xref: /titanic_52/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 12351196aec1e03e183b97d65c2221e93e10ff3b)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /* Portions Copyright 2007 Jeremy Teo */
27 
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/time.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
33 #include <sys/resource.h>
34 #include <sys/vfs.h>
35 #include <sys/vfs_opreg.h>
36 #include <sys/vnode.h>
37 #include <sys/file.h>
38 #include <sys/stat.h>
39 #include <sys/kmem.h>
40 #include <sys/taskq.h>
41 #include <sys/uio.h>
42 #include <sys/vmsystm.h>
43 #include <sys/atomic.h>
44 #include <sys/vm.h>
45 #include <vm/seg_vn.h>
46 #include <vm/pvn.h>
47 #include <vm/as.h>
48 #include <vm/kpm.h>
49 #include <vm/seg_kpm.h>
50 #include <sys/mman.h>
51 #include <sys/pathname.h>
52 #include <sys/cmn_err.h>
53 #include <sys/errno.h>
54 #include <sys/unistd.h>
55 #include <sys/zfs_dir.h>
56 #include <sys/zfs_acl.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/fs/zfs.h>
59 #include <sys/dmu.h>
60 #include <sys/spa.h>
61 #include <sys/txg.h>
62 #include <sys/dbuf.h>
63 #include <sys/zap.h>
64 #include <sys/dirent.h>
65 #include <sys/policy.h>
66 #include <sys/sunddi.h>
67 #include <sys/filio.h>
68 #include "fs/fs_subr.h"
69 #include <sys/zfs_ctldir.h>
70 #include <sys/zfs_fuid.h>
71 #include <sys/dnlc.h>
72 #include <sys/zfs_rlock.h>
73 #include <sys/extdirent.h>
74 #include <sys/kidmap.h>
75 #include <sys/cred_impl.h>
76 #include <sys/attr.h>
77 
78 /*
79  * Programming rules.
80  *
81  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
82  * properly lock its in-core state, create a DMU transaction, do the work,
83  * record this work in the intent log (ZIL), commit the DMU transaction,
84  * and wait for the intent log to commit if it is a synchronous operation.
85  * Moreover, the vnode ops must work in both normal and log replay context.
86  * The ordering of events is important to avoid deadlocks and references
87  * to freed memory.  The example below illustrates the following Big Rules:
88  *
89  *  (1) A check must be made in each zfs thread for a mounted file system.
90  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
91  *      A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
92  *      must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
93  *      can return EIO from the calling function.
94  *
95  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
96  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
97  *	First, if it's the last reference, the vnode/znode
98  *	can be freed, so the zp may point to freed memory.  Second, the last
99  *	reference will call zfs_zinactive(), which may induce a lot of work --
100  *	pushing cached pages (which acquires range locks) and syncing out
101  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
102  *	which could deadlock the system if you were already holding one.
103  *
104  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
105  *	as they can span dmu_tx_assign() calls.
106  *
107  *  (4)	Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign().
108  *	In normal operation, this will be TXG_NOWAIT.  During ZIL replay,
109  *	it will be a specific txg.  Either way, dmu_tx_assign() never blocks.
110  *	This is critical because we don't want to block while holding locks.
111  *	Note, in particular, that if a lock is sometimes acquired before
112  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
113  *	use a non-blocking assign can deadlock the system.  The scenario:
114  *
115  *	Thread A has grabbed a lock before calling dmu_tx_assign().
116  *	Thread B is in an already-assigned tx, and blocks for this lock.
117  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
118  *	forever, because the previous txg can't quiesce until B's tx commits.
119  *
120  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
121  *	then drop all locks, call dmu_tx_wait(), and try again.
122  *
123  *  (5)	If the operation succeeded, generate the intent log entry for it
124  *	before dropping locks.  This ensures that the ordering of events
125  *	in the intent log matches the order in which they actually occurred.
126  *
127  *  (6)	At the end of each vnode op, the DMU tx must always commit,
128  *	regardless of whether there were any errors.
129  *
130  *  (7)	After dropping all locks, invoke zil_commit(zilog, seq, foid)
131  *	to ensure that synchronous semantics are provided when necessary.
132  *
133  * In general, this is how things should be ordered in each vnode op:
134  *
135  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
136  * top:
137  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
138  *	rw_enter(...);			// grab any other locks you need
139  *	tx = dmu_tx_create(...);	// get DMU tx
140  *	dmu_tx_hold_*();		// hold each object you might modify
141  *	error = dmu_tx_assign(tx, zfsvfs->z_assign);	// try to assign
142  *	if (error) {
143  *		rw_exit(...);		// drop locks
144  *		zfs_dirent_unlock(dl);	// unlock directory entry
145  *		VN_RELE(...);		// release held vnodes
146  *		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
147  *			dmu_tx_wait(tx);
148  *			dmu_tx_abort(tx);
149  *			goto top;
150  *		}
151  *		dmu_tx_abort(tx);	// abort DMU tx
152  *		ZFS_EXIT(zfsvfs);	// finished in zfs
153  *		return (error);		// really out of space
154  *	}
155  *	error = do_real_work();		// do whatever this VOP does
156  *	if (error == 0)
157  *		zfs_log_*(...);		// on success, make ZIL entry
158  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
159  *	rw_exit(...);			// drop locks
160  *	zfs_dirent_unlock(dl);		// unlock directory entry
161  *	VN_RELE(...);			// release held vnodes
162  *	zil_commit(zilog, seq, foid);	// synchronous when necessary
163  *	ZFS_EXIT(zfsvfs);		// finished in zfs
164  *	return (error);			// done, report error
165  */
166 
167 /* ARGSUSED */
168 static int
169 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
170 {
171 	znode_t	*zp = VTOZ(*vpp);
172 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
173 
174 	ZFS_ENTER(zfsvfs);
175 	ZFS_VERIFY_ZP(zp);
176 
177 	if ((flag & FWRITE) && (zp->z_phys->zp_flags & ZFS_APPENDONLY) &&
178 	    ((flag & FAPPEND) == 0)) {
179 		ZFS_EXIT(zfsvfs);
180 		return (EPERM);
181 	}
182 
183 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
184 	    ZTOV(zp)->v_type == VREG &&
185 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
186 	    zp->z_phys->zp_size > 0) {
187 		if (fs_vscan(*vpp, cr, 0) != 0) {
188 			ZFS_EXIT(zfsvfs);
189 			return (EACCES);
190 		}
191 	}
192 
193 	/* Keep a count of the synchronous opens in the znode */
194 	if (flag & (FSYNC | FDSYNC))
195 		atomic_inc_32(&zp->z_sync_cnt);
196 
197 	ZFS_EXIT(zfsvfs);
198 	return (0);
199 }
200 
201 /* ARGSUSED */
202 static int
203 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
204     caller_context_t *ct)
205 {
206 	znode_t	*zp = VTOZ(vp);
207 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
208 
209 	ZFS_ENTER(zfsvfs);
210 	ZFS_VERIFY_ZP(zp);
211 
212 	/* Decrement the synchronous opens in the znode */
213 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
214 		atomic_dec_32(&zp->z_sync_cnt);
215 
216 	/*
217 	 * Clean up any locks held by this process on the vp.
218 	 */
219 	cleanlocks(vp, ddi_get_pid(), 0);
220 	cleanshares(vp, ddi_get_pid());
221 
222 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
223 	    ZTOV(zp)->v_type == VREG &&
224 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
225 	    zp->z_phys->zp_size > 0)
226 		VERIFY(fs_vscan(vp, cr, 1) == 0);
227 
228 	ZFS_EXIT(zfsvfs);
229 	return (0);
230 }
231 
232 /*
233  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
234  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
235  */
236 static int
237 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
238 {
239 	znode_t	*zp = VTOZ(vp);
240 	uint64_t noff = (uint64_t)*off; /* new offset */
241 	uint64_t file_sz;
242 	int error;
243 	boolean_t hole;
244 
245 	file_sz = zp->z_phys->zp_size;
246 	if (noff >= file_sz)  {
247 		return (ENXIO);
248 	}
249 
250 	if (cmd == _FIO_SEEK_HOLE)
251 		hole = B_TRUE;
252 	else
253 		hole = B_FALSE;
254 
255 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
256 
257 	/* end of file? */
258 	if ((error == ESRCH) || (noff > file_sz)) {
259 		/*
260 		 * Handle the virtual hole at the end of file.
261 		 */
262 		if (hole) {
263 			*off = file_sz;
264 			return (0);
265 		}
266 		return (ENXIO);
267 	}
268 
269 	if (noff < *off)
270 		return (error);
271 	*off = noff;
272 	return (error);
273 }
274 
275 /* ARGSUSED */
276 static int
277 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
278     int *rvalp, caller_context_t *ct)
279 {
280 	offset_t off;
281 	int error;
282 	zfsvfs_t *zfsvfs;
283 	znode_t *zp;
284 
285 	switch (com) {
286 	case _FIOFFS:
287 		return (zfs_sync(vp->v_vfsp, 0, cred));
288 
289 		/*
290 		 * The following two ioctls are used by bfu.  Faking out,
291 		 * necessary to avoid bfu errors.
292 		 */
293 	case _FIOGDIO:
294 	case _FIOSDIO:
295 		return (0);
296 
297 	case _FIO_SEEK_DATA:
298 	case _FIO_SEEK_HOLE:
299 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
300 			return (EFAULT);
301 
302 		zp = VTOZ(vp);
303 		zfsvfs = zp->z_zfsvfs;
304 		ZFS_ENTER(zfsvfs);
305 		ZFS_VERIFY_ZP(zp);
306 
307 		/* offset parameter is in/out */
308 		error = zfs_holey(vp, com, &off);
309 		ZFS_EXIT(zfsvfs);
310 		if (error)
311 			return (error);
312 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
313 			return (EFAULT);
314 		return (0);
315 	}
316 	return (ENOTTY);
317 }
318 
319 /*
320  * Utility functions to map and unmap a single physical page.  These
321  * are used to manage the mappable copies of ZFS file data, and therefore
322  * do not update ref/mod bits.
323  */
324 caddr_t
325 zfs_map_page(page_t *pp, enum seg_rw rw)
326 {
327 	if (kpm_enable)
328 		return (hat_kpm_mapin(pp, 0));
329 	ASSERT(rw == S_READ || rw == S_WRITE);
330 	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
331 	    (caddr_t)-1));
332 }
333 
334 void
335 zfs_unmap_page(page_t *pp, caddr_t addr)
336 {
337 	if (kpm_enable) {
338 		hat_kpm_mapout(pp, 0, addr);
339 	} else {
340 		ppmapout(addr);
341 	}
342 }
343 
344 /*
345  * When a file is memory mapped, we must keep the IO data synchronized
346  * between the DMU cache and the memory mapped pages.  What this means:
347  *
348  * On Write:	If we find a memory mapped page, we write to *both*
349  *		the page and the dmu buffer.
350  *
351  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
352  *	the file is memory mapped.
353  */
354 static int
355 mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx)
356 {
357 	znode_t	*zp = VTOZ(vp);
358 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
359 	int64_t	start, off;
360 	int len = nbytes;
361 	int error = 0;
362 
363 	start = uio->uio_loffset;
364 	off = start & PAGEOFFSET;
365 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
366 		page_t *pp;
367 		uint64_t bytes = MIN(PAGESIZE - off, len);
368 		uint64_t woff = uio->uio_loffset;
369 
370 		/*
371 		 * We don't want a new page to "appear" in the middle of
372 		 * the file update (because it may not get the write
373 		 * update data), so we grab a lock to block
374 		 * zfs_getpage().
375 		 */
376 		rw_enter(&zp->z_map_lock, RW_WRITER);
377 		if (pp = page_lookup(vp, start, SE_SHARED)) {
378 			caddr_t va;
379 
380 			rw_exit(&zp->z_map_lock);
381 			va = zfs_map_page(pp, S_WRITE);
382 			error = uiomove(va+off, bytes, UIO_WRITE, uio);
383 			if (error == 0) {
384 				dmu_write(zfsvfs->z_os, zp->z_id,
385 				    woff, bytes, va+off, tx);
386 			}
387 			zfs_unmap_page(pp, va);
388 			page_unlock(pp);
389 		} else {
390 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
391 			    uio, bytes, tx);
392 			rw_exit(&zp->z_map_lock);
393 		}
394 		len -= bytes;
395 		off = 0;
396 		if (error)
397 			break;
398 	}
399 	return (error);
400 }
401 
402 /*
403  * When a file is memory mapped, we must keep the IO data synchronized
404  * between the DMU cache and the memory mapped pages.  What this means:
405  *
406  * On Read:	We "read" preferentially from memory mapped pages,
407  *		else we default from the dmu buffer.
408  *
409  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
410  *	the file is memory mapped.
411  */
412 static int
413 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
414 {
415 	znode_t *zp = VTOZ(vp);
416 	objset_t *os = zp->z_zfsvfs->z_os;
417 	int64_t	start, off;
418 	int len = nbytes;
419 	int error = 0;
420 
421 	start = uio->uio_loffset;
422 	off = start & PAGEOFFSET;
423 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
424 		page_t *pp;
425 		uint64_t bytes = MIN(PAGESIZE - off, len);
426 
427 		if (pp = page_lookup(vp, start, SE_SHARED)) {
428 			caddr_t va;
429 
430 			va = zfs_map_page(pp, S_READ);
431 			error = uiomove(va + off, bytes, UIO_READ, uio);
432 			zfs_unmap_page(pp, va);
433 			page_unlock(pp);
434 		} else {
435 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
436 		}
437 		len -= bytes;
438 		off = 0;
439 		if (error)
440 			break;
441 	}
442 	return (error);
443 }
444 
445 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
446 
447 /*
448  * Read bytes from specified file into supplied buffer.
449  *
450  *	IN:	vp	- vnode of file to be read from.
451  *		uio	- structure supplying read location, range info,
452  *			  and return buffer.
453  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
454  *		cr	- credentials of caller.
455  *		ct	- caller context
456  *
457  *	OUT:	uio	- updated offset and range, buffer filled.
458  *
459  *	RETURN:	0 if success
460  *		error code if failure
461  *
462  * Side Effects:
463  *	vp - atime updated if byte count > 0
464  */
465 /* ARGSUSED */
466 static int
467 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
468 {
469 	znode_t		*zp = VTOZ(vp);
470 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
471 	objset_t	*os;
472 	ssize_t		n, nbytes;
473 	int		error;
474 	rl_t		*rl;
475 
476 	ZFS_ENTER(zfsvfs);
477 	ZFS_VERIFY_ZP(zp);
478 	os = zfsvfs->z_os;
479 
480 	if (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) {
481 		ZFS_EXIT(zfsvfs);
482 		return (EACCES);
483 	}
484 
485 	/*
486 	 * Validate file offset
487 	 */
488 	if (uio->uio_loffset < (offset_t)0) {
489 		ZFS_EXIT(zfsvfs);
490 		return (EINVAL);
491 	}
492 
493 	/*
494 	 * Fasttrack empty reads
495 	 */
496 	if (uio->uio_resid == 0) {
497 		ZFS_EXIT(zfsvfs);
498 		return (0);
499 	}
500 
501 	/*
502 	 * Check for mandatory locks
503 	 */
504 	if (MANDMODE((mode_t)zp->z_phys->zp_mode)) {
505 		if (error = chklock(vp, FREAD,
506 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
507 			ZFS_EXIT(zfsvfs);
508 			return (error);
509 		}
510 	}
511 
512 	/*
513 	 * If we're in FRSYNC mode, sync out this znode before reading it.
514 	 */
515 	if (ioflag & FRSYNC)
516 		zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
517 
518 	/*
519 	 * Lock the range against changes.
520 	 */
521 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
522 
523 	/*
524 	 * If we are reading past end-of-file we can skip
525 	 * to the end; but we might still need to set atime.
526 	 */
527 	if (uio->uio_loffset >= zp->z_phys->zp_size) {
528 		error = 0;
529 		goto out;
530 	}
531 
532 	ASSERT(uio->uio_loffset < zp->z_phys->zp_size);
533 	n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset);
534 
535 	while (n > 0) {
536 		nbytes = MIN(n, zfs_read_chunk_size -
537 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
538 
539 		if (vn_has_cached_data(vp))
540 			error = mappedread(vp, nbytes, uio);
541 		else
542 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
543 		if (error) {
544 			/* convert checksum errors into IO errors */
545 			if (error == ECKSUM)
546 				error = EIO;
547 			break;
548 		}
549 
550 		n -= nbytes;
551 	}
552 
553 out:
554 	zfs_range_unlock(rl);
555 
556 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
557 	ZFS_EXIT(zfsvfs);
558 	return (error);
559 }
560 
561 /*
562  * Fault in the pages of the first n bytes specified by the uio structure.
563  * 1 byte in each page is touched and the uio struct is unmodified.
564  * Any error will exit this routine as this is only a best
565  * attempt to get the pages resident. This is a copy of ufs_trans_touch().
566  */
567 static void
568 zfs_prefault_write(ssize_t n, struct uio *uio)
569 {
570 	struct iovec *iov;
571 	ulong_t cnt, incr;
572 	caddr_t p;
573 	uint8_t tmp;
574 
575 	iov = uio->uio_iov;
576 
577 	while (n) {
578 		cnt = MIN(iov->iov_len, n);
579 		if (cnt == 0) {
580 			/* empty iov entry */
581 			iov++;
582 			continue;
583 		}
584 		n -= cnt;
585 		/*
586 		 * touch each page in this segment.
587 		 */
588 		p = iov->iov_base;
589 		while (cnt) {
590 			switch (uio->uio_segflg) {
591 			case UIO_USERSPACE:
592 			case UIO_USERISPACE:
593 				if (fuword8(p, &tmp))
594 					return;
595 				break;
596 			case UIO_SYSSPACE:
597 				if (kcopy(p, &tmp, 1))
598 					return;
599 				break;
600 			}
601 			incr = MIN(cnt, PAGESIZE);
602 			p += incr;
603 			cnt -= incr;
604 		}
605 		/*
606 		 * touch the last byte in case it straddles a page.
