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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
27 /* Portions Copyright 2007 Jeremy Teo */
28 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/resource.h>
37 #include <sys/vfs_opreg.h>
38 #include <sys/vnode.h>
42 #include <sys/taskq.h>
44 #include <sys/vmsystm.h>
45 #include <sys/atomic.h>
47 #include <vm/seg_vn.h>
51 #include <vm/seg_kpm.h>
53 #include <sys/pathname.h>
54 #include <sys/cmn_err.h>
55 #include <sys/errno.h>
56 #include <sys/unistd.h>
57 #include <sys/zfs_dir.h>
58 #include <sys/zfs_acl.h>
59 #include <sys/zfs_ioctl.h>
60 #include <sys/fs/zfs.h>
62 #include <sys/dmu_objset.h>
68 #include <sys/dirent.h>
69 #include <sys/policy.h>
70 #include <sys/sunddi.h>
71 #include <sys/filio.h>
73 #include "fs/fs_subr.h"
74 #include <sys/zfs_ctldir.h>
75 #include <sys/zfs_fuid.h>
76 #include <sys/zfs_sa.h>
78 #include <sys/zfs_rlock.h>
79 #include <sys/extdirent.h>
80 #include <sys/kidmap.h>
87 * Each vnode op performs some logical unit of work. To do this, the ZPL must
88 * properly lock its in-core state, create a DMU transaction, do the work,
89 * record this work in the intent log (ZIL), commit the DMU transaction,
90 * and wait for the intent log to commit if it is a synchronous operation.
91 * Moreover, the vnode ops must work in both normal and log replay context.
92 * The ordering of events is important to avoid deadlocks and references
93 * to freed memory. The example below illustrates the following Big Rules:
95 * (1) A check must be made in each zfs thread for a mounted file system.
96 * This is done avoiding races using ZFS_ENTER(zfsvfs).
97 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes
98 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros
99 * can return EIO from the calling function.
101 * (2) VN_RELE() should always be the last thing except for zil_commit()
102 * (if necessary) and ZFS_EXIT(). This is for 3 reasons:
103 * First, if it's the last reference, the vnode/znode
104 * can be freed, so the zp may point to freed memory. Second, the last
105 * reference will call zfs_zinactive(), which may induce a lot of work --
106 * pushing cached pages (which acquires range locks) and syncing out
107 * cached atime changes. Third, zfs_zinactive() may require a new tx,
108 * which could deadlock the system if you were already holding one.
109 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC().
111 * (3) All range locks must be grabbed before calling dmu_tx_assign(),
112 * as they can span dmu_tx_assign() calls.
114 * (4) Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
115 * This is critical because we don't want to block while holding locks.
116 * Note, in particular, that if a lock is sometimes acquired before
117 * the tx assigns, and sometimes after (e.g. z_lock), then failing to
118 * use a non-blocking assign can deadlock the system. The scenario:
120 * Thread A has grabbed a lock before calling dmu_tx_assign().
121 * Thread B is in an already-assigned tx, and blocks for this lock.
122 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open()
123 * forever, because the previous txg can't quiesce until B's tx commits.
125 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT,
126 * then drop all locks, call dmu_tx_wait(), and try again.
128 * (5) If the operation succeeded, generate the intent log entry for it
129 * before dropping locks. This ensures that the ordering of events
130 * in the intent log matches the order in which they actually occurred.
131 * During ZIL replay the zfs_log_* functions will update the sequence
132 * number to indicate the zil transaction has replayed.
134 * (6) At the end of each vnode op, the DMU tx must always commit,
135 * regardless of whether there were any errors.
137 * (7) After dropping all locks, invoke zil_commit(zilog, foid)
138 * to ensure that synchronous semantics are provided when necessary.
140 * In general, this is how things should be ordered in each vnode op:
142 * ZFS_ENTER(zfsvfs); // exit if unmounted
144 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD())
145 * rw_enter(...); // grab any other locks you need
146 * tx = dmu_tx_create(...); // get DMU tx
147 * dmu_tx_hold_*(); // hold each object you might modify
148 * error = dmu_tx_assign(tx, TXG_NOWAIT); // try to assign
150 * rw_exit(...); // drop locks
151 * zfs_dirent_unlock(dl); // unlock directory entry
152 * VN_RELE(...); // release held vnodes
153 * if (error == ERESTART) {
158 * dmu_tx_abort(tx); // abort DMU tx
159 * ZFS_EXIT(zfsvfs); // finished in zfs
160 * return (error); // really out of space
162 * error = do_real_work(); // do whatever this VOP does
164 * zfs_log_*(...); // on success, make ZIL entry
165 * dmu_tx_commit(tx); // commit DMU tx -- error or not
166 * rw_exit(...); // drop locks
167 * zfs_dirent_unlock(dl); // unlock directory entry
168 * VN_RELE(...); // release held vnodes
169 * zil_commit(zilog, foid); // synchronous when necessary
170 * ZFS_EXIT(zfsvfs); // finished in zfs
171 * return (error); // done, report error
176 zfs_open(vnode_t
**vpp
, int flag
, cred_t
*cr
, caller_context_t
*ct
)
178 znode_t
*zp
= VTOZ(*vpp
);
179 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
184 if ((flag
& FWRITE
) && (zp
->z_pflags
& ZFS_APPENDONLY
) &&
185 ((flag
& FAPPEND
) == 0)) {
187 return (SET_ERROR(EPERM
));
190 if (!zfs_has_ctldir(zp
) && zp
->z_zfsvfs
->z_vscan
&&
191 ZTOV(zp
)->v_type
== VREG
&&
192 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0) {
193 if (fs_vscan(*vpp
, cr
, 0) != 0) {
195 return (SET_ERROR(EACCES
));
199 /* Keep a count of the synchronous opens in the znode */
200 if (flag
& (FSYNC
| FDSYNC
))
201 atomic_inc_32(&zp
->z_sync_cnt
);
209 zfs_close(vnode_t
*vp
, int flag
, int count
, offset_t offset
, cred_t
*cr
,
210 caller_context_t
*ct
)
212 znode_t
*zp
= VTOZ(vp
);
213 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
216 * Clean up any locks held by this process on the vp.
218 cleanlocks(vp
, ddi_get_pid(), 0);
219 cleanshares(vp
, ddi_get_pid());
224 /* Decrement the synchronous opens in the znode */
225 if ((flag
& (FSYNC
| FDSYNC
)) && (count
== 1))
226 atomic_dec_32(&zp
->z_sync_cnt
);
228 if (!zfs_has_ctldir(zp
) && zp
->z_zfsvfs
->z_vscan
&&
229 ZTOV(zp
)->v_type
== VREG
&&
230 !(zp
->z_pflags
& ZFS_AV_QUARANTINED
) && zp
->z_size
> 0)
231 VERIFY(fs_vscan(vp
, cr
, 1) == 0);
238 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and
239 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter.
242 zfs_holey(vnode_t
*vp
, int cmd
, offset_t
*off
)
244 znode_t
*zp
= VTOZ(vp
);
245 uint64_t noff
= (uint64_t)*off
; /* new offset */
250 file_sz
= zp
->z_size
;
251 if (noff
>= file_sz
) {
252 return (SET_ERROR(ENXIO
));
255 if (cmd
== _FIO_SEEK_HOLE
)
260 error
= dmu_offset_next(zp
->z_zfsvfs
->z_os
, zp
->z_id
, hole
, &noff
);
263 if ((error
== ESRCH
) || (noff
> file_sz
)) {
265 * Handle the virtual hole at the end of file.
271 return (SET_ERROR(ENXIO
));
282 zfs_ioctl(vnode_t
*vp
, int com
, intptr_t data
, int flag
, cred_t
*cred
,
283 int *rvalp
, caller_context_t
*ct
)
292 return (zfs_sync(vp
->v_vfsp
, 0, cred
));
295 * The following two ioctls are used by bfu. Faking out,
296 * necessary to avoid bfu errors.
304 if (ddi_copyin((void *)data
, &off
, sizeof (off
), flag
))
305 return (SET_ERROR(EFAULT
));
308 zfsvfs
= zp
->z_zfsvfs
;
312 /* offset parameter is in/out */
313 error
= zfs_holey(vp
, com
, &off
);
317 if (ddi_copyout(&off
, (void *)data
, sizeof (off
), flag
))
318 return (SET_ERROR(EFAULT
));
321 return (SET_ERROR(ENOTTY
));
325 * Utility functions to map and unmap a single physical page. These
326 * are used to manage the mappable copies of ZFS file data, and therefore
327 * do not update ref/mod bits.
330 zfs_map_page(page_t
*pp
, enum seg_rw rw
)
333 return (hat_kpm_mapin(pp
, 0));
334 ASSERT(rw
== S_READ
|| rw
== S_WRITE
);
335 return (ppmapin(pp
, PROT_READ
| ((rw
== S_WRITE
) ? PROT_WRITE
: 0),
340 zfs_unmap_page(page_t
*pp
, caddr_t addr
)
343 hat_kpm_mapout(pp
, 0, addr
);
350 * When a file is memory mapped, we must keep the IO data synchronized
351 * between the DMU cache and the memory mapped pages. What this means:
353 * On Write: If we find a memory mapped page, we write to *both*
354 * the page and the dmu buffer.
357 update_pages(vnode_t
*vp
, int64_t start
, int len
, objset_t
*os
, uint64_t oid
)
361 off
= start
& PAGEOFFSET
;
362 for (start
&= PAGEMASK
; len
> 0; start
+= PAGESIZE
) {
364 uint64_t nbytes
= MIN(PAGESIZE
- off
, len
);
366 if (pp
= page_lookup(vp
, start
, SE_SHARED
)) {
369 va
= zfs_map_page(pp
, S_WRITE
);
370 (void) dmu_read(os
, oid
, start
+off
, nbytes
, va
+off
,
372 zfs_unmap_page(pp
, va
);
381 * When a file is memory mapped, we must keep the IO data synchronized
382 * between the DMU cache and the memory mapped pages. What this means:
384 * On Read: We "read" preferentially from memory mapped pages,
385 * else we default from the dmu buffer.
387 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when
388 * the file is memory mapped.
391 mappedread(vnode_t
*vp
, int nbytes
, uio_t
*uio
)
393 znode_t
*zp
= VTOZ(vp
);
394 objset_t
*os
= zp
->z_zfsvfs
->z_os
;
399 start
= uio
->uio_loffset
;
400 off
= start
& PAGEOFFSET
;
401 for (start
&= PAGEMASK
; len
> 0; start
+= PAGESIZE
) {
403 uint64_t bytes
= MIN(PAGESIZE
- off
, len
);
405 if (pp
= page_lookup(vp
, start
, SE_SHARED
)) {
408 va
= zfs_map_page(pp
, S_READ
);
409 error
= uiomove(va
+ off
, bytes
, UIO_READ
, uio
);
410 zfs_unmap_page(pp
, va
);
413 error
= dmu_read_uio(os
, zp
->z_id
, uio
, bytes
);
423 offset_t zfs_read_chunk_size
= 1024 * 1024; /* Tunable */
426 * Read bytes from specified file into supplied buffer.
428 * IN: vp - vnode of file to be read from.
429 * uio - structure supplying read location, range info,
431 * ioflag - SYNC flags; used to provide FRSYNC semantics.
432 * cr - credentials of caller.
433 * ct - caller context
435 * OUT: uio - updated offset and range, buffer filled.
437 * RETURN: 0 on success, error code on failure.
440 * vp - atime updated if byte count > 0
444 zfs_read(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cr
, caller_context_t
*ct
)
446 znode_t
*zp
= VTOZ(vp
);
447 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
458 if (zp
->z_pflags
& ZFS_AV_QUARANTINED
) {
460 return (SET_ERROR(EACCES
));
464 * Validate file offset
466 if (uio
->uio_loffset
< (offset_t
)0) {
468 return (SET_ERROR(EINVAL
));
472 * Fasttrack empty reads
474 if (uio
->uio_resid
== 0) {
480 * Check for mandatory locks
482 if (MANDMODE(zp
->z_mode
)) {
483 if (error
= chklock(vp
, FREAD
,
484 uio
->uio_loffset
, uio
->uio_resid
, uio
->uio_fmode
, ct
)) {
491 * If we're in FRSYNC mode, sync out this znode before reading it.
493 if (ioflag
& FRSYNC
|| zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
494 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
497 * Lock the range against changes.
499 rl
= zfs_range_lock(zp
, uio
->uio_loffset
, uio
->uio_resid
, RL_READER
);
502 * If we are reading past end-of-file we can skip
503 * to the end; but we might still need to set atime.
505 if (uio
->uio_loffset
>= zp
->z_size
) {
510 ASSERT(uio
->uio_loffset
< zp
->z_size
);
511 n
= MIN(uio
->uio_resid
, zp
->z_size
- uio
->uio_loffset
);
513 if ((uio
->uio_extflg
== UIO_XUIO
) &&
514 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
)) {
516 int blksz
= zp
->z_blksz
;
517 uint64_t offset
= uio
->uio_loffset
;
519 xuio
= (xuio_t
*)uio
;
521 nblk
= (P2ROUNDUP(offset
+ n
, blksz
) - P2ALIGN(offset
,
524 ASSERT(offset
+ n
<= blksz
);
527 (void) dmu_xuio_init(xuio
, nblk
);
529 if (vn_has_cached_data(vp
)) {
531 * For simplicity, we always allocate a full buffer
532 * even if we only expect to read a portion of a block.
534 while (--nblk
>= 0) {
535 (void) dmu_xuio_add(xuio
,
536 dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
543 nbytes
= MIN(n
, zfs_read_chunk_size
-
544 P2PHASE(uio
->uio_loffset
, zfs_read_chunk_size
));
546 if (vn_has_cached_data(vp
))
547 error
= mappedread(vp
, nbytes
, uio
);
549 error
= dmu_read_uio(os
, zp
->z_id
, uio
, nbytes
);
551 /* convert checksum errors into IO errors */
553 error
= SET_ERROR(EIO
);
560 zfs_range_unlock(rl
);
562 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
568 * Write the bytes to a file.
570 * IN: vp - vnode of file to be written to.
571 * uio - structure supplying write location, range info,
573 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is
574 * set if in append mode.
575 * cr - credentials of caller.
576 * ct - caller context (NFS/CIFS fem monitor only)
578 * OUT: uio - updated offset and range.
580 * RETURN: 0 on success, error code on failure.
583 * vp - ctime|mtime updated if byte count > 0
588 zfs_write(vnode_t
*vp
, uio_t
*uio
, int ioflag
, cred_t
*cr
, caller_context_t
*ct
)
590 znode_t
*zp
= VTOZ(vp
);
591 rlim64_t limit
= uio
->uio_llimit
;
592 ssize_t start_resid
= uio
->uio_resid
;
596 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
601 int max_blksz
= zfsvfs
->z_max_blksz
;
604 iovec_t
*aiov
= NULL
;
607 int iovcnt
= uio
->uio_iovcnt
;
608 iovec_t
*iovp
= uio
->uio_iov
;
611 sa_bulk_attr_t bulk
[4];
612 uint64_t mtime
[2], ctime
[2];
615 * Fasttrack empty write
621 if (limit
== RLIM64_INFINITY
|| limit
> MAXOFFSET_T
)
627 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
628 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
629 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_SIZE(zfsvfs
), NULL
,
631 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
635 * If immutable or not appending then return EPERM
637 if ((zp
->z_pflags
& (ZFS_IMMUTABLE
| ZFS_READONLY
)) ||
638 ((zp
->z_pflags
& ZFS_APPENDONLY
) && !(ioflag
& FAPPEND
) &&
639 (uio
->uio_loffset
< zp
->z_size
))) {
641 return (SET_ERROR(EPERM
));
644 zilog
= zfsvfs
->z_log
;
647 * Validate file offset
649 woff
= ioflag
& FAPPEND
? zp
->z_size
: uio
->uio_loffset
;
652 return (SET_ERROR(EINVAL
));
656 * Check for mandatory locks before calling zfs_range_lock()
657 * in order to prevent a deadlock with locks set via fcntl().
659 if (MANDMODE((mode_t
)zp
->z_mode
) &&
660 (error
= chklock(vp
, FWRITE
, woff
, n
, uio
->uio_fmode
, ct
)) != 0) {
666 * Pre-fault the pages to ensure slow (eg NFS) pages
668 * Skip this if uio contains loaned arc_buf.
670 if ((uio
->uio_extflg
== UIO_XUIO
) &&
671 (((xuio_t
*)uio
)->xu_type
== UIOTYPE_ZEROCOPY
))
672 xuio
= (xuio_t
*)uio
;
674 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
677 * If in append mode, set the io offset pointer to eof.
679 if (ioflag
& FAPPEND
) {
681 * Obtain an appending range lock to guarantee file append
682 * semantics. We reset the write offset once we have the lock.
684 rl
= zfs_range_lock(zp
, 0, n
, RL_APPEND
);
686 if (rl
->r_len
== UINT64_MAX
) {
688 * We overlocked the file because this write will cause
689 * the file block size to increase.
690 * Note that zp_size cannot change with this lock held.
694 uio
->uio_loffset
= woff
;
697 * Note that if the file block size will change as a result of
698 * this write, then this range lock will lock the entire file
699 * so that we can re-write the block safely.
701 rl
= zfs_range_lock(zp
, woff
, n
, RL_WRITER
);
705 zfs_range_unlock(rl
);
707 return (SET_ERROR(EFBIG
));
710 if ((woff
+ n
) > limit
|| woff
> (limit
- n
))
713 /* Will this write extend the file length? */
714 write_eof
= (woff
+ n
> zp
->z_size
);
716 end_size
= MAX(zp
->z_size
, woff
+ n
);
719 * Write the file in reasonable size chunks. Each chunk is written
720 * in a separate transaction; this keeps the intent log records small
721 * and allows us to do more fine-grained space accounting.
