4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 * Copyright 2015 Joyent, Inc.
27 * Copyright 2017 Nexenta Systems, Inc.
30 /* Portions Copyright 2007 Jeremy Teo */
31 /* Portions Copyright 2010 Robert Milkowski */
33 #include <sys/types.h>
34 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/resource.h>
40 #include <sys/vnode.h>
44 #include <sys/taskq.h>
46 #include <sys/vmsystm.h>
47 #include <sys/atomic.h>
49 #include <vm/seg_vn.h>
53 #include <vm/seg_kpm.h>
55 #include <sys/pathname.h>
56 #include <sys/cmn_err.h>
57 #include <sys/errno.h>
58 #include <sys/unistd.h>
59 #include <sys/zfs_dir.h>
60 #include <sys/zfs_acl.h>
61 #include <sys/zfs_ioctl.h>
62 #include <sys/fs/zfs.h>
64 #include <sys/dmu_objset.h>
70 #include <sys/dirent.h>
71 #include <sys/policy.h>
72 #include <sys/sunddi.h>
73 #include <sys/filio.h>
75 #include "sys/fs_subr.h"
76 #include <sys/zfs_ctldir.h>
77 #include <sys/zfs_fuid.h>
78 #include <sys/zfs_sa.h>
80 #include <sys/zfs_rlock.h>
81 #include <sys/extdirent.h>
82 #include <sys/kidmap.h>
90 * Each vnode op performs some logical unit of work. To do this, the ZPL must
91 * properly lock its in-core state, create a DMU transaction, do the work,
92 * record this work in the intent log (ZIL), commit the DMU transaction,
93 * and wait for the intent log to commit if it is a synchronous operation.
94 * Moreover, the vnode ops must work in both normal and log replay context.
95 * The ordering of events is important to avoid deadlocks and references
96 * to freed memory. The example below illustrates the following Big Rules:
98 * (1) A check must be made in each zfs thread for a mounted file system.
99 * This is done avoiding races using ZFS_ENTER(zfsvfs).
100 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
101 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
102 * can return EIO from the calling function.
104 * (2) VN_RELE() should always be the last thing except for zil_commit()
105 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
106 * First, if it's the last reference, the vnode/znode
107 * can be freed, so the zp may point to freed memory. Second, the last
108 * reference will call zfs_zinactive(), which may induce a lot of work --
109 * pushing cached pages (which acquires range locks) and syncing out
110 * cached atime changes. Third, zfs_zinactive() may require a new tx,
111 * which could deadlock the system if you were already holding one.
112 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
114 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
115 * as they can span dmu_tx_assign() calls.
117 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
118 * dmu_tx_assign(). This is critical because we don't want to block
119 * while holding locks.
121 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This
122 * reduces lock contention and CPU usage when we must wait (note that if
123 * throughput is constrained by the storage, nearly every transaction
126 * Note, in particular, that if a lock is sometimes acquired before
127 * the tx assigns, and sometimes after (e.g. z_lock), then failing
128 * to use a non-blocking assign can deadlock the system. The scenario:
130 * Thread A has grabbed a lock before calling dmu_tx_assign().
131 * Thread B is in an already-assigned tx, and blocks for this lock.
132 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
133 * forever, because the previous txg can't quiesce until B's tx commits.
135 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
136 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent
137 * calls to dmu_tx_assign(), pass TXG_WAITED rather than TXG_NOWAIT,
138 * to indicate that this operation has already called dmu_tx_wait().
139 * This will ensure that we don't retry forever, waiting a short bit
142 * (5) If the operation succeeded, generate the intent log entry for it
143 * before dropping locks. This ensures that the ordering of events
144 * in the intent log matches the order in which they actually occurred.
145 * During ZIL replay the zfs_log_* functions will update the sequence
146 * number to indicate the zil transaction has replayed.
148 * (6) At the end of each vnode op, the DMU tx must always commit,
149 * regardless of whether there were any errors.
151 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
152 * to ensure that synchronous semantics are provided when necessary.
154 * In general, this is how things should be ordered in each vnode op:
156 * ZFS_ENTER(zfsvfs); // exit if unmounted
158 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
159 * rw_enter(...); // grab any other locks you need
160 * tx = dmu_tx_create(...); // get DMU tx
161 * dmu_tx_hold_*(); // hold each object you might modify
162 * error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
164 * rw_exit(...); // drop locks
165 * zfs_dirent_unlock(dl); // unlock directory entry
166 * VN_RELE(...); // release held vnodes
167 * if (error == ERESTART) {
173 * dmu_tx_abort(tx); // abort DMU tx
174 * ZFS_EXIT(zfsvfs); // finished in zfs
175 * return (error); // really out of space
177 * error = do_real_work(); // do whatever this VOP does
179 * zfs_log_*(...); // on success, make ZIL entry
180 * dmu_tx_commit(tx); // commit DMU tx -- error or not
181 * rw_exit(...); // drop locks
182 * zfs_dirent_unlock(dl); // unlock directory entry
183 * VN_RELE(...); // release held vnodes
184 * zil_commit(zilog, foid); // synchronous when necessary
185 * ZFS_EXIT(zfsvfs); // finished in zfs
186 * return (error); // done, report error
191 zfs_open(vnode_t
**vpp
, int flag
, cred_t
*cr
, caller_context_t
*ct
)
193 znode_t
*zp
= VTOZ(*vpp
);
194 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
199 if ((flag
& FWRITE
) && (zp
->z_pflags
& ZFS_APPENDONLY
) &&
200 ((flag
& FAPPEND
) == 0)) {
202 return (SET_ERROR(EPERM
));
205 if (!zfs_has_ctldir(zp
) && zp
->z_zfsvfs
->z_vscan
&&
206 ZTOV(zp
)->v_type
== VREG
&&
207 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0) {
208 if (fs_vscan(*vpp
, cr
, 0) != 0) {
210 return (SET_ERROR(EACCES
));
214 /* Keep a count of the synchronous opens in the znode */
215 if (flag
& (FSYNC
| FDSYNC
))
216 atomic_inc_32(&zp
->z_sync_cnt
);
224 zfs_close(vnode_t
*vp
, int flag
, int count
, offset_t offset
, cred_t
*cr
,
225 caller_context_t
*ct
)
227 znode_t
*zp
= VTOZ(vp
);
228 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
231 * Clean up any locks held by this process on the vp.
233 cleanlocks(vp
, ddi_get_pid(), 0);
234 cleanshares(vp
, ddi_get_pid());
239 /* Decrement the synchronous opens in the znode */
240 if ((flag
& (FSYNC
| FDSYNC
)) && (count
== 1))
241 atomic_dec_32(&zp
->z_sync_cnt
);
243 if (!zfs_has_ctldir(zp
) && zp
->z_zfsvfs
->z_vscan
&&
244 ZTOV(zp
)->v_type
== VREG
&&
245 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0)
246 VERIFY(fs_vscan(vp
, cr
, 1) == 0);
253 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
254 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
257 zfs_holey(vnode_t
*vp
, int cmd
, offset_t
*off
)
259 znode_t
*zp
= VTOZ(vp
);
260 uint64_t noff
= (uint64_t)*off
; /* new offset */
265 file_sz
= zp
->z_size
;
266 if (noff
>= file_sz
) {
267 return (SET_ERROR(ENXIO
));
270 if (cmd
== _FIO_SEEK_HOLE
)
275 error
= dmu_offset_next(zp
->z_zfsvfs
->z_os
, zp
->z_id
, hole
, &noff
);
278 return (SET_ERROR(ENXIO
));
281 * We could find a hole that begins after the logical end-of-file,
282 * because dmu_offset_next() only works on whole blocks. If the
283 * EOF falls mid-block, then indicate that the "virtual hole"
284 * at the end of the file begins at the logical EOF, rather than
285 * at the end of the last block.
287 if (noff
> file_sz
) {
300 zfs_ioctl(vnode_t
*vp
, int com
, intptr_t data
, int flag
, cred_t
*cred
,
301 int *rvalp
, caller_context_t
*ct
)
305 dmu_object_info_t doi
;
313 return (zfs_sync(vp
->v_vfsp
, 0, cred
));
316 * The following two ioctls are used by bfu. Faking out,
317 * necessary to avoid bfu errors.
329 if (ddi_copyin((void *)data
, &off
, sizeof (off
), flag
))
330 return (SET_ERROR(EFAULT
));
333 zfsvfs
= zp
->z_zfsvfs
;
337 /* offset parameter is in/out */
338 error
= zfs_holey(vp
, com
, &off
);
342 if (ddi_copyout(&off
, (void *)data
, sizeof (off
), flag
))
343 return (SET_ERROR(EFAULT
));
346 case _FIO_COUNT_FILLED
:
349 * _FIO_COUNT_FILLED adds a new ioctl command which
350 * exposes the number of filled blocks in a
354 zfsvfs
= zp
->z_zfsvfs
;
359 * Wait for all dirty blocks for this object
360 * to get synced out to disk, and the DMU info
363 error
= dmu_object_wait_synced(zfsvfs
->z_os
, zp
->z_id
);
370 * Retrieve fill count from DMU object.
372 error
= dmu_object_info(zfsvfs
->z_os
, zp
->z_id
, &doi
);
378 ndata
= doi
.doi_fill_count
;
381 if (ddi_copyout(&ndata
, (void *)data
, sizeof (ndata
), flag
))
382 return (SET_ERROR(EFAULT
));
386 return (SET_ERROR(ENOTTY
));
390 * Utility functions to map and unmap a single physical page. These
391 * are used to manage the mappable copies of ZFS file data, and therefore
392 * do not update ref/mod bits.
395 zfs_map_page(page_t
*pp
, enum seg_rw rw
)
398 return (hat_kpm_mapin(pp
, 0));
399 ASSERT(rw
== S_READ
|| rw
== S_WRITE
);
400 return (ppmapin(pp
, PROT_READ
| ((rw
== S_WRITE
) ? PROT_WRITE
: 0),
405 zfs_unmap_page(page_t
*pp
, caddr_t addr
)
408 hat_kpm_mapout(pp
, 0, addr
);
415 * When a file is memory mapped, we must keep the IO data synchronized
416 * between the DMU cache and the memory mapped pages. What this means:
418 * On Write: If we find a memory mapped page, we write to *both*
419 * the page and the dmu buffer.
422 update_pages(vnode_t
*vp
, int64_t start
, int len
, objset_t
*os
, uint64_t oid
)
426 off
= start
& PAGEOFFSET
;
427 for (start
&= PAGEMASK
; len
> 0; start
+= PAGESIZE
) {
429 uint64_t nbytes
= MIN(PAGESIZE
- off
, len
);
431 if (pp
= page_lookup(&vp
->v_object
, start
, SE_SHARED
)) {
434 va
= zfs_map_page(pp
, S_WRITE
);
435 (void) dmu_read(os
, oid
, start
+off
, nbytes
, va
+off
,
437 zfs_unmap_page(pp
, va
);
446 * When a file is memory mapped, we must keep the IO data synchronized
447 * between the DMU cache and the memory mapped pages. What this means:
449 * On Read: We "read" preferentially from memory mapped pages,
450 * else we default from the dmu buffer.
452 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
453 * the file is memory mapped.
456 mappedread(vnode_t
*vp
, int nbytes
, uio_t
*uio
)
458 znode_t
*zp
= VTOZ(vp
);
463 start
= uio
->uio_loffset
;
464 off
= start
& PAGEOFFSET
;
465 for (start
&= PAGEMASK
; len
> 0; start
+= PAGESIZE
) {
467 uint64_t bytes
= MIN(PAGESIZE
- off
, len
);
469 if (pp
= page_lookup(&vp
->v_object
, start
, SE_SHARED
)) {
472 va
= zfs_map_page(pp
, S_READ
);
473 error
= uiomove(va
+ off
, bytes
, UIO_READ
, uio
);
474 zfs_unmap_page(pp
, va
);
477 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
488 offset_t zfs_read_chunk_size
= 1024 * 1024; /* Tunable */
491 * Read bytes from specified file into supplied buffer.
493 * IN: vp - vnode of file to be read from.
494 * uio - structure supplying read location, range info,
496 * ioflag - SYNC flags; used to provide FRSYNC semantics.
497 * cr - credentials of caller.
498 * ct - caller context
500 * OUT: uio - updated offset and range, buffer filled.
502 * RETURN: 0 on success, error code on failure.
505 * vp - atime updated if byte count > 0
509 zfs_read(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cr
, caller_context_t
*ct
)
511 znode_t
*zp
= VTOZ(vp
);
512 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
521 if (zp
->z_pflags
& ZFS_AV_QUARANTINED
) {
523 return (SET_ERROR(EACCES
));
527 * Validate file offset
529 if (uio
->uio_loffset
< 0) {
531 return (SET_ERROR(EINVAL
));
535 * Fasttrack empty reads
537 if (uio
->uio_resid
== 0) {
543 * Check for mandatory locks
545 if (MANDMODE(zp
->z_mode
)) {
546 if (error
= chklock(vp
, FREAD
,
547 uio
->uio_loffset
, uio
->uio_resid
, uio
->uio_fmode
, ct
)) {
554 * If we're in FRSYNC mode, sync out this znode before reading it.
556 if (ioflag
& FRSYNC
|| zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
557 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
560 * Lock the range against changes.
562 rl
= zfs_range_lock(zp
, uio
->uio_loffset
, uio
->uio_resid
, RL_READER
);
565 * If we are reading past end-of-file we can skip
566 * to the end; but we might still need to set atime.
568 if (uio
->uio_loffset
>= zp
->z_size
) {
573 ASSERT(uio
->uio_loffset
< zp
->z_size
);
574 n
= MIN(uio
->uio_resid
, zp
->z_size
- uio
->uio_loffset
);
576 if ((uio
->uio_extflg
== UIO_XUIO
) &&
577 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
)) {
579 int blksz
= zp
->z_blksz
;
580 uint64_t offset
= uio
->uio_loffset
;
582 xuio
= (xuio_t
*)uio
;
584 nblk
= (P2ROUNDUP(offset
+ n
, blksz
) - P2ALIGN(offset
,
587 ASSERT(offset
+ n
<= blksz
);
590 (void) dmu_xuio_init(xuio
, nblk
);
592 if (vn_has_cached_data(vp
)) {
594 * For simplicity, we always allocate a full buffer
595 * even if we only expect to read a portion of a block.
597 while (--nblk
>= 0) {
598 (void) dmu_xuio_add(xuio
,
599 dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
606 nbytes
= MIN(n
, zfs_read_chunk_size
-
607 P2PHASE(uio
->uio_loffset
, zfs_read_chunk_size
));
609 if (vn_has_cached_data(vp
)) {
610 error
= mappedread(vp
, nbytes
, uio
);
612 error
= dmu_read_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
616 /* convert checksum errors into IO errors */
618 error
= SET_ERROR(EIO
);
625 zfs_range_unlock(rl
);
627 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
633 * Write the bytes to a file.
