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(zsb).
95 * A ZFS_EXIT(zsb) 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 zsb->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(zsb); // 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(zsb); // 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(zsb); // 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 zfs_sb_t
*zsb
= 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
) && zsb
->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
);
229 EXPORT_SYMBOL(zfs_open
);
233 zfs_close(struct inode
*ip
, int flag
, cred_t
*cr
)
235 znode_t
*zp
= ITOZ(ip
);
236 zfs_sb_t
*zsb
= ITOZSB(ip
);
241 /* Decrement the synchronous opens in the znode */
243 atomic_dec_32(&zp
->z_sync_cnt
);
245 if (!zfs_has_ctldir(zp
) && zsb
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
246 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0)
247 VERIFY(zfs_vscan(ip
, cr
, 1) == 0);
252 EXPORT_SYMBOL(zfs_close
);
254 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
256 * Lseek support for finding holes (cmd == SEEK_HOLE) and
257 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
260 zfs_holey_common(struct inode
*ip
, int cmd
, loff_t
*off
)
262 znode_t
*zp
= ITOZ(ip
);
263 uint64_t noff
= (uint64_t)*off
; /* new offset */
268 file_sz
= zp
->z_size
;
269 if (noff
>= file_sz
) {
270 return (SET_ERROR(ENXIO
));
273 if (cmd
== SEEK_HOLE
)
278 error
= dmu_offset_next(ZTOZSB(zp
)->z_os
, zp
->z_id
, hole
, &noff
);
281 return (SET_ERROR(ENXIO
));
284 * We could find a hole that begins after the logical end-of-file,
285 * because dmu_offset_next() only works on whole blocks. If the
286 * EOF falls mid-block, then indicate that the "virtual hole"
287 * at the end of the file begins at the logical EOF, rather than
288 * at the end of the last block.
290 if (noff
> file_sz
) {
302 zfs_holey(struct inode
*ip
, int cmd
, loff_t
*off
)
304 znode_t
*zp
= ITOZ(ip
);
305 zfs_sb_t
*zsb
= ITOZSB(ip
);
311 error
= zfs_holey_common(ip
, cmd
, off
);
316 EXPORT_SYMBOL(zfs_holey
);
317 #endif /* SEEK_HOLE && SEEK_DATA */
321 * When a file is memory mapped, we must keep the IO data synchronized
322 * between the DMU cache and the memory mapped pages. What this means:
324 * On Write: If we find a memory mapped page, we write to *both*
325 * the page and the dmu buffer.
328 update_pages(struct inode
*ip
, int64_t start
, int len
,
329 objset_t
*os
, uint64_t oid
)
331 struct address_space
*mp
= ip
->i_mapping
;
337 off
= start
& (PAGE_SIZE
-1);
338 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
339 nbytes
= MIN(PAGE_SIZE
- off
, len
);
341 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
343 if (mapping_writably_mapped(mp
))
344 flush_dcache_page(pp
);
347 (void) dmu_read(os
, oid
, start
+off
, nbytes
, pb
+off
,
351 if (mapping_writably_mapped(mp
))
352 flush_dcache_page(pp
);
354 mark_page_accessed(pp
);
367 * When a file is memory mapped, we must keep the IO data synchronized
368 * between the DMU cache and the memory mapped pages. What this means:
370 * On Read: We "read" preferentially from memory mapped pages,
371 * else we default from the dmu buffer.
373 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
374 * the file is memory mapped.
377 mappedread(struct inode
*ip
, int nbytes
, uio_t
*uio
)
379 struct address_space
*mp
= ip
->i_mapping
;
381 znode_t
*zp
= ITOZ(ip
);
388 start
= uio
->uio_loffset
;
389 off
= start
& (PAGE_SIZE
-1);
390 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
391 bytes
= MIN(PAGE_SIZE
- off
, len
);
393 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
395 ASSERT(PageUptodate(pp
));
398 error
= uiomove(pb
+ off
, bytes
, UIO_READ
, uio
);
401 if (mapping_writably_mapped(mp
))
402 flush_dcache_page(pp
);
404 mark_page_accessed(pp
);
408 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
421 unsigned long zfs_read_chunk_size
= 1024 * 1024; /* Tunable */
422 unsigned long zfs_delete_blocks
= DMU_MAX_DELETEBLKCNT
;
425 * Read bytes from specified file into supplied buffer.
427 * IN: ip - inode of file to be read from.
428 * uio - structure supplying read location, range info,
430 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
431 * O_DIRECT flag; used to bypass page cache.
432 * cr - credentials of caller.
434 * OUT: uio - updated offset and range, buffer filled.
436 * RETURN: 0 on success, error code on failure.
439 * inode - atime updated if byte count > 0
443 zfs_read(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
445 znode_t
*zp
= ITOZ(ip
);
446 zfs_sb_t
*zsb
= ITOZSB(ip
);
450 #ifdef HAVE_UIO_ZEROCOPY
452 #endif /* HAVE_UIO_ZEROCOPY */
457 if (zp
->z_pflags
& ZFS_AV_QUARANTINED
) {
459 return (SET_ERROR(EACCES
));
463 * Validate file offset
465 if (uio
->uio_loffset
< (offset_t
)0) {
467 return (SET_ERROR(EINVAL
));
471 * Fasttrack empty reads
473 if (uio
->uio_resid
== 0) {
479 * If we're in FRSYNC mode, sync out this znode before reading it.
481 if (ioflag
& FRSYNC
|| zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
482 zil_commit(zsb
->z_log
, zp
->z_id
);
485 * Lock the range against changes.
487 rl
= zfs_range_lock(&zp
->z_range_lock
, uio
->uio_loffset
, uio
->uio_resid
,
491 * If we are reading past end-of-file we can skip
492 * to the end; but we might still need to set atime.
494 if (uio
->uio_loffset
>= zp
->z_size
) {
499 ASSERT(uio
->uio_loffset
< zp
->z_size
);
500 n
= MIN(uio
->uio_resid
, zp
->z_size
- uio
->uio_loffset
);
502 #ifdef HAVE_UIO_ZEROCOPY
503 if ((uio
->uio_extflg
== UIO_XUIO
) &&
504 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
)) {
506 int blksz
= zp
->z_blksz
;
507 uint64_t offset
= uio
->uio_loffset
;
509 xuio
= (xuio_t
*)uio
;
511 nblk
= (P2ROUNDUP(offset
+ n
, blksz
) - P2ALIGN(offset
,
514 ASSERT(offset
+ n
<= blksz
);
517 (void) dmu_xuio_init(xuio
, nblk
);
519 if (vn_has_cached_data(ip
)) {
521 * For simplicity, we always allocate a full buffer
522 * even if we only expect to read a portion of a block.
524 while (--nblk
>= 0) {
525 (void) dmu_xuio_add(xuio
,
526 dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
531 #endif /* HAVE_UIO_ZEROCOPY */
534 nbytes
= MIN(n
, zfs_read_chunk_size
-
535 P2PHASE(uio
->uio_loffset
, zfs_read_chunk_size
));
537 if (zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
)) {
538 error
= mappedread(ip
, nbytes
, uio
);
540 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
545 /* convert checksum errors into IO errors */
547 error
= SET_ERROR(EIO
);
554 zfs_range_unlock(rl
);
559 EXPORT_SYMBOL(zfs_read
);
562 * Write the bytes to a file.
564 * IN: ip - inode of file to be written to.
565 * uio - structure supplying write location, range info,
567 * ioflag - FAPPEND flag set if in append mode.
568 * O_DIRECT flag; used to bypass page cache.
569 * cr - credentials of caller.
571 * OUT: uio - updated offset and range.
573 * RETURN: 0 if success
574 * error code if failure
577 * ip - ctime|mtime updated if byte count > 0
582 zfs_write(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
584 znode_t
*zp
= ITOZ(ip
);
585 rlim64_t limit
= uio
->uio_limit
;
586 ssize_t start_resid
= uio
->uio_resid
;
590 zfs_sb_t
*zsb
= ZTOZSB(zp
);
595 int max_blksz
= zsb
->z_max_blksz
;
598 const iovec_t
*aiov
= NULL
;
602 sa_bulk_attr_t bulk
[4];
603 uint64_t mtime
[2], ctime
[2];
605 #ifdef HAVE_UIO_ZEROCOPY
607 const iovec_t
*iovp
= uio
->uio_iov
;
608 ASSERTV(int iovcnt
= uio
->uio_iovcnt
);
612 * Fasttrack empty write
618 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
624 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
625 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
626 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zsb
), NULL
, &zp
->z_size
, 8);
627 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
631 * Callers might not be able to detect properly that we are read-only,
632 * so check it explicitly here.
634 if (zfs_is_readonly(zsb
)) {
636 return (SET_ERROR(EROFS
));
640 * If immutable or not appending then return EPERM
642 if ((zp
->z_pflags
& (ZFS_IMMUTABLE
| ZFS_READONLY
)) ||
643 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
644 (uio
->uio_loffset
< zp
->z_size
))) {
646 return (SET_ERROR(EPERM
));
652 * Validate file offset
654 woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
657 return (SET_ERROR(EINVAL
));
661 * Pre-fault the pages to ensure slow (eg NFS) pages
663 * Skip this if uio contains loaned arc_buf.
665 #ifdef HAVE_UIO_ZEROCOPY
666 if ((uio
->uio_extflg
== UIO_XUIO
) &&
667 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
))
668 xuio
= (xuio_t
*)uio
;
671 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
674 * If in append mode, set the io offset pointer to eof.
676 if (ioflag
& FAPPEND
) {
678 * Obtain an appending range lock to guarantee file append
679 * semantics. We reset the write offset once we have the lock.
681 rl
= zfs_range_lock(&zp
->z_range_lock
, 0, n
, RL_APPEND
);
683 if (rl
->r_len
== UINT64_MAX
) {
685 * We overlocked the file because this write will cause
686 * the file block size to increase.
687 * Note that zp_size cannot change with this lock held.
691 uio
->uio_loffset
= woff
;
694 * Note that if the file block size will change as a result of
695 * this write, then this range lock will lock the entire file
696 * so that we can re-write the block safely.
698 rl
= zfs_range_lock(&zp
->z_range_lock
, woff
, n
, RL_WRITER
);
702 zfs_range_unlock(rl
);
704 return (SET_ERROR(EFBIG
));
707 if ((woff
+ n
) > limit
|| woff
> (limit
- n
))
710 /* Will this write extend the file length? */
711 write_eof
= (woff
+ n
> zp
->z_size
);
713 end_size
= MAX(zp
->z_size
, woff
+ n
);
716 * Write the file in reasonable size chunks. Each chunk is written
717 * in a separate transaction; this keeps the intent log records small
718 * and allows us to do more fine-grained space accounting.
722 woff
= uio
->uio_loffset
;
723 if (zfs_owner_overquota(zsb
, zp
, B_FALSE
) ||
724 zfs_owner_overquota(zsb
, zp
, B_TRUE
)) {
726 dmu_return_arcbuf(abuf
);
727 error
= SET_ERROR(EDQUOT
);
731 if (xuio
&& abuf
== NULL
) {
732 #ifdef HAVE_UIO_ZEROCOPY
733 ASSERT(i_iov
< iovcnt
);
734 ASSERT3U(uio
->uio_segflg
, !=, UIO_BVEC
);
736 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
737 dmu_xuio_clear(xuio
, i_iov
);
738 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
739 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
740 aiov
->iov_len
== arc_buf_size(abuf
)));
743 } else if (abuf
== NULL
&& n
>= max_blksz
&&
744 woff
>= zp
->z_size
&&
745 P2PHASE(woff
, max_blksz
) == 0 &&
746 zp
->z_blksz
== max_blksz
) {
748 * This write covers a full block. "Borrow" a buffer
749 * from the dmu so that we can fill it before we enter
750 * a transaction. This avoids the possibility of
751 * holding up the transaction if the data copy hangs
752 * up on a pagefault (e.g., from an NFS server mapping).
756 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
758 ASSERT(abuf
!= NULL
);
759 ASSERT(arc_buf_size(abuf
) == max_blksz
);
760 if ((error
= uiocopy(abuf
->b_data
, max_blksz
,
761 UIO_WRITE
, uio
, &cbytes
))) {
762 dmu_return_arcbuf(abuf
);
765 ASSERT(cbytes
== max_blksz
);
769 * Start a transaction.
771 tx
= dmu_tx_create(zsb
->z_os
);
772 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
773 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
774 zfs_sa_upgrade_txholds(tx
, zp
);
775 error
= dmu_tx_assign(tx
, TXG_WAIT
);
779 dmu_return_arcbuf(abuf
);
784 * If zfs_range_lock() over-locked we grow the blocksize
785 * and then reduce the lock range. This will only happen
786 * on the first iteration since zfs_range_reduce() will
787 * shrink down r_len to the appropriate size.
