for a mounted file system.
95  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
96  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
97  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
98  *      can return EIO from the calling function.
99  *
100  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
101  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
102  *	First, if it's the last reference, the vnode/znode
103  *	can be freed, so the zp may point to freed memory.  Second, the last
104  *	reference will call zfs_zinactive(), which may induce a lot of work --
105  *	pushing cached pages (which acquires range locks) and syncing out
106  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
107  *	which could deadlock the system if you were already holding one.
108  *	If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
109  *
110  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
111  *	as they can span dmu_tx_assign() calls.
112  *
113  *  (4)	Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
114  *	This is critical because we don't want to block while holding locks.
115  *	Note, in particular, that if a lock is sometimes acquired before
116  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
117  *	use a non-blocking assign can deadlock the system.  The scenario:
118  *
119  *	Thread A has grabbed a lock before calling dmu_tx_assign().
120  *	Thread B is in an already-assigned tx, and blocks for this lock.
121  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
122  *	forever, because the previous txg can't quiesce until B's tx commits.
123  *
124  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
125  *	then drop all locks, call dmu_tx_wait(), and try again.
126  *
127  *  (5)	If the operation succeeded, generate the intent log entry for it
128  *	before dropping locks.  This ensures that the ordering of events
129  *	in the intent log matches the order in which they actually occurred.
130  *      During ZIL replay the zfs_log_* functions will update the sequence
131  *	number to indicate the zil transaction has replayed.
132  *
133  *  (6)	At the end of each vnode op, the DMU tx must always commit,
134  *	regardless of whether there were any errors.
135  *
136  *  (7)	After dropping all locks, invoke zil_commit(zilog, foid)
137  *	to ensure that synchronous semantics are provided when necessary.
138  *
139  * In general, this is how things should be ordered in each vnode op:
140  *
141  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
142  * top:
143  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
144  *	rw_enter(...);			// grab any other locks you need
145  *	tx = dmu_tx_create(...);	// get DMU tx
146  *	dmu_tx_hold_*();		// hold each object you might modify
147  *	error = dmu_tx_assign(tx, TXG_NOWAIT);	// try to assign
148  *	if (error) {
149  *		rw_exit(...);		// drop locks
150  *		zfs_dirent_unlock(dl);	// unlock directory entry
151  *		VN_RELE(...);		// release held vnodes
152  *		if (error == ERESTART) {
153  *			dmu_tx_wait(tx);
154  *			dmu_tx_abort(tx);
155  *			goto top;
156  *		}
157  *		dmu_tx_abort(tx);	// abort DMU tx
158  *		ZFS_EXIT(zfsvfs);	// finished in zfs
159  *		return (error);		// really out of space
160  *	}
161  *	error = do_real_work();		// do whatever this VOP does
162  *	if (error == 0)
163  *		zfs_log_*(...);		// on success, make ZIL entry
164  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
165  *	rw_exit(...);			// drop locks
166  *	zfs_dirent_unlock(dl);		// unlock directory entry
167  *	VN_RELE(...);			// release held vnodes
168  *	zil_commit(zilog, foid);	// synchronous when necessary
169  *	ZFS_EXIT(zfsvfs);		// finished in zfs
170  *	return (error);			// done, report error
171  */
172 
173 /* ARGSUSED */
174 static int
175 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
176 {
177 	znode_t	*zp = VTOZ(*vpp);
178 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
179 
180 	ZFS_ENTER(zfsvfs);
181 	ZFS_VERIFY_ZP(zp);
182 
183 	if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
184 	    ((flag & FAPPEND) == 0)) {
185 		ZFS_EXIT(zfsvfs);
186 		return (EPERM);
187 	}
188 
189 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
190 	    ZTOV(zp)->v_type == VREG &&
191 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
192 		if (fs_vscan(*vpp, cr, 0) != 0) {
193 			ZFS_EXIT(zfsvfs);
194 			return (EACCES);
195 		}
196 	}
197 
198 	/* Keep a count of the synchronous opens in the znode */
199 	if (flag & (FSYNC | FDSYNC))
200 		atomic_inc_32(&zp->z_sync_cnt);
201 
202 	ZFS_EXIT(zfsvfs);
203 	return (0);
204 }
205 
206 /* ARGSUSED */
207 static int
208 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
209     caller_context_t *ct)
210 {
211 	znode_t	*zp = VTOZ(vp);
212 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
213 
214 	/*
215 	 * Clean up any locks held by this process on the vp.
216 	 */
217 	cleanlocks(vp, ddi_get_pid(), 0);
218 	cleanshares(vp, ddi_get_pid());
219 
220 	ZFS_ENTER(zfsvfs);
221 	ZFS_VERIFY_ZP(zp);
222 
223 	/* Decrement the synchronous opens in the znode */
224 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
225 		atomic_dec_32(&zp->z_sync_cnt);
226 
227 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
228 	    ZTOV(zp)->v_type == VREG &&
229 	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
230 		VERIFY(fs_vscan(vp, cr, 1) == 0);
231 
232 	ZFS_EXIT(zfsvfs);
233 	return (0);
234 }
235 
236 /*
237  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
238  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
239  */
240 static int
241 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
242 {
243 	znode_t	*zp = VTOZ(vp);
244 	uint64_t noff = (uint64_t)*off; /* new offset */
245 	uint64_t file_sz;
246 	int error;
247 	boolean_t hole;
248 
249 	file_sz = zp->z_size;
250 	if (noff >= file_sz)  {
251 		return (ENXIO);
252 	}
253 
254 	if (cmd == _FIO_SEEK_HOLE)
255 		hole = B_TRUE;
256 	else
257 		hole = B_FALSE;
258 
259 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
260 
261 	/* end of file? */
262 	if ((error == ESRCH) || (noff > file_sz)) {
263 		/*
264 		 * Handle the virtual hole at the end of file.
265 		 */
266 		if (hole) {
267 			*off = file_sz;
268 			return (0);
269 		}
270 		return (ENXIO);
271 	}
272 
273 	if (noff < *off)
274 		return (error);
275 	*off = noff;
276 	return (error);
277 }
278 
279 /* ARGSUSED */
280 static int
281 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
282     int *rvalp, caller_context_t *ct)
283 {
284 	offset_t off;
285 	int error;
286 	zfsvfs_t *zfsvfs;
287 	znode_t *zp;
288 
289 	switch (com) {
290 	case _FIOFFS:
291 		return (zfs_sync(vp->v_vfsp, 0, cred));
292 
293 		/*
294 		 * The following two ioctls are used by bfu.  Faking out,
295 		 * necessary to avoid bfu errors.
296 		 */
297 	case _FIOGDIO:
298 	case _FIOSDIO:
299 		return (0);
300 
301 	case _FIO_SEEK_DATA:
302 	case _FIO_SEEK_HOLE:
303 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
304 			return (EFAULT);
305 
306 		zp = VTOZ(vp);
307 		zfsvfs = zp->z_zfsvfs;
308 		ZFS_ENTER(zfsvfs);
309 		ZFS_VERIFY_ZP(zp);
310 
311 		/* offset parameter is in/out */
312 		error = zfs_holey(vp, com, &off);
313 		ZFS_EXIT(zfsvfs);
314 		if (error)
315 			return (error);
316 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
317 			return (EFAULT);
318 		return (0);
319 	}
320 	return (ENOTTY);
321 }
322 
323 /*
324  * Utility functions to map and unmap a single physical page.  These
325  * are used to manage the mappable copies of ZFS file data, and therefore
326  * do not update ref/mod bits.
327  */
328 caddr_t
329 zfs_map_page(page_t *pp, enum seg_rw rw)
330 {
331 	if (kpm_enable)
332 		return (hat_kpm_mapin(pp, 0));
333 	ASSERT(rw == S_READ || rw == S_WRITE);
334 	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
335 	    (caddr_t)-1));
336 }
337 
338 void
339 zfs_unmap_page(page_t *pp, caddr_t addr)
340 {
341 	if (kpm_enable) {
342 		hat_kpm_mapout(pp, 0, addr);
343 	} else {
344 		ppmapout(addr);
345 	}
346 }
347 
348 /*
349  * When a file is memory mapped, we must keep the IO data synchronized
350  * between the DMU cache and the memory mapped pages.  What this means:
351  *
352  * On Write:	If we find a memory mapped page, we write to *both*
353  *		the page and the dmu buffer.
354  */
355 static void
356 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
357 {
358 	int64_t	off;
359 
360 	off = start & PAGEOFFSET;
361 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
362 		page_t *pp;
363 		uint64_t nbytes = MIN(PAGESIZE - off, len);
364 
365 		if (pp = page_lookup(vp, start, SE_SHARED)) {
366 			caddr_t va;
367 
368 			va = zfs_map_page(pp, S_WRITE);
369 			(void) dmu_read(os, oid, start+off, nbytes, va+off,
370 			    DMU_READ_PREFETCH);
371 			zfs_unmap_page(pp, va);
372 			page_unlock(pp);
373 		}
374 		len -= nbytes;
375 		off = 0;
376 	}
377 }
378 
379 /*
380  * When a file is memory mapped, we must keep the IO data synchronized
381  * between the DMU cache and the memory mapped pages.  What this means:
382  *
383  * On Read:	We "read" preferentially from memory mapped pages,
384  *		else we default from the dmu buffer.
385  *
386  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
387  *	the file is memory mapped.
388  */
389 static int
390 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
391 {
392 	znode_t *zp = VTOZ(vp);
393 	objset_t *os = zp->z_zfsvfs->z_os;
394 	int64_t	start, off;
395 	int len = nbytes;
396 	int error = 0;
397 
398 	start = uio->uio_loffset;
399 	off = start & PAGEOFFSET;
400 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
401 		page_t *pp;
402 		uint64_t bytes = MIN(PAGESIZE - off, len);
403 
404 		if (pp = page_lookup(vp, start, SE_SHARED)) {
405 			caddr_t va;
406 
407 			va = zfs_map_page(pp, S_READ);
408 			error = uiomove(va + off, bytes, UIO_READ, uio);
409 			zfs_unmap_page(pp, va);
410 			page_unlock(pp);
411 		} else {
412 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
413 		}
414 		len -= bytes;
415 		off = 0;
416 		if (error)
417 			break;
418 	}
419 	return (error);
420 }
421 
422 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
423 
424 /*
425  * Read bytes from specified file into supplied buffer.
426  *
427  *	IN:	vp	- vnode of file to be read from.
428  *		uio	- structure supplying read location, range info,
429  *			  and return buffer.
430  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
431  *		cr	- credentials of caller.
432  *		ct	- caller context
433  *
434  *	OUT:	uio	- updated offset and range, buffer filled.
435  *
436  *	RETURN:	0 if success
437  *		error code if failure
438  *
439  * Side Effects:
440  *	vp - atime updated if byte count > 0
441  */
442 /* ARGSUSED */
443 static int
444 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
445 {
446 	znode_t		*zp = VTOZ(vp);
447 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
448 	objset_t	*os;
449 	ssize_t		n, nbytes;
450 	int		error;
451 	rl_t		*rl;
452 	xuio_t		*xuio = NULL;
453 
454 	ZFS_ENTER(zfsvfs);
455 	ZFS_VERIFY_ZP(zp);
456 	os = zfsvfs->z_os;
457 
458 	if (zp->z_pflags & ZFS_AV_QUARANTINED) {
459 		ZFS_EXIT(zfsvfs);
460 		return (EACCES);
461 	}
462 
463 	/*
464 	 * Validate file offset
465 	 */
466 	if (uio->uio_loffset < (offset_t)0) {
467 		ZFS_EXIT(zfsvfs);
468 		return (EINVAL);
469 	}
470 
471 	/*
472 	 * Fasttrack empty reads
473 	 */
474 	if (uio->uio_resid == 0) {
475 		ZFS_EXIT(zfsvfs);
476 		return (0);
477 	}
478 
479 	/*
480 	 * Check for mandatory locks
481 	 */
482 	if (MANDMODE(zp->z_mode)) {
483 		if (error = chklock(vp, FREAD,
484 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
485 			ZFS_EXIT(zfsvfs);
486 			return (error);
487 		}
488 	}
489 
490 	/*
491 	 * If we're in FRSYNC mode, sync out this znode before reading it.
492 	 */
493 	if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
494 		zil_commit(zfsvfs->z_log, zp->z_id);
495 
496 	/*
497 	 * Lock the range against changes.
498 	 */
499 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
500 
501 	/*
502 	 * If we are reading past end-of-file we can skip
503 	 * to the end; but we might still need to set atime.
504 	 */
505 	if (uio->uio_loffset >= zp->z_size) {
506 		error = 0;
507 		goto out;
508 	}
509 
510 	ASSERT(uio->uio_loffset < zp->z_size);
511 	n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
512 
513 	if ((uio->uio_extflg == UIO_XUIO) &&
514 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
515 		int nblk;
516 		int blksz = zp->z_blksz;
517 		uint64_t offset = uio->uio_loffset;
518 
519 		xuio = (xuio_t *)uio;
520 		if ((ISP2(blksz))) {
521 			nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
522 			    blksz)) / blksz;
523 		} else {
524 			ASSERT(offset + n <= blksz);
525 			nblk = 1;
526 		}
527 		(void) dmu_xuio_init(xuio, nblk);
528 
529 		if (vn_has_cached_data(vp)) {
530 			/*
531 			 * For simplicity, we always allocate a full buffer
532 			 * even if we only expect to read a portion of a block.
533 			 */
534 			while (--nblk >= 0) {
535 				(void) dmu_xuio_add(xuio,
536 				    dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
537 				    blksz), 0, blksz);
538 			}
539 		}
540 	}
541 
542 	while (n > 0) {
543 		nbytes = MIN(n, zfs_read_chunk_size -
544 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
545 
546 		if (vn_has_cached_data(vp))
547 			error = mappedread(vp, nbytes, uio);
548 		else
549 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
550 		if (error) {
551 			/* convert checksum errors into IO errors */
552 			if (error == ECKSUM)
553 				error = EIO;
554 			break;
555 		}
556 
557 		n -= nbytes;
558 	}
559 out:
560 	zfs_range_unlock(rl);
561 
562 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
563 	ZFS_EXIT(zfsvfs);
564 	return (error);
565 }
566 
567 /*
568  * Write the bytes to a file.
569  *
570  *	IN:	vp	- vnode of file to be written to.
571  *		uio	- structure supplying write location, range info,
572  *			  and data buffer.
573  *		ioflag	- FAPPEND flag set if in append mode.
574  *		cr	- credentials of caller.
575  *		ct	- caller context (NFS/CIFS fem monitor only)
576  *
577  *	OUT:	uio	- updated offset and range.
578  *
579  *	RETURN:	0 if success
580  *		error code if failure
581  *
582  * Timestamps:
583  *	vp - ctime|mtime updated if byte count > 0
584  */
585 
586 /* ARGSUSED */
587 static int
588 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
589 {
590 	znode_t		*zp = VTOZ(vp);
591 	rlim64_t	limit = uio->uio_llimit;
592 	ssize_t		start_resid = uio->uio_resid;
593 	ssize_t		tx_bytes;
594 	uint64_t	end_size;
595 	dmu_tx_t	*tx;
596 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
597 	zilog_t		*zilog;
598 	offset_t	woff;
599 	ssize_t		n, nbytes;
600 	rl_t		*rl;
601 	int		max_blksz = zfsvfs->z_max_blksz;
602 	int		error;
603 	arc_buf_t	*abuf;
604 	iovec_t		*aiov;
605 	xuio_t		*xuio = NULL;
606 	int		i_iov = 0;
607 	int		iovcnt = uio->uio_iovcnt;
608 	iovec_t		*iovp = uio->uio_iov;
609 	int		write_eof;
610 	int		count = 0;
611 	sa_bulk_attr_t	bulk[4];
612 	uint64_t	mtime[2], ctime[2];
613 
614 	/*
615 	 * Fasttrack empty write
616 	 */
617 	n = start_resid;
618 	if (n == 0)
619 		return (0);
620 
621 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
622 		limit = MAXOFFSET_T;
623 
624 	ZFS_ENTER(zfsvfs);
625 	ZFS_VERIFY_ZP(zp);
626 
627 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
628 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
629 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
630 	    &zp->z_size, 8);
631 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
632 	    &zp->z_pflags, 8);
633 
634 	/*
635 	 * If immutable or not appending then return EPERM
636 	 */
637 	if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
638 	    ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
639 	    (uio->uio_loffset < zp->z_size))) {
640 		ZFS_EXIT(zfsvfs);
641 		return (EPERM);
642 	}
643 
644 	zilog = zfsvfs->z_log;
645 
646 	/*
647 	 * Validate file offset
648 	 */
649 	woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
650 	if (woff < 0) {
651 		ZFS_EXIT(zfsvfs);
652 		return (EINVAL);
653 	}
654 
655 	/*
656 	 * Check for mandatory locks before calling zfs_range_lock()
657 	 * in order to prevent a deadlock with locks set via fcntl().
658 	 */
659 	if (MANDMODE((mode_t)zp->z_mode) &&
660 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
661 		ZFS_EXIT(zfsvfs);
662 		return (error);
663 	}
664 
665 	/*
666 	 * Pre-fault the pages to ensure slow (eg NFS) pages
667 	 * don't hold up txg.
668 	 * Skip this if uio contains loaned arc_buf.
669 	 */
670 	if ((uio->uio_extflg == UIO_XUIO) &&
671 	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
672 		xuio = (xuio_t *)uio;
673 	else
674 		uio_prefaultpages(MIN(n, max_blksz), uio);
675 
676 	/*
677 	 * If in append mode, set the io offset pointer to eof.
678 	 */
679 	if (ioflag & FAPPEND) {
680 		/*
681 		 * Obtain an appending range lock to guarantee file append
682 		 * semantics.  We reset the write offset once we have the lock.
683 		 */
684 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
685 		woff = rl->r_off;
686 		if (rl->r_len == UINT64_MAX) {
687 			/*
688 			 * We overlocked the file because this write will cause
689 			 * the file block size to increase.
690 			 * Note that zp_size cannot change with this lock held.
691 			 */
692 			woff = zp->z_size;
693 		}
694 		uio->uio_loffset = woff;
695 	} else {
696 		/*
697 		 * Note that if the file block size will change as a result of
698 		 * this write, then this range lock will lock the entire file
699 		 * so that we can re-write the block safely.
700 		 */
701 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
702 	}
703 
704 	if (woff >= limit) {
705 		zfs_range_unlock(rl);
706 		ZFS_EXIT(zfsvfs);
707 		return (EFBIG);
708 	}
709 
710 	if ((woff + n) > limit || woff > (limit - n))
711 		n = limit - woff;
712 
713 	/* Will this write extend the file length? */
714 	write_eof = (woff + n > zp->z_size);
715 
716 	end_size = MAX(zp->z_size, woff + n);
717 
718 	/*
719 	 * Write the file in reasonable size chunks.  Each chunk is written
720 	 * in a separate transaction; this keeps the intent log records small
721 	 * and allows us to do more fine-grained space accounting.