607 		 */
608 		p--;
609 		switch (uio->uio_segflg) {
610 		case UIO_USERSPACE:
611 		case UIO_USERISPACE:
612 			if (fuword8(p, &tmp))
613 				return;
614 			break;
615 		case UIO_SYSSPACE:
616 			if (kcopy(p, &tmp, 1))
617 				return;
618 			break;
619 		}
620 		iov++;
621 	}
622 }
623 
624 /*
625  * Write the bytes to a file.
626  *
627  *	IN:	vp	- vnode of file to be written to.
628  *		uio	- structure supplying write location, range info,
629  *			  and data buffer.
630  *		ioflag	- FAPPEND flag set if in append mode.
631  *		cr	- credentials of caller.
632  *		ct	- caller context (NFS/CIFS fem monitor only)
633  *
634  *	OUT:	uio	- updated offset and range.
635  *
636  *	RETURN:	0 if success
637  *		error code if failure
638  *
639  * Timestamps:
640  *	vp - ctime|mtime updated if byte count > 0
641  */
642 /* ARGSUSED */
643 static int
644 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
645 {
646 	znode_t		*zp = VTOZ(vp);
647 	rlim64_t	limit = uio->uio_llimit;
648 	ssize_t		start_resid = uio->uio_resid;
649 	ssize_t		tx_bytes;
650 	uint64_t	end_size;
651 	dmu_tx_t	*tx;
652 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
653 	zilog_t		*zilog;
654 	offset_t	woff;
655 	ssize_t		n, nbytes;
656 	rl_t		*rl;
657 	int		max_blksz = zfsvfs->z_max_blksz;
658 	uint64_t	pflags;
659 	int		error;
660 
661 	/*
662 	 * Fasttrack empty write
663 	 */
664 	n = start_resid;
665 	if (n == 0)
666 		return (0);
667 
668 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
669 		limit = MAXOFFSET_T;
670 
671 	ZFS_ENTER(zfsvfs);
672 	ZFS_VERIFY_ZP(zp);
673 
674 	/*
675 	 * If immutable or not appending then return EPERM
676 	 */
677 	pflags = zp->z_phys->zp_flags;
678 	if ((pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
679 	    ((pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
680 	    (uio->uio_loffset < zp->z_phys->zp_size))) {
681 		ZFS_EXIT(zfsvfs);
682 		return (EPERM);
683 	}
684 
685 	zilog = zfsvfs->z_log;
686 
687 	/*
688 	 * Pre-fault the pages to ensure slow (eg NFS) pages
689 	 * don't hold up txg.
690 	 */
691 	zfs_prefault_write(n, uio);
692 
693 	/*
694 	 * If in append mode, set the io offset pointer to eof.
695 	 */
696 	if (ioflag & FAPPEND) {
697 		/*
698 		 * Range lock for a file append:
699 		 * The value for the start of range will be determined by
700 		 * zfs_range_lock() (to guarantee append semantics).
701 		 * If this write will cause the block size to increase,
702 		 * zfs_range_lock() will lock the entire file, so we must
703 		 * later reduce the range after we grow the block size.
704 		 */
705 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
706 		if (rl->r_len == UINT64_MAX) {
707 			/* overlocked, zp_size can't change */
708 			woff = uio->uio_loffset = zp->z_phys->zp_size;
709 		} else {
710 			woff = uio->uio_loffset = rl->r_off;
711 		}
712 	} else {
713 		woff = uio->uio_loffset;
714 		/*
715 		 * Validate file offset
716 		 */
717 		if (woff < 0) {
718 			ZFS_EXIT(zfsvfs);
719 			return (EINVAL);
720 		}
721 
722 		/*
723 		 * If we need to grow the block size then zfs_range_lock()
724 		 * will lock a wider range than we request here.
725 		 * Later after growing the block size we reduce the range.
726 		 */
727 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
728 	}
729 
730 	if (woff >= limit) {
731 		zfs_range_unlock(rl);
732 		ZFS_EXIT(zfsvfs);
733 		return (EFBIG);
734 	}
735 
736 	if ((woff + n) > limit || woff > (limit - n))
737 		n = limit - woff;
738 
739 	/*
740 	 * Check for mandatory locks
741 	 */
742 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) &&
743 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
744 		zfs_range_unlock(rl);
745 		ZFS_EXIT(zfsvfs);
746 		return (error);
747 	}
748 	end_size = MAX(zp->z_phys->zp_size, woff + n);
749 
750 	/*
751 	 * Write the file in reasonable size chunks.  Each chunk is written
752 	 * in a separate transaction; this keeps the intent log records small
753 	 * and allows us to do more fine-grained space accounting.
754 	 */
755 	while (n > 0) {
756 		/*
757 		 * Start a transaction.
758 		 */
759 		woff = uio->uio_loffset;
760 		tx = dmu_tx_create(zfsvfs->z_os);
761 		dmu_tx_hold_bonus(tx, zp->z_id);
762 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
763 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
764 		if (error) {
765 			if (error == ERESTART &&
766 			    zfsvfs->z_assign == TXG_NOWAIT) {
767 				dmu_tx_wait(tx);
768 				dmu_tx_abort(tx);
769 				continue;
770 			}
771 			dmu_tx_abort(tx);
772 			break;
773 		}
774 
775 		/*
776 		 * If zfs_range_lock() over-locked we grow the blocksize
777 		 * and then reduce the lock range.  This will only happen
778 		 * on the first iteration since zfs_range_reduce() will
779 		 * shrink down r_len to the appropriate size.
780 		 */
781 		if (rl->r_len == UINT64_MAX) {
782 			uint64_t new_blksz;
783 
784 			if (zp->z_blksz > max_blksz) {
785 				ASSERT(!ISP2(zp->z_blksz));
786 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
787 			} else {
788 				new_blksz = MIN(end_size, max_blksz);
789 			}
790 			zfs_grow_blocksize(zp, new_blksz, tx);
791 			zfs_range_reduce(rl, woff, n);
792 		}
793 
794 		/*
795 		 * XXX - should we really limit each write to z_max_blksz?
796 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
797 		 */
798 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
799 		rw_enter(&zp->z_map_lock, RW_READER);
800 
801 		tx_bytes = uio->uio_resid;
802 		if (vn_has_cached_data(vp)) {
803 			rw_exit(&zp->z_map_lock);
804 			error = mappedwrite(vp, nbytes, uio, tx);
805 		} else {
806 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
807 			    uio, nbytes, tx);
808 			rw_exit(&zp->z_map_lock);
809 		}
810 		tx_bytes -= uio->uio_resid;
811 
812 		/*
813 		 * If we made no progress, we're done.  If we made even
814 		 * partial progress, update the znode and ZIL accordingly.
815 		 */
816 		if (tx_bytes == 0) {
817 			dmu_tx_commit(tx);
818 			ASSERT(error != 0);
819 			break;
820 		}
821 
822 		/*
823 		 * Clear Set-UID/Set-GID bits on successful write if not
824 		 * privileged and at least one of the excute bits is set.
825 		 *
826 		 * It would be nice to to this after all writes have
827 		 * been done, but that would still expose the ISUID/ISGID
828 		 * to another app after the partial write is committed.
829 		 *
830 		 * Note: we don't call zfs_fuid_map_id() here because
831 		 * user 0 is not an ephemeral uid.
832 		 */
833 		mutex_enter(&zp->z_acl_lock);
834 		if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) |
835 		    (S_IXUSR >> 6))) != 0 &&
836 		    (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 &&
837 		    secpolicy_vnode_setid_retain(cr,
838 		    (zp->z_phys->zp_mode & S_ISUID) != 0 &&
839 		    zp->z_phys->zp_uid == 0) != 0) {
840 			zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID);
841 		}
842 		mutex_exit(&zp->z_acl_lock);
843 
844 		/*
845 		 * Update time stamp.  NOTE: This marks the bonus buffer as
846 		 * dirty, so we don't have to do it again for zp_size.
847 		 */
848 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
849 
850 		/*
851 		 * Update the file size (zp_size) if it has changed;
852 		 * account for possible concurrent updates.
853 		 */
854 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset)
855 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
856 			    uio->uio_loffset);
857 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
858 		dmu_tx_commit(tx);
859 
860 		if (error != 0)
861 			break;
862 		ASSERT(tx_bytes == nbytes);
863 		n -= nbytes;
864 	}
865 
866 	zfs_range_unlock(rl);
867 
868 	/*
869 	 * If we're in replay mode, or we made no progress, return error.
870 	 * Otherwise, it's at least a partial write, so it's successful.
871 	 */
872 	if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) {
873 		ZFS_EXIT(zfsvfs);
874 		return (error);
875 	}
876 
877 	if (ioflag & (FSYNC | FDSYNC))
878 		zil_commit(zilog, zp->z_last_itx, zp->z_id);
879 
880 	ZFS_EXIT(zfsvfs);
881 	return (0);
882 }
883 
884 void
885 zfs_get_done(dmu_buf_t *db, void *vzgd)
886 {
887 	zgd_t *zgd = (zgd_t *)vzgd;
888 	rl_t *rl = zgd->zgd_rl;
889 	vnode_t *vp = ZTOV(rl->r_zp);
890 
891 	dmu_buf_rele(db, vzgd);
892 	zfs_range_unlock(rl);
893 	VN_RELE(vp);
894 	zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
895 	kmem_free(zgd, sizeof (zgd_t));
896 }
897 
898 /*
899  * Get data to generate a TX_WRITE intent log record.
900  */
901 int
902 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
903 {
904 	zfsvfs_t *zfsvfs = arg;
905 	objset_t *os = zfsvfs->z_os;
906 	znode_t *zp;
907 	uint64_t off = lr->lr_offset;
908 	dmu_buf_t *db;
909 	rl_t *rl;
910 	zgd_t *zgd;
911 	int dlen = lr->lr_length;		/* length of user data */
912 	int error = 0;
913 
914 	ASSERT(zio);
915 	ASSERT(dlen != 0);
916 
917 	/*
918 	 * Nothing to do if the file has been removed
919 	 */
920 	if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0)
921 		return (ENOENT);
922 	if (zp->z_unlinked) {
923 		VN_RELE(ZTOV(zp));
924 		return (ENOENT);
925 	}
926 
927 	/*
928 	 * Write records come in two flavors: immediate and indirect.
929 	 * For small writes it's cheaper to store the data with the
930 	 * log record (immediate); for large writes it's cheaper to
931 	 * sync the data and get a pointer to it (indirect) so that
932 	 * we don't have to write the data twice.
933 	 */
934 	if (buf != NULL) { /* immediate write */
935 		rl = zfs_range_lock(zp, off, dlen, RL_READER);
936 		/* test for truncation needs to be done while range locked */
937 		if (off >= zp->z_phys->zp_size) {
938 			error = ENOENT;
939 			goto out;
940 		}
941 		VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf));
942 	} else { /* indirect write */
943 		uint64_t boff; /* block starting offset */
944 
945 		/*
946 		 * Have to lock the whole block to ensure when it's
947 		 * written out and it's checksum is being calculated
948 		 * that no one can change the data. We need to re-check
949 		 * blocksize after we get the lock in case it's changed!
950 		 */
951 		for (;;) {
952 			if (ISP2(zp->z_blksz)) {
953 				boff = P2ALIGN_TYPED(off, zp->z_blksz,
954 				    uint64_t);
955 			} else {
956 				boff = 0;
957 			}
958 			dlen = zp->z_blksz;
959 			rl = zfs_range_lock(zp, boff, dlen, RL_READER);
960 			if (zp->z_blksz == dlen)
961 				break;
962 			zfs_range_unlock(rl);
963 		}
964 		/* test for truncation needs to be done while range locked */
965 		if (off >= zp->z_phys->zp_size) {
966 			error = ENOENT;
967 			goto out;
968 		}
969 		zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
970 		zgd->zgd_rl = rl;
971 		zgd->zgd_zilog = zfsvfs->z_log;
972 		zgd->zgd_bp = &lr->lr_blkptr;
973 		VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db));
974 		ASSERT(boff == db->db_offset);
975 		lr->lr_blkoff = off - boff;
976 		error = dmu_sync(zio, db, &lr->lr_blkptr,
977 		    lr->lr_common.lrc_txg, zfs_get_done, zgd);
978 		ASSERT((error && error != EINPROGRESS) ||
979 		    lr->lr_length <= zp->z_blksz);
980 		if (error == 0)
981 			zil_add_block(zfsvfs->z_log, &lr->lr_blkptr);
982 		/*
983 		 * If we get EINPROGRESS, then we need to wait for a
984 		 * write IO initiated by dmu_sync() to complete before
985 		 * we can release this dbuf.  We will finish everything
986 		 * up in the zfs_get_done() callback.
987 		 */
988 		if (error == EINPROGRESS)
989 			return (0);
990 		dmu_buf_rele(db, zgd);
991 		kmem_free(zgd, sizeof (zgd_t));
992 	}
993 out:
994 	zfs_range_unlock(rl);
995 	VN_RELE(ZTOV(zp));
996 	return (error);
997 }
998 
999 /*ARGSUSED*/
1000 static int
1001 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1002     caller_context_t *ct)
1003 {
1004 	znode_t *zp = VTOZ(vp);
1005 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1006 	int error;
1007 
1008 	ZFS_ENTER(zfsvfs);
1009 	ZFS_VERIFY_ZP(zp);
1010 
1011 	if (flag & V_ACE_MASK)
1012 		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1013 	else
1014 		error = zfs_zaccess_rwx(zp, mode, flag, cr);
1015 
1016 	ZFS_EXIT(zfsvfs);
1017 	return (error);
1018 }
1019 
1020 /*
1021  * Lookup an entry in a directory, or an extended attribute directory.
1022  * If it exists, return a held vnode reference for it.
1023  *
1024  *	IN:	dvp	- vnode of directory to search.
1025  *		nm	- name of entry to lookup.
1026  *		pnp	- full pathname to lookup [UNUSED].
1027  *		flags	- LOOKUP_XATTR set if looking for an attribute.
1028  *		rdir	- root directory vnode [UNUSED].
1029  *		cr	- credentials of caller.
1030  *		ct	- caller context
1031  *		direntflags - directory lookup flags
1032  *		realpnp - returned pathname.
1033  *
1034  *	OUT:	vpp	- vnode of located entry, NULL if not found.
1035  *
1036  *	RETURN:	0 if success
1037  *		error code if failure
1038  *
1039  * Timestamps:
1040  *	NA
1041  */
1042 /* ARGSUSED */
1043 static int
1044 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1045     int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1046     int *direntflags, pathname_t *realpnp)
1047 {
1048 	znode_t *zdp = VTOZ(dvp);
1049 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1050 	int	error;
1051 
1052 	ZFS_ENTER(zfsvfs);
1053 	ZFS_VERIFY_ZP(zdp);
1054 
1055 	*vpp = NULL;
1056 
1057 	if (flags & LOOKUP_XATTR) {
1058 		/*
1059 		 * If the xattr property is off, refuse the lookup request.
1060 		 */
1061 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1062 			ZFS_EXIT(zfsvfs);
1063 			return (EINVAL);
1064 		}
1065 
1066 		/*
1067 		 * We don't allow recursive attributes..
1068 		 * Maybe someday we will.
1069 		 */
1070 		if (zdp->z_phys->zp_flags & ZFS_XATTR) {
1071 			ZFS_EXIT(zfsvfs);
1072 			return (EINVAL);
1073 		}
1074 
1075 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1076 			ZFS_EXIT(zfsvfs);
1077 			return (error);
1078 		}
1079 
1080 		/*
1081 		 * Do we have permission to get into attribute directory?
1082 		 */
1083 
1084 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1085 		    B_FALSE, cr)) {
1086 			VN_RELE(*vpp);
1087 			*vpp = NULL;
1088 		}
1089 
1090 		ZFS_EXIT(zfsvfs);
1091 		return (error);
1092 	}
1093 
1094 	if (dvp->v_type != VDIR) {
1095 		ZFS_EXIT(zfsvfs);
1096 		return (ENOTDIR);
1097 	}
1098 
1099 	/*
1100 	 * Check accessibility of directory.
1101 	 */
1102 
1103 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1104 		ZFS_EXIT(zfsvfs);
1105 		return (error);
1106 	}
1107 
1108 	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1109 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1110 		ZFS_EXIT(zfsvfs);
1111 		return (EILSEQ);
1112 	}
1113 
1114 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1115 	if (error == 0) {
1116 		/*
1117 		 * Convert device special files
1118 		 */
1119 		if (IS_DEVVP(*vpp)) {
1120 			vnode_t	*svp;
1121 
1122 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1123 			VN_RELE(*vpp);
1124 			if (svp == NULL)
1125 				error = ENOSYS;
1126 			else
1127 				*vpp = svp;
1128 		}
1129 	}
1130 
1131 	ZFS_EXIT(zfsvfs);
1132 	return (error);
1133 }
1134 
1135 /*
1136  * Attempt to create a new entry in a directory.  If the entry
1137  * already exists, truncate the file if permissible, else return
1138  * an error.  Return the vp of the created or trunc'd file.
1139  *
1140  *	IN:	dvp	- vnode of directory to put new file entry in.
1141  *		name	- name of new file entry.
1142  *		vap	- attributes of new file.
1143  *		excl	- flag indicating exclusive or non-exclusive mode.
1144  *		mode	- mode to open file with.
1145  *		cr	- credentials of caller.
1146  *		flag	- large file flag [UNUSED].
1147  *		ct	- caller context
1148  *		vsecp 	- ACL to be set
1149  *
1150  *	OUT:	vpp	- vnode of created or trunc'd entry.
