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.
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.
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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
26 * Copyright 2017 Nexenta Systems, Inc.
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
33 #include <sys/types.h>
34 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/resource.h>
40 #include <sys/vfs_opreg.h>
44 #include <sys/taskq.h>
46 #include <sys/vmsystm.h>
47 #include <sys/atomic.h>
49 #include <sys/pathname.h>
50 #include <sys/cmn_err.h>
51 #include <sys/errno.h>
52 #include <sys/unistd.h>
53 #include <sys/zfs_dir.h>
54 #include <sys/zfs_acl.h>
55 #include <sys/zfs_ioctl.h>
56 #include <sys/fs/zfs.h>
58 #include <sys/dmu_objset.h>
64 #include <sys/dirent.h>
65 #include <sys/policy.h>
66 #include <sys/sunddi.h>
69 #include "fs/fs_subr.h"
70 #include <sys/zfs_ctldir.h>
71 #include <sys/zfs_fuid.h>
72 #include <sys/zfs_sa.h>
73 #include <sys/zfs_vnops.h>
75 #include <sys/zfs_rlock.h>
76 #include <sys/extdirent.h>
77 #include <sys/kidmap.h>
86 * Each vnode op performs some logical unit of work. To do this, the ZPL must
87 * properly lock its in-core state, create a DMU transaction, do the work,
88 * record this work in the intent log (ZIL), commit the DMU transaction,
89 * and wait for the intent log to commit if it is a synchronous operation.
90 * Moreover, the vnode ops must work in both normal and log replay context.
91 * The ordering of events is important to avoid deadlocks and references
92 * to freed memory. The example below illustrates the following Big Rules:
94 * (1) A check must be made in each zfs thread for a mounted file system.
95 * This is done avoiding races using ZFS_ENTER(zfsvfs).
96 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
97 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
98 * can return EIO from the calling function.
100 * (2) iput() should always be the last thing except for zil_commit()
101 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
102 * First, if it's the last reference, the vnode/znode
103 * can be freed, so the zp may point to freed memory. Second, the last
104 * reference will call zfs_zinactive(), which may induce a lot of work --
105 * pushing cached pages (which acquires range locks) and syncing out
106 * cached atime changes. Third, zfs_zinactive() may require a new tx,
107 * which could deadlock the system if you were already holding one.
108 * If you must call iput() within a tx then use zfs_iput_async().
110 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
111 * as they can span dmu_tx_assign() calls.
113 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
114 * dmu_tx_assign(). This is critical because we don't want to block
115 * while holding locks.
117 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
118 * reduces lock contention and CPU usage when we must wait (note that if
119 * throughput is constrained by the storage, nearly every transaction
122 * Note, in particular, that if a lock is sometimes acquired before
123 * the tx assigns, and sometimes after (e.g. z_lock), then failing
124 * to use a non-blocking assign can deadlock the system. The scenario:
126 * Thread A has grabbed a lock before calling dmu_tx_assign().
127 * Thread B is in an already-assigned tx, and blocks for this lock.
128 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
129 * forever, because the previous txg can't quiesce until B's tx commits.
131 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
132 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
133 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT,
134 * to indicate that this operation has already called dmu_tx_wait().
135 * This will ensure that we don't retry forever, waiting a short bit
138 * (5) If the operation succeeded, generate the intent log entry for it
139 * before dropping locks. This ensures that the ordering of events
140 * in the intent log matches the order in which they actually occurred.
141 * During ZIL replay the zfs_log_* functions will update the sequence
142 * number to indicate the zil transaction has replayed.
144 * (6) At the end of each vnode op, the DMU tx must always commit,
145 * regardless of whether there were any errors.
147 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
148 * to ensure that synchronous semantics are provided when necessary.
150 * In general, this is how things should be ordered in each vnode op:
152 * ZFS_ENTER(zfsvfs); // exit if unmounted
154 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
155 * rw_enter(...); // grab any other locks you need
156 * tx = dmu_tx_create(...); // get DMU tx
157 * dmu_tx_hold_*(); // hold each object you might modify
158 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
160 * rw_exit(...); // drop locks
161 * zfs_dirent_unlock(dl); // unlock directory entry
162 * iput(...); // release held vnodes
163 * if (error == ERESTART) {
169 * dmu_tx_abort(tx); // abort DMU tx
170 * ZFS_EXIT(zfsvfs); // finished in zfs
171 * return (error); // really out of space
173 * error = do_real_work(); // do whatever this VOP does
175 * zfs_log_*(...); // on success, make ZIL entry
176 * dmu_tx_commit(tx); // commit DMU tx -- error or not
177 * rw_exit(...); // drop locks
178 * zfs_dirent_unlock(dl); // unlock directory entry
179 * iput(...); // release held vnodes
180 * zil_commit(zilog, foid); // synchronous when necessary
181 * ZFS_EXIT(zfsvfs); // finished in zfs
182 * return (error); // done, report error
186 * Virus scanning is unsupported. It would be possible to add a hook
187 * here to performance the required virus scan. This could be done
188 * entirely in the kernel or potentially as an update to invoke a
192 zfs_vscan(struct inode
*ip
, cred_t
*cr
, int async
)
199 zfs_open(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
201 znode_t
*zp
= ITOZ(ip
);
202 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
207 /* Honor ZFS_APPENDONLY file attribute */
208 if ((mode
& FMODE_WRITE
) && (zp
->z_pflags
& ZFS_APPENDONLY
) &&
209 ((flag
& O_APPEND
) == 0)) {
211 return (SET_ERROR(EPERM
));
214 /* Virus scan eligible files on open */
215 if (!zfs_has_ctldir(zp
) && zfsvfs
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
216 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0) {
217 if (zfs_vscan(ip
, cr
, 0) != 0) {
219 return (SET_ERROR(EACCES
));
223 /* Keep a count of the synchronous opens in the znode */
225 atomic_inc_32(&zp
->z_sync_cnt
);
233 zfs_close(struct inode
*ip
, int flag
, cred_t
*cr
)
235 znode_t
*zp
= ITOZ(ip
);
236 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
241 /* Decrement the synchronous opens in the znode */
243 atomic_dec_32(&zp
->z_sync_cnt
);
245 if (!zfs_has_ctldir(zp
) && zfsvfs
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
246 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0)
247 VERIFY(zfs_vscan(ip
, cr
, 1) == 0);
253 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
255 * Lseek support for finding holes (cmd == SEEK_HOLE) and
256 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
259 zfs_holey_common(struct inode
*ip
, int cmd
, loff_t
*off
)
261 znode_t
*zp
= ITOZ(ip
);
262 uint64_t noff
= (uint64_t)*off
; /* new offset */
267 file_sz
= zp
->z_size
;
268 if (noff
>= file_sz
) {
269 return (SET_ERROR(ENXIO
));
272 if (cmd
== SEEK_HOLE
)
277 error
= dmu_offset_next(ZTOZSB(zp
)->z_os
, zp
->z_id
, hole
, &noff
);
280 return (SET_ERROR(ENXIO
));
282 /* file was dirty, so fall back to using generic logic */
283 if (error
== EBUSY
) {
291 * We could find a hole that begins after the logical end-of-file,
292 * because dmu_offset_next() only works on whole blocks. If the
293 * EOF falls mid-block, then indicate that the "virtual hole"
294 * at the end of the file begins at the logical EOF, rather than
295 * at the end of the last block.
297 if (noff
> file_sz
) {
309 zfs_holey(struct inode
*ip
, int cmd
, loff_t
*off
)
311 znode_t
*zp
= ITOZ(ip
);
312 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
318 error
= zfs_holey_common(ip
, cmd
, off
);
323 #endif /* SEEK_HOLE && SEEK_DATA */
327 * When a file is memory mapped, we must keep the IO data synchronized
328 * between the DMU cache and the memory mapped pages. What this means:
330 * On Write: If we find a memory mapped page, we write to *both*
331 * the page and the dmu buffer.
334 update_pages(struct inode
*ip
, int64_t start
, int len
,
335 objset_t
*os
, uint64_t oid
)
337 struct address_space
*mp
= ip
->i_mapping
;
343 off
= start
& (PAGE_SIZE
-1);
344 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
345 nbytes
= MIN(PAGE_SIZE
- off
, len
);
347 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
349 if (mapping_writably_mapped(mp
))
350 flush_dcache_page(pp
);
353 (void) dmu_read(os
, oid
, start
+off
, nbytes
, pb
+off
,
357 if (mapping_writably_mapped(mp
))
358 flush_dcache_page(pp
);
360 mark_page_accessed(pp
);
373 * When a file is memory mapped, we must keep the IO data synchronized
374 * between the DMU cache and the memory mapped pages. What this means:
376 * On Read: We "read" preferentially from memory mapped pages,
377 * else we default from the dmu buffer.
379 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
380 * the file is memory mapped.
383 mappedread(struct inode
*ip
, int nbytes
, uio_t
*uio
)
385 struct address_space
*mp
= ip
->i_mapping
;
387 znode_t
*zp
= ITOZ(ip
);
394 start
= uio
->uio_loffset
;
395 off
= start
& (PAGE_SIZE
-1);
396 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
397 bytes
= MIN(PAGE_SIZE
- off
, len
);
399 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
401 ASSERT(PageUptodate(pp
));
404 error
= uiomove(pb
+ off
, bytes
, UIO_READ
, uio
);
407 if (mapping_writably_mapped(mp
))
408 flush_dcache_page(pp
);
410 mark_page_accessed(pp
);
414 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
427 unsigned long zfs_read_chunk_size
= 1024 * 1024; /* Tunable */
428 unsigned long zfs_delete_blocks
= DMU_MAX_DELETEBLKCNT
;
431 * Read bytes from specified file into supplied buffer.
433 * IN: ip - inode of file to be read from.
434 * uio - structure supplying read location, range info,
436 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
437 * O_DIRECT flag; used to bypass page cache.
438 * cr - credentials of caller.
440 * OUT: uio - updated offset and range, buffer filled.
442 * RETURN: 0 on success, error code on failure.
445 * inode - atime updated if byte count > 0
449 zfs_read(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
451 znode_t
*zp
= ITOZ(ip
);
452 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
456 #ifdef HAVE_UIO_ZEROCOPY
458 #endif /* HAVE_UIO_ZEROCOPY */
463 if (zp
->z_pflags
& ZFS_AV_QUARANTINED
) {
465 return (SET_ERROR(EACCES
));
469 * Validate file offset
471 if (uio
->uio_loffset
< (offset_t
)0) {
473 return (SET_ERROR(EINVAL
));
477 * Fasttrack empty reads
479 if (uio
->uio_resid
== 0) {
485 * If we're in FRSYNC mode, sync out this znode before reading it.
486 * Only do this for non-snapshots.
489 (ioflag
& FRSYNC
|| zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
))
490 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
493 * Lock the range against changes.
495 rl
= zfs_range_lock(&zp
->z_range_lock
, uio
->uio_loffset
, uio
->uio_resid
,
499 * If we are reading past end-of-file we can skip
500 * to the end; but we might still need to set atime.
502 if (uio
->uio_loffset
>= zp
->z_size
) {
507 ASSERT(uio
->uio_loffset
< zp
->z_size
);
508 n
= MIN(uio
->uio_resid
, zp
->z_size
- uio
->uio_loffset
);
510 #ifdef HAVE_UIO_ZEROCOPY
511 if ((uio
->uio_extflg
== UIO_XUIO
) &&
512 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
)) {
514 int blksz
= zp
->z_blksz
;
515 uint64_t offset
= uio
->uio_loffset
;
517 xuio
= (xuio_t
*)uio
;
519 nblk
= (P2ROUNDUP(offset
+ n
, blksz
) - P2ALIGN(offset
,
522 ASSERT(offset
+ n
<= blksz
);
525 (void) dmu_xuio_init(xuio
, nblk
);
527 if (vn_has_cached_data(ip
)) {
529 * For simplicity, we always allocate a full buffer
530 * even if we only expect to read a portion of a block.
532 while (--nblk
>= 0) {
533 (void) dmu_xuio_add(xuio
,
534 dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
539 #endif /* HAVE_UIO_ZEROCOPY */
542 nbytes
= MIN(n
, zfs_read_chunk_size
-
543 P2PHASE(uio
->uio_loffset
, zfs_read_chunk_size
));
545 if (zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
)) {
546 error
= mappedread(ip
, nbytes
, uio
);
548 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
553 /* convert checksum errors into IO errors */
555 error
= SET_ERROR(EIO
);
562 zfs_range_unlock(rl
);
569 * Write the bytes to a file.
571 * IN: ip - inode of file to be written to.
572 * uio - structure supplying write location, range info,
574 * ioflag - FAPPEND flag set if in append mode.
575 * O_DIRECT flag; used to bypass page cache.
576 * cr - credentials of caller.
578 * OUT: uio - updated offset and range.
