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"
61 bhv_vnode_t
*vp
= BHV_TO_VNODE(bdp
);
62 xfs_inode_t
*ip
= XFS_BHVTOI(bdp
);
64 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
65 return XFS_ERROR(EIO
);
68 * If it's a directory with any blocks, read-ahead block 0
69 * as we're almost certain to have the next operation be a read there.
71 if (VN_ISDIR(vp
) && ip
->i_d
.di_nextents
> 0) {
72 mode
= xfs_ilock_map_shared(ip
);
73 if (ip
->i_d
.di_nextents
> 0)
74 (void)xfs_da_reada_buf(NULL
, ip
, 0, XFS_DATA_FORK
);
75 xfs_iunlock(ip
, mode
);
84 lastclose_t lastclose
,
87 bhv_vnode_t
*vp
= BHV_TO_VNODE(bdp
);
88 xfs_inode_t
*ip
= XFS_BHVTOI(bdp
);
90 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
91 return XFS_ERROR(EIO
);
93 if (lastclose
!= L_TRUE
|| !VN_ISREG(vp
))
97 * If we previously truncated this file and removed old data in
98 * the process, we want to initiate "early" writeout on the last
99 * close. This is an attempt to combat the notorious NULL files
100 * problem which is particularly noticable from a truncate down,
101 * buffered (re-)write (delalloc), followed by a crash. What we
102 * are effectively doing here is significantly reducing the time
103 * window where we'd otherwise be exposed to that problem.
105 if (VUNTRUNCATE(vp
) && VN_DIRTY(vp
) && ip
->i_delayed_blks
> 0)
106 return bhv_vop_flush_pages(vp
, 0, -1, XFS_B_ASYNC
, FI_NONE
);
124 vp
= BHV_TO_VNODE(bdp
);
125 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
127 ip
= XFS_BHVTOI(bdp
);
130 if (XFS_FORCED_SHUTDOWN(mp
))
131 return XFS_ERROR(EIO
);
133 if (!(flags
& ATTR_LAZY
))
134 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
136 vap
->va_size
= XFS_ISIZE(ip
);
137 if (vap
->va_mask
== XFS_AT_SIZE
)
141 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
142 vap
->va_nodeid
= ip
->i_ino
;
144 vap
->va_nodeid
+= mp
->m_inoadd
;
146 vap
->va_nlink
= ip
->i_d
.di_nlink
;
149 * Quick exit for non-stat callers
152 ~(XFS_AT_SIZE
|XFS_AT_FSID
|XFS_AT_NODEID
|
153 XFS_AT_NLINK
|XFS_AT_BLKSIZE
)) == 0)
157 * Copy from in-core inode.
159 vap
->va_mode
= ip
->i_d
.di_mode
;
160 vap
->va_uid
= ip
->i_d
.di_uid
;
161 vap
->va_gid
= ip
->i_d
.di_gid
;
162 vap
->va_projid
= ip
->i_d
.di_projid
;
165 * Check vnode type block/char vs. everything else.
167 switch (ip
->i_d
.di_mode
& S_IFMT
) {
170 vap
->va_rdev
= ip
->i_df
.if_u2
.if_rdev
;
171 vap
->va_blocksize
= BLKDEV_IOSIZE
;
176 if (!(ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
)) {
177 vap
->va_blocksize
= xfs_preferred_iosize(mp
);
181 * If the file blocks are being allocated from a
182 * realtime partition, then return the inode's
183 * realtime extent size or the realtime volume's
186 vap
->va_blocksize
= ip
->i_d
.di_extsize
?
187 (ip
->i_d
.di_extsize
<< mp
->m_sb
.sb_blocklog
) :
188 (mp
->m_sb
.sb_rextsize
<< mp
->m_sb
.sb_blocklog
);
193 vn_atime_to_timespec(vp
, &vap
->va_atime
);
194 vap
->va_mtime
.tv_sec
= ip
->i_d
.di_mtime
.t_sec
;
195 vap
->va_mtime
.tv_nsec
= ip
->i_d
.di_mtime
.t_nsec
;
196 vap
->va_ctime
.tv_sec
= ip
->i_d
.di_ctime
.t_sec
;
197 vap
->va_ctime
.tv_nsec
= ip
->i_d
.di_ctime
.t_nsec
;
200 * Exit for stat callers. See if any of the rest of the fields
201 * to be filled in are needed.
204 (XFS_AT_XFLAGS
|XFS_AT_EXTSIZE
|XFS_AT_NEXTENTS
|XFS_AT_ANEXTENTS
|
205 XFS_AT_GENCOUNT
|XFS_AT_VCODE
)) == 0)
209 * Convert di_flags to xflags.
211 vap
->va_xflags
= xfs_ip2xflags(ip
);
214 * Exit for inode revalidate. See if any of the rest of
215 * the fields to be filled in are needed.
218 (XFS_AT_EXTSIZE
|XFS_AT_NEXTENTS
|XFS_AT_ANEXTENTS
|
219 XFS_AT_GENCOUNT
|XFS_AT_VCODE
)) == 0)
222 vap
->va_extsize
= ip
->i_d
.di_extsize
<< mp
->m_sb
.sb_blocklog
;
224 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) ?
225 ip
->i_df
.if_bytes
/ sizeof(xfs_bmbt_rec_t
) :
229 (ip
->i_afp
->if_flags
& XFS_IFEXTENTS
) ?
230 ip
->i_afp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
) :
231 ip
->i_d
.di_anextents
;
233 vap
->va_anextents
= 0;
234 vap
->va_gen
= ip
->i_d
.di_gen
;
237 if (!(flags
& ATTR_LAZY
))
238 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
264 xfs_prid_t projid
=0, iprojid
=0;
265 int mandlock_before
, mandlock_after
;
266 struct xfs_dquot
*udqp
, *gdqp
, *olddquot1
, *olddquot2
;
270 vp
= BHV_TO_VNODE(bdp
);
271 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
273 if (vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
)
274 return XFS_ERROR(EROFS
);
277 * Cannot set certain attributes.
280 if (mask
& XFS_AT_NOSET
) {
281 return XFS_ERROR(EINVAL
);
284 ip
= XFS_BHVTOI(bdp
);
287 if (XFS_FORCED_SHUTDOWN(mp
))
288 return XFS_ERROR(EIO
);
291 * Timestamps do not need to be logged and hence do not
292 * need to be done within a transaction.
294 if (mask
& XFS_AT_UPDTIMES
) {
295 ASSERT((mask
& ~XFS_AT_UPDTIMES
) == 0);
296 timeflags
= ((mask
& XFS_AT_UPDATIME
) ? XFS_ICHGTIME_ACC
: 0) |
297 ((mask
& XFS_AT_UPDCTIME
) ? XFS_ICHGTIME_CHG
: 0) |
298 ((mask
& XFS_AT_UPDMTIME
) ? XFS_ICHGTIME_MOD
: 0);
299 xfs_ichgtime(ip
, timeflags
);
303 olddquot1
= olddquot2
= NULL
;
307 * If disk quotas is on, we make sure that the dquots do exist on disk,
308 * before we start any other transactions. Trying to do this later
309 * is messy. We don't care to take a readlock to look at the ids
310 * in inode here, because we can't hold it across the trans_reserve.
311 * If the IDs do change before we take the ilock, we're covered
312 * because the i_*dquot fields will get updated anyway.
314 if (XFS_IS_QUOTA_ON(mp
) &&
315 (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
))) {
318 if ((mask
& XFS_AT_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
320 qflags
|= XFS_QMOPT_UQUOTA
;
322 uid
= ip
->i_d
.di_uid
;
324 if ((mask
& XFS_AT_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
326 qflags
|= XFS_QMOPT_GQUOTA
;
328 gid
= ip
->i_d
.di_gid
;
330 if ((mask
& XFS_AT_PROJID
) && XFS_IS_PQUOTA_ON(mp
)) {
331 projid
= vap
->va_projid
;
332 qflags
|= XFS_QMOPT_PQUOTA
;
334 projid
= ip
->i_d
.di_projid
;
337 * We take a reference when we initialize udqp and gdqp,
338 * so it is important that we never blindly double trip on
339 * the same variable. See xfs_create() for an example.
341 ASSERT(udqp
== NULL
);
342 ASSERT(gdqp
== NULL
);
343 code
= XFS_QM_DQVOPALLOC(mp
, ip
, uid
, gid
, projid
, qflags
,
350 * For the other attributes, we acquire the inode lock and
351 * first do an error checking pass.
354 lock_flags
= XFS_ILOCK_EXCL
;
355 if (flags
& ATTR_NOLOCK
)
357 if (!(mask
& XFS_AT_SIZE
)) {
358 if ((mask
!= (XFS_AT_CTIME
|XFS_AT_ATIME
|XFS_AT_MTIME
)) ||
359 (mp
->m_flags
& XFS_MOUNT_WSYNC
)) {
360 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
362 if ((code
= xfs_trans_reserve(tp
, 0,
363 XFS_ICHANGE_LOG_RES(mp
), 0,
370 if (DM_EVENT_ENABLED (vp
->v_vfsp
, ip
, DM_EVENT_TRUNCATE
) &&
371 !(flags
& ATTR_DMI
)) {
372 int dmflags
= AT_DELAY_FLAG(flags
) | DM_SEM_FLAG_WR
;
373 code
= XFS_SEND_DATA(mp
, DM_EVENT_TRUNCATE
, vp
,
374 vap
->va_size
, 0, dmflags
, NULL
);
381 lock_flags
|= XFS_IOLOCK_EXCL
;
384 xfs_ilock(ip
, lock_flags
);
386 /* boolean: are we the file owner? */
387 file_owner
= (current_fsuid(credp
) == ip
->i_d
.di_uid
);
390 * Change various properties of a file.
391 * Only the owner or users with CAP_FOWNER
392 * capability may do these things.
395 (XFS_AT_MODE
|XFS_AT_XFLAGS
|XFS_AT_EXTSIZE
|XFS_AT_UID
|
396 XFS_AT_GID
|XFS_AT_PROJID
)) {
398 * CAP_FOWNER overrides the following restrictions:
400 * The user ID of the calling process must be equal
401 * to the file owner ID, except in cases where the
402 * CAP_FSETID capability is applicable.
404 if (!file_owner
&& !capable(CAP_FOWNER
)) {
405 code
= XFS_ERROR(EPERM
);
410 * CAP_FSETID overrides the following restrictions:
412 * The effective user ID of the calling process shall match
413 * the file owner when setting the set-user-ID and
414 * set-group-ID bits on that file.
416 * The effective group ID or one of the supplementary group
417 * IDs of the calling process shall match the group owner of
418 * the file when setting the set-group-ID bit on that file
420 if (mask
& XFS_AT_MODE
) {
423 if ((vap
->va_mode
& S_ISUID
) && !file_owner
)
425 if ((vap
->va_mode
& S_ISGID
) &&
426 !in_group_p((gid_t
)ip
->i_d
.di_gid
))
429 /* Linux allows this, Irix doesn't. */
430 if ((vap
->va_mode
& S_ISVTX
) && !VN_ISDIR(vp
))
433 if (m
&& !capable(CAP_FSETID
))
439 * Change file ownership. Must be the owner or privileged.
440 * If the system was configured with the "restricted_chown"
441 * option, the owner is not permitted to give away the file,
442 * and can change the group id only to a group of which he
443 * or she is a member.
445 if (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
)) {
447 * These IDs could have changed since we last looked at them.
448 * But, we're assured that if the ownership did change
449 * while we didn't have the inode locked, inode's dquot(s)
450 * would have changed also.
452 iuid
= ip
->i_d
.di_uid
;
453 iprojid
= ip
->i_d
.di_projid
;
454 igid
= ip
->i_d
.di_gid
;
455 gid
= (mask
& XFS_AT_GID
) ? vap
->va_gid
: igid
;
456 uid
= (mask
& XFS_AT_UID
) ? vap
->va_uid
: iuid
;
457 projid
= (mask
& XFS_AT_PROJID
) ? (xfs_prid_t
)vap
->va_projid
:
461 * CAP_CHOWN overrides the following restrictions:
463 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
464 * shall override the restriction that a process cannot
465 * change the user ID of a file it owns and the restriction
466 * that the group ID supplied to the chown() function
467 * shall be equal to either the group ID or one of the
468 * supplementary group IDs of the calling process.
470 if (restricted_chown
&&
471 (iuid
!= uid
|| (igid
!= gid
&&
472 !in_group_p((gid_t
)gid
))) &&
473 !capable(CAP_CHOWN
)) {
474 code
= XFS_ERROR(EPERM
);
478 * Do a quota reservation only if uid/projid/gid is actually
481 if ((XFS_IS_UQUOTA_ON(mp
) && iuid
!= uid
) ||
482 (XFS_IS_PQUOTA_ON(mp
) && iprojid
!= projid
) ||
483 (XFS_IS_GQUOTA_ON(mp
) && igid
!= gid
)) {
485 code
= XFS_QM_DQVOPCHOWNRESV(mp
, tp
, ip
, udqp
, gdqp
,
486 capable(CAP_FOWNER
) ?
487 XFS_QMOPT_FORCE_RES
: 0);
488 if (code
) /* out of quota */
494 * Truncate file. Must have write permission and not be a directory.
496 if (mask
& XFS_AT_SIZE
) {
497 /* Short circuit the truncate case for zero length files */
498 if ((vap
->va_size
== 0) &&
499 (ip
->i_size
== 0) && (ip
->i_d
.di_nextents
== 0)) {
500 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
501 lock_flags
&= ~XFS_ILOCK_EXCL
;
502 if (mask
& XFS_AT_CTIME
)
503 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
509 code
= XFS_ERROR(EISDIR
);
511 } else if (!VN_ISREG(vp
)) {
512 code
= XFS_ERROR(EINVAL
);
516 * Make sure that the dquots are attached to the inode.
518 if ((code
= XFS_QM_DQATTACH(mp
, ip
, XFS_QMOPT_ILOCKED
)))
523 * Change file access or modified times.
525 if (mask
& (XFS_AT_ATIME
|XFS_AT_MTIME
)) {
527 if ((flags
& ATTR_UTIME
) &&
528 !capable(CAP_FOWNER
)) {
529 code
= XFS_ERROR(EPERM
);
536 * Change extent size or realtime flag.
538 if (mask
& (XFS_AT_EXTSIZE
|XFS_AT_XFLAGS
)) {
540 * Can't change extent size if any extents are allocated.
542 if (ip
->i_d
.di_nextents
&& (mask
& XFS_AT_EXTSIZE
) &&
543 ((ip
->i_d
.di_extsize
<< mp
->m_sb
.sb_blocklog
) !=
545 code
= XFS_ERROR(EINVAL
); /* EFBIG? */
550 * Can't change realtime flag if any extents are allocated.
552 if ((ip
->i_d
.di_nextents
|| ip
->i_delayed_blks
) &&
553 (mask
& XFS_AT_XFLAGS
) &&
554 (ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) !=
555 (vap
->va_xflags
& XFS_XFLAG_REALTIME
)) {
556 code
= XFS_ERROR(EINVAL
); /* EFBIG? */
560 * Extent size must be a multiple of the appropriate block
561 * size, if set at all.