722 	 */
723 	while (n > 0) {
724 		abuf = NULL;
725 		woff = uio->uio_loffset;
726 again:
727 		if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
728 		    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
729 			if (abuf != NULL)
730 				dmu_return_arcbuf(abuf);
731 			error = EDQUOT;
732 			break;
733 		}
734 
735 		if (xuio && abuf == NULL) {
736 			ASSERT(i_iov < iovcnt);
737 			aiov = &iovp[i_iov];
738 			abuf = dmu_xuio_arcbuf(xuio, i_iov);
739 			dmu_xuio_clear(xuio, i_iov);
740 			DTRACE_PROBE3(zfs_cp_write, int, i_iov,
741 			    iovec_t *, aiov, arc_buf_t *, abuf);
742 			ASSERT((aiov->iov_base == abuf->b_data) ||
743 			    ((char *)aiov->iov_base - (char *)abuf->b_data +
744 			    aiov->iov_len == arc_buf_size(abuf)));
745 			i_iov++;
746 		} else if (abuf == NULL && n >= max_blksz &&
747 		    woff >= zp->z_size &&
748 		    P2PHASE(woff, max_blksz) == 0 &&
749 		    zp->z_blksz == max_blksz) {
750 			/*
751 			 * This write covers a full block.  "Borrow" a buffer
752 			 * from the dmu so that we can fill it before we enter
753 			 * a transaction.  This avoids the possibility of
754 			 * holding up the transaction if the data copy hangs
755 			 * up on a pagefault (e.g., from an NFS server mapping).
756 			 */
757 			size_t cbytes;
758 
759 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
760 			    max_blksz);
761 			ASSERT(abuf != NULL);
762 			ASSERT(arc_buf_size(abuf) == max_blksz);
763 			if (error = uiocopy(abuf->b_data, max_blksz,
764 			    UIO_WRITE, uio, &cbytes)) {
765 				dmu_return_arcbuf(abuf);
766 				break;
767 			}
768 			ASSERT(cbytes == max_blksz);
769 		}
770 
771 		/*
772 		 * Start a transaction.
773 		 */
774 		tx = dmu_tx_create(zfsvfs->z_os);
775 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
776 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
777 		zfs_sa_upgrade_txholds(tx, zp);
778 		error = dmu_tx_assign(tx, TXG_NOWAIT);
779 		if (error) {
780 			if (error == ERESTART) {
781 				dmu_tx_wait(tx);
782 				dmu_tx_abort(tx);
783 				goto again;
784 			}
785 			dmu_tx_abort(tx);
786 			if (abuf != NULL)
787 				dmu_return_arcbuf(abuf);
788 			break;
789 		}
790 
791 		/*
792 		 * If zfs_range_lock() over-locked we grow the blocksize
793 		 * and then reduce the lock range.  This will only happen
794 		 * on the first iteration since zfs_range_reduce() will
795 		 * shrink down r_len to the appropriate size.
796 		 */
797 		if (rl->r_len == UINT64_MAX) {
798 			uint64_t new_blksz;
799 
800 			if (zp->z_blksz > max_blksz) {
801 				ASSERT(!ISP2(zp->z_blksz));
802 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
803 			} else {
804 				new_blksz = MIN(end_size, max_blksz);
805 			}
806 			zfs_grow_blocksize(zp, new_blksz, tx);
807 			zfs_range_reduce(rl, woff, n);
808 		}
809 
810 		/*
811 		 * XXX - should we really limit each write to z_max_blksz?
812 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
813 		 */
814 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
815 
816 		if (abuf == NULL) {
817 			tx_bytes = uio->uio_resid;
818 			error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
819 			    uio, nbytes, tx);
820 			tx_bytes -= uio->uio_resid;
821 		} else {
822 			tx_bytes = nbytes;
823 			ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
824 			/*
825 			 * If this is not a full block write, but we are
826 			 * extending the file past EOF and this data starts
827 			 * block-aligned, use assign_arcbuf().  Otherwise,
828 			 * write via dmu_write().
829 			 */
830 			if (tx_bytes < max_blksz && (!write_eof ||
831 			    aiov->iov_base != abuf->b_data)) {
832 				ASSERT(xuio);
833 				dmu_write(zfsvfs->z_os, zp->z_id, woff,
834 				    aiov->iov_len, aiov->iov_base, tx);
835 				dmu_return_arcbuf(abuf);
836 				xuio_stat_wbuf_copied();
837 			} else {
838 				ASSERT(xuio || tx_bytes == max_blksz);
839 				dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
840 				    woff, abuf, tx);
841 			}
842 			ASSERT(tx_bytes <= uio->uio_resid);
843 			uioskip(uio, tx_bytes);
844 		}
845 		if (tx_bytes && vn_has_cached_data(vp)) {
846 			update_pages(vp, woff,
847 			    tx_bytes, zfsvfs->z_os, zp->z_id);
848 		}
849 
850 		/*
851 		 * If we made no progress, we're done.  If we made even
852 		 * partial progress, update the znode and ZIL accordingly.
853 		 */
854 		if (tx_bytes == 0) {
855 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
856 			    (void *)&zp->z_size, sizeof (uint64_t), tx);
857 			dmu_tx_commit(tx);
858 			ASSERT(error != 0);
859 			break;
860 		}
861 
862 		/*
863 		 * Clear Set-UID/Set-GID bits on successful write if not
864 		 * privileged and at least one of the excute bits is set.
865 		 *
866 		 * It would be nice to to this after all writes have
867 		 * been done, but that would still expose the ISUID/ISGID
868 		 * to another app after the partial write is committed.
869 		 *
870 		 * Note: we don't call zfs_fuid_map_id() here because
871 		 * user 0 is not an ephemeral uid.
872 		 */
873 		mutex_enter(&zp->z_acl_lock);
874 		if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
875 		    (S_IXUSR >> 6))) != 0 &&
876 		    (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
877 		    secpolicy_vnode_setid_retain(cr,
878 		    (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
879 			uint64_t newmode;
880 			zp->z_mode &= ~(S_ISUID | S_ISGID);
881 			newmode = zp->z_mode;
882 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
883 			    (void *)&newmode, sizeof (uint64_t), tx);
884 		}
885 		mutex_exit(&zp->z_acl_lock);
886 
887 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
888 		    B_TRUE);
889 
890 		/*
891 		 * Update the file size (zp_size) if it has changed;
892 		 * account for possible concurrent updates.
893 		 */
894 		while ((end_size = zp->z_size) < uio->uio_loffset) {
895 			(void) atomic_cas_64(&zp->z_size, end_size,
896 			    uio->uio_loffset);
897 			ASSERT(error == 0);
898 		}
899 		/*
900 		 * If we are replaying and eof is non zero then force
901 		 * the file size to the specified eof. Note, there's no
902 		 * concurrency during replay.
903 		 */
904 		if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
905 			zp->z_size = zfsvfs->z_replay_eof;
906 
907 		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
908 
909 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
910 		dmu_tx_commit(tx);
911 
912 		if (error != 0)
913 			break;
914 		ASSERT(tx_bytes == nbytes);
915 		n -= nbytes;
916 
917 		if (!xuio && n > 0)
918 			uio_prefaultpages(MIN(n, max_blksz), uio);
919 	}
920 
921 	zfs_range_unlock(rl);
922 
923 	/*
924 	 * If we're in replay mode, or we made no progress, return error.
925 	 * Otherwise, it's at least a partial write, so it's successful.
926 	 */
927 	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
928 		ZFS_EXIT(zfsvfs);
929 		return (error);
930 	}
931 
932 	if (ioflag & (FSYNC | FDSYNC) ||
933 	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
934 		zil_commit(zilog, zp->z_id);
935 
936 	ZFS_EXIT(zfsvfs);
937 	return (0);
938 }
939 
940 void
941 zfs_get_done(zgd_t *zgd, int error)
942 {
943 	znode_t *zp = zgd->zgd_private;
944 	objset_t *os = zp->z_zfsvfs->z_os;
945 
946 	if (zgd->zgd_db)
947 		dmu_buf_rele(zgd->zgd_db, zgd);
948 
949 	zfs_range_unlock(zgd->zgd_rl);
950 
951 	/*
952 	 * Release the vnode asynchronously as we currently have the
953 	 * txg stopped from syncing.
954 	 */
955 	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
956 
957 	if (error == 0 && zgd->zgd_bp)
958 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
959 
960 	kmem_free(zgd, sizeof (zgd_t));
961 }
962 
963 #ifdef DEBUG
964 static int zil_fault_io = 0;
965 #endif
966 
967 /*
968  * Get data to generate a TX_WRITE intent log record.
969  */
970 int
971 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
972 {
973 	zfsvfs_t *zfsvfs = arg;
974 	objset_t *os = zfsvfs->z_os;
975 	znode_t *zp;
976 	uint64_t object = lr->lr_foid;
977 	uint64_t offset = lr->lr_offset;
978 	uint64_t size = lr->lr_length;
979 	blkptr_t *bp = &lr->lr_blkptr;
980 	dmu_buf_t *db;
981 	zgd_t *zgd;
982 	int error = 0;
983 
984 	ASSERT(zio != NULL);
985 	ASSERT(size != 0);
986 
987 	/*
988 	 * Nothing to do if the file has been removed
989 	 */
990 	if (zfs_zget(zfsvfs, object, &zp) != 0)
991 		return (ENOENT);
992 	if (zp->z_unlinked) {
993 		/*
994 		 * Release the vnode asynchronously as we currently have the
995 		 * txg stopped from syncing.
996 		 */
997 		VN_RELE_ASYNC(ZTOV(zp),
998 		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
999 		return (ENOENT);
1000 	}
1001 
1002 	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1003 	zgd->zgd_zilog = zfsvfs->z_log;
1004 	zgd->zgd_private = zp;
1005 
1006 	/*
1007 	 * Write records come in two flavors: immediate and indirect.
1008 	 * For small writes it's cheaper to store the data with the
1009 	 * log record (immediate); for large writes it's cheaper to
1010 	 * sync the data and get a pointer to it (indirect) so that
1011 	 * we don't have to write the data twice.
1012 	 */
1013 	if (buf != NULL) { /* immediate write */
1014 		zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1015 		/* test for truncation needs to be done while range locked */
1016 		if (offset >= zp->z_size) {
1017 			error = ENOENT;
1018 		} else {
1019 			error = dmu_read(os, object, offset, size, buf,
1020 			    DMU_READ_NO_PREFETCH);
1021 		}
1022 		ASSERT(error == 0 || error == ENOENT);
1023 	} else { /* indirect write */
1024 		/*
1025 		 * Have to lock the whole block to ensure when it's
1026 		 * written out and it's checksum is being calculated
1027 		 * that no one can change the data. We need to re-check
1028 		 * blocksize after we get the lock in case it's changed!
1029 		 */
1030 		for (;;) {
1031 			uint64_t blkoff;
1032 			size = zp->z_blksz;
1033 			blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1034 			offset -= blkoff;
1035 			zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1036 			    RL_READER);
1037 			if (zp->z_blksz == size)
1038 				break;
1039 			offset += blkoff;
1040 			zfs_range_unlock(zgd->zgd_rl);
1041 		}
1042 		/* test for truncation needs to be done while range locked */
1043 		if (lr->lr_offset >= zp->z_size)
1044 			error = ENOENT;
1045 #ifdef DEBUG
1046 		if (zil_fault_io) {
1047 			error = EIO;
1048 			zil_fault_io = 0;
1049 		}
1050 #endif
1051 		if (error == 0)
1052 			error = dmu_buf_hold(os, object, offset, zgd, &db,
1053 			    DMU_READ_NO_PREFETCH);
1054 
1055 		if (error == 0) {
1056 			zgd->zgd_db = db;
1057 			zgd->zgd_bp = bp;
1058 
1059 			ASSERT(db->db_offset == offset);
1060 			ASSERT(db->db_size == size);
1061 
1062 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1063 			    zfs_get_done, zgd);
1064 			ASSERT(error || lr->lr_length <= zp->z_blksz);
1065 
1066 			/*
1067 			 * On success, we need to wait for the write I/O
1068 			 * initiated by dmu_sync() to complete before we can
1069 			 * release this dbuf.  We will finish everything up
1070 			 * in the zfs_get_done() callback.
1071 			 */
1072 			if (error == 0)
1073 				return (0);
1074 
1075 			if (error == EALREADY) {
1076 				lr->lr_common.lrc_txtype = TX_WRITE2;
1077 				error = 0;
1078 			}
1079 		}
1080 	}
1081 
1082 	zfs_get_done(zgd, error);
1083 
1084 	return (error);
1085 }
1086 
1087 /*ARGSUSED*/
1088 static int
1089 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1090     caller_context_t *ct)
1091 {
1092 	znode_t *zp = VTOZ(vp);
1093 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1094 	int error;
1095 
1096 	ZFS_ENTER(zfsvfs);
1097 	ZFS_VERIFY_ZP(zp);
1098 
1099 	if (flag & V_ACE_MASK)
1100 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1101 	else
1102 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
1103 
1104 	ZFS_EXIT(zfsvfs);
1105 	return (error);
1106 }
1107 
1108 /*
1109  * If vnode is for a device return a specfs vnode instead.
1110  */
1111 static int
1112 specvp_check(vnode_t **vpp, cred_t *cr)
1113 {
1114 	int error = 0;
1115 
1116 	if (IS_DEVVP(*vpp)) {
1117 		struct vnode *svp;
1118 
1119 		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1120 		VN_RELE(*vpp);
1121 		if (svp == NULL)
1122 			error = ENOSYS;
1123 		*vpp = svp;
1124 	}
1125 	return (error);
1126 }
1127 
1128 
1129 /*
1130  * Lookup an entry in a directory, or an extended attribute directory.
1131  * If it exists, return a held vnode reference for it.
1132  *
1133  *	IN:	dvp	- vnode of directory to search.
1134  *		nm	- name of entry to lookup.
1135  *		pnp	- full pathname to lookup [UNUSED].
1136  *		flags	- LOOKUP_XATTR set if looking for an attribute.
1137  *		rdir	- root directory vnode [UNUSED].
1138  *		cr	- credentials of caller.
1139  *		ct	- caller context
1140  *		direntflags - directory lookup flags
1141  *		realpnp - returned pathname.
1142  *
1143  *	OUT:	vpp	- vnode of located entry, NULL if not found.
1144  *
1145  *	RETURN:	0 if success
1146  *		error code if failure
1147  *
1148  * Timestamps:
1149  *	NA
1150  */
1151 /* ARGSUSED */
1152 static int
1153 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1154     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1155     int *direntflags, pathname_t *realpnp)
1156 {
1157 	znode_t *zdp = VTOZ(dvp);
1158 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1159 	int	error = 0;
1160 
1161 	/* fast path */
1162 	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1163 
1164 		if (dvp->v_type != VDIR) {
1165 			return (ENOTDIR);
1166 		} else if (zdp->z_sa_hdl == NULL) {
1167 			return (EIO);
1168 		}
1169 
1170 		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1171 			error = zfs_fastaccesschk_execute(zdp, cr);
1172 			if (!error) {
1173 				*vpp = dvp;
1174 				VN_HOLD(*vpp);
1175 				return (0);
1176 			}
1177 			return (error);
1178 		} else {
1179 			vnode_t *tvp = dnlc_lookup(dvp, nm);
1180 
1181 			if (tvp) {
1182 				error = zfs_fastaccesschk_execute(zdp, cr);
1183 				if (error) {
1184 					VN_RELE(tvp);
1185 					return (error);
1186 				}
1187 				if (tvp == DNLC_NO_VNODE) {
1188 					VN_RELE(tvp);
1189 					return (ENOENT);
1190 				} else {
1191 					*vpp = tvp;
1192 					return (specvp_check(vpp, cr));
1193 				}
1194 			}
1195 		}
1196 	}
1197 
1198 	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1199 
1200 	ZFS_ENTER(zfsvfs);
1201 	ZFS_VERIFY_ZP(zdp);
1202 
1203 	*vpp = NULL;
1204 
1205 	if (flags & LOOKUP_XATTR) {
1206 		/*
1207 		 * If the xattr property is off, refuse the lookup request.
1208 		 */
1209 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1210 			ZFS_EXIT(zfsvfs);
1211 			return (EINVAL);
1212 		}
1213 
1214 		/*
1215 		 * We don't allow recursive attributes..
1216 		 * Maybe someday we will.
1217 		 */
1218 		if (zdp->z_pflags & ZFS_XATTR) {
1219 			ZFS_EXIT(zfsvfs);
1220 			return (EINVAL);
1221 		}
1222 
1223 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1224 			ZFS_EXIT(zfsvfs);
1225 			return (error);
1226 		}
1227 
1228 		/*
1229 		 * Do we have permission to get into attribute directory?
1230 		 */
1231 
1232 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1233 		    B_FALSE, cr)) {
1234 			VN_RELE(*vpp);
1235 			*vpp = NULL;
1236 		}
1237 
1238 		ZFS_EXIT(zfsvfs);
1239 		return (error);
1240 	}
1241 
1242 	if (dvp->v_type != VDIR) {
1243 		ZFS_EXIT(zfsvfs);
1244 		return (ENOTDIR);
1245 	}
1246 
1247 	/*
1248 	 * Check accessibility of directory.
1249 	 */
1250 
1251 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1252 		ZFS_EXIT(zfsvfs);
1253 		return (error);
1254 	}
1255 
1256 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1257 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1258 		ZFS_EXIT(zfsvfs);
1259 		return (EILSEQ);
1260 	}
1261 
1262 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1263 	if (error == 0)
1264 		error = specvp_check(vpp, cr);
1265 
1266 	ZFS_EXIT(zfsvfs);
1267 	return (error);
1268 }
1269 
1270 /*
1271  * Attempt to create a new entry in a directory.  If the entry
1272  * already exists, truncate the file if permissible, else return
1273  * an error.  Return the vp of the created or trunc'd file.
1274  *
1275  *	IN:	dvp	- vnode of directory to put new file entry in.
1276  *		name	- name of new file entry.
1277  *		vap	- attributes of new file.
1278  *		excl	- flag indicating exclusive or non-exclusive mode.
1279  *		mode	- mode to open file with.
1280  *		cr	- credentials of caller.
1281  *		flag	- large file flag [UNUSED].
1282  *		ct	- caller context
1283  *		vsecp 	- ACL to be set
1284  *
1285  *	OUT:	vpp	- vnode of created or trunc'd entry.
1286  *
1287  *	RETURN:	0 if success
1288  *		error code if failure
1289  *
1290  * Timestamps:
1291  *	dvp - ctime|mtime updated if new entry created
1292  *	 vp - ctime|mtime always, atime if new
1293  */
1294 
1295 /* ARGSUSED */
1296 static int
1297 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1298     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1299     vsecattr_t *vsecp)
1300 {
1301 	znode_t		*zp, *dzp = VTOZ(dvp);
1302 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1303 	zilog_t		*zilog;
1304 	objset_t	*os;
1305 	zfs_dirlock_t	*dl;
1306 	dmu_tx_t	*tx;
1307 	int		error;
1308 	ksid_t		*ksid;
1309 	uid_t		uid;
1310 	gid_t		gid = crgetgid(cr);
1311 	zfs_acl_ids_t   acl_ids;
1312 	boolean_t	fuid_dirtied;
1313 	boolean_t	have_acl = B_FALSE;
1314 
1315 	/*
1316 	 * If we have an ephemeral id, ACL, or XVATTR then
1317 	 * make sure file system is at proper version
1318 	 */
1319 
1320 	ksid = crgetsid(cr, KSID_OWNER);
1321 	if (ksid)
1322 		uid = ksid_getid(ksid);
1323 	else
1324 		uid = crgetuid(cr);
1325 
1326 	if (zfsvfs->z_use_fuids == B_FALSE &&
1327 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1328 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1329 		return (EINVAL);
1330 
1331 	ZFS_ENTER(zfsvfs);
1332 	ZFS_VERIFY_ZP(dzp);
1333 	os = zfsvfs->z_os;
1334 	zilog = zfsvfs->z_log;
1335 
1336 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1337 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1338 		ZFS_EXIT(zfsvfs);
1339 		return (EILSEQ);
1340 	}
1341 
1342 	if (vap->va_mask & AT_XVATTR) {
1343 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1344 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1345 			ZFS_EXIT(zfsvfs);
1346 			return (error);
1347 		}
1348 	}
1349 top:
1350 	*vpp = NULL;
1351 
1352 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1353 		vap->va_mode &= ~VSVTX;
1354 
1355 	if (*name == '\0') {
1356 		/*
1357 		 * Null component name refers to the directory itself.