789 if (rl
->r_len
== UINT64_MAX
) {
792 if (zp
->z_blksz
> max_blksz
) {
794 * File's blocksize is already larger than the
795 * "recordsize" property. Only let it grow to
796 * the next power of 2.
798 ASSERT(!ISP2(zp
->z_blksz
));
799 new_blksz
= MIN(end_size
,
800 1 << highbit64(zp
->z_blksz
));
802 new_blksz
= MIN(end_size
, max_blksz
);
804 zfs_grow_blocksize(zp
, new_blksz
, tx
);
805 zfs_range_reduce(rl
, woff
, n
);
809 * XXX - should we really limit each write to z_max_blksz?
810 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
812 nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
815 tx_bytes
= uio
->uio_resid
;
816 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
818 tx_bytes
-= uio
->uio_resid
;
821 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
823 * If this is not a full block write, but we are
824 * extending the file past EOF and this data starts
825 * block-aligned, use assign_arcbuf(). Otherwise,
826 * write via dmu_write().
828 if (tx_bytes
< max_blksz
&& (!write_eof
||
829 aiov
->iov_base
!= abuf
->b_data
)) {
831 dmu_write(zsb
->z_os
, zp
->z_id
, woff
,
832 aiov
->iov_len
, aiov
->iov_base
, tx
);
833 dmu_return_arcbuf(abuf
);
834 xuio_stat_wbuf_copied();
836 ASSERT(xuio
|| tx_bytes
== max_blksz
);
837 dmu_assign_arcbuf(sa_get_db(zp
->z_sa_hdl
),
840 ASSERT(tx_bytes
<= uio
->uio_resid
);
841 uioskip(uio
, tx_bytes
);
844 if (tx_bytes
&& zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
))
845 update_pages(ip
, woff
, tx_bytes
, zsb
->z_os
, zp
->z_id
);
848 * If we made no progress, we're done. If we made even
849 * partial progress, update the znode and ZIL accordingly.
852 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
853 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
860 * Clear Set-UID/Set-GID bits on successful write if not
861 * privileged and at least one of the execute bits is set.
863 * It would be nice to to this after all writes have
864 * been done, but that would still expose the ISUID/ISGID
865 * to another app after the partial write is committed.
867 * Note: we don't call zfs_fuid_map_id() here because
868 * user 0 is not an ephemeral uid.
870 mutex_enter(&zp
->z_acl_lock
);
871 uid
= KUID_TO_SUID(ip
->i_uid
);
872 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
873 (S_IXUSR
>> 6))) != 0 &&
874 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
875 secpolicy_vnode_setid_retain(cr
,
876 ((zp
->z_mode
& S_ISUID
) != 0 && uid
== 0)) != 0) {
878 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
879 ip
->i_mode
= newmode
= zp
->z_mode
;
880 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zsb
),
881 (void *)&newmode
, sizeof (uint64_t), tx
);
883 mutex_exit(&zp
->z_acl_lock
);
885 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
);
888 * Update the file size (zp_size) if it has changed;
889 * account for possible concurrent updates.
891 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
892 (void) atomic_cas_64(&zp
->z_size
, end_size
,
897 * If we are replaying and eof is non zero then force
898 * the file size to the specified eof. Note, there's no
899 * concurrency during replay.
901 if (zsb
->z_replay
&& zsb
->z_replay_eof
!= 0)
902 zp
->z_size
= zsb
->z_replay_eof
;
904 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
906 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
,
912 ASSERT(tx_bytes
== nbytes
);
916 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
919 zfs_inode_update(zp
);
920 zfs_range_unlock(rl
);
923 * If we're in replay mode, or we made no progress, return error.
924 * Otherwise, it's at least a partial write, so it's successful.
926 if (zsb
->z_replay
|| uio
->uio_resid
== start_resid
) {
931 if (ioflag
& (FSYNC
| FDSYNC
) ||
932 zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
933 zil_commit(zilog
, zp
->z_id
);
938 EXPORT_SYMBOL(zfs_write
);
941 * Drop a reference on the passed inode asynchronously. This ensures
942 * that the caller will never drop the last reference on an inode in
943 * the current context. Doing so while holding open a tx could result
944 * in a deadlock if iput_final() re-enters the filesystem code.
947 zfs_iput_async(struct inode
*ip
)
949 objset_t
*os
= ITOZSB(ip
)->z_os
;
951 ASSERT(atomic_read(&ip
->i_count
) > 0);
954 if (atomic_read(&ip
->i_count
) == 1)
955 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os
)),
956 (task_func_t
*)iput
, ip
, TQ_SLEEP
) != TASKQID_INVALID
);
962 zfs_get_done(zgd_t
*zgd
, int error
)
964 znode_t
*zp
= zgd
->zgd_private
;
967 dmu_buf_rele(zgd
->zgd_db
, zgd
);
969 zfs_range_unlock(zgd
->zgd_rl
);
972 * Release the vnode asynchronously as we currently have the
973 * txg stopped from syncing.
975 zfs_iput_async(ZTOI(zp
));
977 if (error
== 0 && zgd
->zgd_bp
)
978 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
980 kmem_free(zgd
, sizeof (zgd_t
));
984 static int zil_fault_io
= 0;
988 * Get data to generate a TX_WRITE intent log record.
991 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
994 objset_t
*os
= zsb
->z_os
;
996 uint64_t object
= lr
->lr_foid
;
997 uint64_t offset
= lr
->lr_offset
;
998 uint64_t size
= lr
->lr_length
;
999 blkptr_t
*bp
= &lr
->lr_blkptr
;
1004 ASSERT(zio
!= NULL
);
1008 * Nothing to do if the file has been removed
1010 if (zfs_zget(zsb
, object
, &zp
) != 0)
1011 return (SET_ERROR(ENOENT
));
1012 if (zp
->z_unlinked
) {
1014 * Release the vnode asynchronously as we currently have the
1015 * txg stopped from syncing.
1017 zfs_iput_async(ZTOI(zp
));
1018 return (SET_ERROR(ENOENT
));
1021 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1022 zgd
->zgd_zilog
= zsb
->z_log
;
1023 zgd
->zgd_private
= zp
;
1026 * Write records come in two flavors: immediate and indirect.
1027 * For small writes it's cheaper to store the data with the
1028 * log record (immediate); for large writes it's cheaper to
1029 * sync the data and get a pointer to it (indirect) so that
1030 * we don't have to write the data twice.
1032 if (buf
!= NULL
) { /* immediate write */
1033 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
, size
,
1035 /* test for truncation needs to be done while range locked */
1036 if (offset
>= zp
->z_size
) {
1037 error
= SET_ERROR(ENOENT
);
1039 error
= dmu_read(os
, object
, offset
, size
, buf
,
1040 DMU_READ_NO_PREFETCH
);
1042 ASSERT(error
== 0 || error
== ENOENT
);
1043 } else { /* indirect write */
1045 * Have to lock the whole block to ensure when it's
1046 * written out and it's checksum is being calculated
1047 * that no one can change the data. We need to re-check
1048 * blocksize after we get the lock in case it's changed!
1053 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1055 zgd
->zgd_rl
= zfs_range_lock(&zp
->z_range_lock
, offset
,
1057 if (zp
->z_blksz
== size
)
1060 zfs_range_unlock(zgd
->zgd_rl
);
1062 /* test for truncation needs to be done while range locked */
1063 if (lr
->lr_offset
>= zp
->z_size
)
1064 error
= SET_ERROR(ENOENT
);
1067 error
= SET_ERROR(EIO
);
1072 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1073 DMU_READ_NO_PREFETCH
);
1076 blkptr_t
*obp
= dmu_buf_get_blkptr(db
);
1078 ASSERT(BP_IS_HOLE(bp
));
1085 ASSERT(db
->db_offset
== offset
);
1086 ASSERT(db
->db_size
== size
);
1088 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1090 ASSERT(error
|| lr
->lr_length
<= size
);
1093 * On success, we need to wait for the write I/O
1094 * initiated by dmu_sync() to complete before we can
1095 * release this dbuf. We will finish everything up
1096 * in the zfs_get_done() callback.
1101 if (error
== EALREADY
) {
1102 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
1108 zfs_get_done(zgd
, error
);
1115 zfs_access(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
1117 znode_t
*zp
= ITOZ(ip
);
1118 zfs_sb_t
*zsb
= ITOZSB(ip
);
1124 if (flag
& V_ACE_MASK
)
1125 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1127 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1132 EXPORT_SYMBOL(zfs_access
);
1135 * Lookup an entry in a directory, or an extended attribute directory.
1136 * If it exists, return a held inode reference for it.
1138 * IN: dip - inode of directory to search.
1139 * nm - name of entry to lookup.
1140 * flags - LOOKUP_XATTR set if looking for an attribute.
1141 * cr - credentials of caller.
1142 * direntflags - directory lookup flags
1143 * realpnp - returned pathname.
1145 * OUT: ipp - inode of located entry, NULL if not found.
1147 * RETURN: 0 on success, error code on failure.
1154 zfs_lookup(struct inode
*dip
, char *nm
, struct inode
**ipp
, int flags
,
1155 cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1157 znode_t
*zdp
= ITOZ(dip
);
1158 zfs_sb_t
*zsb
= ITOZSB(dip
);
1162 * Fast path lookup, however we must skip DNLC lookup
1163 * for case folding or normalizing lookups because the
1164 * DNLC code only stores the passed in name. This means
1165 * creating 'a' and removing 'A' on a case insensitive
1166 * file system would work, but DNLC still thinks 'a'
1167 * exists and won't let you create it again on the next
1168 * pass through fast path.
1170 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1172 if (!S_ISDIR(dip
->i_mode
)) {
1173 return (SET_ERROR(ENOTDIR
));
1174 } else if (zdp
->z_sa_hdl
== NULL
) {
1175 return (SET_ERROR(EIO
));
1178 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1179 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1187 } else if (!zdp
->z_zfsvfs
->z_norm
&&
1188 (zdp
->z_zfsvfs
->z_case
== ZFS_CASE_SENSITIVE
)) {
1190 vnode_t
*tvp
= dnlc_lookup(dvp
, nm
);
1193 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1198 if (tvp
== DNLC_NO_VNODE
) {
1200 return (SET_ERROR(ENOENT
));
1203 return (specvp_check(vpp
, cr
));
1206 #endif /* HAVE_DNLC */
1215 if (flags
& LOOKUP_XATTR
) {
1217 * We don't allow recursive attributes..
1218 * Maybe someday we will.
1220 if (zdp
->z_pflags
& ZFS_XATTR
) {
1222 return (SET_ERROR(EINVAL
));
1225 if ((error
= zfs_get_xattrdir(zdp
, ipp
, cr
, flags
))) {
1231 * Do we have permission to get into attribute directory?
1234 if ((error
= zfs_zaccess(ITOZ(*ipp
), ACE_EXECUTE
, 0,
1244 if (!S_ISDIR(dip
->i_mode
)) {
1246 return (SET_ERROR(ENOTDIR
));
1250 * Check accessibility of directory.
1253 if ((error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
))) {
1258 if (zsb
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1259 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1261 return (SET_ERROR(EILSEQ
));
1264 error
= zfs_dirlook(zdp
, nm
, ipp
, flags
, direntflags
, realpnp
);
1265 if ((error
== 0) && (*ipp
))
1266 zfs_inode_update(ITOZ(*ipp
));
1271 EXPORT_SYMBOL(zfs_lookup
);
1274 * Attempt to create a new entry in a directory. If the entry
1275 * already exists, truncate the file if permissible, else return
1276 * an error. Return the ip of the created or trunc'd file.
1278 * IN: dip - inode of directory to put new file entry in.
1279 * name - name of new file entry.
1280 * vap - attributes of new file.
1281 * excl - flag indicating exclusive or non-exclusive mode.
1282 * mode - mode to open file with.
1283 * cr - credentials of caller.
1284 * flag - large file flag [UNUSED].
1285 * vsecp - ACL to be set
1287 * OUT: ipp - inode of created or trunc'd entry.
1289 * RETURN: 0 on success, error code on failure.
1292 * dip - ctime|mtime updated if new entry created
1293 * ip - ctime|mtime always, atime if new
1298 zfs_create(struct inode
*dip
, char *name
, vattr_t
*vap
, int excl
,
1299 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1301 znode_t
*zp
, *dzp
= ITOZ(dip
);
1302 zfs_sb_t
*zsb
= ITOZSB(dip
);
1310 zfs_acl_ids_t acl_ids
;
1311 boolean_t fuid_dirtied
;
1312 boolean_t have_acl
= B_FALSE
;
1313 boolean_t waited
= B_FALSE
;
1316 * If we have an ephemeral id, ACL, or XVATTR then
1317 * make sure file system is at proper version
1323 if (zsb
->z_use_fuids
== B_FALSE
&&
1324 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1325 return (SET_ERROR(EINVAL
));
1328 return (SET_ERROR(EINVAL
));
1335 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
1336 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1338 return (SET_ERROR(EILSEQ
));
1341 if (vap
->va_mask
& ATTR_XVATTR
) {
1342 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1343 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1351 if (*name
== '\0') {
1353 * Null component name refers to the directory itself.
