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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
28 /* Portions Copyright 2007 Jeremy Teo */
29 /* Portions Copyright 2010 Robert Milkowski */
32 #include <sys/types.h>
33 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/sysmacros.h>
37 #include <sys/resource.h>
39 #include <sys/vfs_opreg.h>
43 #include <sys/taskq.h>
45 #include <sys/vmsystm.h>
46 #include <sys/atomic.h>
48 #include <sys/pathname.h>
49 #include <sys/cmn_err.h>
50 #include <sys/errno.h>
51 #include <sys/unistd.h>
52 #include <sys/zfs_dir.h>
53 #include <sys/zfs_acl.h>
54 #include <sys/zfs_ioctl.h>
55 #include <sys/fs/zfs.h>
57 #include <sys/dmu_objset.h>
63 #include <sys/dirent.h>
64 #include <sys/policy.h>
65 #include <sys/sunddi.h>
68 #include "fs/fs_subr.h"
69 #include <sys/zfs_ctldir.h>
70 #include <sys/zfs_fuid.h>
71 #include <sys/zfs_sa.h>
72 #include <sys/zfs_vnops.h>
74 #include <sys/zfs_rlock.h>
75 #include <sys/extdirent.h>
76 #include <sys/kidmap.h>
84 * Each vnode op performs some logical unit of work. To do this, the ZPL must
85 * properly lock its in-core state, create a DMU transaction, do the work,
86 * record this work in the intent log (ZIL), commit the DMU transaction,
87 * and wait for the intent log to commit if it is a synchronous operation.
88 * Moreover, the vnode ops must work in both normal and log replay context.
89 * The ordering of events is important to avoid deadlocks and references
90 * to freed memory. The example below illustrates the following Big Rules:
92 * (1) A check must be made in each zfs thread for a mounted file system.
93 * This is done avoiding races using ZFS_ENTER(zsb).
94 * A ZFS_EXIT(zsb) is needed before all returns. Any znodes
95 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
96 * can return EIO from the calling function.
98 * (2) iput() should always be the last thing except for zil_commit()
99 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
100 * First, if it's the last reference, the vnode/znode
101 * can be freed, so the zp may point to freed memory. Second, the last
102 * reference will call zfs_zinactive(), which may induce a lot of work --
103 * pushing cached pages (which acquires range locks) and syncing out
104 * cached atime changes. Third, zfs_zinactive() may require a new tx,
105 * which could deadlock the system if you were already holding one.
106 * If you must call iput() within a tx then use zfs_iput_async().
108 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
109 * as they can span dmu_tx_assign() calls.
111 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
112 * dmu_tx_assign(). This is critical because we don't want to block
113 * while holding locks.
115 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
116 * reduces lock contention and CPU usage when we must wait (note that if
117 * throughput is constrained by the storage, nearly every transaction
120 * Note, in particular, that if a lock is sometimes acquired before
121 * the tx assigns, and sometimes after (e.g. z_lock), then failing
122 * to use a non-blocking assign can deadlock the system. The scenario:
124 * Thread A has grabbed a lock before calling dmu_tx_assign().
125 * Thread B is in an already-assigned tx, and blocks for this lock.
126 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
127 * forever, because the previous txg can't quiesce until B's tx commits.
129 * If dmu_tx_assign() returns ERESTART and zsb->z_assign is TXG_NOWAIT,
130 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
131 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
132 * to indicate that this operation has already called dmu_tx_wait().
133 * This will ensure that we don't retry forever, waiting a short bit
136 * (5) If the operation succeeded, generate the intent log entry for it
137 * before dropping locks. This ensures that the ordering of events
138 * in the intent log matches the order in which they actually occurred.
139 * During ZIL replay the zfs_log_* functions will update the sequence
140 * number to indicate the zil transaction has replayed.
142 * (6) At the end of each vnode op, the DMU tx must always commit,
143 * regardless of whether there were any errors.
145 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
146 * to ensure that synchronous semantics are provided when necessary.
148 * In general, this is how things should be ordered in each vnode op:
150 * ZFS_ENTER(zsb); // exit if unmounted
152 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
153 * rw_enter(...); // grab any other locks you need
154 * tx = dmu_tx_create(...); // get DMU tx
155 * dmu_tx_hold_*(); // hold each object you might modify
156 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
158 * rw_exit(...); // drop locks
159 * zfs_dirent_unlock(dl); // unlock directory entry
160 * iput(...); // release held vnodes
161 * if (error == ERESTART) {
167 * dmu_tx_abort(tx); // abort DMU tx
168 * ZFS_EXIT(zsb); // finished in zfs
169 * return (error); // really out of space
171 * error = do_real_work(); // do whatever this VOP does
173 * zfs_log_*(...); // on success, make ZIL entry
174 * dmu_tx_commit(tx); // commit DMU tx -- error or not
175 * rw_exit(...); // drop locks
176 * zfs_dirent_unlock(dl); // unlock directory entry
177 * iput(...); // release held vnodes
178 * zil_commit(zilog, foid); // synchronous when necessary
179 * ZFS_EXIT(zsb); // finished in zfs
180 * return (error); // done, report error
184 * Virus scanning is unsupported. It would be possible to add a hook
185 * here to performance the required virus scan. This could be done
186 * entirely in the kernel or potentially as an update to invoke a
190 zfs_vscan(struct inode
*ip
, cred_t
*cr
, int async
)
197 zfs_open(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
199 znode_t
*zp
= ITOZ(ip
);
200 zfs_sb_t
*zsb
= ITOZSB(ip
);
205 /* Honor ZFS_APPENDONLY file attribute */
206 if ((mode
& FMODE_WRITE
) && (zp
->z_pflags
& ZFS_APPENDONLY
) &&
207 ((flag
& O_APPEND
) == 0)) {
209 return (SET_ERROR(EPERM
));
212 /* Virus scan eligible files on open */
213 if (!zfs_has_ctldir(zp
) && zsb
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
214 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0) {
215 if (zfs_vscan(ip
, cr
, 0) != 0) {
217 return (SET_ERROR(EACCES
));
221 /* Keep a count of the synchronous opens in the znode */
223 atomic_inc_32(&zp
->z_sync_cnt
);
228 EXPORT_SYMBOL(zfs_open
);
232 zfs_close(struct inode
*ip
, int flag
, cred_t
*cr
)
234 znode_t
*zp
= ITOZ(ip
);
235 zfs_sb_t
*zsb
= ITOZSB(ip
);
240 /* Decrement the synchronous opens in the znode */
242 atomic_dec_32(&zp
->z_sync_cnt
);
244 if (!zfs_has_ctldir(zp
) && zsb
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
245 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0)
246 VERIFY(zfs_vscan(ip
, cr
, 1) == 0);
251 EXPORT_SYMBOL(zfs_close
);
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
));
283 * We could find a hole that begins after the logical end-of-file,
284 * because dmu_offset_next() only works on whole blocks. If the
285 * EOF falls mid-block, then indicate that the "virtual hole"
286 * at the end of the file begins at the logical EOF, rather than
287 * at the end of the last block.
289 if (noff
> file_sz
) {
301 zfs_holey(struct inode
*ip
, int cmd
, loff_t
*off
)
303 znode_t
*zp
= ITOZ(ip
);
304 zfs_sb_t
*zsb
= ITOZSB(ip
);
310 error
= zfs_holey_common(ip
, cmd
, off
);
315 EXPORT_SYMBOL(zfs_holey
);
316 #endif /* SEEK_HOLE && SEEK_DATA */
320 * When a file is memory mapped, we must keep the IO data synchronized
321 * between the DMU cache and the memory mapped pages. What this means:
323 * On Write: If we find a memory mapped page, we write to *both*
324 * the page and the dmu buffer.
327 update_pages(struct inode
*ip
, int64_t start
, int len
,
328 objset_t
*os
, uint64_t oid
)
330 struct address_space
*mp
= ip
->i_mapping
;
336 off
= start
& (PAGE_SIZE
-1);
337 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
338 nbytes
= MIN(PAGE_SIZE
- off
, len
);
340 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
342 if (mapping_writably_mapped(mp
))
343 flush_dcache_page(pp
);
346 (void) dmu_read(os
, oid
, start
+off
, nbytes
, pb
+off
,
350 if (mapping_writably_mapped(mp
))
351 flush_dcache_page(pp
);
353 mark_page_accessed(pp
);
366 * When a file is memory mapped, we must keep the IO data synchronized
367 * between the DMU cache and the memory mapped pages. What this means:
369 * On Read: We "read" preferentially from memory mapped pages,
370 * else we default from the dmu buffer.
372 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
373 * the file is memory mapped.
376 mappedread(struct inode
*ip
, int nbytes
, uio_t
*uio
)
378 struct address_space
*mp
= ip
->i_mapping
;
380 znode_t
*zp
= ITOZ(ip
);
387 start
= uio
->uio_loffset
;
388 off
= start
& (PAGE_SIZE
-1);
389 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
390 bytes
= MIN(PAGE_SIZE
- off
, len
);
392 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
394 ASSERT(PageUptodate(pp
));
397 error
= uiomove(pb
+ off
, bytes
, UIO_READ
, uio
);
400 if (mapping_writably_mapped(mp
))
401 flush_dcache_page(pp
);
403 mark_page_accessed(pp
);
407 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
420 unsigned long zfs_read_chunk_size
= 1024 * 1024; /* Tunable */
421 unsigned long zfs_delete_blocks
= DMU_MAX_DELETEBLKCNT
;
424 * Read bytes from specified file into supplied buffer.
426 * IN: ip - inode of file to be read from.
427 * uio - structure supplying read location, range info,
429 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
430 * O_DIRECT flag; used to bypass page cache.
431 * cr - credentials of caller.
433 * OUT: uio - updated offset and range, buffer filled.
435 * RETURN: 0 on success, error code on failure.
438 * inode - atime updated if byte count > 0
442 zfs_read(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
444 znode_t
*zp
= ITOZ(ip
);
445 zfs_sb_t
*zsb
= ITOZSB(ip
);
449 #ifdef HAVE_UIO_ZEROCOPY
451 #endif /* HAVE_UIO_ZEROCOPY */
456 if (zp
->z_pflags
& ZFS_AV_QUARANTINED
) {
458 return (SET_ERROR(EACCES
));
462 * Validate file offset
464 if (uio
->uio_loffset
< (offset_t
)0) {
466 return (SET_ERROR(EINVAL
));
470 * Fasttrack empty reads
472 if (uio
->uio_resid
== 0) {
478 * If we're in FRSYNC mode, sync out this znode before reading it.
480 if (ioflag
& FRSYNC
|| zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
481 zil_commit(zsb
->z_log
, zp
->z_id
);
484 * Lock the range against changes.
486 rl
= zfs_range_lock(&zp
->z_range_lock
, uio
->uio_loffset
, uio
->uio_resid
,
490 * If we are reading past end-of-file we can skip
491 * to the end; but we might still need to set atime.
493 if (uio
->uio_loffset
>= zp
->z_size
) {
498 ASSERT(uio
->uio_loffset
< zp
->z_size
);
499 n
= MIN(uio
->uio_resid
, zp
->z_size
- uio
->uio_loffset
);
501 #ifdef HAVE_UIO_ZEROCOPY
502 if ((uio
->uio_extflg
== UIO_XUIO
) &&
503 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
)) {
505 int blksz
= zp
->z_blksz
;
506 uint64_t offset
= uio
->uio_loffset
;
508 xuio
= (xuio_t
*)uio
;
510 nblk
= (P2ROUNDUP(offset
+ n
, blksz
) - P2ALIGN(offset
,
513 ASSERT(offset
+ n
<= blksz
);
516 (void) dmu_xuio_init(xuio
, nblk
);
518 if (vn_has_cached_data(ip
)) {
520 * For simplicity, we always allocate a full buffer
521 * even if we only expect to read a portion of a block.
523 while (--nblk
>= 0) {
524 (void) dmu_xuio_add(xuio
,
525 dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
530 #endif /* HAVE_UIO_ZEROCOPY */
533 nbytes
= MIN(n
, zfs_read_chunk_size
-
534 P2PHASE(uio
->uio_loffset
, zfs_read_chunk_size
));
536 if (zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
)) {
537 error
= mappedread(ip
, nbytes
, uio
);
539 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
544 /* convert checksum errors into IO errors */
546 error
= SET_ERROR(EIO
);
553 zfs_range_unlock(rl
);
558 EXPORT_SYMBOL(zfs_read
);
561 * Write the bytes to a file.
563 * IN: ip - inode of file to be written to.
564 * uio - structure supplying write location, range info,
566 * ioflag - FAPPEND flag set if in append mode.
567 * O_DIRECT flag; used to bypass page cache.
568 * cr - credentials of caller.
570 * OUT: uio - updated offset and range.
