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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
25 * Copyright (c) 2015 by Chunwei Chen. All rights reserved.
26 * Copyright 2017 Nexenta Systems, Inc.
29 /* Portions Copyright 2007 Jeremy Teo */
30 /* Portions Copyright 2010 Robert Milkowski */
33 #include <sys/types.h>
34 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/resource.h>
40 #include <sys/vfs_opreg.h>
44 #include <sys/taskq.h>
46 #include <sys/vmsystm.h>
47 #include <sys/atomic.h>
49 #include <sys/pathname.h>
50 #include <sys/cmn_err.h>
51 #include <sys/errno.h>
52 #include <sys/unistd.h>
53 #include <sys/zfs_dir.h>
54 #include <sys/zfs_acl.h>
55 #include <sys/zfs_ioctl.h>
56 #include <sys/fs/zfs.h>
58 #include <sys/dmu_objset.h>
64 #include <sys/dirent.h>
65 #include <sys/policy.h>
66 #include <sys/sunddi.h>
69 #include "fs/fs_subr.h"
70 #include <sys/zfs_ctldir.h>
71 #include <sys/zfs_fuid.h>
72 #include <sys/zfs_sa.h>
73 #include <sys/zfs_vnops.h>
75 #include <sys/zfs_rlock.h>
76 #include <sys/extdirent.h>
77 #include <sys/kidmap.h>
85 * Each vnode op performs some logical unit of work. To do this, the ZPL must
86 * properly lock its in-core state, create a DMU transaction, do the work,
87 * record this work in the intent log (ZIL), commit the DMU transaction,
88 * and wait for the intent log to commit if it is a synchronous operation.
89 * Moreover, the vnode ops must work in both normal and log replay context.
90 * The ordering of events is important to avoid deadlocks and references
91 * to freed memory. The example below illustrates the following Big Rules:
93 * (1) A check must be made in each zfs thread for a mounted file system.
94 * This is done avoiding races using ZFS_ENTER(zfsvfs).
95 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
96 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
97 * can return EIO from the calling function.
99 * (2) iput() should always be the last thing except for zil_commit()
100 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
101 * First, if it's the last reference, the vnode/znode
102 * can be freed, so the zp may point to freed memory. Second, the last
103 * reference will call zfs_zinactive(), which may induce a lot of work --
104 * pushing cached pages (which acquires range locks) and syncing out
105 * cached atime changes. Third, zfs_zinactive() may require a new tx,
106 * which could deadlock the system if you were already holding one.
107 * If you must call iput() within a tx then use zfs_iput_async().
109 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
110 * as they can span dmu_tx_assign() calls.
112 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
113 * dmu_tx_assign(). This is critical because we don't want to block
114 * while holding locks.
116 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
117 * reduces lock contention and CPU usage when we must wait (note that if
118 * throughput is constrained by the storage, nearly every transaction
121 * Note, in particular, that if a lock is sometimes acquired before
122 * the tx assigns, and sometimes after (e.g. z_lock), then failing
123 * to use a non-blocking assign can deadlock the system. The scenario:
125 * Thread A has grabbed a lock before calling dmu_tx_assign().
126 * Thread B is in an already-assigned tx, and blocks for this lock.
127 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
128 * forever, because the previous txg can't quiesce until B's tx commits.
130 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
131 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
132 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT,
133 * to indicate that this operation has already called dmu_tx_wait().
134 * This will ensure that we don't retry forever, waiting a short bit
137 * (5) If the operation succeeded, generate the intent log entry for it
138 * before dropping locks. This ensures that the ordering of events
139 * in the intent log matches the order in which they actually occurred.
140 * During ZIL replay the zfs_log_* functions will update the sequence
141 * number to indicate the zil transaction has replayed.
143 * (6) At the end of each vnode op, the DMU tx must always commit,
144 * regardless of whether there were any errors.
146 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
147 * to ensure that synchronous semantics are provided when necessary.
149 * In general, this is how things should be ordered in each vnode op:
151 * ZFS_ENTER(zfsvfs); // exit if unmounted
153 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
154 * rw_enter(...); // grab any other locks you need
155 * tx = dmu_tx_create(...); // get DMU tx
156 * dmu_tx_hold_*(); // hold each object you might modify
157 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
159 * rw_exit(...); // drop locks
160 * zfs_dirent_unlock(dl); // unlock directory entry
161 * iput(...); // release held vnodes
162 * if (error == ERESTART) {
168 * dmu_tx_abort(tx); // abort DMU tx
169 * ZFS_EXIT(zfsvfs); // finished in zfs
170 * return (error); // really out of space
172 * error = do_real_work(); // do whatever this VOP does
174 * zfs_log_*(...); // on success, make ZIL entry
175 * dmu_tx_commit(tx); // commit DMU tx -- error or not
176 * rw_exit(...); // drop locks
177 * zfs_dirent_unlock(dl); // unlock directory entry
178 * iput(...); // release held vnodes
179 * zil_commit(zilog, foid); // synchronous when necessary
180 * ZFS_EXIT(zfsvfs); // finished in zfs
181 * return (error); // done, report error
185 * Virus scanning is unsupported. It would be possible to add a hook
186 * here to performance the required virus scan. This could be done
187 * entirely in the kernel or potentially as an update to invoke a
191 zfs_vscan(struct inode
*ip
, cred_t
*cr
, int async
)
198 zfs_open(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
200 znode_t
*zp
= ITOZ(ip
);
201 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
206 /* Honor ZFS_APPENDONLY file attribute */
207 if ((mode
& FMODE_WRITE
) && (zp
->z_pflags
& ZFS_APPENDONLY
) &&
208 ((flag
& O_APPEND
) == 0)) {
210 return (SET_ERROR(EPERM
));
213 /* Virus scan eligible files on open */
214 if (!zfs_has_ctldir(zp
) && zfsvfs
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
215 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0) {
216 if (zfs_vscan(ip
, cr
, 0) != 0) {
218 return (SET_ERROR(EACCES
));
222 /* Keep a count of the synchronous opens in the znode */
224 atomic_inc_32(&zp
->z_sync_cnt
);
232 zfs_close(struct inode
*ip
, int flag
, cred_t
*cr
)
234 znode_t
*zp
= ITOZ(ip
);
235 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
240 /* Decrement the synchronous opens in the znode */
242 atomic_dec_32(&zp
->z_sync_cnt
);
244 if (!zfs_has_ctldir(zp
) && zfsvfs
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
245 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0)
246 VERIFY(zfs_vscan(ip
, cr
, 1) == 0);
252 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
254 * Lseek support for finding holes (cmd == SEEK_HOLE) and
255 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
258 zfs_holey_common(struct inode
*ip
, int cmd
, loff_t
*off
)
260 znode_t
*zp
= ITOZ(ip
);
261 uint64_t noff
= (uint64_t)*off
; /* new offset */
266 file_sz
= zp
->z_size
;
267 if (noff
>= file_sz
) {
268 return (SET_ERROR(ENXIO
));
271 if (cmd
== SEEK_HOLE
)
276 error
= dmu_offset_next(ZTOZSB(zp
)->z_os
, zp
->z_id
, hole
, &noff
);
279 return (SET_ERROR(ENXIO
));
281 /* file was dirty, so fall back to using generic logic */
282 if (error
== EBUSY
) {
290 * We could find a hole that begins after the logical end-of-file,
291 * because dmu_offset_next() only works on whole blocks. If the
292 * EOF falls mid-block, then indicate that the "virtual hole"
293 * at the end of the file begins at the logical EOF, rather than
294 * at the end of the last block.
296 if (noff
> file_sz
) {
308 zfs_holey(struct inode
*ip
, int cmd
, loff_t
*off
)
310 znode_t
*zp
= ITOZ(ip
);
311 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
317 error
= zfs_holey_common(ip
, cmd
, off
);
322 #endif /* SEEK_HOLE && SEEK_DATA */
326 * When a file is memory mapped, we must keep the IO data synchronized
327 * between the DMU cache and the memory mapped pages. What this means:
329 * On Write: If we find a memory mapped page, we write to *both*
330 * the page and the dmu buffer.
333 update_pages(struct inode
*ip
, int64_t start
, int len
,
334 objset_t
*os
, uint64_t oid
)
336 struct address_space
*mp
= ip
->i_mapping
;
342 off
= start
& (PAGE_SIZE
-1);
343 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
344 nbytes
= MIN(PAGE_SIZE
- off
, len
);
346 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
348 if (mapping_writably_mapped(mp
))
349 flush_dcache_page(pp
);
352 (void) dmu_read(os
, oid
, start
+off
, nbytes
, pb
+off
,
356 if (mapping_writably_mapped(mp
))
357 flush_dcache_page(pp
);
359 mark_page_accessed(pp
);
372 * When a file is memory mapped, we must keep the IO data synchronized
373 * between the DMU cache and the memory mapped pages. What this means:
375 * On Read: We "read" preferentially from memory mapped pages,
376 * else we default from the dmu buffer.
378 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
379 * the file is memory mapped.
382 mappedread(struct inode
*ip
, int nbytes
, uio_t
*uio
)
384 struct address_space
*mp
= ip
->i_mapping
;
386 znode_t
*zp
= ITOZ(ip
);
393 start
= uio
->uio_loffset
;
394 off
= start
& (PAGE_SIZE
-1);
395 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
396 bytes
= MIN(PAGE_SIZE
- off
, len
);
398 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
400 ASSERT(PageUptodate(pp
));
404 error
= uiomove(pb
+ off
, bytes
, UIO_READ
, uio
);
407 if (mapping_writably_mapped(mp
))
408 flush_dcache_page(pp
);
410 mark_page_accessed(pp
);
413 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
426 unsigned long zfs_read_chunk_size
= 1024 * 1024; /* Tunable */
427 unsigned long zfs_delete_blocks
= DMU_MAX_DELETEBLKCNT
;
430 * Read bytes from specified file into supplied buffer.
432 * IN: ip - inode of file to be read from.
433 * uio - structure supplying read location, range info,
435 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
436 * O_DIRECT flag; used to bypass page cache.
437 * cr - credentials of caller.
439 * OUT: uio - updated offset and range, buffer filled.
441 * RETURN: 0 on success, error code on failure.
444 * inode - atime updated if byte count > 0
448 zfs_read(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
450 znode_t
*zp
= ITOZ(ip
);
451 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
455 #ifdef HAVE_UIO_ZEROCOPY
457 #endif /* HAVE_UIO_ZEROCOPY */
462 if (zp
->z_pflags
& ZFS_AV_QUARANTINED
) {
464 return (SET_ERROR(EACCES
));
468 * Validate file offset
470 if (uio
->uio_loffset
< (offset_t
)0) {
472 return (SET_ERROR(EINVAL
));
476 * Fasttrack empty reads
478 if (uio
->uio_resid
== 0) {
484 * If we're in FRSYNC mode, sync out this znode before reading it.
485 * Only do this for non-snapshots.
488 (ioflag
& FRSYNC
|| zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
))
489 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
492 * Lock the range against changes.
494 rl
= zfs_range_lock(&zp
->z_range_lock
, uio
->uio_loffset
, uio
->uio_resid
,
498 * If we are reading past end-of-file we can skip
499 * to the end; but we might still need to set atime.
501 if (uio
->uio_loffset
>= zp
->z_size
) {
506 ASSERT(uio
->uio_loffset
< zp
->z_size
);
507 n
= MIN(uio
->uio_resid
, zp
->z_size
- uio
->uio_loffset
);
509 #ifdef HAVE_UIO_ZEROCOPY
510 if ((uio
->uio_extflg
== UIO_XUIO
) &&
511 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
)) {
513 int blksz
= zp
->z_blksz
;
514 uint64_t offset
= uio
->uio_loffset
;
516 xuio
= (xuio_t
*)uio
;
518 nblk
= (P2ROUNDUP(offset
+ n
, blksz
) - P2ALIGN(offset
,
521 ASSERT(offset
+ n
<= blksz
);
524 (void) dmu_xuio_init(xuio
, nblk
);
526 if (vn_has_cached_data(ip
)) {
528 * For simplicity, we always allocate a full buffer
529 * even if we only expect to read a portion of a block.
531 while (--nblk
>= 0) {
532 (void) dmu_xuio_add(xuio
,
533 dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
538 #endif /* HAVE_UIO_ZEROCOPY */
541 nbytes
= MIN(n
, zfs_read_chunk_size
-
542 P2PHASE(uio
->uio_loffset
, zfs_read_chunk_size
));
544 if (zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
)) {
545 error
= mappedread(ip
, nbytes
, uio
);
547 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
552 /* convert checksum errors into IO errors */
554 error
= SET_ERROR(EIO
);
561 zfs_range_unlock(rl
);
568 * Write the bytes to a file.
570 * IN: ip - inode of file to be written to.
571 * uio - structure supplying write location, range info,
573 * ioflag - FAPPEND flag set if in append mode.
574 * O_DIRECT flag; used to bypass page cache.
575 * cr - credentials of caller.
577 * OUT: uio - updated offset and range.
