2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_types.h"
25 #include "xfs_trans.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_itable.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_alloc.h"
47 #include "xfs_error.h"
48 #include "xfs_quota.h"
49 #include "xfs_utils.h"
50 #include "xfs_rtalloc.h"
51 #include "xfs_refcache.h"
52 #include "xfs_trans_space.h"
53 #include "xfs_log_priv.h"
54 #include "xfs_filestream.h"
62 bhv_vnode_t
*vp
= BHV_TO_VNODE(bdp
);
63 xfs_inode_t
*ip
= XFS_BHVTOI(bdp
);
65 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
66 return XFS_ERROR(EIO
);
69 * If it's a directory with any blocks, read-ahead block 0
70 * as we're almost certain to have the next operation be a read there.
72 if (VN_ISDIR(vp
) && ip
->i_d
.di_nextents
> 0) {
73 mode
= xfs_ilock_map_shared(ip
);
74 if (ip
->i_d
.di_nextents
> 0)
75 (void)xfs_da_reada_buf(NULL
, ip
, 0, XFS_DATA_FORK
);
76 xfs_iunlock(ip
, mode
);
95 vp
= BHV_TO_VNODE(bdp
);
96 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
101 if (XFS_FORCED_SHUTDOWN(mp
))
102 return XFS_ERROR(EIO
);
104 if (!(flags
& ATTR_LAZY
))
105 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
107 vap
->va_size
= XFS_ISIZE(ip
);
108 if (vap
->va_mask
== XFS_AT_SIZE
)
112 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
113 vap
->va_nodeid
= ip
->i_ino
;
115 vap
->va_nodeid
+= mp
->m_inoadd
;
117 vap
->va_nlink
= ip
->i_d
.di_nlink
;
120 * Quick exit for non-stat callers
123 ~(XFS_AT_SIZE
|XFS_AT_FSID
|XFS_AT_NODEID
|
124 XFS_AT_NLINK
|XFS_AT_BLKSIZE
)) == 0)
128 * Copy from in-core inode.
130 vap
->va_mode
= ip
->i_d
.di_mode
;
131 vap
->va_uid
= ip
->i_d
.di_uid
;
132 vap
->va_gid
= ip
->i_d
.di_gid
;
133 vap
->va_projid
= ip
->i_d
.di_projid
;
136 * Check vnode type block/char vs. everything else.
138 switch (ip
->i_d
.di_mode
& S_IFMT
) {
141 vap
->va_rdev
= ip
->i_df
.if_u2
.if_rdev
;
142 vap
->va_blocksize
= BLKDEV_IOSIZE
;
147 if (!(ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
)) {
148 vap
->va_blocksize
= xfs_preferred_iosize(mp
);
152 * If the file blocks are being allocated from a
153 * realtime partition, then return the inode's
154 * realtime extent size or the realtime volume's
158 xfs_get_extsz_hint(ip
) << mp
->m_sb
.sb_blocklog
;
163 vn_atime_to_timespec(vp
, &vap
->va_atime
);
164 vap
->va_mtime
.tv_sec
= ip
->i_d
.di_mtime
.t_sec
;
165 vap
->va_mtime
.tv_nsec
= ip
->i_d
.di_mtime
.t_nsec
;
166 vap
->va_ctime
.tv_sec
= ip
->i_d
.di_ctime
.t_sec
;
167 vap
->va_ctime
.tv_nsec
= ip
->i_d
.di_ctime
.t_nsec
;
170 * Exit for stat callers. See if any of the rest of the fields
171 * to be filled in are needed.
174 (XFS_AT_XFLAGS
|XFS_AT_EXTSIZE
|XFS_AT_NEXTENTS
|XFS_AT_ANEXTENTS
|
175 XFS_AT_GENCOUNT
|XFS_AT_VCODE
)) == 0)
179 * Convert di_flags to xflags.
181 vap
->va_xflags
= xfs_ip2xflags(ip
);
184 * Exit for inode revalidate. See if any of the rest of
185 * the fields to be filled in are needed.
188 (XFS_AT_EXTSIZE
|XFS_AT_NEXTENTS
|XFS_AT_ANEXTENTS
|
189 XFS_AT_GENCOUNT
|XFS_AT_VCODE
)) == 0)
192 vap
->va_extsize
= ip
->i_d
.di_extsize
<< mp
->m_sb
.sb_blocklog
;
194 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) ?
195 ip
->i_df
.if_bytes
/ sizeof(xfs_bmbt_rec_t
) :
199 (ip
->i_afp
->if_flags
& XFS_IFEXTENTS
) ?
200 ip
->i_afp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
) :
201 ip
->i_d
.di_anextents
;
203 vap
->va_anextents
= 0;
204 vap
->va_gen
= ip
->i_d
.di_gen
;
207 if (!(flags
& ATTR_LAZY
))
208 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
234 xfs_prid_t projid
=0, iprojid
=0;
235 int mandlock_before
, mandlock_after
;
236 struct xfs_dquot
*udqp
, *gdqp
, *olddquot1
, *olddquot2
;
240 vp
= BHV_TO_VNODE(bdp
);
241 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
243 if (vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
)
244 return XFS_ERROR(EROFS
);
247 * Cannot set certain attributes.
250 if (mask
& XFS_AT_NOSET
) {
251 return XFS_ERROR(EINVAL
);
254 ip
= XFS_BHVTOI(bdp
);
257 if (XFS_FORCED_SHUTDOWN(mp
))
258 return XFS_ERROR(EIO
);
261 * Timestamps do not need to be logged and hence do not
262 * need to be done within a transaction.
264 if (mask
& XFS_AT_UPDTIMES
) {
265 ASSERT((mask
& ~XFS_AT_UPDTIMES
) == 0);
266 timeflags
= ((mask
& XFS_AT_UPDATIME
) ? XFS_ICHGTIME_ACC
: 0) |
267 ((mask
& XFS_AT_UPDCTIME
) ? XFS_ICHGTIME_CHG
: 0) |
268 ((mask
& XFS_AT_UPDMTIME
) ? XFS_ICHGTIME_MOD
: 0);
269 xfs_ichgtime(ip
, timeflags
);
273 olddquot1
= olddquot2
= NULL
;
277 * If disk quotas is on, we make sure that the dquots do exist on disk,
278 * before we start any other transactions. Trying to do this later
279 * is messy. We don't care to take a readlock to look at the ids
280 * in inode here, because we can't hold it across the trans_reserve.
281 * If the IDs do change before we take the ilock, we're covered
282 * because the i_*dquot fields will get updated anyway.
284 if (XFS_IS_QUOTA_ON(mp
) &&
285 (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
))) {
288 if ((mask
& XFS_AT_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
290 qflags
|= XFS_QMOPT_UQUOTA
;
292 uid
= ip
->i_d
.di_uid
;
294 if ((mask
& XFS_AT_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
296 qflags
|= XFS_QMOPT_GQUOTA
;
298 gid
= ip
->i_d
.di_gid
;
300 if ((mask
& XFS_AT_PROJID
) && XFS_IS_PQUOTA_ON(mp
)) {
301 projid
= vap
->va_projid
;
302 qflags
|= XFS_QMOPT_PQUOTA
;
304 projid
= ip
->i_d
.di_projid
;
307 * We take a reference when we initialize udqp and gdqp,
308 * so it is important that we never blindly double trip on
309 * the same variable. See xfs_create() for an example.
311 ASSERT(udqp
== NULL
);
312 ASSERT(gdqp
== NULL
);
313 code
= XFS_QM_DQVOPALLOC(mp
, ip
, uid
, gid
, projid
, qflags
,
320 * For the other attributes, we acquire the inode lock and
321 * first do an error checking pass.
324 lock_flags
= XFS_ILOCK_EXCL
;
325 if (flags
& ATTR_NOLOCK
)
327 if (!(mask
& XFS_AT_SIZE
)) {
328 if ((mask
!= (XFS_AT_CTIME
|XFS_AT_ATIME
|XFS_AT_MTIME
)) ||
329 (mp
->m_flags
& XFS_MOUNT_WSYNC
)) {
330 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
332 if ((code
= xfs_trans_reserve(tp
, 0,
333 XFS_ICHANGE_LOG_RES(mp
), 0,
340 if (DM_EVENT_ENABLED (vp
->v_vfsp
, ip
, DM_EVENT_TRUNCATE
) &&
341 !(flags
& ATTR_DMI
)) {
342 int dmflags
= AT_DELAY_FLAG(flags
) | DM_SEM_FLAG_WR
;
343 code
= XFS_SEND_DATA(mp
, DM_EVENT_TRUNCATE
, vp
,
344 vap
->va_size
, 0, dmflags
, NULL
);
351 lock_flags
|= XFS_IOLOCK_EXCL
;
354 xfs_ilock(ip
, lock_flags
);
356 /* boolean: are we the file owner? */
357 file_owner
= (current_fsuid(credp
) == ip
->i_d
.di_uid
);
360 * Change various properties of a file.
361 * Only the owner or users with CAP_FOWNER
362 * capability may do these things.
365 (XFS_AT_MODE
|XFS_AT_XFLAGS
|XFS_AT_EXTSIZE
|XFS_AT_UID
|
366 XFS_AT_GID
|XFS_AT_PROJID
)) {
368 * CAP_FOWNER overrides the following restrictions:
370 * The user ID of the calling process must be equal
371 * to the file owner ID, except in cases where the
372 * CAP_FSETID capability is applicable.
374 if (!file_owner
&& !capable(CAP_FOWNER
)) {
375 code
= XFS_ERROR(EPERM
);
380 * CAP_FSETID overrides the following restrictions:
382 * The effective user ID of the calling process shall match
383 * the file owner when setting the set-user-ID and
384 * set-group-ID bits on that file.
386 * The effective group ID or one of the supplementary group
387 * IDs of the calling process shall match the group owner of
388 * the file when setting the set-group-ID bit on that file
390 if (mask
& XFS_AT_MODE
) {
393 if ((vap
->va_mode
& S_ISUID
) && !file_owner
)
395 if ((vap
->va_mode
& S_ISGID
) &&
396 !in_group_p((gid_t
)ip
->i_d
.di_gid
))
399 /* Linux allows this, Irix doesn't. */
400 if ((vap
->va_mode
& S_ISVTX
) && !VN_ISDIR(vp
))
403 if (m
&& !capable(CAP_FSETID
))
409 * Change file ownership. Must be the owner or privileged.
410 * If the system was configured with the "restricted_chown"
411 * option, the owner is not permitted to give away the file,
412 * and can change the group id only to a group of which he
413 * or she is a member.
415 if (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
)) {
417 * These IDs could have changed since we last looked at them.
418 * But, we're assured that if the ownership did change
419 * while we didn't have the inode locked, inode's dquot(s)
420 * would have changed also.
422 iuid
= ip
->i_d
.di_uid
;
423 iprojid
= ip
->i_d
.di_projid
;
424 igid
= ip
->i_d
.di_gid
;
425 gid
= (mask
& XFS_AT_GID
) ? vap
->va_gid
: igid
;
426 uid
= (mask
& XFS_AT_UID
) ? vap
->va_uid
: iuid
;
427 projid
= (mask
& XFS_AT_PROJID
) ? (xfs_prid_t
)vap
->va_projid
:
431 * CAP_CHOWN overrides the following restrictions:
433 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
434 * shall override the restriction that a process cannot
435 * change the user ID of a file it owns and the restriction
436 * that the group ID supplied to the chown() function
437 * shall be equal to either the group ID or one of the
438 * supplementary group IDs of the calling process.
440 if (restricted_chown
&&
441 (iuid
!= uid
|| (igid
!= gid
&&
442 !in_group_p((gid_t
)gid
))) &&
443 !capable(CAP_CHOWN
)) {
444 code
= XFS_ERROR(EPERM
);
448 * Do a quota reservation only if uid/projid/gid is actually
451 if ((XFS_IS_UQUOTA_ON(mp
) && iuid
!= uid
) ||
452 (XFS_IS_PQUOTA_ON(mp
) && iprojid
!= projid
) ||
453 (XFS_IS_GQUOTA_ON(mp
) && igid
!= gid
)) {
455 code
= XFS_QM_DQVOPCHOWNRESV(mp
, tp
, ip
, udqp
, gdqp
,
456 capable(CAP_FOWNER
) ?
457 XFS_QMOPT_FORCE_RES
: 0);
458 if (code
) /* out of quota */
464 * Truncate file. Must have write permission and not be a directory.
466 if (mask
& XFS_AT_SIZE
) {
467 /* Short circuit the truncate case for zero length files */
468 if ((vap
->va_size
== 0) &&
469 (ip
->i_size
== 0) && (ip
->i_d
.di_nextents
== 0)) {
470 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
471 lock_flags
&= ~XFS_ILOCK_EXCL
;
472 if (mask
& XFS_AT_CTIME
)
473 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
479 code
= XFS_ERROR(EISDIR
);
481 } else if (!VN_ISREG(vp
)) {
482 code
= XFS_ERROR(EINVAL
);
486 * Make sure that the dquots are attached to the inode.
488 if ((code
= XFS_QM_DQATTACH(mp
, ip
, XFS_QMOPT_ILOCKED
)))
493 * Change file access or modified times.
495 if (mask
& (XFS_AT_ATIME
|XFS_AT_MTIME
)) {
497 if ((flags
& ATTR_UTIME
) &&
498 !capable(CAP_FOWNER
)) {
499 code
= XFS_ERROR(EPERM
);
506 * Change extent size or realtime flag.
508 if (mask
& (XFS_AT_EXTSIZE
|XFS_AT_XFLAGS
)) {
510 * Can't change extent size if any extents are allocated.
512 if (ip
->i_d
.di_nextents
&& (mask
& XFS_AT_EXTSIZE
) &&
513 ((ip
->i_d
.di_extsize
<< mp
->m_sb
.sb_blocklog
) !=
515 code
= XFS_ERROR(EINVAL
); /* EFBIG? */
520 * Can't change realtime flag if any extents are allocated.
522 if ((ip
->i_d
.di_nextents
|| ip
->i_delayed_blks
) &&
523 (mask
& XFS_AT_XFLAGS
) &&
524 (ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) !=
525 (vap
->va_xflags
& XFS_XFLAG_REALTIME
)) {
526 code
= XFS_ERROR(EINVAL
); /* EFBIG? */
530 * Extent size must be a multiple of the appropriate block
531 * size, if set at all.
533 if ((mask
& XFS_AT_EXTSIZE
) && vap
->va_extsize
!= 0) {
536 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) ||
537 ((mask
& XFS_AT_XFLAGS
) &&
538 (vap
->va_xflags
& XFS_XFLAG_REALTIME
))) {
539 size
= mp
->m_sb
.sb_rextsize
<<
540 mp
->m_sb
.sb_blocklog
;
542 size
= mp
->m_sb
.sb_blocksize
;
544 if (vap
->va_extsize
% size
) {
545 code
= XFS_ERROR(EINVAL
);
550 * If realtime flag is set then must have realtime data.
