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_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_itable.h"
37 #include "xfs_ialloc.h"
38 #include "xfs_alloc.h"
43 #include "xfs_error.h"
44 #include "xfs_quota.h"
45 #include "xfs_utils.h"
46 #include "xfs_rtalloc.h"
47 #include "xfs_trans_space.h"
48 #include "xfs_log_priv.h"
49 #include "xfs_filestream.h"
50 #include "xfs_vnodeops.h"
51 #include "xfs_trace.h"
59 xfs_mount_t
*mp
= ip
->i_mount
;
60 struct inode
*inode
= VFS_I(ip
);
61 int mask
= iattr
->ia_valid
;
68 struct xfs_dquot
*udqp
, *gdqp
, *olddquot1
, *olddquot2
;
71 trace_xfs_setattr(ip
);
73 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
74 return XFS_ERROR(EROFS
);
76 if (XFS_FORCED_SHUTDOWN(mp
))
77 return XFS_ERROR(EIO
);
79 code
= -inode_change_ok(inode
, iattr
);
83 olddquot1
= olddquot2
= NULL
;
87 * If disk quotas is on, we make sure that the dquots do exist on disk,
88 * before we start any other transactions. Trying to do this later
89 * is messy. We don't care to take a readlock to look at the ids
90 * in inode here, because we can't hold it across the trans_reserve.
91 * If the IDs do change before we take the ilock, we're covered
92 * because the i_*dquot fields will get updated anyway.
94 if (XFS_IS_QUOTA_ON(mp
) && (mask
& (ATTR_UID
|ATTR_GID
))) {
97 if ((mask
& ATTR_UID
) && XFS_IS_UQUOTA_ON(mp
)) {
99 qflags
|= XFS_QMOPT_UQUOTA
;
101 uid
= ip
->i_d
.di_uid
;
103 if ((mask
& ATTR_GID
) && XFS_IS_GQUOTA_ON(mp
)) {
105 qflags
|= XFS_QMOPT_GQUOTA
;
107 gid
= ip
->i_d
.di_gid
;
111 * We take a reference when we initialize udqp and gdqp,
112 * so it is important that we never blindly double trip on
113 * the same variable. See xfs_create() for an example.
115 ASSERT(udqp
== NULL
);
116 ASSERT(gdqp
== NULL
);
117 code
= xfs_qm_vop_dqalloc(ip
, uid
, gid
, xfs_get_projid(ip
),
118 qflags
, &udqp
, &gdqp
);
124 * For the other attributes, we acquire the inode lock and
125 * first do an error checking pass.
128 lock_flags
= XFS_ILOCK_EXCL
;
129 if (flags
& XFS_ATTR_NOLOCK
)
131 if (!(mask
& ATTR_SIZE
)) {
132 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_NOT_SIZE
);
134 code
= xfs_trans_reserve(tp
, 0, XFS_ICHANGE_LOG_RES(mp
),
142 lock_flags
|= XFS_IOLOCK_EXCL
;
145 xfs_ilock(ip
, lock_flags
);
148 * Change file ownership. Must be the owner or privileged.
150 if (mask
& (ATTR_UID
|ATTR_GID
)) {
152 * These IDs could have changed since we last looked at them.
153 * But, we're assured that if the ownership did change
154 * while we didn't have the inode locked, inode's dquot(s)
155 * would have changed also.
157 iuid
= ip
->i_d
.di_uid
;
158 igid
= ip
->i_d
.di_gid
;
159 gid
= (mask
& ATTR_GID
) ? iattr
->ia_gid
: igid
;
160 uid
= (mask
& ATTR_UID
) ? iattr
->ia_uid
: iuid
;
163 * Do a quota reservation only if uid/gid is actually
166 if (XFS_IS_QUOTA_RUNNING(mp
) &&
167 ((XFS_IS_UQUOTA_ON(mp
) && iuid
!= uid
) ||
168 (XFS_IS_GQUOTA_ON(mp
) && igid
!= gid
))) {
170 code
= xfs_qm_vop_chown_reserve(tp
, ip
, udqp
, gdqp
,
171 capable(CAP_FOWNER
) ?
172 XFS_QMOPT_FORCE_RES
: 0);
173 if (code
) /* out of quota */
179 * Truncate file. Must have write permission and not be a directory.
181 if (mask
& ATTR_SIZE
) {
182 /* Short circuit the truncate case for zero length files */
183 if (iattr
->ia_size
== 0 &&
184 ip
->i_size
== 0 && ip
->i_d
.di_nextents
== 0) {
185 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
186 lock_flags
&= ~XFS_ILOCK_EXCL
;
187 if (mask
& ATTR_CTIME
) {
188 inode
->i_mtime
= inode
->i_ctime
=
189 current_fs_time(inode
->i_sb
);
190 xfs_mark_inode_dirty_sync(ip
);
196 if (S_ISDIR(ip
->i_d
.di_mode
)) {
197 code
= XFS_ERROR(EISDIR
);
199 } else if (!S_ISREG(ip
->i_d
.di_mode
)) {
200 code
= XFS_ERROR(EINVAL
);
205 * Make sure that the dquots are attached to the inode.
207 code
= xfs_qm_dqattach_locked(ip
, 0);
212 * Now we can make the changes. Before we join the inode
213 * to the transaction, if ATTR_SIZE is set then take care of
214 * the part of the truncation that must be done without the
215 * inode lock. This needs to be done before joining the inode
216 * to the transaction, because the inode cannot be unlocked
217 * once it is a part of the transaction.
219 if (iattr
->ia_size
> ip
->i_size
) {
221 * Do the first part of growing a file: zero any data
222 * in the last block that is beyond the old EOF. We
223 * need to do this before the inode is joined to the
224 * transaction to modify the i_size.
226 code
= xfs_zero_eof(ip
, iattr
->ia_size
, ip
->i_size
);
230 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
231 lock_flags
&= ~XFS_ILOCK_EXCL
;
234 * We are going to log the inode size change in this
235 * transaction so any previous writes that are beyond the on
236 * disk EOF and the new EOF that have not been written out need
237 * to be written here. If we do not write the data out, we
238 * expose ourselves to the null files problem.
240 * Only flush from the on disk size to the smaller of the in
241 * memory file size or the new size as that's the range we
242 * really care about here and prevents waiting for other data
243 * not within the range we care about here.
245 if (ip
->i_size
!= ip
->i_d
.di_size
&&
246 iattr
->ia_size
> ip
->i_d
.di_size
) {
247 code
= xfs_flush_pages(ip
,
248 ip
->i_d
.di_size
, iattr
->ia_size
,
254 /* wait for all I/O to complete */
257 code
= -block_truncate_page(inode
->i_mapping
, iattr
->ia_size
,
262 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SETATTR_SIZE
);
263 code
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
264 XFS_TRANS_PERM_LOG_RES
,
265 XFS_ITRUNCATE_LOG_COUNT
);
269 truncate_setsize(inode
, iattr
->ia_size
);
271 commit_flags
= XFS_TRANS_RELEASE_LOG_RES
;
272 lock_flags
|= XFS_ILOCK_EXCL
;
274 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
276 xfs_trans_ijoin(tp
, ip
);
279 * Only change the c/mtime if we are changing the size
280 * or we are explicitly asked to change it. This handles
281 * the semantic difference between truncate() and ftruncate()
282 * as implemented in the VFS.
284 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME
285 * is a special case where we need to update the times despite
286 * not having these flags set. For all other operations the
287 * VFS set these flags explicitly if it wants a timestamp
290 if (iattr
->ia_size
!= ip
->i_size
&&
291 (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))) {
292 iattr
->ia_ctime
= iattr
->ia_mtime
=
293 current_fs_time(inode
->i_sb
);
294 mask
|= ATTR_CTIME
| ATTR_MTIME
;
297 if (iattr
->ia_size
> ip
->i_size
) {
298 ip
->i_d
.di_size
= iattr
->ia_size
;
299 ip
->i_size
= iattr
->ia_size
;
300 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
301 } else if (iattr
->ia_size
<= ip
->i_size
||
302 (iattr
->ia_size
== 0 && ip
->i_d
.di_nextents
)) {
304 * signal a sync transaction unless
305 * we're truncating an already unlinked
306 * file on a wsync filesystem
308 code
= xfs_itruncate_finish(&tp
, ip
, iattr
->ia_size
,
310 ((ip
->i_d
.di_nlink
!= 0 ||
311 !(mp
->m_flags
& XFS_MOUNT_WSYNC
))
316 * Truncated "down", so we're removing references
317 * to old data here - if we now delay flushing for
318 * a long time, we expose ourselves unduly to the
319 * notorious NULL files problem. So, we mark this
320 * vnode and flush it when the file is closed, and
321 * do not wait the usual (long) time for writeout.
323 xfs_iflags_set(ip
, XFS_ITRUNCATED
);
326 xfs_trans_ijoin(tp
, ip
);
330 * Change file ownership. Must be the owner or privileged.
