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, 2018 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/sysmacros.h>
41 #include <sys/taskq.h>
43 #include <sys/vmsystm.h>
44 #include <sys/atomic.h>
45 #include <sys/pathname.h>
46 #include <sys/cmn_err.h>
47 #include <sys/errno.h>
48 #include <sys/zfs_dir.h>
49 #include <sys/zfs_acl.h>
50 #include <sys/zfs_ioctl.h>
51 #include <sys/fs/zfs.h>
53 #include <sys/dmu_objset.h>
59 #include <sys/policy.h>
60 #include <sys/sunddi.h>
63 #include <sys/zfs_ctldir.h>
64 #include <sys/zfs_fuid.h>
65 #include <sys/zfs_sa.h>
66 #include <sys/zfs_vnops.h>
67 #include <sys/zfs_rlock.h>
71 #include <sys/sa_impl.h>
76 * Each vnode op performs some logical unit of work. To do this, the ZPL must
77 * properly lock its in-core state, create a DMU transaction, do the work,
78 * record this work in the intent log (ZIL), commit the DMU transaction,
79 * and wait for the intent log to commit if it is a synchronous operation.
80 * Moreover, the vnode ops must work in both normal and log replay context.
81 * The ordering of events is important to avoid deadlocks and references
82 * to freed memory. The example below illustrates the following Big Rules:
84 * (1) A check must be made in each zfs thread for a mounted file system.
85 * This is done avoiding races using ZFS_ENTER(zfsvfs).
86 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
87 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
88 * can return EIO from the calling function.
90 * (2) iput() should always be the last thing except for zil_commit()
91 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
92 * First, if it's the last reference, the vnode/znode
93 * can be freed, so the zp may point to freed memory. Second, the last
94 * reference will call zfs_zinactive(), which may induce a lot of work --
95 * pushing cached pages (which acquires range locks) and syncing out
96 * cached atime changes. Third, zfs_zinactive() may require a new tx,
97 * which could deadlock the system if you were already holding one.
98 * If you must call iput() within a tx then use zfs_iput_async().
100 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
101 * as they can span dmu_tx_assign() calls.
103 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
104 * dmu_tx_assign(). This is critical because we don't want to block
105 * while holding locks.
107 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
108 * reduces lock contention and CPU usage when we must wait (note that if
109 * throughput is constrained by the storage, nearly every transaction
112 * Note, in particular, that if a lock is sometimes acquired before
113 * the tx assigns, and sometimes after (e.g. z_lock), then failing
114 * to use a non-blocking assign can deadlock the system. The scenario:
116 * Thread A has grabbed a lock before calling dmu_tx_assign().
117 * Thread B is in an already-assigned tx, and blocks for this lock.
118 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
119 * forever, because the previous txg can't quiesce until B's tx commits.
121 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
122 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
123 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT,
124 * to indicate that this operation has already called dmu_tx_wait().
125 * This will ensure that we don't retry forever, waiting a short bit
128 * (5) If the operation succeeded, generate the intent log entry for it
129 * before dropping locks. This ensures that the ordering of events
130 * in the intent log matches the order in which they actually occurred.
131 * During ZIL replay the zfs_log_* functions will update the sequence
132 * number to indicate the zil transaction has replayed.
134 * (6) At the end of each vnode op, the DMU tx must always commit,
135 * regardless of whether there were any errors.
137 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
138 * to ensure that synchronous semantics are provided when necessary.
140 * In general, this is how things should be ordered in each vnode op:
142 * ZFS_ENTER(zfsvfs); // exit if unmounted
144 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab())
145 * rw_enter(...); // grab any other locks you need
146 * tx = dmu_tx_create(...); // get DMU tx
147 * dmu_tx_hold_*(); // hold each object you might modify
148 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT);
150 * rw_exit(...); // drop locks
151 * zfs_dirent_unlock(dl); // unlock directory entry
152 * iput(...); // release held vnodes
153 * if (error == ERESTART) {
159 * dmu_tx_abort(tx); // abort DMU tx
160 * ZFS_EXIT(zfsvfs); // finished in zfs
161 * return (error); // really out of space
163 * error = do_real_work(); // do whatever this VOP does
165 * zfs_log_*(...); // on success, make ZIL entry
166 * dmu_tx_commit(tx); // commit DMU tx -- error or not
167 * rw_exit(...); // drop locks
168 * zfs_dirent_unlock(dl); // unlock directory entry
169 * iput(...); // release held vnodes
170 * zil_commit(zilog, foid); // synchronous when necessary
171 * ZFS_EXIT(zfsvfs); // finished in zfs
172 * return (error); // done, report error
176 * Virus scanning is unsupported. It would be possible to add a hook
177 * here to performance the required virus scan. This could be done
178 * entirely in the kernel or potentially as an update to invoke a
182 zfs_vscan(struct inode
*ip
, cred_t
*cr
, int async
)
189 zfs_open(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
191 znode_t
*zp
= ITOZ(ip
);
192 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
197 /* Honor ZFS_APPENDONLY file attribute */
198 if ((mode
& FMODE_WRITE
) && (zp
->z_pflags
& ZFS_APPENDONLY
) &&
199 ((flag
& O_APPEND
) == 0)) {
201 return (SET_ERROR(EPERM
));
204 /* Virus scan eligible files on open */
205 if (!zfs_has_ctldir(zp
) && zfsvfs
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
206 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0) {
207 if (zfs_vscan(ip
, cr
, 0) != 0) {
209 return (SET_ERROR(EACCES
));
213 /* Keep a count of the synchronous opens in the znode */
215 atomic_inc_32(&zp
->z_sync_cnt
);
223 zfs_close(struct inode
*ip
, int flag
, cred_t
*cr
)
225 znode_t
*zp
= ITOZ(ip
);
226 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
231 /* Decrement the synchronous opens in the znode */
233 atomic_dec_32(&zp
->z_sync_cnt
);
235 if (!zfs_has_ctldir(zp
) && zfsvfs
->z_vscan
&& S_ISREG(ip
->i_mode
) &&
236 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0)
237 VERIFY(zfs_vscan(ip
, cr
, 1) == 0);
243 #if defined(SEEK_HOLE) && defined(SEEK_DATA)
245 * Lseek support for finding holes (cmd == SEEK_HOLE) and
246 * data (cmd == SEEK_DATA). "off" is an in/out parameter.
249 zfs_holey_common(struct inode
*ip
, int cmd
, loff_t
*off
)
251 znode_t
*zp
= ITOZ(ip
);
252 uint64_t noff
= (uint64_t)*off
; /* new offset */
257 file_sz
= zp
->z_size
;
258 if (noff
>= file_sz
) {
259 return (SET_ERROR(ENXIO
));
262 if (cmd
== SEEK_HOLE
)
267 error
= dmu_offset_next(ZTOZSB(zp
)->z_os
, zp
->z_id
, hole
, &noff
);
270 return (SET_ERROR(ENXIO
));
272 /* file was dirty, so fall back to using generic logic */
273 if (error
== EBUSY
) {
281 * We could find a hole that begins after the logical end-of-file,
282 * because dmu_offset_next() only works on whole blocks. If the
283 * EOF falls mid-block, then indicate that the "virtual hole"
284 * at the end of the file begins at the logical EOF, rather than
285 * at the end of the last block.
287 if (noff
> file_sz
) {
299 zfs_holey(struct inode
*ip
, int cmd
, loff_t
*off
)
301 znode_t
*zp
= ITOZ(ip
);
302 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
308 error
= zfs_holey_common(ip
, cmd
, off
);
313 #endif /* SEEK_HOLE && SEEK_DATA */
317 * When a file is memory mapped, we must keep the IO data synchronized
318 * between the DMU cache and the memory mapped pages. What this means:
320 * On Write: If we find a memory mapped page, we write to *both*
321 * the page and the dmu buffer.
324 update_pages(struct inode
*ip
, int64_t start
, int len
,
325 objset_t
*os
, uint64_t oid
)
327 struct address_space
*mp
= ip
->i_mapping
;
333 off
= start
& (PAGE_SIZE
-1);
334 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
335 nbytes
= MIN(PAGE_SIZE
- off
, len
);
337 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
339 if (mapping_writably_mapped(mp
))
340 flush_dcache_page(pp
);
343 (void) dmu_read(os
, oid
, start
+off
, nbytes
, pb
+off
,
347 if (mapping_writably_mapped(mp
))
348 flush_dcache_page(pp
);
350 mark_page_accessed(pp
);
363 * When a file is memory mapped, we must keep the IO data synchronized
364 * between the DMU cache and the memory mapped pages. What this means:
366 * On Read: We "read" preferentially from memory mapped pages,
367 * else we default from the dmu buffer.
369 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
370 * the file is memory mapped.
373 mappedread(struct inode
*ip
, int nbytes
, uio_t
*uio
)
375 struct address_space
*mp
= ip
->i_mapping
;
377 znode_t
*zp
= ITOZ(ip
);
384 start
= uio
->uio_loffset
;
385 off
= start
& (PAGE_SIZE
-1);
386 for (start
&= PAGE_MASK
; len
> 0; start
+= PAGE_SIZE
) {
387 bytes
= MIN(PAGE_SIZE
- off
, len
);
389 pp
= find_lock_page(mp
, start
>> PAGE_SHIFT
);
391 ASSERT(PageUptodate(pp
));
395 error
= uiomove(pb
+ off
, bytes
, UIO_READ
, uio
);
398 if (mapping_writably_mapped(mp
))
399 flush_dcache_page(pp
);
401 mark_page_accessed(pp
);
404 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
417 unsigned long zfs_read_chunk_size
= 1024 * 1024; /* Tunable */
418 unsigned long zfs_delete_blocks
= DMU_MAX_DELETEBLKCNT
;
421 * Read bytes from specified file into supplied buffer.
423 * IN: ip - inode of file to be read from.
424 * uio - structure supplying read location, range info,
426 * ioflag - FSYNC flags; used to provide FRSYNC semantics.
427 * O_DIRECT flag; used to bypass page cache.
428 * cr - credentials of caller.
430 * OUT: uio - updated offset and range, buffer filled.
432 * RETURN: 0 on success, error code on failure.
435 * inode - atime updated if byte count > 0
439 zfs_read(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
443 znode_t
*zp
= ITOZ(ip
);
444 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
448 if (zp
->z_pflags
& ZFS_AV_QUARANTINED
) {
450 return (SET_ERROR(EACCES
));
454 * Validate file offset
456 if (uio
->uio_loffset
< (offset_t
)0) {
458 return (SET_ERROR(EINVAL
));
462 * Fasttrack empty reads
464 if (uio
->uio_resid
== 0) {
470 * If we're in FRSYNC mode, sync out this znode before reading it.
471 * Only do this for non-snapshots.
474 (ioflag
& FRSYNC
|| zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
))
475 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
478 * Lock the range against changes.
480 locked_range_t
*lr
= rangelock_enter(&zp
->z_rangelock
,
481 uio
->uio_loffset
, uio
->uio_resid
, RL_READER
);
484 * If we are reading past end-of-file we can skip
485 * to the end; but we might still need to set atime.
487 if (uio
->uio_loffset
>= zp
->z_size
) {
492 ASSERT(uio
->uio_loffset
< zp
->z_size
);
493 ssize_t n
= MIN(uio
->uio_resid
, zp
->z_size
- uio
->uio_loffset
);
494 ssize_t start_resid
= n
;
496 #ifdef HAVE_UIO_ZEROCOPY
498 if ((uio
->uio_extflg
== UIO_XUIO
) &&
499 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
)) {
501 int blksz
= zp
->z_blksz
;
502 uint64_t offset
= uio
->uio_loffset
;
504 xuio
= (xuio_t
*)uio
;
506 nblk
= (P2ROUNDUP(offset
+ n
, blksz
) - P2ALIGN(offset
,
509 ASSERT(offset
+ n
<= blksz
);
512 (void) dmu_xuio_init(xuio
, nblk
);
514 if (vn_has_cached_data(ip
)) {
516 * For simplicity, we always allocate a full buffer
517 * even if we only expect to read a portion of a block.
519 while (--nblk
>= 0) {
520 (void) dmu_xuio_add(xuio
,
521 dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
526 #endif /* HAVE_UIO_ZEROCOPY */
529 ssize_t nbytes
= MIN(n
, zfs_read_chunk_size
-
530 P2PHASE(uio
->uio_loffset
, zfs_read_chunk_size
));
532 if (zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
)) {
533 error
= mappedread(ip
, nbytes
, uio
);
535 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
540 /* convert checksum errors into IO errors */
542 error
= SET_ERROR(EIO
);
549 int64_t nread
= start_resid
- n
;
550 dataset_kstats_update_read_kstats(&zfsvfs
->z_kstat
, nread
);
551 task_io_account_read(nread
);
560 * Write the bytes to a file.
562 * IN: ip - inode of file to be written to.
563 * uio - structure supplying write location, range info,
565 * ioflag - FAPPEND flag set if in append mode.
566 * O_DIRECT flag; used to bypass page cache.
567 * cr - credentials of caller.
569 * OUT: uio - updated offset and range.
571 * RETURN: 0 if success
572 * error code if failure
575 * ip - ctime|mtime updated if byte count > 0
580 zfs_write(struct inode
*ip
, uio_t
*uio
, int ioflag
, cred_t
*cr
)
583 ssize_t start_resid
= uio
->uio_resid
;
586 * Fasttrack empty write
588 ssize_t n
= start_resid
;
592 rlim64_t limit
= uio
->uio_limit
;
593 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
596 znode_t
*zp
= ITOZ(ip
);
597 zfsvfs_t
*zfsvfs
= ZTOZSB(zp
);
601 sa_bulk_attr_t bulk
[4];
603 uint64_t mtime
[2], ctime
[2];
604 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
605 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
606 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
), NULL
,
608 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
612 * Callers might not be able to detect properly that we are read-only,
613 * so check it explicitly here.
615 if (zfs_is_readonly(zfsvfs
)) {
617 return (SET_ERROR(EROFS
));
621 * If immutable or not appending then return EPERM
623 if ((zp
->z_pflags
& (ZFS_IMMUTABLE
| ZFS_READONLY
)) ||
624 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
625 (uio
->uio_loffset
< zp
->z_size
))) {
627 return (SET_ERROR(EPERM
));
631 * Validate file offset
633 offset_t woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
636 return (SET_ERROR(EINVAL
));
639 int max_blksz
= zfsvfs
->z_max_blksz
;
643 * Pre-fault the pages to ensure slow (eg NFS) pages
645 * Skip this if uio contains loaned arc_buf.
647 #ifdef HAVE_UIO_ZEROCOPY
648 if ((uio
->uio_extflg
== UIO_XUIO
) &&
649 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
))
650 xuio
= (xuio_t
*)uio
;
653 if (uio_prefaultpages(MIN(n
, max_blksz
), uio
)) {
655 return (SET_ERROR(EFAULT
));
659 * If in append mode, set the io offset pointer to eof.
662 if (ioflag
& FAPPEND
) {
664 * Obtain an appending range lock to guarantee file append
665 * semantics. We reset the write offset once we have the lock.
