2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_types.h"
25 #include "xfs_trans.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 #include "xfs_inode.h"
39 #include "xfs_inode_item.h"
40 #include "xfs_itable.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_alloc.h"
48 #include "xfs_error.h"
49 #include "xfs_quota.h"
50 #include "xfs_utils.h"
51 #include "xfs_rtalloc.h"
52 #include "xfs_trans_space.h"
53 #include "xfs_log_priv.h"
54 #include "xfs_filestream.h"
55 #include "xfs_vnodeops.h"
56 #include "xfs_trace.h"
64 xfs_mount_t
*mp
= ip
->i_mount
;
65 struct inode
*inode
= VFS_I(ip
);
66 int mask
= iattr
->ia_valid
;
73 struct xfs_dquot
*udqp
, *gdqp
, *olddquot1
, *olddquot2
;
78 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
79 return XFS_ERROR(EROFS
);
81 if (XFS_FORCED_SHUTDOWN(mp
))
82 return XFS_ERROR(EIO
);
84 code
= -inode_change_ok(inode
, iattr
);
88 olddquot1
= olddquot2
= NULL
;
92 * If disk quotas is on, we make sure that the dquots do exist on disk,
93 * before we start any other transactions. Trying to do this later
94 * is messy. We don't care to take a readlock to look at the ids
95 * in inode here, because we can't hold it across the trans_reserve.
96 * If the IDs do change before we take the ilock, we're covered
97 * because the i_*dquot fields will get updated anyway.
99 if (XFS_IS_QUOTA_ON(mp
) && (mask
& (ATTR_UID
|ATTR_GID
))) {
102 if ((mask
& ATTR_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
104 qflags
|= XFS_QMOPT_UQUOTA
;
106 uid
= ip
->i_d
.di_uid
;
108 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
110 qflags
|= XFS_QMOPT_GQUOTA
;
112 gid
= ip
->i_d
.di_gid
;
116 * We take a reference when we initialize udqp and gdqp,
117 * so it is important that we never blindly double trip on
118 * the same variable. See xfs_create() for an example.
120 ASSERT(udqp
== NULL
);
121 ASSERT(gdqp
== NULL
);
122 code
= xfs_qm_vop_dqalloc(ip
, uid
, gid
, ip
->i_d
.di_projid
,
123 qflags
, &udqp
, &gdqp
);
129 * For the other attributes, we acquire the inode lock and
130 * first do an error checking pass.
133 lock_flags
= XFS_ILOCK_EXCL
;
134 if (flags
& XFS_ATTR_NOLOCK
)
136 if (!(mask
& ATTR_SIZE
)) {
137 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
139 code
= xfs_trans_reserve(tp
, 0, XFS_ICHANGE_LOG_RES(mp
),
146 if (DM_EVENT_ENABLED(ip
, DM_EVENT_TRUNCATE
) &&
147 !(flags
& XFS_ATTR_DMI
)) {
148 int dmflags
= AT_DELAY_FLAG(flags
) | DM_SEM_FLAG_WR
;
149 code
= XFS_SEND_DATA(mp
, DM_EVENT_TRUNCATE
, ip
,
150 iattr
->ia_size
, 0, dmflags
, NULL
);
157 lock_flags
|= XFS_IOLOCK_EXCL
;
160 xfs_ilock(ip
, lock_flags
);
163 * Change file ownership. Must be the owner or privileged.
165 if (mask
& (ATTR_UID
|ATTR_GID
)) {
167 * These IDs could have changed since we last looked at them.
168 * But, we're assured that if the ownership did change
169 * while we didn't have the inode locked, inode's dquot(s)
170 * would have changed also.
172 iuid
= ip
->i_d
.di_uid
;
173 igid
= ip
->i_d
.di_gid
;
174 gid
= (mask
& ATTR_GID
) ? iattr
->ia_gid
: igid
;
175 uid
= (mask
& ATTR_UID
) ? iattr
->ia_uid
: iuid
;
178 * Do a quota reservation only if uid/gid is actually
181 if (XFS_IS_QUOTA_RUNNING(mp
) &&
182 ((XFS_IS_UQUOTA_ON(mp
) && iuid
!= uid
) ||
183 (XFS_IS_GQUOTA_ON(mp
) && igid
!= gid
))) {
185 code
= xfs_qm_vop_chown_reserve(tp
, ip
, udqp
, gdqp
,
186 capable(CAP_FOWNER
) ?
187 XFS_QMOPT_FORCE_RES
: 0);
188 if (code
) /* out of quota */
194 * Truncate file. Must have write permission and not be a directory.
196 if (mask
& ATTR_SIZE
) {
197 /* Short circuit the truncate case for zero length files */
198 if (iattr
->ia_size
== 0 &&
199 ip
->i_size
== 0 && ip
->i_d
.di_nextents
== 0) {
200 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
201 lock_flags
&= ~XFS_ILOCK_EXCL
;
202 if (mask
& ATTR_CTIME
)
203 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
208 if (S_ISDIR(ip
->i_d
.di_mode
)) {
209 code
= XFS_ERROR(EISDIR
);
211 } else if (!S_ISREG(ip
->i_d
.di_mode
)) {
212 code
= XFS_ERROR(EINVAL
);
217 * Make sure that the dquots are attached to the inode.
219 code
= xfs_qm_dqattach_locked(ip
, 0);
224 * Now we can make the changes. Before we join the inode
225 * to the transaction, if ATTR_SIZE is set then take care of
226 * the part of the truncation that must be done without the
227 * inode lock. This needs to be done before joining the inode
228 * to the transaction, because the inode cannot be unlocked
229 * once it is a part of the transaction.
231 if (iattr
->ia_size
> ip
->i_size
) {
233 * Do the first part of growing a file: zero any data
234 * in the last block that is beyond the old EOF. We
235 * need to do this before the inode is joined to the
236 * transaction to modify the i_size.
238 code
= xfs_zero_eof(ip
, iattr
->ia_size
, ip
->i_size
);
240 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
243 * We are going to log the inode size change in this
244 * transaction so any previous writes that are beyond the on
245 * disk EOF and the new EOF that have not been written out need
246 * to be written here. If we do not write the data out, we
247 * expose ourselves to the null files problem.
249 * Only flush from the on disk size to the smaller of the in
250 * memory file size or the new size as that's the range we
251 * really care about here and prevents waiting for other data
252 * not within the range we care about here.
255 ip
->i_size
!= ip
->i_d
.di_size
&&
256 iattr
->ia_size
> ip
->i_d
.di_size
) {
257 code
= xfs_flush_pages(ip
,
258 ip
->i_d
.di_size
, iattr
->ia_size
,
262 /* wait for all I/O to complete */
266 code
= xfs_itruncate_data(ip
, iattr
->ia_size
);
269 lock_flags
&= ~XFS_ILOCK_EXCL
;
270 ASSERT(lock_flags
== XFS_IOLOCK_EXCL
);
273 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
274 if ((code
= xfs_trans_reserve(tp
, 0,
275 XFS_ITRUNCATE_LOG_RES(mp
), 0,
276 XFS_TRANS_PERM_LOG_RES
,
277 XFS_ITRUNCATE_LOG_COUNT
))) {
278 xfs_trans_cancel(tp
, 0);
280 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
283 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
284 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
286 xfs_trans_ijoin(tp
, ip
, lock_flags
);
287 xfs_trans_ihold(tp
, ip
);
290 * Only change the c/mtime if we are changing the size
291 * or we are explicitly asked to change it. This handles
292 * the semantic difference between truncate() and ftruncate()
293 * as implemented in the VFS.
295 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME
296 * is a special case where we need to update the times despite
297 * not having these flags set. For all other operations the
298 * VFS set these flags explicitly if it wants a timestamp
301 if (iattr
->ia_size
!= ip
->i_size
&&
302 (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))) {
303 iattr
->ia_ctime
= iattr
->ia_mtime
=
304 current_fs_time(inode
->i_sb
);
305 mask
|= ATTR_CTIME
| ATTR_MTIME
;
308 if (iattr
->ia_size
> ip
->i_size
) {
309 ip
->i_d
.di_size
= iattr
->ia_size
;
310 ip
->i_size
= iattr
->ia_size
;
311 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
312 } else if (iattr
->ia_size
<= ip
->i_size
||
313 (iattr
->ia_size
== 0 && ip
->i_d
.di_nextents
)) {
315 * signal a sync transaction unless
316 * we're truncating an already unlinked
317 * file on a wsync filesystem
319 code
= xfs_itruncate_finish(&tp
, ip
, iattr
->ia_size
,
321 ((ip
->i_d
.di_nlink
!= 0 ||
322 !(mp
->m_flags
& XFS_MOUNT_WSYNC
))
327 * Truncated "down", so we're removing references
328 * to old data here - if we now delay flushing for
329 * a long time, we expose ourselves unduly to the
330 * notorious NULL files problem. So, we mark this
331 * vnode and flush it when the file is closed, and
332 * do not wait the usual (long) time for writeout.
334 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
337 xfs_trans_ijoin(tp
, ip
, lock_flags
);
338 xfs_trans_ihold(tp
, ip
);
342 * Change file ownership. Must be the owner or privileged.
344 if (mask
& (ATTR_UID
|ATTR_GID
)) {
346 * CAP_FSETID overrides the following restrictions:
348 * The set-user-ID and set-group-ID bits of a file will be
349 * cleared upon successful return from chown()
351 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
352 !capable(CAP_FSETID
)) {
353 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
357 * Change the ownerships and register quota modifications
358 * in the transaction.
