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
20 #include "xfs_format.h"
22 #include "xfs_trans.h"
25 #include "xfs_alloc.h"
26 #include "xfs_quota.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_inode_item.h"
34 #include "xfs_btree.h"
36 #include "xfs_bmap_util.h"
37 #include "xfs_rtalloc.h"
38 #include "xfs_error.h"
39 #include "xfs_itable.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_trans_space.h"
43 #include "xfs_iomap.h"
44 #include "xfs_trace.h"
45 #include "xfs_icache.h"
46 #include "xfs_dquot_item.h"
47 #include "xfs_dquot.h"
50 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
52 #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
55 xfs_iomap_eof_align_last_fsb(
59 xfs_fileoff_t
*last_fsb
)
61 xfs_fileoff_t new_last_fsb
= 0;
62 xfs_extlen_t align
= 0;
65 if (!XFS_IS_REALTIME_INODE(ip
)) {
67 * Round up the allocation request to a stripe unit
68 * (m_dalign) boundary if the file size is >= stripe unit
69 * size, and we are allocating past the allocation eof.
71 * If mounted with the "-o swalloc" option the alignment is
72 * increased from the strip unit size to the stripe width.
74 if (mp
->m_swidth
&& (mp
->m_flags
& XFS_MOUNT_SWALLOC
))
76 else if (mp
->m_dalign
)
79 if (align
&& XFS_ISIZE(ip
) >= XFS_FSB_TO_B(mp
, align
))
80 new_last_fsb
= roundup_64(*last_fsb
, align
);
84 * Always round up the allocation request to an extent boundary
85 * (when file on a real-time subvolume or has di_extsize hint).
89 align
= roundup_64(new_last_fsb
, extsize
);
92 new_last_fsb
= roundup_64(*last_fsb
, align
);
96 error
= xfs_bmap_eof(ip
, new_last_fsb
, XFS_DATA_FORK
, &eof
);
100 *last_fsb
= new_last_fsb
;
106 xfs_alert_fsblock_zero(
108 xfs_bmbt_irec_t
*imap
)
110 xfs_alert_tag(ip
->i_mount
, XFS_PTAG_FSBLOCK_ZERO
,
111 "Access to block zero in inode %llu "
112 "start_block: %llx start_off: %llx "
113 "blkcnt: %llx extent-state: %x\n",
114 (unsigned long long)ip
->i_ino
,
115 (unsigned long long)imap
->br_startblock
,
116 (unsigned long long)imap
->br_startoff
,
117 (unsigned long long)imap
->br_blockcount
,
123 xfs_iomap_write_direct(
127 xfs_bmbt_irec_t
*imap
,
130 xfs_mount_t
*mp
= ip
->i_mount
;
131 xfs_fileoff_t offset_fsb
;
132 xfs_fileoff_t last_fsb
;
133 xfs_filblks_t count_fsb
, resaligned
;
134 xfs_fsblock_t firstfsb
;
135 xfs_extlen_t extsz
, temp
;
141 xfs_bmap_free_t free_list
;
142 uint qblocks
, resblks
, resrtextents
;
146 error
= xfs_qm_dqattach(ip
, 0);
148 return XFS_ERROR(error
);
150 rt
= XFS_IS_REALTIME_INODE(ip
);
151 extsz
= xfs_get_extsz_hint(ip
);
153 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
154 last_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)(offset
+ count
)));
155 if ((offset
+ count
) > XFS_ISIZE(ip
)) {
156 error
= xfs_iomap_eof_align_last_fsb(mp
, ip
, extsz
, &last_fsb
);
158 return XFS_ERROR(error
);
160 if (nmaps
&& (imap
->br_startblock
== HOLESTARTBLOCK
))
161 last_fsb
= MIN(last_fsb
, (xfs_fileoff_t
)
162 imap
->br_blockcount
+
165 count_fsb
= last_fsb
- offset_fsb
;
166 ASSERT(count_fsb
> 0);
168 resaligned
= count_fsb
;
169 if (unlikely(extsz
)) {
170 if ((temp
= do_mod(offset_fsb
, extsz
)))
172 if ((temp
= do_mod(resaligned
, extsz
)))
173 resaligned
+= extsz
- temp
;
177 resrtextents
= qblocks
= resaligned
;
178 resrtextents
/= mp
->m_sb
.sb_rextsize
;
179 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
180 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
183 resblks
= qblocks
= XFS_DIOSTRAT_SPACE_RES(mp
, resaligned
);
184 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
