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_cmn_err_fsblock_zero(
106 xfs_bmbt_irec_t
*imap
)
108 xfs_cmn_err(XFS_PTAG_FSBLOCK_ZERO
, CE_ALERT
, ip
->i_mount
,
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
);
213 bmapi_flag
= XFS_BMAPI_WRITE
;
214 if (offset
< ip
->i_size
|| extsz
)
215 bmapi_flag
|= XFS_BMAPI_PREALLOC
;
218 * Issue the xfs_bmapi() call to allocate the blocks.
220 * From this point onwards we overwrite the imap pointer that the
223 xfs_bmap_init(&free_list
, &firstfsb
);
225 error
= xfs_bmapi(tp
, ip
, offset_fsb
, count_fsb
, bmapi_flag
,
226 &firstfsb
, 0, imap
, &nimaps
, &free_list
);
231 * Complete the transaction
233 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
236 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
241 * Copy any maps to caller's array and return any error.
248 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
))) {
249 error
= xfs_cmn_err_fsblock_zero(ip
, imap
);
255 error0
: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
256 xfs_bmap_cancel(&free_list
);
257 xfs_trans_unreserve_quota_nblks(tp
, ip
, qblocks
, 0, quota_flag
);
259 error1
: /* Just cancel transaction */
260 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
263 return XFS_ERROR(error
);
267 * If the caller is doing a write at the end of the file, then extend the
268 * allocation out to the file system's write iosize. We clean up any extra
269 * space left over when the file is closed in xfs_inactive().
271 * If we find we already have delalloc preallocation beyond EOF, don't do more
272 * preallocation as it it not needed.
275 xfs_iomap_eof_want_preallocate(
280 xfs_bmbt_irec_t
*imap
,
284 xfs_fileoff_t start_fsb
;
285 xfs_filblks_t count_fsb
;
286 xfs_fsblock_t firstblock
;
288 int found_delalloc
= 0;
291 if ((offset
+ count
) <= ip
->i_size
)
295 * If there are any real blocks past eof, then don't
296 * do any speculative allocation.
298 start_fsb
= XFS_B_TO_FSBT(mp
, ((xfs_ufsize_t
)(offset
+ count
- 1)));
299 count_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)XFS_MAXIOFFSET(mp
));
300 while (count_fsb
> 0) {
302 firstblock
= NULLFSBLOCK
;
303 error
= xfs_bmapi(NULL
, ip
, start_fsb
, count_fsb
, 0,
304 &firstblock
, 0, imap
, &imaps
, NULL
);
307 for (n
= 0; n
< imaps
; n
++) {
308 if ((imap
[n
].br_startblock
!= HOLESTARTBLOCK
) &&
309 (imap
[n
].br_startblock
!= DELAYSTARTBLOCK
))
311 start_fsb
+= imap
[n
].br_blockcount
;
312 count_fsb
-= imap
[n
].br_blockcount
;
314 if (imap
[n
].br_startblock
== DELAYSTARTBLOCK
)
324 * If we don't have a user specified preallocation size, dynamically increase
325 * the preallocation size as the size of the file grows. Cap the maximum size
326 * at a single extent or less if the filesystem is near full. The closer the
327 * filesystem is to full, the smaller the maximum prealocation.
330 xfs_iomap_prealloc_size(
331 struct xfs_mount
*mp
,
332 struct xfs_inode
*ip
)
334 xfs_fsblock_t alloc_blocks
= 0;
336 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)) {
341 * rounddown_pow_of_two() returns an undefined result
342 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
343 * ensure we always pass in a non-zero value.
345 alloc_blocks
= XFS_B_TO_FSB(mp
, ip
->i_size
) + 1;
346 alloc_blocks
= XFS_FILEOFF_MIN(MAXEXTLEN
,
347 rounddown_pow_of_two(alloc_blocks
));
349 xfs_icsb_sync_counters(mp
, XFS_ICSB_LAZY_COUNT
);
350 freesp
= mp
->m_sb
.sb_fdblocks
;
351 if (freesp
< mp
->m_low_space
[XFS_LOWSP_5_PCNT
]) {
353 if (freesp
< mp
->m_low_space
[XFS_LOWSP_4_PCNT
])
355 if (freesp
< mp
->m_low_space
[XFS_LOWSP_3_PCNT
])
357 if (freesp
< mp
->m_low_space
[XFS_LOWSP_2_PCNT
])
359 if (freesp
< mp
->m_low_space
[XFS_LOWSP_1_PCNT
])
363 alloc_blocks
>>= shift
;
366 if (alloc_blocks
< mp
->m_writeio_blocks
)
367 alloc_blocks
= mp
->m_writeio_blocks
;
373 xfs_iomap_write_delay(
377 xfs_bmbt_irec_t
*ret_imap
)
379 xfs_mount_t
*mp
= ip
->i_mount
;
380 xfs_fileoff_t offset_fsb
;
381 xfs_fileoff_t last_fsb
;
382 xfs_off_t aligned_offset
;
383 xfs_fileoff_t ioalign
;
384 xfs_fsblock_t firstblock
;
387 xfs_bmbt_irec_t imap
[XFS_WRITE_IMAPS
];
388 int prealloc
, flushed
= 0;
391 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
394 * Make sure that the dquots are there. This doesn't hold
395 * the ilock across a disk read.