635 * IN: vp - vnode of file to be written to.
636 * uio - structure supplying write location, range info,
638 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
639 * set if in append mode.
640 * cr - credentials of caller.
641 * ct - caller context (NFS/CIFS fem monitor only)
643 * OUT: uio - updated offset and range.
645 * RETURN: 0 on success, error code on failure.
648 * vp - ctime|mtime updated if byte count > 0
653 zfs_write(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cr
, caller_context_t
*ct
)
655 znode_t
*zp
= VTOZ(vp
);
656 rlim64_t limit
= uio
->uio_llimit
;
657 ssize_t start_resid
= uio
->uio_resid
;
661 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
666 int max_blksz
= zfsvfs
->z_max_blksz
;
669 iovec_t
*aiov
= NULL
;
672 int iovcnt
= uio
->uio_iovcnt
;
673 iovec_t
*iovp
= uio
->uio_iov
;
676 sa_bulk_attr_t bulk
[4];
677 uint64_t mtime
[2], ctime
[2];
680 * Fasttrack empty write
686 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
692 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
693 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
694 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
), NULL
,
696 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
700 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
701 * callers might not be able to detect properly that we are read-only,
702 * so check it explicitly here.
704 if (zfsvfs
->z_vfs
->vfs_flag
& VFS_RDONLY
) {
706 return (SET_ERROR(EROFS
));
710 * If immutable or not appending then return EPERM.
711 * Intentionally allow ZFS_READONLY through here.
712 * See zfs_zaccess_common()
714 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) ||
715 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
716 (uio
->uio_loffset
< zp
->z_size
))) {
718 return (SET_ERROR(EPERM
));
721 zilog
= zfsvfs
->z_log
;
724 * Validate file offset
726 woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
729 return (SET_ERROR(EINVAL
));
733 * Check for mandatory locks before calling zfs_range_lock()
734 * in order to prevent a deadlock with locks set via fcntl().
736 if (MANDMODE((mode_t
)zp
->z_mode
) &&
737 (error
= chklock(vp
, FWRITE
, woff
, n
, uio
->uio_fmode
, ct
)) != 0) {
743 * Pre-fault the pages to ensure slow (eg NFS) pages
745 * Skip this if uio contains loaned arc_buf.
747 if ((uio
->uio_extflg
== UIO_XUIO
) &&
748 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
))
749 xuio
= (xuio_t
*)uio
;
751 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
754 * If in append mode, set the io offset pointer to eof.
756 if (ioflag
& FAPPEND
) {
758 * Obtain an appending range lock to guarantee file append
759 * semantics. We reset the write offset once we have the lock.
761 rl
= zfs_range_lock(zp
, 0, n
, RL_APPEND
);
763 if (rl
->r_len
== UINT64_MAX
) {
765 * We overlocked the file because this write will cause
766 * the file block size to increase.
767 * Note that zp_size cannot change with this lock held.
771 uio
->uio_loffset
= woff
;
774 * Note that if the file block size will change as a result of
775 * this write, then this range lock will lock the entire file
776 * so that we can re-write the block safely.
778 rl
= zfs_range_lock(zp
, woff
, n
, RL_WRITER
);
782 zfs_range_unlock(rl
);
784 return (SET_ERROR(EFBIG
));
787 if ((woff
+ n
) > limit
|| woff
> (limit
- n
))
790 /* Will this write extend the file length? */
791 write_eof
= (woff
+ n
> zp
->z_size
);
793 end_size
= MAX(zp
->z_size
, woff
+ n
);
796 * Write the file in reasonable size chunks. Each chunk is written
797 * in a separate transaction; this keeps the intent log records small
798 * and allows us to do more fine-grained space accounting.
802 woff
= uio
->uio_loffset
;
803 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
804 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
806 dmu_return_arcbuf(abuf
);
807 error
= SET_ERROR(EDQUOT
);
811 if (xuio
&& abuf
== NULL
) {
812 ASSERT(i_iov
< iovcnt
);
814 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
815 dmu_xuio_clear(xuio
, i_iov
);
816 DTRACE_PROBE3(zfs_cp_write
, int, i_iov
,
817 iovec_t
*, aiov
, arc_buf_t
*, abuf
);
818 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
819 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
820 aiov
->iov_len
== arc_buf_size(abuf
)));
822 } else if (abuf
== NULL
&& n
>= max_blksz
&&
823 woff
>= zp
->z_size
&&
824 P2PHASE(woff
, max_blksz
) == 0 &&
825 zp
->z_blksz
== max_blksz
) {
827 * This write covers a full block. "Borrow" a buffer
828 * from the dmu so that we can fill it before we enter
829 * a transaction. This avoids the possibility of
830 * holding up the transaction if the data copy hangs
831 * up on a pagefault (e.g., from an NFS server mapping).
835 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
837 ASSERT(abuf
!= NULL
);
838 ASSERT(arc_buf_size(abuf
) == max_blksz
);
839 if (error
= uiocopy(abuf
->b_data
, max_blksz
,
840 UIO_WRITE
, uio
, &cbytes
)) {
841 dmu_return_arcbuf(abuf
);
844 ASSERT(cbytes
== max_blksz
);
848 * Start a transaction.
850 tx
= dmu_tx_create(zfsvfs
->z_os
);
851 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
852 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
853 zfs_sa_upgrade_txholds(tx
, zp
);
854 error
= dmu_tx_assign(tx
, TXG_WAIT
);
858 dmu_return_arcbuf(abuf
);
863 * If zfs_range_lock() over-locked we grow the blocksize
864 * and then reduce the lock range. This will only happen
865 * on the first iteration since zfs_range_reduce() will
866 * shrink down r_len to the appropriate size.
868 if (rl
->r_len
== UINT64_MAX
) {
871 if (zp
->z_blksz
> max_blksz
) {
873 * File's blocksize is already larger than the
874 * "recordsize" property. Only let it grow to
875 * the next power of 2.
877 ASSERT(!ISP2(zp
->z_blksz
));
878 new_blksz
= MIN(end_size
,
879 1 << highbit64(zp
->z_blksz
));
881 new_blksz
= MIN(end_size
, max_blksz
);
883 zfs_grow_blocksize(zp
, new_blksz
, tx
);
884 zfs_range_reduce(rl
, woff
, n
);
888 * XXX - should we really limit each write to z_max_blksz?
889 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
891 nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
894 tx_bytes
= uio
->uio_resid
;
895 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
897 tx_bytes
-= uio
->uio_resid
;
900 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
902 * If this is not a full block write, but we are
903 * extending the file past EOF and this data starts
904 * block-aligned, use assign_arcbuf(). Otherwise,
905 * write via dmu_write().
907 if (tx_bytes
< max_blksz
&& (!write_eof
||
908 aiov
->iov_base
!= abuf
->b_data
)) {
910 dmu_write(zfsvfs
->z_os
, zp
->z_id
, woff
,
911 aiov
->iov_len
, aiov
->iov_base
, tx
);
912 dmu_return_arcbuf(abuf
);
913 xuio_stat_wbuf_copied();
915 ASSERT(xuio
|| tx_bytes
== max_blksz
);
916 dmu_assign_arcbuf(sa_get_db(zp
->z_sa_hdl
),
919 ASSERT(tx_bytes
<= uio
->uio_resid
);
920 uioskip(uio
, tx_bytes
);
922 if (tx_bytes
&& vn_has_cached_data(vp
)) {
923 update_pages(vp
, woff
,
924 tx_bytes
, zfsvfs
->z_os
, zp
->z_id
);
928 * If we made no progress, we're done. If we made even
929 * partial progress, update the znode and ZIL accordingly.
932 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
933 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
940 * Clear Set-UID/Set-GID bits on successful write if not
941 * privileged and at least one of the excute bits is set.
943 * It would be nice to to this after all writes have
944 * been done, but that would still expose the ISUID/ISGID
945 * to another app after the partial write is committed.
947 * Note: we don't call zfs_fuid_map_id() here because
948 * user 0 is not an ephemeral uid.
950 mutex_enter(&zp
->z_acl_lock
);
951 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
952 (S_IXUSR
>> 6))) != 0 &&
953 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
954 secpolicy_vnode_setid_retain(cr
,
955 (zp
->z_mode
& S_ISUID
) != 0 && zp
->z_uid
== 0) != 0) {
957 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
958 newmode
= zp
->z_mode
;
959 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zfsvfs
),
960 (void *)&newmode
, sizeof (uint64_t), tx
);
962 mutex_exit(&zp
->z_acl_lock
);
964 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
968 * Update the file size (zp_size) if it has changed;
969 * account for possible concurrent updates.
971 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
972 (void) atomic_cas_64(&zp
->z_size
, end_size
,
977 * If we are replaying and eof is non zero then force
978 * the file size to the specified eof. Note, there's no
979 * concurrency during replay.
981 if (zfsvfs
->z_replay
&& zfsvfs
->z_replay_eof
!= 0)
982 zp
->z_size
= zfsvfs
->z_replay_eof
;
984 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
986 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
);
991 ASSERT(tx_bytes
== nbytes
);
995 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
998 zfs_range_unlock(rl
);
1001 * If we're in replay mode, or we made no progress, return error.
1002 * Otherwise, it's at least a partial write, so it's successful.
1004 if (zfsvfs
->z_replay
|| uio
->uio_resid
== start_resid
) {
1009 if (ioflag
& (FSYNC
| FDSYNC
) ||
1010 zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1011 zil_commit(zilog
, zp
->z_id
);
1018 zfs_get_done(zgd_t
*zgd
, int error
)
1020 znode_t
*zp
= zgd
->zgd_private
;
1021 objset_t
*os
= zp
->z_zfsvfs
->z_os
;
1024 dmu_buf_rele(zgd
->zgd_db
, zgd
);
1026 zfs_range_unlock(zgd
->zgd_rl
);
1029 * Release the vnode asynchronously as we currently have the
1030 * txg stopped from syncing.
1032 VN_RELE_ASYNC(ZTOV(zp
), dsl_pool_vnrele_taskq(dmu_objset_pool(os
)));
1034 if (error
== 0 && zgd
->zgd_bp
)
1035 zil_lwb_add_block(zgd
->zgd_lwb
, zgd
->zgd_bp
);
1037 kmem_free(zgd
, sizeof (zgd_t
));
1041 static int zil_fault_io
= 0;
1045 * Get data to generate a TX_WRITE intent log record.
1048 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, struct lwb
*lwb
, zio_t
*zio
)
1050 zfsvfs_t
*zfsvfs
= arg
;
1051 objset_t
*os
= zfsvfs
->z_os
;
1053 uint64_t object
= lr
->lr_foid
;
1054 uint64_t offset
= lr
->lr_offset
;
1055 uint64_t size
= lr
->lr_length
;
1060 ASSERT3P(lwb
, !=, NULL
);
1061 ASSERT3P(zio
, !=, NULL
);
1062 ASSERT3U(size
, !=, 0);
1065 * Nothing to do if the file has been removed
1067 if (zfs_zget(zfsvfs
, object
, &zp
) != 0)
1068 return (SET_ERROR(ENOENT
));
1069 if (zp
->z_unlinked
) {
1071 * Release the vnode asynchronously as we currently have the
1072 * txg stopped from syncing.
1074 VN_RELE_ASYNC(ZTOV(zp
),
1075 dsl_pool_vnrele_taskq(dmu_objset_pool(os
)));
1076 return (SET_ERROR(ENOENT
));
1079 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1081 zgd
->zgd_private
= zp
;
1084 * Write records come in two flavors: immediate and indirect.
1085 * For small writes it's cheaper to store the data with the
1086 * log record (immediate); for large writes it's cheaper to
1087 * sync the data and get a pointer to it (indirect) so that
1088 * we don't have to write the data twice.
1090 if (buf
!= NULL
) { /* immediate write */
1091 zgd
->zgd_rl
= zfs_range_lock(zp
, offset
, size
, RL_READER
);
1092 /* test for truncation needs to be done while range locked */
1093 if (offset
>= zp
->z_size
) {
1094 error
= SET_ERROR(ENOENT
);
1096 error
= dmu_read(os
, object
, offset
, size
, buf
,
1097 DMU_READ_NO_PREFETCH
);
1099 ASSERT(error
== 0 || error
== ENOENT
);
1100 } else { /* indirect write */
1102 * Have to lock the whole block to ensure when it's
1103 * written out and its checksum is being calculated
1104 * that no one can change the data. We need to re-check
1105 * blocksize after we get the lock in case it's changed!
1110 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1112 zgd
->zgd_rl
= zfs_range_lock(zp
, offset
, size
,
1114 if (zp
->z_blksz
== size
)
1117 zfs_range_unlock(zgd
->zgd_rl
);
1119 /* test for truncation needs to be done while range locked */
1120 if (lr
->lr_offset
>= zp
->z_size
)
1121 error
= SET_ERROR(ENOENT
);
1124 error
= SET_ERROR(EIO
);
1129 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1130 DMU_READ_NO_PREFETCH
);
1133 blkptr_t
*bp
= &lr
->lr_blkptr
;
1138 ASSERT(db
->db_offset
== offset
);
1139 ASSERT(db
->db_size
== size
);
1141 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1143 ASSERT(error
|| lr
->lr_length
<= size
);
1146 * On success, we need to wait for the write I/O
1147 * initiated by dmu_sync() to complete before we can
1148 * release this dbuf. We will finish everything up
1149 * in the zfs_get_done() callback.
1154 if (error
== EALREADY
) {
1155 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
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
& AT_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
& AT_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
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1536 zfs_dirent_unlock(dl
);
1537 if (error
== ERESTART
) {
1543 zfs_acl_ids_free(&acl_ids
);
1548 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1551 zfs_fuid_sync(zfsvfs
, tx
);
1553 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1554 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1555 if (flag
& FIGNORECASE
)
1557 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1558 vsecp
, acl_ids
.z_fuidp
, vap
);
1559 zfs_acl_ids_free(&acl_ids
);
1562 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1565 zfs_acl_ids_free(&acl_ids
);
1569 * A directory entry already exists for this name.
1572 * Can't truncate an existing file if in exclusive mode.
1575 error
= SET_ERROR(EEXIST
);
1579 * Can't open a directory for writing.
1581 if ((ZTOV(zp
)->v_type
== VDIR
) && (mode
& S_IWRITE
)) {
1582 error
= SET_ERROR(EISDIR
);
1586 * Verify requested access to file.
