4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
28 /* Portions Copyright 2007 Jeremy Teo */
29 /* Portions Copyright 2010 Robert Milkowski */
32 #include <sys/types.h>
33 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/sysmacros.h>
37 #include <sys/resource.h>
39 #include <sys/vfs_opreg.h>
43 #include <sys/taskq.h>
45 #include <sys/vmsystm.h>
46 #include <sys/atomic.h>
48 #include <sys/pathname.h>
49 #include <sys/cmn_err.h>
50 #include <sys/errno.h>
51 #include <sys/unistd.h>
52 #include <sys/zfs_dir.h>
53 #include <sys/zfs_acl.h>
54 #include <sys/zfs_ioctl.h>
55 #include <sys/fs/zfs.h>
57 #include <sys/dmu_objset.h>
63 #include <sys/dirent.h>
64 #include <sys/policy.h>
65 #include <sys/sunddi.h>
68 #include "fs/fs_subr.h"
69 #include <sys/zfs_ctldir.h>
70 #include <sys/zfs_fuid.h>
71 #include <sys/zfs_sa.h>
72 #include <sys/zfs_vnops.h>
74 #include <sys/zfs_rlock.h>
75 #include <sys/extdirent.h>
76 #include <sys/kidmap.h>
84 * Each vnode op performs some logical unit of work. To do this, the ZPL must
85 * properly lock its in-core state, create a DMU transaction, do the work,
86 * record this work in the intent log (ZIL), commit the DMU transaction,
87 * and wait for the intent log to commit if it is a synchronous operation.
88 * Moreover, the vnode ops must work in both normal and log replay context.
89 * The ordering of events is important to avoid deadlocks and references
90 * to freed memory. The example below illustrates the following Big Rules:
92 * (1) A check must be made in each zfs thread for a mounted file system.
93 * This is done avoiding races using ZFS_ENTER(zsb).
94 * A ZFS_EXIT(zsb) is needed before all returns. Any znodes
95 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
96 * can return EIO from the calling function.
98 * (2) iput() should always be the last thing except for zil_commit()
99 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
100 * First, if it's the last reference, the vnode/znode
101 * can be freed, so the zp may point to freed memory. Second, the last
102 * reference will call zfs_zinactive(), which may induce a lot of work --
103 * pushing cached pages (which acquires range locks) and syncing out
104 * cached atime changes. Third, zfs_zinactive() may require a new tx,
105 * which could deadlock the system if you were already holding one.
106 * If you must call iput() within a tx then use zfs_iput_async().
108 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
109 * as they can span dmu_tx_assign() calls.
111 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
112 * dmu_tx_assign(). This is critical because we don't want to block
113 * while holding locks.
115 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
116 * reduces lock contention and CPU usage when we must wait (note that if
117 * throughput is constrained by the storage, nearly every transaction
120 * Note, in particular, that if a lock is sometimes acquired before
121 * the tx assigns, and sometimes after (e.g. z_lock), then failing
122 * to use a non-blocking assign can deadlock the system. The scenario:
124 * Thread A has grabbed a lock before calling dmu_tx_assign().
125 * Thread B is in an already-assigned tx, and blocks for this lock.
126 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
127 * forever, because the previous txg can't quiesce until B's tx commits.
129 * If dmu_tx_assign() returns ERESTART and zsb->z_assign is TXG_NOWAIT,
130 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
131 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
132 * to indicate that this operation has already called dmu_tx_wait().
133 * This will ensure that we don't retry forever, waiting a short bit
136 * (5) If the operation succeeded, generate the intent log entry for it
137 * before dropping locks. This ensures that the ordering of events
138 * in the intent log matches the order in which they actually occurred.
139 * During ZIL replay the zfs_log_* functions will update the sequence
140 * number to indicate the zil transaction has replayed.
142 * (6) At the end of each vnode op, the DMU tx must always commit,
143 * regardless of whether there were any errors.
145 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
146 * to ensure that synchronous semantics are provided when necessary.
148 * In general, this is how things should be ordered in each vnode op:
150 * ZFS_ENTER(zsb); // exit if unmounted
152 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
153 * rw_enter(...); // grab any other locks you need
154 * tx = dmu_tx_create(...); // get DMU tx
155 * dmu_tx_hold_*(); // hold each object you might modify
156 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
158 * rw_exit(...); // drop locks
159 * zfs_dirent_unlock(dl); // unlock directory entry
160 * iput(...); // release held vnodes
161 * if (error == ERESTART) {
167 * dmu_tx_abort(tx); // abort DMU tx
168 * ZFS_EXIT(zsb); // finished in zfs
169 * return (error); // really out of space
171 * error = do_real_work(); // do whatever this VOP does
173 * zfs_log_*(...); // on success, make ZIL entry
174 * dmu_tx_commit(tx); // commit DMU tx -- error or not
175 * rw_exit(...); // drop locks
176 * zfs_dirent_unlock(dl); // unlock directory entry
177 * iput(...); // release held vnodes
178 * zil_commit(zilog, foid); // synchronous when necessary
179 * ZFS_EXIT(zsb); // finished in zfs
180 * return (error); // done, report error
184 * Virus scanning is unsupported. It would be possible to add a hook
185 * here to performance the required virus scan. This could be done
186 * entirely in the kernel or potentially as an update to invoke a
190 zfs_vscan(struct inode
*ip
, cred_t
*cr
, int async
)
197 zfs_open(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
199 znode_t
*zp
= ITOZ(ip
);
200 zfs_sb_t
*zsb
= ITOZSB(ip
);
205 /* Honor ZFS_APPENDONLY file attribute */
206 if ((mode
& FMODE_WRITE
) && (zp
->z_pflags
& ZFS_APPENDONLY
) &&
207 ((flag
& O_APPEND
) == 0)) {
209 return (SET_ERROR(EPERM
));
212 /* Virus scan eligible files on open */
213 if (!zfs_has_ctldir(zp
) && zsb
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
214 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0) {
215 if (zfs_vscan(ip
, cr
, 0) != 0) {
217 return (SET_ERROR(EACCES
));
221 /* Keep a count of the synchronous opens in the znode */
223 atomic_inc_32(&zp
->z_sync_cnt
);
228 EXPORT_SYMBOL(zfs_open
);
232 zfs_close(struct inode
*ip
, int flag
, cred_t
*cr
)
234 znode_t
*zp
= ITOZ(ip
);
235 zfs_sb_t
*zsb
= ITOZSB(ip
);
240 /* Decrement the synchronous opens in the znode */
242 atomic_dec_32(&zp
->z_sync_cnt
);
244 if (!zfs_has_ctldir(zp
) && zsb
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
245 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0)
246 VERIFY(zfs_vscan(ip
, cr
, 1) == 0);
251 EXPORT_SYMBOL(zfs_close
);
253 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
255 * Lseek support for finding holes (cmd == SEEK_HOLE) and
256 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
259 zfs_holey_common(struct inode
*ip
, int cmd
, loff_t
*off
)
261 znode_t
*zp
= ITOZ(ip
);
262 uint64_t noff
= (uint64_t)*off
; /* new offset */
267 file_sz
= zp
->z_size
;
268 if (noff
>= file_sz
) {
269 return (SET_ERROR(ENXIO
));
272 if (cmd
== SEEK_HOLE
)
277 error
= dmu_offset_next(ZTOZSB(zp
)->z_os
, zp
->z_id
, hole
, &noff
);
280 return (SET_ERROR(ENXIO
));
283 * We could find a hole that begins after the logical end-of-file,
284 * because dmu_offset_next() only works on whole blocks. If the
285 * EOF falls mid-block, then indicate that the "virtual hole"
286 * at the end of the file begins at the logical EOF, rather than
287 * at the end of the last block.
289 if (noff
> file_sz
) {
301 zfs_holey(struct inode
*ip
, int cmd
, loff_t
*off
)
303 znode_t
*zp
= ITOZ(ip
);
304 zfs_sb_t
*zsb
= ITOZSB(ip
);
310 error
= zfs_holey_common(ip
, cmd
, off
);
315 EXPORT_SYMBOL(zfs_holey
);
316 #endif /* SEEK_HOLE && SEEK_DATA */
320 * When a file is memory mapped, we must keep the IO data synchronized
321 * between the DMU cache and the memory mapped pages. What this means:
323 * On Write: If we find a memory mapped page, we write to *both*
324 * the page and the dmu buffer.
327 update_pages(struct inode
*ip
, int64_t start
, int len
,
328 objset_t
*os
, uint64_t oid
)
330 struct address_space
*mp
= ip
->i_mapping
;
336 off
= start
& (PAGE_SIZE
-1);
337 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
338 nbytes
= MIN(PAGE_SIZE
- off
, len
);
340 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
342 if (mapping_writably_mapped(mp
))
343 flush_dcache_page(pp
);
346 (void) dmu_read(os
, oid
, start
+off
, nbytes
, pb
+off
,
350 if (mapping_writably_mapped(mp
))
351 flush_dcache_page(pp
);
353 mark_page_accessed(pp
);
366 * When a file is memory mapped, we must keep the IO data synchronized
367 * between the DMU cache and the memory mapped pages. What this means:
369 * On Read: We "read" preferentially from memory mapped pages,
370 * else we default from the dmu buffer.
372 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
373 * the file is memory mapped.
376 mappedread(struct inode
*ip
, int nbytes
, uio_t
*uio
)
378 struct address_space
*mp
= ip
->i_mapping
;
380 znode_t
*zp
= ITOZ(ip
);
387 start
= uio
->uio_loffset
;
388 off
= start
& (PAGE_SIZE
-1);
389 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
390 bytes
= MIN(PAGE_SIZE
- off
, len
);
392 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
394 ASSERT(PageUptodate(pp
));
397 error
= uiomove(pb
+ off
, bytes
, UIO_READ
, uio
);
400 if (mapping_writably_mapped(mp
))
401 flush_dcache_page(pp
);
403 mark_page_accessed(pp
);
407 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
420 unsigned long zfs_read_chunk_size
= 1024 * 1024; /* Tunable */
421 unsigned long zfs_delete_blocks
= DMU_MAX_DELETEBLKCNT
;
424 * Read bytes from specified file into supplied buffer.
426 * IN: ip - inode of file to be read from.
427 * uio - structure supplying read location, range info,
429 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
430 * O_DIRECT flag; used to bypass page cache.
431 * cr - credentials of caller.
433 * OUT: uio - updated offset and range, buffer filled.
435 * RETURN: 0 on success, error code on failure.
438 * inode - atime updated if byte count > 0
442 zfs_read(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
444 znode_t
*zp
= ITOZ(ip
);
445 zfs_sb_t
*zsb
= ITOZSB(ip
);
449 #ifdef HAVE_UIO_ZEROCOPY
451 #endif /* HAVE_UIO_ZEROCOPY */
456 if (zp
->z_pflags
& ZFS_AV_QUARANTINED
) {
458 return (SET_ERROR(EACCES
));
462 * Validate file offset
464 if (uio
->uio_loffset
< (offset_t
)0) {
466 return (SET_ERROR(EINVAL
));
470 * Fasttrack empty reads
472 if (uio
->uio_resid
== 0) {
478 * If we're in FRSYNC mode, sync out this znode before reading it.
480 if (ioflag
& FRSYNC
|| zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
481 zil_commit(zsb
->z_log
, zp
->z_id
);
484 * Lock the range against changes.
486 rl
= zfs_range_lock(&zp
->z_range_lock
, uio
->uio_loffset
, uio
->uio_resid
,
490 * If we are reading past end-of-file we can skip
491 * to the end; but we might still need to set atime.
493 if (uio
->uio_loffset
>= zp
->z_size
) {
498 ASSERT(uio
->uio_loffset
< zp
->z_size
);
499 n
= MIN(uio
->uio_resid
, zp
->z_size
- uio
->uio_loffset
);
501 #ifdef HAVE_UIO_ZEROCOPY
502 if ((uio
->uio_extflg
== UIO_XUIO
) &&
503 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
)) {
505 int blksz
= zp
->z_blksz
;
506 uint64_t offset
= uio
->uio_loffset
;
508 xuio
= (xuio_t
*)uio
;
510 nblk
= (P2ROUNDUP(offset
+ n
, blksz
) - P2ALIGN(offset
,
513 ASSERT(offset
+ n
<= blksz
);
516 (void) dmu_xuio_init(xuio
, nblk
);
518 if (vn_has_cached_data(ip
)) {
520 * For simplicity, we always allocate a full buffer
521 * even if we only expect to read a portion of a block.
523 while (--nblk
>= 0) {
524 (void) dmu_xuio_add(xuio
,
525 dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
530 #endif /* HAVE_UIO_ZEROCOPY */
533 nbytes
= MIN(n
, zfs_read_chunk_size
-
534 P2PHASE(uio
->uio_loffset
, zfs_read_chunk_size
));
536 if (zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
)) {
537 error
= mappedread(ip
, nbytes
, uio
);
539 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
544 /* convert checksum errors into IO errors */
546 error
= SET_ERROR(EIO
);
553 zfs_range_unlock(rl
);
558 EXPORT_SYMBOL(zfs_read
);
561 * Write the bytes to a file.
563 * IN: ip - inode of file to be written to.
564 * uio - structure supplying write location, range info,
566 * ioflag - FAPPEND flag set if in append mode.
567 * O_DIRECT flag; used to bypass page cache.
568 * cr - credentials of caller.
570 * OUT: uio - updated offset and range.
