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_WAITED rather than 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_WAITED : 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
));
403 error
= uiomove(pb
+ off
, bytes
, UIO_READ
, uio
);
406 if (mapping_writably_mapped(mp
))
407 flush_dcache_page(pp
);
409 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
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1432 zfs_dirent_unlock(dl
);
1433 if (error
== ERESTART
) {
1439 zfs_acl_ids_free(&acl_ids
);
1444 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1446 error
= zfs_link_create(dl
, zp
, tx
, ZNEW
);
1449 * Since, we failed to add the directory entry for it,
1450 * delete the newly created dnode.
1452 zfs_znode_delete(zp
, tx
);
1453 remove_inode_hash(ZTOI(zp
));
1454 zfs_acl_ids_free(&acl_ids
);
1460 zfs_fuid_sync(zfsvfs
, tx
);
1462 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1463 if (flag
& FIGNORECASE
)
1465 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1466 vsecp
, acl_ids
.z_fuidp
, vap
);
1467 zfs_acl_ids_free(&acl_ids
);
1470 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1473 zfs_acl_ids_free(&acl_ids
);
1477 * A directory entry already exists for this name.
1480 * Can't truncate an existing file if in exclusive mode.
1483 error
= SET_ERROR(EEXIST
);
1487 * Can't open a directory for writing.
1489 if (S_ISDIR(ZTOI(zp
)->i_mode
)) {
1490 error
= SET_ERROR(EISDIR
);
1494 * Verify requested access to file.
1496 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1500 mutex_enter(&dzp
->z_lock
);
1502 mutex_exit(&dzp
->z_lock
);
1505 * Truncate regular files if requested.
1507 if (S_ISREG(ZTOI(zp
)->i_mode
) &&
1508 (vap
->va_mask
& ATTR_SIZE
) && (vap
->va_size
== 0)) {
1509 /* we can't hold any locks when calling zfs_freesp() */
1511 zfs_dirent_unlock(dl
);
1514 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1520 zfs_dirent_unlock(dl
);
1526 zfs_inode_update(dzp
);
1527 zfs_inode_update(zp
);
1531 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1532 zil_commit(zilog
, 0);
1540 zfs_tmpfile(struct inode
*dip
, vattr_t
*vap
, int excl
,
1541 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1543 znode_t
*zp
= NULL
, *dzp
= ITOZ(dip
);
1544 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1550 zfs_acl_ids_t acl_ids
;
1551 boolean_t fuid_dirtied
;
1552 boolean_t have_acl
= B_FALSE
;
1553 boolean_t waited
= B_FALSE
;
1556 * If we have an ephemeral id, ACL, or XVATTR then
1557 * make sure file system is at proper version
1563 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1564 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1565 return (SET_ERROR(EINVAL
));
1571 if (vap
->va_mask
& ATTR_XVATTR
) {
1572 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1573 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1583 * Create a new file object and update the directory
1586 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1588 zfs_acl_ids_free(&acl_ids
);
1592 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1593 cr
, vsecp
, &acl_ids
)) != 0)
1597 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1598 zfs_acl_ids_free(&acl_ids
);
1599 error
= SET_ERROR(EDQUOT
);
1603 tx
= dmu_tx_create(os
);
1605 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1606 ZFS_SA_BASE_ATTR_SIZE
);
1607 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1609 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1611 zfs_fuid_txhold(zfsvfs
, tx
);
1612 if (!zfsvfs
->z_use_sa
&&
1613 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1614 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1615 0, acl_ids
.z_aclp
->z_acl_bytes
);
1617 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1619 if (error
== ERESTART
) {
1625 zfs_acl_ids_free(&acl_ids
);
1630 zfs_mknode(dzp
, vap
, tx
, cr
, IS_TMPFILE
, &zp
, &acl_ids
);
1633 zfs_fuid_sync(zfsvfs
, tx
);
1635 /* Add to unlinked set */
1637 zfs_unlinked_add(zp
, tx
);
1638 zfs_acl_ids_free(&acl_ids
);
1646 zfs_inode_update(dzp
);
1647 zfs_inode_update(zp
);
1656 * Remove an entry from a directory.
1658 * IN: dip - inode of directory to remove entry from.
1659 * name - name of entry to remove.
1660 * cr - credentials of caller.
1662 * RETURN: 0 if success
1663 * error code if failure
1667 * ip - ctime (if nlink > 0)
1670 uint64_t null_xattr
= 0;
1674 zfs_remove(struct inode
*dip
, char *name
, cred_t
*cr
, int flags
)
1676 znode_t
*zp
, *dzp
= ITOZ(dip
);
1679 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1681 uint64_t acl_obj
, xattr_obj
;
1682 uint64_t xattr_obj_unlinked
= 0;
1687 boolean_t may_delete_now
, delete_now
= FALSE
;
1688 boolean_t unlinked
, toobig
= FALSE
;
1690 pathname_t
*realnmp
= NULL
;
1694 boolean_t waited
= B_FALSE
;
1697 return (SET_ERROR(EINVAL
));
1701 zilog
= zfsvfs
->z_log
;
1703 if (flags
& FIGNORECASE
) {
1713 * Attempt to lock directory; fail if entry doesn't exist.
1715 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1725 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1730 * Need to use rmdir for removing directories.
1732 if (S_ISDIR(ip
->i_mode
)) {
1733 error
= SET_ERROR(EPERM
);
1739 dnlc_remove(dvp
, realnmp
->pn_buf
);
1741 dnlc_remove(dvp
, name
);
1742 #endif /* HAVE_DNLC */
1744 mutex_enter(&zp
->z_lock
);
1745 may_delete_now
= atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
);
1746 mutex_exit(&zp
->z_lock
);
1749 * We may delete the znode now, or we may put it in the unlinked set;
1750 * it depends on whether we're the last link, and on whether there are
1751 * other holds on the inode. So we dmu_tx_hold() the right things to
1752 * allow for either case.
1755 tx
= dmu_tx_create(zfsvfs
->z_os
);
1756 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1757 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1758 zfs_sa_upgrade_txholds(tx
, zp
);
1759 zfs_sa_upgrade_txholds(tx
, dzp
);
1760 if (may_delete_now
) {
1761 toobig
= zp
->z_size
> zp
->z_blksz
* zfs_delete_blocks
;
1762 /* if the file is too big, only hold_free a token amount */
1763 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1764 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1767 /* are there any extended attributes? */
1768 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1769 &xattr_obj
, sizeof (xattr_obj
));
1770 if (error
== 0 && xattr_obj
) {
1771 error
= zfs_zget(zfsvfs
, xattr_obj
, &xzp
);
1773 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1774 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1777 mutex_enter(&zp
->z_lock
);
1778 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1779 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1780 mutex_exit(&zp
->z_lock
);
1782 /* charge as an update -- would be nice not to charge at all */
1783 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1786 * Mark this transaction as typically resulting in a net free of space
1788 dmu_tx_mark_netfree(tx
);
1790 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1792 zfs_dirent_unlock(dl
);
1793 if (error
== ERESTART
) {
1813 * Remove the directory entry.
1815 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1824 * Hold z_lock so that we can make sure that the ACL obj
1825 * hasn't changed. Could have been deleted due to
1828 mutex_enter(&zp
->z_lock
);
1829 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1830 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1831 delete_now
= may_delete_now
&& !toobig
&&
1832 atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
) &&
1833 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1838 if (xattr_obj_unlinked
) {
1839 ASSERT3U(ZTOI(xzp
)->i_nlink
, ==, 2);
1840 mutex_enter(&xzp
->z_lock
);
1841 xzp
->z_unlinked
= 1;
1842 clear_nlink(ZTOI(xzp
));
1844 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zfsvfs
),
1845 &links
, sizeof (links
), tx
);
1846 ASSERT3U(error
, ==, 0);
1847 mutex_exit(&xzp
->z_lock
);
1848 zfs_unlinked_add(xzp
, tx
);
1851 error
= sa_remove(zp
->z_sa_hdl
,
1852 SA_ZPL_XATTR(zfsvfs
), tx
);
1854 error
= sa_update(zp
->z_sa_hdl
,
1855 SA_ZPL_XATTR(zfsvfs
), &null_xattr
,
1856 sizeof (uint64_t), tx
);
1860 * Add to the unlinked set because a new reference could be
1861 * taken concurrently resulting in a deferred destruction.
1863 zfs_unlinked_add(zp
, tx
);
1864 mutex_exit(&zp
->z_lock
);
1865 } else if (unlinked
) {
1866 mutex_exit(&zp
->z_lock
);
1867 zfs_unlinked_add(zp
, tx
);
1871 if (flags
& FIGNORECASE
)
1873 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1880 zfs_dirent_unlock(dl
);
1881 zfs_inode_update(dzp
);
1882 zfs_inode_update(zp
);
1890 zfs_inode_update(xzp
);
1891 zfs_iput_async(ZTOI(xzp
));
1894 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1895 zil_commit(zilog
, 0);
1902 * Create a new directory and insert it into dip using the name
1903 * provided. Return a pointer to the inserted directory.
1905 * IN: dip - inode of directory to add subdir to.
1906 * dirname - name of new directory.
1907 * vap - attributes of new directory.
1908 * cr - credentials of caller.
1909 * vsecp - ACL to be set
1911 * OUT: ipp - inode of created directory.