580 * RETURN: 0 if success
581 * error code if failure
584 * ip - ctime|mtime updated if byte count > 0
589 zfs_write(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
591 znode_t
*zp
= ITOZ(ip
);
592 rlim64_t limit
= uio
->uio_limit
;
593 ssize_t start_resid
= uio
->uio_resid
;
597 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
602 int max_blksz
= zfsvfs
->z_max_blksz
;
605 const iovec_t
*aiov
= NULL
;
609 sa_bulk_attr_t bulk
[4];
610 uint64_t mtime
[2], ctime
[2];
612 #ifdef HAVE_UIO_ZEROCOPY
614 const iovec_t
*iovp
= uio
->uio_iov
;
615 ASSERTV(int iovcnt
= uio
->uio_iovcnt
);
619 * Fasttrack empty write
625 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
631 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
632 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
633 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
), NULL
,
635 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
639 * Callers might not be able to detect properly that we are read-only,
640 * so check it explicitly here.
642 if (zfs_is_readonly(zfsvfs
)) {
644 return (SET_ERROR(EROFS
));
648 * If immutable or not appending then return EPERM
650 if ((zp
->z_pflags
& (ZFS_IMMUTABLE
| ZFS_READONLY
)) ||
651 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
652 (uio
->uio_loffset
< zp
->z_size
))) {
654 return (SET_ERROR(EPERM
));
657 zilog
= zfsvfs
->z_log
;
660 * Validate file offset
662 woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
665 return (SET_ERROR(EINVAL
));
669 * Pre-fault the pages to ensure slow (eg NFS) pages
671 * Skip this if uio contains loaned arc_buf.
673 #ifdef HAVE_UIO_ZEROCOPY
674 if ((uio
->uio_extflg
== UIO_XUIO
) &&
675 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
))
676 xuio
= (xuio_t
*)uio
;
679 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
682 * If in append mode, set the io offset pointer to eof.
684 if (ioflag
& FAPPEND
) {
686 * Obtain an appending range lock to guarantee file append
687 * semantics. We reset the write offset once we have the lock.
689 rl
= zfs_range_lock(&zp
->z_range_lock
, 0, n
, RL_APPEND
);
691 if (rl
->r_len
== UINT64_MAX
) {
693 * We overlocked the file because this write will cause
694 * the file block size to increase.
695 * Note that zp_size cannot change with this lock held.
699 uio
->uio_loffset
= woff
;
702 * Note that if the file block size will change as a result of
703 * this write, then this range lock will lock the entire file
704 * so that we can re-write the block safely.
706 rl
= zfs_range_lock(&zp
->z_range_lock
, woff
, n
, RL_WRITER
);
710 zfs_range_unlock(rl
);
712 return (SET_ERROR(EFBIG
));
715 if ((woff
+ n
) > limit
|| woff
> (limit
- n
))
718 /* Will this write extend the file length? */
719 write_eof
= (woff
+ n
> zp
->z_size
);
721 end_size
= MAX(zp
->z_size
, woff
+ n
);
724 * Write the file in reasonable size chunks. Each chunk is written
725 * in a separate transaction; this keeps the intent log records small
726 * and allows us to do more fine-grained space accounting.
730 woff
= uio
->uio_loffset
;
731 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
732 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
734 dmu_return_arcbuf(abuf
);
735 error
= SET_ERROR(EDQUOT
);
739 if (xuio
&& abuf
== NULL
) {
740 #ifdef HAVE_UIO_ZEROCOPY
741 ASSERT(i_iov
< iovcnt
);
742 ASSERT3U(uio
->uio_segflg
, !=, UIO_BVEC
);
744 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
745 dmu_xuio_clear(xuio
, i_iov
);
746 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
747 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
748 aiov
->iov_len
== arc_buf_size(abuf
)));
751 } else if (abuf
== NULL
&& n
>= max_blksz
&&
752 woff
>= zp
->z_size
&&
753 P2PHASE(woff
, max_blksz
) == 0 &&
754 zp
->z_blksz
== max_blksz
) {
756 * This write covers a full block. "Borrow" a buffer
757 * from the dmu so that we can fill it before we enter
758 * a transaction. This avoids the possibility of
759 * holding up the transaction if the data copy hangs
760 * up on a pagefault (e.g., from an NFS server mapping).
764 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
766 ASSERT(abuf
!= NULL
);
767 ASSERT(arc_buf_size(abuf
) == max_blksz
);
768 if ((error
= uiocopy(abuf
->b_data
, max_blksz
,
769 UIO_WRITE
, uio
, &cbytes
))) {
770 dmu_return_arcbuf(abuf
);
773 ASSERT(cbytes
== max_blksz
);
777 * Start a transaction.
779 tx
= dmu_tx_create(zfsvfs
->z_os
);
780 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
781 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
782 zfs_sa_upgrade_txholds(tx
, zp
);
783 error
= dmu_tx_assign(tx
, TXG_WAIT
);
787 dmu_return_arcbuf(abuf
);
792 * If zfs_range_lock() over-locked we grow the blocksize
793 * and then reduce the lock range. This will only happen
794 * on the first iteration since zfs_range_reduce() will
795 * shrink down r_len to the appropriate size.
797 if (rl
->r_len
== UINT64_MAX
) {
800 if (zp
->z_blksz
> max_blksz
) {
802 * File's blocksize is already larger than the
803 * "recordsize" property. Only let it grow to
804 * the next power of 2.
806 ASSERT(!ISP2(zp
->z_blksz
));
807 new_blksz
= MIN(end_size
,
808 1 << highbit64(zp
->z_blksz
));
810 new_blksz
= MIN(end_size
, max_blksz
);
812 zfs_grow_blocksize(zp
, new_blksz
, tx
);
813 zfs_range_reduce(rl
, woff
, n
);
817 * XXX - should we really limit each write to z_max_blksz?
818 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
820 nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
823 tx_bytes
= uio
->uio_resid
;
824 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
826 tx_bytes
-= uio
->uio_resid
;
829 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
831 * If this is not a full block write, but we are
832 * extending the file past EOF and this data starts
833 * block-aligned, use assign_arcbuf(). Otherwise,
834 * write via dmu_write().
836 if (tx_bytes
< max_blksz
&& (!write_eof
||
837 aiov
->iov_base
!= abuf
->b_data
)) {
839 dmu_write(zfsvfs
->z_os
, zp
->z_id
, woff
,
840 /* cppcheck-suppress nullPointer */
841 aiov
->iov_len
, aiov
->iov_base
, tx
);
842 dmu_return_arcbuf(abuf
);
843 xuio_stat_wbuf_copied();
845 ASSERT(xuio
|| tx_bytes
== max_blksz
);
846 dmu_assign_arcbuf_by_dbuf(
847 sa_get_db(zp
->z_sa_hdl
), woff
, abuf
, tx
);
849 ASSERT(tx_bytes
<= uio
->uio_resid
);
850 uioskip(uio
, tx_bytes
);
852 if (tx_bytes
&& zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
)) {
853 update_pages(ip
, woff
,
854 tx_bytes
, zfsvfs
->z_os
, zp
->z_id
);
858 * If we made no progress, we're done. If we made even
859 * partial progress, update the znode and ZIL accordingly.
862 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
863 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
870 * Clear Set-UID/Set-GID bits on successful write if not
871 * privileged and at least one of the execute bits is set.
873 * It would be nice to to this after all writes have
874 * been done, but that would still expose the ISUID/ISGID
875 * to another app after the partial write is committed.
877 * Note: we don't call zfs_fuid_map_id() here because
878 * user 0 is not an ephemeral uid.
880 mutex_enter(&zp
->z_acl_lock
);
881 uid
= KUID_TO_SUID(ip
->i_uid
);
882 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
883 (S_IXUSR
>> 6))) != 0 &&
884 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
885 secpolicy_vnode_setid_retain(cr
,
886 ((zp
->z_mode
& S_ISUID
) != 0 && uid
== 0)) != 0) {
888 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
889 ip
->i_mode
= newmode
= zp
->z_mode
;
890 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zfsvfs
),
891 (void *)&newmode
, sizeof (uint64_t), tx
);
893 mutex_exit(&zp
->z_acl_lock
);
895 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
);
898 * Update the file size (zp_size) if it has changed;
899 * account for possible concurrent updates.
901 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
902 (void) atomic_cas_64(&zp
->z_size
, end_size
,
907 * If we are replaying and eof is non zero then force
908 * the file size to the specified eof. Note, there's no
909 * concurrency during replay.
911 if (zfsvfs
->z_replay
&& zfsvfs
->z_replay_eof
!= 0)
912 zp
->z_size
= zfsvfs
->z_replay_eof
;
914 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
916 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
,
922 ASSERT(tx_bytes
== nbytes
);
926 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
929 zfs_inode_update(zp
);
930 zfs_range_unlock(rl
);
933 * If we're in replay mode, or we made no progress, return error.
934 * Otherwise, it's at least a partial write, so it's successful.
936 if (zfsvfs
->z_replay
|| uio
->uio_resid
== start_resid
) {
941 if (ioflag
& (FSYNC
| FDSYNC
) ||
942 zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
943 zil_commit(zilog
, zp
->z_id
);
950 * Drop a reference on the passed inode asynchronously. This ensures
951 * that the caller will never drop the last reference on an inode in
952 * the current context. Doing so while holding open a tx could result
953 * in a deadlock if iput_final() re-enters the filesystem code.
956 zfs_iput_async(struct inode
*ip
)
958 objset_t
*os
= ITOZSB(ip
)->z_os
;
960 ASSERT(atomic_read(&ip
->i_count
) > 0);
963 if (atomic_read(&ip
->i_count
) == 1)
964 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os
)),
965 (task_func_t
*)iput
, ip
, TQ_SLEEP
) != TASKQID_INVALID
);
971 zfs_get_done(zgd_t
*zgd
, int error
)
973 znode_t
*zp
= zgd
->zgd_private
;
976 dmu_buf_rele(zgd
->zgd_db
, zgd
);
978 zfs_range_unlock(zgd
->zgd_rl
);
981 * Release the vnode asynchronously as we currently have the
982 * txg stopped from syncing.
984 zfs_iput_async(ZTOI(zp
));
986 if (error
== 0 && zgd
->zgd_bp
)
987 zil_lwb_add_block(zgd
->zgd_lwb
, zgd
->zgd_bp
);
989 kmem_free(zgd
, sizeof (zgd_t
));
993 static int zil_fault_io
= 0;
997 * Get data to generate a TX_WRITE intent log record.
1000 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, struct lwb
*lwb
, zio_t
*zio
)
1002 zfsvfs_t
*zfsvfs
= arg
;
1003 objset_t
*os
= zfsvfs
->z_os
;
1005 uint64_t object
= lr
->lr_foid
;
1006 uint64_t offset
= lr
->lr_offset
;
1007 uint64_t size
= lr
->lr_length
;
1012 ASSERT3P(lwb
, !=, NULL
);
1013 ASSERT3P(zio
, !=, NULL
);
1014 ASSERT3U(size
, !=, 0);
1017 * Nothing to do if the file has been removed
1019 if (zfs_zget(zfsvfs
, object
, &zp
) != 0)
1020 return (SET_ERROR(ENOENT
));
1021 if (zp
->z_unlinked
) {
1023 * Release the vnode asynchronously as we currently have the
1024 * txg stopped from syncing.
1026 zfs_iput_async(ZTOI(zp
));
1027 return (SET_ERROR(ENOENT
));
1030 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1032 zgd
->zgd_private
= zp
;
1035 * Write records come in two flavors: immediate and indirect.
1036 * For small writes it's cheaper to store the data with the
1037 * log record (immediate); for large writes it's cheaper to
1038 * sync the data and get a pointer to it (indirect) so that
1039 * we don't have to write the data twice.
1041 if (buf
!= NULL
) { /* immediate write */
1042 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
, size
,
1044 /* test for truncation needs to be done while range locked */
1045 if (offset
>= zp
->z_size
) {
1046 error
= SET_ERROR(ENOENT
);
1048 error
= dmu_read(os
, object
, offset
, size
, buf
,
1049 DMU_READ_NO_PREFETCH
);
1051 ASSERT(error
== 0 || error
== ENOENT
);
1052 } else { /* indirect write */
1054 * Have to lock the whole block to ensure when it's
1055 * written out and its checksum is being calculated
1056 * that no one can change the data. We need to re-check
1057 * blocksize after we get the lock in case it's changed!
1062 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1064 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
,
1066 if (zp
->z_blksz
== size
)
1069 zfs_range_unlock(zgd
->zgd_rl
);
1071 /* test for truncation needs to be done while range locked */
1072 if (lr
->lr_offset
>= zp
->z_size
)
1073 error
= SET_ERROR(ENOENT
);
1076 error
= SET_ERROR(EIO
);
1081 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1082 DMU_READ_NO_PREFETCH
);
1085 blkptr_t
*bp
= &lr
->lr_blkptr
;
1090 ASSERT(db
->db_offset
== offset
);
1091 ASSERT(db
->db_size
== size
);
1093 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1095 ASSERT(error
|| lr
->lr_length
<= size
);
1098 * On success, we need to wait for the write I/O
1099 * initiated by dmu_sync() to complete before we can
1100 * release this dbuf. We will finish everything up
1101 * in the zfs_get_done() callback.
1106 if (error
== EALREADY
) {
1107 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
1113 zfs_get_done(zgd
, error
);
1120 zfs_access(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
1122 znode_t
*zp
= ITOZ(ip
);
1123 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
1129 if (flag
& V_ACE_MASK
)
1130 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1132 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1139 * Lookup an entry in a directory, or an extended attribute directory.
1140 * If it exists, return a held inode reference for it.
