1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
25 */
26
27/* Portions Copyright 2007 Jeremy Teo */
28/* Portions Copyright 2010 Robert Milkowski */
29
30#include <sys/types.h>
31#include <sys/param.h>
32#include <sys/time.h>
33#include <sys/systm.h>
34#include <sys/sysmacros.h>
35#include <sys/resource.h>
36#include <sys/vfs.h>
37#include <sys/vfs_opreg.h>
38#include <sys/vnode.h>
39#include <sys/file.h>
40#include <sys/stat.h>
41#include <sys/kmem.h>
42#include <sys/taskq.h>
43#include <sys/uio.h>
44#include <sys/vmsystm.h>
45#include <sys/atomic.h>
46#include <sys/vm.h>
47#include <vm/seg_vn.h>
48#include <vm/pvn.h>
49#include <vm/as.h>
50#include <vm/kpm.h>
51#include <vm/seg_kpm.h>
52#include <sys/mman.h>
53#include <sys/pathname.h>
54#include <sys/cmn_err.h>
55#include <sys/errno.h>
56#include <sys/unistd.h>
57#include <sys/zfs_dir.h>
58#include <sys/zfs_acl.h>
59#include <sys/zfs_ioctl.h>
60#include <sys/fs/zfs.h>
61#include <sys/dmu.h>
62#include <sys/dmu_objset.h>
63#include <sys/spa.h>
64#include <sys/txg.h>
65#include <sys/dbuf.h>
66#include <sys/zap.h>
67#include <sys/sa.h>
68#include <sys/dirent.h>
69#include <sys/policy.h>
70#include <sys/sunddi.h>
71#include <sys/filio.h>
72#include <sys/sid.h>
73#include "fs/fs_subr.h"
74#include <sys/zfs_ctldir.h>
75#include <sys/zfs_fuid.h>
76#include <sys/zfs_sa.h>
77#include <sys/dnlc.h>
78#include <sys/zfs_rlock.h>
79#include <sys/extdirent.h>
80#include <sys/kidmap.h>
81#include <sys/cred.h>
82#include <sys/attr.h>
83
84/*
85 * Programming rules.
86 *
87 * Each vnode op performs some logical unit of work.  To do this, the ZPL must
88 * properly lock its in-core state, create a DMU transaction, do the work,
89 * record this work in the intent log (ZIL), commit the DMU transaction,
90 * and wait for the intent log to commit if it is a synchronous operation.
91 * Moreover, the vnode ops must work in both normal and log replay context.
92 * The ordering of events is important to avoid deadlocks and references
93 * to freed memory.  The example below illustrates the following Big Rules:
94 *
95 *  (1)	A check must be made in each zfs thread for a mounted file system.
96 *	This is done avoiding races using ZFS_ENTER(zfsvfs).
97 *	A ZFS_EXIT(zfsvfs) is needed before all returns.  Any znodes
98 *	must be checked with ZFS_VERIFY_ZP(zp).  Both of these macros
99 *	can return EIO from the calling function.
100 *
101 *  (2)	VN_RELE() should always be the last thing except for zil_commit()
102 *	(if necessary) and ZFS_EXIT(). This is for 3 reasons:
103 *	First, if it's the last reference, the vnode/znode
104 *	can be freed, so the zp may point to freed memory.  Second, the last
105 *	reference will call zfs_zinactive(), which may induce a lot of work --
106 *	pushing cached pages (which acquires range locks) and syncing out
107 *	cached atime changes.  Third, zfs_zinactive() may require a new tx,
108 *	which could deadlock the system if you were already holding one.
109 *	If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
110 *
111 *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
112 *	as they can span dmu_tx_assign() calls.
113 *
114 *  (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
115 *      dmu_tx_assign().  This is critical because we don't want to block
116 *      while holding locks.
117 *
118 *	If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT.  This
119 *	reduces lock contention and CPU usage when we must wait (note that if
120 *	throughput is constrained by the storage, nearly every transaction
121 *	must wait).
122 *
123 *      Note, in particular, that if a lock is sometimes acquired before
124 *      the tx assigns, and sometimes after (e.g. z_lock), then failing
125 *      to use a non-blocking assign can deadlock the system.  The scenario:
126 *
127 *	Thread A has grabbed a lock before calling dmu_tx_assign().
128 *	Thread B is in an already-assigned tx, and blocks for this lock.
129 *	Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
130 *	forever, because the previous txg can't quiesce until B's tx commits.
131 *
132 *	If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
133 *	then drop all locks, call dmu_tx_wait(), and try again.  On subsequent
134 *	calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
135 *	to indicate that this operation has already called dmu_tx_wait().
136 *	This will ensure that we don't retry forever, waiting a short bit
137 *	each time.
138 *
139 *  (5)	If the operation succeeded, generate the intent log entry for it
140 *	before dropping locks.  This ensures that the ordering of events
141 *	in the intent log matches the order in which they actually occurred.
142 *	During ZIL replay the zfs_log_* functions will update the sequence
143 *	number to indicate the zil transaction has replayed.
144 *
145 *  (6)	At the end of each vnode op, the DMU tx must always commit,
146 *	regardless of whether there were any errors.
147 *
148 *  (7)	After dropping all locks, invoke zil_commit(zilog, foid)
149 *	to ensure that synchronous semantics are provided when necessary.
150 *
151 * In general, this is how things should be ordered in each vnode op:
152 *
153 *	ZFS_ENTER(zfsvfs);		// exit if unmounted
154 * top:
155 *	zfs_dirent_lock(&dl, ...)	// lock directory entry (may VN_HOLD())
156 *	rw_enter(...);			// grab any other locks you need
157 *	tx = dmu_tx_create(...);	// get DMU tx
158 *	dmu_tx_hold_*();		// hold each object you might modify
159 *	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
160 *	if (error) {
161 *		rw_exit(...);		// drop locks
162 *		zfs_dirent_unlock(dl);	// unlock directory entry
163 *		VN_RELE(...);		// release held vnodes
164 *		if (error == ERESTART) {
165 *			waited = B_TRUE;
166 *			dmu_tx_wait(tx);
167 *			dmu_tx_abort(tx);
168 *			goto top;
169 *		}
170 *		dmu_tx_abort(tx);	// abort DMU tx
171 *		ZFS_EXIT(zfsvfs);	// finished in zfs
172 *		return (error);		// really out of space
173 *	}
174 *	error = do_real_work();		// do whatever this VOP does
175 *	if (error == 0)
176 *		zfs_log_*(...);		// on success, make ZIL entry
177 *	dmu_tx_commit(tx);		// commit DMU tx -- error or not
178 *	rw_exit(...);			// drop locks
179 *	zfs_dirent_unlock(dl);		// unlock directory entry
180 *	VN_RELE(...);			// release held vnodes
181 *	zil_commit(zilog, foid);	// synchronous when necessary
182 *	ZFS_EXIT(zfsvfs);		// finished in zfs
183 *	return (error);			// done, report error
184 */
185
186/* ARGSUSED */
187static int
188zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
189{
190	znode_t	*zp = VTOZ(*vpp);
191	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
192
193	ZFS_ENTER(zfsvfs);
194	ZFS_VERIFY_ZP(zp);
195
196	if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) &&
197	    ((flag & FAPPEND) == 0)) {
198		ZFS_EXIT(zfsvfs);
199		return (SET_ERROR(EPERM));
200	}
201
202	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
203	    ZTOV(zp)->v_type == VREG &&
204	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) {
205		if (fs_vscan(*vpp, cr, 0) != 0) {
206			ZFS_EXIT(zfsvfs);
207			return (SET_ERROR(EACCES));
208		}
209	}
210
211	/* Keep a count of the synchronous opens in the znode */
212	if (flag & (FSYNC | FDSYNC))
213		atomic_inc_32(&zp->z_sync_cnt);
214
215	ZFS_EXIT(zfsvfs);
216	return (0);
217}
218
219/* ARGSUSED */
220static int
221zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
222    caller_context_t *ct)
223{
224	znode_t	*zp = VTOZ(vp);
225	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
226
227	/*
228	 * Clean up any locks held by this process on the vp.
229	 */
230	cleanlocks(vp, ddi_get_pid(), 0);
231	cleanshares(vp, ddi_get_pid());
232
233	ZFS_ENTER(zfsvfs);
234	ZFS_VERIFY_ZP(zp);
235
236	/* Decrement the synchronous opens in the znode */
237	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
238		atomic_dec_32(&zp->z_sync_cnt);
239
240	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
241	    ZTOV(zp)->v_type == VREG &&
242	    !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0)
243		VERIFY(fs_vscan(vp, cr, 1) == 0);
244
245	ZFS_EXIT(zfsvfs);
246	return (0);
247}
248
249/*
250 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
251 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
252 */
253static int
254zfs_holey(vnode_t *vp, int cmd, offset_t *off)
255{
256	znode_t	*zp = VTOZ(vp);
257	uint64_t noff = (uint64_t)*off; /* new offset */
258	uint64_t file_sz;
259	int error;
260	boolean_t hole;
261
262	file_sz = zp->z_size;
263	if (noff >= file_sz)  {
264		return (SET_ERROR(ENXIO));
265	}
266
267	if (cmd == _FIO_SEEK_HOLE)
268		hole = B_TRUE;
269	else
270		hole = B_FALSE;
271
272	error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff);
273
274	if (error == ESRCH)
275		return (SET_ERROR(ENXIO));
276
277	/*
278	 * We could find a hole that begins after the logical end-of-file,
279	 * because dmu_offset_next() only works on whole blocks.  If the
280	 * EOF falls mid-block, then indicate that the "virtual hole"
281	 * at the end of the file begins at the logical EOF, rather than
282	 * at the end of the last block.
283	 */
284	if (noff > file_sz) {
285		ASSERT(hole);
286		noff = file_sz;
287	}
288
289	if (noff < *off)
290		return (error);
291	*off = noff;
292	return (error);
293}
294
295/* ARGSUSED */
296static int
297zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred,
298    int *rvalp, caller_context_t *ct)
299{
300	offset_t off;
301	offset_t ndata;
302	dmu_object_info_t doi;
303	int error;
304	zfsvfs_t *zfsvfs;
305	znode_t *zp;
306
307	switch (com) {
308	case _FIOFFS:
309	{
310		return (zfs_sync(vp->v_vfsp, 0, cred));
311
312		/*
313		 * The following two ioctls are used by bfu.  Faking out,
314		 * necessary to avoid bfu errors.
315		 */
316	}
317	case _FIOGDIO:
318	case _FIOSDIO:
319	{
320		return (0);
321	}
322
323	case _FIO_SEEK_DATA:
324	case _FIO_SEEK_HOLE:
325	{
326		if (ddi_copyin((void *)data, &off, sizeof (off), flag))
327			return (SET_ERROR(EFAULT));
328
329		zp = VTOZ(vp);
330		zfsvfs = zp->z_zfsvfs;
331		ZFS_ENTER(zfsvfs);
332		ZFS_VERIFY_ZP(zp);
333
334		/* offset parameter is in/out */
335		error = zfs_holey(vp, com, &off);
336		ZFS_EXIT(zfsvfs);
337		if (error)
338			return (error);
339		if (ddi_copyout(&off, (void *)data, sizeof (off), flag))
340			return (SET_ERROR(EFAULT));
341		return (0);
342	}
343	case _FIO_COUNT_FILLED:
344	{
345		/*
346		 * _FIO_COUNT_FILLED adds a new ioctl command which
347		 * exposes the number of filled blocks in a
348		 * ZFS object.
349		 */
350		zp = VTOZ(vp);
351		zfsvfs = zp->z_zfsvfs;
352		ZFS_ENTER(zfsvfs);
353		ZFS_VERIFY_ZP(zp);
354
355		/*
356		 * Wait for all dirty blocks for this object
357		 * to get synced out to disk, and the DMU info
358		 * updated.
359		 */
360		error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id);
361		if (error) {
362			ZFS_EXIT(zfsvfs);
363			return (error);
364		}
365
366		/*
367		 * Retrieve fill count from DMU object.
368		 */
369		error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi);
370		if (error) {
371			ZFS_EXIT(zfsvfs);
372			return (error);
373		}
374
375		ndata = doi.doi_fill_count;
376
377		ZFS_EXIT(zfsvfs);
378		if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag))
379			return (SET_ERROR(EFAULT));
380		return (0);
381	}
382	}
383	return (SET_ERROR(ENOTTY));
384}
385
386/*
387 * Utility functions to map and unmap a single physical page.  These
388 * are used to manage the mappable copies of ZFS file data, and therefore
389 * do not update ref/mod bits.
390 */
391caddr_t
392zfs_map_page(page_t *pp, enum seg_rw rw)
393{
394	if (kpm_enable)
395		return (hat_kpm_mapin(pp, 0));
396	ASSERT(rw == S_READ || rw == S_WRITE);
397	return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0),
398	    (caddr_t)-1));
399}
400
401void
402zfs_unmap_page(page_t *pp, caddr_t addr)
403{
404	if (kpm_enable) {
405		hat_kpm_mapout(pp, 0, addr);
406	} else {
407		ppmapout(addr);
408	}
409}
410
411/*
412 * When a file is memory mapped, we must keep the IO data synchronized
413 * between the DMU cache and the memory mapped pages.  What this means:
414 *
415 * On Write:	If we find a memory mapped page, we write to *both*
416 *		the page and the dmu buffer.
417 */
418static void
419update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid)
420{
421	int64_t	off;
422
423	off = start & PAGEOFFSET;
424	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
425		page_t *pp;
426		uint64_t nbytes = MIN(PAGESIZE - off, len);
427
428		if (pp = page_lookup(vp, start, SE_SHARED)) {
429			caddr_t va;
430
431			va = zfs_map_page(pp, S_WRITE);
432			(void) dmu_read(os, oid, start+off, nbytes, va+off,
433			    DMU_READ_PREFETCH);
434			zfs_unmap_page(pp, va);
435			page_unlock(pp);
436		}
437		len -= nbytes;
438		off = 0;
439	}
440}
441
442/*
443 * When a file is memory mapped, we must keep the IO data synchronized
444 * between the DMU cache and the memory mapped pages.  What this means:
445 *
446 * On Read:	We "read" preferentially from memory mapped pages,
447 *		else we default from the dmu buffer.
448 *
449 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
450 *	 the file is memory mapped.
451 */
452static int
453mappedread(vnode_t *vp, int nbytes, uio_t *uio)
454{
455	znode_t *zp = VTOZ(vp);
456	int64_t	start, off;
457	int len = nbytes;
458	int error = 0;
459
460	start = uio->uio_loffset;
461	off = start & PAGEOFFSET;
462	for (start &= PAGEMASK; len > 0; start += PAGESIZE) {
463		page_t *pp;
464		uint64_t bytes = MIN(PAGESIZE - off, len);
465
466		if (pp = page_lookup(vp, start, SE_SHARED)) {
467			caddr_t va;
468
469			va = zfs_map_page(pp, S_READ);
470			error = uiomove(va + off, bytes, UIO_READ, uio);
471			zfs_unmap_page(pp, va);
472			page_unlock(pp);
473		} else {
474			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
475			    uio, bytes);
476		}
477		len -= bytes;
478		off = 0;
479		if (error)
480			break;
481	}
482	return (error);
483}
484
485offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */
486
487/*
488 * Read bytes from specified file into supplied buffer.
489 *
490 *	IN:	vp	- vnode of file to be read from.
491 *		uio	- structure supplying read location, range info,
492 *			  and return buffer.
493 *		ioflag	- SYNC flags; used to provide FRSYNC semantics.
494 *		cr	- credentials of caller.
495 *		ct	- caller context
496 *
497 *	OUT:	uio	- updated offset and range, buffer filled.
498 *
499 *	RETURN:	0 on success, error code on failure.
500 *
501 * Side Effects:
502 *	vp - atime updated if byte count > 0
503 */
504/* ARGSUSED */
505static int
506zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
507{
508	znode_t		*zp = VTOZ(vp);
509	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
510	ssize_t		n, nbytes;
511	int		error = 0;
512	rl_t		*rl;
513	xuio_t		*xuio = NULL;
514
515	ZFS_ENTER(zfsvfs);
516	ZFS_VERIFY_ZP(zp);
517
518	if (zp->z_pflags & ZFS_AV_QUARANTINED) {
519		ZFS_EXIT(zfsvfs);
520		return (SET_ERROR(EACCES));
521	}
522
523	/*
524	 * Validate file offset
525	 */
526	if (uio->uio_loffset < (offset_t)0) {
527		ZFS_EXIT(zfsvfs);
528		return (SET_ERROR(EINVAL));
529	}
530
531	/*
532	 * Fasttrack empty reads
533	 */
534	if (uio->uio_resid == 0) {
535		ZFS_EXIT(zfsvfs);
536		return (0);
537	}
538
539	/*
540	 * Check for mandatory locks
541	 */
542	if (MANDMODE(zp->z_mode)) {
543		if (error = chklock(vp, FREAD,
544		    uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) {
545			ZFS_EXIT(zfsvfs);
546			return (error);
547		}
548	}
549
550	/*
551	 * If we're in FRSYNC mode, sync out this znode before reading it.
552	 */
553	if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
554		zil_commit(zfsvfs->z_log, zp->z_id);
555
556	/*
557	 * Lock the range against changes.
558	 */
559	rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER);
560
561	/*
562	 * If we are reading past end-of-file we can skip
563	 * to the end; but we might still need to set atime.
564	 */
565	if (uio->uio_loffset >= zp->z_size) {
566		error = 0;
567		goto out;
568	}
569
570	ASSERT(uio->uio_loffset < zp->z_size);
571	n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset);
572
573	if ((uio->uio_extflg == UIO_XUIO) &&
574	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) {
575		int nblk;
576		int blksz = zp->z_blksz;
577		uint64_t offset = uio->uio_loffset;
578
579		xuio = (xuio_t *)uio;
580		if ((ISP2(blksz))) {
581			nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset,
582			    blksz)) / blksz;
583		} else {
584			ASSERT(offset + n <= blksz);
585			nblk = 1;
586		}
587		(void) dmu_xuio_init(xuio, nblk);
588
589		if (vn_has_cached_data(vp)) {
590			/*
591			 * For simplicity, we always allocate a full buffer
592			 * even if we only expect to read a portion of a block.
593			 */
594			while (--nblk >= 0) {
595				(void) dmu_xuio_add(xuio,
596				    dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
597				    blksz), 0, blksz);
598			}
599		}
600	}
601
602	while (n > 0) {
603		nbytes = MIN(n, zfs_read_chunk_size -
604		    P2PHASE(uio->uio_loffset, zfs_read_chunk_size));
605
606		if (vn_has_cached_data(vp)) {
607			error = mappedread(vp, nbytes, uio);
608		} else {
609			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
610			    uio, nbytes);
611		}
612		if (error) {
613			/* convert checksum errors into IO errors */
614			if (error == ECKSUM)
615				error = SET_ERROR(EIO);
616			break;
617		}
618
619		n -= nbytes;
620	}
621out:
622	zfs_range_unlock(rl);
623
624	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
625	ZFS_EXIT(zfsvfs);
626	return (error);
627}
628
629/*
630 * Write the bytes to a file.
631 *
632 *	IN:	vp	- vnode of file to be written to.
633 *		uio	- structure supplying write location, range info,
634 *			  and data buffer.
635 *		ioflag	- FAPPEND, FSYNC, and/or FDSYNC.  FAPPEND is
636 *			  set if in append mode.
637 *		cr	- credentials of caller.
638 *		ct	- caller context (NFS/CIFS fem monitor only)
639 *
640 *	OUT:	uio	- updated offset and range.
641 *
642 *	RETURN:	0 on success, error code on failure.
643 *
644 * Timestamps:
645 *	vp - ctime|mtime updated if byte count > 0
646 */
647
648/* ARGSUSED */
649static int
650zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
651{
652	znode_t		*zp = VTOZ(vp);
653	rlim64_t	limit = uio->uio_llimit;
654	ssize_t		start_resid = uio->uio_resid;
655	ssize_t		tx_bytes;
656	uint64_t	end_size;
657	dmu_tx_t	*tx;
658	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
659	zilog_t		*zilog;
660	offset_t	woff;
661	ssize_t		n, nbytes;
662	rl_t		*rl;
663	int		max_blksz = zfsvfs->z_max_blksz;
664	int		error = 0;
665	arc_buf_t	*abuf;
666	iovec_t		*aiov = NULL;
667	xuio_t		*xuio = NULL;
668	int		i_iov = 0;
669	int		iovcnt = uio->uio_iovcnt;
670	iovec_t		*iovp = uio->uio_iov;
671	int		write_eof;
672	int		count = 0;
673	sa_bulk_attr_t	bulk[4];
674	uint64_t	mtime[2], ctime[2];
675
676	/*
677	 * Fasttrack empty write
678	 */
679	n = start_resid;
680	if (n == 0)
681		return (0);
682
683	if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
684		limit = MAXOFFSET_T;
685
686	ZFS_ENTER(zfsvfs);
687	ZFS_VERIFY_ZP(zp);
688
689	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
690	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
691	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
692	    &zp->z_size, 8);
693	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
694	    &zp->z_pflags, 8);
695
696	/*
697	 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
698	 * callers might not be able to detect properly that we are read-only,
699	 * so check it explicitly here.