1913 * RETURN: 0 if success
1914 * error code if failure
1917 * dip - ctime|mtime updated
1918 * ipp - ctime|mtime|atime updated
1922 zfs_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
, struct inode
**ipp
,
1923 cred_t
*cr
, int flags
, vsecattr_t
*vsecp
)
1925 znode_t
*zp
, *dzp
= ITOZ(dip
);
1926 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1934 gid_t gid
= crgetgid(cr
);
1935 zfs_acl_ids_t acl_ids
;
1936 boolean_t fuid_dirtied
;
1937 boolean_t waited
= B_FALSE
;
1939 ASSERT(S_ISDIR(vap
->va_mode
));
1942 * If we have an ephemeral id, ACL, or XVATTR then
1943 * make sure file system is at proper version
1947 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1948 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1949 return (SET_ERROR(EINVAL
));
1951 if (dirname
== NULL
)
1952 return (SET_ERROR(EINVAL
));
1956 zilog
= zfsvfs
->z_log
;
1958 if (dzp
->z_pflags
& ZFS_XATTR
) {
1960 return (SET_ERROR(EINVAL
));
1963 if (zfsvfs
->z_utf8
&& u8_validate(dirname
,
1964 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1966 return (SET_ERROR(EILSEQ
));
1968 if (flags
& FIGNORECASE
)
1971 if (vap
->va_mask
& ATTR_XVATTR
) {
1972 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1973 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1979 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1980 vsecp
, &acl_ids
)) != 0) {
1985 * First make sure the new directory doesn't exist.
1987 * Existence is checked first to make sure we don't return
1988 * EACCES instead of EEXIST which can cause some applications
1994 if ((error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1996 zfs_acl_ids_free(&acl_ids
);
2001 if ((error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
))) {
2002 zfs_acl_ids_free(&acl_ids
);
2003 zfs_dirent_unlock(dl
);
2008 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
2009 zfs_acl_ids_free(&acl_ids
);
2010 zfs_dirent_unlock(dl
);
2012 return (SET_ERROR(EDQUOT
));
2016 * Add a new entry to the directory.
2018 tx
= dmu_tx_create(zfsvfs
->z_os
);
2019 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
2020 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
2021 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
2023 zfs_fuid_txhold(zfsvfs
, tx
);
2024 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
2025 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
2026 acl_ids
.z_aclp
->z_acl_bytes
);
2029 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
2030 ZFS_SA_BASE_ATTR_SIZE
);
2032 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
2034 zfs_dirent_unlock(dl
);
2035 if (error
== ERESTART
) {
2041 zfs_acl_ids_free(&acl_ids
);
2050 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
2053 * Now put new name in parent dir.
2055 error
= zfs_link_create(dl
, zp
, tx
, ZNEW
);
2057 zfs_znode_delete(zp
, tx
);
2058 remove_inode_hash(ZTOI(zp
));
2063 zfs_fuid_sync(zfsvfs
, tx
);
2067 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
2068 if (flags
& FIGNORECASE
)
2070 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
2071 acl_ids
.z_fuidp
, vap
);
2074 zfs_acl_ids_free(&acl_ids
);
2078 zfs_dirent_unlock(dl
);
2080 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2081 zil_commit(zilog
, 0);
2086 zfs_inode_update(dzp
);
2087 zfs_inode_update(zp
);
2094 * Remove a directory subdir entry. If the current working
2095 * directory is the same as the subdir to be removed, the
2098 * IN: dip - inode of directory to remove from.
2099 * name - name of directory to be removed.
2100 * cwd - inode of current working directory.
2101 * cr - credentials of caller.
2102 * flags - case flags
2104 * RETURN: 0 on success, error code on failure.
2107 * dip - ctime|mtime updated
2111 zfs_rmdir(struct inode
*dip
, char *name
, struct inode
*cwd
, cred_t
*cr
,
2114 znode_t
*dzp
= ITOZ(dip
);
2117 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
2123 boolean_t waited
= B_FALSE
;
2126 return (SET_ERROR(EINVAL
));
2130 zilog
= zfsvfs
->z_log
;
2132 if (flags
& FIGNORECASE
)
2138 * Attempt to lock directory; fail if entry doesn't exist.
2140 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
2148 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
2152 if (!S_ISDIR(ip
->i_mode
)) {
2153 error
= SET_ERROR(ENOTDIR
);
2158 error
= SET_ERROR(EINVAL
);
2163 * Grab a lock on the directory to make sure that no one is
2164 * trying to add (or lookup) entries while we are removing it.
2166 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
2169 * Grab a lock on the parent pointer to make sure we play well
2170 * with the treewalk and directory rename code.
2172 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
2174 tx
= dmu_tx_create(zfsvfs
->z_os
);
2175 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2176 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2177 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
2178 zfs_sa_upgrade_txholds(tx
, zp
);
2179 zfs_sa_upgrade_txholds(tx
, dzp
);
2180 dmu_tx_mark_netfree(tx
);
2181 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
2183 rw_exit(&zp
->z_parent_lock
);
2184 rw_exit(&zp
->z_name_lock
);
2185 zfs_dirent_unlock(dl
);
2186 if (error
== ERESTART
) {
2199 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2202 uint64_t txtype
= TX_RMDIR
;
2203 if (flags
& FIGNORECASE
)
2205 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2210 rw_exit(&zp
->z_parent_lock
);
2211 rw_exit(&zp
->z_name_lock
);
2213 zfs_dirent_unlock(dl
);
2215 zfs_inode_update(dzp
);
2216 zfs_inode_update(zp
);
2219 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2220 zil_commit(zilog
, 0);
2227 * Read as many directory entries as will fit into the provided
2228 * dirent buffer from the given directory cursor position.
2230 * IN: ip - inode of directory to read.
2231 * dirent - buffer for directory entries.
2233 * OUT: dirent - filler buffer of directory entries.
2235 * RETURN: 0 if success
2236 * error code if failure
2239 * ip - atime updated
2241 * Note that the low 4 bits of the cookie returned by zap is always zero.
2242 * This allows us to use the low range for "special" directory entries:
2243 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2244 * we use the offset 2 for the '.zfs' directory.
2248 zfs_readdir(struct inode
*ip
, struct dir_context
*ctx
, cred_t
*cr
)
2250 znode_t
*zp
= ITOZ(ip
);
2251 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2254 zap_attribute_t zap
;
2260 uint64_t offset
; /* must be unsigned; checks for < 1 */
2265 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
2266 &parent
, sizeof (parent
))) != 0)
2270 * Quit if directory has been removed (posix)
2278 prefetch
= zp
->z_zn_prefetch
;
2281 * Initialize the iterator cursor.
2285 * Start iteration from the beginning of the directory.
2287 zap_cursor_init(&zc
, os
, zp
->z_id
);
2290 * The offset is a serialized cursor.
2292 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2296 * Transform to file-system independent format
2301 * Special case `.', `..', and `.zfs'.
2304 (void) strcpy(zap
.za_name
, ".");
2305 zap
.za_normalization_conflict
= 0;
2308 } else if (offset
== 1) {
2309 (void) strcpy(zap
.za_name
, "..");
2310 zap
.za_normalization_conflict
= 0;
2313 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2314 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2315 zap
.za_normalization_conflict
= 0;
2316 objnum
= ZFSCTL_INO_ROOT
;
2322 if ((error
= zap_cursor_retrieve(&zc
, &zap
))) {
2323 if (error
== ENOENT
)
2330 * Allow multiple entries provided the first entry is
2331 * the object id. Non-zpl consumers may safely make
2332 * use of the additional space.
2334 * XXX: This should be a feature flag for compatibility
2336 if (zap
.za_integer_length
!= 8 ||
2337 zap
.za_num_integers
== 0) {
2338 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2339 "entry, obj = %lld, offset = %lld, "
2340 "length = %d, num = %lld\n",
2341 (u_longlong_t
)zp
->z_id
,
2342 (u_longlong_t
)offset
,
2343 zap
.za_integer_length
,
2344 (u_longlong_t
)zap
.za_num_integers
);
2345 error
= SET_ERROR(ENXIO
);
2349 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2350 type
= ZFS_DIRENT_TYPE(zap
.za_first_integer
);
2353 done
= !dir_emit(ctx
, zap
.za_name
, strlen(zap
.za_name
),
2358 /* Prefetch znode */
2360 dmu_prefetch(os
, objnum
, 0, 0, 0,
2361 ZIO_PRIORITY_SYNC_READ
);
2365 * Move to the next entry, fill in the previous offset.
2367 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2368 zap_cursor_advance(&zc
);
2369 offset
= zap_cursor_serialize(&zc
);
2375 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2378 zap_cursor_fini(&zc
);
2379 if (error
== ENOENT
)
2387 ulong_t zfs_fsync_sync_cnt
= 4;
2390 zfs_fsync(struct inode
*ip
, int syncflag
, cred_t
*cr
)
2392 znode_t
*zp
= ITOZ(ip
);
2393 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2395 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2397 if (zfsvfs
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2400 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
2403 tsd_set(zfs_fsyncer_key
, NULL
);
2410 * Get the requested file attributes and place them in the provided
2413 * IN: ip - inode of file.
2414 * vap - va_mask identifies requested attributes.
2415 * If ATTR_XVATTR set, then optional attrs are requested
2416 * flags - ATTR_NOACLCHECK (CIFS server context)
2417 * cr - credentials of caller.
2419 * OUT: vap - attribute values.
