4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
26 /* Portions Copyright 2007 Jeremy Teo */
27 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
37 #include <sys/vfs_opreg.h>
41 #include <sys/taskq.h>
43 #include <sys/vmsystm.h>
44 #include <sys/atomic.h>
46 #include <sys/pathname.h>
47 #include <sys/cmn_err.h>
48 #include <sys/errno.h>
49 #include <sys/unistd.h>
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_acl.h>
52 #include <sys/zfs_ioctl.h>
53 #include <sys/fs/zfs.h>
55 #include <sys/dmu_objset.h>
61 #include <sys/dirent.h>
62 #include <sys/policy.h>
63 #include <sys/sunddi.h>
66 #include "fs/fs_subr.h"
67 #include <sys/zfs_ctldir.h>
68 #include <sys/zfs_fuid.h>
69 #include <sys/zfs_sa.h>
70 #include <sys/zfs_vnops.h>
72 #include <sys/zfs_rlock.h>
73 #include <sys/extdirent.h>
74 #include <sys/kidmap.h>
82 * Each vnode op performs some logical unit of work. To do this, the ZPL must
83 * properly lock its in-core state, create a DMU transaction, do the work,
84 * record this work in the intent log (ZIL), commit the DMU transaction,
85 * and wait for the intent log to commit if it is a synchronous operation.
86 * Moreover, the vnode ops must work in both normal and log replay context.
87 * The ordering of events is important to avoid deadlocks and references
88 * to freed memory. The example below illustrates the following Big Rules:
90 * (1) A check must be made in each zfs thread for a mounted file system.
91 * This is done avoiding races using ZFS_ENTER(zsb).
92 * A ZFS_EXIT(zsb) is needed before all returns. Any znodes
93 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
94 * can return EIO from the calling function.
96 * (2) iput() should always be the last thing except for zil_commit()
97 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
98 * First, if it's the last reference, the vnode/znode
99 * can be freed, so the zp may point to freed memory. Second, the last
100 * reference will call zfs_zinactive(), which may induce a lot of work --
101 * pushing cached pages (which acquires range locks) and syncing out
102 * cached atime changes. Third, zfs_zinactive() may require a new tx,
103 * which could deadlock the system if you were already holding one.
104 * If you must call iput() within a tx then use iput_ASYNC().
106 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
107 * as they can span dmu_tx_assign() calls.
109 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
110 * dmu_tx_assign(). This is critical because we don't want to block
111 * while holding locks.
113 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
114 * reduces lock contention and CPU usage when we must wait (note that if
115 * throughput is constrained by the storage, nearly every transaction
118 * Note, in particular, that if a lock is sometimes acquired before
119 * the tx assigns, and sometimes after (e.g. z_lock), then failing
120 * to use a non-blocking assign can deadlock the system. The scenario:
122 * Thread A has grabbed a lock before calling dmu_tx_assign().
123 * Thread B is in an already-assigned tx, and blocks for this lock.
124 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
125 * forever, because the previous txg can't quiesce until B's tx commits.
127 * If dmu_tx_assign() returns ERESTART and zsb->z_assign is TXG_NOWAIT,
128 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
129 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
130 * to indicate that this operation has already called dmu_tx_wait().
131 * This will ensure that we don't retry forever, waiting a short bit
134 * (5) If the operation succeeded, generate the intent log entry for it
135 * before dropping locks. This ensures that the ordering of events
136 * in the intent log matches the order in which they actually occurred.
137 * During ZIL replay the zfs_log_* functions will update the sequence
138 * number to indicate the zil transaction has replayed.
140 * (6) At the end of each vnode op, the DMU tx must always commit,
141 * regardless of whether there were any errors.
143 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
144 * to ensure that synchronous semantics are provided when necessary.
146 * In general, this is how things should be ordered in each vnode op:
148 * ZFS_ENTER(zsb); // exit if unmounted
150 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
151 * rw_enter(...); // grab any other locks you need
152 * tx = dmu_tx_create(...); // get DMU tx
153 * dmu_tx_hold_*(); // hold each object you might modify
154 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
156 * rw_exit(...); // drop locks
157 * zfs_dirent_unlock(dl); // unlock directory entry
158 * iput(...); // release held vnodes
159 * if (error == ERESTART) {
165 * dmu_tx_abort(tx); // abort DMU tx
166 * ZFS_EXIT(zsb); // finished in zfs
167 * return (error); // really out of space
169 * error = do_real_work(); // do whatever this VOP does
171 * zfs_log_*(...); // on success, make ZIL entry
172 * dmu_tx_commit(tx); // commit DMU tx -- error or not
173 * rw_exit(...); // drop locks
174 * zfs_dirent_unlock(dl); // unlock directory entry
175 * iput(...); // release held vnodes
176 * zil_commit(zilog, foid); // synchronous when necessary
177 * ZFS_EXIT(zsb); // finished in zfs
178 * return (error); // done, report error
182 * Virus scanning is unsupported. It would be possible to add a hook
183 * here to performance the required virus scan. This could be done
184 * entirely in the kernel or potentially as an update to invoke a
188 zfs_vscan(struct inode
*ip
, cred_t
*cr
, int async
)
195 zfs_open(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
197 znode_t
*zp
= ITOZ(ip
);
198 zfs_sb_t
*zsb
= ITOZSB(ip
);
203 /* Honor ZFS_APPENDONLY file attribute */
204 if ((mode
& FMODE_WRITE
) && (zp
->z_pflags
& ZFS_APPENDONLY
) &&
205 ((flag
& O_APPEND
) == 0)) {
207 return (SET_ERROR(EPERM
));
210 /* Virus scan eligible files on open */
211 if (!zfs_has_ctldir(zp
) && zsb
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
212 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0) {
213 if (zfs_vscan(ip
, cr
, 0) != 0) {
215 return (SET_ERROR(EACCES
));
219 /* Keep a count of the synchronous opens in the znode */
221 atomic_inc_32(&zp
->z_sync_cnt
);
226 EXPORT_SYMBOL(zfs_open
);
230 zfs_close(struct inode
*ip
, int flag
, cred_t
*cr
)
232 znode_t
*zp
= ITOZ(ip
);
233 zfs_sb_t
*zsb
= ITOZSB(ip
);
238 /* Decrement the synchronous opens in the znode */
240 atomic_dec_32(&zp
->z_sync_cnt
);
242 if (!zfs_has_ctldir(zp
) && zsb
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
243 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0)
244 VERIFY(zfs_vscan(ip
, cr
, 1) == 0);
249 EXPORT_SYMBOL(zfs_close
);
251 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
253 * Lseek support for finding holes (cmd == SEEK_HOLE) and
254 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
257 zfs_holey_common(struct inode
*ip
, int cmd
, loff_t
*off
)
259 znode_t
*zp
= ITOZ(ip
);
260 uint64_t noff
= (uint64_t)*off
; /* new offset */
265 file_sz
= zp
->z_size
;
266 if (noff
>= file_sz
) {
267 return (SET_ERROR(ENXIO
));
270 if (cmd
== SEEK_HOLE
)
275 error
= dmu_offset_next(ZTOZSB(zp
)->z_os
, zp
->z_id
, hole
, &noff
);
278 if ((error
== ESRCH
) || (noff
> file_sz
)) {
280 * Handle the virtual hole at the end of file.
286 return (SET_ERROR(ENXIO
));
296 zfs_holey(struct inode
*ip
, int cmd
, loff_t
*off
)
298 znode_t
*zp
= ITOZ(ip
);
299 zfs_sb_t
*zsb
= ITOZSB(ip
);
305 error
= zfs_holey_common(ip
, cmd
, off
);
310 EXPORT_SYMBOL(zfs_holey
);
311 #endif /* SEEK_HOLE && SEEK_DATA */
315 * When a file is memory mapped, we must keep the IO data synchronized
316 * between the DMU cache and the memory mapped pages. What this means:
318 * On Write: If we find a memory mapped page, we write to *both*
319 * the page and the dmu buffer.
322 update_pages(struct inode
*ip
, int64_t start
, int len
,
323 objset_t
*os
, uint64_t oid
)
325 struct address_space
*mp
= ip
->i_mapping
;
331 off
= start
& (PAGE_CACHE_SIZE
-1);
332 for (start
&= PAGE_CACHE_MASK
; len
> 0; start
+= PAGE_CACHE_SIZE
) {
333 nbytes
= MIN(PAGE_CACHE_SIZE
- off
, len
);
335 pp
= find_lock_page(mp
, start
>> PAGE_CACHE_SHIFT
);
337 if (mapping_writably_mapped(mp
))
338 flush_dcache_page(pp
);
341 (void) dmu_read(os
, oid
, start
+off
, nbytes
, pb
+off
,
345 if (mapping_writably_mapped(mp
))
346 flush_dcache_page(pp
);
348 mark_page_accessed(pp
);
352 page_cache_release(pp
);
361 * When a file is memory mapped, we must keep the IO data synchronized
362 * between the DMU cache and the memory mapped pages. What this means:
364 * On Read: We "read" preferentially from memory mapped pages,
365 * else we default from the dmu buffer.
367 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
368 * the file is memory mapped.
371 mappedread(struct inode
*ip
, int nbytes
, uio_t
*uio
)
373 struct address_space
*mp
= ip
->i_mapping
;
375 znode_t
*zp
= ITOZ(ip
);
376 objset_t
*os
= ITOZSB(ip
)->z_os
;
383 start
= uio
->uio_loffset
;
384 off
= start
& (PAGE_CACHE_SIZE
-1);
385 for (start
&= PAGE_CACHE_MASK
; len
> 0; start
+= PAGE_CACHE_SIZE
) {
386 bytes
= MIN(PAGE_CACHE_SIZE
- off
, len
);
388 pp
= find_lock_page(mp
, start
>> PAGE_CACHE_SHIFT
);
390 ASSERT(PageUptodate(pp
));
393 error
= uiomove(pb
+ off
, bytes
, UIO_READ
, uio
);
396 if (mapping_writably_mapped(mp
))
397 flush_dcache_page(pp
);
399 mark_page_accessed(pp
);
401 page_cache_release(pp
);
403 error
= dmu_read_uio(os
, zp
->z_id
, uio
, bytes
);
415 unsigned long zfs_read_chunk_size
= 1024 * 1024; /* Tunable */
418 * Read bytes from specified file into supplied buffer.
420 * IN: ip - inode of file to be read from.
421 * uio - structure supplying read location, range info,
423 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
424 * O_DIRECT flag; used to bypass page cache.
425 * cr - credentials of caller.
427 * OUT: uio - updated offset and range, buffer filled.
429 * RETURN: 0 on success, error code on failure.
432 * inode - atime updated if byte count > 0
436 zfs_read(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
438 znode_t
*zp
= ITOZ(ip
);
439 zfs_sb_t
*zsb
= ITOZSB(ip
);
444 #ifdef HAVE_UIO_ZEROCOPY
446 #endif /* HAVE_UIO_ZEROCOPY */
452 if (zp
->z_pflags
& ZFS_AV_QUARANTINED
) {
454 return (SET_ERROR(EACCES
));
458 * Validate file offset
460 if (uio
->uio_loffset
< (offset_t
)0) {
462 return (SET_ERROR(EINVAL
));
466 * Fasttrack empty reads
468 if (uio
->uio_resid
== 0) {
474 * Check for mandatory locks
476 if (mandatory_lock(ip
) &&
477 !lock_may_read(ip
, uio
->uio_loffset
, uio
->uio_resid
)) {
479 return (SET_ERROR(EAGAIN
));
483 * If we're in FRSYNC mode, sync out this znode before reading it.
485 if (ioflag
& FRSYNC
|| zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
486 zil_commit(zsb
->z_log
, zp
->z_id
);
489 * Lock the range against changes.
491 rl
= zfs_range_lock(zp
, uio
->uio_loffset
, uio
->uio_resid
, RL_READER
);
494 * If we are reading past end-of-file we can skip
495 * to the end; but we might still need to set atime.
