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_trans.h"
24 #include "xfs_alloc.h"
25 #include "xfs_quota.h"
26 #include "xfs_mount.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_alloc_btree.h"
29 #include "xfs_ialloc_btree.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
32 #include "xfs_inode_item.h"
33 #include "xfs_btree.h"
35 #include "xfs_rtalloc.h"
36 #include "xfs_error.h"
37 #include "xfs_itable.h"
39 #include "xfs_buf_item.h"
40 #include "xfs_trans_space.h"
41 #include "xfs_utils.h"
42 #include "xfs_iomap.h"
43 #include "xfs_trace.h"
46 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
48 #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
51 xfs_iomap_eof_align_last_fsb(
55 xfs_fileoff_t
*last_fsb
)
57 xfs_fileoff_t new_last_fsb
= 0;
58 xfs_extlen_t align
= 0;
61 if (!XFS_IS_REALTIME_INODE(ip
)) {
63 * Round up the allocation request to a stripe unit
64 * (m_dalign) boundary if the file size is >= stripe unit
65 * size, and we are allocating past the allocation eof.
67 * If mounted with the "-o swalloc" option the alignment is
68 * increased from the strip unit size to the stripe width.
70 if (mp
->m_swidth
&& (mp
->m_flags
& XFS_MOUNT_SWALLOC
))
72 else if (mp
->m_dalign
)
75 if (align
&& XFS_ISIZE(ip
) >= XFS_FSB_TO_B(mp
, align
))
76 new_last_fsb
= roundup_64(*last_fsb
, align
);
80 * Always round up the allocation request to an extent boundary
81 * (when file on a real-time subvolume or has di_extsize hint).
85 align
= roundup_64(new_last_fsb
, extsize
);
88 new_last_fsb
= roundup_64(*last_fsb
, align
);
92 error
= xfs_bmap_eof(ip
, new_last_fsb
, XFS_DATA_FORK
, &eof
);
96 *last_fsb
= new_last_fsb
;
102 xfs_alert_fsblock_zero(
104 xfs_bmbt_irec_t
*imap
)
106 xfs_alert_tag(ip
->i_mount
, XFS_PTAG_FSBLOCK_ZERO
,
107 "Access to block zero in inode %llu "
108 "start_block: %llx start_off: %llx "
109 "blkcnt: %llx extent-state: %x\n",
110 (unsigned long long)ip
->i_ino
,
111 (unsigned long long)imap
->br_startblock
,
112 (unsigned long long)imap
->br_startoff
,
113 (unsigned long long)imap
->br_blockcount
,
119 xfs_iomap_write_direct(
123 xfs_bmbt_irec_t
*imap
,
126 xfs_mount_t
*mp
= ip
->i_mount
;
127 xfs_fileoff_t offset_fsb
;
128 xfs_fileoff_t last_fsb
;
129 xfs_filblks_t count_fsb
, resaligned
;
130 xfs_fsblock_t firstfsb
;
131 xfs_extlen_t extsz
, temp
;
137 xfs_bmap_free_t free_list
;
138 uint qblocks
, resblks
, resrtextents
;
142 error
= xfs_qm_dqattach(ip
, 0);
144 return XFS_ERROR(error
);
146 rt
= XFS_IS_REALTIME_INODE(ip
);
147 extsz
= xfs_get_extsz_hint(ip
);
149 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
150 last_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)(offset
+ count
)));
151 if ((offset
+ count
) > XFS_ISIZE(ip
)) {
152 error
= xfs_iomap_eof_align_last_fsb(mp
, ip
, extsz
, &last_fsb
);
154 return XFS_ERROR(error
);
156 if (nmaps
&& (imap
->br_startblock
== HOLESTARTBLOCK
))
157 last_fsb
= MIN(last_fsb
, (xfs_fileoff_t
)
158 imap
->br_blockcount
+
161 count_fsb
= last_fsb
- offset_fsb
;
162 ASSERT(count_fsb
> 0);
164 resaligned
= count_fsb
;
165 if (unlikely(extsz
)) {
166 if ((temp
= do_mod(offset_fsb
, extsz
)))
168 if ((temp
= do_mod(resaligned
, extsz
)))
169 resaligned
+= extsz
- temp
;
173 resrtextents
= qblocks
= resaligned
;
174 resrtextents
/= mp
->m_sb
.sb_rextsize
;
175 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
176 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
179 resblks
= qblocks
= XFS_DIOSTRAT_SPACE_RES(mp