2421 * RETURN: 0 (always succeeds)
2425 zfs_getattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2427 znode_t
*zp
= ITOZ(ip
);
2428 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2431 uint64_t atime
[2], mtime
[2], ctime
[2];
2432 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2433 xoptattr_t
*xoap
= NULL
;
2434 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2435 sa_bulk_attr_t bulk
[3];
2441 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2443 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
, &atime
, 16);
2444 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
2445 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
2447 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2453 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2454 * Also, if we are the owner don't bother, since owner should
2455 * always be allowed to read basic attributes of file.
2457 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2458 (vap
->va_uid
!= crgetuid(cr
))) {
2459 if ((error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2467 * Return all attributes. It's cheaper to provide the answer
2468 * than to determine whether we were asked the question.
2471 mutex_enter(&zp
->z_lock
);
2472 vap
->va_type
= vn_mode_to_vtype(zp
->z_mode
);
2473 vap
->va_mode
= zp
->z_mode
;
2474 vap
->va_fsid
= ZTOI(zp
)->i_sb
->s_dev
;
2475 vap
->va_nodeid
= zp
->z_id
;
2476 if ((zp
->z_id
== zfsvfs
->z_root
) && zfs_show_ctldir(zp
))
2477 links
= ZTOI(zp
)->i_nlink
+ 1;
2479 links
= ZTOI(zp
)->i_nlink
;
2480 vap
->va_nlink
= MIN(links
, ZFS_LINK_MAX
);
2481 vap
->va_size
= i_size_read(ip
);
2482 vap
->va_rdev
= ip
->i_rdev
;
2483 vap
->va_seq
= ip
->i_generation
;
2486 * Add in any requested optional attributes and the create time.
2487 * Also set the corresponding bits in the returned attribute bitmap.
2489 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zfsvfs
->z_use_fuids
) {
2490 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2492 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2493 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2496 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2497 xoap
->xoa_readonly
=
2498 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2499 XVA_SET_RTN(xvap
, XAT_READONLY
);
2502 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2504 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2505 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2508 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2510 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2511 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2514 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2515 xoap
->xoa_nounlink
=
2516 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2517 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2520 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2521 xoap
->xoa_immutable
=
2522 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2523 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2526 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2527 xoap
->xoa_appendonly
=
2528 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2529 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2532 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2534 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2535 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2538 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2540 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2541 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2544 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2545 xoap
->xoa_av_quarantined
=
2546 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2547 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2550 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2551 xoap
->xoa_av_modified
=
2552 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2553 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2556 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2557 S_ISREG(ip
->i_mode
)) {
2558 zfs_sa_get_scanstamp(zp
, xvap
);
2561 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2564 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zfsvfs
),
2565 times
, sizeof (times
));
2566 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2567 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2570 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2571 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2572 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2574 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2575 xoap
->xoa_generation
= ip
->i_generation
;
2576 XVA_SET_RTN(xvap
, XAT_GEN
);
2579 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2581 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2582 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2585 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2587 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2588 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2592 ZFS_TIME_DECODE(&vap
->va_atime
, atime
);
2593 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2594 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2596 mutex_exit(&zp
->z_lock
);
2598 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2600 if (zp
->z_blksz
== 0) {
2602 * Block size hasn't been set; suggest maximal I/O transfers.
2604 vap
->va_blksize
= zfsvfs
->z_max_blksz
;
2612 * Get the basic file attributes and place them in the provided kstat
2613 * structure. The inode is assumed to be the authoritative source
2614 * for most of the attributes. However, the znode currently has the
2615 * authoritative atime, blksize, and block count.
2617 * IN: ip - inode of file.
2619 * OUT: sp - kstat values.
2621 * RETURN: 0 (always succeeds)
2625 zfs_getattr_fast(struct inode
*ip
, struct kstat
*sp
)
2627 znode_t
*zp
= ITOZ(ip
);
2628 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2630 u_longlong_t nblocks
;
2635 mutex_enter(&zp
->z_lock
);
2637 generic_fillattr(ip
, sp
);
2639 sa_object_size(zp
->z_sa_hdl
, &blksize
, &nblocks
);
2640 sp
->blksize
= blksize
;
2641 sp
->blocks
= nblocks
;
2643 if (unlikely(zp
->z_blksz
== 0)) {
2645 * Block size hasn't been set; suggest maximal I/O transfers.
2647 sp
->blksize
= zfsvfs
->z_max_blksz
;
2650 mutex_exit(&zp
->z_lock
);
2653 * Required to prevent NFS client from detecting different inode
2654 * numbers of snapshot root dentry before and after snapshot mount.
2656 if (zfsvfs
->z_issnap
) {
2657 if (ip
->i_sb
->s_root
->d_inode
== ip
)
2658 sp
->ino
= ZFSCTL_INO_SNAPDIRS
-
2659 dmu_objset_id(zfsvfs
->z_os
);
2668 * Set the file attributes to the values contained in the
2671 * IN: ip - inode of file to be modified.
2672 * vap - new attribute values.
2673 * If ATTR_XVATTR set, then optional attrs are being set
2674 * flags - ATTR_UTIME set if non-default time values provided.
2675 * - ATTR_NOACLCHECK (CIFS context only).
2676 * cr - credentials of caller.
2678 * RETURN: 0 if success
2679 * error code if failure
2682 * ip - ctime updated, mtime updated if size changed.
2686 zfs_setattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2688 znode_t
*zp
= ITOZ(ip
);
2689 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2693 xvattr_t
*tmpxvattr
;
2694 uint_t mask
= vap
->va_mask
;
2695 uint_t saved_mask
= 0;
2698 uint64_t new_kuid
= 0, new_kgid
= 0, new_uid
, new_gid
;
2700 uint64_t mtime
[2], ctime
[2], atime
[2];
2702 int need_policy
= FALSE
;
2704 zfs_fuid_info_t
*fuidp
= NULL
;
2705 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2708 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2709 boolean_t fuid_dirtied
= B_FALSE
;
2710 sa_bulk_attr_t
*bulk
, *xattr_bulk
;
2711 int count
= 0, xattr_count
= 0;
2719 zilog
= zfsvfs
->z_log
;
2722 * Make sure that if we have ephemeral uid/gid or xvattr specified
2723 * that file system is at proper version level
2726 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
2727 (((mask
& ATTR_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2728 ((mask
& ATTR_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2729 (mask
& ATTR_XVATTR
))) {
2731 return (SET_ERROR(EINVAL
));
2734 if (mask
& ATTR_SIZE
&& S_ISDIR(ip
->i_mode
)) {
2736 return (SET_ERROR(EISDIR
));
2739 if (mask
& ATTR_SIZE
&& !S_ISREG(ip
->i_mode
) && !S_ISFIFO(ip
->i_mode
)) {
2741 return (SET_ERROR(EINVAL
));
2745 * If this is an xvattr_t, then get a pointer to the structure of
2746 * optional attributes. If this is NULL, then we have a vattr_t.
2748 xoap
= xva_getxoptattr(xvap
);
2750 tmpxvattr
= kmem_alloc(sizeof (xvattr_t
), KM_SLEEP
);
2751 xva_init(tmpxvattr
);
2753 bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2754 xattr_bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2757 * Immutable files can only alter immutable bit and atime
2759 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2760 ((mask
& (ATTR_SIZE
|ATTR_UID
|ATTR_GID
|ATTR_MTIME
|ATTR_MODE
)) ||
2761 ((mask
& ATTR_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2766 if ((mask
& ATTR_SIZE
) && (zp
->z_pflags
& ZFS_READONLY
)) {
2772 * Verify timestamps doesn't overflow 32 bits.
2773 * ZFS can handle large timestamps, but 32bit syscalls can't
2774 * handle times greater than 2039. This check should be removed
2775 * once large timestamps are fully supported.
2777 if (mask
& (ATTR_ATIME
| ATTR_MTIME
)) {
2778 if (((mask
& ATTR_ATIME
) &&
2779 TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2780 ((mask
& ATTR_MTIME
) &&
2781 TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2791 /* Can this be moved to before the top label? */
2792 if (zfs_is_readonly(zfsvfs
)) {
2798 * First validate permissions
2801 if (mask
& ATTR_SIZE
) {
2802 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2807 * XXX - Note, we are not providing any open
2808 * mode flags here (like FNDELAY), so we may
2809 * block if there are locks present... this
2810 * should be addressed in openat().
2812 /* XXX - would it be OK to generate a log record here? */
2813 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2818 if (mask
& (ATTR_ATIME
|ATTR_MTIME
) ||
2819 ((mask
& ATTR_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2820 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2821 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2822 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2823 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2824 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2825 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2826 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2830 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2831 int idmask
= (mask
& (ATTR_UID
|ATTR_GID
));
2836 * NOTE: even if a new mode is being set,
2837 * we may clear S_ISUID/S_ISGID bits.
