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_NOTHROTTLE in addition to 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_NOTHROTTLE : 0) | 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
;
1157 * TX_WRITE2 relies on the data previously
1158 * written by the TX_WRITE that caused
1159 * EALREADY. We zero out the BP because
1160 * it is the old, currently-on-disk BP,
1161 * so there's no need to zio_flush() its
1162 * vdevs (flushing would needlesly hurt
1163 * performance, and doesn't work on
1173 zfs_get_done(zgd
, error
);
1180 zfs_access(vnode_t
*vp
, int mode
, int flag
, cred_t
*cr
,
1181 caller_context_t
*ct
)
1183 znode_t
*zp
= VTOZ(vp
);
1184 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
1190 if (flag
& V_ACE_MASK
)
1191 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1193 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1200 * If vnode is for a device return a specfs vnode instead.
1203 specvp_check(vnode_t
**vpp
, cred_t
*cr
)
1207 if (IS_DEVVP(*vpp
)) {
1210 svp
= specvp(*vpp
, (*vpp
)->v_rdev
, (*vpp
)->v_type
, cr
);
1213 error
= SET_ERROR(ENOSYS
);
1221 * Lookup an entry in a directory, or an extended attribute directory.
1222 * If it exists, return a held vnode reference for it.
1224 * IN: dvp - vnode of directory to search.
1225 * nm - name of entry to lookup.
1226 * pnp - full pathname to lookup [UNUSED].
1227 * flags - LOOKUP_XATTR set if looking for an attribute.
1228 * rdir - root directory vnode [UNUSED].
1229 * cr - credentials of caller.
1230 * ct - caller context
1231 * direntflags - directory lookup flags
1232 * realpnp - returned pathname.
1234 * OUT: vpp - vnode of located entry, NULL if not found.
1236 * RETURN: 0 on success, error code on failure.
1243 zfs_lookup(vnode_t
*dvp
, char *nm
, vnode_t
**vpp
, struct pathname
*pnp
,
1244 int flags
, vnode_t
*rdir
, cred_t
*cr
, caller_context_t
*ct
,
1245 int *direntflags
, pathname_t
*realpnp
)
1247 znode_t
*zdp
= VTOZ(dvp
);
1248 zfsvfs_t
*zfsvfs
= zdp
->z_zfsvfs
;
1252 * Fast path lookup, however we must skip DNLC lookup
1253 * for case folding or normalizing lookups because the
1254 * DNLC code only stores the passed in name. This means
1255 * creating 'a' and removing 'A' on a case insensitive
1256 * file system would work, but DNLC still thinks 'a'
1257 * exists and won't let you create it again on the next
1258 * pass through fast path.
1260 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1262 if (dvp
->v_type
!= VDIR
) {
1263 return (SET_ERROR(ENOTDIR
));
1264 } else if (zdp
->z_sa_hdl
== NULL
) {
1265 return (SET_ERROR(EIO
));
1268 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1269 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1276 } else if (!zdp
->z_zfsvfs
->z_norm
&&
1277 (zdp
->z_zfsvfs
->z_case
== ZFS_CASE_SENSITIVE
)) {
1279 vnode_t
*tvp
= dnlc_lookup(dvp
, nm
);
1282 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1287 if (tvp
== DNLC_NO_VNODE
) {
1289 return (SET_ERROR(ENOENT
));
1292 return (specvp_check(vpp
, cr
));
1298 DTRACE_PROBE2(zfs__fastpath__lookup__miss
, vnode_t
*, dvp
, char *, nm
);
1305 if (flags
& LOOKUP_XATTR
) {
1307 * If the xattr property is off, refuse the lookup request.
1309 if (!(zfsvfs
->z_vfs
->vfs_flag
& VFS_XATTR
)) {
1311 return (SET_ERROR(EINVAL
));
1315 * We don't allow recursive attributes..
1316 * Maybe someday we will.
1318 if (zdp
->z_pflags
& ZFS_XATTR
) {
1320 return (SET_ERROR(EINVAL
));
1323 if (error
= zfs_get_xattrdir(VTOZ(dvp
), vpp
, cr
, flags
)) {
1329 * Do we have permission to get into attribute directory?
1332 if (error
= zfs_zaccess(VTOZ(*vpp
), ACE_EXECUTE
, 0,
1342 if (dvp
->v_type
!= VDIR
) {
1344 return (SET_ERROR(ENOTDIR
));
1348 * Check accessibility of directory.
1351 if (error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
)) {
1356 if (zfsvfs
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1357 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1359 return (SET_ERROR(EILSEQ
));
1362 error
= zfs_dirlook(zdp
, nm
, vpp
, flags
, direntflags
, realpnp
);
1364 error
= specvp_check(vpp
, cr
);
1371 * Attempt to create a new entry in a directory. If the entry
1372 * already exists, truncate the file if permissible, else return
1373 * an error. Return the vp of the created or trunc'd file.
1375 * IN: dvp - vnode of directory to put new file entry in.
1376 * name - name of new file entry.
1377 * vap - attributes of new file.
1378 * excl - flag indicating exclusive or non-exclusive mode.
1379 * mode - mode to open file with.
1380 * cr - credentials of caller.
1381 * flag - large file flag [UNUSED].
1382 * ct - caller context
1383 * vsecp - ACL to be set
1385 * OUT: vpp - vnode of created or trunc'd entry.
1387 * RETURN: 0 on success, error code on failure.
1390 * dvp - ctime|mtime updated if new entry created
1391 * vp - ctime|mtime always, atime if new
1396 zfs_create(vnode_t
*dvp
, char *name
, vattr_t
*vap
, vcexcl_t excl
,
1397 int mode
, vnode_t
**vpp
, cred_t
*cr
, int flag
, caller_context_t
*ct
,
1400 znode_t
*zp
, *dzp
= VTOZ(dvp
);
1401 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1409 gid_t gid
= crgetgid(cr
);
1410 zfs_acl_ids_t acl_ids
;
1411 boolean_t fuid_dirtied
;
1412 boolean_t have_acl
= B_FALSE
;
1413 boolean_t waited
= B_FALSE
;
1416 * If we have an ephemeral id, ACL, or XVATTR then
1417 * make sure file system is at proper version
1420 ksid
= crgetsid(cr
, KSID_OWNER
);
1422 uid
= ksid_getid(ksid
);
1426 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1427 (vsecp
|| (vap
->va_mask
& VATTR_XVATTR
) ||
1428 IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1429 return (SET_ERROR(EINVAL
));
1434 zilog
= zfsvfs
->z_log
;
1436 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
1437 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1439 return (SET_ERROR(EILSEQ
));
1442 if (vap
->va_mask
& VATTR_XVATTR
) {
1443 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1444 crgetuid(cr
), cr
, vap
->va_type
)) != 0) {
1452 if ((vap
->va_mode
& VSVTX
) && secpolicy_vnode_stky_modify(cr
))
1453 vap
->va_mode
&= ~VSVTX
;
1455 if (*name
== '\0') {
1457 * Null component name refers to the directory itself.
1464 /* possible VN_HOLD(zp) */
1467 if (flag
& FIGNORECASE
)
1470 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1474 zfs_acl_ids_free(&acl_ids
);
1475 if (strcmp(name
, "..") == 0)
1476 error
= SET_ERROR(EISDIR
);
1486 * Create a new file object and update the directory
1489 if (error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
)) {
1491 zfs_acl_ids_free(&acl_ids
);
1496 * We only support the creation of regular files in
1497 * extended attribute directories.
1500 if ((dzp
->z_pflags
& ZFS_XATTR
) &&
1501 (vap
->va_type
!= VREG
)) {
1503 zfs_acl_ids_free(&acl_ids
);
1504 error
= SET_ERROR(EINVAL
);
1508 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1509 cr
, vsecp
, &acl_ids
)) != 0)
1513 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1514 zfs_acl_ids_free(&acl_ids
);
1515 error
= SET_ERROR(EDQUOT
);
1519 tx
= dmu_tx_create(os
);
1521 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1522 ZFS_SA_BASE_ATTR_SIZE
);
1524 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1526 zfs_fuid_txhold(zfsvfs
, tx
);
1527 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1528 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1529 if (!zfsvfs
->z_use_sa
&&
1530 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1531 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1532 0, acl_ids
.z_aclp
->z_acl_bytes
);
1534 error
= dmu_tx_assign(tx
,
1535 (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
1537 zfs_dirent_unlock(dl
);
1538 if (error
== ERESTART
) {
1544 zfs_acl_ids_free(&acl_ids
);
1549 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1552 zfs_fuid_sync(zfsvfs
, tx
);
1554 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1555 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1556 if (flag
& FIGNORECASE
)
1558 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1559 vsecp
, acl_ids
.z_fuidp
, vap
);
1560 zfs_acl_ids_free(&acl_ids
);
1563 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1566 zfs_acl_ids_free(&acl_ids
);
1570 * A directory entry already exists for this name.
1573 * Can't truncate an existing file if in exclusive mode.
1576 error
= SET_ERROR(EEXIST
);
1580 * Can't open a directory for writing.
1582 if ((ZTOV(zp
)->v_type
== VDIR
) && (mode
& S_IWRITE
)) {
1583 error
= SET_ERROR(EISDIR
);
1587 * Verify requested access to file.
1589 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1593 mutex_enter(&dzp
->z_lock
);
1595 mutex_exit(&dzp
->z_lock
);
1598 * Truncate regular files if requested.
1600 if ((ZTOV(zp
)->v_type
== VREG
) &&
1601 (vap
->va_mask
& VATTR_SIZE
) && (vap
->va_size
== 0)) {
1602 /* we can't hold any locks when calling zfs_freesp() */
1603 zfs_dirent_unlock(dl
);
1605 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1607 vnevent_create(ZTOV(zp
), ct
);
1614 zfs_dirent_unlock(dl
);
1621 error
= specvp_check(vpp
, cr
);
1624 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1625 zil_commit(zilog
, 0);
1632 * Remove an entry from a directory.
1634 * IN: dvp - vnode of directory to remove entry from.
1635 * name - name of entry to remove.
1636 * cr - credentials of caller.
1637 * ct - caller context
1638 * flags - case flags
1640 * RETURN: 0 on success, error code on failure.
1644 * vp - ctime (if nlink > 0)
1647 uint64_t null_xattr
= 0;
1651 zfs_remove(vnode_t
*dvp
, char *name
, cred_t
*cr
, caller_context_t
*ct
,
1654 znode_t
*zp
, *dzp
= VTOZ(dvp
);
1657 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1659 uint64_t acl_obj
, xattr_obj
;
1660 uint64_t xattr_obj_unlinked
= 0;
1664 boolean_t may_delete_now
, delete_now
= FALSE
;
1665 boolean_t unlinked
, toobig
= FALSE
;
1667 pathname_t
*realnmp
= NULL
;
1671 boolean_t waited
= B_FALSE
;
1675 zilog
= zfsvfs
->z_log
;
1677 if (flags
& FIGNORECASE
) {
1687 * Attempt to lock directory; fail if entry doesn't exist.
1689 if (error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1699 if (error
= zfs_zaccess_delete(dzp
, zp
, cr
)) {
1704 * Need to use rmdir for removing directories.
1706 if (vp
->v_type
== VDIR
) {
1707 error
= SET_ERROR(EPERM
);
1711 vnevent_remove(vp
, dvp
, name
, ct
);
1714 dnlc_remove(dvp
, realnmp
->pn_buf
);
1716 dnlc_remove(dvp
, name
);
1718 mutex_enter(&vp
->v_lock
);
1719 may_delete_now
= vp
->v_count
== 1 && !vn_has_cached_data(vp
);
1720 mutex_exit(&vp
->v_lock
);
1723 * We may delete the znode now, or we may put it in the unlinked set;
1724 * it depends on whether we're the last link, and on whether there are
1725 * other holds on the vnode. So we dmu_tx_hold() the right things to
1726 * allow for either case.