700	 */
701	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
702		ZFS_EXIT(zfsvfs);
703		return (SET_ERROR(EROFS));
704	}
705
706	/*
707	 * If immutable or not appending then return EPERM
708	 */
709	if ((zp->z_pflags & (ZFS_IMMUTABLE | ZFS_READONLY)) ||
710	    ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) &&
711	    (uio->uio_loffset < zp->z_size))) {
712		ZFS_EXIT(zfsvfs);
713		return (SET_ERROR(EPERM));
714	}
715
716	zilog = zfsvfs->z_log;
717
718	/*
719	 * Validate file offset
720	 */
721	woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset;
722	if (woff < 0) {
723		ZFS_EXIT(zfsvfs);
724		return (SET_ERROR(EINVAL));
725	}
726
727	/*
728	 * Check for mandatory locks before calling zfs_range_lock()
729	 * in order to prevent a deadlock with locks set via fcntl().
730	 */
731	if (MANDMODE((mode_t)zp->z_mode) &&
732	    (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) {
733		ZFS_EXIT(zfsvfs);
734		return (error);
735	}
736
737	/*
738	 * Pre-fault the pages to ensure slow (eg NFS) pages
739	 * don't hold up txg.
740	 * Skip this if uio contains loaned arc_buf.
741	 */
742	if ((uio->uio_extflg == UIO_XUIO) &&
743	    (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY))
744		xuio = (xuio_t *)uio;
745	else
746		uio_prefaultpages(MIN(n, max_blksz), uio);
747
748	/*
749	 * If in append mode, set the io offset pointer to eof.
750	 */
751	if (ioflag & FAPPEND) {
752		/*
753		 * Obtain an appending range lock to guarantee file append
754		 * semantics.  We reset the write offset once we have the lock.
755		 */
756		rl = zfs_range_lock(zp, 0, n, RL_APPEND);
757		woff = rl->r_off;
758		if (rl->r_len == UINT64_MAX) {
759			/*
760			 * We overlocked the file because this write will cause
761			 * the file block size to increase.
762			 * Note that zp_size cannot change with this lock held.
763			 */
764			woff = zp->z_size;
765		}
766		uio->uio_loffset = woff;
767	} else {
768		/*
769		 * Note that if the file block size will change as a result of
770		 * this write, then this range lock will lock the entire file
771		 * so that we can re-write the block safely.
772		 */
773		rl = zfs_range_lock(zp, woff, n, RL_WRITER);
774	}
775
776	if (woff >= limit) {
777		zfs_range_unlock(rl);
778		ZFS_EXIT(zfsvfs);
779		return (SET_ERROR(EFBIG));
780	}
781
782	if ((woff + n) > limit || woff > (limit - n))
783		n = limit - woff;
784
785	/* Will this write extend the file length? */
786	write_eof = (woff + n > zp->z_size);
787
788	end_size = MAX(zp->z_size, woff + n);
789
790	/*
791	 * Write the file in reasonable size chunks.  Each chunk is written
792	 * in a separate transaction; this keeps the intent log records small
793	 * and allows us to do more fine-grained space accounting.
794	 */
795	while (n > 0) {
796		abuf = NULL;
797		woff = uio->uio_loffset;
798		if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
799		    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
800			if (abuf != NULL)
801				dmu_return_arcbuf(abuf);
802			error = SET_ERROR(EDQUOT);
803			break;
804		}
805
806		if (xuio && abuf == NULL) {
807			ASSERT(i_iov < iovcnt);
808			aiov = &iovp[i_iov];
809			abuf = dmu_xuio_arcbuf(xuio, i_iov);
810			dmu_xuio_clear(xuio, i_iov);
811			DTRACE_PROBE3(zfs_cp_write, int, i_iov,
812			    iovec_t *, aiov, arc_buf_t *, abuf);
813			ASSERT((aiov->iov_base == abuf->b_data) ||
814			    ((char *)aiov->iov_base - (char *)abuf->b_data +
815			    aiov->iov_len == arc_buf_size(abuf)));
816			i_iov++;
817		} else if (abuf == NULL && n >= max_blksz &&
818		    woff >= zp->z_size &&
819		    P2PHASE(woff, max_blksz) == 0 &&
820		    zp->z_blksz == max_blksz) {
821			/*
822			 * This write covers a full block.  "Borrow" a buffer
823			 * from the dmu so that we can fill it before we enter
824			 * a transaction.  This avoids the possibility of
825			 * holding up the transaction if the data copy hangs
826			 * up on a pagefault (e.g., from an NFS server mapping).
827			 */
828			size_t cbytes;
829
830			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
831			    max_blksz);
832			ASSERT(abuf != NULL);
833			ASSERT(arc_buf_size(abuf) == max_blksz);
834			if (error = uiocopy(abuf->b_data, max_blksz,
835			    UIO_WRITE, uio, &cbytes)) {
836				dmu_return_arcbuf(abuf);
837				break;
838			}
839			ASSERT(cbytes == max_blksz);
840		}
841
842		/*
843		 * Start a transaction.
844		 */
845		tx = dmu_tx_create(zfsvfs->z_os);
846		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
847		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
848		zfs_sa_upgrade_txholds(tx, zp);
849		error = dmu_tx_assign(tx, TXG_WAIT);
850		if (error) {
851			dmu_tx_abort(tx);
852			if (abuf != NULL)
853				dmu_return_arcbuf(abuf);
854			break;
855		}
856
857		/*
858		 * If zfs_range_lock() over-locked we grow the blocksize
859		 * and then reduce the lock range.  This will only happen
860		 * on the first iteration since zfs_range_reduce() will
861		 * shrink down r_len to the appropriate size.
862		 */
863		if (rl->r_len == UINT64_MAX) {
864			uint64_t new_blksz;
865
866			if (zp->z_blksz > max_blksz) {
867				/*
868				 * File's blocksize is already larger than the
869				 * "recordsize" property.  Only let it grow to
870				 * the next power of 2.
871				 */
872				ASSERT(!ISP2(zp->z_blksz));
873				new_blksz = MIN(end_size,
874				    1 << highbit64(zp->z_blksz));
875			} else {
876				new_blksz = MIN(end_size, max_blksz);
877			}
878			zfs_grow_blocksize(zp, new_blksz, tx);
879			zfs_range_reduce(rl, woff, n);
880		}
881
882		/*
883		 * XXX - should we really limit each write to z_max_blksz?
884		 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
885		 */
886		nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz));
887
888		if (abuf == NULL) {
889			tx_bytes = uio->uio_resid;
890			error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl),
891			    uio, nbytes, tx);
892			tx_bytes -= uio->uio_resid;
893		} else {
894			tx_bytes = nbytes;
895			ASSERT(xuio == NULL || tx_bytes == aiov->iov_len);
896			/*
897			 * If this is not a full block write, but we are
898			 * extending the file past EOF and this data starts
899			 * block-aligned, use assign_arcbuf().  Otherwise,
900			 * write via dmu_write().
901			 */
902			if (tx_bytes < max_blksz && (!write_eof ||
903			    aiov->iov_base != abuf->b_data)) {
904				ASSERT(xuio);
905				dmu_write(zfsvfs->z_os, zp->z_id, woff,
906				    aiov->iov_len, aiov->iov_base, tx);
907				dmu_return_arcbuf(abuf);
908				xuio_stat_wbuf_copied();
909			} else {
910				ASSERT(xuio || tx_bytes == max_blksz);
911				dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl),
912				    woff, abuf, tx);
913			}
914			ASSERT(tx_bytes <= uio->uio_resid);
915			uioskip(uio, tx_bytes);
916		}
917		if (tx_bytes && vn_has_cached_data(vp)) {
918			update_pages(vp, woff,
919			    tx_bytes, zfsvfs->z_os, zp->z_id);
920		}
921
922		/*
923		 * If we made no progress, we're done.  If we made even
924		 * partial progress, update the znode and ZIL accordingly.
925		 */
926		if (tx_bytes == 0) {
927			(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
928			    (void *)&zp->z_size, sizeof (uint64_t), tx);
929			dmu_tx_commit(tx);
930			ASSERT(error != 0);
931			break;
932		}
933
934		/*
935		 * Clear Set-UID/Set-GID bits on successful write if not
936		 * privileged and at least one of the excute bits is set.
937		 *
938		 * It would be nice to to this after all writes have
939		 * been done, but that would still expose the ISUID/ISGID
940		 * to another app after the partial write is committed.
941		 *
942		 * Note: we don't call zfs_fuid_map_id() here because
943		 * user 0 is not an ephemeral uid.
944		 */
945		mutex_enter(&zp->z_acl_lock);
946		if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) |
947		    (S_IXUSR >> 6))) != 0 &&
948		    (zp->z_mode & (S_ISUID | S_ISGID)) != 0 &&
949		    secpolicy_vnode_setid_retain(cr,
950		    (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) {
951			uint64_t newmode;
952			zp->z_mode &= ~(S_ISUID | S_ISGID);
953			newmode = zp->z_mode;
954			(void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs),
955			    (void *)&newmode, sizeof (uint64_t), tx);
956		}
957		mutex_exit(&zp->z_acl_lock);
958
959		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
960		    B_TRUE);
961
962		/*
963		 * Update the file size (zp_size) if it has changed;
964		 * account for possible concurrent updates.
965		 */
966		while ((end_size = zp->z_size) < uio->uio_loffset) {
967			(void) atomic_cas_64(&zp->z_size, end_size,
968			    uio->uio_loffset);
969			ASSERT(error == 0);
970		}
971		/*
972		 * If we are replaying and eof is non zero then force
973		 * the file size to the specified eof. Note, there's no
974		 * concurrency during replay.
975		 */
976		if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0)
977			zp->z_size = zfsvfs->z_replay_eof;
978
979		error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
980
981		zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag);
982		dmu_tx_commit(tx);
983
984		if (error != 0)
985			break;
986		ASSERT(tx_bytes == nbytes);
987		n -= nbytes;
988
989		if (!xuio && n > 0)
990			uio_prefaultpages(MIN(n, max_blksz), uio);
991	}
992
993	zfs_range_unlock(rl);
994
995	/*
996	 * If we're in replay mode, or we made no progress, return error.
997	 * Otherwise, it's at least a partial write, so it's successful.
998	 */
999	if (zfsvfs->z_replay || uio->uio_resid == start_resid) {
1000		ZFS_EXIT(zfsvfs);
1001		return (error);
1002	}
1003
1004	if (ioflag & (FSYNC | FDSYNC) ||
1005	    zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1006		zil_commit(zilog, zp->z_id);
1007
1008	ZFS_EXIT(zfsvfs);
1009	return (0);
1010}
1011
1012void
1013zfs_get_done(zgd_t *zgd, int error)
1014{
1015	znode_t *zp = zgd->zgd_private;
1016	objset_t *os = zp->z_zfsvfs->z_os;
1017
1018	if (zgd->zgd_db)
1019		dmu_buf_rele(zgd->zgd_db, zgd);
1020
1021	zfs_range_unlock(zgd->zgd_rl);
1022
1023	/*
1024	 * Release the vnode asynchronously as we currently have the
1025	 * txg stopped from syncing.
1026	 */
1027	VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1028
1029	if (error == 0 && zgd->zgd_bp)
1030		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1031
1032	kmem_free(zgd, sizeof (zgd_t));
1033}
1034
1035#ifdef DEBUG
1036static int zil_fault_io = 0;
1037#endif
1038
1039/*
1040 * Get data to generate a TX_WRITE intent log record.
1041 */
1042int
1043zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1044{
1045	zfsvfs_t *zfsvfs = arg;
1046	objset_t *os = zfsvfs->z_os;
1047	znode_t *zp;
1048	uint64_t object = lr->lr_foid;
1049	uint64_t offset = lr->lr_offset;
1050	uint64_t size = lr->lr_length;
1051	blkptr_t *bp = &lr->lr_blkptr;
1052	dmu_buf_t *db;
1053	zgd_t *zgd;
1054	int error = 0;
1055
1056	ASSERT(zio != NULL);
1057	ASSERT(size != 0);
1058
1059	/*
1060	 * Nothing to do if the file has been removed
1061	 */
1062	if (zfs_zget(zfsvfs, object, &zp) != 0)
1063		return (SET_ERROR(ENOENT));
1064	if (zp->z_unlinked) {
1065		/*
1066		 * Release the vnode asynchronously as we currently have the
1067		 * txg stopped from syncing.
1068		 */
1069		VN_RELE_ASYNC(ZTOV(zp),
1070		    dsl_pool_vnrele_taskq(dmu_objset_pool(os)));
1071		return (SET_ERROR(ENOENT));
1072	}
1073
1074	zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1075	zgd->zgd_zilog = zfsvfs->z_log;
1076	zgd->zgd_private = zp;
1077
1078	/*
1079	 * Write records come in two flavors: immediate and indirect.
1080	 * For small writes it's cheaper to store the data with the
1081	 * log record (immediate); for large writes it's cheaper to
1082	 * sync the data and get a pointer to it (indirect) so that
1083	 * we don't have to write the data twice.
1084	 */
1085	if (buf != NULL) { /* immediate write */
1086		zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER);
1087		/* test for truncation needs to be done while range locked */
1088		if (offset >= zp->z_size) {
1089			error = SET_ERROR(ENOENT);
1090		} else {
1091			error = dmu_read(os, object, offset, size, buf,
1092			    DMU_READ_NO_PREFETCH);
1093		}
1094		ASSERT(error == 0 || error == ENOENT);
1095	} else { /* indirect write */
1096		/*
1097		 * Have to lock the whole block to ensure when it's
1098		 * written out and it's checksum is being calculated
1099		 * that no one can change the data. We need to re-check
1100		 * blocksize after we get the lock in case it's changed!
1101		 */
1102		for (;;) {
1103			uint64_t blkoff;
1104			size = zp->z_blksz;
1105			blkoff = ISP2(size) ? P2PHASE(offset, size) : offset;
1106			offset -= blkoff;
1107			zgd->zgd_rl = zfs_range_lock(zp, offset, size,
1108			    RL_READER);
1109			if (zp->z_blksz == size)
1110				break;
1111			offset += blkoff;
1112			zfs_range_unlock(zgd->zgd_rl);
1113		}
1114		/* test for truncation needs to be done while range locked */
1115		if (lr->lr_offset >= zp->z_size)
1116			error = SET_ERROR(ENOENT);
1117#ifdef DEBUG
1118		if (zil_fault_io) {
1119			error = SET_ERROR(EIO);
1120			zil_fault_io = 0;
1121		}
1122#endif
1123		if (error == 0)
1124			error = dmu_buf_hold(os, object, offset, zgd, &db,
1125			    DMU_READ_NO_PREFETCH);
1126
1127		if (error == 0) {
1128			blkptr_t *obp = dmu_buf_get_blkptr(db);
1129			if (obp) {
1130				ASSERT(BP_IS_HOLE(bp));
1131				*bp = *obp;
1132			}
1133
1134			zgd->zgd_db = db;
1135			zgd->zgd_bp = bp;
1136
1137			ASSERT(db->db_offset == offset);
1138			ASSERT(db->db_size == size);
1139
1140			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1141			    zfs_get_done, zgd);
1142			ASSERT(error || lr->lr_length <= zp->z_blksz);
1143
1144			/*
1145			 * On success, we need to wait for the write I/O
1146			 * initiated by dmu_sync() to complete before we can
1147			 * release this dbuf.  We will finish everything up
1148			 * in the zfs_get_done() callback.
1149			 */
1150			if (error == 0)
1151				return (0);
1152
1153			if (error == EALREADY) {
1154				lr->lr_common.lrc_txtype = TX_WRITE2;
1155				error = 0;
1156			}
1157		}
1158	}
1159
1160	zfs_get_done(zgd, error);
1161
1162	return (error);
1163}
1164
1165/*ARGSUSED*/
1166static int
1167zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
1168    caller_context_t *ct)
1169{
1170	znode_t *zp = VTOZ(vp);
1171	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1172	int error;
1173
1174	ZFS_ENTER(zfsvfs);
1175	ZFS_VERIFY_ZP(zp);
1176
1177	if (flag & V_ACE_MASK)
1178		error = zfs_zaccess(zp, mode, flag, B_FALSE, cr);
1179	else
1180		error = zfs_zaccess_rwx(zp, mode, flag, cr);
1181
1182	ZFS_EXIT(zfsvfs);
1183	return (error);
1184}
1185
1186/*
1187 * If vnode is for a device return a specfs vnode instead.
1188 */
1189static int
1190specvp_check(vnode_t **vpp, cred_t *cr)
1191{
1192	int error = 0;
1193
1194	if (IS_DEVVP(*vpp)) {
1195		struct vnode *svp;
1196
1197		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
1198		VN_RELE(*vpp);
1199		if (svp == NULL)
1200			error = SET_ERROR(ENOSYS);
1201		*vpp = svp;
1202	}
1203	return (error);
1204}
1205
1206
1207/*
1208 * Lookup an entry in a directory, or an extended attribute directory.
1209 * If it exists, return a held vnode reference for it.
1210 *
1211 *	IN:	dvp	- vnode of directory to search.
1212 *		nm	- name of entry to lookup.
1213 *		pnp	- full pathname to lookup [UNUSED].
1214 *		flags	- LOOKUP_XATTR set if looking for an attribute.
1215 *		rdir	- root directory vnode [UNUSED].
1216 *		cr	- credentials of caller.
1217 *		ct	- caller context
1218 *		direntflags - directory lookup flags
1219 *		realpnp - returned pathname.
1220 *
1221 *	OUT:	vpp	- vnode of located entry, NULL if not found.
1222 *
1223 *	RETURN:	0 on success, error code on failure.
1224 *
1225 * Timestamps:
1226 *	NA
1227 */
1228/* ARGSUSED */
1229static int
1230zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1231    int flags, vnode_t *rdir, cred_t *cr,  caller_context_t *ct,
1232    int *direntflags, pathname_t *realpnp)
1233{
1234	znode_t *zdp = VTOZ(dvp);
1235	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
1236	int	error = 0;
1237
1238	/* fast path */
1239	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
1240
1241		if (dvp->v_type != VDIR) {
1242			return (SET_ERROR(ENOTDIR));
1243		} else if (zdp->z_sa_hdl == NULL) {
1244			return (SET_ERROR(EIO));
1245		}
1246
1247		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
1248			error = zfs_fastaccesschk_execute(zdp, cr);
1249			if (!error) {
1250				*vpp = dvp;
1251				VN_HOLD(*vpp);
1252				return (0);
1253			}
1254			return (error);
1255		} else {
1256			vnode_t *tvp = dnlc_lookup(dvp, nm);
1257
1258			if (tvp) {
1259				error = zfs_fastaccesschk_execute(zdp, cr);
1260				if (error) {
1261					VN_RELE(tvp);
1262					return (error);
1263				}
1264				if (tvp == DNLC_NO_VNODE) {
1265					VN_RELE(tvp);
1266					return (SET_ERROR(ENOENT));
1267				} else {
1268					*vpp = tvp;
1269					return (specvp_check(vpp, cr));
1270				}
1271			}
1272		}
1273	}
1274
1275	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
1276
1277	ZFS_ENTER(zfsvfs);
1278	ZFS_VERIFY_ZP(zdp);
1279
1280	*vpp = NULL;
1281
1282	if (flags & LOOKUP_XATTR) {
1283		/*
1284		 * If the xattr property is off, refuse the lookup request.
1285		 */
1286		if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) {
1287			ZFS_EXIT(zfsvfs);
1288			return (SET_ERROR(EINVAL));
1289		}
1290
1291		/*
1292		 * We don't allow recursive attributes..
1293		 * Maybe someday we will.
1294		 */
1295		if (zdp->z_pflags & ZFS_XATTR) {
1296			ZFS_EXIT(zfsvfs);
1297			return (SET_ERROR(EINVAL));
1298		}
1299
1300		if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) {
1301			ZFS_EXIT(zfsvfs);
1302			return (error);
1303		}
1304
1305		/*
1306		 * Do we have permission to get into attribute directory?
1307		 */
1308
1309		if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0,
1310		    B_FALSE, cr)) {
1311			VN_RELE(*vpp);
1312			*vpp = NULL;
1313		}
1314
1315		ZFS_EXIT(zfsvfs);
1316		return (error);
1317	}
1318
1319	if (dvp->v_type != VDIR) {
1320		ZFS_EXIT(zfsvfs);
1321		return (SET_ERROR(ENOTDIR));
1322	}
1323
1324	/*
1325	 * Check accessibility of directory.
1326	 */
1327
1328	if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) {
1329		ZFS_EXIT(zfsvfs);
1330		return (error);
1331	}
1332
1333	if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm),
1334	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1335		ZFS_EXIT(zfsvfs);
1336		return (SET_ERROR(EILSEQ));
1337	}
1338
1339	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
1340	if (error == 0)
1341		error = specvp_check(vpp, cr);
1342
1343	ZFS_EXIT(zfsvfs);
1344	return (error);
1345}
1346
1347/*
1348 * Attempt to create a new entry in a directory.  If the entry
1349 * already exists, truncate the file if permissible, else return
1350 * an error.  Return the vp of the created or trunc'd file.
1351 *
1352 *	IN:	dvp	- vnode of directory to put new file entry in.
1353 *		name	- name of new file entry.
1354 *		vap	- attributes of new file.
1355 *		excl	- flag indicating exclusive or non-exclusive mode.
1356 *		mode	- mode to open file with.
1357 *		cr	- credentials of caller.
1358 *		flag	- large file flag [UNUSED].