497 if (uio
->uio_loffset
>= zp
->z_size
) {
502 ASSERT(uio
->uio_loffset
< zp
->z_size
);
503 n
= MIN(uio
->uio_resid
, zp
->z_size
- uio
->uio_loffset
);
505 #ifdef HAVE_UIO_ZEROCOPY
506 if ((uio
->uio_extflg
== UIO_XUIO
) &&
507 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
)) {
509 int blksz
= zp
->z_blksz
;
510 uint64_t offset
= uio
->uio_loffset
;
512 xuio
= (xuio_t
*)uio
;
514 nblk
= (P2ROUNDUP(offset
+ n
, blksz
) - P2ALIGN(offset
,
517 ASSERT(offset
+ n
<= blksz
);
520 (void) dmu_xuio_init(xuio
, nblk
);
522 if (vn_has_cached_data(ip
)) {
524 * For simplicity, we always allocate a full buffer
525 * even if we only expect to read a portion of a block.
527 while (--nblk
>= 0) {
528 (void) dmu_xuio_add(xuio
,
529 dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
534 #endif /* HAVE_UIO_ZEROCOPY */
537 nbytes
= MIN(n
, zfs_read_chunk_size
-
538 P2PHASE(uio
->uio_loffset
, zfs_read_chunk_size
));
540 if (zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
))
541 error
= mappedread(ip
, nbytes
, uio
);
543 error
= dmu_read_uio(os
, zp
->z_id
, uio
, nbytes
);
546 /* convert checksum errors into IO errors */
548 error
= SET_ERROR(EIO
);
555 zfs_range_unlock(rl
);
557 ZFS_ACCESSTIME_STAMP(zsb
, zp
);
561 EXPORT_SYMBOL(zfs_read
);
564 * Write the bytes to a file.
566 * IN: ip - inode of file to be written to.
567 * uio - structure supplying write location, range info,
569 * ioflag - FAPPEND flag set if in append mode.
570 * O_DIRECT flag; used to bypass page cache.
571 * cr - credentials of caller.
573 * OUT: uio - updated offset and range.
575 * RETURN: 0 if success
576 * error code if failure
579 * ip - ctime|mtime updated if byte count > 0
584 zfs_write(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
586 znode_t
*zp
= ITOZ(ip
);
587 rlim64_t limit
= uio
->uio_limit
;
588 ssize_t start_resid
= uio
->uio_resid
;
592 zfs_sb_t
*zsb
= ZTOZSB(zp
);
597 int max_blksz
= zsb
->z_max_blksz
;
600 iovec_t
*aiov
= NULL
;
603 iovec_t
*iovp
= uio
->uio_iov
;
606 sa_bulk_attr_t bulk
[4];
607 uint64_t mtime
[2], ctime
[2];
608 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 * If immutable or not appending then return EPERM
632 if ((zp
->z_pflags
& (ZFS_IMMUTABLE
| ZFS_READONLY
)) ||
633 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
634 (uio
->uio_loffset
< zp
->z_size
))) {
636 return (SET_ERROR(EPERM
));
642 * Validate file offset
644 woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
647 return (SET_ERROR(EINVAL
));
651 * Check for mandatory locks before calling zfs_range_lock()
652 * in order to prevent a deadlock with locks set via fcntl().
654 if (mandatory_lock(ip
) && !lock_may_write(ip
, woff
, n
)) {
656 return (SET_ERROR(EAGAIN
));
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
, 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
, 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 ASSERT(i_iov
< iovcnt
);
733 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
734 dmu_xuio_clear(xuio
, i_iov
);
735 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
736 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
737 aiov
->iov_len
== arc_buf_size(abuf
)));
739 } else if (abuf
== NULL
&& n
>= max_blksz
&&
740 woff
>= zp
->z_size
&&
741 P2PHASE(woff
, max_blksz
) == 0 &&
742 zp
->z_blksz
== max_blksz
) {
744 * This write covers a full block. "Borrow" a buffer
745 * from the dmu so that we can fill it before we enter
746 * a transaction. This avoids the possibility of
747 * holding up the transaction if the data copy hangs
748 * up on a pagefault (e.g., from an NFS server mapping).
752 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
754 ASSERT(abuf
!= NULL
);
755 ASSERT(arc_buf_size(abuf
) == max_blksz
);
756 if ((error
= uiocopy(abuf
->b_data
, max_blksz
,
757 UIO_WRITE
, uio
, &cbytes
))) {
758 dmu_return_arcbuf(abuf
);
761 ASSERT(cbytes
== max_blksz
);
765 * Start a transaction.
767 tx
= dmu_tx_create(zsb
->z_os
);
768 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
769 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
770 zfs_sa_upgrade_txholds(tx
, zp
);
771 error
= dmu_tx_assign(tx
, TXG_WAIT
);
775 dmu_return_arcbuf(abuf
);
780 * If zfs_range_lock() over-locked we grow the blocksize
781 * and then reduce the lock range. This will only happen
782 * on the first iteration since zfs_range_reduce() will
783 * shrink down r_len to the appropriate size.
785 if (rl
->r_len
== UINT64_MAX
) {
788 if (zp
->z_blksz
> max_blksz
) {
789 ASSERT(!ISP2(zp
->z_blksz
));
790 new_blksz
= MIN(end_size
, SPA_MAXBLOCKSIZE
);
792 new_blksz
= MIN(end_size
, max_blksz
);
794 zfs_grow_blocksize(zp
, new_blksz
, tx
);
795 zfs_range_reduce(rl
, woff
, n
);
799 * XXX - should we really limit each write to z_max_blksz?
800 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
802 nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
805 tx_bytes
= uio
->uio_resid
;
806 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
808 tx_bytes
-= uio
->uio_resid
;
811 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
813 * If this is not a full block write, but we are
814 * extending the file past EOF and this data starts
815 * block-aligned, use assign_arcbuf(). Otherwise,
816 * write via dmu_write().
818 if (tx_bytes
< max_blksz
&& (!write_eof
||
819 aiov
->iov_base
!= abuf
->b_data
)) {
821 dmu_write(zsb
->z_os
, zp
->z_id
, woff
,
822 aiov
->iov_len
, aiov
->iov_base
, tx
);
823 dmu_return_arcbuf(abuf
);
824 xuio_stat_wbuf_copied();
826 ASSERT(xuio
|| tx_bytes
== max_blksz
);
827 dmu_assign_arcbuf(sa_get_db(zp
->z_sa_hdl
),
830 ASSERT(tx_bytes
<= uio
->uio_resid
);
831 uioskip(uio
, tx_bytes
);
834 if (tx_bytes
&& zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
))
835 update_pages(ip
, woff
, tx_bytes
, zsb
->z_os
, zp
->z_id
);
838 * If we made no progress, we're done. If we made even
839 * partial progress, update the znode and ZIL accordingly.
842 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
843 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
850 * Clear Set-UID/Set-GID bits on successful write if not
851 * privileged and at least one of the excute bits is set.
853 * It would be nice to to this after all writes have
854 * been done, but that would still expose the ISUID/ISGID
855 * to another app after the partial write is committed.
857 * Note: we don't call zfs_fuid_map_id() here because
858 * user 0 is not an ephemeral uid.
860 mutex_enter(&zp
->z_acl_lock
);
861 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
862 (S_IXUSR
>> 6))) != 0 &&
863 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
864 secpolicy_vnode_setid_retain(cr
,
865 (zp
->z_mode
& S_ISUID
) != 0 && zp
->z_uid
== 0) != 0) {
867 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
868 newmode
= zp
->z_mode
;
869 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zsb
),
870 (void *)&newmode
, sizeof (uint64_t), tx
);
872 mutex_exit(&zp
->z_acl_lock
);
874 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
878 * Update the file size (zp_size) if it has changed;
879 * account for possible concurrent updates.
881 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
882 (void) atomic_cas_64(&zp
->z_size
, end_size
,
887 * If we are replaying and eof is non zero then force
888 * the file size to the specified eof. Note, there's no
889 * concurrency during replay.
891 if (zsb
->z_replay
&& zsb
->z_replay_eof
!= 0)
892 zp
->z_size
= zsb
->z_replay_eof
;
894 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
896 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
,
902 ASSERT(tx_bytes
== nbytes
);
906 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
909 zfs_range_unlock(rl
);
912 * If we're in replay mode, or we made no progress, return error.
913 * Otherwise, it's at least a partial write, so it's successful.
915 if (zsb
->z_replay
|| uio
->uio_resid
== start_resid
) {
920 if (ioflag
& (FSYNC
| FDSYNC
) ||
921 zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
922 zil_commit(zilog
, zp
->z_id
);
924 zfs_inode_update(zp
);
928 EXPORT_SYMBOL(zfs_write
);
931 iput_async(struct inode
*ip
, taskq_t
*taskq
)
933 ASSERT(atomic_read(&ip
->i_count
) > 0);
934 if (atomic_read(&ip
->i_count
) == 1)
935 taskq_dispatch(taskq
, (task_func_t
*)iput
, ip
, TQ_PUSHPAGE
);
941 zfs_get_done(zgd_t
*zgd
, int error
)
943 znode_t
*zp
= zgd
->zgd_private
;
944 objset_t
*os
= ZTOZSB(zp
)->z_os
;
947 dmu_buf_rele(zgd
->zgd_db
, zgd
);
949 zfs_range_unlock(zgd
->zgd_rl
);
952 * Release the vnode asynchronously as we currently have the
953 * txg stopped from syncing.
955 iput_async(ZTOI(zp
), dsl_pool_iput_taskq(dmu_objset_pool(os
)));
957 if (error
== 0 && zgd
->zgd_bp
)
958 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
960 kmem_free(zgd
, sizeof (zgd_t
));
964 static int zil_fault_io
= 0;
968 * Get data to generate a TX_WRITE intent log record.
971 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
974 objset_t
*os
= zsb
->z_os
;
976 uint64_t object
= lr
->lr_foid
;
977 uint64_t offset
= lr
->lr_offset
;
978 uint64_t size
= lr
->lr_length
;
979 blkptr_t
*bp
= &lr
->lr_blkptr
;
988 * Nothing to do if the file has been removed
990 if (zfs_zget(zsb
, object
, &zp
) != 0)
991 return (SET_ERROR(ENOENT
));
992 if (zp
->z_unlinked
) {
994 * Release the vnode asynchronously as we currently have the
995 * txg stopped from syncing.
997 iput_async(ZTOI(zp
), dsl_pool_iput_taskq(dmu_objset_pool(os
)));
998 return (SET_ERROR(ENOENT
));
1001 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_PUSHPAGE
);
1002 zgd
->zgd_zilog
= zsb
->z_log
;
1003 zgd
->zgd_private
= zp
;
1006 * Write records come in two flavors: immediate and indirect.
1007 * For small writes it's cheaper to store the data with the
1008 * log record (immediate); for large writes it's cheaper to
1009 * sync the data and get a pointer to it (indirect) so that
1010 * we don't have to write the data twice.
1012 if (buf
!= NULL
) { /* immediate write */
1013 zgd
->zgd_rl
= zfs_range_lock(zp
, offset
, size
, RL_READER
);
1014 /* test for truncation needs to be done while range locked */
1015 if (offset
>= zp
->z_size
) {
1016 error
= SET_ERROR(ENOENT
);
1018 error
= dmu_read(os
, object
, offset
, size
, buf
,
1019 DMU_READ_NO_PREFETCH
);
1021 ASSERT(error
== 0 || error
== ENOENT
);
1022 } else { /* indirect write */
1024 * Have to lock the whole block to ensure when it's
1025 * written out and it's checksum is being calculated
1026 * that no one can change the data. We need to re-check
1027 * blocksize after we get the lock in case it's changed!
1032 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1034 zgd
->zgd_rl
= zfs_range_lock(zp
, offset
, size
,
1036 if (zp
->z_blksz
== size
)
1039 zfs_range_unlock(zgd
->zgd_rl
);
1041 /* test for truncation needs to be done while range locked */
1042 if (lr
->lr_offset
>= zp
->z_size
)
1043 error
= SET_ERROR(ENOENT
);
1046 error
= SET_ERROR(EIO
);
1051 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1052 DMU_READ_NO_PREFETCH
);
1055 blkptr_t
*obp
= dmu_buf_get_blkptr(db
);
1057 ASSERT(BP_IS_HOLE(bp
));
1064 ASSERT(db
->db_offset
== offset
);
1065 ASSERT(db
->db_size
== size
);
1067 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1069 ASSERT(error
|| lr
->lr_length
<= zp
->z_blksz
);
1072 * On success, we need to wait for the write I/O
1073 * initiated by dmu_sync() to complete before we can
1074 * release this dbuf. We will finish everything up
1075 * in the zfs_get_done() callback.
