xref: /titanic_41/usr/src/uts/common/fs/zfs/zfs_vnops.c (revision 9113a79cf228b8f7bd509b1328adf88659dfe218)
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 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
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/vnode.h>
36 #include <sys/file.h>
37 #include <sys/stat.h>
38 #include <sys/kmem.h>
39 #include <sys/taskq.h>
40 #include <sys/uio.h>
41 #include <sys/vmsystm.h>
42 #include <sys/atomic.h>
43 #include <sys/vm.h>
44 #include <vm/seg_vn.h>
45 #include <vm/pvn.h>
46 #include <vm/as.h>
47 #include <sys/mman.h>
48 #include <sys/pathname.h>
49 #include <sys/cmn_err.h>
50 #include <sys/errno.h>
51 #include <sys/unistd.h>
52 #include <sys/zfs_vfsops.h>
53 #include <sys/zfs_dir.h>
54 #include <sys/zfs_acl.h>
55 #include <sys/zfs_ioctl.h>
56 #include <sys/fs/zfs.h>
57 #include <sys/dmu.h>
58 #include <sys/spa.h>
59 #include <sys/txg.h>
60 #include <sys/dbuf.h>
61 #include <sys/zap.h>
62 #include <sys/dirent.h>
63 #include <sys/policy.h>
64 #include <sys/sunddi.h>
65 #include <sys/filio.h>
66 #include "fs/fs_subr.h"
67 #include <sys/zfs_ctldir.h>
68 #include <sys/dnlc.h>
69 #include <sys/zfs_rlock.h>
70 
71 /*
72  * Programming rules.
73  *
74  * Each vnode op performs some logical unit of work.  To do this, the ZPL must
75  * properly lock its in-core state, create a DMU transaction, do the work,
76  * record this work in the intent log (ZIL), commit the DMU transaction,
77  * and wait the the intent log to commit if it's is a synchronous operation.
78  * Morover, the vnode ops must work in both normal and log replay context.
79  * The ordering of events is important to avoid deadlocks and references
80  * to freed memory.  The example below illustrates the following Big Rules:
81  *
82  *  (1) A check must be made in each zfs thread for a mounted file system.
83  *	This is done avoiding races using ZFS_ENTER(zfsvfs).
84  *	A ZFS_EXIT(zfsvfs) is needed before all returns.
85  *
86  *  (2)	VN_RELE() should always be the last thing except for zil_commit()
87  *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
88  *	First, if it's the last reference, the vnode/znode
89  *	can be freed, so the zp may point to freed memory.  Second, the last
90  *	reference will call zfs_zinactive(), which may induce a lot of work --
91  *	pushing cached pages (which acquires range locks) and syncing out
92  *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
93  *	which could deadlock the system if you were already holding one.
94  *
95  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
96  *	as they can span dmu_tx_assign() calls.
97  *
98  *  (4)	Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign().
99  *	In normal operation, this will be TXG_NOWAIT.  During ZIL replay,
100  *	it will be a specific txg.  Either way, dmu_tx_assign() never blocks.
101  *	This is critical because we don't want to block while holding locks.
102  *	Note, in particular, that if a lock is sometimes acquired before
103  *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
104  *	use a non-blocking assign can deadlock the system.  The scenario:
105  *
106  *	Thread A has grabbed a lock before calling dmu_tx_assign().
107  *	Thread B is in an already-assigned tx, and blocks for this lock.
108  *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
109  *	forever, because the previous txg can't quiesce until B's tx commits.
110  *
111  *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
112  *	then drop all locks, call dmu_tx_wait(), and try again.
113  *
114  *  (5)	If the operation succeeded, generate the intent log entry for it
115  *	before dropping locks.  This ensures that the ordering of events
116  *	in the intent log matches the order in which they actually occurred.
117  *
118  *  (6)	At the end of each vnode op, the DMU tx must always commit,
119  *	regardless of whether there were any errors.
120  *
121  *  (7)	After dropping all locks, invoke zil_commit(zilog, seq, foid)
122  *	to ensure that synchronous semantics are provided when necessary.
123  *
124  * In general, this is how things should be ordered in each vnode op:
125  *
126  *	ZFS_ENTER(zfsvfs);		// exit if unmounted
127  * top:
128  *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
129  *	rw_enter(...);			// grab any other locks you need
130  *	tx = dmu_tx_create(...);	// get DMU tx
131  *	dmu_tx_hold_*();		// hold each object you might modify
132  *	error = dmu_tx_assign(tx, zfsvfs->z_assign);	// try to assign
133  *	if (error) {
134  *		rw_exit(...);		// drop locks
135  *		zfs_dirent_unlock(dl);	// unlock directory entry
136  *		VN_RELE(...);		// release held vnodes
137  *		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
138  *			dmu_tx_wait(tx);
139  *			dmu_tx_abort(tx);
140  *			goto top;
141  *		}
142  *		dmu_tx_abort(tx);	// abort DMU tx
143  *		ZFS_EXIT(zfsvfs);	// finished in zfs
144  *		return (error);		// really out of space
145  *	}
146  *	error = do_real_work();		// do whatever this VOP does
147  *	if (error == 0)
148  *		zfs_log_*(...);		// on success, make ZIL entry
149  *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
150  *	rw_exit(...);			// drop locks
151  *	zfs_dirent_unlock(dl);		// unlock directory entry
152  *	VN_RELE(...);			// release held vnodes
153  *	zil_commit(zilog, seq, foid);	// synchronous when necessary
154  *	ZFS_EXIT(zfsvfs);		// finished in zfs
155  *	return (error);			// done, report error
156  */
157 /* ARGSUSED */
158 static int
159 zfs_open(vnode_t **vpp, int flag, cred_t *cr)
160 {
161 	znode_t	*zp = VTOZ(*vpp);
162 
163 	/* Keep a count of the synchronous opens in the znode */
164 	if (flag & (FSYNC | FDSYNC))
165 		atomic_inc_32(&zp->z_sync_cnt);
166 	return (0);
167 }
168 
169 /* ARGSUSED */
170 static int
171 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr)
172 {
173 	znode_t	*zp = VTOZ(vp);
174 
175 	/* Decrement the synchronous opens in the znode */
176 	if (flag & (FSYNC | FDSYNC))
177 		atomic_dec_32(&zp->z_sync_cnt);
178 
179 	/*
180 	 * Clean up any locks held by this process on the vp.
181 	 */
182 	cleanlocks(vp, ddi_get_pid(), 0);
183 	cleanshares(vp, ddi_get_pid());
184 
185 	return (0);
186 }
187 
188 /*
189  * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
190  * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
191  */
192 static int
193 zfs_holey(vnode_t *vp, int cmd, offset_t *off)
194 {
195 	znode_t	*zp = VTOZ(vp);
196 	uint64_t noff = (uint64_t)*off; /* new offset */
197 	uint64_t file_sz;
198 	int error;
199 	boolean_t hole;
200 
201 	file_sz = zp->z_phys->zp_size;
202 	if (noff >= file_sz)  {
203 		return (ENXIO);
204 	}
205 
206 	if (cmd == _FIO_SEEK_HOLE)
207 		hole = B_TRUE;
208 	else
209 		hole = B_FALSE;
210 
211 	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
212 
213 	/* end of file? */
214 	if ((error == ESRCH) || (noff > file_sz)) {
215 		/*
216 		 * Handle the virtual hole at the end of file.
217 		 */
218 		if (hole) {
219 			*off = file_sz;
220 			return (0);
221 		}
222 		return (ENXIO);
223 	}
224 
225 	if (noff < *off)
226 		return (error);
227 	*off = noff;
228 	return (error);
229 }
230 
231 /* ARGSUSED */
232 static int
233 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
234     int *rvalp)
235 {
236 	offset_t off;
237 	int error;
238 	zfsvfs_t *zfsvfs;
239 
240 	switch (com) {
241 	    case _FIOFFS:
242 		return (zfs_sync(vp->v_vfsp, 0, cred));
243 
244 		/*
245 		 * The following two ioctls are used by bfu.  Faking out,
246 		 * necessary to avoid bfu errors.
247 		 */
248 	    case _FIOGDIO:
249 	    case _FIOSDIO:
250 		return (0);
251 
252 	    case _FIO_SEEK_DATA:
253 	    case _FIO_SEEK_HOLE:
254 		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
255 			return (EFAULT);
256 
257 		zfsvfs = VTOZ(vp)->z_zfsvfs;
258 		ZFS_ENTER(zfsvfs);
259 
260 		/* offset parameter is in/out */
261 		error = zfs_holey(vp, com, &off);
262 		ZFS_EXIT(zfsvfs);
263 		if (error)
264 			return (error);
265 		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
266 			return (EFAULT);
267 		return (0);
268 	}
269 	return (ENOTTY);
270 }
271 
272 /*
273  * When a file is memory mapped, we must keep the IO data synchronized
274  * between the DMU cache and the memory mapped pages.  What this means:
275  *
276  * On Write:	If we find a memory mapped page, we write to *both*
277  *		the page and the dmu buffer.
278  *
279  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
280  *	the file is memory mapped.
281  */
282 static int
283 mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx)
284 {
285 	znode_t	*zp = VTOZ(vp);
286 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
287 	int64_t	start, off;
288 	int len = nbytes;
289 	int error = 0;
290 
291 	start = uio->uio_loffset;
292 	off = start & PAGEOFFSET;
293 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
294 		page_t *pp;
295 		uint64_t bytes = MIN(PAGESIZE - off, len);
296 		uint64_t woff = uio->uio_loffset;
297 
298 		/*
299 		 * We don't want a new page to "appear" in the middle of
300 		 * the file update (because it may not get the write
301 		 * update data), so we grab a lock to block
302 		 * zfs_getpage().
303 		 */
304 		rw_enter(&zp->z_map_lock, RW_WRITER);
305 		if (pp = page_lookup(vp, start, SE_SHARED)) {
306 			caddr_t va;
307 
308 			rw_exit(&zp->z_map_lock);
309 			va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L);
310 			error = uiomove(va+off, bytes, UIO_WRITE, uio);
311 			if (error == 0) {
312 				dmu_write(zfsvfs->z_os, zp->z_id,
313 				    woff, bytes, va+off, tx);
314 			}
315 			ppmapout(va);
316 			page_unlock(pp);
317 		} else {
318 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
319 			    uio, bytes, tx);
320 			rw_exit(&zp->z_map_lock);
321 		}
322 		len -= bytes;
323 		off = 0;
324 		if (error)
325 			break;
326 	}
327 	return (error);
328 }
329 
330 /*
331  * When a file is memory mapped, we must keep the IO data synchronized
332  * between the DMU cache and the memory mapped pages.  What this means:
333  *
334  * On Read:	We "read" preferentially from memory mapped pages,
335  *		else we default from the dmu buffer.
336  *
337  * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
338  *	the file is memory mapped.
339  */
340 static int
341 mappedread(vnode_t *vp, int nbytes, uio_t *uio)
342 {
343 	znode_t *zp = VTOZ(vp);
344 	objset_t *os = zp->z_zfsvfs->z_os;
345 	int64_t	start, off;
346 	int len = nbytes;
347 	int error = 0;
348 
349 	start = uio->uio_loffset;
350 	off = start & PAGEOFFSET;
351 	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
352 		page_t *pp;
353 		uint64_t bytes = MIN(PAGESIZE - off, len);
354 
355 		if (pp = page_lookup(vp, start, SE_SHARED)) {
356 			caddr_t va;
357 
358 			va = ppmapin(pp, PROT_READ, (caddr_t)-1L);
359 			error = uiomove(va + off, bytes, UIO_READ, uio);
360 			ppmapout(va);
361 			page_unlock(pp);
362 		} else {
363 			error = dmu_read_uio(os, zp->z_id, uio, bytes);
364 		}
365 		len -= bytes;
366 		off = 0;
367 		if (error)
368 			break;
369 	}
370 	return (error);
371 }
372 
373 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
374 
375 /*
376  * Read bytes from specified file into supplied buffer.
377  *
378  *	IN:	vp	- vnode of file to be read from.
379  *		uio	- structure supplying read location, range info,
380  *			  and return buffer.
381  *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
382  *		cr	- credentials of caller.
383  *
384  *	OUT:	uio	- updated offset and range, buffer filled.
385  *
386  *	RETURN:	0 if success
387  *		error code if failure
388  *
389  * Side Effects:
390  *	vp - atime updated if byte count > 0
391  */
392 /* ARGSUSED */
393 static int
394 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
395 {
396 	znode_t		*zp = VTOZ(vp);
397 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
398 	objset_t	*os = zfsvfs->z_os;
399 	ssize_t		n, nbytes;
400 	int		error;
401 	rl_t		*rl;
402 
403 	ZFS_ENTER(zfsvfs);
404 
405 	/*
406 	 * Validate file offset
407 	 */
408 	if (uio->uio_loffset < (offset_t)0) {
409 		ZFS_EXIT(zfsvfs);
410 		return (EINVAL);
411 	}
412 
413 	/*
414 	 * Fasttrack empty reads
415 	 */
416 	if (uio->uio_resid == 0) {
417 		ZFS_EXIT(zfsvfs);
418 		return (0);
419 	}
420 
421 	/*
422 	 * Check for mandatory locks
423 	 */
424 	if (MANDMODE((mode_t)zp->z_phys->zp_mode)) {
425 		if (error = chklock(vp, FREAD,
426 		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
427 			ZFS_EXIT(zfsvfs);
428 			return (error);
429 		}
430 	}
431 
432 	/*
433 	 * If we're in FRSYNC mode, sync out this znode before reading it.
434 	 */
435 	if (ioflag & FRSYNC)
436 		zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
437 
438 	/*
439 	 * Lock the range against changes.
440 	 */
441 	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
442 
443 	/*
444 	 * If we are reading past end-of-file we can skip
445 	 * to the end; but we might still need to set atime.
446 	 */
447 	if (uio->uio_loffset >= zp->z_phys->zp_size) {
448 		error = 0;
449 		goto out;
450 	}
451 
452 	ASSERT(uio->uio_loffset < zp->z_phys->zp_size);
453 	n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset);
454 
455 	while (n > 0) {
456 		nbytes = MIN(n, zfs_read_chunk_size -
457 		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
458 
459 		if (vn_has_cached_data(vp))
460 			error = mappedread(vp, nbytes, uio);
461 		else
462 			error = dmu_read_uio(os, zp->z_id, uio, nbytes);
463 		if (error)
464 			break;
465 
466 		n -= nbytes;
467 	}
468 
469 out:
470 	zfs_range_unlock(rl);
471 
472 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
473 	ZFS_EXIT(zfsvfs);
474 	return (error);
475 }
476 
477 /*
478  * Fault in the pages of the first n bytes specified by the uio structure.
479  * 1 byte in each page is touched and the uio struct is unmodified.
480  * Any error will exit this routine as this is only a best
481  * attempt to get the pages resident. This is a copy of ufs_trans_touch().
482  */
483 static void
484 zfs_prefault_write(ssize_t n, struct uio *uio)
485 {
486 	struct iovec *iov;
487 	ulong_t cnt, incr;
488 	caddr_t p;
489 	uint8_t tmp;
490 
491 	iov = uio->uio_iov;
492 
493 	while (n) {
494 		cnt = MIN(iov->iov_len, n);
495 		if (cnt == 0) {
496 			/* empty iov entry */
497 			iov++;
498 			continue;
499 		}
500 		n -= cnt;
501 		/*
502 		 * touch each page in this segment.