1588 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1592 mutex_enter(&dzp
->z_lock
);
1594 mutex_exit(&dzp
->z_lock
);
1597 * Truncate regular files if requested.
1599 if ((ZTOV(zp
)->v_type
== VREG
) &&
1600 (vap
->va_mask
& AT_SIZE
) && (vap
->va_size
== 0)) {
1601 /* we can't hold any locks when calling zfs_freesp() */
1602 zfs_dirent_unlock(dl
);
1604 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1606 vnevent_create(ZTOV(zp
), ct
);
1613 zfs_dirent_unlock(dl
);
1620 error
= specvp_check(vpp
, cr
);
1623 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1624 zil_commit(zilog
, 0);
1631 * Remove an entry from a directory.
1633 * IN: dvp - vnode of directory to remove entry from.
1634 * name - name of entry to remove.
1635 * cr - credentials of caller.
1636 * ct - caller context
1637 * flags - case flags
1639 * RETURN: 0 on success, error code on failure.
1643 * vp - ctime (if nlink > 0)
1646 uint64_t null_xattr
= 0;
1650 zfs_remove(vnode_t
*dvp
, char *name
, cred_t
*cr
, caller_context_t
*ct
,
1653 znode_t
*zp
, *dzp
= VTOZ(dvp
);
1656 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1658 uint64_t acl_obj
, xattr_obj
;
1659 uint64_t xattr_obj_unlinked
= 0;
1663 boolean_t may_delete_now
, delete_now
= FALSE
;
1664 boolean_t unlinked
, toobig
= FALSE
;
1666 pathname_t
*realnmp
= NULL
;
1670 boolean_t waited
= B_FALSE
;
1674 zilog
= zfsvfs
->z_log
;
1676 if (flags
& FIGNORECASE
) {
1686 * Attempt to lock directory; fail if entry doesn't exist.
1688 if (error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1698 if (error
= zfs_zaccess_delete(dzp
, zp
, cr
)) {
1703 * Need to use rmdir for removing directories.
1705 if (vp
->v_type
== VDIR
) {
1706 error
= SET_ERROR(EPERM
);
1710 vnevent_remove(vp
, dvp
, name
, ct
);
1713 dnlc_remove(dvp
, realnmp
->pn_buf
);
1715 dnlc_remove(dvp
, name
);
1717 mutex_enter(&vp
->v_lock
);
1718 may_delete_now
= vp
->v_count
== 1 && !vn_has_cached_data(vp
);
1719 mutex_exit(&vp
->v_lock
);
1722 * We may delete the znode now, or we may put it in the unlinked set;
1723 * it depends on whether we're the last link, and on whether there are
1724 * other holds on the vnode. So we dmu_tx_hold() the right things to
1725 * allow for either case.
1728 tx
= dmu_tx_create(zfsvfs
->z_os
);
1729 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1730 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1731 zfs_sa_upgrade_txholds(tx
, zp
);
1732 zfs_sa_upgrade_txholds(tx
, dzp
);
1733 if (may_delete_now
) {
1735 zp
->z_size
> zp
->z_blksz
* DMU_MAX_DELETEBLKCNT
;
1736 /* if the file is too big, only hold_free a token amount */
1737 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1738 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1741 /* are there any extended attributes? */
1742 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1743 &xattr_obj
, sizeof (xattr_obj
));
1744 if (error
== 0 && xattr_obj
) {
1745 error
= zfs_zget(zfsvfs
, xattr_obj
, &xzp
);
1747 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1748 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1751 mutex_enter(&zp
->z_lock
);
1752 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1753 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1754 mutex_exit(&zp
->z_lock
);
1756 /* charge as an update -- would be nice not to charge at all */
1757 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1760 * Mark this transaction as typically resulting in a net free of space
1762 dmu_tx_mark_netfree(tx
);
1764 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
1766 zfs_dirent_unlock(dl
);
1770 if (error
== ERESTART
) {
1784 * Remove the directory entry.
1786 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1795 * Hold z_lock so that we can make sure that the ACL obj
1796 * hasn't changed. Could have been deleted due to
1799 mutex_enter(&zp
->z_lock
);
1800 mutex_enter(&vp
->v_lock
);
1801 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1802 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1803 delete_now
= may_delete_now
&& !toobig
&&
1804 vp
->v_count
== 1 && !vn_has_cached_data(vp
) &&
1805 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1807 mutex_exit(&vp
->v_lock
);
1811 if (xattr_obj_unlinked
) {
1812 ASSERT3U(xzp
->z_links
, ==, 2);
1813 mutex_enter(&xzp
->z_lock
);
1814 xzp
->z_unlinked
= 1;
1816 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zfsvfs
),
1817 &xzp
->z_links
, sizeof (xzp
->z_links
), tx
);
1818 ASSERT3U(error
, ==, 0);
1819 mutex_exit(&xzp
->z_lock
);
1820 zfs_unlinked_add(xzp
, tx
);
1823 error
= sa_remove(zp
->z_sa_hdl
,
1824 SA_ZPL_XATTR(zfsvfs
), tx
);
1826 error
= sa_update(zp
->z_sa_hdl
,
1827 SA_ZPL_XATTR(zfsvfs
), &null_xattr
,
1828 sizeof (uint64_t), tx
);
1831 mutex_enter(&vp
->v_lock
);
1833 ASSERT0(vp
->v_count
);
1834 mutex_exit(&vp
->v_lock
);
1835 mutex_exit(&zp
->z_lock
);
1836 zfs_znode_delete(zp
, tx
);
1837 } else if (unlinked
) {
1838 mutex_exit(&zp
->z_lock
);
1839 zfs_unlinked_add(zp
, tx
);
1843 if (flags
& FIGNORECASE
)
1845 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1852 zfs_dirent_unlock(dl
);
1859 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1860 zil_commit(zilog
, 0);
1867 * Create a new directory and insert it into dvp using the name
1868 * provided. Return a pointer to the inserted directory.
1870 * IN: dvp - vnode of directory to add subdir to.
1871 * dirname - name of new directory.
1872 * vap - attributes of new directory.
1873 * cr - credentials of caller.
1874 * ct - caller context
1875 * flags - case flags
1876 * vsecp - ACL to be set
1878 * OUT: vpp - vnode of created directory.
1880 * RETURN: 0 on success, error code on failure.
1883 * dvp - ctime|mtime updated
1884 * vp - ctime|mtime|atime updated
1888 zfs_mkdir(vnode_t
*dvp
, char *dirname
, vattr_t
*vap
, vnode_t
**vpp
, cred_t
*cr
,
1889 caller_context_t
*ct
, int flags
, vsecattr_t
*vsecp
)
1891 znode_t
*zp
, *dzp
= VTOZ(dvp
);
1892 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1901 gid_t gid
= crgetgid(cr
);
1902 zfs_acl_ids_t acl_ids
;
1903 boolean_t fuid_dirtied
;
1904 boolean_t waited
= B_FALSE
;
1906 ASSERT(vap
->va_type
== VDIR
);
1909 * If we have an ephemeral id, ACL, or XVATTR then
1910 * make sure file system is at proper version
1913 ksid
= crgetsid(cr
, KSID_OWNER
);
1915 uid
= ksid_getid(ksid
);
1918 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1919 (vsecp
|| (vap
->va_mask
& AT_XVATTR
) ||
1920 IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1921 return (SET_ERROR(EINVAL
));
1925 zilog
= zfsvfs
->z_log
;
1927 if (dzp
->z_pflags
& ZFS_XATTR
) {
1929 return (SET_ERROR(EINVAL
));
1932 if (zfsvfs
->z_utf8
&& u8_validate(dirname
,
1933 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1935 return (SET_ERROR(EILSEQ
));
1937 if (flags
& FIGNORECASE
)
1940 if (vap
->va_mask
& AT_XVATTR
) {
1941 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1942 crgetuid(cr
), cr
, vap
->va_type
)) != 0) {
1948 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1949 vsecp
, &acl_ids
)) != 0) {
1954 * First make sure the new directory doesn't exist.
1956 * Existence is checked first to make sure we don't return
1957 * EACCES instead of EEXIST which can cause some applications
1963 if (error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1965 zfs_acl_ids_free(&acl_ids
);
1970 if (error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
)) {
1971 zfs_acl_ids_free(&acl_ids
);
1972 zfs_dirent_unlock(dl
);
1977 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1978 zfs_acl_ids_free(&acl_ids
);
1979 zfs_dirent_unlock(dl
);
1981 return (SET_ERROR(EDQUOT
));
1985 * Add a new entry to the directory.
1987 tx
= dmu_tx_create(zfsvfs
->z_os
);
1988 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
1989 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1990 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1992 zfs_fuid_txhold(zfsvfs
, tx
);
1993 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1994 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
1995 acl_ids
.z_aclp
->z_acl_bytes
);
1998 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1999 ZFS_SA_BASE_ATTR_SIZE
);
2001 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
2003 zfs_dirent_unlock(dl
);
2004 if (error
== ERESTART
) {
2010 zfs_acl_ids_free(&acl_ids
);
2019 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
2022 zfs_fuid_sync(zfsvfs
, tx
);
2025 * Now put new name in parent dir.
2027 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
2031 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
2032 if (flags
& FIGNORECASE
)
2034 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
2035 acl_ids
.z_fuidp
, vap
);
2037 zfs_acl_ids_free(&acl_ids
);
2041 zfs_dirent_unlock(dl
);
2043 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2044 zil_commit(zilog
, 0);
2051 * Remove a directory subdir entry. If the current working
2052 * directory is the same as the subdir to be removed, the
2055 * IN: dvp - vnode of directory to remove from.
2056 * name - name of directory to be removed.
2057 * cwd - vnode of current working directory.
2058 * cr - credentials of caller.
2059 * ct - caller context
2060 * flags - case flags
2062 * RETURN: 0 on success, error code on failure.
2065 * dvp - ctime|mtime updated
2069 zfs_rmdir(vnode_t
*dvp
, char *name
, vnode_t
*cwd
, cred_t
*cr
,
2070 caller_context_t
*ct
, int flags
)
2072 znode_t
*dzp
= VTOZ(dvp
);
2075 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
2081 boolean_t waited
= B_FALSE
;
2085 zilog
= zfsvfs
->z_log
;
2087 if (flags
& FIGNORECASE
)
2093 * Attempt to lock directory; fail if entry doesn't exist.
2095 if (error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
2103 if (error
= zfs_zaccess_delete(dzp
, zp
, cr
)) {
2107 if (vp
->v_type
!= VDIR
) {
2108 error
= SET_ERROR(ENOTDIR
);
2113 error
= SET_ERROR(EINVAL
);
2117 vnevent_rmdir(vp
, dvp
, name
, ct
);
2120 * Grab a lock on the directory to make sure that noone is
2121 * trying to add (or lookup) entries while we are removing it.
2123 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
2126 * Grab a lock on the parent pointer to make sure we play well
2127 * with the treewalk and directory rename code.
2129 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
2131 tx
= dmu_tx_create(zfsvfs
->z_os
);
2132 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2133 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2134 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
2135 zfs_sa_upgrade_txholds(tx
, zp
);
2136 zfs_sa_upgrade_txholds(tx
, dzp
);
2137 dmu_tx_mark_netfree(tx
);
2138 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
2140 rw_exit(&zp
->z_parent_lock
);
2141 rw_exit(&zp
->z_name_lock
);
2142 zfs_dirent_unlock(dl
);
2144 if (error
== ERESTART
) {
2155 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2158 uint64_t txtype
= TX_RMDIR
;
2159 if (flags
& FIGNORECASE
)
2161 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2166 rw_exit(&zp
->z_parent_lock
);
2167 rw_exit(&zp
->z_name_lock
);
2169 zfs_dirent_unlock(dl
);
2173 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2174 zil_commit(zilog
, 0);
2181 * Read as many directory entries as will fit into the provided
2182 * buffer from the given directory cursor position (specified in
2183 * the uio structure).
2185 * IN: vp - vnode of directory to read.
2186 * uio - structure supplying read location, range info,
2187 * and return buffer.
2188 * cr - credentials of caller.
2189 * ct - caller context
2190 * flags - case flags
2192 * OUT: uio - updated offset and range, buffer filled.
2193 * eofp - set to true if end-of-file detected.
2195 * RETURN: 0 on success, error code on failure.
2198 * vp - atime updated
2200 * Note that the low 4 bits of the cookie returned by zap is always zero.
2201 * This allows us to use the low range for "special" directory entries:
2202 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2203 * we use the offset 2 for the '.zfs' directory.
2207 zfs_readdir(vnode_t
*vp
, uio_t
*uio
, cred_t
*cr
, int *eofp
,
2208 caller_context_t
*ct
, int flags
)
2210 znode_t
*zp
= VTOZ(vp
);
2214 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2219 zap_attribute_t zap
;
2220 uint_t bytes_wanted
;
2221 uint64_t offset
; /* must be unsigned; checks for < 1 */
2227 boolean_t check_sysattrs
;
2232 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
2233 &parent
, sizeof (parent
))) != 0) {
2239 * If we are not given an eof variable,
2246 * Check for valid iov_len.
2248 if (uio
->uio_iov
->iov_len
<= 0) {
2250 return (SET_ERROR(EINVAL
));
2254 * Quit if directory has been removed (posix)
2256 if ((*eofp
= zp
->z_unlinked
) != 0) {
2263 offset
= uio
->uio_loffset
;
2264 prefetch
= zp
->z_zn_prefetch
;
2267 * Initialize the iterator cursor.
2271 * Start iteration from the beginning of the directory.
2273 zap_cursor_init(&zc
, os
, zp
->z_id
);
2276 * The offset is a serialized cursor.
2278 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2282 * Get space to change directory entries into fs independent format.
2284 iovp
= uio
->uio_iov
;
2285 bytes_wanted
= iovp
->iov_len
;
2286 if (uio
->uio_segflg
!= UIO_SYSSPACE
|| uio
->uio_iovcnt
!= 1) {
2287 bufsize
= bytes_wanted
;
2288 outbuf
= kmem_alloc(bufsize
, KM_SLEEP
);
2289 odp
= (struct dirent64
*)outbuf
;
2291 bufsize
= bytes_wanted
;
2293 odp
= (struct dirent64
*)iovp
->iov_base
;
2295 eodp
= (struct edirent
*)odp
;
2298 * If this VFS supports the system attribute view interface; and
2299 * we're looking at an extended attribute directory; and we care
2300 * about normalization conflicts on this vfs; then we must check
2301 * for normalization conflicts with the sysattr name space.