1142 * IN: dip - inode of directory to search.
1143 * nm - name of entry to lookup.
1144 * flags - LOOKUP_XATTR set if looking for an attribute.
1145 * cr - credentials of caller.
1146 * direntflags - directory lookup flags
1147 * realpnp - returned pathname.
1149 * OUT: ipp - inode of located entry, NULL if not found.
1151 * RETURN: 0 on success, error code on failure.
1158 zfs_lookup(struct inode
*dip
, char *nm
, struct inode
**ipp
, int flags
,
1159 cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1161 znode_t
*zdp
= ITOZ(dip
);
1162 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1166 * Fast path lookup, however we must skip DNLC lookup
1167 * for case folding or normalizing lookups because the
1168 * DNLC code only stores the passed in name. This means
1169 * creating 'a' and removing 'A' on a case insensitive
1170 * file system would work, but DNLC still thinks 'a'
1171 * exists and won't let you create it again on the next
1172 * pass through fast path.
1174 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1176 if (!S_ISDIR(dip
->i_mode
)) {
1177 return (SET_ERROR(ENOTDIR
));
1178 } else if (zdp
->z_sa_hdl
== NULL
) {
1179 return (SET_ERROR(EIO
));
1182 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1183 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1191 } else if (!zdp
->z_zfsvfs
->z_norm
&&
1192 (zdp
->z_zfsvfs
->z_case
== ZFS_CASE_SENSITIVE
)) {
1194 vnode_t
*tvp
= dnlc_lookup(dvp
, nm
);
1197 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1202 if (tvp
== DNLC_NO_VNODE
) {
1204 return (SET_ERROR(ENOENT
));
1207 return (specvp_check(vpp
, cr
));
1210 #endif /* HAVE_DNLC */
1219 if (flags
& LOOKUP_XATTR
) {
1221 * We don't allow recursive attributes..
1222 * Maybe someday we will.
1224 if (zdp
->z_pflags
& ZFS_XATTR
) {
1226 return (SET_ERROR(EINVAL
));
1229 if ((error
= zfs_get_xattrdir(zdp
, ipp
, cr
, flags
))) {
1235 * Do we have permission to get into attribute directory?
1238 if ((error
= zfs_zaccess(ITOZ(*ipp
), ACE_EXECUTE
, 0,
1248 if (!S_ISDIR(dip
->i_mode
)) {
1250 return (SET_ERROR(ENOTDIR
));
1254 * Check accessibility of directory.
1257 if ((error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
))) {
1262 if (zfsvfs
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1263 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1265 return (SET_ERROR(EILSEQ
));
1268 error
= zfs_dirlook(zdp
, nm
, ipp
, flags
, direntflags
, realpnp
);
1269 if ((error
== 0) && (*ipp
))
1270 zfs_inode_update(ITOZ(*ipp
));
1277 * Attempt to create a new entry in a directory. If the entry
1278 * already exists, truncate the file if permissible, else return
1279 * an error. Return the ip of the created or trunc'd file.
1281 * IN: dip - inode of directory to put new file entry in.
1282 * name - name of new file entry.
1283 * vap - attributes of new file.
1284 * excl - flag indicating exclusive or non-exclusive mode.
1285 * mode - mode to open file with.
1286 * cr - credentials of caller.
1287 * flag - large file flag [UNUSED].
1288 * vsecp - ACL to be set
1290 * OUT: ipp - inode of created or trunc'd entry.
1292 * RETURN: 0 on success, error code on failure.
1295 * dip - ctime|mtime updated if new entry created
1296 * ip - ctime|mtime always, atime if new
1301 zfs_create(struct inode
*dip
, char *name
, vattr_t
*vap
, int excl
,
1302 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1304 znode_t
*zp
, *dzp
= ITOZ(dip
);
1305 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1313 zfs_acl_ids_t acl_ids
;
1314 boolean_t fuid_dirtied
;
1315 boolean_t have_acl
= B_FALSE
;
1316 boolean_t waited
= B_FALSE
;
1319 * If we have an ephemeral id, ACL, or XVATTR then
1320 * make sure file system is at proper version
1326 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1327 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1328 return (SET_ERROR(EINVAL
));
1331 return (SET_ERROR(EINVAL
));
1336 zilog
= zfsvfs
->z_log
;
1338 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
1339 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1341 return (SET_ERROR(EILSEQ
));
1344 if (vap
->va_mask
& ATTR_XVATTR
) {
1345 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1346 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1354 if (*name
== '\0') {
1356 * Null component name refers to the directory itself.
1363 /* possible igrab(zp) */
1366 if (flag
& FIGNORECASE
)
1369 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1373 zfs_acl_ids_free(&acl_ids
);
1374 if (strcmp(name
, "..") == 0)
1375 error
= SET_ERROR(EISDIR
);
1385 * Create a new file object and update the directory
1388 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1390 zfs_acl_ids_free(&acl_ids
);
1395 * We only support the creation of regular files in
1396 * extended attribute directories.
1399 if ((dzp
->z_pflags
& ZFS_XATTR
) && !S_ISREG(vap
->va_mode
)) {
1401 zfs_acl_ids_free(&acl_ids
);
1402 error
= SET_ERROR(EINVAL
);
1406 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1407 cr
, vsecp
, &acl_ids
)) != 0)
1411 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1412 zfs_acl_ids_free(&acl_ids
);
1413 error
= SET_ERROR(EDQUOT
);
1417 tx
= dmu_tx_create(os
);
1419 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1420 ZFS_SA_BASE_ATTR_SIZE
);
1422 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1424 zfs_fuid_txhold(zfsvfs
, tx
);
1425 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1426 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1427 if (!zfsvfs
->z_use_sa
&&
1428 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1429 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1430 0, acl_ids
.z_aclp
->z_acl_bytes
);
1432 error
= dmu_tx_assign(tx
,
1433 (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
1435 zfs_dirent_unlock(dl
);
1436 if (error
== ERESTART
) {
1442 zfs_acl_ids_free(&acl_ids
);
1447 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1450 zfs_fuid_sync(zfsvfs
, tx
);
1452 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1453 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1454 if (flag
& FIGNORECASE
)
1456 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1457 vsecp
, acl_ids
.z_fuidp
, vap
);
1458 zfs_acl_ids_free(&acl_ids
);
1461 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1464 zfs_acl_ids_free(&acl_ids
);
1468 * A directory entry already exists for this name.
1471 * Can't truncate an existing file if in exclusive mode.
1474 error
= SET_ERROR(EEXIST
);
1478 * Can't open a directory for writing.
1480 if (S_ISDIR(ZTOI(zp
)->i_mode
)) {
1481 error
= SET_ERROR(EISDIR
);
1485 * Verify requested access to file.
1487 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1491 mutex_enter(&dzp
->z_lock
);
1493 mutex_exit(&dzp
->z_lock
);
1496 * Truncate regular files if requested.
1498 if (S_ISREG(ZTOI(zp
)->i_mode
) &&
1499 (vap
->va_mask
& ATTR_SIZE
) && (vap
->va_size
== 0)) {
1500 /* we can't hold any locks when calling zfs_freesp() */
1502 zfs_dirent_unlock(dl
);
1505 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1511 zfs_dirent_unlock(dl
);
1517 zfs_inode_update(dzp
);
1518 zfs_inode_update(zp
);
1522 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1523 zil_commit(zilog
, 0);
1531 zfs_tmpfile(struct inode
*dip
, vattr_t
*vap
, int excl
,
1532 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1534 znode_t
*zp
= NULL
, *dzp
= ITOZ(dip
);
1535 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1541 zfs_acl_ids_t acl_ids
;
1542 boolean_t fuid_dirtied
;
1543 boolean_t have_acl
= B_FALSE
;
1544 boolean_t waited
= B_FALSE
;
1547 * If we have an ephemeral id, ACL, or XVATTR then
1548 * make sure file system is at proper version
1554 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1555 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1556 return (SET_ERROR(EINVAL
));
1562 if (vap
->va_mask
& ATTR_XVATTR
) {
1563 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1564 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1574 * Create a new file object and update the directory
1577 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1579 zfs_acl_ids_free(&acl_ids
);
1583 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1584 cr
, vsecp
, &acl_ids
)) != 0)
1588 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1589 zfs_acl_ids_free(&acl_ids
);
1590 error
= SET_ERROR(EDQUOT
);
1594 tx
= dmu_tx_create(os
);
1596 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1597 ZFS_SA_BASE_ATTR_SIZE
);
1598 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1600 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1602 zfs_fuid_txhold(zfsvfs
, tx
);
1603 if (!zfsvfs
->z_use_sa
&&
1604 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1605 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1606 0, acl_ids
.z_aclp
->z_acl_bytes
);
1608 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
1610 if (error
== ERESTART
) {
1616 zfs_acl_ids_free(&acl_ids
);
1621 zfs_mknode(dzp
, vap
, tx
, cr
, IS_TMPFILE
, &zp
, &acl_ids
);
1624 zfs_fuid_sync(zfsvfs
, tx
);
1626 /* Add to unlinked set */
1628 zfs_unlinked_add(zp
, tx
);
1629 zfs_acl_ids_free(&acl_ids
);
1637 zfs_inode_update(dzp
);
1638 zfs_inode_update(zp
);
1647 * Remove an entry from a directory.
1649 * IN: dip - inode of directory to remove entry from.
1650 * name - name of entry to remove.
1651 * cr - credentials of caller.
1653 * RETURN: 0 if success
1654 * error code if failure
1658 * ip - ctime (if nlink > 0)
1661 uint64_t null_xattr
= 0;
1665 zfs_remove(struct inode
*dip
, char *name
, cred_t
*cr
, int flags
)
1667 znode_t
*zp
, *dzp
= ITOZ(dip
);
1670 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1672 uint64_t acl_obj
, xattr_obj
;
1673 uint64_t xattr_obj_unlinked
= 0;
1678 boolean_t may_delete_now
, delete_now
= FALSE
;
1679 boolean_t unlinked
, toobig
= FALSE
;
1681 pathname_t
*realnmp
= NULL
;
1685 boolean_t waited
= B_FALSE
;
1688 return (SET_ERROR(EINVAL
));
1692 zilog
= zfsvfs
->z_log
;
1694 if (flags
& FIGNORECASE
) {
1704 * Attempt to lock directory; fail if entry doesn't exist.
1706 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1716 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1721 * Need to use rmdir for removing directories.
1723 if (S_ISDIR(ip
->i_mode
)) {
1724 error
= SET_ERROR(EPERM
);
1730 dnlc_remove(dvp
, realnmp
->pn_buf
);
1732 dnlc_remove(dvp
, name
);
1733 #endif /* HAVE_DNLC */
1735 mutex_enter(&zp
->z_lock
);
1736 may_delete_now
= atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
);
1737 mutex_exit(&zp
->z_lock
);
1740 * We may delete the znode now, or we may put it in the unlinked set;
1741 * it depends on whether we're the last link, and on whether there are
1742 * other holds on the inode. So we dmu_tx_hold() the right things to
1743 * allow for either case.
1746 tx
= dmu_tx_create(zfsvfs
->z_os
);
1747 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1748 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1749 zfs_sa_upgrade_txholds(tx
, zp
);
1750 zfs_sa_upgrade_txholds(tx
, dzp
);
1751 if (may_delete_now
) {
1752 toobig
= zp
->z_size
> zp
->z_blksz
* zfs_delete_blocks
;
1753 /* if the file is too big, only hold_free a token amount */
1754 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1755 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1758 /* are there any extended attributes? */
1759 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1760 &xattr_obj
, sizeof (xattr_obj
));
1761 if (error
== 0 && xattr_obj
) {
1762 error
= zfs_zget(zfsvfs
, xattr_obj
, &xzp
);
1764 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1765 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1768 mutex_enter(&zp
->z_lock
);
1769 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1770 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1771 mutex_exit(&zp
->z_lock
);
1773 /* charge as an update -- would be nice not to charge at all */
1774 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1777 * Mark this transaction as typically resulting in a net free of space
1779 dmu_tx_mark_netfree(tx
);
1781 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
1783 zfs_dirent_unlock(dl
);
1784 if (error
== ERESTART
) {
1804 * Remove the directory entry.
1806 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1815 * Hold z_lock so that we can make sure that the ACL obj
1816 * hasn't changed. Could have been deleted due to
1819 mutex_enter(&zp
->z_lock
);
1820 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1821 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1822 delete_now
= may_delete_now
&& !toobig
&&
1823 atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
) &&
1824 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1829 if (xattr_obj_unlinked
) {
1830 ASSERT3U(ZTOI(xzp
)->i_nlink
, ==, 2);
1831 mutex_enter(&xzp
->z_lock
);
1832 xzp
->z_unlinked
= 1;
1833 clear_nlink(ZTOI(xzp
));
1835 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zfsvfs
),
1836 &links
, sizeof (links
), tx
);
1837 ASSERT3U(error
, ==, 0);
1838 mutex_exit(&xzp
->z_lock
);
1839 zfs_unlinked_add(xzp
, tx
);
1842 error
= sa_remove(zp
->z_sa_hdl
,
1843 SA_ZPL_XATTR(zfsvfs
), tx
);
1845 error
= sa_update(zp
->z_sa_hdl
,
1846 SA_ZPL_XATTR(zfsvfs
), &null_xattr
,
1847 sizeof (uint64_t), tx
);
1851 * Add to the unlinked set because a new reference could be
1852 * taken concurrently resulting in a deferred destruction.