503 		 */
504 		p = iov->iov_base;
505 		while (cnt) {
506 			switch (uio->uio_segflg) {
507 			case UIO_USERSPACE:
508 			case UIO_USERISPACE:
509 				if (fuword8(p, &tmp))
510 					return;
511 				break;
512 			case UIO_SYSSPACE:
513 				if (kcopy(p, &tmp, 1))
514 					return;
515 				break;
516 			}
517 			incr = MIN(cnt, PAGESIZE);
518 			p += incr;
519 			cnt -= incr;
520 		}
521 		/*
522 		 * touch the last byte in case it straddles a page.
523 		 */
524 		p--;
525 		switch (uio->uio_segflg) {
526 		case UIO_USERSPACE:
527 		case UIO_USERISPACE:
528 			if (fuword8(p, &tmp))
529 				return;
530 			break;
531 		case UIO_SYSSPACE:
532 			if (kcopy(p, &tmp, 1))
533 				return;
534 			break;
535 		}
536 		iov++;
537 	}
538 }
539 
540 /*
541  * Write the bytes to a file.
542  *
543  *	IN:	vp	- vnode of file to be written to.
544  *		uio	- structure supplying write location, range info,
545  *			  and data buffer.
546  *		ioflag	- FAPPEND flag set if in append mode.
547  *		cr	- credentials of caller.
548  *
549  *	OUT:	uio	- updated offset and range.
550  *
551  *	RETURN:	0 if success
552  *		error code if failure
553  *
554  * Timestamps:
555  *	vp - ctime|mtime updated if byte count > 0
556  */
557 /* ARGSUSED */
558 static int
559 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
560 {
561 	znode_t		*zp = VTOZ(vp);
562 	rlim64_t	limit = uio->uio_llimit;
563 	ssize_t		start_resid = uio->uio_resid;
564 	ssize_t		tx_bytes;
565 	uint64_t	end_size;
566 	dmu_tx_t	*tx;
567 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
568 	zilog_t		*zilog = zfsvfs->z_log;
569 	offset_t	woff;
570 	ssize_t		n, nbytes;
571 	rl_t		*rl;
572 	int		max_blksz = zfsvfs->z_max_blksz;
573 	int		error;
574 
575 	/*
576 	 * Fasttrack empty write
577 	 */
578 	n = start_resid;
579 	if (n == 0)
580 		return (0);
581 
582 	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
583 		limit = MAXOFFSET_T;
584 
585 	ZFS_ENTER(zfsvfs);
586 
587 	/*
588 	 * Pre-fault the pages to ensure slow (eg NFS) pages
589 	 * don't hold up txg.
590 	 */
591 	zfs_prefault_write(n, uio);
592 
593 	/*
594 	 * If in append mode, set the io offset pointer to eof.
595 	 */
596 	if (ioflag & FAPPEND) {
597 		/*
598 		 * Range lock for a file append:
599 		 * The value for the start of range will be determined by
600 		 * zfs_range_lock() (to guarantee append semantics).
601 		 * If this write will cause the block size to increase,
602 		 * zfs_range_lock() will lock the entire file, so we must
603 		 * later reduce the range after we grow the block size.
604 		 */
605 		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
606 		if (rl->r_len == UINT64_MAX) {
607 			/* overlocked, zp_size can't change */
608 			woff = uio->uio_loffset = zp->z_phys->zp_size;
609 		} else {
610 			woff = uio->uio_loffset = rl->r_off;
611 		}
612 	} else {
613 		woff = uio->uio_loffset;
614 		/*
615 		 * Validate file offset
616 		 */
617 		if (woff < 0) {
618 			ZFS_EXIT(zfsvfs);
619 			return (EINVAL);
620 		}
621 
622 		/*
623 		 * If we need to grow the block size then zfs_range_lock()
624 		 * will lock a wider range than we request here.
625 		 * Later after growing the block size we reduce the range.
626 		 */
627 		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
628 	}
629 
630 	if (woff >= limit) {
631 		zfs_range_unlock(rl);
632 		ZFS_EXIT(zfsvfs);
633 		return (EFBIG);
634 	}
635 
636 	if ((woff + n) > limit || woff > (limit - n))
637 		n = limit - woff;
638 
639 	/*
640 	 * Check for mandatory locks
641 	 */
642 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) &&
643 	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
644 		zfs_range_unlock(rl);
645 		ZFS_EXIT(zfsvfs);
646 		return (error);
647 	}
648 	end_size = MAX(zp->z_phys->zp_size, woff + n);
649 
650 	/*
651 	 * Write the file in reasonable size chunks.  Each chunk is written
652 	 * in a separate transaction; this keeps the intent log records small
653 	 * and allows us to do more fine-grained space accounting.
654 	 */
655 	while (n > 0) {
656 		/*
657 		 * Start a transaction.
658 		 */
659 		woff = uio->uio_loffset;
660 		tx = dmu_tx_create(zfsvfs->z_os);
661 		dmu_tx_hold_bonus(tx, zp->z_id);
662 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
663 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
664 		if (error) {
665 			if (error == ERESTART &&
666 			    zfsvfs->z_assign == TXG_NOWAIT) {
667 				dmu_tx_wait(tx);
668 				dmu_tx_abort(tx);
669 				continue;
670 			}
671 			dmu_tx_abort(tx);
672 			break;
673 		}
674 
675 		/*
676 		 * If zfs_range_lock() over-locked we grow the blocksize
677 		 * and then reduce the lock range.  This will only happen
678 		 * on the first iteration since zfs_range_reduce() will
679 		 * shrink down r_len to the appropriate size.
680 		 */
681 		if (rl->r_len == UINT64_MAX) {
682 			uint64_t new_blksz;
683 
684 			if (zp->z_blksz > max_blksz) {
685 				ASSERT(!ISP2(zp->z_blksz));
686 				new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE);
687 			} else {
688 				new_blksz = MIN(end_size, max_blksz);
689 			}
690 			zfs_grow_blocksize(zp, new_blksz, tx);
691 			zfs_range_reduce(rl, woff, n);
692 		}
693 
694 		/*
695 		 * XXX - should we really limit each write to z_max_blksz?
696 		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
697 		 */
698 		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
699 		rw_enter(&zp->z_map_lock, RW_READER);
700 
701 		tx_bytes = uio->uio_resid;
702 		if (vn_has_cached_data(vp)) {
703 			rw_exit(&zp->z_map_lock);
704 			error = mappedwrite(vp, nbytes, uio, tx);
705 		} else {
706 			error = dmu_write_uio(zfsvfs->z_os, zp->z_id,
707 			    uio, nbytes, tx);
708 			rw_exit(&zp->z_map_lock);
709 		}
710 		tx_bytes -= uio->uio_resid;
711 
712 		/*
713 		 * If we made no progress, we're done.  If we made even
714 		 * partial progress, update the znode and ZIL accordingly.
715 		 */
716 		if (tx_bytes == 0) {
717 			ASSERT(error != 0);
718 			break;
719 		}
720 
721 		/*
722 		 * Clear Set-UID/Set-GID bits on successful write if not
723 		 * privileged and at least one of the excute bits is set.
724 		 *
725 		 * It would be nice to to this after all writes have
726 		 * been done, but that would still expose the ISUID/ISGID
727 		 * to another app after the partial write is committed.
728 		 */
729 		mutex_enter(&zp->z_acl_lock);
730 		if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) |
731 		    (S_IXUSR >> 6))) != 0 &&
732 		    (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 &&
733 		    secpolicy_vnode_setid_retain(cr,
734 		    (zp->z_phys->zp_mode & S_ISUID) != 0 &&
735 		    zp->z_phys->zp_uid == 0) != 0) {
736 			    zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID);
737 		}
738 		mutex_exit(&zp->z_acl_lock);
739 
740 		/*
741 		 * Update time stamp.  NOTE: This marks the bonus buffer as
742 		 * dirty, so we don't have to do it again for zp_size.
743 		 */
744 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
745 
746 		/*
747 		 * Update the file size (zp_size) if it has changed;
748 		 * account for possible concurrent updates.
749 		 */
750 		while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset)
751 			(void) atomic_cas_64(&zp->z_phys->zp_size, end_size,
752 			    uio->uio_loffset);
753 		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
754 		dmu_tx_commit(tx);
755 
756 		if (error != 0)
757 			break;
758 		ASSERT(tx_bytes == nbytes);
759 		n -= nbytes;
760 	}
761 
762 	zfs_range_unlock(rl);
763 
764 	/*
765 	 * If we're in replay mode, or we made no progress, return error.
766 	 * Otherwise, it's at least a partial write, so it's successful.
767 	 */
768 	if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) {
769 		ZFS_EXIT(zfsvfs);
770 		return (error);
771 	}
772 
773 	if (ioflag & (FSYNC | FDSYNC))
774 		zil_commit(zilog, zp->z_last_itx, zp->z_id);
775 
776 	ZFS_EXIT(zfsvfs);
777 	return (0);
778 }
779 
780 void
781 zfs_get_done(dmu_buf_t *db, void *vzgd)
782 {
783 	zgd_t *zgd = (zgd_t *)vzgd;
784 	rl_t *rl = zgd->zgd_rl;
785 	vnode_t *vp = ZTOV(rl->r_zp);
786 
787 	dmu_buf_rele(db, vzgd);
788 	zfs_range_unlock(rl);
789 	VN_RELE(vp);
790 	zil_add_vdev(zgd->zgd_zilog, DVA_GET_VDEV(BP_IDENTITY(zgd->zgd_bp)));
791 	kmem_free(zgd, sizeof (zgd_t));
792 }
793 
794 /*
795  * Get data to generate a TX_WRITE intent log record.
796  */
797 int
798 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
799 {
800 	zfsvfs_t *zfsvfs = arg;
801 	objset_t *os = zfsvfs->z_os;
802 	znode_t *zp;
803 	uint64_t off = lr->lr_offset;
804 	dmu_buf_t *db;
805 	rl_t *rl;
806 	zgd_t *zgd;
807 	int dlen = lr->lr_length;		/* length of user data */
808 	int error = 0;
809 
810 	ASSERT(zio);
811 	ASSERT(dlen != 0);
812 
813 	/*
814 	 * Nothing to do if the file has been removed
815 	 */
816 	if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0)
817 		return (ENOENT);
818 	if (zp->z_unlinked) {
819 		VN_RELE(ZTOV(zp));
820 		return (ENOENT);
821 	}
822 
823 	/*
824 	 * Write records come in two flavors: immediate and indirect.
825 	 * For small writes it's cheaper to store the data with the
826 	 * log record (immediate); for large writes it's cheaper to
827 	 * sync the data and get a pointer to it (indirect) so that
828 	 * we don't have to write the data twice.
829 	 */
830 	if (buf != NULL) { /* immediate write */
831 		rl = zfs_range_lock(zp, off, dlen, RL_READER);
832 		/* test for truncation needs to be done while range locked */
833 		if (off >= zp->z_phys->zp_size) {
834 			error = ENOENT;
835 			goto out;
836 		}
837 		VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf));
838 	} else { /* indirect write */
839 		uint64_t boff; /* block starting offset */
840 
841 		ASSERT3U(dlen, <=, zp->z_blksz);
842 		/*
843 		 * Have to lock the whole block to ensure when it's
844 		 * written out and it's checksum is being calculated
845 		 * that no one can change the data. We need to re-check
846 		 * blocksize after we get the lock in case it's changed!
847 		 */
848 		for (;;) {
849 			if (ISP2(zp->z_blksz)) {
850 				boff = P2ALIGN_TYPED(off, zp->z_blksz,
851 				    uint64_t);
852 			} else {
853 				boff = 0;
854 			}
855 			dlen = zp->z_blksz;
856 			rl = zfs_range_lock(zp, boff, dlen, RL_READER);
857 			if (zp->z_blksz == dlen)
858 				break;
859 			zfs_range_unlock(rl);
860 		}
861 		/* test for truncation needs to be done while range locked */
862 		if (off >= zp->z_phys->zp_size) {
863 			error = ENOENT;
864 			goto out;
865 		}
866 		zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP);
867 		zgd->zgd_rl = rl;
868 		zgd->zgd_zilog = zfsvfs->z_log;
869 		zgd->zgd_bp = &lr->lr_blkptr;
870 		VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db));
871 		ASSERT(boff == db->db_offset);
872 		lr->lr_blkoff = off - boff;
873 		error = dmu_sync(zio, db, &lr->lr_blkptr,
874 		    lr->lr_common.lrc_txg, zfs_get_done, zgd);
875 		if (error == 0) {
876 			zil_add_vdev(zfsvfs->z_log,
877 			    DVA_GET_VDEV(BP_IDENTITY(&lr->lr_blkptr)));
878 		}
879 		/*
880 		 * If we get EINPROGRESS, then we need to wait for a
881 		 * write IO initiated by dmu_sync() to complete before
882 		 * we can release this dbuf.  We will finish everything
883 		 * up in the zfs_get_done() callback.
884 		 */
885 		if (error == EINPROGRESS)
886 			return (0);
887 		dmu_buf_rele(db, zgd);
888 		kmem_free(zgd, sizeof (zgd_t));
889 	}
890 out:
891 	zfs_range_unlock(rl);
892 	VN_RELE(ZTOV(zp));
893 	return (error);
894 }
895 
896 /*ARGSUSED*/
897 static int
898 zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr)
899 {
900 	znode_t *zp = VTOZ(vp);
901 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
902 	int error;
903 
904 	ZFS_ENTER(zfsvfs);
905 	error = zfs_zaccess_rwx(zp, mode, cr);
906 	ZFS_EXIT(zfsvfs);
907 	return (error);
908 }
909 
910 /*
911  * Lookup an entry in a directory, or an extended attribute directory.
912  * If it exists, return a held vnode reference for it.
913  *
914  *	IN:	dvp	- vnode of directory to search.
915  *		nm	- name of entry to lookup.
916  *		pnp	- full pathname to lookup [UNUSED].
917  *		flags	- LOOKUP_XATTR set if looking for an attribute.
918  *		rdir	- root directory vnode [UNUSED].
919  *		cr	- credentials of caller.
920  *
921  *	OUT:	vpp	- vnode of located entry, NULL if not found.
922  *
923  *	RETURN:	0 if success
924  *		error code if failure
925  *
926  * Timestamps:
927  *	NA
928  */
929 /* ARGSUSED */
930 static int
931 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
932     int flags, vnode_t *rdir, cred_t *cr)
933 {
934 
935 	znode_t *zdp = VTOZ(dvp);
936 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
937 	int	error;
938 
939 	ZFS_ENTER(zfsvfs);
940 
941 	*vpp = NULL;
942 
943 	if (flags & LOOKUP_XATTR) {
944 		/*
945 		 * If the xattr property is off, refuse the lookup request.
946 		 */
947 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
948 			ZFS_EXIT(zfsvfs);
949 			return (EINVAL);
950 		}
951 
952 		/*
953 		 * We don't allow recursive attributes..
954 		 * Maybe someday we will.
955 		 */
956 		if (zdp->z_phys->zp_flags & ZFS_XATTR) {
957 			ZFS_EXIT(zfsvfs);
958 			return (EINVAL);
959 		}
960 
961 		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
962 			ZFS_EXIT(zfsvfs);
963 			return (error);
964 		}
965 
966 		/*
967 		 * Do we have permission to get into attribute directory?
968 		 */
969 
970 		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) {
971 			VN_RELE(*vpp);
972 		}
973 
974 		ZFS_EXIT(zfsvfs);
975 		return (error);
976 	}
977 
978 	if (dvp->v_type != VDIR) {
979 		ZFS_EXIT(zfsvfs);
980 		return (ENOTDIR);
981 	}
982 
983 	/*
984 	 * Check accessibility of directory.