188 * Allocate and setup the transaction
190 tp
= xfs_trans_alloc(mp
, XFS_TRANS_DIOSTRAT
);
191 error
= xfs_trans_reserve(tp
, resblks
,
192 XFS_WRITE_LOG_RES(mp
), resrtextents
,
193 XFS_TRANS_PERM_LOG_RES
,
194 XFS_WRITE_LOG_COUNT
);
196 * Check for running out of space, note: need lock to return
199 xfs_trans_cancel(tp
, 0);
200 return XFS_ERROR(error
);
203 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
205 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
, 0, quota_flag
);
207 goto out_trans_cancel
;
209 xfs_trans_ijoin(tp
, ip
, 0);
212 if (offset
< XFS_ISIZE(ip
) || extsz
)
213 bmapi_flag
|= XFS_BMAPI_PREALLOC
;
216 * From this point onwards we overwrite the imap pointer that the
219 xfs_bmap_init(&free_list
, &firstfsb
);
221 error
= xfs_bmapi_write(tp
, ip
, offset_fsb
, count_fsb
, bmapi_flag
,
222 &firstfsb
, 0, imap
, &nimaps
, &free_list
);
224 goto out_bmap_cancel
;
227 * Complete the transaction
229 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
231 goto out_bmap_cancel
;
232 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
237 * Copy any maps to caller's array and return any error.
240 error
= XFS_ERROR(ENOSPC
);
244 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
245 error
= xfs_alert_fsblock_zero(ip
, imap
);
248 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
252 xfs_bmap_cancel(&free_list
);
253 xfs_trans_unreserve_quota_nblks(tp
, ip
, (long)qblocks
, 0, quota_flag
);
255 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
260 * If the caller is doing a write at the end of the file, then extend the
261 * allocation out to the file system's write iosize. We clean up any extra
262 * space left over when the file is closed in xfs_inactive().
264 * If we find we already have delalloc preallocation beyond EOF, don't do more
265 * preallocation as it it not needed.
268 xfs_iomap_eof_want_preallocate(
273 xfs_bmbt_irec_t
*imap
,
277 xfs_fileoff_t start_fsb
;
278 xfs_filblks_t count_fsb
;
279 xfs_fsblock_t firstblock
;
281 int found_delalloc
= 0;
284 if (offset
+ count
<= XFS_ISIZE(ip
))
288 * If the file is smaller than the minimum prealloc and we are using
289 * dynamic preallocation, don't do any preallocation at all as it is
290 * likely this is the only write to the file that is going to be done.
292 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
) &&
293 XFS_ISIZE(ip
) < XFS_FSB_TO_B(mp
, mp
->m_writeio_blocks
))
297 * If there are any real blocks past eof, then don't
298 * do any speculative allocation.
300 start_fsb
= XFS_B_TO_FSBT(mp
, ((xfs_ufsize_t
)(offset
+ count
- 1)));
301 count_fsb
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
302 while (count_fsb
> 0) {
304 firstblock
= NULLFSBLOCK
;
305 error
= xfs_bmapi_read(ip
, start_fsb
, count_fsb
, imap
, &imaps
,
309 for (n
= 0; n
< imaps
; n
++) {
310 if ((imap
[n
].br_startblock
!= HOLESTARTBLOCK
) &&
311 (imap
[n
].br_startblock
!= DELAYSTARTBLOCK
))
313 start_fsb
+= imap
[n
].br_blockcount
;
314 count_fsb
-= imap
[n
].br_blockcount
;
316 if (imap
[n
].br_startblock
== DELAYSTARTBLOCK
)
326 * Determine the initial size of the preallocation. We are beyond the current
327 * EOF here, but we need to take into account whether this is a sparse write or
328 * an extending write when determining the preallocation size. Hence we need to
329 * look up the extent that ends at the current write offset and use the result
330 * to determine the preallocation size.
332 * If the extent is a hole, then preallocation is essentially disabled.
