2 * Copyright (c) 2000-2005 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
19 #include <linux/capability.h>
23 #include "xfs_types.h"
27 #include "xfs_trans.h"
32 #include "xfs_dmapi.h"
33 #include "xfs_mount.h"
34 #include "xfs_da_btree.h"
35 #include "xfs_bmap_btree.h"
36 #include "xfs_alloc_btree.h"
37 #include "xfs_ialloc_btree.h"
38 #include "xfs_dir_sf.h"
39 #include "xfs_dir2_sf.h"
40 #include "xfs_attr_sf.h"
41 #include "xfs_dinode.h"
42 #include "xfs_inode.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_dir_leaf.h"
45 #include "xfs_itable.h"
46 #include "xfs_btree.h"
47 #include "xfs_ialloc.h"
48 #include "xfs_alloc.h"
52 #include "xfs_error.h"
53 #include "xfs_quota.h"
54 #include "xfs_utils.h"
55 #include "xfs_rtalloc.h"
56 #include "xfs_refcache.h"
57 #include "xfs_trans_space.h"
58 #include "xfs_log_priv.h"
63 * The maximum pathlen is 1024 bytes. Since the minimum file system
64 * blocksize is 512 bytes, we can get a max of 2 extents back from
67 #define SYMLINK_MAPS 2
70 * For xfs, we check that the file isn't too big to be opened by this kernel.
71 * No other open action is required for regular files. Devices are handled
72 * through the specfs file system, pipes through fifofs. Device and
73 * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
74 * when a new vnode is first looked up or created.
85 vp
= BHV_TO_VNODE(bdp
);
88 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
89 return XFS_ERROR(EIO
);
92 * If it's a directory with any blocks, read-ahead block 0
93 * as we're almost certain to have the next operation be a read there.
95 if (VN_ISDIR(vp
) && ip
->i_d
.di_nextents
> 0) {
96 mode
= xfs_ilock_map_shared(ip
);
97 if (ip
->i_d
.di_nextents
> 0)
98 (void)xfs_da_reada_buf(NULL
, ip
, 0, XFS_DATA_FORK
);
99 xfs_iunlock(ip
, mode
);
119 vp
= BHV_TO_VNODE(bdp
);
120 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
122 ip
= XFS_BHVTOI(bdp
);
125 if (XFS_FORCED_SHUTDOWN(mp
))
126 return XFS_ERROR(EIO
);
128 if (!(flags
& ATTR_LAZY
))
129 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
131 vap
->va_size
= ip
->i_d
.di_size
;
132 if (vap
->va_mask
== XFS_AT_SIZE
)
136 XFS_FSB_TO_BB(mp
, ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
137 vap
->va_nodeid
= ip
->i_ino
;
139 vap
->va_nodeid
+= mp
->m_inoadd
;
141 vap
->va_nlink
= ip
->i_d
.di_nlink
;
144 * Quick exit for non-stat callers
147 ~(XFS_AT_SIZE
|XFS_AT_FSID
|XFS_AT_NODEID
|
148 XFS_AT_NLINK
|XFS_AT_BLKSIZE
)) == 0)
152 * Copy from in-core inode.
154 vap
->va_mode
= ip
->i_d
.di_mode
;
155 vap
->va_uid
= ip
->i_d
.di_uid
;
156 vap
->va_gid
= ip
->i_d
.di_gid
;
157 vap
->va_projid
= ip
->i_d
.di_projid
;
160 * Check vnode type block/char vs. everything else.
162 switch (ip
->i_d
.di_mode
& S_IFMT
) {
165 vap
->va_rdev
= ip
->i_df
.if_u2
.if_rdev
;
166 vap
->va_blocksize
= BLKDEV_IOSIZE
;
171 if (!(ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
)) {
172 vap
->va_blocksize
= xfs_preferred_iosize(mp
);
176 * If the file blocks are being allocated from a
177 * realtime partition, then return the inode's
178 * realtime extent size or the realtime volume's
181 vap
->va_blocksize
= ip
->i_d
.di_extsize
?
182 (ip
->i_d
.di_extsize
<< mp
->m_sb
.sb_blocklog
) :
183 (mp
->m_sb
.sb_rextsize
<< mp
->m_sb
.sb_blocklog
);
188 vn_atime_to_timespec(vp
, &vap
->va_atime
);
189 vap
->va_mtime
.tv_sec
= ip
->i_d
.di_mtime
.t_sec
;
190 vap
->va_mtime
.tv_nsec
= ip
->i_d
.di_mtime
.t_nsec
;
191 vap
->va_ctime
.tv_sec
= ip
->i_d
.di_ctime
.t_sec
;
192 vap
->va_ctime
.tv_nsec
= ip
->i_d
.di_ctime
.t_nsec
;
195 * Exit for stat callers. See if any of the rest of the fields
196 * to be filled in are needed.
199 (XFS_AT_XFLAGS
|XFS_AT_EXTSIZE
|XFS_AT_NEXTENTS
|XFS_AT_ANEXTENTS
|
200 XFS_AT_GENCOUNT
|XFS_AT_VCODE
)) == 0)
204 * Convert di_flags to xflags.
206 vap
->va_xflags
= xfs_ip2xflags(ip
);
209 * Exit for inode revalidate. See if any of the rest of
210 * the fields to be filled in are needed.
213 (XFS_AT_EXTSIZE
|XFS_AT_NEXTENTS
|XFS_AT_ANEXTENTS
|
214 XFS_AT_GENCOUNT
|XFS_AT_VCODE
)) == 0)
217 vap
->va_extsize
= ip
->i_d
.di_extsize
<< mp
->m_sb
.sb_blocklog
;
219 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) ?
220 ip
->i_df
.if_bytes
/ sizeof(xfs_bmbt_rec_t
) :
224 (ip
->i_afp
->if_flags
& XFS_IFEXTENTS
) ?
225 ip
->i_afp
->if_bytes
/ sizeof(xfs_bmbt_rec_t
) :
226 ip
->i_d
.di_anextents
;
228 vap
->va_anextents
= 0;
229 vap
->va_gen
= ip
->i_d
.di_gen
;
232 if (!(flags
& ATTR_LAZY
))
233 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
259 xfs_prid_t projid
=0, iprojid
=0;
260 int mandlock_before
, mandlock_after
;
261 struct xfs_dquot
*udqp
, *gdqp
, *olddquot1
, *olddquot2
;
265 vp
= BHV_TO_VNODE(bdp
);
266 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
268 if (vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
)
269 return XFS_ERROR(EROFS
);
272 * Cannot set certain attributes.
275 if (mask
& XFS_AT_NOSET
) {
276 return XFS_ERROR(EINVAL
);
279 ip
= XFS_BHVTOI(bdp
);
282 if (XFS_FORCED_SHUTDOWN(mp
))
283 return XFS_ERROR(EIO
);
286 * Timestamps do not need to be logged and hence do not
287 * need to be done within a transaction.
289 if (mask
& XFS_AT_UPDTIMES
) {
290 ASSERT((mask
& ~XFS_AT_UPDTIMES
) == 0);
291 timeflags
= ((mask
& XFS_AT_UPDATIME
) ? XFS_ICHGTIME_ACC
: 0) |
292 ((mask
& XFS_AT_UPDCTIME
) ? XFS_ICHGTIME_CHG
: 0) |
293 ((mask
& XFS_AT_UPDMTIME
) ? XFS_ICHGTIME_MOD
: 0);
294 xfs_ichgtime(ip
, timeflags
);
298 olddquot1
= olddquot2
= NULL
;
302 * If disk quotas is on, we make sure that the dquots do exist on disk,
303 * before we start any other transactions. Trying to do this later
304 * is messy. We don't care to take a readlock to look at the ids
305 * in inode here, because we can't hold it across the trans_reserve.
306 * If the IDs do change before we take the ilock, we're covered
307 * because the i_*dquot fields will get updated anyway.
309 if (XFS_IS_QUOTA_ON(mp
) &&
310 (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
))) {
313 if ((mask
& XFS_AT_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
315 qflags
|= XFS_QMOPT_UQUOTA
;
317 uid
= ip
->i_d
.di_uid
;
319 if ((mask
& XFS_AT_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
321 qflags
|= XFS_QMOPT_GQUOTA
;
323 gid
= ip
->i_d
.di_gid
;
325 if ((mask
& XFS_AT_PROJID
) && XFS_IS_PQUOTA_ON(mp
)) {
326 projid
= vap
->va_projid
;
327 qflags
|= XFS_QMOPT_PQUOTA
;
329 projid
= ip
->i_d
.di_projid
;
332 * We take a reference when we initialize udqp and gdqp,
333 * so it is important that we never blindly double trip on
334 * the same variable. See xfs_create() for an example.
336 ASSERT(udqp
== NULL
);
337 ASSERT(gdqp
== NULL
);
338 code
= XFS_QM_DQVOPALLOC(mp
, ip
, uid
, gid
, projid
, qflags
,
345 * For the other attributes, we acquire the inode lock and
346 * first do an error checking pass.
349 lock_flags
= XFS_ILOCK_EXCL
;
350 ASSERT(flags
& ATTR_NOLOCK
? flags
& ATTR_DMI
: 1);
351 if (flags
& ATTR_NOLOCK
)
353 if (!(mask
& XFS_AT_SIZE
)) {
354 if ((mask
!= (XFS_AT_CTIME
|XFS_AT_ATIME
|XFS_AT_MTIME
)) ||
355 (mp
->m_flags
& XFS_MOUNT_WSYNC
)) {
356 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
358 if ((code
= xfs_trans_reserve(tp
, 0,
359 XFS_ICHANGE_LOG_RES(mp
), 0,
366 if (DM_EVENT_ENABLED (vp
->v_vfsp
, ip
, DM_EVENT_TRUNCATE
) &&
367 !(flags
& ATTR_DMI
)) {
368 int dmflags
= AT_DELAY_FLAG(flags
) | DM_SEM_FLAG_WR
;
369 code
= XFS_SEND_DATA(mp
, DM_EVENT_TRUNCATE
, vp
,
370 vap
->va_size
, 0, dmflags
, NULL
);
377 lock_flags
|= XFS_IOLOCK_EXCL
;
380 xfs_ilock(ip
, lock_flags
);
382 /* boolean: are we the file owner? */
383 file_owner
= (current_fsuid(credp
) == ip
->i_d
.di_uid
);
386 * Change various properties of a file.
387 * Only the owner or users with CAP_FOWNER
388 * capability may do these things.
391 (XFS_AT_MODE
|XFS_AT_XFLAGS
|XFS_AT_EXTSIZE
|XFS_AT_UID
|
392 XFS_AT_GID
|XFS_AT_PROJID
)) {
394 * CAP_FOWNER overrides the following restrictions:
396 * The user ID of the calling process must be equal
397 * to the file owner ID, except in cases where the
398 * CAP_FSETID capability is applicable.
400 if (!file_owner
&& !capable(CAP_FOWNER
)) {
401 code
= XFS_ERROR(EPERM
);
406 * CAP_FSETID overrides the following restrictions:
408 * The effective user ID of the calling process shall match
409 * the file owner when setting the set-user-ID and
410 * set-group-ID bits on that file.
412 * The effective group ID or one of the supplementary group
413 * IDs of the calling process shall match the group owner of
414 * the file when setting the set-group-ID bit on that file
416 if (mask
& XFS_AT_MODE
) {
419 if ((vap
->va_mode
& S_ISUID
) && !file_owner
)
421 if ((vap
->va_mode
& S_ISGID
) &&
422 !in_group_p((gid_t
)ip
->i_d
.di_gid
))
425 /* Linux allows this, Irix doesn't. */
426 if ((vap
->va_mode
& S_ISVTX
) && !VN_ISDIR(vp
))
429 if (m
&& !capable(CAP_FSETID
))
435 * Change file ownership. Must be the owner or privileged.
436 * If the system was configured with the "restricted_chown"
437 * option, the owner is not permitted to give away the file,
438 * and can change the group id only to a group of which he
439 * or she is a member.
441 if (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
)) {
443 * These IDs could have changed since we last looked at them.
444 * But, we're assured that if the ownership did change
445 * while we didn't have the inode locked, inode's dquot(s)
446 * would have changed also.
448 iuid
= ip
->i_d
.di_uid
;
449 iprojid
= ip
->i_d
.di_projid
;
450 igid
= ip
->i_d
.di_gid
;
451 gid
= (mask
& XFS_AT_GID
) ? vap
->va_gid
: igid
;
452 uid
= (mask
& XFS_AT_UID
) ? vap
->va_uid
: iuid
;
453 projid
= (mask
& XFS_AT_PROJID
) ? (xfs_prid_t
)vap
->va_projid
:
457 * CAP_CHOWN overrides the following restrictions:
459 * If _POSIX_CHOWN_RESTRICTED is defined, this capability
460 * shall override the restriction that a process cannot
461 * change the user ID of a file it owns and the restriction
462 * that the group ID supplied to the chown() function
463 * shall be equal to either the group ID or one of the
464 * supplementary group IDs of the calling process.
466 if (restricted_chown
&&
467 (iuid
!= uid
|| (igid
!= gid
&&
468 !in_group_p((gid_t
)gid
))) &&
469 !capable(CAP_CHOWN
)) {
470 code
= XFS_ERROR(EPERM
);
474 * Do a quota reservation only if uid/projid/gid is actually
477 if ((XFS_IS_UQUOTA_ON(mp
) && iuid
!= uid
) ||
478 (XFS_IS_PQUOTA_ON(mp
) && iprojid
!= projid
) ||
479 (XFS_IS_GQUOTA_ON(mp
) && igid
!= gid
)) {
481 code
= XFS_QM_DQVOPCHOWNRESV(mp
, tp
, ip
, udqp
, gdqp
,
482 capable(CAP_FOWNER
) ?
483 XFS_QMOPT_FORCE_RES
: 0);
484 if (code
) /* out of quota */
490 * Truncate file. Must have write permission and not be a directory.
492 if (mask
& XFS_AT_SIZE
) {
493 /* Short circuit the truncate case for zero length files */
494 if ((vap
->va_size
== 0) &&
495 (ip
->i_d
.di_size
== 0) && (ip
->i_d
.di_nextents
== 0)) {
496 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
497 lock_flags
&= ~XFS_ILOCK_EXCL
;
498 if (mask
& XFS_AT_CTIME
)
499 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
505 code
= XFS_ERROR(EISDIR
);
507 } else if (!VN_ISREG(vp
)) {
508 code
= XFS_ERROR(EINVAL
);
512 * Make sure that the dquots are attached to the inode.
514 if ((code
= XFS_QM_DQATTACH(mp
, ip
, XFS_QMOPT_ILOCKED
)))
519 * Change file access or modified times.
521 if (mask
& (XFS_AT_ATIME
|XFS_AT_MTIME
)) {
523 if ((flags
& ATTR_UTIME
) &&
524 !capable(CAP_FOWNER
)) {
525 code
= XFS_ERROR(EPERM
);
532 * Change extent size or realtime flag.
534 if (mask
& (XFS_AT_EXTSIZE
|XFS_AT_XFLAGS
)) {
536 * Can't change extent size if any extents are allocated.