985 	 */
986 
987 	if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) {
988 		ZFS_EXIT(zfsvfs);
989 		return (error);
990 	}
991 
992 	if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) {
993 
994 		/*
995 		 * Convert device special files
996 		 */
997 		if (IS_DEVVP(*vpp)) {
998 			vnode_t	*svp;
999 
1000 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1001 			VN_RELE(*vpp);
1002 			if (svp == NULL)
1003 				error = ENOSYS;
1004 			else
1005 				*vpp = svp;
1006 		}
1007 	}
1008 
1009 	ZFS_EXIT(zfsvfs);
1010 	return (error);
1011 }
1012 
1013 /*
1014  * Attempt to create a new entry in a directory.  If the entry
1015  * already exists, truncate the file if permissible, else return
1016  * an error.  Return the vp of the created or trunc'd file.
1017  *
1018  *	IN:	dvp	- vnode of directory to put new file entry in.
1019  *		name	- name of new file entry.
1020  *		vap	- attributes of new file.
1021  *		excl	- flag indicating exclusive or non-exclusive mode.
1022  *		mode	- mode to open file with.
1023  *		cr	- credentials of caller.
1024  *		flag	- large file flag [UNUSED].
1025  *
1026  *	OUT:	vpp	- vnode of created or trunc'd entry.
1027  *
1028  *	RETURN:	0 if success
1029  *		error code if failure
1030  *
1031  * Timestamps:
1032  *	dvp - ctime|mtime updated if new entry created
1033  *	 vp - ctime|mtime always, atime if new
1034  */
1035 /* ARGSUSED */
1036 static int
1037 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1038     int mode, vnode_t **vpp, cred_t *cr, int flag)
1039 {
1040 	znode_t		*zp, *dzp = VTOZ(dvp);
1041 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1042 	zilog_t		*zilog = zfsvfs->z_log;
1043 	objset_t	*os = zfsvfs->z_os;
1044 	zfs_dirlock_t	*dl;
1045 	dmu_tx_t	*tx;
1046 	int		error;
1047 	uint64_t	zoid;
1048 
1049 	ZFS_ENTER(zfsvfs);
1050 
1051 top:
1052 	*vpp = NULL;
1053 
1054 	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1055 		vap->va_mode &= ~VSVTX;
1056 
1057 	if (*name == '\0') {
1058 		/*
1059 		 * Null component name refers to the directory itself.
1060 		 */
1061 		VN_HOLD(dvp);
1062 		zp = dzp;
1063 		dl = NULL;
1064 		error = 0;
1065 	} else {
1066 		/* possible VN_HOLD(zp) */
1067 		if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) {
1068 			if (strcmp(name, "..") == 0)
1069 				error = EISDIR;
1070 			ZFS_EXIT(zfsvfs);
1071 			return (error);
1072 		}
1073 	}
1074 
1075 	zoid = zp ? zp->z_id : -1ULL;
1076 
1077 	if (zp == NULL) {
1078 		/*
1079 		 * Create a new file object and update the directory
1080 		 * to reference it.
1081 		 */
1082 		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
1083 			goto out;
1084 		}
1085 
1086 		/*
1087 		 * We only support the creation of regular files in
1088 		 * extended attribute directories.
1089 		 */
1090 		if ((dzp->z_phys->zp_flags & ZFS_XATTR) &&
1091 		    (vap->va_type != VREG)) {
1092 			error = EINVAL;
1093 			goto out;
1094 		}
1095 
1096 		tx = dmu_tx_create(os);
1097 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1098 		dmu_tx_hold_bonus(tx, dzp->z_id);
1099 		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1100 		if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
1101 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1102 			    0, SPA_MAXBLOCKSIZE);
1103 		error = dmu_tx_assign(tx, zfsvfs->z_assign);
1104 		if (error) {
1105 			zfs_dirent_unlock(dl);
1106 			if (error == ERESTART &&
1107 			    zfsvfs->z_assign == TXG_NOWAIT) {
1108 				dmu_tx_wait(tx);
1109 				dmu_tx_abort(tx);
1110 				goto top;
1111 			}
1112 			dmu_tx_abort(tx);
1113 			ZFS_EXIT(zfsvfs);
1114 			return (error);
1115 		}
1116 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
1117 		ASSERT(zp->z_id == zoid);
1118 		(void) zfs_link_create(dl, zp, tx, ZNEW);
1119 		zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name);
1120 		dmu_tx_commit(tx);
1121 	} else {
1122 		/*
1123 		 * A directory entry already exists for this name.
1124 		 */
1125 		/*
1126 		 * Can't truncate an existing file if in exclusive mode.
1127 		 */
1128 		if (excl == EXCL) {
1129 			error = EEXIST;
1130 			goto out;
1131 		}
1132 		/*
1133 		 * Can't open a directory for writing.
1134 		 */
1135 		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1136 			error = EISDIR;
1137 			goto out;
1138 		}
1139 		/*
1140 		 * Verify requested access to file.
1141 		 */
1142 		if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) {
1143 			goto out;
1144 		}
1145 
1146 		mutex_enter(&dzp->z_lock);
1147 		dzp->z_seq++;
1148 		mutex_exit(&dzp->z_lock);
1149 
1150 		/*
1151 		 * Truncate regular files if requested.
1152 		 */
1153 		if ((ZTOV(zp)->v_type == VREG) &&
1154 		    (zp->z_phys->zp_size != 0) &&
1155 		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1156 			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1157 			if (error == ERESTART &&
1158 			    zfsvfs->z_assign == TXG_NOWAIT) {
1159 				/* NB: we already did dmu_tx_wait() */
1160 				zfs_dirent_unlock(dl);
1161 				VN_RELE(ZTOV(zp));
1162 				goto top;
1163 			}
1164 		}
1165 	}
1166 out:
1167 
1168 	if (dl)
1169 		zfs_dirent_unlock(dl);
1170 
1171 	if (error) {
1172 		if (zp)
1173 			VN_RELE(ZTOV(zp));
1174 	} else {
1175 		*vpp = ZTOV(zp);
1176 		/*
1177 		 * If vnode is for a device return a specfs vnode instead.
1178 		 */
1179 		if (IS_DEVVP(*vpp)) {
1180 			struct vnode *svp;
1181 
1182 			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1183 			VN_RELE(*vpp);
1184 			if (svp == NULL) {
1185 				error = ENOSYS;
1186 			}
1187 			*vpp = svp;
1188 		}
1189 	}
1190 
1191 	ZFS_EXIT(zfsvfs);
1192 	return (error);
1193 }
1194 
1195 /*
1196  * Remove an entry from a directory.
1197  *
1198  *	IN:	dvp	- vnode of directory to remove entry from.
1199  *		name	- name of entry to remove.
1200  *		cr	- credentials of caller.
1201  *
1202  *	RETURN:	0 if success
1203  *		error code if failure
1204  *
1205  * Timestamps:
1206  *	dvp - ctime|mtime
1207  *	 vp - ctime (if nlink > 0)
1208  */
1209 static int
1210 zfs_remove(vnode_t *dvp, char *name, cred_t *cr)
1211 {
1212 	znode_t		*zp, *dzp = VTOZ(dvp);
1213 	znode_t		*xzp = NULL;
1214 	vnode_t		*vp;
1215 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1216 	zilog_t		*zilog = zfsvfs->z_log;
1217 	uint64_t	acl_obj, xattr_obj;
1218 	zfs_dirlock_t	*dl;
1219 	dmu_tx_t	*tx;
1220 	boolean_t	may_delete_now, delete_now = FALSE;
1221 	boolean_t	unlinked;
1222 	int		error;
1223 
1224 	ZFS_ENTER(zfsvfs);
1225 
1226 top:
1227 	/*
1228 	 * Attempt to lock directory; fail if entry doesn't exist.
1229 	 */
1230 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) {
1231 		ZFS_EXIT(zfsvfs);
1232 		return (error);
1233 	}
1234 
1235 	vp = ZTOV(zp);
1236 
1237 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1238 		goto out;
1239 	}
1240 
1241 	/*
1242 	 * Need to use rmdir for removing directories.
1243 	 */
1244 	if (vp->v_type == VDIR) {
1245 		error = EPERM;
1246 		goto out;
1247 	}
1248 
1249 	vnevent_remove(vp);
1250 
1251 	dnlc_remove(dvp, name);
1252 
1253 	mutex_enter(&vp->v_lock);
1254 	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1255 	mutex_exit(&vp->v_lock);
1256 
1257 	/*
1258 	 * We may delete the znode now, or we may put it in the unlinked set;
1259 	 * it depends on whether we're the last link, and on whether there are
1260 	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1261 	 * allow for either case.
1262 	 */
1263 	tx = dmu_tx_create(zfsvfs->z_os);
1264 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1265 	dmu_tx_hold_bonus(tx, zp->z_id);
1266 	if (may_delete_now)
1267 		dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
1268 
1269 	/* are there any extended attributes? */
1270 	if ((xattr_obj = zp->z_phys->zp_xattr) != 0) {
1271 		/* XXX - do we need this if we are deleting? */
1272 		dmu_tx_hold_bonus(tx, xattr_obj);
1273 	}
1274 
1275 	/* are there any additional acls */
1276 	if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 &&
1277 	    may_delete_now)
1278 		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1279 
1280 	/* charge as an update -- would be nice not to charge at all */
1281 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1282 
1283 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1284 	if (error) {
1285 		zfs_dirent_unlock(dl);
1286 		VN_RELE(vp);
1287 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1288 			dmu_tx_wait(tx);
1289 			dmu_tx_abort(tx);
1290 			goto top;
1291 		}
1292 		dmu_tx_abort(tx);
1293 		ZFS_EXIT(zfsvfs);
1294 		return (error);
1295 	}
1296 
1297 	/*
1298 	 * Remove the directory entry.
1299 	 */
1300 	error = zfs_link_destroy(dl, zp, tx, 0, &unlinked);
1301 
1302 	if (error) {
1303 		dmu_tx_commit(tx);
1304 		goto out;
1305 	}
1306 
1307 	if (unlinked) {
1308 		mutex_enter(&vp->v_lock);
1309 		delete_now = may_delete_now &&
1310 		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1311 		    zp->z_phys->zp_xattr == xattr_obj &&
1312 		    zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj;
1313 		mutex_exit(&vp->v_lock);
1314 	}
1315 
1316 	if (delete_now) {
1317 		if (zp->z_phys->zp_xattr) {
1318 			error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp);
1319 			ASSERT3U(error, ==, 0);
1320 			ASSERT3U(xzp->z_phys->zp_links, ==, 2);
1321 			dmu_buf_will_dirty(xzp->z_dbuf, tx);
1322 			mutex_enter(&xzp->z_lock);
1323 			xzp->z_unlinked = 1;
1324 			xzp->z_phys->zp_links = 0;
1325 			mutex_exit(&xzp->z_lock);
1326 			zfs_unlinked_add(xzp, tx);
1327 			zp->z_phys->zp_xattr = 0; /* probably unnecessary */
1328 		}
1329 		mutex_enter(&zp->z_lock);
1330 		mutex_enter(&vp->v_lock);
1331 		vp->v_count--;
1332 		ASSERT3U(vp->v_count, ==, 0);
1333 		mutex_exit(&vp->v_lock);
1334 		mutex_exit(&zp->z_lock);
1335 		zfs_znode_delete(zp, tx);
1336 		VFS_RELE(zfsvfs->z_vfs);
1337 	} else if (unlinked) {
1338 		zfs_unlinked_add(zp, tx);
1339 	}
1340 
1341 	zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name);
1342 
1343 	dmu_tx_commit(tx);
1344 out:
1345 	zfs_dirent_unlock(dl);
1346 
1347 	if (!delete_now) {
1348 		VN_RELE(vp);
1349 	} else if (xzp) {
1350 		/* this rele delayed to prevent nesting transactions */
1351 		VN_RELE(ZTOV(xzp));
1352 	}
1353 
1354 	ZFS_EXIT(zfsvfs);
1355 	return (error);
1356 }
1357 
1358 /*
1359  * Create a new directory and insert it into dvp using the name
1360  * provided.  Return a pointer to the inserted directory.
1361  *
1362  *	IN:	dvp	- vnode of directory to add subdir to.
1363  *		dirname	- name of new directory.
1364  *		vap	- attributes of new directory.
1365  *		cr	- credentials of caller.
1366  *
1367  *	OUT:	vpp	- vnode of created directory.
1368  *
1369  *	RETURN:	0 if success
1370  *		error code if failure
1371  *
1372  * Timestamps:
1373  *	dvp - ctime|mtime updated
1374  *	 vp - ctime|mtime|atime updated
1375  */
1376 static int
1377 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr)
1378 {
1379 	znode_t		*zp, *dzp = VTOZ(dvp);
1380 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1381 	zilog_t		*zilog = zfsvfs->z_log;
1382 	zfs_dirlock_t	*dl;
1383 	uint64_t	zoid = 0;
1384 	dmu_tx_t	*tx;
1385 	int		error;
1386 
1387 	ASSERT(vap->va_type == VDIR);
1388 
1389 	ZFS_ENTER(zfsvfs);
1390 
1391 	if (dzp->z_phys->zp_flags & ZFS_XATTR) {
1392 		ZFS_EXIT(zfsvfs);
1393 		return (EINVAL);
1394 	}
1395 top:
1396 	*vpp = NULL;
1397 
1398 	/*
1399 	 * First make sure the new directory doesn't exist.
1400 	 */
1401 	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) {
1402 		ZFS_EXIT(zfsvfs);
1403 		return (error);
1404 	}
1405 
1406 	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) {
1407 		zfs_dirent_unlock(dl);
1408 		ZFS_EXIT(zfsvfs);
1409 		return (error);
1410 	}
1411 
1412 	/*
1413 	 * Add a new entry to the directory.
1414 	 */
1415 	tx = dmu_tx_create(zfsvfs->z_os);
1416 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1417 	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1418 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
1419 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1420 		    0, SPA_MAXBLOCKSIZE);
1421 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1422 	if (error) {
1423 		zfs_dirent_unlock(dl);
1424 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1425 			dmu_tx_wait(tx);
1426 			dmu_tx_abort(tx);
1427 			goto top;
1428 		}
1429 		dmu_tx_abort(tx);
1430 		ZFS_EXIT(zfsvfs);
1431 		return (error);
1432 	}
1433 
1434 	/*
1435 	 * Create new node.
1436 	 */
1437 	zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
1438 
1439 	/*
1440 	 * Now put new name in parent dir.
1441 	 */
1442 	(void) zfs_link_create(dl, zp, tx, ZNEW);
1443 
1444 	*vpp = ZTOV(zp);
1445 
1446 	zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname);
1447 	dmu_tx_commit(tx);
1448 
1449 	zfs_dirent_unlock(dl);
1450 
1451 	ZFS_EXIT(zfsvfs);
1452 	return (0);
1453 }
1454 
1455 /*
1456  * Remove a directory subdir entry.  If the current working
1457  * directory is the same as the subdir to be removed, the
1458  * remove will fail.
1459  *
1460  *	IN:	dvp	- vnode of directory to remove from.
1461  *		name	- name of directory to be removed.
1462  *		cwd	- vnode of current working directory.
1463  *		cr	- credentials of caller.