333 * Otherwise we take the size of the preceeding data extent as the basis for the
334 * preallocation size. If the size of the extent is greater than half the
335 * maximum extent length, then use the current offset as the basis. This ensures
336 * that for large files the preallocation size always extends to MAXEXTLEN
337 * rather than falling short due to things like stripe unit/width alignment of
341 xfs_iomap_eof_prealloc_initial_size(
342 struct xfs_mount
*mp
,
343 struct xfs_inode
*ip
,
345 xfs_bmbt_irec_t
*imap
,
348 xfs_fileoff_t start_fsb
;
352 ASSERT(nimaps
>= imaps
);
354 /* if we are using a specific prealloc size, return now */
355 if (mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
)
358 /* If the file is small, then use the minimum prealloc */
359 if (XFS_ISIZE(ip
) < XFS_FSB_TO_B(mp
, mp
->m_dalign
))
363 * As we write multiple pages, the offset will always align to the
364 * start of a page and hence point to a hole at EOF. i.e. if the size is
365 * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096)
366 * will return FSB 1. Hence if there are blocks in the file, we want to
367 * point to the block prior to the EOF block and not the hole that maps
368 * directly at @offset.
370 start_fsb
= XFS_B_TO_FSB(mp
, offset
);
373 error
= xfs_bmapi_read(ip
, start_fsb
, 1, imap
, &imaps
, XFS_BMAPI_ENTIRE
);
378 if (imap
[0].br_startblock
== HOLESTARTBLOCK
)
380 if (imap
[0].br_blockcount
<= (MAXEXTLEN
>> 1))
381 return imap
[0].br_blockcount
<< 1;
382 return XFS_B_TO_FSB(mp
, offset
);
386 xfs_quota_need_throttle(
387 struct xfs_inode
*ip
,
389 xfs_fsblock_t alloc_blocks
)
391 struct xfs_dquot
*dq
= xfs_inode_dquot(ip
, type
);
393 if (!dq
|| !xfs_this_quota_on(ip
->i_mount
, type
))
396 /* no hi watermark, no throttle */
397 if (!dq
->q_prealloc_hi_wmark
)
400 /* under the lo watermark, no throttle */
401 if (dq
->q_res_bcount
+ alloc_blocks
< dq
->q_prealloc_lo_wmark
)
408 xfs_quota_calc_throttle(
409 struct xfs_inode
*ip
,
411 xfs_fsblock_t
*qblocks
,
416 struct xfs_dquot
*dq
= xfs_inode_dquot(ip
, type
);
418 /* over hi wmark, squash the prealloc completely */
419 if (dq
->q_res_bcount
>= dq
->q_prealloc_hi_wmark
) {
424 freesp
= dq
->q_prealloc_hi_wmark
- dq
->q_res_bcount
;
425 if (freesp
< dq
->q_low_space
[XFS_QLOWSP_5_PCNT
]) {
427 if (freesp
< dq
->q_low_space
[XFS_QLOWSP_3_PCNT
])
429 if (freesp
< dq
->q_low_space
[XFS_QLOWSP_1_PCNT
])
433 /* only overwrite the throttle values if we are more aggressive */
434 if ((freesp
>> shift
) < (*qblocks
>> *qshift
)) {
441 * If we don't have a user specified preallocation size, dynamically increase
442 * the preallocation size as the size of the file grows. Cap the maximum size
443 * at a single extent or less if the filesystem is near full. The closer the
444 * filesystem is to full, the smaller the maximum prealocation.
447 xfs_iomap_prealloc_size(
448 struct xfs_mount
*mp
,
449 struct xfs_inode
*ip
,
451 struct xfs_bmbt_irec
*imap
,
454 xfs_fsblock_t alloc_blocks
= 0;
457 xfs_fsblock_t qblocks
;
460 alloc_blocks
= xfs_iomap_eof_prealloc_initial_size(mp
, ip
, offset
,
464 qblocks
= alloc_blocks
;
467 * MAXEXTLEN is not a power of two value but we round the prealloc down
468 * to the nearest power of two value after throttling. To prevent the
469 * round down from unconditionally reducing the maximum supported prealloc
470 * size, we round up first, apply appropriate throttling, round down and
471 * cap the value to MAXEXTLEN.
473 alloc_blocks
= XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN
),
476 xfs_icsb_sync_counters(mp
, XFS_ICSB_LAZY_COUNT
);
477 freesp
= mp
->m_sb
.sb_fdblocks
;
478 if (freesp
< mp
->m_low_space
[XFS_LOWSP_5_PCNT
]) {
480 if (freesp
< mp
->m_low_space
[XFS_LOWSP_4_PCNT
])
482 if (freesp
< mp
->m_low_space
[XFS_LOWSP_3_PCNT
])
484 if (freesp
< mp
->m_low_space
[XFS_LOWSP_2_PCNT
])
486 if (freesp
< mp
->m_low_space
[XFS_LOWSP_1_PCNT
])
491 * Check each quota to cap the prealloc size and provide a shift
492 * value to throttle with.
