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_trans_space.h"
52 #include "xfs_log_priv.h"
53 #include "xfs_filestream.h"
54 #include "xfs_vnodeops.h"
62 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
63 return XFS_ERROR(EIO
);
66 * If it's a directory with any blocks, read-ahead block 0
67 * as we're almost certain to have the next operation be a read there.
69 if (S_ISDIR(ip
->i_d
.di_mode
) && ip
->i_d
.di_nextents
> 0) {
70 mode
= xfs_ilock_map_shared(ip
);
71 if (ip
->i_d
.di_nextents
> 0)
72 (void)xfs_da_reada_buf(NULL
, ip
, 0, XFS_DATA_FORK
);
73 xfs_iunlock(ip
, mode
);
87 bhv_vnode_t
*vp
= XFS_ITOV(ip
);
88 xfs_mount_t
*mp
= ip
->i_mount
;
92 if (XFS_FORCED_SHUTDOWN(mp
))
93 return XFS_ERROR(EIO
);
95 if (!(flags
& ATTR_LAZY
))
96 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
98 vap
->va_size
= XFS_ISIZE(ip
);
99 if (vap
->va_mask
== XFS_AT_SIZE
)
103 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
104 vap
->va_nodeid
= ip
->i_ino
;
106 vap
->va_nodeid
+= mp
->m_inoadd
;
108 vap
->va_nlink
= ip
->i_d
.di_nlink
;
111 * Quick exit for non-stat callers
114 ~(XFS_AT_SIZE
|XFS_AT_FSID
|XFS_AT_NODEID
|
115 XFS_AT_NLINK
|XFS_AT_BLKSIZE
)) == 0)
119 * Copy from in-core inode.
121 vap
->va_mode
= ip
->i_d
.di_mode
;
122 vap
->va_uid
= ip
->i_d
.di_uid
;
123 vap
->va_gid
= ip
->i_d
.di_gid
;
124 vap
->va_projid
= ip
->i_d
.di_projid
;
127 * Check vnode type block/char vs. everything else.
129 switch (ip
->i_d
.di_mode
& S_IFMT
) {
132 vap
->va_rdev
= ip
->i_df
.if_u2
.if_rdev
;
133 vap
->va_blocksize
= BLKDEV_IOSIZE
;
138 if (!(XFS_IS_REALTIME_INODE(ip
))) {
139 vap
->va_blocksize
= xfs_preferred_iosize(mp
);
143 * If the file blocks are being allocated from a
144 * realtime partition, then return the inode's
145 * realtime extent size or the realtime volume's
149 xfs_get_extsz_hint(ip
) << mp
->m_sb
.sb_blocklog
;
154 vn_atime_to_timespec(vp
, &vap
->va_atime
);
155 vap
->va_mtime
.tv_sec
= ip
->i_d
.di_mtime
.t_sec
;
156 vap
->va_mtime
.tv_nsec
= ip
->i_d
.di_mtime
.t_nsec
;
157 vap
->va_ctime
.tv_sec
= ip
->i_d
.di_ctime
.t_sec
;
158 vap
->va_ctime
.tv_nsec
= ip
->i_d
.di_ctime
.t_nsec
;
161 * Exit for stat callers. See if any of the rest of the fields
162 * to be filled in are needed.
165 (XFS_AT_XFLAGS
|XFS_AT_EXTSIZE
|XFS_AT_NEXTENTS
|XFS_AT_ANEXTENTS
|
166 XFS_AT_GENCOUNT
|XFS_AT_VCODE
)) == 0)
170 * Convert di_flags to xflags.
172 vap
->va_xflags
= xfs_ip2xflags(ip
);
175 * Exit for inode revalidate. See if any of the rest of
176 * the fields to be filled in are needed.
179 (XFS_AT_EXTSIZE
|XFS_AT_NEXTENTS
|XFS_AT_ANEXTENTS
|
180 XFS_AT_GENCOUNT
|XFS_AT_VCODE
)) == 0)
183 vap
->va_extsize
= ip
->i_d
.di_extsize
<< mp
->m_sb
.sb_blocklog
;
185 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) ?
186 ip
->i_df
.if_bytes
/ sizeof(xfs_bmbt_rec_t
) :
190 (ip
->i_afp
->if_flags
& XFS_IFEXTENTS
) ?
191 ip
->i_afp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
) :
192 ip
->i_d
.di_anextents
;
194 vap
->va_anextents
= 0;
195 vap
->va_gen
= ip
->i_d
.di_gen
;
198 if (!(flags
& ATTR_LAZY
))
199 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
214 bhv_vnode_t
*vp
= XFS_ITOV(ip
);
215 xfs_mount_t
*mp
= ip
->i_mount
;
224 xfs_prid_t projid
=0, iprojid
=0;
225 int mandlock_before
, mandlock_after
;
226 struct xfs_dquot
*udqp
, *gdqp
, *olddquot1
, *olddquot2
;
230 xfs_itrace_entry(ip
);
232 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
233 return XFS_ERROR(EROFS
);
236 * Cannot set certain attributes.
239 if (mask
& XFS_AT_NOSET
) {
240 return XFS_ERROR(EINVAL
);
243 if (XFS_FORCED_SHUTDOWN(mp
))
244 return XFS_ERROR(EIO
);
247 * Timestamps do not need to be logged and hence do not
248 * need to be done within a transaction.
250 if (mask
& XFS_AT_UPDTIMES
) {
251 ASSERT((mask
& ~XFS_AT_UPDTIMES
) == 0);
252 timeflags
= ((mask
& XFS_AT_UPDATIME
) ? XFS_ICHGTIME_ACC
: 0) |
253 ((mask
& XFS_AT_UPDCTIME
) ? XFS_ICHGTIME_CHG
: 0) |
254 ((mask
& XFS_AT_UPDMTIME
) ? XFS_ICHGTIME_MOD
: 0);
255 xfs_ichgtime(ip
, timeflags
);
259 olddquot1
= olddquot2
= NULL
;
263 * If disk quotas is on, we make sure that the dquots do exist on disk,
264 * before we start any other transactions. Trying to do this later
265 * is messy. We don't care to take a readlock to look at the ids
266 * in inode here, because we can't hold it across the trans_reserve.
267 * If the IDs do change before we take the ilock, we're covered
268 * because the i_*dquot fields will get updated anyway.
270 if (XFS_IS_QUOTA_ON(mp
) &&
271 (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
))) {
274 if ((mask
& XFS_AT_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
276 qflags
|= XFS_QMOPT_UQUOTA
;
278 uid
= ip
->i_d
.di_uid
;
280 if ((mask
& XFS_AT_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
282 qflags
|= XFS_QMOPT_GQUOTA
;
284 gid
= ip
->i_d
.di_gid
;
286 if ((mask
& XFS_AT_PROJID
) && XFS_IS_PQUOTA_ON(mp
)) {
287 projid
= vap
->va_projid
;
288 qflags
|= XFS_QMOPT_PQUOTA
;
290 projid
= ip
->i_d
.di_projid
;
293 * We take a reference when we initialize udqp and gdqp,
294 * so it is important that we never blindly double trip on
295 * the same variable. See xfs_create() for an example.
297 ASSERT(udqp
== NULL
);
298 ASSERT(gdqp
== NULL
);
299 code
= XFS_QM_DQVOPALLOC(mp
, ip
, uid
, gid
, projid
, qflags
,
306 * For the other attributes, we acquire the inode lock and
307 * first do an error checking pass.
310 lock_flags
= XFS_ILOCK_EXCL
;
311 if (flags
& ATTR_NOLOCK
)
313 if (!(mask
& XFS_AT_SIZE
)) {
314 if ((mask
!= (XFS_AT_CTIME
|XFS_AT_ATIME
|XFS_AT_MTIME
)) ||
315 (mp
->m_flags
& XFS_MOUNT_WSYNC
)) {
316 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
318 if ((code
= xfs_trans_reserve(tp
, 0,
319 XFS_ICHANGE_LOG_RES(mp
), 0,
326 if (DM_EVENT_ENABLED(ip
, DM_EVENT_TRUNCATE
) &&
327 !(flags
& ATTR_DMI
)) {
328 int dmflags
= AT_DELAY_FLAG(flags
) | DM_SEM_FLAG_WR
;
329 code
= XFS_SEND_DATA(mp
, DM_EVENT_TRUNCATE
, vp
,
330 vap
->va_size
, 0, dmflags
, NULL
);
337 lock_flags
|= XFS_IOLOCK_EXCL
;
340 xfs_ilock(ip
, lock_flags
);
342 /* boolean: are we the file owner? */
343 file_owner
= (current_fsuid(credp
) == ip
->i_d
.di_uid
);
346 * Change various properties of a file.
347 * Only the owner or users with CAP_FOWNER
348 * capability may do these things.
351 (XFS_AT_MODE
|XFS_AT_XFLAGS
|XFS_AT_EXTSIZE
|XFS_AT_UID
|
352 XFS_AT_GID
|XFS_AT_PROJID
)) {
354 * CAP_FOWNER overrides the following restrictions:
356 * The user ID of the calling process must be equal
357 * to the file owner ID, except in cases where the
358 * CAP_FSETID capability is applicable.
360 if (!file_owner
&& !capable(CAP_FOWNER
)) {
361 code
= XFS_ERROR(EPERM
);
366 * CAP_FSETID overrides the following restrictions:
368 * The effective user ID of the calling process shall match
369 * the file owner when setting the set-user-ID and
370 * set-group-ID bits on that file.
372 * The effective group ID or one of the supplementary group
373 * IDs of the calling process shall match the group owner of
374 * the file when setting the set-group-ID bit on that file
376 if (mask
& XFS_AT_MODE
) {
379 if ((vap
->va_mode
& S_ISUID
) && !file_owner
)
381 if ((vap
->va_mode
& S_ISGID
) &&
382 !in_group_p((gid_t
)ip
->i_d
.di_gid
))
385 /* Linux allows this, Irix doesn't. */
386 if ((vap
->va_mode
& S_ISVTX
) && !VN_ISDIR(vp
))
389 if (m
&& !capable(CAP_FSETID
))
395 * Change file ownership. Must be the owner or privileged.
396 * If the system was configured with the "restricted_chown"
397 * option, the owner is not permitted to give away the file,
398 * and can change the group id only to a group of which he
399 * or she is a member.
401 if (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
)) {
403 * These IDs could have changed since we last looked at them.
404 * But, we're assured that if the ownership did change
405 * while we didn't have the inode locked, inode's dquot(s)
406 * would have changed also.
408 iuid
= ip
->i_d
.di_uid
;
409 iprojid
= ip
->i_d
.di_projid
;
410 igid
= ip
->i_d
.di_gid
;
411 gid
= (mask
& XFS_AT_GID
) ? vap
->va_gid
: igid
;
412 uid
= (mask
& XFS_AT_UID
) ? vap
->va_uid
: iuid
;
413 projid
= (mask
& XFS_AT_PROJID
) ? (xfs_prid_t
)vap
->va_projid
:
417 * CAP_CHOWN overrides the following restrictions:
419 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
420 * shall override the restriction that a process cannot
421 * change the user ID of a file it owns and the restriction
422 * that the group ID supplied to the chown() function
423 * shall be equal to either the group ID or one of the
424 * supplementary group IDs of the calling process.
426 if (restricted_chown
&&
427 (iuid
!= uid
|| (igid
!= gid
&&
428 !in_group_p((gid_t
)gid
))) &&
429 !capable(CAP_CHOWN
)) {
430 code
= XFS_ERROR(EPERM
);
434 * Do a quota reservation only if uid/projid/gid is actually
437 if ((XFS_IS_UQUOTA_ON(mp
) && iuid
!= uid
) ||
438 (XFS_IS_PQUOTA_ON(mp
) && iprojid
!= projid
) ||
439 (XFS_IS_GQUOTA_ON(mp
) && igid
!= gid
)) {
441 code
= XFS_QM_DQVOPCHOWNRESV(mp
, tp
, ip
, udqp
, gdqp
,
442 capable(CAP_FOWNER
) ?
443 XFS_QMOPT_FORCE_RES
: 0);
444 if (code
) /* out of quota */
450 * Truncate file. Must have write permission and not be a directory.
452 if (mask
& XFS_AT_SIZE
) {
453 /* Short circuit the truncate case for zero length files */
454 if ((vap
->va_size
== 0) &&
455 (ip
->i_size
== 0) && (ip
->i_d
.di_nextents
== 0)) {
456 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
457 lock_flags
&= ~XFS_ILOCK_EXCL
;
458 if (mask
& XFS_AT_CTIME
)
459 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
465 code
= XFS_ERROR(EISDIR
);
467 } else if (!VN_ISREG(vp
)) {
468 code
= XFS_ERROR(EINVAL
);
472 * Make sure that the dquots are attached to the inode.
474 if ((code
= XFS_QM_DQATTACH(mp
, ip
, XFS_QMOPT_ILOCKED
)))
479 * Change file access or modified times.
481 if (mask
& (XFS_AT_ATIME
|XFS_AT_MTIME
)) {
483 if ((flags
& ATTR_UTIME
) &&
484 !capable(CAP_FOWNER
)) {
485 code
= XFS_ERROR(EPERM
);
492 * Change extent size or realtime flag.
494 if (mask
& (XFS_AT_EXTSIZE
|XFS_AT_XFLAGS
)) {
496 * Can't change extent size if any extents are allocated.
498 if (ip
->i_d
.di_nextents
&& (mask
& XFS_AT_EXTSIZE
) &&
499 ((ip
->i_d
.di_extsize
<< mp
->m_sb
.sb_blocklog
) !=
501 code
= XFS_ERROR(EINVAL
); /* EFBIG? */
506 * Can't change realtime flag if any extents are allocated.
508 if ((ip
->i_d
.di_nextents
|| ip
->i_delayed_blks
) &&
509 (mask
& XFS_AT_XFLAGS
) &&
510 (XFS_IS_REALTIME_INODE(ip
)) !=
511 (vap
->va_xflags
& XFS_XFLAG_REALTIME
)) {
512 code
= XFS_ERROR(EINVAL
); /* EFBIG? */
516 * Extent size must be a multiple of the appropriate block
517 * size, if set at all.