1464  *
1465  *	RETURN:	0 if success
1466  *		error code if failure
1467  *
1468  * Timestamps:
1469  *	dvp - ctime|mtime updated
1470  */
1471 static int
1472 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr)
1473 {
1474 	znode_t		*dzp = VTOZ(dvp);
1475 	znode_t		*zp;
1476 	vnode_t		*vp;
1477 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1478 	zilog_t		*zilog = zfsvfs->z_log;
1479 	zfs_dirlock_t	*dl;
1480 	dmu_tx_t	*tx;
1481 	int		error;
1482 
1483 	ZFS_ENTER(zfsvfs);
1484 
1485 top:
1486 	zp = NULL;
1487 
1488 	/*
1489 	 * Attempt to lock directory; fail if entry doesn't exist.
1490 	 */
1491 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) {
1492 		ZFS_EXIT(zfsvfs);
1493 		return (error);
1494 	}
1495 
1496 	vp = ZTOV(zp);
1497 
1498 	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1499 		goto out;
1500 	}
1501 
1502 	if (vp->v_type != VDIR) {
1503 		error = ENOTDIR;
1504 		goto out;
1505 	}
1506 
1507 	if (vp == cwd) {
1508 		error = EINVAL;
1509 		goto out;
1510 	}
1511 
1512 	vnevent_rmdir(vp);
1513 
1514 	/*
1515 	 * Grab a lock on the parent pointer make sure we play well
1516 	 * with the treewalk and directory rename code.
1517 	 */
1518 	rw_enter(&zp->z_parent_lock, RW_WRITER);
1519 
1520 	tx = dmu_tx_create(zfsvfs->z_os);
1521 	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1522 	dmu_tx_hold_bonus(tx, zp->z_id);
1523 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1524 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
1525 	if (error) {
1526 		rw_exit(&zp->z_parent_lock);
1527 		zfs_dirent_unlock(dl);
1528 		VN_RELE(vp);
1529 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
1530 			dmu_tx_wait(tx);
1531 			dmu_tx_abort(tx);
1532 			goto top;
1533 		}
1534 		dmu_tx_abort(tx);
1535 		ZFS_EXIT(zfsvfs);
1536 		return (error);
1537 	}
1538 
1539 	error = zfs_link_destroy(dl, zp, tx, 0, NULL);
1540 
1541 	if (error == 0)
1542 		zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name);
1543 
1544 	dmu_tx_commit(tx);
1545 
1546 	rw_exit(&zp->z_parent_lock);
1547 out:
1548 	zfs_dirent_unlock(dl);
1549 
1550 	VN_RELE(vp);
1551 
1552 	ZFS_EXIT(zfsvfs);
1553 	return (error);
1554 }
1555 
1556 /*
1557  * Read as many directory entries as will fit into the provided
1558  * buffer from the given directory cursor position (specified in
1559  * the uio structure.
1560  *
1561  *	IN:	vp	- vnode of directory to read.
1562  *		uio	- structure supplying read location, range info,
1563  *			  and return buffer.
1564  *		cr	- credentials of caller.
1565  *
1566  *	OUT:	uio	- updated offset and range, buffer filled.
1567  *		eofp	- set to true if end-of-file detected.
1568  *
1569  *	RETURN:	0 if success
1570  *		error code if failure
1571  *
1572  * Timestamps:
1573  *	vp - atime updated
1574  *
1575  * Note that the low 4 bits of the cookie returned by zap is always zero.
1576  * This allows us to use the low range for "special" directory entries:
1577  * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
1578  * we use the offset 2 for the '.zfs' directory.
1579  */
1580 /* ARGSUSED */
1581 static int
1582 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp)
1583 {
1584 	znode_t		*zp = VTOZ(vp);
1585 	iovec_t		*iovp;
1586 	dirent64_t	*odp;
1587 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1588 	objset_t	*os;
1589 	caddr_t		outbuf;
1590 	size_t		bufsize;
1591 	zap_cursor_t	zc;
1592 	zap_attribute_t	zap;
1593 	uint_t		bytes_wanted;
1594 	ushort_t	this_reclen;
1595 	uint64_t	offset; /* must be unsigned; checks for < 1 */
1596 	off64_t		*next;
1597 	int		local_eof;
1598 	int		outcount;
1599 	int		error;
1600 	uint8_t		prefetch;
1601 
1602 	ZFS_ENTER(zfsvfs);
1603 
1604 	/*
1605 	 * If we are not given an eof variable,
1606 	 * use a local one.
1607 	 */
1608 	if (eofp == NULL)
1609 		eofp = &local_eof;
1610 
1611 	/*
1612 	 * Check for valid iov_len.
1613 	 */
1614 	if (uio->uio_iov->iov_len <= 0) {
1615 		ZFS_EXIT(zfsvfs);
1616 		return (EINVAL);
1617 	}
1618 
1619 	/*
1620 	 * Quit if directory has been removed (posix)
1621 	 */
1622 	if ((*eofp = zp->z_unlinked) != 0) {
1623 		ZFS_EXIT(zfsvfs);
1624 		return (0);
1625 	}
1626 
1627 	error = 0;
1628 	os = zfsvfs->z_os;
1629 	offset = uio->uio_loffset;
1630 	prefetch = zp->z_zn_prefetch;
1631 
1632 	/*
1633 	 * Initialize the iterator cursor.
1634 	 */
1635 	if (offset <= 3) {
1636 		/*
1637 		 * Start iteration from the beginning of the directory.
1638 		 */
1639 		zap_cursor_init(&zc, os, zp->z_id);
1640 	} else {
1641 		/*
1642 		 * The offset is a serialized cursor.
1643 		 */
1644 		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
1645 	}
1646 
1647 	/*
1648 	 * Get space to change directory entries into fs independent format.
1649 	 */
1650 	iovp = uio->uio_iov;
1651 	bytes_wanted = iovp->iov_len;
1652 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
1653 		bufsize = bytes_wanted;
1654 		outbuf = kmem_alloc(bufsize, KM_SLEEP);
1655 		odp = (struct dirent64 *)outbuf;
1656 	} else {
1657 		bufsize = bytes_wanted;
1658 		odp = (struct dirent64 *)iovp->iov_base;
1659 	}
1660 
1661 	/*
1662 	 * Transform to file-system independent format
1663 	 */
1664 	outcount = 0;
1665 	while (outcount < bytes_wanted) {
1666 		/*
1667 		 * Special case `.', `..', and `.zfs'.
1668 		 */
1669 		if (offset == 0) {
1670 			(void) strcpy(zap.za_name, ".");
1671 			zap.za_first_integer = zp->z_id;
1672 			this_reclen = DIRENT64_RECLEN(1);
1673 		} else if (offset == 1) {
1674 			(void) strcpy(zap.za_name, "..");
1675 			zap.za_first_integer = zp->z_phys->zp_parent;
1676 			this_reclen = DIRENT64_RECLEN(2);
1677 		} else if (offset == 2 && zfs_show_ctldir(zp)) {
1678 			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
1679 			zap.za_first_integer = ZFSCTL_INO_ROOT;
1680 			this_reclen =
1681 			    DIRENT64_RECLEN(sizeof (ZFS_CTLDIR_NAME) - 1);
1682 		} else {
1683 			/*
1684 			 * Grab next entry.
1685 			 */
1686 			if (error = zap_cursor_retrieve(&zc, &zap)) {
1687 				if ((*eofp = (error == ENOENT)) != 0)
1688 					break;
1689 				else
1690 					goto update;
1691 			}
1692 
1693 			if (zap.za_integer_length != 8 ||
1694 			    zap.za_num_integers != 1) {
1695 				cmn_err(CE_WARN, "zap_readdir: bad directory "
1696 				    "entry, obj = %lld, offset = %lld\n",
1697 				    (u_longlong_t)zp->z_id,
1698 				    (u_longlong_t)offset);
1699 				error = ENXIO;
1700 				goto update;
1701 			}
1702 			this_reclen = DIRENT64_RECLEN(strlen(zap.za_name));
1703 		}
1704 
1705 		/*
1706 		 * Will this entry fit in the buffer?
1707 		 */
1708 		if (outcount + this_reclen > bufsize) {
1709 			/*
1710 			 * Did we manage to fit anything in the buffer?
1711 			 */
1712 			if (!outcount) {
1713 				error = EINVAL;
1714 				goto update;
1715 			}
1716 			break;
1717 		}
1718 		/*
1719 		 * Add this entry:
1720 		 */
1721 		odp->d_ino = (ino64_t)zap.za_first_integer;
1722 		odp->d_reclen = (ushort_t)this_reclen;
1723 		/* NOTE: d_off is the offset for the *next* entry */
1724 		next = &(odp->d_off);
1725 		(void) strncpy(odp->d_name, zap.za_name,
1726 		    DIRENT64_NAMELEN(this_reclen));
1727 		outcount += this_reclen;
1728 		odp = (dirent64_t *)((intptr_t)odp + this_reclen);
1729 
1730 		ASSERT(outcount <= bufsize);
1731 
1732 		/* Prefetch znode */
1733 		if (prefetch)
1734 			dmu_prefetch(os, zap.za_first_integer, 0, 0);
1735 
1736 		/*
1737 		 * Move to the next entry, fill in the previous offset.
1738 		 */
1739 		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
1740 			zap_cursor_advance(&zc);
1741 			offset = zap_cursor_serialize(&zc);
1742 		} else {
1743 			offset += 1;
1744 		}
1745 		*next = offset;
1746 	}
1747 	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
1748 
1749 	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
1750 		iovp->iov_base += outcount;
1751 		iovp->iov_len -= outcount;
1752 		uio->uio_resid -= outcount;
1753 	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
1754 		/*
1755 		 * Reset the pointer.
1756 		 */
1757 		offset = uio->uio_loffset;
1758 	}
1759 
1760 update:
1761 	zap_cursor_fini(&zc);
1762 	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
1763 		kmem_free(outbuf, bufsize);
1764 
1765 	if (error == ENOENT)
1766 		error = 0;
1767 
1768 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
1769 
1770 	uio->uio_loffset = offset;
1771 	ZFS_EXIT(zfsvfs);
1772 	return (error);
1773 }
1774 
1775 static int
1776 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr)
1777 {
1778 	znode_t	*zp = VTOZ(vp);
1779 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1780 
1781 	/*
1782 	 * Regardless of whether this is required for standards conformance,
1783 	 * this is the logical behavior when fsync() is called on a file with
1784 	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
1785 	 * going to be pushed out as part of the zil_commit().
1786 	 */
1787 	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
1788 	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
1789 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr);
1790 
1791 	ZFS_ENTER(zfsvfs);
1792 	zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id);
1793 	ZFS_EXIT(zfsvfs);
1794 	return (0);
1795 }
1796 
1797 /*
1798  * Get the requested file attributes and place them in the provided
1799  * vattr structure.
1800  *
1801  *	IN:	vp	- vnode of file.
1802  *		vap	- va_mask identifies requested attributes.
1803  *		flags	- [UNUSED]
1804  *		cr	- credentials of caller.
1805  *
1806  *	OUT:	vap	- attribute values.
1807  *
1808  *	RETURN:	0 (always succeeds)
1809  */
1810 /* ARGSUSED */
1811 static int
1812 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr)
1813 {
1814 	znode_t *zp = VTOZ(vp);
1815 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1816 	znode_phys_t *pzp = zp->z_phys;
1817 	int	error;
1818 
1819 	ZFS_ENTER(zfsvfs);
1820 
1821 	/*
1822 	 * Return all attributes.  It's cheaper to provide the answer
1823 	 * than to determine whether we were asked the question.
1824 	 */
1825 	mutex_enter(&zp->z_lock);
1826 
1827 	vap->va_type = vp->v_type;
1828 	vap->va_mode = pzp->zp_mode & MODEMASK;
1829 	vap->va_uid = zp->z_phys->zp_uid;
1830 	vap->va_gid = zp->z_phys->zp_gid;
1831 	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
1832 	vap->va_nodeid = zp->z_id;
1833 	vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX);	/* nlink_t limit! */
1834 	vap->va_size = pzp->zp_size;
1835 	vap->va_rdev = vp->v_rdev;
1836 	vap->va_seq = zp->z_seq;
1837 
1838 	ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime);
1839 	ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime);
1840 	ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime);
1841 
1842 	/*
1843 	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
1844 	 * Also, if we are the owner don't bother, since owner should
1845 	 * always be allowed to read basic attributes of file.
1846 	 */
1847 	if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) &&
1848 	    (zp->z_phys->zp_uid != crgetuid(cr))) {
1849 		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) {
1850 			mutex_exit(&zp->z_lock);
1851 			ZFS_EXIT(zfsvfs);
1852 			return (error);
1853 		}
1854 	}
1855 
1856 	mutex_exit(&zp->z_lock);
1857 
1858 	dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks);
1859 
1860 	if (zp->z_blksz == 0) {
1861 		/*
1862 		 * Block size hasn't been set; suggest maximal I/O transfers.
1863 		 */
1864 		vap->va_blksize = zfsvfs->z_max_blksz;
1865 	}
1866 
1867 	ZFS_EXIT(zfsvfs);
1868 	return (0);
1869 }
1870 
1871 /*
1872  * Set the file attributes to the values contained in the
1873  * vattr structure.
1874  *
1875  *	IN:	vp	- vnode of file to be modified.
1876  *		vap	- new attribute values.
1877  *		flags	- ATTR_UTIME set if non-default time values provided.
1878  *		cr	- credentials of caller.
1879  *
1880  *	RETURN:	0 if success
1881  *		error code if failure
1882  *
1883  * Timestamps:
1884  *	vp - ctime updated, mtime updated if size changed.
1885  */
1886 /* ARGSUSED */
1887 static int
1888 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1889 	caller_context_t *ct)
1890 {
1891 	struct znode	*zp = VTOZ(vp);
1892 	znode_phys_t	*pzp = zp->z_phys;
1893 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1894 	zilog_t		*zilog = zfsvfs->z_log;
1895 	dmu_tx_t	*tx;
1896 	vattr_t		oldva;
1897 	uint_t		mask = vap->va_mask;
1898 	uint_t		saved_mask;
1899 	int		trim_mask = 0;
1900 	uint64_t	new_mode;
1901 	znode_t		*attrzp;
1902 	int		need_policy = FALSE;
1903 	int		err;
1904 
1905 	if (mask == 0)
1906 		return (0);
1907 
1908 	if (mask & AT_NOSET)
1909 		return (EINVAL);
1910 
1911 	if (mask & AT_SIZE && vp->v_type == VDIR)
1912 		return (EISDIR);
1913 
1914 	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO)
1915 		return (EINVAL);
1916 
1917 	ZFS_ENTER(zfsvfs);
1918 
1919 top:
1920 	attrzp = NULL;
1921 
1922 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
1923 		ZFS_EXIT(zfsvfs);
1924 		return (EROFS);
1925 	}
1926 
1927 	/*
1928 	 * First validate permissions
1929 	 */
1930 
1931 	if (mask & AT_SIZE) {
1932 		err = zfs_zaccess(zp, ACE_WRITE_DATA, cr);
1933 		if (err) {
1934 			ZFS_EXIT(zfsvfs);
1935 			return (err);
1936 		}
1937 		/*
1938 		 * XXX - Note, we are not providing any open
1939 		 * mode flags here (like FNDELAY), so we may
1940 		 * block if there are locks present... this
1941 		 * should be addressed in openat().
1942 		 */
1943 		do {
1944 			err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
1945 			/* NB: we already did dmu_tx_wait() if necessary */
1946 		} while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
1947 		if (err) {
1948 			ZFS_EXIT(zfsvfs);
1949 			return (err);
1950 		}
1951 	}
1952 
1953 	if (mask & (AT_ATIME|AT_MTIME))
1954 		need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr);
1955 
1956 	if (mask & (AT_UID|AT_GID)) {
1957 		int	idmask = (mask & (AT_UID|AT_GID));
1958 		int	take_owner;
1959 		int	take_group;
1960 
1961 		/*
1962 		 * NOTE: even if a new mode is being set,
1963 		 * we may clear S_ISUID/S_ISGID bits.