552 if ((mask
& XFS_AT_XFLAGS
) &&
553 (vap
->va_xflags
& XFS_XFLAG_REALTIME
)) {
554 if ((mp
->m_sb
.sb_rblocks
== 0) ||
555 (mp
->m_sb
.sb_rextsize
== 0) ||
556 (ip
->i_d
.di_extsize
% mp
->m_sb
.sb_rextsize
)) {
557 code
= XFS_ERROR(EINVAL
);
563 * Can't modify an immutable/append-only file unless
564 * we have appropriate permission.
566 if ((mask
& XFS_AT_XFLAGS
) &&
568 (XFS_DIFLAG_IMMUTABLE
|XFS_DIFLAG_APPEND
) ||
570 (XFS_XFLAG_IMMUTABLE
| XFS_XFLAG_APPEND
))) &&
571 !capable(CAP_LINUX_IMMUTABLE
)) {
572 code
= XFS_ERROR(EPERM
);
578 * Now we can make the changes. Before we join the inode
579 * to the transaction, if XFS_AT_SIZE is set then take care of
580 * the part of the truncation that must be done without the
581 * inode lock. This needs to be done before joining the inode
582 * to the transaction, because the inode cannot be unlocked
583 * once it is a part of the transaction.
585 if (mask
& XFS_AT_SIZE
) {
587 if ((vap
->va_size
> ip
->i_size
) &&
588 (flags
& ATTR_NOSIZETOK
) == 0) {
589 code
= xfs_igrow_start(ip
, vap
->va_size
, credp
);
591 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
592 vn_iowait(vp
); /* wait for the completion of any pending DIOs */
594 code
= xfs_itruncate_data(ip
, vap
->va_size
);
597 lock_flags
&= ~XFS_ILOCK_EXCL
;
598 ASSERT(lock_flags
== XFS_IOLOCK_EXCL
);
601 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
602 if ((code
= xfs_trans_reserve(tp
, 0,
603 XFS_ITRUNCATE_LOG_RES(mp
), 0,
604 XFS_TRANS_PERM_LOG_RES
,
605 XFS_ITRUNCATE_LOG_COUNT
))) {
606 xfs_trans_cancel(tp
, 0);
608 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
611 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
612 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
616 xfs_trans_ijoin(tp
, ip
, lock_flags
);
617 xfs_trans_ihold(tp
, ip
);
620 /* determine whether mandatory locking mode changes */
621 mandlock_before
= MANDLOCK(vp
, ip
->i_d
.di_mode
);
624 * Truncate file. Must have write permission and not be a directory.
626 if (mask
& XFS_AT_SIZE
) {
627 if (vap
->va_size
> ip
->i_size
) {
628 xfs_igrow_finish(tp
, ip
, vap
->va_size
,
629 !(flags
& ATTR_DMI
));
630 } else if ((vap
->va_size
<= ip
->i_size
) ||
631 ((vap
->va_size
== 0) && ip
->i_d
.di_nextents
)) {
633 * signal a sync transaction unless
634 * we're truncating an already unlinked
635 * file on a wsync filesystem
637 code
= xfs_itruncate_finish(&tp
, ip
,
638 (xfs_fsize_t
)vap
->va_size
,
640 ((ip
->i_d
.di_nlink
!= 0 ||
641 !(mp
->m_flags
& XFS_MOUNT_WSYNC
))
646 * Truncated "down", so we're removing references
647 * to old data here - if we now delay flushing for
648 * a long time, we expose ourselves unduly to the
649 * notorious NULL files problem. So, we mark this
650 * vnode and flush it when the file is closed, and
651 * do not wait the usual (long) time for writeout.
656 * Have to do this even if the file's size doesn't change.
658 timeflags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
662 * Change file access modes.
664 if (mask
& XFS_AT_MODE
) {
665 ip
->i_d
.di_mode
&= S_IFMT
;
666 ip
->i_d
.di_mode
|= vap
->va_mode
& ~S_IFMT
;
668 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
669 timeflags
|= XFS_ICHGTIME_CHG
;
673 * Change file ownership. Must be the owner or privileged.
674 * If the system was configured with the "restricted_chown"
675 * option, the owner is not permitted to give away the file,
676 * and can change the group id only to a group of which he
677 * or she is a member.
679 if (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
)) {
681 * CAP_FSETID overrides the following restrictions:
683 * The set-user-ID and set-group-ID bits of a file will be
684 * cleared upon successful return from chown()
686 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
687 !capable(CAP_FSETID
)) {
688 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
692 * Change the ownerships and register quota modifications
693 * in the transaction.
696 if (XFS_IS_UQUOTA_ON(mp
)) {
697 ASSERT(mask
& XFS_AT_UID
);
699 olddquot1
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
700 &ip
->i_udquot
, udqp
);
702 ip
->i_d
.di_uid
= uid
;
705 if (XFS_IS_GQUOTA_ON(mp
)) {
706 ASSERT(!XFS_IS_PQUOTA_ON(mp
));
707 ASSERT(mask
& XFS_AT_GID
);
709 olddquot2
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
710 &ip
->i_gdquot
, gdqp
);
712 ip
->i_d
.di_gid
= gid
;
714 if (iprojid
!= projid
) {
715 if (XFS_IS_PQUOTA_ON(mp
)) {
716 ASSERT(!XFS_IS_GQUOTA_ON(mp
));
717 ASSERT(mask
& XFS_AT_PROJID
);
719 olddquot2
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
720 &ip
->i_gdquot
, gdqp
);
722 ip
->i_d
.di_projid
= projid
;
724 * We may have to rev the inode as well as
725 * the superblock version number since projids didn't
726 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
728 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
)
729 xfs_bump_ino_vers2(tp
, ip
);
732 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
733 timeflags
|= XFS_ICHGTIME_CHG
;
738 * Change file access or modified times.
740 if (mask
& (XFS_AT_ATIME
|XFS_AT_MTIME
)) {
741 if (mask
& XFS_AT_ATIME
) {
742 ip
->i_d
.di_atime
.t_sec
= vap
->va_atime
.tv_sec
;
743 ip
->i_d
.di_atime
.t_nsec
= vap
->va_atime
.tv_nsec
;
744 ip
->i_update_core
= 1;
745 timeflags
&= ~XFS_ICHGTIME_ACC
;
747 if (mask
& XFS_AT_MTIME
) {
748 ip
->i_d
.di_mtime
.t_sec
= vap
->va_mtime
.tv_sec
;
749 ip
->i_d
.di_mtime
.t_nsec
= vap
->va_mtime
.tv_nsec
;
750 timeflags
&= ~XFS_ICHGTIME_MOD
;
751 timeflags
|= XFS_ICHGTIME_CHG
;
753 if (tp
&& (flags
& ATTR_UTIME
))
754 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
758 * Change XFS-added attributes.
760 if (mask
& (XFS_AT_EXTSIZE
|XFS_AT_XFLAGS
)) {
761 if (mask
& XFS_AT_EXTSIZE
) {
763 * Converting bytes to fs blocks.
765 ip
->i_d
.di_extsize
= vap
->va_extsize
>>
766 mp
->m_sb
.sb_blocklog
;
768 if (mask
& XFS_AT_XFLAGS
) {
771 /* can't set PREALLOC this way, just preserve it */
772 di_flags
= (ip
->i_d
.di_flags
& XFS_DIFLAG_PREALLOC
);
773 if (vap
->va_xflags
& XFS_XFLAG_IMMUTABLE
)
774 di_flags
|= XFS_DIFLAG_IMMUTABLE
;
775 if (vap
->va_xflags
& XFS_XFLAG_APPEND
)
776 di_flags
|= XFS_DIFLAG_APPEND
;
777 if (vap
->va_xflags
& XFS_XFLAG_SYNC
)
778 di_flags
|= XFS_DIFLAG_SYNC
;
779 if (vap
->va_xflags
& XFS_XFLAG_NOATIME
)
780 di_flags
|= XFS_DIFLAG_NOATIME
;
781 if (vap
->va_xflags
& XFS_XFLAG_NODUMP
)
782 di_flags
|= XFS_DIFLAG_NODUMP
;
783 if (vap
->va_xflags
& XFS_XFLAG_PROJINHERIT
)
784 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
785 if (vap
->va_xflags
& XFS_XFLAG_NODEFRAG
)
786 di_flags
|= XFS_DIFLAG_NODEFRAG
;
787 if (vap
->va_xflags
& XFS_XFLAG_FILESTREAM
)
788 di_flags
|= XFS_DIFLAG_FILESTREAM
;
789 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
790 if (vap
->va_xflags
& XFS_XFLAG_RTINHERIT
)
791 di_flags
|= XFS_DIFLAG_RTINHERIT
;
792 if (vap
->va_xflags
& XFS_XFLAG_NOSYMLINKS
)
793 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
794 if (vap
->va_xflags
& XFS_XFLAG_EXTSZINHERIT
)
795 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
796 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
797 if (vap
->va_xflags
& XFS_XFLAG_REALTIME
) {
798 di_flags
|= XFS_DIFLAG_REALTIME
;
799 ip
->i_iocore
.io_flags
|= XFS_IOCORE_RT
;
801 ip
->i_iocore
.io_flags
&= ~XFS_IOCORE_RT
;
803 if (vap
->va_xflags
& XFS_XFLAG_EXTSIZE
)
804 di_flags
|= XFS_DIFLAG_EXTSIZE
;
806 ip
->i_d
.di_flags
= di_flags
;
808 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
809 timeflags
|= XFS_ICHGTIME_CHG
;
813 * Change file inode change time only if XFS_AT_CTIME set
814 * AND we have been called by a DMI function.
817 if ( (flags
& ATTR_DMI
) && (mask
& XFS_AT_CTIME
) ) {
818 ip
->i_d
.di_ctime
.t_sec
= vap
->va_ctime
.tv_sec
;
819 ip
->i_d
.di_ctime
.t_nsec
= vap
->va_ctime
.tv_nsec
;
820 ip
->i_update_core
= 1;
821 timeflags
&= ~XFS_ICHGTIME_CHG
;
825 * Send out timestamp changes that need to be set to the
826 * current time. Not done when called by a DMI function.
828 if (timeflags
&& !(flags
& ATTR_DMI
))
829 xfs_ichgtime(ip
, timeflags
);
831 XFS_STATS_INC(xs_ig_attrchg
);
834 * If this is a synchronous mount, make sure that the
835 * transaction goes to disk before returning to the user.
836 * This is slightly sub-optimal in that truncates require
837 * two sync transactions instead of one for wsync filesystems.
838 * One for the truncate and one for the timestamps since we
839 * don't want to change the timestamps unless we're sure the
840 * truncate worked. Truncates are less than 1% of the laddis
841 * mix so this probably isn't worth the trouble to optimize.
845 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
846 xfs_trans_set_sync(tp
);
848 code
= xfs_trans_commit(tp
, commit_flags
);
852 * If the (regular) file's mandatory locking mode changed, then
853 * notify the vnode. We do this under the inode lock to prevent
854 * racing calls to vop_vnode_change.
856 mandlock_after
= MANDLOCK(vp
, ip
->i_d
.di_mode
);
857 if (mandlock_before
!= mandlock_after
) {
858 bhv_vop_vnode_change(vp
, VCHANGE_FLAGS_ENF_LOCKING
,
862 xfs_iunlock(ip
, lock_flags
);
865 * Release any dquot(s) the inode had kept before chown.
867 XFS_QM_DQRELE(mp
, olddquot1
);
868 XFS_QM_DQRELE(mp
, olddquot2
);
869 XFS_QM_DQRELE(mp
, udqp
);
870 XFS_QM_DQRELE(mp
, gdqp
);
876 if (DM_EVENT_ENABLED(vp
->v_vfsp
, ip
, DM_EVENT_ATTRIBUTE
) &&
877 !(flags
& ATTR_DMI
)) {
878 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_ATTRIBUTE
, vp
, DM_RIGHT_NULL
,
879 NULL
, DM_RIGHT_NULL
, NULL
, NULL
,
880 0, 0, AT_DELAY_FLAG(flags
));
885 commit_flags
|= XFS_TRANS_ABORT
;
888 XFS_QM_DQRELE(mp
, udqp
);
889 XFS_QM_DQRELE(mp
, gdqp
);
891 xfs_trans_cancel(tp
, commit_flags
);
893 if (lock_flags
!= 0) {
894 xfs_iunlock(ip
, lock_flags
);
902 * Null conversion from vnode mode bits to inode mode bits, as in efs.
913 vn_trace_entry(BHV_TO_VNODE(bdp
), __FUNCTION__
,
914 (inst_t
*)__return_address
);
916 ip
= XFS_BHVTOI(bdp
);
917 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
918 error
= xfs_iaccess(ip
, mode
, credp
);
919 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
925 * The maximum pathlen is 1024 bytes. Since the minimum file system
926 * blocksize is 512 bytes, we can get a max of 2 extents back from
929 #define SYMLINK_MAPS 2
950 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
956 vp
= BHV_TO_VNODE(bdp
);
957 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
959 ip
= XFS_BHVTOI(bdp
);
962 if (XFS_FORCED_SHUTDOWN(mp
))
963 return XFS_ERROR(EIO
);
965 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
967 ASSERT((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
);
969 offset
= uiop
->uio_offset
;
970 count
= uiop
->uio_resid
;
973 error
= XFS_ERROR(EINVAL
);
982 * See if the symlink is stored inline.
984 pathlen
= (int)ip
->i_d
.di_size
;
986 if (ip
->i_df
.if_flags
& XFS_IFINLINE
) {
987 error
= xfs_uio_read(ip
->i_df
.if_u1
.if_data
, pathlen
, uiop
);
991 * Symlink not inline. Call bmap to get it in.
993 nmaps
= SYMLINK_MAPS
;
995 error
= xfs_bmapi(NULL
, ip
, 0, XFS_B_TO_FSB(mp
, pathlen
),
996 0, NULL
, 0, mval
, &nmaps
, NULL
, NULL
);
1002 for (n
= 0; n
< nmaps
; n
++) {
1003 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
1004 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
1005 bp
= xfs_buf_read(mp
->m_ddev_targp
, d
,
1006 BTOBB(byte_cnt
), 0);
1007 error
= XFS_BUF_GETERROR(bp
);
1009 xfs_ioerror_alert("xfs_readlink",
1010 ip
->i_mount
, bp
, XFS_BUF_ADDR(bp
));
1014 if (pathlen
< byte_cnt
)
1016 pathlen
-= byte_cnt
;
1018 error
= xfs_uio_read(XFS_BUF_PTR(bp
), byte_cnt
, uiop
);
1025 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1033 * This is called to sync the inode and its data out to disk.
1034 * We need to hold the I/O lock while flushing the data, and
1035 * the inode lock while flushing the inode. The inode lock CANNOT
1036 * be held while flushing the data, so acquire after we're done
1050 int log_flushed
= 0, changed
= 1;
1052 vn_trace_entry(BHV_TO_VNODE(bdp
),
1053 __FUNCTION__
, (inst_t
*)__return_address
);
1055 ip
= XFS_BHVTOI(bdp
);
1057 ASSERT(start
>= 0 && stop
>= -1);
1059 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
1060 return XFS_ERROR(EIO
);
1063 * We always need to make sure that the required inode state
1064 * is safe on disk. The vnode might be clean but because
1065 * of committed transactions that haven't hit the disk yet.