361 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_UQUOTA_ON(mp
)) {
362 ASSERT(mask
& ATTR_UID
);
364 olddquot1
= xfs_qm_vop_chown(tp
, ip
,
365 &ip
->i_udquot
, udqp
);
367 ip
->i_d
.di_uid
= uid
;
371 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_GQUOTA_ON(mp
)) {
372 ASSERT(!XFS_IS_PQUOTA_ON(mp
));
373 ASSERT(mask
& ATTR_GID
);
375 olddquot2
= xfs_qm_vop_chown(tp
, ip
,
376 &ip
->i_gdquot
, gdqp
);
378 ip
->i_d
.di_gid
= gid
;
384 * Change file access modes.
386 if (mask
& ATTR_MODE
) {
387 umode_t mode
= iattr
->ia_mode
;
389 if (!in_group_p(inode
->i_gid
) && !capable(CAP_FSETID
))
392 ip
->i_d
.di_mode
&= S_IFMT
;
393 ip
->i_d
.di_mode
|= mode
& ~S_IFMT
;
395 inode
->i_mode
&= S_IFMT
;
396 inode
->i_mode
|= mode
& ~S_IFMT
;
400 * Change file access or modified times.
402 if (mask
& ATTR_ATIME
) {
403 inode
->i_atime
= iattr
->ia_atime
;
404 ip
->i_d
.di_atime
.t_sec
= iattr
->ia_atime
.tv_sec
;
405 ip
->i_d
.di_atime
.t_nsec
= iattr
->ia_atime
.tv_nsec
;
406 ip
->i_update_core
= 1;
408 if (mask
& ATTR_CTIME
) {
409 inode
->i_ctime
= iattr
->ia_ctime
;
410 ip
->i_d
.di_ctime
.t_sec
= iattr
->ia_ctime
.tv_sec
;
411 ip
->i_d
.di_ctime
.t_nsec
= iattr
->ia_ctime
.tv_nsec
;
412 ip
->i_update_core
= 1;
414 if (mask
& ATTR_MTIME
) {
415 inode
->i_mtime
= iattr
->ia_mtime
;
416 ip
->i_d
.di_mtime
.t_sec
= iattr
->ia_mtime
.tv_sec
;
417 ip
->i_d
.di_mtime
.t_nsec
= iattr
->ia_mtime
.tv_nsec
;
418 ip
->i_update_core
= 1;
422 * And finally, log the inode core if any attribute in it
425 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
|
426 ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
427 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
429 XFS_STATS_INC(xs_ig_attrchg
);
432 * If this is a synchronous mount, make sure that the
433 * transaction goes to disk before returning to the user.
434 * This is slightly sub-optimal in that truncates require
435 * two sync transactions instead of one for wsync filesystems.
436 * One for the truncate and one for the timestamps since we
437 * don't want to change the timestamps unless we're sure the
438 * truncate worked. Truncates are less than 1% of the laddis
439 * mix so this probably isn't worth the trouble to optimize.
442 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
443 xfs_trans_set_sync(tp
);
445 code
= xfs_trans_commit(tp
, commit_flags
);
447 xfs_iunlock(ip
, lock_flags
);
450 * Release any dquot(s) the inode had kept before chown.
452 xfs_qm_dqrele(olddquot1
);
453 xfs_qm_dqrele(olddquot2
);
461 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
462 * update. We could avoid this with linked transactions
463 * and passing down the transaction pointer all the way
464 * to attr_set. No previous user of the generic
465 * Posix ACL code seems to care about this issue either.
467 if ((mask
& ATTR_MODE
) && !(flags
& XFS_ATTR_NOACL
)) {
468 code
= -xfs_acl_chmod(inode
);
470 return XFS_ERROR(code
);
473 if (DM_EVENT_ENABLED(ip
, DM_EVENT_ATTRIBUTE
) &&
474 !(flags
& XFS_ATTR_DMI
)) {
475 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_ATTRIBUTE
, ip
, DM_RIGHT_NULL
,
476 NULL
, DM_RIGHT_NULL
, NULL
, NULL
,
477 0, 0, AT_DELAY_FLAG(flags
));
482 commit_flags
|= XFS_TRANS_ABORT
;
488 xfs_trans_cancel(tp
, commit_flags
);
490 if (lock_flags
!= 0) {
491 xfs_iunlock(ip
, lock_flags
);
497 * The maximum pathlen is 1024 bytes. Since the minimum file system
498 * blocksize is 512 bytes, we can get a max of 2 extents back from
501 #define SYMLINK_MAPS 2
508 xfs_mount_t
*mp
= ip
->i_mount
;
509 int pathlen
= ip
->i_d
.di_size
;
510 int nmaps
= SYMLINK_MAPS
;
511 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
518 error
= xfs_bmapi(NULL
, ip
, 0, XFS_B_TO_FSB(mp
, pathlen
), 0, NULL
, 0,
519 mval
, &nmaps
, NULL
, NULL
);
523 for (n
= 0; n
< nmaps
; n
++) {
524 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
525 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
527 bp
= xfs_buf_read(mp
->m_ddev_targp
, d
, BTOBB(byte_cnt
),
528 XBF_LOCK
| XBF_MAPPED
| XBF_DONT_BLOCK
);
529 error
= XFS_BUF_GETERROR(bp
);
531 xfs_ioerror_alert("xfs_readlink",
532 ip
->i_mount
, bp
, XFS_BUF_ADDR(bp
));
536 if (pathlen
< byte_cnt
)
540 memcpy(link
, XFS_BUF_PTR(bp
), byte_cnt
);
544 link
[ip
->i_d
.di_size
] = '\0';
556 xfs_mount_t
*mp
= ip
->i_mount
;
560 xfs_itrace_entry(ip
);
562 if (XFS_FORCED_SHUTDOWN(mp
))
563 return XFS_ERROR(EIO
);
565 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
567 ASSERT((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
);
568 ASSERT(ip
->i_d
.di_size
<= MAXPATHLEN
);
570 pathlen
= ip
->i_d
.di_size
;
574 if (ip
->i_df
.if_flags
& XFS_IFINLINE
) {
575 memcpy(link
, ip
->i_df
.if_u1
.if_data
, pathlen
);
576 link
[pathlen
] = '\0';
578 error
= xfs_readlink_bmap(ip
, link
);
582 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
589 * This is called to sync the inode and its data out to disk. We need to hold
590 * the I/O lock while flushing the data, and the inode lock while flushing the
591 * inode. The inode lock CANNOT be held while flushing the data, so acquire
592 * after we're done with that.
602 xfs_itrace_entry(ip
);
604 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
605 return XFS_ERROR(EIO
);
608 * We always need to make sure that the required inode state is safe on
609 * disk. The inode might be clean but we still might need to force the
610 * log because of committed transactions that haven't hit the disk yet.
611 * Likewise, there could be unflushed non-transactional changes to the
612 * inode core that have to go to disk and this requires us to issue
613 * a synchronous transaction to capture these changes correctly.
615 * This code relies on the assumption that if the update_* fields
616 * of the inode are clear and the inode is unpinned then it is clean
617 * and no action is required.
619 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
621 if (!ip
->i_update_core
) {
623 * Timestamps/size haven't changed since last inode flush or
624 * inode transaction commit. That means either nothing got
625 * written or a transaction committed which caught the updates.
626 * If the latter happened and the transaction hasn't hit the
627 * disk yet, the inode will be still be pinned. If it is,
630 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
631 if (xfs_ipincount(ip
)) {
632 error
= _xfs_log_force(ip
->i_mount
, XFS_LOG_SYNC
,
637 * Kick off a transaction to log the inode core to get the
638 * updates. The sync transaction will also force the log.
640 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
641 tp
= xfs_trans_alloc(ip
->i_mount
, XFS_TRANS_FSYNC_TS
);
642 error
= xfs_trans_reserve(tp
, 0,
643 XFS_FSYNC_TS_LOG_RES(ip
->i_mount
), 0, 0, 0);
645 xfs_trans_cancel(tp
, 0);
648 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
651 * Note - it's possible that we might have pushed ourselves out
652 * of the way during trans_reserve which would flush the inode.
653 * But there's no guarantee that the inode buffer has actually
654 * gone out yet (it's delwri). Plus the buffer could be pinned
655 * anyway if it's part of an inode in another recent
656 * transaction. So we play it safe and fire off the
657 * transaction anyway.
659 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
660 xfs_trans_ihold(tp
, ip
);
661 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
662 xfs_trans_set_sync(tp
);
663 error
= _xfs_trans_commit(tp
, 0, &log_flushed
);
665 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
668 if (ip
->i_mount
->m_flags
& XFS_MOUNT_BARRIER
) {
670 * If the log write didn't issue an ordered tag we need
671 * to flush the disk cache for the data device now.
674 xfs_blkdev_issue_flush(ip
->i_mount
->m_ddev_targp
);
677 * If this inode is on the RT dev we need to flush that
680 if (XFS_IS_REALTIME_INODE(ip
))
681 xfs_blkdev_issue_flush(ip
->i_mount
->m_rtdev_targp
);
688 * Flags for xfs_free_eofblocks
690 #define XFS_FREE_EOF_TRYLOCK (1<<0)
693 * This is called by xfs_inactive to free any blocks beyond eof
694 * when the link count isn't zero and by xfs_dm_punch_hole() when
695 * punching a hole to EOF.
705 xfs_fileoff_t end_fsb
;
706 xfs_fileoff_t last_fsb
;
707 xfs_filblks_t map_len
;
709 xfs_bmbt_irec_t imap
;
712 * Figure out if there are any blocks beyond the end
713 * of the file. If not, then there is nothing to do.