667 lr
= rangelock_enter(&zp
->z_rangelock
, 0, n
, RL_APPEND
);
668 woff
= lr
->lr_offset
;
669 if (lr
->lr_length
== UINT64_MAX
) {
671 * We overlocked the file because this write will cause
672 * the file block size to increase.
673 * Note that zp_size cannot change with this lock held.
677 uio
->uio_loffset
= woff
;
680 * Note that if the file block size will change as a result of
681 * this write, then this range lock will lock the entire file
682 * so that we can re-write the block safely.
684 lr
= rangelock_enter(&zp
->z_rangelock
, woff
, n
, RL_WRITER
);
690 return (SET_ERROR(EFBIG
));
693 if ((woff
+ n
) > limit
|| woff
> (limit
- n
))
696 /* Will this write extend the file length? */
697 int write_eof
= (woff
+ n
> zp
->z_size
);
699 uint64_t end_size
= MAX(zp
->z_size
, woff
+ n
);
700 zilog_t
*zilog
= zfsvfs
->z_log
;
701 #ifdef HAVE_UIO_ZEROCOPY
703 const iovec_t
*iovp
= uio
->uio_iov
;
704 ASSERTV(int iovcnt
= uio
->uio_iovcnt
);
709 * Write the file in reasonable size chunks. Each chunk is written
710 * in a separate transaction; this keeps the intent log records small
711 * and allows us to do more fine-grained space accounting.
714 woff
= uio
->uio_loffset
;
716 if (zfs_id_overblockquota(zfsvfs
, DMU_USERUSED_OBJECT
,
717 KUID_TO_SUID(ip
->i_uid
)) ||
718 zfs_id_overblockquota(zfsvfs
, DMU_GROUPUSED_OBJECT
,
719 KGID_TO_SGID(ip
->i_gid
)) ||
720 (zp
->z_projid
!= ZFS_DEFAULT_PROJID
&&
721 zfs_id_overblockquota(zfsvfs
, DMU_PROJECTUSED_OBJECT
,
723 error
= SET_ERROR(EDQUOT
);
727 arc_buf_t
*abuf
= NULL
;
728 const iovec_t
*aiov
= NULL
;
730 #ifdef HAVE_UIO_ZEROCOPY
731 ASSERT(i_iov
< iovcnt
);
732 ASSERT3U(uio
->uio_segflg
, !=, UIO_BVEC
);
734 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
735 dmu_xuio_clear(xuio
, i_iov
);
736 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
737 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
738 aiov
->iov_len
== arc_buf_size(abuf
)));
741 } else if (n
>= max_blksz
&& woff
>= zp
->z_size
&&
742 P2PHASE(woff
, max_blksz
) == 0 &&
743 zp
->z_blksz
== max_blksz
) {
745 * This write covers a full block. "Borrow" a buffer
746 * from the dmu so that we can fill it before we enter
747 * a transaction. This avoids the possibility of
748 * holding up the transaction if the data copy hangs
749 * up on a pagefault (e.g., from an NFS server mapping).
753 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
755 ASSERT(abuf
!= NULL
);
756 ASSERT(arc_buf_size(abuf
) == max_blksz
);
757 if ((error
= uiocopy(abuf
->b_data
, max_blksz
,
758 UIO_WRITE
, uio
, &cbytes
))) {
759 dmu_return_arcbuf(abuf
);
762 ASSERT(cbytes
== max_blksz
);
766 * Start a transaction.
768 dmu_tx_t
*tx
= dmu_tx_create(zfsvfs
->z_os
);
769 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
770 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
771 zfs_sa_upgrade_txholds(tx
, zp
);
772 error
= dmu_tx_assign(tx
, TXG_WAIT
);
776 dmu_return_arcbuf(abuf
);
781 * If rangelock_enter() over-locked we grow the blocksize
782 * and then reduce the lock range. This will only happen
783 * on the first iteration since rangelock_reduce() will
784 * shrink down lr_length to the appropriate size.
786 if (lr
->lr_length
== UINT64_MAX
) {
789 if (zp
->z_blksz
> max_blksz
) {
791 * File's blocksize is already larger than the
792 * "recordsize" property. Only let it grow to
793 * the next power of 2.
795 ASSERT(!ISP2(zp
->z_blksz
));
796 new_blksz
= MIN(end_size
,
797 1 << highbit64(zp
->z_blksz
));
799 new_blksz
= MIN(end_size
, max_blksz
);
801 zfs_grow_blocksize(zp
, new_blksz
, tx
);
802 rangelock_reduce(lr
, woff
, n
);
806 * XXX - should we really limit each write to z_max_blksz?
807 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
809 ssize_t nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
813 tx_bytes
= uio
->uio_resid
;
814 uio
->uio_fault_disable
= B_TRUE
;
815 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
817 if (error
== EFAULT
) {
819 if (uio_prefaultpages(MIN(n
, max_blksz
), uio
)) {
823 } else if (error
!= 0) {
827 tx_bytes
-= uio
->uio_resid
;
830 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
832 * If this is not a full block write, but we are
833 * extending the file past EOF and this data starts
834 * block-aligned, use assign_arcbuf(). Otherwise,
835 * write via dmu_write().
837 if (tx_bytes
< max_blksz
&& (!write_eof
||
838 aiov
->iov_base
!= abuf
->b_data
)) {
840 dmu_write(zfsvfs
->z_os
, zp
->z_id
, woff
,
841 /* cppcheck-suppress nullPointer */
842 aiov
->iov_len
, aiov
->iov_base
, tx
);
843 dmu_return_arcbuf(abuf
);
844 xuio_stat_wbuf_copied();
846 ASSERT(xuio
|| tx_bytes
== max_blksz
);
847 error
= dmu_assign_arcbuf_by_dbuf(
848 sa_get_db(zp
->z_sa_hdl
), woff
, abuf
, tx
);
850 dmu_return_arcbuf(abuf
);
855 ASSERT(tx_bytes
<= uio
->uio_resid
);
856 uioskip(uio
, tx_bytes
);
858 if (tx_bytes
&& zp
->z_is_mapped
&& !(ioflag
& O_DIRECT
)) {
859 update_pages(ip
, woff
,
860 tx_bytes
, zfsvfs
->z_os
, zp
->z_id
);
864 * If we made no progress, we're done. If we made even
865 * partial progress, update the znode and ZIL accordingly.
868 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
869 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
876 * Clear Set-UID/Set-GID bits on successful write if not
877 * privileged and at least one of the execute bits is set.
879 * It would be nice to to this after all writes have
880 * been done, but that would still expose the ISUID/ISGID
881 * to another app after the partial write is committed.
883 * Note: we don't call zfs_fuid_map_id() here because
884 * user 0 is not an ephemeral uid.
886 mutex_enter(&zp
->z_acl_lock
);
887 uint32_t uid
= KUID_TO_SUID(ip
->i_uid
);
888 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
889 (S_IXUSR
>> 6))) != 0 &&
890 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
891 secpolicy_vnode_setid_retain(cr
,
892 ((zp
->z_mode
& S_ISUID
) != 0 && uid
== 0)) != 0) {
894 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
895 ip
->i_mode
= newmode
= zp
->z_mode
;
896 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zfsvfs
),
897 (void *)&newmode
, sizeof (uint64_t), tx
);
899 mutex_exit(&zp
->z_acl_lock
);
901 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
);
904 * Update the file size (zp_size) if it has changed;
905 * account for possible concurrent updates.
907 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
908 (void) atomic_cas_64(&zp
->z_size
, end_size
,
913 * If we are replaying and eof is non zero then force
914 * the file size to the specified eof. Note, there's no
915 * concurrency during replay.
917 if (zfsvfs
->z_replay
&& zfsvfs
->z_replay_eof
!= 0)
918 zp
->z_size
= zfsvfs
->z_replay_eof
;
920 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
922 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
,
928 ASSERT(tx_bytes
== nbytes
);
931 if (!xuio
&& n
> 0) {
932 if (uio_prefaultpages(MIN(n
, max_blksz
), uio
)) {
939 zfs_inode_update(zp
);
943 * If we're in replay mode, or we made no progress, return error.
944 * Otherwise, it's at least a partial write, so it's successful.
946 if (zfsvfs
->z_replay
|| uio
->uio_resid
== start_resid
) {
951 if (ioflag
& (FSYNC
| FDSYNC
) ||
952 zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
953 zil_commit(zilog
, zp
->z_id
);
955 int64_t nwritten
= start_resid
- uio
->uio_resid
;
956 dataset_kstats_update_write_kstats(&zfsvfs
->z_kstat
, nwritten
);
957 task_io_account_write(nwritten
);
964 * Drop a reference on the passed inode asynchronously. This ensures
965 * that the caller will never drop the last reference on an inode in
966 * the current context. Doing so while holding open a tx could result
967 * in a deadlock if iput_final() re-enters the filesystem code.
970 zfs_iput_async(struct inode
*ip
)
972 objset_t
*os
= ITOZSB(ip
)->z_os
;
974 ASSERT(atomic_read(&ip
->i_count
) > 0);
977 if (atomic_read(&ip
->i_count
) == 1)
978 VERIFY(taskq_dispatch(dsl_pool_iput_taskq(dmu_objset_pool(os
)),
979 (task_func_t
*)iput
, ip
, TQ_SLEEP
) != TASKQID_INVALID
);
986 zfs_get_done(zgd_t
*zgd
, int error
)
988 znode_t
*zp
= zgd
->zgd_private
;
991 dmu_buf_rele(zgd
->zgd_db
, zgd
);
993 rangelock_exit(zgd
->zgd_lr
);
996 * Release the vnode asynchronously as we currently have the
997 * txg stopped from syncing.
999 zfs_iput_async(ZTOI(zp
));
1001 kmem_free(zgd
, sizeof (zgd_t
));
1005 static int zil_fault_io
= 0;
1009 * Get data to generate a TX_WRITE intent log record.
1012 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, struct lwb
*lwb
, zio_t
*zio
)
1014 zfsvfs_t
*zfsvfs
= arg
;
1015 objset_t
*os
= zfsvfs
->z_os
;
1017 uint64_t object
= lr
->lr_foid
;
1018 uint64_t offset
= lr
->lr_offset
;
1019 uint64_t size
= lr
->lr_length
;
1024 ASSERT3P(lwb
, !=, NULL
);
1025 ASSERT3P(zio
, !=, NULL
);
1026 ASSERT3U(size
, !=, 0);
1029 * Nothing to do if the file has been removed
1031 if (zfs_zget(zfsvfs
, object
, &zp
) != 0)
1032 return (SET_ERROR(ENOENT
));
1033 if (zp
->z_unlinked
) {
1035 * Release the vnode asynchronously as we currently have the
1036 * txg stopped from syncing.
1038 zfs_iput_async(ZTOI(zp
));
1039 return (SET_ERROR(ENOENT
));
1042 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1044 zgd
->zgd_private
= zp
;
1047 * Write records come in two flavors: immediate and indirect.
1048 * For small writes it's cheaper to store the data with the
1049 * log record (immediate); for large writes it's cheaper to
1050 * sync the data and get a pointer to it (indirect) so that
1051 * we don't have to write the data twice.
1053 if (buf
!= NULL
) { /* immediate write */
1054 zgd
->zgd_lr
= rangelock_enter(&zp
->z_rangelock
,
1055 offset
, size
, RL_READER
);
1056 /* test for truncation needs to be done while range locked */
1057 if (offset
>= zp
->z_size
) {
1058 error
= SET_ERROR(ENOENT
);
1060 error
= dmu_read(os
, object
, offset
, size
, buf
,
1061 DMU_READ_NO_PREFETCH
);
1063 ASSERT(error
== 0 || error
== ENOENT
);
1064 } else { /* indirect write */
1066 * Have to lock the whole block to ensure when it's
1067 * written out and its checksum is being calculated
1068 * that no one can change the data. We need to re-check
1069 * blocksize after we get the lock in case it's changed!
1074 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1076 zgd
->zgd_lr
= rangelock_enter(&zp
->z_rangelock
,
1077 offset
, size
, RL_READER
);
1078 if (zp
->z_blksz
== size
)
1081 rangelock_exit(zgd
->zgd_lr
);
1083 /* test for truncation needs to be done while range locked */
1084 if (lr
->lr_offset
>= zp
->z_size
)
1085 error
= SET_ERROR(ENOENT
);
1088 error
= SET_ERROR(EIO
);
1093 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1094 DMU_READ_NO_PREFETCH
);
1097 blkptr_t
*bp
= &lr
->lr_blkptr
;
1102 ASSERT(db
->db_offset
== offset
);
1103 ASSERT(db
->db_size
== size
);
1105 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1107 ASSERT(error
|| lr
->lr_length
<= size
);
1110 * On success, we need to wait for the write I/O
1111 * initiated by dmu_sync() to complete before we can
1112 * release this dbuf. We will finish everything up
1113 * in the zfs_get_done() callback.
1118 if (error
== EALREADY
) {
1119 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
1121 * TX_WRITE2 relies on the data previously
1122 * written by the TX_WRITE that caused
1123 * EALREADY. We zero out the BP because
1124 * it is the old, currently-on-disk BP.
1133 zfs_get_done(zgd
, error
);
1140 zfs_access(struct inode
*ip
, int mode
, int flag
, cred_t
*cr
)
1142 znode_t
*zp
= ITOZ(ip
);
1143 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
1149 if (flag
& V_ACE_MASK
)
1150 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1152 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1159 * Lookup an entry in a directory, or an extended attribute directory.
1160 * If it exists, return a held inode reference for it.
1162 * IN: dip - inode of directory to search.
1163 * nm - name of entry to lookup.
1164 * flags - LOOKUP_XATTR set if looking for an attribute.
1165 * cr - credentials of caller.
1166 * direntflags - directory lookup flags
1167 * realpnp - returned pathname.
1169 * OUT: ipp - inode of located entry, NULL if not found.
1171 * RETURN: 0 on success, error code on failure.
1178 zfs_lookup(struct inode
*dip
, char *nm
, struct inode
**ipp
, int flags
,
1179 cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1181 znode_t
*zdp
= ITOZ(dip
);
1182 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1186 * Fast path lookup, however we must skip DNLC lookup
1187 * for case folding or normalizing lookups because the
1188 * DNLC code only stores the passed in name. This means
1189 * creating 'a' and removing 'A' on a case insensitive
1190 * file system would work, but DNLC still thinks 'a'
1191 * exists and won't let you create it again on the next
1192 * pass through fast path.
1194 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1196 if (!S_ISDIR(dip
->i_mode
)) {
1197 return (SET_ERROR(ENOTDIR
));
1198 } else if (zdp
->z_sa_hdl
== NULL
) {
1199 return (SET_ERROR(EIO
));
1202 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1203 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1218 if (flags
& LOOKUP_XATTR
) {
1220 * We don't allow recursive attributes..
1221 * Maybe someday we will.
1223 if (zdp
->z_pflags
& ZFS_XATTR
) {
1225 return (SET_ERROR(EINVAL
));
1228 if ((error
= zfs_get_xattrdir(zdp
, ipp
, cr
, flags
))) {
1234 * Do we have permission to get into attribute directory?