725 woff
= uio
->uio_loffset
;
727 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
728 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
730 dmu_return_arcbuf(abuf
);
731 error
= SET_ERROR(EDQUOT
);
735 if (xuio
&& abuf
== NULL
) {
736 ASSERT(i_iov
< iovcnt
);
738 abuf
= dmu_xuio_arcbuf(xuio
, i_iov
);
739 dmu_xuio_clear(xuio
, i_iov
);
740 DTRACE_PROBE3(zfs_cp_write
, int, i_iov
,
741 iovec_t
*, aiov
, arc_buf_t
*, abuf
);
742 ASSERT((aiov
->iov_base
== abuf
->b_data
) ||
743 ((char *)aiov
->iov_base
- (char *)abuf
->b_data
+
744 aiov
->iov_len
== arc_buf_size(abuf
)));
746 } else if (abuf
== NULL
&& n
>= max_blksz
&&
747 woff
>= zp
->z_size
&&
748 P2PHASE(woff
, max_blksz
) == 0 &&
749 zp
->z_blksz
== max_blksz
) {
751 * This write covers a full block. "Borrow" a buffer
752 * from the dmu so that we can fill it before we enter
753 * a transaction. This avoids the possibility of
754 * holding up the transaction if the data copy hangs
755 * up on a pagefault (e.g., from an NFS server mapping).
759 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
761 ASSERT(abuf
!= NULL
);
762 ASSERT(arc_buf_size(abuf
) == max_blksz
);
763 if (error
= uiocopy(abuf
->b_data
, max_blksz
,
764 UIO_WRITE
, uio
, &cbytes
)) {
765 dmu_return_arcbuf(abuf
);
768 ASSERT(cbytes
== max_blksz
);
772 * Start a transaction.
774 tx
= dmu_tx_create(zfsvfs
->z_os
);
775 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
776 dmu_tx_hold_write(tx
, zp
->z_id
, woff
, MIN(n
, max_blksz
));
777 zfs_sa_upgrade_txholds(tx
, zp
);
778 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
780 if (error
== ERESTART
) {
787 dmu_return_arcbuf(abuf
);
792 * If zfs_range_lock() over-locked we grow the blocksize
793 * and then reduce the lock range. This will only happen
794 * on the first iteration since zfs_range_reduce() will
795 * shrink down r_len to the appropriate size.
797 if (rl
->r_len
== UINT64_MAX
) {
800 if (zp
->z_blksz
> max_blksz
) {
801 ASSERT(!ISP2(zp
->z_blksz
));
802 new_blksz
= MIN(end_size
, SPA_MAXBLOCKSIZE
);
804 new_blksz
= MIN(end_size
, max_blksz
);
806 zfs_grow_blocksize(zp
, new_blksz
, tx
);
807 zfs_range_reduce(rl
, woff
, n
);
811 * XXX - should we really limit each write to z_max_blksz?
812 * Perhaps we should use SPA_MAXBLOCKSIZE chunks?
814 nbytes
= MIN(n
, max_blksz
- P2PHASE(woff
, max_blksz
));
817 tx_bytes
= uio
->uio_resid
;
818 error
= dmu_write_uio_dbuf(sa_get_db(zp
->z_sa_hdl
),
820 tx_bytes
-= uio
->uio_resid
;
823 ASSERT(xuio
== NULL
|| tx_bytes
== aiov
->iov_len
);
825 * If this is not a full block write, but we are
826 * extending the file past EOF and this data starts
827 * block-aligned, use assign_arcbuf(). Otherwise,
828 * write via dmu_write().
830 if (tx_bytes
< max_blksz
&& (!write_eof
||
831 aiov
->iov_base
!= abuf
->b_data
)) {
833 dmu_write(zfsvfs
->z_os
, zp
->z_id
, woff
,
834 aiov
->iov_len
, aiov
->iov_base
, tx
);
835 dmu_return_arcbuf(abuf
);
836 xuio_stat_wbuf_copied();
838 ASSERT(xuio
|| tx_bytes
== max_blksz
);
839 dmu_assign_arcbuf(sa_get_db(zp
->z_sa_hdl
),
842 ASSERT(tx_bytes
<= uio
->uio_resid
);
843 uioskip(uio
, tx_bytes
);
845 if (tx_bytes
&& vn_has_cached_data(vp
)) {
846 update_pages(vp
, woff
,
847 tx_bytes
, zfsvfs
->z_os
, zp
->z_id
);
851 * If we made no progress, we're done. If we made even
852 * partial progress, update the znode and ZIL accordingly.
855 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
856 (void *)&zp
->z_size
, sizeof (uint64_t), tx
);
863 * Clear Set-UID/Set-GID bits on successful write if not
864 * privileged and at least one of the excute bits is set.
866 * It would be nice to to this after all writes have
867 * been done, but that would still expose the ISUID/ISGID
868 * to another app after the partial write is committed.
870 * Note: we don't call zfs_fuid_map_id() here because
871 * user 0 is not an ephemeral uid.
873 mutex_enter(&zp
->z_acl_lock
);
874 if ((zp
->z_mode
& (S_IXUSR
| (S_IXUSR
>> 3) |
875 (S_IXUSR
>> 6))) != 0 &&
876 (zp
->z_mode
& (S_ISUID
| S_ISGID
)) != 0 &&
877 secpolicy_vnode_setid_retain(cr
,
878 (zp
->z_mode
& S_ISUID
) != 0 && zp
->z_uid
== 0) != 0) {
880 zp
->z_mode
&= ~(S_ISUID
| S_ISGID
);
881 newmode
= zp
->z_mode
;
882 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_MODE(zfsvfs
),
883 (void *)&newmode
, sizeof (uint64_t), tx
);
885 mutex_exit(&zp
->z_acl_lock
);
887 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
891 * Update the file size (zp_size) if it has changed;
892 * account for possible concurrent updates.
894 while ((end_size
= zp
->z_size
) < uio
->uio_loffset
) {
895 (void) atomic_cas_64(&zp
->z_size
, end_size
,
900 * If we are replaying and eof is non zero then force
901 * the file size to the specified eof. Note, there's no
902 * concurrency during replay.
904 if (zfsvfs
->z_replay
&& zfsvfs
->z_replay_eof
!= 0)
905 zp
->z_size
= zfsvfs
->z_replay_eof
;
907 error
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
909 zfs_log_write(zilog
, tx
, TX_WRITE
, zp
, woff
, tx_bytes
, ioflag
);
914 ASSERT(tx_bytes
== nbytes
);
918 uio_prefaultpages(MIN(n
, max_blksz
), uio
);
921 zfs_range_unlock(rl
);
924 * If we're in replay mode, or we made no progress, return error.
925 * Otherwise, it's at least a partial write, so it's successful.
927 if (zfsvfs
->z_replay
|| uio
->uio_resid
== start_resid
) {
932 if (ioflag
& (FSYNC
| FDSYNC
) ||
933 zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
934 zil_commit(zilog
, zp
->z_id
);
941 zfs_get_done(zgd_t
*zgd
, int error
)
943 znode_t
*zp
= zgd
->zgd_private
;
944 objset_t
*os
= zp
->z_zfsvfs
->z_os
;
947 dmu_buf_rele(zgd
->zgd_db
, zgd
);
949 zfs_range_unlock(zgd
->zgd_rl
);
952 * Release the vnode asynchronously as we currently have the
953 * txg stopped from syncing.
955 VN_RELE_ASYNC(ZTOV(zp
), dsl_pool_vnrele_taskq(dmu_objset_pool(os
)));
957 if (error
== 0 && zgd
->zgd_bp
)
958 zil_add_block(zgd
->zgd_zilog
, zgd
->zgd_bp
);
960 kmem_free(zgd
, sizeof (zgd_t
));
964 static int zil_fault_io
= 0;
968 * Get data to generate a TX_WRITE intent log record.
971 zfs_get_data(void *arg
, lr_write_t
*lr
, char *buf
, zio_t
*zio
)
973 zfsvfs_t
*zfsvfs
= arg
;
974 objset_t
*os
= zfsvfs
->z_os
;
976 uint64_t object
= lr
->lr_foid
;
977 uint64_t offset
= lr
->lr_offset
;
978 uint64_t size
= lr
->lr_length
;
979 blkptr_t
*bp
= &lr
->lr_blkptr
;
988 * Nothing to do if the file has been removed
990 if (zfs_zget(zfsvfs
, object
, &zp
) != 0)
991 return (SET_ERROR(ENOENT
));
992 if (zp
->z_unlinked
) {
994 * Release the vnode asynchronously as we currently have the
995 * txg stopped from syncing.
997 VN_RELE_ASYNC(ZTOV(zp
),
998 dsl_pool_vnrele_taskq(dmu_objset_pool(os
)));
999 return (SET_ERROR(ENOENT
));
1002 zgd
= (zgd_t
*)kmem_zalloc(sizeof (zgd_t
), KM_SLEEP
);
1003 zgd
->zgd_zilog
= zfsvfs
->z_log
;
1004 zgd
->zgd_private
= zp
;
1007 * Write records come in two flavors: immediate and indirect.
1008 * For small writes it's cheaper to store the data with the
1009 * log record (immediate); for large writes it's cheaper to
1010 * sync the data and get a pointer to it (indirect) so that
1011 * we don't have to write the data twice.
1013 if (buf
!= NULL
) { /* immediate write */
1014 zgd
->zgd_rl
= zfs_range_lock(zp
, offset
, size
, RL_READER
);
1015 /* test for truncation needs to be done while range locked */
1016 if (offset
>= zp
->z_size
) {
1017 error
= SET_ERROR(ENOENT
);
1019 error
= dmu_read(os
, object
, offset
, size
, buf
,
1020 DMU_READ_NO_PREFETCH
);
1022 ASSERT(error
== 0 || error
== ENOENT
);
1023 } else { /* indirect write */
1025 * Have to lock the whole block to ensure when it's
1026 * written out and it's checksum is being calculated
1027 * that no one can change the data. We need to re-check
1028 * blocksize after we get the lock in case it's changed!
1033 blkoff
= ISP2(size
) ? P2PHASE(offset
, size
) : offset
;
1035 zgd
->zgd_rl
= zfs_range_lock(zp
, offset
, size
,
1037 if (zp
->z_blksz
== size
)
1040 zfs_range_unlock(zgd
->zgd_rl
);
1042 /* test for truncation needs to be done while range locked */
1043 if (lr
->lr_offset
>= zp
->z_size
)
1044 error
= SET_ERROR(ENOENT
);
1047 error
= SET_ERROR(EIO
);
1052 error
= dmu_buf_hold(os
, object
, offset
, zgd
, &db
,
1053 DMU_READ_NO_PREFETCH
);
1056 blkptr_t
*obp
= dmu_buf_get_blkptr(db
);
1058 ASSERT(BP_IS_HOLE(bp
));
1065 ASSERT(db
->db_offset
== offset
);
1066 ASSERT(db
->db_size
== size
);
1068 error
= dmu_sync(zio
, lr
->lr_common
.lrc_txg
,
1070 ASSERT(error
|| lr
->lr_length
<= zp
->z_blksz
);
1073 * On success, we need to wait for the write I/O
1074 * initiated by dmu_sync() to complete before we can
1075 * release this dbuf. We will finish everything up
1076 * in the zfs_get_done() callback.
1081 if (error
== EALREADY
) {
1082 lr
->lr_common
.lrc_txtype
= TX_WRITE2
;
1088 zfs_get_done(zgd
, error
);
1095 zfs_access(vnode_t
*vp
, int mode
, int flag
, cred_t
*cr
,
1096 caller_context_t
*ct
)
1098 znode_t
*zp
= VTOZ(vp
);
1099 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
1105 if (flag
& V_ACE_MASK
)
1106 error
= zfs_zaccess(zp
, mode
, flag
, B_FALSE
, cr
);
1108 error
= zfs_zaccess_rwx(zp
, mode
, flag
, cr
);
1115 * If vnode is for a device return a specfs vnode instead.
1118 specvp_check(vnode_t
**vpp
, cred_t
*cr
)
1122 if (IS_DEVVP(*vpp
)) {
1125 svp
= specvp(*vpp
, (*vpp
)->v_rdev
, (*vpp
)->v_type
, cr
);
1128 error
= SET_ERROR(ENOSYS
);
1136 * Lookup an entry in a directory, or an extended attribute directory.
1137 * If it exists, return a held vnode reference for it.
1139 * IN: dvp - vnode of directory to search.
1140 * nm - name of entry to lookup.
1141 * pnp - full pathname to lookup [UNUSED].
1142 * flags - LOOKUP_XATTR set if looking for an attribute.
1143 * rdir - root directory vnode [UNUSED].
1144 * cr - credentials of caller.
1145 * ct - caller context
1146 * direntflags - directory lookup flags
1147 * realpnp - returned pathname.
1149 * OUT: vpp - vnode of located entry, NULL if not found.
1151 * RETURN: 0 on success, error code on failure.
1158 zfs_lookup(vnode_t
*dvp
, char *nm
, vnode_t
**vpp
, struct pathname
*pnp
,
1159 int flags
, vnode_t
*rdir
, cred_t
*cr
, caller_context_t
*ct
,
1160 int *direntflags
, pathname_t
*realpnp
)
1162 znode_t
*zdp
= VTOZ(dvp
);
1163 zfsvfs_t
*zfsvfs
= zdp
->z_zfsvfs
;
1167 if (!(flags
& (LOOKUP_XATTR
| FIGNORECASE
))) {
1169 if (dvp
->v_type
!= VDIR
) {
1170 return (SET_ERROR(ENOTDIR
));
1171 } else if (zdp
->z_sa_hdl
== NULL
) {
1172 return (SET_ERROR(EIO
));
1175 if (nm
[0] == 0 || (nm
[0] == '.' && nm
[1] == '\0')) {
1176 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1184 vnode_t
*tvp
= dnlc_lookup(dvp
, nm
);
1187 error
= zfs_fastaccesschk_execute(zdp
, cr
);
1192 if (tvp
== DNLC_NO_VNODE
) {
1194 return (SET_ERROR(ENOENT
));
1197 return (specvp_check(vpp
, cr
));
1203 DTRACE_PROBE2(zfs__fastpath__lookup__miss
, vnode_t
*, dvp
, char *, nm
);
1210 if (flags
& LOOKUP_XATTR
) {
1212 * If the xattr property is off, refuse the lookup request.
1214 if (!(zfsvfs
->z_vfs
->vfs_flag
& VFS_XATTR
)) {
1216 return (SET_ERROR(EINVAL
));
1220 * We don't allow recursive attributes..
1221 * Maybe someday we will.
1223 if (zdp
->z_pflags
& ZFS_XATTR
) {
1225 return (SET_ERROR(EINVAL
));
1228 if (error
= zfs_get_xattrdir(VTOZ(dvp
), vpp
, cr
, flags
)) {
1234 * Do we have permission to get into attribute directory?
1237 if (error
= zfs_zaccess(VTOZ(*vpp
), ACE_EXECUTE
, 0,
1247 if (dvp
->v_type
!= VDIR
) {
1249 return (SET_ERROR(ENOTDIR
));
1253 * Check accessibility of directory.
1256 if (error
= zfs_zaccess(zdp
, ACE_EXECUTE
, 0, B_FALSE
, cr
)) {
1261 if (zfsvfs
->z_utf8
&& u8_validate(nm
, strlen(nm
),
1262 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1264 return (SET_ERROR(EILSEQ
));
1267 error
= zfs_dirlook(zdp
, nm
, vpp
, flags
, direntflags
, realpnp
);
1269 error
= specvp_check(vpp
, cr
);
1276 * Attempt to create a new entry in a directory. If the entry
1277 * already exists, truncate the file if permissible, else return
1278 * an error. Return the vp of the created or trunc'd file.
1280 * IN: dvp - vnode of directory to put new file entry in.
1281 * name - name of new file entry.
1282 * vap - attributes of new file.
1283 * excl - flag indicating exclusive or non-exclusive mode.
1284 * mode - mode to open file with.
1285 * cr - credentials of caller.
1286 * flag - large file flag [UNUSED].
1287 * ct - caller context
1288 * vsecp - ACL to be set
1290 * OUT: vpp - vnode of created or trunc'd entry.
1292 * RETURN: 0 on success, error code on failure.
1295 * dvp - ctime|mtime updated if new entry created
1296 * vp - ctime|mtime always, atime if new
1301 zfs_create(vnode_t
*dvp
, char *name
, vattr_t
*vap
, vcexcl_t excl
,
1302 int mode
, vnode_t
**vpp
, cred_t
*cr
, int flag
, caller_context_t
*ct
,
1305 znode_t
*zp
, *dzp
= VTOZ(dvp
);
1306 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1314 gid_t gid
= crgetgid(cr
);
1315 zfs_acl_ids_t acl_ids
;
1316 boolean_t fuid_dirtied
;
1317 boolean_t have_acl
= B_FALSE
;
1320 * If we have an ephemeral id, ACL, or XVATTR then
1321 * make sure file system is at proper version
1324 ksid
= crgetsid(cr
, KSID_OWNER
);
1326 uid
= ksid_getid(ksid
);
1330 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1331 (vsecp
|| (vap
->va_mask
& AT_XVATTR
) ||
1332 IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1333 return (SET_ERROR(EINVAL
));
1338 zilog
= zfsvfs
->z_log
;
1340 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
1341 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1343 return (SET_ERROR(EILSEQ
));
1346 if (vap
->va_mask
& AT_XVATTR
) {
1347 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1348 crgetuid(cr
), cr
, vap
->va_type
)) != 0) {
1356 if ((vap
->va_mode
& VSVTX
) && secpolicy_vnode_stky_modify(cr
))
1357 vap
->va_mode
&= ~VSVTX
;
1359 if (*name
== '\0') {
1361 * Null component name refers to the directory itself.
1368 /* possible VN_HOLD(zp) */
1371 if (flag
& FIGNORECASE
)
1374 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1378 zfs_acl_ids_free(&acl_ids
);
1379 if (strcmp(name
, "..") == 0)
1380 error
= SET_ERROR(EISDIR
);
1390 * Create a new file object and update the directory
1393 if (error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
)) {
1395 zfs_acl_ids_free(&acl_ids
);
1400 * We only support the creation of regular files in
1401 * extended attribute directories.