563 if ((mask
& XFS_AT_EXTSIZE
) && vap
->va_extsize
!= 0) {
566 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) ||
567 ((mask
& XFS_AT_XFLAGS
) &&
568 (vap
->va_xflags
& XFS_XFLAG_REALTIME
))) {
569 size
= mp
->m_sb
.sb_rextsize
<<
570 mp
->m_sb
.sb_blocklog
;
572 size
= mp
->m_sb
.sb_blocksize
;
574 if (vap
->va_extsize
% size
) {
575 code
= XFS_ERROR(EINVAL
);
580 * If realtime flag is set then must have realtime data.
582 if ((mask
& XFS_AT_XFLAGS
) &&
583 (vap
->va_xflags
& XFS_XFLAG_REALTIME
)) {
584 if ((mp
->m_sb
.sb_rblocks
== 0) ||
585 (mp
->m_sb
.sb_rextsize
== 0) ||
586 (ip
->i_d
.di_extsize
% mp
->m_sb
.sb_rextsize
)) {
587 code
= XFS_ERROR(EINVAL
);
593 * Can't modify an immutable/append-only file unless
594 * we have appropriate permission.
596 if ((mask
& XFS_AT_XFLAGS
) &&
598 (XFS_DIFLAG_IMMUTABLE
|XFS_DIFLAG_APPEND
) ||
600 (XFS_XFLAG_IMMUTABLE
| XFS_XFLAG_APPEND
))) &&
601 !capable(CAP_LINUX_IMMUTABLE
)) {
602 code
= XFS_ERROR(EPERM
);
608 * Now we can make the changes. Before we join the inode
609 * to the transaction, if XFS_AT_SIZE is set then take care of
610 * the part of the truncation that must be done without the
611 * inode lock. This needs to be done before joining the inode
612 * to the transaction, because the inode cannot be unlocked
613 * once it is a part of the transaction.
615 if (mask
& XFS_AT_SIZE
) {
617 if ((vap
->va_size
> ip
->i_size
) &&
618 (flags
& ATTR_NOSIZETOK
) == 0) {
619 code
= xfs_igrow_start(ip
, vap
->va_size
, credp
);
621 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
622 vn_iowait(vp
); /* wait for the completion of any pending DIOs */
624 code
= xfs_itruncate_data(ip
, vap
->va_size
);
627 lock_flags
&= ~XFS_ILOCK_EXCL
;
628 ASSERT(lock_flags
== XFS_IOLOCK_EXCL
);
631 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
632 if ((code
= xfs_trans_reserve(tp
, 0,
633 XFS_ITRUNCATE_LOG_RES(mp
), 0,
634 XFS_TRANS_PERM_LOG_RES
,
635 XFS_ITRUNCATE_LOG_COUNT
))) {
636 xfs_trans_cancel(tp
, 0);
638 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
641 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
642 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
646 xfs_trans_ijoin(tp
, ip
, lock_flags
);
647 xfs_trans_ihold(tp
, ip
);
650 /* determine whether mandatory locking mode changes */
651 mandlock_before
= MANDLOCK(vp
, ip
->i_d
.di_mode
);
654 * Truncate file. Must have write permission and not be a directory.
656 if (mask
& XFS_AT_SIZE
) {
657 if (vap
->va_size
> ip
->i_size
) {
658 xfs_igrow_finish(tp
, ip
, vap
->va_size
,
659 !(flags
& ATTR_DMI
));
660 } else if ((vap
->va_size
<= ip
->i_size
) ||
661 ((vap
->va_size
== 0) && ip
->i_d
.di_nextents
)) {
663 * signal a sync transaction unless
664 * we're truncating an already unlinked
665 * file on a wsync filesystem
667 code
= xfs_itruncate_finish(&tp
, ip
,
668 (xfs_fsize_t
)vap
->va_size
,
670 ((ip
->i_d
.di_nlink
!= 0 ||
671 !(mp
->m_flags
& XFS_MOUNT_WSYNC
))
676 * Truncated "down", so we're removing references
677 * to old data here - if we now delay flushing for
678 * a long time, we expose ourselves unduly to the
679 * notorious NULL files problem. So, we mark this
680 * vnode and flush it when the file is closed, and
681 * do not wait the usual (long) time for writeout.
686 * Have to do this even if the file's size doesn't change.
688 timeflags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
692 * Change file access modes.
694 if (mask
& XFS_AT_MODE
) {
695 ip
->i_d
.di_mode
&= S_IFMT
;
696 ip
->i_d
.di_mode
|= vap
->va_mode
& ~S_IFMT
;
698 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
699 timeflags
|= XFS_ICHGTIME_CHG
;
703 * Change file ownership. Must be the owner or privileged.
704 * If the system was configured with the "restricted_chown"
705 * option, the owner is not permitted to give away the file,
706 * and can change the group id only to a group of which he
707 * or she is a member.
709 if (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
)) {
711 * CAP_FSETID overrides the following restrictions:
713 * The set-user-ID and set-group-ID bits of a file will be
714 * cleared upon successful return from chown()
716 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
717 !capable(CAP_FSETID
)) {
718 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
722 * Change the ownerships and register quota modifications
723 * in the transaction.
726 if (XFS_IS_UQUOTA_ON(mp
)) {
727 ASSERT(mask
& XFS_AT_UID
);
729 olddquot1
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
730 &ip
->i_udquot
, udqp
);
732 ip
->i_d
.di_uid
= uid
;
735 if (XFS_IS_GQUOTA_ON(mp
)) {
736 ASSERT(!XFS_IS_PQUOTA_ON(mp
));
737 ASSERT(mask
& XFS_AT_GID
);
739 olddquot2
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
740 &ip
->i_gdquot
, gdqp
);
742 ip
->i_d
.di_gid
= gid
;
744 if (iprojid
!= projid
) {
745 if (XFS_IS_PQUOTA_ON(mp
)) {
746 ASSERT(!XFS_IS_GQUOTA_ON(mp
));
747 ASSERT(mask
& XFS_AT_PROJID
);
749 olddquot2
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
750 &ip
->i_gdquot
, gdqp
);
752 ip
->i_d
.di_projid
= projid
;
754 * We may have to rev the inode as well as
755 * the superblock version number since projids didn't
756 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
758 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
)
759 xfs_bump_ino_vers2(tp
, ip
);
762 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
763 timeflags
|= XFS_ICHGTIME_CHG
;
768 * Change file access or modified times.
770 if (mask
& (XFS_AT_ATIME
|XFS_AT_MTIME
)) {
771 if (mask
& XFS_AT_ATIME
) {
772 ip
->i_d
.di_atime
.t_sec
= vap
->va_atime
.tv_sec
;
773 ip
->i_d
.di_atime
.t_nsec
= vap
->va_atime
.tv_nsec
;
774 ip
->i_update_core
= 1;
775 timeflags
&= ~XFS_ICHGTIME_ACC
;
777 if (mask
& XFS_AT_MTIME
) {
778 ip
->i_d
.di_mtime
.t_sec
= vap
->va_mtime
.tv_sec
;
779 ip
->i_d
.di_mtime
.t_nsec
= vap
->va_mtime
.tv_nsec
;
780 timeflags
&= ~XFS_ICHGTIME_MOD
;
781 timeflags
|= XFS_ICHGTIME_CHG
;
783 if (tp
&& (flags
& ATTR_UTIME
))
784 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
788 * Change XFS-added attributes.
790 if (mask
& (XFS_AT_EXTSIZE
|XFS_AT_XFLAGS
)) {
791 if (mask
& XFS_AT_EXTSIZE
) {
793 * Converting bytes to fs blocks.
795 ip
->i_d
.di_extsize
= vap
->va_extsize
>>
796 mp
->m_sb
.sb_blocklog
;
798 if (mask
& XFS_AT_XFLAGS
) {
801 /* can't set PREALLOC this way, just preserve it */
802 di_flags
= (ip
->i_d
.di_flags
& XFS_DIFLAG_PREALLOC
);
803 if (vap
->va_xflags
& XFS_XFLAG_IMMUTABLE
)
804 di_flags
|= XFS_DIFLAG_IMMUTABLE
;
805 if (vap
->va_xflags
& XFS_XFLAG_APPEND
)
806 di_flags
|= XFS_DIFLAG_APPEND
;
807 if (vap
->va_xflags
& XFS_XFLAG_SYNC
)
808 di_flags
|= XFS_DIFLAG_SYNC
;
809 if (vap
->va_xflags
& XFS_XFLAG_NOATIME
)
810 di_flags
|= XFS_DIFLAG_NOATIME
;
811 if (vap
->va_xflags
& XFS_XFLAG_NODUMP
)
812 di_flags
|= XFS_DIFLAG_NODUMP
;
813 if (vap
->va_xflags
& XFS_XFLAG_PROJINHERIT
)
814 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
815 if (vap
->va_xflags
& XFS_XFLAG_NODEFRAG
)
816 di_flags
|= XFS_DIFLAG_NODEFRAG
;
817 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
818 if (vap
->va_xflags
& XFS_XFLAG_RTINHERIT
)
819 di_flags
|= XFS_DIFLAG_RTINHERIT
;
820 if (vap
->va_xflags
& XFS_XFLAG_NOSYMLINKS
)
821 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
822 if (vap
->va_xflags
& XFS_XFLAG_EXTSZINHERIT
)
823 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
824 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
825 if (vap
->va_xflags
& XFS_XFLAG_REALTIME
) {
826 di_flags
|= XFS_DIFLAG_REALTIME
;
827 ip
->i_iocore
.io_flags
|= XFS_IOCORE_RT
;
829 ip
->i_iocore
.io_flags
&= ~XFS_IOCORE_RT
;
831 if (vap
->va_xflags
& XFS_XFLAG_EXTSIZE
)
832 di_flags
|= XFS_DIFLAG_EXTSIZE
;
834 ip
->i_d
.di_flags
= di_flags
;
836 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
837 timeflags
|= XFS_ICHGTIME_CHG
;
841 * Change file inode change time only if XFS_AT_CTIME set
842 * AND we have been called by a DMI function.
845 if ( (flags
& ATTR_DMI
) && (mask
& XFS_AT_CTIME
) ) {
846 ip
->i_d
.di_ctime
.t_sec
= vap
->va_ctime
.tv_sec
;
847 ip
->i_d
.di_ctime
.t_nsec
= vap
->va_ctime
.tv_nsec
;
848 ip
->i_update_core
= 1;
849 timeflags
&= ~XFS_ICHGTIME_CHG
;
853 * Send out timestamp changes that need to be set to the
854 * current time. Not done when called by a DMI function.
856 if (timeflags
&& !(flags
& ATTR_DMI
))
857 xfs_ichgtime(ip
, timeflags
);
859 XFS_STATS_INC(xs_ig_attrchg
);
862 * If this is a synchronous mount, make sure that the
863 * transaction goes to disk before returning to the user.
864 * This is slightly sub-optimal in that truncates require
865 * two sync transactions instead of one for wsync filesystems.
866 * One for the truncate and one for the timestamps since we
867 * don't want to change the timestamps unless we're sure the
868 * truncate worked. Truncates are less than 1% of the laddis
869 * mix so this probably isn't worth the trouble to optimize.
873 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
874 xfs_trans_set_sync(tp
);
876 code
= xfs_trans_commit(tp
, commit_flags
);
880 * If the (regular) file's mandatory locking mode changed, then
881 * notify the vnode. We do this under the inode lock to prevent
882 * racing calls to vop_vnode_change.
884 mandlock_after
= MANDLOCK(vp
, ip
->i_d
.di_mode
);
885 if (mandlock_before
!= mandlock_after
) {
886 bhv_vop_vnode_change(vp
, VCHANGE_FLAGS_ENF_LOCKING
,
890 xfs_iunlock(ip
, lock_flags
);
893 * Release any dquot(s) the inode had kept before chown.
895 XFS_QM_DQRELE(mp
, olddquot1
);
896 XFS_QM_DQRELE(mp
, olddquot2
);
897 XFS_QM_DQRELE(mp
, udqp
);
898 XFS_QM_DQRELE(mp
, gdqp
);
904 if (DM_EVENT_ENABLED(vp
->v_vfsp
, ip
, DM_EVENT_ATTRIBUTE
) &&
905 !(flags
& ATTR_DMI
)) {
906 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_ATTRIBUTE
, vp
, DM_RIGHT_NULL
,
907 NULL
, DM_RIGHT_NULL
, NULL
, NULL
,
908 0, 0, AT_DELAY_FLAG(flags
));
913 commit_flags
|= XFS_TRANS_ABORT
;
916 XFS_QM_DQRELE(mp
, udqp
);
917 XFS_QM_DQRELE(mp
, gdqp
);
919 xfs_trans_cancel(tp
, commit_flags
);
921 if (lock_flags
!= 0) {
922 xfs_iunlock(ip
, lock_flags
);
930 * Null conversion from vnode mode bits to inode mode bits, as in efs.
941 vn_trace_entry(BHV_TO_VNODE(bdp
), __FUNCTION__
,
942 (inst_t
*)__return_address
);
944 ip
= XFS_BHVTOI(bdp
);
945 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
946 error
= xfs_iaccess(ip
, mode
, credp
);
947 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
953 * The maximum pathlen is 1024 bytes. Since the minimum file system
954 * blocksize is 512 bytes, we can get a max of 2 extents back from
957 #define SYMLINK_MAPS 2
978 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
984 vp
= BHV_TO_VNODE(bdp
);
985 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
987 ip
= XFS_BHVTOI(bdp
);
990 if (XFS_FORCED_SHUTDOWN(mp
))
991 return XFS_ERROR(EIO
);
993 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
995 ASSERT((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
);
997 offset
= uiop
->uio_offset
;
998 count
= uiop
->uio_resid
;
1001 error
= XFS_ERROR(EINVAL
);
1010 * See if the symlink is stored inline.
1012 pathlen
= (int)ip
->i_d
.di_size
;
1014 if (ip
->i_df
.if_flags
& XFS_IFINLINE
) {
1015 error
= xfs_uio_read(ip
->i_df
.if_u1
.if_data
, pathlen
, uiop
);
1019 * Symlink not inline. Call bmap to get it in.
1021 nmaps
= SYMLINK_MAPS
;
1023 error
= xfs_bmapi(NULL
, ip
, 0, XFS_B_TO_FSB(mp
, pathlen
),
1024 0, NULL
, 0, mval
, &nmaps
, NULL
, NULL
);
1030 for (n
= 0; n
< nmaps
; n
++) {
1031 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
1032 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
1033 bp
= xfs_buf_read(mp
->m_ddev_targp
, d
,
1034 BTOBB(byte_cnt
), 0);
1035 error
= XFS_BUF_GETERROR(bp
);
1037 xfs_ioerror_alert("xfs_readlink",
1038 ip
->i_mount
, bp
, XFS_BUF_ADDR(bp
));
1042 if (pathlen
< byte_cnt
)
1044 pathlen
-= byte_cnt
;
1046 error
= xfs_uio_read(XFS_BUF_PTR(bp
), byte_cnt
, uiop
);
1053 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1061 * This is called to sync the inode and its data out to disk.