2303 check_sysattrs
= vfs_has_feature(vp
->v_vfsp
, VFSFT_SYSATTR_VIEWS
) &&
2304 (vp
->v_flag
& V_XATTRDIR
) && zfsvfs
->z_norm
&&
2305 (flags
& V_RDDIR_ENTFLAGS
);
2308 * Transform to file-system independent format
2311 while (outcount
< bytes_wanted
) {
2314 off64_t
*next
= NULL
;
2317 * Special case `.', `..', and `.zfs'.
2320 (void) strcpy(zap
.za_name
, ".");
2321 zap
.za_normalization_conflict
= 0;
2323 } else if (offset
== 1) {
2324 (void) strcpy(zap
.za_name
, "..");
2325 zap
.za_normalization_conflict
= 0;
2327 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2328 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2329 zap
.za_normalization_conflict
= 0;
2330 objnum
= ZFSCTL_INO_ROOT
;
2335 if (error
= zap_cursor_retrieve(&zc
, &zap
)) {
2336 if ((*eofp
= (error
== ENOENT
)) != 0)
2342 if (zap
.za_integer_length
!= 8 ||
2343 zap
.za_num_integers
!= 1) {
2344 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2345 "entry, obj = %lld, offset = %lld\n",
2346 (u_longlong_t
)zp
->z_id
,
2347 (u_longlong_t
)offset
);
2348 error
= SET_ERROR(ENXIO
);
2352 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2354 * MacOS X can extract the object type here such as:
2355 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2358 if (check_sysattrs
&& !zap
.za_normalization_conflict
) {
2359 zap
.za_normalization_conflict
=
2360 xattr_sysattr_casechk(zap
.za_name
);
2364 if (flags
& V_RDDIR_ACCFILTER
) {
2366 * If we have no access at all, don't include
2367 * this entry in the returned information
2370 if (zfs_zget(zp
->z_zfsvfs
, objnum
, &ezp
) != 0)
2372 if (!zfs_has_access(ezp
, cr
)) {
2379 if (flags
& V_RDDIR_ENTFLAGS
)
2380 reclen
= EDIRENT_RECLEN(strlen(zap
.za_name
));
2382 reclen
= DIRENT64_RECLEN(strlen(zap
.za_name
));
2385 * Will this entry fit in the buffer?
2387 if (outcount
+ reclen
> bufsize
) {
2389 * Did we manage to fit anything in the buffer?
2392 error
= SET_ERROR(EINVAL
);
2397 if (flags
& V_RDDIR_ENTFLAGS
) {
2399 * Add extended flag entry:
2401 eodp
->ed_ino
= objnum
;
2402 eodp
->ed_reclen
= reclen
;
2403 /* NOTE: ed_off is the offset for the *next* entry */
2404 next
= &(eodp
->ed_off
);
2405 eodp
->ed_eflags
= zap
.za_normalization_conflict
?
2406 ED_CASE_CONFLICT
: 0;
2407 (void) strncpy(eodp
->ed_name
, zap
.za_name
,
2408 EDIRENT_NAMELEN(reclen
));
2409 eodp
= (edirent_t
*)((intptr_t)eodp
+ reclen
);
2414 odp
->d_ino
= objnum
;
2415 odp
->d_reclen
= reclen
;
2416 /* NOTE: d_off is the offset for the *next* entry */
2417 next
= &(odp
->d_off
);
2418 (void) strncpy(odp
->d_name
, zap
.za_name
,
2419 DIRENT64_NAMELEN(reclen
));
2420 odp
= (dirent64_t
*)((intptr_t)odp
+ reclen
);
2424 ASSERT(outcount
<= bufsize
);
2426 /* Prefetch znode */
2428 dmu_prefetch(os
, objnum
, 0, 0, 0,
2429 ZIO_PRIORITY_SYNC_READ
);
2433 * Move to the next entry, fill in the previous offset.
2435 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2436 zap_cursor_advance(&zc
);
2437 offset
= zap_cursor_serialize(&zc
);
2444 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2446 if (uio
->uio_segflg
== UIO_SYSSPACE
&& uio
->uio_iovcnt
== 1) {
2447 iovp
->iov_base
+= outcount
;
2448 iovp
->iov_len
-= outcount
;
2449 uio
->uio_resid
-= outcount
;
2450 } else if (error
= uiomove(outbuf
, (long)outcount
, UIO_READ
, uio
)) {
2452 * Reset the pointer.
2454 offset
= uio
->uio_loffset
;
2458 zap_cursor_fini(&zc
);
2459 if (uio
->uio_segflg
!= UIO_SYSSPACE
|| uio
->uio_iovcnt
!= 1)
2460 kmem_free(outbuf
, bufsize
);
2462 if (error
== ENOENT
)
2465 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
2467 uio
->uio_loffset
= offset
;
2472 ulong_t zfs_fsync_sync_cnt
= 4;
2475 zfs_fsync(vnode_t
*vp
, int syncflag
, cred_t
*cr
, caller_context_t
*ct
)
2477 znode_t
*zp
= VTOZ(vp
);
2478 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2481 * Regardless of whether this is required for standards conformance,
2482 * this is the logical behavior when fsync() is called on a file with
2483 * dirty pages. We use B_ASYNC since the ZIL transactions are already
2484 * going to be pushed out as part of the zil_commit().
2486 if (vn_has_cached_data(vp
) && !(syncflag
& FNODSYNC
) &&
2487 (vp
->v_type
== VREG
) && !(IS_SWAPVP(vp
)))
2488 (void) fop_putpage(vp
, 0, (size_t)0, B_ASYNC
, cr
, ct
);
2490 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2492 if (zfsvfs
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2495 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
2503 * Get the requested file attributes and place them in the provided
2506 * IN: vp - vnode of file.
2507 * vap - va_mask identifies requested attributes.
2508 * If AT_XVATTR set, then optional attrs are requested
2509 * flags - ATTR_NOACLCHECK (CIFS server context)
2510 * cr - credentials of caller.
2511 * ct - caller context
2513 * OUT: vap - attribute values.
2515 * RETURN: 0 (always succeeds).
2519 zfs_getattr(vnode_t
*vp
, vattr_t
*vap
, int flags
, cred_t
*cr
,
2520 caller_context_t
*ct
)
2522 znode_t
*zp
= VTOZ(vp
);
2523 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2526 uint64_t mtime
[2], ctime
[2];
2527 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2528 xoptattr_t
*xoap
= NULL
;
2529 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2530 sa_bulk_attr_t bulk
[2];
2536 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2538 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
2539 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
2541 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2547 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2548 * Also, if we are the owner don't bother, since owner should
2549 * always be allowed to read basic attributes of file.
2551 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2552 (vap
->va_uid
!= crgetuid(cr
))) {
2553 if (error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2561 * Return all attributes. It's cheaper to provide the answer
2562 * than to determine whether we were asked the question.
2565 mutex_enter(&zp
->z_lock
);
2566 vap
->va_type
= vp
->v_type
;
2567 vap
->va_mode
= zp
->z_mode
& MODEMASK
;
2568 vap
->va_fsid
= zp
->z_zfsvfs
->z_vfs
->vfs_dev
;
2569 vap
->va_nodeid
= zp
->z_id
;
2570 if ((vp
->v_flag
& VROOT
) && zfs_show_ctldir(zp
))
2571 links
= zp
->z_links
+ 1;
2573 links
= zp
->z_links
;
2574 vap
->va_nlink
= MIN(links
, UINT32_MAX
); /* nlink_t limit! */
2575 vap
->va_size
= zp
->z_size
;
2576 vap
->va_rdev
= vp
->v_rdev
;
2577 vap
->va_seq
= zp
->z_seq
;
2580 * Add in any requested optional attributes and the create time.
2581 * Also set the corresponding bits in the returned attribute bitmap.
2583 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zfsvfs
->z_use_fuids
) {
2584 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2586 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2587 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2590 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2591 xoap
->xoa_readonly
=
2592 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2593 XVA_SET_RTN(xvap
, XAT_READONLY
);
2596 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2598 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2599 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2602 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2604 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2605 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2608 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2609 xoap
->xoa_nounlink
=
2610 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2611 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2614 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2615 xoap
->xoa_immutable
=
2616 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2617 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2620 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2621 xoap
->xoa_appendonly
=
2622 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2623 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2626 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2628 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2629 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2632 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2634 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2635 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2638 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2639 xoap
->xoa_av_quarantined
=
2640 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2641 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2644 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2645 xoap
->xoa_av_modified
=
2646 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2647 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2650 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2651 vp
->v_type
== VREG
) {
2652 zfs_sa_get_scanstamp(zp
, xvap
);
2655 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2658 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zfsvfs
),
2659 times
, sizeof (times
));
2660 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2661 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2664 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2665 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2666 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2668 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2669 xoap
->xoa_generation
= zp
->z_gen
;
2670 XVA_SET_RTN(xvap
, XAT_GEN
);
2673 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2675 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2676 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2679 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2681 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2682 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2686 ZFS_TIME_DECODE(&vap
->va_atime
, zp
->z_atime
);
2687 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2688 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2690 mutex_exit(&zp
->z_lock
);
2692 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2694 if (zp
->z_blksz
== 0) {
2696 * Block size hasn't been set; suggest maximal I/O transfers.
2698 vap
->va_blksize
= zfsvfs
->z_max_blksz
;
2706 * Set the file attributes to the values contained in the
2709 * IN: vp - vnode of file to be modified.
2710 * vap - new attribute values.
2711 * If AT_XVATTR set, then optional attrs are being set
2712 * flags - ATTR_UTIME set if non-default time values provided.
2713 * - ATTR_NOACLCHECK (CIFS context only).
2714 * cr - credentials of caller.
2715 * ct - caller context
2717 * RETURN: 0 on success, error code on failure.
2720 * vp - ctime updated, mtime updated if size changed.
2724 zfs_setattr(vnode_t
*vp
, vattr_t
*vap
, int flags
, cred_t
*cr
,
2725 caller_context_t
*ct
)
2727 znode_t
*zp
= VTOZ(vp
);
2728 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2733 uint_t mask
= vap
->va_mask
;
2734 uint_t saved_mask
= 0;
2737 uint64_t new_uid
, new_gid
;
2739 uint64_t mtime
[2], ctime
[2];
2741 int need_policy
= FALSE
;
2743 zfs_fuid_info_t
*fuidp
= NULL
;
2744 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2747 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2748 boolean_t fuid_dirtied
= B_FALSE
;
2749 sa_bulk_attr_t bulk
[7], xattr_bulk
[7];
2750 int count
= 0, xattr_count
= 0;
2755 if (mask
& AT_NOSET
)
2756 return (SET_ERROR(EINVAL
));
2761 zilog
= zfsvfs
->z_log
;
2764 * Make sure that if we have ephemeral uid/gid or xvattr specified
2765 * that file system is at proper version level
2768 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
2769 (((mask
& AT_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2770 ((mask
& AT_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2771 (mask
& AT_XVATTR
))) {
2773 return (SET_ERROR(EINVAL
));
2776 if (mask
& AT_SIZE
&& vp
->v_type
== VDIR
) {
2778 return (SET_ERROR(EISDIR
));
2781 if (mask
& AT_SIZE
&& vp
->v_type
!= VREG
&& vp
->v_type
!= VFIFO
) {
2783 return (SET_ERROR(EINVAL
));
2787 * If this is an xvattr_t, then get a pointer to the structure of
2788 * optional attributes. If this is NULL, then we have a vattr_t.
2790 xoap
= xva_getxoptattr(xvap
);
2792 xva_init(&tmpxvattr
);
2795 * Immutable files can only alter immutable bit and atime
2797 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2798 ((mask
& (AT_SIZE
|AT_UID
|AT_GID
|AT_MTIME
|AT_MODE
)) ||
2799 ((mask
& AT_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2801 return (SET_ERROR(EPERM
));
2805 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
2809 * Verify timestamps doesn't overflow 32 bits.
2810 * ZFS can handle large timestamps, but 32bit syscalls can't
2811 * handle times greater than 2039. This check should be removed
2812 * once large timestamps are fully supported.
2814 if (mask
& (AT_ATIME
| AT_MTIME
)) {
2815 if (((mask
& AT_ATIME
) && TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2816 ((mask
& AT_MTIME
) && TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2818 return (SET_ERROR(EOVERFLOW
));
2826 /* Can this be moved to before the top label? */
2827 if (zfsvfs
->z_vfs
->vfs_flag
& VFS_RDONLY
) {
2829 return (SET_ERROR(EROFS
));
2833 * First validate permissions
2836 if (mask
& AT_SIZE
) {
2837 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2843 * XXX - Note, we are not providing any open
2844 * mode flags here (like FNDELAY), so we may
2845 * block if there are locks present... this
2846 * should be addressed in openat().
2848 /* XXX - would it be OK to generate a log record here? */
2849 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2855 if (vap
->va_size
== 0)
2856 vnevent_truncate(ZTOV(zp
), ct
);
2859 if (mask
& (AT_ATIME
|AT_MTIME
) ||
2860 ((mask
& AT_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2861 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2862 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2863 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2864 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2865 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2866 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2867 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2871 if (mask
& (AT_UID
|AT_GID
)) {
2872 int idmask
= (mask
& (AT_UID
|AT_GID
));
2877 * NOTE: even if a new mode is being set,
2878 * we may clear S_ISUID/S_ISGID bits.
2881 if (!(mask
& AT_MODE
))
2882 vap
->va_mode
= zp
->z_mode
;
2885 * Take ownership or chgrp to group we are a member of
2888 take_owner
= (mask
& AT_UID
) && (vap
->va_uid
== crgetuid(cr
));
2889 take_group
= (mask
& AT_GID
) &&
2890 zfs_groupmember(zfsvfs
, vap
->va_gid
, cr
);
2893 * If both AT_UID and AT_GID are set then take_owner and
2894 * take_group must both be set in order to allow taking
2897 * Otherwise, send the check through secpolicy_vnode_setattr()
2901 if (((idmask
== (AT_UID
|AT_GID
)) && take_owner
&& take_group
) ||
2902 ((idmask
== AT_UID
) && take_owner
) ||
2903 ((idmask
== AT_GID
) && take_group
)) {
2904 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2905 skipaclchk
, cr
) == 0) {
2907 * Remove setuid/setgid for non-privileged users
2909 secpolicy_setid_clear(vap
, cr
);
2910 trim_mask
= (mask
& (AT_UID
|AT_GID
));
2919 mutex_enter(&zp
->z_lock
);
2920 oldva
.va_mode
= zp
->z_mode
;
2921 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2922 if (mask
& AT_XVATTR
) {
2924 * Update xvattr mask to include only those attributes
2925 * that are actually changing.
2927 * the bits will be restored prior to actually setting
2928 * the attributes so the caller thinks they were set.
