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
23 #include "xfs_trans.h"
26 #include "xfs_alloc.h"
27 #include "xfs_quota.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_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_btree.h"
37 #include "xfs_rtalloc.h"
38 #include "xfs_error.h"
39 #include "xfs_itable.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_trans_space.h"
44 #include "xfs_utils.h"
45 #include "xfs_iomap.h"
46 #include "xfs_trace.h"
49 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
51 #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
54 xfs_iomap_eof_align_last_fsb(
58 xfs_fileoff_t
*last_fsb
)
60 xfs_fileoff_t new_last_fsb
= 0;
61 xfs_extlen_t align
= 0;
64 if (!XFS_IS_REALTIME_INODE(ip
)) {
66 * Round up the allocation request to a stripe unit
67 * (m_dalign) boundary if the file size is >= stripe unit
68 * size, and we are allocating past the allocation eof.
70 * If mounted with the "-o swalloc" option the alignment is
71 * increased from the strip unit size to the stripe width.
73 if (mp
->m_swidth
&& (mp
->m_flags
& XFS_MOUNT_SWALLOC
))
75 else if (mp
->m_dalign
)
78 if (align
&& XFS_ISIZE(ip
) >= XFS_FSB_TO_B(mp
, align
))
79 new_last_fsb
= roundup_64(*last_fsb
, align
);
83 * Always round up the allocation request to an extent boundary
84 * (when file on a real-time subvolume or has di_extsize hint).
88 align
= roundup_64(new_last_fsb
, extsize
);
91 new_last_fsb
= roundup_64(*last_fsb
, align
);
95 error
= xfs_bmap_eof(ip
, new_last_fsb
, XFS_DATA_FORK
, &eof
);
99 *last_fsb
= new_last_fsb
;
105 xfs_alert_fsblock_zero(
107 xfs_bmbt_irec_t
*imap
)
109 xfs_alert_tag(ip
->i_mount
, XFS_PTAG_FSBLOCK_ZERO
,
110 "Access to block zero in inode %llu "
111 "start_block: %llx start_off: %llx "
112 "blkcnt: %llx extent-state: %x\n",
113 (unsigned long long)ip
->i_ino
,
114 (unsigned long long)imap
->br_startblock
,
115 (unsigned long long)imap
->br_startoff
,
116 (unsigned long long)imap
->br_blockcount
,
122 xfs_iomap_write_direct(
126 xfs_bmbt_irec_t
*imap
,
129 xfs_mount_t
*mp
= ip
->i_mount
;
130 xfs_fileoff_t offset_fsb
;
131 xfs_fileoff_t last_fsb
;
132 xfs_filblks_t count_fsb
, resaligned
;
133 xfs_fsblock_t firstfsb
;
134 xfs_extlen_t extsz
, temp
;
140 xfs_bmap_free_t free_list
;
141 uint qblocks
, resblks
, resrtextents
;
146 * Make sure that the dquots are there. This doesn't hold
147 * the ilock across a disk read.