1080 if (error
== EALREADY
) {
1081 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
1087 zfs_get_done(zgd
, error
);
1094 zfs_access(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
1096 znode_t
*zp
= ITOZ(ip
);
1097 zfs_sb_t
*zsb
= ITOZSB(ip
);
1103 if (flag
& V_ACE_MASK
)
1104 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1106 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1111 EXPORT_SYMBOL(zfs_access
);
1114 * Lookup an entry in a directory, or an extended attribute directory.
1115 * If it exists, return a held inode reference for it.
1117 * IN: dip - inode of directory to search.
1118 * nm - name of entry to lookup.
1119 * flags - LOOKUP_XATTR set if looking for an attribute.
1120 * cr - credentials of caller.
1121 * direntflags - directory lookup flags
1122 * realpnp - returned pathname.
1124 * OUT: ipp - inode of located entry, NULL if not found.
1126 * RETURN: 0 on success, error code on failure.
1133 zfs_lookup(struct inode
*dip
, char *nm
, struct inode
**ipp
, int flags
,
1134 cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1136 znode_t
*zdp
= ITOZ(dip
);
1137 zfs_sb_t
*zsb
= ITOZSB(dip
);
1141 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1143 if (!S_ISDIR(dip
->i_mode
)) {
1144 return (SET_ERROR(ENOTDIR
));
1145 } else if (zdp
->z_sa_hdl
== NULL
) {
1146 return (SET_ERROR(EIO
));
1149 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1150 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1159 vnode_t
*tvp
= dnlc_lookup(dvp
, nm
);
1162 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1167 if (tvp
== DNLC_NO_VNODE
) {
1169 return (SET_ERROR(ENOENT
));
1172 return (specvp_check(vpp
, cr
));
1175 #endif /* HAVE_DNLC */
1184 if (flags
& LOOKUP_XATTR
) {
1186 * We don't allow recursive attributes..
1187 * Maybe someday we will.
1189 if (zdp
->z_pflags
& ZFS_XATTR
) {
1191 return (SET_ERROR(EINVAL
));
1194 if ((error
= zfs_get_xattrdir(zdp
, ipp
, cr
, flags
))) {
1200 * Do we have permission to get into attribute directory?
1203 if ((error
= zfs_zaccess(ITOZ(*ipp
), ACE_EXECUTE
, 0,
1213 if (!S_ISDIR(dip
->i_mode
)) {
1215 return (SET_ERROR(ENOTDIR
));
1219 * Check accessibility of directory.
1222 if ((error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
))) {
1227 if (zsb
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1228 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1230 return (SET_ERROR(EILSEQ
));
1233 error
= zfs_dirlook(zdp
, nm
, ipp
, flags
, direntflags
, realpnp
);
1234 if ((error
== 0) && (*ipp
))
1235 zfs_inode_update(ITOZ(*ipp
));
1240 EXPORT_SYMBOL(zfs_lookup
);
1243 * Attempt to create a new entry in a directory. If the entry
1244 * already exists, truncate the file if permissible, else return
1245 * an error. Return the ip of the created or trunc'd file.
1247 * IN: dip - inode of directory to put new file entry in.
1248 * name - name of new file entry.
1249 * vap - attributes of new file.
1250 * excl - flag indicating exclusive or non-exclusive mode.
1251 * mode - mode to open file with.
1252 * cr - credentials of caller.
1253 * flag - large file flag [UNUSED].
1254 * vsecp - ACL to be set
1256 * OUT: ipp - inode of created or trunc'd entry.
1258 * RETURN: 0 on success, error code on failure.
1261 * dip - ctime|mtime updated if new entry created
1262 * ip - ctime|mtime always, atime if new
1267 zfs_create(struct inode
*dip
, char *name
, vattr_t
*vap
, int excl
,
1268 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1270 znode_t
*zp
, *dzp
= ITOZ(dip
);
1271 zfs_sb_t
*zsb
= ITOZSB(dip
);
1279 zfs_acl_ids_t acl_ids
;
1280 boolean_t fuid_dirtied
;
1281 boolean_t have_acl
= B_FALSE
;
1282 boolean_t waited
= B_FALSE
;
1285 * If we have an ephemeral id, ACL, or XVATTR then
1286 * make sure file system is at proper version
1292 if (zsb
->z_use_fuids
== B_FALSE
&&
1293 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1294 return (SET_ERROR(EINVAL
));
1301 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
1302 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1304 return (SET_ERROR(EILSEQ
));
1307 if (vap
->va_mask
& ATTR_XVATTR
) {
1308 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1309 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1317 if (*name
== '\0') {
1319 * Null component name refers to the directory itself.
1326 /* possible igrab(zp) */
1329 if (flag
& FIGNORECASE
)
1332 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1336 zfs_acl_ids_free(&acl_ids
);
1337 if (strcmp(name
, "..") == 0)
1338 error
= SET_ERROR(EISDIR
);
1348 * Create a new file object and update the directory
1351 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1353 zfs_acl_ids_free(&acl_ids
);
1358 * We only support the creation of regular files in
1359 * extended attribute directories.
1362 if ((dzp
->z_pflags
& ZFS_XATTR
) && !S_ISREG(vap
->va_mode
)) {
1364 zfs_acl_ids_free(&acl_ids
);
1365 error
= SET_ERROR(EINVAL
);
1369 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1370 cr
, vsecp
, &acl_ids
)) != 0)
1374 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1375 zfs_acl_ids_free(&acl_ids
);
1376 error
= SET_ERROR(EDQUOT
);
1380 tx
= dmu_tx_create(os
);
1382 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1383 ZFS_SA_BASE_ATTR_SIZE
);
1385 fuid_dirtied
= zsb
->z_fuid_dirty
;
1387 zfs_fuid_txhold(zsb
, tx
);
1388 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1389 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1390 if (!zsb
->z_use_sa
&&
1391 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1392 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1393 0, acl_ids
.z_aclp
->z_acl_bytes
);
1395 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1397 zfs_dirent_unlock(dl
);
1398 if (error
== ERESTART
) {
1404 zfs_acl_ids_free(&acl_ids
);
1409 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1412 zfs_fuid_sync(zsb
, tx
);
1414 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1415 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1416 if (flag
& FIGNORECASE
)
1418 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1419 vsecp
, acl_ids
.z_fuidp
, vap
);
1420 zfs_acl_ids_free(&acl_ids
);
1423 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1426 zfs_acl_ids_free(&acl_ids
);
1430 * A directory entry already exists for this name.
1433 * Can't truncate an existing file if in exclusive mode.
1436 error
= SET_ERROR(EEXIST
);
1440 * Can't open a directory for writing.
1442 if (S_ISDIR(ZTOI(zp
)->i_mode
)) {
1443 error
= SET_ERROR(EISDIR
);
1447 * Verify requested access to file.
1449 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1453 mutex_enter(&dzp
->z_lock
);
1455 mutex_exit(&dzp
->z_lock
);
1458 * Truncate regular files if requested.
1460 if (S_ISREG(ZTOI(zp
)->i_mode
) &&
1461 (vap
->va_mask
& ATTR_SIZE
) && (vap
->va_size
== 0)) {
1462 /* we can't hold any locks when calling zfs_freesp() */
1463 zfs_dirent_unlock(dl
);
1465 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1471 zfs_dirent_unlock(dl
);
1477 zfs_inode_update(dzp
);
1478 zfs_inode_update(zp
);
1482 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1483 zil_commit(zilog
, 0);
1488 EXPORT_SYMBOL(zfs_create
);
1491 * Remove an entry from a directory.
1493 * IN: dip - inode of directory to remove entry from.
1494 * name - name of entry to remove.
1495 * cr - credentials of caller.
1497 * RETURN: 0 if success
1498 * error code if failure
1502 * ip - ctime (if nlink > 0)
1505 uint64_t null_xattr
= 0;
1509 zfs_remove(struct inode
*dip
, char *name
, cred_t
*cr
)
1511 znode_t
*zp
, *dzp
= ITOZ(dip
);
1514 zfs_sb_t
*zsb
= ITOZSB(dip
);
1517 uint64_t xattr_obj_unlinked
= 0;
1523 pathname_t
*realnmp
= NULL
;
1524 #ifdef HAVE_PN_UTILS
1526 #endif /* HAVE_PN_UTILS */
1529 boolean_t waited
= B_FALSE
;
1535 #ifdef HAVE_PN_UTILS
1536 if (flags
& FIGNORECASE
) {
1541 #endif /* HAVE_PN_UTILS */
1547 * Attempt to lock directory; fail if entry doesn't exist.
1549 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1551 #ifdef HAVE_PN_UTILS
1554 #endif /* HAVE_PN_UTILS */
1561 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1566 * Need to use rmdir for removing directories.
1568 if (S_ISDIR(ip
->i_mode
)) {
1569 error
= SET_ERROR(EPERM
);
1575 dnlc_remove(dvp
, realnmp
->pn_buf
);
1577 dnlc_remove(dvp
, name
);
1578 #endif /* HAVE_DNLC */
1581 * We never delete the znode and always place it in the unlinked
1582 * set. The dentry cache will always hold the last reference and
1583 * is responsible for safely freeing the znode.
1586 tx
= dmu_tx_create(zsb
->z_os
);
1587 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1588 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1589 zfs_sa_upgrade_txholds(tx
, zp
);
1590 zfs_sa_upgrade_txholds(tx
, dzp
);
1592 /* are there any extended attributes? */
1593 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1594 &xattr_obj
, sizeof (xattr_obj
));
1595 if (error
== 0 && xattr_obj
) {
1596 error
= zfs_zget(zsb
, xattr_obj
, &xzp
);
1598 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1599 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1602 /* charge as an update -- would be nice not to charge at all */
1603 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
1605 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1607 zfs_dirent_unlock(dl
);
1611 if (error
== ERESTART
) {
1617 #ifdef HAVE_PN_UTILS
1620 #endif /* HAVE_PN_UTILS */
1627 * Remove the directory entry.
1629 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1638 * Hold z_lock so that we can make sure that the ACL obj
1639 * hasn't changed. Could have been deleted due to
1642 mutex_enter(&zp
->z_lock
);
1643 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1644 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1645 mutex_exit(&zp
->z_lock
);
1646 zfs_unlinked_add(zp
, tx
);
1650 #ifdef HAVE_PN_UTILS
1651 if (flags
& FIGNORECASE
)
1653 #endif /* HAVE_PN_UTILS */
1654 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1658 #ifdef HAVE_PN_UTILS
1661 #endif /* HAVE_PN_UTILS */
1663 zfs_dirent_unlock(dl
);
1664 zfs_inode_update(dzp
);
1665 zfs_inode_update(zp
);
1667 zfs_inode_update(xzp
);
1673 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1674 zil_commit(zilog
, 0);
1679 EXPORT_SYMBOL(zfs_remove
);
1682 * Create a new directory and insert it into dip using the name
1683 * provided. Return a pointer to the inserted directory.
1685 * IN: dip - inode of directory to add subdir to.
1686 * dirname - name of new directory.
1687 * vap - attributes of new directory.
1688 * cr - credentials of caller.
1689 * vsecp - ACL to be set
1691 * OUT: ipp - inode of created directory.