1062 * We need to hold the I/O lock while flushing the data, and
1063 * the inode lock while flushing the inode. The inode lock CANNOT
1064 * be held while flushing the data, so acquire after we're done
1078 int log_flushed
= 0, changed
= 1;
1080 vn_trace_entry(BHV_TO_VNODE(bdp
),
1081 __FUNCTION__
, (inst_t
*)__return_address
);
1083 ip
= XFS_BHVTOI(bdp
);
1085 ASSERT(start
>= 0 && stop
>= -1);
1087 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
1088 return XFS_ERROR(EIO
);
1091 * We always need to make sure that the required inode state
1092 * is safe on disk. The vnode might be clean but because
1093 * of committed transactions that haven't hit the disk yet.
1094 * Likewise, there could be unflushed non-transactional
1095 * changes to the inode core that have to go to disk.
1097 * The following code depends on one assumption: that
1098 * any transaction that changes an inode logs the core
1099 * because it has to change some field in the inode core
1100 * (typically nextents or nblocks). That assumption
1101 * implies that any transactions against an inode will
1102 * catch any non-transactional updates. If inode-altering
1103 * transactions exist that violate this assumption, the
1104 * code breaks. Right now, it figures that if the involved
1105 * update_* field is clear and the inode is unpinned, the
1106 * inode is clean. Either it's been flushed or it's been
1107 * committed and the commit has hit the disk unpinning the inode.
1108 * (Note that xfs_inode_item_format() called at commit clears
1109 * the update_* fields.)
1111 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1113 /* If we are flushing data then we care about update_size
1114 * being set, otherwise we care about update_core
1116 if ((flag
& FSYNC_DATA
) ?
1117 (ip
->i_update_size
== 0) :
1118 (ip
->i_update_core
== 0)) {
1120 * Timestamps/size haven't changed since last inode
1121 * flush or inode transaction commit. That means
1122 * either nothing got written or a transaction
1123 * committed which caught the updates. If the
1124 * latter happened and the transaction hasn't
1125 * hit the disk yet, the inode will be still
1126 * be pinned. If it is, force the log.
1129 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1131 if (xfs_ipincount(ip
)) {
1132 _xfs_log_force(ip
->i_mount
, (xfs_lsn_t
)0,
1134 ((flag
& FSYNC_WAIT
)
1135 ? XFS_LOG_SYNC
: 0),
1139 * If the inode is not pinned and nothing
1140 * has changed we don't need to flush the
1148 * Kick off a transaction to log the inode
1149 * core to get the updates. Make it
1150 * sync if FSYNC_WAIT is passed in (which
1151 * is done by everybody but specfs). The
1152 * sync transaction will also force the log.
1154 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1155 tp
= xfs_trans_alloc(ip
->i_mount
, XFS_TRANS_FSYNC_TS
);
1156 if ((error
= xfs_trans_reserve(tp
, 0,
1157 XFS_FSYNC_TS_LOG_RES(ip
->i_mount
),
1159 xfs_trans_cancel(tp
, 0);
1162 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1165 * Note - it's possible that we might have pushed
1166 * ourselves out of the way during trans_reserve
1167 * which would flush the inode. But there's no
1168 * guarantee that the inode buffer has actually
1169 * gone out yet (it's delwri). Plus the buffer
1170 * could be pinned anyway if it's part of an
1171 * inode in another recent transaction. So we
1172 * play it safe and fire off the transaction anyway.
1174 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1175 xfs_trans_ihold(tp
, ip
);
1176 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1177 if (flag
& FSYNC_WAIT
)
1178 xfs_trans_set_sync(tp
);
1179 error
= _xfs_trans_commit(tp
, 0, &log_flushed
);
1181 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1184 if ((ip
->i_mount
->m_flags
& XFS_MOUNT_BARRIER
) && changed
) {
1186 * If the log write didn't issue an ordered tag we need
1187 * to flush the disk cache for the data device now.
1190 xfs_blkdev_issue_flush(ip
->i_mount
->m_ddev_targp
);
1193 * If this inode is on the RT dev we need to flush that
1196 if (ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
)
1197 xfs_blkdev_issue_flush(ip
->i_mount
->m_rtdev_targp
);
1204 * This is called by xfs_inactive to free any blocks beyond eof
1205 * when the link count isn't zero and by xfs_dm_punch_hole() when
1206 * punching a hole to EOF.
1216 xfs_fileoff_t end_fsb
;
1217 xfs_fileoff_t last_fsb
;
1218 xfs_filblks_t map_len
;
1220 xfs_bmbt_irec_t imap
;
1221 int use_iolock
= (flags
& XFS_FREE_EOF_LOCK
);
1224 * Figure out if there are any blocks beyond the end
1225 * of the file. If not, then there is nothing to do.
1227 end_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)ip
->i_size
));
1228 last_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1229 map_len
= last_fsb
- end_fsb
;
1234 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1235 error
= XFS_BMAPI(mp
, NULL
, &ip
->i_iocore
, end_fsb
, map_len
, 0,
1236 NULL
, 0, &imap
, &nimaps
, NULL
, NULL
);
1237 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1239 if (!error
&& (nimaps
!= 0) &&
1240 (imap
.br_startblock
!= HOLESTARTBLOCK
||
1241 ip
->i_delayed_blks
)) {
1243 * Attach the dquots to the inode up front.
1245 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
1249 * There are blocks after the end of file.
1250 * Free them up now by truncating the file to
1253 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1256 * Do the xfs_itruncate_start() call before
1257 * reserving any log space because
1258 * itruncate_start will call into the buffer
1259 * cache and we can't
1260 * do that within a transaction.
1263 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1264 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
,
1268 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1272 error
= xfs_trans_reserve(tp
, 0,
1273 XFS_ITRUNCATE_LOG_RES(mp
),
1274 0, XFS_TRANS_PERM_LOG_RES
,
1275 XFS_ITRUNCATE_LOG_COUNT
);
1277 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1278 xfs_trans_cancel(tp
, 0);
1279 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1283 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1284 xfs_trans_ijoin(tp
, ip
,
1287 xfs_trans_ihold(tp
, ip
);
1289 error
= xfs_itruncate_finish(&tp
, ip
,
1294 * If we get an error at this point we
1295 * simply don't bother truncating the file.
1298 xfs_trans_cancel(tp
,
1299 (XFS_TRANS_RELEASE_LOG_RES
|
1302 error
= xfs_trans_commit(tp
,
1303 XFS_TRANS_RELEASE_LOG_RES
);
1305 xfs_iunlock(ip
, (use_iolock
? (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)
1312 * Free a symlink that has blocks associated with it.
1315 xfs_inactive_symlink_rmt(
1323 xfs_fsblock_t first_block
;
1324 xfs_bmap_free_t free_list
;
1327 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
1335 ASSERT(ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
));
1337 * We're freeing a symlink that has some
1338 * blocks allocated to it. Free the
1339 * blocks here. We know that we've got
1340 * either 1 or 2 extents and that we can
1341 * free them all in one bunmapi call.
1343 ASSERT(ip
->i_d
.di_nextents
> 0 && ip
->i_d
.di_nextents
<= 2);
1344 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1345 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
1346 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1347 xfs_trans_cancel(tp
, 0);
1352 * Lock the inode, fix the size, and join it to the transaction.
1353 * Hold it so in the normal path, we still have it locked for
1354 * the second transaction. In the error paths we need it
1355 * held so the cancel won't rele it, see below.
1357 xfs_ilock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1358 size
= (int)ip
->i_d
.di_size
;
1359 ip
->i_d
.di_size
= 0;
1360 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1361 xfs_trans_ihold(tp
, ip
);
1362 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1364 * Find the block(s) so we can inval and unmap them.
1367 XFS_BMAP_INIT(&free_list
, &first_block
);
1368 nmaps
= ARRAY_SIZE(mval
);
1369 if ((error
= xfs_bmapi(tp
, ip
, 0, XFS_B_TO_FSB(mp
, size
),
1370 XFS_BMAPI_METADATA
, &first_block
, 0, mval
, &nmaps
,
1374 * Invalidate the block(s).
1376 for (i
= 0; i
< nmaps
; i
++) {
1377 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
1378 XFS_FSB_TO_DADDR(mp
, mval
[i
].br_startblock
),
1379 XFS_FSB_TO_BB(mp
, mval
[i
].br_blockcount
), 0);
1380 xfs_trans_binval(tp
, bp
);
1383 * Unmap the dead block(s) to the free_list.
1385 if ((error
= xfs_bunmapi(tp
, ip
, 0, size
, XFS_BMAPI_METADATA
, nmaps
,
1386 &first_block
, &free_list
, NULL
, &done
)))
1390 * Commit the first transaction. This logs the EFI and the inode.
1392 if ((error
= xfs_bmap_finish(&tp
, &free_list
, &committed
)))
1395 * The transaction must have been committed, since there were
1396 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1397 * The new tp has the extent freeing and EFDs.
1401 * The first xact was committed, so add the inode to the new one.
1402 * Mark it dirty so it will be logged and moved forward in the log as
1403 * part of every commit.
1405 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1406 xfs_trans_ihold(tp
, ip
);
1407 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1409 * Get a new, empty transaction to return to our caller.
1411 ntp
= xfs_trans_dup(tp
);
1413 * Commit the transaction containing extent freeing and EFDs.
1414 * If we get an error on the commit here or on the reserve below,
1415 * we need to unlock the inode since the new transaction doesn't
1416 * have the inode attached.
1418 error
= xfs_trans_commit(tp
, 0);
1421 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1425 * Remove the memory for extent descriptions (just bookkeeping).
1427 if (ip
->i_df
.if_bytes
)
1428 xfs_idata_realloc(ip
, -ip
->i_df
.if_bytes
, XFS_DATA_FORK
);
1429 ASSERT(ip
->i_df
.if_bytes
== 0);
1431 * Put an itruncate log reservation in the new transaction
1434 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1435 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
1436 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1440 * Return with the inode locked but not joined to the transaction.
1446 xfs_bmap_cancel(&free_list
);
1449 * Have to come here with the inode locked and either
1450 * (held and in the transaction) or (not in the transaction).
1451 * If the inode isn't held then cancel would iput it, but
1452 * that's wrong since this is inactive and the vnode ref
1453 * count is 0 already.
1454 * Cancel won't do anything to the inode if held, but it still
1455 * needs to be locked until the cancel is done, if it was
1456 * joined to the transaction.
1458 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1459 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1466 xfs_inactive_symlink_local(
1472 ASSERT(ip
->i_d
.di_size
<= XFS_IFORK_DSIZE(ip
));
1474 * We're freeing a symlink which fit into
1475 * the inode. Just free the memory used
1476 * to hold the old symlink.
1478 error
= xfs_trans_reserve(*tpp
, 0,
1479 XFS_ITRUNCATE_LOG_RES(ip
->i_mount
),
1480 0, XFS_TRANS_PERM_LOG_RES
,
1481 XFS_ITRUNCATE_LOG_COUNT
);
1484 xfs_trans_cancel(*tpp
, 0);
1488 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1491 * Zero length symlinks _can_ exist.
1493 if (ip
->i_df
.if_bytes
> 0) {
1494 xfs_idata_realloc(ip
,
1495 -(ip
->i_df
.if_bytes
),
1497 ASSERT(ip
->i_df
.if_bytes
== 0);
1511 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
));
1514 ASSERT(ip
->i_d
.di_forkoff
!= 0);
1515 xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1516 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1518 error
= xfs_attr_inactive(ip
);
1521 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1522 return error
; /* goto out */
1525 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1526 error
= xfs_trans_reserve(tp
, 0,
1527 XFS_IFREE_LOG_RES(mp
),
1528 0, XFS_TRANS_PERM_LOG_RES
,
1529 XFS_INACTIVE_LOG_COUNT
);
1531 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1532 xfs_trans_cancel(tp
, 0);
1534 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1538 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1539 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1540 xfs_trans_ihold(tp
, ip
);
1541 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1543 ASSERT(ip
->i_d
.di_anextents
== 0);
1558 vp
= BHV_TO_VNODE(bdp
);
1559 ip
= XFS_BHVTOI(bdp
);
1562 if (!VN_ISREG(vp
) || (ip
->i_d
.di_mode
== 0))
1565 /* If this is a read-only mount, don't do this (would generate I/O) */
1566 if (vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
)
1569 #ifdef HAVE_REFCACHE
1570 /* If we are in the NFS reference cache then don't do this now */
1575 if (ip
->i_d
.di_nlink
!= 0) {
1576 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1577 ((ip
->i_size
> 0) || (VN_CACHED(vp
) > 0 ||
1578 ip
->i_delayed_blks
> 0)) &&
1579 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
)) &&
1580 (!(ip
->i_d
.di_flags
&
1581 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)))) {
1582 error
= xfs_free_eofblocks(mp
, ip
, XFS_FREE_EOF_LOCK
);
1585 /* Update linux inode block count after free above */
1586 vn_to_inode(vp
)->i_blocks
= XFS_FSB_TO_BB(mp
,
1587 ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
1597 * This is called when the vnode reference count for the vnode
1598 * goes to zero. If the file has been unlinked, then it must
1599 * now be truncated. Also, we clear all of the read-ahead state
1600 * kept for the inode here since the file is now closed.
1609 xfs_bmap_free_t free_list
;
1610 xfs_fsblock_t first_block
;
1617 vp
= BHV_TO_VNODE(bdp
);
1618 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
1620 ip
= XFS_BHVTOI(bdp
);
1623 * If the inode is already free, then there can be nothing
1626 if (ip
->i_d
.di_mode
== 0 || VN_BAD(vp
)) {
1627 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1628 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1629 return VN_INACTIVE_CACHE
;
1633 * Only do a truncate if it's a regular file with
1634 * some actual space in it. It's OK to look at the
1635 * inode's fields without the lock because we're the
1636 * only one with a reference to the inode.