332 if (mask
& (ATTR_UID
|ATTR_GID
)) {
334 * CAP_FSETID overrides the following restrictions:
336 * The set-user-ID and set-group-ID bits of a file will be
337 * cleared upon successful return from chown()
339 if ((ip
->i_d
.di_mode
& (S_ISUID
|S_ISGID
)) &&
340 !capable(CAP_FSETID
)) {
341 ip
->i_d
.di_mode
&= ~(S_ISUID
|S_ISGID
);
345 * Change the ownerships and register quota modifications
346 * in the transaction.
349 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_UQUOTA_ON(mp
)) {
350 ASSERT(mask
& ATTR_UID
);
352 olddquot1
= xfs_qm_vop_chown(tp
, ip
,
353 &ip
->i_udquot
, udqp
);
355 ip
->i_d
.di_uid
= uid
;
359 if (XFS_IS_QUOTA_RUNNING(mp
) && XFS_IS_GQUOTA_ON(mp
)) {
360 ASSERT(!XFS_IS_PQUOTA_ON(mp
));
361 ASSERT(mask
& ATTR_GID
);
363 olddquot2
= xfs_qm_vop_chown(tp
, ip
,
364 &ip
->i_gdquot
, gdqp
);
366 ip
->i_d
.di_gid
= gid
;
372 * Change file access modes.
374 if (mask
& ATTR_MODE
) {
375 umode_t mode
= iattr
->ia_mode
;
377 if (!in_group_p(inode
->i_gid
) && !capable(CAP_FSETID
))
380 ip
->i_d
.di_mode
&= S_IFMT
;
381 ip
->i_d
.di_mode
|= mode
& ~S_IFMT
;
383 inode
->i_mode
&= S_IFMT
;
384 inode
->i_mode
|= mode
& ~S_IFMT
;
388 * Change file access or modified times.
390 if (mask
& ATTR_ATIME
) {
391 inode
->i_atime
= iattr
->ia_atime
;
392 ip
->i_d
.di_atime
.t_sec
= iattr
->ia_atime
.tv_sec
;
393 ip
->i_d
.di_atime
.t_nsec
= iattr
->ia_atime
.tv_nsec
;
394 ip
->i_update_core
= 1;
396 if (mask
& ATTR_CTIME
) {
397 inode
->i_ctime
= iattr
->ia_ctime
;
398 ip
->i_d
.di_ctime
.t_sec
= iattr
->ia_ctime
.tv_sec
;
399 ip
->i_d
.di_ctime
.t_nsec
= iattr
->ia_ctime
.tv_nsec
;
400 ip
->i_update_core
= 1;
402 if (mask
& ATTR_MTIME
) {
403 inode
->i_mtime
= iattr
->ia_mtime
;
404 ip
->i_d
.di_mtime
.t_sec
= iattr
->ia_mtime
.tv_sec
;
405 ip
->i_d
.di_mtime
.t_nsec
= iattr
->ia_mtime
.tv_nsec
;
406 ip
->i_update_core
= 1;
410 * And finally, log the inode core if any attribute in it
413 if (mask
& (ATTR_UID
|ATTR_GID
|ATTR_MODE
|
414 ATTR_ATIME
|ATTR_CTIME
|ATTR_MTIME
))
415 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
417 XFS_STATS_INC(xs_ig_attrchg
);
420 * If this is a synchronous mount, make sure that the
421 * transaction goes to disk before returning to the user.
422 * This is slightly sub-optimal in that truncates require
423 * two sync transactions instead of one for wsync filesystems.
424 * One for the truncate and one for the timestamps since we
425 * don't want to change the timestamps unless we're sure the
426 * truncate worked. Truncates are less than 1% of the laddis
427 * mix so this probably isn't worth the trouble to optimize.
430 if (mp
->m_flags
& XFS_MOUNT_WSYNC
)
431 xfs_trans_set_sync(tp
);
433 code
= xfs_trans_commit(tp
, commit_flags
);
435 xfs_iunlock(ip
, lock_flags
);
438 * Release any dquot(s) the inode had kept before chown.
440 xfs_qm_dqrele(olddquot1
);
441 xfs_qm_dqrele(olddquot2
);
449 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
450 * update. We could avoid this with linked transactions
451 * and passing down the transaction pointer all the way
452 * to attr_set. No previous user of the generic
453 * Posix ACL code seems to care about this issue either.
455 if ((mask
& ATTR_MODE
) && !(flags
& XFS_ATTR_NOACL
)) {
456 code
= -xfs_acl_chmod(inode
);
458 return XFS_ERROR(code
);
464 commit_flags
|= XFS_TRANS_ABORT
;
469 xfs_trans_cancel(tp
, commit_flags
);
471 if (lock_flags
!= 0) {
472 xfs_iunlock(ip
, lock_flags
);
478 * The maximum pathlen is 1024 bytes. Since the minimum file system
479 * blocksize is 512 bytes, we can get a max of 2 extents back from
482 #define SYMLINK_MAPS 2
489 xfs_mount_t
*mp
= ip
->i_mount
;
490 int pathlen
= ip
->i_d
.di_size
;
491 int nmaps
= SYMLINK_MAPS
;
492 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
499 error
= xfs_bmapi(NULL
, ip
, 0, XFS_B_TO_FSB(mp
, pathlen
), 0, NULL
, 0,
504 for (n
= 0; n
< nmaps
; n
++) {
505 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
506 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
508 bp
= xfs_buf_read(mp
->m_ddev_targp
, d
, BTOBB(byte_cnt
),
509 XBF_LOCK
| XBF_MAPPED
| XBF_DONT_BLOCK
);
510 error
= XFS_BUF_GETERROR(bp
);
512 xfs_ioerror_alert("xfs_readlink",
513 ip
->i_mount
, bp
, XFS_BUF_ADDR(bp
));
517 if (pathlen
< byte_cnt
)
521 memcpy(link
, XFS_BUF_PTR(bp
), byte_cnt
);
525 link
[ip
->i_d
.di_size
] = '\0';
537 xfs_mount_t
*mp
= ip
->i_mount
;
541 trace_xfs_readlink(ip
);
543 if (XFS_FORCED_SHUTDOWN(mp
))
544 return XFS_ERROR(EIO
);
546 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
548 pathlen
= ip
->i_d
.di_size
;
552 if (pathlen
< 0 || pathlen
> MAXPATHLEN
) {
553 xfs_alert(mp
, "%s: inode (%llu) bad symlink length (%lld)",
554 __func__
, (unsigned long long) ip
->i_ino
,
555 (long long) pathlen
);
557 return XFS_ERROR(EFSCORRUPTED
);
561 if (ip
->i_df
.if_flags
& XFS_IFINLINE
) {
562 memcpy(link
, ip
->i_df
.if_u1
.if_data
, pathlen
);
563 link
[pathlen
] = '\0';
565 error
= xfs_readlink_bmap(ip
, link
);
569 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
574 * Flags for xfs_free_eofblocks
576 #define XFS_FREE_EOF_TRYLOCK (1<<0)
579 * This is called by xfs_inactive to free any blocks beyond eof
580 * when the link count isn't zero and by xfs_dm_punch_hole() when
581 * punching a hole to EOF.
591 xfs_fileoff_t end_fsb
;
592 xfs_fileoff_t last_fsb
;
593 xfs_filblks_t map_len
;
595 xfs_bmbt_irec_t imap
;
598 * Figure out if there are any blocks beyond the end
599 * of the file. If not, then there is nothing to do.
601 end_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)ip
->i_size
));
602 last_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
603 if (last_fsb
<= end_fsb
)
605 map_len
= last_fsb
- end_fsb
;
608 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
609 error
= xfs_bmapi(NULL
, ip
, end_fsb
, map_len
, 0,
610 NULL
, 0, &imap
, &nimaps
, NULL
);
611 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
613 if (!error
&& (nimaps
!= 0) &&
614 (imap
.br_startblock
!= HOLESTARTBLOCK
||
615 ip
->i_delayed_blks
)) {
617 * Attach the dquots to the inode up front.
619 error
= xfs_qm_dqattach(ip
, 0);
624 * There are blocks after the end of file.
625 * Free them up now by truncating the file to
628 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
631 * Do the xfs_itruncate_start() call before
632 * reserving any log space because
633 * itruncate_start will call into the buffer
635 * do that within a transaction.
637 if (flags
& XFS_FREE_EOF_TRYLOCK
) {
638 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
639 xfs_trans_cancel(tp
, 0);
643 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
645 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
,
648 xfs_trans_cancel(tp
, 0);
649 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
653 error
= xfs_trans_reserve(tp
, 0,
654 XFS_ITRUNCATE_LOG_RES(mp
),
655 0, XFS_TRANS_PERM_LOG_RES
,
656 XFS_ITRUNCATE_LOG_COUNT
);
658 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
659 xfs_trans_cancel(tp
, 0);
660 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
664 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
665 xfs_trans_ijoin(tp
, ip
);
667 error
= xfs_itruncate_finish(&tp
, ip
,
672 * If we get an error at this point we
673 * simply don't bother truncating the file.