1151  *
1152  *	RETURN:	0 if success
1153  *		error code if failure
1154  *
1155  * Timestamps:
1156  *	dvp - ctime|mtime updated if new entry created
1157  *	 vp - ctime|mtime always, atime if new
1158  */
1159 
1160 /* ARGSUSED */
1161 static int
1162 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1163     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1164     vsecattr_t *vsecp)
1165 {
1166 	znode_t		*zp, *dzp = VTOZ(dvp);
1167 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1168 	zilog_t		*zilog;
1169 	objset_t	*os;
1170 	zfs_dirlock_t	*dl;
1171 	dmu_tx_t	*tx;
1172 	int		error;
1173 	zfs_acl_t	*aclp = NULL;
1174 	zfs_fuid_info_t *fuidp = NULL;
1175 
1176 	/*
1177 	 * If we have an ephemeral id, ACL, or XVATTR then
1178 	 * make sure file system is at proper version
1179 	 */
1180 
1181 	if (zfsvfs->z_use_fuids == B_FALSE &&
1182 	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1183 	    IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))))
1184 		return (EINVAL);
1185 
1186 	ZFS_ENTER(zfsvfs);
1187 	ZFS_VERIFY_ZP(dzp);
1188 	os = zfsvfs->z_os;
1189 	zilog = zfsvfs->z_log;
1190 
1191 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1192 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1193 		ZFS_EXIT(zfsvfs);
1194 		return (EILSEQ);
1195 	}
1196 
1197 	if (vap->va_mask & AT_XVATTR) {
1198 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1199 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1200 			ZFS_EXIT(zfsvfs);
1201 			return (error);
1202 		}
1203 	}
1204 top:
1205 	*vpp = NULL;
1206 
1207 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1208 		vap->va_mode &= ~VSVTX;
1209 
1210 	if (*name == '\0') {
1211 		/*
1212 		 * Null component name refers to the directory itself.
1213 		 */
1214 		VN_HOLD(dvp);
1215 		zp = dzp;
1216 		dl = NULL;
1217 		error = 0;
1218 	} else {
1219 		/* possible VN_HOLD(zp) */
1220 		int zflg = 0;
1221 
1222 		if (flag & FIGNORECASE)
1223 			zflg |= ZCILOOK;
1224 
1225 		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1226 		    NULL, NULL);
1227 		if (error) {
1228 			if (strcmp(name, "..") == 0)
1229 				error = EISDIR;
1230 			ZFS_EXIT(zfsvfs);
1231 			if (aclp)
1232 				zfs_acl_free(aclp);
1233 			return (error);
1234 		}
1235 	}
1236 	if (vsecp && aclp == NULL) {
1237 		error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp);
1238 		if (error) {
1239 			ZFS_EXIT(zfsvfs);
1240 			if (dl)
1241 				zfs_dirent_unlock(dl);
1242 			return (error);
1243 		}
1244 	}
1245 
1246 	if (zp == NULL) {
1247 		uint64_t txtype;
1248 
1249 		/*
1250 		 * Create a new file object and update the directory
1251 		 * to reference it.
1252 		 */
1253 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1254 			goto out;
1255 		}
1256 
1257 		/*
1258 		 * We only support the creation of regular files in
1259 		 * extended attribute directories.
1260 		 */
1261 		if ((dzp->z_phys->zp_flags & ZFS_XATTR) &&
1262 		    (vap->va_type != VREG)) {
1263 			error = EINVAL;
1264 			goto out;
1265 		}
1266 
1267 		tx = dmu_tx_create(os);
1268 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1269 		if ((aclp && aclp->z_has_fuids) || IS_EPHEMERAL(crgetuid(cr)) ||
1270 		    IS_EPHEMERAL(crgetgid(cr))) {
1271 			if (zfsvfs->z_fuid_obj == 0) {
1272 				dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1273 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1274 				    FUID_SIZE_ESTIMATE(zfsvfs));
1275 				dmu_tx_hold_zap(tx, MASTER_NODE_OBJ,
1276 				    FALSE, NULL);
1277 			} else {
1278 				dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
1279 				dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
1280 				    FUID_SIZE_ESTIMATE(zfsvfs));
1281 			}
1282 		}
1283 		dmu_tx_hold_bonus(tx, dzp->z_id);
1284 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1285 		if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp) {
1286 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1287 			    0, SPA_MAXBLOCKSIZE);
1288 		}
1289 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
1290 		if (error) {
1291 			zfs_dirent_unlock(dl);
1292 			if (error == ERESTART &&
1293 			    zfsvfs->z_assign == TXG_NOWAIT) {
1294 				dmu_tx_wait(tx);
1295 				dmu_tx_abort(tx);
1296 				goto top;
1297 			}
1298 			dmu_tx_abort(tx);
1299 			ZFS_EXIT(zfsvfs);
1300 			if (aclp)
1301 				zfs_acl_free(aclp);
1302 			return (error);
1303 		}
1304 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp);
1305 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1306 		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1307 		if (flag & FIGNORECASE)
1308 			txtype |= TX_CI;
1309 		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1310 		    vsecp, fuidp, vap);
1311 		if (fuidp)
1312 			zfs_fuid_info_free(fuidp);
1313 		dmu_tx_commit(tx);
1314 	} else {
1315 		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1316 
1317 		/*
1318 		 * A directory entry already exists for this name.
1319 		 */
1320 		/*
1321 		 * Can't truncate an existing file if in exclusive mode.
1322 		 */
1323 		if (excl == EXCL) {
1324 			error = EEXIST;
1325 			goto out;
1326 		}
1327 		/*
1328 		 * Can't open a directory for writing.
1329 		 */
1330 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1331 			error = EISDIR;
1332 			goto out;
1333 		}
1334 		/*
1335 		 * Verify requested access to file.
1336 		 */
1337 		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1338 			goto out;
1339 		}
1340 
1341 		mutex_enter(&dzp->z_lock);
1342 		dzp->z_seq++;
1343 		mutex_exit(&dzp->z_lock);
1344 
1345 		/*
1346 		 * Truncate regular files if requested.
1347 		 */
1348 		if ((ZTOV(zp)->v_type == VREG) &&
1349 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1350 			/* we can't hold any locks when calling zfs_freesp() */
1351 			zfs_dirent_unlock(dl);
1352 			dl = NULL;
1353 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1354 			if (error == 0) {
1355 				vnevent_create(ZTOV(zp), ct);
1356 			}
1357 		}
1358 	}
1359 out:
1360 
1361 	if (dl)
1362 		zfs_dirent_unlock(dl);
1363 
1364 	if (error) {
1365 		if (zp)
1366 			VN_RELE(ZTOV(zp));
1367 	} else {
1368 		*vpp = ZTOV(zp);
1369 		/*
1370 		 * If vnode is for a device return a specfs vnode instead.
1371 		 */
1372 		if (IS_DEVVP(*vpp)) {
1373 			struct vnode *svp;
1374 
1375 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1376 			VN_RELE(*vpp);
1377 			if (svp == NULL) {
1378 				error = ENOSYS;
1379 			}
1380 			*vpp = svp;
1381 		}
1382 	}
1383 	if (aclp)
1384 		zfs_acl_free(aclp);
1385 
1386 	ZFS_EXIT(zfsvfs);
1387 	return (error);
1388 }
1389 
1390 /*
1391  * Remove an entry from a directory.
1392  *
1393  *	IN:	dvp	- vnode of directory to remove entry from.
1394  *		name	- name of entry to remove.
1395  *		cr	- credentials of caller.
1396  *		ct	- caller context
1397  *		flags	- case flags
1398  *
1399  *	RETURN:	0 if success
1400  *		error code if failure
1401  *
1402  * Timestamps:
1403  *	dvp - ctime|mtime
1404  *	 vp - ctime (if nlink > 0)
1405  */
1406 /*ARGSUSED*/
1407 static int
1408 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1409     int flags)
1410 {
1411 	znode_t		*zp, *dzp = VTOZ(dvp);
1412 	znode_t		*xzp = NULL;
1413 	vnode_t		*vp;
1414 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1415 	zilog_t		*zilog;
1416 	uint64_t	acl_obj, xattr_obj;
1417 	zfs_dirlock_t	*dl;
1418 	dmu_tx_t	*tx;
1419 	boolean_t	may_delete_now, delete_now = FALSE;
1420 	boolean_t	unlinked, toobig = FALSE;
1421 	uint64_t	txtype;
1422 	pathname_t	*realnmp = NULL;
1423 	pathname_t	realnm;
1424 	int		error;
1425 	int		zflg = ZEXISTS;
1426 
1427 	ZFS_ENTER(zfsvfs);
1428 	ZFS_VERIFY_ZP(dzp);
1429 	zilog = zfsvfs->z_log;
1430 
1431 	if (flags & FIGNORECASE) {
1432 		zflg |= ZCILOOK;
1433 		pn_alloc(&realnm);
1434 		realnmp = &realnm;
1435 	}
1436 
1437 top:
1438 	/*
1439 	 * Attempt to lock directory; fail if entry doesn't exist.
1440 	 */
1441 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1442 	    NULL, realnmp)) {
1443 		if (realnmp)
1444 			pn_free(realnmp);
1445 		ZFS_EXIT(zfsvfs);
1446 		return (error);
1447 	}
1448 
1449 	vp = ZTOV(zp);
1450 
1451 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1452 		goto out;
1453 	}
1454 
1455 	/*
1456 	 * Need to use rmdir for removing directories.
1457 	 */
1458 	if (vp->v_type == VDIR) {
1459 		error = EPERM;
1460 		goto out;
1461 	}
1462 
1463 	vnevent_remove(vp, dvp, name, ct);
1464 
1465 	if (realnmp)
1466 		dnlc_remove(dvp, realnmp->pn_buf);
1467 	else
1468 		dnlc_remove(dvp, name);
1469 
1470 	mutex_enter(&vp->v_lock);
1471 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1472 	mutex_exit(&vp->v_lock);
1473 
1474 	/*
1475 	 * We may delete the znode now, or we may put it in the unlinked set;
1476 	 * it depends on whether we're the last link, and on whether there are
1477 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1478 	 * allow for either case.
1479 	 */
1480 	tx = dmu_tx_create(zfsvfs->z_os);
1481 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1482 	dmu_tx_hold_bonus(tx, zp->z_id);
1483 	if (may_delete_now) {
1484 		toobig =
1485 		    zp->z_phys->zp_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1486 		/* if the file is too big, only hold_free a token amount */
1487 		dmu_tx_hold_free(tx, zp->z_id, 0,
1488 		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1489 	}
1490 
1491 	/* are there any extended attributes? */
1492 	if ((xattr_obj = zp->z_phys->zp_xattr) != 0) {
1493 		/* XXX - do we need this if we are deleting? */
1494 		dmu_tx_hold_bonus(tx, xattr_obj);
1495 	}
1496 
1497 	/* are there any additional acls */
1498 	if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 &&
1499 	    may_delete_now)
1500 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1501 
1502 	/* charge as an update -- would be nice not to charge at all */
1503 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1504 
1505 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1506 	if (error) {
1507 		zfs_dirent_unlock(dl);
1508 		VN_RELE(vp);
1509 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1510 			dmu_tx_wait(tx);
1511 			dmu_tx_abort(tx);
1512 			goto top;
1513 		}
1514 		if (realnmp)
1515 			pn_free(realnmp);
1516 		dmu_tx_abort(tx);
1517 		ZFS_EXIT(zfsvfs);
1518 		return (error);
1519 	}
1520 
1521 	/*
1522 	 * Remove the directory entry.
1523 	 */
1524 	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1525 
1526 	if (error) {
1527 		dmu_tx_commit(tx);
1528 		goto out;
1529 	}
1530 
1531 	if (unlinked) {
1532 		mutex_enter(&vp->v_lock);
1533 		delete_now = may_delete_now && !toobig &&
1534 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1535 		    zp->z_phys->zp_xattr == xattr_obj &&
1536 		    zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj;
1537 		mutex_exit(&vp->v_lock);
1538 	}
1539 
1540 	if (delete_now) {
1541 		if (zp->z_phys->zp_xattr) {
1542 			error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
1543 			ASSERT3U(error, ==, 0);
1544 			ASSERT3U(xzp->z_phys->zp_links, ==, 2);
1545 			dmu_buf_will_dirty(xzp->z_dbuf, tx);
1546 			mutex_enter(&xzp->z_lock);
1547 			xzp->z_unlinked = 1;
1548 			xzp->z_phys->zp_links = 0;
1549 			mutex_exit(&xzp->z_lock);
1550 			zfs_unlinked_add(xzp, tx);
1551 			zp->z_phys->zp_xattr = 0; /* probably unnecessary */
1552 		}
1553 		mutex_enter(&zp->z_lock);
1554 		mutex_enter(&vp->v_lock);
1555 		vp->v_count--;
1556 		ASSERT3U(vp->v_count, ==, 0);
1557 		mutex_exit(&vp->v_lock);
1558 		mutex_exit(&zp->z_lock);
1559 		zfs_znode_delete(zp, tx);
1560 	} else if (unlinked) {
1561 		zfs_unlinked_add(zp, tx);
1562 	}
1563 
1564 	txtype = TX_REMOVE;
1565 	if (flags & FIGNORECASE)
1566 		txtype |= TX_CI;
1567 	zfs_log_remove(zilog, tx, txtype, dzp, name);
1568 
1569 	dmu_tx_commit(tx);
1570 out:
1571 	if (realnmp)
1572 		pn_free(realnmp);
1573 
1574 	zfs_dirent_unlock(dl);
1575 
1576 	if (!delete_now) {
1577 		VN_RELE(vp);
1578 	} else if (xzp) {
1579 		/* this rele is delayed to prevent nesting transactions */
1580 		VN_RELE(ZTOV(xzp));
1581 	}
1582 
1583 	ZFS_EXIT(zfsvfs);
1584 	return (error);
1585 }
1586 
1587 /*
1588  * Create a new directory and insert it into dvp using the name
1589  * provided.  Return a pointer to the inserted directory.
1590  *
1591  *	IN:	dvp	- vnode of directory to add subdir to.
1592  *		dirname	- name of new directory.
1593  *		vap	- attributes of new directory.
1594  *		cr	- credentials of caller.
1595  *		ct	- caller context
1596  *		vsecp	- ACL to be set
1597  *
1598  *	OUT:	vpp	- vnode of created directory.
1599  *
1600  *	RETURN:	0 if success
1601  *		error code if failure
1602  *
1603  * Timestamps:
1604  *	dvp - ctime|mtime updated
1605  *	 vp - ctime|mtime|atime updated
1606  */
1607 /*ARGSUSED*/
1608 static int
1609 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1610     caller_context_t *ct, int flags, vsecattr_t *vsecp)
1611 {
1612 	znode_t		*zp, *dzp = VTOZ(dvp);
1613 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1614 	zilog_t		*zilog;
1615 	zfs_dirlock_t	*dl;
1616 	uint64_t	txtype;
1617 	dmu_tx_t	*tx;
1618 	int		error;
1619 	zfs_acl_t	*aclp = NULL;
1620 	zfs_fuid_info_t	*fuidp = NULL;
1621 	int		zf = ZNEW;
1622 
1623 	ASSERT(vap->va_type == VDIR);
1624 
1625 	/*
1626 	 * If we have an ephemeral id, ACL, or XVATTR then
1627 	 * make sure file system is at proper version
1628 	 */
1629 
1630 	if (zfsvfs->z_use_fuids == B_FALSE &&
1631 	    (vsecp || (vap->va_mask & AT_XVATTR) || IS_EPHEMERAL(crgetuid(cr))||
1632 	    IS_EPHEMERAL(crgetgid(cr))))
1633 		return (EINVAL);
1634 
1635 	ZFS_ENTER(zfsvfs);
1636 	ZFS_VERIFY_ZP(dzp);
1637 	zilog = zfsvfs->z_log;
1638 
1639 	if (dzp->z_phys->zp_flags & ZFS_XATTR) {
1640 		ZFS_EXIT(zfsvfs);
1641 		return (EINVAL);
1642 	}
1643 
1644 	if (zfsvfs->z_utf8 && u8_validate(dirname,
1645 	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1646 		ZFS_EXIT(zfsvfs);
1647 		return (EILSEQ);
1648 	}
1649 	if (flags & FIGNORECASE)
1650 		zf |= ZCILOOK;
1651 
1652 	if (vap->va_mask & AT_XVATTR)
1653 		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1654 		    crgetuid(cr), cr, vap->va_type)) != 0) {
1655 			ZFS_EXIT(zfsvfs);
1656 			return (error);
1657 		}
1658 
1659 	/*
1660 	 * First make sure the new directory doesn't exist.
1661 	 */
1662 top:
1663 	*vpp = NULL;
1664 
1665 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1666 	    NULL, NULL)) {
1667 		ZFS_EXIT(zfsvfs);
1668 		return (error);
1669 	}
1670 
1671 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1672 		zfs_dirent_unlock(dl);
1673 		ZFS_EXIT(zfsvfs);
1674 		return (error);
1675 	}
1676 
1677 	if (vsecp && aclp == NULL) {
1678 		error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, &aclp);
1679 		if (error) {
1680 			zfs_dirent_unlock(dl);
1681 			ZFS_EXIT(zfsvfs);
1682 			return (error);
1683 		}
1684 	}
1685 	/*
1686 	 * Add a new entry to the directory.
1687 	 */
1688 	tx = dmu_tx_create(zfsvfs->z_os);
1689 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1690 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1691 	if ((aclp && aclp->z_has_fuids) || IS_EPHEMERAL(crgetuid(cr)) ||
1692 	    IS_EPHEMERAL(crgetgid(cr))) {
1693 		if (zfsvfs->z_fuid_obj == 0) {
1694 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1695 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1696 			    FUID_SIZE_ESTIMATE(zfsvfs));
1697 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
1698 		} else {
1699 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
1700 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
1701 			    FUID_SIZE_ESTIMATE(zfsvfs));
1702 		}
1703 	}
1704 	if ((dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) || aclp)
1705 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1706 		    0, SPA_MAXBLOCKSIZE);
1707 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1708 	if (error) {
1709 		zfs_dirent_unlock(dl);
1710 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1711 			dmu_tx_wait(tx);
1712 			dmu_tx_abort(tx);
1713 			goto top;
1714 		}
1715 		dmu_tx_abort(tx);
1716 		ZFS_EXIT(zfsvfs);
1717 		if (aclp)
1718 			zfs_acl_free(aclp);
1719 		return (error);
1720 	}
1721 
1722 	/*
1723 	 * Create new node.
1724 	 */
1725 	zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, aclp, &fuidp);
1726 
1727 	if (aclp)
1728 		zfs_acl_free(aclp);
1729 
1730 	/*
1731 	 * Now put new name in parent dir.