1404 if ((dzp
->z_pflags
& ZFS_XATTR
) &&
1405 (vap
->va_type
!= VREG
)) {
1407 zfs_acl_ids_free(&acl_ids
);
1408 error
= SET_ERROR(EINVAL
);
1412 if (!have_acl
&& (error
= zfs_acl_ids_create(dzp
, 0, vap
,
1413 cr
, vsecp
, &acl_ids
)) != 0)
1417 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1418 zfs_acl_ids_free(&acl_ids
);
1419 error
= SET_ERROR(EDQUOT
);
1423 tx
= dmu_tx_create(os
);
1425 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1426 ZFS_SA_BASE_ATTR_SIZE
);
1428 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1430 zfs_fuid_txhold(zfsvfs
, tx
);
1431 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
1432 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
1433 if (!zfsvfs
->z_use_sa
&&
1434 acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1435 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
1436 0, acl_ids
.z_aclp
->z_acl_bytes
);
1438 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
1440 zfs_dirent_unlock(dl
);
1441 if (error
== ERESTART
) {
1446 zfs_acl_ids_free(&acl_ids
);
1451 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1454 zfs_fuid_sync(zfsvfs
, tx
);
1456 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1457 txtype
= zfs_log_create_txtype(Z_FILE
, vsecp
, vap
);
1458 if (flag
& FIGNORECASE
)
1460 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, name
,
1461 vsecp
, acl_ids
.z_fuidp
, vap
);
1462 zfs_acl_ids_free(&acl_ids
);
1465 int aflags
= (flag
& FAPPEND
) ? V_APPEND
: 0;
1468 zfs_acl_ids_free(&acl_ids
);
1472 * A directory entry already exists for this name.
1475 * Can't truncate an existing file if in exclusive mode.
1478 error
= SET_ERROR(EEXIST
);
1482 * Can't open a directory for writing.
1484 if ((ZTOV(zp
)->v_type
== VDIR
) && (mode
& S_IWRITE
)) {
1485 error
= SET_ERROR(EISDIR
);
1489 * Verify requested access to file.
1491 if (mode
&& (error
= zfs_zaccess_rwx(zp
, mode
, aflags
, cr
))) {
1495 mutex_enter(&dzp
->z_lock
);
1497 mutex_exit(&dzp
->z_lock
);
1500 * Truncate regular files if requested.
1502 if ((ZTOV(zp
)->v_type
== VREG
) &&
1503 (vap
->va_mask
& AT_SIZE
) && (vap
->va_size
== 0)) {
1504 /* we can't hold any locks when calling zfs_freesp() */
1505 zfs_dirent_unlock(dl
);
1507 error
= zfs_freesp(zp
, 0, 0, mode
, TRUE
);
1509 vnevent_create(ZTOV(zp
), ct
);
1516 zfs_dirent_unlock(dl
);
1523 error
= specvp_check(vpp
, cr
);
1526 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1527 zil_commit(zilog
, 0);
1534 * Remove an entry from a directory.
1536 * IN: dvp - vnode of directory to remove entry from.
1537 * name - name of entry to remove.
1538 * cr - credentials of caller.
1539 * ct - caller context
1540 * flags - case flags
1542 * RETURN: 0 on success, error code on failure.
1546 * vp - ctime (if nlink > 0)
1549 uint64_t null_xattr
= 0;
1553 zfs_remove(vnode_t
*dvp
, char *name
, cred_t
*cr
, caller_context_t
*ct
,
1556 znode_t
*zp
, *dzp
= VTOZ(dvp
);
1559 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1561 uint64_t acl_obj
, xattr_obj
;
1562 uint64_t xattr_obj_unlinked
= 0;
1566 boolean_t may_delete_now
, delete_now
= FALSE
;
1567 boolean_t unlinked
, toobig
= FALSE
;
1569 pathname_t
*realnmp
= NULL
;
1576 zilog
= zfsvfs
->z_log
;
1578 if (flags
& FIGNORECASE
) {
1588 * Attempt to lock directory; fail if entry doesn't exist.
1590 if (error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1600 if (error
= zfs_zaccess_delete(dzp
, zp
, cr
)) {
1605 * Need to use rmdir for removing directories.
1607 if (vp
->v_type
== VDIR
) {
1608 error
= SET_ERROR(EPERM
);
1612 vnevent_remove(vp
, dvp
, name
, ct
);
1615 dnlc_remove(dvp
, realnmp
->pn_buf
);
1617 dnlc_remove(dvp
, name
);
1619 mutex_enter(&vp
->v_lock
);
1620 may_delete_now
= vp
->v_count
== 1 && !vn_has_cached_data(vp
);
1621 mutex_exit(&vp
->v_lock
);
1624 * We may delete the znode now, or we may put it in the unlinked set;
1625 * it depends on whether we're the last link, and on whether there are
1626 * other holds on the vnode. So we dmu_tx_hold() the right things to
1627 * allow for either case.
1630 tx
= dmu_tx_create(zfsvfs
->z_os
);
1631 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
1632 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
1633 zfs_sa_upgrade_txholds(tx
, zp
);
1634 zfs_sa_upgrade_txholds(tx
, dzp
);
1635 if (may_delete_now
) {
1637 zp
->z_size
> zp
->z_blksz
* DMU_MAX_DELETEBLKCNT
;
1638 /* if the file is too big, only hold_free a token amount */
1639 dmu_tx_hold_free(tx
, zp
->z_id
, 0,
1640 (toobig
? DMU_MAX_ACCESS
: DMU_OBJECT_END
));
1643 /* are there any extended attributes? */
1644 error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1645 &xattr_obj
, sizeof (xattr_obj
));
1646 if (error
== 0 && xattr_obj
) {
1647 error
= zfs_zget(zfsvfs
, xattr_obj
, &xzp
);
1649 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
1650 dmu_tx_hold_sa(tx
, xzp
->z_sa_hdl
, B_FALSE
);
1653 mutex_enter(&zp
->z_lock
);
1654 if ((acl_obj
= zfs_external_acl(zp
)) != 0 && may_delete_now
)
1655 dmu_tx_hold_free(tx
, acl_obj
, 0, DMU_OBJECT_END
);
1656 mutex_exit(&zp
->z_lock
);
1658 /* charge as an update -- would be nice not to charge at all */
1659 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
1661 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
1663 zfs_dirent_unlock(dl
);
1667 if (error
== ERESTART
) {
1680 * Remove the directory entry.
1682 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, &unlinked
);
1692 * Hold z_lock so that we can make sure that the ACL obj
1693 * hasn't changed. Could have been deleted due to
1696 mutex_enter(&zp
->z_lock
);
1697 mutex_enter(&vp
->v_lock
);
1698 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
1699 &xattr_obj_unlinked
, sizeof (xattr_obj_unlinked
));
1700 delete_now
= may_delete_now
&& !toobig
&&
1701 vp
->v_count
== 1 && !vn_has_cached_data(vp
) &&
1702 xattr_obj
== xattr_obj_unlinked
&& zfs_external_acl(zp
) ==
1704 mutex_exit(&vp
->v_lock
);
1708 if (xattr_obj_unlinked
) {
1709 ASSERT3U(xzp
->z_links
, ==, 2);
1710 mutex_enter(&xzp
->z_lock
);
1711 xzp
->z_unlinked
= 1;
1713 error
= sa_update(xzp
->z_sa_hdl
, SA_ZPL_LINKS(zfsvfs
),
1714 &xzp
->z_links
, sizeof (xzp
->z_links
), tx
);
1715 ASSERT3U(error
, ==, 0);
1716 mutex_exit(&xzp
->z_lock
);
1717 zfs_unlinked_add(xzp
, tx
);
1720 error
= sa_remove(zp
->z_sa_hdl
,
1721 SA_ZPL_XATTR(zfsvfs
), tx
);
1723 error
= sa_update(zp
->z_sa_hdl
,
1724 SA_ZPL_XATTR(zfsvfs
), &null_xattr
,
1725 sizeof (uint64_t), tx
);
1728 mutex_enter(&vp
->v_lock
);
1730 ASSERT0(vp
->v_count
);
1731 mutex_exit(&vp
->v_lock
);
1732 mutex_exit(&zp
->z_lock
);
1733 zfs_znode_delete(zp
, tx
);
1734 } else if (unlinked
) {
1735 mutex_exit(&zp
->z_lock
);
1736 zfs_unlinked_add(zp
, tx
);
1740 if (flags
& FIGNORECASE
)
1742 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, obj
);
1749 zfs_dirent_unlock(dl
);
1756 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1757 zil_commit(zilog
, 0);
1764 * Create a new directory and insert it into dvp using the name
1765 * provided. Return a pointer to the inserted directory.
1767 * IN: dvp - vnode of directory to add subdir to.
1768 * dirname - name of new directory.
1769 * vap - attributes of new directory.
1770 * cr - credentials of caller.
1771 * ct - caller context
1772 * flags - case flags
1773 * vsecp - ACL to be set
1775 * OUT: vpp - vnode of created directory.
1777 * RETURN: 0 on success, error code on failure.
1780 * dvp - ctime|mtime updated
1781 * vp - ctime|mtime|atime updated
1785 zfs_mkdir(vnode_t
*dvp
, char *dirname
, vattr_t
*vap
, vnode_t
**vpp
, cred_t
*cr
,
1786 caller_context_t
*ct
, int flags
, vsecattr_t
*vsecp
)
1788 znode_t
*zp
, *dzp
= VTOZ(dvp
);
1789 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1798 gid_t gid
= crgetgid(cr
);
1799 zfs_acl_ids_t acl_ids
;
1800 boolean_t fuid_dirtied
;
1802 ASSERT(vap
->va_type
== VDIR
);
1805 * If we have an ephemeral id, ACL, or XVATTR then
1806 * make sure file system is at proper version
1809 ksid
= crgetsid(cr
, KSID_OWNER
);
1811 uid
= ksid_getid(ksid
);
1814 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
1815 (vsecp
|| (vap
->va_mask
& AT_XVATTR
) ||
1816 IS_EPHEMERAL(uid
) || IS_EPHEMERAL(gid
)))
1817 return (SET_ERROR(EINVAL
));
1821 zilog
= zfsvfs
->z_log
;
1823 if (dzp
->z_pflags
& ZFS_XATTR
) {
1825 return (SET_ERROR(EINVAL
));
1828 if (zfsvfs
->z_utf8
&& u8_validate(dirname
,
1829 strlen(dirname
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
1831 return (SET_ERROR(EILSEQ
));
1833 if (flags
& FIGNORECASE
)
1836 if (vap
->va_mask
& AT_XVATTR
) {
1837 if ((error
= secpolicy_xvattr((xvattr_t
*)vap
,
1838 crgetuid(cr
), cr
, vap
->va_type
)) != 0) {
1844 if ((error
= zfs_acl_ids_create(dzp
, 0, vap
, cr
,
1845 vsecp
, &acl_ids
)) != 0) {
1850 * First make sure the new directory doesn't exist.
1852 * Existence is checked first to make sure we don't return
1853 * EACCES instead of EEXIST which can cause some applications
1859 if (error
= zfs_dirent_lock(&dl
, dzp
, dirname
, &zp
, zf
,
1861 zfs_acl_ids_free(&acl_ids
);
1866 if (error
= zfs_zaccess(dzp
, ACE_ADD_SUBDIRECTORY
, 0, B_FALSE
, cr
)) {
1867 zfs_acl_ids_free(&acl_ids
);
1868 zfs_dirent_unlock(dl
);
1873 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
1874 zfs_acl_ids_free(&acl_ids
);
1875 zfs_dirent_unlock(dl
);
1877 return (SET_ERROR(EDQUOT
));
1881 * Add a new entry to the directory.
1883 tx
= dmu_tx_create(zfsvfs
->z_os
);
1884 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, dirname
);
1885 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1886 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
1888 zfs_fuid_txhold(zfsvfs
, tx
);
1889 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
1890 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
1891 acl_ids
.z_aclp
->z_acl_bytes
);
1894 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
1895 ZFS_SA_BASE_ATTR_SIZE
);
1897 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
1899 zfs_dirent_unlock(dl
);
1900 if (error
== ERESTART
) {
1905 zfs_acl_ids_free(&acl_ids
);
1914 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
1917 zfs_fuid_sync(zfsvfs
, tx
);
1920 * Now put new name in parent dir.
1922 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
1926 txtype
= zfs_log_create_txtype(Z_DIR
, vsecp
, vap
);
1927 if (flags
& FIGNORECASE
)
1929 zfs_log_create(zilog
, tx
, txtype
, dzp
, zp
, dirname
, vsecp
,
1930 acl_ids
.z_fuidp
, vap
);
1932 zfs_acl_ids_free(&acl_ids
);
1936 zfs_dirent_unlock(dl
);
1938 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
1939 zil_commit(zilog
, 0);
1946 * Remove a directory subdir entry. If the current working
1947 * directory is the same as the subdir to be removed, the
1950 * IN: dvp - vnode of directory to remove from.
1951 * name - name of directory to be removed.
1952 * cwd - vnode of current working directory.
1953 * cr - credentials of caller.
1954 * ct - caller context
1955 * flags - case flags
1957 * RETURN: 0 on success, error code on failure.
1960 * dvp - ctime|mtime updated
1964 zfs_rmdir(vnode_t
*dvp
, char *name
, vnode_t
*cwd
, cred_t
*cr
,
1965 caller_context_t
*ct
, int flags
)
1967 znode_t
*dzp
= VTOZ(dvp
);
1970 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
1979 zilog
= zfsvfs
->z_log
;
1981 if (flags
& FIGNORECASE
)
1987 * Attempt to lock directory; fail if entry doesn't exist.
1989 if (error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
,
1997 if (error
= zfs_zaccess_delete(dzp
, zp
, cr
)) {
2001 if (vp
->v_type
!= VDIR
) {
2002 error
= SET_ERROR(ENOTDIR
);
2007 error
= SET_ERROR(EINVAL
);
2011 vnevent_rmdir(vp
, dvp
, name
, ct
);
2014 * Grab a lock on the directory to make sure that noone is
2015 * trying to add (or lookup) entries while we are removing it.
2017 rw_enter(&zp
->z_name_lock
, RW_WRITER
);
2020 * Grab a lock on the parent pointer to make sure we play well
2021 * with the treewalk and directory rename code.
2023 rw_enter(&zp
->z_parent_lock
, RW_WRITER
);
2025 tx
= dmu_tx_create(zfsvfs
->z_os
);
2026 dmu_tx_hold_zap(tx
, dzp
->z_id
, FALSE
, name
);
2027 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
2028 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
2029 zfs_sa_upgrade_txholds(tx
, zp
);
2030 zfs_sa_upgrade_txholds(tx
, dzp
);
2031 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
2033 rw_exit(&zp
->z_parent_lock
);
2034 rw_exit(&zp
->z_name_lock
);
2035 zfs_dirent_unlock(dl
);
2037 if (error
== ERESTART
) {
2047 error
= zfs_link_destroy(dl
, zp
, tx
, zflg
, NULL
);
2050 uint64_t txtype
= TX_RMDIR
;
2051 if (flags
& FIGNORECASE
)
2053 zfs_log_remove(zilog
, tx
, txtype
, dzp
, name
, ZFS_NO_OBJECT
);
2058 rw_exit(&zp
->z_parent_lock
);
2059 rw_exit(&zp
->z_name_lock
);
2061 zfs_dirent_unlock(dl
);
2065 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
2066 zil_commit(zilog
, 0);
2073 * Read as many directory entries as will fit into the provided
2074 * buffer from the given directory cursor position (specified in
2075 * the uio structure).
2077 * IN: vp - vnode of directory to read.
2078 * uio - structure supplying read location, range info,
2079 * and return buffer.
2080 * cr - credentials of caller.
2081 * ct - caller context
2082 * flags - case flags
2084 * OUT: uio - updated offset and range, buffer filled.
2085 * eofp - set to true if end-of-file detected.
2087 * RETURN: 0 on success, error code on failure.
2090 * vp - atime updated
2092 * Note that the low 4 bits of the cookie returned by zap is always zero.
2093 * This allows us to use the low range for "special" directory entries:
2094 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem,
2095 * we use the offset 2 for the '.zfs' directory.
2099 zfs_readdir(vnode_t
*vp
, uio_t
*uio
, cred_t
*cr
, int *eofp
,
2100 caller_context_t
*ct
, int flags
)
2102 znode_t
*zp
= VTOZ(vp
);
2106 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2111 zap_attribute_t zap
;
2112 uint_t bytes_wanted
;
2113 uint64_t offset
; /* must be unsigned; checks for < 1 */
2119 boolean_t check_sysattrs
;
2124 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
2125 &parent
, sizeof (parent
))) != 0) {
2131 * If we are not given an eof variable,
2138 * Check for valid iov_len.
2140 if (uio
->uio_iov
->iov_len
<= 0) {
2142 return (SET_ERROR(EINVAL
));
2146 * Quit if directory has been removed (posix)
2148 if ((*eofp
= zp
->z_unlinked
) != 0) {
2155 offset
= uio
->uio_loffset
;
2156 prefetch
= zp
->z_zn_prefetch
;
2159 * Initialize the iterator cursor.
2163 * Start iteration from the beginning of the directory.
2165 zap_cursor_init(&zc
, os
, zp
->z_id
);
2168 * The offset is a serialized cursor.
2170 zap_cursor_init_serialized(&zc
, os
, zp
->z_id
, offset
);
2174 * Get space to change directory entries into fs independent format.
2176 iovp
= uio
->uio_iov
;
2177 bytes_wanted
= iovp
->iov_len
;
2178 if (uio
->uio_segflg
!= UIO_SYSSPACE
|| uio
->uio_iovcnt
!= 1) {
2179 bufsize
= bytes_wanted
;
2180 outbuf
= kmem_alloc(bufsize
, KM_SLEEP
);
2181 odp
= (struct dirent64
*)outbuf
;
2183 bufsize
= bytes_wanted
;
2185 odp
= (struct dirent64
*)iovp
->iov_base
;
2187 eodp
= (struct edirent
*)odp
;
2190 * If this VFS supports the system attribute view interface; and
2191 * we're looking at an extended attribute directory; and we care
2192 * about normalization conflicts on this vfs; then we must check
2193 * for normalization conflicts with the sysattr name space.
2195 check_sysattrs
= vfs_has_feature(vp
->v_vfsp
, VFSFT_SYSATTR_VIEWS
) &&
2196 (vp
->v_flag
& V_XATTRDIR
) && zfsvfs
->z_norm
&&
2197 (flags
& V_RDDIR_ENTFLAGS
);
2200 * Transform to file-system independent format
2203 while (outcount
< bytes_wanted
) {
2206 off64_t
*next
= NULL
;
2209 * Special case `.', `..', and `.zfs'.
