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) 2013 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_CACHE_SIZE
-1);
337 for (start
&= PAGE_CACHE_MASK
; len
> 0; start
+= PAGE_CACHE_SIZE
) {
338 nbytes
= MIN(PAGE_CACHE_SIZE
- off
, len
);
340 pp
= find_lock_page(mp
, start
>> PAGE_CACHE_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
);
357 page_cache_release(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_CACHE_SIZE
-1);
389 for (start
&= PAGE_CACHE_MASK
; len
> 0; start
+= PAGE_CACHE_SIZE
) {
390 bytes
= MIN(PAGE_CACHE_SIZE
- off
, len
);
392 pp
= find_lock_page(mp
, start
>> PAGE_CACHE_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
);
405 page_cache_release(pp
);
407 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
420 unsigned long zfs_read_chunk_size
= 1024 * 1024; /* Tunable */
423 * Read bytes from specified file into supplied buffer.
425 * IN: ip - inode of file to be read from.
426 * uio - structure supplying read location, range info,
428 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
429 * O_DIRECT flag; used to bypass page cache.
430 * cr - credentials of caller.
432 * OUT: uio - updated offset and range, buffer filled.
434 * RETURN: 0 on success, error code on failure.
437 * inode - atime updated if byte count > 0
441 zfs_read(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
443 znode_t
*zp
= ITOZ(ip
);
444 zfs_sb_t
*zsb
= ITOZSB(ip
);
448 #ifdef HAVE_UIO_ZEROCOPY
450 #endif /* HAVE_UIO_ZEROCOPY */
455 if (zp
->z_pflags
& ZFS_AV_QUARANTINED
) {
457 return (SET_ERROR(EACCES
));
461 * Validate file offset
463 if (uio
->uio_loffset
< (offset_t
)0) {
465 return (SET_ERROR(EINVAL
));
469 * Fasttrack empty reads
471 if (uio
->uio_resid
== 0) {
477 * If we're in FRSYNC mode, sync out this znode before reading it.
479 if (ioflag
& FRSYNC
|| zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
480 zil_commit(zsb
->z_log
, zp
->z_id
);
483 * Lock the range against changes.
485 rl
= zfs_range_lock(zp
, uio
->uio_loffset
, uio
->uio_resid
, RL_READER
);
488 * If we are reading past end-of-file we can skip
489 * to the end; but we might still need to set atime.
491 if (uio
->uio_loffset
>= zp
->z_size
) {
496 ASSERT(uio
->uio_loffset
< zp
->z_size
);
497 n
= MIN(uio
->uio_resid
, zp
->z_size
- uio
->uio_loffset
);
499 #ifdef HAVE_UIO_ZEROCOPY
500 if ((uio
->uio_extflg
== UIO_XUIO
) &&
501 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
)) {
503 int blksz
= zp
->z_blksz
;
504 uint64_t offset
= uio
->uio_loffset
;
506 xuio
= (xuio_t
*)uio
;
508 nblk
= (P2ROUNDUP(offset
+ n
, blksz
) - P2ALIGN(offset
,
511 ASSERT(offset
+ n
<= blksz
);
514 (void) dmu_xuio_init(xuio
, nblk
);
516 if (vn_has_cached_data(ip
)) {
518 * For simplicity, we always allocate a full buffer
519 * even if we only expect to read a portion of a block.
521 while (--nblk
>= 0) {
522 (void) dmu_xuio_add(xuio
,
523 dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
528 #endif /* HAVE_UIO_ZEROCOPY */
531 nbytes
= MIN(n
, zfs_read_chunk_size
-
532 P2PHASE(uio
->uio_loffset
, zfs_read_chunk_size
));
534 if (zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
)) {
535 error
= mappedread(ip
, nbytes
, uio
);
537 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
542 /* convert checksum errors into IO errors */
544 error
= SET_ERROR(EIO
);
551 zfs_range_unlock(rl
);
553 ZFS_ACCESSTIME_STAMP(zsb
, zp
);
557 EXPORT_SYMBOL(zfs_read
);
560 * Write the bytes to a file.
562 * IN: ip - inode of file to be written to.
563 * uio - structure supplying write location, range info,
565 * ioflag - FAPPEND flag set if in append mode.
566 * O_DIRECT flag; used to bypass page cache.
567 * cr - credentials of caller.
569 * OUT: uio - updated offset and range.
571 * RETURN: 0 if success
572 * error code if failure
575 * ip - ctime|mtime updated if byte count > 0
580 zfs_write(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
582 znode_t
*zp
= ITOZ(ip
);
583 rlim64_t limit
= uio
->uio_limit
;
584 ssize_t start_resid
= uio
->uio_resid
;
588 zfs_sb_t
*zsb
= ZTOZSB(zp
);
593 int max_blksz
= zsb
->z_max_blksz
;
596 const iovec_t
*aiov
= NULL
;
599 const iovec_t
*iovp
= uio
->uio_iov
;
602 sa_bulk_attr_t bulk
[4];
603 uint64_t mtime
[2], ctime
[2];
604 ASSERTV(int iovcnt
= uio
->uio_iovcnt
);
607 * Fasttrack empty write
613 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
619 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
620 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
621 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zsb
), NULL
, &zp
->z_size
, 8);
622 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
626 * Callers might not be able to detect properly that we are read-only,
627 * so check it explicitly here.
629 if (zfs_is_readonly(zsb
)) {
631 return (SET_ERROR(EROFS
));
635 * If immutable or not appending then return EPERM
637 if ((zp
->z_pflags
& (ZFS_IMMUTABLE
| ZFS_READONLY
)) ||
638 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
639 (uio
->uio_loffset
< zp
->z_size
))) {
641 return (SET_ERROR(EPERM
));
647 * Validate file offset
649 woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
652 return (SET_ERROR(EINVAL
));
656 * Pre-fault the pages to ensure slow (eg NFS) pages
658 * Skip this if uio contains loaned arc_buf.
660 #ifdef HAVE_UIO_ZEROCOPY
661 if ((uio
->uio_extflg
== UIO_XUIO
) &&
662 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
))
663 xuio
= (xuio_t
*)uio
;
666 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
669 * If in append mode, set the io offset pointer to eof.
671 if (ioflag
& FAPPEND
) {
673 * Obtain an appending range lock to guarantee file append
674 * semantics. We reset the write offset once we have the lock.
676 rl
= zfs_range_lock(zp
, 0, n
, RL_APPEND
);
678 if (rl
->r_len
== UINT64_MAX
) {
680 * We overlocked the file because this write will cause
681 * the file block size to increase.
682 * Note that zp_size cannot change with this lock held.
686 uio
->uio_loffset
= woff
;
689 * Note that if the file block size will change as a result of
690 * this write, then this range lock will lock the entire file
691 * so that we can re-write the block safely.
693 rl
= zfs_range_lock(zp
, woff
, n
, RL_WRITER
);
697 zfs_range_unlock(rl
);
699 return (SET_ERROR(EFBIG
));
702 if ((woff
+ n
) > limit
|| woff
> (limit
- n
))
705 /* Will this write extend the file length? */
706 write_eof
= (woff
+ n
> zp
->z_size
);
708 end_size
= MAX(zp
->z_size
, woff
+ n
);
711 * Write the file in reasonable size chunks. Each chunk is written
712 * in a separate transaction; this keeps the intent log records small
713 * and allows us to do more fine-grained space accounting.
717 woff
= uio
->uio_loffset
;
718 if (zfs_owner_overquota(zsb
, zp
, B_FALSE
) ||
719 zfs_owner_overquota(zsb
, zp
, B_TRUE
)) {
721 dmu_return_arcbuf(abuf
);
722 error
= SET_ERROR(EDQUOT
);
726 if (xuio
&& abuf
== NULL
) {
727 ASSERT(i_iov
< iovcnt
);
728 ASSERT3U(uio
->uio_segflg
, !=, UIO_BVEC
);
730 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
731 dmu_xuio_clear(xuio
, i_iov
);
732 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
733 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
734 aiov
->iov_len
== arc_buf_size(abuf
)));
736 } else if (abuf
== NULL
&& n
>= max_blksz
&&
737 woff
>= zp
->z_size
&&
738 P2PHASE(woff
, max_blksz
) == 0 &&
739 zp
->z_blksz
== max_blksz
) {
741 * This write covers a full block. "Borrow" a buffer
742 * from the dmu so that we can fill it before we enter
743 * a transaction. This avoids the possibility of
744 * holding up the transaction if the data copy hangs
745 * up on a pagefault (e.g., from an NFS server mapping).
749 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
751 ASSERT(abuf
!= NULL
);
752 ASSERT(arc_buf_size(abuf
) == max_blksz
);
753 if ((error
= uiocopy(abuf
->b_data
, max_blksz
,
754 UIO_WRITE
, uio
, &cbytes
))) {
755 dmu_return_arcbuf(abuf
);
758 ASSERT(cbytes
== max_blksz
);
762 * Start a transaction.
764 tx
= dmu_tx_create(zsb
->z_os
);
765 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
766 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
767 zfs_sa_upgrade_txholds(tx
, zp
);
768 error
= dmu_tx_assign(tx
, TXG_WAIT
);
772 dmu_return_arcbuf(abuf
);
777 * If zfs_range_lock() over-locked we grow the blocksize
778 * and then reduce the lock range. This will only happen
779 * on the first iteration since zfs_range_reduce() will
780 * shrink down r_len to the appropriate size.
782 if (rl
->r_len
== UINT64_MAX
) {
785 if (zp
->z_blksz
> max_blksz
) {
787 * File's blocksize is already larger than the
788 * "recordsize" property. Only let it grow to
789 * the next power of 2.
791 ASSERT(!ISP2(zp
->z_blksz
));
792 new_blksz
= MIN(end_size
,
793 1 << highbit64(zp
->z_blksz
));
795 new_blksz
= MIN(end_size
, max_blksz
);
797 zfs_grow_blocksize(zp
, new_blksz
, tx
);
798 zfs_range_reduce(rl
, woff
, n
);
802 * XXX - should we really limit each write to z_max_blksz?
803 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
805 nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
808 tx_bytes
= uio
->uio_resid
;
809 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
811 tx_bytes
-= uio
->uio_resid
;
814 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
816 * If this is not a full block write, but we are
817 * extending the file past EOF and this data starts
818 * block-aligned, use assign_arcbuf(). Otherwise,
819 * write via dmu_write().
821 if (tx_bytes
< max_blksz
&& (!write_eof
||
822 aiov
->iov_base
!= abuf
->b_data
)) {
824 dmu_write(zsb
->z_os
, zp
->z_id
, woff
,
825 aiov
->iov_len
, aiov
->iov_base
, tx
);
826 dmu_return_arcbuf(abuf
);
827 xuio_stat_wbuf_copied();
829 ASSERT(xuio
|| tx_bytes
== max_blksz
);
830 dmu_assign_arcbuf(sa_get_db(zp
->z_sa_hdl
),
833 ASSERT(tx_bytes
<= uio
->uio_resid
);
834 uioskip(uio
, tx_bytes
);
837 if (tx_bytes
&& zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
))
838 update_pages(ip
, woff
, tx_bytes
, zsb
->z_os
, zp
->z_id
);
841 * If we made no progress, we're done. If we made even
842 * partial progress, update the znode and ZIL accordingly.
845 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
846 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
853 * Clear Set-UID/Set-GID bits on successful write if not
854 * privileged and at least one of the excute bits is set.
856 * It would be nice to to this after all writes have
857 * been done, but that would still expose the ISUID/ISGID
858 * to another app after the partial write is committed.