579 * RETURN: 0 if success
580 * error code if failure
583 * ip - ctime|mtime updated if byte count > 0
588 zfs_write(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
590 znode_t
*zp
= ITOZ(ip
);
591 rlim64_t limit
= uio
->uio_limit
;
592 ssize_t start_resid
= uio
->uio_resid
;
596 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
601 int max_blksz
= zfsvfs
->z_max_blksz
;
604 const iovec_t
*aiov
= NULL
;
608 sa_bulk_attr_t bulk
[4];
609 uint64_t mtime
[2], ctime
[2];
611 #ifdef HAVE_UIO_ZEROCOPY
613 const iovec_t
*iovp
= uio
->uio_iov
;
614 ASSERTV(int iovcnt
= uio
->uio_iovcnt
);
618 * Fasttrack empty write
624 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
630 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
631 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
632 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
), NULL
,
634 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
638 * Callers might not be able to detect properly that we are read-only,
639 * so check it explicitly here.
641 if (zfs_is_readonly(zfsvfs
)) {
643 return (SET_ERROR(EROFS
));
647 * If immutable or not appending then return EPERM
649 if ((zp
->z_pflags
& (ZFS_IMMUTABLE
| ZFS_READONLY
)) ||
650 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
651 (uio
->uio_loffset
< zp
->z_size
))) {
653 return (SET_ERROR(EPERM
));
656 zilog
= zfsvfs
->z_log
;
659 * Validate file offset
661 woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
664 return (SET_ERROR(EINVAL
));
668 * Pre-fault the pages to ensure slow (eg NFS) pages
670 * Skip this if uio contains loaned arc_buf.
672 #ifdef HAVE_UIO_ZEROCOPY
673 if ((uio
->uio_extflg
== UIO_XUIO
) &&
674 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
))
675 xuio
= (xuio_t
*)uio
;
678 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
681 * If in append mode, set the io offset pointer to eof.
683 if (ioflag
& FAPPEND
) {
685 * Obtain an appending range lock to guarantee file append
686 * semantics. We reset the write offset once we have the lock.
688 rl
= zfs_range_lock(&zp
->z_range_lock
, 0, n
, RL_APPEND
);
690 if (rl
->r_len
== UINT64_MAX
) {
692 * We overlocked the file because this write will cause
693 * the file block size to increase.
694 * Note that zp_size cannot change with this lock held.
698 uio
->uio_loffset
= woff
;
701 * Note that if the file block size will change as a result of
702 * this write, then this range lock will lock the entire file
703 * so that we can re-write the block safely.
705 rl
= zfs_range_lock(&zp
->z_range_lock
, woff
, n
, RL_WRITER
);
709 zfs_range_unlock(rl
);
711 return (SET_ERROR(EFBIG
));
714 if ((woff
+ n
) > limit
|| woff
> (limit
- n
))
717 /* Will this write extend the file length? */
718 write_eof
= (woff
+ n
> zp
->z_size
);
720 end_size
= MAX(zp
->z_size
, woff
+ n
);
723 * Write the file in reasonable size chunks. Each chunk is written
724 * in a separate transaction; this keeps the intent log records small
725 * and allows us to do more fine-grained space accounting.
729 woff
= uio
->uio_loffset
;
730 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
731 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
733 dmu_return_arcbuf(abuf
);
734 error
= SET_ERROR(EDQUOT
);
738 if (xuio
&& abuf
== NULL
) {
739 #ifdef HAVE_UIO_ZEROCOPY
740 ASSERT(i_iov
< iovcnt
);
741 ASSERT3U(uio
->uio_segflg
, !=, UIO_BVEC
);
743 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
744 dmu_xuio_clear(xuio
, i_iov
);
745 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
746 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
747 aiov
->iov_len
== arc_buf_size(abuf
)));
750 } else if (abuf
== NULL
&& n
>= max_blksz
&&
751 woff
>= zp
->z_size
&&
752 P2PHASE(woff
, max_blksz
) == 0 &&
753 zp
->z_blksz
== max_blksz
) {
755 * This write covers a full block. "Borrow" a buffer
756 * from the dmu so that we can fill it before we enter
757 * a transaction. This avoids the possibility of
758 * holding up the transaction if the data copy hangs
759 * up on a pagefault (e.g., from an NFS server mapping).
763 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
765 ASSERT(abuf
!= NULL
);
766 ASSERT(arc_buf_size(abuf
) == max_blksz
);
767 if ((error
= uiocopy(abuf
->b_data
, max_blksz
,
768 UIO_WRITE
, uio
, &cbytes
))) {
769 dmu_return_arcbuf(abuf
);
772 ASSERT(cbytes
== max_blksz
);
776 * Start a transaction.
778 tx
= dmu_tx_create(zfsvfs
->z_os
);
779 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
780 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
781 zfs_sa_upgrade_txholds(tx
, zp
);
782 error
= dmu_tx_assign(tx
, TXG_WAIT
);
786 dmu_return_arcbuf(abuf
);
791 * If zfs_range_lock() over-locked we grow the blocksize
792 * and then reduce the lock range. This will only happen
793 * on the first iteration since zfs_range_reduce() will
794 * shrink down r_len to the appropriate size.
796 if (rl
->r_len
== UINT64_MAX
) {
799 if (zp
->z_blksz
> max_blksz
) {
801 * File's blocksize is already larger than the
802 * "recordsize" property. Only let it grow to
803 * the next power of 2.
805 ASSERT(!ISP2(zp
->z_blksz
));
806 new_blksz
= MIN(end_size
,
807 1 << highbit64(zp
->z_blksz
));
809 new_blksz
= MIN(end_size
, max_blksz
);
811 zfs_grow_blocksize(zp
, new_blksz
, tx
);
812 zfs_range_reduce(rl
, woff
, n
);
816 * XXX - should we really limit each write to z_max_blksz?
817 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
819 nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
822 tx_bytes
= uio
->uio_resid
;
823 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
825 tx_bytes
-= uio
->uio_resid
;
828 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
830 * If this is not a full block write, but we are
831 * extending the file past EOF and this data starts
832 * block-aligned, use assign_arcbuf(). Otherwise,
833 * write via dmu_write().
835 if (tx_bytes
< max_blksz
&& (!write_eof
||
836 aiov
->iov_base
!= abuf
->b_data
)) {
838 dmu_write(zfsvfs
->z_os
, zp
->z_id
, woff
,
839 /* cppcheck-suppress nullPointer */
840 aiov
->iov_len
, aiov
->iov_base
, tx
);
841 dmu_return_arcbuf(abuf
);
842 xuio_stat_wbuf_copied();
844 ASSERT(xuio
|| tx_bytes
== max_blksz
);
845 dmu_assign_arcbuf(sa_get_db(zp
->z_sa_hdl
),
848 ASSERT(tx_bytes
<= uio
->uio_resid
);
849 uioskip(uio
, tx_bytes
);
851 if (tx_bytes
&& zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
)) {
852 update_pages(ip
, woff
,
853 tx_bytes
, zfsvfs
->z_os
, zp
->z_id
);
857 * If we made no progress, we're done. If we made even
858 * partial progress, update the znode and ZIL accordingly.
861 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
862 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
869 * Clear Set-UID/Set-GID bits on successful write if not
870 * privileged and at least one of the execute bits is set.
872 * It would be nice to to this after all writes have
873 * been done, but that would still expose the ISUID/ISGID
874 * to another app after the partial write is committed.
876 * Note: we don't call zfs_fuid_map_id() here because
877 * user 0 is not an ephemeral uid.
879 mutex_enter(&zp
->z_acl_lock
);
880 uid
= KUID_TO_SUID(ip
->i_uid
);
881 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
882 (S_IXUSR
>> 6))) != 0 &&
883 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
884 secpolicy_vnode_setid_retain(cr
,
885 ((zp
->z_mode
& S_ISUID
) != 0 && uid
== 0)) != 0) {
887 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
888 ip
->i_mode
= newmode
= zp
->z_mode
;
889 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zfsvfs
),
890 (void *)&newmode
, sizeof (uint64_t), tx
);
892 mutex_exit(&zp
->z_acl_lock
);
894 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
);
897 * Update the file size (zp_size) if it has changed;
898 * account for possible concurrent updates.
900 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
901 (void) atomic_cas_64(&zp
->z_size
, end_size
,
906 * If we are replaying and eof is non zero then force
907 * the file size to the specified eof. Note, there's no
908 * concurrency during replay.
910 if (zfsvfs
->z_replay
&& zfsvfs
->z_replay_eof
!= 0)
911 zp
->z_size
= zfsvfs
->z_replay_eof
;
913 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
915 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
,
921 ASSERT(tx_bytes
== nbytes
);
925 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
928 zfs_inode_update(zp
);
929 zfs_range_unlock(rl
);
932 * If we're in replay mode, or we made no progress, return error.
933 * Otherwise, it's at least a partial write, so it's successful.
935 if (zfsvfs
->z_replay
|| uio
->uio_resid
== start_resid
) {
940 if (ioflag
& (FSYNC
| FDSYNC
) ||
941 zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
942 zil_commit(zilog
, zp
->z_id
);
949 * Drop a reference on the passed inode asynchronously. This ensures
950 * that the caller will never drop the last reference on an inode in
951 * the current context. Doing so while holding open a tx could result
952 * in a deadlock if iput_final() re-enters the filesystem code.
955 zfs_iput_async(struct inode
*ip
)
957 objset_t
*os
= ITOZSB(ip
)->z_os
;
959 ASSERT(atomic_read(&ip
->i_count
) > 0);
962 if (atomic_read(&ip
->i_count
) == 1)
963 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os
)),
964 (task_func_t
*)iput
, ip
, TQ_SLEEP
) != TASKQID_INVALID
);
970 zfs_get_done(zgd_t
*zgd
, int error
)
972 znode_t
*zp
= zgd
->zgd_private
;
975 dmu_buf_rele(zgd
->zgd_db
, zgd
);
977 zfs_range_unlock(zgd
->zgd_rl
);
980 * Release the vnode asynchronously as we currently have the
981 * txg stopped from syncing.
983 zfs_iput_async(ZTOI(zp
));
985 if (error
== 0 && zgd
->zgd_bp
)
986 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
988 kmem_free(zgd
, sizeof (zgd_t
));
992 static int zil_fault_io
= 0;
996 * Get data to generate a TX_WRITE intent log record.
999 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
1001 zfsvfs_t
*zfsvfs
= arg
;
1002 objset_t
*os
= zfsvfs
->z_os
;
1004 uint64_t object
= lr
->lr_foid
;
1005 uint64_t offset
= lr
->lr_offset
;
1006 uint64_t size
= lr
->lr_length
;
1011 ASSERT(zio
!= NULL
);
1015 * Nothing to do if the file has been removed
1017 if (zfs_zget(zfsvfs
, object
, &zp
) != 0)
1018 return (SET_ERROR(ENOENT
));
1019 if (zp
->z_unlinked
) {
1021 * Release the vnode asynchronously as we currently have the
1022 * txg stopped from syncing.
1024 zfs_iput_async(ZTOI(zp
));
1025 return (SET_ERROR(ENOENT
));
1028 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1029 zgd
->zgd_zilog
= zfsvfs
->z_log
;
1030 zgd
->zgd_private
= zp
;
1033 * Write records come in two flavors: immediate and indirect.
1034 * For small writes it's cheaper to store the data with the
1035 * log record (immediate); for large writes it's cheaper to
1036 * sync the data and get a pointer to it (indirect) so that
1037 * we don't have to write the data twice.
1039 if (buf
!= NULL
) { /* immediate write */
1040 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
, size
,
1042 /* test for truncation needs to be done while range locked */
1043 if (offset
>= zp
->z_size
) {
1044 error
= SET_ERROR(ENOENT
);
1046 error
= dmu_read(os
, object
, offset
, size
, buf
,
1047 DMU_READ_NO_PREFETCH
);
1049 ASSERT(error
== 0 || error
== ENOENT
);
1050 } else { /* indirect write */
1052 * Have to lock the whole block to ensure when it's
1053 * written out and its checksum is being calculated
1054 * that no one can change the data. We need to re-check
1055 * blocksize after we get the lock in case it's changed!
1060 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1062 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
,
1064 if (zp
->z_blksz
== size
)
1067 zfs_range_unlock(zgd
->zgd_rl
);
1069 /* test for truncation needs to be done while range locked */
1070 if (lr
->lr_offset
>= zp
->z_size
)
1071 error
= SET_ERROR(ENOENT
);
1074 error
= SET_ERROR(EIO
);
1079 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1080 DMU_READ_NO_PREFETCH
);
1083 blkptr_t
*bp
= &lr
->lr_blkptr
;
1088 ASSERT(db
->db_offset
== offset
);
1089 ASSERT(db
->db_size
== size
);
1091 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1093 ASSERT(error
|| lr
->lr_length
<= size
);
1096 * On success, we need to wait for the write I/O
1097 * initiated by dmu_sync() to complete before we can
1098 * release this dbuf. We will finish everything up
1099 * in the zfs_get_done() callback.
1104 if (error
== EALREADY
) {
1105 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
1111 zfs_get_done(zgd
, error
);
1118 zfs_access(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
1120 znode_t
*zp
= ITOZ(ip
);
1121 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
1127 if (flag
& V_ACE_MASK
)
1128 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1130 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1137 * Lookup an entry in a directory, or an extended attribute directory.
1138 * If it exists, return a held inode reference for it.