677 (XFS_TRANS_RELEASE_LOG_RES
|
680 error
= xfs_trans_commit(tp
,
681 XFS_TRANS_RELEASE_LOG_RES
);
683 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
);
689 * Free a symlink that has blocks associated with it.
692 xfs_inactive_symlink_rmt(
700 xfs_fsblock_t first_block
;
701 xfs_bmap_free_t free_list
;
704 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
712 ASSERT(ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
));
714 * We're freeing a symlink that has some
715 * blocks allocated to it. Free the
716 * blocks here. We know that we've got
717 * either 1 or 2 extents and that we can
718 * free them all in one bunmapi call.
720 ASSERT(ip
->i_d
.di_nextents
> 0 && ip
->i_d
.di_nextents
<= 2);
721 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
722 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
723 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
724 xfs_trans_cancel(tp
, 0);
729 * Lock the inode, fix the size, and join it to the transaction.
730 * Hold it so in the normal path, we still have it locked for
731 * the second transaction. In the error paths we need it
732 * held so the cancel won't rele it, see below.
734 xfs_ilock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
735 size
= (int)ip
->i_d
.di_size
;
737 xfs_trans_ijoin(tp
, ip
);
738 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
740 * Find the block(s) so we can inval and unmap them.
743 xfs_bmap_init(&free_list
, &first_block
);
744 nmaps
= ARRAY_SIZE(mval
);
745 if ((error
= xfs_bmapi(tp
, ip
, 0, XFS_B_TO_FSB(mp
, size
),
746 XFS_BMAPI_METADATA
, &first_block
, 0, mval
, &nmaps
,
750 * Invalidate the block(s).
752 for (i
= 0; i
< nmaps
; i
++) {
753 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
754 XFS_FSB_TO_DADDR(mp
, mval
[i
].br_startblock
),
755 XFS_FSB_TO_BB(mp
, mval
[i
].br_blockcount
), 0);
756 xfs_trans_binval(tp
, bp
);
759 * Unmap the dead block(s) to the free_list.
761 if ((error
= xfs_bunmapi(tp
, ip
, 0, size
, XFS_BMAPI_METADATA
, nmaps
,
762 &first_block
, &free_list
, &done
)))
766 * Commit the first transaction. This logs the EFI and the inode.
768 if ((error
= xfs_bmap_finish(&tp
, &free_list
, &committed
)))
771 * The transaction must have been committed, since there were
772 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
773 * The new tp has the extent freeing and EFDs.
777 * The first xact was committed, so add the inode to the new one.
778 * Mark it dirty so it will be logged and moved forward in the log as
779 * part of every commit.
781 xfs_trans_ijoin(tp
, ip
);
782 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
784 * Get a new, empty transaction to return to our caller.
786 ntp
= xfs_trans_dup(tp
);
788 * Commit the transaction containing extent freeing and EFDs.
789 * If we get an error on the commit here or on the reserve below,
790 * we need to unlock the inode since the new transaction doesn't
791 * have the inode attached.
793 error
= xfs_trans_commit(tp
, 0);
796 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
800 * transaction commit worked ok so we can drop the extra ticket
801 * reference that we gained in xfs_trans_dup()
803 xfs_log_ticket_put(tp
->t_ticket
);
806 * Remove the memory for extent descriptions (just bookkeeping).
808 if (ip
->i_df
.if_bytes
)
809 xfs_idata_realloc(ip
, -ip
->i_df
.if_bytes
, XFS_DATA_FORK
);
810 ASSERT(ip
->i_df
.if_bytes
== 0);
812 * Put an itruncate log reservation in the new transaction
815 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
816 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
817 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
821 * Return with the inode locked but not joined to the transaction.
827 xfs_bmap_cancel(&free_list
);
830 * Have to come here with the inode locked and either
831 * (held and in the transaction) or (not in the transaction).
832 * If the inode isn't held then cancel would iput it, but
833 * that's wrong since this is inactive and the vnode ref
834 * count is 0 already.
835 * Cancel won't do anything to the inode if held, but it still
836 * needs to be locked until the cancel is done, if it was
837 * joined to the transaction.
839 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
840 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
847 xfs_inactive_symlink_local(
853 ASSERT(ip
->i_d
.di_size
<= XFS_IFORK_DSIZE(ip
));
855 * We're freeing a symlink which fit into
856 * the inode. Just free the memory used
857 * to hold the old symlink.
859 error
= xfs_trans_reserve(*tpp
, 0,
860 XFS_ITRUNCATE_LOG_RES(ip
->i_mount
),
861 0, XFS_TRANS_PERM_LOG_RES
,
862 XFS_ITRUNCATE_LOG_COUNT
);
865 xfs_trans_cancel(*tpp
, 0);
869 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
872 * Zero length symlinks _can_ exist.
874 if (ip
->i_df
.if_bytes
> 0) {
875 xfs_idata_realloc(ip
,
876 -(ip
->i_df
.if_bytes
),
878 ASSERT(ip
->i_df
.if_bytes
== 0);
892 ASSERT(xfs_isilocked(ip
, XFS_IOLOCK_EXCL
));
895 ASSERT(ip
->i_d
.di_forkoff
!= 0);
896 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
897 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
901 error
= xfs_attr_inactive(ip
);
905 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
906 error
= xfs_trans_reserve(tp
, 0,
907 XFS_IFREE_LOG_RES(mp
),
908 0, XFS_TRANS_PERM_LOG_RES
,
909 XFS_INACTIVE_LOG_COUNT
);
913 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
914 xfs_trans_ijoin(tp
, ip
);
915 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
917 ASSERT(ip
->i_d
.di_anextents
== 0);
923 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
924 xfs_trans_cancel(tp
, 0);
927 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
935 xfs_mount_t
*mp
= ip
->i_mount
;
938 if (!S_ISREG(ip
->i_d
.di_mode
) || (ip
->i_d
.di_mode
== 0))
941 /* If this is a read-only mount, don't do this (would generate I/O) */
942 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
945 if (!XFS_FORCED_SHUTDOWN(mp
)) {
949 * If we are using filestreams, and we have an unlinked
950 * file that we are processing the last close on, then nothing
951 * will be able to reopen and write to this file. Purge this
952 * inode from the filestreams cache so that it doesn't delay
953 * teardown of the inode.
955 if ((ip
->i_d
.di_nlink
== 0) && xfs_inode_is_filestream(ip
))
956 xfs_filestream_deassociate(ip
);
959 * If we previously truncated this file and removed old data
960 * in the process, we want to initiate "early" writeout on
961 * the last close. This is an attempt to combat the notorious
962 * NULL files problem which is particularly noticeable from a
963 * truncate down, buffered (re-)write (delalloc), followed by
964 * a crash. What we are effectively doing here is
965 * significantly reducing the time window where we'd otherwise
966 * be exposed to that problem.
968 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
970 xfs_iflags_clear(ip
, XFS_IDIRTY_RELEASE
);
971 if (VN_DIRTY(VFS_I(ip
)) && ip
->i_delayed_blks
> 0)
972 xfs_flush_pages(ip
, 0, -1, XBF_ASYNC
, FI_NONE
);
976 if (ip
->i_d
.di_nlink
== 0)
979 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
980 ((ip
->i_size
> 0) || (VN_CACHED(VFS_I(ip
)) > 0 ||
981 ip
->i_delayed_blks
> 0)) &&
982 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
)) &&
983 (!(ip
->i_d
.di_flags
& (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)))) {
986 * If we can't get the iolock just skip truncating the blocks
987 * past EOF because we could deadlock with the mmap_sem
988 * otherwise. We'll get another chance to drop them once the
989 * last reference to the inode is dropped, so we'll never leak
990 * blocks permanently.
992 * Further, check if the inode is being opened, written and
993 * closed frequently and we have delayed allocation blocks
994 * outstanding (e.g. streaming writes from the NFS server),
995 * truncating the blocks past EOF will cause fragmentation to
998 * In this case don't do the truncation, either, but we have to
999 * be careful how we detect this case. Blocks beyond EOF show
1000 * up as i_delayed_blks even when the inode is clean, so we
1001 * need to truncate them away first before checking for a dirty
1002 * release. Hence on the first dirty close we will still remove
1003 * the speculative allocation, but after that we will leave it
1006 if (xfs_iflags_test(ip
, XFS_IDIRTY_RELEASE
))
1009 error
= xfs_free_eofblocks(mp
, ip
,
1010 XFS_FREE_EOF_TRYLOCK
);
1014 /* delalloc blocks after truncation means it really is dirty */
1015 if (ip
->i_delayed_blks
)
1016 xfs_iflags_set(ip
, XFS_IDIRTY_RELEASE
);
1024 * This is called when the vnode reference count for the vnode
1025 * goes to zero. If the file has been unlinked, then it must
1026 * now be truncated. Also, we clear all of the read-ahead state
1027 * kept for the inode here since the file is now closed.
