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"
24 #include "xfs_trans.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_itable.h"
36 #include "xfs_ialloc.h"
37 #include "xfs_alloc.h"
41 #include "xfs_error.h"
42 #include "xfs_quota.h"
43 #include "xfs_utils.h"
44 #include "xfs_rtalloc.h"
45 #include "xfs_trans_space.h"
46 #include "xfs_log_priv.h"
47 #include "xfs_filestream.h"
48 #include "xfs_vnodeops.h"
49 #include "xfs_trace.h"
50 #include "xfs_icache.h"
53 * The maximum pathlen is 1024 bytes. Since the minimum file system
54 * blocksize is 512 bytes, we can get a max of 2 extents back from
57 #define SYMLINK_MAPS 2
64 xfs_mount_t
*mp
= ip
->i_mount
;
65 int pathlen
= ip
->i_d
.di_size
;
66 int nmaps
= SYMLINK_MAPS
;
67 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
74 error
= xfs_bmapi_read(ip
, 0, XFS_B_TO_FSB(mp
, pathlen
), mval
, &nmaps
,
79 for (n
= 0; n
< nmaps
; n
++) {
80 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
81 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
83 bp
= xfs_buf_read(mp
->m_ddev_targp
, d
, BTOBB(byte_cnt
), 0);
85 return XFS_ERROR(ENOMEM
);
88 xfs_buf_ioerror_alert(bp
, __func__
);
92 if (pathlen
< byte_cnt
)
96 memcpy(link
, bp
->b_addr
, byte_cnt
);
100 link
[ip
->i_d
.di_size
] = '\0';
112 xfs_mount_t
*mp
= ip
->i_mount
;
116 trace_xfs_readlink(ip
);
118 if (XFS_FORCED_SHUTDOWN(mp
))
119 return XFS_ERROR(EIO
);
121 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
123 pathlen
= ip
->i_d
.di_size
;
127 if (pathlen
< 0 || pathlen
> MAXPATHLEN
) {
128 xfs_alert(mp
, "%s: inode (%llu) bad symlink length (%lld)",
129 __func__
, (unsigned long long) ip
->i_ino
,
130 (long long) pathlen
);
132 error
= XFS_ERROR(EFSCORRUPTED
);
137 if (ip
->i_df
.if_flags
& XFS_IFINLINE
) {
138 memcpy(link
, ip
->i_df
.if_u1
.if_data
, pathlen
);
139 link
[pathlen
] = '\0';
141 error
= xfs_readlink_bmap(ip
, link
);
145 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
150 * This is called by xfs_inactive to free any blocks beyond eof
151 * when the link count isn't zero and by xfs_dm_punch_hole() when
152 * punching a hole to EOF.
162 xfs_fileoff_t end_fsb
;
163 xfs_fileoff_t last_fsb
;
164 xfs_filblks_t map_len
;
166 xfs_bmbt_irec_t imap
;
169 * Figure out if there are any blocks beyond the end
170 * of the file. If not, then there is nothing to do.
172 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_ISIZE(ip
));
173 last_fsb
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
174 if (last_fsb
<= end_fsb
)
176 map_len
= last_fsb
- end_fsb
;
179 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
180 error
= xfs_bmapi_read(ip
, end_fsb
, map_len
, &imap
, &nimaps
, 0);
181 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
183 if (!error
&& (nimaps
!= 0) &&
184 (imap
.br_startblock
!= HOLESTARTBLOCK
||
185 ip
->i_delayed_blks
)) {
187 * Attach the dquots to the inode up front.
189 error
= xfs_qm_dqattach(ip
, 0);
194 * There are blocks after the end of file.
195 * Free them up now by truncating the file to
198 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
201 if (!xfs_ilock_nowait(ip
, XFS_IOLOCK_EXCL
)) {
202 xfs_trans_cancel(tp
, 0);
207 error
= xfs_trans_reserve(tp
, 0,
208 XFS_ITRUNCATE_LOG_RES(mp
),
209 0, XFS_TRANS_PERM_LOG_RES
,
210 XFS_ITRUNCATE_LOG_COUNT
);
212 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
213 xfs_trans_cancel(tp
, 0);
215 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
219 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
220 xfs_trans_ijoin(tp
, ip
, 0);
223 * Do not update the on-disk file size. If we update the
224 * on-disk file size and then the system crashes before the
225 * contents of the file are flushed to disk then the files
226 * may be full of holes (ie NULL files bug).
228 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
,
232 * If we get an error at this point we simply don't
233 * bother truncating the file.
236 (XFS_TRANS_RELEASE_LOG_RES
|
239 error
= xfs_trans_commit(tp
,
240 XFS_TRANS_RELEASE_LOG_RES
);
242 xfs_inode_clear_eofblocks_tag(ip
);
245 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
247 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
253 * Free a symlink that has blocks associated with it.
256 xfs_inactive_symlink_rmt(
264 xfs_fsblock_t first_block
;
265 xfs_bmap_free_t free_list
;
268 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
276 ASSERT(ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
));
278 * We're freeing a symlink that has some
279 * blocks allocated to it. Free the
280 * blocks here. We know that we've got
281 * either 1 or 2 extents and that we can
282 * free them all in one bunmapi call.
284 ASSERT(ip
->i_d
.di_nextents
> 0 && ip
->i_d
.di_nextents
<= 2);
287 * Lock the inode, fix the size, and join it to the transaction.
288 * Hold it so in the normal path, we still have it locked for
289 * the second transaction. In the error paths we need it
290 * held so the cancel won't rele it, see below.
292 size
= (int)ip
->i_d
.di_size
;
294 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
296 * Find the block(s) so we can inval and unmap them.
299 xfs_bmap_init(&free_list
, &first_block
);
300 nmaps
= ARRAY_SIZE(mval
);
301 error
= xfs_bmapi_read(ip
, 0, XFS_B_TO_FSB(mp
, size
),
306 * Invalidate the block(s).
308 for (i
= 0; i
< nmaps
; i
++) {
309 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
,
310 XFS_FSB_TO_DADDR(mp
, mval
[i
].br_startblock
),
311 XFS_FSB_TO_BB(mp
, mval
[i
].br_blockcount
), 0);
316 xfs_trans_binval(tp
, bp
);
319 * Unmap the dead block(s) to the free_list.
321 if ((error
= xfs_bunmapi(tp
, ip
, 0, size
, XFS_BMAPI_METADATA
, nmaps
,
322 &first_block
, &free_list
, &done
)))
326 * Commit the first transaction. This logs the EFI and the inode.
328 if ((error
= xfs_bmap_finish(&tp
, &free_list
, &committed
)))
331 * The transaction must have been committed, since there were
332 * actually extents freed by xfs_bunmapi. See xfs_bmap_finish.
333 * The new tp has the extent freeing and EFDs.
337 * The first xact was committed, so add the inode to the new one.
338 * Mark it dirty so it will be logged and moved forward in the log as
339 * part of every commit.
341 xfs_trans_ijoin(tp
, ip
, 0);
342 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
344 * Get a new, empty transaction to return to our caller.
346 ntp
= xfs_trans_dup(tp
);
348 * Commit the transaction containing extent freeing and EFDs.
349 * If we get an error on the commit here or on the reserve below,
350 * we need to unlock the inode since the new transaction doesn't
351 * have the inode attached.