1358 		 */
1359 		VN_HOLD(dvp);
1360 		zp = dzp;
1361 		dl = NULL;
1362 		error = 0;
1363 	} else {
1364 		/* possible VN_HOLD(zp) */
1365 		int zflg = 0;
1366 
1367 		if (flag & FIGNORECASE)
1368 			zflg |= ZCILOOK;
1369 
1370 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1371 		    NULL, NULL);
1372 		if (error) {
1373 			if (have_acl)
1374 				zfs_acl_ids_free(&acl_ids);
1375 			if (strcmp(name, "..") == 0)
1376 				error = EISDIR;
1377 			ZFS_EXIT(zfsvfs);
1378 			return (error);
1379 		}
1380 	}
1381 
1382 	if (zp == NULL) {
1383 		uint64_t txtype;
1384 
1385 		/*
1386 		 * Create a new file object and update the directory
1387 		 * to reference it.
1388 		 */
1389 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1390 			if (have_acl)
1391 				zfs_acl_ids_free(&acl_ids);
1392 			goto out;
1393 		}
1394 
1395 		/*
1396 		 * We only support the creation of regular files in
1397 		 * extended attribute directories.
1398 		 */
1399 
1400 		if ((dzp->z_pflags & ZFS_XATTR) &&
1401 		    (vap->va_type != VREG)) {
1402 			if (have_acl)
1403 				zfs_acl_ids_free(&acl_ids);
1404 			error = EINVAL;
1405 			goto out;
1406 		}
1407 
1408 		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1409 		    cr, vsecp, &acl_ids)) != 0)
1410 			goto out;
1411 		have_acl = B_TRUE;
1412 
1413 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1414 			zfs_acl_ids_free(&acl_ids);
1415 			error = EDQUOT;
1416 			goto out;
1417 		}
1418 
1419 		tx = dmu_tx_create(os);
1420 
1421 		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1422 		    ZFS_SA_BASE_ATTR_SIZE);
1423 
1424 		fuid_dirtied = zfsvfs->z_fuid_dirty;
1425 		if (fuid_dirtied)
1426 			zfs_fuid_txhold(zfsvfs, tx);
1427 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1428 		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1429 		if (!zfsvfs->z_use_sa &&
1430 		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1431 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1432 			    0, acl_ids.z_aclp->z_acl_bytes);
1433 		}
1434 		error = dmu_tx_assign(tx, TXG_NOWAIT);
1435 		if (error) {
1436 			zfs_dirent_unlock(dl);
1437 			if (error == ERESTART) {
1438 				dmu_tx_wait(tx);
1439 				dmu_tx_abort(tx);
1440 				goto top;
1441 			}
1442 			zfs_acl_ids_free(&acl_ids);
1443 			dmu_tx_abort(tx);
1444 			ZFS_EXIT(zfsvfs);
1445 			return (error);
1446 		}
1447 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1448 
1449 		if (fuid_dirtied)
1450 			zfs_fuid_sync(zfsvfs, tx);
1451 
1452 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1453 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1454 		if (flag & FIGNORECASE)
1455 			txtype |= TX_CI;
1456 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1457 		    vsecp, acl_ids.z_fuidp, vap);
1458 		zfs_acl_ids_free(&acl_ids);
1459 		dmu_tx_commit(tx);
1460 	} else {
1461 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1462 
1463 		if (have_acl)
1464 			zfs_acl_ids_free(&acl_ids);
1465 		have_acl = B_FALSE;
1466 
1467 		/*
1468 		 * A directory entry already exists for this name.
1469 		 */
1470 		/*
1471 		 * Can't truncate an existing file if in exclusive mode.
1472 		 */
1473 		if (excl == EXCL) {
1474 			error = EEXIST;
1475 			goto out;
1476 		}
1477 		/*
1478 		 * Can't open a directory for writing.
1479 		 */
1480 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1481 			error = EISDIR;
1482 			goto out;
1483 		}
1484 		/*
1485 		 * Verify requested access to file.
1486 		 */
1487 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1488 			goto out;
1489 		}
1490 
1491 		mutex_enter(&dzp->z_lock);
1492 		dzp->z_seq++;
1493 		mutex_exit(&dzp->z_lock);
1494 
1495 		/*
1496 		 * Truncate regular files if requested.
1497 		 */
1498 		if ((ZTOV(zp)->v_type == VREG) &&
1499 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1500 			/* we can't hold any locks when calling zfs_freesp() */
1501 			zfs_dirent_unlock(dl);
1502 			dl = NULL;
1503 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1504 			if (error == 0) {
1505 				vnevent_create(ZTOV(zp), ct);
1506 			}
1507 		}
1508 	}
1509 out:
1510 
1511 	if (dl)
1512 		zfs_dirent_unlock(dl);
1513 
1514 	if (error) {
1515 		if (zp)
1516 			VN_RELE(ZTOV(zp));
1517 	} else {
1518 		*vpp = ZTOV(zp);
1519 		error = specvp_check(vpp, cr);
1520 	}
1521 
1522 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1523 		zil_commit(zilog, 0);
1524 
1525 	ZFS_EXIT(zfsvfs);
1526 	return (error);
1527 }
1528 
1529 /*
1530  * Remove an entry from a directory.
1531  *
1532  *	IN:	dvp	- vnode of directory to remove entry from.
1533  *		name	- name of entry to remove.
1534  *		cr	- credentials of caller.
1535  *		ct	- caller context
1536  *		flags	- case flags
1537  *
1538  *	RETURN:	0 if success
1539  *		error code if failure
1540  *
1541  * Timestamps:
1542  *	dvp - ctime|mtime
1543  *	 vp - ctime (if nlink > 0)
1544  */
1545 
1546 uint64_t null_xattr = 0;
1547 
1548 /*ARGSUSED*/
1549 static int
1550 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1551     int flags)
1552 {
1553 	znode_t		*zp, *dzp = VTOZ(dvp);
1554 	znode_t		*xzp;
1555 	vnode_t		*vp;
1556 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1557 	zilog_t		*zilog;
1558 	uint64_t	acl_obj, xattr_obj;
1559 	uint64_t 	xattr_obj_unlinked = 0;
1560 	uint64_t	obj = 0;
1561 	zfs_dirlock_t	*dl;
1562 	dmu_tx_t	*tx;
1563 	boolean_t	may_delete_now, delete_now = FALSE;
1564 	boolean_t	unlinked, toobig = FALSE;
1565 	uint64_t	txtype;
1566 	pathname_t	*realnmp = NULL;
1567 	pathname_t	realnm;
1568 	int		error;
1569 	int		zflg = ZEXISTS;
1570 
1571 	ZFS_ENTER(zfsvfs);
1572 	ZFS_VERIFY_ZP(dzp);
1573 	zilog = zfsvfs->z_log;
1574 
1575 	if (flags & FIGNORECASE) {
1576 		zflg |= ZCILOOK;
1577 		pn_alloc(&realnm);
1578 		realnmp = &realnm;
1579 	}
1580 
1581 top:
1582 	xattr_obj = 0;
1583 	xzp = NULL;
1584 	/*
1585 	 * Attempt to lock directory; fail if entry doesn't exist.
1586 	 */
1587 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1588 	    NULL, realnmp)) {
1589 		if (realnmp)
1590 			pn_free(realnmp);
1591 		ZFS_EXIT(zfsvfs);
1592 		return (error);
1593 	}
1594 
1595 	vp = ZTOV(zp);
1596 
1597 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1598 		goto out;
1599 	}
1600 
1601 	/*
1602 	 * Need to use rmdir for removing directories.
1603 	 */
1604 	if (vp->v_type == VDIR) {
1605 		error = EPERM;
1606 		goto out;
1607 	}
1608 
1609 	vnevent_remove(vp, dvp, name, ct);
1610 
1611 	if (realnmp)
1612 		dnlc_remove(dvp, realnmp->pn_buf);
1613 	else
1614 		dnlc_remove(dvp, name);
1615 
1616 	mutex_enter(&vp->v_lock);
1617 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1618 	mutex_exit(&vp->v_lock);
1619 
1620 	/*
1621 	 * We may delete the znode now, or we may put it in the unlinked set;
1622 	 * it depends on whether we're the last link, and on whether there are
1623 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1624 	 * allow for either case.
1625 	 */
1626 	obj = zp->z_id;
1627 	tx = dmu_tx_create(zfsvfs->z_os);
1628 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1629 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1630 	zfs_sa_upgrade_txholds(tx, zp);
1631 	zfs_sa_upgrade_txholds(tx, dzp);
1632 	if (may_delete_now) {
1633 		toobig =
1634 		    zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1635 		/* if the file is too big, only hold_free a token amount */
1636 		dmu_tx_hold_free(tx, zp->z_id, 0,
1637 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1638 	}
1639 
1640 	/* are there any extended attributes? */
1641 	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1642 	    &xattr_obj, sizeof (xattr_obj));
1643 	if (error == 0 && xattr_obj) {
1644 		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1645 		ASSERT0(error);
1646 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1647 		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1648 	}
1649 
1650 	mutex_enter(&zp->z_lock);
1651 	if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1652 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1653 	mutex_exit(&zp->z_lock);
1654 
1655 	/* charge as an update -- would be nice not to charge at all */
1656 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1657 
1658 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1659 	if (error) {
1660 		zfs_dirent_unlock(dl);
1661 		VN_RELE(vp);
1662 		if (xzp)
1663 			VN_RELE(ZTOV(xzp));
1664 		if (error == ERESTART) {
1665 			dmu_tx_wait(tx);
1666 			dmu_tx_abort(tx);
1667 			goto top;
1668 		}
1669 		if (realnmp)
1670 			pn_free(realnmp);
1671 		dmu_tx_abort(tx);
1672 		ZFS_EXIT(zfsvfs);
1673 		return (error);
1674 	}
1675 
1676 	/*
1677 	 * Remove the directory entry.
1678 	 */
1679 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1680 
1681 	if (error) {
1682 		dmu_tx_commit(tx);
1683 		goto out;
1684 	}
1685 
1686 	if (unlinked) {
1687 
1688 		/*
1689 		 * Hold z_lock so that we can make sure that the ACL obj
1690 		 * hasn't changed.  Could have been deleted due to
1691 		 * zfs_sa_upgrade().
1692 		 */
1693 		mutex_enter(&zp->z_lock);
1694 		mutex_enter(&vp->v_lock);
1695 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1696 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1697 		delete_now = may_delete_now && !toobig &&
1698 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1699 		    xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1700 		    acl_obj;
1701 		mutex_exit(&vp->v_lock);
1702 	}
1703 
1704 	if (delete_now) {
1705 		if (xattr_obj_unlinked) {
1706 			ASSERT3U(xzp->z_links, ==, 2);
1707 			mutex_enter(&xzp->z_lock);
1708 			xzp->z_unlinked = 1;
1709 			xzp->z_links = 0;
1710 			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1711 			    &xzp->z_links, sizeof (xzp->z_links), tx);
1712 			ASSERT3U(error,  ==,  0);
1713 			mutex_exit(&xzp->z_lock);
1714 			zfs_unlinked_add(xzp, tx);
1715 
1716 			if (zp->z_is_sa)
1717 				error = sa_remove(zp->z_sa_hdl,
1718 				    SA_ZPL_XATTR(zfsvfs), tx);
1719 			else
1720 				error = sa_update(zp->z_sa_hdl,
1721 				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
1722 				    sizeof (uint64_t), tx);
1723 			ASSERT0(error);
1724 		}
1725 		mutex_enter(&vp->v_lock);
1726 		vp->v_count--;
1727 		ASSERT0(vp->v_count);
1728 		mutex_exit(&vp->v_lock);
1729 		mutex_exit(&zp->z_lock);
1730 		zfs_znode_delete(zp, tx);
1731 	} else if (unlinked) {
1732 		mutex_exit(&zp->z_lock);
1733 		zfs_unlinked_add(zp, tx);
1734 	}
1735 
1736 	txtype = TX_REMOVE;
1737 	if (flags & FIGNORECASE)
1738 		txtype |= TX_CI;
1739 	zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1740 
1741 	dmu_tx_commit(tx);
1742 out:
1743 	if (realnmp)
1744 		pn_free(realnmp);
1745 
1746 	zfs_dirent_unlock(dl);
1747 
1748 	if (!delete_now)
1749 		VN_RELE(vp);
1750 	if (xzp)
1751 		VN_RELE(ZTOV(xzp));
1752 
1753 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1754 		zil_commit(zilog, 0);
1755 
1756 	ZFS_EXIT(zfsvfs);
1757 	return (error);
1758 }
1759 
1760 /*
1761  * Create a new directory and insert it into dvp using the name
1762  * provided.  Return a pointer to the inserted directory.
1763  *
1764  *	IN:	dvp	- vnode of directory to add subdir to.
1765  *		dirname	- name of new directory.
1766  *		vap	- attributes of new directory.
1767  *		cr	- credentials of caller.
1768  *		ct	- caller context
1769  *		vsecp	- ACL to be set
1770  *
1771  *	OUT:	vpp	- vnode of created directory.
1772  *
1773  *	RETURN:	0 if success
1774  *		error code if failure
1775  *
1776  * Timestamps:
1777  *	dvp - ctime|mtime updated
1778  *	 vp - ctime|mtime|atime updated
1779  */
1780 /*ARGSUSED*/
1781 static int
1782 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1783     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1784 {
1785 	znode_t		*zp, *dzp = VTOZ(dvp);
1786 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1787 	zilog_t		*zilog;
1788 	zfs_dirlock_t	*dl;
1789 	uint64_t	txtype;
1790 	dmu_tx_t	*tx;
1791 	int		error;
1792 	int		zf = ZNEW;
1793 	ksid_t		*ksid;
1794 	uid_t		uid;
1795 	gid_t		gid = crgetgid(cr);
1796 	zfs_acl_ids_t   acl_ids;
1797 	boolean_t	fuid_dirtied;
1798 
1799 	ASSERT(vap->va_type == VDIR);
1800 
1801 	/*
1802 	 * If we have an ephemeral id, ACL, or XVATTR then
1803 	 * make sure file system is at proper version
1804 	 */
1805 
1806 	ksid = crgetsid(cr, KSID_OWNER);
1807 	if (ksid)
1808 		uid = ksid_getid(ksid);
1809 	else
1810 		uid = crgetuid(cr);
1811 	if (zfsvfs->z_use_fuids == B_FALSE &&
1812 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1813 	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1814 		return (EINVAL);
1815 
1816 	ZFS_ENTER(zfsvfs);
1817 	ZFS_VERIFY_ZP(dzp);
1818 	zilog = zfsvfs->z_log;
1819 
1820 	if (dzp->z_pflags & ZFS_XATTR) {
1821 		ZFS_EXIT(zfsvfs);
1822 		return (EINVAL);
1823 	}
1824 
1825 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1826 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1827 		ZFS_EXIT(zfsvfs);
1828 		return (EILSEQ);
1829 	}
1830 	if (flags & FIGNORECASE)
1831 		zf |= ZCILOOK;
1832 
1833 	if (vap->va_mask & AT_XVATTR) {
1834 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1835 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1836 			ZFS_EXIT(zfsvfs);
1837 			return (error);
1838 		}
1839 	}
1840 
1841 	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1842 	    vsecp, &acl_ids)) != 0) {
1843 		ZFS_EXIT(zfsvfs);
1844 		return (error);
1845 	}
1846 	/*
1847 	 * First make sure the new directory doesn't exist.
1848 	 *
1849 	 * Existence is checked first to make sure we don't return
1850 	 * EACCES instead of EEXIST which can cause some applications
1851 	 * to fail.
1852 	 */
1853 top:
1854 	*vpp = NULL;
1855 
1856 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1857 	    NULL, NULL)) {
1858 		zfs_acl_ids_free(&acl_ids);
1859 		ZFS_EXIT(zfsvfs);
1860 		return (error);
1861 	}
1862 
1863 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1864 		zfs_acl_ids_free(&acl_ids);
1865 		zfs_dirent_unlock(dl);
1866 		ZFS_EXIT(zfsvfs);
1867 		return (error);
1868 	}
1869 
1870 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1871 		zfs_acl_ids_free(&acl_ids);
1872 		zfs_dirent_unlock(dl);
1873 		ZFS_EXIT(zfsvfs);
1874 		return (EDQUOT);
1875 	}
1876 
1877 	/*
1878 	 * Add a new entry to the directory.
1879 	 */
1880 	tx = dmu_tx_create(zfsvfs->z_os);
1881 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1882 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1883 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1884 	if (fuid_dirtied)
1885 		zfs_fuid_txhold(zfsvfs, tx);
1886 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1887 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1888 		    acl_ids.z_aclp->z_acl_bytes);
1889 	}
1890 
1891 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1892 	    ZFS_SA_BASE_ATTR_SIZE);
1893 
1894 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1895 	if (error) {
1896 		zfs_dirent_unlock(dl);
1897 		if (error == ERESTART) {
1898 			dmu_tx_wait(tx);
1899 			dmu_tx_abort(tx);
1900 			goto top;
1901 		}
1902 		zfs_acl_ids_free(&acl_ids);
1903 		dmu_tx_abort(tx);
1904 		ZFS_EXIT(zfsvfs);
1905 		return (error);
1906 	}
1907 
1908 	/*
1909 	 * Create new node.
1910 	 */
1911 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1912 
1913 	if (fuid_dirtied)
1914 		zfs_fuid_sync(zfsvfs, tx);
1915 
1916 	/*
1917 	 * Now put new name in parent dir.
1918 	 */
1919 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1920 
1921 	*vpp = ZTOV(zp);
1922 
1923 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1924 	if (flags & FIGNORECASE)
1925 		txtype |= TX_CI;
1926 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
1927 	    acl_ids.z_fuidp, vap);
1928 
1929 	zfs_acl_ids_free(&acl_ids);
1930 
1931 	dmu_tx_commit(tx);
1932 
1933 	zfs_dirent_unlock(dl);
1934 
1935 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1936 		zil_commit(zilog, 0);
1937 
1938 	ZFS_EXIT(zfsvfs);
1939 	return (0);
1940 }
1941 
1942 /*
1943  * Remove a directory subdir entry.  If the current working
1944  * directory is the same as the subdir to be removed, the
1945  * remove will fail.
1946  *
1947  *	IN:	dvp	- vnode of directory to remove from.
1948  *		name	- name of directory to be removed.
1949  *		cwd	- vnode of current working directory.
1950  *		cr	- credentials of caller.
1951  *		ct	- caller context
1952  *		flags	- case flags
1953  *
1954  *	RETURN:	0 if success
1955  *		error code if failure
1956  *
1957  * Timestamps:
1958  *	dvp - ctime|mtime updated
1959  */
1960 /*ARGSUSED*/
1961 static int
1962 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1963     caller_context_t *ct, int flags)
1964 {
1965 	znode_t		*dzp = VTOZ(dvp);
1966 	znode_t		*zp;
1967 	vnode_t		*vp;
1968 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1969 	zilog_t		*zilog;
1970 	zfs_dirlock_t	*dl;
1971 	dmu_tx_t	*tx;
1972 	int		error;
1973 	int		zflg = ZEXISTS;
1974 
1975 	ZFS_ENTER(zfsvfs);
1976 	ZFS_VERIFY_ZP(dzp);
1977 	zilog = zfsvfs->z_log;
1978 
1979 	if (flags & FIGNORECASE)
1980 		zflg |= ZCILOOK;
1981 top:
1982 	zp = NULL;
1983 
1984 	/*
1985 	 * Attempt to lock directory; fail if entry doesn't exist.