1360 /* possible igrab(zp) */
1363 if (flag
& FIGNORECASE
)
1366 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1370 zfs_acl_ids_free(&acl_ids
);
1371 if (strcmp(name
, "..") == 0)
1372 error
= SET_ERROR(EISDIR
);
1382 * Create a new file object and update the directory
1385 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1387 zfs_acl_ids_free(&acl_ids
);
1392 * We only support the creation of regular files in
1393 * extended attribute directories.
1396 if ((dzp
->z_pflags
& ZFS_XATTR
) && !S_ISREG(vap
->va_mode
)) {
1398 zfs_acl_ids_free(&acl_ids
);
1399 error
= SET_ERROR(EINVAL
);
1403 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1404 cr
, vsecp
, &acl_ids
)) != 0)
1408 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1409 zfs_acl_ids_free(&acl_ids
);
1410 error
= SET_ERROR(EDQUOT
);
1414 tx
= dmu_tx_create(os
);
1416 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1417 ZFS_SA_BASE_ATTR_SIZE
);
1419 fuid_dirtied
= zsb
->z_fuid_dirty
;
1421 zfs_fuid_txhold(zsb
, tx
);
1422 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1423 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1424 if (!zsb
->z_use_sa
&&
1425 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1426 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1427 0, acl_ids
.z_aclp
->z_acl_bytes
);
1429 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1431 zfs_dirent_unlock(dl
);
1432 if (error
== ERESTART
) {
1438 zfs_acl_ids_free(&acl_ids
);
1443 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1446 zfs_fuid_sync(zsb
, tx
);
1448 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1449 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1450 if (flag
& FIGNORECASE
)
1452 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1453 vsecp
, acl_ids
.z_fuidp
, vap
);
1454 zfs_acl_ids_free(&acl_ids
);
1457 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1460 zfs_acl_ids_free(&acl_ids
);
1464 * A directory entry already exists for this name.
1467 * Can't truncate an existing file if in exclusive mode.
1470 error
= SET_ERROR(EEXIST
);
1474 * Can't open a directory for writing.
1476 if (S_ISDIR(ZTOI(zp
)->i_mode
)) {
1477 error
= SET_ERROR(EISDIR
);
1481 * Verify requested access to file.
1483 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1487 mutex_enter(&dzp
->z_lock
);
1489 mutex_exit(&dzp
->z_lock
);
1492 * Truncate regular files if requested.
1494 if (S_ISREG(ZTOI(zp
)->i_mode
) &&
1495 (vap
->va_mask
& ATTR_SIZE
) && (vap
->va_size
== 0)) {
1496 /* we can't hold any locks when calling zfs_freesp() */
1498 zfs_dirent_unlock(dl
);
1501 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1507 zfs_dirent_unlock(dl
);
1513 zfs_inode_update(dzp
);
1514 zfs_inode_update(zp
);
1518 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1519 zil_commit(zilog
, 0);
1524 EXPORT_SYMBOL(zfs_create
);
1528 zfs_tmpfile(struct inode
*dip
, vattr_t
*vap
, int excl
,
1529 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1531 znode_t
*zp
= NULL
, *dzp
= ITOZ(dip
);
1532 zfs_sb_t
*zsb
= ITOZSB(dip
);
1538 zfs_acl_ids_t acl_ids
;
1539 boolean_t fuid_dirtied
;
1540 boolean_t have_acl
= B_FALSE
;
1541 boolean_t waited
= B_FALSE
;
1544 * If we have an ephemeral id, ACL, or XVATTR then
1545 * make sure file system is at proper version
1551 if (zsb
->z_use_fuids
== B_FALSE
&&
1552 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1553 return (SET_ERROR(EINVAL
));
1559 if (vap
->va_mask
& ATTR_XVATTR
) {
1560 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1561 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1571 * Create a new file object and update the directory
1574 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1576 zfs_acl_ids_free(&acl_ids
);
1580 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1581 cr
, vsecp
, &acl_ids
)) != 0)
1585 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1586 zfs_acl_ids_free(&acl_ids
);
1587 error
= SET_ERROR(EDQUOT
);
1591 tx
= dmu_tx_create(os
);
1593 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1594 ZFS_SA_BASE_ATTR_SIZE
);
1595 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
1597 fuid_dirtied
= zsb
->z_fuid_dirty
;
1599 zfs_fuid_txhold(zsb
, tx
);
1600 if (!zsb
->z_use_sa
&&
1601 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1602 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1603 0, acl_ids
.z_aclp
->z_acl_bytes
);
1605 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1607 if (error
== ERESTART
) {
1613 zfs_acl_ids_free(&acl_ids
);
1618 zfs_mknode(dzp
, vap
, tx
, cr
, IS_TMPFILE
, &zp
, &acl_ids
);
1621 zfs_fuid_sync(zsb
, tx
);
1623 /* Add to unlinked set */
1625 zfs_unlinked_add(zp
, tx
);
1626 zfs_acl_ids_free(&acl_ids
);
1634 zfs_inode_update(dzp
);
1635 zfs_inode_update(zp
);
1644 * Remove an entry from a directory.
1646 * IN: dip - inode of directory to remove entry from.
1647 * name - name of entry to remove.
1648 * cr - credentials of caller.
1650 * RETURN: 0 if success
1651 * error code if failure
1655 * ip - ctime (if nlink > 0)
1658 uint64_t null_xattr
= 0;
1662 zfs_remove(struct inode
*dip
, char *name
, cred_t
*cr
, int flags
)
1664 znode_t
*zp
, *dzp
= ITOZ(dip
);
1667 zfs_sb_t
*zsb
= ITOZSB(dip
);
1669 uint64_t acl_obj
, xattr_obj
;
1670 uint64_t xattr_obj_unlinked
= 0;
1675 boolean_t may_delete_now
, delete_now
= FALSE
;
1676 boolean_t unlinked
, toobig
= FALSE
;
1678 pathname_t
*realnmp
= NULL
;
1682 boolean_t waited
= B_FALSE
;
1685 return (SET_ERROR(EINVAL
));
1691 if (flags
& FIGNORECASE
) {
1701 * Attempt to lock directory; fail if entry doesn't exist.
1703 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1713 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1718 * Need to use rmdir for removing directories.
1720 if (S_ISDIR(ip
->i_mode
)) {
1721 error
= SET_ERROR(EPERM
);
1727 dnlc_remove(dvp
, realnmp
->pn_buf
);
1729 dnlc_remove(dvp
, name
);
1730 #endif /* HAVE_DNLC */
1732 mutex_enter(&zp
->z_lock
);
1733 may_delete_now
= atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
);
1734 mutex_exit(&zp
->z_lock
);
1737 * We may delete the znode now, or we may put it in the unlinked set;
1738 * it depends on whether we're the last link, and on whether there are
1739 * other holds on the inode. So we dmu_tx_hold() the right things to
1740 * allow for either case.
1743 tx
= dmu_tx_create(zsb
->z_os
);
1744 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1745 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1746 zfs_sa_upgrade_txholds(tx
, zp
);
1747 zfs_sa_upgrade_txholds(tx
, dzp
);
1748 if (may_delete_now
) {
1749 toobig
= zp
->z_size
> zp
->z_blksz
* zfs_delete_blocks
;
1750 /* if the file is too big, only hold_free a token amount */
1751 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1752 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1755 /* are there any extended attributes? */
1756 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1757 &xattr_obj
, sizeof (xattr_obj
));
1758 if (error
== 0 && xattr_obj
) {
1759 error
= zfs_zget(zsb
, xattr_obj
, &xzp
);
1761 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1762 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1765 mutex_enter(&zp
->z_lock
);
1766 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1767 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1768 mutex_exit(&zp
->z_lock
);
1770 /* charge as an update -- would be nice not to charge at all */
1771 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
1774 * Mark this transaction as typically resulting in a net free of space
1776 dmu_tx_mark_netfree(tx
);
1778 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1780 zfs_dirent_unlock(dl
);
1781 if (error
== ERESTART
) {
1801 * Remove the directory entry.
1803 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1812 * Hold z_lock so that we can make sure that the ACL obj
1813 * hasn't changed. Could have been deleted due to
1816 mutex_enter(&zp
->z_lock
);
1817 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
1818 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1819 delete_now
= may_delete_now
&& !toobig
&&
1820 atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
) &&
1821 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1826 if (xattr_obj_unlinked
) {
1827 ASSERT3U(ZTOI(xzp
)->i_nlink
, ==, 2);
1828 mutex_enter(&xzp
->z_lock
);
1829 xzp
->z_unlinked
= 1;
1830 clear_nlink(ZTOI(xzp
));
1832 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zsb
),
1833 &links
, sizeof (links
), tx
);
1834 ASSERT3U(error
, ==, 0);
1835 mutex_exit(&xzp
->z_lock
);
1836 zfs_unlinked_add(xzp
, tx
);
1839 error
= sa_remove(zp
->z_sa_hdl
,
1840 SA_ZPL_XATTR(zsb
), tx
);
1842 error
= sa_update(zp
->z_sa_hdl
,
1843 SA_ZPL_XATTR(zsb
), &null_xattr
,
1844 sizeof (uint64_t), tx
);
1848 * Add to the unlinked set because a new reference could be
1849 * taken concurrently resulting in a deferred destruction.
1851 zfs_unlinked_add(zp
, tx
);
1852 mutex_exit(&zp
->z_lock
);
1853 } else if (unlinked
) {
1854 mutex_exit(&zp
->z_lock
);
1855 zfs_unlinked_add(zp
, tx
);
1859 if (flags
& FIGNORECASE
)
1861 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1868 zfs_dirent_unlock(dl
);
1869 zfs_inode_update(dzp
);
1870 zfs_inode_update(zp
);
1878 zfs_inode_update(xzp
);
1879 zfs_iput_async(ZTOI(xzp
));
1882 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1883 zil_commit(zilog
, 0);
1888 EXPORT_SYMBOL(zfs_remove
);
1891 * Create a new directory and insert it into dip using the name
1892 * provided. Return a pointer to the inserted directory.
1894 * IN: dip - inode of directory to add subdir to.
1895 * dirname - name of new directory.
1896 * vap - attributes of new directory.
1897 * cr - credentials of caller.
1898 * vsecp - ACL to be set
1900 * OUT: ipp - inode of created directory.
1902 * RETURN: 0 if success
1903 * error code if failure
1906 * dip - ctime|mtime updated
1907 * ipp - ctime|mtime|atime updated
1911 zfs_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
, struct inode
**ipp
,
1912 cred_t
*cr
, int flags
, vsecattr_t
*vsecp
)
1914 znode_t
*zp
, *dzp
= ITOZ(dip
);
1915 zfs_sb_t
*zsb
= ITOZSB(dip
);
1923 gid_t gid
= crgetgid(cr
);
1924 zfs_acl_ids_t acl_ids
;
1925 boolean_t fuid_dirtied
;
1926 boolean_t waited
= B_FALSE
;
1928 ASSERT(S_ISDIR(vap
->va_mode
));
1931 * If we have an ephemeral id, ACL, or XVATTR then
1932 * make sure file system is at proper version
1936 if (zsb
->z_use_fuids
== B_FALSE
&&
1937 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1938 return (SET_ERROR(EINVAL
));
1940 if (dirname
== NULL
)
1941 return (SET_ERROR(EINVAL
));
1947 if (dzp
->z_pflags
& ZFS_XATTR
) {
1949 return (SET_ERROR(EINVAL
));
1952 if (zsb
->z_utf8
&& u8_validate(dirname
,
1953 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1955 return (SET_ERROR(EILSEQ
));
1957 if (flags
& FIGNORECASE
)
1960 if (vap
->va_mask
& ATTR_XVATTR
) {
1961 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1962 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1968 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1969 vsecp
, &acl_ids
)) != 0) {
1974 * First make sure the new directory doesn't exist.
1976 * Existence is checked first to make sure we don't return
1977 * EACCES instead of EEXIST which can cause some applications
1983 if ((error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1985 zfs_acl_ids_free(&acl_ids
);
1990 if ((error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
))) {
1991 zfs_acl_ids_free(&acl_ids
);
1992 zfs_dirent_unlock(dl
);
1997 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
1998 zfs_acl_ids_free(&acl_ids
);
1999 zfs_dirent_unlock(dl
);
2001 return (SET_ERROR(EDQUOT
));
2005 * Add a new entry to the directory.