1066 * Likewise, there could be unflushed non-transactional
1067 * changes to the inode core that have to go to disk.
1069 * The following code depends on one assumption: that
1070 * any transaction that changes an inode logs the core
1071 * because it has to change some field in the inode core
1072 * (typically nextents or nblocks). That assumption
1073 * implies that any transactions against an inode will
1074 * catch any non-transactional updates. If inode-altering
1075 * transactions exist that violate this assumption, the
1076 * code breaks. Right now, it figures that if the involved
1077 * update_* field is clear and the inode is unpinned, the
1078 * inode is clean. Either it's been flushed or it's been
1079 * committed and the commit has hit the disk unpinning the inode.
1080 * (Note that xfs_inode_item_format() called at commit clears
1081 * the update_* fields.)
1083 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1085 /* If we are flushing data then we care about update_size
1086 * being set, otherwise we care about update_core
1088 if ((flag
& FSYNC_DATA
) ?
1089 (ip
->i_update_size
== 0) :
1090 (ip
->i_update_core
== 0)) {
1092 * Timestamps/size haven't changed since last inode
1093 * flush or inode transaction commit. That means
1094 * either nothing got written or a transaction
1095 * committed which caught the updates. If the
1096 * latter happened and the transaction hasn't
1097 * hit the disk yet, the inode will be still
1098 * be pinned. If it is, force the log.
1101 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1103 if (xfs_ipincount(ip
)) {
1104 _xfs_log_force(ip
->i_mount
, (xfs_lsn_t
)0,
1106 ((flag
& FSYNC_WAIT
)
1107 ? XFS_LOG_SYNC
: 0),
1111 * If the inode is not pinned and nothing
1112 * has changed we don't need to flush the
1120 * Kick off a transaction to log the inode
1121 * core to get the updates. Make it
1122 * sync if FSYNC_WAIT is passed in (which
1123 * is done by everybody but specfs). The
1124 * sync transaction will also force the log.
1126 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1127 tp
= xfs_trans_alloc(ip
->i_mount
, XFS_TRANS_FSYNC_TS
);
1128 if ((error
= xfs_trans_reserve(tp
, 0,
1129 XFS_FSYNC_TS_LOG_RES(ip
->i_mount
),
1131 xfs_trans_cancel(tp
, 0);
1134 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1137 * Note - it's possible that we might have pushed
1138 * ourselves out of the way during trans_reserve
1139 * which would flush the inode. But there's no
1140 * guarantee that the inode buffer has actually
1141 * gone out yet (it's delwri). Plus the buffer
1142 * could be pinned anyway if it's part of an
1143 * inode in another recent transaction. So we
1144 * play it safe and fire off the transaction anyway.
1146 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1147 xfs_trans_ihold(tp
, ip
);
1148 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1149 if (flag
& FSYNC_WAIT
)
1150 xfs_trans_set_sync(tp
);
1151 error
= _xfs_trans_commit(tp
, 0, &log_flushed
);
1153 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1156 if ((ip
->i_mount
->m_flags
& XFS_MOUNT_BARRIER
) && changed
) {
1158 * If the log write didn't issue an ordered tag we need
1159 * to flush the disk cache for the data device now.
1162 xfs_blkdev_issue_flush(ip
->i_mount
->m_ddev_targp
);
1165 * If this inode is on the RT dev we need to flush that
1168 if (ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
)
1169 xfs_blkdev_issue_flush(ip
->i_mount
->m_rtdev_targp
);
1176 * This is called by xfs_inactive to free any blocks beyond eof
1177 * when the link count isn't zero and by xfs_dm_punch_hole() when
1178 * punching a hole to EOF.
1188 xfs_fileoff_t end_fsb
;
1189 xfs_fileoff_t last_fsb
;
1190 xfs_filblks_t map_len
;
1192 xfs_bmbt_irec_t imap
;
1193 int use_iolock
= (flags
& XFS_FREE_EOF_LOCK
);
1196 * Figure out if there are any blocks beyond the end
1197 * of the file. If not, then there is nothing to do.
1199 end_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)ip
->i_size
));
1200 last_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1201 map_len
= last_fsb
- end_fsb
;
1206 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1207 error
= XFS_BMAPI(mp
, NULL
, &ip
->i_iocore
, end_fsb
, map_len
, 0,
1208 NULL
, 0, &imap
, &nimaps
, NULL
, NULL
);
1209 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1211 if (!error
&& (nimaps
!= 0) &&
1212 (imap
.br_startblock
!= HOLESTARTBLOCK
||
1213 ip
->i_delayed_blks
)) {
1215 * Attach the dquots to the inode up front.
1217 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
1221 * There are blocks after the end of file.
1222 * Free them up now by truncating the file to
1225 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1228 * Do the xfs_itruncate_start() call before
1229 * reserving any log space because
1230 * itruncate_start will call into the buffer
1231 * cache and we can't
1232 * do that within a transaction.
1235 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1236 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
,
1239 xfs_trans_cancel(tp
, 0);
1241 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1245 error
= xfs_trans_reserve(tp
, 0,
1246 XFS_ITRUNCATE_LOG_RES(mp
),
1247 0, XFS_TRANS_PERM_LOG_RES
,
1248 XFS_ITRUNCATE_LOG_COUNT
);
1250 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1251 xfs_trans_cancel(tp
, 0);
1252 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1256 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1257 xfs_trans_ijoin(tp
, ip
,
1260 xfs_trans_ihold(tp
, ip
);
1262 error
= xfs_itruncate_finish(&tp
, ip
,
1267 * If we get an error at this point we
1268 * simply don't bother truncating the file.
1271 xfs_trans_cancel(tp
,
1272 (XFS_TRANS_RELEASE_LOG_RES
|
1275 error
= xfs_trans_commit(tp
,
1276 XFS_TRANS_RELEASE_LOG_RES
);
1278 xfs_iunlock(ip
, (use_iolock
? (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)
1285 * Free a symlink that has blocks associated with it.
1288 xfs_inactive_symlink_rmt(
1296 xfs_fsblock_t first_block
;
1297 xfs_bmap_free_t free_list
;
1300 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
1308 ASSERT(ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
));
1310 * We're freeing a symlink that has some
1311 * blocks allocated to it. Free the
1312 * blocks here. We know that we've got
1313 * either 1 or 2 extents and that we can
1314 * free them all in one bunmapi call.
1316 ASSERT(ip
->i_d
.di_nextents
> 0 && ip
->i_d
.di_nextents
<= 2);
1317 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1318 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
1319 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1320 xfs_trans_cancel(tp
, 0);
1325 * Lock the inode, fix the size, and join it to the transaction.
1326 * Hold it so in the normal path, we still have it locked for
1327 * the second transaction. In the error paths we need it
1328 * held so the cancel won't rele it, see below.
1330 xfs_ilock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1331 size
= (int)ip
->i_d
.di_size
;
1332 ip
->i_d
.di_size
= 0;
1333 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1334 xfs_trans_ihold(tp
, ip
);
1335 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1337 * Find the block(s) so we can inval and unmap them.
1340 XFS_BMAP_INIT(&free_list
, &first_block
);
1341 nmaps
= ARRAY_SIZE(mval
);
1342 if ((error
= xfs_bmapi(tp
, ip
, 0, XFS_B_TO_FSB(mp
, size
),
1343 XFS_BMAPI_METADATA
, &first_block
, 0, mval
, &nmaps
,
1347 * Invalidate the block(s).
1349 for (i
= 0; i
< nmaps
; i
++) {
1350 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
1351 XFS_FSB_TO_DADDR(mp
, mval
[i
].br_startblock
),
1352 XFS_FSB_TO_BB(mp
, mval
[i
].br_blockcount
), 0);
1353 xfs_trans_binval(tp
, bp
);
1356 * Unmap the dead block(s) to the free_list.
1358 if ((error
= xfs_bunmapi(tp
, ip
, 0, size
, XFS_BMAPI_METADATA
, nmaps
,
1359 &first_block
, &free_list
, NULL
, &done
)))
1363 * Commit the first transaction. This logs the EFI and the inode.
1365 if ((error
= xfs_bmap_finish(&tp
, &free_list
, &committed
)))
1368 * The transaction must have been committed, since there were
1369 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1370 * The new tp has the extent freeing and EFDs.
1374 * The first xact was committed, so add the inode to the new one.
1375 * Mark it dirty so it will be logged and moved forward in the log as
1376 * part of every commit.
1378 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1379 xfs_trans_ihold(tp
, ip
);
1380 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1382 * Get a new, empty transaction to return to our caller.
1384 ntp
= xfs_trans_dup(tp
);
1386 * Commit the transaction containing extent freeing and EFDs.
1387 * If we get an error on the commit here or on the reserve below,
1388 * we need to unlock the inode since the new transaction doesn't
1389 * have the inode attached.
1391 error
= xfs_trans_commit(tp
, 0);
1394 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1398 * Remove the memory for extent descriptions (just bookkeeping).
1400 if (ip
->i_df
.if_bytes
)
1401 xfs_idata_realloc(ip
, -ip
->i_df
.if_bytes
, XFS_DATA_FORK
);
1402 ASSERT(ip
->i_df
.if_bytes
== 0);
1404 * Put an itruncate log reservation in the new transaction
1407 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1408 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
1409 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1413 * Return with the inode locked but not joined to the transaction.
1419 xfs_bmap_cancel(&free_list
);
1422 * Have to come here with the inode locked and either
1423 * (held and in the transaction) or (not in the transaction).
1424 * If the inode isn't held then cancel would iput it, but
1425 * that's wrong since this is inactive and the vnode ref
1426 * count is 0 already.
1427 * Cancel won't do anything to the inode if held, but it still
1428 * needs to be locked until the cancel is done, if it was
1429 * joined to the transaction.
1431 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1432 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1439 xfs_inactive_symlink_local(
1445 ASSERT(ip
->i_d
.di_size
<= XFS_IFORK_DSIZE(ip
));
1447 * We're freeing a symlink which fit into
1448 * the inode. Just free the memory used
1449 * to hold the old symlink.
1451 error
= xfs_trans_reserve(*tpp
, 0,
1452 XFS_ITRUNCATE_LOG_RES(ip
->i_mount
),
1453 0, XFS_TRANS_PERM_LOG_RES
,
1454 XFS_ITRUNCATE_LOG_COUNT
);
1457 xfs_trans_cancel(*tpp
, 0);
1461 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1464 * Zero length symlinks _can_ exist.
1466 if (ip
->i_df
.if_bytes
> 0) {
1467 xfs_idata_realloc(ip
,
1468 -(ip
->i_df
.if_bytes
),
1470 ASSERT(ip
->i_df
.if_bytes
== 0);
1484 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
));
1487 ASSERT(ip
->i_d
.di_forkoff
!= 0);
1488 xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1489 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1491 error
= xfs_attr_inactive(ip
);
1494 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1495 return error
; /* goto out */
1498 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1499 error
= xfs_trans_reserve(tp
, 0,
1500 XFS_IFREE_LOG_RES(mp
),
1501 0, XFS_TRANS_PERM_LOG_RES
,
1502 XFS_INACTIVE_LOG_COUNT
);
1504 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1505 xfs_trans_cancel(tp
, 0);
1507 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1511 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1512 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1513 xfs_trans_ihold(tp
, ip
);
1514 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1516 ASSERT(ip
->i_d
.di_anextents
== 0);
1531 vp
= BHV_TO_VNODE(bdp
);
1532 ip
= XFS_BHVTOI(bdp
);
1535 if (!VN_ISREG(vp
) || (ip
->i_d
.di_mode
== 0))
1538 /* If this is a read-only mount, don't do this (would generate I/O) */
1539 if (vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
)
1542 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1544 * If we are using filestreams, and we have an unlinked
1545 * file that we are processing the last close on, then nothing
1546 * will be able to reopen and write to this file. Purge this
1547 * inode from the filestreams cache so that it doesn't delay
1548 * teardown of the inode.
1550 if ((ip
->i_d
.di_nlink
== 0) && xfs_inode_is_filestream(ip
))
1551 xfs_filestream_deassociate(ip
);
1554 * If we previously truncated this file and removed old data
1555 * in the process, we want to initiate "early" writeout on
1556 * the last close. This is an attempt to combat the notorious
1557 * NULL files problem which is particularly noticable from a
1558 * truncate down, buffered (re-)write (delalloc), followed by
1559 * a crash. What we are effectively doing here is
1560 * significantly reducing the time window where we'd otherwise
1561 * be exposed to that problem.
1563 if (VUNTRUNCATE(vp
) && VN_DIRTY(vp
) && ip
->i_delayed_blks
> 0)
1564 bhv_vop_flush_pages(vp
, 0, -1, XFS_B_ASYNC
, FI_NONE
);
1567 #ifdef HAVE_REFCACHE
1568 /* If we are in the NFS reference cache then don't do this now */
1573 if (ip
->i_d
.di_nlink
!= 0) {
1574 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1575 ((ip
->i_size
> 0) || (VN_CACHED(vp
) > 0 ||
1576 ip
->i_delayed_blks
> 0)) &&
1577 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
)) &&
1578 (!(ip
->i_d
.di_flags
&
1579 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)))) {
1580 error
= xfs_free_eofblocks(mp
, ip
, XFS_FREE_EOF_LOCK
);
1583 /* Update linux inode block count after free above */
1584 vn_to_inode(vp
)->i_blocks
= XFS_FSB_TO_BB(mp
,
1585 ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
1595 * This is called when the vnode reference count for the vnode
1596 * goes to zero. If the file has been unlinked, then it must
1597 * now be truncated. Also, we clear all of the read-ahead state
1598 * kept for the inode here since the file is now closed.
1607 xfs_bmap_free_t free_list
;
1608 xfs_fsblock_t first_block
;
1615 vp
= BHV_TO_VNODE(bdp
);
1616 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
1618 ip
= XFS_BHVTOI(bdp
);
1621 * If the inode is already free, then there can be nothing
1624 if (ip
->i_d
.di_mode
== 0 || VN_BAD(vp
)) {
1625 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1626 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1627 return VN_INACTIVE_CACHE
;
1631 * Only do a truncate if it's a regular file with
1632 * some actual space in it. It's OK to look at the
1633 * inode's fields without the lock because we're the
1634 * only one with a reference to the inode.