1729 tx
= dmu_tx_create(zfsvfs
->z_os
);
1730 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1731 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1732 zfs_sa_upgrade_txholds(tx
, zp
);
1733 zfs_sa_upgrade_txholds(tx
, dzp
);
1734 if (may_delete_now
) {
1736 zp
->z_size
> zp
->z_blksz
* DMU_MAX_DELETEBLKCNT
;
1737 /* if the file is too big, only hold_free a token amount */
1738 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1739 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1742 /* are there any extended attributes? */
1743 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1744 &xattr_obj
, sizeof (xattr_obj
));
1745 if (error
== 0 && xattr_obj
) {
1746 error
= zfs_zget(zfsvfs
, xattr_obj
, &xzp
);
1748 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1749 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1752 mutex_enter(&zp
->z_lock
);
1753 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1754 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1755 mutex_exit(&zp
->z_lock
);
1757 /* charge as an update -- would be nice not to charge at all */
1758 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1761 * Mark this transaction as typically resulting in a net free of space
1763 dmu_tx_mark_netfree(tx
);
1765 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
1767 zfs_dirent_unlock(dl
);
1771 if (error
== ERESTART
) {
1785 * Remove the directory entry.
1787 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1796 * Hold z_lock so that we can make sure that the ACL obj
1797 * hasn't changed. Could have been deleted due to
1800 mutex_enter(&zp
->z_lock
);
1801 mutex_enter(&vp
->v_lock
);
1802 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1803 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1804 delete_now
= may_delete_now
&& !toobig
&&
1805 vp
->v_count
== 1 && !vn_has_cached_data(vp
) &&
1806 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1808 mutex_exit(&vp
->v_lock
);
1812 if (xattr_obj_unlinked
) {
1813 ASSERT3U(xzp
->z_links
, ==, 2);
1814 mutex_enter(&xzp
->z_lock
);
1815 xzp
->z_unlinked
= 1;
1817 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zfsvfs
),
1818 &xzp
->z_links
, sizeof (xzp
->z_links
), tx
);
1819 ASSERT3U(error
, ==, 0);
1820 mutex_exit(&xzp
->z_lock
);
1821 zfs_unlinked_add(xzp
, tx
);
1824 error
= sa_remove(zp
->z_sa_hdl
,
1825 SA_ZPL_XATTR(zfsvfs
), tx
);
1827 error
= sa_update(zp
->z_sa_hdl
,
1828 SA_ZPL_XATTR(zfsvfs
), &null_xattr
,
1829 sizeof (uint64_t), tx
);
1832 mutex_enter(&vp
->v_lock
);
1834 ASSERT0(vp
->v_count
);
1835 mutex_exit(&vp
->v_lock
);
1836 mutex_exit(&zp
->z_lock
);
1837 zfs_znode_delete(zp
, tx
);
1838 } else if (unlinked
) {
1839 mutex_exit(&zp
->z_lock
);
1840 zfs_unlinked_add(zp
, tx
);
1844 if (flags
& FIGNORECASE
)
1846 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1853 zfs_dirent_unlock(dl
);
1860 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1861 zil_commit(zilog
, 0);
1868 * Create a new directory and insert it into dvp using the name
1869 * provided. Return a pointer to the inserted directory.
1871 * IN: dvp - vnode of directory to add subdir to.
1872 * dirname - name of new directory.
1873 * vap - attributes of new directory.
1874 * cr - credentials of caller.
1875 * ct - caller context
1876 * flags - case flags
1877 * vsecp - ACL to be set
1879 * OUT: vpp - vnode of created directory.
1881 * RETURN: 0 on success, error code on failure.
1884 * dvp - ctime|mtime updated
1885 * vp - ctime|mtime|atime updated
1889 zfs_mkdir(vnode_t
*dvp
, char *dirname
, vattr_t
*vap
, vnode_t
**vpp
, cred_t
*cr
,
1890 caller_context_t
*ct
, int flags
, vsecattr_t
*vsecp
)
1892 znode_t
*zp
, *dzp
= VTOZ(dvp
);
1893 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1902 gid_t gid
= crgetgid(cr
);
1903 zfs_acl_ids_t acl_ids
;
1904 boolean_t fuid_dirtied
;
1905 boolean_t waited
= B_FALSE
;
1907 ASSERT(vap
->va_type
== VDIR
);
1910 * If we have an ephemeral id, ACL, or XVATTR then
1911 * make sure file system is at proper version
1914 ksid
= crgetsid(cr
, KSID_OWNER
);
1916 uid
= ksid_getid(ksid
);
1919 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1920 (vsecp
|| (vap
->va_mask
& VATTR_XVATTR
) ||
1921 IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1922 return (SET_ERROR(EINVAL
));
1926 zilog
= zfsvfs
->z_log
;
1928 if (dzp
->z_pflags
& ZFS_XATTR
) {
1930 return (SET_ERROR(EINVAL
));
1933 if (zfsvfs
->z_utf8
&& u8_validate(dirname
,
1934 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1936 return (SET_ERROR(EILSEQ
));
1938 if (flags
& FIGNORECASE
)
1941 if (vap
->va_mask
& VATTR_XVATTR
) {
1942 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1943 crgetuid(cr
), cr
, vap
->va_type
)) != 0) {
1949 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1950 vsecp
, &acl_ids
)) != 0) {
1955 * First make sure the new directory doesn't exist.
1957 * Existence is checked first to make sure we don't return
1958 * EACCES instead of EEXIST which can cause some applications
1964 if (error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1966 zfs_acl_ids_free(&acl_ids
);
1971 if (error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
)) {
1972 zfs_acl_ids_free(&acl_ids
);
1973 zfs_dirent_unlock(dl
);
1978 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1979 zfs_acl_ids_free(&acl_ids
);
1980 zfs_dirent_unlock(dl
);
1982 return (SET_ERROR(EDQUOT
));
1986 * Add a new entry to the directory.
1988 tx
= dmu_tx_create(zfsvfs
->z_os
);
1989 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
1990 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1991 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1993 zfs_fuid_txhold(zfsvfs
, tx
);
1994 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1995 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
1996 acl_ids
.z_aclp
->z_acl_bytes
);
1999 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
2000 ZFS_SA_BASE_ATTR_SIZE
);
2002 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
2004 zfs_dirent_unlock(dl
);
2005 if (error
== ERESTART
) {
2011 zfs_acl_ids_free(&acl_ids
);
2020 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
2023 zfs_fuid_sync(zfsvfs
, tx
);
2026 * Now put new name in parent dir.
2028 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
2032 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
2033 if (flags
& FIGNORECASE
)
2035 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
2036 acl_ids
.z_fuidp
, vap
);
2038 zfs_acl_ids_free(&acl_ids
);
2042 zfs_dirent_unlock(dl
);
2044 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2045 zil_commit(zilog
, 0);
2052 * Remove a directory subdir entry. If the current working
2053 * directory is the same as the subdir to be removed, the
2056 * IN: dvp - vnode of directory to remove from.
2057 * name - name of directory to be removed.
2058 * cwd - vnode of current working directory.
2059 * cr - credentials of caller.
2060 * ct - caller context
2061 * flags - case flags
2063 * RETURN: 0 on success, error code on failure.
2066 * dvp - ctime|mtime updated
2070 zfs_rmdir(vnode_t
*dvp
, char *name
, vnode_t
*cwd
, cred_t
*cr
,
2071 caller_context_t
*ct
, int flags
)
2073 znode_t
*dzp
= VTOZ(dvp
);
2076 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
2082 boolean_t waited
= B_FALSE
;
2086 zilog
= zfsvfs
->z_log
;
2088 if (flags
& FIGNORECASE
)
2094 * Attempt to lock directory; fail if entry doesn't exist.
2096 if (error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
2104 if (error
= zfs_zaccess_delete(dzp
, zp
, cr
)) {
2108 if (vp
->v_type
!= VDIR
) {
2109 error
= SET_ERROR(ENOTDIR
);
2114 error
= SET_ERROR(EINVAL
);
2118 vnevent_rmdir(vp
, dvp
, name
, ct
);
2121 * Grab a lock on the directory to make sure that noone is
2122 * trying to add (or lookup) entries while we are removing it.
2124 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
2127 * Grab a lock on the parent pointer to make sure we play well
2128 * with the treewalk and directory rename code.
2130 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
2132 tx
= dmu_tx_create(zfsvfs
->z_os
);
2133 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2134 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2135 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
2136 zfs_sa_upgrade_txholds(tx
, zp
);
2137 zfs_sa_upgrade_txholds(tx
, dzp
);
2138 dmu_tx_mark_netfree(tx
);
2139 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
2141 rw_exit(&zp
->z_parent_lock
);
2142 rw_exit(&zp
->z_name_lock
);
2143 zfs_dirent_unlock(dl
);
2145 if (error
== ERESTART
) {
2156 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2159 uint64_t txtype
= TX_RMDIR
;
2160 if (flags
& FIGNORECASE
)
2162 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2167 rw_exit(&zp
->z_parent_lock
);
2168 rw_exit(&zp
->z_name_lock
);
2170 zfs_dirent_unlock(dl
);
2174 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2175 zil_commit(zilog
, 0);
2182 * Read as many directory entries as will fit into the provided
2183 * buffer from the given directory cursor position (specified in
2184 * the uio structure).
2186 * IN: vp - vnode of directory to read.
2187 * uio - structure supplying read location, range info,
2188 * and return buffer.
2189 * cr - credentials of caller.
2190 * ct - caller context
2191 * flags - case flags
2193 * OUT: uio - updated offset and range, buffer filled.
2194 * eofp - set to true if end-of-file detected.
2196 * RETURN: 0 on success, error code on failure.
2199 * vp - atime updated
2201 * Note that the low 4 bits of the cookie returned by zap is always zero.
2202 * This allows us to use the low range for "special" directory entries:
2203 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2204 * we use the offset 2 for the '.zfs' directory.
2208 zfs_readdir(vnode_t
*vp
, uio_t
*uio
, cred_t
*cr
, int *eofp
,
2209 caller_context_t
*ct
, int flags
)
2211 znode_t
*zp
= VTOZ(vp
);
2215 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2220 zap_attribute_t zap
;
2221 uint_t bytes_wanted
;
2222 uint64_t offset
; /* must be unsigned; checks for < 1 */
2228 boolean_t check_sysattrs
;
2233 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
2234 &parent
, sizeof (parent
))) != 0) {
2240 * If we are not given an eof variable,
2247 * Check for valid iov_len.
2249 if (uio
->uio_iov
->iov_len
<= 0) {
2251 return (SET_ERROR(EINVAL
));
2255 * Quit if directory has been removed (posix)
2257 if ((*eofp
= zp
->z_unlinked
) != 0) {
2264 offset
= uio
->uio_loffset
;
2265 prefetch
= zp
->z_zn_prefetch
;
2268 * Initialize the iterator cursor.
2272 * Start iteration from the beginning of the directory.
2274 zap_cursor_init(&zc
, os
, zp
->z_id
);
2277 * The offset is a serialized cursor.
2279 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2283 * Get space to change directory entries into fs independent format.
2285 iovp
= uio
->uio_iov
;
2286 bytes_wanted
= iovp
->iov_len
;
2287 if (uio
->uio_segflg
!= UIO_SYSSPACE
|| uio
->uio_iovcnt
!= 1) {
2288 bufsize
= bytes_wanted
;
2289 outbuf
= kmem_alloc(bufsize
, KM_SLEEP
);
2290 odp
= (struct dirent64
*)outbuf
;
2292 bufsize
= bytes_wanted
;
2294 odp
= (struct dirent64
*)iovp
->iov_base
;
2296 eodp
= (struct edirent
*)odp
;
2299 * If this VFS supports the system attribute view interface; and
2300 * we're looking at an extended attribute directory; and we care
2301 * about normalization conflicts on this vfs; then we must check
2302 * for normalization conflicts with the sysattr name space.
2304 check_sysattrs
= vfs_has_feature(vp
->v_vfsp
, VFSFT_SYSATTR_VIEWS
) &&
2305 (vp
->v_flag
& V_XATTRDIR
) && zfsvfs
->z_norm
&&
2306 (flags
& V_RDDIR_ENTFLAGS
);
2309 * Transform to file-system independent format
2312 while (outcount
< bytes_wanted
) {
2315 off64_t
*next
= NULL
;
2318 * Special case `.', `..', and `.zfs'.