2007 tx
= dmu_tx_create(zsb
->z_os
);
2008 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
2009 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
2010 fuid_dirtied
= zsb
->z_fuid_dirty
;
2012 zfs_fuid_txhold(zsb
, tx
);
2013 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
2014 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
2015 acl_ids
.z_aclp
->z_acl_bytes
);
2018 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
2019 ZFS_SA_BASE_ATTR_SIZE
);
2021 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
2023 zfs_dirent_unlock(dl
);
2024 if (error
== ERESTART
) {
2030 zfs_acl_ids_free(&acl_ids
);
2039 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
2042 zfs_fuid_sync(zsb
, tx
);
2045 * Now put new name in parent dir.
2047 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
2051 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
2052 if (flags
& FIGNORECASE
)
2054 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
2055 acl_ids
.z_fuidp
, vap
);
2057 zfs_acl_ids_free(&acl_ids
);
2061 zfs_dirent_unlock(dl
);
2063 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2064 zil_commit(zilog
, 0);
2066 zfs_inode_update(dzp
);
2067 zfs_inode_update(zp
);
2071 EXPORT_SYMBOL(zfs_mkdir
);
2074 * Remove a directory subdir entry. If the current working
2075 * directory is the same as the subdir to be removed, the
2078 * IN: dip - inode of directory to remove from.
2079 * name - name of directory to be removed.
2080 * cwd - inode of current working directory.
2081 * cr - credentials of caller.
2082 * flags - case flags
2084 * RETURN: 0 on success, error code on failure.
2087 * dip - ctime|mtime updated
2091 zfs_rmdir(struct inode
*dip
, char *name
, struct inode
*cwd
, cred_t
*cr
,
2094 znode_t
*dzp
= ITOZ(dip
);
2097 zfs_sb_t
*zsb
= ITOZSB(dip
);
2103 boolean_t waited
= B_FALSE
;
2106 return (SET_ERROR(EINVAL
));
2112 if (flags
& FIGNORECASE
)
2118 * Attempt to lock directory; fail if entry doesn't exist.
2120 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
2128 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
2132 if (!S_ISDIR(ip
->i_mode
)) {
2133 error
= SET_ERROR(ENOTDIR
);
2138 error
= SET_ERROR(EINVAL
);
2143 * Grab a lock on the directory to make sure that no one is
2144 * trying to add (or lookup) entries while we are removing it.
2146 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
2149 * Grab a lock on the parent pointer to make sure we play well
2150 * with the treewalk and directory rename code.
2152 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
2154 tx
= dmu_tx_create(zsb
->z_os
);
2155 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2156 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2157 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
2158 zfs_sa_upgrade_txholds(tx
, zp
);
2159 zfs_sa_upgrade_txholds(tx
, dzp
);
2160 dmu_tx_mark_netfree(tx
);
2161 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
2163 rw_exit(&zp
->z_parent_lock
);
2164 rw_exit(&zp
->z_name_lock
);
2165 zfs_dirent_unlock(dl
);
2166 if (error
== ERESTART
) {
2179 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2182 uint64_t txtype
= TX_RMDIR
;
2183 if (flags
& FIGNORECASE
)
2185 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2190 rw_exit(&zp
->z_parent_lock
);
2191 rw_exit(&zp
->z_name_lock
);
2193 zfs_dirent_unlock(dl
);
2195 zfs_inode_update(dzp
);
2196 zfs_inode_update(zp
);
2199 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2200 zil_commit(zilog
, 0);
2205 EXPORT_SYMBOL(zfs_rmdir
);
2208 * Read as many directory entries as will fit into the provided
2209 * dirent buffer from the given directory cursor position.
2211 * IN: ip - inode of directory to read.
2212 * dirent - buffer for directory entries.
2214 * OUT: dirent - filler buffer of directory entries.
2216 * RETURN: 0 if success
2217 * error code if failure
2220 * ip - atime updated
2222 * Note that the low 4 bits of the cookie returned by zap is always zero.
2223 * This allows us to use the low range for "special" directory entries:
2224 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2225 * we use the offset 2 for the '.zfs' directory.
2229 zfs_readdir(struct inode
*ip
, struct dir_context
*ctx
, cred_t
*cr
)
2231 znode_t
*zp
= ITOZ(ip
);
2232 zfs_sb_t
*zsb
= ITOZSB(ip
);
2235 zap_attribute_t zap
;
2241 uint64_t offset
; /* must be unsigned; checks for < 1 */
2246 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
2247 &parent
, sizeof (parent
))) != 0)
2251 * Quit if directory has been removed (posix)
2259 prefetch
= zp
->z_zn_prefetch
;
2262 * Initialize the iterator cursor.
2266 * Start iteration from the beginning of the directory.
2268 zap_cursor_init(&zc
, os
, zp
->z_id
);
2271 * The offset is a serialized cursor.
2273 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2277 * Transform to file-system independent format
2282 * Special case `.', `..', and `.zfs'.
2285 (void) strcpy(zap
.za_name
, ".");
2286 zap
.za_normalization_conflict
= 0;
2289 } else if (offset
== 1) {
2290 (void) strcpy(zap
.za_name
, "..");
2291 zap
.za_normalization_conflict
= 0;
2294 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2295 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2296 zap
.za_normalization_conflict
= 0;
2297 objnum
= ZFSCTL_INO_ROOT
;
2303 if ((error
= zap_cursor_retrieve(&zc
, &zap
))) {
2304 if (error
== ENOENT
)
2311 * Allow multiple entries provided the first entry is
2312 * the object id. Non-zpl consumers may safely make
2313 * use of the additional space.
2315 * XXX: This should be a feature flag for compatibility
2317 if (zap
.za_integer_length
!= 8 ||
2318 zap
.za_num_integers
== 0) {
2319 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2320 "entry, obj = %lld, offset = %lld, "
2321 "length = %d, num = %lld\n",
2322 (u_longlong_t
)zp
->z_id
,
2323 (u_longlong_t
)offset
,
2324 zap
.za_integer_length
,
2325 (u_longlong_t
)zap
.za_num_integers
);
2326 error
= SET_ERROR(ENXIO
);
2330 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2331 type
= ZFS_DIRENT_TYPE(zap
.za_first_integer
);
2334 done
= !dir_emit(ctx
, zap
.za_name
, strlen(zap
.za_name
),
2339 /* Prefetch znode */
2341 dmu_prefetch(os
, objnum
, 0, 0, 0,
2342 ZIO_PRIORITY_SYNC_READ
);
2346 * Move to the next entry, fill in the previous offset.
2348 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2349 zap_cursor_advance(&zc
);
2350 offset
= zap_cursor_serialize(&zc
);
2356 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2359 zap_cursor_fini(&zc
);
2360 if (error
== ENOENT
)
2367 EXPORT_SYMBOL(zfs_readdir
);
2369 ulong_t zfs_fsync_sync_cnt
= 4;
2372 zfs_fsync(struct inode
*ip
, int syncflag
, cred_t
*cr
)
2374 znode_t
*zp
= ITOZ(ip
);
2375 zfs_sb_t
*zsb
= ITOZSB(ip
);
2377 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2379 if (zsb
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2382 zil_commit(zsb
->z_log
, zp
->z_id
);
2385 tsd_set(zfs_fsyncer_key
, NULL
);
2389 EXPORT_SYMBOL(zfs_fsync
);
2393 * Get the requested file attributes and place them in the provided
2396 * IN: ip - inode of file.
2397 * vap - va_mask identifies requested attributes.
2398 * If ATTR_XVATTR set, then optional attrs are requested
2399 * flags - ATTR_NOACLCHECK (CIFS server context)
2400 * cr - credentials of caller.
2402 * OUT: vap - attribute values.
2404 * RETURN: 0 (always succeeds)
2408 zfs_getattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2410 znode_t
*zp
= ITOZ(ip
);
2411 zfs_sb_t
*zsb
= ITOZSB(ip
);
2414 uint64_t atime
[2], mtime
[2], ctime
[2];
2415 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2416 xoptattr_t
*xoap
= NULL
;
2417 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2418 sa_bulk_attr_t bulk
[3];
2424 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2426 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zsb
), NULL
, &atime
, 16);
2427 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
2428 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
2430 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2436 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2437 * Also, if we are the owner don't bother, since owner should
2438 * always be allowed to read basic attributes of file.
2440 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2441 (vap
->va_uid
!= crgetuid(cr
))) {
2442 if ((error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2450 * Return all attributes. It's cheaper to provide the answer
2451 * than to determine whether we were asked the question.
2454 mutex_enter(&zp
->z_lock
);
2455 vap
->va_type
= vn_mode_to_vtype(zp
->z_mode
);
2456 vap
->va_mode
= zp
->z_mode
;
2457 vap
->va_fsid
= ZTOI(zp
)->i_sb
->s_dev
;
2458 vap
->va_nodeid
= zp
->z_id
;
2459 if ((zp
->z_id
== zsb
->z_root
) && zfs_show_ctldir(zp
))
2460 links
= ZTOI(zp
)->i_nlink
+ 1;
2462 links
= ZTOI(zp
)->i_nlink
;
2463 vap
->va_nlink
= MIN(links
, ZFS_LINK_MAX
);
2464 vap
->va_size
= i_size_read(ip
);
2465 vap
->va_rdev
= ip
->i_rdev
;
2466 vap
->va_seq
= ip
->i_generation
;
2469 * Add in any requested optional attributes and the create time.
2470 * Also set the corresponding bits in the returned attribute bitmap.
2472 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zsb
->z_use_fuids
) {
2473 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2475 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2476 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2479 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2480 xoap
->xoa_readonly
=
2481 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2482 XVA_SET_RTN(xvap
, XAT_READONLY
);
2485 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2487 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2488 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2491 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2493 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2494 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2497 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2498 xoap
->xoa_nounlink
=
2499 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2500 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2503 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2504 xoap
->xoa_immutable
=
2505 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2506 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2509 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2510 xoap
->xoa_appendonly
=
2511 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2512 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2515 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2517 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2518 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2521 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2523 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2524 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2527 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2528 xoap
->xoa_av_quarantined
=
2529 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2530 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2533 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2534 xoap
->xoa_av_modified
=
2535 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2536 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2539 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2540 S_ISREG(ip
->i_mode
)) {
2541 zfs_sa_get_scanstamp(zp
, xvap
);
2544 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2547 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zsb
),
2548 times
, sizeof (times
));
2549 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2550 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2553 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2554 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2555 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2557 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2558 xoap
->xoa_generation
= ip
->i_generation
;
2559 XVA_SET_RTN(xvap
, XAT_GEN
);
2562 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2564 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2565 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2568 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2570 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2571 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2575 ZFS_TIME_DECODE(&vap
->va_atime
, atime
);
2576 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2577 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2579 mutex_exit(&zp
->z_lock
);
2581 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2583 if (zp
->z_blksz
== 0) {
2585 * Block size hasn't been set; suggest maximal I/O transfers.
2587 vap
->va_blksize
= zsb
->z_max_blksz
;
2593 EXPORT_SYMBOL(zfs_getattr
);
2596 * Get the basic file attributes and place them in the provided kstat
2597 * structure. The inode is assumed to be the authoritative source
2598 * for most of the attributes. However, the znode currently has the
2599 * authoritative atime, blksize, and block count.
2601 * IN: ip - inode of file.
2603 * OUT: sp - kstat values.
2605 * RETURN: 0 (always succeeds)
2609 zfs_getattr_fast(struct inode
*ip
, struct kstat
*sp
)
2611 znode_t
*zp
= ITOZ(ip
);
2612 zfs_sb_t
*zsb
= ITOZSB(ip
);
2614 u_longlong_t nblocks
;
2619 mutex_enter(&zp
->z_lock
);
2621 generic_fillattr(ip
, sp
);
2623 sa_object_size(zp
->z_sa_hdl
, &blksize
, &nblocks
);
2624 sp
->blksize
= blksize
;
2625 sp
->blocks
= nblocks
;
2627 if (unlikely(zp
->z_blksz
== 0)) {
2629 * Block size hasn't been set; suggest maximal I/O transfers.
2631 sp
->blksize
= zsb
->z_max_blksz
;
2634 mutex_exit(&zp
->z_lock
);
2637 * Required to prevent NFS client from detecting different inode
2638 * numbers of snapshot root dentry before and after snapshot mount.
2640 if (zsb
->z_issnap
) {
2641 if (ip
->i_sb
->s_root
->d_inode
== ip
)
2642 sp
->ino
= ZFSCTL_INO_SNAPDIRS
-
2643 dmu_objset_id(zsb
->z_os
);
2650 EXPORT_SYMBOL(zfs_getattr_fast
);
2653 * Set the file attributes to the values contained in the
2656 * IN: ip - inode of file to be modified.
2657 * vap - new attribute values.
2658 * If ATTR_XVATTR set, then optional attrs are being set
2659 * flags - ATTR_UTIME set if non-default time values provided.
2660 * - ATTR_NOACLCHECK (CIFS context only).
2661 * cr - credentials of caller.
2663 * RETURN: 0 if success
2664 * error code if failure
2667 * ip - ctime updated, mtime updated if size changed.