1854 zfs_unlinked_add(zp
, tx
);
1855 mutex_exit(&zp
->z_lock
);
1856 } else if (unlinked
) {
1857 mutex_exit(&zp
->z_lock
);
1858 zfs_unlinked_add(zp
, tx
);
1862 if (flags
& FIGNORECASE
)
1864 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1871 zfs_dirent_unlock(dl
);
1872 zfs_inode_update(dzp
);
1873 zfs_inode_update(zp
);
1881 zfs_inode_update(xzp
);
1882 zfs_iput_async(ZTOI(xzp
));
1885 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1886 zil_commit(zilog
, 0);
1893 * Create a new directory and insert it into dip using the name
1894 * provided. Return a pointer to the inserted directory.
1896 * IN: dip - inode of directory to add subdir to.
1897 * dirname - name of new directory.
1898 * vap - attributes of new directory.
1899 * cr - credentials of caller.
1900 * vsecp - ACL to be set
1902 * OUT: ipp - inode of created directory.
1904 * RETURN: 0 if success
1905 * error code if failure
1908 * dip - ctime|mtime updated
1909 * ipp - ctime|mtime|atime updated
1913 zfs_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
, struct inode
**ipp
,
1914 cred_t
*cr
, int flags
, vsecattr_t
*vsecp
)
1916 znode_t
*zp
, *dzp
= ITOZ(dip
);
1917 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1925 gid_t gid
= crgetgid(cr
);
1926 zfs_acl_ids_t acl_ids
;
1927 boolean_t fuid_dirtied
;
1928 boolean_t waited
= B_FALSE
;
1930 ASSERT(S_ISDIR(vap
->va_mode
));
1933 * If we have an ephemeral id, ACL, or XVATTR then
1934 * make sure file system is at proper version
1938 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1939 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1940 return (SET_ERROR(EINVAL
));
1942 if (dirname
== NULL
)
1943 return (SET_ERROR(EINVAL
));
1947 zilog
= zfsvfs
->z_log
;
1949 if (dzp
->z_pflags
& ZFS_XATTR
) {
1951 return (SET_ERROR(EINVAL
));
1954 if (zfsvfs
->z_utf8
&& u8_validate(dirname
,
1955 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1957 return (SET_ERROR(EILSEQ
));
1959 if (flags
& FIGNORECASE
)
1962 if (vap
->va_mask
& ATTR_XVATTR
) {
1963 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1964 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1970 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1971 vsecp
, &acl_ids
)) != 0) {
1976 * First make sure the new directory doesn't exist.
1978 * Existence is checked first to make sure we don't return
1979 * EACCES instead of EEXIST which can cause some applications
1985 if ((error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1987 zfs_acl_ids_free(&acl_ids
);
1992 if ((error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
))) {
1993 zfs_acl_ids_free(&acl_ids
);
1994 zfs_dirent_unlock(dl
);
1999 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
2000 zfs_acl_ids_free(&acl_ids
);
2001 zfs_dirent_unlock(dl
);
2003 return (SET_ERROR(EDQUOT
));
2007 * Add a new entry to the directory.
2009 tx
= dmu_tx_create(zfsvfs
->z_os
);
2010 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
2011 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
2012 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
2014 zfs_fuid_txhold(zfsvfs
, tx
);
2015 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
2016 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
2017 acl_ids
.z_aclp
->z_acl_bytes
);
2020 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
2021 ZFS_SA_BASE_ATTR_SIZE
);
2023 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
2025 zfs_dirent_unlock(dl
);
2026 if (error
== ERESTART
) {
2032 zfs_acl_ids_free(&acl_ids
);
2041 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
2044 zfs_fuid_sync(zfsvfs
, tx
);
2047 * Now put new name in parent dir.
2049 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
2053 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
2054 if (flags
& FIGNORECASE
)
2056 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
2057 acl_ids
.z_fuidp
, vap
);
2059 zfs_acl_ids_free(&acl_ids
);
2063 zfs_dirent_unlock(dl
);
2065 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2066 zil_commit(zilog
, 0);
2068 zfs_inode_update(dzp
);
2069 zfs_inode_update(zp
);
2075 * Remove a directory subdir entry. If the current working
2076 * directory is the same as the subdir to be removed, the
2079 * IN: dip - inode of directory to remove from.
2080 * name - name of directory to be removed.
2081 * cwd - inode of current working directory.
2082 * cr - credentials of caller.
2083 * flags - case flags
2085 * RETURN: 0 on success, error code on failure.
2088 * dip - ctime|mtime updated
2092 zfs_rmdir(struct inode
*dip
, char *name
, struct inode
*cwd
, cred_t
*cr
,
2095 znode_t
*dzp
= ITOZ(dip
);
2098 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
2104 boolean_t waited
= B_FALSE
;
2107 return (SET_ERROR(EINVAL
));
2111 zilog
= zfsvfs
->z_log
;
2113 if (flags
& FIGNORECASE
)
2119 * Attempt to lock directory; fail if entry doesn't exist.
2121 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
2129 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
2133 if (!S_ISDIR(ip
->i_mode
)) {
2134 error
= SET_ERROR(ENOTDIR
);
2139 error
= SET_ERROR(EINVAL
);
2144 * Grab a lock on the directory to make sure that no one is
2145 * trying to add (or lookup) entries while we are removing it.
2147 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
2150 * Grab a lock on the parent pointer to make sure we play well
2151 * with the treewalk and directory rename code.
2153 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
2155 tx
= dmu_tx_create(zfsvfs
->z_os
);
2156 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2157 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2158 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
2159 zfs_sa_upgrade_txholds(tx
, zp
);
2160 zfs_sa_upgrade_txholds(tx
, dzp
);
2161 dmu_tx_mark_netfree(tx
);
2162 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
2164 rw_exit(&zp
->z_parent_lock
);
2165 rw_exit(&zp
->z_name_lock
);
2166 zfs_dirent_unlock(dl
);
2167 if (error
== ERESTART
) {
2180 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2183 uint64_t txtype
= TX_RMDIR
;
2184 if (flags
& FIGNORECASE
)
2186 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2191 rw_exit(&zp
->z_parent_lock
);
2192 rw_exit(&zp
->z_name_lock
);
2194 zfs_dirent_unlock(dl
);
2196 zfs_inode_update(dzp
);
2197 zfs_inode_update(zp
);
2200 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2201 zil_commit(zilog
, 0);
2208 * Read as many directory entries as will fit into the provided
2209 * dirent buffer from the given directory cursor position.
2211 * IN: ip - inode of directory to read.
2212 * dirent - buffer for directory entries.
2214 * OUT: dirent - filler buffer of directory entries.
2216 * RETURN: 0 if success
2217 * error code if failure
2220 * ip - atime updated
2222 * Note that the low 4 bits of the cookie returned by zap is always zero.
2223 * This allows us to use the low range for "special" directory entries:
2224 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2225 * we use the offset 2 for the '.zfs' directory.
2229 zfs_readdir(struct inode
*ip
, struct dir_context
*ctx
, cred_t
*cr
)
2231 znode_t
*zp
= ITOZ(ip
);
2232 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2235 zap_attribute_t zap
;
2241 uint64_t offset
; /* must be unsigned; checks for < 1 */
2246 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
2247 &parent
, sizeof (parent
))) != 0)
2251 * Quit if directory has been removed (posix)
2259 prefetch
= zp
->z_zn_prefetch
;
2262 * Initialize the iterator cursor.
2266 * Start iteration from the beginning of the directory.
2268 zap_cursor_init(&zc
, os
, zp
->z_id
);
2271 * The offset is a serialized cursor.
2273 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2277 * Transform to file-system independent format
2282 * Special case `.', `..', and `.zfs'.
2285 (void) strcpy(zap
.za_name
, ".");
2286 zap
.za_normalization_conflict
= 0;
2289 } else if (offset
== 1) {
2290 (void) strcpy(zap
.za_name
, "..");
2291 zap
.za_normalization_conflict
= 0;
2294 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2295 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2296 zap
.za_normalization_conflict
= 0;
2297 objnum
= ZFSCTL_INO_ROOT
;
2303 if ((error
= zap_cursor_retrieve(&zc
, &zap
))) {
2304 if (error
== ENOENT
)
2311 * Allow multiple entries provided the first entry is
2312 * the object id. Non-zpl consumers may safely make
2313 * use of the additional space.
2315 * XXX: This should be a feature flag for compatibility
2317 if (zap
.za_integer_length
!= 8 ||
2318 zap
.za_num_integers
== 0) {
2319 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2320 "entry, obj = %lld, offset = %lld, "
2321 "length = %d, num = %lld\n",
2322 (u_longlong_t
)zp
->z_id
,
2323 (u_longlong_t
)offset
,
2324 zap
.za_integer_length
,
2325 (u_longlong_t
)zap
.za_num_integers
);
2326 error
= SET_ERROR(ENXIO
);
2330 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2331 type
= ZFS_DIRENT_TYPE(zap
.za_first_integer
);
2334 done
= !dir_emit(ctx
, zap
.za_name
, strlen(zap
.za_name
),
2339 /* Prefetch znode */
2341 dmu_prefetch(os
, objnum
, 0, 0, 0,
2342 ZIO_PRIORITY_SYNC_READ
);
2346 * Move to the next entry, fill in the previous offset.
2348 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2349 zap_cursor_advance(&zc
);
2350 offset
= zap_cursor_serialize(&zc
);
2356 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2359 zap_cursor_fini(&zc
);
2360 if (error
== ENOENT
)
2368 ulong_t zfs_fsync_sync_cnt
= 4;
2371 zfs_fsync(struct inode
*ip
, int syncflag
, cred_t
*cr
)
2373 znode_t
*zp
= ITOZ(ip
);
2374 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2376 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2378 if (zfsvfs
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2381 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
2384 tsd_set(zfs_fsyncer_key
, NULL
);
2391 * Get the requested file attributes and place them in the provided
2394 * IN: ip - inode of file.
2395 * vap - va_mask identifies requested attributes.
2396 * If ATTR_XVATTR set, then optional attrs are requested
2397 * flags - ATTR_NOACLCHECK (CIFS server context)
2398 * cr - credentials of caller.
2400 * OUT: vap - attribute values.
2402 * RETURN: 0 (always succeeds)
2406 zfs_getattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2408 znode_t
*zp
= ITOZ(ip
);
2409 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2412 uint64_t atime
[2], mtime
[2], ctime
[2];
2413 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2414 xoptattr_t
*xoap
= NULL
;
2415 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2416 sa_bulk_attr_t bulk
[3];
2422 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2424 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
, &atime
, 16);
2425 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
2426 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
2428 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2434 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2435 * Also, if we are the owner don't bother, since owner should
2436 * always be allowed to read basic attributes of file.
2438 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2439 (vap
->va_uid
!= crgetuid(cr
))) {
2440 if ((error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2448 * Return all attributes. It's cheaper to provide the answer
2449 * than to determine whether we were asked the question.
2452 mutex_enter(&zp
->z_lock
);
2453 vap
->va_type
= vn_mode_to_vtype(zp
->z_mode
);
2454 vap
->va_mode
= zp
->z_mode
;
2455 vap
->va_fsid
= ZTOI(zp
)->i_sb
->s_dev
;
2456 vap
->va_nodeid
= zp
->z_id
;
2457 if ((zp
->z_id
== zfsvfs
->z_root
) && zfs_show_ctldir(zp
))
2458 links
= ZTOI(zp
)->i_nlink
+ 1;
2460 links
= ZTOI(zp
)->i_nlink
;
2461 vap
->va_nlink
= MIN(links
, ZFS_LINK_MAX
);
2462 vap
->va_size
= i_size_read(ip
);
2463 vap
->va_rdev
= ip
->i_rdev
;
2464 vap
->va_seq
= ip
->i_generation
;
2467 * Add in any requested optional attributes and the create time.
2468 * Also set the corresponding bits in the returned attribute bitmap.
2470 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zfsvfs
->z_use_fuids
) {
2471 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2473 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2474 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2477 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2478 xoap
->xoa_readonly
=
2479 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2480 XVA_SET_RTN(xvap
, XAT_READONLY
);
2483 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2485 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2486 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2489 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2491 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2492 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2495 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2496 xoap
->xoa_nounlink
=
2497 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2498 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2501 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2502 xoap
->xoa_immutable
=
2503 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2504 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2507 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2508 xoap
->xoa_appendonly
=
2509 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2510 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2513 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2515 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2516 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2519 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2521 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2522 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2525 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2526 xoap
->xoa_av_quarantined
=
2527 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2528 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2531 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2532 xoap
->xoa_av_modified
=
2533 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2534 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2537 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2538 S_ISREG(ip
->i_mode
)) {
2539 zfs_sa_get_scanstamp(zp
, xvap
);
2542 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2545 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zfsvfs
),
2546 times
, sizeof (times
));
2547 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2548 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2551 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2552 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2553 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2555 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2556 xoap
->xoa_generation
= ip
->i_generation
;
2557 XVA_SET_RTN(xvap
, XAT_GEN
);
2560 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2562 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2563 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2566 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2568 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2569 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2573 ZFS_TIME_DECODE(&vap
->va_atime
, atime
);
2574 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2575 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2577 mutex_exit(&zp
->z_lock
);
2579 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2581 if (zp
->z_blksz
== 0) {
2583 * Block size hasn't been set; suggest maximal I/O transfers.
