2 * Copyright (c) 2000-2005 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
24 #include "xfs_trans.h"
25 #include "xfs_mount.h"
26 #include "xfs_bmap_btree.h"
27 #include "xfs_dinode.h"
28 #include "xfs_inode.h"
29 #include "xfs_inode_item.h"
30 #include "xfs_alloc.h"
31 #include "xfs_error.h"
33 #include "xfs_iomap.h"
34 #include "xfs_vnodeops.h"
35 #include "xfs_trace.h"
37 #include <linux/gfp.h>
38 #include <linux/mpage.h>
39 #include <linux/pagevec.h>
40 #include <linux/writeback.h>
48 struct buffer_head
*bh
, *head
;
50 *delalloc
= *unwritten
= 0;
52 bh
= head
= page_buffers(page
);
54 if (buffer_unwritten(bh
))
56 else if (buffer_delay(bh
))
58 } while ((bh
= bh
->b_this_page
) != head
);
61 STATIC
struct block_device
*
62 xfs_find_bdev_for_inode(
65 struct xfs_inode
*ip
= XFS_I(inode
);
66 struct xfs_mount
*mp
= ip
->i_mount
;
68 if (XFS_IS_REALTIME_INODE(ip
))
69 return mp
->m_rtdev_targp
->bt_bdev
;
71 return mp
->m_ddev_targp
->bt_bdev
;
75 * We're now finished for good with this ioend structure.
76 * Update the page state via the associated buffer_heads,
77 * release holds on the inode and bio, and finally free
78 * up memory. Do not use the ioend after this.
84 struct buffer_head
*bh
, *next
;
86 for (bh
= ioend
->io_buffer_head
; bh
; bh
= next
) {
88 bh
->b_end_io(bh
, !ioend
->io_error
);
92 if (ioend
->io_isasync
) {
93 aio_complete(ioend
->io_iocb
, ioend
->io_error
?
94 ioend
->io_error
: ioend
->io_result
, 0);
96 inode_dio_done(ioend
->io_inode
);
99 mempool_free(ioend
, xfs_ioend_pool
);
103 * Fast and loose check if this write could update the on-disk inode size.
105 static inline bool xfs_ioend_is_append(struct xfs_ioend
*ioend
)
107 return ioend
->io_offset
+ ioend
->io_size
>
108 XFS_I(ioend
->io_inode
)->i_d
.di_size
;
112 xfs_setfilesize_trans_alloc(
113 struct xfs_ioend
*ioend
)
115 struct xfs_mount
*mp
= XFS_I(ioend
->io_inode
)->i_mount
;
116 struct xfs_trans
*tp
;
119 tp
= xfs_trans_alloc(mp
, XFS_TRANS_FSYNC_TS
);
121 error
= xfs_trans_reserve(tp
, 0, XFS_FSYNC_TS_LOG_RES(mp
), 0, 0, 0);
123 xfs_trans_cancel(tp
, 0);
127 ioend
->io_append_trans
= tp
;
130 * We hand off the transaction to the completion thread now, so
131 * clear the flag here.
133 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
138 * Update on-disk file size now that data has been written to disk.
142 struct xfs_ioend
*ioend
)
144 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
145 struct xfs_trans
*tp
= ioend
->io_append_trans
;
149 * The transaction was allocated in the I/O submission thread,
150 * thus we need to mark ourselves as beeing in a transaction
153 current_set_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
155 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
156 isize
= xfs_new_eof(ip
, ioend
->io_offset
+ ioend
->io_size
);
158 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
159 xfs_trans_cancel(tp
, 0);
163 trace_xfs_setfilesize(ip
, ioend
->io_offset
, ioend
->io_size
);
165 ip
->i_d
.di_size
= isize
;
166 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
167 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
169 return xfs_trans_commit(tp
, 0);
173 * Schedule IO completion handling on the final put of an ioend.
175 * If there is no work to do we might as well call it a day and free the
180 struct xfs_ioend
*ioend
)
182 if (atomic_dec_and_test(&ioend
->io_remaining
)) {
183 struct xfs_mount
*mp
= XFS_I(ioend
->io_inode
)->i_mount
;
185 if (ioend
->io_type
== IO_UNWRITTEN
)
186 queue_work(mp
->m_unwritten_workqueue
, &ioend
->io_work
);
187 else if (ioend
->io_append_trans
)
188 queue_work(mp
->m_data_workqueue
, &ioend
->io_work
);
190 xfs_destroy_ioend(ioend
);
195 * IO write completion.
199 struct work_struct
*work
)
201 xfs_ioend_t
*ioend
= container_of(work
, xfs_ioend_t
, io_work
);
202 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
205 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
206 ioend
->io_error
= -EIO
;
213 * For unwritten extents we need to issue transactions to convert a
214 * range to normal written extens after the data I/O has finished.