1986 	 */
1987 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1988 	    NULL, NULL)) {
1989 		ZFS_EXIT(zfsvfs);
1990 		return (error);
1991 	}
1992 
1993 	vp = ZTOV(zp);
1994 
1995 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1996 		goto out;
1997 	}
1998 
1999 	if (vp->v_type != VDIR) {
2000 		error = ENOTDIR;
2001 		goto out;
2002 	}
2003 
2004 	if (vp == cwd) {
2005 		error = EINVAL;
2006 		goto out;
2007 	}
2008 
2009 	vnevent_rmdir(vp, dvp, name, ct);
2010 
2011 	/*
2012 	 * Grab a lock on the directory to make sure that noone is
2013 	 * trying to add (or lookup) entries while we are removing it.
2014 	 */
2015 	rw_enter(&zp->z_name_lock, RW_WRITER);
2016 
2017 	/*
2018 	 * Grab a lock on the parent pointer to make sure we play well
2019 	 * with the treewalk and directory rename code.
2020 	 */
2021 	rw_enter(&zp->z_parent_lock, RW_WRITER);
2022 
2023 	tx = dmu_tx_create(zfsvfs->z_os);
2024 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2025 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2026 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2027 	zfs_sa_upgrade_txholds(tx, zp);
2028 	zfs_sa_upgrade_txholds(tx, dzp);
2029 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2030 	if (error) {
2031 		rw_exit(&zp->z_parent_lock);
2032 		rw_exit(&zp->z_name_lock);
2033 		zfs_dirent_unlock(dl);
2034 		VN_RELE(vp);
2035 		if (error == ERESTART) {
2036 			dmu_tx_wait(tx);
2037 			dmu_tx_abort(tx);
2038 			goto top;
2039 		}
2040 		dmu_tx_abort(tx);
2041 		ZFS_EXIT(zfsvfs);
2042 		return (error);
2043 	}
2044 
2045 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2046 
2047 	if (error == 0) {
2048 		uint64_t txtype = TX_RMDIR;
2049 		if (flags & FIGNORECASE)
2050 			txtype |= TX_CI;
2051 		zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2052 	}
2053 
2054 	dmu_tx_commit(tx);
2055 
2056 	rw_exit(&zp->z_parent_lock);
2057 	rw_exit(&zp->z_name_lock);
2058 out:
2059 	zfs_dirent_unlock(dl);
2060 
2061 	VN_RELE(vp);
2062 
2063 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2064 		zil_commit(zilog, 0);
2065 
2066 	ZFS_EXIT(zfsvfs);
2067 	return (error);
2068 }
2069 
2070 /*
2071  * Read as many directory entries as will fit into the provided
2072  * buffer from the given directory cursor position (specified in
2073  * the uio structure.
2074  *
2075  *	IN:	vp	- vnode of directory to read.
2076  *		uio	- structure supplying read location, range info,
2077  *			  and return buffer.
2078  *		cr	- credentials of caller.
2079  *		ct	- caller context
2080  *		flags	- case flags
2081  *
2082  *	OUT:	uio	- updated offset and range, buffer filled.
2083  *		eofp	- set to true if end-of-file detected.
2084  *
2085  *	RETURN:	0 if success
2086  *		error code if failure
2087  *
2088  * Timestamps:
2089  *	vp - atime updated
2090  *
2091  * Note that the low 4 bits of the cookie returned by zap is always zero.
2092  * This allows us to use the low range for "special" directory entries:
2093  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2094  * we use the offset 2 for the '.zfs' directory.
2095  */
2096 /* ARGSUSED */
2097 static int
2098 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2099     caller_context_t *ct, int flags)
2100 {
2101 	znode_t		*zp = VTOZ(vp);
2102 	iovec_t		*iovp;
2103 	edirent_t	*eodp;
2104 	dirent64_t	*odp;
2105 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2106 	objset_t	*os;
2107 	caddr_t		outbuf;
2108 	size_t		bufsize;
2109 	zap_cursor_t	zc;
2110 	zap_attribute_t	zap;
2111 	uint_t		bytes_wanted;
2112 	uint64_t	offset; /* must be unsigned; checks for < 1 */
2113 	uint64_t	parent;
2114 	int		local_eof;
2115 	int		outcount;
2116 	int		error;
2117 	uint8_t		prefetch;
2118 	boolean_t	check_sysattrs;
2119 
2120 	ZFS_ENTER(zfsvfs);
2121 	ZFS_VERIFY_ZP(zp);
2122 
2123 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2124 	    &parent, sizeof (parent))) != 0) {
2125 		ZFS_EXIT(zfsvfs);
2126 		return (error);
2127 	}
2128 
2129 	/*
2130 	 * If we are not given an eof variable,
2131 	 * use a local one.
2132 	 */
2133 	if (eofp == NULL)
2134 		eofp = &local_eof;
2135 
2136 	/*
2137 	 * Check for valid iov_len.
2138 	 */
2139 	if (uio->uio_iov->iov_len <= 0) {
2140 		ZFS_EXIT(zfsvfs);
2141 		return (EINVAL);
2142 	}
2143 
2144 	/*
2145 	 * Quit if directory has been removed (posix)
2146 	 */
2147 	if ((*eofp = zp->z_unlinked) != 0) {
2148 		ZFS_EXIT(zfsvfs);
2149 		return (0);
2150 	}
2151 
2152 	error = 0;
2153 	os = zfsvfs->z_os;
2154 	offset = uio->uio_loffset;
2155 	prefetch = zp->z_zn_prefetch;
2156 
2157 	/*
2158 	 * Initialize the iterator cursor.
2159 	 */
2160 	if (offset <= 3) {
2161 		/*
2162 		 * Start iteration from the beginning of the directory.
2163 		 */
2164 		zap_cursor_init(&zc, os, zp->z_id);
2165 	} else {
2166 		/*
2167 		 * The offset is a serialized cursor.
2168 		 */
2169 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2170 	}
2171 
2172 	/*
2173 	 * Get space to change directory entries into fs independent format.
2174 	 */
2175 	iovp = uio->uio_iov;
2176 	bytes_wanted = iovp->iov_len;
2177 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2178 		bufsize = bytes_wanted;
2179 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
2180 		odp = (struct dirent64 *)outbuf;
2181 	} else {
2182 		bufsize = bytes_wanted;
2183 		odp = (struct dirent64 *)iovp->iov_base;
2184 	}
2185 	eodp = (struct edirent *)odp;
2186 
2187 	/*
2188 	 * If this VFS supports the system attribute view interface; and
2189 	 * we're looking at an extended attribute directory; and we care
2190 	 * about normalization conflicts on this vfs; then we must check
2191 	 * for normalization conflicts with the sysattr name space.
2192 	 */
2193 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2194 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2195 	    (flags & V_RDDIR_ENTFLAGS);
2196 
2197 	/*
2198 	 * Transform to file-system independent format
2199 	 */
2200 	outcount = 0;
2201 	while (outcount < bytes_wanted) {
2202 		ino64_t objnum;
2203 		ushort_t reclen;
2204 		off64_t *next = NULL;
2205 
2206 		/*
2207 		 * Special case `.', `..', and `.zfs'.
2208 		 */
2209 		if (offset == 0) {
2210 			(void) strcpy(zap.za_name, ".");
2211 			zap.za_normalization_conflict = 0;
2212 			objnum = zp->z_id;
2213 		} else if (offset == 1) {
2214 			(void) strcpy(zap.za_name, "..");
2215 			zap.za_normalization_conflict = 0;
2216 			objnum = parent;
2217 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2218 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2219 			zap.za_normalization_conflict = 0;
2220 			objnum = ZFSCTL_INO_ROOT;
2221 		} else {
2222 			/*
2223 			 * Grab next entry.
2224 			 */
2225 			if (error = zap_cursor_retrieve(&zc, &zap)) {
2226 				if ((*eofp = (error == ENOENT)) != 0)
2227 					break;
2228 				else
2229 					goto update;
2230 			}
2231 
2232 			if (zap.za_integer_length != 8 ||
2233 			    zap.za_num_integers != 1) {
2234 				cmn_err(CE_WARN, "zap_readdir: bad directory "
2235 				    "entry, obj = %lld, offset = %lld\n",
2236 				    (u_longlong_t)zp->z_id,
2237 				    (u_longlong_t)offset);
2238 				error = ENXIO;
2239 				goto update;
2240 			}
2241 
2242 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2243 			/*
2244 			 * MacOS X can extract the object type here such as:
2245 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2246 			 */
2247 
2248 			if (check_sysattrs && !zap.za_normalization_conflict) {
2249 				zap.za_normalization_conflict =
2250 				    xattr_sysattr_casechk(zap.za_name);
2251 			}
2252 		}
2253 
2254 		if (flags & V_RDDIR_ACCFILTER) {
2255 			/*
2256 			 * If we have no access at all, don't include
2257 			 * this entry in the returned information
2258 			 */
2259 			znode_t	*ezp;
2260 			if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2261 				goto skip_entry;
2262 			if (!zfs_has_access(ezp, cr)) {
2263 				VN_RELE(ZTOV(ezp));
2264 				goto skip_entry;
2265 			}
2266 			VN_RELE(ZTOV(ezp));
2267 		}
2268 
2269 		if (flags & V_RDDIR_ENTFLAGS)
2270 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2271 		else
2272 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2273 
2274 		/*
2275 		 * Will this entry fit in the buffer?
2276 		 */
2277 		if (outcount + reclen > bufsize) {
2278 			/*
2279 			 * Did we manage to fit anything in the buffer?
2280 			 */
2281 			if (!outcount) {
2282 				error = EINVAL;
2283 				goto update;
2284 			}
2285 			break;
2286 		}
2287 		if (flags & V_RDDIR_ENTFLAGS) {
2288 			/*
2289 			 * Add extended flag entry:
2290 			 */
2291 			eodp->ed_ino = objnum;
2292 			eodp->ed_reclen = reclen;
2293 			/* NOTE: ed_off is the offset for the *next* entry */
2294 			next = &(eodp->ed_off);
2295 			eodp->ed_eflags = zap.za_normalization_conflict ?
2296 			    ED_CASE_CONFLICT : 0;
2297 			(void) strncpy(eodp->ed_name, zap.za_name,
2298 			    EDIRENT_NAMELEN(reclen));
2299 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2300 		} else {
2301 			/*
2302 			 * Add normal entry:
2303 			 */
2304 			odp->d_ino = objnum;
2305 			odp->d_reclen = reclen;
2306 			/* NOTE: d_off is the offset for the *next* entry */
2307 			next = &(odp->d_off);
2308 			(void) strncpy(odp->d_name, zap.za_name,
2309 			    DIRENT64_NAMELEN(reclen));
2310 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2311 		}
2312 		outcount += reclen;
2313 
2314 		ASSERT(outcount <= bufsize);
2315 
2316 		/* Prefetch znode */
2317 		if (prefetch)
2318 			dmu_prefetch(os, objnum, 0, 0);
2319 
2320 	skip_entry:
2321 		/*
2322 		 * Move to the next entry, fill in the previous offset.
2323 		 */
2324 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2325 			zap_cursor_advance(&zc);
2326 			offset = zap_cursor_serialize(&zc);
2327 		} else {
2328 			offset += 1;
2329 		}
2330 		if (next)
2331 			*next = offset;
2332 	}
2333 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2334 
2335 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2336 		iovp->iov_base += outcount;
2337 		iovp->iov_len -= outcount;
2338 		uio->uio_resid -= outcount;
2339 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2340 		/*
2341 		 * Reset the pointer.
2342 		 */
2343 		offset = uio->uio_loffset;
2344 	}
2345 
2346 update:
2347 	zap_cursor_fini(&zc);
2348 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2349 		kmem_free(outbuf, bufsize);
2350 
2351 	if (error == ENOENT)
2352 		error = 0;
2353 
2354 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2355 
2356 	uio->uio_loffset = offset;
2357 	ZFS_EXIT(zfsvfs);
2358 	return (error);
2359 }
2360 
2361 ulong_t zfs_fsync_sync_cnt = 4;
2362 
2363 static int
2364 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2365 {
2366 	znode_t	*zp = VTOZ(vp);
2367 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2368 
2369 	/*
2370 	 * Regardless of whether this is required for standards conformance,
2371 	 * this is the logical behavior when fsync() is called on a file with
2372 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2373 	 * going to be pushed out as part of the zil_commit().
2374 	 */
2375 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2376 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2377 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2378 
2379 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2380 
2381 	if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2382 		ZFS_ENTER(zfsvfs);
2383 		ZFS_VERIFY_ZP(zp);
2384 		zil_commit(zfsvfs->z_log, zp->z_id);
2385 		ZFS_EXIT(zfsvfs);
2386 	}
2387 	return (0);
2388 }
2389 
2390 
2391 /*
2392  * Get the requested file attributes and place them in the provided
2393  * vattr structure.
2394  *
2395  *	IN:	vp	- vnode of file.
2396  *		vap	- va_mask identifies requested attributes.
2397  *			  If AT_XVATTR set, then optional attrs are requested
2398  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2399  *		cr	- credentials of caller.
2400  *		ct	- caller context
2401  *
2402  *	OUT:	vap	- attribute values.
2403  *
2404  *	RETURN:	0 (always succeeds)
2405  */
2406 /* ARGSUSED */
2407 static int
2408 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2409     caller_context_t *ct)
2410 {
2411 	znode_t *zp = VTOZ(vp);
2412 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2413 	int	error = 0;
2414 	uint64_t links;
2415 	uint64_t mtime[2], ctime[2];
2416 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2417 	xoptattr_t *xoap = NULL;
2418 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2419 	sa_bulk_attr_t bulk[2];
2420 	int count = 0;
2421 
2422 	ZFS_ENTER(zfsvfs);
2423 	ZFS_VERIFY_ZP(zp);
2424 
2425 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2426 
2427 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2428 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2429 
2430 	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2431 		ZFS_EXIT(zfsvfs);
2432 		return (error);
2433 	}
2434 
2435 	/*
2436 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2437 	 * Also, if we are the owner don't bother, since owner should
2438 	 * always be allowed to read basic attributes of file.
2439 	 */
2440 	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2441 	    (vap->va_uid != crgetuid(cr))) {
2442 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2443 		    skipaclchk, cr)) {
2444 			ZFS_EXIT(zfsvfs);
2445 			return (error);
2446 		}
2447 	}
2448 
2449 	/*
2450 	 * Return all attributes.  It's cheaper to provide the answer
2451 	 * than to determine whether we were asked the question.
2452 	 */
2453 
2454 	mutex_enter(&zp->z_lock);
2455 	vap->va_type = vp->v_type;
2456 	vap->va_mode = zp->z_mode & MODEMASK;
2457 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2458 	vap->va_nodeid = zp->z_id;
2459 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2460 		links = zp->z_links + 1;
2461 	else
2462 		links = zp->z_links;
2463 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2464 	vap->va_size = zp->z_size;
2465 	vap->va_rdev = vp->v_rdev;
2466 	vap->va_seq = zp->z_seq;
2467 
2468 	/*
2469 	 * Add in any requested optional attributes and the create time.
2470 	 * Also set the corresponding bits in the returned attribute bitmap.
2471 	 */
2472 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2473 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2474 			xoap->xoa_archive =
2475 			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2476 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2477 		}
2478 
2479 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2480 			xoap->xoa_readonly =
2481 			    ((zp->z_pflags & ZFS_READONLY) != 0);
2482 			XVA_SET_RTN(xvap, XAT_READONLY);
2483 		}
2484 
2485 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2486 			xoap->xoa_system =
2487 			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2488 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2489 		}
2490 
2491 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2492 			xoap->xoa_hidden =
2493 			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2494 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2495 		}
2496 
2497 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2498 			xoap->xoa_nounlink =
2499 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2500 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2501 		}
2502 
2503 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2504 			xoap->xoa_immutable =
2505 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2506 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2507 		}
2508 
2509 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2510 			xoap->xoa_appendonly =
2511 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2512 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2513 		}
2514 
2515 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2516 			xoap->xoa_nodump =
2517 			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2518 			XVA_SET_RTN(xvap, XAT_NODUMP);
2519 		}
2520 
2521 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2522 			xoap->xoa_opaque =
2523 			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2524 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2525 		}
2526 
2527 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2528 			xoap->xoa_av_quarantined =
2529 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2530 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2531 		}
2532 
2533 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2534 			xoap->xoa_av_modified =
2535 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2536 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2537 		}
2538 
2539 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2540 		    vp->v_type == VREG) {
2541 			zfs_sa_get_scanstamp(zp, xvap);
2542 		}
2543 
2544 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2545 			uint64_t times[2];
2546 
2547 			(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2548 			    times, sizeof (times));
2549 			ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2550 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2551 		}
2552 
2553 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2554 			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2555 			XVA_SET_RTN(xvap, XAT_REPARSE);
2556 		}
2557 		if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2558 			xoap->xoa_generation = zp->z_gen;
2559 			XVA_SET_RTN(xvap, XAT_GEN);
2560 		}
2561 
2562 		if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2563 			xoap->xoa_offline =
2564 			    ((zp->z_pflags & ZFS_OFFLINE) != 0);
2565 			XVA_SET_RTN(xvap, XAT_OFFLINE);
2566 		}
2567 
2568 		if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2569 			xoap->xoa_sparse =
2570 			    ((zp->z_pflags & ZFS_SPARSE) != 0);
2571 			XVA_SET_RTN(xvap, XAT_SPARSE);
2572 		}
2573 	}
2574 
2575 	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2576 	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2577 	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2578 
2579 	mutex_exit(&zp->z_lock);
2580 
2581 	sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2582 
2583 	if (zp->z_blksz == 0) {
2584 		/*
2585 		 * Block size hasn't been set; suggest maximal I/O transfers.
2586 		 */
2587 		vap->va_blksize = zfsvfs->z_max_blksz;
2588 	}
2589 
2590 	ZFS_EXIT(zfsvfs);
2591 	return (0);
2592 }
2593 
2594 /*
2595  * Set the file attributes to the values contained in the
2596  * vattr structure.
2597  *
2598  *	IN:	vp	- vnode of file to be modified.
2599  *		vap	- new attribute values.
2600  *			  If AT_XVATTR set, then optional attrs are being set
2601  *		flags	- ATTR_UTIME set if non-default time values provided.
2602  *			- ATTR_NOACLCHECK (CIFS context only).
2603  *		cr	- credentials of caller.
2604  *		ct	- caller context
2605  *
2606  *	RETURN:	0 if success
2607  *		error code if failure
2608  *
2609  * Timestamps:
2610  *	vp - ctime updated, mtime updated if size changed.
