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, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2015 Joyent, Inc.
27 * Copyright 2017 Nexenta Systems, Inc.
30 /* Portions Copyright 2007 Jeremy Teo */
31 /* Portions Copyright 2010 Robert Milkowski */
33 #include <sys/types.h>
34 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/resource.h>
40 #include <sys/vnode.h>
44 #include <sys/taskq.h>
46 #include <sys/vmsystm.h>
47 #include <sys/atomic.h>
49 #include <vm/seg_vn.h>
53 #include <vm/seg_kpm.h>
55 #include <sys/pathname.h>
56 #include <sys/cmn_err.h>
57 #include <sys/errno.h>
58 #include <sys/unistd.h>
59 #include <sys/zfs_dir.h>
60 #include <sys/zfs_acl.h>
61 #include <sys/zfs_ioctl.h>
62 #include <sys/fs/zfs.h>
64 #include <sys/dmu_objset.h>
70 #include <sys/dirent.h>
71 #include <sys/policy.h>
72 #include <sys/sunddi.h>
73 #include <sys/filio.h>
75 #include "sys/fs_subr.h"
76 #include <sys/zfs_ctldir.h>
77 #include <sys/zfs_fuid.h>
78 #include <sys/zfs_sa.h>
80 #include <sys/zfs_rlock.h>
81 #include <sys/extdirent.h>
82 #include <sys/kidmap.h>
90 * Each vnode op performs some logical unit of work. To do this, the ZPL must
91 * properly lock its in-core state, create a DMU transaction, do the work,
92 * record this work in the intent log (ZIL), commit the DMU transaction,
93 * and wait for the intent log to commit if it is a synchronous operation.
94 * Moreover, the vnode ops must work in both normal and log replay context.
95 * The ordering of events is important to avoid deadlocks and references
96 * to freed memory. The example below illustrates the following Big Rules:
98 * (1) A check must be made in each zfs thread for a mounted file system.
99 * This is done avoiding races using ZFS_ENTER(zfsvfs).
100 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
101 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
102 * can return EIO from the calling function.
104 * (2) VN_RELE() should always be the last thing except for zil_commit()
105 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
106 * First, if it's the last reference, the vnode/znode
107 * can be freed, so the zp may point to freed memory. Second, the last
108 * reference will call zfs_zinactive(), which may induce a lot of work --
109 * pushing cached pages (which acquires range locks) and syncing out
110 * cached atime changes. Third, zfs_zinactive() may require a new tx,
111 * which could deadlock the system if you were already holding one.
112 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
114 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
115 * as they can span dmu_tx_assign() calls.
117 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
118 * dmu_tx_assign(). This is critical because we don't want to block
119 * while holding locks.
121 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
122 * reduces lock contention and CPU usage when we must wait (note that if
123 * throughput is constrained by the storage, nearly every transaction
126 * Note, in particular, that if a lock is sometimes acquired before
127 * the tx assigns, and sometimes after (e.g. z_lock), then failing
128 * to use a non-blocking assign can deadlock the system. The scenario:
130 * Thread A has grabbed a lock before calling dmu_tx_assign().
131 * Thread B is in an already-assigned tx, and blocks for this lock.
132 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
133 * forever, because the previous txg can't quiesce until B's tx commits.
135 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
136 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
137 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
138 * to indicate that this operation has already called dmu_tx_wait().
139 * This will ensure that we don't retry forever, waiting a short bit
142 * (5) If the operation succeeded, generate the intent log entry for it
143 * before dropping locks. This ensures that the ordering of events
144 * in the intent log matches the order in which they actually occurred.
145 * During ZIL replay the zfs_log_* functions will update the sequence
146 * number to indicate the zil transaction has replayed.
148 * (6) At the end of each vnode op, the DMU tx must always commit,
149 * regardless of whether there were any errors.
151 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
152 * to ensure that synchronous semantics are provided when necessary.
154 * In general, this is how things should be ordered in each vnode op:
156 * ZFS_ENTER(zfsvfs); // exit if unmounted
158 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
159 * rw_enter(...); // grab any other locks you need
160 * tx = dmu_tx_create(...); // get DMU tx
161 * dmu_tx_hold_*(); // hold each object you might modify
162 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
164 * rw_exit(...); // drop locks
165 * zfs_dirent_unlock(dl); // unlock directory entry
166 * VN_RELE(...); // release held vnodes
167 * if (error == ERESTART) {
173 * dmu_tx_abort(tx); // abort DMU tx
174 * ZFS_EXIT(zfsvfs); // finished in zfs
175 * return (error); // really out of space
177 * error = do_real_work(); // do whatever this VOP does
179 * zfs_log_*(...); // on success, make ZIL entry
180 * dmu_tx_commit(tx); // commit DMU tx -- error or not
181 * rw_exit(...); // drop locks
182 * zfs_dirent_unlock(dl); // unlock directory entry
183 * VN_RELE(...); // release held vnodes
184 * zil_commit(zilog, foid); // synchronous when necessary
185 * ZFS_EXIT(zfsvfs); // finished in zfs
186 * return (error); // done, report error
191 zfs_open(vnode_t
**vpp
, int flag
, cred_t
*cr
, caller_context_t
*ct
)
193 znode_t
*zp
= VTOZ(*vpp
);
194 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
199 if ((flag
& FWRITE
) && (zp
->z_pflags
& ZFS_APPENDONLY
) &&
200 ((flag
& FAPPEND
) == 0)) {
202 return (SET_ERROR(EPERM
));
205 if (!zfs_has_ctldir(zp
) && zp
->z_zfsvfs
->z_vscan
&&
206 ZTOV(zp
)->v_type
== VREG
&&
207 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0) {
208 if (fs_vscan(*vpp
, cr
, 0) != 0) {
210 return (SET_ERROR(EACCES
));
214 /* Keep a count of the synchronous opens in the znode */
215 if (flag
& (FSYNC
| FDSYNC
))
216 atomic_inc_32(&zp
->z_sync_cnt
);
224 zfs_close(vnode_t
*vp
, int flag
, int count
, offset_t offset
, cred_t
*cr
,
225 caller_context_t
*ct
)
227 znode_t
*zp
= VTOZ(vp
);
228 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
231 * Clean up any locks held by this process on the vp.
233 cleanlocks(vp
, ddi_get_pid(), 0);
234 cleanshares(vp
, ddi_get_pid());
239 /* Decrement the synchronous opens in the znode */
240 if ((flag
& (FSYNC
| FDSYNC
)) && (count
== 1))
241 atomic_dec_32(&zp
->z_sync_cnt
);
243 if (!zfs_has_ctldir(zp
) && zp
->z_zfsvfs
->z_vscan
&&
244 ZTOV(zp
)->v_type
== VREG
&&
245 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0)
246 VERIFY(fs_vscan(vp
, cr
, 1) == 0);
253 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
254 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
257 zfs_holey(vnode_t
*vp
, int cmd
, offset_t
*off
)
259 znode_t
*zp
= VTOZ(vp
);
260 uint64_t noff
= (uint64_t)*off
; /* new offset */
265 file_sz
= zp
->z_size
;
266 if (noff
>= file_sz
) {
267 return (SET_ERROR(ENXIO
));
270 if (cmd
== _FIO_SEEK_HOLE
)
275 error
= dmu_offset_next(zp
->z_zfsvfs
->z_os
, zp
->z_id
, hole
, &noff
);
278 return (SET_ERROR(ENXIO
));
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
) {
300 zfs_ioctl(vnode_t
*vp
, int com
, intptr_t data
, int flag
, cred_t
*cred
,
301 int *rvalp
, caller_context_t
*ct
)
305 dmu_object_info_t doi
;
313 return (zfs_sync(vp
->v_vfsp
, 0, cred
));
316 * The following two ioctls are used by bfu. Faking out,
317 * necessary to avoid bfu errors.
329 if (ddi_copyin((void *)data
, &off
, sizeof (off
), flag
))
330 return (SET_ERROR(EFAULT
));
333 zfsvfs
= zp
->z_zfsvfs
;
337 /* offset parameter is in/out */
338 error
= zfs_holey(vp
, com
, &off
);
342 if (ddi_copyout(&off
, (void *)data
, sizeof (off
), flag
))
343 return (SET_ERROR(EFAULT
));
346 case _FIO_COUNT_FILLED
:
349 * _FIO_COUNT_FILLED adds a new ioctl command which
350 * exposes the number of filled blocks in a
354 zfsvfs
= zp
->z_zfsvfs
;
359 * Wait for all dirty blocks for this object
360 * to get synced out to disk, and the DMU info
363 error
= dmu_object_wait_synced(zfsvfs
->z_os
, zp
->z_id
);
370 * Retrieve fill count from DMU object.
372 error
= dmu_object_info(zfsvfs
->z_os
, zp
->z_id
, &doi
);
378 ndata
= doi
.doi_fill_count
;
381 if (ddi_copyout(&ndata
, (void *)data
, sizeof (ndata
), flag
))
382 return (SET_ERROR(EFAULT
));
386 return (SET_ERROR(ENOTTY
));
390 * Utility functions to map and unmap a single physical page. These
391 * are used to manage the mappable copies of ZFS file data, and therefore
392 * do not update ref/mod bits.
395 zfs_map_page(page_t
*pp
, enum seg_rw rw
)
398 return (hat_kpm_mapin(pp
, 0));
399 ASSERT(rw
== S_READ
|| rw
== S_WRITE
);
400 return (ppmapin(pp
, PROT_READ
| ((rw
== S_WRITE
) ? PROT_WRITE
: 0),
405 zfs_unmap_page(page_t
*pp
, caddr_t addr
)
408 hat_kpm_mapout(pp
, 0, addr
);
415 * When a file is memory mapped, we must keep the IO data synchronized
416 * between the DMU cache and the memory mapped pages. What this means:
418 * On Write: If we find a memory mapped page, we write to *both*
419 * the page and the dmu buffer.
422 update_pages(vnode_t
*vp
, int64_t start
, int len
, objset_t
*os
, uint64_t oid
)
426 off
= start
& PAGEOFFSET
;
427 for (start
&= PAGEMASK
; len
> 0; start
+= PAGESIZE
) {
429 uint64_t nbytes
= MIN(PAGESIZE
- off
, len
);
431 if (pp
= page_lookup(&vp
->v_object
, start
, SE_SHARED
)) {
434 va
= zfs_map_page(pp
, S_WRITE
);
435 (void) dmu_read(os
, oid
, start
+off
, nbytes
, va
+off
,
437 zfs_unmap_page(pp
, va
);
446 * When a file is memory mapped, we must keep the IO data synchronized
447 * between the DMU cache and the memory mapped pages. What this means:
449 * On Read: We "read" preferentially from memory mapped pages,
450 * else we default from the dmu buffer.
452 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
453 * the file is memory mapped.
456 mappedread(vnode_t
*vp
, int nbytes
, uio_t
*uio
)
458 znode_t
*zp
= VTOZ(vp
);
463 start
= uio
->uio_loffset
;
464 off
= start
& PAGEOFFSET
;
465 for (start
&= PAGEMASK
; len
> 0; start
+= PAGESIZE
) {
467 uint64_t bytes
= MIN(PAGESIZE
- off
, len
);
469 if (pp
= page_lookup(&vp
->v_object
, start
, SE_SHARED
)) {
472 va
= zfs_map_page(pp
, S_READ
);
473 error
= uiomove(va
+ off
, bytes
, UIO_READ
, uio
);
474 zfs_unmap_page(pp
, va
);
477 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
488 offset_t zfs_read_chunk_size
= 1024 * 1024; /* Tunable */
491 * Read bytes from specified file into supplied buffer.
493 * IN: vp - vnode of file to be read from.
494 * uio - structure supplying read location, range info,
496 * ioflag - SYNC flags; used to provide FRSYNC semantics.
497 * cr - credentials of caller.
498 * ct - caller context
500 * OUT: uio - updated offset and range, buffer filled.
502 * RETURN: 0 on success, error code on failure.
505 * vp - atime updated if byte count > 0
509 zfs_read(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cr
, caller_context_t
*ct
)
511 znode_t
*zp
= VTOZ(vp
);
512 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
521 if (zp
->z_pflags
& ZFS_AV_QUARANTINED
) {
523 return (SET_ERROR(EACCES
));
527 * Validate file offset
529 if (uio
->uio_loffset
< 0) {
531 return (SET_ERROR(EINVAL
));
535 * Fasttrack empty reads
537 if (uio
->uio_resid
== 0) {
543 * Check for mandatory locks
545 if (MANDMODE(zp
->z_mode
)) {
546 if (error
= chklock(vp
, FREAD
,
547 uio
->uio_loffset
, uio
->uio_resid
, uio
->uio_fmode
, ct
)) {
554 * If we're in FRSYNC mode, sync out this znode before reading it.
556 if (ioflag
& FRSYNC
|| zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
557 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
560 * Lock the range against changes.
562 rl
= zfs_range_lock(zp
, uio
->uio_loffset
, uio
->uio_resid
, RL_READER
);
565 * If we are reading past end-of-file we can skip
566 * to the end; but we might still need to set atime.
568 if (uio
->uio_loffset
>= zp
->z_size
) {
573 ASSERT(uio
->uio_loffset
< zp
->z_size
);
574 n
= MIN(uio
->uio_resid
, zp
->z_size
- uio
->uio_loffset
);
576 if ((uio
->uio_extflg
== UIO_XUIO
) &&
577 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
)) {
579 int blksz
= zp
->z_blksz
;
580 uint64_t offset
= uio
->uio_loffset
;
582 xuio
= (xuio_t
*)uio
;
584 nblk
= (P2ROUNDUP(offset
+ n
, blksz
) - P2ALIGN(offset
,
587 ASSERT(offset
+ n
<= blksz
);
590 (void) dmu_xuio_init(xuio
, nblk
);
592 if (vn_has_cached_data(vp
)) {
594 * For simplicity, we always allocate a full buffer
595 * even if we only expect to read a portion of a block.
597 while (--nblk
>= 0) {
598 (void) dmu_xuio_add(xuio
,
599 dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
606 nbytes
= MIN(n
, zfs_read_chunk_size
-
607 P2PHASE(uio
->uio_loffset
, zfs_read_chunk_size
));
609 if (vn_has_cached_data(vp
)) {
610 error
= mappedread(vp
, nbytes
, uio
);
612 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
616 /* convert checksum errors into IO errors */
618 error
= SET_ERROR(EIO
);
625 zfs_range_unlock(rl
);
627 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
633 * Write the bytes to a file.
635 * IN: vp - vnode of file to be written to.
636 * uio - structure supplying write location, range info,
638 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
639 * set if in append mode.
640 * cr - credentials of caller.
641 * ct - caller context (NFS/CIFS fem monitor only)
643 * OUT: uio - updated offset and range.
645 * RETURN: 0 on success, error code on failure.
648 * vp - ctime|mtime updated if byte count > 0
653 zfs_write(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cr
, caller_context_t
*ct
)
655 znode_t
*zp
= VTOZ(vp
);
656 rlim64_t limit
= uio
->uio_llimit
;
657 ssize_t start_resid
= uio
->uio_resid
;
661 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
666 int max_blksz
= zfsvfs
->z_max_blksz
;
669 iovec_t
*aiov
= NULL
;
672 int iovcnt
= uio
->uio_iovcnt
;
673 iovec_t
*iovp
= uio
->uio_iov
;
676 sa_bulk_attr_t bulk
[4];
677 uint64_t mtime
[2], ctime
[2];
680 * Fasttrack empty write
686 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
692 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
693 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
694 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
), NULL
,
696 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
700 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
701 * callers might not be able to detect properly that we are read-only,
702 * so check it explicitly here.