, resaligned
);
180 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
184 * Allocate and setup the transaction
186 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
187 error
= xfs_trans_reserve(tp
, resblks
,
188 XFS_WRITE_LOG_RES(mp
), resrtextents
,
189 XFS_TRANS_PERM_LOG_RES
,
190 XFS_WRITE_LOG_COUNT
);
192 * Check for running out of space, note: need lock to return
195 xfs_trans_cancel(tp
, 0);
196 return XFS_ERROR(error
);
199 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
201 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
, 0, quota_flag
);
203 goto out_trans_cancel
;
205 xfs_trans_ijoin(tp
, ip
, 0);
208 if (offset
< XFS_ISIZE(ip
) || extsz
)
209 bmapi_flag
|= XFS_BMAPI_PREALLOC
;
212 * From this point onwards we overwrite the imap pointer that the
215 xfs_bmap_init(&free_list
, &firstfsb
);
217 error
= xfs_bmapi_write(tp
, ip
, offset_fsb
, count_fsb
, bmapi_flag
,
218 &firstfsb
, 0, imap
, &nimaps
, &free_list
);
220 goto out_bmap_cancel
;
223 * Complete the transaction
225 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
227 goto out_bmap_cancel
;
228 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
233 * Copy any maps to caller's array and return any error.
236 error
= XFS_ERROR(ENOSPC
);
240 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
241 error
= xfs_alert_fsblock_zero(ip
, imap
);
244 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
248 xfs_bmap_cancel(&free_list
);
249 xfs_trans_unreserve_quota_nblks(tp
, ip
, (long)qblocks
, 0, quota_flag
);
251 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
256 * If the caller is doing a write at the end of the file, then extend the
257 * allocation out to the file system's write iosize. We clean up any extra
258 * space left over when the file is closed in xfs_inactive().
260 * If we find we already have delalloc preallocation beyond EOF, don't do more
261 * preallocation as it it not needed.
264 xfs_iomap_eof_want_preallocate(
269 xfs_bmbt_irec_t
*imap
,
273 xfs_fileoff_t start_fsb
;
274 xfs_filblks_t count_fsb
;
275 xfs_fsblock_t firstblock
;
277 int found_delalloc
= 0;
280 if (offset
+ count
<= XFS_ISIZE(ip
))
284 * If there are any real blocks past eof, then don't
285 * do any speculative allocation.
287 start_fsb
= XFS_B_TO_FSBT(mp
, ((xfs_ufsize_t
)(offset
+ count
- 1)));
288 count_fsb
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
289 while (count_fsb
> 0) {
291 firstblock
= NULLFSBLOCK
;
292 error
= xfs_bmapi_read(ip
, start_fsb
, count_fsb
, imap
, &imaps
,
296 for (n
= 0; n
< imaps
; n
++) {
297 if ((imap
[n
].br_startblock
!= HOLESTARTBLOCK
) &&
298 (imap
[n
].br_startblock
!= DELAYSTARTBLOCK
))
300 start_fsb
+= imap
[n
].br_blockcount
;
301 count_fsb
-= imap
[n
].br_blockcount
;
303 if (imap
[n
].br_startblock
== DELAYSTARTBLOCK
)
313 * If we don't have a user specified preallocation size, dynamically increase
314 * the preallocation size as the size of the file grows. Cap the maximum size
315 * at a single extent or less if the filesystem is near full. The closer the
316 * filesystem is to full, the smaller the maximum prealocation.
319 xfs_iomap_prealloc_size(
320 struct xfs_mount
*mp
,
321 struct xfs_inode
*ip
)
323 xfs_fsblock_t alloc_blocks
= 0;
325 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
330 * rounddown_pow_of_two() returns an undefined result
331 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
332 * ensure we always pass in a non-zero value.