572 * RETURN: 0 if success
573 * error code if failure
576 * ip - ctime|mtime updated if byte count > 0
581 zfs_write(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
583 znode_t
*zp
= ITOZ(ip
);
584 rlim64_t limit
= uio
->uio_limit
;
585 ssize_t start_resid
= uio
->uio_resid
;
589 zfs_sb_t
*zsb
= ZTOZSB(zp
);
594 int max_blksz
= zsb
->z_max_blksz
;
597 const iovec_t
*aiov
= NULL
;
600 const iovec_t
*iovp
= uio
->uio_iov
;
603 sa_bulk_attr_t bulk
[4];
604 uint64_t mtime
[2], ctime
[2];
605 ASSERTV(int iovcnt
= uio
->uio_iovcnt
);
608 * Fasttrack empty write
614 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
620 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
621 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
622 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zsb
), NULL
, &zp
->z_size
, 8);
623 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
627 * Callers might not be able to detect properly that we are read-only,
628 * so check it explicitly here.
630 if (zfs_is_readonly(zsb
)) {
632 return (SET_ERROR(EROFS
));
636 * If immutable or not appending then return EPERM
638 if ((zp
->z_pflags
& (ZFS_IMMUTABLE
| ZFS_READONLY
)) ||
639 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
640 (uio
->uio_loffset
< zp
->z_size
))) {
642 return (SET_ERROR(EPERM
));
648 * Validate file offset
650 woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
653 return (SET_ERROR(EINVAL
));
657 * Pre-fault the pages to ensure slow (eg NFS) pages
659 * Skip this if uio contains loaned arc_buf.
661 #ifdef HAVE_UIO_ZEROCOPY
662 if ((uio
->uio_extflg
== UIO_XUIO
) &&
663 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
))
664 xuio
= (xuio_t
*)uio
;
667 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
670 * If in append mode, set the io offset pointer to eof.
672 if (ioflag
& FAPPEND
) {
674 * Obtain an appending range lock to guarantee file append
675 * semantics. We reset the write offset once we have the lock.
677 rl
= zfs_range_lock(&zp
->z_range_lock
, 0, n
, RL_APPEND
);
679 if (rl
->r_len
== UINT64_MAX
) {
681 * We overlocked the file because this write will cause
682 * the file block size to increase.
683 * Note that zp_size cannot change with this lock held.
687 uio
->uio_loffset
= woff
;
690 * Note that if the file block size will change as a result of
691 * this write, then this range lock will lock the entire file
692 * so that we can re-write the block safely.
694 rl
= zfs_range_lock(&zp
->z_range_lock
, woff
, n
, RL_WRITER
);
698 zfs_range_unlock(rl
);
700 return (SET_ERROR(EFBIG
));
703 if ((woff
+ n
) > limit
|| woff
> (limit
- n
))
706 /* Will this write extend the file length? */
707 write_eof
= (woff
+ n
> zp
->z_size
);
709 end_size
= MAX(zp
->z_size
, woff
+ n
);
712 * Write the file in reasonable size chunks. Each chunk is written
713 * in a separate transaction; this keeps the intent log records small
714 * and allows us to do more fine-grained space accounting.
718 woff
= uio
->uio_loffset
;
719 if (zfs_owner_overquota(zsb
, zp
, B_FALSE
) ||
720 zfs_owner_overquota(zsb
, zp
, B_TRUE
)) {
722 dmu_return_arcbuf(abuf
);
723 error
= SET_ERROR(EDQUOT
);
727 if (xuio
&& abuf
== NULL
) {
728 ASSERT(i_iov
< iovcnt
);
729 ASSERT3U(uio
->uio_segflg
, !=, UIO_BVEC
);
731 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
732 dmu_xuio_clear(xuio
, i_iov
);
733 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
734 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
735 aiov
->iov_len
== arc_buf_size(abuf
)));
737 } else if (abuf
== NULL
&& n
>= max_blksz
&&
738 woff
>= zp
->z_size
&&
739 P2PHASE(woff
, max_blksz
) == 0 &&
740 zp
->z_blksz
== max_blksz
) {
742 * This write covers a full block. "Borrow" a buffer
743 * from the dmu so that we can fill it before we enter
744 * a transaction. This avoids the possibility of
745 * holding up the transaction if the data copy hangs
746 * up on a pagefault (e.g., from an NFS server mapping).
750 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
752 ASSERT(abuf
!= NULL
);
753 ASSERT(arc_buf_size(abuf
) == max_blksz
);
754 if ((error
= uiocopy(abuf
->b_data
, max_blksz
,
755 UIO_WRITE
, uio
, &cbytes
))) {
756 dmu_return_arcbuf(abuf
);
759 ASSERT(cbytes
== max_blksz
);
763 * Start a transaction.
765 tx
= dmu_tx_create(zsb
->z_os
);
766 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
767 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
768 zfs_sa_upgrade_txholds(tx
, zp
);
769 error
= dmu_tx_assign(tx
, TXG_WAIT
);
773 dmu_return_arcbuf(abuf
);
778 * If zfs_range_lock() over-locked we grow the blocksize
779 * and then reduce the lock range. This will only happen
780 * on the first iteration since zfs_range_reduce() will
781 * shrink down r_len to the appropriate size.
783 if (rl
->r_len
== UINT64_MAX
) {
786 if (zp
->z_blksz
> max_blksz
) {
788 * File's blocksize is already larger than the
789 * "recordsize" property. Only let it grow to
790 * the next power of 2.
792 ASSERT(!ISP2(zp
->z_blksz
));
793 new_blksz
= MIN(end_size
,
794 1 << highbit64(zp
->z_blksz
));
796 new_blksz
= MIN(end_size
, max_blksz
);
798 zfs_grow_blocksize(zp
, new_blksz
, tx
);
799 zfs_range_reduce(rl
, woff
, n
);
803 * XXX - should we really limit each write to z_max_blksz?
804 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
806 nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
809 tx_bytes
= uio
->uio_resid
;
810 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
812 tx_bytes
-= uio
->uio_resid
;
815 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
817 * If this is not a full block write, but we are
818 * extending the file past EOF and this data starts
819 * block-aligned, use assign_arcbuf(). Otherwise,
820 * write via dmu_write().
822 if (tx_bytes
< max_blksz
&& (!write_eof
||
823 aiov
->iov_base
!= abuf
->b_data
)) {
825 dmu_write(zsb
->z_os
, zp
->z_id
, woff
,
826 aiov
->iov_len
, aiov
->iov_base
, tx
);
827 dmu_return_arcbuf(abuf
);
828 xuio_stat_wbuf_copied();
830 ASSERT(xuio
|| tx_bytes
== max_blksz
);
831 dmu_assign_arcbuf(sa_get_db(zp
->z_sa_hdl
),
834 ASSERT(tx_bytes
<= uio
->uio_resid
);
835 uioskip(uio
, tx_bytes
);
838 if (tx_bytes
&& zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
))
839 update_pages(ip
, woff
, tx_bytes
, zsb
->z_os
, zp
->z_id
);
842 * If we made no progress, we're done. If we made even
843 * partial progress, update the znode and ZIL accordingly.
846 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
847 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
854 * Clear Set-UID/Set-GID bits on successful write if not
855 * privileged and at least one of the excute bits is set.
857 * It would be nice to to this after all writes have
858 * been done, but that would still expose the ISUID/ISGID
859 * to another app after the partial write is committed.
861 * Note: we don't call zfs_fuid_map_id() here because
862 * user 0 is not an ephemeral uid.
864 mutex_enter(&zp
->z_acl_lock
);
865 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
866 (S_IXUSR
>> 6))) != 0 &&
867 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
868 secpolicy_vnode_setid_retain(cr
,
869 (zp
->z_mode
& S_ISUID
) != 0 && zp
->z_uid
== 0) != 0) {
871 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
872 newmode
= zp
->z_mode
;
873 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zsb
),
874 (void *)&newmode
, sizeof (uint64_t), tx
);
876 mutex_exit(&zp
->z_acl_lock
);
878 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
);
881 * Update the file size (zp_size) if it has changed;
882 * account for possible concurrent updates.
884 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
885 (void) atomic_cas_64(&zp
->z_size
, end_size
,
890 * If we are replaying and eof is non zero then force
891 * the file size to the specified eof. Note, there's no
892 * concurrency during replay.
894 if (zsb
->z_replay
&& zsb
->z_replay_eof
!= 0)
895 zp
->z_size
= zsb
->z_replay_eof
;
897 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
899 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
,
905 ASSERT(tx_bytes
== nbytes
);
909 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
912 zfs_inode_update(zp
);
913 zfs_range_unlock(rl
);
916 * If we're in replay mode, or we made no progress, return error.
917 * Otherwise, it's at least a partial write, so it's successful.
919 if (zsb
->z_replay
|| uio
->uio_resid
== start_resid
) {
924 if (ioflag
& (FSYNC
| FDSYNC
) ||
925 zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
926 zil_commit(zilog
, zp
->z_id
);
931 EXPORT_SYMBOL(zfs_write
);
934 zfs_iput_async(struct inode
*ip
)
936 objset_t
*os
= ITOZSB(ip
)->z_os
;
938 ASSERT(atomic_read(&ip
->i_count
) > 0);
941 if (atomic_read(&ip
->i_count
) == 1)
942 taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os
)),
943 (task_func_t
*)iput
, ip
, TQ_SLEEP
);
949 zfs_get_done(zgd_t
*zgd
, int error
)
951 znode_t
*zp
= zgd
->zgd_private
;
954 dmu_buf_rele(zgd
->zgd_db
, zgd
);
956 zfs_range_unlock(zgd
->zgd_rl
);
959 * Release the vnode asynchronously as we currently have the
960 * txg stopped from syncing.
962 zfs_iput_async(ZTOI(zp
));
964 if (error
== 0 && zgd
->zgd_bp
)
965 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
967 kmem_free(zgd
, sizeof (zgd_t
));
971 static int zil_fault_io
= 0;
975 * Get data to generate a TX_WRITE intent log record.
978 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
981 objset_t
*os
= zsb
->z_os
;
983 uint64_t object
= lr
->lr_foid
;
984 uint64_t offset
= lr
->lr_offset
;
985 uint64_t size
= lr
->lr_length
;
986 blkptr_t
*bp
= &lr
->lr_blkptr
;
995 * Nothing to do if the file has been removed
997 if (zfs_zget(zsb
, object
, &zp
) != 0)
998 return (SET_ERROR(ENOENT
));
999 if (zp
->z_unlinked
) {
1001 * Release the vnode asynchronously as we currently have the
1002 * txg stopped from syncing.
1004 zfs_iput_async(ZTOI(zp
));
1005 return (SET_ERROR(ENOENT
));
1008 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1009 zgd
->zgd_zilog
= zsb
->z_log
;
1010 zgd
->zgd_private
= zp
;
1013 * Write records come in two flavors: immediate and indirect.
1014 * For small writes it's cheaper to store the data with the
1015 * log record (immediate); for large writes it's cheaper to
1016 * sync the data and get a pointer to it (indirect) so that
1017 * we don't have to write the data twice.
1019 if (buf
!= NULL
) { /* immediate write */
1020 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
, size
,
1022 /* test for truncation needs to be done while range locked */
1023 if (offset
>= zp
->z_size
) {
1024 error
= SET_ERROR(ENOENT
);
1026 error
= dmu_read(os
, object
, offset
, size
, buf
,
1027 DMU_READ_NO_PREFETCH
);
1029 ASSERT(error
== 0 || error
== ENOENT
);
1030 } else { /* indirect write */
1032 * Have to lock the whole block to ensure when it's
1033 * written out and it's checksum is being calculated
1034 * that no one can change the data. We need to re-check
1035 * blocksize after we get the lock in case it's changed!
1040 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1042 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
,
1044 if (zp
->z_blksz
== size
)
1047 zfs_range_unlock(zgd
->zgd_rl
);
1049 /* test for truncation needs to be done while range locked */
1050 if (lr
->lr_offset
>= zp
->z_size
)
1051 error
= SET_ERROR(ENOENT
);
1054 error
= SET_ERROR(EIO
);
1059 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1060 DMU_READ_NO_PREFETCH
);
1063 blkptr_t
*obp
= dmu_buf_get_blkptr(db
);
1065 ASSERT(BP_IS_HOLE(bp
));
1072 ASSERT(db
->db_offset
== offset
);
1073 ASSERT(db
->db_size
== size
);
1075 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1077 ASSERT(error
|| lr
->lr_length
<= zp
->z_blksz
);
1080 * On success, we need to wait for the write I/O
1081 * initiated by dmu_sync() to complete before we can
1082 * release this dbuf. We will finish everything up
1083 * in the zfs_get_done() callback.