715 end_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)ip
->i_size
));
716 last_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
717 map_len
= last_fsb
- end_fsb
;
722 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
723 error
= xfs_bmapi(NULL
, ip
, end_fsb
, map_len
, 0,
724 NULL
, 0, &imap
, &nimaps
, NULL
, NULL
);
725 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
727 if (!error
&& (nimaps
!= 0) &&
728 (imap
.br_startblock
!= HOLESTARTBLOCK
||
729 ip
->i_delayed_blks
)) {
731 * Attach the dquots to the inode up front.
733 error
= xfs_qm_dqattach(ip
, 0);
738 * There are blocks after the end of file.
739 * Free them up now by truncating the file to
742 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
745 * Do the xfs_itruncate_start() call before
746 * reserving any log space because
747 * itruncate_start will call into the buffer
749 * do that within a transaction.
751 if (flags
& XFS_FREE_EOF_TRYLOCK
) {
752 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
753 xfs_trans_cancel(tp
, 0);
757 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
759 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
,
762 xfs_trans_cancel(tp
, 0);
763 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
767 error
= xfs_trans_reserve(tp
, 0,
768 XFS_ITRUNCATE_LOG_RES(mp
),
769 0, XFS_TRANS_PERM_LOG_RES
,
770 XFS_ITRUNCATE_LOG_COUNT
);
772 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
773 xfs_trans_cancel(tp
, 0);
774 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
778 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
779 xfs_trans_ijoin(tp
, ip
,
782 xfs_trans_ihold(tp
, ip
);
784 error
= xfs_itruncate_finish(&tp
, ip
,
789 * If we get an error at this point we
790 * simply don't bother truncating the file.
794 (XFS_TRANS_RELEASE_LOG_RES
|
797 error
= xfs_trans_commit(tp
,
798 XFS_TRANS_RELEASE_LOG_RES
);
800 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
);
806 * Free a symlink that has blocks associated with it.
809 xfs_inactive_symlink_rmt(
817 xfs_fsblock_t first_block
;
818 xfs_bmap_free_t free_list
;
821 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
829 ASSERT(ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
));
831 * We're freeing a symlink that has some
832 * blocks allocated to it. Free the
833 * blocks here. We know that we've got
834 * either 1 or 2 extents and that we can
835 * free them all in one bunmapi call.
837 ASSERT(ip
->i_d
.di_nextents
> 0 && ip
->i_d
.di_nextents
<= 2);
838 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
839 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
840 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
841 xfs_trans_cancel(tp
, 0);
846 * Lock the inode, fix the size, and join it to the transaction.
847 * Hold it so in the normal path, we still have it locked for
848 * the second transaction. In the error paths we need it
849 * held so the cancel won't rele it, see below.
851 xfs_ilock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
852 size
= (int)ip
->i_d
.di_size
;
854 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
855 xfs_trans_ihold(tp
, ip
);
856 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
858 * Find the block(s) so we can inval and unmap them.
861 xfs_bmap_init(&free_list
, &first_block
);
862 nmaps
= ARRAY_SIZE(mval
);
863 if ((error
= xfs_bmapi(tp
, ip
, 0, XFS_B_TO_FSB(mp
, size
),
864 XFS_BMAPI_METADATA
, &first_block
, 0, mval
, &nmaps
,
868 * Invalidate the block(s).
870 for (i
= 0; i
< nmaps
; i
++) {
871 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
872 XFS_FSB_TO_DADDR(mp
, mval
[i
].br_startblock
),
873 XFS_FSB_TO_BB(mp
, mval
[i
].br_blockcount
), 0);
874 xfs_trans_binval(tp
, bp
);
877 * Unmap the dead block(s) to the free_list.
879 if ((error
= xfs_bunmapi(tp
, ip
, 0, size
, XFS_BMAPI_METADATA
, nmaps
,
880 &first_block
, &free_list
, NULL
, &done
)))
884 * Commit the first transaction. This logs the EFI and the inode.
886 if ((error
= xfs_bmap_finish(&tp
, &free_list
, &committed
)))
889 * The transaction must have been committed, since there were
890 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
891 * The new tp has the extent freeing and EFDs.
895 * The first xact was committed, so add the inode to the new one.
896 * Mark it dirty so it will be logged and moved forward in the log as
897 * part of every commit.
899 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
900 xfs_trans_ihold(tp
, ip
);
901 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
903 * Get a new, empty transaction to return to our caller.
905 ntp
= xfs_trans_dup(tp
);
907 * Commit the transaction containing extent freeing and EFDs.
908 * If we get an error on the commit here or on the reserve below,
909 * we need to unlock the inode since the new transaction doesn't
910 * have the inode attached.
912 error
= xfs_trans_commit(tp
, 0);
915 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
919 * transaction commit worked ok so we can drop the extra ticket
920 * reference that we gained in xfs_trans_dup()
922 xfs_log_ticket_put(tp
->t_ticket
);
925 * Remove the memory for extent descriptions (just bookkeeping).
927 if (ip
->i_df
.if_bytes
)
928 xfs_idata_realloc(ip
, -ip
->i_df
.if_bytes
, XFS_DATA_FORK
);
929 ASSERT(ip
->i_df
.if_bytes
== 0);
931 * Put an itruncate log reservation in the new transaction
934 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
935 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
936 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
940 * Return with the inode locked but not joined to the transaction.
946 xfs_bmap_cancel(&free_list
);
949 * Have to come here with the inode locked and either
950 * (held and in the transaction) or (not in the transaction).
951 * If the inode isn't held then cancel would iput it, but
952 * that's wrong since this is inactive and the vnode ref
953 * count is 0 already.
954 * Cancel won't do anything to the inode if held, but it still
955 * needs to be locked until the cancel is done, if it was
956 * joined to the transaction.
958 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
959 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
966 xfs_inactive_symlink_local(
972 ASSERT(ip
->i_d
.di_size
<= XFS_IFORK_DSIZE(ip
));
974 * We're freeing a symlink which fit into
975 * the inode. Just free the memory used
976 * to hold the old symlink.
978 error
= xfs_trans_reserve(*tpp
, 0,
979 XFS_ITRUNCATE_LOG_RES(ip
->i_mount
),
980 0, XFS_TRANS_PERM_LOG_RES
,
981 XFS_ITRUNCATE_LOG_COUNT
);
984 xfs_trans_cancel(*tpp
, 0);
988 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
991 * Zero length symlinks _can_ exist.
993 if (ip
->i_df
.if_bytes
> 0) {
994 xfs_idata_realloc(ip
,
995 -(ip
->i_df
.if_bytes
),
997 ASSERT(ip
->i_df
.if_bytes
== 0);
1011 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
1014 ASSERT(ip
->i_d
.di_forkoff
!= 0);
1015 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1016 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1020 error
= xfs_attr_inactive(ip
);
1024 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1025 error
= xfs_trans_reserve(tp
, 0,
1026 XFS_IFREE_LOG_RES(mp
),
1027 0, XFS_TRANS_PERM_LOG_RES
,
1028 XFS_INACTIVE_LOG_COUNT
);
1032 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1033 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1034 xfs_trans_ihold(tp
, ip
);
1035 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1037 ASSERT(ip
->i_d
.di_anextents
== 0);
1043 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1044 xfs_trans_cancel(tp
, 0);
1047 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1055 xfs_mount_t
*mp
= ip
->i_mount
;
1058 if (!S_ISREG(ip
->i_d
.di_mode
) || (ip
->i_d
.di_mode
== 0))
1061 /* If this is a read-only mount, don't do this (would generate I/O) */
1062 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1065 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1069 * If we are using filestreams, and we have an unlinked
1070 * file that we are processing the last close on, then nothing
1071 * will be able to reopen and write to this file. Purge this
1072 * inode from the filestreams cache so that it doesn't delay
1073 * teardown of the inode.
1075 if ((ip
->i_d
.di_nlink
== 0) && xfs_inode_is_filestream(ip
))
1076 xfs_filestream_deassociate(ip
);
1079 * If we previously truncated this file and removed old data
1080 * in the process, we want to initiate "early" writeout on
1081 * the last close. This is an attempt to combat the notorious
1082 * NULL files problem which is particularly noticable from a
1083 * truncate down, buffered (re-)write (delalloc), followed by
1084 * a crash. What we are effectively doing here is
1085 * significantly reducing the time window where we'd otherwise
1086 * be exposed to that problem.
1088 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
1089 if (truncated
&& VN_DIRTY(VFS_I(ip
)) && ip
->i_delayed_blks
> 0)
1090 xfs_flush_pages(ip
, 0, -1, XBF_ASYNC
, FI_NONE
);
1093 if (ip
->i_d
.di_nlink
!= 0) {
1094 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1095 ((ip
->i_size
> 0) || (VN_CACHED(VFS_I(ip
)) > 0 ||
1096 ip
->i_delayed_blks
> 0)) &&
1097 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
)) &&
1098 (!(ip
->i_d
.di_flags
&
1099 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)))) {
1102 * If we can't get the iolock just skip truncating
1103 * the blocks past EOF because we could deadlock
1104 * with the mmap_sem otherwise. We'll get another
1105 * chance to drop them once the last reference to
1106 * the inode is dropped, so we'll never leak blocks
1109 error
= xfs_free_eofblocks(mp
, ip
,
1110 XFS_FREE_EOF_TRYLOCK
);
1122 * This is called when the vnode reference count for the vnode
1123 * goes to zero. If the file has been unlinked, then it must
1124 * now be truncated. Also, we clear all of the read-ahead state
1125 * kept for the inode here since the file is now closed.
1131 xfs_bmap_free_t free_list
;
1132 xfs_fsblock_t first_block
;
1139 xfs_itrace_entry(ip
);
1142 * If the inode is already free, then there can be nothing
1145 if (ip
->i_d
.di_mode
== 0 || is_bad_inode(VFS_I(ip
))) {
1146 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1147 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1148 return VN_INACTIVE_CACHE
;
1152 * Only do a truncate if it's a regular file with
1153 * some actual space in it. It's OK to look at the
1154 * inode's fields without the lock because we're the
1155 * only one with a reference to the inode.