704 if (zfsvfs
->z_vfs
->vfs_flag
& VFS_RDONLY
) {
706 return (SET_ERROR(EROFS
));
710 * If immutable or not appending then return EPERM.
711 * Intentionally allow ZFS_READONLY through here.
712 * See zfs_zaccess_common()
714 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) ||
715 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
716 (uio
->uio_loffset
< zp
->z_size
))) {
718 return (SET_ERROR(EPERM
));
721 zilog
= zfsvfs
->z_log
;
724 * Validate file offset
726 woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
729 return (SET_ERROR(EINVAL
));
733 * Check for mandatory locks before calling zfs_range_lock()
734 * in order to prevent a deadlock with locks set via fcntl().
736 if (MANDMODE((mode_t
)zp
->z_mode
) &&
737 (error
= chklock(vp
, FWRITE
, woff
, n
, uio
->uio_fmode
, ct
)) != 0) {
743 * Pre-fault the pages to ensure slow (eg NFS) pages
745 * Skip this if uio contains loaned arc_buf.
747 if ((uio
->uio_extflg
== UIO_XUIO
) &&
748 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
))
749 xuio
= (xuio_t
*)uio
;
751 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
754 * If in append mode, set the io offset pointer to eof.
756 if (ioflag
& FAPPEND
) {
758 * Obtain an appending range lock to guarantee file append
759 * semantics. We reset the write offset once we have the lock.
761 rl
= zfs_range_lock(zp
, 0, n
, RL_APPEND
);
763 if (rl
->r_len
== UINT64_MAX
) {
765 * We overlocked the file because this write will cause
766 * the file block size to increase.
767 * Note that zp_size cannot change with this lock held.
771 uio
->uio_loffset
= woff
;
774 * Note that if the file block size will change as a result of
775 * this write, then this range lock will lock the entire file
776 * so that we can re-write the block safely.
778 rl
= zfs_range_lock(zp
, woff
, n
, RL_WRITER
);
782 zfs_range_unlock(rl
);
784 return (SET_ERROR(EFBIG
));
787 if ((woff
+ n
) > limit
|| woff
> (limit
- n
))
790 /* Will this write extend the file length? */
791 write_eof
= (woff
+ n
> zp
->z_size
);
793 end_size
= MAX(zp
->z_size
, woff
+ n
);
796 * Write the file in reasonable size chunks. Each chunk is written
797 * in a separate transaction; this keeps the intent log records small
798 * and allows us to do more fine-grained space accounting.
802 woff
= uio
->uio_loffset
;
803 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
804 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
806 dmu_return_arcbuf(abuf
);
807 error
= SET_ERROR(EDQUOT
);
811 if (xuio
&& abuf
== NULL
) {
812 ASSERT(i_iov
< iovcnt
);
814 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
815 dmu_xuio_clear(xuio
, i_iov
);
816 DTRACE_PROBE3(zfs_cp_write
, int, i_iov
,
817 iovec_t
*, aiov
, arc_buf_t
*, abuf
);
818 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
819 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
820 aiov
->iov_len
== arc_buf_size(abuf
)));
822 } else if (abuf
== NULL
&& n
>= max_blksz
&&
823 woff
>= zp
->z_size
&&
824 P2PHASE(woff
, max_blksz
) == 0 &&
825 zp
->z_blksz
== max_blksz
) {
827 * This write covers a full block. "Borrow" a buffer
828 * from the dmu so that we can fill it before we enter
829 * a transaction. This avoids the possibility of
830 * holding up the transaction if the data copy hangs
831 * up on a pagefault (e.g., from an NFS server mapping).
835 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
837 ASSERT(abuf
!= NULL
);
838 ASSERT(arc_buf_size(abuf
) == max_blksz
);
839 if (error
= uiocopy(abuf
->b_data
, max_blksz
,
840 UIO_WRITE
, uio
, &cbytes
)) {
841 dmu_return_arcbuf(abuf
);
844 ASSERT(cbytes
== max_blksz
);
848 * Start a transaction.
850 tx
= dmu_tx_create(zfsvfs
->z_os
);
851 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
852 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
853 zfs_sa_upgrade_txholds(tx
, zp
);
854 error
= dmu_tx_assign(tx
, TXG_WAIT
);
858 dmu_return_arcbuf(abuf
);
863 * If zfs_range_lock() over-locked we grow the blocksize
864 * and then reduce the lock range. This will only happen
865 * on the first iteration since zfs_range_reduce() will
866 * shrink down r_len to the appropriate size.
868 if (rl
->r_len
== UINT64_MAX
) {
871 if (zp
->z_blksz
> max_blksz
) {
873 * File's blocksize is already larger than the
874 * "recordsize" property. Only let it grow to
875 * the next power of 2.
877 ASSERT(!ISP2(zp
->z_blksz
));
878 new_blksz
= MIN(end_size
,
879 1 << highbit64(zp
->z_blksz
));
881 new_blksz
= MIN(end_size
, max_blksz
);
883 zfs_grow_blocksize(zp
, new_blksz
, tx
);
884 zfs_range_reduce(rl
, woff
, n
);
888 * XXX - should we really limit each write to z_max_blksz?
889 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
891 nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
894 tx_bytes
= uio
->uio_resid
;
895 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
897 tx_bytes
-= uio
->uio_resid
;
900 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
902 * If this is not a full block write, but we are
903 * extending the file past EOF and this data starts
904 * block-aligned, use assign_arcbuf(). Otherwise,
905 * write via dmu_write().
907 if (tx_bytes
< max_blksz
&& (!write_eof
||
908 aiov
->iov_base
!= abuf
->b_data
)) {
910 dmu_write(zfsvfs
->z_os
, zp
->z_id
, woff
,
911 aiov
->iov_len
, aiov
->iov_base
, tx
);
912 dmu_return_arcbuf(abuf
);
913 xuio_stat_wbuf_copied();
915 ASSERT(xuio
|| tx_bytes
== max_blksz
);
916 dmu_assign_arcbuf(sa_get_db(zp
->z_sa_hdl
),
919 ASSERT(tx_bytes
<= uio
->uio_resid
);
920 uioskip(uio
, tx_bytes
);
922 if (tx_bytes
&& vn_has_cached_data(vp
)) {
923 update_pages(vp
, woff
,
924 tx_bytes
, zfsvfs
->z_os
, zp
->z_id
);
928 * If we made no progress, we're done. If we made even
929 * partial progress, update the znode and ZIL accordingly.
932 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
933 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
940 * Clear Set-UID/Set-GID bits on successful write if not
941 * privileged and at least one of the excute bits is set.
943 * It would be nice to to this after all writes have
944 * been done, but that would still expose the ISUID/ISGID
945 * to another app after the partial write is committed.
947 * Note: we don't call zfs_fuid_map_id() here because
948 * user 0 is not an ephemeral uid.
950 mutex_enter(&zp
->z_acl_lock
);
951 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
952 (S_IXUSR
>> 6))) != 0 &&
953 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
954 secpolicy_vnode_setid_retain(cr
,
955 (zp
->z_mode
& S_ISUID
) != 0 && zp
->z_uid
== 0) != 0) {
957 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
958 newmode
= zp
->z_mode
;
959 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zfsvfs
),
960 (void *)&newmode
, sizeof (uint64_t), tx
);
962 mutex_exit(&zp
->z_acl_lock
);
964 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
968 * Update the file size (zp_size) if it has changed;
969 * account for possible concurrent updates.
971 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
972 (void) atomic_cas_64(&zp
->z_size
, end_size
,
977 * If we are replaying and eof is non zero then force
978 * the file size to the specified eof. Note, there's no
979 * concurrency during replay.
981 if (zfsvfs
->z_replay
&& zfsvfs
->z_replay_eof
!= 0)
982 zp
->z_size
= zfsvfs
->z_replay_eof
;
984 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
986 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
);
991 ASSERT(tx_bytes
== nbytes
);
995 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
998 zfs_range_unlock(rl
);
1001 * If we're in replay mode, or we made no progress, return error.
1002 * Otherwise, it's at least a partial write, so it's successful.
1004 if (zfsvfs
->z_replay
|| uio
->uio_resid
== start_resid
) {
1009 if (ioflag
& (FSYNC
| FDSYNC
) ||
1010 zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1011 zil_commit(zilog
, zp
->z_id
);
1018 zfs_get_done(zgd_t
*zgd
, int error
)
1020 znode_t
*zp
= zgd
->zgd_private
;
1021 objset_t
*os
= zp
->z_zfsvfs
->z_os
;
1024 dmu_buf_rele(zgd
->zgd_db
, zgd
);
1026 zfs_range_unlock(zgd
->zgd_rl
);
1029 * Release the vnode asynchronously as we currently have the
1030 * txg stopped from syncing.
1032 VN_RELE_ASYNC(ZTOV(zp
), dsl_pool_vnrele_taskq(dmu_objset_pool(os
)));
1034 if (error
== 0 && zgd
->zgd_bp
)
1035 zil_lwb_add_block(zgd
->zgd_lwb
, zgd
->zgd_bp
);
1037 kmem_free(zgd
, sizeof (zgd_t
));
1041 static int zil_fault_io
= 0;
1045 * Get data to generate a TX_WRITE intent log record.
1048 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, struct lwb
*lwb
, zio_t
*zio
)
1050 zfsvfs_t
*zfsvfs
= arg
;
1051 objset_t
*os
= zfsvfs
->z_os
;
1053 uint64_t object
= lr
->lr_foid
;
1054 uint64_t offset
= lr
->lr_offset
;
1055 uint64_t size
= lr
->lr_length
;
1060 ASSERT3P(lwb
, !=, NULL
);
1061 ASSERT3P(zio
, !=, NULL
);
1062 ASSERT3U(size
, !=, 0);
1065 * Nothing to do if the file has been removed
1067 if (zfs_zget(zfsvfs
, object
, &zp
) != 0)
1068 return (SET_ERROR(ENOENT
));
1069 if (zp
->z_unlinked
) {
1071 * Release the vnode asynchronously as we currently have the
1072 * txg stopped from syncing.
1074 VN_RELE_ASYNC(ZTOV(zp
),
1075 dsl_pool_vnrele_taskq(dmu_objset_pool(os
)));
1076 return (SET_ERROR(ENOENT
));
1079 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1081 zgd
->zgd_private
= zp
;
1084 * Write records come in two flavors: immediate and indirect.
1085 * For small writes it's cheaper to store the data with the
1086 * log record (immediate); for large writes it's cheaper to
1087 * sync the data and get a pointer to it (indirect) so that
1088 * we don't have to write the data twice.
1090 if (buf
!= NULL
) { /* immediate write */
1091 zgd
->zgd_rl
= zfs_range_lock(zp
, offset
, size
, RL_READER
);
1092 /* test for truncation needs to be done while range locked */
1093 if (offset
>= zp
->z_size
) {
1094 error
= SET_ERROR(ENOENT
);
1096 error
= dmu_read(os
, object
, offset
, size
, buf
,
1097 DMU_READ_NO_PREFETCH
);
1099 ASSERT(error
== 0 || error
== ENOENT
);
1100 } else { /* indirect write */
1102 * Have to lock the whole block to ensure when it's
1103 * written out and its checksum is being calculated
1104 * that no one can change the data. We need to re-check
1105 * blocksize after we get the lock in case it's changed!
1110 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1112 zgd
->zgd_rl
= zfs_range_lock(zp
, offset
, size
,
1114 if (zp
->z_blksz
== size
)
1117 zfs_range_unlock(zgd
->zgd_rl
);
1119 /* test for truncation needs to be done while range locked */
1120 if (lr
->lr_offset
>= zp
->z_size
)
1121 error
= SET_ERROR(ENOENT
);
1124 error
= SET_ERROR(EIO
);
1129 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1130 DMU_READ_NO_PREFETCH
);
1133 blkptr_t
*bp
= &lr
->lr_blkptr
;
1138 ASSERT(db
->db_offset
== offset
);
1139 ASSERT(db
->db_size
== size
);
1141 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1143 ASSERT(error
|| lr
->lr_length
<= size
);
1146 * On success, we need to wait for the write I/O
1147 * initiated by dmu_sync() to complete before we can
1148 * release this dbuf. We will finish everything up
1149 * in the zfs_get_done() callback.
1154 if (error
== EALREADY
) {
1155 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
1161 zfs_get_done(zgd
, error
);
1168 zfs_access(vnode_t
*vp
, int mode
, int flag
, cred_t
*cr
,
1169 caller_context_t
*ct
)
1171 znode_t
*zp
= VTOZ(vp
);
1172 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
1178 if (flag
& V_ACE_MASK
)
1179 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1181 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1188 * If vnode is for a device return a specfs vnode instead.
1191 specvp_check(vnode_t
**vpp
, cred_t
*cr
)
1195 if (IS_DEVVP(*vpp
)) {
1198 svp
= specvp(*vpp
, (*vpp
)->v_rdev
, (*vpp
)->v_type
, cr
);
1201 error
= SET_ERROR(ENOSYS
);
1209 * Lookup an entry in a directory, or an extended attribute directory.
1210 * If it exists, return a held vnode reference for it.
1212 * IN: dvp - vnode of directory to search.
1213 * nm - name of entry to lookup.
1214 * pnp - full pathname to lookup [UNUSED].
1215 * flags - LOOKUP_XATTR set if looking for an attribute.
1216 * rdir - root directory vnode [UNUSED].
1217 * cr - credentials of caller.
1218 * ct - caller context
1219 * direntflags - directory lookup flags
1220 * realpnp - returned pathname.
1222 * OUT: vpp - vnode of located entry, NULL if not found.
1224 * RETURN: 0 on success, error code on failure.
1231 zfs_lookup(vnode_t
*dvp
, char *nm
, vnode_t
**vpp
, struct pathname
*pnp
,
1232 int flags
, vnode_t
*rdir
, cred_t
*cr
, caller_context_t
*ct
,
1233 int *direntflags
, pathname_t
*realpnp
)
1235 znode_t
*zdp
= VTOZ(dvp
);
1236 zfsvfs_t
*zfsvfs
= zdp
->z_zfsvfs
;
1240 * Fast path lookup, however we must skip DNLC lookup
1241 * for case folding or normalizing lookups because the
1242 * DNLC code only stores the passed in name. This means
1243 * creating 'a' and removing 'A' on a case insensitive
1244 * file system would work, but DNLC still thinks 'a'
1245 * exists and won't let you create it again on the next
1246 * pass through fast path.
1248 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1250 if (dvp
->v_type
!= VDIR
) {
1251 return (SET_ERROR(ENOTDIR
));
1252 } else if (zdp
->z_sa_hdl
== NULL
) {
1253 return (SET_ERROR(EIO
));
1256 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1257 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1264 } else if (!zdp
->z_zfsvfs
->z_norm
&&
1265 (zdp
->z_zfsvfs
->z_case
== ZFS_CASE_SENSITIVE
)) {
1267 vnode_t
*tvp
= dnlc_lookup(dvp
, nm
);
1270 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1275 if (tvp
== DNLC_NO_VNODE
) {
1277 return (SET_ERROR(ENOENT
));
1280 return (specvp_check(vpp
, cr
));
1286 DTRACE_PROBE2(zfs__fastpath__lookup__miss
, vnode_t
*, dvp
, char *, nm
);
1293 if (flags
& LOOKUP_XATTR
) {
1295 * If the xattr property is off, refuse the lookup request.