519 if ((mask
& XFS_AT_EXTSIZE
) && vap
->va_extsize
!= 0) {
522 if (XFS_IS_REALTIME_INODE(ip
) ||
523 ((mask
& XFS_AT_XFLAGS
) &&
524 (vap
->va_xflags
& XFS_XFLAG_REALTIME
))) {
525 size
= mp
->m_sb
.sb_rextsize
<<
526 mp
->m_sb
.sb_blocklog
;
528 size
= mp
->m_sb
.sb_blocksize
;
530 if (vap
->va_extsize
% size
) {
531 code
= XFS_ERROR(EINVAL
);
536 * If realtime flag is set then must have realtime data.
538 if ((mask
& XFS_AT_XFLAGS
) &&
539 (vap
->va_xflags
& XFS_XFLAG_REALTIME
)) {
540 if ((mp
->m_sb
.sb_rblocks
== 0) ||
541 (mp
->m_sb
.sb_rextsize
== 0) ||
542 (ip
->i_d
.di_extsize
% mp
->m_sb
.sb_rextsize
)) {
543 code
= XFS_ERROR(EINVAL
);
549 * Can't modify an immutable/append-only file unless
550 * we have appropriate permission.
552 if ((mask
& XFS_AT_XFLAGS
) &&
554 (XFS_DIFLAG_IMMUTABLE
|XFS_DIFLAG_APPEND
) ||
556 (XFS_XFLAG_IMMUTABLE
| XFS_XFLAG_APPEND
))) &&
557 !capable(CAP_LINUX_IMMUTABLE
)) {
558 code
= XFS_ERROR(EPERM
);
564 * Now we can make the changes. Before we join the inode
565 * to the transaction, if XFS_AT_SIZE is set then take care of
566 * the part of the truncation that must be done without the
567 * inode lock. This needs to be done before joining the inode
568 * to the transaction, because the inode cannot be unlocked
569 * once it is a part of the transaction.
571 if (mask
& XFS_AT_SIZE
) {
573 if ((vap
->va_size
> ip
->i_size
) &&
574 (flags
& ATTR_NOSIZETOK
) == 0) {
575 code
= xfs_igrow_start(ip
, vap
->va_size
, credp
);
577 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
580 * We are going to log the inode size change in this
581 * transaction so any previous writes that are beyond the on
582 * disk EOF and the new EOF that have not been written out need
583 * to be written here. If we do not write the data out, we
584 * expose ourselves to the null files problem.
586 * Only flush from the on disk size to the smaller of the in
587 * memory file size or the new size as that's the range we
588 * really care about here and prevents waiting for other data
589 * not within the range we care about here.
592 (ip
->i_size
!= ip
->i_d
.di_size
) &&
593 (vap
->va_size
> ip
->i_d
.di_size
)) {
594 code
= xfs_flush_pages(ip
,
595 ip
->i_d
.di_size
, vap
->va_size
,
596 XFS_B_ASYNC
, FI_NONE
);
599 /* wait for all I/O to complete */
603 code
= xfs_itruncate_data(ip
, vap
->va_size
);
606 lock_flags
&= ~XFS_ILOCK_EXCL
;
607 ASSERT(lock_flags
== XFS_IOLOCK_EXCL
);
610 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
611 if ((code
= xfs_trans_reserve(tp
, 0,
612 XFS_ITRUNCATE_LOG_RES(mp
), 0,
613 XFS_TRANS_PERM_LOG_RES
,
614 XFS_ITRUNCATE_LOG_COUNT
))) {
615 xfs_trans_cancel(tp
, 0);
617 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
620 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
621 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
625 xfs_trans_ijoin(tp
, ip
, lock_flags
);
626 xfs_trans_ihold(tp
, ip
);
629 /* determine whether mandatory locking mode changes */
630 mandlock_before
= MANDLOCK(vp
, ip
->i_d
.di_mode
);
633 * Truncate file. Must have write permission and not be a directory.
635 if (mask
& XFS_AT_SIZE
) {
636 if (vap
->va_size
> ip
->i_size
) {
637 xfs_igrow_finish(tp
, ip
, vap
->va_size
,
638 !(flags
& ATTR_DMI
));
639 } else if ((vap
->va_size
<= ip
->i_size
) ||
640 ((vap
->va_size
== 0) && ip
->i_d
.di_nextents
)) {
642 * signal a sync transaction unless
643 * we're truncating an already unlinked
644 * file on a wsync filesystem
646 code
= xfs_itruncate_finish(&tp
, ip
,
647 (xfs_fsize_t
)vap
->va_size
,
649 ((ip
->i_d
.di_nlink
!= 0 ||
650 !(mp
->m_flags
& XFS_MOUNT_WSYNC
))
655 * Truncated "down", so we're removing references
656 * to old data here - if we now delay flushing for
657 * a long time, we expose ourselves unduly to the
658 * notorious NULL files problem. So, we mark this
659 * vnode and flush it when the file is closed, and
660 * do not wait the usual (long) time for writeout.
662 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
665 * Have to do this even if the file's size doesn't change.
667 timeflags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
671 * Change file access modes.
673 if (mask
& XFS_AT_MODE
) {
674 ip
->i_d
.di_mode
&= S_IFMT
;
675 ip
->i_d
.di_mode
|= vap
->va_mode
& ~S_IFMT
;
677 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
678 timeflags
|= XFS_ICHGTIME_CHG
;
682 * Change file ownership. Must be the owner or privileged.
683 * If the system was configured with the "restricted_chown"
684 * option, the owner is not permitted to give away the file,
685 * and can change the group id only to a group of which he
686 * or she is a member.
688 if (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
)) {
690 * CAP_FSETID overrides the following restrictions:
692 * The set-user-ID and set-group-ID bits of a file will be
693 * cleared upon successful return from chown()
695 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
696 !capable(CAP_FSETID
)) {
697 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
701 * Change the ownerships and register quota modifications
702 * in the transaction.
705 if (XFS_IS_UQUOTA_ON(mp
)) {
706 ASSERT(mask
& XFS_AT_UID
);
708 olddquot1
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
709 &ip
->i_udquot
, udqp
);
711 ip
->i_d
.di_uid
= uid
;
714 if (XFS_IS_GQUOTA_ON(mp
)) {
715 ASSERT(!XFS_IS_PQUOTA_ON(mp
));
716 ASSERT(mask
& XFS_AT_GID
);
718 olddquot2
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
719 &ip
->i_gdquot
, gdqp
);
721 ip
->i_d
.di_gid
= gid
;
723 if (iprojid
!= projid
) {
724 if (XFS_IS_PQUOTA_ON(mp
)) {
725 ASSERT(!XFS_IS_GQUOTA_ON(mp
));
726 ASSERT(mask
& XFS_AT_PROJID
);
728 olddquot2
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
729 &ip
->i_gdquot
, gdqp
);
731 ip
->i_d
.di_projid
= projid
;
733 * We may have to rev the inode as well as
734 * the superblock version number since projids didn't
735 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
737 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
)
738 xfs_bump_ino_vers2(tp
, ip
);
741 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
742 timeflags
|= XFS_ICHGTIME_CHG
;
747 * Change file access or modified times.
749 if (mask
& (XFS_AT_ATIME
|XFS_AT_MTIME
)) {
750 if (mask
& XFS_AT_ATIME
) {
751 ip
->i_d
.di_atime
.t_sec
= vap
->va_atime
.tv_sec
;
752 ip
->i_d
.di_atime
.t_nsec
= vap
->va_atime
.tv_nsec
;
753 ip
->i_update_core
= 1;
754 timeflags
&= ~XFS_ICHGTIME_ACC
;
756 if (mask
& XFS_AT_MTIME
) {
757 ip
->i_d
.di_mtime
.t_sec
= vap
->va_mtime
.tv_sec
;
758 ip
->i_d
.di_mtime
.t_nsec
= vap
->va_mtime
.tv_nsec
;
759 timeflags
&= ~XFS_ICHGTIME_MOD
;
760 timeflags
|= XFS_ICHGTIME_CHG
;
762 if (tp
&& (flags
& ATTR_UTIME
))
763 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
767 * Change XFS-added attributes.
769 if (mask
& (XFS_AT_EXTSIZE
|XFS_AT_XFLAGS
)) {
770 if (mask
& XFS_AT_EXTSIZE
) {
772 * Converting bytes to fs blocks.
774 ip
->i_d
.di_extsize
= vap
->va_extsize
>>
775 mp
->m_sb
.sb_blocklog
;
777 if (mask
& XFS_AT_XFLAGS
) {
780 /* can't set PREALLOC this way, just preserve it */
781 di_flags
= (ip
->i_d
.di_flags
& XFS_DIFLAG_PREALLOC
);
782 if (vap
->va_xflags
& XFS_XFLAG_IMMUTABLE
)
783 di_flags
|= XFS_DIFLAG_IMMUTABLE
;
784 if (vap
->va_xflags
& XFS_XFLAG_APPEND
)
785 di_flags
|= XFS_DIFLAG_APPEND
;
786 if (vap
->va_xflags
& XFS_XFLAG_SYNC
)
787 di_flags
|= XFS_DIFLAG_SYNC
;
788 if (vap
->va_xflags
& XFS_XFLAG_NOATIME
)
789 di_flags
|= XFS_DIFLAG_NOATIME
;
790 if (vap
->va_xflags
& XFS_XFLAG_NODUMP
)
791 di_flags
|= XFS_DIFLAG_NODUMP
;
792 if (vap
->va_xflags
& XFS_XFLAG_PROJINHERIT
)
793 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
794 if (vap
->va_xflags
& XFS_XFLAG_NODEFRAG
)
795 di_flags
|= XFS_DIFLAG_NODEFRAG
;
796 if (vap
->va_xflags
& XFS_XFLAG_FILESTREAM
)
797 di_flags
|= XFS_DIFLAG_FILESTREAM
;
798 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
799 if (vap
->va_xflags
& XFS_XFLAG_RTINHERIT
)
800 di_flags
|= XFS_DIFLAG_RTINHERIT
;
801 if (vap
->va_xflags
& XFS_XFLAG_NOSYMLINKS
)
802 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
803 if (vap
->va_xflags
& XFS_XFLAG_EXTSZINHERIT
)
804 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
805 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
806 if (vap
->va_xflags
& XFS_XFLAG_REALTIME
)
807 di_flags
|= XFS_DIFLAG_REALTIME
;
808 if (vap
->va_xflags
& XFS_XFLAG_EXTSIZE
)
809 di_flags
|= XFS_DIFLAG_EXTSIZE
;
811 ip
->i_d
.di_flags
= di_flags
;
813 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
814 timeflags
|= XFS_ICHGTIME_CHG
;
818 * Change file inode change time only if XFS_AT_CTIME set
819 * AND we have been called by a DMI function.
822 if ( (flags
& ATTR_DMI
) && (mask
& XFS_AT_CTIME
) ) {
823 ip
->i_d
.di_ctime
.t_sec
= vap
->va_ctime
.tv_sec
;
824 ip
->i_d
.di_ctime
.t_nsec
= vap
->va_ctime
.tv_nsec
;
825 ip
->i_update_core
= 1;
826 timeflags
&= ~XFS_ICHGTIME_CHG
;
830 * Send out timestamp changes that need to be set to the
831 * current time. Not done when called by a DMI function.
833 if (timeflags
&& !(flags
& ATTR_DMI
))
834 xfs_ichgtime(ip
, timeflags
);
836 XFS_STATS_INC(xs_ig_attrchg
);
839 * If this is a synchronous mount, make sure that the
840 * transaction goes to disk before returning to the user.
841 * This is slightly sub-optimal in that truncates require
842 * two sync transactions instead of one for wsync filesystems.
843 * One for the truncate and one for the timestamps since we
844 * don't want to change the timestamps unless we're sure the
845 * truncate worked. Truncates are less than 1% of the laddis
846 * mix so this probably isn't worth the trouble to optimize.
850 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
851 xfs_trans_set_sync(tp
);
853 code
= xfs_trans_commit(tp
, commit_flags
);
857 * If the (regular) file's mandatory locking mode changed, then
858 * notify the vnode. We do this under the inode lock to prevent
859 * racing calls to vop_vnode_change.
861 mandlock_after
= MANDLOCK(vp
, ip
->i_d
.di_mode
);
863 xfs_iunlock(ip
, lock_flags
);
866 * Release any dquot(s) the inode had kept before chown.
868 XFS_QM_DQRELE(mp
, olddquot1
);
869 XFS_QM_DQRELE(mp
, olddquot2
);
870 XFS_QM_DQRELE(mp
, udqp
);
871 XFS_QM_DQRELE(mp
, gdqp
);
877 if (DM_EVENT_ENABLED(ip
, DM_EVENT_ATTRIBUTE
) &&
878 !(flags
& ATTR_DMI
)) {
879 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_ATTRIBUTE
, vp
, DM_RIGHT_NULL
,
880 NULL
, DM_RIGHT_NULL
, NULL
, NULL
,
881 0, 0, AT_DELAY_FLAG(flags
));
886 commit_flags
|= XFS_TRANS_ABORT
;
889 XFS_QM_DQRELE(mp
, udqp
);
890 XFS_QM_DQRELE(mp
, gdqp
);
892 xfs_trans_cancel(tp
, commit_flags
);
894 if (lock_flags
!= 0) {
895 xfs_iunlock(ip
, lock_flags
);
901 * The maximum pathlen is 1024 bytes. Since the minimum file system
902 * blocksize is 512 bytes, we can get a max of 2 extents back from
905 #define SYMLINK_MAPS 2
912 xfs_mount_t
*mp
= ip
->i_mount
;
913 int pathlen
= ip
->i_d
.di_size
;
914 int nmaps
= SYMLINK_MAPS
;
915 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
922 error
= xfs_bmapi(NULL
, ip
, 0, XFS_B_TO_FSB(mp
, pathlen
), 0, NULL
, 0,
923 mval
, &nmaps
, NULL
, NULL
);
927 for (n
= 0; n
< nmaps
; n
++) {
928 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
929 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
931 bp
= xfs_buf_read(mp
->m_ddev_targp
, d
, BTOBB(byte_cnt
), 0);
932 error
= XFS_BUF_GETERROR(bp
);
934 xfs_ioerror_alert("xfs_readlink",
935 ip
->i_mount
, bp
, XFS_BUF_ADDR(bp
));
939 if (pathlen
< byte_cnt
)
943 memcpy(link
, XFS_BUF_PTR(bp
), byte_cnt
);
947 link
[ip
->i_d
.di_size
] = '\0';
959 xfs_mount_t
*mp
= ip
->i_mount
;
963 xfs_itrace_entry(ip
);
965 if (XFS_FORCED_SHUTDOWN(mp
))
966 return XFS_ERROR(EIO
);
968 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
970 ASSERT((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
);
971 ASSERT(ip
->i_d
.di_size
<= MAXPATHLEN
);
973 pathlen
= ip
->i_d
.di_size
;
977 if (ip
->i_df
.if_flags
& XFS_IFINLINE
) {
978 memcpy(link
, ip
->i_df
.if_u1
.if_data
, pathlen
);
979 link
[pathlen
] = '\0';
981 error
= xfs_readlink_bmap(ip
, link
);
985 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
992 * This is called to sync the inode and its data out to disk.