149 error
= xfs_qm_dqattach_locked(ip
, 0);
151 return XFS_ERROR(error
);
153 rt
= XFS_IS_REALTIME_INODE(ip
);
154 extsz
= xfs_get_extsz_hint(ip
);
156 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
157 last_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)(offset
+ count
)));
158 if ((offset
+ count
) > XFS_ISIZE(ip
)) {
159 error
= xfs_iomap_eof_align_last_fsb(mp
, ip
, extsz
, &last_fsb
);
163 if (nmaps
&& (imap
->br_startblock
== HOLESTARTBLOCK
))
164 last_fsb
= MIN(last_fsb
, (xfs_fileoff_t
)
165 imap
->br_blockcount
+
168 count_fsb
= last_fsb
- offset_fsb
;
169 ASSERT(count_fsb
> 0);
171 resaligned
= count_fsb
;
172 if (unlikely(extsz
)) {
173 if ((temp
= do_mod(offset_fsb
, extsz
)))
175 if ((temp
= do_mod(resaligned
, extsz
)))
176 resaligned
+= extsz
- temp
;
180 resrtextents
= qblocks
= resaligned
;
181 resrtextents
/= mp
->m_sb
.sb_rextsize
;
182 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
183 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
186 resblks
= qblocks
= XFS_DIOSTRAT_SPACE_RES(mp
, resaligned
);
187 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
191 * Allocate and setup the transaction
193 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
194 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
195 error
= xfs_trans_reserve(tp
, resblks
,
196 XFS_WRITE_LOG_RES(mp
), resrtextents
,
197 XFS_TRANS_PERM_LOG_RES
,
198 XFS_WRITE_LOG_COUNT
);
200 * Check for running out of space, note: need lock to return
203 xfs_trans_cancel(tp
, 0);
204 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
208 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
, 0, quota_flag
);
212 xfs_trans_ijoin(tp
, ip
, 0);
215 if (offset
< XFS_ISIZE(ip
) || extsz
)
216 bmapi_flag
|= XFS_BMAPI_PREALLOC
;
219 * From this point onwards we overwrite the imap pointer that the
222 xfs_bmap_init(&free_list
, &firstfsb
);
224 error
= xfs_bmapi_write(tp
, ip
, offset_fsb
, count_fsb
, bmapi_flag
,
225 &firstfsb
, 0, imap
, &nimaps
, &free_list
);
230 * Complete the transaction
232 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
235 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
240 * Copy any maps to caller's array and return any error.
247 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
))) {
248 error
= xfs_alert_fsblock_zero(ip
, imap
);
254 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
255 xfs_bmap_cancel(&free_list
);
256 xfs_trans_unreserve_quota_nblks(tp
, ip
, qblocks
, 0, quota_flag
);
258 error1
: /* Just cancel transaction */
259 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
262 return XFS_ERROR(error
);
266 * If the caller is doing a write at the end of the file, then extend the
267 * allocation out to the file system's write iosize. We clean up any extra
268 * space left over when the file is closed in xfs_inactive().
270 * If we find we already have delalloc preallocation beyond EOF, don't do more
271 * preallocation as it it not needed.
274 xfs_iomap_eof_want_preallocate(
279 xfs_bmbt_irec_t
*imap
,
283 xfs_fileoff_t start_fsb
;
284 xfs_filblks_t count_fsb
;
285 xfs_fsblock_t firstblock
;
287 int found_delalloc
= 0;
290 if (offset
+ count
<= XFS_ISIZE(ip
))
294 * If there are any real blocks past eof, then don't
295 * do any speculative allocation.
297 start_fsb
= XFS_B_TO_FSBT(mp
, ((xfs_ufsize_t
)(offset
+ count
- 1)));
298 count_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
299 while (count_fsb
> 0) {
301 firstblock
= NULLFSBLOCK
;
302 error
= xfs_bmapi_read(ip
, start_fsb
, count_fsb
, imap
, &imaps
,
306 for (n
= 0; n
< imaps
; n
++) {
307 if ((imap
[n
].br_startblock
!= HOLESTARTBLOCK
) &&
308 (imap
[n
].br_startblock
!= DELAYSTARTBLOCK
))
310 start_fsb
+= imap
[n
].br_blockcount
;
311 count_fsb
-= imap
[n
].br_blockcount
;
313 if (imap
[n
].br_startblock
== DELAYSTARTBLOCK
)
323 * If we don't have a user specified preallocation size, dynamically increase
324 * the preallocation size as the size of the file grows. Cap the maximum size
325 * at a single extent or less if the filesystem is near full. The closer the
326 * filesystem is to full, the smaller the maximum prealocation.
329 xfs_iomap_prealloc_size(
330 struct xfs_mount
*mp
,
331 struct xfs_inode
*ip
)
333 xfs_fsblock_t alloc_blocks
= 0;
335 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
340 * rounddown_pow_of_two() returns an undefined result
341 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
342 * ensure we always pass in a non-zero value.
