2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * Copyright (c) 2016 Christoph Hellwig.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 #include <linux/iomap.h>
22 #include "xfs_shared.h"
23 #include "xfs_format.h"
24 #include "xfs_log_format.h"
25 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_inode.h"
29 #include "xfs_btree.h"
30 #include "xfs_bmap_btree.h"
32 #include "xfs_bmap_util.h"
33 #include "xfs_error.h"
34 #include "xfs_trans.h"
35 #include "xfs_trans_space.h"
36 #include "xfs_iomap.h"
37 #include "xfs_trace.h"
38 #include "xfs_icache.h"
39 #include "xfs_quota.h"
40 #include "xfs_dquot_item.h"
41 #include "xfs_dquot.h"
42 #include "xfs_reflink.h"
45 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
52 struct xfs_bmbt_irec
*imap
)
54 struct xfs_mount
*mp
= ip
->i_mount
;
56 if (imap
->br_startblock
== HOLESTARTBLOCK
) {
57 iomap
->blkno
= IOMAP_NULL_BLOCK
;
58 iomap
->type
= IOMAP_HOLE
;
59 } else if (imap
->br_startblock
== DELAYSTARTBLOCK
) {
60 iomap
->blkno
= IOMAP_NULL_BLOCK
;
61 iomap
->type
= IOMAP_DELALLOC
;
63 iomap
->blkno
= xfs_fsb_to_db(ip
, imap
->br_startblock
);
64 if (imap
->br_state
== XFS_EXT_UNWRITTEN
)
65 iomap
->type
= IOMAP_UNWRITTEN
;
67 iomap
->type
= IOMAP_MAPPED
;
69 iomap
->offset
= XFS_FSB_TO_B(mp
, imap
->br_startoff
);
70 iomap
->length
= XFS_FSB_TO_B(mp
, imap
->br_blockcount
);
71 iomap
->bdev
= xfs_find_bdev_for_inode(VFS_I(ip
));
79 struct xfs_mount
*mp
= ip
->i_mount
;
80 xfs_extlen_t align
= 0;
82 if (!XFS_IS_REALTIME_INODE(ip
)) {
84 * Round up the allocation request to a stripe unit
85 * (m_dalign) boundary if the file size is >= stripe unit
86 * size, and we are allocating past the allocation eof.
88 * If mounted with the "-o swalloc" option the alignment is
89 * increased from the strip unit size to the stripe width.
91 if (mp
->m_swidth
&& (mp
->m_flags
& XFS_MOUNT_SWALLOC
))
93 else if (mp
->m_dalign
)
96 if (align
&& XFS_ISIZE(ip
) < XFS_FSB_TO_B(mp
, align
))
101 * Always round up the allocation request to an extent boundary
102 * (when file on a real-time subvolume or has di_extsize hint).
106 align
= roundup_64(align
, extsize
);
115 xfs_iomap_eof_align_last_fsb(
116 struct xfs_inode
*ip
,
117 xfs_extlen_t extsize
,
118 xfs_fileoff_t
*last_fsb
)
120 xfs_extlen_t align
= xfs_eof_alignment(ip
, extsize
);
123 xfs_fileoff_t new_last_fsb
= roundup_64(*last_fsb
, align
);
126 error
= xfs_bmap_eof(ip
, new_last_fsb
, XFS_DATA_FORK
, &eof
);
130 *last_fsb
= new_last_fsb
;
136 xfs_alert_fsblock_zero(
138 xfs_bmbt_irec_t
*imap
)
140 xfs_alert_tag(ip
->i_mount
, XFS_PTAG_FSBLOCK_ZERO
,
141 "Access to block zero in inode %llu "
142 "start_block: %llx start_off: %llx "
143 "blkcnt: %llx extent-state: %x",
144 (unsigned long long)ip
->i_ino
,
145 (unsigned long long)imap
->br_startblock
,
146 (unsigned long long)imap
->br_startoff
,
147 (unsigned long long)imap
->br_blockcount
,
149 return -EFSCORRUPTED
;
153 xfs_iomap_write_direct(
157 xfs_bmbt_irec_t
*imap
,
160 xfs_mount_t
*mp
= ip
->i_mount
;
161 xfs_fileoff_t offset_fsb
;
162 xfs_fileoff_t last_fsb
;
163 xfs_filblks_t count_fsb
, resaligned
;
164 xfs_fsblock_t firstfsb
;
170 struct xfs_defer_ops dfops
;
171 uint qblocks
, resblks
, resrtextents
;
174 int bmapi_flags
= XFS_BMAPI_PREALLOC
;
177 rt
= XFS_IS_REALTIME_INODE(ip
);
178 extsz
= xfs_get_extsz_hint(ip
);
179 lockmode
= XFS_ILOCK_SHARED
; /* locked by caller */
181 ASSERT(xfs_isilocked(ip
, lockmode
));
183 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
184 last_fsb
= XFS_B_TO_FSB(mp
, ((xfs_ufsize_t
)(offset
+ count
)));
185 if ((offset
+ count
) > XFS_ISIZE(ip
)) {
187 * Assert that the in-core extent list is present since this can
188 * call xfs_iread_extents() and we only have the ilock shared.