1088 if (error
== EALREADY
) {
1089 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
1095 zfs_get_done(zgd
, error
);
1102 zfs_access(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
1104 znode_t
*zp
= ITOZ(ip
);
1105 zfs_sb_t
*zsb
= ITOZSB(ip
);
1111 if (flag
& V_ACE_MASK
)
1112 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1114 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1119 EXPORT_SYMBOL(zfs_access
);
1122 * Lookup an entry in a directory, or an extended attribute directory.
1123 * If it exists, return a held inode reference for it.
1125 * IN: dip - inode of directory to search.
1126 * nm - name of entry to lookup.
1127 * flags - LOOKUP_XATTR set if looking for an attribute.
1128 * cr - credentials of caller.
1129 * direntflags - directory lookup flags
1130 * realpnp - returned pathname.
1132 * OUT: ipp - inode of located entry, NULL if not found.
1134 * RETURN: 0 on success, error code on failure.
1141 zfs_lookup(struct inode
*dip
, char *nm
, struct inode
**ipp
, int flags
,
1142 cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1144 znode_t
*zdp
= ITOZ(dip
);
1145 zfs_sb_t
*zsb
= ITOZSB(dip
);
1149 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1151 if (!S_ISDIR(dip
->i_mode
)) {
1152 return (SET_ERROR(ENOTDIR
));
1153 } else if (zdp
->z_sa_hdl
== NULL
) {
1154 return (SET_ERROR(EIO
));
1157 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1158 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1167 vnode_t
*tvp
= dnlc_lookup(dvp
, nm
);
1170 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1175 if (tvp
== DNLC_NO_VNODE
) {
1177 return (SET_ERROR(ENOENT
));
1180 return (specvp_check(vpp
, cr
));
1183 #endif /* HAVE_DNLC */
1192 if (flags
& LOOKUP_XATTR
) {
1194 * We don't allow recursive attributes..
1195 * Maybe someday we will.
1197 if (zdp
->z_pflags
& ZFS_XATTR
) {
1199 return (SET_ERROR(EINVAL
));
1202 if ((error
= zfs_get_xattrdir(zdp
, ipp
, cr
, flags
))) {
1208 * Do we have permission to get into attribute directory?
1211 if ((error
= zfs_zaccess(ITOZ(*ipp
), ACE_EXECUTE
, 0,
1221 if (!S_ISDIR(dip
->i_mode
)) {
1223 return (SET_ERROR(ENOTDIR
));
1227 * Check accessibility of directory.
1230 if ((error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
))) {
1235 if (zsb
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1236 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1238 return (SET_ERROR(EILSEQ
));
1241 error
= zfs_dirlook(zdp
, nm
, ipp
, flags
, direntflags
, realpnp
);
1242 if ((error
== 0) && (*ipp
))
1243 zfs_inode_update(ITOZ(*ipp
));
1248 EXPORT_SYMBOL(zfs_lookup
);
1251 * Attempt to create a new entry in a directory. If the entry
1252 * already exists, truncate the file if permissible, else return
1253 * an error. Return the ip of the created or trunc'd file.
1255 * IN: dip - inode of directory to put new file entry in.
1256 * name - name of new file entry.
1257 * vap - attributes of new file.
1258 * excl - flag indicating exclusive or non-exclusive mode.
1259 * mode - mode to open file with.
1260 * cr - credentials of caller.
1261 * flag - large file flag [UNUSED].
1262 * vsecp - ACL to be set
1264 * OUT: ipp - inode of created or trunc'd entry.
1266 * RETURN: 0 on success, error code on failure.
1269 * dip - ctime|mtime updated if new entry created
1270 * ip - ctime|mtime always, atime if new
1275 zfs_create(struct inode
*dip
, char *name
, vattr_t
*vap
, int excl
,
1276 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1278 znode_t
*zp
, *dzp
= ITOZ(dip
);
1279 zfs_sb_t
*zsb
= ITOZSB(dip
);
1287 zfs_acl_ids_t acl_ids
;
1288 boolean_t fuid_dirtied
;
1289 boolean_t have_acl
= B_FALSE
;
1290 boolean_t waited
= B_FALSE
;
1293 * If we have an ephemeral id, ACL, or XVATTR then
1294 * make sure file system is at proper version
1300 if (zsb
->z_use_fuids
== B_FALSE
&&
1301 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1302 return (SET_ERROR(EINVAL
));
1309 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
1310 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1312 return (SET_ERROR(EILSEQ
));
1315 if (vap
->va_mask
& ATTR_XVATTR
) {
1316 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1317 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1325 if (*name
== '\0') {
1327 * Null component name refers to the directory itself.
1334 /* possible igrab(zp) */
1337 if (flag
& FIGNORECASE
)
1340 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1344 zfs_acl_ids_free(&acl_ids
);
1345 if (strcmp(name
, "..") == 0)
1346 error
= SET_ERROR(EISDIR
);
1356 * Create a new file object and update the directory
1359 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1361 zfs_acl_ids_free(&acl_ids
);
1366 * We only support the creation of regular files in
1367 * extended attribute directories.
1370 if ((dzp
->z_pflags
& ZFS_XATTR
) && !S_ISREG(vap
->va_mode
)) {
1372 zfs_acl_ids_free(&acl_ids
);
1373 error
= SET_ERROR(EINVAL
);
1377 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1378 cr
, vsecp
, &acl_ids
)) != 0)
1382 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1383 zfs_acl_ids_free(&acl_ids
);
1384 error
= SET_ERROR(EDQUOT
);
1388 tx
= dmu_tx_create(os
);
1390 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1391 ZFS_SA_BASE_ATTR_SIZE
);
1393 fuid_dirtied
= zsb
->z_fuid_dirty
;
1395 zfs_fuid_txhold(zsb
, tx
);
1396 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1397 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1398 if (!zsb
->z_use_sa
&&
1399 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1400 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1401 0, acl_ids
.z_aclp
->z_acl_bytes
);
1403 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1405 zfs_dirent_unlock(dl
);
1406 if (error
== ERESTART
) {
1412 zfs_acl_ids_free(&acl_ids
);
1417 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1420 zfs_fuid_sync(zsb
, tx
);
1422 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1423 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1424 if (flag
& FIGNORECASE
)
1426 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1427 vsecp
, acl_ids
.z_fuidp
, vap
);
1428 zfs_acl_ids_free(&acl_ids
);
1431 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1434 zfs_acl_ids_free(&acl_ids
);
1438 * A directory entry already exists for this name.
1441 * Can't truncate an existing file if in exclusive mode.
1444 error
= SET_ERROR(EEXIST
);
1448 * Can't open a directory for writing.
1450 if (S_ISDIR(ZTOI(zp
)->i_mode
)) {
1451 error
= SET_ERROR(EISDIR
);
1455 * Verify requested access to file.
1457 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1461 mutex_enter(&dzp
->z_lock
);
1463 mutex_exit(&dzp
->z_lock
);
1466 * Truncate regular files if requested.
1468 if (S_ISREG(ZTOI(zp
)->i_mode
) &&
1469 (vap
->va_mask
& ATTR_SIZE
) && (vap
->va_size
== 0)) {
1470 /* we can't hold any locks when calling zfs_freesp() */
1471 zfs_dirent_unlock(dl
);
1473 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1479 zfs_dirent_unlock(dl
);
1485 zfs_inode_update(dzp
);
1486 zfs_inode_update(zp
);
1490 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1491 zil_commit(zilog
, 0);
1496 EXPORT_SYMBOL(zfs_create
);
1499 * Remove an entry from a directory.
1501 * IN: dip - inode of directory to remove entry from.
1502 * name - name of entry to remove.
1503 * cr - credentials of caller.
1505 * RETURN: 0 if success
1506 * error code if failure
1510 * ip - ctime (if nlink > 0)
1513 uint64_t null_xattr
= 0;
1517 zfs_remove(struct inode
*dip
, char *name
, cred_t
*cr
, int flags
)
1519 znode_t
*zp
, *dzp
= ITOZ(dip
);
1522 zfs_sb_t
*zsb
= ITOZSB(dip
);
1524 uint64_t acl_obj
, xattr_obj
;
1525 uint64_t xattr_obj_unlinked
= 0;
1530 boolean_t may_delete_now
, delete_now
= FALSE
;
1531 boolean_t unlinked
, toobig
= FALSE
;
1533 pathname_t
*realnmp
= NULL
;
1537 boolean_t waited
= B_FALSE
;
1543 if (flags
& FIGNORECASE
) {
1553 * Attempt to lock directory; fail if entry doesn't exist.
1555 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1565 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1570 * Need to use rmdir for removing directories.
1572 if (S_ISDIR(ip
->i_mode
)) {
1573 error
= SET_ERROR(EPERM
);
1579 dnlc_remove(dvp
, realnmp
->pn_buf
);
1581 dnlc_remove(dvp
, name
);
1582 #endif /* HAVE_DNLC */
1584 mutex_enter(&zp
->z_lock
);
1585 may_delete_now
= atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
);
1586 mutex_exit(&zp
->z_lock
);
1589 * We may delete the znode now, or we may put it in the unlinked set;
1590 * it depends on whether we're the last link, and on whether there are
1591 * other holds on the inode. So we dmu_tx_hold() the right things to
1592 * allow for either case.
1595 tx
= dmu_tx_create(zsb
->z_os
);
1596 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1597 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1598 zfs_sa_upgrade_txholds(tx
, zp
);
1599 zfs_sa_upgrade_txholds(tx
, dzp
);
1600 if (may_delete_now
) {
1601 toobig
= zp
->z_size
> zp
->z_blksz
* zfs_delete_blocks
;
1602 /* if the file is too big, only hold_free a token amount */
1603 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1604 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1607 /* are there any extended attributes? */
1608 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1609 &xattr_obj
, sizeof (xattr_obj
));
1610 if (error
== 0 && xattr_obj
) {
1611 error
= zfs_zget(zsb
, xattr_obj
, &xzp
);
1613 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1614 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1617 mutex_enter(&zp
->z_lock
);
1618 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1619 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1620 mutex_exit(&zp
->z_lock
);
1622 /* charge as an update -- would be nice not to charge at all */
1623 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
1626 * Mark this transaction as typically resulting in a net free of space
1628 dmu_tx_mark_netfree(tx
);
1630 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1632 zfs_dirent_unlock(dl
);
1636 if (error
== ERESTART
) {
1650 * Remove the directory entry.
1652 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1661 * Hold z_lock so that we can make sure that the ACL obj
1662 * hasn't changed. Could have been deleted due to
1665 mutex_enter(&zp
->z_lock
);
1666 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1667 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1668 delete_now
= may_delete_now
&& !toobig
&&
1669 atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
) &&
1670 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1675 if (xattr_obj_unlinked
) {
1676 ASSERT3U(ZTOI(xzp
)->i_nlink
, ==, 2);
1677 mutex_enter(&xzp
->z_lock
);
1678 xzp
->z_unlinked
= 1;
1679 clear_nlink(ZTOI(xzp
));
1681 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zsb
),
1682 &links
, sizeof (links
), tx
);
1683 ASSERT3U(error
, ==, 0);
1684 mutex_exit(&xzp
->z_lock
);
1685 zfs_unlinked_add(xzp
, tx
);
1688 error
= sa_remove(zp
->z_sa_hdl
,
1689 SA_ZPL_XATTR(zsb
), tx
);
1691 error
= sa_update(zp
->z_sa_hdl
,
1692 SA_ZPL_XATTR(zsb
), &null_xattr
,
1693 sizeof (uint64_t), tx
);
1697 * Add to the unlinked set because a new reference could be
1698 * taken concurrently resulting in a deferred destruction.
1700 zfs_unlinked_add(zp
, tx
);
1701 mutex_exit(&zp
->z_lock
);
1702 zfs_inode_update(zp
);
1704 } else if (unlinked
) {
1705 mutex_exit(&zp
->z_lock
);
1706 zfs_unlinked_add(zp
, tx
);
1710 if (flags
& FIGNORECASE
)
1712 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1719 zfs_dirent_unlock(dl
);
1720 zfs_inode_update(dzp
);
1723 zfs_inode_update(zp
);
1728 zfs_inode_update(xzp
);
1729 zfs_iput_async(ZTOI(xzp
));
1732 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1733 zil_commit(zilog
, 0);
1738 EXPORT_SYMBOL(zfs_remove
);
1741 * Create a new directory and insert it into dip using the name
1742 * provided. Return a pointer to the inserted directory.
1744 * IN: dip - inode of directory to add subdir to.
1745 * dirname - name of new directory.
1746 * vap - attributes of new directory.
1747 * cr - credentials of caller.
1748 * vsecp - ACL to be set
1750 * OUT: ipp - inode of created directory.