353 error
= xfs_trans_commit(tp
, 0);
356 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
360 * transaction commit worked ok so we can drop the extra ticket
361 * reference that we gained in xfs_trans_dup()
363 xfs_log_ticket_put(tp
->t_ticket
);
366 * Remove the memory for extent descriptions (just bookkeeping).
368 if (ip
->i_df
.if_bytes
)
369 xfs_idata_realloc(ip
, -ip
->i_df
.if_bytes
, XFS_DATA_FORK
);
370 ASSERT(ip
->i_df
.if_bytes
== 0);
372 * Put an itruncate log reservation in the new transaction
375 if ((error
= xfs_trans_reserve(tp
, 0, XFS_ITRUNCATE_LOG_RES(mp
), 0,
376 XFS_TRANS_PERM_LOG_RES
, XFS_ITRUNCATE_LOG_COUNT
))) {
377 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
381 xfs_trans_ijoin(tp
, ip
, 0);
386 xfs_bmap_cancel(&free_list
);
395 xfs_mount_t
*mp
= ip
->i_mount
;
398 if (!S_ISREG(ip
->i_d
.di_mode
) || (ip
->i_d
.di_mode
== 0))
401 /* If this is a read-only mount, don't do this (would generate I/O) */
402 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
405 if (!XFS_FORCED_SHUTDOWN(mp
)) {
409 * If we are using filestreams, and we have an unlinked
410 * file that we are processing the last close on, then nothing
411 * will be able to reopen and write to this file. Purge this
412 * inode from the filestreams cache so that it doesn't delay
413 * teardown of the inode.
415 if ((ip
->i_d
.di_nlink
== 0) && xfs_inode_is_filestream(ip
))
416 xfs_filestream_deassociate(ip
);
419 * If we previously truncated this file and removed old data
420 * in the process, we want to initiate "early" writeout on
421 * the last close. This is an attempt to combat the notorious
422 * NULL files problem which is particularly noticeable from a
423 * truncate down, buffered (re-)write (delalloc), followed by
424 * a crash. What we are effectively doing here is
425 * significantly reducing the time window where we'd otherwise
426 * be exposed to that problem.
428 truncated
= xfs_iflags_test_and_clear(ip
, XFS_ITRUNCATED
);
430 xfs_iflags_clear(ip
, XFS_IDIRTY_RELEASE
);
431 if (VN_DIRTY(VFS_I(ip
)) && ip
->i_delayed_blks
> 0)
432 xfs_flush_pages(ip
, 0, -1, XBF_ASYNC
, FI_NONE
);
436 if (ip
->i_d
.di_nlink
== 0)
439 if ((S_ISREG(ip
->i_d
.di_mode
) &&
440 (VFS_I(ip
)->i_size
> 0 ||
441 (VN_CACHED(VFS_I(ip
)) > 0 || ip
->i_delayed_blks
> 0)) &&
442 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
)) &&
443 (!(ip
->i_d
.di_flags
& (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)))) {
446 * If we can't get the iolock just skip truncating the blocks
447 * past EOF because we could deadlock with the mmap_sem
448 * otherwise. We'll get another chance to drop them once the
449 * last reference to the inode is dropped, so we'll never leak
450 * blocks permanently.
452 * Further, check if the inode is being opened, written and
453 * closed frequently and we have delayed allocation blocks
454 * outstanding (e.g. streaming writes from the NFS server),
455 * truncating the blocks past EOF will cause fragmentation to
458 * In this case don't do the truncation, either, but we have to
459 * be careful how we detect this case. Blocks beyond EOF show
460 * up as i_delayed_blks even when the inode is clean, so we
461 * need to truncate them away first before checking for a dirty
462 * release. Hence on the first dirty close we will still remove
463 * the speculative allocation, but after that we will leave it
466 if (xfs_iflags_test(ip
, XFS_IDIRTY_RELEASE
))
469 error
= xfs_free_eofblocks(mp
, ip
, true);
473 /* delalloc blocks after truncation means it really is dirty */
474 if (ip
->i_delayed_blks
)
475 xfs_iflags_set(ip
, XFS_IDIRTY_RELEASE
);
483 * This is called when the vnode reference count for the vnode
484 * goes to zero. If the file has been unlinked, then it must
485 * now be truncated. Also, we clear all of the read-ahead state
486 * kept for the inode here since the file is now closed.
492 xfs_bmap_free_t free_list
;
493 xfs_fsblock_t first_block
;
501 * If the inode is already free, then there can be nothing
504 if (ip
->i_d
.di_mode
== 0 || is_bad_inode(VFS_I(ip
))) {
505 ASSERT(ip
->i_df
.if_real_bytes
== 0);
506 ASSERT(ip
->i_df
.if_broot_bytes
== 0);
507 return VN_INACTIVE_CACHE
;
514 /* If this is a read-only mount, don't do this (would generate I/O) */
515 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
518 if (ip
->i_d
.di_nlink
!= 0) {
519 if ((S_ISREG(ip
->i_d
.di_mode
) &&
520 (VFS_I(ip
)->i_size
> 0 ||
521 (VN_CACHED(VFS_I(ip
)) > 0 || ip
->i_delayed_blks
> 0)) &&
522 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
) &&
523 (!(ip
->i_d
.di_flags
&
524 (XFS_DIFLAG_PREALLOC
| XFS_DIFLAG_APPEND
)) ||
525 ip
->i_delayed_blks
!= 0))) {
526 error
= xfs_free_eofblocks(mp
, ip
, false);
528 return VN_INACTIVE_CACHE
;
533 if (S_ISREG(ip
->i_d
.di_mode
) &&
534 (ip
->i_d
.di_size
!= 0 || XFS_ISIZE(ip
) != 0 ||
535 ip
->i_d
.di_nextents
> 0 || ip
->i_delayed_blks
> 0))
538 error
= xfs_qm_dqattach(ip
, 0);
540 return VN_INACTIVE_CACHE
;
542 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
543 error
= xfs_trans_reserve(tp
, 0,
544 (truncate
|| S_ISLNK(ip
->i_d
.di_mode
)) ?
545 XFS_ITRUNCATE_LOG_RES(mp
) :
546 XFS_IFREE_LOG_RES(mp
),
548 XFS_TRANS_PERM_LOG_RES
,
549 XFS_ITRUNCATE_LOG_COUNT
);
551 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
552 xfs_trans_cancel(tp
, 0);
553 return VN_INACTIVE_CACHE
;
556 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
557 xfs_trans_ijoin(tp
, ip
, 0);
559 if (S_ISLNK(ip
->i_d
.di_mode
)) {
561 * Zero length symlinks _can_ exist.
563 if (ip
->i_d
.di_size
> XFS_IFORK_DSIZE(ip
)) {
564 error
= xfs_inactive_symlink_rmt(ip
, &tp
);
567 } else if (ip
->i_df
.if_bytes
> 0) {
568 xfs_idata_realloc(ip
, -(ip
->i_df
.if_bytes
),
570 ASSERT(ip
->i_df
.if_bytes
== 0);
572 } else if (truncate
) {
574 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
576 error
= xfs_itruncate_extents(&tp
, ip
, XFS_DATA_FORK
, 0);
580 ASSERT(ip
->i_d
.di_nextents
== 0);
584 * If there are attributes associated with the file then blow them away
585 * now. The code calls a routine that recursively deconstructs the
586 * attribute fork. We need to just commit the current transaction
587 * because we can't use it for xfs_attr_inactive().