1297 if (!(zfsvfs
->z_vfs
->vfs_flag
& VFS_XATTR
)) {
1299 return (SET_ERROR(EINVAL
));
1303 * We don't allow recursive attributes..
1304 * Maybe someday we will.
1306 if (zdp
->z_pflags
& ZFS_XATTR
) {
1308 return (SET_ERROR(EINVAL
));
1311 if (error
= zfs_get_xattrdir(VTOZ(dvp
), vpp
, cr
, flags
)) {
1317 * Do we have permission to get into attribute directory?
1320 if (error
= zfs_zaccess(VTOZ(*vpp
), ACE_EXECUTE
, 0,
1330 if (dvp
->v_type
!= VDIR
) {
1332 return (SET_ERROR(ENOTDIR
));
1336 * Check accessibility of directory.
1339 if (error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
)) {
1344 if (zfsvfs
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1345 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1347 return (SET_ERROR(EILSEQ
));
1350 error
= zfs_dirlook(zdp
, nm
, vpp
, flags
, direntflags
, realpnp
);
1352 error
= specvp_check(vpp
, cr
);
1359 * Attempt to create a new entry in a directory. If the entry
1360 * already exists, truncate the file if permissible, else return
1361 * an error. Return the vp of the created or trunc'd file.
1363 * IN: dvp - vnode of directory to put new file entry in.
1364 * name - name of new file entry.
1365 * vap - attributes of new file.
1366 * excl - flag indicating exclusive or non-exclusive mode.
1367 * mode - mode to open file with.
1368 * cr - credentials of caller.
1369 * flag - large file flag [UNUSED].
1370 * ct - caller context
1371 * vsecp - ACL to be set
1373 * OUT: vpp - vnode of created or trunc'd entry.
1375 * RETURN: 0 on success, error code on failure.
1378 * dvp - ctime|mtime updated if new entry created
1379 * vp - ctime|mtime always, atime if new
1384 zfs_create(vnode_t
*dvp
, char *name
, vattr_t
*vap
, vcexcl_t excl
,
1385 int mode
, vnode_t
**vpp
, cred_t
*cr
, int flag
, caller_context_t
*ct
,
1388 znode_t
*zp
, *dzp
= VTOZ(dvp
);
1389 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1397 gid_t gid
= crgetgid(cr
);
1398 zfs_acl_ids_t acl_ids
;
1399 boolean_t fuid_dirtied
;
1400 boolean_t have_acl
= B_FALSE
;
1401 boolean_t waited
= B_FALSE
;
1404 * If we have an ephemeral id, ACL, or XVATTR then
1405 * make sure file system is at proper version
1408 ksid
= crgetsid(cr
, KSID_OWNER
);
1410 uid
= ksid_getid(ksid
);
1414 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1415 (vsecp
|| (vap
->va_mask
& AT_XVATTR
) ||
1416 IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1417 return (SET_ERROR(EINVAL
));
1422 zilog
= zfsvfs
->z_log
;
1424 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
1425 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1427 return (SET_ERROR(EILSEQ
));
1430 if (vap
->va_mask
& AT_XVATTR
) {
1431 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1432 crgetuid(cr
), cr
, vap
->va_type
)) != 0) {
1440 if ((vap
->va_mode
& VSVTX
) && secpolicy_vnode_stky_modify(cr
))
1441 vap
->va_mode
&= ~VSVTX
;
1443 if (*name
== '\0') {
1445 * Null component name refers to the directory itself.
1452 /* possible VN_HOLD(zp) */
1455 if (flag
& FIGNORECASE
)
1458 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1462 zfs_acl_ids_free(&acl_ids
);
1463 if (strcmp(name
, "..") == 0)
1464 error
= SET_ERROR(EISDIR
);
1474 * Create a new file object and update the directory
1477 if (error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
)) {
1479 zfs_acl_ids_free(&acl_ids
);
1484 * We only support the creation of regular files in
1485 * extended attribute directories.
1488 if ((dzp
->z_pflags
& ZFS_XATTR
) &&
1489 (vap
->va_type
!= VREG
)) {
1491 zfs_acl_ids_free(&acl_ids
);
1492 error
= SET_ERROR(EINVAL
);
1496 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1497 cr
, vsecp
, &acl_ids
)) != 0)
1501 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1502 zfs_acl_ids_free(&acl_ids
);
1503 error
= SET_ERROR(EDQUOT
);
1507 tx
= dmu_tx_create(os
);
1509 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1510 ZFS_SA_BASE_ATTR_SIZE
);
1512 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1514 zfs_fuid_txhold(zfsvfs
, tx
);
1515 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1516 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1517 if (!zfsvfs
->z_use_sa
&&
1518 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1519 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1520 0, acl_ids
.z_aclp
->z_acl_bytes
);
1522 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1524 zfs_dirent_unlock(dl
);
1525 if (error
== ERESTART
) {
1531 zfs_acl_ids_free(&acl_ids
);
1536 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1539 zfs_fuid_sync(zfsvfs
, tx
);
1541 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1542 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1543 if (flag
& FIGNORECASE
)
1545 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1546 vsecp
, acl_ids
.z_fuidp
, vap
);
1547 zfs_acl_ids_free(&acl_ids
);
1550 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1553 zfs_acl_ids_free(&acl_ids
);
1557 * A directory entry already exists for this name.
1560 * Can't truncate an existing file if in exclusive mode.
1563 error
= SET_ERROR(EEXIST
);
1567 * Can't open a directory for writing.
1569 if ((ZTOV(zp
)->v_type
== VDIR
) && (mode
& S_IWRITE
)) {
1570 error
= SET_ERROR(EISDIR
);
1574 * Verify requested access to file.
1576 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1580 mutex_enter(&dzp
->z_lock
);
1582 mutex_exit(&dzp
->z_lock
);
1585 * Truncate regular files if requested.
1587 if ((ZTOV(zp
)->v_type
== VREG
) &&
1588 (vap
->va_mask
& AT_SIZE
) && (vap
->va_size
== 0)) {
1589 /* we can't hold any locks when calling zfs_freesp() */
1590 zfs_dirent_unlock(dl
);
1592 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1594 vnevent_create(ZTOV(zp
), ct
);
1601 zfs_dirent_unlock(dl
);
1608 error
= specvp_check(vpp
, cr
);
1611 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1612 zil_commit(zilog
, 0);
1619 * Remove an entry from a directory.
1621 * IN: dvp - vnode of directory to remove entry from.
1622 * name - name of entry to remove.
1623 * cr - credentials of caller.
1624 * ct - caller context
1625 * flags - case flags
1627 * RETURN: 0 on success, error code on failure.
1631 * vp - ctime (if nlink > 0)
1634 uint64_t null_xattr
= 0;
1638 zfs_remove(vnode_t
*dvp
, char *name
, cred_t
*cr
, caller_context_t
*ct
,
1641 znode_t
*zp
, *dzp
= VTOZ(dvp
);
1644 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1646 uint64_t acl_obj
, xattr_obj
;
1647 uint64_t xattr_obj_unlinked
= 0;
1651 boolean_t may_delete_now
, delete_now
= FALSE
;
1652 boolean_t unlinked
, toobig
= FALSE
;
1654 pathname_t
*realnmp
= NULL
;
1658 boolean_t waited
= B_FALSE
;
1662 zilog
= zfsvfs
->z_log
;
1664 if (flags
& FIGNORECASE
) {
1674 * Attempt to lock directory; fail if entry doesn't exist.
1676 if (error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1686 if (error
= zfs_zaccess_delete(dzp
, zp
, cr
)) {
1691 * Need to use rmdir for removing directories.
1693 if (vp
->v_type
== VDIR
) {
1694 error
= SET_ERROR(EPERM
);
1698 vnevent_remove(vp
, dvp
, name
, ct
);
1701 dnlc_remove(dvp
, realnmp
->pn_buf
);
1703 dnlc_remove(dvp
, name
);
1705 mutex_enter(&vp
->v_lock
);
1706 may_delete_now
= vp
->v_count
== 1 && !vn_has_cached_data(vp
);
1707 mutex_exit(&vp
->v_lock
);
1710 * We may delete the znode now, or we may put it in the unlinked set;
1711 * it depends on whether we're the last link, and on whether there are
1712 * other holds on the vnode. So we dmu_tx_hold() the right things to
1713 * allow for either case.
1716 tx
= dmu_tx_create(zfsvfs
->z_os
);
1717 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1718 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1719 zfs_sa_upgrade_txholds(tx
, zp
);
1720 zfs_sa_upgrade_txholds(tx
, dzp
);
1721 if (may_delete_now
) {
1723 zp
->z_size
> zp
->z_blksz
* DMU_MAX_DELETEBLKCNT
;
1724 /* if the file is too big, only hold_free a token amount */
1725 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1726 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1729 /* are there any extended attributes? */
1730 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1731 &xattr_obj
, sizeof (xattr_obj
));
1732 if (error
== 0 && xattr_obj
) {
1733 error
= zfs_zget(zfsvfs
, xattr_obj
, &xzp
);
1735 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1736 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1739 mutex_enter(&zp
->z_lock
);
1740 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1741 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1742 mutex_exit(&zp
->z_lock
);
1744 /* charge as an update -- would be nice not to charge at all */
1745 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1748 * Mark this transaction as typically resulting in a net free of space
1750 dmu_tx_mark_netfree(tx
);
1752 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1754 zfs_dirent_unlock(dl
);
1758 if (error
== ERESTART
) {
1772 * Remove the directory entry.
1774 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1783 * Hold z_lock so that we can make sure that the ACL obj
1784 * hasn't changed. Could have been deleted due to
1787 mutex_enter(&zp
->z_lock
);
1788 mutex_enter(&vp
->v_lock
);
1789 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1790 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1791 delete_now
= may_delete_now
&& !toobig
&&
1792 vp
->v_count
== 1 && !vn_has_cached_data(vp
) &&
1793 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1795 mutex_exit(&vp
->v_lock
);
1799 if (xattr_obj_unlinked
) {
1800 ASSERT3U(xzp
->z_links
, ==, 2);
1801 mutex_enter(&xzp
->z_lock
);
1802 xzp
->z_unlinked
= 1;
1804 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zfsvfs
),
1805 &xzp
->z_links
, sizeof (xzp
->z_links
), tx
);
1806 ASSERT3U(error
, ==, 0);
1807 mutex_exit(&xzp
->z_lock
);
1808 zfs_unlinked_add(xzp
, tx
);
1811 error
= sa_remove(zp
->z_sa_hdl
,
1812 SA_ZPL_XATTR(zfsvfs
), tx
);
1814 error
= sa_update(zp
->z_sa_hdl
,
1815 SA_ZPL_XATTR(zfsvfs
), &null_xattr
,
1816 sizeof (uint64_t), tx
);
1819 mutex_enter(&vp
->v_lock
);
1821 ASSERT0(vp
->v_count
);
1822 mutex_exit(&vp
->v_lock
);
1823 mutex_exit(&zp
->z_lock
);
1824 zfs_znode_delete(zp
, tx
);
1825 } else if (unlinked
) {
1826 mutex_exit(&zp
->z_lock
);
1827 zfs_unlinked_add(zp
, tx
);
1831 if (flags
& FIGNORECASE
)
1833 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1840 zfs_dirent_unlock(dl
);
1847 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1848 zil_commit(zilog
, 0);
1855 * Create a new directory and insert it into dvp using the name
1856 * provided. Return a pointer to the inserted directory.
1858 * IN: dvp - vnode of directory to add subdir to.
1859 * dirname - name of new directory.
1860 * vap - attributes of new directory.
1861 * cr - credentials of caller.
1862 * ct - caller context
1863 * flags - case flags
1864 * vsecp - ACL to be set
1866 * OUT: vpp - vnode of created directory.
1868 * RETURN: 0 on success, error code on failure.
1871 * dvp - ctime|mtime updated
1872 * vp - ctime|mtime|atime updated
1876 zfs_mkdir(vnode_t
*dvp
, char *dirname
, vattr_t
*vap
, vnode_t
**vpp
, cred_t
*cr
,
1877 caller_context_t
*ct
, int flags
, vsecattr_t
*vsecp
)
1879 znode_t
*zp
, *dzp
= VTOZ(dvp
);
1880 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1889 gid_t gid
= crgetgid(cr
);
1890 zfs_acl_ids_t acl_ids
;
1891 boolean_t fuid_dirtied
;
1892 boolean_t waited
= B_FALSE
;
1894 ASSERT(vap
->va_type
== VDIR
);
1897 * If we have an ephemeral id, ACL, or XVATTR then
1898 * make sure file system is at proper version
1901 ksid
= crgetsid(cr
, KSID_OWNER
);
1903 uid
= ksid_getid(ksid
);
1906 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1907 (vsecp
|| (vap
->va_mask
& AT_XVATTR
) ||
1908 IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1909 return (SET_ERROR(EINVAL
));
1913 zilog
= zfsvfs
->z_log
;
1915 if (dzp
->z_pflags
& ZFS_XATTR
) {
1917 return (SET_ERROR(EINVAL
));
1920 if (zfsvfs
->z_utf8
&& u8_validate(dirname
,
1921 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1923 return (SET_ERROR(EILSEQ
));
1925 if (flags
& FIGNORECASE
)
1928 if (vap
->va_mask
& AT_XVATTR
) {
1929 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1930 crgetuid(cr
), cr
, vap
->va_type
)) != 0) {
1936 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1937 vsecp
, &acl_ids
)) != 0) {
1942 * First make sure the new directory doesn't exist.
1944 * Existence is checked first to make sure we don't return
1945 * EACCES instead of EEXIST which can cause some applications
1951 if (error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1953 zfs_acl_ids_free(&acl_ids
);
1958 if (error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
)) {
1959 zfs_acl_ids_free(&acl_ids
);
1960 zfs_dirent_unlock(dl
);
1965 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1966 zfs_acl_ids_free(&acl_ids
);
1967 zfs_dirent_unlock(dl
);
1969 return (SET_ERROR(EDQUOT
));
1973 * Add a new entry to the directory.
1975 tx
= dmu_tx_create(zfsvfs
->z_os
);
1976 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
1977 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1978 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1980 zfs_fuid_txhold(zfsvfs
, tx
);
1981 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1982 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
1983 acl_ids
.z_aclp
->z_acl_bytes
);
1986 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1987 ZFS_SA_BASE_ATTR_SIZE
);
1989 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1991 zfs_dirent_unlock(dl
);
1992 if (error
== ERESTART
) {
1998 zfs_acl_ids_free(&acl_ids
);
2007 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
2010 zfs_fuid_sync(zfsvfs
, tx
);
2013 * Now put new name in parent dir.
2015 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
2019 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
2020 if (flags
& FIGNORECASE
)
2022 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
2023 acl_ids
.z_fuidp
, vap
);
2025 zfs_acl_ids_free(&acl_ids
);
2029 zfs_dirent_unlock(dl
);
2031 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2032 zil_commit(zilog
, 0);
2039 * Remove a directory subdir entry. If the current working
2040 * directory is the same as the subdir to be removed, the
2043 * IN: dvp - vnode of directory to remove from.
2044 * name - name of directory to be removed.
2045 * cwd - vnode of current working directory.
2046 * cr - credentials of caller.
2047 * ct - caller context
2048 * flags - case flags
2050 * RETURN: 0 on success, error code on failure.