216 if (ioend
->io_type
== IO_UNWRITTEN
) {
218 * For buffered I/O we never preallocate a transaction when
219 * doing the unwritten extent conversion, but for direct I/O
220 * we do not know if we are converting an unwritten extent
221 * or not at the point where we preallocate the transaction.
223 if (ioend
->io_append_trans
) {
224 ASSERT(ioend
->io_isdirect
);
226 current_set_flags_nested(
227 &ioend
->io_append_trans
->t_pflags
, PF_FSTRANS
);
228 xfs_trans_cancel(ioend
->io_append_trans
, 0);
231 error
= xfs_iomap_write_unwritten(ip
, ioend
->io_offset
,
234 ioend
->io_error
= -error
;
237 } else if (ioend
->io_append_trans
) {
238 error
= xfs_setfilesize(ioend
);
240 ioend
->io_error
= -error
;
242 ASSERT(!xfs_ioend_is_append(ioend
));
246 xfs_destroy_ioend(ioend
);
250 * Call IO completion handling in caller context on the final put of an ioend.
253 xfs_finish_ioend_sync(
254 struct xfs_ioend
*ioend
)
256 if (atomic_dec_and_test(&ioend
->io_remaining
))
257 xfs_end_io(&ioend
->io_work
);
261 * Allocate and initialise an IO completion structure.
262 * We need to track unwritten extent write completion here initially.
263 * We'll need to extend this for updating the ondisk inode size later
273 ioend
= mempool_alloc(xfs_ioend_pool
, GFP_NOFS
);
276 * Set the count to 1 initially, which will prevent an I/O
277 * completion callback from happening before we have started
278 * all the I/O from calling the completion routine too early.
280 atomic_set(&ioend
->io_remaining
, 1);
281 ioend
->io_isasync
= 0;
282 ioend
->io_isdirect
= 0;
284 ioend
->io_list
= NULL
;
285 ioend
->io_type
= type
;
286 ioend
->io_inode
= inode
;
287 ioend
->io_buffer_head
= NULL
;
288 ioend
->io_buffer_tail
= NULL
;
289 ioend
->io_offset
= 0;
291 ioend
->io_iocb
= NULL
;
292 ioend
->io_result
= 0;
293 ioend
->io_append_trans
= NULL
;
295 INIT_WORK(&ioend
->io_work
, xfs_end_io
);
303 struct xfs_bmbt_irec
*imap
,
307 struct xfs_inode
*ip
= XFS_I(inode
);
308 struct xfs_mount
*mp
= ip
->i_mount
;
309 ssize_t count
= 1 << inode
->i_blkbits
;
310 xfs_fileoff_t offset_fsb
, end_fsb
;
312 int bmapi_flags
= XFS_BMAPI_ENTIRE
;
315 if (XFS_FORCED_SHUTDOWN(mp
))
316 return -XFS_ERROR(EIO
);
318 if (type
== IO_UNWRITTEN
)
319 bmapi_flags
|= XFS_BMAPI_IGSTATE
;
321 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_SHARED
)) {
323 return -XFS_ERROR(EAGAIN
);
324 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
327 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
328 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
));
329 ASSERT(offset
<= mp
->m_maxioffset
);
331 if (offset
+ count
> mp
->m_maxioffset
)
332 count
= mp
->m_maxioffset
- offset
;
333 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ count
);
334 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
335 error
= xfs_bmapi_read(ip
, offset_fsb
, end_fsb
- offset_fsb
,
336 imap
, &nimaps
, bmapi_flags
);
337 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
340 return -XFS_ERROR(error
);
342 if (type
== IO_DELALLOC
&&
343 (!nimaps
|| isnullstartblock(imap
->br_startblock
))) {
344 error
= xfs_iomap_write_allocate(ip
, offset
, count
, imap
);
346 trace_xfs_map_blocks_alloc(ip
, offset
, count
, type
, imap
);
347 return -XFS_ERROR(error
);
351 if (type
== IO_UNWRITTEN
) {
353 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
354 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
358 trace_xfs_map_blocks_found(ip
, offset
, count
, type
, imap
);
365 struct xfs_bmbt_irec
*imap
,
368 offset
>>= inode
->i_blkbits
;
370 return offset
>= imap
->br_startoff
&&
371 offset
< imap
->br_startoff
+ imap
->br_blockcount
;
375 * BIO completion handler for buffered IO.