589 if (ip
->i_d
.di_anextents
> 0) {
590 ASSERT(ip
->i_d
.di_forkoff
!= 0);
592 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
596 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
598 error
= xfs_attr_inactive(ip
);
602 tp
= xfs_trans_alloc(mp
, XFS_TRANS_INACTIVE
);
603 error
= xfs_trans_reserve(tp
, 0,
604 XFS_IFREE_LOG_RES(mp
),
605 0, XFS_TRANS_PERM_LOG_RES
,
606 XFS_INACTIVE_LOG_COUNT
);
608 xfs_trans_cancel(tp
, 0);
612 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
613 xfs_trans_ijoin(tp
, ip
, 0);
617 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
619 ASSERT(ip
->i_d
.di_anextents
== 0);
624 xfs_bmap_init(&free_list
, &first_block
);
625 error
= xfs_ifree(tp
, ip
, &free_list
);
628 * If we fail to free the inode, shut down. The cancel
629 * might do that, we need to make sure. Otherwise the
630 * inode might be lost for a long time or forever.
632 if (!XFS_FORCED_SHUTDOWN(mp
)) {
633 xfs_notice(mp
, "%s: xfs_ifree returned error %d",
635 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
637 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_ABORT
);
640 * Credit the quota account(s). The inode is gone.
642 xfs_trans_mod_dquot_byino(tp
, ip
, XFS_TRANS_DQ_ICOUNT
, -1);
645 * Just ignore errors at this point. There is nothing we can
646 * do except to try to keep going. Make sure it's not a silent
649 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
651 xfs_notice(mp
, "%s: xfs_bmap_finish returned error %d",
653 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
655 xfs_notice(mp
, "%s: xfs_trans_commit returned error %d",
660 * Release the dquots held by inode, if any.
664 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
666 return VN_INACTIVE_CACHE
;
668 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
673 * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
674 * is allowed, otherwise it has to be an exact match. If a CI match is found,
675 * ci_name->name will point to a the actual name (caller must free) or
676 * will be set to NULL if an exact match is found.
681 struct xfs_name
*name
,
683 struct xfs_name
*ci_name
)
689 trace_xfs_lookup(dp
, name
);
691 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
692 return XFS_ERROR(EIO
);
694 lock_mode
= xfs_ilock_map_shared(dp
);
695 error
= xfs_dir_lookup(NULL
, dp
, name
, &inum
, ci_name
);
696 xfs_iunlock_map_shared(dp
, lock_mode
);
701 error
= xfs_iget(dp
->i_mount
, NULL
, inum
, 0, 0, ipp
);
709 kmem_free(ci_name
->name
);
718 struct xfs_name
*name
,
723 int is_dir
= S_ISDIR(mode
);
724 struct xfs_mount
*mp
= dp
->i_mount
;
725 struct xfs_inode
*ip
= NULL
;
726 struct xfs_trans
*tp
= NULL
;
728 xfs_bmap_free_t free_list
;
729 xfs_fsblock_t first_block
;
730 boolean_t unlock_dp_on_error
= B_FALSE
;
734 struct xfs_dquot
*udqp
= NULL
;
735 struct xfs_dquot
*gdqp
= NULL
;
740 trace_xfs_create(dp
, name
);
742 if (XFS_FORCED_SHUTDOWN(mp
))
743 return XFS_ERROR(EIO
);
745 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
746 prid
= xfs_get_projid(dp
);
748 prid
= XFS_PROJID_DEFAULT
;
751 * Make sure that we have allocated dquot(s) on disk.
753 error
= xfs_qm_vop_dqalloc(dp
, current_fsuid(), current_fsgid(), prid
,
754 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
760 resblks
= XFS_MKDIR_SPACE_RES(mp
, name
->len
);
761 log_res
= XFS_MKDIR_LOG_RES(mp
);
762 log_count
= XFS_MKDIR_LOG_COUNT
;
763 tp
= xfs_trans_alloc(mp
, XFS_TRANS_MKDIR
);
765 resblks
= XFS_CREATE_SPACE_RES(mp
, name
->len
);
766 log_res
= XFS_CREATE_LOG_RES(mp
);
767 log_count
= XFS_CREATE_LOG_COUNT
;
768 tp
= xfs_trans_alloc(mp
, XFS_TRANS_CREATE
);
771 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
774 * Initially assume that the file does not exist and
775 * reserve the resources for that case. If that is not
776 * the case we'll drop the one we have and get a more
777 * appropriate transaction later.
779 error
= xfs_trans_reserve(tp
, resblks
, log_res
, 0,
780 XFS_TRANS_PERM_LOG_RES
, log_count
);
781 if (error
== ENOSPC
) {
782 /* flush outstanding delalloc blocks and retry */
783 xfs_flush_inodes(mp
);
784 error
= xfs_trans_reserve(tp
, resblks
, log_res
, 0,
785 XFS_TRANS_PERM_LOG_RES
, log_count
);
787 if (error
== ENOSPC
) {
788 /* No space at all so try a "no-allocation" reservation */
790 error
= xfs_trans_reserve(tp
, 0, log_res
, 0,
791 XFS_TRANS_PERM_LOG_RES
, log_count
);
795 goto out_trans_cancel
;
798 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
799 unlock_dp_on_error
= B_TRUE
;
801 xfs_bmap_init(&free_list
, &first_block
);
804 * Reserve disk quota and the inode.
806 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
, resblks
, 1, 0);
808 goto out_trans_cancel
;
810 error
= xfs_dir_canenter(tp
, dp
, name
, resblks
);
812 goto out_trans_cancel
;
815 * A newly created regular or special file just has one directory
816 * entry pointing to them, but a directory also the "." entry
817 * pointing to itself.
819 error
= xfs_dir_ialloc(&tp
, dp
, mode
, is_dir
? 2 : 1, rdev
,
820 prid
, resblks
> 0, &ip
, &committed
);
823 goto out_trans_cancel
;
824 goto out_trans_abort
;
828 * Now we join the directory inode to the transaction. We do not do it
829 * earlier because xfs_dir_ialloc might commit the previous transaction
830 * (and release all the locks). An error from here on will result in
831 * the transaction cancel unlocking dp so don't do it explicitly in the
834 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
835 unlock_dp_on_error
= B_FALSE
;
837 error
= xfs_dir_createname(tp
, dp
, name
, ip
->i_ino
,
838 &first_block
, &free_list
, resblks
?
839 resblks
- XFS_IALLOC_SPACE_RES(mp
) : 0);
841 ASSERT(error
!= ENOSPC
);
842 goto out_trans_abort
;
844 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
845 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
848 error
= xfs_dir_init(tp
, ip
, dp
);
850 goto out_bmap_cancel
;
852 error
= xfs_bumplink(tp
, dp
);
854 goto out_bmap_cancel
;
858 * If this is a synchronous mount, make sure that the
859 * create transaction goes to disk before returning to
862 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
863 xfs_trans_set_sync(tp
);
866 * Attach the dquot(s) to the inodes and modify them incore.
867 * These ids of the inode couldn't have changed since the new
868 * inode has been locked ever since it was created.
870 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
);
872 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
874 goto out_bmap_cancel
;
876 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
878 goto out_release_inode
;
887 xfs_bmap_cancel(&free_list
);
889 cancel_flags
|= XFS_TRANS_ABORT
;
891 xfs_trans_cancel(tp
, cancel_flags
);
894 * Wait until after the current transaction is aborted to
895 * release the inode. This prevents recursive transactions
896 * and deadlocks from xfs_inactive.