1157 truncate
= ((ip
->i_d
.di_nlink
== 0) &&
1158 ((ip
->i_d
.di_size
!= 0) || (ip
->i_size
!= 0) ||
1159 (ip
->i_d
.di_nextents
> 0) || (ip
->i_delayed_blks
> 0)) &&
1160 ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
));
1164 if (ip
->i_d
.di_nlink
== 0 && DM_EVENT_ENABLED(ip
, DM_EVENT_DESTROY
))
1165 XFS_SEND_DESTROY(mp
, ip
, DM_RIGHT_NULL
);
1169 /* If this is a read-only mount, don't do this (would generate I/O) */
1170 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1173 if (ip
->i_d
.di_nlink
!= 0) {
1174 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1175 ((ip
->i_size
> 0) || (VN_CACHED(VFS_I(ip
)) > 0 ||
1176 ip
->i_delayed_blks
> 0)) &&
1177 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) &&
1178 (!(ip
->i_d
.di_flags
&
1179 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)) ||
1180 (ip
->i_delayed_blks
!= 0)))) {
1181 error
= xfs_free_eofblocks(mp
, ip
, 0);
1183 return VN_INACTIVE_CACHE
;
1188 ASSERT(ip
->i_d
.di_nlink
== 0);
1190 error
= xfs_qm_dqattach(ip
, 0);
1192 return VN_INACTIVE_CACHE
;
1194 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1197 * Do the xfs_itruncate_start() call before
1198 * reserving any log space because itruncate_start
1199 * will call into the buffer cache and we can't
1200 * do that within a transaction.
1202 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1204 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
, 0);
1206 xfs_trans_cancel(tp
, 0);
1207 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1208 return VN_INACTIVE_CACHE
;
1211 error
= xfs_trans_reserve(tp
, 0,
1212 XFS_ITRUNCATE_LOG_RES(mp
),
1213 0, XFS_TRANS_PERM_LOG_RES
,
1214 XFS_ITRUNCATE_LOG_COUNT
);
1216 /* Don't call itruncate_cleanup */
1217 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1218 xfs_trans_cancel(tp
, 0);
1219 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1220 return VN_INACTIVE_CACHE
;
1223 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1224 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1225 xfs_trans_ihold(tp
, ip
);
1228 * normally, we have to run xfs_itruncate_finish sync.
1229 * But if filesystem is wsync and we're in the inactive
1230 * path, then we know that nlink == 0, and that the
1231 * xaction that made nlink == 0 is permanently committed
1232 * since xfs_remove runs as a synchronous transaction.
1234 error
= xfs_itruncate_finish(&tp
, ip
, 0, XFS_DATA_FORK
,
1235 (!(mp
->m_flags
& XFS_MOUNT_WSYNC
) ? 1 : 0));
1238 xfs_trans_cancel(tp
,
1239 XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1240 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1241 return VN_INACTIVE_CACHE
;
1243 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
) {
1246 * If we get an error while cleaning up a
1247 * symlink we bail out.
1249 error
= (ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
)) ?
1250 xfs_inactive_symlink_rmt(ip
, &tp
) :
1251 xfs_inactive_symlink_local(ip
, &tp
);
1255 return VN_INACTIVE_CACHE
;
1258 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1259 xfs_trans_ihold(tp
, ip
);
1261 error
= xfs_trans_reserve(tp
, 0,
1262 XFS_IFREE_LOG_RES(mp
),
1263 0, XFS_TRANS_PERM_LOG_RES
,
1264 XFS_INACTIVE_LOG_COUNT
);
1266 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1267 xfs_trans_cancel(tp
, 0);
1268 return VN_INACTIVE_CACHE
;
1271 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1272 xfs_trans_ijoin(tp
, ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1273 xfs_trans_ihold(tp
, ip
);
1277 * If there are attributes associated with the file
1278 * then blow them away now. The code calls a routine
1279 * that recursively deconstructs the attribute fork.
1280 * We need to just commit the current transaction
1281 * because we can't use it for xfs_attr_inactive().
1283 if (ip
->i_d
.di_anextents
> 0) {
1284 error
= xfs_inactive_attrs(ip
, &tp
);
1286 * If we got an error, the transaction is already
1287 * cancelled, and the inode is unlocked. Just get out.
1290 return VN_INACTIVE_CACHE
;
1291 } else if (ip
->i_afp
) {
1292 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1298 xfs_bmap_init(&free_list
, &first_block
);
1299 error
= xfs_ifree(tp
, ip
, &free_list
);
1302 * If we fail to free the inode, shut down. The cancel
1303 * might do that, we need to make sure. Otherwise the
1304 * inode might be lost for a long time or forever.
1306 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1308 "xfs_inactive: xfs_ifree() returned an error = %d on %s",
1309 error
, mp
->m_fsname
);
1310 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1312 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
1315 * Credit the quota account(s). The inode is gone.
1317 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1320 * Just ignore errors at this point. There is nothing we can
1321 * do except to try to keep going. Make sure it's not a silent
1324 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1326 xfs_fs_cmn_err(CE_NOTE
, mp
, "xfs_inactive: "
1327 "xfs_bmap_finish() returned error %d", error
);
1328 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1330 xfs_fs_cmn_err(CE_NOTE
, mp
, "xfs_inactive: "
1331 "xfs_trans_commit() returned error %d", error
);
1335 * Release the dquots held by inode, if any.
1337 xfs_qm_dqdetach(ip
);
1338 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1341 return VN_INACTIVE_CACHE
;
1345 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1346 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1347 * ci_name->name will point to a the actual name (caller must free) or
1348 * will be set to NULL if an exact match is found.
1353 struct xfs_name
*name
,
1355 struct xfs_name
*ci_name
)
1361 xfs_itrace_entry(dp
);
1363 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
1364 return XFS_ERROR(EIO
);
1366 lock_mode
= xfs_ilock_map_shared(dp
);
1367 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
1368 xfs_iunlock_map_shared(dp
, lock_mode
);
1373 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
, 0);
1381 kmem_free(ci_name
->name
);
1390 struct xfs_name
*name
,
1396 int is_dir
= S_ISDIR(mode
);
1397 struct xfs_mount
*mp
= dp
->i_mount
;
1398 struct xfs_inode
*ip
= NULL
;
1399 struct xfs_trans
*tp
= NULL
;
1401 xfs_bmap_free_t free_list
;
1402 xfs_fsblock_t first_block
;
1403 boolean_t unlock_dp_on_error
= B_FALSE
;
1407 struct xfs_dquot
*udqp
= NULL
;
1408 struct xfs_dquot
*gdqp
= NULL
;
1413 xfs_itrace_entry(dp
);
1415 if (XFS_FORCED_SHUTDOWN(mp
))
1416 return XFS_ERROR(EIO
);
1418 if (DM_EVENT_ENABLED(dp
, DM_EVENT_CREATE
)) {
1419 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_CREATE
,
1420 dp
, DM_RIGHT_NULL
, NULL
,
1421 DM_RIGHT_NULL
, name
->name
, NULL
,
1428 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1429 prid
= dp
->i_d
.di_projid
;
1434 * Make sure that we have allocated dquot(s) on disk.
1436 error
= xfs_qm_vop_dqalloc(dp
, current_fsuid(), current_fsgid(), prid
,
1437 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
1443 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1444 log_res
= XFS_MKDIR_LOG_RES(mp
);
1445 log_count
= XFS_MKDIR_LOG_COUNT
;
1446 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
1448 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1449 log_res
= XFS_CREATE_LOG_RES(mp
);
1450 log_count
= XFS_CREATE_LOG_COUNT
;
1451 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1454 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1457 * Initially assume that the file does not exist and
1458 * reserve the resources for that case. If that is not
1459 * the case we'll drop the one we have and get a more
1460 * appropriate transaction later.
1462 error
= xfs_trans_reserve(tp
, resblks
, log_res
, 0,
1463 XFS_TRANS_PERM_LOG_RES
, log_count
);
1464 if (error
== ENOSPC
) {
1465 /* flush outstanding delalloc blocks and retry */
1466 xfs_flush_inodes(dp
);
1467 error
= xfs_trans_reserve(tp
, resblks
, log_res
, 0,
1468 XFS_TRANS_PERM_LOG_RES
, log_count
);
1470 if (error
== ENOSPC
) {
1471 /* No space at all so try a "no-allocation" reservation */
1473 error
= xfs_trans_reserve(tp
, 0, log_res
, 0,
1474 XFS_TRANS_PERM_LOG_RES
, log_count
);
1478 goto out_trans_cancel
;
1481 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1482 unlock_dp_on_error
= B_TRUE
;
1485 * Check for directory link count overflow.
1487 if (is_dir
&& dp
->i_d
.di_nlink
>= XFS_MAXLINK
) {
1488 error
= XFS_ERROR(EMLINK
);
1489 goto out_trans_cancel
;
1492 xfs_bmap_init(&free_list
, &first_block
);
1495 * Reserve disk quota and the inode.
1497 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
, resblks
, 1, 0);
1499 goto out_trans_cancel
;
1501 error
= xfs_dir_canenter(tp
, dp
, name
, resblks
);
1503 goto out_trans_cancel
;
1506 * A newly created regular or special file just has one directory
1507 * entry pointing to them, but a directory also the "." entry
1508 * pointing to itself.