1964 		 */
1965 
1966 		if (!(mask & AT_MODE))
1967 			vap->va_mode = pzp->zp_mode;
1968 
1969 		/*
1970 		 * Take ownership or chgrp to group we are a member of
1971 		 */
1972 
1973 		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
1974 		take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr);
1975 
1976 		/*
1977 		 * If both AT_UID and AT_GID are set then take_owner and
1978 		 * take_group must both be set in order to allow taking
1979 		 * ownership.
1980 		 *
1981 		 * Otherwise, send the check through secpolicy_vnode_setattr()
1982 		 *
1983 		 */
1984 
1985 		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
1986 		    ((idmask == AT_UID) && take_owner) ||
1987 		    ((idmask == AT_GID) && take_group)) {
1988 			if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) {
1989 				/*
1990 				 * Remove setuid/setgid for non-privileged users
1991 				 */
1992 				secpolicy_setid_clear(vap, cr);
1993 				trim_mask = (mask & (AT_UID|AT_GID));
1994 			} else {
1995 				need_policy =  TRUE;
1996 			}
1997 		} else {
1998 			need_policy =  TRUE;
1999 		}
2000 	}
2001 
2002 	mutex_enter(&zp->z_lock);
2003 	oldva.va_mode = pzp->zp_mode;
2004 	oldva.va_uid = zp->z_phys->zp_uid;
2005 	oldva.va_gid = zp->z_phys->zp_gid;
2006 	mutex_exit(&zp->z_lock);
2007 
2008 	if (mask & AT_MODE) {
2009 		if (zfs_zaccess_v4_perm(zp, ACE_WRITE_ACL, cr) == 0) {
2010 			err = secpolicy_setid_setsticky_clear(vp, vap,
2011 			    &oldva, cr);
2012 			if (err) {
2013 				ZFS_EXIT(zfsvfs);
2014 				return (err);
2015 			}
2016 			trim_mask |= AT_MODE;
2017 		} else {
2018 			need_policy = TRUE;
2019 		}
2020 	}
2021 
2022 	if (need_policy) {
2023 		/*
2024 		 * If trim_mask is set then take ownership
2025 		 * has been granted or write_acl is present and user
2026 		 * has the ability to modify mode.  In that case remove
2027 		 * UID|GID and or MODE from mask so that
2028 		 * secpolicy_vnode_setattr() doesn't revoke it.
2029 		 */
2030 
2031 		if (trim_mask) {
2032 			saved_mask = vap->va_mask;
2033 			vap->va_mask &= ~trim_mask;
2034 
2035 		}
2036 		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2037 		    (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp);
2038 		if (err) {
2039 			ZFS_EXIT(zfsvfs);
2040 			return (err);
2041 		}
2042 
2043 		if (trim_mask)
2044 			vap->va_mask |= saved_mask;
2045 	}
2046 
2047 	/*
2048 	 * secpolicy_vnode_setattr, or take ownership may have
2049 	 * changed va_mask
2050 	 */
2051 	mask = vap->va_mask;
2052 
2053 	tx = dmu_tx_create(zfsvfs->z_os);
2054 	dmu_tx_hold_bonus(tx, zp->z_id);
2055 
2056 	if (mask & AT_MODE) {
2057 		uint64_t pmode = pzp->zp_mode;
2058 
2059 		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
2060 
2061 		if (zp->z_phys->zp_acl.z_acl_extern_obj)
2062 			dmu_tx_hold_write(tx,
2063 			    pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE);
2064 		else
2065 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2066 			    0, ZFS_ACL_SIZE(MAX_ACL_SIZE));
2067 	}
2068 
2069 	if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) {
2070 		err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp);
2071 		if (err) {
2072 			dmu_tx_abort(tx);
2073 			ZFS_EXIT(zfsvfs);
2074 			return (err);
2075 		}
2076 		dmu_tx_hold_bonus(tx, attrzp->z_id);
2077 	}
2078 
2079 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2080 	if (err) {
2081 		if (attrzp)
2082 			VN_RELE(ZTOV(attrzp));
2083 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2084 			dmu_tx_wait(tx);
2085 			dmu_tx_abort(tx);
2086 			goto top;
2087 		}
2088 		dmu_tx_abort(tx);
2089 		ZFS_EXIT(zfsvfs);
2090 		return (err);
2091 	}
2092 
2093 	dmu_buf_will_dirty(zp->z_dbuf, tx);
2094 
2095 	/*
2096 	 * Set each attribute requested.
2097 	 * We group settings according to the locks they need to acquire.
2098 	 *
2099 	 * Note: you cannot set ctime directly, although it will be
2100 	 * updated as a side-effect of calling this function.
2101 	 */
2102 
2103 	mutex_enter(&zp->z_lock);
2104 
2105 	if (mask & AT_MODE) {
2106 		err = zfs_acl_chmod_setattr(zp, new_mode, tx);
2107 		ASSERT3U(err, ==, 0);
2108 	}
2109 
2110 	if (attrzp)
2111 		mutex_enter(&attrzp->z_lock);
2112 
2113 	if (mask & AT_UID) {
2114 		zp->z_phys->zp_uid = (uint64_t)vap->va_uid;
2115 		if (attrzp) {
2116 			attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid;
2117 		}
2118 	}
2119 
2120 	if (mask & AT_GID) {
2121 		zp->z_phys->zp_gid = (uint64_t)vap->va_gid;
2122 		if (attrzp)
2123 			attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid;
2124 	}
2125 
2126 	if (attrzp)
2127 		mutex_exit(&attrzp->z_lock);
2128 
2129 	if (mask & AT_ATIME)
2130 		ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime);
2131 
2132 	if (mask & AT_MTIME)
2133 		ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime);
2134 
2135 	if (mask & AT_SIZE)
2136 		zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx);
2137 	else if (mask != 0)
2138 		zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2139 
2140 	if (mask != 0)
2141 		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask);
2142 
2143 	mutex_exit(&zp->z_lock);
2144 
2145 	if (attrzp)
2146 		VN_RELE(ZTOV(attrzp));
2147 
2148 	dmu_tx_commit(tx);
2149 
2150 	ZFS_EXIT(zfsvfs);
2151 	return (err);
2152 }
2153 
2154 typedef struct zfs_zlock {
2155 	krwlock_t	*zl_rwlock;	/* lock we acquired */
2156 	znode_t		*zl_znode;	/* znode we held */
2157 	struct zfs_zlock *zl_next;	/* next in list */
2158 } zfs_zlock_t;
2159 
2160 /*
2161  * Drop locks and release vnodes that were held by zfs_rename_lock().
2162  */
2163 static void
2164 zfs_rename_unlock(zfs_zlock_t **zlpp)
2165 {
2166 	zfs_zlock_t *zl;
2167 
2168 	while ((zl = *zlpp) != NULL) {
2169 		if (zl->zl_znode != NULL)
2170 			VN_RELE(ZTOV(zl->zl_znode));
2171 		rw_exit(zl->zl_rwlock);
2172 		*zlpp = zl->zl_next;
2173 		kmem_free(zl, sizeof (*zl));
2174 	}
2175 }
2176 
2177 /*
2178  * Search back through the directory tree, using the ".." entries.
2179  * Lock each directory in the chain to prevent concurrent renames.
2180  * Fail any attempt to move a directory into one of its own descendants.
2181  * XXX - z_parent_lock can overlap with map or grow locks
2182  */
2183 static int
2184 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
2185 {
2186 	zfs_zlock_t	*zl;
2187 	znode_t		*zp = tdzp;
2188 	uint64_t	rootid = zp->z_zfsvfs->z_root;
2189 	uint64_t	*oidp = &zp->z_id;
2190 	krwlock_t	*rwlp = &szp->z_parent_lock;
2191 	krw_t		rw = RW_WRITER;
2192 
2193 	/*
2194 	 * First pass write-locks szp and compares to zp->z_id.
2195 	 * Later passes read-lock zp and compare to zp->z_parent.
2196 	 */
2197 	do {
2198 		if (!rw_tryenter(rwlp, rw)) {
2199 			/*
2200 			 * Another thread is renaming in this path.
2201 			 * Note that if we are a WRITER, we don't have any
2202 			 * parent_locks held yet.
2203 			 */
2204 			if (rw == RW_READER && zp->z_id > szp->z_id) {
2205 				/*
2206 				 * Drop our locks and restart
2207 				 */
2208 				zfs_rename_unlock(&zl);
2209 				*zlpp = NULL;
2210 				zp = tdzp;
2211 				oidp = &zp->z_id;
2212 				rwlp = &szp->z_parent_lock;
2213 				rw = RW_WRITER;
2214 				continue;
2215 			} else {
2216 				/*
2217 				 * Wait for other thread to drop its locks
2218 				 */
2219 				rw_enter(rwlp, rw);
2220 			}
2221 		}
2222 
2223 		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
2224 		zl->zl_rwlock = rwlp;
2225 		zl->zl_znode = NULL;
2226 		zl->zl_next = *zlpp;
2227 		*zlpp = zl;
2228 
2229 		if (*oidp == szp->z_id)		/* We're a descendant of szp */
2230 			return (EINVAL);
2231 
2232 		if (*oidp == rootid)		/* We've hit the top */
2233 			return (0);
2234 
2235 		if (rw == RW_READER) {		/* i.e. not the first pass */
2236 			int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp);
2237 			if (error)
2238 				return (error);
2239 			zl->zl_znode = zp;
2240 		}
2241 		oidp = &zp->z_phys->zp_parent;
2242 		rwlp = &zp->z_parent_lock;
2243 		rw = RW_READER;
2244 
2245 	} while (zp->z_id != sdzp->z_id);
2246 
2247 	return (0);
2248 }
2249 
2250 /*
2251  * Move an entry from the provided source directory to the target
2252  * directory.  Change the entry name as indicated.
2253  *
2254  *	IN:	sdvp	- Source directory containing the "old entry".
2255  *		snm	- Old entry name.
2256  *		tdvp	- Target directory to contain the "new entry".
2257  *		tnm	- New entry name.
2258  *		cr	- credentials of caller.
2259  *
2260  *	RETURN:	0 if success
2261  *		error code if failure
2262  *
2263  * Timestamps:
2264  *	sdvp,tdvp - ctime|mtime updated
2265  */
2266 static int
2267 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr)
2268 {
2269 	znode_t		*tdzp, *szp, *tzp;
2270 	znode_t		*sdzp = VTOZ(sdvp);
2271 	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
2272 	zilog_t		*zilog = zfsvfs->z_log;
2273 	vnode_t		*realvp;
2274 	zfs_dirlock_t	*sdl, *tdl;
2275 	dmu_tx_t	*tx;
2276 	zfs_zlock_t	*zl;
2277 	int		cmp, serr, terr, error;
2278 
2279 	ZFS_ENTER(zfsvfs);
2280 
2281 	/*
2282 	 * Make sure we have the real vp for the target directory.
2283 	 */
2284 	if (VOP_REALVP(tdvp, &realvp) == 0)
2285 		tdvp = realvp;
2286 
2287 	if (tdvp->v_vfsp != sdvp->v_vfsp) {
2288 		ZFS_EXIT(zfsvfs);
2289 		return (EXDEV);
2290 	}
2291 
2292 	tdzp = VTOZ(tdvp);
2293 top:
2294 	szp = NULL;
2295 	tzp = NULL;
2296 	zl = NULL;
2297 
2298 	/*
2299 	 * This is to prevent the creation of links into attribute space
2300 	 * by renaming a linked file into/outof an attribute directory.
2301 	 * See the comment in zfs_link() for why this is considered bad.
2302 	 */
2303 	if ((tdzp->z_phys->zp_flags & ZFS_XATTR) !=
2304 	    (sdzp->z_phys->zp_flags & ZFS_XATTR)) {
2305 		ZFS_EXIT(zfsvfs);
2306 		return (EINVAL);
2307 	}
2308 
2309 	/*
2310 	 * Lock source and target directory entries.  To prevent deadlock,
2311 	 * a lock ordering must be defined.  We lock the directory with
2312 	 * the smallest object id first, or if it's a tie, the one with
2313 	 * the lexically first name.
2314 	 */
2315 	if (sdzp->z_id < tdzp->z_id) {
2316 		cmp = -1;
2317 	} else if (sdzp->z_id > tdzp->z_id) {
2318 		cmp = 1;
2319 	} else {
2320 		cmp = strcmp(snm, tnm);
2321 		if (cmp == 0) {
2322 			/*
2323 			 * POSIX: "If the old argument and the new argument
2324 			 * both refer to links to the same existing file,
2325 			 * the rename() function shall return successfully
2326 			 * and perform no other action."
2327 			 */
2328 			ZFS_EXIT(zfsvfs);
2329 			return (0);
2330 		}
2331 	}
2332 	if (cmp < 0) {
2333 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS);
2334 		terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0);
2335 	} else {
2336 		terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0);
2337 		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS);
2338 	}
2339 
2340 	if (serr) {
2341 		/*
2342 		 * Source entry invalid or not there.
2343 		 */
2344 		if (!terr) {
2345 			zfs_dirent_unlock(tdl);
2346 			if (tzp)
2347 				VN_RELE(ZTOV(tzp));
2348 		}
2349 		if (strcmp(snm, "..") == 0)
2350 			serr = EINVAL;
2351 		ZFS_EXIT(zfsvfs);
2352 		return (serr);
2353 	}
2354 	if (terr) {
2355 		zfs_dirent_unlock(sdl);
2356 		VN_RELE(ZTOV(szp));
2357 		if (strcmp(tnm, "..") == 0)
2358 			terr = EINVAL;
2359 		ZFS_EXIT(zfsvfs);
2360 		return (terr);
2361 	}
2362 
2363 	/*
2364 	 * Must have write access at the source to remove the old entry
2365 	 * and write access at the target to create the new entry.
2366 	 * Note that if target and source are the same, this can be
2367 	 * done in a single check.
2368 	 */
2369 
2370 	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
2371 		goto out;
2372 
2373 	if (ZTOV(szp)->v_type == VDIR) {
2374 		/*
2375 		 * Check to make sure rename is valid.
2376 		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
2377 		 */
2378 		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
2379 			goto out;
2380 	}
2381 
2382 	/*
2383 	 * Does target exist?
2384 	 */
2385 	if (tzp) {
2386 		/*
2387 		 * Source and target must be the same type.
2388 		 */
2389 		if (ZTOV(szp)->v_type == VDIR) {
2390 			if (ZTOV(tzp)->v_type != VDIR) {
2391 				error = ENOTDIR;
2392 				goto out;
2393 			}
2394 		} else {
2395 			if (ZTOV(tzp)->v_type == VDIR) {
2396 				error = EISDIR;
2397 				goto out;
2398 			}
2399 		}
2400 		/*
2401 		 * POSIX dictates that when the source and target
2402 		 * entries refer to the same file object, rename
2403 		 * must do nothing and exit without error.