1732 	 */
1733 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1734 
1735 	*vpp = ZTOV(zp);
1736 
1737 	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
1738 	if (flags & FIGNORECASE)
1739 		txtype |= TX_CI;
1740 	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, fuidp, vap);
1741 
1742 	if (fuidp)
1743 		zfs_fuid_info_free(fuidp);
1744 	dmu_tx_commit(tx);
1745 
1746 	zfs_dirent_unlock(dl);
1747 
1748 	ZFS_EXIT(zfsvfs);
1749 	return (0);
1750 }
1751 
1752 /*
1753  * Remove a directory subdir entry.  If the current working
1754  * directory is the same as the subdir to be removed, the
1755  * remove will fail.
1756  *
1757  *	IN:	dvp	- vnode of directory to remove from.
1758  *		name	- name of directory to be removed.
1759  *		cwd	- vnode of current working directory.
1760  *		cr	- credentials of caller.
1761  *		ct	- caller context
1762  *		flags	- case flags
1763  *
1764  *	RETURN:	0 if success
1765  *		error code if failure
1766  *
1767  * Timestamps:
1768  *	dvp - ctime|mtime updated
1769  */
1770 /*ARGSUSED*/
1771 static int
1772 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
1773     caller_context_t *ct, int flags)
1774 {
1775 	znode_t		*dzp = VTOZ(dvp);
1776 	znode_t		*zp;
1777 	vnode_t		*vp;
1778 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1779 	zilog_t		*zilog;
1780 	zfs_dirlock_t	*dl;
1781 	dmu_tx_t	*tx;
1782 	int		error;
1783 	int		zflg = ZEXISTS;
1784 
1785 	ZFS_ENTER(zfsvfs);
1786 	ZFS_VERIFY_ZP(dzp);
1787 	zilog = zfsvfs->z_log;
1788 
1789 	if (flags & FIGNORECASE)
1790 		zflg |= ZCILOOK;
1791 top:
1792 	zp = NULL;
1793 
1794 	/*
1795 	 * Attempt to lock directory; fail if entry doesn't exist.
1796 	 */
1797 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1798 	    NULL, NULL)) {
1799 		ZFS_EXIT(zfsvfs);
1800 		return (error);
1801 	}
1802 
1803 	vp = ZTOV(zp);
1804 
1805 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1806 		goto out;
1807 	}
1808 
1809 	if (vp->v_type != VDIR) {
1810 		error = ENOTDIR;
1811 		goto out;
1812 	}
1813 
1814 	if (vp == cwd) {
1815 		error = EINVAL;
1816 		goto out;
1817 	}
1818 
1819 	vnevent_rmdir(vp, dvp, name, ct);
1820 
1821 	/*
1822 	 * Grab a lock on the directory to make sure that noone is
1823 	 * trying to add (or lookup) entries while we are removing it.
1824 	 */
1825 	rw_enter(&zp->z_name_lock, RW_WRITER);
1826 
1827 	/*
1828 	 * Grab a lock on the parent pointer to make sure we play well
1829 	 * with the treewalk and directory rename code.
1830 	 */
1831 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1832 
1833 	tx = dmu_tx_create(zfsvfs->z_os);
1834 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1835 	dmu_tx_hold_bonus(tx, zp->z_id);
1836 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1837 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1838 	if (error) {
1839 		rw_exit(&zp->z_parent_lock);
1840 		rw_exit(&zp->z_name_lock);
1841 		zfs_dirent_unlock(dl);
1842 		VN_RELE(vp);
1843 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1844 			dmu_tx_wait(tx);
1845 			dmu_tx_abort(tx);
1846 			goto top;
1847 		}
1848 		dmu_tx_abort(tx);
1849 		ZFS_EXIT(zfsvfs);
1850 		return (error);
1851 	}
1852 
1853 	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
1854 
1855 	if (error == 0) {
1856 		uint64_t txtype = TX_RMDIR;
1857 		if (flags & FIGNORECASE)
1858 			txtype |= TX_CI;
1859 		zfs_log_remove(zilog, tx, txtype, dzp, name);
1860 	}
1861 
1862 	dmu_tx_commit(tx);
1863 
1864 	rw_exit(&zp->z_parent_lock);
1865 	rw_exit(&zp->z_name_lock);
1866 out:
1867 	zfs_dirent_unlock(dl);
1868 
1869 	VN_RELE(vp);
1870 
1871 	ZFS_EXIT(zfsvfs);
1872 	return (error);
1873 }
1874 
1875 /*
1876  * Read as many directory entries as will fit into the provided
1877  * buffer from the given directory cursor position (specified in
1878  * the uio structure.
1879  *
1880  *	IN:	vp	- vnode of directory to read.
1881  *		uio	- structure supplying read location, range info,
1882  *			  and return buffer.
1883  *		cr	- credentials of caller.
1884  *		ct	- caller context
1885  *		flags	- case flags
1886  *
1887  *	OUT:	uio	- updated offset and range, buffer filled.
1888  *		eofp	- set to true if end-of-file detected.
1889  *
1890  *	RETURN:	0 if success
1891  *		error code if failure
1892  *
1893  * Timestamps:
1894  *	vp - atime updated
1895  *
1896  * Note that the low 4 bits of the cookie returned by zap is always zero.
1897  * This allows us to use the low range for "special" directory entries:
1898  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1899  * we use the offset 2 for the '.zfs' directory.
1900  */
1901 /* ARGSUSED */
1902 static int
1903 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
1904     caller_context_t *ct, int flags)
1905 {
1906 	znode_t		*zp = VTOZ(vp);
1907 	iovec_t		*iovp;
1908 	edirent_t	*eodp;
1909 	dirent64_t	*odp;
1910 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1911 	objset_t	*os;
1912 	caddr_t		outbuf;
1913 	size_t		bufsize;
1914 	zap_cursor_t	zc;
1915 	zap_attribute_t	zap;
1916 	uint_t		bytes_wanted;
1917 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1918 	int		local_eof;
1919 	int		outcount;
1920 	int		error;
1921 	uint8_t		prefetch;
1922 	boolean_t	check_sysattrs;
1923 
1924 	ZFS_ENTER(zfsvfs);
1925 	ZFS_VERIFY_ZP(zp);
1926 
1927 	/*
1928 	 * If we are not given an eof variable,
1929 	 * use a local one.
1930 	 */
1931 	if (eofp == NULL)
1932 		eofp = &local_eof;
1933 
1934 	/*
1935 	 * Check for valid iov_len.
1936 	 */
1937 	if (uio->uio_iov->iov_len <= 0) {
1938 		ZFS_EXIT(zfsvfs);
1939 		return (EINVAL);
1940 	}
1941 
1942 	/*
1943 	 * Quit if directory has been removed (posix)
1944 	 */
1945 	if ((*eofp = zp->z_unlinked) != 0) {
1946 		ZFS_EXIT(zfsvfs);
1947 		return (0);
1948 	}
1949 
1950 	error = 0;
1951 	os = zfsvfs->z_os;
1952 	offset = uio->uio_loffset;
1953 	prefetch = zp->z_zn_prefetch;
1954 
1955 	/*
1956 	 * Initialize the iterator cursor.
1957 	 */
1958 	if (offset <= 3) {
1959 		/*
1960 		 * Start iteration from the beginning of the directory.
1961 		 */
1962 		zap_cursor_init(&zc, os, zp->z_id);
1963 	} else {
1964 		/*
1965 		 * The offset is a serialized cursor.
1966 		 */
1967 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1968 	}
1969 
1970 	/*
1971 	 * Get space to change directory entries into fs independent format.
1972 	 */
1973 	iovp = uio->uio_iov;
1974 	bytes_wanted = iovp->iov_len;
1975 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
1976 		bufsize = bytes_wanted;
1977 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
1978 		odp = (struct dirent64 *)outbuf;
1979 	} else {
1980 		bufsize = bytes_wanted;
1981 		odp = (struct dirent64 *)iovp->iov_base;
1982 	}
1983 	eodp = (struct edirent *)odp;
1984 
1985 	/*
1986 	 * If this VFS supports the system attribute view interface; and
1987 	 * we're looking at an extended attribute directory; and we care
1988 	 * about normalization conflicts on this vfs; then we must check
1989 	 * for normalization conflicts with the sysattr name space.
1990 	 */
1991 	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
1992 	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
1993 	    (flags & V_RDDIR_ENTFLAGS);
1994 
1995 	/*
1996 	 * Transform to file-system independent format
1997 	 */
1998 	outcount = 0;
1999 	while (outcount < bytes_wanted) {
2000 		ino64_t objnum;
2001 		ushort_t reclen;
2002 		off64_t *next;
2003 
2004 		/*
2005 		 * Special case `.', `..', and `.zfs'.
2006 		 */
2007 		if (offset == 0) {
2008 			(void) strcpy(zap.za_name, ".");
2009 			zap.za_normalization_conflict = 0;
2010 			objnum = zp->z_id;
2011 		} else if (offset == 1) {
2012 			(void) strcpy(zap.za_name, "..");
2013 			zap.za_normalization_conflict = 0;
2014 			objnum = zp->z_phys->zp_parent;
2015 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2016 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2017 			zap.za_normalization_conflict = 0;
2018 			objnum = ZFSCTL_INO_ROOT;
2019 		} else {
2020 			/*
2021 			 * Grab next entry.
2022 			 */
2023 			if (error = zap_cursor_retrieve(&zc, &zap)) {
2024 				if ((*eofp = (error == ENOENT)) != 0)
2025 					break;
2026 				else
2027 					goto update;
2028 			}
2029 
2030 			if (zap.za_integer_length != 8 ||
2031 			    zap.za_num_integers != 1) {
2032 				cmn_err(CE_WARN, "zap_readdir: bad directory "
2033 				    "entry, obj = %lld, offset = %lld\n",
2034 				    (u_longlong_t)zp->z_id,
2035 				    (u_longlong_t)offset);
2036 				error = ENXIO;
2037 				goto update;
2038 			}
2039 
2040 			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2041 			/*
2042 			 * MacOS X can extract the object type here such as:
2043 			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2044 			 */
2045 
2046 			if (check_sysattrs && !zap.za_normalization_conflict) {
2047 				zap.za_normalization_conflict =
2048 				    xattr_sysattr_casechk(zap.za_name);
2049 			}
2050 		}
2051 
2052 		if (flags & V_RDDIR_ENTFLAGS)
2053 			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2054 		else
2055 			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2056 
2057 		/*
2058 		 * Will this entry fit in the buffer?
2059 		 */
2060 		if (outcount + reclen > bufsize) {
2061 			/*
2062 			 * Did we manage to fit anything in the buffer?
2063 			 */
2064 			if (!outcount) {
2065 				error = EINVAL;
2066 				goto update;
2067 			}
2068 			break;
2069 		}
2070 		if (flags & V_RDDIR_ENTFLAGS) {
2071 			/*
2072 			 * Add extended flag entry:
2073 			 */
2074 			eodp->ed_ino = objnum;
2075 			eodp->ed_reclen = reclen;
2076 			/* NOTE: ed_off is the offset for the *next* entry */
2077 			next = &(eodp->ed_off);
2078 			eodp->ed_eflags = zap.za_normalization_conflict ?
2079 			    ED_CASE_CONFLICT : 0;
2080 			(void) strncpy(eodp->ed_name, zap.za_name,
2081 			    EDIRENT_NAMELEN(reclen));
2082 			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2083 		} else {
2084 			/*
2085 			 * Add normal entry:
2086 			 */
2087 			odp->d_ino = objnum;
2088 			odp->d_reclen = reclen;
2089 			/* NOTE: d_off is the offset for the *next* entry */
2090 			next = &(odp->d_off);
2091 			(void) strncpy(odp->d_name, zap.za_name,
2092 			    DIRENT64_NAMELEN(reclen));
2093 			odp = (dirent64_t *)((intptr_t)odp + reclen);
2094 		}
2095 		outcount += reclen;
2096 
2097 		ASSERT(outcount <= bufsize);
2098 
2099 		/* Prefetch znode */
2100 		if (prefetch)
2101 			dmu_prefetch(os, objnum, 0, 0);
2102 
2103 		/*
2104 		 * Move to the next entry, fill in the previous offset.
2105 		 */
2106 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2107 			zap_cursor_advance(&zc);
2108 			offset = zap_cursor_serialize(&zc);
2109 		} else {
2110 			offset += 1;
2111 		}
2112 		*next = offset;
2113 	}
2114 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2115 
2116 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2117 		iovp->iov_base += outcount;
2118 		iovp->iov_len -= outcount;
2119 		uio->uio_resid -= outcount;
2120 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2121 		/*
2122 		 * Reset the pointer.
2123 		 */
2124 		offset = uio->uio_loffset;
2125 	}
2126 
2127 update:
2128 	zap_cursor_fini(&zc);
2129 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2130 		kmem_free(outbuf, bufsize);
2131 
2132 	if (error == ENOENT)
2133 		error = 0;
2134 
2135 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2136 
2137 	uio->uio_loffset = offset;
2138 	ZFS_EXIT(zfsvfs);
2139 	return (error);
2140 }
2141 
2142 ulong_t zfs_fsync_sync_cnt = 4;
2143 
2144 static int
2145 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2146 {
2147 	znode_t	*zp = VTOZ(vp);
2148 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2149 
2150 	/*
2151 	 * Regardless of whether this is required for standards conformance,
2152 	 * this is the logical behavior when fsync() is called on a file with
2153 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2154 	 * going to be pushed out as part of the zil_commit().
2155 	 */
2156 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2157 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2158 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2159 
2160 	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2161 
2162 	ZFS_ENTER(zfsvfs);
2163 	ZFS_VERIFY_ZP(zp);
2164 	zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
2165 	ZFS_EXIT(zfsvfs);
2166 	return (0);
2167 }
2168 
2169 
2170 /*
2171  * Get the requested file attributes and place them in the provided
2172  * vattr structure.
2173  *
2174  *	IN:	vp	- vnode of file.
2175  *		vap	- va_mask identifies requested attributes.
2176  *			  If AT_XVATTR set, then optional attrs are requested
2177  *		flags	- ATTR_NOACLCHECK (CIFS server context)
2178  *		cr	- credentials of caller.
2179  *		ct	- caller context
2180  *
2181  *	OUT:	vap	- attribute values.
2182  *
2183  *	RETURN:	0 (always succeeds)
2184  */
2185 /* ARGSUSED */
2186 static int
2187 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2188     caller_context_t *ct)
2189 {
2190 	znode_t *zp = VTOZ(vp);
2191 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2192 	znode_phys_t *pzp;
2193 	int	error = 0;
2194 	uint64_t links;
2195 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2196 	xoptattr_t *xoap = NULL;
2197 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2198 
2199 	ZFS_ENTER(zfsvfs);
2200 	ZFS_VERIFY_ZP(zp);
2201 	pzp = zp->z_phys;
2202 
2203 	mutex_enter(&zp->z_lock);
2204 
2205 	/*
2206 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2207 	 * Also, if we are the owner don't bother, since owner should
2208 	 * always be allowed to read basic attributes of file.
2209 	 */
2210 	if (!(pzp->zp_flags & ZFS_ACL_TRIVIAL) &&
2211 	    (pzp->zp_uid != crgetuid(cr))) {
2212 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2213 		    skipaclchk, cr)) {
2214 			mutex_exit(&zp->z_lock);
2215 			ZFS_EXIT(zfsvfs);
2216 			return (error);
2217 		}
2218 	}
2219 
2220 	/*
2221 	 * Return all attributes.  It's cheaper to provide the answer
2222 	 * than to determine whether we were asked the question.
2223 	 */
2224 
2225 	vap->va_type = vp->v_type;
2226 	vap->va_mode = pzp->zp_mode & MODEMASK;
2227 	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2228 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2229 	vap->va_nodeid = zp->z_id;
2230 	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2231 		links = pzp->zp_links + 1;
2232 	else
2233 		links = pzp->zp_links;
2234 	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2235 	vap->va_size = pzp->zp_size;
2236 	vap->va_rdev = vp->v_rdev;
2237 	vap->va_seq = zp->z_seq;
2238 
2239 	/*
2240 	 * Add in any requested optional attributes and the create time.
2241 	 * Also set the corresponding bits in the returned attribute bitmap.
2242 	 */
2243 	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2244 		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2245 			xoap->xoa_archive =
2246 			    ((pzp->zp_flags & ZFS_ARCHIVE) != 0);
2247 			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2248 		}
2249 
2250 		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2251 			xoap->xoa_readonly =
2252 			    ((pzp->zp_flags & ZFS_READONLY) != 0);
2253 			XVA_SET_RTN(xvap, XAT_READONLY);
2254 		}
2255 
2256 		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2257 			xoap->xoa_system =
2258 			    ((pzp->zp_flags & ZFS_SYSTEM) != 0);
2259 			XVA_SET_RTN(xvap, XAT_SYSTEM);
2260 		}
2261 
2262 		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2263 			xoap->xoa_hidden =
2264 			    ((pzp->zp_flags & ZFS_HIDDEN) != 0);
2265 			XVA_SET_RTN(xvap, XAT_HIDDEN);
2266 		}
2267 
2268 		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2269 			xoap->xoa_nounlink =
2270 			    ((pzp->zp_flags & ZFS_NOUNLINK) != 0);
2271 			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2272 		}
2273 
2274 		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2275 			xoap->xoa_immutable =
2276 			    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0);
2277 			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2278 		}
2279 
2280 		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2281 			xoap->xoa_appendonly =
2282 			    ((pzp->zp_flags & ZFS_APPENDONLY) != 0);
2283 			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2284 		}
2285 
2286 		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2287 			xoap->xoa_nodump =
2288 			    ((pzp->zp_flags & ZFS_NODUMP) != 0);
2289 			XVA_SET_RTN(xvap, XAT_NODUMP);
2290 		}
2291 
2292 		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2293 			xoap->xoa_opaque =
2294 			    ((pzp->zp_flags & ZFS_OPAQUE) != 0);
2295 			XVA_SET_RTN(xvap, XAT_OPAQUE);
2296 		}
2297 
2298 		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2299 			xoap->xoa_av_quarantined =
2300 			    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0);
2301 			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2302 		}
2303 
2304 		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2305 			xoap->xoa_av_modified =
2306 			    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0);
2307 			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2308 		}
2309 
2310 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2311 		    vp->v_type == VREG &&
2312 		    (pzp->zp_flags & ZFS_BONUS_SCANSTAMP)) {
2313 			size_t len;
2314 			dmu_object_info_t doi;
2315 
2316 			/*
2317 			 * Only VREG files have anti-virus scanstamps, so we
2318 			 * won't conflict with symlinks in the bonus buffer.