2585 vap
->va_blksize
= zfsvfs
->z_max_blksz
;
2593 * Get the basic file attributes and place them in the provided kstat
2594 * structure. The inode is assumed to be the authoritative source
2595 * for most of the attributes. However, the znode currently has the
2596 * authoritative atime, blksize, and block count.
2598 * IN: ip - inode of file.
2600 * OUT: sp - kstat values.
2602 * RETURN: 0 (always succeeds)
2606 zfs_getattr_fast(struct inode
*ip
, struct kstat
*sp
)
2608 znode_t
*zp
= ITOZ(ip
);
2609 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2611 u_longlong_t nblocks
;
2616 mutex_enter(&zp
->z_lock
);
2618 generic_fillattr(ip
, sp
);
2620 sa_object_size(zp
->z_sa_hdl
, &blksize
, &nblocks
);
2621 sp
->blksize
= blksize
;
2622 sp
->blocks
= nblocks
;
2624 if (unlikely(zp
->z_blksz
== 0)) {
2626 * Block size hasn't been set; suggest maximal I/O transfers.
2628 sp
->blksize
= zfsvfs
->z_max_blksz
;
2631 mutex_exit(&zp
->z_lock
);
2634 * Required to prevent NFS client from detecting different inode
2635 * numbers of snapshot root dentry before and after snapshot mount.
2637 if (zfsvfs
->z_issnap
) {
2638 if (ip
->i_sb
->s_root
->d_inode
== ip
)
2639 sp
->ino
= ZFSCTL_INO_SNAPDIRS
-
2640 dmu_objset_id(zfsvfs
->z_os
);
2649 * Set the file attributes to the values contained in the
2652 * IN: ip - inode of file to be modified.
2653 * vap - new attribute values.
2654 * If ATTR_XVATTR set, then optional attrs are being set
2655 * flags - ATTR_UTIME set if non-default time values provided.
2656 * - ATTR_NOACLCHECK (CIFS context only).
2657 * cr - credentials of caller.
2659 * RETURN: 0 if success
2660 * error code if failure
2663 * ip - ctime updated, mtime updated if size changed.
2667 zfs_setattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2669 znode_t
*zp
= ITOZ(ip
);
2670 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2674 xvattr_t
*tmpxvattr
;
2675 uint_t mask
= vap
->va_mask
;
2676 uint_t saved_mask
= 0;
2679 uint64_t new_kuid
= 0, new_kgid
= 0, new_uid
, new_gid
;
2681 uint64_t mtime
[2], ctime
[2], atime
[2];
2683 int need_policy
= FALSE
;
2685 zfs_fuid_info_t
*fuidp
= NULL
;
2686 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2689 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2690 boolean_t fuid_dirtied
= B_FALSE
;
2691 sa_bulk_attr_t
*bulk
, *xattr_bulk
;
2692 int count
= 0, xattr_count
= 0;
2700 zilog
= zfsvfs
->z_log
;
2703 * Make sure that if we have ephemeral uid/gid or xvattr specified
2704 * that file system is at proper version level
2707 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
2708 (((mask
& ATTR_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2709 ((mask
& ATTR_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2710 (mask
& ATTR_XVATTR
))) {
2712 return (SET_ERROR(EINVAL
));
2715 if (mask
& ATTR_SIZE
&& S_ISDIR(ip
->i_mode
)) {
2717 return (SET_ERROR(EISDIR
));
2720 if (mask
& ATTR_SIZE
&& !S_ISREG(ip
->i_mode
) && !S_ISFIFO(ip
->i_mode
)) {
2722 return (SET_ERROR(EINVAL
));
2726 * If this is an xvattr_t, then get a pointer to the structure of
2727 * optional attributes. If this is NULL, then we have a vattr_t.
2729 xoap
= xva_getxoptattr(xvap
);
2731 tmpxvattr
= kmem_alloc(sizeof (xvattr_t
), KM_SLEEP
);
2732 xva_init(tmpxvattr
);
2734 bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2735 xattr_bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2738 * Immutable files can only alter immutable bit and atime
2740 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2741 ((mask
& (ATTR_SIZE
|ATTR_UID
|ATTR_GID
|ATTR_MTIME
|ATTR_MODE
)) ||
2742 ((mask
& ATTR_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2743 err
= SET_ERROR(EPERM
);
2747 if ((mask
& ATTR_SIZE
) && (zp
->z_pflags
& ZFS_READONLY
)) {
2748 err
= SET_ERROR(EPERM
);
2753 * Verify timestamps doesn't overflow 32 bits.
2754 * ZFS can handle large timestamps, but 32bit syscalls can't
2755 * handle times greater than 2039. This check should be removed
2756 * once large timestamps are fully supported.
2758 if (mask
& (ATTR_ATIME
| ATTR_MTIME
)) {
2759 if (((mask
& ATTR_ATIME
) &&
2760 TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2761 ((mask
& ATTR_MTIME
) &&
2762 TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2763 err
= SET_ERROR(EOVERFLOW
);
2772 /* Can this be moved to before the top label? */
2773 if (zfs_is_readonly(zfsvfs
)) {
2774 err
= SET_ERROR(EROFS
);
2779 * First validate permissions
2782 if (mask
& ATTR_SIZE
) {
2783 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2788 * XXX - Note, we are not providing any open
2789 * mode flags here (like FNDELAY), so we may
2790 * block if there are locks present... this
2791 * should be addressed in openat().
2793 /* XXX - would it be OK to generate a log record here? */
2794 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2799 if (mask
& (ATTR_ATIME
|ATTR_MTIME
) ||
2800 ((mask
& ATTR_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2801 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2802 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2803 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2804 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2805 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2806 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2807 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2811 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2812 int idmask
= (mask
& (ATTR_UID
|ATTR_GID
));
2817 * NOTE: even if a new mode is being set,
2818 * we may clear S_ISUID/S_ISGID bits.
2821 if (!(mask
& ATTR_MODE
))
2822 vap
->va_mode
= zp
->z_mode
;
2825 * Take ownership or chgrp to group we are a member of
2828 take_owner
= (mask
& ATTR_UID
) && (vap
->va_uid
== crgetuid(cr
));
2829 take_group
= (mask
& ATTR_GID
) &&
2830 zfs_groupmember(zfsvfs
, vap
->va_gid
, cr
);
2833 * If both ATTR_UID and ATTR_GID are set then take_owner and
2834 * take_group must both be set in order to allow taking
2837 * Otherwise, send the check through secpolicy_vnode_setattr()
2841 if (((idmask
== (ATTR_UID
|ATTR_GID
)) &&
2842 take_owner
&& take_group
) ||
2843 ((idmask
== ATTR_UID
) && take_owner
) ||
2844 ((idmask
== ATTR_GID
) && take_group
)) {
2845 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2846 skipaclchk
, cr
) == 0) {
2848 * Remove setuid/setgid for non-privileged users
2850 (void) secpolicy_setid_clear(vap
, cr
);
2851 trim_mask
= (mask
& (ATTR_UID
|ATTR_GID
));
2860 mutex_enter(&zp
->z_lock
);
2861 oldva
.va_mode
= zp
->z_mode
;
2862 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2863 if (mask
& ATTR_XVATTR
) {
2865 * Update xvattr mask to include only those attributes
2866 * that are actually changing.
2868 * the bits will be restored prior to actually setting
2869 * the attributes so the caller thinks they were set.
2871 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2872 if (xoap
->xoa_appendonly
!=
2873 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2876 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2877 XVA_SET_REQ(tmpxvattr
, XAT_APPENDONLY
);
2881 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2882 if (xoap
->xoa_nounlink
!=
2883 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2886 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2887 XVA_SET_REQ(tmpxvattr
, XAT_NOUNLINK
);
2891 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2892 if (xoap
->xoa_immutable
!=
2893 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2896 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2897 XVA_SET_REQ(tmpxvattr
, XAT_IMMUTABLE
);
2901 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2902 if (xoap
->xoa_nodump
!=
2903 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2906 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2907 XVA_SET_REQ(tmpxvattr
, XAT_NODUMP
);
2911 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2912 if (xoap
->xoa_av_modified
!=
2913 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2916 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2917 XVA_SET_REQ(tmpxvattr
, XAT_AV_MODIFIED
);
2921 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2922 if ((!S_ISREG(ip
->i_mode
) &&
2923 xoap
->xoa_av_quarantined
) ||
2924 xoap
->xoa_av_quarantined
!=
2925 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2928 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2929 XVA_SET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
);
2933 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2934 mutex_exit(&zp
->z_lock
);
2935 err
= SET_ERROR(EPERM
);
2939 if (need_policy
== FALSE
&&
2940 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
2941 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
2946 mutex_exit(&zp
->z_lock
);
2948 if (mask
& ATTR_MODE
) {
2949 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
2950 err
= secpolicy_setid_setsticky_clear(ip
, vap
,
2955 trim_mask
|= ATTR_MODE
;
2963 * If trim_mask is set then take ownership
2964 * has been granted or write_acl is present and user
2965 * has the ability to modify mode. In that case remove
2966 * UID|GID and or MODE from mask so that
2967 * secpolicy_vnode_setattr() doesn't revoke it.
2971 saved_mask
= vap
->va_mask
;
2972 vap
->va_mask
&= ~trim_mask
;
2974 err
= secpolicy_vnode_setattr(cr
, ip
, vap
, &oldva
, flags
,
2975 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
2980 vap
->va_mask
|= saved_mask
;
2984 * secpolicy_vnode_setattr, or take ownership may have
2987 mask
= vap
->va_mask
;
2989 if ((mask
& (ATTR_UID
| ATTR_GID
))) {
2990 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
2991 &xattr_obj
, sizeof (xattr_obj
));
2993 if (err
== 0 && xattr_obj
) {
2994 err
= zfs_zget(ZTOZSB(zp
), xattr_obj
, &attrzp
);
2998 if (mask
& ATTR_UID
) {
2999 new_kuid
= zfs_fuid_create(zfsvfs
,
3000 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
3001 if (new_kuid
!= KUID_TO_SUID(ZTOI(zp
)->i_uid
) &&
3002 zfs_fuid_overquota(zfsvfs
, B_FALSE
, new_kuid
)) {
3005 err
= SET_ERROR(EDQUOT
);
3010 if (mask
& ATTR_GID
) {
3011 new_kgid
= zfs_fuid_create(zfsvfs
,
3012 (uint64_t)vap
->va_gid
, cr
, ZFS_GROUP
, &fuidp
);
3013 if (new_kgid
!= KGID_TO_SGID(ZTOI(zp
)->i_gid
) &&
3014 zfs_fuid_overquota(zfsvfs
, B_TRUE
, new_kgid
)) {
3017 err
= SET_ERROR(EDQUOT
);
3022 tx
= dmu_tx_create(zfsvfs
->z_os
);
3024 if (mask
& ATTR_MODE
) {
3025 uint64_t pmode
= zp
->z_mode
;
3027 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
3029 zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
);
3031 mutex_enter(&zp
->z_lock
);
3032 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
3034 * Are we upgrading ACL from old V0 format
3037 if (zfsvfs
->z_version
>= ZPL_VERSION_FUID
&&
3038 zfs_znode_acl_version(zp
) ==
3039 ZFS_ACL_VERSION_INITIAL
) {
3040 dmu_tx_hold_free(tx
, acl_obj
, 0,
3042 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3043 0, aclp
->z_acl_bytes
);
3045 dmu_tx_hold_write(tx
, acl_obj
, 0,
3048 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3049 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3050 0, aclp
->z_acl_bytes
);
3052 mutex_exit(&zp
->z_lock
);
3053 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3055 if ((mask
& ATTR_XVATTR
) &&
3056 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3057 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3059 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3063 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
3066 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3068 zfs_fuid_txhold(zfsvfs
, tx
);
3070 zfs_sa_upgrade_txholds(tx
, zp
);
3072 err
= dmu_tx_assign(tx
, TXG_WAIT
);
3078 * Set each attribute requested.
3079 * We group settings according to the locks they need to acquire.
3081 * Note: you cannot set ctime directly, although it will be
3082 * updated as a side-effect of calling this function.