189 * This should be safe because the lock was held around a bmapi
190 * call in the caller and we only need it to access the in-core
193 ASSERT(XFS_IFORK_PTR(ip
, XFS_DATA_FORK
)->if_flags
&
195 error
= xfs_iomap_eof_align_last_fsb(ip
, extsz
, &last_fsb
);
199 if (nmaps
&& (imap
->br_startblock
== HOLESTARTBLOCK
))
200 last_fsb
= MIN(last_fsb
, (xfs_fileoff_t
)
201 imap
->br_blockcount
+
204 count_fsb
= last_fsb
- offset_fsb
;
205 ASSERT(count_fsb
> 0);
206 resaligned
= xfs_aligned_fsb_count(offset_fsb
, count_fsb
, extsz
);
209 resrtextents
= qblocks
= resaligned
;
210 resrtextents
/= mp
->m_sb
.sb_rextsize
;
211 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0);
212 quota_flag
= XFS_QMOPT_RES_RTBLKS
;
215 resblks
= qblocks
= XFS_DIOSTRAT_SPACE_RES(mp
, resaligned
);
216 quota_flag
= XFS_QMOPT_RES_REGBLKS
;
220 * Drop the shared lock acquired by the caller, attach the dquot if
221 * necessary and move on to transaction setup.
223 xfs_iunlock(ip
, lockmode
);
224 error
= xfs_qm_dqattach(ip
, 0);
229 * For DAX, we do not allocate unwritten extents, but instead we zero
230 * the block before we commit the transaction. Ideally we'd like to do
231 * this outside the transaction context, but if we commit and then crash
232 * we may not have zeroed the blocks and this will be exposed on
233 * recovery of the allocation. Hence we must zero before commit.
235 * Further, if we are mapping unwritten extents here, we need to zero
236 * and convert them to written so that we don't need an unwritten extent
237 * callback for DAX. This also means that we need to be able to dip into
238 * the reserve block pool for bmbt block allocation if there is no space
239 * left but we need to do unwritten extent conversion.
241 if (IS_DAX(VFS_I(ip
))) {
242 bmapi_flags
= XFS_BMAPI_CONVERT
| XFS_BMAPI_ZERO
;
243 if (imap
->br_state
== XFS_EXT_UNWRITTEN
) {
244 tflags
|= XFS_TRANS_RESERVE
;
245 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0) << 1;
248 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
, resblks
, resrtextents
,
253 lockmode
= XFS_ILOCK_EXCL
;
254 xfs_ilock(ip
, lockmode
);
256 error
= xfs_trans_reserve_quota_nblks(tp
, ip
, qblocks
, 0, quota_flag
);
258 goto out_trans_cancel
;
260 xfs_trans_ijoin(tp
, ip
, 0);
263 * From this point onwards we overwrite the imap pointer that the
266 xfs_defer_init(&dfops
, &firstfsb
);
268 error
= xfs_bmapi_write(tp
, ip
, offset_fsb
, count_fsb
,
269 bmapi_flags
, &firstfsb
, resblks
, imap
,
272 goto out_bmap_cancel
;
275 * Complete the transaction
277 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
279 goto out_bmap_cancel
;
281 error
= xfs_trans_commit(tp
);
286 * Copy any maps to caller's array and return any error.