572 * RETURN: 0 if success
573 * error code if failure
576 * ip - ctime|mtime updated if byte count > 0
581 zfs_write(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
583 znode_t
*zp
= ITOZ(ip
);
584 rlim64_t limit
= uio
->uio_limit
;
585 ssize_t start_resid
= uio
->uio_resid
;
589 zfs_sb_t
*zsb
= ZTOZSB(zp
);
594 int max_blksz
= zsb
->z_max_blksz
;
597 const iovec_t
*aiov
= NULL
;
601 sa_bulk_attr_t bulk
[4];
602 uint64_t mtime
[2], ctime
[2];
604 #ifdef HAVE_UIO_ZEROCOPY
606 const iovec_t
*iovp
= uio
->uio_iov
;
607 ASSERTV(int iovcnt
= uio
->uio_iovcnt
);
611 * Fasttrack empty write
617 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
623 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
624 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
625 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zsb
), NULL
, &zp
->z_size
, 8);
626 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
630 * Callers might not be able to detect properly that we are read-only,
631 * so check it explicitly here.
633 if (zfs_is_readonly(zsb
)) {
635 return (SET_ERROR(EROFS
));
639 * If immutable or not appending then return EPERM
641 if ((zp
->z_pflags
& (ZFS_IMMUTABLE
| ZFS_READONLY
)) ||
642 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
643 (uio
->uio_loffset
< zp
->z_size
))) {
645 return (SET_ERROR(EPERM
));
651 * Validate file offset
653 woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
656 return (SET_ERROR(EINVAL
));
660 * Pre-fault the pages to ensure slow (eg NFS) pages
662 * Skip this if uio contains loaned arc_buf.
664 #ifdef HAVE_UIO_ZEROCOPY
665 if ((uio
->uio_extflg
== UIO_XUIO
) &&
666 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
))
667 xuio
= (xuio_t
*)uio
;
670 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
673 * If in append mode, set the io offset pointer to eof.
675 if (ioflag
& FAPPEND
) {
677 * Obtain an appending range lock to guarantee file append
678 * semantics. We reset the write offset once we have the lock.
680 rl
= zfs_range_lock(&zp
->z_range_lock
, 0, n
, RL_APPEND
);
682 if (rl
->r_len
== UINT64_MAX
) {
684 * We overlocked the file because this write will cause
685 * the file block size to increase.
686 * Note that zp_size cannot change with this lock held.
690 uio
->uio_loffset
= woff
;
693 * Note that if the file block size will change as a result of
694 * this write, then this range lock will lock the entire file
695 * so that we can re-write the block safely.
697 rl
= zfs_range_lock(&zp
->z_range_lock
, woff
, n
, RL_WRITER
);
701 zfs_range_unlock(rl
);
703 return (SET_ERROR(EFBIG
));
706 if ((woff
+ n
) > limit
|| woff
> (limit
- n
))
709 /* Will this write extend the file length? */
710 write_eof
= (woff
+ n
> zp
->z_size
);
712 end_size
= MAX(zp
->z_size
, woff
+ n
);
715 * Write the file in reasonable size chunks. Each chunk is written
716 * in a separate transaction; this keeps the intent log records small
717 * and allows us to do more fine-grained space accounting.
721 woff
= uio
->uio_loffset
;
722 if (zfs_owner_overquota(zsb
, zp
, B_FALSE
) ||
723 zfs_owner_overquota(zsb
, zp
, B_TRUE
)) {
725 dmu_return_arcbuf(abuf
);
726 error
= SET_ERROR(EDQUOT
);
730 if (xuio
&& abuf
== NULL
) {
731 #ifdef HAVE_UIO_ZEROCOPY
732 ASSERT(i_iov
< iovcnt
);
733 ASSERT3U(uio
->uio_segflg
, !=, UIO_BVEC
);
735 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
736 dmu_xuio_clear(xuio
, i_iov
);
737 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
738 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
739 aiov
->iov_len
== arc_buf_size(abuf
)));
742 } else if (abuf
== NULL
&& n
>= max_blksz
&&
743 woff
>= zp
->z_size
&&
744 P2PHASE(woff
, max_blksz
) == 0 &&
745 zp
->z_blksz
== max_blksz
) {
747 * This write covers a full block. "Borrow" a buffer
748 * from the dmu so that we can fill it before we enter
749 * a transaction. This avoids the possibility of
750 * holding up the transaction if the data copy hangs
751 * up on a pagefault (e.g., from an NFS server mapping).
755 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
757 ASSERT(abuf
!= NULL
);
758 ASSERT(arc_buf_size(abuf
) == max_blksz
);
759 if ((error
= uiocopy(abuf
->b_data
, max_blksz
,
760 UIO_WRITE
, uio
, &cbytes
))) {
761 dmu_return_arcbuf(abuf
);
764 ASSERT(cbytes
== max_blksz
);
768 * Start a transaction.
770 tx
= dmu_tx_create(zsb
->z_os
);
771 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
772 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
773 zfs_sa_upgrade_txholds(tx
, zp
);
774 error
= dmu_tx_assign(tx
, TXG_WAIT
);
778 dmu_return_arcbuf(abuf
);
783 * If zfs_range_lock() over-locked we grow the blocksize
784 * and then reduce the lock range. This will only happen
785 * on the first iteration since zfs_range_reduce() will
786 * shrink down r_len to the appropriate size.
788 if (rl
->r_len
== UINT64_MAX
) {
791 if (zp
->z_blksz
> max_blksz
) {
793 * File's blocksize is already larger than the
794 * "recordsize" property. Only let it grow to
795 * the next power of 2.
797 ASSERT(!ISP2(zp
->z_blksz
));
798 new_blksz
= MIN(end_size
,
799 1 << highbit64(zp
->z_blksz
));
801 new_blksz
= MIN(end_size
, max_blksz
);
803 zfs_grow_blocksize(zp
, new_blksz
, tx
);
804 zfs_range_reduce(rl
, woff
, n
);
808 * XXX - should we really limit each write to z_max_blksz?
809 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
811 nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
814 tx_bytes
= uio
->uio_resid
;
815 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
817 tx_bytes
-= uio
->uio_resid
;
820 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
822 * If this is not a full block write, but we are
823 * extending the file past EOF and this data starts
824 * block-aligned, use assign_arcbuf(). Otherwise,
825 * write via dmu_write().
827 if (tx_bytes
< max_blksz
&& (!write_eof
||
828 aiov
->iov_base
!= abuf
->b_data
)) {
830 dmu_write(zsb
->z_os
, zp
->z_id
, woff
,
831 aiov
->iov_len
, aiov
->iov_base
, tx
);
832 dmu_return_arcbuf(abuf
);
833 xuio_stat_wbuf_copied();
835 ASSERT(xuio
|| tx_bytes
== max_blksz
);
836 dmu_assign_arcbuf(sa_get_db(zp
->z_sa_hdl
),
839 ASSERT(tx_bytes
<= uio
->uio_resid
);
840 uioskip(uio
, tx_bytes
);
843 if (tx_bytes
&& zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
))
844 update_pages(ip
, woff
, tx_bytes
, zsb
->z_os
, zp
->z_id
);
847 * If we made no progress, we're done. If we made even
848 * partial progress, update the znode and ZIL accordingly.
851 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
852 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
859 * Clear Set-UID/Set-GID bits on successful write if not
860 * privileged and at least one of the excute bits is set.
862 * It would be nice to to this after all writes have
863 * been done, but that would still expose the ISUID/ISGID
864 * to another app after the partial write is committed.
866 * Note: we don't call zfs_fuid_map_id() here because
867 * user 0 is not an ephemeral uid.
869 mutex_enter(&zp
->z_acl_lock
);
870 uid
= KUID_TO_SUID(ip
->i_uid
);
871 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
872 (S_IXUSR
>> 6))) != 0 &&
873 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
874 secpolicy_vnode_setid_retain(cr
,
875 ((zp
->z_mode
& S_ISUID
) != 0 && uid
== 0)) != 0) {
877 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
878 ip
->i_mode
= newmode
= zp
->z_mode
;
879 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zsb
),
880 (void *)&newmode
, sizeof (uint64_t), tx
);
882 mutex_exit(&zp
->z_acl_lock
);
884 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
);
887 * Update the file size (zp_size) if it has changed;
888 * account for possible concurrent updates.
890 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
891 (void) atomic_cas_64(&zp
->z_size
, end_size
,
896 * If we are replaying and eof is non zero then force
897 * the file size to the specified eof. Note, there's no
898 * concurrency during replay.
900 if (zsb
->z_replay
&& zsb
->z_replay_eof
!= 0)
901 zp
->z_size
= zsb
->z_replay_eof
;
903 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
905 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
,
911 ASSERT(tx_bytes
== nbytes
);
915 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
918 zfs_inode_update(zp
);
919 zfs_range_unlock(rl
);
922 * If we're in replay mode, or we made no progress, return error.
923 * Otherwise, it's at least a partial write, so it's successful.
925 if (zsb
->z_replay
|| uio
->uio_resid
== start_resid
) {
930 if (ioflag
& (FSYNC
| FDSYNC
) ||
931 zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
932 zil_commit(zilog
, zp
->z_id
);
937 EXPORT_SYMBOL(zfs_write
);
940 * Drop a reference on the passed inode asynchronously. This ensures
941 * that the caller will never drop the last reference on an inode in
942 * the current context. Doing so while holding open a tx could result
943 * in a deadlock if iput_final() re-enters the filesystem code.
946 zfs_iput_async(struct inode
*ip
)
948 objset_t
*os
= ITOZSB(ip
)->z_os
;
950 ASSERT(atomic_read(&ip
->i_count
) > 0);
953 if (atomic_read(&ip
->i_count
) == 1)
954 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os
)),
955 (task_func_t
*)iput
, ip
, TQ_SLEEP
) != TASKQID_INVALID
);
961 zfs_get_done(zgd_t
*zgd
, int error
)
963 znode_t
*zp
= zgd
->zgd_private
;
966 dmu_buf_rele(zgd
->zgd_db
, zgd
);
968 zfs_range_unlock(zgd
->zgd_rl
);
971 * Release the vnode asynchronously as we currently have the
972 * txg stopped from syncing.
974 zfs_iput_async(ZTOI(zp
));
976 if (error
== 0 && zgd
->zgd_bp
)
977 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
979 kmem_free(zgd
, sizeof (zgd_t
));
983 static int zil_fault_io
= 0;
987 * Get data to generate a TX_WRITE intent log record.
990 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
993 objset_t
*os
= zsb
->z_os
;
995 uint64_t object
= lr
->lr_foid
;
996 uint64_t offset
= lr
->lr_offset
;
997 uint64_t size
= lr
->lr_length
;
998 blkptr_t
*bp
= &lr
->lr_blkptr
;
1003 ASSERT(zio
!= NULL
);
1007 * Nothing to do if the file has been removed
1009 if (zfs_zget(zsb
, object
, &zp
) != 0)
1010 return (SET_ERROR(ENOENT
));
1011 if (zp
->z_unlinked
) {
1013 * Release the vnode asynchronously as we currently have the
1014 * txg stopped from syncing.
1016 zfs_iput_async(ZTOI(zp
));
1017 return (SET_ERROR(ENOENT
));
1020 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1021 zgd
->zgd_zilog
= zsb
->z_log
;
1022 zgd
->zgd_private
= zp
;
1025 * Write records come in two flavors: immediate and indirect.
1026 * For small writes it's cheaper to store the data with the
1027 * log record (immediate); for large writes it's cheaper to
1028 * sync the data and get a pointer to it (indirect) so that
1029 * we don't have to write the data twice.
1031 if (buf
!= NULL
) { /* immediate write */
1032 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
, size
,
1034 /* test for truncation needs to be done while range locked */
1035 if (offset
>= zp
->z_size
) {
1036 error
= SET_ERROR(ENOENT
);
1038 error
= dmu_read(os
, object
, offset
, size
, buf
,
1039 DMU_READ_NO_PREFETCH
);
1041 ASSERT(error
== 0 || error
== ENOENT
);
1042 } else { /* indirect write */
1044 * Have to lock the whole block to ensure when it's
1045 * written out and it's checksum is being calculated
1046 * that no one can change the data. We need to re-check
1047 * blocksize after we get the lock in case it's changed!
1052 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1054 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
,
1056 if (zp
->z_blksz
== size
)
1059 zfs_range_unlock(zgd
->zgd_rl
);
1061 /* test for truncation needs to be done while range locked */
1062 if (lr
->lr_offset
>= zp
->z_size
)
1063 error
= SET_ERROR(ENOENT
);
1066 error
= SET_ERROR(EIO
);
1071 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1072 DMU_READ_NO_PREFETCH
);
1075 blkptr_t
*obp
= dmu_buf_get_blkptr(db
);
1077 ASSERT(BP_IS_HOLE(bp
));
1084 ASSERT(db
->db_offset
== offset
);
1085 ASSERT(db
->db_size
== size
);
1087 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1089 ASSERT(error
|| lr
->lr_length
<= zp
->z_blksz
);
1092 * On success, we need to wait for the write I/O
1093 * initiated by dmu_sync() to complete before we can
1094 * release this dbuf. We will finish everything up
1095 * in the zfs_get_done() callback.
1100 if (error
== EALREADY
) {
1101 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
1107 zfs_get_done(zgd
, error
);
1114 zfs_access(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
1116 znode_t
*zp
= ITOZ(ip
);
1117 zfs_sb_t
*zsb
= ITOZSB(ip
);
1123 if (flag
& V_ACE_MASK
)
1124 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1126 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1131 EXPORT_SYMBOL(zfs_access
);
1134 * Lookup an entry in a directory, or an extended attribute directory.