1237 if ((error
= zfs_zaccess(ITOZ(*ipp
), ACE_EXECUTE
, 0,
1247 if (!S_ISDIR(dip
->i_mode
)) {
1249 return (SET_ERROR(ENOTDIR
));
1253 * Check accessibility of directory.
1256 if ((error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
))) {
1261 if (zfsvfs
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1262 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1264 return (SET_ERROR(EILSEQ
));
1267 error
= zfs_dirlook(zdp
, nm
, ipp
, flags
, direntflags
, realpnp
);
1268 if ((error
== 0) && (*ipp
))
1269 zfs_inode_update(ITOZ(*ipp
));
1276 * Attempt to create a new entry in a directory. If the entry
1277 * already exists, truncate the file if permissible, else return
1278 * an error. Return the ip of the created or trunc'd file.
1280 * IN: dip - inode of directory to put new file entry in.
1281 * name - name of new file entry.
1282 * vap - attributes of new file.
1283 * excl - flag indicating exclusive or non-exclusive mode.
1284 * mode - mode to open file with.
1285 * cr - credentials of caller.
1286 * flag - large file flag [UNUSED].
1287 * vsecp - ACL to be set
1289 * OUT: ipp - inode of created or trunc'd entry.
1291 * RETURN: 0 on success, error code on failure.
1294 * dip - ctime|mtime updated if new entry created
1295 * ip - ctime|mtime always, atime if new
1300 zfs_create(struct inode
*dip
, char *name
, vattr_t
*vap
, int excl
,
1301 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1303 znode_t
*zp
, *dzp
= ITOZ(dip
);
1304 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1312 zfs_acl_ids_t acl_ids
;
1313 boolean_t fuid_dirtied
;
1314 boolean_t have_acl
= B_FALSE
;
1315 boolean_t waited
= B_FALSE
;
1318 * If we have an ephemeral id, ACL, or XVATTR then
1319 * make sure file system is at proper version
1325 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1326 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1327 return (SET_ERROR(EINVAL
));
1330 return (SET_ERROR(EINVAL
));
1335 zilog
= zfsvfs
->z_log
;
1337 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
1338 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1340 return (SET_ERROR(EILSEQ
));
1343 if (vap
->va_mask
& ATTR_XVATTR
) {
1344 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1345 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1353 if (*name
== '\0') {
1355 * Null component name refers to the directory itself.
1362 /* possible igrab(zp) */
1365 if (flag
& FIGNORECASE
)
1368 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1372 zfs_acl_ids_free(&acl_ids
);
1373 if (strcmp(name
, "..") == 0)
1374 error
= SET_ERROR(EISDIR
);
1382 uint64_t projid
= ZFS_DEFAULT_PROJID
;
1385 * Create a new file object and update the directory
1388 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1390 zfs_acl_ids_free(&acl_ids
);
1395 * We only support the creation of regular files in
1396 * extended attribute directories.
1399 if ((dzp
->z_pflags
& ZFS_XATTR
) && !S_ISREG(vap
->va_mode
)) {
1401 zfs_acl_ids_free(&acl_ids
);
1402 error
= SET_ERROR(EINVAL
);
1406 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1407 cr
, vsecp
, &acl_ids
)) != 0)
1411 if (S_ISREG(vap
->va_mode
) || S_ISDIR(vap
->va_mode
))
1412 projid
= zfs_inherit_projid(dzp
);
1413 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
, projid
)) {
1414 zfs_acl_ids_free(&acl_ids
);
1415 error
= SET_ERROR(EDQUOT
);
1419 tx
= dmu_tx_create(os
);
1421 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1422 ZFS_SA_BASE_ATTR_SIZE
);
1424 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1426 zfs_fuid_txhold(zfsvfs
, tx
);
1427 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1428 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1429 if (!zfsvfs
->z_use_sa
&&
1430 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1431 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1432 0, acl_ids
.z_aclp
->z_acl_bytes
);
1435 error
= dmu_tx_assign(tx
,
1436 (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
1438 zfs_dirent_unlock(dl
);
1439 if (error
== ERESTART
) {
1445 zfs_acl_ids_free(&acl_ids
);
1450 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1452 error
= zfs_link_create(dl
, zp
, tx
, ZNEW
);
1455 * Since, we failed to add the directory entry for it,
1456 * delete the newly created dnode.
1458 zfs_znode_delete(zp
, tx
);
1459 remove_inode_hash(ZTOI(zp
));
1460 zfs_acl_ids_free(&acl_ids
);
1466 zfs_fuid_sync(zfsvfs
, tx
);
1468 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1469 if (flag
& FIGNORECASE
)
1471 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1472 vsecp
, acl_ids
.z_fuidp
, vap
);
1473 zfs_acl_ids_free(&acl_ids
);
1476 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1479 zfs_acl_ids_free(&acl_ids
);
1483 * A directory entry already exists for this name.
1486 * Can't truncate an existing file if in exclusive mode.
1489 error
= SET_ERROR(EEXIST
);
1493 * Can't open a directory for writing.
1495 if (S_ISDIR(ZTOI(zp
)->i_mode
)) {
1496 error
= SET_ERROR(EISDIR
);
1500 * Verify requested access to file.
1502 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1506 mutex_enter(&dzp
->z_lock
);
1508 mutex_exit(&dzp
->z_lock
);
1511 * Truncate regular files if requested.
1513 if (S_ISREG(ZTOI(zp
)->i_mode
) &&
1514 (vap
->va_mask
& ATTR_SIZE
) && (vap
->va_size
== 0)) {
1515 /* we can't hold any locks when calling zfs_freesp() */
1517 zfs_dirent_unlock(dl
);
1520 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1526 zfs_dirent_unlock(dl
);
1532 zfs_inode_update(dzp
);
1533 zfs_inode_update(zp
);
1537 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1538 zil_commit(zilog
, 0);
1546 zfs_tmpfile(struct inode
*dip
, vattr_t
*vap
, int excl
,
1547 int mode
, struct inode
**ipp
, cred_t
*cr
, int flag
, vsecattr_t
*vsecp
)
1549 znode_t
*zp
= NULL
, *dzp
= ITOZ(dip
);
1550 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1556 zfs_acl_ids_t acl_ids
;
1557 uint64_t projid
= ZFS_DEFAULT_PROJID
;
1558 boolean_t fuid_dirtied
;
1559 boolean_t have_acl
= B_FALSE
;
1560 boolean_t waited
= B_FALSE
;
1563 * If we have an ephemeral id, ACL, or XVATTR then
1564 * make sure file system is at proper version
1570 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1571 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1572 return (SET_ERROR(EINVAL
));
1578 if (vap
->va_mask
& ATTR_XVATTR
) {
1579 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1580 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1590 * Create a new file object and update the directory
1593 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
1595 zfs_acl_ids_free(&acl_ids
);
1599 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1600 cr
, vsecp
, &acl_ids
)) != 0)
1604 if (S_ISREG(vap
->va_mode
) || S_ISDIR(vap
->va_mode
))
1605 projid
= zfs_inherit_projid(dzp
);
1606 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
, projid
)) {
1607 zfs_acl_ids_free(&acl_ids
);
1608 error
= SET_ERROR(EDQUOT
);
1612 tx
= dmu_tx_create(os
);
1614 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1615 ZFS_SA_BASE_ATTR_SIZE
);
1616 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1618 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1620 zfs_fuid_txhold(zfsvfs
, tx
);
1621 if (!zfsvfs
->z_use_sa
&&
1622 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1623 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1624 0, acl_ids
.z_aclp
->z_acl_bytes
);
1626 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
1628 if (error
== ERESTART
) {
1634 zfs_acl_ids_free(&acl_ids
);
1639 zfs_mknode(dzp
, vap
, tx
, cr
, IS_TMPFILE
, &zp
, &acl_ids
);
1642 zfs_fuid_sync(zfsvfs
, tx
);
1644 /* Add to unlinked set */
1646 zfs_unlinked_add(zp
, tx
);
1647 zfs_acl_ids_free(&acl_ids
);
1655 zfs_inode_update(dzp
);
1656 zfs_inode_update(zp
);
1665 * Remove an entry from a directory.
1667 * IN: dip - inode of directory to remove entry from.
1668 * name - name of entry to remove.
1669 * cr - credentials of caller.
1671 * RETURN: 0 if success
1672 * error code if failure
1676 * ip - ctime (if nlink > 0)
1679 uint64_t null_xattr
= 0;
1683 zfs_remove(struct inode
*dip
, char *name
, cred_t
*cr
, int flags
)
1685 znode_t
*zp
, *dzp
= ITOZ(dip
);
1688 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1690 uint64_t acl_obj
, xattr_obj
;
1691 uint64_t xattr_obj_unlinked
= 0;
1696 boolean_t may_delete_now
, delete_now
= FALSE
;
1697 boolean_t unlinked
, toobig
= FALSE
;
1699 pathname_t
*realnmp
= NULL
;
1703 boolean_t waited
= B_FALSE
;
1706 return (SET_ERROR(EINVAL
));
1710 zilog
= zfsvfs
->z_log
;
1712 if (flags
& FIGNORECASE
) {
1722 * Attempt to lock directory; fail if entry doesn't exist.
1724 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1734 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
1739 * Need to use rmdir for removing directories.
1741 if (S_ISDIR(ip
->i_mode
)) {
1742 error
= SET_ERROR(EPERM
);
1746 mutex_enter(&zp
->z_lock
);
1747 may_delete_now
= atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
);
1748 mutex_exit(&zp
->z_lock
);
1751 * We may delete the znode now, or we may put it in the unlinked set;
1752 * it depends on whether we're the last link, and on whether there are
1753 * other holds on the inode. So we dmu_tx_hold() the right things to
1754 * allow for either case.
1757 tx
= dmu_tx_create(zfsvfs
->z_os
);
1758 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1759 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1760 zfs_sa_upgrade_txholds(tx
, zp
);
1761 zfs_sa_upgrade_txholds(tx
, dzp
);
1762 if (may_delete_now
) {
1763 toobig
= zp
->z_size
> zp
->z_blksz
* zfs_delete_blocks
;
1764 /* if the file is too big, only hold_free a token amount */
1765 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1766 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1769 /* are there any extended attributes? */
1770 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1771 &xattr_obj
, sizeof (xattr_obj
));
1772 if (error
== 0 && xattr_obj
) {
1773 error
= zfs_zget(zfsvfs
, xattr_obj
, &xzp
);
1775 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1776 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1779 mutex_enter(&zp
->z_lock
);
1780 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1781 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1782 mutex_exit(&zp
->z_lock
);
1784 /* charge as an update -- would be nice not to charge at all */
1785 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1788 * Mark this transaction as typically resulting in a net free of space
1790 dmu_tx_mark_netfree(tx
);
1792 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
1794 zfs_dirent_unlock(dl
);
1795 if (error
== ERESTART
) {
1815 * Remove the directory entry.
1817 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1826 * Hold z_lock so that we can make sure that the ACL obj
1827 * hasn't changed. Could have been deleted due to
1830 mutex_enter(&zp
->z_lock
);
1831 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1832 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1833 delete_now
= may_delete_now
&& !toobig
&&
1834 atomic_read(&ip
->i_count
) == 1 && !(zp
->z_is_mapped
) &&
1835 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1840 if (xattr_obj_unlinked
) {
1841 ASSERT3U(ZTOI(xzp
)->i_nlink
, ==, 2);
1842 mutex_enter(&xzp
->z_lock
);
1843 xzp
->z_unlinked
= 1;
1844 clear_nlink(ZTOI(xzp
));
1846 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zfsvfs
),
1847 &links
, sizeof (links
), tx
);
1848 ASSERT3U(error
, ==, 0);
1849 mutex_exit(&xzp
->z_lock
);
1850 zfs_unlinked_add(xzp
, tx
);
1853 error
= sa_remove(zp
->z_sa_hdl
,
1854 SA_ZPL_XATTR(zfsvfs
), tx
);
1856 error
= sa_update(zp
->z_sa_hdl
,
1857 SA_ZPL_XATTR(zfsvfs
), &null_xattr
,
1858 sizeof (uint64_t), tx
);
1862 * Add to the unlinked set because a new reference could be
1863 * taken concurrently resulting in a deferred destruction.
1865 zfs_unlinked_add(zp
, tx
);
1866 mutex_exit(&zp
->z_lock
);
1867 } else if (unlinked
) {
1868 mutex_exit(&zp
->z_lock
);
1869 zfs_unlinked_add(zp
, tx
);
1873 if (flags
& FIGNORECASE
)
1875 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1882 zfs_dirent_unlock(dl
);
1883 zfs_inode_update(dzp
);
1884 zfs_inode_update(zp
);
1892 zfs_inode_update(xzp
);
1893 zfs_iput_async(ZTOI(xzp
));
1896 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1897 zil_commit(zilog
, 0);
1904 * Create a new directory and insert it into dip using the name
1905 * provided. Return a pointer to the inserted directory.
1907 * IN: dip - inode of directory to add subdir to.
1908 * dirname - name of new directory.
1909 * vap - attributes of new directory.
1910 * cr - credentials of caller.
1911 * vsecp - ACL to be set
1913 * OUT: ipp - inode of created directory.
1915 * RETURN: 0 if success
1916 * error code if failure
1919 * dip - ctime|mtime updated
1920 * ipp - ctime|mtime|atime updated
1924 zfs_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
, struct inode
**ipp
,
1925 cred_t
*cr
, int flags
, vsecattr_t
*vsecp
)
1927 znode_t
*zp
, *dzp
= ITOZ(dip
);
1928 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1936 gid_t gid
= crgetgid(cr
);
1937 zfs_acl_ids_t acl_ids
;
1938 boolean_t fuid_dirtied
;
1939 boolean_t waited
= B_FALSE
;
1941 ASSERT(S_ISDIR(vap
->va_mode
));
1944 * If we have an ephemeral id, ACL, or XVATTR then
1945 * make sure file system is at proper version
1949 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1950 (vsecp
|| IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1951 return (SET_ERROR(EINVAL
));
1953 if (dirname
== NULL
)
1954 return (SET_ERROR(EINVAL
));
1958 zilog
= zfsvfs
->z_log
;
1960 if (dzp
->z_pflags
& ZFS_XATTR
) {
1962 return (SET_ERROR(EINVAL
));
1965 if (zfsvfs
->z_utf8
&& u8_validate(dirname
,
1966 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1968 return (SET_ERROR(EILSEQ
));
1970 if (flags
& FIGNORECASE
)
1973 if (vap
->va_mask
& ATTR_XVATTR
) {
1974 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1975 crgetuid(cr
), cr
, vap
->va_mode
)) != 0) {
1981 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1982 vsecp
, &acl_ids
)) != 0) {
1987 * First make sure the new directory doesn't exist.