293 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
294 error
= xfs_alert_fsblock_zero(ip
, imap
);
297 xfs_iunlock(ip
, lockmode
);
301 xfs_defer_cancel(&dfops
);
302 xfs_trans_unreserve_quota_nblks(tp
, ip
, (long)qblocks
, 0, quota_flag
);
304 xfs_trans_cancel(tp
);
309 xfs_quota_need_throttle(
310 struct xfs_inode
*ip
,
312 xfs_fsblock_t alloc_blocks
)
314 struct xfs_dquot
*dq
= xfs_inode_dquot(ip
, type
);
316 if (!dq
|| !xfs_this_quota_on(ip
->i_mount
, type
))
319 /* no hi watermark, no throttle */
320 if (!dq
->q_prealloc_hi_wmark
)
323 /* under the lo watermark, no throttle */
324 if (dq
->q_res_bcount
+ alloc_blocks
< dq
->q_prealloc_lo_wmark
)
331 xfs_quota_calc_throttle(
332 struct xfs_inode
*ip
,
334 xfs_fsblock_t
*qblocks
,
340 struct xfs_dquot
*dq
= xfs_inode_dquot(ip
, type
);
342 /* no dq, or over hi wmark, squash the prealloc completely */
343 if (!dq
|| dq
->q_res_bcount
>= dq
->q_prealloc_hi_wmark
) {
349 freesp
= dq
->q_prealloc_hi_wmark
- dq
->q_res_bcount
;
350 if (freesp
< dq
->q_low_space
[XFS_QLOWSP_5_PCNT
]) {
352 if (freesp
< dq
->q_low_space
[XFS_QLOWSP_3_PCNT
])
354 if (freesp
< dq
->q_low_space
[XFS_QLOWSP_1_PCNT
])
358 if (freesp
< *qfreesp
)
361 /* only overwrite the throttle values if we are more aggressive */
362 if ((freesp
>> shift
) < (*qblocks
>> *qshift
)) {
369 * If we are doing a write at the end of the file and there are no allocations
370 * past this one, then extend the allocation out to the file system's write
373 * If we don't have a user specified preallocation size, dynamically increase
374 * the preallocation size as the size of the file grows. Cap the maximum size
375 * at a single extent or less if the filesystem is near full. The closer the
376 * filesystem is to full, the smaller the maximum prealocation.
378 * As an exception we don't do any preallocation at all if the file is smaller
379 * than the minimum preallocation and we are using the default dynamic
380 * preallocation scheme, as it is likely this is the only write to the file that
381 * is going to be done.
383 * We clean up any extra space left over when the file is closed in
387 xfs_iomap_prealloc_size(
388 struct xfs_inode
*ip
,
393 struct xfs_mount
*mp
= ip
->i_mount
;
394 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
395 xfs_fileoff_t offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
396 struct xfs_bmbt_irec prev
;
399 xfs_fsblock_t qblocks
;
401 xfs_fsblock_t alloc_blocks
= 0;
403 if (offset
+ count
<= XFS_ISIZE(ip
))
406 if (!(mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
) &&
407 (XFS_ISIZE(ip
) < XFS_FSB_TO_B(mp
, mp
->m_writeio_blocks
)))
411 * If an explicit allocsize is set, the file is small, or we
412 * are writing behind a hole, then use the minimum prealloc:
414 if ((mp
->m_flags
& XFS_MOUNT_DFLT_IOSIZE
) ||
415 XFS_ISIZE(ip
) < XFS_FSB_TO_B(mp
, mp
->m_dalign
) ||
416 !xfs_iext_get_extent(ifp
, idx
- 1, &prev
) ||
417 prev
.br_startoff
+ prev
.br_blockcount
< offset_fsb
)
418 return mp
->m_writeio_blocks
;
421 * Determine the initial size of the preallocation. We are beyond the
422 * current EOF here, but we need to take into account whether this is
423 * a sparse write or an extending write when determining the
424 * preallocation size. Hence we need to look up the extent that ends
425 * at the current write offset and use the result to determine the
426 * preallocation size.
428 * If the extent is a hole, then preallocation is essentially disabled.
429 * Otherwise we take the size of the preceding data extent as the basis
430 * for the preallocation size. If the size of the extent is greater than
431 * half the maximum extent length, then use the current offset as the
432 * basis. This ensures that for large files the preallocation size
433 * always extends to MAXEXTLEN rather than falling short due to things
434 * like stripe unit/width alignment of real extents.
436 if (prev
.br_blockcount
<= (MAXEXTLEN
>> 1))
437 alloc_blocks
= prev
.br_blockcount
<< 1;
439 alloc_blocks
= XFS_B_TO_FSB(mp
, offset
);
442 qblocks
= alloc_blocks
;
445 * MAXEXTLEN is not a power of two value but we round the prealloc down
446 * to the nearest power of two value after throttling. To prevent the
447 * round down from unconditionally reducing the maximum supported prealloc
448 * size, we round up first, apply appropriate throttling, round down and
449 * cap the value to MAXEXTLEN.
451 alloc_blocks
= XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN
),
454 freesp
= percpu_counter_read_positive(&mp
->m_fdblocks
);
455 if (freesp
< mp
->m_low_space
[XFS_LOWSP_5_PCNT
]) {
457 if (freesp
< mp
->m_low_space
[XFS_LOWSP_4_PCNT
])
459 if (freesp
< mp
->m_low_space
[XFS_LOWSP_3_PCNT
])
461 if (freesp
< mp
->m_low_space
[XFS_LOWSP_2_PCNT
])
463 if (freesp
< mp
->m_low_space
[XFS_LOWSP_1_PCNT
])
468 * Check each quota to cap the prealloc size, provide a shift value to
469 * throttle with and adjust amount of available space.