860 * Note: we don't call zfs_fuid_map_id() here because
861 * user 0 is not an ephemeral uid.
863 mutex_enter(&zp
->z_acl_lock
);
864 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
865 (S_IXUSR
>> 6))) != 0 &&
866 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
867 secpolicy_vnode_setid_retain(cr
,
868 (zp
->z_mode
& S_ISUID
) != 0 && zp
->z_uid
== 0) != 0) {
870 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
871 newmode
= zp
->z_mode
;
872 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zsb
),
873 (void *)&newmode
, sizeof (uint64_t), tx
);
875 mutex_exit(&zp
->z_acl_lock
);
877 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
881 * Update the file size (zp_size) if it has changed;
882 * account for possible concurrent updates.
884 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
885 (void) atomic_cas_64(&zp
->z_size
, end_size
,
890 * If we are replaying and eof is non zero then force
891 * the file size to the specified eof. Note, there's no
892 * concurrency during replay.
894 if (zsb
->z_replay
&& zsb
->z_replay_eof
!= 0)
895 zp
->z_size
= zsb
->z_replay_eof
;
897 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
899 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
,
905 ASSERT(tx_bytes
== nbytes
);
909 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
912 zfs_inode_update(zp
);
913 zfs_range_unlock(rl
);
916 * If we're in replay mode, or we made no progress, return error.
917 * Otherwise, it's at least a partial write, so it's successful.
919 if (zsb
->z_replay
|| uio
->uio_resid
== start_resid
) {
924 if (ioflag
& (FSYNC
| FDSYNC
) ||
925 zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
926 zil_commit(zilog
, zp
->z_id
);
931 EXPORT_SYMBOL(zfs_write
);
934 zfs_iput_async(struct inode
*ip
)
936 objset_t
*os
= ITOZSB(ip
)->z_os
;
938 ASSERT(atomic_read(&ip
->i_count
) > 0);
941 if (atomic_read(&ip
->i_count
) == 1)
942 taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os
)),
943 (task_func_t
*)iput
, ip
, TQ_SLEEP
);
949 zfs_get_done(zgd_t
*zgd
, int error
)
951 znode_t
*zp
= zgd
->zgd_private
;
954 dmu_buf_rele(zgd
->zgd_db
, zgd
);
956 zfs_range_unlock(zgd
->zgd_rl
);
959 * Release the vnode asynchronously as we currently have the
960 * txg stopped from syncing.
962 zfs_iput_async(ZTOI(zp
));
964 if (error
== 0 && zgd
->zgd_bp
)
965 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
967 kmem_free(zgd
, sizeof (zgd_t
));
971 static int zil_fault_io
= 0;
975 * Get data to generate a TX_WRITE intent log record.
978 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
981 objset_t
*os
= zsb
->z_os
;
983 uint64_t object
= lr
->lr_foid
;
984 uint64_t offset
= lr
->lr_offset
;
985 uint64_t size
= lr
->lr_length
;
986 blkptr_t
*bp
= &lr
->lr_blkptr
;
995 * Nothing to do if the file has been removed
997 if (zfs_zget(zsb
, object
, &zp
) != 0)
998 return (SET_ERROR(ENOENT
));
999 if (zp
->z_unlinked
) {
1001 * Release the vnode asynchronously as we currently have the
1002 * txg stopped from syncing.
1004 zfs_iput_async(ZTOI(zp
));
1005 return (SET_ERROR(ENOENT
));
1008 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1009 zgd
->zgd_zilog
= zsb
->z_log
;
1010 zgd
->zgd_private
= zp
;
1013 * Write records come in two flavors: immediate and indirect.
1014 * For small writes it's cheaper to store the data with the
1015 * log record (immediate); for large writes it's cheaper to
1016 * sync the data and get a pointer to it (indirect) so that
1017 * we don't have to write the data twice.
1019 if (buf
!= NULL
) { /* immediate write */
1020 zgd
->zgd_rl
= zfs_range_lock(zp
, offset
, size
, RL_READER
);
1021 /* test for truncation needs to be done while range locked */
1022 if (offset
>= zp
->z_size
) {
1023 error
= SET_ERROR(ENOENT
);
1025 error
= dmu_read(os
, object
, offset
, size
, buf
,
1026 DMU_READ_NO_PREFETCH
);
1028 ASSERT(error
== 0 || error
== ENOENT
);
1029 } else { /* indirect write */
1031 * Have to lock the whole block to ensure when it's
1032 * written out and it's checksum is being calculated
1033 * that no one can change the data. We need to re-check
1034 * blocksize after we get the lock in case it's changed!
1039 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1041 zgd
->zgd_rl
= zfs_range_lock(zp
, offset
, size
,
1043 if (zp
->z_blksz
== size
)
1046 zfs_range_unlock(zgd
->zgd_rl
);
1048 /* test for truncation needs to be done while range locked */
1049 if (lr
->lr_offset
>= zp
->z_size
)
1050 error
= SET_ERROR(ENOENT
);
1053 error
= SET_ERROR(EIO
);
1058 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1059 DMU_READ_NO_PREFETCH
);
1062 blkptr_t
*obp
= dmu_buf_get_blkptr(db
);
1064 ASSERT(BP_IS_HOLE(bp
));
1071 ASSERT(db
->db_offset
== offset
);
1072 ASSERT(db
->db_size
== size
);
1074 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1076 ASSERT(error
|| lr
->lr_length
<= zp
->z_blksz
);
1079 * On success, we need to wait for the write I/O
1080 * initiated by dmu_sync() to complete before we can
1081 * release this dbuf. We will finish everything up
1082 * in the zfs_get_done() callback.
1087 if (error
== EALREADY
) {
1088 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
1094 zfs_get_done(zgd
, error
);
1101 zfs_access(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
1103 znode_t
*zp
= ITOZ(ip
);
1104 zfs_sb_t
*zsb
= ITOZSB(ip
);
1110 if (flag
& V_ACE_MASK
)
1111 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1113 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1118 EXPORT_SYMBOL(zfs_access
);
1121 * Lookup an entry in a directory, or an extended attribute directory.
1122 * If it exists, return a held inode reference for it.
1124 * IN: dip - inode of directory to search.
1125 * nm - name of entry to lookup.
1126 * flags - LOOKUP_XATTR set if looking for an attribute.
1127 * cr - credentials of caller.
1128 * direntflags - directory lookup flags
1129 * realpnp - returned pathname.
1131 * OUT: ipp - inode of located entry, NULL if not found.
1133 * RETURN: 0 on success, error code on failure.
1140 zfs_lookup(struct inode
*dip
, char *nm
, struct inode
**ipp
, int flags
,
1141 cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1143 znode_t
*zdp
= ITOZ(dip
);
1144 zfs_sb_t
*zsb
= ITOZSB(dip
);
1148 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1150 if (!S_ISDIR(dip
->i_mode
)) {
1151 return (SET_ERROR(ENOTDIR
));
1152 } else if (zdp
->z_sa_hdl
== NULL
) {
1153 return (SET_ERROR(EIO
));
1156 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1157 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1166 vnode_t
*tvp
= dnlc_lookup(dvp
, nm
);
1169 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1174 if (tvp
== DNLC_NO_VNODE
) {
1176 return (SET_ERROR(ENOENT
));
1179 return (specvp_check(vpp
, cr
));
1182 #endif /* HAVE_DNLC */
1191 if (flags
& LOOKUP_XATTR
) {
1193 * We don't allow recursive attributes..
1194 * Maybe someday we will.
1196 if (zdp
->z_pflags
& ZFS_XATTR
) {
1198 return (SET_ERROR(EINVAL
));
1201 if ((error
= zfs_get_xattrdir(zdp
, ipp
, cr
, flags
))) {
1207 * Do we have permission to get into attribute directory?
1210 if ((error
= zfs_zaccess(ITOZ(*ipp
), ACE_EXECUTE
, 0,
1220 if (!S_ISDIR(dip
->i_mode
)) {
1222 return (SET_ERROR(ENOTDIR
));
1226 * Check accessibility of directory.
1229 if ((error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
))) {
1234 if (zsb
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1235 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1237 return (SET_ERROR(EILSEQ
));
1240 error
= zfs_dirlook(zdp
, nm
, ipp
, flags
, direntflags
, realpnp
);
1241 if ((error
== 0) && (*ipp
))
1242 zfs_inode_update(ITOZ(*ipp
));
1247 EXPORT_SYMBOL(zfs_lookup
);
1250 * Attempt to create a new entry in a directory. If the entry
1251 * already exists, truncate the file if permissible, else return
1252 * an error. Return the ip of the created or trunc'd file.
1254 * IN: dip - inode of directory to put new file entry in.
1255 * name - name of new file entry.
1256 * vap - attributes of new file.
1257 * excl - flag indicating exclusive or non-exclusive mode.
1258 * mode - mode to open file with.
1259 * cr - credentials of caller.
1260 * flag - large file flag [UNUSED].
1261 * vsecp - ACL to be set
1263 * OUT: ipp - inode of created or trunc'd entry.
1265 * RETURN: 0 on success, error code on failure.
1268 * dip - ctime|mtime updated if new entry created
1269 * ip - ctime|mtime always, atime if new
1274 zfs_create(struct inode
*dip
, char *name
, vattr_t
*vap
, int excl
,
1275 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1277 znode_t
*zp
, *dzp
= ITOZ(dip
);
1278 zfs_sb_t
*zsb
= ITOZSB(dip
);
1286 zfs_acl_ids_t acl_ids
;
1287 boolean_t fuid_dirtied
;
1288 boolean_t have_acl
= B_FALSE
;
1289 boolean_t waited
= B_FALSE
;
1292 * If we have an ephemeral id, ACL, or XVATTR then
1293 * make sure file system is at proper version
1299 if (zsb
->z_use_fuids
== B_FALSE
&&
1300 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1301 return (SET_ERROR(EINVAL
));
1308 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
1309 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1311 return (SET_ERROR(EILSEQ
));
1314 if (vap
->va_mask
& ATTR_XVATTR
) {
1315 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1316 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1324 if (*name
== '\0') {
1326 * Null component name refers to the directory itself.
1333 /* possible igrab(zp) */
1336 if (flag
& FIGNORECASE
)
1339 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1343 zfs_acl_ids_free(&acl_ids
);
1344 if (strcmp(name
, "..") == 0)
1345 error
= SET_ERROR(EISDIR
);
1355 * Create a new file object and update the directory
1358 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1360 zfs_acl_ids_free(&acl_ids
);
1365 * We only support the creation of regular files in
1366 * extended attribute directories.
1369 if ((dzp
->z_pflags
& ZFS_XATTR
) && !S_ISREG(vap
->va_mode
)) {
1371 zfs_acl_ids_free(&acl_ids
);
1372 error
= SET_ERROR(EINVAL
);
1376 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1377 cr
, vsecp
, &acl_ids
)) != 0)
1381 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1382 zfs_acl_ids_free(&acl_ids
);
1383 error
= SET_ERROR(EDQUOT
);
1387 tx
= dmu_tx_create(os
);
1389 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1390 ZFS_SA_BASE_ATTR_SIZE
);
1392 fuid_dirtied
= zsb
->z_fuid_dirty
;
1394 zfs_fuid_txhold(zsb
, tx
);
1395 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1396 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1397 if (!zsb
->z_use_sa
&&
1398 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1399 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1400 0, acl_ids
.z_aclp
->z_acl_bytes
);
1402 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1404 zfs_dirent_unlock(dl
);
1405 if (error
== ERESTART
) {
1411 zfs_acl_ids_free(&acl_ids
);
1416 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1419 zfs_fuid_sync(zsb
, tx
);
1421 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1422 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1423 if (flag
& FIGNORECASE
)
1425 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1426 vsecp
, acl_ids
.z_fuidp
, vap
);
1427 zfs_acl_ids_free(&acl_ids
);
1430 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1433 zfs_acl_ids_free(&acl_ids
);
1437 * A directory entry already exists for this name.