1033 xfs_bmap_free_t free_list
;
1034 xfs_fsblock_t first_block
;
1042 * If the inode is already free, then there can be nothing
1045 if (ip
->i_d
.di_mode
== 0 || is_bad_inode(VFS_I(ip
))) {
1046 ASSERT(ip
->i_df
.if_real_bytes
== 0);
1047 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
1048 return VN_INACTIVE_CACHE
;
1052 * Only do a truncate if it's a regular file with
1053 * some actual space in it. It's OK to look at the
1054 * inode's fields without the lock because we're the
1055 * only one with a reference to the inode.
1057 truncate
= ((ip
->i_d
.di_nlink
== 0) &&
1058 ((ip
->i_d
.di_size
!= 0) || (ip
->i_size
!= 0) ||
1059 (ip
->i_d
.di_nextents
> 0) || (ip
->i_delayed_blks
> 0)) &&
1060 ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
));
1066 /* If this is a read-only mount, don't do this (would generate I/O) */
1067 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
1070 if (ip
->i_d
.di_nlink
!= 0) {
1071 if ((((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
) &&
1072 ((ip
->i_size
> 0) || (VN_CACHED(VFS_I(ip
)) > 0 ||
1073 ip
->i_delayed_blks
> 0)) &&
1074 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) &&
1075 (!(ip
->i_d
.di_flags
&
1076 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)) ||
1077 (ip
->i_delayed_blks
!= 0)))) {
1078 error
= xfs_free_eofblocks(mp
, ip
, 0);
1080 return VN_INACTIVE_CACHE
;
1085 ASSERT(ip
->i_d
.di_nlink
== 0);
1087 error
= xfs_qm_dqattach(ip
, 0);
1089 return VN_INACTIVE_CACHE
;
1091 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
1094 * Do the xfs_itruncate_start() call before
1095 * reserving any log space because itruncate_start
1096 * will call into the buffer cache and we can't
1097 * do that within a transaction.
1099 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1101 error
= xfs_itruncate_start(ip
, XFS_ITRUNC_DEFINITE
, 0);
1103 xfs_trans_cancel(tp
, 0);
1104 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1105 return VN_INACTIVE_CACHE
;
1108 error
= xfs_trans_reserve(tp
, 0,
1109 XFS_ITRUNCATE_LOG_RES(mp
),
1110 0, XFS_TRANS_PERM_LOG_RES
,
1111 XFS_ITRUNCATE_LOG_COUNT
);
1113 /* Don't call itruncate_cleanup */
1114 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1115 xfs_trans_cancel(tp
, 0);
1116 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
1117 return VN_INACTIVE_CACHE
;
1120 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1121 xfs_trans_ijoin(tp
, ip
);
1124 * normally, we have to run xfs_itruncate_finish sync.
1125 * But if filesystem is wsync and we're in the inactive
1126 * path, then we know that nlink == 0, and that the
1127 * xaction that made nlink == 0 is permanently committed
1128 * since xfs_remove runs as a synchronous transaction.
1130 error
= xfs_itruncate_finish(&tp
, ip
, 0, XFS_DATA_FORK
,
1131 (!(mp
->m_flags
& XFS_MOUNT_WSYNC
) ? 1 : 0));
1134 xfs_trans_cancel(tp
,
1135 XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1136 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1137 return VN_INACTIVE_CACHE
;
1139 } else if ((ip
->i_d
.di_mode
& S_IFMT
) == S_IFLNK
) {
1142 * If we get an error while cleaning up a
1143 * symlink we bail out.
1145 error
= (ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
)) ?
1146 xfs_inactive_symlink_rmt(ip
, &tp
) :
1147 xfs_inactive_symlink_local(ip
, &tp
);
1151 return VN_INACTIVE_CACHE
;
1154 xfs_trans_ijoin(tp
, ip
);
1156 error
= xfs_trans_reserve(tp
, 0,
1157 XFS_IFREE_LOG_RES(mp
),
1158 0, XFS_TRANS_PERM_LOG_RES
,
1159 XFS_INACTIVE_LOG_COUNT
);
1161 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1162 xfs_trans_cancel(tp
, 0);
1163 return VN_INACTIVE_CACHE
;
1166 xfs_ilock(ip
, XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
);
1167 xfs_trans_ijoin(tp
, ip
);
1171 * If there are attributes associated with the file
1172 * then blow them away now. The code calls a routine
1173 * that recursively deconstructs the attribute fork.
1174 * We need to just commit the current transaction
1175 * because we can't use it for xfs_attr_inactive().
1177 if (ip
->i_d
.di_anextents
> 0) {
1178 error
= xfs_inactive_attrs(ip
, &tp
);
1180 * If we got an error, the transaction is already
1181 * cancelled, and the inode is unlocked. Just get out.
1184 return VN_INACTIVE_CACHE
;
1185 } else if (ip
->i_afp
) {
1186 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
1192 xfs_bmap_init(&free_list
, &first_block
);
1193 error
= xfs_ifree(tp
, ip
, &free_list
);
1196 * If we fail to free the inode, shut down. The cancel
1197 * might do that, we need to make sure. Otherwise the
1198 * inode might be lost for a long time or forever.
1200 if (!XFS_FORCED_SHUTDOWN(mp
)) {
1201 xfs_notice(mp
, "%s: xfs_ifree returned error %d",
1203 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1205 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
1208 * Credit the quota account(s). The inode is gone.
1210 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
1213 * Just ignore errors at this point. There is nothing we can
1214 * do except to try to keep going. Make sure it's not a silent
1217 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1219 xfs_notice(mp
, "%s: xfs_bmap_finish returned error %d",
1221 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1223 xfs_notice(mp
, "%s: xfs_trans_commit returned error %d",
1228 * Release the dquots held by inode, if any.
1230 xfs_qm_dqdetach(ip
);
1231 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
| XFS_ILOCK_EXCL
);
1234 return VN_INACTIVE_CACHE
;
1238 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
1239 * is allowed, otherwise it has to be an exact match. If a CI match is found,
1240 * ci_name->name will point to a the actual name (caller must free) or
1241 * will be set to NULL if an exact match is found.
1246 struct xfs_name
*name
,
1248 struct xfs_name
*ci_name
)
1254 trace_xfs_lookup(dp
, name
);
1256 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
1257 return XFS_ERROR(EIO
);
1259 lock_mode
= xfs_ilock_map_shared(dp
);
1260 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
1261 xfs_iunlock_map_shared(dp
, lock_mode
);
1266 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
);
1274 kmem_free(ci_name
->name
);
1283 struct xfs_name
*name
,
1288 int is_dir
= S_ISDIR(mode
);
1289 struct xfs_mount
*mp
= dp
->i_mount
;
1290 struct xfs_inode
*ip
= NULL
;
1291 struct xfs_trans
*tp
= NULL
;
1293 xfs_bmap_free_t free_list
;
1294 xfs_fsblock_t first_block
;
1295 boolean_t unlock_dp_on_error
= B_FALSE
;
1299 struct xfs_dquot
*udqp
= NULL
;
1300 struct xfs_dquot
*gdqp
= NULL
;
1305 trace_xfs_create(dp
, name
);
1307 if (XFS_FORCED_SHUTDOWN(mp
))
1308 return XFS_ERROR(EIO
);
1310 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1311 prid
= xfs_get_projid(dp
);
1313 prid
= XFS_PROJID_DEFAULT
;
1316 * Make sure that we have allocated dquot(s) on disk.
1318 error
= xfs_qm_vop_dqalloc(dp
, current_fsuid(), current_fsgid(), prid
,
1319 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
1325 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
1326 log_res
= XFS_MKDIR_LOG_RES(mp
);
1327 log_count
= XFS_MKDIR_LOG_COUNT
;
1328 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
1330 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
1331 log_res
= XFS_CREATE_LOG_RES(mp
);
1332 log_count
= XFS_CREATE_LOG_COUNT
;
1333 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
1336 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1339 * Initially assume that the file does not exist and
1340 * reserve the resources for that case. If that is not
1341 * the case we'll drop the one we have and get a more
1342 * appropriate transaction later.
1344 error
= xfs_trans_reserve(tp
, resblks
, log_res
, 0,
1345 XFS_TRANS_PERM_LOG_RES
, log_count
);
1346 if (error
== ENOSPC
) {
1347 /* flush outstanding delalloc blocks and retry */
1348 xfs_flush_inodes(dp
);
1349 error
= xfs_trans_reserve(tp
, resblks
, log_res
, 0,
1350 XFS_TRANS_PERM_LOG_RES
, log_count
);
1352 if (error
== ENOSPC
) {
1353 /* No space at all so try a "no-allocation" reservation */
1355 error
= xfs_trans_reserve(tp
, 0, log_res
, 0,
1356 XFS_TRANS_PERM_LOG_RES
, log_count
);
1360 goto out_trans_cancel
;
1363 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1364 unlock_dp_on_error
= B_TRUE
;
1367 * Check for directory link count overflow.