1140 * IN: dip - inode of directory to search.
1141 * nm - name of entry to lookup.
1142 * flags - LOOKUP_XATTR set if looking for an attribute.
1143 * cr - credentials of caller.
1144 * direntflags - directory lookup flags
1145 * realpnp - returned pathname.
1147 * OUT: ipp - inode of located entry, NULL if not found.
1149 * RETURN: 0 on success, error code on failure.
1156 zfs_lookup(struct inode
*dip
, char *nm
, struct inode
**ipp
, int flags
,
1157 cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1159 znode_t
*zdp
= ITOZ(dip
);
1160 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1164 * Fast path lookup, however we must skip DNLC lookup
1165 * for case folding or normalizing lookups because the
1166 * DNLC code only stores the passed in name. This means
1167 * creating 'a' and removing 'A' on a case insensitive
1168 * file system would work, but DNLC still thinks 'a'
1169 * exists and won't let you create it again on the next
1170 * pass through fast path.
1172 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1174 if (!S_ISDIR(dip
->i_mode
)) {
1175 return (SET_ERROR(ENOTDIR
));
1176 } else if (zdp
->z_sa_hdl
== NULL
) {
1177 return (SET_ERROR(EIO
));
1180 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1181 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1189 } else if (!zdp
->z_zfsvfs
->z_norm
&&
1190 (zdp
->z_zfsvfs
->z_case
== ZFS_CASE_SENSITIVE
)) {
1192 vnode_t
*tvp
= dnlc_lookup(dvp
, nm
);
1195 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1200 if (tvp
== DNLC_NO_VNODE
) {
1202 return (SET_ERROR(ENOENT
));
1205 return (specvp_check(vpp
, cr
));
1208 #endif /* HAVE_DNLC */
1217 if (flags
& LOOKUP_XATTR
) {
1219 * We don't allow recursive attributes..
1220 * Maybe someday we will.
1222 if (zdp
->z_pflags
& ZFS_XATTR
) {
1224 return (SET_ERROR(EINVAL
));
1227 if ((error
= zfs_get_xattrdir(zdp
, ipp
, cr
, flags
))) {
1233 * Do we have permission to get into attribute directory?
1236 if ((error
= zfs_zaccess(ITOZ(*ipp
), ACE_EXECUTE
, 0,
1246 if (!S_ISDIR(dip
->i_mode
)) {
1248 return (SET_ERROR(ENOTDIR
));
1252 * Check accessibility of directory.
1255 if ((error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
))) {
1260 if (zfsvfs
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1261 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1263 return (SET_ERROR(EILSEQ
));
1266 error
= zfs_dirlook(zdp
, nm
, ipp
, flags
, direntflags
, realpnp
);
1267 if ((error
== 0) && (*ipp
))
1268 zfs_inode_update(ITOZ(*ipp
));
1275 * Attempt to create a new entry in a directory. If the entry
1276 * already exists, truncate the file if permissible, else return
1277 * an error. Return the ip of the created or trunc'd file.
1279 * IN: dip - inode of directory to put new file entry in.
1280 * name - name of new file entry.
1281 * vap - attributes of new file.
1282 * excl - flag indicating exclusive or non-exclusive mode.
1283 * mode - mode to open file with.
1284 * cr - credentials of caller.
1285 * flag - large file flag [UNUSED].
1286 * vsecp - ACL to be set
1288 * OUT: ipp - inode of created or trunc'd entry.
1290 * RETURN: 0 on success, error code on failure.
1293 * dip - ctime|mtime updated if new entry created
1294 * ip - ctime|mtime always, atime if new
1299 zfs_create(struct inode
*dip
, char *name
, vattr_t
*vap
, int excl
,
1300 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1302 znode_t
*zp
, *dzp
= ITOZ(dip
);
1303 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1311 zfs_acl_ids_t acl_ids
;
1312 boolean_t fuid_dirtied
;
1313 boolean_t have_acl
= B_FALSE
;
1314 boolean_t waited
= B_FALSE
;
1317 * If we have an ephemeral id, ACL, or XVATTR then
1318 * make sure file system is at proper version
1324 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1325 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1326 return (SET_ERROR(EINVAL
));
1329 return (SET_ERROR(EINVAL
));
1334 zilog
= zfsvfs
->z_log
;
1336 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
1337 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1339 return (SET_ERROR(EILSEQ
));
1342 if (vap
->va_mask
& ATTR_XVATTR
) {
1343 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1344 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1352 if (*name
== '\0') {
1354 * Null component name refers to the directory itself.
1361 /* possible igrab(zp) */
1364 if (flag
& FIGNORECASE
)
1367 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1371 zfs_acl_ids_free(&acl_ids
);
1372 if (strcmp(name
, "..") == 0)
1373 error
= SET_ERROR(EISDIR
);
1383 * Create a new file object and update the directory
1386 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1388 zfs_acl_ids_free(&acl_ids
);
1393 * We only support the creation of regular files in
1394 * extended attribute directories.
1397 if ((dzp
->z_pflags
& ZFS_XATTR
) && !S_ISREG(vap
->va_mode
)) {
1399 zfs_acl_ids_free(&acl_ids
);
1400 error
= SET_ERROR(EINVAL
);
1404 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1405 cr
, vsecp
, &acl_ids
)) != 0)
1409 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1410 zfs_acl_ids_free(&acl_ids
);
1411 error
= SET_ERROR(EDQUOT
);
1415 tx
= dmu_tx_create(os
);
1417 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1418 ZFS_SA_BASE_ATTR_SIZE
);
1420 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1422 zfs_fuid_txhold(zfsvfs
, tx
);
1423 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1424 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1425 if (!zfsvfs
->z_use_sa
&&
1426 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1427 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1428 0, acl_ids
.z_aclp
->z_acl_bytes
);
1430 error
= dmu_tx_assign(tx
,
1431 (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
1433 zfs_dirent_unlock(dl
);
1434 if (error
== ERESTART
) {
1440 zfs_acl_ids_free(&acl_ids
);
1445 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1448 zfs_fuid_sync(zfsvfs
, tx
);
1450 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1451 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1452 if (flag
& FIGNORECASE
)
1454 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1455 vsecp
, acl_ids
.z_fuidp
, vap
);
1456 zfs_acl_ids_free(&acl_ids
);
1459 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1462 zfs_acl_ids_free(&acl_ids
);
1466 * A directory entry already exists for this name.
1469 * Can't truncate an existing file if in exclusive mode.
1472 error
= SET_ERROR(EEXIST
);
1476 * Can't open a directory for writing.
1478 if (S_ISDIR(ZTOI(zp
)->i_mode
)) {
1479 error
= SET_ERROR(EISDIR
);
1483 * Verify requested access to file.
1485 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1489 mutex_enter(&dzp
->z_lock
);
1491 mutex_exit(&dzp
->z_lock
);
1494 * Truncate regular files if requested.
1496 if (S_ISREG(ZTOI(zp
)->i_mode
) &&
1497 (vap
->va_mask
& ATTR_SIZE
) && (vap
->va_size
== 0)) {
1498 /* we can't hold any locks when calling zfs_freesp() */
1500 zfs_dirent_unlock(dl
);
1503 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1509 zfs_dirent_unlock(dl
);
1515 zfs_inode_update(dzp
);
1516 zfs_inode_update(zp
);
1520 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1521 zil_commit(zilog
, 0);
1529 zfs_tmpfile(struct inode
*dip
, vattr_t
*vap
, int excl
,
1530 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1532 znode_t
*zp
= NULL
, *dzp
= ITOZ(dip
);
1533 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1539 zfs_acl_ids_t acl_ids
;
1540 boolean_t fuid_dirtied
;
1541 boolean_t have_acl
= B_FALSE
;
1542 boolean_t waited
= B_FALSE
;
1545 * If we have an ephemeral id, ACL, or XVATTR then
1546 * make sure file system is at proper version
1552 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1553 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1554 return (SET_ERROR(EINVAL
));
1560 if (vap
->va_mask
& ATTR_XVATTR
) {
1561 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1562 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1572 * Create a new file object and update the directory
1575 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1577 zfs_acl_ids_free(&acl_ids
);
1581 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1582 cr
, vsecp
, &acl_ids
)) != 0)
1586 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1587 zfs_acl_ids_free(&acl_ids
);
1588 error
= SET_ERROR(EDQUOT
);
1592 tx
= dmu_tx_create(os
);
1594 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1595 ZFS_SA_BASE_ATTR_SIZE
);
1596 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1598 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1600 zfs_fuid_txhold(zfsvfs
, tx
);
1601 if (!zfsvfs
->z_use_sa
&&
1602 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1603 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1604 0, acl_ids
.z_aclp
->z_acl_bytes
);
1606 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
1608 if (error
== ERESTART
) {
1614 zfs_acl_ids_free(&acl_ids
);
1619 zfs_mknode(dzp
, vap
, tx
, cr
, IS_TMPFILE
, &zp
, &acl_ids
);
1622 zfs_fuid_sync(zfsvfs
, tx
);
1624 /* Add to unlinked set */
1626 zfs_unlinked_add(zp
, tx
);
1627 zfs_acl_ids_free(&acl_ids
);
1635 zfs_inode_update(dzp
);
1636 zfs_inode_update(zp
);
1645 * Remove an entry from a directory.
1647 * IN: dip - inode of directory to remove entry from.
1648 * name - name of entry to remove.
1649 * cr - credentials of caller.
1651 * RETURN: 0 if success
1652 * error code if failure
1656 * ip - ctime (if nlink > 0)
1659 uint64_t null_xattr
= 0;
1663 zfs_remove(struct inode
*dip
, char *name
, cred_t
*cr
, int flags
)
1665 znode_t
*zp
, *dzp
= ITOZ(dip
);
1668 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1670 uint64_t acl_obj
, xattr_obj
;
1671 uint64_t xattr_obj_unlinked
= 0;
1676 boolean_t may_delete_now
, delete_now
= FALSE
;
1677 boolean_t unlinked
, toobig
= FALSE
;
1679 pathname_t
*realnmp
= NULL
;
1683 boolean_t waited
= B_FALSE
;
1686 return (SET_ERROR(EINVAL
));
1690 zilog
= zfsvfs
->z_log
;
1692 if (flags
& FIGNORECASE
) {
1702 * Attempt to lock directory; fail if entry doesn't exist.
1704 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1714 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1719 * Need to use rmdir for removing directories.
1721 if (S_ISDIR(ip
->i_mode
)) {
1722 error
= SET_ERROR(EPERM
);
1728 dnlc_remove(dvp
, realnmp
->pn_buf
);
1730 dnlc_remove(dvp
, name
);
1731 #endif /* HAVE_DNLC */
1733 mutex_enter(&zp
->z_lock
);
1734 may_delete_now
= atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
);
1735 mutex_exit(&zp
->z_lock
);
1738 * We may delete the znode now, or we may put it in the unlinked set;
1739 * it depends on whether we're the last link, and on whether there are
1740 * other holds on the inode. So we dmu_tx_hold() the right things to
1741 * allow for either case.
1744 tx
= dmu_tx_create(zfsvfs
->z_os
);
1745 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1746 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1747 zfs_sa_upgrade_txholds(tx
, zp
);
1748 zfs_sa_upgrade_txholds(tx
, dzp
);
1749 if (may_delete_now
) {
1750 toobig
= zp
->z_size
> zp
->z_blksz
* zfs_delete_blocks
;
1751 /* if the file is too big, only hold_free a token amount */
1752 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1753 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1756 /* are there any extended attributes? */
1757 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1758 &xattr_obj
, sizeof (xattr_obj
));
1759 if (error
== 0 && xattr_obj
) {
1760 error
= zfs_zget(zfsvfs
, xattr_obj
, &xzp
);
1762 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1763 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1766 mutex_enter(&zp
->z_lock
);
1767 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1768 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1769 mutex_exit(&zp
->z_lock
);
1771 /* charge as an update -- would be nice not to charge at all */
1772 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1775 * Mark this transaction as typically resulting in a net free of space
1777 dmu_tx_mark_netfree(tx
);
1779 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
1781 zfs_dirent_unlock(dl
);
1782 if (error
== ERESTART
) {
1802 * Remove the directory entry.
1804 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1813 * Hold z_lock so that we can make sure that the ACL obj
1814 * hasn't changed. Could have been deleted due to
1817 mutex_enter(&zp
->z_lock
);
1818 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1819 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1820 delete_now
= may_delete_now
&& !toobig
&&
1821 atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
) &&
1822 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1827 if (xattr_obj_unlinked
) {
1828 ASSERT3U(ZTOI(xzp
)->i_nlink
, ==, 2);
1829 mutex_enter(&xzp
->z_lock
);
1830 xzp
->z_unlinked
= 1;
1831 clear_nlink(ZTOI(xzp
));
1833 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zfsvfs
),
1834 &links
, sizeof (links
), tx
);
1835 ASSERT3U(error
, ==, 0);
1836 mutex_exit(&xzp
->z_lock
);
1837 zfs_unlinked_add(xzp
, tx
);
1840 error
= sa_remove(zp
->z_sa_hdl
,
1841 SA_ZPL_XATTR(zfsvfs
), tx
);
1843 error
= sa_update(zp
->z_sa_hdl
,
1844 SA_ZPL_XATTR(zfsvfs
), &null_xattr
,
1845 sizeof (uint64_t), tx
);
1849 * Add to the unlinked set because a new reference could be
1850 * taken concurrently resulting in a deferred destruction.