2321 (void) strcpy(zap
.za_name
, ".");
2322 zap
.za_normalization_conflict
= 0;
2324 } else if (offset
== 1) {
2325 (void) strcpy(zap
.za_name
, "..");
2326 zap
.za_normalization_conflict
= 0;
2328 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2329 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2330 zap
.za_normalization_conflict
= 0;
2331 objnum
= ZFSCTL_INO_ROOT
;
2336 if (error
= zap_cursor_retrieve(&zc
, &zap
)) {
2337 if ((*eofp
= (error
== ENOENT
)) != 0)
2343 if (zap
.za_integer_length
!= 8 ||
2344 zap
.za_num_integers
!= 1) {
2345 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2346 "entry, obj = %lld, offset = %lld\n",
2347 (u_longlong_t
)zp
->z_id
,
2348 (u_longlong_t
)offset
);
2349 error
= SET_ERROR(ENXIO
);
2353 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2355 * MacOS X can extract the object type here such as:
2356 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2359 if (check_sysattrs
&& !zap
.za_normalization_conflict
) {
2360 zap
.za_normalization_conflict
=
2361 xattr_sysattr_casechk(zap
.za_name
);
2365 if (flags
& V_RDDIR_ACCFILTER
) {
2367 * If we have no access at all, don't include
2368 * this entry in the returned information
2371 if (zfs_zget(zp
->z_zfsvfs
, objnum
, &ezp
) != 0)
2373 if (!zfs_has_access(ezp
, cr
)) {
2380 if (flags
& V_RDDIR_ENTFLAGS
)
2381 reclen
= EDIRENT_RECLEN(strlen(zap
.za_name
));
2383 reclen
= DIRENT64_RECLEN(strlen(zap
.za_name
));
2386 * Will this entry fit in the buffer?
2388 if (outcount
+ reclen
> bufsize
) {
2390 * Did we manage to fit anything in the buffer?
2393 error
= SET_ERROR(EINVAL
);
2398 if (flags
& V_RDDIR_ENTFLAGS
) {
2400 * Add extended flag entry:
2402 eodp
->ed_ino
= objnum
;
2403 eodp
->ed_reclen
= reclen
;
2404 /* NOTE: ed_off is the offset for the *next* entry */
2405 next
= &(eodp
->ed_off
);
2406 eodp
->ed_eflags
= zap
.za_normalization_conflict
?
2407 ED_CASE_CONFLICT
: 0;
2408 (void) strncpy(eodp
->ed_name
, zap
.za_name
,
2409 EDIRENT_NAMELEN(reclen
));
2410 eodp
= (edirent_t
*)((intptr_t)eodp
+ reclen
);
2415 odp
->d_ino
= objnum
;
2416 odp
->d_reclen
= reclen
;
2417 /* NOTE: d_off is the offset for the *next* entry */
2418 next
= &(odp
->d_off
);
2419 (void) strncpy(odp
->d_name
, zap
.za_name
,
2420 DIRENT64_NAMELEN(reclen
));
2421 odp
= (dirent64_t
*)((intptr_t)odp
+ reclen
);
2425 ASSERT(outcount
<= bufsize
);
2427 /* Prefetch znode */
2429 dmu_prefetch(os
, objnum
, 0, 0, 0,
2430 ZIO_PRIORITY_SYNC_READ
);
2434 * Move to the next entry, fill in the previous offset.
2436 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2437 zap_cursor_advance(&zc
);
2438 offset
= zap_cursor_serialize(&zc
);
2445 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2447 if (uio
->uio_segflg
== UIO_SYSSPACE
&& uio
->uio_iovcnt
== 1) {
2448 iovp
->iov_base
+= outcount
;
2449 iovp
->iov_len
-= outcount
;
2450 uio
->uio_resid
-= outcount
;
2451 } else if (error
= uiomove(outbuf
, (long)outcount
, UIO_READ
, uio
)) {
2453 * Reset the pointer.
2455 offset
= uio
->uio_loffset
;
2459 zap_cursor_fini(&zc
);
2460 if (uio
->uio_segflg
!= UIO_SYSSPACE
|| uio
->uio_iovcnt
!= 1)
2461 kmem_free(outbuf
, bufsize
);
2463 if (error
== ENOENT
)
2466 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
2468 uio
->uio_loffset
= offset
;
2473 ulong_t zfs_fsync_sync_cnt
= 4;
2476 zfs_fsync(vnode_t
*vp
, int syncflag
, cred_t
*cr
, caller_context_t
*ct
)
2478 znode_t
*zp
= VTOZ(vp
);
2479 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2482 * Regardless of whether this is required for standards conformance,
2483 * this is the logical behavior when fsync() is called on a file with
2484 * dirty pages. We use B_ASYNC since the ZIL transactions are already
2485 * going to be pushed out as part of the zil_commit().
2487 if (vn_has_cached_data(vp
) && !(syncflag
& FNODSYNC
) &&
2488 (vp
->v_type
== VREG
) && !(IS_SWAPVP(vp
)))
2489 (void) fop_putpage(vp
, 0, (size_t)0, B_ASYNC
, cr
, ct
);
2491 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2493 if (zfsvfs
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2496 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
2504 * Get the requested file attributes and place them in the provided
2507 * IN: vp - vnode of file.
2508 * vap - va_mask identifies requested attributes.
2509 * If VATTR_XVATTR set, then optional attrs are requested
2510 * flags - ATTR_NOACLCHECK (CIFS server context)
2511 * cr - credentials of caller.
2512 * ct - caller context
2514 * OUT: vap - attribute values.
2516 * RETURN: 0 (always succeeds).
2520 zfs_getattr(vnode_t
*vp
, vattr_t
*vap
, int flags
, cred_t
*cr
,
2521 caller_context_t
*ct
)
2523 znode_t
*zp
= VTOZ(vp
);
2524 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2527 uint64_t mtime
[2], ctime
[2];
2528 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2529 xoptattr_t
*xoap
= NULL
;
2530 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2531 sa_bulk_attr_t bulk
[2];
2537 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2539 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
2540 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
2542 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2548 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2549 * Also, if we are the owner don't bother, since owner should
2550 * always be allowed to read basic attributes of file.
2552 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2553 (vap
->va_uid
!= crgetuid(cr
))) {
2554 if (error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2562 * Return all attributes. It's cheaper to provide the answer
2563 * than to determine whether we were asked the question.
2566 mutex_enter(&zp
->z_lock
);
2567 vap
->va_type
= vp
->v_type
;
2568 vap
->va_mode
= zp
->z_mode
& MODEMASK
;
2569 vap
->va_fsid
= zp
->z_zfsvfs
->z_vfs
->vfs_dev
;
2570 vap
->va_nodeid
= zp
->z_id
;
2571 if ((vp
->v_flag
& VROOT
) && zfs_show_ctldir(zp
))
2572 links
= zp
->z_links
+ 1;
2574 links
= zp
->z_links
;
2575 vap
->va_nlink
= MIN(links
, UINT32_MAX
); /* nlink_t limit! */
2576 vap
->va_size
= zp
->z_size
;
2577 vap
->va_rdev
= vp
->v_rdev
;
2578 vap
->va_seq
= zp
->z_seq
;
2581 * Add in any requested optional attributes and the create time.
2582 * Also set the corresponding bits in the returned attribute bitmap.
2584 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zfsvfs
->z_use_fuids
) {
2585 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2587 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2588 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2591 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2592 xoap
->xoa_readonly
=
2593 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2594 XVA_SET_RTN(xvap
, XAT_READONLY
);
2597 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2599 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2600 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2603 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2605 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2606 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2609 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2610 xoap
->xoa_nounlink
=
2611 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2612 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2615 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2616 xoap
->xoa_immutable
=
2617 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2618 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2621 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2622 xoap
->xoa_appendonly
=
2623 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2624 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2627 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2629 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2630 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2633 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2635 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2636 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2639 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2640 xoap
->xoa_av_quarantined
=
2641 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2642 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2645 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2646 xoap
->xoa_av_modified
=
2647 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2648 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2651 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2652 vp
->v_type
== VREG
) {
2653 zfs_sa_get_scanstamp(zp
, xvap
);
2656 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2659 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zfsvfs
),
2660 times
, sizeof (times
));
2661 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2662 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2665 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2666 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2667 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2669 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2670 xoap
->xoa_generation
= zp
->z_gen
;
2671 XVA_SET_RTN(xvap
, XAT_GEN
);
2674 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2676 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2677 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2680 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2682 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2683 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2687 ZFS_TIME_DECODE(&vap
->va_atime
, zp
->z_atime
);
2688 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2689 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2691 mutex_exit(&zp
->z_lock
);
2693 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2695 if (zp
->z_blksz
== 0) {
2697 * Block size hasn't been set; suggest maximal I/O transfers.
2699 vap
->va_blksize
= zfsvfs
->z_max_blksz
;
2707 * Set the file attributes to the values contained in the
2710 * IN: vp - vnode of file to be modified.
2711 * vap - new attribute values.
2712 * If VATTR_XVATTR set, then optional attrs are being set
2713 * flags - ATTR_UTIME set if non-default time values provided.
2714 * - ATTR_NOACLCHECK (CIFS context only).
2715 * cr - credentials of caller.
2716 * ct - caller context
2718 * RETURN: 0 on success, error code on failure.
2721 * vp - ctime updated, mtime updated if size changed.
2725 zfs_setattr(vnode_t
*vp
, vattr_t
*vap
, int flags
, cred_t
*cr
,
2726 caller_context_t
*ct
)
2728 znode_t
*zp
= VTOZ(vp
);
2729 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2734 uint_t mask
= vap
->va_mask
;
2735 uint_t saved_mask
= 0;
2738 uint64_t new_uid
, new_gid
;
2740 uint64_t mtime
[2], ctime
[2];
2742 int need_policy
= FALSE
;
2744 zfs_fuid_info_t
*fuidp
= NULL
;
2745 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2748 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2749 boolean_t fuid_dirtied
= B_FALSE
;
2750 sa_bulk_attr_t bulk
[7], xattr_bulk
[7];
2751 int count
= 0, xattr_count
= 0;
2756 if (mask
& VATTR_NOSET
)
2757 return (SET_ERROR(EINVAL
));
2762 zilog
= zfsvfs
->z_log
;
2765 * Make sure that if we have ephemeral uid/gid or xvattr specified
2766 * that file system is at proper version level
2769 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
2770 (((mask
& VATTR_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2771 ((mask
& VATTR_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2772 (mask
& VATTR_XVATTR
))) {
2774 return (SET_ERROR(EINVAL
));
2777 if (mask
& VATTR_SIZE
&& vp
->v_type
== VDIR
) {
2779 return (SET_ERROR(EISDIR
));
2782 if (mask
& VATTR_SIZE
&& vp
->v_type
!= VREG
&& vp
->v_type
!= VFIFO
) {
2784 return (SET_ERROR(EINVAL
));
2788 * If this is an xvattr_t, then get a pointer to the structure of
2789 * optional attributes. If this is NULL, then we have a vattr_t.
2791 xoap
= xva_getxoptattr(xvap
);
2793 xva_init(&tmpxvattr
);
2796 * Immutable files can only alter immutable bit and atime
2798 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2799 ((mask
& (VATTR_SIZE
|VATTR_UID
|VATTR_GID
|VATTR_MTIME
|VATTR_MODE
)) ||
2800 ((mask
& VATTR_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2802 return (SET_ERROR(EPERM
));
2806 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
2810 * Verify timestamps doesn't overflow 32 bits.
2811 * ZFS can handle large timestamps, but 32bit syscalls can't
2812 * handle times greater than 2039. This check should be removed
2813 * once large timestamps are fully supported.
2815 if (mask
& (VATTR_ATIME
| VATTR_MTIME
)) {
2816 if (((mask
& VATTR_ATIME
) && TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2817 ((mask
& VATTR_MTIME
) && TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2819 return (SET_ERROR(EOVERFLOW
));
2827 /* Can this be moved to before the top label? */
2828 if (zfsvfs
->z_vfs
->vfs_flag
& VFS_RDONLY
) {
2830 return (SET_ERROR(EROFS
));
2834 * First validate permissions
2837 if (mask
& VATTR_SIZE
) {
2838 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2844 * XXX - Note, we are not providing any open
2845 * mode flags here (like FNDELAY), so we may
2846 * block if there are locks present... this
2847 * should be addressed in openat().