1369 if (is_dir
&& dp
->i_d
.di_nlink
>= XFS_MAXLINK
) {
1370 error
= XFS_ERROR(EMLINK
);
1371 goto out_trans_cancel
;
1374 xfs_bmap_init(&free_list
, &first_block
);
1377 * Reserve disk quota and the inode.
1379 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
, resblks
, 1, 0);
1381 goto out_trans_cancel
;
1383 error
= xfs_dir_canenter(tp
, dp
, name
, resblks
);
1385 goto out_trans_cancel
;
1388 * A newly created regular or special file just has one directory
1389 * entry pointing to them, but a directory also the "." entry
1390 * pointing to itself.
1392 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
,
1393 prid
, resblks
> 0, &ip
, &committed
);
1395 if (error
== ENOSPC
)
1396 goto out_trans_cancel
;
1397 goto out_trans_abort
;
1401 * Now we join the directory inode to the transaction. We do not do it
1402 * earlier because xfs_dir_ialloc might commit the previous transaction
1403 * (and release all the locks). An error from here on will result in
1404 * the transaction cancel unlocking dp so don't do it explicitly in the
1407 xfs_trans_ijoin_ref(tp
, dp
, XFS_ILOCK_EXCL
);
1408 unlock_dp_on_error
= B_FALSE
;
1410 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
1411 &first_block
, &free_list
, resblks
?
1412 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
1414 ASSERT(error
!= ENOSPC
);
1415 goto out_trans_abort
;
1417 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1418 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1421 error
= xfs_dir_init(tp
, ip
, dp
);
1423 goto out_bmap_cancel
;
1425 error
= xfs_bumplink(tp
, dp
);
1427 goto out_bmap_cancel
;
1431 * If this is a synchronous mount, make sure that the
1432 * create transaction goes to disk before returning to
1435 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1436 xfs_trans_set_sync(tp
);
1439 * Attach the dquot(s) to the inodes and modify them incore.
1440 * These ids of the inode couldn't have changed since the new
1441 * inode has been locked ever since it was created.
1443 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
);
1445 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1447 goto out_bmap_cancel
;
1449 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1451 goto out_release_inode
;
1453 xfs_qm_dqrele(udqp
);
1454 xfs_qm_dqrele(gdqp
);
1460 xfs_bmap_cancel(&free_list
);
1462 cancel_flags
|= XFS_TRANS_ABORT
;
1464 xfs_trans_cancel(tp
, cancel_flags
);
1467 * Wait until after the current transaction is aborted to
1468 * release the inode. This prevents recursive transactions
1469 * and deadlocks from xfs_inactive.
1474 xfs_qm_dqrele(udqp
);
1475 xfs_qm_dqrele(gdqp
);
1477 if (unlock_dp_on_error
)
1478 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1484 int xfs_small_retries
;
1485 int xfs_middle_retries
;
1486 int xfs_lots_retries
;
1487 int xfs_lock_delays
;
1491 * Bump the subclass so xfs_lock_inodes() acquires each lock with
1495 xfs_lock_inumorder(int lock_mode
, int subclass
)
1497 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
1498 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_IOLOCK_SHIFT
;
1499 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
))
1500 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_ILOCK_SHIFT
;
1506 * The following routine will lock n inodes in exclusive mode.
1507 * We assume the caller calls us with the inodes in i_ino order.
1509 * We need to detect deadlock where an inode that we lock
1510 * is in the AIL and we start waiting for another inode that is locked
1511 * by a thread in a long running transaction (such as truncate). This can
1512 * result in deadlock since the long running trans might need to wait
1513 * for the inode we just locked in order to push the tail and free space
1522 int attempts
= 0, i
, j
, try_lock
;
1525 ASSERT(ips
&& (inodes
>= 2)); /* we need at least two */
1531 for (; i
< inodes
; i
++) {
1534 if (i
&& (ips
[i
] == ips
[i
-1])) /* Already locked */
1538 * If try_lock is not set yet, make sure all locked inodes
1539 * are not in the AIL.
1540 * If any are, set try_lock to be used later.
1544 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
1545 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
1546 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
1553 * If any of the previous locks we have locked is in the AIL,
1554 * we must TRY to get the second and subsequent locks. If
1555 * we can't get any, we must release all we have
1560 /* try_lock must be 0 if i is 0. */
1562 * try_lock means we have an inode locked
1563 * that is in the AIL.
1566 if (!xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
))) {
1570 * Unlock all previous guys and try again.
1571 * xfs_iunlock will try to push the tail
1572 * if the inode is in the AIL.
1575 for(j
= i
- 1; j
>= 0; j
--) {
1578 * Check to see if we've already
1579 * unlocked this one.
1580 * Not the first one going back,
1581 * and the inode ptr is the same.
1583 if ((j
!= (i
- 1)) && ips
[j
] ==
1587 xfs_iunlock(ips
[j
], lock_mode
);
1590 if ((attempts
% 5) == 0) {
1591 delay(1); /* Don't just spin the CPU */
1601 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
1607 if (attempts
< 5) xfs_small_retries
++;
1608 else if (attempts
< 100) xfs_middle_retries
++;
1609 else xfs_lots_retries
++;
1617 * xfs_lock_two_inodes() can only be used to lock one type of lock
1618 * at a time - the iolock or the ilock, but not both at once. If
1619 * we lock both at once, lockdep will report false positives saying
1620 * we have violated locking orders.
1623 xfs_lock_two_inodes(
1632 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
1633 ASSERT((lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)) == 0);
1634 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
1636 if (ip0
->i_ino
> ip1
->i_ino
) {
1643 xfs_ilock(ip0
, xfs_lock_inumorder(lock_mode
, 0));
1646 * If the first lock we have locked is in the AIL, we must TRY to get
1647 * the second lock. If we can't get it, we must release the first one
1650 lp
= (xfs_log_item_t
*)ip0
->i_itemp
;
1651 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
1652 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(lock_mode
, 1))) {
1653 xfs_iunlock(ip0
, lock_mode
);
1654 if ((++attempts
% 5) == 0)
1655 delay(1); /* Don't just spin the CPU */
1659 xfs_ilock(ip1
, xfs_lock_inumorder(lock_mode
, 1));
1666 struct xfs_name
*name
,
1669 xfs_mount_t
*mp
= dp
->i_mount
;
1670 xfs_trans_t
*tp
= NULL
;
1671 int is_dir
= S_ISDIR(ip
->i_d
.di_mode
);
1673 xfs_bmap_free_t free_list
;
1674 xfs_fsblock_t first_block
;
1681 trace_xfs_remove(dp
, name
);
1683 if (XFS_FORCED_SHUTDOWN(mp
))
1684 return XFS_ERROR(EIO
);
1686 error
= xfs_qm_dqattach(dp
, 0);
1690 error
= xfs_qm_dqattach(ip
, 0);
1695 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
1696 log_count
= XFS_DEFAULT_LOG_COUNT
;
1698 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
1699 log_count
= XFS_REMOVE_LOG_COUNT
;
1701 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1704 * We try to get the real space reservation first,
1705 * allowing for directory btree deletion(s) implying
1706 * possible bmap insert(s). If we can't get the space
1707 * reservation then we use 0 instead, and avoid the bmap
1708 * btree insert(s) in the directory code by, if the bmap
1709 * insert tries to happen, instead trimming the LAST
1710 * block from the directory.
1712 resblks
= XFS_REMOVE_SPACE_RES(mp
);
1713 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
1714 XFS_TRANS_PERM_LOG_RES
, log_count
);
1715 if (error
== ENOSPC
) {
1717 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
1718 XFS_TRANS_PERM_LOG_RES
, log_count
);
1721 ASSERT(error
!= ENOSPC
);
1723 goto out_trans_cancel
;
1726 xfs_lock_two_inodes(dp
, ip
, XFS_ILOCK_EXCL
);
1728 xfs_trans_ijoin_ref(tp
, dp
, XFS_ILOCK_EXCL
);
1729 xfs_trans_ijoin_ref(tp
, ip
, XFS_ILOCK_EXCL
);
1732 * If we're removing a directory perform some additional validation.
1735 ASSERT(ip
->i_d
.di_nlink
>= 2);
1736 if (ip
->i_d
.di_nlink
!= 2) {
1737 error
= XFS_ERROR(ENOTEMPTY
);
1738 goto out_trans_cancel
;
1740 if (!xfs_dir_isempty(ip
)) {
1741 error
= XFS_ERROR(ENOTEMPTY
);
1742 goto out_trans_cancel
;
1746 xfs_bmap_init(&free_list
, &first_block
);
1747 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
,
1748 &first_block
, &free_list
, resblks
);
1750 ASSERT(error
!= ENOENT
);
1751 goto out_bmap_cancel
;
1753 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1757 * Drop the link from ip's "..".
1759 error
= xfs_droplink(tp
, dp
);
1761 goto out_bmap_cancel
;
1764 * Drop the "." link from ip to self.