1359 *		ct	- caller context
1360 *		vsecp	- ACL to be set
1361 *
1362 *	OUT:	vpp	- vnode of created or trunc'd entry.
1363 *
1364 *	RETURN:	0 on success, error code on failure.
1365 *
1366 * Timestamps:
1367 *	dvp - ctime|mtime updated if new entry created
1368 *	 vp - ctime|mtime always, atime if new
1369 */
1370
1371/* ARGSUSED */
1372static int
1373zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl,
1374    int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
1375    vsecattr_t *vsecp)
1376{
1377	znode_t		*zp, *dzp = VTOZ(dvp);
1378	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1379	zilog_t		*zilog;
1380	objset_t	*os;
1381	zfs_dirlock_t	*dl;
1382	dmu_tx_t	*tx;
1383	int		error;
1384	ksid_t		*ksid;
1385	uid_t		uid;
1386	gid_t		gid = crgetgid(cr);
1387	zfs_acl_ids_t   acl_ids;
1388	boolean_t	fuid_dirtied;
1389	boolean_t	have_acl = B_FALSE;
1390	boolean_t	waited = B_FALSE;
1391
1392	/*
1393	 * If we have an ephemeral id, ACL, or XVATTR then
1394	 * make sure file system is at proper version
1395	 */
1396
1397	ksid = crgetsid(cr, KSID_OWNER);
1398	if (ksid)
1399		uid = ksid_getid(ksid);
1400	else
1401		uid = crgetuid(cr);
1402
1403	if (zfsvfs->z_use_fuids == B_FALSE &&
1404	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1405	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1406		return (SET_ERROR(EINVAL));
1407
1408	ZFS_ENTER(zfsvfs);
1409	ZFS_VERIFY_ZP(dzp);
1410	os = zfsvfs->z_os;
1411	zilog = zfsvfs->z_log;
1412
1413	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
1414	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1415		ZFS_EXIT(zfsvfs);
1416		return (SET_ERROR(EILSEQ));
1417	}
1418
1419	if (vap->va_mask & AT_XVATTR) {
1420		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1421		    crgetuid(cr), cr, vap->va_type)) != 0) {
1422			ZFS_EXIT(zfsvfs);
1423			return (error);
1424		}
1425	}
1426top:
1427	*vpp = NULL;
1428
1429	if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr))
1430		vap->va_mode &= ~VSVTX;
1431
1432	if (*name == '\0') {
1433		/*
1434		 * Null component name refers to the directory itself.
1435		 */
1436		VN_HOLD(dvp);
1437		zp = dzp;
1438		dl = NULL;
1439		error = 0;
1440	} else {
1441		/* possible VN_HOLD(zp) */
1442		int zflg = 0;
1443
1444		if (flag & FIGNORECASE)
1445			zflg |= ZCILOOK;
1446
1447		error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1448		    NULL, NULL);
1449		if (error) {
1450			if (have_acl)
1451				zfs_acl_ids_free(&acl_ids);
1452			if (strcmp(name, "..") == 0)
1453				error = SET_ERROR(EISDIR);
1454			ZFS_EXIT(zfsvfs);
1455			return (error);
1456		}
1457	}
1458
1459	if (zp == NULL) {
1460		uint64_t txtype;
1461
1462		/*
1463		 * Create a new file object and update the directory
1464		 * to reference it.
1465		 */
1466		if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
1467			if (have_acl)
1468				zfs_acl_ids_free(&acl_ids);
1469			goto out;
1470		}
1471
1472		/*
1473		 * We only support the creation of regular files in
1474		 * extended attribute directories.
1475		 */
1476
1477		if ((dzp->z_pflags & ZFS_XATTR) &&
1478		    (vap->va_type != VREG)) {
1479			if (have_acl)
1480				zfs_acl_ids_free(&acl_ids);
1481			error = SET_ERROR(EINVAL);
1482			goto out;
1483		}
1484
1485		if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap,
1486		    cr, vsecp, &acl_ids)) != 0)
1487			goto out;
1488		have_acl = B_TRUE;
1489
1490		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1491			zfs_acl_ids_free(&acl_ids);
1492			error = SET_ERROR(EDQUOT);
1493			goto out;
1494		}
1495
1496		tx = dmu_tx_create(os);
1497
1498		dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1499		    ZFS_SA_BASE_ATTR_SIZE);
1500
1501		fuid_dirtied = zfsvfs->z_fuid_dirty;
1502		if (fuid_dirtied)
1503			zfs_fuid_txhold(zfsvfs, tx);
1504		dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
1505		dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
1506		if (!zfsvfs->z_use_sa &&
1507		    acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1508			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
1509			    0, acl_ids.z_aclp->z_acl_bytes);
1510		}
1511		error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1512		if (error) {
1513			zfs_dirent_unlock(dl);
1514			if (error == ERESTART) {
1515				waited = B_TRUE;
1516				dmu_tx_wait(tx);
1517				dmu_tx_abort(tx);
1518				goto top;
1519			}
1520			zfs_acl_ids_free(&acl_ids);
1521			dmu_tx_abort(tx);
1522			ZFS_EXIT(zfsvfs);
1523			return (error);
1524		}
1525		zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
1526
1527		if (fuid_dirtied)
1528			zfs_fuid_sync(zfsvfs, tx);
1529
1530		(void) zfs_link_create(dl, zp, tx, ZNEW);
1531		txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap);
1532		if (flag & FIGNORECASE)
1533			txtype |= TX_CI;
1534		zfs_log_create(zilog, tx, txtype, dzp, zp, name,
1535		    vsecp, acl_ids.z_fuidp, vap);
1536		zfs_acl_ids_free(&acl_ids);
1537		dmu_tx_commit(tx);
1538	} else {
1539		int aflags = (flag & FAPPEND) ? V_APPEND : 0;
1540
1541		if (have_acl)
1542			zfs_acl_ids_free(&acl_ids);
1543		have_acl = B_FALSE;
1544
1545		/*
1546		 * A directory entry already exists for this name.
1547		 */
1548		/*
1549		 * Can't truncate an existing file if in exclusive mode.
1550		 */
1551		if (excl == EXCL) {
1552			error = SET_ERROR(EEXIST);
1553			goto out;
1554		}
1555		/*
1556		 * Can't open a directory for writing.
1557		 */
1558		if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) {
1559			error = SET_ERROR(EISDIR);
1560			goto out;
1561		}
1562		/*
1563		 * Verify requested access to file.
1564		 */
1565		if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) {
1566			goto out;
1567		}
1568
1569		mutex_enter(&dzp->z_lock);
1570		dzp->z_seq++;
1571		mutex_exit(&dzp->z_lock);
1572
1573		/*
1574		 * Truncate regular files if requested.
1575		 */
1576		if ((ZTOV(zp)->v_type == VREG) &&
1577		    (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) {
1578			/* we can't hold any locks when calling zfs_freesp() */
1579			zfs_dirent_unlock(dl);
1580			dl = NULL;
1581			error = zfs_freesp(zp, 0, 0, mode, TRUE);
1582			if (error == 0) {
1583				vnevent_create(ZTOV(zp), ct);
1584			}
1585		}
1586	}
1587out:
1588
1589	if (dl)
1590		zfs_dirent_unlock(dl);
1591
1592	if (error) {
1593		if (zp)
1594			VN_RELE(ZTOV(zp));
1595	} else {
1596		*vpp = ZTOV(zp);
1597		error = specvp_check(vpp, cr);
1598	}
1599
1600	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1601		zil_commit(zilog, 0);
1602
1603	ZFS_EXIT(zfsvfs);
1604	return (error);
1605}
1606
1607/*
1608 * Remove an entry from a directory.
1609 *
1610 *	IN:	dvp	- vnode of directory to remove entry from.
1611 *		name	- name of entry to remove.
1612 *		cr	- credentials of caller.
1613 *		ct	- caller context
1614 *		flags	- case flags
1615 *
1616 *	RETURN:	0 on success, error code on failure.
1617 *
1618 * Timestamps:
1619 *	dvp - ctime|mtime
1620 *	 vp - ctime (if nlink > 0)
1621 */
1622
1623uint64_t null_xattr = 0;
1624
1625/*ARGSUSED*/
1626static int
1627zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct,
1628    int flags)
1629{
1630	znode_t		*zp, *dzp = VTOZ(dvp);
1631	znode_t		*xzp;
1632	vnode_t		*vp;
1633	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1634	zilog_t		*zilog;
1635	uint64_t	acl_obj, xattr_obj;
1636	uint64_t	xattr_obj_unlinked = 0;
1637	uint64_t	obj = 0;
1638	zfs_dirlock_t	*dl;
1639	dmu_tx_t	*tx;
1640	boolean_t	may_delete_now, delete_now = FALSE;
1641	boolean_t	unlinked, toobig = FALSE;
1642	uint64_t	txtype;
1643	pathname_t	*realnmp = NULL;
1644	pathname_t	realnm;
1645	int		error;
1646	int		zflg = ZEXISTS;
1647	boolean_t	waited = B_FALSE;
1648
1649	ZFS_ENTER(zfsvfs);
1650	ZFS_VERIFY_ZP(dzp);
1651	zilog = zfsvfs->z_log;
1652
1653	if (flags & FIGNORECASE) {
1654		zflg |= ZCILOOK;
1655		pn_alloc(&realnm);
1656		realnmp = &realnm;
1657	}
1658
1659top:
1660	xattr_obj = 0;
1661	xzp = NULL;
1662	/*
1663	 * Attempt to lock directory; fail if entry doesn't exist.
1664	 */
1665	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
1666	    NULL, realnmp)) {
1667		if (realnmp)
1668			pn_free(realnmp);
1669		ZFS_EXIT(zfsvfs);
1670		return (error);
1671	}
1672
1673	vp = ZTOV(zp);
1674
1675	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
1676		goto out;
1677	}
1678
1679	/*
1680	 * Need to use rmdir for removing directories.
1681	 */
1682	if (vp->v_type == VDIR) {
1683		error = SET_ERROR(EPERM);
1684		goto out;
1685	}
1686
1687	vnevent_remove(vp, dvp, name, ct);
1688
1689	if (realnmp)
1690		dnlc_remove(dvp, realnmp->pn_buf);
1691	else
1692		dnlc_remove(dvp, name);
1693
1694	mutex_enter(&vp->v_lock);
1695	may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp);
1696	mutex_exit(&vp->v_lock);
1697
1698	/*
1699	 * We may delete the znode now, or we may put it in the unlinked set;
1700	 * it depends on whether we're the last link, and on whether there are
1701	 * other holds on the vnode.  So we dmu_tx_hold() the right things to
1702	 * allow for either case.
1703	 */
1704	obj = zp->z_id;
1705	tx = dmu_tx_create(zfsvfs->z_os);
1706	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
1707	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
1708	zfs_sa_upgrade_txholds(tx, zp);
1709	zfs_sa_upgrade_txholds(tx, dzp);
1710	if (may_delete_now) {
1711		toobig =
1712		    zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT;
1713		/* if the file is too big, only hold_free a token amount */
1714		dmu_tx_hold_free(tx, zp->z_id, 0,
1715		    (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END));
1716	}
1717
1718	/* are there any extended attributes? */
1719	error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1720	    &xattr_obj, sizeof (xattr_obj));
1721	if (error == 0 && xattr_obj) {
1722		error = zfs_zget(zfsvfs, xattr_obj, &xzp);
1723		ASSERT0(error);
1724		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1725		dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
1726	}
1727
1728	mutex_enter(&zp->z_lock);
1729	if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now)
1730		dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
1731	mutex_exit(&zp->z_lock);
1732
1733	/* charge as an update -- would be nice not to charge at all */
1734	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
1735
1736	/*
1737	 * Mark this transaction as typically resulting in a net free of
1738	 * space, unless object removal will be delayed indefinitely
1739	 * (due to active holds on the vnode due to the file being open).
1740	 */
1741	if (may_delete_now)
1742		dmu_tx_mark_netfree(tx);
1743
1744	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1745	if (error) {
1746		zfs_dirent_unlock(dl);
1747		VN_RELE(vp);
1748		if (xzp)
1749			VN_RELE(ZTOV(xzp));
1750		if (error == ERESTART) {
1751			waited = B_TRUE;
1752			dmu_tx_wait(tx);
1753			dmu_tx_abort(tx);
1754			goto top;
1755		}
1756		if (realnmp)
1757			pn_free(realnmp);
1758		dmu_tx_abort(tx);
1759		ZFS_EXIT(zfsvfs);
1760		return (error);
1761	}
1762
1763	/*
1764	 * Remove the directory entry.
1765	 */
1766	error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked);
1767
1768	if (error) {
1769		dmu_tx_commit(tx);
1770		goto out;
1771	}
1772
1773	if (unlinked) {
1774		/*
1775		 * Hold z_lock so that we can make sure that the ACL obj
1776		 * hasn't changed.  Could have been deleted due to
1777		 * zfs_sa_upgrade().
1778		 */
1779		mutex_enter(&zp->z_lock);
1780		mutex_enter(&vp->v_lock);
1781		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
1782		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
1783		delete_now = may_delete_now && !toobig &&
1784		    vp->v_count == 1 && !vn_has_cached_data(vp) &&
1785		    xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) ==
1786		    acl_obj;
1787		mutex_exit(&vp->v_lock);
1788	}
1789
1790	if (delete_now) {
1791		if (xattr_obj_unlinked) {
1792			ASSERT3U(xzp->z_links, ==, 2);
1793			mutex_enter(&xzp->z_lock);
1794			xzp->z_unlinked = 1;
1795			xzp->z_links = 0;
1796			error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
1797			    &xzp->z_links, sizeof (xzp->z_links), tx);
1798			ASSERT3U(error,  ==,  0);
1799			mutex_exit(&xzp->z_lock);
1800			zfs_unlinked_add(xzp, tx);
1801
1802			if (zp->z_is_sa)
1803				error = sa_remove(zp->z_sa_hdl,
1804				    SA_ZPL_XATTR(zfsvfs), tx);
1805			else
1806				error = sa_update(zp->z_sa_hdl,
1807				    SA_ZPL_XATTR(zfsvfs), &null_xattr,
1808				    sizeof (uint64_t), tx);
1809			ASSERT0(error);
1810		}
1811		mutex_enter(&vp->v_lock);
1812		vp->v_count--;
1813		ASSERT0(vp->v_count);
1814		mutex_exit(&vp->v_lock);
1815		mutex_exit(&zp->z_lock);
1816		zfs_znode_delete(zp, tx);
1817	} else if (unlinked) {
1818		mutex_exit(&zp->z_lock);
1819		zfs_unlinked_add(zp, tx);
1820	}
1821
1822	txtype = TX_REMOVE;
1823	if (flags & FIGNORECASE)
1824		txtype |= TX_CI;
1825	zfs_log_remove(zilog, tx, txtype, dzp, name, obj);
1826
1827	dmu_tx_commit(tx);
1828out:
1829	if (realnmp)
1830		pn_free(realnmp);
1831
1832	zfs_dirent_unlock(dl);
1833
1834	if (!delete_now)
1835		VN_RELE(vp);
1836	if (xzp)
1837		VN_RELE(ZTOV(xzp));
1838
1839	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
1840		zil_commit(zilog, 0);
1841
1842	ZFS_EXIT(zfsvfs);
1843	return (error);
1844}
1845
1846/*
1847 * Create a new directory and insert it into dvp using the name
1848 * provided.  Return a pointer to the inserted directory.
1849 *
1850 *	IN:	dvp	- vnode of directory to add subdir to.
1851 *		dirname	- name of new directory.
1852 *		vap	- attributes of new directory.
1853 *		cr	- credentials of caller.
1854 *		ct	- caller context
1855 *		flags	- case flags
1856 *		vsecp	- ACL to be set
1857 *
1858 *	OUT:	vpp	- vnode of created directory.
1859 *
1860 *	RETURN:	0 on success, error code on failure.
1861 *
1862 * Timestamps:
1863 *	dvp - ctime|mtime updated
1864 *	 vp - ctime|mtime|atime updated
1865 */
1866/*ARGSUSED*/
1867static int
1868zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr,
1869    caller_context_t *ct, int flags, vsecattr_t *vsecp)
1870{
1871	znode_t		*zp, *dzp = VTOZ(dvp);
1872	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1873	zilog_t		*zilog;
1874	zfs_dirlock_t	*dl;
1875	uint64_t	txtype;
1876	dmu_tx_t	*tx;
1877	int		error;
1878	int		zf = ZNEW;
1879	ksid_t		*ksid;
1880	uid_t		uid;
1881	gid_t		gid = crgetgid(cr);
1882	zfs_acl_ids_t   acl_ids;
1883	boolean_t	fuid_dirtied;
1884	boolean_t	waited = B_FALSE;
1885
1886	ASSERT(vap->va_type == VDIR);
1887
1888	/*
1889	 * If we have an ephemeral id, ACL, or XVATTR then
1890	 * make sure file system is at proper version
1891	 */
1892
1893	ksid = crgetsid(cr, KSID_OWNER);
1894	if (ksid)
1895		uid = ksid_getid(ksid);
1896	else
1897		uid = crgetuid(cr);
1898	if (zfsvfs->z_use_fuids == B_FALSE &&
1899	    (vsecp || (vap->va_mask & AT_XVATTR) ||
1900	    IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid)))
1901		return (SET_ERROR(EINVAL));
1902
1903	ZFS_ENTER(zfsvfs);
1904	ZFS_VERIFY_ZP(dzp);
1905	zilog = zfsvfs->z_log;
1906
1907	if (dzp->z_pflags & ZFS_XATTR) {
1908		ZFS_EXIT(zfsvfs);
1909		return (SET_ERROR(EINVAL));
1910	}
1911
1912	if (zfsvfs->z_utf8 && u8_validate(dirname,
1913	    strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
1914		ZFS_EXIT(zfsvfs);
1915		return (SET_ERROR(EILSEQ));
1916	}
1917	if (flags & FIGNORECASE)
1918		zf |= ZCILOOK;
1919
1920	if (vap->va_mask & AT_XVATTR) {
1921		if ((error = secpolicy_xvattr((xvattr_t *)vap,
1922		    crgetuid(cr), cr, vap->va_type)) != 0) {
1923			ZFS_EXIT(zfsvfs);
1924			return (error);
1925		}
1926	}
1927
1928	if ((error = zfs_acl_ids_create(dzp, 0, vap, cr,
1929	    vsecp, &acl_ids)) != 0) {
1930		ZFS_EXIT(zfsvfs);
1931		return (error);
1932	}
1933	/*
1934	 * First make sure the new directory doesn't exist.
1935	 *
1936	 * Existence is checked first to make sure we don't return
1937	 * EACCES instead of EEXIST which can cause some applications
1938	 * to fail.
1939	 */
1940top:
1941	*vpp = NULL;
1942
1943	if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf,
1944	    NULL, NULL)) {
1945		zfs_acl_ids_free(&acl_ids);
1946		ZFS_EXIT(zfsvfs);
1947		return (error);
1948	}
1949
1950	if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) {
1951		zfs_acl_ids_free(&acl_ids);
1952		zfs_dirent_unlock(dl);
1953		ZFS_EXIT(zfsvfs);
1954		return (error);
1955	}
1956
1957	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1958		zfs_acl_ids_free(&acl_ids);
1959		zfs_dirent_unlock(dl);
1960		ZFS_EXIT(zfsvfs);
1961		return (SET_ERROR(EDQUOT));
1962	}
1963
1964	/*
1965	 * Add a new entry to the directory.
1966	 */
1967	tx = dmu_tx_create(zfsvfs->z_os);
1968	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname);
1969	dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1970	fuid_dirtied = zfsvfs->z_fuid_dirty;
1971	if (fuid_dirtied)
1972		zfs_fuid_txhold(zfsvfs, tx);
1973	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1974		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1975		    acl_ids.z_aclp->z_acl_bytes);
1976	}
1977
1978	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1979	    ZFS_SA_BASE_ATTR_SIZE);
1980
1981	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
1982	if (error) {
1983		zfs_dirent_unlock(dl);
1984		if (error == ERESTART) {
1985			waited = B_TRUE;
1986			dmu_tx_wait(tx);
1987			dmu_tx_abort(tx);
1988			goto top;
1989		}
1990		zfs_acl_ids_free(&acl_ids);
1991		dmu_tx_abort(tx);
1992		ZFS_EXIT(zfsvfs);
1993		return (error);
1994	}
1995
1996	/*
1997	 * Create new node.
1998	 */
1999	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
2000
2001	if (fuid_dirtied)
2002		zfs_fuid_sync(zfsvfs, tx);
2003
2004	/*
2005	 * Now put new name in parent dir.
2006	 */
2007	(void) zfs_link_create(dl, zp, tx, ZNEW);
2008
2009	*vpp = ZTOV(zp);
2010
2011	txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap);
2012	if (flags & FIGNORECASE)
2013		txtype |= TX_CI;
2014	zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp,
2015	    acl_ids.z_fuidp, vap);
2016
2017	zfs_acl_ids_free(&acl_ids);
2018
2019	dmu_tx_commit(tx);
2020
2021	zfs_dirent_unlock(dl);
2022
2023	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2024		zil_commit(zilog, 0);
2025
2026	ZFS_EXIT(zfsvfs);
2027	return (0);
2028}
2029
2030/*
2031 * Remove a directory subdir entry.  If the current working
2032 * directory is the same as the subdir to be removed, the
2033 * remove will fail.
2034 *
2035 *	IN:	dvp	- vnode of directory to remove from.