1852 zfs_unlinked_add(zp
, tx
);
1853 mutex_exit(&zp
->z_lock
);
1854 } else if (unlinked
) {
1855 mutex_exit(&zp
->z_lock
);
1856 zfs_unlinked_add(zp
, tx
);
1860 if (flags
& FIGNORECASE
)
1862 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1869 zfs_dirent_unlock(dl
);
1870 zfs_inode_update(dzp
);
1871 zfs_inode_update(zp
);
1879 zfs_inode_update(xzp
);
1880 zfs_iput_async(ZTOI(xzp
));
1883 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1884 zil_commit(zilog
, 0);
1891 * Create a new directory and insert it into dip using the name
1892 * provided. Return a pointer to the inserted directory.
1894 * IN: dip - inode of directory to add subdir to.
1895 * dirname - name of new directory.
1896 * vap - attributes of new directory.
1897 * cr - credentials of caller.
1898 * vsecp - ACL to be set
1900 * OUT: ipp - inode of created directory.
1902 * RETURN: 0 if success
1903 * error code if failure
1906 * dip - ctime|mtime updated
1907 * ipp - ctime|mtime|atime updated
1911 zfs_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
, struct inode
**ipp
,
1912 cred_t
*cr
, int flags
, vsecattr_t
*vsecp
)
1914 znode_t
*zp
, *dzp
= ITOZ(dip
);
1915 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1923 gid_t gid
= crgetgid(cr
);
1924 zfs_acl_ids_t acl_ids
;
1925 boolean_t fuid_dirtied
;
1926 boolean_t waited
= B_FALSE
;
1928 ASSERT(S_ISDIR(vap
->va_mode
));
1931 * If we have an ephemeral id, ACL, or XVATTR then
1932 * make sure file system is at proper version
1936 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1937 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1938 return (SET_ERROR(EINVAL
));
1940 if (dirname
== NULL
)
1941 return (SET_ERROR(EINVAL
));
1945 zilog
= zfsvfs
->z_log
;
1947 if (dzp
->z_pflags
& ZFS_XATTR
) {
1949 return (SET_ERROR(EINVAL
));
1952 if (zfsvfs
->z_utf8
&& u8_validate(dirname
,
1953 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1955 return (SET_ERROR(EILSEQ
));
1957 if (flags
& FIGNORECASE
)
1960 if (vap
->va_mask
& ATTR_XVATTR
) {
1961 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1962 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1968 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1969 vsecp
, &acl_ids
)) != 0) {
1974 * First make sure the new directory doesn't exist.
1976 * Existence is checked first to make sure we don't return
1977 * EACCES instead of EEXIST which can cause some applications
1983 if ((error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1985 zfs_acl_ids_free(&acl_ids
);
1990 if ((error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
))) {
1991 zfs_acl_ids_free(&acl_ids
);
1992 zfs_dirent_unlock(dl
);
1997 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1998 zfs_acl_ids_free(&acl_ids
);
1999 zfs_dirent_unlock(dl
);
2001 return (SET_ERROR(EDQUOT
));
2005 * Add a new entry to the directory.
2007 tx
= dmu_tx_create(zfsvfs
->z_os
);
2008 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
2009 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
2010 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
2012 zfs_fuid_txhold(zfsvfs
, tx
);
2013 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
2014 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
2015 acl_ids
.z_aclp
->z_acl_bytes
);
2018 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
2019 ZFS_SA_BASE_ATTR_SIZE
);
2021 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
2023 zfs_dirent_unlock(dl
);
2024 if (error
== ERESTART
) {
2030 zfs_acl_ids_free(&acl_ids
);
2039 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
2042 zfs_fuid_sync(zfsvfs
, tx
);
2045 * Now put new name in parent dir.
2047 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
2051 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
2052 if (flags
& FIGNORECASE
)
2054 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
2055 acl_ids
.z_fuidp
, vap
);
2057 zfs_acl_ids_free(&acl_ids
);
2061 zfs_dirent_unlock(dl
);
2063 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2064 zil_commit(zilog
, 0);
2066 zfs_inode_update(dzp
);
2067 zfs_inode_update(zp
);
2073 * Remove a directory subdir entry. If the current working
2074 * directory is the same as the subdir to be removed, the
2077 * IN: dip - inode of directory to remove from.
2078 * name - name of directory to be removed.
2079 * cwd - inode of current working directory.
2080 * cr - credentials of caller.
2081 * flags - case flags
2083 * RETURN: 0 on success, error code on failure.
2086 * dip - ctime|mtime updated
2090 zfs_rmdir(struct inode
*dip
, char *name
, struct inode
*cwd
, cred_t
*cr
,
2093 znode_t
*dzp
= ITOZ(dip
);
2096 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
2102 boolean_t waited
= B_FALSE
;
2105 return (SET_ERROR(EINVAL
));
2109 zilog
= zfsvfs
->z_log
;
2111 if (flags
& FIGNORECASE
)
2117 * Attempt to lock directory; fail if entry doesn't exist.
2119 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
2127 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
2131 if (!S_ISDIR(ip
->i_mode
)) {
2132 error
= SET_ERROR(ENOTDIR
);
2137 error
= SET_ERROR(EINVAL
);
2142 * Grab a lock on the directory to make sure that no one is
2143 * trying to add (or lookup) entries while we are removing it.
2145 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
2148 * Grab a lock on the parent pointer to make sure we play well
2149 * with the treewalk and directory rename code.
2151 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
2153 tx
= dmu_tx_create(zfsvfs
->z_os
);
2154 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2155 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2156 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
2157 zfs_sa_upgrade_txholds(tx
, zp
);
2158 zfs_sa_upgrade_txholds(tx
, dzp
);
2159 dmu_tx_mark_netfree(tx
);
2160 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
2162 rw_exit(&zp
->z_parent_lock
);
2163 rw_exit(&zp
->z_name_lock
);
2164 zfs_dirent_unlock(dl
);
2165 if (error
== ERESTART
) {
2178 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2181 uint64_t txtype
= TX_RMDIR
;
2182 if (flags
& FIGNORECASE
)
2184 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2189 rw_exit(&zp
->z_parent_lock
);
2190 rw_exit(&zp
->z_name_lock
);
2192 zfs_dirent_unlock(dl
);
2194 zfs_inode_update(dzp
);
2195 zfs_inode_update(zp
);
2198 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2199 zil_commit(zilog
, 0);
2206 * Read as many directory entries as will fit into the provided
2207 * dirent buffer from the given directory cursor position.
2209 * IN: ip - inode of directory to read.
2210 * dirent - buffer for directory entries.
2212 * OUT: dirent - filler buffer of directory entries.
2214 * RETURN: 0 if success
2215 * error code if failure
2218 * ip - atime updated
2220 * Note that the low 4 bits of the cookie returned by zap is always zero.
2221 * This allows us to use the low range for "special" directory entries:
2222 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2223 * we use the offset 2 for the '.zfs' directory.
2227 zfs_readdir(struct inode
*ip
, zpl_dir_context_t
*ctx
, cred_t
*cr
)
2229 znode_t
*zp
= ITOZ(ip
);
2230 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2233 zap_attribute_t zap
;
2239 uint64_t offset
; /* must be unsigned; checks for < 1 */
2244 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
2245 &parent
, sizeof (parent
))) != 0)
2249 * Quit if directory has been removed (posix)
2257 prefetch
= zp
->z_zn_prefetch
;
2260 * Initialize the iterator cursor.
2264 * Start iteration from the beginning of the directory.
2266 zap_cursor_init(&zc
, os
, zp
->z_id
);
2269 * The offset is a serialized cursor.
2271 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2275 * Transform to file-system independent format
2280 * Special case `.', `..', and `.zfs'.
2283 (void) strcpy(zap
.za_name
, ".");
2284 zap
.za_normalization_conflict
= 0;
2287 } else if (offset
== 1) {
2288 (void) strcpy(zap
.za_name
, "..");
2289 zap
.za_normalization_conflict
= 0;
2292 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2293 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2294 zap
.za_normalization_conflict
= 0;
2295 objnum
= ZFSCTL_INO_ROOT
;
2301 if ((error
= zap_cursor_retrieve(&zc
, &zap
))) {
2302 if (error
== ENOENT
)
2309 * Allow multiple entries provided the first entry is
2310 * the object id. Non-zpl consumers may safely make
2311 * use of the additional space.
2313 * XXX: This should be a feature flag for compatibility
2315 if (zap
.za_integer_length
!= 8 ||
2316 zap
.za_num_integers
== 0) {
2317 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2318 "entry, obj = %lld, offset = %lld, "
2319 "length = %d, num = %lld\n",
2320 (u_longlong_t
)zp
->z_id
,
2321 (u_longlong_t
)offset
,
2322 zap
.za_integer_length
,
2323 (u_longlong_t
)zap
.za_num_integers
);
2324 error
= SET_ERROR(ENXIO
);
2328 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2329 type
= ZFS_DIRENT_TYPE(zap
.za_first_integer
);
2332 done
= !zpl_dir_emit(ctx
, zap
.za_name
, strlen(zap
.za_name
),
2337 /* Prefetch znode */
2339 dmu_prefetch(os
, objnum
, 0, 0, 0,
2340 ZIO_PRIORITY_SYNC_READ
);
2344 * Move to the next entry, fill in the previous offset.
2346 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2347 zap_cursor_advance(&zc
);
2348 offset
= zap_cursor_serialize(&zc
);
2354 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2357 zap_cursor_fini(&zc
);
2358 if (error
== ENOENT
)
2366 ulong_t zfs_fsync_sync_cnt
= 4;
2369 zfs_fsync(struct inode
*ip
, int syncflag
, cred_t
*cr
)
2371 znode_t
*zp
= ITOZ(ip
);
2372 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2374 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2376 if (zfsvfs
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2379 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
2382 tsd_set(zfs_fsyncer_key
, NULL
);
2389 * Get the requested file attributes and place them in the provided
2392 * IN: ip - inode of file.
2393 * vap - va_mask identifies requested attributes.
2394 * If ATTR_XVATTR set, then optional attrs are requested
2395 * flags - ATTR_NOACLCHECK (CIFS server context)
2396 * cr - credentials of caller.
2398 * OUT: vap - attribute values.
2400 * RETURN: 0 (always succeeds)
2404 zfs_getattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2406 znode_t
*zp
= ITOZ(ip
);
2407 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2410 uint64_t atime
[2], mtime
[2], ctime
[2];
2411 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2412 xoptattr_t
*xoap
= NULL
;
2413 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2414 sa_bulk_attr_t bulk
[3];
2420 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2422 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
, &atime
, 16);
2423 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
2424 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
2426 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2432 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2433 * Also, if we are the owner don't bother, since owner should
2434 * always be allowed to read basic attributes of file.
2436 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2437 (vap
->va_uid
!= crgetuid(cr
))) {
2438 if ((error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2446 * Return all attributes. It's cheaper to provide the answer
2447 * than to determine whether we were asked the question.
2450 mutex_enter(&zp
->z_lock
);
2451 vap
->va_type
= vn_mode_to_vtype(zp
->z_mode
);
2452 vap
->va_mode
= zp
->z_mode
;
2453 vap
->va_fsid
= ZTOI(zp
)->i_sb
->s_dev
;
2454 vap
->va_nodeid
= zp
->z_id
;
2455 if ((zp
->z_id
== zfsvfs
->z_root
) && zfs_show_ctldir(zp
))
2456 links
= ZTOI(zp
)->i_nlink
+ 1;
2458 links
= ZTOI(zp
)->i_nlink
;
2459 vap
->va_nlink
= MIN(links
, ZFS_LINK_MAX
);
2460 vap
->va_size
= i_size_read(ip
);
2461 vap
->va_rdev
= ip
->i_rdev
;
2462 vap
->va_seq
= ip
->i_generation
;
2465 * Add in any requested optional attributes and the create time.
2466 * Also set the corresponding bits in the returned attribute bitmap.
2468 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zfsvfs
->z_use_fuids
) {
2469 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2471 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2472 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2475 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2476 xoap
->xoa_readonly
=
2477 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2478 XVA_SET_RTN(xvap
, XAT_READONLY
);
2481 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2483 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2484 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2487 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2489 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2490 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2493 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2494 xoap
->xoa_nounlink
=
2495 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2496 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2499 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2500 xoap
->xoa_immutable
=
2501 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2502 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2505 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2506 xoap
->xoa_appendonly
=
2507 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2508 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2511 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2513 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2514 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2517 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2519 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2520 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2523 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2524 xoap
->xoa_av_quarantined
=
2525 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2526 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2529 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2530 xoap
->xoa_av_modified
=
2531 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2532 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2535 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2536 S_ISREG(ip
->i_mode
)) {
2537 zfs_sa_get_scanstamp(zp
, xvap
);
2540 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2543 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zfsvfs
),
2544 times
, sizeof (times
));
2545 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2546 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2549 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2550 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2551 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2553 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2554 xoap
->xoa_generation
= ip
->i_generation
;
2555 XVA_SET_RTN(xvap
, XAT_GEN
);
2558 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2560 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2561 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2564 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2566 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2567 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2571 ZFS_TIME_DECODE(&vap
->va_atime
, atime
);
2572 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2573 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2575 mutex_exit(&zp
->z_lock
);
2577 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2579 if (zp
->z_blksz
== 0) {
2581 * Block size hasn't been set; suggest maximal I/O transfers.