2849 /* XXX - would it be OK to generate a log record here? */
2850 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2856 if (vap
->va_size
== 0)
2857 vnevent_truncate(ZTOV(zp
), ct
);
2860 if (mask
& (VATTR_ATIME
|VATTR_MTIME
) ||
2861 ((mask
& VATTR_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2862 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2863 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2864 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2865 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2866 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2867 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2868 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2872 if (mask
& (VATTR_UID
|VATTR_GID
)) {
2873 int idmask
= (mask
& (VATTR_UID
|VATTR_GID
));
2878 * NOTE: even if a new mode is being set,
2879 * we may clear S_ISUID/S_ISGID bits.
2882 if (!(mask
& VATTR_MODE
))
2883 vap
->va_mode
= zp
->z_mode
;
2886 * Take ownership or chgrp to group we are a member of
2889 take_owner
= (mask
& VATTR_UID
) && (vap
->va_uid
== crgetuid(cr
));
2890 take_group
= (mask
& VATTR_GID
) &&
2891 zfs_groupmember(zfsvfs
, vap
->va_gid
, cr
);
2894 * If both VATTR_UID and VATTR_GID are set then take_owner and
2895 * take_group must both be set in order to allow taking
2898 * Otherwise, send the check through secpolicy_vnode_setattr()
2902 if (((idmask
== (VATTR_UID
|VATTR_GID
)) && take_owner
&& take_group
) ||
2903 ((idmask
== VATTR_UID
) && take_owner
) ||
2904 ((idmask
== VATTR_GID
) && take_group
)) {
2905 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2906 skipaclchk
, cr
) == 0) {
2908 * Remove setuid/setgid for non-privileged users
2910 secpolicy_setid_clear(vap
, cr
);
2911 trim_mask
= (mask
& (VATTR_UID
|VATTR_GID
));
2920 mutex_enter(&zp
->z_lock
);
2921 oldva
.va_mode
= zp
->z_mode
;
2922 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2923 if (mask
& VATTR_XVATTR
) {
2925 * Update xvattr mask to include only those attributes
2926 * that are actually changing.
2928 * the bits will be restored prior to actually setting
2929 * the attributes so the caller thinks they were set.
2931 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2932 if (xoap
->xoa_appendonly
!=
2933 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2936 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2937 XVA_SET_REQ(&tmpxvattr
, XAT_APPENDONLY
);
2941 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2942 if (xoap
->xoa_nounlink
!=
2943 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2946 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2947 XVA_SET_REQ(&tmpxvattr
, XAT_NOUNLINK
);
2951 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2952 if (xoap
->xoa_immutable
!=
2953 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2956 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2957 XVA_SET_REQ(&tmpxvattr
, XAT_IMMUTABLE
);
2961 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2962 if (xoap
->xoa_nodump
!=
2963 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2966 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2967 XVA_SET_REQ(&tmpxvattr
, XAT_NODUMP
);
2971 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2972 if (xoap
->xoa_av_modified
!=
2973 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2976 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2977 XVA_SET_REQ(&tmpxvattr
, XAT_AV_MODIFIED
);
2981 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2982 if ((vp
->v_type
!= VREG
&&
2983 xoap
->xoa_av_quarantined
) ||
2984 xoap
->xoa_av_quarantined
!=
2985 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2988 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2989 XVA_SET_REQ(&tmpxvattr
, XAT_AV_QUARANTINED
);
2993 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2994 mutex_exit(&zp
->z_lock
);
2996 return (SET_ERROR(EPERM
));
2999 if (need_policy
== FALSE
&&
3000 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
3001 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
3006 mutex_exit(&zp
->z_lock
);
3008 if (mask
& VATTR_MODE
) {
3009 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
3010 err
= secpolicy_setid_setsticky_clear(vp
, vap
,
3016 trim_mask
|= VATTR_MODE
;
3024 * If trim_mask is set then take ownership
3025 * has been granted or write_acl is present and user
3026 * has the ability to modify mode. In that case remove
3027 * UID|GID and or MODE from mask so that
3028 * secpolicy_vnode_setattr() doesn't revoke it.
3032 saved_mask
= vap
->va_mask
;
3033 vap
->va_mask
&= ~trim_mask
;
3035 err
= secpolicy_vnode_setattr(cr
, vp
, vap
, &oldva
, flags
,
3036 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
3043 vap
->va_mask
|= saved_mask
;
3047 * secpolicy_vnode_setattr, or take ownership may have
3050 mask
= vap
->va_mask
;
3052 if ((mask
& (VATTR_UID
| VATTR_GID
))) {
3053 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
3054 &xattr_obj
, sizeof (xattr_obj
));
3056 if (err
== 0 && xattr_obj
) {
3057 err
= zfs_zget(zp
->z_zfsvfs
, xattr_obj
, &attrzp
);
3061 if (mask
& VATTR_UID
) {
3062 new_uid
= zfs_fuid_create(zfsvfs
,
3063 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
3064 if (new_uid
!= zp
->z_uid
&&
3065 zfs_fuid_overquota(zfsvfs
, B_FALSE
, new_uid
)) {
3067 VN_RELE(ZTOV(attrzp
));
3068 err
= SET_ERROR(EDQUOT
);
3073 if (mask
& VATTR_GID
) {
3074 new_gid
= zfs_fuid_create(zfsvfs
, (uint64_t)vap
->va_gid
,
3075 cr
, ZFS_GROUP
, &fuidp
);
3076 if (new_gid
!= zp
->z_gid
&&
3077 zfs_fuid_overquota(zfsvfs
, B_TRUE
, new_gid
)) {
3079 VN_RELE(ZTOV(attrzp
));
3080 err
= SET_ERROR(EDQUOT
);
3085 tx
= dmu_tx_create(zfsvfs
->z_os
);
3087 if (mask
& VATTR_MODE
) {
3088 uint64_t pmode
= zp
->z_mode
;
3090 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
3092 if (zp
->z_zfsvfs
->z_acl_mode
== ZFS_ACL_RESTRICTED
&&
3093 !(zp
->z_pflags
& ZFS_ACL_TRIVIAL
)) {
3094 err
= SET_ERROR(EPERM
);
3098 if (err
= zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
))
3101 mutex_enter(&zp
->z_lock
);
3102 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
3104 * Are we upgrading ACL from old V0 format
3107 if (zfsvfs
->z_version
>= ZPL_VERSION_FUID
&&
3108 zfs_znode_acl_version(zp
) ==
3109 ZFS_ACL_VERSION_INITIAL
) {
3110 dmu_tx_hold_free(tx
, acl_obj
, 0,
3112 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3113 0, aclp
->z_acl_bytes
);
3115 dmu_tx_hold_write(tx
, acl_obj
, 0,
3118 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3119 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3120 0, aclp
->z_acl_bytes
);
3122 mutex_exit(&zp
->z_lock
);
3123 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3125 if ((mask
& VATTR_XVATTR
) &&
3126 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3127 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3129 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3133 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
3136 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3138 zfs_fuid_txhold(zfsvfs
, tx
);
3140 zfs_sa_upgrade_txholds(tx
, zp
);
3142 err
= dmu_tx_assign(tx
, TXG_WAIT
);
3148 * Set each attribute requested.
3149 * We group settings according to the locks they need to acquire.
3151 * Note: you cannot set ctime directly, although it will be
3152 * updated as a side-effect of calling this function.
3156 if (mask
& (VATTR_UID
|VATTR_GID
|VATTR_MODE
))
3157 mutex_enter(&zp
->z_acl_lock
);
3158 mutex_enter(&zp
->z_lock
);
3160 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
3161 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
3164 if (mask
& (VATTR_UID
|VATTR_GID
|VATTR_MODE
))
3165 mutex_enter(&attrzp
->z_acl_lock
);
3166 mutex_enter(&attrzp
->z_lock
);
3167 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3168 SA_ZPL_FLAGS(zfsvfs
), NULL
, &attrzp
->z_pflags
,
3169 sizeof (attrzp
->z_pflags
));
3172 if (mask
& (VATTR_UID
|VATTR_GID
)) {
3174 if (mask
& VATTR_UID
) {
3175 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zfsvfs
), NULL
,
3176 &new_uid
, sizeof (new_uid
));
3177 zp
->z_uid
= new_uid
;
3179 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3180 SA_ZPL_UID(zfsvfs
), NULL
, &new_uid
,
3182 attrzp
->z_uid
= new_uid
;
3186 if (mask
& VATTR_GID
) {
3187 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zfsvfs
),
3188 NULL
, &new_gid
, sizeof (new_gid
));
3189 zp
->z_gid
= new_gid
;
3191 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3192 SA_ZPL_GID(zfsvfs
), NULL
, &new_gid
,
3194 attrzp
->z_gid
= new_gid
;
3197 if (!(mask
& VATTR_MODE
)) {
3198 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
),
3199 NULL
, &new_mode
, sizeof (new_mode
));
3200 new_mode
= zp
->z_mode
;
3202 err
= zfs_acl_chown_setattr(zp
);
3205 err
= zfs_acl_chown_setattr(attrzp
);
3210 if (mask
& VATTR_MODE
) {
3211 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
), NULL
,
3212 &new_mode
, sizeof (new_mode
));
3213 zp
->z_mode
= new_mode
;
3214 ASSERT3U((uintptr_t)aclp
, !=, (uintptr_t)NULL
);
3215 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
3217 if (zp
->z_acl_cached
)
3218 zfs_acl_free(zp
->z_acl_cached
);
3219 zp
->z_acl_cached
= aclp
;
3224 if (mask
& VATTR_ATIME
) {
3225 ZFS_TIME_ENCODE(&vap
->va_atime
, zp
->z_atime
);
3226 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
,
3227 &zp
->z_atime
, sizeof (zp
->z_atime
));
3230 if (mask
& VATTR_MTIME
) {
3231 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3232 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
3233 mtime
, sizeof (mtime
));
3236 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3237 if (mask
& VATTR_SIZE
&& !(mask
& VATTR_MTIME
)) {
3238 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
),
3239 NULL
, mtime
, sizeof (mtime
));
3240 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3241 &ctime
, sizeof (ctime
));
3242 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
3244 } else if (mask
!= 0) {
3245 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3246 &ctime
, sizeof (ctime
));
3247 zfs_tstamp_update_setup(zp
, STATE_CHANGED
, mtime
, ctime
,
3250 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3251 SA_ZPL_CTIME(zfsvfs
), NULL
,
3252 &ctime
, sizeof (ctime
));
3253 zfs_tstamp_update_setup(attrzp
, STATE_CHANGED
,
3254 mtime
, ctime
, B_TRUE
);
3258 * Do this after setting timestamps to prevent timestamp
3259 * update from toggling bit
3262 if (xoap
&& (mask
& VATTR_XVATTR
)) {
3265 * restore trimmed off masks
3266 * so that return masks can be set for caller.
3269 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_APPENDONLY
)) {
3270 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3272 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_NOUNLINK
)) {
3273 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3275 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_IMMUTABLE
)) {
3276 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3278 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_NODUMP
)) {
3279 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3281 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_AV_MODIFIED
)) {
3282 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3284 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_AV_QUARANTINED
)) {
3285 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3288 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3289 ASSERT(vp
->v_type
== VREG
);
3291 zfs_xvattr_set(zp
, xvap
, tx
);
3295 zfs_fuid_sync(zfsvfs
, tx
);
3298 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3300 mutex_exit(&zp
->z_lock
);
3301 if (mask
& (VATTR_UID
|VATTR_GID
|VATTR_MODE
))
3302 mutex_exit(&zp
->z_acl_lock
);
3305 if (mask
& (VATTR_UID
|VATTR_GID
|VATTR_MODE
))
3306 mutex_exit(&attrzp
->z_acl_lock
);
3307 mutex_exit(&attrzp
->z_lock
);
3310 if (err
== 0 && attrzp
) {
3311 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3317 VN_RELE(ZTOV(attrzp
));
3323 zfs_fuid_info_free(fuidp
);
3329 if (err
== ERESTART
)
3332 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3337 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3338 zil_commit(zilog
, 0);
3344 typedef struct zfs_zlock
{
3345 krwlock_t
*zl_rwlock
; /* lock we acquired */
3346 znode_t
*zl_znode
; /* znode we held */
3347 struct zfs_zlock
*zl_next
; /* next in list */
3351 * Drop locks and release vnodes that were held by zfs_rename_lock().