494 if (xfs_quota_need_throttle(ip
, XFS_DQ_USER
, alloc_blocks
))
495 xfs_quota_calc_throttle(ip
, XFS_DQ_USER
, &qblocks
, &qshift
);
496 if (xfs_quota_need_throttle(ip
, XFS_DQ_GROUP
, alloc_blocks
))
497 xfs_quota_calc_throttle(ip
, XFS_DQ_GROUP
, &qblocks
, &qshift
);
498 if (xfs_quota_need_throttle(ip
, XFS_DQ_PROJ
, alloc_blocks
))
499 xfs_quota_calc_throttle(ip
, XFS_DQ_PROJ
, &qblocks
, &qshift
);
502 * The final prealloc size is set to the minimum of free space available
503 * in each of the quotas and the overall filesystem.
505 * The shift throttle value is set to the maximum value as determined by
506 * the global low free space values and per-quota low free space values.
508 alloc_blocks
= MIN(alloc_blocks
, qblocks
);
509 shift
= MAX(shift
, qshift
);
512 alloc_blocks
>>= shift
;
514 * rounddown_pow_of_two() returns an undefined result if we pass in
518 alloc_blocks
= rounddown_pow_of_two(alloc_blocks
);
519 if (alloc_blocks
> MAXEXTLEN
)
520 alloc_blocks
= MAXEXTLEN
;
523 * If we are still trying to allocate more space than is
524 * available, squash the prealloc hard. This can happen if we
525 * have a large file on a small filesystem and the above
526 * lowspace thresholds are smaller than MAXEXTLEN.
528 while (alloc_blocks
&& alloc_blocks
>= freesp
)
532 if (alloc_blocks
< mp
->m_writeio_blocks
)
533 alloc_blocks
= mp
->m_writeio_blocks
;
535 trace_xfs_iomap_prealloc_size(ip
, alloc_blocks
, shift
,
536 mp
->m_writeio_blocks
);
542 xfs_iomap_write_delay(
546 xfs_bmbt_irec_t
*ret_imap
)
548 xfs_mount_t
*mp
= ip
->i_mount
;
549 xfs_fileoff_t offset_fsb
;
550 xfs_fileoff_t last_fsb
;
551 xfs_off_t aligned_offset
;
552 xfs_fileoff_t ioalign
;
555 xfs_bmbt_irec_t imap
[XFS_WRITE_IMAPS
];
559 ASSERT(xfs_isilocked(ip
, XFS_ILOCK_EXCL
));
562 * Make sure that the dquots are there. This doesn't hold
563 * the ilock across a disk read.
565 error
= xfs_qm_dqattach_locked(ip
, 0);
567 return XFS_ERROR(error
);
569 extsz
= xfs_get_extsz_hint(ip
);
570 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
572 error
= xfs_iomap_eof_want_preallocate(mp
, ip
, offset
, count
,
573 imap
, XFS_WRITE_IMAPS
, &prealloc
);
579 xfs_fsblock_t alloc_blocks
;
581 alloc_blocks
= xfs_iomap_prealloc_size(mp
, ip
, offset
, imap
,
584 aligned_offset
= XFS_WRITEIO_ALIGN(mp
, (offset
+ count
- 1));
585 ioalign
= XFS_B_TO_FSBT(mp
, aligned_offset
);
586 last_fsb
= ioalign
+ alloc_blocks
;
588 last_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)(offset
+ count
)));
591 if (prealloc
|| extsz
) {
592 error
= xfs_iomap_eof_align_last_fsb(mp
, ip
, extsz
, &last_fsb
);
598 * Make sure preallocation does not create extents beyond the range we
599 * actually support in this filesystem.
601 if (last_fsb
> XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
))
602 last_fsb
= XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
604 ASSERT(last_fsb
> offset_fsb
);
606 nimaps
= XFS_WRITE_IMAPS
;
607 error
= xfs_bmapi_delay(ip
, offset_fsb
, last_fsb
- offset_fsb
,
608 imap
, &nimaps
, XFS_BMAPI_ENTIRE
);
615 return XFS_ERROR(error
);
619 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
620 * without EOF preallocation.
623 trace_xfs_delalloc_enospc(ip
, offset
, count
);
629 return XFS_ERROR(error
? error
: ENOSPC
);
632 if (!(imap
[0].br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
633 return xfs_alert_fsblock_zero(ip
, &imap
[0]);
636 * Tag the inode as speculatively preallocated so we can reclaim this
637 * space on demand, if necessary.