1989 * Existence is checked first to make sure we don't return
1990 * EACCES instead of EEXIST which can cause some applications
1996 if ((error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1998 zfs_acl_ids_free(&acl_ids
);
2003 if ((error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
))) {
2004 zfs_acl_ids_free(&acl_ids
);
2005 zfs_dirent_unlock(dl
);
2010 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
, zfs_inherit_projid(dzp
))) {
2011 zfs_acl_ids_free(&acl_ids
);
2012 zfs_dirent_unlock(dl
);
2014 return (SET_ERROR(EDQUOT
));
2018 * Add a new entry to the directory.
2020 tx
= dmu_tx_create(zfsvfs
->z_os
);
2021 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
2022 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
2023 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
2025 zfs_fuid_txhold(zfsvfs
, tx
);
2026 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
2027 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
2028 acl_ids
.z_aclp
->z_acl_bytes
);
2031 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
2032 ZFS_SA_BASE_ATTR_SIZE
);
2034 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
2036 zfs_dirent_unlock(dl
);
2037 if (error
== ERESTART
) {
2043 zfs_acl_ids_free(&acl_ids
);
2052 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
2055 * Now put new name in parent dir.
2057 error
= zfs_link_create(dl
, zp
, tx
, ZNEW
);
2059 zfs_znode_delete(zp
, tx
);
2060 remove_inode_hash(ZTOI(zp
));
2065 zfs_fuid_sync(zfsvfs
, tx
);
2069 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
2070 if (flags
& FIGNORECASE
)
2072 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
2073 acl_ids
.z_fuidp
, vap
);
2076 zfs_acl_ids_free(&acl_ids
);
2080 zfs_dirent_unlock(dl
);
2082 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2083 zil_commit(zilog
, 0);
2088 zfs_inode_update(dzp
);
2089 zfs_inode_update(zp
);
2096 * Remove a directory subdir entry. If the current working
2097 * directory is the same as the subdir to be removed, the
2100 * IN: dip - inode of directory to remove from.
2101 * name - name of directory to be removed.
2102 * cwd - inode of current working directory.
2103 * cr - credentials of caller.
2104 * flags - case flags
2106 * RETURN: 0 on success, error code on failure.
2109 * dip - ctime|mtime updated
2113 zfs_rmdir(struct inode
*dip
, char *name
, struct inode
*cwd
, cred_t
*cr
,
2116 znode_t
*dzp
= ITOZ(dip
);
2119 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
2125 boolean_t waited
= B_FALSE
;
2128 return (SET_ERROR(EINVAL
));
2132 zilog
= zfsvfs
->z_log
;
2134 if (flags
& FIGNORECASE
)
2140 * Attempt to lock directory; fail if entry doesn't exist.
2142 if ((error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
2150 if ((error
= zfs_zaccess_delete(dzp
, zp
, cr
))) {
2154 if (!S_ISDIR(ip
->i_mode
)) {
2155 error
= SET_ERROR(ENOTDIR
);
2160 error
= SET_ERROR(EINVAL
);
2165 * Grab a lock on the directory to make sure that no one is
2166 * trying to add (or lookup) entries while we are removing it.
2168 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
2171 * Grab a lock on the parent pointer to make sure we play well
2172 * with the treewalk and directory rename code.
2174 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
2176 tx
= dmu_tx_create(zfsvfs
->z_os
);
2177 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2178 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2179 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
2180 zfs_sa_upgrade_txholds(tx
, zp
);
2181 zfs_sa_upgrade_txholds(tx
, dzp
);
2182 dmu_tx_mark_netfree(tx
);
2183 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
2185 rw_exit(&zp
->z_parent_lock
);
2186 rw_exit(&zp
->z_name_lock
);
2187 zfs_dirent_unlock(dl
);
2188 if (error
== ERESTART
) {
2201 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2204 uint64_t txtype
= TX_RMDIR
;
2205 if (flags
& FIGNORECASE
)
2207 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2212 rw_exit(&zp
->z_parent_lock
);
2213 rw_exit(&zp
->z_name_lock
);
2215 zfs_dirent_unlock(dl
);
2217 zfs_inode_update(dzp
);
2218 zfs_inode_update(zp
);
2221 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2222 zil_commit(zilog
, 0);
2229 * Read as many directory entries as will fit into the provided
2230 * dirent buffer from the given directory cursor position.
2232 * IN: ip - inode of directory to read.
2233 * dirent - buffer for directory entries.
2235 * OUT: dirent - filler buffer of directory entries.
2237 * RETURN: 0 if success
2238 * error code if failure
2241 * ip - atime updated
2243 * Note that the low 4 bits of the cookie returned by zap is always zero.
2244 * This allows us to use the low range for "special" directory entries:
2245 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2246 * we use the offset 2 for the '.zfs' directory.
2250 zfs_readdir(struct inode
*ip
, zpl_dir_context_t
*ctx
, cred_t
*cr
)
2252 znode_t
*zp
= ITOZ(ip
);
2253 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2256 zap_attribute_t zap
;
2262 uint64_t offset
; /* must be unsigned; checks for < 1 */
2267 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
2268 &parent
, sizeof (parent
))) != 0)
2272 * Quit if directory has been removed (posix)
2280 prefetch
= zp
->z_zn_prefetch
;
2283 * Initialize the iterator cursor.
2287 * Start iteration from the beginning of the directory.
2289 zap_cursor_init(&zc
, os
, zp
->z_id
);
2292 * The offset is a serialized cursor.
2294 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2298 * Transform to file-system independent format
2303 * Special case `.', `..', and `.zfs'.
2306 (void) strcpy(zap
.za_name
, ".");
2307 zap
.za_normalization_conflict
= 0;
2310 } else if (offset
== 1) {
2311 (void) strcpy(zap
.za_name
, "..");
2312 zap
.za_normalization_conflict
= 0;
2315 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2316 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2317 zap
.za_normalization_conflict
= 0;
2318 objnum
= ZFSCTL_INO_ROOT
;
2324 if ((error
= zap_cursor_retrieve(&zc
, &zap
))) {
2325 if (error
== ENOENT
)
2332 * Allow multiple entries provided the first entry is
2333 * the object id. Non-zpl consumers may safely make
2334 * use of the additional space.
2336 * XXX: This should be a feature flag for compatibility
2338 if (zap
.za_integer_length
!= 8 ||
2339 zap
.za_num_integers
== 0) {
2340 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2341 "entry, obj = %lld, offset = %lld, "
2342 "length = %d, num = %lld\n",
2343 (u_longlong_t
)zp
->z_id
,
2344 (u_longlong_t
)offset
,
2345 zap
.za_integer_length
,
2346 (u_longlong_t
)zap
.za_num_integers
);
2347 error
= SET_ERROR(ENXIO
);
2351 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2352 type
= ZFS_DIRENT_TYPE(zap
.za_first_integer
);
2355 done
= !zpl_dir_emit(ctx
, zap
.za_name
, strlen(zap
.za_name
),
2360 /* Prefetch znode */
2362 dmu_prefetch(os
, objnum
, 0, 0, 0,
2363 ZIO_PRIORITY_SYNC_READ
);
2367 * Move to the next entry, fill in the previous offset.
2369 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2370 zap_cursor_advance(&zc
);
2371 offset
= zap_cursor_serialize(&zc
);
2377 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2380 zap_cursor_fini(&zc
);
2381 if (error
== ENOENT
)
2389 ulong_t zfs_fsync_sync_cnt
= 4;
2392 zfs_fsync(struct inode
*ip
, int syncflag
, cred_t
*cr
)
2394 znode_t
*zp
= ITOZ(ip
);
2395 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2397 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2399 if (zfsvfs
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2402 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
2405 tsd_set(zfs_fsyncer_key
, NULL
);
2412 * Get the requested file attributes and place them in the provided
2415 * IN: ip - inode of file.
2416 * vap - va_mask identifies requested attributes.
2417 * If ATTR_XVATTR set, then optional attrs are requested
2418 * flags - ATTR_NOACLCHECK (CIFS server context)
2419 * cr - credentials of caller.
2421 * OUT: vap - attribute values.
2423 * RETURN: 0 (always succeeds)
2427 zfs_getattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2429 znode_t
*zp
= ITOZ(ip
);
2430 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2433 uint64_t atime
[2], mtime
[2], ctime
[2];
2434 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2435 xoptattr_t
*xoap
= NULL
;
2436 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2437 sa_bulk_attr_t bulk
[3];
2443 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2445 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
, &atime
, 16);
2446 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
2447 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
2449 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2455 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2456 * Also, if we are the owner don't bother, since owner should
2457 * always be allowed to read basic attributes of file.
2459 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2460 (vap
->va_uid
!= crgetuid(cr
))) {
2461 if ((error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2469 * Return all attributes. It's cheaper to provide the answer
2470 * than to determine whether we were asked the question.
2473 mutex_enter(&zp
->z_lock
);
2474 vap
->va_type
= vn_mode_to_vtype(zp
->z_mode
);
2475 vap
->va_mode
= zp
->z_mode
;
2476 vap
->va_fsid
= ZTOI(zp
)->i_sb
->s_dev
;
2477 vap
->va_nodeid
= zp
->z_id
;
2478 if ((zp
->z_id
== zfsvfs
->z_root
) && zfs_show_ctldir(zp
))
2479 links
= ZTOI(zp
)->i_nlink
+ 1;
2481 links
= ZTOI(zp
)->i_nlink
;
2482 vap
->va_nlink
= MIN(links
, ZFS_LINK_MAX
);
2483 vap
->va_size
= i_size_read(ip
);
2484 vap
->va_rdev
= ip
->i_rdev
;
2485 vap
->va_seq
= ip
->i_generation
;
2488 * Add in any requested optional attributes and the create time.
2489 * Also set the corresponding bits in the returned attribute bitmap.
2491 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zfsvfs
->z_use_fuids
) {
2492 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2494 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2495 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2498 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2499 xoap
->xoa_readonly
=
2500 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2501 XVA_SET_RTN(xvap
, XAT_READONLY
);
2504 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2506 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2507 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2510 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2512 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2513 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2516 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2517 xoap
->xoa_nounlink
=
2518 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2519 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2522 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2523 xoap
->xoa_immutable
=
2524 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2525 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2528 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2529 xoap
->xoa_appendonly
=
2530 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2531 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2534 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2536 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2537 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2540 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2542 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2543 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2546 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2547 xoap
->xoa_av_quarantined
=
2548 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2549 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2552 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2553 xoap
->xoa_av_modified
=
2554 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2555 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2558 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2559 S_ISREG(ip
->i_mode
)) {
2560 zfs_sa_get_scanstamp(zp
, xvap
);
2563 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2566 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zfsvfs
),
2567 times
, sizeof (times
));
2568 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2569 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2572 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2573 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2574 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2576 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2577 xoap
->xoa_generation
= ip
->i_generation
;
2578 XVA_SET_RTN(xvap
, XAT_GEN
);
2581 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2583 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2584 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2587 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2589 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2590 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2593 if (XVA_ISSET_REQ(xvap
, XAT_PROJINHERIT
)) {
2594 xoap
->xoa_projinherit
=
2595 ((zp
->z_pflags
& ZFS_PROJINHERIT
) != 0);
2596 XVA_SET_RTN(xvap
, XAT_PROJINHERIT
);
2599 if (XVA_ISSET_REQ(xvap
, XAT_PROJID
)) {
2600 xoap
->xoa_projid
= zp
->z_projid
;
2601 XVA_SET_RTN(xvap
, XAT_PROJID
);
2605 ZFS_TIME_DECODE(&vap
->va_atime
, atime
);
2606 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2607 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2609 mutex_exit(&zp
->z_lock
);
2611 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2613 if (zp
->z_blksz
== 0) {
2615 * Block size hasn't been set; suggest maximal I/O transfers.
2617 vap
->va_blksize
= zfsvfs
->z_max_blksz
;
2625 * Get the basic file attributes and place them in the provided kstat
2626 * structure. The inode is assumed to be the authoritative source
2627 * for most of the attributes. However, the znode currently has the
2628 * authoritative atime, blksize, and block count.
2630 * IN: ip - inode of file.
2632 * OUT: sp - kstat values.
2634 * RETURN: 0 (always succeeds)
2638 zfs_getattr_fast(struct inode
*ip
, struct kstat
*sp
)
2640 znode_t
*zp
= ITOZ(ip
);
2641 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2643 u_longlong_t nblocks
;
2648 mutex_enter(&zp
->z_lock
);
2650 generic_fillattr(ip
, sp
);
2652 sa_object_size(zp
->z_sa_hdl
, &blksize
, &nblocks
);
2653 sp
->blksize
= blksize
;
2654 sp
->blocks
= nblocks
;
2656 if (unlikely(zp
->z_blksz
== 0)) {
2658 * Block size hasn't been set; suggest maximal I/O transfers.
2660 sp
->blksize
= zfsvfs
->z_max_blksz
;
2663 mutex_exit(&zp
->z_lock
);
2666 * Required to prevent NFS client from detecting different inode
2667 * numbers of snapshot root dentry before and after snapshot mount.
2669 if (zfsvfs
->z_issnap
) {
2670 if (ip
->i_sb
->s_root
->d_inode
== ip
)
2671 sp
->ino
= ZFSCTL_INO_SNAPDIRS
-
2672 dmu_objset_id(zfsvfs
->z_os
);
2681 * For the operation of changing file's user/group/project, we need to
2682 * handle not only the main object that is assigned to the file directly,
2683 * but also the ones that are used by the file via hidden xattr directory.
2685 * Because the xattr directory may contains many EA entries, as to it may
2686 * be impossible to change all of them via the transaction of changing the
2687 * main object's user/group/project attributes. Then we have to change them
2688 * via other multiple independent transactions one by one. It may be not good
2689 * solution, but we have no better idea yet.