1766 error
= xfs_droplink(tp
, ip
);
1768 goto out_bmap_cancel
;
1771 * When removing a non-directory we need to log the parent
1772 * inode here. For a directory this is done implicitly
1773 * by the xfs_droplink call for the ".." entry.
1775 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1779 * Drop the link from dp to ip.
1781 error
= xfs_droplink(tp
, ip
);
1783 goto out_bmap_cancel
;
1786 * Determine if this is the last link while
1787 * we are in the transaction.
1789 link_zero
= (ip
->i_d
.di_nlink
== 0);
1792 * If this is a synchronous mount, make sure that the
1793 * remove transaction goes to disk before returning to
1796 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1797 xfs_trans_set_sync(tp
);
1799 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1801 goto out_bmap_cancel
;
1803 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1808 * If we are using filestreams, kill the stream association.
1809 * If the file is still open it may get a new one but that
1810 * will get killed on last close in xfs_close() so we don't
1811 * have to worry about that.
1813 if (!is_dir
&& link_zero
&& xfs_inode_is_filestream(ip
))
1814 xfs_filestream_deassociate(ip
);
1819 xfs_bmap_cancel(&free_list
);
1820 cancel_flags
|= XFS_TRANS_ABORT
;
1822 xfs_trans_cancel(tp
, cancel_flags
);
1831 struct xfs_name
*target_name
)
1833 xfs_mount_t
*mp
= tdp
->i_mount
;
1836 xfs_bmap_free_t free_list
;
1837 xfs_fsblock_t first_block
;
1842 trace_xfs_link(tdp
, target_name
);
1844 ASSERT(!S_ISDIR(sip
->i_d
.di_mode
));
1846 if (XFS_FORCED_SHUTDOWN(mp
))
1847 return XFS_ERROR(EIO
);
1849 error
= xfs_qm_dqattach(sip
, 0);
1853 error
= xfs_qm_dqattach(tdp
, 0);
1857 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
1858 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1859 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
1860 error
= xfs_trans_reserve(tp
, resblks
, XFS_LINK_LOG_RES(mp
), 0,
1861 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
1862 if (error
== ENOSPC
) {
1864 error
= xfs_trans_reserve(tp
, 0, XFS_LINK_LOG_RES(mp
), 0,
1865 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
1872 xfs_lock_two_inodes(sip
, tdp
, XFS_ILOCK_EXCL
);
1874 xfs_trans_ijoin_ref(tp
, sip
, XFS_ILOCK_EXCL
);
1875 xfs_trans_ijoin_ref(tp
, tdp
, XFS_ILOCK_EXCL
);
1878 * If the source has too many links, we can't make any more to it.
1880 if (sip
->i_d
.di_nlink
>= XFS_MAXLINK
) {
1881 error
= XFS_ERROR(EMLINK
);
1886 * If we are using project inheritance, we only allow hard link
1887 * creation in our tree when the project IDs are the same; else
1888 * the tree quota mechanism could be circumvented.
1890 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
1891 (xfs_get_projid(tdp
) != xfs_get_projid(sip
)))) {
1892 error
= XFS_ERROR(EXDEV
);
1896 error
= xfs_dir_canenter(tp
, tdp
, target_name
, resblks
);
1900 xfs_bmap_init(&free_list
, &first_block
);
1902 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
1903 &first_block
, &free_list
, resblks
);
1906 xfs_trans_ichgtime(tp
, tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1907 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
1909 error
= xfs_bumplink(tp
, sip
);
1914 * If this is a synchronous mount, make sure that the
1915 * link transaction goes to disk before returning to
1918 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
1919 xfs_trans_set_sync(tp
);
1922 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
1924 xfs_bmap_cancel(&free_list
);
1928 return xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1931 cancel_flags
|= XFS_TRANS_ABORT
;
1933 xfs_trans_cancel(tp
, cancel_flags
);
1941 struct xfs_name
*link_name
,
1942 const char *target_path
,
1946 xfs_mount_t
*mp
= dp
->i_mount
;
1951 xfs_bmap_free_t free_list
;
1952 xfs_fsblock_t first_block
;
1953 boolean_t unlock_dp_on_error
= B_FALSE
;
1956 xfs_fileoff_t first_fsb
;
1957 xfs_filblks_t fs_blocks
;
1959 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
1961 const char *cur_chunk
;
1966 struct xfs_dquot
*udqp
, *gdqp
;
1974 trace_xfs_symlink(dp
, link_name
);
1976 if (XFS_FORCED_SHUTDOWN(mp
))
1977 return XFS_ERROR(EIO
);
1980 * Check component lengths of the target path name.
1982 pathlen
= strlen(target_path
);
1983 if (pathlen
>= MAXPATHLEN
) /* total string too long */
1984 return XFS_ERROR(ENAMETOOLONG
);
1987 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1988 prid
= xfs_get_projid(dp
);
1990 prid
= XFS_PROJID_DEFAULT
;
1993 * Make sure that we have allocated dquot(s) on disk.
1995 error
= xfs_qm_vop_dqalloc(dp
, current_fsuid(), current_fsgid(), prid
,
1996 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
2000 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SYMLINK
);
2001 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
2003 * The symlink will fit into the inode data fork?
2004 * There can't be any attributes so we get the whole variable part.
2006 if (pathlen
<= XFS_LITINO(mp
))
2009 fs_blocks
= XFS_B_TO_FSB(mp
, pathlen
);
2010 resblks
= XFS_SYMLINK_SPACE_RES(mp
, link_name
->len
, fs_blocks
);
2011 error
= xfs_trans_reserve(tp
, resblks
, XFS_SYMLINK_LOG_RES(mp
), 0,
2012 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
2013 if (error
== ENOSPC
&& fs_blocks
== 0) {
2015 error
= xfs_trans_reserve(tp
, 0, XFS_SYMLINK_LOG_RES(mp
), 0,
2016 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
2023 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
2024 unlock_dp_on_error
= B_TRUE
;
2027 * Check whether the directory allows new symlinks or not.
2029 if (dp
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) {
2030 error
= XFS_ERROR(EPERM
);
2035 * Reserve disk quota : blocks and inode.
2037 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
, resblks
, 1, 0);
2042 * Check for ability to enter directory entry, if no space reserved.
2044 error
= xfs_dir_canenter(tp
, dp
, link_name
, resblks
);
2048 * Initialize the bmap freelist prior to calling either
2049 * bmapi or the directory create code.
2051 xfs_bmap_init(&free_list
, &first_block
);
2054 * Allocate an inode for the symlink.
2056 error
= xfs_dir_ialloc(&tp
, dp
, S_IFLNK
| (mode
& ~S_IFMT
), 1, 0,
2057 prid
, resblks
> 0, &ip
, NULL
);
2059 if (error
== ENOSPC
)
2065 * An error after we've joined dp to the transaction will result in the
2066 * transaction cancel unlocking dp so don't do it explicitly in the
2069 xfs_trans_ijoin_ref(tp
, dp
, XFS_ILOCK_EXCL
);
2070 unlock_dp_on_error
= B_FALSE
;
2073 * Also attach the dquot(s) to it, if applicable.
2075 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
);
2078 resblks
-= XFS_IALLOC_SPACE_RES(mp
);
2080 * If the symlink will fit into the inode, write it inline.
2082 if (pathlen
<= XFS_IFORK_DSIZE(ip
)) {
2083 xfs_idata_realloc(ip
, pathlen
, XFS_DATA_FORK
);
2084 memcpy(ip
->i_df
.if_u1
.if_data
, target_path
, pathlen
);
2085 ip
->i_d
.di_size
= pathlen
;
2088 * The inode was initially created in extent format.
2090 ip
->i_df
.if_flags
&= ~(XFS_IFEXTENTS
| XFS_IFBROOT
);
2091 ip
->i_df
.if_flags
|= XFS_IFINLINE
;
2093 ip
->i_d
.di_format
= XFS_DINODE_FMT_LOCAL
;
2094 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_DDATA
| XFS_ILOG_CORE
);
2098 nmaps
= SYMLINK_MAPS
;
2100 error
= xfs_bmapi(tp
, ip
, first_fsb
, fs_blocks
,
2101 XFS_BMAPI_WRITE
| XFS_BMAPI_METADATA
,
2102 &first_block
, resblks
, mval
, &nmaps
,
2108 resblks
-= fs_blocks
;
2109 ip
->i_d
.di_size
= pathlen
;
2110 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2112 cur_chunk
= target_path
;
2113 for (n
= 0; n
< nmaps
; n
++) {
2114 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
2115 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
2116 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
2117 BTOBB(byte_cnt
), 0);
2118 ASSERT(bp
&& !XFS_BUF_GETERROR(bp
));
2119 if (pathlen
< byte_cnt
) {
2122 pathlen
-= byte_cnt
;
2124 memcpy(XFS_BUF_PTR(bp
), cur_chunk
, byte_cnt
);
2125 cur_chunk
+= byte_cnt
;
2127 xfs_trans_log_buf(tp
, bp
, 0, byte_cnt
- 1);
2132 * Create the directory entry for the symlink.