1510 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
, credp
,
1511 prid
, resblks
> 0, &ip
, &committed
);
1513 if (error
== ENOSPC
)
1514 goto out_trans_cancel
;
1515 goto out_trans_abort
;
1519 * At this point, we've gotten a newly allocated inode.
1520 * It is locked (and joined to the transaction).
1522 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
1525 * Now we join the directory inode to the transaction. We do not do it
1526 * earlier because xfs_dir_ialloc might commit the previous transaction
1527 * (and release all the locks). An error from here on will result in
1528 * the transaction cancel unlocking dp so don't do it explicitly in the
1532 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1533 unlock_dp_on_error
= B_FALSE
;
1535 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1536 &first_block
, &free_list
, resblks
?
1537 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1539 ASSERT(error
!= ENOSPC
);
1540 goto out_trans_abort
;
1542 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1543 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1546 error
= xfs_dir_init(tp
, ip
, dp
);
1548 goto out_bmap_cancel
;
1550 error
= xfs_bumplink(tp
, dp
);
1552 goto out_bmap_cancel
;
1556 * If this is a synchronous mount, make sure that the
1557 * create transaction goes to disk before returning to
1560 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1561 xfs_trans_set_sync(tp
);
1564 * Attach the dquot(s) to the inodes and modify them incore.
1565 * These ids of the inode couldn't have changed since the new
1566 * inode has been locked ever since it was created.
1568 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
);
1571 * xfs_trans_commit normally decrements the vnode ref count
1572 * when it unlocks the inode. Since we want to return the
1573 * vnode to the caller, we bump the vnode ref count now.
1577 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1579 goto out_abort_rele
;
1581 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1587 xfs_qm_dqrele(udqp
);
1588 xfs_qm_dqrele(gdqp
);
1592 /* Fallthrough to std_return with error = 0 */
1594 if (DM_EVENT_ENABLED(dp
, DM_EVENT_POSTCREATE
)) {
1595 XFS_SEND_NAMESP(mp
, DM_EVENT_POSTCREATE
, dp
, DM_RIGHT_NULL
,
1596 ip
, DM_RIGHT_NULL
, name
->name
, NULL
, mode
,
1603 xfs_bmap_cancel(&free_list
);
1605 cancel_flags
|= XFS_TRANS_ABORT
;
1607 xfs_trans_cancel(tp
, cancel_flags
);
1609 xfs_qm_dqrele(udqp
);
1610 xfs_qm_dqrele(gdqp
);
1612 if (unlock_dp_on_error
)
1613 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1619 * Wait until after the current transaction is aborted to
1620 * release the inode. This prevents recursive transactions
1621 * and deadlocks from xfs_inactive.
1623 xfs_bmap_cancel(&free_list
);
1624 cancel_flags
|= XFS_TRANS_ABORT
;
1625 xfs_trans_cancel(tp
, cancel_flags
);
1627 unlock_dp_on_error
= B_FALSE
;
1633 int xfs_small_retries
;
1634 int xfs_middle_retries
;
1635 int xfs_lots_retries
;
1636 int xfs_lock_delays
;
1640 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1644 xfs_lock_inumorder(int lock_mode
, int subclass
)
1646 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
1647 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_IOLOCK_SHIFT
;
1648 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
))
1649 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_ILOCK_SHIFT
;
1655 * The following routine will lock n inodes in exclusive mode.
1656 * We assume the caller calls us with the inodes in i_ino order.
1658 * We need to detect deadlock where an inode that we lock
1659 * is in the AIL and we start waiting for another inode that is locked
1660 * by a thread in a long running transaction (such as truncate). This can
1661 * result in deadlock since the long running trans might need to wait
1662 * for the inode we just locked in order to push the tail and free space
1671 int attempts
= 0, i
, j
, try_lock
;
1674 ASSERT(ips
&& (inodes
>= 2)); /* we need at least two */
1680 for (; i
< inodes
; i
++) {
1683 if (i
&& (ips
[i
] == ips
[i
-1])) /* Already locked */
1687 * If try_lock is not set yet, make sure all locked inodes
1688 * are not in the AIL.
1689 * If any are, set try_lock to be used later.
1693 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
1694 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
1695 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
1702 * If any of the previous locks we have locked is in the AIL,
1703 * we must TRY to get the second and subsequent locks. If
1704 * we can't get any, we must release all we have
1709 /* try_lock must be 0 if i is 0. */
1711 * try_lock means we have an inode locked
1712 * that is in the AIL.
1715 if (!xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
))) {
1719 * Unlock all previous guys and try again.
1720 * xfs_iunlock will try to push the tail
1721 * if the inode is in the AIL.
1724 for(j
= i
- 1; j
>= 0; j
--) {
1727 * Check to see if we've already
1728 * unlocked this one.
1729 * Not the first one going back,
1730 * and the inode ptr is the same.
1732 if ((j
!= (i
- 1)) && ips
[j
] ==
1736 xfs_iunlock(ips
[j
], lock_mode
);
1739 if ((attempts
% 5) == 0) {
1740 delay(1); /* Don't just spin the CPU */
1750 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
1756 if (attempts
< 5) xfs_small_retries
++;
1757 else if (attempts
< 100) xfs_middle_retries
++;
1758 else xfs_lots_retries
++;
1766 * xfs_lock_two_inodes() can only be used to lock one type of lock
1767 * at a time - the iolock or the ilock, but not both at once. If
1768 * we lock both at once, lockdep will report false positives saying
1769 * we have violated locking orders.
1772 xfs_lock_two_inodes(
1781 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
1782 ASSERT((lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)) == 0);
1783 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
1785 if (ip0
->i_ino
> ip1
->i_ino
) {
1792 xfs_ilock(ip0
, xfs_lock_inumorder(lock_mode
, 0));
1795 * If the first lock we have locked is in the AIL, we must TRY to get
1796 * the second lock. If we can't get it, we must release the first one
1799 lp
= (xfs_log_item_t
*)ip0
->i_itemp
;
1800 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
1801 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(lock_mode
, 1))) {
1802 xfs_iunlock(ip0
, lock_mode
);
1803 if ((++attempts
% 5) == 0)
1804 delay(1); /* Don't just spin the CPU */
1808 xfs_ilock(ip1
, xfs_lock_inumorder(lock_mode
, 1));
1815 struct xfs_name
*name
,
1818 xfs_mount_t
*mp
= dp
->i_mount
;
1819 xfs_trans_t
*tp
= NULL
;
1820 int is_dir
= S_ISDIR(ip
->i_d
.di_mode
);
1822 xfs_bmap_free_t free_list
;
1823 xfs_fsblock_t first_block
;
1830 xfs_itrace_entry(dp
);
1831 xfs_itrace_entry(ip
);
1833 if (XFS_FORCED_SHUTDOWN(mp
))
1834 return XFS_ERROR(EIO
);
1836 if (DM_EVENT_ENABLED(dp
, DM_EVENT_REMOVE
)) {
1837 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_REMOVE
, dp
, DM_RIGHT_NULL
,
1838 NULL
, DM_RIGHT_NULL
, name
->name
, NULL
,
1839 ip
->i_d
.di_mode
, 0, 0);
1844 error
= xfs_qm_dqattach(dp
, 0);
1848 error
= xfs_qm_dqattach(ip
, 0);
1853 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
1854 log_count
= XFS_DEFAULT_LOG_COUNT
;
1856 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
1857 log_count
= XFS_REMOVE_LOG_COUNT
;
1859 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1862 * We try to get the real space reservation first,
1863 * allowing for directory btree deletion(s) implying
1864 * possible bmap insert(s). If we can't get the space
1865 * reservation then we use 0 instead, and avoid the bmap
1866 * btree insert(s) in the directory code by, if the bmap
1867 * insert tries to happen, instead trimming the LAST
1868 * block from the directory.
1870 resblks
= XFS_REMOVE_SPACE_RES(mp
);
1871 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
1872 XFS_TRANS_PERM_LOG_RES
, log_count
);
1873 if (error
== ENOSPC
) {
1875 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
1876 XFS_TRANS_PERM_LOG_RES
, log_count
);
1879 ASSERT(error
!= ENOSPC
);
1881 goto out_trans_cancel
;
1884 xfs_lock_two_inodes(dp
, ip
, XFS_ILOCK_EXCL
);
1887 * At this point, we've gotten both the directory and the entry
1891 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1894 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1897 * If we're removing a directory perform some additional validation.
1900 ASSERT(ip
->i_d
.di_nlink
>= 2);
1901 if (ip
->i_d
.di_nlink
!= 2) {
1902 error
= XFS_ERROR(ENOTEMPTY
);
1903 goto out_trans_cancel
;
1905 if (!xfs_dir_isempty(ip
)) {
1906 error
= XFS_ERROR(ENOTEMPTY
);
1907 goto out_trans_cancel
;
1911 xfs_bmap_init(&free_list
, &first_block
);
1912 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
,
1913 &first_block
, &free_list
, resblks
);
1915 ASSERT(error
!= ENOENT
);
1916 goto out_bmap_cancel
;
1918 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1922 * Drop the link from ip's "..".
1924 error
= xfs_droplink(tp
, dp
);
1926 goto out_bmap_cancel
;
1929 * Drop the "." link from ip to self.
1931 error
= xfs_droplink(tp
, ip
);
1933 goto out_bmap_cancel
;
1936 * When removing a non-directory we need to log the parent
1937 * inode here. For a directory this is done implicitly
1938 * by the xfs_droplink call for the ".." entry.
1940 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1944 * Drop the link from dp to ip.
1946 error
= xfs_droplink(tp
, ip
);
1948 goto out_bmap_cancel
;
1951 * Determine if this is the last link while
1952 * we are in the transaction.