2319 			 */
2320 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2321 			len = sizeof (xoap->xoa_av_scanstamp) +
2322 			    sizeof (znode_phys_t);
2323 			if (len <= doi.doi_bonus_size) {
2324 				/*
2325 				 * pzp points to the start of the
2326 				 * znode_phys_t. pzp + 1 points to the
2327 				 * first byte after the znode_phys_t.
2328 				 */
2329 				(void) memcpy(xoap->xoa_av_scanstamp,
2330 				    pzp + 1,
2331 				    sizeof (xoap->xoa_av_scanstamp));
2332 				XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP);
2333 			}
2334 		}
2335 
2336 		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2337 			ZFS_TIME_DECODE(&xoap->xoa_createtime, pzp->zp_crtime);
2338 			XVA_SET_RTN(xvap, XAT_CREATETIME);
2339 		}
2340 	}
2341 
2342 	ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime);
2343 	ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime);
2344 	ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime);
2345 
2346 	mutex_exit(&zp->z_lock);
2347 
2348 	dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks);
2349 
2350 	if (zp->z_blksz == 0) {
2351 		/*
2352 		 * Block size hasn't been set; suggest maximal I/O transfers.
2353 		 */
2354 		vap->va_blksize = zfsvfs->z_max_blksz;
2355 	}
2356 
2357 	ZFS_EXIT(zfsvfs);
2358 	return (0);
2359 }
2360 
2361 /*
2362  * Set the file attributes to the values contained in the
2363  * vattr structure.
2364  *
2365  *	IN:	vp	- vnode of file to be modified.
2366  *		vap	- new attribute values.
2367  *			  If AT_XVATTR set, then optional attrs are being set
2368  *		flags	- ATTR_UTIME set if non-default time values provided.
2369  *			- ATTR_NOACLCHECK (CIFS context only).
2370  *		cr	- credentials of caller.
2371  *		ct	- caller context
2372  *
2373  *	RETURN:	0 if success
2374  *		error code if failure
2375  *
2376  * Timestamps:
2377  *	vp - ctime updated, mtime updated if size changed.
2378  */
2379 /* ARGSUSED */
2380 static int
2381 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2382 	caller_context_t *ct)
2383 {
2384 	znode_t		*zp = VTOZ(vp);
2385 	znode_phys_t	*pzp;
2386 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2387 	zilog_t		*zilog;
2388 	dmu_tx_t	*tx;
2389 	vattr_t		oldva;
2390 	uint_t		mask = vap->va_mask;
2391 	uint_t		saved_mask;
2392 	int		trim_mask = 0;
2393 	uint64_t	new_mode;
2394 	znode_t		*attrzp;
2395 	int		need_policy = FALSE;
2396 	int		err;
2397 	zfs_fuid_info_t *fuidp = NULL;
2398 	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2399 	xoptattr_t	*xoap;
2400 	zfs_acl_t	*aclp = NULL;
2401 	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2402 
2403 	if (mask == 0)
2404 		return (0);
2405 
2406 	if (mask & AT_NOSET)
2407 		return (EINVAL);
2408 
2409 	ZFS_ENTER(zfsvfs);
2410 	ZFS_VERIFY_ZP(zp);
2411 
2412 	pzp = zp->z_phys;
2413 	zilog = zfsvfs->z_log;
2414 
2415 	/*
2416 	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2417 	 * that file system is at proper version level
2418 	 */
2419 
2420 	if (zfsvfs->z_use_fuids == B_FALSE &&
2421 	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2422 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2423 	    (mask & AT_XVATTR))) {
2424 		ZFS_EXIT(zfsvfs);
2425 		return (EINVAL);
2426 	}
2427 
2428 	if (mask & AT_SIZE && vp->v_type == VDIR) {
2429 		ZFS_EXIT(zfsvfs);
2430 		return (EISDIR);
2431 	}
2432 
2433 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2434 		ZFS_EXIT(zfsvfs);
2435 		return (EINVAL);
2436 	}
2437 
2438 	/*
2439 	 * If this is an xvattr_t, then get a pointer to the structure of
2440 	 * optional attributes.  If this is NULL, then we have a vattr_t.
2441 	 */
2442 	xoap = xva_getxoptattr(xvap);
2443 
2444 	/*
2445 	 * Immutable files can only alter immutable bit and atime
2446 	 */
2447 	if ((pzp->zp_flags & ZFS_IMMUTABLE) &&
2448 	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2449 	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2450 		ZFS_EXIT(zfsvfs);
2451 		return (EPERM);
2452 	}
2453 
2454 	if ((mask & AT_SIZE) && (pzp->zp_flags & ZFS_READONLY)) {
2455 		ZFS_EXIT(zfsvfs);
2456 		return (EPERM);
2457 	}
2458 
2459 	/*
2460 	 * Verify timestamps doesn't overflow 32 bits.
2461 	 * ZFS can handle large timestamps, but 32bit syscalls can't
2462 	 * handle times greater than 2039.  This check should be removed
2463 	 * once large timestamps are fully supported.
2464 	 */
2465 	if (mask & (AT_ATIME | AT_MTIME)) {
2466 		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2467 		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2468 			ZFS_EXIT(zfsvfs);
2469 			return (EOVERFLOW);
2470 		}
2471 	}
2472 
2473 top:
2474 	attrzp = NULL;
2475 
2476 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2477 		ZFS_EXIT(zfsvfs);
2478 		return (EROFS);
2479 	}
2480 
2481 	/*
2482 	 * First validate permissions
2483 	 */
2484 
2485 	if (mask & AT_SIZE) {
2486 		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2487 		if (err) {
2488 			ZFS_EXIT(zfsvfs);
2489 			return (err);
2490 		}
2491 		/*
2492 		 * XXX - Note, we are not providing any open
2493 		 * mode flags here (like FNDELAY), so we may
2494 		 * block if there are locks present... this
2495 		 * should be addressed in openat().
2496 		 */
2497 		/* XXX - would it be OK to generate a log record here? */
2498 		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2499 		if (err) {
2500 			ZFS_EXIT(zfsvfs);
2501 			return (err);
2502 		}
2503 	}
2504 
2505 	if (mask & (AT_ATIME|AT_MTIME) ||
2506 	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2507 	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2508 	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2509 	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2510 	    XVA_ISSET_REQ(xvap, XAT_SYSTEM))))
2511 		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2512 		    skipaclchk, cr);
2513 
2514 	if (mask & (AT_UID|AT_GID)) {
2515 		int	idmask = (mask & (AT_UID|AT_GID));
2516 		int	take_owner;
2517 		int	take_group;
2518 
2519 		/*
2520 		 * NOTE: even if a new mode is being set,
2521 		 * we may clear S_ISUID/S_ISGID bits.
2522 		 */
2523 
2524 		if (!(mask & AT_MODE))
2525 			vap->va_mode = pzp->zp_mode;
2526 
2527 		/*
2528 		 * Take ownership or chgrp to group we are a member of
2529 		 */
2530 
2531 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2532 		take_group = (mask & AT_GID) &&
2533 		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2534 
2535 		/*
2536 		 * If both AT_UID and AT_GID are set then take_owner and
2537 		 * take_group must both be set in order to allow taking
2538 		 * ownership.
2539 		 *
2540 		 * Otherwise, send the check through secpolicy_vnode_setattr()
2541 		 *
2542 		 */
2543 
2544 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2545 		    ((idmask == AT_UID) && take_owner) ||
2546 		    ((idmask == AT_GID) && take_group)) {
2547 			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2548 			    skipaclchk, cr) == 0) {
2549 				/*
2550 				 * Remove setuid/setgid for non-privileged users
2551 				 */
2552 				secpolicy_setid_clear(vap, cr);
2553 				trim_mask = (mask & (AT_UID|AT_GID));
2554 			} else {
2555 				need_policy =  TRUE;
2556 			}
2557 		} else {
2558 			need_policy =  TRUE;
2559 		}
2560 	}
2561 
2562 	mutex_enter(&zp->z_lock);
2563 	oldva.va_mode = pzp->zp_mode;
2564 	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2565 	if (mask & AT_XVATTR) {
2566 		if ((need_policy == FALSE) &&
2567 		    (XVA_ISSET_REQ(xvap, XAT_APPENDONLY) &&
2568 		    xoap->xoa_appendonly !=
2569 		    ((pzp->zp_flags & ZFS_APPENDONLY) != 0)) ||
2570 		    (XVA_ISSET_REQ(xvap, XAT_NOUNLINK) &&
2571 		    xoap->xoa_nounlink !=
2572 		    ((pzp->zp_flags & ZFS_NOUNLINK) != 0)) ||
2573 		    (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE) &&
2574 		    xoap->xoa_immutable !=
2575 		    ((pzp->zp_flags & ZFS_IMMUTABLE) != 0)) ||
2576 		    (XVA_ISSET_REQ(xvap, XAT_NODUMP) &&
2577 		    xoap->xoa_nodump !=
2578 		    ((pzp->zp_flags & ZFS_NODUMP) != 0)) ||
2579 		    (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED) &&
2580 		    xoap->xoa_av_modified !=
2581 		    ((pzp->zp_flags & ZFS_AV_MODIFIED) != 0)) ||
2582 		    ((XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED) &&
2583 		    ((vp->v_type != VREG && xoap->xoa_av_quarantined) ||
2584 		    xoap->xoa_av_quarantined !=
2585 		    ((pzp->zp_flags & ZFS_AV_QUARANTINED) != 0)))) ||
2586 		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) ||
2587 		    (XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2588 			need_policy = TRUE;
2589 		}
2590 	}
2591 
2592 	mutex_exit(&zp->z_lock);
2593 
2594 	if (mask & AT_MODE) {
2595 		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2596 			err = secpolicy_setid_setsticky_clear(vp, vap,
2597 			    &oldva, cr);
2598 			if (err) {
2599 				ZFS_EXIT(zfsvfs);
2600 				return (err);
2601 			}
2602 			trim_mask |= AT_MODE;
2603 		} else {
2604 			need_policy = TRUE;
2605 		}
2606 	}
2607 
2608 	if (need_policy) {
2609 		/*
2610 		 * If trim_mask is set then take ownership
2611 		 * has been granted or write_acl is present and user
2612 		 * has the ability to modify mode.  In that case remove
2613 		 * UID|GID and or MODE from mask so that
2614 		 * secpolicy_vnode_setattr() doesn't revoke it.
2615 		 */
2616 
2617 		if (trim_mask) {
2618 			saved_mask = vap->va_mask;
2619 			vap->va_mask &= ~trim_mask;
2620 		}
2621 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2622 		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
2623 		if (err) {
2624 			ZFS_EXIT(zfsvfs);
2625 			return (err);
2626 		}
2627 
2628 		if (trim_mask)
2629 			vap->va_mask |= saved_mask;
2630 	}
2631 
2632 	/*
2633 	 * secpolicy_vnode_setattr, or take ownership may have
2634 	 * changed va_mask
2635 	 */
2636 	mask = vap->va_mask;
2637 
2638 	tx = dmu_tx_create(zfsvfs->z_os);
2639 	dmu_tx_hold_bonus(tx, zp->z_id);
2640 	if (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2641 	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid))) {
2642 		if (zfsvfs->z_fuid_obj == 0) {
2643 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
2644 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2645 			    FUID_SIZE_ESTIMATE(zfsvfs));
2646 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
2647 		} else {
2648 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
2649 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
2650 			    FUID_SIZE_ESTIMATE(zfsvfs));
2651 		}
2652 	}
2653 
2654 	if (mask & AT_MODE) {
2655 		uint64_t pmode = pzp->zp_mode;
2656 
2657 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2658 
2659 		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) {
2660 			dmu_tx_abort(tx);
2661 			ZFS_EXIT(zfsvfs);
2662 			return (err);
2663 		}
2664 		if (pzp->zp_acl.z_acl_extern_obj) {
2665 			/* Are we upgrading ACL from old V0 format to new V1 */
2666 			if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2667 			    pzp->zp_acl.z_acl_version ==
2668 			    ZFS_ACL_VERSION_INITIAL) {
2669 				dmu_tx_hold_free(tx,
2670 				    pzp->zp_acl.z_acl_extern_obj, 0,
2671 				    DMU_OBJECT_END);
2672 				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2673 				    0, aclp->z_acl_bytes);
2674 			} else {
2675 				dmu_tx_hold_write(tx,
2676 				    pzp->zp_acl.z_acl_extern_obj, 0,
2677 				    aclp->z_acl_bytes);
2678 			}
2679 		} else if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2680 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2681 			    0, aclp->z_acl_bytes);
2682 		}
2683 	}
2684 
2685 	if ((mask & (AT_UID | AT_GID)) && pzp->zp_xattr != 0) {
2686 		err = zfs_zget(zp->z_zfsvfs, pzp->zp_xattr, &attrzp);
2687 		if (err) {
2688 			dmu_tx_abort(tx);
2689 			ZFS_EXIT(zfsvfs);
2690 			if (aclp)
2691 				zfs_acl_free(aclp);
2692 			return (err);
2693 		}
2694 		dmu_tx_hold_bonus(tx, attrzp->z_id);
2695 	}
2696 
2697 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2698 	if (err) {
2699 		if (attrzp)
2700 			VN_RELE(ZTOV(attrzp));
2701 
2702 		if (aclp) {
2703 			zfs_acl_free(aclp);
2704 			aclp = NULL;
2705 		}
2706 
2707 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2708 			dmu_tx_wait(tx);
2709 			dmu_tx_abort(tx);
2710 			goto top;
2711 		}
2712 		dmu_tx_abort(tx);
2713 		ZFS_EXIT(zfsvfs);
2714 		return (err);
2715 	}
2716 
2717 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2718 
2719 	/*
2720 	 * Set each attribute requested.
2721 	 * We group settings according to the locks they need to acquire.
2722 	 *
2723 	 * Note: you cannot set ctime directly, although it will be
2724 	 * updated as a side-effect of calling this function.
2725 	 */
2726 
2727 	mutex_enter(&zp->z_lock);
2728 
2729 	if (mask & AT_MODE) {
2730 		mutex_enter(&zp->z_acl_lock);
2731 		zp->z_phys->zp_mode = new_mode;
2732 		err = zfs_aclset_common(zp, aclp, cr, &fuidp, tx);
2733 		ASSERT3U(err, ==, 0);
2734 		mutex_exit(&zp->z_acl_lock);
2735 	}
2736 
2737 	if (attrzp)
2738 		mutex_enter(&attrzp->z_lock);
2739 
2740 	if (mask & AT_UID) {
2741 		pzp->zp_uid = zfs_fuid_create(zfsvfs,
2742 		    vap->va_uid, cr, ZFS_OWNER, tx, &fuidp);
2743 		if (attrzp) {
2744 			attrzp->z_phys->zp_uid = zfs_fuid_create(zfsvfs,
2745 			    vap->va_uid,  cr, ZFS_OWNER, tx, &fuidp);
2746 		}
2747 	}
2748 
2749 	if (mask & AT_GID) {
2750 		pzp->zp_gid = zfs_fuid_create(zfsvfs, vap->va_gid,
2751 		    cr, ZFS_GROUP, tx, &fuidp);
2752 		if (attrzp)
2753 			attrzp->z_phys->zp_gid = zfs_fuid_create(zfsvfs,
2754 			    vap->va_gid, cr, ZFS_GROUP, tx, &fuidp);
2755 	}
2756 
2757 	if (aclp)
2758 		zfs_acl_free(aclp);
2759 
2760 	if (attrzp)
2761 		mutex_exit(&attrzp->z_lock);
2762 
2763 	if (mask & AT_ATIME)
2764 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2765 
2766 	if (mask & AT_MTIME)
2767 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2768 
2769 	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2770 	if (mask & AT_SIZE)
2771 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2772 	else if (mask != 0)
2773 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2774 	/*
2775 	 * Do this after setting timestamps to prevent timestamp
2776 	 * update from toggling bit
2777 	 */
2778 
2779 	if (xoap && (mask & AT_XVATTR)) {
2780 		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
2781 			size_t len;
2782 			dmu_object_info_t doi;
2783 
2784 			ASSERT(vp->v_type == VREG);
2785 
2786 			/* Grow the bonus buffer if necessary. */
2787 			dmu_object_info_from_db(zp->z_dbuf, &doi);
2788 			len = sizeof (xoap->xoa_av_scanstamp) +
2789 			    sizeof (znode_phys_t);
2790 			if (len > doi.doi_bonus_size)
2791 				VERIFY(dmu_set_bonus(zp->z_dbuf, len, tx) == 0);
2792 		}
2793 		zfs_xvattr_set(zp, xvap);
2794 	}
2795 
2796 	if (mask != 0)
2797 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
2798 
2799 	if (fuidp)
2800 		zfs_fuid_info_free(fuidp);
2801 	mutex_exit(&zp->z_lock);
2802 
2803 	if (attrzp)
2804 		VN_RELE(ZTOV(attrzp));
2805 
2806 	dmu_tx_commit(tx);
2807 
2808 	ZFS_EXIT(zfsvfs);
2809 	return (err);
2810 }
2811 
2812 typedef struct zfs_zlock {
2813 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2814 	znode_t		*zl_znode;	/* znode we held */
2815 	struct zfs_zlock *zl_next;	/* next in list */
2816 } zfs_zlock_t;
2817 
2818 /*
2819  * Drop locks and release vnodes that were held by zfs_rename_lock().
2820  */
2821 static void
2822 zfs_rename_unlock(zfs_zlock_t **zlpp)
2823 {
2824 	zfs_zlock_t *zl;
2825 
2826 	while ((zl = *zlpp) != NULL) {
2827 		if (zl->zl_znode != NULL)
2828 			VN_RELE(ZTOV(zl->zl_znode));
2829 		rw_exit(zl->zl_rwlock);
2830 		*zlpp = zl->zl_next;
2831 		kmem_free(zl, sizeof (*zl));
2832 	}
2833 }
2834 
2835 /*
2836  * Search back through the directory tree, using the ".." entries.
2837  * Lock each directory in the chain to prevent concurrent renames.
2838  * Fail any attempt to move a directory into one of its own descendants.