2611  */
2612 /* ARGSUSED */
2613 static int
2614 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2615 	caller_context_t *ct)
2616 {
2617 	znode_t		*zp = VTOZ(vp);
2618 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2619 	zilog_t		*zilog;
2620 	dmu_tx_t	*tx;
2621 	vattr_t		oldva;
2622 	xvattr_t	tmpxvattr;
2623 	uint_t		mask = vap->va_mask;
2624 	uint_t		saved_mask;
2625 	int		trim_mask = 0;
2626 	uint64_t	new_mode;
2627 	uint64_t	new_uid, new_gid;
2628 	uint64_t	xattr_obj;
2629 	uint64_t	mtime[2], ctime[2];
2630 	znode_t		*attrzp;
2631 	int		need_policy = FALSE;
2632 	int		err, err2;
2633 	zfs_fuid_info_t *fuidp = NULL;
2634 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2635 	xoptattr_t	*xoap;
2636 	zfs_acl_t	*aclp;
2637 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2638 	boolean_t	fuid_dirtied = B_FALSE;
2639 	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
2640 	int		count = 0, xattr_count = 0;
2641 
2642 	if (mask == 0)
2643 		return (0);
2644 
2645 	if (mask & AT_NOSET)
2646 		return (EINVAL);
2647 
2648 	ZFS_ENTER(zfsvfs);
2649 	ZFS_VERIFY_ZP(zp);
2650 
2651 	zilog = zfsvfs->z_log;
2652 
2653 	/*
2654 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2655 	 * that file system is at proper version level
2656 	 */
2657 
2658 	if (zfsvfs->z_use_fuids == B_FALSE &&
2659 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2660 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2661 	    (mask & AT_XVATTR))) {
2662 		ZFS_EXIT(zfsvfs);
2663 		return (EINVAL);
2664 	}
2665 
2666 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2667 		ZFS_EXIT(zfsvfs);
2668 		return (EISDIR);
2669 	}
2670 
2671 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2672 		ZFS_EXIT(zfsvfs);
2673 		return (EINVAL);
2674 	}
2675 
2676 	/*
2677 	 * If this is an xvattr_t, then get a pointer to the structure of
2678 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2679 	 */
2680 	xoap = xva_getxoptattr(xvap);
2681 
2682 	xva_init(&tmpxvattr);
2683 
2684 	/*
2685 	 * Immutable files can only alter immutable bit and atime
2686 	 */
2687 	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2688 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2689 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2690 		ZFS_EXIT(zfsvfs);
2691 		return (EPERM);
2692 	}
2693 
2694 	if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2695 		ZFS_EXIT(zfsvfs);
2696 		return (EPERM);
2697 	}
2698 
2699 	/*
2700 	 * Verify timestamps doesn't overflow 32 bits.
2701 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2702 	 * handle times greater than 2039.  This check should be removed
2703 	 * once large timestamps are fully supported.
2704 	 */
2705 	if (mask & (AT_ATIME | AT_MTIME)) {
2706 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2707 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2708 			ZFS_EXIT(zfsvfs);
2709 			return (EOVERFLOW);
2710 		}
2711 	}
2712 
2713 top:
2714 	attrzp = NULL;
2715 	aclp = NULL;
2716 
2717 	/* Can this be moved to before the top label? */
2718 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2719 		ZFS_EXIT(zfsvfs);
2720 		return (EROFS);
2721 	}
2722 
2723 	/*
2724 	 * First validate permissions
2725 	 */
2726 
2727 	if (mask & AT_SIZE) {
2728 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2729 		if (err) {
2730 			ZFS_EXIT(zfsvfs);
2731 			return (err);
2732 		}
2733 		/*
2734 		 * XXX - Note, we are not providing any open
2735 		 * mode flags here (like FNDELAY), so we may
2736 		 * block if there are locks present... this
2737 		 * should be addressed in openat().
2738 		 */
2739 		/* XXX - would it be OK to generate a log record here? */
2740 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2741 		if (err) {
2742 			ZFS_EXIT(zfsvfs);
2743 			return (err);
2744 		}
2745 	}
2746 
2747 	if (mask & (AT_ATIME|AT_MTIME) ||
2748 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2749 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2750 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2751 	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2752 	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2753 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2754 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2755 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2756 		    skipaclchk, cr);
2757 	}
2758 
2759 	if (mask & (AT_UID|AT_GID)) {
2760 		int	idmask = (mask & (AT_UID|AT_GID));
2761 		int	take_owner;
2762 		int	take_group;
2763 
2764 		/*
2765 		 * NOTE: even if a new mode is being set,
2766 		 * we may clear S_ISUID/S_ISGID bits.
2767 		 */
2768 
2769 		if (!(mask & AT_MODE))
2770 			vap->va_mode = zp->z_mode;
2771 
2772 		/*
2773 		 * Take ownership or chgrp to group we are a member of
2774 		 */
2775 
2776 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2777 		take_group = (mask & AT_GID) &&
2778 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2779 
2780 		/*
2781 		 * If both AT_UID and AT_GID are set then take_owner and
2782 		 * take_group must both be set in order to allow taking
2783 		 * ownership.
2784 		 *
2785 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2786 		 *
2787 		 */
2788 
2789 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2790 		    ((idmask == AT_UID) && take_owner) ||
2791 		    ((idmask == AT_GID) && take_group)) {
2792 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2793 			    skipaclchk, cr) == 0) {
2794 				/*
2795 				 * Remove setuid/setgid for non-privileged users
2796 				 */
2797 				secpolicy_setid_clear(vap, cr);
2798 				trim_mask = (mask & (AT_UID|AT_GID));
2799 			} else {
2800 				need_policy =  TRUE;
2801 			}
2802 		} else {
2803 			need_policy =  TRUE;
2804 		}
2805 	}
2806 
2807 	mutex_enter(&zp->z_lock);
2808 	oldva.va_mode = zp->z_mode;
2809 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2810 	if (mask & AT_XVATTR) {
2811 		/*
2812 		 * Update xvattr mask to include only those attributes
2813 		 * that are actually changing.
2814 		 *
2815 		 * the bits will be restored prior to actually setting
2816 		 * the attributes so the caller thinks they were set.
2817 		 */
2818 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2819 			if (xoap->xoa_appendonly !=
2820 			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2821 				need_policy = TRUE;
2822 			} else {
2823 				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2824 				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2825 			}
2826 		}
2827 
2828 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2829 			if (xoap->xoa_nounlink !=
2830 			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2831 				need_policy = TRUE;
2832 			} else {
2833 				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2834 				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2835 			}
2836 		}
2837 
2838 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2839 			if (xoap->xoa_immutable !=
2840 			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2841 				need_policy = TRUE;
2842 			} else {
2843 				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2844 				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2845 			}
2846 		}
2847 
2848 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2849 			if (xoap->xoa_nodump !=
2850 			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2851 				need_policy = TRUE;
2852 			} else {
2853 				XVA_CLR_REQ(xvap, XAT_NODUMP);
2854 				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2855 			}
2856 		}
2857 
2858 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2859 			if (xoap->xoa_av_modified !=
2860 			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2861 				need_policy = TRUE;
2862 			} else {
2863 				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2864 				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2865 			}
2866 		}
2867 
2868 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2869 			if ((vp->v_type != VREG &&
2870 			    xoap->xoa_av_quarantined) ||
2871 			    xoap->xoa_av_quarantined !=
2872 			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2873 				need_policy = TRUE;
2874 			} else {
2875 				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2876 				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2877 			}
2878 		}
2879 
2880 		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2881 			mutex_exit(&zp->z_lock);
2882 			ZFS_EXIT(zfsvfs);
2883 			return (EPERM);
2884 		}
2885 
2886 		if (need_policy == FALSE &&
2887 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2888 		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2889 			need_policy = TRUE;
2890 		}
2891 	}
2892 
2893 	mutex_exit(&zp->z_lock);
2894 
2895 	if (mask & AT_MODE) {
2896 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2897 			err = secpolicy_setid_setsticky_clear(vp, vap,
2898 			    &oldva, cr);
2899 			if (err) {
2900 				ZFS_EXIT(zfsvfs);
2901 				return (err);
2902 			}
2903 			trim_mask |= AT_MODE;
2904 		} else {
2905 			need_policy = TRUE;
2906 		}
2907 	}
2908 
2909 	if (need_policy) {
2910 		/*
2911 		 * If trim_mask is set then take ownership
2912 		 * has been granted or write_acl is present and user
2913 		 * has the ability to modify mode.  In that case remove
2914 		 * UID|GID and or MODE from mask so that
2915 		 * secpolicy_vnode_setattr() doesn't revoke it.
2916 		 */
2917 
2918 		if (trim_mask) {
2919 			saved_mask = vap->va_mask;
2920 			vap->va_mask &= ~trim_mask;
2921 		}
2922 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2923 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2924 		if (err) {
2925 			ZFS_EXIT(zfsvfs);
2926 			return (err);
2927 		}
2928 
2929 		if (trim_mask)
2930 			vap->va_mask |= saved_mask;
2931 	}
2932 
2933 	/*
2934 	 * secpolicy_vnode_setattr, or take ownership may have
2935 	 * changed va_mask
2936 	 */
2937 	mask = vap->va_mask;
2938 
2939 	if ((mask & (AT_UID | AT_GID))) {
2940 		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
2941 		    &xattr_obj, sizeof (xattr_obj));
2942 
2943 		if (err == 0 && xattr_obj) {
2944 			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
2945 			if (err)
2946 				goto out2;
2947 		}
2948 		if (mask & AT_UID) {
2949 			new_uid = zfs_fuid_create(zfsvfs,
2950 			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
2951 			if (new_uid != zp->z_uid &&
2952 			    zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
2953 				if (attrzp)
2954 					VN_RELE(ZTOV(attrzp));
2955 				err = EDQUOT;
2956 				goto out2;
2957 			}
2958 		}
2959 
2960 		if (mask & AT_GID) {
2961 			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
2962 			    cr, ZFS_GROUP, &fuidp);
2963 			if (new_gid != zp->z_gid &&
2964 			    zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
2965 				if (attrzp)
2966 					VN_RELE(ZTOV(attrzp));
2967 				err = EDQUOT;
2968 				goto out2;
2969 			}
2970 		}
2971 	}
2972 	tx = dmu_tx_create(zfsvfs->z_os);
2973 
2974 	if (mask & AT_MODE) {
2975 		uint64_t pmode = zp->z_mode;
2976 		uint64_t acl_obj;
2977 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2978 
2979 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
2980 			goto out;
2981 
2982 		mutex_enter(&zp->z_lock);
2983 		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
2984 			/*
2985 			 * Are we upgrading ACL from old V0 format
2986 			 * to V1 format?
2987 			 */
2988 			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2989 			    zfs_znode_acl_version(zp) ==
2990 			    ZFS_ACL_VERSION_INITIAL) {
2991 				dmu_tx_hold_free(tx, acl_obj, 0,
2992 				    DMU_OBJECT_END);
2993 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2994 				    0, aclp->z_acl_bytes);
2995 			} else {
2996 				dmu_tx_hold_write(tx, acl_obj, 0,
2997 				    aclp->z_acl_bytes);
2998 			}
2999 		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3000 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3001 			    0, aclp->z_acl_bytes);
3002 		}
3003 		mutex_exit(&zp->z_lock);
3004 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3005 	} else {
3006 		if ((mask & AT_XVATTR) &&
3007 		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3008 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3009 		else
3010 			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3011 	}
3012 
3013 	if (attrzp) {
3014 		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3015 	}
3016 
3017 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3018 	if (fuid_dirtied)
3019 		zfs_fuid_txhold(zfsvfs, tx);
3020 
3021 	zfs_sa_upgrade_txholds(tx, zp);
3022 
3023 	err = dmu_tx_assign(tx, TXG_NOWAIT);
3024 	if (err) {
3025 		if (err == ERESTART)
3026 			dmu_tx_wait(tx);
3027 		goto out;
3028 	}
3029 
3030 	count = 0;
3031 	/*
3032 	 * Set each attribute requested.
3033 	 * We group settings according to the locks they need to acquire.
3034 	 *
3035 	 * Note: you cannot set ctime directly, although it will be
3036 	 * updated as a side-effect of calling this function.
3037 	 */
3038 
3039 
3040 	if (mask & (AT_UID|AT_GID|AT_MODE))
3041 		mutex_enter(&zp->z_acl_lock);
3042 	mutex_enter(&zp->z_lock);
3043 
3044 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3045 	    &zp->z_pflags, sizeof (zp->z_pflags));
3046 
3047 	if (attrzp) {
3048 		if (mask & (AT_UID|AT_GID|AT_MODE))
3049 			mutex_enter(&attrzp->z_acl_lock);
3050 		mutex_enter(&attrzp->z_lock);
3051 		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3052 		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3053 		    sizeof (attrzp->z_pflags));
3054 	}
3055 
3056 	if (mask & (AT_UID|AT_GID)) {
3057 
3058 		if (mask & AT_UID) {
3059 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3060 			    &new_uid, sizeof (new_uid));
3061 			zp->z_uid = new_uid;
3062 			if (attrzp) {
3063 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3064 				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3065 				    sizeof (new_uid));
3066 				attrzp->z_uid = new_uid;
3067 			}
3068 		}
3069 
3070 		if (mask & AT_GID) {
3071 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3072 			    NULL, &new_gid, sizeof (new_gid));
3073 			zp->z_gid = new_gid;
3074 			if (attrzp) {
3075 				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3076 				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3077 				    sizeof (new_gid));
3078 				attrzp->z_gid = new_gid;
3079 			}
3080 		}
3081 		if (!(mask & AT_MODE)) {
3082 			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3083 			    NULL, &new_mode, sizeof (new_mode));
3084 			new_mode = zp->z_mode;
3085 		}
3086 		err = zfs_acl_chown_setattr(zp);
3087 		ASSERT(err == 0);
3088 		if (attrzp) {
3089 			err = zfs_acl_chown_setattr(attrzp);
3090 			ASSERT(err == 0);
3091 		}
3092 	}
3093 
3094 	if (mask & AT_MODE) {
3095 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3096 		    &new_mode, sizeof (new_mode));
3097 		zp->z_mode = new_mode;
3098 		ASSERT3U((uintptr_t)aclp, !=, NULL);
3099 		err = zfs_aclset_common(zp, aclp, cr, tx);
3100 		ASSERT0(err);
3101 		if (zp->z_acl_cached)
3102 			zfs_acl_free(zp->z_acl_cached);
3103 		zp->z_acl_cached = aclp;
3104 		aclp = NULL;
3105 	}
3106 
3107 
3108 	if (mask & AT_ATIME) {
3109 		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3110 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3111 		    &zp->z_atime, sizeof (zp->z_atime));
3112 	}
3113 
3114 	if (mask & AT_MTIME) {
3115 		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3116 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3117 		    mtime, sizeof (mtime));
3118 	}
3119 
3120 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3121 	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3122 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3123 		    NULL, mtime, sizeof (mtime));
3124 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3125 		    &ctime, sizeof (ctime));
3126 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3127 		    B_TRUE);
3128 	} else if (mask != 0) {
3129 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3130 		    &ctime, sizeof (ctime));
3131 		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3132 		    B_TRUE);
3133 		if (attrzp) {
3134 			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3135 			    SA_ZPL_CTIME(zfsvfs), NULL,
3136 			    &ctime, sizeof (ctime));
3137 			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3138 			    mtime, ctime, B_TRUE);
3139 		}
3140 	}
3141 	/*
3142 	 * Do this after setting timestamps to prevent timestamp
3143 	 * update from toggling bit
3144 	 */
3145 
3146 	if (xoap && (mask & AT_XVATTR)) {
3147 
3148 		/*
3149 		 * restore trimmed off masks
3150 		 * so that return masks can be set for caller.
3151 		 */
3152 
3153 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3154 			XVA_SET_REQ(xvap, XAT_APPENDONLY);
3155 		}
3156 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3157 			XVA_SET_REQ(xvap, XAT_NOUNLINK);
3158 		}
3159 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3160 			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3161 		}
3162 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3163 			XVA_SET_REQ(xvap, XAT_NODUMP);
3164 		}
3165 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3166 			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3167 		}
3168 		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3169 			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3170 		}
3171 
3172 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3173 			ASSERT(vp->v_type == VREG);
3174 
3175 		zfs_xvattr_set(zp, xvap, tx);
3176 	}
3177 
3178 	if (fuid_dirtied)
3179 		zfs_fuid_sync(zfsvfs, tx);
3180 
3181 	if (mask != 0)
3182 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3183 
3184 	mutex_exit(&zp->z_lock);
3185 	if (mask & (AT_UID|AT_GID|AT_MODE))
3186 		mutex_exit(&zp->z_acl_lock);
3187 
3188 	if (attrzp) {
3189 		if (mask & (AT_UID|AT_GID|AT_MODE))
3190 			mutex_exit(&attrzp->z_acl_lock);
3191 		mutex_exit(&attrzp->z_lock);
3192 	}
3193 out:
3194 	if (err == 0 && attrzp) {
3195 		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3196 		    xattr_count, tx);
3197 		ASSERT(err2 == 0);
3198 	}
3199 
3200 	if (attrzp)
3201 		VN_RELE(ZTOV(attrzp));
3202 	if (aclp)
3203 		zfs_acl_free(aclp);
3204 
3205 	if (fuidp) {
3206 		zfs_fuid_info_free(fuidp);
3207 		fuidp = NULL;
3208 	}
3209 
3210 	if (err) {
3211 		dmu_tx_abort(tx);
3212 		if (err == ERESTART)
3213 			goto top;
3214 	} else {
3215 		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3216 		dmu_tx_commit(tx);
3217 	}
3218 
3219 out2:
3220 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3221 		zil_commit(zilog, 0);
3222 
3223 	ZFS_EXIT(zfsvfs);
3224 	return (err);
3225 }
3226 
3227 typedef struct zfs_zlock {
3228 	krwlock_t	*zl_rwlock;	/* lock we acquired */
3229 	znode_t		*zl_znode;	/* znode we held */
3230 	struct zfs_zlock *zl_next;	/* next in list */
3231 } zfs_zlock_t;
3232 
3233 /*
3234  * Drop locks and release vnodes that were held by zfs_rename_lock().
3235  */
3236 static void
3237 zfs_rename_unlock(zfs_zlock_t **zlpp)
3238 {
3239 	zfs_zlock_t *zl;
3240 
3241 	while ((zl = *zlpp) != NULL) {
3242 		if (zl->zl_znode != NULL)
3243 			VN_RELE(ZTOV(zl->zl_znode));
3244 		rw_exit(zl->zl_rwlock);
3245 		*zlpp = zl->zl_next;
3246 		kmem_free(zl, sizeof (*zl));
3247 	}
3248 }
3249 
3250 /*
3251  * Search back through the directory tree, using the ".." entries.
3252  * Lock each directory in the chain to prevent concurrent renames.
3253  * Fail any attempt to move a directory into one of its own descendants.
3254  * XXX - z_parent_lock can overlap with map or grow locks
3255  */
3256 static int
3257 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3258 {
3259 	zfs_zlock_t	*zl;
3260 	znode_t		*zp = tdzp;
3261 	uint64_t	rootid = zp->z_zfsvfs->z_root;
3262 	uint64_t	oidp = zp->z_id;
3263 	krwlock_t	*rwlp = &szp->z_parent_lock;
3264 	krw_t		rw = RW_WRITER;
3265 
3266 	/*
3267 	 * First pass write-locks szp and compares to zp->z_id.
3268 	 * Later passes read-lock zp and compare to zp->z_parent.
3269 	 */
3270 	do {
3271 		if (!rw_tryenter(rwlp, rw)) {
3272 			/*
3273 			 * Another thread is renaming in this path.
3274 			 * Note that if we are a WRITER, we don't have any
3275 			 * parent_locks held yet.
3276 			 */
3277 			if (rw == RW_READER && zp->z_id > szp->z_id) {
3278 				/*
3279 				 * Drop our locks and restart
3280 				 */
3281 				zfs_rename_unlock(&zl);
3282 				*zlpp = NULL;
3283 				zp = tdzp;
3284 				oidp = zp->z_id;
3285 				rwlp = &szp->z_parent_lock;
3286 				rw = RW_WRITER;
3287 				continue;
3288 			} else {
3289 				/*
3290 				 * Wait for other thread to drop its locks
3291 				 */
3292 				rw_enter(rwlp, rw);
3293 			}
3294 		}
3295 
3296 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3297 		zl->zl_rwlock = rwlp;
3298 		zl->zl_znode = NULL;
3299 		zl->zl_next = *zlpp;
3300 		*zlpp = zl;
3301 
3302 		if (oidp == szp->z_id)		/* We're a descendant of szp */
3303 			return (EINVAL);
3304 
3305 		if (oidp == rootid)		/* We've hit the top */
3306 			return (0);
3307 
3308 		if (rw == RW_READER) {		/* i.e. not the first pass */
3309 			int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3310 			if (error)
3311 				return (error);
3312 			zl->zl_znode = zp;
3313 		}
3314 		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3315 		    &oidp, sizeof (oidp));
3316 		rwlp = &zp->z_parent_lock;
3317 		rw = RW_READER;
3318 
3319 	} while (zp->z_id != sdzp->z_id);
3320 
3321 	return (0);
3322 }
3323 
3324 /*
3325  * Move an entry from the provided source directory to the target
3326  * directory.  Change the entry name as indicated.