2212 (void) strcpy(zap
.za_name
, ".");
2213 zap
.za_normalization_conflict
= 0;
2215 } else if (offset
== 1) {
2216 (void) strcpy(zap
.za_name
, "..");
2217 zap
.za_normalization_conflict
= 0;
2219 } else if (offset
== 2 && zfs_show_ctldir(zp
)) {
2220 (void) strcpy(zap
.za_name
, ZFS_CTLDIR_NAME
);
2221 zap
.za_normalization_conflict
= 0;
2222 objnum
= ZFSCTL_INO_ROOT
;
2227 if (error
= zap_cursor_retrieve(&zc
, &zap
)) {
2228 if ((*eofp
= (error
== ENOENT
)) != 0)
2234 if (zap
.za_integer_length
!= 8 ||
2235 zap
.za_num_integers
!= 1) {
2236 cmn_err(CE_WARN
, "zap_readdir: bad directory "
2237 "entry, obj = %lld, offset = %lld\n",
2238 (u_longlong_t
)zp
->z_id
,
2239 (u_longlong_t
)offset
);
2240 error
= SET_ERROR(ENXIO
);
2244 objnum
= ZFS_DIRENT_OBJ(zap
.za_first_integer
);
2246 * MacOS X can extract the object type here such as:
2247 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer);
2250 if (check_sysattrs
&& !zap
.za_normalization_conflict
) {
2251 zap
.za_normalization_conflict
=
2252 xattr_sysattr_casechk(zap
.za_name
);
2256 if (flags
& V_RDDIR_ACCFILTER
) {
2258 * If we have no access at all, don't include
2259 * this entry in the returned information
2262 if (zfs_zget(zp
->z_zfsvfs
, objnum
, &ezp
) != 0)
2264 if (!zfs_has_access(ezp
, cr
)) {
2271 if (flags
& V_RDDIR_ENTFLAGS
)
2272 reclen
= EDIRENT_RECLEN(strlen(zap
.za_name
));
2274 reclen
= DIRENT64_RECLEN(strlen(zap
.za_name
));
2277 * Will this entry fit in the buffer?
2279 if (outcount
+ reclen
> bufsize
) {
2281 * Did we manage to fit anything in the buffer?
2284 error
= SET_ERROR(EINVAL
);
2289 if (flags
& V_RDDIR_ENTFLAGS
) {
2291 * Add extended flag entry:
2293 eodp
->ed_ino
= objnum
;
2294 eodp
->ed_reclen
= reclen
;
2295 /* NOTE: ed_off is the offset for the *next* entry */
2296 next
= &(eodp
->ed_off
);
2297 eodp
->ed_eflags
= zap
.za_normalization_conflict
?
2298 ED_CASE_CONFLICT
: 0;
2299 (void) strncpy(eodp
->ed_name
, zap
.za_name
,
2300 EDIRENT_NAMELEN(reclen
));
2301 eodp
= (edirent_t
*)((intptr_t)eodp
+ reclen
);
2306 odp
->d_ino
= objnum
;
2307 odp
->d_reclen
= reclen
;
2308 /* NOTE: d_off is the offset for the *next* entry */
2309 next
= &(odp
->d_off
);
2310 (void) strncpy(odp
->d_name
, zap
.za_name
,
2311 DIRENT64_NAMELEN(reclen
));
2312 odp
= (dirent64_t
*)((intptr_t)odp
+ reclen
);
2316 ASSERT(outcount
<= bufsize
);
2318 /* Prefetch znode */
2320 dmu_prefetch(os
, objnum
, 0, 0);
2324 * Move to the next entry, fill in the previous offset.
2326 if (offset
> 2 || (offset
== 2 && !zfs_show_ctldir(zp
))) {
2327 zap_cursor_advance(&zc
);
2328 offset
= zap_cursor_serialize(&zc
);
2335 zp
->z_zn_prefetch
= B_FALSE
; /* a lookup will re-enable pre-fetching */
2337 if (uio
->uio_segflg
== UIO_SYSSPACE
&& uio
->uio_iovcnt
== 1) {
2338 iovp
->iov_base
+= outcount
;
2339 iovp
->iov_len
-= outcount
;
2340 uio
->uio_resid
-= outcount
;
2341 } else if (error
= uiomove(outbuf
, (long)outcount
, UIO_READ
, uio
)) {
2343 * Reset the pointer.
2345 offset
= uio
->uio_loffset
;
2349 zap_cursor_fini(&zc
);
2350 if (uio
->uio_segflg
!= UIO_SYSSPACE
|| uio
->uio_iovcnt
!= 1)
2351 kmem_free(outbuf
, bufsize
);
2353 if (error
== ENOENT
)
2356 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
2358 uio
->uio_loffset
= offset
;
2363 ulong_t zfs_fsync_sync_cnt
= 4;
2366 zfs_fsync(vnode_t
*vp
, int syncflag
, cred_t
*cr
, caller_context_t
*ct
)
2368 znode_t
*zp
= VTOZ(vp
);
2369 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2372 * Regardless of whether this is required for standards conformance,
2373 * this is the logical behavior when fsync() is called on a file with
2374 * dirty pages. We use B_ASYNC since the ZIL transactions are already
2375 * going to be pushed out as part of the zil_commit().
2377 if (vn_has_cached_data(vp
) && !(syncflag
& FNODSYNC
) &&
2378 (vp
->v_type
== VREG
) && !(IS_SWAPVP(vp
)))
2379 (void) VOP_PUTPAGE(vp
, (offset_t
)0, (size_t)0, B_ASYNC
, cr
, ct
);
2381 (void) tsd_set(zfs_fsyncer_key
, (void *)zfs_fsync_sync_cnt
);
2383 if (zfsvfs
->z_os
->os_sync
!= ZFS_SYNC_DISABLED
) {
2386 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
2394 * Get the requested file attributes and place them in the provided
2397 * IN: vp - vnode of file.
2398 * vap - va_mask identifies requested attributes.
2399 * If AT_XVATTR set, then optional attrs are requested
2400 * flags - ATTR_NOACLCHECK (CIFS server context)
2401 * cr - credentials of caller.
2402 * ct - caller context
2404 * OUT: vap - attribute values.
2406 * RETURN: 0 (always succeeds).
2410 zfs_getattr(vnode_t
*vp
, vattr_t
*vap
, int flags
, cred_t
*cr
,
2411 caller_context_t
*ct
)
2413 znode_t
*zp
= VTOZ(vp
);
2414 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2417 uint64_t mtime
[2], ctime
[2];
2418 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2419 xoptattr_t
*xoap
= NULL
;
2420 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2421 sa_bulk_attr_t bulk
[2];
2427 zfs_fuid_map_ids(zp
, cr
, &vap
->va_uid
, &vap
->va_gid
);
2429 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
, &mtime
, 16);
2430 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
, &ctime
, 16);
2432 if ((error
= sa_bulk_lookup(zp
->z_sa_hdl
, bulk
, count
)) != 0) {
2438 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES.
2439 * Also, if we are the owner don't bother, since owner should
2440 * always be allowed to read basic attributes of file.
2442 if (!(zp
->z_pflags
& ZFS_ACL_TRIVIAL
) &&
2443 (vap
->va_uid
!= crgetuid(cr
))) {
2444 if (error
= zfs_zaccess(zp
, ACE_READ_ATTRIBUTES
, 0,
2452 * Return all attributes. It's cheaper to provide the answer
2453 * than to determine whether we were asked the question.
2456 mutex_enter(&zp
->z_lock
);
2457 vap
->va_type
= vp
->v_type
;
2458 vap
->va_mode
= zp
->z_mode
& MODEMASK
;
2459 vap
->va_fsid
= zp
->z_zfsvfs
->z_vfs
->vfs_dev
;
2460 vap
->va_nodeid
= zp
->z_id
;
2461 if ((vp
->v_flag
& VROOT
) && zfs_show_ctldir(zp
))
2462 links
= zp
->z_links
+ 1;
2464 links
= zp
->z_links
;
2465 vap
->va_nlink
= MIN(links
, UINT32_MAX
); /* nlink_t limit! */
2466 vap
->va_size
= zp
->z_size
;
2467 vap
->va_rdev
= vp
->v_rdev
;
2468 vap
->va_seq
= zp
->z_seq
;
2471 * Add in any requested optional attributes and the create time.
2472 * Also set the corresponding bits in the returned attribute bitmap.
2474 if ((xoap
= xva_getxoptattr(xvap
)) != NULL
&& zfsvfs
->z_use_fuids
) {
2475 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
2477 ((zp
->z_pflags
& ZFS_ARCHIVE
) != 0);
2478 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
2481 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
2482 xoap
->xoa_readonly
=
2483 ((zp
->z_pflags
& ZFS_READONLY
) != 0);
2484 XVA_SET_RTN(xvap
, XAT_READONLY
);
2487 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
2489 ((zp
->z_pflags
& ZFS_SYSTEM
) != 0);
2490 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
2493 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
2495 ((zp
->z_pflags
& ZFS_HIDDEN
) != 0);
2496 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
2499 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2500 xoap
->xoa_nounlink
=
2501 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0);
2502 XVA_SET_RTN(xvap
, XAT_NOUNLINK
);
2505 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2506 xoap
->xoa_immutable
=
2507 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0);
2508 XVA_SET_RTN(xvap
, XAT_IMMUTABLE
);
2511 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2512 xoap
->xoa_appendonly
=
2513 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0);
2514 XVA_SET_RTN(xvap
, XAT_APPENDONLY
);
2517 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2519 ((zp
->z_pflags
& ZFS_NODUMP
) != 0);
2520 XVA_SET_RTN(xvap
, XAT_NODUMP
);
2523 if (XVA_ISSET_REQ(xvap
, XAT_OPAQUE
)) {
2525 ((zp
->z_pflags
& ZFS_OPAQUE
) != 0);
2526 XVA_SET_RTN(xvap
, XAT_OPAQUE
);
2529 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2530 xoap
->xoa_av_quarantined
=
2531 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0);
2532 XVA_SET_RTN(xvap
, XAT_AV_QUARANTINED
);
2535 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2536 xoap
->xoa_av_modified
=
2537 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0);
2538 XVA_SET_RTN(xvap
, XAT_AV_MODIFIED
);
2541 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) &&
2542 vp
->v_type
== VREG
) {
2543 zfs_sa_get_scanstamp(zp
, xvap
);
2546 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
2549 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_CRTIME(zfsvfs
),
2550 times
, sizeof (times
));
2551 ZFS_TIME_DECODE(&xoap
->xoa_createtime
, times
);
2552 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
2555 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2556 xoap
->xoa_reparse
= ((zp
->z_pflags
& ZFS_REPARSE
) != 0);
2557 XVA_SET_RTN(xvap
, XAT_REPARSE
);
2559 if (XVA_ISSET_REQ(xvap
, XAT_GEN
)) {
2560 xoap
->xoa_generation
= zp
->z_gen
;
2561 XVA_SET_RTN(xvap
, XAT_GEN
);
2564 if (XVA_ISSET_REQ(xvap
, XAT_OFFLINE
)) {
2566 ((zp
->z_pflags
& ZFS_OFFLINE
) != 0);
2567 XVA_SET_RTN(xvap
, XAT_OFFLINE
);
2570 if (XVA_ISSET_REQ(xvap
, XAT_SPARSE
)) {
2572 ((zp
->z_pflags
& ZFS_SPARSE
) != 0);
2573 XVA_SET_RTN(xvap
, XAT_SPARSE
);
2577 ZFS_TIME_DECODE(&vap
->va_atime
, zp
->z_atime
);
2578 ZFS_TIME_DECODE(&vap
->va_mtime
, mtime
);
2579 ZFS_TIME_DECODE(&vap
->va_ctime
, ctime
);
2581 mutex_exit(&zp
->z_lock
);
2583 sa_object_size(zp
->z_sa_hdl
, &vap
->va_blksize
, &vap
->va_nblocks
);
2585 if (zp
->z_blksz
== 0) {
2587 * Block size hasn't been set; suggest maximal I/O transfers.
2589 vap
->va_blksize
= zfsvfs
->z_max_blksz
;
2597 * Set the file attributes to the values contained in the
2600 * IN: vp - vnode of file to be modified.
2601 * vap - new attribute values.
2602 * If AT_XVATTR set, then optional attrs are being set
2603 * flags - ATTR_UTIME set if non-default time values provided.
2604 * - ATTR_NOACLCHECK (CIFS context only).
2605 * cr - credentials of caller.
2606 * ct - caller context
2608 * RETURN: 0 on success, error code on failure.
2611 * vp - ctime updated, mtime updated if size changed.
2615 zfs_setattr(vnode_t
*vp
, vattr_t
*vap
, int flags
, cred_t
*cr
,
2616 caller_context_t
*ct
)
2618 znode_t
*zp
= VTOZ(vp
);
2619 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
2624 uint_t mask
= vap
->va_mask
;
2625 uint_t saved_mask
= 0;
2628 uint64_t new_uid
, new_gid
;
2630 uint64_t mtime
[2], ctime
[2];
2632 int need_policy
= FALSE
;
2634 zfs_fuid_info_t
*fuidp
= NULL
;
2635 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* vap may be an xvattr_t * */
2638 boolean_t skipaclchk
= (flags
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
2639 boolean_t fuid_dirtied
= B_FALSE
;
2640 sa_bulk_attr_t bulk
[7], xattr_bulk
[7];
2641 int count
= 0, xattr_count
= 0;
2646 if (mask
& AT_NOSET
)
2647 return (SET_ERROR(EINVAL
));
2652 zilog
= zfsvfs
->z_log
;
2655 * Make sure that if we have ephemeral uid/gid or xvattr specified
2656 * that file system is at proper version level
2659 if (zfsvfs
->z_use_fuids
== B_FALSE
&&
2660 (((mask
& AT_UID
) && IS_EPHEMERAL(vap
->va_uid
)) ||
2661 ((mask
& AT_GID
) && IS_EPHEMERAL(vap
->va_gid
)) ||
2662 (mask
& AT_XVATTR
))) {
2664 return (SET_ERROR(EINVAL
));
2667 if (mask
& AT_SIZE
&& vp
->v_type
== VDIR
) {
2669 return (SET_ERROR(EISDIR
));
2672 if (mask
& AT_SIZE
&& vp
->v_type
!= VREG
&& vp
->v_type
!= VFIFO
) {
2674 return (SET_ERROR(EINVAL
));
2678 * If this is an xvattr_t, then get a pointer to the structure of
2679 * optional attributes. If this is NULL, then we have a vattr_t.
2681 xoap
= xva_getxoptattr(xvap
);
2683 xva_init(&tmpxvattr
);
2686 * Immutable files can only alter immutable bit and atime
2688 if ((zp
->z_pflags
& ZFS_IMMUTABLE
) &&
2689 ((mask
& (AT_SIZE
|AT_UID
|AT_GID
|AT_MTIME
|AT_MODE
)) ||
2690 ((mask
& AT_XVATTR
) && XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)))) {
2692 return (SET_ERROR(EPERM
));
2695 if ((mask
& AT_SIZE
) && (zp
->z_pflags
& ZFS_READONLY
)) {
2697 return (SET_ERROR(EPERM
));
2701 * Verify timestamps doesn't overflow 32 bits.
2702 * ZFS can handle large timestamps, but 32bit syscalls can't
2703 * handle times greater than 2039. This check should be removed
2704 * once large timestamps are fully supported.
2706 if (mask
& (AT_ATIME
| AT_MTIME
)) {
2707 if (((mask
& AT_ATIME
) && TIMESPEC_OVERFLOW(&vap
->va_atime
)) ||
2708 ((mask
& AT_MTIME
) && TIMESPEC_OVERFLOW(&vap
->va_mtime
))) {
2710 return (SET_ERROR(EOVERFLOW
));
2718 /* Can this be moved to before the top label? */
2719 if (zfsvfs
->z_vfs
->vfs_flag
& VFS_RDONLY
) {
2721 return (SET_ERROR(EROFS
));
2725 * First validate permissions
2728 if (mask
& AT_SIZE
) {
2729 err
= zfs_zaccess(zp
, ACE_WRITE_DATA
, 0, skipaclchk
, cr
);
2735 * XXX - Note, we are not providing any open
2736 * mode flags here (like FNDELAY), so we may
2737 * block if there are locks present... this
2738 * should be addressed in openat().
2740 /* XXX - would it be OK to generate a log record here? */
2741 err
= zfs_freesp(zp
, vap
->va_size
, 0, 0, FALSE
);
2748 if (mask
& (AT_ATIME
|AT_MTIME
) ||
2749 ((mask
& AT_XVATTR
) && (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
) ||
2750 XVA_ISSET_REQ(xvap
, XAT_READONLY
) ||
2751 XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
) ||
2752 XVA_ISSET_REQ(xvap
, XAT_OFFLINE
) ||
2753 XVA_ISSET_REQ(xvap
, XAT_SPARSE
) ||
2754 XVA_ISSET_REQ(xvap
, XAT_CREATETIME
) ||
2755 XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)))) {
2756 need_policy
= zfs_zaccess(zp
, ACE_WRITE_ATTRIBUTES
, 0,
2760 if (mask
& (AT_UID
|AT_GID
)) {
2761 int idmask
= (mask
& (AT_UID
|AT_GID
));
2766 * NOTE: even if a new mode is being set,
2767 * we may clear S_ISUID/S_ISGID bits.
2770 if (!(mask
& AT_MODE
))
2771 vap
->va_mode
= zp
->z_mode
;
2774 * Take ownership or chgrp to group we are a member of
2777 take_owner
= (mask
& AT_UID
) && (vap
->va_uid
== crgetuid(cr
));
2778 take_group
= (mask
& AT_GID
) &&
2779 zfs_groupmember(zfsvfs
, vap
->va_gid
, cr
);
2782 * If both AT_UID and AT_GID are set then take_owner and
2783 * take_group must both be set in order to allow taking
2786 * Otherwise, send the check through secpolicy_vnode_setattr()
2790 if (((idmask
== (AT_UID
|AT_GID
)) && take_owner
&& take_group
) ||
2791 ((idmask
== AT_UID
) && take_owner
) ||
2792 ((idmask
== AT_GID
) && take_group
)) {
2793 if (zfs_zaccess(zp
, ACE_WRITE_OWNER
, 0,
2794 skipaclchk
, cr
) == 0) {
2796 * Remove setuid/setgid for non-privileged users
2798 secpolicy_setid_clear(vap
, cr
);
2799 trim_mask
= (mask
& (AT_UID
|AT_GID
));
2808 mutex_enter(&zp
->z_lock
);
2809 oldva
.va_mode
= zp
->z_mode
;
2810 zfs_fuid_map_ids(zp
, cr
, &oldva
.va_uid
, &oldva
.va_gid
);
2811 if (mask
& AT_XVATTR
) {
2813 * Update xvattr mask to include only those attributes
2814 * that are actually changing.