2583 vap
->va_blksize
= zfsvfs
->z_max_blksz
;
2591 * Get the basic file attributes and place them in the provided kstat
2592 * structure. The inode is assumed to be the authoritative source
2593 * for most of the attributes. However, the znode currently has the
2594 * authoritative atime, blksize, and block count.
2596 * IN: ip - inode of file.
2598 * OUT: sp - kstat values.
2600 * RETURN: 0 (always succeeds)
2604 zfs_getattr_fast(struct inode
*ip
, struct kstat
*sp
)
2606 znode_t
*zp
= ITOZ(ip
);
2607 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2609 u_longlong_t nblocks
;
2614 mutex_enter(&zp
->z_lock
);
2616 generic_fillattr(ip
, sp
);
2618 sa_object_size(zp
->z_sa_hdl
, &blksize
, &nblocks
);
2619 sp
->blksize
= blksize
;
2620 sp
->blocks
= nblocks
;
2622 if (unlikely(zp
->z_blksz
== 0)) {
2624 * Block size hasn't been set; suggest maximal I/O transfers.
2626 sp
->blksize
= zfsvfs
->z_max_blksz
;
2629 mutex_exit(&zp
->z_lock
);
2632 * Required to prevent NFS client from detecting different inode
2633 * numbers of snapshot root dentry before and after snapshot mount.
2635 if (zfsvfs
->z_issnap
) {
2636 if (ip
->i_sb
->s_root
->d_inode
== ip
)
2637 sp
->ino
= ZFSCTL_INO_SNAPDIRS
-
2638 dmu_objset_id(zfsvfs
->z_os
);
2647 * Set the file attributes to the values contained in the
2650 * IN: ip - inode of file to be modified.
2651 * vap - new attribute values.
2652 * If ATTR_XVATTR set, then optional attrs are being set
2653 * flags - ATTR_UTIME set if non-default time values provided.
2654 * - ATTR_NOACLCHECK (CIFS context only).
2655 * cr - credentials of caller.
2657 * RETURN: 0 if success
2658 * error code if failure
2661 * ip - ctime updated, mtime updated if size changed.
2665 zfs_setattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2667 znode_t
*zp
= ITOZ(ip
);
2668 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2672 xvattr_t
*tmpxvattr
;
2673 uint_t mask
= vap
->va_mask
;
2674 uint_t saved_mask
= 0;
2677 uint64_t new_kuid
= 0, new_kgid
= 0, new_uid
, new_gid
;
2679 uint64_t mtime
[2], ctime
[2], atime
[2];
2681 int need_policy
= FALSE
;
2683 zfs_fuid_info_t
*fuidp
= NULL
;
2684 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2687 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2688 boolean_t fuid_dirtied
= B_FALSE
;
2689 sa_bulk_attr_t
*bulk
, *xattr_bulk
;
2690 int count
= 0, xattr_count
= 0;
2698 zilog
= zfsvfs
->z_log
;
2701 * Make sure that if we have ephemeral uid/gid or xvattr specified
2702 * that file system is at proper version level
2705 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
2706 (((mask
& ATTR_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2707 ((mask
& ATTR_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2708 (mask
& ATTR_XVATTR
))) {
2710 return (SET_ERROR(EINVAL
));
2713 if (mask
& ATTR_SIZE
&& S_ISDIR(ip
->i_mode
)) {
2715 return (SET_ERROR(EISDIR
));
2718 if (mask
& ATTR_SIZE
&& !S_ISREG(ip
->i_mode
) && !S_ISFIFO(ip
->i_mode
)) {
2720 return (SET_ERROR(EINVAL
));
2724 * If this is an xvattr_t, then get a pointer to the structure of
2725 * optional attributes. If this is NULL, then we have a vattr_t.
2727 xoap
= xva_getxoptattr(xvap
);
2729 tmpxvattr
= kmem_alloc(sizeof (xvattr_t
), KM_SLEEP
);
2730 xva_init(tmpxvattr
);
2732 bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2733 xattr_bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2736 * Immutable files can only alter immutable bit and atime
2738 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2739 ((mask
& (ATTR_SIZE
|ATTR_UID
|ATTR_GID
|ATTR_MTIME
|ATTR_MODE
)) ||
2740 ((mask
& ATTR_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2745 if ((mask
& ATTR_SIZE
) && (zp
->z_pflags
& ZFS_READONLY
)) {
2751 * Verify timestamps doesn't overflow 32 bits.
2752 * ZFS can handle large timestamps, but 32bit syscalls can't
2753 * handle times greater than 2039. This check should be removed
2754 * once large timestamps are fully supported.
2756 if (mask
& (ATTR_ATIME
| ATTR_MTIME
)) {
2757 if (((mask
& ATTR_ATIME
) &&
2758 TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2759 ((mask
& ATTR_MTIME
) &&
2760 TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2770 /* Can this be moved to before the top label? */
2771 if (zfs_is_readonly(zfsvfs
)) {
2777 * First validate permissions
2780 if (mask
& ATTR_SIZE
) {
2781 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2786 * XXX - Note, we are not providing any open
2787 * mode flags here (like FNDELAY), so we may
2788 * block if there are locks present... this
2789 * should be addressed in openat().
2791 /* XXX - would it be OK to generate a log record here? */
2792 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2797 if (mask
& (ATTR_ATIME
|ATTR_MTIME
) ||
2798 ((mask
& ATTR_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2799 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2800 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2801 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2802 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2803 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2804 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2805 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2809 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2810 int idmask
= (mask
& (ATTR_UID
|ATTR_GID
));
2815 * NOTE: even if a new mode is being set,
2816 * we may clear S_ISUID/S_ISGID bits.
2819 if (!(mask
& ATTR_MODE
))
2820 vap
->va_mode
= zp
->z_mode
;
2823 * Take ownership or chgrp to group we are a member of
2826 take_owner
= (mask
& ATTR_UID
) && (vap
->va_uid
== crgetuid(cr
));
2827 take_group
= (mask
& ATTR_GID
) &&
2828 zfs_groupmember(zfsvfs
, vap
->va_gid
, cr
);
2831 * If both ATTR_UID and ATTR_GID are set then take_owner and
2832 * take_group must both be set in order to allow taking
2835 * Otherwise, send the check through secpolicy_vnode_setattr()
2839 if (((idmask
== (ATTR_UID
|ATTR_GID
)) &&
2840 take_owner
&& take_group
) ||
2841 ((idmask
== ATTR_UID
) && take_owner
) ||
2842 ((idmask
== ATTR_GID
) && take_group
)) {
2843 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2844 skipaclchk
, cr
) == 0) {
2846 * Remove setuid/setgid for non-privileged users
2848 (void) secpolicy_setid_clear(vap
, cr
);
2849 trim_mask
= (mask
& (ATTR_UID
|ATTR_GID
));
2858 mutex_enter(&zp
->z_lock
);
2859 oldva
.va_mode
= zp
->z_mode
;
2860 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2861 if (mask
& ATTR_XVATTR
) {
2863 * Update xvattr mask to include only those attributes
2864 * that are actually changing.
2866 * the bits will be restored prior to actually setting
2867 * the attributes so the caller thinks they were set.
2869 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2870 if (xoap
->xoa_appendonly
!=
2871 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2874 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2875 XVA_SET_REQ(tmpxvattr
, XAT_APPENDONLY
);
2879 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2880 if (xoap
->xoa_nounlink
!=
2881 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2884 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2885 XVA_SET_REQ(tmpxvattr
, XAT_NOUNLINK
);
2889 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2890 if (xoap
->xoa_immutable
!=
2891 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2894 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2895 XVA_SET_REQ(tmpxvattr
, XAT_IMMUTABLE
);
2899 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2900 if (xoap
->xoa_nodump
!=
2901 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2904 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2905 XVA_SET_REQ(tmpxvattr
, XAT_NODUMP
);
2909 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2910 if (xoap
->xoa_av_modified
!=
2911 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2914 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2915 XVA_SET_REQ(tmpxvattr
, XAT_AV_MODIFIED
);
2919 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2920 if ((!S_ISREG(ip
->i_mode
) &&
2921 xoap
->xoa_av_quarantined
) ||
2922 xoap
->xoa_av_quarantined
!=
2923 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2926 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2927 XVA_SET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
);
2931 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2932 mutex_exit(&zp
->z_lock
);
2937 if (need_policy
== FALSE
&&
2938 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
2939 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
2944 mutex_exit(&zp
->z_lock
);
2946 if (mask
& ATTR_MODE
) {
2947 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
2948 err
= secpolicy_setid_setsticky_clear(ip
, vap
,
2953 trim_mask
|= ATTR_MODE
;
2961 * If trim_mask is set then take ownership
2962 * has been granted or write_acl is present and user
2963 * has the ability to modify mode. In that case remove
2964 * UID|GID and or MODE from mask so that
2965 * secpolicy_vnode_setattr() doesn't revoke it.
2969 saved_mask
= vap
->va_mask
;
2970 vap
->va_mask
&= ~trim_mask
;
2972 err
= secpolicy_vnode_setattr(cr
, ip
, vap
, &oldva
, flags
,
2973 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
2978 vap
->va_mask
|= saved_mask
;
2982 * secpolicy_vnode_setattr, or take ownership may have
2985 mask
= vap
->va_mask
;
2987 if ((mask
& (ATTR_UID
| ATTR_GID
))) {
2988 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
2989 &xattr_obj
, sizeof (xattr_obj
));
2991 if (err
== 0 && xattr_obj
) {
2992 err
= zfs_zget(ZTOZSB(zp
), xattr_obj
, &attrzp
);
2996 if (mask
& ATTR_UID
) {
2997 new_kuid
= zfs_fuid_create(zfsvfs
,
2998 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
2999 if (new_kuid
!= KUID_TO_SUID(ZTOI(zp
)->i_uid
) &&
3000 zfs_fuid_overquota(zfsvfs
, B_FALSE
, new_kuid
)) {
3008 if (mask
& ATTR_GID
) {
3009 new_kgid
= zfs_fuid_create(zfsvfs
,
3010 (uint64_t)vap
->va_gid
, cr
, ZFS_GROUP
, &fuidp
);
3011 if (new_kgid
!= KGID_TO_SGID(ZTOI(zp
)->i_gid
) &&
3012 zfs_fuid_overquota(zfsvfs
, B_TRUE
, new_kgid
)) {
3020 tx
= dmu_tx_create(zfsvfs
->z_os
);
3022 if (mask
& ATTR_MODE
) {
3023 uint64_t pmode
= zp
->z_mode
;
3025 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
3027 zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
);
3029 mutex_enter(&zp
->z_lock
);
3030 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
3032 * Are we upgrading ACL from old V0 format
3035 if (zfsvfs
->z_version
>= ZPL_VERSION_FUID
&&
3036 zfs_znode_acl_version(zp
) ==
3037 ZFS_ACL_VERSION_INITIAL
) {
3038 dmu_tx_hold_free(tx
, acl_obj
, 0,
3040 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3041 0, aclp
->z_acl_bytes
);
3043 dmu_tx_hold_write(tx
, acl_obj
, 0,
3046 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3047 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3048 0, aclp
->z_acl_bytes
);
3050 mutex_exit(&zp
->z_lock
);
3051 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3053 if ((mask
& ATTR_XVATTR
) &&
3054 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3055 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3057 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3061 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
3064 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3066 zfs_fuid_txhold(zfsvfs
, tx
);
3068 zfs_sa_upgrade_txholds(tx
, zp
);
3070 err
= dmu_tx_assign(tx
, TXG_WAIT
);
3076 * Set each attribute requested.
3077 * We group settings according to the locks they need to acquire.
3079 * Note: you cannot set ctime directly, although it will be
3080 * updated as a side-effect of calling this function.