2053 * dvp - ctime|mtime updated
2057 zfs_rmdir(vnode_t
*dvp
, char *name
, vnode_t
*cwd
, cred_t
*cr
,
2058 caller_context_t
*ct
, int flags
)
2060 znode_t
*dzp
= VTOZ(dvp
);
2063 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
2069 boolean_t waited
= B_FALSE
;
2073 zilog
= zfsvfs
->z_log
;
2075 if (flags
& FIGNORECASE
)
2081 * Attempt to lock directory; fail if entry doesn't exist.
2083 if (error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
2091 if (error
= zfs_zaccess_delete(dzp
, zp
, cr
)) {
2095 if (vp
->v_type
!= VDIR
) {
2096 error
= SET_ERROR(ENOTDIR
);
2101 error
= SET_ERROR(EINVAL
);
2105 vnevent_rmdir(vp
, dvp
, name
, ct
);
2108 * Grab a lock on the directory to make sure that noone is
2109 * trying to add (or lookup) entries while we are removing it.
2111 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
2114 * Grab a lock on the parent pointer to make sure we play well
2115 * with the treewalk and directory rename code.
2117 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
2119 tx
= dmu_tx_create(zfsvfs
->z_os
);
2120 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2121 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2122 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
2123 zfs_sa_upgrade_txholds(tx
, zp
);
2124 zfs_sa_upgrade_txholds(tx
, dzp
);
2125 dmu_tx_mark_netfree(tx
);
2126 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
2128 rw_exit(&zp
->z_parent_lock
);
2129 rw_exit(&zp
->z_name_lock
);
2130 zfs_dirent_unlock(dl
);
2132 if (error
== ERESTART
) {
2143 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2146 uint64_t txtype
= TX_RMDIR
;
2147 if (flags
& FIGNORECASE
)
2149 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2154 rw_exit(&zp
->z_parent_lock
);
2155 rw_exit(&zp
->z_name_lock
);
2157 zfs_dirent_unlock(dl
);
2161 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2162 zil_commit(zilog
, 0);
2169 * Read as many directory entries as will fit into the provided
2170 * buffer from the given directory cursor position (specified in
2171 * the uio structure).
2173 * IN: vp - vnode of directory to read.
2174 * uio - structure supplying read location, range info,
2175 * and return buffer.
2176 * cr - credentials of caller.
2177 * ct - caller context
2178 * flags - case flags
2180 * OUT: uio - updated offset and range, buffer filled.
2181 * eofp - set to true if end-of-file detected.
2183 * RETURN: 0 on success, error code on failure.
2186 * vp - atime updated
2188 * Note that the low 4 bits of the cookie returned by zap is always zero.
2189 * This allows us to use the low range for "special" directory entries:
2190 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2191 * we use the offset 2 for the '.zfs' directory.
2195 zfs_readdir(vnode_t
*vp
, uio_t
*uio
, cred_t
*cr
, int *eofp
,
2196 caller_context_t
*ct
, int flags
)
2198 znode_t
*zp
= VTOZ(vp
);
2202 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2207 zap_attribute_t zap
;
2208 uint_t bytes_wanted
;
2209 uint64_t offset
; /* must be unsigned; checks for < 1 */
2215 boolean_t check_sysattrs
;
2220 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
2221 &parent
, sizeof (parent
))) != 0) {
2227 * If we are not given an eof variable,
2234 * Check for valid iov_len.
2236 if (uio
->uio_iov
->iov_len
<= 0) {
2238 return (SET_ERROR(EINVAL
));
2242 * Quit if directory has been removed (posix)
2244 if ((*eofp
= zp
->z_unlinked
) != 0) {
2251 offset
= uio
->uio_loffset
;
2252 prefetch
= zp
->z_zn_prefetch
;
2255 * Initialize the iterator cursor.
2259 * Start iteration from the beginning of the directory.
2261 zap_cursor_init(&zc
, os
, zp
->z_id
);
2264 * The offset is a serialized cursor.
2266 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2270 * Get space to change directory entries into fs independent format.
2272 iovp
= uio
->uio_iov
;
2273 bytes_wanted
= iovp
->iov_len
;
2274 if (uio
->uio_segflg
!= UIO_SYSSPACE
|| uio
->uio_iovcnt
!= 1) {
2275 bufsize
= bytes_wanted
;
2276 outbuf
= kmem_alloc(bufsize
, KM_SLEEP
);
2277 odp
= (struct dirent64
*)outbuf
;
2279 bufsize
= bytes_wanted
;
2281 odp
= (struct dirent64
*)iovp
->iov_base
;
2283 eodp
= (struct edirent
*)odp
;
2286 * If this VFS supports the system attribute view interface; and
2287 * we're looking at an extended attribute directory; and we care
2288 * about normalization conflicts on this vfs; then we must check
2289 * for normalization conflicts with the sysattr name space.
2291 check_sysattrs
= vfs_has_feature(vp
->v_vfsp
, VFSFT_SYSATTR_VIEWS
) &&
2292 (vp
->v_flag
& V_XATTRDIR
) && zfsvfs
->z_norm
&&
2293 (flags
& V_RDDIR_ENTFLAGS
);
2296 * Transform to file-system independent format
2299 while (outcount
< bytes_wanted
) {
2302 off64_t
*next
= NULL
;
2305 * Special case `.', `..', and `.zfs'.
2308 (void) strcpy(zap
.za_name
, ".");
2309 zap
.za_normalization_conflict
= 0;
2311 } else if (offset
== 1) {
2312 (void) strcpy(zap
.za_name
, "..");
2313 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
;
2323 if (error
= zap_cursor_retrieve(&zc
, &zap
)) {
2324 if ((*eofp
= (error
== ENOENT
)) != 0)
2330 if (zap
.za_integer_length
!= 8 ||
2331 zap
.za_num_integers
!= 1) {
2332 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2333 "entry, obj = %lld, offset = %lld\n",
2334 (u_longlong_t
)zp
->z_id
,
2335 (u_longlong_t
)offset
);
2336 error
= SET_ERROR(ENXIO
);
2340 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2342 * MacOS X can extract the object type here such as:
2343 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2346 if (check_sysattrs
&& !zap
.za_normalization_conflict
) {
2347 zap
.za_normalization_conflict
=
2348 xattr_sysattr_casechk(zap
.za_name
);
2352 if (flags
& V_RDDIR_ACCFILTER
) {
2354 * If we have no access at all, don't include
2355 * this entry in the returned information
2358 if (zfs_zget(zp
->z_zfsvfs
, objnum
, &ezp
) != 0)
2360 if (!zfs_has_access(ezp
, cr
)) {
2367 if (flags
& V_RDDIR_ENTFLAGS
)
2368 reclen
= EDIRENT_RECLEN(strlen(zap
.za_name
));
2370 reclen
= DIRENT64_RECLEN(strlen(zap
.za_name
));
2373 * Will this entry fit in the buffer?
2375 if (outcount
+ reclen
> bufsize
) {
2377 * Did we manage to fit anything in the buffer?
2380 error
= SET_ERROR(EINVAL
);
2385 if (flags
& V_RDDIR_ENTFLAGS
) {
2387 * Add extended flag entry:
2389 eodp
->ed_ino
= objnum
;
2390 eodp
->ed_reclen
= reclen
;
2391 /* NOTE: ed_off is the offset for the *next* entry */
2392 next
= &(eodp
->ed_off
);
2393 eodp
->ed_eflags
= zap
.za_normalization_conflict
?
2394 ED_CASE_CONFLICT
: 0;
2395 (void) strncpy(eodp
->ed_name
, zap
.za_name
,
2396 EDIRENT_NAMELEN(reclen
));
2397 eodp
= (edirent_t
*)((intptr_t)eodp
+ reclen
);
2402 odp
->d_ino
= objnum
;
2403 odp
->d_reclen
= reclen
;
2404 /* NOTE: d_off is the offset for the *next* entry */
2405 next
= &(odp
->d_off
);
2406 (void) strncpy(odp
->d_name
, zap
.za_name
,
2407 DIRENT64_NAMELEN(reclen
));
2408 odp
= (dirent64_t
*)((intptr_t)odp
+ reclen
);
2412 ASSERT(outcount
<= bufsize
);
2414 /* Prefetch znode */
2416 dmu_prefetch(os
, objnum
, 0, 0, 0,
2417 ZIO_PRIORITY_SYNC_READ
);
2421 * Move to the next entry, fill in the previous offset.
2423 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2424 zap_cursor_advance(&zc
);
2425 offset
= zap_cursor_serialize(&zc
);
2432 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2434 if (uio
->uio_segflg
== UIO_SYSSPACE
&& uio
->uio_iovcnt
== 1) {
2435 iovp
->iov_base
+= outcount
;
2436 iovp
->iov_len
-= outcount
;
2437 uio
->uio_resid
-= outcount
;
2438 } else if (error
= uiomove(outbuf
, (long)outcount
, UIO_READ
, uio
)) {
2440 * Reset the pointer.
2442 offset
= uio
->uio_loffset
;
2446 zap_cursor_fini(&zc
);
2447 if (uio
->uio_segflg
!= UIO_SYSSPACE
|| uio
->uio_iovcnt
!= 1)
2448 kmem_free(outbuf
, bufsize
);
2450 if (error
== ENOENT
)
2453 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
2455 uio
->uio_loffset
= offset
;
2460 ulong_t zfs_fsync_sync_cnt
= 4;
2463 zfs_fsync(vnode_t
*vp
, int syncflag
, cred_t
*cr
, caller_context_t
*ct
)
2465 znode_t
*zp
= VTOZ(vp
);
2466 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2469 * Regardless of whether this is required for standards conformance,
2470 * this is the logical behavior when fsync() is called on a file with
2471 * dirty pages. We use B_ASYNC since the ZIL transactions are already
2472 * going to be pushed out as part of the zil_commit().
2474 if (vn_has_cached_data(vp
) && !(syncflag
& FNODSYNC
) &&
2475 (vp
->v_type
== VREG
) && !(IS_SWAPVP(vp
)))
2476 (void) fop_putpage(vp
, 0, (size_t)0, B_ASYNC
, cr
, ct
);
2478 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2480 if (zfsvfs
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2483 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
2491 * Get the requested file attributes and place them in the provided
2494 * IN: vp - vnode of file.
2495 * vap - va_mask identifies requested attributes.
2496 * If AT_XVATTR set, then optional attrs are requested
2497 * flags - ATTR_NOACLCHECK (CIFS server context)
2498 * cr - credentials of caller.
2499 * ct - caller context
2501 * OUT: vap - attribute values.
2503 * RETURN: 0 (always succeeds).
2507 zfs_getattr(vnode_t
*vp
, vattr_t
*vap
, int flags
, cred_t
*cr
,
2508 caller_context_t
*ct
)
2510 znode_t
*zp
= VTOZ(vp
);
2511 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2514 uint64_t mtime
[2], ctime
[2];
2515 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2516 xoptattr_t
*xoap
= NULL
;
2517 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2518 sa_bulk_attr_t bulk
[2];
2524 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2526 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
2527 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
2529 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2535 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2536 * Also, if we are the owner don't bother, since owner should
2537 * always be allowed to read basic attributes of file.
2539 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2540 (vap
->va_uid
!= crgetuid(cr
))) {
2541 if (error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2549 * Return all attributes. It's cheaper to provide the answer
2550 * than to determine whether we were asked the question.
2553 mutex_enter(&zp
->z_lock
);
2554 vap
->va_type
= vp
->v_type
;
2555 vap
->va_mode
= zp
->z_mode
& MODEMASK
;
2556 vap
->va_fsid
= zp
->z_zfsvfs
->z_vfs
->vfs_dev
;
2557 vap
->va_nodeid
= zp
->z_id
;
2558 if ((vp
->v_flag
& VROOT
) && zfs_show_ctldir(zp
))
2559 links
= zp
->z_links
+ 1;
2561 links
= zp
->z_links
;
2562 vap
->va_nlink
= MIN(links
, UINT32_MAX
); /* nlink_t limit! */
2563 vap
->va_size
= zp
->z_size
;
2564 vap
->va_rdev
= vp
->v_rdev
;
2565 vap
->va_seq
= zp
->z_seq
;
2568 * Add in any requested optional attributes and the create time.
2569 * Also set the corresponding bits in the returned attribute bitmap.
2571 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zfsvfs
->z_use_fuids
) {
2572 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2574 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2575 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2578 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2579 xoap
->xoa_readonly
=
2580 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2581 XVA_SET_RTN(xvap
, XAT_READONLY
);
2584 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2586 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2587 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2590 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2592 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2593 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2596 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2597 xoap
->xoa_nounlink
=
2598 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2599 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2602 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2603 xoap
->xoa_immutable
=
2604 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2605 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2608 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2609 xoap
->xoa_appendonly
=
2610 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2611 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2614 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2616 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2617 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2620 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2622 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2623 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2626 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2627 xoap
->xoa_av_quarantined
=
2628 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2629 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2632 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2633 xoap
->xoa_av_modified
=
2634 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2635 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2638 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2639 vp
->v_type
== VREG
) {
2640 zfs_sa_get_scanstamp(zp
, xvap
);
2643 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2646 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zfsvfs
),
2647 times
, sizeof (times
));
2648 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2649 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2652 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2653 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2654 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2656 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2657 xoap
->xoa_generation
= zp
->z_gen
;
2658 XVA_SET_RTN(xvap
, XAT_GEN
);
2661 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2663 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2664 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2667 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2669 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2670 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2674 ZFS_TIME_DECODE(&vap
->va_atime
, zp
->z_atime
);
2675 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2676 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2678 mutex_exit(&zp
->z_lock
);
2680 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2682 if (zp
->z_blksz
== 0) {
2684 * Block size hasn't been set; suggest maximal I/O transfers.
2686 vap
->va_blksize
= zfsvfs
->z_max_blksz
;
2694 * Set the file attributes to the values contained in the
2697 * IN: vp - vnode of file to be modified.
2698 * vap - new attribute values.
2699 * If AT_XVATTR set, then optional attrs are being set
2700 * flags - ATTR_UTIME set if non-default time values provided.
2701 * - ATTR_NOACLCHECK (CIFS context only).
2702 * cr - credentials of caller.
2703 * ct - caller context
2705 * RETURN: 0 on success, error code on failure.
2708 * vp - ctime updated, mtime updated if size changed.
2712 zfs_setattr(vnode_t
*vp
, vattr_t
*vap
, int flags
, cred_t
*cr
,
2713 caller_context_t
*ct
)
2715 znode_t
*zp
= VTOZ(vp
);
2716 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2721 uint_t mask
= vap
->va_mask
;
2722 uint_t saved_mask
= 0;
2725 uint64_t new_uid
, new_gid
;
2727 uint64_t mtime
[2], ctime
[2];
2729 int need_policy
= FALSE
;
2731 zfs_fuid_info_t
*fuidp
= NULL
;
2732 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2735 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2736 boolean_t fuid_dirtied
= B_FALSE
;
2737 sa_bulk_attr_t bulk
[7], xattr_bulk
[7];
2738 int count
= 0, xattr_count
= 0;
2743 if (mask
& AT_NOSET
)
2744 return (SET_ERROR(EINVAL
));
2749 zilog
= zfsvfs
->z_log
;
2752 * Make sure that if we have ephemeral uid/gid or xvattr specified
2753 * that file system is at proper version level
2756 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
2757 (((mask
& AT_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2758 ((mask
& AT_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2759 (mask
& AT_XVATTR
))) {
2761 return (SET_ERROR(EINVAL
));
2764 if (mask
& AT_SIZE
&& vp
->v_type
== VDIR
) {
2766 return (SET_ERROR(EISDIR
));
2769 if (mask
& AT_SIZE
&& vp
->v_type
!= VREG
&& vp
->v_type
!= VFIFO
) {
2771 return (SET_ERROR(EINVAL
));
2775 * If this is an xvattr_t, then get a pointer to the structure of
2776 * optional attributes. If this is NULL, then we have a vattr_t.