2840 if (!(mask
& ATTR_MODE
))
2841 vap
->va_mode
= zp
->z_mode
;
2844 * Take ownership or chgrp to group we are a member of
2847 take_owner
= (mask
& ATTR_UID
) && (vap
->va_uid
== crgetuid(cr
));
2848 take_group
= (mask
& ATTR_GID
) &&
2849 zfs_groupmember(zfsvfs
, vap
->va_gid
, cr
);
2852 * If both ATTR_UID and ATTR_GID are set then take_owner and
2853 * take_group must both be set in order to allow taking
2856 * Otherwise, send the check through secpolicy_vnode_setattr()
2860 if (((idmask
== (ATTR_UID
|ATTR_GID
)) &&
2861 take_owner
&& take_group
) ||
2862 ((idmask
== ATTR_UID
) && take_owner
) ||
2863 ((idmask
== ATTR_GID
) && take_group
)) {
2864 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2865 skipaclchk
, cr
) == 0) {
2867 * Remove setuid/setgid for non-privileged users
2869 (void) secpolicy_setid_clear(vap
, cr
);
2870 trim_mask
= (mask
& (ATTR_UID
|ATTR_GID
));
2879 mutex_enter(&zp
->z_lock
);
2880 oldva
.va_mode
= zp
->z_mode
;
2881 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2882 if (mask
& ATTR_XVATTR
) {
2884 * Update xvattr mask to include only those attributes
2885 * that are actually changing.
2887 * the bits will be restored prior to actually setting
2888 * the attributes so the caller thinks they were set.
2890 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2891 if (xoap
->xoa_appendonly
!=
2892 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2895 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2896 XVA_SET_REQ(tmpxvattr
, XAT_APPENDONLY
);
2900 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2901 if (xoap
->xoa_nounlink
!=
2902 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2905 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2906 XVA_SET_REQ(tmpxvattr
, XAT_NOUNLINK
);
2910 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2911 if (xoap
->xoa_immutable
!=
2912 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2915 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2916 XVA_SET_REQ(tmpxvattr
, XAT_IMMUTABLE
);
2920 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2921 if (xoap
->xoa_nodump
!=
2922 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2925 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2926 XVA_SET_REQ(tmpxvattr
, XAT_NODUMP
);
2930 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2931 if (xoap
->xoa_av_modified
!=
2932 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2935 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2936 XVA_SET_REQ(tmpxvattr
, XAT_AV_MODIFIED
);
2940 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2941 if ((!S_ISREG(ip
->i_mode
) &&
2942 xoap
->xoa_av_quarantined
) ||
2943 xoap
->xoa_av_quarantined
!=
2944 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2947 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2948 XVA_SET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
);
2952 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2953 mutex_exit(&zp
->z_lock
);
2958 if (need_policy
== FALSE
&&
2959 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
2960 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
2965 mutex_exit(&zp
->z_lock
);
2967 if (mask
& ATTR_MODE
) {
2968 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
2969 err
= secpolicy_setid_setsticky_clear(ip
, vap
,
2974 trim_mask
|= ATTR_MODE
;
2982 * If trim_mask is set then take ownership
2983 * has been granted or write_acl is present and user
2984 * has the ability to modify mode. In that case remove
2985 * UID|GID and or MODE from mask so that
2986 * secpolicy_vnode_setattr() doesn't revoke it.
2990 saved_mask
= vap
->va_mask
;
2991 vap
->va_mask
&= ~trim_mask
;
2993 err
= secpolicy_vnode_setattr(cr
, ip
, vap
, &oldva
, flags
,
2994 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
2999 vap
->va_mask
|= saved_mask
;
3003 * secpolicy_vnode_setattr, or take ownership may have
3006 mask
= vap
->va_mask
;
3008 if ((mask
& (ATTR_UID
| ATTR_GID
))) {
3009 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
3010 &xattr_obj
, sizeof (xattr_obj
));
3012 if (err
== 0 && xattr_obj
) {
3013 err
= zfs_zget(ZTOZSB(zp
), xattr_obj
, &attrzp
);
3017 if (mask
& ATTR_UID
) {
3018 new_kuid
= zfs_fuid_create(zfsvfs
,
3019 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
3020 if (new_kuid
!= KUID_TO_SUID(ZTOI(zp
)->i_uid
) &&
3021 zfs_fuid_overquota(zfsvfs
, B_FALSE
, new_kuid
)) {
3029 if (mask
& ATTR_GID
) {
3030 new_kgid
= zfs_fuid_create(zfsvfs
,
3031 (uint64_t)vap
->va_gid
, cr
, ZFS_GROUP
, &fuidp
);
3032 if (new_kgid
!= KGID_TO_SGID(ZTOI(zp
)->i_gid
) &&
3033 zfs_fuid_overquota(zfsvfs
, B_TRUE
, new_kgid
)) {
3041 tx
= dmu_tx_create(zfsvfs
->z_os
);
3043 if (mask
& ATTR_MODE
) {
3044 uint64_t pmode
= zp
->z_mode
;
3046 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
3048 zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
);
3050 mutex_enter(&zp
->z_lock
);
3051 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
3053 * Are we upgrading ACL from old V0 format
3056 if (zfsvfs
->z_version
>= ZPL_VERSION_FUID
&&
3057 zfs_znode_acl_version(zp
) ==
3058 ZFS_ACL_VERSION_INITIAL
) {
3059 dmu_tx_hold_free(tx
, acl_obj
, 0,
3061 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3062 0, aclp
->z_acl_bytes
);
3064 dmu_tx_hold_write(tx
, acl_obj
, 0,
3067 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3068 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3069 0, aclp
->z_acl_bytes
);
3071 mutex_exit(&zp
->z_lock
);
3072 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3074 if ((mask
& ATTR_XVATTR
) &&
3075 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3076 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3078 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3082 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
3085 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3087 zfs_fuid_txhold(zfsvfs
, tx
);
3089 zfs_sa_upgrade_txholds(tx
, zp
);
3091 err
= dmu_tx_assign(tx
, TXG_WAIT
);
3097 * Set each attribute requested.
3098 * We group settings according to the locks they need to acquire.
3100 * Note: you cannot set ctime directly, although it will be
3101 * updated as a side-effect of calling this function.
3105 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3106 mutex_enter(&zp
->z_acl_lock
);
3107 mutex_enter(&zp
->z_lock
);
3109 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
3110 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
3113 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3114 mutex_enter(&attrzp
->z_acl_lock
);
3115 mutex_enter(&attrzp
->z_lock
);
3116 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3117 SA_ZPL_FLAGS(zfsvfs
), NULL
, &attrzp
->z_pflags
,
3118 sizeof (attrzp
->z_pflags
));
3121 if (mask
& (ATTR_UID
|ATTR_GID
)) {
3123 if (mask
& ATTR_UID
) {
3124 ZTOI(zp
)->i_uid
= SUID_TO_KUID(new_kuid
);
3125 new_uid
= zfs_uid_read(ZTOI(zp
));
3126 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zfsvfs
), NULL
,
3127 &new_uid
, sizeof (new_uid
));
3129 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3130 SA_ZPL_UID(zfsvfs
), NULL
, &new_uid
,
3132 ZTOI(attrzp
)->i_uid
= SUID_TO_KUID(new_uid
);
3136 if (mask
& ATTR_GID
) {
3137 ZTOI(zp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3138 new_gid
= zfs_gid_read(ZTOI(zp
));
3139 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zfsvfs
),
3140 NULL
, &new_gid
, sizeof (new_gid
));
3142 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3143 SA_ZPL_GID(zfsvfs
), NULL
, &new_gid
,
3145 ZTOI(attrzp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3148 if (!(mask
& ATTR_MODE
)) {
3149 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
),
3150 NULL
, &new_mode
, sizeof (new_mode
));
3151 new_mode
= zp
->z_mode
;
3153 err
= zfs_acl_chown_setattr(zp
);
3156 err
= zfs_acl_chown_setattr(attrzp
);
3161 if (mask
& ATTR_MODE
) {
3162 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
), NULL
,
3163 &new_mode
, sizeof (new_mode
));
3164 zp
->z_mode
= ZTOI(zp
)->i_mode
= new_mode
;
3165 ASSERT3P(aclp
, !=, NULL
);
3166 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
3168 if (zp
->z_acl_cached
)
3169 zfs_acl_free(zp
->z_acl_cached
);
3170 zp
->z_acl_cached
= aclp
;
3174 if ((mask
& ATTR_ATIME
) || zp
->z_atime_dirty
) {
3175 zp
->z_atime_dirty
= 0;
3176 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
3177 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
,
3178 &atime
, sizeof (atime
));
3181 if (mask
& (ATTR_MTIME
| ATTR_SIZE
)) {
3182 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3183 ZTOI(zp
)->i_mtime
= timespec_trunc(vap
->va_mtime
,
3184 ZTOI(zp
)->i_sb
->s_time_gran
);
3186 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
3187 mtime
, sizeof (mtime
));
3190 if (mask
& (ATTR_CTIME
| ATTR_SIZE
)) {
3191 ZFS_TIME_ENCODE(&vap
->va_ctime
, ctime
);
3192 ZTOI(zp
)->i_ctime
= timespec_trunc(vap
->va_ctime
,
3193 ZTOI(zp
)->i_sb
->s_time_gran
);
3194 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3195 ctime
, sizeof (ctime
));
3198 if (attrzp
&& mask
) {
3199 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3200 SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
,
3205 * Do this after setting timestamps to prevent timestamp
3206 * update from toggling bit
3209 if (xoap
&& (mask
& ATTR_XVATTR
)) {
3212 * restore trimmed off masks
3213 * so that return masks can be set for caller.