3354 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3358 while ((zl
= *zlpp
) != NULL
) {
3359 if (zl
->zl_znode
!= NULL
)
3360 VN_RELE(ZTOV(zl
->zl_znode
));
3361 rw_exit(zl
->zl_rwlock
);
3362 *zlpp
= zl
->zl_next
;
3363 kmem_free(zl
, sizeof (*zl
));
3368 * Search back through the directory tree, using the ".." entries.
3369 * Lock each directory in the chain to prevent concurrent renames.
3370 * Fail any attempt to move a directory into one of its own descendants.
3371 * XXX - z_parent_lock can overlap with map or grow locks
3374 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3378 uint64_t rootid
= zp
->z_zfsvfs
->z_root
;
3379 uint64_t oidp
= zp
->z_id
;
3380 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3381 krw_t rw
= RW_WRITER
;
3384 * First pass write-locks szp and compares to zp->z_id.
3385 * Later passes read-lock zp and compare to zp->z_parent.
3388 if (!rw_tryenter(rwlp
, rw
)) {
3390 * Another thread is renaming in this path.
3391 * Note that if we are a WRITER, we don't have any
3392 * parent_locks held yet.
3394 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3396 * Drop our locks and restart
3398 zfs_rename_unlock(&zl
);
3402 rwlp
= &szp
->z_parent_lock
;
3407 * Wait for other thread to drop its locks
3413 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3414 zl
->zl_rwlock
= rwlp
;
3415 zl
->zl_znode
= NULL
;
3416 zl
->zl_next
= *zlpp
;
3419 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3420 return (SET_ERROR(EINVAL
));
3422 if (oidp
== rootid
) /* We've hit the top */
3425 if (rw
== RW_READER
) { /* i.e. not the first pass */
3426 int error
= zfs_zget(zp
->z_zfsvfs
, oidp
, &zp
);
3431 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zp
->z_zfsvfs
),
3432 &oidp
, sizeof (oidp
));
3433 rwlp
= &zp
->z_parent_lock
;
3436 } while (zp
->z_id
!= sdzp
->z_id
);
3442 * Move an entry from the provided source directory to the target
3443 * directory. Change the entry name as indicated.
3445 * IN: sdvp - Source directory containing the "old entry".
3446 * snm - Old entry name.
3447 * tdvp - Target directory to contain the "new entry".
3448 * tnm - New entry name.
3449 * cr - credentials of caller.
3450 * ct - caller context
3451 * flags - case flags
3453 * RETURN: 0 on success, error code on failure.
3456 * sdvp,tdvp - ctime|mtime updated
3460 zfs_rename(vnode_t
*sdvp
, char *snm
, vnode_t
*tdvp
, char *tnm
, cred_t
*cr
,
3461 caller_context_t
*ct
, int flags
)
3463 znode_t
*tdzp
, *szp
, *tzp
;
3464 znode_t
*sdzp
= VTOZ(sdvp
);
3465 zfsvfs_t
*zfsvfs
= sdzp
->z_zfsvfs
;
3468 zfs_dirlock_t
*sdl
, *tdl
;
3471 int cmp
, serr
, terr
;
3472 int error
= 0, rm_err
= 0;
3474 boolean_t waited
= B_FALSE
;
3477 ZFS_VERIFY_ZP(sdzp
);
3478 zilog
= zfsvfs
->z_log
;
3481 * Make sure we have the real vp for the target directory.
3483 if (fop_realvp(tdvp
, &realvp
, ct
) == 0)
3487 ZFS_VERIFY_ZP(tdzp
);
3490 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3491 * ctldir appear to have the same v_vfsp.
3493 if (tdzp
->z_zfsvfs
!= zfsvfs
|| zfsctl_is_node(tdvp
)) {
3495 return (SET_ERROR(EXDEV
));
3498 if (zfsvfs
->z_utf8
&& u8_validate(tnm
,
3499 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3501 return (SET_ERROR(EILSEQ
));
3504 if (flags
& FIGNORECASE
)
3513 * This is to prevent the creation of links into attribute space
3514 * by renaming a linked file into/outof an attribute directory.
3515 * See the comment in zfs_link() for why this is considered bad.
3517 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3519 return (SET_ERROR(EINVAL
));
3523 * Lock source and target directory entries. To prevent deadlock,
3524 * a lock ordering must be defined. We lock the directory with
3525 * the smallest object id first, or if it's a tie, the one with
3526 * the lexically first name.
3528 if (sdzp
->z_id
< tdzp
->z_id
) {
3530 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3534 * First compare the two name arguments without
3535 * considering any case folding.
3537 int nofold
= (zfsvfs
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3539 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3540 ASSERT(error
== 0 || !zfsvfs
->z_utf8
);
3543 * POSIX: "If the old argument and the new argument
3544 * both refer to links to the same existing file,
3545 * the rename() function shall return successfully
3546 * and perform no other action."
3552 * If the file system is case-folding, then we may
3553 * have some more checking to do. A case-folding file
3554 * system is either supporting mixed case sensitivity
3555 * access or is completely case-insensitive. Note
3556 * that the file system is always case preserving.
3558 * In mixed sensitivity mode case sensitive behavior
3559 * is the default. FIGNORECASE must be used to
3560 * explicitly request case insensitive behavior.
3562 * If the source and target names provided differ only
3563 * by case (e.g., a request to rename 'tim' to 'Tim'),
3564 * we will treat this as a special case in the
3565 * case-insensitive mode: as long as the source name
3566 * is an exact match, we will allow this to proceed as
3567 * a name-change request.
3569 if ((zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
||
3570 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&&
3571 flags
& FIGNORECASE
)) &&
3572 u8_strcmp(snm
, tnm
, 0, zfsvfs
->z_norm
, U8_UNICODE_LATEST
,
3575 * case preserving rename request, require exact
3584 * If the source and destination directories are the same, we should
3585 * grab the z_name_lock of that directory only once.
3589 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3593 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3594 ZEXISTS
| zflg
, NULL
, NULL
);
3595 terr
= zfs_dirent_lock(&tdl
,
3596 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3598 terr
= zfs_dirent_lock(&tdl
,
3599 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3600 serr
= zfs_dirent_lock(&sdl
,
3601 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3607 * Source entry invalid or not there.
3610 zfs_dirent_unlock(tdl
);
3616 rw_exit(&sdzp
->z_name_lock
);
3618 if (strcmp(snm
, "..") == 0)
3619 serr
= SET_ERROR(EINVAL
);
3624 zfs_dirent_unlock(sdl
);
3628 rw_exit(&sdzp
->z_name_lock
);
3630 if (strcmp(tnm
, "..") == 0)
3631 terr
= SET_ERROR(EINVAL
);
3637 * Must have write access at the source to remove the old entry
3638 * and write access at the target to create the new entry.
3639 * Note that if target and source are the same, this can be
3640 * done in a single check.
3643 if (error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
))
3646 if (ZTOV(szp
)->v_type
== VDIR
) {
3648 * Check to make sure rename is valid.
3649 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3651 if (error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
))
3656 * Does target exist?
3660 * Source and target must be the same type.
3662 if (ZTOV(szp
)->v_type
== VDIR
) {
3663 if (ZTOV(tzp
)->v_type
!= VDIR
) {
3664 error
= SET_ERROR(ENOTDIR
);
3668 if (ZTOV(tzp
)->v_type
== VDIR
) {
3669 error
= SET_ERROR(EISDIR
);
3674 * POSIX dictates that when the source and target
3675 * entries refer to the same file object, rename
3676 * must do nothing and exit without error.
3678 if (szp
->z_id
== tzp
->z_id
) {
3684 vnevent_pre_rename_src(ZTOV(szp
), sdvp
, snm
, ct
);
3686 vnevent_pre_rename_dest(ZTOV(tzp
), tdvp
, tnm
, ct
);
3689 * notify the target directory if it is not the same
3690 * as source directory.
3693 vnevent_pre_rename_dest_dir(tdvp
, ZTOV(szp
), tnm
, ct
);
3696 tx
= dmu_tx_create(zfsvfs
->z_os
);
3697 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3698 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3699 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3700 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3702 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3703 zfs_sa_upgrade_txholds(tx
, tdzp
);
3706 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3707 zfs_sa_upgrade_txholds(tx
, tzp
);
3710 zfs_sa_upgrade_txholds(tx
, szp
);
3711 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
3712 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
3715 zfs_rename_unlock(&zl
);
3716 zfs_dirent_unlock(sdl
);
3717 zfs_dirent_unlock(tdl
);
3720 rw_exit(&sdzp
->z_name_lock
);
3725 if (error
== ERESTART
) {
3736 if (tzp
) /* Attempt to remove the existing target */
3737 error
= rm_err
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3740 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3742 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3744 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zfsvfs
),
3745 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3748 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3750 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3751 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3752 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3755 * Update path information for the target vnode
3757 vn_renamepath(tdvp
, ZTOV(szp
), tnm
,
3761 * At this point, we have successfully created
3762 * the target name, but have failed to remove
3763 * the source name. Since the create was done
3764 * with the ZRENAMING flag, there are
3765 * complications; for one, the link count is
3766 * wrong. The easiest way to deal with this
3767 * is to remove the newly created target, and
3768 * return the original error. This must
3769 * succeed; fortunately, it is very unlikely to
3770 * fail, since we just created it.
3772 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3773 ZRENAMING
, NULL
), ==, 0);
3780 if (tzp
&& rm_err
== 0)
3781 vnevent_rename_dest(ZTOV(tzp
), tdvp
, tnm
, ct
);
3784 vnevent_rename_src(ZTOV(szp
), sdvp
, snm
, ct
);
3785 /* notify the target dir if it is not the same as source dir */
3787 vnevent_rename_dest_dir(tdvp
, ct
);
3791 zfs_rename_unlock(&zl
);
3793 zfs_dirent_unlock(sdl
);
3794 zfs_dirent_unlock(tdl
);
3797 rw_exit(&sdzp
->z_name_lock
);
3804 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3805 zil_commit(zilog
, 0);
3812 * Insert the indicated symbolic reference entry into the directory.
3814 * IN: dvp - Directory to contain new symbolic link.
3815 * link - Name for new symlink entry.
3816 * vap - Attributes of new entry.
3817 * cr - credentials of caller.
3818 * ct - caller context
3819 * flags - case flags
3821 * RETURN: 0 on success, error code on failure.
3824 * dvp - ctime|mtime updated
3828 zfs_symlink(vnode_t
*dvp
, char *name
, vattr_t
*vap
, char *link
, cred_t
*cr
,
3829 caller_context_t
*ct
, int flags
)
3831 znode_t
*zp
, *dzp
= VTOZ(dvp
);
3834 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
3836 uint64_t len
= strlen(link
);
3839 zfs_acl_ids_t acl_ids
;
3840 boolean_t fuid_dirtied
;
3841 uint64_t txtype
= TX_SYMLINK
;
3842 boolean_t waited
= B_FALSE
;
3844 ASSERT(vap
->va_type
== VLNK
);
3848 zilog
= zfsvfs
->z_log
;
3850 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
3851 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3853 return (SET_ERROR(EILSEQ
));
3855 if (flags
& FIGNORECASE
)
3858 if (len
> MAXPATHLEN
) {
3860 return (SET_ERROR(ENAMETOOLONG
));
3863 if ((error
= zfs_acl_ids_create(dzp
, 0,
3864 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3870 * Attempt to lock directory; fail if entry already exists.
3872 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3874 zfs_acl_ids_free(&acl_ids
);
3879 if (error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
)) {
3880 zfs_acl_ids_free(&acl_ids
);
3881 zfs_dirent_unlock(dl
);
3886 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
3887 zfs_acl_ids_free(&acl_ids
);
3888 zfs_dirent_unlock(dl
);
3890 return (SET_ERROR(EDQUOT
));
3892 tx
= dmu_tx_create(zfsvfs
->z_os
);
3893 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3894 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3895 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3896 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3897 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3898 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3899 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3900 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3901 acl_ids
.z_aclp
->z_acl_bytes
);
3904 zfs_fuid_txhold(zfsvfs
, tx
);
3905 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
3907 zfs_dirent_unlock(dl
);
3908 if (error
== ERESTART
) {
3914 zfs_acl_ids_free(&acl_ids
);
3921 * Create a new object for the symlink.