2036 *		name	- name of directory to be removed.
2037 *		cwd	- vnode of current working directory.
2038 *		cr	- credentials of caller.
2039 *		ct	- caller context
2040 *		flags	- case flags
2041 *
2042 *	RETURN:	0 on success, error code on failure.
2043 *
2044 * Timestamps:
2045 *	dvp - ctime|mtime updated
2046 */
2047/*ARGSUSED*/
2048static int
2049zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
2050    caller_context_t *ct, int flags)
2051{
2052	znode_t		*dzp = VTOZ(dvp);
2053	znode_t		*zp;
2054	vnode_t		*vp;
2055	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
2056	zilog_t		*zilog;
2057	zfs_dirlock_t	*dl;
2058	dmu_tx_t	*tx;
2059	int		error;
2060	int		zflg = ZEXISTS;
2061	boolean_t	waited = B_FALSE;
2062
2063	ZFS_ENTER(zfsvfs);
2064	ZFS_VERIFY_ZP(dzp);
2065	zilog = zfsvfs->z_log;
2066
2067	if (flags & FIGNORECASE)
2068		zflg |= ZCILOOK;
2069top:
2070	zp = NULL;
2071
2072	/*
2073	 * Attempt to lock directory; fail if entry doesn't exist.
2074	 */
2075	if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg,
2076	    NULL, NULL)) {
2077		ZFS_EXIT(zfsvfs);
2078		return (error);
2079	}
2080
2081	vp = ZTOV(zp);
2082
2083	if (error = zfs_zaccess_delete(dzp, zp, cr)) {
2084		goto out;
2085	}
2086
2087	if (vp->v_type != VDIR) {
2088		error = SET_ERROR(ENOTDIR);
2089		goto out;
2090	}
2091
2092	if (vp == cwd) {
2093		error = SET_ERROR(EINVAL);
2094		goto out;
2095	}
2096
2097	vnevent_rmdir(vp, dvp, name, ct);
2098
2099	/*
2100	 * Grab a lock on the directory to make sure that noone is
2101	 * trying to add (or lookup) entries while we are removing it.
2102	 */
2103	rw_enter(&zp->z_name_lock, RW_WRITER);
2104
2105	/*
2106	 * Grab a lock on the parent pointer to make sure we play well
2107	 * with the treewalk and directory rename code.
2108	 */
2109	rw_enter(&zp->z_parent_lock, RW_WRITER);
2110
2111	tx = dmu_tx_create(zfsvfs->z_os);
2112	dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name);
2113	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
2114	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
2115	zfs_sa_upgrade_txholds(tx, zp);
2116	zfs_sa_upgrade_txholds(tx, dzp);
2117	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
2118	if (error) {
2119		rw_exit(&zp->z_parent_lock);
2120		rw_exit(&zp->z_name_lock);
2121		zfs_dirent_unlock(dl);
2122		VN_RELE(vp);
2123		if (error == ERESTART) {
2124			waited = B_TRUE;
2125			dmu_tx_wait(tx);
2126			dmu_tx_abort(tx);
2127			goto top;
2128		}
2129		dmu_tx_abort(tx);
2130		ZFS_EXIT(zfsvfs);
2131		return (error);
2132	}
2133
2134	error = zfs_link_destroy(dl, zp, tx, zflg, NULL);
2135
2136	if (error == 0) {
2137		uint64_t txtype = TX_RMDIR;
2138		if (flags & FIGNORECASE)
2139			txtype |= TX_CI;
2140		zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT);
2141	}
2142
2143	dmu_tx_commit(tx);
2144
2145	rw_exit(&zp->z_parent_lock);
2146	rw_exit(&zp->z_name_lock);
2147out:
2148	zfs_dirent_unlock(dl);
2149
2150	VN_RELE(vp);
2151
2152	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
2153		zil_commit(zilog, 0);
2154
2155	ZFS_EXIT(zfsvfs);
2156	return (error);
2157}
2158
2159/*
2160 * Read as many directory entries as will fit into the provided
2161 * buffer from the given directory cursor position (specified in
2162 * the uio structure).
2163 *
2164 *	IN:	vp	- vnode of directory to read.
2165 *		uio	- structure supplying read location, range info,
2166 *			  and return buffer.
2167 *		cr	- credentials of caller.
2168 *		ct	- caller context
2169 *		flags	- case flags
2170 *
2171 *	OUT:	uio	- updated offset and range, buffer filled.
2172 *		eofp	- set to true if end-of-file detected.
2173 *
2174 *	RETURN:	0 on success, error code on failure.
2175 *
2176 * Timestamps:
2177 *	vp - atime updated
2178 *
2179 * Note that the low 4 bits of the cookie returned by zap is always zero.
2180 * This allows us to use the low range for "special" directory entries:
2181 * We use 0 for '.', and 1 for '..'.  If this is the root of the filesystem,
2182 * we use the offset 2 for the '.zfs' directory.
2183 */
2184/* ARGSUSED */
2185static int
2186zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp,
2187    caller_context_t *ct, int flags)
2188{
2189	znode_t		*zp = VTOZ(vp);
2190	iovec_t		*iovp;
2191	edirent_t	*eodp;
2192	dirent64_t	*odp;
2193	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2194	objset_t	*os;
2195	caddr_t		outbuf;
2196	size_t		bufsize;
2197	zap_cursor_t	zc;
2198	zap_attribute_t	zap;
2199	uint_t		bytes_wanted;
2200	uint64_t	offset; /* must be unsigned; checks for < 1 */
2201	uint64_t	parent;
2202	int		local_eof;
2203	int		outcount;
2204	int		error;
2205	uint8_t		prefetch;
2206	boolean_t	check_sysattrs;
2207
2208	ZFS_ENTER(zfsvfs);
2209	ZFS_VERIFY_ZP(zp);
2210
2211	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
2212	    &parent, sizeof (parent))) != 0) {
2213		ZFS_EXIT(zfsvfs);
2214		return (error);
2215	}
2216
2217	/*
2218	 * If we are not given an eof variable,
2219	 * use a local one.
2220	 */
2221	if (eofp == NULL)
2222		eofp = &local_eof;
2223
2224	/*
2225	 * Check for valid iov_len.
2226	 */
2227	if (uio->uio_iov->iov_len <= 0) {
2228		ZFS_EXIT(zfsvfs);
2229		return (SET_ERROR(EINVAL));
2230	}
2231
2232	/*
2233	 * Quit if directory has been removed (posix)
2234	 */
2235	if ((*eofp = zp->z_unlinked) != 0) {
2236		ZFS_EXIT(zfsvfs);
2237		return (0);
2238	}
2239
2240	error = 0;
2241	os = zfsvfs->z_os;
2242	offset = uio->uio_loffset;
2243	prefetch = zp->z_zn_prefetch;
2244
2245	/*
2246	 * Initialize the iterator cursor.
2247	 */
2248	if (offset <= 3) {
2249		/*
2250		 * Start iteration from the beginning of the directory.
2251		 */
2252		zap_cursor_init(&zc, os, zp->z_id);
2253	} else {
2254		/*
2255		 * The offset is a serialized cursor.
2256		 */
2257		zap_cursor_init_serialized(&zc, os, zp->z_id, offset);
2258	}
2259
2260	/*
2261	 * Get space to change directory entries into fs independent format.
2262	 */
2263	iovp = uio->uio_iov;
2264	bytes_wanted = iovp->iov_len;
2265	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) {
2266		bufsize = bytes_wanted;
2267		outbuf = kmem_alloc(bufsize, KM_SLEEP);
2268		odp = (struct dirent64 *)outbuf;
2269	} else {
2270		bufsize = bytes_wanted;
2271		outbuf = NULL;
2272		odp = (struct dirent64 *)iovp->iov_base;
2273	}
2274	eodp = (struct edirent *)odp;
2275
2276	/*
2277	 * If this VFS supports the system attribute view interface; and
2278	 * we're looking at an extended attribute directory; and we care
2279	 * about normalization conflicts on this vfs; then we must check
2280	 * for normalization conflicts with the sysattr name space.
2281	 */
2282	check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
2283	    (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm &&
2284	    (flags & V_RDDIR_ENTFLAGS);
2285
2286	/*
2287	 * Transform to file-system independent format
2288	 */
2289	outcount = 0;
2290	while (outcount < bytes_wanted) {
2291		ino64_t objnum;
2292		ushort_t reclen;
2293		off64_t *next = NULL;
2294
2295		/*
2296		 * Special case `.', `..', and `.zfs'.
2297		 */
2298		if (offset == 0) {
2299			(void) strcpy(zap.za_name, ".");
2300			zap.za_normalization_conflict = 0;
2301			objnum = zp->z_id;
2302		} else if (offset == 1) {
2303			(void) strcpy(zap.za_name, "..");
2304			zap.za_normalization_conflict = 0;
2305			objnum = parent;
2306		} else if (offset == 2 && zfs_show_ctldir(zp)) {
2307			(void) strcpy(zap.za_name, ZFS_CTLDIR_NAME);
2308			zap.za_normalization_conflict = 0;
2309			objnum = ZFSCTL_INO_ROOT;
2310		} else {
2311			/*
2312			 * Grab next entry.
2313			 */
2314			if (error = zap_cursor_retrieve(&zc, &zap)) {
2315				if ((*eofp = (error == ENOENT)) != 0)
2316					break;
2317				else
2318					goto update;
2319			}
2320
2321			if (zap.za_integer_length != 8 ||
2322			    zap.za_num_integers != 1) {
2323				cmn_err(CE_WARN, "zap_readdir: bad directory "
2324				    "entry, obj = %lld, offset = %lld\n",
2325				    (u_longlong_t)zp->z_id,
2326				    (u_longlong_t)offset);
2327				error = SET_ERROR(ENXIO);
2328				goto update;
2329			}
2330
2331			objnum = ZFS_DIRENT_OBJ(zap.za_first_integer);
2332			/*
2333			 * MacOS X can extract the object type here such as:
2334			 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2335			 */
2336
2337			if (check_sysattrs && !zap.za_normalization_conflict) {
2338				zap.za_normalization_conflict =
2339				    xattr_sysattr_casechk(zap.za_name);
2340			}
2341		}
2342
2343		if (flags & V_RDDIR_ACCFILTER) {
2344			/*
2345			 * If we have no access at all, don't include
2346			 * this entry in the returned information
2347			 */
2348			znode_t	*ezp;
2349			if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0)
2350				goto skip_entry;
2351			if (!zfs_has_access(ezp, cr)) {
2352				VN_RELE(ZTOV(ezp));
2353				goto skip_entry;
2354			}
2355			VN_RELE(ZTOV(ezp));
2356		}
2357
2358		if (flags & V_RDDIR_ENTFLAGS)
2359			reclen = EDIRENT_RECLEN(strlen(zap.za_name));
2360		else
2361			reclen = DIRENT64_RECLEN(strlen(zap.za_name));
2362
2363		/*
2364		 * Will this entry fit in the buffer?
2365		 */
2366		if (outcount + reclen > bufsize) {
2367			/*
2368			 * Did we manage to fit anything in the buffer?
2369			 */
2370			if (!outcount) {
2371				error = SET_ERROR(EINVAL);
2372				goto update;
2373			}
2374			break;
2375		}
2376		if (flags & V_RDDIR_ENTFLAGS) {
2377			/*
2378			 * Add extended flag entry:
2379			 */
2380			eodp->ed_ino = objnum;
2381			eodp->ed_reclen = reclen;
2382			/* NOTE: ed_off is the offset for the *next* entry */
2383			next = &(eodp->ed_off);
2384			eodp->ed_eflags = zap.za_normalization_conflict ?
2385			    ED_CASE_CONFLICT : 0;
2386			(void) strncpy(eodp->ed_name, zap.za_name,
2387			    EDIRENT_NAMELEN(reclen));
2388			eodp = (edirent_t *)((intptr_t)eodp + reclen);
2389		} else {
2390			/*
2391			 * Add normal entry:
2392			 */
2393			odp->d_ino = objnum;
2394			odp->d_reclen = reclen;
2395			/* NOTE: d_off is the offset for the *next* entry */
2396			next = &(odp->d_off);
2397			(void) strncpy(odp->d_name, zap.za_name,
2398			    DIRENT64_NAMELEN(reclen));
2399			odp = (dirent64_t *)((intptr_t)odp + reclen);
2400		}
2401		outcount += reclen;
2402
2403		ASSERT(outcount <= bufsize);
2404
2405		/* Prefetch znode */
2406		if (prefetch)
2407			dmu_prefetch(os, objnum, 0, 0);
2408
2409	skip_entry:
2410		/*
2411		 * Move to the next entry, fill in the previous offset.
2412		 */
2413		if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) {
2414			zap_cursor_advance(&zc);
2415			offset = zap_cursor_serialize(&zc);
2416		} else {
2417			offset += 1;
2418		}
2419		if (next)
2420			*next = offset;
2421	}
2422	zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */
2423
2424	if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) {
2425		iovp->iov_base += outcount;
2426		iovp->iov_len -= outcount;
2427		uio->uio_resid -= outcount;
2428	} else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) {
2429		/*
2430		 * Reset the pointer.
2431		 */
2432		offset = uio->uio_loffset;
2433	}
2434
2435update:
2436	zap_cursor_fini(&zc);
2437	if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1)
2438		kmem_free(outbuf, bufsize);
2439
2440	if (error == ENOENT)
2441		error = 0;
2442
2443	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
2444
2445	uio->uio_loffset = offset;
2446	ZFS_EXIT(zfsvfs);
2447	return (error);
2448}
2449
2450ulong_t zfs_fsync_sync_cnt = 4;
2451
2452static int
2453zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
2454{
2455	znode_t	*zp = VTOZ(vp);
2456	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2457
2458	/*
2459	 * Regardless of whether this is required for standards conformance,
2460	 * this is the logical behavior when fsync() is called on a file with
2461	 * dirty pages.  We use B_ASYNC since the ZIL transactions are already
2462	 * going to be pushed out as part of the zil_commit().
2463	 */
2464	if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2465	    (vp->v_type == VREG) && !(IS_SWAPVP(vp)))
2466		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct);
2467
2468	(void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt);
2469
2470	if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) {
2471		ZFS_ENTER(zfsvfs);
2472		ZFS_VERIFY_ZP(zp);
2473		zil_commit(zfsvfs->z_log, zp->z_id);
2474		ZFS_EXIT(zfsvfs);
2475	}
2476	return (0);
2477}
2478
2479
2480/*
2481 * Get the requested file attributes and place them in the provided
2482 * vattr structure.
2483 *
2484 *	IN:	vp	- vnode of file.
2485 *		vap	- va_mask identifies requested attributes.
2486 *			  If AT_XVATTR set, then optional attrs are requested
2487 *		flags	- ATTR_NOACLCHECK (CIFS server context)
2488 *		cr	- credentials of caller.
2489 *		ct	- caller context
2490 *
2491 *	OUT:	vap	- attribute values.
2492 *
2493 *	RETURN:	0 (always succeeds).
2494 */
2495/* ARGSUSED */
2496static int
2497zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2498    caller_context_t *ct)
2499{
2500	znode_t *zp = VTOZ(vp);
2501	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2502	int	error = 0;
2503	uint64_t links;
2504	uint64_t mtime[2], ctime[2];
2505	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2506	xoptattr_t *xoap = NULL;
2507	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2508	sa_bulk_attr_t bulk[2];
2509	int count = 0;
2510
2511	ZFS_ENTER(zfsvfs);
2512	ZFS_VERIFY_ZP(zp);
2513
2514	zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid);
2515
2516	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16);
2517	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16);
2518
2519	if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
2520		ZFS_EXIT(zfsvfs);
2521		return (error);
2522	}
2523
2524	/*
2525	 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2526	 * Also, if we are the owner don't bother, since owner should
2527	 * always be allowed to read basic attributes of file.
2528	 */
2529	if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) &&
2530	    (vap->va_uid != crgetuid(cr))) {
2531		if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0,
2532		    skipaclchk, cr)) {
2533			ZFS_EXIT(zfsvfs);
2534			return (error);
2535		}
2536	}
2537
2538	/*
2539	 * Return all attributes.  It's cheaper to provide the answer
2540	 * than to determine whether we were asked the question.
2541	 */
2542
2543	mutex_enter(&zp->z_lock);
2544	vap->va_type = vp->v_type;
2545	vap->va_mode = zp->z_mode & MODEMASK;
2546	vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev;
2547	vap->va_nodeid = zp->z_id;
2548	if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp))
2549		links = zp->z_links + 1;
2550	else
2551		links = zp->z_links;
2552	vap->va_nlink = MIN(links, UINT32_MAX);	/* nlink_t limit! */
2553	vap->va_size = zp->z_size;
2554	vap->va_rdev = vp->v_rdev;
2555	vap->va_seq = zp->z_seq;
2556
2557	/*
2558	 * Add in any requested optional attributes and the create time.
2559	 * Also set the corresponding bits in the returned attribute bitmap.
2560	 */
2561	if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) {
2562		if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) {
2563			xoap->xoa_archive =
2564			    ((zp->z_pflags & ZFS_ARCHIVE) != 0);
2565			XVA_SET_RTN(xvap, XAT_ARCHIVE);
2566		}
2567
2568		if (XVA_ISSET_REQ(xvap, XAT_READONLY)) {
2569			xoap->xoa_readonly =
2570			    ((zp->z_pflags & ZFS_READONLY) != 0);
2571			XVA_SET_RTN(xvap, XAT_READONLY);
2572		}
2573
2574		if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) {
2575			xoap->xoa_system =
2576			    ((zp->z_pflags & ZFS_SYSTEM) != 0);
2577			XVA_SET_RTN(xvap, XAT_SYSTEM);
2578		}
2579
2580		if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) {
2581			xoap->xoa_hidden =
2582			    ((zp->z_pflags & ZFS_HIDDEN) != 0);
2583			XVA_SET_RTN(xvap, XAT_HIDDEN);
2584		}
2585
2586		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2587			xoap->xoa_nounlink =
2588			    ((zp->z_pflags & ZFS_NOUNLINK) != 0);
2589			XVA_SET_RTN(xvap, XAT_NOUNLINK);
2590		}
2591
2592		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2593			xoap->xoa_immutable =
2594			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0);
2595			XVA_SET_RTN(xvap, XAT_IMMUTABLE);
2596		}
2597
2598		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2599			xoap->xoa_appendonly =
2600			    ((zp->z_pflags & ZFS_APPENDONLY) != 0);
2601			XVA_SET_RTN(xvap, XAT_APPENDONLY);
2602		}
2603
2604		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2605			xoap->xoa_nodump =
2606			    ((zp->z_pflags & ZFS_NODUMP) != 0);
2607			XVA_SET_RTN(xvap, XAT_NODUMP);
2608		}
2609
2610		if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) {
2611			xoap->xoa_opaque =
2612			    ((zp->z_pflags & ZFS_OPAQUE) != 0);
2613			XVA_SET_RTN(xvap, XAT_OPAQUE);
2614		}
2615
2616		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2617			xoap->xoa_av_quarantined =
2618			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0);
2619			XVA_SET_RTN(xvap, XAT_AV_QUARANTINED);
2620		}
2621
2622		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2623			xoap->xoa_av_modified =
2624			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0);
2625			XVA_SET_RTN(xvap, XAT_AV_MODIFIED);
2626		}
2627
2628		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) &&
2629		    vp->v_type == VREG) {
2630			zfs_sa_get_scanstamp(zp, xvap);
2631		}
2632
2633		if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) {
2634			uint64_t times[2];
2635
2636			(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
2637			    times, sizeof (times));
2638			ZFS_TIME_DECODE(&xoap->xoa_createtime, times);
2639			XVA_SET_RTN(xvap, XAT_CREATETIME);
2640		}
2641
2642		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2643			xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0);
2644			XVA_SET_RTN(xvap, XAT_REPARSE);
2645		}
2646		if (XVA_ISSET_REQ(xvap, XAT_GEN)) {
2647			xoap->xoa_generation = zp->z_gen;
2648			XVA_SET_RTN(xvap, XAT_GEN);
2649		}
2650
2651		if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) {
2652			xoap->xoa_offline =
2653			    ((zp->z_pflags & ZFS_OFFLINE) != 0);
2654			XVA_SET_RTN(xvap, XAT_OFFLINE);
2655		}
2656
2657		if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) {
2658			xoap->xoa_sparse =
2659			    ((zp->z_pflags & ZFS_SPARSE) != 0);
2660			XVA_SET_RTN(xvap, XAT_SPARSE);
2661		}
2662	}
2663
2664	ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime);
2665	ZFS_TIME_DECODE(&vap->va_mtime, mtime);
2666	ZFS_TIME_DECODE(&vap->va_ctime, ctime);
2667
2668	mutex_exit(&zp->z_lock);
2669
2670	sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks);
2671
2672	if (zp->z_blksz == 0) {
2673		/*
2674		 * Block size hasn't been set; suggest maximal I/O transfers.
2675		 */
2676		vap->va_blksize = zfsvfs->z_max_blksz;
2677	}
2678
2679	ZFS_EXIT(zfsvfs);
2680	return (0);
2681}
2682
2683/*
2684 * Set the file attributes to the values contained in the
2685 * vattr structure.
2686 *
2687 *	IN:	vp	- vnode of file to be modified.
2688 *		vap	- new attribute values.
2689 *			  If AT_XVATTR set, then optional attrs are being set
2690 *		flags	- ATTR_UTIME set if non-default time values provided.