3216 if (XVA_ISSET_REQ(tmpxvattr
, XAT_APPENDONLY
)) {
3217 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3219 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NOUNLINK
)) {
3220 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3222 if (XVA_ISSET_REQ(tmpxvattr
, XAT_IMMUTABLE
)) {
3223 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3225 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NODUMP
)) {
3226 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3228 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_MODIFIED
)) {
3229 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3231 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
)) {
3232 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3235 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3236 ASSERT(S_ISREG(ip
->i_mode
));
3238 zfs_xvattr_set(zp
, xvap
, tx
);
3242 zfs_fuid_sync(zfsvfs
, tx
);
3245 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3247 mutex_exit(&zp
->z_lock
);
3248 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3249 mutex_exit(&zp
->z_acl_lock
);
3252 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3253 mutex_exit(&attrzp
->z_acl_lock
);
3254 mutex_exit(&attrzp
->z_lock
);
3257 if (err
== 0 && attrzp
) {
3258 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3267 zfs_fuid_info_free(fuidp
);
3275 if (err
== ERESTART
)
3278 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3282 zfs_inode_update(zp
);
3286 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3287 zil_commit(zilog
, 0);
3290 kmem_free(xattr_bulk
, sizeof (sa_bulk_attr_t
) * 7);
3291 kmem_free(bulk
, sizeof (sa_bulk_attr_t
) * 7);
3292 kmem_free(tmpxvattr
, sizeof (xvattr_t
));
3297 typedef struct zfs_zlock
{
3298 krwlock_t
*zl_rwlock
; /* lock we acquired */
3299 znode_t
*zl_znode
; /* znode we held */
3300 struct zfs_zlock
*zl_next
; /* next in list */
3304 * Drop locks and release vnodes that were held by zfs_rename_lock().
3307 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3311 while ((zl
= *zlpp
) != NULL
) {
3312 if (zl
->zl_znode
!= NULL
)
3313 zfs_iput_async(ZTOI(zl
->zl_znode
));
3314 rw_exit(zl
->zl_rwlock
);
3315 *zlpp
= zl
->zl_next
;
3316 kmem_free(zl
, sizeof (*zl
));
3321 * Search back through the directory tree, using the ".." entries.
3322 * Lock each directory in the chain to prevent concurrent renames.
3323 * Fail any attempt to move a directory into one of its own descendants.
3324 * XXX - z_parent_lock can overlap with map or grow locks
3327 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3331 uint64_t rootid
= ZTOZSB(zp
)->z_root
;
3332 uint64_t oidp
= zp
->z_id
;
3333 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3334 krw_t rw
= RW_WRITER
;
3337 * First pass write-locks szp and compares to zp->z_id.
3338 * Later passes read-lock zp and compare to zp->z_parent.
3341 if (!rw_tryenter(rwlp
, rw
)) {
3343 * Another thread is renaming in this path.
3344 * Note that if we are a WRITER, we don't have any
3345 * parent_locks held yet.
3347 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3349 * Drop our locks and restart
3351 zfs_rename_unlock(&zl
);
3355 rwlp
= &szp
->z_parent_lock
;
3360 * Wait for other thread to drop its locks
3366 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3367 zl
->zl_rwlock
= rwlp
;
3368 zl
->zl_znode
= NULL
;
3369 zl
->zl_next
= *zlpp
;
3372 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3373 return (SET_ERROR(EINVAL
));
3375 if (oidp
== rootid
) /* We've hit the top */
3378 if (rw
== RW_READER
) { /* i.e. not the first pass */
3379 int error
= zfs_zget(ZTOZSB(zp
), oidp
, &zp
);
3384 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(ZTOZSB(zp
)),
3385 &oidp
, sizeof (oidp
));
3386 rwlp
= &zp
->z_parent_lock
;
3389 } while (zp
->z_id
!= sdzp
->z_id
);
3395 * Move an entry from the provided source directory to the target
3396 * directory. Change the entry name as indicated.
3398 * IN: sdip - Source directory containing the "old entry".
3399 * snm - Old entry name.
3400 * tdip - Target directory to contain the "new entry".
3401 * tnm - New entry name.
3402 * cr - credentials of caller.
3403 * flags - case flags
3405 * RETURN: 0 on success, error code on failure.
3408 * sdip,tdip - ctime|mtime updated
3412 zfs_rename(struct inode
*sdip
, char *snm
, struct inode
*tdip
, char *tnm
,
3413 cred_t
*cr
, int flags
)
3415 znode_t
*tdzp
, *szp
, *tzp
;
3416 znode_t
*sdzp
= ITOZ(sdip
);
3417 zfsvfs_t
*zfsvfs
= ITOZSB(sdip
);
3419 zfs_dirlock_t
*sdl
, *tdl
;
3422 int cmp
, serr
, terr
;
3425 boolean_t waited
= B_FALSE
;
3427 if (snm
== NULL
|| tnm
== NULL
)
3428 return (SET_ERROR(EINVAL
));
3431 ZFS_VERIFY_ZP(sdzp
);
3432 zilog
= zfsvfs
->z_log
;
3435 ZFS_VERIFY_ZP(tdzp
);
3438 * We check i_sb because snapshots and the ctldir must have different
3441 if (tdip
->i_sb
!= sdip
->i_sb
|| zfsctl_is_node(tdip
)) {
3443 return (SET_ERROR(EXDEV
));
3446 if (zfsvfs
->z_utf8
&& u8_validate(tnm
,
3447 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3449 return (SET_ERROR(EILSEQ
));
3452 if (flags
& FIGNORECASE
)
3461 * This is to prevent the creation of links into attribute space
3462 * by renaming a linked file into/outof an attribute directory.
3463 * See the comment in zfs_link() for why this is considered bad.
3465 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3467 return (SET_ERROR(EINVAL
));
3471 * Lock source and target directory entries. To prevent deadlock,
3472 * a lock ordering must be defined. We lock the directory with
3473 * the smallest object id first, or if it's a tie, the one with
3474 * the lexically first name.
3476 if (sdzp
->z_id
< tdzp
->z_id
) {
3478 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3482 * First compare the two name arguments without
3483 * considering any case folding.
3485 int nofold
= (zfsvfs
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3487 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3488 ASSERT(error
== 0 || !zfsvfs
->z_utf8
);
3491 * POSIX: "If the old argument and the new argument
3492 * both refer to links to the same existing file,
3493 * the rename() function shall return successfully
3494 * and perform no other action."
3500 * If the file system is case-folding, then we may
3501 * have some more checking to do. A case-folding file
3502 * system is either supporting mixed case sensitivity
3503 * access or is completely case-insensitive. Note
3504 * that the file system is always case preserving.
3506 * In mixed sensitivity mode case sensitive behavior
3507 * is the default. FIGNORECASE must be used to
3508 * explicitly request case insensitive behavior.
3510 * If the source and target names provided differ only
3511 * by case (e.g., a request to rename 'tim' to 'Tim'),
3512 * we will treat this as a special case in the
3513 * case-insensitive mode: as long as the source name
3514 * is an exact match, we will allow this to proceed as
3515 * a name-change request.
3517 if ((zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
||
3518 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&&
3519 flags
& FIGNORECASE
)) &&
3520 u8_strcmp(snm
, tnm
, 0, zfsvfs
->z_norm
, U8_UNICODE_LATEST
,
3523 * case preserving rename request, require exact
3532 * If the source and destination directories are the same, we should
3533 * grab the z_name_lock of that directory only once.
3537 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3541 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3542 ZEXISTS
| zflg
, NULL
, NULL
);
3543 terr
= zfs_dirent_lock(&tdl
,
3544 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3546 terr
= zfs_dirent_lock(&tdl
,
3547 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3548 serr
= zfs_dirent_lock(&sdl
,
3549 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3555 * Source entry invalid or not there.
3558 zfs_dirent_unlock(tdl
);
3564 rw_exit(&sdzp
->z_name_lock
);
3566 if (strcmp(snm
, "..") == 0)
3572 zfs_dirent_unlock(sdl
);
3576 rw_exit(&sdzp
->z_name_lock
);
3578 if (strcmp(tnm
, "..") == 0)
3585 * Must have write access at the source to remove the old entry
3586 * and write access at the target to create the new entry.
3587 * Note that if target and source are the same, this can be
3588 * done in a single check.
3591 if ((error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
)))
3594 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3596 * Check to make sure rename is valid.
3597 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3599 if ((error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
)))
3604 * Does target exist?
3608 * Source and target must be the same type.
3610 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3611 if (!S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3612 error
= SET_ERROR(ENOTDIR
);
3616 if (S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3617 error
= SET_ERROR(EISDIR
);
3622 * POSIX dictates that when the source and target
3623 * entries refer to the same file object, rename
3624 * must do nothing and exit without error.
3626 if (szp
->z_id
== tzp
->z_id
) {
3632 tx
= dmu_tx_create(zfsvfs
->z_os
);
3633 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3634 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3635 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3636 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3638 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3639 zfs_sa_upgrade_txholds(tx
, tdzp
);
3642 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3643 zfs_sa_upgrade_txholds(tx
, tzp
);
3646 zfs_sa_upgrade_txholds(tx
, szp
);
3647 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
3648 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3651 zfs_rename_unlock(&zl
);
3652 zfs_dirent_unlock(sdl
);
3653 zfs_dirent_unlock(tdl
);
3656 rw_exit(&sdzp
->z_name_lock
);
3658 if (error
== ERESTART
) {
3675 if (tzp
) /* Attempt to remove the existing target */
3676 error
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3679 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3681 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3683 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zfsvfs
),
3684 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3687 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3689 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3690 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3691 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3694 * At this point, we have successfully created
3695 * the target name, but have failed to remove
3696 * the source name. Since the create was done
3697 * with the ZRENAMING flag, there are
3698 * complications; for one, the link count is
3699 * wrong. The easiest way to deal with this
3700 * is to remove the newly created target, and
3701 * return the original error. This must
3702 * succeed; fortunately, it is very unlikely to
3703 * fail, since we just created it.