2134 error
= xfs_dir_createname(tp
, dp
, link_name
, ip
->i_ino
,
2135 &first_block
, &free_list
, resblks
);
2138 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2139 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
2142 * If this is a synchronous mount, make sure that the
2143 * symlink transaction goes to disk before returning to
2146 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
2147 xfs_trans_set_sync(tp
);
2150 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2154 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2155 xfs_qm_dqrele(udqp
);
2156 xfs_qm_dqrele(gdqp
);
2164 xfs_bmap_cancel(&free_list
);
2165 cancel_flags
|= XFS_TRANS_ABORT
;
2167 xfs_trans_cancel(tp
, cancel_flags
);
2168 xfs_qm_dqrele(udqp
);
2169 xfs_qm_dqrele(gdqp
);
2171 if (unlock_dp_on_error
)
2172 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
2183 xfs_mount_t
*mp
= ip
->i_mount
;
2187 if (!capable(CAP_SYS_ADMIN
))
2188 return XFS_ERROR(EPERM
);
2190 if (XFS_FORCED_SHUTDOWN(mp
))
2191 return XFS_ERROR(EIO
);
2193 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SET_DMATTRS
);
2194 error
= xfs_trans_reserve(tp
, 0, XFS_ICHANGE_LOG_RES (mp
), 0, 0, 0);
2196 xfs_trans_cancel(tp
, 0);
2199 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2200 xfs_trans_ijoin_ref(tp
, ip
, XFS_ILOCK_EXCL
);
2202 ip
->i_d
.di_dmevmask
= evmask
;
2203 ip
->i_d
.di_dmstate
= state
;
2205 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2206 error
= xfs_trans_commit(tp
, 0);
2212 * xfs_alloc_file_space()
2213 * This routine allocates disk space for the given file.
2215 * If alloc_type == 0, this request is for an ALLOCSP type
2216 * request which will change the file size. In this case, no
2217 * DMAPI event will be generated by the call. A TRUNCATE event
2218 * will be generated later by xfs_setattr.
2220 * If alloc_type != 0, this request is for a RESVSP type
2221 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
2222 * lower block boundary byte address is less than the file's
2231 xfs_alloc_file_space(
2238 xfs_mount_t
*mp
= ip
->i_mount
;
2240 xfs_filblks_t allocated_fsb
;
2241 xfs_filblks_t allocatesize_fsb
;
2242 xfs_extlen_t extsz
, temp
;
2243 xfs_fileoff_t startoffset_fsb
;
2244 xfs_fsblock_t firstfsb
;
2250 xfs_bmbt_irec_t imaps
[1], *imapp
;
2251 xfs_bmap_free_t free_list
;
2252 uint qblocks
, resblks
, resrtextents
;
2256 trace_xfs_alloc_file_space(ip
);
2258 if (XFS_FORCED_SHUTDOWN(mp
))
2259 return XFS_ERROR(EIO
);
2261 error
= xfs_qm_dqattach(ip
, 0);
2266 return XFS_ERROR(EINVAL
);
2268 rt
= XFS_IS_REALTIME_INODE(ip
);
2269 extsz
= xfs_get_extsz_hint(ip
);
2274 bmapi_flag
= XFS_BMAPI_WRITE
| alloc_type
;
2275 startoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
2276 allocatesize_fsb
= XFS_B_TO_FSB(mp
, count
);
2279 * Allocate file space until done or until there is an error
2281 while (allocatesize_fsb
&& !error
) {
2285 * Determine space reservations for data/realtime.
2287 if (unlikely(extsz
)) {
2288 s
= startoffset_fsb
;
2291 e
= startoffset_fsb
+ allocatesize_fsb
;
2292 if ((temp
= do_mod(startoffset_fsb
, extsz
)))
2294 if ((temp
= do_mod(e
, extsz
)))
2298 e
= allocatesize_fsb
;
2302 * The transaction reservation is limited to a 32-bit block
2303 * count, hence we need to limit the number of blocks we are
2304 * trying to reserve to avoid an overflow. We can't allocate
2305 * more than @nimaps extents, and an extent is limited on disk
2306 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
2308 resblks
= min_t(xfs_fileoff_t
, (e
- s
), (MAXEXTLEN
* nimaps
));
2310 resrtextents
= qblocks
= resblks
;
2311 resrtextents
/= mp
->m_sb
.sb_rextsize
;
2312 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
2313 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
2316 resblks
= qblocks
= XFS_DIOSTRAT_SPACE_RES(mp
, resblks
);
2317 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
2321 * Allocate and setup the transaction.
2323 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
2324 error
= xfs_trans_reserve(tp
, resblks
,
2325 XFS_WRITE_LOG_RES(mp
), resrtextents
,
2326 XFS_TRANS_PERM_LOG_RES
,
2327 XFS_WRITE_LOG_COUNT
);
2329 * Check for running out of space
2333 * Free the transaction structure.
2335 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
2336 xfs_trans_cancel(tp
, 0);
2339 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2340 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
,
2345 xfs_trans_ijoin(tp
, ip
);
2348 * Issue the xfs_bmapi() call to allocate the blocks
2350 xfs_bmap_init(&free_list
, &firstfsb
);
2351 error
= xfs_bmapi(tp
, ip
, startoffset_fsb
,
2352 allocatesize_fsb
, bmapi_flag
,
2353 &firstfsb
, 0, imapp
, &nimaps
,
2360 * Complete the transaction
2362 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2367 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2368 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2373 allocated_fsb
= imapp
->br_blockcount
;
2376 error
= XFS_ERROR(ENOSPC
);
2380 startoffset_fsb
+= allocated_fsb
;
2381 allocatesize_fsb
-= allocated_fsb
;
2386 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
2387 xfs_bmap_cancel(&free_list
);
2388 xfs_trans_unreserve_quota_nblks(tp
, ip
, qblocks
, 0, quota_flag
);
2390 error1
: /* Just cancel transaction */
2391 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
2392 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2397 * Zero file bytes between startoff and endoff inclusive.
2398 * The iolock is held exclusive and no blocks are buffered.
2400 * This function is used by xfs_free_file_space() to zero
2401 * partial blocks when the range to free is not block aligned.
2402 * When unreserving space with boundaries that are not block
2403 * aligned we round up the start and round down the end
2404 * boundaries and then use this function to zero the parts of
2405 * the blocks that got dropped during the rounding.
2408 xfs_zero_remaining_bytes(
2413 xfs_bmbt_irec_t imap
;
2414 xfs_fileoff_t offset_fsb
;
2415 xfs_off_t lastoffset
;
2418 xfs_mount_t
*mp
= ip
->i_mount
;
2423 * Avoid doing I/O beyond eof - it's not necessary
2424 * since nothing can read beyond eof. The space will
2425 * be zeroed when the file is extended anyway.
2427 if (startoff
>= ip
->i_size
)
2430 if (endoff
> ip
->i_size
)
2431 endoff
= ip
->i_size
;
2433 bp
= xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip
) ?
2434 mp
->m_rtdev_targp
: mp
->m_ddev_targp
,
2435 mp
->m_sb
.sb_blocksize
, XBF_DONT_BLOCK
);
2437 return XFS_ERROR(ENOMEM
);
2439 for (offset
= startoff
; offset
<= endoff
; offset
= lastoffset
+ 1) {
2440 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
2442 error
= xfs_bmapi(NULL
, ip
, offset_fsb
, 1, 0,
2443 NULL
, 0, &imap
, &nimap
, NULL
);
2444 if (error
|| nimap
< 1)
2446 ASSERT(imap
.br_blockcount
>= 1);
2447 ASSERT(imap
.br_startoff
== offset_fsb
);
2448 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+ 1) - 1;
2449 if (lastoffset
> endoff
)
2450 lastoffset
= endoff
;
2451 if (imap
.br_startblock
== HOLESTARTBLOCK
)
2453 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
2454 if (imap
.br_state
== XFS_EXT_UNWRITTEN
)
2457 XFS_BUF_UNWRITE(bp
);
2459 XFS_BUF_SET_ADDR(bp
, xfs_fsb_to_db(ip
, imap
.br_startblock
));
2461 error
= xfs_buf_iowait(bp
);
2463 xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
2464 mp
, bp
, XFS_BUF_ADDR(bp
));
2467 memset(XFS_BUF_PTR(bp
) +
2468 (offset
- XFS_FSB_TO_B(mp
, imap
.br_startoff
)),
2469 0, lastoffset
- offset
+ 1);
2474 error
= xfs_buf_iowait(bp
);
2476 xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
2477 mp
, bp
, XFS_BUF_ADDR(bp
));
2486 * xfs_free_file_space()
2487 * This routine frees disk space for the given file.
2489 * This routine is only called by xfs_change_file_space
2490 * for an UNRESVSP type call.