2691 *			- ATTR_NOACLCHECK (CIFS context only).
2692 *		cr	- credentials of caller.
2693 *		ct	- caller context
2694 *
2695 *	RETURN:	0 on success, error code on failure.
2696 *
2697 * Timestamps:
2698 *	vp - ctime updated, mtime updated if size changed.
2699 */
2700/* ARGSUSED */
2701static int
2702zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2703    caller_context_t *ct)
2704{
2705	znode_t		*zp = VTOZ(vp);
2706	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2707	zilog_t		*zilog;
2708	dmu_tx_t	*tx;
2709	vattr_t		oldva;
2710	xvattr_t	tmpxvattr;
2711	uint_t		mask = vap->va_mask;
2712	uint_t		saved_mask = 0;
2713	int		trim_mask = 0;
2714	uint64_t	new_mode;
2715	uint64_t	new_uid, new_gid;
2716	uint64_t	xattr_obj;
2717	uint64_t	mtime[2], ctime[2];
2718	znode_t		*attrzp;
2719	int		need_policy = FALSE;
2720	int		err, err2;
2721	zfs_fuid_info_t *fuidp = NULL;
2722	xvattr_t *xvap = (xvattr_t *)vap;	/* vap may be an xvattr_t * */
2723	xoptattr_t	*xoap;
2724	zfs_acl_t	*aclp;
2725	boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
2726	boolean_t	fuid_dirtied = B_FALSE;
2727	sa_bulk_attr_t	bulk[7], xattr_bulk[7];
2728	int		count = 0, xattr_count = 0;
2729
2730	if (mask == 0)
2731		return (0);
2732
2733	if (mask & AT_NOSET)
2734		return (SET_ERROR(EINVAL));
2735
2736	ZFS_ENTER(zfsvfs);
2737	ZFS_VERIFY_ZP(zp);
2738
2739	zilog = zfsvfs->z_log;
2740
2741	/*
2742	 * Make sure that if we have ephemeral uid/gid or xvattr specified
2743	 * that file system is at proper version level
2744	 */
2745
2746	if (zfsvfs->z_use_fuids == B_FALSE &&
2747	    (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) ||
2748	    ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) ||
2749	    (mask & AT_XVATTR))) {
2750		ZFS_EXIT(zfsvfs);
2751		return (SET_ERROR(EINVAL));
2752	}
2753
2754	if (mask & AT_SIZE && vp->v_type == VDIR) {
2755		ZFS_EXIT(zfsvfs);
2756		return (SET_ERROR(EISDIR));
2757	}
2758
2759	if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) {
2760		ZFS_EXIT(zfsvfs);
2761		return (SET_ERROR(EINVAL));
2762	}
2763
2764	/*
2765	 * If this is an xvattr_t, then get a pointer to the structure of
2766	 * optional attributes.  If this is NULL, then we have a vattr_t.
2767	 */
2768	xoap = xva_getxoptattr(xvap);
2769
2770	xva_init(&tmpxvattr);
2771
2772	/*
2773	 * Immutable files can only alter immutable bit and atime
2774	 */
2775	if ((zp->z_pflags & ZFS_IMMUTABLE) &&
2776	    ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) ||
2777	    ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) {
2778		ZFS_EXIT(zfsvfs);
2779		return (SET_ERROR(EPERM));
2780	}
2781
2782	if ((mask & AT_SIZE) && (zp->z_pflags & ZFS_READONLY)) {
2783		ZFS_EXIT(zfsvfs);
2784		return (SET_ERROR(EPERM));
2785	}
2786
2787	/*
2788	 * Verify timestamps doesn't overflow 32 bits.
2789	 * ZFS can handle large timestamps, but 32bit syscalls can't
2790	 * handle times greater than 2039.  This check should be removed
2791	 * once large timestamps are fully supported.
2792	 */
2793	if (mask & (AT_ATIME | AT_MTIME)) {
2794		if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2795		    ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2796			ZFS_EXIT(zfsvfs);
2797			return (SET_ERROR(EOVERFLOW));
2798		}
2799	}
2800
2801top:
2802	attrzp = NULL;
2803	aclp = NULL;
2804
2805	/* Can this be moved to before the top label? */
2806	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
2807		ZFS_EXIT(zfsvfs);
2808		return (SET_ERROR(EROFS));
2809	}
2810
2811	/*
2812	 * First validate permissions
2813	 */
2814
2815	if (mask & AT_SIZE) {
2816		err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr);
2817		if (err) {
2818			ZFS_EXIT(zfsvfs);
2819			return (err);
2820		}
2821		/*
2822		 * XXX - Note, we are not providing any open
2823		 * mode flags here (like FNDELAY), so we may
2824		 * block if there are locks present... this
2825		 * should be addressed in openat().
2826		 */
2827		/* XXX - would it be OK to generate a log record here? */
2828		err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE);
2829		if (err) {
2830			ZFS_EXIT(zfsvfs);
2831			return (err);
2832		}
2833
2834		if (vap->va_size == 0)
2835			vnevent_truncate(ZTOV(zp), ct);
2836	}
2837
2838	if (mask & (AT_ATIME|AT_MTIME) ||
2839	    ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) ||
2840	    XVA_ISSET_REQ(xvap, XAT_READONLY) ||
2841	    XVA_ISSET_REQ(xvap, XAT_ARCHIVE) ||
2842	    XVA_ISSET_REQ(xvap, XAT_OFFLINE) ||
2843	    XVA_ISSET_REQ(xvap, XAT_SPARSE) ||
2844	    XVA_ISSET_REQ(xvap, XAT_CREATETIME) ||
2845	    XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) {
2846		need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0,
2847		    skipaclchk, cr);
2848	}
2849
2850	if (mask & (AT_UID|AT_GID)) {
2851		int	idmask = (mask & (AT_UID|AT_GID));
2852		int	take_owner;
2853		int	take_group;
2854
2855		/*
2856		 * NOTE: even if a new mode is being set,
2857		 * we may clear S_ISUID/S_ISGID bits.
2858		 */
2859
2860		if (!(mask & AT_MODE))
2861			vap->va_mode = zp->z_mode;
2862
2863		/*
2864		 * Take ownership or chgrp to group we are a member of
2865		 */
2866
2867		take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr));
2868		take_group = (mask & AT_GID) &&
2869		    zfs_groupmember(zfsvfs, vap->va_gid, cr);
2870
2871		/*
2872		 * If both AT_UID and AT_GID are set then take_owner and
2873		 * take_group must both be set in order to allow taking
2874		 * ownership.
2875		 *
2876		 * Otherwise, send the check through secpolicy_vnode_setattr()
2877		 *
2878		 */
2879
2880		if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) ||
2881		    ((idmask == AT_UID) && take_owner) ||
2882		    ((idmask == AT_GID) && take_group)) {
2883			if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0,
2884			    skipaclchk, cr) == 0) {
2885				/*
2886				 * Remove setuid/setgid for non-privileged users
2887				 */
2888				secpolicy_setid_clear(vap, cr);
2889				trim_mask = (mask & (AT_UID|AT_GID));
2890			} else {
2891				need_policy =  TRUE;
2892			}
2893		} else {
2894			need_policy =  TRUE;
2895		}
2896	}
2897
2898	mutex_enter(&zp->z_lock);
2899	oldva.va_mode = zp->z_mode;
2900	zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid);
2901	if (mask & AT_XVATTR) {
2902		/*
2903		 * Update xvattr mask to include only those attributes
2904		 * that are actually changing.
2905		 *
2906		 * the bits will be restored prior to actually setting
2907		 * the attributes so the caller thinks they were set.
2908		 */
2909		if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) {
2910			if (xoap->xoa_appendonly !=
2911			    ((zp->z_pflags & ZFS_APPENDONLY) != 0)) {
2912				need_policy = TRUE;
2913			} else {
2914				XVA_CLR_REQ(xvap, XAT_APPENDONLY);
2915				XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY);
2916			}
2917		}
2918
2919		if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) {
2920			if (xoap->xoa_nounlink !=
2921			    ((zp->z_pflags & ZFS_NOUNLINK) != 0)) {
2922				need_policy = TRUE;
2923			} else {
2924				XVA_CLR_REQ(xvap, XAT_NOUNLINK);
2925				XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK);
2926			}
2927		}
2928
2929		if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) {
2930			if (xoap->xoa_immutable !=
2931			    ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) {
2932				need_policy = TRUE;
2933			} else {
2934				XVA_CLR_REQ(xvap, XAT_IMMUTABLE);
2935				XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE);
2936			}
2937		}
2938
2939		if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) {
2940			if (xoap->xoa_nodump !=
2941			    ((zp->z_pflags & ZFS_NODUMP) != 0)) {
2942				need_policy = TRUE;
2943			} else {
2944				XVA_CLR_REQ(xvap, XAT_NODUMP);
2945				XVA_SET_REQ(&tmpxvattr, XAT_NODUMP);
2946			}
2947		}
2948
2949		if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) {
2950			if (xoap->xoa_av_modified !=
2951			    ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) {
2952				need_policy = TRUE;
2953			} else {
2954				XVA_CLR_REQ(xvap, XAT_AV_MODIFIED);
2955				XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED);
2956			}
2957		}
2958
2959		if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) {
2960			if ((vp->v_type != VREG &&
2961			    xoap->xoa_av_quarantined) ||
2962			    xoap->xoa_av_quarantined !=
2963			    ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) {
2964				need_policy = TRUE;
2965			} else {
2966				XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED);
2967				XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED);
2968			}
2969		}
2970
2971		if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) {
2972			mutex_exit(&zp->z_lock);
2973			ZFS_EXIT(zfsvfs);
2974			return (SET_ERROR(EPERM));
2975		}
2976
2977		if (need_policy == FALSE &&
2978		    (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) ||
2979		    XVA_ISSET_REQ(xvap, XAT_OPAQUE))) {
2980			need_policy = TRUE;
2981		}
2982	}
2983
2984	mutex_exit(&zp->z_lock);
2985
2986	if (mask & AT_MODE) {
2987		if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) {
2988			err = secpolicy_setid_setsticky_clear(vp, vap,
2989			    &oldva, cr);
2990			if (err) {
2991				ZFS_EXIT(zfsvfs);
2992				return (err);
2993			}
2994			trim_mask |= AT_MODE;
2995		} else {
2996			need_policy = TRUE;
2997		}
2998	}
2999
3000	if (need_policy) {
3001		/*
3002		 * If trim_mask is set then take ownership
3003		 * has been granted or write_acl is present and user
3004		 * has the ability to modify mode.  In that case remove
3005		 * UID|GID and or MODE from mask so that
3006		 * secpolicy_vnode_setattr() doesn't revoke it.
3007		 */
3008
3009		if (trim_mask) {
3010			saved_mask = vap->va_mask;
3011			vap->va_mask &= ~trim_mask;
3012		}
3013		err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
3014		    (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp);
3015		if (err) {
3016			ZFS_EXIT(zfsvfs);
3017			return (err);
3018		}
3019
3020		if (trim_mask)
3021			vap->va_mask |= saved_mask;
3022	}
3023
3024	/*
3025	 * secpolicy_vnode_setattr, or take ownership may have
3026	 * changed va_mask
3027	 */
3028	mask = vap->va_mask;
3029
3030	if ((mask & (AT_UID | AT_GID))) {
3031		err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
3032		    &xattr_obj, sizeof (xattr_obj));
3033
3034		if (err == 0 && xattr_obj) {
3035			err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp);
3036			if (err)
3037				goto out2;
3038		}
3039		if (mask & AT_UID) {
3040			new_uid = zfs_fuid_create(zfsvfs,
3041			    (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp);
3042			if (new_uid != zp->z_uid &&
3043			    zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) {
3044				if (attrzp)
3045					VN_RELE(ZTOV(attrzp));
3046				err = SET_ERROR(EDQUOT);
3047				goto out2;
3048			}
3049		}
3050
3051		if (mask & AT_GID) {
3052			new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
3053			    cr, ZFS_GROUP, &fuidp);
3054			if (new_gid != zp->z_gid &&
3055			    zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) {
3056				if (attrzp)
3057					VN_RELE(ZTOV(attrzp));
3058				err = SET_ERROR(EDQUOT);
3059				goto out2;
3060			}
3061		}
3062	}
3063	tx = dmu_tx_create(zfsvfs->z_os);
3064
3065	if (mask & AT_MODE) {
3066		uint64_t pmode = zp->z_mode;
3067		uint64_t acl_obj;
3068		new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT);
3069
3070		if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
3071		    !(zp->z_pflags & ZFS_ACL_TRIVIAL)) {
3072			err = SET_ERROR(EPERM);
3073			goto out;
3074		}
3075
3076		if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))
3077			goto out;
3078
3079		mutex_enter(&zp->z_lock);
3080		if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) {
3081			/*
3082			 * Are we upgrading ACL from old V0 format
3083			 * to V1 format?
3084			 */
3085			if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
3086			    zfs_znode_acl_version(zp) ==
3087			    ZFS_ACL_VERSION_INITIAL) {
3088				dmu_tx_hold_free(tx, acl_obj, 0,
3089				    DMU_OBJECT_END);
3090				dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3091				    0, aclp->z_acl_bytes);
3092			} else {
3093				dmu_tx_hold_write(tx, acl_obj, 0,
3094				    aclp->z_acl_bytes);
3095			}
3096		} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3097			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
3098			    0, aclp->z_acl_bytes);
3099		}
3100		mutex_exit(&zp->z_lock);
3101		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3102	} else {
3103		if ((mask & AT_XVATTR) &&
3104		    XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3105			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
3106		else
3107			dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
3108	}
3109
3110	if (attrzp) {
3111		dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE);
3112	}
3113
3114	fuid_dirtied = zfsvfs->z_fuid_dirty;
3115	if (fuid_dirtied)
3116		zfs_fuid_txhold(zfsvfs, tx);
3117
3118	zfs_sa_upgrade_txholds(tx, zp);
3119
3120	err = dmu_tx_assign(tx, TXG_WAIT);
3121	if (err)
3122		goto out;
3123
3124	count = 0;
3125	/*
3126	 * Set each attribute requested.
3127	 * We group settings according to the locks they need to acquire.
3128	 *
3129	 * Note: you cannot set ctime directly, although it will be
3130	 * updated as a side-effect of calling this function.
3131	 */
3132
3133
3134	if (mask & (AT_UID|AT_GID|AT_MODE))
3135		mutex_enter(&zp->z_acl_lock);
3136	mutex_enter(&zp->z_lock);
3137
3138	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
3139	    &zp->z_pflags, sizeof (zp->z_pflags));
3140
3141	if (attrzp) {
3142		if (mask & (AT_UID|AT_GID|AT_MODE))
3143			mutex_enter(&attrzp->z_acl_lock);
3144		mutex_enter(&attrzp->z_lock);
3145		SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3146		    SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags,
3147		    sizeof (attrzp->z_pflags));
3148	}
3149
3150	if (mask & (AT_UID|AT_GID)) {
3151
3152		if (mask & AT_UID) {
3153			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL,
3154			    &new_uid, sizeof (new_uid));
3155			zp->z_uid = new_uid;
3156			if (attrzp) {
3157				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3158				    SA_ZPL_UID(zfsvfs), NULL, &new_uid,
3159				    sizeof (new_uid));
3160				attrzp->z_uid = new_uid;
3161			}
3162		}
3163
3164		if (mask & AT_GID) {
3165			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs),
3166			    NULL, &new_gid, sizeof (new_gid));
3167			zp->z_gid = new_gid;
3168			if (attrzp) {
3169				SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3170				    SA_ZPL_GID(zfsvfs), NULL, &new_gid,
3171				    sizeof (new_gid));
3172				attrzp->z_gid = new_gid;
3173			}
3174		}
3175		if (!(mask & AT_MODE)) {
3176			SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs),
3177			    NULL, &new_mode, sizeof (new_mode));
3178			new_mode = zp->z_mode;
3179		}
3180		err = zfs_acl_chown_setattr(zp);
3181		ASSERT(err == 0);
3182		if (attrzp) {
3183			err = zfs_acl_chown_setattr(attrzp);
3184			ASSERT(err == 0);
3185		}
3186	}
3187
3188	if (mask & AT_MODE) {
3189		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
3190		    &new_mode, sizeof (new_mode));
3191		zp->z_mode = new_mode;
3192		ASSERT3U((uintptr_t)aclp, !=, NULL);
3193		err = zfs_aclset_common(zp, aclp, cr, tx);
3194		ASSERT0(err);
3195		if (zp->z_acl_cached)
3196			zfs_acl_free(zp->z_acl_cached);
3197		zp->z_acl_cached = aclp;
3198		aclp = NULL;
3199	}
3200
3201
3202	if (mask & AT_ATIME) {
3203		ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime);
3204		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL,
3205		    &zp->z_atime, sizeof (zp->z_atime));
3206	}
3207
3208	if (mask & AT_MTIME) {
3209		ZFS_TIME_ENCODE(&vap->va_mtime, mtime);
3210		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
3211		    mtime, sizeof (mtime));
3212	}
3213
3214	/* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3215	if (mask & AT_SIZE && !(mask & AT_MTIME)) {
3216		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
3217		    NULL, mtime, sizeof (mtime));
3218		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3219		    &ctime, sizeof (ctime));
3220		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
3221		    B_TRUE);
3222	} else if (mask != 0) {
3223		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
3224		    &ctime, sizeof (ctime));
3225		zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
3226		    B_TRUE);
3227		if (attrzp) {
3228			SA_ADD_BULK_ATTR(xattr_bulk, xattr_count,
3229			    SA_ZPL_CTIME(zfsvfs), NULL,
3230			    &ctime, sizeof (ctime));
3231			zfs_tstamp_update_setup(attrzp, STATE_CHANGED,
3232			    mtime, ctime, B_TRUE);
3233		}
3234	}
3235	/*
3236	 * Do this after setting timestamps to prevent timestamp
3237	 * update from toggling bit
3238	 */
3239
3240	if (xoap && (mask & AT_XVATTR)) {
3241
3242		/*
3243		 * restore trimmed off masks
3244		 * so that return masks can be set for caller.
3245		 */
3246
3247		if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) {
3248			XVA_SET_REQ(xvap, XAT_APPENDONLY);
3249		}
3250		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) {
3251			XVA_SET_REQ(xvap, XAT_NOUNLINK);
3252		}
3253		if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) {
3254			XVA_SET_REQ(xvap, XAT_IMMUTABLE);
3255		}
3256		if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) {
3257			XVA_SET_REQ(xvap, XAT_NODUMP);
3258		}
3259		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) {
3260			XVA_SET_REQ(xvap, XAT_AV_MODIFIED);
3261		}
3262		if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) {
3263			XVA_SET_REQ(xvap, XAT_AV_QUARANTINED);
3264		}
3265
3266		if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP))
3267			ASSERT(vp->v_type == VREG);
3268
3269		zfs_xvattr_set(zp, xvap, tx);
3270	}
3271
3272	if (fuid_dirtied)
3273		zfs_fuid_sync(zfsvfs, tx);
3274
3275	if (mask != 0)
3276		zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp);
3277
3278	mutex_exit(&zp->z_lock);
3279	if (mask & (AT_UID|AT_GID|AT_MODE))
3280		mutex_exit(&zp->z_acl_lock);
3281
3282	if (attrzp) {
3283		if (mask & (AT_UID|AT_GID|AT_MODE))
3284			mutex_exit(&attrzp->z_acl_lock);
3285		mutex_exit(&attrzp->z_lock);
3286	}
3287out:
3288	if (err == 0 && attrzp) {
3289		err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk,
3290		    xattr_count, tx);
3291		ASSERT(err2 == 0);
3292	}
3293
3294	if (attrzp)
3295		VN_RELE(ZTOV(attrzp));
3296
3297	if (aclp)
3298		zfs_acl_free(aclp);
3299
3300	if (fuidp) {
3301		zfs_fuid_info_free(fuidp);
3302		fuidp = NULL;
3303	}
3304
3305	if (err) {
3306		dmu_tx_abort(tx);
3307		if (err == ERESTART)
3308			goto top;
3309	} else {
3310		err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
3311		dmu_tx_commit(tx);
3312	}
3313
3314out2:
3315	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3316		zil_commit(zilog, 0);
3317
3318	ZFS_EXIT(zfsvfs);
3319	return (err);
3320}
3321
3322typedef struct zfs_zlock {
3323	krwlock_t	*zl_rwlock;	/* lock we acquired */
3324	znode_t		*zl_znode;	/* znode we held */
3325	struct zfs_zlock *zl_next;	/* next in list */
3326} zfs_zlock_t;
3327
3328/*
3329 * Drop locks and release vnodes that were held by zfs_rename_lock().
3330 */
3331static void
3332zfs_rename_unlock(zfs_zlock_t **zlpp)
3333{
3334	zfs_zlock_t *zl;
3335
3336	while ((zl = *zlpp) != NULL) {
3337		if (zl->zl_znode != NULL)
3338			VN_RELE(ZTOV(zl->zl_znode));
3339		rw_exit(zl->zl_rwlock);
3340		*zlpp = zl->zl_next;
3341		kmem_free(zl, sizeof (*zl));
3342	}
3343}
3344
3345/*
3346 * Search back through the directory tree, using the ".." entries.
3347 * Lock each directory in the chain to prevent concurrent renames.
3348 * Fail any attempt to move a directory into one of its own descendants.