640 xfs_inode_set_eofblocks_tag(ip
);
647 * Pass in a delayed allocate extent, convert it to real extents;
648 * return to the caller the extent we create which maps on top of
649 * the originating callers request.
651 * Called without a lock on the inode.
653 * We no longer bother to look at the incoming map - all we have to
654 * guarantee is that whatever we allocate fills the required range.
657 xfs_iomap_write_allocate(
661 xfs_bmbt_irec_t
*imap
)
663 xfs_mount_t
*mp
= ip
->i_mount
;
664 xfs_fileoff_t offset_fsb
, last_block
;
665 xfs_fileoff_t end_fsb
, map_start_fsb
;
666 xfs_fsblock_t first_block
;
667 xfs_bmap_free_t free_list
;
668 xfs_filblks_t count_fsb
;
670 int nimaps
, committed
;
675 * Make sure that the dquots are there.
677 error
= xfs_qm_dqattach(ip
, 0);
679 return XFS_ERROR(error
);
681 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
682 count_fsb
= imap
->br_blockcount
;
683 map_start_fsb
= imap
->br_startoff
;
685 XFS_STATS_ADD(xs_xstrat_bytes
, XFS_FSB_TO_B(mp
, count_fsb
));
687 while (count_fsb
!= 0) {
689 * Set up a transaction with which to allocate the
690 * backing store for the file. Do allocations in a
691 * loop until we get some space in the range we are
692 * interested in. The other space that might be allocated
693 * is in the delayed allocation extent on which we sit
694 * but before our buffer starts.
698 while (nimaps
== 0) {
699 tp
= xfs_trans_alloc(mp
, XFS_TRANS_STRAT_WRITE
);
700 tp
->t_flags
|= XFS_TRANS_RESERVE
;
701 nres
= XFS_EXTENTADD_SPACE_RES(mp
, XFS_DATA_FORK
);
702 error
= xfs_trans_reserve(tp
, nres
,
703 XFS_WRITE_LOG_RES(mp
),
704 0, XFS_TRANS_PERM_LOG_RES
,
705 XFS_WRITE_LOG_COUNT
);
707 xfs_trans_cancel(tp
, 0);
708 return XFS_ERROR(error
);
710 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
711 xfs_trans_ijoin(tp
, ip
, 0);
713 xfs_bmap_init(&free_list
, &first_block
);
716 * it is possible that the extents have changed since
717 * we did the read call as we dropped the ilock for a
718 * while. We have to be careful about truncates or hole
719 * punchs here - we are not allowed to allocate
720 * non-delalloc blocks here.
722 * The only protection against truncation is the pages
723 * for the range we are being asked to convert are
724 * locked and hence a truncate will block on them
727 * As a result, if we go beyond the range we really
728 * need and hit an delalloc extent boundary followed by
729 * a hole while we have excess blocks in the map, we
730 * will fill the hole incorrectly and overrun the
731 * transaction reservation.
733 * Using a single map prevents this as we are forced to
734 * check each map we look for overlap with the desired
735 * range and abort as soon as we find it. Also, given
736 * that we only return a single map, having one beyond
737 * what we can return is probably a bit silly.
739 * We also need to check that we don't go beyond EOF;
740 * this is a truncate optimisation as a truncate sets
741 * the new file size before block on the pages we
742 * currently have locked under writeback. Because they
743 * are about to be tossed, we don't need to write them
747 end_fsb
= XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
));
748 error
= xfs_bmap_last_offset(NULL
, ip
, &last_block
,
753 last_block
= XFS_FILEOFF_MAX(last_block
, end_fsb
);
754 if ((map_start_fsb
+ count_fsb
) > last_block
) {
755 count_fsb
= last_block
- map_start_fsb
;
756 if (count_fsb
== 0) {
763 * From this point onwards we overwrite the imap
764 * pointer that the caller gave to us.
766 error
= xfs_bmapi_write(tp
, ip
, map_start_fsb
,
768 XFS_BMAPI_STACK_SWITCH
,
770 imap
, &nimaps
, &free_list
);
774 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
778 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
782 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
786 * See if we were able to allocate an extent that
787 * covers at least part of the callers request
789 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
790 return xfs_alert_fsblock_zero(ip
, imap
);
792 if ((offset_fsb
>= imap
->br_startoff
) &&
793 (offset_fsb
< (imap
->br_startoff
+
794 imap
->br_blockcount
))) {
795 XFS_STATS_INC(xs_xstrat_quick
);
800 * So far we have not mapped the requested part of the
801 * file, just surrounding data, try again.