1954 link_zero
= (ip
->i_d
.di_nlink
== 0);
1957 * If this is a synchronous mount, make sure that the
1958 * remove transaction goes to disk before returning to
1961 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1962 xfs_trans_set_sync(tp
);
1964 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1966 goto out_bmap_cancel
;
1968 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1973 * If we are using filestreams, kill the stream association.
1974 * If the file is still open it may get a new one but that
1975 * will get killed on last close in xfs_close() so we don't
1976 * have to worry about that.
1978 if (!is_dir
&& link_zero
&& xfs_inode_is_filestream(ip
))
1979 xfs_filestream_deassociate(ip
);
1982 if (DM_EVENT_ENABLED(dp
, DM_EVENT_POSTREMOVE
)) {
1983 XFS_SEND_NAMESP(mp
, DM_EVENT_POSTREMOVE
, dp
, DM_RIGHT_NULL
,
1984 NULL
, DM_RIGHT_NULL
, name
->name
, NULL
,
1985 ip
->i_d
.di_mode
, error
, 0);
1991 xfs_bmap_cancel(&free_list
);
1992 cancel_flags
|= XFS_TRANS_ABORT
;
1994 xfs_trans_cancel(tp
, cancel_flags
);
2002 struct xfs_name
*target_name
)
2004 xfs_mount_t
*mp
= tdp
->i_mount
;
2007 xfs_bmap_free_t free_list
;
2008 xfs_fsblock_t first_block
;
2013 xfs_itrace_entry(tdp
);
2014 xfs_itrace_entry(sip
);
2016 ASSERT(!S_ISDIR(sip
->i_d
.di_mode
));
2018 if (XFS_FORCED_SHUTDOWN(mp
))
2019 return XFS_ERROR(EIO
);
2021 if (DM_EVENT_ENABLED(tdp
, DM_EVENT_LINK
)) {
2022 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_LINK
,
2025 target_name
->name
, NULL
, 0, 0, 0);
2030 /* Return through std_return after this point. */
2032 error
= xfs_qm_dqattach(sip
, 0);
2036 error
= xfs_qm_dqattach(tdp
, 0);
2040 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
2041 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2042 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
2043 error
= xfs_trans_reserve(tp
, resblks
, XFS_LINK_LOG_RES(mp
), 0,
2044 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2045 if (error
== ENOSPC
) {
2047 error
= xfs_trans_reserve(tp
, 0, XFS_LINK_LOG_RES(mp
), 0,
2048 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
2055 xfs_lock_two_inodes(sip
, tdp
, XFS_ILOCK_EXCL
);
2058 * Increment vnode ref counts since xfs_trans_commit &
2059 * xfs_trans_cancel will both unlock the inodes and
2060 * decrement the associated ref counts.
2064 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
2065 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
2068 * If the source has too many links, we can't make any more to it.
2070 if (sip
->i_d
.di_nlink
>= XFS_MAXLINK
) {
2071 error
= XFS_ERROR(EMLINK
);
2076 * If we are using project inheritance, we only allow hard link
2077 * creation in our tree when the project IDs are the same; else
2078 * the tree quota mechanism could be circumvented.
2080 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
2081 (tdp
->i_d
.di_projid
!= sip
->i_d
.di_projid
))) {
2082 error
= XFS_ERROR(EXDEV
);
2086 error
= xfs_dir_canenter(tp
, tdp
, target_name
, resblks
);
2090 xfs_bmap_init(&free_list
, &first_block
);
2092 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
2093 &first_block
, &free_list
, resblks
);
2096 xfs_ichgtime(tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2097 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
2099 error
= xfs_bumplink(tp
, sip
);
2104 * If this is a synchronous mount, make sure that the
2105 * link transaction goes to disk before returning to
2108 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2109 xfs_trans_set_sync(tp
);
2112 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
2114 xfs_bmap_cancel(&free_list
);
2118 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2122 /* Fall through to std_return with error = 0. */
2124 if (DM_EVENT_ENABLED(sip
, DM_EVENT_POSTLINK
)) {
2125 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTLINK
,
2128 target_name
->name
, NULL
, 0, error
, 0);
2133 cancel_flags
|= XFS_TRANS_ABORT
;
2137 xfs_trans_cancel(tp
, cancel_flags
);
2144 struct xfs_name
*link_name
,
2145 const char *target_path
,
2150 xfs_mount_t
*mp
= dp
->i_mount
;
2155 xfs_bmap_free_t free_list
;
2156 xfs_fsblock_t first_block
;
2157 boolean_t unlock_dp_on_error
= B_FALSE
;
2160 xfs_fileoff_t first_fsb
;
2161 xfs_filblks_t fs_blocks
;
2163 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
2165 const char *cur_chunk
;
2170 struct xfs_dquot
*udqp
, *gdqp
;
2178 xfs_itrace_entry(dp
);
2180 if (XFS_FORCED_SHUTDOWN(mp
))
2181 return XFS_ERROR(EIO
);
2184 * Check component lengths of the target path name.
2186 pathlen
= strlen(target_path
);
2187 if (pathlen
>= MAXPATHLEN
) /* total string too long */
2188 return XFS_ERROR(ENAMETOOLONG
);
2190 if (DM_EVENT_ENABLED(dp
, DM_EVENT_SYMLINK
)) {
2191 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_SYMLINK
, dp
,
2192 DM_RIGHT_NULL
, NULL
, DM_RIGHT_NULL
,
2194 (unsigned char *)target_path
, 0, 0, 0);
2199 /* Return through std_return after this point. */
2202 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
2203 prid
= dp
->i_d
.di_projid
;
2205 prid
= (xfs_prid_t
)dfltprid
;
2208 * Make sure that we have allocated dquot(s) on disk.
2210 error
= xfs_qm_vop_dqalloc(dp
, current_fsuid(), current_fsgid(), prid
,
2211 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
2215 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SYMLINK
);
2216 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2218 * The symlink will fit into the inode data fork?
2219 * There can't be any attributes so we get the whole variable part.
2221 if (pathlen
<= XFS_LITINO(mp
))
2224 fs_blocks
= XFS_B_TO_FSB(mp
, pathlen
);
2225 resblks
= XFS_SYMLINK_SPACE_RES(mp
, link_name
->len
, fs_blocks
);
2226 error
= xfs_trans_reserve(tp
, resblks
, XFS_SYMLINK_LOG_RES(mp
), 0,
2227 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
2228 if (error
== ENOSPC
&& fs_blocks
== 0) {
2230 error
= xfs_trans_reserve(tp
, 0, XFS_SYMLINK_LOG_RES(mp
), 0,
2231 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
2238 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
2239 unlock_dp_on_error
= B_TRUE
;
2242 * Check whether the directory allows new symlinks or not.
2244 if (dp
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) {
2245 error
= XFS_ERROR(EPERM
);
2250 * Reserve disk quota : blocks and inode.
2252 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
, resblks
, 1, 0);
2257 * Check for ability to enter directory entry, if no space reserved.
2259 error
= xfs_dir_canenter(tp
, dp
, link_name
, resblks
);
2263 * Initialize the bmap freelist prior to calling either
2264 * bmapi or the directory create code.
2266 xfs_bmap_init(&free_list
, &first_block
);
2269 * Allocate an inode for the symlink.
2271 error
= xfs_dir_ialloc(&tp
, dp
, S_IFLNK
| (mode
& ~S_IFMT
),
2272 1, 0, credp
, prid
, resblks
> 0, &ip
, NULL
);
2274 if (error
== ENOSPC
)
2280 * An error after we've joined dp to the transaction will result in the
2281 * transaction cancel unlocking dp so don't do it explicitly in the
2285 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
2286 unlock_dp_on_error
= B_FALSE
;
2289 * Also attach the dquot(s) to it, if applicable.
2291 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
);
2294 resblks
-= XFS_IALLOC_SPACE_RES(mp
);
2296 * If the symlink will fit into the inode, write it inline.
2298 if (pathlen
<= XFS_IFORK_DSIZE(ip
)) {
2299 xfs_idata_realloc(ip
, pathlen
, XFS_DATA_FORK
);
2300 memcpy(ip
->i_df
.if_u1
.if_data
, target_path
, pathlen
);
2301 ip
->i_d
.di_size
= pathlen
;
2304 * The inode was initially created in extent format.
2306 ip
->i_df
.if_flags
&= ~(XFS_IFEXTENTS
| XFS_IFBROOT
);
2307 ip
->i_df
.if_flags
|= XFS_IFINLINE
;
2309 ip
->i_d
.di_format
= XFS_DINODE_FMT_LOCAL
;
2310 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_DDATA
| XFS_ILOG_CORE
);
2314 nmaps
= SYMLINK_MAPS
;
2316 error
= xfs_bmapi(tp
, ip
, first_fsb
, fs_blocks
,
2317 XFS_BMAPI_WRITE
| XFS_BMAPI_METADATA
,
2318 &first_block
, resblks
, mval
, &nmaps
,
2325 resblks
-= fs_blocks
;
2326 ip
->i_d
.di_size
= pathlen
;
2327 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2329 cur_chunk
= target_path
;
2330 for (n
= 0; n
< nmaps
; n
++) {
2331 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
2332 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
2333 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
2334 BTOBB(byte_cnt
), 0);
2335 ASSERT(bp
&& !XFS_BUF_GETERROR(bp
));
2336 if (pathlen
< byte_cnt
) {
2339 pathlen
-= byte_cnt
;
2341 memcpy(XFS_BUF_PTR(bp
), cur_chunk
, byte_cnt
);
2342 cur_chunk
+= byte_cnt
;
2344 xfs_trans_log_buf(tp
, bp
, 0, byte_cnt
- 1);
2349 * Create the directory entry for the symlink.