2778 xoap
= xva_getxoptattr(xvap
);
2780 xva_init(&tmpxvattr
);
2783 * Immutable files can only alter immutable bit and atime
2785 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2786 ((mask
& (AT_SIZE
|AT_UID
|AT_GID
|AT_MTIME
|AT_MODE
)) ||
2787 ((mask
& AT_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2789 return (SET_ERROR(EPERM
));
2793 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
2797 * Verify timestamps doesn't overflow 32 bits.
2798 * ZFS can handle large timestamps, but 32bit syscalls can't
2799 * handle times greater than 2039. This check should be removed
2800 * once large timestamps are fully supported.
2802 if (mask
& (AT_ATIME
| AT_MTIME
)) {
2803 if (((mask
& AT_ATIME
) && TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2804 ((mask
& AT_MTIME
) && TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2806 return (SET_ERROR(EOVERFLOW
));
2814 /* Can this be moved to before the top label? */
2815 if (zfsvfs
->z_vfs
->vfs_flag
& VFS_RDONLY
) {
2817 return (SET_ERROR(EROFS
));
2821 * First validate permissions
2824 if (mask
& AT_SIZE
) {
2825 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2831 * XXX - Note, we are not providing any open
2832 * mode flags here (like FNDELAY), so we may
2833 * block if there are locks present... this
2834 * should be addressed in openat().
2836 /* XXX - would it be OK to generate a log record here? */
2837 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2843 if (vap
->va_size
== 0)
2844 vnevent_truncate(ZTOV(zp
), ct
);
2847 if (mask
& (AT_ATIME
|AT_MTIME
) ||
2848 ((mask
& AT_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2849 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2850 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2851 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2852 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2853 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2854 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2855 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2859 if (mask
& (AT_UID
|AT_GID
)) {
2860 int idmask
= (mask
& (AT_UID
|AT_GID
));
2865 * NOTE: even if a new mode is being set,
2866 * we may clear S_ISUID/S_ISGID bits.
2869 if (!(mask
& AT_MODE
))
2870 vap
->va_mode
= zp
->z_mode
;
2873 * Take ownership or chgrp to group we are a member of
2876 take_owner
= (mask
& AT_UID
) && (vap
->va_uid
== crgetuid(cr
));
2877 take_group
= (mask
& AT_GID
) &&
2878 zfs_groupmember(zfsvfs
, vap
->va_gid
, cr
);
2881 * If both AT_UID and AT_GID are set then take_owner and
2882 * take_group must both be set in order to allow taking
2885 * Otherwise, send the check through secpolicy_vnode_setattr()
2889 if (((idmask
== (AT_UID
|AT_GID
)) && take_owner
&& take_group
) ||
2890 ((idmask
== AT_UID
) && take_owner
) ||
2891 ((idmask
== AT_GID
) && take_group
)) {
2892 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2893 skipaclchk
, cr
) == 0) {
2895 * Remove setuid/setgid for non-privileged users
2897 secpolicy_setid_clear(vap
, cr
);
2898 trim_mask
= (mask
& (AT_UID
|AT_GID
));
2907 mutex_enter(&zp
->z_lock
);
2908 oldva
.va_mode
= zp
->z_mode
;
2909 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2910 if (mask
& AT_XVATTR
) {
2912 * Update xvattr mask to include only those attributes
2913 * that are actually changing.
2915 * the bits will be restored prior to actually setting
2916 * the attributes so the caller thinks they were set.
2918 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2919 if (xoap
->xoa_appendonly
!=
2920 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2923 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2924 XVA_SET_REQ(&tmpxvattr
, XAT_APPENDONLY
);
2928 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2929 if (xoap
->xoa_nounlink
!=
2930 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2933 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2934 XVA_SET_REQ(&tmpxvattr
, XAT_NOUNLINK
);
2938 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2939 if (xoap
->xoa_immutable
!=
2940 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2943 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2944 XVA_SET_REQ(&tmpxvattr
, XAT_IMMUTABLE
);
2948 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2949 if (xoap
->xoa_nodump
!=
2950 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2953 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2954 XVA_SET_REQ(&tmpxvattr
, XAT_NODUMP
);
2958 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2959 if (xoap
->xoa_av_modified
!=
2960 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2963 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2964 XVA_SET_REQ(&tmpxvattr
, XAT_AV_MODIFIED
);
2968 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2969 if ((vp
->v_type
!= VREG
&&
2970 xoap
->xoa_av_quarantined
) ||
2971 xoap
->xoa_av_quarantined
!=
2972 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2975 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2976 XVA_SET_REQ(&tmpxvattr
, XAT_AV_QUARANTINED
);
2980 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2981 mutex_exit(&zp
->z_lock
);
2983 return (SET_ERROR(EPERM
));
2986 if (need_policy
== FALSE
&&
2987 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
2988 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
2993 mutex_exit(&zp
->z_lock
);
2995 if (mask
& AT_MODE
) {
2996 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
2997 err
= secpolicy_setid_setsticky_clear(vp
, vap
,
3003 trim_mask
|= AT_MODE
;
3011 * If trim_mask is set then take ownership
3012 * has been granted or write_acl is present and user
3013 * has the ability to modify mode. In that case remove
3014 * UID|GID and or MODE from mask so that
3015 * secpolicy_vnode_setattr() doesn't revoke it.
3019 saved_mask
= vap
->va_mask
;
3020 vap
->va_mask
&= ~trim_mask
;
3022 err
= secpolicy_vnode_setattr(cr
, vp
, vap
, &oldva
, flags
,
3023 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
3030 vap
->va_mask
|= saved_mask
;
3034 * secpolicy_vnode_setattr, or take ownership may have
3037 mask
= vap
->va_mask
;
3039 if ((mask
& (AT_UID
| AT_GID
))) {
3040 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
3041 &xattr_obj
, sizeof (xattr_obj
));
3043 if (err
== 0 && xattr_obj
) {
3044 err
= zfs_zget(zp
->z_zfsvfs
, xattr_obj
, &attrzp
);
3048 if (mask
& AT_UID
) {
3049 new_uid
= zfs_fuid_create(zfsvfs
,
3050 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
3051 if (new_uid
!= zp
->z_uid
&&
3052 zfs_fuid_overquota(zfsvfs
, B_FALSE
, new_uid
)) {
3054 VN_RELE(ZTOV(attrzp
));
3055 err
= SET_ERROR(EDQUOT
);
3060 if (mask
& AT_GID
) {
3061 new_gid
= zfs_fuid_create(zfsvfs
, (uint64_t)vap
->va_gid
,
3062 cr
, ZFS_GROUP
, &fuidp
);
3063 if (new_gid
!= zp
->z_gid
&&
3064 zfs_fuid_overquota(zfsvfs
, B_TRUE
, new_gid
)) {
3066 VN_RELE(ZTOV(attrzp
));
3067 err
= SET_ERROR(EDQUOT
);
3072 tx
= dmu_tx_create(zfsvfs
->z_os
);
3074 if (mask
& AT_MODE
) {
3075 uint64_t pmode
= zp
->z_mode
;
3077 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
3079 if (zp
->z_zfsvfs
->z_acl_mode
== ZFS_ACL_RESTRICTED
&&
3080 !(zp
->z_pflags
& ZFS_ACL_TRIVIAL
)) {
3081 err
= SET_ERROR(EPERM
);
3085 if (err
= zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
))
3088 mutex_enter(&zp
->z_lock
);
3089 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
3091 * Are we upgrading ACL from old V0 format
3094 if (zfsvfs
->z_version
>= ZPL_VERSION_FUID
&&
3095 zfs_znode_acl_version(zp
) ==
3096 ZFS_ACL_VERSION_INITIAL
) {
3097 dmu_tx_hold_free(tx
, acl_obj
, 0,
3099 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3100 0, aclp
->z_acl_bytes
);
3102 dmu_tx_hold_write(tx
, acl_obj
, 0,
3105 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3106 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3107 0, aclp
->z_acl_bytes
);
3109 mutex_exit(&zp
->z_lock
);
3110 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3112 if ((mask
& AT_XVATTR
) &&
3113 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3114 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3116 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3120 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
3123 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3125 zfs_fuid_txhold(zfsvfs
, tx
);
3127 zfs_sa_upgrade_txholds(tx
, zp
);
3129 err
= dmu_tx_assign(tx
, TXG_WAIT
);
3135 * Set each attribute requested.
3136 * We group settings according to the locks they need to acquire.
3138 * Note: you cannot set ctime directly, although it will be
3139 * updated as a side-effect of calling this function.
3143 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3144 mutex_enter(&zp
->z_acl_lock
);
3145 mutex_enter(&zp
->z_lock
);
3147 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
3148 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
3151 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3152 mutex_enter(&attrzp
->z_acl_lock
);
3153 mutex_enter(&attrzp
->z_lock
);
3154 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3155 SA_ZPL_FLAGS(zfsvfs
), NULL
, &attrzp
->z_pflags
,
3156 sizeof (attrzp
->z_pflags
));
3159 if (mask
& (AT_UID
|AT_GID
)) {
3161 if (mask
& AT_UID
) {
3162 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zfsvfs
), NULL
,
3163 &new_uid
, sizeof (new_uid
));
3164 zp
->z_uid
= new_uid
;
3166 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3167 SA_ZPL_UID(zfsvfs
), NULL
, &new_uid
,
3169 attrzp
->z_uid
= new_uid
;
3173 if (mask
& AT_GID
) {
3174 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zfsvfs
),
3175 NULL
, &new_gid
, sizeof (new_gid
));
3176 zp
->z_gid
= new_gid
;
3178 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3179 SA_ZPL_GID(zfsvfs
), NULL
, &new_gid
,
3181 attrzp
->z_gid
= new_gid
;
3184 if (!(mask
& AT_MODE
)) {
3185 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
),
3186 NULL
, &new_mode
, sizeof (new_mode
));
3187 new_mode
= zp
->z_mode
;
3189 err
= zfs_acl_chown_setattr(zp
);
3192 err
= zfs_acl_chown_setattr(attrzp
);
3197 if (mask
& AT_MODE
) {
3198 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
), NULL
,
3199 &new_mode
, sizeof (new_mode
));
3200 zp
->z_mode
= new_mode
;
3201 ASSERT3U((uintptr_t)aclp
, !=, (uintptr_t)NULL
);
3202 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
3204 if (zp
->z_acl_cached
)
3205 zfs_acl_free(zp
->z_acl_cached
);
3206 zp
->z_acl_cached
= aclp
;
3211 if (mask
& AT_ATIME
) {
3212 ZFS_TIME_ENCODE(&vap
->va_atime
, zp
->z_atime
);
3213 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
,
3214 &zp
->z_atime
, sizeof (zp
->z_atime
));
3217 if (mask
& AT_MTIME
) {
3218 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3219 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
3220 mtime
, sizeof (mtime
));
3223 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3224 if (mask
& AT_SIZE
&& !(mask
& AT_MTIME
)) {
3225 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
),
3226 NULL
, mtime
, sizeof (mtime
));
3227 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3228 &ctime
, sizeof (ctime
));
3229 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
3231 } else if (mask
!= 0) {
3232 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3233 &ctime
, sizeof (ctime
));
3234 zfs_tstamp_update_setup(zp
, STATE_CHANGED
, mtime
, ctime
,
3237 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3238 SA_ZPL_CTIME(zfsvfs
), NULL
,
3239 &ctime
, sizeof (ctime
));
3240 zfs_tstamp_update_setup(attrzp
, STATE_CHANGED
,
3241 mtime
, ctime
, B_TRUE
);
3245 * Do this after setting timestamps to prevent timestamp
3246 * update from toggling bit
3249 if (xoap
&& (mask
& AT_XVATTR
)) {
3252 * restore trimmed off masks
3253 * so that return masks can be set for caller.
3256 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_APPENDONLY
)) {
3257 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3259 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_NOUNLINK
)) {
3260 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3262 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_IMMUTABLE
)) {
3263 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3265 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_NODUMP
)) {
3266 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3268 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_AV_MODIFIED
)) {
3269 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3271 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_AV_QUARANTINED
)) {
3272 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3275 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3276 ASSERT(vp
->v_type
== VREG
);
3278 zfs_xvattr_set(zp
, xvap
, tx
);
3282 zfs_fuid_sync(zfsvfs
, tx
);
3285 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3287 mutex_exit(&zp
->z_lock
);
3288 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3289 mutex_exit(&zp
->z_acl_lock
);
3292 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3293 mutex_exit(&attrzp
->z_acl_lock
);
3294 mutex_exit(&attrzp
->z_lock
);
3297 if (err
== 0 && attrzp
) {
3298 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3304 VN_RELE(ZTOV(attrzp
));
3310 zfs_fuid_info_free(fuidp
);
3316 if (err
== ERESTART
)
3319 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3324 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3325 zil_commit(zilog
, 0);
3331 typedef struct zfs_zlock
{
3332 krwlock_t
*zl_rwlock
; /* lock we acquired */
3333 znode_t
*zl_znode
; /* znode we held */
3334 struct zfs_zlock
*zl_next
; /* next in list */
3338 * Drop locks and release vnodes that were held by zfs_rename_lock().
3341 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3345 while ((zl
= *zlpp
) != NULL
) {
3346 if (zl
->zl_znode
!= NULL
)
3347 VN_RELE(ZTOV(zl
->zl_znode
));
3348 rw_exit(zl
->zl_rwlock
);
3349 *zlpp
= zl
->zl_next
;
3350 kmem_free(zl
, sizeof (*zl
));
3355 * Search back through the directory tree, using the ".." entries.
3356 * Lock each directory in the chain to prevent concurrent renames.
3357 * Fail any attempt to move a directory into one of its own descendants.
3358 * XXX - z_parent_lock can overlap with map or grow locks
3361 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3365 uint64_t rootid
= zp
->z_zfsvfs
->z_root
;
3366 uint64_t oidp
= zp
->z_id
;
3367 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3368 krw_t rw
= RW_WRITER
;
3371 * First pass write-locks szp and compares to zp->z_id.
3372 * Later passes read-lock zp and compare to zp->z_parent.
3375 if (!rw_tryenter(rwlp
, rw
)) {
3377 * Another thread is renaming in this path.
3378 * Note that if we are a WRITER, we don't have any
3379 * parent_locks held yet.
3381 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3383 * Drop our locks and restart
3385 zfs_rename_unlock(&zl
);
3389 rwlp
= &szp
->z_parent_lock
;
3394 * Wait for other thread to drop its locks
3400 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3401 zl
->zl_rwlock
= rwlp
;
3402 zl
->zl_znode
= NULL
;
3403 zl
->zl_next
= *zlpp
;
3406 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3407 return (SET_ERROR(EINVAL
));
3409 if (oidp
== rootid
) /* We've hit the top */
3412 if (rw
== RW_READER
) { /* i.e. not the first pass */
3413 int error
= zfs_zget(zp
->z_zfsvfs
, oidp
, &zp
);
3418 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zp
->z_zfsvfs
),
3419 &oidp
, sizeof (oidp
));
3420 rwlp
= &zp
->z_parent_lock
;
3423 } while (zp
->z_id
!= sdzp
->z_id
);
3429 * Move an entry from the provided source directory to the target
3430 * directory. Change the entry name as indicated.