1636 truncate
= ((ip
->i_d
.di_nlink
== 0) &&
1637 ((ip
->i_d
.di_size
!= 0) || (ip
->i_size
!= 0) ||
1638 (ip
->i_d
.di_nextents
> 0) || (ip
->i_delayed_blks
> 0)) &&
1639 ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
));
1643 if (ip
->i_d
.di_nlink
== 0 &&
1644 DM_EVENT_ENABLED(vp
->v_vfsp
, ip
, DM_EVENT_DESTROY
)) {
1645 (void) XFS_SEND_DESTROY(mp
, vp
, DM_RIGHT_NULL
);
1650 /* If this is a read-only mount, don't do this (would generate I/O) */
1651 if (vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
)
1654 if (ip
->i_d
.di_nlink
!= 0) {
1655 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1656 ((ip
->i_size
> 0) || (VN_CACHED(vp
) > 0 ||
1657 ip
->i_delayed_blks
> 0)) &&
1658 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) &&
1659 (!(ip
->i_d
.di_flags
&
1660 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)) ||
1661 (ip
->i_delayed_blks
!= 0)))) {
1662 error
= xfs_free_eofblocks(mp
, ip
, XFS_FREE_EOF_LOCK
);
1664 return VN_INACTIVE_CACHE
;
1665 /* Update linux inode block count after free above */
1666 vn_to_inode(vp
)->i_blocks
= XFS_FSB_TO_BB(mp
,
1667 ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
1672 ASSERT(ip
->i_d
.di_nlink
== 0);
1674 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
1675 return VN_INACTIVE_CACHE
;
1677 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1680 * Do the xfs_itruncate_start() call before
1681 * reserving any log space because itruncate_start
1682 * will call into the buffer cache and we can't
1683 * do that within a transaction.
1685 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1687 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
, 0);
1689 xfs_trans_cancel(tp
, 0);
1690 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1691 return VN_INACTIVE_CACHE
;
1694 error
= xfs_trans_reserve(tp
, 0,
1695 XFS_ITRUNCATE_LOG_RES(mp
),
1696 0, XFS_TRANS_PERM_LOG_RES
,
1697 XFS_ITRUNCATE_LOG_COUNT
);
1699 /* Don't call itruncate_cleanup */
1700 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1701 xfs_trans_cancel(tp
, 0);
1702 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1703 return VN_INACTIVE_CACHE
;
1706 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1707 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1708 xfs_trans_ihold(tp
, ip
);
1711 * normally, we have to run xfs_itruncate_finish sync.
1712 * But if filesystem is wsync and we're in the inactive
1713 * path, then we know that nlink == 0, and that the
1714 * xaction that made nlink == 0 is permanently committed
1715 * since xfs_remove runs as a synchronous transaction.
1717 error
= xfs_itruncate_finish(&tp
, ip
, 0, XFS_DATA_FORK
,
1718 (!(mp
->m_flags
& XFS_MOUNT_WSYNC
) ? 1 : 0));
1721 xfs_trans_cancel(tp
,
1722 XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1723 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1724 return VN_INACTIVE_CACHE
;
1726 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
) {
1729 * If we get an error while cleaning up a
1730 * symlink we bail out.
1732 error
= (ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
)) ?
1733 xfs_inactive_symlink_rmt(ip
, &tp
) :
1734 xfs_inactive_symlink_local(ip
, &tp
);
1738 return VN_INACTIVE_CACHE
;
1741 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1742 xfs_trans_ihold(tp
, ip
);
1744 error
= xfs_trans_reserve(tp
, 0,
1745 XFS_IFREE_LOG_RES(mp
),
1746 0, XFS_TRANS_PERM_LOG_RES
,
1747 XFS_INACTIVE_LOG_COUNT
);
1749 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1750 xfs_trans_cancel(tp
, 0);
1751 return VN_INACTIVE_CACHE
;
1754 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1755 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1756 xfs_trans_ihold(tp
, ip
);
1760 * If there are attributes associated with the file
1761 * then blow them away now. The code calls a routine
1762 * that recursively deconstructs the attribute fork.
1763 * We need to just commit the current transaction
1764 * because we can't use it for xfs_attr_inactive().
1766 if (ip
->i_d
.di_anextents
> 0) {
1767 error
= xfs_inactive_attrs(ip
, &tp
);
1769 * If we got an error, the transaction is already
1770 * cancelled, and the inode is unlocked. Just get out.
1773 return VN_INACTIVE_CACHE
;
1774 } else if (ip
->i_afp
) {
1775 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1781 XFS_BMAP_INIT(&free_list
, &first_block
);
1782 error
= xfs_ifree(tp
, ip
, &free_list
);
1785 * If we fail to free the inode, shut down. The cancel
1786 * might do that, we need to make sure. Otherwise the
1787 * inode might be lost for a long time or forever.
1789 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1791 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1792 error
, mp
->m_fsname
);
1793 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1795 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
1798 * Credit the quota account(s). The inode is gone.
1800 XFS_TRANS_MOD_DQUOT_BYINO(mp
, tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1803 * Just ignore errors at this point. There is
1804 * nothing we can do except to try to keep going.
1806 (void) xfs_bmap_finish(&tp
, &free_list
, &committed
);
1807 (void) xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1810 * Release the dquots held by inode, if any.
1812 XFS_QM_DQDETACH(mp
, ip
);
1814 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1817 return VN_INACTIVE_CACHE
;
1826 bhv_desc_t
*dir_bdp
,
1827 bhv_vname_t
*dentry
,
1833 xfs_inode_t
*dp
, *ip
;
1837 bhv_vnode_t
*dir_vp
;
1839 dir_vp
= BHV_TO_VNODE(dir_bdp
);
1840 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
1842 dp
= XFS_BHVTOI(dir_bdp
);
1844 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
1845 return XFS_ERROR(EIO
);
1847 lock_mode
= xfs_ilock_map_shared(dp
);
1848 error
= xfs_dir_lookup_int(dir_bdp
, lock_mode
, dentry
, &e_inum
, &ip
);
1850 *vpp
= XFS_ITOV(ip
);
1853 xfs_iunlock_map_shared(dp
, lock_mode
);
1859 * xfs_create (create a new file).
1863 bhv_desc_t
*dir_bdp
,
1864 bhv_vname_t
*dentry
,
1869 char *name
= VNAME(dentry
);
1870 bhv_vnode_t
*dir_vp
;
1871 xfs_inode_t
*dp
, *ip
;
1872 bhv_vnode_t
*vp
= NULL
;
1877 xfs_bmap_free_t free_list
;
1878 xfs_fsblock_t first_block
;
1879 boolean_t dp_joined_to_trans
;
1880 int dm_event_sent
= 0;
1884 struct xfs_dquot
*udqp
, *gdqp
;
1890 dir_vp
= BHV_TO_VNODE(dir_bdp
);
1891 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
1893 dp
= XFS_BHVTOI(dir_bdp
);
1896 dm_di_mode
= vap
->va_mode
;
1897 namelen
= VNAMELEN(dentry
);
1899 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_CREATE
)) {
1900 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_CREATE
,
1901 dir_vp
, DM_RIGHT_NULL
, NULL
,
1902 DM_RIGHT_NULL
, name
, NULL
,
1910 if (XFS_FORCED_SHUTDOWN(mp
))
1911 return XFS_ERROR(EIO
);
1913 /* Return through std_return after this point. */
1916 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1917 prid
= dp
->i_d
.di_projid
;
1918 else if (vap
->va_mask
& XFS_AT_PROJID
)
1919 prid
= (xfs_prid_t
)vap
->va_projid
;
1921 prid
= (xfs_prid_t
)dfltprid
;
1924 * Make sure that we have allocated dquot(s) on disk.
1926 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
1927 current_fsuid(credp
), current_fsgid(credp
), prid
,
1928 XFS_QMOPT_QUOTALL
|XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
1933 dp_joined_to_trans
= B_FALSE
;
1935 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1936 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1937 resblks
= XFS_CREATE_SPACE_RES(mp
, namelen
);
1939 * Initially assume that the file does not exist and
1940 * reserve the resources for that case. If that is not
1941 * the case we'll drop the one we have and get a more
1942 * appropriate transaction later.
1944 error
= xfs_trans_reserve(tp
, resblks
, XFS_CREATE_LOG_RES(mp
), 0,
1945 XFS_TRANS_PERM_LOG_RES
, XFS_CREATE_LOG_COUNT
);
1946 if (error
== ENOSPC
) {
1948 error
= xfs_trans_reserve(tp
, 0, XFS_CREATE_LOG_RES(mp
), 0,
1949 XFS_TRANS_PERM_LOG_RES
, XFS_CREATE_LOG_COUNT
);
1957 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1959 XFS_BMAP_INIT(&free_list
, &first_block
);
1964 * Reserve disk quota and the inode.
1966 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
1970 if (resblks
== 0 && (error
= xfs_dir_canenter(tp
, dp
, name
, namelen
)))
1972 rdev
= (vap
->va_mask
& XFS_AT_RDEV
) ? vap
->va_rdev
: 0;
1973 error
= xfs_dir_ialloc(&tp
, dp
, vap
->va_mode
, 1,
1974 rdev
, credp
, prid
, resblks
> 0,
1977 if (error
== ENOSPC
)
1984 * At this point, we've gotten a newly allocated inode.
1985 * It is locked (and joined to the transaction).
1988 ASSERT(ismrlocked (&ip
->i_lock
, MR_UPDATE
));
1991 * Now we join the directory inode to the transaction.
1992 * We do not do it earlier because xfs_dir_ialloc
1993 * might commit the previous transaction (and release
1998 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1999 dp_joined_to_trans
= B_TRUE
;
2001 error
= xfs_dir_createname(tp
, dp
, name
, namelen
, ip
->i_ino
,
2002 &first_block
, &free_list
, resblks
?
2003 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
2005 ASSERT(error
!= ENOSPC
);
2008 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2009 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2012 * If this is a synchronous mount, make sure that the
2013 * create transaction goes to disk before returning to
2016 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2017 xfs_trans_set_sync(tp
);
2023 * Attach the dquot(s) to the inodes and modify them incore.
2024 * These ids of the inode couldn't have changed since the new
2025 * inode has been locked ever since it was created.
2027 XFS_QM_DQVOPCREATE(mp
, tp
, ip
, udqp
, gdqp
);
2030 * xfs_trans_commit normally decrements the vnode ref count
2031 * when it unlocks the inode. Since we want to return the
2032 * vnode to the caller, we bump the vnode ref count now.
2037 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2039 xfs_bmap_cancel(&free_list
);
2043 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2050 XFS_QM_DQRELE(mp
, udqp
);
2051 XFS_QM_DQRELE(mp
, gdqp
);
2054 * Propagate the fact that the vnode changed after the
2055 * xfs_inode locks have been released.
2057 bhv_vop_vnode_change(vp
, VCHANGE_FLAGS_TRUNCATED
, 3);
2061 /* Fallthrough to std_return with error = 0 */
2064 if ( (*vpp
|| (error
!= 0 && dm_event_sent
!= 0)) &&
2065 DM_EVENT_ENABLED(dir_vp
->v_vfsp
, XFS_BHVTOI(dir_bdp
),
2066 DM_EVENT_POSTCREATE
)) {
2067 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTCREATE
,
2068 dir_vp
, DM_RIGHT_NULL
,
2070 DM_RIGHT_NULL
, name
, NULL
,
2071 dm_di_mode
, error
, 0);
2076 cancel_flags
|= XFS_TRANS_ABORT
;
2081 xfs_trans_cancel(tp
, cancel_flags
);
2083 if (!dp_joined_to_trans
&& (dp
!= NULL
))
2084 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2085 XFS_QM_DQRELE(mp
, udqp
);
2086 XFS_QM_DQRELE(mp
, gdqp
);
2092 * Wait until after the current transaction is aborted to
2093 * release the inode. This prevents recursive transactions
2094 * and deadlocks from xfs_inactive.
2096 cancel_flags
|= XFS_TRANS_ABORT
;
2097 xfs_trans_cancel(tp
, cancel_flags
);
2100 XFS_QM_DQRELE(mp
, udqp
);
2101 XFS_QM_DQRELE(mp
, gdqp
);
2108 * Some counters to see if (and how often) we are hitting some deadlock
2109 * prevention code paths.
2113 int xfs_rm_lock_delays
;
2114 int xfs_rm_attempts
;
2118 * The following routine will lock the inodes associated with the
2119 * directory and the named entry in the directory. The locks are
2120 * acquired in increasing inode number.
2122 * If the entry is "..", then only the directory is locked. The
2123 * vnode ref count will still include that from the .. entry in
2126 * There is a deadlock we need to worry about. If the locked directory is
2127 * in the AIL, it might be blocking up the log. The next inode we lock
2128 * could be already locked by another thread waiting for log space (e.g
2129 * a permanent log reservation with a long running transaction (see
2130 * xfs_itruncate_finish)). To solve this, we must check if the directory
2131 * is in the ail and use lock_nowait. If we can't lock, we need to
2132 * drop the inode lock on the directory and try again. xfs_iunlock will
2133 * potentially push the tail if we were holding up the log.
2136 xfs_lock_dir_and_entry(
2138 xfs_inode_t
*ip
) /* inode of entry 'name' */
2142 xfs_inode_t
*ips
[2];
2151 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
2158 * We want to lock in increasing inum. Since we've already
2159 * acquired the lock on the directory, we may need to release
2160 * if if the inum of the entry turns out to be less.
2162 if (e_inum
> dp
->i_ino
) {
2164 * We are already in the right order, so just
2165 * lock on the inode of the entry.
2166 * We need to use nowait if dp is in the AIL.
2169 lp
= (xfs_log_item_t
*)dp
->i_itemp
;
2170 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
2171 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2178 * Unlock dp and try again.
2179 * xfs_iunlock will try to push the tail
2180 * if the inode is in the AIL.
2183 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2185 if ((attempts
% 5) == 0) {
2186 delay(1); /* Don't just spin the CPU */
2188 xfs_rm_lock_delays
++;
2194 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2196 } else if (e_inum
< dp
->i_ino
) {
2197 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2201 xfs_lock_inodes(ips
, 2, 0, XFS_ILOCK_EXCL
);
2203 /* else e_inum == dp->i_ino */
2204 /* This can happen if we're asked to lock /x/..
2205 * the entry is "..", which is also the parent directory.
2213 int xfs_small_retries
;
2214 int xfs_middle_retries
;
2215 int xfs_lots_retries
;
2216 int xfs_lock_delays
;
2220 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2224 xfs_lock_inumorder(int lock_mode
, int subclass
)
2226 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
2227 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_IOLOCK_SHIFT
;
2228 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
))
2229 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_ILOCK_SHIFT
;
2235 * The following routine will lock n inodes in exclusive mode.
2236 * We assume the caller calls us with the inodes in i_ino order.
2238 * We need to detect deadlock where an inode that we lock
2239 * is in the AIL and we start waiting for another inode that is locked
2240 * by a thread in a long running transaction (such as truncate). This can
2241 * result in deadlock since the long running trans might need to wait
2242 * for the inode we just locked in order to push the tail and free space
2252 int attempts
= 0, i
, j
, try_lock
;
2255 ASSERT(ips
&& (inodes
>= 2)); /* we need at least two */
2266 for (; i
< inodes
; i
++) {
2269 if (i
&& (ips
[i
] == ips
[i
-1])) /* Already locked */
2273 * If try_lock is not set yet, make sure all locked inodes
2274 * are not in the AIL.
2275 * If any are, set try_lock to be used later.