2692 zfs_setattr_dir(znode_t
*dzp
)
2694 struct inode
*dxip
= ZTOI(dzp
);
2695 struct inode
*xip
= NULL
;
2696 zfsvfs_t
*zfsvfs
= ITOZSB(dxip
);
2697 objset_t
*os
= zfsvfs
->z_os
;
2699 zap_attribute_t zap
;
2702 dmu_tx_t
*tx
= NULL
;
2704 sa_bulk_attr_t bulk
[4];
2708 zap_cursor_init(&zc
, os
, dzp
->z_id
);
2709 while ((err
= zap_cursor_retrieve(&zc
, &zap
)) == 0) {
2710 if (zap
.za_integer_length
!= 8 || zap
.za_num_integers
!= 1) {
2715 err
= zfs_dirent_lock(&dl
, dzp
, (char *)zap
.za_name
, &zp
,
2716 ZEXISTS
, NULL
, NULL
);
2723 if (KUID_TO_SUID(xip
->i_uid
) == KUID_TO_SUID(dxip
->i_uid
) &&
2724 KGID_TO_SGID(xip
->i_gid
) == KGID_TO_SGID(dxip
->i_gid
) &&
2725 zp
->z_projid
== dzp
->z_projid
)
2728 tx
= dmu_tx_create(os
);
2729 if (!(zp
->z_pflags
& ZFS_PROJID
))
2730 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
2732 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2734 err
= dmu_tx_assign(tx
, TXG_WAIT
);
2738 mutex_enter(&dzp
->z_lock
);
2740 if (KUID_TO_SUID(xip
->i_uid
) != KUID_TO_SUID(dxip
->i_uid
)) {
2741 xip
->i_uid
= dxip
->i_uid
;
2742 uid
= zfs_uid_read(dxip
);
2743 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zfsvfs
), NULL
,
2744 &uid
, sizeof (uid
));
2747 if (KGID_TO_SGID(xip
->i_gid
) != KGID_TO_SGID(dxip
->i_gid
)) {
2748 xip
->i_gid
= dxip
->i_gid
;
2749 gid
= zfs_gid_read(dxip
);
2750 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zfsvfs
), NULL
,
2751 &gid
, sizeof (gid
));
2754 if (zp
->z_projid
!= dzp
->z_projid
) {
2755 if (!(zp
->z_pflags
& ZFS_PROJID
)) {
2756 zp
->z_pflags
|= ZFS_PROJID
;
2757 SA_ADD_BULK_ATTR(bulk
, count
,
2758 SA_ZPL_FLAGS(zfsvfs
), NULL
, &zp
->z_pflags
,
2759 sizeof (zp
->z_pflags
));
2762 zp
->z_projid
= dzp
->z_projid
;
2763 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_PROJID(zfsvfs
),
2764 NULL
, &zp
->z_projid
, sizeof (zp
->z_projid
));
2767 mutex_exit(&dzp
->z_lock
);
2769 if (likely(count
> 0)) {
2770 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
2776 if (err
!= 0 && err
!= ENOENT
)
2783 zfs_dirent_unlock(dl
);
2785 zap_cursor_advance(&zc
);
2792 zfs_dirent_unlock(dl
);
2794 zap_cursor_fini(&zc
);
2796 return (err
== ENOENT
? 0 : err
);
2800 * Set the file attributes to the values contained in the
2803 * IN: ip - inode of file to be modified.
2804 * vap - new attribute values.
2805 * If ATTR_XVATTR set, then optional attrs are being set
2806 * flags - ATTR_UTIME set if non-default time values provided.
2807 * - ATTR_NOACLCHECK (CIFS context only).
2808 * cr - credentials of caller.
2810 * RETURN: 0 if success
2811 * error code if failure
2814 * ip - ctime updated, mtime updated if size changed.
2818 zfs_setattr(struct inode
*ip
, vattr_t
*vap
, int flags
, cred_t
*cr
)
2820 znode_t
*zp
= ITOZ(ip
);
2821 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
2822 objset_t
*os
= zfsvfs
->z_os
;
2826 xvattr_t
*tmpxvattr
;
2827 uint_t mask
= vap
->va_mask
;
2828 uint_t saved_mask
= 0;
2831 uint64_t new_kuid
= 0, new_kgid
= 0, new_uid
, new_gid
;
2833 uint64_t mtime
[2], ctime
[2], atime
[2];
2834 uint64_t projid
= ZFS_INVALID_PROJID
;
2836 int need_policy
= FALSE
;
2838 zfs_fuid_info_t
*fuidp
= NULL
;
2839 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2842 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2843 boolean_t fuid_dirtied
= B_FALSE
;
2844 boolean_t handle_eadir
= B_FALSE
;
2845 sa_bulk_attr_t
*bulk
, *xattr_bulk
;
2846 int count
= 0, xattr_count
= 0, bulks
= 8;
2855 * If this is a xvattr_t, then get a pointer to the structure of
2856 * optional attributes. If this is NULL, then we have a vattr_t.
2858 xoap
= xva_getxoptattr(xvap
);
2859 if (xoap
!= NULL
&& (mask
& ATTR_XVATTR
)) {
2860 if (XVA_ISSET_REQ(xvap
, XAT_PROJID
)) {
2861 if (!dmu_objset_projectquota_enabled(os
) ||
2862 (!S_ISREG(ip
->i_mode
) && !S_ISDIR(ip
->i_mode
))) {
2864 return (SET_ERROR(ENOTSUP
));
2867 projid
= xoap
->xoa_projid
;
2868 if (unlikely(projid
== ZFS_INVALID_PROJID
)) {
2870 return (SET_ERROR(EINVAL
));
2873 if (projid
== zp
->z_projid
&& zp
->z_pflags
& ZFS_PROJID
)
2874 projid
= ZFS_INVALID_PROJID
;
2879 if (XVA_ISSET_REQ(xvap
, XAT_PROJINHERIT
) &&
2880 (xoap
->xoa_projinherit
!=
2881 ((zp
->z_pflags
& ZFS_PROJINHERIT
) != 0)) &&
2882 (!dmu_objset_projectquota_enabled(os
) ||
2883 (!S_ISREG(ip
->i_mode
) && !S_ISDIR(ip
->i_mode
)))) {
2885 return (SET_ERROR(ENOTSUP
));
2889 zilog
= zfsvfs
->z_log
;
2892 * Make sure that if we have ephemeral uid/gid or xvattr specified
2893 * that file system is at proper version level
2896 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
2897 (((mask
& ATTR_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2898 ((mask
& ATTR_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2899 (mask
& ATTR_XVATTR
))) {
2901 return (SET_ERROR(EINVAL
));
2904 if (mask
& ATTR_SIZE
&& S_ISDIR(ip
->i_mode
)) {
2906 return (SET_ERROR(EISDIR
));
2909 if (mask
& ATTR_SIZE
&& !S_ISREG(ip
->i_mode
) && !S_ISFIFO(ip
->i_mode
)) {
2911 return (SET_ERROR(EINVAL
));
2914 tmpxvattr
= kmem_alloc(sizeof (xvattr_t
), KM_SLEEP
);
2915 xva_init(tmpxvattr
);
2917 bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * bulks
, KM_SLEEP
);
2918 xattr_bulk
= kmem_alloc(sizeof (sa_bulk_attr_t
) * bulks
, KM_SLEEP
);
2921 * Immutable files can only alter immutable bit and atime
2923 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2924 ((mask
& (ATTR_SIZE
|ATTR_UID
|ATTR_GID
|ATTR_MTIME
|ATTR_MODE
)) ||
2925 ((mask
& ATTR_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2926 err
= SET_ERROR(EPERM
);
2930 if ((mask
& ATTR_SIZE
) && (zp
->z_pflags
& ZFS_READONLY
)) {
2931 err
= SET_ERROR(EPERM
);
2936 * Verify timestamps doesn't overflow 32 bits.
2937 * ZFS can handle large timestamps, but 32bit syscalls can't
2938 * handle times greater than 2039. This check should be removed
2939 * once large timestamps are fully supported.
2941 if (mask
& (ATTR_ATIME
| ATTR_MTIME
)) {
2942 if (((mask
& ATTR_ATIME
) &&
2943 TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2944 ((mask
& ATTR_MTIME
) &&
2945 TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2946 err
= SET_ERROR(EOVERFLOW
);
2955 /* Can this be moved to before the top label? */
2956 if (zfs_is_readonly(zfsvfs
)) {
2957 err
= SET_ERROR(EROFS
);
2962 * First validate permissions
2965 if (mask
& ATTR_SIZE
) {
2966 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2971 * XXX - Note, we are not providing any open
2972 * mode flags here (like FNDELAY), so we may
2973 * block if there are locks present... this
2974 * should be addressed in openat().
2976 /* XXX - would it be OK to generate a log record here? */
2977 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2982 if (mask
& (ATTR_ATIME
|ATTR_MTIME
) ||
2983 ((mask
& ATTR_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2984 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2985 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2986 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2987 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2988 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2989 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2990 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2994 if (mask
& (ATTR_UID
|ATTR_GID
)) {
2995 int idmask
= (mask
& (ATTR_UID
|ATTR_GID
));
3000 * NOTE: even if a new mode is being set,
3001 * we may clear S_ISUID/S_ISGID bits.
3004 if (!(mask
& ATTR_MODE
))
3005 vap
->va_mode
= zp
->z_mode
;
3008 * Take ownership or chgrp to group we are a member of
3011 take_owner
= (mask
& ATTR_UID
) && (vap
->va_uid
== crgetuid(cr
));
3012 take_group
= (mask
& ATTR_GID
) &&
3013 zfs_groupmember(zfsvfs
, vap
->va_gid
, cr
);
3016 * If both ATTR_UID and ATTR_GID are set then take_owner and
3017 * take_group must both be set in order to allow taking
3020 * Otherwise, send the check through secpolicy_vnode_setattr()
3024 if (((idmask
== (ATTR_UID
|ATTR_GID
)) &&
3025 take_owner
&& take_group
) ||
3026 ((idmask
== ATTR_UID
) && take_owner
) ||
3027 ((idmask
== ATTR_GID
) && take_group
)) {
3028 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
3029 skipaclchk
, cr
) == 0) {
3031 * Remove setuid/setgid for non-privileged users
3033 (void) secpolicy_setid_clear(vap
, cr
);
3034 trim_mask
= (mask
& (ATTR_UID
|ATTR_GID
));
3043 mutex_enter(&zp
->z_lock
);
3044 oldva
.va_mode
= zp
->z_mode
;
3045 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
3046 if (mask
& ATTR_XVATTR
) {
3048 * Update xvattr mask to include only those attributes
3049 * that are actually changing.
3051 * the bits will be restored prior to actually setting
3052 * the attributes so the caller thinks they were set.
3054 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
3055 if (xoap
->xoa_appendonly
!=
3056 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
3059 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
3060 XVA_SET_REQ(tmpxvattr
, XAT_APPENDONLY
);
3064 if (XVA_ISSET_REQ(xvap
, XAT_PROJINHERIT
)) {
3065 if (xoap
->xoa_projinherit
!=
3066 ((zp
->z_pflags
& ZFS_PROJINHERIT
) != 0)) {
3069 XVA_CLR_REQ(xvap
, XAT_PROJINHERIT
);
3070 XVA_SET_REQ(tmpxvattr
, XAT_PROJINHERIT
);
3074 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
3075 if (xoap
->xoa_nounlink
!=
3076 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
3079 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
3080 XVA_SET_REQ(tmpxvattr
, XAT_NOUNLINK
);
3084 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
3085 if (xoap
->xoa_immutable
!=
3086 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
3089 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
3090 XVA_SET_REQ(tmpxvattr
, XAT_IMMUTABLE
);
3094 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
3095 if (xoap
->xoa_nodump
!=
3096 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
3099 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
3100 XVA_SET_REQ(tmpxvattr
, XAT_NODUMP
);
3104 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
3105 if (xoap
->xoa_av_modified
!=
3106 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
3109 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
3110 XVA_SET_REQ(tmpxvattr
, XAT_AV_MODIFIED
);
3114 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
3115 if ((!S_ISREG(ip
->i_mode
) &&
3116 xoap
->xoa_av_quarantined
) ||
3117 xoap
->xoa_av_quarantined
!=
3118 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
3121 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
3122 XVA_SET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
);
3126 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
3127 mutex_exit(&zp
->z_lock
);
3128 err
= SET_ERROR(EPERM
);
3132 if (need_policy
== FALSE
&&
3133 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
3134 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
3139 mutex_exit(&zp
->z_lock
);
3141 if (mask
& ATTR_MODE
) {
3142 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
3143 err
= secpolicy_setid_setsticky_clear(ip
, vap
,
3148 trim_mask
|= ATTR_MODE
;
3156 * If trim_mask is set then take ownership
3157 * has been granted or write_acl is present and user
3158 * has the ability to modify mode. In that case remove
3159 * UID|GID and or MODE from mask so that
3160 * secpolicy_vnode_setattr() doesn't revoke it.
3164 saved_mask
= vap
->va_mask
;
3165 vap
->va_mask
&= ~trim_mask
;
3167 err
= secpolicy_vnode_setattr(cr
, ip
, vap
, &oldva
, flags
,
3168 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
3173 vap
->va_mask
|= saved_mask
;
3177 * secpolicy_vnode_setattr, or take ownership may have
3180 mask
= vap
->va_mask
;
3182 if ((mask
& (ATTR_UID
| ATTR_GID
)) || projid
!= ZFS_INVALID_PROJID
) {
3183 handle_eadir
= B_TRUE
;
3184 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
3185 &xattr_obj
, sizeof (xattr_obj
));
3187 if (err
== 0 && xattr_obj
) {
3188 err
= zfs_zget(ZTOZSB(zp
), xattr_obj
, &attrzp
);
3192 if (mask
& ATTR_UID
) {
3193 new_kuid
= zfs_fuid_create(zfsvfs
,
3194 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
3195 if (new_kuid
!= KUID_TO_SUID(ZTOI(zp
)->i_uid
) &&
3196 zfs_id_overquota(zfsvfs
, DMU_USERUSED_OBJECT
,
3200 err
= SET_ERROR(EDQUOT
);
3205 if (mask
& ATTR_GID
) {
3206 new_kgid
= zfs_fuid_create(zfsvfs
,
3207 (uint64_t)vap
->va_gid
, cr
, ZFS_GROUP
, &fuidp
);
3208 if (new_kgid
!= KGID_TO_SGID(ZTOI(zp
)->i_gid
) &&
3209 zfs_id_overquota(zfsvfs
, DMU_GROUPUSED_OBJECT
,
3213 err
= SET_ERROR(EDQUOT
);
3218 if (projid
!= ZFS_INVALID_PROJID
&&
3219 zfs_id_overquota(zfsvfs
, DMU_PROJECTUSED_OBJECT
, projid
)) {
3226 tx
= dmu_tx_create(os
);
3228 if (mask
& ATTR_MODE
) {
3229 uint64_t pmode
= zp
->z_mode
;
3231 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
3233 zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
);
3235 mutex_enter(&zp
->z_lock
);
3236 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
3238 * Are we upgrading ACL from old V0 format
3241 if (zfsvfs
->z_version
>= ZPL_VERSION_FUID
&&
3242 zfs_znode_acl_version(zp
) ==
3243 ZFS_ACL_VERSION_INITIAL
) {
3244 dmu_tx_hold_free(tx
, acl_obj
, 0,
3246 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3247 0, aclp
->z_acl_bytes
);
3249 dmu_tx_hold_write(tx
, acl_obj
, 0,
3252 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3253 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3254 0, aclp
->z_acl_bytes
);
3256 mutex_exit(&zp
->z_lock
);
3257 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3259 if (((mask
& ATTR_XVATTR
) &&
3260 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
)) ||
3261 (projid
!= ZFS_INVALID_PROJID
&&
3262 !(zp
->z_pflags
& ZFS_PROJID
)))
3263 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3265 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3269 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
3272 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3274 zfs_fuid_txhold(zfsvfs
, tx
);
3276 zfs_sa_upgrade_txholds(tx
, zp
);
3278 err
= dmu_tx_assign(tx
, TXG_WAIT
);
3284 * Set each attribute requested.
3285 * We group settings according to the locks they need to acquire.
3287 * Note: you cannot set ctime directly, although it will be
3288 * updated as a side-effect of calling this function.
3291 if (projid
!= ZFS_INVALID_PROJID
&& !(zp
->z_pflags
& ZFS_PROJID
)) {
3293 * For the existed object that is upgraded from old system,
3294 * its on-disk layout has no slot for the project ID attribute.