2930 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2931 if (xoap
->xoa_appendonly
!=
2932 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2935 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2936 XVA_SET_REQ(&tmpxvattr
, XAT_APPENDONLY
);
2940 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2941 if (xoap
->xoa_nounlink
!=
2942 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2945 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2946 XVA_SET_REQ(&tmpxvattr
, XAT_NOUNLINK
);
2950 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2951 if (xoap
->xoa_immutable
!=
2952 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2955 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2956 XVA_SET_REQ(&tmpxvattr
, XAT_IMMUTABLE
);
2960 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2961 if (xoap
->xoa_nodump
!=
2962 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2965 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2966 XVA_SET_REQ(&tmpxvattr
, XAT_NODUMP
);
2970 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2971 if (xoap
->xoa_av_modified
!=
2972 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2975 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2976 XVA_SET_REQ(&tmpxvattr
, XAT_AV_MODIFIED
);
2980 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2981 if ((vp
->v_type
!= VREG
&&
2982 xoap
->xoa_av_quarantined
) ||
2983 xoap
->xoa_av_quarantined
!=
2984 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2987 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2988 XVA_SET_REQ(&tmpxvattr
, XAT_AV_QUARANTINED
);
2992 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2993 mutex_exit(&zp
->z_lock
);
2995 return (SET_ERROR(EPERM
));
2998 if (need_policy
== FALSE
&&
2999 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
3000 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
3005 mutex_exit(&zp
->z_lock
);
3007 if (mask
& AT_MODE
) {
3008 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
3009 err
= secpolicy_setid_setsticky_clear(vp
, vap
,
3015 trim_mask
|= AT_MODE
;
3023 * If trim_mask is set then take ownership
3024 * has been granted or write_acl is present and user
3025 * has the ability to modify mode. In that case remove
3026 * UID|GID and or MODE from mask so that
3027 * secpolicy_vnode_setattr() doesn't revoke it.
3031 saved_mask
= vap
->va_mask
;
3032 vap
->va_mask
&= ~trim_mask
;
3034 err
= secpolicy_vnode_setattr(cr
, vp
, vap
, &oldva
, flags
,
3035 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
3042 vap
->va_mask
|= saved_mask
;
3046 * secpolicy_vnode_setattr, or take ownership may have
3049 mask
= vap
->va_mask
;
3051 if ((mask
& (AT_UID
| AT_GID
))) {
3052 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
3053 &xattr_obj
, sizeof (xattr_obj
));
3055 if (err
== 0 && xattr_obj
) {
3056 err
= zfs_zget(zp
->z_zfsvfs
, xattr_obj
, &attrzp
);
3060 if (mask
& AT_UID
) {
3061 new_uid
= zfs_fuid_create(zfsvfs
,
3062 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
3063 if (new_uid
!= zp
->z_uid
&&
3064 zfs_fuid_overquota(zfsvfs
, B_FALSE
, new_uid
)) {
3066 VN_RELE(ZTOV(attrzp
));
3067 err
= SET_ERROR(EDQUOT
);
3072 if (mask
& AT_GID
) {
3073 new_gid
= zfs_fuid_create(zfsvfs
, (uint64_t)vap
->va_gid
,
3074 cr
, ZFS_GROUP
, &fuidp
);
3075 if (new_gid
!= zp
->z_gid
&&
3076 zfs_fuid_overquota(zfsvfs
, B_TRUE
, new_gid
)) {
3078 VN_RELE(ZTOV(attrzp
));
3079 err
= SET_ERROR(EDQUOT
);
3084 tx
= dmu_tx_create(zfsvfs
->z_os
);
3086 if (mask
& AT_MODE
) {
3087 uint64_t pmode
= zp
->z_mode
;
3089 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
3091 if (zp
->z_zfsvfs
->z_acl_mode
== ZFS_ACL_RESTRICTED
&&
3092 !(zp
->z_pflags
& ZFS_ACL_TRIVIAL
)) {
3093 err
= SET_ERROR(EPERM
);
3097 if (err
= zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
))
3100 mutex_enter(&zp
->z_lock
);
3101 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
3103 * Are we upgrading ACL from old V0 format
3106 if (zfsvfs
->z_version
>= ZPL_VERSION_FUID
&&
3107 zfs_znode_acl_version(zp
) ==
3108 ZFS_ACL_VERSION_INITIAL
) {
3109 dmu_tx_hold_free(tx
, acl_obj
, 0,
3111 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3112 0, aclp
->z_acl_bytes
);
3114 dmu_tx_hold_write(tx
, acl_obj
, 0,
3117 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3118 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3119 0, aclp
->z_acl_bytes
);
3121 mutex_exit(&zp
->z_lock
);
3122 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3124 if ((mask
& AT_XVATTR
) &&
3125 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3126 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3128 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3132 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
3135 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3137 zfs_fuid_txhold(zfsvfs
, tx
);
3139 zfs_sa_upgrade_txholds(tx
, zp
);
3141 err
= dmu_tx_assign(tx
, TXG_WAIT
);
3147 * Set each attribute requested.
3148 * We group settings according to the locks they need to acquire.
3150 * Note: you cannot set ctime directly, although it will be
3151 * updated as a side-effect of calling this function.
3155 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3156 mutex_enter(&zp
->z_acl_lock
);
3157 mutex_enter(&zp
->z_lock
);
3159 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
3160 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
3163 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3164 mutex_enter(&attrzp
->z_acl_lock
);
3165 mutex_enter(&attrzp
->z_lock
);
3166 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3167 SA_ZPL_FLAGS(zfsvfs
), NULL
, &attrzp
->z_pflags
,
3168 sizeof (attrzp
->z_pflags
));
3171 if (mask
& (AT_UID
|AT_GID
)) {
3173 if (mask
& AT_UID
) {
3174 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zfsvfs
), NULL
,
3175 &new_uid
, sizeof (new_uid
));
3176 zp
->z_uid
= new_uid
;
3178 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3179 SA_ZPL_UID(zfsvfs
), NULL
, &new_uid
,
3181 attrzp
->z_uid
= new_uid
;
3185 if (mask
& AT_GID
) {
3186 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zfsvfs
),
3187 NULL
, &new_gid
, sizeof (new_gid
));
3188 zp
->z_gid
= new_gid
;
3190 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3191 SA_ZPL_GID(zfsvfs
), NULL
, &new_gid
,
3193 attrzp
->z_gid
= new_gid
;
3196 if (!(mask
& AT_MODE
)) {
3197 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
),
3198 NULL
, &new_mode
, sizeof (new_mode
));
3199 new_mode
= zp
->z_mode
;
3201 err
= zfs_acl_chown_setattr(zp
);
3204 err
= zfs_acl_chown_setattr(attrzp
);
3209 if (mask
& AT_MODE
) {
3210 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
), NULL
,
3211 &new_mode
, sizeof (new_mode
));
3212 zp
->z_mode
= new_mode
;
3213 ASSERT3U((uintptr_t)aclp
, !=, (uintptr_t)NULL
);
3214 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
3216 if (zp
->z_acl_cached
)
3217 zfs_acl_free(zp
->z_acl_cached
);
3218 zp
->z_acl_cached
= aclp
;
3223 if (mask
& AT_ATIME
) {
3224 ZFS_TIME_ENCODE(&vap
->va_atime
, zp
->z_atime
);
3225 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
,
3226 &zp
->z_atime
, sizeof (zp
->z_atime
));
3229 if (mask
& AT_MTIME
) {
3230 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3231 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
3232 mtime
, sizeof (mtime
));
3235 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3236 if (mask
& AT_SIZE
&& !(mask
& AT_MTIME
)) {
3237 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
),
3238 NULL
, mtime
, sizeof (mtime
));
3239 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3240 &ctime
, sizeof (ctime
));
3241 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
3243 } else if (mask
!= 0) {
3244 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3245 &ctime
, sizeof (ctime
));
3246 zfs_tstamp_update_setup(zp
, STATE_CHANGED
, mtime
, ctime
,
3249 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3250 SA_ZPL_CTIME(zfsvfs
), NULL
,
3251 &ctime
, sizeof (ctime
));
3252 zfs_tstamp_update_setup(attrzp
, STATE_CHANGED
,
3253 mtime
, ctime
, B_TRUE
);
3257 * Do this after setting timestamps to prevent timestamp
3258 * update from toggling bit
3261 if (xoap
&& (mask
& AT_XVATTR
)) {
3264 * restore trimmed off masks
3265 * so that return masks can be set for caller.
3268 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_APPENDONLY
)) {
3269 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3271 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_NOUNLINK
)) {
3272 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3274 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_IMMUTABLE
)) {
3275 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3277 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_NODUMP
)) {
3278 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3280 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_AV_MODIFIED
)) {
3281 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3283 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_AV_QUARANTINED
)) {
3284 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3287 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3288 ASSERT(vp
->v_type
== VREG
);
3290 zfs_xvattr_set(zp
, xvap
, tx
);
3294 zfs_fuid_sync(zfsvfs
, tx
);
3297 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3299 mutex_exit(&zp
->z_lock
);
3300 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3301 mutex_exit(&zp
->z_acl_lock
);
3304 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3305 mutex_exit(&attrzp
->z_acl_lock
);
3306 mutex_exit(&attrzp
->z_lock
);
3309 if (err
== 0 && attrzp
) {
3310 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3316 VN_RELE(ZTOV(attrzp
));
3322 zfs_fuid_info_free(fuidp
);
3328 if (err
== ERESTART
)
3331 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3336 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3337 zil_commit(zilog
, 0);
3343 typedef struct zfs_zlock
{
3344 krwlock_t
*zl_rwlock
; /* lock we acquired */
3345 znode_t
*zl_znode
; /* znode we held */
3346 struct zfs_zlock
*zl_next
; /* next in list */
3350 * Drop locks and release vnodes that were held by zfs_rename_lock().
3353 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3357 while ((zl
= *zlpp
) != NULL
) {
3358 if (zl
->zl_znode
!= NULL
)
3359 VN_RELE(ZTOV(zl
->zl_znode
));
3360 rw_exit(zl
->zl_rwlock
);
3361 *zlpp
= zl
->zl_next
;
3362 kmem_free(zl
, sizeof (*zl
));
3367 * Search back through the directory tree, using the ".." entries.
3368 * Lock each directory in the chain to prevent concurrent renames.
3369 * Fail any attempt to move a directory into one of its own descendants.
3370 * XXX - z_parent_lock can overlap with map or grow locks
3373 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3377 uint64_t rootid
= zp
->z_zfsvfs
->z_root
;
3378 uint64_t oidp
= zp
->z_id
;
3379 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3380 krw_t rw
= RW_WRITER
;
3383 * First pass write-locks szp and compares to zp->z_id.
3384 * Later passes read-lock zp and compare to zp->z_parent.
3387 if (!rw_tryenter(rwlp
, rw
)) {
3389 * Another thread is renaming in this path.
3390 * Note that if we are a WRITER, we don't have any
3391 * parent_locks held yet.
3393 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3395 * Drop our locks and restart
3397 zfs_rename_unlock(&zl
);
3401 rwlp
= &szp
->z_parent_lock
;
3406 * Wait for other thread to drop its locks
3412 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3413 zl
->zl_rwlock
= rwlp
;
3414 zl
->zl_znode
= NULL
;
3415 zl
->zl_next
= *zlpp
;
3418 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3419 return (SET_ERROR(EINVAL
));
3421 if (oidp
== rootid
) /* We've hit the top */
3424 if (rw
== RW_READER
) { /* i.e. not the first pass */
3425 int error
= zfs_zget(zp
->z_zfsvfs
, oidp
, &zp
);
3430 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zp
->z_zfsvfs
),
3431 &oidp
, sizeof (oidp
));
3432 rwlp
= &zp
->z_parent_lock
;
3435 } while (zp
->z_id
!= sdzp
->z_id
);
3441 * Move an entry from the provided source directory to the target
3442 * directory. Change the entry name as indicated.
3444 * IN: sdvp - Source directory containing the "old entry".
3445 * snm - Old entry name.
3446 * tdvp - Target directory to contain the "new entry".
3447 * tnm - New entry name.
3448 * cr - credentials of caller.
3449 * ct - caller context
3450 * flags - case flags
3452 * RETURN: 0 on success, error code on failure.
3455 * sdvp,tdvp - ctime|mtime updated
3459 zfs_rename(vnode_t
*sdvp
, char *snm
, vnode_t
*tdvp
, char *tnm
, cred_t
*cr
,
3460 caller_context_t
*ct
, int flags
)
3462 znode_t
*tdzp
, *szp
, *tzp
;
3463 znode_t
*sdzp
= VTOZ(sdvp
);
3464 zfsvfs_t
*zfsvfs
= sdzp
->z_zfsvfs
;
3467 zfs_dirlock_t
*sdl
, *tdl
;
3470 int cmp
, serr
, terr
;
3471 int error
= 0, rm_err
= 0;
3473 boolean_t waited
= B_FALSE
;
3476 ZFS_VERIFY_ZP(sdzp
);
3477 zilog
= zfsvfs
->z_log
;
3480 * Make sure we have the real vp for the target directory.
3482 if (fop_realvp(tdvp
, &realvp
, ct
) == 0)
3486 ZFS_VERIFY_ZP(tdzp
);
3489 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3490 * ctldir appear to have the same v_vfsp.
3492 if (tdzp
->z_zfsvfs
!= zfsvfs
|| zfsctl_is_node(tdvp
)) {
3494 return (SET_ERROR(EXDEV
));
3497 if (zfsvfs
->z_utf8
&& u8_validate(tnm
,
3498 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3500 return (SET_ERROR(EILSEQ
));
3503 if (flags
& FIGNORECASE
)
3512 * This is to prevent the creation of links into attribute space
3513 * by renaming a linked file into/outof an attribute directory.
3514 * See the comment in zfs_link() for why this is considered bad.
3516 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3518 return (SET_ERROR(EINVAL
));
3522 * Lock source and target directory entries. To prevent deadlock,
3523 * a lock ordering must be defined. We lock the directory with
3524 * the smallest object id first, or if it's a tie, the one with
3525 * the lexically first name.
3527 if (sdzp
->z_id
< tdzp
->z_id
) {
3529 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3533 * First compare the two name arguments without
3534 * considering any case folding.
3536 int nofold
= (zfsvfs
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3538 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3539 ASSERT(error
== 0 || !zfsvfs
->z_utf8
);
3542 * POSIX: "If the old argument and the new argument
3543 * both refer to links to the same existing file,
3544 * the rename() function shall return successfully
3545 * and perform no other action."
3551 * If the file system is case-folding, then we may
3552 * have some more checking to do. A case-folding file
3553 * system is either supporting mixed case sensitivity
3554 * access or is completely case-insensitive. Note
3555 * that the file system is always case preserving.