3327  *
3328  *	IN:	sdvp	- Source directory containing the "old entry".
3329  *		snm	- Old entry name.
3330  *		tdvp	- Target directory to contain the "new entry".
3331  *		tnm	- New entry name.
3332  *		cr	- credentials of caller.
3333  *		ct	- caller context
3334  *		flags	- case flags
3335  *
3336  *	RETURN:	0 if success
3337  *		error code if failure
3338  *
3339  * Timestamps:
3340  *	sdvp,tdvp - ctime|mtime updated
3341  */
3342 /*ARGSUSED*/
3343 static int
3344 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3345     caller_context_t *ct, int flags)
3346 {
3347 	znode_t		*tdzp, *szp, *tzp;
3348 	znode_t		*sdzp = VTOZ(sdvp);
3349 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
3350 	zilog_t		*zilog;
3351 	vnode_t		*realvp;
3352 	zfs_dirlock_t	*sdl, *tdl;
3353 	dmu_tx_t	*tx;
3354 	zfs_zlock_t	*zl;
3355 	int		cmp, serr, terr;
3356 	int		error = 0;
3357 	int		zflg = 0;
3358 
3359 	ZFS_ENTER(zfsvfs);
3360 	ZFS_VERIFY_ZP(sdzp);
3361 	zilog = zfsvfs->z_log;
3362 
3363 	/*
3364 	 * Make sure we have the real vp for the target directory.
3365 	 */
3366 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3367 		tdvp = realvp;
3368 
3369 	if (tdvp->v_vfsp != sdvp->v_vfsp || zfsctl_is_node(tdvp)) {
3370 		ZFS_EXIT(zfsvfs);
3371 		return (EXDEV);
3372 	}
3373 
3374 	tdzp = VTOZ(tdvp);
3375 	ZFS_VERIFY_ZP(tdzp);
3376 	if (zfsvfs->z_utf8 && u8_validate(tnm,
3377 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3378 		ZFS_EXIT(zfsvfs);
3379 		return (EILSEQ);
3380 	}
3381 
3382 	if (flags & FIGNORECASE)
3383 		zflg |= ZCILOOK;
3384 
3385 top:
3386 	szp = NULL;
3387 	tzp = NULL;
3388 	zl = NULL;
3389 
3390 	/*
3391 	 * This is to prevent the creation of links into attribute space
3392 	 * by renaming a linked file into/outof an attribute directory.
3393 	 * See the comment in zfs_link() for why this is considered bad.
3394 	 */
3395 	if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3396 		ZFS_EXIT(zfsvfs);
3397 		return (EINVAL);
3398 	}
3399 
3400 	/*
3401 	 * Lock source and target directory entries.  To prevent deadlock,
3402 	 * a lock ordering must be defined.  We lock the directory with
3403 	 * the smallest object id first, or if it's a tie, the one with
3404 	 * the lexically first name.
3405 	 */
3406 	if (sdzp->z_id < tdzp->z_id) {
3407 		cmp = -1;
3408 	} else if (sdzp->z_id > tdzp->z_id) {
3409 		cmp = 1;
3410 	} else {
3411 		/*
3412 		 * First compare the two name arguments without
3413 		 * considering any case folding.
3414 		 */
3415 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3416 
3417 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3418 		ASSERT(error == 0 || !zfsvfs->z_utf8);
3419 		if (cmp == 0) {
3420 			/*
3421 			 * POSIX: "If the old argument and the new argument
3422 			 * both refer to links to the same existing file,
3423 			 * the rename() function shall return successfully
3424 			 * and perform no other action."
3425 			 */
3426 			ZFS_EXIT(zfsvfs);
3427 			return (0);
3428 		}
3429 		/*
3430 		 * If the file system is case-folding, then we may
3431 		 * have some more checking to do.  A case-folding file
3432 		 * system is either supporting mixed case sensitivity
3433 		 * access or is completely case-insensitive.  Note
3434 		 * that the file system is always case preserving.
3435 		 *
3436 		 * In mixed sensitivity mode case sensitive behavior
3437 		 * is the default.  FIGNORECASE must be used to
3438 		 * explicitly request case insensitive behavior.
3439 		 *
3440 		 * If the source and target names provided differ only
3441 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
3442 		 * we will treat this as a special case in the
3443 		 * case-insensitive mode: as long as the source name
3444 		 * is an exact match, we will allow this to proceed as
3445 		 * a name-change request.
3446 		 */
3447 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3448 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
3449 		    flags & FIGNORECASE)) &&
3450 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3451 		    &error) == 0) {
3452 			/*
3453 			 * case preserving rename request, require exact
3454 			 * name matches
3455 			 */
3456 			zflg |= ZCIEXACT;
3457 			zflg &= ~ZCILOOK;
3458 		}
3459 	}
3460 
3461 	/*
3462 	 * If the source and destination directories are the same, we should
3463 	 * grab the z_name_lock of that directory only once.
3464 	 */
3465 	if (sdzp == tdzp) {
3466 		zflg |= ZHAVELOCK;
3467 		rw_enter(&sdzp->z_name_lock, RW_READER);
3468 	}
3469 
3470 	if (cmp < 0) {
3471 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3472 		    ZEXISTS | zflg, NULL, NULL);
3473 		terr = zfs_dirent_lock(&tdl,
3474 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3475 	} else {
3476 		terr = zfs_dirent_lock(&tdl,
3477 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3478 		serr = zfs_dirent_lock(&sdl,
3479 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3480 		    NULL, NULL);
3481 	}
3482 
3483 	if (serr) {
3484 		/*
3485 		 * Source entry invalid or not there.
3486 		 */
3487 		if (!terr) {
3488 			zfs_dirent_unlock(tdl);
3489 			if (tzp)
3490 				VN_RELE(ZTOV(tzp));
3491 		}
3492 
3493 		if (sdzp == tdzp)
3494 			rw_exit(&sdzp->z_name_lock);
3495 
3496 		if (strcmp(snm, "..") == 0)
3497 			serr = EINVAL;
3498 		ZFS_EXIT(zfsvfs);
3499 		return (serr);
3500 	}
3501 	if (terr) {
3502 		zfs_dirent_unlock(sdl);
3503 		VN_RELE(ZTOV(szp));
3504 
3505 		if (sdzp == tdzp)
3506 			rw_exit(&sdzp->z_name_lock);
3507 
3508 		if (strcmp(tnm, "..") == 0)
3509 			terr = EINVAL;
3510 		ZFS_EXIT(zfsvfs);
3511 		return (terr);
3512 	}
3513 
3514 	/*
3515 	 * Must have write access at the source to remove the old entry
3516 	 * and write access at the target to create the new entry.
3517 	 * Note that if target and source are the same, this can be
3518 	 * done in a single check.
3519 	 */
3520 
3521 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3522 		goto out;
3523 
3524 	if (ZTOV(szp)->v_type == VDIR) {
3525 		/*
3526 		 * Check to make sure rename is valid.
3527 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3528 		 */
3529 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3530 			goto out;
3531 	}
3532 
3533 	/*
3534 	 * Does target exist?
3535 	 */
3536 	if (tzp) {
3537 		/*
3538 		 * Source and target must be the same type.
3539 		 */
3540 		if (ZTOV(szp)->v_type == VDIR) {
3541 			if (ZTOV(tzp)->v_type != VDIR) {
3542 				error = ENOTDIR;
3543 				goto out;
3544 			}
3545 		} else {
3546 			if (ZTOV(tzp)->v_type == VDIR) {
3547 				error = EISDIR;
3548 				goto out;
3549 			}
3550 		}
3551 		/*
3552 		 * POSIX dictates that when the source and target
3553 		 * entries refer to the same file object, rename
3554 		 * must do nothing and exit without error.
3555 		 */
3556 		if (szp->z_id == tzp->z_id) {
3557 			error = 0;
3558 			goto out;
3559 		}
3560 	}
3561 
3562 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3563 	if (tzp)
3564 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3565 
3566 	/*
3567 	 * notify the target directory if it is not the same
3568 	 * as source directory.
3569 	 */
3570 	if (tdvp != sdvp) {
3571 		vnevent_rename_dest_dir(tdvp, ct);
3572 	}
3573 
3574 	tx = dmu_tx_create(zfsvfs->z_os);
3575 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3576 	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3577 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3578 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3579 	if (sdzp != tdzp) {
3580 		dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3581 		zfs_sa_upgrade_txholds(tx, tdzp);
3582 	}
3583 	if (tzp) {
3584 		dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3585 		zfs_sa_upgrade_txholds(tx, tzp);
3586 	}
3587 
3588 	zfs_sa_upgrade_txholds(tx, szp);
3589 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3590 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3591 	if (error) {
3592 		if (zl != NULL)
3593 			zfs_rename_unlock(&zl);
3594 		zfs_dirent_unlock(sdl);
3595 		zfs_dirent_unlock(tdl);
3596 
3597 		if (sdzp == tdzp)
3598 			rw_exit(&sdzp->z_name_lock);
3599 
3600 		VN_RELE(ZTOV(szp));
3601 		if (tzp)
3602 			VN_RELE(ZTOV(tzp));
3603 		if (error == ERESTART) {
3604 			dmu_tx_wait(tx);
3605 			dmu_tx_abort(tx);
3606 			goto top;
3607 		}
3608 		dmu_tx_abort(tx);
3609 		ZFS_EXIT(zfsvfs);
3610 		return (error);
3611 	}
3612 
3613 	if (tzp)	/* Attempt to remove the existing target */
3614 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3615 
3616 	if (error == 0) {
3617 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3618 		if (error == 0) {
3619 			szp->z_pflags |= ZFS_AV_MODIFIED;
3620 
3621 			error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3622 			    (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3623 			ASSERT0(error);
3624 
3625 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3626 			if (error == 0) {
3627 				zfs_log_rename(zilog, tx, TX_RENAME |
3628 				    (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3629 				    sdl->dl_name, tdzp, tdl->dl_name, szp);
3630 
3631 				/*
3632 				 * Update path information for the target vnode
3633 				 */
3634 				vn_renamepath(tdvp, ZTOV(szp), tnm,
3635 				    strlen(tnm));
3636 			} else {
3637 				/*
3638 				 * At this point, we have successfully created
3639 				 * the target name, but have failed to remove
3640 				 * the source name.  Since the create was done
3641 				 * with the ZRENAMING flag, there are
3642 				 * complications; for one, the link count is
3643 				 * wrong.  The easiest way to deal with this
3644 				 * is to remove the newly created target, and
3645 				 * return the original error.  This must
3646 				 * succeed; fortunately, it is very unlikely to
3647 				 * fail, since we just created it.
3648 				 */
3649 				VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3650 				    ZRENAMING, NULL), ==, 0);
3651 			}
3652 		}
3653 	}
3654 
3655 	dmu_tx_commit(tx);
3656 out:
3657 	if (zl != NULL)
3658 		zfs_rename_unlock(&zl);
3659 
3660 	zfs_dirent_unlock(sdl);
3661 	zfs_dirent_unlock(tdl);
3662 
3663 	if (sdzp == tdzp)
3664 		rw_exit(&sdzp->z_name_lock);
3665 
3666 
3667 	VN_RELE(ZTOV(szp));
3668 	if (tzp)
3669 		VN_RELE(ZTOV(tzp));
3670 
3671 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3672 		zil_commit(zilog, 0);
3673 
3674 	ZFS_EXIT(zfsvfs);
3675 	return (error);
3676 }
3677 
3678 /*
3679  * Insert the indicated symbolic reference entry into the directory.
3680  *
3681  *	IN:	dvp	- Directory to contain new symbolic link.
3682  *		link	- Name for new symlink entry.
3683  *		vap	- Attributes of new entry.
3684  *		target	- Target path of new symlink.
3685  *		cr	- credentials of caller.
3686  *		ct	- caller context
3687  *		flags	- case flags
3688  *
3689  *	RETURN:	0 if success
3690  *		error code if failure
3691  *
3692  * Timestamps:
3693  *	dvp - ctime|mtime updated
3694  */
3695 /*ARGSUSED*/
3696 static int
3697 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3698     caller_context_t *ct, int flags)
3699 {
3700 	znode_t		*zp, *dzp = VTOZ(dvp);
3701 	zfs_dirlock_t	*dl;
3702 	dmu_tx_t	*tx;
3703 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3704 	zilog_t		*zilog;
3705 	uint64_t	len = strlen(link);
3706 	int		error;
3707 	int		zflg = ZNEW;
3708 	zfs_acl_ids_t	acl_ids;
3709 	boolean_t	fuid_dirtied;
3710 	uint64_t	txtype = TX_SYMLINK;
3711 
3712 	ASSERT(vap->va_type == VLNK);
3713 
3714 	ZFS_ENTER(zfsvfs);
3715 	ZFS_VERIFY_ZP(dzp);
3716 	zilog = zfsvfs->z_log;
3717 
3718 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3719 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3720 		ZFS_EXIT(zfsvfs);
3721 		return (EILSEQ);
3722 	}
3723 	if (flags & FIGNORECASE)
3724 		zflg |= ZCILOOK;
3725 
3726 	if (len > MAXPATHLEN) {
3727 		ZFS_EXIT(zfsvfs);
3728 		return (ENAMETOOLONG);
3729 	}
3730 
3731 	if ((error = zfs_acl_ids_create(dzp, 0,
3732 	    vap, cr, NULL, &acl_ids)) != 0) {
3733 		ZFS_EXIT(zfsvfs);
3734 		return (error);
3735 	}
3736 top:
3737 	/*
3738 	 * Attempt to lock directory; fail if entry already exists.
3739 	 */
3740 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3741 	if (error) {
3742 		zfs_acl_ids_free(&acl_ids);
3743 		ZFS_EXIT(zfsvfs);
3744 		return (error);
3745 	}
3746 
3747 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3748 		zfs_acl_ids_free(&acl_ids);
3749 		zfs_dirent_unlock(dl);
3750 		ZFS_EXIT(zfsvfs);
3751 		return (error);
3752 	}
3753 
3754 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3755 		zfs_acl_ids_free(&acl_ids);
3756 		zfs_dirent_unlock(dl);
3757 		ZFS_EXIT(zfsvfs);
3758 		return (EDQUOT);
3759 	}
3760 	tx = dmu_tx_create(zfsvfs->z_os);
3761 	fuid_dirtied = zfsvfs->z_fuid_dirty;
3762 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3763 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3764 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3765 	    ZFS_SA_BASE_ATTR_SIZE + len);
3766 	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3767 	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3768 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3769 		    acl_ids.z_aclp->z_acl_bytes);
3770 	}
3771 	if (fuid_dirtied)
3772 		zfs_fuid_txhold(zfsvfs, tx);
3773 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3774 	if (error) {
3775 		zfs_dirent_unlock(dl);
3776 		if (error == ERESTART) {
3777 			dmu_tx_wait(tx);
3778 			dmu_tx_abort(tx);
3779 			goto top;
3780 		}
3781 		zfs_acl_ids_free(&acl_ids);
3782 		dmu_tx_abort(tx);
3783 		ZFS_EXIT(zfsvfs);
3784 		return (error);
3785 	}
3786 
3787 	/*
3788 	 * Create a new object for the symlink.
3789 	 * for version 4 ZPL datsets the symlink will be an SA attribute
3790 	 */
3791 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3792 
3793 	if (fuid_dirtied)
3794 		zfs_fuid_sync(zfsvfs, tx);
3795 
3796 	mutex_enter(&zp->z_lock);
3797 	if (zp->z_is_sa)
3798 		error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3799 		    link, len, tx);
3800 	else
3801 		zfs_sa_symlink(zp, link, len, tx);
3802 	mutex_exit(&zp->z_lock);
3803 
3804 	zp->z_size = len;
3805 	(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3806 	    &zp->z_size, sizeof (zp->z_size), tx);
3807 	/*
3808 	 * Insert the new object into the directory.
3809 	 */
3810 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3811 
3812 	if (flags & FIGNORECASE)
3813 		txtype |= TX_CI;
3814 	zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3815 
3816 	zfs_acl_ids_free(&acl_ids);
3817 
3818 	dmu_tx_commit(tx);
3819 
3820 	zfs_dirent_unlock(dl);
3821 
3822 	VN_RELE(ZTOV(zp));
3823 
3824 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3825 		zil_commit(zilog, 0);
3826 
3827 	ZFS_EXIT(zfsvfs);
3828 	return (error);
3829 }
3830 
3831 /*
3832  * Return, in the buffer contained in the provided uio structure,
3833  * the symbolic path referred to by vp.
3834  *
3835  *	IN:	vp	- vnode of symbolic link.
3836  *		uoip	- structure to contain the link path.
3837  *		cr	- credentials of caller.
3838  *		ct	- caller context
3839  *
3840  *	OUT:	uio	- structure to contain the link path.
3841  *
3842  *	RETURN:	0 if success
3843  *		error code if failure
3844  *
3845  * Timestamps:
3846  *	vp - atime updated
3847  */
3848 /* ARGSUSED */
3849 static int
3850 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3851 {
3852 	znode_t		*zp = VTOZ(vp);
3853 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3854 	int		error;
3855 
3856 	ZFS_ENTER(zfsvfs);
3857 	ZFS_VERIFY_ZP(zp);
3858 
3859 	mutex_enter(&zp->z_lock);
3860 	if (zp->z_is_sa)
3861 		error = sa_lookup_uio(zp->z_sa_hdl,
3862 		    SA_ZPL_SYMLINK(zfsvfs), uio);
3863 	else
3864 		error = zfs_sa_readlink(zp, uio);
3865 	mutex_exit(&zp->z_lock);
3866 
3867 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3868 
3869 	ZFS_EXIT(zfsvfs);
3870 	return (error);
3871 }
3872 
3873 /*
3874  * Insert a new entry into directory tdvp referencing svp.
3875  *
3876  *	IN:	tdvp	- Directory to contain new entry.
3877  *		svp	- vnode of new entry.
3878  *		name	- name of new entry.
3879  *		cr	- credentials of caller.
3880  *		ct	- caller context
3881  *
3882  *	RETURN:	0 if success
3883  *		error code if failure
3884  *
3885  * Timestamps:
3886  *	tdvp - ctime|mtime updated
3887  *	 svp - ctime updated
3888  */
3889 /* ARGSUSED */
3890 static int
3891 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3892     caller_context_t *ct, int flags)
3893 {
3894 	znode_t		*dzp = VTOZ(tdvp);
3895 	znode_t		*tzp, *szp;
3896 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3897 	zilog_t		*zilog;
3898 	zfs_dirlock_t	*dl;
3899 	dmu_tx_t	*tx;
3900 	vnode_t		*realvp;
3901 	int		error;
3902 	int		zf = ZNEW;
3903 	uint64_t	parent;
3904 	uid_t		owner;
3905 
3906 	ASSERT(tdvp->v_type == VDIR);
3907 
3908 	ZFS_ENTER(zfsvfs);
3909 	ZFS_VERIFY_ZP(dzp);
3910 	zilog = zfsvfs->z_log;
3911 
3912 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3913 		svp = realvp;
3914 
3915 	/*
3916 	 * POSIX dictates that we return EPERM here.
3917 	 * Better choices include ENOTSUP or EISDIR.