904 if (unlock_dp_on_error
)
905 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
911 int xfs_small_retries
;
912 int xfs_middle_retries
;
913 int xfs_lots_retries
;
918 * Bump the subclass so xfs_lock_inodes() acquires each lock with
922 xfs_lock_inumorder(int lock_mode
, int subclass
)
924 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
925 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_IOLOCK_SHIFT
;
926 if (lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
))
927 lock_mode
|= (subclass
+ XFS_LOCK_INUMORDER
) << XFS_ILOCK_SHIFT
;
933 * The following routine will lock n inodes in exclusive mode.
934 * We assume the caller calls us with the inodes in i_ino order.
936 * We need to detect deadlock where an inode that we lock
937 * is in the AIL and we start waiting for another inode that is locked
938 * by a thread in a long running transaction (such as truncate). This can
939 * result in deadlock since the long running trans might need to wait
940 * for the inode we just locked in order to push the tail and free space
949 int attempts
= 0, i
, j
, try_lock
;
952 ASSERT(ips
&& (inodes
>= 2)); /* we need at least two */
958 for (; i
< inodes
; i
++) {
961 if (i
&& (ips
[i
] == ips
[i
-1])) /* Already locked */
965 * If try_lock is not set yet, make sure all locked inodes
966 * are not in the AIL.
967 * If any are, set try_lock to be used later.
971 for (j
= (i
- 1); j
>= 0 && !try_lock
; j
--) {
972 lp
= (xfs_log_item_t
*)ips
[j
]->i_itemp
;
973 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
980 * If any of the previous locks we have locked is in the AIL,
981 * we must TRY to get the second and subsequent locks. If
982 * we can't get any, we must release all we have
987 /* try_lock must be 0 if i is 0. */
989 * try_lock means we have an inode locked
990 * that is in the AIL.
993 if (!xfs_ilock_nowait(ips
[i
], xfs_lock_inumorder(lock_mode
, i
))) {
997 * Unlock all previous guys and try again.
998 * xfs_iunlock will try to push the tail
999 * if the inode is in the AIL.
1002 for(j
= i
- 1; j
>= 0; j
--) {
1005 * Check to see if we've already
1006 * unlocked this one.
1007 * Not the first one going back,
1008 * and the inode ptr is the same.
1010 if ((j
!= (i
- 1)) && ips
[j
] ==
1014 xfs_iunlock(ips
[j
], lock_mode
);
1017 if ((attempts
% 5) == 0) {
1018 delay(1); /* Don't just spin the CPU */
1028 xfs_ilock(ips
[i
], xfs_lock_inumorder(lock_mode
, i
));
1034 if (attempts
< 5) xfs_small_retries
++;
1035 else if (attempts
< 100) xfs_middle_retries
++;
1036 else xfs_lots_retries
++;
1044 * xfs_lock_two_inodes() can only be used to lock one type of lock
1045 * at a time - the iolock or the ilock, but not both at once. If
1046 * we lock both at once, lockdep will report false positives saying
1047 * we have violated locking orders.
1050 xfs_lock_two_inodes(
1059 if (lock_mode
& (XFS_IOLOCK_SHARED
|XFS_IOLOCK_EXCL
))
1060 ASSERT((lock_mode
& (XFS_ILOCK_SHARED
|XFS_ILOCK_EXCL
)) == 0);
1061 ASSERT(ip0
->i_ino
!= ip1
->i_ino
);
1063 if (ip0
->i_ino
> ip1
->i_ino
) {
1070 xfs_ilock(ip0
, xfs_lock_inumorder(lock_mode
, 0));
1073 * If the first lock we have locked is in the AIL, we must TRY to get
1074 * the second lock. If we can't get it, we must release the first one
1077 lp
= (xfs_log_item_t
*)ip0
->i_itemp
;
1078 if (lp
&& (lp
->li_flags
& XFS_LI_IN_AIL
)) {
1079 if (!xfs_ilock_nowait(ip1
, xfs_lock_inumorder(lock_mode
, 1))) {
1080 xfs_iunlock(ip0
, lock_mode
);
1081 if ((++attempts
% 5) == 0)
1082 delay(1); /* Don't just spin the CPU */
1086 xfs_ilock(ip1
, xfs_lock_inumorder(lock_mode
, 1));
1093 struct xfs_name
*name
,
1096 xfs_mount_t
*mp
= dp
->i_mount
;
1097 xfs_trans_t
*tp
= NULL
;
1098 int is_dir
= S_ISDIR(ip
->i_d
.di_mode
);
1100 xfs_bmap_free_t free_list
;
1101 xfs_fsblock_t first_block
;
1108 trace_xfs_remove(dp
, name
);
1110 if (XFS_FORCED_SHUTDOWN(mp
))
1111 return XFS_ERROR(EIO
);
1113 error
= xfs_qm_dqattach(dp
, 0);
1117 error
= xfs_qm_dqattach(ip
, 0);
1122 tp
= xfs_trans_alloc(mp
, XFS_TRANS_RMDIR
);
1123 log_count
= XFS_DEFAULT_LOG_COUNT
;
1125 tp
= xfs_trans_alloc(mp
, XFS_TRANS_REMOVE
);
1126 log_count
= XFS_REMOVE_LOG_COUNT
;
1128 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1131 * We try to get the real space reservation first,
1132 * allowing for directory btree deletion(s) implying
1133 * possible bmap insert(s). If we can't get the space
1134 * reservation then we use 0 instead, and avoid the bmap
1135 * btree insert(s) in the directory code by, if the bmap
1136 * insert tries to happen, instead trimming the LAST
1137 * block from the directory.
1139 resblks
= XFS_REMOVE_SPACE_RES(mp
);
1140 error
= xfs_trans_reserve(tp
, resblks
, XFS_REMOVE_LOG_RES(mp
), 0,
1141 XFS_TRANS_PERM_LOG_RES
, log_count
);
1142 if (error
== ENOSPC
) {
1144 error
= xfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
1145 XFS_TRANS_PERM_LOG_RES
, log_count
);
1148 ASSERT(error
!= ENOSPC
);
1150 goto out_trans_cancel
;
1153 xfs_lock_two_inodes(dp
, ip
, XFS_ILOCK_EXCL
);
1155 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1156 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1159 * If we're removing a directory perform some additional validation.
1162 ASSERT(ip
->i_d
.di_nlink
>= 2);
1163 if (ip
->i_d
.di_nlink
!= 2) {
1164 error
= XFS_ERROR(ENOTEMPTY
);
1165 goto out_trans_cancel
;
1167 if (!xfs_dir_isempty(ip
)) {
1168 error
= XFS_ERROR(ENOTEMPTY
);
1169 goto out_trans_cancel
;
1173 xfs_bmap_init(&free_list
, &first_block
);
1174 error
= xfs_dir_removename(tp
, dp
, name
, ip
->i_ino
,
1175 &first_block
, &free_list
, resblks
);
1177 ASSERT(error
!= ENOENT
);
1178 goto out_bmap_cancel
;
1180 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1184 * Drop the link from ip's "..".
1186 error
= xfs_droplink(tp
, dp
);
1188 goto out_bmap_cancel
;
1191 * Drop the "." link from ip to self.