538 if (ip
->i_d
.di_nextents
&& (mask
& XFS_AT_EXTSIZE
) &&
539 ((ip
->i_d
.di_extsize
<< mp
->m_sb
.sb_blocklog
) !=
541 code
= XFS_ERROR(EINVAL
); /* EFBIG? */
546 * Can't change realtime flag if any extents are allocated.
548 if ((ip
->i_d
.di_nextents
|| ip
->i_delayed_blks
) &&
549 (mask
& XFS_AT_XFLAGS
) &&
550 (ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) !=
551 (vap
->va_xflags
& XFS_XFLAG_REALTIME
)) {
552 code
= XFS_ERROR(EINVAL
); /* EFBIG? */
556 * Extent size must be a multiple of the appropriate block
557 * size, if set at all.
559 if ((mask
& XFS_AT_EXTSIZE
) && vap
->va_extsize
!= 0) {
562 if ((ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
) ||
563 ((mask
& XFS_AT_XFLAGS
) &&
564 (vap
->va_xflags
& XFS_XFLAG_REALTIME
))) {
565 size
= mp
->m_sb
.sb_rextsize
<<
566 mp
->m_sb
.sb_blocklog
;
568 size
= mp
->m_sb
.sb_blocksize
;
570 if (vap
->va_extsize
% size
) {
571 code
= XFS_ERROR(EINVAL
);
576 * If realtime flag is set then must have realtime data.
578 if ((mask
& XFS_AT_XFLAGS
) &&
579 (vap
->va_xflags
& XFS_XFLAG_REALTIME
)) {
580 if ((mp
->m_sb
.sb_rblocks
== 0) ||
581 (mp
->m_sb
.sb_rextsize
== 0) ||
582 (ip
->i_d
.di_extsize
% mp
->m_sb
.sb_rextsize
)) {
583 code
= XFS_ERROR(EINVAL
);
589 * Can't modify an immutable/append-only file unless
590 * we have appropriate permission.
592 if ((mask
& XFS_AT_XFLAGS
) &&
594 (XFS_DIFLAG_IMMUTABLE
|XFS_DIFLAG_APPEND
) ||
596 (XFS_XFLAG_IMMUTABLE
| XFS_XFLAG_APPEND
))) &&
597 !capable(CAP_LINUX_IMMUTABLE
)) {
598 code
= XFS_ERROR(EPERM
);
604 * Now we can make the changes. Before we join the inode
605 * to the transaction, if XFS_AT_SIZE is set then take care of
606 * the part of the truncation that must be done without the
607 * inode lock. This needs to be done before joining the inode
608 * to the transaction, because the inode cannot be unlocked
609 * once it is a part of the transaction.
611 if (mask
& XFS_AT_SIZE
) {
613 if ((vap
->va_size
> ip
->i_d
.di_size
) &&
614 (flags
& ATTR_NOSIZETOK
) == 0) {
615 code
= xfs_igrow_start(ip
, vap
->va_size
, credp
);
617 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
619 code
= xfs_itruncate_data(ip
, vap
->va_size
);
622 lock_flags
&= ~XFS_ILOCK_EXCL
;
623 ASSERT(lock_flags
== XFS_IOLOCK_EXCL
);
626 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
627 if ((code
= xfs_trans_reserve(tp
, 0,
628 XFS_ITRUNCATE_LOG_RES(mp
), 0,
629 XFS_TRANS_PERM_LOG_RES
,
630 XFS_ITRUNCATE_LOG_COUNT
))) {
631 xfs_trans_cancel(tp
, 0);
633 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
636 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
637 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
641 xfs_trans_ijoin(tp
, ip
, lock_flags
);
642 xfs_trans_ihold(tp
, ip
);
645 /* determine whether mandatory locking mode changes */
646 mandlock_before
= MANDLOCK(vp
, ip
->i_d
.di_mode
);
649 * Truncate file. Must have write permission and not be a directory.
651 if (mask
& XFS_AT_SIZE
) {
652 if (vap
->va_size
> ip
->i_d
.di_size
) {
653 xfs_igrow_finish(tp
, ip
, vap
->va_size
,
654 !(flags
& ATTR_DMI
));
655 } else if ((vap
->va_size
<= ip
->i_d
.di_size
) ||
656 ((vap
->va_size
== 0) && ip
->i_d
.di_nextents
)) {
658 * signal a sync transaction unless
659 * we're truncating an already unlinked
660 * file on a wsync filesystem
662 code
= xfs_itruncate_finish(&tp
, ip
,
663 (xfs_fsize_t
)vap
->va_size
,
665 ((ip
->i_d
.di_nlink
!= 0 ||
666 !(mp
->m_flags
& XFS_MOUNT_WSYNC
))
673 * Have to do this even if the file's size doesn't change.
675 timeflags
|= XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
;
679 * Change file access modes.
681 if (mask
& XFS_AT_MODE
) {
682 ip
->i_d
.di_mode
&= S_IFMT
;
683 ip
->i_d
.di_mode
|= vap
->va_mode
& ~S_IFMT
;
685 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
686 timeflags
|= XFS_ICHGTIME_CHG
;
690 * Change file ownership. Must be the owner or privileged.
691 * If the system was configured with the "restricted_chown"
692 * option, the owner is not permitted to give away the file,
693 * and can change the group id only to a group of which he
694 * or she is a member.
696 if (mask
& (XFS_AT_UID
|XFS_AT_GID
|XFS_AT_PROJID
)) {
698 * CAP_FSETID overrides the following restrictions:
700 * The set-user-ID and set-group-ID bits of a file will be
701 * cleared upon successful return from chown()
703 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
704 !capable(CAP_FSETID
)) {
705 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
709 * Change the ownerships and register quota modifications
710 * in the transaction.
713 if (XFS_IS_UQUOTA_ON(mp
)) {
714 ASSERT(mask
& XFS_AT_UID
);
716 olddquot1
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
717 &ip
->i_udquot
, udqp
);
719 ip
->i_d
.di_uid
= uid
;
722 if (XFS_IS_GQUOTA_ON(mp
)) {
723 ASSERT(!XFS_IS_PQUOTA_ON(mp
));
724 ASSERT(mask
& XFS_AT_GID
);
726 olddquot2
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
727 &ip
->i_gdquot
, gdqp
);
729 ip
->i_d
.di_gid
= gid
;
731 if (iprojid
!= projid
) {
732 if (XFS_IS_PQUOTA_ON(mp
)) {
733 ASSERT(!XFS_IS_GQUOTA_ON(mp
));
734 ASSERT(mask
& XFS_AT_PROJID
);
736 olddquot2
= XFS_QM_DQVOPCHOWN(mp
, tp
, ip
,
737 &ip
->i_gdquot
, gdqp
);
739 ip
->i_d
.di_projid
= projid
;
741 * We may have to rev the inode as well as
742 * the superblock version number since projids didn't
743 * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
745 if (ip
->i_d
.di_version
== XFS_DINODE_VERSION_1
)
746 xfs_bump_ino_vers2(tp
, ip
);
749 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
750 timeflags
|= XFS_ICHGTIME_CHG
;
755 * Change file access or modified times.
757 if (mask
& (XFS_AT_ATIME
|XFS_AT_MTIME
)) {
758 if (mask
& XFS_AT_ATIME
) {
759 ip
->i_d
.di_atime
.t_sec
= vap
->va_atime
.tv_sec
;
760 ip
->i_d
.di_atime
.t_nsec
= vap
->va_atime
.tv_nsec
;
761 ip
->i_update_core
= 1;
762 timeflags
&= ~XFS_ICHGTIME_ACC
;
764 if (mask
& XFS_AT_MTIME
) {
765 ip
->i_d
.di_mtime
.t_sec
= vap
->va_mtime
.tv_sec
;
766 ip
->i_d
.di_mtime
.t_nsec
= vap
->va_mtime
.tv_nsec
;
767 timeflags
&= ~XFS_ICHGTIME_MOD
;
768 timeflags
|= XFS_ICHGTIME_CHG
;
770 if (tp
&& (flags
& ATTR_UTIME
))
771 xfs_trans_log_inode (tp
, ip
, XFS_ILOG_CORE
);
775 * Change XFS-added attributes.
777 if (mask
& (XFS_AT_EXTSIZE
|XFS_AT_XFLAGS
)) {
778 if (mask
& XFS_AT_EXTSIZE
) {
780 * Converting bytes to fs blocks.
782 ip
->i_d
.di_extsize
= vap
->va_extsize
>>
783 mp
->m_sb
.sb_blocklog
;
785 if (mask
& XFS_AT_XFLAGS
) {
788 /* can't set PREALLOC this way, just preserve it */
789 di_flags
= (ip
->i_d
.di_flags
& XFS_DIFLAG_PREALLOC
);
790 if (vap
->va_xflags
& XFS_XFLAG_IMMUTABLE
)
791 di_flags
|= XFS_DIFLAG_IMMUTABLE
;
792 if (vap
->va_xflags
& XFS_XFLAG_APPEND
)
793 di_flags
|= XFS_DIFLAG_APPEND
;
794 if (vap
->va_xflags
& XFS_XFLAG_SYNC
)
795 di_flags
|= XFS_DIFLAG_SYNC
;
796 if (vap
->va_xflags
& XFS_XFLAG_NOATIME
)
797 di_flags
|= XFS_DIFLAG_NOATIME
;
798 if (vap
->va_xflags
& XFS_XFLAG_NODUMP
)
799 di_flags
|= XFS_DIFLAG_NODUMP
;
800 if (vap
->va_xflags
& XFS_XFLAG_PROJINHERIT
)
801 di_flags
|= XFS_DIFLAG_PROJINHERIT
;
802 if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFDIR
) {
803 if (vap
->va_xflags
& XFS_XFLAG_RTINHERIT
)
804 di_flags
|= XFS_DIFLAG_RTINHERIT
;
805 if (vap
->va_xflags
& XFS_XFLAG_NOSYMLINKS
)
806 di_flags
|= XFS_DIFLAG_NOSYMLINKS
;
807 if (vap
->va_xflags
& XFS_XFLAG_EXTSZINHERIT
)
808 di_flags
|= XFS_DIFLAG_EXTSZINHERIT
;
809 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) {
810 if (vap
->va_xflags
& XFS_XFLAG_REALTIME
) {
811 di_flags
|= XFS_DIFLAG_REALTIME
;
812 ip
->i_iocore
.io_flags
|= XFS_IOCORE_RT
;
814 ip
->i_iocore
.io_flags
&= ~XFS_IOCORE_RT
;
816 if (vap
->va_xflags
& XFS_XFLAG_EXTSIZE
)
817 di_flags
|= XFS_DIFLAG_EXTSIZE
;
819 ip
->i_d
.di_flags
= di_flags
;
821 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
822 timeflags
|= XFS_ICHGTIME_CHG
;
826 * Change file inode change time only if XFS_AT_CTIME set
827 * AND we have been called by a DMI function.
830 if ( (flags
& ATTR_DMI
) && (mask
& XFS_AT_CTIME
) ) {
831 ip
->i_d
.di_ctime
.t_sec
= vap
->va_ctime
.tv_sec
;
832 ip
->i_d
.di_ctime
.t_nsec
= vap
->va_ctime
.tv_nsec
;
833 ip
->i_update_core
= 1;
834 timeflags
&= ~XFS_ICHGTIME_CHG
;
838 * Send out timestamp changes that need to be set to the
839 * current time. Not done when called by a DMI function.
841 if (timeflags
&& !(flags
& ATTR_DMI
))
842 xfs_ichgtime(ip
, timeflags
);
844 XFS_STATS_INC(xs_ig_attrchg
);
847 * If this is a synchronous mount, make sure that the
848 * transaction goes to disk before returning to the user.
849 * This is slightly sub-optimal in that truncates require
850 * two sync transactions instead of one for wsync filesytems.
851 * One for the truncate and one for the timestamps since we
852 * don't want to change the timestamps unless we're sure the
853 * truncate worked. Truncates are less than 1% of the laddis
854 * mix so this probably isn't worth the trouble to optimize.
858 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
859 xfs_trans_set_sync(tp
);
861 code
= xfs_trans_commit(tp
, commit_flags
, NULL
);
865 * If the (regular) file's mandatory locking mode changed, then
866 * notify the vnode. We do this under the inode lock to prevent
867 * racing calls to vop_vnode_change.
869 mandlock_after
= MANDLOCK(vp
, ip
->i_d
.di_mode
);
870 if (mandlock_before
!= mandlock_after
) {
871 VOP_VNODE_CHANGE(vp
, VCHANGE_FLAGS_ENF_LOCKING
,
875 xfs_iunlock(ip
, lock_flags
);
878 * Release any dquot(s) the inode had kept before chown.
880 XFS_QM_DQRELE(mp
, olddquot1
);
881 XFS_QM_DQRELE(mp
, olddquot2
);
882 XFS_QM_DQRELE(mp
, udqp
);
883 XFS_QM_DQRELE(mp
, gdqp
);
889 if (DM_EVENT_ENABLED(vp
->v_vfsp
, ip
, DM_EVENT_ATTRIBUTE
) &&
890 !(flags
& ATTR_DMI
)) {
891 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_ATTRIBUTE
, vp
, DM_RIGHT_NULL
,
892 NULL
, DM_RIGHT_NULL
, NULL
, NULL
,
893 0, 0, AT_DELAY_FLAG(flags
));
898 commit_flags
|= XFS_TRANS_ABORT
;
901 XFS_QM_DQRELE(mp
, udqp
);
902 XFS_QM_DQRELE(mp
, gdqp
);
904 xfs_trans_cancel(tp
, commit_flags
);
906 if (lock_flags
!= 0) {
907 xfs_iunlock(ip
, lock_flags
);
915 * Null conversion from vnode mode bits to inode mode bits, as in efs.
926 vn_trace_entry(BHV_TO_VNODE(bdp
), __FUNCTION__
,
927 (inst_t
*)__return_address
);
929 ip
= XFS_BHVTOI(bdp
);
930 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
931 error
= xfs_iaccess(ip
, mode
, credp
);
932 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
956 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
962 vp
= BHV_TO_VNODE(bdp
);
963 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
965 ip
= XFS_BHVTOI(bdp
);
968 if (XFS_FORCED_SHUTDOWN(mp
))
969 return XFS_ERROR(EIO
);
971 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
973 ASSERT((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
);
975 offset
= uiop
->uio_offset
;
976 count
= uiop
->uio_resid
;
979 error
= XFS_ERROR(EINVAL
);
988 * See if the symlink is stored inline.
990 pathlen
= (int)ip
->i_d
.di_size
;
992 if (ip
->i_df
.if_flags
& XFS_IFINLINE
) {
993 error
= uio_read(ip
->i_df
.if_u1
.if_data
, pathlen
, uiop
);
997 * Symlink not inline. Call bmap to get it in.