2671 zfs_setattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2673 znode_t
*zp
= ITOZ(ip
);
2674 zfs_sb_t
*zsb
= ITOZSB(ip
);
2678 xvattr_t
*tmpxvattr
;
2679 uint_t mask
= vap
->va_mask
;
2680 uint_t saved_mask
= 0;
2683 uint64_t new_kuid
= 0, new_kgid
= 0, new_uid
, new_gid
;
2685 uint64_t mtime
[2], ctime
[2], atime
[2];
2687 int need_policy
= FALSE
;
2689 zfs_fuid_info_t
*fuidp
= NULL
;
2690 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2693 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2694 boolean_t fuid_dirtied
= B_FALSE
;
2695 sa_bulk_attr_t
*bulk
, *xattr_bulk
;
2696 int count
= 0, xattr_count
= 0;
2707 * Make sure that if we have ephemeral uid/gid or xvattr specified
2708 * that file system is at proper version level
2711 if (zsb
->z_use_fuids
== B_FALSE
&&
2712 (((mask
& ATTR_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2713 ((mask
& ATTR_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2714 (mask
& ATTR_XVATTR
))) {
2716 return (SET_ERROR(EINVAL
));
2719 if (mask
& ATTR_SIZE
&& S_ISDIR(ip
->i_mode
)) {
2721 return (SET_ERROR(EISDIR
));
2724 if (mask
& ATTR_SIZE
&& !S_ISREG(ip
->i_mode
) && !S_ISFIFO(ip
->i_mode
)) {
2726 return (SET_ERROR(EINVAL
));
2730 * If this is an xvattr_t, then get a pointer to the structure of
2731 * optional attributes. If this is NULL, then we have a vattr_t.
2733 xoap
= xva_getxoptattr(xvap
);
2735 tmpxvattr
= kmem_alloc(sizeof (xvattr_t
), KM_SLEEP
);
2736 xva_init(tmpxvattr
);
2738 bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2739 xattr_bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * 7, KM_SLEEP
);
2742 * Immutable files can only alter immutable bit and atime
2744 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2745 ((mask
& (ATTR_SIZE
|ATTR_UID
|ATTR_GID
|ATTR_MTIME
|ATTR_MODE
)) ||
2746 ((mask
& ATTR_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2751 if ((mask
& ATTR_SIZE
) && (zp
->z_pflags
& ZFS_READONLY
)) {
2757 * Verify timestamps doesn't overflow 32 bits.
2758 * ZFS can handle large timestamps, but 32bit syscalls can't
2759 * handle times greater than 2039. This check should be removed
2760 * once large timestamps are fully supported.
2762 if (mask
& (ATTR_ATIME
| ATTR_MTIME
)) {
2763 if (((mask
& ATTR_ATIME
) &&
2764 TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2765 ((mask
& ATTR_MTIME
) &&
2766 TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2776 /* Can this be moved to before the top label? */
2777 if (zfs_is_readonly(zsb
)) {
2783 * First validate permissions
2786 if (mask
& ATTR_SIZE
) {
2787 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2792 * XXX - Note, we are not providing any open
2793 * mode flags here (like FNDELAY), so we may
2794 * block if there are locks present... this
2795 * should be addressed in openat().
2797 /* XXX - would it be OK to generate a log record here? */
2798 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2803 if (mask
& (ATTR_ATIME
|ATTR_MTIME
) ||
2804 ((mask
& ATTR_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2805 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2806 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2807 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2808 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2809 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2810 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2811 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2815 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2816 int idmask
= (mask
& (ATTR_UID
|ATTR_GID
));
2821 * NOTE: even if a new mode is being set,
2822 * we may clear S_ISUID/S_ISGID bits.
2825 if (!(mask
& ATTR_MODE
))
2826 vap
->va_mode
= zp
->z_mode
;
2829 * Take ownership or chgrp to group we are a member of
2832 take_owner
= (mask
& ATTR_UID
) && (vap
->va_uid
== crgetuid(cr
));
2833 take_group
= (mask
& ATTR_GID
) &&
2834 zfs_groupmember(zsb
, vap
->va_gid
, cr
);
2837 * If both ATTR_UID and ATTR_GID are set then take_owner and
2838 * take_group must both be set in order to allow taking
2841 * Otherwise, send the check through secpolicy_vnode_setattr()
2845 if (((idmask
== (ATTR_UID
|ATTR_GID
)) &&
2846 take_owner
&& take_group
) ||
2847 ((idmask
== ATTR_UID
) && take_owner
) ||
2848 ((idmask
== ATTR_GID
) && take_group
)) {
2849 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2850 skipaclchk
, cr
) == 0) {
2852 * Remove setuid/setgid for non-privileged users
2854 (void) secpolicy_setid_clear(vap
, cr
);
2855 trim_mask
= (mask
& (ATTR_UID
|ATTR_GID
));
2864 mutex_enter(&zp
->z_lock
);
2865 oldva
.va_mode
= zp
->z_mode
;
2866 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2867 if (mask
& ATTR_XVATTR
) {
2869 * Update xvattr mask to include only those attributes
2870 * that are actually changing.
2872 * the bits will be restored prior to actually setting
2873 * the attributes so the caller thinks they were set.
2875 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2876 if (xoap
->xoa_appendonly
!=
2877 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2880 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2881 XVA_SET_REQ(tmpxvattr
, XAT_APPENDONLY
);
2885 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2886 if (xoap
->xoa_nounlink
!=
2887 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2890 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2891 XVA_SET_REQ(tmpxvattr
, XAT_NOUNLINK
);
2895 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2896 if (xoap
->xoa_immutable
!=
2897 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2900 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2901 XVA_SET_REQ(tmpxvattr
, XAT_IMMUTABLE
);
2905 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2906 if (xoap
->xoa_nodump
!=
2907 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2910 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2911 XVA_SET_REQ(tmpxvattr
, XAT_NODUMP
);
2915 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2916 if (xoap
->xoa_av_modified
!=
2917 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2920 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2921 XVA_SET_REQ(tmpxvattr
, XAT_AV_MODIFIED
);
2925 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2926 if ((!S_ISREG(ip
->i_mode
) &&
2927 xoap
->xoa_av_quarantined
) ||
2928 xoap
->xoa_av_quarantined
!=
2929 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2932 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2933 XVA_SET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
);
2937 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2938 mutex_exit(&zp
->z_lock
);
2943 if (need_policy
== FALSE
&&
2944 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
2945 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
2950 mutex_exit(&zp
->z_lock
);
2952 if (mask
& ATTR_MODE
) {
2953 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
2954 err
= secpolicy_setid_setsticky_clear(ip
, vap
,
2959 trim_mask
|= ATTR_MODE
;
2967 * If trim_mask is set then take ownership
2968 * has been granted or write_acl is present and user
2969 * has the ability to modify mode. In that case remove
2970 * UID|GID and or MODE from mask so that
2971 * secpolicy_vnode_setattr() doesn't revoke it.
2975 saved_mask
= vap
->va_mask
;
2976 vap
->va_mask
&= ~trim_mask
;
2978 err
= secpolicy_vnode_setattr(cr
, ip
, vap
, &oldva
, flags
,
2979 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
2984 vap
->va_mask
|= saved_mask
;
2988 * secpolicy_vnode_setattr, or take ownership may have
2991 mask
= vap
->va_mask
;
2993 if ((mask
& (ATTR_UID
| ATTR_GID
))) {
2994 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zsb
),
2995 &xattr_obj
, sizeof (xattr_obj
));
2997 if (err
== 0 && xattr_obj
) {
2998 err
= zfs_zget(ZTOZSB(zp
), xattr_obj
, &attrzp
);
3002 if (mask
& ATTR_UID
) {
3003 new_kuid
= zfs_fuid_create(zsb
,
3004 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
3005 if (new_kuid
!= KUID_TO_SUID(ZTOI(zp
)->i_uid
) &&
3006 zfs_fuid_overquota(zsb
, B_FALSE
, new_kuid
)) {
3014 if (mask
& ATTR_GID
) {
3015 new_kgid
= zfs_fuid_create(zsb
, (uint64_t)vap
->va_gid
,
3016 cr
, ZFS_GROUP
, &fuidp
);
3017 if (new_kgid
!= KGID_TO_SGID(ZTOI(zp
)->i_gid
) &&
3018 zfs_fuid_overquota(zsb
, B_TRUE
, new_kgid
)) {
3026 tx
= dmu_tx_create(zsb
->z_os
);
3028 if (mask
& ATTR_MODE
) {
3029 uint64_t pmode
= zp
->z_mode
;
3031 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
3033 zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
);
3035 mutex_enter(&zp
->z_lock
);
3036 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
3038 * Are we upgrading ACL from old V0 format
3041 if (zsb
->z_version
>= ZPL_VERSION_FUID
&&
3042 zfs_znode_acl_version(zp
) ==
3043 ZFS_ACL_VERSION_INITIAL
) {
3044 dmu_tx_hold_free(tx
, acl_obj
, 0,
3046 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3047 0, aclp
->z_acl_bytes
);
3049 dmu_tx_hold_write(tx
, acl_obj
, 0,
3052 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3053 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3054 0, aclp
->z_acl_bytes
);
3056 mutex_exit(&zp
->z_lock
);
3057 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3059 if ((mask
& ATTR_XVATTR
) &&
3060 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3061 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3063 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3067 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
3070 fuid_dirtied
= zsb
->z_fuid_dirty
;
3072 zfs_fuid_txhold(zsb
, tx
);
3074 zfs_sa_upgrade_txholds(tx
, zp
);
3076 err
= dmu_tx_assign(tx
, TXG_WAIT
);
3082 * Set each attribute requested.
3083 * We group settings according to the locks they need to acquire.
3085 * Note: you cannot set ctime directly, although it will be
3086 * updated as a side-effect of calling this function.
3090 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3091 mutex_enter(&zp
->z_acl_lock
);
3092 mutex_enter(&zp
->z_lock
);
3094 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zsb
), NULL
,
3095 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
3098 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3099 mutex_enter(&attrzp
->z_acl_lock
);
3100 mutex_enter(&attrzp
->z_lock
);
3101 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3102 SA_ZPL_FLAGS(zsb
), NULL
, &attrzp
->z_pflags
,
3103 sizeof (attrzp
->z_pflags
));
3106 if (mask
& (ATTR_UID
|ATTR_GID
)) {
3108 if (mask
& ATTR_UID
) {
3109 ZTOI(zp
)->i_uid
= SUID_TO_KUID(new_kuid
);
3110 new_uid
= zfs_uid_read(ZTOI(zp
));
3111 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zsb
), NULL
,
3112 &new_uid
, sizeof (new_uid
));
3114 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3115 SA_ZPL_UID(zsb
), NULL
, &new_uid
,
3117 ZTOI(attrzp
)->i_uid
= SUID_TO_KUID(new_uid
);
3121 if (mask
& ATTR_GID
) {
3122 ZTOI(zp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3123 new_gid
= zfs_gid_read(ZTOI(zp
));
3124 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zsb
),
3125 NULL
, &new_gid
, sizeof (new_gid
));
3127 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3128 SA_ZPL_GID(zsb
), NULL
, &new_gid
,
3130 ZTOI(attrzp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3133 if (!(mask
& ATTR_MODE
)) {
3134 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
),
3135 NULL
, &new_mode
, sizeof (new_mode
));
3136 new_mode
= zp
->z_mode
;
3138 err
= zfs_acl_chown_setattr(zp
);
3141 err
= zfs_acl_chown_setattr(attrzp
);
3146 if (mask
& ATTR_MODE
) {
3147 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zsb
), NULL
,
3148 &new_mode
, sizeof (new_mode
));
3149 zp
->z_mode
= ZTOI(zp
)->i_mode
= new_mode
;
3150 ASSERT3P(aclp
, !=, NULL
);
3151 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
3153 if (zp
->z_acl_cached
)
3154 zfs_acl_free(zp
->z_acl_cached
);
3155 zp
->z_acl_cached
= aclp
;
3159 if ((mask
& ATTR_ATIME
) || zp
->z_atime_dirty
) {
3160 zp
->z_atime_dirty
= 0;
3161 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
3162 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zsb
), NULL
,
3163 &atime
, sizeof (atime
));
3166 if (mask
& ATTR_MTIME
) {
3167 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3168 ZTOI(zp
)->i_mtime
= timespec_trunc(vap
->va_mtime
,
3169 ZTOI(zp
)->i_sb
->s_time_gran
);
3171 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zsb
), NULL
,
3172 mtime
, sizeof (mtime
));
3175 if (mask
& ATTR_CTIME
) {
3176 ZFS_TIME_ENCODE(&vap
->va_ctime
, ctime
);
3177 ZTOI(zp
)->i_ctime
= timespec_trunc(vap
->va_ctime
,
3178 ZTOI(zp
)->i_sb
->s_time_gran
);
3179 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zsb
), NULL
,
3180 ctime
, sizeof (ctime
));
3183 if (attrzp
&& mask
) {
3184 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3185 SA_ZPL_CTIME(zsb
), NULL
, &ctime
,
3190 * Do this after setting timestamps to prevent timestamp
3191 * update from toggling bit
3194 if (xoap
&& (mask
& ATTR_XVATTR
)) {
3197 * restore trimmed off masks
3198 * so that return masks can be set for caller.