2816 * the bits will be restored prior to actually setting
2817 * the attributes so the caller thinks they were set.
2819 if (XVA_ISSET_REQ(xvap
, XAT_APPENDONLY
)) {
2820 if (xoap
->xoa_appendonly
!=
2821 ((zp
->z_pflags
& ZFS_APPENDONLY
) != 0)) {
2824 XVA_CLR_REQ(xvap
, XAT_APPENDONLY
);
2825 XVA_SET_REQ(&tmpxvattr
, XAT_APPENDONLY
);
2829 if (XVA_ISSET_REQ(xvap
, XAT_NOUNLINK
)) {
2830 if (xoap
->xoa_nounlink
!=
2831 ((zp
->z_pflags
& ZFS_NOUNLINK
) != 0)) {
2834 XVA_CLR_REQ(xvap
, XAT_NOUNLINK
);
2835 XVA_SET_REQ(&tmpxvattr
, XAT_NOUNLINK
);
2839 if (XVA_ISSET_REQ(xvap
, XAT_IMMUTABLE
)) {
2840 if (xoap
->xoa_immutable
!=
2841 ((zp
->z_pflags
& ZFS_IMMUTABLE
) != 0)) {
2844 XVA_CLR_REQ(xvap
, XAT_IMMUTABLE
);
2845 XVA_SET_REQ(&tmpxvattr
, XAT_IMMUTABLE
);
2849 if (XVA_ISSET_REQ(xvap
, XAT_NODUMP
)) {
2850 if (xoap
->xoa_nodump
!=
2851 ((zp
->z_pflags
& ZFS_NODUMP
) != 0)) {
2854 XVA_CLR_REQ(xvap
, XAT_NODUMP
);
2855 XVA_SET_REQ(&tmpxvattr
, XAT_NODUMP
);
2859 if (XVA_ISSET_REQ(xvap
, XAT_AV_MODIFIED
)) {
2860 if (xoap
->xoa_av_modified
!=
2861 ((zp
->z_pflags
& ZFS_AV_MODIFIED
) != 0)) {
2864 XVA_CLR_REQ(xvap
, XAT_AV_MODIFIED
);
2865 XVA_SET_REQ(&tmpxvattr
, XAT_AV_MODIFIED
);
2869 if (XVA_ISSET_REQ(xvap
, XAT_AV_QUARANTINED
)) {
2870 if ((vp
->v_type
!= VREG
&&
2871 xoap
->xoa_av_quarantined
) ||
2872 xoap
->xoa_av_quarantined
!=
2873 ((zp
->z_pflags
& ZFS_AV_QUARANTINED
) != 0)) {
2876 XVA_CLR_REQ(xvap
, XAT_AV_QUARANTINED
);
2877 XVA_SET_REQ(&tmpxvattr
, XAT_AV_QUARANTINED
);
2881 if (XVA_ISSET_REQ(xvap
, XAT_REPARSE
)) {
2882 mutex_exit(&zp
->z_lock
);
2884 return (SET_ERROR(EPERM
));
2887 if (need_policy
== FALSE
&&
2888 (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
) ||
2889 XVA_ISSET_REQ(xvap
, XAT_OPAQUE
))) {
2894 mutex_exit(&zp
->z_lock
);
2896 if (mask
& AT_MODE
) {
2897 if (zfs_zaccess(zp
, ACE_WRITE_ACL
, 0, skipaclchk
, cr
) == 0) {
2898 err
= secpolicy_setid_setsticky_clear(vp
, vap
,
2904 trim_mask
|= AT_MODE
;
2912 * If trim_mask is set then take ownership
2913 * has been granted or write_acl is present and user
2914 * has the ability to modify mode. In that case remove
2915 * UID|GID and or MODE from mask so that
2916 * secpolicy_vnode_setattr() doesn't revoke it.
2920 saved_mask
= vap
->va_mask
;
2921 vap
->va_mask
&= ~trim_mask
;
2923 err
= secpolicy_vnode_setattr(cr
, vp
, vap
, &oldva
, flags
,
2924 (int (*)(void *, int, cred_t
*))zfs_zaccess_unix
, zp
);
2931 vap
->va_mask
|= saved_mask
;
2935 * secpolicy_vnode_setattr, or take ownership may have
2938 mask
= vap
->va_mask
;
2940 if ((mask
& (AT_UID
| AT_GID
))) {
2941 err
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_XATTR(zfsvfs
),
2942 &xattr_obj
, sizeof (xattr_obj
));
2944 if (err
== 0 && xattr_obj
) {
2945 err
= zfs_zget(zp
->z_zfsvfs
, xattr_obj
, &attrzp
);
2949 if (mask
& AT_UID
) {
2950 new_uid
= zfs_fuid_create(zfsvfs
,
2951 (uint64_t)vap
->va_uid
, cr
, ZFS_OWNER
, &fuidp
);
2952 if (new_uid
!= zp
->z_uid
&&
2953 zfs_fuid_overquota(zfsvfs
, B_FALSE
, new_uid
)) {
2955 VN_RELE(ZTOV(attrzp
));
2956 err
= SET_ERROR(EDQUOT
);
2961 if (mask
& AT_GID
) {
2962 new_gid
= zfs_fuid_create(zfsvfs
, (uint64_t)vap
->va_gid
,
2963 cr
, ZFS_GROUP
, &fuidp
);
2964 if (new_gid
!= zp
->z_gid
&&
2965 zfs_fuid_overquota(zfsvfs
, B_TRUE
, new_gid
)) {
2967 VN_RELE(ZTOV(attrzp
));
2968 err
= SET_ERROR(EDQUOT
);
2973 tx
= dmu_tx_create(zfsvfs
->z_os
);
2975 if (mask
& AT_MODE
) {
2976 uint64_t pmode
= zp
->z_mode
;
2978 new_mode
= (pmode
& S_IFMT
) | (vap
->va_mode
& ~S_IFMT
);
2980 if (zp
->z_zfsvfs
->z_acl_mode
== ZFS_ACL_RESTRICTED
&&
2981 !(zp
->z_pflags
& ZFS_ACL_TRIVIAL
)) {
2982 err
= SET_ERROR(EPERM
);
2986 if (err
= zfs_acl_chmod_setattr(zp
, &aclp
, new_mode
))
2989 mutex_enter(&zp
->z_lock
);
2990 if (!zp
->z_is_sa
&& ((acl_obj
= zfs_external_acl(zp
)) != 0)) {
2992 * Are we upgrading ACL from old V0 format
2995 if (zfsvfs
->z_version
>= ZPL_VERSION_FUID
&&
2996 zfs_znode_acl_version(zp
) ==
2997 ZFS_ACL_VERSION_INITIAL
) {
2998 dmu_tx_hold_free(tx
, acl_obj
, 0,
3000 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3001 0, aclp
->z_acl_bytes
);
3003 dmu_tx_hold_write(tx
, acl_obj
, 0,
3006 } else if (!zp
->z_is_sa
&& aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3007 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
,
3008 0, aclp
->z_acl_bytes
);
3010 mutex_exit(&zp
->z_lock
);
3011 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3013 if ((mask
& AT_XVATTR
) &&
3014 XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3015 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_TRUE
);
3017 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
3021 dmu_tx_hold_sa(tx
, attrzp
->z_sa_hdl
, B_FALSE
);
3024 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3026 zfs_fuid_txhold(zfsvfs
, tx
);
3028 zfs_sa_upgrade_txholds(tx
, zp
);
3030 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
3032 if (err
== ERESTART
)
3039 * Set each attribute requested.
3040 * We group settings according to the locks they need to acquire.
3042 * Note: you cannot set ctime directly, although it will be
3043 * updated as a side-effect of calling this function.
3047 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3048 mutex_enter(&zp
->z_acl_lock
);
3049 mutex_enter(&zp
->z_lock
);
3051 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
3052 &zp
->z_pflags
, sizeof (zp
->z_pflags
));
3055 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3056 mutex_enter(&attrzp
->z_acl_lock
);
3057 mutex_enter(&attrzp
->z_lock
);
3058 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3059 SA_ZPL_FLAGS(zfsvfs
), NULL
, &attrzp
->z_pflags
,
3060 sizeof (attrzp
->z_pflags
));
3063 if (mask
& (AT_UID
|AT_GID
)) {
3065 if (mask
& AT_UID
) {
3066 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_UID(zfsvfs
), NULL
,
3067 &new_uid
, sizeof (new_uid
));
3068 zp
->z_uid
= new_uid
;
3070 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3071 SA_ZPL_UID(zfsvfs
), NULL
, &new_uid
,
3073 attrzp
->z_uid
= new_uid
;
3077 if (mask
& AT_GID
) {
3078 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_GID(zfsvfs
),
3079 NULL
, &new_gid
, sizeof (new_gid
));
3080 zp
->z_gid
= new_gid
;
3082 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3083 SA_ZPL_GID(zfsvfs
), NULL
, &new_gid
,
3085 attrzp
->z_gid
= new_gid
;
3088 if (!(mask
& AT_MODE
)) {
3089 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
),
3090 NULL
, &new_mode
, sizeof (new_mode
));
3091 new_mode
= zp
->z_mode
;
3093 err
= zfs_acl_chown_setattr(zp
);
3096 err
= zfs_acl_chown_setattr(attrzp
);
3101 if (mask
& AT_MODE
) {
3102 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MODE(zfsvfs
), NULL
,
3103 &new_mode
, sizeof (new_mode
));
3104 zp
->z_mode
= new_mode
;
3105 ASSERT3U((uintptr_t)aclp
, !=, NULL
);
3106 err
= zfs_aclset_common(zp
, aclp
, cr
, tx
);
3108 if (zp
->z_acl_cached
)
3109 zfs_acl_free(zp
->z_acl_cached
);
3110 zp
->z_acl_cached
= aclp
;
3115 if (mask
& AT_ATIME
) {
3116 ZFS_TIME_ENCODE(&vap
->va_atime
, zp
->z_atime
);
3117 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_ATIME(zfsvfs
), NULL
,
3118 &zp
->z_atime
, sizeof (zp
->z_atime
));
3121 if (mask
& AT_MTIME
) {
3122 ZFS_TIME_ENCODE(&vap
->va_mtime
, mtime
);
3123 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
3124 mtime
, sizeof (mtime
));
3127 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */
3128 if (mask
& AT_SIZE
&& !(mask
& AT_MTIME
)) {
3129 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
),
3130 NULL
, mtime
, sizeof (mtime
));
3131 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3132 &ctime
, sizeof (ctime
));
3133 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
3135 } else if (mask
!= 0) {
3136 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
3137 &ctime
, sizeof (ctime
));
3138 zfs_tstamp_update_setup(zp
, STATE_CHANGED
, mtime
, ctime
,
3141 SA_ADD_BULK_ATTR(xattr_bulk
, xattr_count
,
3142 SA_ZPL_CTIME(zfsvfs
), NULL
,
3143 &ctime
, sizeof (ctime
));
3144 zfs_tstamp_update_setup(attrzp
, STATE_CHANGED
,
3145 mtime
, ctime
, B_TRUE
);
3149 * Do this after setting timestamps to prevent timestamp
3150 * update from toggling bit
3153 if (xoap
&& (mask
& AT_XVATTR
)) {
3156 * restore trimmed off masks
3157 * so that return masks can be set for caller.
3160 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_APPENDONLY
)) {
3161 XVA_SET_REQ(xvap
, XAT_APPENDONLY
);
3163 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_NOUNLINK
)) {
3164 XVA_SET_REQ(xvap
, XAT_NOUNLINK
);
3166 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_IMMUTABLE
)) {
3167 XVA_SET_REQ(xvap
, XAT_IMMUTABLE
);
3169 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_NODUMP
)) {
3170 XVA_SET_REQ(xvap
, XAT_NODUMP
);
3172 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_AV_MODIFIED
)) {
3173 XVA_SET_REQ(xvap
, XAT_AV_MODIFIED
);
3175 if (XVA_ISSET_REQ(&tmpxvattr
, XAT_AV_QUARANTINED
)) {
3176 XVA_SET_REQ(xvap
, XAT_AV_QUARANTINED
);
3179 if (XVA_ISSET_REQ(xvap
, XAT_AV_SCANSTAMP
))
3180 ASSERT(vp
->v_type
== VREG
);
3182 zfs_xvattr_set(zp
, xvap
, tx
);
3186 zfs_fuid_sync(zfsvfs
, tx
);
3189 zfs_log_setattr(zilog
, tx
, TX_SETATTR
, zp
, vap
, mask
, fuidp
);
3191 mutex_exit(&zp
->z_lock
);
3192 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3193 mutex_exit(&zp
->z_acl_lock
);
3196 if (mask
& (AT_UID
|AT_GID
|AT_MODE
))
3197 mutex_exit(&attrzp
->z_acl_lock
);
3198 mutex_exit(&attrzp
->z_lock
);
3201 if (err
== 0 && attrzp
) {
3202 err2
= sa_bulk_update(attrzp
->z_sa_hdl
, xattr_bulk
,
3208 VN_RELE(ZTOV(attrzp
));
3214 zfs_fuid_info_free(fuidp
);
3220 if (err
== ERESTART
)
3223 err2
= sa_bulk_update(zp
->z_sa_hdl
, bulk
, count
, tx
);
3228 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3229 zil_commit(zilog
, 0);
3235 typedef struct zfs_zlock
{
3236 krwlock_t
*zl_rwlock
; /* lock we acquired */
3237 znode_t
*zl_znode
; /* znode we held */
3238 struct zfs_zlock
*zl_next
; /* next in list */
3242 * Drop locks and release vnodes that were held by zfs_rename_lock().
3245 zfs_rename_unlock(zfs_zlock_t
**zlpp
)
3249 while ((zl
= *zlpp
) != NULL
) {
3250 if (zl
->zl_znode
!= NULL
)
3251 VN_RELE(ZTOV(zl
->zl_znode
));
3252 rw_exit(zl
->zl_rwlock
);
3253 *zlpp
= zl
->zl_next
;
3254 kmem_free(zl
, sizeof (*zl
));
3259 * Search back through the directory tree, using the ".." entries.
3260 * Lock each directory in the chain to prevent concurrent renames.
3261 * Fail any attempt to move a directory into one of its own descendants.
3262 * XXX - z_parent_lock can overlap with map or grow locks
3265 zfs_rename_lock(znode_t
*szp
, znode_t
*tdzp
, znode_t
*sdzp
, zfs_zlock_t
**zlpp
)
3269 uint64_t rootid
= zp
->z_zfsvfs
->z_root
;
3270 uint64_t oidp
= zp
->z_id
;
3271 krwlock_t
*rwlp
= &szp
->z_parent_lock
;
3272 krw_t rw
= RW_WRITER
;
3275 * First pass write-locks szp and compares to zp->z_id.
3276 * Later passes read-lock zp and compare to zp->z_parent.
3279 if (!rw_tryenter(rwlp
, rw
)) {
3281 * Another thread is renaming in this path.
3282 * Note that if we are a WRITER, we don't have any
3283 * parent_locks held yet.
3285 if (rw
== RW_READER
&& zp
->z_id
> szp
->z_id
) {
3287 * Drop our locks and restart
3289 zfs_rename_unlock(&zl
);
3293 rwlp
= &szp
->z_parent_lock
;
3298 * Wait for other thread to drop its locks
3304 zl
= kmem_alloc(sizeof (*zl
), KM_SLEEP
);
3305 zl
->zl_rwlock
= rwlp
;
3306 zl
->zl_znode
= NULL
;
3307 zl
->zl_next
= *zlpp
;
3310 if (oidp
== szp
->z_id
) /* We're a descendant of szp */
3311 return (SET_ERROR(EINVAL
));
3313 if (oidp
== rootid
) /* We've hit the top */
3316 if (rw
== RW_READER
) { /* i.e. not the first pass */
3317 int error
= zfs_zget(zp
->z_zfsvfs
, oidp
, &zp
);
3322 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_PARENT(zp
->z_zfsvfs
),
3323 &oidp
, sizeof (oidp
));
3324 rwlp
= &zp
->z_parent_lock
;
3327 } while (zp
->z_id
!= sdzp
->z_id
);
3333 * Move an entry from the provided source directory to the target
3334 * directory. Change the entry name as indicated.
3336 * IN: sdvp - Source directory containing the "old entry".
3337 * snm - Old entry name.
3338 * tdvp - Target directory to contain the "new entry".
3339 * tnm - New entry name.
3340 * cr - credentials of caller.
3341 * ct - caller context
3342 * flags - case flags
3344 * RETURN: 0 on success, error code on failure.
3347 * sdvp,tdvp - ctime|mtime updated
3351 zfs_rename(vnode_t
*sdvp
, char *snm
, vnode_t
*tdvp
, char *tnm
, cred_t
*cr
,
3352 caller_context_t
*ct
, int flags
)
3354 znode_t
*tdzp
, *szp
, *tzp
;
3355 znode_t
*sdzp
= VTOZ(sdvp
);
3356 zfsvfs_t
*zfsvfs
= sdzp
->z_zfsvfs
;
3359 zfs_dirlock_t
*sdl
, *tdl
;
3362 int cmp
, serr
, terr
;
3367 ZFS_VERIFY_ZP(sdzp
);
3368 zilog
= zfsvfs
->z_log
;
3371 * Make sure we have the real vp for the target directory.
3373 if (VOP_REALVP(tdvp
, &realvp
, ct
) == 0)
3377 ZFS_VERIFY_ZP(tdzp
);
3380 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3381 * ctldir appear to have the same v_vfsp.
3383 if (tdzp
->z_zfsvfs
!= zfsvfs
|| zfsctl_is_node(tdvp
)) {
3385 return (SET_ERROR(EXDEV
));
3388 if (zfsvfs
->z_utf8
&& u8_validate(tnm
,
3389 strlen(tnm
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3391 return (SET_ERROR(EILSEQ
));
3394 if (flags
& FIGNORECASE
)
3403 * This is to prevent the creation of links into attribute space
3404 * by renaming a linked file into/outof an attribute directory.
3405 * See the comment in zfs_link() for why this is considered bad.
3407 if ((tdzp
->z_pflags
& ZFS_XATTR
) != (sdzp
->z_pflags
& ZFS_XATTR
)) {
3409 return (SET_ERROR(EINVAL
));
3413 * Lock source and target directory entries. To prevent deadlock,
3414 * a lock ordering must be defined. We lock the directory with
3415 * the smallest object id first, or if it's a tie, the one with
3416 * the lexically first name.