2404 		 */
2405 		if (szp->z_id == tzp->z_id) {
2406 			error = 0;
2407 			goto out;
2408 		}
2409 	}
2410 
2411 	vnevent_rename_src(ZTOV(szp));
2412 	if (tzp)
2413 		vnevent_rename_dest(ZTOV(tzp));
2414 
2415 	tx = dmu_tx_create(zfsvfs->z_os);
2416 	dmu_tx_hold_bonus(tx, szp->z_id);	/* nlink changes */
2417 	dmu_tx_hold_bonus(tx, sdzp->z_id);	/* nlink changes */
2418 	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
2419 	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
2420 	if (sdzp != tdzp)
2421 		dmu_tx_hold_bonus(tx, tdzp->z_id);	/* nlink changes */
2422 	if (tzp)
2423 		dmu_tx_hold_bonus(tx, tzp->z_id);	/* parent changes */
2424 	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2425 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2426 	if (error) {
2427 		if (zl != NULL)
2428 			zfs_rename_unlock(&zl);
2429 		zfs_dirent_unlock(sdl);
2430 		zfs_dirent_unlock(tdl);
2431 		VN_RELE(ZTOV(szp));
2432 		if (tzp)
2433 			VN_RELE(ZTOV(tzp));
2434 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2435 			dmu_tx_wait(tx);
2436 			dmu_tx_abort(tx);
2437 			goto top;
2438 		}
2439 		dmu_tx_abort(tx);
2440 		ZFS_EXIT(zfsvfs);
2441 		return (error);
2442 	}
2443 
2444 	if (tzp)	/* Attempt to remove the existing target */
2445 		error = zfs_link_destroy(tdl, tzp, tx, 0, NULL);
2446 
2447 	if (error == 0) {
2448 		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
2449 		if (error == 0) {
2450 			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
2451 			ASSERT(error == 0);
2452 			zfs_log_rename(zilog, tx, TX_RENAME, sdzp,
2453 			    sdl->dl_name, tdzp, tdl->dl_name, szp);
2454 		}
2455 	}
2456 
2457 	dmu_tx_commit(tx);
2458 out:
2459 	if (zl != NULL)
2460 		zfs_rename_unlock(&zl);
2461 
2462 	zfs_dirent_unlock(sdl);
2463 	zfs_dirent_unlock(tdl);
2464 
2465 	VN_RELE(ZTOV(szp));
2466 	if (tzp)
2467 		VN_RELE(ZTOV(tzp));
2468 
2469 	ZFS_EXIT(zfsvfs);
2470 	return (error);
2471 }
2472 
2473 /*
2474  * Insert the indicated symbolic reference entry into the directory.
2475  *
2476  *	IN:	dvp	- Directory to contain new symbolic link.
2477  *		link	- Name for new symlink entry.
2478  *		vap	- Attributes of new entry.
2479  *		target	- Target path of new symlink.
2480  *		cr	- credentials of caller.
2481  *
2482  *	RETURN:	0 if success
2483  *		error code if failure
2484  *
2485  * Timestamps:
2486  *	dvp - ctime|mtime updated
2487  */
2488 static int
2489 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr)
2490 {
2491 	znode_t		*zp, *dzp = VTOZ(dvp);
2492 	zfs_dirlock_t	*dl;
2493 	dmu_tx_t	*tx;
2494 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2495 	zilog_t		*zilog = zfsvfs->z_log;
2496 	uint64_t	zoid;
2497 	int		len = strlen(link);
2498 	int		error;
2499 
2500 	ASSERT(vap->va_type == VLNK);
2501 
2502 	ZFS_ENTER(zfsvfs);
2503 top:
2504 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
2505 		ZFS_EXIT(zfsvfs);
2506 		return (error);
2507 	}
2508 
2509 	if (len > MAXPATHLEN) {
2510 		ZFS_EXIT(zfsvfs);
2511 		return (ENAMETOOLONG);
2512 	}
2513 
2514 	/*
2515 	 * Attempt to lock directory; fail if entry already exists.
2516 	 */
2517 	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) {
2518 		ZFS_EXIT(zfsvfs);
2519 		return (error);
2520 	}
2521 
2522 	tx = dmu_tx_create(zfsvfs->z_os);
2523 	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
2524 	dmu_tx_hold_bonus(tx, dzp->z_id);
2525 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
2526 	if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)
2527 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE);
2528 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2529 	if (error) {
2530 		zfs_dirent_unlock(dl);
2531 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2532 			dmu_tx_wait(tx);
2533 			dmu_tx_abort(tx);
2534 			goto top;
2535 		}
2536 		dmu_tx_abort(tx);
2537 		ZFS_EXIT(zfsvfs);
2538 		return (error);
2539 	}
2540 
2541 	dmu_buf_will_dirty(dzp->z_dbuf, tx);
2542 
2543 	/*
2544 	 * Create a new object for the symlink.
2545 	 * Put the link content into bonus buffer if it will fit;
2546 	 * otherwise, store it just like any other file data.
2547 	 */
2548 	zoid = 0;
2549 	if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) {
2550 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len);
2551 		if (len != 0)
2552 			bcopy(link, zp->z_phys + 1, len);
2553 	} else {
2554 		dmu_buf_t *dbp;
2555 
2556 		zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0);
2557 
2558 		/*
2559 		 * Nothing can access the znode yet so no locking needed
2560 		 * for growing the znode's blocksize.
2561 		 */
2562 		zfs_grow_blocksize(zp, len, tx);
2563 
2564 		VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp));
2565 		dmu_buf_will_dirty(dbp, tx);
2566 
2567 		ASSERT3U(len, <=, dbp->db_size);
2568 		bcopy(link, dbp->db_data, len);
2569 		dmu_buf_rele(dbp, FTAG);
2570 	}
2571 	zp->z_phys->zp_size = len;
2572 
2573 	/*
2574 	 * Insert the new object into the directory.
2575 	 */
2576 	(void) zfs_link_create(dl, zp, tx, ZNEW);
2577 out:
2578 	if (error == 0)
2579 		zfs_log_symlink(zilog, tx, TX_SYMLINK, dzp, zp, name, link);
2580 
2581 	dmu_tx_commit(tx);
2582 
2583 	zfs_dirent_unlock(dl);
2584 
2585 	VN_RELE(ZTOV(zp));
2586 
2587 	ZFS_EXIT(zfsvfs);
2588 	return (error);
2589 }
2590 
2591 /*
2592  * Return, in the buffer contained in the provided uio structure,
2593  * the symbolic path referred to by vp.
2594  *
2595  *	IN:	vp	- vnode of symbolic link.
2596  *		uoip	- structure to contain the link path.
2597  *		cr	- credentials of caller.
2598  *
2599  *	OUT:	uio	- structure to contain the link path.
2600  *
2601  *	RETURN:	0 if success
2602  *		error code if failure
2603  *
2604  * Timestamps:
2605  *	vp - atime updated
2606  */
2607 /* ARGSUSED */
2608 static int
2609 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr)
2610 {
2611 	znode_t		*zp = VTOZ(vp);
2612 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2613 	size_t		bufsz;
2614 	int		error;
2615 
2616 	ZFS_ENTER(zfsvfs);
2617 
2618 	bufsz = (size_t)zp->z_phys->zp_size;
2619 	if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) {
2620 		error = uiomove(zp->z_phys + 1,
2621 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
2622 	} else {
2623 		dmu_buf_t *dbp;
2624 		error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp);
2625 		if (error) {
2626 			ZFS_EXIT(zfsvfs);
2627 			return (error);
2628 		}
2629 		error = uiomove(dbp->db_data,
2630 		    MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio);
2631 		dmu_buf_rele(dbp, FTAG);
2632 	}
2633 
2634 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2635 	ZFS_EXIT(zfsvfs);
2636 	return (error);
2637 }
2638 
2639 /*
2640  * Insert a new entry into directory tdvp referencing svp.
2641  *
2642  *	IN:	tdvp	- Directory to contain new entry.
2643  *		svp	- vnode of new entry.
2644  *		name	- name of new entry.
2645  *		cr	- credentials of caller.
2646  *
2647  *	RETURN:	0 if success
2648  *		error code if failure
2649  *
2650  * Timestamps:
2651  *	tdvp - ctime|mtime updated
2652  *	 svp - ctime updated
2653  */
2654 /* ARGSUSED */
2655 static int
2656 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr)
2657 {
2658 	znode_t		*dzp = VTOZ(tdvp);
2659 	znode_t		*tzp, *szp;
2660 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2661 	zilog_t		*zilog = zfsvfs->z_log;
2662 	zfs_dirlock_t	*dl;
2663 	dmu_tx_t	*tx;
2664 	vnode_t		*realvp;
2665 	int		error;
2666 
2667 	ASSERT(tdvp->v_type == VDIR);
2668 
2669 	ZFS_ENTER(zfsvfs);
2670 
2671 	if (VOP_REALVP(svp, &realvp) == 0)
2672 		svp = realvp;
2673 
2674 	if (svp->v_vfsp != tdvp->v_vfsp) {
2675 		ZFS_EXIT(zfsvfs);
2676 		return (EXDEV);
2677 	}
2678 
2679 	szp = VTOZ(svp);
2680 top:
2681 	/*
2682 	 * We do not support links between attributes and non-attributes
2683 	 * because of the potential security risk of creating links
2684 	 * into "normal" file space in order to circumvent restrictions
2685 	 * imposed in attribute space.
2686 	 */
2687 	if ((szp->z_phys->zp_flags & ZFS_XATTR) !=
2688 	    (dzp->z_phys->zp_flags & ZFS_XATTR)) {
2689 		ZFS_EXIT(zfsvfs);
2690 		return (EINVAL);
2691 	}
2692 
2693 	/*
2694 	 * POSIX dictates that we return EPERM here.
2695 	 * Better choices include ENOTSUP or EISDIR.
2696 	 */
2697 	if (svp->v_type == VDIR) {
2698 		ZFS_EXIT(zfsvfs);
2699 		return (EPERM);
2700 	}
2701 
2702 	if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) &&
2703 	    secpolicy_basic_link(cr) != 0) {
2704 		ZFS_EXIT(zfsvfs);
2705 		return (EPERM);
2706 	}
2707 
2708 	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) {
2709 		ZFS_EXIT(zfsvfs);
2710 		return (error);
2711 	}
2712 
2713 	/*
2714 	 * Attempt to lock directory; fail if entry already exists.
2715 	 */
2716 	if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) {
2717 		ZFS_EXIT(zfsvfs);
2718 		return (error);
2719 	}
2720 
2721 	tx = dmu_tx_create(zfsvfs->z_os);
2722 	dmu_tx_hold_bonus(tx, szp->z_id);
2723 	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
2724 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2725 	if (error) {
2726 		zfs_dirent_unlock(dl);
2727 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2728 			dmu_tx_wait(tx);
2729 			dmu_tx_abort(tx);
2730 			goto top;
2731 		}
2732 		dmu_tx_abort(tx);
2733 		ZFS_EXIT(zfsvfs);
2734 		return (error);
2735 	}
2736 
2737 	error = zfs_link_create(dl, szp, tx, 0);
2738 
2739 	if (error == 0)
2740 		zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name);
2741 
2742 	dmu_tx_commit(tx);
2743 
2744 	zfs_dirent_unlock(dl);
2745 
2746 	ZFS_EXIT(zfsvfs);
2747 	return (error);
2748 }
2749 
2750 /*
2751  * zfs_null_putapage() is used when the file system has been force
2752  * unmounted. It just drops the pages.
2753  */
2754 /* ARGSUSED */
2755 static int
2756 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
2757 		size_t *lenp, int flags, cred_t *cr)
2758 {
2759 	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
2760 	return (0);
2761 }
2762 
2763 /*
2764  * Push a page out to disk, klustering if possible.
2765  *
2766  *	IN:	vp	- file to push page to.
2767  *		pp	- page to push.
2768  *		flags	- additional flags.
2769  *		cr	- credentials of caller.
2770  *
2771  *	OUT:	offp	- start of range pushed.
2772  *		lenp	- len of range pushed.
2773  *
2774  *	RETURN:	0 if success
2775  *		error code if failure
2776  *
2777  * NOTE: callers must have locked the page to be pushed.  On
2778  * exit, the page (and all other pages in the kluster) must be
2779  * unlocked.
2780  */
2781 /* ARGSUSED */
2782 static int
2783 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
2784 		size_t *lenp, int flags, cred_t *cr)
2785 {
2786 	znode_t		*zp = VTOZ(vp);
2787 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2788 	zilog_t		*zilog = zfsvfs->z_log;
2789 	dmu_tx_t	*tx;
2790 	rl_t		*rl;
2791 	u_offset_t	off, koff;
2792 	size_t		len, klen;
2793 	int		err;
2794 
2795 	off = pp->p_offset;
2796 	len = PAGESIZE;
2797 	/*
2798 	 * If our blocksize is bigger than the page size, try to kluster
2799 	 * muiltiple pages so that we write a full block (thus avoiding
2800 	 * a read-modify-write).
2801 	 */
2802 	if (zp->z_blksz > PAGESIZE) {
2803 		uint64_t filesz = zp->z_phys->zp_size;
2804 
2805 		if (!ISP2(zp->z_blksz)) {
2806 			/* Only one block in the file. */
2807 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
2808 			koff = 0;
2809 		} else {
2810 			klen = zp->z_blksz;
2811 			koff = P2ALIGN(off, (u_offset_t)klen);
2812 		}
2813 		ASSERT(koff <= filesz);
2814 		if (koff + klen > filesz)
2815 			klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE);
2816 		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
2817 	}
2818 	ASSERT3U(btop(len), ==, btopr(len));
2819 top:
2820 	rl = zfs_range_lock(zp, off, len, RL_WRITER);
2821 	/*
2822 	 * Can't push pages past end-of-file.
2823 	 */
2824 	if (off >= zp->z_phys->zp_size) {
2825 		/* discard all pages */
2826 		flags |= B_INVAL;
2827 		err = 0;
2828 		goto out;
2829 	} else if (off + len > zp->z_phys->zp_size) {
2830 		int npages = btopr(zp->z_phys->zp_size - off);
2831 		page_t *trunc;
2832 
2833 		page_list_break(&pp, &trunc, npages);
2834 		/* discard pages past end of file */
2835 		if (trunc)
2836 			pvn_write_done(trunc, B_INVAL | flags);
2837 		len = zp->z_phys->zp_size - off;
2838 	}
2839 
2840 	tx = dmu_tx_create(zfsvfs->z_os);
2841 	dmu_tx_hold_write(tx, zp->z_id, off, len);
2842 	dmu_tx_hold_bonus(tx, zp->z_id);
2843 	err = dmu_tx_assign(tx, zfsvfs->z_assign);
2844 	if (err != 0) {
2845 		if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2846 			zfs_range_unlock(rl);
2847 			dmu_tx_wait(tx);
2848 			dmu_tx_abort(tx);
2849 			err = 0;
2850 			goto top;
2851 		}
2852 		dmu_tx_abort(tx);
2853 		goto out;
2854 	}
2855 
2856 	if (zp->z_blksz <= PAGESIZE) {
2857 		caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1);
2858 		ASSERT3U(len, <=, PAGESIZE);
2859 		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
2860 		ppmapout(va);
2861 	} else {
2862 		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
2863 	}
2864 
2865 	if (err == 0) {
2866 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
2867 		zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0);
2868 		dmu_tx_commit(tx);
2869 	}
2870 
2871 out:
2872 	zfs_range_unlock(rl);
2873 	pvn_write_done(pp, (err ? B_ERROR : 0) | B_WRITE | flags);
2874 	if (offp)
2875 		*offp = off;
2876 	if (lenp)
2877 		*lenp = len;
2878 
2879 	return (err);
2880 }
2881 
2882 /*
2883  * Copy the portion of the file indicated from pages into the file.
2884  * The pages are stored in a page list attached to the files vnode.