3432 * IN: sdvp - Source directory containing the "old entry".
3433 * snm - Old entry name.
3434 * tdvp - Target directory to contain the "new entry".
3435 * tnm - New entry name.
3436 * cr - credentials of caller.
3437 * ct - caller context
3438 * flags - case flags
3440 * RETURN: 0 on success, error code on failure.
3443 * sdvp,tdvp - ctime|mtime updated
3447 zfs_rename(vnode_t
*sdvp
, char *snm
, vnode_t
*tdvp
, char *tnm
, cred_t
*cr
,
3448 caller_context_t
*ct
, int flags
)
3450 znode_t
*tdzp
, *szp
, *tzp
;
3451 znode_t
*sdzp
= VTOZ(sdvp
);
3452 zfsvfs_t
*zfsvfs
= sdzp
->z_zfsvfs
;
3455 zfs_dirlock_t
*sdl
, *tdl
;
3458 int cmp
, serr
, terr
;
3459 int error
= 0, rm_err
= 0;
3461 boolean_t waited
= B_FALSE
;
3464 ZFS_VERIFY_ZP(sdzp
);
3465 zilog
= zfsvfs
->z_log
;
3468 * Make sure we have the real vp for the target directory.
3470 if (fop_realvp(tdvp
, &realvp
, ct
) == 0)
3474 ZFS_VERIFY_ZP(tdzp
);
3477 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3478 * ctldir appear to have the same v_vfsp.
3480 if (tdzp
->z_zfsvfs
!= zfsvfs
|| zfsctl_is_node(tdvp
)) {
3482 return (SET_ERROR(EXDEV
));
3485 if (zfsvfs
->z_utf8
&& u8_validate(tnm
,
3486 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3488 return (SET_ERROR(EILSEQ
));
3491 if (flags
& FIGNORECASE
)
3500 * This is to prevent the creation of links into attribute space
3501 * by renaming a linked file into/outof an attribute directory.
3502 * See the comment in zfs_link() for why this is considered bad.
3504 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3506 return (SET_ERROR(EINVAL
));
3510 * Lock source and target directory entries. To prevent deadlock,
3511 * a lock ordering must be defined. We lock the directory with
3512 * the smallest object id first, or if it's a tie, the one with
3513 * the lexically first name.
3515 if (sdzp
->z_id
< tdzp
->z_id
) {
3517 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3521 * First compare the two name arguments without
3522 * considering any case folding.
3524 int nofold
= (zfsvfs
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3526 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3527 ASSERT(error
== 0 || !zfsvfs
->z_utf8
);
3530 * POSIX: "If the old argument and the new argument
3531 * both refer to links to the same existing file,
3532 * the rename() function shall return successfully
3533 * and perform no other action."
3539 * If the file system is case-folding, then we may
3540 * have some more checking to do. A case-folding file
3541 * system is either supporting mixed case sensitivity
3542 * access or is completely case-insensitive. Note
3543 * that the file system is always case preserving.
3545 * In mixed sensitivity mode case sensitive behavior
3546 * is the default. FIGNORECASE must be used to
3547 * explicitly request case insensitive behavior.
3549 * If the source and target names provided differ only
3550 * by case (e.g., a request to rename 'tim' to 'Tim'),
3551 * we will treat this as a special case in the
3552 * case-insensitive mode: as long as the source name
3553 * is an exact match, we will allow this to proceed as
3554 * a name-change request.
3556 if ((zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
||
3557 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&&
3558 flags
& FIGNORECASE
)) &&
3559 u8_strcmp(snm
, tnm
, 0, zfsvfs
->z_norm
, U8_UNICODE_LATEST
,
3562 * case preserving rename request, require exact
3571 * If the source and destination directories are the same, we should
3572 * grab the z_name_lock of that directory only once.
3576 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3580 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3581 ZEXISTS
| zflg
, NULL
, NULL
);
3582 terr
= zfs_dirent_lock(&tdl
,
3583 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3585 terr
= zfs_dirent_lock(&tdl
,
3586 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3587 serr
= zfs_dirent_lock(&sdl
,
3588 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3594 * Source entry invalid or not there.
3597 zfs_dirent_unlock(tdl
);
3603 rw_exit(&sdzp
->z_name_lock
);
3605 if (strcmp(snm
, "..") == 0)
3606 serr
= SET_ERROR(EINVAL
);
3611 zfs_dirent_unlock(sdl
);
3615 rw_exit(&sdzp
->z_name_lock
);
3617 if (strcmp(tnm
, "..") == 0)
3618 terr
= SET_ERROR(EINVAL
);
3624 * Must have write access at the source to remove the old entry
3625 * and write access at the target to create the new entry.
3626 * Note that if target and source are the same, this can be
3627 * done in a single check.
3630 if (error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
))
3633 if (ZTOV(szp
)->v_type
== VDIR
) {
3635 * Check to make sure rename is valid.
3636 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3638 if (error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
))
3643 * Does target exist?
3647 * Source and target must be the same type.
3649 if (ZTOV(szp
)->v_type
== VDIR
) {
3650 if (ZTOV(tzp
)->v_type
!= VDIR
) {
3651 error
= SET_ERROR(ENOTDIR
);
3655 if (ZTOV(tzp
)->v_type
== VDIR
) {
3656 error
= SET_ERROR(EISDIR
);
3661 * POSIX dictates that when the source and target
3662 * entries refer to the same file object, rename
3663 * must do nothing and exit without error.
3665 if (szp
->z_id
== tzp
->z_id
) {
3671 vnevent_pre_rename_src(ZTOV(szp
), sdvp
, snm
, ct
);
3673 vnevent_pre_rename_dest(ZTOV(tzp
), tdvp
, tnm
, ct
);
3676 * notify the target directory if it is not the same
3677 * as source directory.
3680 vnevent_pre_rename_dest_dir(tdvp
, ZTOV(szp
), tnm
, ct
);
3683 tx
= dmu_tx_create(zfsvfs
->z_os
);
3684 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3685 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3686 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3687 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3689 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3690 zfs_sa_upgrade_txholds(tx
, tdzp
);
3693 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3694 zfs_sa_upgrade_txholds(tx
, tzp
);
3697 zfs_sa_upgrade_txholds(tx
, szp
);
3698 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
3699 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3702 zfs_rename_unlock(&zl
);
3703 zfs_dirent_unlock(sdl
);
3704 zfs_dirent_unlock(tdl
);
3707 rw_exit(&sdzp
->z_name_lock
);
3712 if (error
== ERESTART
) {
3723 if (tzp
) /* Attempt to remove the existing target */
3724 error
= rm_err
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3727 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3729 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3731 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zfsvfs
),
3732 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3735 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3737 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3738 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3739 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3742 * Update path information for the target vnode
3744 vn_renamepath(tdvp
, ZTOV(szp
), tnm
,
3748 * At this point, we have successfully created
3749 * the target name, but have failed to remove
3750 * the source name. Since the create was done
3751 * with the ZRENAMING flag, there are
3752 * complications; for one, the link count is
3753 * wrong. The easiest way to deal with this
3754 * is to remove the newly created target, and
3755 * return the original error. This must
3756 * succeed; fortunately, it is very unlikely to
3757 * fail, since we just created it.
3759 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3760 ZRENAMING
, NULL
), ==, 0);
3767 if (tzp
&& rm_err
== 0)
3768 vnevent_rename_dest(ZTOV(tzp
), tdvp
, tnm
, ct
);
3771 vnevent_rename_src(ZTOV(szp
), sdvp
, snm
, ct
);
3772 /* notify the target dir if it is not the same as source dir */
3774 vnevent_rename_dest_dir(tdvp
, ct
);
3778 zfs_rename_unlock(&zl
);
3780 zfs_dirent_unlock(sdl
);
3781 zfs_dirent_unlock(tdl
);
3784 rw_exit(&sdzp
->z_name_lock
);
3791 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3792 zil_commit(zilog
, 0);
3799 * Insert the indicated symbolic reference entry into the directory.
3801 * IN: dvp - Directory to contain new symbolic link.
3802 * link - Name for new symlink entry.
3803 * vap - Attributes of new entry.
3804 * cr - credentials of caller.
3805 * ct - caller context
3806 * flags - case flags
3808 * RETURN: 0 on success, error code on failure.
3811 * dvp - ctime|mtime updated
3815 zfs_symlink(vnode_t
*dvp
, char *name
, vattr_t
*vap
, char *link
, cred_t
*cr
,
3816 caller_context_t
*ct
, int flags
)
3818 znode_t
*zp
, *dzp
= VTOZ(dvp
);
3821 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
3823 uint64_t len
= strlen(link
);
3826 zfs_acl_ids_t acl_ids
;
3827 boolean_t fuid_dirtied
;
3828 uint64_t txtype
= TX_SYMLINK
;
3829 boolean_t waited
= B_FALSE
;
3831 ASSERT(vap
->va_type
== VLNK
);
3835 zilog
= zfsvfs
->z_log
;
3837 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
3838 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3840 return (SET_ERROR(EILSEQ
));
3842 if (flags
& FIGNORECASE
)
3845 if (len
> MAXPATHLEN
) {
3847 return (SET_ERROR(ENAMETOOLONG
));
3850 if ((error
= zfs_acl_ids_create(dzp
, 0,
3851 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3857 * Attempt to lock directory; fail if entry already exists.
3859 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3861 zfs_acl_ids_free(&acl_ids
);
3866 if (error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
)) {
3867 zfs_acl_ids_free(&acl_ids
);
3868 zfs_dirent_unlock(dl
);
3873 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
3874 zfs_acl_ids_free(&acl_ids
);
3875 zfs_dirent_unlock(dl
);
3877 return (SET_ERROR(EDQUOT
));
3879 tx
= dmu_tx_create(zfsvfs
->z_os
);
3880 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3881 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3882 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3883 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3884 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3885 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3886 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3887 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3888 acl_ids
.z_aclp
->z_acl_bytes
);
3891 zfs_fuid_txhold(zfsvfs
, tx
);
3892 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3894 zfs_dirent_unlock(dl
);
3895 if (error
== ERESTART
) {
3901 zfs_acl_ids_free(&acl_ids
);
3908 * Create a new object for the symlink.
3909 * for version 4 ZPL datsets the symlink will be an SA attribute
3911 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3914 zfs_fuid_sync(zfsvfs
, tx
);
3916 mutex_enter(&zp
->z_lock
);
3918 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zfsvfs
),
3921 zfs_sa_symlink(zp
, link
, len
, tx
);
3922 mutex_exit(&zp
->z_lock
);
3925 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
3926 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3928 * Insert the new object into the directory.
3930 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3932 if (flags
& FIGNORECASE
)
3934 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3936 zfs_acl_ids_free(&acl_ids
);
3940 zfs_dirent_unlock(dl
);
3944 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3945 zil_commit(zilog
, 0);
3952 * Return, in the buffer contained in the provided uio structure,
3953 * the symbolic path referred to by vp.
3955 * IN: vp - vnode of symbolic link.
3956 * uio - structure to contain the link path.
3957 * cr - credentials of caller.
3958 * ct - caller context
3960 * OUT: uio - structure containing the link path.
3962 * RETURN: 0 on success, error code on failure.
3965 * vp - atime updated
3969 zfs_readlink(vnode_t
*vp
, uio_t
*uio
, cred_t
*cr
, caller_context_t
*ct
)
3971 znode_t
*zp
= VTOZ(vp
);
3972 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
3978 mutex_enter(&zp
->z_lock
);
3980 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3981 SA_ZPL_SYMLINK(zfsvfs
), uio
);
3983 error
= zfs_sa_readlink(zp
, uio
);
3984 mutex_exit(&zp
->z_lock
);
3986 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
3993 * Insert a new entry into directory tdvp referencing svp.
3995 * IN: tdvp - Directory to contain new entry.
3996 * svp - vnode of new entry.
3997 * name - name of new entry.
3998 * cr - credentials of caller.
3999 * ct - caller context
4001 * RETURN: 0 on success, error code on failure.
4004 * tdvp - ctime|mtime updated
4005 * svp - ctime updated
4009 zfs_link(vnode_t
*tdvp
, vnode_t
*svp
, char *name
, cred_t
*cr
,
4010 caller_context_t
*ct
, int flags
)
4012 znode_t
*dzp
= VTOZ(tdvp
);
4014 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
4023 boolean_t waited
= B_FALSE
;
4025 ASSERT(tdvp
->v_type
== VDIR
);
4029 zilog
= zfsvfs
->z_log
;
4031 if (fop_realvp(svp
, &realvp
, ct
) == 0)
4035 * POSIX dictates that we return EPERM here.
4036 * Better choices include ENOTSUP or EISDIR.
4038 if (svp
->v_type
== VDIR
) {
4040 return (SET_ERROR(EPERM
));
4047 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4048 * ctldir appear to have the same v_vfsp.
4050 if (szp
->z_zfsvfs
!= zfsvfs
|| zfsctl_is_node(svp
)) {
4052 return (SET_ERROR(EXDEV
));
4055 /* Prevent links to .zfs/shares files */
4057 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
4058 &parent
, sizeof (uint64_t))) != 0) {
4062 if (parent
== zfsvfs
->z_shares_dir
) {
4064 return (SET_ERROR(EPERM
));
4067 if (zfsvfs
->z_utf8
&& u8_validate(name
,
4068 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
4070 return (SET_ERROR(EILSEQ
));
4072 if (flags
& FIGNORECASE
)
4076 * We do not support links between attributes and non-attributes
4077 * because of the potential security risk of creating links
4078 * into "normal" file space in order to circumvent restrictions
4079 * imposed in attribute space.
4081 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
4083 return (SET_ERROR(EINVAL
));
4087 owner
= zfs_fuid_map_id(zfsvfs
, szp
->z_uid
, cr
, ZFS_OWNER
);
4088 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
4090 return (SET_ERROR(EPERM
));
4093 if (error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
)) {
4100 * Attempt to lock directory; fail if entry already exists.
4102 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
4108 tx
= dmu_tx_create(zfsvfs
->z_os
);
4109 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
4110 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
4111 zfs_sa_upgrade_txholds(tx
, szp
);
4112 zfs_sa_upgrade_txholds(tx
, dzp
);
4113 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
4115 zfs_dirent_unlock(dl
);
4116 if (error
== ERESTART
) {
4127 error
= zfs_link_create(dl
, szp
, tx
, 0);
4130 uint64_t txtype
= TX_LINK
;
4131 if (flags
& FIGNORECASE
)
4133 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
4138 zfs_dirent_unlock(dl
);
4141 vnevent_link(svp
, ct
);
4144 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4145 zil_commit(zilog
, 0);
4152 * zfs_null_putapage() is used when the file system has been force
4153 * unmounted. It just drops the pages.
4157 zfs_null_putapage(vnode_t
*vp
, page_t
*pp
, uoff_t
*offp
,
4158 size_t *lenp
, int flags
, cred_t
*cr
)
4160 pvn_write_done(pp
, B_INVAL
|B_FORCE
|B_ERROR
);
4165 * Push a page out to disk, klustering if possible.
4167 * IN: vp - file to push page to.
4168 * pp - page to push.
4169 * flags - additional flags.
4170 * cr - credentials of caller.
4172 * OUT: offp - start of range pushed.
4173 * lenp - len of range pushed.
4175 * RETURN: 0 on success, error code on failure.