3418 if (sdzp
->z_id
< tdzp
->z_id
) {
3420 } else if (sdzp
->z_id
> tdzp
->z_id
) {
3424 * First compare the two name arguments without
3425 * considering any case folding.
3427 int nofold
= (zfsvfs
->z_norm
& ~U8_TEXTPREP_TOUPPER
);
3429 cmp
= u8_strcmp(snm
, tnm
, 0, nofold
, U8_UNICODE_LATEST
, &error
);
3430 ASSERT(error
== 0 || !zfsvfs
->z_utf8
);
3433 * POSIX: "If the old argument and the new argument
3434 * both refer to links to the same existing file,
3435 * the rename() function shall return successfully
3436 * and perform no other action."
3442 * If the file system is case-folding, then we may
3443 * have some more checking to do. A case-folding file
3444 * system is either supporting mixed case sensitivity
3445 * access or is completely case-insensitive. Note
3446 * that the file system is always case preserving.
3448 * In mixed sensitivity mode case sensitive behavior
3449 * is the default. FIGNORECASE must be used to
3450 * explicitly request case insensitive behavior.
3452 * If the source and target names provided differ only
3453 * by case (e.g., a request to rename 'tim' to 'Tim'),
3454 * we will treat this as a special case in the
3455 * case-insensitive mode: as long as the source name
3456 * is an exact match, we will allow this to proceed as
3457 * a name-change request.
3459 if ((zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
||
3460 (zfsvfs
->z_case
== ZFS_CASE_MIXED
&&
3461 flags
& FIGNORECASE
)) &&
3462 u8_strcmp(snm
, tnm
, 0, zfsvfs
->z_norm
, U8_UNICODE_LATEST
,
3465 * case preserving rename request, require exact
3474 * If the source and destination directories are the same, we should
3475 * grab the z_name_lock of that directory only once.
3479 rw_enter(&sdzp
->z_name_lock
, RW_READER
);
3483 serr
= zfs_dirent_lock(&sdl
, sdzp
, snm
, &szp
,
3484 ZEXISTS
| zflg
, NULL
, NULL
);
3485 terr
= zfs_dirent_lock(&tdl
,
3486 tdzp
, tnm
, &tzp
, ZRENAMING
| zflg
, NULL
, NULL
);
3488 terr
= zfs_dirent_lock(&tdl
,
3489 tdzp
, tnm
, &tzp
, zflg
, NULL
, NULL
);
3490 serr
= zfs_dirent_lock(&sdl
,
3491 sdzp
, snm
, &szp
, ZEXISTS
| ZRENAMING
| zflg
,
3497 * Source entry invalid or not there.
3500 zfs_dirent_unlock(tdl
);
3506 rw_exit(&sdzp
->z_name_lock
);
3508 if (strcmp(snm
, "..") == 0)
3509 serr
= SET_ERROR(EINVAL
);
3514 zfs_dirent_unlock(sdl
);
3518 rw_exit(&sdzp
->z_name_lock
);
3520 if (strcmp(tnm
, "..") == 0)
3521 terr
= SET_ERROR(EINVAL
);
3527 * Must have write access at the source to remove the old entry
3528 * and write access at the target to create the new entry.
3529 * Note that if target and source are the same, this can be
3530 * done in a single check.
3533 if (error
= zfs_zaccess_rename(sdzp
, szp
, tdzp
, tzp
, cr
))
3536 if (ZTOV(szp
)->v_type
== VDIR
) {
3538 * Check to make sure rename is valid.
3539 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d
3541 if (error
= zfs_rename_lock(szp
, tdzp
, sdzp
, &zl
))
3546 * Does target exist?
3550 * Source and target must be the same type.
3552 if (ZTOV(szp
)->v_type
== VDIR
) {
3553 if (ZTOV(tzp
)->v_type
!= VDIR
) {
3554 error
= SET_ERROR(ENOTDIR
);
3558 if (ZTOV(tzp
)->v_type
== VDIR
) {
3559 error
= SET_ERROR(EISDIR
);
3564 * POSIX dictates that when the source and target
3565 * entries refer to the same file object, rename
3566 * must do nothing and exit without error.
3568 if (szp
->z_id
== tzp
->z_id
) {
3574 vnevent_rename_src(ZTOV(szp
), sdvp
, snm
, ct
);
3576 vnevent_rename_dest(ZTOV(tzp
), tdvp
, tnm
, ct
);
3579 * notify the target directory if it is not the same
3580 * as source directory.
3583 vnevent_rename_dest_dir(tdvp
, ct
);
3586 tx
= dmu_tx_create(zfsvfs
->z_os
);
3587 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3588 dmu_tx_hold_sa(tx
, sdzp
->z_sa_hdl
, B_FALSE
);
3589 dmu_tx_hold_zap(tx
, sdzp
->z_id
, FALSE
, snm
);
3590 dmu_tx_hold_zap(tx
, tdzp
->z_id
, TRUE
, tnm
);
3592 dmu_tx_hold_sa(tx
, tdzp
->z_sa_hdl
, B_FALSE
);
3593 zfs_sa_upgrade_txholds(tx
, tdzp
);
3596 dmu_tx_hold_sa(tx
, tzp
->z_sa_hdl
, B_FALSE
);
3597 zfs_sa_upgrade_txholds(tx
, tzp
);
3600 zfs_sa_upgrade_txholds(tx
, szp
);
3601 dmu_tx_hold_zap(tx
, zfsvfs
->z_unlinkedobj
, FALSE
, NULL
);
3602 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
3605 zfs_rename_unlock(&zl
);
3606 zfs_dirent_unlock(sdl
);
3607 zfs_dirent_unlock(tdl
);
3610 rw_exit(&sdzp
->z_name_lock
);
3615 if (error
== ERESTART
) {
3625 if (tzp
) /* Attempt to remove the existing target */
3626 error
= zfs_link_destroy(tdl
, tzp
, tx
, zflg
, NULL
);
3629 error
= zfs_link_create(tdl
, szp
, tx
, ZRENAMING
);
3631 szp
->z_pflags
|= ZFS_AV_MODIFIED
;
3633 error
= sa_update(szp
->z_sa_hdl
, SA_ZPL_FLAGS(zfsvfs
),
3634 (void *)&szp
->z_pflags
, sizeof (uint64_t), tx
);
3637 error
= zfs_link_destroy(sdl
, szp
, tx
, ZRENAMING
, NULL
);
3639 zfs_log_rename(zilog
, tx
, TX_RENAME
|
3640 (flags
& FIGNORECASE
? TX_CI
: 0), sdzp
,
3641 sdl
->dl_name
, tdzp
, tdl
->dl_name
, szp
);
3644 * Update path information for the target vnode
3646 vn_renamepath(tdvp
, ZTOV(szp
), tnm
,
3650 * At this point, we have successfully created
3651 * the target name, but have failed to remove
3652 * the source name. Since the create was done
3653 * with the ZRENAMING flag, there are
3654 * complications; for one, the link count is
3655 * wrong. The easiest way to deal with this
3656 * is to remove the newly created target, and
3657 * return the original error. This must
3658 * succeed; fortunately, it is very unlikely to
3659 * fail, since we just created it.
3661 VERIFY3U(zfs_link_destroy(tdl
, szp
, tx
,
3662 ZRENAMING
, NULL
), ==, 0);
3670 zfs_rename_unlock(&zl
);
3672 zfs_dirent_unlock(sdl
);
3673 zfs_dirent_unlock(tdl
);
3676 rw_exit(&sdzp
->z_name_lock
);
3683 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3684 zil_commit(zilog
, 0);
3691 * Insert the indicated symbolic reference entry into the directory.
3693 * IN: dvp - Directory to contain new symbolic link.
3694 * link - Name for new symlink entry.
3695 * vap - Attributes of new entry.
3696 * cr - credentials of caller.
3697 * ct - caller context
3698 * flags - case flags
3700 * RETURN: 0 on success, error code on failure.
3703 * dvp - ctime|mtime updated
3707 zfs_symlink(vnode_t
*dvp
, char *name
, vattr_t
*vap
, char *link
, cred_t
*cr
,
3708 caller_context_t
*ct
, int flags
)
3710 znode_t
*zp
, *dzp
= VTOZ(dvp
);
3713 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
3715 uint64_t len
= strlen(link
);
3718 zfs_acl_ids_t acl_ids
;
3719 boolean_t fuid_dirtied
;
3720 uint64_t txtype
= TX_SYMLINK
;
3722 ASSERT(vap
->va_type
== VLNK
);
3726 zilog
= zfsvfs
->z_log
;
3728 if (zfsvfs
->z_utf8
&& u8_validate(name
, strlen(name
),
3729 NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3731 return (SET_ERROR(EILSEQ
));
3733 if (flags
& FIGNORECASE
)
3736 if (len
> MAXPATHLEN
) {
3738 return (SET_ERROR(ENAMETOOLONG
));
3741 if ((error
= zfs_acl_ids_create(dzp
, 0,
3742 vap
, cr
, NULL
, &acl_ids
)) != 0) {
3748 * Attempt to lock directory; fail if entry already exists.
3750 error
= zfs_dirent_lock(&dl
, dzp
, name
, &zp
, zflg
, NULL
, NULL
);
3752 zfs_acl_ids_free(&acl_ids
);
3757 if (error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
)) {
3758 zfs_acl_ids_free(&acl_ids
);
3759 zfs_dirent_unlock(dl
);
3764 if (zfs_acl_ids_overquota(zfsvfs
, &acl_ids
)) {
3765 zfs_acl_ids_free(&acl_ids
);
3766 zfs_dirent_unlock(dl
);
3768 return (SET_ERROR(EDQUOT
));
3770 tx
= dmu_tx_create(zfsvfs
->z_os
);
3771 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
3772 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0, MAX(1, len
));
3773 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
3774 dmu_tx_hold_sa_create(tx
, acl_ids
.z_aclp
->z_acl_bytes
+
3775 ZFS_SA_BASE_ATTR_SIZE
+ len
);
3776 dmu_tx_hold_sa(tx
, dzp
->z_sa_hdl
, B_FALSE
);
3777 if (!zfsvfs
->z_use_sa
&& acl_ids
.z_aclp
->z_acl_bytes
> ZFS_ACE_SPACE
) {
3778 dmu_tx_hold_write(tx
, DMU_NEW_OBJECT
, 0,
3779 acl_ids
.z_aclp
->z_acl_bytes
);
3782 zfs_fuid_txhold(zfsvfs
, tx
);
3783 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
3785 zfs_dirent_unlock(dl
);
3786 if (error
== ERESTART
) {
3791 zfs_acl_ids_free(&acl_ids
);
3798 * Create a new object for the symlink.
3799 * for version 4 ZPL datsets the symlink will be an SA attribute
3801 zfs_mknode(dzp
, vap
, tx
, cr
, 0, &zp
, &acl_ids
);
3804 zfs_fuid_sync(zfsvfs
, tx
);
3806 mutex_enter(&zp
->z_lock
);
3808 error
= sa_update(zp
->z_sa_hdl
, SA_ZPL_SYMLINK(zfsvfs
),
3811 zfs_sa_symlink(zp
, link
, len
, tx
);
3812 mutex_exit(&zp
->z_lock
);
3815 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_SIZE(zfsvfs
),
3816 &zp
->z_size
, sizeof (zp
->z_size
), tx
);
3818 * Insert the new object into the directory.
3820 (void) zfs_link_create(dl
, zp
, tx
, ZNEW
);
3822 if (flags
& FIGNORECASE
)
3824 zfs_log_symlink(zilog
, tx
, txtype
, dzp
, zp
, name
, link
);
3826 zfs_acl_ids_free(&acl_ids
);
3830 zfs_dirent_unlock(dl
);
3834 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
3835 zil_commit(zilog
, 0);
3842 * Return, in the buffer contained in the provided uio structure,
3843 * the symbolic path referred to by vp.
3845 * IN: vp - vnode of symbolic link.
3846 * uio - structure to contain the link path.
3847 * cr - credentials of caller.
3848 * ct - caller context
3850 * OUT: uio - structure containing the link path.
3852 * RETURN: 0 on success, error code on failure.
3855 * vp - atime updated
3859 zfs_readlink(vnode_t
*vp
, uio_t
*uio
, cred_t
*cr
, caller_context_t
*ct
)
3861 znode_t
*zp
= VTOZ(vp
);
3862 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
3868 mutex_enter(&zp
->z_lock
);
3870 error
= sa_lookup_uio(zp
->z_sa_hdl
,
3871 SA_ZPL_SYMLINK(zfsvfs
), uio
);
3873 error
= zfs_sa_readlink(zp
, uio
);
3874 mutex_exit(&zp
->z_lock
);
3876 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
3883 * Insert a new entry into directory tdvp referencing svp.
3885 * IN: tdvp - Directory to contain new entry.
3886 * svp - vnode of new entry.
3887 * name - name of new entry.
3888 * cr - credentials of caller.
3889 * ct - caller context
3891 * RETURN: 0 on success, error code on failure.
3894 * tdvp - ctime|mtime updated
3895 * svp - ctime updated
3899 zfs_link(vnode_t
*tdvp
, vnode_t
*svp
, char *name
, cred_t
*cr
,
3900 caller_context_t
*ct
, int flags
)
3902 znode_t
*dzp
= VTOZ(tdvp
);
3904 zfsvfs_t
*zfsvfs
= dzp
->z_zfsvfs
;
3914 ASSERT(tdvp
->v_type
== VDIR
);
3918 zilog
= zfsvfs
->z_log
;
3920 if (VOP_REALVP(svp
, &realvp
, ct
) == 0)
3924 * POSIX dictates that we return EPERM here.
3925 * Better choices include ENOTSUP or EISDIR.
3927 if (svp
->v_type
== VDIR
) {
3929 return (SET_ERROR(EPERM
));
3936 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the
3937 * ctldir appear to have the same v_vfsp.
3939 if (szp
->z_zfsvfs
!= zfsvfs
|| zfsctl_is_node(svp
)) {
3941 return (SET_ERROR(EXDEV
));
3944 /* Prevent links to .zfs/shares files */
3946 if ((error
= sa_lookup(szp
->z_sa_hdl
, SA_ZPL_PARENT(zfsvfs
),
3947 &parent
, sizeof (uint64_t))) != 0) {
3951 if (parent
== zfsvfs
->z_shares_dir
) {
3953 return (SET_ERROR(EPERM
));
3956 if (zfsvfs
->z_utf8
&& u8_validate(name
,
3957 strlen(name
), NULL
, U8_VALIDATE_ENTIRE
, &error
) < 0) {
3959 return (SET_ERROR(EILSEQ
));
3961 if (flags
& FIGNORECASE
)
3965 * We do not support links between attributes and non-attributes
3966 * because of the potential security risk of creating links
3967 * into "normal" file space in order to circumvent restrictions
3968 * imposed in attribute space.
3970 if ((szp
->z_pflags
& ZFS_XATTR
) != (dzp
->z_pflags
& ZFS_XATTR
)) {
3972 return (SET_ERROR(EINVAL
));
3976 owner
= zfs_fuid_map_id(zfsvfs
, szp
->z_uid
, cr
, ZFS_OWNER
);
3977 if (owner
!= crgetuid(cr
) && secpolicy_basic_link(cr
) != 0) {
3979 return (SET_ERROR(EPERM
));
3982 if (error
= zfs_zaccess(dzp
, ACE_ADD_FILE
, 0, B_FALSE
, cr
)) {
3989 * Attempt to lock directory; fail if entry already exists.
3991 error
= zfs_dirent_lock(&dl
, dzp
, name
, &tzp
, zf
, NULL
, NULL
);
3997 tx
= dmu_tx_create(zfsvfs
->z_os
);
3998 dmu_tx_hold_sa(tx
, szp
->z_sa_hdl
, B_FALSE
);
3999 dmu_tx_hold_zap(tx
, dzp
->z_id
, TRUE
, name
);
4000 zfs_sa_upgrade_txholds(tx
, szp
);
4001 zfs_sa_upgrade_txholds(tx
, dzp
);
4002 error
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4004 zfs_dirent_unlock(dl
);
4005 if (error
== ERESTART
) {
4015 error
= zfs_link_create(dl
, szp
, tx
, 0);
4018 uint64_t txtype
= TX_LINK
;
4019 if (flags
& FIGNORECASE
)
4021 zfs_log_link(zilog
, tx
, txtype
, dzp
, szp
, name
);
4026 zfs_dirent_unlock(dl
);
4029 vnevent_link(svp
, ct
);
4032 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4033 zil_commit(zilog
, 0);
4040 * zfs_null_putapage() is used when the file system has been force
4041 * unmounted. It just drops the pages.
4045 zfs_null_putapage(vnode_t
*vp
, page_t
*pp
, u_offset_t
*offp
,
4046 size_t *lenp
, int flags
, cred_t
*cr
)
4048 pvn_write_done(pp
, B_INVAL
|B_FORCE
|B_ERROR
);
4053 * Push a page out to disk, klustering if possible.
4055 * IN: vp - file to push page to.
4056 * pp - page to push.
4057 * flags - additional flags.
4058 * cr - credentials of caller.
4060 * OUT: offp - start of range pushed.
4061 * lenp - len of range pushed.
4063 * RETURN: 0 on success, error code on failure.
4065 * NOTE: callers must have locked the page to be pushed. On
4066 * exit, the page (and all other pages in the kluster) must be
4071 zfs_putapage(vnode_t
*vp
, page_t
*pp
, u_offset_t
*offp
,
4072 size_t *lenp
, int flags
, cred_t
*cr
)
4074 znode_t
*zp
= VTOZ(vp
);
4075 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4077 u_offset_t off
, koff
;
4084 * If our blocksize is bigger than the page size, try to kluster
4085 * multiple pages so that we write a full block (thus avoiding
4086 * a read-modify-write).
4088 if (off
< zp
->z_size
&& zp
->z_blksz
> PAGESIZE
) {
4089 klen
= P2ROUNDUP((ulong_t
)zp
->z_blksz
, PAGESIZE
);
4090 koff
= ISP2(klen
) ? P2ALIGN(off
, (u_offset_t
)klen
) : 0;
4091 ASSERT(koff
<= zp
->z_size
);
4092 if (koff
+ klen
> zp
->z_size
)
4093 klen
= P2ROUNDUP(zp
->z_size
- koff
, (uint64_t)PAGESIZE
);
4094 pp
= pvn_write_kluster(vp
, pp
, &off
, &len
, koff
, klen
, flags
);
4096 ASSERT3U(btop(len
), ==, btopr(len
));
4099 * Can't push pages past end-of-file.
