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];
606 ASSERTV(int iovcnt
= uio
->uio_iovcnt
);
609 * Fasttrack empty write
615 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
621 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
622 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
623 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zsb
), NULL
, &zp
->z_size
, 8);
624 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
628 * Callers might not be able to detect properly that we are read-only,
629 * so check it explicitly here.
631 if (zfs_is_readonly(zsb
)) {
633 return (SET_ERROR(EROFS
));
637 * If immutable or not appending then return EPERM
639 if ((zp
->z_pflags
& (ZFS_IMMUTABLE
| ZFS_READONLY
)) ||
640 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
641 (uio
->uio_loffset
< zp
->z_size
))) {
643 return (SET_ERROR(EPERM
));
649 * Validate file offset
651 woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
654 return (SET_ERROR(EINVAL
));
658 * Pre-fault the pages to ensure slow (eg NFS) pages
660 * Skip this if uio contains loaned arc_buf.
662 #ifdef HAVE_UIO_ZEROCOPY
663 if ((uio
->uio_extflg
== UIO_XUIO
) &&
664 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
))
665 xuio
= (xuio_t
*)uio
;
668 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
671 * If in append mode, set the io offset pointer to eof.
673 if (ioflag
& FAPPEND
) {
675 * Obtain an appending range lock to guarantee file append
676 * semantics. We reset the write offset once we have the lock.
678 rl
= zfs_range_lock(&zp
->z_range_lock
, 0, n
, RL_APPEND
);
680 if (rl
->r_len
== UINT64_MAX
) {
682 * We overlocked the file because this write will cause
683 * the file block size to increase.
684 * Note that zp_size cannot change with this lock held.
688 uio
->uio_loffset
= woff
;
691 * Note that if the file block size will change as a result of
692 * this write, then this range lock will lock the entire file
693 * so that we can re-write the block safely.
695 rl
= zfs_range_lock(&zp
->z_range_lock
, woff
, n
, RL_WRITER
);
699 zfs_range_unlock(rl
);
701 return (SET_ERROR(EFBIG
));
704 if ((woff
+ n
) > limit
|| woff
> (limit
- n
))
707 /* Will this write extend the file length? */
708 write_eof
= (woff
+ n
> zp
->z_size
);
710 end_size
= MAX(zp
->z_size
, woff
+ n
);
713 * Write the file in reasonable size chunks. Each chunk is written
714 * in a separate transaction; this keeps the intent log records small
715 * and allows us to do more fine-grained space accounting.
719 woff
= uio
->uio_loffset
;
720 if (zfs_owner_overquota(zsb
, zp
, B_FALSE
) ||
721 zfs_owner_overquota(zsb
, zp
, B_TRUE
)) {
723 dmu_return_arcbuf(abuf
);
724 error
= SET_ERROR(EDQUOT
);
728 if (xuio
&& abuf
== NULL
) {
729 ASSERT(i_iov
< iovcnt
);
730 ASSERT3U(uio
->uio_segflg
, !=, UIO_BVEC
);
732 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
733 dmu_xuio_clear(xuio
, i_iov
);
734 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
735 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
736 aiov
->iov_len
== arc_buf_size(abuf
)));
738 } else if (abuf
== NULL
&& n
>= max_blksz
&&
739 woff
>= zp
->z_size
&&
740 P2PHASE(woff
, max_blksz
) == 0 &&
741 zp
->z_blksz
== max_blksz
) {
743 * This write covers a full block. "Borrow" a buffer
744 * from the dmu so that we can fill it before we enter
745 * a transaction. This avoids the possibility of
746 * holding up the transaction if the data copy hangs
747 * up on a pagefault (e.g., from an NFS server mapping).
751 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
753 ASSERT(abuf
!= NULL
);
754 ASSERT(arc_buf_size(abuf
) == max_blksz
);
755 if ((error
= uiocopy(abuf
->b_data
, max_blksz
,
756 UIO_WRITE
, uio
, &cbytes
))) {
757 dmu_return_arcbuf(abuf
);
760 ASSERT(cbytes
== max_blksz
);
764 * Start a transaction.
766 tx
= dmu_tx_create(zsb
->z_os
);
767 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
768 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
769 zfs_sa_upgrade_txholds(tx
, zp
);
770 error
= dmu_tx_assign(tx
, TXG_WAIT
);
774 dmu_return_arcbuf(abuf
);
779 * If zfs_range_lock() over-locked we grow the blocksize
780 * and then reduce the lock range. This will only happen
781 * on the first iteration since zfs_range_reduce() will
782 * shrink down r_len to the appropriate size.
784 if (rl
->r_len
== UINT64_MAX
) {
787 if (zp
->z_blksz
> max_blksz
) {
789 * File's blocksize is already larger than the
790 * "recordsize" property. Only let it grow to
791 * the next power of 2.
793 ASSERT(!ISP2(zp
->z_blksz
));
794 new_blksz
= MIN(end_size
,
795 1 << highbit64(zp
->z_blksz
));
797 new_blksz
= MIN(end_size
, max_blksz
);
799 zfs_grow_blocksize(zp
, new_blksz
, tx
);
800 zfs_range_reduce(rl
, woff
, n
);
804 * XXX - should we really limit each write to z_max_blksz?
805 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
807 nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
810 tx_bytes
= uio
->uio_resid
;
811 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
813 tx_bytes
-= uio
->uio_resid
;
816 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
818 * If this is not a full block write, but we are
819 * extending the file past EOF and this data starts
820 * block-aligned, use assign_arcbuf(). Otherwise,
821 * write via dmu_write().
823 if (tx_bytes
< max_blksz
&& (!write_eof
||
824 aiov
->iov_base
!= abuf
->b_data
)) {
826 dmu_write(zsb
->z_os
, zp
->z_id
, woff
,
827 aiov
->iov_len
, aiov
->iov_base
, tx
);
828 dmu_return_arcbuf(abuf
);
829 xuio_stat_wbuf_copied();
831 ASSERT(xuio
|| tx_bytes
== max_blksz
);
832 dmu_assign_arcbuf(sa_get_db(zp
->z_sa_hdl
),
835 ASSERT(tx_bytes
<= uio
->uio_resid
);
836 uioskip(uio
, tx_bytes
);
839 if (tx_bytes
&& zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
))
840 update_pages(ip
, woff
, tx_bytes
, zsb
->z_os
, zp
->z_id
);
843 * If we made no progress, we're done. If we made even
844 * partial progress, update the znode and ZIL accordingly.
847 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
848 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
855 * Clear Set-UID/Set-GID bits on successful write if not
856 * privileged and at least one of the excute bits is set.
858 * It would be nice to to this after all writes have
859 * been done, but that would still expose the ISUID/ISGID
860 * to another app after the partial write is committed.
862 * Note: we don't call zfs_fuid_map_id() here because
863 * user 0 is not an ephemeral uid.
865 mutex_enter(&zp
->z_acl_lock
);
866 uid
= KUID_TO_SUID(ip
->i_uid
);
867 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
868 (S_IXUSR
>> 6))) != 0 &&
869 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
870 secpolicy_vnode_setid_retain(cr
,
871 ((zp
->z_mode
& S_ISUID
) != 0 && uid
== 0)) != 0) {
873 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
874 ip
->i_mode
= newmode
= zp
->z_mode
;
875 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zsb
),
876 (void *)&newmode
, sizeof (uint64_t), tx
);
878 mutex_exit(&zp
->z_acl_lock
);
880 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
);
883 * Update the file size (zp_size) if it has changed;
884 * account for possible concurrent updates.
886 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
887 (void) atomic_cas_64(&zp
->z_size
, end_size
,
892 * If we are replaying and eof is non zero then force
893 * the file size to the specified eof. Note, there's no
894 * concurrency during replay.
896 if (zsb
->z_replay
&& zsb
->z_replay_eof
!= 0)
897 zp
->z_size
= zsb
->z_replay_eof
;
899 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
901 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
,
907 ASSERT(tx_bytes
== nbytes
);
911 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
914 zfs_inode_update(zp
);
915 zfs_range_unlock(rl
);
918 * If we're in replay mode, or we made no progress, return error.
919 * Otherwise, it's at least a partial write, so it's successful.
921 if (zsb
->z_replay
|| uio
->uio_resid
== start_resid
) {
926 if (ioflag
& (FSYNC
| FDSYNC
) ||
927 zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
928 zil_commit(zilog
, zp
->z_id
);
933 EXPORT_SYMBOL(zfs_write
);
936 * Drop a reference on the passed inode asynchronously. This ensures
937 * that the caller will never drop the last reference on an inode in
938 * the current context. Doing so while holding open a tx could result
939 * in a deadlock if iput_final() re-enters the filesystem code.
942 zfs_iput_async(struct inode
*ip
)
944 objset_t
*os
= ITOZSB(ip
)->z_os
;
946 ASSERT(atomic_read(&ip
->i_count
) > 0);
949 if (atomic_read(&ip
->i_count
) == 1)
950 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os
)),
951 (task_func_t
*)iput
, ip
, TQ_SLEEP
) != 0);
957 zfs_get_done(zgd_t
*zgd
, int error
)
959 znode_t
*zp
= zgd
->zgd_private
;
962 dmu_buf_rele(zgd
->zgd_db
, zgd
);
964 zfs_range_unlock(zgd
->zgd_rl
);
967 * Release the vnode asynchronously as we currently have the
968 * txg stopped from syncing.
970 zfs_iput_async(ZTOI(zp
));
972 if (error
== 0 && zgd
->zgd_bp
)
973 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
975 kmem_free(zgd
, sizeof (zgd_t
));
979 static int zil_fault_io
= 0;
983 * Get data to generate a TX_WRITE intent log record.
986 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
989 objset_t
*os
= zsb
->z_os
;
991 uint64_t object
= lr
->lr_foid
;
992 uint64_t offset
= lr
->lr_offset
;
993 uint64_t size
= lr
->lr_length
;
994 blkptr_t
*bp
= &lr
->lr_blkptr
;
1003 * Nothing to do if the file has been removed
1005 if (zfs_zget(zsb
, object
, &zp
) != 0)
1006 return (SET_ERROR(ENOENT
));
1007 if (zp
->z_unlinked
) {
1009 * Release the vnode asynchronously as we currently have the
1010 * txg stopped from syncing.
1012 zfs_iput_async(ZTOI(zp
));
1013 return (SET_ERROR(ENOENT
));
1016 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1017 zgd
->zgd_zilog
= zsb
->z_log
;
1018 zgd
->zgd_private
= zp
;
1021 * Write records come in two flavors: immediate and indirect.
1022 * For small writes it's cheaper to store the data with the
1023 * log record (immediate); for large writes it's cheaper to
1024 * sync the data and get a pointer to it (indirect) so that
1025 * we don't have to write the data twice.
1027 if (buf
!= NULL
) { /* immediate write */
1028 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
, size
,
1030 /* test for truncation needs to be done while range locked */
1031 if (offset
>= zp
->z_size
) {
1032 error
= SET_ERROR(ENOENT
);
1034 error
= dmu_read(os
, object
, offset
, size
, buf
,
1035 DMU_READ_NO_PREFETCH
);
1037 ASSERT(error
== 0 || error
== ENOENT
);
1038 } else { /* indirect write */
1040 * Have to lock the whole block to ensure when it's
1041 * written out and it's checksum is being calculated
1042 * that no one can change the data. We need to re-check
1043 * blocksize after we get the lock in case it's changed!