344 alloc_blocks
= XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
)) + 1;
345 alloc_blocks
= XFS_FILEOFF_MIN(MAXEXTLEN
,
346 rounddown_pow_of_two(alloc_blocks
));
348 xfs_icsb_sync_counters(mp
, XFS_ICSB_LAZY_COUNT
);
349 freesp
= mp
->m_sb
.sb_fdblocks
;
350 if (freesp
< mp
->m_low_space
[XFS_LOWSP_5_PCNT
]) {
352 if (freesp
< mp
->m_low_space
[XFS_LOWSP_4_PCNT
])
354 if (freesp
< mp
->m_low_space
[XFS_LOWSP_3_PCNT
])
356 if (freesp
< mp
->m_low_space
[XFS_LOWSP_2_PCNT
])
358 if (freesp
< mp
->m_low_space
[XFS_LOWSP_1_PCNT
])
362 alloc_blocks
>>= shift
;
365 if (alloc_blocks
< mp
->m_writeio_blocks
)
366 alloc_blocks
= mp
->m_writeio_blocks
;
372 xfs_iomap_write_delay(
376 xfs_bmbt_irec_t
*ret_imap
)
378 xfs_mount_t
*mp
= ip
->i_mount
;
379 xfs_fileoff_t offset_fsb
;
380 xfs_fileoff_t last_fsb
;
381 xfs_off_t aligned_offset
;
382 xfs_fileoff_t ioalign
;
385 xfs_bmbt_irec_t imap
[XFS_WRITE_IMAPS
];
386 int prealloc
, flushed
= 0;
389 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
392 * Make sure that the dquots are there. This doesn't hold
393 * the ilock across a disk read.
395 error
= xfs_qm_dqattach_locked(ip
, 0);
397 return XFS_ERROR(error
);
399 extsz
= xfs_get_extsz_hint(ip
);
400 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
403 error
= xfs_iomap_eof_want_preallocate(mp
, ip
, offset
, count
,
404 imap
, XFS_WRITE_IMAPS
, &prealloc
);
410 xfs_fsblock_t alloc_blocks
= xfs_iomap_prealloc_size(mp
, ip
);
412 aligned_offset
= XFS_WRITEIO_ALIGN(mp
, (offset
+ count
- 1));
413 ioalign
= XFS_B_TO_FSBT(mp
, aligned_offset
);
414 last_fsb
= ioalign
+ alloc_blocks
;
416 last_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)(offset
+ count
)));
419 if (prealloc
|| extsz
) {
420 error
= xfs_iomap_eof_align_last_fsb(mp
, ip
, extsz
, &last_fsb
);
425 nimaps
= XFS_WRITE_IMAPS
;
426 error
= xfs_bmapi_delay(ip
, offset_fsb
, last_fsb
- offset_fsb
,
427 imap
, &nimaps
, XFS_BMAPI_ENTIRE
);
434 return XFS_ERROR(error
);
438 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. For
439 * ENOSPC, * flush all other inodes with delalloc blocks to free up
440 * some of the excess reserved metadata space. For both cases, retry
441 * without EOF preallocation.
444 trace_xfs_delalloc_enospc(ip
, offset
, count
);
446 return XFS_ERROR(error
? error
: ENOSPC
);
448 if (error
== ENOSPC
) {
449 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
450 xfs_flush_inodes(ip
);
451 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
460 if (!(imap
[0].br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
461 return xfs_alert_fsblock_zero(ip
, &imap
[0]);
468 * Pass in a delayed allocate extent, convert it to real extents;
469 * return to the caller the extent we create which maps on top of
470 * the originating callers request.
472 * Called without a lock on the inode.
474 * We no longer bother to look at the incoming map - all we have to
475 * guarantee is that whatever we allocate fills the required range.