803 count_fsb
-= imap
->br_blockcount
;
804 map_start_fsb
= imap
->br_startoff
+ imap
->br_blockcount
;
808 xfs_bmap_cancel(&free_list
);
809 xfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
);
811 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
812 return XFS_ERROR(error
);
816 xfs_iomap_write_unwritten(
821 xfs_mount_t
*mp
= ip
->i_mount
;
822 xfs_fileoff_t offset_fsb
;
823 xfs_filblks_t count_fsb
;
824 xfs_filblks_t numblks_fsb
;
825 xfs_fsblock_t firstfsb
;
828 xfs_bmbt_irec_t imap
;
829 xfs_bmap_free_t free_list
;
835 trace_xfs_unwritten_convert(ip
, offset
, count
);
837 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
838 count_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ count
);
839 count_fsb
= (xfs_filblks_t
)(count_fsb
- offset_fsb
);
842 * Reserve enough blocks in this transaction for two complete extent
843 * btree splits. We may be converting the middle part of an unwritten
844 * extent and in this case we will insert two new extents in the btree
845 * each of which could cause a full split.
847 * This reservation amount will be used in the first call to
848 * xfs_bmbt_split() to select an AG with enough space to satisfy the
849 * rest of the operation.
851 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0) << 1;
855 * set up a transaction to convert the range of extents
856 * from unwritten to real. Do allocations in a loop until
857 * we have covered the range passed in.
859 * Note that we open code the transaction allocation here
860 * to pass KM_NOFS--we can't risk to recursing back into
861 * the filesystem here as we might be asked to write out
862 * the same inode that we complete here and might deadlock
865 sb_start_intwrite(mp
->m_super
);
866 tp
= _xfs_trans_alloc(mp
, XFS_TRANS_STRAT_WRITE
, KM_NOFS
);
867 tp
->t_flags
|= XFS_TRANS_RESERVE
| XFS_TRANS_FREEZE_PROT
;
868 error
= xfs_trans_reserve(tp
, resblks
,
869 XFS_WRITE_LOG_RES(mp
), 0,
870 XFS_TRANS_PERM_LOG_RES
,
871 XFS_WRITE_LOG_COUNT
);
873 xfs_trans_cancel(tp
, 0);
874 return XFS_ERROR(error
);
877 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
878 xfs_trans_ijoin(tp
, ip
, 0);
881 * Modify the unwritten extent state of the buffer.
883 xfs_bmap_init(&free_list
, &firstfsb
);
885 error
= xfs_bmapi_write(tp
, ip
, offset_fsb
, count_fsb
,
886 XFS_BMAPI_CONVERT
, &firstfsb
,
887 1, &imap
, &nimaps
, &free_list
);
889 goto error_on_bmapi_transaction
;
892 * Log the updated inode size as we go. We have to be careful
893 * to only log it up to the actual write offset if it is
894 * halfway into a block.
896 i_size
= XFS_FSB_TO_B(mp
, offset_fsb
+ count_fsb
);
897 if (i_size
> offset
+ count
)
898 i_size
= offset
+ count
;
900 i_size
= xfs_new_eof(ip
, i_size
);
902 ip
->i_d
.di_size
= i_size
;
903 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
906 error
= xfs_bmap_finish(&tp
, &free_list
, &committed
);
908 goto error_on_bmapi_transaction
;
910 error
= xfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
);
911 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
913 return XFS_ERROR(error
);
915 if (!(imap
.br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
916 return xfs_alert_fsblock_zero(ip
, &imap
);
918 if ((numblks_fsb
= imap
.br_blockcount
) == 0) {
920 * The numblks_fsb value should always get
921 * smaller, otherwise the loop is stuck.
923 ASSERT(imap
.br_blockcount
);
926 offset_fsb
+= numblks_fsb
;
927 count_fsb
-= numblks_fsb
;
928 } while (count_fsb
> 0);
932 error_on_bmapi_transaction
:
933 xfs_bmap_cancel(&free_list
);
934 xfs_trans_cancel(tp
, (XFS_TRANS_RELEASE_LOG_RES
| XFS_TRANS_ABORT
));
935 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
936 return XFS_ERROR(error
);