1440 * Can't truncate an existing file if in exclusive mode.
1443 error
= SET_ERROR(EEXIST
);
1447 * Can't open a directory for writing.
1449 if (S_ISDIR(ZTOI(zp
)->i_mode
)) {
1450 error
= SET_ERROR(EISDIR
);
1454 * Verify requested access to file.
1456 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1460 mutex_enter(&dzp
->z_lock
);
1462 mutex_exit(&dzp
->z_lock
);
1465 * Truncate regular files if requested.
1467 if (S_ISREG(ZTOI(zp
)->i_mode
) &&
1468 (vap
->va_mask
& ATTR_SIZE
) && (vap
->va_size
== 0)) {
1469 /* we can't hold any locks when calling zfs_freesp() */
1470 zfs_dirent_unlock(dl
);
1472 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1478 zfs_dirent_unlock(dl
);
1484 zfs_inode_update(dzp
);
1485 zfs_inode_update(zp
);
1489 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1490 zil_commit(zilog
, 0);
1495 EXPORT_SYMBOL(zfs_create
);
1498 * Remove an entry from a directory.
1500 * IN: dip - inode of directory to remove entry from.
1501 * name - name of entry to remove.
1502 * cr - credentials of caller.
1504 * RETURN: 0 if success
1505 * error code if failure
1509 * ip - ctime (if nlink > 0)
1512 uint64_t null_xattr
= 0;
1516 zfs_remove(struct inode
*dip
, char *name
, cred_t
*cr
)
1518 znode_t
*zp
, *dzp
= ITOZ(dip
);
1521 zfs_sb_t
*zsb
= ITOZSB(dip
);
1524 uint64_t xattr_obj_unlinked
= 0;
1530 pathname_t
*realnmp
= NULL
;
1531 #ifdef HAVE_PN_UTILS
1533 #endif /* HAVE_PN_UTILS */
1536 boolean_t waited
= B_FALSE
;
1542 #ifdef HAVE_PN_UTILS
1543 if (flags
& FIGNORECASE
) {
1548 #endif /* HAVE_PN_UTILS */
1554 * Attempt to lock directory; fail if entry doesn't exist.
1556 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1558 #ifdef HAVE_PN_UTILS
1561 #endif /* HAVE_PN_UTILS */
1568 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1573 * Need to use rmdir for removing directories.
1575 if (S_ISDIR(ip
->i_mode
)) {
1576 error
= SET_ERROR(EPERM
);
1582 dnlc_remove(dvp
, realnmp
->pn_buf
);
1584 dnlc_remove(dvp
, name
);
1585 #endif /* HAVE_DNLC */
1588 * We never delete the znode and always place it in the unlinked
1589 * set. The dentry cache will always hold the last reference and
1590 * is responsible for safely freeing the znode.
1593 tx
= dmu_tx_create(zsb
->z_os
);
1594 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1595 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1596 zfs_sa_upgrade_txholds(tx
, zp
);
1597 zfs_sa_upgrade_txholds(tx
, dzp
);
1599 /* are there any extended attributes? */
1600 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1601 &xattr_obj
, sizeof (xattr_obj
));
1602 if (error
== 0 && xattr_obj
) {
1603 error
= zfs_zget(zsb
, xattr_obj
, &xzp
);
1605 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1606 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1609 /* charge as an update -- would be nice not to charge at all */
1610 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
1612 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1614 zfs_dirent_unlock(dl
);
1618 if (error
== ERESTART
) {
1624 #ifdef HAVE_PN_UTILS
1627 #endif /* HAVE_PN_UTILS */
1634 * Remove the directory entry.
1636 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1645 * Hold z_lock so that we can make sure that the ACL obj
1646 * hasn't changed. Could have been deleted due to
1649 mutex_enter(&zp
->z_lock
);
1650 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1651 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1652 mutex_exit(&zp
->z_lock
);
1653 zfs_unlinked_add(zp
, tx
);
1657 #ifdef HAVE_PN_UTILS
1658 if (flags
& FIGNORECASE
)
1660 #endif /* HAVE_PN_UTILS */
1661 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1665 #ifdef HAVE_PN_UTILS
1668 #endif /* HAVE_PN_UTILS */
1670 zfs_dirent_unlock(dl
);
1671 zfs_inode_update(dzp
);
1672 zfs_inode_update(zp
);
1674 zfs_inode_update(xzp
);
1680 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1681 zil_commit(zilog
, 0);
1686 EXPORT_SYMBOL(zfs_remove
);
1689 * Create a new directory and insert it into dip using the name
1690 * provided. Return a pointer to the inserted directory.
1692 * IN: dip - inode of directory to add subdir to.
1693 * dirname - name of new directory.
1694 * vap - attributes of new directory.
1695 * cr - credentials of caller.
1696 * vsecp - ACL to be set
1698 * OUT: ipp - inode of created directory.
1700 * RETURN: 0 if success
1701 * error code if failure
1704 * dip - ctime|mtime updated
1705 * ipp - ctime|mtime|atime updated
1709 zfs_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
, struct inode
**ipp
,
1710 cred_t
*cr
, int flags
, vsecattr_t
*vsecp
)
1712 znode_t
*zp
, *dzp
= ITOZ(dip
);
1713 zfs_sb_t
*zsb
= ITOZSB(dip
);
1721 gid_t gid
= crgetgid(cr
);
1722 zfs_acl_ids_t acl_ids
;
1723 boolean_t fuid_dirtied
;
1724 boolean_t waited
= B_FALSE
;
1726 ASSERT(S_ISDIR(vap
->va_mode
));
1729 * If we have an ephemeral id, ACL, or XVATTR then
1730 * make sure file system is at proper version
1734 if (zsb
->z_use_fuids
== B_FALSE
&&
1735 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1736 return (SET_ERROR(EINVAL
));
1742 if (dzp
->z_pflags
& ZFS_XATTR
) {
1744 return (SET_ERROR(EINVAL
));
1747 if (zsb
->z_utf8
&& u8_validate(dirname
,
1748 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1750 return (SET_ERROR(EILSEQ
));
1752 if (flags
& FIGNORECASE
)
1755 if (vap
->va_mask
& ATTR_XVATTR
) {
1756 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1757 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1763 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1764 vsecp
, &acl_ids
)) != 0) {
1769 * First make sure the new directory doesn't exist.
1771 * Existence is checked first to make sure we don't return
1772 * EACCES instead of EEXIST which can cause some applications
1778 if ((error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1780 zfs_acl_ids_free(&acl_ids
);
1785 if ((error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
))) {
1786 zfs_acl_ids_free(&acl_ids
);
1787 zfs_dirent_unlock(dl
);
1792 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1793 zfs_acl_ids_free(&acl_ids
);
1794 zfs_dirent_unlock(dl
);
1796 return (SET_ERROR(EDQUOT
));
1800 * Add a new entry to the directory.
1802 tx
= dmu_tx_create(zsb
->z_os
);
1803 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
1804 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1805 fuid_dirtied
= zsb
->z_fuid_dirty
;
1807 zfs_fuid_txhold(zsb
, tx
);
1808 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1809 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
1810 acl_ids
.z_aclp
->z_acl_bytes
);
1813 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1814 ZFS_SA_BASE_ATTR_SIZE
);
1816 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1818 zfs_dirent_unlock(dl
);
1819 if (error
== ERESTART
) {
1825 zfs_acl_ids_free(&acl_ids
);
1834 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1837 zfs_fuid_sync(zsb
, tx
);
1840 * Now put new name in parent dir.
1842 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1846 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
1847 if (flags
& FIGNORECASE
)
1849 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
1850 acl_ids
.z_fuidp
, vap
);
1852 zfs_acl_ids_free(&acl_ids
);
1856 zfs_dirent_unlock(dl
);
1858 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1859 zil_commit(zilog
, 0);
1861 zfs_inode_update(dzp
);
1862 zfs_inode_update(zp
);
1866 EXPORT_SYMBOL(zfs_mkdir
);
1869 * Remove a directory subdir entry. If the current working
1870 * directory is the same as the subdir to be removed, the
1873 * IN: dip - inode of directory to remove from.
1874 * name - name of directory to be removed.
1875 * cwd - inode of current working directory.
1876 * cr - credentials of caller.
1877 * flags - case flags
1879 * RETURN: 0 on success, error code on failure.
1882 * dip - ctime|mtime updated
1886 zfs_rmdir(struct inode
*dip
, char *name
, struct inode
*cwd
, cred_t
*cr
,
1889 znode_t
*dzp
= ITOZ(dip
);
1892 zfs_sb_t
*zsb
= ITOZSB(dip
);
1898 boolean_t waited
= B_FALSE
;
1904 if (flags
& FIGNORECASE
)
1910 * Attempt to lock directory; fail if entry doesn't exist.
1912 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1920 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1924 if (!S_ISDIR(ip
->i_mode
)) {
1925 error
= SET_ERROR(ENOTDIR
);
1930 error
= SET_ERROR(EINVAL
);
1935 * Grab a lock on the directory to make sure that noone is
1936 * trying to add (or lookup) entries while we are removing it.
1938 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
1941 * Grab a lock on the parent pointer to make sure we play well
1942 * with the treewalk and directory rename code.
1944 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
1946 tx
= dmu_tx_create(zsb
->z_os
);
1947 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1948 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1949 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
1950 zfs_sa_upgrade_txholds(tx
, zp
);
1951 zfs_sa_upgrade_txholds(tx
, dzp
);
1952 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1954 rw_exit(&zp
->z_parent_lock
);
1955 rw_exit(&zp
->z_name_lock
);
1956 zfs_dirent_unlock(dl
);
1958 if (error
== ERESTART
) {
1969 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
1972 uint64_t txtype
= TX_RMDIR
;
1973 if (flags
& FIGNORECASE
)
1975 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
1980 rw_exit(&zp
->z_parent_lock
);
1981 rw_exit(&zp
->z_name_lock
);
1983 zfs_dirent_unlock(dl
);
1985 zfs_inode_update(dzp
);
1986 zfs_inode_update(zp
);
1989 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1990 zil_commit(zilog
, 0);
1995 EXPORT_SYMBOL(zfs_rmdir
);
1998 * Read as many directory entries as will fit into the provided
1999 * dirent buffer from the given directory cursor position.
2001 * IN: ip - inode of directory to read.
2002 * dirent - buffer for directory entries.
2004 * OUT: dirent - filler buffer of directory entries.
2006 * RETURN: 0 if success
2007 * error code if failure
2010 * ip - atime updated
2012 * Note that the low 4 bits of the cookie returned by zap is always zero.
2013 * This allows us to use the low range for "special" directory entries:
2014 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2015 * we use the offset 2 for the '.zfs' directory.
2019 zfs_readdir(struct inode
*ip
, struct dir_context
*ctx
, cred_t
*cr
)
2021 znode_t
*zp
= ITOZ(ip
);
2022 zfs_sb_t
*zsb
= ITOZSB(ip
);
2025 zap_attribute_t zap
;
2031 uint64_t offset
; /* must be unsigned; checks for < 1 */
2036 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
2037 &parent
, sizeof (parent
))) != 0)
2041 * Quit if directory has been removed (posix)
2049 prefetch
= zp
->z_zn_prefetch
;
2052 * Initialize the iterator cursor.