3295 * But quota accounting logic needs to access related slots by
3296 * offset directly. So we need to adjust old objects' layout
3297 * to make the project ID to some unified and fixed offset.
3300 err
= sa_add_projid(attrzp
->z_sa_hdl
, tx
, projid
);
3302 err
= sa_add_projid(zp
->z_sa_hdl
, tx
, projid
);
3304 if (unlikely(err
== EEXIST
))
3309 projid
= ZFS_INVALID_PROJID
;
3312 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3313 mutex_enter(&zp
->z_acl_lock
);
3314 mutex_enter(&zp
->z_lock
);
3316 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
3317 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
3320 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3321 mutex_enter(&attrzp
->z_acl_lock
);
3322 mutex_enter(&attrzp
->z_lock
);
3323 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3324 SA_ZPL_FLAGS(zfsvfs
), NULL
, &attrzp
->z_pflags
,
3325 sizeof (attrzp
->z_pflags
));
3326 if (projid
!= ZFS_INVALID_PROJID
) {
3327 attrzp
->z_projid
= projid
;
3328 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3329 SA_ZPL_PROJID(zfsvfs
), NULL
, &attrzp
->z_projid
,
3330 sizeof (attrzp
->z_projid
));
3334 if (mask
& (ATTR_UID
|ATTR_GID
)) {
3336 if (mask
& ATTR_UID
) {
3337 ZTOI(zp
)->i_uid
= SUID_TO_KUID(new_kuid
);
3338 new_uid
= zfs_uid_read(ZTOI(zp
));
3339 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zfsvfs
), NULL
,
3340 &new_uid
, sizeof (new_uid
));
3342 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3343 SA_ZPL_UID(zfsvfs
), NULL
, &new_uid
,
3345 ZTOI(attrzp
)->i_uid
= SUID_TO_KUID(new_uid
);
3349 if (mask
& ATTR_GID
) {
3350 ZTOI(zp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3351 new_gid
= zfs_gid_read(ZTOI(zp
));
3352 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zfsvfs
),
3353 NULL
, &new_gid
, sizeof (new_gid
));
3355 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3356 SA_ZPL_GID(zfsvfs
), NULL
, &new_gid
,
3358 ZTOI(attrzp
)->i_gid
= SGID_TO_KGID(new_kgid
);
3361 if (!(mask
& ATTR_MODE
)) {
3362 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
),
3363 NULL
, &new_mode
, sizeof (new_mode
));
3364 new_mode
= zp
->z_mode
;
3366 err
= zfs_acl_chown_setattr(zp
);
3369 err
= zfs_acl_chown_setattr(attrzp
);
3374 if (mask
& ATTR_MODE
) {
3375 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
), NULL
,
3376 &new_mode
, sizeof (new_mode
));
3377 zp
->z_mode
= ZTOI(zp
)->i_mode
= new_mode
;
3378 ASSERT3P(aclp
, !=, NULL
);
3379 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
3381 if (zp
->z_acl_cached
)
3382 zfs_acl_free(zp
->z_acl_cached
);
3383 zp
->z_acl_cached
= aclp
;
3387 if ((mask
& ATTR_ATIME
) || zp
->z_atime_dirty
) {
3388 zp
->z_atime_dirty
= 0;
3389 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
3390 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
,
3391 &atime
, sizeof (atime
));
3394 if (mask
& (ATTR_MTIME
| ATTR_SIZE
)) {
3395 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3396 ZTOI(zp
)->i_mtime
= zpl_inode_timespec_trunc(vap
->va_mtime
,
3397 ZTOI(zp
)->i_sb
->s_time_gran
);
3399 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
3400 mtime
, sizeof (mtime
));
3403 if (mask
& (ATTR_CTIME
| ATTR_SIZE
)) {
3404 ZFS_TIME_ENCODE(&vap
->va_ctime
, ctime
);
3405 ZTOI(zp
)->i_ctime
= zpl_inode_timespec_trunc(vap
->va_ctime
,
3406 ZTOI(zp
)->i_sb
->s_time_gran
);
3407 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3408 ctime
, sizeof (ctime
));
3411 if (projid
!= ZFS_INVALID_PROJID
) {
3412 zp
->z_projid
= projid
;
3413 SA_ADD_BULK_ATTR(bulk
, count
,
3414 SA_ZPL_PROJID(zfsvfs
), NULL
, &zp
->z_projid
,
3415 sizeof (zp
->z_projid
));
3418 if (attrzp
&& mask
) {
3419 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3420 SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
,
3425 * Do this after setting timestamps to prevent timestamp
3426 * update from toggling bit
3429 if (xoap
&& (mask
& ATTR_XVATTR
)) {
3432 * restore trimmed off masks
3433 * so that return masks can be set for caller.
3436 if (XVA_ISSET_REQ(tmpxvattr
, XAT_APPENDONLY
)) {
3437 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3439 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NOUNLINK
)) {
3440 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3442 if (XVA_ISSET_REQ(tmpxvattr
, XAT_IMMUTABLE
)) {
3443 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3445 if (XVA_ISSET_REQ(tmpxvattr
, XAT_NODUMP
)) {
3446 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3448 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_MODIFIED
)) {
3449 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3451 if (XVA_ISSET_REQ(tmpxvattr
, XAT_AV_QUARANTINED
)) {
3452 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3454 if (XVA_ISSET_REQ(tmpxvattr
, XAT_PROJINHERIT
)) {
3455 XVA_SET_REQ(xvap
, XAT_PROJINHERIT
);
3458 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3459 ASSERT(S_ISREG(ip
->i_mode
));
3461 zfs_xvattr_set(zp
, xvap
, tx
);
3465 zfs_fuid_sync(zfsvfs
, tx
);
3468 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3470 mutex_exit(&zp
->z_lock
);
3471 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3472 mutex_exit(&zp
->z_acl_lock
);
3475 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
))
3476 mutex_exit(&attrzp
->z_acl_lock
);
3477 mutex_exit(&attrzp
->z_lock
);
3480 if (err
== 0 && xattr_count
> 0) {
3481 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3490 zfs_fuid_info_free(fuidp
);
3498 if (err
== ERESTART
)
3502 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3505 if (err2
== 0 && handle_eadir
)
3506 err2
= zfs_setattr_dir(attrzp
);
3509 zfs_inode_update(zp
);
3513 if (os
->os_sync
== ZFS_SYNC_ALWAYS
)
3514 zil_commit(zilog
, 0);
3517 kmem_free(xattr_bulk
, sizeof (sa_bulk_attr_t
) * bulks
);
3518 kmem_free(bulk
, sizeof (sa_bulk_attr_t
) * bulks
);
3519 kmem_free(tmpxvattr
, sizeof (xvattr_t
));
3524 typedef struct zfs_zlock
{
3525 krwlock_t
*zl_rwlock
; /* lock we acquired */
3526 znode_t
*zl_znode
; /* znode we held */
3527 struct zfs_zlock
*zl_next
; /* next in list */
3531 * Drop locks and release vnodes that were held by zfs_rename_lock().
3534 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3538 while ((zl
= *zlpp
) != NULL
) {
3539 if (zl
->zl_znode
!= NULL
)
3540 zfs_iput_async(ZTOI(zl
->zl_znode
));
3541 rw_exit(zl
->zl_rwlock
);
3542 *zlpp
= zl
->zl_next
;
3543 kmem_free(zl
, sizeof (*zl
));
3548 * Search back through the directory tree, using the ".." entries.
3549 * Lock each directory in the chain to prevent concurrent renames.
3550 * Fail any attempt to move a directory into one of its own descendants.
3551 * XXX - z_parent_lock can overlap with map or grow locks
3554 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3558 uint64_t rootid
= ZTOZSB(zp
)->z_root
;
3559 uint64_t oidp
= zp
->z_id
;
3560 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3561 krw_t rw
= RW_WRITER
;
3564 * First pass write-locks szp and compares to zp->z_id.
3565 * Later passes read-lock zp and compare to zp->z_parent.
3568 if (!rw_tryenter(rwlp
, rw
)) {
3570 * Another thread is renaming in this path.
3571 * Note that if we are a WRITER, we don't have any
3572 * parent_locks held yet.
3574 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3576 * Drop our locks and restart
3578 zfs_rename_unlock(&zl
);
3582 rwlp
= &szp
->z_parent_lock
;
3587 * Wait for other thread to drop its locks
3593 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3594 zl
->zl_rwlock
= rwlp
;
3595 zl
->zl_znode
= NULL
;
3596 zl
->zl_next
= *zlpp
;
3599 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3600 return (SET_ERROR(EINVAL
));
3602 if (oidp
== rootid
) /* We've hit the top */
3605 if (rw
== RW_READER
) { /* i.e. not the first pass */
3606 int error
= zfs_zget(ZTOZSB(zp
), oidp
, &zp
);
3611 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(ZTOZSB(zp
)),
3612 &oidp
, sizeof (oidp
));
3613 rwlp
= &zp
->z_parent_lock
;
3616 } while (zp
->z_id
!= sdzp
->z_id
);
3622 * Move an entry from the provided source directory to the target
3623 * directory. Change the entry name as indicated.
3625 * IN: sdip - Source directory containing the "old entry".
3626 * snm - Old entry name.
3627 * tdip - Target directory to contain the "new entry".
3628 * tnm - New entry name.
3629 * cr - credentials of caller.
3630 * flags - case flags
3632 * RETURN: 0 on success, error code on failure.
3635 * sdip,tdip - ctime|mtime updated
3639 zfs_rename(struct inode
*sdip
, char *snm
, struct inode
*tdip
, char *tnm
,
3640 cred_t
*cr
, int flags
)
3642 znode_t
*tdzp
, *szp
, *tzp
;
3643 znode_t
*sdzp
= ITOZ(sdip
);
3644 zfsvfs_t
*zfsvfs
= ITOZSB(sdip
);
3646 zfs_dirlock_t
*sdl
, *tdl
;
3649 int cmp
, serr
, terr
;
3652 boolean_t waited
= B_FALSE
;
3654 if (snm
== NULL
|| tnm
== NULL
)
3655 return (SET_ERROR(EINVAL
));
3658 ZFS_VERIFY_ZP(sdzp
);
3659 zilog
= zfsvfs
->z_log
;
3662 ZFS_VERIFY_ZP(tdzp
);
3665 * We check i_sb because snapshots and the ctldir must have different
3668 if (tdip
->i_sb
!= sdip
->i_sb
|| zfsctl_is_node(tdip
)) {
3670 return (SET_ERROR(EXDEV
));
3673 if (zfsvfs
->z_utf8
&& u8_validate(tnm
,
3674 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3676 return (SET_ERROR(EILSEQ
));
3679 if (flags
& FIGNORECASE
)
3688 * This is to prevent the creation of links into attribute space
3689 * by renaming a linked file into/outof an attribute directory.
3690 * See the comment in zfs_link() for why this is considered bad.
3692 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3694 return (SET_ERROR(EINVAL
));
3698 * Lock source and target directory entries. To prevent deadlock,
3699 * a lock ordering must be defined. We lock the directory with
3700 * the smallest object id first, or if it's a tie, the one with
3701 * the lexically first name.
3703 if (sdzp
->z_id
< tdzp
->z_id
) {
3705 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3709 * First compare the two name arguments without
3710 * considering any case folding.
3712 int nofold
= (zfsvfs
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3714 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3715 ASSERT(error
== 0 || !zfsvfs
->z_utf8
);
3718 * POSIX: "If the old argument and the new argument
3719 * both refer to links to the same existing file,
3720 * the rename() function shall return successfully
3721 * and perform no other action."
3727 * If the file system is case-folding, then we may
3728 * have some more checking to do. A case-folding file
3729 * system is either supporting mixed case sensitivity
3730 * access or is completely case-insensitive. Note
3731 * that the file system is always case preserving.
3733 * In mixed sensitivity mode case sensitive behavior
3734 * is the default. FIGNORECASE must be used to
3735 * explicitly request case insensitive behavior.
3737 * If the source and target names provided differ only
3738 * by case (e.g., a request to rename 'tim' to 'Tim'),
3739 * we will treat this as a special case in the
3740 * case-insensitive mode: as long as the source name
3741 * is an exact match, we will allow this to proceed as
3742 * a name-change request.
3744 if ((zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
||
3745 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&&
3746 flags
& FIGNORECASE
)) &&
3747 u8_strcmp(snm
, tnm
, 0, zfsvfs
->z_norm
, U8_UNICODE_LATEST
,
3750 * case preserving rename request, require exact
3759 * If the source and destination directories are the same, we should
3760 * grab the z_name_lock of that directory only once.
3764 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3768 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3769 ZEXISTS
| zflg
, NULL
, NULL
);
3770 terr
= zfs_dirent_lock(&tdl
,
3771 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3773 terr
= zfs_dirent_lock(&tdl
,
3774 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3775 serr
= zfs_dirent_lock(&sdl
,
3776 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3782 * Source entry invalid or not there.
3785 zfs_dirent_unlock(tdl
);
3791 rw_exit(&sdzp
->z_name_lock
);
3793 if (strcmp(snm
, "..") == 0)
3799 zfs_dirent_unlock(sdl
);
3803 rw_exit(&sdzp
->z_name_lock
);
3805 if (strcmp(tnm
, "..") == 0)
3812 * If we are using project inheritance, means if the directory has
3813 * ZFS_PROJINHERIT set, then its descendant directories will inherit
3814 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
3815 * such case, we only allow renames into our tree when the project
3818 if (tdzp
->z_pflags
& ZFS_PROJINHERIT
&&
3819 tdzp
->z_projid
!= szp
->z_projid
) {
3820 error
= SET_ERROR(EXDEV
);
3825 * Must have write access at the source to remove the old entry
3826 * and write access at the target to create the new entry.
3827 * Note that if target and source are the same, this can be
3828 * done in a single check.
3831 if ((error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
)))
3834 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3836 * Check to make sure rename is valid.
3837 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3839 if ((error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
)))
3844 * Does target exist?
3848 * Source and target must be the same type.
3850 if (S_ISDIR(ZTOI(szp
)->i_mode
)) {
3851 if (!S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3852 error
= SET_ERROR(ENOTDIR
);
3856 if (S_ISDIR(ZTOI(tzp
)->i_mode
)) {
3857 error
= SET_ERROR(EISDIR
);
3862 * POSIX dictates that when the source and target
3863 * entries refer to the same file object, rename
3864 * must do nothing and exit without error.