1048 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1050 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
,
1052 if (zp
->z_blksz
== size
)
1055 zfs_range_unlock(zgd
->zgd_rl
);
1057 /* test for truncation needs to be done while range locked */
1058 if (lr
->lr_offset
>= zp
->z_size
)
1059 error
= SET_ERROR(ENOENT
);
1062 error
= SET_ERROR(EIO
);
1067 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1068 DMU_READ_NO_PREFETCH
);
1071 blkptr_t
*obp
= dmu_buf_get_blkptr(db
);
1073 ASSERT(BP_IS_HOLE(bp
));
1080 ASSERT(db
->db_offset
== offset
);
1081 ASSERT(db
->db_size
== size
);
1083 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1085 ASSERT(error
|| lr
->lr_length
<= zp
->z_blksz
);
1088 * On success, we need to wait for the write I/O
1089 * initiated by dmu_sync() to complete before we can
1090 * release this dbuf. We will finish everything up
1091 * in the zfs_get_done() callback.
1096 if (error
== EALREADY
) {
1097 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
1103 zfs_get_done(zgd
, error
);
1110 zfs_access(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
1112 znode_t
*zp
= ITOZ(ip
);
1113 zfs_sb_t
*zsb
= ITOZSB(ip
);
1119 if (flag
& V_ACE_MASK
)
1120 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1122 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1127 EXPORT_SYMBOL(zfs_access
);
1130 * Lookup an entry in a directory, or an extended attribute directory.
1131 * If it exists, return a held inode reference for it.
1133 * IN: dip - inode of directory to search.
1134 * nm - name of entry to lookup.
1135 * flags - LOOKUP_XATTR set if looking for an attribute.
1136 * cr - credentials of caller.
1137 * direntflags - directory lookup flags
1138 * realpnp - returned pathname.
1140 * OUT: ipp - inode of located entry, NULL if not found.
1142 * RETURN: 0 on success, error code on failure.
1149 zfs_lookup(struct inode
*dip
, char *nm
, struct inode
**ipp
, int flags
,
1150 cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1152 znode_t
*zdp
= ITOZ(dip
);
1153 zfs_sb_t
*zsb
= ITOZSB(dip
);
1157 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1159 if (!S_ISDIR(dip
->i_mode
)) {
1160 return (SET_ERROR(ENOTDIR
));
1161 } else if (zdp
->z_sa_hdl
== NULL
) {
1162 return (SET_ERROR(EIO
));
1165 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1166 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1175 vnode_t
*tvp
= dnlc_lookup(dvp
, nm
);
1178 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1183 if (tvp
== DNLC_NO_VNODE
) {
1185 return (SET_ERROR(ENOENT
));
1188 return (specvp_check(vpp
, cr
));
1191 #endif /* HAVE_DNLC */
1200 if (flags
& LOOKUP_XATTR
) {
1202 * We don't allow recursive attributes..
1203 * Maybe someday we will.
1205 if (zdp
->z_pflags
& ZFS_XATTR
) {
1207 return (SET_ERROR(EINVAL
));
1210 if ((error
= zfs_get_xattrdir(zdp
, ipp
, cr
, flags
))) {
1216 * Do we have permission to get into attribute directory?
1219 if ((error
= zfs_zaccess(ITOZ(*ipp
), ACE_EXECUTE
, 0,
1229 if (!S_ISDIR(dip
->i_mode
)) {
1231 return (SET_ERROR(ENOTDIR
));
1235 * Check accessibility of directory.
1238 if ((error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
))) {
1243 if (zsb
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1244 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1246 return (SET_ERROR(EILSEQ
));
1249 error
= zfs_dirlook(zdp
, nm
, ipp
, flags
, direntflags
, realpnp
);
1250 if ((error
== 0) && (*ipp
))
1251 zfs_inode_update(ITOZ(*ipp
));
1256 EXPORT_SYMBOL(zfs_lookup
);
1259 * Attempt to create a new entry in a directory. If the entry
1260 * already exists, truncate the file if permissible, else return
1261 * an error. Return the ip of the created or trunc'd file.
1263 * IN: dip - inode of directory to put new file entry in.
1264 * name - name of new file entry.
1265 * vap - attributes of new file.
1266 * excl - flag indicating exclusive or non-exclusive mode.
1267 * mode - mode to open file with.
1268 * cr - credentials of caller.
1269 * flag - large file flag [UNUSED].
1270 * vsecp - ACL to be set
1272 * OUT: ipp - inode of created or trunc'd entry.
1274 * RETURN: 0 on success, error code on failure.
1277 * dip - ctime|mtime updated if new entry created
1278 * ip - ctime|mtime always, atime if new
1283 zfs_create(struct inode
*dip
, char *name
, vattr_t
*vap
, int excl
,
1284 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1286 znode_t
*zp
, *dzp
= ITOZ(dip
);
1287 zfs_sb_t
*zsb
= ITOZSB(dip
);
1295 zfs_acl_ids_t acl_ids
;
1296 boolean_t fuid_dirtied
;
1297 boolean_t have_acl
= B_FALSE
;
1298 boolean_t waited
= B_FALSE
;
1301 * If we have an ephemeral id, ACL, or XVATTR then
1302 * make sure file system is at proper version
1308 if (zsb
->z_use_fuids
== B_FALSE
&&
1309 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1310 return (SET_ERROR(EINVAL
));
1317 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
1318 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1320 return (SET_ERROR(EILSEQ
));
1323 if (vap
->va_mask
& ATTR_XVATTR
) {
1324 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1325 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1333 if (*name
== '\0') {
1335 * Null component name refers to the directory itself.
1342 /* possible igrab(zp) */
1345 if (flag
& FIGNORECASE
)
1348 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1352 zfs_acl_ids_free(&acl_ids
);
1353 if (strcmp(name
, "..") == 0)
1354 error
= SET_ERROR(EISDIR
);
1364 * Create a new file object and update the directory
1367 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1369 zfs_acl_ids_free(&acl_ids
);
1374 * We only support the creation of regular files in
1375 * extended attribute directories.
1378 if ((dzp
->z_pflags
& ZFS_XATTR
) && !S_ISREG(vap
->va_mode
)) {
1380 zfs_acl_ids_free(&acl_ids
);
1381 error
= SET_ERROR(EINVAL
);
1385 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1386 cr
, vsecp
, &acl_ids
)) != 0)
1390 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1391 zfs_acl_ids_free(&acl_ids
);
1392 error
= SET_ERROR(EDQUOT
);
1396 tx
= dmu_tx_create(os
);
1398 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1399 ZFS_SA_BASE_ATTR_SIZE
);
1401 fuid_dirtied
= zsb
->z_fuid_dirty
;
1403 zfs_fuid_txhold(zsb
, tx
);
1404 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1405 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1406 if (!zsb
->z_use_sa
&&
1407 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1408 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1409 0, acl_ids
.z_aclp
->z_acl_bytes
);
1411 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1413 zfs_dirent_unlock(dl
);
1414 if (error
== ERESTART
) {
1420 zfs_acl_ids_free(&acl_ids
);
1425 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1428 zfs_fuid_sync(zsb
, tx
);
1430 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1431 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1432 if (flag
& FIGNORECASE
)
1434 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1435 vsecp
, acl_ids
.z_fuidp
, vap
);
1436 zfs_acl_ids_free(&acl_ids
);
1439 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1442 zfs_acl_ids_free(&acl_ids
);
1446 * A directory entry already exists for this name.
1449 * Can't truncate an existing file if in exclusive mode.
1452 error
= SET_ERROR(EEXIST
);
1456 * Can't open a directory for writing.
1458 if (S_ISDIR(ZTOI(zp
)->i_mode
)) {
1459 error
= SET_ERROR(EISDIR
);
1463 * Verify requested access to file.
1465 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1469 mutex_enter(&dzp
->z_lock
);
1471 mutex_exit(&dzp
->z_lock
);
1474 * Truncate regular files if requested.
1476 if (S_ISREG(ZTOI(zp
)->i_mode
) &&
1477 (vap
->va_mask
& ATTR_SIZE
) && (vap
->va_size
== 0)) {
1478 /* we can't hold any locks when calling zfs_freesp() */
1480 zfs_dirent_unlock(dl
);
1483 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1489 zfs_dirent_unlock(dl
);
1495 zfs_inode_update(dzp
);
1496 zfs_inode_update(zp
);
1500 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1501 zil_commit(zilog
, 0);
1506 EXPORT_SYMBOL(zfs_create
);
1509 * Remove an entry from a directory.
1511 * IN: dip - inode of directory to remove entry from.
1512 * name - name of entry to remove.
1513 * cr - credentials of caller.
1515 * RETURN: 0 if success
1516 * error code if failure
1520 * ip - ctime (if nlink > 0)
1523 uint64_t null_xattr
= 0;
1527 zfs_remove(struct inode
*dip
, char *name
, cred_t
*cr
, int flags
)
1529 znode_t
*zp
, *dzp
= ITOZ(dip
);
1532 zfs_sb_t
*zsb
= ITOZSB(dip
);
1534 uint64_t acl_obj
, xattr_obj
;
1535 uint64_t xattr_obj_unlinked
= 0;
1540 boolean_t may_delete_now
, delete_now
= FALSE
;
1541 boolean_t unlinked
, toobig
= FALSE
;
1543 pathname_t
*realnmp
= NULL
;
1547 boolean_t waited
= B_FALSE
;
1553 if (flags
& FIGNORECASE
) {
1563 * Attempt to lock directory; fail if entry doesn't exist.
1565 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1575 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1580 * Need to use rmdir for removing directories.
1582 if (S_ISDIR(ip
->i_mode
)) {
1583 error
= SET_ERROR(EPERM
);
1589 dnlc_remove(dvp
, realnmp
->pn_buf
);
1591 dnlc_remove(dvp
, name
);
1592 #endif /* HAVE_DNLC */
1594 mutex_enter(&zp
->z_lock
);
1595 may_delete_now
= atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
);
1596 mutex_exit(&zp
->z_lock
);
1599 * We may delete the znode now, or we may put it in the unlinked set;
1600 * it depends on whether we're the last link, and on whether there are
1601 * other holds on the inode. So we dmu_tx_hold() the right things to
1602 * allow for either case.
1605 tx
= dmu_tx_create(zsb
->z_os
);
1606 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1607 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1608 zfs_sa_upgrade_txholds(tx
, zp
);
1609 zfs_sa_upgrade_txholds(tx
, dzp
);
1610 if (may_delete_now
) {
1611 toobig
= zp
->z_size
> zp
->z_blksz
* zfs_delete_blocks
;
1612 /* if the file is too big, only hold_free a token amount */
1613 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1614 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1617 /* are there any extended attributes? */
1618 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1619 &xattr_obj
, sizeof (xattr_obj
));
1620 if (error
== 0 && xattr_obj
) {
1621 error
= zfs_zget(zsb
, xattr_obj
, &xzp
);
1623 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1624 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1627 mutex_enter(&zp
->z_lock
);
1628 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1629 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1630 mutex_exit(&zp
->z_lock
);
1632 /* charge as an update -- would be nice not to charge at all */
1633 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
1636 * Mark this transaction as typically resulting in a net free of space
1638 dmu_tx_mark_netfree(tx
);
1640 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1642 zfs_dirent_unlock(dl
);
1646 if (error
== ERESTART
) {
1660 * Remove the directory entry.
