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_btree.h"
36 #include "xfs_rtalloc.h"
37 #include "xfs_error.h"
38 #include "xfs_itable.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_trans_space.h"
43 #include "xfs_utils.h"
44 #include "xfs_iomap.h"
45 #include "xfs_trace.h"
48 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
50 #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
53 xfs_iomap_eof_align_last_fsb(
57 xfs_fileoff_t
*last_fsb
)
59 xfs_fileoff_t new_last_fsb
= 0;
63 if (XFS_IS_REALTIME_INODE(ip
))
66 * If mounted with the "-o swalloc" option, roundup the allocation
67 * request to a stripe width boundary if the file size is >=
68 * stripe width and we are allocating past the allocation eof.
70 else if (mp
->m_swidth
&& (mp
->m_flags
& XFS_MOUNT_SWALLOC
) &&
71 (ip
->i_size
>= XFS_FSB_TO_B(mp
, mp
->m_swidth
)))
72 new_last_fsb
= roundup_64(*last_fsb
, mp
->m_swidth
);
74 * Roundup the allocation request to a stripe unit (m_dalign) boundary
75 * if the file size is >= stripe unit size, and we are allocating past
78 else if (mp
->m_dalign
&& (ip
->i_size
>= XFS_FSB_TO_B(mp
, mp
->m_dalign
)))
79 new_last_fsb
= roundup_64(*last_fsb
, mp
->m_dalign
);
82 * Always round up the allocation request to an extent boundary
83 * (when file on a real-time subvolume or has di_extsize hint).
87 align
= roundup_64(new_last_fsb
, extsize
);
90 new_last_fsb
= roundup_64(*last_fsb
, align
);
94 error
= xfs_bmap_eof(ip
, new_last_fsb
, XFS_DATA_FORK
, &eof
);
98 *last_fsb
= new_last_fsb
;
104 xfs_alert_fsblock_zero(
106 xfs_bmbt_irec_t
*imap
)
108 xfs_alert_tag(ip
->i_mount
, XFS_PTAG_FSBLOCK_ZERO
,
109 "Access to block zero in inode %llu "
110 "start_block: %llx start_off: %llx "
111 "blkcnt: %llx extent-state: %x\n",
112 (unsigned long long)ip
->i_ino
,
113 (unsigned long long)imap
->br_startblock
,
114 (unsigned long long)imap
->br_startoff
,
115 (unsigned long long)imap
->br_blockcount
,
121 xfs_iomap_write_direct(
125 xfs_bmbt_irec_t
*imap
,
128 xfs_mount_t
*mp
= ip
->i_mount
;
129 xfs_fileoff_t offset_fsb
;
130 xfs_fileoff_t last_fsb
;
131 xfs_filblks_t count_fsb
, resaligned
;
132 xfs_fsblock_t firstfsb
;
133 xfs_extlen_t extsz
, temp
;
139 xfs_bmap_free_t free_list
;
140 uint qblocks
, resblks
, resrtextents
;
145 * Make sure that the dquots are there. This doesn't hold
146 * the ilock across a disk read.
148 error
= xfs_qm_dqattach_locked(ip
, 0);
150 return XFS_ERROR(error
);
152 rt
= XFS_IS_REALTIME_INODE(ip
);
153 extsz
= xfs_get_extsz_hint(ip
);
155 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
156 last_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)(offset
+ count
)));
157 if ((offset
+ count
) > ip
->i_size
) {
158 error
= xfs_iomap_eof_align_last_fsb(mp
, ip
, extsz
, &last_fsb
);
162 if (nmaps
&& (imap
->br_startblock
== HOLESTARTBLOCK
))
163 last_fsb
= MIN(last_fsb
, (xfs_fileoff_t
)
164 imap
->br_blockcount
+
167 count_fsb
= last_fsb
- offset_fsb
;
168 ASSERT(count_fsb
> 0);
170 resaligned
= count_fsb
;
171 if (unlikely(extsz
)) {
172 if ((temp
= do_mod(offset_fsb
, extsz
)))
174 if ((temp
= do_mod(resaligned
, extsz
)))
175 resaligned
+= extsz
- temp
;
179 resrtextents
= qblocks
= resaligned
;
180 resrtextents
/= mp
->m_sb
.sb_rextsize
;
181 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
182 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
185 resblks
= qblocks
= XFS_DIOSTRAT_SPACE_RES(mp
, resaligned
);
186 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
190 * Allocate and setup the transaction
192 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
193 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
194 error
= xfs_trans_reserve(tp
, resblks
,
195 XFS_WRITE_LOG_RES(mp
), resrtextents
,
196 XFS_TRANS_PERM_LOG_RES
,
197 XFS_WRITE_LOG_COUNT
);
199 * Check for running out of space, note: need lock to return
202 xfs_trans_cancel(tp
, 0);
203 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
207 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
, 0, quota_flag
);
211 xfs_trans_ijoin(tp
, ip
, 0);
214 if (offset
< ip
->i_size
|| extsz
)
215 bmapi_flag
|= XFS_BMAPI_PREALLOC
;
218 * From this point onwards we overwrite the imap pointer that the
221 xfs_bmap_init(&free_list
, &firstfsb
);
223 error
= xfs_bmapi_write(tp
, ip
, offset_fsb
, count_fsb
, bmapi_flag
,
224 &firstfsb
, 0, imap
, &nimaps
, &free_list
);
229 * Complete the transaction
231 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
234 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
239 * Copy any maps to caller's array and return any error.