999 nmaps
= SYMLINK_MAPS
;
1001 error
= xfs_bmapi(NULL
, ip
, 0, XFS_B_TO_FSB(mp
, pathlen
),
1002 0, NULL
, 0, mval
, &nmaps
, NULL
);
1008 for (n
= 0; n
< nmaps
; n
++) {
1009 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
1010 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
1011 bp
= xfs_buf_read(mp
->m_ddev_targp
, d
,
1012 BTOBB(byte_cnt
), 0);
1013 error
= XFS_BUF_GETERROR(bp
);
1015 xfs_ioerror_alert("xfs_readlink",
1016 ip
->i_mount
, bp
, XFS_BUF_ADDR(bp
));
1020 if (pathlen
< byte_cnt
)
1022 pathlen
-= byte_cnt
;
1024 error
= uio_read(XFS_BUF_PTR(bp
), byte_cnt
, uiop
);
1031 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1039 * This is called to sync the inode and its data out to disk.
1040 * We need to hold the I/O lock while flushing the data, and
1041 * the inode lock while flushing the inode. The inode lock CANNOT
1042 * be held while flushing the data, so acquire after we're done
1056 int log_flushed
= 0, changed
= 1;
1058 vn_trace_entry(BHV_TO_VNODE(bdp
),
1059 __FUNCTION__
, (inst_t
*)__return_address
);
1061 ip
= XFS_BHVTOI(bdp
);
1063 ASSERT(start
>= 0 && stop
>= -1);
1065 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
1066 return XFS_ERROR(EIO
);
1069 * We always need to make sure that the required inode state
1070 * is safe on disk. The vnode might be clean but because
1071 * of committed transactions that haven't hit the disk yet.
1072 * Likewise, there could be unflushed non-transactional
1073 * changes to the inode core that have to go to disk.
1075 * The following code depends on one assumption: that
1076 * any transaction that changes an inode logs the core
1077 * because it has to change some field in the inode core
1078 * (typically nextents or nblocks). That assumption
1079 * implies that any transactions against an inode will
1080 * catch any non-transactional updates. If inode-altering
1081 * transactions exist that violate this assumption, the
1082 * code breaks. Right now, it figures that if the involved
1083 * update_* field is clear and the inode is unpinned, the
1084 * inode is clean. Either it's been flushed or it's been
1085 * committed and the commit has hit the disk unpinning the inode.
1086 * (Note that xfs_inode_item_format() called at commit clears
1087 * the update_* fields.)
1089 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1091 /* If we are flushing data then we care about update_size
1092 * being set, otherwise we care about update_core
1094 if ((flag
& FSYNC_DATA
) ?
1095 (ip
->i_update_size
== 0) :
1096 (ip
->i_update_core
== 0)) {
1098 * Timestamps/size haven't changed since last inode
1099 * flush or inode transaction commit. That means
1100 * either nothing got written or a transaction
1101 * committed which caught the updates. If the
1102 * latter happened and the transaction hasn't
1103 * hit the disk yet, the inode will be still
1104 * be pinned. If it is, force the log.
1107 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1109 if (xfs_ipincount(ip
)) {
1110 _xfs_log_force(ip
->i_mount
, (xfs_lsn_t
)0,
1112 ((flag
& FSYNC_WAIT
)
1113 ? XFS_LOG_SYNC
: 0),
1117 * If the inode is not pinned and nothing
1118 * has changed we don't need to flush the
1126 * Kick off a transaction to log the inode
1127 * core to get the updates. Make it
1128 * sync if FSYNC_WAIT is passed in (which
1129 * is done by everybody but specfs). The
1130 * sync transaction will also force the log.
1132 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1133 tp
= xfs_trans_alloc(ip
->i_mount
, XFS_TRANS_FSYNC_TS
);
1134 if ((error
= xfs_trans_reserve(tp
, 0,
1135 XFS_FSYNC_TS_LOG_RES(ip
->i_mount
),
1137 xfs_trans_cancel(tp
, 0);
1140 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1143 * Note - it's possible that we might have pushed
1144 * ourselves out of the way during trans_reserve
1145 * which would flush the inode. But there's no
1146 * guarantee that the inode buffer has actually
1147 * gone out yet (it's delwri). Plus the buffer
1148 * could be pinned anyway if it's part of an
1149 * inode in another recent transaction. So we
1150 * play it safe and fire off the transaction anyway.
1152 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1153 xfs_trans_ihold(tp
, ip
);
1154 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1155 if (flag
& FSYNC_WAIT
)
1156 xfs_trans_set_sync(tp
);
1157 error
= _xfs_trans_commit(tp
, 0, NULL
, &log_flushed
);
1159 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1162 if ((ip
->i_mount
->m_flags
& XFS_MOUNT_BARRIER
) && changed
) {
1164 * If the log write didn't issue an ordered tag we need
1165 * to flush the disk cache for the data device now.
1168 xfs_blkdev_issue_flush(ip
->i_mount
->m_ddev_targp
);
1171 * If this inode is on the RT dev we need to flush that
1174 if (ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
)
1175 xfs_blkdev_issue_flush(ip
->i_mount
->m_rtdev_targp
);
1182 * This is called by xfs_inactive to free any blocks beyond eof,
1183 * when the link count isn't zero.
1186 xfs_inactive_free_eofblocks(
1192 xfs_fileoff_t end_fsb
;
1193 xfs_fileoff_t last_fsb
;
1194 xfs_filblks_t map_len
;
1196 xfs_bmbt_irec_t imap
;
1199 * Figure out if there are any blocks beyond the end
1200 * of the file. If not, then there is nothing to do.
1202 end_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)ip
->i_d
.di_size
));
1203 last_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
1204 map_len
= last_fsb
- end_fsb
;
1209 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
1210 error
= xfs_bmapi(NULL
, ip
, end_fsb
, map_len
, 0,
1211 NULL
, 0, &imap
, &nimaps
, NULL
);
1212 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
1214 if (!error
&& (nimaps
!= 0) &&
1215 (imap
.br_startblock
!= HOLESTARTBLOCK
||
1216 ip
->i_delayed_blks
)) {
1218 * Attach the dquots to the inode up front.
1220 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
1224 * There are blocks after the end of file.
1225 * Free them up now by truncating the file to
1228 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1231 * Do the xfs_itruncate_start() call before
1232 * reserving any log space because
1233 * itruncate_start will call into the buffer
1234 * cache and we can't
1235 * do that within a transaction.
1237 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1238 xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
,
1241 error
= xfs_trans_reserve(tp
, 0,
1242 XFS_ITRUNCATE_LOG_RES(mp
),
1243 0, XFS_TRANS_PERM_LOG_RES
,
1244 XFS_ITRUNCATE_LOG_COUNT
);
1246 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1247 xfs_trans_cancel(tp
, 0);
1248 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1252 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1253 xfs_trans_ijoin(tp
, ip
,
1256 xfs_trans_ihold(tp
, ip
);
1258 error
= xfs_itruncate_finish(&tp
, ip
,
1263 * If we get an error at this point we
1264 * simply don't bother truncating the file.
1267 xfs_trans_cancel(tp
,
1268 (XFS_TRANS_RELEASE_LOG_RES
|
1271 error
= xfs_trans_commit(tp
,
1272 XFS_TRANS_RELEASE_LOG_RES
,
1275 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1281 * Free a symlink that has blocks associated with it.
1284 xfs_inactive_symlink_rmt(
1292 xfs_fsblock_t first_block
;
1293 xfs_bmap_free_t free_list
;
1296 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
1304 ASSERT(ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
));
1306 * We're freeing a symlink that has some
1307 * blocks allocated to it. Free the
1308 * blocks here. We know that we've got
1309 * either 1 or 2 extents and that we can
1310 * free them all in one bunmapi call.
1312 ASSERT(ip
->i_d
.di_nextents
> 0 && ip
->i_d
.di_nextents
<= 2);
1313 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1314 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
1315 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1316 xfs_trans_cancel(tp
, 0);
1321 * Lock the inode, fix the size, and join it to the transaction.
1322 * Hold it so in the normal path, we still have it locked for
1323 * the second transaction. In the error paths we need it
1324 * held so the cancel won't rele it, see below.
1326 xfs_ilock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1327 size
= (int)ip
->i_d
.di_size
;
1328 ip
->i_d
.di_size
= 0;
1329 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1330 xfs_trans_ihold(tp
, ip
);
1331 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1333 * Find the block(s) so we can inval and unmap them.
1336 XFS_BMAP_INIT(&free_list
, &first_block
);
1337 nmaps
= sizeof(mval
) / sizeof(mval
[0]);
1338 if ((error
= xfs_bmapi(tp
, ip
, 0, XFS_B_TO_FSB(mp
, size
),
1339 XFS_BMAPI_METADATA
, &first_block
, 0, mval
, &nmaps
,
1343 * Invalidate the block(s).
1345 for (i
= 0; i
< nmaps
; i
++) {
1346 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
1347 XFS_FSB_TO_DADDR(mp
, mval
[i
].br_startblock
),
1348 XFS_FSB_TO_BB(mp
, mval
[i
].br_blockcount
), 0);
1349 xfs_trans_binval(tp
, bp
);
1352 * Unmap the dead block(s) to the free_list.
1354 if ((error
= xfs_bunmapi(tp
, ip
, 0, size
, XFS_BMAPI_METADATA
, nmaps
,
1355 &first_block
, &free_list
, &done
)))
1359 * Commit the first transaction. This logs the EFI and the inode.
1361 if ((error
= xfs_bmap_finish(&tp
, &free_list
, first_block
, &committed
)))
1364 * The transaction must have been committed, since there were
1365 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
1366 * The new tp has the extent freeing and EFDs.
1370 * The first xact was committed, so add the inode to the new one.
1371 * Mark it dirty so it will be logged and moved forward in the log as
1372 * part of every commit.
1374 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1375 xfs_trans_ihold(tp
, ip
);
1376 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1378 * Get a new, empty transaction to return to our caller.
1380 ntp
= xfs_trans_dup(tp
);
1382 * Commit the transaction containing extent freeing and EFD's.
1383 * If we get an error on the commit here or on the reserve below,
1384 * we need to unlock the inode since the new transaction doesn't
1385 * have the inode attached.
1387 error
= xfs_trans_commit(tp
, 0, NULL
);
1390 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1394 * Remove the memory for extent descriptions (just bookkeeping).
1396 if (ip
->i_df
.if_bytes
)
1397 xfs_idata_realloc(ip
, -ip
->i_df
.if_bytes
, XFS_DATA_FORK
);
1398 ASSERT(ip
->i_df
.if_bytes
== 0);
1400 * Put an itruncate log reservation in the new transaction
1403 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
1404 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
1405 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1409 * Return with the inode locked but not joined to the transaction.
1415 xfs_bmap_cancel(&free_list
);
1418 * Have to come here with the inode locked and either
1419 * (held and in the transaction) or (not in the transaction).
1420 * If the inode isn't held then cancel would iput it, but
1421 * that's wrong since this is inactive and the vnode ref
1422 * count is 0 already.
1423 * Cancel won't do anything to the inode if held, but it still
1424 * needs to be locked until the cancel is done, if it was
1425 * joined to the transaction.
1427 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1428 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1435 xfs_inactive_symlink_local(
1441 ASSERT(ip
->i_d
.di_size
<= XFS_IFORK_DSIZE(ip
));
1443 * We're freeing a symlink which fit into
1444 * the inode. Just free the memory used
1445 * to hold the old symlink.
1447 error
= xfs_trans_reserve(*tpp
, 0,
1448 XFS_ITRUNCATE_LOG_RES(ip
->i_mount
),
1449 0, XFS_TRANS_PERM_LOG_RES
,
1450 XFS_ITRUNCATE_LOG_COUNT
);
1453 xfs_trans_cancel(*tpp
, 0);
1457 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1460 * Zero length symlinks _can_ exist.
1462 if (ip
->i_df
.if_bytes
> 0) {
1463 xfs_idata_realloc(ip
,
1464 -(ip
->i_df
.if_bytes
),
1466 ASSERT(ip
->i_df
.if_bytes
== 0);
1483 ASSERT(ismrlocked(&ip
->i_iolock
, MR_UPDATE
));
1486 ASSERT(ip
->i_d
.di_forkoff
!= 0);
1487 xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
, NULL
);
1488 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1490 error
= xfs_attr_inactive(ip
);
1493 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1494 return error
; /* goto out */
1497 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1498 error
= xfs_trans_reserve(tp
, 0,
1499 XFS_IFREE_LOG_RES(mp
),
1500 0, XFS_TRANS_PERM_LOG_RES
,
1501 XFS_INACTIVE_LOG_COUNT
);
1503 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1504 xfs_trans_cancel(tp
, 0);
1506 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1510 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1511 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1512 xfs_trans_ihold(tp
, ip
);
1513 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1515 ASSERT(ip
->i_d
.di_anextents
== 0);
1530 vp
= BHV_TO_VNODE(bdp
);
1531 ip
= XFS_BHVTOI(bdp
);
1533 if (!VN_ISREG(vp
) || (ip
->i_d
.di_mode
== 0)) {
1537 /* If this is a read-only mount, don't do this (would generate I/O) */
1538 if (vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
)
1541 #ifdef HAVE_REFCACHE
1542 /* If we are in the NFS reference cache then don't do this now */
1549 if (ip
->i_d
.di_nlink
!= 0) {
1550 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1551 ((ip
->i_d
.di_size
> 0) || (VN_CACHED(vp
) > 0 ||
1552 ip
->i_delayed_blks
> 0)) &&
1553 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
)) &&
1554 (!(ip
->i_d
.di_flags
&
1555 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)))) {
1556 if ((error
= xfs_inactive_free_eofblocks(mp
, ip
)))
1558 /* Update linux inode block count after free above */
1559 LINVFS_GET_IP(vp
)->i_blocks
= XFS_FSB_TO_BB(mp
,
1560 ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
1570 * This is called when the vnode reference count for the vnode
1571 * goes to zero. If the file has been unlinked, then it must
1572 * now be truncated. Also, we clear all of the read-ahead state
1573 * kept for the inode here since the file is now closed.
1582 xfs_bmap_free_t free_list
;
1583 xfs_fsblock_t first_block
;
1590 vp
= BHV_TO_VNODE(bdp
);
1591 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
1593 ip
= XFS_BHVTOI(bdp
);
1596 * If the inode is already free, then there can be nothing
1599 if (ip
->i_d
.di_mode
== 0 || VN_BAD(vp
)) {
1600 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1601 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1602 return VN_INACTIVE_CACHE
;
1606 * Only do a truncate if it's a regular file with
1607 * some actual space in it. It's OK to look at the
1608 * inode's fields without the lock because we're the
1609 * only one with a reference to the inode.