1662 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1671 * Hold z_lock so that we can make sure that the ACL obj
1672 * hasn't changed. Could have been deleted due to
1675 mutex_enter(&zp
->z_lock
);
1676 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1677 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1678 delete_now
= may_delete_now
&& !toobig
&&
1679 atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
) &&
1680 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1685 if (xattr_obj_unlinked
) {
1686 ASSERT3U(ZTOI(xzp
)->i_nlink
, ==, 2);
1687 mutex_enter(&xzp
->z_lock
);
1688 xzp
->z_unlinked
= 1;
1689 clear_nlink(ZTOI(xzp
));
1691 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zsb
),
1692 &links
, sizeof (links
), tx
);
1693 ASSERT3U(error
, ==, 0);
1694 mutex_exit(&xzp
->z_lock
);
1695 zfs_unlinked_add(xzp
, tx
);
1698 error
= sa_remove(zp
->z_sa_hdl
,
1699 SA_ZPL_XATTR(zsb
), tx
);
1701 error
= sa_update(zp
->z_sa_hdl
,
1702 SA_ZPL_XATTR(zsb
), &null_xattr
,
1703 sizeof (uint64_t), tx
);
1707 * Add to the unlinked set because a new reference could be
1708 * taken concurrently resulting in a deferred destruction.
1710 zfs_unlinked_add(zp
, tx
);
1711 mutex_exit(&zp
->z_lock
);
1712 zfs_inode_update(zp
);
1714 } else if (unlinked
) {
1715 mutex_exit(&zp
->z_lock
);
1716 zfs_unlinked_add(zp
, tx
);
1720 if (flags
& FIGNORECASE
)
1722 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1729 zfs_dirent_unlock(dl
);
1730 zfs_inode_update(dzp
);
1733 zfs_inode_update(zp
);
1738 zfs_inode_update(xzp
);
1739 zfs_iput_async(ZTOI(xzp
));
1742 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1743 zil_commit(zilog
, 0);
1748 EXPORT_SYMBOL(zfs_remove
);
1751 * Create a new directory and insert it into dip using the name
1752 * provided. Return a pointer to the inserted directory.
1754 * IN: dip - inode of directory to add subdir to.
1755 * dirname - name of new directory.
1756 * vap - attributes of new directory.
1757 * cr - credentials of caller.
1758 * vsecp - ACL to be set
1760 * OUT: ipp - inode of created directory.
1762 * RETURN: 0 if success
1763 * error code if failure
1766 * dip - ctime|mtime updated
1767 * ipp - ctime|mtime|atime updated
1771 zfs_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
, struct inode
**ipp
,
1772 cred_t
*cr
, int flags
, vsecattr_t
*vsecp
)
1774 znode_t
*zp
, *dzp
= ITOZ(dip
);
1775 zfs_sb_t
*zsb
= ITOZSB(dip
);
1783 gid_t gid
= crgetgid(cr
);
1784 zfs_acl_ids_t acl_ids
;
1785 boolean_t fuid_dirtied
;
1786 boolean_t waited
= B_FALSE
;
1788 ASSERT(S_ISDIR(vap
->va_mode
));
1791 * If we have an ephemeral id, ACL, or XVATTR then
1792 * make sure file system is at proper version
1796 if (zsb
->z_use_fuids
== B_FALSE
&&
1797 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1798 return (SET_ERROR(EINVAL
));
1804 if (dzp
->z_pflags
& ZFS_XATTR
) {
1806 return (SET_ERROR(EINVAL
));
1809 if (zsb
->z_utf8
&& u8_validate(dirname
,
1810 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1812 return (SET_ERROR(EILSEQ
));
1814 if (flags
& FIGNORECASE
)
1817 if (vap
->va_mask
& ATTR_XVATTR
) {
1818 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1819 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1825 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1826 vsecp
, &acl_ids
)) != 0) {
1831 * First make sure the new directory doesn't exist.
1833 * Existence is checked first to make sure we don't return
1834 * EACCES instead of EEXIST which can cause some applications
1840 if ((error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1842 zfs_acl_ids_free(&acl_ids
);
1847 if ((error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
))) {
1848 zfs_acl_ids_free(&acl_ids
);
1849 zfs_dirent_unlock(dl
);
1854 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1855 zfs_acl_ids_free(&acl_ids
);
1856 zfs_dirent_unlock(dl
);
1858 return (SET_ERROR(EDQUOT
));
1862 * Add a new entry to the directory.
1864 tx
= dmu_tx_create(zsb
->z_os
);
1865 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
1866 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1867 fuid_dirtied
= zsb
->z_fuid_dirty
;
1869 zfs_fuid_txhold(zsb
, tx
);
1870 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1871 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
1872 acl_ids
.z_aclp
->z_acl_bytes
);
1875 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1876 ZFS_SA_BASE_ATTR_SIZE
);
1878 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1880 zfs_dirent_unlock(dl
);
1881 if (error
== ERESTART
) {
1887 zfs_acl_ids_free(&acl_ids
);
1896 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1899 zfs_fuid_sync(zsb
, tx
);
1902 * Now put new name in parent dir.
1904 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1908 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
1909 if (flags
& FIGNORECASE
)
1911 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
1912 acl_ids
.z_fuidp
, vap
);
1914 zfs_acl_ids_free(&acl_ids
);
1918 zfs_dirent_unlock(dl
);
1920 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1921 zil_commit(zilog
, 0);
1923 zfs_inode_update(dzp
);
1924 zfs_inode_update(zp
);
1928 EXPORT_SYMBOL(zfs_mkdir
);
1931 * Remove a directory subdir entry. If the current working
1932 * directory is the same as the subdir to be removed, the
1935 * IN: dip - inode of directory to remove from.
1936 * name - name of directory to be removed.
1937 * cwd - inode of current working directory.
1938 * cr - credentials of caller.
1939 * flags - case flags
1941 * RETURN: 0 on success, error code on failure.
1944 * dip - ctime|mtime updated
1948 zfs_rmdir(struct inode
*dip
, char *name
, struct inode
*cwd
, cred_t
*cr
,
1951 znode_t
*dzp
= ITOZ(dip
);
1954 zfs_sb_t
*zsb
= ITOZSB(dip
);
1960 boolean_t waited
= B_FALSE
;
1966 if (flags
& FIGNORECASE
)
1972 * Attempt to lock directory; fail if entry doesn't exist.
1974 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1982 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1986 if (!S_ISDIR(ip
->i_mode
)) {
1987 error
= SET_ERROR(ENOTDIR
);
1992 error
= SET_ERROR(EINVAL
);
1997 * Grab a lock on the directory to make sure that noone is
1998 * trying to add (or lookup) entries while we are removing it.
2000 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
2003 * Grab a lock on the parent pointer to make sure we play well
2004 * with the treewalk and directory rename code.
2006 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
2008 tx
= dmu_tx_create(zsb
->z_os
);
2009 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2010 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2011 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
2012 zfs_sa_upgrade_txholds(tx
, zp
);
2013 zfs_sa_upgrade_txholds(tx
, dzp
);
2014 dmu_tx_mark_netfree(tx
);
2015 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
2017 rw_exit(&zp
->z_parent_lock
);
2018 rw_exit(&zp
->z_name_lock
);
2019 zfs_dirent_unlock(dl
);
2021 if (error
== ERESTART
) {
2032 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2035 uint64_t txtype
= TX_RMDIR
;
2036 if (flags
& FIGNORECASE
)
2038 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2043 rw_exit(&zp
->z_parent_lock
);
2044 rw_exit(&zp
->z_name_lock
);
2046 zfs_dirent_unlock(dl
);
2048 zfs_inode_update(dzp
);
2049 zfs_inode_update(zp
);
2052 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2053 zil_commit(zilog
, 0);
2058 EXPORT_SYMBOL(zfs_rmdir
);
2061 * Read as many directory entries as will fit into the provided
2062 * dirent buffer from the given directory cursor position.
2064 * IN: ip - inode of directory to read.
2065 * dirent - buffer for directory entries.
2067 * OUT: dirent - filler buffer of directory entries.
2069 * RETURN: 0 if success
2070 * error code if failure
2073 * ip - atime updated
2075 * Note that the low 4 bits of the cookie returned by zap is always zero.
2076 * This allows us to use the low range for "special" directory entries:
2077 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2078 * we use the offset 2 for the '.zfs' directory.
2082 zfs_readdir(struct inode
*ip
, struct dir_context
*ctx
, cred_t
*cr
)
2084 znode_t
*zp
= ITOZ(ip
);
2085 zfs_sb_t
*zsb
= ITOZSB(ip
);
2088 zap_attribute_t zap
;
2094 uint64_t offset
; /* must be unsigned; checks for < 1 */
2099 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
2100 &parent
, sizeof (parent
))) != 0)
2104 * Quit if directory has been removed (posix)
2112 prefetch
= zp
->z_zn_prefetch
;
2115 * Initialize the iterator cursor.
2119 * Start iteration from the beginning of the directory.
2121 zap_cursor_init(&zc
, os
, zp
->z_id
);
2124 * The offset is a serialized cursor.
2126 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2130 * Transform to file-system independent format
2135 * Special case `.', `..', and `.zfs'.
2138 (void) strcpy(zap
.za_name
, ".");
2139 zap
.za_normalization_conflict
= 0;
2142 } else if (offset
== 1) {
2143 (void) strcpy(zap
.za_name
, "..");
2144 zap
.za_normalization_conflict
= 0;
2147 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2148 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2149 zap
.za_normalization_conflict
= 0;
2150 objnum
= ZFSCTL_INO_ROOT
;
2156 if ((error
= zap_cursor_retrieve(&zc
, &zap
))) {
2157 if (error
== ENOENT
)
2164 * Allow multiple entries provided the first entry is
2165 * the object id. Non-zpl consumers may safely make
2166 * use of the additional space.
2168 * XXX: This should be a feature flag for compatibility
2170 if (zap
.za_integer_length
!= 8 ||
2171 zap
.za_num_integers
== 0) {
2172 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2173 "entry, obj = %lld, offset = %lld, "
2174 "length = %d, num = %lld\n",
2175 (u_longlong_t
)zp
->z_id
,
2176 (u_longlong_t
)offset
,
2177 zap
.za_integer_length
,
2178 (u_longlong_t
)zap
.za_num_integers
);
2179 error
= SET_ERROR(ENXIO
);
2183 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2184 type
= ZFS_DIRENT_TYPE(zap
.za_first_integer
);
2187 done
= !dir_emit(ctx
, zap
.za_name
, strlen(zap
.za_name
),
2192 /* Prefetch znode */
2194 dmu_prefetch(os
, objnum
, 0, 0, 0,
2195 ZIO_PRIORITY_SYNC_READ
);
2199 * Move to the next entry, fill in the previous offset.