1752 * RETURN: 0 if success
1753 * error code if failure
1756 * dip - ctime|mtime updated
1757 * ipp - ctime|mtime|atime updated
1761 zfs_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
, struct inode
**ipp
,
1762 cred_t
*cr
, int flags
, vsecattr_t
*vsecp
)
1764 znode_t
*zp
, *dzp
= ITOZ(dip
);
1765 zfs_sb_t
*zsb
= ITOZSB(dip
);
1773 gid_t gid
= crgetgid(cr
);
1774 zfs_acl_ids_t acl_ids
;
1775 boolean_t fuid_dirtied
;
1776 boolean_t waited
= B_FALSE
;
1778 ASSERT(S_ISDIR(vap
->va_mode
));
1781 * If we have an ephemeral id, ACL, or XVATTR then
1782 * make sure file system is at proper version
1786 if (zsb
->z_use_fuids
== B_FALSE
&&
1787 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1788 return (SET_ERROR(EINVAL
));
1794 if (dzp
->z_pflags
& ZFS_XATTR
) {
1796 return (SET_ERROR(EINVAL
));
1799 if (zsb
->z_utf8
&& u8_validate(dirname
,
1800 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1802 return (SET_ERROR(EILSEQ
));
1804 if (flags
& FIGNORECASE
)
1807 if (vap
->va_mask
& ATTR_XVATTR
) {
1808 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1809 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1815 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1816 vsecp
, &acl_ids
)) != 0) {
1821 * First make sure the new directory doesn't exist.
1823 * Existence is checked first to make sure we don't return
1824 * EACCES instead of EEXIST which can cause some applications
1830 if ((error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1832 zfs_acl_ids_free(&acl_ids
);
1837 if ((error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
))) {
1838 zfs_acl_ids_free(&acl_ids
);
1839 zfs_dirent_unlock(dl
);
1844 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1845 zfs_acl_ids_free(&acl_ids
);
1846 zfs_dirent_unlock(dl
);
1848 return (SET_ERROR(EDQUOT
));
1852 * Add a new entry to the directory.
1854 tx
= dmu_tx_create(zsb
->z_os
);
1855 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
1856 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1857 fuid_dirtied
= zsb
->z_fuid_dirty
;
1859 zfs_fuid_txhold(zsb
, tx
);
1860 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1861 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
1862 acl_ids
.z_aclp
->z_acl_bytes
);
1865 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1866 ZFS_SA_BASE_ATTR_SIZE
);
1868 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1870 zfs_dirent_unlock(dl
);
1871 if (error
== ERESTART
) {
1877 zfs_acl_ids_free(&acl_ids
);
1886 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1889 zfs_fuid_sync(zsb
, tx
);
1892 * Now put new name in parent dir.
1894 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1898 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
1899 if (flags
& FIGNORECASE
)
1901 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
1902 acl_ids
.z_fuidp
, vap
);
1904 zfs_acl_ids_free(&acl_ids
);
1908 zfs_dirent_unlock(dl
);
1910 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1911 zil_commit(zilog
, 0);
1913 zfs_inode_update(dzp
);
1914 zfs_inode_update(zp
);
1918 EXPORT_SYMBOL(zfs_mkdir
);
1921 * Remove a directory subdir entry. If the current working
1922 * directory is the same as the subdir to be removed, the
1925 * IN: dip - inode of directory to remove from.
1926 * name - name of directory to be removed.
1927 * cwd - inode of current working directory.
1928 * cr - credentials of caller.
1929 * flags - case flags
1931 * RETURN: 0 on success, error code on failure.
1934 * dip - ctime|mtime updated
1938 zfs_rmdir(struct inode
*dip
, char *name
, struct inode
*cwd
, cred_t
*cr
,
1941 znode_t
*dzp
= ITOZ(dip
);
1944 zfs_sb_t
*zsb
= ITOZSB(dip
);
1950 boolean_t waited
= B_FALSE
;
1956 if (flags
& FIGNORECASE
)
1962 * Attempt to lock directory; fail if entry doesn't exist.
1964 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1972 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1976 if (!S_ISDIR(ip
->i_mode
)) {
1977 error
= SET_ERROR(ENOTDIR
);
1982 error
= SET_ERROR(EINVAL
);
1987 * Grab a lock on the directory to make sure that noone is
1988 * trying to add (or lookup) entries while we are removing it.
1990 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
1993 * Grab a lock on the parent pointer to make sure we play well
1994 * with the treewalk and directory rename code.
1996 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
1998 tx
= dmu_tx_create(zsb
->z_os
);
1999 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2000 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2001 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
2002 zfs_sa_upgrade_txholds(tx
, zp
);
2003 zfs_sa_upgrade_txholds(tx
, dzp
);
2004 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
2006 rw_exit(&zp
->z_parent_lock
);
2007 rw_exit(&zp
->z_name_lock
);
2008 zfs_dirent_unlock(dl
);
2010 if (error
== ERESTART
) {
2021 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2024 uint64_t txtype
= TX_RMDIR
;
2025 if (flags
& FIGNORECASE
)
2027 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2032 rw_exit(&zp
->z_parent_lock
);
2033 rw_exit(&zp
->z_name_lock
);
2035 zfs_dirent_unlock(dl
);
2037 zfs_inode_update(dzp
);
2038 zfs_inode_update(zp
);
2041 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2042 zil_commit(zilog
, 0);
2047 EXPORT_SYMBOL(zfs_rmdir
);
2050 * Read as many directory entries as will fit into the provided
2051 * dirent buffer from the given directory cursor position.
2053 * IN: ip - inode of directory to read.
2054 * dirent - buffer for directory entries.
2056 * OUT: dirent - filler buffer of directory entries.
2058 * RETURN: 0 if success
2059 * error code if failure
2062 * ip - atime updated
2064 * Note that the low 4 bits of the cookie returned by zap is always zero.
2065 * This allows us to use the low range for "special" directory entries:
2066 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2067 * we use the offset 2 for the '.zfs' directory.
2071 zfs_readdir(struct inode
*ip
, struct dir_context
*ctx
, cred_t
*cr
)
2073 znode_t
*zp
= ITOZ(ip
);
2074 zfs_sb_t
*zsb
= ITOZSB(ip
);
2077 zap_attribute_t zap
;
2083 uint64_t offset
; /* must be unsigned; checks for < 1 */
2088 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
2089 &parent
, sizeof (parent
))) != 0)
2093 * Quit if directory has been removed (posix)
2101 prefetch
= zp
->z_zn_prefetch
;
2104 * Initialize the iterator cursor.
2108 * Start iteration from the beginning of the directory.
2110 zap_cursor_init(&zc
, os
, zp
->z_id
);
2113 * The offset is a serialized cursor.
2115 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2119 * Transform to file-system independent format
2124 * Special case `.', `..', and `.zfs'.
2127 (void) strcpy(zap
.za_name
, ".");
2128 zap
.za_normalization_conflict
= 0;
2131 } else if (offset
== 1) {
2132 (void) strcpy(zap
.za_name
, "..");
2133 zap
.za_normalization_conflict
= 0;
2136 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2137 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2138 zap
.za_normalization_conflict
= 0;
2139 objnum
= ZFSCTL_INO_ROOT
;
2145 if ((error
= zap_cursor_retrieve(&zc
, &zap
))) {
2146 if (error
== ENOENT
)
2153 * Allow multiple entries provided the first entry is
2154 * the object id. Non-zpl consumers may safely make
2155 * use of the additional space.
2157 * XXX: This should be a feature flag for compatibility
2159 if (zap
.za_integer_length
!= 8 ||
2160 zap
.za_num_integers
== 0) {
2161 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2162 "entry, obj = %lld, offset = %lld, "
2163 "length = %d, num = %lld\n",
2164 (u_longlong_t
)zp
->z_id
,
2165 (u_longlong_t
)offset
,
2166 zap
.za_integer_length
,
2167 (u_longlong_t
)zap
.za_num_integers
);
2168 error
= SET_ERROR(ENXIO
);
2172 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2173 type
= ZFS_DIRENT_TYPE(zap
.za_first_integer
);
2176 done
= !dir_emit(ctx
, zap
.za_name
, strlen(zap
.za_name
),
2181 /* Prefetch znode */
2183 dmu_prefetch(os
, objnum
, 0, 0, 0,
2184 ZIO_PRIORITY_SYNC_READ
);
2188 * Move to the next entry, fill in the previous offset.
2190 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2191 zap_cursor_advance(&zc
);
2192 offset
= zap_cursor_serialize(&zc
);
2198 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2201 zap_cursor_fini(&zc
);
2202 if (error
== ENOENT
)
2209 EXPORT_SYMBOL(zfs_readdir
);
2211 ulong_t zfs_fsync_sync_cnt
= 4;
2214 zfs_fsync(struct inode
*ip
, int syncflag
, cred_t
*cr
)
2216 znode_t
*zp
= ITOZ(ip
);
2217 zfs_sb_t
*zsb
= ITOZSB(ip
);
2219 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2221 if (zsb
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2224 zil_commit(zsb
->z_log
, zp
->z_id
);
2227 tsd_set(zfs_fsyncer_key
, NULL
);
2231 EXPORT_SYMBOL(zfs_fsync
);
2235 * Get the requested file attributes and place them in the provided
2238 * IN: ip - inode of file.
2239 * vap - va_mask identifies requested attributes.
2240 * If ATTR_XVATTR set, then optional attrs are requested
2241 * flags - ATTR_NOACLCHECK (CIFS server context)
2242 * cr - credentials of caller.
2244 * OUT: vap - attribute values.
2246 * RETURN: 0 (always succeeds)
2250 zfs_getattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2252 znode_t
*zp
= ITOZ(ip
);
2253 zfs_sb_t
*zsb
= ITOZSB(ip
);
2256 uint64_t atime
[2], mtime
[2], ctime
[2];
2257 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2258 xoptattr_t
*xoap
= NULL
;
2259 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2260 sa_bulk_attr_t bulk
[3];
2266 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2268 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zsb
), NULL
, &atime
, 16);
2269 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
2270 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
2272 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2278 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2279 * Also, if we are the owner don't bother, since owner should
2280 * always be allowed to read basic attributes of file.
2282 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2283 (vap
->va_uid
!= crgetuid(cr
))) {
2284 if ((error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2292 * Return all attributes. It's cheaper to provide the answer
2293 * than to determine whether we were asked the question.
2296 mutex_enter(&zp
->z_lock
);
2297 vap
->va_type
= vn_mode_to_vtype(zp
->z_mode
);
2298 vap
->va_mode
= zp
->z_mode
;
2299 vap
->va_fsid
= ZTOI(zp
)->i_sb
->s_dev
;
2300 vap
->va_nodeid
= zp
->z_id
;
2301 if ((zp
->z_id
== zsb
->z_root
) && zfs_show_ctldir(zp
))
2302 links
= ZTOI(zp
)->i_nlink
+ 1;
2304 links
= ZTOI(zp
)->i_nlink
;
2305 vap
->va_nlink
= MIN(links
, ZFS_LINK_MAX
);
2306 vap
->va_size
= i_size_read(ip
);
2307 vap
->va_rdev
= ip
->i_rdev
;
2308 vap
->va_seq
= ip
->i_generation
;
2311 * Add in any requested optional attributes and the create time.
2312 * Also set the corresponding bits in the returned attribute bitmap.
2314 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zsb
->z_use_fuids
) {
2315 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2317 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2318 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2321 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2322 xoap
->xoa_readonly
=
2323 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2324 XVA_SET_RTN(xvap
, XAT_READONLY
);
2327 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2329 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2330 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2333 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2335 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2336 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2339 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2340 xoap
->xoa_nounlink
=
2341 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2342 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2345 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2346 xoap
->xoa_immutable
=
2347 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2348 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2351 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2352 xoap
->xoa_appendonly
=
2353 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2354 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2357 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2359 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2360 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2363 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2365 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2366 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2369 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2370 xoap
->xoa_av_quarantined
=
2371 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2372 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2375 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2376 xoap
->xoa_av_modified
=
2377 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2378 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2381 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2382 S_ISREG(ip
->i_mode
)) {
2383 zfs_sa_get_scanstamp(zp
, xvap
);
2386 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2389 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zsb
),
2390 times
, sizeof (times
));
2391 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2392 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2395 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2396 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2397 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2399 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2400 xoap
->xoa_generation
= ip
->i_generation
;
2401 XVA_SET_RTN(xvap
, XAT_GEN
);
2404 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2406 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2407 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2410 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2412 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2413 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2417 ZFS_TIME_DECODE(&vap
->va_atime
, atime
);
2418 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2419 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2421 mutex_exit(&zp
->z_lock
);
2423 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2425 if (zp
->z_blksz
== 0) {
2427 * Block size hasn't been set; suggest maximal I/O transfers.
2429 vap
->va_blksize
= zsb
->z_max_blksz
;
2435 EXPORT_SYMBOL(zfs_getattr
);
2438 * Get the basic file attributes and place them in the provided kstat
2439 * structure. The inode is assumed to be the authoritative source
2440 * for most of the attributes. However, the znode currently has the
2441 * authoritative atime, blksize, and block count.