993 * We need to hold the I/O lock while flushing the data, and
994 * the inode lock while flushing the inode. The inode lock CANNOT
995 * be held while flushing the data, so acquire after we're done
1007 int log_flushed
= 0, changed
= 1;
1009 xfs_itrace_entry(ip
);
1011 ASSERT(start
>= 0 && stop
>= -1);
1013 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
1014 return XFS_ERROR(EIO
);
1016 if (flag
& FSYNC_DATA
)
1017 filemap_fdatawait(vn_to_inode(XFS_ITOV(ip
))->i_mapping
);
1020 * We always need to make sure that the required inode state
1021 * is safe on disk. The vnode might be clean but because
1022 * of committed transactions that haven't hit the disk yet.
1023 * Likewise, there could be unflushed non-transactional
1024 * changes to the inode core that have to go to disk.
1026 * The following code depends on one assumption: that
1027 * any transaction that changes an inode logs the core
1028 * because it has to change some field in the inode core
1029 * (typically nextents or nblocks). That assumption
1030 * implies that any transactions against an inode will
1031 * catch any non-transactional updates. If inode-altering
1032 * transactions exist that violate this assumption, the
1033 * code breaks. Right now, it figures that if the involved
1034 * update_* field is clear and the inode is unpinned, the
1035 * inode is clean. Either it's been flushed or it's been
1036 * committed and the commit has hit the disk unpinning the inode.
1037 * (Note that xfs_inode_item_format() called at commit clears
1038 * the update_* fields.)
1040 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1042 /* If we are flushing data then we care about update_size
1043 * being set, otherwise we care about update_core
1045 if ((flag
& FSYNC_DATA
) ?
1046 (ip
->i_update_size
== 0) :
1047 (ip
->i_update_core
== 0)) {
1049 * Timestamps/size haven't changed since last inode
1050 * flush or inode transaction commit. That means
1051 * either nothing got written or a transaction
1052 * committed which caught the updates. If the
1053 * latter happened and the transaction hasn't
1054 * hit the disk yet, the inode will be still
1055 * be pinned. If it is, force the log.
1058 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1060 if (xfs_ipincount(ip
)) {
1061 _xfs_log_force(ip
->i_mount
, (xfs_lsn_t
)0,
1063 ((flag
& FSYNC_WAIT
)
1064 ? XFS_LOG_SYNC
: 0),
1068 * If the inode is not pinned and nothing
1069 * has changed we don't need to flush the
1077 * Kick off a transaction to log the inode
1078 * core to get the updates. Make it
1079 * sync if FSYNC_WAIT is passed in (which
1080 * is done by everybody but specfs). The
1081 * sync transaction will also force the log.
1083 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1084 tp
= xfs_trans_alloc(ip
->i_mount
, XFS_TRANS_FSYNC_TS
);
1085 if ((error
= xfs_trans_reserve(tp
, 0,
1086 XFS_FSYNC_TS_LOG_RES(ip
->i_mount
),
1088 xfs_trans_cancel(tp
, 0);
1091 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1094 * Note - it's possible that we might have pushed
1095 * ourselves out of the way during trans_reserve
1096 * which would flush the inode. But there's no
1097 * guarantee that the inode buffer has actually
1098 * gone out yet (it's delwri). Plus the buffer
1099 * could be pinned anyway if it's part of an
1100 * inode in another recent transaction. So we
1101 * play it safe and fire off the transaction anyway.
1103 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1104 xfs_trans_ihold(tp
, ip
);
1105 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1106 if (flag
& FSYNC_WAIT
)
1107 xfs_trans_set_sync(tp
);
1108 error
= _xfs_trans_commit(tp
, 0, &log_flushed
);
1110 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1113 if ((ip
->i_mount
->m_flags
& XFS_MOUNT_BARRIER
) && changed
) {
1115 * If the log write didn't issue an ordered tag we need
1116 * to flush the disk cache for the data device now.
1119 xfs_blkdev_issue_flush(ip
->i_mount
->m_ddev_targp
);
1122 * If this inode is on the RT dev we need to flush that
1125 if (XFS_IS_REALTIME_INODE(ip
))
1126 xfs_blkdev_issue_flush(ip
->i_mount
->m_rtdev_targp
);
1133 * This is called by xfs_inactive to free any blocks beyond eof
1134 * when the link count isn't zero and by xfs_dm_punch_hole() when
1135 * punching a hole to EOF.
1145 xfs_fileoff_t end_fsb
;
1146 xfs_fileoff_t last_fsb
;
1147 xfs_filblks_t map_len
;
1149 xfs_bmbt_irec_t imap
;
1150 int use_iolock
= (flags
& XFS_FREE_EOF_LOCK
);
1153 * Figure out if there are any blocks beyond the end
1154 * of the file. If not, then there is nothing to do.
1156 end_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)ip
->i_size
));
1157 last_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1158 map_len
= last_fsb
- end_fsb
;
1163 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1164 error
= xfs_bmapi(NULL
, ip
, end_fsb
, map_len
, 0,
1165 NULL
, 0, &imap
, &nimaps
, NULL
, NULL
);
1166 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1168 if (!error
&& (nimaps
!= 0) &&
1169 (imap
.br_startblock
!= HOLESTARTBLOCK
||
1170 ip
->i_delayed_blks
)) {
1172 * Attach the dquots to the inode up front.
1174 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
1178 * There are blocks after the end of file.
1179 * Free them up now by truncating the file to
1182 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1185 * Do the xfs_itruncate_start() call before
1186 * reserving any log space because
1187 * itruncate_start will call into the buffer
1188 * cache and we can't
1189 * do that within a transaction.
1192 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1193 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
,
1196 xfs_trans_cancel(tp
, 0);
1198 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1202 error
= xfs_trans_reserve(tp
, 0,
1203 XFS_ITRUNCATE_LOG_RES(mp
),
1204 0, XFS_TRANS_PERM_LOG_RES
,
1205 XFS_ITRUNCATE_LOG_COUNT
);
1207 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1208 xfs_trans_cancel(tp
, 0);
1209 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1213 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1214 xfs_trans_ijoin(tp
, ip
,
1217 xfs_trans_ihold(tp
, ip
);
1219 error
= xfs_itruncate_finish(&tp
, ip
,
1224 * If we get an error at this point we
1225 * simply don't bother truncating the file.
1228 xfs_trans_cancel(tp
,
1229 (XFS_TRANS_RELEASE_LOG_RES
|
1232 error
= xfs_trans_commit(tp
,
1233 XFS_TRANS_RELEASE_LOG_RES
);
1235 xfs_iunlock(ip
, (use_iolock
? (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)
1242 * Free a symlink that has blocks associated with it.
1245 xfs_inactive_symlink_rmt(
1253 xfs_fsblock_t first_block
;
1254 xfs_bmap_free_t free_list
;
1257 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
1265 ASSERT(ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
));
1267 * We're freeing a symlink that has some
1268 * blocks allocated to it. Free the
1269 * blocks here. We know that we've got
1270 * either 1 or 2 extents and that we can
1271 * free them all in one bunmapi call.
1273 ASSERT(ip
->i_d
.di_nextents
> 0 && ip
->i_d
.di_nextents
<= 2);
1274 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1275 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
1276 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1277 xfs_trans_cancel(tp
, 0);
1282 * Lock the inode, fix the size, and join it to the transaction.
1283 * Hold it so in the normal path, we still have it locked for
1284 * the second transaction. In the error paths we need it
1285 * held so the cancel won't rele it, see below.
1287 xfs_ilock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1288 size
= (int)ip
->i_d
.di_size
;
1289 ip
->i_d
.di_size
= 0;
1290 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1291 xfs_trans_ihold(tp
, ip
);
1292 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1294 * Find the block(s) so we can inval and unmap them.
1297 XFS_BMAP_INIT(&free_list
, &first_block
);
1298 nmaps
= ARRAY_SIZE(mval
);
1299 if ((error
= xfs_bmapi(tp
, ip
, 0, XFS_B_TO_FSB(mp
, size
),
1300 XFS_BMAPI_METADATA
, &first_block
, 0, mval
, &nmaps
,
1304 * Invalidate the block(s).
1306 for (i
= 0; i
< nmaps
; i
++) {
1307 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
1308 XFS_FSB_TO_DADDR(mp
, mval
[i
].br_startblock
),
1309 XFS_FSB_TO_BB(mp
, mval
[i
].br_blockcount
), 0);
1310 xfs_trans_binval(tp
, bp
);
1313 * Unmap the dead block(s) to the free_list.
1315 if ((error
= xfs_bunmapi(tp
, ip
, 0, size
, XFS_BMAPI_METADATA
, nmaps
,
1316 &first_block
, &free_list
, NULL
, &done
)))
1320 * Commit the first transaction. This logs the EFI and the inode.
1322 if ((error
= xfs_bmap_finish(&tp
, &free_list
, &committed
)))
1325 * The transaction must have been committed, since there were
1326 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1327 * The new tp has the extent freeing and EFDs.
1331 * The first xact was committed, so add the inode to the new one.
1332 * Mark it dirty so it will be logged and moved forward in the log as
1333 * part of every commit.
1335 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1336 xfs_trans_ihold(tp
, ip
);
1337 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1339 * Get a new, empty transaction to return to our caller.
1341 ntp
= xfs_trans_dup(tp
);
1343 * Commit the transaction containing extent freeing and EFDs.
1344 * If we get an error on the commit here or on the reserve below,
1345 * we need to unlock the inode since the new transaction doesn't
1346 * have the inode attached.
1348 error
= xfs_trans_commit(tp
, 0);
1351 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1355 * Remove the memory for extent descriptions (just bookkeeping).
1357 if (ip
->i_df
.if_bytes
)
1358 xfs_idata_realloc(ip
, -ip
->i_df
.if_bytes
, XFS_DATA_FORK
);
1359 ASSERT(ip
->i_df
.if_bytes
== 0);
1361 * Put an itruncate log reservation in the new transaction
1364 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1365 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
1366 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1370 * Return with the inode locked but not joined to the transaction.
1376 xfs_bmap_cancel(&free_list
);
1379 * Have to come here with the inode locked and either
1380 * (held and in the transaction) or (not in the transaction).
1381 * If the inode isn't held then cancel would iput it, but
1382 * that's wrong since this is inactive and the vnode ref
1383 * count is 0 already.
1384 * Cancel won't do anything to the inode if held, but it still
1385 * needs to be locked until the cancel is done, if it was
1386 * joined to the transaction.
1388 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1389 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1396 xfs_inactive_symlink_local(
1402 ASSERT(ip
->i_d
.di_size
<= XFS_IFORK_DSIZE(ip
));
1404 * We're freeing a symlink which fit into
1405 * the inode. Just free the memory used
1406 * to hold the old symlink.
1408 error
= xfs_trans_reserve(*tpp
, 0,
1409 XFS_ITRUNCATE_LOG_RES(ip
->i_mount
),
1410 0, XFS_TRANS_PERM_LOG_RES
,
1411 XFS_ITRUNCATE_LOG_COUNT
);
1414 xfs_trans_cancel(*tpp
, 0);
1418 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1421 * Zero length symlinks _can_ exist.
1423 if (ip
->i_df
.if_bytes
> 0) {
1424 xfs_idata_realloc(ip
,
1425 -(ip
->i_df
.if_bytes
),
1427 ASSERT(ip
->i_df
.if_bytes
== 0);
1441 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
));
1444 ASSERT(ip
->i_d
.di_forkoff
!= 0);
1445 xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1446 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1448 error
= xfs_attr_inactive(ip
);
1451 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1452 return error
; /* goto out */
1455 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1456 error
= xfs_trans_reserve(tp
, 0,
1457 XFS_IFREE_LOG_RES(mp
),
1458 0, XFS_TRANS_PERM_LOG_RES
,
1459 XFS_INACTIVE_LOG_COUNT
);
1461 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1462 xfs_trans_cancel(tp
, 0);
1464 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1468 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1469 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1470 xfs_trans_ihold(tp
, ip
);
1471 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1473 ASSERT(ip
->i_d
.di_anextents
== 0);
1483 bhv_vnode_t
*vp
= XFS_ITOV(ip
);
1484 xfs_mount_t
*mp
= ip
->i_mount
;
1487 if (!VN_ISREG(vp
) || (ip
->i_d
.di_mode
== 0))
1490 /* If this is a read-only mount, don't do this (would generate I/O) */
1491 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1494 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1498 * If we are using filestreams, and we have an unlinked
1499 * file that we are processing the last close on, then nothing
1500 * will be able to reopen and write to this file. Purge this
1501 * inode from the filestreams cache so that it doesn't delay
1502 * teardown of the inode.