2201 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2202 zap_cursor_advance(&zc
);
2203 offset
= zap_cursor_serialize(&zc
);
2209 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2212 zap_cursor_fini(&zc
);
2213 if (error
== ENOENT
)
2220 EXPORT_SYMBOL(zfs_readdir
);
2222 ulong_t zfs_fsync_sync_cnt
= 4;
2225 zfs_fsync(struct inode
*ip
, int syncflag
, cred_t
*cr
)
2227 znode_t
*zp
= ITOZ(ip
);
2228 zfs_sb_t
*zsb
= ITOZSB(ip
);
2230 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2232 if (zsb
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2235 zil_commit(zsb
->z_log
, zp
->z_id
);
2238 tsd_set(zfs_fsyncer_key
, NULL
);
2242 EXPORT_SYMBOL(zfs_fsync
);
2246 * Get the requested file attributes and place them in the provided
2249 * IN: ip - inode of file.
2250 * vap - va_mask identifies requested attributes.
2251 * If ATTR_XVATTR set, then optional attrs are requested
2252 * flags - ATTR_NOACLCHECK (CIFS server context)
2253 * cr - credentials of caller.
2255 * OUT: vap - attribute values.
2257 * RETURN: 0 (always succeeds)
2261 zfs_getattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2263 znode_t
*zp
= ITOZ(ip
);
2264 zfs_sb_t
*zsb
= ITOZSB(ip
);
2267 uint64_t atime
[2], mtime
[2], ctime
[2];
2268 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2269 xoptattr_t
*xoap
= NULL
;
2270 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2271 sa_bulk_attr_t bulk
[3];
2277 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2279 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zsb
), NULL
, &atime
, 16);
2280 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
2281 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
2283 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2289 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2290 * Also, if we are the owner don't bother, since owner should
2291 * always be allowed to read basic attributes of file.
2293 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2294 (vap
->va_uid
!= crgetuid(cr
))) {
2295 if ((error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2303 * Return all attributes. It's cheaper to provide the answer
2304 * than to determine whether we were asked the question.
2307 mutex_enter(&zp
->z_lock
);
2308 vap
->va_type
= vn_mode_to_vtype(zp
->z_mode
);
2309 vap
->va_mode
= zp
->z_mode
;
2310 vap
->va_fsid
= ZTOI(zp
)->i_sb
->s_dev
;
2311 vap
->va_nodeid
= zp
->z_id
;
2312 if ((zp
->z_id
== zsb
->z_root
) && zfs_show_ctldir(zp
))
2313 links
= ZTOI(zp
)->i_nlink
+ 1;
2315 links
= ZTOI(zp
)->i_nlink
;
2316 vap
->va_nlink
= MIN(links
, ZFS_LINK_MAX
);
2317 vap
->va_size
= i_size_read(ip
);
2318 vap
->va_rdev
= ip
->i_rdev
;
2319 vap
->va_seq
= ip
->i_generation
;
2322 * Add in any requested optional attributes and the create time.
2323 * Also set the corresponding bits in the returned attribute bitmap.
2325 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zsb
->z_use_fuids
) {
2326 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2328 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2329 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2332 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2333 xoap
->xoa_readonly
=
2334 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2335 XVA_SET_RTN(xvap
, XAT_READONLY
);
2338 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2340 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2341 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2344 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2346 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2347 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2350 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2351 xoap
->xoa_nounlink
=
2352 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2353 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2356 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2357 xoap
->xoa_immutable
=
2358 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2359 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2362 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2363 xoap
->xoa_appendonly
=
2364 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2365 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2368 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2370 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2371 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2374 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2376 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2377 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2380 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2381 xoap
->xoa_av_quarantined
=
2382 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2383 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2386 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2387 xoap
->xoa_av_modified
=
2388 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2389 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2392 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2393 S_ISREG(ip
->i_mode
)) {
2394 zfs_sa_get_scanstamp(zp
, xvap
);
2397 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2400 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zsb
),
2401 times
, sizeof (times
));
2402 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2403 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2406 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2407 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2408 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2410 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2411 xoap
->xoa_generation
= ip
->i_generation
;
2412 XVA_SET_RTN(xvap
, XAT_GEN
);
2415 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2417 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2418 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2421 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2423 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2424 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2428 ZFS_TIME_DECODE(&vap
->va_atime
, atime
);
2429 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2430 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2432 mutex_exit(&zp
->z_lock
);
2434 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2436 if (zp
->z_blksz
== 0) {
2438 * Block size hasn't been set; suggest maximal I/O transfers.
2440 vap
->va_blksize
= zsb
->z_max_blksz
;
2446 EXPORT_SYMBOL(zfs_getattr
);
2449 * Get the basic file attributes and place them in the provided kstat
2450 * structure. The inode is assumed to be the authoritative source
2451 * for most of the attributes. However, the znode currently has the
2452 * authoritative atime, blksize, and block count.
2454 * IN: ip - inode of file.
2456 * OUT: sp - kstat values.
2458 * RETURN: 0 (always succeeds)
2462 zfs_getattr_fast(struct inode
*ip
, struct kstat
*sp
)
2464 znode_t
*zp
= ITOZ(ip
);
2465 zfs_sb_t
*zsb
= ITOZSB(ip
);
2467 u_longlong_t nblocks
;
2472 mutex_enter(&zp
->z_lock
);
2474 generic_fillattr(ip
, sp
);
2476 sa_object_size(zp
->z_sa_hdl
, &blksize
, &nblocks
);
2477 sp
->blksize
= blksize
;
2478 sp
->blocks
= nblocks
;
2480 if (unlikely(zp
->z_blksz
== 0)) {
2482 * Block size hasn't been set; suggest maximal I/O transfers.
2484 sp
->blksize
= zsb
->z_max_blksz
;
2487 mutex_exit(&zp
->z_lock
);
2490 * Required to prevent NFS client from detecting different inode
2491 * numbers of snapshot root dentry before and after snapshot mount.
2493 if (zsb
->z_issnap
) {
2494 if (ip
->i_sb
->s_root
->d_inode
== ip
)
2495 sp
->ino
= ZFSCTL_INO_SNAPDIRS
-
2496 dmu_objset_id(zsb
->z_os
);
2503 EXPORT_SYMBOL(zfs_getattr_fast
);
2506 * Set the file attributes to the values contained in the
2509 * IN: ip - inode of file to be modified.
2510 * vap - new attribute values.
2511 * If ATTR_XVATTR set, then optional attrs are being set
2512 * flags - ATTR_UTIME set if non-default time values provided.
2513 * - ATTR_NOACLCHECK (CIFS context only).
2514 * cr - credentials of caller.
2516 * RETURN: 0 if success
2517 * error code if failure
2520 * ip - ctime updated, mtime updated if size changed.
2524 zfs_setattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2526 znode_t
*zp
= ITOZ(ip
);
2527 zfs_sb_t
*zsb
= ITOZSB(ip
);
2531 xvattr_t
*tmpxvattr
;
2532 uint_t mask
= vap
->va_mask
;
2533 uint_t saved_mask
= 0;
2536 uint64_t new_kuid
= 0, new_kgid
= 0, new_uid
, new_gid
;
2538 uint64_t mtime
[2], ctime
[2], atime
[2];
2540 int need_policy
= FALSE
;
2542 zfs_fuid_info_t
*fuidp
= NULL
;
2543 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2546 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2547 boolean_t fuid_dirtied
= B_FALSE
;
2548 sa_bulk_attr_t
*bulk
, *xattr_bulk
;
2549 int count
= 0, xattr_count
= 0;
2560 * Make sure that if we have ephemeral uid/gid or xvattr specified
2561 * that file system is at proper version level
2564 if (zsb
->z_use_fuids
== B_FALSE
&&
2565 (((mask
& ATTR_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2566 ((mask
& ATTR_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2567 (mask
& ATTR_XVATTR
))) {
2569 return (SET_ERROR(EINVAL
));
2572 if (mask
& ATTR_SIZE
&& S_ISDIR(ip
->i_mode
)) {
2574 return (SET_ERROR(EISDIR
));
2577 if (mask
& ATTR_SIZE
&& !S_ISREG(ip
->i_mode
) && !S_ISFIFO(ip
->i_mode
)) {
2579 return (SET_ERROR(EINVAL
));
2583 * If this is an xvattr_t, then get a pointer to the structure of
2584 * optional attributes. If this is NULL, then we have a vattr_t.
2586 xoap
= xva_getxoptattr(xvap
);
2588 tmpxvattr
= kmem_alloc(sizeof (xvattr_t
), KM_SLEEP
);
2589 xva_init(tmpxvattr
);
2591 bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2592 xattr_bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2595 * Immutable files can only alter immutable bit and atime
2597 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2598 ((mask
& (ATTR_SIZE
|ATTR_UID
|ATTR_GID
|ATTR_MTIME
|ATTR_MODE
)) ||
2599 ((mask
& ATTR_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2604 if ((mask
& ATTR_SIZE
) && (zp
->z_pflags
& ZFS_READONLY
)) {
2610 * Verify timestamps doesn't overflow 32 bits.
2611 * ZFS can handle large timestamps, but 32bit syscalls can't
2612 * handle times greater than 2039. This check should be removed
2613 * once large timestamps are fully supported.
2615 if (mask
& (ATTR_ATIME
| ATTR_MTIME
)) {
2616 if (((mask
& ATTR_ATIME
) &&
2617 TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2618 ((mask
& ATTR_MTIME
) &&
2619 TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2629 /* Can this be moved to before the top label? */
2630 if (zfs_is_readonly(zsb
)) {
2636 * First validate permissions
2639 if (mask
& ATTR_SIZE
) {
2640 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2645 * XXX - Note, we are not providing any open
2646 * mode flags here (like FNDELAY), so we may
2647 * block if there are locks present... this
2648 * should be addressed in openat().
2650 /* XXX - would it be OK to generate a log record here? */
2651 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2656 if (mask
& (ATTR_ATIME
|ATTR_MTIME
) ||
2657 ((mask
& ATTR_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2658 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2659 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2660 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2661 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2662 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2663 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2664 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2668 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2669 int idmask
= (mask
& (ATTR_UID
|ATTR_GID
));
2674 * NOTE: even if a new mode is being set,
2675 * we may clear S_ISUID/S_ISGID bits.
2678 if (!(mask
& ATTR_MODE
))
2679 vap
->va_mode
= zp
->z_mode
;
2682 * Take ownership or chgrp to group we are a member of
2685 take_owner
= (mask
& ATTR_UID
) && (vap
->va_uid
== crgetuid(cr
));
2686 take_group
= (mask
& ATTR_GID
) &&
2687 zfs_groupmember(zsb
, vap
->va_gid
, cr
);
2690 * If both ATTR_UID and ATTR_GID are set then take_owner and
2691 * take_group must both be set in order to allow taking
2694 * Otherwise, send the check through secpolicy_vnode_setattr()
2698 if (((idmask
== (ATTR_UID
|ATTR_GID
)) &&
2699 take_owner
&& take_group
) ||
2700 ((idmask
== ATTR_UID
) && take_owner
) ||
2701 ((idmask
== ATTR_GID
) && take_group
)) {
2702 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2703 skipaclchk
, cr
) == 0) {
2705 * Remove setuid/setgid for non-privileged users
2707 (void) secpolicy_setid_clear(vap
, cr
);
2708 trim_mask
= (mask
& (ATTR_UID
|ATTR_GID
));
2717 mutex_enter(&zp
->z_lock
);
2718 oldva
.va_mode
= zp
->z_mode
;
2719 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2720 if (mask
& ATTR_XVATTR
) {
2722 * Update xvattr mask to include only those attributes
2723 * that are actually changing.