1193 error
= xfs_droplink(tp
, ip
);
1195 goto out_bmap_cancel
;
1198 * When removing a non-directory we need to log the parent
1199 * inode here. For a directory this is done implicitly
1200 * by the xfs_droplink call for the ".." entry.
1202 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1206 * Drop the link from dp to ip.
1208 error
= xfs_droplink(tp
, ip
);
1210 goto out_bmap_cancel
;
1213 * Determine if this is the last link while
1214 * we are in the transaction.
1216 link_zero
= (ip
->i_d
.di_nlink
== 0);
1219 * If this is a synchronous mount, make sure that the
1220 * remove transaction goes to disk before returning to
1223 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
))
1224 xfs_trans_set_sync(tp
);
1226 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1228 goto out_bmap_cancel
;
1230 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1235 * If we are using filestreams, kill the stream association.
1236 * If the file is still open it may get a new one but that
1237 * will get killed on last close in xfs_close() so we don't
1238 * have to worry about that.
1240 if (!is_dir
&& link_zero
&& xfs_inode_is_filestream(ip
))
1241 xfs_filestream_deassociate(ip
);
1246 xfs_bmap_cancel(&free_list
);
1247 cancel_flags
|= XFS_TRANS_ABORT
;
1249 xfs_trans_cancel(tp
, cancel_flags
);
1258 struct xfs_name
*target_name
)
1260 xfs_mount_t
*mp
= tdp
->i_mount
;
1263 xfs_bmap_free_t free_list
;
1264 xfs_fsblock_t first_block
;
1269 trace_xfs_link(tdp
, target_name
);
1271 ASSERT(!S_ISDIR(sip
->i_d
.di_mode
));
1273 if (XFS_FORCED_SHUTDOWN(mp
))
1274 return XFS_ERROR(EIO
);
1276 error
= xfs_qm_dqattach(sip
, 0);
1280 error
= xfs_qm_dqattach(tdp
, 0);
1284 tp
= xfs_trans_alloc(mp
, XFS_TRANS_LINK
);
1285 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1286 resblks
= XFS_LINK_SPACE_RES(mp
, target_name
->len
);
1287 error
= xfs_trans_reserve(tp
, resblks
, XFS_LINK_LOG_RES(mp
), 0,
1288 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
1289 if (error
== ENOSPC
) {
1291 error
= xfs_trans_reserve(tp
, 0, XFS_LINK_LOG_RES(mp
), 0,
1292 XFS_TRANS_PERM_LOG_RES
, XFS_LINK_LOG_COUNT
);
1299 xfs_lock_two_inodes(sip
, tdp
, XFS_ILOCK_EXCL
);
1301 xfs_trans_ijoin(tp
, sip
, XFS_ILOCK_EXCL
);
1302 xfs_trans_ijoin(tp
, tdp
, XFS_ILOCK_EXCL
);
1305 * If we are using project inheritance, we only allow hard link
1306 * creation in our tree when the project IDs are the same; else
1307 * the tree quota mechanism could be circumvented.
1309 if (unlikely((tdp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
) &&
1310 (xfs_get_projid(tdp
) != xfs_get_projid(sip
)))) {
1311 error
= XFS_ERROR(EXDEV
);
1315 error
= xfs_dir_canenter(tp
, tdp
, target_name
, resblks
);
1319 xfs_bmap_init(&free_list
, &first_block
);
1321 error
= xfs_dir_createname(tp
, tdp
, target_name
, sip
->i_ino
,
1322 &first_block
, &free_list
, resblks
);
1325 xfs_trans_ichgtime(tp
, tdp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1326 xfs_trans_log_inode(tp
, tdp
, XFS_ILOG_CORE
);
1328 error
= xfs_bumplink(tp
, sip
);
1333 * If this is a synchronous mount, make sure that the
1334 * link transaction goes to disk before returning to
1337 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
1338 xfs_trans_set_sync(tp
);
1341 error
= xfs_bmap_finish (&tp
, &free_list
, &committed
);
1343 xfs_bmap_cancel(&free_list
);
1347 return xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1350 cancel_flags
|= XFS_TRANS_ABORT
;
1352 xfs_trans_cancel(tp
, cancel_flags
);
1360 struct xfs_name
*link_name
,
1361 const char *target_path
,
1365 xfs_mount_t
*mp
= dp
->i_mount
;
1370 xfs_bmap_free_t free_list
;
1371 xfs_fsblock_t first_block
;
1372 boolean_t unlock_dp_on_error
= B_FALSE
;
1375 xfs_fileoff_t first_fsb
;
1376 xfs_filblks_t fs_blocks
;
1378 xfs_bmbt_irec_t mval
[SYMLINK_MAPS
];
1380 const char *cur_chunk
;
1385 struct xfs_dquot
*udqp
, *gdqp
;
1393 trace_xfs_symlink(dp
, link_name
);
1395 if (XFS_FORCED_SHUTDOWN(mp
))
1396 return XFS_ERROR(EIO
);
1399 * Check component lengths of the target path name.
1401 pathlen
= strlen(target_path
);
1402 if (pathlen
>= MAXPATHLEN
) /* total string too long */
1403 return XFS_ERROR(ENAMETOOLONG
);
1406 if (dp
->i_d
.di_flags
& XFS_DIFLAG_PROJINHERIT
)
1407 prid
= xfs_get_projid(dp
);
1409 prid
= XFS_PROJID_DEFAULT
;
1412 * Make sure that we have allocated dquot(s) on disk.
1414 error
= xfs_qm_vop_dqalloc(dp
, current_fsuid(), current_fsgid(), prid
,
1415 XFS_QMOPT_QUOTALL
| XFS_QMOPT_INHERIT
, &udqp
, &gdqp
);
1419 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SYMLINK
);
1420 cancel_flags
= XFS_TRANS_RELEASE_LOG_RES
;
1422 * The symlink will fit into the inode data fork?
1423 * There can't be any attributes so we get the whole variable part.
1425 if (pathlen
<= XFS_LITINO(mp
))
1428 fs_blocks
= XFS_B_TO_FSB(mp
, pathlen
);
1429 resblks
= XFS_SYMLINK_SPACE_RES(mp
, link_name
->len
, fs_blocks
);
1430 error
= xfs_trans_reserve(tp
, resblks
, XFS_SYMLINK_LOG_RES(mp
), 0,
1431 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
1432 if (error
== ENOSPC
&& fs_blocks
== 0) {
1434 error
= xfs_trans_reserve(tp
, 0, XFS_SYMLINK_LOG_RES(mp
), 0,
1435 XFS_TRANS_PERM_LOG_RES
, XFS_SYMLINK_LOG_COUNT
);
1442 xfs_ilock(dp
, XFS_ILOCK_EXCL
| XFS_ILOCK_PARENT
);
1443 unlock_dp_on_error
= B_TRUE
;
1446 * Check whether the directory allows new symlinks or not.
1448 if (dp
->i_d
.di_flags
& XFS_DIFLAG_NOSYMLINKS
) {
1449 error
= XFS_ERROR(EPERM
);
1454 * Reserve disk quota : blocks and inode.
1456 error
= xfs_trans_reserve_quota(tp
, mp
, udqp
, gdqp
, resblks
, 1, 0);
1461 * Check for ability to enter directory entry, if no space reserved.
1463 error
= xfs_dir_canenter(tp
, dp
, link_name
, resblks
);
1467 * Initialize the bmap freelist prior to calling either
1468 * bmapi or the directory create code.