1693 * RETURN: 0 if success
1694 * error code if failure
1697 * dip - ctime|mtime updated
1698 * ipp - ctime|mtime|atime updated
1702 zfs_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
, struct inode
**ipp
,
1703 cred_t
*cr
, int flags
, vsecattr_t
*vsecp
)
1705 znode_t
*zp
, *dzp
= ITOZ(dip
);
1706 zfs_sb_t
*zsb
= ITOZSB(dip
);
1714 gid_t gid
= crgetgid(cr
);
1715 zfs_acl_ids_t acl_ids
;
1716 boolean_t fuid_dirtied
;
1717 boolean_t waited
= B_FALSE
;
1719 ASSERT(S_ISDIR(vap
->va_mode
));
1722 * If we have an ephemeral id, ACL, or XVATTR then
1723 * make sure file system is at proper version
1727 if (zsb
->z_use_fuids
== B_FALSE
&&
1728 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1729 return (SET_ERROR(EINVAL
));
1735 if (dzp
->z_pflags
& ZFS_XATTR
) {
1737 return (SET_ERROR(EINVAL
));
1740 if (zsb
->z_utf8
&& u8_validate(dirname
,
1741 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1743 return (SET_ERROR(EILSEQ
));
1745 if (flags
& FIGNORECASE
)
1748 if (vap
->va_mask
& ATTR_XVATTR
) {
1749 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1750 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1756 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1757 vsecp
, &acl_ids
)) != 0) {
1762 * First make sure the new directory doesn't exist.
1764 * Existence is checked first to make sure we don't return
1765 * EACCES instead of EEXIST which can cause some applications
1771 if ((error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1773 zfs_acl_ids_free(&acl_ids
);
1778 if ((error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
))) {
1779 zfs_acl_ids_free(&acl_ids
);
1780 zfs_dirent_unlock(dl
);
1785 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1786 zfs_acl_ids_free(&acl_ids
);
1787 zfs_dirent_unlock(dl
);
1789 return (SET_ERROR(EDQUOT
));
1793 * Add a new entry to the directory.
1795 tx
= dmu_tx_create(zsb
->z_os
);
1796 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
1797 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1798 fuid_dirtied
= zsb
->z_fuid_dirty
;
1800 zfs_fuid_txhold(zsb
, tx
);
1801 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1802 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
1803 acl_ids
.z_aclp
->z_acl_bytes
);
1806 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1807 ZFS_SA_BASE_ATTR_SIZE
);
1809 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1811 zfs_dirent_unlock(dl
);
1812 if (error
== ERESTART
) {
1818 zfs_acl_ids_free(&acl_ids
);
1827 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1830 zfs_fuid_sync(zsb
, tx
);
1833 * Now put new name in parent dir.
1835 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1839 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
1840 if (flags
& FIGNORECASE
)
1842 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
1843 acl_ids
.z_fuidp
, vap
);
1845 zfs_acl_ids_free(&acl_ids
);
1849 zfs_dirent_unlock(dl
);
1851 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1852 zil_commit(zilog
, 0);
1854 zfs_inode_update(dzp
);
1855 zfs_inode_update(zp
);
1859 EXPORT_SYMBOL(zfs_mkdir
);
1862 * Remove a directory subdir entry. If the current working
1863 * directory is the same as the subdir to be removed, the
1866 * IN: dip - inode of directory to remove from.
1867 * name - name of directory to be removed.
1868 * cwd - inode of current working directory.
1869 * cr - credentials of caller.
1870 * flags - case flags
1872 * RETURN: 0 on success, error code on failure.
1875 * dip - ctime|mtime updated
1879 zfs_rmdir(struct inode
*dip
, char *name
, struct inode
*cwd
, cred_t
*cr
,
1882 znode_t
*dzp
= ITOZ(dip
);
1885 zfs_sb_t
*zsb
= ITOZSB(dip
);
1891 boolean_t waited
= B_FALSE
;
1897 if (flags
& FIGNORECASE
)
1903 * Attempt to lock directory; fail if entry doesn't exist.
1905 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1913 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1917 if (!S_ISDIR(ip
->i_mode
)) {
1918 error
= SET_ERROR(ENOTDIR
);
1923 error
= SET_ERROR(EINVAL
);
1928 * Grab a lock on the directory to make sure that noone is
1929 * trying to add (or lookup) entries while we are removing it.
1931 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
1934 * Grab a lock on the parent pointer to make sure we play well
1935 * with the treewalk and directory rename code.
1937 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
1939 tx
= dmu_tx_create(zsb
->z_os
);
1940 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1941 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1942 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
1943 zfs_sa_upgrade_txholds(tx
, zp
);
1944 zfs_sa_upgrade_txholds(tx
, dzp
);
1945 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1947 rw_exit(&zp
->z_parent_lock
);
1948 rw_exit(&zp
->z_name_lock
);
1949 zfs_dirent_unlock(dl
);
1951 if (error
== ERESTART
) {
1962 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
1965 uint64_t txtype
= TX_RMDIR
;
1966 if (flags
& FIGNORECASE
)
1968 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
1973 rw_exit(&zp
->z_parent_lock
);
1974 rw_exit(&zp
->z_name_lock
);
1976 zfs_dirent_unlock(dl
);
1978 zfs_inode_update(dzp
);
1979 zfs_inode_update(zp
);
1982 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1983 zil_commit(zilog
, 0);
1988 EXPORT_SYMBOL(zfs_rmdir
);
1991 * Read as many directory entries as will fit into the provided
1992 * dirent buffer from the given directory cursor position.
1994 * IN: ip - inode of directory to read.
1995 * dirent - buffer for directory entries.
1997 * OUT: dirent - filler buffer of directory entries.
1999 * RETURN: 0 if success
2000 * error code if failure
2003 * ip - atime updated
2005 * Note that the low 4 bits of the cookie returned by zap is always zero.
2006 * This allows us to use the low range for "special" directory entries:
2007 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2008 * we use the offset 2 for the '.zfs' directory.
2012 zfs_readdir(struct inode
*ip
, struct dir_context
*ctx
, cred_t
*cr
)
2014 znode_t
*zp
= ITOZ(ip
);
2015 zfs_sb_t
*zsb
= ITOZSB(ip
);
2018 zap_attribute_t zap
;
2024 uint64_t offset
; /* must be unsigned; checks for < 1 */
2029 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
2030 &parent
, sizeof (parent
))) != 0)
2034 * Quit if directory has been removed (posix)
2042 prefetch
= zp
->z_zn_prefetch
;
2045 * Initialize the iterator cursor.
2049 * Start iteration from the beginning of the directory.
2051 zap_cursor_init(&zc
, os
, zp
->z_id
);
2054 * The offset is a serialized cursor.
2056 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2060 * Transform to file-system independent format
2065 * Special case `.', `..', and `.zfs'.
2068 (void) strcpy(zap
.za_name
, ".");
2069 zap
.za_normalization_conflict
= 0;
2072 } else if (offset
== 1) {
2073 (void) strcpy(zap
.za_name
, "..");
2074 zap
.za_normalization_conflict
= 0;
2077 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2078 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2079 zap
.za_normalization_conflict
= 0;
2080 objnum
= ZFSCTL_INO_ROOT
;
2086 if ((error
= zap_cursor_retrieve(&zc
, &zap
))) {
2087 if (error
== ENOENT
)
2094 * Allow multiple entries provided the first entry is
2095 * the object id. Non-zpl consumers may safely make
2096 * use of the additional space.
2098 * XXX: This should be a feature flag for compatibility
2100 if (zap
.za_integer_length
!= 8 ||
2101 zap
.za_num_integers
== 0) {
2102 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2103 "entry, obj = %lld, offset = %lld, "
2104 "length = %d, num = %lld\n",
2105 (u_longlong_t
)zp
->z_id
,
2106 (u_longlong_t
)offset
,
2107 zap
.za_integer_length
,
2108 (u_longlong_t
)zap
.za_num_integers
);
2109 error
= SET_ERROR(ENXIO
);
2113 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2114 type
= ZFS_DIRENT_TYPE(zap
.za_first_integer
);
2117 done
= !dir_emit(ctx
, zap
.za_name
, strlen(zap
.za_name
),
2122 /* Prefetch znode */
2124 dmu_prefetch(os
, objnum
, 0, 0);
2128 * Move to the next entry, fill in the previous offset.
2130 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2131 zap_cursor_advance(&zc
);
2132 offset
= zap_cursor_serialize(&zc
);
2138 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2141 zap_cursor_fini(&zc
);
2142 if (error
== ENOENT
)
2145 ZFS_ACCESSTIME_STAMP(zsb
, zp
);
2152 EXPORT_SYMBOL(zfs_readdir
);
2154 ulong_t zfs_fsync_sync_cnt
= 4;
2157 zfs_fsync(struct inode
*ip
, int syncflag
, cred_t
*cr
)
2159 znode_t
*zp
= ITOZ(ip
);
2160 zfs_sb_t
*zsb
= ITOZSB(ip
);
2162 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2164 if (zsb
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2167 zil_commit(zsb
->z_log
, zp
->z_id
);
2172 EXPORT_SYMBOL(zfs_fsync
);
2176 * Get the requested file attributes and place them in the provided
2179 * IN: ip - inode of file.
2180 * vap - va_mask identifies requested attributes.
2181 * If ATTR_XVATTR set, then optional attrs are requested
2182 * flags - ATTR_NOACLCHECK (CIFS server context)
2183 * cr - credentials of caller.
2185 * OUT: vap - attribute values.
2187 * RETURN: 0 (always succeeds)
2191 zfs_getattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2193 znode_t
*zp
= ITOZ(ip
);
2194 zfs_sb_t
*zsb
= ITOZSB(ip
);
2197 uint64_t mtime
[2], ctime
[2];
2198 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2199 xoptattr_t
*xoap
= NULL
;
2200 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2201 sa_bulk_attr_t bulk
[2];
2207 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2209 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
2210 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
2212 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2218 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2219 * Also, if we are the owner don't bother, since owner should
2220 * always be allowed to read basic attributes of file.
2222 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2223 (vap
->va_uid
!= crgetuid(cr
))) {
2224 if ((error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2232 * Return all attributes. It's cheaper to provide the answer
2233 * than to determine whether we were asked the question.
2236 mutex_enter(&zp
->z_lock
);
2237 vap
->va_type
= vn_mode_to_vtype(zp
->z_mode
);
2238 vap
->va_mode
= zp
->z_mode
;
2239 vap
->va_fsid
= ZTOI(zp
)->i_sb
->s_dev
;
2240 vap
->va_nodeid
= zp
->z_id
;
2241 if ((zp
->z_id
== zsb
->z_root
) && zfs_show_ctldir(zp
))
2242 links
= zp
->z_links
+ 1;
2244 links
= zp
->z_links
;
2245 vap
->va_nlink
= MIN(links
, ZFS_LINK_MAX
);
2246 vap
->va_size
= i_size_read(ip
);
2247 vap
->va_rdev
= ip
->i_rdev
;
2248 vap
->va_seq
= ip
->i_generation
;
2251 * Add in any requested optional attributes and the create time.
2252 * Also set the corresponding bits in the returned attribute bitmap.
2254 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zsb
->z_use_fuids
) {
2255 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2257 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2258 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2261 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2262 xoap
->xoa_readonly
=
2263 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2264 XVA_SET_RTN(xvap
, XAT_READONLY
);
2267 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2269 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2270 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2273 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2275 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2276 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2279 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2280 xoap
->xoa_nounlink
=
2281 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2282 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2285 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2286 xoap
->xoa_immutable
=
2287 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2288 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2291 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2292 xoap
->xoa_appendonly
=
2293 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2294 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2297 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2299 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2300 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2303 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2305 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2306 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2309 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2310 xoap
->xoa_av_quarantined
=
2311 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2312 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2315 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2316 xoap
->xoa_av_modified
=
2317 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2318 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2321 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2322 S_ISREG(ip
->i_mode
)) {
2323 zfs_sa_get_scanstamp(zp
, xvap
);
2326 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2329 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zsb
),
2330 times
, sizeof (times
));
2331 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2332 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2335 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2336 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2337 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2339 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2340 xoap
->xoa_generation
= zp
->z_gen
;
2341 XVA_SET_RTN(xvap
, XAT_GEN
);
2344 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2346 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2347 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2350 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2352 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2353 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2357 ZFS_TIME_DECODE(&vap
->va_atime
, zp
->z_atime
);
2358 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2359 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2361 mutex_exit(&zp
->z_lock
);
2363 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2365 if (zp
->z_blksz
== 0) {
2367 * Block size hasn't been set; suggest maximal I/O transfers.
2369 vap
->va_blksize
= zsb
->z_max_blksz
;
2375 EXPORT_SYMBOL(zfs_getattr
);
2378 * Get the basic file attributes and place them in the provided kstat
2379 * structure. The inode is assumed to be the authoritative source
2380 * for most of the attributes. However, the znode currently has the
2381 * authoritative atime, blksize, and block count.