1135 * If it exists, return a held inode reference for it.
1137 * IN: dip - inode of directory to search.
1138 * nm - name of entry to lookup.
1139 * flags - LOOKUP_XATTR set if looking for an attribute.
1140 * cr - credentials of caller.
1141 * direntflags - directory lookup flags
1142 * realpnp - returned pathname.
1144 * OUT: ipp - inode of located entry, NULL if not found.
1146 * RETURN: 0 on success, error code on failure.
1153 zfs_lookup(struct inode
*dip
, char *nm
, struct inode
**ipp
, int flags
,
1154 cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1156 znode_t
*zdp
= ITOZ(dip
);
1157 zfs_sb_t
*zsb
= ITOZSB(dip
);
1161 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1163 if (!S_ISDIR(dip
->i_mode
)) {
1164 return (SET_ERROR(ENOTDIR
));
1165 } else if (zdp
->z_sa_hdl
== NULL
) {
1166 return (SET_ERROR(EIO
));
1169 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1170 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1179 vnode_t
*tvp
= dnlc_lookup(dvp
, nm
);
1182 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1187 if (tvp
== DNLC_NO_VNODE
) {
1189 return (SET_ERROR(ENOENT
));
1192 return (specvp_check(vpp
, cr
));
1195 #endif /* HAVE_DNLC */
1204 if (flags
& LOOKUP_XATTR
) {
1206 * We don't allow recursive attributes..
1207 * Maybe someday we will.
1209 if (zdp
->z_pflags
& ZFS_XATTR
) {
1211 return (SET_ERROR(EINVAL
));
1214 if ((error
= zfs_get_xattrdir(zdp
, ipp
, cr
, flags
))) {
1220 * Do we have permission to get into attribute directory?
1223 if ((error
= zfs_zaccess(ITOZ(*ipp
), ACE_EXECUTE
, 0,
1233 if (!S_ISDIR(dip
->i_mode
)) {
1235 return (SET_ERROR(ENOTDIR
));
1239 * Check accessibility of directory.
1242 if ((error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
))) {
1247 if (zsb
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1248 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1250 return (SET_ERROR(EILSEQ
));
1253 error
= zfs_dirlook(zdp
, nm
, ipp
, flags
, direntflags
, realpnp
);
1254 if ((error
== 0) && (*ipp
))
1255 zfs_inode_update(ITOZ(*ipp
));
1260 EXPORT_SYMBOL(zfs_lookup
);
1263 * Attempt to create a new entry in a directory. If the entry
1264 * already exists, truncate the file if permissible, else return
1265 * an error. Return the ip of the created or trunc'd file.
1267 * IN: dip - inode of directory to put new file entry in.
1268 * name - name of new file entry.
1269 * vap - attributes of new file.
1270 * excl - flag indicating exclusive or non-exclusive mode.
1271 * mode - mode to open file with.
1272 * cr - credentials of caller.
1273 * flag - large file flag [UNUSED].
1274 * vsecp - ACL to be set
1276 * OUT: ipp - inode of created or trunc'd entry.
1278 * RETURN: 0 on success, error code on failure.
1281 * dip - ctime|mtime updated if new entry created
1282 * ip - ctime|mtime always, atime if new
1287 zfs_create(struct inode
*dip
, char *name
, vattr_t
*vap
, int excl
,
1288 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1290 znode_t
*zp
, *dzp
= ITOZ(dip
);
1291 zfs_sb_t
*zsb
= ITOZSB(dip
);
1299 zfs_acl_ids_t acl_ids
;
1300 boolean_t fuid_dirtied
;
1301 boolean_t have_acl
= B_FALSE
;
1302 boolean_t waited
= B_FALSE
;
1305 * If we have an ephemeral id, ACL, or XVATTR then
1306 * make sure file system is at proper version
1312 if (zsb
->z_use_fuids
== B_FALSE
&&
1313 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1314 return (SET_ERROR(EINVAL
));
1317 return (SET_ERROR(EINVAL
));
1324 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
1325 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1327 return (SET_ERROR(EILSEQ
));
1330 if (vap
->va_mask
& ATTR_XVATTR
) {
1331 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1332 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1340 if (*name
== '\0') {
1342 * Null component name refers to the directory itself.
1349 /* possible igrab(zp) */
1352 if (flag
& FIGNORECASE
)
1355 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1359 zfs_acl_ids_free(&acl_ids
);
1360 if (strcmp(name
, "..") == 0)
1361 error
= SET_ERROR(EISDIR
);
1371 * Create a new file object and update the directory
1374 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1376 zfs_acl_ids_free(&acl_ids
);
1381 * We only support the creation of regular files in
1382 * extended attribute directories.
1385 if ((dzp
->z_pflags
& ZFS_XATTR
) && !S_ISREG(vap
->va_mode
)) {
1387 zfs_acl_ids_free(&acl_ids
);
1388 error
= SET_ERROR(EINVAL
);
1392 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1393 cr
, vsecp
, &acl_ids
)) != 0)
1397 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1398 zfs_acl_ids_free(&acl_ids
);
1399 error
= SET_ERROR(EDQUOT
);
1403 tx
= dmu_tx_create(os
);
1405 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1406 ZFS_SA_BASE_ATTR_SIZE
);
1408 fuid_dirtied
= zsb
->z_fuid_dirty
;
1410 zfs_fuid_txhold(zsb
, tx
);
1411 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1412 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1413 if (!zsb
->z_use_sa
&&
1414 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1415 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1416 0, acl_ids
.z_aclp
->z_acl_bytes
);
1418 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1420 zfs_dirent_unlock(dl
);
1421 if (error
== ERESTART
) {
1427 zfs_acl_ids_free(&acl_ids
);
1432 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1435 zfs_fuid_sync(zsb
, tx
);
1437 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1438 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1439 if (flag
& FIGNORECASE
)
1441 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1442 vsecp
, acl_ids
.z_fuidp
, vap
);
1443 zfs_acl_ids_free(&acl_ids
);
1446 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1449 zfs_acl_ids_free(&acl_ids
);
1453 * A directory entry already exists for this name.
1456 * Can't truncate an existing file if in exclusive mode.
1459 error
= SET_ERROR(EEXIST
);
1463 * Can't open a directory for writing.
1465 if (S_ISDIR(ZTOI(zp
)->i_mode
)) {
1466 error
= SET_ERROR(EISDIR
);
1470 * Verify requested access to file.
1472 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1476 mutex_enter(&dzp
->z_lock
);
1478 mutex_exit(&dzp
->z_lock
);
1481 * Truncate regular files if requested.
1483 if (S_ISREG(ZTOI(zp
)->i_mode
) &&
1484 (vap
->va_mask
& ATTR_SIZE
) && (vap
->va_size
== 0)) {
1485 /* we can't hold any locks when calling zfs_freesp() */
1487 zfs_dirent_unlock(dl
);
1490 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1496 zfs_dirent_unlock(dl
);
1502 zfs_inode_update(dzp
);
1503 zfs_inode_update(zp
);
1507 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1508 zil_commit(zilog
, 0);
1513 EXPORT_SYMBOL(zfs_create
);
1517 zfs_tmpfile(struct inode
*dip
, vattr_t
*vap
, int excl
,
1518 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1520 znode_t
*zp
= NULL
, *dzp
= ITOZ(dip
);
1521 zfs_sb_t
*zsb
= ITOZSB(dip
);
1527 zfs_acl_ids_t acl_ids
;
1528 boolean_t fuid_dirtied
;
1529 boolean_t have_acl
= B_FALSE
;
1530 boolean_t waited
= B_FALSE
;
1533 * If we have an ephemeral id, ACL, or XVATTR then
1534 * make sure file system is at proper version
1540 if (zsb
->z_use_fuids
== B_FALSE
&&
1541 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1542 return (SET_ERROR(EINVAL
));
1548 if (vap
->va_mask
& ATTR_XVATTR
) {
1549 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1550 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1560 * Create a new file object and update the directory
1563 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1565 zfs_acl_ids_free(&acl_ids
);
1569 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1570 cr
, vsecp
, &acl_ids
)) != 0)
1574 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1575 zfs_acl_ids_free(&acl_ids
);
1576 error
= SET_ERROR(EDQUOT
);
1580 tx
= dmu_tx_create(os
);
1582 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1583 ZFS_SA_BASE_ATTR_SIZE
);
1584 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
1586 fuid_dirtied
= zsb
->z_fuid_dirty
;
1588 zfs_fuid_txhold(zsb
, tx
);
1589 if (!zsb
->z_use_sa
&&
1590 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1591 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1592 0, acl_ids
.z_aclp
->z_acl_bytes
);
1594 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1596 if (error
== ERESTART
) {
1602 zfs_acl_ids_free(&acl_ids
);
1607 zfs_mknode(dzp
, vap
, tx
, cr
, IS_TMPFILE
, &zp
, &acl_ids
);
1610 zfs_fuid_sync(zsb
, tx
);
1612 /* Add to unlinked set */
1614 zfs_unlinked_add(zp
, tx
);
1615 zfs_acl_ids_free(&acl_ids
);
1623 zfs_inode_update(dzp
);
1624 zfs_inode_update(zp
);
1633 * Remove an entry from a directory.
1635 * IN: dip - inode of directory to remove entry from.
1636 * name - name of entry to remove.
1637 * cr - credentials of caller.
1639 * RETURN: 0 if success
1640 * error code if failure
1644 * ip - ctime (if nlink > 0)
1647 uint64_t null_xattr
= 0;
1651 zfs_remove(struct inode
*dip
, char *name
, cred_t
*cr
, int flags
)
1653 znode_t
*zp
, *dzp
= ITOZ(dip
);
1656 zfs_sb_t
*zsb
= ITOZSB(dip
);
1658 uint64_t acl_obj
, xattr_obj
;
1659 uint64_t xattr_obj_unlinked
= 0;
1664 boolean_t may_delete_now
, delete_now
= FALSE
;
1665 boolean_t unlinked
, toobig
= FALSE
;
1667 pathname_t
*realnmp
= NULL
;
1671 boolean_t waited
= B_FALSE
;
1674 return (SET_ERROR(EINVAL
));
1680 if (flags
& FIGNORECASE
) {
1690 * Attempt to lock directory; fail if entry doesn't exist.
1692 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1702 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1707 * Need to use rmdir for removing directories.
1709 if (S_ISDIR(ip
->i_mode
)) {
1710 error
= SET_ERROR(EPERM
);
1716 dnlc_remove(dvp
, realnmp
->pn_buf
);
1718 dnlc_remove(dvp
, name
);
1719 #endif /* HAVE_DNLC */
1721 mutex_enter(&zp
->z_lock
);
1722 may_delete_now
= atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
);
1723 mutex_exit(&zp
->z_lock
);
1726 * We may delete the znode now, or we may put it in the unlinked set;
1727 * it depends on whether we're the last link, and on whether there are
1728 * other holds on the inode. So we dmu_tx_hold() the right things to
1729 * allow for either case.
1732 tx
= dmu_tx_create(zsb
->z_os
);
1733 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1734 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1735 zfs_sa_upgrade_txholds(tx
, zp
);
1736 zfs_sa_upgrade_txholds(tx
, dzp
);
1737 if (may_delete_now
) {
1738 toobig
= zp
->z_size
> zp
->z_blksz
* zfs_delete_blocks
;
1739 /* if the file is too big, only hold_free a token amount */
1740 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1741 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1744 /* are there any extended attributes? */
1745 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1746 &xattr_obj
, sizeof (xattr_obj
));
1747 if (error
== 0 && xattr_obj
) {
1748 error
= zfs_zget(zsb
, xattr_obj
, &xzp
);
1750 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1751 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1754 mutex_enter(&zp
->z_lock
);
1755 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1756 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1757 mutex_exit(&zp
->z_lock
);
1759 /* charge as an update -- would be nice not to charge at all */
1760 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
1763 * Mark this transaction as typically resulting in a net free of space
1765 dmu_tx_mark_netfree(tx
);
1767 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1769 zfs_dirent_unlock(dl
);
1773 if (error
== ERESTART
) {
1787 * Remove the directory entry.
1789 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1798 * Hold z_lock so that we can make sure that the ACL obj
1799 * hasn't changed. Could have been deleted due to
1802 mutex_enter(&zp
->z_lock
);
1803 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1804 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1805 delete_now
= may_delete_now
&& !toobig
&&
1806 atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
) &&
1807 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1812 if (xattr_obj_unlinked
) {
1813 ASSERT3U(ZTOI(xzp
)->i_nlink
, ==, 2);
1814 mutex_enter(&xzp
->z_lock
);
1815 xzp
->z_unlinked
= 1;
1816 clear_nlink(ZTOI(xzp
));
1818 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zsb
),
1819 &links
, sizeof (links
), tx
);
1820 ASSERT3U(error
, ==, 0);
1821 mutex_exit(&xzp
->z_lock
);
1822 zfs_unlinked_add(xzp
, tx
);
1825 error
= sa_remove(zp
->z_sa_hdl
,
1826 SA_ZPL_XATTR(zsb
), tx
);
1828 error
= sa_update(zp
->z_sa_hdl
,
1829 SA_ZPL_XATTR(zsb
), &null_xattr
,
1830 sizeof (uint64_t), tx
);
1834 * Add to the unlinked set because a new reference could be
1835 * taken concurrently resulting in a deferred destruction.