1470 xfs_bmap_init(&free_list
, &first_block
);
1473 * Allocate an inode for the symlink.
1475 error
= xfs_dir_ialloc(&tp
, dp
, S_IFLNK
| (mode
& ~S_IFMT
), 1, 0,
1476 prid
, resblks
> 0, &ip
, NULL
);
1478 if (error
== ENOSPC
)
1484 * An error after we've joined dp to the transaction will result in the
1485 * transaction cancel unlocking dp so don't do it explicitly in the
1488 xfs_trans_ijoin(tp
, dp
, XFS_ILOCK_EXCL
);
1489 unlock_dp_on_error
= B_FALSE
;
1492 * Also attach the dquot(s) to it, if applicable.
1494 xfs_qm_vop_create_dqattach(tp
, ip
, udqp
, gdqp
);
1497 resblks
-= XFS_IALLOC_SPACE_RES(mp
);
1499 * If the symlink will fit into the inode, write it inline.
1501 if (pathlen
<= XFS_IFORK_DSIZE(ip
)) {
1502 xfs_idata_realloc(ip
, pathlen
, XFS_DATA_FORK
);
1503 memcpy(ip
->i_df
.if_u1
.if_data
, target_path
, pathlen
);
1504 ip
->i_d
.di_size
= pathlen
;
1507 * The inode was initially created in extent format.
1509 ip
->i_df
.if_flags
&= ~(XFS_IFEXTENTS
| XFS_IFBROOT
);
1510 ip
->i_df
.if_flags
|= XFS_IFINLINE
;
1512 ip
->i_d
.di_format
= XFS_DINODE_FMT_LOCAL
;
1513 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_DDATA
| XFS_ILOG_CORE
);
1517 nmaps
= SYMLINK_MAPS
;
1519 error
= xfs_bmapi_write(tp
, ip
, first_fsb
, fs_blocks
,
1520 XFS_BMAPI_METADATA
, &first_block
, resblks
,
1521 mval
, &nmaps
, &free_list
);
1526 resblks
-= fs_blocks
;
1527 ip
->i_d
.di_size
= pathlen
;
1528 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1530 cur_chunk
= target_path
;
1531 for (n
= 0; n
< nmaps
; n
++) {
1532 d
= XFS_FSB_TO_DADDR(mp
, mval
[n
].br_startblock
);
1533 byte_cnt
= XFS_FSB_TO_B(mp
, mval
[n
].br_blockcount
);
1534 bp
= xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
,
1535 BTOBB(byte_cnt
), 0);
1540 if (pathlen
< byte_cnt
) {
1543 pathlen
-= byte_cnt
;
1545 memcpy(bp
->b_addr
, cur_chunk
, byte_cnt
);
1546 cur_chunk
+= byte_cnt
;
1548 xfs_trans_log_buf(tp
, bp
, 0, byte_cnt
- 1);
1553 * Create the directory entry for the symlink.
1555 error
= xfs_dir_createname(tp
, dp
, link_name
, ip
->i_ino
,
1556 &first_block
, &free_list
, resblks
);
1559 xfs_trans_ichgtime(tp
, dp
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
1560 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1563 * If this is a synchronous mount, make sure that the
1564 * symlink transaction goes to disk before returning to
1567 if (mp
->m_flags
& (XFS_MOUNT_WSYNC
|XFS_MOUNT_DIRSYNC
)) {
1568 xfs_trans_set_sync(tp
);
1571 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1575 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1576 xfs_qm_dqrele(udqp
);
1577 xfs_qm_dqrele(gdqp
);
1585 xfs_bmap_cancel(&free_list
);
1586 cancel_flags
|= XFS_TRANS_ABORT
;
1588 xfs_trans_cancel(tp
, cancel_flags
);
1589 xfs_qm_dqrele(udqp
);
1590 xfs_qm_dqrele(gdqp
);
1592 if (unlock_dp_on_error
)
1593 xfs_iunlock(dp
, XFS_ILOCK_EXCL
);
1604 xfs_mount_t
*mp
= ip
->i_mount
;
1608 if (!capable(CAP_SYS_ADMIN
))
1609 return XFS_ERROR(EPERM
);
1611 if (XFS_FORCED_SHUTDOWN(mp
))
1612 return XFS_ERROR(EIO
);
1614 tp
= xfs_trans_alloc(mp
, XFS_TRANS_SET_DMATTRS
);
1615 error
= xfs_trans_reserve(tp
, 0, XFS_ICHANGE_LOG_RES (mp
), 0, 0, 0);
1617 xfs_trans_cancel(tp
, 0);
1620 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1621 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
1623 ip
->i_d
.di_dmevmask
= evmask
;
1624 ip
->i_d
.di_dmstate
= state
;
1626 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
1627 error
= xfs_trans_commit(tp
, 0);
1633 * xfs_alloc_file_space()
1634 * This routine allocates disk space for the given file.
1636 * If alloc_type == 0, this request is for an ALLOCSP type
1637 * request which will change the file size. In this case, no
1638 * DMAPI event will be generated by the call. A TRUNCATE event
1639 * will be generated later by xfs_setattr.
1641 * If alloc_type != 0, this request is for a RESVSP type
1642 * request, and a DMAPI DM_EVENT_WRITE will be generated if the
1643 * lower block boundary byte address is less than the file's
1652 xfs_alloc_file_space(
1659 xfs_mount_t
*mp
= ip
->i_mount
;
1661 xfs_filblks_t allocated_fsb
;
1662 xfs_filblks_t allocatesize_fsb
;
1663 xfs_extlen_t extsz
, temp
;
1664 xfs_fileoff_t startoffset_fsb
;
1665 xfs_fsblock_t firstfsb
;
1670 xfs_bmbt_irec_t imaps
[1], *imapp
;
1671 xfs_bmap_free_t free_list
;
1672 uint qblocks
, resblks
, resrtextents
;
1676 trace_xfs_alloc_file_space(ip
);
1678 if (XFS_FORCED_SHUTDOWN(mp
))
1679 return XFS_ERROR(EIO
);
1681 error
= xfs_qm_dqattach(ip
, 0);
1686 return XFS_ERROR(EINVAL
);
1688 rt
= XFS_IS_REALTIME_INODE(ip
);
1689 extsz
= xfs_get_extsz_hint(ip
);
1694 startoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1695 allocatesize_fsb
= XFS_B_TO_FSB(mp
, count
);
1698 * Allocate file space until done or until there is an error
1700 while (allocatesize_fsb
&& !error
) {
1704 * Determine space reservations for data/realtime.
1706 if (unlikely(extsz
)) {
1707 s
= startoffset_fsb
;
1710 e
= startoffset_fsb
+ allocatesize_fsb
;
1711 if ((temp
= do_mod(startoffset_fsb
, extsz
)))
1713 if ((temp
= do_mod(e
, extsz
)))
1717 e
= allocatesize_fsb
;
1721 * The transaction reservation is limited to a 32-bit block
1722 * count, hence we need to limit the number of blocks we are
1723 * trying to reserve to avoid an overflow. We can't allocate
1724 * more than @nimaps extents, and an extent is limited on disk
1725 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1727 resblks
= min_t(xfs_fileoff_t
, (e
- s
), (MAXEXTLEN
* nimaps
));
1729 resrtextents
= qblocks
= resblks
;
1730 resrtextents
/= mp
->m_sb
.sb_rextsize
;
1731 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
1732 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
1735 resblks
= qblocks
= XFS_DIOSTRAT_SPACE_RES(mp
, resblks
);
1736 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
1740 * Allocate and setup the transaction.