3557 * In mixed sensitivity mode case sensitive behavior
3558 * is the default. FIGNORECASE must be used to
3559 * explicitly request case insensitive behavior.
3561 * If the source and target names provided differ only
3562 * by case (e.g., a request to rename 'tim' to 'Tim'),
3563 * we will treat this as a special case in the
3564 * case-insensitive mode: as long as the source name
3565 * is an exact match, we will allow this to proceed as
3566 * a name-change request.
3568 if ((zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
||
3569 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&&
3570 flags
& FIGNORECASE
)) &&
3571 u8_strcmp(snm
, tnm
, 0, zfsvfs
->z_norm
, U8_UNICODE_LATEST
,
3574 * case preserving rename request, require exact
3583 * If the source and destination directories are the same, we should
3584 * grab the z_name_lock of that directory only once.
3588 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3592 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3593 ZEXISTS
| zflg
, NULL
, NULL
);
3594 terr
= zfs_dirent_lock(&tdl
,
3595 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3597 terr
= zfs_dirent_lock(&tdl
,
3598 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3599 serr
= zfs_dirent_lock(&sdl
,
3600 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3606 * Source entry invalid or not there.
3609 zfs_dirent_unlock(tdl
);
3615 rw_exit(&sdzp
->z_name_lock
);
3617 if (strcmp(snm
, "..") == 0)
3618 serr
= SET_ERROR(EINVAL
);
3623 zfs_dirent_unlock(sdl
);
3627 rw_exit(&sdzp
->z_name_lock
);
3629 if (strcmp(tnm
, "..") == 0)
3630 terr
= SET_ERROR(EINVAL
);
3636 * Must have write access at the source to remove the old entry
3637 * and write access at the target to create the new entry.
3638 * Note that if target and source are the same, this can be
3639 * done in a single check.
3642 if (error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
))
3645 if (ZTOV(szp
)->v_type
== VDIR
) {
3647 * Check to make sure rename is valid.
3648 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3650 if (error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
))
3655 * Does target exist?
3659 * Source and target must be the same type.
3661 if (ZTOV(szp
)->v_type
== VDIR
) {
3662 if (ZTOV(tzp
)->v_type
!= VDIR
) {
3663 error
= SET_ERROR(ENOTDIR
);
3667 if (ZTOV(tzp
)->v_type
== VDIR
) {
3668 error
= SET_ERROR(EISDIR
);
3673 * POSIX dictates that when the source and target
3674 * entries refer to the same file object, rename
3675 * must do nothing and exit without error.
3677 if (szp
->z_id
== tzp
->z_id
) {
3683 vnevent_pre_rename_src(ZTOV(szp
), sdvp
, snm
, ct
);
3685 vnevent_pre_rename_dest(ZTOV(tzp
), tdvp
, tnm
, ct
);
3688 * notify the target directory if it is not the same
3689 * as source directory.
3692 vnevent_pre_rename_dest_dir(tdvp
, ZTOV(szp
), tnm
, ct
);
3695 tx
= dmu_tx_create(zfsvfs
->z_os
);
3696 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3697 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3698 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3699 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3701 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3702 zfs_sa_upgrade_txholds(tx
, tdzp
);
3705 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3706 zfs_sa_upgrade_txholds(tx
, tzp
);
3709 zfs_sa_upgrade_txholds(tx
, szp
);
3710 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
3711 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3714 zfs_rename_unlock(&zl
);
3715 zfs_dirent_unlock(sdl
);
3716 zfs_dirent_unlock(tdl
);
3719 rw_exit(&sdzp
->z_name_lock
);
3724 if (error
== ERESTART
) {
3735 if (tzp
) /* Attempt to remove the existing target */
3736 error
= rm_err
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3739 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3741 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3743 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zfsvfs
),
3744 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3747 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3749 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3750 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3751 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3754 * Update path information for the target vnode
3756 vn_renamepath(tdvp
, ZTOV(szp
), tnm
,
3760 * At this point, we have successfully created
3761 * the target name, but have failed to remove
3762 * the source name. Since the create was done
3763 * with the ZRENAMING flag, there are
3764 * complications; for one, the link count is
3765 * wrong. The easiest way to deal with this
3766 * is to remove the newly created target, and
3767 * return the original error. This must
3768 * succeed; fortunately, it is very unlikely to
3769 * fail, since we just created it.
3771 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3772 ZRENAMING
, NULL
), ==, 0);
3779 if (tzp
&& rm_err
== 0)
3780 vnevent_rename_dest(ZTOV(tzp
), tdvp
, tnm
, ct
);
3783 vnevent_rename_src(ZTOV(szp
), sdvp
, snm
, ct
);
3784 /* notify the target dir if it is not the same as source dir */
3786 vnevent_rename_dest_dir(tdvp
, ct
);
3790 zfs_rename_unlock(&zl
);
3792 zfs_dirent_unlock(sdl
);
3793 zfs_dirent_unlock(tdl
);
3796 rw_exit(&sdzp
->z_name_lock
);
3803 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3804 zil_commit(zilog
, 0);
3811 * Insert the indicated symbolic reference entry into the directory.
3813 * IN: dvp - Directory to contain new symbolic link.
3814 * link - Name for new symlink entry.
3815 * vap - Attributes of new entry.
3816 * cr - credentials of caller.
3817 * ct - caller context
3818 * flags - case flags
3820 * RETURN: 0 on success, error code on failure.
3823 * dvp - ctime|mtime updated
3827 zfs_symlink(vnode_t
*dvp
, char *name
, vattr_t
*vap
, char *link
, cred_t
*cr
,
3828 caller_context_t
*ct
, int flags
)
3830 znode_t
*zp
, *dzp
= VTOZ(dvp
);
3833 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
3835 uint64_t len
= strlen(link
);
3838 zfs_acl_ids_t acl_ids
;
3839 boolean_t fuid_dirtied
;
3840 uint64_t txtype
= TX_SYMLINK
;
3841 boolean_t waited
= B_FALSE
;
3843 ASSERT(vap
->va_type
== VLNK
);
3847 zilog
= zfsvfs
->z_log
;
3849 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
3850 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3852 return (SET_ERROR(EILSEQ
));
3854 if (flags
& FIGNORECASE
)
3857 if (len
> MAXPATHLEN
) {
3859 return (SET_ERROR(ENAMETOOLONG
));
3862 if ((error
= zfs_acl_ids_create(dzp
, 0,
3863 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3869 * Attempt to lock directory; fail if entry already exists.
3871 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3873 zfs_acl_ids_free(&acl_ids
);
3878 if (error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
)) {
3879 zfs_acl_ids_free(&acl_ids
);
3880 zfs_dirent_unlock(dl
);
3885 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
3886 zfs_acl_ids_free(&acl_ids
);
3887 zfs_dirent_unlock(dl
);
3889 return (SET_ERROR(EDQUOT
));
3891 tx
= dmu_tx_create(zfsvfs
->z_os
);
3892 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3893 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3894 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3895 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3896 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3897 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3898 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3899 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3900 acl_ids
.z_aclp
->z_acl_bytes
);
3903 zfs_fuid_txhold(zfsvfs
, tx
);
3904 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
3906 zfs_dirent_unlock(dl
);
3907 if (error
== ERESTART
) {
3913 zfs_acl_ids_free(&acl_ids
);
3920 * Create a new object for the symlink.
3921 * for version 4 ZPL datsets the symlink will be an SA attribute
3923 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3926 zfs_fuid_sync(zfsvfs
, tx
);
3928 mutex_enter(&zp
->z_lock
);
3930 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zfsvfs
),
3933 zfs_sa_symlink(zp
, link
, len
, tx
);
3934 mutex_exit(&zp
->z_lock
);
3937 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
3938 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3940 * Insert the new object into the directory.
3942 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3944 if (flags
& FIGNORECASE
)
3946 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3948 zfs_acl_ids_free(&acl_ids
);
3952 zfs_dirent_unlock(dl
);
3956 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3957 zil_commit(zilog
, 0);
3964 * Return, in the buffer contained in the provided uio structure,
3965 * the symbolic path referred to by vp.
3967 * IN: vp - vnode of symbolic link.
3968 * uio - structure to contain the link path.
3969 * cr - credentials of caller.
3970 * ct - caller context
3972 * OUT: uio - structure containing the link path.
3974 * RETURN: 0 on success, error code on failure.
3977 * vp - atime updated
3981 zfs_readlink(vnode_t
*vp
, uio_t
*uio
, cred_t
*cr
, caller_context_t
*ct
)
3983 znode_t
*zp
= VTOZ(vp
);
3984 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
3990 mutex_enter(&zp
->z_lock
);
3992 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3993 SA_ZPL_SYMLINK(zfsvfs
), uio
);
3995 error
= zfs_sa_readlink(zp
, uio
);
3996 mutex_exit(&zp
->z_lock
);
3998 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
4005 * Insert a new entry into directory tdvp referencing svp.
4007 * IN: tdvp - Directory to contain new entry.
4008 * svp - vnode of new entry.
4009 * name - name of new entry.
4010 * cr - credentials of caller.
4011 * ct - caller context
4013 * RETURN: 0 on success, error code on failure.
4016 * tdvp - ctime|mtime updated
4017 * svp - ctime updated
4021 zfs_link(vnode_t
*tdvp
, vnode_t
*svp
, char *name
, cred_t
*cr
,
4022 caller_context_t
*ct
, int flags
)
4024 znode_t
*dzp
= VTOZ(tdvp
);
4026 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
4035 boolean_t waited
= B_FALSE
;
4037 ASSERT(tdvp
->v_type
== VDIR
);
4041 zilog
= zfsvfs
->z_log
;
4043 if (fop_realvp(svp
, &realvp
, ct
) == 0)
4047 * POSIX dictates that we return EPERM here.
4048 * Better choices include ENOTSUP or EISDIR.
4050 if (svp
->v_type
== VDIR
) {
4052 return (SET_ERROR(EPERM
));
4059 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
4060 * ctldir appear to have the same v_vfsp.
4062 if (szp
->z_zfsvfs
!= zfsvfs
|| zfsctl_is_node(svp
)) {
4064 return (SET_ERROR(EXDEV
));
4067 /* Prevent links to .zfs/shares files */
4069 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
4070 &parent
, sizeof (uint64_t))) != 0) {
4074 if (parent
== zfsvfs
->z_shares_dir
) {
4076 return (SET_ERROR(EPERM
));
4079 if (zfsvfs
->z_utf8
&& u8_validate(name
,
4080 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
4082 return (SET_ERROR(EILSEQ
));
4084 if (flags
& FIGNORECASE
)
4088 * We do not support links between attributes and non-attributes
4089 * because of the potential security risk of creating links
4090 * into "normal" file space in order to circumvent restrictions
4091 * imposed in attribute space.
4093 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
4095 return (SET_ERROR(EINVAL
));
4099 owner
= zfs_fuid_map_id(zfsvfs
, szp
->z_uid
, cr
, ZFS_OWNER
);
4100 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
4102 return (SET_ERROR(EPERM
));
4105 if (error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
)) {
4112 * Attempt to lock directory; fail if entry already exists.
4114 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
4120 tx
= dmu_tx_create(zfsvfs
->z_os
);
4121 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
4122 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
4123 zfs_sa_upgrade_txholds(tx
, szp
);
4124 zfs_sa_upgrade_txholds(tx
, dzp
);
4125 error
= dmu_tx_assign(tx
, waited
? TXG_WAITED
: TXG_NOWAIT
);
4127 zfs_dirent_unlock(dl
);
4128 if (error
== ERESTART
) {
4139 error
= zfs_link_create(dl
, szp
, tx
, 0);
4142 uint64_t txtype
= TX_LINK
;
4143 if (flags
& FIGNORECASE
)
4145 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
4150 zfs_dirent_unlock(dl
);
4153 vnevent_link(svp
, ct
);
4156 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4157 zil_commit(zilog
, 0);
4164 * zfs_null_putapage() is used when the file system has been force
4165 * unmounted. It just drops the pages.
4169 zfs_null_putapage(vnode_t
*vp
, page_t
*pp
, uoff_t
*offp
,
4170 size_t *lenp
, int flags
, cred_t
*cr
)
4172 pvn_write_done(pp
, B_INVAL
|B_FORCE
|B_ERROR
);
4177 * Push a page out to disk, klustering if possible.
4179 * IN: vp - file to push page to.
4180 * pp - page to push.
4181 * flags - additional flags.
4182 * cr - credentials of caller.
4184 * OUT: offp - start of range pushed.
4185 * lenp - len of range pushed.
4187 * RETURN: 0 on success, error code on failure.
4189 * NOTE: callers must have locked the page to be pushed. On
4190 * exit, the page (and all other pages in the kluster) must be
4195 zfs_putapage(vnode_t
*vp
, page_t
*pp
, uoff_t
*offp
,
4196 size_t *lenp
, int flags
, cred_t
*cr
)
4198 znode_t
*zp
= VTOZ(vp
);
4199 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4208 * If our blocksize is bigger than the page size, try to kluster
4209 * multiple pages so that we write a full block (thus avoiding
4210 * a read-modify-write).
4212 if (off
< zp
->z_size
&& zp
->z_blksz
> PAGESIZE
) {
4213 klen
= P2ROUNDUP((ulong_t
)zp
->z_blksz
, PAGESIZE
);
4214 koff
= ISP2(klen
) ? P2ALIGN(off
, (uoff_t
)klen
) : 0;
4215 ASSERT(koff
<= zp
->z_size
);
4216 if (koff
+ klen
> zp
->z_size
)
4217 klen
= P2ROUNDUP(zp
->z_size
- koff
, (uint64_t)PAGESIZE
);
4218 pp
= pvn_write_kluster(vp
, pp
, &off
, &len
, koff
, klen
, flags
);
4220 ASSERT3U(btop(len
), ==, btopr(len
));
4223 * Can't push pages past end-of-file.