334 alloc_blocks
= XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
)) + 1;
335 alloc_blocks
= XFS_FILEOFF_MIN(MAXEXTLEN
,
336 rounddown_pow_of_two(alloc_blocks
));
338 xfs_icsb_sync_counters(mp
, XFS_ICSB_LAZY_COUNT
);
339 freesp
= mp
->m_sb
.sb_fdblocks
;
340 if (freesp
< mp
->m_low_space
[XFS_LOWSP_5_PCNT
]) {
342 if (freesp
< mp
->m_low_space
[XFS_LOWSP_4_PCNT
])
344 if (freesp
< mp
->m_low_space
[XFS_LOWSP_3_PCNT
])
346 if (freesp
< mp
->m_low_space
[XFS_LOWSP_2_PCNT
])
348 if (freesp
< mp
->m_low_space
[XFS_LOWSP_1_PCNT
])
352 alloc_blocks
>>= shift
;
355 if (alloc_blocks
< mp
->m_writeio_blocks
)
356 alloc_blocks
= mp
->m_writeio_blocks
;
362 xfs_iomap_write_delay(
366 xfs_bmbt_irec_t
*ret_imap
)
368 xfs_mount_t
*mp
= ip
->i_mount
;
369 xfs_fileoff_t offset_fsb
;
370 xfs_fileoff_t last_fsb
;
371 xfs_off_t aligned_offset
;
372 xfs_fileoff_t ioalign
;
375 xfs_bmbt_irec_t imap
[XFS_WRITE_IMAPS
];
376 int prealloc
, flushed
= 0;
379 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
382 * Make sure that the dquots are there. This doesn't hold
383 * the ilock across a disk read.
385 error
= xfs_qm_dqattach_locked(ip
, 0);
387 return XFS_ERROR(error
);
389 extsz
= xfs_get_extsz_hint(ip
);
390 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
393 error
= xfs_iomap_eof_want_preallocate(mp
, ip
, offset
, count
,
394 imap
, XFS_WRITE_IMAPS
, &prealloc
);
400 xfs_fsblock_t alloc_blocks
= xfs_iomap_prealloc_size(mp
, ip
);
402 aligned_offset
= XFS_WRITEIO_ALIGN(mp
, (offset
+ count
- 1));
403 ioalign
= XFS_B_TO_FSBT(mp
, aligned_offset
);
404 last_fsb
= ioalign
+ alloc_blocks
;
406 last_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)(offset
+ count
)));
409 if (prealloc
|| extsz
) {
410 error
= xfs_iomap_eof_align_last_fsb(mp
, ip
, extsz
, &last_fsb
);
416 * Make sure preallocation does not create extents beyond the range we
417 * actually support in this filesystem.
419 if (last_fsb
> XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
))
420 last_fsb
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
422 ASSERT(last_fsb
> offset_fsb
);
424 nimaps
= XFS_WRITE_IMAPS
;
425 error
= xfs_bmapi_delay(ip
, offset_fsb
, last_fsb
- offset_fsb
,
426 imap
, &nimaps
, XFS_BMAPI_ENTIRE
);
433 return XFS_ERROR(error
);
437 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. For
438 * ENOSPC, * flush all other inodes with delalloc blocks to free up
439 * some of the excess reserved metadata space. For both cases, retry
440 * without EOF preallocation.
443 trace_xfs_delalloc_enospc(ip
, offset
, count
);
445 return XFS_ERROR(error
? error
: ENOSPC
);
447 if (error
== ENOSPC
) {
448 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
449 xfs_flush_inodes(ip
);
450 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
459 if (!(imap
[0].br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
460 return xfs_alert_fsblock_zero(ip
, &imap
[0]);
467 * Pass in a delayed allocate extent, convert it to real extents;
468 * return to the caller the extent we create which maps on top of
469 * the originating callers request.
471 * Called without a lock on the inode.