2383 * IN: ip - inode of file.
2385 * OUT: sp - kstat values.
2387 * RETURN: 0 (always succeeds)
2391 zfs_getattr_fast(struct inode
*ip
, struct kstat
*sp
)
2393 znode_t
*zp
= ITOZ(ip
);
2394 zfs_sb_t
*zsb
= ITOZSB(ip
);
2396 u_longlong_t nblocks
;
2401 mutex_enter(&zp
->z_lock
);
2403 generic_fillattr(ip
, sp
);
2404 ZFS_TIME_DECODE(&sp
->atime
, zp
->z_atime
);
2406 sa_object_size(zp
->z_sa_hdl
, &blksize
, &nblocks
);
2407 sp
->blksize
= blksize
;
2408 sp
->blocks
= nblocks
;
2410 if (unlikely(zp
->z_blksz
== 0)) {
2412 * Block size hasn't been set; suggest maximal I/O transfers.
2414 sp
->blksize
= zsb
->z_max_blksz
;
2417 mutex_exit(&zp
->z_lock
);
2423 EXPORT_SYMBOL(zfs_getattr_fast
);
2426 * Set the file attributes to the values contained in the
2429 * IN: ip - inode of file to be modified.
2430 * vap - new attribute values.
2431 * If ATTR_XVATTR set, then optional attrs are being set
2432 * flags - ATTR_UTIME set if non-default time values provided.
2433 * - ATTR_NOACLCHECK (CIFS context only).
2434 * cr - credentials of caller.
2436 * RETURN: 0 if success
2437 * error code if failure
2440 * ip - ctime updated, mtime updated if size changed.
2444 zfs_setattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2446 znode_t
*zp
= ITOZ(ip
);
2447 zfs_sb_t
*zsb
= ITOZSB(ip
);
2451 xvattr_t
*tmpxvattr
;
2452 uint_t mask
= vap
->va_mask
;
2453 uint_t saved_mask
= 0;
2456 uint64_t new_uid
, new_gid
;
2458 uint64_t mtime
[2], ctime
[2];
2460 int need_policy
= FALSE
;
2462 zfs_fuid_info_t
*fuidp
= NULL
;
2463 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2466 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2467 boolean_t fuid_dirtied
= B_FALSE
;
2468 sa_bulk_attr_t
*bulk
, *xattr_bulk
;
2469 int count
= 0, xattr_count
= 0;
2480 * Make sure that if we have ephemeral uid/gid or xvattr specified
2481 * that file system is at proper version level
2484 if (zsb
->z_use_fuids
== B_FALSE
&&
2485 (((mask
& ATTR_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2486 ((mask
& ATTR_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2487 (mask
& ATTR_XVATTR
))) {
2489 return (SET_ERROR(EINVAL
));
2492 if (mask
& ATTR_SIZE
&& S_ISDIR(ip
->i_mode
)) {
2494 return (SET_ERROR(EISDIR
));
2497 if (mask
& ATTR_SIZE
&& !S_ISREG(ip
->i_mode
) && !S_ISFIFO(ip
->i_mode
)) {
2499 return (SET_ERROR(EINVAL
));
2503 * If this is an xvattr_t, then get a pointer to the structure of
2504 * optional attributes. If this is NULL, then we have a vattr_t.
2506 xoap
= xva_getxoptattr(xvap
);
2508 tmpxvattr
= kmem_alloc(sizeof (xvattr_t
), KM_SLEEP
);
2509 xva_init(tmpxvattr
);
2511 bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2512 xattr_bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2515 * Immutable files can only alter immutable bit and atime
2517 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2518 ((mask
& (ATTR_SIZE
|ATTR_UID
|ATTR_GID
|ATTR_MTIME
|ATTR_MODE
)) ||
2519 ((mask
& ATTR_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2524 if ((mask
& ATTR_SIZE
) && (zp
->z_pflags
& ZFS_READONLY
)) {
2530 * Verify timestamps doesn't overflow 32 bits.
2531 * ZFS can handle large timestamps, but 32bit syscalls can't
2532 * handle times greater than 2039. This check should be removed
2533 * once large timestamps are fully supported.
2535 if (mask
& (ATTR_ATIME
| ATTR_MTIME
)) {
2536 if (((mask
& ATTR_ATIME
) &&
2537 TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2538 ((mask
& ATTR_MTIME
) &&
2539 TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2549 /* Can this be moved to before the top label? */
2550 if (zfs_is_readonly(zsb
)) {
2556 * First validate permissions
2559 if (mask
& ATTR_SIZE
) {
2560 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2564 truncate_setsize(ip
, vap
->va_size
);
2567 * XXX - Note, we are not providing any open
2568 * mode flags here (like FNDELAY), so we may
2569 * block if there are locks present... this
2570 * should be addressed in openat().
2572 /* XXX - would it be OK to generate a log record here? */
2573 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2578 if (mask
& (ATTR_ATIME
|ATTR_MTIME
) ||
2579 ((mask
& ATTR_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2580 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2581 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2582 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2583 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2584 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2585 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2586 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2590 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2591 int idmask
= (mask
& (ATTR_UID
|ATTR_GID
));
2596 * NOTE: even if a new mode is being set,
2597 * we may clear S_ISUID/S_ISGID bits.
2600 if (!(mask
& ATTR_MODE
))
2601 vap
->va_mode
= zp
->z_mode
;
2604 * Take ownership or chgrp to group we are a member of
2607 take_owner
= (mask
& ATTR_UID
) && (vap
->va_uid
== crgetuid(cr
));
2608 take_group
= (mask
& ATTR_GID
) &&
2609 zfs_groupmember(zsb
, vap
->va_gid
, cr
);
2612 * If both ATTR_UID and ATTR_GID are set then take_owner and
2613 * take_group must both be set in order to allow taking
2616 * Otherwise, send the check through secpolicy_vnode_setattr()
2620 if (((idmask
== (ATTR_UID
|ATTR_GID
)) &&
2621 take_owner
&& take_group
) ||
2622 ((idmask
== ATTR_UID
) && take_owner
) ||
2623 ((idmask
== ATTR_GID
) && take_group
)) {
2624 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2625 skipaclchk
, cr
) == 0) {
2627 * Remove setuid/setgid for non-privileged users
2629 (void) secpolicy_setid_clear(vap
, cr
);
2630 trim_mask
= (mask
& (ATTR_UID
|ATTR_GID
));
2639 mutex_enter(&zp
->z_lock
);
2640 oldva
.va_mode
= zp
->z_mode
;
2641 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2642 if (mask
& ATTR_XVATTR
) {
2644 * Update xvattr mask to include only those attributes
2645 * that are actually changing.
2647 * the bits will be restored prior to actually setting
2648 * the attributes so the caller thinks they were set.
2650 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2651 if (xoap
->xoa_appendonly
!=
2652 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2655 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2656 XVA_SET_REQ(tmpxvattr
, XAT_APPENDONLY
);
2660 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2661 if (xoap
->xoa_nounlink
!=
2662 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2665 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2666 XVA_SET_REQ(tmpxvattr
, XAT_NOUNLINK
);
2670 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2671 if (xoap
->xoa_immutable
!=
2672 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2675 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2676 XVA_SET_REQ(tmpxvattr
, XAT_IMMUTABLE
);
2680 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2681 if (xoap
->xoa_nodump
!=
2682 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2685 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2686 XVA_SET_REQ(tmpxvattr
, XAT_NODUMP
);
2690 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2691 if (xoap
->xoa_av_modified
!=
2692 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2695 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2696 XVA_SET_REQ(tmpxvattr
, XAT_AV_MODIFIED
);
2700 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2701 if ((!S_ISREG(ip
->i_mode
) &&
2702 xoap
->xoa_av_quarantined
) ||
2703 xoap
->xoa_av_quarantined
!=
2704 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2707 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2708 XVA_SET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
);
2712 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2713 mutex_exit(&zp
->z_lock
);
2718 if (need_policy
== FALSE
&&
2719 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
2720 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
2725 mutex_exit(&zp
->z_lock
);
2727 if (mask
& ATTR_MODE
) {
2728 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
2729 err
= secpolicy_setid_setsticky_clear(ip
, vap
,
2734 trim_mask
|= ATTR_MODE
;
2742 * If trim_mask is set then take ownership
2743 * has been granted or write_acl is present and user
2744 * has the ability to modify mode. In that case remove
2745 * UID|GID and or MODE from mask so that
2746 * secpolicy_vnode_setattr() doesn't revoke it.
2750 saved_mask
= vap
->va_mask
;
2751 vap
->va_mask
&= ~trim_mask
;
2753 err
= secpolicy_vnode_setattr(cr
, ip
, vap
, &oldva
, flags
,
2754 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
2759 vap
->va_mask
|= saved_mask
;
2763 * secpolicy_vnode_setattr, or take ownership may have
2766 mask
= vap
->va_mask
;
2768 if ((mask
& (ATTR_UID
| ATTR_GID
))) {
2769 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
2770 &xattr_obj
, sizeof (xattr_obj
));
2772 if (err
== 0 && xattr_obj
) {
2773 err
= zfs_zget(ZTOZSB(zp
), xattr_obj
, &attrzp
);
2777 if (mask
& ATTR_UID
) {
2778 new_uid
= zfs_fuid_create(zsb
,
2779 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
2780 if (new_uid
!= zp
->z_uid
&&
2781 zfs_fuid_overquota(zsb
, B_FALSE
, new_uid
)) {
2789 if (mask
& ATTR_GID
) {
2790 new_gid
= zfs_fuid_create(zsb
, (uint64_t)vap
->va_gid
,
2791 cr
, ZFS_GROUP
, &fuidp
);
2792 if (new_gid
!= zp
->z_gid
&&
2793 zfs_fuid_overquota(zsb
, B_TRUE
, new_gid
)) {
2801 tx
= dmu_tx_create(zsb
->z_os
);
2803 if (mask
& ATTR_MODE
) {
2804 uint64_t pmode
= zp
->z_mode
;
2806 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
2808 zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
);
2810 mutex_enter(&zp
->z_lock
);
2811 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
2813 * Are we upgrading ACL from old V0 format
2816 if (zsb
->z_version
>= ZPL_VERSION_FUID
&&
2817 zfs_znode_acl_version(zp
) ==
2818 ZFS_ACL_VERSION_INITIAL
) {
2819 dmu_tx_hold_free(tx
, acl_obj
, 0,
2821 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
2822 0, aclp
->z_acl_bytes
);
2824 dmu_tx_hold_write(tx
, acl_obj
, 0,
2827 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
2828 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
2829 0, aclp
->z_acl_bytes
);
2831 mutex_exit(&zp
->z_lock
);
2832 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
2834 if ((mask
& ATTR_XVATTR
) &&
2835 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
2836 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
2838 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2842 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
2845 fuid_dirtied
= zsb
->z_fuid_dirty
;
2847 zfs_fuid_txhold(zsb
, tx
);
2849 zfs_sa_upgrade_txholds(tx
, zp
);
2851 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2857 * Set each attribute requested.
2858 * We group settings according to the locks they need to acquire.
2860 * Note: you cannot set ctime directly, although it will be
2861 * updated as a side-effect of calling this function.