246 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
))) {
247 error
= xfs_alert_fsblock_zero(ip
, imap
);
253 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
254 xfs_bmap_cancel(&free_list
);
255 xfs_trans_unreserve_quota_nblks(tp
, ip
, qblocks
, 0, quota_flag
);
257 error1
: /* Just cancel transaction */
258 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
261 return XFS_ERROR(error
);
265 * If the caller is doing a write at the end of the file, then extend the
266 * allocation out to the file system's write iosize. We clean up any extra
267 * space left over when the file is closed in xfs_inactive().
269 * If we find we already have delalloc preallocation beyond EOF, don't do more
270 * preallocation as it it not needed.
273 xfs_iomap_eof_want_preallocate(
278 xfs_bmbt_irec_t
*imap
,
282 xfs_fileoff_t start_fsb
;
283 xfs_filblks_t count_fsb
;
284 xfs_fsblock_t firstblock
;
286 int found_delalloc
= 0;
289 if ((offset
+ count
) <= ip
->i_size
)
293 * If there are any real blocks past eof, then don't
294 * do any speculative allocation.
296 start_fsb
= XFS_B_TO_FSBT(mp
, ((xfs_ufsize_t
)(offset
+ count
- 1)));
297 count_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
298 while (count_fsb
> 0) {
300 firstblock
= NULLFSBLOCK
;
301 error
= xfs_bmapi_read(ip
, start_fsb
, count_fsb
, imap
, &imaps
,
305 for (n
= 0; n
< imaps
; n
++) {
306 if ((imap
[n
].br_startblock
!= HOLESTARTBLOCK
) &&
307 (imap
[n
].br_startblock
!= DELAYSTARTBLOCK
))
309 start_fsb
+= imap
[n
].br_blockcount
;
310 count_fsb
-= imap
[n
].br_blockcount
;
312 if (imap
[n
].br_startblock
== DELAYSTARTBLOCK
)
322 * If we don't have a user specified preallocation size, dynamically increase
323 * the preallocation size as the size of the file grows. Cap the maximum size
324 * at a single extent or less if the filesystem is near full. The closer the
325 * filesystem is to full, the smaller the maximum prealocation.
328 xfs_iomap_prealloc_size(
329 struct xfs_mount
*mp
,
330 struct xfs_inode
*ip
)
332 xfs_fsblock_t alloc_blocks
= 0;
334 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
339 * rounddown_pow_of_two() returns an undefined result
340 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
341 * ensure we always pass in a non-zero value.
343 alloc_blocks
= XFS_B_TO_FSB(mp
, ip
->i_size
) + 1;
344 alloc_blocks
= XFS_FILEOFF_MIN(MAXEXTLEN
,
345 rounddown_pow_of_two(alloc_blocks
));
347 xfs_icsb_sync_counters(mp
, XFS_ICSB_LAZY_COUNT
);
348 freesp
= mp
->m_sb
.sb_fdblocks
;
349 if (freesp
< mp
->m_low_space
[XFS_LOWSP_5_PCNT
]) {
351 if (freesp
< mp
->m_low_space
[XFS_LOWSP_4_PCNT
])
353 if (freesp
< mp
->m_low_space
[XFS_LOWSP_3_PCNT
])
355 if (freesp
< mp
->m_low_space
[XFS_LOWSP_2_PCNT
])
357 if (freesp
< mp
->m_low_space
[XFS_LOWSP_1_PCNT
])
361 alloc_blocks
>>= shift
;
364 if (alloc_blocks
< mp
->m_writeio_blocks
)
365 alloc_blocks
= mp
->m_writeio_blocks
;
371 xfs_iomap_write_delay(
375 xfs_bmbt_irec_t
*ret_imap
)
377 xfs_mount_t
*mp
= ip
->i_mount
;
378 xfs_fileoff_t offset_fsb
;
379 xfs_fileoff_t last_fsb
;
380 xfs_off_t aligned_offset
;
381 xfs_fileoff_t ioalign
;
384 xfs_bmbt_irec_t imap
[XFS_WRITE_IMAPS
];
385 int prealloc
, flushed
= 0;
388 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
391 * Make sure that the dquots are there. This doesn't hold
392 * the ilock across a disk read.