3084 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3085 mutex_enter(&zp
->z_acl_lock
);
3086 mutex_enter(&zp
->z_lock
);
3088 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
3089 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
3092 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3093 mutex_enter(&attrzp
->z_acl_lock
);
3094 mutex_enter(&attrzp
->z_lock
);
3095 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3096 SA_ZPL_FLAGS(zfsvfs
), NULL
, &attrzp
->z_pflags
,
3097 sizeof (attrzp
->z_pflags
));
3100 if (mask
& (ATTR_UID
|ATTR_GID
)) {
3102 if (mask
& ATTR_UID
) {
3103 ZTOI(zp
)->i_uid
= SUID_TO_KUID(new_kuid
);
3104 new_uid
= zfs_uid_read(ZTOI(zp
));
3105 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zfsvfs
), NULL
,
3106 &new_uid
, sizeof (new_uid
));
3108 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3109 SA_ZPL_UID(zfsvfs
), NULL
, &new_uid
,
3111 ZTOI(attrzp
)->i_uid
= SUID_TO_KUID(new_uid
);
3115 if (mask
& ATTR_GID
) {
3116 ZTOI(zp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3117 new_gid
= zfs_gid_read(ZTOI(zp
));
3118 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zfsvfs
),
3119 NULL
, &new_gid
, sizeof (new_gid
));
3121 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3122 SA_ZPL_GID(zfsvfs
), NULL
, &new_gid
,
3124 ZTOI(attrzp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3127 if (!(mask
& ATTR_MODE
)) {
3128 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
),
3129 NULL
, &new_mode
, sizeof (new_mode
));
3130 new_mode
= zp
->z_mode
;
3132 err
= zfs_acl_chown_setattr(zp
);
3135 err
= zfs_acl_chown_setattr(attrzp
);
3140 if (mask
& ATTR_MODE
) {
3141 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
), NULL
,
3142 &new_mode
, sizeof (new_mode
));
3143 zp
->z_mode
= ZTOI(zp
)->i_mode
= new_mode
;
3144 ASSERT3P(aclp
, !=, NULL
);
3145 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
3147 if (zp
->z_acl_cached
)
3148 zfs_acl_free(zp
->z_acl_cached
);
3149 zp
->z_acl_cached
= aclp
;
3153 if ((mask
& ATTR_ATIME
) || zp
->z_atime_dirty
) {
3154 zp
->z_atime_dirty
= 0;
3155 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
3156 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
,
3157 &atime
, sizeof (atime
));
3160 if (mask
& (ATTR_MTIME
| ATTR_SIZE
)) {
3161 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3162 ZTOI(zp
)->i_mtime
= zpl_inode_timespec_trunc(vap
->va_mtime
,
3163 ZTOI(zp
)->i_sb
->s_time_gran
);
3165 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
3166 mtime
, sizeof (mtime
));
3169 if (mask
& (ATTR_CTIME
| ATTR_SIZE
)) {
3170 ZFS_TIME_ENCODE(&vap
->va_ctime
, ctime
);
3171 ZTOI(zp
)->i_ctime
= zpl_inode_timespec_trunc(vap
->va_ctime
,
3172 ZTOI(zp
)->i_sb
->s_time_gran
);
3173 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3174 ctime
, sizeof (ctime
));
3177 if (attrzp
&& mask
) {
3178 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3179 SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
,
3184 * Do this after setting timestamps to prevent timestamp
3185 * update from toggling bit
3188 if (xoap
&& (mask
& ATTR_XVATTR
)) {
3191 * restore trimmed off masks
3192 * so that return masks can be set for caller.
3195 if (XVA_ISSET_REQ(tmpxvattr
, XAT_APPENDONLY
)) {
3196 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3198 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NOUNLINK
)) {
3199 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3201 if (XVA_ISSET_REQ(tmpxvattr
, XAT_IMMUTABLE
)) {
3202 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3204 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NODUMP
)) {
3205 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3207 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_MODIFIED
)) {
3208 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3210 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
)) {
3211 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3214 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3215 ASSERT(S_ISREG(ip
->i_mode
));
3217 zfs_xvattr_set(zp
, xvap
, tx
);
3221 zfs_fuid_sync(zfsvfs
, tx
);
3224 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3226 mutex_exit(&zp
->z_lock
);
3227 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3228 mutex_exit(&zp
->z_acl_lock
);
3231 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3232 mutex_exit(&attrzp
->z_acl_lock
);
3233 mutex_exit(&attrzp
->z_lock
);
3236 if (err
== 0 && attrzp
) {
3237 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3246 zfs_fuid_info_free(fuidp
);
3254 if (err
== ERESTART
)
3257 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3261 zfs_inode_update(zp
);
3265 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3266 zil_commit(zilog
, 0);
3269 kmem_free(xattr_bulk
, sizeof (sa_bulk_attr_t
) * 7);
3270 kmem_free(bulk
, sizeof (sa_bulk_attr_t
) * 7);
3271 kmem_free(tmpxvattr
, sizeof (xvattr_t
));
3276 typedef struct zfs_zlock
{
3277 krwlock_t
*zl_rwlock
; /* lock we acquired */
3278 znode_t
*zl_znode
; /* znode we held */
3279 struct zfs_zlock
*zl_next
; /* next in list */
3283 * Drop locks and release vnodes that were held by zfs_rename_lock().
3286 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3290 while ((zl
= *zlpp
) != NULL
) {
3291 if (zl
->zl_znode
!= NULL
)
3292 zfs_iput_async(ZTOI(zl
->zl_znode
));
3293 rw_exit(zl
->zl_rwlock
);
3294 *zlpp
= zl
->zl_next
;
3295 kmem_free(zl
, sizeof (*zl
));
3300 * Search back through the directory tree, using the ".." entries.
3301 * Lock each directory in the chain to prevent concurrent renames.
3302 * Fail any attempt to move a directory into one of its own descendants.
3303 * XXX - z_parent_lock can overlap with map or grow locks
3306 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3310 uint64_t rootid
= ZTOZSB(zp
)->z_root
;
3311 uint64_t oidp
= zp
->z_id
;
3312 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3313 krw_t rw
= RW_WRITER
;
3316 * First pass write-locks szp and compares to zp->z_id.
3317 * Later passes read-lock zp and compare to zp->z_parent.
3320 if (!rw_tryenter(rwlp
, rw
)) {
3322 * Another thread is renaming in this path.
3323 * Note that if we are a WRITER, we don't have any
3324 * parent_locks held yet.
3326 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3328 * Drop our locks and restart
3330 zfs_rename_unlock(&zl
);
3334 rwlp
= &szp
->z_parent_lock
;
3339 * Wait for other thread to drop its locks
3345 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3346 zl
->zl_rwlock
= rwlp
;
3347 zl
->zl_znode
= NULL
;
3348 zl
->zl_next
= *zlpp
;
3351 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3352 return (SET_ERROR(EINVAL
));
3354 if (oidp
== rootid
) /* We've hit the top */
3357 if (rw
== RW_READER
) { /* i.e. not the first pass */
3358 int error
= zfs_zget(ZTOZSB(zp
), oidp
, &zp
);
3363 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(ZTOZSB(zp
)),
3364 &oidp
, sizeof (oidp
));
3365 rwlp
= &zp
->z_parent_lock
;
3368 } while (zp
->z_id
!= sdzp
->z_id
);
3374 * Move an entry from the provided source directory to the target
3375 * directory. Change the entry name as indicated.
3377 * IN: sdip - Source directory containing the "old entry".
3378 * snm - Old entry name.
3379 * tdip - Target directory to contain the "new entry".
3380 * tnm - New entry name.
3381 * cr - credentials of caller.
3382 * flags - case flags
3384 * RETURN: 0 on success, error code on failure.
3387 * sdip,tdip - ctime|mtime updated
3391 zfs_rename(struct inode
*sdip
, char *snm
, struct inode
*tdip
, char *tnm
,
3392 cred_t
*cr
, int flags
)
3394 znode_t
*tdzp
, *szp
, *tzp
;
3395 znode_t
*sdzp
= ITOZ(sdip
);
3396 zfsvfs_t
*zfsvfs
= ITOZSB(sdip
);
3398 zfs_dirlock_t
*sdl
, *tdl
;
3401 int cmp
, serr
, terr
;
3404 boolean_t waited
= B_FALSE
;
3406 if (snm
== NULL
|| tnm
== NULL
)
3407 return (SET_ERROR(EINVAL
));
3410 ZFS_VERIFY_ZP(sdzp
);
3411 zilog
= zfsvfs
->z_log
;
3414 ZFS_VERIFY_ZP(tdzp
);
3417 * We check i_sb because snapshots and the ctldir must have different
3420 if (tdip
->i_sb
!= sdip
->i_sb
|| zfsctl_is_node(tdip
)) {
3422 return (SET_ERROR(EXDEV
));
3425 if (zfsvfs
->z_utf8
&& u8_validate(tnm
,
3426 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3428 return (SET_ERROR(EILSEQ
));
3431 if (flags
& FIGNORECASE
)
3440 * This is to prevent the creation of links into attribute space
3441 * by renaming a linked file into/outof an attribute directory.
3442 * See the comment in zfs_link() for why this is considered bad.
3444 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3446 return (SET_ERROR(EINVAL
));
3450 * Lock source and target directory entries. To prevent deadlock,
3451 * a lock ordering must be defined. We lock the directory with
3452 * the smallest object id first, or if it's a tie, the one with
3453 * the lexically first name.
3455 if (sdzp
->z_id
< tdzp
->z_id
) {
3457 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3461 * First compare the two name arguments without
3462 * considering any case folding.
3464 int nofold
= (zfsvfs
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3466 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3467 ASSERT(error
== 0 || !zfsvfs
->z_utf8
);
3470 * POSIX: "If the old argument and the new argument
3471 * both refer to links to the same existing file,
3472 * the rename() function shall return successfully
3473 * and perform no other action."
3479 * If the file system is case-folding, then we may
3480 * have some more checking to do. A case-folding file
3481 * system is either supporting mixed case sensitivity
3482 * access or is completely case-insensitive. Note
3483 * that the file system is always case preserving.
3485 * In mixed sensitivity mode case sensitive behavior
3486 * is the default. FIGNORECASE must be used to
3487 * explicitly request case insensitive behavior.
3489 * If the source and target names provided differ only
3490 * by case (e.g., a request to rename 'tim' to 'Tim'),
3491 * we will treat this as a special case in the
3492 * case-insensitive mode: as long as the source name
3493 * is an exact match, we will allow this to proceed as
3494 * a name-change request.
3496 if ((zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
||
3497 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&&
3498 flags
& FIGNORECASE
)) &&
3499 u8_strcmp(snm
, tnm
, 0, zfsvfs
->z_norm
, U8_UNICODE_LATEST
,
3502 * case preserving rename request, require exact
3511 * If the source and destination directories are the same, we should
3512 * grab the z_name_lock of that directory only once.
3516 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3520 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3521 ZEXISTS
| zflg
, NULL
, NULL
);
3522 terr
= zfs_dirent_lock(&tdl
,
3523 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3525 terr
= zfs_dirent_lock(&tdl
,
3526 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3527 serr
= zfs_dirent_lock(&sdl
,
3528 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3534 * Source entry invalid or not there.
3537 zfs_dirent_unlock(tdl
);
3543 rw_exit(&sdzp
->z_name_lock
);
3545 if (strcmp(snm
, "..") == 0)
3551 zfs_dirent_unlock(sdl
);
3555 rw_exit(&sdzp
->z_name_lock
);
3557 if (strcmp(tnm
, "..") == 0)
3564 * Must have write access at the source to remove the old entry
3565 * and write access at the target to create the new entry.
3566 * Note that if target and source are the same, this can be
3567 * done in a single check.
3570 if ((error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
)))
3573 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3575 * Check to make sure rename is valid.
3576 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3578 if ((error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
)))
3583 * Does target exist?
3587 * Source and target must be the same type.
3589 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3590 if (!S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3591 error
= SET_ERROR(ENOTDIR
);
3595 if (S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3596 error
= SET_ERROR(EISDIR
);
3601 * POSIX dictates that when the source and target
3602 * entries refer to the same file object, rename
3603 * must do nothing and exit without error.
3605 if (szp
->z_id
== tzp
->z_id
) {
3611 tx
= dmu_tx_create(zfsvfs
->z_os
);
3612 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3613 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3614 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3615 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3617 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3618 zfs_sa_upgrade_txholds(tx
, tdzp
);
3621 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3622 zfs_sa_upgrade_txholds(tx
, tzp
);
3625 zfs_sa_upgrade_txholds(tx
, szp
);
3626 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
3627 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
3630 zfs_rename_unlock(&zl
);
3631 zfs_dirent_unlock(sdl
);
3632 zfs_dirent_unlock(tdl
);
3635 rw_exit(&sdzp
->z_name_lock
);
3637 if (error
== ERESTART
) {
3654 if (tzp
) /* Attempt to remove the existing target */
3655 error
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3658 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3660 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3662 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zfsvfs
),
3663 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3666 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3668 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3669 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3670 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3673 * At this point, we have successfully created
3674 * the target name, but have failed to remove
3675 * the source name. Since the create was done
3676 * with the ZRENAMING flag, there are
3677 * complications; for one, the link count is
3678 * wrong. The easiest way to deal with this
3679 * is to remove the newly created target, and
3680 * return the original error. This must
3681 * succeed; fortunately, it is very unlikely to
3682 * fail, since we just created it.
3684 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3685 ZRENAMING
, NULL
), ==, 0);
3693 zfs_rename_unlock(&zl
);
3695 zfs_dirent_unlock(sdl
);
3696 zfs_dirent_unlock(tdl
);
3698 zfs_inode_update(sdzp
);
3700 rw_exit(&sdzp
->z_name_lock
);
3703 zfs_inode_update(tdzp
);
3705 zfs_inode_update(szp
);
3708 zfs_inode_update(tzp
);
3712 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3713 zil_commit(zilog
, 0);
3720 * Insert the indicated symbolic reference entry into the directory.
3722 * IN: dip - Directory to contain new symbolic link.
3723 * link - Name for new symlink entry.
3724 * vap - Attributes of new entry.
3725 * target - Target path of new symlink.
3727 * cr - credentials of caller.