3705 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3706 ZRENAMING
, NULL
), ==, 0);
3710 * If we had removed the existing target, subsequent
3711 * call to zfs_link_create() to add back the same entry
3712 * but, the new dnode (szp) should not fail.
3714 ASSERT(tzp
== NULL
);
3721 zfs_rename_unlock(&zl
);
3723 zfs_dirent_unlock(sdl
);
3724 zfs_dirent_unlock(tdl
);
3726 zfs_inode_update(sdzp
);
3728 rw_exit(&sdzp
->z_name_lock
);
3731 zfs_inode_update(tdzp
);
3733 zfs_inode_update(szp
);
3736 zfs_inode_update(tzp
);
3740 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3741 zil_commit(zilog
, 0);
3748 * Insert the indicated symbolic reference entry into the directory.
3750 * IN: dip - Directory to contain new symbolic link.
3751 * link - Name for new symlink entry.
3752 * vap - Attributes of new entry.
3753 * target - Target path of new symlink.
3755 * cr - credentials of caller.
3756 * flags - case flags
3758 * RETURN: 0 on success, error code on failure.
3761 * dip - ctime|mtime updated
3765 zfs_symlink(struct inode
*dip
, char *name
, vattr_t
*vap
, char *link
,
3766 struct inode
**ipp
, cred_t
*cr
, int flags
)
3768 znode_t
*zp
, *dzp
= ITOZ(dip
);
3771 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
3773 uint64_t len
= strlen(link
);
3776 zfs_acl_ids_t acl_ids
;
3777 boolean_t fuid_dirtied
;
3778 uint64_t txtype
= TX_SYMLINK
;
3779 boolean_t waited
= B_FALSE
;
3781 ASSERT(S_ISLNK(vap
->va_mode
));
3784 return (SET_ERROR(EINVAL
));
3788 zilog
= zfsvfs
->z_log
;
3790 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
3791 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3793 return (SET_ERROR(EILSEQ
));
3795 if (flags
& FIGNORECASE
)
3798 if (len
> MAXPATHLEN
) {
3800 return (SET_ERROR(ENAMETOOLONG
));
3803 if ((error
= zfs_acl_ids_create(dzp
, 0,
3804 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3812 * Attempt to lock directory; fail if entry already exists.
3814 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3816 zfs_acl_ids_free(&acl_ids
);
3821 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3822 zfs_acl_ids_free(&acl_ids
);
3823 zfs_dirent_unlock(dl
);
3828 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
3829 zfs_acl_ids_free(&acl_ids
);
3830 zfs_dirent_unlock(dl
);
3832 return (SET_ERROR(EDQUOT
));
3834 tx
= dmu_tx_create(zfsvfs
->z_os
);
3835 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3836 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3837 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3838 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3839 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3840 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3841 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3842 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3843 acl_ids
.z_aclp
->z_acl_bytes
);
3846 zfs_fuid_txhold(zfsvfs
, tx
);
3847 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3849 zfs_dirent_unlock(dl
);
3850 if (error
== ERESTART
) {
3856 zfs_acl_ids_free(&acl_ids
);
3863 * Create a new object for the symlink.
3864 * for version 4 ZPL datsets the symlink will be an SA attribute
3866 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3869 zfs_fuid_sync(zfsvfs
, tx
);
3871 mutex_enter(&zp
->z_lock
);
3873 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zfsvfs
),
3876 zfs_sa_symlink(zp
, link
, len
, tx
);
3877 mutex_exit(&zp
->z_lock
);
3880 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
3881 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3883 * Insert the new object into the directory.
3885 error
= zfs_link_create(dl
, zp
, tx
, ZNEW
);
3887 zfs_znode_delete(zp
, tx
);
3888 remove_inode_hash(ZTOI(zp
));
3890 if (flags
& FIGNORECASE
)
3892 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3894 zfs_inode_update(dzp
);
3895 zfs_inode_update(zp
);
3898 zfs_acl_ids_free(&acl_ids
);
3902 zfs_dirent_unlock(dl
);
3907 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3908 zil_commit(zilog
, 0);
3918 * Return, in the buffer contained in the provided uio structure,
3919 * the symbolic path referred to by ip.
3921 * IN: ip - inode of symbolic link
3922 * uio - structure to contain the link path.
3923 * cr - credentials of caller.
3925 * RETURN: 0 if success
3926 * error code if failure
3929 * ip - atime updated
3933 zfs_readlink(struct inode
*ip
, uio_t
*uio
, cred_t
*cr
)
3935 znode_t
*zp
= ITOZ(ip
);
3936 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
3942 mutex_enter(&zp
->z_lock
);
3944 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3945 SA_ZPL_SYMLINK(zfsvfs
), uio
);
3947 error
= zfs_sa_readlink(zp
, uio
);
3948 mutex_exit(&zp
->z_lock
);
3955 * Insert a new entry into directory tdip referencing sip.
3957 * IN: tdip - Directory to contain new entry.
3958 * sip - inode of new entry.
3959 * name - name of new entry.
3960 * cr - credentials of caller.
3962 * RETURN: 0 if success
3963 * error code if failure
3966 * tdip - ctime|mtime updated
3967 * sip - ctime updated
3971 zfs_link(struct inode
*tdip
, struct inode
*sip
, char *name
, cred_t
*cr
,
3974 znode_t
*dzp
= ITOZ(tdip
);
3976 zfsvfs_t
*zfsvfs
= ITOZSB(tdip
);
3984 boolean_t waited
= B_FALSE
;
3985 boolean_t is_tmpfile
= 0;
3988 is_tmpfile
= (sip
->i_nlink
== 0 && (sip
->i_state
& I_LINKABLE
));
3990 ASSERT(S_ISDIR(tdip
->i_mode
));
3993 return (SET_ERROR(EINVAL
));
3997 zilog
= zfsvfs
->z_log
;
4000 * POSIX dictates that we return EPERM here.
4001 * Better choices include ENOTSUP or EISDIR.
4003 if (S_ISDIR(sip
->i_mode
)) {
4005 return (SET_ERROR(EPERM
));
4012 * We check i_sb because snapshots and the ctldir must have different
4015 if (sip
->i_sb
!= tdip
->i_sb
|| zfsctl_is_node(sip
)) {
4017 return (SET_ERROR(EXDEV
));
4020 /* Prevent links to .zfs/shares files */
4022 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
4023 &parent
, sizeof (uint64_t))) != 0) {
4027 if (parent
== zfsvfs
->z_shares_dir
) {
4029 return (SET_ERROR(EPERM
));
4032 if (zfsvfs
->z_utf8
&& u8_validate(name
,
4033 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
4035 return (SET_ERROR(EILSEQ
));
4037 if (flags
& FIGNORECASE
)
4041 * We do not support links between attributes and non-attributes
4042 * because of the potential security risk of creating links
4043 * into "normal" file space in order to circumvent restrictions
4044 * imposed in attribute space.
4046 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
4048 return (SET_ERROR(EINVAL
));
4051 owner
= zfs_fuid_map_id(zfsvfs
, KUID_TO_SUID(sip
->i_uid
),
4053 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
4055 return (SET_ERROR(EPERM
));
4058 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
4065 * Attempt to lock directory; fail if entry already exists.
4067 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
4073 tx
= dmu_tx_create(zfsvfs
->z_os
);
4074 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
4075 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
4077 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
4079 zfs_sa_upgrade_txholds(tx
, szp
);
4080 zfs_sa_upgrade_txholds(tx
, dzp
);
4081 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
4083 zfs_dirent_unlock(dl
);
4084 if (error
== ERESTART
) {
4094 /* unmark z_unlinked so zfs_link_create will not reject */
4096 szp
->z_unlinked
= 0;
4097 error
= zfs_link_create(dl
, szp
, tx
, 0);
4100 uint64_t txtype
= TX_LINK
;
4102 * tmpfile is created to be in z_unlinkedobj, so remove it.
4103 * Also, we don't log in ZIL, be cause all previous file
4104 * operation on the tmpfile are ignored by ZIL. Instead we
4105 * always wait for txg to sync to make sure all previous
4106 * operation are sync safe.
4109 VERIFY(zap_remove_int(zfsvfs
->z_os
,
4110 zfsvfs
->z_unlinkedobj
, szp
->z_id
, tx
) == 0);
4112 if (flags
& FIGNORECASE
)
4114 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
4116 } else if (is_tmpfile
) {
4117 /* restore z_unlinked since when linking failed */
4118 szp
->z_unlinked
= 1;
4120 txg
= dmu_tx_get_txg(tx
);
4123 zfs_dirent_unlock(dl
);
4125 if (!is_tmpfile
&& zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4126 zil_commit(zilog
, 0);
4129 txg_wait_synced(dmu_objset_pool(zfsvfs
->z_os
), txg
);
4131 zfs_inode_update(dzp
);
4132 zfs_inode_update(szp
);
4138 zfs_putpage_commit_cb(void *arg
)
4140 struct page
*pp
= arg
;
4143 end_page_writeback(pp
);
4147 * Push a page out to disk, once the page is on stable storage the
4148 * registered commit callback will be run as notification of completion.