2351 error
= xfs_dir_createname(tp
, dp
, link_name
, ip
->i_ino
,
2352 &first_block
, &free_list
, resblks
);
2355 xfs_ichgtime(dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2356 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2359 * If this is a synchronous mount, make sure that the
2360 * symlink transaction goes to disk before returning to
2363 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2364 xfs_trans_set_sync(tp
);
2368 * xfs_trans_commit normally decrements the vnode ref count
2369 * when it unlocks the inode. Since we want to return the
2370 * vnode to the caller, we bump the vnode ref count now.
2374 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2378 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2379 xfs_qm_dqrele(udqp
);
2380 xfs_qm_dqrele(gdqp
);
2382 /* Fall through to std_return with error = 0 or errno from
2383 * xfs_trans_commit */
2385 if (DM_EVENT_ENABLED(dp
, DM_EVENT_POSTSYMLINK
)) {
2386 (void) XFS_SEND_NAMESP(mp
, DM_EVENT_POSTSYMLINK
,
2389 DM_RIGHT_NULL
, link_name
->name
,
2390 (unsigned char *)target_path
,
2401 xfs_bmap_cancel(&free_list
);
2402 cancel_flags
|= XFS_TRANS_ABORT
;
2404 xfs_trans_cancel(tp
, cancel_flags
);
2405 xfs_qm_dqrele(udqp
);
2406 xfs_qm_dqrele(gdqp
);
2408 if (unlock_dp_on_error
)
2409 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2420 xfs_mount_t
*mp
= ip
->i_mount
;
2424 if (!capable(CAP_SYS_ADMIN
))
2425 return XFS_ERROR(EPERM
);
2427 if (XFS_FORCED_SHUTDOWN(mp
))
2428 return XFS_ERROR(EIO
);
2430 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SET_DMATTRS
);
2431 error
= xfs_trans_reserve(tp
, 0, XFS_ICHANGE_LOG_RES (mp
), 0, 0, 0);
2433 xfs_trans_cancel(tp
, 0);
2436 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2437 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2439 ip
->i_d
.di_dmevmask
= evmask
;
2440 ip
->i_d
.di_dmstate
= state
;
2442 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2444 error
= xfs_trans_commit(tp
, 0);
2450 * xfs_alloc_file_space()
2451 * This routine allocates disk space for the given file.
2453 * If alloc_type == 0, this request is for an ALLOCSP type
2454 * request which will change the file size. In this case, no
2455 * DMAPI event will be generated by the call. A TRUNCATE event
2456 * will be generated later by xfs_setattr.
2458 * If alloc_type != 0, this request is for a RESVSP type
2459 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2460 * lower block boundary byte address is less than the file's
2469 xfs_alloc_file_space(
2476 xfs_mount_t
*mp
= ip
->i_mount
;
2478 xfs_filblks_t allocated_fsb
;
2479 xfs_filblks_t allocatesize_fsb
;
2480 xfs_extlen_t extsz
, temp
;
2481 xfs_fileoff_t startoffset_fsb
;
2482 xfs_fsblock_t firstfsb
;
2488 xfs_bmbt_irec_t imaps
[1], *imapp
;
2489 xfs_bmap_free_t free_list
;
2490 uint qblocks
, resblks
, resrtextents
;
2494 xfs_itrace_entry(ip
);
2496 if (XFS_FORCED_SHUTDOWN(mp
))
2497 return XFS_ERROR(EIO
);
2499 error
= xfs_qm_dqattach(ip
, 0);
2504 return XFS_ERROR(EINVAL
);
2506 rt
= XFS_IS_REALTIME_INODE(ip
);
2507 extsz
= xfs_get_extsz_hint(ip
);
2512 bmapi_flag
= XFS_BMAPI_WRITE
| (alloc_type
? XFS_BMAPI_PREALLOC
: 0);
2513 startoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
2514 allocatesize_fsb
= XFS_B_TO_FSB(mp
, count
);
2516 /* Generate a DMAPI event if needed. */
2517 if (alloc_type
!= 0 && offset
< ip
->i_size
&&
2518 (attr_flags
& XFS_ATTR_DMI
) == 0 &&
2519 DM_EVENT_ENABLED(ip
, DM_EVENT_WRITE
)) {
2520 xfs_off_t end_dmi_offset
;
2522 end_dmi_offset
= offset
+len
;
2523 if (end_dmi_offset
> ip
->i_size
)
2524 end_dmi_offset
= ip
->i_size
;
2525 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, ip
, offset
,
2526 end_dmi_offset
- offset
, 0, NULL
);
2532 * Allocate file space until done or until there is an error
2535 while (allocatesize_fsb
&& !error
) {
2539 * Determine space reservations for data/realtime.
2541 if (unlikely(extsz
)) {
2542 s
= startoffset_fsb
;
2545 e
= startoffset_fsb
+ allocatesize_fsb
;
2546 if ((temp
= do_mod(startoffset_fsb
, extsz
)))
2548 if ((temp
= do_mod(e
, extsz
)))
2552 e
= allocatesize_fsb
;
2556 resrtextents
= qblocks
= (uint
)(e
- s
);
2557 resrtextents
/= mp
->m_sb
.sb_rextsize
;
2558 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
2559 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
2562 resblks
= qblocks
= \
2563 XFS_DIOSTRAT_SPACE_RES(mp
, (uint
)(e
- s
));
2564 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
2568 * Allocate and setup the transaction.
2570 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
2571 error
= xfs_trans_reserve(tp
, resblks
,
2572 XFS_WRITE_LOG_RES(mp
), resrtextents
,
2573 XFS_TRANS_PERM_LOG_RES
,
2574 XFS_WRITE_LOG_COUNT
);
2576 * Check for running out of space
2580 * Free the transaction structure.
2582 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
2583 xfs_trans_cancel(tp
, 0);
2586 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2587 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
,
2592 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2593 xfs_trans_ihold(tp
, ip
);
2596 * Issue the xfs_bmapi() call to allocate the blocks
2598 xfs_bmap_init(&free_list
, &firstfsb
);
2599 error
= xfs_bmapi(tp
, ip
, startoffset_fsb
,
2600 allocatesize_fsb
, bmapi_flag
,
2601 &firstfsb
, 0, imapp
, &nimaps
,
2608 * Complete the transaction
2610 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2615 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2616 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2621 allocated_fsb
= imapp
->br_blockcount
;
2624 error
= XFS_ERROR(ENOSPC
);
2628 startoffset_fsb
+= allocated_fsb
;
2629 allocatesize_fsb
-= allocated_fsb
;
2632 if (error
== ENOSPC
&& (attr_flags
& XFS_ATTR_DMI
) == 0 &&
2633 DM_EVENT_ENABLED(ip
, DM_EVENT_NOSPACE
)) {
2634 error
= XFS_SEND_NAMESP(mp
, DM_EVENT_NOSPACE
,
2637 NULL
, NULL
, 0, 0, 0); /* Delay flag intentionally unused */
2639 goto retry
; /* Maybe DMAPI app. has made space */
2640 /* else fall through with error from XFS_SEND_DATA */
2645 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
2646 xfs_bmap_cancel(&free_list
);
2647 xfs_trans_unreserve_quota_nblks(tp
, ip
, qblocks
, 0, quota_flag
);
2649 error1
: /* Just cancel transaction */
2650 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
2651 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2652 goto dmapi_enospc_check
;
2656 * Zero file bytes between startoff and endoff inclusive.
2657 * The iolock is held exclusive and no blocks are buffered.
2659 * This function is used by xfs_free_file_space() to zero
2660 * partial blocks when the range to free is not block aligned.
2661 * When unreserving space with boundaries that are not block
2662 * aligned we round up the start and round down the end
2663 * boundaries and then use this function to zero the parts of
2664 * the blocks that got dropped during the rounding.
2667 xfs_zero_remaining_bytes(
2672 xfs_bmbt_irec_t imap
;
2673 xfs_fileoff_t offset_fsb
;
2674 xfs_off_t lastoffset
;
2677 xfs_mount_t
*mp
= ip
->i_mount
;
2682 * Avoid doing I/O beyond eof - it's not necessary
2683 * since nothing can read beyond eof. The space will
2684 * be zeroed when the file is extended anyway.
2686 if (startoff
>= ip
->i_size
)
2689 if (endoff
> ip
->i_size
)
2690 endoff
= ip
->i_size
;
2692 bp
= xfs_buf_get_noaddr(mp
->m_sb
.sb_blocksize
,
2693 XFS_IS_REALTIME_INODE(ip
) ?
2694 mp
->m_rtdev_targp
: mp
->m_ddev_targp
);
2696 return XFS_ERROR(ENOMEM
);
2698 for (offset
= startoff
; offset
<= endoff
; offset
= lastoffset
+ 1) {
2699 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
2701 error
= xfs_bmapi(NULL
, ip
, offset_fsb
, 1, 0,
2702 NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
2703 if (error
|| nimap
< 1)
2705 ASSERT(imap
.br_blockcount
>= 1);
2706 ASSERT(imap
.br_startoff
== offset_fsb
);
2707 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+ 1) - 1;
2708 if (lastoffset
> endoff
)
2709 lastoffset
= endoff
;
2710 if (imap
.br_startblock
== HOLESTARTBLOCK
)
2712 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
2713 if (imap
.br_state
== XFS_EXT_UNWRITTEN
)
2716 XFS_BUF_UNWRITE(bp
);
2718 XFS_BUF_SET_ADDR(bp
, xfs_fsb_to_db(ip
, imap
.br_startblock
));
2720 error
= xfs_iowait(bp
);
2722 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
2723 mp
, bp
, XFS_BUF_ADDR(bp
));
2726 memset(XFS_BUF_PTR(bp
) +
2727 (offset
- XFS_FSB_TO_B(mp
, imap
.br_startoff
)),
2728 0, lastoffset
- offset
+ 1);
2733 error
= xfs_iowait(bp
);
2735 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
2736 mp
, bp
, XFS_BUF_ADDR(bp
));
2745 * xfs_free_file_space()
2746 * This routine frees disk space for the given file.