471 if (xfs_quota_need_throttle(ip
, XFS_DQ_USER
, alloc_blocks
))
472 xfs_quota_calc_throttle(ip
, XFS_DQ_USER
, &qblocks
, &qshift
,
474 if (xfs_quota_need_throttle(ip
, XFS_DQ_GROUP
, alloc_blocks
))
475 xfs_quota_calc_throttle(ip
, XFS_DQ_GROUP
, &qblocks
, &qshift
,
477 if (xfs_quota_need_throttle(ip
, XFS_DQ_PROJ
, alloc_blocks
))
478 xfs_quota_calc_throttle(ip
, XFS_DQ_PROJ
, &qblocks
, &qshift
,
482 * The final prealloc size is set to the minimum of free space available
483 * in each of the quotas and the overall filesystem.
485 * The shift throttle value is set to the maximum value as determined by
486 * the global low free space values and per-quota low free space values.
488 alloc_blocks
= MIN(alloc_blocks
, qblocks
);
489 shift
= MAX(shift
, qshift
);
492 alloc_blocks
>>= shift
;
494 * rounddown_pow_of_two() returns an undefined result if we pass in
498 alloc_blocks
= rounddown_pow_of_two(alloc_blocks
);
499 if (alloc_blocks
> MAXEXTLEN
)
500 alloc_blocks
= MAXEXTLEN
;
503 * If we are still trying to allocate more space than is
504 * available, squash the prealloc hard. This can happen if we
505 * have a large file on a small filesystem and the above
506 * lowspace thresholds are smaller than MAXEXTLEN.
508 while (alloc_blocks
&& alloc_blocks
>= freesp
)
511 if (alloc_blocks
< mp
->m_writeio_blocks
)
512 alloc_blocks
= mp
->m_writeio_blocks
;
513 trace_xfs_iomap_prealloc_size(ip
, alloc_blocks
, shift
,
514 mp
->m_writeio_blocks
);
519 xfs_file_iomap_begin_delay(
526 struct xfs_inode
*ip
= XFS_I(inode
);
527 struct xfs_mount
*mp
= ip
->i_mount
;
528 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, XFS_DATA_FORK
);
529 xfs_fileoff_t offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
530 xfs_fileoff_t maxbytes_fsb
=
531 XFS_B_TO_FSB(mp
, mp
->m_super
->s_maxbytes
);
532 xfs_fileoff_t end_fsb
;
533 int error
= 0, eof
= 0;
534 struct xfs_bmbt_irec got
;
536 xfs_fsblock_t prealloc_blocks
= 0;
538 ASSERT(!XFS_IS_REALTIME_INODE(ip
));
539 ASSERT(!xfs_get_extsz_hint(ip
));
541 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
543 if (unlikely(XFS_TEST_ERROR(
544 (XFS_IFORK_FORMAT(ip
, XFS_DATA_FORK
) != XFS_DINODE_FMT_EXTENTS
&&
545 XFS_IFORK_FORMAT(ip
, XFS_DATA_FORK
) != XFS_DINODE_FMT_BTREE
),
546 mp
, XFS_ERRTAG_BMAPIFORMAT
, XFS_RANDOM_BMAPIFORMAT
))) {
547 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_LOW
, mp
);
548 error
= -EFSCORRUPTED
;
552 XFS_STATS_INC(mp
, xs_blk_mapw
);
554 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
555 error
= xfs_iread_extents(NULL
, ip
, XFS_DATA_FORK
);
560 eof
= !xfs_iext_lookup_extent(ip
, ifp
, offset_fsb
, &idx
, &got
);
561 if (!eof
&& got
.br_startoff
<= offset_fsb
) {
562 if (xfs_is_reflink_inode(ip
)) {
565 end_fsb
= min(XFS_B_TO_FSB(mp
, offset
+ count
),
567 xfs_trim_extent(&got
, offset_fsb
, end_fsb
- offset_fsb
);
568 error
= xfs_reflink_reserve_cow(ip
, &got
, &shared
);
573 trace_xfs_iomap_found(ip
, offset
, count
, 0, &got
);
577 error
= xfs_qm_dqattach_locked(ip
, 0);
582 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages
583 * to keep the chunks of work done where somewhat symmetric with the
584 * work writeback does. This is a completely arbitrary number pulled
585 * out of thin air as a best guess for initial testing.
587 * Note that the values needs to be less than 32-bits wide until
588 * the lower level functions are updated.