1504 if ((ip
->i_d
.di_nlink
== 0) && xfs_inode_is_filestream(ip
))
1505 xfs_filestream_deassociate(ip
);
1508 * If we previously truncated this file and removed old data
1509 * in the process, we want to initiate "early" writeout on
1510 * the last close. This is an attempt to combat the notorious
1511 * NULL files problem which is particularly noticable from a
1512 * truncate down, buffered (re-)write (delalloc), followed by
1513 * a crash. What we are effectively doing here is
1514 * significantly reducing the time window where we'd otherwise
1515 * be exposed to that problem.
1517 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1518 if (truncated
&& VN_DIRTY(vp
) && ip
->i_delayed_blks
> 0)
1519 xfs_flush_pages(ip
, 0, -1, XFS_B_ASYNC
, FI_NONE
);
1522 if (ip
->i_d
.di_nlink
!= 0) {
1523 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1524 ((ip
->i_size
> 0) || (VN_CACHED(vp
) > 0 ||
1525 ip
->i_delayed_blks
> 0)) &&
1526 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
)) &&
1527 (!(ip
->i_d
.di_flags
&
1528 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)))) {
1529 error
= xfs_free_eofblocks(mp
, ip
, XFS_FREE_EOF_LOCK
);
1541 * This is called when the vnode reference count for the vnode
1542 * goes to zero. If the file has been unlinked, then it must
1543 * now be truncated. Also, we clear all of the read-ahead state
1544 * kept for the inode here since the file is now closed.
1550 bhv_vnode_t
*vp
= XFS_ITOV(ip
);
1551 xfs_bmap_free_t free_list
;
1552 xfs_fsblock_t first_block
;
1559 xfs_itrace_entry(ip
);
1562 * If the inode is already free, then there can be nothing
1565 if (ip
->i_d
.di_mode
== 0 || VN_BAD(vp
)) {
1566 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1567 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1568 return VN_INACTIVE_CACHE
;
1572 * Only do a truncate if it's a regular file with
1573 * some actual space in it. It's OK to look at the
1574 * inode's fields without the lock because we're the
1575 * only one with a reference to the inode.
1577 truncate
= ((ip
->i_d
.di_nlink
== 0) &&
1578 ((ip
->i_d
.di_size
!= 0) || (ip
->i_size
!= 0) ||
1579 (ip
->i_d
.di_nextents
> 0) || (ip
->i_delayed_blks
> 0)) &&
1580 ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
));
1584 if (ip
->i_d
.di_nlink
== 0 && DM_EVENT_ENABLED(ip
, DM_EVENT_DESTROY
)) {
1585 (void) XFS_SEND_DESTROY(mp
, vp
, DM_RIGHT_NULL
);
1590 /* If this is a read-only mount, don't do this (would generate I/O) */
1591 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1594 if (ip
->i_d
.di_nlink
!= 0) {
1595 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1596 ((ip
->i_size
> 0) || (VN_CACHED(vp
) > 0 ||
1597 ip
->i_delayed_blks
> 0)) &&
1598 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) &&
1599 (!(ip
->i_d
.di_flags
&
1600 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)) ||
1601 (ip
->i_delayed_blks
!= 0)))) {
1602 error
= xfs_free_eofblocks(mp
, ip
, XFS_FREE_EOF_LOCK
);
1604 return VN_INACTIVE_CACHE
;
1609 ASSERT(ip
->i_d
.di_nlink
== 0);
1611 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
1612 return VN_INACTIVE_CACHE
;
1614 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1617 * Do the xfs_itruncate_start() call before
1618 * reserving any log space because itruncate_start
1619 * will call into the buffer cache and we can't
1620 * do that within a transaction.
1622 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1624 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
, 0);
1626 xfs_trans_cancel(tp
, 0);
1627 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1628 return VN_INACTIVE_CACHE
;
1631 error
= xfs_trans_reserve(tp
, 0,
1632 XFS_ITRUNCATE_LOG_RES(mp
),
1633 0, XFS_TRANS_PERM_LOG_RES
,
1634 XFS_ITRUNCATE_LOG_COUNT
);
1636 /* Don't call itruncate_cleanup */
1637 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1638 xfs_trans_cancel(tp
, 0);
1639 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1640 return VN_INACTIVE_CACHE
;
1643 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1644 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1645 xfs_trans_ihold(tp
, ip
);
1648 * normally, we have to run xfs_itruncate_finish sync.
1649 * But if filesystem is wsync and we're in the inactive
1650 * path, then we know that nlink == 0, and that the
1651 * xaction that made nlink == 0 is permanently committed
1652 * since xfs_remove runs as a synchronous transaction.
1654 error
= xfs_itruncate_finish(&tp
, ip
, 0, XFS_DATA_FORK
,
1655 (!(mp
->m_flags
& XFS_MOUNT_WSYNC
) ? 1 : 0));
1658 xfs_trans_cancel(tp
,
1659 XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1660 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1661 return VN_INACTIVE_CACHE
;
1663 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
) {
1666 * If we get an error while cleaning up a
1667 * symlink we bail out.
1669 error
= (ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
)) ?
1670 xfs_inactive_symlink_rmt(ip
, &tp
) :
1671 xfs_inactive_symlink_local(ip
, &tp
);
1675 return VN_INACTIVE_CACHE
;
1678 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1679 xfs_trans_ihold(tp
, ip
);
1681 error
= xfs_trans_reserve(tp
, 0,
1682 XFS_IFREE_LOG_RES(mp
),
1683 0, XFS_TRANS_PERM_LOG_RES
,
1684 XFS_INACTIVE_LOG_COUNT
);
1686 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1687 xfs_trans_cancel(tp
, 0);
1688 return VN_INACTIVE_CACHE
;
1691 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1692 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1693 xfs_trans_ihold(tp
, ip
);
1697 * If there are attributes associated with the file
1698 * then blow them away now. The code calls a routine
1699 * that recursively deconstructs the attribute fork.
1700 * We need to just commit the current transaction
1701 * because we can't use it for xfs_attr_inactive().
1703 if (ip
->i_d
.di_anextents
> 0) {
1704 error
= xfs_inactive_attrs(ip
, &tp
);
1706 * If we got an error, the transaction is already
1707 * cancelled, and the inode is unlocked. Just get out.
1710 return VN_INACTIVE_CACHE
;
1711 } else if (ip
->i_afp
) {
1712 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1718 XFS_BMAP_INIT(&free_list
, &first_block
);
1719 error
= xfs_ifree(tp
, ip
, &free_list
);
1722 * If we fail to free the inode, shut down. The cancel
1723 * might do that, we need to make sure. Otherwise the
1724 * inode might be lost for a long time or forever.
1726 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1728 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1729 error
, mp
->m_fsname
);
1730 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1732 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
1735 * Credit the quota account(s). The inode is gone.
1737 XFS_TRANS_MOD_DQUOT_BYINO(mp
, tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1740 * Just ignore errors at this point. There is
1741 * nothing we can do except to try to keep going.
1743 (void) xfs_bmap_finish(&tp
, &free_list
, &committed
);
1744 (void) xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1747 * Release the dquots held by inode, if any.
1749 XFS_QM_DQDETACH(mp
, ip
);
1751 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1754 return VN_INACTIVE_CACHE
;
1761 bhv_vname_t
*dentry
,
1769 xfs_itrace_entry(dp
);
1771 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
1772 return XFS_ERROR(EIO
);
1774 lock_mode
= xfs_ilock_map_shared(dp
);
1775 error
= xfs_dir_lookup_int(dp
, lock_mode
, dentry
, &e_inum
, &ip
);
1777 *vpp
= XFS_ITOV(ip
);
1780 xfs_iunlock_map_shared(dp
, lock_mode
);
1787 bhv_vname_t
*dentry
,
1793 char *name
= VNAME(dentry
);
1794 xfs_mount_t
*mp
= dp
->i_mount
;
1795 bhv_vnode_t
*dir_vp
= XFS_ITOV(dp
);
1797 bhv_vnode_t
*vp
= NULL
;
1800 xfs_bmap_free_t free_list
;
1801 xfs_fsblock_t first_block
;
1802 boolean_t unlock_dp_on_error
= B_FALSE
;
1803 int dm_event_sent
= 0;
1807 struct xfs_dquot
*udqp
, *gdqp
;
1812 xfs_itrace_entry(dp
);
1814 namelen
= VNAMELEN(dentry
);
1816 if (DM_EVENT_ENABLED(dp
, DM_EVENT_CREATE
)) {
1817 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_CREATE
,
1818 dir_vp
, DM_RIGHT_NULL
, NULL
,
1819 DM_RIGHT_NULL
, name
, NULL
,
1827 if (XFS_FORCED_SHUTDOWN(mp
))
1828 return XFS_ERROR(EIO
);
1830 /* Return through std_return after this point. */
1833 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1834 prid
= dp
->i_d
.di_projid
;
1836 prid
= (xfs_prid_t
)dfltprid
;
1839 * Make sure that we have allocated dquot(s) on disk.
1841 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
1842 current_fsuid(credp
), current_fsgid(credp
), prid
,
1843 XFS_QMOPT_QUOTALL
|XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
1849 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1850 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1851 resblks
= XFS_CREATE_SPACE_RES(mp
, namelen
);
1853 * Initially assume that the file does not exist and
1854 * reserve the resources for that case. If that is not
1855 * the case we'll drop the one we have and get a more
1856 * appropriate transaction later.
1858 error
= xfs_trans_reserve(tp
, resblks
, XFS_CREATE_LOG_RES(mp
), 0,
1859 XFS_TRANS_PERM_LOG_RES
, XFS_CREATE_LOG_COUNT
);
1860 if (error
== ENOSPC
) {
1862 error
= xfs_trans_reserve(tp
, 0, XFS_CREATE_LOG_RES(mp
), 0,
1863 XFS_TRANS_PERM_LOG_RES
, XFS_CREATE_LOG_COUNT
);
1870 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1871 unlock_dp_on_error
= B_TRUE
;
1873 XFS_BMAP_INIT(&free_list
, &first_block
);
1878 * Reserve disk quota and the inode.
1880 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
1884 if (resblks
== 0 && (error
= xfs_dir_canenter(tp
, dp
, name
, namelen
)))
1886 error
= xfs_dir_ialloc(&tp
, dp
, mode
, 1,
1887 rdev
, credp
, prid
, resblks
> 0,
1890 if (error
== ENOSPC
)
1897 * At this point, we've gotten a newly allocated inode.
1898 * It is locked (and joined to the transaction).
1901 ASSERT(ismrlocked (&ip
->i_lock
, MR_UPDATE
));
1904 * Now we join the directory inode to the transaction. We do not do it
1905 * earlier because xfs_dir_ialloc might commit the previous transaction
1906 * (and release all the locks). An error from here on will result in
1907 * the transaction cancel unlocking dp so don't do it explicitly in the
1911 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1912 unlock_dp_on_error
= B_FALSE
;
1914 error
= xfs_dir_createname(tp
, dp
, name
, namelen
, ip
->i_ino
,
1915 &first_block
, &free_list
, resblks
?
1916 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1918 ASSERT(error
!= ENOSPC
);
1921 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1922 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1925 * If this is a synchronous mount, make sure that the
1926 * create transaction goes to disk before returning to
1929 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
1930 xfs_trans_set_sync(tp
);
1936 * Attach the dquot(s) to the inodes and modify them incore.
1937 * These ids of the inode couldn't have changed since the new
1938 * inode has been locked ever since it was created.
1940 XFS_QM_DQVOPCREATE(mp
, tp
, ip
, udqp
, gdqp
);
1943 * xfs_trans_commit normally decrements the vnode ref count
1944 * when it unlocks the inode. Since we want to return the
1945 * vnode to the caller, we bump the vnode ref count now.
1950 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1952 xfs_bmap_cancel(&free_list
);
1956 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1963 XFS_QM_DQRELE(mp
, udqp
);
1964 XFS_QM_DQRELE(mp
, gdqp
);
1968 /* Fallthrough to std_return with error = 0 */
1971 if ((*vpp
|| (error
!= 0 && dm_event_sent
!= 0)) &&
1972 DM_EVENT_ENABLED(dp
, DM_EVENT_POSTCREATE
)) {
1973 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTCREATE
,
1974 dir_vp
, DM_RIGHT_NULL
,
1976 DM_RIGHT_NULL
, name
, NULL
,
1982 cancel_flags
|= XFS_TRANS_ABORT
;
1987 xfs_trans_cancel(tp
, cancel_flags
);
1989 XFS_QM_DQRELE(mp
, udqp
);
1990 XFS_QM_DQRELE(mp
, gdqp
);
1992 if (unlock_dp_on_error
)
1993 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1999 * Wait until after the current transaction is aborted to
2000 * release the inode. This prevents recursive transactions
2001 * and deadlocks from xfs_inactive.
2003 cancel_flags
|= XFS_TRANS_ABORT
;
2004 xfs_trans_cancel(tp
, cancel_flags
);
2007 XFS_QM_DQRELE(mp
, udqp
);
2008 XFS_QM_DQRELE(mp
, gdqp
);
2015 * Some counters to see if (and how often) we are hitting some deadlock
2016 * prevention code paths.
2020 int xfs_rm_lock_delays
;
2021 int xfs_rm_attempts
;
2025 * The following routine will lock the inodes associated with the
2026 * directory and the named entry in the directory. The locks are
2027 * acquired in increasing inode number.
2029 * If the entry is "..", then only the directory is locked. The
2030 * vnode ref count will still include that from the .. entry in
2033 * There is a deadlock we need to worry about. If the locked directory is
2034 * in the AIL, it might be blocking up the log. The next inode we lock
2035 * could be already locked by another thread waiting for log space (e.g
2036 * a permanent log reservation with a long running transaction (see
2037 * xfs_itruncate_finish)). To solve this, we must check if the directory
2038 * is in the ail and use lock_nowait. If we can't lock, we need to
2039 * drop the inode lock on the directory and try again. xfs_iunlock will
2040 * potentially push the tail if we were holding up the log.