1638 truncate
= ((ip
->i_d
.di_nlink
== 0) &&
1639 ((ip
->i_d
.di_size
!= 0) || (ip
->i_size
!= 0) ||
1640 (ip
->i_d
.di_nextents
> 0) || (ip
->i_delayed_blks
> 0)) &&
1641 ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
));
1645 if (ip
->i_d
.di_nlink
== 0 &&
1646 DM_EVENT_ENABLED(vp
->v_vfsp
, ip
, DM_EVENT_DESTROY
)) {
1647 (void) XFS_SEND_DESTROY(mp
, vp
, DM_RIGHT_NULL
);
1652 /* If this is a read-only mount, don't do this (would generate I/O) */
1653 if (vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
)
1656 if (ip
->i_d
.di_nlink
!= 0) {
1657 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1658 ((ip
->i_size
> 0) || (VN_CACHED(vp
) > 0 ||
1659 ip
->i_delayed_blks
> 0)) &&
1660 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) &&
1661 (!(ip
->i_d
.di_flags
&
1662 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)) ||
1663 (ip
->i_delayed_blks
!= 0)))) {
1664 error
= xfs_free_eofblocks(mp
, ip
, XFS_FREE_EOF_LOCK
);
1666 return VN_INACTIVE_CACHE
;
1667 /* Update linux inode block count after free above */
1668 vn_to_inode(vp
)->i_blocks
= XFS_FSB_TO_BB(mp
,
1669 ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
1674 ASSERT(ip
->i_d
.di_nlink
== 0);
1676 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
1677 return VN_INACTIVE_CACHE
;
1679 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1682 * Do the xfs_itruncate_start() call before
1683 * reserving any log space because itruncate_start
1684 * will call into the buffer cache and we can't
1685 * do that within a transaction.
1687 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1689 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
, 0);
1691 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1692 return VN_INACTIVE_CACHE
;
1695 error
= xfs_trans_reserve(tp
, 0,
1696 XFS_ITRUNCATE_LOG_RES(mp
),
1697 0, XFS_TRANS_PERM_LOG_RES
,
1698 XFS_ITRUNCATE_LOG_COUNT
);
1700 /* Don't call itruncate_cleanup */
1701 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1702 xfs_trans_cancel(tp
, 0);
1703 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1704 return VN_INACTIVE_CACHE
;
1707 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1708 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1709 xfs_trans_ihold(tp
, ip
);
1712 * normally, we have to run xfs_itruncate_finish sync.
1713 * But if filesystem is wsync and we're in the inactive
1714 * path, then we know that nlink == 0, and that the
1715 * xaction that made nlink == 0 is permanently committed
1716 * since xfs_remove runs as a synchronous transaction.
1718 error
= xfs_itruncate_finish(&tp
, ip
, 0, XFS_DATA_FORK
,
1719 (!(mp
->m_flags
& XFS_MOUNT_WSYNC
) ? 1 : 0));
1722 xfs_trans_cancel(tp
,
1723 XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1724 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1725 return VN_INACTIVE_CACHE
;
1727 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
) {
1730 * If we get an error while cleaning up a
1731 * symlink we bail out.
1733 error
= (ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
)) ?
1734 xfs_inactive_symlink_rmt(ip
, &tp
) :
1735 xfs_inactive_symlink_local(ip
, &tp
);
1739 return VN_INACTIVE_CACHE
;
1742 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1743 xfs_trans_ihold(tp
, ip
);
1745 error
= xfs_trans_reserve(tp
, 0,
1746 XFS_IFREE_LOG_RES(mp
),
1747 0, XFS_TRANS_PERM_LOG_RES
,
1748 XFS_INACTIVE_LOG_COUNT
);
1750 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1751 xfs_trans_cancel(tp
, 0);
1752 return VN_INACTIVE_CACHE
;
1755 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1756 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1757 xfs_trans_ihold(tp
, ip
);
1761 * If there are attributes associated with the file
1762 * then blow them away now. The code calls a routine
1763 * that recursively deconstructs the attribute fork.
1764 * We need to just commit the current transaction
1765 * because we can't use it for xfs_attr_inactive().
1767 if (ip
->i_d
.di_anextents
> 0) {
1768 error
= xfs_inactive_attrs(ip
, &tp
);
1770 * If we got an error, the transaction is already
1771 * cancelled, and the inode is unlocked. Just get out.
1774 return VN_INACTIVE_CACHE
;
1775 } else if (ip
->i_afp
) {
1776 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1782 XFS_BMAP_INIT(&free_list
, &first_block
);
1783 error
= xfs_ifree(tp
, ip
, &free_list
);
1786 * If we fail to free the inode, shut down. The cancel
1787 * might do that, we need to make sure. Otherwise the
1788 * inode might be lost for a long time or forever.
1790 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1792 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1793 error
, mp
->m_fsname
);
1794 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1796 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
1799 * Credit the quota account(s). The inode is gone.
1801 XFS_TRANS_MOD_DQUOT_BYINO(mp
, tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1804 * Just ignore errors at this point. There is
1805 * nothing we can do except to try to keep going.
1807 (void) xfs_bmap_finish(&tp
, &free_list
, &committed
);
1808 (void) xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1811 * Release the dquots held by inode, if any.
1813 XFS_QM_DQDETACH(mp
, ip
);
1815 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1818 return VN_INACTIVE_CACHE
;
1827 bhv_desc_t
*dir_bdp
,
1828 bhv_vname_t
*dentry
,
1834 xfs_inode_t
*dp
, *ip
;
1838 bhv_vnode_t
*dir_vp
;
1840 dir_vp
= BHV_TO_VNODE(dir_bdp
);
1841 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
1843 dp
= XFS_BHVTOI(dir_bdp
);
1845 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
1846 return XFS_ERROR(EIO
);
1848 lock_mode
= xfs_ilock_map_shared(dp
);
1849 error
= xfs_dir_lookup_int(dir_bdp
, lock_mode
, dentry
, &e_inum
, &ip
);
1851 *vpp
= XFS_ITOV(ip
);
1854 xfs_iunlock_map_shared(dp
, lock_mode
);
1860 * xfs_create (create a new file).
1864 bhv_desc_t
*dir_bdp
,
1865 bhv_vname_t
*dentry
,
1870 char *name
= VNAME(dentry
);
1871 bhv_vnode_t
*dir_vp
;
1872 xfs_inode_t
*dp
, *ip
;
1873 bhv_vnode_t
*vp
= NULL
;
1878 xfs_bmap_free_t free_list
;
1879 xfs_fsblock_t first_block
;
1880 boolean_t dp_joined_to_trans
;
1881 int dm_event_sent
= 0;
1885 struct xfs_dquot
*udqp
, *gdqp
;
1891 dir_vp
= BHV_TO_VNODE(dir_bdp
);
1892 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
1894 dp
= XFS_BHVTOI(dir_bdp
);
1897 dm_di_mode
= vap
->va_mode
;
1898 namelen
= VNAMELEN(dentry
);
1900 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_CREATE
)) {
1901 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_CREATE
,
1902 dir_vp
, DM_RIGHT_NULL
, NULL
,
1903 DM_RIGHT_NULL
, name
, NULL
,
1911 if (XFS_FORCED_SHUTDOWN(mp
))
1912 return XFS_ERROR(EIO
);
1914 /* Return through std_return after this point. */
1917 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1918 prid
= dp
->i_d
.di_projid
;
1919 else if (vap
->va_mask
& XFS_AT_PROJID
)
1920 prid
= (xfs_prid_t
)vap
->va_projid
;
1922 prid
= (xfs_prid_t
)dfltprid
;
1925 * Make sure that we have allocated dquot(s) on disk.
1927 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
1928 current_fsuid(credp
), current_fsgid(credp
), prid
,
1929 XFS_QMOPT_QUOTALL
|XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
1934 dp_joined_to_trans
= B_FALSE
;
1936 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1937 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1938 resblks
= XFS_CREATE_SPACE_RES(mp
, namelen
);
1940 * Initially assume that the file does not exist and
1941 * reserve the resources for that case. If that is not
1942 * the case we'll drop the one we have and get a more
1943 * appropriate transaction later.
1945 error
= xfs_trans_reserve(tp
, resblks
, XFS_CREATE_LOG_RES(mp
), 0,
1946 XFS_TRANS_PERM_LOG_RES
, XFS_CREATE_LOG_COUNT
);
1947 if (error
== ENOSPC
) {
1949 error
= xfs_trans_reserve(tp
, 0, XFS_CREATE_LOG_RES(mp
), 0,
1950 XFS_TRANS_PERM_LOG_RES
, XFS_CREATE_LOG_COUNT
);
1958 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1960 XFS_BMAP_INIT(&free_list
, &first_block
);
1965 * Reserve disk quota and the inode.
1967 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
1971 if (resblks
== 0 && (error
= xfs_dir_canenter(tp
, dp
, name
, namelen
)))
1973 rdev
= (vap
->va_mask
& XFS_AT_RDEV
) ? vap
->va_rdev
: 0;
1974 error
= xfs_dir_ialloc(&tp
, dp
, vap
->va_mode
, 1,
1975 rdev
, credp
, prid
, resblks
> 0,
1978 if (error
== ENOSPC
)
1985 * At this point, we've gotten a newly allocated inode.
1986 * It is locked (and joined to the transaction).
1989 ASSERT(ismrlocked (&ip
->i_lock
, MR_UPDATE
));
1992 * Now we join the directory inode to the transaction.
1993 * We do not do it earlier because xfs_dir_ialloc
1994 * might commit the previous transaction (and release
1999 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2000 dp_joined_to_trans
= B_TRUE
;
2002 error
= xfs_dir_createname(tp
, dp
, name
, namelen
, ip
->i_ino
,
2003 &first_block
, &free_list
, resblks
?
2004 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
2006 ASSERT(error
!= ENOSPC
);
2009 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2010 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2013 * If this is a synchronous mount, make sure that the
2014 * create transaction goes to disk before returning to
2017 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2018 xfs_trans_set_sync(tp
);
2024 * Attach the dquot(s) to the inodes and modify them incore.
2025 * These ids of the inode couldn't have changed since the new
2026 * inode has been locked ever since it was created.
2028 XFS_QM_DQVOPCREATE(mp
, tp
, ip
, udqp
, gdqp
);
2031 * xfs_trans_commit normally decrements the vnode ref count
2032 * when it unlocks the inode. Since we want to return the
2033 * vnode to the caller, we bump the vnode ref count now.
2038 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2040 xfs_bmap_cancel(&free_list
);
2044 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2051 XFS_QM_DQRELE(mp
, udqp
);
2052 XFS_QM_DQRELE(mp
, gdqp
);
2055 * Propagate the fact that the vnode changed after the
2056 * xfs_inode locks have been released.
2058 bhv_vop_vnode_change(vp
, VCHANGE_FLAGS_TRUNCATED
, 3);
2062 /* Fallthrough to std_return with error = 0 */
2065 if ( (*vpp
|| (error
!= 0 && dm_event_sent
!= 0)) &&
2066 DM_EVENT_ENABLED(dir_vp
->v_vfsp
, XFS_BHVTOI(dir_bdp
),
2067 DM_EVENT_POSTCREATE
)) {
2068 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTCREATE
,
2069 dir_vp
, DM_RIGHT_NULL
,
2071 DM_RIGHT_NULL
, name
, NULL
,
2072 dm_di_mode
, error
, 0);
2077 cancel_flags
|= XFS_TRANS_ABORT
;
2082 xfs_trans_cancel(tp
, cancel_flags
);
2084 if (!dp_joined_to_trans
&& (dp
!= NULL
))
2085 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2086 XFS_QM_DQRELE(mp
, udqp
);
2087 XFS_QM_DQRELE(mp
, gdqp
);
2093 * Wait until after the current transaction is aborted to
2094 * release the inode. This prevents recursive transactions
2095 * and deadlocks from xfs_inactive.
2097 cancel_flags
|= XFS_TRANS_ABORT
;
2098 xfs_trans_cancel(tp
, cancel_flags
);
2101 XFS_QM_DQRELE(mp
, udqp
);
2102 XFS_QM_DQRELE(mp
, gdqp
);
2109 * Some counters to see if (and how often) we are hitting some deadlock
2110 * prevention code paths.
2114 int xfs_rm_lock_delays
;
2115 int xfs_rm_attempts
;
2119 * The following routine will lock the inodes associated with the
2120 * directory and the named entry in the directory. The locks are
2121 * acquired in increasing inode number.
2123 * If the entry is "..", then only the directory is locked. The
2124 * vnode ref count will still include that from the .. entry in
2127 * There is a deadlock we need to worry about. If the locked directory is
2128 * in the AIL, it might be blocking up the log. The next inode we lock
2129 * could be already locked by another thread waiting for log space (e.g
2130 * a permanent log reservation with a long running transaction (see
2131 * xfs_itruncate_finish)). To solve this, we must check if the directory
2132 * is in the ail and use lock_nowait. If we can't lock, we need to
2133 * drop the inode lock on the directory and try again. xfs_iunlock will
2134 * potentially push the tail if we were holding up the log.
2137 xfs_lock_dir_and_entry(
2139 xfs_inode_t
*ip
) /* inode of entry 'name' */
2143 xfs_inode_t
*ips
[2];
2152 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
2159 * We want to lock in increasing inum. Since we've already
2160 * acquired the lock on the directory, we may need to release
2161 * if if the inum of the entry turns out to be less.
2163 if (e_inum
> dp
->i_ino
) {
2165 * We are already in the right order, so just
2166 * lock on the inode of the entry.
2167 * We need to use nowait if dp is in the AIL.
2170 lp
= (xfs_log_item_t
*)dp
->i_itemp
;
2171 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
2172 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2179 * Unlock dp and try again.
2180 * xfs_iunlock will try to push the tail
2181 * if the inode is in the AIL.
2184 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2186 if ((attempts
% 5) == 0) {
2187 delay(1); /* Don't just spin the CPU */
2189 xfs_rm_lock_delays
++;
2195 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2197 } else if (e_inum
< dp
->i_ino
) {
2198 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2202 xfs_lock_inodes(ips
, 2, 0, XFS_ILOCK_EXCL
);
2204 /* else e_inum == dp->i_ino */
2205 /* This can happen if we're asked to lock /x/..
2206 * the entry is "..", which is also the parent directory.
2214 int xfs_small_retries
;
2215 int xfs_middle_retries
;
2216 int xfs_lots_retries
;
2217 int xfs_lock_delays
;
2221 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2225 xfs_lock_inumorder(int lock_mode
, int subclass
)
2227 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
2228 lock_mode
|= (subclass
+ XFS_IOLOCK_INUMORDER
) << XFS_IOLOCK_SHIFT
;
2229 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
))
2230 lock_mode
|= (subclass
+ XFS_ILOCK_INUMORDER
) << XFS_ILOCK_SHIFT
;
2236 * The following routine will lock n inodes in exclusive mode.
2237 * We assume the caller calls us with the inodes in i_ino order.
2239 * We need to detect deadlock where an inode that we lock
2240 * is in the AIL and we start waiting for another inode that is locked
2241 * by a thread in a long running transaction (such as truncate). This can
2242 * result in deadlock since the long running trans might need to wait
2243 * for the inode we just locked in order to push the tail and free space
2253 int attempts
= 0, i
, j
, try_lock
;
2256 ASSERT(ips
&& (inodes
>= 2)); /* we need at least two */
2267 for (; i
< inodes
; i
++) {
2270 if (i
&& (ips
[i
] == ips
[i
-1])) /* Already locked */
2274 * If try_lock is not set yet, make sure all locked inodes
2275 * are not in the AIL.
2276 * If any are, set try_lock to be used later.
2280 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
2281 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
2282 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
2289 * If any of the previous locks we have locked is in the AIL,
2290 * we must TRY to get the second and subsequent locks. If
2291 * we can't get any, we must release all we have
2296 /* try_lock must be 0 if i is 0. */
2298 * try_lock means we have an inode locked
2299 * that is in the AIL.