2056 * Start iteration from the beginning of the directory.
2058 zap_cursor_init(&zc
, os
, zp
->z_id
);
2061 * The offset is a serialized cursor.
2063 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2067 * Transform to file-system independent format
2072 * Special case `.', `..', and `.zfs'.
2075 (void) strcpy(zap
.za_name
, ".");
2076 zap
.za_normalization_conflict
= 0;
2079 } else if (offset
== 1) {
2080 (void) strcpy(zap
.za_name
, "..");
2081 zap
.za_normalization_conflict
= 0;
2084 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2085 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2086 zap
.za_normalization_conflict
= 0;
2087 objnum
= ZFSCTL_INO_ROOT
;
2093 if ((error
= zap_cursor_retrieve(&zc
, &zap
))) {
2094 if (error
== ENOENT
)
2101 * Allow multiple entries provided the first entry is
2102 * the object id. Non-zpl consumers may safely make
2103 * use of the additional space.
2105 * XXX: This should be a feature flag for compatibility
2107 if (zap
.za_integer_length
!= 8 ||
2108 zap
.za_num_integers
== 0) {
2109 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2110 "entry, obj = %lld, offset = %lld, "
2111 "length = %d, num = %lld\n",
2112 (u_longlong_t
)zp
->z_id
,
2113 (u_longlong_t
)offset
,
2114 zap
.za_integer_length
,
2115 (u_longlong_t
)zap
.za_num_integers
);
2116 error
= SET_ERROR(ENXIO
);
2120 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2121 type
= ZFS_DIRENT_TYPE(zap
.za_first_integer
);
2124 done
= !dir_emit(ctx
, zap
.za_name
, strlen(zap
.za_name
),
2129 /* Prefetch znode */
2131 dmu_prefetch(os
, objnum
, 0, 0, 0,
2132 ZIO_PRIORITY_SYNC_READ
);
2136 * Move to the next entry, fill in the previous offset.
2138 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2139 zap_cursor_advance(&zc
);
2140 offset
= zap_cursor_serialize(&zc
);
2146 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2149 zap_cursor_fini(&zc
);
2150 if (error
== ENOENT
)
2153 ZFS_ACCESSTIME_STAMP(zsb
, zp
);
2160 EXPORT_SYMBOL(zfs_readdir
);
2162 ulong_t zfs_fsync_sync_cnt
= 4;
2165 zfs_fsync(struct inode
*ip
, int syncflag
, cred_t
*cr
)
2167 znode_t
*zp
= ITOZ(ip
);
2168 zfs_sb_t
*zsb
= ITOZSB(ip
);
2170 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2172 if (zsb
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2175 zil_commit(zsb
->z_log
, zp
->z_id
);
2178 tsd_set(zfs_fsyncer_key
, NULL
);
2182 EXPORT_SYMBOL(zfs_fsync
);
2186 * Get the requested file attributes and place them in the provided
2189 * IN: ip - inode of file.
2190 * vap - va_mask identifies requested attributes.
2191 * If ATTR_XVATTR set, then optional attrs are requested
2192 * flags - ATTR_NOACLCHECK (CIFS server context)
2193 * cr - credentials of caller.
2195 * OUT: vap - attribute values.
2197 * RETURN: 0 (always succeeds)
2201 zfs_getattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2203 znode_t
*zp
= ITOZ(ip
);
2204 zfs_sb_t
*zsb
= ITOZSB(ip
);
2207 uint64_t mtime
[2], ctime
[2];
2208 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2209 xoptattr_t
*xoap
= NULL
;
2210 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2211 sa_bulk_attr_t bulk
[2];
2217 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2219 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
2220 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
2222 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2228 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2229 * Also, if we are the owner don't bother, since owner should
2230 * always be allowed to read basic attributes of file.
2232 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2233 (vap
->va_uid
!= crgetuid(cr
))) {
2234 if ((error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2242 * Return all attributes. It's cheaper to provide the answer
2243 * than to determine whether we were asked the question.
2246 mutex_enter(&zp
->z_lock
);
2247 vap
->va_type
= vn_mode_to_vtype(zp
->z_mode
);
2248 vap
->va_mode
= zp
->z_mode
;
2249 vap
->va_fsid
= ZTOI(zp
)->i_sb
->s_dev
;
2250 vap
->va_nodeid
= zp
->z_id
;
2251 if ((zp
->z_id
== zsb
->z_root
) && zfs_show_ctldir(zp
))
2252 links
= zp
->z_links
+ 1;
2254 links
= zp
->z_links
;
2255 vap
->va_nlink
= MIN(links
, ZFS_LINK_MAX
);
2256 vap
->va_size
= i_size_read(ip
);
2257 vap
->va_rdev
= ip
->i_rdev
;
2258 vap
->va_seq
= ip
->i_generation
;
2261 * Add in any requested optional attributes and the create time.
2262 * Also set the corresponding bits in the returned attribute bitmap.
2264 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zsb
->z_use_fuids
) {
2265 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2267 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2268 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2271 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2272 xoap
->xoa_readonly
=
2273 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2274 XVA_SET_RTN(xvap
, XAT_READONLY
);
2277 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2279 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2280 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2283 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2285 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2286 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2289 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2290 xoap
->xoa_nounlink
=
2291 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2292 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2295 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2296 xoap
->xoa_immutable
=
2297 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2298 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2301 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2302 xoap
->xoa_appendonly
=
2303 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2304 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2307 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2309 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2310 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2313 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2315 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2316 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2319 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2320 xoap
->xoa_av_quarantined
=
2321 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2322 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2325 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2326 xoap
->xoa_av_modified
=
2327 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2328 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2331 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2332 S_ISREG(ip
->i_mode
)) {
2333 zfs_sa_get_scanstamp(zp
, xvap
);
2336 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2339 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zsb
),
2340 times
, sizeof (times
));
2341 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2342 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2345 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2346 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2347 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2349 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2350 xoap
->xoa_generation
= zp
->z_gen
;
2351 XVA_SET_RTN(xvap
, XAT_GEN
);
2354 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2356 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2357 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2360 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2362 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2363 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2367 ZFS_TIME_DECODE(&vap
->va_atime
, zp
->z_atime
);
2368 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2369 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2371 mutex_exit(&zp
->z_lock
);
2373 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2375 if (zp
->z_blksz
== 0) {
2377 * Block size hasn't been set; suggest maximal I/O transfers.
2379 vap
->va_blksize
= zsb
->z_max_blksz
;
2385 EXPORT_SYMBOL(zfs_getattr
);
2388 * Get the basic file attributes and place them in the provided kstat
2389 * structure. The inode is assumed to be the authoritative source
2390 * for most of the attributes. However, the znode currently has the
2391 * authoritative atime, blksize, and block count.
2393 * IN: ip - inode of file.
2395 * OUT: sp - kstat values.
2397 * RETURN: 0 (always succeeds)
2401 zfs_getattr_fast(struct inode
*ip
, struct kstat
*sp
)
2403 znode_t
*zp
= ITOZ(ip
);
2404 zfs_sb_t
*zsb
= ITOZSB(ip
);
2406 u_longlong_t nblocks
;
2411 mutex_enter(&zp
->z_lock
);
2413 generic_fillattr(ip
, sp
);
2414 ZFS_TIME_DECODE(&sp
->atime
, zp
->z_atime
);
2416 sa_object_size(zp
->z_sa_hdl
, &blksize
, &nblocks
);
2417 sp
->blksize
= blksize
;
2418 sp
->blocks
= nblocks
;
2420 if (unlikely(zp
->z_blksz
== 0)) {
2422 * Block size hasn't been set; suggest maximal I/O transfers.
2424 sp
->blksize
= zsb
->z_max_blksz
;
2427 mutex_exit(&zp
->z_lock
);
2430 * Required to prevent NFS client from detecting different inode
2431 * numbers of snapshot root dentry before and after snapshot mount.
2433 if (zsb
->z_issnap
) {
2434 if (ip
->i_sb
->s_root
->d_inode
== ip
)
2435 sp
->ino
= ZFSCTL_INO_SNAPDIRS
-
2436 dmu_objset_id(zsb
->z_os
);
2443 EXPORT_SYMBOL(zfs_getattr_fast
);
2446 * Set the file attributes to the values contained in the
2449 * IN: ip - inode of file to be modified.
2450 * vap - new attribute values.
2451 * If ATTR_XVATTR set, then optional attrs are being set
2452 * flags - ATTR_UTIME set if non-default time values provided.
2453 * - ATTR_NOACLCHECK (CIFS context only).
2454 * cr - credentials of caller.
2456 * RETURN: 0 if success
2457 * error code if failure
2460 * ip - ctime updated, mtime updated if size changed.
2464 zfs_setattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2466 znode_t
*zp
= ITOZ(ip
);
2467 zfs_sb_t
*zsb
= ITOZSB(ip
);
2471 xvattr_t
*tmpxvattr
;
2472 uint_t mask
= vap
->va_mask
;
2473 uint_t saved_mask
= 0;
2476 uint64_t new_uid
, new_gid
;
2478 uint64_t mtime
[2], ctime
[2];
2480 int need_policy
= FALSE
;
2482 zfs_fuid_info_t
*fuidp
= NULL
;
2483 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2486 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2487 boolean_t fuid_dirtied
= B_FALSE
;
2488 sa_bulk_attr_t
*bulk
, *xattr_bulk
;
2489 int count
= 0, xattr_count
= 0;
2500 * Make sure that if we have ephemeral uid/gid or xvattr specified
2501 * that file system is at proper version level
2504 if (zsb
->z_use_fuids
== B_FALSE
&&
2505 (((mask
& ATTR_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2506 ((mask
& ATTR_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2507 (mask
& ATTR_XVATTR
))) {
2509 return (SET_ERROR(EINVAL
));
2512 if (mask
& ATTR_SIZE
&& S_ISDIR(ip
->i_mode
)) {
2514 return (SET_ERROR(EISDIR
));
2517 if (mask
& ATTR_SIZE
&& !S_ISREG(ip
->i_mode
) && !S_ISFIFO(ip
->i_mode
)) {
2519 return (SET_ERROR(EINVAL
));
2523 * If this is an xvattr_t, then get a pointer to the structure of
2524 * optional attributes. If this is NULL, then we have a vattr_t.
2526 xoap
= xva_getxoptattr(xvap
);
2528 tmpxvattr
= kmem_alloc(sizeof (xvattr_t
), KM_SLEEP
);
2529 xva_init(tmpxvattr
);
2531 bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2532 xattr_bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2535 * Immutable files can only alter immutable bit and atime
2537 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2538 ((mask
& (ATTR_SIZE
|ATTR_UID
|ATTR_GID
|ATTR_MTIME
|ATTR_MODE
)) ||
2539 ((mask
& ATTR_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2544 if ((mask
& ATTR_SIZE
) && (zp
->z_pflags
& ZFS_READONLY
)) {
2550 * Verify timestamps doesn't overflow 32 bits.
2551 * ZFS can handle large timestamps, but 32bit syscalls can't
2552 * handle times greater than 2039. This check should be removed
2553 * once large timestamps are fully supported.
2555 if (mask
& (ATTR_ATIME
| ATTR_MTIME
)) {
2556 if (((mask
& ATTR_ATIME
) &&
2557 TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2558 ((mask
& ATTR_MTIME
) &&
2559 TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2569 /* Can this be moved to before the top label? */
2570 if (zfs_is_readonly(zsb
)) {
2576 * First validate permissions
2579 if (mask
& ATTR_SIZE
) {
2580 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2585 * XXX - Note, we are not providing any open
2586 * mode flags here (like FNDELAY), so we may
2587 * block if there are locks present... this
2588 * should be addressed in openat().