2443 * IN: ip - inode of file.
2445 * OUT: sp - kstat values.
2447 * RETURN: 0 (always succeeds)
2451 zfs_getattr_fast(struct inode
*ip
, struct kstat
*sp
)
2453 znode_t
*zp
= ITOZ(ip
);
2454 zfs_sb_t
*zsb
= ITOZSB(ip
);
2456 u_longlong_t nblocks
;
2461 mutex_enter(&zp
->z_lock
);
2463 generic_fillattr(ip
, sp
);
2465 sa_object_size(zp
->z_sa_hdl
, &blksize
, &nblocks
);
2466 sp
->blksize
= blksize
;
2467 sp
->blocks
= nblocks
;
2469 if (unlikely(zp
->z_blksz
== 0)) {
2471 * Block size hasn't been set; suggest maximal I/O transfers.
2473 sp
->blksize
= zsb
->z_max_blksz
;
2476 mutex_exit(&zp
->z_lock
);
2479 * Required to prevent NFS client from detecting different inode
2480 * numbers of snapshot root dentry before and after snapshot mount.
2482 if (zsb
->z_issnap
) {
2483 if (ip
->i_sb
->s_root
->d_inode
== ip
)
2484 sp
->ino
= ZFSCTL_INO_SNAPDIRS
-
2485 dmu_objset_id(zsb
->z_os
);
2492 EXPORT_SYMBOL(zfs_getattr_fast
);
2495 * Set the file attributes to the values contained in the
2498 * IN: ip - inode of file to be modified.
2499 * vap - new attribute values.
2500 * If ATTR_XVATTR set, then optional attrs are being set
2501 * flags - ATTR_UTIME set if non-default time values provided.
2502 * - ATTR_NOACLCHECK (CIFS context only).
2503 * cr - credentials of caller.
2505 * RETURN: 0 if success
2506 * error code if failure
2509 * ip - ctime updated, mtime updated if size changed.
2513 zfs_setattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2515 znode_t
*zp
= ITOZ(ip
);
2516 zfs_sb_t
*zsb
= ITOZSB(ip
);
2520 xvattr_t
*tmpxvattr
;
2521 uint_t mask
= vap
->va_mask
;
2522 uint_t saved_mask
= 0;
2525 uint64_t new_uid
, new_gid
;
2527 uint64_t mtime
[2], ctime
[2], atime
[2];
2529 int need_policy
= FALSE
;
2531 zfs_fuid_info_t
*fuidp
= NULL
;
2532 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2535 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2536 boolean_t fuid_dirtied
= B_FALSE
;
2537 sa_bulk_attr_t
*bulk
, *xattr_bulk
;
2538 int count
= 0, xattr_count
= 0;
2549 * Make sure that if we have ephemeral uid/gid or xvattr specified
2550 * that file system is at proper version level
2553 if (zsb
->z_use_fuids
== B_FALSE
&&
2554 (((mask
& ATTR_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2555 ((mask
& ATTR_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2556 (mask
& ATTR_XVATTR
))) {
2558 return (SET_ERROR(EINVAL
));
2561 if (mask
& ATTR_SIZE
&& S_ISDIR(ip
->i_mode
)) {
2563 return (SET_ERROR(EISDIR
));
2566 if (mask
& ATTR_SIZE
&& !S_ISREG(ip
->i_mode
) && !S_ISFIFO(ip
->i_mode
)) {
2568 return (SET_ERROR(EINVAL
));
2572 * If this is an xvattr_t, then get a pointer to the structure of
2573 * optional attributes. If this is NULL, then we have a vattr_t.
2575 xoap
= xva_getxoptattr(xvap
);
2577 tmpxvattr
= kmem_alloc(sizeof (xvattr_t
), KM_SLEEP
);
2578 xva_init(tmpxvattr
);
2580 bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2581 xattr_bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2584 * Immutable files can only alter immutable bit and atime
2586 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2587 ((mask
& (ATTR_SIZE
|ATTR_UID
|ATTR_GID
|ATTR_MTIME
|ATTR_MODE
)) ||
2588 ((mask
& ATTR_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2593 if ((mask
& ATTR_SIZE
) && (zp
->z_pflags
& ZFS_READONLY
)) {
2599 * Verify timestamps doesn't overflow 32 bits.
2600 * ZFS can handle large timestamps, but 32bit syscalls can't
2601 * handle times greater than 2039. This check should be removed
2602 * once large timestamps are fully supported.
2604 if (mask
& (ATTR_ATIME
| ATTR_MTIME
)) {
2605 if (((mask
& ATTR_ATIME
) &&
2606 TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2607 ((mask
& ATTR_MTIME
) &&
2608 TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2618 /* Can this be moved to before the top label? */
2619 if (zfs_is_readonly(zsb
)) {
2625 * First validate permissions
2628 if (mask
& ATTR_SIZE
) {
2629 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2634 * XXX - Note, we are not providing any open
2635 * mode flags here (like FNDELAY), so we may
2636 * block if there are locks present... this
2637 * should be addressed in openat().
2639 /* XXX - would it be OK to generate a log record here? */
2640 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2645 if (mask
& (ATTR_ATIME
|ATTR_MTIME
) ||
2646 ((mask
& ATTR_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2647 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2648 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2649 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2650 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2651 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2652 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2653 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2657 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2658 int idmask
= (mask
& (ATTR_UID
|ATTR_GID
));
2663 * NOTE: even if a new mode is being set,
2664 * we may clear S_ISUID/S_ISGID bits.
2667 if (!(mask
& ATTR_MODE
))
2668 vap
->va_mode
= zp
->z_mode
;
2671 * Take ownership or chgrp to group we are a member of
2674 take_owner
= (mask
& ATTR_UID
) && (vap
->va_uid
== crgetuid(cr
));
2675 take_group
= (mask
& ATTR_GID
) &&
2676 zfs_groupmember(zsb
, vap
->va_gid
, cr
);
2679 * If both ATTR_UID and ATTR_GID are set then take_owner and
2680 * take_group must both be set in order to allow taking
2683 * Otherwise, send the check through secpolicy_vnode_setattr()
2687 if (((idmask
== (ATTR_UID
|ATTR_GID
)) &&
2688 take_owner
&& take_group
) ||
2689 ((idmask
== ATTR_UID
) && take_owner
) ||
2690 ((idmask
== ATTR_GID
) && take_group
)) {
2691 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2692 skipaclchk
, cr
) == 0) {
2694 * Remove setuid/setgid for non-privileged users
2696 (void) secpolicy_setid_clear(vap
, cr
);
2697 trim_mask
= (mask
& (ATTR_UID
|ATTR_GID
));
2706 mutex_enter(&zp
->z_lock
);
2707 oldva
.va_mode
= zp
->z_mode
;
2708 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2709 if (mask
& ATTR_XVATTR
) {
2711 * Update xvattr mask to include only those attributes
2712 * that are actually changing.
2714 * the bits will be restored prior to actually setting
2715 * the attributes so the caller thinks they were set.
2717 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2718 if (xoap
->xoa_appendonly
!=
2719 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2722 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2723 XVA_SET_REQ(tmpxvattr
, XAT_APPENDONLY
);
2727 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2728 if (xoap
->xoa_nounlink
!=
2729 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2732 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2733 XVA_SET_REQ(tmpxvattr
, XAT_NOUNLINK
);
2737 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2738 if (xoap
->xoa_immutable
!=
2739 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2742 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2743 XVA_SET_REQ(tmpxvattr
, XAT_IMMUTABLE
);
2747 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2748 if (xoap
->xoa_nodump
!=
2749 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2752 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2753 XVA_SET_REQ(tmpxvattr
, XAT_NODUMP
);
2757 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2758 if (xoap
->xoa_av_modified
!=
2759 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2762 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2763 XVA_SET_REQ(tmpxvattr
, XAT_AV_MODIFIED
);
2767 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2768 if ((!S_ISREG(ip
->i_mode
) &&
2769 xoap
->xoa_av_quarantined
) ||
2770 xoap
->xoa_av_quarantined
!=
2771 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2774 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2775 XVA_SET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
);
2779 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2780 mutex_exit(&zp
->z_lock
);
2785 if (need_policy
== FALSE
&&
2786 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
2787 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
2792 mutex_exit(&zp
->z_lock
);
2794 if (mask
& ATTR_MODE
) {
2795 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
2796 err
= secpolicy_setid_setsticky_clear(ip
, vap
,
2801 trim_mask
|= ATTR_MODE
;
2809 * If trim_mask is set then take ownership
2810 * has been granted or write_acl is present and user
2811 * has the ability to modify mode. In that case remove
2812 * UID|GID and or MODE from mask so that
2813 * secpolicy_vnode_setattr() doesn't revoke it.
2817 saved_mask
= vap
->va_mask
;
2818 vap
->va_mask
&= ~trim_mask
;
2820 err
= secpolicy_vnode_setattr(cr
, ip
, vap
, &oldva
, flags
,
2821 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
2826 vap
->va_mask
|= saved_mask
;
2830 * secpolicy_vnode_setattr, or take ownership may have
2833 mask
= vap
->va_mask
;
2835 if ((mask
& (ATTR_UID
| ATTR_GID
))) {
2836 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
2837 &xattr_obj
, sizeof (xattr_obj
));
2839 if (err
== 0 && xattr_obj
) {
2840 err
= zfs_zget(ZTOZSB(zp
), xattr_obj
, &attrzp
);
2844 if (mask
& ATTR_UID
) {
2845 new_uid
= zfs_fuid_create(zsb
,
2846 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
2847 if (new_uid
!= zp
->z_uid
&&
2848 zfs_fuid_overquota(zsb
, B_FALSE
, new_uid
)) {
2856 if (mask
& ATTR_GID
) {
2857 new_gid
= zfs_fuid_create(zsb
, (uint64_t)vap
->va_gid
,
2858 cr
, ZFS_GROUP
, &fuidp
);
2859 if (new_gid
!= zp
->z_gid
&&
2860 zfs_fuid_overquota(zsb
, B_TRUE
, new_gid
)) {
2868 tx
= dmu_tx_create(zsb
->z_os
);
2870 if (mask
& ATTR_MODE
) {
2871 uint64_t pmode
= zp
->z_mode
;
2873 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
2875 zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
);
2877 mutex_enter(&zp
->z_lock
);
2878 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
2880 * Are we upgrading ACL from old V0 format
2883 if (zsb
->z_version
>= ZPL_VERSION_FUID
&&
2884 zfs_znode_acl_version(zp
) ==
2885 ZFS_ACL_VERSION_INITIAL
) {
2886 dmu_tx_hold_free(tx
, acl_obj
, 0,
2888 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
2889 0, aclp
->z_acl_bytes
);
2891 dmu_tx_hold_write(tx
, acl_obj
, 0,
2894 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
2895 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
2896 0, aclp
->z_acl_bytes
);
2898 mutex_exit(&zp
->z_lock
);
2899 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
2901 if ((mask
& ATTR_XVATTR
) &&
2902 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
2903 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
2905 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2909 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
2912 fuid_dirtied
= zsb
->z_fuid_dirty
;
2914 zfs_fuid_txhold(zsb
, tx
);
2916 zfs_sa_upgrade_txholds(tx
, zp
);
2918 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2924 * Set each attribute requested.
2925 * We group settings according to the locks they need to acquire.
2927 * Note: you cannot set ctime directly, although it will be
2928 * updated as a side-effect of calling this function.