2302 if (!xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
))) {
2306 * Unlock all previous guys and try again.
2307 * xfs_iunlock will try to push the tail
2308 * if the inode is in the AIL.
2311 for(j
= i
- 1; j
>= 0; j
--) {
2314 * Check to see if we've already
2315 * unlocked this one.
2316 * Not the first one going back,
2317 * and the inode ptr is the same.
2319 if ((j
!= (i
- 1)) && ips
[j
] ==
2323 xfs_iunlock(ips
[j
], lock_mode
);
2326 if ((attempts
% 5) == 0) {
2327 delay(1); /* Don't just spin the CPU */
2337 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
2343 if (attempts
< 5) xfs_small_retries
++;
2344 else if (attempts
< 100) xfs_middle_retries
++;
2345 else xfs_lots_retries
++;
2353 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2354 int remove_which_error_return
= 0;
2356 #define REMOVE_DEBUG_TRACE(x)
2357 #endif /* ! DEBUG */
2366 bhv_desc_t
*dir_bdp
,
2367 bhv_vname_t
*dentry
,
2370 bhv_vnode_t
*dir_vp
;
2371 char *name
= VNAME(dentry
);
2372 xfs_inode_t
*dp
, *ip
;
2373 xfs_trans_t
*tp
= NULL
;
2376 xfs_bmap_free_t free_list
;
2377 xfs_fsblock_t first_block
;
2385 dir_vp
= BHV_TO_VNODE(dir_bdp
);
2386 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2388 dp
= XFS_BHVTOI(dir_bdp
);
2391 if (XFS_FORCED_SHUTDOWN(mp
))
2392 return XFS_ERROR(EIO
);
2394 namelen
= VNAMELEN(dentry
);
2396 if (!xfs_get_dir_entry(dentry
, &ip
)) {
2397 dm_di_mode
= ip
->i_d
.di_mode
;
2401 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_REMOVE
)) {
2402 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_REMOVE
, dir_vp
,
2403 DM_RIGHT_NULL
, NULL
, DM_RIGHT_NULL
,
2404 name
, NULL
, dm_di_mode
, 0, 0);
2409 /* From this point on, return through std_return */
2413 * We need to get a reference to ip before we get our log
2414 * reservation. The reason for this is that we cannot call
2415 * xfs_iget for an inode for which we do not have a reference
2416 * once we've acquired a log reservation. This is because the
2417 * inode we are trying to get might be in xfs_inactive going
2418 * for a log reservation. Since we'll have to wait for the
2419 * inactive code to complete before returning from xfs_iget,
2420 * we need to make sure that we don't have log space reserved
2421 * when we call xfs_iget. Instead we get an unlocked reference
2422 * to the inode before getting our log reservation.
2424 error
= xfs_get_dir_entry(dentry
, &ip
);
2426 REMOVE_DEBUG_TRACE(__LINE__
);
2430 dm_di_mode
= ip
->i_d
.di_mode
;
2432 vn_trace_entry(XFS_ITOV(ip
), __FUNCTION__
, (inst_t
*)__return_address
);
2436 error
= XFS_QM_DQATTACH(mp
, dp
, 0);
2437 if (!error
&& dp
!= ip
)
2438 error
= XFS_QM_DQATTACH(mp
, ip
, 0);
2440 REMOVE_DEBUG_TRACE(__LINE__
);
2445 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
2446 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2448 * We try to get the real space reservation first,
2449 * allowing for directory btree deletion(s) implying
2450 * possible bmap insert(s). If we can't get the space
2451 * reservation then we use 0 instead, and avoid the bmap
2452 * btree insert(s) in the directory code by, if the bmap
2453 * insert tries to happen, instead trimming the LAST
2454 * block from the directory.
2456 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2457 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
2458 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
2459 if (error
== ENOSPC
) {
2461 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
2462 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
2465 ASSERT(error
!= ENOSPC
);
2466 REMOVE_DEBUG_TRACE(__LINE__
);
2467 xfs_trans_cancel(tp
, 0);
2472 error
= xfs_lock_dir_and_entry(dp
, ip
);
2474 REMOVE_DEBUG_TRACE(__LINE__
);
2475 xfs_trans_cancel(tp
, cancel_flags
);
2481 * At this point, we've gotten both the directory and the entry
2484 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2487 * Increment vnode ref count only in this case since
2488 * there's an extra vnode reference in the case where
2492 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2496 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2498 XFS_BMAP_INIT(&free_list
, &first_block
);
2499 error
= xfs_dir_removename(tp
, dp
, name
, namelen
, ip
->i_ino
,
2500 &first_block
, &free_list
, 0);
2502 ASSERT(error
!= ENOENT
);
2503 REMOVE_DEBUG_TRACE(__LINE__
);
2506 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2509 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2511 error
= xfs_droplink(tp
, ip
);
2513 REMOVE_DEBUG_TRACE(__LINE__
);
2517 /* Determine if this is the last link while
2518 * we are in the transaction.
2520 link_zero
= (ip
)->i_d
.di_nlink
==0;
2523 * Take an extra ref on the inode so that it doesn't
2524 * go to xfs_inactive() from within the commit.
2529 * If this is a synchronous mount, make sure that the
2530 * remove transaction goes to disk before returning to
2533 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2534 xfs_trans_set_sync(tp
);
2537 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2539 REMOVE_DEBUG_TRACE(__LINE__
);
2543 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2550 * Before we drop our extra reference to the inode, purge it
2551 * from the refcache if it is there. By waiting until afterwards
2552 * to do the IRELE, we ensure that we won't go inactive in the
2553 * xfs_refcache_purge_ip routine (although that would be OK).
2555 xfs_refcache_purge_ip(ip
);
2557 vn_trace_exit(XFS_ITOV(ip
), __FUNCTION__
, (inst_t
*)__return_address
);
2560 * Let interposed file systems know about removed links.
2562 bhv_vop_link_removed(XFS_ITOV(ip
), dir_vp
, link_zero
);
2566 /* Fall through to std_return with error = 0 */
2568 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
,
2569 DM_EVENT_POSTREMOVE
)) {
2570 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTREMOVE
,
2571 dir_vp
, DM_RIGHT_NULL
,
2572 NULL
, DM_RIGHT_NULL
,
2573 name
, NULL
, dm_di_mode
, error
, 0);
2578 xfs_bmap_cancel(&free_list
);
2579 cancel_flags
|= XFS_TRANS_ABORT
;
2580 xfs_trans_cancel(tp
, cancel_flags
);
2585 * In this case make sure to not release the inode until after
2586 * the current transaction is aborted. Releasing it beforehand
2587 * can cause us to go to xfs_inactive and start a recursive
2588 * transaction which can easily deadlock with the current one.
2590 xfs_bmap_cancel(&free_list
);
2591 cancel_flags
|= XFS_TRANS_ABORT
;
2592 xfs_trans_cancel(tp
, cancel_flags
);
2595 * Before we drop our extra reference to the inode, purge it
2596 * from the refcache if it is there. By waiting until afterwards
2597 * to do the IRELE, we ensure that we won't go inactive in the
2598 * xfs_refcache_purge_ip routine (although that would be OK).
2600 xfs_refcache_purge_ip(ip
);
2614 bhv_desc_t
*target_dir_bdp
,
2615 bhv_vnode_t
*src_vp
,
2616 bhv_vname_t
*dentry
,
2619 xfs_inode_t
*tdp
, *sip
;
2622 xfs_inode_t
*ips
[2];
2624 xfs_bmap_free_t free_list
;
2625 xfs_fsblock_t first_block
;
2628 bhv_vnode_t
*target_dir_vp
;
2630 char *target_name
= VNAME(dentry
);
2633 target_dir_vp
= BHV_TO_VNODE(target_dir_bdp
);
2634 vn_trace_entry(target_dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2635 vn_trace_entry(src_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2637 target_namelen
= VNAMELEN(dentry
);
2638 ASSERT(!VN_ISDIR(src_vp
));
2640 sip
= xfs_vtoi(src_vp
);
2641 tdp
= XFS_BHVTOI(target_dir_bdp
);
2643 if (XFS_FORCED_SHUTDOWN(mp
))
2644 return XFS_ERROR(EIO
);
2646 if (DM_EVENT_ENABLED(src_vp
->v_vfsp
, tdp
, DM_EVENT_LINK
)) {
2647 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_LINK
,
2648 target_dir_vp
, DM_RIGHT_NULL
,
2649 src_vp
, DM_RIGHT_NULL
,
2650 target_name
, NULL
, 0, 0, 0);
2655 /* Return through std_return after this point. */
2657 error
= XFS_QM_DQATTACH(mp
, sip
, 0);
2658 if (!error
&& sip
!= tdp
)
2659 error
= XFS_QM_DQATTACH(mp
, tdp
, 0);
2663 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
2664 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2665 resblks
= XFS_LINK_SPACE_RES(mp
, target_namelen
);
2666 error
= xfs_trans_reserve(tp
, resblks
, XFS_LINK_LOG_RES(mp
), 0,
2667 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2668 if (error
== ENOSPC
) {
2670 error
= xfs_trans_reserve(tp
, 0, XFS_LINK_LOG_RES(mp
), 0,
2671 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2678 if (sip
->i_ino
< tdp
->i_ino
) {
2686 xfs_lock_inodes(ips
, 2, 0, XFS_ILOCK_EXCL
);
2689 * Increment vnode ref counts since xfs_trans_commit &
2690 * xfs_trans_cancel will both unlock the inodes and
2691 * decrement the associated ref counts.
2694 VN_HOLD(target_dir_vp
);
2695 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
2696 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
2699 * If the source has too many links, we can't make any more to it.
2701 if (sip
->i_d
.di_nlink
>= XFS_MAXLINK
) {
2702 error
= XFS_ERROR(EMLINK
);
2707 * If we are using project inheritance, we only allow hard link
2708 * creation in our tree when the project IDs are the same; else
2709 * the tree quota mechanism could be circumvented.
2711 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2712 (tdp
->i_d
.di_projid
!= sip
->i_d
.di_projid
))) {
2713 error
= XFS_ERROR(EXDEV
);
2718 (error
= xfs_dir_canenter(tp
, tdp
, target_name
, target_namelen
)))
2721 XFS_BMAP_INIT(&free_list
, &first_block
);
2723 error
= xfs_dir_createname(tp
, tdp
, target_name
, target_namelen
,
2724 sip
->i_ino
, &first_block
, &free_list
,
2728 xfs_ichgtime(tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2730 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
2732 error
= xfs_bumplink(tp
, sip
);
2737 * If this is a synchronous mount, make sure that the
2738 * link transaction goes to disk before returning to
2741 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2742 xfs_trans_set_sync(tp
);
2745 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
2747 xfs_bmap_cancel(&free_list
);
2751 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2755 /* Fall through to std_return with error = 0. */
2757 if (DM_EVENT_ENABLED(src_vp
->v_vfsp
, sip
,
2758 DM_EVENT_POSTLINK
)) {
2759 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTLINK
,
2760 target_dir_vp
, DM_RIGHT_NULL
,
2761 src_vp
, DM_RIGHT_NULL
,
2762 target_name
, NULL
, 0, error
, 0);
2767 cancel_flags
|= XFS_TRANS_ABORT
;
2771 xfs_trans_cancel(tp
, cancel_flags
);
2782 bhv_desc_t
*dir_bdp
,
2783 bhv_vname_t
*dentry
,
2788 char *dir_name
= VNAME(dentry
);
2790 xfs_inode_t
*cdp
; /* inode of created dir */
2791 bhv_vnode_t
*cvp
; /* vnode of created dir */
2797 xfs_bmap_free_t free_list
;
2798 xfs_fsblock_t first_block
;
2799 bhv_vnode_t
*dir_vp
;
2800 boolean_t dp_joined_to_trans
;
2801 boolean_t created
= B_FALSE
;
2802 int dm_event_sent
= 0;
2804 struct xfs_dquot
*udqp
, *gdqp
;
2809 dir_vp
= BHV_TO_VNODE(dir_bdp
);
2810 dp
= XFS_BHVTOI(dir_bdp
);
2813 if (XFS_FORCED_SHUTDOWN(mp
))
2814 return XFS_ERROR(EIO
);
2816 dir_namelen
= VNAMELEN(dentry
);
2819 dp_joined_to_trans
= B_FALSE
;
2820 dm_di_mode
= vap
->va_mode
;
2822 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_CREATE
)) {
2823 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_CREATE
,
2824 dir_vp
, DM_RIGHT_NULL
, NULL
,
2825 DM_RIGHT_NULL
, dir_name
, NULL
,
2832 /* Return through std_return after this point. */
2834 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2838 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
2839 prid
= dp
->i_d
.di_projid
;
2840 else if (vap
->va_mask
& XFS_AT_PROJID
)
2841 prid
= (xfs_prid_t
)vap
->va_projid
;
2843 prid
= (xfs_prid_t
)dfltprid
;
2846 * Make sure that we have allocated dquot(s) on disk.
2848 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
2849 current_fsuid(credp
), current_fsgid(credp
), prid
,
2850 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
2854 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
2855 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2856 resblks
= XFS_MKDIR_SPACE_RES(mp
, dir_namelen
);
2857 error
= xfs_trans_reserve(tp
, resblks
, XFS_MKDIR_LOG_RES(mp
), 0,
2858 XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
);
2859 if (error
== ENOSPC
) {
2861 error
= xfs_trans_reserve(tp
, 0, XFS_MKDIR_LOG_RES(mp
), 0,
2862 XFS_TRANS_PERM_LOG_RES
,
2863 XFS_MKDIR_LOG_COUNT
);
2871 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
2874 * Check for directory link count overflow.
2876 if (dp
->i_d
.di_nlink
>= XFS_MAXLINK
) {
2877 error
= XFS_ERROR(EMLINK
);
2882 * Reserve disk quota and the inode.
2884 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
2889 (error
= xfs_dir_canenter(tp
, dp
, dir_name
, dir_namelen
)))
2892 * create the directory inode.
2894 error
= xfs_dir_ialloc(&tp
, dp
, vap
->va_mode
, 2,
2895 0, credp
, prid
, resblks
> 0,
2898 if (error
== ENOSPC
)
2905 * Now we add the directory inode to the transaction.
2906 * We waited until now since xfs_dir_ialloc might start
2907 * a new transaction. Had we joined the transaction
2908 * earlier, the locks might have gotten released.
2911 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2912 dp_joined_to_trans
= B_TRUE
;
2914 XFS_BMAP_INIT(&free_list
, &first_block
);
2916 error
= xfs_dir_createname(tp
, dp
, dir_name
, dir_namelen
, cdp
->i_ino
,
2917 &first_block
, &free_list
, resblks
?
2918 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
2920 ASSERT(error
!= ENOSPC
);
2923 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2926 * Bump the in memory version number of the parent directory
2927 * so that other processes accessing it will recognize that
2928 * the directory has changed.