4177 * NOTE: callers must have locked the page to be pushed. On
4178 * exit, the page (and all other pages in the kluster) must be
4183 zfs_putapage(vnode_t
*vp
, page_t
*pp
, uoff_t
*offp
,
4184 size_t *lenp
, int flags
, cred_t
*cr
)
4186 znode_t
*zp
= VTOZ(vp
);
4187 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4196 * If our blocksize is bigger than the page size, try to kluster
4197 * multiple pages so that we write a full block (thus avoiding
4198 * a read-modify-write).
4200 if (off
< zp
->z_size
&& zp
->z_blksz
> PAGESIZE
) {
4201 klen
= P2ROUNDUP((ulong_t
)zp
->z_blksz
, PAGESIZE
);
4202 koff
= ISP2(klen
) ? P2ALIGN(off
, (uoff_t
)klen
) : 0;
4203 ASSERT(koff
<= zp
->z_size
);
4204 if (koff
+ klen
> zp
->z_size
)
4205 klen
= P2ROUNDUP(zp
->z_size
- koff
, (uint64_t)PAGESIZE
);
4206 pp
= pvn_write_kluster(vp
, pp
, &off
, &len
, koff
, klen
, flags
);
4208 ASSERT3U(btop(len
), ==, btopr(len
));
4211 * Can't push pages past end-of-file.
4213 if (off
>= zp
->z_size
) {
4214 /* ignore all pages */
4217 } else if (off
+ len
> zp
->z_size
) {
4218 int npages
= btopr(zp
->z_size
- off
);
4221 page_list_break(&pp
, &trunc
, npages
);
4222 /* ignore pages past end of file */
4224 pvn_write_done(trunc
, flags
);
4225 len
= zp
->z_size
- off
;
4228 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
4229 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
4230 err
= SET_ERROR(EDQUOT
);
4233 tx
= dmu_tx_create(zfsvfs
->z_os
);
4234 dmu_tx_hold_write(tx
, zp
->z_id
, off
, len
);
4236 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4237 zfs_sa_upgrade_txholds(tx
, zp
);
4238 err
= dmu_tx_assign(tx
, TXG_WAIT
);
4244 if (zp
->z_blksz
<= PAGESIZE
) {
4245 caddr_t va
= zfs_map_page(pp
, S_READ
);
4246 ASSERT3U(len
, <=, PAGESIZE
);
4247 dmu_write(zfsvfs
->z_os
, zp
->z_id
, off
, len
, va
, tx
);
4248 zfs_unmap_page(pp
, va
);
4250 err
= dmu_write_pages(zfsvfs
->z_os
, zp
->z_id
, off
, len
, pp
, tx
);
4254 uint64_t mtime
[2], ctime
[2];
4255 sa_bulk_attr_t bulk
[3];
4258 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
4260 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
4262 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
4264 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
4266 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
4268 zfs_log_write(zfsvfs
->z_log
, tx
, TX_WRITE
, zp
, off
, len
, 0);
4273 pvn_write_done(pp
, (err
? B_ERROR
: 0) | flags
);
4283 * Copy the portion of the file indicated from pages into the file.
4284 * The pages are stored in a page list attached to the files vnode.
4286 * IN: vp - vnode of file to push page data to.
4287 * off - position in file to put data.
4288 * len - amount of data to write.
4289 * flags - flags to control the operation.
4290 * cr - credentials of caller.
4291 * ct - caller context.
4293 * RETURN: 0 on success, error code on failure.
4296 * vp - ctime|mtime updated
4300 zfs_putpage(vnode_t
*vp
, offset_t off
, size_t len
, int flags
, cred_t
*cr
,
4301 caller_context_t
*ct
)
4303 znode_t
*zp
= VTOZ(vp
);
4304 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4316 * There's nothing to do if no data is cached.
4318 if (!vn_has_cached_data(vp
)) {
4324 * Align this request to the file block size in case we kluster.
4325 * XXX - this can result in pretty aggresive locking, which can
4326 * impact simultanious read/write access. One option might be
4327 * to break up long requests (len == 0) into block-by-block
4328 * operations to get narrower locking.
4330 blksz
= zp
->z_blksz
;
4332 io_off
= P2ALIGN_TYPED(off
, blksz
, uoff_t
);
4335 if (len
> 0 && ISP2(blksz
))
4336 io_len
= P2ROUNDUP_TYPED(len
+ (off
- io_off
), blksz
, size_t);
4342 * Search the entire vp list for pages >= io_off.
4344 rl
= zfs_range_lock(zp
, io_off
, UINT64_MAX
, RL_WRITER
);
4345 error
= pvn_vplist_dirty(vp
, io_off
, zfs_putapage
, flags
, cr
);
4348 rl
= zfs_range_lock(zp
, io_off
, io_len
, RL_WRITER
);
4350 if (off
> zp
->z_size
) {
4351 /* past end of file */
4352 zfs_range_unlock(rl
);
4357 len
= MIN(io_len
, P2ROUNDUP(zp
->z_size
, PAGESIZE
) - io_off
);
4359 for (off
= io_off
; io_off
< off
+ len
; io_off
+= io_len
) {
4360 if ((flags
& B_INVAL
) || ((flags
& B_ASYNC
) == 0)) {
4361 pp
= page_lookup(&vp
->v_object
, io_off
,
4362 (flags
& (B_INVAL
| B_FREE
)) ? SE_EXCL
: SE_SHARED
);
4364 pp
= page_lookup_nowait(&vp
->v_object
, io_off
,
4365 (flags
& B_FREE
) ? SE_EXCL
: SE_SHARED
);
4368 if (pp
!= NULL
&& pvn_getdirty(pp
, flags
)) {
4372 * Found a dirty page to push
4374 err
= zfs_putapage(vp
, pp
, &io_off
, &io_len
, flags
, cr
);
4382 zfs_range_unlock(rl
);
4383 if ((flags
& B_ASYNC
) == 0 || zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4384 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
4391 zfs_inactive(vnode_t
*vp
, cred_t
*cr
, caller_context_t
*ct
)
4393 znode_t
*zp
= VTOZ(vp
);
4394 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4397 rw_enter(&zfsvfs
->z_teardown_inactive_lock
, RW_READER
);
4398 if (zp
->z_sa_hdl
== NULL
) {
4400 * The fs has been unmounted, or we did a
4401 * suspend/resume and this file no longer exists.
4403 if (vn_has_cached_data(vp
)) {
4404 (void) pvn_vplist_dirty(vp
, 0, zfs_null_putapage
,
4408 mutex_enter(&zp
->z_lock
);
4409 mutex_enter(&vp
->v_lock
);
4410 ASSERT(vp
->v_count
== 1);
4412 mutex_exit(&vp
->v_lock
);
4413 mutex_exit(&zp
->z_lock
);
4414 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4420 * Attempt to push any data in the page cache. If this fails
4421 * we will get kicked out later in zfs_zinactive().
4423 if (vn_has_cached_data(vp
)) {
4424 (void) pvn_vplist_dirty(vp
, 0, zfs_putapage
, B_INVAL
|B_ASYNC
,
4428 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4429 dmu_tx_t
*tx
= dmu_tx_create(zfsvfs
->z_os
);
4431 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4432 zfs_sa_upgrade_txholds(tx
, zp
);
4433 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4437 mutex_enter(&zp
->z_lock
);
4438 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zfsvfs
),
4439 (void *)&zp
->z_atime
, sizeof (zp
->z_atime
), tx
);
4440 zp
->z_atime_dirty
= 0;
4441 mutex_exit(&zp
->z_lock
);
4447 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4451 * Bounds-check the seek operation.
4453 * IN: vp - vnode seeking within
4454 * ooff - old file offset
4455 * noffp - pointer to new file offset
4456 * ct - caller context
4458 * RETURN: 0 on success, EINVAL if new offset invalid.
4462 zfs_seek(vnode_t
*vp
, offset_t ooff
, offset_t
*noffp
,
4463 caller_context_t
*ct
)
4465 if (vp
->v_type
== VDIR
)
4467 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4471 * Pre-filter the generic locking function to trap attempts to place
4472 * a mandatory lock on a memory mapped file.
4475 zfs_frlock(vnode_t
*vp
, int cmd
, flock64_t
*bfp
, int flag
, offset_t offset
,
4476 flk_callback_t
*flk_cbp
, cred_t
*cr
, caller_context_t
*ct
)
4478 znode_t
*zp
= VTOZ(vp
);
4479 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4485 * We are following the UFS semantics with respect to mapcnt
4486 * here: If we see that the file is mapped already, then we will
4487 * return an error, but we don't worry about races between this
4488 * function and zfs_map().
4490 if (zp
->z_mapcnt
> 0 && MANDMODE(zp
->z_mode
)) {
4492 return (SET_ERROR(EAGAIN
));
4495 return (fs_frlock(vp
, cmd
, bfp
, flag
, offset
, flk_cbp
, cr
, ct
));
4499 * If we can't find a page in the cache, we will create a new page
4500 * and fill it with file data. For efficiency, we may try to fill
4501 * multiple pages at once (klustering) to fill up the supplied page
4502 * list. Note that the pages to be filled are held with an exclusive
4503 * lock to prevent access by other threads while they are being filled.
4506 zfs_fillpage(vnode_t
*vp
, uoff_t off
, struct seg
*seg
,
4507 caddr_t addr
, page_t
*pl
[], size_t plsz
, enum seg_rw rw
)
4509 znode_t
*zp
= VTOZ(vp
);
4510 page_t
*pp
, *cur_pp
;
4511 objset_t
*os
= zp
->z_zfsvfs
->z_os
;
4512 uoff_t io_off
, total
;
4516 if (plsz
== PAGESIZE
|| zp
->z_blksz
<= PAGESIZE
) {
4518 * We only have a single page, don't bother klustering
4522 pp
= page_create_va(&vp
->v_object
, io_off
, io_len
,
4523 PG_EXCL
| PG_WAIT
, seg
, addr
);
4526 * Try to find enough pages to fill the page list
4528 pp
= pvn_read_kluster(vp
, off
, seg
, addr
, &io_off
,
4529 &io_len
, off
, plsz
, 0);
4533 * The page already exists, nothing to do here.
4540 * Fill the pages in the kluster.
4543 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4546 ASSERT3U(io_off
, ==, cur_pp
->p_offset
);
4547 va
= zfs_map_page(cur_pp
, S_WRITE
);
4548 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4550 zfs_unmap_page(cur_pp
, va
);
4552 /* On error, toss the entire kluster */
4553 pvn_read_done(pp
, B_ERROR
);
4554 /* convert checksum errors into IO errors */
4556 err
= SET_ERROR(EIO
);
4559 cur_pp
= cur_pp
->p_next
;
4563 * Fill in the page list array from the kluster starting
4564 * from the desired offset `off'.
4565 * NOTE: the page list will always be null terminated.
4567 pvn_plist_init(pp
, pl
, plsz
, off
, io_len
, rw
);
4568 ASSERT(pl
== NULL
|| (*pl
)->p_offset
== off
);
4574 * Return pointers to the pages for the file region [off, off + len]
4575 * in the pl array. If plsz is greater than len, this function may
4576 * also return page pointers from after the specified region
4577 * (i.e. the region [off, off + plsz]). These additional pages are
4578 * only returned if they are already in the cache, or were created as
4579 * part of a klustered read.
4581 * IN: vp - vnode of file to get data from.
4582 * off - position in file to get data from.
4583 * len - amount of data to retrieve.
4584 * plsz - length of provided page list.
4585 * seg - segment to obtain pages for.
4586 * addr - virtual address of fault.
4587 * rw - mode of created pages.
4588 * cr - credentials of caller.
4589 * ct - caller context.
4591 * OUT: protp - protection mode of created pages.
4592 * pl - list of pages created.
4594 * RETURN: 0 on success, error code on failure.
4597 * vp - atime updated
4601 zfs_getpage(vnode_t
*vp
, offset_t off
, size_t len
, uint_t
*protp
,
4602 page_t
*pl
[], size_t plsz
, struct seg
*seg
, caddr_t addr
,
4603 enum seg_rw rw
, cred_t
*cr
, caller_context_t
*ct
)
4605 znode_t
*zp
= VTOZ(vp
);
4606 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4610 /* we do our own caching, faultahead is unnecessary */
4613 else if (len
> plsz
)
4616 len
= P2ROUNDUP(len
, PAGESIZE
);
4617 ASSERT(plsz
>= len
);
4626 * Loop through the requested range [off, off + len) looking
4627 * for pages. If we don't find a page, we will need to create
4628 * a new page and fill it with data from the file.
4631 if (*pl
= page_lookup(&vp
->v_object
, off
, SE_SHARED
))
4633 else if (err
= zfs_fillpage(vp
, off
, seg
, addr
, pl
, plsz
, rw
))
4636 ASSERT3U((*pl
)->p_offset
, ==, off
);
4640 ASSERT3U(len
, >=, PAGESIZE
);
4643 ASSERT3U(plsz
, >=, PAGESIZE
);
4650 * Fill out the page array with any pages already in the cache.
4653 (*pl
++ = page_lookup_nowait(&vp
->v_object
, off
, SE_SHARED
))) {
4660 * Release any pages we have previously locked.
4665 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
4675 * Request a memory map for a section of a file. This code interacts
4676 * with common code and the VM system as follows:
4678 * - common code calls mmap(), which ends up in smmap_common()
4679 * - this calls fop_map(), which takes you into (say) zfs
4680 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4681 * - segvn_create() creates the new segment and calls fop_addmap()
4682 * - zfs_addmap() updates z_mapcnt
4686 zfs_map(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t
*addrp
,
4687 size_t len
, uchar_t prot
, uchar_t maxprot
, uint_t flags
, cred_t
*cr
,
4688 caller_context_t
*ct
)
4690 znode_t
*zp
= VTOZ(vp
);
4691 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4692 segvn_crargs_t vn_a
;
4699 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
4702 if ((prot
& PROT_WRITE
) && (zp
->z_pflags
&
4703 (ZFS_IMMUTABLE
| ZFS_APPENDONLY
))) {
4705 return (SET_ERROR(EPERM
));
4708 if ((prot
& (PROT_READ
| PROT_EXEC
)) &&
4709 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4711 return (SET_ERROR(EACCES
));
4714 if (vp
->v_flag
& VNOMAP
) {
4716 return (SET_ERROR(ENOSYS
));
4719 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4721 return (SET_ERROR(ENXIO
));
4724 if (vp
->v_type
!= VREG
) {
4726 return (SET_ERROR(ENODEV
));
4730 * If file is locked, disallow mapping.