2590 /* XXX - would it be OK to generate a log record here? */
2591 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2596 if (mask
& (ATTR_ATIME
|ATTR_MTIME
) ||
2597 ((mask
& ATTR_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2598 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2599 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2600 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2601 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2602 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2603 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2604 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2608 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2609 int idmask
= (mask
& (ATTR_UID
|ATTR_GID
));
2614 * NOTE: even if a new mode is being set,
2615 * we may clear S_ISUID/S_ISGID bits.
2618 if (!(mask
& ATTR_MODE
))
2619 vap
->va_mode
= zp
->z_mode
;
2622 * Take ownership or chgrp to group we are a member of
2625 take_owner
= (mask
& ATTR_UID
) && (vap
->va_uid
== crgetuid(cr
));
2626 take_group
= (mask
& ATTR_GID
) &&
2627 zfs_groupmember(zsb
, vap
->va_gid
, cr
);
2630 * If both ATTR_UID and ATTR_GID are set then take_owner and
2631 * take_group must both be set in order to allow taking
2634 * Otherwise, send the check through secpolicy_vnode_setattr()
2638 if (((idmask
== (ATTR_UID
|ATTR_GID
)) &&
2639 take_owner
&& take_group
) ||
2640 ((idmask
== ATTR_UID
) && take_owner
) ||
2641 ((idmask
== ATTR_GID
) && take_group
)) {
2642 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2643 skipaclchk
, cr
) == 0) {
2645 * Remove setuid/setgid for non-privileged users
2647 (void) secpolicy_setid_clear(vap
, cr
);
2648 trim_mask
= (mask
& (ATTR_UID
|ATTR_GID
));
2657 mutex_enter(&zp
->z_lock
);
2658 oldva
.va_mode
= zp
->z_mode
;
2659 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2660 if (mask
& ATTR_XVATTR
) {
2662 * Update xvattr mask to include only those attributes
2663 * that are actually changing.
2665 * the bits will be restored prior to actually setting
2666 * the attributes so the caller thinks they were set.
2668 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2669 if (xoap
->xoa_appendonly
!=
2670 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2673 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2674 XVA_SET_REQ(tmpxvattr
, XAT_APPENDONLY
);
2678 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2679 if (xoap
->xoa_nounlink
!=
2680 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2683 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2684 XVA_SET_REQ(tmpxvattr
, XAT_NOUNLINK
);
2688 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2689 if (xoap
->xoa_immutable
!=
2690 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2693 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2694 XVA_SET_REQ(tmpxvattr
, XAT_IMMUTABLE
);
2698 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2699 if (xoap
->xoa_nodump
!=
2700 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2703 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2704 XVA_SET_REQ(tmpxvattr
, XAT_NODUMP
);
2708 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2709 if (xoap
->xoa_av_modified
!=
2710 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2713 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2714 XVA_SET_REQ(tmpxvattr
, XAT_AV_MODIFIED
);
2718 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2719 if ((!S_ISREG(ip
->i_mode
) &&
2720 xoap
->xoa_av_quarantined
) ||
2721 xoap
->xoa_av_quarantined
!=
2722 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2725 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2726 XVA_SET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
);
2730 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2731 mutex_exit(&zp
->z_lock
);
2736 if (need_policy
== FALSE
&&
2737 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
2738 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
2743 mutex_exit(&zp
->z_lock
);
2745 if (mask
& ATTR_MODE
) {
2746 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
2747 err
= secpolicy_setid_setsticky_clear(ip
, vap
,
2752 trim_mask
|= ATTR_MODE
;
2760 * If trim_mask is set then take ownership
2761 * has been granted or write_acl is present and user
2762 * has the ability to modify mode. In that case remove
2763 * UID|GID and or MODE from mask so that
2764 * secpolicy_vnode_setattr() doesn't revoke it.
2768 saved_mask
= vap
->va_mask
;
2769 vap
->va_mask
&= ~trim_mask
;
2771 err
= secpolicy_vnode_setattr(cr
, ip
, vap
, &oldva
, flags
,
2772 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
2777 vap
->va_mask
|= saved_mask
;
2781 * secpolicy_vnode_setattr, or take ownership may have
2784 mask
= vap
->va_mask
;
2786 if ((mask
& (ATTR_UID
| ATTR_GID
))) {
2787 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
2788 &xattr_obj
, sizeof (xattr_obj
));
2790 if (err
== 0 && xattr_obj
) {
2791 err
= zfs_zget(ZTOZSB(zp
), xattr_obj
, &attrzp
);
2795 if (mask
& ATTR_UID
) {
2796 new_uid
= zfs_fuid_create(zsb
,
2797 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
2798 if (new_uid
!= zp
->z_uid
&&
2799 zfs_fuid_overquota(zsb
, B_FALSE
, new_uid
)) {
2807 if (mask
& ATTR_GID
) {
2808 new_gid
= zfs_fuid_create(zsb
, (uint64_t)vap
->va_gid
,
2809 cr
, ZFS_GROUP
, &fuidp
);
2810 if (new_gid
!= zp
->z_gid
&&
2811 zfs_fuid_overquota(zsb
, B_TRUE
, new_gid
)) {
2819 tx
= dmu_tx_create(zsb
->z_os
);
2821 if (mask
& ATTR_MODE
) {
2822 uint64_t pmode
= zp
->z_mode
;
2824 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
2826 zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
);
2828 mutex_enter(&zp
->z_lock
);
2829 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
2831 * Are we upgrading ACL from old V0 format
2834 if (zsb
->z_version
>= ZPL_VERSION_FUID
&&
2835 zfs_znode_acl_version(zp
) ==
2836 ZFS_ACL_VERSION_INITIAL
) {
2837 dmu_tx_hold_free(tx
, acl_obj
, 0,
2839 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
2840 0, aclp
->z_acl_bytes
);
2842 dmu_tx_hold_write(tx
, acl_obj
, 0,
2845 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
2846 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
2847 0, aclp
->z_acl_bytes
);
2849 mutex_exit(&zp
->z_lock
);
2850 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
2852 if ((mask
& ATTR_XVATTR
) &&
2853 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
2854 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
2856 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2860 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
2863 fuid_dirtied
= zsb
->z_fuid_dirty
;
2865 zfs_fuid_txhold(zsb
, tx
);
2867 zfs_sa_upgrade_txholds(tx
, zp
);
2869 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2875 * Set each attribute requested.
2876 * We group settings according to the locks they need to acquire.
2878 * Note: you cannot set ctime directly, although it will be
2879 * updated as a side-effect of calling this function.
2883 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
2884 mutex_enter(&zp
->z_acl_lock
);
2885 mutex_enter(&zp
->z_lock
);
2887 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
2888 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
2891 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
2892 mutex_enter(&attrzp
->z_acl_lock
);
2893 mutex_enter(&attrzp
->z_lock
);
2894 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2895 SA_ZPL_FLAGS(zsb
), NULL
, &attrzp
->z_pflags
,
2896 sizeof (attrzp
->z_pflags
));
2899 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2901 if (mask
& ATTR_UID
) {
2902 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zsb
), NULL
,
2903 &new_uid
, sizeof (new_uid
));
2904 zp
->z_uid
= new_uid
;
2906 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2907 SA_ZPL_UID(zsb
), NULL
, &new_uid
,
2909 attrzp
->z_uid
= new_uid
;
2913 if (mask
& ATTR_GID
) {
2914 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zsb
),
2915 NULL
, &new_gid
, sizeof (new_gid
));
2916 zp
->z_gid
= new_gid
;
2918 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2919 SA_ZPL_GID(zsb
), NULL
, &new_gid
,
2921 attrzp
->z_gid
= new_gid
;
2924 if (!(mask
& ATTR_MODE
)) {
2925 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
),
2926 NULL
, &new_mode
, sizeof (new_mode
));
2927 new_mode
= zp
->z_mode
;
2929 err
= zfs_acl_chown_setattr(zp
);
2932 err
= zfs_acl_chown_setattr(attrzp
);
2937 if (mask
& ATTR_MODE
) {
2938 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
), NULL
,
2939 &new_mode
, sizeof (new_mode
));
2940 zp
->z_mode
= new_mode
;
2941 ASSERT3P(aclp
, !=, NULL
);
2942 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
2944 if (zp
->z_acl_cached
)
2945 zfs_acl_free(zp
->z_acl_cached
);
2946 zp
->z_acl_cached
= aclp
;
2951 if (mask
& ATTR_ATIME
) {
2952 ZFS_TIME_ENCODE(&vap
->va_atime
, zp
->z_atime
);
2953 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zsb
), NULL
,
2954 &zp
->z_atime
, sizeof (zp
->z_atime
));
2957 if (mask
& ATTR_MTIME
) {
2958 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
2959 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
,
2960 mtime
, sizeof (mtime
));
2963 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2964 if (mask
& ATTR_SIZE
&& !(mask
& ATTR_MTIME
)) {
2965 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
),
2966 NULL
, mtime
, sizeof (mtime
));
2967 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
,
2968 &ctime
, sizeof (ctime
));
2969 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
2971 } else if (mask
!= 0) {
2972 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
,
2973 &ctime
, sizeof (ctime
));
2974 zfs_tstamp_update_setup(zp
, STATE_CHANGED
, mtime
, ctime
,
2977 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2978 SA_ZPL_CTIME(zsb
), NULL
,
2979 &ctime
, sizeof (ctime
));
2980 zfs_tstamp_update_setup(attrzp
, STATE_CHANGED
,
2981 mtime
, ctime
, B_TRUE
);
2985 * Do this after setting timestamps to prevent timestamp
2986 * update from toggling bit
2989 if (xoap
&& (mask
& ATTR_XVATTR
)) {
2992 * restore trimmed off masks
2993 * so that return masks can be set for caller.
2996 if (XVA_ISSET_REQ(tmpxvattr
, XAT_APPENDONLY
)) {
2997 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
2999 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NOUNLINK
)) {
3000 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3002 if (XVA_ISSET_REQ(tmpxvattr
, XAT_IMMUTABLE
)) {
3003 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3005 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NODUMP
)) {
3006 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3008 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_MODIFIED
)) {
3009 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3011 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
)) {
3012 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3015 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3016 ASSERT(S_ISREG(ip
->i_mode
));
3018 zfs_xvattr_set(zp
, xvap
, tx
);
3022 zfs_fuid_sync(zsb
, tx
);
3025 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3027 mutex_exit(&zp
->z_lock
);
3028 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3029 mutex_exit(&zp
->z_acl_lock
);
3032 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3033 mutex_exit(&attrzp
->z_acl_lock
);
3034 mutex_exit(&attrzp
->z_lock
);
3037 if (err
== 0 && attrzp
) {
3038 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3049 zfs_fuid_info_free(fuidp
);
3055 if (err
== ERESTART
)
3058 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3060 zfs_inode_update(zp
);
3064 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3065 zil_commit(zilog
, 0);
3068 kmem_free(xattr_bulk
, sizeof (sa_bulk_attr_t
) * 7);
3069 kmem_free(bulk
, sizeof (sa_bulk_attr_t
) * 7);
3070 kmem_free(tmpxvattr
, sizeof (xvattr_t
));
3074 EXPORT_SYMBOL(zfs_setattr
);
3076 typedef struct zfs_zlock
{
3077 krwlock_t
*zl_rwlock
; /* lock we acquired */
3078 znode_t
*zl_znode
; /* znode we held */
3079 struct zfs_zlock
*zl_next
; /* next in list */
3083 * Drop locks and release vnodes that were held by zfs_rename_lock().