2885  *
2886  *	IN:	vp	- vnode of file to push page data to.
2887  *		off	- position in file to put data.
2888  *		len	- amount of data to write.
2889  *		flags	- flags to control the operation.
2890  *		cr	- credentials of caller.
2891  *
2892  *	RETURN:	0 if success
2893  *		error code if failure
2894  *
2895  * Timestamps:
2896  *	vp - ctime|mtime updated
2897  */
2898 static int
2899 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr)
2900 {
2901 	znode_t		*zp = VTOZ(vp);
2902 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2903 	page_t		*pp;
2904 	size_t		io_len;
2905 	u_offset_t	io_off;
2906 	uint64_t	filesz;
2907 	int		error = 0;
2908 
2909 	ZFS_ENTER(zfsvfs);
2910 
2911 	ASSERT(zp->z_dbuf_held && zp->z_phys);
2912 
2913 	if (len == 0) {
2914 		/*
2915 		 * Search the entire vp list for pages >= off.
2916 		 */
2917 		error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage,
2918 		    flags, cr);
2919 		goto out;
2920 	}
2921 
2922 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
2923 	if (off > filesz) {
2924 		/* past end of file */
2925 		ZFS_EXIT(zfsvfs);
2926 		return (0);
2927 	}
2928 
2929 	len = MIN(len, filesz - off);
2930 
2931 	for (io_off = off; io_off < off + len; io_off += io_len) {
2932 		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
2933 			pp = page_lookup(vp, io_off,
2934 				(flags & (B_INVAL | B_FREE)) ?
2935 					SE_EXCL : SE_SHARED);
2936 		} else {
2937 			pp = page_lookup_nowait(vp, io_off,
2938 				(flags & B_FREE) ? SE_EXCL : SE_SHARED);
2939 		}
2940 
2941 		if (pp != NULL && pvn_getdirty(pp, flags)) {
2942 			int err;
2943 
2944 			/*
2945 			 * Found a dirty page to push
2946 			 */
2947 			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
2948 			if (err)
2949 				error = err;
2950 		} else {
2951 			io_len = PAGESIZE;
2952 		}
2953 	}
2954 out:
2955 	if ((flags & B_ASYNC) == 0)
2956 		zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id);
2957 	ZFS_EXIT(zfsvfs);
2958 	return (error);
2959 }
2960 
2961 void
2962 zfs_inactive(vnode_t *vp, cred_t *cr)
2963 {
2964 	znode_t	*zp = VTOZ(vp);
2965 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2966 	int error;
2967 
2968 	rw_enter(&zfsvfs->z_um_lock, RW_READER);
2969 	if (zfsvfs->z_unmounted2) {
2970 		ASSERT(zp->z_dbuf_held == 0);
2971 
2972 		if (vn_has_cached_data(vp)) {
2973 			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
2974 			    B_INVAL, cr);
2975 		}
2976 
2977 		mutex_enter(&zp->z_lock);
2978 		vp->v_count = 0; /* count arrives as 1 */
2979 		if (zp->z_dbuf == NULL) {
2980 			mutex_exit(&zp->z_lock);
2981 			zfs_znode_free(zp);
2982 		} else {
2983 			mutex_exit(&zp->z_lock);
2984 		}
2985 		rw_exit(&zfsvfs->z_um_lock);
2986 		VFS_RELE(zfsvfs->z_vfs);
2987 		return;
2988 	}
2989 
2990 	/*
2991 	 * Attempt to push any data in the page cache.  If this fails
2992 	 * we will get kicked out later in zfs_zinactive().
2993 	 */
2994 	if (vn_has_cached_data(vp)) {
2995 		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
2996 		    cr);
2997 	}
2998 
2999 	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
3000 		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
3001 
3002 		dmu_tx_hold_bonus(tx, zp->z_id);
3003 		error = dmu_tx_assign(tx, TXG_WAIT);
3004 		if (error) {
3005 			dmu_tx_abort(tx);
3006 		} else {
3007 			dmu_buf_will_dirty(zp->z_dbuf, tx);
3008 			mutex_enter(&zp->z_lock);
3009 			zp->z_atime_dirty = 0;
3010 			mutex_exit(&zp->z_lock);
3011 			dmu_tx_commit(tx);
3012 		}
3013 	}
3014 
3015 	zfs_zinactive(zp);
3016 	rw_exit(&zfsvfs->z_um_lock);
3017 }
3018 
3019 /*
3020  * Bounds-check the seek operation.
3021  *
3022  *	IN:	vp	- vnode seeking within
3023  *		ooff	- old file offset
3024  *		noffp	- pointer to new file offset
3025  *
3026  *	RETURN:	0 if success
3027  *		EINVAL if new offset invalid
3028  */
3029 /* ARGSUSED */
3030 static int
3031 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp)
3032 {
3033 	if (vp->v_type == VDIR)
3034 		return (0);
3035 	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
3036 }
3037 
3038 /*
3039  * Pre-filter the generic locking function to trap attempts to place
3040  * a mandatory lock on a memory mapped file.
3041  */
3042 static int
3043 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
3044     flk_callback_t *flk_cbp, cred_t *cr)
3045 {
3046 	znode_t *zp = VTOZ(vp);
3047 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3048 	int error;
3049 
3050 	ZFS_ENTER(zfsvfs);
3051 
3052 	/*
3053 	 * We are following the UFS semantics with respect to mapcnt
3054 	 * here: If we see that the file is mapped already, then we will
3055 	 * return an error, but we don't worry about races between this
3056 	 * function and zfs_map().
3057 	 */
3058 	if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) {
3059 		ZFS_EXIT(zfsvfs);
3060 		return (EAGAIN);
3061 	}
3062 	error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr);
3063 	ZFS_EXIT(zfsvfs);
3064 	return (error);
3065 }
3066 
3067 /*
3068  * If we can't find a page in the cache, we will create a new page
3069  * and fill it with file data.  For efficiency, we may try to fill
3070  * multiple pages at once (klustering).
3071  */
3072 static int
3073 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
3074     caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
3075 {
3076 	znode_t *zp = VTOZ(vp);
3077 	page_t *pp, *cur_pp;
3078 	objset_t *os = zp->z_zfsvfs->z_os;
3079 	caddr_t va;
3080 	u_offset_t io_off, total;
3081 	uint64_t oid = zp->z_id;
3082 	size_t io_len;
3083 	uint64_t filesz;
3084 	int err;
3085 
3086 	/*
3087 	 * If we are only asking for a single page don't bother klustering.
3088 	 */
3089 	filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */
3090 	if (off >= filesz)
3091 		return (EFAULT);
3092 	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
3093 		io_off = off;
3094 		io_len = PAGESIZE;
3095 		pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr);
3096 	} else {
3097 		/*
3098 		 * Try to fill a kluster of pages (a blocks worth).
3099 		 */
3100 		size_t klen;
3101 		u_offset_t koff;
3102 
3103 		if (!ISP2(zp->z_blksz)) {
3104 			/* Only one block in the file. */
3105 			klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
3106 			koff = 0;
3107 		} else {
3108 			/*
3109 			 * It would be ideal to align our offset to the
3110 			 * blocksize but doing so has resulted in some
3111 			 * strange application crashes. For now, we
3112 			 * leave the offset as is and only adjust the
3113 			 * length if we are off the end of the file.
3114 			 */
3115 			koff = off;
3116 			klen = plsz;
3117 		}
3118 		ASSERT(koff <= filesz);
3119 		if (koff + klen > filesz)
3120 			klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff;
3121 		ASSERT3U(off, >=, koff);
3122 		ASSERT3U(off, <, koff + klen);
3123 		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3124 			    &io_len, koff, klen, 0);
3125 	}
3126 	if (pp == NULL) {
3127 		/*
3128 		 * Some other thread entered the page before us.
3129 		 * Return to zfs_getpage to retry the lookup.
3130 		 */
3131 		*pl = NULL;
3132 		return (0);
3133 	}
3134 
3135 	/*
3136 	 * Fill the pages in the kluster.
3137 	 */
3138 	cur_pp = pp;
3139 	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
3140 		ASSERT3U(io_off, ==, cur_pp->p_offset);
3141 		va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1);
3142 		err = dmu_read(os, oid, io_off, PAGESIZE, va);
3143 		ppmapout(va);
3144 		if (err) {
3145 			/* On error, toss the entire kluster */
3146 			pvn_read_done(pp, B_ERROR);
3147 			return (err);
3148 		}
3149 		cur_pp = cur_pp->p_next;
3150 	}
3151 out:
3152 	/*
3153 	 * Fill in the page list array from the kluster.  If
3154 	 * there are too many pages in the kluster, return
3155 	 * as many pages as possible starting from the desired
3156 	 * offset `off'.
3157 	 * NOTE: the page list will always be null terminated.
3158 	 */
3159 	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3160 
3161 	return (0);
3162 }
3163 
3164 /*
3165  * Return pointers to the pages for the file region [off, off + len]
3166  * in the pl array.  If plsz is greater than len, this function may
3167  * also return page pointers from before or after the specified
3168  * region (i.e. some region [off', off' + plsz]).  These additional
3169  * pages are only returned if they are already in the cache, or were
3170  * created as part of a klustered read.
3171  *
3172  *	IN:	vp	- vnode of file to get data from.
3173  *		off	- position in file to get data from.
3174  *		len	- amount of data to retrieve.
3175  *		plsz	- length of provided page list.
3176  *		seg	- segment to obtain pages for.
3177  *		addr	- virtual address of fault.
3178  *		rw	- mode of created pages.
3179  *		cr	- credentials of caller.
3180  *
3181  *	OUT:	protp	- protection mode of created pages.
3182  *		pl	- list of pages created.
3183  *
3184  *	RETURN:	0 if success
3185  *		error code if failure
3186  *
3187  * Timestamps:
3188  *	vp - atime updated
3189  */
3190 /* ARGSUSED */
3191 static int
3192 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3193 	page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3194 	enum seg_rw rw, cred_t *cr)
3195 {
3196 	znode_t		*zp = VTOZ(vp);
3197 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3198 	page_t		*pp, **pl0 = pl;
3199 	int		need_unlock = 0, err = 0;
3200 	offset_t	orig_off;
3201 
3202 	ZFS_ENTER(zfsvfs);
3203 
3204 	if (protp)
3205 		*protp = PROT_ALL;
3206 
3207 	ASSERT(zp->z_dbuf_held && zp->z_phys);
3208 
3209 	/* no faultahead (for now) */
3210 	if (pl == NULL) {
3211 		ZFS_EXIT(zfsvfs);
3212 		return (0);
3213 	}
3214 
3215 	/* can't fault past EOF */
3216 	if (off >= zp->z_phys->zp_size) {
3217 		ZFS_EXIT(zfsvfs);
3218 		return (EFAULT);
3219 	}
3220 	orig_off = off;
3221 
3222 	/*
3223 	 * If we already own the lock, then we must be page faulting
3224 	 * in the middle of a write to this file (i.e., we are writing
3225 	 * to this file using data from a mapped region of the file).
3226 	 */
3227 	if (rw_owner(&zp->z_map_lock) != curthread) {
3228 		rw_enter(&zp->z_map_lock, RW_WRITER);
3229 		need_unlock = TRUE;
3230 	}
3231 
3232 	/*
3233 	 * Loop through the requested range [off, off + len] looking
3234 	 * for pages.  If we don't find a page, we will need to create
3235 	 * a new page and fill it with data from the file.
3236 	 */
3237 	while (len > 0) {
3238 		if (plsz < PAGESIZE)
3239 			break;
3240 		if (pp = page_lookup(vp, off, SE_SHARED)) {
3241 			*pl++ = pp;
3242 			off += PAGESIZE;
3243 			addr += PAGESIZE;
3244 			len -= PAGESIZE;
3245 			plsz -= PAGESIZE;
3246 		} else {
3247 			err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw);
3248 			if (err)
3249 				goto out;
3250 			/*
3251 			 * klustering may have changed our region
3252 			 * to be block aligned.
3253 			 */
3254 			if (((pp = *pl) != 0) && (off != pp->p_offset)) {
3255 				int delta = off - pp->p_offset;
3256 				len += delta;
3257 				off -= delta;
3258 				addr -= delta;
3259 			}
3260 			while (*pl) {
3261 				pl++;
3262 				off += PAGESIZE;
3263 				addr += PAGESIZE;
3264 				plsz -= PAGESIZE;
3265 				if (len > PAGESIZE)
3266 					len -= PAGESIZE;
3267 				else
3268 					len = 0;
3269 			}
3270 		}
3271 	}
3272 
3273 	/*
3274 	 * Fill out the page array with any pages already in the cache.
3275 	 */
3276 	while (plsz > 0) {
3277 		pp = page_lookup_nowait(vp, off, SE_SHARED);
3278 		if (pp == NULL)
3279 			break;
3280 		*pl++ = pp;
3281 		off += PAGESIZE;
3282 		plsz -= PAGESIZE;
3283 	}
3284 
3285 	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3286 out:
3287 	/*
3288 	 * We can't grab the range lock for the page as reader which would
3289 	 * stop truncation as this leads to deadlock. So we need to recheck
3290 	 * the file size.
3291 	 */
3292 	if (orig_off >= zp->z_phys->zp_size)
3293 		err = EFAULT;
3294 	if (err) {
3295 		/*
3296 		 * Release any pages we have previously locked.
3297 		 */
3298 		while (pl > pl0)
3299 			page_unlock(*--pl);
3300 	}
3301 
3302 	*pl = NULL;
3303 
3304 	if (need_unlock)
3305 		rw_exit(&zp->z_map_lock);
3306 
3307 	ZFS_EXIT(zfsvfs);
3308 	return (err);
3309 }
3310 
3311 /*
3312  * Request a memory map for a section of a file.  This code interacts
3313  * with common code and the VM system as follows:
3314  *
3315  *	common code calls mmap(), which ends up in smmap_common()
3316  *
3317  *	this calls VOP_MAP(), which takes you into (say) zfs
3318  *
3319  *	zfs_map() calls as_map(), passing segvn_create() as the callback
3320  *
3321  *	segvn_create() creates the new segment and calls VOP_ADDMAP()
3322  *
3323  *	zfs_addmap() updates z_mapcnt
3324  */
3325 static int
3326 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
3327     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3328 {
3329 	znode_t *zp = VTOZ(vp);
3330 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3331 	segvn_crargs_t	vn_a;
3332 	int		error;
3333 
3334 	ZFS_ENTER(zfsvfs);
3335 
3336 	if (vp->v_flag & VNOMAP) {
3337 		ZFS_EXIT(zfsvfs);
3338 		return (ENOSYS);
3339 	}
3340 
3341 	if (off < 0 || len > MAXOFFSET_T - off) {
3342 		ZFS_EXIT(zfsvfs);
3343 		return (ENXIO);
3344 	}
3345 
3346 	if (vp->v_type != VREG) {
3347 		ZFS_EXIT(zfsvfs);
3348 		return (ENODEV);
3349 	}
3350 
3351 	/*
3352 	 * If file is locked, disallow mapping.