4732 if (MANDMODE(zp
->z_mode
) && vn_has_flocks(vp
)) {
4734 return (SET_ERROR(EAGAIN
));
4738 error
= choose_addr(as
, addrp
, len
, off
, ADDR_VACALIGN
, flags
);
4746 vn_a
.offset
= (uoff_t
)off
;
4747 vn_a
.type
= flags
& MAP_TYPE
;
4749 vn_a
.maxprot
= maxprot
;
4752 vn_a
.flags
= flags
& ~MAP_TYPE
;
4754 vn_a
.lgrp_mem_policy_flags
= 0;
4756 error
= as_map(as
, *addrp
, len
, segvn_create
, &vn_a
);
4765 zfs_addmap(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t addr
,
4766 size_t len
, uchar_t prot
, uchar_t maxprot
, uint_t flags
, cred_t
*cr
,
4767 caller_context_t
*ct
)
4769 uint64_t pages
= btopr(len
);
4771 atomic_add_64(&VTOZ(vp
)->z_mapcnt
, pages
);
4776 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4777 * more accurate mtime for the associated file. Since we don't have a way of
4778 * detecting when the data was actually modified, we have to resort to
4779 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4780 * last page is pushed. The problem occurs when the msync() call is omitted,
4781 * which by far the most common case:
4789 * putpage() via fsflush
4791 * If we wait until fsflush to come along, we can have a modification time that
4792 * is some arbitrary point in the future. In order to prevent this in the
4793 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4798 zfs_delmap(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t addr
,
4799 size_t len
, uint_t prot
, uint_t maxprot
, uint_t flags
, cred_t
*cr
,
4800 caller_context_t
*ct
)
4802 uint64_t pages
= btopr(len
);
4804 ASSERT3U(VTOZ(vp
)->z_mapcnt
, >=, pages
);
4805 atomic_add_64(&VTOZ(vp
)->z_mapcnt
, -pages
);
4807 if ((flags
& MAP_SHARED
) && (prot
& PROT_WRITE
) &&
4808 vn_has_cached_data(vp
))
4809 (void) fop_putpage(vp
, off
, len
, B_ASYNC
, cr
, ct
);
4815 * Free or allocate space in a file. Currently, this function only
4816 * supports the `F_FREESP' command. However, this command is somewhat
4817 * misnamed, as its functionality includes the ability to allocate as
4818 * well as free space.
4820 * IN: vp - vnode of file to free data in.
4821 * cmd - action to take (only F_FREESP supported).
4822 * bfp - section of file to free/alloc.
4823 * flag - current file open mode flags.
4824 * offset - current file offset.
4825 * cr - credentials of caller [UNUSED].
4826 * ct - caller context.
4828 * RETURN: 0 on success, error code on failure.
4831 * vp - ctime|mtime updated
4835 zfs_space(vnode_t
*vp
, int cmd
, flock64_t
*bfp
, int flag
,
4836 offset_t offset
, cred_t
*cr
, caller_context_t
*ct
)
4838 znode_t
*zp
= VTOZ(vp
);
4839 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4846 if (cmd
!= F_FREESP
) {
4848 return (SET_ERROR(EINVAL
));
4852 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
4853 * callers might not be able to detect properly that we are read-only,
4854 * so check it explicitly here.
4856 if (zfsvfs
->z_vfs
->vfs_flag
& VFS_RDONLY
) {
4858 return (SET_ERROR(EROFS
));
4861 if (error
= convoff(vp
, bfp
, 0, offset
)) {
4866 if (bfp
->l_len
< 0) {
4868 return (SET_ERROR(EINVAL
));
4872 len
= bfp
->l_len
; /* 0 means from off to end of file */
4874 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4876 if (error
== 0 && off
== 0 && len
== 0)
4877 vnevent_truncate(ZTOV(zp
), ct
);
4885 zfs_fid(vnode_t
*vp
, fid_t
*fidp
, caller_context_t
*ct
)
4887 znode_t
*zp
= VTOZ(vp
);
4888 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4891 uint64_t object
= zp
->z_id
;
4898 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zfsvfs
),
4899 &gen64
, sizeof (uint64_t))) != 0) {
4904 gen
= (uint32_t)gen64
;
4906 size
= (zfsvfs
->z_parent
!= zfsvfs
) ? LONG_FID_LEN
: SHORT_FID_LEN
;
4907 if (fidp
->fid_len
< size
) {
4908 fidp
->fid_len
= size
;
4910 return (SET_ERROR(ENOSPC
));
4913 zfid
= (zfid_short_t
*)fidp
;
4915 zfid
->zf_len
= size
;
4917 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4918 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4920 /* Must have a non-zero generation number to distinguish from .zfs */
4923 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4924 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4926 if (size
== LONG_FID_LEN
) {
4927 uint64_t objsetid
= dmu_objset_id(zfsvfs
->z_os
);
4930 zlfid
= (zfid_long_t
*)fidp
;
4932 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
4933 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
4935 /* XXX - this should be the generation number for the objset */
4936 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
4937 zlfid
->zf_setgen
[i
] = 0;
4945 zfs_pathconf(vnode_t
*vp
, int cmd
, ulong_t
*valp
, cred_t
*cr
,
4946 caller_context_t
*ct
)
4958 case _PC_FILESIZEBITS
:
4962 case _PC_XATTR_EXISTS
:
4964 zfsvfs
= zp
->z_zfsvfs
;
4968 error
= zfs_dirent_lock(&dl
, zp
, "", &xzp
,
4969 ZXATTR
| ZEXISTS
| ZSHARED
, NULL
, NULL
);
4971 zfs_dirent_unlock(dl
);
4972 if (!zfs_dirempty(xzp
))
4975 } else if (error
== ENOENT
) {
4977 * If there aren't extended attributes, it's the
4978 * same as having zero of them.
4985 case _PC_SATTR_ENABLED
:
4986 case _PC_SATTR_EXISTS
:
4987 *valp
= vfs_has_feature(vp
->v_vfsp
, VFSFT_SYSATTR_VIEWS
) &&
4988 (vp
->v_type
== VREG
|| vp
->v_type
== VDIR
);
4991 case _PC_ACCESS_FILTERING
:
4992 *valp
= vfs_has_feature(vp
->v_vfsp
, VFSFT_ACCESS_FILTER
) &&
4996 case _PC_ACL_ENABLED
:
4997 *valp
= _ACL_ACE_ENABLED
;
5000 case _PC_MIN_HOLE_SIZE
:
5001 *valp
= (ulong_t
)SPA_MINBLOCKSIZE
;
5004 case _PC_TIMESTAMP_RESOLUTION
:
5005 /* nanosecond timestamp resolution */
5010 return (fs_pathconf(vp
, cmd
, valp
, cr
, ct
));
5016 zfs_getsecattr(vnode_t
*vp
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
,
5017 caller_context_t
*ct
)
5019 znode_t
*zp
= VTOZ(vp
);
5020 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
5022 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
5026 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
5034 zfs_setsecattr(vnode_t
*vp
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
,
5035 caller_context_t
*ct
)
5037 znode_t
*zp
= VTOZ(vp
);
5038 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
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);
5056 * The smallest read we may consider to loan out an arcbuf.
5057 * This must be a power of 2.
5059 int zcr_blksz_min
= (1 << 10); /* 1K */
5061 * If set to less than the file block size, allow loaning out of an
5062 * arcbuf for a partial block read. This must be a power of 2.
5064 int zcr_blksz_max
= (1 << 17); /* 128K */
5068 zfs_reqzcbuf(vnode_t
*vp
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
,
5069 caller_context_t
*ct
)
5071 znode_t
*zp
= VTOZ(vp
);
5072 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
5073 int max_blksz
= zfsvfs
->z_max_blksz
;
5074 uio_t
*uio
= &xuio
->xu_uio
;
5075 ssize_t size
= uio
->uio_resid
;
5076 offset_t offset
= uio
->uio_loffset
;
5081 int preamble
, postamble
;
5083 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
5084 return (SET_ERROR(EINVAL
));
5091 * Loan out an arc_buf for write if write size is bigger than
5092 * max_blksz, and the file's block size is also max_blksz.
5095 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
5097 return (SET_ERROR(EINVAL
));
5100 * Caller requests buffers for write before knowing where the
5101 * write offset might be (e.g. NFS TCP write).
5106 preamble
= P2PHASE(offset
, blksz
);
5108 preamble
= blksz
- preamble
;
5113 postamble
= P2PHASE(size
, blksz
);
5116 fullblk
= size
/ blksz
;
5117 (void) dmu_xuio_init(xuio
,
5118 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
5119 DTRACE_PROBE3(zfs_reqzcbuf_align
, int, preamble
,
5120 int, postamble
, int,
5121 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
5124 * Have to fix iov base/len for partial buffers. They
5125 * currently represent full arc_buf's.
5128 /* data begins in the middle of the arc_buf */
5129 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5132 (void) dmu_xuio_add(xuio
, abuf
,
5133 blksz
- preamble
, preamble
);
5136 for (i
= 0; i
< fullblk
; i
++) {
5137 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5140 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
5144 /* data ends in the middle of the arc_buf */
5145 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5148 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
5153 * Loan out an arc_buf for read if the read size is larger than
5154 * the current file block size. Block alignment is not
5155 * considered. Partial arc_buf will be loaned out for read.
5157 blksz
= zp
->z_blksz
;
5158 if (blksz
< zcr_blksz_min
)
5159 blksz
= zcr_blksz_min
;
5160 if (blksz
> zcr_blksz_max
)
5161 blksz
= zcr_blksz_max
;
5162 /* avoid potential complexity of dealing with it */
5163 if (blksz
> max_blksz
) {
5165 return (SET_ERROR(EINVAL
));
5168 maxsize
= zp
->z_size
- uio
->uio_loffset
;
5172 if (size
< blksz
|| vn_has_cached_data(vp
)) {
5174 return (SET_ERROR(EINVAL
));
5179 return (SET_ERROR(EINVAL
));
5182 uio
->uio_extflg
= UIO_XUIO
;
5183 XUIO_XUZC_RW(xuio
) = ioflag
;
5190 zfs_retzcbuf(vnode_t
*vp
, xuio_t
*xuio
, cred_t
*cr
, caller_context_t
*ct
)
5194 int ioflag
= XUIO_XUZC_RW(xuio
);
5196 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
5198 i
= dmu_xuio_cnt(xuio
);
5200 abuf
= dmu_xuio_arcbuf(xuio
, i
);
5202 * if abuf == NULL, it must be a write buffer
5203 * that has been returned in zfs_write().
5206 dmu_return_arcbuf(abuf
);
5207 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
5210 dmu_xuio_fini(xuio
);
5215 * Predeclare these here so that the compiler assumes that
5216 * this is an "old style" function declaration that does
5217 * not include arguments => we won't get type mismatch errors
5218 * in the initializations that follow.
5220 static int zfs_inval();
5221 static int zfs_isdir();
5226 return (SET_ERROR(EINVAL
));
5232 return (SET_ERROR(EISDIR
));
5236 * Directory vnode operations
5238 const struct vnodeops zfs_dvnodeops
= {
5240 .vop_open
= zfs_open
,
5241 .vop_close
= zfs_close
,
5242 .vop_read
= zfs_isdir
,
5243 .vop_write
= zfs_isdir
,
5244 .vop_ioctl
= zfs_ioctl
,
5245 .vop_getattr
= zfs_getattr
,
5246 .vop_setattr
= zfs_setattr
,
5247 .vop_access
= zfs_access
,
5248 .vop_lookup
= zfs_lookup
,
5249 .vop_create
= zfs_create
,
5250 .vop_remove
= zfs_remove
,
5251 .vop_link
= zfs_link
,
5252 .vop_rename
= zfs_rename
,
5253 .vop_mkdir
= zfs_mkdir
,
5254 .vop_rmdir
= zfs_rmdir
,
5255 .vop_readdir
= zfs_readdir
,
5256 .vop_symlink
= zfs_symlink
,
5257 .vop_fsync
= zfs_fsync
,
5258 .vop_inactive
= zfs_inactive
,
5260 .vop_seek
= zfs_seek
,
5261 .vop_pathconf
= zfs_pathconf
,
5262 .vop_getsecattr
= zfs_getsecattr
,
5263 .vop_setsecattr
= zfs_setsecattr
,
5264 .vop_vnevent
= fs_vnevent_support
,
5268 * Regular file vnode operations
5270 const struct vnodeops zfs_fvnodeops
= {
5272 .vop_open
= zfs_open
,
5273 .vop_close
= zfs_close
,
5274 .vop_read
= zfs_read
,
5275 .vop_write
= zfs_write
,
5276 .vop_ioctl
= zfs_ioctl
,
5277 .vop_getattr
= zfs_getattr
,
5278 .vop_setattr
= zfs_setattr
,
5279 .vop_access
= zfs_access
,
5280 .vop_lookup
= zfs_lookup
,
5281 .vop_rename
= zfs_rename
,
5282 .vop_fsync
= zfs_fsync
,
5283 .vop_inactive
= zfs_inactive
,
5285 .vop_seek
= zfs_seek
,
5286 .vop_frlock
= zfs_frlock
,
5287 .vop_space
= zfs_space
,
5288 .vop_getpage
= zfs_getpage
,
5289 .vop_putpage
= zfs_putpage
,
5291 .vop_addmap
= zfs_addmap
,
5292 .vop_delmap
= zfs_delmap
,
5293 .vop_pathconf
= zfs_pathconf
,
5294 .vop_getsecattr
= zfs_getsecattr
,
5295 .vop_setsecattr
= zfs_setsecattr
,
5296 .vop_vnevent
= fs_vnevent_support
,
5297 .vop_reqzcbuf
= zfs_reqzcbuf
,
5298 .vop_retzcbuf
= zfs_retzcbuf
,
5302 * Symbolic link vnode operations
5304 const struct vnodeops zfs_symvnodeops
= {
5306 .vop_getattr
= zfs_getattr
,
5307 .vop_setattr
= zfs_setattr
,
5308 .vop_access
= zfs_access
,
5309 .vop_rename
= zfs_rename
,
5310 .vop_readlink
= zfs_readlink
,
5311 .vop_inactive
= zfs_inactive
,
5313 .vop_pathconf
= zfs_pathconf
,
5314 .vop_vnevent
= fs_vnevent_support
,
5318 * special share hidden files vnode operations
5320 const struct vnodeops zfs_sharevnodeops
= {
5322 .vop_getattr
= zfs_getattr
,
5323 .vop_access
= zfs_access
,
5324 .vop_inactive
= zfs_inactive
,
5326 .vop_pathconf
= zfs_pathconf
,
5327 .vop_getsecattr
= zfs_getsecattr
,
5328 .vop_setsecattr
= zfs_setsecattr
,
5329 .vop_vnevent
= fs_vnevent_support
,
5333 * Extended attribute directory vnode operations
5335 * These ops are identical to the directory vnode
5336 * operations except for restricted operations:
5340 * Note that there are other restrictions embedded in:
5341 * zfs_create() - restrict type to VREG
5342 * zfs_link() - no links into/out of attribute space
5343 * zfs_rename() - no moves into/out of attribute space
5345 const struct vnodeops zfs_xdvnodeops
= {
5347 .vop_open
= zfs_open
,
5348 .vop_close
= zfs_close
,
5349 .vop_ioctl
= zfs_ioctl
,
5350 .vop_getattr
= zfs_getattr
,
5351 .vop_setattr
= zfs_setattr
,
5352 .vop_access
= zfs_access
,
5353 .vop_lookup
= zfs_lookup
,
5354 .vop_create
= zfs_create
,
5355 .vop_remove
= zfs_remove
,
5356 .vop_link
= zfs_link
,
5357 .vop_rename
= zfs_rename
,
5358 .vop_mkdir
= zfs_inval
,
5359 .vop_rmdir
= zfs_rmdir
,
5360 .vop_readdir
= zfs_readdir
,
5361 .vop_symlink
= zfs_inval
,
5362 .vop_fsync
= zfs_fsync
,
5363 .vop_inactive
= zfs_inactive
,
5365 .vop_seek
= zfs_seek
,
5366 .vop_pathconf
= zfs_pathconf
,
5367 .vop_getsecattr
= zfs_getsecattr
,
5368 .vop_setsecattr
= zfs_setsecattr
,
5369 .vop_vnevent
= fs_vnevent_support
,
5373 * Error vnode operations
5375 const struct vnodeops zfs_evnodeops
= {
5377 .vop_inactive
= zfs_inactive
,
5378 .vop_pathconf
= zfs_pathconf
,