3349 * XXX - z_parent_lock can overlap with map or grow locks
3350 */
3351static int
3352zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp)
3353{
3354	zfs_zlock_t	*zl;
3355	znode_t		*zp = tdzp;
3356	uint64_t	rootid = zp->z_zfsvfs->z_root;
3357	uint64_t	oidp = zp->z_id;
3358	krwlock_t	*rwlp = &szp->z_parent_lock;
3359	krw_t		rw = RW_WRITER;
3360
3361	/*
3362	 * First pass write-locks szp and compares to zp->z_id.
3363	 * Later passes read-lock zp and compare to zp->z_parent.
3364	 */
3365	do {
3366		if (!rw_tryenter(rwlp, rw)) {
3367			/*
3368			 * Another thread is renaming in this path.
3369			 * Note that if we are a WRITER, we don't have any
3370			 * parent_locks held yet.
3371			 */
3372			if (rw == RW_READER && zp->z_id > szp->z_id) {
3373				/*
3374				 * Drop our locks and restart
3375				 */
3376				zfs_rename_unlock(&zl);
3377				*zlpp = NULL;
3378				zp = tdzp;
3379				oidp = zp->z_id;
3380				rwlp = &szp->z_parent_lock;
3381				rw = RW_WRITER;
3382				continue;
3383			} else {
3384				/*
3385				 * Wait for other thread to drop its locks
3386				 */
3387				rw_enter(rwlp, rw);
3388			}
3389		}
3390
3391		zl = kmem_alloc(sizeof (*zl), KM_SLEEP);
3392		zl->zl_rwlock = rwlp;
3393		zl->zl_znode = NULL;
3394		zl->zl_next = *zlpp;
3395		*zlpp = zl;
3396
3397		if (oidp == szp->z_id)		/* We're a descendant of szp */
3398			return (SET_ERROR(EINVAL));
3399
3400		if (oidp == rootid)		/* We've hit the top */
3401			return (0);
3402
3403		if (rw == RW_READER) {		/* i.e. not the first pass */
3404			int error = zfs_zget(zp->z_zfsvfs, oidp, &zp);
3405			if (error)
3406				return (error);
3407			zl->zl_znode = zp;
3408		}
3409		(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs),
3410		    &oidp, sizeof (oidp));
3411		rwlp = &zp->z_parent_lock;
3412		rw = RW_READER;
3413
3414	} while (zp->z_id != sdzp->z_id);
3415
3416	return (0);
3417}
3418
3419/*
3420 * Move an entry from the provided source directory to the target
3421 * directory.  Change the entry name as indicated.
3422 *
3423 *	IN:	sdvp	- Source directory containing the "old entry".
3424 *		snm	- Old entry name.
3425 *		tdvp	- Target directory to contain the "new entry".
3426 *		tnm	- New entry name.
3427 *		cr	- credentials of caller.
3428 *		ct	- caller context
3429 *		flags	- case flags
3430 *
3431 *	RETURN:	0 on success, error code on failure.
3432 *
3433 * Timestamps:
3434 *	sdvp,tdvp - ctime|mtime updated
3435 */
3436/*ARGSUSED*/
3437static int
3438zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
3439    caller_context_t *ct, int flags)
3440{
3441	znode_t		*tdzp, *szp, *tzp;
3442	znode_t		*sdzp = VTOZ(sdvp);
3443	zfsvfs_t	*zfsvfs = sdzp->z_zfsvfs;
3444	zilog_t		*zilog;
3445	vnode_t		*realvp;
3446	zfs_dirlock_t	*sdl, *tdl;
3447	dmu_tx_t	*tx;
3448	zfs_zlock_t	*zl;
3449	int		cmp, serr, terr;
3450	int		error = 0;
3451	int		zflg = 0;
3452	boolean_t	waited = B_FALSE;
3453
3454	ZFS_ENTER(zfsvfs);
3455	ZFS_VERIFY_ZP(sdzp);
3456	zilog = zfsvfs->z_log;
3457
3458	/*
3459	 * Make sure we have the real vp for the target directory.
3460	 */
3461	if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3462		tdvp = realvp;
3463
3464	tdzp = VTOZ(tdvp);
3465	ZFS_VERIFY_ZP(tdzp);
3466
3467	/*
3468	 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3469	 * ctldir appear to have the same v_vfsp.
3470	 */
3471	if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) {
3472		ZFS_EXIT(zfsvfs);
3473		return (SET_ERROR(EXDEV));
3474	}
3475
3476	if (zfsvfs->z_utf8 && u8_validate(tnm,
3477	    strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3478		ZFS_EXIT(zfsvfs);
3479		return (SET_ERROR(EILSEQ));
3480	}
3481
3482	if (flags & FIGNORECASE)
3483		zflg |= ZCILOOK;
3484
3485top:
3486	szp = NULL;
3487	tzp = NULL;
3488	zl = NULL;
3489
3490	/*
3491	 * This is to prevent the creation of links into attribute space
3492	 * by renaming a linked file into/outof an attribute directory.
3493	 * See the comment in zfs_link() for why this is considered bad.
3494	 */
3495	if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) {
3496		ZFS_EXIT(zfsvfs);
3497		return (SET_ERROR(EINVAL));
3498	}
3499
3500	/*
3501	 * Lock source and target directory entries.  To prevent deadlock,
3502	 * a lock ordering must be defined.  We lock the directory with
3503	 * the smallest object id first, or if it's a tie, the one with
3504	 * the lexically first name.
3505	 */
3506	if (sdzp->z_id < tdzp->z_id) {
3507		cmp = -1;
3508	} else if (sdzp->z_id > tdzp->z_id) {
3509		cmp = 1;
3510	} else {
3511		/*
3512		 * First compare the two name arguments without
3513		 * considering any case folding.
3514		 */
3515		int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER);
3516
3517		cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error);
3518		ASSERT(error == 0 || !zfsvfs->z_utf8);
3519		if (cmp == 0) {
3520			/*
3521			 * POSIX: "If the old argument and the new argument
3522			 * both refer to links to the same existing file,
3523			 * the rename() function shall return successfully
3524			 * and perform no other action."
3525			 */
3526			ZFS_EXIT(zfsvfs);
3527			return (0);
3528		}
3529		/*
3530		 * If the file system is case-folding, then we may
3531		 * have some more checking to do.  A case-folding file
3532		 * system is either supporting mixed case sensitivity
3533		 * access or is completely case-insensitive.  Note
3534		 * that the file system is always case preserving.
3535		 *
3536		 * In mixed sensitivity mode case sensitive behavior
3537		 * is the default.  FIGNORECASE must be used to
3538		 * explicitly request case insensitive behavior.
3539		 *
3540		 * If the source and target names provided differ only
3541		 * by case (e.g., a request to rename 'tim' to 'Tim'),
3542		 * we will treat this as a special case in the
3543		 * case-insensitive mode: as long as the source name
3544		 * is an exact match, we will allow this to proceed as
3545		 * a name-change request.
3546		 */
3547		if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
3548		    (zfsvfs->z_case == ZFS_CASE_MIXED &&
3549		    flags & FIGNORECASE)) &&
3550		    u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST,
3551		    &error) == 0) {
3552			/*
3553			 * case preserving rename request, require exact
3554			 * name matches
3555			 */
3556			zflg |= ZCIEXACT;
3557			zflg &= ~ZCILOOK;
3558		}
3559	}
3560
3561	/*
3562	 * If the source and destination directories are the same, we should
3563	 * grab the z_name_lock of that directory only once.
3564	 */
3565	if (sdzp == tdzp) {
3566		zflg |= ZHAVELOCK;
3567		rw_enter(&sdzp->z_name_lock, RW_READER);
3568	}
3569
3570	if (cmp < 0) {
3571		serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp,
3572		    ZEXISTS | zflg, NULL, NULL);
3573		terr = zfs_dirent_lock(&tdl,
3574		    tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL);
3575	} else {
3576		terr = zfs_dirent_lock(&tdl,
3577		    tdzp, tnm, &tzp, zflg, NULL, NULL);
3578		serr = zfs_dirent_lock(&sdl,
3579		    sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg,
3580		    NULL, NULL);
3581	}
3582
3583	if (serr) {
3584		/*
3585		 * Source entry invalid or not there.
3586		 */
3587		if (!terr) {
3588			zfs_dirent_unlock(tdl);
3589			if (tzp)
3590				VN_RELE(ZTOV(tzp));
3591		}
3592
3593		if (sdzp == tdzp)
3594			rw_exit(&sdzp->z_name_lock);
3595
3596		if (strcmp(snm, "..") == 0)
3597			serr = SET_ERROR(EINVAL);
3598		ZFS_EXIT(zfsvfs);
3599		return (serr);
3600	}
3601	if (terr) {
3602		zfs_dirent_unlock(sdl);
3603		VN_RELE(ZTOV(szp));
3604
3605		if (sdzp == tdzp)
3606			rw_exit(&sdzp->z_name_lock);
3607
3608		if (strcmp(tnm, "..") == 0)
3609			terr = SET_ERROR(EINVAL);
3610		ZFS_EXIT(zfsvfs);
3611		return (terr);
3612	}
3613
3614	/*
3615	 * Must have write access at the source to remove the old entry
3616	 * and write access at the target to create the new entry.
3617	 * Note that if target and source are the same, this can be
3618	 * done in a single check.
3619	 */
3620
3621	if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr))
3622		goto out;
3623
3624	if (ZTOV(szp)->v_type == VDIR) {
3625		/*
3626		 * Check to make sure rename is valid.
3627		 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3628		 */
3629		if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl))
3630			goto out;
3631	}
3632
3633	/*
3634	 * Does target exist?
3635	 */
3636	if (tzp) {
3637		/*
3638		 * Source and target must be the same type.
3639		 */
3640		if (ZTOV(szp)->v_type == VDIR) {
3641			if (ZTOV(tzp)->v_type != VDIR) {
3642				error = SET_ERROR(ENOTDIR);
3643				goto out;
3644			}
3645		} else {
3646			if (ZTOV(tzp)->v_type == VDIR) {
3647				error = SET_ERROR(EISDIR);
3648				goto out;
3649			}
3650		}
3651		/*
3652		 * POSIX dictates that when the source and target
3653		 * entries refer to the same file object, rename
3654		 * must do nothing and exit without error.
3655		 */
3656		if (szp->z_id == tzp->z_id) {
3657			error = 0;
3658			goto out;
3659		}
3660	}
3661
3662	vnevent_rename_src(ZTOV(szp), sdvp, snm, ct);
3663	if (tzp)
3664		vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct);
3665
3666	/*
3667	 * notify the target directory if it is not the same
3668	 * as source directory.
3669	 */
3670	if (tdvp != sdvp) {
3671		vnevent_rename_dest_dir(tdvp, ct);
3672	}
3673
3674	tx = dmu_tx_create(zfsvfs->z_os);
3675	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
3676	dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE);
3677	dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm);
3678	dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm);
3679	if (sdzp != tdzp) {
3680		dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE);
3681		zfs_sa_upgrade_txholds(tx, tdzp);
3682	}
3683	if (tzp) {
3684		dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE);
3685		zfs_sa_upgrade_txholds(tx, tzp);
3686	}
3687
3688	zfs_sa_upgrade_txholds(tx, szp);
3689	dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
3690	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3691	if (error) {
3692		if (zl != NULL)
3693			zfs_rename_unlock(&zl);
3694		zfs_dirent_unlock(sdl);
3695		zfs_dirent_unlock(tdl);
3696
3697		if (sdzp == tdzp)
3698			rw_exit(&sdzp->z_name_lock);
3699
3700		VN_RELE(ZTOV(szp));
3701		if (tzp)
3702			VN_RELE(ZTOV(tzp));
3703		if (error == ERESTART) {
3704			waited = B_TRUE;
3705			dmu_tx_wait(tx);
3706			dmu_tx_abort(tx);
3707			goto top;
3708		}
3709		dmu_tx_abort(tx);
3710		ZFS_EXIT(zfsvfs);
3711		return (error);
3712	}
3713
3714	if (tzp)	/* Attempt to remove the existing target */
3715		error = zfs_link_destroy(tdl, tzp, tx, zflg, NULL);
3716
3717	if (error == 0) {
3718		error = zfs_link_create(tdl, szp, tx, ZRENAMING);
3719		if (error == 0) {
3720			szp->z_pflags |= ZFS_AV_MODIFIED;
3721
3722			error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs),
3723			    (void *)&szp->z_pflags, sizeof (uint64_t), tx);
3724			ASSERT0(error);
3725
3726			error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL);
3727			if (error == 0) {
3728				zfs_log_rename(zilog, tx, TX_RENAME |
3729				    (flags & FIGNORECASE ? TX_CI : 0), sdzp,
3730				    sdl->dl_name, tdzp, tdl->dl_name, szp);
3731
3732				/*
3733				 * Update path information for the target vnode
3734				 */
3735				vn_renamepath(tdvp, ZTOV(szp), tnm,
3736				    strlen(tnm));
3737			} else {
3738				/*
3739				 * At this point, we have successfully created
3740				 * the target name, but have failed to remove
3741				 * the source name.  Since the create was done
3742				 * with the ZRENAMING flag, there are
3743				 * complications; for one, the link count is
3744				 * wrong.  The easiest way to deal with this
3745				 * is to remove the newly created target, and
3746				 * return the original error.  This must
3747				 * succeed; fortunately, it is very unlikely to
3748				 * fail, since we just created it.
3749				 */
3750				VERIFY3U(zfs_link_destroy(tdl, szp, tx,
3751				    ZRENAMING, NULL), ==, 0);
3752			}
3753		}
3754	}
3755
3756	dmu_tx_commit(tx);
3757out:
3758	if (zl != NULL)
3759		zfs_rename_unlock(&zl);
3760
3761	zfs_dirent_unlock(sdl);
3762	zfs_dirent_unlock(tdl);
3763
3764	if (sdzp == tdzp)
3765		rw_exit(&sdzp->z_name_lock);
3766
3767
3768	VN_RELE(ZTOV(szp));
3769	if (tzp)
3770		VN_RELE(ZTOV(tzp));
3771
3772	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3773		zil_commit(zilog, 0);
3774
3775	ZFS_EXIT(zfsvfs);
3776	return (error);
3777}
3778
3779/*
3780 * Insert the indicated symbolic reference entry into the directory.
3781 *
3782 *	IN:	dvp	- Directory to contain new symbolic link.
3783 *		link	- Name for new symlink entry.
3784 *		vap	- Attributes of new entry.
3785 *		cr	- credentials of caller.
3786 *		ct	- caller context
3787 *		flags	- case flags
3788 *
3789 *	RETURN:	0 on success, error code on failure.
3790 *
3791 * Timestamps:
3792 *	dvp - ctime|mtime updated
3793 */
3794/*ARGSUSED*/
3795static int
3796zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr,
3797    caller_context_t *ct, int flags)
3798{
3799	znode_t		*zp, *dzp = VTOZ(dvp);
3800	zfs_dirlock_t	*dl;
3801	dmu_tx_t	*tx;
3802	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3803	zilog_t		*zilog;
3804	uint64_t	len = strlen(link);
3805	int		error;
3806	int		zflg = ZNEW;
3807	zfs_acl_ids_t	acl_ids;
3808	boolean_t	fuid_dirtied;
3809	uint64_t	txtype = TX_SYMLINK;
3810	boolean_t	waited = B_FALSE;
3811
3812	ASSERT(vap->va_type == VLNK);
3813
3814	ZFS_ENTER(zfsvfs);
3815	ZFS_VERIFY_ZP(dzp);
3816	zilog = zfsvfs->z_log;
3817
3818	if (zfsvfs->z_utf8 && u8_validate(name, strlen(name),
3819	    NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
3820		ZFS_EXIT(zfsvfs);
3821		return (SET_ERROR(EILSEQ));
3822	}
3823	if (flags & FIGNORECASE)
3824		zflg |= ZCILOOK;
3825
3826	if (len > MAXPATHLEN) {
3827		ZFS_EXIT(zfsvfs);
3828		return (SET_ERROR(ENAMETOOLONG));
3829	}
3830
3831	if ((error = zfs_acl_ids_create(dzp, 0,
3832	    vap, cr, NULL, &acl_ids)) != 0) {
3833		ZFS_EXIT(zfsvfs);
3834		return (error);
3835	}
3836top:
3837	/*
3838	 * Attempt to lock directory; fail if entry already exists.
3839	 */
3840	error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL);
3841	if (error) {
3842		zfs_acl_ids_free(&acl_ids);
3843		ZFS_EXIT(zfsvfs);
3844		return (error);
3845	}
3846
3847	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
3848		zfs_acl_ids_free(&acl_ids);
3849		zfs_dirent_unlock(dl);
3850		ZFS_EXIT(zfsvfs);
3851		return (error);
3852	}
3853
3854	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
3855		zfs_acl_ids_free(&acl_ids);
3856		zfs_dirent_unlock(dl);
3857		ZFS_EXIT(zfsvfs);
3858		return (SET_ERROR(EDQUOT));
3859	}
3860	tx = dmu_tx_create(zfsvfs->z_os);
3861	fuid_dirtied = zfsvfs->z_fuid_dirty;
3862	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len));
3863	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
3864	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
3865	    ZFS_SA_BASE_ATTR_SIZE + len);
3866	dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
3867	if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) {
3868		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
3869		    acl_ids.z_aclp->z_acl_bytes);
3870	}
3871	if (fuid_dirtied)
3872		zfs_fuid_txhold(zfsvfs, tx);
3873	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
3874	if (error) {
3875		zfs_dirent_unlock(dl);
3876		if (error == ERESTART) {
3877			waited = B_TRUE;
3878			dmu_tx_wait(tx);
3879			dmu_tx_abort(tx);
3880			goto top;
3881		}
3882		zfs_acl_ids_free(&acl_ids);
3883		dmu_tx_abort(tx);
3884		ZFS_EXIT(zfsvfs);
3885		return (error);
3886	}
3887
3888	/*
3889	 * Create a new object for the symlink.
3890	 * for version 4 ZPL datsets the symlink will be an SA attribute
3891	 */
3892	zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids);
3893
3894	if (fuid_dirtied)
3895		zfs_fuid_sync(zfsvfs, tx);
3896
3897	mutex_enter(&zp->z_lock);
3898	if (zp->z_is_sa)
3899		error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs),
3900		    link, len, tx);
3901	else
3902		zfs_sa_symlink(zp, link, len, tx);
3903	mutex_exit(&zp->z_lock);
3904
3905	zp->z_size = len;
3906	(void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs),
3907	    &zp->z_size, sizeof (zp->z_size), tx);
3908	/*
3909	 * Insert the new object into the directory.
3910	 */
3911	(void) zfs_link_create(dl, zp, tx, ZNEW);
3912
3913	if (flags & FIGNORECASE)
3914		txtype |= TX_CI;
3915	zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link);
3916
3917	zfs_acl_ids_free(&acl_ids);
3918
3919	dmu_tx_commit(tx);
3920
3921	zfs_dirent_unlock(dl);
3922
3923	VN_RELE(ZTOV(zp));
3924
3925	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
3926		zil_commit(zilog, 0);
3927
3928	ZFS_EXIT(zfsvfs);
3929	return (error);
3930}
3931
3932/*
3933 * Return, in the buffer contained in the provided uio structure,
3934 * the symbolic path referred to by vp.
3935 *
3936 *	IN:	vp	- vnode of symbolic link.
3937 *		uio	- structure to contain the link path.
3938 *		cr	- credentials of caller.
3939 *		ct	- caller context
3940 *
3941 *	OUT:	uio	- structure containing the link path.
3942 *
3943 *	RETURN:	0 on success, error code on failure.
3944 *
3945 * Timestamps:
3946 *	vp - atime updated
3947 */
3948/* ARGSUSED */
3949static int
3950zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct)
3951{
3952	znode_t		*zp = VTOZ(vp);
3953	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
3954	int		error;
3955
3956	ZFS_ENTER(zfsvfs);
3957	ZFS_VERIFY_ZP(zp);
3958
3959	mutex_enter(&zp->z_lock);
3960	if (zp->z_is_sa)
3961		error = sa_lookup_uio(zp->z_sa_hdl,
3962		    SA_ZPL_SYMLINK(zfsvfs), uio);
3963	else
3964		error = zfs_sa_readlink(zp, uio);
3965	mutex_exit(&zp->z_lock);
3966
3967	ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
3968
3969	ZFS_EXIT(zfsvfs);
3970	return (error);
3971}
3972
3973/*
3974 * Insert a new entry into directory tdvp referencing svp.
3975 *
3976 *	IN:	tdvp	- Directory to contain new entry.
3977 *		svp	- vnode of new entry.
3978 *		name	- name of new entry.
3979 *		cr	- credentials of caller.
3980 *		ct	- caller context
3981 *
3982 *	RETURN:	0 on success, error code on failure.
3983 *
3984 * Timestamps:
3985 *	tdvp - ctime|mtime updated
3986 *	 svp - ctime updated
3987 */
3988/* ARGSUSED */
3989static int
3990zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr,
3991    caller_context_t *ct, int flags)
3992{
3993	znode_t		*dzp = VTOZ(tdvp);
3994	znode_t		*tzp, *szp;
3995	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
3996	zilog_t		*zilog;
3997	zfs_dirlock_t	*dl;
3998	dmu_tx_t	*tx;
3999	vnode_t		*realvp;
4000	int		error;
4001	int		zf = ZNEW;
4002	uint64_t	parent;
4003	uid_t		owner;
4004	boolean_t	waited = B_FALSE;
4005
4006	ASSERT(tdvp->v_type == VDIR);
4007
4008	ZFS_ENTER(zfsvfs);
4009	ZFS_VERIFY_ZP(dzp);
4010	zilog = zfsvfs->z_log;
4011
4012	if (VOP_REALVP(svp, &realvp, ct) == 0)
4013		svp = realvp;
4014
4015	/*
4016	 * POSIX dictates that we return EPERM here.