397 error
= xfs_qm_dqattach_locked(ip
, 0);
399 return XFS_ERROR(error
);
401 extsz
= xfs_get_extsz_hint(ip
);
402 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
405 error
= xfs_iomap_eof_want_preallocate(mp
, ip
, offset
, count
,
406 imap
, XFS_WRITE_IMAPS
, &prealloc
);
412 xfs_fsblock_t alloc_blocks
= xfs_iomap_prealloc_size(mp
, ip
);
414 aligned_offset
= XFS_WRITEIO_ALIGN(mp
, (offset
+ count
- 1));
415 ioalign
= XFS_B_TO_FSBT(mp
, aligned_offset
);
416 last_fsb
= ioalign
+ alloc_blocks
;
418 last_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)(offset
+ count
)));
421 if (prealloc
|| extsz
) {
422 error
= xfs_iomap_eof_align_last_fsb(mp
, ip
, extsz
, &last_fsb
);
427 nimaps
= XFS_WRITE_IMAPS
;
428 firstblock
= NULLFSBLOCK
;
429 error
= xfs_bmapi(NULL
, ip
, offset_fsb
,
430 (xfs_filblks_t
)(last_fsb
- offset_fsb
),
431 XFS_BMAPI_DELAY
| XFS_BMAPI_WRITE
|
432 XFS_BMAPI_ENTIRE
, &firstblock
, 1, imap
,
440 return XFS_ERROR(error
);
444 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. For
445 * ENOSPC, * flush all other inodes with delalloc blocks to free up
446 * some of the excess reserved metadata space. For both cases, retry
447 * without EOF preallocation.
450 trace_xfs_delalloc_enospc(ip
, offset
, count
);
452 return XFS_ERROR(error
? error
: ENOSPC
);
454 if (error
== ENOSPC
) {
455 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
456 xfs_flush_inodes(ip
);
457 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
466 if (!(imap
[0].br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
467 return xfs_cmn_err_fsblock_zero(ip
, &imap
[0]);
474 * Pass in a delayed allocate extent, convert it to real extents;
475 * return to the caller the extent we create which maps on top of
476 * the originating callers request.
478 * Called without a lock on the inode.
480 * We no longer bother to look at the incoming map - all we have to
481 * guarantee is that whatever we allocate fills the required range.
484 xfs_iomap_write_allocate(
488 xfs_bmbt_irec_t
*imap
)
490 xfs_mount_t
*mp
= ip
->i_mount
;
491 xfs_fileoff_t offset_fsb
, last_block
;
492 xfs_fileoff_t end_fsb
, map_start_fsb
;
493 xfs_fsblock_t first_block
;
494 xfs_bmap_free_t free_list
;
495 xfs_filblks_t count_fsb
;
497 int nimaps
, committed
;
502 * Make sure that the dquots are there.
504 error
= xfs_qm_dqattach(ip
, 0);
506 return XFS_ERROR(error
);
508 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
509 count_fsb
= imap
->br_blockcount
;
510 map_start_fsb
= imap
->br_startoff
;
512 XFS_STATS_ADD(xs_xstrat_bytes
, XFS_FSB_TO_B(mp
, count_fsb
));
514 while (count_fsb
!= 0) {
516 * Set up a transaction with which to allocate the
517 * backing store for the file. Do allocations in a
518 * loop until we get some space in the range we are
519 * interested in. The other space that might be allocated
520 * is in the delayed allocation extent on which we sit
521 * but before our buffer starts.
525 while (nimaps
== 0) {
526 tp
= xfs_trans_alloc(mp
, XFS_TRANS_STRAT_WRITE
);
527 tp
->t_flags
|= XFS_TRANS_RESERVE
;
528 nres
= XFS_EXTENTADD_SPACE_RES(mp
, XFS_DATA_FORK
);
529 error
= xfs_trans_reserve(tp
, nres
,
530 XFS_WRITE_LOG_RES(mp
),
531 0, XFS_TRANS_PERM_LOG_RES
,
532 XFS_WRITE_LOG_COUNT
);
534 xfs_trans_cancel(tp
, 0);
535 return XFS_ERROR(error
);
537 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
538 xfs_trans_ijoin(tp
, ip
);
540 xfs_bmap_init(&free_list
, &first_block
);
543 * it is possible that the extents have changed since
544 * we did the read call as we dropped the ilock for a
545 * while. We have to be careful about truncates or hole
546 * punchs here - we are not allowed to allocate
547 * non-delalloc blocks here.
549 * The only protection against truncation is the pages
550 * for the range we are being asked to convert are
551 * locked and hence a truncate will block on them
554 * As a result, if we go beyond the range we really
555 * need and hit an delalloc extent boundary followed by
556 * a hole while we have excess blocks in the map, we
557 * will fill the hole incorrectly and overrun the
558 * transaction reservation.