3353 	 */
3354 	if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) {
3355 		ZFS_EXIT(zfsvfs);
3356 		return (EAGAIN);
3357 	}
3358 
3359 	as_rangelock(as);
3360 	if ((flags & MAP_FIXED) == 0) {
3361 		map_addr(addrp, len, off, 1, flags);
3362 		if (*addrp == NULL) {
3363 			as_rangeunlock(as);
3364 			ZFS_EXIT(zfsvfs);
3365 			return (ENOMEM);
3366 		}
3367 	} else {
3368 		/*
3369 		 * User specified address - blow away any previous mappings
3370 		 */
3371 		(void) as_unmap(as, *addrp, len);
3372 	}
3373 
3374 	vn_a.vp = vp;
3375 	vn_a.offset = (u_offset_t)off;
3376 	vn_a.type = flags & MAP_TYPE;
3377 	vn_a.prot = prot;
3378 	vn_a.maxprot = maxprot;
3379 	vn_a.cred = cr;
3380 	vn_a.amp = NULL;
3381 	vn_a.flags = flags & ~MAP_TYPE;
3382 	vn_a.szc = 0;
3383 	vn_a.lgrp_mem_policy_flags = 0;
3384 
3385 	error = as_map(as, *addrp, len, segvn_create, &vn_a);
3386 
3387 	as_rangeunlock(as);
3388 	ZFS_EXIT(zfsvfs);
3389 	return (error);
3390 }
3391 
3392 /* ARGSUSED */
3393 static int
3394 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3395     size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr)
3396 {
3397 	uint64_t pages = btopr(len);
3398 
3399 	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
3400 	return (0);
3401 }
3402 
3403 /*
3404  * The reason we push dirty pages as part of zfs_delmap() is so that we get a
3405  * more accurate mtime for the associated file.  Since we don't have a way of
3406  * detecting when the data was actually modified, we have to resort to
3407  * heuristics.  If an explicit msync() is done, then we mark the mtime when the
3408  * last page is pushed.  The problem occurs when the msync() call is omitted,
3409  * which by far the most common case:
3410  *
3411  * 	open()
3412  * 	mmap()
3413  * 	<modify memory>
3414  * 	munmap()
3415  * 	close()
3416  * 	<time lapse>
3417  * 	putpage() via fsflush
3418  *
3419  * If we wait until fsflush to come along, we can have a modification time that
3420  * is some arbitrary point in the future.  In order to prevent this in the
3421  * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
3422  * torn down.
3423  */
3424 /* ARGSUSED */
3425 static int
3426 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
3427     size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr)
3428 {
3429 	uint64_t pages = btopr(len);
3430 
3431 	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
3432 	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
3433 
3434 	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
3435 	    vn_has_cached_data(vp))
3436 		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr);
3437 
3438 	return (0);
3439 }
3440 
3441 /*
3442  * Free or allocate space in a file.  Currently, this function only
3443  * supports the `F_FREESP' command.  However, this command is somewhat
3444  * misnamed, as its functionality includes the ability to allocate as
3445  * well as free space.
3446  *
3447  *	IN:	vp	- vnode of file to free data in.
3448  *		cmd	- action to take (only F_FREESP supported).
3449  *		bfp	- section of file to free/alloc.
3450  *		flag	- current file open mode flags.
3451  *		offset	- current file offset.
3452  *		cr	- credentials of caller [UNUSED].
3453  *
3454  *	RETURN:	0 if success
3455  *		error code if failure
3456  *
3457  * Timestamps:
3458  *	vp - ctime|mtime updated
3459  */
3460 /* ARGSUSED */
3461 static int
3462 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
3463     offset_t offset, cred_t *cr, caller_context_t *ct)
3464 {
3465 	znode_t		*zp = VTOZ(vp);
3466 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3467 	uint64_t	off, len;
3468 	int		error;
3469 
3470 	ZFS_ENTER(zfsvfs);
3471 
3472 top:
3473 	if (cmd != F_FREESP) {
3474 		ZFS_EXIT(zfsvfs);
3475 		return (EINVAL);
3476 	}
3477 
3478 	if (error = convoff(vp, bfp, 0, offset)) {
3479 		ZFS_EXIT(zfsvfs);
3480 		return (error);
3481 	}
3482 
3483 	if (bfp->l_len < 0) {
3484 		ZFS_EXIT(zfsvfs);
3485 		return (EINVAL);
3486 	}
3487 
3488 	off = bfp->l_start;
3489 	len = bfp->l_len; /* 0 means from off to end of file */
3490 
3491 	do {
3492 		error = zfs_freesp(zp, off, len, flag, TRUE);
3493 		/* NB: we already did dmu_tx_wait() if necessary */
3494 	} while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT);
3495 
3496 	ZFS_EXIT(zfsvfs);
3497 	return (error);
3498 }
3499 
3500 static int
3501 zfs_fid(vnode_t *vp, fid_t *fidp)
3502 {
3503 	znode_t		*zp = VTOZ(vp);
3504 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3505 	uint32_t	gen = (uint32_t)zp->z_phys->zp_gen;
3506 	uint64_t	object = zp->z_id;
3507 	zfid_short_t	*zfid;
3508 	int		size, i;
3509 
3510 	ZFS_ENTER(zfsvfs);
3511 
3512 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
3513 	if (fidp->fid_len < size) {
3514 		fidp->fid_len = size;
3515 		ZFS_EXIT(zfsvfs);
3516 		return (ENOSPC);
3517 	}
3518 
3519 	zfid = (zfid_short_t *)fidp;
3520 
3521 	zfid->zf_len = size;
3522 
3523 	for (i = 0; i < sizeof (zfid->zf_object); i++)
3524 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
3525 
3526 	/* Must have a non-zero generation number to distinguish from .zfs */
3527 	if (gen == 0)
3528 		gen = 1;
3529 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
3530 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
3531 
3532 	if (size == LONG_FID_LEN) {
3533 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
3534 		zfid_long_t	*zlfid;
3535 
3536 		zlfid = (zfid_long_t *)fidp;
3537 
3538 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
3539 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
3540 
3541 		/* XXX - this should be the generation number for the objset */
3542 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
3543 			zlfid->zf_setgen[i] = 0;
3544 	}
3545 
3546 	ZFS_EXIT(zfsvfs);
3547 	return (0);
3548 }
3549 
3550 static int
3551 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr)
3552 {
3553 	znode_t		*zp, *xzp;
3554 	zfsvfs_t	*zfsvfs;
3555 	zfs_dirlock_t	*dl;
3556 	int		error;
3557 
3558 	switch (cmd) {
3559 	case _PC_LINK_MAX:
3560 		*valp = ULONG_MAX;
3561 		return (0);
3562 
3563 	case _PC_FILESIZEBITS:
3564 		*valp = 64;
3565 		return (0);
3566 
3567 	case _PC_XATTR_EXISTS:
3568 		zp = VTOZ(vp);
3569 		zfsvfs = zp->z_zfsvfs;
3570 		ZFS_ENTER(zfsvfs);
3571 		*valp = 0;
3572 		error = zfs_dirent_lock(&dl, zp, "", &xzp,
3573 		    ZXATTR | ZEXISTS | ZSHARED);
3574 		if (error == 0) {
3575 			zfs_dirent_unlock(dl);
3576 			if (!zfs_dirempty(xzp))
3577 				*valp = 1;
3578 			VN_RELE(ZTOV(xzp));
3579 		} else if (error == ENOENT) {
3580 			/*
3581 			 * If there aren't extended attributes, it's the
3582 			 * same as having zero of them.
3583 			 */
3584 			error = 0;
3585 		}
3586 		ZFS_EXIT(zfsvfs);
3587 		return (error);
3588 
3589 	case _PC_ACL_ENABLED:
3590 		*valp = _ACL_ACE_ENABLED;
3591 		return (0);
3592 
3593 	case _PC_MIN_HOLE_SIZE:
3594 		*valp = (ulong_t)SPA_MINBLOCKSIZE;
3595 		return (0);
3596 
3597 	default:
3598 		return (fs_pathconf(vp, cmd, valp, cr));
3599 	}
3600 }
3601 
3602 /*ARGSUSED*/
3603 static int
3604 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr)
3605 {
3606 	znode_t *zp = VTOZ(vp);
3607 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3608 	int error;
3609 
3610 	ZFS_ENTER(zfsvfs);
3611 	error = zfs_getacl(zp, vsecp, cr);
3612 	ZFS_EXIT(zfsvfs);
3613 
3614 	return (error);
3615 }
3616 
3617 /*ARGSUSED*/
3618 static int
3619 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr)
3620 {
3621 	znode_t *zp = VTOZ(vp);
3622 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
3623 	int error;
3624 
3625 	ZFS_ENTER(zfsvfs);
3626 	error = zfs_setacl(zp, vsecp, cr);
3627 	ZFS_EXIT(zfsvfs);
3628 	return (error);
3629 }
3630 
3631 /*
3632  * Predeclare these here so that the compiler assumes that
3633  * this is an "old style" function declaration that does
3634  * not include arguments => we won't get type mismatch errors
3635  * in the initializations that follow.
3636  */
3637 static int zfs_inval();
3638 static int zfs_isdir();
3639 
3640 static int
3641 zfs_inval()
3642 {
3643 	return (EINVAL);
3644 }
3645 
3646 static int
3647 zfs_isdir()
3648 {
3649 	return (EISDIR);
3650 }
3651 /*
3652  * Directory vnode operations template
3653  */
3654 vnodeops_t *zfs_dvnodeops;
3655 const fs_operation_def_t zfs_dvnodeops_template[] = {
3656 	VOPNAME_OPEN, zfs_open,
3657 	VOPNAME_CLOSE, zfs_close,
3658 	VOPNAME_READ, zfs_isdir,
3659 	VOPNAME_WRITE, zfs_isdir,
3660 	VOPNAME_IOCTL, zfs_ioctl,
3661 	VOPNAME_GETATTR, zfs_getattr,
3662 	VOPNAME_SETATTR, zfs_setattr,
3663 	VOPNAME_ACCESS, zfs_access,
3664 	VOPNAME_LOOKUP, zfs_lookup,
3665 	VOPNAME_CREATE, zfs_create,
3666 	VOPNAME_REMOVE, zfs_remove,
3667 	VOPNAME_LINK, zfs_link,
3668 	VOPNAME_RENAME, zfs_rename,
3669 	VOPNAME_MKDIR, zfs_mkdir,
3670 	VOPNAME_RMDIR, zfs_rmdir,
3671 	VOPNAME_READDIR, zfs_readdir,
3672 	VOPNAME_SYMLINK, zfs_symlink,
3673 	VOPNAME_FSYNC, zfs_fsync,
3674 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3675 	VOPNAME_FID, zfs_fid,
3676 	VOPNAME_SEEK, zfs_seek,
3677 	VOPNAME_PATHCONF, zfs_pathconf,
3678 	VOPNAME_GETSECATTR, zfs_getsecattr,
3679 	VOPNAME_SETSECATTR, zfs_setsecattr,
3680 	NULL, NULL
3681 };
3682 
3683 /*
3684  * Regular file vnode operations template
3685  */
3686 vnodeops_t *zfs_fvnodeops;
3687 const fs_operation_def_t zfs_fvnodeops_template[] = {
3688 	VOPNAME_OPEN, zfs_open,
3689 	VOPNAME_CLOSE, zfs_close,
3690 	VOPNAME_READ, zfs_read,
3691 	VOPNAME_WRITE, zfs_write,
3692 	VOPNAME_IOCTL, zfs_ioctl,
3693 	VOPNAME_GETATTR, zfs_getattr,
3694 	VOPNAME_SETATTR, zfs_setattr,
3695 	VOPNAME_ACCESS, zfs_access,
3696 	VOPNAME_LOOKUP, zfs_lookup,
3697 	VOPNAME_RENAME, zfs_rename,
3698 	VOPNAME_FSYNC, zfs_fsync,
3699 	VOPNAME_INACTIVE, (fs_generic_func_p)zfs_inactive,
3700 	VOPNAME_FID, zfs_fid,
3701 	VOPNAME_SEEK, zfs_seek,
3702 	VOPNAME_FRLOCK, zfs_frlock,
3703 	VOPNAME_SPACE, zfs_space,
3704 	VOPNAME_GETPAGE, zfs_getpage,
3705 	VOPNAME_PUTPAGE, zfs_putpage,
3706 	VOPNAME_MAP, (fs_generic_func_p) zfs_map,
3707 	VOPNAME_ADDMAP, (fs_generic_func_p) zfs_addmap,
3708 	VOPNAME_DELMAP, zfs_delmap,
3709 	VOPNAME_PATHCONF, zfs_pathconf,
3710 	VOPNAME_GETSECATTR, zfs_getsecattr,
3711 	VOPNAME_SETSECATTR, zfs_setsecattr,
3712 	VOPNAME_VNEVENT, fs_vnevent_support,
3713 	NULL, NULL
3714 };
3715 
3716 /*
3717  * Symbolic link vnode operations template
3718  */
3719 vnodeops_t *zfs_symvnodeops;
3720 const fs_operation_def_t zfs_symvnodeops_template[] = {
3721 	VOPNAME_GETATTR, zfs_getattr,
3722 	VOPNAME_SETATTR, zfs_setattr,
3723 	VOPNAME_ACCESS, zfs_access,
3724 	VOPNAME_RENAME, zfs_rename,
3725 	VOPNAME_READLINK, zfs_readlink,
3726 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3727 	VOPNAME_FID, zfs_fid,
3728 	VOPNAME_PATHCONF, zfs_pathconf,
3729 	VOPNAME_VNEVENT, fs_vnevent_support,
3730 	NULL, NULL
3731 };
3732 
3733 /*
3734  * Extended attribute directory vnode operations template
3735  *	This template is identical to the directory vnodes
3736  *	operation template except for restricted operations:
3737  *		VOP_MKDIR()
3738  *		VOP_SYMLINK()
3739  * Note that there are other restrictions embedded in:
3740  *	zfs_create()	- restrict type to VREG
3741  *	zfs_link()	- no links into/out of attribute space
3742  *	zfs_rename()	- no moves into/out of attribute space
3743  */
3744 vnodeops_t *zfs_xdvnodeops;
3745 const fs_operation_def_t zfs_xdvnodeops_template[] = {
3746 	VOPNAME_OPEN, zfs_open,
3747 	VOPNAME_CLOSE, zfs_close,
3748 	VOPNAME_IOCTL, zfs_ioctl,
3749 	VOPNAME_GETATTR, zfs_getattr,
3750 	VOPNAME_SETATTR, zfs_setattr,
3751 	VOPNAME_ACCESS, zfs_access,
3752 	VOPNAME_LOOKUP, zfs_lookup,
3753 	VOPNAME_CREATE, zfs_create,
3754 	VOPNAME_REMOVE, zfs_remove,
3755 	VOPNAME_LINK, zfs_link,
3756 	VOPNAME_RENAME, zfs_rename,
3757 	VOPNAME_MKDIR, zfs_inval,
3758 	VOPNAME_RMDIR, zfs_rmdir,
3759 	VOPNAME_READDIR, zfs_readdir,
3760 	VOPNAME_SYMLINK, zfs_inval,
3761 	VOPNAME_FSYNC, zfs_fsync,
3762 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3763 	VOPNAME_FID, zfs_fid,
3764 	VOPNAME_SEEK, zfs_seek,
3765 	VOPNAME_PATHCONF, zfs_pathconf,
3766 	VOPNAME_GETSECATTR, zfs_getsecattr,
3767 	VOPNAME_SETSECATTR, zfs_setsecattr,
3768 	VOPNAME_VNEVENT, fs_vnevent_support,
3769 	NULL, NULL
3770 };
3771 
3772 /*
3773  * Error vnode operations template
3774  */
3775 vnodeops_t *zfs_evnodeops;
3776 const fs_operation_def_t zfs_evnodeops_template[] = {
3777 	VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive,
3778 	VOPNAME_PATHCONF, zfs_pathconf,
3779 	NULL, NULL
3780 };
3781