382 xfs_ioend_t
*ioend
= bio
->bi_private
;
384 ASSERT(atomic_read(&bio
->bi_cnt
) >= 1);
385 ioend
->io_error
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
) ? 0 : error
;
387 /* Toss bio and pass work off to an xfsdatad thread */
388 bio
->bi_private
= NULL
;
389 bio
->bi_end_io
= NULL
;
392 xfs_finish_ioend(ioend
);
396 xfs_submit_ioend_bio(
397 struct writeback_control
*wbc
,
401 atomic_inc(&ioend
->io_remaining
);
402 bio
->bi_private
= ioend
;
403 bio
->bi_end_io
= xfs_end_bio
;
404 submit_bio(wbc
->sync_mode
== WB_SYNC_ALL
? WRITE_SYNC
: WRITE
, bio
);
409 struct buffer_head
*bh
)
411 int nvecs
= bio_get_nr_vecs(bh
->b_bdev
);
412 struct bio
*bio
= bio_alloc(GFP_NOIO
, nvecs
);
414 ASSERT(bio
->bi_private
== NULL
);
415 bio
->bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
416 bio
->bi_bdev
= bh
->b_bdev
;
421 xfs_start_buffer_writeback(
422 struct buffer_head
*bh
)
424 ASSERT(buffer_mapped(bh
));
425 ASSERT(buffer_locked(bh
));
426 ASSERT(!buffer_delay(bh
));
427 ASSERT(!buffer_unwritten(bh
));
429 mark_buffer_async_write(bh
);
430 set_buffer_uptodate(bh
);
431 clear_buffer_dirty(bh
);
435 xfs_start_page_writeback(
440 ASSERT(PageLocked(page
));
441 ASSERT(!PageWriteback(page
));
443 clear_page_dirty_for_io(page
);
444 set_page_writeback(page
);
446 /* If no buffers on the page are to be written, finish it here */
448 end_page_writeback(page
);
451 static inline int bio_add_buffer(struct bio
*bio
, struct buffer_head
*bh
)
453 return bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
457 * Submit all of the bios for all of the ioends we have saved up, covering the
458 * initial writepage page and also any probed pages.
460 * Because we may have multiple ioends spanning a page, we need to start
461 * writeback on all the buffers before we submit them for I/O. If we mark the
462 * buffers as we got, then we can end up with a page that only has buffers
463 * marked async write and I/O complete on can occur before we mark the other
464 * buffers async write.
466 * The end result of this is that we trip a bug in end_page_writeback() because
467 * we call it twice for the one page as the code in end_buffer_async_write()
468 * assumes that all buffers on the page are started at the same time.
470 * The fix is two passes across the ioend list - one to start writeback on the
471 * buffer_heads, and then submit them for I/O on the second pass.
475 struct writeback_control
*wbc
,
478 xfs_ioend_t
*head
= ioend
;
480 struct buffer_head
*bh
;
482 sector_t lastblock
= 0;
484 /* Pass 1 - start writeback */
486 next
= ioend
->io_list
;
487 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
)
488 xfs_start_buffer_writeback(bh
);
489 } while ((ioend
= next
) != NULL
);
491 /* Pass 2 - submit I/O */
494 next
= ioend
->io_list
;
497 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
501 bio
= xfs_alloc_ioend_bio(bh
);
502 } else if (bh
->b_blocknr
!= lastblock
+ 1) {
503 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
507 if (bio_add_buffer(bio
, bh
) != bh
->b_size
) {
508 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
512 lastblock
= bh
->b_blocknr
;
515 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
516 xfs_finish_ioend(ioend
);
517 } while ((ioend
= next
) != NULL
);
521 * Cancel submission of all buffer_heads so far in this endio.
522 * Toss the endio too. Only ever called for the initial page
523 * in a writepage request, so only ever one page.
530 struct buffer_head
*bh
, *next_bh
;
533 next
= ioend
->io_list
;
534 bh
= ioend
->io_buffer_head
;
536 next_bh
= bh
->b_private
;
537 clear_buffer_async_write(bh
);
539 } while ((bh
= next_bh
) != NULL
);
541 mempool_free(ioend
, xfs_ioend_pool
);
542 } while ((ioend
= next
) != NULL
);
546 * Test to see if we've been building up a completion structure for
547 * earlier buffers -- if so, we try to append to this ioend if we
548 * can, otherwise we finish off any current ioend and start another.
549 * Return true if we've finished the given ioend.