2865 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
2866 mutex_enter(&zp
->z_acl_lock
);
2867 mutex_enter(&zp
->z_lock
);
2869 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
2870 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
2873 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
2874 mutex_enter(&attrzp
->z_acl_lock
);
2875 mutex_enter(&attrzp
->z_lock
);
2876 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2877 SA_ZPL_FLAGS(zsb
), NULL
, &attrzp
->z_pflags
,
2878 sizeof (attrzp
->z_pflags
));
2881 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2883 if (mask
& ATTR_UID
) {
2884 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zsb
), NULL
,
2885 &new_uid
, sizeof (new_uid
));
2886 zp
->z_uid
= new_uid
;
2888 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2889 SA_ZPL_UID(zsb
), NULL
, &new_uid
,
2891 attrzp
->z_uid
= new_uid
;
2895 if (mask
& ATTR_GID
) {
2896 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zsb
),
2897 NULL
, &new_gid
, sizeof (new_gid
));
2898 zp
->z_gid
= new_gid
;
2900 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2901 SA_ZPL_GID(zsb
), NULL
, &new_gid
,
2903 attrzp
->z_gid
= new_gid
;
2906 if (!(mask
& ATTR_MODE
)) {
2907 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
),
2908 NULL
, &new_mode
, sizeof (new_mode
));
2909 new_mode
= zp
->z_mode
;
2911 err
= zfs_acl_chown_setattr(zp
);
2914 err
= zfs_acl_chown_setattr(attrzp
);
2919 if (mask
& ATTR_MODE
) {
2920 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
), NULL
,
2921 &new_mode
, sizeof (new_mode
));
2922 zp
->z_mode
= new_mode
;
2923 ASSERT3P(aclp
, !=, NULL
);
2924 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
2926 if (zp
->z_acl_cached
)
2927 zfs_acl_free(zp
->z_acl_cached
);
2928 zp
->z_acl_cached
= aclp
;
2933 if (mask
& ATTR_ATIME
) {
2934 ZFS_TIME_ENCODE(&vap
->va_atime
, zp
->z_atime
);
2935 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zsb
), NULL
,
2936 &zp
->z_atime
, sizeof (zp
->z_atime
));
2939 if (mask
& ATTR_MTIME
) {
2940 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
2941 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
,
2942 mtime
, sizeof (mtime
));
2945 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
2946 if (mask
& ATTR_SIZE
&& !(mask
& ATTR_MTIME
)) {
2947 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
),
2948 NULL
, mtime
, sizeof (mtime
));
2949 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
,
2950 &ctime
, sizeof (ctime
));
2951 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
2953 } else if (mask
!= 0) {
2954 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
,
2955 &ctime
, sizeof (ctime
));
2956 zfs_tstamp_update_setup(zp
, STATE_CHANGED
, mtime
, ctime
,
2959 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
2960 SA_ZPL_CTIME(zsb
), NULL
,
2961 &ctime
, sizeof (ctime
));
2962 zfs_tstamp_update_setup(attrzp
, STATE_CHANGED
,
2963 mtime
, ctime
, B_TRUE
);
2967 * Do this after setting timestamps to prevent timestamp
2968 * update from toggling bit
2971 if (xoap
&& (mask
& ATTR_XVATTR
)) {
2974 * restore trimmed off masks
2975 * so that return masks can be set for caller.
2978 if (XVA_ISSET_REQ(tmpxvattr
, XAT_APPENDONLY
)) {
2979 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
2981 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NOUNLINK
)) {
2982 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
2984 if (XVA_ISSET_REQ(tmpxvattr
, XAT_IMMUTABLE
)) {
2985 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
2987 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NODUMP
)) {
2988 XVA_SET_REQ(xvap
, XAT_NODUMP
);
2990 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_MODIFIED
)) {
2991 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
2993 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
)) {
2994 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
2997 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
2998 ASSERT(S_ISREG(ip
->i_mode
));
3000 zfs_xvattr_set(zp
, xvap
, tx
);
3004 zfs_fuid_sync(zsb
, tx
);
3007 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3009 mutex_exit(&zp
->z_lock
);
3010 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3011 mutex_exit(&zp
->z_acl_lock
);
3014 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3015 mutex_exit(&attrzp
->z_acl_lock
);
3016 mutex_exit(&attrzp
->z_lock
);
3019 if (err
== 0 && attrzp
) {
3020 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3031 zfs_fuid_info_free(fuidp
);
3037 if (err
== ERESTART
)
3040 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3042 zfs_inode_update(zp
);
3046 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3047 zil_commit(zilog
, 0);
3050 kmem_free(xattr_bulk
, sizeof (sa_bulk_attr_t
) * 7);
3051 kmem_free(bulk
, sizeof (sa_bulk_attr_t
) * 7);
3052 kmem_free(tmpxvattr
, sizeof (xvattr_t
));
3056 EXPORT_SYMBOL(zfs_setattr
);
3058 typedef struct zfs_zlock
{
3059 krwlock_t
*zl_rwlock
; /* lock we acquired */
3060 znode_t
*zl_znode
; /* znode we held */
3061 struct zfs_zlock
*zl_next
; /* next in list */
3065 * Drop locks and release vnodes that were held by zfs_rename_lock().
3068 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3072 while ((zl
= *zlpp
) != NULL
) {
3073 if (zl
->zl_znode
!= NULL
)
3074 iput(ZTOI(zl
->zl_znode
));
3075 rw_exit(zl
->zl_rwlock
);
3076 *zlpp
= zl
->zl_next
;
3077 kmem_free(zl
, sizeof (*zl
));
3082 * Search back through the directory tree, using the ".." entries.
3083 * Lock each directory in the chain to prevent concurrent renames.
3084 * Fail any attempt to move a directory into one of its own descendants.
3085 * XXX - z_parent_lock can overlap with map or grow locks
3088 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3092 uint64_t rootid
= ZTOZSB(zp
)->z_root
;
3093 uint64_t oidp
= zp
->z_id
;
3094 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3095 krw_t rw
= RW_WRITER
;
3098 * First pass write-locks szp and compares to zp->z_id.
3099 * Later passes read-lock zp and compare to zp->z_parent.
3102 if (!rw_tryenter(rwlp
, rw
)) {
3104 * Another thread is renaming in this path.
3105 * Note that if we are a WRITER, we don't have any
3106 * parent_locks held yet.
3108 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3110 * Drop our locks and restart
3112 zfs_rename_unlock(&zl
);
3116 rwlp
= &szp
->z_parent_lock
;
3121 * Wait for other thread to drop its locks
3127 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3128 zl
->zl_rwlock
= rwlp
;
3129 zl
->zl_znode
= NULL
;
3130 zl
->zl_next
= *zlpp
;
3133 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3134 return (SET_ERROR(EINVAL
));
3136 if (oidp
== rootid
) /* We've hit the top */
3139 if (rw
== RW_READER
) { /* i.e. not the first pass */
3140 int error
= zfs_zget(ZTOZSB(zp
), oidp
, &zp
);
3145 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(ZTOZSB(zp
)),
3146 &oidp
, sizeof (oidp
));
3147 rwlp
= &zp
->z_parent_lock
;
3150 } while (zp
->z_id
!= sdzp
->z_id
);
3156 * Move an entry from the provided source directory to the target
3157 * directory. Change the entry name as indicated.
3159 * IN: sdip - Source directory containing the "old entry".
3160 * snm - Old entry name.
3161 * tdip - Target directory to contain the "new entry".
3162 * tnm - New entry name.
3163 * cr - credentials of caller.
3164 * flags - case flags
3166 * RETURN: 0 on success, error code on failure.
3169 * sdip,tdip - ctime|mtime updated
3173 zfs_rename(struct inode
*sdip
, char *snm
, struct inode
*tdip
, char *tnm
,
3174 cred_t
*cr
, int flags
)
3176 znode_t
*tdzp
, *szp
, *tzp
;
3177 znode_t
*sdzp
= ITOZ(sdip
);
3178 zfs_sb_t
*zsb
= ITOZSB(sdip
);
3180 zfs_dirlock_t
*sdl
, *tdl
;
3183 int cmp
, serr
, terr
;
3186 boolean_t waited
= B_FALSE
;
3189 ZFS_VERIFY_ZP(sdzp
);
3192 if (tdip
->i_sb
!= sdip
->i_sb
|| zfsctl_is_node(tdip
)) {
3194 return (SET_ERROR(EXDEV
));
3198 ZFS_VERIFY_ZP(tdzp
);
3199 if (zsb
->z_utf8
&& u8_validate(tnm
,
3200 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3202 return (SET_ERROR(EILSEQ
));
3205 if (flags
& FIGNORECASE
)
3214 * This is to prevent the creation of links into attribute space
3215 * by renaming a linked file into/outof an attribute directory.
3216 * See the comment in zfs_link() for why this is considered bad.
3218 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3220 return (SET_ERROR(EINVAL
));
3224 * Lock source and target directory entries. To prevent deadlock,
3225 * a lock ordering must be defined. We lock the directory with
3226 * the smallest object id first, or if it's a tie, the one with
3227 * the lexically first name.
3229 if (sdzp
->z_id
< tdzp
->z_id
) {
3231 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3235 * First compare the two name arguments without
3236 * considering any case folding.
3238 int nofold
= (zsb
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3240 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3241 ASSERT(error
== 0 || !zsb
->z_utf8
);
3244 * POSIX: "If the old argument and the new argument
3245 * both refer to links to the same existing file,
3246 * the rename() function shall return successfully
3247 * and perform no other action."
3253 * If the file system is case-folding, then we may
3254 * have some more checking to do. A case-folding file
3255 * system is either supporting mixed case sensitivity
3256 * access or is completely case-insensitive. Note
3257 * that the file system is always case preserving.
3259 * In mixed sensitivity mode case sensitive behavior
3260 * is the default. FIGNORECASE must be used to
3261 * explicitly request case insensitive behavior.
3263 * If the source and target names provided differ only
3264 * by case (e.g., a request to rename 'tim' to 'Tim'),
3265 * we will treat this as a special case in the
3266 * case-insensitive mode: as long as the source name
3267 * is an exact match, we will allow this to proceed as
3268 * a name-change request.
3270 if ((zsb
->z_case
== ZFS_CASE_INSENSITIVE
||
3271 (zsb
->z_case
== ZFS_CASE_MIXED
&&
3272 flags
& FIGNORECASE
)) &&
3273 u8_strcmp(snm
, tnm
, 0, zsb
->z_norm
, U8_UNICODE_LATEST
,
3276 * case preserving rename request, require exact
3285 * If the source and destination directories are the same, we should
3286 * grab the z_name_lock of that directory only once.
3290 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3294 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3295 ZEXISTS
| zflg
, NULL
, NULL
);
3296 terr
= zfs_dirent_lock(&tdl
,
3297 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3299 terr
= zfs_dirent_lock(&tdl
,
3300 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3301 serr
= zfs_dirent_lock(&sdl
,
3302 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3308 * Source entry invalid or not there.
3311 zfs_dirent_unlock(tdl
);
3317 rw_exit(&sdzp
->z_name_lock
);
3319 if (strcmp(snm
, "..") == 0)
3325 zfs_dirent_unlock(sdl
);
3329 rw_exit(&sdzp
->z_name_lock
);
3331 if (strcmp(tnm
, "..") == 0)
3338 * Must have write access at the source to remove the old entry
3339 * and write access at the target to create the new entry.
3340 * Note that if target and source are the same, this can be
3341 * done in a single check.
3344 if ((error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
)))
3347 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3349 * Check to make sure rename is valid.
3350 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3352 if ((error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
)))
3357 * Does target exist?
3361 * Source and target must be the same type.
3363 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3364 if (!S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3365 error
= SET_ERROR(ENOTDIR
);
3369 if (S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3370 error
= SET_ERROR(EISDIR
);
3375 * POSIX dictates that when the source and target
3376 * entries refer to the same file object, rename
3377 * must do nothing and exit without error.
3379 if (szp
->z_id
== tzp
->z_id
) {
3385 tx
= dmu_tx_create(zsb
->z_os
);
3386 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3387 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3388 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3389 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3391 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3392 zfs_sa_upgrade_txholds(tx
, tdzp
);
3395 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3396 zfs_sa_upgrade_txholds(tx
, tzp
);
3399 zfs_sa_upgrade_txholds(tx
, szp
);
3400 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
3401 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3404 zfs_rename_unlock(&zl
);
3405 zfs_dirent_unlock(sdl
);
3406 zfs_dirent_unlock(tdl
);
3409 rw_exit(&sdzp
->z_name_lock
);
3414 if (error
== ERESTART
) {
3425 if (tzp
) /* Attempt to remove the existing target */
3426 error
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3429 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3431 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3433 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zsb
),
3434 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3437 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3439 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3440 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3441 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3444 * At this point, we have successfully created
3445 * the target name, but have failed to remove
3446 * the source name. Since the create was done
3447 * with the ZRENAMING flag, there are
3448 * complications; for one, the link count is
3449 * wrong. The easiest way to deal with this
3450 * is to remove the newly created target, and
3451 * return the original error. This must
3452 * succeed; fortunately, it is very unlikely to
3453 * fail, since we just created it.