2725 * the bits will be restored prior to actually setting
2726 * the attributes so the caller thinks they were set.
2728 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2729 if (xoap
->xoa_appendonly
!=
2730 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2733 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2734 XVA_SET_REQ(tmpxvattr
, XAT_APPENDONLY
);
2738 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2739 if (xoap
->xoa_nounlink
!=
2740 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2743 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2744 XVA_SET_REQ(tmpxvattr
, XAT_NOUNLINK
);
2748 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2749 if (xoap
->xoa_immutable
!=
2750 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2753 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2754 XVA_SET_REQ(tmpxvattr
, XAT_IMMUTABLE
);
2758 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2759 if (xoap
->xoa_nodump
!=
2760 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2763 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2764 XVA_SET_REQ(tmpxvattr
, XAT_NODUMP
);
2768 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2769 if (xoap
->xoa_av_modified
!=
2770 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2773 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2774 XVA_SET_REQ(tmpxvattr
, XAT_AV_MODIFIED
);
2778 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2779 if ((!S_ISREG(ip
->i_mode
) &&
2780 xoap
->xoa_av_quarantined
) ||
2781 xoap
->xoa_av_quarantined
!=
2782 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2785 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2786 XVA_SET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
);
2790 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2791 mutex_exit(&zp
->z_lock
);
2796 if (need_policy
== FALSE
&&
2797 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
2798 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
2803 mutex_exit(&zp
->z_lock
);
2805 if (mask
& ATTR_MODE
) {
2806 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
2807 err
= secpolicy_setid_setsticky_clear(ip
, vap
,
2812 trim_mask
|= ATTR_MODE
;
2820 * If trim_mask is set then take ownership
2821 * has been granted or write_acl is present and user
2822 * has the ability to modify mode. In that case remove
2823 * UID|GID and or MODE from mask so that
2824 * secpolicy_vnode_setattr() doesn't revoke it.
2828 saved_mask
= vap
->va_mask
;
2829 vap
->va_mask
&= ~trim_mask
;
2831 err
= secpolicy_vnode_setattr(cr
, ip
, vap
, &oldva
, flags
,
2832 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
2837 vap
->va_mask
|= saved_mask
;
2841 * secpolicy_vnode_setattr, or take ownership may have
2844 mask
= vap
->va_mask
;
2846 if ((mask
& (ATTR_UID
| ATTR_GID
))) {
2847 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
2848 &xattr_obj
, sizeof (xattr_obj
));
2850 if (err
== 0 && xattr_obj
) {
2851 err
= zfs_zget(ZTOZSB(zp
), xattr_obj
, &attrzp
);
2855 if (mask
& ATTR_UID
) {
2856 new_kuid
= zfs_fuid_create(zsb
,
2857 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
2858 if (new_kuid
!= KUID_TO_SUID(ZTOI(zp
)->i_uid
) &&
2859 zfs_fuid_overquota(zsb
, B_FALSE
, new_kuid
)) {
2867 if (mask
& ATTR_GID
) {
2868 new_kgid
= zfs_fuid_create(zsb
, (uint64_t)vap
->va_gid
,
2869 cr
, ZFS_GROUP
, &fuidp
);
2870 if (new_kgid
!= KGID_TO_SGID(ZTOI(zp
)->i_gid
) &&
2871 zfs_fuid_overquota(zsb
, B_TRUE
, new_kgid
)) {
2879 tx
= dmu_tx_create(zsb
->z_os
);
2881 if (mask
& ATTR_MODE
) {
2882 uint64_t pmode
= zp
->z_mode
;
2884 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
2886 zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
);
2888 mutex_enter(&zp
->z_lock
);
2889 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
2891 * Are we upgrading ACL from old V0 format
2894 if (zsb
->z_version
>= ZPL_VERSION_FUID
&&
2895 zfs_znode_acl_version(zp
) ==
2896 ZFS_ACL_VERSION_INITIAL
) {
2897 dmu_tx_hold_free(tx
, acl_obj
, 0,
2899 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
2900 0, aclp
->z_acl_bytes
);
2902 dmu_tx_hold_write(tx
, acl_obj
, 0,
2905 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
2906 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
2907 0, aclp
->z_acl_bytes
);
2909 mutex_exit(&zp
->z_lock
);
2910 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
2912 if ((mask
& ATTR_XVATTR
) &&
2913 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
2914 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
2916 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2920 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
2923 fuid_dirtied
= zsb
->z_fuid_dirty
;
2925 zfs_fuid_txhold(zsb
, tx
);
2927 zfs_sa_upgrade_txholds(tx
, zp
);
2929 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2935 * Set each attribute requested.
2936 * We group settings according to the locks they need to acquire.
2938 * Note: you cannot set ctime directly, although it will be
2939 * updated as a side-effect of calling this function.
2943 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
2944 mutex_enter(&zp
->z_acl_lock
);
2945 mutex_enter(&zp
->z_lock
);
2947 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
2948 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
2951 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
2952 mutex_enter(&attrzp
->z_acl_lock
);
2953 mutex_enter(&attrzp
->z_lock
);
2954 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2955 SA_ZPL_FLAGS(zsb
), NULL
, &attrzp
->z_pflags
,
2956 sizeof (attrzp
->z_pflags
));
2959 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2961 if (mask
& ATTR_UID
) {
2962 ZTOI(zp
)->i_uid
= SUID_TO_KUID(new_kuid
);
2963 new_uid
= zfs_uid_read(ZTOI(zp
));
2964 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zsb
), NULL
,
2965 &new_uid
, sizeof (new_uid
));
2967 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2968 SA_ZPL_UID(zsb
), NULL
, &new_uid
,
2970 ZTOI(attrzp
)->i_uid
= SUID_TO_KUID(new_uid
);
2974 if (mask
& ATTR_GID
) {
2975 ZTOI(zp
)->i_gid
= SGID_TO_KGID(new_kgid
);
2976 new_gid
= zfs_gid_read(ZTOI(zp
));
2977 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zsb
),
2978 NULL
, &new_gid
, sizeof (new_gid
));
2980 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2981 SA_ZPL_GID(zsb
), NULL
, &new_gid
,
2983 ZTOI(attrzp
)->i_gid
= SGID_TO_KGID(new_kgid
);
2986 if (!(mask
& ATTR_MODE
)) {
2987 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
),
2988 NULL
, &new_mode
, sizeof (new_mode
));
2989 new_mode
= zp
->z_mode
;
2991 err
= zfs_acl_chown_setattr(zp
);
2994 err
= zfs_acl_chown_setattr(attrzp
);
2999 if (mask
& ATTR_MODE
) {
3000 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
), NULL
,
3001 &new_mode
, sizeof (new_mode
));
3002 zp
->z_mode
= ZTOI(zp
)->i_mode
= new_mode
;
3003 ASSERT3P(aclp
, !=, NULL
);
3004 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
3006 if (zp
->z_acl_cached
)
3007 zfs_acl_free(zp
->z_acl_cached
);
3008 zp
->z_acl_cached
= aclp
;
3012 if ((mask
& ATTR_ATIME
) || zp
->z_atime_dirty
) {
3013 zp
->z_atime_dirty
= 0;
3014 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
3015 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zsb
), NULL
,
3016 &atime
, sizeof (atime
));
3019 if (mask
& ATTR_MTIME
) {
3020 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3021 ZTOI(zp
)->i_mtime
= timespec_trunc(vap
->va_mtime
,
3022 ZTOI(zp
)->i_sb
->s_time_gran
);
3024 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
,
3025 mtime
, sizeof (mtime
));
3028 if (mask
& ATTR_CTIME
) {
3029 ZFS_TIME_ENCODE(&vap
->va_ctime
, ctime
);
3030 ZTOI(zp
)->i_ctime
= timespec_trunc(vap
->va_ctime
,
3031 ZTOI(zp
)->i_sb
->s_time_gran
);
3032 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
,
3033 ctime
, sizeof (ctime
));
3036 if (attrzp
&& mask
) {
3037 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3038 SA_ZPL_CTIME(zsb
), NULL
, &ctime
,
3043 * Do this after setting timestamps to prevent timestamp
3044 * update from toggling bit
3047 if (xoap
&& (mask
& ATTR_XVATTR
)) {
3050 * restore trimmed off masks
3051 * so that return masks can be set for caller.
3054 if (XVA_ISSET_REQ(tmpxvattr
, XAT_APPENDONLY
)) {
3055 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3057 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NOUNLINK
)) {
3058 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3060 if (XVA_ISSET_REQ(tmpxvattr
, XAT_IMMUTABLE
)) {
3061 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3063 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NODUMP
)) {
3064 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3066 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_MODIFIED
)) {
3067 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3069 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
)) {
3070 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3073 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3074 ASSERT(S_ISREG(ip
->i_mode
));
3076 zfs_xvattr_set(zp
, xvap
, tx
);
3080 zfs_fuid_sync(zsb
, tx
);
3083 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3085 mutex_exit(&zp
->z_lock
);
3086 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3087 mutex_exit(&zp
->z_acl_lock
);
3090 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3091 mutex_exit(&attrzp
->z_acl_lock
);
3092 mutex_exit(&attrzp
->z_lock
);
3095 if (err
== 0 && attrzp
) {
3096 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3107 zfs_fuid_info_free(fuidp
);
3113 if (err
== ERESTART
)
3116 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3118 zfs_inode_update(zp
);
3122 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3123 zil_commit(zilog
, 0);
3126 kmem_free(xattr_bulk
, sizeof (sa_bulk_attr_t
) * 7);
3127 kmem_free(bulk
, sizeof (sa_bulk_attr_t
) * 7);
3128 kmem_free(tmpxvattr
, sizeof (xvattr_t
));
3132 EXPORT_SYMBOL(zfs_setattr
);
3134 typedef struct zfs_zlock
{
3135 krwlock_t
*zl_rwlock
; /* lock we acquired */
3136 znode_t
*zl_znode
; /* znode we held */
3137 struct zfs_zlock
*zl_next
; /* next in list */
3141 * Drop locks and release vnodes that were held by zfs_rename_lock().
3144 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3148 while ((zl
= *zlpp
) != NULL
) {
3149 if (zl
->zl_znode
!= NULL
)
3150 iput(ZTOI(zl
->zl_znode
));
3151 rw_exit(zl
->zl_rwlock
);
3152 *zlpp
= zl
->zl_next
;
3153 kmem_free(zl
, sizeof (*zl
));
3158 * Search back through the directory tree, using the ".." entries.
3159 * Lock each directory in the chain to prevent concurrent renames.
3160 * Fail any attempt to move a directory into one of its own descendants.
3161 * XXX - z_parent_lock can overlap with map or grow locks
3164 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3168 uint64_t rootid
= ZTOZSB(zp
)->z_root
;
3169 uint64_t oidp
= zp
->z_id
;
3170 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3171 krw_t rw
= RW_WRITER
;
3174 * First pass write-locks szp and compares to zp->z_id.
3175 * Later passes read-lock zp and compare to zp->z_parent.
3178 if (!rw_tryenter(rwlp
, rw
)) {
3180 * Another thread is renaming in this path.