2279 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
2280 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
2281 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
2288 * If any of the previous locks we have locked is in the AIL,
2289 * we must TRY to get the second and subsequent locks. If
2290 * we can't get any, we must release all we have
2295 /* try_lock must be 0 if i is 0. */
2297 * try_lock means we have an inode locked
2298 * that is in the AIL.
2301 if (!xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
))) {
2305 * Unlock all previous guys and try again.
2306 * xfs_iunlock will try to push the tail
2307 * if the inode is in the AIL.
2310 for(j
= i
- 1; j
>= 0; j
--) {
2313 * Check to see if we've already
2314 * unlocked this one.
2315 * Not the first one going back,
2316 * and the inode ptr is the same.
2318 if ((j
!= (i
- 1)) && ips
[j
] ==
2322 xfs_iunlock(ips
[j
], lock_mode
);
2325 if ((attempts
% 5) == 0) {
2326 delay(1); /* Don't just spin the CPU */
2336 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
2342 if (attempts
< 5) xfs_small_retries
++;
2343 else if (attempts
< 100) xfs_middle_retries
++;
2344 else xfs_lots_retries
++;
2352 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2353 int remove_which_error_return
= 0;
2355 #define REMOVE_DEBUG_TRACE(x)
2356 #endif /* ! DEBUG */
2365 bhv_desc_t
*dir_bdp
,
2366 bhv_vname_t
*dentry
,
2369 bhv_vnode_t
*dir_vp
;
2370 char *name
= VNAME(dentry
);
2371 xfs_inode_t
*dp
, *ip
;
2372 xfs_trans_t
*tp
= NULL
;
2375 xfs_bmap_free_t free_list
;
2376 xfs_fsblock_t first_block
;
2384 dir_vp
= BHV_TO_VNODE(dir_bdp
);
2385 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2387 dp
= XFS_BHVTOI(dir_bdp
);
2390 if (XFS_FORCED_SHUTDOWN(mp
))
2391 return XFS_ERROR(EIO
);
2393 namelen
= VNAMELEN(dentry
);
2395 if (!xfs_get_dir_entry(dentry
, &ip
)) {
2396 dm_di_mode
= ip
->i_d
.di_mode
;
2400 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_REMOVE
)) {
2401 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_REMOVE
, dir_vp
,
2402 DM_RIGHT_NULL
, NULL
, DM_RIGHT_NULL
,
2403 name
, NULL
, dm_di_mode
, 0, 0);
2408 /* From this point on, return through std_return */
2412 * We need to get a reference to ip before we get our log
2413 * reservation. The reason for this is that we cannot call
2414 * xfs_iget for an inode for which we do not have a reference
2415 * once we've acquired a log reservation. This is because the
2416 * inode we are trying to get might be in xfs_inactive going
2417 * for a log reservation. Since we'll have to wait for the
2418 * inactive code to complete before returning from xfs_iget,
2419 * we need to make sure that we don't have log space reserved
2420 * when we call xfs_iget. Instead we get an unlocked reference
2421 * to the inode before getting our log reservation.
2423 error
= xfs_get_dir_entry(dentry
, &ip
);
2425 REMOVE_DEBUG_TRACE(__LINE__
);
2429 dm_di_mode
= ip
->i_d
.di_mode
;
2431 vn_trace_entry(XFS_ITOV(ip
), __FUNCTION__
, (inst_t
*)__return_address
);
2435 error
= XFS_QM_DQATTACH(mp
, dp
, 0);
2436 if (!error
&& dp
!= ip
)
2437 error
= XFS_QM_DQATTACH(mp
, ip
, 0);
2439 REMOVE_DEBUG_TRACE(__LINE__
);
2444 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
2445 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2447 * We try to get the real space reservation first,
2448 * allowing for directory btree deletion(s) implying
2449 * possible bmap insert(s). If we can't get the space
2450 * reservation then we use 0 instead, and avoid the bmap
2451 * btree insert(s) in the directory code by, if the bmap
2452 * insert tries to happen, instead trimming the LAST
2453 * block from the directory.
2455 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2456 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
2457 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
2458 if (error
== ENOSPC
) {
2460 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
2461 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
2464 ASSERT(error
!= ENOSPC
);
2465 REMOVE_DEBUG_TRACE(__LINE__
);
2466 xfs_trans_cancel(tp
, 0);
2471 error
= xfs_lock_dir_and_entry(dp
, ip
);
2473 REMOVE_DEBUG_TRACE(__LINE__
);
2474 xfs_trans_cancel(tp
, cancel_flags
);
2480 * At this point, we've gotten both the directory and the entry
2483 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2486 * Increment vnode ref count only in this case since
2487 * there's an extra vnode reference in the case where
2491 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2495 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2497 XFS_BMAP_INIT(&free_list
, &first_block
);
2498 error
= xfs_dir_removename(tp
, dp
, name
, namelen
, ip
->i_ino
,
2499 &first_block
, &free_list
, 0);
2501 ASSERT(error
!= ENOENT
);
2502 REMOVE_DEBUG_TRACE(__LINE__
);
2505 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2508 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2510 error
= xfs_droplink(tp
, ip
);
2512 REMOVE_DEBUG_TRACE(__LINE__
);
2516 /* Determine if this is the last link while
2517 * we are in the transaction.
2519 link_zero
= (ip
)->i_d
.di_nlink
==0;
2522 * Take an extra ref on the inode so that it doesn't
2523 * go to xfs_inactive() from within the commit.
2528 * If this is a synchronous mount, make sure that the
2529 * remove transaction goes to disk before returning to
2532 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2533 xfs_trans_set_sync(tp
);
2536 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2538 REMOVE_DEBUG_TRACE(__LINE__
);
2542 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2549 * Before we drop our extra reference to the inode, purge it
2550 * from the refcache if it is there. By waiting until afterwards
2551 * to do the IRELE, we ensure that we won't go inactive in the
2552 * xfs_refcache_purge_ip routine (although that would be OK).
2554 xfs_refcache_purge_ip(ip
);
2557 * If we are using filestreams, kill the stream association.
2558 * If the file is still open it may get a new one but that
2559 * will get killed on last close in xfs_close() so we don't
2560 * have to worry about that.
2562 if (link_zero
&& xfs_inode_is_filestream(ip
))
2563 xfs_filestream_deassociate(ip
);
2565 vn_trace_exit(XFS_ITOV(ip
), __FUNCTION__
, (inst_t
*)__return_address
);
2568 * Let interposed file systems know about removed links.
2570 bhv_vop_link_removed(XFS_ITOV(ip
), dir_vp
, link_zero
);
2574 /* Fall through to std_return with error = 0 */
2576 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
,
2577 DM_EVENT_POSTREMOVE
)) {
2578 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTREMOVE
,
2579 dir_vp
, DM_RIGHT_NULL
,
2580 NULL
, DM_RIGHT_NULL
,
2581 name
, NULL
, dm_di_mode
, error
, 0);
2586 xfs_bmap_cancel(&free_list
);
2587 cancel_flags
|= XFS_TRANS_ABORT
;
2588 xfs_trans_cancel(tp
, cancel_flags
);
2593 * In this case make sure to not release the inode until after
2594 * the current transaction is aborted. Releasing it beforehand
2595 * can cause us to go to xfs_inactive and start a recursive
2596 * transaction which can easily deadlock with the current one.
2598 xfs_bmap_cancel(&free_list
);
2599 cancel_flags
|= XFS_TRANS_ABORT
;
2600 xfs_trans_cancel(tp
, cancel_flags
);
2603 * Before we drop our extra reference to the inode, purge it
2604 * from the refcache if it is there. By waiting until afterwards
2605 * to do the IRELE, we ensure that we won't go inactive in the
2606 * xfs_refcache_purge_ip routine (although that would be OK).
2608 xfs_refcache_purge_ip(ip
);
2622 bhv_desc_t
*target_dir_bdp
,
2623 bhv_vnode_t
*src_vp
,
2624 bhv_vname_t
*dentry
,
2627 xfs_inode_t
*tdp
, *sip
;
2630 xfs_inode_t
*ips
[2];
2632 xfs_bmap_free_t free_list
;
2633 xfs_fsblock_t first_block
;
2636 bhv_vnode_t
*target_dir_vp
;
2638 char *target_name
= VNAME(dentry
);
2641 target_dir_vp
= BHV_TO_VNODE(target_dir_bdp
);
2642 vn_trace_entry(target_dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2643 vn_trace_entry(src_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2645 target_namelen
= VNAMELEN(dentry
);
2646 ASSERT(!VN_ISDIR(src_vp
));
2648 sip
= xfs_vtoi(src_vp
);
2649 tdp
= XFS_BHVTOI(target_dir_bdp
);
2651 if (XFS_FORCED_SHUTDOWN(mp
))
2652 return XFS_ERROR(EIO
);
2654 if (DM_EVENT_ENABLED(src_vp
->v_vfsp
, tdp
, DM_EVENT_LINK
)) {
2655 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_LINK
,
2656 target_dir_vp
, DM_RIGHT_NULL
,
2657 src_vp
, DM_RIGHT_NULL
,
2658 target_name
, NULL
, 0, 0, 0);
2663 /* Return through std_return after this point. */
2665 error
= XFS_QM_DQATTACH(mp
, sip
, 0);
2666 if (!error
&& sip
!= tdp
)
2667 error
= XFS_QM_DQATTACH(mp
, tdp
, 0);
2671 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
2672 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2673 resblks
= XFS_LINK_SPACE_RES(mp
, target_namelen
);
2674 error
= xfs_trans_reserve(tp
, resblks
, XFS_LINK_LOG_RES(mp
), 0,
2675 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2676 if (error
== ENOSPC
) {
2678 error
= xfs_trans_reserve(tp
, 0, XFS_LINK_LOG_RES(mp
), 0,
2679 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2686 if (sip
->i_ino
< tdp
->i_ino
) {
2694 xfs_lock_inodes(ips
, 2, 0, XFS_ILOCK_EXCL
);
2697 * Increment vnode ref counts since xfs_trans_commit &
2698 * xfs_trans_cancel will both unlock the inodes and
2699 * decrement the associated ref counts.
2702 VN_HOLD(target_dir_vp
);
2703 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
2704 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
2707 * If the source has too many links, we can't make any more to it.
2709 if (sip
->i_d
.di_nlink
>= XFS_MAXLINK
) {
2710 error
= XFS_ERROR(EMLINK
);
2715 * If we are using project inheritance, we only allow hard link
2716 * creation in our tree when the project IDs are the same; else
2717 * the tree quota mechanism could be circumvented.
2719 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2720 (tdp
->i_d
.di_projid
!= sip
->i_d
.di_projid
))) {
2721 error
= XFS_ERROR(EXDEV
);
2726 (error
= xfs_dir_canenter(tp
, tdp
, target_name
, target_namelen
)))
2729 XFS_BMAP_INIT(&free_list
, &first_block
);
2731 error
= xfs_dir_createname(tp
, tdp
, target_name
, target_namelen
,
2732 sip
->i_ino
, &first_block
, &free_list
,
2736 xfs_ichgtime(tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2738 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
2740 error
= xfs_bumplink(tp
, sip
);
2745 * If this is a synchronous mount, make sure that the
2746 * link transaction goes to disk before returning to
2749 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2750 xfs_trans_set_sync(tp
);
2753 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
2755 xfs_bmap_cancel(&free_list
);
2759 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2763 /* Fall through to std_return with error = 0. */
2765 if (DM_EVENT_ENABLED(src_vp
->v_vfsp
, sip
,
2766 DM_EVENT_POSTLINK
)) {
2767 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTLINK
,
2768 target_dir_vp
, DM_RIGHT_NULL
,
2769 src_vp
, DM_RIGHT_NULL
,
2770 target_name
, NULL
, 0, error
, 0);
2775 cancel_flags
|= XFS_TRANS_ABORT
;
2779 xfs_trans_cancel(tp
, cancel_flags
);
2790 bhv_desc_t
*dir_bdp
,
2791 bhv_vname_t
*dentry
,
2796 char *dir_name
= VNAME(dentry
);
2798 xfs_inode_t
*cdp
; /* inode of created dir */
2799 bhv_vnode_t
*cvp
; /* vnode of created dir */
2805 xfs_bmap_free_t free_list
;
2806 xfs_fsblock_t first_block
;
2807 bhv_vnode_t
*dir_vp
;
2808 boolean_t dp_joined_to_trans
;
2809 boolean_t created
= B_FALSE
;
2810 int dm_event_sent
= 0;
2812 struct xfs_dquot
*udqp
, *gdqp
;
2817 dir_vp
= BHV_TO_VNODE(dir_bdp
);
2818 dp
= XFS_BHVTOI(dir_bdp
);
2821 if (XFS_FORCED_SHUTDOWN(mp
))
2822 return XFS_ERROR(EIO
);
2824 dir_namelen
= VNAMELEN(dentry
);
2827 dp_joined_to_trans
= B_FALSE
;
2828 dm_di_mode
= vap
->va_mode
;
2830 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_CREATE
)) {
2831 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_CREATE
,
2832 dir_vp
, DM_RIGHT_NULL
, NULL
,
2833 DM_RIGHT_NULL
, dir_name
, NULL
,
2840 /* Return through std_return after this point. */
2842 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2846 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
2847 prid
= dp
->i_d
.di_projid
;
2848 else if (vap
->va_mask
& XFS_AT_PROJID
)
2849 prid
= (xfs_prid_t
)vap
->va_projid
;
2851 prid
= (xfs_prid_t
)dfltprid
;
2854 * Make sure that we have allocated dquot(s) on disk.
2856 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
2857 current_fsuid(credp
), current_fsgid(credp
), prid
,
2858 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
2862 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
2863 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2864 resblks
= XFS_MKDIR_SPACE_RES(mp
, dir_namelen
);
2865 error
= xfs_trans_reserve(tp
, resblks
, XFS_MKDIR_LOG_RES(mp
), 0,
2866 XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
);
2867 if (error
== ENOSPC
) {
2869 error
= xfs_trans_reserve(tp
, 0, XFS_MKDIR_LOG_RES(mp
), 0,
2870 XFS_TRANS_PERM_LOG_RES
,
2871 XFS_MKDIR_LOG_COUNT
);
2879 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
2882 * Check for directory link count overflow.
2884 if (dp
->i_d
.di_nlink
>= XFS_MAXLINK
) {
2885 error
= XFS_ERROR(EMLINK
);
2890 * Reserve disk quota and the inode.
2892 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
2897 (error
= xfs_dir_canenter(tp
, dp
, dir_name
, dir_namelen
)))
2900 * create the directory inode.
2902 error
= xfs_dir_ialloc(&tp
, dp
, vap
->va_mode
, 2,
2903 0, credp
, prid
, resblks
> 0,
2906 if (error
== ENOSPC
)
2913 * Now we add the directory inode to the transaction.
2914 * We waited until now since xfs_dir_ialloc might start
2915 * a new transaction. Had we joined the transaction
2916 * earlier, the locks might have gotten released.