4150 * IN: ip - page mapped for inode.
4151 * pp - page to push (page is locked)
4152 * wbc - writeback control data
4154 * RETURN: 0 if success
4155 * error code if failure
4158 * ip - ctime|mtime updated
4162 zfs_putpage(struct inode
*ip
, struct page
*pp
, struct writeback_control
*wbc
)
4164 znode_t
*zp
= ITOZ(ip
);
4165 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4173 uint64_t mtime
[2], ctime
[2];
4174 sa_bulk_attr_t bulk
[3];
4176 struct address_space
*mapping
;
4181 ASSERT(PageLocked(pp
));
4183 pgoff
= page_offset(pp
); /* Page byte-offset in file */
4184 offset
= i_size_read(ip
); /* File length in bytes */
4185 pglen
= MIN(PAGE_SIZE
, /* Page length in bytes */
4186 P2ROUNDUP(offset
, PAGE_SIZE
)-pgoff
);
4188 /* Page is beyond end of file */
4189 if (pgoff
>= offset
) {
4195 /* Truncate page length to end of file */
4196 if (pgoff
+ pglen
> offset
)
4197 pglen
= offset
- pgoff
;
4201 * FIXME: Allow mmap writes past its quota. The correct fix
4202 * is to register a page_mkwrite() handler to count the page
4203 * against its quota when it is about to be dirtied.
4205 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
4206 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
4212 * The ordering here is critical and must adhere to the following
4213 * rules in order to avoid deadlocking in either zfs_read() or
4214 * zfs_free_range() due to a lock inversion.
4216 * 1) The page must be unlocked prior to acquiring the range lock.
4217 * This is critical because zfs_read() calls find_lock_page()
4218 * which may block on the page lock while holding the range lock.
4220 * 2) Before setting or clearing write back on a page the range lock
4221 * must be held in order to prevent a lock inversion with the
4222 * zfs_free_range() function.
4224 * This presents a problem because upon entering this function the
4225 * page lock is already held. To safely acquire the range lock the
4226 * page lock must be dropped. This creates a window where another
4227 * process could truncate, invalidate, dirty, or write out the page.
4229 * Therefore, after successfully reacquiring the range and page locks
4230 * the current page state is checked. In the common case everything
4231 * will be as is expected and it can be written out. However, if
4232 * the page state has changed it must be handled accordingly.
4234 mapping
= pp
->mapping
;
4235 redirty_page_for_writepage(wbc
, pp
);
4238 rl
= zfs_range_lock(&zp
->z_range_lock
, pgoff
, pglen
, RL_WRITER
);
4241 /* Page mapping changed or it was no longer dirty, we're done */
4242 if (unlikely((mapping
!= pp
->mapping
) || !PageDirty(pp
))) {
4244 zfs_range_unlock(rl
);
4249 /* Another process started write block if required */
4250 if (PageWriteback(pp
)) {
4252 zfs_range_unlock(rl
);
4254 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
4255 wait_on_page_writeback(pp
);
4261 /* Clear the dirty flag the required locks are held */
4262 if (!clear_page_dirty_for_io(pp
)) {
4264 zfs_range_unlock(rl
);
4270 * Counterpart for redirty_page_for_writepage() above. This page
4271 * was in fact not skipped and should not be counted as if it were.
4273 wbc
->pages_skipped
--;
4274 set_page_writeback(pp
);
4277 tx
= dmu_tx_create(zfsvfs
->z_os
);
4278 dmu_tx_hold_write(tx
, zp
->z_id
, pgoff
, pglen
);
4279 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4280 zfs_sa_upgrade_txholds(tx
, zp
);
4282 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4284 if (err
== ERESTART
)
4288 __set_page_dirty_nobuffers(pp
);
4290 end_page_writeback(pp
);
4291 zfs_range_unlock(rl
);
4297 ASSERT3U(pglen
, <=, PAGE_SIZE
);
4298 dmu_write(zfsvfs
->z_os
, zp
->z_id
, pgoff
, pglen
, va
, tx
);
4301 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
4302 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
4303 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
4306 /* Preserve the mtime and ctime provided by the inode */
4307 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4308 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4309 zp
->z_atime_dirty
= 0;
4312 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4314 zfs_log_write(zfsvfs
->z_log
, tx
, TX_WRITE
, zp
, pgoff
, pglen
, 0,
4315 zfs_putpage_commit_cb
, pp
);
4318 zfs_range_unlock(rl
);
4320 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4322 * Note that this is rarely called under writepages(), because
4323 * writepages() normally handles the entire commit for
4324 * performance reasons.
4326 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
4334 * Update the system attributes when the inode has been dirtied. For the
4335 * moment we only update the mode, atime, mtime, and ctime.
4338 zfs_dirty_inode(struct inode
*ip
, int flags
)
4340 znode_t
*zp
= ITOZ(ip
);
4341 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4343 uint64_t mode
, atime
[2], mtime
[2], ctime
[2];
4344 sa_bulk_attr_t bulk
[4];
4348 if (zfs_is_readonly(zfsvfs
) || dmu_objset_is_snapshot(zfsvfs
->z_os
))
4356 * This is the lazytime semantic indroduced in Linux 4.0
4357 * This flag will only be called from update_time when lazytime is set.
4358 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4359 * Fortunately mtime and ctime are managed within ZFS itself, so we
4360 * only need to dirty atime.
4362 if (flags
== I_DIRTY_TIME
) {
4363 zp
->z_atime_dirty
= 1;
4368 tx
= dmu_tx_create(zfsvfs
->z_os
);
4370 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4371 zfs_sa_upgrade_txholds(tx
, zp
);
4373 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4379 mutex_enter(&zp
->z_lock
);
4380 zp
->z_atime_dirty
= 0;
4382 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MODE(zfsvfs
), NULL
, &mode
, 8);
4383 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_ATIME(zfsvfs
), NULL
, &atime
, 16);
4384 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
4385 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
4387 /* Preserve the mode, mtime and ctime provided by the inode */
4388 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4389 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4390 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4395 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4396 mutex_exit(&zp
->z_lock
);
4406 zfs_inactive(struct inode
*ip
)
4408 znode_t
*zp
= ITOZ(ip
);
4409 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4412 int need_unlock
= 0;
4414 /* Only read lock if we haven't already write locked, e.g. rollback */
4415 if (!RW_WRITE_HELD(&zfsvfs
->z_teardown_inactive_lock
)) {
4417 rw_enter(&zfsvfs
->z_teardown_inactive_lock
, RW_READER
);
4419 if (zp
->z_sa_hdl
== NULL
) {
4421 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4425 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4426 dmu_tx_t
*tx
= dmu_tx_create(zfsvfs
->z_os
);
4428 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4429 zfs_sa_upgrade_txholds(tx
, zp
);
4430 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4434 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4435 mutex_enter(&zp
->z_lock
);
4436 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zfsvfs
),
4437 (void *)&atime
, sizeof (atime
), tx
);
4438 zp
->z_atime_dirty
= 0;
4439 mutex_exit(&zp
->z_lock
);
4446 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4450 * Bounds-check the seek operation.
4452 * IN: ip - inode seeking within
4453 * ooff - old file offset
4454 * noffp - pointer to new file offset
4455 * ct - caller context
4457 * RETURN: 0 if success
4458 * EINVAL if new offset invalid
4462 zfs_seek(struct inode
*ip
, offset_t ooff
, offset_t
*noffp
)
4464 if (S_ISDIR(ip
->i_mode
))
4466 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4470 * Fill pages with data from the disk.
4473 zfs_fillpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4475 znode_t
*zp
= ITOZ(ip
);
4476 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4478 struct page
*cur_pp
;
4479 u_offset_t io_off
, total
;
4486 io_len
= nr_pages
<< PAGE_SHIFT
;
4487 i_size
= i_size_read(ip
);
4488 io_off
= page_offset(pl
[0]);
4490 if (io_off
+ io_len
> i_size
)
4491 io_len
= i_size
- io_off
;
4494 * Iterate over list of pages and read each page individually.
4497 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4500 cur_pp
= pl
[page_idx
++];
4502 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4506 /* convert checksum errors into IO errors */
4508 err
= SET_ERROR(EIO
);
4517 * Uses zfs_fillpage to read data from the file and fill the pages.
4519 * IN: ip - inode of file to get data from.
4520 * pl - list of pages to read
4521 * nr_pages - number of pages to read
4523 * RETURN: 0 on success, error code on failure.
4526 * vp - atime updated
4530 zfs_getpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4532 znode_t
*zp
= ITOZ(ip
);
4533 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4542 err
= zfs_fillpage(ip
, pl
, nr_pages
);
4549 * Check ZFS specific permissions to memory map a section of a file.