3201 if (XVA_ISSET_REQ(tmpxvattr
, XAT_APPENDONLY
)) {
3202 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3204 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NOUNLINK
)) {
3205 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3207 if (XVA_ISSET_REQ(tmpxvattr
, XAT_IMMUTABLE
)) {
3208 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3210 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NODUMP
)) {
3211 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3213 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_MODIFIED
)) {
3214 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3216 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
)) {
3217 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3220 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3221 ASSERT(S_ISREG(ip
->i_mode
));
3223 zfs_xvattr_set(zp
, xvap
, tx
);
3227 zfs_fuid_sync(zsb
, tx
);
3230 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3232 mutex_exit(&zp
->z_lock
);
3233 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3234 mutex_exit(&zp
->z_acl_lock
);
3237 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3238 mutex_exit(&attrzp
->z_acl_lock
);
3239 mutex_exit(&attrzp
->z_lock
);
3242 if (err
== 0 && attrzp
) {
3243 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3252 zfs_fuid_info_free(fuidp
);
3260 if (err
== ERESTART
)
3263 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3267 zfs_inode_update(zp
);
3271 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3272 zil_commit(zilog
, 0);
3275 kmem_free(xattr_bulk
, sizeof (sa_bulk_attr_t
) * 7);
3276 kmem_free(bulk
, sizeof (sa_bulk_attr_t
) * 7);
3277 kmem_free(tmpxvattr
, sizeof (xvattr_t
));
3281 EXPORT_SYMBOL(zfs_setattr
);
3283 typedef struct zfs_zlock
{
3284 krwlock_t
*zl_rwlock
; /* lock we acquired */
3285 znode_t
*zl_znode
; /* znode we held */
3286 struct zfs_zlock
*zl_next
; /* next in list */
3290 * Drop locks and release vnodes that were held by zfs_rename_lock().
3293 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3297 while ((zl
= *zlpp
) != NULL
) {
3298 if (zl
->zl_znode
!= NULL
)
3299 zfs_iput_async(ZTOI(zl
->zl_znode
));
3300 rw_exit(zl
->zl_rwlock
);
3301 *zlpp
= zl
->zl_next
;
3302 kmem_free(zl
, sizeof (*zl
));
3307 * Search back through the directory tree, using the ".." entries.
3308 * Lock each directory in the chain to prevent concurrent renames.
3309 * Fail any attempt to move a directory into one of its own descendants.
3310 * XXX - z_parent_lock can overlap with map or grow locks
3313 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3317 uint64_t rootid
= ZTOZSB(zp
)->z_root
;
3318 uint64_t oidp
= zp
->z_id
;
3319 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3320 krw_t rw
= RW_WRITER
;
3323 * First pass write-locks szp and compares to zp->z_id.
3324 * Later passes read-lock zp and compare to zp->z_parent.
3327 if (!rw_tryenter(rwlp
, rw
)) {
3329 * Another thread is renaming in this path.
3330 * Note that if we are a WRITER, we don't have any
3331 * parent_locks held yet.
3333 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3335 * Drop our locks and restart
3337 zfs_rename_unlock(&zl
);
3341 rwlp
= &szp
->z_parent_lock
;
3346 * Wait for other thread to drop its locks
3352 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3353 zl
->zl_rwlock
= rwlp
;
3354 zl
->zl_znode
= NULL
;
3355 zl
->zl_next
= *zlpp
;
3358 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3359 return (SET_ERROR(EINVAL
));
3361 if (oidp
== rootid
) /* We've hit the top */
3364 if (rw
== RW_READER
) { /* i.e. not the first pass */
3365 int error
= zfs_zget(ZTOZSB(zp
), oidp
, &zp
);
3370 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(ZTOZSB(zp
)),
3371 &oidp
, sizeof (oidp
));
3372 rwlp
= &zp
->z_parent_lock
;
3375 } while (zp
->z_id
!= sdzp
->z_id
);
3381 * Move an entry from the provided source directory to the target
3382 * directory. Change the entry name as indicated.
3384 * IN: sdip - Source directory containing the "old entry".
3385 * snm - Old entry name.
3386 * tdip - Target directory to contain the "new entry".
3387 * tnm - New entry name.
3388 * cr - credentials of caller.
3389 * flags - case flags
3391 * RETURN: 0 on success, error code on failure.
3394 * sdip,tdip - ctime|mtime updated
3398 zfs_rename(struct inode
*sdip
, char *snm
, struct inode
*tdip
, char *tnm
,
3399 cred_t
*cr
, int flags
)
3401 znode_t
*tdzp
, *szp
, *tzp
;
3402 znode_t
*sdzp
= ITOZ(sdip
);
3403 zfs_sb_t
*zsb
= ITOZSB(sdip
);
3405 zfs_dirlock_t
*sdl
, *tdl
;
3408 int cmp
, serr
, terr
;
3411 boolean_t waited
= B_FALSE
;
3413 if (snm
== NULL
|| tnm
== NULL
)
3414 return (SET_ERROR(EINVAL
));
3417 ZFS_VERIFY_ZP(sdzp
);
3421 ZFS_VERIFY_ZP(tdzp
);
3424 * We check i_sb because snapshots and the ctldir must have different
3427 if (tdip
->i_sb
!= sdip
->i_sb
|| zfsctl_is_node(tdip
)) {
3429 return (SET_ERROR(EXDEV
));
3432 if (zsb
->z_utf8
&& u8_validate(tnm
,
3433 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3435 return (SET_ERROR(EILSEQ
));
3438 if (flags
& FIGNORECASE
)
3447 * This is to prevent the creation of links into attribute space
3448 * by renaming a linked file into/outof an attribute directory.
3449 * See the comment in zfs_link() for why this is considered bad.
3451 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3453 return (SET_ERROR(EINVAL
));
3457 * Lock source and target directory entries. To prevent deadlock,
3458 * a lock ordering must be defined. We lock the directory with
3459 * the smallest object id first, or if it's a tie, the one with
3460 * the lexically first name.
3462 if (sdzp
->z_id
< tdzp
->z_id
) {
3464 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3468 * First compare the two name arguments without
3469 * considering any case folding.
3471 int nofold
= (zsb
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3473 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3474 ASSERT(error
== 0 || !zsb
->z_utf8
);
3477 * POSIX: "If the old argument and the new argument
3478 * both refer to links to the same existing file,
3479 * the rename() function shall return successfully
3480 * and perform no other action."
3486 * If the file system is case-folding, then we may
3487 * have some more checking to do. A case-folding file
3488 * system is either supporting mixed case sensitivity
3489 * access or is completely case-insensitive. Note
3490 * that the file system is always case preserving.
3492 * In mixed sensitivity mode case sensitive behavior
3493 * is the default. FIGNORECASE must be used to
3494 * explicitly request case insensitive behavior.
3496 * If the source and target names provided differ only
3497 * by case (e.g., a request to rename 'tim' to 'Tim'),
3498 * we will treat this as a special case in the
3499 * case-insensitive mode: as long as the source name
3500 * is an exact match, we will allow this to proceed as
3501 * a name-change request.
3503 if ((zsb
->z_case
== ZFS_CASE_INSENSITIVE
||
3504 (zsb
->z_case
== ZFS_CASE_MIXED
&&
3505 flags
& FIGNORECASE
)) &&
3506 u8_strcmp(snm
, tnm
, 0, zsb
->z_norm
, U8_UNICODE_LATEST
,
3509 * case preserving rename request, require exact
3518 * If the source and destination directories are the same, we should
3519 * grab the z_name_lock of that directory only once.
3523 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3527 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3528 ZEXISTS
| zflg
, NULL
, NULL
);
3529 terr
= zfs_dirent_lock(&tdl
,
3530 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3532 terr
= zfs_dirent_lock(&tdl
,
3533 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3534 serr
= zfs_dirent_lock(&sdl
,
3535 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3541 * Source entry invalid or not there.
3544 zfs_dirent_unlock(tdl
);
3550 rw_exit(&sdzp
->z_name_lock
);
3552 if (strcmp(snm
, "..") == 0)
3558 zfs_dirent_unlock(sdl
);
3562 rw_exit(&sdzp
->z_name_lock
);
3564 if (strcmp(tnm
, "..") == 0)
3571 * Must have write access at the source to remove the old entry
3572 * and write access at the target to create the new entry.
3573 * Note that if target and source are the same, this can be
3574 * done in a single check.
3577 if ((error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
)))
3580 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3582 * Check to make sure rename is valid.
3583 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3585 if ((error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
)))
3590 * Does target exist?
3594 * Source and target must be the same type.
3596 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3597 if (!S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3598 error
= SET_ERROR(ENOTDIR
);
3602 if (S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3603 error
= SET_ERROR(EISDIR
);
3608 * POSIX dictates that when the source and target
3609 * entries refer to the same file object, rename
3610 * must do nothing and exit without error.
3612 if (szp
->z_id
== tzp
->z_id
) {
3618 tx
= dmu_tx_create(zsb
->z_os
);
3619 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3620 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3621 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3622 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3624 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3625 zfs_sa_upgrade_txholds(tx
, tdzp
);
3628 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3629 zfs_sa_upgrade_txholds(tx
, tzp
);
3632 zfs_sa_upgrade_txholds(tx
, szp
);
3633 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
3634 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3637 zfs_rename_unlock(&zl
);
3638 zfs_dirent_unlock(sdl
);
3639 zfs_dirent_unlock(tdl
);
3642 rw_exit(&sdzp
->z_name_lock
);
3644 if (error
== ERESTART
) {
3661 if (tzp
) /* Attempt to remove the existing target */
3662 error
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3665 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3667 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3669 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zsb
),
3670 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3673 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3675 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3676 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3677 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3680 * At this point, we have successfully created
3681 * the target name, but have failed to remove
3682 * the source name. Since the create was done
3683 * with the ZRENAMING flag, there are
3684 * complications; for one, the link count is
3685 * wrong. The easiest way to deal with this
3686 * is to remove the newly created target, and
3687 * return the original error. This must
3688 * succeed; fortunately, it is very unlikely to
3689 * fail, since we just created it.
3691 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3692 ZRENAMING
, NULL
), ==, 0);
3700 zfs_rename_unlock(&zl
);
3702 zfs_dirent_unlock(sdl
);
3703 zfs_dirent_unlock(tdl
);
3705 zfs_inode_update(sdzp
);
3707 rw_exit(&sdzp
->z_name_lock
);
3710 zfs_inode_update(tdzp
);
3712 zfs_inode_update(szp
);
3715 zfs_inode_update(tzp
);
3719 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3720 zil_commit(zilog
, 0);
3725 EXPORT_SYMBOL(zfs_rename
);
3728 * Insert the indicated symbolic reference entry into the directory.
3730 * IN: dip - Directory to contain new symbolic link.
3731 * link - Name for new symlink entry.
3732 * vap - Attributes of new entry.
3733 * target - Target path of new symlink.
3735 * cr - credentials of caller.
3736 * flags - case flags
3738 * RETURN: 0 on success, error code on failure.
3741 * dip - ctime|mtime updated
3745 zfs_symlink(struct inode
*dip
, char *name
, vattr_t
*vap
, char *link
,
3746 struct inode
**ipp
, cred_t
*cr
, int flags
)
3748 znode_t
*zp
, *dzp
= ITOZ(dip
);
3751 zfs_sb_t
*zsb
= ITOZSB(dip
);
3753 uint64_t len
= strlen(link
);
3756 zfs_acl_ids_t acl_ids
;
3757 boolean_t fuid_dirtied
;
3758 uint64_t txtype
= TX_SYMLINK
;
3759 boolean_t waited
= B_FALSE
;
3761 ASSERT(S_ISLNK(vap
->va_mode
));
3764 return (SET_ERROR(EINVAL
));
3770 if (zsb
->z_utf8
&& u8_validate(name
, strlen(name
),
3771 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3773 return (SET_ERROR(EILSEQ
));
3775 if (flags
& FIGNORECASE
)
3778 if (len
> MAXPATHLEN
) {
3780 return (SET_ERROR(ENAMETOOLONG
));
3783 if ((error
= zfs_acl_ids_create(dzp
, 0,
3784 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3792 * Attempt to lock directory; fail if entry already exists.