2043 xfs_lock_dir_and_entry(
2045 xfs_inode_t
*ip
) /* inode of entry 'name' */
2049 xfs_inode_t
*ips
[2];
2058 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
2065 * We want to lock in increasing inum. Since we've already
2066 * acquired the lock on the directory, we may need to release
2067 * if if the inum of the entry turns out to be less.
2069 if (e_inum
> dp
->i_ino
) {
2071 * We are already in the right order, so just
2072 * lock on the inode of the entry.
2073 * We need to use nowait if dp is in the AIL.
2076 lp
= (xfs_log_item_t
*)dp
->i_itemp
;
2077 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
2078 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2085 * Unlock dp and try again.
2086 * xfs_iunlock will try to push the tail
2087 * if the inode is in the AIL.
2090 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2092 if ((attempts
% 5) == 0) {
2093 delay(1); /* Don't just spin the CPU */
2095 xfs_rm_lock_delays
++;
2101 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2103 } else if (e_inum
< dp
->i_ino
) {
2104 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2108 xfs_lock_inodes(ips
, 2, 0, XFS_ILOCK_EXCL
);
2110 /* else e_inum == dp->i_ino */
2111 /* This can happen if we're asked to lock /x/..
2112 * the entry is "..", which is also the parent directory.
2120 int xfs_small_retries
;
2121 int xfs_middle_retries
;
2122 int xfs_lots_retries
;
2123 int xfs_lock_delays
;
2127 * Bump the subclass so xfs_lock_inodes() acquires each lock with
2131 xfs_lock_inumorder(int lock_mode
, int subclass
)
2133 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
2134 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_IOLOCK_SHIFT
;
2135 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
))
2136 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_ILOCK_SHIFT
;
2142 * The following routine will lock n inodes in exclusive mode.
2143 * We assume the caller calls us with the inodes in i_ino order.
2145 * We need to detect deadlock where an inode that we lock
2146 * is in the AIL and we start waiting for another inode that is locked
2147 * by a thread in a long running transaction (such as truncate). This can
2148 * result in deadlock since the long running trans might need to wait
2149 * for the inode we just locked in order to push the tail and free space
2159 int attempts
= 0, i
, j
, try_lock
;
2162 ASSERT(ips
&& (inodes
>= 2)); /* we need at least two */
2173 for (; i
< inodes
; i
++) {
2176 if (i
&& (ips
[i
] == ips
[i
-1])) /* Already locked */
2180 * If try_lock is not set yet, make sure all locked inodes
2181 * are not in the AIL.
2182 * If any are, set try_lock to be used later.
2186 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
2187 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
2188 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
2195 * If any of the previous locks we have locked is in the AIL,
2196 * we must TRY to get the second and subsequent locks. If
2197 * we can't get any, we must release all we have
2202 /* try_lock must be 0 if i is 0. */
2204 * try_lock means we have an inode locked
2205 * that is in the AIL.
2208 if (!xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
))) {
2212 * Unlock all previous guys and try again.
2213 * xfs_iunlock will try to push the tail
2214 * if the inode is in the AIL.
2217 for(j
= i
- 1; j
>= 0; j
--) {
2220 * Check to see if we've already
2221 * unlocked this one.
2222 * Not the first one going back,
2223 * and the inode ptr is the same.
2225 if ((j
!= (i
- 1)) && ips
[j
] ==
2229 xfs_iunlock(ips
[j
], lock_mode
);
2232 if ((attempts
% 5) == 0) {
2233 delay(1); /* Don't just spin the CPU */
2243 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
2249 if (attempts
< 5) xfs_small_retries
++;
2250 else if (attempts
< 100) xfs_middle_retries
++;
2251 else xfs_lots_retries
++;
2259 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2260 int remove_which_error_return
= 0;
2262 #define REMOVE_DEBUG_TRACE(x)
2263 #endif /* ! DEBUG */
2268 bhv_vname_t
*dentry
)
2270 bhv_vnode_t
*dir_vp
= XFS_ITOV(dp
);
2271 char *name
= VNAME(dentry
);
2272 xfs_mount_t
*mp
= dp
->i_mount
;
2274 xfs_trans_t
*tp
= NULL
;
2276 xfs_bmap_free_t free_list
;
2277 xfs_fsblock_t first_block
;
2285 xfs_itrace_entry(dp
);
2287 if (XFS_FORCED_SHUTDOWN(mp
))
2288 return XFS_ERROR(EIO
);
2290 namelen
= VNAMELEN(dentry
);
2292 if (!xfs_get_dir_entry(dentry
, &ip
)) {
2293 dm_di_mode
= ip
->i_d
.di_mode
;
2297 if (DM_EVENT_ENABLED(dp
, DM_EVENT_REMOVE
)) {
2298 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_REMOVE
, dir_vp
,
2299 DM_RIGHT_NULL
, NULL
, DM_RIGHT_NULL
,
2300 name
, NULL
, dm_di_mode
, 0, 0);
2305 /* From this point on, return through std_return */
2309 * We need to get a reference to ip before we get our log
2310 * reservation. The reason for this is that we cannot call
2311 * xfs_iget for an inode for which we do not have a reference
2312 * once we've acquired a log reservation. This is because the
2313 * inode we are trying to get might be in xfs_inactive going
2314 * for a log reservation. Since we'll have to wait for the
2315 * inactive code to complete before returning from xfs_iget,
2316 * we need to make sure that we don't have log space reserved
2317 * when we call xfs_iget. Instead we get an unlocked reference
2318 * to the inode before getting our log reservation.
2320 error
= xfs_get_dir_entry(dentry
, &ip
);
2322 REMOVE_DEBUG_TRACE(__LINE__
);
2326 dm_di_mode
= ip
->i_d
.di_mode
;
2328 xfs_itrace_entry(ip
);
2331 error
= XFS_QM_DQATTACH(mp
, dp
, 0);
2332 if (!error
&& dp
!= ip
)
2333 error
= XFS_QM_DQATTACH(mp
, ip
, 0);
2335 REMOVE_DEBUG_TRACE(__LINE__
);
2340 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
2341 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2343 * We try to get the real space reservation first,
2344 * allowing for directory btree deletion(s) implying
2345 * possible bmap insert(s). If we can't get the space
2346 * reservation then we use 0 instead, and avoid the bmap
2347 * btree insert(s) in the directory code by, if the bmap
2348 * insert tries to happen, instead trimming the LAST
2349 * block from the directory.
2351 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2352 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
2353 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
2354 if (error
== ENOSPC
) {
2356 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
2357 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
2360 ASSERT(error
!= ENOSPC
);
2361 REMOVE_DEBUG_TRACE(__LINE__
);
2362 xfs_trans_cancel(tp
, 0);
2367 error
= xfs_lock_dir_and_entry(dp
, ip
);
2369 REMOVE_DEBUG_TRACE(__LINE__
);
2370 xfs_trans_cancel(tp
, cancel_flags
);
2376 * At this point, we've gotten both the directory and the entry
2379 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2382 * Increment vnode ref count only in this case since
2383 * there's an extra vnode reference in the case where
2387 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2391 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2393 XFS_BMAP_INIT(&free_list
, &first_block
);
2394 error
= xfs_dir_removename(tp
, dp
, name
, namelen
, ip
->i_ino
,
2395 &first_block
, &free_list
, 0);
2397 ASSERT(error
!= ENOENT
);
2398 REMOVE_DEBUG_TRACE(__LINE__
);
2401 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2404 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2406 error
= xfs_droplink(tp
, ip
);
2408 REMOVE_DEBUG_TRACE(__LINE__
);
2412 /* Determine if this is the last link while
2413 * we are in the transaction.
2415 link_zero
= (ip
)->i_d
.di_nlink
==0;
2418 * Take an extra ref on the inode so that it doesn't
2419 * go to xfs_inactive() from within the commit.
2424 * If this is a synchronous mount, make sure that the
2425 * remove transaction goes to disk before returning to
2428 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2429 xfs_trans_set_sync(tp
);
2432 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2434 REMOVE_DEBUG_TRACE(__LINE__
);
2438 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2445 * If we are using filestreams, kill the stream association.
2446 * If the file is still open it may get a new one but that
2447 * will get killed on last close in xfs_close() so we don't
2448 * have to worry about that.
2450 if (link_zero
&& xfs_inode_is_filestream(ip
))
2451 xfs_filestream_deassociate(ip
);
2453 xfs_itrace_exit(ip
);
2456 /* Fall through to std_return with error = 0 */
2458 if (DM_EVENT_ENABLED(dp
, DM_EVENT_POSTREMOVE
)) {
2459 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTREMOVE
,
2460 dir_vp
, DM_RIGHT_NULL
,
2461 NULL
, DM_RIGHT_NULL
,
2462 name
, NULL
, dm_di_mode
, error
, 0);
2467 xfs_bmap_cancel(&free_list
);
2468 cancel_flags
|= XFS_TRANS_ABORT
;
2469 xfs_trans_cancel(tp
, cancel_flags
);
2474 * In this case make sure to not release the inode until after
2475 * the current transaction is aborted. Releasing it beforehand
2476 * can cause us to go to xfs_inactive and start a recursive
2477 * transaction which can easily deadlock with the current one.
2479 xfs_bmap_cancel(&free_list
);
2480 cancel_flags
|= XFS_TRANS_ABORT
;
2481 xfs_trans_cancel(tp
, cancel_flags
);
2491 bhv_vnode_t
*src_vp
,
2492 bhv_vname_t
*dentry
)
2494 bhv_vnode_t
*target_dir_vp
= XFS_ITOV(tdp
);
2495 xfs_mount_t
*mp
= tdp
->i_mount
;
2496 xfs_inode_t
*sip
= xfs_vtoi(src_vp
);
2498 xfs_inode_t
*ips
[2];
2500 xfs_bmap_free_t free_list
;
2501 xfs_fsblock_t first_block
;
2505 char *target_name
= VNAME(dentry
);
2508 xfs_itrace_entry(tdp
);
2509 xfs_itrace_entry(xfs_vtoi(src_vp
));
2511 target_namelen
= VNAMELEN(dentry
);
2512 ASSERT(!VN_ISDIR(src_vp
));
2514 if (XFS_FORCED_SHUTDOWN(mp
))
2515 return XFS_ERROR(EIO
);
2517 if (DM_EVENT_ENABLED(tdp
, DM_EVENT_LINK
)) {
2518 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_LINK
,
2519 target_dir_vp
, DM_RIGHT_NULL
,
2520 src_vp
, DM_RIGHT_NULL
,
2521 target_name
, NULL
, 0, 0, 0);
2526 /* Return through std_return after this point. */
2528 error
= XFS_QM_DQATTACH(mp
, sip
, 0);
2529 if (!error
&& sip
!= tdp
)
2530 error
= XFS_QM_DQATTACH(mp
, tdp
, 0);
2534 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
2535 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2536 resblks
= XFS_LINK_SPACE_RES(mp
, target_namelen
);
2537 error
= xfs_trans_reserve(tp
, resblks
, XFS_LINK_LOG_RES(mp
), 0,
2538 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2539 if (error
== ENOSPC
) {
2541 error
= xfs_trans_reserve(tp
, 0, XFS_LINK_LOG_RES(mp
), 0,
2542 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2549 if (sip
->i_ino
< tdp
->i_ino
) {
2557 xfs_lock_inodes(ips
, 2, 0, XFS_ILOCK_EXCL
);
2560 * Increment vnode ref counts since xfs_trans_commit &
2561 * xfs_trans_cancel will both unlock the inodes and
2562 * decrement the associated ref counts.
2565 VN_HOLD(target_dir_vp
);
2566 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
2567 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
2570 * If the source has too many links, we can't make any more to it.
2572 if (sip
->i_d
.di_nlink
>= XFS_MAXLINK
) {
2573 error
= XFS_ERROR(EMLINK
);
2578 * If we are using project inheritance, we only allow hard link
2579 * creation in our tree when the project IDs are the same; else
2580 * the tree quota mechanism could be circumvented.
2582 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2583 (tdp
->i_d
.di_projid
!= sip
->i_d
.di_projid
))) {
2584 error
= XFS_ERROR(EXDEV
);
2589 (error
= xfs_dir_canenter(tp
, tdp
, target_name
, target_namelen
)))
2592 XFS_BMAP_INIT(&free_list
, &first_block
);
2594 error
= xfs_dir_createname(tp
, tdp
, target_name
, target_namelen
,
2595 sip
->i_ino
, &first_block
, &free_list
,
2599 xfs_ichgtime(tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2601 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
2603 error
= xfs_bumplink(tp
, sip
);
2608 * If this is a synchronous mount, make sure that the
2609 * link transaction goes to disk before returning to
2612 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2613 xfs_trans_set_sync(tp
);
2616 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
2618 xfs_bmap_cancel(&free_list
);
2622 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2626 /* Fall through to std_return with error = 0. */
2628 if (DM_EVENT_ENABLED(sip
, DM_EVENT_POSTLINK
)) {
2629 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTLINK
,
2630 target_dir_vp
, DM_RIGHT_NULL
,
2631 src_vp
, DM_RIGHT_NULL
,
2632 target_name
, NULL
, 0, error
, 0);
2637 cancel_flags
|= XFS_TRANS_ABORT
;
2641 xfs_trans_cancel(tp
, cancel_flags
);
2649 bhv_vname_t
*dentry
,
2654 bhv_vnode_t
*dir_vp
= XFS_ITOV(dp
);
2655 char *dir_name
= VNAME(dentry
);
2656 int dir_namelen
= VNAMELEN(dentry
);
2657 xfs_mount_t
*mp
= dp
->i_mount
;
2658 xfs_inode_t
*cdp
; /* inode of created dir */
2659 bhv_vnode_t
*cvp
; /* vnode of created dir */
2664 xfs_bmap_free_t free_list
;
2665 xfs_fsblock_t first_block
;
2666 boolean_t unlock_dp_on_error
= B_FALSE
;
2667 boolean_t created
= B_FALSE
;
2668 int dm_event_sent
= 0;
2670 struct xfs_dquot
*udqp
, *gdqp
;
2673 if (XFS_FORCED_SHUTDOWN(mp
))
2674 return XFS_ERROR(EIO
);
2678 if (DM_EVENT_ENABLED(dp
, DM_EVENT_CREATE
)) {
2679 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_CREATE
,
2680 dir_vp
, DM_RIGHT_NULL
, NULL
,
2681 DM_RIGHT_NULL
, dir_name
, NULL
,
2688 /* Return through std_return after this point. */
2690 xfs_itrace_entry(dp
);
2694 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
2695 prid
= dp
->i_d
.di_projid
;
2697 prid
= (xfs_prid_t
)dfltprid
;
2700 * Make sure that we have allocated dquot(s) on disk.