1742 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
1743 error
= xfs_trans_reserve(tp
, resblks
,
1744 XFS_WRITE_LOG_RES(mp
), resrtextents
,
1745 XFS_TRANS_PERM_LOG_RES
,
1746 XFS_WRITE_LOG_COUNT
);
1748 * Check for running out of space
1752 * Free the transaction structure.
1754 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
1755 xfs_trans_cancel(tp
, 0);
1758 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1759 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
,
1764 xfs_trans_ijoin(tp
, ip
, 0);
1766 xfs_bmap_init(&free_list
, &firstfsb
);
1767 error
= xfs_bmapi_write(tp
, ip
, startoffset_fsb
,
1768 allocatesize_fsb
, alloc_type
, &firstfsb
,
1769 0, imapp
, &nimaps
, &free_list
);
1775 * Complete the transaction
1777 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
1782 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
1783 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1788 allocated_fsb
= imapp
->br_blockcount
;
1791 error
= XFS_ERROR(ENOSPC
);
1795 startoffset_fsb
+= allocated_fsb
;
1796 allocatesize_fsb
-= allocated_fsb
;
1801 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1802 xfs_bmap_cancel(&free_list
);
1803 xfs_trans_unreserve_quota_nblks(tp
, ip
, (long)qblocks
, 0, quota_flag
);
1805 error1
: /* Just cancel transaction */
1806 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
1807 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1812 * Zero file bytes between startoff and endoff inclusive.
1813 * The iolock is held exclusive and no blocks are buffered.
1815 * This function is used by xfs_free_file_space() to zero
1816 * partial blocks when the range to free is not block aligned.
1817 * When unreserving space with boundaries that are not block
1818 * aligned we round up the start and round down the end
1819 * boundaries and then use this function to zero the parts of
1820 * the blocks that got dropped during the rounding.
1823 xfs_zero_remaining_bytes(
1828 xfs_bmbt_irec_t imap
;
1829 xfs_fileoff_t offset_fsb
;
1830 xfs_off_t lastoffset
;
1833 xfs_mount_t
*mp
= ip
->i_mount
;
1838 * Avoid doing I/O beyond eof - it's not necessary
1839 * since nothing can read beyond eof. The space will
1840 * be zeroed when the file is extended anyway.
1842 if (startoff
>= XFS_ISIZE(ip
))
1845 if (endoff
> XFS_ISIZE(ip
))
1846 endoff
= XFS_ISIZE(ip
);
1848 bp
= xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip
) ?
1849 mp
->m_rtdev_targp
: mp
->m_ddev_targp
,
1850 BTOBB(mp
->m_sb
.sb_blocksize
), 0);
1852 return XFS_ERROR(ENOMEM
);
1856 for (offset
= startoff
; offset
<= endoff
; offset
= lastoffset
+ 1) {
1857 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1859 error
= xfs_bmapi_read(ip
, offset_fsb
, 1, &imap
, &nimap
, 0);
1860 if (error
|| nimap
< 1)
1862 ASSERT(imap
.br_blockcount
>= 1);
1863 ASSERT(imap
.br_startoff
== offset_fsb
);
1864 lastoffset
= XFS_FSB_TO_B(mp
, imap
.br_startoff
+ 1) - 1;
1865 if (lastoffset
> endoff
)
1866 lastoffset
= endoff
;
1867 if (imap
.br_startblock
== HOLESTARTBLOCK
)
1869 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
1870 if (imap
.br_state
== XFS_EXT_UNWRITTEN
)
1873 XFS_BUF_UNWRITE(bp
);
1875 XFS_BUF_SET_ADDR(bp
, xfs_fsb_to_db(ip
, imap
.br_startblock
));
1877 error
= xfs_buf_iowait(bp
);
1879 xfs_buf_ioerror_alert(bp
,
1880 "xfs_zero_remaining_bytes(read)");
1884 (offset
- XFS_FSB_TO_B(mp
, imap
.br_startoff
)),
1885 0, lastoffset
- offset
+ 1);
1890 error
= xfs_buf_iowait(bp
);
1892 xfs_buf_ioerror_alert(bp
,
1893 "xfs_zero_remaining_bytes(write)");
1902 * xfs_free_file_space()
1903 * This routine frees disk space for the given file.
1905 * This routine is only called by xfs_change_file_space
1906 * for an UNRESVSP type call.
1914 xfs_free_file_space(
1922 xfs_fileoff_t endoffset_fsb
;
1924 xfs_fsblock_t firstfsb
;
1925 xfs_bmap_free_t free_list
;
1926 xfs_bmbt_irec_t imap
;
1934 xfs_fileoff_t startoffset_fsb
;
1936 int need_iolock
= 1;
1940 trace_xfs_free_file_space(ip
);
1942 error
= xfs_qm_dqattach(ip
, 0);
1947 if (len
<= 0) /* if nothing being freed */
1949 rt
= XFS_IS_REALTIME_INODE(ip
);
1950 startoffset_fsb
= XFS_B_TO_FSB(mp
, offset
);
1951 endoffset_fsb
= XFS_B_TO_FSBT(mp
, offset
+ len
);
1953 if (attr_flags
& XFS_ATTR_NOLOCK
)
1956 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
1957 /* wait for the completion of any pending DIOs */
1958 inode_dio_wait(VFS_I(ip
));
1961 rounding
= max_t(uint
, 1 << mp
->m_sb
.sb_blocklog
, PAGE_CACHE_SIZE
);
1962 ioffset
= offset
& ~(rounding
- 1);
1964 if (VN_CACHED(VFS_I(ip
)) != 0) {
1965 error
= xfs_flushinval_pages(ip
, ioffset
, -1, FI_REMAPF_LOCKED
);
1967 goto out_unlock_iolock
;
1971 * Need to zero the stuff we're not freeing, on disk.
1972 * If it's a realtime file & can't use unwritten extents then we
1973 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1974 * will take care of it for us.
1976 if (rt
&& !xfs_sb_version_hasextflgbit(&mp
->m_sb
)) {
1978 error
= xfs_bmapi_read(ip
, startoffset_fsb
, 1,
1981 goto out_unlock_iolock
;
1982 ASSERT(nimap
== 0 || nimap
== 1);
1983 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
1986 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
1987 block
= imap
.br_startblock
;
1988 mod
= do_div(block
, mp
->m_sb
.sb_rextsize
);
1990 startoffset_fsb
+= mp
->m_sb
.sb_rextsize
- mod
;
1993 error
= xfs_bmapi_read(ip
, endoffset_fsb
- 1, 1,
1996 goto out_unlock_iolock
;
1997 ASSERT(nimap
== 0 || nimap
== 1);
1998 if (nimap
&& imap
.br_startblock
!= HOLESTARTBLOCK
) {
1999 ASSERT(imap
.br_startblock
!= DELAYSTARTBLOCK
);
2001 if (mod
&& (mod
!= mp
->m_sb
.sb_rextsize
))
2002 endoffset_fsb
-= mod
;
2005 if ((done
= (endoffset_fsb
<= startoffset_fsb
)))
2007 * One contiguous piece to clear
2009 error
= xfs_zero_remaining_bytes(ip
, offset
, offset
+ len
- 1);
2012 * Some full blocks, possibly two pieces to clear
2014 if (offset
< XFS_FSB_TO_B(mp
, startoffset_fsb
))
2015 error
= xfs_zero_remaining_bytes(ip
, offset
,
2016 XFS_FSB_TO_B(mp
, startoffset_fsb
) - 1);
2018 XFS_FSB_TO_B(mp
, endoffset_fsb
) < offset
+ len
)
2019 error
= xfs_zero_remaining_bytes(ip
,
2020 XFS_FSB_TO_B(mp
, endoffset_fsb
),
2025 * free file space until done or until there is an error
2027 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
2028 while (!error
&& !done
) {
2031 * allocate and setup the transaction. Allow this
2032 * transaction to dip into the reserve blocks to ensure
2033 * the freeing of the space succeeds at ENOSPC.