4017	 * Better choices include ENOTSUP or EISDIR.
4018	 */
4019	if (svp->v_type == VDIR) {
4020		ZFS_EXIT(zfsvfs);
4021		return (SET_ERROR(EPERM));
4022	}
4023
4024	szp = VTOZ(svp);
4025	ZFS_VERIFY_ZP(szp);
4026
4027	/*
4028	 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4029	 * ctldir appear to have the same v_vfsp.
4030	 */
4031	if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) {
4032		ZFS_EXIT(zfsvfs);
4033		return (SET_ERROR(EXDEV));
4034	}
4035
4036	/* Prevent links to .zfs/shares files */
4037
4038	if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
4039	    &parent, sizeof (uint64_t))) != 0) {
4040		ZFS_EXIT(zfsvfs);
4041		return (error);
4042	}
4043	if (parent == zfsvfs->z_shares_dir) {
4044		ZFS_EXIT(zfsvfs);
4045		return (SET_ERROR(EPERM));
4046	}
4047
4048	if (zfsvfs->z_utf8 && u8_validate(name,
4049	    strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) {
4050		ZFS_EXIT(zfsvfs);
4051		return (SET_ERROR(EILSEQ));
4052	}
4053	if (flags & FIGNORECASE)
4054		zf |= ZCILOOK;
4055
4056	/*
4057	 * We do not support links between attributes and non-attributes
4058	 * because of the potential security risk of creating links
4059	 * into "normal" file space in order to circumvent restrictions
4060	 * imposed in attribute space.
4061	 */
4062	if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) {
4063		ZFS_EXIT(zfsvfs);
4064		return (SET_ERROR(EINVAL));
4065	}
4066
4067
4068	owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER);
4069	if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) {
4070		ZFS_EXIT(zfsvfs);
4071		return (SET_ERROR(EPERM));
4072	}
4073
4074	if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) {
4075		ZFS_EXIT(zfsvfs);
4076		return (error);
4077	}
4078
4079top:
4080	/*
4081	 * Attempt to lock directory; fail if entry already exists.
4082	 */
4083	error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL);
4084	if (error) {
4085		ZFS_EXIT(zfsvfs);
4086		return (error);
4087	}
4088
4089	tx = dmu_tx_create(zfsvfs->z_os);
4090	dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE);
4091	dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name);
4092	zfs_sa_upgrade_txholds(tx, szp);
4093	zfs_sa_upgrade_txholds(tx, dzp);
4094	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
4095	if (error) {
4096		zfs_dirent_unlock(dl);
4097		if (error == ERESTART) {
4098			waited = B_TRUE;
4099			dmu_tx_wait(tx);
4100			dmu_tx_abort(tx);
4101			goto top;
4102		}
4103		dmu_tx_abort(tx);
4104		ZFS_EXIT(zfsvfs);
4105		return (error);
4106	}
4107
4108	error = zfs_link_create(dl, szp, tx, 0);
4109
4110	if (error == 0) {
4111		uint64_t txtype = TX_LINK;
4112		if (flags & FIGNORECASE)
4113			txtype |= TX_CI;
4114		zfs_log_link(zilog, tx, txtype, dzp, szp, name);
4115	}
4116
4117	dmu_tx_commit(tx);
4118
4119	zfs_dirent_unlock(dl);
4120
4121	if (error == 0) {
4122		vnevent_link(svp, ct);
4123	}
4124
4125	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4126		zil_commit(zilog, 0);
4127
4128	ZFS_EXIT(zfsvfs);
4129	return (error);
4130}
4131
4132/*
4133 * zfs_null_putapage() is used when the file system has been force
4134 * unmounted. It just drops the pages.
4135 */
4136/* ARGSUSED */
4137static int
4138zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4139		size_t *lenp, int flags, cred_t *cr)
4140{
4141	pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR);
4142	return (0);
4143}
4144
4145/*
4146 * Push a page out to disk, klustering if possible.
4147 *
4148 *	IN:	vp	- file to push page to.
4149 *		pp	- page to push.
4150 *		flags	- additional flags.
4151 *		cr	- credentials of caller.
4152 *
4153 *	OUT:	offp	- start of range pushed.
4154 *		lenp	- len of range pushed.
4155 *
4156 *	RETURN:	0 on success, error code on failure.
4157 *
4158 * NOTE: callers must have locked the page to be pushed.  On
4159 * exit, the page (and all other pages in the kluster) must be
4160 * unlocked.
4161 */
4162/* ARGSUSED */
4163static int
4164zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
4165		size_t *lenp, int flags, cred_t *cr)
4166{
4167	znode_t		*zp = VTOZ(vp);
4168	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4169	dmu_tx_t	*tx;
4170	u_offset_t	off, koff;
4171	size_t		len, klen;
4172	int		err;
4173
4174	off = pp->p_offset;
4175	len = PAGESIZE;
4176	/*
4177	 * If our blocksize is bigger than the page size, try to kluster
4178	 * multiple pages so that we write a full block (thus avoiding
4179	 * a read-modify-write).
4180	 */
4181	if (off < zp->z_size && zp->z_blksz > PAGESIZE) {
4182		klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE);
4183		koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0;
4184		ASSERT(koff <= zp->z_size);
4185		if (koff + klen > zp->z_size)
4186			klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE);
4187		pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags);
4188	}
4189	ASSERT3U(btop(len), ==, btopr(len));
4190
4191	/*
4192	 * Can't push pages past end-of-file.
4193	 */
4194	if (off >= zp->z_size) {
4195		/* ignore all pages */
4196		err = 0;
4197		goto out;
4198	} else if (off + len > zp->z_size) {
4199		int npages = btopr(zp->z_size - off);
4200		page_t *trunc;
4201
4202		page_list_break(&pp, &trunc, npages);
4203		/* ignore pages past end of file */
4204		if (trunc)
4205			pvn_write_done(trunc, flags);
4206		len = zp->z_size - off;
4207	}
4208
4209	if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
4210	    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
4211		err = SET_ERROR(EDQUOT);
4212		goto out;
4213	}
4214	tx = dmu_tx_create(zfsvfs->z_os);
4215	dmu_tx_hold_write(tx, zp->z_id, off, len);
4216
4217	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4218	zfs_sa_upgrade_txholds(tx, zp);
4219	err = dmu_tx_assign(tx, TXG_WAIT);
4220	if (err != 0) {
4221		dmu_tx_abort(tx);
4222		goto out;
4223	}
4224
4225	if (zp->z_blksz <= PAGESIZE) {
4226		caddr_t va = zfs_map_page(pp, S_READ);
4227		ASSERT3U(len, <=, PAGESIZE);
4228		dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx);
4229		zfs_unmap_page(pp, va);
4230	} else {
4231		err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx);
4232	}
4233
4234	if (err == 0) {
4235		uint64_t mtime[2], ctime[2];
4236		sa_bulk_attr_t bulk[3];
4237		int count = 0;
4238
4239		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
4240		    &mtime, 16);
4241		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
4242		    &ctime, 16);
4243		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
4244		    &zp->z_pflags, 8);
4245		zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime,
4246		    B_TRUE);
4247		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
4248	}
4249	dmu_tx_commit(tx);
4250
4251out:
4252	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);
4253	if (offp)
4254		*offp = off;
4255	if (lenp)
4256		*lenp = len;
4257
4258	return (err);
4259}
4260
4261/*
4262 * Copy the portion of the file indicated from pages into the file.
4263 * The pages are stored in a page list attached to the files vnode.
4264 *
4265 *	IN:	vp	- vnode of file to push page data to.
4266 *		off	- position in file to put data.
4267 *		len	- amount of data to write.
4268 *		flags	- flags to control the operation.
4269 *		cr	- credentials of caller.
4270 *		ct	- caller context.
4271 *
4272 *	RETURN:	0 on success, error code on failure.
4273 *
4274 * Timestamps:
4275 *	vp - ctime|mtime updated
4276 */
4277/*ARGSUSED*/
4278static int
4279zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4280    caller_context_t *ct)
4281{
4282	znode_t		*zp = VTOZ(vp);
4283	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4284	page_t		*pp;
4285	size_t		io_len;
4286	u_offset_t	io_off;
4287	uint_t		blksz;
4288	rl_t		*rl;
4289	int		error = 0;
4290
4291	ZFS_ENTER(zfsvfs);
4292	ZFS_VERIFY_ZP(zp);
4293
4294	/*
4295	 * There's nothing to do if no data is cached.
4296	 */
4297	if (!vn_has_cached_data(vp)) {
4298		ZFS_EXIT(zfsvfs);
4299		return (0);
4300	}
4301
4302	/*
4303	 * Align this request to the file block size in case we kluster.
4304	 * XXX - this can result in pretty aggresive locking, which can
4305	 * impact simultanious read/write access.  One option might be
4306	 * to break up long requests (len == 0) into block-by-block
4307	 * operations to get narrower locking.
4308	 */
4309	blksz = zp->z_blksz;
4310	if (ISP2(blksz))
4311		io_off = P2ALIGN_TYPED(off, blksz, u_offset_t);
4312	else
4313		io_off = 0;
4314	if (len > 0 && ISP2(blksz))
4315		io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t);
4316	else
4317		io_len = 0;
4318
4319	if (io_len == 0) {
4320		/*
4321		 * Search the entire vp list for pages >= io_off.
4322		 */
4323		rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER);
4324		error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr);
4325		goto out;
4326	}
4327	rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER);
4328
4329	if (off > zp->z_size) {
4330		/* past end of file */
4331		zfs_range_unlock(rl);
4332		ZFS_EXIT(zfsvfs);
4333		return (0);
4334	}
4335
4336	len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off);
4337
4338	for (off = io_off; io_off < off + len; io_off += io_len) {
4339		if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
4340			pp = page_lookup(vp, io_off,
4341			    (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED);
4342		} else {
4343			pp = page_lookup_nowait(vp, io_off,
4344			    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
4345		}
4346
4347		if (pp != NULL && pvn_getdirty(pp, flags)) {
4348			int err;
4349
4350			/*
4351			 * Found a dirty page to push
4352			 */
4353			err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr);
4354			if (err)
4355				error = err;
4356		} else {
4357			io_len = PAGESIZE;
4358		}
4359	}
4360out:
4361	zfs_range_unlock(rl);
4362	if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
4363		zil_commit(zfsvfs->z_log, zp->z_id);
4364	ZFS_EXIT(zfsvfs);
4365	return (error);
4366}
4367
4368/*ARGSUSED*/
4369void
4370zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
4371{
4372	znode_t	*zp = VTOZ(vp);
4373	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4374	int error;
4375
4376	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER);
4377	if (zp->z_sa_hdl == NULL) {
4378		/*
4379		 * The fs has been unmounted, or we did a
4380		 * suspend/resume and this file no longer exists.
4381		 */
4382		if (vn_has_cached_data(vp)) {
4383			(void) pvn_vplist_dirty(vp, 0, zfs_null_putapage,
4384			    B_INVAL, cr);
4385		}
4386
4387		mutex_enter(&zp->z_lock);
4388		mutex_enter(&vp->v_lock);
4389		ASSERT(vp->v_count == 1);
4390		vp->v_count = 0;
4391		mutex_exit(&vp->v_lock);
4392		mutex_exit(&zp->z_lock);
4393		rw_exit(&zfsvfs->z_teardown_inactive_lock);
4394		zfs_znode_free(zp);
4395		return;
4396	}
4397
4398	/*
4399	 * Attempt to push any data in the page cache.  If this fails
4400	 * we will get kicked out later in zfs_zinactive().
4401	 */
4402	if (vn_has_cached_data(vp)) {
4403		(void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC,
4404		    cr);
4405	}
4406
4407	if (zp->z_atime_dirty && zp->z_unlinked == 0) {
4408		dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os);
4409
4410		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
4411		zfs_sa_upgrade_txholds(tx, zp);
4412		error = dmu_tx_assign(tx, TXG_WAIT);
4413		if (error) {
4414			dmu_tx_abort(tx);
4415		} else {
4416			mutex_enter(&zp->z_lock);
4417			(void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
4418			    (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
4419			zp->z_atime_dirty = 0;
4420			mutex_exit(&zp->z_lock);
4421			dmu_tx_commit(tx);
4422		}
4423	}
4424
4425	zfs_zinactive(zp);
4426	rw_exit(&zfsvfs->z_teardown_inactive_lock);
4427}
4428
4429/*
4430 * Bounds-check the seek operation.
4431 *
4432 *	IN:	vp	- vnode seeking within
4433 *		ooff	- old file offset
4434 *		noffp	- pointer to new file offset
4435 *		ct	- caller context
4436 *
4437 *	RETURN:	0 on success, EINVAL if new offset invalid.
4438 */
4439/* ARGSUSED */
4440static int
4441zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp,
4442    caller_context_t *ct)
4443{
4444	if (vp->v_type == VDIR)
4445		return (0);
4446	return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4447}
4448
4449/*
4450 * Pre-filter the generic locking function to trap attempts to place
4451 * a mandatory lock on a memory mapped file.
4452 */
4453static int
4454zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset,
4455    flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct)
4456{
4457	znode_t *zp = VTOZ(vp);
4458	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4459
4460	ZFS_ENTER(zfsvfs);
4461	ZFS_VERIFY_ZP(zp);
4462
4463	/*
4464	 * We are following the UFS semantics with respect to mapcnt
4465	 * here: If we see that the file is mapped already, then we will
4466	 * return an error, but we don't worry about races between this
4467	 * function and zfs_map().
4468	 */
4469	if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) {
4470		ZFS_EXIT(zfsvfs);
4471		return (SET_ERROR(EAGAIN));
4472	}
4473	ZFS_EXIT(zfsvfs);
4474	return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4475}
4476
4477/*
4478 * If we can't find a page in the cache, we will create a new page
4479 * and fill it with file data.  For efficiency, we may try to fill
4480 * multiple pages at once (klustering) to fill up the supplied page
4481 * list.  Note that the pages to be filled are held with an exclusive
4482 * lock to prevent access by other threads while they are being filled.
4483 */
4484static int
4485zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg,
4486    caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw)
4487{
4488	znode_t *zp = VTOZ(vp);
4489	page_t *pp, *cur_pp;
4490	objset_t *os = zp->z_zfsvfs->z_os;
4491	u_offset_t io_off, total;
4492	size_t io_len;
4493	int err;
4494
4495	if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) {
4496		/*
4497		 * We only have a single page, don't bother klustering
4498		 */
4499		io_off = off;
4500		io_len = PAGESIZE;
4501		pp = page_create_va(vp, io_off, io_len,
4502		    PG_EXCL | PG_WAIT, seg, addr);
4503	} else {
4504		/*
4505		 * Try to find enough pages to fill the page list
4506		 */
4507		pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4508		    &io_len, off, plsz, 0);
4509	}
4510	if (pp == NULL) {
4511		/*
4512		 * The page already exists, nothing to do here.
4513		 */
4514		*pl = NULL;
4515		return (0);
4516	}
4517
4518	/*
4519	 * Fill the pages in the kluster.
4520	 */
4521	cur_pp = pp;
4522	for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) {
4523		caddr_t va;
4524
4525		ASSERT3U(io_off, ==, cur_pp->p_offset);
4526		va = zfs_map_page(cur_pp, S_WRITE);
4527		err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va,
4528		    DMU_READ_PREFETCH);
4529		zfs_unmap_page(cur_pp, va);
4530		if (err) {
4531			/* On error, toss the entire kluster */
4532			pvn_read_done(pp, B_ERROR);
4533			/* convert checksum errors into IO errors */
4534			if (err == ECKSUM)
4535				err = SET_ERROR(EIO);
4536			return (err);
4537		}
4538		cur_pp = cur_pp->p_next;
4539	}
4540
4541	/*
4542	 * Fill in the page list array from the kluster starting
4543	 * from the desired offset `off'.
4544	 * NOTE: the page list will always be null terminated.
4545	 */
4546	pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4547	ASSERT(pl == NULL || (*pl)->p_offset == off);
4548
4549	return (0);
4550}
4551
4552/*
4553 * Return pointers to the pages for the file region [off, off + len]
4554 * in the pl array.  If plsz is greater than len, this function may
4555 * also return page pointers from after the specified region
4556 * (i.e. the region [off, off + plsz]).  These additional pages are
4557 * only returned if they are already in the cache, or were created as
4558 * part of a klustered read.
4559 *
4560 *	IN:	vp	- vnode of file to get data from.
4561 *		off	- position in file to get data from.
4562 *		len	- amount of data to retrieve.
4563 *		plsz	- length of provided page list.
4564 *		seg	- segment to obtain pages for.
4565 *		addr	- virtual address of fault.
4566 *		rw	- mode of created pages.
4567 *		cr	- credentials of caller.
4568 *		ct	- caller context.
4569 *
4570 *	OUT:	protp	- protection mode of created pages.
4571 *		pl	- list of pages created.
4572 *
4573 *	RETURN:	0 on success, error code on failure.
4574 *
4575 * Timestamps:
4576 *	vp - atime updated
4577 */
4578/* ARGSUSED */
4579static int
4580zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
4581    page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
4582    enum seg_rw rw, cred_t *cr, caller_context_t *ct)
4583{
4584	znode_t		*zp = VTOZ(vp);
4585	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4586	page_t		**pl0 = pl;
4587	int		err = 0;
4588
4589	/* we do our own caching, faultahead is unnecessary */
4590	if (pl == NULL)
4591		return (0);
4592	else if (len > plsz)
4593		len = plsz;
4594	else
4595		len = P2ROUNDUP(len, PAGESIZE);
4596	ASSERT(plsz >= len);
4597
4598	ZFS_ENTER(zfsvfs);
4599	ZFS_VERIFY_ZP(zp);
4600
4601	if (protp)
4602		*protp = PROT_ALL;
4603
4604	/*
4605	 * Loop through the requested range [off, off + len) looking
4606	 * for pages.  If we don't find a page, we will need to create
4607	 * a new page and fill it with data from the file.
4608	 */
4609	while (len > 0) {
4610		if (*pl = page_lookup(vp, off, SE_SHARED))
4611			*(pl+1) = NULL;
4612		else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw))
4613			goto out;
4614		while (*pl) {
4615			ASSERT3U((*pl)->p_offset, ==, off);
4616			off += PAGESIZE;
4617			addr += PAGESIZE;
4618			if (len > 0) {
4619				ASSERT3U(len, >=, PAGESIZE);
4620				len -= PAGESIZE;
4621			}
4622			ASSERT3U(plsz, >=, PAGESIZE);
4623			plsz -= PAGESIZE;
4624			pl++;
4625		}
4626	}
4627
4628	/*
4629	 * Fill out the page array with any pages already in the cache.
4630	 */
4631	while (plsz > 0 &&
4632	    (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) {
4633			off += PAGESIZE;
4634			plsz -= PAGESIZE;
4635	}
4636out:
4637	if (err) {
4638		/*
4639		 * Release any pages we have previously locked.
4640		 */
4641		while (pl > pl0)
4642			page_unlock(*--pl);
4643	} else {
4644		ZFS_ACCESSTIME_STAMP(zfsvfs, zp);
4645	}
4646
4647	*pl = NULL;
4648
4649	ZFS_EXIT(zfsvfs);
4650	return (err);
4651}
4652
4653/*
4654 * Request a memory map for a section of a file.  This code interacts
4655 * with common code and the VM system as follows:
4656 *
4657 * - common code calls mmap(), which ends up in smmap_common()
4658 * - this calls VOP_MAP(), which takes you into (say) zfs
4659 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4660 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4661 * - zfs_addmap() updates z_mapcnt
4662 */
4663/*ARGSUSED*/
4664static int
4665zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4666    size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4667    caller_context_t *ct)
4668{
4669	znode_t *zp = VTOZ(vp);
4670	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4671	segvn_crargs_t	vn_a;
4672	int		error;
4673
4674	ZFS_ENTER(zfsvfs);
4675	ZFS_VERIFY_ZP(zp);
4676
4677	if ((prot & PROT_WRITE) && (zp->z_pflags &
4678	    (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) {
4679		ZFS_EXIT(zfsvfs);
4680		return (SET_ERROR(EPERM));
4681	}
4682
4683	if ((prot & (PROT_READ | PROT_EXEC)) &&
4684	    (zp->z_pflags & ZFS_AV_QUARANTINED)) {
4685		ZFS_EXIT(zfsvfs);
4686		return (SET_ERROR(EACCES));
4687	}
4688
4689	if (vp->v_flag & VNOMAP) {
4690		ZFS_EXIT(zfsvfs);
4691		return (SET_ERROR(ENOSYS));
4692	}
4693
4694	if (off < 0 || len > MAXOFFSET_T - off) {
4695		ZFS_EXIT(zfsvfs);
4696		return (SET_ERROR(ENXIO));
4697	}
4698
4699	if (vp->v_type != VREG) {
4700		ZFS_EXIT(zfsvfs);
4701		return (SET_ERROR(ENODEV));
4702	}
4703
4704	/*
4705	 * If file is locked, disallow mapping.