3086 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3090 while ((zl
= *zlpp
) != NULL
) {
3091 if (zl
->zl_znode
!= NULL
)
3092 iput(ZTOI(zl
->zl_znode
));
3093 rw_exit(zl
->zl_rwlock
);
3094 *zlpp
= zl
->zl_next
;
3095 kmem_free(zl
, sizeof (*zl
));
3100 * Search back through the directory tree, using the ".." entries.
3101 * Lock each directory in the chain to prevent concurrent renames.
3102 * Fail any attempt to move a directory into one of its own descendants.
3103 * XXX - z_parent_lock can overlap with map or grow locks
3106 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3110 uint64_t rootid
= ZTOZSB(zp
)->z_root
;
3111 uint64_t oidp
= zp
->z_id
;
3112 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3113 krw_t rw
= RW_WRITER
;
3116 * First pass write-locks szp and compares to zp->z_id.
3117 * Later passes read-lock zp and compare to zp->z_parent.
3120 if (!rw_tryenter(rwlp
, rw
)) {
3122 * Another thread is renaming in this path.
3123 * Note that if we are a WRITER, we don't have any
3124 * parent_locks held yet.
3126 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3128 * Drop our locks and restart
3130 zfs_rename_unlock(&zl
);
3134 rwlp
= &szp
->z_parent_lock
;
3139 * Wait for other thread to drop its locks
3145 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3146 zl
->zl_rwlock
= rwlp
;
3147 zl
->zl_znode
= NULL
;
3148 zl
->zl_next
= *zlpp
;
3151 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3152 return (SET_ERROR(EINVAL
));
3154 if (oidp
== rootid
) /* We've hit the top */
3157 if (rw
== RW_READER
) { /* i.e. not the first pass */
3158 int error
= zfs_zget(ZTOZSB(zp
), oidp
, &zp
);
3163 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(ZTOZSB(zp
)),
3164 &oidp
, sizeof (oidp
));
3165 rwlp
= &zp
->z_parent_lock
;
3168 } while (zp
->z_id
!= sdzp
->z_id
);
3174 * Move an entry from the provided source directory to the target
3175 * directory. Change the entry name as indicated.
3177 * IN: sdip - Source directory containing the "old entry".
3178 * snm - Old entry name.
3179 * tdip - Target directory to contain the "new entry".
3180 * tnm - New entry name.
3181 * cr - credentials of caller.
3182 * flags - case flags
3184 * RETURN: 0 on success, error code on failure.
3187 * sdip,tdip - ctime|mtime updated
3191 zfs_rename(struct inode
*sdip
, char *snm
, struct inode
*tdip
, char *tnm
,
3192 cred_t
*cr
, int flags
)
3194 znode_t
*tdzp
, *szp
, *tzp
;
3195 znode_t
*sdzp
= ITOZ(sdip
);
3196 zfs_sb_t
*zsb
= ITOZSB(sdip
);
3198 zfs_dirlock_t
*sdl
, *tdl
;
3201 int cmp
, serr
, terr
;
3204 boolean_t waited
= B_FALSE
;
3207 ZFS_VERIFY_ZP(sdzp
);
3211 ZFS_VERIFY_ZP(tdzp
);
3214 * We check i_sb because snapshots and the ctldir must have different
3217 if (tdip
->i_sb
!= sdip
->i_sb
|| zfsctl_is_node(tdip
)) {
3219 return (SET_ERROR(EXDEV
));
3222 if (zsb
->z_utf8
&& u8_validate(tnm
,
3223 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3225 return (SET_ERROR(EILSEQ
));
3228 if (flags
& FIGNORECASE
)
3237 * This is to prevent the creation of links into attribute space
3238 * by renaming a linked file into/outof an attribute directory.
3239 * See the comment in zfs_link() for why this is considered bad.
3241 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3243 return (SET_ERROR(EINVAL
));
3247 * Lock source and target directory entries. To prevent deadlock,
3248 * a lock ordering must be defined. We lock the directory with
3249 * the smallest object id first, or if it's a tie, the one with
3250 * the lexically first name.
3252 if (sdzp
->z_id
< tdzp
->z_id
) {
3254 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3258 * First compare the two name arguments without
3259 * considering any case folding.
3261 int nofold
= (zsb
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3263 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3264 ASSERT(error
== 0 || !zsb
->z_utf8
);
3267 * POSIX: "If the old argument and the new argument
3268 * both refer to links to the same existing file,
3269 * the rename() function shall return successfully
3270 * and perform no other action."
3276 * If the file system is case-folding, then we may
3277 * have some more checking to do. A case-folding file
3278 * system is either supporting mixed case sensitivity
3279 * access or is completely case-insensitive. Note
3280 * that the file system is always case preserving.
3282 * In mixed sensitivity mode case sensitive behavior
3283 * is the default. FIGNORECASE must be used to
3284 * explicitly request case insensitive behavior.
3286 * If the source and target names provided differ only
3287 * by case (e.g., a request to rename 'tim' to 'Tim'),
3288 * we will treat this as a special case in the
3289 * case-insensitive mode: as long as the source name
3290 * is an exact match, we will allow this to proceed as
3291 * a name-change request.
3293 if ((zsb
->z_case
== ZFS_CASE_INSENSITIVE
||
3294 (zsb
->z_case
== ZFS_CASE_MIXED
&&
3295 flags
& FIGNORECASE
)) &&
3296 u8_strcmp(snm
, tnm
, 0, zsb
->z_norm
, U8_UNICODE_LATEST
,
3299 * case preserving rename request, require exact
3308 * If the source and destination directories are the same, we should
3309 * grab the z_name_lock of that directory only once.
3313 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3317 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3318 ZEXISTS
| zflg
, NULL
, NULL
);
3319 terr
= zfs_dirent_lock(&tdl
,
3320 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3322 terr
= zfs_dirent_lock(&tdl
,
3323 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3324 serr
= zfs_dirent_lock(&sdl
,
3325 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3331 * Source entry invalid or not there.
3334 zfs_dirent_unlock(tdl
);
3340 rw_exit(&sdzp
->z_name_lock
);
3342 if (strcmp(snm
, "..") == 0)
3348 zfs_dirent_unlock(sdl
);
3352 rw_exit(&sdzp
->z_name_lock
);
3354 if (strcmp(tnm
, "..") == 0)
3361 * Must have write access at the source to remove the old entry
3362 * and write access at the target to create the new entry.
3363 * Note that if target and source are the same, this can be
3364 * done in a single check.
3367 if ((error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
)))
3370 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3372 * Check to make sure rename is valid.
3373 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3375 if ((error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
)))
3380 * Does target exist?
3384 * Source and target must be the same type.
3386 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3387 if (!S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3388 error
= SET_ERROR(ENOTDIR
);
3392 if (S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3393 error
= SET_ERROR(EISDIR
);
3398 * POSIX dictates that when the source and target
3399 * entries refer to the same file object, rename
3400 * must do nothing and exit without error.
3402 if (szp
->z_id
== tzp
->z_id
) {
3408 tx
= dmu_tx_create(zsb
->z_os
);
3409 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3410 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3411 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3412 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3414 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3415 zfs_sa_upgrade_txholds(tx
, tdzp
);
3418 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3419 zfs_sa_upgrade_txholds(tx
, tzp
);
3422 zfs_sa_upgrade_txholds(tx
, szp
);
3423 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
3424 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3427 zfs_rename_unlock(&zl
);
3428 zfs_dirent_unlock(sdl
);
3429 zfs_dirent_unlock(tdl
);
3432 rw_exit(&sdzp
->z_name_lock
);
3437 if (error
== ERESTART
) {
3448 if (tzp
) /* Attempt to remove the existing target */
3449 error
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3452 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3454 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3456 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zsb
),
3457 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3460 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3462 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3463 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3464 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3467 * At this point, we have successfully created
3468 * the target name, but have failed to remove
3469 * the source name. Since the create was done
3470 * with the ZRENAMING flag, there are
3471 * complications; for one, the link count is
3472 * wrong. The easiest way to deal with this
3473 * is to remove the newly created target, and
3474 * return the original error. This must
3475 * succeed; fortunately, it is very unlikely to
3476 * fail, since we just created it.
3478 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3479 ZRENAMING
, NULL
), ==, 0);
3487 zfs_rename_unlock(&zl
);
3489 zfs_dirent_unlock(sdl
);
3490 zfs_dirent_unlock(tdl
);
3492 zfs_inode_update(sdzp
);
3494 rw_exit(&sdzp
->z_name_lock
);
3497 zfs_inode_update(tdzp
);
3499 zfs_inode_update(szp
);
3502 zfs_inode_update(tzp
);
3506 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3507 zil_commit(zilog
, 0);
3512 EXPORT_SYMBOL(zfs_rename
);
3515 * Insert the indicated symbolic reference entry into the directory.
3517 * IN: dip - Directory to contain new symbolic link.
3518 * link - Name for new symlink entry.
3519 * vap - Attributes of new entry.
3520 * target - Target path of new symlink.
3522 * cr - credentials of caller.
3523 * flags - case flags
3525 * RETURN: 0 on success, error code on failure.
3528 * dip - ctime|mtime updated
3532 zfs_symlink(struct inode
*dip
, char *name
, vattr_t
*vap
, char *link
,
3533 struct inode
**ipp
, cred_t
*cr
, int flags
)
3535 znode_t
*zp
, *dzp
= ITOZ(dip
);
3538 zfs_sb_t
*zsb
= ITOZSB(dip
);
3540 uint64_t len
= strlen(link
);
3543 zfs_acl_ids_t acl_ids
;
3544 boolean_t fuid_dirtied
;
3545 uint64_t txtype
= TX_SYMLINK
;
3546 boolean_t waited
= B_FALSE
;
3548 ASSERT(S_ISLNK(vap
->va_mode
));
3554 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
3555 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3557 return (SET_ERROR(EILSEQ
));
3559 if (flags
& FIGNORECASE
)
3562 if (len
> MAXPATHLEN
) {
3564 return (SET_ERROR(ENAMETOOLONG
));
3567 if ((error
= zfs_acl_ids_create(dzp
, 0,
3568 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3576 * Attempt to lock directory; fail if entry already exists.
3578 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3580 zfs_acl_ids_free(&acl_ids
);
3585 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3586 zfs_acl_ids_free(&acl_ids
);
3587 zfs_dirent_unlock(dl
);
3592 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
3593 zfs_acl_ids_free(&acl_ids
);
3594 zfs_dirent_unlock(dl
);
3596 return (SET_ERROR(EDQUOT
));
3598 tx
= dmu_tx_create(zsb
->z_os
);
3599 fuid_dirtied
= zsb
->z_fuid_dirty
;
3600 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3601 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3602 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3603 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3604 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3605 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3606 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3607 acl_ids
.z_aclp
->z_acl_bytes
);
3610 zfs_fuid_txhold(zsb
, tx
);
3611 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3613 zfs_dirent_unlock(dl
);
3614 if (error
== ERESTART
) {
3620 zfs_acl_ids_free(&acl_ids
);
3627 * Create a new object for the symlink.
3628 * for version 4 ZPL datsets the symlink will be an SA attribute
3630 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3633 zfs_fuid_sync(zsb
, tx
);
3635 mutex_enter(&zp
->z_lock
);
3637 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zsb
),
3640 zfs_sa_symlink(zp
, link
, len
, tx
);
3641 mutex_exit(&zp
->z_lock
);
3644 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
3645 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3647 * Insert the new object into the directory.