1611 truncate
= ((ip
->i_d
.di_nlink
== 0) &&
1612 ((ip
->i_d
.di_size
!= 0) || (ip
->i_d
.di_nextents
> 0) ||
1613 (ip
->i_delayed_blks
> 0)) &&
1614 ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
));
1618 if (ip
->i_d
.di_nlink
== 0 &&
1619 DM_EVENT_ENABLED(vp
->v_vfsp
, ip
, DM_EVENT_DESTROY
)) {
1620 (void) XFS_SEND_DESTROY(mp
, vp
, DM_RIGHT_NULL
);
1625 /* If this is a read-only mount, don't do this (would generate I/O) */
1626 if (vp
->v_vfsp
->vfs_flag
& VFS_RDONLY
)
1629 if (ip
->i_d
.di_nlink
!= 0) {
1630 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1631 ((ip
->i_d
.di_size
> 0) || (VN_CACHED(vp
) > 0 ||
1632 ip
->i_delayed_blks
> 0)) &&
1633 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) &&
1634 (!(ip
->i_d
.di_flags
&
1635 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)) ||
1636 (ip
->i_delayed_blks
!= 0)))) {
1637 if ((error
= xfs_inactive_free_eofblocks(mp
, ip
)))
1638 return VN_INACTIVE_CACHE
;
1639 /* Update linux inode block count after free above */
1640 LINVFS_GET_IP(vp
)->i_blocks
= XFS_FSB_TO_BB(mp
,
1641 ip
->i_d
.di_nblocks
+ ip
->i_delayed_blks
);
1646 ASSERT(ip
->i_d
.di_nlink
== 0);
1648 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
1649 return VN_INACTIVE_CACHE
;
1651 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1654 * Do the xfs_itruncate_start() call before
1655 * reserving any log space because itruncate_start
1656 * will call into the buffer cache and we can't
1657 * do that within a transaction.
1659 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1661 xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
, 0);
1663 error
= xfs_trans_reserve(tp
, 0,
1664 XFS_ITRUNCATE_LOG_RES(mp
),
1665 0, XFS_TRANS_PERM_LOG_RES
,
1666 XFS_ITRUNCATE_LOG_COUNT
);
1668 /* Don't call itruncate_cleanup */
1669 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1670 xfs_trans_cancel(tp
, 0);
1671 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1672 return VN_INACTIVE_CACHE
;
1675 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1676 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1677 xfs_trans_ihold(tp
, ip
);
1680 * normally, we have to run xfs_itruncate_finish sync.
1681 * But if filesystem is wsync and we're in the inactive
1682 * path, then we know that nlink == 0, and that the
1683 * xaction that made nlink == 0 is permanently committed
1684 * since xfs_remove runs as a synchronous transaction.
1686 error
= xfs_itruncate_finish(&tp
, ip
, 0, XFS_DATA_FORK
,
1687 (!(mp
->m_flags
& XFS_MOUNT_WSYNC
) ? 1 : 0));
1690 xfs_trans_cancel(tp
,
1691 XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1692 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1693 return VN_INACTIVE_CACHE
;
1695 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
) {
1698 * If we get an error while cleaning up a
1699 * symlink we bail out.
1701 error
= (ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
)) ?
1702 xfs_inactive_symlink_rmt(ip
, &tp
) :
1703 xfs_inactive_symlink_local(ip
, &tp
);
1707 return VN_INACTIVE_CACHE
;
1710 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1711 xfs_trans_ihold(tp
, ip
);
1713 error
= xfs_trans_reserve(tp
, 0,
1714 XFS_IFREE_LOG_RES(mp
),
1715 0, XFS_TRANS_PERM_LOG_RES
,
1716 XFS_INACTIVE_LOG_COUNT
);
1718 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1719 xfs_trans_cancel(tp
, 0);
1720 return VN_INACTIVE_CACHE
;
1723 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1724 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1725 xfs_trans_ihold(tp
, ip
);
1729 * If there are attributes associated with the file
1730 * then blow them away now. The code calls a routine
1731 * that recursively deconstructs the attribute fork.
1732 * We need to just commit the current transaction
1733 * because we can't use it for xfs_attr_inactive().
1735 if (ip
->i_d
.di_anextents
> 0) {
1736 error
= xfs_inactive_attrs(ip
, &tp
);
1738 * If we got an error, the transaction is already
1739 * cancelled, and the inode is unlocked. Just get out.
1742 return VN_INACTIVE_CACHE
;
1743 } else if (ip
->i_afp
) {
1744 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1750 XFS_BMAP_INIT(&free_list
, &first_block
);
1751 error
= xfs_ifree(tp
, ip
, &free_list
);
1754 * If we fail to free the inode, shut down. The cancel
1755 * might do that, we need to make sure. Otherwise the
1756 * inode might be lost for a long time or forever.
1758 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1760 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1761 error
, mp
->m_fsname
);
1762 xfs_force_shutdown(mp
, XFS_METADATA_IO_ERROR
);
1764 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
1767 * Credit the quota account(s). The inode is gone.
1769 XFS_TRANS_MOD_DQUOT_BYINO(mp
, tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1772 * Just ignore errors at this point. There is
1773 * nothing we can do except to try to keep going.
1775 (void) xfs_bmap_finish(&tp
, &free_list
, first_block
,
1777 (void) xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
, NULL
);
1780 * Release the dquots held by inode, if any.
1782 XFS_QM_DQDETACH(mp
, ip
);
1784 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1787 return VN_INACTIVE_CACHE
;
1796 bhv_desc_t
*dir_bdp
,
1803 xfs_inode_t
*dp
, *ip
;
1809 dir_vp
= BHV_TO_VNODE(dir_bdp
);
1810 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
1812 dp
= XFS_BHVTOI(dir_bdp
);
1814 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
1815 return XFS_ERROR(EIO
);
1817 lock_mode
= xfs_ilock_map_shared(dp
);
1818 error
= xfs_dir_lookup_int(dir_bdp
, lock_mode
, dentry
, &e_inum
, &ip
);
1820 *vpp
= XFS_ITOV(ip
);
1823 xfs_iunlock_map_shared(dp
, lock_mode
);
1829 * xfs_create (create a new file).
1833 bhv_desc_t
*dir_bdp
,
1839 char *name
= VNAME(dentry
);
1841 xfs_inode_t
*dp
, *ip
;
1847 xfs_bmap_free_t free_list
;
1848 xfs_fsblock_t first_block
;
1849 boolean_t dp_joined_to_trans
;
1850 int dm_event_sent
= 0;
1854 struct xfs_dquot
*udqp
, *gdqp
;
1860 dir_vp
= BHV_TO_VNODE(dir_bdp
);
1861 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
1863 dp
= XFS_BHVTOI(dir_bdp
);
1866 dm_di_mode
= vap
->va_mode
;
1867 namelen
= VNAMELEN(dentry
);
1869 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_CREATE
)) {
1870 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_CREATE
,
1871 dir_vp
, DM_RIGHT_NULL
, NULL
,
1872 DM_RIGHT_NULL
, name
, NULL
,
1880 if (XFS_FORCED_SHUTDOWN(mp
))
1881 return XFS_ERROR(EIO
);
1883 /* Return through std_return after this point. */
1886 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1887 prid
= dp
->i_d
.di_projid
;
1888 else if (vap
->va_mask
& XFS_AT_PROJID
)
1889 prid
= (xfs_prid_t
)vap
->va_projid
;
1891 prid
= (xfs_prid_t
)dfltprid
;
1894 * Make sure that we have allocated dquot(s) on disk.
1896 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
1897 current_fsuid(credp
), current_fsgid(credp
), prid
,
1898 XFS_QMOPT_QUOTALL
|XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
1903 dp_joined_to_trans
= B_FALSE
;
1905 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1906 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1907 resblks
= XFS_CREATE_SPACE_RES(mp
, namelen
);
1909 * Initially assume that the file does not exist and
1910 * reserve the resources for that case. If that is not
1911 * the case we'll drop the one we have and get a more
1912 * appropriate transaction later.
1914 error
= xfs_trans_reserve(tp
, resblks
, XFS_CREATE_LOG_RES(mp
), 0,
1915 XFS_TRANS_PERM_LOG_RES
, XFS_CREATE_LOG_COUNT
);
1916 if (error
== ENOSPC
) {
1918 error
= xfs_trans_reserve(tp
, 0, XFS_CREATE_LOG_RES(mp
), 0,
1919 XFS_TRANS_PERM_LOG_RES
, XFS_CREATE_LOG_COUNT
);
1927 xfs_ilock(dp
, XFS_ILOCK_EXCL
);
1929 XFS_BMAP_INIT(&free_list
, &first_block
);
1934 * Reserve disk quota and the inode.
1936 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
1941 (error
= XFS_DIR_CANENTER(mp
, tp
, dp
, name
, namelen
)))
1943 rdev
= (vap
->va_mask
& XFS_AT_RDEV
) ? vap
->va_rdev
: 0;
1944 error
= xfs_dir_ialloc(&tp
, dp
, vap
->va_mode
, 1,
1945 rdev
, credp
, prid
, resblks
> 0,
1948 if (error
== ENOSPC
)
1955 * At this point, we've gotten a newly allocated inode.
1956 * It is locked (and joined to the transaction).
1959 ASSERT(ismrlocked (&ip
->i_lock
, MR_UPDATE
));
1962 * Now we join the directory inode to the transaction.
1963 * We do not do it earlier because xfs_dir_ialloc
1964 * might commit the previous transaction (and release
1969 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1970 dp_joined_to_trans
= B_TRUE
;
1972 error
= XFS_DIR_CREATENAME(mp
, tp
, dp
, name
, namelen
, ip
->i_ino
,
1973 &first_block
, &free_list
,
1974 resblks
? resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1976 ASSERT(error
!= ENOSPC
);
1979 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1980 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1983 * If this is a synchronous mount, make sure that the
1984 * create transaction goes to disk before returning to
1987 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
1988 xfs_trans_set_sync(tp
);
1994 * Attach the dquot(s) to the inodes and modify them incore.
1995 * These ids of the inode couldn't have changed since the new
1996 * inode has been locked ever since it was created.
1998 XFS_QM_DQVOPCREATE(mp
, tp
, ip
, udqp
, gdqp
);
2001 * xfs_trans_commit normally decrements the vnode ref count
2002 * when it unlocks the inode. Since we want to return the
2003 * vnode to the caller, we bump the vnode ref count now.
2008 error
= xfs_bmap_finish(&tp
, &free_list
, first_block
, &committed
);
2010 xfs_bmap_cancel(&free_list
);
2014 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
, NULL
);
2021 XFS_QM_DQRELE(mp
, udqp
);
2022 XFS_QM_DQRELE(mp
, gdqp
);
2025 * Propogate the fact that the vnode changed after the
2026 * xfs_inode locks have been released.
2028 VOP_VNODE_CHANGE(vp
, VCHANGE_FLAGS_TRUNCATED
, 3);
2032 /* Fallthrough to std_return with error = 0 */
2035 if ( (*vpp
|| (error
!= 0 && dm_event_sent
!= 0)) &&
2036 DM_EVENT_ENABLED(dir_vp
->v_vfsp
, XFS_BHVTOI(dir_bdp
),
2037 DM_EVENT_POSTCREATE
)) {
2038 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTCREATE
,
2039 dir_vp
, DM_RIGHT_NULL
,
2041 DM_RIGHT_NULL
, name
, NULL
,
2042 dm_di_mode
, error
, 0);
2047 cancel_flags
|= XFS_TRANS_ABORT
;
2052 xfs_trans_cancel(tp
, cancel_flags
);
2054 if (!dp_joined_to_trans
&& (dp
!= NULL
))
2055 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2056 XFS_QM_DQRELE(mp
, udqp
);
2057 XFS_QM_DQRELE(mp
, gdqp
);
2063 * Wait until after the current transaction is aborted to
2064 * release the inode. This prevents recursive transactions
2065 * and deadlocks from xfs_inactive.
2067 cancel_flags
|= XFS_TRANS_ABORT
;
2068 xfs_trans_cancel(tp
, cancel_flags
);
2071 XFS_QM_DQRELE(mp
, udqp
);
2072 XFS_QM_DQRELE(mp
, gdqp
);
2079 * Some counters to see if (and how often) we are hitting some deadlock
2080 * prevention code paths.
2084 int xfs_rm_lock_delays
;
2085 int xfs_rm_attempts
;
2089 * The following routine will lock the inodes associated with the
2090 * directory and the named entry in the directory. The locks are
2091 * acquired in increasing inode number.
2093 * If the entry is "..", then only the directory is locked. The
2094 * vnode ref count will still include that from the .. entry in
2097 * There is a deadlock we need to worry about. If the locked directory is
2098 * in the AIL, it might be blocking up the log. The next inode we lock
2099 * could be already locked by another thread waiting for log space (e.g
2100 * a permanent log reservation with a long running transaction (see
2101 * xfs_itruncate_finish)). To solve this, we must check if the directory
2102 * is in the ail and use lock_nowait. If we can't lock, we need to
2103 * drop the inode lock on the directory and try again. xfs_iunlock will
2104 * potentially push the tail if we were holding up the log.
2107 xfs_lock_dir_and_entry(
2110 xfs_inode_t
*ip
) /* inode of entry 'name' */
2114 xfs_inode_t
*ips
[2];
2123 xfs_ilock(dp
, XFS_ILOCK_EXCL
);
2130 * We want to lock in increasing inum. Since we've already
2131 * acquired the lock on the directory, we may need to release
2132 * if if the inum of the entry turns out to be less.
2134 if (e_inum
> dp
->i_ino
) {
2136 * We are already in the right order, so just
2137 * lock on the inode of the entry.
2138 * We need to use nowait if dp is in the AIL.
2141 lp
= (xfs_log_item_t
*)dp
->i_itemp
;
2142 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
2143 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
)) {
2150 * Unlock dp and try again.
2151 * xfs_iunlock will try to push the tail
2152 * if the inode is in the AIL.
2155 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2157 if ((attempts
% 5) == 0) {
2158 delay(1); /* Don't just spin the CPU */
2160 xfs_rm_lock_delays
++;
2166 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2168 } else if (e_inum
< dp
->i_ino
) {
2169 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2173 xfs_lock_inodes(ips
, 2, 0, XFS_ILOCK_EXCL
);
2175 /* else e_inum == dp->i_ino */
2176 /* This can happen if we're asked to lock /x/..
2177 * the entry is "..", which is also the parent directory.
2185 int xfs_small_retries
;
2186 int xfs_middle_retries
;
2187 int xfs_lots_retries
;
2188 int xfs_lock_delays
;
2192 * The following routine will lock n inodes in exclusive mode.
2193 * We assume the caller calls us with the inodes in i_ino order.
2195 * We need to detect deadlock where an inode that we lock
2196 * is in the AIL and we start waiting for another inode that is locked
2197 * by a thread in a long running transaction (such as truncate). This can
2198 * result in deadlock since the long running trans might need to wait
2199 * for the inode we just locked in order to push the tail and free space
2209 int attempts
= 0, i
, j
, try_lock
;
2212 ASSERT(ips
&& (inodes
>= 2)); /* we need at least two */
2223 for (; i
< inodes
; i
++) {
2226 if (i
&& (ips
[i
] == ips
[i
-1])) /* Already locked */
2230 * If try_lock is not set yet, make sure all locked inodes
2231 * are not in the AIL.
2232 * If any are, set try_lock to be used later.
2236 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
2237 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
2238 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
2245 * If any of the previous locks we have locked is in the AIL,
2246 * we must TRY to get the second and subsequent locks. If
2247 * we can't get any, we must release all we have
2252 /* try_lock must be 0 if i is 0. */
2254 * try_lock means we have an inode locked
2255 * that is in the AIL.
2258 if (!xfs_ilock_nowait(ips
[i
], lock_mode
)) {
2262 * Unlock all previous guys and try again.
2263 * xfs_iunlock will try to push the tail
2264 * if the inode is in the AIL.