2839  * XXX - z_parent_lock can overlap with map or grow locks
2840  */
2841 static int
2842 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2843 {
2844 	zfs_zlock_t	*zl;
2845 	znode_t		*zp = tdzp;
2846 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2847 	uint64_t	*oidp = &zp->z_id;
2848 	krwlock_t	*rwlp = &szp->z_parent_lock;
2849 	krw_t		rw = RW_WRITER;
2850 
2851 	/*
2852 	 * First pass write-locks szp and compares to zp->z_id.
2853 	 * Later passes read-lock zp and compare to zp->z_parent.
2854 	 */
2855 	do {
2856 		if (!rw_tryenter(rwlp, rw)) {
2857 			/*
2858 			 * Another thread is renaming in this path.
2859 			 * Note that if we are a WRITER, we don't have any
2860 			 * parent_locks held yet.
2861 			 */
2862 			if (rw == RW_READER && zp->z_id > szp->z_id) {
2863 				/*
2864 				 * Drop our locks and restart
2865 				 */
2866 				zfs_rename_unlock(&zl);
2867 				*zlpp = NULL;
2868 				zp = tdzp;
2869 				oidp = &zp->z_id;
2870 				rwlp = &szp->z_parent_lock;
2871 				rw = RW_WRITER;
2872 				continue;
2873 			} else {
2874 				/*
2875 				 * Wait for other thread to drop its locks
2876 				 */
2877 				rw_enter(rwlp, rw);
2878 			}
2879 		}
2880 
2881 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2882 		zl->zl_rwlock = rwlp;
2883 		zl->zl_znode = NULL;
2884 		zl->zl_next = *zlpp;
2885 		*zlpp = zl;
2886 
2887 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2888 			return (EINVAL);
2889 
2890 		if (*oidp == rootid)		/* We've hit the top */
2891 			return (0);
2892 
2893 		if (rw == RW_READER) {		/* i.e. not the first pass */
2894 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2895 			if (error)
2896 				return (error);
2897 			zl->zl_znode = zp;
2898 		}
2899 		oidp = &zp->z_phys->zp_parent;
2900 		rwlp = &zp->z_parent_lock;
2901 		rw = RW_READER;
2902 
2903 	} while (zp->z_id != sdzp->z_id);
2904 
2905 	return (0);
2906 }
2907 
2908 /*
2909  * Move an entry from the provided source directory to the target
2910  * directory.  Change the entry name as indicated.
2911  *
2912  *	IN:	sdvp	- Source directory containing the "old entry".
2913  *		snm	- Old entry name.
2914  *		tdvp	- Target directory to contain the "new entry".
2915  *		tnm	- New entry name.
2916  *		cr	- credentials of caller.
2917  *		ct	- caller context
2918  *		flags	- case flags
2919  *
2920  *	RETURN:	0 if success
2921  *		error code if failure
2922  *
2923  * Timestamps:
2924  *	sdvp,tdvp - ctime|mtime updated
2925  */
2926 /*ARGSUSED*/
2927 static int
2928 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
2929     caller_context_t *ct, int flags)
2930 {
2931 	znode_t		*tdzp, *szp, *tzp;
2932 	znode_t		*sdzp = VTOZ(sdvp);
2933 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2934 	zilog_t		*zilog;
2935 	vnode_t		*realvp;
2936 	zfs_dirlock_t	*sdl, *tdl;
2937 	dmu_tx_t	*tx;
2938 	zfs_zlock_t	*zl;
2939 	int		cmp, serr, terr;
2940 	int		error = 0;
2941 	int		zflg = 0;
2942 
2943 	ZFS_ENTER(zfsvfs);
2944 	ZFS_VERIFY_ZP(sdzp);
2945 	zilog = zfsvfs->z_log;
2946 
2947 	/*
2948 	 * Make sure we have the real vp for the target directory.
2949 	 */
2950 	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
2951 		tdvp = realvp;
2952 
2953 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
2954 		ZFS_EXIT(zfsvfs);
2955 		return (EXDEV);
2956 	}
2957 
2958 	tdzp = VTOZ(tdvp);
2959 	ZFS_VERIFY_ZP(tdzp);
2960 	if (zfsvfs->z_utf8 && u8_validate(tnm,
2961 	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
2962 		ZFS_EXIT(zfsvfs);
2963 		return (EILSEQ);
2964 	}
2965 
2966 	if (flags & FIGNORECASE)
2967 		zflg |= ZCILOOK;
2968 
2969 top:
2970 	szp = NULL;
2971 	tzp = NULL;
2972 	zl = NULL;
2973 
2974 	/*
2975 	 * This is to prevent the creation of links into attribute space
2976 	 * by renaming a linked file into/outof an attribute directory.
2977 	 * See the comment in zfs_link() for why this is considered bad.
2978 	 */
2979 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
2980 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
2981 		ZFS_EXIT(zfsvfs);
2982 		return (EINVAL);
2983 	}
2984 
2985 	/*
2986 	 * Lock source and target directory entries.  To prevent deadlock,
2987 	 * a lock ordering must be defined.  We lock the directory with
2988 	 * the smallest object id first, or if it's a tie, the one with
2989 	 * the lexically first name.
2990 	 */
2991 	if (sdzp->z_id < tdzp->z_id) {
2992 		cmp = -1;
2993 	} else if (sdzp->z_id > tdzp->z_id) {
2994 		cmp = 1;
2995 	} else {
2996 		/*
2997 		 * First compare the two name arguments without
2998 		 * considering any case folding.
2999 		 */
3000 		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3001 
3002 		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3003 		ASSERT(error == 0 || !zfsvfs->z_utf8);
3004 		if (cmp == 0) {
3005 			/*
3006 			 * POSIX: "If the old argument and the new argument
3007 			 * both refer to links to the same existing file,
3008 			 * the rename() function shall return successfully
3009 			 * and perform no other action."
3010 			 */
3011 			ZFS_EXIT(zfsvfs);
3012 			return (0);
3013 		}
3014 		/*
3015 		 * If the file system is case-folding, then we may
3016 		 * have some more checking to do.  A case-folding file
3017 		 * system is either supporting mixed case sensitivity
3018 		 * access or is completely case-insensitive.  Note
3019 		 * that the file system is always case preserving.
3020 		 *
3021 		 * In mixed sensitivity mode case sensitive behavior
3022 		 * is the default.  FIGNORECASE must be used to
3023 		 * explicitly request case insensitive behavior.
3024 		 *
3025 		 * If the source and target names provided differ only
3026 		 * by case (e.g., a request to rename 'tim' to 'Tim'),
3027 		 * we will treat this as a special case in the
3028 		 * case-insensitive mode: as long as the source name
3029 		 * is an exact match, we will allow this to proceed as
3030 		 * a name-change request.
3031 		 */
3032 		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3033 		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
3034 		    flags & FIGNORECASE)) &&
3035 		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3036 		    &error) == 0) {
3037 			/*
3038 			 * case preserving rename request, require exact
3039 			 * name matches
3040 			 */
3041 			zflg |= ZCIEXACT;
3042 			zflg &= ~ZCILOOK;
3043 		}
3044 	}
3045 
3046 	if (cmp < 0) {
3047 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3048 		    ZEXISTS | zflg, NULL, NULL);
3049 		terr = zfs_dirent_lock(&tdl,
3050 		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3051 	} else {
3052 		terr = zfs_dirent_lock(&tdl,
3053 		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3054 		serr = zfs_dirent_lock(&sdl,
3055 		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3056 		    NULL, NULL);
3057 	}
3058 
3059 	if (serr) {
3060 		/*
3061 		 * Source entry invalid or not there.
3062 		 */
3063 		if (!terr) {
3064 			zfs_dirent_unlock(tdl);
3065 			if (tzp)
3066 				VN_RELE(ZTOV(tzp));
3067 		}
3068 		if (strcmp(snm, "..") == 0)
3069 			serr = EINVAL;
3070 		ZFS_EXIT(zfsvfs);
3071 		return (serr);
3072 	}
3073 	if (terr) {
3074 		zfs_dirent_unlock(sdl);
3075 		VN_RELE(ZTOV(szp));
3076 		if (strcmp(tnm, "..") == 0)
3077 			terr = EINVAL;
3078 		ZFS_EXIT(zfsvfs);
3079 		return (terr);
3080 	}
3081 
3082 	/*
3083 	 * Must have write access at the source to remove the old entry
3084 	 * and write access at the target to create the new entry.
3085 	 * Note that if target and source are the same, this can be
3086 	 * done in a single check.
3087 	 */
3088 
3089 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3090 		goto out;
3091 
3092 	if (ZTOV(szp)->v_type == VDIR) {
3093 		/*
3094 		 * Check to make sure rename is valid.
3095 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3096 		 */
3097 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3098 			goto out;
3099 	}
3100 
3101 	/*
3102 	 * Does target exist?
3103 	 */
3104 	if (tzp) {
3105 		/*
3106 		 * Source and target must be the same type.
3107 		 */
3108 		if (ZTOV(szp)->v_type == VDIR) {
3109 			if (ZTOV(tzp)->v_type != VDIR) {
3110 				error = ENOTDIR;
3111 				goto out;
3112 			}
3113 		} else {
3114 			if (ZTOV(tzp)->v_type == VDIR) {
3115 				error = EISDIR;
3116 				goto out;
3117 			}
3118 		}
3119 		/*
3120 		 * POSIX dictates that when the source and target
3121 		 * entries refer to the same file object, rename
3122 		 * must do nothing and exit without error.
3123 		 */
3124 		if (szp->z_id == tzp->z_id) {
3125 			error = 0;
3126 			goto out;
3127 		}
3128 	}
3129 
3130 	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3131 	if (tzp)
3132 		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3133 
3134 	/*
3135 	 * notify the target directory if it is not the same
3136 	 * as source directory.
3137 	 */
3138 	if (tdvp != sdvp) {
3139 		vnevent_rename_dest_dir(tdvp, ct);
3140 	}
3141 
3142 	tx = dmu_tx_create(zfsvfs->z_os);
3143 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
3144 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
3145 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3146 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3147 	if (sdzp != tdzp)
3148 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
3149 	if (tzp)
3150 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* parent changes */
3151 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3152 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3153 	if (error) {
3154 		if (zl != NULL)
3155 			zfs_rename_unlock(&zl);
3156 		zfs_dirent_unlock(sdl);
3157 		zfs_dirent_unlock(tdl);
3158 		VN_RELE(ZTOV(szp));
3159 		if (tzp)
3160 			VN_RELE(ZTOV(tzp));
3161 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3162 			dmu_tx_wait(tx);
3163 			dmu_tx_abort(tx);
3164 			goto top;
3165 		}
3166 		dmu_tx_abort(tx);
3167 		ZFS_EXIT(zfsvfs);
3168 		return (error);
3169 	}
3170 
3171 	if (tzp)	/* Attempt to remove the existing target */
3172 		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3173 
3174 	if (error == 0) {
3175 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3176 		if (error == 0) {
3177 			szp->z_phys->zp_flags |= ZFS_AV_MODIFIED;
3178 
3179 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3180 			ASSERT(error == 0);
3181 
3182 			zfs_log_rename(zilog, tx,
3183 			    TX_RENAME | (flags & FIGNORECASE ? TX_CI : 0),
3184 			    sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp);
3185 
3186 			/* Update path information for the target vnode */
3187 			vn_renamepath(tdvp, ZTOV(szp), tnm, strlen(tnm));
3188 		}
3189 	}
3190 
3191 	dmu_tx_commit(tx);
3192 out:
3193 	if (zl != NULL)
3194 		zfs_rename_unlock(&zl);
3195 
3196 	zfs_dirent_unlock(sdl);
3197 	zfs_dirent_unlock(tdl);
3198 
3199 	VN_RELE(ZTOV(szp));
3200 	if (tzp)
3201 		VN_RELE(ZTOV(tzp));
3202 
3203 	ZFS_EXIT(zfsvfs);
3204 	return (error);
3205 }
3206 
3207 /*
3208  * Insert the indicated symbolic reference entry into the directory.
3209  *
3210  *	IN:	dvp	- Directory to contain new symbolic link.
3211  *		link	- Name for new symlink entry.
3212  *		vap	- Attributes of new entry.
3213  *		target	- Target path of new symlink.
3214  *		cr	- credentials of caller.
3215  *		ct	- caller context
3216  *		flags	- case flags
3217  *
3218  *	RETURN:	0 if success
3219  *		error code if failure
3220  *
3221  * Timestamps:
3222  *	dvp - ctime|mtime updated
3223  */
3224 /*ARGSUSED*/
3225 static int
3226 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3227     caller_context_t *ct, int flags)
3228 {
3229 	znode_t		*zp, *dzp = VTOZ(dvp);
3230 	zfs_dirlock_t	*dl;
3231 	dmu_tx_t	*tx;
3232 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3233 	zilog_t		*zilog;
3234 	int		len = strlen(link);
3235 	int		error;
3236 	int		zflg = ZNEW;
3237 	zfs_fuid_info_t *fuidp = NULL;
3238 
3239 	ASSERT(vap->va_type == VLNK);
3240 
3241 	ZFS_ENTER(zfsvfs);
3242 	ZFS_VERIFY_ZP(dzp);
3243 	zilog = zfsvfs->z_log;
3244 
3245 	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3246 	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3247 		ZFS_EXIT(zfsvfs);
3248 		return (EILSEQ);
3249 	}
3250 	if (flags & FIGNORECASE)
3251 		zflg |= ZCILOOK;
3252 top:
3253 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3254 		ZFS_EXIT(zfsvfs);
3255 		return (error);
3256 	}
3257 
3258 	if (len > MAXPATHLEN) {
3259 		ZFS_EXIT(zfsvfs);
3260 		return (ENAMETOOLONG);
3261 	}
3262 
3263 	/*
3264 	 * Attempt to lock directory; fail if entry already exists.
3265 	 */
3266 	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3267 	if (error) {
3268 		ZFS_EXIT(zfsvfs);
3269 		return (error);
3270 	}
3271 
3272 	tx = dmu_tx_create(zfsvfs->z_os);
3273 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3274 	dmu_tx_hold_bonus(tx, dzp->z_id);
3275 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3276 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
3277 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
3278 	if (IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))) {
3279 		if (zfsvfs->z_fuid_obj == 0) {
3280 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
3281 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3282 			    FUID_SIZE_ESTIMATE(zfsvfs));
3283 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
3284 		} else {
3285 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
3286 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
3287 			    FUID_SIZE_ESTIMATE(zfsvfs));
3288 		}
3289 	}
3290 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3291 	if (error) {
3292 		zfs_dirent_unlock(dl);
3293 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3294 			dmu_tx_wait(tx);
3295 			dmu_tx_abort(tx);
3296 			goto top;
3297 		}
3298 		dmu_tx_abort(tx);
3299 		ZFS_EXIT(zfsvfs);
3300 		return (error);
3301 	}
3302 
3303 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
3304 
3305 	/*
3306 	 * Create a new object for the symlink.
3307 	 * Put the link content into bonus buffer if it will fit;
3308 	 * otherwise, store it just like any other file data.
3309 	 */
3310 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
3311 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, len, NULL, &fuidp);
3312 		if (len != 0)
3313 			bcopy(link, zp->z_phys + 1, len);
3314 	} else {
3315 		dmu_buf_t *dbp;
3316 
3317 		zfs_mknode(dzp, vap, tx, cr, 0, &zp, 0, NULL, &fuidp);
3318 		/*
3319 		 * Nothing can access the znode yet so no locking needed
3320 		 * for growing the znode's blocksize.
3321 		 */
3322 		zfs_grow_blocksize(zp, len, tx);
3323 
3324 		VERIFY(0 == dmu_buf_hold(zfsvfs->z_os,
3325 		    zp->z_id, 0, FTAG, &dbp));
3326 		dmu_buf_will_dirty(dbp, tx);
3327 
3328 		ASSERT3U(len, <=, dbp->db_size);
3329 		bcopy(link, dbp->db_data, len);
3330 		dmu_buf_rele(dbp, FTAG);
3331 	}
3332 	zp->z_phys->zp_size = len;
3333 
3334 	/*
3335 	 * Insert the new object into the directory.
3336 	 */
3337 	(void) zfs_link_create(dl, zp, tx, ZNEW);
3338 out:
3339 	if (error == 0) {
3340 		uint64_t txtype = TX_SYMLINK;
3341 		if (flags & FIGNORECASE)
3342 			txtype |= TX_CI;
3343 		zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3344 	}
3345 	if (fuidp)
3346 		zfs_fuid_info_free(fuidp);
3347 
3348 	dmu_tx_commit(tx);
3349 
3350 	zfs_dirent_unlock(dl);
3351 
3352 	VN_RELE(ZTOV(zp));
3353 
3354 	ZFS_EXIT(zfsvfs);
3355 	return (error);
3356 }
3357 
3358 /*
3359  * Return, in the buffer contained in the provided uio structure,
3360  * the symbolic path referred to by vp.
3361  *
3362  *	IN:	vp	- vnode of symbolic link.
3363  *		uoip	- structure to contain the link path.
3364  *		cr	- credentials of caller.
3365  *		ct	- caller context
3366  *
3367  *	OUT:	uio	- structure to contain the link path.
3368  *
3369  *	RETURN:	0 if success
3370  *		error code if failure
3371  *
3372  * Timestamps:
3373  *	vp - atime updated
3374  */
3375 /* ARGSUSED */
3376 static int
3377 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3378 {
3379 	znode_t		*zp = VTOZ(vp);
3380 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3381 	size_t		bufsz;
3382 	int		error;
3383 
3384 	ZFS_ENTER(zfsvfs);
3385 	ZFS_VERIFY_ZP(zp);
3386 
3387 	bufsz = (size_t)zp->z_phys->zp_size;
3388 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
3389 		error = uiomove(zp->z_phys + 1,
3390 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3391 	} else {
3392 		dmu_buf_t *dbp;
3393 		error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp);
3394 		if (error) {
3395 			ZFS_EXIT(zfsvfs);
3396 			return (error);
3397 		}
3398 		error = uiomove(dbp->db_data,
3399 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
3400 		dmu_buf_rele(dbp, FTAG);
3401 	}
3402 
3403 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3404 	ZFS_EXIT(zfsvfs);
3405 	return (error);
3406 }
3407 
3408 /*
3409  * Insert a new entry into directory tdvp referencing svp.