394 error
= xfs_qm_dqattach_locked(ip
, 0);
396 return XFS_ERROR(error
);
398 extsz
= xfs_get_extsz_hint(ip
);
399 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
402 error
= xfs_iomap_eof_want_preallocate(mp
, ip
, offset
, count
,
403 imap
, XFS_WRITE_IMAPS
, &prealloc
);
409 xfs_fsblock_t alloc_blocks
= xfs_iomap_prealloc_size(mp
, ip
);
411 aligned_offset
= XFS_WRITEIO_ALIGN(mp
, (offset
+ count
- 1));
412 ioalign
= XFS_B_TO_FSBT(mp
, aligned_offset
);
413 last_fsb
= ioalign
+ alloc_blocks
;
415 last_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)(offset
+ count
)));
418 if (prealloc
|| extsz
) {
419 error
= xfs_iomap_eof_align_last_fsb(mp
, ip
, extsz
, &last_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
, ip
->i_size
);
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
,
587 count_fsb
, 0, &first_block
, 1,
588 imap
, &nimaps
, &free_list
);
592 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
596 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
600 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
604 * See if we were able to allocate an extent that
605 * covers at least part of the callers request
607 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
608 return xfs_alert_fsblock_zero(ip
, imap
);
610 if ((offset_fsb
>= imap
->br_startoff
) &&
611 (offset_fsb
< (imap
->br_startoff
+
612 imap
->br_blockcount
))) {
613 XFS_STATS_INC(xs_xstrat_quick
);
618 * So far we have not mapped the requested part of the
619 * file, just surrounding data, try again.
621 count_fsb
-= imap
->br_blockcount
;
622 map_start_fsb
= imap
->br_startoff
+ imap
->br_blockcount
;
626 xfs_bmap_cancel(&free_list
);
627 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
629 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
630 return XFS_ERROR(error
);
634 xfs_iomap_write_unwritten(
639 xfs_mount_t
*mp
= ip
->i_mount
;
640 xfs_fileoff_t offset_fsb
;
641 xfs_filblks_t count_fsb
;
642 xfs_filblks_t numblks_fsb
;
643 xfs_fsblock_t firstfsb
;
646 xfs_bmbt_irec_t imap
;
647 xfs_bmap_free_t free_list
;
652 trace_xfs_unwritten_convert(ip
, offset
, count
);
654 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
655 count_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ count
);
656 count_fsb
= (xfs_filblks_t
)(count_fsb
- offset_fsb
);
659 * Reserve enough blocks in this transaction for two complete extent
660 * btree splits. We may be converting the middle part of an unwritten
661 * extent and in this case we will insert two new extents in the btree
662 * each of which could cause a full split.
664 * This reservation amount will be used in the first call to
665 * xfs_bmbt_split() to select an AG with enough space to satisfy the
666 * rest of the operation.
668 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0) << 1;
672 * set up a transaction to convert the range of extents
673 * from unwritten to real. Do allocations in a loop until
674 * we have covered the range passed in.
676 * Note that we open code the transaction allocation here
677 * to pass KM_NOFS--we can't risk to recursing back into
678 * the filesystem here as we might be asked to write out
679 * the same inode that we complete here and might deadlock
682 xfs_wait_for_freeze(mp
, SB_FREEZE_TRANS
);
683 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_STRAT_WRITE
, KM_NOFS
);
684 tp
->t_flags
|= XFS_TRANS_RESERVE
;
685 error
= xfs_trans_reserve(tp
, resblks
,
686 XFS_WRITE_LOG_RES(mp
), 0,
687 XFS_TRANS_PERM_LOG_RES
,
688 XFS_WRITE_LOG_COUNT
);
690 xfs_trans_cancel(tp
, 0);
691 return XFS_ERROR(error
);
694 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
695 xfs_trans_ijoin(tp
, ip
, 0);
698 * Modify the unwritten extent state of the buffer.
700 xfs_bmap_init(&free_list
, &firstfsb
);
702 error
= xfs_bmapi_write(tp
, ip
, offset_fsb
, count_fsb
,
703 XFS_BMAPI_CONVERT
, &firstfsb
,
704 1, &imap
, &nimaps
, &free_list
);
706 goto error_on_bmapi_transaction
;
708 error
= xfs_bmap_finish(&(tp
), &(free_list
), &committed
);
710 goto error_on_bmapi_transaction
;
712 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
713 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
715 return XFS_ERROR(error
);
717 if (!(imap
.br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
718 return xfs_alert_fsblock_zero(ip
, &imap
);
720 if ((numblks_fsb
= imap
.br_blockcount
) == 0) {
722 * The numblks_fsb value should always get
723 * smaller, otherwise the loop is stuck.
725 ASSERT(imap
.br_blockcount
);
728 offset_fsb
+= numblks_fsb
;
729 count_fsb
-= numblks_fsb
;
730 } while (count_fsb
> 0);
734 error_on_bmapi_transaction
:
735 xfs_bmap_cancel(&free_list
);
736 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
));
737 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
738 return XFS_ERROR(error
);