1837 zfs_unlinked_add(zp
, tx
);
1838 mutex_exit(&zp
->z_lock
);
1839 zfs_inode_update(zp
);
1841 } else if (unlinked
) {
1842 mutex_exit(&zp
->z_lock
);
1843 zfs_unlinked_add(zp
, tx
);
1847 if (flags
& FIGNORECASE
)
1849 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1856 zfs_dirent_unlock(dl
);
1857 zfs_inode_update(dzp
);
1860 zfs_inode_update(zp
);
1865 zfs_inode_update(xzp
);
1866 zfs_iput_async(ZTOI(xzp
));
1869 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1870 zil_commit(zilog
, 0);
1875 EXPORT_SYMBOL(zfs_remove
);
1878 * Create a new directory and insert it into dip using the name
1879 * provided. Return a pointer to the inserted directory.
1881 * IN: dip - inode of directory to add subdir to.
1882 * dirname - name of new directory.
1883 * vap - attributes of new directory.
1884 * cr - credentials of caller.
1885 * vsecp - ACL to be set
1887 * OUT: ipp - inode of created directory.
1889 * RETURN: 0 if success
1890 * error code if failure
1893 * dip - ctime|mtime updated
1894 * ipp - ctime|mtime|atime updated
1898 zfs_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
, struct inode
**ipp
,
1899 cred_t
*cr
, int flags
, vsecattr_t
*vsecp
)
1901 znode_t
*zp
, *dzp
= ITOZ(dip
);
1902 zfs_sb_t
*zsb
= ITOZSB(dip
);
1910 gid_t gid
= crgetgid(cr
);
1911 zfs_acl_ids_t acl_ids
;
1912 boolean_t fuid_dirtied
;
1913 boolean_t waited
= B_FALSE
;
1915 ASSERT(S_ISDIR(vap
->va_mode
));
1918 * If we have an ephemeral id, ACL, or XVATTR then
1919 * make sure file system is at proper version
1923 if (zsb
->z_use_fuids
== B_FALSE
&&
1924 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1925 return (SET_ERROR(EINVAL
));
1927 if (dirname
== NULL
)
1928 return (SET_ERROR(EINVAL
));
1934 if (dzp
->z_pflags
& ZFS_XATTR
) {
1936 return (SET_ERROR(EINVAL
));
1939 if (zsb
->z_utf8
&& u8_validate(dirname
,
1940 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1942 return (SET_ERROR(EILSEQ
));
1944 if (flags
& FIGNORECASE
)
1947 if (vap
->va_mask
& ATTR_XVATTR
) {
1948 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1949 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1955 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1956 vsecp
, &acl_ids
)) != 0) {
1961 * First make sure the new directory doesn't exist.
1963 * Existence is checked first to make sure we don't return
1964 * EACCES instead of EEXIST which can cause some applications
1970 if ((error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1972 zfs_acl_ids_free(&acl_ids
);
1977 if ((error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
))) {
1978 zfs_acl_ids_free(&acl_ids
);
1979 zfs_dirent_unlock(dl
);
1984 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1985 zfs_acl_ids_free(&acl_ids
);
1986 zfs_dirent_unlock(dl
);
1988 return (SET_ERROR(EDQUOT
));
1992 * Add a new entry to the directory.
1994 tx
= dmu_tx_create(zsb
->z_os
);
1995 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
1996 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1997 fuid_dirtied
= zsb
->z_fuid_dirty
;
1999 zfs_fuid_txhold(zsb
, tx
);
2000 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
2001 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
2002 acl_ids
.z_aclp
->z_acl_bytes
);
2005 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
2006 ZFS_SA_BASE_ATTR_SIZE
);
2008 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
2010 zfs_dirent_unlock(dl
);
2011 if (error
== ERESTART
) {
2017 zfs_acl_ids_free(&acl_ids
);
2026 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
2029 zfs_fuid_sync(zsb
, tx
);
2032 * Now put new name in parent dir.
2034 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
2038 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
2039 if (flags
& FIGNORECASE
)
2041 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
2042 acl_ids
.z_fuidp
, vap
);
2044 zfs_acl_ids_free(&acl_ids
);
2048 zfs_dirent_unlock(dl
);
2050 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2051 zil_commit(zilog
, 0);
2053 zfs_inode_update(dzp
);
2054 zfs_inode_update(zp
);
2058 EXPORT_SYMBOL(zfs_mkdir
);
2061 * Remove a directory subdir entry. If the current working
2062 * directory is the same as the subdir to be removed, the
2065 * IN: dip - inode of directory to remove from.
2066 * name - name of directory to be removed.
2067 * cwd - inode of current working directory.
2068 * cr - credentials of caller.
2069 * flags - case flags
2071 * RETURN: 0 on success, error code on failure.
2074 * dip - ctime|mtime updated
2078 zfs_rmdir(struct inode
*dip
, char *name
, struct inode
*cwd
, cred_t
*cr
,
2081 znode_t
*dzp
= ITOZ(dip
);
2084 zfs_sb_t
*zsb
= ITOZSB(dip
);
2090 boolean_t waited
= B_FALSE
;
2093 return (SET_ERROR(EINVAL
));
2099 if (flags
& FIGNORECASE
)
2105 * Attempt to lock directory; fail if entry doesn't exist.
2107 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
2115 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
2119 if (!S_ISDIR(ip
->i_mode
)) {
2120 error
= SET_ERROR(ENOTDIR
);
2125 error
= SET_ERROR(EINVAL
);
2130 * Grab a lock on the directory to make sure that noone is
2131 * trying to add (or lookup) entries while we are removing it.
2133 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
2136 * Grab a lock on the parent pointer to make sure we play well
2137 * with the treewalk and directory rename code.
2139 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
2141 tx
= dmu_tx_create(zsb
->z_os
);
2142 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2143 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2144 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
2145 zfs_sa_upgrade_txholds(tx
, zp
);
2146 zfs_sa_upgrade_txholds(tx
, dzp
);
2147 dmu_tx_mark_netfree(tx
);
2148 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
2150 rw_exit(&zp
->z_parent_lock
);
2151 rw_exit(&zp
->z_name_lock
);
2152 zfs_dirent_unlock(dl
);
2154 if (error
== ERESTART
) {
2165 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2168 uint64_t txtype
= TX_RMDIR
;
2169 if (flags
& FIGNORECASE
)
2171 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2176 rw_exit(&zp
->z_parent_lock
);
2177 rw_exit(&zp
->z_name_lock
);
2179 zfs_dirent_unlock(dl
);
2181 zfs_inode_update(dzp
);
2182 zfs_inode_update(zp
);
2185 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2186 zil_commit(zilog
, 0);
2191 EXPORT_SYMBOL(zfs_rmdir
);
2194 * Read as many directory entries as will fit into the provided
2195 * dirent buffer from the given directory cursor position.
2197 * IN: ip - inode of directory to read.
2198 * dirent - buffer for directory entries.
2200 * OUT: dirent - filler buffer of directory entries.
2202 * RETURN: 0 if success
2203 * error code if failure
2206 * ip - atime updated
2208 * Note that the low 4 bits of the cookie returned by zap is always zero.
2209 * This allows us to use the low range for "special" directory entries:
2210 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2211 * we use the offset 2 for the '.zfs' directory.
2215 zfs_readdir(struct inode
*ip
, struct dir_context
*ctx
, cred_t
*cr
)
2217 znode_t
*zp
= ITOZ(ip
);
2218 zfs_sb_t
*zsb
= ITOZSB(ip
);
2221 zap_attribute_t zap
;
2227 uint64_t offset
; /* must be unsigned; checks for < 1 */
2232 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
2233 &parent
, sizeof (parent
))) != 0)
2237 * Quit if directory has been removed (posix)
2245 prefetch
= zp
->z_zn_prefetch
;
2248 * Initialize the iterator cursor.
2252 * Start iteration from the beginning of the directory.
2254 zap_cursor_init(&zc
, os
, zp
->z_id
);
2257 * The offset is a serialized cursor.
2259 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2263 * Transform to file-system independent format
2268 * Special case `.', `..', and `.zfs'.
2271 (void) strcpy(zap
.za_name
, ".");
2272 zap
.za_normalization_conflict
= 0;
2275 } else if (offset
== 1) {
2276 (void) strcpy(zap
.za_name
, "..");
2277 zap
.za_normalization_conflict
= 0;
2280 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2281 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2282 zap
.za_normalization_conflict
= 0;
2283 objnum
= ZFSCTL_INO_ROOT
;
2289 if ((error
= zap_cursor_retrieve(&zc
, &zap
))) {
2290 if (error
== ENOENT
)
2297 * Allow multiple entries provided the first entry is
2298 * the object id. Non-zpl consumers may safely make
2299 * use of the additional space.
2301 * XXX: This should be a feature flag for compatibility
2303 if (zap
.za_integer_length
!= 8 ||
2304 zap
.za_num_integers
== 0) {
2305 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2306 "entry, obj = %lld, offset = %lld, "
2307 "length = %d, num = %lld\n",
2308 (u_longlong_t
)zp
->z_id
,
2309 (u_longlong_t
)offset
,
2310 zap
.za_integer_length
,
2311 (u_longlong_t
)zap
.za_num_integers
);
2312 error
= SET_ERROR(ENXIO
);
2316 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2317 type
= ZFS_DIRENT_TYPE(zap
.za_first_integer
);
2320 done
= !dir_emit(ctx
, zap
.za_name
, strlen(zap
.za_name
),
2325 /* Prefetch znode */
2327 dmu_prefetch(os
, objnum
, 0, 0, 0,
2328 ZIO_PRIORITY_SYNC_READ
);
2332 * Move to the next entry, fill in the previous offset.
2334 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2335 zap_cursor_advance(&zc
);
2336 offset
= zap_cursor_serialize(&zc
);
2342 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2345 zap_cursor_fini(&zc
);
2346 if (error
== ENOENT
)
2353 EXPORT_SYMBOL(zfs_readdir
);
2355 ulong_t zfs_fsync_sync_cnt
= 4;
2358 zfs_fsync(struct inode
*ip
, int syncflag
, cred_t
*cr
)
2360 znode_t
*zp
= ITOZ(ip
);
2361 zfs_sb_t
*zsb
= ITOZSB(ip
);
2363 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2365 if (zsb
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2368 zil_commit(zsb
->z_log
, zp
->z_id
);
2371 tsd_set(zfs_fsyncer_key
, NULL
);
2375 EXPORT_SYMBOL(zfs_fsync
);
2379 * Get the requested file attributes and place them in the provided
2382 * IN: ip - inode of file.
2383 * vap - va_mask identifies requested attributes.
2384 * If ATTR_XVATTR set, then optional attrs are requested
2385 * flags - ATTR_NOACLCHECK (CIFS server context)
2386 * cr - credentials of caller.
2388 * OUT: vap - attribute values.
2390 * RETURN: 0 (always succeeds)
2394 zfs_getattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2396 znode_t
*zp
= ITOZ(ip
);
2397 zfs_sb_t
*zsb
= ITOZSB(ip
);
2400 uint64_t atime
[2], mtime
[2], ctime
[2];
2401 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2402 xoptattr_t
*xoap
= NULL
;
2403 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2404 sa_bulk_attr_t bulk
[3];
2410 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2412 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zsb
), NULL
, &atime
, 16);
2413 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
2414 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
2416 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2422 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2423 * Also, if we are the owner don't bother, since owner should
2424 * always be allowed to read basic attributes of file.
2426 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2427 (vap
->va_uid
!= crgetuid(cr
))) {
2428 if ((error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2436 * Return all attributes. It's cheaper to provide the answer
2437 * than to determine whether we were asked the question.
2440 mutex_enter(&zp
->z_lock
);
2441 vap
->va_type
= vn_mode_to_vtype(zp
->z_mode
);
2442 vap
->va_mode
= zp
->z_mode
;
2443 vap
->va_fsid
= ZTOI(zp
)->i_sb
->s_dev
;
2444 vap
->va_nodeid
= zp
->z_id
;
2445 if ((zp
->z_id
== zsb
->z_root
) && zfs_show_ctldir(zp
))
2446 links
= ZTOI(zp
)->i_nlink
+ 1;
2448 links
= ZTOI(zp
)->i_nlink
;
2449 vap
->va_nlink
= MIN(links
, ZFS_LINK_MAX
);
2450 vap
->va_size
= i_size_read(ip
);
2451 vap
->va_rdev
= ip
->i_rdev
;
2452 vap
->va_seq
= ip
->i_generation
;
2455 * Add in any requested optional attributes and the create time.
2456 * Also set the corresponding bits in the returned attribute bitmap.
2458 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zsb
->z_use_fuids
) {
2459 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2461 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2462 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2465 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2466 xoap
->xoa_readonly
=
2467 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2468 XVA_SET_RTN(xvap
, XAT_READONLY
);
2471 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2473 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2474 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2477 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2479 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2480 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2483 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2484 xoap
->xoa_nounlink
=
2485 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2486 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2489 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2490 xoap
->xoa_immutable
=
2491 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2492 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2495 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2496 xoap
->xoa_appendonly
=
2497 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2498 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2501 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2503 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2504 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2507 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2509 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2510 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2513 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2514 xoap
->xoa_av_quarantined
=
2515 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2516 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2519 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2520 xoap
->xoa_av_modified
=
2521 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2522 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2525 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2526 S_ISREG(ip
->i_mode
)) {
2527 zfs_sa_get_scanstamp(zp
, xvap
);
2530 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2533 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zsb
),
2534 times
, sizeof (times
));
2535 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2536 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2539 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2540 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2541 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2543 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2544 xoap
->xoa_generation
= ip
->i_generation
;
2545 XVA_SET_RTN(xvap
, XAT_GEN
);
2548 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2550 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2551 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2554 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2556 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2557 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2561 ZFS_TIME_DECODE(&vap
->va_atime
, atime
);
2562 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2563 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2565 mutex_exit(&zp
->z_lock
);
2567 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2569 if (zp
->z_blksz
== 0) {
2571 * Block size hasn't been set; suggest maximal I/O transfers.