3086 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3087 mutex_enter(&zp
->z_acl_lock
);
3088 mutex_enter(&zp
->z_lock
);
3090 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
3091 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
3094 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3095 mutex_enter(&attrzp
->z_acl_lock
);
3096 mutex_enter(&attrzp
->z_lock
);
3097 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3098 SA_ZPL_FLAGS(zfsvfs
), NULL
, &attrzp
->z_pflags
,
3099 sizeof (attrzp
->z_pflags
));
3102 if (mask
& (ATTR_UID
|ATTR_GID
)) {
3104 if (mask
& ATTR_UID
) {
3105 ZTOI(zp
)->i_uid
= SUID_TO_KUID(new_kuid
);
3106 new_uid
= zfs_uid_read(ZTOI(zp
));
3107 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zfsvfs
), NULL
,
3108 &new_uid
, sizeof (new_uid
));
3110 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3111 SA_ZPL_UID(zfsvfs
), NULL
, &new_uid
,
3113 ZTOI(attrzp
)->i_uid
= SUID_TO_KUID(new_uid
);
3117 if (mask
& ATTR_GID
) {
3118 ZTOI(zp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3119 new_gid
= zfs_gid_read(ZTOI(zp
));
3120 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zfsvfs
),
3121 NULL
, &new_gid
, sizeof (new_gid
));
3123 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3124 SA_ZPL_GID(zfsvfs
), NULL
, &new_gid
,
3126 ZTOI(attrzp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3129 if (!(mask
& ATTR_MODE
)) {
3130 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
),
3131 NULL
, &new_mode
, sizeof (new_mode
));
3132 new_mode
= zp
->z_mode
;
3134 err
= zfs_acl_chown_setattr(zp
);
3137 err
= zfs_acl_chown_setattr(attrzp
);
3142 if (mask
& ATTR_MODE
) {
3143 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
), NULL
,
3144 &new_mode
, sizeof (new_mode
));
3145 zp
->z_mode
= ZTOI(zp
)->i_mode
= new_mode
;
3146 ASSERT3P(aclp
, !=, NULL
);
3147 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
3149 if (zp
->z_acl_cached
)
3150 zfs_acl_free(zp
->z_acl_cached
);
3151 zp
->z_acl_cached
= aclp
;
3155 if ((mask
& ATTR_ATIME
) || zp
->z_atime_dirty
) {
3156 zp
->z_atime_dirty
= 0;
3157 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
3158 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
,
3159 &atime
, sizeof (atime
));
3162 if (mask
& (ATTR_MTIME
| ATTR_SIZE
)) {
3163 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3164 ZTOI(zp
)->i_mtime
= timespec_trunc(vap
->va_mtime
,
3165 ZTOI(zp
)->i_sb
->s_time_gran
);
3167 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
3168 mtime
, sizeof (mtime
));
3171 if (mask
& (ATTR_CTIME
| ATTR_SIZE
)) {
3172 ZFS_TIME_ENCODE(&vap
->va_ctime
, ctime
);
3173 ZTOI(zp
)->i_ctime
= timespec_trunc(vap
->va_ctime
,
3174 ZTOI(zp
)->i_sb
->s_time_gran
);
3175 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3176 ctime
, sizeof (ctime
));
3179 if (attrzp
&& mask
) {
3180 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3181 SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
,
3186 * Do this after setting timestamps to prevent timestamp
3187 * update from toggling bit
3190 if (xoap
&& (mask
& ATTR_XVATTR
)) {
3193 * restore trimmed off masks
3194 * so that return masks can be set for caller.
3197 if (XVA_ISSET_REQ(tmpxvattr
, XAT_APPENDONLY
)) {
3198 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3200 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NOUNLINK
)) {
3201 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3203 if (XVA_ISSET_REQ(tmpxvattr
, XAT_IMMUTABLE
)) {
3204 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3206 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NODUMP
)) {
3207 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3209 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_MODIFIED
)) {
3210 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3212 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
)) {
3213 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3216 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3217 ASSERT(S_ISREG(ip
->i_mode
));
3219 zfs_xvattr_set(zp
, xvap
, tx
);
3223 zfs_fuid_sync(zfsvfs
, tx
);
3226 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3228 mutex_exit(&zp
->z_lock
);
3229 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3230 mutex_exit(&zp
->z_acl_lock
);
3233 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3234 mutex_exit(&attrzp
->z_acl_lock
);
3235 mutex_exit(&attrzp
->z_lock
);
3238 if (err
== 0 && attrzp
) {
3239 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3248 zfs_fuid_info_free(fuidp
);
3256 if (err
== ERESTART
)
3259 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3263 zfs_inode_update(zp
);
3267 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3268 zil_commit(zilog
, 0);
3271 kmem_free(xattr_bulk
, sizeof (sa_bulk_attr_t
) * 7);
3272 kmem_free(bulk
, sizeof (sa_bulk_attr_t
) * 7);
3273 kmem_free(tmpxvattr
, sizeof (xvattr_t
));
3278 typedef struct zfs_zlock
{
3279 krwlock_t
*zl_rwlock
; /* lock we acquired */
3280 znode_t
*zl_znode
; /* znode we held */
3281 struct zfs_zlock
*zl_next
; /* next in list */
3285 * Drop locks and release vnodes that were held by zfs_rename_lock().
3288 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3292 while ((zl
= *zlpp
) != NULL
) {
3293 if (zl
->zl_znode
!= NULL
)
3294 zfs_iput_async(ZTOI(zl
->zl_znode
));
3295 rw_exit(zl
->zl_rwlock
);
3296 *zlpp
= zl
->zl_next
;
3297 kmem_free(zl
, sizeof (*zl
));
3302 * Search back through the directory tree, using the ".." entries.
3303 * Lock each directory in the chain to prevent concurrent renames.
3304 * Fail any attempt to move a directory into one of its own descendants.
3305 * XXX - z_parent_lock can overlap with map or grow locks
3308 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3312 uint64_t rootid
= ZTOZSB(zp
)->z_root
;
3313 uint64_t oidp
= zp
->z_id
;
3314 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3315 krw_t rw
= RW_WRITER
;
3318 * First pass write-locks szp and compares to zp->z_id.
3319 * Later passes read-lock zp and compare to zp->z_parent.
3322 if (!rw_tryenter(rwlp
, rw
)) {
3324 * Another thread is renaming in this path.
3325 * Note that if we are a WRITER, we don't have any
3326 * parent_locks held yet.
3328 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3330 * Drop our locks and restart
3332 zfs_rename_unlock(&zl
);
3336 rwlp
= &szp
->z_parent_lock
;
3341 * Wait for other thread to drop its locks
3347 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3348 zl
->zl_rwlock
= rwlp
;
3349 zl
->zl_znode
= NULL
;
3350 zl
->zl_next
= *zlpp
;
3353 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3354 return (SET_ERROR(EINVAL
));
3356 if (oidp
== rootid
) /* We've hit the top */
3359 if (rw
== RW_READER
) { /* i.e. not the first pass */
3360 int error
= zfs_zget(ZTOZSB(zp
), oidp
, &zp
);
3365 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(ZTOZSB(zp
)),
3366 &oidp
, sizeof (oidp
));
3367 rwlp
= &zp
->z_parent_lock
;
3370 } while (zp
->z_id
!= sdzp
->z_id
);
3376 * Move an entry from the provided source directory to the target
3377 * directory. Change the entry name as indicated.
3379 * IN: sdip - Source directory containing the "old entry".
3380 * snm - Old entry name.
3381 * tdip - Target directory to contain the "new entry".
3382 * tnm - New entry name.
3383 * cr - credentials of caller.
3384 * flags - case flags
3386 * RETURN: 0 on success, error code on failure.
3389 * sdip,tdip - ctime|mtime updated
3393 zfs_rename(struct inode
*sdip
, char *snm
, struct inode
*tdip
, char *tnm
,
3394 cred_t
*cr
, int flags
)
3396 znode_t
*tdzp
, *szp
, *tzp
;
3397 znode_t
*sdzp
= ITOZ(sdip
);
3398 zfsvfs_t
*zfsvfs
= ITOZSB(sdip
);
3400 zfs_dirlock_t
*sdl
, *tdl
;
3403 int cmp
, serr
, terr
;
3406 boolean_t waited
= B_FALSE
;
3408 if (snm
== NULL
|| tnm
== NULL
)
3409 return (SET_ERROR(EINVAL
));
3412 ZFS_VERIFY_ZP(sdzp
);
3413 zilog
= zfsvfs
->z_log
;
3416 ZFS_VERIFY_ZP(tdzp
);
3419 * We check i_sb because snapshots and the ctldir must have different
3422 if (tdip
->i_sb
!= sdip
->i_sb
|| zfsctl_is_node(tdip
)) {
3424 return (SET_ERROR(EXDEV
));
3427 if (zfsvfs
->z_utf8
&& u8_validate(tnm
,
3428 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3430 return (SET_ERROR(EILSEQ
));
3433 if (flags
& FIGNORECASE
)
3442 * This is to prevent the creation of links into attribute space
3443 * by renaming a linked file into/outof an attribute directory.
3444 * See the comment in zfs_link() for why this is considered bad.
3446 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3448 return (SET_ERROR(EINVAL
));
3452 * Lock source and target directory entries. To prevent deadlock,
3453 * a lock ordering must be defined. We lock the directory with
3454 * the smallest object id first, or if it's a tie, the one with
3455 * the lexically first name.
3457 if (sdzp
->z_id
< tdzp
->z_id
) {
3459 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3463 * First compare the two name arguments without
3464 * considering any case folding.
3466 int nofold
= (zfsvfs
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3468 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3469 ASSERT(error
== 0 || !zfsvfs
->z_utf8
);
3472 * POSIX: "If the old argument and the new argument
3473 * both refer to links to the same existing file,
3474 * the rename() function shall return successfully
3475 * and perform no other action."
3481 * If the file system is case-folding, then we may
3482 * have some more checking to do. A case-folding file
3483 * system is either supporting mixed case sensitivity
3484 * access or is completely case-insensitive. Note
3485 * that the file system is always case preserving.
3487 * In mixed sensitivity mode case sensitive behavior
3488 * is the default. FIGNORECASE must be used to
3489 * explicitly request case insensitive behavior.
3491 * If the source and target names provided differ only
3492 * by case (e.g., a request to rename 'tim' to 'Tim'),
3493 * we will treat this as a special case in the
3494 * case-insensitive mode: as long as the source name
3495 * is an exact match, we will allow this to proceed as
3496 * a name-change request.
3498 if ((zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
||
3499 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&&
3500 flags
& FIGNORECASE
)) &&
3501 u8_strcmp(snm
, tnm
, 0, zfsvfs
->z_norm
, U8_UNICODE_LATEST
,
3504 * case preserving rename request, require exact
3513 * If the source and destination directories are the same, we should
3514 * grab the z_name_lock of that directory only once.
3518 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3522 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3523 ZEXISTS
| zflg
, NULL
, NULL
);
3524 terr
= zfs_dirent_lock(&tdl
,
3525 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3527 terr
= zfs_dirent_lock(&tdl
,
3528 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3529 serr
= zfs_dirent_lock(&sdl
,
3530 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3536 * Source entry invalid or not there.
3539 zfs_dirent_unlock(tdl
);
3545 rw_exit(&sdzp
->z_name_lock
);
3547 if (strcmp(snm
, "..") == 0)
3553 zfs_dirent_unlock(sdl
);
3557 rw_exit(&sdzp
->z_name_lock
);
3559 if (strcmp(tnm
, "..") == 0)
3566 * Must have write access at the source to remove the old entry
3567 * and write access at the target to create the new entry.
3568 * Note that if target and source are the same, this can be
3569 * done in a single check.
3572 if ((error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
)))
3575 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3577 * Check to make sure rename is valid.
3578 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3580 if ((error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
)))
3585 * Does target exist?
3589 * Source and target must be the same type.
3591 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3592 if (!S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3593 error
= SET_ERROR(ENOTDIR
);
3597 if (S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3598 error
= SET_ERROR(EISDIR
);
3603 * POSIX dictates that when the source and target
3604 * entries refer to the same file object, rename
3605 * must do nothing and exit without error.
3607 if (szp
->z_id
== tzp
->z_id
) {
3613 tx
= dmu_tx_create(zfsvfs
->z_os
);
3614 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3615 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3616 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3617 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3619 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3620 zfs_sa_upgrade_txholds(tx
, tdzp
);
3623 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3624 zfs_sa_upgrade_txholds(tx
, tzp
);
3627 zfs_sa_upgrade_txholds(tx
, szp
);
3628 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
3629 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
3632 zfs_rename_unlock(&zl
);
3633 zfs_dirent_unlock(sdl
);
3634 zfs_dirent_unlock(tdl
);
3637 rw_exit(&sdzp
->z_name_lock
);
3639 if (error
== ERESTART
) {
3656 if (tzp
) /* Attempt to remove the existing target */
3657 error
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3660 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3662 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3664 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zfsvfs
),
3665 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3668 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3670 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3671 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3672 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3675 * At this point, we have successfully created
3676 * the target name, but have failed to remove
3677 * the source name. Since the create was done
3678 * with the ZRENAMING flag, there are
3679 * complications; for one, the link count is
3680 * wrong. The easiest way to deal with this
3681 * is to remove the newly created target, and
3682 * return the original error. This must
3683 * succeed; fortunately, it is very unlikely to
3684 * fail, since we just created it.
3686 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3687 ZRENAMING
, NULL
), ==, 0);
3695 zfs_rename_unlock(&zl
);
3697 zfs_dirent_unlock(sdl
);
3698 zfs_dirent_unlock(tdl
);
3700 zfs_inode_update(sdzp
);
3702 rw_exit(&sdzp
->z_name_lock
);
3705 zfs_inode_update(tdzp
);
3707 zfs_inode_update(szp
);
3710 zfs_inode_update(tzp
);
3714 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3715 zil_commit(zilog
, 0);
3722 * Insert the indicated symbolic reference entry into the directory.
3724 * IN: dip - Directory to contain new symbolic link.
3725 * link - Name for new symlink entry.