2267 for(j
= i
- 1; j
>= 0; j
--) {
2270 * Check to see if we've already
2271 * unlocked this one.
2272 * Not the first one going back,
2273 * and the inode ptr is the same.
2275 if ((j
!= (i
- 1)) && ips
[j
] ==
2279 xfs_iunlock(ips
[j
], lock_mode
);
2282 if ((attempts
% 5) == 0) {
2283 delay(1); /* Don't just spin the CPU */
2293 xfs_ilock(ips
[i
], lock_mode
);
2299 if (attempts
< 5) xfs_small_retries
++;
2300 else if (attempts
< 100) xfs_middle_retries
++;
2301 else xfs_lots_retries
++;
2309 #define REMOVE_DEBUG_TRACE(x) {remove_which_error_return = (x);}
2310 int remove_which_error_return
= 0;
2312 #define REMOVE_DEBUG_TRACE(x)
2313 #endif /* ! DEBUG */
2322 bhv_desc_t
*dir_bdp
,
2327 char *name
= VNAME(dentry
);
2328 xfs_inode_t
*dp
, *ip
;
2329 xfs_trans_t
*tp
= NULL
;
2332 xfs_bmap_free_t free_list
;
2333 xfs_fsblock_t first_block
;
2341 dir_vp
= BHV_TO_VNODE(dir_bdp
);
2342 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2344 dp
= XFS_BHVTOI(dir_bdp
);
2347 if (XFS_FORCED_SHUTDOWN(mp
))
2348 return XFS_ERROR(EIO
);
2350 namelen
= VNAMELEN(dentry
);
2352 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_REMOVE
)) {
2353 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_REMOVE
, dir_vp
,
2354 DM_RIGHT_NULL
, NULL
, DM_RIGHT_NULL
,
2355 name
, NULL
, 0, 0, 0);
2360 /* From this point on, return through std_return */
2364 * We need to get a reference to ip before we get our log
2365 * reservation. The reason for this is that we cannot call
2366 * xfs_iget for an inode for which we do not have a reference
2367 * once we've acquired a log reservation. This is because the
2368 * inode we are trying to get might be in xfs_inactive going
2369 * for a log reservation. Since we'll have to wait for the
2370 * inactive code to complete before returning from xfs_iget,
2371 * we need to make sure that we don't have log space reserved
2372 * when we call xfs_iget. Instead we get an unlocked referece
2373 * to the inode before getting our log reservation.
2375 error
= xfs_get_dir_entry(dentry
, &ip
);
2377 REMOVE_DEBUG_TRACE(__LINE__
);
2381 dm_di_mode
= ip
->i_d
.di_mode
;
2383 vn_trace_entry(XFS_ITOV(ip
), __FUNCTION__
, (inst_t
*)__return_address
);
2387 error
= XFS_QM_DQATTACH(mp
, dp
, 0);
2388 if (!error
&& dp
!= ip
)
2389 error
= XFS_QM_DQATTACH(mp
, ip
, 0);
2391 REMOVE_DEBUG_TRACE(__LINE__
);
2396 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
2397 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2399 * We try to get the real space reservation first,
2400 * allowing for directory btree deletion(s) implying
2401 * possible bmap insert(s). If we can't get the space
2402 * reservation then we use 0 instead, and avoid the bmap
2403 * btree insert(s) in the directory code by, if the bmap
2404 * insert tries to happen, instead trimming the LAST
2405 * block from the directory.
2407 resblks
= XFS_REMOVE_SPACE_RES(mp
);
2408 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
2409 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
2410 if (error
== ENOSPC
) {
2412 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
2413 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
2416 ASSERT(error
!= ENOSPC
);
2417 REMOVE_DEBUG_TRACE(__LINE__
);
2418 xfs_trans_cancel(tp
, 0);
2423 error
= xfs_lock_dir_and_entry(dp
, dentry
, ip
);
2425 REMOVE_DEBUG_TRACE(__LINE__
);
2426 xfs_trans_cancel(tp
, cancel_flags
);
2432 * At this point, we've gotten both the directory and the entry
2435 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2438 * Increment vnode ref count only in this case since
2439 * there's an extra vnode reference in the case where
2443 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2447 * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2449 XFS_BMAP_INIT(&free_list
, &first_block
);
2450 error
= XFS_DIR_REMOVENAME(mp
, tp
, dp
, name
, namelen
, ip
->i_ino
,
2451 &first_block
, &free_list
, 0);
2453 ASSERT(error
!= ENOENT
);
2454 REMOVE_DEBUG_TRACE(__LINE__
);
2457 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2460 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2462 error
= xfs_droplink(tp
, ip
);
2464 REMOVE_DEBUG_TRACE(__LINE__
);
2468 /* Determine if this is the last link while
2469 * we are in the transaction.
2471 link_zero
= (ip
)->i_d
.di_nlink
==0;
2474 * Take an extra ref on the inode so that it doesn't
2475 * go to xfs_inactive() from within the commit.
2480 * If this is a synchronous mount, make sure that the
2481 * remove transaction goes to disk before returning to
2484 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2485 xfs_trans_set_sync(tp
);
2488 error
= xfs_bmap_finish(&tp
, &free_list
, first_block
, &committed
);
2490 REMOVE_DEBUG_TRACE(__LINE__
);
2494 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
, NULL
);
2501 * Before we drop our extra reference to the inode, purge it
2502 * from the refcache if it is there. By waiting until afterwards
2503 * to do the IRELE, we ensure that we won't go inactive in the
2504 * xfs_refcache_purge_ip routine (although that would be OK).
2506 xfs_refcache_purge_ip(ip
);
2508 vn_trace_exit(XFS_ITOV(ip
), __FUNCTION__
, (inst_t
*)__return_address
);
2511 * Let interposed file systems know about removed links.
2513 VOP_LINK_REMOVED(XFS_ITOV(ip
), dir_vp
, link_zero
);
2517 /* Fall through to std_return with error = 0 */
2519 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
,
2520 DM_EVENT_POSTREMOVE
)) {
2521 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTREMOVE
,
2522 dir_vp
, DM_RIGHT_NULL
,
2523 NULL
, DM_RIGHT_NULL
,
2524 name
, NULL
, dm_di_mode
, error
, 0);
2529 xfs_bmap_cancel(&free_list
);
2530 cancel_flags
|= XFS_TRANS_ABORT
;
2531 xfs_trans_cancel(tp
, cancel_flags
);
2536 * In this case make sure to not release the inode until after
2537 * the current transaction is aborted. Releasing it beforehand
2538 * can cause us to go to xfs_inactive and start a recursive
2539 * transaction which can easily deadlock with the current one.
2541 xfs_bmap_cancel(&free_list
);
2542 cancel_flags
|= XFS_TRANS_ABORT
;
2543 xfs_trans_cancel(tp
, cancel_flags
);
2546 * Before we drop our extra reference to the inode, purge it
2547 * from the refcache if it is there. By waiting until afterwards
2548 * to do the IRELE, we ensure that we won't go inactive in the
2549 * xfs_refcache_purge_ip routine (although that would be OK).
2551 xfs_refcache_purge_ip(ip
);
2565 bhv_desc_t
*target_dir_bdp
,
2570 xfs_inode_t
*tdp
, *sip
;
2573 xfs_inode_t
*ips
[2];
2575 xfs_bmap_free_t free_list
;
2576 xfs_fsblock_t first_block
;
2579 vnode_t
*target_dir_vp
;
2581 char *target_name
= VNAME(dentry
);
2584 target_dir_vp
= BHV_TO_VNODE(target_dir_bdp
);
2585 vn_trace_entry(target_dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2586 vn_trace_entry(src_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2588 target_namelen
= VNAMELEN(dentry
);
2589 if (VN_ISDIR(src_vp
))
2590 return XFS_ERROR(EPERM
);
2592 sip
= xfs_vtoi(src_vp
);
2593 tdp
= XFS_BHVTOI(target_dir_bdp
);
2595 if (XFS_FORCED_SHUTDOWN(mp
))
2596 return XFS_ERROR(EIO
);
2598 if (DM_EVENT_ENABLED(src_vp
->v_vfsp
, tdp
, DM_EVENT_LINK
)) {
2599 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_LINK
,
2600 target_dir_vp
, DM_RIGHT_NULL
,
2601 src_vp
, DM_RIGHT_NULL
,
2602 target_name
, NULL
, 0, 0, 0);
2607 /* Return through std_return after this point. */
2609 error
= XFS_QM_DQATTACH(mp
, sip
, 0);
2610 if (!error
&& sip
!= tdp
)
2611 error
= XFS_QM_DQATTACH(mp
, tdp
, 0);
2615 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
2616 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2617 resblks
= XFS_LINK_SPACE_RES(mp
, target_namelen
);
2618 error
= xfs_trans_reserve(tp
, resblks
, XFS_LINK_LOG_RES(mp
), 0,
2619 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2620 if (error
== ENOSPC
) {
2622 error
= xfs_trans_reserve(tp
, 0, XFS_LINK_LOG_RES(mp
), 0,
2623 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2630 if (sip
->i_ino
< tdp
->i_ino
) {
2638 xfs_lock_inodes(ips
, 2, 0, XFS_ILOCK_EXCL
);
2641 * Increment vnode ref counts since xfs_trans_commit &
2642 * xfs_trans_cancel will both unlock the inodes and
2643 * decrement the associated ref counts.
2646 VN_HOLD(target_dir_vp
);
2647 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
2648 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
2651 * If the source has too many links, we can't make any more to it.
2653 if (sip
->i_d
.di_nlink
>= XFS_MAXLINK
) {
2654 error
= XFS_ERROR(EMLINK
);
2659 * If we are using project inheritance, we only allow hard link
2660 * creation in our tree when the project IDs are the same; else
2661 * the tree quota mechanism could be circumvented.
2663 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2664 (tdp
->i_d
.di_projid
!= sip
->i_d
.di_projid
))) {
2665 error
= XFS_ERROR(EPERM
);
2670 (error
= XFS_DIR_CANENTER(mp
, tp
, tdp
, target_name
,
2674 XFS_BMAP_INIT(&free_list
, &first_block
);
2676 error
= XFS_DIR_CREATENAME(mp
, tp
, tdp
, target_name
, target_namelen
,
2677 sip
->i_ino
, &first_block
, &free_list
,
2681 xfs_ichgtime(tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2683 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
2685 error
= xfs_bumplink(tp
, sip
);
2691 * If this is a synchronous mount, make sure that the
2692 * link transaction goes to disk before returning to
2695 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2696 xfs_trans_set_sync(tp
);
2699 error
= xfs_bmap_finish (&tp
, &free_list
, first_block
, &committed
);
2701 xfs_bmap_cancel(&free_list
);
2705 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
, NULL
);
2710 /* Fall through to std_return with error = 0. */
2712 if (DM_EVENT_ENABLED(src_vp
->v_vfsp
, sip
,
2713 DM_EVENT_POSTLINK
)) {
2714 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTLINK
,
2715 target_dir_vp
, DM_RIGHT_NULL
,
2716 src_vp
, DM_RIGHT_NULL
,
2717 target_name
, NULL
, 0, error
, 0);
2722 cancel_flags
|= XFS_TRANS_ABORT
;
2726 xfs_trans_cancel(tp
, cancel_flags
);
2735 bhv_desc_t
*dir_bdp
,
2741 char *dir_name
= VNAME(dentry
);
2743 xfs_inode_t
*cdp
; /* inode of created dir */
2744 vnode_t
*cvp
; /* vnode of created dir */
2750 xfs_bmap_free_t free_list
;
2751 xfs_fsblock_t first_block
;
2753 boolean_t dp_joined_to_trans
;
2754 boolean_t created
= B_FALSE
;
2755 int dm_event_sent
= 0;
2757 struct xfs_dquot
*udqp
, *gdqp
;
2762 dir_vp
= BHV_TO_VNODE(dir_bdp
);
2763 dp
= XFS_BHVTOI(dir_bdp
);
2766 if (XFS_FORCED_SHUTDOWN(mp
))
2767 return XFS_ERROR(EIO
);
2769 dir_namelen
= VNAMELEN(dentry
);
2772 dp_joined_to_trans
= B_FALSE
;
2773 dm_di_mode
= vap
->va_mode
;
2775 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_CREATE
)) {
2776 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_CREATE
,
2777 dir_vp
, DM_RIGHT_NULL
, NULL
,
2778 DM_RIGHT_NULL
, dir_name
, NULL
,
2785 /* Return through std_return after this point. */
2787 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2791 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
2792 prid
= dp
->i_d
.di_projid
;
2793 else if (vap
->va_mask
& XFS_AT_PROJID
)
2794 prid
= (xfs_prid_t
)vap
->va_projid
;
2796 prid
= (xfs_prid_t
)dfltprid
;
2799 * Make sure that we have allocated dquot(s) on disk.
2801 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
2802 current_fsuid(credp
), current_fsgid(credp
), prid
,
2803 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
2807 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
2808 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2809 resblks
= XFS_MKDIR_SPACE_RES(mp
, dir_namelen
);
2810 error
= xfs_trans_reserve(tp
, resblks
, XFS_MKDIR_LOG_RES(mp
), 0,
2811 XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
);
2812 if (error
== ENOSPC
) {
2814 error
= xfs_trans_reserve(tp
, 0, XFS_MKDIR_LOG_RES(mp
), 0,
2815 XFS_TRANS_PERM_LOG_RES
,
2816 XFS_MKDIR_LOG_COUNT
);
2824 xfs_ilock(dp
, XFS_ILOCK_EXCL
);
2827 * Check for directory link count overflow.
2829 if (dp
->i_d
.di_nlink
>= XFS_MAXLINK
) {
2830 error
= XFS_ERROR(EMLINK
);
2835 * Reserve disk quota and the inode.
2837 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
2842 (error
= XFS_DIR_CANENTER(mp
, tp
, dp
, dir_name
, dir_namelen
)))
2845 * create the directory inode.
2847 error
= xfs_dir_ialloc(&tp
, dp
, vap
->va_mode
, 2,
2848 0, credp
, prid
, resblks
> 0,
2851 if (error
== ENOSPC
)
2858 * Now we add the directory inode to the transaction.
2859 * We waited until now since xfs_dir_ialloc might start
2860 * a new transaction. Had we joined the transaction
2861 * earlier, the locks might have gotten released.
2864 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2865 dp_joined_to_trans
= B_TRUE
;
2867 XFS_BMAP_INIT(&free_list
, &first_block
);
2869 error
= XFS_DIR_CREATENAME(mp
, tp
, dp
, dir_name
, dir_namelen
,
2870 cdp
->i_ino
, &first_block
, &free_list
,
2871 resblks
? resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
2873 ASSERT(error
!= ENOSPC
);
2876 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2879 * Bump the in memory version number of the parent directory
2880 * so that other processes accessing it will recognize that
2881 * the directory has changed.
2885 error
= XFS_DIR_INIT(mp
, tp
, cdp
, dp
);
2891 error
= xfs_bumplink(tp
, dp
);
2896 cvp
= XFS_ITOV(cdp
);
2904 * Attach the dquots to the new inode and modify the icount incore.