2573 vap
->va_blksize
= zsb
->z_max_blksz
;
2579 EXPORT_SYMBOL(zfs_getattr
);
2582 * Get the basic file attributes and place them in the provided kstat
2583 * structure. The inode is assumed to be the authoritative source
2584 * for most of the attributes. However, the znode currently has the
2585 * authoritative atime, blksize, and block count.
2587 * IN: ip - inode of file.
2589 * OUT: sp - kstat values.
2591 * RETURN: 0 (always succeeds)
2595 zfs_getattr_fast(struct inode
*ip
, struct kstat
*sp
)
2597 znode_t
*zp
= ITOZ(ip
);
2598 zfs_sb_t
*zsb
= ITOZSB(ip
);
2600 u_longlong_t nblocks
;
2605 mutex_enter(&zp
->z_lock
);
2607 generic_fillattr(ip
, sp
);
2609 sa_object_size(zp
->z_sa_hdl
, &blksize
, &nblocks
);
2610 sp
->blksize
= blksize
;
2611 sp
->blocks
= nblocks
;
2613 if (unlikely(zp
->z_blksz
== 0)) {
2615 * Block size hasn't been set; suggest maximal I/O transfers.
2617 sp
->blksize
= zsb
->z_max_blksz
;
2620 mutex_exit(&zp
->z_lock
);
2623 * Required to prevent NFS client from detecting different inode
2624 * numbers of snapshot root dentry before and after snapshot mount.
2626 if (zsb
->z_issnap
) {
2627 if (ip
->i_sb
->s_root
->d_inode
== ip
)
2628 sp
->ino
= ZFSCTL_INO_SNAPDIRS
-
2629 dmu_objset_id(zsb
->z_os
);
2636 EXPORT_SYMBOL(zfs_getattr_fast
);
2639 * Set the file attributes to the values contained in the
2642 * IN: ip - inode of file to be modified.
2643 * vap - new attribute values.
2644 * If ATTR_XVATTR set, then optional attrs are being set
2645 * flags - ATTR_UTIME set if non-default time values provided.
2646 * - ATTR_NOACLCHECK (CIFS context only).
2647 * cr - credentials of caller.
2649 * RETURN: 0 if success
2650 * error code if failure
2653 * ip - ctime updated, mtime updated if size changed.
2657 zfs_setattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2659 znode_t
*zp
= ITOZ(ip
);
2660 zfs_sb_t
*zsb
= ITOZSB(ip
);
2664 xvattr_t
*tmpxvattr
;
2665 uint_t mask
= vap
->va_mask
;
2666 uint_t saved_mask
= 0;
2669 uint64_t new_kuid
= 0, new_kgid
= 0, new_uid
, new_gid
;
2671 uint64_t mtime
[2], ctime
[2], atime
[2];
2673 int need_policy
= FALSE
;
2675 zfs_fuid_info_t
*fuidp
= NULL
;
2676 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2679 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2680 boolean_t fuid_dirtied
= B_FALSE
;
2681 sa_bulk_attr_t
*bulk
, *xattr_bulk
;
2682 int count
= 0, xattr_count
= 0;
2693 * Make sure that if we have ephemeral uid/gid or xvattr specified
2694 * that file system is at proper version level
2697 if (zsb
->z_use_fuids
== B_FALSE
&&
2698 (((mask
& ATTR_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2699 ((mask
& ATTR_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2700 (mask
& ATTR_XVATTR
))) {
2702 return (SET_ERROR(EINVAL
));
2705 if (mask
& ATTR_SIZE
&& S_ISDIR(ip
->i_mode
)) {
2707 return (SET_ERROR(EISDIR
));
2710 if (mask
& ATTR_SIZE
&& !S_ISREG(ip
->i_mode
) && !S_ISFIFO(ip
->i_mode
)) {
2712 return (SET_ERROR(EINVAL
));
2716 * If this is an xvattr_t, then get a pointer to the structure of
2717 * optional attributes. If this is NULL, then we have a vattr_t.
2719 xoap
= xva_getxoptattr(xvap
);
2721 tmpxvattr
= kmem_alloc(sizeof (xvattr_t
), KM_SLEEP
);
2722 xva_init(tmpxvattr
);
2724 bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2725 xattr_bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2728 * Immutable files can only alter immutable bit and atime
2730 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2731 ((mask
& (ATTR_SIZE
|ATTR_UID
|ATTR_GID
|ATTR_MTIME
|ATTR_MODE
)) ||
2732 ((mask
& ATTR_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2737 if ((mask
& ATTR_SIZE
) && (zp
->z_pflags
& ZFS_READONLY
)) {
2743 * Verify timestamps doesn't overflow 32 bits.
2744 * ZFS can handle large timestamps, but 32bit syscalls can't
2745 * handle times greater than 2039. This check should be removed
2746 * once large timestamps are fully supported.
2748 if (mask
& (ATTR_ATIME
| ATTR_MTIME
)) {
2749 if (((mask
& ATTR_ATIME
) &&
2750 TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2751 ((mask
& ATTR_MTIME
) &&
2752 TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2762 /* Can this be moved to before the top label? */
2763 if (zfs_is_readonly(zsb
)) {
2769 * First validate permissions
2772 if (mask
& ATTR_SIZE
) {
2773 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2778 * XXX - Note, we are not providing any open
2779 * mode flags here (like FNDELAY), so we may
2780 * block if there are locks present... this
2781 * should be addressed in openat().
2783 /* XXX - would it be OK to generate a log record here? */
2784 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2789 if (mask
& (ATTR_ATIME
|ATTR_MTIME
) ||
2790 ((mask
& ATTR_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2791 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2792 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2793 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2794 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2795 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2796 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2797 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2801 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2802 int idmask
= (mask
& (ATTR_UID
|ATTR_GID
));
2807 * NOTE: even if a new mode is being set,
2808 * we may clear S_ISUID/S_ISGID bits.
2811 if (!(mask
& ATTR_MODE
))
2812 vap
->va_mode
= zp
->z_mode
;
2815 * Take ownership or chgrp to group we are a member of
2818 take_owner
= (mask
& ATTR_UID
) && (vap
->va_uid
== crgetuid(cr
));
2819 take_group
= (mask
& ATTR_GID
) &&
2820 zfs_groupmember(zsb
, vap
->va_gid
, cr
);
2823 * If both ATTR_UID and ATTR_GID are set then take_owner and
2824 * take_group must both be set in order to allow taking
2827 * Otherwise, send the check through secpolicy_vnode_setattr()
2831 if (((idmask
== (ATTR_UID
|ATTR_GID
)) &&
2832 take_owner
&& take_group
) ||
2833 ((idmask
== ATTR_UID
) && take_owner
) ||
2834 ((idmask
== ATTR_GID
) && take_group
)) {
2835 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2836 skipaclchk
, cr
) == 0) {
2838 * Remove setuid/setgid for non-privileged users
2840 (void) secpolicy_setid_clear(vap
, cr
);
2841 trim_mask
= (mask
& (ATTR_UID
|ATTR_GID
));
2850 mutex_enter(&zp
->z_lock
);
2851 oldva
.va_mode
= zp
->z_mode
;
2852 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2853 if (mask
& ATTR_XVATTR
) {
2855 * Update xvattr mask to include only those attributes
2856 * that are actually changing.
2858 * the bits will be restored prior to actually setting
2859 * the attributes so the caller thinks they were set.
2861 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2862 if (xoap
->xoa_appendonly
!=
2863 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2866 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2867 XVA_SET_REQ(tmpxvattr
, XAT_APPENDONLY
);
2871 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2872 if (xoap
->xoa_nounlink
!=
2873 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2876 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2877 XVA_SET_REQ(tmpxvattr
, XAT_NOUNLINK
);
2881 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2882 if (xoap
->xoa_immutable
!=
2883 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2886 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2887 XVA_SET_REQ(tmpxvattr
, XAT_IMMUTABLE
);
2891 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2892 if (xoap
->xoa_nodump
!=
2893 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2896 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2897 XVA_SET_REQ(tmpxvattr
, XAT_NODUMP
);
2901 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2902 if (xoap
->xoa_av_modified
!=
2903 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2906 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2907 XVA_SET_REQ(tmpxvattr
, XAT_AV_MODIFIED
);
2911 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2912 if ((!S_ISREG(ip
->i_mode
) &&
2913 xoap
->xoa_av_quarantined
) ||
2914 xoap
->xoa_av_quarantined
!=
2915 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2918 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2919 XVA_SET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
);
2923 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2924 mutex_exit(&zp
->z_lock
);
2929 if (need_policy
== FALSE
&&
2930 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
2931 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
2936 mutex_exit(&zp
->z_lock
);
2938 if (mask
& ATTR_MODE
) {
2939 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
2940 err
= secpolicy_setid_setsticky_clear(ip
, vap
,
2945 trim_mask
|= ATTR_MODE
;
2953 * If trim_mask is set then take ownership
2954 * has been granted or write_acl is present and user
2955 * has the ability to modify mode. In that case remove
2956 * UID|GID and or MODE from mask so that
2957 * secpolicy_vnode_setattr() doesn't revoke it.
2961 saved_mask
= vap
->va_mask
;
2962 vap
->va_mask
&= ~trim_mask
;
2964 err
= secpolicy_vnode_setattr(cr
, ip
, vap
, &oldva
, flags
,
2965 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
2970 vap
->va_mask
|= saved_mask
;
2974 * secpolicy_vnode_setattr, or take ownership may have
2977 mask
= vap
->va_mask
;
2979 if ((mask
& (ATTR_UID
| ATTR_GID
))) {
2980 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
2981 &xattr_obj
, sizeof (xattr_obj
));
2983 if (err
== 0 && xattr_obj
) {
2984 err
= zfs_zget(ZTOZSB(zp
), xattr_obj
, &attrzp
);
2988 if (mask
& ATTR_UID
) {
2989 new_kuid
= zfs_fuid_create(zsb
,
2990 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
2991 if (new_kuid
!= KUID_TO_SUID(ZTOI(zp
)->i_uid
) &&
2992 zfs_fuid_overquota(zsb
, B_FALSE
, new_kuid
)) {
3000 if (mask
& ATTR_GID
) {
3001 new_kgid
= zfs_fuid_create(zsb
, (uint64_t)vap
->va_gid
,
3002 cr
, ZFS_GROUP
, &fuidp
);
3003 if (new_kgid
!= KGID_TO_SGID(ZTOI(zp
)->i_gid
) &&
3004 zfs_fuid_overquota(zsb
, B_TRUE
, new_kgid
)) {
3012 tx
= dmu_tx_create(zsb
->z_os
);
3014 if (mask
& ATTR_MODE
) {
3015 uint64_t pmode
= zp
->z_mode
;
3017 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
3019 zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
);
3021 mutex_enter(&zp
->z_lock
);
3022 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
3024 * Are we upgrading ACL from old V0 format
3027 if (zsb
->z_version
>= ZPL_VERSION_FUID
&&
3028 zfs_znode_acl_version(zp
) ==
3029 ZFS_ACL_VERSION_INITIAL
) {
3030 dmu_tx_hold_free(tx
, acl_obj
, 0,
3032 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3033 0, aclp
->z_acl_bytes
);
3035 dmu_tx_hold_write(tx
, acl_obj
, 0,
3038 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3039 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3040 0, aclp
->z_acl_bytes
);
3042 mutex_exit(&zp
->z_lock
);
3043 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3045 if ((mask
& ATTR_XVATTR
) &&
3046 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3047 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3049 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3053 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
3056 fuid_dirtied
= zsb
->z_fuid_dirty
;
3058 zfs_fuid_txhold(zsb
, tx
);
3060 zfs_sa_upgrade_txholds(tx
, zp
);
3062 err
= dmu_tx_assign(tx
, TXG_WAIT
);
3068 * Set each attribute requested.
3069 * We group settings according to the locks they need to acquire.
3071 * Note: you cannot set ctime directly, although it will be
3072 * updated as a side-effect of calling this function.