590 count
= min_t(loff_t
, count
, 1024 * PAGE_SIZE
);
591 end_fsb
= min(XFS_B_TO_FSB(mp
, offset
+ count
), maxbytes_fsb
);
594 prealloc_blocks
= xfs_iomap_prealloc_size(ip
, offset
, count
, idx
);
595 if (prealloc_blocks
) {
597 xfs_off_t end_offset
;
598 xfs_fileoff_t p_end_fsb
;
600 end_offset
= XFS_WRITEIO_ALIGN(mp
, offset
+ count
- 1);
601 p_end_fsb
= XFS_B_TO_FSBT(mp
, end_offset
) +
604 align
= xfs_eof_alignment(ip
, 0);
606 p_end_fsb
= roundup_64(p_end_fsb
, align
);
608 p_end_fsb
= min(p_end_fsb
, maxbytes_fsb
);
609 ASSERT(p_end_fsb
> offset_fsb
);
610 prealloc_blocks
= p_end_fsb
- end_fsb
;
615 error
= xfs_bmapi_reserve_delalloc(ip
, XFS_DATA_FORK
, offset_fsb
,
616 end_fsb
- offset_fsb
, prealloc_blocks
, &got
, &idx
, eof
);
622 /* retry without any preallocation */
623 trace_xfs_delalloc_enospc(ip
, offset
, count
);
624 if (prealloc_blocks
) {
634 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
635 * them out if the write happens to fail.
637 iomap
->flags
= IOMAP_F_NEW
;
638 trace_xfs_iomap_alloc(ip
, offset
, count
, 0, &got
);
640 if (isnullstartblock(got
.br_startblock
))
641 got
.br_startblock
= DELAYSTARTBLOCK
;
643 if (!got
.br_startblock
) {
644 error
= xfs_alert_fsblock_zero(ip
, &got
);
649 xfs_bmbt_to_iomap(ip
, iomap
, &got
);
652 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
657 * Pass in a delayed allocate extent, convert it to real extents;
658 * return to the caller the extent we create which maps on top of
659 * the originating callers request.
661 * Called without a lock on the inode.
663 * We no longer bother to look at the incoming map - all we have to
664 * guarantee is that whatever we allocate fills the required range.
667 xfs_iomap_write_allocate(
671 xfs_bmbt_irec_t
*imap
)
673 xfs_mount_t
*mp
= ip
->i_mount
;
674 xfs_fileoff_t offset_fsb
, last_block
;
675 xfs_fileoff_t end_fsb
, map_start_fsb
;
676 xfs_fsblock_t first_block
;
677 struct xfs_defer_ops dfops
;
678 xfs_filblks_t count_fsb
;
682 int flags
= XFS_BMAPI_DELALLOC
;
685 if (whichfork
== XFS_COW_FORK
)
686 flags
|= XFS_BMAPI_COWFORK
| XFS_BMAPI_PREALLOC
;
689 * Make sure that the dquots are there.
691 error
= xfs_qm_dqattach(ip
, 0);
695 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
696 count_fsb
= imap
->br_blockcount
;
697 map_start_fsb
= imap
->br_startoff
;
699 XFS_STATS_ADD(mp
, xs_xstrat_bytes
, XFS_FSB_TO_B(mp
, count_fsb
));
701 while (count_fsb
!= 0) {
703 * Set up a transaction with which to allocate the
704 * backing store for the file. Do allocations in a
705 * loop until we get some space in the range we are
706 * interested in. The other space that might be allocated
707 * is in the delayed allocation extent on which we sit
708 * but before our buffer starts.
711 while (nimaps
== 0) {
712 nres
= XFS_EXTENTADD_SPACE_RES(mp
, XFS_DATA_FORK
);
714 * We have already reserved space for the extent and any
715 * indirect blocks when creating the delalloc extent,
716 * there is no need to reserve space in this transaction
719 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
, 0,
720 0, XFS_TRANS_RESERVE
, &tp
);
724 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
725 xfs_trans_ijoin(tp
, ip
, 0);
727 xfs_defer_init(&dfops
, &first_block
);
730 * it is possible that the extents have changed since
731 * we did the read call as we dropped the ilock for a
732 * while. We have to be careful about truncates or hole
733 * punchs here - we are not allowed to allocate
734 * non-delalloc blocks here.
736 * The only protection against truncation is the pages
737 * for the range we are being asked to convert are
738 * locked and hence a truncate will block on them
741 * As a result, if we go beyond the range we really
742 * need and hit an delalloc extent boundary followed by
743 * a hole while we have excess blocks in the map, we
744 * will fill the hole incorrectly and overrun the
745 * transaction reservation.
747 * Using a single map prevents this as we are forced to
748 * check each map we look for overlap with the desired
749 * range and abort as soon as we find it. Also, given
750 * that we only return a single map, having one beyond
751 * what we can return is probably a bit silly.