2906 XFS_QM_DQVOPCREATE(mp
, tp
, cdp
, udqp
, gdqp
);
2909 * If this is a synchronous mount, make sure that the
2910 * mkdir transaction goes to disk before returning to
2913 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2914 xfs_trans_set_sync(tp
);
2917 error
= xfs_bmap_finish(&tp
, &free_list
, first_block
, &committed
);
2923 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
, NULL
);
2924 XFS_QM_DQRELE(mp
, udqp
);
2925 XFS_QM_DQRELE(mp
, gdqp
);
2930 /* Fall through to std_return with error = 0 or errno from
2931 * xfs_trans_commit. */
2934 if ( (created
|| (error
!= 0 && dm_event_sent
!= 0)) &&
2935 DM_EVENT_ENABLED(dir_vp
->v_vfsp
, XFS_BHVTOI(dir_bdp
),
2936 DM_EVENT_POSTCREATE
)) {
2937 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTCREATE
,
2938 dir_vp
, DM_RIGHT_NULL
,
2939 created
? XFS_ITOV(cdp
):NULL
,
2942 dm_di_mode
, error
, 0);
2948 xfs_bmap_cancel(&free_list
);
2950 cancel_flags
|= XFS_TRANS_ABORT
;
2952 xfs_trans_cancel(tp
, cancel_flags
);
2953 XFS_QM_DQRELE(mp
, udqp
);
2954 XFS_QM_DQRELE(mp
, gdqp
);
2956 if (!dp_joined_to_trans
&& (dp
!= NULL
)) {
2957 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2970 bhv_desc_t
*dir_bdp
,
2974 char *name
= VNAME(dentry
);
2976 xfs_inode_t
*cdp
; /* child directory */
2980 xfs_bmap_free_t free_list
;
2981 xfs_fsblock_t first_block
;
2990 dir_vp
= BHV_TO_VNODE(dir_bdp
);
2991 dp
= XFS_BHVTOI(dir_bdp
);
2994 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
2996 if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp
)->i_mount
))
2997 return XFS_ERROR(EIO
);
2998 namelen
= VNAMELEN(dentry
);
3000 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_REMOVE
)) {
3001 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_REMOVE
,
3002 dir_vp
, DM_RIGHT_NULL
,
3003 NULL
, DM_RIGHT_NULL
,
3004 name
, NULL
, 0, 0, 0);
3006 return XFS_ERROR(error
);
3009 /* Return through std_return after this point. */
3014 * We need to get a reference to cdp before we get our log
3015 * reservation. The reason for this is that we cannot call
3016 * xfs_iget for an inode for which we do not have a reference
3017 * once we've acquired a log reservation. This is because the
3018 * inode we are trying to get might be in xfs_inactive going
3019 * for a log reservation. Since we'll have to wait for the
3020 * inactive code to complete before returning from xfs_iget,
3021 * we need to make sure that we don't have log space reserved
3022 * when we call xfs_iget. Instead we get an unlocked referece
3023 * to the inode before getting our log reservation.
3025 error
= xfs_get_dir_entry(dentry
, &cdp
);
3027 REMOVE_DEBUG_TRACE(__LINE__
);
3031 dm_di_mode
= cdp
->i_d
.di_mode
;
3034 * Get the dquots for the inodes.
3036 error
= XFS_QM_DQATTACH(mp
, dp
, 0);
3037 if (!error
&& dp
!= cdp
)
3038 error
= XFS_QM_DQATTACH(mp
, cdp
, 0);
3041 REMOVE_DEBUG_TRACE(__LINE__
);
3045 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
3046 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
3048 * We try to get the real space reservation first,
3049 * allowing for directory btree deletion(s) implying
3050 * possible bmap insert(s). If we can't get the space
3051 * reservation then we use 0 instead, and avoid the bmap
3052 * btree insert(s) in the directory code by, if the bmap
3053 * insert tries to happen, instead trimming the LAST
3054 * block from the directory.
3056 resblks
= XFS_REMOVE_SPACE_RES(mp
);
3057 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
3058 XFS_TRANS_PERM_LOG_RES
, XFS_DEFAULT_LOG_COUNT
);
3059 if (error
== ENOSPC
) {
3061 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
3062 XFS_TRANS_PERM_LOG_RES
, XFS_DEFAULT_LOG_COUNT
);
3065 ASSERT(error
!= ENOSPC
);
3070 XFS_BMAP_INIT(&free_list
, &first_block
);
3073 * Now lock the child directory inode and the parent directory
3074 * inode in the proper order. This will take care of validating
3075 * that the directory entry for the child directory inode has
3076 * not changed while we were obtaining a log reservation.
3078 error
= xfs_lock_dir_and_entry(dp
, dentry
, cdp
);
3080 xfs_trans_cancel(tp
, cancel_flags
);
3085 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
3088 * Only increment the parent directory vnode count if
3089 * we didn't bump it in looking up cdp. The only time
3090 * we don't bump it is when we're looking up ".".
3096 xfs_trans_ijoin(tp
, cdp
, XFS_ILOCK_EXCL
);
3098 ASSERT(cdp
->i_d
.di_nlink
>= 2);
3099 if (cdp
->i_d
.di_nlink
!= 2) {
3100 error
= XFS_ERROR(ENOTEMPTY
);
3103 if (!XFS_DIR_ISEMPTY(mp
, cdp
)) {
3104 error
= XFS_ERROR(ENOTEMPTY
);
3108 error
= XFS_DIR_REMOVENAME(mp
, tp
, dp
, name
, namelen
, cdp
->i_ino
,
3109 &first_block
, &free_list
, resblks
);
3114 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3117 * Bump the in memory generation count on the parent
3118 * directory so that other can know that it has changed.
3123 * Drop the link from cdp's "..".
3125 error
= xfs_droplink(tp
, dp
);
3131 * Drop the link from dp to cdp.
3133 error
= xfs_droplink(tp
, cdp
);
3139 * Drop the "." link from cdp to self.
3141 error
= xfs_droplink(tp
, cdp
);
3146 /* Determine these before committing transaction */
3147 last_cdp_link
= (cdp
)->i_d
.di_nlink
==0;
3150 * Take an extra ref on the child vnode so that it
3151 * does not go to xfs_inactive() from within the commit.
3156 * If this is a synchronous mount, make sure that the
3157 * rmdir transaction goes to disk before returning to
3160 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
3161 xfs_trans_set_sync(tp
);
3164 error
= xfs_bmap_finish (&tp
, &free_list
, first_block
, &committed
);
3166 xfs_bmap_cancel(&free_list
);
3167 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
|
3173 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
, NULL
);
3181 * Let interposed file systems know about removed links.
3183 VOP_LINK_REMOVED(XFS_ITOV(cdp
), dir_vp
, last_cdp_link
);
3187 /* Fall through to std_return with error = 0 or the errno
3188 * from xfs_trans_commit. */
3190 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_POSTREMOVE
)) {
3191 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTREMOVE
,
3192 dir_vp
, DM_RIGHT_NULL
,
3193 NULL
, DM_RIGHT_NULL
,
3194 name
, NULL
, dm_di_mode
,
3200 xfs_bmap_cancel(&free_list
);
3201 cancel_flags
|= XFS_TRANS_ABORT
;
3205 xfs_trans_cancel(tp
, cancel_flags
);
3213 * Read dp's entries starting at uiop->uio_offset and translate them into
3214 * bufsize bytes worth of struct dirents starting at bufbase.
3218 bhv_desc_t
*dir_bdp
,
3224 xfs_trans_t
*tp
= NULL
;
3228 vn_trace_entry(BHV_TO_VNODE(dir_bdp
), __FUNCTION__
,
3229 (inst_t
*)__return_address
);
3230 dp
= XFS_BHVTOI(dir_bdp
);
3232 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
)) {
3233 return XFS_ERROR(EIO
);
3236 lock_mode
= xfs_ilock_map_shared(dp
);
3237 error
= XFS_DIR_GETDENTS(dp
->i_mount
, tp
, dp
, uiop
, eofp
);
3238 xfs_iunlock_map_shared(dp
, lock_mode
);
3249 bhv_desc_t
*dir_bdp
,
3262 xfs_bmap_free_t free_list
;
3263 xfs_fsblock_t first_block
;
3264 boolean_t dp_joined_to_trans
;
3268 xfs_fileoff_t first_fsb
;
3269 xfs_filblks_t fs_blocks
;
3271 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
3278 struct xfs_dquot
*udqp
, *gdqp
;
3280 char *link_name
= VNAME(dentry
);
3284 dir_vp
= BHV_TO_VNODE(dir_bdp
);
3285 dp
= XFS_BHVTOI(dir_bdp
);
3286 dp_joined_to_trans
= B_FALSE
;
3291 vn_trace_entry(dir_vp
, __FUNCTION__
, (inst_t
*)__return_address
);
3295 if (XFS_FORCED_SHUTDOWN(mp
))
3296 return XFS_ERROR(EIO
);
3298 link_namelen
= VNAMELEN(dentry
);
3301 * Check component lengths of the target path name.
3303 pathlen
= strlen(target_path
);
3304 if (pathlen
>= MAXPATHLEN
) /* total string too long */
3305 return XFS_ERROR(ENAMETOOLONG
);
3306 if (pathlen
>= MAXNAMELEN
) { /* is any component too long? */
3310 for(total
= 0, path
= target_path
; total
< pathlen
;) {
3314 while(*path
== '/') {
3320 * Count up to the next slash or end of path.
3321 * Error out if the component is bigger than MAXNAMELEN.
3323 for(len
= 0; *path
!= '/' && total
< pathlen
;total
++, path
++) {
3324 if (++len
>= MAXNAMELEN
) {
3325 error
= ENAMETOOLONG
;
3332 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, dp
, DM_EVENT_SYMLINK
)) {
3333 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_SYMLINK
, dir_vp
,
3334 DM_RIGHT_NULL
, NULL
, DM_RIGHT_NULL
,
3335 link_name
, target_path
, 0, 0, 0);
3340 /* Return through std_return after this point. */
3343 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
3344 prid
= dp
->i_d
.di_projid
;
3345 else if (vap
->va_mask
& XFS_AT_PROJID
)
3346 prid
= (xfs_prid_t
)vap
->va_projid
;
3348 prid
= (xfs_prid_t
)dfltprid
;
3351 * Make sure that we have allocated dquot(s) on disk.
3353 error
= XFS_QM_DQVOPALLOC(mp
, dp
,
3354 current_fsuid(credp
), current_fsgid(credp
), prid
,
3355 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
3359 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SYMLINK
);
3360 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
3362 * The symlink will fit into the inode data fork?
3363 * There can't be any attributes so we get the whole variable part.
3365 if (pathlen
<= XFS_LITINO(mp
))
3368 fs_blocks
= XFS_B_TO_FSB(mp
, pathlen
);
3369 resblks
= XFS_SYMLINK_SPACE_RES(mp
, link_namelen
, fs_blocks
);
3370 error
= xfs_trans_reserve(tp
, resblks
, XFS_SYMLINK_LOG_RES(mp
), 0,
3371 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
3372 if (error
== ENOSPC
&& fs_blocks
== 0) {
3374 error
= xfs_trans_reserve(tp
, 0, XFS_SYMLINK_LOG_RES(mp
), 0,
3375 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
3383 xfs_ilock(dp
, XFS_ILOCK_EXCL
);
3386 * Check whether the directory allows new symlinks or not.
3388 if (dp
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) {
3389 error
= XFS_ERROR(EPERM
);
3394 * Reserve disk quota : blocks and inode.
3396 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
, udqp
, gdqp
, resblks
, 1, 0);
3401 * Check for ability to enter directory entry, if no space reserved.
3404 (error
= XFS_DIR_CANENTER(mp
, tp
, dp
, link_name
, link_namelen
)))
3407 * Initialize the bmap freelist prior to calling either
3408 * bmapi or the directory create code.
3410 XFS_BMAP_INIT(&free_list
, &first_block
);
3413 * Allocate an inode for the symlink.
3415 error
= xfs_dir_ialloc(&tp
, dp
, S_IFLNK
| (vap
->va_mode
&~S_IFMT
),
3416 1, 0, credp
, prid
, resblks
> 0, &ip
, NULL
);
3418 if (error
== ENOSPC
)
3425 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
3426 dp_joined_to_trans
= B_TRUE
;
3429 * Also attach the dquot(s) to it, if applicable.
3431 XFS_QM_DQVOPCREATE(mp
, tp
, ip
, udqp
, gdqp
);
3434 resblks
-= XFS_IALLOC_SPACE_RES(mp
);
3436 * If the symlink will fit into the inode, write it inline.
3438 if (pathlen
<= XFS_IFORK_DSIZE(ip
)) {
3439 xfs_idata_realloc(ip
, pathlen
, XFS_DATA_FORK
);
3440 memcpy(ip
->i_df
.if_u1
.if_data
, target_path
, pathlen
);
3441 ip
->i_d
.di_size
= pathlen
;
3444 * The inode was initially created in extent format.
3446 ip
->i_df
.if_flags
&= ~(XFS_IFEXTENTS
| XFS_IFBROOT
);
3447 ip
->i_df
.if_flags
|= XFS_IFINLINE
;
3449 ip
->i_d
.di_format
= XFS_DINODE_FMT_LOCAL
;
3450 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_DDATA
| XFS_ILOG_CORE
);
3454 nmaps
= SYMLINK_MAPS
;
3456 error
= xfs_bmapi(tp
, ip
, first_fsb
, fs_blocks
,
3457 XFS_BMAPI_WRITE
| XFS_BMAPI_METADATA
,
3458 &first_block
, resblks
, mval
, &nmaps
,
3465 resblks
-= fs_blocks
;
3466 ip
->i_d
.di_size
= pathlen
;
3467 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3469 cur_chunk
= target_path
;
3470 for (n
= 0; n
< nmaps
; n
++) {
3471 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
3472 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
3473 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
3474 BTOBB(byte_cnt
), 0);
3475 ASSERT(bp
&& !XFS_BUF_GETERROR(bp
));
3476 if (pathlen
< byte_cnt
) {
3479 pathlen
-= byte_cnt
;
3481 memcpy(XFS_BUF_PTR(bp
), cur_chunk
, byte_cnt
);
3482 cur_chunk
+= byte_cnt
;
3484 xfs_trans_log_buf(tp
, bp
, 0, byte_cnt
- 1);
3489 * Create the directory entry for the symlink.
3491 error
= XFS_DIR_CREATENAME(mp
, tp
, dp
, link_name
, link_namelen
,
3492 ip
->i_ino
, &first_block
, &free_list
, resblks
);
3496 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3497 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
3500 * Bump the in memory version number of the parent directory
3501 * so that other processes accessing it will recognize that
3502 * the directory has changed.
3507 * If this is a synchronous mount, make sure that the
3508 * symlink transaction goes to disk before returning to
3511 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
3512 xfs_trans_set_sync(tp
);
3516 * xfs_trans_commit normally decrements the vnode ref count
3517 * when it unlocks the inode. Since we want to return the
3518 * vnode to the caller, we bump the vnode ref count now.