2932 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
2933 mutex_enter(&zp
->z_acl_lock
);
2934 mutex_enter(&zp
->z_lock
);
2936 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
2937 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
2940 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
2941 mutex_enter(&attrzp
->z_acl_lock
);
2942 mutex_enter(&attrzp
->z_lock
);
2943 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2944 SA_ZPL_FLAGS(zsb
), NULL
, &attrzp
->z_pflags
,
2945 sizeof (attrzp
->z_pflags
));
2948 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2950 if (mask
& ATTR_UID
) {
2951 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zsb
), NULL
,
2952 &new_uid
, sizeof (new_uid
));
2953 zp
->z_uid
= new_uid
;
2955 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2956 SA_ZPL_UID(zsb
), NULL
, &new_uid
,
2958 attrzp
->z_uid
= new_uid
;
2962 if (mask
& ATTR_GID
) {
2963 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zsb
),
2964 NULL
, &new_gid
, sizeof (new_gid
));
2965 zp
->z_gid
= new_gid
;
2967 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2968 SA_ZPL_GID(zsb
), NULL
, &new_gid
,
2970 attrzp
->z_gid
= new_gid
;
2973 if (!(mask
& ATTR_MODE
)) {
2974 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
),
2975 NULL
, &new_mode
, sizeof (new_mode
));
2976 new_mode
= zp
->z_mode
;
2978 err
= zfs_acl_chown_setattr(zp
);
2981 err
= zfs_acl_chown_setattr(attrzp
);
2986 if (mask
& ATTR_MODE
) {
2987 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
), NULL
,
2988 &new_mode
, sizeof (new_mode
));
2989 zp
->z_mode
= new_mode
;
2990 ASSERT3P(aclp
, !=, NULL
);
2991 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
2993 if (zp
->z_acl_cached
)
2994 zfs_acl_free(zp
->z_acl_cached
);
2995 zp
->z_acl_cached
= aclp
;
3000 if ((mask
& ATTR_ATIME
) || zp
->z_atime_dirty
) {
3001 zp
->z_atime_dirty
= 0;
3002 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
3003 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zsb
), NULL
,
3004 &atime
, sizeof (atime
));
3007 if (mask
& ATTR_MTIME
) {
3008 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3009 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
,
3010 mtime
, sizeof (mtime
));
3013 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3014 if (mask
& ATTR_SIZE
&& !(mask
& ATTR_MTIME
)) {
3015 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
),
3016 NULL
, mtime
, sizeof (mtime
));
3017 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
,
3018 &ctime
, sizeof (ctime
));
3019 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
);
3020 } else if (mask
!= 0) {
3021 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
,
3022 &ctime
, sizeof (ctime
));
3023 zfs_tstamp_update_setup(zp
, STATE_CHANGED
, mtime
, ctime
);
3025 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3026 SA_ZPL_CTIME(zsb
), NULL
,
3027 &ctime
, sizeof (ctime
));
3028 zfs_tstamp_update_setup(attrzp
, STATE_CHANGED
,
3033 * Do this after setting timestamps to prevent timestamp
3034 * update from toggling bit
3037 if (xoap
&& (mask
& ATTR_XVATTR
)) {
3040 * restore trimmed off masks
3041 * so that return masks can be set for caller.
3044 if (XVA_ISSET_REQ(tmpxvattr
, XAT_APPENDONLY
)) {
3045 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3047 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NOUNLINK
)) {
3048 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3050 if (XVA_ISSET_REQ(tmpxvattr
, XAT_IMMUTABLE
)) {
3051 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3053 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NODUMP
)) {
3054 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3056 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_MODIFIED
)) {
3057 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3059 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
)) {
3060 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3063 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3064 ASSERT(S_ISREG(ip
->i_mode
));
3066 zfs_xvattr_set(zp
, xvap
, tx
);
3070 zfs_fuid_sync(zsb
, tx
);
3073 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3075 mutex_exit(&zp
->z_lock
);
3076 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3077 mutex_exit(&zp
->z_acl_lock
);
3080 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3081 mutex_exit(&attrzp
->z_acl_lock
);
3082 mutex_exit(&attrzp
->z_lock
);
3085 if (err
== 0 && attrzp
) {
3086 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3097 zfs_fuid_info_free(fuidp
);
3103 if (err
== ERESTART
)
3106 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3108 zfs_inode_update(zp
);
3112 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3113 zil_commit(zilog
, 0);
3116 kmem_free(xattr_bulk
, sizeof (sa_bulk_attr_t
) * 7);
3117 kmem_free(bulk
, sizeof (sa_bulk_attr_t
) * 7);
3118 kmem_free(tmpxvattr
, sizeof (xvattr_t
));
3122 EXPORT_SYMBOL(zfs_setattr
);
3124 typedef struct zfs_zlock
{
3125 krwlock_t
*zl_rwlock
; /* lock we acquired */
3126 znode_t
*zl_znode
; /* znode we held */
3127 struct zfs_zlock
*zl_next
; /* next in list */
3131 * Drop locks and release vnodes that were held by zfs_rename_lock().
3134 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3138 while ((zl
= *zlpp
) != NULL
) {
3139 if (zl
->zl_znode
!= NULL
)
3140 iput(ZTOI(zl
->zl_znode
));
3141 rw_exit(zl
->zl_rwlock
);
3142 *zlpp
= zl
->zl_next
;
3143 kmem_free(zl
, sizeof (*zl
));
3148 * Search back through the directory tree, using the ".." entries.
3149 * Lock each directory in the chain to prevent concurrent renames.
3150 * Fail any attempt to move a directory into one of its own descendants.
3151 * XXX - z_parent_lock can overlap with map or grow locks
3154 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3158 uint64_t rootid
= ZTOZSB(zp
)->z_root
;
3159 uint64_t oidp
= zp
->z_id
;
3160 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3161 krw_t rw
= RW_WRITER
;
3164 * First pass write-locks szp and compares to zp->z_id.
3165 * Later passes read-lock zp and compare to zp->z_parent.
3168 if (!rw_tryenter(rwlp
, rw
)) {
3170 * Another thread is renaming in this path.
3171 * Note that if we are a WRITER, we don't have any
3172 * parent_locks held yet.
3174 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3176 * Drop our locks and restart
3178 zfs_rename_unlock(&zl
);
3182 rwlp
= &szp
->z_parent_lock
;
3187 * Wait for other thread to drop its locks
3193 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3194 zl
->zl_rwlock
= rwlp
;
3195 zl
->zl_znode
= NULL
;
3196 zl
->zl_next
= *zlpp
;
3199 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3200 return (SET_ERROR(EINVAL
));
3202 if (oidp
== rootid
) /* We've hit the top */
3205 if (rw
== RW_READER
) { /* i.e. not the first pass */
3206 int error
= zfs_zget(ZTOZSB(zp
), oidp
, &zp
);
3211 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(ZTOZSB(zp
)),
3212 &oidp
, sizeof (oidp
));
3213 rwlp
= &zp
->z_parent_lock
;
3216 } while (zp
->z_id
!= sdzp
->z_id
);
3222 * Move an entry from the provided source directory to the target
3223 * directory. Change the entry name as indicated.
3225 * IN: sdip - Source directory containing the "old entry".
3226 * snm - Old entry name.
3227 * tdip - Target directory to contain the "new entry".
3228 * tnm - New entry name.
3229 * cr - credentials of caller.
3230 * flags - case flags
3232 * RETURN: 0 on success, error code on failure.
3235 * sdip,tdip - ctime|mtime updated
3239 zfs_rename(struct inode
*sdip
, char *snm
, struct inode
*tdip
, char *tnm
,
3240 cred_t
*cr
, int flags
)
3242 znode_t
*tdzp
, *szp
, *tzp
;
3243 znode_t
*sdzp
= ITOZ(sdip
);
3244 zfs_sb_t
*zsb
= ITOZSB(sdip
);
3246 zfs_dirlock_t
*sdl
, *tdl
;
3249 int cmp
, serr
, terr
;
3252 boolean_t waited
= B_FALSE
;
3255 ZFS_VERIFY_ZP(sdzp
);
3259 ZFS_VERIFY_ZP(tdzp
);
3262 * We check i_sb because snapshots and the ctldir must have different
3265 if (tdip
->i_sb
!= sdip
->i_sb
|| zfsctl_is_node(tdip
)) {
3267 return (SET_ERROR(EXDEV
));
3270 if (zsb
->z_utf8
&& u8_validate(tnm
,
3271 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3273 return (SET_ERROR(EILSEQ
));
3276 if (flags
& FIGNORECASE
)
3285 * This is to prevent the creation of links into attribute space
3286 * by renaming a linked file into/outof an attribute directory.
3287 * See the comment in zfs_link() for why this is considered bad.
3289 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3291 return (SET_ERROR(EINVAL
));
3295 * Lock source and target directory entries. To prevent deadlock,
3296 * a lock ordering must be defined. We lock the directory with
3297 * the smallest object id first, or if it's a tie, the one with
3298 * the lexically first name.
3300 if (sdzp
->z_id
< tdzp
->z_id
) {
3302 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3306 * First compare the two name arguments without
3307 * considering any case folding.
3309 int nofold
= (zsb
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3311 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3312 ASSERT(error
== 0 || !zsb
->z_utf8
);
3315 * POSIX: "If the old argument and the new argument
3316 * both refer to links to the same existing file,
3317 * the rename() function shall return successfully
3318 * and perform no other action."
3324 * If the file system is case-folding, then we may
3325 * have some more checking to do. A case-folding file
3326 * system is either supporting mixed case sensitivity
3327 * access or is completely case-insensitive. Note
3328 * that the file system is always case preserving.
3330 * In mixed sensitivity mode case sensitive behavior
3331 * is the default. FIGNORECASE must be used to
3332 * explicitly request case insensitive behavior.
3334 * If the source and target names provided differ only
3335 * by case (e.g., a request to rename 'tim' to 'Tim'),
3336 * we will treat this as a special case in the
3337 * case-insensitive mode: as long as the source name
3338 * is an exact match, we will allow this to proceed as
3339 * a name-change request.
3341 if ((zsb
->z_case
== ZFS_CASE_INSENSITIVE
||
3342 (zsb
->z_case
== ZFS_CASE_MIXED
&&
3343 flags
& FIGNORECASE
)) &&
3344 u8_strcmp(snm
, tnm
, 0, zsb
->z_norm
, U8_UNICODE_LATEST
,
3347 * case preserving rename request, require exact
3356 * If the source and destination directories are the same, we should
3357 * grab the z_name_lock of that directory only once.
3361 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3365 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3366 ZEXISTS
| zflg
, NULL
, NULL
);
3367 terr
= zfs_dirent_lock(&tdl
,
3368 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3370 terr
= zfs_dirent_lock(&tdl
,
3371 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3372 serr
= zfs_dirent_lock(&sdl
,
3373 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3379 * Source entry invalid or not there.
3382 zfs_dirent_unlock(tdl
);
3388 rw_exit(&sdzp
->z_name_lock
);
3390 if (strcmp(snm
, "..") == 0)
3396 zfs_dirent_unlock(sdl
);
3400 rw_exit(&sdzp
->z_name_lock
);
3402 if (strcmp(tnm
, "..") == 0)
3409 * Must have write access at the source to remove the old entry
3410 * and write access at the target to create the new entry.
3411 * Note that if target and source are the same, this can be
3412 * done in a single check.
3415 if ((error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
)))
3418 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3420 * Check to make sure rename is valid.
3421 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3423 if ((error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
)))
3428 * Does target exist?
3432 * Source and target must be the same type.
3434 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3435 if (!S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3436 error
= SET_ERROR(ENOTDIR
);
3440 if (S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3441 error
= SET_ERROR(EISDIR
);
3446 * POSIX dictates that when the source and target
3447 * entries refer to the same file object, rename
3448 * must do nothing and exit without error.
3450 if (szp
->z_id
== tzp
->z_id
) {
3456 tx
= dmu_tx_create(zsb
->z_os
);
3457 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3458 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3459 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3460 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3462 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3463 zfs_sa_upgrade_txholds(tx
, tdzp
);
3466 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3467 zfs_sa_upgrade_txholds(tx
, tzp
);
3470 zfs_sa_upgrade_txholds(tx
, szp
);
3471 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
3472 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3475 zfs_rename_unlock(&zl
);
3476 zfs_dirent_unlock(sdl
);
3477 zfs_dirent_unlock(tdl
);
3480 rw_exit(&sdzp
->z_name_lock
);
3485 if (error
== ERESTART
) {
3496 if (tzp
) /* Attempt to remove the existing target */
3497 error
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3500 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3502 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3504 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zsb
),
3505 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3508 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3510 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3511 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3512 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3515 * At this point, we have successfully created
3516 * the target name, but have failed to remove
3517 * the source name. Since the create was done
3518 * with the ZRENAMING flag, there are
3519 * complications; for one, the link count is
3520 * wrong. The easiest way to deal with this
3521 * is to remove the newly created target, and
3522 * return the original error. This must
3523 * succeed; fortunately, it is very unlikely to
3524 * fail, since we just created it.
3526 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3527 ZRENAMING
, NULL
), ==, 0);
3535 zfs_rename_unlock(&zl
);
3537 zfs_dirent_unlock(sdl
);
3538 zfs_dirent_unlock(tdl
);
3540 zfs_inode_update(sdzp
);
3542 rw_exit(&sdzp
->z_name_lock
);
3545 zfs_inode_update(tdzp
);
3547 zfs_inode_update(szp
);
3550 zfs_inode_update(tzp
);
3554 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3555 zil_commit(zilog
, 0);
3560 EXPORT_SYMBOL(zfs_rename
);
3563 * Insert the indicated symbolic reference entry into the directory.
3565 * IN: dip - Directory to contain new symbolic link.
3566 * link - Name for new symlink entry.
3567 * vap - Attributes of new entry.
3568 * target - Target path of new symlink.
3570 * cr - credentials of caller.
3571 * flags - case flags
3573 * RETURN: 0 on success, error code on failure.