2932 error
= xfs_dir_init(tp
, cdp
, dp
);
2937 error
= xfs_bumplink(tp
, dp
);
2941 cvp
= XFS_ITOV(cdp
);
2949 * Attach the dquots to the new inode and modify the icount incore.
2951 XFS_QM_DQVOPCREATE(mp
, tp
, cdp
, udqp
, gdqp
);
2954 * If this is a synchronous mount, make sure that the
2955 * mkdir transaction goes to disk before returning to
2958 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2959 xfs_trans_set_sync(tp
);
2962 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2968 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2969 XFS_QM_DQRELE(mp
, udqp
);
2970 XFS_QM_DQRELE(mp
, gdqp
);
2975 /* Fall through to std_return with error = 0 or errno from
2976 * xfs_trans_commit. */
2979 if ( (created
|| (error
!= 0 && dm_event_sent
!= 0)) &&
2980 DM_EVENT_ENABLED(dir_vp
->v_vfsp
, XFS_BHVTOI(dir_bdp
),
2981 DM_EVENT_POSTCREATE
)) {
2982 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTCREATE
,
2983 dir_vp
, DM_RIGHT_NULL
,
2984 created
? XFS_ITOV(cdp
):NULL
,
2987 dm_di_mode
, error
, 0);
2993 xfs_bmap_cancel(&free_list
);
2995 cancel_flags
|= XFS_TRANS_ABORT
;
2997 xfs_trans_cancel(tp
, cancel_flags
);
2998 XFS_QM_DQRELE(mp
, udqp
);
2999 XFS_QM_DQRELE(mp
, gdqp
);
3001 if (!dp_joined_to_trans
&& (dp
!= NULL
)) {
3002 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
3015 bhv_desc_t
*dir_bdp
,
3016 bhv_vname_t
*dentry
,
3019 char *name
= VNAME(dentry
);
3021 xfs_inode_t
*cdp
; /* child directory */
3025 xfs_bmap_free_t free_list
;
3026 xfs_fsblock_t first_block
;
3029 bhv_vnode_t
*dir_vp
;
3030 int dm_di_mode
= S_IFDIR
;
3035 dir_vp
= BHV_TO_VNODE(dir_bdp
);
3036 dp
= XFS_BHVTOI(dir_bdp
);
3039 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
3041 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp
)->i_mount
))
3042 return XFS_ERROR(EIO
);
3043 namelen
= VNAMELEN(dentry
);
3045 if (!xfs_get_dir_entry(dentry
, &cdp
)) {
3046 dm_di_mode
= cdp
->i_d
.di_mode
;
3050 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_REMOVE
)) {
3051 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_REMOVE
,
3052 dir_vp
, DM_RIGHT_NULL
,
3053 NULL
, DM_RIGHT_NULL
,
3054 name
, NULL
, dm_di_mode
, 0, 0);
3056 return XFS_ERROR(error
);
3059 /* Return through std_return after this point. */
3064 * We need to get a reference to cdp before we get our log
3065 * reservation. The reason for this is that we cannot call
3066 * xfs_iget for an inode for which we do not have a reference
3067 * once we've acquired a log reservation. This is because the
3068 * inode we are trying to get might be in xfs_inactive going
3069 * for a log reservation. Since we'll have to wait for the
3070 * inactive code to complete before returning from xfs_iget,
3071 * we need to make sure that we don't have log space reserved
3072 * when we call xfs_iget. Instead we get an unlocked reference
3073 * to the inode before getting our log reservation.
3075 error
= xfs_get_dir_entry(dentry
, &cdp
);
3077 REMOVE_DEBUG_TRACE(__LINE__
);
3081 dm_di_mode
= cdp
->i_d
.di_mode
;
3084 * Get the dquots for the inodes.
3086 error
= XFS_QM_DQATTACH(mp
, dp
, 0);
3087 if (!error
&& dp
!= cdp
)
3088 error
= XFS_QM_DQATTACH(mp
, cdp
, 0);
3091 REMOVE_DEBUG_TRACE(__LINE__
);
3095 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
3096 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
3098 * We try to get the real space reservation first,
3099 * allowing for directory btree deletion(s) implying
3100 * possible bmap insert(s). If we can't get the space
3101 * reservation then we use 0 instead, and avoid the bmap
3102 * btree insert(s) in the directory code by, if the bmap
3103 * insert tries to happen, instead trimming the LAST
3104 * block from the directory.
3106 resblks
= XFS_REMOVE_SPACE_RES(mp
);
3107 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
3108 XFS_TRANS_PERM_LOG_RES
, XFS_DEFAULT_LOG_COUNT
);
3109 if (error
== ENOSPC
) {
3111 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
3112 XFS_TRANS_PERM_LOG_RES
, XFS_DEFAULT_LOG_COUNT
);
3115 ASSERT(error
!= ENOSPC
);
3120 XFS_BMAP_INIT(&free_list
, &first_block
);
3123 * Now lock the child directory inode and the parent directory
3124 * inode in the proper order. This will take care of validating
3125 * that the directory entry for the child directory inode has
3126 * not changed while we were obtaining a log reservation.
3128 error
= xfs_lock_dir_and_entry(dp
, cdp
);
3130 xfs_trans_cancel(tp
, cancel_flags
);
3135 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
3138 * Only increment the parent directory vnode count if
3139 * we didn't bump it in looking up cdp. The only time
3140 * we don't bump it is when we're looking up ".".
3146 xfs_trans_ijoin(tp
, cdp
, XFS_ILOCK_EXCL
);
3148 ASSERT(cdp
->i_d
.di_nlink
>= 2);
3149 if (cdp
->i_d
.di_nlink
!= 2) {
3150 error
= XFS_ERROR(ENOTEMPTY
);
3153 if (!xfs_dir_isempty(cdp
)) {
3154 error
= XFS_ERROR(ENOTEMPTY
);
3158 error
= xfs_dir_removename(tp
, dp
, name
, namelen
, cdp
->i_ino
,
3159 &first_block
, &free_list
, resblks
);
3163 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3166 * Bump the in memory generation count on the parent
3167 * directory so that other can know that it has changed.
3172 * Drop the link from cdp's "..".
3174 error
= xfs_droplink(tp
, dp
);
3180 * Drop the link from dp to cdp.
3182 error
= xfs_droplink(tp
, cdp
);
3188 * Drop the "." link from cdp to self.
3190 error
= xfs_droplink(tp
, cdp
);
3195 /* Determine these before committing transaction */
3196 last_cdp_link
= (cdp
)->i_d
.di_nlink
==0;
3199 * Take an extra ref on the child vnode so that it
3200 * does not go to xfs_inactive() from within the commit.
3205 * If this is a synchronous mount, make sure that the
3206 * rmdir transaction goes to disk before returning to
3209 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
3210 xfs_trans_set_sync(tp
);
3213 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
3215 xfs_bmap_cancel(&free_list
);
3216 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
|
3222 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
3230 * Let interposed file systems know about removed links.
3232 bhv_vop_link_removed(XFS_ITOV(cdp
), dir_vp
, last_cdp_link
);
3236 /* Fall through to std_return with error = 0 or the errno
3237 * from xfs_trans_commit. */
3239 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_POSTREMOVE
)) {
3240 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTREMOVE
,
3241 dir_vp
, DM_RIGHT_NULL
,
3242 NULL
, DM_RIGHT_NULL
,
3243 name
, NULL
, dm_di_mode
,
3249 xfs_bmap_cancel(&free_list
);
3250 cancel_flags
|= XFS_TRANS_ABORT
;
3254 xfs_trans_cancel(tp
, cancel_flags
);
3260 * Read dp's entries starting at uiop->uio_offset and translate them into
3261 * bufsize bytes worth of struct dirents starting at bufbase.
3265 bhv_desc_t
*dir_bdp
,
3271 xfs_trans_t
*tp
= NULL
;
3275 vn_trace_entry(BHV_TO_VNODE(dir_bdp
), __FUNCTION__
,
3276 (inst_t
*)__return_address
);
3277 dp
= XFS_BHVTOI(dir_bdp
);
3279 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
3280 return XFS_ERROR(EIO
);
3282 lock_mode
= xfs_ilock_map_shared(dp
);
3283 error
= xfs_dir_getdents(tp
, dp
, uiop
, eofp
);
3284 xfs_iunlock_map_shared(dp
, lock_mode
);
3291 bhv_desc_t
*dir_bdp
,
3292 bhv_vname_t
*dentry
,
3304 xfs_bmap_free_t free_list
;
3305 xfs_fsblock_t first_block
;
3306 boolean_t dp_joined_to_trans
;
3307 bhv_vnode_t
*dir_vp
;
3310 xfs_fileoff_t first_fsb
;
3311 xfs_filblks_t fs_blocks
;
3313 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
3320 struct xfs_dquot
*udqp
, *gdqp
;
3322 char *link_name
= VNAME(dentry
);
3326 dir_vp
= BHV_TO_VNODE(dir_bdp
);
3327 dp
= XFS_BHVTOI(dir_bdp
);
3328 dp_joined_to_trans
= B_FALSE
;
3333 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
3337 if (XFS_FORCED_SHUTDOWN(mp
))
3338 return XFS_ERROR(EIO
);
3340 link_namelen
= VNAMELEN(dentry
);
3343 * Check component lengths of the target path name.
3345 pathlen
= strlen(target_path
);
3346 if (pathlen
>= MAXPATHLEN
) /* total string too long */
3347 return XFS_ERROR(ENAMETOOLONG
);
3348 if (pathlen
>= MAXNAMELEN
) { /* is any component too long? */
3352 for (total
= 0, path
= target_path
; total
< pathlen
;) {
3356 while(*path
== '/') {
3362 * Count up to the next slash or end of path.
3363 * Error out if the component is bigger than MAXNAMELEN.
3365 for(len
= 0; *path
!= '/' && total
< pathlen
;total
++, path
++) {
3366 if (++len
>= MAXNAMELEN
) {
3367 error
= ENAMETOOLONG
;
3374 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_SYMLINK
)) {
3375 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_SYMLINK
, dir_vp
,
3376 DM_RIGHT_NULL
, NULL
, DM_RIGHT_NULL
,
3377 link_name
, target_path
, 0, 0, 0);
3382 /* Return through std_return after this point. */
3385 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
3386 prid
= dp
->i_d
.di_projid
;
3387 else if (vap
->va_mask
& XFS_AT_PROJID
)
3388 prid
= (xfs_prid_t
)vap
->va_projid
;
3390 prid
= (xfs_prid_t
)dfltprid
;
3393 * Make sure that we have allocated dquot(s) on disk.
3395 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
3396 current_fsuid(credp
), current_fsgid(credp
), prid
,
3397 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
3401 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SYMLINK
);
3402 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
3404 * The symlink will fit into the inode data fork?
3405 * There can't be any attributes so we get the whole variable part.
3407 if (pathlen
<= XFS_LITINO(mp
))
3410 fs_blocks
= XFS_B_TO_FSB(mp
, pathlen
);
3411 resblks
= XFS_SYMLINK_SPACE_RES(mp
, link_namelen
, fs_blocks
);
3412 error
= xfs_trans_reserve(tp
, resblks
, XFS_SYMLINK_LOG_RES(mp
), 0,
3413 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
3414 if (error
== ENOSPC
&& fs_blocks
== 0) {
3416 error
= xfs_trans_reserve(tp
, 0, XFS_SYMLINK_LOG_RES(mp
), 0,
3417 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
3425 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
3428 * Check whether the directory allows new symlinks or not.
3430 if (dp
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) {
3431 error
= XFS_ERROR(EPERM
);
3436 * Reserve disk quota : blocks and inode.
3438 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
3443 * Check for ability to enter directory entry, if no space reserved.
3446 (error
= xfs_dir_canenter(tp
, dp
, link_name
, link_namelen
)))
3449 * Initialize the bmap freelist prior to calling either
3450 * bmapi or the directory create code.
3452 XFS_BMAP_INIT(&free_list
, &first_block
);
3455 * Allocate an inode for the symlink.
3457 error
= xfs_dir_ialloc(&tp
, dp
, S_IFLNK
| (vap
->va_mode
&~S_IFMT
),
3458 1, 0, credp
, prid
, resblks
> 0, &ip
, NULL
);
3460 if (error
== ENOSPC
)
3467 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
3468 dp_joined_to_trans
= B_TRUE
;
3471 * Also attach the dquot(s) to it, if applicable.
3473 XFS_QM_DQVOPCREATE(mp
, tp
, ip
, udqp
, gdqp
);
3476 resblks
-= XFS_IALLOC_SPACE_RES(mp
);
3478 * If the symlink will fit into the inode, write it inline.
3480 if (pathlen
<= XFS_IFORK_DSIZE(ip
)) {
3481 xfs_idata_realloc(ip
, pathlen
, XFS_DATA_FORK
);
3482 memcpy(ip
->i_df
.if_u1
.if_data
, target_path
, pathlen
);
3483 ip
->i_d
.di_size
= pathlen
;
3486 * The inode was initially created in extent format.
3488 ip
->i_df
.if_flags
&= ~(XFS_IFEXTENTS
| XFS_IFBROOT
);
3489 ip
->i_df
.if_flags
|= XFS_IFINLINE
;
3491 ip
->i_d
.di_format
= XFS_DINODE_FMT_LOCAL
;
3492 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_DDATA
| XFS_ILOG_CORE
);
3496 nmaps
= SYMLINK_MAPS
;
3498 error
= xfs_bmapi(tp
, ip
, first_fsb
, fs_blocks
,
3499 XFS_BMAPI_WRITE
| XFS_BMAPI_METADATA
,
3500 &first_block
, resblks
, mval
, &nmaps
,
3507 resblks
-= fs_blocks
;
3508 ip
->i_d
.di_size
= pathlen
;
3509 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3511 cur_chunk
= target_path
;
3512 for (n
= 0; n
< nmaps
; n
++) {
3513 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
3514 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
3515 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
3516 BTOBB(byte_cnt
), 0);
3517 ASSERT(bp
&& !XFS_BUF_GETERROR(bp
));
3518 if (pathlen
< byte_cnt
) {
3521 pathlen
-= byte_cnt
;
3523 memcpy(XFS_BUF_PTR(bp
), cur_chunk
, byte_cnt
);
3524 cur_chunk
+= byte_cnt
;
3526 xfs_trans_log_buf(tp
, bp
, 0, byte_cnt
- 1);
3531 * Create the directory entry for the symlink.
3533 error
= xfs_dir_createname(tp
, dp
, link_name
, link_namelen
, ip
->i_ino
,
3534 &first_block
, &free_list
, resblks
);
3537 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3538 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
3541 * Bump the in memory version number of the parent directory
3542 * so that other processes accessing it will recognize that
3543 * the directory has changed.
3548 * If this is a synchronous mount, make sure that the
3549 * symlink transaction goes to disk before returning to
3552 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
3553 xfs_trans_set_sync(tp
);
3557 * xfs_trans_commit normally decrements the vnode ref count
3558 * when it unlocks the inode. Since we want to return the
3559 * vnode to the caller, we bump the vnode ref count now.