2748 * This routine is only called by xfs_change_file_space
2749 * for an UNRESVSP type call.
2757 xfs_free_file_space(
2765 xfs_off_t end_dmi_offset
;
2766 xfs_fileoff_t endoffset_fsb
;
2768 xfs_fsblock_t firstfsb
;
2769 xfs_bmap_free_t free_list
;
2770 xfs_bmbt_irec_t imap
;
2778 xfs_fileoff_t startoffset_fsb
;
2780 int need_iolock
= 1;
2784 xfs_itrace_entry(ip
);
2786 error
= xfs_qm_dqattach(ip
, 0);
2791 if (len
<= 0) /* if nothing being freed */
2793 rt
= XFS_IS_REALTIME_INODE(ip
);
2794 startoffset_fsb
= XFS_B_TO_FSB(mp
, offset
);
2795 end_dmi_offset
= offset
+ len
;
2796 endoffset_fsb
= XFS_B_TO_FSBT(mp
, end_dmi_offset
);
2798 if (offset
< ip
->i_size
&& (attr_flags
& XFS_ATTR_DMI
) == 0 &&
2799 DM_EVENT_ENABLED(ip
, DM_EVENT_WRITE
)) {
2800 if (end_dmi_offset
> ip
->i_size
)
2801 end_dmi_offset
= ip
->i_size
;
2802 error
= XFS_SEND_DATA(mp
, DM_EVENT_WRITE
, ip
,
2803 offset
, end_dmi_offset
- offset
,
2804 AT_DELAY_FLAG(attr_flags
), NULL
);
2809 if (attr_flags
& XFS_ATTR_NOLOCK
)
2812 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
2813 /* wait for the completion of any pending DIOs */
2817 rounding
= max_t(uint
, 1 << mp
->m_sb
.sb_blocklog
, PAGE_CACHE_SIZE
);
2818 ioffset
= offset
& ~(rounding
- 1);
2820 if (VN_CACHED(VFS_I(ip
)) != 0) {
2821 error
= xfs_flushinval_pages(ip
, ioffset
, -1, FI_REMAPF_LOCKED
);
2823 goto out_unlock_iolock
;
2827 * Need to zero the stuff we're not freeing, on disk.
2828 * If it's a realtime file & can't use unwritten extents then we
2829 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2830 * will take care of it for us.
2832 if (rt
&& !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
2834 error
= xfs_bmapi(NULL
, ip
, startoffset_fsb
,
2835 1, 0, NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
2837 goto out_unlock_iolock
;
2838 ASSERT(nimap
== 0 || nimap
== 1);
2839 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
2842 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
2843 block
= imap
.br_startblock
;
2844 mod
= do_div(block
, mp
->m_sb
.sb_rextsize
);
2846 startoffset_fsb
+= mp
->m_sb
.sb_rextsize
- mod
;
2849 error
= xfs_bmapi(NULL
, ip
, endoffset_fsb
- 1,
2850 1, 0, NULL
, 0, &imap
, &nimap
, NULL
, NULL
);
2852 goto out_unlock_iolock
;
2853 ASSERT(nimap
== 0 || nimap
== 1);
2854 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
2855 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
2857 if (mod
&& (mod
!= mp
->m_sb
.sb_rextsize
))
2858 endoffset_fsb
-= mod
;
2861 if ((done
= (endoffset_fsb
<= startoffset_fsb
)))
2863 * One contiguous piece to clear
2865 error
= xfs_zero_remaining_bytes(ip
, offset
, offset
+ len
- 1);
2868 * Some full blocks, possibly two pieces to clear
2870 if (offset
< XFS_FSB_TO_B(mp
, startoffset_fsb
))
2871 error
= xfs_zero_remaining_bytes(ip
, offset
,
2872 XFS_FSB_TO_B(mp
, startoffset_fsb
) - 1);
2874 XFS_FSB_TO_B(mp
, endoffset_fsb
) < offset
+ len
)
2875 error
= xfs_zero_remaining_bytes(ip
,
2876 XFS_FSB_TO_B(mp
, endoffset_fsb
),
2881 * free file space until done or until there is an error
2883 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
2884 while (!error
&& !done
) {
2887 * allocate and setup the transaction. Allow this
2888 * transaction to dip into the reserve blocks to ensure
2889 * the freeing of the space succeeds at ENOSPC.
2891 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
2892 tp
->t_flags
|= XFS_TRANS_RESERVE
;
2893 error
= xfs_trans_reserve(tp
,
2895 XFS_WRITE_LOG_RES(mp
),
2897 XFS_TRANS_PERM_LOG_RES
,
2898 XFS_WRITE_LOG_COUNT
);
2901 * check for running out of space
2905 * Free the transaction structure.
2907 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
2908 xfs_trans_cancel(tp
, 0);
2911 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2912 error
= xfs_trans_reserve_quota(tp
, mp
,
2913 ip
->i_udquot
, ip
->i_gdquot
,
2914 resblks
, 0, XFS_QMOPT_RES_REGBLKS
);
2918 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2919 xfs_trans_ihold(tp
, ip
);
2922 * issue the bunmapi() call to free the blocks
2924 xfs_bmap_init(&free_list
, &firstfsb
);
2925 error
= xfs_bunmapi(tp
, ip
, startoffset_fsb
,
2926 endoffset_fsb
- startoffset_fsb
,
2927 0, 2, &firstfsb
, &free_list
, NULL
, &done
);
2933 * complete the transaction
2935 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2940 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2941 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2946 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
2950 xfs_bmap_cancel(&free_list
);
2952 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
2953 xfs_iunlock(ip
, need_iolock
? (XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
) :
2959 * xfs_change_file_space()
2960 * This routine allocates or frees disk space for the given file.
2961 * The user specified parameters are checked for alignment and size
2970 xfs_change_file_space(
2977 xfs_mount_t
*mp
= ip
->i_mount
;
2982 xfs_off_t startoffset
;
2987 xfs_itrace_entry(ip
);
2989 if (!S_ISREG(ip
->i_d
.di_mode
))
2990 return XFS_ERROR(EINVAL
);
2992 switch (bf
->l_whence
) {
2993 case 0: /*SEEK_SET*/
2995 case 1: /*SEEK_CUR*/
2996 bf
->l_start
+= offset
;
2998 case 2: /*SEEK_END*/
2999 bf
->l_start
+= ip
->i_size
;
3002 return XFS_ERROR(EINVAL
);
3005 llen
= bf
->l_len
> 0 ? bf
->l_len
- 1 : bf
->l_len
;
3007 if ( (bf
->l_start
< 0)
3008 || (bf
->l_start
> XFS_MAXIOFFSET(mp
))
3009 || (bf
->l_start
+ llen
< 0)
3010 || (bf
->l_start
+ llen
> XFS_MAXIOFFSET(mp
)))
3011 return XFS_ERROR(EINVAL
);
3015 startoffset
= bf
->l_start
;
3019 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
3021 * These calls do NOT zero the data space allocated to the file,
3022 * nor do they change the file size.
3024 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
3026 * These calls cause the new file data to be zeroed and the file
3027 * size to be changed.
3029 setprealloc
= clrprealloc
= 0;
3032 case XFS_IOC_RESVSP
:
3033 case XFS_IOC_RESVSP64
:
3034 error
= xfs_alloc_file_space(ip
, startoffset
, bf
->l_len
,
3041 case XFS_IOC_UNRESVSP
:
3042 case XFS_IOC_UNRESVSP64
:
3043 if ((error
= xfs_free_file_space(ip
, startoffset
, bf
->l_len
,
3048 case XFS_IOC_ALLOCSP
:
3049 case XFS_IOC_ALLOCSP64
:
3050 case XFS_IOC_FREESP
:
3051 case XFS_IOC_FREESP64
:
3052 if (startoffset
> fsize
) {
3053 error
= xfs_alloc_file_space(ip
, fsize
,
3054 startoffset
- fsize
, 0, attr_flags
);
3059 iattr
.ia_valid
= ATTR_SIZE
;
3060 iattr
.ia_size
= startoffset
;
3062 error
= xfs_setattr(ip
, &iattr
, attr_flags
);
3072 return XFS_ERROR(EINVAL
);
3076 * update the inode timestamp, mode, and prealloc flag bits
3078 tp
= xfs_trans_alloc(mp
, XFS_TRANS_WRITEID
);
3080 if ((error
= xfs_trans_reserve(tp
, 0, XFS_WRITEID_LOG_RES(mp
),
3083 xfs_trans_cancel(tp
, 0);
3087 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
3089 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
3090 xfs_trans_ihold(tp
, ip
);
3092 if ((attr_flags
& XFS_ATTR_DMI
) == 0) {
3093 ip
->i_d
.di_mode
&= ~S_ISUID
;
3096 * Note that we don't have to worry about mandatory
3097 * file locking being disabled here because we only
3098 * clear the S_ISGID bit if the Group execute bit is
3099 * on, but if it was on then mandatory locking wouldn't
3100 * have been enabled.
3102 if (ip
->i_d
.di_mode
& S_IXGRP
)
3103 ip
->i_d
.di_mode
&= ~S_ISGID
;
3105 xfs_ichgtime(ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
3108 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
3109 else if (clrprealloc
)
3110 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
3112 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
3113 xfs_trans_set_sync(tp
);
3115 error
= xfs_trans_commit(tp
, 0);
3117 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);