3455 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3456 ZRENAMING
, NULL
), ==, 0);
3464 zfs_rename_unlock(&zl
);
3466 zfs_dirent_unlock(sdl
);
3467 zfs_dirent_unlock(tdl
);
3469 zfs_inode_update(sdzp
);
3471 rw_exit(&sdzp
->z_name_lock
);
3474 zfs_inode_update(tdzp
);
3476 zfs_inode_update(szp
);
3479 zfs_inode_update(tzp
);
3483 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3484 zil_commit(zilog
, 0);
3489 EXPORT_SYMBOL(zfs_rename
);
3492 * Insert the indicated symbolic reference entry into the directory.
3494 * IN: dip - Directory to contain new symbolic link.
3495 * link - Name for new symlink entry.
3496 * vap - Attributes of new entry.
3497 * target - Target path of new symlink.
3499 * cr - credentials of caller.
3500 * flags - case flags
3502 * RETURN: 0 on success, error code on failure.
3505 * dip - ctime|mtime updated
3509 zfs_symlink(struct inode
*dip
, char *name
, vattr_t
*vap
, char *link
,
3510 struct inode
**ipp
, cred_t
*cr
, int flags
)
3512 znode_t
*zp
, *dzp
= ITOZ(dip
);
3515 zfs_sb_t
*zsb
= ITOZSB(dip
);
3517 uint64_t len
= strlen(link
);
3520 zfs_acl_ids_t acl_ids
;
3521 boolean_t fuid_dirtied
;
3522 uint64_t txtype
= TX_SYMLINK
;
3523 boolean_t waited
= B_FALSE
;
3525 ASSERT(S_ISLNK(vap
->va_mode
));
3531 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
3532 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3534 return (SET_ERROR(EILSEQ
));
3536 if (flags
& FIGNORECASE
)
3539 if (len
> MAXPATHLEN
) {
3541 return (SET_ERROR(ENAMETOOLONG
));
3544 if ((error
= zfs_acl_ids_create(dzp
, 0,
3545 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3553 * Attempt to lock directory; fail if entry already exists.
3555 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3557 zfs_acl_ids_free(&acl_ids
);
3562 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3563 zfs_acl_ids_free(&acl_ids
);
3564 zfs_dirent_unlock(dl
);
3569 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
3570 zfs_acl_ids_free(&acl_ids
);
3571 zfs_dirent_unlock(dl
);
3573 return (SET_ERROR(EDQUOT
));
3575 tx
= dmu_tx_create(zsb
->z_os
);
3576 fuid_dirtied
= zsb
->z_fuid_dirty
;
3577 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3578 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3579 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3580 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3581 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3582 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3583 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3584 acl_ids
.z_aclp
->z_acl_bytes
);
3587 zfs_fuid_txhold(zsb
, tx
);
3588 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3590 zfs_dirent_unlock(dl
);
3591 if (error
== ERESTART
) {
3597 zfs_acl_ids_free(&acl_ids
);
3604 * Create a new object for the symlink.
3605 * for version 4 ZPL datsets the symlink will be an SA attribute
3607 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3610 zfs_fuid_sync(zsb
, tx
);
3612 mutex_enter(&zp
->z_lock
);
3614 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zsb
),
3617 zfs_sa_symlink(zp
, link
, len
, tx
);
3618 mutex_exit(&zp
->z_lock
);
3621 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
3622 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3624 * Insert the new object into the directory.
3626 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3628 if (flags
& FIGNORECASE
)
3630 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3632 zfs_inode_update(dzp
);
3633 zfs_inode_update(zp
);
3635 zfs_acl_ids_free(&acl_ids
);
3639 zfs_dirent_unlock(dl
);
3643 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3644 zil_commit(zilog
, 0);
3649 EXPORT_SYMBOL(zfs_symlink
);
3652 * Return, in the buffer contained in the provided uio structure,
3653 * the symbolic path referred to by ip.
3655 * IN: ip - inode of symbolic link
3656 * uio - structure to contain the link path.
3657 * cr - credentials of caller.
3659 * RETURN: 0 if success
3660 * error code if failure
3663 * ip - atime updated
3667 zfs_readlink(struct inode
*ip
, uio_t
*uio
, cred_t
*cr
)
3669 znode_t
*zp
= ITOZ(ip
);
3670 zfs_sb_t
*zsb
= ITOZSB(ip
);
3676 mutex_enter(&zp
->z_lock
);
3678 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3679 SA_ZPL_SYMLINK(zsb
), uio
);
3681 error
= zfs_sa_readlink(zp
, uio
);
3682 mutex_exit(&zp
->z_lock
);
3684 ZFS_ACCESSTIME_STAMP(zsb
, zp
);
3688 EXPORT_SYMBOL(zfs_readlink
);
3691 * Insert a new entry into directory tdip referencing sip.
3693 * IN: tdip - Directory to contain new entry.
3694 * sip - inode of new entry.
3695 * name - name of new entry.
3696 * cr - credentials of caller.
3698 * RETURN: 0 if success
3699 * error code if failure
3702 * tdip - ctime|mtime updated
3703 * sip - ctime updated
3707 zfs_link(struct inode
*tdip
, struct inode
*sip
, char *name
, cred_t
*cr
)
3709 znode_t
*dzp
= ITOZ(tdip
);
3711 zfs_sb_t
*zsb
= ITOZSB(tdip
);
3719 boolean_t waited
= B_FALSE
;
3721 ASSERT(S_ISDIR(tdip
->i_mode
));
3728 * POSIX dictates that we return EPERM here.
3729 * Better choices include ENOTSUP or EISDIR.
3731 if (S_ISDIR(sip
->i_mode
)) {
3733 return (SET_ERROR(EPERM
));
3736 if (sip
->i_sb
!= tdip
->i_sb
|| zfsctl_is_node(sip
)) {
3738 return (SET_ERROR(EXDEV
));
3744 /* Prevent links to .zfs/shares files */
3746 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
3747 &parent
, sizeof (uint64_t))) != 0) {
3751 if (parent
== zsb
->z_shares_dir
) {
3753 return (SET_ERROR(EPERM
));
3756 if (zsb
->z_utf8
&& u8_validate(name
,
3757 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3759 return (SET_ERROR(EILSEQ
));
3761 #ifdef HAVE_PN_UTILS
3762 if (flags
& FIGNORECASE
)
3764 #endif /* HAVE_PN_UTILS */
3767 * We do not support links between attributes and non-attributes
3768 * because of the potential security risk of creating links
3769 * into "normal" file space in order to circumvent restrictions
3770 * imposed in attribute space.
3772 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
3774 return (SET_ERROR(EINVAL
));
3777 owner
= zfs_fuid_map_id(zsb
, szp
->z_uid
, cr
, ZFS_OWNER
);
3778 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
3780 return (SET_ERROR(EPERM
));
3783 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3790 * Attempt to lock directory; fail if entry already exists.
3792 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
3798 tx
= dmu_tx_create(zsb
->z_os
);
3799 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3800 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3801 zfs_sa_upgrade_txholds(tx
, szp
);
3802 zfs_sa_upgrade_txholds(tx
, dzp
);
3803 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3805 zfs_dirent_unlock(dl
);
3806 if (error
== ERESTART
) {
3817 error
= zfs_link_create(dl
, szp
, tx
, 0);
3820 uint64_t txtype
= TX_LINK
;
3821 #ifdef HAVE_PN_UTILS
3822 if (flags
& FIGNORECASE
)
3824 #endif /* HAVE_PN_UTILS */
3825 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
3830 zfs_dirent_unlock(dl
);
3832 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3833 zil_commit(zilog
, 0);
3835 zfs_inode_update(dzp
);
3836 zfs_inode_update(szp
);
3840 EXPORT_SYMBOL(zfs_link
);
3843 zfs_putpage_commit_cb(void *arg
)
3845 struct page
*pp
= arg
;
3848 end_page_writeback(pp
);
3852 * Push a page out to disk, once the page is on stable storage the
3853 * registered commit callback will be run as notification of completion.
3855 * IN: ip - page mapped for inode.
3856 * pp - page to push (page is locked)
3857 * wbc - writeback control data
3859 * RETURN: 0 if success
3860 * error code if failure
3863 * ip - ctime|mtime updated
3867 zfs_putpage(struct inode
*ip
, struct page
*pp
, struct writeback_control
*wbc
)
3869 znode_t
*zp
= ITOZ(ip
);
3870 zfs_sb_t
*zsb
= ITOZSB(ip
);
3878 uint64_t mtime
[2], ctime
[2];
3879 sa_bulk_attr_t bulk
[3];
3885 ASSERT(PageLocked(pp
));
3887 pgoff
= page_offset(pp
); /* Page byte-offset in file */
3888 offset
= i_size_read(ip
); /* File length in bytes */
3889 pglen
= MIN(PAGE_CACHE_SIZE
, /* Page length in bytes */
3890 P2ROUNDUP(offset
, PAGE_CACHE_SIZE
)-pgoff
);
3892 /* Page is beyond end of file */
3893 if (pgoff
>= offset
) {
3899 /* Truncate page length to end of file */
3900 if (pgoff
+ pglen
> offset
)
3901 pglen
= offset
- pgoff
;
3905 * FIXME: Allow mmap writes past its quota. The correct fix
3906 * is to register a page_mkwrite() handler to count the page
3907 * against its quota when it is about to be dirtied.
3909 if (zfs_owner_overquota(zsb
, zp
, B_FALSE
) ||
3910 zfs_owner_overquota(zsb
, zp
, B_TRUE
)) {
3915 set_page_writeback(pp
);
3918 rl
= zfs_range_lock(zp
, pgoff
, pglen
, RL_WRITER
);
3919 tx
= dmu_tx_create(zsb
->z_os
);
3921 dmu_tx_hold_write(tx
, zp
->z_id
, pgoff
, pglen
);
3923 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3924 zfs_sa_upgrade_txholds(tx
, zp
);
3925 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
3927 if (err
== ERESTART
)
3931 __set_page_dirty_nobuffers(pp
);
3933 end_page_writeback(pp
);
3934 zfs_range_unlock(rl
);
3940 ASSERT3U(pglen
, <=, PAGE_CACHE_SIZE
);
3941 dmu_write(zsb
->z_os
, zp
->z_id
, pgoff
, pglen
, va
, tx
);
3944 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
3945 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
3946 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_FLAGS(zsb
), NULL
, &zp
->z_pflags
, 8);
3948 /* Preserve the mtime and ctime provided by the inode */
3949 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
3950 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
3951 zp
->z_atime_dirty
= 0;
3954 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
3956 zfs_log_write(zsb
->z_log
, tx
, TX_WRITE
, zp
, pgoff
, pglen
, 0,
3957 zfs_putpage_commit_cb
, pp
);
3960 zfs_range_unlock(rl
);
3962 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
3964 * Note that this is rarely called under writepages(), because
3965 * writepages() normally handles the entire commit for
3966 * performance reasons.
3968 if (zsb
->z_log
!= NULL
)
3969 zil_commit(zsb
->z_log
, zp
->z_id
);
3977 * Update the system attributes when the inode has been dirtied. For the
3978 * moment we only update the mode, atime, mtime, and ctime.