473 * We no longer bother to look at the incoming map - all we have to
474 * guarantee is that whatever we allocate fills the required range.
477 xfs_iomap_write_allocate(
481 xfs_bmbt_irec_t
*imap
)
483 xfs_mount_t
*mp
= ip
->i_mount
;
484 xfs_fileoff_t offset_fsb
, last_block
;
485 xfs_fileoff_t end_fsb
, map_start_fsb
;
486 xfs_fsblock_t first_block
;
487 xfs_bmap_free_t free_list
;
488 xfs_filblks_t count_fsb
;
490 int nimaps
, committed
;
495 * Make sure that the dquots are there.
497 error
= xfs_qm_dqattach(ip
, 0);
499 return XFS_ERROR(error
);
501 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
502 count_fsb
= imap
->br_blockcount
;
503 map_start_fsb
= imap
->br_startoff
;
505 XFS_STATS_ADD(xs_xstrat_bytes
, XFS_FSB_TO_B(mp
, count_fsb
));
507 while (count_fsb
!= 0) {
509 * Set up a transaction with which to allocate the
510 * backing store for the file. Do allocations in a
511 * loop until we get some space in the range we are
512 * interested in. The other space that might be allocated
513 * is in the delayed allocation extent on which we sit
514 * but before our buffer starts.
518 while (nimaps
== 0) {
519 tp
= xfs_trans_alloc(mp
, XFS_TRANS_STRAT_WRITE
);
520 tp
->t_flags
|= XFS_TRANS_RESERVE
;
521 nres
= XFS_EXTENTADD_SPACE_RES(mp
, XFS_DATA_FORK
);
522 error
= xfs_trans_reserve(tp
, nres
,
523 XFS_WRITE_LOG_RES(mp
),
524 0, XFS_TRANS_PERM_LOG_RES
,
525 XFS_WRITE_LOG_COUNT
);
527 xfs_trans_cancel(tp
, 0);
528 return XFS_ERROR(error
);
530 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
531 xfs_trans_ijoin(tp
, ip
, 0);
533 xfs_bmap_init(&free_list
, &first_block
);
536 * it is possible that the extents have changed since
537 * we did the read call as we dropped the ilock for a
538 * while. We have to be careful about truncates or hole
539 * punchs here - we are not allowed to allocate
540 * non-delalloc blocks here.
542 * The only protection against truncation is the pages
543 * for the range we are being asked to convert are
544 * locked and hence a truncate will block on them
547 * As a result, if we go beyond the range we really
548 * need and hit an delalloc extent boundary followed by
549 * a hole while we have excess blocks in the map, we
550 * will fill the hole incorrectly and overrun the
551 * transaction reservation.
553 * Using a single map prevents this as we are forced to
554 * check each map we look for overlap with the desired
555 * range and abort as soon as we find it. Also, given
556 * that we only return a single map, having one beyond
557 * what we can return is probably a bit silly.
559 * We also need to check that we don't go beyond EOF;
560 * this is a truncate optimisation as a truncate sets
561 * the new file size before block on the pages we
562 * currently have locked under writeback. Because they
563 * are about to be tossed, we don't need to write them
567 end_fsb
= XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
));
568 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
573 last_block
= XFS_FILEOFF_MAX(last_block
, end_fsb
);
574 if ((map_start_fsb
+ count_fsb
) > last_block
) {
575 count_fsb
= last_block
- map_start_fsb
;
576 if (count_fsb
== 0) {
583 * From this point onwards we overwrite the imap
584 * pointer that the caller gave to us.
586 error
= xfs_bmapi_write(tp
, ip
, map_start_fsb
,
588 XFS_BMAPI_STACK_SWITCH
,
590 imap
, &nimaps
, &free_list
);
594 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
598 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
602 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
606 * See if we were able to allocate an extent that
607 * covers at least part of the callers request
609 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
610 return xfs_alert_fsblock_zero(ip
, imap
);
612 if ((offset_fsb
>= imap
->br_startoff
) &&
613 (offset_fsb
< (imap
->br_startoff
+
614 imap
->br_blockcount
))) {
615 XFS_STATS_INC(xs_xstrat_quick
);
620 * So far we have not mapped the requested part of the
621 * file, just surrounding data, try again.