753 * We also need to check that we don't go beyond EOF;
754 * this is a truncate optimisation as a truncate sets
755 * the new file size before block on the pages we
756 * currently have locked under writeback. Because they
757 * are about to be tossed, we don't need to write them
761 end_fsb
= XFS_B_TO_FSB(mp
, XFS_ISIZE(ip
));
762 error
= xfs_bmap_last_offset(ip
, &last_block
,
767 last_block
= XFS_FILEOFF_MAX(last_block
, end_fsb
);
768 if ((map_start_fsb
+ count_fsb
) > last_block
) {
769 count_fsb
= last_block
- map_start_fsb
;
770 if (count_fsb
== 0) {
777 * From this point onwards we overwrite the imap
778 * pointer that the caller gave to us.
780 error
= xfs_bmapi_write(tp
, ip
, map_start_fsb
,
781 count_fsb
, flags
, &first_block
,
787 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
791 error
= xfs_trans_commit(tp
);
795 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
799 * See if we were able to allocate an extent that
800 * covers at least part of the callers request
802 if (!(imap
->br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
803 return xfs_alert_fsblock_zero(ip
, imap
);
805 if ((offset_fsb
>= imap
->br_startoff
) &&
806 (offset_fsb
< (imap
->br_startoff
+
807 imap
->br_blockcount
))) {
808 XFS_STATS_INC(mp
, xs_xstrat_quick
);
813 * So far we have not mapped the requested part of the
814 * file, just surrounding data, try again.
816 count_fsb
-= imap
->br_blockcount
;
817 map_start_fsb
= imap
->br_startoff
+ imap
->br_blockcount
;
821 xfs_defer_cancel(&dfops
);
822 xfs_trans_cancel(tp
);
824 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
829 xfs_iomap_write_unwritten(
834 xfs_mount_t
*mp
= ip
->i_mount
;
835 xfs_fileoff_t offset_fsb
;
836 xfs_filblks_t count_fsb
;
837 xfs_filblks_t numblks_fsb
;
838 xfs_fsblock_t firstfsb
;
841 xfs_bmbt_irec_t imap
;
842 struct xfs_defer_ops dfops
;
847 trace_xfs_unwritten_convert(ip
, offset
, count
);
849 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
850 count_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ count
);
851 count_fsb
= (xfs_filblks_t
)(count_fsb
- offset_fsb
);
854 * Reserve enough blocks in this transaction for two complete extent
855 * btree splits. We may be converting the middle part of an unwritten
856 * extent and in this case we will insert two new extents in the btree
857 * each of which could cause a full split.
859 * This reservation amount will be used in the first call to
860 * xfs_bmbt_split() to select an AG with enough space to satisfy the
861 * rest of the operation.
863 resblks
= XFS_DIOSTRAT_SPACE_RES(mp
, 0) << 1;
867 * Set up a transaction to convert the range of extents
868 * from unwritten to real. Do allocations in a loop until
869 * we have covered the range passed in.
871 * Note that we can't risk to recursing back into the filesystem
872 * here as we might be asked to write out the same inode that we
873 * complete here and might deadlock on the iolock.
875 error
= xfs_trans_alloc(mp
, &M_RES(mp
)->tr_write
, resblks
, 0,
876 XFS_TRANS_RESERVE
| XFS_TRANS_NOFS
, &tp
);
880 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
881 xfs_trans_ijoin(tp
, ip
, 0);
884 * Modify the unwritten extent state of the buffer.
886 xfs_defer_init(&dfops
, &firstfsb
);
888 error
= xfs_bmapi_write(tp
, ip
, offset_fsb
, count_fsb
,
889 XFS_BMAPI_CONVERT
, &firstfsb
, resblks
,
890 &imap
, &nimaps
, &dfops
);
892 goto error_on_bmapi_transaction
;
895 * Log the updated inode size as we go. We have to be careful
896 * to only log it up to the actual write offset if it is
897 * halfway into a block.
899 i_size
= XFS_FSB_TO_B(mp
, offset_fsb
+ count_fsb
);
900 if (i_size
> offset
+ count
)
901 i_size
= offset
+ count
;
903 i_size
= xfs_new_eof(ip
, i_size
);
905 ip
->i_d
.di_size
= i_size
;
906 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
909 error
= xfs_defer_finish(&tp
, &dfops
, NULL
);
911 goto error_on_bmapi_transaction
;
913 error
= xfs_trans_commit(tp
);
914 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
918 if (!(imap
.br_startblock
|| XFS_IS_REALTIME_INODE(ip
)))
919 return xfs_alert_fsblock_zero(ip
, &imap
);
921 if ((numblks_fsb
= imap
.br_blockcount
) == 0) {
923 * The numblks_fsb value should always get
924 * smaller, otherwise the loop is stuck.