3410  *
3411  *	IN:	tdvp	- Directory to contain new entry.
3412  *		svp	- vnode of new entry.
3413  *		name	- name of new entry.
3414  *		cr	- credentials of caller.
3415  *		ct	- caller context
3416  *
3417  *	RETURN:	0 if success
3418  *		error code if failure
3419  *
3420  * Timestamps:
3421  *	tdvp - ctime|mtime updated
3422  *	 svp - ctime updated
3423  */
3424 /* ARGSUSED */
3425 static int
3426 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3427     caller_context_t *ct, int flags)
3428 {
3429 	znode_t		*dzp = VTOZ(tdvp);
3430 	znode_t		*tzp, *szp;
3431 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3432 	zilog_t		*zilog;
3433 	zfs_dirlock_t	*dl;
3434 	dmu_tx_t	*tx;
3435 	vnode_t		*realvp;
3436 	int		error;
3437 	int		zf = ZNEW;
3438 	uid_t		owner;
3439 
3440 	ASSERT(tdvp->v_type == VDIR);
3441 
3442 	ZFS_ENTER(zfsvfs);
3443 	ZFS_VERIFY_ZP(dzp);
3444 	zilog = zfsvfs->z_log;
3445 
3446 	if (VOP_REALVP(svp, &realvp, ct) == 0)
3447 		svp = realvp;
3448 
3449 	if (svp->v_vfsp != tdvp->v_vfsp) {
3450 		ZFS_EXIT(zfsvfs);
3451 		return (EXDEV);
3452 	}
3453 	szp = VTOZ(svp);
3454 	ZFS_VERIFY_ZP(szp);
3455 
3456 	if (zfsvfs->z_utf8 && u8_validate(name,
3457 	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3458 		ZFS_EXIT(zfsvfs);
3459 		return (EILSEQ);
3460 	}
3461 	if (flags & FIGNORECASE)
3462 		zf |= ZCILOOK;
3463 
3464 top:
3465 	/*
3466 	 * We do not support links between attributes and non-attributes
3467 	 * because of the potential security risk of creating links
3468 	 * into "normal" file space in order to circumvent restrictions
3469 	 * imposed in attribute space.
3470 	 */
3471 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
3472 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
3473 		ZFS_EXIT(zfsvfs);
3474 		return (EINVAL);
3475 	}
3476 
3477 	/*
3478 	 * POSIX dictates that we return EPERM here.
3479 	 * Better choices include ENOTSUP or EISDIR.
3480 	 */
3481 	if (svp->v_type == VDIR) {
3482 		ZFS_EXIT(zfsvfs);
3483 		return (EPERM);
3484 	}
3485 
3486 	owner = zfs_fuid_map_id(zfsvfs, szp->z_phys->zp_uid, cr, ZFS_OWNER);
3487 	if (owner != crgetuid(cr) &&
3488 	    secpolicy_basic_link(cr) != 0) {
3489 		ZFS_EXIT(zfsvfs);
3490 		return (EPERM);
3491 	}
3492 
3493 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3494 		ZFS_EXIT(zfsvfs);
3495 		return (error);
3496 	}
3497 
3498 	/*
3499 	 * Attempt to lock directory; fail if entry already exists.
3500 	 */
3501 	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
3502 	if (error) {
3503 		ZFS_EXIT(zfsvfs);
3504 		return (error);
3505 	}
3506 
3507 	tx = dmu_tx_create(zfsvfs->z_os);
3508 	dmu_tx_hold_bonus(tx, szp->z_id);
3509 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3510 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
3511 	if (error) {
3512 		zfs_dirent_unlock(dl);
3513 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3514 			dmu_tx_wait(tx);
3515 			dmu_tx_abort(tx);
3516 			goto top;
3517 		}
3518 		dmu_tx_abort(tx);
3519 		ZFS_EXIT(zfsvfs);
3520 		return (error);
3521 	}
3522 
3523 	error = zfs_link_create(dl, szp, tx, 0);
3524 
3525 	if (error == 0) {
3526 		uint64_t txtype = TX_LINK;
3527 		if (flags & FIGNORECASE)
3528 			txtype |= TX_CI;
3529 		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
3530 	}
3531 
3532 	dmu_tx_commit(tx);
3533 
3534 	zfs_dirent_unlock(dl);
3535 
3536 	if (error == 0) {
3537 		vnevent_link(svp, ct);
3538 	}
3539 
3540 	ZFS_EXIT(zfsvfs);
3541 	return (error);
3542 }
3543 
3544 /*
3545  * zfs_null_putapage() is used when the file system has been force
3546  * unmounted. It just drops the pages.
3547  */
3548 /* ARGSUSED */
3549 static int
3550 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3551 		size_t *lenp, int flags, cred_t *cr)
3552 {
3553 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
3554 	return (0);
3555 }
3556 
3557 /*
3558  * Push a page out to disk, klustering if possible.
3559  *
3560  *	IN:	vp	- file to push page to.
3561  *		pp	- page to push.
3562  *		flags	- additional flags.
3563  *		cr	- credentials of caller.
3564  *
3565  *	OUT:	offp	- start of range pushed.
3566  *		lenp	- len of range pushed.
3567  *
3568  *	RETURN:	0 if success
3569  *		error code if failure
3570  *
3571  * NOTE: callers must have locked the page to be pushed.  On
3572  * exit, the page (and all other pages in the kluster) must be
3573  * unlocked.
3574  */
3575 /* ARGSUSED */
3576 static int
3577 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
3578 		size_t *lenp, int flags, cred_t *cr)
3579 {
3580 	znode_t		*zp = VTOZ(vp);
3581 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3582 	zilog_t		*zilog = zfsvfs->z_log;
3583 	dmu_tx_t	*tx;
3584 	rl_t		*rl;
3585 	u_offset_t	off, koff;
3586 	size_t		len, klen;
3587 	uint64_t	filesz;
3588 	int		err;
3589 
3590 	filesz = zp->z_phys->zp_size;
3591 	off = pp->p_offset;
3592 	len = PAGESIZE;
3593 	/*
3594 	 * If our blocksize is bigger than the page size, try to kluster
3595 	 * muiltiple pages so that we write a full block (thus avoiding
3596 	 * a read-modify-write).
3597 	 */
3598 	if (off < filesz && zp->z_blksz > PAGESIZE) {
3599 		if (!ISP2(zp->z_blksz)) {
3600 			/* Only one block in the file. */
3601 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3602 			koff = 0;
3603 		} else {
3604 			klen = zp->z_blksz;
3605 			koff = P2ALIGN(off, (u_offset_t)klen);
3606 		}
3607 		ASSERT(koff <= filesz);
3608 		if (koff + klen > filesz)
3609 			klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE);
3610 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
3611 	}
3612 	ASSERT3U(btop(len), ==, btopr(len));
3613 top:
3614 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
3615 	/*
3616 	 * Can't push pages past end-of-file.
3617 	 */
3618 	filesz = zp->z_phys->zp_size;
3619 	if (off >= filesz) {
3620 		/* ignore all pages */
3621 		err = 0;
3622 		goto out;
3623 	} else if (off + len > filesz) {
3624 		int npages = btopr(filesz - off);
3625 		page_t *trunc;
3626 
3627 		page_list_break(&pp, &trunc, npages);
3628 		/* ignore pages past end of file */
3629 		if (trunc)
3630 			pvn_write_done(trunc, flags);
3631 		len = filesz - off;
3632 	}
3633 
3634 	tx = dmu_tx_create(zfsvfs->z_os);
3635 	dmu_tx_hold_write(tx, zp->z_id, off, len);
3636 	dmu_tx_hold_bonus(tx, zp->z_id);
3637 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
3638 	if (err != 0) {
3639 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
3640 			zfs_range_unlock(rl);
3641 			dmu_tx_wait(tx);
3642 			dmu_tx_abort(tx);
3643 			err = 0;
3644 			goto top;
3645 		}
3646 		dmu_tx_abort(tx);
3647 		goto out;
3648 	}
3649 
3650 	if (zp->z_blksz <= PAGESIZE) {
3651 		caddr_t va = zfs_map_page(pp, S_READ);
3652 		ASSERT3U(len, <=, PAGESIZE);
3653 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
3654 		zfs_unmap_page(pp, va);
3655 	} else {
3656 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
3657 	}
3658 
3659 	if (err == 0) {
3660 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
3661 		zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0);
3662 		dmu_tx_commit(tx);
3663 	}
3664 
3665 out:
3666 	zfs_range_unlock(rl);
3667 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
3668 	if (offp)
3669 		*offp = off;
3670 	if (lenp)
3671 		*lenp = len;
3672 
3673 	return (err);
3674 }
3675 
3676 /*
3677  * Copy the portion of the file indicated from pages into the file.
3678  * The pages are stored in a page list attached to the files vnode.
3679  *
3680  *	IN:	vp	- vnode of file to push page data to.
3681  *		off	- position in file to put data.
3682  *		len	- amount of data to write.
3683  *		flags	- flags to control the operation.
3684  *		cr	- credentials of caller.
3685  *		ct	- caller context.
3686  *
3687  *	RETURN:	0 if success
3688  *		error code if failure
3689  *
3690  * Timestamps:
3691  *	vp - ctime|mtime updated
3692  */
3693 /*ARGSUSED*/
3694 static int
3695 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
3696     caller_context_t *ct)
3697 {
3698 	znode_t		*zp = VTOZ(vp);
3699 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3700 	page_t		*pp;
3701 	size_t		io_len;
3702 	u_offset_t	io_off;
3703 	uint64_t	filesz;
3704 	int		error = 0;
3705 
3706 	ZFS_ENTER(zfsvfs);
3707 	ZFS_VERIFY_ZP(zp);
3708 
3709 	if (len == 0) {
3710 		/*
3711 		 * Search the entire vp list for pages >= off.
3712 		 */
3713 		error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage,
3714 		    flags, cr);
3715 		goto out;
3716 	}
3717 
3718 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3719 	if (off > filesz) {
3720 		/* past end of file */
3721 		ZFS_EXIT(zfsvfs);
3722 		return (0);
3723 	}
3724 
3725 	len = MIN(len, filesz - off);
3726 
3727 	for (io_off = off; io_off < off + len; io_off += io_len) {
3728 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
3729 			pp = page_lookup(vp, io_off,
3730 			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
3731 		} else {
3732 			pp = page_lookup_nowait(vp, io_off,
3733 			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
3734 		}
3735 
3736 		if (pp != NULL && pvn_getdirty(pp, flags)) {
3737 			int err;
3738 
3739 			/*
3740 			 * Found a dirty page to push
3741 			 */
3742 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
3743 			if (err)
3744 				error = err;
3745 		} else {
3746 			io_len = PAGESIZE;
3747 		}
3748 	}
3749 out:
3750 	if ((flags & B_ASYNC) == 0)
3751 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
3752 	ZFS_EXIT(zfsvfs);
3753 	return (error);
3754 }
3755 
3756 /*ARGSUSED*/
3757 void
3758 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
3759 {
3760 	znode_t	*zp = VTOZ(vp);
3761 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3762 	int error;
3763 
3764 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
3765 	if (zp->z_dbuf == NULL) {
3766 		/*
3767 		 * The fs has been unmounted, or we did a
3768 		 * suspend/resume and this file no longer exists.
3769 		 */
3770 		if (vn_has_cached_data(vp)) {
3771 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
3772 			    B_INVAL, cr);
3773 		}
3774 
3775 		mutex_enter(&zp->z_lock);
3776 		vp->v_count = 0; /* count arrives as 1 */
3777 		mutex_exit(&zp->z_lock);
3778 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
3779 		zfs_znode_free(zp);
3780 		return;
3781 	}
3782 
3783 	/*
3784 	 * Attempt to push any data in the page cache.  If this fails
3785 	 * we will get kicked out later in zfs_zinactive().
3786 	 */
3787 	if (vn_has_cached_data(vp)) {
3788 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
3789 		    cr);
3790 	}
3791 
3792 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3793 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3794 
3795 		dmu_tx_hold_bonus(tx, zp->z_id);
3796 		error = dmu_tx_assign(tx, TXG_WAIT);
3797 		if (error) {
3798 			dmu_tx_abort(tx);
3799 		} else {
3800 			dmu_buf_will_dirty(zp->z_dbuf, tx);
3801 			mutex_enter(&zp->z_lock);
3802 			zp->z_atime_dirty = 0;
3803 			mutex_exit(&zp->z_lock);
3804 			dmu_tx_commit(tx);
3805 		}
3806 	}
3807 
3808 	zfs_zinactive(zp);
3809 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
3810 }
3811 
3812 /*
3813  * Bounds-check the seek operation.
3814  *
3815  *	IN:	vp	- vnode seeking within
3816  *		ooff	- old file offset
3817  *		noffp	- pointer to new file offset
3818  *		ct	- caller context
3819  *
3820  *	RETURN:	0 if success
3821  *		EINVAL if new offset invalid
3822  */
3823 /* ARGSUSED */
3824 static int
3825 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
3826     caller_context_t *ct)
3827 {
3828 	if (vp->v_type == VDIR)
3829 		return (0);
3830 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3831 }
3832 
3833 /*
3834  * Pre-filter the generic locking function to trap attempts to place
3835  * a mandatory lock on a memory mapped file.
3836  */
3837 static int
3838 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3839     flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
3840 {
3841 	znode_t *zp = VTOZ(vp);
3842 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3843 	int error;
3844 
3845 	ZFS_ENTER(zfsvfs);
3846 	ZFS_VERIFY_ZP(zp);
3847 
3848 	/*
3849 	 * We are following the UFS semantics with respect to mapcnt
3850 	 * here: If we see that the file is mapped already, then we will
3851 	 * return an error, but we don't worry about races between this
3852 	 * function and zfs_map().
3853 	 */
3854 	if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3855 		ZFS_EXIT(zfsvfs);
3856 		return (EAGAIN);
3857 	}
3858 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct);
3859 	ZFS_EXIT(zfsvfs);
3860 	return (error);
3861 }
3862 
3863 /*
3864  * If we can't find a page in the cache, we will create a new page
3865  * and fill it with file data.  For efficiency, we may try to fill
3866  * multiple pages at once (klustering).
3867  */
3868 static int
3869 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3870     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3871 {
3872 	znode_t *zp = VTOZ(vp);
3873 	page_t *pp, *cur_pp;
3874 	objset_t *os = zp->z_zfsvfs->z_os;
3875 	caddr_t va;
3876 	u_offset_t io_off, total;
3877 	uint64_t oid = zp->z_id;
3878 	size_t io_len;
3879 	uint64_t filesz;
3880 	int err;
3881 
3882 	/*
3883 	 * If we are only asking for a single page don't bother klustering.
3884 	 */
3885 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3886 	if (off >= filesz)
3887 		return (EFAULT);
3888 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3889 		io_off = off;
3890 		io_len = PAGESIZE;
3891 		pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr);
3892 	} else {
3893 		/*
3894 		 * Try to fill a kluster of pages (a blocks worth).
3895 		 */
3896 		size_t klen;
3897 		u_offset_t koff;
3898 
3899 		if (!ISP2(zp->z_blksz)) {
3900 			/* Only one block in the file. */
3901 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3902 			koff = 0;
3903 		} else {
3904 			/*
3905 			 * It would be ideal to align our offset to the
3906 			 * blocksize but doing so has resulted in some
3907 			 * strange application crashes. For now, we
3908 			 * leave the offset as is and only adjust the
3909 			 * length if we are off the end of the file.
3910 			 */
3911 			koff = off;
3912 			klen = plsz;
3913 		}
3914 		ASSERT(koff <= filesz);
3915 		if (koff + klen > filesz)
3916 			klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff;
3917 		ASSERT3U(off, >=, koff);
3918 		ASSERT3U(off, <, koff + klen);
3919 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3920 		    &io_len, koff, klen, 0);
3921 	}
3922 	if (pp == NULL) {
3923 		/*
3924 		 * Some other thread entered the page before us.
3925 		 * Return to zfs_getpage to retry the lookup.
3926 		 */
3927 		*pl = NULL;
3928 		return (0);
3929 	}
3930 
3931 	/*
3932 	 * Fill the pages in the kluster.
3933 	 */
3934 	cur_pp = pp;
3935 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3936 		ASSERT3U(io_off, ==, cur_pp->p_offset);
3937 		va = zfs_map_page(cur_pp, S_WRITE);
3938 		err = dmu_read(os, oid, io_off, PAGESIZE, va);
3939 		zfs_unmap_page(cur_pp, va);
3940 		if (err) {
3941 			/* On error, toss the entire kluster */
3942 			pvn_read_done(pp, B_ERROR);
3943 			/* convert checksum errors into IO errors */
3944 			if (err == ECKSUM)
3945 				err = EIO;
3946 			return (err);
3947 		}
3948 		cur_pp = cur_pp->p_next;
3949 	}
3950 out:
3951 	/*
3952 	 * Fill in the page list array from the kluster.  If
3953 	 * there are too many pages in the kluster, return
3954 	 * as many pages as possible starting from the desired
3955 	 * offset `off'.
3956 	 * NOTE: the page list will always be null terminated.
3957 	 */
3958 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3959 
3960 	return (0);
3961 }
3962 
3963 /*
3964  * Return pointers to the pages for the file region [off, off + len]
3965  * in the pl array.  If plsz is greater than len, this function may
3966  * also return page pointers from before or after the specified
3967  * region (i.e. some region [off', off' + plsz]).  These additional
3968  * pages are only returned if they are already in the cache, or were
3969  * created as part of a klustered read.
3970  *
3971  *	IN:	vp	- vnode of file to get data from.
3972  *		off	- position in file to get data from.
3973  *		len	- amount of data to retrieve.
3974  *		plsz	- length of provided page list.
3975  *		seg	- segment to obtain pages for.
3976  *		addr	- virtual address of fault.
3977  *		rw	- mode of created pages.
3978  *		cr	- credentials of caller.
3979  *		ct	- caller context.
3980  *
3981  *	OUT:	protp	- protection mode of created pages.
3982  *		pl	- list of pages created.