2498 xfs_free_file_space(
2506 xfs_fileoff_t endoffset_fsb
;
2508 xfs_fsblock_t firstfsb
;
2509 xfs_bmap_free_t free_list
;
2510 xfs_bmbt_irec_t imap
;
2518 xfs_fileoff_t startoffset_fsb
;
2520 int need_iolock
= 1;
2524 trace_xfs_free_file_space(ip
);
2526 error
= xfs_qm_dqattach(ip
, 0);
2531 if (len
<= 0) /* if nothing being freed */
2533 rt
= XFS_IS_REALTIME_INODE(ip
);
2534 startoffset_fsb
= XFS_B_TO_FSB(mp
, offset
);
2535 endoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
+ len
);
2537 if (attr_flags
& XFS_ATTR_NOLOCK
)
2540 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
2541 /* wait for the completion of any pending DIOs */
2545 rounding
= max_t(uint
, 1 << mp
->m_sb
.sb_blocklog
, PAGE_CACHE_SIZE
);
2546 ioffset
= offset
& ~(rounding
- 1);
2548 if (VN_CACHED(VFS_I(ip
)) != 0) {
2549 error
= xfs_flushinval_pages(ip
, ioffset
, -1, FI_REMAPF_LOCKED
);
2551 goto out_unlock_iolock
;
2555 * Need to zero the stuff we're not freeing, on disk.
2556 * If it's a realtime file & can't use unwritten extents then we
2557 * actually need to zero the extent edges. Otherwise xfs_bunmapi
2558 * will take care of it for us.
2560 if (rt
&& !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
2562 error
= xfs_bmapi(NULL
, ip
, startoffset_fsb
,
2563 1, 0, NULL
, 0, &imap
, &nimap
, NULL
);
2565 goto out_unlock_iolock
;
2566 ASSERT(nimap
== 0 || nimap
== 1);
2567 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
2570 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
2571 block
= imap
.br_startblock
;
2572 mod
= do_div(block
, mp
->m_sb
.sb_rextsize
);
2574 startoffset_fsb
+= mp
->m_sb
.sb_rextsize
- mod
;
2577 error
= xfs_bmapi(NULL
, ip
, endoffset_fsb
- 1,
2578 1, 0, NULL
, 0, &imap
, &nimap
, NULL
);
2580 goto out_unlock_iolock
;
2581 ASSERT(nimap
== 0 || nimap
== 1);
2582 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
2583 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
2585 if (mod
&& (mod
!= mp
->m_sb
.sb_rextsize
))
2586 endoffset_fsb
-= mod
;
2589 if ((done
= (endoffset_fsb
<= startoffset_fsb
)))
2591 * One contiguous piece to clear
2593 error
= xfs_zero_remaining_bytes(ip
, offset
, offset
+ len
- 1);
2596 * Some full blocks, possibly two pieces to clear
2598 if (offset
< XFS_FSB_TO_B(mp
, startoffset_fsb
))
2599 error
= xfs_zero_remaining_bytes(ip
, offset
,
2600 XFS_FSB_TO_B(mp
, startoffset_fsb
) - 1);
2602 XFS_FSB_TO_B(mp
, endoffset_fsb
) < offset
+ len
)
2603 error
= xfs_zero_remaining_bytes(ip
,
2604 XFS_FSB_TO_B(mp
, endoffset_fsb
),
2609 * free file space until done or until there is an error
2611 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
2612 while (!error
&& !done
) {
2615 * allocate and setup the transaction. Allow this
2616 * transaction to dip into the reserve blocks to ensure
2617 * the freeing of the space succeeds at ENOSPC.
2619 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
2620 tp
->t_flags
|= XFS_TRANS_RESERVE
;
2621 error
= xfs_trans_reserve(tp
,
2623 XFS_WRITE_LOG_RES(mp
),
2625 XFS_TRANS_PERM_LOG_RES
,
2626 XFS_WRITE_LOG_COUNT
);
2629 * check for running out of space
2633 * Free the transaction structure.
2635 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
2636 xfs_trans_cancel(tp
, 0);
2639 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2640 error
= xfs_trans_reserve_quota(tp
, mp
,
2641 ip
->i_udquot
, ip
->i_gdquot
,
2642 resblks
, 0, XFS_QMOPT_RES_REGBLKS
);
2646 xfs_trans_ijoin(tp
, ip
);
2649 * issue the bunmapi() call to free the blocks
2651 xfs_bmap_init(&free_list
, &firstfsb
);
2652 error
= xfs_bunmapi(tp
, ip
, startoffset_fsb
,
2653 endoffset_fsb
- startoffset_fsb
,
2654 0, 2, &firstfsb
, &free_list
, &done
);
2660 * complete the transaction
2662 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2667 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2668 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2673 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
2677 xfs_bmap_cancel(&free_list
);
2679 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
2680 xfs_iunlock(ip
, need_iolock
? (XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
) :
2686 * xfs_change_file_space()
2687 * This routine allocates or frees disk space for the given file.
2688 * The user specified parameters are checked for alignment and size
2697 xfs_change_file_space(
2704 xfs_mount_t
*mp
= ip
->i_mount
;
2709 xfs_off_t startoffset
;
2715 if (!S_ISREG(ip
->i_d
.di_mode
))
2716 return XFS_ERROR(EINVAL
);
2718 switch (bf
->l_whence
) {
2719 case 0: /*SEEK_SET*/
2721 case 1: /*SEEK_CUR*/
2722 bf
->l_start
+= offset
;
2724 case 2: /*SEEK_END*/
2725 bf
->l_start
+= ip
->i_size
;
2728 return XFS_ERROR(EINVAL
);
2731 llen
= bf
->l_len
> 0 ? bf
->l_len
- 1 : bf
->l_len
;
2733 if ( (bf
->l_start
< 0)
2734 || (bf
->l_start
> XFS_MAXIOFFSET(mp
))
2735 || (bf
->l_start
+ llen
< 0)
2736 || (bf
->l_start
+ llen
> XFS_MAXIOFFSET(mp
)))
2737 return XFS_ERROR(EINVAL
);
2741 startoffset
= bf
->l_start
;
2745 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
2747 * These calls do NOT zero the data space allocated to the file,
2748 * nor do they change the file size.
2750 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
2752 * These calls cause the new file data to be zeroed and the file
2753 * size to be changed.
2755 setprealloc
= clrprealloc
= 0;
2756 prealloc_type
= XFS_BMAPI_PREALLOC
;
2759 case XFS_IOC_ZERO_RANGE
:
2760 prealloc_type
|= XFS_BMAPI_CONVERT
;
2761 xfs_tosspages(ip
, startoffset
, startoffset
+ bf
->l_len
, 0);
2763 case XFS_IOC_RESVSP
:
2764 case XFS_IOC_RESVSP64
:
2765 error
= xfs_alloc_file_space(ip
, startoffset
, bf
->l_len
,
2766 prealloc_type
, attr_flags
);
2772 case XFS_IOC_UNRESVSP
:
2773 case XFS_IOC_UNRESVSP64
:
2774 if ((error
= xfs_free_file_space(ip
, startoffset
, bf
->l_len
,
2779 case XFS_IOC_ALLOCSP
:
2780 case XFS_IOC_ALLOCSP64
:
2781 case XFS_IOC_FREESP
:
2782 case XFS_IOC_FREESP64
:
2783 if (startoffset
> fsize
) {
2784 error
= xfs_alloc_file_space(ip
, fsize
,
2785 startoffset
- fsize
, 0, attr_flags
);
2790 iattr
.ia_valid
= ATTR_SIZE
;
2791 iattr
.ia_size
= startoffset
;
2793 error
= xfs_setattr(ip
, &iattr
, attr_flags
);
2803 return XFS_ERROR(EINVAL
);
2807 * update the inode timestamp, mode, and prealloc flag bits
2809 tp
= xfs_trans_alloc(mp
, XFS_TRANS_WRITEID
);
2811 if ((error
= xfs_trans_reserve(tp
, 0, XFS_WRITEID_LOG_RES(mp
),
2814 xfs_trans_cancel(tp
, 0);
2818 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2820 xfs_trans_ijoin(tp
, ip
);
2822 if ((attr_flags
& XFS_ATTR_DMI
) == 0) {
2823 ip
->i_d
.di_mode
&= ~S_ISUID
;
2826 * Note that we don't have to worry about mandatory
2827 * file locking being disabled here because we only
2828 * clear the S_ISGID bit if the Group execute bit is
2829 * on, but if it was on then mandatory locking wouldn't
2830 * have been enabled.
2832 if (ip
->i_d
.di_mode
& S_IXGRP
)
2833 ip
->i_d
.di_mode
&= ~S_ISGID
;
2835 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2838 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
2839 else if (clrprealloc
)
2840 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
2842 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2843 if (attr_flags
& XFS_ATTR_SYNC
)
2844 xfs_trans_set_sync(tp
);
2846 error
= xfs_trans_commit(tp
, 0);
2848 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);