623 count_fsb
-= imap
->br_blockcount
;
624 map_start_fsb
= imap
->br_startoff
+ imap
->br_blockcount
;
628 xfs_bmap_cancel(&free_list
);
629 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
631 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
632 return XFS_ERROR(error
);
636 xfs_iomap_write_unwritten(
641 xfs_mount_t
*mp
= ip
->i_mount
;
642 xfs_fileoff_t offset_fsb
;
643 xfs_filblks_t count_fsb
;
644 xfs_filblks_t numblks_fsb
;
645 xfs_fsblock_t firstfsb
;
648 xfs_bmbt_irec_t imap
;
649 xfs_bmap_free_t free_list
;
655 trace_xfs_unwritten_convert(ip
, offset
, count
);
657 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
658 count_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ count
);
659 count_fsb
= (xfs_filblks_t
)(count_fsb
- offset_fsb
);
662 * Reserve enough blocks in this transaction for two complete extent
663 * btree splits. We may be converting the middle part of an unwritten
664 * extent and in this case we will insert two new extents in the btree
665 * each of which could cause a full split.
667 * This reservation amount will be used in the first call to
668 * xfs_bmbt_split() to select an AG with enough space to satisfy the
669 * rest of the operation.
671 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0) << 1;
675 * set up a transaction to convert the range of extents
676 * from unwritten to real. Do allocations in a loop until
677 * we have covered the range passed in.
679 * Note that we open code the transaction allocation here
680 * to pass KM_NOFS--we can't risk to recursing back into
681 * the filesystem here as we might be asked to write out
682 * the same inode that we complete here and might deadlock
685 sb_start_intwrite(mp
->m_super
);
686 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_STRAT_WRITE
, KM_NOFS
);
687 tp
->t_flags
|= XFS_TRANS_RESERVE
| XFS_TRANS_FREEZE_PROT
;
688 error
= xfs_trans_reserve(tp
, resblks
,
689 XFS_WRITE_LOG_RES(mp
), 0,
690 XFS_TRANS_PERM_LOG_RES
,
691 XFS_WRITE_LOG_COUNT
);
693 xfs_trans_cancel(tp
, 0);
694 return XFS_ERROR(error
);
697 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
698 xfs_trans_ijoin(tp
, ip
, 0);
701 * Modify the unwritten extent state of the buffer.
703 xfs_bmap_init(&free_list
, &firstfsb
);
705 error
= xfs_bmapi_write(tp
, ip
, offset_fsb
, count_fsb
,
706 XFS_BMAPI_CONVERT
, &firstfsb
,
707 1, &imap
, &nimaps
, &free_list
);
709 goto error_on_bmapi_transaction
;
712 * Log the updated inode size as we go. We have to be careful
713 * to only log it up to the actual write offset if it is
714 * halfway into a block.
716 i_size
= XFS_FSB_TO_B(mp
, offset_fsb
+ count_fsb
);
717 if (i_size
> offset
+ count
)
718 i_size
= offset
+ count
;
720 i_size
= xfs_new_eof(ip
, i_size
);
722 ip
->i_d
.di_size
= i_size
;
723 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
726 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
728 goto error_on_bmapi_transaction
;
730 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
731 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
733 return XFS_ERROR(error
);
735 if (!(imap
.br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
736 return xfs_alert_fsblock_zero(ip
, &imap
);
738 if ((numblks_fsb
= imap
.br_blockcount
) == 0) {
740 * The numblks_fsb value should always get
741 * smaller, otherwise the loop is stuck.
743 ASSERT(imap
.br_blockcount
);
746 offset_fsb
+= numblks_fsb
;
747 count_fsb
-= numblks_fsb
;
748 } while (count_fsb
> 0);
752 error_on_bmapi_transaction
:
753 xfs_bmap_cancel(&free_list
);
754 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
));
755 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
756 return XFS_ERROR(error
);