3181 * Note that if we are a WRITER, we don't have any
3182 * parent_locks held yet.
3184 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3186 * Drop our locks and restart
3188 zfs_rename_unlock(&zl
);
3192 rwlp
= &szp
->z_parent_lock
;
3197 * Wait for other thread to drop its locks
3203 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3204 zl
->zl_rwlock
= rwlp
;
3205 zl
->zl_znode
= NULL
;
3206 zl
->zl_next
= *zlpp
;
3209 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3210 return (SET_ERROR(EINVAL
));
3212 if (oidp
== rootid
) /* We've hit the top */
3215 if (rw
== RW_READER
) { /* i.e. not the first pass */
3216 int error
= zfs_zget(ZTOZSB(zp
), oidp
, &zp
);
3221 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(ZTOZSB(zp
)),
3222 &oidp
, sizeof (oidp
));
3223 rwlp
= &zp
->z_parent_lock
;
3226 } while (zp
->z_id
!= sdzp
->z_id
);
3232 * Move an entry from the provided source directory to the target
3233 * directory. Change the entry name as indicated.
3235 * IN: sdip - Source directory containing the "old entry".
3236 * snm - Old entry name.
3237 * tdip - Target directory to contain the "new entry".
3238 * tnm - New entry name.
3239 * cr - credentials of caller.
3240 * flags - case flags
3242 * RETURN: 0 on success, error code on failure.
3245 * sdip,tdip - ctime|mtime updated
3249 zfs_rename(struct inode
*sdip
, char *snm
, struct inode
*tdip
, char *tnm
,
3250 cred_t
*cr
, int flags
)
3252 znode_t
*tdzp
, *szp
, *tzp
;
3253 znode_t
*sdzp
= ITOZ(sdip
);
3254 zfs_sb_t
*zsb
= ITOZSB(sdip
);
3256 zfs_dirlock_t
*sdl
, *tdl
;
3259 int cmp
, serr
, terr
;
3262 boolean_t waited
= B_FALSE
;
3265 ZFS_VERIFY_ZP(sdzp
);
3269 ZFS_VERIFY_ZP(tdzp
);
3272 * We check i_sb because snapshots and the ctldir must have different
3275 if (tdip
->i_sb
!= sdip
->i_sb
|| zfsctl_is_node(tdip
)) {
3277 return (SET_ERROR(EXDEV
));
3280 if (zsb
->z_utf8
&& u8_validate(tnm
,
3281 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3283 return (SET_ERROR(EILSEQ
));
3286 if (flags
& FIGNORECASE
)
3295 * This is to prevent the creation of links into attribute space
3296 * by renaming a linked file into/outof an attribute directory.
3297 * See the comment in zfs_link() for why this is considered bad.
3299 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3301 return (SET_ERROR(EINVAL
));
3305 * Lock source and target directory entries. To prevent deadlock,
3306 * a lock ordering must be defined. We lock the directory with
3307 * the smallest object id first, or if it's a tie, the one with
3308 * the lexically first name.
3310 if (sdzp
->z_id
< tdzp
->z_id
) {
3312 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3316 * First compare the two name arguments without
3317 * considering any case folding.
3319 int nofold
= (zsb
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3321 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3322 ASSERT(error
== 0 || !zsb
->z_utf8
);
3325 * POSIX: "If the old argument and the new argument
3326 * both refer to links to the same existing file,
3327 * the rename() function shall return successfully
3328 * and perform no other action."
3334 * If the file system is case-folding, then we may
3335 * have some more checking to do. A case-folding file
3336 * system is either supporting mixed case sensitivity
3337 * access or is completely case-insensitive. Note
3338 * that the file system is always case preserving.
3340 * In mixed sensitivity mode case sensitive behavior
3341 * is the default. FIGNORECASE must be used to
3342 * explicitly request case insensitive behavior.
3344 * If the source and target names provided differ only
3345 * by case (e.g., a request to rename 'tim' to 'Tim'),
3346 * we will treat this as a special case in the
3347 * case-insensitive mode: as long as the source name
3348 * is an exact match, we will allow this to proceed as
3349 * a name-change request.
3351 if ((zsb
->z_case
== ZFS_CASE_INSENSITIVE
||
3352 (zsb
->z_case
== ZFS_CASE_MIXED
&&
3353 flags
& FIGNORECASE
)) &&
3354 u8_strcmp(snm
, tnm
, 0, zsb
->z_norm
, U8_UNICODE_LATEST
,
3357 * case preserving rename request, require exact
3366 * If the source and destination directories are the same, we should
3367 * grab the z_name_lock of that directory only once.
3371 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3375 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3376 ZEXISTS
| zflg
, NULL
, NULL
);
3377 terr
= zfs_dirent_lock(&tdl
,
3378 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3380 terr
= zfs_dirent_lock(&tdl
,
3381 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3382 serr
= zfs_dirent_lock(&sdl
,
3383 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3389 * Source entry invalid or not there.
3392 zfs_dirent_unlock(tdl
);
3398 rw_exit(&sdzp
->z_name_lock
);
3400 if (strcmp(snm
, "..") == 0)
3406 zfs_dirent_unlock(sdl
);
3410 rw_exit(&sdzp
->z_name_lock
);
3412 if (strcmp(tnm
, "..") == 0)
3419 * Must have write access at the source to remove the old entry
3420 * and write access at the target to create the new entry.
3421 * Note that if target and source are the same, this can be
3422 * done in a single check.
3425 if ((error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
)))
3428 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3430 * Check to make sure rename is valid.
3431 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3433 if ((error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
)))
3438 * Does target exist?
3442 * Source and target must be the same type.
3444 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3445 if (!S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3446 error
= SET_ERROR(ENOTDIR
);
3450 if (S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3451 error
= SET_ERROR(EISDIR
);
3456 * POSIX dictates that when the source and target
3457 * entries refer to the same file object, rename
3458 * must do nothing and exit without error.
3460 if (szp
->z_id
== tzp
->z_id
) {
3466 tx
= dmu_tx_create(zsb
->z_os
);
3467 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3468 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3469 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3470 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3472 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3473 zfs_sa_upgrade_txholds(tx
, tdzp
);
3476 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3477 zfs_sa_upgrade_txholds(tx
, tzp
);
3480 zfs_sa_upgrade_txholds(tx
, szp
);
3481 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
3482 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3485 zfs_rename_unlock(&zl
);
3486 zfs_dirent_unlock(sdl
);
3487 zfs_dirent_unlock(tdl
);
3490 rw_exit(&sdzp
->z_name_lock
);
3495 if (error
== ERESTART
) {
3506 if (tzp
) /* Attempt to remove the existing target */
3507 error
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3510 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3512 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3514 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zsb
),
3515 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3518 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3520 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3521 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3522 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3525 * At this point, we have successfully created
3526 * the target name, but have failed to remove
3527 * the source name. Since the create was done
3528 * with the ZRENAMING flag, there are
3529 * complications; for one, the link count is
3530 * wrong. The easiest way to deal with this
3531 * is to remove the newly created target, and
3532 * return the original error. This must
3533 * succeed; fortunately, it is very unlikely to
3534 * fail, since we just created it.
3536 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3537 ZRENAMING
, NULL
), ==, 0);
3545 zfs_rename_unlock(&zl
);
3547 zfs_dirent_unlock(sdl
);
3548 zfs_dirent_unlock(tdl
);
3550 zfs_inode_update(sdzp
);
3552 rw_exit(&sdzp
->z_name_lock
);
3555 zfs_inode_update(tdzp
);
3557 zfs_inode_update(szp
);
3560 zfs_inode_update(tzp
);
3564 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3565 zil_commit(zilog
, 0);
3570 EXPORT_SYMBOL(zfs_rename
);
3573 * Insert the indicated symbolic reference entry into the directory.
3575 * IN: dip - Directory to contain new symbolic link.
3576 * link - Name for new symlink entry.
3577 * vap - Attributes of new entry.
3578 * target - Target path of new symlink.
3580 * cr - credentials of caller.
3581 * flags - case flags
3583 * RETURN: 0 on success, error code on failure.
3586 * dip - ctime|mtime updated
3590 zfs_symlink(struct inode
*dip
, char *name
, vattr_t
*vap
, char *link
,
3591 struct inode
**ipp
, cred_t
*cr
, int flags
)
3593 znode_t
*zp
, *dzp
= ITOZ(dip
);
3596 zfs_sb_t
*zsb
= ITOZSB(dip
);
3598 uint64_t len
= strlen(link
);
3601 zfs_acl_ids_t acl_ids
;
3602 boolean_t fuid_dirtied
;
3603 uint64_t txtype
= TX_SYMLINK
;
3604 boolean_t waited
= B_FALSE
;
3606 ASSERT(S_ISLNK(vap
->va_mode
));
3612 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
3613 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3615 return (SET_ERROR(EILSEQ
));
3617 if (flags
& FIGNORECASE
)
3620 if (len
> MAXPATHLEN
) {
3622 return (SET_ERROR(ENAMETOOLONG
));
3625 if ((error
= zfs_acl_ids_create(dzp
, 0,
3626 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3634 * Attempt to lock directory; fail if entry already exists.
3636 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3638 zfs_acl_ids_free(&acl_ids
);
3643 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3644 zfs_acl_ids_free(&acl_ids
);
3645 zfs_dirent_unlock(dl
);
3650 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
3651 zfs_acl_ids_free(&acl_ids
);
3652 zfs_dirent_unlock(dl
);
3654 return (SET_ERROR(EDQUOT
));
3656 tx
= dmu_tx_create(zsb
->z_os
);
3657 fuid_dirtied
= zsb
->z_fuid_dirty
;
3658 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3659 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3660 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3661 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3662 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3663 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3664 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3665 acl_ids
.z_aclp
->z_acl_bytes
);
3668 zfs_fuid_txhold(zsb
, tx
);
3669 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3671 zfs_dirent_unlock(dl
);
3672 if (error
== ERESTART
) {
3678 zfs_acl_ids_free(&acl_ids
);
3685 * Create a new object for the symlink.
3686 * for version 4 ZPL datsets the symlink will be an SA attribute
3688 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3691 zfs_fuid_sync(zsb
, tx
);
3693 mutex_enter(&zp
->z_lock
);
3695 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zsb
),
3698 zfs_sa_symlink(zp
, link
, len
, tx
);
3699 mutex_exit(&zp
->z_lock
);
3702 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
3703 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3705 * Insert the new object into the directory.
3707 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3709 if (flags
& FIGNORECASE
)
3711 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3713 zfs_inode_update(dzp
);
3714 zfs_inode_update(zp
);
3716 zfs_acl_ids_free(&acl_ids
);
3720 zfs_dirent_unlock(dl
);
3724 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3725 zil_commit(zilog
, 0);
3730 EXPORT_SYMBOL(zfs_symlink
);
3733 * Return, in the buffer contained in the provided uio structure,
3734 * the symbolic path referred to by ip.
3736 * IN: ip - inode of symbolic link
3737 * uio - structure to contain the link path.
3738 * cr - credentials of caller.