3918 	 */
3919 	if (svp->v_type == VDIR) {
3920 		ZFS_EXIT(zfsvfs);
3921 		return (EPERM);
3922 	}
3923 
3924 	if (svp->v_vfsp != tdvp->v_vfsp || zfsctl_is_node(svp)) {
3925 		ZFS_EXIT(zfsvfs);
3926 		return (EXDEV);
3927 	}
3928 
3929 	szp = VTOZ(svp);
3930 	ZFS_VERIFY_ZP(szp);
3931 
3932 	/* Prevent links to .zfs/shares files */
3933 
3934 	if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
3935 	    &parent, sizeof (uint64_t))) != 0) {
3936 		ZFS_EXIT(zfsvfs);
3937 		return (error);
3938 	}
3939 	if (parent == zfsvfs->z_shares_dir) {
3940 		ZFS_EXIT(zfsvfs);
3941 		return (EPERM);
3942 	}
3943 
3944 	if (zfsvfs->z_utf8 && u8_validate(name,
3945 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3946 		ZFS_EXIT(zfsvfs);
3947 		return (EILSEQ);
3948 	}
3949 	if (flags & FIGNORECASE)
3950 		zf |= ZCILOOK;
3951 
3952 	/*
3953 	 * We do not support links between attributes and non-attributes
3954 	 * because of the potential security risk of creating links
3955 	 * into "normal" file space in order to circumvent restrictions
3956 	 * imposed in attribute space.
3957 	 */
3958 	if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
3959 		ZFS_EXIT(zfsvfs);
3960 		return (EINVAL);
3961 	}
3962 
3963 
3964 	owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
3965 	if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
3966 		ZFS_EXIT(zfsvfs);
3967 		return (EPERM);
3968 	}
3969 
3970 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3971 		ZFS_EXIT(zfsvfs);
3972 		return (error);
3973 	}
3974 
3975 top:
3976 	/*
3977 	 * Attempt to lock directory; fail if entry already exists.
3978 	 */
3979 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3980 	if (error) {
3981 		ZFS_EXIT(zfsvfs);
3982 		return (error);
3983 	}
3984 
3985 	tx = dmu_tx_create(zfsvfs->z_os);
3986 	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3987 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3988 	zfs_sa_upgrade_txholds(tx, szp);
3989 	zfs_sa_upgrade_txholds(tx, dzp);
3990 	error = dmu_tx_assign(tx, TXG_NOWAIT);
3991 	if (error) {
3992 		zfs_dirent_unlock(dl);
3993 		if (error == ERESTART) {
3994 			dmu_tx_wait(tx);
3995 			dmu_tx_abort(tx);
3996 			goto top;
3997 		}
3998 		dmu_tx_abort(tx);
3999 		ZFS_EXIT(zfsvfs);
4000 		return (error);
4001 	}
4002 
4003 	error = zfs_link_create(dl, szp, tx, 0);
4004 
4005 	if (error == 0) {
4006 		uint64_t txtype = TX_LINK;
4007 		if (flags & FIGNORECASE)
4008 			txtype |= TX_CI;
4009 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4010 	}
4011 
4012 	dmu_tx_commit(tx);
4013 
4014 	zfs_dirent_unlock(dl);
4015 
4016 	if (error == 0) {
4017 		vnevent_link(svp, ct);
4018 	}
4019 
4020 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4021 		zil_commit(zilog, 0);
4022 
4023 	ZFS_EXIT(zfsvfs);
4024 	return (error);
4025 }
4026 
4027 /*
4028  * zfs_null_putapage() is used when the file system has been force
4029  * unmounted. It just drops the pages.
4030  */
4031 /* ARGSUSED */
4032 static int
4033 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4034 		size_t *lenp, int flags, cred_t *cr)
4035 {
4036 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4037 	return (0);
4038 }
4039 
4040 /*
4041  * Push a page out to disk, klustering if possible.
4042  *
4043  *	IN:	vp	- file to push page to.
4044  *		pp	- page to push.
4045  *		flags	- additional flags.
4046  *		cr	- credentials of caller.
4047  *
4048  *	OUT:	offp	- start of range pushed.
4049  *		lenp	- len of range pushed.
4050  *
4051  *	RETURN:	0 if success
4052  *		error code if failure
4053  *
4054  * NOTE: callers must have locked the page to be pushed.  On
4055  * exit, the page (and all other pages in the kluster) must be
4056  * unlocked.
4057  */
4058 /* ARGSUSED */
4059 static int
4060 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4061 		size_t *lenp, int flags, cred_t *cr)
4062 {
4063 	znode_t		*zp = VTOZ(vp);
4064 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4065 	dmu_tx_t	*tx;
4066 	u_offset_t	off, koff;
4067 	size_t		len, klen;
4068 	int		err;
4069 
4070 	off = pp->p_offset;
4071 	len = PAGESIZE;
4072 	/*
4073 	 * If our blocksize is bigger than the page size, try to kluster
4074 	 * multiple pages so that we write a full block (thus avoiding
4075 	 * a read-modify-write).
4076 	 */
4077 	if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4078 		klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4079 		koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4080 		ASSERT(koff <= zp->z_size);
4081 		if (koff + klen > zp->z_size)
4082 			klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4083 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4084 	}
4085 	ASSERT3U(btop(len), ==, btopr(len));
4086 
4087 	/*
4088 	 * Can't push pages past end-of-file.
4089 	 */
4090 	if (off >= zp->z_size) {
4091 		/* ignore all pages */
4092 		err = 0;
4093 		goto out;
4094 	} else if (off + len > zp->z_size) {
4095 		int npages = btopr(zp->z_size - off);
4096 		page_t *trunc;
4097 
4098 		page_list_break(&pp, &trunc, npages);
4099 		/* ignore pages past end of file */
4100 		if (trunc)
4101 			pvn_write_done(trunc, flags);
4102 		len = zp->z_size - off;
4103 	}
4104 
4105 	if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4106 	    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4107 		err = EDQUOT;
4108 		goto out;
4109 	}
4110 top:
4111 	tx = dmu_tx_create(zfsvfs->z_os);
4112 	dmu_tx_hold_write(tx, zp->z_id, off, len);
4113 
4114 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4115 	zfs_sa_upgrade_txholds(tx, zp);
4116 	err = dmu_tx_assign(tx, TXG_NOWAIT);
4117 	if (err != 0) {
4118 		if (err == ERESTART) {
4119 			dmu_tx_wait(tx);
4120 			dmu_tx_abort(tx);
4121 			goto top;
4122 		}
4123 		dmu_tx_abort(tx);
4124 		goto out;
4125 	}
4126 
4127 	if (zp->z_blksz <= PAGESIZE) {
4128 		caddr_t va = zfs_map_page(pp, S_READ);
4129 		ASSERT3U(len, <=, PAGESIZE);
4130 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4131 		zfs_unmap_page(pp, va);
4132 	} else {
4133 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4134 	}
4135 
4136 	if (err == 0) {
4137 		uint64_t mtime[2], ctime[2];
4138 		sa_bulk_attr_t bulk[3];
4139 		int count = 0;
4140 
4141 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4142 		    &mtime, 16);
4143 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4144 		    &ctime, 16);
4145 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4146 		    &zp->z_pflags, 8);
4147 		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4148 		    B_TRUE);
4149 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4150 	}
4151 	dmu_tx_commit(tx);
4152 
4153 out:
4154 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4155 	if (offp)
4156 		*offp = off;
4157 	if (lenp)
4158 		*lenp = len;
4159 
4160 	return (err);
4161 }
4162 
4163 /*
4164  * Copy the portion of the file indicated from pages into the file.
4165  * The pages are stored in a page list attached to the files vnode.
4166  *
4167  *	IN:	vp	- vnode of file to push page data to.
4168  *		off	- position in file to put data.
4169  *		len	- amount of data to write.
4170  *		flags	- flags to control the operation.
4171  *		cr	- credentials of caller.
4172  *		ct	- caller context.
4173  *
4174  *	RETURN:	0 if success
4175  *		error code if failure
4176  *
4177  * Timestamps:
4178  *	vp - ctime|mtime updated
4179  */
4180 /*ARGSUSED*/
4181 static int
4182 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4183     caller_context_t *ct)
4184 {
4185 	znode_t		*zp = VTOZ(vp);
4186 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4187 	page_t		*pp;
4188 	size_t		io_len;
4189 	u_offset_t	io_off;
4190 	uint_t		blksz;
4191 	rl_t		*rl;
4192 	int		error = 0;
4193 
4194 	ZFS_ENTER(zfsvfs);
4195 	ZFS_VERIFY_ZP(zp);
4196 
4197 	/*
4198 	 * There's nothing to do if no data is cached.
4199 	 */
4200 	if (!vn_has_cached_data(vp)) {
4201 		ZFS_EXIT(zfsvfs);
4202 		return (0);
4203 	}
4204 
4205 	/*
4206 	 * Align this request to the file block size in case we kluster.
4207 	 * XXX - this can result in pretty aggresive locking, which can
4208 	 * impact simultanious read/write access.  One option might be
4209 	 * to break up long requests (len == 0) into block-by-block
4210 	 * operations to get narrower locking.
4211 	 */
4212 	blksz = zp->z_blksz;
4213 	if (ISP2(blksz))
4214 		io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4215 	else
4216 		io_off = 0;
4217 	if (len > 0 && ISP2(blksz))
4218 		io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4219 	else
4220 		io_len = 0;
4221 
4222 	if (io_len == 0) {
4223 		/*
4224 		 * Search the entire vp list for pages >= io_off.
4225 		 */
4226 		rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4227 		error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4228 		goto out;
4229 	}
4230 	rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4231 
4232 	if (off > zp->z_size) {
4233 		/* past end of file */
4234 		zfs_range_unlock(rl);
4235 		ZFS_EXIT(zfsvfs);
4236 		return (0);
4237 	}
4238 
4239 	len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4240 
4241 	for (off = io_off; io_off < off + len; io_off += io_len) {
4242 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4243 			pp = page_lookup(vp, io_off,
4244 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4245 		} else {
4246 			pp = page_lookup_nowait(vp, io_off,
4247 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4248 		}
4249 
4250 		if (pp != NULL && pvn_getdirty(pp, flags)) {
4251 			int err;
4252 
4253 			/*
4254 			 * Found a dirty page to push
4255 			 */
4256 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4257 			if (err)
4258 				error = err;
4259 		} else {
4260 			io_len = PAGESIZE;
4261 		}
4262 	}
4263 out:
4264 	zfs_range_unlock(rl);
4265 	if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4266 		zil_commit(zfsvfs->z_log, zp->z_id);
4267 	ZFS_EXIT(zfsvfs);
4268 	return (error);
4269 }
4270 
4271 /*ARGSUSED*/
4272 void
4273 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4274 {
4275 	znode_t	*zp = VTOZ(vp);
4276 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4277 	int error;
4278 
4279 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4280 	if (zp->z_sa_hdl == NULL) {
4281 		/*
4282 		 * The fs has been unmounted, or we did a
4283 		 * suspend/resume and this file no longer exists.
4284 		 */
4285 		if (vn_has_cached_data(vp)) {
4286 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4287 			    B_INVAL, cr);
4288 		}
4289 
4290 		mutex_enter(&zp->z_lock);
4291 		mutex_enter(&vp->v_lock);
4292 		ASSERT(vp->v_count == 1);
4293 		vp->v_count = 0;
4294 		mutex_exit(&vp->v_lock);
4295 		mutex_exit(&zp->z_lock);
4296 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
4297 		zfs_znode_free(zp);
4298 		return;
4299 	}
4300 
4301 	/*
4302 	 * Attempt to push any data in the page cache.  If this fails
4303 	 * we will get kicked out later in zfs_zinactive().
4304 	 */
4305 	if (vn_has_cached_data(vp)) {
4306 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4307 		    cr);
4308 	}
4309 
4310 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4311 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4312 
4313 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4314 		zfs_sa_upgrade_txholds(tx, zp);
4315 		error = dmu_tx_assign(tx, TXG_WAIT);
4316 		if (error) {
4317 			dmu_tx_abort(tx);
4318 		} else {
4319 			mutex_enter(&zp->z_lock);
4320 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4321 			    (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4322 			zp->z_atime_dirty = 0;
4323 			mutex_exit(&zp->z_lock);
4324 			dmu_tx_commit(tx);
4325 		}
4326 	}
4327 
4328 	zfs_zinactive(zp);
4329 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
4330 }
4331 
4332 /*
4333  * Bounds-check the seek operation.
4334  *
4335  *	IN:	vp	- vnode seeking within
4336  *		ooff	- old file offset
4337  *		noffp	- pointer to new file offset
4338  *		ct	- caller context
4339  *
4340  *	RETURN:	0 if success
4341  *		EINVAL if new offset invalid
4342  */
4343 /* ARGSUSED */
4344 static int
4345 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4346     caller_context_t *ct)
4347 {
4348 	if (vp->v_type == VDIR)
4349 		return (0);
4350 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4351 }
4352 
4353 /*
4354  * Pre-filter the generic locking function to trap attempts to place
4355  * a mandatory lock on a memory mapped file.
4356  */
4357 static int
4358 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4359     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4360 {
4361 	znode_t *zp = VTOZ(vp);
4362 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4363 
4364 	ZFS_ENTER(zfsvfs);
4365 	ZFS_VERIFY_ZP(zp);
4366 
4367 	/*
4368 	 * We are following the UFS semantics with respect to mapcnt
4369 	 * here: If we see that the file is mapped already, then we will
4370 	 * return an error, but we don't worry about races between this
4371 	 * function and zfs_map().
4372 	 */
4373 	if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4374 		ZFS_EXIT(zfsvfs);
4375 		return (EAGAIN);
4376 	}
4377 	ZFS_EXIT(zfsvfs);
4378 	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4379 }
4380 
4381 /*
4382  * If we can't find a page in the cache, we will create a new page
4383  * and fill it with file data.  For efficiency, we may try to fill
4384  * multiple pages at once (klustering) to fill up the supplied page
4385  * list.  Note that the pages to be filled are held with an exclusive
4386  * lock to prevent access by other threads while they are being filled.
4387  */
4388 static int
4389 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4390     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4391 {
4392 	znode_t *zp = VTOZ(vp);
4393 	page_t *pp, *cur_pp;
4394 	objset_t *os = zp->z_zfsvfs->z_os;
4395 	u_offset_t io_off, total;
4396 	size_t io_len;
4397 	int err;
4398 
4399 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4400 		/*
4401 		 * We only have a single page, don't bother klustering
4402 		 */
4403 		io_off = off;
4404 		io_len = PAGESIZE;
4405 		pp = page_create_va(vp, io_off, io_len,
4406 		    PG_EXCL | PG_WAIT, seg, addr);
4407 	} else {
4408 		/*
4409 		 * Try to find enough pages to fill the page list
4410 		 */
4411 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4412 		    &io_len, off, plsz, 0);
4413 	}
4414 	if (pp == NULL) {
4415 		/*
4416 		 * The page already exists, nothing to do here.
4417 		 */
4418 		*pl = NULL;
4419 		return (0);
4420 	}
4421 
4422 	/*
4423 	 * Fill the pages in the kluster.
4424 	 */
4425 	cur_pp = pp;
4426 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4427 		caddr_t va;
4428 
4429 		ASSERT3U(io_off, ==, cur_pp->p_offset);
4430 		va = zfs_map_page(cur_pp, S_WRITE);
4431 		err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4432 		    DMU_READ_PREFETCH);
4433 		zfs_unmap_page(cur_pp, va);
4434 		if (err) {
4435 			/* On error, toss the entire kluster */
4436 			pvn_read_done(pp, B_ERROR);
4437 			/* convert checksum errors into IO errors */
4438 			if (err == ECKSUM)
4439 				err = EIO;
4440 			return (err);
4441 		}
4442 		cur_pp = cur_pp->p_next;
4443 	}
4444 
4445 	/*
4446 	 * Fill in the page list array from the kluster starting
4447 	 * from the desired offset `off'.
4448 	 * NOTE: the page list will always be null terminated.
4449 	 */
4450 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4451 	ASSERT(pl == NULL || (*pl)->p_offset == off);
4452 
4453 	return (0);
4454 }
4455 
4456 /*
4457  * Return pointers to the pages for the file region [off, off + len]
4458  * in the pl array.  If plsz is greater than len, this function may
4459  * also return page pointers from after the specified region
4460  * (i.e. the region [off, off + plsz]).  These additional pages are
4461  * only returned if they are already in the cache, or were created as
4462  * part of a klustered read.
4463  *
4464  *	IN:	vp	- vnode of file to get data from.
4465  *		off	- position in file to get data from.
4466  *		len	- amount of data to retrieve.
4467  *		plsz	- length of provided page list.
4468  *		seg	- segment to obtain pages for.
4469  *		addr	- virtual address of fault.
4470  *		rw	- mode of created pages.
4471  *		cr	- credentials of caller.
4472  *		ct	- caller context.
4473  *
4474  *	OUT:	protp	- protection mode of created pages.
4475  *		pl	- list of pages created.
4476  *
4477  *	RETURN:	0 if success
4478  *		error code if failure
4479  *
4480  * Timestamps:
4481  *	vp - atime updated
4482  */
4483 /* ARGSUSED */
4484 static int
4485 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4486 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4487 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4488 {
4489 	znode_t		*zp = VTOZ(vp);
4490 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4491 	page_t		**pl0 = pl;
4492 	int		err = 0;
4493 
4494 	/* we do our own caching, faultahead is unnecessary */
4495 	if (pl == NULL)
4496 		return (0);
4497 	else if (len > plsz)
4498 		len = plsz;
4499 	else
4500 		len = P2ROUNDUP(len, PAGESIZE);
4501 	ASSERT(plsz >= len);
4502 
4503 	ZFS_ENTER(zfsvfs);
4504 	ZFS_VERIFY_ZP(zp);
4505 
4506 	if (protp)
4507 		*protp = PROT_ALL;
4508 
4509 	/*
4510 	 * Loop through the requested range [off, off + len) looking
4511 	 * for pages.  If we don't find a page, we will need to create
4512 	 * a new page and fill it with data from the file.
4513 	 */
4514 	while (len > 0) {
4515 		if (*pl = page_lookup(vp, off, SE_SHARED))
4516 			*(pl+1) = NULL;
4517 		else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4518 			goto out;
4519 		while (*pl) {
4520 			ASSERT3U((*pl)->p_offset, ==, off);
4521 			off += PAGESIZE;
4522 			addr += PAGESIZE;
4523 			if (len > 0) {
4524 				ASSERT3U(len, >=, PAGESIZE);
4525 				len -= PAGESIZE;
4526 			}
4527 			ASSERT3U(plsz, >=, PAGESIZE);
4528 			plsz -= PAGESIZE;
4529 			pl++;
4530 		}
4531 	}
4532 
4533 	/*
4534 	 * Fill out the page array with any pages already in the cache.
4535 	 */
4536 	while (plsz > 0 &&
4537 	    (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4538 			off += PAGESIZE;
4539 			plsz -= PAGESIZE;
4540 	}
4541 out:
4542 	if (err) {
4543 		/*
4544 		 * Release any pages we have previously locked.