3649 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3651 if (flags
& FIGNORECASE
)
3653 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3655 zfs_inode_update(dzp
);
3656 zfs_inode_update(zp
);
3658 zfs_acl_ids_free(&acl_ids
);
3662 zfs_dirent_unlock(dl
);
3666 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3667 zil_commit(zilog
, 0);
3672 EXPORT_SYMBOL(zfs_symlink
);
3675 * Return, in the buffer contained in the provided uio structure,
3676 * the symbolic path referred to by ip.
3678 * IN: ip - inode of symbolic link
3679 * uio - structure to contain the link path.
3680 * cr - credentials of caller.
3682 * RETURN: 0 if success
3683 * error code if failure
3686 * ip - atime updated
3690 zfs_readlink(struct inode
*ip
, uio_t
*uio
, cred_t
*cr
)
3692 znode_t
*zp
= ITOZ(ip
);
3693 zfs_sb_t
*zsb
= ITOZSB(ip
);
3699 mutex_enter(&zp
->z_lock
);
3701 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3702 SA_ZPL_SYMLINK(zsb
), uio
);
3704 error
= zfs_sa_readlink(zp
, uio
);
3705 mutex_exit(&zp
->z_lock
);
3707 ZFS_ACCESSTIME_STAMP(zsb
, zp
);
3711 EXPORT_SYMBOL(zfs_readlink
);
3714 * Insert a new entry into directory tdip referencing sip.
3716 * IN: tdip - Directory to contain new entry.
3717 * sip - inode of new entry.
3718 * name - name of new entry.
3719 * cr - credentials of caller.
3721 * RETURN: 0 if success
3722 * error code if failure
3725 * tdip - ctime|mtime updated
3726 * sip - ctime updated
3730 zfs_link(struct inode
*tdip
, struct inode
*sip
, char *name
, cred_t
*cr
)
3732 znode_t
*dzp
= ITOZ(tdip
);
3734 zfs_sb_t
*zsb
= ITOZSB(tdip
);
3742 boolean_t waited
= B_FALSE
;
3744 ASSERT(S_ISDIR(tdip
->i_mode
));
3751 * POSIX dictates that we return EPERM here.
3752 * Better choices include ENOTSUP or EISDIR.
3754 if (S_ISDIR(sip
->i_mode
)) {
3756 return (SET_ERROR(EPERM
));
3763 * We check i_sb because snapshots and the ctldir must have different
3766 if (sip
->i_sb
!= tdip
->i_sb
|| zfsctl_is_node(sip
)) {
3768 return (SET_ERROR(EXDEV
));
3771 /* Prevent links to .zfs/shares files */
3773 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
3774 &parent
, sizeof (uint64_t))) != 0) {
3778 if (parent
== zsb
->z_shares_dir
) {
3780 return (SET_ERROR(EPERM
));
3783 if (zsb
->z_utf8
&& u8_validate(name
,
3784 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3786 return (SET_ERROR(EILSEQ
));
3788 #ifdef HAVE_PN_UTILS
3789 if (flags
& FIGNORECASE
)
3791 #endif /* HAVE_PN_UTILS */
3794 * We do not support links between attributes and non-attributes
3795 * because of the potential security risk of creating links
3796 * into "normal" file space in order to circumvent restrictions
3797 * imposed in attribute space.
3799 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
3801 return (SET_ERROR(EINVAL
));
3804 owner
= zfs_fuid_map_id(zsb
, szp
->z_uid
, cr
, ZFS_OWNER
);
3805 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
3807 return (SET_ERROR(EPERM
));
3810 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3817 * Attempt to lock directory; fail if entry already exists.
3819 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
3825 tx
= dmu_tx_create(zsb
->z_os
);
3826 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3827 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3828 zfs_sa_upgrade_txholds(tx
, szp
);
3829 zfs_sa_upgrade_txholds(tx
, dzp
);
3830 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3832 zfs_dirent_unlock(dl
);
3833 if (error
== ERESTART
) {
3844 error
= zfs_link_create(dl
, szp
, tx
, 0);
3847 uint64_t txtype
= TX_LINK
;
3848 #ifdef HAVE_PN_UTILS
3849 if (flags
& FIGNORECASE
)
3851 #endif /* HAVE_PN_UTILS */
3852 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
3857 zfs_dirent_unlock(dl
);
3859 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3860 zil_commit(zilog
, 0);
3862 zfs_inode_update(dzp
);
3863 zfs_inode_update(szp
);
3867 EXPORT_SYMBOL(zfs_link
);
3870 zfs_putpage_commit_cb(void *arg
)
3872 struct page
*pp
= arg
;
3875 end_page_writeback(pp
);
3879 * Push a page out to disk, once the page is on stable storage the
3880 * registered commit callback will be run as notification of completion.
3882 * IN: ip - page mapped for inode.
3883 * pp - page to push (page is locked)
3884 * wbc - writeback control data
3886 * RETURN: 0 if success
3887 * error code if failure
3890 * ip - ctime|mtime updated
3894 zfs_putpage(struct inode
*ip
, struct page
*pp
, struct writeback_control
*wbc
)
3896 znode_t
*zp
= ITOZ(ip
);
3897 zfs_sb_t
*zsb
= ITOZSB(ip
);
3905 uint64_t mtime
[2], ctime
[2];
3906 sa_bulk_attr_t bulk
[3];
3908 struct address_space
*mapping
;
3913 ASSERT(PageLocked(pp
));
3915 pgoff
= page_offset(pp
); /* Page byte-offset in file */
3916 offset
= i_size_read(ip
); /* File length in bytes */
3917 pglen
= MIN(PAGE_CACHE_SIZE
, /* Page length in bytes */
3918 P2ROUNDUP(offset
, PAGE_CACHE_SIZE
)-pgoff
);
3920 /* Page is beyond end of file */
3921 if (pgoff
>= offset
) {
3927 /* Truncate page length to end of file */
3928 if (pgoff
+ pglen
> offset
)
3929 pglen
= offset
- pgoff
;
3933 * FIXME: Allow mmap writes past its quota. The correct fix
3934 * is to register a page_mkwrite() handler to count the page
3935 * against its quota when it is about to be dirtied.
3937 if (zfs_owner_overquota(zsb
, zp
, B_FALSE
) ||
3938 zfs_owner_overquota(zsb
, zp
, B_TRUE
)) {
3944 * The ordering here is critical and must adhere to the following
3945 * rules in order to avoid deadlocking in either zfs_read() or
3946 * zfs_free_range() due to a lock inversion.
3948 * 1) The page must be unlocked prior to acquiring the range lock.
3949 * This is critical because zfs_read() calls find_lock_page()
3950 * which may block on the page lock while holding the range lock.
3952 * 2) Before setting or clearing write back on a page the range lock
3953 * must be held in order to prevent a lock inversion with the
3954 * zfs_free_range() function.
3956 * This presents a problem because upon entering this function the
3957 * page lock is already held. To safely acquire the range lock the
3958 * page lock must be dropped. This creates a window where another
3959 * process could truncate, invalidate, dirty, or write out the page.
3961 * Therefore, after successfully reacquiring the range and page locks
3962 * the current page state is checked. In the common case everything
3963 * will be as is expected and it can be written out. However, if
3964 * the page state has changed it must be handled accordingly.
3966 mapping
= pp
->mapping
;
3967 redirty_page_for_writepage(wbc
, pp
);
3970 rl
= zfs_range_lock(zp
, pgoff
, pglen
, RL_WRITER
);
3973 /* Page mapping changed or it was no longer dirty, we're done */
3974 if (unlikely((mapping
!= pp
->mapping
) || !PageDirty(pp
))) {
3976 zfs_range_unlock(rl
);
3981 /* Another process started write block if required */
3982 if (PageWriteback(pp
)) {
3984 zfs_range_unlock(rl
);
3986 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
3987 wait_on_page_writeback(pp
);
3993 /* Clear the dirty flag the required locks are held */
3994 if (!clear_page_dirty_for_io(pp
)) {
3996 zfs_range_unlock(rl
);
4002 * Counterpart for redirty_page_for_writepage() above. This page
4003 * was in fact not skipped and should not be counted as if it were.
4005 wbc
->pages_skipped
--;
4006 set_page_writeback(pp
);
4009 tx
= dmu_tx_create(zsb
->z_os
);
4010 dmu_tx_hold_write(tx
, zp
->z_id
, pgoff
, pglen
);
4011 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4012 zfs_sa_upgrade_txholds(tx
, zp
);
4014 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4016 if (err
== ERESTART
)
4020 __set_page_dirty_nobuffers(pp
);
4022 end_page_writeback(pp
);
4023 zfs_range_unlock(rl
);
4029 ASSERT3U(pglen
, <=, PAGE_CACHE_SIZE
);
4030 dmu_write(zsb
->z_os
, zp
->z_id
, pgoff
, pglen
, va
, tx
);
4033 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
4034 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
4035 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_FLAGS(zsb
), NULL
, &zp
->z_pflags
, 8);
4037 /* Preserve the mtime and ctime provided by the inode */
4038 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4039 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4040 zp
->z_atime_dirty
= 0;
4043 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4045 zfs_log_write(zsb
->z_log
, tx
, TX_WRITE
, zp
, pgoff
, pglen
, 0,
4046 zfs_putpage_commit_cb
, pp
);
4049 zfs_range_unlock(rl
);
4051 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4053 * Note that this is rarely called under writepages(), because
4054 * writepages() normally handles the entire commit for
4055 * performance reasons.
4057 if (zsb
->z_log
!= NULL
)
4058 zil_commit(zsb
->z_log
, zp
->z_id
);
4066 * Update the system attributes when the inode has been dirtied. For the
4067 * moment we only update the mode, atime, mtime, and ctime.
4070 zfs_dirty_inode(struct inode
*ip
, int flags
)
4072 znode_t
*zp
= ITOZ(ip
);
4073 zfs_sb_t
*zsb
= ITOZSB(ip
);
4075 uint64_t mode
, atime
[2], mtime
[2], ctime
[2];
4076 sa_bulk_attr_t bulk
[4];
4080 if (zfs_is_readonly(zsb
) || dmu_objset_is_snapshot(zsb
->z_os
))
4086 tx
= dmu_tx_create(zsb
->z_os
);
4088 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4089 zfs_sa_upgrade_txholds(tx
, zp
);
4091 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4097 mutex_enter(&zp
->z_lock
);
4098 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MODE(zsb
), NULL
, &mode
, 8);
4099 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_ATIME(zsb
), NULL
, &atime
, 16);
4100 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
4101 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
4103 /* Preserve the mode, mtime and ctime provided by the inode */
4104 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4105 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4106 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4110 zp
->z_atime_dirty
= 0;
4112 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4113 mutex_exit(&zp
->z_lock
);
4120 EXPORT_SYMBOL(zfs_dirty_inode
);
4124 zfs_inactive(struct inode
*ip
)
4126 znode_t
*zp
= ITOZ(ip
);
4127 zfs_sb_t
*zsb
= ITOZSB(ip
);
4129 int need_unlock
= 0;
4131 /* Only read lock if we haven't already write locked, e.g. rollback */
4132 if (!RW_WRITE_HELD(&zsb
->z_teardown_inactive_lock
)) {
4134 rw_enter(&zsb
->z_teardown_inactive_lock
, RW_READER
);
4136 if (zp
->z_sa_hdl
== NULL
) {
4138 rw_exit(&zsb
->z_teardown_inactive_lock
);
4142 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4143 dmu_tx_t
*tx
= dmu_tx_create(zsb
->z_os
);
4145 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4146 zfs_sa_upgrade_txholds(tx
, zp
);
4147 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4151 mutex_enter(&zp
->z_lock
);
4152 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zsb
),
4153 (void *)&zp
->z_atime
, sizeof (zp
->z_atime
), tx
);
4154 zp
->z_atime_dirty
= 0;
4155 mutex_exit(&zp
->z_lock
);
4162 rw_exit(&zsb
->z_teardown_inactive_lock
);
4164 EXPORT_SYMBOL(zfs_inactive
);
4167 * Bounds-check the seek operation.