3740 * RETURN: 0 if success
3741 * error code if failure
3744 * ip - atime updated
3748 zfs_readlink(struct inode
*ip
, uio_t
*uio
, cred_t
*cr
)
3750 znode_t
*zp
= ITOZ(ip
);
3751 zfs_sb_t
*zsb
= ITOZSB(ip
);
3757 mutex_enter(&zp
->z_lock
);
3759 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3760 SA_ZPL_SYMLINK(zsb
), uio
);
3762 error
= zfs_sa_readlink(zp
, uio
);
3763 mutex_exit(&zp
->z_lock
);
3768 EXPORT_SYMBOL(zfs_readlink
);
3771 * Insert a new entry into directory tdip referencing sip.
3773 * IN: tdip - Directory to contain new entry.
3774 * sip - inode of new entry.
3775 * name - name of new entry.
3776 * cr - credentials of caller.
3778 * RETURN: 0 if success
3779 * error code if failure
3782 * tdip - ctime|mtime updated
3783 * sip - ctime updated
3787 zfs_link(struct inode
*tdip
, struct inode
*sip
, char *name
, cred_t
*cr
,
3790 znode_t
*dzp
= ITOZ(tdip
);
3792 zfs_sb_t
*zsb
= ITOZSB(tdip
);
3800 boolean_t waited
= B_FALSE
;
3802 ASSERT(S_ISDIR(tdip
->i_mode
));
3809 * POSIX dictates that we return EPERM here.
3810 * Better choices include ENOTSUP or EISDIR.
3812 if (S_ISDIR(sip
->i_mode
)) {
3814 return (SET_ERROR(EPERM
));
3821 * We check i_sb because snapshots and the ctldir must have different
3824 if (sip
->i_sb
!= tdip
->i_sb
|| zfsctl_is_node(sip
)) {
3826 return (SET_ERROR(EXDEV
));
3829 /* Prevent links to .zfs/shares files */
3831 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
3832 &parent
, sizeof (uint64_t))) != 0) {
3836 if (parent
== zsb
->z_shares_dir
) {
3838 return (SET_ERROR(EPERM
));
3841 if (zsb
->z_utf8
&& u8_validate(name
,
3842 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3844 return (SET_ERROR(EILSEQ
));
3846 if (flags
& FIGNORECASE
)
3850 * We do not support links between attributes and non-attributes
3851 * because of the potential security risk of creating links
3852 * into "normal" file space in order to circumvent restrictions
3853 * imposed in attribute space.
3855 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
3857 return (SET_ERROR(EINVAL
));
3860 owner
= zfs_fuid_map_id(zsb
, KUID_TO_SUID(sip
->i_uid
), cr
, ZFS_OWNER
);
3861 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
3863 return (SET_ERROR(EPERM
));
3866 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3873 * Attempt to lock directory; fail if entry already exists.
3875 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
3881 tx
= dmu_tx_create(zsb
->z_os
);
3882 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3883 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3884 zfs_sa_upgrade_txholds(tx
, szp
);
3885 zfs_sa_upgrade_txholds(tx
, dzp
);
3886 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3888 zfs_dirent_unlock(dl
);
3889 if (error
== ERESTART
) {
3900 error
= zfs_link_create(dl
, szp
, tx
, 0);
3903 uint64_t txtype
= TX_LINK
;
3904 if (flags
& FIGNORECASE
)
3906 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
3911 zfs_dirent_unlock(dl
);
3913 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3914 zil_commit(zilog
, 0);
3916 zfs_inode_update(dzp
);
3917 zfs_inode_update(szp
);
3921 EXPORT_SYMBOL(zfs_link
);
3924 zfs_putpage_commit_cb(void *arg
)
3926 struct page
*pp
= arg
;
3929 end_page_writeback(pp
);
3933 * Push a page out to disk, once the page is on stable storage the
3934 * registered commit callback will be run as notification of completion.
3936 * IN: ip - page mapped for inode.
3937 * pp - page to push (page is locked)
3938 * wbc - writeback control data
3940 * RETURN: 0 if success
3941 * error code if failure
3944 * ip - ctime|mtime updated
3948 zfs_putpage(struct inode
*ip
, struct page
*pp
, struct writeback_control
*wbc
)
3950 znode_t
*zp
= ITOZ(ip
);
3951 zfs_sb_t
*zsb
= ITOZSB(ip
);
3959 uint64_t mtime
[2], ctime
[2];
3960 sa_bulk_attr_t bulk
[3];
3962 struct address_space
*mapping
;
3967 ASSERT(PageLocked(pp
));
3969 pgoff
= page_offset(pp
); /* Page byte-offset in file */
3970 offset
= i_size_read(ip
); /* File length in bytes */
3971 pglen
= MIN(PAGE_SIZE
, /* Page length in bytes */
3972 P2ROUNDUP(offset
, PAGE_SIZE
)-pgoff
);
3974 /* Page is beyond end of file */
3975 if (pgoff
>= offset
) {
3981 /* Truncate page length to end of file */
3982 if (pgoff
+ pglen
> offset
)
3983 pglen
= offset
- pgoff
;
3987 * FIXME: Allow mmap writes past its quota. The correct fix
3988 * is to register a page_mkwrite() handler to count the page
3989 * against its quota when it is about to be dirtied.
3991 if (zfs_owner_overquota(zsb
, zp
, B_FALSE
) ||
3992 zfs_owner_overquota(zsb
, zp
, B_TRUE
)) {
3998 * The ordering here is critical and must adhere to the following
3999 * rules in order to avoid deadlocking in either zfs_read() or
4000 * zfs_free_range() due to a lock inversion.
4002 * 1) The page must be unlocked prior to acquiring the range lock.
4003 * This is critical because zfs_read() calls find_lock_page()
4004 * which may block on the page lock while holding the range lock.
4006 * 2) Before setting or clearing write back on a page the range lock
4007 * must be held in order to prevent a lock inversion with the
4008 * zfs_free_range() function.
4010 * This presents a problem because upon entering this function the
4011 * page lock is already held. To safely acquire the range lock the
4012 * page lock must be dropped. This creates a window where another
4013 * process could truncate, invalidate, dirty, or write out the page.
4015 * Therefore, after successfully reacquiring the range and page locks
4016 * the current page state is checked. In the common case everything
4017 * will be as is expected and it can be written out. However, if
4018 * the page state has changed it must be handled accordingly.
4020 mapping
= pp
->mapping
;
4021 redirty_page_for_writepage(wbc
, pp
);
4024 rl
= zfs_range_lock(&zp
->z_range_lock
, pgoff
, pglen
, RL_WRITER
);
4027 /* Page mapping changed or it was no longer dirty, we're done */
4028 if (unlikely((mapping
!= pp
->mapping
) || !PageDirty(pp
))) {
4030 zfs_range_unlock(rl
);
4035 /* Another process started write block if required */
4036 if (PageWriteback(pp
)) {
4038 zfs_range_unlock(rl
);
4040 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
4041 wait_on_page_writeback(pp
);
4047 /* Clear the dirty flag the required locks are held */
4048 if (!clear_page_dirty_for_io(pp
)) {
4050 zfs_range_unlock(rl
);
4056 * Counterpart for redirty_page_for_writepage() above. This page
4057 * was in fact not skipped and should not be counted as if it were.
4059 wbc
->pages_skipped
--;
4060 set_page_writeback(pp
);
4063 tx
= dmu_tx_create(zsb
->z_os
);
4064 dmu_tx_hold_write(tx
, zp
->z_id
, pgoff
, pglen
);
4065 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4066 zfs_sa_upgrade_txholds(tx
, zp
);
4068 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4070 if (err
== ERESTART
)
4074 __set_page_dirty_nobuffers(pp
);
4076 end_page_writeback(pp
);
4077 zfs_range_unlock(rl
);
4083 ASSERT3U(pglen
, <=, PAGE_SIZE
);
4084 dmu_write(zsb
->z_os
, zp
->z_id
, pgoff
, pglen
, va
, tx
);
4087 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
4088 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
4089 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_FLAGS(zsb
), NULL
, &zp
->z_pflags
, 8);
4091 /* Preserve the mtime and ctime provided by the inode */
4092 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4093 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4094 zp
->z_atime_dirty
= 0;
4097 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4099 zfs_log_write(zsb
->z_log
, tx
, TX_WRITE
, zp
, pgoff
, pglen
, 0,
4100 zfs_putpage_commit_cb
, pp
);
4103 zfs_range_unlock(rl
);
4105 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4107 * Note that this is rarely called under writepages(), because
4108 * writepages() normally handles the entire commit for
4109 * performance reasons.
4111 if (zsb
->z_log
!= NULL
)
4112 zil_commit(zsb
->z_log
, zp
->z_id
);
4120 * Update the system attributes when the inode has been dirtied. For the
4121 * moment we only update the mode, atime, mtime, and ctime.
4124 zfs_dirty_inode(struct inode
*ip
, int flags
)
4126 znode_t
*zp
= ITOZ(ip
);
4127 zfs_sb_t
*zsb
= ITOZSB(ip
);
4129 uint64_t mode
, atime
[2], mtime
[2], ctime
[2];
4130 sa_bulk_attr_t bulk
[4];
4134 if (zfs_is_readonly(zsb
) || dmu_objset_is_snapshot(zsb
->z_os
))
4142 * This is the lazytime semantic indroduced in Linux 4.0
4143 * This flag will only be called from update_time when lazytime is set.
4144 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4145 * Fortunately mtime and ctime are managed within ZFS itself, so we
4146 * only need to dirty atime.
4148 if (flags
== I_DIRTY_TIME
) {
4149 zp
->z_atime_dirty
= 1;
4154 tx
= dmu_tx_create(zsb
->z_os
);
4156 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4157 zfs_sa_upgrade_txholds(tx
, zp
);
4159 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4165 mutex_enter(&zp
->z_lock
);
4166 zp
->z_atime_dirty
= 0;
4168 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MODE(zsb
), NULL
, &mode
, 8);
4169 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_ATIME(zsb
), NULL
, &atime
, 16);
4170 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
4171 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
4173 /* Preserve the mode, mtime and ctime provided by the inode */
4174 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4175 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4176 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4181 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4182 mutex_exit(&zp
->z_lock
);
4189 EXPORT_SYMBOL(zfs_dirty_inode
);
4193 zfs_inactive(struct inode
*ip
)
4195 znode_t
*zp
= ITOZ(ip
);
4196 zfs_sb_t
*zsb
= ITOZSB(ip
);
4199 int need_unlock
= 0;
4201 /* Only read lock if we haven't already write locked, e.g. rollback */
4202 if (!RW_WRITE_HELD(&zsb
->z_teardown_inactive_lock
)) {
4204 rw_enter(&zsb
->z_teardown_inactive_lock
, RW_READER
);
4206 if (zp
->z_sa_hdl
== NULL
) {
4208 rw_exit(&zsb
->z_teardown_inactive_lock
);
4212 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4213 dmu_tx_t
*tx
= dmu_tx_create(zsb
->z_os
);
4215 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4216 zfs_sa_upgrade_txholds(tx
, zp
);
4217 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4221 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4222 mutex_enter(&zp
->z_lock
);
4223 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zsb
),
4224 (void *)&atime
, sizeof (atime
), tx
);
4225 zp
->z_atime_dirty
= 0;
4226 mutex_exit(&zp
->z_lock
);
4233 rw_exit(&zsb
->z_teardown_inactive_lock
);
4235 EXPORT_SYMBOL(zfs_inactive
);
4238 * Bounds-check the seek operation.