2919 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2920 dp_joined_to_trans
= B_TRUE
;
2922 XFS_BMAP_INIT(&free_list
, &first_block
);
2924 error
= xfs_dir_createname(tp
, dp
, dir_name
, dir_namelen
, cdp
->i_ino
,
2925 &first_block
, &free_list
, resblks
?
2926 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
2928 ASSERT(error
!= ENOSPC
);
2931 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2934 * Bump the in memory version number of the parent directory
2935 * so that other processes accessing it will recognize that
2936 * the directory has changed.
2940 error
= xfs_dir_init(tp
, cdp
, dp
);
2945 error
= xfs_bumplink(tp
, dp
);
2949 cvp
= XFS_ITOV(cdp
);
2957 * Attach the dquots to the new inode and modify the icount incore.
2959 XFS_QM_DQVOPCREATE(mp
, tp
, cdp
, udqp
, gdqp
);
2962 * If this is a synchronous mount, make sure that the
2963 * mkdir transaction goes to disk before returning to
2966 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2967 xfs_trans_set_sync(tp
);
2970 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2976 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2977 XFS_QM_DQRELE(mp
, udqp
);
2978 XFS_QM_DQRELE(mp
, gdqp
);
2983 /* Fall through to std_return with error = 0 or errno from
2984 * xfs_trans_commit. */
2987 if ( (created
|| (error
!= 0 && dm_event_sent
!= 0)) &&
2988 DM_EVENT_ENABLED(dir_vp
->v_vfsp
, XFS_BHVTOI(dir_bdp
),
2989 DM_EVENT_POSTCREATE
)) {
2990 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTCREATE
,
2991 dir_vp
, DM_RIGHT_NULL
,
2992 created
? XFS_ITOV(cdp
):NULL
,
2995 dm_di_mode
, error
, 0);
3001 xfs_bmap_cancel(&free_list
);
3003 cancel_flags
|= XFS_TRANS_ABORT
;
3005 xfs_trans_cancel(tp
, cancel_flags
);
3006 XFS_QM_DQRELE(mp
, udqp
);
3007 XFS_QM_DQRELE(mp
, gdqp
);
3009 if (!dp_joined_to_trans
&& (dp
!= NULL
)) {
3010 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
3023 bhv_desc_t
*dir_bdp
,
3024 bhv_vname_t
*dentry
,
3027 char *name
= VNAME(dentry
);
3029 xfs_inode_t
*cdp
; /* child directory */
3033 xfs_bmap_free_t free_list
;
3034 xfs_fsblock_t first_block
;
3037 bhv_vnode_t
*dir_vp
;
3038 int dm_di_mode
= S_IFDIR
;
3043 dir_vp
= BHV_TO_VNODE(dir_bdp
);
3044 dp
= XFS_BHVTOI(dir_bdp
);
3047 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
3049 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp
)->i_mount
))
3050 return XFS_ERROR(EIO
);
3051 namelen
= VNAMELEN(dentry
);
3053 if (!xfs_get_dir_entry(dentry
, &cdp
)) {
3054 dm_di_mode
= cdp
->i_d
.di_mode
;
3058 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_REMOVE
)) {
3059 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_REMOVE
,
3060 dir_vp
, DM_RIGHT_NULL
,
3061 NULL
, DM_RIGHT_NULL
,
3062 name
, NULL
, dm_di_mode
, 0, 0);
3064 return XFS_ERROR(error
);
3067 /* Return through std_return after this point. */
3072 * We need to get a reference to cdp before we get our log
3073 * reservation. The reason for this is that we cannot call
3074 * xfs_iget for an inode for which we do not have a reference
3075 * once we've acquired a log reservation. This is because the
3076 * inode we are trying to get might be in xfs_inactive going
3077 * for a log reservation. Since we'll have to wait for the
3078 * inactive code to complete before returning from xfs_iget,
3079 * we need to make sure that we don't have log space reserved
3080 * when we call xfs_iget. Instead we get an unlocked reference
3081 * to the inode before getting our log reservation.
3083 error
= xfs_get_dir_entry(dentry
, &cdp
);
3085 REMOVE_DEBUG_TRACE(__LINE__
);
3089 dm_di_mode
= cdp
->i_d
.di_mode
;
3092 * Get the dquots for the inodes.
3094 error
= XFS_QM_DQATTACH(mp
, dp
, 0);
3095 if (!error
&& dp
!= cdp
)
3096 error
= XFS_QM_DQATTACH(mp
, cdp
, 0);
3099 REMOVE_DEBUG_TRACE(__LINE__
);
3103 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
3104 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
3106 * We try to get the real space reservation first,
3107 * allowing for directory btree deletion(s) implying
3108 * possible bmap insert(s). If we can't get the space
3109 * reservation then we use 0 instead, and avoid the bmap
3110 * btree insert(s) in the directory code by, if the bmap
3111 * insert tries to happen, instead trimming the LAST
3112 * block from the directory.
3114 resblks
= XFS_REMOVE_SPACE_RES(mp
);
3115 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
3116 XFS_TRANS_PERM_LOG_RES
, XFS_DEFAULT_LOG_COUNT
);
3117 if (error
== ENOSPC
) {
3119 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
3120 XFS_TRANS_PERM_LOG_RES
, XFS_DEFAULT_LOG_COUNT
);
3123 ASSERT(error
!= ENOSPC
);
3128 XFS_BMAP_INIT(&free_list
, &first_block
);
3131 * Now lock the child directory inode and the parent directory
3132 * inode in the proper order. This will take care of validating
3133 * that the directory entry for the child directory inode has
3134 * not changed while we were obtaining a log reservation.
3136 error
= xfs_lock_dir_and_entry(dp
, cdp
);
3138 xfs_trans_cancel(tp
, cancel_flags
);
3143 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
3146 * Only increment the parent directory vnode count if
3147 * we didn't bump it in looking up cdp. The only time
3148 * we don't bump it is when we're looking up ".".
3154 xfs_trans_ijoin(tp
, cdp
, XFS_ILOCK_EXCL
);
3156 ASSERT(cdp
->i_d
.di_nlink
>= 2);
3157 if (cdp
->i_d
.di_nlink
!= 2) {
3158 error
= XFS_ERROR(ENOTEMPTY
);
3161 if (!xfs_dir_isempty(cdp
)) {
3162 error
= XFS_ERROR(ENOTEMPTY
);
3166 error
= xfs_dir_removename(tp
, dp
, name
, namelen
, cdp
->i_ino
,
3167 &first_block
, &free_list
, resblks
);
3171 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3174 * Bump the in memory generation count on the parent
3175 * directory so that other can know that it has changed.
3180 * Drop the link from cdp's "..".
3182 error
= xfs_droplink(tp
, dp
);
3188 * Drop the link from dp to cdp.
3190 error
= xfs_droplink(tp
, cdp
);
3196 * Drop the "." link from cdp to self.
3198 error
= xfs_droplink(tp
, cdp
);
3203 /* Determine these before committing transaction */
3204 last_cdp_link
= (cdp
)->i_d
.di_nlink
==0;
3207 * Take an extra ref on the child vnode so that it
3208 * does not go to xfs_inactive() from within the commit.
3213 * If this is a synchronous mount, make sure that the
3214 * rmdir transaction goes to disk before returning to
3217 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
3218 xfs_trans_set_sync(tp
);
3221 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
3223 xfs_bmap_cancel(&free_list
);
3224 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
|
3230 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
3238 * Let interposed file systems know about removed links.
3240 bhv_vop_link_removed(XFS_ITOV(cdp
), dir_vp
, last_cdp_link
);
3244 /* Fall through to std_return with error = 0 or the errno
3245 * from xfs_trans_commit. */
3247 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_POSTREMOVE
)) {
3248 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTREMOVE
,
3249 dir_vp
, DM_RIGHT_NULL
,
3250 NULL
, DM_RIGHT_NULL
,
3251 name
, NULL
, dm_di_mode
,
3257 xfs_bmap_cancel(&free_list
);
3258 cancel_flags
|= XFS_TRANS_ABORT
;
3262 xfs_trans_cancel(tp
, cancel_flags
);
3268 * Read dp's entries starting at uiop->uio_offset and translate them into
3269 * bufsize bytes worth of struct dirents starting at bufbase.
3273 bhv_desc_t
*dir_bdp
,
3279 xfs_trans_t
*tp
= NULL
;
3283 vn_trace_entry(BHV_TO_VNODE(dir_bdp
), __FUNCTION__
,
3284 (inst_t
*)__return_address
);
3285 dp
= XFS_BHVTOI(dir_bdp
);
3287 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
3288 return XFS_ERROR(EIO
);
3290 lock_mode
= xfs_ilock_map_shared(dp
);
3291 error
= xfs_dir_getdents(tp
, dp
, uiop
, eofp
);
3292 xfs_iunlock_map_shared(dp
, lock_mode
);
3299 bhv_desc_t
*dir_bdp
,
3300 bhv_vname_t
*dentry
,
3312 xfs_bmap_free_t free_list
;
3313 xfs_fsblock_t first_block
;
3314 boolean_t dp_joined_to_trans
;
3315 bhv_vnode_t
*dir_vp
;
3318 xfs_fileoff_t first_fsb
;
3319 xfs_filblks_t fs_blocks
;
3321 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
3328 struct xfs_dquot
*udqp
, *gdqp
;
3330 char *link_name
= VNAME(dentry
);
3334 dir_vp
= BHV_TO_VNODE(dir_bdp
);
3335 dp
= XFS_BHVTOI(dir_bdp
);
3336 dp_joined_to_trans
= B_FALSE
;
3341 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
3345 if (XFS_FORCED_SHUTDOWN(mp
))
3346 return XFS_ERROR(EIO
);
3348 link_namelen
= VNAMELEN(dentry
);
3351 * Check component lengths of the target path name.
3353 pathlen
= strlen(target_path
);
3354 if (pathlen
>= MAXPATHLEN
) /* total string too long */
3355 return XFS_ERROR(ENAMETOOLONG
);
3356 if (pathlen
>= MAXNAMELEN
) { /* is any component too long? */
3360 for (total
= 0, path
= target_path
; total
< pathlen
;) {
3364 while(*path
== '/') {
3370 * Count up to the next slash or end of path.
3371 * Error out if the component is bigger than MAXNAMELEN.
3373 for(len
= 0; *path
!= '/' && total
< pathlen
;total
++, path
++) {
3374 if (++len
>= MAXNAMELEN
) {
3375 error
= ENAMETOOLONG
;
3382 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_SYMLINK
)) {
3383 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_SYMLINK
, dir_vp
,
3384 DM_RIGHT_NULL
, NULL
, DM_RIGHT_NULL
,
3385 link_name
, target_path
, 0, 0, 0);
3390 /* Return through std_return after this point. */
3393 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
3394 prid
= dp
->i_d
.di_projid
;
3395 else if (vap
->va_mask
& XFS_AT_PROJID
)
3396 prid
= (xfs_prid_t
)vap
->va_projid
;
3398 prid
= (xfs_prid_t
)dfltprid
;
3401 * Make sure that we have allocated dquot(s) on disk.
3403 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
3404 current_fsuid(credp
), current_fsgid(credp
), prid
,
3405 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
3409 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SYMLINK
);
3410 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
3412 * The symlink will fit into the inode data fork?
3413 * There can't be any attributes so we get the whole variable part.
3415 if (pathlen
<= XFS_LITINO(mp
))
3418 fs_blocks
= XFS_B_TO_FSB(mp
, pathlen
);
3419 resblks
= XFS_SYMLINK_SPACE_RES(mp
, link_namelen
, fs_blocks
);
3420 error
= xfs_trans_reserve(tp
, resblks
, XFS_SYMLINK_LOG_RES(mp
), 0,
3421 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
3422 if (error
== ENOSPC
&& fs_blocks
== 0) {
3424 error
= xfs_trans_reserve(tp
, 0, XFS_SYMLINK_LOG_RES(mp
), 0,
3425 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
3433 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
3436 * Check whether the directory allows new symlinks or not.
3438 if (dp
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) {
3439 error
= XFS_ERROR(EPERM
);
3444 * Reserve disk quota : blocks and inode.
3446 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
3451 * Check for ability to enter directory entry, if no space reserved.
3454 (error
= xfs_dir_canenter(tp
, dp
, link_name
, link_namelen
)))
3457 * Initialize the bmap freelist prior to calling either
3458 * bmapi or the directory create code.
3460 XFS_BMAP_INIT(&free_list
, &first_block
);
3463 * Allocate an inode for the symlink.
3465 error
= xfs_dir_ialloc(&tp
, dp
, S_IFLNK
| (vap
->va_mode
&~S_IFMT
),
3466 1, 0, credp
, prid
, resblks
> 0, &ip
, NULL
);
3468 if (error
== ENOSPC
)
3475 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
3476 dp_joined_to_trans
= B_TRUE
;
3479 * Also attach the dquot(s) to it, if applicable.
3481 XFS_QM_DQVOPCREATE(mp
, tp
, ip
, udqp
, gdqp
);
3484 resblks
-= XFS_IALLOC_SPACE_RES(mp
);
3486 * If the symlink will fit into the inode, write it inline.
3488 if (pathlen
<= XFS_IFORK_DSIZE(ip
)) {
3489 xfs_idata_realloc(ip
, pathlen
, XFS_DATA_FORK
);
3490 memcpy(ip
->i_df
.if_u1
.if_data
, target_path
, pathlen
);
3491 ip
->i_d
.di_size
= pathlen
;
3494 * The inode was initially created in extent format.
3496 ip
->i_df
.if_flags
&= ~(XFS_IFEXTENTS
| XFS_IFBROOT
);
3497 ip
->i_df
.if_flags
|= XFS_IFINLINE
;
3499 ip
->i_d
.di_format
= XFS_DINODE_FMT_LOCAL
;
3500 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_DDATA
| XFS_ILOG_CORE
);
3504 nmaps
= SYMLINK_MAPS
;
3506 error
= xfs_bmapi(tp
, ip
, first_fsb
, fs_blocks
,
3507 XFS_BMAPI_WRITE
| XFS_BMAPI_METADATA
,
3508 &first_block
, resblks
, mval
, &nmaps
,
3515 resblks
-= fs_blocks
;
3516 ip
->i_d
.di_size
= pathlen
;
3517 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3519 cur_chunk
= target_path
;
3520 for (n
= 0; n
< nmaps
; n
++) {
3521 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
3522 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
3523 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
3524 BTOBB(byte_cnt
), 0);
3525 ASSERT(bp
&& !XFS_BUF_GETERROR(bp
));
3526 if (pathlen
< byte_cnt
) {
3529 pathlen
-= byte_cnt
;
3531 memcpy(XFS_BUF_PTR(bp
), cur_chunk
, byte_cnt
);
3532 cur_chunk
+= byte_cnt
;
3534 xfs_trans_log_buf(tp
, bp
, 0, byte_cnt
- 1);
3539 * Create the directory entry for the symlink.
3541 error
= xfs_dir_createname(tp
, dp
, link_name
, link_namelen
, ip
->i_ino
,
3542 &first_block
, &free_list
, resblks
);
3545 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3546 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
3549 * Bump the in memory version number of the parent directory
3550 * so that other processes accessing it will recognize that
3551 * the directory has changed.