3563 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
3567 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
3568 XFS_QM_DQRELE(mp
, udqp
);
3569 XFS_QM_DQRELE(mp
, gdqp
);
3571 /* Fall through to std_return with error = 0 or errno from
3572 * xfs_trans_commit */
3574 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, XFS_BHVTOI(dir_bdp
),
3575 DM_EVENT_POSTSYMLINK
)) {
3576 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTSYMLINK
,
3577 dir_vp
, DM_RIGHT_NULL
,
3578 error
? NULL
: XFS_ITOV(ip
),
3579 DM_RIGHT_NULL
, link_name
, target_path
,
3595 xfs_bmap_cancel(&free_list
);
3596 cancel_flags
|= XFS_TRANS_ABORT
;
3598 xfs_trans_cancel(tp
, cancel_flags
);
3599 XFS_QM_DQRELE(mp
, udqp
);
3600 XFS_QM_DQRELE(mp
, gdqp
);
3602 if (!dp_joined_to_trans
&& (dp
!= NULL
)) {
3603 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
3613 * A fid routine that takes a pointer to a previously allocated
3614 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3624 vn_trace_entry(BHV_TO_VNODE(bdp
), __FUNCTION__
,
3625 (inst_t
*)__return_address
);
3626 ASSERT(sizeof(fid_t
) >= sizeof(xfs_fid2_t
));
3628 xfid
= (xfs_fid2_t
*)fidp
;
3629 ip
= XFS_BHVTOI(bdp
);
3630 xfid
->fid_len
= sizeof(xfs_fid2_t
) - sizeof(xfid
->fid_len
);
3633 * use memcpy because the inode is a long long and there's no
3634 * assurance that xfid->fid_ino is properly aligned.
3636 memcpy(&xfid
->fid_ino
, &ip
->i_ino
, sizeof(xfid
->fid_ino
));
3637 xfid
->fid_gen
= ip
->i_d
.di_gen
;
3649 bhv_vrwlock_t locktype
)
3654 vp
= BHV_TO_VNODE(bdp
);
3657 ip
= XFS_BHVTOI(bdp
);
3658 if (locktype
== VRWLOCK_WRITE
) {
3659 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
3660 } else if (locktype
== VRWLOCK_TRY_READ
) {
3661 return xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
);
3662 } else if (locktype
== VRWLOCK_TRY_WRITE
) {
3663 return xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
);
3665 ASSERT((locktype
== VRWLOCK_READ
) ||
3666 (locktype
== VRWLOCK_WRITE_DIRECT
));
3667 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
3680 bhv_vrwlock_t locktype
)
3685 vp
= BHV_TO_VNODE(bdp
);
3688 ip
= XFS_BHVTOI(bdp
);
3689 if (locktype
== VRWLOCK_WRITE
) {
3691 * In the write case, we may have added a new entry to
3692 * the reference cache. This might store a pointer to
3693 * an inode to be released in this inode. If it is there,
3694 * clear the pointer and release the inode after unlocking
3697 xfs_refcache_iunlock(ip
, XFS_IOLOCK_EXCL
);
3699 ASSERT((locktype
== VRWLOCK_READ
) ||
3700 (locktype
== VRWLOCK_WRITE_DIRECT
));
3701 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
3713 xfs_inode_log_item_t
*iip
;
3716 ip
= XFS_BHVTOI(bdp
);
3720 if (XFS_FORCED_SHUTDOWN(mp
))
3721 return XFS_ERROR(EIO
);
3724 * Bypass inodes which have already been cleaned by
3725 * the inode flush clustering code inside xfs_iflush
3727 if ((ip
->i_update_core
== 0) &&
3728 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)))
3731 if (flags
& FLUSH_LOG
) {
3732 if (iip
&& iip
->ili_last_lsn
) {
3733 xlog_t
*log
= mp
->m_log
;
3735 int s
, log_flags
= XFS_LOG_FORCE
;
3737 s
= GRANT_LOCK(log
);
3738 sync_lsn
= log
->l_last_sync_lsn
;
3739 GRANT_UNLOCK(log
, s
);
3741 if ((XFS_LSN_CMP(iip
->ili_last_lsn
, sync_lsn
) <= 0))
3744 if (flags
& FLUSH_SYNC
)
3745 log_flags
|= XFS_LOG_SYNC
;
3746 return xfs_log_force(mp
, iip
->ili_last_lsn
, log_flags
);
3751 * We make this non-blocking if the inode is contended,
3752 * return EAGAIN to indicate to the caller that they
3753 * did not succeed. This prevents the flush path from
3754 * blocking on inodes inside another operation right
3755 * now, they get caught later by xfs_sync.
3757 if (flags
& FLUSH_INODE
) {
3760 if (xfs_ipincount(ip
))
3763 if (flags
& FLUSH_SYNC
) {
3764 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
3766 } else if (xfs_ilock_nowait(ip
, XFS_ILOCK_SHARED
)) {
3767 if (xfs_ipincount(ip
) || !xfs_iflock_nowait(ip
)) {
3768 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3775 if (flags
& FLUSH_SYNC
)
3776 flush_flags
= XFS_IFLUSH_SYNC
;
3778 flush_flags
= XFS_IFLUSH_ASYNC
;
3780 error
= xfs_iflush(ip
, flush_flags
);
3781 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3799 if (!capable(CAP_SYS_ADMIN
))
3800 return XFS_ERROR(EPERM
);
3802 ip
= XFS_BHVTOI(bdp
);
3805 if (XFS_FORCED_SHUTDOWN(mp
))
3806 return XFS_ERROR(EIO
);
3808 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SET_DMATTRS
);
3809 error
= xfs_trans_reserve(tp
, 0, XFS_ICHANGE_LOG_RES (mp
), 0, 0, 0);
3811 xfs_trans_cancel(tp
, 0);
3814 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3815 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
3817 ip
->i_iocore
.io_dmevmask
= ip
->i_d
.di_dmevmask
= evmask
;
3818 ip
->i_iocore
.io_dmstate
= ip
->i_d
.di_dmstate
= state
;
3820 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3822 error
= xfs_trans_commit(tp
, 0);
3834 vp
= BHV_TO_VNODE(bdp
);
3835 ip
= XFS_BHVTOI(bdp
);
3837 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
3839 ASSERT(!VN_MAPPED(vp
));
3841 /* bad inode, get out here ASAP */
3849 ASSERT(XFS_FORCED_SHUTDOWN(ip
->i_mount
) || ip
->i_delayed_blks
== 0);
3852 * Make sure the atime in the XFS inode is correct before freeing the
3855 xfs_synchronize_atime(ip
);
3858 * If we have nothing to flush with this inode then complete the
3859 * teardown now, otherwise break the link between the xfs inode and the
3860 * linux inode and clean up the xfs inode later. This avoids flushing
3861 * the inode to disk during the delete operation itself.
3863 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3864 * first to ensure that xfs_iunpin() will never see an xfs inode
3865 * that has a linux inode being reclaimed. Synchronisation is provided
3866 * by the i_flags_lock.
3868 if (!ip
->i_update_core
&& (ip
->i_itemp
== NULL
)) {
3869 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3871 return xfs_finish_reclaim(ip
, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC
);
3873 xfs_mount_t
*mp
= ip
->i_mount
;
3875 /* Protect sync and unpin from us */
3876 XFS_MOUNT_ILOCK(mp
);
3877 spin_lock(&ip
->i_flags_lock
);
3878 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
3879 vn_bhv_remove(VN_BHV_HEAD(vp
), XFS_ITOBHV(ip
));
3880 spin_unlock(&ip
->i_flags_lock
);
3881 list_add_tail(&ip
->i_reclaim
, &mp
->m_del_inodes
);
3882 XFS_MOUNT_IUNLOCK(mp
);
3893 xfs_ihash_t
*ih
= ip
->i_hash
;
3894 bhv_vnode_t
*vp
= XFS_ITOV_NULL(ip
);
3897 if (vp
&& VN_BAD(vp
))
3900 /* The hash lock here protects a thread in xfs_iget_core from
3901 * racing with us on linking the inode back with a vnode.
3902 * Once we have the XFS_IRECLAIM flag set it will not touch
3905 write_lock(&ih
->ih_lock
);
3906 spin_lock(&ip
->i_flags_lock
);
3907 if (__xfs_iflags_test(ip
, XFS_IRECLAIM
) ||
3908 (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) && vp
== NULL
)) {
3909 spin_unlock(&ip
->i_flags_lock
);
3910 write_unlock(&ih
->ih_lock
);
3913 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3917 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
3918 spin_unlock(&ip
->i_flags_lock
);
3919 write_unlock(&ih
->ih_lock
);
3922 * If the inode is still dirty, then flush it out. If the inode
3923 * is not in the AIL, then it will be OK to flush it delwri as
3924 * long as xfs_iflush() does not keep any references to the inode.
3925 * We leave that decision up to xfs_iflush() since it has the
3926 * knowledge of whether it's OK to simply do a delwri flush of
3927 * the inode or whether we need to wait until the inode is
3928 * pulled from the AIL.
3929 * We get the flush lock regardless, though, just to make sure
3930 * we don't free it while it is being flushed.
3932 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
3934 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3938 if (ip
->i_update_core
||
3939 ((ip
->i_itemp
!= NULL
) &&
3940 (ip
->i_itemp
->ili_format
.ilf_fields
!= 0))) {
3941 error
= xfs_iflush(ip
, sync_mode
);
3943 * If we hit an error, typically because of filesystem
3944 * shutdown, we don't need to let vn_reclaim to know
3945 * because we're gonna reclaim the inode anyway.
3948 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3951 xfs_iflock(ip
); /* synchronize with xfs_iflush_done */
3954 ASSERT(ip
->i_update_core
== 0);
3955 ASSERT(ip
->i_itemp
== NULL
||
3956 ip
->i_itemp
->ili_format
.ilf_fields
== 0);
3957 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3958 } else if (locked
) {
3960 * We are not interested in doing an iflush if we're
3961 * in the process of shutting down the filesystem forcibly.
3962 * So, just reclaim the inode.
3965 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3974 xfs_finish_reclaim_all(xfs_mount_t
*mp
, int noblock
)
3977 xfs_inode_t
*ip
, *n
;
3982 XFS_MOUNT_ILOCK(mp
);
3983 list_for_each_entry_safe(ip
, n
, &mp
->m_del_inodes
, i_reclaim
) {
3985 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
) == 0)
3987 if (xfs_ipincount(ip
) ||
3988 !xfs_iflock_nowait(ip
)) {
3989 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3993 XFS_MOUNT_IUNLOCK(mp
);
3994 if (xfs_finish_reclaim(ip
, noblock
,
3995 XFS_IFLUSH_DELWRI_ELSE_ASYNC
))
4004 XFS_MOUNT_IUNLOCK(mp
);
4009 * xfs_alloc_file_space()
4010 * This routine allocates disk space for the given file.
4012 * If alloc_type == 0, this request is for an ALLOCSP type
4013 * request which will change the file size. In this case, no
4014 * DMAPI event will be generated by the call. A TRUNCATE event
4015 * will be generated later by xfs_setattr.
4017 * If alloc_type != 0, this request is for a RESVSP type
4018 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
4019 * lower block boundary byte address is less than the file's
4028 xfs_alloc_file_space(
4035 xfs_mount_t
*mp
= ip
->i_mount
;
4037 xfs_filblks_t allocated_fsb
;
4038 xfs_filblks_t allocatesize_fsb
;
4039 xfs_extlen_t extsz
, temp
;
4040 xfs_fileoff_t startoffset_fsb
;
4041 xfs_fsblock_t firstfsb
;
4047 xfs_bmbt_irec_t imaps
[1], *imapp
;
4048 xfs_bmap_free_t free_list
;
4049 uint qblocks
, resblks
, resrtextents
;
4053 vn_trace_entry(XFS_ITOV(ip
), __FUNCTION__
, (inst_t
*)__return_address
);
4055 if (XFS_FORCED_SHUTDOWN(mp
))
4056 return XFS_ERROR(EIO
);
4058 rt
= XFS_IS_REALTIME_INODE(ip
);
4060 if (!(extsz
= ip
->i_d
.di_extsize
))
4061 extsz
= mp
->m_sb
.sb_rextsize
;
4063 extsz
= ip
->i_d
.di_extsize
;
4066 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
4070 return XFS_ERROR(EINVAL
);
4076 bmapi_flag
= XFS_BMAPI_WRITE
| (alloc_type
? XFS_BMAPI_PREALLOC
: 0);
4077 startoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
4078 allocatesize_fsb
= XFS_B_TO_FSB(mp
, count
);
4080 /* Generate a DMAPI event if needed. */
4081 if (alloc_type
!= 0 && offset
< ip
->i_size
&&
4082 (attr_flags
&ATTR_DMI
) == 0 &&
4083 DM_EVENT_ENABLED(XFS_MTOVFS(mp
), ip
, DM_EVENT_WRITE
)) {
4084 xfs_off_t end_dmi_offset
;
4086 end_dmi_offset
= offset
+len
;
4087 if (end_dmi_offset
> ip
->i_size
)
4088 end_dmi_offset
= ip
->i_size
;
4089 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, XFS_ITOV(ip
),
4090 offset
, end_dmi_offset
- offset
,
4097 * Allocate file space until done or until there is an error
4100 while (allocatesize_fsb
&& !error
) {
4104 * Determine space reservations for data/realtime.
4106 if (unlikely(extsz
)) {
4107 s
= startoffset_fsb
;
4110 e
= startoffset_fsb
+ allocatesize_fsb
;
4111 if ((temp
= do_mod(startoffset_fsb
, extsz
)))
4113 if ((temp
= do_mod(e
, extsz
)))
4117 e
= allocatesize_fsb
;
4121 resrtextents
= qblocks
= (uint
)(e
- s
);
4122 resrtextents
/= mp
->m_sb
.sb_rextsize
;
4123 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
4124 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
4127 resblks
= qblocks
= \
4128 XFS_DIOSTRAT_SPACE_RES(mp
, (uint
)(e
- s
));
4129 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
4133 * Allocate and setup the transaction.
4135 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
4136 error
= xfs_trans_reserve(tp
, resblks
,
4137 XFS_WRITE_LOG_RES(mp
), resrtextents
,
4138 XFS_TRANS_PERM_LOG_RES
,
4139 XFS_WRITE_LOG_COUNT
);
4141 * Check for running out of space
4145 * Free the transaction structure.