4706	 */
4707	if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) {
4708		ZFS_EXIT(zfsvfs);
4709		return (SET_ERROR(EAGAIN));
4710	}
4711
4712	as_rangelock(as);
4713	error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4714	if (error != 0) {
4715		as_rangeunlock(as);
4716		ZFS_EXIT(zfsvfs);
4717		return (error);
4718	}
4719
4720	vn_a.vp = vp;
4721	vn_a.offset = (u_offset_t)off;
4722	vn_a.type = flags & MAP_TYPE;
4723	vn_a.prot = prot;
4724	vn_a.maxprot = maxprot;
4725	vn_a.cred = cr;
4726	vn_a.amp = NULL;
4727	vn_a.flags = flags & ~MAP_TYPE;
4728	vn_a.szc = 0;
4729	vn_a.lgrp_mem_policy_flags = 0;
4730
4731	error = as_map(as, *addrp, len, segvn_create, &vn_a);
4732
4733	as_rangeunlock(as);
4734	ZFS_EXIT(zfsvfs);
4735	return (error);
4736}
4737
4738/* ARGSUSED */
4739static int
4740zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4741    size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4742    caller_context_t *ct)
4743{
4744	uint64_t pages = btopr(len);
4745
4746	atomic_add_64(&VTOZ(vp)->z_mapcnt, pages);
4747	return (0);
4748}
4749
4750/*
4751 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4752 * more accurate mtime for the associated file.  Since we don't have a way of
4753 * detecting when the data was actually modified, we have to resort to
4754 * heuristics.  If an explicit msync() is done, then we mark the mtime when the
4755 * last page is pushed.  The problem occurs when the msync() call is omitted,
4756 * which by far the most common case:
4757 *
4758 *	open()
4759 *	mmap()
4760 *	<modify memory>
4761 *	munmap()
4762 *	close()
4763 *	<time lapse>
4764 *	putpage() via fsflush
4765 *
4766 * If we wait until fsflush to come along, we can have a modification time that
4767 * is some arbitrary point in the future.  In order to prevent this in the
4768 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4769 * torn down.
4770 */
4771/* ARGSUSED */
4772static int
4773zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4774    size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4775    caller_context_t *ct)
4776{
4777	uint64_t pages = btopr(len);
4778
4779	ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages);
4780	atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages);
4781
4782	if ((flags & MAP_SHARED) && (prot & PROT_WRITE) &&
4783	    vn_has_cached_data(vp))
4784		(void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct);
4785
4786	return (0);
4787}
4788
4789/*
4790 * Free or allocate space in a file.  Currently, this function only
4791 * supports the `F_FREESP' command.  However, this command is somewhat
4792 * misnamed, as its functionality includes the ability to allocate as
4793 * well as free space.
4794 *
4795 *	IN:	vp	- vnode of file to free data in.
4796 *		cmd	- action to take (only F_FREESP supported).
4797 *		bfp	- section of file to free/alloc.
4798 *		flag	- current file open mode flags.
4799 *		offset	- current file offset.
4800 *		cr	- credentials of caller [UNUSED].
4801 *		ct	- caller context.
4802 *
4803 *	RETURN:	0 on success, error code on failure.
4804 *
4805 * Timestamps:
4806 *	vp - ctime|mtime updated
4807 */
4808/* ARGSUSED */
4809static int
4810zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag,
4811    offset_t offset, cred_t *cr, caller_context_t *ct)
4812{
4813	znode_t		*zp = VTOZ(vp);
4814	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4815	uint64_t	off, len;
4816	int		error;
4817
4818	ZFS_ENTER(zfsvfs);
4819	ZFS_VERIFY_ZP(zp);
4820
4821	if (cmd != F_FREESP) {
4822		ZFS_EXIT(zfsvfs);
4823		return (SET_ERROR(EINVAL));
4824	}
4825
4826	/*
4827	 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
4828	 * callers might not be able to detect properly that we are read-only,
4829	 * so check it explicitly here.
4830	 */
4831	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
4832		ZFS_EXIT(zfsvfs);
4833		return (SET_ERROR(EROFS));
4834	}
4835
4836	if (error = convoff(vp, bfp, 0, offset)) {
4837		ZFS_EXIT(zfsvfs);
4838		return (error);
4839	}
4840
4841	if (bfp->l_len < 0) {
4842		ZFS_EXIT(zfsvfs);
4843		return (SET_ERROR(EINVAL));
4844	}
4845
4846	off = bfp->l_start;
4847	len = bfp->l_len; /* 0 means from off to end of file */
4848
4849	error = zfs_freesp(zp, off, len, flag, TRUE);
4850
4851	if (error == 0 && off == 0 && len == 0)
4852		vnevent_truncate(ZTOV(zp), ct);
4853
4854	ZFS_EXIT(zfsvfs);
4855	return (error);
4856}
4857
4858/*ARGSUSED*/
4859static int
4860zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
4861{
4862	znode_t		*zp = VTOZ(vp);
4863	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
4864	uint32_t	gen;
4865	uint64_t	gen64;
4866	uint64_t	object = zp->z_id;
4867	zfid_short_t	*zfid;
4868	int		size, i, error;
4869
4870	ZFS_ENTER(zfsvfs);
4871	ZFS_VERIFY_ZP(zp);
4872
4873	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
4874	    &gen64, sizeof (uint64_t))) != 0) {
4875		ZFS_EXIT(zfsvfs);
4876		return (error);
4877	}
4878
4879	gen = (uint32_t)gen64;
4880
4881	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
4882	if (fidp->fid_len < size) {
4883		fidp->fid_len = size;
4884		ZFS_EXIT(zfsvfs);
4885		return (SET_ERROR(ENOSPC));
4886	}
4887
4888	zfid = (zfid_short_t *)fidp;
4889
4890	zfid->zf_len = size;
4891
4892	for (i = 0; i < sizeof (zfid->zf_object); i++)
4893		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
4894
4895	/* Must have a non-zero generation number to distinguish from .zfs */
4896	if (gen == 0)
4897		gen = 1;
4898	for (i = 0; i < sizeof (zfid->zf_gen); i++)
4899		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
4900
4901	if (size == LONG_FID_LEN) {
4902		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
4903		zfid_long_t	*zlfid;
4904
4905		zlfid = (zfid_long_t *)fidp;
4906
4907		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
4908			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
4909
4910		/* XXX - this should be the generation number for the objset */
4911		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
4912			zlfid->zf_setgen[i] = 0;
4913	}
4914
4915	ZFS_EXIT(zfsvfs);
4916	return (0);
4917}
4918
4919static int
4920zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4921    caller_context_t *ct)
4922{
4923	znode_t		*zp, *xzp;
4924	zfsvfs_t	*zfsvfs;
4925	zfs_dirlock_t	*dl;
4926	int		error;
4927
4928	switch (cmd) {
4929	case _PC_LINK_MAX:
4930		*valp = ULONG_MAX;
4931		return (0);
4932
4933	case _PC_FILESIZEBITS:
4934		*valp = 64;
4935		return (0);
4936
4937	case _PC_XATTR_EXISTS:
4938		zp = VTOZ(vp);
4939		zfsvfs = zp->z_zfsvfs;
4940		ZFS_ENTER(zfsvfs);
4941		ZFS_VERIFY_ZP(zp);
4942		*valp = 0;
4943		error = zfs_dirent_lock(&dl, zp, "", &xzp,
4944		    ZXATTR | ZEXISTS | ZSHARED, NULL, NULL);
4945		if (error == 0) {
4946			zfs_dirent_unlock(dl);
4947			if (!zfs_dirempty(xzp))
4948				*valp = 1;
4949			VN_RELE(ZTOV(xzp));
4950		} else if (error == ENOENT) {
4951			/*
4952			 * If there aren't extended attributes, it's the
4953			 * same as having zero of them.
4954			 */
4955			error = 0;
4956		}
4957		ZFS_EXIT(zfsvfs);
4958		return (error);
4959
4960	case _PC_SATTR_ENABLED:
4961	case _PC_SATTR_EXISTS:
4962		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
4963		    (vp->v_type == VREG || vp->v_type == VDIR);
4964		return (0);
4965
4966	case _PC_ACCESS_FILTERING:
4967		*valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) &&
4968		    vp->v_type == VDIR;
4969		return (0);
4970
4971	case _PC_ACL_ENABLED:
4972		*valp = _ACL_ACE_ENABLED;
4973		return (0);
4974
4975	case _PC_MIN_HOLE_SIZE:
4976		*valp = (ulong_t)SPA_MINBLOCKSIZE;
4977		return (0);
4978
4979	case _PC_TIMESTAMP_RESOLUTION:
4980		/* nanosecond timestamp resolution */
4981		*valp = 1L;
4982		return (0);
4983
4984	default:
4985		return (fs_pathconf(vp, cmd, valp, cr, ct));
4986	}
4987}
4988
4989/*ARGSUSED*/
4990static int
4991zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
4992    caller_context_t *ct)
4993{
4994	znode_t *zp = VTOZ(vp);
4995	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
4996	int error;
4997	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
4998
4999	ZFS_ENTER(zfsvfs);
5000	ZFS_VERIFY_ZP(zp);
5001	error = zfs_getacl(zp, vsecp, skipaclchk, cr);
5002	ZFS_EXIT(zfsvfs);
5003
5004	return (error);
5005}
5006
5007/*ARGSUSED*/
5008static int
5009zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr,
5010    caller_context_t *ct)
5011{
5012	znode_t *zp = VTOZ(vp);
5013	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5014	int error;
5015	boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE;
5016	zilog_t	*zilog = zfsvfs->z_log;
5017
5018	ZFS_ENTER(zfsvfs);
5019	ZFS_VERIFY_ZP(zp);
5020
5021	error = zfs_setacl(zp, vsecp, skipaclchk, cr);
5022
5023	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
5024		zil_commit(zilog, 0);
5025
5026	ZFS_EXIT(zfsvfs);
5027	return (error);
5028}
5029
5030/*
5031 * The smallest read we may consider to loan out an arcbuf.
5032 * This must be a power of 2.
5033 */
5034int zcr_blksz_min = (1 << 10);	/* 1K */
5035/*
5036 * If set to less than the file block size, allow loaning out of an
5037 * arcbuf for a partial block read.  This must be a power of 2.
5038 */
5039int zcr_blksz_max = (1 << 17);	/* 128K */
5040
5041/*ARGSUSED*/
5042static int
5043zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr,
5044    caller_context_t *ct)
5045{
5046	znode_t	*zp = VTOZ(vp);
5047	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
5048	int max_blksz = zfsvfs->z_max_blksz;
5049	uio_t *uio = &xuio->xu_uio;
5050	ssize_t size = uio->uio_resid;
5051	offset_t offset = uio->uio_loffset;
5052	int blksz;
5053	int fullblk, i;
5054	arc_buf_t *abuf;
5055	ssize_t maxsize;
5056	int preamble, postamble;
5057
5058	if (xuio->xu_type != UIOTYPE_ZEROCOPY)
5059		return (SET_ERROR(EINVAL));
5060
5061	ZFS_ENTER(zfsvfs);
5062	ZFS_VERIFY_ZP(zp);
5063	switch (ioflag) {
5064	case UIO_WRITE:
5065		/*
5066		 * Loan out an arc_buf for write if write size is bigger than
5067		 * max_blksz, and the file's block size is also max_blksz.
5068		 */
5069		blksz = max_blksz;
5070		if (size < blksz || zp->z_blksz != blksz) {
5071			ZFS_EXIT(zfsvfs);
5072			return (SET_ERROR(EINVAL));
5073		}
5074		/*
5075		 * Caller requests buffers for write before knowing where the
5076		 * write offset might be (e.g. NFS TCP write).
5077		 */
5078		if (offset == -1) {
5079			preamble = 0;
5080		} else {
5081			preamble = P2PHASE(offset, blksz);
5082			if (preamble) {
5083				preamble = blksz - preamble;
5084				size -= preamble;
5085			}
5086		}
5087
5088		postamble = P2PHASE(size, blksz);
5089		size -= postamble;
5090
5091		fullblk = size / blksz;
5092		(void) dmu_xuio_init(xuio,
5093		    (preamble != 0) + fullblk + (postamble != 0));
5094		DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble,
5095		    int, postamble, int,
5096		    (preamble != 0) + fullblk + (postamble != 0));
5097
5098		/*
5099		 * Have to fix iov base/len for partial buffers.  They
5100		 * currently represent full arc_buf's.
5101		 */
5102		if (preamble) {
5103			/* data begins in the middle of the arc_buf */
5104			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5105			    blksz);
5106			ASSERT(abuf);
5107			(void) dmu_xuio_add(xuio, abuf,
5108			    blksz - preamble, preamble);
5109		}
5110
5111		for (i = 0; i < fullblk; i++) {
5112			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5113			    blksz);
5114			ASSERT(abuf);
5115			(void) dmu_xuio_add(xuio, abuf, 0, blksz);
5116		}
5117
5118		if (postamble) {
5119			/* data ends in the middle of the arc_buf */
5120			abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl),
5121			    blksz);
5122			ASSERT(abuf);
5123			(void) dmu_xuio_add(xuio, abuf, 0, postamble);
5124		}
5125		break;
5126	case UIO_READ:
5127		/*
5128		 * Loan out an arc_buf for read if the read size is larger than
5129		 * the current file block size.  Block alignment is not
5130		 * considered.  Partial arc_buf will be loaned out for read.
5131		 */
5132		blksz = zp->z_blksz;
5133		if (blksz < zcr_blksz_min)
5134			blksz = zcr_blksz_min;
5135		if (blksz > zcr_blksz_max)
5136			blksz = zcr_blksz_max;
5137		/* avoid potential complexity of dealing with it */
5138		if (blksz > max_blksz) {
5139			ZFS_EXIT(zfsvfs);
5140			return (SET_ERROR(EINVAL));
5141		}
5142
5143		maxsize = zp->z_size - uio->uio_loffset;
5144		if (size > maxsize)
5145			size = maxsize;
5146
5147		if (size < blksz || vn_has_cached_data(vp)) {
5148			ZFS_EXIT(zfsvfs);
5149			return (SET_ERROR(EINVAL));
5150		}
5151		break;
5152	default:
5153		ZFS_EXIT(zfsvfs);
5154		return (SET_ERROR(EINVAL));
5155	}
5156
5157	uio->uio_extflg = UIO_XUIO;
5158	XUIO_XUZC_RW(xuio) = ioflag;
5159	ZFS_EXIT(zfsvfs);
5160	return (0);
5161}
5162
5163/*ARGSUSED*/
5164static int
5165zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct)
5166{
5167	int i;
5168	arc_buf_t *abuf;
5169	int ioflag = XUIO_XUZC_RW(xuio);
5170
5171	ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY);
5172
5173	i = dmu_xuio_cnt(xuio);
5174	while (i-- > 0) {
5175		abuf = dmu_xuio_arcbuf(xuio, i);
5176		/*
5177		 * if abuf == NULL, it must be a write buffer
5178		 * that has been returned in zfs_write().
5179		 */
5180		if (abuf)
5181			dmu_return_arcbuf(abuf);
5182		ASSERT(abuf || ioflag == UIO_WRITE);
5183	}
5184
5185	dmu_xuio_fini(xuio);
5186	return (0);
5187}
5188
5189/*
5190 * Predeclare these here so that the compiler assumes that
5191 * this is an "old style" function declaration that does
5192 * not include arguments => we won't get type mismatch errors
5193 * in the initializations that follow.
5194 */
5195static int zfs_inval();
5196static int zfs_isdir();
5197
5198static int
5199zfs_inval()
5200{
5201	return (SET_ERROR(EINVAL));
5202}
5203
5204static int
5205zfs_isdir()
5206{
5207	return (SET_ERROR(EISDIR));
5208}
5209/*
5210 * Directory vnode operations template
5211 */
5212vnodeops_t *zfs_dvnodeops;
5213const fs_operation_def_t zfs_dvnodeops_template[] = {
5214	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5215	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5216	VOPNAME_READ,		{ .error = zfs_isdir },
5217	VOPNAME_WRITE,		{ .error = zfs_isdir },
5218	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5219	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5220	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5221	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5222	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5223	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5224	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5225	VOPNAME_LINK,		{ .vop_link = zfs_link },
5226	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5227	VOPNAME_MKDIR,		{ .vop_mkdir = zfs_mkdir },
5228	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5229	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5230	VOPNAME_SYMLINK,	{ .vop_symlink = zfs_symlink },
5231	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5232	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5233	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5234	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5235	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5236	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5237	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5238	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5239	NULL,			NULL
5240};
5241
5242/*
5243 * Regular file vnode operations template
5244 */
5245vnodeops_t *zfs_fvnodeops;
5246const fs_operation_def_t zfs_fvnodeops_template[] = {
5247	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5248	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5249	VOPNAME_READ,		{ .vop_read = zfs_read },
5250	VOPNAME_WRITE,		{ .vop_write = zfs_write },
5251	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5252	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5253	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5254	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5255	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5256	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5257	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5258	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5259	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5260	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5261	VOPNAME_FRLOCK,		{ .vop_frlock = zfs_frlock },
5262	VOPNAME_SPACE,		{ .vop_space = zfs_space },
5263	VOPNAME_GETPAGE,	{ .vop_getpage = zfs_getpage },
5264	VOPNAME_PUTPAGE,	{ .vop_putpage = zfs_putpage },
5265	VOPNAME_MAP,		{ .vop_map = zfs_map },
5266	VOPNAME_ADDMAP,		{ .vop_addmap = zfs_addmap },
5267	VOPNAME_DELMAP,		{ .vop_delmap = zfs_delmap },
5268	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5269	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5270	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5271	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5272	VOPNAME_REQZCBUF,	{ .vop_reqzcbuf = zfs_reqzcbuf },
5273	VOPNAME_RETZCBUF,	{ .vop_retzcbuf = zfs_retzcbuf },
5274	NULL,			NULL
5275};
5276
5277/*
5278 * Symbolic link vnode operations template
5279 */
5280vnodeops_t *zfs_symvnodeops;
5281const fs_operation_def_t zfs_symvnodeops_template[] = {
5282	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5283	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5284	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5285	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5286	VOPNAME_READLINK,	{ .vop_readlink = zfs_readlink },
5287	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5288	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5289	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5290	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5291	NULL,			NULL
5292};
5293
5294/*
5295 * special share hidden files vnode operations template
5296 */
5297vnodeops_t *zfs_sharevnodeops;
5298const fs_operation_def_t zfs_sharevnodeops_template[] = {
5299	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5300	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5301	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5302	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5303	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5304	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5305	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5306	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5307	NULL,			NULL
5308};
5309
5310/*
5311 * Extended attribute directory vnode operations template
5312 *
5313 * This template is identical to the directory vnodes
5314 * operation template except for restricted operations:
5315 *	VOP_MKDIR()
5316 *	VOP_SYMLINK()
5317 *
5318 * Note that there are other restrictions embedded in:
5319 *	zfs_create()	- restrict type to VREG
5320 *	zfs_link()	- no links into/out of attribute space
5321 *	zfs_rename()	- no moves into/out of attribute space
5322 */
5323vnodeops_t *zfs_xdvnodeops;
5324const fs_operation_def_t zfs_xdvnodeops_template[] = {
5325	VOPNAME_OPEN,		{ .vop_open = zfs_open },
5326	VOPNAME_CLOSE,		{ .vop_close = zfs_close },
5327	VOPNAME_IOCTL,		{ .vop_ioctl = zfs_ioctl },
5328	VOPNAME_GETATTR,	{ .vop_getattr = zfs_getattr },
5329	VOPNAME_SETATTR,	{ .vop_setattr = zfs_setattr },
5330	VOPNAME_ACCESS,		{ .vop_access = zfs_access },
5331	VOPNAME_LOOKUP,		{ .vop_lookup = zfs_lookup },
5332	VOPNAME_CREATE,		{ .vop_create = zfs_create },
5333	VOPNAME_REMOVE,		{ .vop_remove = zfs_remove },
5334	VOPNAME_LINK,		{ .vop_link = zfs_link },
5335	VOPNAME_RENAME,		{ .vop_rename = zfs_rename },
5336	VOPNAME_MKDIR,		{ .error = zfs_inval },
5337	VOPNAME_RMDIR,		{ .vop_rmdir = zfs_rmdir },
5338	VOPNAME_READDIR,	{ .vop_readdir = zfs_readdir },
5339	VOPNAME_SYMLINK,	{ .error = zfs_inval },
5340	VOPNAME_FSYNC,		{ .vop_fsync = zfs_fsync },
5341	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5342	VOPNAME_FID,		{ .vop_fid = zfs_fid },
5343	VOPNAME_SEEK,		{ .vop_seek = zfs_seek },
5344	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5345	VOPNAME_GETSECATTR,	{ .vop_getsecattr = zfs_getsecattr },
5346	VOPNAME_SETSECATTR,	{ .vop_setsecattr = zfs_setsecattr },
5347	VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
5348	NULL,			NULL
5349};
5350
5351/*
5352 * Error vnode operations template
5353 */
5354vnodeops_t *zfs_evnodeops;
5355const fs_operation_def_t zfs_evnodeops_template[] = {
5356	VOPNAME_INACTIVE,	{ .vop_inactive = zfs_inactive },
5357	VOPNAME_PATHCONF,	{ .vop_pathconf = zfs_pathconf },
5358	NULL,			NULL
5359};
5360