3556 * If this is a synchronous mount, make sure that the
3557 * symlink transaction goes to disk before returning to
3560 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
3561 xfs_trans_set_sync(tp
);
3565 * xfs_trans_commit normally decrements the vnode ref count
3566 * when it unlocks the inode. Since we want to return the
3567 * vnode to the caller, we bump the vnode ref count now.
3571 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
3575 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
3576 XFS_QM_DQRELE(mp
, udqp
);
3577 XFS_QM_DQRELE(mp
, gdqp
);
3579 /* Fall through to std_return with error = 0 or errno from
3580 * xfs_trans_commit */
3582 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, XFS_BHVTOI(dir_bdp
),
3583 DM_EVENT_POSTSYMLINK
)) {
3584 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTSYMLINK
,
3585 dir_vp
, DM_RIGHT_NULL
,
3586 error
? NULL
: XFS_ITOV(ip
),
3587 DM_RIGHT_NULL
, link_name
, target_path
,
3603 xfs_bmap_cancel(&free_list
);
3604 cancel_flags
|= XFS_TRANS_ABORT
;
3606 xfs_trans_cancel(tp
, cancel_flags
);
3607 XFS_QM_DQRELE(mp
, udqp
);
3608 XFS_QM_DQRELE(mp
, gdqp
);
3610 if (!dp_joined_to_trans
&& (dp
!= NULL
)) {
3611 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
3621 * A fid routine that takes a pointer to a previously allocated
3622 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3632 vn_trace_entry(BHV_TO_VNODE(bdp
), __FUNCTION__
,
3633 (inst_t
*)__return_address
);
3634 ASSERT(sizeof(fid_t
) >= sizeof(xfs_fid2_t
));
3636 xfid
= (xfs_fid2_t
*)fidp
;
3637 ip
= XFS_BHVTOI(bdp
);
3638 xfid
->fid_len
= sizeof(xfs_fid2_t
) - sizeof(xfid
->fid_len
);
3641 * use memcpy because the inode is a long long and there's no
3642 * assurance that xfid->fid_ino is properly aligned.
3644 memcpy(&xfid
->fid_ino
, &ip
->i_ino
, sizeof(xfid
->fid_ino
));
3645 xfid
->fid_gen
= ip
->i_d
.di_gen
;
3657 bhv_vrwlock_t locktype
)
3662 vp
= BHV_TO_VNODE(bdp
);
3665 ip
= XFS_BHVTOI(bdp
);
3666 if (locktype
== VRWLOCK_WRITE
) {
3667 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
3668 } else if (locktype
== VRWLOCK_TRY_READ
) {
3669 return xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
);
3670 } else if (locktype
== VRWLOCK_TRY_WRITE
) {
3671 return xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
);
3673 ASSERT((locktype
== VRWLOCK_READ
) ||
3674 (locktype
== VRWLOCK_WRITE_DIRECT
));
3675 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
3688 bhv_vrwlock_t locktype
)
3693 vp
= BHV_TO_VNODE(bdp
);
3696 ip
= XFS_BHVTOI(bdp
);
3697 if (locktype
== VRWLOCK_WRITE
) {
3699 * In the write case, we may have added a new entry to
3700 * the reference cache. This might store a pointer to
3701 * an inode to be released in this inode. If it is there,
3702 * clear the pointer and release the inode after unlocking
3705 xfs_refcache_iunlock(ip
, XFS_IOLOCK_EXCL
);
3707 ASSERT((locktype
== VRWLOCK_READ
) ||
3708 (locktype
== VRWLOCK_WRITE_DIRECT
));
3709 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
3721 xfs_inode_log_item_t
*iip
;
3724 ip
= XFS_BHVTOI(bdp
);
3728 if (XFS_FORCED_SHUTDOWN(mp
))
3729 return XFS_ERROR(EIO
);
3732 * Bypass inodes which have already been cleaned by
3733 * the inode flush clustering code inside xfs_iflush
3735 if ((ip
->i_update_core
== 0) &&
3736 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)))
3739 if (flags
& FLUSH_LOG
) {
3740 if (iip
&& iip
->ili_last_lsn
) {
3741 xlog_t
*log
= mp
->m_log
;
3743 int s
, log_flags
= XFS_LOG_FORCE
;
3745 s
= GRANT_LOCK(log
);
3746 sync_lsn
= log
->l_last_sync_lsn
;
3747 GRANT_UNLOCK(log
, s
);
3749 if ((XFS_LSN_CMP(iip
->ili_last_lsn
, sync_lsn
) <= 0))
3752 if (flags
& FLUSH_SYNC
)
3753 log_flags
|= XFS_LOG_SYNC
;
3754 return xfs_log_force(mp
, iip
->ili_last_lsn
, log_flags
);
3759 * We make this non-blocking if the inode is contended,
3760 * return EAGAIN to indicate to the caller that they
3761 * did not succeed. This prevents the flush path from
3762 * blocking on inodes inside another operation right
3763 * now, they get caught later by xfs_sync.
3765 if (flags
& FLUSH_INODE
) {
3768 if (xfs_ipincount(ip
))
3771 if (flags
& FLUSH_SYNC
) {
3772 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
3774 } else if (xfs_ilock_nowait(ip
, XFS_ILOCK_SHARED
)) {
3775 if (xfs_ipincount(ip
) || !xfs_iflock_nowait(ip
)) {
3776 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3783 if (flags
& FLUSH_SYNC
)
3784 flush_flags
= XFS_IFLUSH_SYNC
;
3786 flush_flags
= XFS_IFLUSH_ASYNC
;
3788 error
= xfs_iflush(ip
, flush_flags
);
3789 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3807 if (!capable(CAP_SYS_ADMIN
))
3808 return XFS_ERROR(EPERM
);
3810 ip
= XFS_BHVTOI(bdp
);
3813 if (XFS_FORCED_SHUTDOWN(mp
))
3814 return XFS_ERROR(EIO
);
3816 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SET_DMATTRS
);
3817 error
= xfs_trans_reserve(tp
, 0, XFS_ICHANGE_LOG_RES (mp
), 0, 0, 0);
3819 xfs_trans_cancel(tp
, 0);
3822 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3823 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
3825 ip
->i_iocore
.io_dmevmask
= ip
->i_d
.di_dmevmask
= evmask
;
3826 ip
->i_iocore
.io_dmstate
= ip
->i_d
.di_dmstate
= state
;
3828 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3830 error
= xfs_trans_commit(tp
, 0);
3842 vp
= BHV_TO_VNODE(bdp
);
3843 ip
= XFS_BHVTOI(bdp
);
3845 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
3847 ASSERT(!VN_MAPPED(vp
));
3849 /* bad inode, get out here ASAP */
3857 ASSERT(XFS_FORCED_SHUTDOWN(ip
->i_mount
) || ip
->i_delayed_blks
== 0);
3860 * Make sure the atime in the XFS inode is correct before freeing the
3863 xfs_synchronize_atime(ip
);
3866 * If we have nothing to flush with this inode then complete the
3867 * teardown now, otherwise break the link between the xfs inode and the
3868 * linux inode and clean up the xfs inode later. This avoids flushing
3869 * the inode to disk during the delete operation itself.
3871 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3872 * first to ensure that xfs_iunpin() will never see an xfs inode
3873 * that has a linux inode being reclaimed. Synchronisation is provided
3874 * by the i_flags_lock.
3876 if (!ip
->i_update_core
&& (ip
->i_itemp
== NULL
)) {
3877 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3879 return xfs_finish_reclaim(ip
, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC
);
3881 xfs_mount_t
*mp
= ip
->i_mount
;
3883 /* Protect sync and unpin from us */
3884 XFS_MOUNT_ILOCK(mp
);
3885 spin_lock(&ip
->i_flags_lock
);
3886 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
3887 vn_bhv_remove(VN_BHV_HEAD(vp
), XFS_ITOBHV(ip
));
3888 spin_unlock(&ip
->i_flags_lock
);
3889 list_add_tail(&ip
->i_reclaim
, &mp
->m_del_inodes
);
3890 XFS_MOUNT_IUNLOCK(mp
);
3901 xfs_ihash_t
*ih
= ip
->i_hash
;
3902 bhv_vnode_t
*vp
= XFS_ITOV_NULL(ip
);
3905 if (vp
&& VN_BAD(vp
))
3908 /* The hash lock here protects a thread in xfs_iget_core from
3909 * racing with us on linking the inode back with a vnode.
3910 * Once we have the XFS_IRECLAIM flag set it will not touch
3913 write_lock(&ih
->ih_lock
);
3914 spin_lock(&ip
->i_flags_lock
);
3915 if (__xfs_iflags_test(ip
, XFS_IRECLAIM
) ||
3916 (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) && vp
== NULL
)) {
3917 spin_unlock(&ip
->i_flags_lock
);
3918 write_unlock(&ih
->ih_lock
);
3921 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3925 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
3926 spin_unlock(&ip
->i_flags_lock
);
3927 write_unlock(&ih
->ih_lock
);
3930 * If the inode is still dirty, then flush it out. If the inode
3931 * is not in the AIL, then it will be OK to flush it delwri as
3932 * long as xfs_iflush() does not keep any references to the inode.
3933 * We leave that decision up to xfs_iflush() since it has the
3934 * knowledge of whether it's OK to simply do a delwri flush of
3935 * the inode or whether we need to wait until the inode is
3936 * pulled from the AIL.
3937 * We get the flush lock regardless, though, just to make sure
3938 * we don't free it while it is being flushed.
3940 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
3942 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3946 if (ip
->i_update_core
||
3947 ((ip
->i_itemp
!= NULL
) &&
3948 (ip
->i_itemp
->ili_format
.ilf_fields
!= 0))) {
3949 error
= xfs_iflush(ip
, sync_mode
);
3951 * If we hit an error, typically because of filesystem
3952 * shutdown, we don't need to let vn_reclaim to know
3953 * because we're gonna reclaim the inode anyway.
3956 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3959 xfs_iflock(ip
); /* synchronize with xfs_iflush_done */
3962 ASSERT(ip
->i_update_core
== 0);
3963 ASSERT(ip
->i_itemp
== NULL
||
3964 ip
->i_itemp
->ili_format
.ilf_fields
== 0);
3965 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3966 } else if (locked
) {
3968 * We are not interested in doing an iflush if we're
3969 * in the process of shutting down the filesystem forcibly.
3970 * So, just reclaim the inode.
3973 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3982 xfs_finish_reclaim_all(xfs_mount_t
*mp
, int noblock
)
3985 xfs_inode_t
*ip
, *n
;
3990 XFS_MOUNT_ILOCK(mp
);
3991 list_for_each_entry_safe(ip
, n
, &mp
->m_del_inodes
, i_reclaim
) {
3993 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
) == 0)
3995 if (xfs_ipincount(ip
) ||
3996 !xfs_iflock_nowait(ip
)) {
3997 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4001 XFS_MOUNT_IUNLOCK(mp
);
4002 if (xfs_finish_reclaim(ip
, noblock
,
4003 XFS_IFLUSH_DELWRI_ELSE_ASYNC
))
4012 XFS_MOUNT_IUNLOCK(mp
);
4017 * xfs_alloc_file_space()
4018 * This routine allocates disk space for the given file.
4020 * If alloc_type == 0, this request is for an ALLOCSP type
4021 * request which will change the file size. In this case, no
4022 * DMAPI event will be generated by the call. A TRUNCATE event
4023 * will be generated later by xfs_setattr.
4025 * If alloc_type != 0, this request is for a RESVSP type
4026 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4027 * lower block boundary byte address is less than the file's
4036 xfs_alloc_file_space(
4043 xfs_mount_t
*mp
= ip
->i_mount
;
4045 xfs_filblks_t allocated_fsb
;
4046 xfs_filblks_t allocatesize_fsb
;
4047 xfs_extlen_t extsz
, temp
;
4048 xfs_fileoff_t startoffset_fsb
;
4049 xfs_fsblock_t firstfsb
;
4055 xfs_bmbt_irec_t imaps
[1], *imapp
;
4056 xfs_bmap_free_t free_list
;
4057 uint qblocks
, resblks
, resrtextents
;
4061 vn_trace_entry(XFS_ITOV(ip
), __FUNCTION__
, (inst_t
*)__return_address
);
4063 if (XFS_FORCED_SHUTDOWN(mp
))
4064 return XFS_ERROR(EIO
);
4066 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
4070 return XFS_ERROR(EINVAL
);
4072 rt
= XFS_IS_REALTIME_INODE(ip
);
4073 extsz
= xfs_get_extsz_hint(ip
);
4078 bmapi_flag
= XFS_BMAPI_WRITE
| (alloc_type
? XFS_BMAPI_PREALLOC
: 0);
4079 startoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
4080 allocatesize_fsb
= XFS_B_TO_FSB(mp
, count
);
4082 /* Generate a DMAPI event if needed. */
4083 if (alloc_type
!= 0 && offset
< ip
->i_size
&&
4084 (attr_flags
&ATTR_DMI
) == 0 &&
4085 DM_EVENT_ENABLED(XFS_MTOVFS(mp
), ip
, DM_EVENT_WRITE
)) {
4086 xfs_off_t end_dmi_offset
;
4088 end_dmi_offset
= offset
+len
;
4089 if (end_dmi_offset
> ip
->i_size
)
4090 end_dmi_offset
= ip
->i_size
;
4091 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, XFS_ITOV(ip
),
4092 offset
, end_dmi_offset
- offset
,
4099 * Allocate file space until done or until there is an error
4102 while (allocatesize_fsb
&& !error
) {
4106 * Determine space reservations for data/realtime.
4108 if (unlikely(extsz
)) {
4109 s
= startoffset_fsb
;
4112 e
= startoffset_fsb
+ allocatesize_fsb
;
4113 if ((temp
= do_mod(startoffset_fsb
, extsz
)))
4115 if ((temp
= do_mod(e
, extsz
)))
4119 e
= allocatesize_fsb
;
4123 resrtextents
= qblocks
= (uint
)(e
- s
);
4124 resrtextents
/= mp
->m_sb
.sb_rextsize
;
4125 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
4126 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
4129 resblks
= qblocks
= \
4130 XFS_DIOSTRAT_SPACE_RES(mp
, (uint
)(e
- s
));
4131 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
4135 * Allocate and setup the transaction.
4137 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
4138 error
= xfs_trans_reserve(tp
, resblks
,
4139 XFS_WRITE_LOG_RES(mp
), resrtextents
,
4140 XFS_TRANS_PERM_LOG_RES
,
4141 XFS_WRITE_LOG_COUNT
);
4143 * Check for running out of space
4147 * Free the transaction structure.