3726 * vap - Attributes of new entry.
3727 * target - Target path of new symlink.
3729 * cr - credentials of caller.
3730 * flags - case flags
3732 * RETURN: 0 on success, error code on failure.
3735 * dip - ctime|mtime updated
3739 zfs_symlink(struct inode
*dip
, char *name
, vattr_t
*vap
, char *link
,
3740 struct inode
**ipp
, cred_t
*cr
, int flags
)
3742 znode_t
*zp
, *dzp
= ITOZ(dip
);
3745 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
3747 uint64_t len
= strlen(link
);
3750 zfs_acl_ids_t acl_ids
;
3751 boolean_t fuid_dirtied
;
3752 uint64_t txtype
= TX_SYMLINK
;
3753 boolean_t waited
= B_FALSE
;
3755 ASSERT(S_ISLNK(vap
->va_mode
));
3758 return (SET_ERROR(EINVAL
));
3762 zilog
= zfsvfs
->z_log
;
3764 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
3765 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3767 return (SET_ERROR(EILSEQ
));
3769 if (flags
& FIGNORECASE
)
3772 if (len
> MAXPATHLEN
) {
3774 return (SET_ERROR(ENAMETOOLONG
));
3777 if ((error
= zfs_acl_ids_create(dzp
, 0,
3778 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3786 * Attempt to lock directory; fail if entry already exists.
3788 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3790 zfs_acl_ids_free(&acl_ids
);
3795 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3796 zfs_acl_ids_free(&acl_ids
);
3797 zfs_dirent_unlock(dl
);
3802 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
3803 zfs_acl_ids_free(&acl_ids
);
3804 zfs_dirent_unlock(dl
);
3806 return (SET_ERROR(EDQUOT
));
3808 tx
= dmu_tx_create(zfsvfs
->z_os
);
3809 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3810 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3811 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3812 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3813 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3814 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3815 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3816 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3817 acl_ids
.z_aclp
->z_acl_bytes
);
3820 zfs_fuid_txhold(zfsvfs
, tx
);
3821 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
3823 zfs_dirent_unlock(dl
);
3824 if (error
== ERESTART
) {
3830 zfs_acl_ids_free(&acl_ids
);
3837 * Create a new object for the symlink.
3838 * for version 4 ZPL datsets the symlink will be an SA attribute
3840 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3843 zfs_fuid_sync(zfsvfs
, tx
);
3845 mutex_enter(&zp
->z_lock
);
3847 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zfsvfs
),
3850 zfs_sa_symlink(zp
, link
, len
, tx
);
3851 mutex_exit(&zp
->z_lock
);
3854 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
3855 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3857 * Insert the new object into the directory.
3859 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3861 if (flags
& FIGNORECASE
)
3863 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3865 zfs_inode_update(dzp
);
3866 zfs_inode_update(zp
);
3868 zfs_acl_ids_free(&acl_ids
);
3872 zfs_dirent_unlock(dl
);
3876 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3877 zil_commit(zilog
, 0);
3884 * Return, in the buffer contained in the provided uio structure,
3885 * the symbolic path referred to by ip.
3887 * IN: ip - inode of symbolic link
3888 * uio - structure to contain the link path.
3889 * cr - credentials of caller.
3891 * RETURN: 0 if success
3892 * error code if failure
3895 * ip - atime updated
3899 zfs_readlink(struct inode
*ip
, uio_t
*uio
, cred_t
*cr
)
3901 znode_t
*zp
= ITOZ(ip
);
3902 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
3908 mutex_enter(&zp
->z_lock
);
3910 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3911 SA_ZPL_SYMLINK(zfsvfs
), uio
);
3913 error
= zfs_sa_readlink(zp
, uio
);
3914 mutex_exit(&zp
->z_lock
);
3921 * Insert a new entry into directory tdip referencing sip.
3923 * IN: tdip - Directory to contain new entry.
3924 * sip - inode of new entry.
3925 * name - name of new entry.
3926 * cr - credentials of caller.
3928 * RETURN: 0 if success
3929 * error code if failure
3932 * tdip - ctime|mtime updated
3933 * sip - ctime updated
3937 zfs_link(struct inode
*tdip
, struct inode
*sip
, char *name
, cred_t
*cr
,
3940 znode_t
*dzp
= ITOZ(tdip
);
3942 zfsvfs_t
*zfsvfs
= ITOZSB(tdip
);
3950 boolean_t waited
= B_FALSE
;
3951 boolean_t is_tmpfile
= 0;
3954 is_tmpfile
= (sip
->i_nlink
== 0 && (sip
->i_state
& I_LINKABLE
));
3956 ASSERT(S_ISDIR(tdip
->i_mode
));
3959 return (SET_ERROR(EINVAL
));
3963 zilog
= zfsvfs
->z_log
;
3966 * POSIX dictates that we return EPERM here.
3967 * Better choices include ENOTSUP or EISDIR.
3969 if (S_ISDIR(sip
->i_mode
)) {
3971 return (SET_ERROR(EPERM
));
3978 * We check i_sb because snapshots and the ctldir must have different
3981 if (sip
->i_sb
!= tdip
->i_sb
|| zfsctl_is_node(sip
)) {
3983 return (SET_ERROR(EXDEV
));
3986 /* Prevent links to .zfs/shares files */
3988 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
3989 &parent
, sizeof (uint64_t))) != 0) {
3993 if (parent
== zfsvfs
->z_shares_dir
) {
3995 return (SET_ERROR(EPERM
));
3998 if (zfsvfs
->z_utf8
&& u8_validate(name
,
3999 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
4001 return (SET_ERROR(EILSEQ
));
4003 if (flags
& FIGNORECASE
)
4007 * We do not support links between attributes and non-attributes
4008 * because of the potential security risk of creating links
4009 * into "normal" file space in order to circumvent restrictions
4010 * imposed in attribute space.
4012 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
4014 return (SET_ERROR(EINVAL
));
4017 owner
= zfs_fuid_map_id(zfsvfs
, KUID_TO_SUID(sip
->i_uid
),
4019 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
4021 return (SET_ERROR(EPERM
));
4024 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
4031 * Attempt to lock directory; fail if entry already exists.
4033 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
4039 tx
= dmu_tx_create(zfsvfs
->z_os
);
4040 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
4041 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
4043 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
4045 zfs_sa_upgrade_txholds(tx
, szp
);
4046 zfs_sa_upgrade_txholds(tx
, dzp
);
4047 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
4049 zfs_dirent_unlock(dl
);
4050 if (error
== ERESTART
) {
4060 /* unmark z_unlinked so zfs_link_create will not reject */
4062 szp
->z_unlinked
= 0;
4063 error
= zfs_link_create(dl
, szp
, tx
, 0);
4066 uint64_t txtype
= TX_LINK
;
4068 * tmpfile is created to be in z_unlinkedobj, so remove it.
4069 * Also, we don't log in ZIL, be cause all previous file
4070 * operation on the tmpfile are ignored by ZIL. Instead we
4071 * always wait for txg to sync to make sure all previous
4072 * operation are sync safe.
4075 VERIFY(zap_remove_int(zfsvfs
->z_os
,
4076 zfsvfs
->z_unlinkedobj
, szp
->z_id
, tx
) == 0);
4078 if (flags
& FIGNORECASE
)
4080 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
4082 } else if (is_tmpfile
) {
4083 /* restore z_unlinked since when linking failed */
4084 szp
->z_unlinked
= 1;
4086 txg
= dmu_tx_get_txg(tx
);
4089 zfs_dirent_unlock(dl
);
4091 if (!is_tmpfile
&& zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4092 zil_commit(zilog
, 0);
4095 txg_wait_synced(dmu_objset_pool(zfsvfs
->z_os
), txg
);
4097 zfs_inode_update(dzp
);
4098 zfs_inode_update(szp
);
4104 zfs_putpage_commit_cb(void *arg
)
4106 struct page
*pp
= arg
;
4109 end_page_writeback(pp
);
4113 * Push a page out to disk, once the page is on stable storage the
4114 * registered commit callback will be run as notification of completion.
4116 * IN: ip - page mapped for inode.
4117 * pp - page to push (page is locked)
4118 * wbc - writeback control data
4120 * RETURN: 0 if success
4121 * error code if failure
4124 * ip - ctime|mtime updated
4128 zfs_putpage(struct inode
*ip
, struct page
*pp
, struct writeback_control
*wbc
)
4130 znode_t
*zp
= ITOZ(ip
);
4131 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4139 uint64_t mtime
[2], ctime
[2];
4140 sa_bulk_attr_t bulk
[3];
4142 struct address_space
*mapping
;
4147 ASSERT(PageLocked(pp
));
4149 pgoff
= page_offset(pp
); /* Page byte-offset in file */
4150 offset
= i_size_read(ip
); /* File length in bytes */
4151 pglen
= MIN(PAGE_SIZE
, /* Page length in bytes */
4152 P2ROUNDUP(offset
, PAGE_SIZE
)-pgoff
);
4154 /* Page is beyond end of file */
4155 if (pgoff
>= offset
) {
4161 /* Truncate page length to end of file */
4162 if (pgoff
+ pglen
> offset
)
4163 pglen
= offset
- pgoff
;
4167 * FIXME: Allow mmap writes past its quota. The correct fix
4168 * is to register a page_mkwrite() handler to count the page
4169 * against its quota when it is about to be dirtied.
4171 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
4172 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
4178 * The ordering here is critical and must adhere to the following
4179 * rules in order to avoid deadlocking in either zfs_read() or
4180 * zfs_free_range() due to a lock inversion.
4182 * 1) The page must be unlocked prior to acquiring the range lock.
4183 * This is critical because zfs_read() calls find_lock_page()
4184 * which may block on the page lock while holding the range lock.
4186 * 2) Before setting or clearing write back on a page the range lock
4187 * must be held in order to prevent a lock inversion with the
4188 * zfs_free_range() function.
4190 * This presents a problem because upon entering this function the
4191 * page lock is already held. To safely acquire the range lock the
4192 * page lock must be dropped. This creates a window where another
4193 * process could truncate, invalidate, dirty, or write out the page.
4195 * Therefore, after successfully reacquiring the range and page locks
4196 * the current page state is checked. In the common case everything
4197 * will be as is expected and it can be written out. However, if
4198 * the page state has changed it must be handled accordingly.
4200 mapping
= pp
->mapping
;
4201 redirty_page_for_writepage(wbc
, pp
);
4204 rl
= zfs_range_lock(&zp
->z_range_lock
, pgoff
, pglen
, RL_WRITER
);
4207 /* Page mapping changed or it was no longer dirty, we're done */
4208 if (unlikely((mapping
!= pp
->mapping
) || !PageDirty(pp
))) {
4210 zfs_range_unlock(rl
);
4215 /* Another process started write block if required */
4216 if (PageWriteback(pp
)) {
4218 zfs_range_unlock(rl
);
4220 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
4221 wait_on_page_writeback(pp
);
4227 /* Clear the dirty flag the required locks are held */
4228 if (!clear_page_dirty_for_io(pp
)) {
4230 zfs_range_unlock(rl
);
4236 * Counterpart for redirty_page_for_writepage() above. This page
4237 * was in fact not skipped and should not be counted as if it were.
4239 wbc
->pages_skipped
--;
4240 set_page_writeback(pp
);
4243 tx
= dmu_tx_create(zfsvfs
->z_os
);
4244 dmu_tx_hold_write(tx
, zp
->z_id
, pgoff
, pglen
);
4245 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4246 zfs_sa_upgrade_txholds(tx
, zp
);
4248 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4250 if (err
== ERESTART
)
4254 __set_page_dirty_nobuffers(pp
);
4256 end_page_writeback(pp
);
4257 zfs_range_unlock(rl
);
4263 ASSERT3U(pglen
, <=, PAGE_SIZE
);
4264 dmu_write(zfsvfs
->z_os
, zp
->z_id
, pgoff
, pglen
, va
, tx
);
4267 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
4268 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
4269 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
4272 /* Preserve the mtime and ctime provided by the inode */
4273 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4274 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4275 zp
->z_atime_dirty
= 0;
4278 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4280 zfs_log_write(zfsvfs
->z_log
, tx
, TX_WRITE
, zp
, pgoff
, pglen
, 0,
4281 zfs_putpage_commit_cb
, pp
);
4284 zfs_range_unlock(rl
);
4286 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4288 * Note that this is rarely called under writepages(), because
4289 * writepages() normally handles the entire commit for
4290 * performance reasons.
4292 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
4300 * Update the system attributes when the inode has been dirtied. For the
4301 * moment we only update the mode, atime, mtime, and ctime.
4304 zfs_dirty_inode(struct inode
*ip
, int flags
)
4306 znode_t
*zp
= ITOZ(ip
);
4307 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4309 uint64_t mode
, atime
[2], mtime
[2], ctime
[2];
4310 sa_bulk_attr_t bulk
[4];
4314 if (zfs_is_readonly(zfsvfs
) || dmu_objset_is_snapshot(zfsvfs
->z_os
))
4322 * This is the lazytime semantic indroduced in Linux 4.0
4323 * This flag will only be called from update_time when lazytime is set.
4324 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4325 * Fortunately mtime and ctime are managed within ZFS itself, so we
4326 * only need to dirty atime.