3922 * for version 4 ZPL datsets the symlink will be an SA attribute
3924 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3927 zfs_fuid_sync(zfsvfs
, tx
);
3929 mutex_enter(&zp
->z_lock
);
3931 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zfsvfs
),
3934 zfs_sa_symlink(zp
, link
, len
, tx
);
3935 mutex_exit(&zp
->z_lock
);
3938 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
3939 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3941 * Insert the new object into the directory.
3943 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3945 if (flags
& FIGNORECASE
)
3947 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3949 zfs_acl_ids_free(&acl_ids
);
3953 zfs_dirent_unlock(dl
);
3957 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3958 zil_commit(zilog
, 0);
3965 * Return, in the buffer contained in the provided uio structure,
3966 * the symbolic path referred to by vp.
3968 * IN: vp - vnode of symbolic link.
3969 * uio - structure to contain the link path.
3970 * cr - credentials of caller.
3971 * ct - caller context
3973 * OUT: uio - structure containing the link path.
3975 * RETURN: 0 on success, error code on failure.
3978 * vp - atime updated
3982 zfs_readlink(vnode_t
*vp
, uio_t
*uio
, cred_t
*cr
, caller_context_t
*ct
)
3984 znode_t
*zp
= VTOZ(vp
);
3985 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
3991 mutex_enter(&zp
->z_lock
);
3993 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3994 SA_ZPL_SYMLINK(zfsvfs
), uio
);
3996 error
= zfs_sa_readlink(zp
, uio
);
3997 mutex_exit(&zp
->z_lock
);
3999 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
4006 * Insert a new entry into directory tdvp referencing svp.
4008 * IN: tdvp - Directory to contain new entry.
4009 * svp - vnode of new entry.
4010 * name - name of new entry.
4011 * cr - credentials of caller.
4012 * ct - caller context
4014 * RETURN: 0 on success, error code on failure.
4017 * tdvp - ctime|mtime updated
4018 * svp - ctime updated
4022 zfs_link(vnode_t
*tdvp
, vnode_t
*svp
, char *name
, cred_t
*cr
,
4023 caller_context_t
*ct
, int flags
)
4025 znode_t
*dzp
= VTOZ(tdvp
);
4027 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
4036 boolean_t waited
= B_FALSE
;
4038 ASSERT(tdvp
->v_type
== VDIR
);
4042 zilog
= zfsvfs
->z_log
;
4044 if (fop_realvp(svp
, &realvp
, ct
) == 0)
4048 * POSIX dictates that we return EPERM here.
4049 * Better choices include ENOTSUP or EISDIR.
4051 if (svp
->v_type
== VDIR
) {
4053 return (SET_ERROR(EPERM
));
4060 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4061 * ctldir appear to have the same v_vfsp.
4063 if (szp
->z_zfsvfs
!= zfsvfs
|| zfsctl_is_node(svp
)) {
4065 return (SET_ERROR(EXDEV
));
4068 /* Prevent links to .zfs/shares files */
4070 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
4071 &parent
, sizeof (uint64_t))) != 0) {
4075 if (parent
== zfsvfs
->z_shares_dir
) {
4077 return (SET_ERROR(EPERM
));
4080 if (zfsvfs
->z_utf8
&& u8_validate(name
,
4081 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
4083 return (SET_ERROR(EILSEQ
));
4085 if (flags
& FIGNORECASE
)
4089 * We do not support links between attributes and non-attributes
4090 * because of the potential security risk of creating links
4091 * into "normal" file space in order to circumvent restrictions
4092 * imposed in attribute space.
4094 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
4096 return (SET_ERROR(EINVAL
));
4100 owner
= zfs_fuid_map_id(zfsvfs
, szp
->z_uid
, cr
, ZFS_OWNER
);
4101 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
4103 return (SET_ERROR(EPERM
));
4106 if (error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
)) {
4113 * Attempt to lock directory; fail if entry already exists.
4115 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
4121 tx
= dmu_tx_create(zfsvfs
->z_os
);
4122 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
4123 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
4124 zfs_sa_upgrade_txholds(tx
, szp
);
4125 zfs_sa_upgrade_txholds(tx
, dzp
);
4126 error
= dmu_tx_assign(tx
, (waited
? TXG_NOTHROTTLE
: 0) | TXG_NOWAIT
);
4128 zfs_dirent_unlock(dl
);
4129 if (error
== ERESTART
) {
4140 error
= zfs_link_create(dl
, szp
, tx
, 0);
4143 uint64_t txtype
= TX_LINK
;
4144 if (flags
& FIGNORECASE
)
4146 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
4151 zfs_dirent_unlock(dl
);
4154 vnevent_link(svp
, ct
);
4157 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4158 zil_commit(zilog
, 0);
4165 * zfs_null_putapage() is used when the file system has been force
4166 * unmounted. It just drops the pages.
4170 zfs_null_putapage(vnode_t
*vp
, page_t
*pp
, uoff_t
*offp
,
4171 size_t *lenp
, int flags
, cred_t
*cr
)
4173 pvn_write_done(pp
, B_INVAL
|B_FORCE
|B_ERROR
);
4178 * Push a page out to disk, klustering if possible.
4180 * IN: vp - file to push page to.
4181 * pp - page to push.
4182 * flags - additional flags.
4183 * cr - credentials of caller.
4185 * OUT: offp - start of range pushed.
4186 * lenp - len of range pushed.
4188 * RETURN: 0 on success, error code on failure.
4190 * NOTE: callers must have locked the page to be pushed. On
4191 * exit, the page (and all other pages in the kluster) must be
4196 zfs_putapage(vnode_t
*vp
, page_t
*pp
, uoff_t
*offp
,
4197 size_t *lenp
, int flags
, cred_t
*cr
)
4199 znode_t
*zp
= VTOZ(vp
);
4200 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4209 * If our blocksize is bigger than the page size, try to kluster
4210 * multiple pages so that we write a full block (thus avoiding
4211 * a read-modify-write).
4213 if (off
< zp
->z_size
&& zp
->z_blksz
> PAGESIZE
) {
4214 klen
= P2ROUNDUP((ulong_t
)zp
->z_blksz
, PAGESIZE
);
4215 koff
= ISP2(klen
) ? P2ALIGN(off
, (uoff_t
)klen
) : 0;
4216 ASSERT(koff
<= zp
->z_size
);
4217 if (koff
+ klen
> zp
->z_size
)
4218 klen
= P2ROUNDUP(zp
->z_size
- koff
, (uint64_t)PAGESIZE
);
4219 pp
= pvn_write_kluster(vp
, pp
, &off
, &len
, koff
, klen
, flags
);
4221 ASSERT3U(btop(len
), ==, btopr(len
));
4224 * Can't push pages past end-of-file.
4226 if (off
>= zp
->z_size
) {
4227 /* ignore all pages */
4230 } else if (off
+ len
> zp
->z_size
) {
4231 int npages
= btopr(zp
->z_size
- off
);
4234 page_list_break(&pp
, &trunc
, npages
);
4235 /* ignore pages past end of file */
4237 pvn_write_done(trunc
, flags
);
4238 len
= zp
->z_size
- off
;
4241 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
4242 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
4243 err
= SET_ERROR(EDQUOT
);
4246 tx
= dmu_tx_create(zfsvfs
->z_os
);
4247 dmu_tx_hold_write(tx
, zp
->z_id
, off
, len
);
4249 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4250 zfs_sa_upgrade_txholds(tx
, zp
);
4251 err
= dmu_tx_assign(tx
, TXG_WAIT
);
4257 if (zp
->z_blksz
<= PAGESIZE
) {
4258 caddr_t va
= zfs_map_page(pp
, S_READ
);
4259 ASSERT3U(len
, <=, PAGESIZE
);
4260 dmu_write(zfsvfs
->z_os
, zp
->z_id
, off
, len
, va
, tx
);
4261 zfs_unmap_page(pp
, va
);
4263 err
= dmu_write_pages(zfsvfs
->z_os
, zp
->z_id
, off
, len
, pp
, tx
);
4267 uint64_t mtime
[2], ctime
[2];
4268 sa_bulk_attr_t bulk
[3];
4271 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
4273 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
4275 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
4277 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
4279 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
4281 zfs_log_write(zfsvfs
->z_log
, tx
, TX_WRITE
, zp
, off
, len
, 0);
4286 pvn_write_done(pp
, (err
? B_ERROR
: 0) | flags
);
4296 * Copy the portion of the file indicated from pages into the file.
4297 * The pages are stored in a page list attached to the files vnode.
4299 * IN: vp - vnode of file to push page data to.
4300 * off - position in file to put data.
4301 * len - amount of data to write.
4302 * flags - flags to control the operation.
4303 * cr - credentials of caller.
4304 * ct - caller context.
4306 * RETURN: 0 on success, error code on failure.
4309 * vp - ctime|mtime updated
4313 zfs_putpage(vnode_t
*vp
, offset_t off
, size_t len
, int flags
, cred_t
*cr
,
4314 caller_context_t
*ct
)
4316 znode_t
*zp
= VTOZ(vp
);
4317 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4329 * There's nothing to do if no data is cached.
4331 if (!vn_has_cached_data(vp
)) {
4337 * Align this request to the file block size in case we kluster.
4338 * XXX - this can result in pretty aggresive locking, which can
4339 * impact simultanious read/write access. One option might be
4340 * to break up long requests (len == 0) into block-by-block
4341 * operations to get narrower locking.
4343 blksz
= zp
->z_blksz
;
4345 io_off
= P2ALIGN_TYPED(off
, blksz
, uoff_t
);
4348 if (len
> 0 && ISP2(blksz
))
4349 io_len
= P2ROUNDUP_TYPED(len
+ (off
- io_off
), blksz
, size_t);
4355 * Search the entire vp list for pages >= io_off.
4357 rl
= zfs_range_lock(zp
, io_off
, UINT64_MAX
, RL_WRITER
);
4358 error
= pvn_vplist_dirty(vp
, io_off
, zfs_putapage
, flags
, cr
);
4361 rl
= zfs_range_lock(zp
, io_off
, io_len
, RL_WRITER
);
4363 if (off
> zp
->z_size
) {
4364 /* past end of file */
4365 zfs_range_unlock(rl
);
4370 len
= MIN(io_len
, P2ROUNDUP(zp
->z_size
, PAGESIZE
) - io_off
);
4372 for (off
= io_off
; io_off
< off
+ len
; io_off
+= io_len
) {
4373 if ((flags
& B_INVAL
) || ((flags
& B_ASYNC
) == 0)) {
4374 pp
= page_lookup(&vp
->v_object
, io_off
,
4375 (flags
& (B_INVAL
| B_FREE
)) ? SE_EXCL
: SE_SHARED
);
4377 pp
= page_lookup_nowait(&vp
->v_object
, io_off
,
4378 (flags
& B_FREE
) ? SE_EXCL
: SE_SHARED
);
4381 if (pp
!= NULL
&& pvn_getdirty(pp
, flags
)) {
4385 * Found a dirty page to push
4387 err
= zfs_putapage(vp
, pp
, &io_off
, &io_len
, flags
, cr
);
4395 zfs_range_unlock(rl
);
4396 if ((flags
& B_ASYNC
) == 0 || zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4397 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
4404 zfs_inactive(vnode_t
*vp
, cred_t
*cr
, caller_context_t
*ct
)
4406 znode_t
*zp
= VTOZ(vp
);
4407 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4410 rw_enter(&zfsvfs
->z_teardown_inactive_lock
, RW_READER
);
4411 if (zp
->z_sa_hdl
== NULL
) {
4413 * The fs has been unmounted, or we did a
4414 * suspend/resume and this file no longer exists.
4416 if (vn_has_cached_data(vp
)) {
4417 (void) pvn_vplist_dirty(vp
, 0, zfs_null_putapage
,
4421 mutex_enter(&zp
->z_lock
);
4422 mutex_enter(&vp
->v_lock
);
4423 ASSERT(vp
->v_count
== 1);
4425 mutex_exit(&vp
->v_lock
);
4426 mutex_exit(&zp
->z_lock
);
4427 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4433 * Attempt to push any data in the page cache. If this fails
4434 * we will get kicked out later in zfs_zinactive().