4101 if (off
>= zp
->z_size
) {
4102 /* ignore all pages */
4105 } else if (off
+ len
> zp
->z_size
) {
4106 int npages
= btopr(zp
->z_size
- off
);
4109 page_list_break(&pp
, &trunc
, npages
);
4110 /* ignore pages past end of file */
4112 pvn_write_done(trunc
, flags
);
4113 len
= zp
->z_size
- off
;
4116 if (zfs_owner_overquota(zfsvfs
, zp
, B_FALSE
) ||
4117 zfs_owner_overquota(zfsvfs
, zp
, B_TRUE
)) {
4118 err
= SET_ERROR(EDQUOT
);
4122 tx
= dmu_tx_create(zfsvfs
->z_os
);
4123 dmu_tx_hold_write(tx
, zp
->z_id
, off
, len
);
4125 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4126 zfs_sa_upgrade_txholds(tx
, zp
);
4127 err
= dmu_tx_assign(tx
, TXG_NOWAIT
);
4129 if (err
== ERESTART
) {
4138 if (zp
->z_blksz
<= PAGESIZE
) {
4139 caddr_t va
= zfs_map_page(pp
, S_READ
);
4140 ASSERT3U(len
, <=, PAGESIZE
);
4141 dmu_write(zfsvfs
->z_os
, zp
->z_id
, off
, len
, va
, tx
);
4142 zfs_unmap_page(pp
, va
);
4144 err
= dmu_write_pages(zfsvfs
->z_os
, zp
->z_id
, off
, len
, pp
, tx
);
4148 uint64_t mtime
[2], ctime
[2];
4149 sa_bulk_attr_t bulk
[3];
4152 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_MTIME(zfsvfs
), NULL
,
4154 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_CTIME(zfsvfs
), NULL
,
4156 SA_ADD_BULK_ATTR(bulk
, count
, SA_ZPL_FLAGS(zfsvfs
), NULL
,
4158 zfs_tstamp_update_setup(zp
, CONTENT_MODIFIED
, mtime
, ctime
,
4160 zfs_log_write(zfsvfs
->z_log
, tx
, TX_WRITE
, zp
, off
, len
, 0);
4165 pvn_write_done(pp
, (err
? B_ERROR
: 0) | flags
);
4175 * Copy the portion of the file indicated from pages into the file.
4176 * The pages are stored in a page list attached to the files vnode.
4178 * IN: vp - vnode of file to push page data to.
4179 * off - position in file to put data.
4180 * len - amount of data to write.
4181 * flags - flags to control the operation.
4182 * cr - credentials of caller.
4183 * ct - caller context.
4185 * RETURN: 0 on success, error code on failure.
4188 * vp - ctime|mtime updated
4192 zfs_putpage(vnode_t
*vp
, offset_t off
, size_t len
, int flags
, cred_t
*cr
,
4193 caller_context_t
*ct
)
4195 znode_t
*zp
= VTOZ(vp
);
4196 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4208 * There's nothing to do if no data is cached.
4210 if (!vn_has_cached_data(vp
)) {
4216 * Align this request to the file block size in case we kluster.
4217 * XXX - this can result in pretty aggresive locking, which can
4218 * impact simultanious read/write access. One option might be
4219 * to break up long requests (len == 0) into block-by-block
4220 * operations to get narrower locking.
4222 blksz
= zp
->z_blksz
;
4224 io_off
= P2ALIGN_TYPED(off
, blksz
, u_offset_t
);
4227 if (len
> 0 && ISP2(blksz
))
4228 io_len
= P2ROUNDUP_TYPED(len
+ (off
- io_off
), blksz
, size_t);
4234 * Search the entire vp list for pages >= io_off.
4236 rl
= zfs_range_lock(zp
, io_off
, UINT64_MAX
, RL_WRITER
);
4237 error
= pvn_vplist_dirty(vp
, io_off
, zfs_putapage
, flags
, cr
);
4240 rl
= zfs_range_lock(zp
, io_off
, io_len
, RL_WRITER
);
4242 if (off
> zp
->z_size
) {
4243 /* past end of file */
4244 zfs_range_unlock(rl
);
4249 len
= MIN(io_len
, P2ROUNDUP(zp
->z_size
, PAGESIZE
) - io_off
);
4251 for (off
= io_off
; io_off
< off
+ len
; io_off
+= io_len
) {
4252 if ((flags
& B_INVAL
) || ((flags
& B_ASYNC
) == 0)) {
4253 pp
= page_lookup(vp
, io_off
,
4254 (flags
& (B_INVAL
| B_FREE
)) ? SE_EXCL
: SE_SHARED
);
4256 pp
= page_lookup_nowait(vp
, io_off
,
4257 (flags
& B_FREE
) ? SE_EXCL
: SE_SHARED
);
4260 if (pp
!= NULL
&& pvn_getdirty(pp
, flags
)) {
4264 * Found a dirty page to push
4266 err
= zfs_putapage(vp
, pp
, &io_off
, &io_len
, flags
, cr
);
4274 zfs_range_unlock(rl
);
4275 if ((flags
& B_ASYNC
) == 0 || zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4276 zil_commit(zfsvfs
->z_log
, zp
->z_id
);
4283 zfs_inactive(vnode_t
*vp
, cred_t
*cr
, caller_context_t
*ct
)
4285 znode_t
*zp
= VTOZ(vp
);
4286 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4289 rw_enter(&zfsvfs
->z_teardown_inactive_lock
, RW_READER
);
4290 if (zp
->z_sa_hdl
== NULL
) {
4292 * The fs has been unmounted, or we did a
4293 * suspend/resume and this file no longer exists.
4295 if (vn_has_cached_data(vp
)) {
4296 (void) pvn_vplist_dirty(vp
, 0, zfs_null_putapage
,
4300 mutex_enter(&zp
->z_lock
);
4301 mutex_enter(&vp
->v_lock
);
4302 ASSERT(vp
->v_count
== 1);
4304 mutex_exit(&vp
->v_lock
);
4305 mutex_exit(&zp
->z_lock
);
4306 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4312 * Attempt to push any data in the page cache. If this fails
4313 * we will get kicked out later in zfs_zinactive().
4315 if (vn_has_cached_data(vp
)) {
4316 (void) pvn_vplist_dirty(vp
, 0, zfs_putapage
, B_INVAL
|B_ASYNC
,
4320 if (zp
->z_atime_dirty
&& zp
->z_unlinked
== 0) {
4321 dmu_tx_t
*tx
= dmu_tx_create(zfsvfs
->z_os
);
4323 dmu_tx_hold_sa(tx
, zp
->z_sa_hdl
, B_FALSE
);
4324 zfs_sa_upgrade_txholds(tx
, zp
);
4325 error
= dmu_tx_assign(tx
, TXG_WAIT
);
4329 mutex_enter(&zp
->z_lock
);
4330 (void) sa_update(zp
->z_sa_hdl
, SA_ZPL_ATIME(zfsvfs
),
4331 (void *)&zp
->z_atime
, sizeof (zp
->z_atime
), tx
);
4332 zp
->z_atime_dirty
= 0;
4333 mutex_exit(&zp
->z_lock
);
4339 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
4343 * Bounds-check the seek operation.
4345 * IN: vp - vnode seeking within
4346 * ooff - old file offset
4347 * noffp - pointer to new file offset
4348 * ct - caller context
4350 * RETURN: 0 on success, EINVAL if new offset invalid.
4354 zfs_seek(vnode_t
*vp
, offset_t ooff
, offset_t
*noffp
,
4355 caller_context_t
*ct
)
4357 if (vp
->v_type
== VDIR
)
4359 return ((*noffp
< 0 || *noffp
> MAXOFFSET_T
) ? EINVAL
: 0);
4363 * Pre-filter the generic locking function to trap attempts to place
4364 * a mandatory lock on a memory mapped file.
4367 zfs_frlock(vnode_t
*vp
, int cmd
, flock64_t
*bfp
, int flag
, offset_t offset
,
4368 flk_callback_t
*flk_cbp
, cred_t
*cr
, caller_context_t
*ct
)
4370 znode_t
*zp
= VTOZ(vp
);
4371 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4377 * We are following the UFS semantics with respect to mapcnt
4378 * here: If we see that the file is mapped already, then we will
4379 * return an error, but we don't worry about races between this
4380 * function and zfs_map().
4382 if (zp
->z_mapcnt
> 0 && MANDMODE(zp
->z_mode
)) {
4384 return (SET_ERROR(EAGAIN
));
4387 return (fs_frlock(vp
, cmd
, bfp
, flag
, offset
, flk_cbp
, cr
, ct
));
4391 * If we can't find a page in the cache, we will create a new page
4392 * and fill it with file data. For efficiency, we may try to fill
4393 * multiple pages at once (klustering) to fill up the supplied page
4394 * list. Note that the pages to be filled are held with an exclusive
4395 * lock to prevent access by other threads while they are being filled.
4398 zfs_fillpage(vnode_t
*vp
, u_offset_t off
, struct seg
*seg
,
4399 caddr_t addr
, page_t
*pl
[], size_t plsz
, enum seg_rw rw
)
4401 znode_t
*zp
= VTOZ(vp
);
4402 page_t
*pp
, *cur_pp
;
4403 objset_t
*os
= zp
->z_zfsvfs
->z_os
;
4404 u_offset_t io_off
, total
;
4408 if (plsz
== PAGESIZE
|| zp
->z_blksz
<= PAGESIZE
) {
4410 * We only have a single page, don't bother klustering
4414 pp
= page_create_va(vp
, io_off
, io_len
,
4415 PG_EXCL
| PG_WAIT
, seg
, addr
);
4418 * Try to find enough pages to fill the page list
4420 pp
= pvn_read_kluster(vp
, off
, seg
, addr
, &io_off
,
4421 &io_len
, off
, plsz
, 0);
4425 * The page already exists, nothing to do here.
4432 * Fill the pages in the kluster.
4435 for (total
= io_off
+ io_len
; io_off
< total
; io_off
+= PAGESIZE
) {
4438 ASSERT3U(io_off
, ==, cur_pp
->p_offset
);
4439 va
= zfs_map_page(cur_pp
, S_WRITE
);
4440 err
= dmu_read(os
, zp
->z_id
, io_off
, PAGESIZE
, va
,
4442 zfs_unmap_page(cur_pp
, va
);
4444 /* On error, toss the entire kluster */
4445 pvn_read_done(pp
, B_ERROR
);
4446 /* convert checksum errors into IO errors */
4448 err
= SET_ERROR(EIO
);
4451 cur_pp
= cur_pp
->p_next
;
4455 * Fill in the page list array from the kluster starting
4456 * from the desired offset `off'.
4457 * NOTE: the page list will always be null terminated.
4459 pvn_plist_init(pp
, pl
, plsz
, off
, io_len
, rw
);
4460 ASSERT(pl
== NULL
|| (*pl
)->p_offset
== off
);
4466 * Return pointers to the pages for the file region [off, off + len]
4467 * in the pl array. If plsz is greater than len, this function may
4468 * also return page pointers from after the specified region
4469 * (i.e. the region [off, off + plsz]). These additional pages are
4470 * only returned if they are already in the cache, or were created as
4471 * part of a klustered read.
4473 * IN: vp - vnode of file to get data from.
4474 * off - position in file to get data from.
4475 * len - amount of data to retrieve.
4476 * plsz - length of provided page list.
4477 * seg - segment to obtain pages for.
4478 * addr - virtual address of fault.
4479 * rw - mode of created pages.
4480 * cr - credentials of caller.
4481 * ct - caller context.
4483 * OUT: protp - protection mode of created pages.
4484 * pl - list of pages created.
4486 * RETURN: 0 on success, error code on failure.
4489 * vp - atime updated
4493 zfs_getpage(vnode_t
*vp
, offset_t off
, size_t len
, uint_t
*protp
,
4494 page_t
*pl
[], size_t plsz
, struct seg
*seg
, caddr_t addr
,
4495 enum seg_rw rw
, cred_t
*cr
, caller_context_t
*ct
)
4497 znode_t
*zp
= VTOZ(vp
);
4498 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4502 /* we do our own caching, faultahead is unnecessary */
4505 else if (len
> plsz
)
4508 len
= P2ROUNDUP(len
, PAGESIZE
);
4509 ASSERT(plsz
>= len
);
4518 * Loop through the requested range [off, off + len) looking
4519 * for pages. If we don't find a page, we will need to create
4520 * a new page and fill it with data from the file.
4523 if (*pl
= page_lookup(vp
, off
, SE_SHARED
))
4525 else if (err
= zfs_fillpage(vp
, off
, seg
, addr
, pl
, plsz
, rw
))
4528 ASSERT3U((*pl
)->p_offset
, ==, off
);
4532 ASSERT3U(len
, >=, PAGESIZE
);
4535 ASSERT3U(plsz
, >=, PAGESIZE
);
4542 * Fill out the page array with any pages already in the cache.
4545 (*pl
++ = page_lookup_nowait(vp
, off
, SE_SHARED
))) {
4552 * Release any pages we have previously locked.
4557 ZFS_ACCESSTIME_STAMP(zfsvfs
, zp
);
4567 * Request a memory map for a section of a file. This code interacts
4568 * with common code and the VM system as follows:
4570 * - common code calls mmap(), which ends up in smmap_common()
4571 * - this calls VOP_MAP(), which takes you into (say) zfs
4572 * - zfs_map() calls as_map(), passing segvn_create() as the callback
4573 * - segvn_create() creates the new segment and calls VOP_ADDMAP()
4574 * - zfs_addmap() updates z_mapcnt
4578 zfs_map(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t
*addrp
,
4579 size_t len
, uchar_t prot
, uchar_t maxprot
, uint_t flags
, cred_t
*cr
,
4580 caller_context_t
*ct
)
4582 znode_t
*zp
= VTOZ(vp
);
4583 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4584 segvn_crargs_t vn_a
;
4590 if ((prot
& PROT_WRITE
) && (zp
->z_pflags
&
4591 (ZFS_IMMUTABLE
| ZFS_READONLY
| ZFS_APPENDONLY
))) {
4593 return (SET_ERROR(EPERM
));
4596 if ((prot
& (PROT_READ
| PROT_EXEC
)) &&
4597 (zp
->z_pflags
& ZFS_AV_QUARANTINED
)) {
4599 return (SET_ERROR(EACCES
));
4602 if (vp
->v_flag
& VNOMAP
) {
4604 return (SET_ERROR(ENOSYS
));
4607 if (off
< 0 || len
> MAXOFFSET_T
- off
) {
4609 return (SET_ERROR(ENXIO
));
4612 if (vp
->v_type
!= VREG
) {
4614 return (SET_ERROR(ENODEV
));
4618 * If file is locked, disallow mapping.
4620 if (MANDMODE(zp
->z_mode
) && vn_has_flocks(vp
)) {
4622 return (SET_ERROR(EAGAIN
));
4626 error
= choose_addr(as
, addrp
, len
, off
, ADDR_VACALIGN
, flags
);
4634 vn_a
.offset
= (u_offset_t
)off
;
4635 vn_a
.type
= flags
& MAP_TYPE
;
4637 vn_a
.maxprot
= maxprot
;
4640 vn_a
.flags
= flags
& ~MAP_TYPE
;
4642 vn_a
.lgrp_mem_policy_flags
= 0;
4644 error
= as_map(as
, *addrp
, len
, segvn_create
, &vn_a
);
4653 zfs_addmap(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t addr
,
4654 size_t len
, uchar_t prot
, uchar_t maxprot
, uint_t flags
, cred_t
*cr
,
4655 caller_context_t
*ct
)
4657 uint64_t pages
= btopr(len
);
4659 atomic_add_64(&VTOZ(vp
)->z_mapcnt
, pages
);
4664 * The reason we push dirty pages as part of zfs_delmap() is so that we get a
4665 * more accurate mtime for the associated file. Since we don't have a way of
4666 * detecting when the data was actually modified, we have to resort to
4667 * heuristics. If an explicit msync() is done, then we mark the mtime when the
4668 * last page is pushed. The problem occurs when the msync() call is omitted,
4669 * which by far the most common case:
4677 * putpage() via fsflush
4679 * If we wait until fsflush to come along, we can have a modification time that
4680 * is some arbitrary point in the future. In order to prevent this in the
4681 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is
4686 zfs_delmap(vnode_t
*vp
, offset_t off
, struct as
*as
, caddr_t addr
,
4687 size_t len
, uint_t prot
, uint_t maxprot
, uint_t flags
, cred_t
*cr
,
4688 caller_context_t
*ct
)
4690 uint64_t pages
= btopr(len
);
4692 ASSERT3U(VTOZ(vp
)->z_mapcnt
, >=, pages
);
4693 atomic_add_64(&VTOZ(vp
)->z_mapcnt
, -pages
);
4695 if ((flags
& MAP_SHARED
) && (prot
& PROT_WRITE
) &&
4696 vn_has_cached_data(vp
))
4697 (void) VOP_PUTPAGE(vp
, off
, len
, B_ASYNC
, cr
, ct
);
4703 * Free or allocate space in a file. Currently, this function only
4704 * supports the `F_FREESP' command. However, this command is somewhat
4705 * misnamed, as its functionality includes the ability to allocate as
4706 * well as free space.
4708 * IN: vp - vnode of file to free data in.
4709 * cmd - action to take (only F_FREESP supported).
4710 * bfp - section of file to free/alloc.
4711 * flag - current file open mode flags.
4712 * offset - current file offset.
4713 * cr - credentials of caller [UNUSED].
4714 * ct - caller context.
4716 * RETURN: 0 on success, error code on failure.