4545 		 */
4546 		while (pl > pl0)
4547 			page_unlock(*--pl);
4548 	} else {
4549 		ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4550 	}
4551 
4552 	*pl = NULL;
4553 
4554 	ZFS_EXIT(zfsvfs);
4555 	return (err);
4556 }
4557 
4558 /*
4559  * Request a memory map for a section of a file.  This code interacts
4560  * with common code and the VM system as follows:
4561  *
4562  *	common code calls mmap(), which ends up in smmap_common()
4563  *
4564  *	this calls VOP_MAP(), which takes you into (say) zfs
4565  *
4566  *	zfs_map() calls as_map(), passing segvn_create() as the callback
4567  *
4568  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
4569  *
4570  *	zfs_addmap() updates z_mapcnt
4571  */
4572 /*ARGSUSED*/
4573 static int
4574 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4575     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4576     caller_context_t *ct)
4577 {
4578 	znode_t *zp = VTOZ(vp);
4579 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4580 	segvn_crargs_t	vn_a;
4581 	int		error;
4582 
4583 	ZFS_ENTER(zfsvfs);
4584 	ZFS_VERIFY_ZP(zp);
4585 
4586 	if ((prot & PROT_WRITE) && (zp->z_pflags &
4587 	    (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4588 		ZFS_EXIT(zfsvfs);
4589 		return (EPERM);
4590 	}
4591 
4592 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4593 	    (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4594 		ZFS_EXIT(zfsvfs);
4595 		return (EACCES);
4596 	}
4597 
4598 	if (vp->v_flag & VNOMAP) {
4599 		ZFS_EXIT(zfsvfs);
4600 		return (ENOSYS);
4601 	}
4602 
4603 	if (off < 0 || len > MAXOFFSET_T - off) {
4604 		ZFS_EXIT(zfsvfs);
4605 		return (ENXIO);
4606 	}
4607 
4608 	if (vp->v_type != VREG) {
4609 		ZFS_EXIT(zfsvfs);
4610 		return (ENODEV);
4611 	}
4612 
4613 	/*
4614 	 * If file is locked, disallow mapping.
4615 	 */
4616 	if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4617 		ZFS_EXIT(zfsvfs);
4618 		return (EAGAIN);
4619 	}
4620 
4621 	as_rangelock(as);
4622 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4623 	if (error != 0) {
4624 		as_rangeunlock(as);
4625 		ZFS_EXIT(zfsvfs);
4626 		return (error);
4627 	}
4628 
4629 	vn_a.vp = vp;
4630 	vn_a.offset = (u_offset_t)off;
4631 	vn_a.type = flags & MAP_TYPE;
4632 	vn_a.prot = prot;
4633 	vn_a.maxprot = maxprot;
4634 	vn_a.cred = cr;
4635 	vn_a.amp = NULL;
4636 	vn_a.flags = flags & ~MAP_TYPE;
4637 	vn_a.szc = 0;
4638 	vn_a.lgrp_mem_policy_flags = 0;
4639 
4640 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4641 
4642 	as_rangeunlock(as);
4643 	ZFS_EXIT(zfsvfs);
4644 	return (error);
4645 }
4646 
4647 /* ARGSUSED */
4648 static int
4649 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4650     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4651     caller_context_t *ct)
4652 {
4653 	uint64_t pages = btopr(len);
4654 
4655 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4656 	return (0);
4657 }
4658 
4659 /*
4660  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4661  * more accurate mtime for the associated file.  Since we don't have a way of
4662  * detecting when the data was actually modified, we have to resort to
4663  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4664  * last page is pushed.  The problem occurs when the msync() call is omitted,
4665  * which by far the most common case:
4666  *
4667  * 	open()
4668  * 	mmap()
4669  * 	<modify memory>
4670  * 	munmap()
4671  * 	close()
4672  * 	<time lapse>
4673  * 	putpage() via fsflush
4674  *
4675  * If we wait until fsflush to come along, we can have a modification time that
4676  * is some arbitrary point in the future.  In order to prevent this in the
4677  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4678  * torn down.
4679  */
4680 /* ARGSUSED */
4681 static int
4682 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4683     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4684     caller_context_t *ct)
4685 {
4686 	uint64_t pages = btopr(len);
4687 
4688 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4689 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4690 
4691 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4692 	    vn_has_cached_data(vp))
4693 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4694 
4695 	return (0);
4696 }
4697 
4698 /*
4699  * Free or allocate space in a file.  Currently, this function only
4700  * supports the `F_FREESP' command.  However, this command is somewhat
4701  * misnamed, as its functionality includes the ability to allocate as
4702  * well as free space.
4703  *
4704  *	IN:	vp	- vnode of file to free data in.
4705  *		cmd	- action to take (only F_FREESP supported).
4706  *		bfp	- section of file to free/alloc.
4707  *		flag	- current file open mode flags.
4708  *		offset	- current file offset.
4709  *		cr	- credentials of caller [UNUSED].
4710  *		ct	- caller context.
4711  *
4712  *	RETURN:	0 if success
4713  *		error code if failure
4714  *
4715  * Timestamps:
4716  *	vp - ctime|mtime updated
4717  */
4718 /* ARGSUSED */
4719 static int
4720 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4721     offset_t offset, cred_t *cr, caller_context_t *ct)
4722 {
4723 	znode_t		*zp = VTOZ(vp);
4724 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4725 	uint64_t	off, len;
4726 	int		error;
4727 
4728 	ZFS_ENTER(zfsvfs);
4729 	ZFS_VERIFY_ZP(zp);
4730 
4731 	if (cmd != F_FREESP) {
4732 		ZFS_EXIT(zfsvfs);
4733 		return (EINVAL);
4734 	}
4735 
4736 	if (error = convoff(vp, bfp, 0, offset)) {
4737 		ZFS_EXIT(zfsvfs);
4738 		return (error);
4739 	}
4740 
4741 	if (bfp->l_len < 0) {
4742 		ZFS_EXIT(zfsvfs);
4743 		return (EINVAL);
4744 	}
4745 
4746 	off = bfp->l_start;
4747 	len = bfp->l_len; /* 0 means from off to end of file */
4748 
4749 	error = zfs_freesp(zp, off, len, flag, TRUE);
4750 
4751 	ZFS_EXIT(zfsvfs);
4752 	return (error);
4753 }
4754 
4755 /*ARGSUSED*/
4756 static int
4757 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4758 {
4759 	znode_t		*zp = VTOZ(vp);
4760 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4761 	uint32_t	gen;
4762 	uint64_t	gen64;
4763 	uint64_t	object = zp->z_id;
4764 	zfid_short_t	*zfid;
4765 	int		size, i, error;
4766 
4767 	ZFS_ENTER(zfsvfs);
4768 	ZFS_VERIFY_ZP(zp);
4769 
4770 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4771 	    &gen64, sizeof (uint64_t))) != 0) {
4772 		ZFS_EXIT(zfsvfs);
4773 		return (error);
4774 	}
4775 
4776 	gen = (uint32_t)gen64;
4777 
4778 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4779 	if (fidp->fid_len < size) {
4780 		fidp->fid_len = size;
4781 		ZFS_EXIT(zfsvfs);
4782 		return (ENOSPC);
4783 	}
4784 
4785 	zfid = (zfid_short_t *)fidp;
4786 
4787 	zfid->zf_len = size;
4788 
4789 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4790 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4791 
4792 	/* Must have a non-zero generation number to distinguish from .zfs */
4793 	if (gen == 0)
4794 		gen = 1;
4795 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4796 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4797 
4798 	if (size == LONG_FID_LEN) {
4799 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4800 		zfid_long_t	*zlfid;
4801 
4802 		zlfid = (zfid_long_t *)fidp;
4803 
4804 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4805 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4806 
4807 		/* XXX - this should be the generation number for the objset */
4808 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4809 			zlfid->zf_setgen[i] = 0;
4810 	}
4811 
4812 	ZFS_EXIT(zfsvfs);
4813 	return (0);
4814 }
4815 
4816 static int
4817 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4818     caller_context_t *ct)
4819 {
4820 	znode_t		*zp, *xzp;
4821 	zfsvfs_t	*zfsvfs;
4822 	zfs_dirlock_t	*dl;
4823 	int		error;
4824 
4825 	switch (cmd) {
4826 	case _PC_LINK_MAX:
4827 		*valp = ULONG_MAX;
4828 		return (0);
4829 
4830 	case _PC_FILESIZEBITS:
4831 		*valp = 64;
4832 		return (0);
4833 
4834 	case _PC_XATTR_EXISTS:
4835 		zp = VTOZ(vp);
4836 		zfsvfs = zp->z_zfsvfs;
4837 		ZFS_ENTER(zfsvfs);
4838 		ZFS_VERIFY_ZP(zp);
4839 		*valp = 0;
4840 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4841 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4842 		if (error == 0) {
4843 			zfs_dirent_unlock(dl);
4844 			if (!zfs_dirempty(xzp))
4845 				*valp = 1;
4846 			VN_RELE(ZTOV(xzp));
4847 		} else if (error == ENOENT) {
4848 			/*
4849 			 * If there aren't extended attributes, it's the
4850 			 * same as having zero of them.
4851 			 */
4852 			error = 0;
4853 		}
4854 		ZFS_EXIT(zfsvfs);
4855 		return (error);
4856 
4857 	case _PC_SATTR_ENABLED:
4858 	case _PC_SATTR_EXISTS:
4859 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4860 		    (vp->v_type == VREG || vp->v_type == VDIR);
4861 		return (0);
4862 
4863 	case _PC_ACCESS_FILTERING:
4864 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4865 		    vp->v_type == VDIR;
4866 		return (0);
4867 
4868 	case _PC_ACL_ENABLED:
4869 		*valp = _ACL_ACE_ENABLED;
4870 		return (0);
4871 
4872 	case _PC_MIN_HOLE_SIZE:
4873 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4874 		return (0);
4875 
4876 	case _PC_TIMESTAMP_RESOLUTION:
4877 		/* nanosecond timestamp resolution */
4878 		*valp = 1L;
4879 		return (0);
4880 
4881 	default:
4882 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4883 	}
4884 }
4885 
4886 /*ARGSUSED*/
4887 static int
4888 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4889     caller_context_t *ct)
4890 {
4891 	znode_t *zp = VTOZ(vp);
4892 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4893 	int error;
4894 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4895 
4896 	ZFS_ENTER(zfsvfs);
4897 	ZFS_VERIFY_ZP(zp);
4898 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4899 	ZFS_EXIT(zfsvfs);
4900 
4901 	return (error);
4902 }
4903 
4904 /*ARGSUSED*/
4905 static int
4906 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4907     caller_context_t *ct)
4908 {
4909 	znode_t *zp = VTOZ(vp);
4910 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4911 	int error;
4912 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4913 	zilog_t	*zilog = zfsvfs->z_log;
4914 
4915 	ZFS_ENTER(zfsvfs);
4916 	ZFS_VERIFY_ZP(zp);
4917 
4918 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4919 
4920 	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4921 		zil_commit(zilog, 0);
4922 
4923 	ZFS_EXIT(zfsvfs);
4924 	return (error);
4925 }
4926 
4927 /*
4928  * Tunable, both must be a power of 2.
4929  *
4930  * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4931  * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4932  *                an arcbuf for a partial block read
4933  */
4934 int zcr_blksz_min = (1 << 10);	/* 1K */
4935 int zcr_blksz_max = (1 << 17);	/* 128K */
4936 
4937 /*ARGSUSED*/
4938 static int
4939 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
4940     caller_context_t *ct)
4941 {
4942 	znode_t	*zp = VTOZ(vp);
4943 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4944 	int max_blksz = zfsvfs->z_max_blksz;
4945 	uio_t *uio = &xuio->xu_uio;
4946 	ssize_t size = uio->uio_resid;
4947 	offset_t offset = uio->uio_loffset;
4948 	int blksz;
4949 	int fullblk, i;
4950 	arc_buf_t *abuf;
4951 	ssize_t maxsize;
4952 	int preamble, postamble;
4953 
4954 	if (xuio->xu_type != UIOTYPE_ZEROCOPY)
4955 		return (EINVAL);
4956 
4957 	ZFS_ENTER(zfsvfs);
4958 	ZFS_VERIFY_ZP(zp);
4959 	switch (ioflag) {
4960 	case UIO_WRITE:
4961 		/*
4962 		 * Loan out an arc_buf for write if write size is bigger than
4963 		 * max_blksz, and the file's block size is also max_blksz.
4964 		 */
4965 		blksz = max_blksz;
4966 		if (size < blksz || zp->z_blksz != blksz) {
4967 			ZFS_EXIT(zfsvfs);
4968 			return (EINVAL);
4969 		}
4970 		/*
4971 		 * Caller requests buffers for write before knowing where the
4972 		 * write offset might be (e.g. NFS TCP write).
4973 		 */
4974 		if (offset == -1) {
4975 			preamble = 0;
4976 		} else {
4977 			preamble = P2PHASE(offset, blksz);
4978 			if (preamble) {
4979 				preamble = blksz - preamble;
4980 				size -= preamble;
4981 			}
4982 		}
4983 
4984 		postamble = P2PHASE(size, blksz);
4985 		size -= postamble;
4986 
4987 		fullblk = size / blksz;
4988 		(void) dmu_xuio_init(xuio,
4989 		    (preamble != 0) + fullblk + (postamble != 0));
4990 		DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
4991 		    int, postamble, int,
4992 		    (preamble != 0) + fullblk + (postamble != 0));
4993 
4994 		/*
4995 		 * Have to fix iov base/len for partial buffers.  They
4996 		 * currently represent full arc_buf's.
4997 		 */
4998 		if (preamble) {
4999 			/* data begins in the middle of the arc_buf */
5000 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5001 			    blksz);
5002 			ASSERT(abuf);
5003 			(void) dmu_xuio_add(xuio, abuf,
5004 			    blksz - preamble, preamble);
5005 		}
5006 
5007 		for (i = 0; i < fullblk; i++) {
5008 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5009 			    blksz);
5010 			ASSERT(abuf);
5011 			(void) dmu_xuio_add(xuio, abuf, 0, blksz);
5012 		}
5013 
5014 		if (postamble) {
5015 			/* data ends in the middle of the arc_buf */
5016 			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5017 			    blksz);
5018 			ASSERT(abuf);
5019 			(void) dmu_xuio_add(xuio, abuf, 0, postamble);
5020 		}
5021 		break;
5022 	case UIO_READ:
5023 		/*
5024 		 * Loan out an arc_buf for read if the read size is larger than
5025 		 * the current file block size.  Block alignment is not
5026 		 * considered.  Partial arc_buf will be loaned out for read.
5027 		 */
5028 		blksz = zp->z_blksz;
5029 		if (blksz < zcr_blksz_min)
5030 			blksz = zcr_blksz_min;
5031 		if (blksz > zcr_blksz_max)
5032 			blksz = zcr_blksz_max;
5033 		/* avoid potential complexity of dealing with it */
5034 		if (blksz > max_blksz) {
5035 			ZFS_EXIT(zfsvfs);
5036 			return (EINVAL);
5037 		}
5038 
5039 		maxsize = zp->z_size - uio->uio_loffset;
5040 		if (size > maxsize)
5041 			size = maxsize;
5042 
5043 		if (size < blksz || vn_has_cached_data(vp)) {
5044 			ZFS_EXIT(zfsvfs);
5045 			return (EINVAL);
5046 		}
5047 		break;
5048 	default:
5049 		ZFS_EXIT(zfsvfs);
5050 		return (EINVAL);
5051 	}
5052 
5053 	uio->uio_extflg = UIO_XUIO;
5054 	XUIO_XUZC_RW(xuio) = ioflag;
5055 	ZFS_EXIT(zfsvfs);
5056 	return (0);
5057 }
5058 
5059 /*ARGSUSED*/
5060 static int
5061 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5062 {
5063 	int i;
5064 	arc_buf_t *abuf;
5065 	int ioflag = XUIO_XUZC_RW(xuio);
5066 
5067 	ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5068 
5069 	i = dmu_xuio_cnt(xuio);
5070 	while (i-- > 0) {
5071 		abuf = dmu_xuio_arcbuf(xuio, i);
5072 		/*
5073 		 * if abuf == NULL, it must be a write buffer
5074 		 * that has been returned in zfs_write().
5075 		 */
5076 		if (abuf)
5077 			dmu_return_arcbuf(abuf);
5078 		ASSERT(abuf || ioflag == UIO_WRITE);
5079 	}
5080 
5081 	dmu_xuio_fini(xuio);
5082 	return (0);
5083 }
5084 
5085 /*
5086  * Predeclare these here so that the compiler assumes that
5087  * this is an "old style" function declaration that does
5088  * not include arguments => we won't get type mismatch errors
5089  * in the initializations that follow.
5090  */
5091 static int zfs_inval();
5092 static int zfs_isdir();
5093 
5094 static int
5095 zfs_inval()
5096 {
5097 	return (EINVAL);
5098 }
5099 
5100 static int
5101 zfs_isdir()
5102 {
5103 	return (EISDIR);
5104 }
5105 /*
5106  * Directory vnode operations template
5107  */
5108 vnodeops_t *zfs_dvnodeops;
5109 const fs_operation_def_t zfs_dvnodeops_template[] = {
5110 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5111 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5112 	VOPNAME_READ,		{ .error = zfs_isdir },
5113 	VOPNAME_WRITE,		{ .error = zfs_isdir },
5114 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5115 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5116 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5117 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5118 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5119 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5120 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5121 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5122 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5123 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
5124 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5125 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5126 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
5127 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5128 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5129 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5130 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5131 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5132 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5133 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5134 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
5135 	NULL,			NULL
5136 };
5137 
5138 /*
5139  * Regular file vnode operations template
5140  */
5141 vnodeops_t *zfs_fvnodeops;
5142 const fs_operation_def_t zfs_fvnodeops_template[] = {
5143 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5144 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5145 	VOPNAME_READ,		{ .vop_read = zfs_read },
5146 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
5147 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5148 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5149 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5150 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5151 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5152 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5153 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5154 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5155 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5156 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5157 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
5158 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
5159 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
5160 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
5161 	VOPNAME_MAP,		{ .vop_map = zfs_map },
5162 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
5163 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
5164 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5165 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5166 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5167 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5168 	VOPNAME_REQZCBUF, 	{ .vop_reqzcbuf = zfs_reqzcbuf },
5169 	VOPNAME_RETZCBUF, 	{ .vop_retzcbuf = zfs_retzcbuf },
5170 	NULL,			NULL
5171 };
5172 
5173 /*
5174  * Symbolic link vnode operations template
5175  */
5176 vnodeops_t *zfs_symvnodeops;
5177 const fs_operation_def_t zfs_symvnodeops_template[] = {
5178 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5179 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5180 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5181 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5182 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
5183 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5184 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5185 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5186 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5187 	NULL,			NULL
5188 };
5189 
5190 /*
5191  * special share hidden files vnode operations template
5192  */
5193 vnodeops_t *zfs_sharevnodeops;
5194 const fs_operation_def_t zfs_sharevnodeops_template[] = {
5195 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5196 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5197 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5198 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5199 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5200 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5201 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5202 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5203 	NULL,			NULL
5204 };
5205 
5206 /*
5207  * Extended attribute directory vnode operations template
5208  *	This template is identical to the directory vnodes
5209  *	operation template except for restricted operations:
5210  *		VOP_MKDIR()
5211  *		VOP_SYMLINK()
5212  * Note that there are other restrictions embedded in:
5213  *	zfs_create()	- restrict type to VREG
5214  *	zfs_link()	- no links into/out of attribute space
5215  *	zfs_rename()	- no moves into/out of attribute space
5216  */
5217 vnodeops_t *zfs_xdvnodeops;
5218 const fs_operation_def_t zfs_xdvnodeops_template[] = {
5219 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5220 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5221 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5222 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5223 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5224 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5225 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5226 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5227 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5228 	VOPNAME_LINK,		{ .vop_link = zfs_link },
5229 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5230 	VOPNAME_MKDIR,		{ .error = zfs_inval },
5231 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5232 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5233 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
5234 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5235 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5236 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5237 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5238 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5239 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5240 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5241 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5242 	NULL,			NULL
5243 };
5244 
5245 /*
5246  * Error vnode operations template
5247  */
5248 vnodeops_t *zfs_evnodeops;
5249 const fs_operation_def_t zfs_evnodeops_template[] = {
5250 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5251 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5252 	NULL,			NULL
5253 };
5254