4225 if (off
>= zp
->z_size
) {
4226 /* ignore all pages */
4229 } else if (off
+ len
> zp
->z_size
) {
4230 int npages
= btopr(zp
->z_size
- off
);
4233 page_list_break(&pp
, &trunc
, npages
);
4234 /* ignore pages past end of file */
4236 pvn_write_done(trunc
, flags
);
4237 len
= zp
->z_size
- off
;
4240 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
4241 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
4242 err
= SET_ERROR(EDQUOT
);
4245 tx
= dmu_tx_create(zfsvfs
->z_os
);
4246 dmu_tx_hold_write(tx
, zp
->z_id
, off
, len
);
4248 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4249 zfs_sa_upgrade_txholds(tx
, zp
);
4250 err
= dmu_tx_assign(tx
, TXG_WAIT
);
4256 if (zp
->z_blksz
<= PAGESIZE
) {
4257 caddr_t va
= zfs_map_page(pp
, S_READ
);
4258 ASSERT3U(len
, <=, PAGESIZE
);
4259 dmu_write(zfsvfs
->z_os
, zp
->z_id
, off
, len
, va
, tx
);
4260 zfs_unmap_page(pp
, va
);
4262 err
= dmu_write_pages(zfsvfs
->z_os
, zp
->z_id
, off
, len
, pp
, tx
);
4266 uint64_t mtime
[2], ctime
[2];
4267 sa_bulk_attr_t bulk
[3];
4270 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
4272 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
4274 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
4276 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
4278 err
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
4280 zfs_log_write(zfsvfs
->z_log
, tx
, TX_WRITE
, zp
, off
, len
, 0);
4285 pvn_write_done(pp
, (err
? B_ERROR
: 0) | flags
);
4295 * Copy the portion of the file indicated from pages into the file.
4296 * The pages are stored in a page list attached to the files vnode.
4298 * IN: vp - vnode of file to push page data to.
4299 * off - position in file to put data.
4300 * len - amount of data to write.
4301 * flags - flags to control the operation.
4302 * cr - credentials of caller.
4303 * ct - caller context.
4305 * RETURN: 0 on success, error code on failure.
4308 * vp - ctime|mtime updated
4312 zfs_putpage(vnode_t
*vp
, offset_t off
, size_t len
, int flags
, cred_t
*cr
,
4313 caller_context_t
*ct
)
4315 znode_t
*zp
= VTOZ(vp
);
4316 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4328 * There's nothing to do if no data is cached.
4330 if (!vn_has_cached_data(vp
)) {
4336 * Align this request to the file block size in case we kluster.
4337 * XXX - this can result in pretty aggresive locking, which can
4338 * impact simultanious read/write access. One option might be
4339 * to break up long requests (len == 0) into block-by-block
4340 * operations to get narrower locking.
4342 blksz
= zp
->z_blksz
;
4344 io_off
= P2ALIGN_TYPED(off
, blksz
, uoff_t
);
4347 if (len
> 0 && ISP2(blksz
))
4348 io_len
= P2ROUNDUP_TYPED(len
+ (off
- io_off
), blksz
, size_t);
4354 * Search the entire vp list for pages >= io_off.
4356 rl
= zfs_range_lock(zp
, io_off
, UINT64_MAX
, RL_WRITER
);
4357 error
= pvn_vplist_dirty(vp
, io_off
, zfs_putapage
, flags
, cr
);
4360 rl
= zfs_range_lock(zp
, io_off
, io_len
, RL_WRITER
);
4362 if (off
> zp
->z_size
) {
4363 /* past end of file */
4364 zfs_range_unlock(rl
);
4369 len
= MIN(io_len
, P2ROUNDUP(zp
->z_size
, PAGESIZE
) - io_off
);
4371 for (off
= io_off
; io_off
< off
+ len
; io_off
+= io_len
) {
4372 if ((flags
& B_INVAL
) || ((flags
& B_ASYNC
) == 0)) {
4373 pp
= page_lookup(&vp
->v_object
, io_off
,
4374 (flags
& (B_INVAL
| B_FREE
)) ? SE_EXCL
: SE_SHARED
);
4376 pp
= page_lookup_nowait(&vp
->v_object
, io_off
,
4377 (flags
& B_FREE
) ? SE_EXCL
: SE_SHARED
);
4380 if (pp
!= NULL
&& pvn_getdirty(pp
, flags
)) {
4384 * Found a dirty page to push
4386 err
= zfs_putapage(vp
, pp
, &io_off
, &io_len
, flags
, cr
);
4394 zfs_range_unlock(rl
);
4395 if ((flags
& B_ASYNC
) == 0 || zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4396 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
4403 zfs_inactive(vnode_t
*vp
, cred_t
*cr
, caller_context_t
*ct
)
4405 znode_t
*zp
= VTOZ(vp
);
4406 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4409 rw_enter(&zfsvfs
->z_teardown_inactive_lock
, RW_READER
);
4410 if (zp
->z_sa_hdl
== NULL
) {
4412 * The fs has been unmounted, or we did a
4413 * suspend/resume and this file no longer exists.
4415 if (vn_has_cached_data(vp
)) {
4416 (void) pvn_vplist_dirty(vp
, 0, zfs_null_putapage
,
4420 mutex_enter(&zp
->z_lock
);
4421 mutex_enter(&vp
->v_lock
);
4422 ASSERT(vp
->v_count
== 1);
4424 mutex_exit(&vp
->v_lock
);
4425 mutex_exit(&zp
->z_lock
);
4426 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4432 * Attempt to push any data in the page cache. If this fails
4433 * we will get kicked out later in zfs_zinactive().
4435 if (vn_has_cached_data(vp
)) {
4436 (void) pvn_vplist_dirty(vp
, 0, zfs_putapage
, B_INVAL
|B_ASYNC
,
4440 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4441 dmu_tx_t
*tx
= dmu_tx_create(zfsvfs
->z_os
);
4443 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4444 zfs_sa_upgrade_txholds(tx
, zp
);
4445 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4449 mutex_enter(&zp
->z_lock
);
4450 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zfsvfs
),
4451 (void *)&zp
->z_atime
, sizeof (zp
->z_atime
), tx
);
4452 zp
->z_atime_dirty
= 0;
4453 mutex_exit(&zp
->z_lock
);
4459 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4463 * Bounds-check the seek operation.
4465 * IN: vp - vnode seeking within
4466 * ooff - old file offset
4467 * noffp - pointer to new file offset
4468 * ct - caller context
4470 * RETURN: 0 on success, EINVAL if new offset invalid.
4474 zfs_seek(vnode_t
*vp
, offset_t ooff
, offset_t
*noffp
,
4475 caller_context_t
*ct
)
4477 if (vp
->v_type
== VDIR
)
4479 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4483 * Pre-filter the generic locking function to trap attempts to place
4484 * a mandatory lock on a memory mapped file.
4487 zfs_frlock(vnode_t
*vp
, int cmd
, flock64_t
*bfp
, int flag
, offset_t offset
,
4488 flk_callback_t
*flk_cbp
, cred_t
*cr
, caller_context_t
*ct
)
4490 znode_t
*zp
= VTOZ(vp
);
4491 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4497 * We are following the UFS semantics with respect to mapcnt
4498 * here: If we see that the file is mapped already, then we will
4499 * return an error, but we don't worry about races between this
4500 * function and zfs_map().
4502 if (zp
->z_mapcnt
> 0 && MANDMODE(zp
->z_mode
)) {
4504 return (SET_ERROR(EAGAIN
));
4507 return (fs_frlock(vp
, cmd
, bfp
, flag
, offset
, flk_cbp
, cr
, ct
));
4511 * If we can't find a page in the cache, we will create a new page
4512 * and fill it with file data. For efficiency, we may try to fill
4513 * multiple pages at once (klustering) to fill up the supplied page
4514 * list. Note that the pages to be filled are held with an exclusive
4515 * lock to prevent access by other threads while they are being filled.
4518 zfs_fillpage(vnode_t
*vp
, uoff_t off
, struct seg
*seg
,
4519 caddr_t addr
, page_t
*pl
[], size_t plsz
, enum seg_rw rw
)
4521 znode_t
*zp
= VTOZ(vp
);
4522 page_t
*pp
, *cur_pp
;
4523 objset_t
*os
= zp
->z_zfsvfs
->z_os
;
4524 uoff_t io_off
, total
;
4528 if (plsz
== PAGESIZE
|| zp
->z_blksz
<= PAGESIZE
) {
4530 * We only have a single page, don't bother klustering
4534 pp
= page_create_va(&vp
->v_object
, io_off
, io_len
,
4535 PG_EXCL
| PG_WAIT
, seg
, addr
);
4538 * Try to find enough pages to fill the page list
4540 pp
= pvn_read_kluster(vp
, off
, seg
, addr
, &io_off
,
4541 &io_len
, off
, plsz
, 0);
4545 * The page already exists, nothing to do here.
4552 * Fill the pages in the kluster.
4555 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4558 ASSERT3U(io_off
, ==, cur_pp
->p_offset
);
4559 va
= zfs_map_page(cur_pp
, S_WRITE
);
4560 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4562 zfs_unmap_page(cur_pp
, va
);
4564 /* On error, toss the entire kluster */
4565 pvn_read_done(pp
, B_ERROR
);
4566 /* convert checksum errors into IO errors */
4568 err
= SET_ERROR(EIO
);
4571 cur_pp
= cur_pp
->p_next
;
4575 * Fill in the page list array from the kluster starting
4576 * from the desired offset `off'.
4577 * NOTE: the page list will always be null terminated.
4579 pvn_plist_init(pp
, pl
, plsz
, off
, io_len
, rw
);
4580 ASSERT(pl
== NULL
|| (*pl
)->p_offset
== off
);
4586 * Return pointers to the pages for the file region [off, off + len]
4587 * in the pl array. If plsz is greater than len, this function may
4588 * also return page pointers from after the specified region
4589 * (i.e. the region [off, off + plsz]). These additional pages are
4590 * only returned if they are already in the cache, or were created as
4591 * part of a klustered read.
4593 * IN: vp - vnode of file to get data from.
4594 * off - position in file to get data from.
4595 * len - amount of data to retrieve.
4596 * plsz - length of provided page list.
4597 * seg - segment to obtain pages for.
4598 * addr - virtual address of fault.
4599 * rw - mode of created pages.
4600 * cr - credentials of caller.
4601 * ct - caller context.
4603 * OUT: protp - protection mode of created pages.
4604 * pl - list of pages created.
4606 * RETURN: 0 on success, error code on failure.
4609 * vp - atime updated
4613 zfs_getpage(vnode_t
*vp
, offset_t off
, size_t len
, uint_t
*protp
,
4614 page_t
*pl
[], size_t plsz
, struct seg
*seg
, caddr_t addr
,
4615 enum seg_rw rw
, cred_t
*cr
, caller_context_t
*ct
)
4617 znode_t
*zp
= VTOZ(vp
);
4618 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4622 /* we do our own caching, faultahead is unnecessary */
4625 else if (len
> plsz
)
4628 len
= P2ROUNDUP(len
, PAGESIZE
);
4629 ASSERT(plsz
>= len
);
4638 * Loop through the requested range [off, off + len) looking
4639 * for pages. If we don't find a page, we will need to create
4640 * a new page and fill it with data from the file.
4643 if (*pl
= page_lookup(&vp
->v_object
, off
, SE_SHARED
))
4645 else if (err
= zfs_fillpage(vp
, off
, seg
, addr
, pl
, plsz
, rw
))
4648 ASSERT3U((*pl
)->p_offset
, ==, off
);
4652 ASSERT3U(len
, >=, PAGESIZE
);
4655 ASSERT3U(plsz
, >=, PAGESIZE
);
4662 * Fill out the page array with any pages already in the cache.
4665 (*pl
++ = page_lookup_nowait(&vp
->v_object
, off
, SE_SHARED
))) {
4672 * Release any pages we have previously locked.
4677 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
4687 * Request a memory map for a section of a file. This code interacts
4688 * with common code and the VM system as follows:
4690 * - common code calls mmap(), which ends up in smmap_common()
4691 * - this calls fop_map(), which takes you into (say) zfs
4692 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4693 * - segvn_create() creates the new segment and calls fop_addmap()
4694 * - zfs_addmap() updates z_mapcnt
4698 zfs_map(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t
*addrp
,
4699 size_t len
, uchar_t prot
, uchar_t maxprot
, uint_t flags
, cred_t
*cr
,
4700 caller_context_t
*ct
)
4702 znode_t
*zp
= VTOZ(vp
);
4703 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4704 segvn_crargs_t vn_a
;
4711 * Note: ZFS_READONLY is handled in zfs_zaccess_common.
4714 if ((prot
& PROT_WRITE
) && (zp
->z_pflags
&
4715 (ZFS_IMMUTABLE
| ZFS_APPENDONLY
))) {
4717 return (SET_ERROR(EPERM
));
4720 if ((prot
& (PROT_READ
| PROT_EXEC
)) &&
4721 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4723 return (SET_ERROR(EACCES
));
4726 if (vp
->v_flag
& VNOMAP
) {
4728 return (SET_ERROR(ENOSYS
));
4731 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4733 return (SET_ERROR(ENXIO
));
4736 if (vp
->v_type
!= VREG
) {
4738 return (SET_ERROR(ENODEV
));
4742 * If file is locked, disallow mapping.
4744 if (MANDMODE(zp
->z_mode
) && vn_has_flocks(vp
)) {
4746 return (SET_ERROR(EAGAIN
));
4750 error
= choose_addr(as
, addrp
, len
, off
, ADDR_VACALIGN
, flags
);
4758 vn_a
.offset
= (uoff_t
)off
;
4759 vn_a
.type
= flags
& MAP_TYPE
;
4761 vn_a
.maxprot
= maxprot
;
4764 vn_a
.flags
= flags
& ~MAP_TYPE
;
4766 vn_a
.lgrp_mem_policy_flags
= 0;
4768 error
= as_map(as
, *addrp
, len
, segvn_create
, &vn_a
);
4777 zfs_addmap(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t addr
,
4778 size_t len
, uchar_t prot
, uchar_t maxprot
, uint_t flags
, cred_t
*cr
,
4779 caller_context_t
*ct
)
4781 uint64_t pages
= btopr(len
);
4783 atomic_add_64(&VTOZ(vp
)->z_mapcnt
, pages
);
4788 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4789 * more accurate mtime for the associated file. Since we don't have a way of
4790 * detecting when the data was actually modified, we have to resort to
4791 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4792 * last page is pushed. The problem occurs when the msync() call is omitted,
4793 * which by far the most common case:
4801 * putpage() via fsflush
4803 * If we wait until fsflush to come along, we can have a modification time that
4804 * is some arbitrary point in the future. In order to prevent this in the
4805 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4810 zfs_delmap(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t addr
,
4811 size_t len
, uint_t prot
, uint_t maxprot
, uint_t flags
, cred_t
*cr
,
4812 caller_context_t
*ct
)
4814 uint64_t pages
= btopr(len
);
4816 ASSERT3U(VTOZ(vp
)->z_mapcnt
, >=, pages
);
4817 atomic_add_64(&VTOZ(vp
)->z_mapcnt
, -pages
);
4819 if ((flags
& MAP_SHARED
) && (prot
& PROT_WRITE
) &&
4820 vn_has_cached_data(vp
))
4821 (void) fop_putpage(vp
, off
, len
, B_ASYNC
, cr
, ct
);
4827 * Free or allocate space in a file. Currently, this function only
4828 * supports the `F_FREESP' command. However, this command is somewhat
4829 * misnamed, as its functionality includes the ability to allocate as
4830 * well as free space.