3866 if (szp
->z_id
== tzp
->z_id
) {
3872 tx
= dmu_tx_create(zfsvfs
->z_os
);
3873 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3874 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3875 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3876 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3878 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3879 zfs_sa_upgrade_txholds(tx
, tdzp
);
3882 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3883 zfs_sa_upgrade_txholds(tx
, tzp
);
3886 zfs_sa_upgrade_txholds(tx
, szp
);
3887 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
3888 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
3891 zfs_rename_unlock(&zl
);
3892 zfs_dirent_unlock(sdl
);
3893 zfs_dirent_unlock(tdl
);
3896 rw_exit(&sdzp
->z_name_lock
);
3898 if (error
== ERESTART
) {
3915 if (tzp
) /* Attempt to remove the existing target */
3916 error
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3919 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3921 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3922 if (tdzp
->z_pflags
& ZFS_PROJINHERIT
)
3923 szp
->z_pflags
|= ZFS_PROJINHERIT
;
3925 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zfsvfs
),
3926 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3929 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3931 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3932 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3933 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3936 * At this point, we have successfully created
3937 * the target name, but have failed to remove
3938 * the source name. Since the create was done
3939 * with the ZRENAMING flag, there are
3940 * complications; for one, the link count is
3941 * wrong. The easiest way to deal with this
3942 * is to remove the newly created target, and
3943 * return the original error. This must
3944 * succeed; fortunately, it is very unlikely to
3945 * fail, since we just created it.
3947 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3948 ZRENAMING
, NULL
), ==, 0);
3952 * If we had removed the existing target, subsequent
3953 * call to zfs_link_create() to add back the same entry
3954 * but, the new dnode (szp) should not fail.
3956 ASSERT(tzp
== NULL
);
3963 zfs_rename_unlock(&zl
);
3965 zfs_dirent_unlock(sdl
);
3966 zfs_dirent_unlock(tdl
);
3968 zfs_inode_update(sdzp
);
3970 rw_exit(&sdzp
->z_name_lock
);
3973 zfs_inode_update(tdzp
);
3975 zfs_inode_update(szp
);
3978 zfs_inode_update(tzp
);
3982 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3983 zil_commit(zilog
, 0);
3990 * Insert the indicated symbolic reference entry into the directory.
3992 * IN: dip - Directory to contain new symbolic link.
3993 * link - Name for new symlink entry.
3994 * vap - Attributes of new entry.
3995 * target - Target path of new symlink.
3997 * cr - credentials of caller.
3998 * flags - case flags
4000 * RETURN: 0 on success, error code on failure.
4003 * dip - ctime|mtime updated
4007 zfs_symlink(struct inode
*dip
, char *name
, vattr_t
*vap
, char *link
,
4008 struct inode
**ipp
, cred_t
*cr
, int flags
)
4010 znode_t
*zp
, *dzp
= ITOZ(dip
);
4013 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
4015 uint64_t len
= strlen(link
);
4018 zfs_acl_ids_t acl_ids
;
4019 boolean_t fuid_dirtied
;
4020 uint64_t txtype
= TX_SYMLINK
;
4021 boolean_t waited
= B_FALSE
;
4023 ASSERT(S_ISLNK(vap
->va_mode
));
4026 return (SET_ERROR(EINVAL
));
4030 zilog
= zfsvfs
->z_log
;
4032 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
4033 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
4035 return (SET_ERROR(EILSEQ
));
4037 if (flags
& FIGNORECASE
)
4040 if (len
> MAXPATHLEN
) {
4042 return (SET_ERROR(ENAMETOOLONG
));
4045 if ((error
= zfs_acl_ids_create(dzp
, 0,
4046 vap
, cr
, NULL
, &acl_ids
)) != 0) {
4054 * Attempt to lock directory; fail if entry already exists.
4056 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
4058 zfs_acl_ids_free(&acl_ids
);
4063 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
4064 zfs_acl_ids_free(&acl_ids
);
4065 zfs_dirent_unlock(dl
);
4070 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
, ZFS_DEFAULT_PROJID
)) {
4071 zfs_acl_ids_free(&acl_ids
);
4072 zfs_dirent_unlock(dl
);
4074 return (SET_ERROR(EDQUOT
));
4076 tx
= dmu_tx_create(zfsvfs
->z_os
);
4077 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
4078 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
4079 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
4080 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
4081 ZFS_SA_BASE_ATTR_SIZE
+ len
);
4082 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
4083 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
4084 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
4085 acl_ids
.z_aclp
->z_acl_bytes
);
4088 zfs_fuid_txhold(zfsvfs
, tx
);
4089 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
4091 zfs_dirent_unlock(dl
);
4092 if (error
== ERESTART
) {
4098 zfs_acl_ids_free(&acl_ids
);
4105 * Create a new object for the symlink.
4106 * for version 4 ZPL datsets the symlink will be an SA attribute
4108 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
4111 zfs_fuid_sync(zfsvfs
, tx
);
4113 mutex_enter(&zp
->z_lock
);
4115 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zfsvfs
),
4118 zfs_sa_symlink(zp
, link
, len
, tx
);
4119 mutex_exit(&zp
->z_lock
);
4122 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
4123 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
4125 * Insert the new object into the directory.
4127 error
= zfs_link_create(dl
, zp
, tx
, ZNEW
);
4129 zfs_znode_delete(zp
, tx
);
4130 remove_inode_hash(ZTOI(zp
));
4132 if (flags
& FIGNORECASE
)
4134 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
4136 zfs_inode_update(dzp
);
4137 zfs_inode_update(zp
);
4140 zfs_acl_ids_free(&acl_ids
);
4144 zfs_dirent_unlock(dl
);
4149 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4150 zil_commit(zilog
, 0);
4160 * Return, in the buffer contained in the provided uio structure,
4161 * the symbolic path referred to by ip.
4163 * IN: ip - inode of symbolic link
4164 * uio - structure to contain the link path.
4165 * cr - credentials of caller.
4167 * RETURN: 0 if success
4168 * error code if failure
4171 * ip - atime updated
4175 zfs_readlink(struct inode
*ip
, uio_t
*uio
, cred_t
*cr
)
4177 znode_t
*zp
= ITOZ(ip
);
4178 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4184 mutex_enter(&zp
->z_lock
);
4186 error
= sa_lookup_uio(zp
->z_sa_hdl
,
4187 SA_ZPL_SYMLINK(zfsvfs
), uio
);
4189 error
= zfs_sa_readlink(zp
, uio
);
4190 mutex_exit(&zp
->z_lock
);
4197 * Insert a new entry into directory tdip referencing sip.
4199 * IN: tdip - Directory to contain new entry.
4200 * sip - inode of new entry.
4201 * name - name of new entry.
4202 * cr - credentials of caller.
4204 * RETURN: 0 if success
4205 * error code if failure
4208 * tdip - ctime|mtime updated
4209 * sip - ctime updated
4213 zfs_link(struct inode
*tdip
, struct inode
*sip
, char *name
, cred_t
*cr
,
4216 znode_t
*dzp
= ITOZ(tdip
);
4218 zfsvfs_t
*zfsvfs
= ITOZSB(tdip
);
4226 boolean_t waited
= B_FALSE
;
4227 boolean_t is_tmpfile
= 0;
4230 is_tmpfile
= (sip
->i_nlink
== 0 && (sip
->i_state
& I_LINKABLE
));
4232 ASSERT(S_ISDIR(tdip
->i_mode
));
4235 return (SET_ERROR(EINVAL
));
4239 zilog
= zfsvfs
->z_log
;
4242 * POSIX dictates that we return EPERM here.
4243 * Better choices include ENOTSUP or EISDIR.
4245 if (S_ISDIR(sip
->i_mode
)) {
4247 return (SET_ERROR(EPERM
));
4254 * If we are using project inheritance, means if the directory has
4255 * ZFS_PROJINHERIT set, then its descendant directories will inherit
4256 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under
4257 * such case, we only allow hard link creation in our tree when the
4258 * project IDs are the same.
4260 if (dzp
->z_pflags
& ZFS_PROJINHERIT
&& dzp
->z_projid
!= szp
->z_projid
) {
4262 return (SET_ERROR(EXDEV
));
4266 * We check i_sb because snapshots and the ctldir must have different
4269 if (sip
->i_sb
!= tdip
->i_sb
|| zfsctl_is_node(sip
)) {
4271 return (SET_ERROR(EXDEV
));
4274 /* Prevent links to .zfs/shares files */
4276 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
4277 &parent
, sizeof (uint64_t))) != 0) {
4281 if (parent
== zfsvfs
->z_shares_dir
) {
4283 return (SET_ERROR(EPERM
));
4286 if (zfsvfs
->z_utf8
&& u8_validate(name
,
4287 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
4289 return (SET_ERROR(EILSEQ
));
4291 if (flags
& FIGNORECASE
)
4295 * We do not support links between attributes and non-attributes
4296 * because of the potential security risk of creating links
4297 * into "normal" file space in order to circumvent restrictions
4298 * imposed in attribute space.
4300 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
4302 return (SET_ERROR(EINVAL
));
4305 owner
= zfs_fuid_map_id(zfsvfs
, KUID_TO_SUID(sip
->i_uid
),
4307 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
4309 return (SET_ERROR(EPERM
));
4312 if ((error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
))) {
4319 * Attempt to lock directory; fail if entry already exists.
4321 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
4327 tx
= dmu_tx_create(zfsvfs
->z_os
);
4328 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
4329 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
4331 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
4333 zfs_sa_upgrade_txholds(tx
, szp
);
4334 zfs_sa_upgrade_txholds(tx
, dzp
);
4335 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
4337 zfs_dirent_unlock(dl
);
4338 if (error
== ERESTART
) {
4348 /* unmark z_unlinked so zfs_link_create will not reject */
4350 szp
->z_unlinked
= 0;
4351 error
= zfs_link_create(dl
, szp
, tx
, 0);
4354 uint64_t txtype
= TX_LINK
;
4356 * tmpfile is created to be in z_unlinkedobj, so remove it.
4357 * Also, we don't log in ZIL, be cause all previous file
4358 * operation on the tmpfile are ignored by ZIL. Instead we
4359 * always wait for txg to sync to make sure all previous
4360 * operation are sync safe.
4363 VERIFY(zap_remove_int(zfsvfs
->z_os
,
4364 zfsvfs
->z_unlinkedobj
, szp
->z_id
, tx
) == 0);
4366 if (flags
& FIGNORECASE
)
4368 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
4370 } else if (is_tmpfile
) {
4371 /* restore z_unlinked since when linking failed */
4372 szp
->z_unlinked
= 1;
4374 txg
= dmu_tx_get_txg(tx
);
4377 zfs_dirent_unlock(dl
);
4379 if (!is_tmpfile
&& zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4380 zil_commit(zilog
, 0);
4383 txg_wait_synced(dmu_objset_pool(zfsvfs
->z_os
), txg
);
4385 zfs_inode_update(dzp
);
4386 zfs_inode_update(szp
);
4392 zfs_putpage_commit_cb(void *arg
)
4394 struct page
*pp
= arg
;
4397 end_page_writeback(pp
);
4401 * Push a page out to disk, once the page is on stable storage the
4402 * registered commit callback will be run as notification of completion.
4404 * IN: ip - page mapped for inode.
4405 * pp - page to push (page is locked)
4406 * wbc - writeback control data
4408 * RETURN: 0 if success
4409 * error code if failure
4412 * ip - ctime|mtime updated
4416 zfs_putpage(struct inode
*ip
, struct page
*pp
, struct writeback_control
*wbc
)
4418 znode_t
*zp
= ITOZ(ip
);
4419 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4426 uint64_t mtime
[2], ctime
[2];
4427 sa_bulk_attr_t bulk
[3];
4429 struct address_space
*mapping
;
4434 ASSERT(PageLocked(pp
));
4436 pgoff
= page_offset(pp
); /* Page byte-offset in file */
4437 offset
= i_size_read(ip
); /* File length in bytes */
4438 pglen
= MIN(PAGE_SIZE
, /* Page length in bytes */
4439 P2ROUNDUP(offset
, PAGE_SIZE
)-pgoff
);
4441 /* Page is beyond end of file */
4442 if (pgoff
>= offset
) {
4448 /* Truncate page length to end of file */
4449 if (pgoff
+ pglen
> offset
)
4450 pglen
= offset
- pgoff
;
4454 * FIXME: Allow mmap writes past its quota. The correct fix
4455 * is to register a page_mkwrite() handler to count the page
4456 * against its quota when it is about to be dirtied.
4458 if (zfs_id_overblockquota(zfsvfs
, DMU_USERUSED_OBJECT
,
4459 KUID_TO_SUID(ip
->i_uid
)) ||
4460 zfs_id_overblockquota(zfsvfs
, DMU_GROUPUSED_OBJECT
,
4461 KGID_TO_SGID(ip
->i_gid
)) ||
4462 (zp
->z_projid
!= ZFS_DEFAULT_PROJID
&&
4463 zfs_id_overblockquota(zfsvfs
, DMU_PROJECTUSED_OBJECT
,
4470 * The ordering here is critical and must adhere to the following
4471 * rules in order to avoid deadlocking in either zfs_read() or
4472 * zfs_free_range() due to a lock inversion.
4474 * 1) The page must be unlocked prior to acquiring the range lock.
4475 * This is critical because zfs_read() calls find_lock_page()
4476 * which may block on the page lock while holding the range lock.
4478 * 2) Before setting or clearing write back on a page the range lock
4479 * must be held in order to prevent a lock inversion with the
4480 * zfs_free_range() function.
4482 * This presents a problem because upon entering this function the
4483 * page lock is already held. To safely acquire the range lock the
4484 * page lock must be dropped. This creates a window where another
4485 * process could truncate, invalidate, dirty, or write out the page.
4487 * Therefore, after successfully reacquiring the range and page locks
4488 * the current page state is checked. In the common case everything
4489 * will be as is expected and it can be written out. However, if
4490 * the page state has changed it must be handled accordingly.
4492 mapping
= pp
->mapping
;
4493 redirty_page_for_writepage(wbc
, pp
);
4496 locked_range_t
*lr
= rangelock_enter(&zp
->z_rangelock
,
4497 pgoff
, pglen
, RL_WRITER
);
4500 /* Page mapping changed or it was no longer dirty, we're done */
4501 if (unlikely((mapping
!= pp
->mapping
) || !PageDirty(pp
))) {
4508 /* Another process started write block if required */
4509 if (PageWriteback(pp
)) {
4513 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
4514 wait_on_page_writeback(pp
);
4520 /* Clear the dirty flag the required locks are held */
4521 if (!clear_page_dirty_for_io(pp
)) {
4529 * Counterpart for redirty_page_for_writepage() above. This page
4530 * was in fact not skipped and should not be counted as if it were.
4532 wbc
->pages_skipped
--;
4533 set_page_writeback(pp
);
4536 tx
= dmu_tx_create(zfsvfs
->z_os
);
4537 dmu_tx_hold_write(tx
, zp
->z_id
, pgoff
, pglen
);
4538 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4539 zfs_sa_upgrade_txholds(tx
, zp
);
4541 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4543 if (err
== ERESTART
)
4547 __set_page_dirty_nobuffers(pp
);
4549 end_page_writeback(pp
);
4556 ASSERT3U(pglen
, <=, PAGE_SIZE
);
4557 dmu_write(zfsvfs
->z_os
, zp
->z_id
, pgoff
, pglen
, va
, tx
);
4560 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
4561 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
4562 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
4565 /* Preserve the mtime and ctime provided by the inode */
4566 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4567 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4568 zp
->z_atime_dirty
= 0;
4571 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4573 zfs_log_write(zfsvfs
->z_log
, tx
, TX_WRITE
, zp
, pgoff
, pglen
, 0,
4574 zfs_putpage_commit_cb
, pp
);
4579 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
4581 * Note that this is rarely called under writepages(), because
4582 * writepages() normally handles the entire commit for
4583 * performance reasons.