4240 * IN: ip - inode seeking within
4241 * ooff - old file offset
4242 * noffp - pointer to new file offset
4243 * ct - caller context
4245 * RETURN: 0 if success
4246 * EINVAL if new offset invalid
4250 zfs_seek(struct inode
*ip
, offset_t ooff
, offset_t
*noffp
)
4252 if (S_ISDIR(ip
->i_mode
))
4254 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4256 EXPORT_SYMBOL(zfs_seek
);
4259 * Fill pages with data from the disk.
4262 zfs_fillpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4264 znode_t
*zp
= ITOZ(ip
);
4265 zfs_sb_t
*zsb
= ITOZSB(ip
);
4267 struct page
*cur_pp
;
4268 u_offset_t io_off
, total
;
4275 io_len
= nr_pages
<< PAGE_SHIFT
;
4276 i_size
= i_size_read(ip
);
4277 io_off
= page_offset(pl
[0]);
4279 if (io_off
+ io_len
> i_size
)
4280 io_len
= i_size
- io_off
;
4283 * Iterate over list of pages and read each page individually.
4286 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4289 cur_pp
= pl
[page_idx
++];
4291 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4295 /* convert checksum errors into IO errors */
4297 err
= SET_ERROR(EIO
);
4306 * Uses zfs_fillpage to read data from the file and fill the pages.
4308 * IN: ip - inode of file to get data from.
4309 * pl - list of pages to read
4310 * nr_pages - number of pages to read
4312 * RETURN: 0 on success, error code on failure.
4315 * vp - atime updated
4319 zfs_getpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4321 znode_t
*zp
= ITOZ(ip
);
4322 zfs_sb_t
*zsb
= ITOZSB(ip
);
4331 err
= zfs_fillpage(ip
, pl
, nr_pages
);
4336 EXPORT_SYMBOL(zfs_getpage
);
4339 * Check ZFS specific permissions to memory map a section of a file.
4341 * IN: ip - inode of the file to mmap
4343 * addrp - start address in memory region
4344 * len - length of memory region
4345 * vm_flags- address flags
4347 * RETURN: 0 if success
4348 * error code if failure
4352 zfs_map(struct inode
*ip
, offset_t off
, caddr_t
*addrp
, size_t len
,
4353 unsigned long vm_flags
)
4355 znode_t
*zp
= ITOZ(ip
);
4356 zfs_sb_t
*zsb
= ITOZSB(ip
);
4361 if ((vm_flags
& VM_WRITE
) && (zp
->z_pflags
&
4362 (ZFS_IMMUTABLE
| ZFS_READONLY
| ZFS_APPENDONLY
))) {
4364 return (SET_ERROR(EPERM
));
4367 if ((vm_flags
& (VM_READ
| VM_EXEC
)) &&
4368 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4370 return (SET_ERROR(EACCES
));
4373 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4375 return (SET_ERROR(ENXIO
));
4381 EXPORT_SYMBOL(zfs_map
);
4384 * convoff - converts the given data (start, whence) to the
4388 convoff(struct inode
*ip
, flock64_t
*lckdat
, int whence
, offset_t offset
)
4393 if ((lckdat
->l_whence
== 2) || (whence
== 2)) {
4394 if ((error
= zfs_getattr(ip
, &vap
, 0, CRED()) != 0))
4398 switch (lckdat
->l_whence
) {
4400 lckdat
->l_start
+= offset
;
4403 lckdat
->l_start
+= vap
.va_size
;
4408 return (SET_ERROR(EINVAL
));
4411 if (lckdat
->l_start
< 0)
4412 return (SET_ERROR(EINVAL
));
4416 lckdat
->l_start
-= offset
;
4419 lckdat
->l_start
-= vap
.va_size
;
4424 return (SET_ERROR(EINVAL
));
4427 lckdat
->l_whence
= (short)whence
;
4432 * Free or allocate space in a file. Currently, this function only
4433 * supports the `F_FREESP' command. However, this command is somewhat
4434 * misnamed, as its functionality includes the ability to allocate as
4435 * well as free space.
4437 * IN: ip - inode of file to free data in.
4438 * cmd - action to take (only F_FREESP supported).
4439 * bfp - section of file to free/alloc.
4440 * flag - current file open mode flags.
4441 * offset - current file offset.
4442 * cr - credentials of caller [UNUSED].
4444 * RETURN: 0 on success, error code on failure.
4447 * ip - ctime|mtime updated
4451 zfs_space(struct inode
*ip
, int cmd
, flock64_t
*bfp
, int flag
,
4452 offset_t offset
, cred_t
*cr
)
4454 znode_t
*zp
= ITOZ(ip
);
4455 zfs_sb_t
*zsb
= ITOZSB(ip
);
4462 if (cmd
!= F_FREESP
) {
4464 return (SET_ERROR(EINVAL
));
4468 * Callers might not be able to detect properly that we are read-only,
4469 * so check it explicitly here.
4471 if (zfs_is_readonly(zsb
)) {
4473 return (SET_ERROR(EROFS
));
4476 if ((error
= convoff(ip
, bfp
, 0, offset
))) {
4481 if (bfp
->l_len
< 0) {
4483 return (SET_ERROR(EINVAL
));
4487 * Permissions aren't checked on Solaris because on this OS
4488 * zfs_space() can only be called with an opened file handle.
4489 * On Linux we can get here through truncate_range() which
4490 * operates directly on inodes, so we need to check access rights.
4492 if ((error
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
))) {
4498 len
= bfp
->l_len
; /* 0 means from off to end of file */
4500 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4505 EXPORT_SYMBOL(zfs_space
);
4509 zfs_fid(struct inode
*ip
, fid_t
*fidp
)
4511 znode_t
*zp
= ITOZ(ip
);
4512 zfs_sb_t
*zsb
= ITOZSB(ip
);
4515 uint64_t object
= zp
->z_id
;
4522 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zsb
),
4523 &gen64
, sizeof (uint64_t))) != 0) {
4528 gen
= (uint32_t)gen64
;
4530 size
= (zsb
->z_parent
!= zsb
) ? LONG_FID_LEN
: SHORT_FID_LEN
;
4531 if (fidp
->fid_len
< size
) {
4532 fidp
->fid_len
= size
;
4534 return (SET_ERROR(ENOSPC
));
4537 zfid
= (zfid_short_t
*)fidp
;
4539 zfid
->zf_len
= size
;
4541 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4542 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4544 /* Must have a non-zero generation number to distinguish from .zfs */
4547 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4548 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4550 if (size
== LONG_FID_LEN
) {
4551 uint64_t objsetid
= dmu_objset_id(zsb
->z_os
);
4554 zlfid
= (zfid_long_t
*)fidp
;
4556 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
4557 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
4559 /* XXX - this should be the generation number for the objset */
4560 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
4561 zlfid
->zf_setgen
[i
] = 0;
4567 EXPORT_SYMBOL(zfs_fid
);
4571 zfs_getsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4573 znode_t
*zp
= ITOZ(ip
);
4574 zfs_sb_t
*zsb
= ITOZSB(ip
);
4576 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4580 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
4585 EXPORT_SYMBOL(zfs_getsecattr
);
4589 zfs_setsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4591 znode_t
*zp
= ITOZ(ip
);
4592 zfs_sb_t
*zsb
= ITOZSB(ip
);
4594 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4595 zilog_t
*zilog
= zsb
->z_log
;
4600 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
4602 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4603 zil_commit(zilog
, 0);
4608 EXPORT_SYMBOL(zfs_setsecattr
);
4610 #ifdef HAVE_UIO_ZEROCOPY
4612 * Tunable, both must be a power of 2.
4614 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4615 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4616 * an arcbuf for a partial block read
4618 int zcr_blksz_min
= (1 << 10); /* 1K */
4619 int zcr_blksz_max
= (1 << 17); /* 128K */
4623 zfs_reqzcbuf(struct inode
*ip
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
)
4625 znode_t
*zp
= ITOZ(ip
);
4626 zfs_sb_t
*zsb
= ITOZSB(ip
);
4627 int max_blksz
= zsb
->z_max_blksz
;
4628 uio_t
*uio
= &xuio
->xu_uio
;
4629 ssize_t size
= uio
->uio_resid
;
4630 offset_t offset
= uio
->uio_loffset
;
4635 int preamble
, postamble
;
4637 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
4638 return (SET_ERROR(EINVAL
));
4645 * Loan out an arc_buf for write if write size is bigger than
4646 * max_blksz, and the file's block size is also max_blksz.
4649 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
4651 return (SET_ERROR(EINVAL
));
4654 * Caller requests buffers for write before knowing where the
4655 * write offset might be (e.g. NFS TCP write).
4660 preamble
= P2PHASE(offset
, blksz
);
4662 preamble
= blksz
- preamble
;
4667 postamble
= P2PHASE(size
, blksz
);
4670 fullblk
= size
/ blksz
;
4671 (void) dmu_xuio_init(xuio
,
4672 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
4675 * Have to fix iov base/len for partial buffers. They
4676 * currently represent full arc_buf's.
4679 /* data begins in the middle of the arc_buf */
4680 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4683 (void) dmu_xuio_add(xuio
, abuf
,
4684 blksz
- preamble
, preamble
);
4687 for (i
= 0; i
< fullblk
; i
++) {
4688 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4691 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
4695 /* data ends in the middle of the arc_buf */
4696 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4699 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
4704 * Loan out an arc_buf for read if the read size is larger than
4705 * the current file block size. Block alignment is not
4706 * considered. Partial arc_buf will be loaned out for read.
4708 blksz
= zp
->z_blksz
;
4709 if (blksz
< zcr_blksz_min
)
4710 blksz
= zcr_blksz_min
;
4711 if (blksz
> zcr_blksz_max
)
4712 blksz
= zcr_blksz_max
;
4713 /* avoid potential complexity of dealing with it */
4714 if (blksz
> max_blksz
) {
4716 return (SET_ERROR(EINVAL
));
4719 maxsize
= zp
->z_size
- uio
->uio_loffset
;
4725 return (SET_ERROR(EINVAL
));
4730 return (SET_ERROR(EINVAL
));
4733 uio
->uio_extflg
= UIO_XUIO
;
4734 XUIO_XUZC_RW(xuio
) = ioflag
;
4741 zfs_retzcbuf(struct inode
*ip
, xuio_t
*xuio
, cred_t
*cr
)
4745 int ioflag
= XUIO_XUZC_RW(xuio
);
4747 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
4749 i
= dmu_xuio_cnt(xuio
);
4751 abuf
= dmu_xuio_arcbuf(xuio
, i
);
4753 * if abuf == NULL, it must be a write buffer
4754 * that has been returned in zfs_write().
4757 dmu_return_arcbuf(abuf
);
4758 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
4761 dmu_xuio_fini(xuio
);
4764 #endif /* HAVE_UIO_ZEROCOPY */
4766 #if defined(_KERNEL) && defined(HAVE_SPL)
4767 module_param(zfs_delete_blocks
, ulong
, 0644);
4768 MODULE_PARM_DESC(zfs_delete_blocks
, "Delete files larger than N blocks async");
4769 module_param(zfs_read_chunk_size
, long, 0644);
4770 MODULE_PARM_DESC(zfs_read_chunk_size
, "Bytes to read per chunk");