4832 * IN: vp - vnode of file to free data in.
4833 * cmd - action to take (only F_FREESP supported).
4834 * bfp - section of file to free/alloc.
4835 * flag - current file open mode flags.
4836 * offset - current file offset.
4837 * cr - credentials of caller [UNUSED].
4838 * ct - caller context.
4840 * RETURN: 0 on success, error code on failure.
4843 * vp - ctime|mtime updated
4847 zfs_space(vnode_t
*vp
, int cmd
, flock64_t
*bfp
, int flag
,
4848 offset_t offset
, cred_t
*cr
, caller_context_t
*ct
)
4850 znode_t
*zp
= VTOZ(vp
);
4851 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4858 if (cmd
!= F_FREESP
) {
4860 return (SET_ERROR(EINVAL
));
4864 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our
4865 * callers might not be able to detect properly that we are read-only,
4866 * so check it explicitly here.
4868 if (zfsvfs
->z_vfs
->vfs_flag
& VFS_RDONLY
) {
4870 return (SET_ERROR(EROFS
));
4873 if (error
= convoff(vp
, bfp
, 0, offset
)) {
4878 if (bfp
->l_len
< 0) {
4880 return (SET_ERROR(EINVAL
));
4884 len
= bfp
->l_len
; /* 0 means from off to end of file */
4886 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4888 if (error
== 0 && off
== 0 && len
== 0)
4889 vnevent_truncate(ZTOV(zp
), ct
);
4897 zfs_fid(vnode_t
*vp
, fid_t
*fidp
, caller_context_t
*ct
)
4899 znode_t
*zp
= VTOZ(vp
);
4900 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4903 uint64_t object
= zp
->z_id
;
4910 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zfsvfs
),
4911 &gen64
, sizeof (uint64_t))) != 0) {
4916 gen
= (uint32_t)gen64
;
4918 size
= (zfsvfs
->z_parent
!= zfsvfs
) ? LONG_FID_LEN
: SHORT_FID_LEN
;
4919 if (fidp
->fid_len
< size
) {
4920 fidp
->fid_len
= size
;
4922 return (SET_ERROR(ENOSPC
));
4925 zfid
= (zfid_short_t
*)fidp
;
4927 zfid
->zf_len
= size
;
4929 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4930 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4932 /* Must have a non-zero generation number to distinguish from .zfs */
4935 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4936 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4938 if (size
== LONG_FID_LEN
) {
4939 uint64_t objsetid
= dmu_objset_id(zfsvfs
->z_os
);
4942 zlfid
= (zfid_long_t
*)fidp
;
4944 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
4945 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
4947 /* XXX - this should be the generation number for the objset */
4948 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
4949 zlfid
->zf_setgen
[i
] = 0;
4957 zfs_pathconf(vnode_t
*vp
, int cmd
, ulong_t
*valp
, cred_t
*cr
,
4958 caller_context_t
*ct
)
4970 case _PC_FILESIZEBITS
:
4974 case _PC_XATTR_EXISTS
:
4976 zfsvfs
= zp
->z_zfsvfs
;
4980 error
= zfs_dirent_lock(&dl
, zp
, "", &xzp
,
4981 ZXATTR
| ZEXISTS
| ZSHARED
, NULL
, NULL
);
4983 zfs_dirent_unlock(dl
);
4984 if (!zfs_dirempty(xzp
))
4987 } else if (error
== ENOENT
) {
4989 * If there aren't extended attributes, it's the
4990 * same as having zero of them.
4997 case _PC_SATTR_ENABLED
:
4998 case _PC_SATTR_EXISTS
:
4999 *valp
= vfs_has_feature(vp
->v_vfsp
, VFSFT_SYSATTR_VIEWS
) &&
5000 (vp
->v_type
== VREG
|| vp
->v_type
== VDIR
);
5003 case _PC_ACCESS_FILTERING
:
5004 *valp
= vfs_has_feature(vp
->v_vfsp
, VFSFT_ACCESS_FILTER
) &&
5008 case _PC_ACL_ENABLED
:
5009 *valp
= _ACL_ACE_ENABLED
;
5012 case _PC_MIN_HOLE_SIZE
:
5013 *valp
= (ulong_t
)SPA_MINBLOCKSIZE
;
5016 case _PC_TIMESTAMP_RESOLUTION
:
5017 /* nanosecond timestamp resolution */
5022 return (fs_pathconf(vp
, cmd
, valp
, cr
, ct
));
5028 zfs_getsecattr(vnode_t
*vp
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
,
5029 caller_context_t
*ct
)
5031 znode_t
*zp
= VTOZ(vp
);
5032 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
5034 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
5038 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
5046 zfs_setsecattr(vnode_t
*vp
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
,
5047 caller_context_t
*ct
)
5049 znode_t
*zp
= VTOZ(vp
);
5050 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
5052 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
5053 zilog_t
*zilog
= zfsvfs
->z_log
;
5058 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
5060 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
5061 zil_commit(zilog
, 0);
5068 * The smallest read we may consider to loan out an arcbuf.
5069 * This must be a power of 2.
5071 int zcr_blksz_min
= (1 << 10); /* 1K */
5073 * If set to less than the file block size, allow loaning out of an
5074 * arcbuf for a partial block read. This must be a power of 2.
5076 int zcr_blksz_max
= (1 << 17); /* 128K */
5080 zfs_reqzcbuf(vnode_t
*vp
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
,
5081 caller_context_t
*ct
)
5083 znode_t
*zp
= VTOZ(vp
);
5084 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
5085 int max_blksz
= zfsvfs
->z_max_blksz
;
5086 uio_t
*uio
= &xuio
->xu_uio
;
5087 ssize_t size
= uio
->uio_resid
;
5088 offset_t offset
= uio
->uio_loffset
;
5093 int preamble
, postamble
;
5095 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
5096 return (SET_ERROR(EINVAL
));
5103 * Loan out an arc_buf for write if write size is bigger than
5104 * max_blksz, and the file's block size is also max_blksz.
5107 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
5109 return (SET_ERROR(EINVAL
));
5112 * Caller requests buffers for write before knowing where the
5113 * write offset might be (e.g. NFS TCP write).
5118 preamble
= P2PHASE(offset
, blksz
);
5120 preamble
= blksz
- preamble
;
5125 postamble
= P2PHASE(size
, blksz
);
5128 fullblk
= size
/ blksz
;
5129 (void) dmu_xuio_init(xuio
,
5130 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
5131 DTRACE_PROBE3(zfs_reqzcbuf_align
, int, preamble
,
5132 int, postamble
, int,
5133 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
5136 * Have to fix iov base/len for partial buffers. They
5137 * currently represent full arc_buf's.
5140 /* data begins in the middle of the arc_buf */
5141 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5144 (void) dmu_xuio_add(xuio
, abuf
,
5145 blksz
- preamble
, preamble
);
5148 for (i
= 0; i
< fullblk
; i
++) {
5149 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5152 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
5156 /* data ends in the middle of the arc_buf */
5157 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5160 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
5165 * Loan out an arc_buf for read if the read size is larger than
5166 * the current file block size. Block alignment is not
5167 * considered. Partial arc_buf will be loaned out for read.
5169 blksz
= zp
->z_blksz
;
5170 if (blksz
< zcr_blksz_min
)
5171 blksz
= zcr_blksz_min
;
5172 if (blksz
> zcr_blksz_max
)
5173 blksz
= zcr_blksz_max
;
5174 /* avoid potential complexity of dealing with it */
5175 if (blksz
> max_blksz
) {
5177 return (SET_ERROR(EINVAL
));
5180 maxsize
= zp
->z_size
- uio
->uio_loffset
;
5184 if (size
< blksz
|| vn_has_cached_data(vp
)) {
5186 return (SET_ERROR(EINVAL
));
5191 return (SET_ERROR(EINVAL
));
5194 uio
->uio_extflg
= UIO_XUIO
;
5195 XUIO_XUZC_RW(xuio
) = ioflag
;
5202 zfs_retzcbuf(vnode_t
*vp
, xuio_t
*xuio
, cred_t
*cr
, caller_context_t
*ct
)
5206 int ioflag
= XUIO_XUZC_RW(xuio
);
5208 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
5210 i
= dmu_xuio_cnt(xuio
);
5212 abuf
= dmu_xuio_arcbuf(xuio
, i
);
5214 * if abuf == NULL, it must be a write buffer
5215 * that has been returned in zfs_write().
5218 dmu_return_arcbuf(abuf
);
5219 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
5222 dmu_xuio_fini(xuio
);
5227 * Predeclare these here so that the compiler assumes that
5228 * this is an "old style" function declaration that does
5229 * not include arguments => we won't get type mismatch errors
5230 * in the initializations that follow.
5232 static int zfs_inval();
5233 static int zfs_isdir();
5238 return (SET_ERROR(EINVAL
));
5244 return (SET_ERROR(EISDIR
));
5248 * Directory vnode operations
5250 const struct vnodeops zfs_dvnodeops
= {
5252 .vop_open
= zfs_open
,
5253 .vop_close
= zfs_close
,
5254 .vop_read
= zfs_isdir
,
5255 .vop_write
= zfs_isdir
,
5256 .vop_ioctl
= zfs_ioctl
,
5257 .vop_getattr
= zfs_getattr
,
5258 .vop_setattr
= zfs_setattr
,
5259 .vop_access
= zfs_access
,
5260 .vop_lookup
= zfs_lookup
,
5261 .vop_create
= zfs_create
,
5262 .vop_remove
= zfs_remove
,
5263 .vop_link
= zfs_link
,
5264 .vop_rename
= zfs_rename
,
5265 .vop_mkdir
= zfs_mkdir
,
5266 .vop_rmdir
= zfs_rmdir
,
5267 .vop_readdir
= zfs_readdir
,
5268 .vop_symlink
= zfs_symlink
,
5269 .vop_fsync
= zfs_fsync
,
5270 .vop_inactive
= zfs_inactive
,
5272 .vop_seek
= zfs_seek
,
5273 .vop_pathconf
= zfs_pathconf
,
5274 .vop_getsecattr
= zfs_getsecattr
,
5275 .vop_setsecattr
= zfs_setsecattr
,
5276 .vop_vnevent
= fs_vnevent_support
,
5280 * Regular file vnode operations
5282 const struct vnodeops zfs_fvnodeops
= {
5284 .vop_open
= zfs_open
,
5285 .vop_close
= zfs_close
,
5286 .vop_read
= zfs_read
,
5287 .vop_write
= zfs_write
,
5288 .vop_ioctl
= zfs_ioctl
,
5289 .vop_getattr
= zfs_getattr
,
5290 .vop_setattr
= zfs_setattr
,
5291 .vop_access
= zfs_access
,
5292 .vop_lookup
= zfs_lookup
,
5293 .vop_rename
= zfs_rename
,
5294 .vop_fsync
= zfs_fsync
,
5295 .vop_inactive
= zfs_inactive
,
5297 .vop_seek
= zfs_seek
,
5298 .vop_frlock
= zfs_frlock
,
5299 .vop_space
= zfs_space
,
5300 .vop_getpage
= zfs_getpage
,
5301 .vop_putpage
= zfs_putpage
,
5303 .vop_addmap
= zfs_addmap
,
5304 .vop_delmap
= zfs_delmap
,
5305 .vop_pathconf
= zfs_pathconf
,
5306 .vop_getsecattr
= zfs_getsecattr
,
5307 .vop_setsecattr
= zfs_setsecattr
,
5308 .vop_vnevent
= fs_vnevent_support
,
5309 .vop_reqzcbuf
= zfs_reqzcbuf
,
5310 .vop_retzcbuf
= zfs_retzcbuf
,
5314 * Symbolic link vnode operations
5316 const struct vnodeops zfs_symvnodeops
= {
5318 .vop_getattr
= zfs_getattr
,
5319 .vop_setattr
= zfs_setattr
,
5320 .vop_access
= zfs_access
,
5321 .vop_rename
= zfs_rename
,
5322 .vop_readlink
= zfs_readlink
,
5323 .vop_inactive
= zfs_inactive
,
5325 .vop_pathconf
= zfs_pathconf
,
5326 .vop_vnevent
= fs_vnevent_support
,
5330 * special share hidden files vnode operations
5332 const struct vnodeops zfs_sharevnodeops
= {
5334 .vop_getattr
= zfs_getattr
,
5335 .vop_access
= zfs_access
,
5336 .vop_inactive
= zfs_inactive
,
5338 .vop_pathconf
= zfs_pathconf
,
5339 .vop_getsecattr
= zfs_getsecattr
,
5340 .vop_setsecattr
= zfs_setsecattr
,
5341 .vop_vnevent
= fs_vnevent_support
,
5345 * Extended attribute directory vnode operations
5347 * These ops are identical to the directory vnode
5348 * operations except for restricted operations:
5352 * Note that there are other restrictions embedded in:
5353 * zfs_create() - restrict type to VREG
5354 * zfs_link() - no links into/out of attribute space
5355 * zfs_rename() - no moves into/out of attribute space
5357 const struct vnodeops zfs_xdvnodeops
= {
5359 .vop_open
= zfs_open
,
5360 .vop_close
= zfs_close
,
5361 .vop_ioctl
= zfs_ioctl
,
5362 .vop_getattr
= zfs_getattr
,
5363 .vop_setattr
= zfs_setattr
,
5364 .vop_access
= zfs_access
,
5365 .vop_lookup
= zfs_lookup
,
5366 .vop_create
= zfs_create
,
5367 .vop_remove
= zfs_remove
,
5368 .vop_link
= zfs_link
,
5369 .vop_rename
= zfs_rename
,
5370 .vop_mkdir
= zfs_inval
,
5371 .vop_rmdir
= zfs_rmdir
,
5372 .vop_readdir
= zfs_readdir
,
5373 .vop_symlink
= zfs_inval
,
5374 .vop_fsync
= zfs_fsync
,
5375 .vop_inactive
= zfs_inactive
,
5377 .vop_seek
= zfs_seek
,
5378 .vop_pathconf
= zfs_pathconf
,
5379 .vop_getsecattr
= zfs_getsecattr
,
5380 .vop_setsecattr
= zfs_setsecattr
,
5381 .vop_vnevent
= fs_vnevent_support
,
5385 * Error vnode operations
5387 const struct vnodeops zfs_evnodeops
= {
5389 .vop_inactive
= zfs_inactive
,
5390 .vop_pathconf
= zfs_pathconf
,