3076 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3077 mutex_enter(&zp
->z_acl_lock
);
3078 mutex_enter(&zp
->z_lock
);
3080 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
3081 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
3084 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3085 mutex_enter(&attrzp
->z_acl_lock
);
3086 mutex_enter(&attrzp
->z_lock
);
3087 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3088 SA_ZPL_FLAGS(zsb
), NULL
, &attrzp
->z_pflags
,
3089 sizeof (attrzp
->z_pflags
));
3092 if (mask
& (ATTR_UID
|ATTR_GID
)) {
3094 if (mask
& ATTR_UID
) {
3095 ZTOI(zp
)->i_uid
= SUID_TO_KUID(new_kuid
);
3096 new_uid
= zfs_uid_read(ZTOI(zp
));
3097 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zsb
), NULL
,
3098 &new_uid
, sizeof (new_uid
));
3100 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3101 SA_ZPL_UID(zsb
), NULL
, &new_uid
,
3103 ZTOI(attrzp
)->i_uid
= SUID_TO_KUID(new_uid
);
3107 if (mask
& ATTR_GID
) {
3108 ZTOI(zp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3109 new_gid
= zfs_gid_read(ZTOI(zp
));
3110 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zsb
),
3111 NULL
, &new_gid
, sizeof (new_gid
));
3113 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3114 SA_ZPL_GID(zsb
), NULL
, &new_gid
,
3116 ZTOI(attrzp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3119 if (!(mask
& ATTR_MODE
)) {
3120 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
),
3121 NULL
, &new_mode
, sizeof (new_mode
));
3122 new_mode
= zp
->z_mode
;
3124 err
= zfs_acl_chown_setattr(zp
);
3127 err
= zfs_acl_chown_setattr(attrzp
);
3132 if (mask
& ATTR_MODE
) {
3133 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
), NULL
,
3134 &new_mode
, sizeof (new_mode
));
3135 zp
->z_mode
= ZTOI(zp
)->i_mode
= new_mode
;
3136 ASSERT3P(aclp
, !=, NULL
);
3137 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
3139 if (zp
->z_acl_cached
)
3140 zfs_acl_free(zp
->z_acl_cached
);
3141 zp
->z_acl_cached
= aclp
;
3145 if ((mask
& ATTR_ATIME
) || zp
->z_atime_dirty
) {
3146 zp
->z_atime_dirty
= 0;
3147 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
3148 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zsb
), NULL
,
3149 &atime
, sizeof (atime
));
3152 if (mask
& ATTR_MTIME
) {
3153 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3154 ZTOI(zp
)->i_mtime
= timespec_trunc(vap
->va_mtime
,
3155 ZTOI(zp
)->i_sb
->s_time_gran
);
3157 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
,
3158 mtime
, sizeof (mtime
));
3161 if (mask
& ATTR_CTIME
) {
3162 ZFS_TIME_ENCODE(&vap
->va_ctime
, ctime
);
3163 ZTOI(zp
)->i_ctime
= timespec_trunc(vap
->va_ctime
,
3164 ZTOI(zp
)->i_sb
->s_time_gran
);
3165 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
,
3166 ctime
, sizeof (ctime
));
3169 if (attrzp
&& mask
) {
3170 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3171 SA_ZPL_CTIME(zsb
), NULL
, &ctime
,
3176 * Do this after setting timestamps to prevent timestamp
3177 * update from toggling bit
3180 if (xoap
&& (mask
& ATTR_XVATTR
)) {
3183 * restore trimmed off masks
3184 * so that return masks can be set for caller.
3187 if (XVA_ISSET_REQ(tmpxvattr
, XAT_APPENDONLY
)) {
3188 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3190 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NOUNLINK
)) {
3191 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3193 if (XVA_ISSET_REQ(tmpxvattr
, XAT_IMMUTABLE
)) {
3194 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3196 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NODUMP
)) {
3197 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3199 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_MODIFIED
)) {
3200 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3202 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
)) {
3203 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3206 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3207 ASSERT(S_ISREG(ip
->i_mode
));
3209 zfs_xvattr_set(zp
, xvap
, tx
);
3213 zfs_fuid_sync(zsb
, tx
);
3216 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3218 mutex_exit(&zp
->z_lock
);
3219 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3220 mutex_exit(&zp
->z_acl_lock
);
3223 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3224 mutex_exit(&attrzp
->z_acl_lock
);
3225 mutex_exit(&attrzp
->z_lock
);
3228 if (err
== 0 && attrzp
) {
3229 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3240 zfs_fuid_info_free(fuidp
);
3246 if (err
== ERESTART
)
3249 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3251 zfs_inode_update(zp
);
3255 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3256 zil_commit(zilog
, 0);
3259 kmem_free(xattr_bulk
, sizeof (sa_bulk_attr_t
) * 7);
3260 kmem_free(bulk
, sizeof (sa_bulk_attr_t
) * 7);
3261 kmem_free(tmpxvattr
, sizeof (xvattr_t
));
3265 EXPORT_SYMBOL(zfs_setattr
);
3267 typedef struct zfs_zlock
{
3268 krwlock_t
*zl_rwlock
; /* lock we acquired */
3269 znode_t
*zl_znode
; /* znode we held */
3270 struct zfs_zlock
*zl_next
; /* next in list */
3274 * Drop locks and release vnodes that were held by zfs_rename_lock().
3277 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3281 while ((zl
= *zlpp
) != NULL
) {
3282 if (zl
->zl_znode
!= NULL
)
3283 iput(ZTOI(zl
->zl_znode
));
3284 rw_exit(zl
->zl_rwlock
);
3285 *zlpp
= zl
->zl_next
;
3286 kmem_free(zl
, sizeof (*zl
));
3291 * Search back through the directory tree, using the ".." entries.
3292 * Lock each directory in the chain to prevent concurrent renames.
3293 * Fail any attempt to move a directory into one of its own descendants.
3294 * XXX - z_parent_lock can overlap with map or grow locks
3297 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3301 uint64_t rootid
= ZTOZSB(zp
)->z_root
;
3302 uint64_t oidp
= zp
->z_id
;
3303 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3304 krw_t rw
= RW_WRITER
;
3307 * First pass write-locks szp and compares to zp->z_id.
3308 * Later passes read-lock zp and compare to zp->z_parent.
3311 if (!rw_tryenter(rwlp
, rw
)) {
3313 * Another thread is renaming in this path.
3314 * Note that if we are a WRITER, we don't have any
3315 * parent_locks held yet.
3317 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3319 * Drop our locks and restart
3321 zfs_rename_unlock(&zl
);
3325 rwlp
= &szp
->z_parent_lock
;
3330 * Wait for other thread to drop its locks
3336 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3337 zl
->zl_rwlock
= rwlp
;
3338 zl
->zl_znode
= NULL
;
3339 zl
->zl_next
= *zlpp
;
3342 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3343 return (SET_ERROR(EINVAL
));
3345 if (oidp
== rootid
) /* We've hit the top */
3348 if (rw
== RW_READER
) { /* i.e. not the first pass */
3349 int error
= zfs_zget(ZTOZSB(zp
), oidp
, &zp
);
3354 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(ZTOZSB(zp
)),
3355 &oidp
, sizeof (oidp
));
3356 rwlp
= &zp
->z_parent_lock
;
3359 } while (zp
->z_id
!= sdzp
->z_id
);
3365 * Move an entry from the provided source directory to the target
3366 * directory. Change the entry name as indicated.
3368 * IN: sdip - Source directory containing the "old entry".
3369 * snm - Old entry name.
3370 * tdip - Target directory to contain the "new entry".
3371 * tnm - New entry name.
3372 * cr - credentials of caller.
3373 * flags - case flags
3375 * RETURN: 0 on success, error code on failure.
3378 * sdip,tdip - ctime|mtime updated
3382 zfs_rename(struct inode
*sdip
, char *snm
, struct inode
*tdip
, char *tnm
,
3383 cred_t
*cr
, int flags
)
3385 znode_t
*tdzp
, *szp
, *tzp
;
3386 znode_t
*sdzp
= ITOZ(sdip
);
3387 zfs_sb_t
*zsb
= ITOZSB(sdip
);
3389 zfs_dirlock_t
*sdl
, *tdl
;
3392 int cmp
, serr
, terr
;
3395 boolean_t waited
= B_FALSE
;
3397 if (snm
== NULL
|| tnm
== NULL
)
3398 return (SET_ERROR(EINVAL
));
3401 ZFS_VERIFY_ZP(sdzp
);
3405 ZFS_VERIFY_ZP(tdzp
);
3408 * We check i_sb because snapshots and the ctldir must have different
3411 if (tdip
->i_sb
!= sdip
->i_sb
|| zfsctl_is_node(tdip
)) {
3413 return (SET_ERROR(EXDEV
));
3416 if (zsb
->z_utf8
&& u8_validate(tnm
,
3417 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3419 return (SET_ERROR(EILSEQ
));
3422 if (flags
& FIGNORECASE
)
3431 * This is to prevent the creation of links into attribute space
3432 * by renaming a linked file into/outof an attribute directory.
3433 * See the comment in zfs_link() for why this is considered bad.
3435 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3437 return (SET_ERROR(EINVAL
));
3441 * Lock source and target directory entries. To prevent deadlock,
3442 * a lock ordering must be defined. We lock the directory with
3443 * the smallest object id first, or if it's a tie, the one with
3444 * the lexically first name.
3446 if (sdzp
->z_id
< tdzp
->z_id
) {
3448 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3452 * First compare the two name arguments without
3453 * considering any case folding.
3455 int nofold
= (zsb
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3457 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3458 ASSERT(error
== 0 || !zsb
->z_utf8
);
3461 * POSIX: "If the old argument and the new argument
3462 * both refer to links to the same existing file,
3463 * the rename() function shall return successfully
3464 * and perform no other action."
3470 * If the file system is case-folding, then we may
3471 * have some more checking to do. A case-folding file
3472 * system is either supporting mixed case sensitivity
3473 * access or is completely case-insensitive. Note
3474 * that the file system is always case preserving.
3476 * In mixed sensitivity mode case sensitive behavior
3477 * is the default. FIGNORECASE must be used to
3478 * explicitly request case insensitive behavior.
3480 * If the source and target names provided differ only
3481 * by case (e.g., a request to rename 'tim' to 'Tim'),
3482 * we will treat this as a special case in the
3483 * case-insensitive mode: as long as the source name
3484 * is an exact match, we will allow this to proceed as
3485 * a name-change request.
3487 if ((zsb
->z_case
== ZFS_CASE_INSENSITIVE
||
3488 (zsb
->z_case
== ZFS_CASE_MIXED
&&
3489 flags
& FIGNORECASE
)) &&
3490 u8_strcmp(snm
, tnm
, 0, zsb
->z_norm
, U8_UNICODE_LATEST
,
3493 * case preserving rename request, require exact
3502 * If the source and destination directories are the same, we should
3503 * grab the z_name_lock of that directory only once.
3507 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3511 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3512 ZEXISTS
| zflg
, NULL
, NULL
);
3513 terr
= zfs_dirent_lock(&tdl
,
3514 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3516 terr
= zfs_dirent_lock(&tdl
,
3517 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3518 serr
= zfs_dirent_lock(&sdl
,
3519 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3525 * Source entry invalid or not there.
3528 zfs_dirent_unlock(tdl
);
3534 rw_exit(&sdzp
->z_name_lock
);
3536 if (strcmp(snm
, "..") == 0)
3542 zfs_dirent_unlock(sdl
);
3546 rw_exit(&sdzp
->z_name_lock
);
3548 if (strcmp(tnm
, "..") == 0)
3555 * Must have write access at the source to remove the old entry
3556 * and write access at the target to create the new entry.
3557 * Note that if target and source are the same, this can be
3558 * done in a single check.
3561 if ((error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
)))
3564 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3566 * Check to make sure rename is valid.
3567 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3569 if ((error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
)))
3574 * Does target exist?
3578 * Source and target must be the same type.
3580 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3581 if (!S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3582 error
= SET_ERROR(ENOTDIR
);
3586 if (S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3587 error
= SET_ERROR(EISDIR
);
3592 * POSIX dictates that when the source and target
3593 * entries refer to the same file object, rename
3594 * must do nothing and exit without error.
3596 if (szp
->z_id
== tzp
->z_id
) {
3602 tx
= dmu_tx_create(zsb
->z_os
);
3603 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3604 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3605 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3606 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3608 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3609 zfs_sa_upgrade_txholds(tx
, tdzp
);
3612 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3613 zfs_sa_upgrade_txholds(tx
, tzp
);
3616 zfs_sa_upgrade_txholds(tx
, szp
);
3617 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
3618 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3621 zfs_rename_unlock(&zl
);
3622 zfs_dirent_unlock(sdl
);
3623 zfs_dirent_unlock(tdl
);
3626 rw_exit(&sdzp
->z_name_lock
);
3631 if (error
== ERESTART
) {
3642 if (tzp
) /* Attempt to remove the existing target */
3643 error
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3646 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3648 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3650 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zsb
),
3651 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3654 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3656 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3657 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3658 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3661 * At this point, we have successfully created
3662 * the target name, but have failed to remove
3663 * the source name. Since the create was done
3664 * with the ZRENAMING flag, there are
3665 * complications; for one, the link count is
3666 * wrong. The easiest way to deal with this
3667 * is to remove the newly created target, and
3668 * return the original error. This must
3669 * succeed; fortunately, it is very unlikely to
3670 * fail, since we just created it.
3672 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3673 ZRENAMING
, NULL
), ==, 0);
3681 zfs_rename_unlock(&zl
);
3683 zfs_dirent_unlock(sdl
);
3684 zfs_dirent_unlock(tdl
);
3686 zfs_inode_update(sdzp
);
3688 rw_exit(&sdzp
->z_name_lock
);
3691 zfs_inode_update(tdzp
);
3693 zfs_inode_update(szp
);
3696 zfs_inode_update(tzp
);
3700 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3701 zil_commit(zilog
, 0);
3706 EXPORT_SYMBOL(zfs_rename
);
3709 * Insert the indicated symbolic reference entry into the directory.
3711 * IN: dip - Directory to contain new symbolic link.
3712 * link - Name for new symlink entry.
3713 * vap - Attributes of new entry.
3714 * target - Target path of new symlink.
3716 * cr - credentials of caller.