4147 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
4148 xfs_trans_cancel(tp
, 0);
4151 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4152 error
= XFS_TRANS_RESERVE_QUOTA_NBLKS(mp
, tp
, ip
,
4153 qblocks
, 0, quota_flag
);
4157 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
4158 xfs_trans_ihold(tp
, ip
);
4161 * Issue the xfs_bmapi() call to allocate the blocks
4163 XFS_BMAP_INIT(&free_list
, &firstfsb
);
4164 error
= XFS_BMAPI(mp
, tp
, &ip
->i_iocore
, startoffset_fsb
,
4165 allocatesize_fsb
, bmapi_flag
,
4166 &firstfsb
, 0, imapp
, &nimaps
,
4173 * Complete the transaction
4175 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
4180 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
4181 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4186 allocated_fsb
= imapp
->br_blockcount
;
4189 error
= XFS_ERROR(ENOSPC
);
4193 startoffset_fsb
+= allocated_fsb
;
4194 allocatesize_fsb
-= allocated_fsb
;
4197 if (error
== ENOSPC
&& (attr_flags
&ATTR_DMI
) == 0 &&
4198 DM_EVENT_ENABLED(XFS_MTOVFS(mp
), ip
, DM_EVENT_NOSPACE
)) {
4200 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_NOSPACE
,
4201 XFS_ITOV(ip
), DM_RIGHT_NULL
,
4202 XFS_ITOV(ip
), DM_RIGHT_NULL
,
4203 NULL
, NULL
, 0, 0, 0); /* Delay flag intentionally unused */
4205 goto retry
; /* Maybe DMAPI app. has made space */
4206 /* else fall through with error from XFS_SEND_DATA */
4211 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4212 xfs_bmap_cancel(&free_list
);
4213 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp
, tp
, ip
, qblocks
, 0, quota_flag
);
4215 error1
: /* Just cancel transaction */
4216 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
4217 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4218 goto dmapi_enospc_check
;
4222 * Zero file bytes between startoff and endoff inclusive.
4223 * The iolock is held exclusive and no blocks are buffered.
4226 xfs_zero_remaining_bytes(
4231 xfs_bmbt_irec_t imap
;
4232 xfs_fileoff_t offset_fsb
;
4233 xfs_off_t lastoffset
;
4236 xfs_mount_t
*mp
= ip
->i_mount
;
4240 bp
= xfs_buf_get_noaddr(mp
->m_sb
.sb_blocksize
,
4241 ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
?
4242 mp
->m_rtdev_targp
: mp
->m_ddev_targp
);
4244 for (offset
= startoff
; offset
<= endoff
; offset
= lastoffset
+ 1) {
4245 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
4247 error
= XFS_BMAPI(mp
, NULL
, &ip
->i_iocore
, offset_fsb
, 1, 0,
4248 NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
4249 if (error
|| nimap
< 1)
4251 ASSERT(imap
.br_blockcount
>= 1);
4252 ASSERT(imap
.br_startoff
== offset_fsb
);
4253 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+ 1) - 1;
4254 if (lastoffset
> endoff
)
4255 lastoffset
= endoff
;
4256 if (imap
.br_startblock
== HOLESTARTBLOCK
)
4258 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
4259 if (imap
.br_state
== XFS_EXT_UNWRITTEN
)
4262 XFS_BUF_UNWRITE(bp
);
4264 XFS_BUF_SET_ADDR(bp
, XFS_FSB_TO_DB(ip
, imap
.br_startblock
));
4266 if ((error
= xfs_iowait(bp
))) {
4267 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4268 mp
, bp
, XFS_BUF_ADDR(bp
));
4271 memset(XFS_BUF_PTR(bp
) +
4272 (offset
- XFS_FSB_TO_B(mp
, imap
.br_startoff
)),
4273 0, lastoffset
- offset
+ 1);
4278 if ((error
= xfs_iowait(bp
))) {
4279 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4280 mp
, bp
, XFS_BUF_ADDR(bp
));
4289 * xfs_free_file_space()
4290 * This routine frees disk space for the given file.
4292 * This routine is only called by xfs_change_file_space
4293 * for an UNRESVSP type call.
4301 xfs_free_file_space(
4310 xfs_off_t end_dmi_offset
;
4311 xfs_fileoff_t endoffset_fsb
;
4313 xfs_fsblock_t firstfsb
;
4314 xfs_bmap_free_t free_list
;
4315 xfs_bmbt_irec_t imap
;
4323 xfs_fileoff_t startoffset_fsb
;
4325 int need_iolock
= 1;
4330 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
4332 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
4336 if (len
<= 0) /* if nothing being freed */
4338 rt
= (ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
);
4339 startoffset_fsb
= XFS_B_TO_FSB(mp
, offset
);
4340 end_dmi_offset
= offset
+ len
;
4341 endoffset_fsb
= XFS_B_TO_FSBT(mp
, end_dmi_offset
);
4343 if (offset
< ip
->i_size
&&
4344 (attr_flags
& ATTR_DMI
) == 0 &&
4345 DM_EVENT_ENABLED(XFS_MTOVFS(mp
), ip
, DM_EVENT_WRITE
)) {
4346 if (end_dmi_offset
> ip
->i_size
)
4347 end_dmi_offset
= ip
->i_size
;
4348 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, vp
,
4349 offset
, end_dmi_offset
- offset
,
4350 AT_DELAY_FLAG(attr_flags
), NULL
);
4355 if (attr_flags
& ATTR_NOLOCK
)
4358 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
4359 vn_iowait(vp
); /* wait for the completion of any pending DIOs */
4362 rounding
= max_t(uint
, 1 << mp
->m_sb
.sb_blocklog
, NBPP
);
4363 ioffset
= offset
& ~(rounding
- 1);
4365 if (VN_CACHED(vp
) != 0) {
4366 xfs_inval_cached_trace(&ip
->i_iocore
, ioffset
, -1,
4367 ctooff(offtoct(ioffset
)), -1);
4368 error
= bhv_vop_flushinval_pages(vp
, ctooff(offtoct(ioffset
)),
4369 -1, FI_REMAPF_LOCKED
);
4371 goto out_unlock_iolock
;
4375 * Need to zero the stuff we're not freeing, on disk.
4376 * If its a realtime file & can't use unwritten extents then we
4377 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4378 * will take care of it for us.
4380 if (rt
&& !XFS_SB_VERSION_HASEXTFLGBIT(&mp
->m_sb
)) {
4382 error
= XFS_BMAPI(mp
, NULL
, &ip
->i_iocore
, startoffset_fsb
,
4383 1, 0, NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
4385 goto out_unlock_iolock
;
4386 ASSERT(nimap
== 0 || nimap
== 1);
4387 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
4390 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
4391 block
= imap
.br_startblock
;
4392 mod
= do_div(block
, mp
->m_sb
.sb_rextsize
);
4394 startoffset_fsb
+= mp
->m_sb
.sb_rextsize
- mod
;
4397 error
= XFS_BMAPI(mp
, NULL
, &ip
->i_iocore
, endoffset_fsb
- 1,
4398 1, 0, NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
4400 goto out_unlock_iolock
;
4401 ASSERT(nimap
== 0 || nimap
== 1);
4402 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
4403 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
4405 if (mod
&& (mod
!= mp
->m_sb
.sb_rextsize
))
4406 endoffset_fsb
-= mod
;
4409 if ((done
= (endoffset_fsb
<= startoffset_fsb
)))
4411 * One contiguous piece to clear
4413 error
= xfs_zero_remaining_bytes(ip
, offset
, offset
+ len
- 1);
4416 * Some full blocks, possibly two pieces to clear
4418 if (offset
< XFS_FSB_TO_B(mp
, startoffset_fsb
))
4419 error
= xfs_zero_remaining_bytes(ip
, offset
,
4420 XFS_FSB_TO_B(mp
, startoffset_fsb
) - 1);
4422 XFS_FSB_TO_B(mp
, endoffset_fsb
) < offset
+ len
)
4423 error
= xfs_zero_remaining_bytes(ip
,
4424 XFS_FSB_TO_B(mp
, endoffset_fsb
),
4429 * free file space until done or until there is an error
4431 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
4432 while (!error
&& !done
) {
4435 * allocate and setup the transaction
4437 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
4438 error
= xfs_trans_reserve(tp
,
4440 XFS_WRITE_LOG_RES(mp
),
4442 XFS_TRANS_PERM_LOG_RES
,
4443 XFS_WRITE_LOG_COUNT
);
4446 * check for running out of space
4450 * Free the transaction structure.
4452 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
4453 xfs_trans_cancel(tp
, 0);
4456 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4457 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
,
4458 ip
->i_udquot
, ip
->i_gdquot
, resblks
, 0,
4459 XFS_QMOPT_RES_REGBLKS
);
4463 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
4464 xfs_trans_ihold(tp
, ip
);
4467 * issue the bunmapi() call to free the blocks
4469 XFS_BMAP_INIT(&free_list
, &firstfsb
);
4470 error
= XFS_BUNMAPI(mp
, tp
, &ip
->i_iocore
, startoffset_fsb
,
4471 endoffset_fsb
- startoffset_fsb
,
4472 0, 2, &firstfsb
, &free_list
, NULL
, &done
);
4478 * complete the transaction
4480 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
4485 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
4486 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4491 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
4495 xfs_bmap_cancel(&free_list
);
4497 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
4498 xfs_iunlock(ip
, need_iolock
? (XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
) :
4504 * xfs_change_file_space()
4505 * This routine allocates or frees disk space for the given file.
4506 * The user specified parameters are checked for alignment and size
4515 xfs_change_file_space(
4529 xfs_off_t startoffset
;
4535 vp
= BHV_TO_VNODE(bdp
);
4536 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
4538 ip
= XFS_BHVTOI(bdp
);
4542 * must be a regular file and have write permission
4545 return XFS_ERROR(EINVAL
);
4547 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
4549 if ((error
= xfs_iaccess(ip
, S_IWUSR
, credp
))) {
4550 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
4554 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
4556 switch (bf
->l_whence
) {
4557 case 0: /*SEEK_SET*/
4559 case 1: /*SEEK_CUR*/
4560 bf
->l_start
+= offset
;
4562 case 2: /*SEEK_END*/
4563 bf
->l_start
+= ip
->i_size
;
4566 return XFS_ERROR(EINVAL
);
4569 llen
= bf
->l_len
> 0 ? bf
->l_len
- 1 : bf
->l_len
;
4571 if ( (bf
->l_start
< 0)
4572 || (bf
->l_start
> XFS_MAXIOFFSET(mp
))
4573 || (bf
->l_start
+ llen
< 0)
4574 || (bf
->l_start
+ llen
> XFS_MAXIOFFSET(mp
)))
4575 return XFS_ERROR(EINVAL
);
4579 startoffset
= bf
->l_start
;
4583 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4585 * These calls do NOT zero the data space allocated to the file,
4586 * nor do they change the file size.
4588 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4590 * These calls cause the new file data to be zeroed and the file
4591 * size to be changed.
4593 setprealloc
= clrprealloc
= 0;
4596 case XFS_IOC_RESVSP
:
4597 case XFS_IOC_RESVSP64
:
4598 error
= xfs_alloc_file_space(ip
, startoffset
, bf
->l_len
,
4605 case XFS_IOC_UNRESVSP
:
4606 case XFS_IOC_UNRESVSP64
:
4607 if ((error
= xfs_free_file_space(ip
, startoffset
, bf
->l_len
,
4612 case XFS_IOC_ALLOCSP
:
4613 case XFS_IOC_ALLOCSP64
:
4614 case XFS_IOC_FREESP
:
4615 case XFS_IOC_FREESP64
:
4616 if (startoffset
> fsize
) {
4617 error
= xfs_alloc_file_space(ip
, fsize
,
4618 startoffset
- fsize
, 0, attr_flags
);
4623 va
.va_mask
= XFS_AT_SIZE
;
4624 va
.va_size
= startoffset
;
4626 error
= xfs_setattr(bdp
, &va
, attr_flags
, credp
);
4636 return XFS_ERROR(EINVAL
);
4640 * update the inode timestamp, mode, and prealloc flag bits
4642 tp
= xfs_trans_alloc(mp
, XFS_TRANS_WRITEID
);
4644 if ((error
= xfs_trans_reserve(tp
, 0, XFS_WRITEID_LOG_RES(mp
),
4647 xfs_trans_cancel(tp
, 0);
4651 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4653 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
4654 xfs_trans_ihold(tp
, ip
);
4656 if ((attr_flags
& ATTR_DMI
) == 0) {
4657 ip
->i_d
.di_mode
&= ~S_ISUID
;
4660 * Note that we don't have to worry about mandatory
4661 * file locking being disabled here because we only
4662 * clear the S_ISGID bit if the Group execute bit is
4663 * on, but if it was on then mandatory locking wouldn't
4664 * have been enabled.
4666 if (ip
->i_d
.di_mode
& S_IXGRP
)
4667 ip
->i_d
.di_mode
&= ~S_ISGID
;
4669 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
4672 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
4673 else if (clrprealloc
)
4674 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
4676 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
4677 xfs_trans_set_sync(tp
);
4679 error
= xfs_trans_commit(tp
, 0);
4681 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4686 bhv_vnodeops_t xfs_vnodeops
= {
4687 BHV_IDENTITY_INIT(VN_BHV_XFS
,VNODE_POSITION_XFS
),
4688 .vop_open
= xfs_open
,
4689 .vop_close
= xfs_close
,
4690 .vop_read
= xfs_read
,
4692 .vop_splice_read
= xfs_splice_read
,
4693 .vop_splice_write
= xfs_splice_write
,
4695 .vop_write
= xfs_write
,
4696 .vop_ioctl
= xfs_ioctl
,
4697 .vop_getattr
= xfs_getattr
,
4698 .vop_setattr
= xfs_setattr
,
4699 .vop_access
= xfs_access
,
4700 .vop_lookup
= xfs_lookup
,
4701 .vop_create
= xfs_create
,
4702 .vop_remove
= xfs_remove
,
4703 .vop_link
= xfs_link
,
4704 .vop_rename
= xfs_rename
,
4705 .vop_mkdir
= xfs_mkdir
,
4706 .vop_rmdir
= xfs_rmdir
,
4707 .vop_readdir
= xfs_readdir
,
4708 .vop_symlink
= xfs_symlink
,
4709 .vop_readlink
= xfs_readlink
,
4710 .vop_fsync
= xfs_fsync
,
4711 .vop_inactive
= xfs_inactive
,
4712 .vop_fid2
= xfs_fid2
,
4713 .vop_rwlock
= xfs_rwlock
,
4714 .vop_rwunlock
= xfs_rwunlock
,
4715 .vop_bmap
= xfs_bmap
,
4716 .vop_reclaim
= xfs_reclaim
,
4717 .vop_attr_get
= xfs_attr_get
,
4718 .vop_attr_set
= xfs_attr_set
,
4719 .vop_attr_remove
= xfs_attr_remove
,
4720 .vop_attr_list
= xfs_attr_list
,
4721 .vop_link_removed
= (vop_link_removed_t
)fs_noval
,
4722 .vop_vnode_change
= (vop_vnode_change_t
)fs_noval
,
4723 .vop_tosspages
= fs_tosspages
,
4724 .vop_flushinval_pages
= fs_flushinval_pages
,
4725 .vop_flush_pages
= fs_flush_pages
,
4726 .vop_release
= xfs_release
,
4727 .vop_iflush
= xfs_inode_flush
,