3983  *
3984  *	RETURN:	0 if success
3985  *		error code if failure
3986  *
3987  * Timestamps:
3988  *	vp - atime updated
3989  */
3990 /* ARGSUSED */
3991 static int
3992 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3993 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3994 	enum seg_rw rw, cred_t *cr, caller_context_t *ct)
3995 {
3996 	znode_t		*zp = VTOZ(vp);
3997 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3998 	page_t		*pp, **pl0 = pl;
3999 	int		need_unlock = 0, err = 0;
4000 	offset_t	orig_off;
4001 
4002 	ZFS_ENTER(zfsvfs);
4003 	ZFS_VERIFY_ZP(zp);
4004 
4005 	if (protp)
4006 		*protp = PROT_ALL;
4007 
4008 	/* no faultahead (for now) */
4009 	if (pl == NULL) {
4010 		ZFS_EXIT(zfsvfs);
4011 		return (0);
4012 	}
4013 
4014 	/* can't fault past EOF */
4015 	if (off >= zp->z_phys->zp_size) {
4016 		ZFS_EXIT(zfsvfs);
4017 		return (EFAULT);
4018 	}
4019 	orig_off = off;
4020 
4021 	/*
4022 	 * If we already own the lock, then we must be page faulting
4023 	 * in the middle of a write to this file (i.e., we are writing
4024 	 * to this file using data from a mapped region of the file).
4025 	 */
4026 	if (rw_owner(&zp->z_map_lock) != curthread) {
4027 		rw_enter(&zp->z_map_lock, RW_WRITER);
4028 		need_unlock = TRUE;
4029 	}
4030 
4031 	/*
4032 	 * Loop through the requested range [off, off + len] looking
4033 	 * for pages.  If we don't find a page, we will need to create
4034 	 * a new page and fill it with data from the file.
4035 	 */
4036 	while (len > 0) {
4037 		if (plsz < PAGESIZE)
4038 			break;
4039 		if (pp = page_lookup(vp, off, SE_SHARED)) {
4040 			*pl++ = pp;
4041 			off += PAGESIZE;
4042 			addr += PAGESIZE;
4043 			len -= PAGESIZE;
4044 			plsz -= PAGESIZE;
4045 		} else {
4046 			err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw);
4047 			if (err)
4048 				goto out;
4049 			/*
4050 			 * klustering may have changed our region
4051 			 * to be block aligned.
4052 			 */
4053 			if (((pp = *pl) != 0) && (off != pp->p_offset)) {
4054 				int delta = off - pp->p_offset;
4055 				len += delta;
4056 				off -= delta;
4057 				addr -= delta;
4058 			}
4059 			while (*pl) {
4060 				pl++;
4061 				off += PAGESIZE;
4062 				addr += PAGESIZE;
4063 				plsz -= PAGESIZE;
4064 				if (len > PAGESIZE)
4065 					len -= PAGESIZE;
4066 				else
4067 					len = 0;
4068 			}
4069 		}
4070 	}
4071 
4072 	/*
4073 	 * Fill out the page array with any pages already in the cache.
4074 	 */
4075 	while (plsz > 0) {
4076 		pp = page_lookup_nowait(vp, off, SE_SHARED);
4077 		if (pp == NULL)
4078 			break;
4079 		*pl++ = pp;
4080 		off += PAGESIZE;
4081 		plsz -= PAGESIZE;
4082 	}
4083 
4084 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4085 out:
4086 	/*
4087 	 * We can't grab the range lock for the page as reader which would
4088 	 * stop truncation as this leads to deadlock. So we need to recheck
4089 	 * the file size.
4090 	 */
4091 	if (orig_off >= zp->z_phys->zp_size)
4092 		err = EFAULT;
4093 	if (err) {
4094 		/*
4095 		 * Release any pages we have previously locked.
4096 		 */
4097 		while (pl > pl0)
4098 			page_unlock(*--pl);
4099 	}
4100 
4101 	*pl = NULL;
4102 
4103 	if (need_unlock)
4104 		rw_exit(&zp->z_map_lock);
4105 
4106 	ZFS_EXIT(zfsvfs);
4107 	return (err);
4108 }
4109 
4110 /*
4111  * Request a memory map for a section of a file.  This code interacts
4112  * with common code and the VM system as follows:
4113  *
4114  *	common code calls mmap(), which ends up in smmap_common()
4115  *
4116  *	this calls VOP_MAP(), which takes you into (say) zfs
4117  *
4118  *	zfs_map() calls as_map(), passing segvn_create() as the callback
4119  *
4120  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
4121  *
4122  *	zfs_addmap() updates z_mapcnt
4123  */
4124 /*ARGSUSED*/
4125 static int
4126 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4127     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4128     caller_context_t *ct)
4129 {
4130 	znode_t *zp = VTOZ(vp);
4131 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4132 	segvn_crargs_t	vn_a;
4133 	int		error;
4134 
4135 	ZFS_ENTER(zfsvfs);
4136 	ZFS_VERIFY_ZP(zp);
4137 
4138 	if ((prot & PROT_WRITE) &&
4139 	    (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_READONLY |
4140 	    ZFS_APPENDONLY))) {
4141 		ZFS_EXIT(zfsvfs);
4142 		return (EPERM);
4143 	}
4144 
4145 	if ((prot & (PROT_READ | PROT_EXEC)) &&
4146 	    (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED)) {
4147 		ZFS_EXIT(zfsvfs);
4148 		return (EACCES);
4149 	}
4150 
4151 	if (vp->v_flag & VNOMAP) {
4152 		ZFS_EXIT(zfsvfs);
4153 		return (ENOSYS);
4154 	}
4155 
4156 	if (off < 0 || len > MAXOFFSET_T - off) {
4157 		ZFS_EXIT(zfsvfs);
4158 		return (ENXIO);
4159 	}
4160 
4161 	if (vp->v_type != VREG) {
4162 		ZFS_EXIT(zfsvfs);
4163 		return (ENODEV);
4164 	}
4165 
4166 	/*
4167 	 * If file is locked, disallow mapping.
4168 	 */
4169 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) {
4170 		ZFS_EXIT(zfsvfs);
4171 		return (EAGAIN);
4172 	}
4173 
4174 	as_rangelock(as);
4175 	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4176 	if (error != 0) {
4177 		as_rangeunlock(as);
4178 		ZFS_EXIT(zfsvfs);
4179 		return (error);
4180 	}
4181 
4182 	vn_a.vp = vp;
4183 	vn_a.offset = (u_offset_t)off;
4184 	vn_a.type = flags & MAP_TYPE;
4185 	vn_a.prot = prot;
4186 	vn_a.maxprot = maxprot;
4187 	vn_a.cred = cr;
4188 	vn_a.amp = NULL;
4189 	vn_a.flags = flags & ~MAP_TYPE;
4190 	vn_a.szc = 0;
4191 	vn_a.lgrp_mem_policy_flags = 0;
4192 
4193 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4194 
4195 	as_rangeunlock(as);
4196 	ZFS_EXIT(zfsvfs);
4197 	return (error);
4198 }
4199 
4200 /* ARGSUSED */
4201 static int
4202 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4203     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4204     caller_context_t *ct)
4205 {
4206 	uint64_t pages = btopr(len);
4207 
4208 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4209 	return (0);
4210 }
4211 
4212 /*
4213  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4214  * more accurate mtime for the associated file.  Since we don't have a way of
4215  * detecting when the data was actually modified, we have to resort to
4216  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4217  * last page is pushed.  The problem occurs when the msync() call is omitted,
4218  * which by far the most common case:
4219  *
4220  * 	open()
4221  * 	mmap()
4222  * 	<modify memory>
4223  * 	munmap()
4224  * 	close()
4225  * 	<time lapse>
4226  * 	putpage() via fsflush
4227  *
4228  * If we wait until fsflush to come along, we can have a modification time that
4229  * is some arbitrary point in the future.  In order to prevent this in the
4230  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4231  * torn down.
4232  */
4233 /* ARGSUSED */
4234 static int
4235 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4236     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4237     caller_context_t *ct)
4238 {
4239 	uint64_t pages = btopr(len);
4240 
4241 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4242 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4243 
4244 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4245 	    vn_has_cached_data(vp))
4246 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4247 
4248 	return (0);
4249 }
4250 
4251 /*
4252  * Free or allocate space in a file.  Currently, this function only
4253  * supports the `F_FREESP' command.  However, this command is somewhat
4254  * misnamed, as its functionality includes the ability to allocate as
4255  * well as free space.
4256  *
4257  *	IN:	vp	- vnode of file to free data in.
4258  *		cmd	- action to take (only F_FREESP supported).
4259  *		bfp	- section of file to free/alloc.
4260  *		flag	- current file open mode flags.
4261  *		offset	- current file offset.
4262  *		cr	- credentials of caller [UNUSED].
4263  *		ct	- caller context.
4264  *
4265  *	RETURN:	0 if success
4266  *		error code if failure
4267  *
4268  * Timestamps:
4269  *	vp - ctime|mtime updated
4270  */
4271 /* ARGSUSED */
4272 static int
4273 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4274     offset_t offset, cred_t *cr, caller_context_t *ct)
4275 {
4276 	znode_t		*zp = VTOZ(vp);
4277 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4278 	uint64_t	off, len;
4279 	int		error;
4280 
4281 	ZFS_ENTER(zfsvfs);
4282 	ZFS_VERIFY_ZP(zp);
4283 
4284 	if (cmd != F_FREESP) {
4285 		ZFS_EXIT(zfsvfs);
4286 		return (EINVAL);
4287 	}
4288 
4289 	if (error = convoff(vp, bfp, 0, offset)) {
4290 		ZFS_EXIT(zfsvfs);
4291 		return (error);
4292 	}
4293 
4294 	if (bfp->l_len < 0) {
4295 		ZFS_EXIT(zfsvfs);
4296 		return (EINVAL);
4297 	}
4298 
4299 	off = bfp->l_start;
4300 	len = bfp->l_len; /* 0 means from off to end of file */
4301 
4302 	error = zfs_freesp(zp, off, len, flag, TRUE);
4303 
4304 	ZFS_EXIT(zfsvfs);
4305 	return (error);
4306 }
4307 
4308 /*ARGSUSED*/
4309 static int
4310 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4311 {
4312 	znode_t		*zp = VTOZ(vp);
4313 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4314 	uint32_t	gen;
4315 	uint64_t	object = zp->z_id;
4316 	zfid_short_t	*zfid;
4317 	int		size, i;
4318 
4319 	ZFS_ENTER(zfsvfs);
4320 	ZFS_VERIFY_ZP(zp);
4321 	gen = (uint32_t)zp->z_gen;
4322 
4323 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4324 	if (fidp->fid_len < size) {
4325 		fidp->fid_len = size;
4326 		ZFS_EXIT(zfsvfs);
4327 		return (ENOSPC);
4328 	}
4329 
4330 	zfid = (zfid_short_t *)fidp;
4331 
4332 	zfid->zf_len = size;
4333 
4334 	for (i = 0; i < sizeof (zfid->zf_object); i++)
4335 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4336 
4337 	/* Must have a non-zero generation number to distinguish from .zfs */
4338 	if (gen == 0)
4339 		gen = 1;
4340 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4341 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4342 
4343 	if (size == LONG_FID_LEN) {
4344 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4345 		zfid_long_t	*zlfid;
4346 
4347 		zlfid = (zfid_long_t *)fidp;
4348 
4349 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4350 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4351 
4352 		/* XXX - this should be the generation number for the objset */
4353 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4354 			zlfid->zf_setgen[i] = 0;
4355 	}
4356 
4357 	ZFS_EXIT(zfsvfs);
4358 	return (0);
4359 }
4360 
4361 static int
4362 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4363     caller_context_t *ct)
4364 {
4365 	znode_t		*zp, *xzp;
4366 	zfsvfs_t	*zfsvfs;
4367 	zfs_dirlock_t	*dl;
4368 	int		error;
4369 
4370 	switch (cmd) {
4371 	case _PC_LINK_MAX:
4372 		*valp = ULONG_MAX;
4373 		return (0);
4374 
4375 	case _PC_FILESIZEBITS:
4376 		*valp = 64;
4377 		return (0);
4378 
4379 	case _PC_XATTR_EXISTS:
4380 		zp = VTOZ(vp);
4381 		zfsvfs = zp->z_zfsvfs;
4382 		ZFS_ENTER(zfsvfs);
4383 		ZFS_VERIFY_ZP(zp);
4384 		*valp = 0;
4385 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4386 		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4387 		if (error == 0) {
4388 			zfs_dirent_unlock(dl);
4389 			if (!zfs_dirempty(xzp))
4390 				*valp = 1;
4391 			VN_RELE(ZTOV(xzp));
4392 		} else if (error == ENOENT) {
4393 			/*
4394 			 * If there aren't extended attributes, it's the
4395 			 * same as having zero of them.
4396 			 */
4397 			error = 0;
4398 		}
4399 		ZFS_EXIT(zfsvfs);
4400 		return (error);
4401 
4402 	case _PC_SATTR_ENABLED:
4403 	case _PC_SATTR_EXISTS:
4404 		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4405 		    (vp->v_type == VREG || vp->v_type == VDIR);
4406 		return (0);
4407 
4408 	case _PC_ACL_ENABLED:
4409 		*valp = _ACL_ACE_ENABLED;
4410 		return (0);
4411 
4412 	case _PC_MIN_HOLE_SIZE:
4413 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4414 		return (0);
4415 
4416 	default:
4417 		return (fs_pathconf(vp, cmd, valp, cr, ct));
4418 	}
4419 }
4420 
4421 /*ARGSUSED*/
4422 static int
4423 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4424     caller_context_t *ct)
4425 {
4426 	znode_t *zp = VTOZ(vp);
4427 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4428 	int error;
4429 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4430 
4431 	ZFS_ENTER(zfsvfs);
4432 	ZFS_VERIFY_ZP(zp);
4433 	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
4434 	ZFS_EXIT(zfsvfs);
4435 
4436 	return (error);
4437 }
4438 
4439 /*ARGSUSED*/
4440 static int
4441 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4442     caller_context_t *ct)
4443 {
4444 	znode_t *zp = VTOZ(vp);
4445 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4446 	int error;
4447 	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4448 
4449 	ZFS_ENTER(zfsvfs);
4450 	ZFS_VERIFY_ZP(zp);
4451 	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
4452 	ZFS_EXIT(zfsvfs);
4453 	return (error);
4454 }
4455 
4456 /*
4457  * Predeclare these here so that the compiler assumes that
4458  * this is an "old style" function declaration that does
4459  * not include arguments => we won't get type mismatch errors
4460  * in the initializations that follow.
4461  */
4462 static int zfs_inval();
4463 static int zfs_isdir();
4464 
4465 static int
4466 zfs_inval()
4467 {
4468 	return (EINVAL);
4469 }
4470 
4471 static int
4472 zfs_isdir()
4473 {
4474 	return (EISDIR);
4475 }
4476 /*
4477  * Directory vnode operations template
4478  */
4479 vnodeops_t *zfs_dvnodeops;
4480 const fs_operation_def_t zfs_dvnodeops_template[] = {
4481 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4482 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4483 	VOPNAME_READ,		{ .error = zfs_isdir },
4484 	VOPNAME_WRITE,		{ .error = zfs_isdir },
4485 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4486 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4487 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4488 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4489 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4490 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4491 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4492 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4493 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4494 	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
4495 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4496 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4497 	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
4498 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4499 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4500 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4501 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4502 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4503 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4504 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4505 	VOPNAME_VNEVENT, 	{ .vop_vnevent = fs_vnevent_support },
4506 	NULL,			NULL
4507 };
4508 
4509 /*
4510  * Regular file vnode operations template
4511  */
4512 vnodeops_t *zfs_fvnodeops;
4513 const fs_operation_def_t zfs_fvnodeops_template[] = {
4514 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4515 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4516 	VOPNAME_READ,		{ .vop_read = zfs_read },
4517 	VOPNAME_WRITE,		{ .vop_write = zfs_write },
4518 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4519 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4520 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4521 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4522 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4523 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4524 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4525 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4526 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4527 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4528 	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
4529 	VOPNAME_SPACE,		{ .vop_space = zfs_space },
4530 	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
4531 	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
4532 	VOPNAME_MAP,		{ .vop_map = zfs_map },
4533 	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
4534 	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
4535 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4536 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4537 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4538 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4539 	NULL,			NULL
4540 };
4541 
4542 /*
4543  * Symbolic link vnode operations template
4544  */
4545 vnodeops_t *zfs_symvnodeops;
4546 const fs_operation_def_t zfs_symvnodeops_template[] = {
4547 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4548 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4549 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4550 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4551 	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
4552 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4553 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4554 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4555 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4556 	NULL,			NULL
4557 };
4558 
4559 /*
4560  * Extended attribute directory vnode operations template
4561  *	This template is identical to the directory vnodes
4562  *	operation template except for restricted operations:
4563  *		VOP_MKDIR()
4564  *		VOP_SYMLINK()
4565  * Note that there are other restrictions embedded in:
4566  *	zfs_create()	- restrict type to VREG
4567  *	zfs_link()	- no links into/out of attribute space
4568  *	zfs_rename()	- no moves into/out of attribute space
4569  */
4570 vnodeops_t *zfs_xdvnodeops;
4571 const fs_operation_def_t zfs_xdvnodeops_template[] = {
4572 	VOPNAME_OPEN,		{ .vop_open = zfs_open },
4573 	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
4574 	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
4575 	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
4576 	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
4577 	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
4578 	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
4579 	VOPNAME_CREATE,		{ .vop_create = zfs_create },
4580 	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
4581 	VOPNAME_LINK,		{ .vop_link = zfs_link },
4582 	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
4583 	VOPNAME_MKDIR,		{ .error = zfs_inval },
4584 	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
4585 	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
4586 	VOPNAME_SYMLINK,	{ .error = zfs_inval },
4587 	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
4588 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4589 	VOPNAME_FID,		{ .vop_fid = zfs_fid },
4590 	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
4591 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4592 	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
4593 	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
4594 	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
4595 	NULL,			NULL
4596 };
4597 
4598 /*
4599  * Error vnode operations template
4600  */
4601 vnodeops_t *zfs_evnodeops;
4602 const fs_operation_def_t zfs_evnodeops_template[] = {
4603 	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
4604 	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
4605 	NULL,			NULL
4606 };
4607