4328 if (flags
== I_DIRTY_TIME
) {
4329 zp
->z_atime_dirty
= 1;
4334 tx
= dmu_tx_create(zfsvfs
->z_os
);
4336 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4337 zfs_sa_upgrade_txholds(tx
, zp
);
4339 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4345 mutex_enter(&zp
->z_lock
);
4346 zp
->z_atime_dirty
= 0;
4348 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MODE(zfsvfs
), NULL
, &mode
, 8);
4349 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_ATIME(zfsvfs
), NULL
, &atime
, 16);
4350 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
4351 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
4353 /* Preserve the mode, mtime and ctime provided by the inode */
4354 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4355 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4356 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4361 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4362 mutex_exit(&zp
->z_lock
);
4372 zfs_inactive(struct inode
*ip
)
4374 znode_t
*zp
= ITOZ(ip
);
4375 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4378 int need_unlock
= 0;
4380 /* Only read lock if we haven't already write locked, e.g. rollback */
4381 if (!RW_WRITE_HELD(&zfsvfs
->z_teardown_inactive_lock
)) {
4383 rw_enter(&zfsvfs
->z_teardown_inactive_lock
, RW_READER
);
4385 if (zp
->z_sa_hdl
== NULL
) {
4387 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4391 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4392 dmu_tx_t
*tx
= dmu_tx_create(zfsvfs
->z_os
);
4394 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4395 zfs_sa_upgrade_txholds(tx
, zp
);
4396 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4400 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4401 mutex_enter(&zp
->z_lock
);
4402 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zfsvfs
),
4403 (void *)&atime
, sizeof (atime
), tx
);
4404 zp
->z_atime_dirty
= 0;
4405 mutex_exit(&zp
->z_lock
);
4412 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4416 * Bounds-check the seek operation.
4418 * IN: ip - inode seeking within
4419 * ooff - old file offset
4420 * noffp - pointer to new file offset
4421 * ct - caller context
4423 * RETURN: 0 if success
4424 * EINVAL if new offset invalid
4428 zfs_seek(struct inode
*ip
, offset_t ooff
, offset_t
*noffp
)
4430 if (S_ISDIR(ip
->i_mode
))
4432 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4436 * Fill pages with data from the disk.
4439 zfs_fillpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4441 znode_t
*zp
= ITOZ(ip
);
4442 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4444 struct page
*cur_pp
;
4445 u_offset_t io_off
, total
;
4452 io_len
= nr_pages
<< PAGE_SHIFT
;
4453 i_size
= i_size_read(ip
);
4454 io_off
= page_offset(pl
[0]);
4456 if (io_off
+ io_len
> i_size
)
4457 io_len
= i_size
- io_off
;
4460 * Iterate over list of pages and read each page individually.
4463 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4466 cur_pp
= pl
[page_idx
++];
4468 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4472 /* convert checksum errors into IO errors */
4474 err
= SET_ERROR(EIO
);
4483 * Uses zfs_fillpage to read data from the file and fill the pages.
4485 * IN: ip - inode of file to get data from.
4486 * pl - list of pages to read
4487 * nr_pages - number of pages to read
4489 * RETURN: 0 on success, error code on failure.
4492 * vp - atime updated
4496 zfs_getpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4498 znode_t
*zp
= ITOZ(ip
);
4499 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4508 err
= zfs_fillpage(ip
, pl
, nr_pages
);
4515 * Check ZFS specific permissions to memory map a section of a file.
4517 * IN: ip - inode of the file to mmap
4519 * addrp - start address in memory region
4520 * len - length of memory region
4521 * vm_flags- address flags
4523 * RETURN: 0 if success
4524 * error code if failure
4528 zfs_map(struct inode
*ip
, offset_t off
, caddr_t
*addrp
, size_t len
,
4529 unsigned long vm_flags
)
4531 znode_t
*zp
= ITOZ(ip
);
4532 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4537 if ((vm_flags
& VM_WRITE
) && (zp
->z_pflags
&
4538 (ZFS_IMMUTABLE
| ZFS_READONLY
| ZFS_APPENDONLY
))) {
4540 return (SET_ERROR(EPERM
));
4543 if ((vm_flags
& (VM_READ
| VM_EXEC
)) &&
4544 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4546 return (SET_ERROR(EACCES
));
4549 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4551 return (SET_ERROR(ENXIO
));
4559 * convoff - converts the given data (start, whence) to the
4563 convoff(struct inode
*ip
, flock64_t
*lckdat
, int whence
, offset_t offset
)
4568 if ((lckdat
->l_whence
== 2) || (whence
== 2)) {
4569 if ((error
= zfs_getattr(ip
, &vap
, 0, CRED())))
4573 switch (lckdat
->l_whence
) {
4575 lckdat
->l_start
+= offset
;
4578 lckdat
->l_start
+= vap
.va_size
;
4583 return (SET_ERROR(EINVAL
));
4586 if (lckdat
->l_start
< 0)
4587 return (SET_ERROR(EINVAL
));
4591 lckdat
->l_start
-= offset
;
4594 lckdat
->l_start
-= vap
.va_size
;
4599 return (SET_ERROR(EINVAL
));
4602 lckdat
->l_whence
= (short)whence
;
4607 * Free or allocate space in a file. Currently, this function only
4608 * supports the `F_FREESP' command. However, this command is somewhat
4609 * misnamed, as its functionality includes the ability to allocate as
4610 * well as free space.
4612 * IN: ip - inode of file to free data in.
4613 * cmd - action to take (only F_FREESP supported).
4614 * bfp - section of file to free/alloc.
4615 * flag - current file open mode flags.
4616 * offset - current file offset.
4617 * cr - credentials of caller [UNUSED].
4619 * RETURN: 0 on success, error code on failure.
4622 * ip - ctime|mtime updated
4626 zfs_space(struct inode
*ip
, int cmd
, flock64_t
*bfp
, int flag
,
4627 offset_t offset
, cred_t
*cr
)
4629 znode_t
*zp
= ITOZ(ip
);
4630 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4637 if (cmd
!= F_FREESP
) {
4639 return (SET_ERROR(EINVAL
));
4643 * Callers might not be able to detect properly that we are read-only,
4644 * so check it explicitly here.
4646 if (zfs_is_readonly(zfsvfs
)) {
4648 return (SET_ERROR(EROFS
));
4651 if ((error
= convoff(ip
, bfp
, 0, offset
))) {
4656 if (bfp
->l_len
< 0) {
4658 return (SET_ERROR(EINVAL
));
4662 * Permissions aren't checked on Solaris because on this OS
4663 * zfs_space() can only be called with an opened file handle.
4664 * On Linux we can get here through truncate_range() which
4665 * operates directly on inodes, so we need to check access rights.
4667 if ((error
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
))) {
4673 len
= bfp
->l_len
; /* 0 means from off to end of file */
4675 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4683 zfs_fid(struct inode
*ip
, fid_t
*fidp
)
4685 znode_t
*zp
= ITOZ(ip
);
4686 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4689 uint64_t object
= zp
->z_id
;
4696 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zfsvfs
),
4697 &gen64
, sizeof (uint64_t))) != 0) {
4702 gen
= (uint32_t)gen64
;
4704 size
= SHORT_FID_LEN
;
4706 zfid
= (zfid_short_t
*)fidp
;
4708 zfid
->zf_len
= size
;
4710 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4711 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4713 /* Must have a non-zero generation number to distinguish from .zfs */
4716 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4717 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4725 zfs_getsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4727 znode_t
*zp
= ITOZ(ip
);
4728 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4730 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4734 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
4742 zfs_setsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4744 znode_t
*zp
= ITOZ(ip
);
4745 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4747 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4748 zilog_t
*zilog
= zfsvfs
->z_log
;
4753 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
4755 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4756 zil_commit(zilog
, 0);
4762 #ifdef HAVE_UIO_ZEROCOPY
4764 * Tunable, both must be a power of 2.
4766 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4767 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4768 * an arcbuf for a partial block read
4770 int zcr_blksz_min
= (1 << 10); /* 1K */
4771 int zcr_blksz_max
= (1 << 17); /* 128K */
4775 zfs_reqzcbuf(struct inode
*ip
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
)
4777 znode_t
*zp
= ITOZ(ip
);
4778 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4779 int max_blksz
= zfsvfs
->z_max_blksz
;
4780 uio_t
*uio
= &xuio
->xu_uio
;
4781 ssize_t size
= uio
->uio_resid
;
4782 offset_t offset
= uio
->uio_loffset
;
4787 int preamble
, postamble
;
4789 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
4790 return (SET_ERROR(EINVAL
));
4797 * Loan out an arc_buf for write if write size is bigger than
4798 * max_blksz, and the file's block size is also max_blksz.
4801 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
4803 return (SET_ERROR(EINVAL
));
4806 * Caller requests buffers for write before knowing where the
4807 * write offset might be (e.g. NFS TCP write).
4812 preamble
= P2PHASE(offset
, blksz
);
4814 preamble
= blksz
- preamble
;
4819 postamble
= P2PHASE(size
, blksz
);
4822 fullblk
= size
/ blksz
;
4823 (void) dmu_xuio_init(xuio
,
4824 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
4827 * Have to fix iov base/len for partial buffers. They
4828 * currently represent full arc_buf's.
4831 /* data begins in the middle of the arc_buf */
4832 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4835 (void) dmu_xuio_add(xuio
, abuf
,
4836 blksz
- preamble
, preamble
);
4839 for (i
= 0; i
< fullblk
; i
++) {
4840 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4843 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
4847 /* data ends in the middle of the arc_buf */
4848 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4851 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
4856 * Loan out an arc_buf for read if the read size is larger than
4857 * the current file block size. Block alignment is not
4858 * considered. Partial arc_buf will be loaned out for read.
4860 blksz
= zp
->z_blksz
;
4861 if (blksz
< zcr_blksz_min
)
4862 blksz
= zcr_blksz_min
;
4863 if (blksz
> zcr_blksz_max
)
4864 blksz
= zcr_blksz_max
;
4865 /* avoid potential complexity of dealing with it */
4866 if (blksz
> max_blksz
) {
4868 return (SET_ERROR(EINVAL
));
4871 maxsize
= zp
->z_size
- uio
->uio_loffset
;
4877 return (SET_ERROR(EINVAL
));
4882 return (SET_ERROR(EINVAL
));
4885 uio
->uio_extflg
= UIO_XUIO
;
4886 XUIO_XUZC_RW(xuio
) = ioflag
;
4893 zfs_retzcbuf(struct inode
*ip
, xuio_t
*xuio
, cred_t
*cr
)
4897 int ioflag
= XUIO_XUZC_RW(xuio
);
4899 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
4901 i
= dmu_xuio_cnt(xuio
);
4903 abuf
= dmu_xuio_arcbuf(xuio
, i
);
4905 * if abuf == NULL, it must be a write buffer
4906 * that has been returned in zfs_write().
4909 dmu_return_arcbuf(abuf
);
4910 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
4913 dmu_xuio_fini(xuio
);
4916 #endif /* HAVE_UIO_ZEROCOPY */
4918 #if defined(_KERNEL) && defined(HAVE_SPL)
4919 EXPORT_SYMBOL(zfs_open
);
4920 EXPORT_SYMBOL(zfs_close
);
4921 EXPORT_SYMBOL(zfs_read
);
4922 EXPORT_SYMBOL(zfs_write
);
4923 EXPORT_SYMBOL(zfs_access
);
4924 EXPORT_SYMBOL(zfs_lookup
);
4925 EXPORT_SYMBOL(zfs_create
);
4926 EXPORT_SYMBOL(zfs_tmpfile
);
4927 EXPORT_SYMBOL(zfs_remove
);
4928 EXPORT_SYMBOL(zfs_mkdir
);
4929 EXPORT_SYMBOL(zfs_rmdir
);
4930 EXPORT_SYMBOL(zfs_readdir
);
4931 EXPORT_SYMBOL(zfs_fsync
);
4932 EXPORT_SYMBOL(zfs_getattr
);
4933 EXPORT_SYMBOL(zfs_getattr_fast
);
4934 EXPORT_SYMBOL(zfs_setattr
);
4935 EXPORT_SYMBOL(zfs_rename
);
4936 EXPORT_SYMBOL(zfs_symlink
);
4937 EXPORT_SYMBOL(zfs_readlink
);
4938 EXPORT_SYMBOL(zfs_link
);
4939 EXPORT_SYMBOL(zfs_inactive
);
4940 EXPORT_SYMBOL(zfs_space
);
4941 EXPORT_SYMBOL(zfs_fid
);
4942 EXPORT_SYMBOL(zfs_getsecattr
);
4943 EXPORT_SYMBOL(zfs_setsecattr
);
4944 EXPORT_SYMBOL(zfs_getpage
);
4945 EXPORT_SYMBOL(zfs_putpage
);
4946 EXPORT_SYMBOL(zfs_dirty_inode
);
4947 EXPORT_SYMBOL(zfs_map
);
4950 module_param(zfs_delete_blocks
, ulong
, 0644);
4951 MODULE_PARM_DESC(zfs_delete_blocks
, "Delete files larger than N blocks async");
4952 module_param(zfs_read_chunk_size
, long, 0644);
4953 MODULE_PARM_DESC(zfs_read_chunk_size
, "Bytes to read per chunk");