478 xfs_iomap_write_allocate(
482 xfs_bmbt_irec_t
*imap
)
484 xfs_mount_t
*mp
= ip
->i_mount
;
485 xfs_fileoff_t offset_fsb
, last_block
;
486 xfs_fileoff_t end_fsb
, map_start_fsb
;
487 xfs_fsblock_t first_block
;
488 xfs_bmap_free_t free_list
;
489 xfs_filblks_t count_fsb
;
491 int nimaps
, committed
;
496 * Make sure that the dquots are there.
498 error
= xfs_qm_dqattach(ip
, 0);
500 return XFS_ERROR(error
);
502 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
503 count_fsb
= imap
->br_blockcount
;
504 map_start_fsb
= imap
->br_startoff
;
506 XFS_STATS_ADD(xs_xstrat_bytes
, XFS_FSB_TO_B(mp
, count_fsb
));
508 while (count_fsb
!= 0) {
510 * Set up a transaction with which to allocate the
511 * backing store for the file. Do allocations in a
512 * loop until we get some space in the range we are
513 * interested in. The other space that might be allocated
514 * is in the delayed allocation extent on which we sit
515 * but before our buffer starts.
519 while (nimaps
== 0) {
520 tp
= xfs_trans_alloc(mp
, XFS_TRANS_STRAT_WRITE
);
521 tp
->t_flags
|= XFS_TRANS_RESERVE
;
522 nres
= XFS_EXTENTADD_SPACE_RES(mp
, XFS_DATA_FORK
);
523 error
= xfs_trans_reserve(tp
, nres
,
524 XFS_WRITE_LOG_RES(mp
),
525 0, XFS_TRANS_PERM_LOG_RES
,
526 XFS_WRITE_LOG_COUNT
);
528 xfs_trans_cancel(tp
, 0);
529 return XFS_ERROR(error
);
531 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
532 xfs_trans_ijoin(tp
, ip
, 0);
534 xfs_bmap_init(&free_list
, &first_block
);
537 * it is possible that the extents have changed since
538 * we did the read call as we dropped the ilock for a
539 * while. We have to be careful about truncates or hole
540 * punchs here - we are not allowed to allocate
541 * non-delalloc blocks here.
543 * The only protection against truncation is the pages
544 * for the range we are being asked to convert are
545 * locked and hence a truncate will block on them
548 * As a result, if we go beyond the range we really
549 * need and hit an delalloc extent boundary followed by
550 * a hole while we have excess blocks in the map, we
551 * will fill the hole incorrectly and overrun the
552 * transaction reservation.
554 * Using a single map prevents this as we are forced to
555 * check each map we look for overlap with the desired
556 * range and abort as soon as we find it. Also, given
557 * that we only return a single map, having one beyond
558 * what we can return is probably a bit silly.
560 * We also need to check that we don't go beyond EOF;
561 * this is a truncate optimisation as a truncate sets
562 * the new file size before block on the pages we
563 * currently have locked under writeback. Because they
564 * are about to be tossed, we don't need to write them
568 end_fsb
= XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
));
569 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
574 last_block
= XFS_FILEOFF_MAX(last_block
, end_fsb
);
575 if ((map_start_fsb
+ count_fsb
) > last_block
) {
576 count_fsb
= last_block
- map_start_fsb
;
577 if (count_fsb
== 0) {
584 * From this point onwards we overwrite the imap
585 * pointer that the caller gave to us.
587 error
= xfs_bmapi_write(tp
, ip
, map_start_fsb
,
588 count_fsb
, 0, &first_block
, 1,
589 imap
, &nimaps
, &free_list
);
593 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
597 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
601 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
605 * See if we were able to allocate an extent that
606 * covers at least part of the callers request
608 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
609 return xfs_alert_fsblock_zero(ip
, imap
);
611 if ((offset_fsb
>= imap
->br_startoff
) &&
612 (offset_fsb
< (imap
->br_startoff
+
613 imap
->br_blockcount
))) {
614 XFS_STATS_INC(xs_xstrat_quick
);
619 * So far we have not mapped the requested part of the
620 * file, just surrounding data, try again.