3794 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3796 zfs_acl_ids_free(&acl_ids
);
3801 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
3802 zfs_acl_ids_free(&acl_ids
);
3803 zfs_dirent_unlock(dl
);
3808 if (zfs_acl_ids_overquota(zsb
, &acl_ids
)) {
3809 zfs_acl_ids_free(&acl_ids
);
3810 zfs_dirent_unlock(dl
);
3812 return (SET_ERROR(EDQUOT
));
3814 tx
= dmu_tx_create(zsb
->z_os
);
3815 fuid_dirtied
= zsb
->z_fuid_dirty
;
3816 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3817 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3818 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3819 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3820 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3821 if (!zsb
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3822 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3823 acl_ids
.z_aclp
->z_acl_bytes
);
3826 zfs_fuid_txhold(zsb
, tx
);
3827 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3829 zfs_dirent_unlock(dl
);
3830 if (error
== ERESTART
) {
3836 zfs_acl_ids_free(&acl_ids
);
3843 * Create a new object for the symlink.
3844 * for version 4 ZPL datsets the symlink will be an SA attribute
3846 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3849 zfs_fuid_sync(zsb
, tx
);
3851 mutex_enter(&zp
->z_lock
);
3853 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zsb
),
3856 zfs_sa_symlink(zp
, link
, len
, tx
);
3857 mutex_exit(&zp
->z_lock
);
3860 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zsb
),
3861 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3863 * Insert the new object into the directory.
3865 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3867 if (flags
& FIGNORECASE
)
3869 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3871 zfs_inode_update(dzp
);
3872 zfs_inode_update(zp
);
3874 zfs_acl_ids_free(&acl_ids
);
3878 zfs_dirent_unlock(dl
);
3882 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3883 zil_commit(zilog
, 0);
3888 EXPORT_SYMBOL(zfs_symlink
);
3891 * Return, in the buffer contained in the provided uio structure,
3892 * the symbolic path referred to by ip.
3894 * IN: ip - inode of symbolic link
3895 * uio - structure to contain the link path.
3896 * cr - credentials of caller.
3898 * RETURN: 0 if success
3899 * error code if failure
3902 * ip - atime updated
3906 zfs_readlink(struct inode
*ip
, uio_t
*uio
, cred_t
*cr
)
3908 znode_t
*zp
= ITOZ(ip
);
3909 zfs_sb_t
*zsb
= ITOZSB(ip
);
3915 mutex_enter(&zp
->z_lock
);
3917 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3918 SA_ZPL_SYMLINK(zsb
), uio
);
3920 error
= zfs_sa_readlink(zp
, uio
);
3921 mutex_exit(&zp
->z_lock
);
3926 EXPORT_SYMBOL(zfs_readlink
);
3929 * Insert a new entry into directory tdip referencing sip.
3931 * IN: tdip - Directory to contain new entry.
3932 * sip - inode of new entry.
3933 * name - name of new entry.
3934 * cr - credentials of caller.
3936 * RETURN: 0 if success
3937 * error code if failure
3940 * tdip - ctime|mtime updated
3941 * sip - ctime updated
3945 zfs_link(struct inode
*tdip
, struct inode
*sip
, char *name
, cred_t
*cr
,
3948 znode_t
*dzp
= ITOZ(tdip
);
3950 zfs_sb_t
*zsb
= ITOZSB(tdip
);
3958 boolean_t waited
= B_FALSE
;
3959 boolean_t is_tmpfile
= 0;
3962 is_tmpfile
= (sip
->i_nlink
== 0 && (sip
->i_state
& I_LINKABLE
));
3964 ASSERT(S_ISDIR(tdip
->i_mode
));
3967 return (SET_ERROR(EINVAL
));
3974 * POSIX dictates that we return EPERM here.
3975 * Better choices include ENOTSUP or EISDIR.
3977 if (S_ISDIR(sip
->i_mode
)) {
3979 return (SET_ERROR(EPERM
));
3986 * We check i_sb because snapshots and the ctldir must have different
3989 if (sip
->i_sb
!= tdip
->i_sb
|| zfsctl_is_node(sip
)) {
3991 return (SET_ERROR(EXDEV
));
3994 /* Prevent links to .zfs/shares files */
3996 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zsb
),
3997 &parent
, sizeof (uint64_t))) != 0) {
4001 if (parent
== zsb
->z_shares_dir
) {
4003 return (SET_ERROR(EPERM
));
4006 if (zsb
->z_utf8
&& u8_validate(name
,
4007 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
4009 return (SET_ERROR(EILSEQ
));
4011 if (flags
& FIGNORECASE
)
4015 * We do not support links between attributes and non-attributes
4016 * because of the potential security risk of creating links
4017 * into "normal" file space in order to circumvent restrictions
4018 * imposed in attribute space.
4020 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
4022 return (SET_ERROR(EINVAL
));
4025 owner
= zfs_fuid_map_id(zsb
, KUID_TO_SUID(sip
->i_uid
), cr
, ZFS_OWNER
);
4026 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
4028 return (SET_ERROR(EPERM
));
4031 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
4038 * Attempt to lock directory; fail if entry already exists.
4040 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
4046 tx
= dmu_tx_create(zsb
->z_os
);
4047 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
4048 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
4050 dmu_tx_hold_zap(tx
, zsb
->z_unlinkedobj
, FALSE
, NULL
);
4052 zfs_sa_upgrade_txholds(tx
, szp
);
4053 zfs_sa_upgrade_txholds(tx
, dzp
);
4054 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
4056 zfs_dirent_unlock(dl
);
4057 if (error
== ERESTART
) {
4067 /* unmark z_unlinked so zfs_link_create will not reject */
4069 szp
->z_unlinked
= 0;
4070 error
= zfs_link_create(dl
, szp
, tx
, 0);
4073 uint64_t txtype
= TX_LINK
;
4075 * tmpfile is created to be in z_unlinkedobj, so remove it.
4076 * Also, we don't log in ZIL, be cause all previous file
4077 * operation on the tmpfile are ignored by ZIL. Instead we
4078 * always wait for txg to sync to make sure all previous
4079 * operation are sync safe.
4082 VERIFY(zap_remove_int(zsb
->z_os
, zsb
->z_unlinkedobj
,
4083 szp
->z_id
, tx
) == 0);
4085 if (flags
& FIGNORECASE
)
4087 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
4089 } else if (is_tmpfile
) {
4090 /* restore z_unlinked since when linking failed */
4091 szp
->z_unlinked
= 1;
4093 txg
= dmu_tx_get_txg(tx
);
4096 zfs_dirent_unlock(dl
);
4098 if (!is_tmpfile
&& zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4099 zil_commit(zilog
, 0);
4102 txg_wait_synced(dmu_objset_pool(zsb
->z_os
), txg
);
4104 zfs_inode_update(dzp
);
4105 zfs_inode_update(szp
);
4109 EXPORT_SYMBOL(zfs_link
);
4112 zfs_putpage_commit_cb(void *arg
)
4114 struct page
*pp
= arg
;
4117 end_page_writeback(pp
);
4121 * Push a page out to disk, once the page is on stable storage the
4122 * registered commit callback will be run as notification of completion.
4124 * IN: ip - page mapped for inode.
4125 * pp - page to push (page is locked)
4126 * wbc - writeback control data
4128 * RETURN: 0 if success
4129 * error code if failure
4132 * ip - ctime|mtime updated
4136 zfs_putpage(struct inode
*ip
, struct page
*pp
, struct writeback_control
*wbc
)
4138 znode_t
*zp
= ITOZ(ip
);
4139 zfs_sb_t
*zsb
= ITOZSB(ip
);
4147 uint64_t mtime
[2], ctime
[2];
4148 sa_bulk_attr_t bulk
[3];
4150 struct address_space
*mapping
;
4155 ASSERT(PageLocked(pp
));
4157 pgoff
= page_offset(pp
); /* Page byte-offset in file */
4158 offset
= i_size_read(ip
); /* File length in bytes */
4159 pglen
= MIN(PAGE_SIZE
, /* Page length in bytes */
4160 P2ROUNDUP(offset
, PAGE_SIZE
)-pgoff
);
4162 /* Page is beyond end of file */
4163 if (pgoff
>= offset
) {
4169 /* Truncate page length to end of file */
4170 if (pgoff
+ pglen
> offset
)
4171 pglen
= offset
- pgoff
;
4175 * FIXME: Allow mmap writes past its quota. The correct fix
4176 * is to register a page_mkwrite() handler to count the page
4177 * against its quota when it is about to be dirtied.
4179 if (zfs_owner_overquota(zsb
, zp
, B_FALSE
) ||
4180 zfs_owner_overquota(zsb
, zp
, B_TRUE
)) {
4186 * The ordering here is critical and must adhere to the following
4187 * rules in order to avoid deadlocking in either zfs_read() or
4188 * zfs_free_range() due to a lock inversion.
4190 * 1) The page must be unlocked prior to acquiring the range lock.
4191 * This is critical because zfs_read() calls find_lock_page()
4192 * which may block on the page lock while holding the range lock.
4194 * 2) Before setting or clearing write back on a page the range lock
4195 * must be held in order to prevent a lock inversion with the
4196 * zfs_free_range() function.
4198 * This presents a problem because upon entering this function the
4199 * page lock is already held. To safely acquire the range lock the
4200 * page lock must be dropped. This creates a window where another
4201 * process could truncate, invalidate, dirty, or write out the page.
4203 * Therefore, after successfully reacquiring the range and page locks
4204 * the current page state is checked. In the common case everything
4205 * will be as is expected and it can be written out. However, if
4206 * the page state has changed it must be handled accordingly.
4208 mapping
= pp
->mapping
;
4209 redirty_page_for_writepage(wbc
, pp
);
4212 rl
= zfs_range_lock(&zp
->z_range_lock
, pgoff
, pglen
, RL_WRITER
);
4215 /* Page mapping changed or it was no longer dirty, we're done */
4216 if (unlikely((mapping
!= pp
->mapping
) || !PageDirty(pp
))) {
4218 zfs_range_unlock(rl
);
4223 /* Another process started write block if required */
4224 if (PageWriteback(pp
)) {
4226 zfs_range_unlock(rl
);
4228 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
4229 wait_on_page_writeback(pp
);
4235 /* Clear the dirty flag the required locks are held */
4236 if (!clear_page_dirty_for_io(pp
)) {
4238 zfs_range_unlock(rl
);
4244 * Counterpart for redirty_page_for_writepage() above. This page
4245 * was in fact not skipped and should not be counted as if it were.
4247 wbc
->pages_skipped
--;
4248 set_page_writeback(pp
);
4251 tx
= dmu_tx_create(zsb
->z_os
);
4252 dmu_tx_hold_write(tx
, zp
->z_id
, pgoff
, pglen
);
4253 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4254 zfs_sa_upgrade_txholds(tx
, zp
);
4256 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4258 if (err
== ERESTART
)
4262 __set_page_dirty_nobuffers(pp
);
4264 end_page_writeback(pp
);
4265 zfs_range_unlock(rl
);
4271 ASSERT3U(pglen
, <=, PAGE_SIZE
);
4272 dmu_write(zsb
->z_os
, zp
->z_id
, pgoff
, pglen
, va
, tx
);
4275 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
4276 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
4277 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_FLAGS(zsb
), NULL
, &zp
->z_pflags
, 8);
4279 /* Preserve the mtime and ctime provided by the inode */
4280 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4281 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4282 zp
->z_atime_dirty
= 0;
4285 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4287 zfs_log_write(zsb
->z_log
, tx
, TX_WRITE
, zp
, pgoff
, pglen
, 0,
4288 zfs_putpage_commit_cb
, pp
);
4291 zfs_range_unlock(rl
);
4293 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4295 * Note that this is rarely called under writepages(), because
4296 * writepages() normally handles the entire commit for
4297 * performance reasons.
4299 zil_commit(zsb
->z_log
, zp
->z_id
);
4307 * Update the system attributes when the inode has been dirtied. For the
4308 * moment we only update the mode, atime, mtime, and ctime.
4311 zfs_dirty_inode(struct inode
*ip
, int flags
)
4313 znode_t
*zp
= ITOZ(ip
);
4314 zfs_sb_t
*zsb
= ITOZSB(ip
);
4316 uint64_t mode
, atime
[2], mtime
[2], ctime
[2];
4317 sa_bulk_attr_t bulk
[4];
4321 if (zfs_is_readonly(zsb
) || dmu_objset_is_snapshot(zsb
->z_os
))
4329 * This is the lazytime semantic indroduced in Linux 4.0
4330 * This flag will only be called from update_time when lazytime is set.
4331 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4332 * Fortunately mtime and ctime are managed within ZFS itself, so we
4333 * only need to dirty atime.