4436 if (vn_has_cached_data(vp
)) {
4437 (void) pvn_vplist_dirty(vp
, 0, zfs_putapage
, B_INVAL
|B_ASYNC
,
4441 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4442 dmu_tx_t
*tx
= dmu_tx_create(zfsvfs
->z_os
);
4444 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4445 zfs_sa_upgrade_txholds(tx
, zp
);
4446 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4450 mutex_enter(&zp
->z_lock
);
4451 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zfsvfs
),
4452 (void *)&zp
->z_atime
, sizeof (zp
->z_atime
), tx
);
4453 zp
->z_atime_dirty
= 0;
4454 mutex_exit(&zp
->z_lock
);
4460 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4464 * Bounds-check the seek operation.
4466 * IN: vp - vnode seeking within
4467 * ooff - old file offset
4468 * noffp - pointer to new file offset
4469 * ct - caller context
4471 * RETURN: 0 on success, EINVAL if new offset invalid.
4475 zfs_seek(vnode_t
*vp
, offset_t ooff
, offset_t
*noffp
,
4476 caller_context_t
*ct
)
4478 if (vp
->v_type
== VDIR
)
4480 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4484 * Pre-filter the generic locking function to trap attempts to place
4485 * a mandatory lock on a memory mapped file.
4488 zfs_frlock(vnode_t
*vp
, int cmd
, flock64_t
*bfp
, int flag
, offset_t offset
,
4489 flk_callback_t
*flk_cbp
, cred_t
*cr
, caller_context_t
*ct
)
4491 znode_t
*zp
= VTOZ(vp
);
4492 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4498 * We are following the UFS semantics with respect to mapcnt
4499 * here: If we see that the file is mapped already, then we will
4500 * return an error, but we don't worry about races between this
4501 * function and zfs_map().
4503 if (zp
->z_mapcnt
> 0 && MANDMODE(zp
->z_mode
)) {
4505 return (SET_ERROR(EAGAIN
));
4508 return (fs_frlock(vp
, cmd
, bfp
, flag
, offset
, flk_cbp
, cr
, ct
));
4512 * If we can't find a page in the cache, we will create a new page
4513 * and fill it with file data. For efficiency, we may try to fill
4514 * multiple pages at once (klustering) to fill up the supplied page
4515 * list. Note that the pages to be filled are held with an exclusive
4516 * lock to prevent access by other threads while they are being filled.
4519 zfs_fillpage(vnode_t
*vp
, uoff_t off
, struct seg
*seg
,
4520 caddr_t addr
, page_t
*pl
[], size_t plsz
, enum seg_rw rw
)
4522 znode_t
*zp
= VTOZ(vp
);
4523 page_t
*pp
, *cur_pp
;
4524 objset_t
*os
= zp
->z_zfsvfs
->z_os
;
4525 uoff_t io_off
, total
;
4529 if (plsz
== PAGESIZE
|| zp
->z_blksz
<= PAGESIZE
) {
4531 * We only have a single page, don't bother klustering
4535 pp
= page_create_va(&vp
->v_object
, io_off
, io_len
,
4536 PG_EXCL
| PG_WAIT
, seg
, addr
);
4539 * Try to find enough pages to fill the page list
4541 pp
= pvn_read_kluster(vp
, off
, seg
, addr
, &io_off
,
4542 &io_len
, off
, plsz
, 0);
4546 * The page already exists, nothing to do here.
4553 * Fill the pages in the kluster.
4556 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4559 ASSERT3U(io_off
, ==, cur_pp
->p_offset
);
4560 va
= zfs_map_page(cur_pp
, S_WRITE
);
4561 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4563 zfs_unmap_page(cur_pp
, va
);
4565 /* On error, toss the entire kluster */
4566 pvn_read_done(pp
, B_ERROR
);
4567 /* convert checksum errors into IO errors */
4569 err
= SET_ERROR(EIO
);
4572 cur_pp
= cur_pp
->p_next
;
4576 * Fill in the page list array from the kluster starting
4577 * from the desired offset `off'.
4578 * NOTE: the page list will always be null terminated.
4580 pvn_plist_init(pp
, pl
, plsz
, off
, io_len
, rw
);
4581 ASSERT(pl
== NULL
|| (*pl
)->p_offset
== off
);
4587 * Return pointers to the pages for the file region [off, off + len]
4588 * in the pl array. If plsz is greater than len, this function may
4589 * also return page pointers from after the specified region
4590 * (i.e. the region [off, off + plsz]). These additional pages are
4591 * only returned if they are already in the cache, or were created as
4592 * part of a klustered read.
4594 * IN: vp - vnode of file to get data from.
4595 * off - position in file to get data from.
4596 * len - amount of data to retrieve.
4597 * plsz - length of provided page list.
4598 * seg - segment to obtain pages for.
4599 * addr - virtual address of fault.
4600 * rw - mode of created pages.
4601 * cr - credentials of caller.
4602 * ct - caller context.
4604 * OUT: protp - protection mode of created pages.
4605 * pl - list of pages created.
4607 * RETURN: 0 on success, error code on failure.
4610 * vp - atime updated
4614 zfs_getpage(vnode_t
*vp
, offset_t off
, size_t len
, uint_t
*protp
,
4615 page_t
*pl
[], size_t plsz
, struct seg
*seg
, caddr_t addr
,
4616 enum seg_rw rw
, cred_t
*cr
, caller_context_t
*ct
)
4618 znode_t
*zp
= VTOZ(vp
);
4619 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4623 /* we do our own caching, faultahead is unnecessary */
4626 else if (len
> plsz
)
4629 len
= P2ROUNDUP(len
, PAGESIZE
);
4630 ASSERT(plsz
>= len
);
4639 * Loop through the requested range [off, off + len) looking
4640 * for pages. If we don't find a page, we will need to create
4641 * a new page and fill it with data from the file.
4644 if (*pl
= page_lookup(&vp
->v_object
, off
, SE_SHARED
))
4646 else if (err
= zfs_fillpage(vp
, off
, seg
, addr
, pl
, plsz
, rw
))
4649 ASSERT3U((*pl
)->p_offset
, ==, off
);
4653 ASSERT3U(len
, >=, PAGESIZE
);
4656 ASSERT3U(plsz
, >=, PAGESIZE
);
4663 * Fill out the page array with any pages already in the cache.
4666 (*pl
++ = page_lookup_nowait(&vp
->v_object
, off
, SE_SHARED
))) {
4673 * Release any pages we have previously locked.
4678 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
4688 * Request a memory map for a section of a file. This code interacts
4689 * with common code and the VM system as follows:
4691 * - common code calls mmap(), which ends up in smmap_common()
4692 * - this calls fop_map(), which takes you into (say) zfs
4693 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4694 * - segvn_create() creates the new segment and calls fop_addmap()
4695 * - zfs_addmap() updates z_mapcnt
4699 zfs_map(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t
*addrp
,
4700 size_t len
, uchar_t prot
, uchar_t maxprot
, uint_t flags
, cred_t
*cr
,
4701 caller_context_t
*ct
)
4703 znode_t
*zp
= VTOZ(vp
);
4704 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4705 segvn_crargs_t vn_a
;
4712 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
4715 if ((prot
& PROT_WRITE
) && (zp
->z_pflags
&
4716 (ZFS_IMMUTABLE
| ZFS_APPENDONLY
))) {
4718 return (SET_ERROR(EPERM
));
4721 if ((prot
& (PROT_READ
| PROT_EXEC
)) &&
4722 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4724 return (SET_ERROR(EACCES
));
4727 if (vp
->v_flag
& VNOMAP
) {
4729 return (SET_ERROR(ENOSYS
));
4732 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4734 return (SET_ERROR(ENXIO
));
4737 if (vp
->v_type
!= VREG
) {
4739 return (SET_ERROR(ENODEV
));
4743 * If file is locked, disallow mapping.
4745 if (MANDMODE(zp
->z_mode
) && vn_has_flocks(vp
)) {
4747 return (SET_ERROR(EAGAIN
));
4751 error
= choose_addr(as
, addrp
, len
, off
, ADDR_VACALIGN
, flags
);
4759 vn_a
.offset
= (uoff_t
)off
;
4760 vn_a
.type
= flags
& MAP_TYPE
;
4762 vn_a
.maxprot
= maxprot
;
4765 vn_a
.flags
= flags
& ~MAP_TYPE
;
4767 vn_a
.lgrp_mem_policy_flags
= 0;
4769 error
= as_map(as
, *addrp
, len
, segvn_create
, &vn_a
);
4778 zfs_addmap(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t addr
,
4779 size_t len
, uchar_t prot
, uchar_t maxprot
, uint_t flags
, cred_t
*cr
,
4780 caller_context_t
*ct
)
4782 uint64_t pages
= btopr(len
);
4784 atomic_add_64(&VTOZ(vp
)->z_mapcnt
, pages
);
4789 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4790 * more accurate mtime for the associated file. Since we don't have a way of
4791 * detecting when the data was actually modified, we have to resort to
4792 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4793 * last page is pushed. The problem occurs when the msync() call is omitted,
4794 * which by far the most common case:
4802 * putpage() via fsflush
4804 * If we wait until fsflush to come along, we can have a modification time that
4805 * is some arbitrary point in the future. In order to prevent this in the
4806 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4811 zfs_delmap(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t addr
,
4812 size_t len
, uint_t prot
, uint_t maxprot
, uint_t flags
, cred_t
*cr
,
4813 caller_context_t
*ct
)
4815 uint64_t pages
= btopr(len
);
4817 ASSERT3U(VTOZ(vp
)->z_mapcnt
, >=, pages
);
4818 atomic_add_64(&VTOZ(vp
)->z_mapcnt
, -pages
);
4820 if ((flags
& MAP_SHARED
) && (prot
& PROT_WRITE
) &&
4821 vn_has_cached_data(vp
))
4822 (void) fop_putpage(vp
, off
, len
, B_ASYNC
, cr
, ct
);
4828 * Free or allocate space in a file. Currently, this function only
4829 * supports the `F_FREESP' command. However, this command is somewhat
4830 * misnamed, as its functionality includes the ability to allocate as
4831 * well as free space.
4833 * IN: vp - vnode of file to free data in.
4834 * cmd - action to take (only F_FREESP supported).
4835 * bfp - section of file to free/alloc.
4836 * flag - current file open mode flags.
4837 * offset - current file offset.
4838 * cr - credentials of caller [UNUSED].
4839 * ct - caller context.
4841 * RETURN: 0 on success, error code on failure.
4844 * vp - ctime|mtime updated
4848 zfs_space(vnode_t
*vp
, int cmd
, flock64_t
*bfp
, int flag
,
4849 offset_t offset
, cred_t
*cr
, caller_context_t
*ct
)
4851 znode_t
*zp
= VTOZ(vp
);
4852 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4859 if (cmd
!= F_FREESP
) {
4861 return (SET_ERROR(EINVAL
));
4865 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
4866 * callers might not be able to detect properly that we are read-only,
4867 * so check it explicitly here.
4869 if (zfsvfs
->z_vfs
->vfs_flag
& VFS_RDONLY
) {
4871 return (SET_ERROR(EROFS
));
4874 if (error
= convoff(vp
, bfp
, 0, offset
)) {
4879 if (bfp
->l_len
< 0) {
4881 return (SET_ERROR(EINVAL
));
4885 len
= bfp
->l_len
; /* 0 means from off to end of file */
4887 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4889 if (error
== 0 && off
== 0 && len
== 0)
4890 vnevent_truncate(ZTOV(zp
), ct
);
4898 zfs_fid(vnode_t
*vp
, fid_t
*fidp
, caller_context_t
*ct
)
4900 znode_t
*zp
= VTOZ(vp
);
4901 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4904 uint64_t object
= zp
->z_id
;
4911 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zfsvfs
),
4912 &gen64
, sizeof (uint64_t))) != 0) {
4917 gen
= (uint32_t)gen64
;
4919 size
= (zfsvfs
->z_parent
!= zfsvfs
) ? LONG_FID_LEN
: SHORT_FID_LEN
;
4920 if (fidp
->fid_len
< size
) {
4921 fidp
->fid_len
= size
;
4923 return (SET_ERROR(ENOSPC
));
4926 zfid
= (zfid_short_t
*)fidp
;
4928 zfid
->zf_len
= size
;
4930 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4931 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4933 /* Must have a non-zero generation number to distinguish from .zfs */
4936 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4937 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4939 if (size
== LONG_FID_LEN
) {
4940 uint64_t objsetid
= dmu_objset_id(zfsvfs
->z_os
);
4943 zlfid
= (zfid_long_t
*)fidp
;
4945 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
4946 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
4948 /* XXX - this should be the generation number for the objset */
4949 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
4950 zlfid
->zf_setgen
[i
] = 0;
4958 zfs_pathconf(vnode_t
*vp
, int cmd
, ulong_t
*valp
, cred_t
*cr
,
4959 caller_context_t
*ct
)
4971 case _PC_FILESIZEBITS
:
4975 case _PC_XATTR_EXISTS
:
4977 zfsvfs
= zp
->z_zfsvfs
;
4981 error
= zfs_dirent_lock(&dl
, zp
, "", &xzp
,
4982 ZXATTR
| ZEXISTS
| ZSHARED
, NULL
, NULL
);
4984 zfs_dirent_unlock(dl
);
4985 if (!zfs_dirempty(xzp
))
4988 } else if (error
== ENOENT
) {
4990 * If there aren't extended attributes, it's the
4991 * same as having zero of them.