560 * Using a single map prevents this as we are forced to
561 * check each map we look for overlap with the desired
562 * range and abort as soon as we find it. Also, given
563 * that we only return a single map, having one beyond
564 * what we can return is probably a bit silly.
566 * We also need to check that we don't go beyond EOF;
567 * this is a truncate optimisation as a truncate sets
568 * the new file size before block on the pages we
569 * currently have locked under writeback. Because they
570 * are about to be tossed, we don't need to write them
574 end_fsb
= XFS_B_TO_FSB(mp
, ip
->i_size
);
575 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
580 last_block
= XFS_FILEOFF_MAX(last_block
, end_fsb
);
581 if ((map_start_fsb
+ count_fsb
) > last_block
) {
582 count_fsb
= last_block
- map_start_fsb
;
583 if (count_fsb
== 0) {
590 * Go get the actual blocks.
592 * From this point onwards we overwrite the imap
593 * pointer that the caller gave to us.
595 error
= xfs_bmapi(tp
, ip
, map_start_fsb
, count_fsb
,
596 XFS_BMAPI_WRITE
, &first_block
, 1,
597 imap
, &nimaps
, &free_list
);
601 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
605 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
609 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
613 * See if we were able to allocate an extent that
614 * covers at least part of the callers request
616 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
617 return xfs_cmn_err_fsblock_zero(ip
, imap
);
619 if ((offset_fsb
>= imap
->br_startoff
) &&
620 (offset_fsb
< (imap
->br_startoff
+
621 imap
->br_blockcount
))) {
622 XFS_STATS_INC(xs_xstrat_quick
);
627 * So far we have not mapped the requested part of the
628 * file, just surrounding data, try again.
630 count_fsb
-= imap
->br_blockcount
;
631 map_start_fsb
= imap
->br_startoff
+ imap
->br_blockcount
;
635 xfs_bmap_cancel(&free_list
);
636 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
638 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
639 return XFS_ERROR(error
);
643 xfs_iomap_write_unwritten(
648 xfs_mount_t
*mp
= ip
->i_mount
;
649 xfs_fileoff_t offset_fsb
;
650 xfs_filblks_t count_fsb
;
651 xfs_filblks_t numblks_fsb
;
652 xfs_fsblock_t firstfsb
;
655 xfs_bmbt_irec_t imap
;
656 xfs_bmap_free_t free_list
;
661 trace_xfs_unwritten_convert(ip
, offset
, count
);
663 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
664 count_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ count
);
665 count_fsb
= (xfs_filblks_t
)(count_fsb
- offset_fsb
);
668 * Reserve enough blocks in this transaction for two complete extent
669 * btree splits. We may be converting the middle part of an unwritten
670 * extent and in this case we will insert two new extents in the btree
671 * each of which could cause a full split.
673 * This reservation amount will be used in the first call to
674 * xfs_bmbt_split() to select an AG with enough space to satisfy the
675 * rest of the operation.
677 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0) << 1;
681 * set up a transaction to convert the range of extents
682 * from unwritten to real. Do allocations in a loop until
683 * we have covered the range passed in.
685 * Note that we open code the transaction allocation here
686 * to pass KM_NOFS--we can't risk to recursing back into
687 * the filesystem here as we might be asked to write out
688 * the same inode that we complete here and might deadlock
691 xfs_wait_for_freeze(mp
, SB_FREEZE_TRANS
);
692 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_STRAT_WRITE
, KM_NOFS
);
693 tp
->t_flags
|= XFS_TRANS_RESERVE
;
694 error
= xfs_trans_reserve(tp
, resblks
,
695 XFS_WRITE_LOG_RES(mp
), 0,
696 XFS_TRANS_PERM_LOG_RES
,
697 XFS_WRITE_LOG_COUNT
);
699 xfs_trans_cancel(tp
, 0);
700 return XFS_ERROR(error
);
703 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
704 xfs_trans_ijoin(tp
, ip
);
707 * Modify the unwritten extent state of the buffer.
709 xfs_bmap_init(&free_list
, &firstfsb
);
711 error
= xfs_bmapi(tp
, ip
, offset_fsb
, count_fsb
,
712 XFS_BMAPI_WRITE
|XFS_BMAPI_CONVERT
, &firstfsb
,
713 1, &imap
, &nimaps
, &free_list
);
715 goto error_on_bmapi_transaction
;
717 error
= xfs_bmap_finish(&(tp
), &(free_list
), &committed
);
719 goto error_on_bmapi_transaction
;
721 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
722 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
724 return XFS_ERROR(error
);
726 if (!(imap
.br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
727 return xfs_cmn_err_fsblock_zero(ip
, &imap
);
729 if ((numblks_fsb
= imap
.br_blockcount
) == 0) {
731 * The numblks_fsb value should always get
732 * smaller, otherwise the loop is stuck.
734 ASSERT(imap
.br_blockcount
);
737 offset_fsb
+= numblks_fsb
;
738 count_fsb
-= numblks_fsb
;
739 } while (count_fsb
> 0);
743 error_on_bmapi_transaction
:
744 xfs_bmap_cancel(&free_list
);
745 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
));
746 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
747 return XFS_ERROR(error
);