4719 * vp - ctime|mtime updated
4723 zfs_space(vnode_t
*vp
, int cmd
, flock64_t
*bfp
, int flag
,
4724 offset_t offset
, cred_t
*cr
, caller_context_t
*ct
)
4726 znode_t
*zp
= VTOZ(vp
);
4727 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4734 if (cmd
!= F_FREESP
) {
4736 return (SET_ERROR(EINVAL
));
4739 if (error
= convoff(vp
, bfp
, 0, offset
)) {
4744 if (bfp
->l_len
< 0) {
4746 return (SET_ERROR(EINVAL
));
4750 len
= bfp
->l_len
; /* 0 means from off to end of file */
4752 error
= zfs_freesp(zp
, off
, len
, flag
, TRUE
);
4760 zfs_fid(vnode_t
*vp
, fid_t
*fidp
, caller_context_t
*ct
)
4762 znode_t
*zp
= VTOZ(vp
);
4763 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4766 uint64_t object
= zp
->z_id
;
4773 if ((error
= sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zfsvfs
),
4774 &gen64
, sizeof (uint64_t))) != 0) {
4779 gen
= (uint32_t)gen64
;
4781 size
= (zfsvfs
->z_parent
!= zfsvfs
) ? LONG_FID_LEN
: SHORT_FID_LEN
;
4782 if (fidp
->fid_len
< size
) {
4783 fidp
->fid_len
= size
;
4785 return (SET_ERROR(ENOSPC
));
4788 zfid
= (zfid_short_t
*)fidp
;
4790 zfid
->zf_len
= size
;
4792 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
4793 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
4795 /* Must have a non-zero generation number to distinguish from .zfs */
4798 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
4799 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
4801 if (size
== LONG_FID_LEN
) {
4802 uint64_t objsetid
= dmu_objset_id(zfsvfs
->z_os
);
4805 zlfid
= (zfid_long_t
*)fidp
;
4807 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
4808 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
4810 /* XXX - this should be the generation number for the objset */
4811 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
4812 zlfid
->zf_setgen
[i
] = 0;
4820 zfs_pathconf(vnode_t
*vp
, int cmd
, ulong_t
*valp
, cred_t
*cr
,
4821 caller_context_t
*ct
)
4833 case _PC_FILESIZEBITS
:
4837 case _PC_XATTR_EXISTS
:
4839 zfsvfs
= zp
->z_zfsvfs
;
4843 error
= zfs_dirent_lock(&dl
, zp
, "", &xzp
,
4844 ZXATTR
| ZEXISTS
| ZSHARED
, NULL
, NULL
);
4846 zfs_dirent_unlock(dl
);
4847 if (!zfs_dirempty(xzp
))
4850 } else if (error
== ENOENT
) {
4852 * If there aren't extended attributes, it's the
4853 * same as having zero of them.
4860 case _PC_SATTR_ENABLED
:
4861 case _PC_SATTR_EXISTS
:
4862 *valp
= vfs_has_feature(vp
->v_vfsp
, VFSFT_SYSATTR_VIEWS
) &&
4863 (vp
->v_type
== VREG
|| vp
->v_type
== VDIR
);
4866 case _PC_ACCESS_FILTERING
:
4867 *valp
= vfs_has_feature(vp
->v_vfsp
, VFSFT_ACCESS_FILTER
) &&
4871 case _PC_ACL_ENABLED
:
4872 *valp
= _ACL_ACE_ENABLED
;
4875 case _PC_MIN_HOLE_SIZE
:
4876 *valp
= (ulong_t
)SPA_MINBLOCKSIZE
;
4879 case _PC_TIMESTAMP_RESOLUTION
:
4880 /* nanosecond timestamp resolution */
4885 return (fs_pathconf(vp
, cmd
, valp
, cr
, ct
));
4891 zfs_getsecattr(vnode_t
*vp
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
,
4892 caller_context_t
*ct
)
4894 znode_t
*zp
= VTOZ(vp
);
4895 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4897 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4901 error
= zfs_getacl(zp
, vsecp
, skipaclchk
, cr
);
4909 zfs_setsecattr(vnode_t
*vp
, vsecattr_t
*vsecp
, int flag
, cred_t
*cr
,
4910 caller_context_t
*ct
)
4912 znode_t
*zp
= VTOZ(vp
);
4913 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4915 boolean_t skipaclchk
= (flag
& ATTR_NOACLCHECK
) ? B_TRUE
: B_FALSE
;
4916 zilog_t
*zilog
= zfsvfs
->z_log
;
4921 error
= zfs_setacl(zp
, vsecp
, skipaclchk
, cr
);
4923 if (zfsvfs
->z_os
->os_sync
== ZFS_SYNC_ALWAYS
)
4924 zil_commit(zilog
, 0);
4931 * The smallest read we may consider to loan out an arcbuf.
4932 * This must be a power of 2.
4934 int zcr_blksz_min
= (1 << 10); /* 1K */
4936 * If set to less than the file block size, allow loaning out of an
4937 * arcbuf for a partial block read. This must be a power of 2.
4939 int zcr_blksz_max
= (1 << 17); /* 128K */
4943 zfs_reqzcbuf(vnode_t
*vp
, enum uio_rw ioflag
, xuio_t
*xuio
, cred_t
*cr
,
4944 caller_context_t
*ct
)
4946 znode_t
*zp
= VTOZ(vp
);
4947 zfsvfs_t
*zfsvfs
= zp
->z_zfsvfs
;
4948 int max_blksz
= zfsvfs
->z_max_blksz
;
4949 uio_t
*uio
= &xuio
->xu_uio
;
4950 ssize_t size
= uio
->uio_resid
;
4951 offset_t offset
= uio
->uio_loffset
;
4956 int preamble
, postamble
;
4958 if (xuio
->xu_type
!= UIOTYPE_ZEROCOPY
)
4959 return (SET_ERROR(EINVAL
));
4966 * Loan out an arc_buf for write if write size is bigger than
4967 * max_blksz, and the file's block size is also max_blksz.
4970 if (size
< blksz
|| zp
->z_blksz
!= blksz
) {
4972 return (SET_ERROR(EINVAL
));
4975 * Caller requests buffers for write before knowing where the
4976 * write offset might be (e.g. NFS TCP write).
4981 preamble
= P2PHASE(offset
, blksz
);
4983 preamble
= blksz
- preamble
;
4988 postamble
= P2PHASE(size
, blksz
);
4991 fullblk
= size
/ blksz
;
4992 (void) dmu_xuio_init(xuio
,
4993 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
4994 DTRACE_PROBE3(zfs_reqzcbuf_align
, int, preamble
,
4995 int, postamble
, int,
4996 (preamble
!= 0) + fullblk
+ (postamble
!= 0));
4999 * Have to fix iov base/len for partial buffers. They
5000 * currently represent full arc_buf's.
5003 /* data begins in the middle of the arc_buf */
5004 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5007 (void) dmu_xuio_add(xuio
, abuf
,
5008 blksz
- preamble
, preamble
);
5011 for (i
= 0; i
< fullblk
; i
++) {
5012 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5015 (void) dmu_xuio_add(xuio
, abuf
, 0, blksz
);
5019 /* data ends in the middle of the arc_buf */
5020 abuf
= dmu_request_arcbuf(sa_get_db(zp
->z_sa_hdl
),
5023 (void) dmu_xuio_add(xuio
, abuf
, 0, postamble
);
5028 * Loan out an arc_buf for read if the read size is larger than
5029 * the current file block size. Block alignment is not
5030 * considered. Partial arc_buf will be loaned out for read.
5032 blksz
= zp
->z_blksz
;
5033 if (blksz
< zcr_blksz_min
)
5034 blksz
= zcr_blksz_min
;
5035 if (blksz
> zcr_blksz_max
)
5036 blksz
= zcr_blksz_max
;
5037 /* avoid potential complexity of dealing with it */
5038 if (blksz
> max_blksz
) {
5040 return (SET_ERROR(EINVAL
));
5043 maxsize
= zp
->z_size
- uio
->uio_loffset
;
5047 if (size
< blksz
|| vn_has_cached_data(vp
)) {
5049 return (SET_ERROR(EINVAL
));
5054 return (SET_ERROR(EINVAL
));
5057 uio
->uio_extflg
= UIO_XUIO
;
5058 XUIO_XUZC_RW(xuio
) = ioflag
;
5065 zfs_retzcbuf(vnode_t
*vp
, xuio_t
*xuio
, cred_t
*cr
, caller_context_t
*ct
)
5069 int ioflag
= XUIO_XUZC_RW(xuio
);
5071 ASSERT(xuio
->xu_type
== UIOTYPE_ZEROCOPY
);
5073 i
= dmu_xuio_cnt(xuio
);
5075 abuf
= dmu_xuio_arcbuf(xuio
, i
);
5077 * if abuf == NULL, it must be a write buffer
5078 * that has been returned in zfs_write().
5081 dmu_return_arcbuf(abuf
);
5082 ASSERT(abuf
|| ioflag
== UIO_WRITE
);
5085 dmu_xuio_fini(xuio
);
5090 * Predeclare these here so that the compiler assumes that
5091 * this is an "old style" function declaration that does
5092 * not include arguments => we won't get type mismatch errors
5093 * in the initializations that follow.
5095 static int zfs_inval();
5096 static int zfs_isdir();
5101 return (SET_ERROR(EINVAL
));
5107 return (SET_ERROR(EISDIR
));
5110 * Directory vnode operations template
5112 vnodeops_t
*zfs_dvnodeops
;
5113 const fs_operation_def_t zfs_dvnodeops_template
[] = {
5114 VOPNAME_OPEN
, { .vop_open
= zfs_open
},
5115 VOPNAME_CLOSE
, { .vop_close
= zfs_close
},
5116 VOPNAME_READ
, { .error
= zfs_isdir
},
5117 VOPNAME_WRITE
, { .error
= zfs_isdir
},
5118 VOPNAME_IOCTL
, { .vop_ioctl
= zfs_ioctl
},
5119 VOPNAME_GETATTR
, { .vop_getattr
= zfs_getattr
},
5120 VOPNAME_SETATTR
, { .vop_setattr
= zfs_setattr
},
5121 VOPNAME_ACCESS
, { .vop_access
= zfs_access
},
5122 VOPNAME_LOOKUP
, { .vop_lookup
= zfs_lookup
},
5123 VOPNAME_CREATE
, { .vop_create
= zfs_create
},
5124 VOPNAME_REMOVE
, { .vop_remove
= zfs_remove
},
5125 VOPNAME_LINK
, { .vop_link
= zfs_link
},
5126 VOPNAME_RENAME
, { .vop_rename
= zfs_rename
},
5127 VOPNAME_MKDIR
, { .vop_mkdir
= zfs_mkdir
},
5128 VOPNAME_RMDIR
, { .vop_rmdir
= zfs_rmdir
},
5129 VOPNAME_READDIR
, { .vop_readdir
= zfs_readdir
},
5130 VOPNAME_SYMLINK
, { .vop_symlink
= zfs_symlink
},
5131 VOPNAME_FSYNC
, { .vop_fsync
= zfs_fsync
},
5132 VOPNAME_INACTIVE
, { .vop_inactive
= zfs_inactive
},
5133 VOPNAME_FID
, { .vop_fid
= zfs_fid
},
5134 VOPNAME_SEEK
, { .vop_seek
= zfs_seek
},
5135 VOPNAME_PATHCONF
, { .vop_pathconf
= zfs_pathconf
},
5136 VOPNAME_GETSECATTR
, { .vop_getsecattr
= zfs_getsecattr
},
5137 VOPNAME_SETSECATTR
, { .vop_setsecattr
= zfs_setsecattr
},
5138 VOPNAME_VNEVENT
, { .vop_vnevent
= fs_vnevent_support
},
5143 * Regular file vnode operations template
5145 vnodeops_t
*zfs_fvnodeops
;
5146 const fs_operation_def_t zfs_fvnodeops_template
[] = {
5147 VOPNAME_OPEN
, { .vop_open
= zfs_open
},
5148 VOPNAME_CLOSE
, { .vop_close
= zfs_close
},
5149 VOPNAME_READ
, { .vop_read
= zfs_read
},
5150 VOPNAME_WRITE
, { .vop_write
= zfs_write
},
5151 VOPNAME_IOCTL
, { .vop_ioctl
= zfs_ioctl
},
5152 VOPNAME_GETATTR
, { .vop_getattr
= zfs_getattr
},
5153 VOPNAME_SETATTR
, { .vop_setattr
= zfs_setattr
},
5154 VOPNAME_ACCESS
, { .vop_access
= zfs_access
},
5155 VOPNAME_LOOKUP
, { .vop_lookup
= zfs_lookup
},
5156 VOPNAME_RENAME
, { .vop_rename
= zfs_rename
},
5157 VOPNAME_FSYNC
, { .vop_fsync
= zfs_fsync
},
5158 VOPNAME_INACTIVE
, { .vop_inactive
= zfs_inactive
},
5159 VOPNAME_FID
, { .vop_fid
= zfs_fid
},
5160 VOPNAME_SEEK
, { .vop_seek
= zfs_seek
},
5161 VOPNAME_FRLOCK
, { .vop_frlock
= zfs_frlock
},
5162 VOPNAME_SPACE
, { .vop_space
= zfs_space
},
5163 VOPNAME_GETPAGE
, { .vop_getpage
= zfs_getpage
},
5164 VOPNAME_PUTPAGE
, { .vop_putpage
= zfs_putpage
},
5165 VOPNAME_MAP
, { .vop_map
= zfs_map
},
5166 VOPNAME_ADDMAP
, { .vop_addmap
= zfs_addmap
},
5167 VOPNAME_DELMAP
, { .vop_delmap
= zfs_delmap
},
5168 VOPNAME_PATHCONF
, { .vop_pathconf
= zfs_pathconf
},
5169 VOPNAME_GETSECATTR
, { .vop_getsecattr
= zfs_getsecattr
},
5170 VOPNAME_SETSECATTR
, { .vop_setsecattr
= zfs_setsecattr
},
5171 VOPNAME_VNEVENT
, { .vop_vnevent
= fs_vnevent_support
},
5172 VOPNAME_REQZCBUF
, { .vop_reqzcbuf
= zfs_reqzcbuf
},
5173 VOPNAME_RETZCBUF
, { .vop_retzcbuf
= zfs_retzcbuf
},
5178 * Symbolic link vnode operations template
5180 vnodeops_t
*zfs_symvnodeops
;
5181 const fs_operation_def_t zfs_symvnodeops_template
[] = {
5182 VOPNAME_GETATTR
, { .vop_getattr
= zfs_getattr
},
5183 VOPNAME_SETATTR
, { .vop_setattr
= zfs_setattr
},
5184 VOPNAME_ACCESS
, { .vop_access
= zfs_access
},
5185 VOPNAME_RENAME
, { .vop_rename
= zfs_rename
},
5186 VOPNAME_READLINK
, { .vop_readlink
= zfs_readlink
},
5187 VOPNAME_INACTIVE
, { .vop_inactive
= zfs_inactive
},
5188 VOPNAME_FID
, { .vop_fid
= zfs_fid
},
5189 VOPNAME_PATHCONF
, { .vop_pathconf
= zfs_pathconf
},
5190 VOPNAME_VNEVENT
, { .vop_vnevent
= fs_vnevent_support
},
5195 * special share hidden files vnode operations template
5197 vnodeops_t
*zfs_sharevnodeops
;
5198 const fs_operation_def_t zfs_sharevnodeops_template
[] = {
5199 VOPNAME_GETATTR
, { .vop_getattr
= zfs_getattr
},
5200 VOPNAME_ACCESS
, { .vop_access
= zfs_access
},
5201 VOPNAME_INACTIVE
, { .vop_inactive
= zfs_inactive
},
5202 VOPNAME_FID
, { .vop_fid
= zfs_fid
},
5203 VOPNAME_PATHCONF
, { .vop_pathconf
= zfs_pathconf
},
5204 VOPNAME_GETSECATTR
, { .vop_getsecattr
= zfs_getsecattr
},
5205 VOPNAME_SETSECATTR
, { .vop_setsecattr
= zfs_setsecattr
},
5206 VOPNAME_VNEVENT
, { .vop_vnevent
= fs_vnevent_support
},
5211 * Extended attribute directory vnode operations template
5213 * This template is identical to the directory vnodes
5214 * operation template except for restricted operations:
5218 * Note that there are other restrictions embedded in:
5219 * zfs_create() - restrict type to VREG
5220 * zfs_link() - no links into/out of attribute space
5221 * zfs_rename() - no moves into/out of attribute space
5223 vnodeops_t
*zfs_xdvnodeops
;
5224 const fs_operation_def_t zfs_xdvnodeops_template
[] = {
5225 VOPNAME_OPEN
, { .vop_open
= zfs_open
},
5226 VOPNAME_CLOSE
, { .vop_close
= zfs_close
},
5227 VOPNAME_IOCTL
, { .vop_ioctl
= zfs_ioctl
},
5228 VOPNAME_GETATTR
, { .vop_getattr
= zfs_getattr
},
5229 VOPNAME_SETATTR
, { .vop_setattr
= zfs_setattr
},
5230 VOPNAME_ACCESS
, { .vop_access
= zfs_access
},
5231 VOPNAME_LOOKUP
, { .vop_lookup
= zfs_lookup
},
5232 VOPNAME_CREATE
, { .vop_create
= zfs_create
},
5233 VOPNAME_REMOVE
, { .vop_remove
= zfs_remove
},
5234 VOPNAME_LINK
, { .vop_link
= zfs_link
},
5235 VOPNAME_RENAME
, { .vop_rename
= zfs_rename
},
5236 VOPNAME_MKDIR
, { .error
= zfs_inval
},
5237 VOPNAME_RMDIR
, { .vop_rmdir
= zfs_rmdir
},
5238 VOPNAME_READDIR
, { .vop_readdir
= zfs_readdir
},
5239 VOPNAME_SYMLINK
, { .error
= zfs_inval
},
5240 VOPNAME_FSYNC
, { .vop_fsync
= zfs_fsync
},
5241 VOPNAME_INACTIVE
, { .vop_inactive
= zfs_inactive
},
5242 VOPNAME_FID
, { .vop_fid
= zfs_fid
},
5243 VOPNAME_SEEK
, { .vop_seek
= zfs_seek
},
5244 VOPNAME_PATHCONF
, { .vop_pathconf
= zfs_pathconf
},
5245 VOPNAME_GETSECATTR
, { .vop_getsecattr
= zfs_getsecattr
},
5246 VOPNAME_SETSECATTR
, { .vop_setsecattr
= zfs_setsecattr
},
5247 VOPNAME_VNEVENT
, { .vop_vnevent
= fs_vnevent_support
},
5252 * Error vnode operations template
5254 vnodeops_t
*zfs_evnodeops
;
5255 const fs_operation_def_t zfs_evnodeops_template
[] = {
5256 VOPNAME_INACTIVE
, { .vop_inactive
= zfs_inactive
},
5257 VOPNAME_PATHCONF
, { .vop_pathconf
= zfs_pathconf
},