2702 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
2703 current_fsuid(credp
), current_fsgid(credp
), prid
,
2704 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
2708 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
2709 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2710 resblks
= XFS_MKDIR_SPACE_RES(mp
, dir_namelen
);
2711 error
= xfs_trans_reserve(tp
, resblks
, XFS_MKDIR_LOG_RES(mp
), 0,
2712 XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
);
2713 if (error
== ENOSPC
) {
2715 error
= xfs_trans_reserve(tp
, 0, XFS_MKDIR_LOG_RES(mp
), 0,
2716 XFS_TRANS_PERM_LOG_RES
,
2717 XFS_MKDIR_LOG_COUNT
);
2724 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
2725 unlock_dp_on_error
= B_TRUE
;
2728 * Check for directory link count overflow.
2730 if (dp
->i_d
.di_nlink
>= XFS_MAXLINK
) {
2731 error
= XFS_ERROR(EMLINK
);
2736 * Reserve disk quota and the inode.
2738 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
2743 (error
= xfs_dir_canenter(tp
, dp
, dir_name
, dir_namelen
)))
2746 * create the directory inode.
2748 error
= xfs_dir_ialloc(&tp
, dp
, mode
, 2,
2749 0, credp
, prid
, resblks
> 0,
2752 if (error
== ENOSPC
)
2756 xfs_itrace_ref(cdp
);
2759 * Now we add the directory inode to the transaction.
2760 * We waited until now since xfs_dir_ialloc might start
2761 * a new transaction. Had we joined the transaction
2762 * earlier, the locks might have gotten released. An error
2763 * from here on will result in the transaction cancel
2764 * unlocking dp so don't do it explicitly in the error path.
2767 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2768 unlock_dp_on_error
= B_FALSE
;
2770 XFS_BMAP_INIT(&free_list
, &first_block
);
2772 error
= xfs_dir_createname(tp
, dp
, dir_name
, dir_namelen
, cdp
->i_ino
,
2773 &first_block
, &free_list
, resblks
?
2774 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
2776 ASSERT(error
!= ENOSPC
);
2779 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2782 * Bump the in memory version number of the parent directory
2783 * so that other processes accessing it will recognize that
2784 * the directory has changed.
2788 error
= xfs_dir_init(tp
, cdp
, dp
);
2793 error
= xfs_bumplink(tp
, dp
);
2797 cvp
= XFS_ITOV(cdp
);
2805 * Attach the dquots to the new inode and modify the icount incore.
2807 XFS_QM_DQVOPCREATE(mp
, tp
, cdp
, udqp
, gdqp
);
2810 * If this is a synchronous mount, make sure that the
2811 * mkdir transaction goes to disk before returning to
2814 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2815 xfs_trans_set_sync(tp
);
2818 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2824 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2825 XFS_QM_DQRELE(mp
, udqp
);
2826 XFS_QM_DQRELE(mp
, gdqp
);
2831 /* Fall through to std_return with error = 0 or errno from
2832 * xfs_trans_commit. */
2835 if ((created
|| (error
!= 0 && dm_event_sent
!= 0)) &&
2836 DM_EVENT_ENABLED(dp
, DM_EVENT_POSTCREATE
)) {
2837 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTCREATE
,
2838 dir_vp
, DM_RIGHT_NULL
,
2839 created
? XFS_ITOV(cdp
):NULL
,
2848 xfs_bmap_cancel(&free_list
);
2850 cancel_flags
|= XFS_TRANS_ABORT
;
2852 xfs_trans_cancel(tp
, cancel_flags
);
2853 XFS_QM_DQRELE(mp
, udqp
);
2854 XFS_QM_DQRELE(mp
, gdqp
);
2856 if (unlock_dp_on_error
)
2857 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2865 bhv_vname_t
*dentry
)
2867 bhv_vnode_t
*dir_vp
= XFS_ITOV(dp
);
2868 char *name
= VNAME(dentry
);
2869 int namelen
= VNAMELEN(dentry
);
2870 xfs_mount_t
*mp
= dp
->i_mount
;
2871 xfs_inode_t
*cdp
; /* child directory */
2874 xfs_bmap_free_t free_list
;
2875 xfs_fsblock_t first_block
;
2878 int dm_di_mode
= S_IFDIR
;
2882 xfs_itrace_entry(dp
);
2884 if (XFS_FORCED_SHUTDOWN(mp
))
2885 return XFS_ERROR(EIO
);
2887 if (!xfs_get_dir_entry(dentry
, &cdp
)) {
2888 dm_di_mode
= cdp
->i_d
.di_mode
;
2892 if (DM_EVENT_ENABLED(dp
, DM_EVENT_REMOVE
)) {
2893 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_REMOVE
,
2894 dir_vp
, DM_RIGHT_NULL
,
2895 NULL
, DM_RIGHT_NULL
,
2896 name
, NULL
, dm_di_mode
, 0, 0);
2898 return XFS_ERROR(error
);
2901 /* Return through std_return after this point. */
2906 * We need to get a reference to cdp before we get our log
2907 * reservation. The reason for this is that we cannot call
2908 * xfs_iget for an inode for which we do not have a reference
2909 * once we've acquired a log reservation. This is because the
2910 * inode we are trying to get might be in xfs_inactive going
2911 * for a log reservation. Since we'll have to wait for the
2912 * inactive code to complete before returning from xfs_iget,
2913 * we need to make sure that we don't have log space reserved
2914 * when we call xfs_iget. Instead we get an unlocked reference
2915 * to the inode before getting our log reservation.
2917 error
= xfs_get_dir_entry(dentry
, &cdp
);
2919 REMOVE_DEBUG_TRACE(__LINE__
);
2923 dm_di_mode
= cdp
->i_d
.di_mode
;
2926 * Get the dquots for the inodes.
2928 error
= XFS_QM_DQATTACH(mp
, dp
, 0);
2929 if (!error
&& dp
!= cdp
)
2930 error
= XFS_QM_DQATTACH(mp
, cdp
, 0);
2933 REMOVE_DEBUG_TRACE(__LINE__
);
2937 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
2938 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2940 * We try to get the real space reservation first,
2941 * allowing for directory btree deletion(s) implying
2942 * possible bmap insert(s). If we can't get the space
2943 * reservation then we use 0 instead, and avoid the bmap
2944 * btree insert(s) in the directory code by, if the bmap
2945 * insert tries to happen, instead trimming the LAST
2946 * block from the directory.
2948 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2949 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
2950 XFS_TRANS_PERM_LOG_RES
, XFS_DEFAULT_LOG_COUNT
);
2951 if (error
== ENOSPC
) {
2953 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
2954 XFS_TRANS_PERM_LOG_RES
, XFS_DEFAULT_LOG_COUNT
);
2957 ASSERT(error
!= ENOSPC
);
2962 XFS_BMAP_INIT(&free_list
, &first_block
);
2965 * Now lock the child directory inode and the parent directory
2966 * inode in the proper order. This will take care of validating
2967 * that the directory entry for the child directory inode has
2968 * not changed while we were obtaining a log reservation.
2970 error
= xfs_lock_dir_and_entry(dp
, cdp
);
2972 xfs_trans_cancel(tp
, cancel_flags
);
2977 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2980 * Only increment the parent directory vnode count if
2981 * we didn't bump it in looking up cdp. The only time
2982 * we don't bump it is when we're looking up ".".
2987 xfs_itrace_ref(cdp
);
2988 xfs_trans_ijoin(tp
, cdp
, XFS_ILOCK_EXCL
);
2990 ASSERT(cdp
->i_d
.di_nlink
>= 2);
2991 if (cdp
->i_d
.di_nlink
!= 2) {
2992 error
= XFS_ERROR(ENOTEMPTY
);
2995 if (!xfs_dir_isempty(cdp
)) {
2996 error
= XFS_ERROR(ENOTEMPTY
);
3000 error
= xfs_dir_removename(tp
, dp
, name
, namelen
, cdp
->i_ino
,
3001 &first_block
, &free_list
, resblks
);
3005 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3008 * Bump the in memory generation count on the parent
3009 * directory so that other can know that it has changed.
3014 * Drop the link from cdp's "..".
3016 error
= xfs_droplink(tp
, dp
);
3022 * Drop the link from dp to cdp.
3024 error
= xfs_droplink(tp
, cdp
);
3030 * Drop the "." link from cdp to self.
3032 error
= xfs_droplink(tp
, cdp
);
3037 /* Determine these before committing transaction */
3038 last_cdp_link
= (cdp
)->i_d
.di_nlink
==0;
3041 * Take an extra ref on the child vnode so that it
3042 * does not go to xfs_inactive() from within the commit.
3047 * If this is a synchronous mount, make sure that the
3048 * rmdir transaction goes to disk before returning to
3051 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
3052 xfs_trans_set_sync(tp
);
3055 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
3057 xfs_bmap_cancel(&free_list
);
3058 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
|
3064 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
3073 /* Fall through to std_return with error = 0 or the errno
3074 * from xfs_trans_commit. */
3076 if (DM_EVENT_ENABLED(dp
, DM_EVENT_POSTREMOVE
)) {
3077 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTREMOVE
,
3078 dir_vp
, DM_RIGHT_NULL
,
3079 NULL
, DM_RIGHT_NULL
,
3080 name
, NULL
, dm_di_mode
,
3086 xfs_bmap_cancel(&free_list
);
3087 cancel_flags
|= XFS_TRANS_ABORT
;
3091 xfs_trans_cancel(tp
, cancel_flags
);
3098 bhv_vname_t
*dentry
,
3104 bhv_vnode_t
*dir_vp
= XFS_ITOV(dp
);
3105 xfs_mount_t
*mp
= dp
->i_mount
;
3110 xfs_bmap_free_t free_list
;
3111 xfs_fsblock_t first_block
;
3112 boolean_t unlock_dp_on_error
= B_FALSE
;
3115 xfs_fileoff_t first_fsb
;
3116 xfs_filblks_t fs_blocks
;
3118 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
3125 struct xfs_dquot
*udqp
, *gdqp
;
3127 char *link_name
= VNAME(dentry
);
3135 xfs_itrace_entry(dp
);
3137 if (XFS_FORCED_SHUTDOWN(mp
))
3138 return XFS_ERROR(EIO
);
3140 link_namelen
= VNAMELEN(dentry
);
3143 * Check component lengths of the target path name.
3145 pathlen
= strlen(target_path
);
3146 if (pathlen
>= MAXPATHLEN
) /* total string too long */
3147 return XFS_ERROR(ENAMETOOLONG
);
3148 if (pathlen
>= MAXNAMELEN
) { /* is any component too long? */
3152 for (total
= 0, path
= target_path
; total
< pathlen
;) {
3156 while(*path
== '/') {
3162 * Count up to the next slash or end of path.
3163 * Error out if the component is bigger than MAXNAMELEN.
3165 for(len
= 0; *path
!= '/' && total
< pathlen
;total
++, path
++) {
3166 if (++len
>= MAXNAMELEN
) {
3167 error
= ENAMETOOLONG
;
3174 if (DM_EVENT_ENABLED(dp
, DM_EVENT_SYMLINK
)) {
3175 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_SYMLINK
, dir_vp
,
3176 DM_RIGHT_NULL
, NULL
, DM_RIGHT_NULL
,
3177 link_name
, target_path
, 0, 0, 0);
3182 /* Return through std_return after this point. */
3185 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
3186 prid
= dp
->i_d
.di_projid
;
3188 prid
= (xfs_prid_t
)dfltprid
;
3191 * Make sure that we have allocated dquot(s) on disk.
3193 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
3194 current_fsuid(credp
), current_fsgid(credp
), prid
,
3195 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
3199 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SYMLINK
);
3200 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
3202 * The symlink will fit into the inode data fork?
3203 * There can't be any attributes so we get the whole variable part.
3205 if (pathlen
<= XFS_LITINO(mp
))
3208 fs_blocks
= XFS_B_TO_FSB(mp
, pathlen
);
3209 resblks
= XFS_SYMLINK_SPACE_RES(mp
, link_namelen
, fs_blocks
);
3210 error
= xfs_trans_reserve(tp
, resblks
, XFS_SYMLINK_LOG_RES(mp
), 0,
3211 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
3212 if (error
== ENOSPC
&& fs_blocks
== 0) {
3214 error
= xfs_trans_reserve(tp
, 0, XFS_SYMLINK_LOG_RES(mp
), 0,
3215 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
3222 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
3223 unlock_dp_on_error
= B_TRUE
;
3226 * Check whether the directory allows new symlinks or not.
3228 if (dp
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) {
3229 error
= XFS_ERROR(EPERM
);
3234 * Reserve disk quota : blocks and inode.
3236 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
3241 * Check for ability to enter directory entry, if no space reserved.
3244 (error
= xfs_dir_canenter(tp
, dp
, link_name
, link_namelen
)))
3247 * Initialize the bmap freelist prior to calling either
3248 * bmapi or the directory create code.
3250 XFS_BMAP_INIT(&free_list
, &first_block
);
3253 * Allocate an inode for the symlink.