4335 if (flags
== I_DIRTY_TIME
) {
4336 zp
->z_atime_dirty
= 1;
4341 tx
= dmu_tx_create(zsb
->z_os
);
4343 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4344 zfs_sa_upgrade_txholds(tx
, zp
);
4346 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4352 mutex_enter(&zp
->z_lock
);
4353 zp
->z_atime_dirty
= 0;
4355 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MODE(zsb
), NULL
, &mode
, 8);
4356 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_ATIME(zsb
), NULL
, &atime
, 16);
4357 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zsb
), NULL
, &mtime
, 16);
4358 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zsb
), NULL
, &ctime
, 16);
4360 /* Preserve the mode, mtime and ctime provided by the inode */
4361 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4362 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4363 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4368 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4369 mutex_exit(&zp
->z_lock
);
4376 EXPORT_SYMBOL(zfs_dirty_inode
);
4380 zfs_inactive(struct inode
*ip
)
4382 znode_t
*zp
= ITOZ(ip
);
4383 zfs_sb_t
*zsb
= ITOZSB(ip
);
4386 int need_unlock
= 0;
4388 /* Only read lock if we haven't already write locked, e.g. rollback */
4389 if (!RW_WRITE_HELD(&zsb
->z_teardown_inactive_lock
)) {
4391 rw_enter(&zsb
->z_teardown_inactive_lock
, RW_READER
);
4393 if (zp
->z_sa_hdl
== NULL
) {
4395 rw_exit(&zsb
->z_teardown_inactive_lock
);
4399 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4400 dmu_tx_t
*tx
= dmu_tx_create(zsb
->z_os
);
4402 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4403 zfs_sa_upgrade_txholds(tx
, zp
);
4404 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4408 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4409 mutex_enter(&zp
->z_lock
);
4410 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zsb
),
4411 (void *)&atime
, sizeof (atime
), tx
);
4412 zp
->z_atime_dirty
= 0;
4413 mutex_exit(&zp
->z_lock
);
4420 rw_exit(&zsb
->z_teardown_inactive_lock
);
4422 EXPORT_SYMBOL(zfs_inactive
);
4425 * Bounds-check the seek operation.
4427 * IN: ip - inode seeking within
4428 * ooff - old file offset
4429 * noffp - pointer to new file offset
4430 * ct - caller context
4432 * RETURN: 0 if success
4433 * EINVAL if new offset invalid
4437 zfs_seek(struct inode
*ip
, offset_t ooff
, offset_t
*noffp
)
4439 if (S_ISDIR(ip
->i_mode
))
4441 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4443 EXPORT_SYMBOL(zfs_seek
);
4446 * Fill pages with data from the disk.
4449 zfs_fillpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4451 znode_t
*zp
= ITOZ(ip
);
4452 zfs_sb_t
*zsb
= ITOZSB(ip
);
4454 struct page
*cur_pp
;
4455 u_offset_t io_off
, total
;
4462 io_len
= nr_pages
<< PAGE_SHIFT
;
4463 i_size
= i_size_read(ip
);
4464 io_off
= page_offset(pl
[0]);
4466 if (io_off
+ io_len
> i_size
)
4467 io_len
= i_size
- io_off
;
4470 * Iterate over list of pages and read each page individually.
4473 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4476 cur_pp
= pl
[page_idx
++];
4478 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4482 /* convert checksum errors into IO errors */
4484 err
= SET_ERROR(EIO
);
4493 * Uses zfs_fillpage to read data from the file and fill the pages.
4495 * IN: ip - inode of file to get data from.
4496 * pl - list of pages to read
4497 * nr_pages - number of pages to read
4499 * RETURN: 0 on success, error code on failure.
4502 * vp - atime updated
4506 zfs_getpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4508 znode_t
*zp
= ITOZ(ip
);
4509 zfs_sb_t
*zsb
= ITOZSB(ip
);
4518 err
= zfs_fillpage(ip
, pl
, nr_pages
);
4523 EXPORT_SYMBOL(zfs_getpage
);
4526 * Check ZFS specific permissions to memory map a section of a file.
4528 * IN: ip - inode of the file to mmap
4530 * addrp - start address in memory region
4531 * len - length of memory region
4532 * vm_flags- address flags
4534 * RETURN: 0 if success
4535 * error code if failure
4539 zfs_map(struct inode
*ip
, offset_t off
, caddr_t
*addrp
, size_t len
,
4540 unsigned long vm_flags
)
4542 znode_t
*zp
= ITOZ(ip
);
4543 zfs_sb_t
*zsb
= ITOZSB(ip
);
4548 if ((vm_flags
& VM_WRITE
) && (zp
->z_pflags
&
4549 (ZFS_IMMUTABLE
| ZFS_READONLY
| ZFS_APPENDONLY
))) {
4551 return (SET_ERROR(EPERM
));
4554 if ((vm_flags
& (VM_READ
| VM_EXEC
)) &&
4555 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4557 return (SET_ERROR(EACCES
));
4560 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4562 return (SET_ERROR(ENXIO
));
4568 EXPORT_SYMBOL(zfs_map
);
4571 * convoff - converts the given data (start, whence) to the
4575 convoff(struct inode
*ip
, flock64_t
*lckdat
, int whence
, offset_t offset
)
4580 if ((lckdat
->l_whence
== 2) || (whence
== 2)) {
4581 if ((error
= zfs_getattr(ip
, &vap
, 0, CRED()) != 0))
4585 switch (lckdat
->l_whence
) {
4587 lckdat
->l_start
+= offset
;
4590 lckdat
->l_start
+= vap
.va_size
;
4595 return (SET_ERROR(EINVAL
));
4598 if (lckdat
->l_start
< 0)
4599 return (SET_ERROR(EINVAL
));
4603 lckdat
->l_start
-= offset
;
4606 lckdat
->l_start
-= vap
.va_size
;
4611 return (SET_ERROR(EINVAL
));
4614 lckdat
->l_whence
= (short)whence
;
4619 * Free or allocate space in a file. Currently, this function only
4620 * supports the `F_FREESP' command. However, this command is somewhat
4621 * misnamed, as its functionality includes the ability to allocate as
4622 * well as free space.
4624 * IN: ip - inode of file to free data in.
4625 * cmd - action to take (only F_FREESP supported).
4626 * bfp - section of file to free/alloc.
4627 * flag - current file open mode flags.
4628 * offset - current file offset.
4629 * cr - credentials of caller [UNUSED].
4631 * RETURN: 0 on success, error code on failure.
4634 * ip - ctime|mtime updated
4638 zfs_space(struct inode
*ip
, int cmd
, flock64_t
*bfp
, int flag
,
4639 offset_t offset
, cred_t
*cr
)
4641 znode_t
*zp
= ITOZ(ip
);
4642 zfs_sb_t
*zsb
= ITOZSB(ip
);
4649 if (cmd
!= F_FREESP
) {
4651 return (SET_ERROR(EINVAL
));
4655 * Callers might not be able to detect properly that we are read-only,
4656 * so check it explicitly here.
4658 if (zfs_is_readonly(zsb
)) {
4660 return (SET_ERROR(EROFS
));
4663 if ((error
= convoff(ip
, bfp
, 0, offset
))) {
4668 if (bfp
->l_len
< 0) {
4670 return (SET_ERROR(EINVAL
));
4674 * Permissions aren't checked on Solaris because on this OS
4675 * zfs_space() can only be called with an opened file handle.
4676 * On Linux we can get here through truncate_range() which
4677 * operates directly on inodes, so we need to check access rights.
4679 if ((error
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
))) {
4685 len
= bfp
->l_len
; /* 0 means from off to end of file */
4687 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4692 EXPORT_SYMBOL(zfs_space
);
4696 zfs_fid(struct inode
*ip
, fid_t
*fidp
)
4698 znode_t
*zp
= ITOZ(ip
);
4699 zfs_sb_t
*zsb
= ITOZSB(ip
);
4702 uint64_t object
= zp
->z_id
;
4709 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zsb
),
4710 &gen64
, sizeof (uint64_t))) != 0) {
4715 gen
= (uint32_t)gen64
;
4717 size
= (zsb
->z_parent
!= zsb
) ? LONG_FID_LEN
: SHORT_FID_LEN
;
4718 if (fidp
->fid_len
< size
) {
4719 fidp
->fid_len
= size
;
4721 return (SET_ERROR(ENOSPC
));
4724 zfid
= (zfid_short_t
*)fidp
;
4726 zfid
->zf_len
= size
;
4728 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4729 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4731 /* Must have a non-zero generation number to distinguish from .zfs */
4734 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4735 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4737 if (size
== LONG_FID_LEN
) {
4738 uint64_t objsetid
= dmu_objset_id(zsb
->z_os
);
4741 zlfid
= (zfid_long_t
*)fidp
;
4743 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
4744 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
4746 /* XXX - this should be the generation number for the objset */
4747 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
4748 zlfid
->zf_setgen
[i
] = 0;
4754 EXPORT_SYMBOL(zfs_fid
);
4758 zfs_getsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4760 znode_t
*zp
= ITOZ(ip
);
4761 zfs_sb_t
*zsb
= ITOZSB(ip
);
4763 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4767 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
4772 EXPORT_SYMBOL(zfs_getsecattr
);
4776 zfs_setsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
4778 znode_t
*zp
= ITOZ(ip
);
4779 zfs_sb_t
*zsb
= ITOZSB(ip
);
4781 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4782 zilog_t
*zilog
= zsb
->z_log
;
4787 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
4789 if (zsb
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4790 zil_commit(zilog
, 0);
4795 EXPORT_SYMBOL(zfs_setsecattr
);
4797 #ifdef HAVE_UIO_ZEROCOPY
4799 * Tunable, both must be a power of 2.
4801 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
4802 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
4803 * an arcbuf for a partial block read
4805 int zcr_blksz_min
= (1 << 10); /* 1K */
4806 int zcr_blksz_max
= (1 << 17); /* 128K */
4810 zfs_reqzcbuf(struct inode
*ip
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
)
4812 znode_t
*zp
= ITOZ(ip
);
4813 zfs_sb_t
*zsb
= ITOZSB(ip
);
4814 int max_blksz
= zsb
->z_max_blksz
;
4815 uio_t
*uio
= &xuio
->xu_uio
;
4816 ssize_t size
= uio
->uio_resid
;
4817 offset_t offset
= uio
->uio_loffset
;
4822 int preamble
, postamble
;
4824 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
4825 return (SET_ERROR(EINVAL
));
4832 * Loan out an arc_buf for write if write size is bigger than
4833 * max_blksz, and the file's block size is also max_blksz.
4836 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
4838 return (SET_ERROR(EINVAL
));
4841 * Caller requests buffers for write before knowing where the
4842 * write offset might be (e.g. NFS TCP write).
4847 preamble
= P2PHASE(offset
, blksz
);
4849 preamble
= blksz
- preamble
;
4854 postamble
= P2PHASE(size
, blksz
);
4857 fullblk
= size
/ blksz
;
4858 (void) dmu_xuio_init(xuio
,
4859 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
4862 * Have to fix iov base/len for partial buffers. They
4863 * currently represent full arc_buf's.
4866 /* data begins in the middle of the arc_buf */
4867 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4870 (void) dmu_xuio_add(xuio
, abuf
,
4871 blksz
- preamble
, preamble
);
4874 for (i
= 0; i
< fullblk
; i
++) {
4875 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4878 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
4882 /* data ends in the middle of the arc_buf */
4883 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
4886 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
4891 * Loan out an arc_buf for read if the read size is larger than
4892 * the current file block size. Block alignment is not
4893 * considered. Partial arc_buf will be loaned out for read.
4895 blksz
= zp
->z_blksz
;
4896 if (blksz
< zcr_blksz_min
)
4897 blksz
= zcr_blksz_min
;
4898 if (blksz
> zcr_blksz_max
)
4899 blksz
= zcr_blksz_max
;
4900 /* avoid potential complexity of dealing with it */
4901 if (blksz
> max_blksz
) {
4903 return (SET_ERROR(EINVAL
));
4906 maxsize
= zp
->z_size
- uio
->uio_loffset
;
4912 return (SET_ERROR(EINVAL
));
4917 return (SET_ERROR(EINVAL
));
4920 uio
->uio_extflg
= UIO_XUIO
;
4921 XUIO_XUZC_RW(xuio
) = ioflag
;
4928 zfs_retzcbuf(struct inode
*ip
, xuio_t
*xuio
, cred_t
*cr
)
4932 int ioflag
= XUIO_XUZC_RW(xuio
);
4934 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
4936 i
= dmu_xuio_cnt(xuio
);
4938 abuf
= dmu_xuio_arcbuf(xuio
, i
);
4940 * if abuf == NULL, it must be a write buffer
4941 * that has been returned in zfs_write().
4944 dmu_return_arcbuf(abuf
);
4945 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
4948 dmu_xuio_fini(xuio
);
4951 #endif /* HAVE_UIO_ZEROCOPY */
4953 #if defined(_KERNEL) && defined(HAVE_SPL)
4955 module_param(zfs_delete_blocks
, ulong
, 0644);
4956 MODULE_PARM_DESC(zfs_delete_blocks
, "Delete files larger than N blocks async");
4957 module_param(zfs_read_chunk_size
, long, 0644);
4958 MODULE_PARM_DESC(zfs_read_chunk_size
, "Bytes to read per chunk");