3728 * flags - case flags
3730 * RETURN: 0 on success, error code on failure.
3733 * dip - ctime|mtime updated
3737 zfs_symlink(struct inode
*dip
, char *name
, vattr_t
*vap
, char *link
,
3738 struct inode
**ipp
, cred_t
*cr
, int flags
)
3740 znode_t
*zp
, *dzp
= ITOZ(dip
);
3743 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
3745 uint64_t len
= strlen(link
);
3748 zfs_acl_ids_t acl_ids
;
3749 boolean_t fuid_dirtied
;
3750 uint64_t txtype
= TX_SYMLINK
;
3751 boolean_t waited
= B_FALSE
;
3753 ASSERT(S_ISLNK(vap
->va_mode
));
3756 return (SET_ERROR(EINVAL
));
3760 zilog
= zfsvfs
->z_log
;
3762 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
3763 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3765 return (SET_ERROR(EILSEQ
));
3767 if (flags
& FIGNORECASE
)
3770 if (len
> MAXPATHLEN
) {
3772 return (SET_ERROR(ENAMETOOLONG
));
3775 if ((error
= zfs_acl_ids_create(dzp
, 0,
3776 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3784 * Attempt to lock directory; fail if entry already exists.
3786 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3788 zfs_acl_ids_free(&acl_ids
);
3793 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3794 zfs_acl_ids_free(&acl_ids
);
3795 zfs_dirent_unlock(dl
);
3800 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
3801 zfs_acl_ids_free(&acl_ids
);
3802 zfs_dirent_unlock(dl
);
3804 return (SET_ERROR(EDQUOT
));
3806 tx
= dmu_tx_create(zfsvfs
->z_os
);
3807 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3808 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3809 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3810 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3811 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3812 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3813 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3814 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3815 acl_ids
.z_aclp
->z_acl_bytes
);
3818 zfs_fuid_txhold(zfsvfs
, tx
);
3819 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
3821 zfs_dirent_unlock(dl
);
3822 if (error
== ERESTART
) {
3828 zfs_acl_ids_free(&acl_ids
);
3835 * Create a new object for the symlink.
3836 * for version 4 ZPL datsets the symlink will be an SA attribute
3838 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3841 zfs_fuid_sync(zfsvfs
, tx
);
3843 mutex_enter(&zp
->z_lock
);
3845 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zfsvfs
),
3848 zfs_sa_symlink(zp
, link
, len
, tx
);
3849 mutex_exit(&zp
->z_lock
);
3852 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
3853 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3855 * Insert the new object into the directory.
3857 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3859 if (flags
& FIGNORECASE
)
3861 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3863 zfs_inode_update(dzp
);
3864 zfs_inode_update(zp
);
3866 zfs_acl_ids_free(&acl_ids
);
3870 zfs_dirent_unlock(dl
);
3874 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3875 zil_commit(zilog
, 0);
3882 * Return, in the buffer contained in the provided uio structure,
3883 * the symbolic path referred to by ip.
3885 * IN: ip - inode of symbolic link
3886 * uio - structure to contain the link path.
3887 * cr - credentials of caller.
3889 * RETURN: 0 if success
3890 * error code if failure
3893 * ip - atime updated
3897 zfs_readlink(struct inode
*ip
, uio_t
*uio
, cred_t
*cr
)
3899 znode_t
*zp
= ITOZ(ip
);
3900 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
3906 mutex_enter(&zp
->z_lock
);
3908 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3909 SA_ZPL_SYMLINK(zfsvfs
), uio
);
3911 error
= zfs_sa_readlink(zp
, uio
);
3912 mutex_exit(&zp
->z_lock
);
3919 * Insert a new entry into directory tdip referencing sip.
3921 * IN: tdip - Directory to contain new entry.
3922 * sip - inode of new entry.
3923 * name - name of new entry.
3924 * cr - credentials of caller.
3926 * RETURN: 0 if success
3927 * error code if failure
3930 * tdip - ctime|mtime updated
3931 * sip - ctime updated
3935 zfs_link(struct inode
*tdip
, struct inode
*sip
, char *name
, cred_t
*cr
,
3938 znode_t
*dzp
= ITOZ(tdip
);
3940 zfsvfs_t
*zfsvfs
= ITOZSB(tdip
);
3948 boolean_t waited
= B_FALSE
;
3949 boolean_t is_tmpfile
= 0;
3952 is_tmpfile
= (sip
->i_nlink
== 0 && (sip
->i_state
& I_LINKABLE
));
3954 ASSERT(S_ISDIR(tdip
->i_mode
));
3957 return (SET_ERROR(EINVAL
));
3961 zilog
= zfsvfs
->z_log
;
3964 * POSIX dictates that we return EPERM here.
3965 * Better choices include ENOTSUP or EISDIR.
3967 if (S_ISDIR(sip
->i_mode
)) {
3969 return (SET_ERROR(EPERM
));
3976 * We check i_sb because snapshots and the ctldir must have different
3979 if (sip
->i_sb
!= tdip
->i_sb
|| zfsctl_is_node(sip
)) {
3981 return (SET_ERROR(EXDEV
));
3984 /* Prevent links to .zfs/shares files */
3986 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
3987 &parent
, sizeof (uint64_t))) != 0) {
3991 if (parent
== zfsvfs
->z_shares_dir
) {
3993 return (SET_ERROR(EPERM
));
3996 if (zfsvfs
->z_utf8
&& u8_validate(name
,
3997 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3999 return (SET_ERROR(EILSEQ
));
4001 if (flags
& FIGNORECASE
)
4005 * We do not support links between attributes and non-attributes
4006 * because of the potential security risk of creating links
4007 * into "normal" file space in order to circumvent restrictions
4008 * imposed in attribute space.
4010 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
4012 return (SET_ERROR(EINVAL
));
4015 owner
= zfs_fuid_map_id(zfsvfs
, KUID_TO_SUID(sip
->i_uid
),
4017 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
4019 return (SET_ERROR(EPERM
));
4022 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
4029 * Attempt to lock directory; fail if entry already exists.
4031 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
4037 tx
= dmu_tx_create(zfsvfs
->z_os
);
4038 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
4039 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
4041 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
4043 zfs_sa_upgrade_txholds(tx
, szp
);
4044 zfs_sa_upgrade_txholds(tx
, dzp
);
4045 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
4047 zfs_dirent_unlock(dl
);
4048 if (error
== ERESTART
) {
4058 /* unmark z_unlinked so zfs_link_create will not reject */
4060 szp
->z_unlinked
= 0;
4061 error
= zfs_link_create(dl
, szp
, tx
, 0);
4064 uint64_t txtype
= TX_LINK
;
4066 * tmpfile is created to be in z_unlinkedobj, so remove it.
4067 * Also, we don't log in ZIL, be cause all previous file
4068 * operation on the tmpfile are ignored by ZIL. Instead we
4069 * always wait for txg to sync to make sure all previous
4070 * operation are sync safe.
4073 VERIFY(zap_remove_int(zfsvfs
->z_os
,
4074 zfsvfs
->z_unlinkedobj
, szp
->z_id
, tx
) == 0);
4076 if (flags
& FIGNORECASE
)
4078 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
4080 } else if (is_tmpfile
) {
4081 /* restore z_unlinked since when linking failed */
4082 szp
->z_unlinked
= 1;
4084 txg
= dmu_tx_get_txg(tx
);
4087 zfs_dirent_unlock(dl
);
4089 if (!is_tmpfile
&& zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4090 zil_commit(zilog
, 0);
4093 txg_wait_synced(dmu_objset_pool(zfsvfs
->z_os
), txg
);
4095 zfs_inode_update(dzp
);
4096 zfs_inode_update(szp
);
4102 zfs_putpage_commit_cb(void *arg
)
4104 struct page
*pp
= arg
;
4107 end_page_writeback(pp
);
4111 * Push a page out to disk, once the page is on stable storage the
4112 * registered commit callback will be run as notification of completion.
4114 * IN: ip - page mapped for inode.
4115 * pp - page to push (page is locked)
4116 * wbc - writeback control data
4118 * RETURN: 0 if success
4119 * error code if failure
4122 * ip - ctime|mtime updated
4126 zfs_putpage(struct inode
*ip
, struct page
*pp
, struct writeback_control
*wbc
)
4128 znode_t
*zp
= ITOZ(ip
);
4129 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4137 uint64_t mtime
[2], ctime
[2];
4138 sa_bulk_attr_t bulk
[3];
4140 struct address_space
*mapping
;
4145 ASSERT(PageLocked(pp
));
4147 pgoff
= page_offset(pp
); /* Page byte-offset in file */
4148 offset
= i_size_read(ip
); /* File length in bytes */
4149 pglen
= MIN(PAGE_SIZE
, /* Page length in bytes */
4150 P2ROUNDUP(offset
, PAGE_SIZE
)-pgoff
);
4152 /* Page is beyond end of file */
4153 if (pgoff
>= offset
) {
4159 /* Truncate page length to end of file */
4160 if (pgoff
+ pglen
> offset
)
4161 pglen
= offset
- pgoff
;
4165 * FIXME: Allow mmap writes past its quota. The correct fix
4166 * is to register a page_mkwrite() handler to count the page
4167 * against its quota when it is about to be dirtied.
4169 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
4170 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
4176 * The ordering here is critical and must adhere to the following
4177 * rules in order to avoid deadlocking in either zfs_read() or
4178 * zfs_free_range() due to a lock inversion.
4180 * 1) The page must be unlocked prior to acquiring the range lock.
4181 * This is critical because zfs_read() calls find_lock_page()
4182 * which may block on the page lock while holding the range lock.
4184 * 2) Before setting or clearing write back on a page the range lock
4185 * must be held in order to prevent a lock inversion with the
4186 * zfs_free_range() function.
4188 * This presents a problem because upon entering this function the
4189 * page lock is already held. To safely acquire the range lock the
4190 * page lock must be dropped. This creates a window where another
4191 * process could truncate, invalidate, dirty, or write out the page.
4193 * Therefore, after successfully reacquiring the range and page locks
4194 * the current page state is checked. In the common case everything
4195 * will be as is expected and it can be written out. However, if
4196 * the page state has changed it must be handled accordingly.
4198 mapping
= pp
->mapping
;
4199 redirty_page_for_writepage(wbc
, pp
);
4202 rl
= zfs_range_lock(&zp
->z_range_lock
, pgoff
, pglen
, RL_WRITER
);
4205 /* Page mapping changed or it was no longer dirty, we're done */
4206 if (unlikely((mapping
!= pp
->mapping
) || !PageDirty(pp
))) {
4208 zfs_range_unlock(rl
);
4213 /* Another process started write block if required */
4214 if (PageWriteback(pp
)) {
4216 zfs_range_unlock(rl
);
4218 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
4219 wait_on_page_writeback(pp
);
4225 /* Clear the dirty flag the required locks are held */
4226 if (!clear_page_dirty_for_io(pp
)) {
4228 zfs_range_unlock(rl
);
4234 * Counterpart for redirty_page_for_writepage() above. This page
4235 * was in fact not skipped and should not be counted as if it were.
4237 wbc
->pages_skipped
--;
4238 set_page_writeback(pp
);
4241 tx
= dmu_tx_create(zfsvfs
->z_os
);
4242 dmu_tx_hold_write(tx
, zp
->z_id
, pgoff
, pglen
);
4243 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4244 zfs_sa_upgrade_txholds(tx
, zp
);
4246 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4248 if (err
== ERESTART
)
4252 __set_page_dirty_nobuffers(pp
);
4254 end_page_writeback(pp
);
4255 zfs_range_unlock(rl
);
4261 ASSERT3U(pglen
, <=, PAGE_SIZE
);
4262 dmu_write(zfsvfs
->z_os
, zp
->z_id
, pgoff
, pglen
, va
, tx
);
4265 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
4266 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
4267 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
4270 /* Preserve the mtime and ctime provided by the inode */
4271 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4272 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4273 zp
->z_atime_dirty
= 0;
4276 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4278 zfs_log_write(zfsvfs
->z_log
, tx
, TX_WRITE
, zp
, pgoff
, pglen
, 0,
4279 zfs_putpage_commit_cb
, pp
);
4282 zfs_range_unlock(rl
);
4284 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4286 * Note that this is rarely called under writepages(), because
4287 * writepages() normally handles the entire commit for
4288 * performance reasons.
4290 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
4298 * Update the system attributes when the inode has been dirtied. For the
4299 * moment we only update the mode, atime, mtime, and ctime.
4302 zfs_dirty_inode(struct inode
*ip
, int flags
)
4304 znode_t
*zp
= ITOZ(ip
);
4305 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4307 uint64_t mode
, atime
[2], mtime
[2], ctime
[2];
4308 sa_bulk_attr_t bulk
[4];
4312 if (zfs_is_readonly(zfsvfs
) || dmu_objset_is_snapshot(zfsvfs
->z_os
))
4320 * This is the lazytime semantic indroduced in Linux 4.0
4321 * This flag will only be called from update_time when lazytime is set.
4322 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4323 * Fortunately mtime and ctime are managed within ZFS itself, so we
4324 * only need to dirty atime.