3576 * dip - ctime|mtime updated
3580 zfs_symlink(struct inode
*dip
, char *name
, vattr_t
*vap
, char *link
,
3581 struct inode
**ipp
, cred_t
*cr
, int flags
)
3583 znode_t
*zp
, *dzp
= ITOZ(dip
);
3586 zfs_sb_t
*zsb
= ITOZSB(dip
);
3588 uint64_t len
= strlen(link
);
3591 zfs_acl_ids_t acl_ids
;
3592 boolean_t fuid_dirtied
;
3593 uint64_t txtype
= TX_SYMLINK
;
3594 boolean_t waited
= B_FALSE
;
3596 ASSERT(S_ISLNK(vap
->va_mode
));
3602 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
3603 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3605 return (SET_ERROR(EILSEQ
));
3607 if (flags
& FIGNORECASE
)
3610 if (len
> MAXPATHLEN
) {
3612 return (SET_ERROR(ENAMETOOLONG
));
3615 if ((error
= zfs_acl_ids_create(dzp
, 0,
3616 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3624 * Attempt to lock directory; fail if entry already exists.
3626 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3628 zfs_acl_ids_free(&acl_ids
);
3633 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3634 zfs_acl_ids_free(&acl_ids
);
3635 zfs_dirent_unlock(dl
);
3640 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
3641 zfs_acl_ids_free(&acl_ids
);
3642 zfs_dirent_unlock(dl
);
3644 return (SET_ERROR(EDQUOT
));
3646 tx
= dmu_tx_create(zsb
->z_os
);
3647 fuid_dirtied
= zsb
->z_fuid_dirty
;
3648 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3649 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3650 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3651 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3652 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3653 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3654 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3655 acl_ids
.z_aclp
->z_acl_bytes
);
3658 zfs_fuid_txhold(zsb
, tx
);
3659 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3661 zfs_dirent_unlock(dl
);
3662 if (error
== ERESTART
) {
3668 zfs_acl_ids_free(&acl_ids
);
3675 * Create a new object for the symlink.
3676 * for version 4 ZPL datsets the symlink will be an SA attribute
3678 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3681 zfs_fuid_sync(zsb
, tx
);
3683 mutex_enter(&zp
->z_lock
);
3685 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zsb
),
3688 zfs_sa_symlink(zp
, link
, len
, tx
);
3689 mutex_exit(&zp
->z_lock
);
3692 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
3693 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3695 * Insert the new object into the directory.
3697 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3699 if (flags
& FIGNORECASE
)
3701 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3703 zfs_inode_update(dzp
);
3704 zfs_inode_update(zp
);
3706 zfs_acl_ids_free(&acl_ids
);
3710 zfs_dirent_unlock(dl
);
3714 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3715 zil_commit(zilog
, 0);
3720 EXPORT_SYMBOL(zfs_symlink
);
3723 * Return, in the buffer contained in the provided uio structure,
3724 * the symbolic path referred to by ip.
3726 * IN: ip - inode of symbolic link
3727 * uio - structure to contain the link path.
3728 * cr - credentials of caller.
3730 * RETURN: 0 if success
3731 * error code if failure
3734 * ip - atime updated
3738 zfs_readlink(struct inode
*ip
, uio_t
*uio
, cred_t
*cr
)
3740 znode_t
*zp
= ITOZ(ip
);
3741 zfs_sb_t
*zsb
= ITOZSB(ip
);
3747 mutex_enter(&zp
->z_lock
);
3749 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3750 SA_ZPL_SYMLINK(zsb
), uio
);
3752 error
= zfs_sa_readlink(zp
, uio
);
3753 mutex_exit(&zp
->z_lock
);
3758 EXPORT_SYMBOL(zfs_readlink
);
3761 * Insert a new entry into directory tdip referencing sip.
3763 * IN: tdip - Directory to contain new entry.
3764 * sip - inode of new entry.
3765 * name - name of new entry.
3766 * cr - credentials of caller.
3768 * RETURN: 0 if success
3769 * error code if failure
3772 * tdip - ctime|mtime updated
3773 * sip - ctime updated
3777 zfs_link(struct inode
*tdip
, struct inode
*sip
, char *name
, cred_t
*cr
,
3780 znode_t
*dzp
= ITOZ(tdip
);
3782 zfs_sb_t
*zsb
= ITOZSB(tdip
);
3790 boolean_t waited
= B_FALSE
;
3792 ASSERT(S_ISDIR(tdip
->i_mode
));
3799 * POSIX dictates that we return EPERM here.
3800 * Better choices include ENOTSUP or EISDIR.
3802 if (S_ISDIR(sip
->i_mode
)) {
3804 return (SET_ERROR(EPERM
));
3811 * We check i_sb because snapshots and the ctldir must have different
3814 if (sip
->i_sb
!= tdip
->i_sb
|| zfsctl_is_node(sip
)) {
3816 return (SET_ERROR(EXDEV
));
3819 /* Prevent links to .zfs/shares files */
3821 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
3822 &parent
, sizeof (uint64_t))) != 0) {
3826 if (parent
== zsb
->z_shares_dir
) {
3828 return (SET_ERROR(EPERM
));
3831 if (zsb
->z_utf8
&& u8_validate(name
,
3832 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3834 return (SET_ERROR(EILSEQ
));
3836 if (flags
& FIGNORECASE
)
3840 * We do not support links between attributes and non-attributes
3841 * because of the potential security risk of creating links
3842 * into "normal" file space in order to circumvent restrictions
3843 * imposed in attribute space.
3845 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
3847 return (SET_ERROR(EINVAL
));
3850 owner
= zfs_fuid_map_id(zsb
, szp
->z_uid
, cr
, ZFS_OWNER
);
3851 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
3853 return (SET_ERROR(EPERM
));
3856 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3863 * Attempt to lock directory; fail if entry already exists.
3865 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
3871 tx
= dmu_tx_create(zsb
->z_os
);
3872 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3873 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3874 zfs_sa_upgrade_txholds(tx
, szp
);
3875 zfs_sa_upgrade_txholds(tx
, dzp
);
3876 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3878 zfs_dirent_unlock(dl
);
3879 if (error
== ERESTART
) {
3890 error
= zfs_link_create(dl
, szp
, tx
, 0);
3893 uint64_t txtype
= TX_LINK
;
3894 if (flags
& FIGNORECASE
)
3896 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
3901 zfs_dirent_unlock(dl
);
3903 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3904 zil_commit(zilog
, 0);
3906 zfs_inode_update(dzp
);
3907 zfs_inode_update(szp
);
3911 EXPORT_SYMBOL(zfs_link
);
3914 zfs_putpage_commit_cb(void *arg
)
3916 struct page
*pp
= arg
;
3919 end_page_writeback(pp
);
3923 * Push a page out to disk, once the page is on stable storage the
3924 * registered commit callback will be run as notification of completion.
3926 * IN: ip - page mapped for inode.
3927 * pp - page to push (page is locked)
3928 * wbc - writeback control data
3930 * RETURN: 0 if success
3931 * error code if failure
3934 * ip - ctime|mtime updated
3938 zfs_putpage(struct inode
*ip
, struct page
*pp
, struct writeback_control
*wbc
)
3940 znode_t
*zp
= ITOZ(ip
);
3941 zfs_sb_t
*zsb
= ITOZSB(ip
);
3949 uint64_t mtime
[2], ctime
[2];
3950 sa_bulk_attr_t bulk
[3];
3952 struct address_space
*mapping
;
3957 ASSERT(PageLocked(pp
));
3959 pgoff
= page_offset(pp
); /* Page byte-offset in file */
3960 offset
= i_size_read(ip
); /* File length in bytes */
3961 pglen
= MIN(PAGE_SIZE
, /* Page length in bytes */
3962 P2ROUNDUP(offset
, PAGE_SIZE
)-pgoff
);
3964 /* Page is beyond end of file */
3965 if (pgoff
>= offset
) {
3971 /* Truncate page length to end of file */
3972 if (pgoff
+ pglen
> offset
)
3973 pglen
= offset
- pgoff
;
3977 * FIXME: Allow mmap writes past its quota. The correct fix
3978 * is to register a page_mkwrite() handler to count the page
3979 * against its quota when it is about to be dirtied.
3981 if (zfs_owner_overquota(zsb
, zp
, B_FALSE
) ||
3982 zfs_owner_overquota(zsb
, zp
, B_TRUE
)) {
3988 * The ordering here is critical and must adhere to the following
3989 * rules in order to avoid deadlocking in either zfs_read() or
3990 * zfs_free_range() due to a lock inversion.
3992 * 1) The page must be unlocked prior to acquiring the range lock.
3993 * This is critical because zfs_read() calls find_lock_page()
3994 * which may block on the page lock while holding the range lock.
3996 * 2) Before setting or clearing write back on a page the range lock
3997 * must be held in order to prevent a lock inversion with the
3998 * zfs_free_range() function.
4000 * This presents a problem because upon entering this function the
4001 * page lock is already held. To safely acquire the range lock the
4002 * page lock must be dropped. This creates a window where another
4003 * process could truncate, invalidate, dirty, or write out the page.
4005 * Therefore, after successfully reacquiring the range and page locks
4006 * the current page state is checked. In the common case everything
4007 * will be as is expected and it can be written out. However, if
4008 * the page state has changed it must be handled accordingly.
4010 mapping
= pp
->mapping
;
4011 redirty_page_for_writepage(wbc
, pp
);
4014 rl
= zfs_range_lock(&zp
->z_range_lock
, pgoff
, pglen
, RL_WRITER
);
4017 /* Page mapping changed or it was no longer dirty, we're done */
4018 if (unlikely((mapping
!= pp
->mapping
) || !PageDirty(pp
))) {
4020 zfs_range_unlock(rl
);
4025 /* Another process started write block if required */
4026 if (PageWriteback(pp
)) {
4028 zfs_range_unlock(rl
);
4030 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
4031 wait_on_page_writeback(pp
);
4037 /* Clear the dirty flag the required locks are held */
4038 if (!clear_page_dirty_for_io(pp
)) {
4040 zfs_range_unlock(rl
);
4046 * Counterpart for redirty_page_for_writepage() above. This page
4047 * was in fact not skipped and should not be counted as if it were.
4049 wbc
->pages_skipped
--;
4050 set_page_writeback(pp
);
4053 tx
= dmu_tx_create(zsb
->z_os
);
4054 dmu_tx_hold_write(tx
, zp
->z_id
, pgoff
, pglen
);
4055 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4056 zfs_sa_upgrade_txholds(tx
, zp
);
4058 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4060 if (err
== ERESTART
)
4064 __set_page_dirty_nobuffers(pp
);
4066 end_page_writeback(pp
);
4067 zfs_range_unlock(rl
);
4073 ASSERT3U(pglen
, <=, PAGE_SIZE
);
4074 dmu_write(zsb
->z_os
, zp
->z_id
, pgoff
, pglen
, va
, tx
);
4077 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
4078 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
4079 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_FLAGS(zsb
), NULL
, &zp
->z_pflags
, 8);
4081 /* Preserve the mtime and ctime provided by the inode */
4082 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4083 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4084 zp
->z_atime_dirty
= 0;
4087 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4089 zfs_log_write(zsb
->z_log
, tx
, TX_WRITE
, zp
, pgoff
, pglen
, 0,
4090 zfs_putpage_commit_cb
, pp
);
4093 zfs_range_unlock(rl
);
4095 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4097 * Note that this is rarely called under writepages(), because
4098 * writepages() normally handles the entire commit for
4099 * performance reasons.
4101 if (zsb
->z_log
!= NULL
)
4102 zil_commit(zsb
->z_log
, zp
->z_id
);
4110 * Update the system attributes when the inode has been dirtied. For the
4111 * moment we only update the mode, atime, mtime, and ctime.
4114 zfs_dirty_inode(struct inode
*ip
, int flags
)
4116 znode_t
*zp
= ITOZ(ip
);
4117 zfs_sb_t
*zsb
= ITOZSB(ip
);
4119 uint64_t mode
, atime
[2], mtime
[2], ctime
[2];
4120 sa_bulk_attr_t bulk
[4];
4124 if (zfs_is_readonly(zsb
) || dmu_objset_is_snapshot(zsb
->z_os
))
4132 * This is the lazytime semantic indroduced in Linux 4.0
4133 * This flag will only be called from update_time when lazytime is set.
4134 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4135 * Fortunately mtime and ctime are managed within ZFS itself, so we
4136 * only need to dirty atime.