3522 error
= xfs_bmap_finish(&tp
, &free_list
, first_block
, &committed
);
3526 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
, NULL
);
3527 XFS_QM_DQRELE(mp
, udqp
);
3528 XFS_QM_DQRELE(mp
, gdqp
);
3530 /* Fall through to std_return with error = 0 or errno from
3531 * xfs_trans_commit */
3533 if (DM_EVENT_ENABLED(dir_vp
->v_vfsp
, XFS_BHVTOI(dir_bdp
),
3534 DM_EVENT_POSTSYMLINK
)) {
3535 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTSYMLINK
,
3536 dir_vp
, DM_RIGHT_NULL
,
3537 error
? NULL
: XFS_ITOV(ip
),
3538 DM_RIGHT_NULL
, link_name
, target_path
,
3554 xfs_bmap_cancel(&free_list
);
3555 cancel_flags
|= XFS_TRANS_ABORT
;
3557 xfs_trans_cancel(tp
, cancel_flags
);
3558 XFS_QM_DQRELE(mp
, udqp
);
3559 XFS_QM_DQRELE(mp
, gdqp
);
3561 if (!dp_joined_to_trans
&& (dp
!= NULL
)) {
3562 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
3572 * A fid routine that takes a pointer to a previously allocated
3573 * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3583 vn_trace_entry(BHV_TO_VNODE(bdp
), __FUNCTION__
,
3584 (inst_t
*)__return_address
);
3585 ASSERT(sizeof(fid_t
) >= sizeof(xfs_fid2_t
));
3587 xfid
= (xfs_fid2_t
*)fidp
;
3588 ip
= XFS_BHVTOI(bdp
);
3589 xfid
->fid_len
= sizeof(xfs_fid2_t
) - sizeof(xfid
->fid_len
);
3592 * use memcpy because the inode is a long long and there's no
3593 * assurance that xfid->fid_ino is properly aligned.
3595 memcpy(&xfid
->fid_ino
, &ip
->i_ino
, sizeof(xfid
->fid_ino
));
3596 xfid
->fid_gen
= ip
->i_d
.di_gen
;
3613 vp
= BHV_TO_VNODE(bdp
);
3616 ip
= XFS_BHVTOI(bdp
);
3617 if (locktype
== VRWLOCK_WRITE
) {
3618 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
3619 } else if (locktype
== VRWLOCK_TRY_READ
) {
3620 return xfs_ilock_nowait(ip
, XFS_IOLOCK_SHARED
);
3621 } else if (locktype
== VRWLOCK_TRY_WRITE
) {
3622 return xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
);
3624 ASSERT((locktype
== VRWLOCK_READ
) ||
3625 (locktype
== VRWLOCK_WRITE_DIRECT
));
3626 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
3644 vp
= BHV_TO_VNODE(bdp
);
3647 ip
= XFS_BHVTOI(bdp
);
3648 if (locktype
== VRWLOCK_WRITE
) {
3650 * In the write case, we may have added a new entry to
3651 * the reference cache. This might store a pointer to
3652 * an inode to be released in this inode. If it is there,
3653 * clear the pointer and release the inode after unlocking
3656 xfs_refcache_iunlock(ip
, XFS_IOLOCK_EXCL
);
3658 ASSERT((locktype
== VRWLOCK_READ
) ||
3659 (locktype
== VRWLOCK_WRITE_DIRECT
));
3660 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
3672 xfs_inode_log_item_t
*iip
;
3675 ip
= XFS_BHVTOI(bdp
);
3679 if (XFS_FORCED_SHUTDOWN(mp
))
3680 return XFS_ERROR(EIO
);
3683 * Bypass inodes which have already been cleaned by
3684 * the inode flush clustering code inside xfs_iflush
3686 if ((ip
->i_update_core
== 0) &&
3687 ((iip
== NULL
) || !(iip
->ili_format
.ilf_fields
& XFS_ILOG_ALL
)))
3690 if (flags
& FLUSH_LOG
) {
3691 if (iip
&& iip
->ili_last_lsn
) {
3692 xlog_t
*log
= mp
->m_log
;
3694 int s
, log_flags
= XFS_LOG_FORCE
;
3696 s
= GRANT_LOCK(log
);
3697 sync_lsn
= log
->l_last_sync_lsn
;
3698 GRANT_UNLOCK(log
, s
);
3700 if ((XFS_LSN_CMP(iip
->ili_last_lsn
, sync_lsn
) <= 0))
3703 if (flags
& FLUSH_SYNC
)
3704 log_flags
|= XFS_LOG_SYNC
;
3705 return xfs_log_force(mp
, iip
->ili_last_lsn
, log_flags
);
3710 * We make this non-blocking if the inode is contended,
3711 * return EAGAIN to indicate to the caller that they
3712 * did not succeed. This prevents the flush path from
3713 * blocking on inodes inside another operation right
3714 * now, they get caught later by xfs_sync.
3716 if (flags
& FLUSH_INODE
) {
3719 if (xfs_ipincount(ip
))
3722 if (flags
& FLUSH_SYNC
) {
3723 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
3725 } else if (xfs_ilock_nowait(ip
, XFS_ILOCK_SHARED
)) {
3726 if (xfs_ipincount(ip
) || !xfs_iflock_nowait(ip
)) {
3727 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3734 if (flags
& FLUSH_SYNC
)
3735 flush_flags
= XFS_IFLUSH_SYNC
;
3737 flush_flags
= XFS_IFLUSH_ASYNC
;
3739 error
= xfs_iflush(ip
, flush_flags
);
3740 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
3759 if (!capable(CAP_SYS_ADMIN
))
3760 return XFS_ERROR(EPERM
);
3762 ip
= XFS_BHVTOI(bdp
);
3765 if (XFS_FORCED_SHUTDOWN(mp
))
3766 return XFS_ERROR(EIO
);
3768 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SET_DMATTRS
);
3769 error
= xfs_trans_reserve(tp
, 0, XFS_ICHANGE_LOG_RES (mp
), 0, 0, 0);
3771 xfs_trans_cancel(tp
, 0);
3774 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3775 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
3777 ip
->i_iocore
.io_dmevmask
= ip
->i_d
.di_dmevmask
= evmask
;
3778 ip
->i_iocore
.io_dmstate
= ip
->i_d
.di_dmstate
= state
;
3780 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3782 error
= xfs_trans_commit(tp
, 0, NULL
);
3798 vp
= BHV_TO_VNODE(bdp
);
3799 ip
= XFS_BHVTOI(bdp
);
3801 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
3803 ASSERT(!VN_MAPPED(vp
));
3805 /* bad inode, get out here ASAP */
3813 ASSERT(XFS_FORCED_SHUTDOWN(ip
->i_mount
) || ip
->i_delayed_blks
== 0);
3816 * Make sure the atime in the XFS inode is correct before freeing the
3819 xfs_synchronize_atime(ip
);
3821 /* If we have nothing to flush with this inode then complete the
3822 * teardown now, otherwise break the link between the xfs inode
3823 * and the linux inode and clean up the xfs inode later. This
3824 * avoids flushing the inode to disk during the delete operation
3827 if (!ip
->i_update_core
&& (ip
->i_itemp
== NULL
)) {
3828 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3830 return xfs_finish_reclaim(ip
, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC
);
3832 xfs_mount_t
*mp
= ip
->i_mount
;
3834 /* Protect sync from us */
3835 XFS_MOUNT_ILOCK(mp
);
3836 vn_bhv_remove(VN_BHV_HEAD(vp
), XFS_ITOBHV(ip
));
3837 list_add_tail(&ip
->i_reclaim
, &mp
->m_del_inodes
);
3838 ip
->i_flags
|= XFS_IRECLAIMABLE
;
3839 XFS_MOUNT_IUNLOCK(mp
);
3850 xfs_ihash_t
*ih
= ip
->i_hash
;
3851 vnode_t
*vp
= XFS_ITOV_NULL(ip
);
3854 if (vp
&& VN_BAD(vp
))
3857 /* The hash lock here protects a thread in xfs_iget_core from
3858 * racing with us on linking the inode back with a vnode.
3859 * Once we have the XFS_IRECLAIM flag set it will not touch
3862 write_lock(&ih
->ih_lock
);
3863 if ((ip
->i_flags
& XFS_IRECLAIM
) ||
3864 (!(ip
->i_flags
& XFS_IRECLAIMABLE
) && vp
== NULL
)) {
3865 write_unlock(&ih
->ih_lock
);
3868 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3872 ip
->i_flags
|= XFS_IRECLAIM
;
3873 write_unlock(&ih
->ih_lock
);
3876 * If the inode is still dirty, then flush it out. If the inode
3877 * is not in the AIL, then it will be OK to flush it delwri as
3878 * long as xfs_iflush() does not keep any references to the inode.
3879 * We leave that decision up to xfs_iflush() since it has the
3880 * knowledge of whether it's OK to simply do a delwri flush of
3881 * the inode or whether we need to wait until the inode is
3882 * pulled from the AIL.
3883 * We get the flush lock regardless, though, just to make sure
3884 * we don't free it while it is being flushed.
3886 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
3888 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3892 if (ip
->i_update_core
||
3893 ((ip
->i_itemp
!= NULL
) &&
3894 (ip
->i_itemp
->ili_format
.ilf_fields
!= 0))) {
3895 error
= xfs_iflush(ip
, sync_mode
);
3897 * If we hit an error, typically because of filesystem
3898 * shutdown, we don't need to let vn_reclaim to know
3899 * because we're gonna reclaim the inode anyway.
3902 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3905 xfs_iflock(ip
); /* synchronize with xfs_iflush_done */
3908 ASSERT(ip
->i_update_core
== 0);
3909 ASSERT(ip
->i_itemp
== NULL
||
3910 ip
->i_itemp
->ili_format
.ilf_fields
== 0);
3911 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3912 } else if (locked
) {
3914 * We are not interested in doing an iflush if we're
3915 * in the process of shutting down the filesystem forcibly.
3916 * So, just reclaim the inode.
3919 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3928 xfs_finish_reclaim_all(xfs_mount_t
*mp
, int noblock
)
3931 xfs_inode_t
*ip
, *n
;
3936 XFS_MOUNT_ILOCK(mp
);
3937 list_for_each_entry_safe(ip
, n
, &mp
->m_del_inodes
, i_reclaim
) {
3939 if (xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
) == 0)
3941 if (xfs_ipincount(ip
) ||
3942 !xfs_iflock_nowait(ip
)) {
3943 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
3947 XFS_MOUNT_IUNLOCK(mp
);
3948 if (xfs_finish_reclaim(ip
, noblock
,
3949 XFS_IFLUSH_DELWRI_ELSE_ASYNC
))
3958 XFS_MOUNT_IUNLOCK(mp
);
3963 * xfs_alloc_file_space()
3964 * This routine allocates disk space for the given file.
3966 * If alloc_type == 0, this request is for an ALLOCSP type
3967 * request which will change the file size. In this case, no
3968 * DMAPI event will be generated by the call. A TRUNCATE event
3969 * will be generated later by xfs_setattr.
3971 * If alloc_type != 0, this request is for a RESVSP type
3972 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
3973 * lower block boundary byte address is less than the file's
3982 xfs_alloc_file_space(
3989 xfs_mount_t
*mp
= ip
->i_mount
;
3991 xfs_filblks_t allocated_fsb
;
3992 xfs_filblks_t allocatesize_fsb
;
3993 xfs_extlen_t extsz
, temp
;
3994 xfs_fileoff_t startoffset_fsb
;
3995 xfs_fsblock_t firstfsb
;
4001 xfs_bmbt_irec_t imaps
[1], *imapp
;
4002 xfs_bmap_free_t free_list
;
4003 uint qblocks
, resblks
, resrtextents
;
4007 vn_trace_entry(XFS_ITOV(ip
), __FUNCTION__
, (inst_t
*)__return_address
);
4009 if (XFS_FORCED_SHUTDOWN(mp
))
4010 return XFS_ERROR(EIO
);
4012 rt
= XFS_IS_REALTIME_INODE(ip
);
4014 if (!(extsz
= ip
->i_d
.di_extsize
))
4015 extsz
= mp
->m_sb
.sb_rextsize
;
4017 extsz
= ip
->i_d
.di_extsize
;
4020 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
4024 return XFS_ERROR(EINVAL
);
4030 bmapi_flag
= XFS_BMAPI_WRITE
| (alloc_type
? XFS_BMAPI_PREALLOC
: 0);
4031 startoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
4032 allocatesize_fsb
= XFS_B_TO_FSB(mp
, count
);
4034 /* Generate a DMAPI event if needed. */
4035 if (alloc_type
!= 0 && offset
< ip
->i_d
.di_size
&&
4036 (attr_flags
&ATTR_DMI
) == 0 &&
4037 DM_EVENT_ENABLED(XFS_MTOVFS(mp
), ip
, DM_EVENT_WRITE
)) {
4038 xfs_off_t end_dmi_offset
;
4040 end_dmi_offset
= offset
+len
;
4041 if (end_dmi_offset
> ip
->i_d
.di_size
)
4042 end_dmi_offset
= ip
->i_d
.di_size
;
4043 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, XFS_ITOV(ip
),
4044 offset
, end_dmi_offset
- offset
,
4051 * Allocate file space until done or until there is an error
4054 while (allocatesize_fsb
&& !error
) {
4058 * Determine space reservations for data/realtime.
4060 if (unlikely(extsz
)) {
4061 s
= startoffset_fsb
;
4064 e
= startoffset_fsb
+ allocatesize_fsb
;
4065 if ((temp
= do_mod(startoffset_fsb
, extsz
)))
4067 if ((temp
= do_mod(e
, extsz
)))
4071 e
= allocatesize_fsb
;
4075 resrtextents
= qblocks
= (uint
)(e
- s
);
4076 resrtextents
/= mp
->m_sb
.sb_rextsize
;
4077 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
4078 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
4081 resblks
= qblocks
= \
4082 XFS_DIOSTRAT_SPACE_RES(mp
, (uint
)(e
- s
));
4083 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
4087 * Allocate and setup the transaction.
4089 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
4090 error
= xfs_trans_reserve(tp
, resblks
,
4091 XFS_WRITE_LOG_RES(mp
), resrtextents
,
4092 XFS_TRANS_PERM_LOG_RES
,
4093 XFS_WRITE_LOG_COUNT
);
4095 * Check for running out of space
4099 * Free the transaction structure.