4149 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
4150 xfs_trans_cancel(tp
, 0);
4153 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4154 error
= XFS_TRANS_RESERVE_QUOTA_NBLKS(mp
, tp
, ip
,
4155 qblocks
, 0, quota_flag
);
4159 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
4160 xfs_trans_ihold(tp
, ip
);
4163 * Issue the xfs_bmapi() call to allocate the blocks
4165 XFS_BMAP_INIT(&free_list
, &firstfsb
);
4166 error
= XFS_BMAPI(mp
, tp
, &ip
->i_iocore
, startoffset_fsb
,
4167 allocatesize_fsb
, bmapi_flag
,
4168 &firstfsb
, 0, imapp
, &nimaps
,
4175 * Complete the transaction
4177 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
4182 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
4183 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4188 allocated_fsb
= imapp
->br_blockcount
;
4191 error
= XFS_ERROR(ENOSPC
);
4195 startoffset_fsb
+= allocated_fsb
;
4196 allocatesize_fsb
-= allocated_fsb
;
4199 if (error
== ENOSPC
&& (attr_flags
&ATTR_DMI
) == 0 &&
4200 DM_EVENT_ENABLED(XFS_MTOVFS(mp
), ip
, DM_EVENT_NOSPACE
)) {
4202 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_NOSPACE
,
4203 XFS_ITOV(ip
), DM_RIGHT_NULL
,
4204 XFS_ITOV(ip
), DM_RIGHT_NULL
,
4205 NULL
, NULL
, 0, 0, 0); /* Delay flag intentionally unused */
4207 goto retry
; /* Maybe DMAPI app. has made space */
4208 /* else fall through with error from XFS_SEND_DATA */
4213 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4214 xfs_bmap_cancel(&free_list
);
4215 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp
, tp
, ip
, qblocks
, 0, quota_flag
);
4217 error1
: /* Just cancel transaction */
4218 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
4219 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4220 goto dmapi_enospc_check
;
4224 * Zero file bytes between startoff and endoff inclusive.
4225 * The iolock is held exclusive and no blocks are buffered.
4228 xfs_zero_remaining_bytes(
4233 xfs_bmbt_irec_t imap
;
4234 xfs_fileoff_t offset_fsb
;
4235 xfs_off_t lastoffset
;
4238 xfs_mount_t
*mp
= ip
->i_mount
;
4242 bp
= xfs_buf_get_noaddr(mp
->m_sb
.sb_blocksize
,
4243 ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
?
4244 mp
->m_rtdev_targp
: mp
->m_ddev_targp
);
4246 for (offset
= startoff
; offset
<= endoff
; offset
= lastoffset
+ 1) {
4247 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
4249 error
= XFS_BMAPI(mp
, NULL
, &ip
->i_iocore
, offset_fsb
, 1, 0,
4250 NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
4251 if (error
|| nimap
< 1)
4253 ASSERT(imap
.br_blockcount
>= 1);
4254 ASSERT(imap
.br_startoff
== offset_fsb
);
4255 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+ 1) - 1;
4256 if (lastoffset
> endoff
)
4257 lastoffset
= endoff
;
4258 if (imap
.br_startblock
== HOLESTARTBLOCK
)
4260 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
4261 if (imap
.br_state
== XFS_EXT_UNWRITTEN
)
4264 XFS_BUF_UNWRITE(bp
);
4266 XFS_BUF_SET_ADDR(bp
, XFS_FSB_TO_DB(ip
, imap
.br_startblock
));
4268 if ((error
= xfs_iowait(bp
))) {
4269 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4270 mp
, bp
, XFS_BUF_ADDR(bp
));
4273 memset(XFS_BUF_PTR(bp
) +
4274 (offset
- XFS_FSB_TO_B(mp
, imap
.br_startoff
)),
4275 0, lastoffset
- offset
+ 1);
4280 if ((error
= xfs_iowait(bp
))) {
4281 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4282 mp
, bp
, XFS_BUF_ADDR(bp
));
4291 * xfs_free_file_space()
4292 * This routine frees disk space for the given file.
4294 * This routine is only called by xfs_change_file_space
4295 * for an UNRESVSP type call.
4303 xfs_free_file_space(
4312 xfs_off_t end_dmi_offset
;
4313 xfs_fileoff_t endoffset_fsb
;
4315 xfs_fsblock_t firstfsb
;
4316 xfs_bmap_free_t free_list
;
4317 xfs_bmbt_irec_t imap
;
4325 xfs_fileoff_t startoffset_fsb
;
4327 int need_iolock
= 1;
4332 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
4334 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
4338 if (len
<= 0) /* if nothing being freed */
4340 rt
= (ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
);
4341 startoffset_fsb
= XFS_B_TO_FSB(mp
, offset
);
4342 end_dmi_offset
= offset
+ len
;
4343 endoffset_fsb
= XFS_B_TO_FSBT(mp
, end_dmi_offset
);
4345 if (offset
< ip
->i_size
&&
4346 (attr_flags
& ATTR_DMI
) == 0 &&
4347 DM_EVENT_ENABLED(XFS_MTOVFS(mp
), ip
, DM_EVENT_WRITE
)) {
4348 if (end_dmi_offset
> ip
->i_size
)
4349 end_dmi_offset
= ip
->i_size
;
4350 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, vp
,
4351 offset
, end_dmi_offset
- offset
,
4352 AT_DELAY_FLAG(attr_flags
), NULL
);
4357 if (attr_flags
& ATTR_NOLOCK
)
4360 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
4361 vn_iowait(vp
); /* wait for the completion of any pending DIOs */
4364 rounding
= max_t(uint
, 1 << mp
->m_sb
.sb_blocklog
, NBPP
);
4365 ioffset
= offset
& ~(rounding
- 1);
4367 if (VN_CACHED(vp
) != 0) {
4368 xfs_inval_cached_trace(&ip
->i_iocore
, ioffset
, -1,
4369 ctooff(offtoct(ioffset
)), -1);
4370 error
= bhv_vop_flushinval_pages(vp
, ctooff(offtoct(ioffset
)),
4371 -1, FI_REMAPF_LOCKED
);
4373 goto out_unlock_iolock
;
4377 * Need to zero the stuff we're not freeing, on disk.
4378 * If its a realtime file & can't use unwritten extents then we
4379 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4380 * will take care of it for us.
4382 if (rt
&& !XFS_SB_VERSION_HASEXTFLGBIT(&mp
->m_sb
)) {
4384 error
= XFS_BMAPI(mp
, NULL
, &ip
->i_iocore
, startoffset_fsb
,
4385 1, 0, NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
4387 goto out_unlock_iolock
;
4388 ASSERT(nimap
== 0 || nimap
== 1);
4389 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
4392 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
4393 block
= imap
.br_startblock
;
4394 mod
= do_div(block
, mp
->m_sb
.sb_rextsize
);
4396 startoffset_fsb
+= mp
->m_sb
.sb_rextsize
- mod
;
4399 error
= XFS_BMAPI(mp
, NULL
, &ip
->i_iocore
, endoffset_fsb
- 1,
4400 1, 0, NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
4402 goto out_unlock_iolock
;
4403 ASSERT(nimap
== 0 || nimap
== 1);
4404 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
4405 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
4407 if (mod
&& (mod
!= mp
->m_sb
.sb_rextsize
))
4408 endoffset_fsb
-= mod
;
4411 if ((done
= (endoffset_fsb
<= startoffset_fsb
)))
4413 * One contiguous piece to clear
4415 error
= xfs_zero_remaining_bytes(ip
, offset
, offset
+ len
- 1);
4418 * Some full blocks, possibly two pieces to clear
4420 if (offset
< XFS_FSB_TO_B(mp
, startoffset_fsb
))
4421 error
= xfs_zero_remaining_bytes(ip
, offset
,
4422 XFS_FSB_TO_B(mp
, startoffset_fsb
) - 1);
4424 XFS_FSB_TO_B(mp
, endoffset_fsb
) < offset
+ len
)
4425 error
= xfs_zero_remaining_bytes(ip
,
4426 XFS_FSB_TO_B(mp
, endoffset_fsb
),
4431 * free file space until done or until there is an error
4433 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
4434 while (!error
&& !done
) {
4437 * allocate and setup the transaction
4439 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
4440 error
= xfs_trans_reserve(tp
,
4442 XFS_WRITE_LOG_RES(mp
),
4444 XFS_TRANS_PERM_LOG_RES
,
4445 XFS_WRITE_LOG_COUNT
);
4448 * check for running out of space
4452 * Free the transaction structure.
4454 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
4455 xfs_trans_cancel(tp
, 0);
4458 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4459 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
,
4460 ip
->i_udquot
, ip
->i_gdquot
, resblks
, 0,
4461 XFS_QMOPT_RES_REGBLKS
);
4465 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
4466 xfs_trans_ihold(tp
, ip
);
4469 * issue the bunmapi() call to free the blocks
4471 XFS_BMAP_INIT(&free_list
, &firstfsb
);
4472 error
= XFS_BUNMAPI(mp
, tp
, &ip
->i_iocore
, startoffset_fsb
,
4473 endoffset_fsb
- startoffset_fsb
,
4474 0, 2, &firstfsb
, &free_list
, NULL
, &done
);
4480 * complete the transaction
4482 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
4487 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
4488 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4493 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
4497 xfs_bmap_cancel(&free_list
);
4499 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
4500 xfs_iunlock(ip
, need_iolock
? (XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
) :
4506 * xfs_change_file_space()
4507 * This routine allocates or frees disk space for the given file.
4508 * The user specified parameters are checked for alignment and size
4517 xfs_change_file_space(
4531 xfs_off_t startoffset
;
4537 vp
= BHV_TO_VNODE(bdp
);
4538 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
4540 ip
= XFS_BHVTOI(bdp
);
4544 * must be a regular file and have write permission
4547 return XFS_ERROR(EINVAL
);
4549 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
4551 if ((error
= xfs_iaccess(ip
, S_IWUSR
, credp
))) {
4552 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
4556 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
4558 switch (bf
->l_whence
) {
4559 case 0: /*SEEK_SET*/
4561 case 1: /*SEEK_CUR*/
4562 bf
->l_start
+= offset
;
4564 case 2: /*SEEK_END*/
4565 bf
->l_start
+= ip
->i_size
;
4568 return XFS_ERROR(EINVAL
);
4571 llen
= bf
->l_len
> 0 ? bf
->l_len
- 1 : bf
->l_len
;
4573 if ( (bf
->l_start
< 0)
4574 || (bf
->l_start
> XFS_MAXIOFFSET(mp
))
4575 || (bf
->l_start
+ llen
< 0)
4576 || (bf
->l_start
+ llen
> XFS_MAXIOFFSET(mp
)))
4577 return XFS_ERROR(EINVAL
);
4581 startoffset
= bf
->l_start
;
4585 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4587 * These calls do NOT zero the data space allocated to the file,
4588 * nor do they change the file size.
4590 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4592 * These calls cause the new file data to be zeroed and the file
4593 * size to be changed.
4595 setprealloc
= clrprealloc
= 0;
4598 case XFS_IOC_RESVSP
:
4599 case XFS_IOC_RESVSP64
:
4600 error
= xfs_alloc_file_space(ip
, startoffset
, bf
->l_len
,
4607 case XFS_IOC_UNRESVSP
:
4608 case XFS_IOC_UNRESVSP64
:
4609 if ((error
= xfs_free_file_space(ip
, startoffset
, bf
->l_len
,
4614 case XFS_IOC_ALLOCSP
:
4615 case XFS_IOC_ALLOCSP64
:
4616 case XFS_IOC_FREESP
:
4617 case XFS_IOC_FREESP64
:
4618 if (startoffset
> fsize
) {
4619 error
= xfs_alloc_file_space(ip
, fsize
,
4620 startoffset
- fsize
, 0, attr_flags
);
4625 va
.va_mask
= XFS_AT_SIZE
;
4626 va
.va_size
= startoffset
;
4628 error
= xfs_setattr(bdp
, &va
, attr_flags
, credp
);
4638 return XFS_ERROR(EINVAL
);
4642 * update the inode timestamp, mode, and prealloc flag bits
4644 tp
= xfs_trans_alloc(mp
, XFS_TRANS_WRITEID
);
4646 if ((error
= xfs_trans_reserve(tp
, 0, XFS_WRITEID_LOG_RES(mp
),
4649 xfs_trans_cancel(tp
, 0);
4653 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4655 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
4656 xfs_trans_ihold(tp
, ip
);
4658 if ((attr_flags
& ATTR_DMI
) == 0) {
4659 ip
->i_d
.di_mode
&= ~S_ISUID
;
4662 * Note that we don't have to worry about mandatory
4663 * file locking being disabled here because we only
4664 * clear the S_ISGID bit if the Group execute bit is
4665 * on, but if it was on then mandatory locking wouldn't
4666 * have been enabled.
4668 if (ip
->i_d
.di_mode
& S_IXGRP
)
4669 ip
->i_d
.di_mode
&= ~S_ISGID
;
4671 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
4674 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
4675 else if (clrprealloc
)
4676 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
4678 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
4679 xfs_trans_set_sync(tp
);
4681 error
= xfs_trans_commit(tp
, 0);
4683 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4688 bhv_vnodeops_t xfs_vnodeops
= {
4689 BHV_IDENTITY_INIT(VN_BHV_XFS
,VNODE_POSITION_XFS
),
4690 .vop_open
= xfs_open
,
4691 .vop_read
= xfs_read
,
4693 .vop_splice_read
= xfs_splice_read
,
4694 .vop_splice_write
= xfs_splice_write
,
4696 .vop_write
= xfs_write
,
4697 .vop_ioctl
= xfs_ioctl
,
4698 .vop_getattr
= xfs_getattr
,
4699 .vop_setattr
= xfs_setattr
,
4700 .vop_access
= xfs_access
,
4701 .vop_lookup
= xfs_lookup
,
4702 .vop_create
= xfs_create
,
4703 .vop_remove
= xfs_remove
,
4704 .vop_link
= xfs_link
,
4705 .vop_rename
= xfs_rename
,
4706 .vop_mkdir
= xfs_mkdir
,
4707 .vop_rmdir
= xfs_rmdir
,
4708 .vop_readdir
= xfs_readdir
,
4709 .vop_symlink
= xfs_symlink
,
4710 .vop_readlink
= xfs_readlink
,
4711 .vop_fsync
= xfs_fsync
,
4712 .vop_inactive
= xfs_inactive
,
4713 .vop_fid2
= xfs_fid2
,
4714 .vop_rwlock
= xfs_rwlock
,
4715 .vop_rwunlock
= xfs_rwunlock
,
4716 .vop_bmap
= xfs_bmap
,
4717 .vop_reclaim
= xfs_reclaim
,
4718 .vop_attr_get
= xfs_attr_get
,
4719 .vop_attr_set
= xfs_attr_set
,
4720 .vop_attr_remove
= xfs_attr_remove
,
4721 .vop_attr_list
= xfs_attr_list
,
4722 .vop_link_removed
= (vop_link_removed_t
)fs_noval
,
4723 .vop_vnode_change
= (vop_vnode_change_t
)fs_noval
,
4724 .vop_tosspages
= fs_tosspages
,
4725 .vop_flushinval_pages
= fs_flushinval_pages
,
4726 .vop_flush_pages
= fs_flush_pages
,
4727 .vop_release
= xfs_release
,
4728 .vop_iflush
= xfs_inode_flush
,