2035 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
2036 tp
->t_flags
|= XFS_TRANS_RESERVE
;
2037 error
= xfs_trans_reserve(tp
,
2039 XFS_WRITE_LOG_RES(mp
),
2041 XFS_TRANS_PERM_LOG_RES
,
2042 XFS_WRITE_LOG_COUNT
);
2045 * check for running out of space
2049 * Free the transaction structure.
2051 ASSERT(error
== ENOSPC
|| XFS_FORCED_SHUTDOWN(mp
));
2052 xfs_trans_cancel(tp
, 0);
2055 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2056 error
= xfs_trans_reserve_quota(tp
, mp
,
2057 ip
->i_udquot
, ip
->i_gdquot
,
2058 resblks
, 0, XFS_QMOPT_RES_REGBLKS
);
2062 xfs_trans_ijoin(tp
, ip
, 0);
2065 * issue the bunmapi() call to free the blocks
2067 xfs_bmap_init(&free_list
, &firstfsb
);
2068 error
= xfs_bunmapi(tp
, ip
, startoffset_fsb
,
2069 endoffset_fsb
- startoffset_fsb
,
2070 0, 2, &firstfsb
, &free_list
, &done
);
2076 * complete the transaction
2078 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
2083 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
2084 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
2089 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
2093 xfs_bmap_cancel(&free_list
);
2095 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
2096 xfs_iunlock(ip
, need_iolock
? (XFS_ILOCK_EXCL
| XFS_IOLOCK_EXCL
) :
2102 * xfs_change_file_space()
2103 * This routine allocates or frees disk space for the given file.
2104 * The user specified parameters are checked for alignment and size
2113 xfs_change_file_space(
2120 xfs_mount_t
*mp
= ip
->i_mount
;
2125 xfs_off_t startoffset
;
2131 if (!S_ISREG(ip
->i_d
.di_mode
))
2132 return XFS_ERROR(EINVAL
);
2134 switch (bf
->l_whence
) {
2135 case 0: /*SEEK_SET*/
2137 case 1: /*SEEK_CUR*/
2138 bf
->l_start
+= offset
;
2140 case 2: /*SEEK_END*/
2141 bf
->l_start
+= XFS_ISIZE(ip
);
2144 return XFS_ERROR(EINVAL
);
2147 llen
= bf
->l_len
> 0 ? bf
->l_len
- 1 : bf
->l_len
;
2149 if (bf
->l_start
< 0 ||
2150 bf
->l_start
> mp
->m_super
->s_maxbytes
||
2151 bf
->l_start
+ llen
< 0 ||
2152 bf
->l_start
+ llen
> mp
->m_super
->s_maxbytes
)
2153 return XFS_ERROR(EINVAL
);
2157 startoffset
= bf
->l_start
;
2158 fsize
= XFS_ISIZE(ip
);
2161 * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
2163 * These calls do NOT zero the data space allocated to the file,
2164 * nor do they change the file size.
2166 * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
2168 * These calls cause the new file data to be zeroed and the file
2169 * size to be changed.
2171 setprealloc
= clrprealloc
= 0;
2172 prealloc_type
= XFS_BMAPI_PREALLOC
;
2175 case XFS_IOC_ZERO_RANGE
:
2176 prealloc_type
|= XFS_BMAPI_CONVERT
;
2177 xfs_tosspages(ip
, startoffset
, startoffset
+ bf
->l_len
, 0);
2179 case XFS_IOC_RESVSP
:
2180 case XFS_IOC_RESVSP64
:
2181 error
= xfs_alloc_file_space(ip
, startoffset
, bf
->l_len
,
2182 prealloc_type
, attr_flags
);
2188 case XFS_IOC_UNRESVSP
:
2189 case XFS_IOC_UNRESVSP64
:
2190 if ((error
= xfs_free_file_space(ip
, startoffset
, bf
->l_len
,
2195 case XFS_IOC_ALLOCSP
:
2196 case XFS_IOC_ALLOCSP64
:
2197 case XFS_IOC_FREESP
:
2198 case XFS_IOC_FREESP64
:
2200 * These operations actually do IO when extending the file, but
2201 * the allocation is done seperately to the zeroing that is
2202 * done. This set of operations need to be serialised against
2203 * other IO operations, such as truncate and buffered IO. We
2204 * need to take the IOLOCK here to serialise the allocation and
2205 * zeroing IO to prevent other IOLOCK holders (e.g. getbmap,
2206 * truncate, direct IO) from racing against the transient
2207 * allocated but not written state we can have here.
2209 xfs_ilock(ip
, XFS_IOLOCK_EXCL
);
2210 if (startoffset
> fsize
) {
2211 error
= xfs_alloc_file_space(ip
, fsize
,
2212 startoffset
- fsize
, 0,
2213 attr_flags
| XFS_ATTR_NOLOCK
);
2215 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
2220 iattr
.ia_valid
= ATTR_SIZE
;
2221 iattr
.ia_size
= startoffset
;
2223 error
= xfs_setattr_size(ip
, &iattr
,
2224 attr_flags
| XFS_ATTR_NOLOCK
);
2225 xfs_iunlock(ip
, XFS_IOLOCK_EXCL
);
2235 return XFS_ERROR(EINVAL
);
2239 * update the inode timestamp, mode, and prealloc flag bits
2241 tp
= xfs_trans_alloc(mp
, XFS_TRANS_WRITEID
);
2243 if ((error
= xfs_trans_reserve(tp
, 0, XFS_WRITEID_LOG_RES(mp
),
2246 xfs_trans_cancel(tp
, 0);
2250 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
2251 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
2253 if ((attr_flags
& XFS_ATTR_DMI
) == 0) {
2254 ip
->i_d
.di_mode
&= ~S_ISUID
;
2257 * Note that we don't have to worry about mandatory
2258 * file locking being disabled here because we only
2259 * clear the S_ISGID bit if the Group execute bit is
2260 * on, but if it was on then mandatory locking wouldn't
2261 * have been enabled.
2263 if (ip
->i_d
.di_mode
& S_IXGRP
)
2264 ip
->i_d
.di_mode
&= ~S_ISGID
;
2266 xfs_trans_ichgtime(tp
, ip
, XFS_ICHGTIME_MOD
| XFS_ICHGTIME_CHG
);
2269 ip
->i_d
.di_flags
|= XFS_DIFLAG_PREALLOC
;
2270 else if (clrprealloc
)
2271 ip
->i_d
.di_flags
&= ~XFS_DIFLAG_PREALLOC
;
2273 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2274 if (attr_flags
& XFS_ATTR_SYNC
)
2275 xfs_trans_set_sync(tp
);
2276 return xfs_trans_commit(tp
, 0);