4101 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
4102 xfs_trans_cancel(tp
, 0);
4105 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4106 error
= XFS_TRANS_RESERVE_QUOTA_NBLKS(mp
, tp
, ip
,
4107 qblocks
, 0, quota_flag
);
4111 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
4112 xfs_trans_ihold(tp
, ip
);
4115 * Issue the xfs_bmapi() call to allocate the blocks
4117 XFS_BMAP_INIT(&free_list
, &firstfsb
);
4118 error
= xfs_bmapi(tp
, ip
, startoffset_fsb
,
4119 allocatesize_fsb
, bmapi_flag
,
4120 &firstfsb
, 0, imapp
, &nimaps
,
4127 * Complete the transaction
4129 error
= xfs_bmap_finish(&tp
, &free_list
, firstfsb
, &committed
);
4134 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
, NULL
);
4135 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4140 allocated_fsb
= imapp
->br_blockcount
;
4143 error
= XFS_ERROR(ENOSPC
);
4147 startoffset_fsb
+= allocated_fsb
;
4148 allocatesize_fsb
-= allocated_fsb
;
4151 if (error
== ENOSPC
&& (attr_flags
&ATTR_DMI
) == 0 &&
4152 DM_EVENT_ENABLED(XFS_MTOVFS(mp
), ip
, DM_EVENT_NOSPACE
)) {
4154 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_NOSPACE
,
4155 XFS_ITOV(ip
), DM_RIGHT_NULL
,
4156 XFS_ITOV(ip
), DM_RIGHT_NULL
,
4157 NULL
, NULL
, 0, 0, 0); /* Delay flag intentionally unused */
4159 goto retry
; /* Maybe DMAPI app. has made space */
4160 /* else fall through with error from XFS_SEND_DATA */
4165 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
4166 xfs_bmap_cancel(&free_list
);
4167 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(mp
, tp
, ip
, qblocks
, 0, quota_flag
);
4169 error1
: /* Just cancel transaction */
4170 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
4171 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4172 goto dmapi_enospc_check
;
4176 * Zero file bytes between startoff and endoff inclusive.
4177 * The iolock is held exclusive and no blocks are buffered.
4180 xfs_zero_remaining_bytes(
4185 xfs_bmbt_irec_t imap
;
4186 xfs_fileoff_t offset_fsb
;
4187 xfs_off_t lastoffset
;
4190 xfs_mount_t
*mp
= ip
->i_mount
;
4194 bp
= xfs_buf_get_noaddr(mp
->m_sb
.sb_blocksize
,
4195 ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
?
4196 mp
->m_rtdev_targp
: mp
->m_ddev_targp
);
4198 for (offset
= startoff
; offset
<= endoff
; offset
= lastoffset
+ 1) {
4199 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
4201 error
= xfs_bmapi(NULL
, ip
, offset_fsb
, 1, 0, NULL
, 0, &imap
,
4203 if (error
|| nimap
< 1)
4205 ASSERT(imap
.br_blockcount
>= 1);
4206 ASSERT(imap
.br_startoff
== offset_fsb
);
4207 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+ 1) - 1;
4208 if (lastoffset
> endoff
)
4209 lastoffset
= endoff
;
4210 if (imap
.br_startblock
== HOLESTARTBLOCK
)
4212 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
4213 if (imap
.br_state
== XFS_EXT_UNWRITTEN
)
4216 XFS_BUF_UNWRITE(bp
);
4218 XFS_BUF_SET_ADDR(bp
, XFS_FSB_TO_DB(ip
, imap
.br_startblock
));
4220 if ((error
= xfs_iowait(bp
))) {
4221 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4222 mp
, bp
, XFS_BUF_ADDR(bp
));
4225 memset(XFS_BUF_PTR(bp
) +
4226 (offset
- XFS_FSB_TO_B(mp
, imap
.br_startoff
)),
4227 0, lastoffset
- offset
+ 1);
4232 if ((error
= xfs_iowait(bp
))) {
4233 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4234 mp
, bp
, XFS_BUF_ADDR(bp
));
4243 * xfs_free_file_space()
4244 * This routine frees disk space for the given file.
4246 * This routine is only called by xfs_change_file_space
4247 * for an UNRESVSP type call.
4255 xfs_free_file_space(
4264 xfs_off_t end_dmi_offset
;
4265 xfs_fileoff_t endoffset_fsb
;
4267 xfs_fsblock_t firstfsb
;
4268 xfs_bmap_free_t free_list
;
4270 xfs_bmbt_irec_t imap
;
4278 xfs_fileoff_t startoffset_fsb
;
4280 int need_iolock
= 1;
4285 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
4287 if ((error
= XFS_QM_DQATTACH(mp
, ip
, 0)))
4291 if (len
<= 0) /* if nothing being freed */
4293 rt
= (ip
->i_d
.di_flags
& XFS_DIFLAG_REALTIME
);
4294 startoffset_fsb
= XFS_B_TO_FSB(mp
, offset
);
4295 end_dmi_offset
= offset
+ len
;
4296 endoffset_fsb
= XFS_B_TO_FSBT(mp
, end_dmi_offset
);
4298 if (offset
< ip
->i_d
.di_size
&&
4299 (attr_flags
& ATTR_DMI
) == 0 &&
4300 DM_EVENT_ENABLED(XFS_MTOVFS(mp
), ip
, DM_EVENT_WRITE
)) {
4301 if (end_dmi_offset
> ip
->i_d
.di_size
)
4302 end_dmi_offset
= ip
->i_d
.di_size
;
4303 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, vp
,
4304 offset
, end_dmi_offset
- offset
,
4305 AT_DELAY_FLAG(attr_flags
), NULL
);
4310 ASSERT(attr_flags
& ATTR_NOLOCK
? attr_flags
& ATTR_DMI
: 1);
4311 if (attr_flags
& ATTR_NOLOCK
)
4314 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
4316 rounding
= MAX((__uint8_t
)(1 << mp
->m_sb
.sb_blocklog
),
4318 ilen
= len
+ (offset
& (rounding
- 1));
4319 ioffset
= offset
& ~(rounding
- 1);
4320 if (ilen
& (rounding
- 1))
4321 ilen
= (ilen
+ rounding
) & ~(rounding
- 1);
4323 if (VN_CACHED(vp
) != 0) {
4324 xfs_inval_cached_trace(&ip
->i_iocore
, ioffset
, -1,
4325 ctooff(offtoct(ioffset
)), -1);
4326 VOP_FLUSHINVAL_PAGES(vp
, ctooff(offtoct(ioffset
)),
4327 -1, FI_REMAPF_LOCKED
);
4331 * Need to zero the stuff we're not freeing, on disk.
4332 * If its a realtime file & can't use unwritten extents then we
4333 * actually need to zero the extent edges. Otherwise xfs_bunmapi
4334 * will take care of it for us.
4336 if (rt
&& !XFS_SB_VERSION_HASEXTFLGBIT(&mp
->m_sb
)) {
4338 error
= xfs_bmapi(NULL
, ip
, startoffset_fsb
, 1, 0, NULL
, 0,
4339 &imap
, &nimap
, NULL
);
4341 goto out_unlock_iolock
;
4342 ASSERT(nimap
== 0 || nimap
== 1);
4343 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
4346 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
4347 block
= imap
.br_startblock
;
4348 mod
= do_div(block
, mp
->m_sb
.sb_rextsize
);
4350 startoffset_fsb
+= mp
->m_sb
.sb_rextsize
- mod
;
4353 error
= xfs_bmapi(NULL
, ip
, endoffset_fsb
- 1, 1, 0, NULL
, 0,
4354 &imap
, &nimap
, NULL
);
4356 goto out_unlock_iolock
;
4357 ASSERT(nimap
== 0 || nimap
== 1);
4358 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
4359 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
4361 if (mod
&& (mod
!= mp
->m_sb
.sb_rextsize
))
4362 endoffset_fsb
-= mod
;
4365 if ((done
= (endoffset_fsb
<= startoffset_fsb
)))
4367 * One contiguous piece to clear
4369 error
= xfs_zero_remaining_bytes(ip
, offset
, offset
+ len
- 1);
4372 * Some full blocks, possibly two pieces to clear
4374 if (offset
< XFS_FSB_TO_B(mp
, startoffset_fsb
))
4375 error
= xfs_zero_remaining_bytes(ip
, offset
,
4376 XFS_FSB_TO_B(mp
, startoffset_fsb
) - 1);
4378 XFS_FSB_TO_B(mp
, endoffset_fsb
) < offset
+ len
)
4379 error
= xfs_zero_remaining_bytes(ip
,
4380 XFS_FSB_TO_B(mp
, endoffset_fsb
),
4385 * free file space until done or until there is an error
4387 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
4388 while (!error
&& !done
) {
4391 * allocate and setup the transaction
4393 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
4394 error
= xfs_trans_reserve(tp
,
4396 XFS_WRITE_LOG_RES(mp
),
4398 XFS_TRANS_PERM_LOG_RES
,
4399 XFS_WRITE_LOG_COUNT
);
4402 * check for running out of space
4406 * Free the transaction structure.
4408 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
4409 xfs_trans_cancel(tp
, 0);
4412 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4413 error
= XFS_TRANS_RESERVE_QUOTA(mp
, tp
,
4414 ip
->i_udquot
, ip
->i_gdquot
, resblks
, 0,
4415 XFS_QMOPT_RES_REGBLKS
);
4419 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
4420 xfs_trans_ihold(tp
, ip
);
4423 * issue the bunmapi() call to free the blocks
4425 XFS_BMAP_INIT(&free_list
, &firstfsb
);
4426 error
= xfs_bunmapi(tp
, ip
, startoffset_fsb
,
4427 endoffset_fsb
- startoffset_fsb
,
4428 0, 2, &firstfsb
, &free_list
, &done
);
4434 * complete the transaction
4436 error
= xfs_bmap_finish(&tp
, &free_list
, firstfsb
, &committed
);
4441 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
, NULL
);
4442 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4447 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
4451 xfs_bmap_cancel(&free_list
);
4453 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
4454 xfs_iunlock(ip
, need_iolock
? (XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
) :
4460 * xfs_change_file_space()
4461 * This routine allocates or frees disk space for the given file.
4462 * The user specified parameters are checked for alignment and size
4471 xfs_change_file_space(
4485 xfs_off_t startoffset
;
4491 vp
= BHV_TO_VNODE(bdp
);
4492 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
4494 ip
= XFS_BHVTOI(bdp
);
4498 * must be a regular file and have write permission
4501 return XFS_ERROR(EINVAL
);
4503 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
4505 if ((error
= xfs_iaccess(ip
, S_IWUSR
, credp
))) {
4506 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
4510 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
4512 switch (bf
->l_whence
) {
4513 case 0: /*SEEK_SET*/
4515 case 1: /*SEEK_CUR*/
4516 bf
->l_start
+= offset
;
4518 case 2: /*SEEK_END*/
4519 bf
->l_start
+= ip
->i_d
.di_size
;
4522 return XFS_ERROR(EINVAL
);
4525 llen
= bf
->l_len
> 0 ? bf
->l_len
- 1 : bf
->l_len
;
4527 if ( (bf
->l_start
< 0)
4528 || (bf
->l_start
> XFS_MAXIOFFSET(mp
))
4529 || (bf
->l_start
+ llen
< 0)
4530 || (bf
->l_start
+ llen
> XFS_MAXIOFFSET(mp
)))
4531 return XFS_ERROR(EINVAL
);
4535 startoffset
= bf
->l_start
;
4536 fsize
= ip
->i_d
.di_size
;
4539 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4541 * These calls do NOT zero the data space allocated to the file,
4542 * nor do they change the file size.
4544 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4546 * These calls cause the new file data to be zeroed and the file
4547 * size to be changed.
4549 setprealloc
= clrprealloc
= 0;
4552 case XFS_IOC_RESVSP
:
4553 case XFS_IOC_RESVSP64
:
4554 error
= xfs_alloc_file_space(ip
, startoffset
, bf
->l_len
,
4561 case XFS_IOC_UNRESVSP
:
4562 case XFS_IOC_UNRESVSP64
:
4563 if ((error
= xfs_free_file_space(ip
, startoffset
, bf
->l_len
,
4568 case XFS_IOC_ALLOCSP
:
4569 case XFS_IOC_ALLOCSP64
:
4570 case XFS_IOC_FREESP
:
4571 case XFS_IOC_FREESP64
:
4572 if (startoffset
> fsize
) {
4573 error
= xfs_alloc_file_space(ip
, fsize
,
4574 startoffset
- fsize
, 0, attr_flags
);
4579 va
.va_mask
= XFS_AT_SIZE
;
4580 va
.va_size
= startoffset
;
4582 error
= xfs_setattr(bdp
, &va
, attr_flags
, credp
);
4592 return XFS_ERROR(EINVAL
);
4596 * update the inode timestamp, mode, and prealloc flag bits
4598 tp
= xfs_trans_alloc(mp
, XFS_TRANS_WRITEID
);
4600 if ((error
= xfs_trans_reserve(tp
, 0, XFS_WRITEID_LOG_RES(mp
),
4603 xfs_trans_cancel(tp
, 0);
4607 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
4609 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
4610 xfs_trans_ihold(tp
, ip
);
4612 if ((attr_flags
& ATTR_DMI
) == 0) {
4613 ip
->i_d
.di_mode
&= ~S_ISUID
;
4616 * Note that we don't have to worry about mandatory
4617 * file locking being disabled here because we only
4618 * clear the S_ISGID bit if the Group execute bit is
4619 * on, but if it was on then mandatory locking wouldn't
4620 * have been enabled.
4622 if (ip
->i_d
.di_mode
& S_IXGRP
)
4623 ip
->i_d
.di_mode
&= ~S_ISGID
;
4625 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
4628 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
4629 else if (clrprealloc
)
4630 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
4632 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
4633 xfs_trans_set_sync(tp
);
4635 error
= xfs_trans_commit(tp
, 0, NULL
);
4637 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
4642 vnodeops_t xfs_vnodeops
= {
4643 BHV_IDENTITY_INIT(VN_BHV_XFS
,VNODE_POSITION_XFS
),
4644 .vop_open
= xfs_open
,
4645 .vop_read
= xfs_read
,
4646 #ifdef HAVE_SENDFILE
4647 .vop_sendfile
= xfs_sendfile
,
4649 .vop_write
= xfs_write
,
4650 .vop_ioctl
= xfs_ioctl
,
4651 .vop_getattr
= xfs_getattr
,
4652 .vop_setattr
= xfs_setattr
,
4653 .vop_access
= xfs_access
,
4654 .vop_lookup
= xfs_lookup
,
4655 .vop_create
= xfs_create
,
4656 .vop_remove
= xfs_remove
,
4657 .vop_link
= xfs_link
,
4658 .vop_rename
= xfs_rename
,
4659 .vop_mkdir
= xfs_mkdir
,
4660 .vop_rmdir
= xfs_rmdir
,
4661 .vop_readdir
= xfs_readdir
,
4662 .vop_symlink
= xfs_symlink
,
4663 .vop_readlink
= xfs_readlink
,
4664 .vop_fsync
= xfs_fsync
,
4665 .vop_inactive
= xfs_inactive
,
4666 .vop_fid2
= xfs_fid2
,
4667 .vop_rwlock
= xfs_rwlock
,
4668 .vop_rwunlock
= xfs_rwunlock
,
4669 .vop_bmap
= xfs_bmap
,
4670 .vop_reclaim
= xfs_reclaim
,
4671 .vop_attr_get
= xfs_attr_get
,
4672 .vop_attr_set
= xfs_attr_set
,
4673 .vop_attr_remove
= xfs_attr_remove
,
4674 .vop_attr_list
= xfs_attr_list
,
4675 .vop_link_removed
= (vop_link_removed_t
)fs_noval
,
4676 .vop_vnode_change
= (vop_vnode_change_t
)fs_noval
,
4677 .vop_tosspages
= fs_tosspages
,
4678 .vop_flushinval_pages
= fs_flushinval_pages
,
4679 .vop_flush_pages
= fs_flush_pages
,
4680 .vop_release
= xfs_release
,
4681 .vop_iflush
= xfs_inode_flush
,