3255 error
= xfs_dir_ialloc(&tp
, dp
, S_IFLNK
| (mode
& ~S_IFMT
),
3256 1, 0, credp
, prid
, resblks
> 0, &ip
, NULL
);
3258 if (error
== ENOSPC
)
3265 * An error after we've joined dp to the transaction will result in the
3266 * transaction cancel unlocking dp so don't do it explicitly in the
3270 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
3271 unlock_dp_on_error
= B_FALSE
;
3274 * Also attach the dquot(s) to it, if applicable.
3276 XFS_QM_DQVOPCREATE(mp
, tp
, ip
, udqp
, gdqp
);
3279 resblks
-= XFS_IALLOC_SPACE_RES(mp
);
3281 * If the symlink will fit into the inode, write it inline.
3283 if (pathlen
<= XFS_IFORK_DSIZE(ip
)) {
3284 xfs_idata_realloc(ip
, pathlen
, XFS_DATA_FORK
);
3285 memcpy(ip
->i_df
.if_u1
.if_data
, target_path
, pathlen
);
3286 ip
->i_d
.di_size
= pathlen
;
3289 * The inode was initially created in extent format.
3291 ip
->i_df
.if_flags
&= ~(XFS_IFEXTENTS
| XFS_IFBROOT
);
3292 ip
->i_df
.if_flags
|= XFS_IFINLINE
;
3294 ip
->i_d
.di_format
= XFS_DINODE_FMT_LOCAL
;
3295 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_DDATA
| XFS_ILOG_CORE
);
3299 nmaps
= SYMLINK_MAPS
;
3301 error
= xfs_bmapi(tp
, ip
, first_fsb
, fs_blocks
,
3302 XFS_BMAPI_WRITE
| XFS_BMAPI_METADATA
,
3303 &first_block
, resblks
, mval
, &nmaps
,
3310 resblks
-= fs_blocks
;
3311 ip
->i_d
.di_size
= pathlen
;
3312 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3314 cur_chunk
= target_path
;
3315 for (n
= 0; n
< nmaps
; n
++) {
3316 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
3317 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
3318 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
3319 BTOBB(byte_cnt
), 0);
3320 ASSERT(bp
&& !XFS_BUF_GETERROR(bp
));
3321 if (pathlen
< byte_cnt
) {
3324 pathlen
-= byte_cnt
;
3326 memcpy(XFS_BUF_PTR(bp
), cur_chunk
, byte_cnt
);
3327 cur_chunk
+= byte_cnt
;
3329 xfs_trans_log_buf(tp
, bp
, 0, byte_cnt
- 1);
3334 * Create the directory entry for the symlink.
3336 error
= xfs_dir_createname(tp
, dp
, link_name
, link_namelen
, ip
->i_ino
,
3337 &first_block
, &free_list
, resblks
);
3340 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3341 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
3344 * Bump the in memory version number of the parent directory
3345 * so that other processes accessing it will recognize that
3346 * the directory has changed.
3351 * If this is a synchronous mount, make sure that the
3352 * symlink transaction goes to disk before returning to
3355 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
3356 xfs_trans_set_sync(tp
);
3360 * xfs_trans_commit normally decrements the vnode ref count
3361 * when it unlocks the inode. Since we want to return the
3362 * vnode to the caller, we bump the vnode ref count now.
3366 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
3370 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
3371 XFS_QM_DQRELE(mp
, udqp
);
3372 XFS_QM_DQRELE(mp
, gdqp
);
3374 /* Fall through to std_return with error = 0 or errno from
3375 * xfs_trans_commit */
3377 if (DM_EVENT_ENABLED(dp
, DM_EVENT_POSTSYMLINK
)) {
3378 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTSYMLINK
,
3379 dir_vp
, DM_RIGHT_NULL
,
3380 error
? NULL
: XFS_ITOV(ip
),
3381 DM_RIGHT_NULL
, link_name
, target_path
,
3397 xfs_bmap_cancel(&free_list
);
3398 cancel_flags
|= XFS_TRANS_ABORT
;
3400 xfs_trans_cancel(tp
, cancel_flags
);
3401 XFS_QM_DQRELE(mp
, udqp
);
3402 XFS_QM_DQRELE(mp
, gdqp
);
3404 if (unlock_dp_on_error
)
3405 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
3413 bhv_vrwlock_t locktype
)
3415 if (S_ISDIR(ip
->i_d
.di_mode
))
3417 if (locktype
== VRWLOCK_WRITE
) {
3418 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
3419 } else if (locktype
== VRWLOCK_TRY_READ
) {
3420 return xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
);
3421 } else if (locktype
== VRWLOCK_TRY_WRITE
) {
3422 return xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
);
3424 ASSERT((locktype
== VRWLOCK_READ
) ||
3425 (locktype
== VRWLOCK_WRITE_DIRECT
));
3426 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
3436 bhv_vrwlock_t locktype
)
3438 if (S_ISDIR(ip
->i_d
.di_mode
))
3440 if (locktype
== VRWLOCK_WRITE
) {
3441 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
3443 ASSERT((locktype
== VRWLOCK_READ
) ||
3444 (locktype
== VRWLOCK_WRITE_DIRECT
));
3445 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
3456 xfs_mount_t
*mp
= ip
->i_mount
;
3459 if (XFS_FORCED_SHUTDOWN(mp
))
3460 return XFS_ERROR(EIO
);
3463 * Bypass inodes which have already been cleaned by
3464 * the inode flush clustering code inside xfs_iflush
3466 if (xfs_inode_clean(ip
))
3470 * We make this non-blocking if the inode is contended,
3471 * return EAGAIN to indicate to the caller that they
3472 * did not succeed. This prevents the flush path from
3473 * blocking on inodes inside another operation right
3474 * now, they get caught later by xfs_sync.
3476 if (flags
& FLUSH_SYNC
) {
3477 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
3479 } else if (xfs_ilock_nowait(ip
, XFS_ILOCK_SHARED
)) {
3480 if (xfs_ipincount(ip
) || !xfs_iflock_nowait(ip
)) {
3481 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3488 error
= xfs_iflush(ip
, (flags
& FLUSH_SYNC
) ? XFS_IFLUSH_SYNC
3489 : XFS_IFLUSH_ASYNC_NOBLOCK
);
3490 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3502 xfs_mount_t
*mp
= ip
->i_mount
;
3506 if (!capable(CAP_SYS_ADMIN
))
3507 return XFS_ERROR(EPERM
);
3509 if (XFS_FORCED_SHUTDOWN(mp
))
3510 return XFS_ERROR(EIO
);
3512 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SET_DMATTRS
);
3513 error
= xfs_trans_reserve(tp
, 0, XFS_ICHANGE_LOG_RES (mp
), 0, 0, 0);
3515 xfs_trans_cancel(tp
, 0);
3518 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3519 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
3521 ip
->i_d
.di_dmevmask
= evmask
;
3522 ip
->i_d
.di_dmstate
= state
;
3524 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3526 error
= xfs_trans_commit(tp
, 0);
3535 bhv_vnode_t
*vp
= XFS_ITOV(ip
);
3537 xfs_itrace_entry(ip
);
3539 ASSERT(!VN_MAPPED(vp
));
3541 /* bad inode, get out here ASAP */
3549 ASSERT(XFS_FORCED_SHUTDOWN(ip
->i_mount
) || ip
->i_delayed_blks
== 0);
3552 * Make sure the atime in the XFS inode is correct before freeing the
3555 xfs_synchronize_atime(ip
);
3558 * If we have nothing to flush with this inode then complete the
3559 * teardown now, otherwise break the link between the xfs inode and the
3560 * linux inode and clean up the xfs inode later. This avoids flushing
3561 * the inode to disk during the delete operation itself.
3563 * When breaking the link, we need to set the XFS_IRECLAIMABLE flag
3564 * first to ensure that xfs_iunpin() will never see an xfs inode
3565 * that has a linux inode being reclaimed. Synchronisation is provided
3566 * by the i_flags_lock.
3568 if (!ip
->i_update_core
&& (ip
->i_itemp
== NULL
)) {
3569 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3571 return xfs_finish_reclaim(ip
, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC
);
3573 xfs_mount_t
*mp
= ip
->i_mount
;
3575 /* Protect sync and unpin from us */
3576 XFS_MOUNT_ILOCK(mp
);
3577 spin_lock(&ip
->i_flags_lock
);
3578 __xfs_iflags_set(ip
, XFS_IRECLAIMABLE
);
3579 vn_to_inode(vp
)->i_private
= NULL
;
3581 spin_unlock(&ip
->i_flags_lock
);
3582 list_add_tail(&ip
->i_reclaim
, &mp
->m_del_inodes
);
3583 XFS_MOUNT_IUNLOCK(mp
);
3594 xfs_perag_t
*pag
= xfs_get_perag(ip
->i_mount
, ip
->i_ino
);
3595 bhv_vnode_t
*vp
= XFS_ITOV_NULL(ip
);
3598 if (vp
&& VN_BAD(vp
))
3601 /* The hash lock here protects a thread in xfs_iget_core from
3602 * racing with us on linking the inode back with a vnode.
3603 * Once we have the XFS_IRECLAIM flag set it will not touch
3606 write_lock(&pag
->pag_ici_lock
);
3607 spin_lock(&ip
->i_flags_lock
);
3608 if (__xfs_iflags_test(ip
, XFS_IRECLAIM
) ||
3609 (!__xfs_iflags_test(ip
, XFS_IRECLAIMABLE
) && vp
== NULL
)) {
3610 spin_unlock(&ip
->i_flags_lock
);
3611 write_unlock(&pag
->pag_ici_lock
);
3614 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3618 __xfs_iflags_set(ip
, XFS_IRECLAIM
);
3619 spin_unlock(&ip
->i_flags_lock
);
3620 write_unlock(&pag
->pag_ici_lock
);
3621 xfs_put_perag(ip
->i_mount
, pag
);
3624 * If the inode is still dirty, then flush it out. If the inode
3625 * is not in the AIL, then it will be OK to flush it delwri as
3626 * long as xfs_iflush() does not keep any references to the inode.
3627 * We leave that decision up to xfs_iflush() since it has the
3628 * knowledge of whether it's OK to simply do a delwri flush of
3629 * the inode or whether we need to wait until the inode is
3630 * pulled from the AIL.
3631 * We get the flush lock regardless, though, just to make sure
3632 * we don't free it while it is being flushed.
3635 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3639 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
3640 if (ip
->i_update_core
||
3641 ((ip
->i_itemp
!= NULL
) &&
3642 (ip
->i_itemp
->ili_format
.ilf_fields
!= 0))) {
3643 error
= xfs_iflush(ip
, sync_mode
);
3645 * If we hit an error, typically because of filesystem
3646 * shutdown, we don't need to let vn_reclaim to know
3647 * because we're gonna reclaim the inode anyway.
3650 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3653 xfs_iflock(ip
); /* synchronize with xfs_iflush_done */
3656 ASSERT(ip
->i_update_core
== 0);
3657 ASSERT(ip
->i_itemp
== NULL
||
3658 ip
->i_itemp
->ili_format
.ilf_fields
== 0);
3662 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3670 xfs_finish_reclaim_all(xfs_mount_t
*mp
, int noblock
)
3673 xfs_inode_t
*ip
, *n
;
3678 XFS_MOUNT_ILOCK(mp
);
3679 list_for_each_entry_safe(ip
, n
, &mp
->m_del_inodes
, i_reclaim
) {
3681 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
) == 0)
3683 if (xfs_ipincount(ip
) ||
3684 !xfs_iflock_nowait(ip
)) {
3685 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3689 XFS_MOUNT_IUNLOCK(mp
);
3690 if (xfs_finish_reclaim(ip
, noblock
,
3691 XFS_IFLUSH_DELWRI_ELSE_ASYNC
))
3700 XFS_MOUNT_IUNLOCK(mp
);
3705 * xfs_alloc_file_space()
3706 * This routine allocates disk space for the given file.
3708 * If alloc_type == 0, this request is for an ALLOCSP type
3709 * request which will change the file size. In this case, no
3710 * DMAPI event will be generated by the call. A TRUNCATE event
3711 * will be generated later by xfs_setattr.
3713 * If alloc_type != 0, this request is for a RESVSP type
3714 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3715 * lower block boundary byte address is less than the file's
3724 xfs_alloc_file_space(
3731 xfs_mount_t
*mp
= ip
->i_mount
;
3733 xfs_filblks_t allocated_fsb
;
3734 xfs_filblks_t allocatesize_fsb
;
3735 xfs_extlen_t extsz
, temp
;
3736 xfs_fileoff_t startoffset_fsb
;
3737 xfs_fsblock_t firstfsb
;
3743 xfs_bmbt_irec_t imaps
[1], *imapp
;
3744 xfs_bmap_free_t free_list
;
3745 uint qblocks
, resblks
, resrtextents
;
3749 xfs_itrace_entry(ip
);
3751 if (XFS_FORCED_SHUTDOWN(mp
))
3752 return XFS_ERROR(EIO
);
3754 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
3758 return XFS_ERROR(EINVAL
);
3760 rt
= XFS_IS_REALTIME_INODE(ip
);
3761 extsz
= xfs_get_extsz_hint(ip
);
3766 bmapi_flag
= XFS_BMAPI_WRITE
| (alloc_type
? XFS_BMAPI_PREALLOC
: 0);
3767 startoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
3768 allocatesize_fsb
= XFS_B_TO_FSB(mp
, count
);
3770 /* Generate a DMAPI event if needed. */
3771 if (alloc_type
!= 0 && offset
< ip
->i_size
&&
3772 (attr_flags
&ATTR_DMI
) == 0 &&
3773 DM_EVENT_ENABLED(ip
, DM_EVENT_WRITE
)) {
3774 xfs_off_t end_dmi_offset
;
3776 end_dmi_offset
= offset
+len
;
3777 if (end_dmi_offset
> ip
->i_size
)
3778 end_dmi_offset
= ip
->i_size
;
3779 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, XFS_ITOV(ip
),
3780 offset
, end_dmi_offset
- offset
,
3787 * Allocate file space until done or until there is an error
3790 while (allocatesize_fsb
&& !error
) {
3794 * Determine space reservations for data/realtime.