926 ASSERT(imap
.br_blockcount
);
929 offset_fsb
+= numblks_fsb
;
930 count_fsb
-= numblks_fsb
;
931 } while (count_fsb
> 0);
935 error_on_bmapi_transaction
:
936 xfs_defer_cancel(&dfops
);
937 xfs_trans_cancel(tp
);
938 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
942 static inline bool imap_needs_alloc(struct inode
*inode
,
943 struct xfs_bmbt_irec
*imap
, int nimaps
)
946 imap
->br_startblock
== HOLESTARTBLOCK
||
947 imap
->br_startblock
== DELAYSTARTBLOCK
||
948 (IS_DAX(inode
) && imap
->br_state
== XFS_EXT_UNWRITTEN
);
951 static inline bool need_excl_ilock(struct xfs_inode
*ip
, unsigned flags
)
954 * COW writes will allocate delalloc space, so we need to make sure
955 * to take the lock exclusively here.
957 if (xfs_is_reflink_inode(ip
) && (flags
& (IOMAP_WRITE
| IOMAP_ZERO
)))
959 if ((flags
& IOMAP_DIRECT
) && (flags
& IOMAP_WRITE
))
965 xfs_file_iomap_begin(
972 struct xfs_inode
*ip
= XFS_I(inode
);
973 struct xfs_mount
*mp
= ip
->i_mount
;
974 struct xfs_bmbt_irec imap
;
975 xfs_fileoff_t offset_fsb
, end_fsb
;
976 int nimaps
= 1, error
= 0;
977 bool shared
= false, trimmed
= false;
979 struct block_device
*bdev
;
981 if (XFS_FORCED_SHUTDOWN(mp
))
984 if (((flags
& (IOMAP_WRITE
| IOMAP_DIRECT
)) == IOMAP_WRITE
) &&
985 !IS_DAX(inode
) && !xfs_get_extsz_hint(ip
)) {
986 /* Reserve delalloc blocks for regular writeback. */
987 return xfs_file_iomap_begin_delay(inode
, offset
, length
, flags
,
991 if (need_excl_ilock(ip
, flags
)) {
992 lockmode
= XFS_ILOCK_EXCL
;
993 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
995 lockmode
= xfs_ilock_data_map_shared(ip
);
998 ASSERT(offset
<= mp
->m_super
->s_maxbytes
);
999 if ((xfs_fsize_t
)offset
+ length
> mp
->m_super
->s_maxbytes
)
1000 length
= mp
->m_super
->s_maxbytes
- offset
;
1001 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1002 end_fsb
= XFS_B_TO_FSB(mp
, offset
+ length
);
1004 error
= xfs_bmapi_read(ip
, offset_fsb
, end_fsb
- offset_fsb
, &imap
,
1009 if (flags
& IOMAP_REPORT
) {
1010 /* Trim the mapping to the nearest shared extent boundary. */
1011 error
= xfs_reflink_trim_around_shared(ip
, &imap
, &shared
,
1017 if ((flags
& (IOMAP_WRITE
| IOMAP_ZERO
)) && xfs_is_reflink_inode(ip
)) {
1018 if (flags
& IOMAP_DIRECT
) {
1019 /* may drop and re-acquire the ilock */
1020 error
= xfs_reflink_allocate_cow(ip
, &imap
, &shared
,
1025 error
= xfs_reflink_reserve_cow(ip
, &imap
, &shared
);
1030 end_fsb
= imap
.br_startoff
+ imap
.br_blockcount
;
1031 length
= XFS_FSB_TO_B(mp
, end_fsb
) - offset
;
1034 if ((flags
& IOMAP_WRITE
) && imap_needs_alloc(inode
, &imap
, nimaps
)) {
1036 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
1037 * pages to keep the chunks of work done where somewhat symmetric
1038 * with the work writeback does. This is a completely arbitrary
1039 * number pulled out of thin air as a best guess for initial
1042 * Note that the values needs to be less than 32-bits wide until
1043 * the lower level functions are updated.
1045 length
= min_t(loff_t
, length
, 1024 * PAGE_SIZE
);
1047 * xfs_iomap_write_direct() expects the shared lock. It
1048 * is unlocked on return.