622 count_fsb
-= imap
->br_blockcount
;
623 map_start_fsb
= imap
->br_startoff
+ imap
->br_blockcount
;
627 xfs_bmap_cancel(&free_list
);
628 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
630 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
631 return XFS_ERROR(error
);
635 xfs_iomap_write_unwritten(
640 xfs_mount_t
*mp
= ip
->i_mount
;
641 xfs_fileoff_t offset_fsb
;
642 xfs_filblks_t count_fsb
;
643 xfs_filblks_t numblks_fsb
;
644 xfs_fsblock_t firstfsb
;
647 xfs_bmbt_irec_t imap
;
648 xfs_bmap_free_t free_list
;
654 trace_xfs_unwritten_convert(ip
, offset
, count
);
656 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
657 count_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ count
);
658 count_fsb
= (xfs_filblks_t
)(count_fsb
- offset_fsb
);
661 * Reserve enough blocks in this transaction for two complete extent
662 * btree splits. We may be converting the middle part of an unwritten
663 * extent and in this case we will insert two new extents in the btree
664 * each of which could cause a full split.
666 * This reservation amount will be used in the first call to
667 * xfs_bmbt_split() to select an AG with enough space to satisfy the
668 * rest of the operation.
670 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0) << 1;
674 * set up a transaction to convert the range of extents
675 * from unwritten to real. Do allocations in a loop until
676 * we have covered the range passed in.
678 * Note that we open code the transaction allocation here
679 * to pass KM_NOFS--we can't risk to recursing back into
680 * the filesystem here as we might be asked to write out
681 * the same inode that we complete here and might deadlock
684 xfs_wait_for_freeze(mp
, SB_FREEZE_TRANS
);
685 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_STRAT_WRITE
, KM_NOFS
);
686 tp
->t_flags
|= XFS_TRANS_RESERVE
;
687 error
= xfs_trans_reserve(tp
, resblks
,
688 XFS_WRITE_LOG_RES(mp
), 0,
689 XFS_TRANS_PERM_LOG_RES
,
690 XFS_WRITE_LOG_COUNT
);
692 xfs_trans_cancel(tp
, 0);
693 return XFS_ERROR(error
);
696 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
697 xfs_trans_ijoin(tp
, ip
, 0);
700 * Modify the unwritten extent state of the buffer.
702 xfs_bmap_init(&free_list
, &firstfsb
);
704 error
= xfs_bmapi_write(tp
, ip
, offset_fsb
, count_fsb
,
705 XFS_BMAPI_CONVERT
, &firstfsb
,
706 1, &imap
, &nimaps
, &free_list
);
708 goto error_on_bmapi_transaction
;
711 * Log the updated inode size as we go. We have to be careful
712 * to only log it up to the actual write offset if it is
713 * halfway into a block.
715 i_size
= XFS_FSB_TO_B(mp
, offset_fsb
+ count_fsb
);
716 if (i_size
> offset
+ count
)
717 i_size
= offset
+ count
;
719 i_size
= xfs_new_eof(ip
, i_size
);
721 ip
->i_d
.di_size
= i_size
;
722 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
725 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
727 goto error_on_bmapi_transaction
;
729 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
730 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
732 return XFS_ERROR(error
);
734 if (!(imap
.br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
735 return xfs_alert_fsblock_zero(ip
, &imap
);
737 if ((numblks_fsb
= imap
.br_blockcount
) == 0) {
739 * The numblks_fsb value should always get
740 * smaller, otherwise the loop is stuck.
742 ASSERT(imap
.br_blockcount
);
745 offset_fsb
+= numblks_fsb
;
746 count_fsb
-= numblks_fsb
;
747 } while (count_fsb
> 0);
751 error_on_bmapi_transaction
:
752 xfs_bmap_cancel(&free_list
);
753 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
));
754 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
755 return XFS_ERROR(error
);