3796 if (unlikely(extsz
)) {
3797 s
= startoffset_fsb
;
3800 e
= startoffset_fsb
+ allocatesize_fsb
;
3801 if ((temp
= do_mod(startoffset_fsb
, extsz
)))
3803 if ((temp
= do_mod(e
, extsz
)))
3807 e
= allocatesize_fsb
;
3811 resrtextents
= qblocks
= (uint
)(e
- s
);
3812 resrtextents
/= mp
->m_sb
.sb_rextsize
;
3813 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
3814 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
3817 resblks
= qblocks
= \
3818 XFS_DIOSTRAT_SPACE_RES(mp
, (uint
)(e
- s
));
3819 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
3823 * Allocate and setup the transaction.
3825 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
3826 error
= xfs_trans_reserve(tp
, resblks
,
3827 XFS_WRITE_LOG_RES(mp
), resrtextents
,
3828 XFS_TRANS_PERM_LOG_RES
,
3829 XFS_WRITE_LOG_COUNT
);
3831 * Check for running out of space
3835 * Free the transaction structure.
3837 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
3838 xfs_trans_cancel(tp
, 0);
3841 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3842 error
= XFS_TRANS_RESERVE_QUOTA_NBLKS(mp
, tp
, ip
,
3843 qblocks
, 0, quota_flag
);
3847 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
3848 xfs_trans_ihold(tp
, ip
);
3851 * Issue the xfs_bmapi() call to allocate the blocks
3853 XFS_BMAP_INIT(&free_list
, &firstfsb
);
3854 error
= xfs_bmapi(tp
, ip
, startoffset_fsb
,
3855 allocatesize_fsb
, bmapi_flag
,
3856 &firstfsb
, 0, imapp
, &nimaps
,
3863 * Complete the transaction
3865 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
3870 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
3871 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3876 allocated_fsb
= imapp
->br_blockcount
;
3879 error
= XFS_ERROR(ENOSPC
);
3883 startoffset_fsb
+= allocated_fsb
;
3884 allocatesize_fsb
-= allocated_fsb
;
3887 if (error
== ENOSPC
&& (attr_flags
& ATTR_DMI
) == 0 &&
3888 DM_EVENT_ENABLED(ip
, DM_EVENT_NOSPACE
)) {
3889 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_NOSPACE
,
3890 XFS_ITOV(ip
), DM_RIGHT_NULL
,
3891 XFS_ITOV(ip
), DM_RIGHT_NULL
,
3892 NULL
, NULL
, 0, 0, 0); /* Delay flag intentionally unused */
3894 goto retry
; /* Maybe DMAPI app. has made space */
3895 /* else fall through with error from XFS_SEND_DATA */
3900 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
3901 xfs_bmap_cancel(&free_list
);
3902 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp
, tp
, ip
, qblocks
, 0, quota_flag
);
3904 error1
: /* Just cancel transaction */
3905 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
3906 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3907 goto dmapi_enospc_check
;
3911 * Zero file bytes between startoff and endoff inclusive.
3912 * The iolock is held exclusive and no blocks are buffered.
3915 xfs_zero_remaining_bytes(
3920 xfs_bmbt_irec_t imap
;
3921 xfs_fileoff_t offset_fsb
;
3922 xfs_off_t lastoffset
;
3925 xfs_mount_t
*mp
= ip
->i_mount
;
3929 bp
= xfs_buf_get_noaddr(mp
->m_sb
.sb_blocksize
,
3930 XFS_IS_REALTIME_INODE(ip
) ?
3931 mp
->m_rtdev_targp
: mp
->m_ddev_targp
);
3933 for (offset
= startoff
; offset
<= endoff
; offset
= lastoffset
+ 1) {
3934 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
3936 error
= xfs_bmapi(NULL
, ip
, offset_fsb
, 1, 0,
3937 NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
3938 if (error
|| nimap
< 1)
3940 ASSERT(imap
.br_blockcount
>= 1);
3941 ASSERT(imap
.br_startoff
== offset_fsb
);
3942 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+ 1) - 1;
3943 if (lastoffset
> endoff
)
3944 lastoffset
= endoff
;
3945 if (imap
.br_startblock
== HOLESTARTBLOCK
)
3947 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
3948 if (imap
.br_state
== XFS_EXT_UNWRITTEN
)
3951 XFS_BUF_UNWRITE(bp
);
3953 XFS_BUF_SET_ADDR(bp
, XFS_FSB_TO_DB(ip
, imap
.br_startblock
));
3955 if ((error
= xfs_iowait(bp
))) {
3956 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
3957 mp
, bp
, XFS_BUF_ADDR(bp
));
3960 memset(XFS_BUF_PTR(bp
) +
3961 (offset
- XFS_FSB_TO_B(mp
, imap
.br_startoff
)),
3962 0, lastoffset
- offset
+ 1);
3967 if ((error
= xfs_iowait(bp
))) {
3968 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
3969 mp
, bp
, XFS_BUF_ADDR(bp
));
3978 * xfs_free_file_space()
3979 * This routine frees disk space for the given file.
3981 * This routine is only called by xfs_change_file_space
3982 * for an UNRESVSP type call.
3990 xfs_free_file_space(
3999 xfs_off_t end_dmi_offset
;
4000 xfs_fileoff_t endoffset_fsb
;
4002 xfs_fsblock_t firstfsb
;
4003 xfs_bmap_free_t free_list
;
4004 xfs_bmbt_irec_t imap
;
4012 xfs_fileoff_t startoffset_fsb
;
4014 int need_iolock
= 1;
4019 xfs_itrace_entry(ip
);
4021 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
4025 if (len
<= 0) /* if nothing being freed */
4027 rt
= XFS_IS_REALTIME_INODE(ip
);
4028 startoffset_fsb
= XFS_B_TO_FSB(mp
, offset
);
4029 end_dmi_offset
= offset
+ len
;
4030 endoffset_fsb
= XFS_B_TO_FSBT(mp
, end_dmi_offset
);
4032 if (offset
< ip
->i_size
&& (attr_flags
& ATTR_DMI
) == 0 &&
4033 DM_EVENT_ENABLED(ip
, DM_EVENT_WRITE
)) {
4034 if (end_dmi_offset
> ip
->i_size
)
4035 end_dmi_offset
= ip
->i_size
;
4036 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, vp
,
4037 offset
, end_dmi_offset
- offset
,
4038 AT_DELAY_FLAG(attr_flags
), NULL
);
4043 if (attr_flags
& ATTR_NOLOCK
)
4046 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
4047 vn_iowait(ip
); /* wait for the completion of any pending DIOs */
4050 rounding
= max_t(uint
, 1 << mp
->m_sb
.sb_blocklog
, PAGE_CACHE_SIZE
);
4051 ioffset
= offset
& ~(rounding
- 1);
4053 if (VN_CACHED(vp
) != 0) {
4054 xfs_inval_cached_trace(ip
, ioffset
, -1, ioffset
, -1);
4055 error
= xfs_flushinval_pages(ip
, ioffset
, -1, FI_REMAPF_LOCKED
);
4057 goto out_unlock_iolock
;
4061 * Need to zero the stuff we're not freeing, on disk.
4062 * If its a realtime file & can't use unwritten extents then we
4063 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4064 * will take care of it for us.
4066 if (rt
&& !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
4068 error
= xfs_bmapi(NULL
, ip
, startoffset_fsb
,
4069 1, 0, NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
4071 goto out_unlock_iolock
;
4072 ASSERT(nimap
== 0 || nimap
== 1);
4073 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
4076 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
4077 block
= imap
.br_startblock
;
4078 mod
= do_div(block
, mp
->m_sb
.sb_rextsize
);
4080 startoffset_fsb
+= mp
->m_sb
.sb_rextsize
- mod
;
4083 error
= xfs_bmapi(NULL
, ip
, endoffset_fsb
- 1,
4084 1, 0, NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
4086 goto out_unlock_iolock
;
4087 ASSERT(nimap
== 0 || nimap
== 1);
4088 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
4089 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
4091 if (mod
&& (mod
!= mp
->m_sb
.sb_rextsize
))
4092 endoffset_fsb
-= mod
;
4095 if ((done
= (endoffset_fsb
<= startoffset_fsb
)))
4097 * One contiguous piece to clear
4099 error
= xfs_zero_remaining_bytes(ip
, offset
, offset
+ len
- 1);
4102 * Some full blocks, possibly two pieces to clear
4104 if (offset
< XFS_FSB_TO_B(mp
, startoffset_fsb
))
4105 error
= xfs_zero_remaining_bytes(ip
, offset
,
4106 XFS_FSB_TO_B(mp
, startoffset_fsb
) - 1);
4108 XFS_FSB_TO_B(mp
, endoffset_fsb
) < offset
+ len
)
4109 error
= xfs_zero_remaining_bytes(ip
,
4110 XFS_FSB_TO_B(mp
, endoffset_fsb
),
4115 * free file space until done or until there is an error
4117 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
4118 while (!error
&& !done
) {
4121 * allocate and setup the transaction. Allow this
4122 * transaction to dip into the reserve blocks to ensure
4123 * the freeing of the space succeeds at ENOSPC.
4125 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
4126 tp
->t_flags
|= XFS_TRANS_RESERVE
;
4127 error
= xfs_trans_reserve(tp
,
4129 XFS_WRITE_LOG_RES(mp
),
4131 XFS_TRANS_PERM_LOG_RES
,
4132 XFS_WRITE_LOG_COUNT
);
4135 * check for running out of space
4139 * Free the transaction structure.
4141 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
4142 xfs_trans_cancel(tp
, 0);
4145 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4146 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
,
4147 ip
->i_udquot
, ip
->i_gdquot
, resblks
, 0,
4148 XFS_QMOPT_RES_REGBLKS
);
4152 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
4153 xfs_trans_ihold(tp
, ip
);
4156 * issue the bunmapi() call to free the blocks
4158 XFS_BMAP_INIT(&free_list
, &firstfsb
);
4159 error
= xfs_bunmapi(tp
, ip
, startoffset_fsb
,
4160 endoffset_fsb
- startoffset_fsb
,
4161 0, 2, &firstfsb
, &free_list
, NULL
, &done
);
4167 * complete the transaction
4169 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
4174 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
4175 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4180 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
4184 xfs_bmap_cancel(&free_list
);
4186 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
4187 xfs_iunlock(ip
, need_iolock
? (XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
) :
4193 * xfs_change_file_space()
4194 * This routine allocates or frees disk space for the given file.
4195 * The user specified parameters are checked for alignment and size
4204 xfs_change_file_space(
4212 xfs_mount_t
*mp
= ip
->i_mount
;
4217 xfs_off_t startoffset
;
4222 xfs_itrace_entry(ip
);
4224 if (!S_ISREG(ip
->i_d
.di_mode
))
4225 return XFS_ERROR(EINVAL
);
4227 switch (bf
->l_whence
) {
4228 case 0: /*SEEK_SET*/
4230 case 1: /*SEEK_CUR*/
4231 bf
->l_start
+= offset
;
4233 case 2: /*SEEK_END*/
4234 bf
->l_start
+= ip
->i_size
;
4237 return XFS_ERROR(EINVAL
);
4240 llen
= bf
->l_len
> 0 ? bf
->l_len
- 1 : bf
->l_len
;
4242 if ( (bf
->l_start
< 0)
4243 || (bf
->l_start
> XFS_MAXIOFFSET(mp
))
4244 || (bf
->l_start
+ llen
< 0)
4245 || (bf
->l_start
+ llen
> XFS_MAXIOFFSET(mp
)))
4246 return XFS_ERROR(EINVAL
);
4250 startoffset
= bf
->l_start
;
4254 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4256 * These calls do NOT zero the data space allocated to the file,
4257 * nor do they change the file size.
4259 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4261 * These calls cause the new file data to be zeroed and the file
4262 * size to be changed.
4264 setprealloc
= clrprealloc
= 0;
4267 case XFS_IOC_RESVSP
:
4268 case XFS_IOC_RESVSP64
:
4269 error
= xfs_alloc_file_space(ip
, startoffset
, bf
->l_len
,
4276 case XFS_IOC_UNRESVSP
:
4277 case XFS_IOC_UNRESVSP64
:
4278 if ((error
= xfs_free_file_space(ip
, startoffset
, bf
->l_len
,
4283 case XFS_IOC_ALLOCSP
:
4284 case XFS_IOC_ALLOCSP64
:
4285 case XFS_IOC_FREESP
:
4286 case XFS_IOC_FREESP64
:
4287 if (startoffset
> fsize
) {
4288 error
= xfs_alloc_file_space(ip
, fsize
,
4289 startoffset
- fsize
, 0, attr_flags
);
4294 va
.va_mask
= XFS_AT_SIZE
;
4295 va
.va_size
= startoffset
;
4297 error
= xfs_setattr(ip
, &va
, attr_flags
, credp
);
4307 return XFS_ERROR(EINVAL
);
4311 * update the inode timestamp, mode, and prealloc flag bits
4313 tp
= xfs_trans_alloc(mp
, XFS_TRANS_WRITEID
);
4315 if ((error
= xfs_trans_reserve(tp
, 0, XFS_WRITEID_LOG_RES(mp
),
4318 xfs_trans_cancel(tp
, 0);
4322 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4324 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
4325 xfs_trans_ihold(tp
, ip
);
4327 if ((attr_flags
& ATTR_DMI
) == 0) {
4328 ip
->i_d
.di_mode
&= ~S_ISUID
;
4331 * Note that we don't have to worry about mandatory
4332 * file locking being disabled here because we only
4333 * clear the S_ISGID bit if the Group execute bit is
4334 * on, but if it was on then mandatory locking wouldn't
4335 * have been enabled.
4337 if (ip
->i_d
.di_mode
& S_IXGRP
)
4338 ip
->i_d
.di_mode
&= ~S_ISGID
;
4340 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
4343 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
4344 else if (clrprealloc
)
4345 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
4347 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
4348 xfs_trans_set_sync(tp
);
4350 error
= xfs_trans_commit(tp
, 0);
4352 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);