3717 * flags - case flags
3719 * RETURN: 0 on success, error code on failure.
3722 * dip - ctime|mtime updated
3726 zfs_symlink(struct inode
*dip
, char *name
, vattr_t
*vap
, char *link
,
3727 struct inode
**ipp
, cred_t
*cr
, int flags
)
3729 znode_t
*zp
, *dzp
= ITOZ(dip
);
3732 zfs_sb_t
*zsb
= ITOZSB(dip
);
3734 uint64_t len
= strlen(link
);
3737 zfs_acl_ids_t acl_ids
;
3738 boolean_t fuid_dirtied
;
3739 uint64_t txtype
= TX_SYMLINK
;
3740 boolean_t waited
= B_FALSE
;
3742 ASSERT(S_ISLNK(vap
->va_mode
));
3745 return (SET_ERROR(EINVAL
));
3751 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
3752 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3754 return (SET_ERROR(EILSEQ
));
3756 if (flags
& FIGNORECASE
)
3759 if (len
> MAXPATHLEN
) {
3761 return (SET_ERROR(ENAMETOOLONG
));
3764 if ((error
= zfs_acl_ids_create(dzp
, 0,
3765 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3773 * Attempt to lock directory; fail if entry already exists.
3775 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3777 zfs_acl_ids_free(&acl_ids
);
3782 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3783 zfs_acl_ids_free(&acl_ids
);
3784 zfs_dirent_unlock(dl
);
3789 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
3790 zfs_acl_ids_free(&acl_ids
);
3791 zfs_dirent_unlock(dl
);
3793 return (SET_ERROR(EDQUOT
));
3795 tx
= dmu_tx_create(zsb
->z_os
);
3796 fuid_dirtied
= zsb
->z_fuid_dirty
;
3797 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3798 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3799 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3800 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3801 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3802 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3803 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3804 acl_ids
.z_aclp
->z_acl_bytes
);
3807 zfs_fuid_txhold(zsb
, tx
);
3808 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3810 zfs_dirent_unlock(dl
);
3811 if (error
== ERESTART
) {
3817 zfs_acl_ids_free(&acl_ids
);
3824 * Create a new object for the symlink.
3825 * for version 4 ZPL datsets the symlink will be an SA attribute
3827 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3830 zfs_fuid_sync(zsb
, tx
);
3832 mutex_enter(&zp
->z_lock
);
3834 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zsb
),
3837 zfs_sa_symlink(zp
, link
, len
, tx
);
3838 mutex_exit(&zp
->z_lock
);
3841 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
3842 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3844 * Insert the new object into the directory.
3846 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3848 if (flags
& FIGNORECASE
)
3850 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3852 zfs_inode_update(dzp
);
3853 zfs_inode_update(zp
);
3855 zfs_acl_ids_free(&acl_ids
);
3859 zfs_dirent_unlock(dl
);
3863 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3864 zil_commit(zilog
, 0);
3869 EXPORT_SYMBOL(zfs_symlink
);
3872 * Return, in the buffer contained in the provided uio structure,
3873 * the symbolic path referred to by ip.
3875 * IN: ip - inode of symbolic link
3876 * uio - structure to contain the link path.
3877 * cr - credentials of caller.
3879 * RETURN: 0 if success
3880 * error code if failure
3883 * ip - atime updated
3887 zfs_readlink(struct inode
*ip
, uio_t
*uio
, cred_t
*cr
)
3889 znode_t
*zp
= ITOZ(ip
);
3890 zfs_sb_t
*zsb
= ITOZSB(ip
);
3896 mutex_enter(&zp
->z_lock
);
3898 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3899 SA_ZPL_SYMLINK(zsb
), uio
);
3901 error
= zfs_sa_readlink(zp
, uio
);
3902 mutex_exit(&zp
->z_lock
);
3907 EXPORT_SYMBOL(zfs_readlink
);
3910 * Insert a new entry into directory tdip referencing sip.
3912 * IN: tdip - Directory to contain new entry.
3913 * sip - inode of new entry.
3914 * name - name of new entry.
3915 * cr - credentials of caller.
3917 * RETURN: 0 if success
3918 * error code if failure
3921 * tdip - ctime|mtime updated
3922 * sip - ctime updated
3926 zfs_link(struct inode
*tdip
, struct inode
*sip
, char *name
, cred_t
*cr
,
3929 znode_t
*dzp
= ITOZ(tdip
);
3931 zfs_sb_t
*zsb
= ITOZSB(tdip
);
3939 boolean_t waited
= B_FALSE
;
3940 boolean_t is_tmpfile
= 0;
3943 is_tmpfile
= (sip
->i_nlink
== 0 && (sip
->i_state
& I_LINKABLE
));
3945 ASSERT(S_ISDIR(tdip
->i_mode
));
3948 return (SET_ERROR(EINVAL
));
3955 * POSIX dictates that we return EPERM here.
3956 * Better choices include ENOTSUP or EISDIR.
3958 if (S_ISDIR(sip
->i_mode
)) {
3960 return (SET_ERROR(EPERM
));
3967 * We check i_sb because snapshots and the ctldir must have different
3970 if (sip
->i_sb
!= tdip
->i_sb
|| zfsctl_is_node(sip
)) {
3972 return (SET_ERROR(EXDEV
));
3975 /* Prevent links to .zfs/shares files */
3977 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
3978 &parent
, sizeof (uint64_t))) != 0) {
3982 if (parent
== zsb
->z_shares_dir
) {
3984 return (SET_ERROR(EPERM
));
3987 if (zsb
->z_utf8
&& u8_validate(name
,
3988 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3990 return (SET_ERROR(EILSEQ
));
3992 if (flags
& FIGNORECASE
)
3996 * We do not support links between attributes and non-attributes
3997 * because of the potential security risk of creating links
3998 * into "normal" file space in order to circumvent restrictions
3999 * imposed in attribute space.
4001 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
4003 return (SET_ERROR(EINVAL
));
4006 owner
= zfs_fuid_map_id(zsb
, KUID_TO_SUID(sip
->i_uid
), cr
, ZFS_OWNER
);
4007 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
4009 return (SET_ERROR(EPERM
));
4012 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
4019 * Attempt to lock directory; fail if entry already exists.
4021 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
4027 tx
= dmu_tx_create(zsb
->z_os
);
4028 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
4029 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
4031 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
4033 zfs_sa_upgrade_txholds(tx
, szp
);
4034 zfs_sa_upgrade_txholds(tx
, dzp
);
4035 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
4037 zfs_dirent_unlock(dl
);
4038 if (error
== ERESTART
) {
4048 /* unmark z_unlinked so zfs_link_create will not reject */
4050 szp
->z_unlinked
= 0;
4051 error
= zfs_link_create(dl
, szp
, tx
, 0);
4054 uint64_t txtype
= TX_LINK
;
4056 * tmpfile is created to be in z_unlinkedobj, so remove it.
4057 * Also, we don't log in ZIL, be cause all previous file
4058 * operation on the tmpfile are ignored by ZIL. Instead we
4059 * always wait for txg to sync to make sure all previous
4060 * operation are sync safe.
4063 VERIFY(zap_remove_int(zsb
->z_os
, zsb
->z_unlinkedobj
,
4064 szp
->z_id
, tx
) == 0);
4066 if (flags
& FIGNORECASE
)
4068 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
4070 } else if (is_tmpfile
) {
4071 /* restore z_unlinked since when linking failed */
4072 szp
->z_unlinked
= 1;
4074 txg
= dmu_tx_get_txg(tx
);
4077 zfs_dirent_unlock(dl
);
4079 if (!is_tmpfile
&& zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4080 zil_commit(zilog
, 0);
4083 txg_wait_synced(dmu_objset_pool(zsb
->z_os
), txg
);
4085 zfs_inode_update(dzp
);
4086 zfs_inode_update(szp
);
4090 EXPORT_SYMBOL(zfs_link
);
4093 zfs_putpage_commit_cb(void *arg
)
4095 struct page
*pp
= arg
;
4098 end_page_writeback(pp
);
4102 * Push a page out to disk, once the page is on stable storage the
4103 * registered commit callback will be run as notification of completion.
4105 * IN: ip - page mapped for inode.
4106 * pp - page to push (page is locked)
4107 * wbc - writeback control data
4109 * RETURN: 0 if success
4110 * error code if failure
4113 * ip - ctime|mtime updated
4117 zfs_putpage(struct inode
*ip
, struct page
*pp
, struct writeback_control
*wbc
)
4119 znode_t
*zp
= ITOZ(ip
);
4120 zfs_sb_t
*zsb
= ITOZSB(ip
);
4128 uint64_t mtime
[2], ctime
[2];
4129 sa_bulk_attr_t bulk
[3];
4131 struct address_space
*mapping
;
4136 ASSERT(PageLocked(pp
));
4138 pgoff
= page_offset(pp
); /* Page byte-offset in file */
4139 offset
= i_size_read(ip
); /* File length in bytes */
4140 pglen
= MIN(PAGE_SIZE
, /* Page length in bytes */
4141 P2ROUNDUP(offset
, PAGE_SIZE
)-pgoff
);
4143 /* Page is beyond end of file */
4144 if (pgoff
>= offset
) {
4150 /* Truncate page length to end of file */
4151 if (pgoff
+ pglen
> offset
)
4152 pglen
= offset
- pgoff
;
4156 * FIXME: Allow mmap writes past its quota. The correct fix
4157 * is to register a page_mkwrite() handler to count the page
4158 * against its quota when it is about to be dirtied.
4160 if (zfs_owner_overquota(zsb
, zp
, B_FALSE
) ||
4161 zfs_owner_overquota(zsb
, zp
, B_TRUE
)) {
4167 * The ordering here is critical and must adhere to the following
4168 * rules in order to avoid deadlocking in either zfs_read() or
4169 * zfs_free_range() due to a lock inversion.
4171 * 1) The page must be unlocked prior to acquiring the range lock.
4172 * This is critical because zfs_read() calls find_lock_page()
4173 * which may block on the page lock while holding the range lock.
4175 * 2) Before setting or clearing write back on a page the range lock
4176 * must be held in order to prevent a lock inversion with the
4177 * zfs_free_range() function.
4179 * This presents a problem because upon entering this function the
4180 * page lock is already held. To safely acquire the range lock the
4181 * page lock must be dropped. This creates a window where another
4182 * process could truncate, invalidate, dirty, or write out the page.
4184 * Therefore, after successfully reacquiring the range and page locks
4185 * the current page state is checked. In the common case everything
4186 * will be as is expected and it can be written out. However, if
4187 * the page state has changed it must be handled accordingly.
4189 mapping
= pp
->mapping
;
4190 redirty_page_for_writepage(wbc
, pp
);
4193 rl
= zfs_range_lock(&zp
->z_range_lock
, pgoff
, pglen
, RL_WRITER
);
4196 /* Page mapping changed or it was no longer dirty, we're done */
4197 if (unlikely((mapping
!= pp
->mapping
) || !PageDirty(pp
))) {
4199 zfs_range_unlock(rl
);
4204 /* Another process started write block if required */
4205 if (PageWriteback(pp
)) {
4207 zfs_range_unlock(rl
);
4209 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
4210 wait_on_page_writeback(pp
);
4216 /* Clear the dirty flag the required locks are held */
4217 if (!clear_page_dirty_for_io(pp
)) {
4219 zfs_range_unlock(rl
);
4225 * Counterpart for redirty_page_for_writepage() above. This page
4226 * was in fact not skipped and should not be counted as if it were.
4228 wbc
->pages_skipped
--;
4229 set_page_writeback(pp
);
4232 tx
= dmu_tx_create(zsb
->z_os
);
4233 dmu_tx_hold_write(tx
, zp
->z_id
, pgoff
, pglen
);
4234 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4235 zfs_sa_upgrade_txholds(tx
, zp
);
4237 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4239 if (err
== ERESTART
)
4243 __set_page_dirty_nobuffers(pp
);
4245 end_page_writeback(pp
);
4246 zfs_range_unlock(rl
);
4252 ASSERT3U(pglen
, <=, PAGE_SIZE
);
4253 dmu_write(zsb
->z_os
, zp
->z_id
, pgoff
, pglen
, va
, tx
);
4256 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
4257 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
4258 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_FLAGS(zsb
), NULL
, &zp
->z_pflags
, 8);
4260 /* Preserve the mtime and ctime provided by the inode */
4261 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4262 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4263 zp
->z_atime_dirty
= 0;
4266 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4268 zfs_log_write(zsb
->z_log
, tx
, TX_WRITE
, zp
, pgoff
, pglen
, 0,
4269 zfs_putpage_commit_cb
, pp
);
4272 zfs_range_unlock(rl
);
4274 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4276 * Note that this is rarely called under writepages(), because
4277 * writepages() normally handles the entire commit for
4278 * performance reasons.
4280 zil_commit(zsb
->z_log
, zp
->z_id
);
4288 * Update the system attributes when the inode has been dirtied. For the
4289 * moment we only update the mode, atime, mtime, and ctime.
4292 zfs_dirty_inode(struct inode
*ip
, int flags
)
4294 znode_t
*zp
= ITOZ(ip
);
4295 zfs_sb_t
*zsb
= ITOZSB(ip
);
4297 uint64_t mode
, atime
[2], mtime
[2], ctime
[2];
4298 sa_bulk_attr_t bulk
[4];
4302 if (zfs_is_readonly(zsb
) || dmu_objset_is_snapshot(zsb
->z_os
))
4310 * This is the lazytime semantic indroduced in Linux 4.0
4311 * This flag will only be called from update_time when lazytime is set.