4169 * IN: ip - inode seeking within
4170 * ooff - old file offset
4171 * noffp - pointer to new file offset
4172 * ct - caller context
4174 * RETURN: 0 if success
4175 * EINVAL if new offset invalid
4179 zfs_seek(struct inode
*ip
, offset_t ooff
, offset_t
*noffp
)
4181 if (S_ISDIR(ip
->i_mode
))
4183 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4185 EXPORT_SYMBOL(zfs_seek
);
4188 * Fill pages with data from the disk.
4191 zfs_fillpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4193 znode_t
*zp
= ITOZ(ip
);
4194 zfs_sb_t
*zsb
= ITOZSB(ip
);
4196 struct page
*cur_pp
;
4197 u_offset_t io_off
, total
;
4204 io_len
= nr_pages
<< PAGE_CACHE_SHIFT
;
4205 i_size
= i_size_read(ip
);
4206 io_off
= page_offset(pl
[0]);
4208 if (io_off
+ io_len
> i_size
)
4209 io_len
= i_size
- io_off
;
4212 * Iterate over list of pages and read each page individually.
4216 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4220 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4224 /* convert checksum errors into IO errors */
4226 err
= SET_ERROR(EIO
);
4229 cur_pp
= pl
[++page_idx
];
4236 * Uses zfs_fillpage to read data from the file and fill the pages.
4238 * IN: ip - inode of file to get data from.
4239 * pl - list of pages to read
4240 * nr_pages - number of pages to read
4242 * RETURN: 0 on success, error code on failure.
4245 * vp - atime updated
4249 zfs_getpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4251 znode_t
*zp
= ITOZ(ip
);
4252 zfs_sb_t
*zsb
= ITOZSB(ip
);
4261 err
= zfs_fillpage(ip
, pl
, nr_pages
);
4264 ZFS_ACCESSTIME_STAMP(zsb
, zp
);
4269 EXPORT_SYMBOL(zfs_getpage
);
4272 * Check ZFS specific permissions to memory map a section of a file.
4274 * IN: ip - inode of the file to mmap
4276 * addrp - start address in memory region
4277 * len - length of memory region
4278 * vm_flags- address flags
4280 * RETURN: 0 if success
4281 * error code if failure
4285 zfs_map(struct inode
*ip
, offset_t off
, caddr_t
*addrp
, size_t len
,
4286 unsigned long vm_flags
)
4288 znode_t
*zp
= ITOZ(ip
);
4289 zfs_sb_t
*zsb
= ITOZSB(ip
);
4294 if ((vm_flags
& VM_WRITE
) && (zp
->z_pflags
&
4295 (ZFS_IMMUTABLE
| ZFS_READONLY
| ZFS_APPENDONLY
))) {
4297 return (SET_ERROR(EPERM
));
4300 if ((vm_flags
& (VM_READ
| VM_EXEC
)) &&
4301 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4303 return (SET_ERROR(EACCES
));
4306 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4308 return (SET_ERROR(ENXIO
));
4314 EXPORT_SYMBOL(zfs_map
);
4317 * convoff - converts the given data (start, whence) to the
4321 convoff(struct inode
*ip
, flock64_t
*lckdat
, int whence
, offset_t offset
)
4326 if ((lckdat
->l_whence
== 2) || (whence
== 2)) {
4327 if ((error
= zfs_getattr(ip
, &vap
, 0, CRED()) != 0))
4331 switch (lckdat
->l_whence
) {
4333 lckdat
->l_start
+= offset
;
4336 lckdat
->l_start
+= vap
.va_size
;
4341 return (SET_ERROR(EINVAL
));
4344 if (lckdat
->l_start
< 0)
4345 return (SET_ERROR(EINVAL
));
4349 lckdat
->l_start
-= offset
;
4352 lckdat
->l_start
-= vap
.va_size
;
4357 return (SET_ERROR(EINVAL
));
4360 lckdat
->l_whence
= (short)whence
;
4365 * Free or allocate space in a file. Currently, this function only
4366 * supports the `F_FREESP' command. However, this command is somewhat
4367 * misnamed, as its functionality includes the ability to allocate as
4368 * well as free space.
4370 * IN: ip - inode of file to free data in.
4371 * cmd - action to take (only F_FREESP supported).
4372 * bfp - section of file to free/alloc.
4373 * flag - current file open mode flags.
4374 * offset - current file offset.
4375 * cr - credentials of caller [UNUSED].
4377 * RETURN: 0 on success, error code on failure.
4380 * ip - ctime|mtime updated
4384 zfs_space(struct inode
*ip
, int cmd
, flock64_t
*bfp
, int flag
,
4385 offset_t offset
, cred_t
*cr
)
4387 znode_t
*zp
= ITOZ(ip
);
4388 zfs_sb_t
*zsb
= ITOZSB(ip
);
4395 if (cmd
!= F_FREESP
) {
4397 return (SET_ERROR(EINVAL
));
4401 * Callers might not be able to detect properly that we are read-only,
4402 * so check it explicitly here.
4404 if (zfs_is_readonly(zsb
)) {
4406 return (SET_ERROR(EROFS
));
4409 if ((error
= convoff(ip
, bfp
, 0, offset
))) {
4414 if (bfp
->l_len
< 0) {
4416 return (SET_ERROR(EINVAL
));
4420 * Permissions aren't checked on Solaris because on this OS
4421 * zfs_space() can only be called with an opened file handle.
4422 * On Linux we can get here through truncate_range() which
4423 * operates directly on inodes, so we need to check access rights.
4425 if ((error
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
))) {
4431 len
= bfp
->l_len
; /* 0 means from off to end of file */
4433 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4438 EXPORT_SYMBOL(zfs_space
);
4442 zfs_fid(struct inode
*ip
, fid_t
*fidp
)
4444 znode_t
*zp
= ITOZ(ip
);
4445 zfs_sb_t
*zsb
= ITOZSB(ip
);
4448 uint64_t object
= zp
->z_id
;
4455 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zsb
),
4456 &gen64
, sizeof (uint64_t))) != 0) {
4461 gen
= (uint32_t)gen64
;
4463 size
= (zsb
->z_parent
!= zsb
) ? LONG_FID_LEN
: SHORT_FID_LEN
;
4464 if (fidp
->fid_len
< size
) {
4465 fidp
->fid_len
= size
;
4467 return (SET_ERROR(ENOSPC
));
4470 zfid
= (zfid_short_t
*)fidp
;
4472 zfid
->zf_len
= size
;
4474 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4475 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4477 /* Must have a non-zero generation number to distinguish from .zfs */
4480 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4481 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4483 if (size
== LONG_FID_LEN
) {
4484 uint64_t objsetid
= dmu_objset_id(zsb
->z_os
);
4487 zlfid
= (zfid_long_t
*)fidp
;
4489 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
4490 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
4492 /* XXX - this should be the generation number for the objset */
4493 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
4494 zlfid
->zf_setgen
[i
] = 0;
4500 EXPORT_SYMBOL(zfs_fid
);
4504 zfs_getsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4506 znode_t
*zp
= ITOZ(ip
);
4507 zfs_sb_t
*zsb
= ITOZSB(ip
);
4509 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4513 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
4518 EXPORT_SYMBOL(zfs_getsecattr
);
4522 zfs_setsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4524 znode_t
*zp
= ITOZ(ip
);
4525 zfs_sb_t
*zsb
= ITOZSB(ip
);
4527 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4528 zilog_t
*zilog
= zsb
->z_log
;
4533 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
4535 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4536 zil_commit(zilog
, 0);
4541 EXPORT_SYMBOL(zfs_setsecattr
);
4543 #ifdef HAVE_UIO_ZEROCOPY
4545 * Tunable, both must be a power of 2.
4547 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4548 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4549 * an arcbuf for a partial block read
4551 int zcr_blksz_min
= (1 << 10); /* 1K */
4552 int zcr_blksz_max
= (1 << 17); /* 128K */
4556 zfs_reqzcbuf(struct inode
*ip
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
)
4558 znode_t
*zp
= ITOZ(ip
);
4559 zfs_sb_t
*zsb
= ITOZSB(ip
);
4560 int max_blksz
= zsb
->z_max_blksz
;
4561 uio_t
*uio
= &xuio
->xu_uio
;
4562 ssize_t size
= uio
->uio_resid
;
4563 offset_t offset
= uio
->uio_loffset
;
4568 int preamble
, postamble
;
4570 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
4571 return (SET_ERROR(EINVAL
));
4578 * Loan out an arc_buf for write if write size is bigger than
4579 * max_blksz, and the file's block size is also max_blksz.
4582 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
4584 return (SET_ERROR(EINVAL
));
4587 * Caller requests buffers for write before knowing where the
4588 * write offset might be (e.g. NFS TCP write).
4593 preamble
= P2PHASE(offset
, blksz
);
4595 preamble
= blksz
- preamble
;
4600 postamble
= P2PHASE(size
, blksz
);
4603 fullblk
= size
/ blksz
;
4604 (void) dmu_xuio_init(xuio
,
4605 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
4608 * Have to fix iov base/len for partial buffers. They
4609 * currently represent full arc_buf's.
4612 /* data begins in the middle of the arc_buf */
4613 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4616 (void) dmu_xuio_add(xuio
, abuf
,
4617 blksz
- preamble
, preamble
);
4620 for (i
= 0; i
< fullblk
; i
++) {
4621 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4624 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
4628 /* data ends in the middle of the arc_buf */
4629 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4632 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
4637 * Loan out an arc_buf for read if the read size is larger than
4638 * the current file block size. Block alignment is not
4639 * considered. Partial arc_buf will be loaned out for read.
4641 blksz
= zp
->z_blksz
;
4642 if (blksz
< zcr_blksz_min
)
4643 blksz
= zcr_blksz_min
;
4644 if (blksz
> zcr_blksz_max
)
4645 blksz
= zcr_blksz_max
;
4646 /* avoid potential complexity of dealing with it */
4647 if (blksz
> max_blksz
) {
4649 return (SET_ERROR(EINVAL
));
4652 maxsize
= zp
->z_size
- uio
->uio_loffset
;
4658 return (SET_ERROR(EINVAL
));
4663 return (SET_ERROR(EINVAL
));
4666 uio
->uio_extflg
= UIO_XUIO
;
4667 XUIO_XUZC_RW(xuio
) = ioflag
;
4674 zfs_retzcbuf(struct inode
*ip
, xuio_t
*xuio
, cred_t
*cr
)
4678 int ioflag
= XUIO_XUZC_RW(xuio
);
4680 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
4682 i
= dmu_xuio_cnt(xuio
);
4684 abuf
= dmu_xuio_arcbuf(xuio
, i
);
4686 * if abuf == NULL, it must be a write buffer
4687 * that has been returned in zfs_write().
4690 dmu_return_arcbuf(abuf
);
4691 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
4694 dmu_xuio_fini(xuio
);
4697 #endif /* HAVE_UIO_ZEROCOPY */
4699 #if defined(_KERNEL) && defined(HAVE_SPL)
4700 module_param(zfs_read_chunk_size
, long, 0644);
4701 MODULE_PARM_DESC(zfs_read_chunk_size
, "Bytes to read per chunk");