4998 case _PC_SATTR_ENABLED
:
4999 case _PC_SATTR_EXISTS
:
5000 *valp
= vfs_has_feature(vp
->v_vfsp
, VFSFT_SYSATTR_VIEWS
) &&
5001 (vp
->v_type
== VREG
|| vp
->v_type
== VDIR
);
5004 case _PC_ACCESS_FILTERING
:
5005 *valp
= vfs_has_feature(vp
->v_vfsp
, VFSFT_ACCESS_FILTER
) &&
5009 case _PC_ACL_ENABLED
:
5010 *valp
= _ACL_ACE_ENABLED
;
5013 case _PC_MIN_HOLE_SIZE
:
5014 *valp
= (ulong_t
)SPA_MINBLOCKSIZE
;
5017 case _PC_TIMESTAMP_RESOLUTION
:
5018 /* nanosecond timestamp resolution */
5023 return (fs_pathconf(vp
, cmd
, valp
, cr
, ct
));
5029 zfs_getsecattr(vnode_t
*vp
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
,
5030 caller_context_t
*ct
)
5032 znode_t
*zp
= VTOZ(vp
);
5033 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
5035 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
5039 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
5047 zfs_setsecattr(vnode_t
*vp
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
,
5048 caller_context_t
*ct
)
5050 znode_t
*zp
= VTOZ(vp
);
5051 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
5053 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
5054 zilog_t
*zilog
= zfsvfs
->z_log
;
5059 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
5061 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
5062 zil_commit(zilog
, 0);
5069 * The smallest read we may consider to loan out an arcbuf.
5070 * This must be a power of 2.
5072 int zcr_blksz_min
= (1 << 10); /* 1K */
5074 * If set to less than the file block size, allow loaning out of an
5075 * arcbuf for a partial block read. This must be a power of 2.
5077 int zcr_blksz_max
= (1 << 17); /* 128K */
5081 zfs_reqzcbuf(vnode_t
*vp
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
,
5082 caller_context_t
*ct
)
5084 znode_t
*zp
= VTOZ(vp
);
5085 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
5086 int max_blksz
= zfsvfs
->z_max_blksz
;
5087 uio_t
*uio
= &xuio
->xu_uio
;
5088 ssize_t size
= uio
->uio_resid
;
5089 offset_t offset
= uio
->uio_loffset
;
5094 int preamble
, postamble
;
5096 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
5097 return (SET_ERROR(EINVAL
));
5104 * Loan out an arc_buf for write if write size is bigger than
5105 * max_blksz, and the file's block size is also max_blksz.
5108 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
5110 return (SET_ERROR(EINVAL
));
5113 * Caller requests buffers for write before knowing where the
5114 * write offset might be (e.g. NFS TCP write).
5119 preamble
= P2PHASE(offset
, blksz
);
5121 preamble
= blksz
- preamble
;
5126 postamble
= P2PHASE(size
, blksz
);
5129 fullblk
= size
/ blksz
;
5130 (void) dmu_xuio_init(xuio
,
5131 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
5132 DTRACE_PROBE3(zfs_reqzcbuf_align
, int, preamble
,
5133 int, postamble
, int,
5134 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
5137 * Have to fix iov base/len for partial buffers. They
5138 * currently represent full arc_buf's.
5141 /* data begins in the middle of the arc_buf */
5142 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5145 (void) dmu_xuio_add(xuio
, abuf
,
5146 blksz
- preamble
, preamble
);
5149 for (i
= 0; i
< fullblk
; i
++) {
5150 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5153 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
5157 /* data ends in the middle of the arc_buf */
5158 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5161 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
5166 * Loan out an arc_buf for read if the read size is larger than
5167 * the current file block size. Block alignment is not
5168 * considered. Partial arc_buf will be loaned out for read.
5170 blksz
= zp
->z_blksz
;
5171 if (blksz
< zcr_blksz_min
)
5172 blksz
= zcr_blksz_min
;
5173 if (blksz
> zcr_blksz_max
)
5174 blksz
= zcr_blksz_max
;
5175 /* avoid potential complexity of dealing with it */
5176 if (blksz
> max_blksz
) {
5178 return (SET_ERROR(EINVAL
));
5181 maxsize
= zp
->z_size
- uio
->uio_loffset
;
5185 if (size
< blksz
|| vn_has_cached_data(vp
)) {
5187 return (SET_ERROR(EINVAL
));
5192 return (SET_ERROR(EINVAL
));
5195 uio
->uio_extflg
= UIO_XUIO
;
5196 XUIO_XUZC_RW(xuio
) = ioflag
;
5203 zfs_retzcbuf(vnode_t
*vp
, xuio_t
*xuio
, cred_t
*cr
, caller_context_t
*ct
)
5207 int ioflag
= XUIO_XUZC_RW(xuio
);
5209 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
5211 i
= dmu_xuio_cnt(xuio
);
5213 abuf
= dmu_xuio_arcbuf(xuio
, i
);
5215 * if abuf == NULL, it must be a write buffer
5216 * that has been returned in zfs_write().
5219 dmu_return_arcbuf(abuf
);
5220 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
5223 dmu_xuio_fini(xuio
);
5228 * Predeclare these here so that the compiler assumes that
5229 * this is an "old style" function declaration that does
5230 * not include arguments => we won't get type mismatch errors
5231 * in the initializations that follow.
5233 static int zfs_inval();
5234 static int zfs_isdir();
5239 return (SET_ERROR(EINVAL
));
5245 return (SET_ERROR(EISDIR
));
5249 * Directory vnode operations
5251 const struct vnodeops zfs_dvnodeops
= {
5253 .vop_open
= zfs_open
,
5254 .vop_close
= zfs_close
,
5255 .vop_read
= zfs_isdir
,
5256 .vop_write
= zfs_isdir
,
5257 .vop_ioctl
= zfs_ioctl
,
5258 .vop_getattr
= zfs_getattr
,
5259 .vop_setattr
= zfs_setattr
,
5260 .vop_access
= zfs_access
,
5261 .vop_lookup
= zfs_lookup
,
5262 .vop_create
= zfs_create
,
5263 .vop_remove
= zfs_remove
,
5264 .vop_link
= zfs_link
,
5265 .vop_rename
= zfs_rename
,
5266 .vop_mkdir
= zfs_mkdir
,
5267 .vop_rmdir
= zfs_rmdir
,
5268 .vop_readdir
= zfs_readdir
,
5269 .vop_symlink
= zfs_symlink
,
5270 .vop_fsync
= zfs_fsync
,
5271 .vop_inactive
= zfs_inactive
,
5273 .vop_seek
= zfs_seek
,
5274 .vop_pathconf
= zfs_pathconf
,
5275 .vop_getsecattr
= zfs_getsecattr
,
5276 .vop_setsecattr
= zfs_setsecattr
,
5277 .vop_vnevent
= fs_vnevent_support
,
5281 * Regular file vnode operations
5283 const struct vnodeops zfs_fvnodeops
= {
5285 .vop_open
= zfs_open
,
5286 .vop_close
= zfs_close
,
5287 .vop_read
= zfs_read
,
5288 .vop_write
= zfs_write
,
5289 .vop_ioctl
= zfs_ioctl
,
5290 .vop_getattr
= zfs_getattr
,
5291 .vop_setattr
= zfs_setattr
,
5292 .vop_access
= zfs_access
,
5293 .vop_lookup
= zfs_lookup
,
5294 .vop_rename
= zfs_rename
,
5295 .vop_fsync
= zfs_fsync
,
5296 .vop_inactive
= zfs_inactive
,
5298 .vop_seek
= zfs_seek
,
5299 .vop_frlock
= zfs_frlock
,
5300 .vop_space
= zfs_space
,
5301 .vop_getpage
= zfs_getpage
,
5302 .vop_putpage
= zfs_putpage
,
5304 .vop_addmap
= zfs_addmap
,
5305 .vop_delmap
= zfs_delmap
,
5306 .vop_pathconf
= zfs_pathconf
,
5307 .vop_getsecattr
= zfs_getsecattr
,
5308 .vop_setsecattr
= zfs_setsecattr
,
5309 .vop_vnevent
= fs_vnevent_support
,
5310 .vop_reqzcbuf
= zfs_reqzcbuf
,
5311 .vop_retzcbuf
= zfs_retzcbuf
,
5315 * Symbolic link vnode operations
5317 const struct vnodeops zfs_symvnodeops
= {
5319 .vop_getattr
= zfs_getattr
,
5320 .vop_setattr
= zfs_setattr
,
5321 .vop_access
= zfs_access
,
5322 .vop_rename
= zfs_rename
,
5323 .vop_readlink
= zfs_readlink
,
5324 .vop_inactive
= zfs_inactive
,
5326 .vop_pathconf
= zfs_pathconf
,
5327 .vop_vnevent
= fs_vnevent_support
,
5331 * special share hidden files vnode operations
5333 const struct vnodeops zfs_sharevnodeops
= {
5335 .vop_getattr
= zfs_getattr
,
5336 .vop_access
= zfs_access
,
5337 .vop_inactive
= zfs_inactive
,
5339 .vop_pathconf
= zfs_pathconf
,
5340 .vop_getsecattr
= zfs_getsecattr
,
5341 .vop_setsecattr
= zfs_setsecattr
,
5342 .vop_vnevent
= fs_vnevent_support
,
5346 * Extended attribute directory vnode operations
5348 * These ops are identical to the directory vnode
5349 * operations except for restricted operations:
5353 * Note that there are other restrictions embedded in:
5354 * zfs_create() - restrict type to VREG
5355 * zfs_link() - no links into/out of attribute space
5356 * zfs_rename() - no moves into/out of attribute space
5358 const struct vnodeops zfs_xdvnodeops
= {
5360 .vop_open
= zfs_open
,
5361 .vop_close
= zfs_close
,
5362 .vop_ioctl
= zfs_ioctl
,
5363 .vop_getattr
= zfs_getattr
,
5364 .vop_setattr
= zfs_setattr
,
5365 .vop_access
= zfs_access
,
5366 .vop_lookup
= zfs_lookup
,
5367 .vop_create
= zfs_create
,
5368 .vop_remove
= zfs_remove
,
5369 .vop_link
= zfs_link
,
5370 .vop_rename
= zfs_rename
,
5371 .vop_mkdir
= zfs_inval
,
5372 .vop_rmdir
= zfs_rmdir
,
5373 .vop_readdir
= zfs_readdir
,
5374 .vop_symlink
= zfs_inval
,
5375 .vop_fsync
= zfs_fsync
,
5376 .vop_inactive
= zfs_inactive
,
5378 .vop_seek
= zfs_seek
,
5379 .vop_pathconf
= zfs_pathconf
,
5380 .vop_getsecattr
= zfs_getsecattr
,
5381 .vop_setsecattr
= zfs_setsecattr
,
5382 .vop_vnevent
= fs_vnevent_support
,
5386 * Error vnode operations
5388 const struct vnodeops zfs_evnodeops
= {
5390 .vop_inactive
= zfs_inactive
,
5391 .vop_pathconf
= zfs_pathconf
,