4312 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4313 * Fortunately mtime and ctime are managed within ZFS itself, so we
4314 * only need to dirty atime.
4316 if (flags
== I_DIRTY_TIME
) {
4317 zp
->z_atime_dirty
= 1;
4322 tx
= dmu_tx_create(zsb
->z_os
);
4324 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4325 zfs_sa_upgrade_txholds(tx
, zp
);
4327 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4333 mutex_enter(&zp
->z_lock
);
4334 zp
->z_atime_dirty
= 0;
4336 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MODE(zsb
), NULL
, &mode
, 8);
4337 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_ATIME(zsb
), NULL
, &atime
, 16);
4338 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
4339 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
4341 /* Preserve the mode, mtime and ctime provided by the inode */
4342 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4343 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4344 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4349 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4350 mutex_exit(&zp
->z_lock
);
4357 EXPORT_SYMBOL(zfs_dirty_inode
);
4361 zfs_inactive(struct inode
*ip
)
4363 znode_t
*zp
= ITOZ(ip
);
4364 zfs_sb_t
*zsb
= ITOZSB(ip
);
4367 int need_unlock
= 0;
4369 /* Only read lock if we haven't already write locked, e.g. rollback */
4370 if (!RW_WRITE_HELD(&zsb
->z_teardown_inactive_lock
)) {
4372 rw_enter(&zsb
->z_teardown_inactive_lock
, RW_READER
);
4374 if (zp
->z_sa_hdl
== NULL
) {
4376 rw_exit(&zsb
->z_teardown_inactive_lock
);
4380 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4381 dmu_tx_t
*tx
= dmu_tx_create(zsb
->z_os
);
4383 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4384 zfs_sa_upgrade_txholds(tx
, zp
);
4385 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4389 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4390 mutex_enter(&zp
->z_lock
);
4391 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zsb
),
4392 (void *)&atime
, sizeof (atime
), tx
);
4393 zp
->z_atime_dirty
= 0;
4394 mutex_exit(&zp
->z_lock
);
4401 rw_exit(&zsb
->z_teardown_inactive_lock
);
4403 EXPORT_SYMBOL(zfs_inactive
);
4406 * Bounds-check the seek operation.
4408 * IN: ip - inode seeking within
4409 * ooff - old file offset
4410 * noffp - pointer to new file offset
4411 * ct - caller context
4413 * RETURN: 0 if success
4414 * EINVAL if new offset invalid
4418 zfs_seek(struct inode
*ip
, offset_t ooff
, offset_t
*noffp
)
4420 if (S_ISDIR(ip
->i_mode
))
4422 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4424 EXPORT_SYMBOL(zfs_seek
);
4427 * Fill pages with data from the disk.
4430 zfs_fillpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4432 znode_t
*zp
= ITOZ(ip
);
4433 zfs_sb_t
*zsb
= ITOZSB(ip
);
4435 struct page
*cur_pp
;
4436 u_offset_t io_off
, total
;
4443 io_len
= nr_pages
<< PAGE_SHIFT
;
4444 i_size
= i_size_read(ip
);
4445 io_off
= page_offset(pl
[0]);
4447 if (io_off
+ io_len
> i_size
)
4448 io_len
= i_size
- io_off
;
4451 * Iterate over list of pages and read each page individually.
4454 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4457 cur_pp
= pl
[page_idx
++];
4459 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4463 /* convert checksum errors into IO errors */
4465 err
= SET_ERROR(EIO
);
4474 * Uses zfs_fillpage to read data from the file and fill the pages.
4476 * IN: ip - inode of file to get data from.
4477 * pl - list of pages to read
4478 * nr_pages - number of pages to read
4480 * RETURN: 0 on success, error code on failure.
4483 * vp - atime updated
4487 zfs_getpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4489 znode_t
*zp
= ITOZ(ip
);
4490 zfs_sb_t
*zsb
= ITOZSB(ip
);
4499 err
= zfs_fillpage(ip
, pl
, nr_pages
);
4504 EXPORT_SYMBOL(zfs_getpage
);
4507 * Check ZFS specific permissions to memory map a section of a file.
4509 * IN: ip - inode of the file to mmap
4511 * addrp - start address in memory region
4512 * len - length of memory region
4513 * vm_flags- address flags
4515 * RETURN: 0 if success
4516 * error code if failure
4520 zfs_map(struct inode
*ip
, offset_t off
, caddr_t
*addrp
, size_t len
,
4521 unsigned long vm_flags
)
4523 znode_t
*zp
= ITOZ(ip
);
4524 zfs_sb_t
*zsb
= ITOZSB(ip
);
4529 if ((vm_flags
& VM_WRITE
) && (zp
->z_pflags
&
4530 (ZFS_IMMUTABLE
| ZFS_READONLY
| ZFS_APPENDONLY
))) {
4532 return (SET_ERROR(EPERM
));
4535 if ((vm_flags
& (VM_READ
| VM_EXEC
)) &&
4536 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4538 return (SET_ERROR(EACCES
));
4541 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4543 return (SET_ERROR(ENXIO
));
4549 EXPORT_SYMBOL(zfs_map
);
4552 * convoff - converts the given data (start, whence) to the
4556 convoff(struct inode
*ip
, flock64_t
*lckdat
, int whence
, offset_t offset
)
4561 if ((lckdat
->l_whence
== 2) || (whence
== 2)) {
4562 if ((error
= zfs_getattr(ip
, &vap
, 0, CRED()) != 0))
4566 switch (lckdat
->l_whence
) {
4568 lckdat
->l_start
+= offset
;
4571 lckdat
->l_start
+= vap
.va_size
;
4576 return (SET_ERROR(EINVAL
));
4579 if (lckdat
->l_start
< 0)
4580 return (SET_ERROR(EINVAL
));
4584 lckdat
->l_start
-= offset
;
4587 lckdat
->l_start
-= vap
.va_size
;
4592 return (SET_ERROR(EINVAL
));
4595 lckdat
->l_whence
= (short)whence
;
4600 * Free or allocate space in a file. Currently, this function only
4601 * supports the `F_FREESP' command. However, this command is somewhat
4602 * misnamed, as its functionality includes the ability to allocate as
4603 * well as free space.
4605 * IN: ip - inode of file to free data in.
4606 * cmd - action to take (only F_FREESP supported).
4607 * bfp - section of file to free/alloc.
4608 * flag - current file open mode flags.
4609 * offset - current file offset.
4610 * cr - credentials of caller [UNUSED].
4612 * RETURN: 0 on success, error code on failure.
4615 * ip - ctime|mtime updated
4619 zfs_space(struct inode
*ip
, int cmd
, flock64_t
*bfp
, int flag
,
4620 offset_t offset
, cred_t
*cr
)
4622 znode_t
*zp
= ITOZ(ip
);
4623 zfs_sb_t
*zsb
= ITOZSB(ip
);
4630 if (cmd
!= F_FREESP
) {
4632 return (SET_ERROR(EINVAL
));
4636 * Callers might not be able to detect properly that we are read-only,
4637 * so check it explicitly here.
4639 if (zfs_is_readonly(zsb
)) {
4641 return (SET_ERROR(EROFS
));
4644 if ((error
= convoff(ip
, bfp
, 0, offset
))) {
4649 if (bfp
->l_len
< 0) {
4651 return (SET_ERROR(EINVAL
));
4655 * Permissions aren't checked on Solaris because on this OS
4656 * zfs_space() can only be called with an opened file handle.
4657 * On Linux we can get here through truncate_range() which
4658 * operates directly on inodes, so we need to check access rights.
4660 if ((error
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
))) {
4666 len
= bfp
->l_len
; /* 0 means from off to end of file */
4668 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4673 EXPORT_SYMBOL(zfs_space
);
4677 zfs_fid(struct inode
*ip
, fid_t
*fidp
)
4679 znode_t
*zp
= ITOZ(ip
);
4680 zfs_sb_t
*zsb
= ITOZSB(ip
);
4683 uint64_t object
= zp
->z_id
;
4690 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zsb
),
4691 &gen64
, sizeof (uint64_t))) != 0) {
4696 gen
= (uint32_t)gen64
;
4698 size
= (zsb
->z_parent
!= zsb
) ? LONG_FID_LEN
: SHORT_FID_LEN
;
4699 if (fidp
->fid_len
< size
) {
4700 fidp
->fid_len
= size
;
4702 return (SET_ERROR(ENOSPC
));
4705 zfid
= (zfid_short_t
*)fidp
;
4707 zfid
->zf_len
= size
;
4709 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4710 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4712 /* Must have a non-zero generation number to distinguish from .zfs */
4715 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4716 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4718 if (size
== LONG_FID_LEN
) {
4719 uint64_t objsetid
= dmu_objset_id(zsb
->z_os
);
4722 zlfid
= (zfid_long_t
*)fidp
;
4724 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
4725 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
4727 /* XXX - this should be the generation number for the objset */
4728 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
4729 zlfid
->zf_setgen
[i
] = 0;
4735 EXPORT_SYMBOL(zfs_fid
);
4739 zfs_getsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4741 znode_t
*zp
= ITOZ(ip
);
4742 zfs_sb_t
*zsb
= ITOZSB(ip
);
4744 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4748 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
4753 EXPORT_SYMBOL(zfs_getsecattr
);
4757 zfs_setsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4759 znode_t
*zp
= ITOZ(ip
);
4760 zfs_sb_t
*zsb
= ITOZSB(ip
);
4762 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4763 zilog_t
*zilog
= zsb
->z_log
;
4768 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
4770 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4771 zil_commit(zilog
, 0);
4776 EXPORT_SYMBOL(zfs_setsecattr
);
4778 #ifdef HAVE_UIO_ZEROCOPY
4780 * Tunable, both must be a power of 2.
4782 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4783 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4784 * an arcbuf for a partial block read
4786 int zcr_blksz_min
= (1 << 10); /* 1K */
4787 int zcr_blksz_max
= (1 << 17); /* 128K */
4791 zfs_reqzcbuf(struct inode
*ip
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
)
4793 znode_t
*zp
= ITOZ(ip
);
4794 zfs_sb_t
*zsb
= ITOZSB(ip
);
4795 int max_blksz
= zsb
->z_max_blksz
;
4796 uio_t
*uio
= &xuio
->xu_uio
;
4797 ssize_t size
= uio
->uio_resid
;
4798 offset_t offset
= uio
->uio_loffset
;
4803 int preamble
, postamble
;
4805 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
4806 return (SET_ERROR(EINVAL
));
4813 * Loan out an arc_buf for write if write size is bigger than
4814 * max_blksz, and the file's block size is also max_blksz.
4817 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
4819 return (SET_ERROR(EINVAL
));
4822 * Caller requests buffers for write before knowing where the
4823 * write offset might be (e.g. NFS TCP write).
4828 preamble
= P2PHASE(offset
, blksz
);
4830 preamble
= blksz
- preamble
;
4835 postamble
= P2PHASE(size
, blksz
);
4838 fullblk
= size
/ blksz
;
4839 (void) dmu_xuio_init(xuio
,
4840 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
4843 * Have to fix iov base/len for partial buffers. They
4844 * currently represent full arc_buf's.
4847 /* data begins in the middle of the arc_buf */
4848 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4851 (void) dmu_xuio_add(xuio
, abuf
,
4852 blksz
- preamble
, preamble
);
4855 for (i
= 0; i
< fullblk
; i
++) {
4856 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4859 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
4863 /* data ends in the middle of the arc_buf */
4864 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4867 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
4872 * Loan out an arc_buf for read if the read size is larger than
4873 * the current file block size. Block alignment is not
4874 * considered. Partial arc_buf will be loaned out for read.
4876 blksz
= zp
->z_blksz
;
4877 if (blksz
< zcr_blksz_min
)
4878 blksz
= zcr_blksz_min
;
4879 if (blksz
> zcr_blksz_max
)
4880 blksz
= zcr_blksz_max
;
4881 /* avoid potential complexity of dealing with it */
4882 if (blksz
> max_blksz
) {
4884 return (SET_ERROR(EINVAL
));
4887 maxsize
= zp
->z_size
- uio
->uio_loffset
;
4893 return (SET_ERROR(EINVAL
));
4898 return (SET_ERROR(EINVAL
));
4901 uio
->uio_extflg
= UIO_XUIO
;
4902 XUIO_XUZC_RW(xuio
) = ioflag
;
4909 zfs_retzcbuf(struct inode
*ip
, xuio_t
*xuio
, cred_t
*cr
)
4913 int ioflag
= XUIO_XUZC_RW(xuio
);
4915 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
4917 i
= dmu_xuio_cnt(xuio
);
4919 abuf
= dmu_xuio_arcbuf(xuio
, i
);
4921 * if abuf == NULL, it must be a write buffer
4922 * that has been returned in zfs_write().
4925 dmu_return_arcbuf(abuf
);
4926 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
4929 dmu_xuio_fini(xuio
);
4932 #endif /* HAVE_UIO_ZEROCOPY */
4934 #if defined(_KERNEL) && defined(HAVE_SPL)
4935 module_param(zfs_delete_blocks
, ulong
, 0644);
4936 MODULE_PARM_DESC(zfs_delete_blocks
, "Delete files larger than N blocks async");
4937 module_param(zfs_read_chunk_size
, long, 0644);
4938 MODULE_PARM_DESC(zfs_read_chunk_size
, "Bytes to read per chunk");