4138 if (flags
== I_DIRTY_TIME
) {
4139 zp
->z_atime_dirty
= 1;
4144 tx
= dmu_tx_create(zsb
->z_os
);
4146 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4147 zfs_sa_upgrade_txholds(tx
, zp
);
4149 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4155 mutex_enter(&zp
->z_lock
);
4156 zp
->z_atime_dirty
= 0;
4158 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MODE(zsb
), NULL
, &mode
, 8);
4159 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_ATIME(zsb
), NULL
, &atime
, 16);
4160 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
4161 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
4163 /* Preserve the mode, mtime and ctime provided by the inode */
4164 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4165 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4166 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4171 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4172 mutex_exit(&zp
->z_lock
);
4179 EXPORT_SYMBOL(zfs_dirty_inode
);
4183 zfs_inactive(struct inode
*ip
)
4185 znode_t
*zp
= ITOZ(ip
);
4186 zfs_sb_t
*zsb
= ITOZSB(ip
);
4189 int need_unlock
= 0;
4191 /* Only read lock if we haven't already write locked, e.g. rollback */
4192 if (!RW_WRITE_HELD(&zsb
->z_teardown_inactive_lock
)) {
4194 rw_enter(&zsb
->z_teardown_inactive_lock
, RW_READER
);
4196 if (zp
->z_sa_hdl
== NULL
) {
4198 rw_exit(&zsb
->z_teardown_inactive_lock
);
4202 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4203 dmu_tx_t
*tx
= dmu_tx_create(zsb
->z_os
);
4205 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4206 zfs_sa_upgrade_txholds(tx
, zp
);
4207 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4211 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4212 mutex_enter(&zp
->z_lock
);
4213 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zsb
),
4214 (void *)&atime
, sizeof (atime
), tx
);
4215 zp
->z_atime_dirty
= 0;
4216 mutex_exit(&zp
->z_lock
);
4223 rw_exit(&zsb
->z_teardown_inactive_lock
);
4225 EXPORT_SYMBOL(zfs_inactive
);
4228 * Bounds-check the seek operation.
4230 * IN: ip - inode seeking within
4231 * ooff - old file offset
4232 * noffp - pointer to new file offset
4233 * ct - caller context
4235 * RETURN: 0 if success
4236 * EINVAL if new offset invalid
4240 zfs_seek(struct inode
*ip
, offset_t ooff
, offset_t
*noffp
)
4242 if (S_ISDIR(ip
->i_mode
))
4244 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4246 EXPORT_SYMBOL(zfs_seek
);
4249 * Fill pages with data from the disk.
4252 zfs_fillpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4254 znode_t
*zp
= ITOZ(ip
);
4255 zfs_sb_t
*zsb
= ITOZSB(ip
);
4257 struct page
*cur_pp
;
4258 u_offset_t io_off
, total
;
4265 io_len
= nr_pages
<< PAGE_SHIFT
;
4266 i_size
= i_size_read(ip
);
4267 io_off
= page_offset(pl
[0]);
4269 if (io_off
+ io_len
> i_size
)
4270 io_len
= i_size
- io_off
;
4273 * Iterate over list of pages and read each page individually.
4276 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4279 cur_pp
= pl
[page_idx
++];
4281 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4285 /* convert checksum errors into IO errors */
4287 err
= SET_ERROR(EIO
);
4296 * Uses zfs_fillpage to read data from the file and fill the pages.
4298 * IN: ip - inode of file to get data from.
4299 * pl - list of pages to read
4300 * nr_pages - number of pages to read
4302 * RETURN: 0 on success, error code on failure.
4305 * vp - atime updated
4309 zfs_getpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4311 znode_t
*zp
= ITOZ(ip
);
4312 zfs_sb_t
*zsb
= ITOZSB(ip
);
4321 err
= zfs_fillpage(ip
, pl
, nr_pages
);
4326 EXPORT_SYMBOL(zfs_getpage
);
4329 * Check ZFS specific permissions to memory map a section of a file.
4331 * IN: ip - inode of the file to mmap
4333 * addrp - start address in memory region
4334 * len - length of memory region
4335 * vm_flags- address flags
4337 * RETURN: 0 if success
4338 * error code if failure
4342 zfs_map(struct inode
*ip
, offset_t off
, caddr_t
*addrp
, size_t len
,
4343 unsigned long vm_flags
)
4345 znode_t
*zp
= ITOZ(ip
);
4346 zfs_sb_t
*zsb
= ITOZSB(ip
);
4351 if ((vm_flags
& VM_WRITE
) && (zp
->z_pflags
&
4352 (ZFS_IMMUTABLE
| ZFS_READONLY
| ZFS_APPENDONLY
))) {
4354 return (SET_ERROR(EPERM
));
4357 if ((vm_flags
& (VM_READ
| VM_EXEC
)) &&
4358 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4360 return (SET_ERROR(EACCES
));
4363 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4365 return (SET_ERROR(ENXIO
));
4371 EXPORT_SYMBOL(zfs_map
);
4374 * convoff - converts the given data (start, whence) to the
4378 convoff(struct inode
*ip
, flock64_t
*lckdat
, int whence
, offset_t offset
)
4383 if ((lckdat
->l_whence
== 2) || (whence
== 2)) {
4384 if ((error
= zfs_getattr(ip
, &vap
, 0, CRED()) != 0))
4388 switch (lckdat
->l_whence
) {
4390 lckdat
->l_start
+= offset
;
4393 lckdat
->l_start
+= vap
.va_size
;
4398 return (SET_ERROR(EINVAL
));
4401 if (lckdat
->l_start
< 0)
4402 return (SET_ERROR(EINVAL
));
4406 lckdat
->l_start
-= offset
;
4409 lckdat
->l_start
-= vap
.va_size
;
4414 return (SET_ERROR(EINVAL
));
4417 lckdat
->l_whence
= (short)whence
;
4422 * Free or allocate space in a file. Currently, this function only
4423 * supports the `F_FREESP' command. However, this command is somewhat
4424 * misnamed, as its functionality includes the ability to allocate as
4425 * well as free space.
4427 * IN: ip - inode of file to free data in.
4428 * cmd - action to take (only F_FREESP supported).
4429 * bfp - section of file to free/alloc.
4430 * flag - current file open mode flags.
4431 * offset - current file offset.
4432 * cr - credentials of caller [UNUSED].
4434 * RETURN: 0 on success, error code on failure.
4437 * ip - ctime|mtime updated
4441 zfs_space(struct inode
*ip
, int cmd
, flock64_t
*bfp
, int flag
,
4442 offset_t offset
, cred_t
*cr
)
4444 znode_t
*zp
= ITOZ(ip
);
4445 zfs_sb_t
*zsb
= ITOZSB(ip
);
4452 if (cmd
!= F_FREESP
) {
4454 return (SET_ERROR(EINVAL
));
4458 * Callers might not be able to detect properly that we are read-only,
4459 * so check it explicitly here.
4461 if (zfs_is_readonly(zsb
)) {
4463 return (SET_ERROR(EROFS
));
4466 if ((error
= convoff(ip
, bfp
, 0, offset
))) {
4471 if (bfp
->l_len
< 0) {
4473 return (SET_ERROR(EINVAL
));
4477 * Permissions aren't checked on Solaris because on this OS
4478 * zfs_space() can only be called with an opened file handle.
4479 * On Linux we can get here through truncate_range() which
4480 * operates directly on inodes, so we need to check access rights.
4482 if ((error
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
))) {
4488 len
= bfp
->l_len
; /* 0 means from off to end of file */
4490 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4495 EXPORT_SYMBOL(zfs_space
);
4499 zfs_fid(struct inode
*ip
, fid_t
*fidp
)
4501 znode_t
*zp
= ITOZ(ip
);
4502 zfs_sb_t
*zsb
= ITOZSB(ip
);
4505 uint64_t object
= zp
->z_id
;
4512 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zsb
),
4513 &gen64
, sizeof (uint64_t))) != 0) {
4518 gen
= (uint32_t)gen64
;
4520 size
= (zsb
->z_parent
!= zsb
) ? LONG_FID_LEN
: SHORT_FID_LEN
;
4521 if (fidp
->fid_len
< size
) {
4522 fidp
->fid_len
= size
;
4524 return (SET_ERROR(ENOSPC
));
4527 zfid
= (zfid_short_t
*)fidp
;
4529 zfid
->zf_len
= size
;
4531 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4532 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4534 /* Must have a non-zero generation number to distinguish from .zfs */
4537 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4538 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4540 if (size
== LONG_FID_LEN
) {
4541 uint64_t objsetid
= dmu_objset_id(zsb
->z_os
);
4544 zlfid
= (zfid_long_t
*)fidp
;
4546 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
4547 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
4549 /* XXX - this should be the generation number for the objset */
4550 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
4551 zlfid
->zf_setgen
[i
] = 0;
4557 EXPORT_SYMBOL(zfs_fid
);
4561 zfs_getsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4563 znode_t
*zp
= ITOZ(ip
);
4564 zfs_sb_t
*zsb
= ITOZSB(ip
);
4566 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4570 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
4575 EXPORT_SYMBOL(zfs_getsecattr
);
4579 zfs_setsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4581 znode_t
*zp
= ITOZ(ip
);
4582 zfs_sb_t
*zsb
= ITOZSB(ip
);
4584 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4585 zilog_t
*zilog
= zsb
->z_log
;
4590 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
4592 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4593 zil_commit(zilog
, 0);
4598 EXPORT_SYMBOL(zfs_setsecattr
);
4600 #ifdef HAVE_UIO_ZEROCOPY
4602 * Tunable, both must be a power of 2.
4604 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4605 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4606 * an arcbuf for a partial block read
4608 int zcr_blksz_min
= (1 << 10); /* 1K */
4609 int zcr_blksz_max
= (1 << 17); /* 128K */
4613 zfs_reqzcbuf(struct inode
*ip
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
)
4615 znode_t
*zp
= ITOZ(ip
);
4616 zfs_sb_t
*zsb
= ITOZSB(ip
);
4617 int max_blksz
= zsb
->z_max_blksz
;
4618 uio_t
*uio
= &xuio
->xu_uio
;
4619 ssize_t size
= uio
->uio_resid
;
4620 offset_t offset
= uio
->uio_loffset
;
4625 int preamble
, postamble
;
4627 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
4628 return (SET_ERROR(EINVAL
));
4635 * Loan out an arc_buf for write if write size is bigger than
4636 * max_blksz, and the file's block size is also max_blksz.
4639 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
4641 return (SET_ERROR(EINVAL
));
4644 * Caller requests buffers for write before knowing where the
4645 * write offset might be (e.g. NFS TCP write).
4650 preamble
= P2PHASE(offset
, blksz
);
4652 preamble
= blksz
- preamble
;
4657 postamble
= P2PHASE(size
, blksz
);
4660 fullblk
= size
/ blksz
;
4661 (void) dmu_xuio_init(xuio
,
4662 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
4665 * Have to fix iov base/len for partial buffers. They
4666 * currently represent full arc_buf's.
4669 /* data begins in the middle of the arc_buf */
4670 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4673 (void) dmu_xuio_add(xuio
, abuf
,
4674 blksz
- preamble
, preamble
);
4677 for (i
= 0; i
< fullblk
; i
++) {
4678 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4681 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
4685 /* data ends in the middle of the arc_buf */
4686 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4689 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
4694 * Loan out an arc_buf for read if the read size is larger than
4695 * the current file block size. Block alignment is not
4696 * considered. Partial arc_buf will be loaned out for read.
4698 blksz
= zp
->z_blksz
;
4699 if (blksz
< zcr_blksz_min
)
4700 blksz
= zcr_blksz_min
;
4701 if (blksz
> zcr_blksz_max
)
4702 blksz
= zcr_blksz_max
;
4703 /* avoid potential complexity of dealing with it */
4704 if (blksz
> max_blksz
) {
4706 return (SET_ERROR(EINVAL
));
4709 maxsize
= zp
->z_size
- uio
->uio_loffset
;
4715 return (SET_ERROR(EINVAL
));
4720 return (SET_ERROR(EINVAL
));
4723 uio
->uio_extflg
= UIO_XUIO
;
4724 XUIO_XUZC_RW(xuio
) = ioflag
;
4731 zfs_retzcbuf(struct inode
*ip
, xuio_t
*xuio
, cred_t
*cr
)
4735 int ioflag
= XUIO_XUZC_RW(xuio
);
4737 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
4739 i
= dmu_xuio_cnt(xuio
);
4741 abuf
= dmu_xuio_arcbuf(xuio
, i
);
4743 * if abuf == NULL, it must be a write buffer
4744 * that has been returned in zfs_write().
4747 dmu_return_arcbuf(abuf
);
4748 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
4751 dmu_xuio_fini(xuio
);
4754 #endif /* HAVE_UIO_ZEROCOPY */
4756 #if defined(_KERNEL) && defined(HAVE_SPL)
4757 module_param(zfs_delete_blocks
, ulong
, 0644);
4758 MODULE_PARM_DESC(zfs_delete_blocks
, "Delete files larger than N blocks async");
4759 module_param(zfs_read_chunk_size
, long, 0644);
4760 MODULE_PARM_DESC(zfs_read_chunk_size
, "Bytes to read per chunk");