1050 if (lockmode
== XFS_ILOCK_EXCL
)
1051 xfs_ilock_demote(ip
, lockmode
);
1052 error
= xfs_iomap_write_direct(ip
, offset
, length
, &imap
,
1057 iomap
->flags
= IOMAP_F_NEW
;
1058 trace_xfs_iomap_alloc(ip
, offset
, length
, 0, &imap
);
1062 xfs_iunlock(ip
, lockmode
);
1063 trace_xfs_iomap_found(ip
, offset
, length
, 0, &imap
);
1066 xfs_bmbt_to_iomap(ip
, iomap
, &imap
);
1068 /* optionally associate a dax device with the iomap bdev */
1070 if (blk_queue_dax(bdev
->bd_queue
))
1071 iomap
->dax_dev
= dax_get_by_host(bdev
->bd_disk
->disk_name
);
1073 iomap
->dax_dev
= NULL
;
1076 iomap
->flags
|= IOMAP_F_SHARED
;
1079 xfs_iunlock(ip
, lockmode
);
1084 xfs_file_iomap_end_delalloc(
1085 struct xfs_inode
*ip
,
1089 struct iomap
*iomap
)
1091 struct xfs_mount
*mp
= ip
->i_mount
;
1092 xfs_fileoff_t start_fsb
;
1093 xfs_fileoff_t end_fsb
;
1097 * Behave as if the write failed if drop writes is enabled. Set the NEW
1098 * flag to force delalloc cleanup.
1100 if (xfs_mp_drop_writes(mp
)) {
1101 iomap
->flags
|= IOMAP_F_NEW
;
1106 * start_fsb refers to the first unused block after a short write. If
1107 * nothing was written, round offset down to point at the first block in
1110 if (unlikely(!written
))
1111 start_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1113 start_fsb
= XFS_B_TO_FSB(mp
, offset
+ written
);
1114 end_fsb
= XFS_B_TO_FSB(mp
, offset
+ length
);
1117 * Trim delalloc blocks if they were allocated by this write and we
1118 * didn't manage to write the whole range.
1120 * We don't need to care about racing delalloc as we hold i_mutex
1121 * across the reserve/allocate/unreserve calls. If there are delalloc
1122 * blocks in the range, they are ours.
1124 if ((iomap
->flags
& IOMAP_F_NEW
) && start_fsb
< end_fsb
) {
1125 truncate_pagecache_range(VFS_I(ip
), XFS_FSB_TO_B(mp
, start_fsb
),
1126 XFS_FSB_TO_B(mp
, end_fsb
) - 1);
1128 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1129 error
= xfs_bmap_punch_delalloc_range(ip
, start_fsb
,
1130 end_fsb
- start_fsb
);
1131 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1133 if (error
&& !XFS_FORCED_SHUTDOWN(mp
)) {
1134 xfs_alert(mp
, "%s: unable to clean up ino %lld",
1135 __func__
, ip
->i_ino
);
1145 struct inode
*inode
,
1150 struct iomap
*iomap
)
1152 put_dax(iomap
->dax_dev
);
1153 if ((flags
& IOMAP_WRITE
) && iomap
->type
== IOMAP_DELALLOC
)
1154 return xfs_file_iomap_end_delalloc(XFS_I(inode
), offset
,
1155 length
, written
, iomap
);
1159 const struct iomap_ops xfs_iomap_ops
= {
1160 .iomap_begin
= xfs_file_iomap_begin
,
1161 .iomap_end
= xfs_file_iomap_end
,
1165 xfs_xattr_iomap_begin(
1166 struct inode
*inode
,
1170 struct iomap
*iomap
)
1172 struct xfs_inode
*ip
= XFS_I(inode
);
1173 struct xfs_mount
*mp
= ip
->i_mount
;
1174 xfs_fileoff_t offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1175 xfs_fileoff_t end_fsb
= XFS_B_TO_FSB(mp
, offset
+ length
);
1176 struct xfs_bmbt_irec imap
;
1177 int nimaps
= 1, error
= 0;
1180 if (XFS_FORCED_SHUTDOWN(mp
))
1183 lockmode
= xfs_ilock_attr_map_shared(ip
);
1185 /* if there are no attribute fork or extents, return ENOENT */
1186 if (!XFS_IFORK_Q(ip
) || !ip
->i_d
.di_anextents
) {
1191 ASSERT(ip
->i_d
.di_aformat
!= XFS_DINODE_FMT_LOCAL
);
1192 error
= xfs_bmapi_read(ip
, offset_fsb
, end_fsb
- offset_fsb
, &imap
,
1193 &nimaps
, XFS_BMAPI_ENTIRE
| XFS_BMAPI_ATTRFORK
);
1195 xfs_iunlock(ip
, lockmode
);
1199 xfs_bmbt_to_iomap(ip
, iomap
, &imap
);
1205 const struct iomap_ops xfs_xattr_iomap_ops
= {
1206 .iomap_begin
= xfs_xattr_iomap_begin
,