4585 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
4593 * Update the system attributes when the inode has been dirtied. For the
4594 * moment we only update the mode, atime, mtime, and ctime.
4597 zfs_dirty_inode(struct inode
*ip
, int flags
)
4599 znode_t
*zp
= ITOZ(ip
);
4600 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4602 uint64_t mode
, atime
[2], mtime
[2], ctime
[2];
4603 sa_bulk_attr_t bulk
[4];
4607 if (zfs_is_readonly(zfsvfs
) || dmu_objset_is_snapshot(zfsvfs
->z_os
))
4615 * This is the lazytime semantic indroduced in Linux 4.0
4616 * This flag will only be called from update_time when lazytime is set.
4617 * (Note, I_DIRTY_SYNC will also set if not lazytime)
4618 * Fortunately mtime and ctime are managed within ZFS itself, so we
4619 * only need to dirty atime.
4621 if (flags
== I_DIRTY_TIME
) {
4622 zp
->z_atime_dirty
= 1;
4627 tx
= dmu_tx_create(zfsvfs
->z_os
);
4629 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4630 zfs_sa_upgrade_txholds(tx
, zp
);
4632 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4638 mutex_enter(&zp
->z_lock
);
4639 zp
->z_atime_dirty
= 0;
4641 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MODE(zfsvfs
), NULL
, &mode
, 8);
4642 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_ATIME(zfsvfs
), NULL
, &atime
, 16);
4643 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
4644 SA_ADD_BULK_ATTR(bulk
, cnt
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
4646 /* Preserve the mode, mtime and ctime provided by the inode */
4647 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4648 ZFS_TIME_ENCODE(&ip
->i_mtime
, mtime
);
4649 ZFS_TIME_ENCODE(&ip
->i_ctime
, ctime
);
4654 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, cnt
, tx
);
4655 mutex_exit(&zp
->z_lock
);
4665 zfs_inactive(struct inode
*ip
)
4667 znode_t
*zp
= ITOZ(ip
);
4668 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4671 int need_unlock
= 0;
4673 /* Only read lock if we haven't already write locked, e.g. rollback */
4674 if (!RW_WRITE_HELD(&zfsvfs
->z_teardown_inactive_lock
)) {
4676 rw_enter(&zfsvfs
->z_teardown_inactive_lock
, RW_READER
);
4678 if (zp
->z_sa_hdl
== NULL
) {
4680 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4684 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4685 dmu_tx_t
*tx
= dmu_tx_create(zfsvfs
->z_os
);
4687 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4688 zfs_sa_upgrade_txholds(tx
, zp
);
4689 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4693 ZFS_TIME_ENCODE(&ip
->i_atime
, atime
);
4694 mutex_enter(&zp
->z_lock
);
4695 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zfsvfs
),
4696 (void *)&atime
, sizeof (atime
), tx
);
4697 zp
->z_atime_dirty
= 0;
4698 mutex_exit(&zp
->z_lock
);
4705 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4709 * Bounds-check the seek operation.
4711 * IN: ip - inode seeking within
4712 * ooff - old file offset
4713 * noffp - pointer to new file offset
4714 * ct - caller context
4716 * RETURN: 0 if success
4717 * EINVAL if new offset invalid
4721 zfs_seek(struct inode
*ip
, offset_t ooff
, offset_t
*noffp
)
4723 if (S_ISDIR(ip
->i_mode
))
4725 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4729 * Fill pages with data from the disk.
4732 zfs_fillpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4734 znode_t
*zp
= ITOZ(ip
);
4735 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4737 struct page
*cur_pp
;
4738 u_offset_t io_off
, total
;
4745 io_len
= nr_pages
<< PAGE_SHIFT
;
4746 i_size
= i_size_read(ip
);
4747 io_off
= page_offset(pl
[0]);
4749 if (io_off
+ io_len
> i_size
)
4750 io_len
= i_size
- io_off
;
4753 * Iterate over list of pages and read each page individually.
4756 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4759 cur_pp
= pl
[page_idx
++];
4761 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4765 /* convert checksum errors into IO errors */
4767 err
= SET_ERROR(EIO
);
4776 * Uses zfs_fillpage to read data from the file and fill the pages.
4778 * IN: ip - inode of file to get data from.
4779 * pl - list of pages to read
4780 * nr_pages - number of pages to read
4782 * RETURN: 0 on success, error code on failure.
4785 * vp - atime updated
4789 zfs_getpage(struct inode
*ip
, struct page
*pl
[], int nr_pages
)
4791 znode_t
*zp
= ITOZ(ip
);
4792 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4801 err
= zfs_fillpage(ip
, pl
, nr_pages
);
4808 * Check ZFS specific permissions to memory map a section of a file.
4810 * IN: ip - inode of the file to mmap
4812 * addrp - start address in memory region
4813 * len - length of memory region
4814 * vm_flags- address flags
4816 * RETURN: 0 if success
4817 * error code if failure
4821 zfs_map(struct inode
*ip
, offset_t off
, caddr_t
*addrp
, size_t len
,
4822 unsigned long vm_flags
)
4824 znode_t
*zp
= ITOZ(ip
);
4825 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4830 if ((vm_flags
& VM_WRITE
) && (zp
->z_pflags
&
4831 (ZFS_IMMUTABLE
| ZFS_READONLY
| ZFS_APPENDONLY
))) {
4833 return (SET_ERROR(EPERM
));
4836 if ((vm_flags
& (VM_READ
| VM_EXEC
)) &&
4837 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4839 return (SET_ERROR(EACCES
));
4842 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4844 return (SET_ERROR(ENXIO
));
4852 * convoff - converts the given data (start, whence) to the
4856 convoff(struct inode
*ip
, flock64_t
*lckdat
, int whence
, offset_t offset
)
4861 if ((lckdat
->l_whence
== 2) || (whence
== 2)) {
4862 if ((error
= zfs_getattr(ip
, &vap
, 0, CRED())))
4866 switch (lckdat
->l_whence
) {
4868 lckdat
->l_start
+= offset
;
4871 lckdat
->l_start
+= vap
.va_size
;
4876 return (SET_ERROR(EINVAL
));
4879 if (lckdat
->l_start
< 0)
4880 return (SET_ERROR(EINVAL
));
4884 lckdat
->l_start
-= offset
;
4887 lckdat
->l_start
-= vap
.va_size
;
4892 return (SET_ERROR(EINVAL
));
4895 lckdat
->l_whence
= (short)whence
;
4900 * Free or allocate space in a file. Currently, this function only
4901 * supports the `F_FREESP' command. However, this command is somewhat
4902 * misnamed, as its functionality includes the ability to allocate as
4903 * well as free space.
4905 * IN: ip - inode of file to free data in.
4906 * cmd - action to take (only F_FREESP supported).
4907 * bfp - section of file to free/alloc.
4908 * flag - current file open mode flags.
4909 * offset - current file offset.
4910 * cr - credentials of caller [UNUSED].
4912 * RETURN: 0 on success, error code on failure.
4915 * ip - ctime|mtime updated
4919 zfs_space(struct inode
*ip
, int cmd
, flock64_t
*bfp
, int flag
,
4920 offset_t offset
, cred_t
*cr
)
4922 znode_t
*zp
= ITOZ(ip
);
4923 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4930 if (cmd
!= F_FREESP
) {
4932 return (SET_ERROR(EINVAL
));
4936 * Callers might not be able to detect properly that we are read-only,
4937 * so check it explicitly here.
4939 if (zfs_is_readonly(zfsvfs
)) {
4941 return (SET_ERROR(EROFS
));
4944 if ((error
= convoff(ip
, bfp
, 0, offset
))) {
4949 if (bfp
->l_len
< 0) {
4951 return (SET_ERROR(EINVAL
));
4955 * Permissions aren't checked on Solaris because on this OS
4956 * zfs_space() can only be called with an opened file handle.
4957 * On Linux we can get here through truncate_range() which
4958 * operates directly on inodes, so we need to check access rights.
4960 if ((error
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, B_FALSE
, cr
))) {
4966 len
= bfp
->l_len
; /* 0 means from off to end of file */
4968 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4976 zfs_fid(struct inode
*ip
, fid_t
*fidp
)
4978 znode_t
*zp
= ITOZ(ip
);
4979 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
4982 uint64_t object
= zp
->z_id
;
4989 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zfsvfs
),
4990 &gen64
, sizeof (uint64_t))) != 0) {
4995 gen
= (uint32_t)gen64
;
4997 size
= SHORT_FID_LEN
;
4999 zfid
= (zfid_short_t
*)fidp
;
5001 zfid
->zf_len
= size
;
5003 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
5004 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
5006 /* Must have a non-zero generation number to distinguish from .zfs */
5009 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
5010 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
5018 zfs_getsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
5020 znode_t
*zp
= ITOZ(ip
);
5021 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
5023 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
5027 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
5035 zfs_setsecattr(struct inode
*ip
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
)
5037 znode_t
*zp
= ITOZ(ip
);
5038 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
5040 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
5041 zilog_t
*zilog
= zfsvfs
->z_log
;
5046 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
5048 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
5049 zil_commit(zilog
, 0);
5055 #ifdef HAVE_UIO_ZEROCOPY
5057 * Tunable, both must be a power of 2.
5059 * zcr_blksz_min: the smallest read we may consider to loan out an arcbuf
5060 * zcr_blksz_max: if set to less than the file block size, allow loaning out of
5061 * an arcbuf for a partial block read
5063 int zcr_blksz_min
= (1 << 10); /* 1K */
5064 int zcr_blksz_max
= (1 << 17); /* 128K */
5068 zfs_reqzcbuf(struct inode
*ip
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
)
5070 znode_t
*zp
= ITOZ(ip
);
5071 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
5072 int max_blksz
= zfsvfs
->z_max_blksz
;
5073 uio_t
*uio
= &xuio
->xu_uio
;
5074 ssize_t size
= uio
->uio_resid
;
5075 offset_t offset
= uio
->uio_loffset
;
5080 int preamble
, postamble
;
5082 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
5083 return (SET_ERROR(EINVAL
));
5090 * Loan out an arc_buf for write if write size is bigger than
5091 * max_blksz, and the file's block size is also max_blksz.
5094 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
5096 return (SET_ERROR(EINVAL
));
5099 * Caller requests buffers for write before knowing where the
5100 * write offset might be (e.g. NFS TCP write).
5105 preamble
= P2PHASE(offset
, blksz
);
5107 preamble
= blksz
- preamble
;
5112 postamble
= P2PHASE(size
, blksz
);
5115 fullblk
= size
/ blksz
;
5116 (void) dmu_xuio_init(xuio
,
5117 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
5120 * Have to fix iov base/len for partial buffers. They
5121 * currently represent full arc_buf's.
5124 /* data begins in the middle of the arc_buf */
5125 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5128 (void) dmu_xuio_add(xuio
, abuf
,
5129 blksz
- preamble
, preamble
);
5132 for (i
= 0; i
< fullblk
; i
++) {
5133 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5136 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
5140 /* data ends in the middle of the arc_buf */
5141 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5144 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
5149 * Loan out an arc_buf for read if the read size is larger than
5150 * the current file block size. Block alignment is not
5151 * considered. Partial arc_buf will be loaned out for read.
5153 blksz
= zp
->z_blksz
;
5154 if (blksz
< zcr_blksz_min
)
5155 blksz
= zcr_blksz_min
;
5156 if (blksz
> zcr_blksz_max
)
5157 blksz
= zcr_blksz_max
;
5158 /* avoid potential complexity of dealing with it */
5159 if (blksz
> max_blksz
) {
5161 return (SET_ERROR(EINVAL
));
5164 maxsize
= zp
->z_size
- uio
->uio_loffset
;
5170 return (SET_ERROR(EINVAL
));
5175 return (SET_ERROR(EINVAL
));
5178 uio
->uio_extflg
= UIO_XUIO
;
5179 XUIO_XUZC_RW(xuio
) = ioflag
;
5186 zfs_retzcbuf(struct inode
*ip
, xuio_t
*xuio
, cred_t
*cr
)
5190 int ioflag
= XUIO_XUZC_RW(xuio
);
5192 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
5194 i
= dmu_xuio_cnt(xuio
);
5196 abuf
= dmu_xuio_arcbuf(xuio
, i
);
5198 * if abuf == NULL, it must be a write buffer
5199 * that has been returned in zfs_write().
5202 dmu_return_arcbuf(abuf
);
5203 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
5206 dmu_xuio_fini(xuio
);
5209 #endif /* HAVE_UIO_ZEROCOPY */
5211 #if defined(_KERNEL)
5212 EXPORT_SYMBOL(zfs_open
);
5213 EXPORT_SYMBOL(zfs_close
);
5214 EXPORT_SYMBOL(zfs_read
);
5215 EXPORT_SYMBOL(zfs_write
);
5216 EXPORT_SYMBOL(zfs_access
);
5217 EXPORT_SYMBOL(zfs_lookup
);
5218 EXPORT_SYMBOL(zfs_create
);
5219 EXPORT_SYMBOL(zfs_tmpfile
);
5220 EXPORT_SYMBOL(zfs_remove
);
5221 EXPORT_SYMBOL(zfs_mkdir
);
5222 EXPORT_SYMBOL(zfs_rmdir
);
5223 EXPORT_SYMBOL(zfs_readdir
);
5224 EXPORT_SYMBOL(zfs_fsync
);
5225 EXPORT_SYMBOL(zfs_getattr
);
5226 EXPORT_SYMBOL(zfs_getattr_fast
);
5227 EXPORT_SYMBOL(zfs_setattr
);
5228 EXPORT_SYMBOL(zfs_rename
);
5229 EXPORT_SYMBOL(zfs_symlink
);
5230 EXPORT_SYMBOL(zfs_readlink
);
5231 EXPORT_SYMBOL(zfs_link
);
5232 EXPORT_SYMBOL(zfs_inactive
);
5233 EXPORT_SYMBOL(zfs_space
);
5234 EXPORT_SYMBOL(zfs_fid
);
5235 EXPORT_SYMBOL(zfs_getsecattr
);
5236 EXPORT_SYMBOL(zfs_setsecattr
);
5237 EXPORT_SYMBOL(zfs_getpage
);
5238 EXPORT_SYMBOL(zfs_putpage
);
5239 EXPORT_SYMBOL(zfs_dirty_inode
);
5240 EXPORT_SYMBOL(zfs_map
);
5243 module_param(zfs_delete_blocks
, ulong
, 0644);
5244 MODULE_PARM_DESC(zfs_delete_blocks
, "Delete files larger than N blocks async");
5245 module_param(zfs_read_chunk_size
, long, 0644);
5246 MODULE_PARM_DESC(zfs_read_chunk_size
, "Bytes to read per chunk");