4551 * IN: ip - inode of the file to mmap
4553 * addrp - start address in memory region
4554 * len - length of memory region
4555 * vm_flags- address flags
4557 * RETURN: 0 if success
4558 * error code if failure
4562 zfs_map(struct inode
*ip
, offset_t off
, caddr_t
*addrp
, size_t len
,
4563 unsigned long vm_flags
)
4565 znode_t
*zp
= ITOZ(ip
);
4566 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4571 if ((vm_flags
& VM_WRITE
) && (zp
->z_pflags
&
4572 (ZFS_IMMUTABLE
| ZFS_READONLY
| ZFS_APPENDONLY
))) {
4574 return (SET_ERROR(EPERM
));
4577 if ((vm_flags
& (VM_READ
| VM_EXEC
)) &&
4578 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4580 return (SET_ERROR(EACCES
));
4583 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4585 return (SET_ERROR(ENXIO
));
4593 * convoff - converts the given data (start, whence) to the
4597 convoff(struct inode
*ip
, flock64_t
*lckdat
, int whence
, offset_t offset
)
4602 if ((lckdat
->l_whence
== 2) || (whence
== 2)) {
4603 if ((error
= zfs_getattr(ip
, &vap
, 0, CRED())))
4607 switch (lckdat
->l_whence
) {
4609 lckdat
->l_start
+= offset
;
4612 lckdat
->l_start
+= vap
.va_size
;
4617 return (SET_ERROR(EINVAL
));
4620 if (lckdat
->l_start
< 0)
4621 return (SET_ERROR(EINVAL
));
4625 lckdat
->l_start
-= offset
;
4628 lckdat
->l_start
-= vap
.va_size
;
4633 return (SET_ERROR(EINVAL
));
4636 lckdat
->l_whence
= (short)whence
;
4641 * Free or allocate space in a file. Currently, this function only
4642 * supports the `F_FREESP' command. However, this command is somewhat
4643 * misnamed, as its functionality includes the ability to allocate as
4644 * well as free space.
4646 * IN: ip - inode of file to free data in.
4647 * cmd - action to take (only F_FREESP supported).
4648 * bfp - section of file to free/alloc.
4649 * flag - current file open mode flags.
4650 * offset - current file offset.
4651 * cr - credentials of caller [UNUSED].
4653 * RETURN: 0 on success, error code on failure.
4656 * ip - ctime|mtime updated
4660 zfs_space(struct inode
*ip
, int cmd
, flock64_t
*bfp
, int flag
,
4661 offset_t offset
, cred_t
*cr
)
4663 znode_t
*zp
= ITOZ(ip
);
4664 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4671 if (cmd
!= F_FREESP
) {
4673 return (SET_ERROR(EINVAL
));
4677 * Callers might not be able to detect properly that we are read-only,
4678 * so check it explicitly here.
4680 if (zfs_is_readonly(zfsvfs
)) {
4682 return (SET_ERROR(EROFS
));
4685 if ((error
= convoff(ip
, bfp
, 0, offset
))) {
4690 if (bfp
->l_len
< 0) {
4692 return (SET_ERROR(EINVAL
));
4696 * Permissions aren't checked on Solaris because on this OS
4697 * zfs_space() can only be called with an opened file handle.
4698 * On Linux we can get here through truncate_range() which
4699 * operates directly on inodes, so we need to check access rights.
4701 if ((error
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
))) {
4707 len
= bfp
->l_len
; /* 0 means from off to end of file */
4709 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4717 zfs_fid(struct inode
*ip
, fid_t
*fidp
)
4719 znode_t
*zp
= ITOZ(ip
);
4720 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4723 uint64_t object
= zp
->z_id
;
4730 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zfsvfs
),
4731 &gen64
, sizeof (uint64_t))) != 0) {
4736 gen
= (uint32_t)gen64
;
4738 size
= SHORT_FID_LEN
;
4740 zfid
= (zfid_short_t
*)fidp
;
4742 zfid
->zf_len
= size
;
4744 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4745 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4747 /* Must have a non-zero generation number to distinguish from .zfs */
4750 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4751 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4759 zfs_getsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4761 znode_t
*zp
= ITOZ(ip
);
4762 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4764 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4768 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
4776 zfs_setsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4778 znode_t
*zp
= ITOZ(ip
);
4779 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4781 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4782 zilog_t
*zilog
= zfsvfs
->z_log
;
4787 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
4789 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4790 zil_commit(zilog
, 0);
4796 #ifdef HAVE_UIO_ZEROCOPY
4798 * Tunable, both must be a power of 2.
4800 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4801 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4802 * an arcbuf for a partial block read
4804 int zcr_blksz_min
= (1 << 10); /* 1K */
4805 int zcr_blksz_max
= (1 << 17); /* 128K */
4809 zfs_reqzcbuf(struct inode
*ip
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
)
4811 znode_t
*zp
= ITOZ(ip
);
4812 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4813 int max_blksz
= zfsvfs
->z_max_blksz
;
4814 uio_t
*uio
= &xuio
->xu_uio
;
4815 ssize_t size
= uio
->uio_resid
;
4816 offset_t offset
= uio
->uio_loffset
;
4821 int preamble
, postamble
;
4823 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
4824 return (SET_ERROR(EINVAL
));
4831 * Loan out an arc_buf for write if write size is bigger than
4832 * max_blksz, and the file's block size is also max_blksz.
4835 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
4837 return (SET_ERROR(EINVAL
));
4840 * Caller requests buffers for write before knowing where the
4841 * write offset might be (e.g. NFS TCP write).
4846 preamble
= P2PHASE(offset
, blksz
);
4848 preamble
= blksz
- preamble
;
4853 postamble
= P2PHASE(size
, blksz
);
4856 fullblk
= size
/ blksz
;
4857 (void) dmu_xuio_init(xuio
,
4858 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
4861 * Have to fix iov base/len for partial buffers. They
4862 * currently represent full arc_buf's.
4865 /* data begins in the middle of the arc_buf */
4866 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4869 (void) dmu_xuio_add(xuio
, abuf
,
4870 blksz
- preamble
, preamble
);
4873 for (i
= 0; i
< fullblk
; i
++) {
4874 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4877 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
4881 /* data ends in the middle of the arc_buf */
4882 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4885 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
4890 * Loan out an arc_buf for read if the read size is larger than
4891 * the current file block size. Block alignment is not
4892 * considered. Partial arc_buf will be loaned out for read.
4894 blksz
= zp
->z_blksz
;
4895 if (blksz
< zcr_blksz_min
)
4896 blksz
= zcr_blksz_min
;
4897 if (blksz
> zcr_blksz_max
)
4898 blksz
= zcr_blksz_max
;
4899 /* avoid potential complexity of dealing with it */
4900 if (blksz
> max_blksz
) {
4902 return (SET_ERROR(EINVAL
));
4905 maxsize
= zp
->z_size
- uio
->uio_loffset
;
4911 return (SET_ERROR(EINVAL
));
4916 return (SET_ERROR(EINVAL
));
4919 uio
->uio_extflg
= UIO_XUIO
;
4920 XUIO_XUZC_RW(xuio
) = ioflag
;
4927 zfs_retzcbuf(struct inode
*ip
, xuio_t
*xuio
, cred_t
*cr
)
4931 int ioflag
= XUIO_XUZC_RW(xuio
);
4933 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
4935 i
= dmu_xuio_cnt(xuio
);
4937 abuf
= dmu_xuio_arcbuf(xuio
, i
);
4939 * if abuf == NULL, it must be a write buffer
4940 * that has been returned in zfs_write().
4943 dmu_return_arcbuf(abuf
);
4944 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
4947 dmu_xuio_fini(xuio
);
4950 #endif /* HAVE_UIO_ZEROCOPY */
4952 #if defined(_KERNEL) && defined(HAVE_SPL)
4953 EXPORT_SYMBOL(zfs_open
);
4954 EXPORT_SYMBOL(zfs_close
);
4955 EXPORT_SYMBOL(zfs_read
);
4956 EXPORT_SYMBOL(zfs_write
);
4957 EXPORT_SYMBOL(zfs_access
);
4958 EXPORT_SYMBOL(zfs_lookup
);
4959 EXPORT_SYMBOL(zfs_create
);
4960 EXPORT_SYMBOL(zfs_tmpfile
);
4961 EXPORT_SYMBOL(zfs_remove
);
4962 EXPORT_SYMBOL(zfs_mkdir
);
4963 EXPORT_SYMBOL(zfs_rmdir
);
4964 EXPORT_SYMBOL(zfs_readdir
);
4965 EXPORT_SYMBOL(zfs_fsync
);
4966 EXPORT_SYMBOL(zfs_getattr
);
4967 EXPORT_SYMBOL(zfs_getattr_fast
);
4968 EXPORT_SYMBOL(zfs_setattr
);
4969 EXPORT_SYMBOL(zfs_rename
);
4970 EXPORT_SYMBOL(zfs_symlink
);
4971 EXPORT_SYMBOL(zfs_readlink
);
4972 EXPORT_SYMBOL(zfs_link
);
4973 EXPORT_SYMBOL(zfs_inactive
);
4974 EXPORT_SYMBOL(zfs_space
);
4975 EXPORT_SYMBOL(zfs_fid
);
4976 EXPORT_SYMBOL(zfs_getsecattr
);
4977 EXPORT_SYMBOL(zfs_setsecattr
);
4978 EXPORT_SYMBOL(zfs_getpage
);
4979 EXPORT_SYMBOL(zfs_putpage
);
4980 EXPORT_SYMBOL(zfs_dirty_inode
);
4981 EXPORT_SYMBOL(zfs_map
);
4984 module_param(zfs_delete_blocks
, ulong
, 0644);
4985 MODULE_PARM_DESC(zfs_delete_blocks
, "Delete files larger than N blocks async");
4986 module_param(zfs_read_chunk_size
, long, 0644);
4987 MODULE_PARM_DESC(zfs_read_chunk_size
, "Bytes to read per chunk");