4326 if (flags
== I_DIRTY_TIME
) {
4327 zp
->z_atime_dirty
= 1;
4332 tx
= dmu_tx_create(zfsvfs
->z_os
);
4334 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4335 zfs_sa_upgrade_txholds(tx
, zp
);
4337 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4343 mutex_enter(&zp
->z_lock
);
4344 zp
->z_atime_dirty
= 0;
4346 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MODE(zfsvfs
), NULL
, &mode
, 8);
4347 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_ATIME(zfsvfs
), NULL
, &atime
, 16);
4348 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
4349 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
4351 /* Preserve the mode, mtime and ctime provided by the inode */
4352 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4353 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4354 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4359 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4360 mutex_exit(&zp
->z_lock
);
4370 zfs_inactive(struct inode
*ip
)
4372 znode_t
*zp
= ITOZ(ip
);
4373 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4376 int need_unlock
= 0;
4378 /* Only read lock if we haven't already write locked, e.g. rollback */
4379 if (!RW_WRITE_HELD(&zfsvfs
->z_teardown_inactive_lock
)) {
4381 rw_enter(&zfsvfs
->z_teardown_inactive_lock
, RW_READER
);
4383 if (zp
->z_sa_hdl
== NULL
) {
4385 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4389 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4390 dmu_tx_t
*tx
= dmu_tx_create(zfsvfs
->z_os
);
4392 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4393 zfs_sa_upgrade_txholds(tx
, zp
);
4394 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4398 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4399 mutex_enter(&zp
->z_lock
);
4400 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zfsvfs
),
4401 (void *)&atime
, sizeof (atime
), tx
);
4402 zp
->z_atime_dirty
= 0;
4403 mutex_exit(&zp
->z_lock
);
4410 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4414 * Bounds-check the seek operation.
4416 * IN: ip - inode seeking within
4417 * ooff - old file offset
4418 * noffp - pointer to new file offset
4419 * ct - caller context
4421 * RETURN: 0 if success
4422 * EINVAL if new offset invalid
4426 zfs_seek(struct inode
*ip
, offset_t ooff
, offset_t
*noffp
)
4428 if (S_ISDIR(ip
->i_mode
))
4430 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4434 * Fill pages with data from the disk.
4437 zfs_fillpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4439 znode_t
*zp
= ITOZ(ip
);
4440 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4442 struct page
*cur_pp
;
4443 u_offset_t io_off
, total
;
4450 io_len
= nr_pages
<< PAGE_SHIFT
;
4451 i_size
= i_size_read(ip
);
4452 io_off
= page_offset(pl
[0]);
4454 if (io_off
+ io_len
> i_size
)
4455 io_len
= i_size
- io_off
;
4458 * Iterate over list of pages and read each page individually.
4461 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4464 cur_pp
= pl
[page_idx
++];
4466 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4470 /* convert checksum errors into IO errors */
4472 err
= SET_ERROR(EIO
);
4481 * Uses zfs_fillpage to read data from the file and fill the pages.
4483 * IN: ip - inode of file to get data from.
4484 * pl - list of pages to read
4485 * nr_pages - number of pages to read
4487 * RETURN: 0 on success, error code on failure.
4490 * vp - atime updated
4494 zfs_getpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4496 znode_t
*zp
= ITOZ(ip
);
4497 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4506 err
= zfs_fillpage(ip
, pl
, nr_pages
);
4513 * Check ZFS specific permissions to memory map a section of a file.
4515 * IN: ip - inode of the file to mmap
4517 * addrp - start address in memory region
4518 * len - length of memory region
4519 * vm_flags- address flags
4521 * RETURN: 0 if success
4522 * error code if failure
4526 zfs_map(struct inode
*ip
, offset_t off
, caddr_t
*addrp
, size_t len
,
4527 unsigned long vm_flags
)
4529 znode_t
*zp
= ITOZ(ip
);
4530 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4535 if ((vm_flags
& VM_WRITE
) && (zp
->z_pflags
&
4536 (ZFS_IMMUTABLE
| ZFS_READONLY
| ZFS_APPENDONLY
))) {
4538 return (SET_ERROR(EPERM
));
4541 if ((vm_flags
& (VM_READ
| VM_EXEC
)) &&
4542 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4544 return (SET_ERROR(EACCES
));
4547 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4549 return (SET_ERROR(ENXIO
));
4557 * convoff - converts the given data (start, whence) to the
4561 convoff(struct inode
*ip
, flock64_t
*lckdat
, int whence
, offset_t offset
)
4566 if ((lckdat
->l_whence
== 2) || (whence
== 2)) {
4567 if ((error
= zfs_getattr(ip
, &vap
, 0, CRED())))
4571 switch (lckdat
->l_whence
) {
4573 lckdat
->l_start
+= offset
;
4576 lckdat
->l_start
+= vap
.va_size
;
4581 return (SET_ERROR(EINVAL
));
4584 if (lckdat
->l_start
< 0)
4585 return (SET_ERROR(EINVAL
));
4589 lckdat
->l_start
-= offset
;
4592 lckdat
->l_start
-= vap
.va_size
;
4597 return (SET_ERROR(EINVAL
));
4600 lckdat
->l_whence
= (short)whence
;
4605 * Free or allocate space in a file. Currently, this function only
4606 * supports the `F_FREESP' command. However, this command is somewhat
4607 * misnamed, as its functionality includes the ability to allocate as
4608 * well as free space.
4610 * IN: ip - inode of file to free data in.
4611 * cmd - action to take (only F_FREESP supported).
4612 * bfp - section of file to free/alloc.
4613 * flag - current file open mode flags.
4614 * offset - current file offset.
4615 * cr - credentials of caller [UNUSED].
4617 * RETURN: 0 on success, error code on failure.
4620 * ip - ctime|mtime updated
4624 zfs_space(struct inode
*ip
, int cmd
, flock64_t
*bfp
, int flag
,
4625 offset_t offset
, cred_t
*cr
)
4627 znode_t
*zp
= ITOZ(ip
);
4628 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4635 if (cmd
!= F_FREESP
) {
4637 return (SET_ERROR(EINVAL
));
4641 * Callers might not be able to detect properly that we are read-only,
4642 * so check it explicitly here.
4644 if (zfs_is_readonly(zfsvfs
)) {
4646 return (SET_ERROR(EROFS
));
4649 if ((error
= convoff(ip
, bfp
, 0, offset
))) {
4654 if (bfp
->l_len
< 0) {
4656 return (SET_ERROR(EINVAL
));
4660 * Permissions aren't checked on Solaris because on this OS
4661 * zfs_space() can only be called with an opened file handle.
4662 * On Linux we can get here through truncate_range() which
4663 * operates directly on inodes, so we need to check access rights.
4665 if ((error
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
))) {
4671 len
= bfp
->l_len
; /* 0 means from off to end of file */
4673 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4681 zfs_fid(struct inode
*ip
, fid_t
*fidp
)
4683 znode_t
*zp
= ITOZ(ip
);
4684 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4687 uint64_t object
= zp
->z_id
;
4694 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zfsvfs
),
4695 &gen64
, sizeof (uint64_t))) != 0) {
4700 gen
= (uint32_t)gen64
;
4702 size
= SHORT_FID_LEN
;
4704 zfid
= (zfid_short_t
*)fidp
;
4706 zfid
->zf_len
= size
;
4708 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4709 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4711 /* Must have a non-zero generation number to distinguish from .zfs */
4714 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4715 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4723 zfs_getsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4725 znode_t
*zp
= ITOZ(ip
);
4726 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4728 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4732 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
4740 zfs_setsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4742 znode_t
*zp
= ITOZ(ip
);
4743 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4745 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4746 zilog_t
*zilog
= zfsvfs
->z_log
;
4751 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
4753 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4754 zil_commit(zilog
, 0);
4760 #ifdef HAVE_UIO_ZEROCOPY
4762 * Tunable, both must be a power of 2.
4764 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4765 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4766 * an arcbuf for a partial block read
4768 int zcr_blksz_min
= (1 << 10); /* 1K */
4769 int zcr_blksz_max
= (1 << 17); /* 128K */
4773 zfs_reqzcbuf(struct inode
*ip
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
)
4775 znode_t
*zp
= ITOZ(ip
);
4776 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4777 int max_blksz
= zfsvfs
->z_max_blksz
;
4778 uio_t
*uio
= &xuio
->xu_uio
;
4779 ssize_t size
= uio
->uio_resid
;
4780 offset_t offset
= uio
->uio_loffset
;
4785 int preamble
, postamble
;
4787 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
4788 return (SET_ERROR(EINVAL
));
4795 * Loan out an arc_buf for write if write size is bigger than
4796 * max_blksz, and the file's block size is also max_blksz.
4799 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
4801 return (SET_ERROR(EINVAL
));
4804 * Caller requests buffers for write before knowing where the
4805 * write offset might be (e.g. NFS TCP write).
4810 preamble
= P2PHASE(offset
, blksz
);
4812 preamble
= blksz
- preamble
;
4817 postamble
= P2PHASE(size
, blksz
);
4820 fullblk
= size
/ blksz
;
4821 (void) dmu_xuio_init(xuio
,
4822 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
4825 * Have to fix iov base/len for partial buffers. They
4826 * currently represent full arc_buf's.
4829 /* data begins in the middle of the arc_buf */
4830 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4833 (void) dmu_xuio_add(xuio
, abuf
,
4834 blksz
- preamble
, preamble
);
4837 for (i
= 0; i
< fullblk
; i
++) {
4838 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4841 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
4845 /* data ends in the middle of the arc_buf */
4846 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4849 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
4854 * Loan out an arc_buf for read if the read size is larger than
4855 * the current file block size. Block alignment is not
4856 * considered. Partial arc_buf will be loaned out for read.
4858 blksz
= zp
->z_blksz
;
4859 if (blksz
< zcr_blksz_min
)
4860 blksz
= zcr_blksz_min
;
4861 if (blksz
> zcr_blksz_max
)
4862 blksz
= zcr_blksz_max
;
4863 /* avoid potential complexity of dealing with it */
4864 if (blksz
> max_blksz
) {
4866 return (SET_ERROR(EINVAL
));
4869 maxsize
= zp
->z_size
- uio
->uio_loffset
;
4875 return (SET_ERROR(EINVAL
));
4880 return (SET_ERROR(EINVAL
));
4883 uio
->uio_extflg
= UIO_XUIO
;
4884 XUIO_XUZC_RW(xuio
) = ioflag
;
4891 zfs_retzcbuf(struct inode
*ip
, xuio_t
*xuio
, cred_t
*cr
)
4895 int ioflag
= XUIO_XUZC_RW(xuio
);
4897 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
4899 i
= dmu_xuio_cnt(xuio
);
4901 abuf
= dmu_xuio_arcbuf(xuio
, i
);
4903 * if abuf == NULL, it must be a write buffer
4904 * that has been returned in zfs_write().
4907 dmu_return_arcbuf(abuf
);
4908 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
4911 dmu_xuio_fini(xuio
);
4914 #endif /* HAVE_UIO_ZEROCOPY */
4916 #if defined(_KERNEL) && defined(HAVE_SPL)
4917 EXPORT_SYMBOL(zfs_open
);
4918 EXPORT_SYMBOL(zfs_close
);
4919 EXPORT_SYMBOL(zfs_read
);
4920 EXPORT_SYMBOL(zfs_write
);
4921 EXPORT_SYMBOL(zfs_access
);
4922 EXPORT_SYMBOL(zfs_lookup
);
4923 EXPORT_SYMBOL(zfs_create
);
4924 EXPORT_SYMBOL(zfs_tmpfile
);
4925 EXPORT_SYMBOL(zfs_remove
);
4926 EXPORT_SYMBOL(zfs_mkdir
);
4927 EXPORT_SYMBOL(zfs_rmdir
);
4928 EXPORT_SYMBOL(zfs_readdir
);
4929 EXPORT_SYMBOL(zfs_fsync
);
4930 EXPORT_SYMBOL(zfs_getattr
);
4931 EXPORT_SYMBOL(zfs_getattr_fast
);
4932 EXPORT_SYMBOL(zfs_setattr
);
4933 EXPORT_SYMBOL(zfs_rename
);
4934 EXPORT_SYMBOL(zfs_symlink
);
4935 EXPORT_SYMBOL(zfs_readlink
);
4936 EXPORT_SYMBOL(zfs_link
);
4937 EXPORT_SYMBOL(zfs_inactive
);
4938 EXPORT_SYMBOL(zfs_space
);
4939 EXPORT_SYMBOL(zfs_fid
);
4940 EXPORT_SYMBOL(zfs_getsecattr
);
4941 EXPORT_SYMBOL(zfs_setsecattr
);
4942 EXPORT_SYMBOL(zfs_getpage
);
4943 EXPORT_SYMBOL(zfs_putpage
);
4944 EXPORT_SYMBOL(zfs_dirty_inode
);
4945 EXPORT_SYMBOL(zfs_map
);
4948 module_param(zfs_delete_blocks
, ulong
, 0644);
4949 MODULE_PARM_DESC(zfs_delete_blocks
, "Delete files larger than N blocks async");
4950 module_param(zfs_read_chunk_size
, long, 0644);
4951 MODULE_PARM_DESC(zfs_read_chunk_size
, "Bytes to read per chunk");