3981 zfs_dirty_inode(struct inode
*ip
, int flags
)
3983 znode_t
*zp
= ITOZ(ip
);
3984 zfs_sb_t
*zsb
= ITOZSB(ip
);
3986 uint64_t mode
, atime
[2], mtime
[2], ctime
[2];
3987 sa_bulk_attr_t bulk
[4];
3994 tx
= dmu_tx_create(zsb
->z_os
);
3996 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3997 zfs_sa_upgrade_txholds(tx
, zp
);
3999 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4005 mutex_enter(&zp
->z_lock
);
4006 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MODE(zsb
), NULL
, &mode
, 8);
4007 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_ATIME(zsb
), NULL
, &atime
, 16);
4008 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
4009 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
4011 /* Preserve the mode, mtime and ctime provided by the inode */
4012 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4013 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4014 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4018 zp
->z_atime_dirty
= 0;
4020 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4021 mutex_exit(&zp
->z_lock
);
4028 EXPORT_SYMBOL(zfs_dirty_inode
);
4032 zfs_inactive(struct inode
*ip
)
4034 znode_t
*zp
= ITOZ(ip
);
4035 zfs_sb_t
*zsb
= ITOZSB(ip
);
4038 if (zfsctl_is_node(ip
)) {
4039 zfsctl_inode_inactive(ip
);
4043 rw_enter(&zsb
->z_teardown_inactive_lock
, RW_READER
);
4044 if (zp
->z_sa_hdl
== NULL
) {
4045 rw_exit(&zsb
->z_teardown_inactive_lock
);
4049 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4050 dmu_tx_t
*tx
= dmu_tx_create(zsb
->z_os
);
4052 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4053 zfs_sa_upgrade_txholds(tx
, zp
);
4054 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4058 mutex_enter(&zp
->z_lock
);
4059 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zsb
),
4060 (void *)&zp
->z_atime
, sizeof (zp
->z_atime
), tx
);
4061 zp
->z_atime_dirty
= 0;
4062 mutex_exit(&zp
->z_lock
);
4068 rw_exit(&zsb
->z_teardown_inactive_lock
);
4070 EXPORT_SYMBOL(zfs_inactive
);
4073 * Bounds-check the seek operation.
4075 * IN: ip - inode seeking within
4076 * ooff - old file offset
4077 * noffp - pointer to new file offset
4078 * ct - caller context
4080 * RETURN: 0 if success
4081 * EINVAL if new offset invalid
4085 zfs_seek(struct inode
*ip
, offset_t ooff
, offset_t
*noffp
)
4087 if (S_ISDIR(ip
->i_mode
))
4089 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4091 EXPORT_SYMBOL(zfs_seek
);
4094 * Fill pages with data from the disk.
4097 zfs_fillpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4099 znode_t
*zp
= ITOZ(ip
);
4100 zfs_sb_t
*zsb
= ITOZSB(ip
);
4102 struct page
*cur_pp
;
4103 u_offset_t io_off
, total
;
4110 io_len
= nr_pages
<< PAGE_CACHE_SHIFT
;
4111 i_size
= i_size_read(ip
);
4112 io_off
= page_offset(pl
[0]);
4114 if (io_off
+ io_len
> i_size
)
4115 io_len
= i_size
- io_off
;
4118 * Iterate over list of pages and read each page individually.
4122 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4126 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4130 /* convert checksum errors into IO errors */
4132 err
= SET_ERROR(EIO
);
4135 cur_pp
= pl
[++page_idx
];
4142 * Uses zfs_fillpage to read data from the file and fill the pages.
4144 * IN: ip - inode of file to get data from.
4145 * pl - list of pages to read
4146 * nr_pages - number of pages to read
4148 * RETURN: 0 on success, error code on failure.
4151 * vp - atime updated
4155 zfs_getpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4157 znode_t
*zp
= ITOZ(ip
);
4158 zfs_sb_t
*zsb
= ITOZSB(ip
);
4167 err
= zfs_fillpage(ip
, pl
, nr_pages
);
4170 ZFS_ACCESSTIME_STAMP(zsb
, zp
);
4175 EXPORT_SYMBOL(zfs_getpage
);
4178 * Check ZFS specific permissions to memory map a section of a file.
4180 * IN: ip - inode of the file to mmap
4182 * addrp - start address in memory region
4183 * len - length of memory region
4184 * vm_flags- address flags
4186 * RETURN: 0 if success
4187 * error code if failure
4191 zfs_map(struct inode
*ip
, offset_t off
, caddr_t
*addrp
, size_t len
,
4192 unsigned long vm_flags
)
4194 znode_t
*zp
= ITOZ(ip
);
4195 zfs_sb_t
*zsb
= ITOZSB(ip
);
4200 if ((vm_flags
& VM_WRITE
) && (zp
->z_pflags
&
4201 (ZFS_IMMUTABLE
| ZFS_READONLY
| ZFS_APPENDONLY
))) {
4203 return (SET_ERROR(EPERM
));
4206 if ((vm_flags
& (VM_READ
| VM_EXEC
)) &&
4207 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4209 return (SET_ERROR(EACCES
));
4212 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4214 return (SET_ERROR(ENXIO
));
4220 EXPORT_SYMBOL(zfs_map
);
4223 * convoff - converts the given data (start, whence) to the
4227 convoff(struct inode
*ip
, flock64_t
*lckdat
, int whence
, offset_t offset
)
4232 if ((lckdat
->l_whence
== 2) || (whence
== 2)) {
4233 if ((error
= zfs_getattr(ip
, &vap
, 0, CRED()) != 0))
4237 switch (lckdat
->l_whence
) {
4239 lckdat
->l_start
+= offset
;
4242 lckdat
->l_start
+= vap
.va_size
;
4247 return (SET_ERROR(EINVAL
));
4250 if (lckdat
->l_start
< 0)
4251 return (SET_ERROR(EINVAL
));
4255 lckdat
->l_start
-= offset
;
4258 lckdat
->l_start
-= vap
.va_size
;
4263 return (SET_ERROR(EINVAL
));
4266 lckdat
->l_whence
= (short)whence
;
4271 * Free or allocate space in a file. Currently, this function only
4272 * supports the `F_FREESP' command. However, this command is somewhat
4273 * misnamed, as its functionality includes the ability to allocate as
4274 * well as free space.
4276 * IN: ip - inode of file to free data in.
4277 * cmd - action to take (only F_FREESP supported).
4278 * bfp - section of file to free/alloc.
4279 * flag - current file open mode flags.
4280 * offset - current file offset.
4281 * cr - credentials of caller [UNUSED].
4283 * RETURN: 0 on success, error code on failure.
4286 * ip - ctime|mtime updated
4290 zfs_space(struct inode
*ip
, int cmd
, flock64_t
*bfp
, int flag
,
4291 offset_t offset
, cred_t
*cr
)
4293 znode_t
*zp
= ITOZ(ip
);
4294 zfs_sb_t
*zsb
= ITOZSB(ip
);
4301 if (cmd
!= F_FREESP
) {
4303 return (SET_ERROR(EINVAL
));
4306 if ((error
= convoff(ip
, bfp
, 0, offset
))) {
4311 if (bfp
->l_len
< 0) {
4313 return (SET_ERROR(EINVAL
));
4317 * Permissions aren't checked on Solaris because on this OS
4318 * zfs_space() can only be called with an opened file handle.
4319 * On Linux we can get here through truncate_range() which
4320 * operates directly on inodes, so we need to check access rights.
4322 if ((error
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
))) {
4328 len
= bfp
->l_len
; /* 0 means from off to end of file */
4330 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4335 EXPORT_SYMBOL(zfs_space
);
4339 zfs_fid(struct inode
*ip
, fid_t
*fidp
)
4341 znode_t
*zp
= ITOZ(ip
);
4342 zfs_sb_t
*zsb
= ITOZSB(ip
);
4345 uint64_t object
= zp
->z_id
;
4352 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zsb
),
4353 &gen64
, sizeof (uint64_t))) != 0) {
4358 gen
= (uint32_t)gen64
;
4360 size
= (zsb
->z_parent
!= zsb
) ? LONG_FID_LEN
: SHORT_FID_LEN
;
4361 if (fidp
->fid_len
< size
) {
4362 fidp
->fid_len
= size
;
4364 return (SET_ERROR(ENOSPC
));
4367 zfid
= (zfid_short_t
*)fidp
;
4369 zfid
->zf_len
= size
;
4371 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4372 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4374 /* Must have a non-zero generation number to distinguish from .zfs */
4377 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4378 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4380 if (size
== LONG_FID_LEN
) {
4381 uint64_t objsetid
= dmu_objset_id(zsb
->z_os
);
4384 zlfid
= (zfid_long_t
*)fidp
;
4386 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
4387 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
4389 /* XXX - this should be the generation number for the objset */
4390 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
4391 zlfid
->zf_setgen
[i
] = 0;
4397 EXPORT_SYMBOL(zfs_fid
);
4401 zfs_getsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4403 znode_t
*zp
= ITOZ(ip
);
4404 zfs_sb_t
*zsb
= ITOZSB(ip
);
4406 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4410 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
4415 EXPORT_SYMBOL(zfs_getsecattr
);
4419 zfs_setsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4421 znode_t
*zp
= ITOZ(ip
);
4422 zfs_sb_t
*zsb
= ITOZSB(ip
);
4424 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4425 zilog_t
*zilog
= zsb
->z_log
;
4430 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
4432 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4433 zil_commit(zilog
, 0);
4438 EXPORT_SYMBOL(zfs_setsecattr
);
4440 #ifdef HAVE_UIO_ZEROCOPY
4442 * Tunable, both must be a power of 2.
4444 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4445 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4446 * an arcbuf for a partial block read
4448 int zcr_blksz_min
= (1 << 10); /* 1K */
4449 int zcr_blksz_max
= (1 << 17); /* 128K */
4453 zfs_reqzcbuf(struct inode
*ip
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
)
4455 znode_t
*zp
= ITOZ(ip
);
4456 zfs_sb_t
*zsb
= ITOZSB(ip
);
4457 int max_blksz
= zsb
->z_max_blksz
;
4458 uio_t
*uio
= &xuio
->xu_uio
;
4459 ssize_t size
= uio
->uio_resid
;
4460 offset_t offset
= uio
->uio_loffset
;
4465 int preamble
, postamble
;
4467 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
4468 return (SET_ERROR(EINVAL
));
4475 * Loan out an arc_buf for write if write size is bigger than
4476 * max_blksz, and the file's block size is also max_blksz.
4479 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
4481 return (SET_ERROR(EINVAL
));
4484 * Caller requests buffers for write before knowing where the
4485 * write offset might be (e.g. NFS TCP write).
4490 preamble
= P2PHASE(offset
, blksz
);
4492 preamble
= blksz
- preamble
;
4497 postamble
= P2PHASE(size
, blksz
);
4500 fullblk
= size
/ blksz
;
4501 (void) dmu_xuio_init(xuio
,
4502 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
4505 * Have to fix iov base/len for partial buffers. They
4506 * currently represent full arc_buf's.
4509 /* data begins in the middle of the arc_buf */
4510 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4513 (void) dmu_xuio_add(xuio
, abuf
,
4514 blksz
- preamble
, preamble
);
4517 for (i
= 0; i
< fullblk
; i
++) {
4518 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4521 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
4525 /* data ends in the middle of the arc_buf */
4526 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4529 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
4534 * Loan out an arc_buf for read if the read size is larger than
4535 * the current file block size. Block alignment is not
4536 * considered. Partial arc_buf will be loaned out for read.
4538 blksz
= zp
->z_blksz
;
4539 if (blksz
< zcr_blksz_min
)
4540 blksz
= zcr_blksz_min
;
4541 if (blksz
> zcr_blksz_max
)
4542 blksz
= zcr_blksz_max
;
4543 /* avoid potential complexity of dealing with it */
4544 if (blksz
> max_blksz
) {
4546 return (SET_ERROR(EINVAL
));
4549 maxsize
= zp
->z_size
- uio
->uio_loffset
;
4555 return (SET_ERROR(EINVAL
));
4560 return (SET_ERROR(EINVAL
));
4563 uio
->uio_extflg
= UIO_XUIO
;
4564 XUIO_XUZC_RW(xuio
) = ioflag
;
4571 zfs_retzcbuf(struct inode
*ip
, xuio_t
*xuio
, cred_t
*cr
)
4575 int ioflag
= XUIO_XUZC_RW(xuio
);
4577 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
4579 i
= dmu_xuio_cnt(xuio
);
4581 abuf
= dmu_xuio_arcbuf(xuio
, i
);
4583 * if abuf == NULL, it must be a write buffer
4584 * that has been returned in zfs_write().
4587 dmu_return_arcbuf(abuf
);
4588 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
4591 dmu_xuio_fini(xuio
);
4594 #endif /* HAVE_UIO_ZEROCOPY */
4596 #if defined(_KERNEL) && defined(HAVE_SPL)
4597 module_param(zfs_read_chunk_size
, long, 0644);
4598 MODULE_PARM_DESC(zfs_read_chunk_size
, "Bytes to read per chunk");