554 struct buffer_head
*bh
,
557 xfs_ioend_t
**result
,
560 xfs_ioend_t
*ioend
= *result
;
562 if (!ioend
|| need_ioend
|| type
!= ioend
->io_type
) {
563 xfs_ioend_t
*previous
= *result
;
565 ioend
= xfs_alloc_ioend(inode
, type
);
566 ioend
->io_offset
= offset
;
567 ioend
->io_buffer_head
= bh
;
568 ioend
->io_buffer_tail
= bh
;
570 previous
->io_list
= ioend
;
573 ioend
->io_buffer_tail
->b_private
= bh
;
574 ioend
->io_buffer_tail
= bh
;
577 bh
->b_private
= NULL
;
578 ioend
->io_size
+= bh
->b_size
;
584 struct buffer_head
*bh
,
585 struct xfs_bmbt_irec
*imap
,
589 struct xfs_mount
*m
= XFS_I(inode
)->i_mount
;
590 xfs_off_t iomap_offset
= XFS_FSB_TO_B(m
, imap
->br_startoff
);
591 xfs_daddr_t iomap_bn
= xfs_fsb_to_db(XFS_I(inode
), imap
->br_startblock
);
593 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
594 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
596 bn
= (iomap_bn
>> (inode
->i_blkbits
- BBSHIFT
)) +
597 ((offset
- iomap_offset
) >> inode
->i_blkbits
);
599 ASSERT(bn
|| XFS_IS_REALTIME_INODE(XFS_I(inode
)));
602 set_buffer_mapped(bh
);
608 struct buffer_head
*bh
,
609 struct xfs_bmbt_irec
*imap
,
612 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
613 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
615 xfs_map_buffer(inode
, bh
, imap
, offset
);
616 set_buffer_mapped(bh
);
617 clear_buffer_delay(bh
);
618 clear_buffer_unwritten(bh
);
622 * Test if a given page is suitable for writing as part of an unwritten
623 * or delayed allocate extent.
630 if (PageWriteback(page
))
633 if (page
->mapping
&& page_has_buffers(page
)) {
634 struct buffer_head
*bh
, *head
;
637 bh
= head
= page_buffers(page
);
639 if (buffer_unwritten(bh
))
640 acceptable
= (type
== IO_UNWRITTEN
);
641 else if (buffer_delay(bh
))
642 acceptable
= (type
== IO_DELALLOC
);
643 else if (buffer_dirty(bh
) && buffer_mapped(bh
))
644 acceptable
= (type
== IO_OVERWRITE
);
647 } while ((bh
= bh
->b_this_page
) != head
);
657 * Allocate & map buffers for page given the extent map. Write it out.
658 * except for the original page of a writepage, this is called on
659 * delalloc/unwritten pages only, for the original page it is possible
660 * that the page has no mapping at all.
667 struct xfs_bmbt_irec
*imap
,
668 xfs_ioend_t
**ioendp
,
669 struct writeback_control
*wbc
)
671 struct buffer_head
*bh
, *head
;
672 xfs_off_t end_offset
;
673 unsigned long p_offset
;
676 int count
= 0, done
= 0, uptodate
= 1;
677 xfs_off_t offset
= page_offset(page
);
679 if (page
->index
!= tindex
)
681 if (!trylock_page(page
))
683 if (PageWriteback(page
))
684 goto fail_unlock_page
;
685 if (page
->mapping
!= inode
->i_mapping
)
686 goto fail_unlock_page
;
687 if (!xfs_is_delayed_page(page
, (*ioendp
)->io_type
))
688 goto fail_unlock_page
;
691 * page_dirty is initially a count of buffers on the page before
692 * EOF and is decremented as we move each into a cleanable state.
696 * End offset is the highest offset that this page should represent.
697 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
698 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
699 * hence give us the correct page_dirty count. On any other page,
700 * it will be zero and in that case we need page_dirty to be the
701 * count of buffers on the page.
703 end_offset
= min_t(unsigned long long,
704 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
707 len
= 1 << inode
->i_blkbits
;
708 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
710 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
711 page_dirty
= p_offset
/ len
;
713 bh
= head
= page_buffers(page
);
715 if (offset
>= end_offset
)
717 if (!buffer_uptodate(bh
))
719 if (!(PageUptodate(page
) || buffer_uptodate(bh
))) {
724 if (buffer_unwritten(bh
) || buffer_delay(bh
) ||
726 if (buffer_unwritten(bh
))
728 else if (buffer_delay(bh
))
733 if (!xfs_imap_valid(inode
, imap
, offset
)) {
739 if (type
!= IO_OVERWRITE
)
740 xfs_map_at_offset(inode
, bh
, imap
, offset
);
741 xfs_add_to_ioend(inode
, bh
, offset
, type
,
749 } while (offset
+= len
, (bh
= bh
->b_this_page
) != head
);
751 if (uptodate
&& bh
== head
)
752 SetPageUptodate(page
);
755 if (--wbc
->nr_to_write
<= 0 &&
756 wbc
->sync_mode
== WB_SYNC_NONE
)
759 xfs_start_page_writeback(page
, !page_dirty
, count
);
769 * Convert & write out a cluster of pages in the same extent as defined
770 * by mp and following the start page.
776 struct xfs_bmbt_irec
*imap
,
777 xfs_ioend_t
**ioendp
,
778 struct writeback_control
*wbc
,
784 pagevec_init(&pvec
, 0);
785 while (!done
&& tindex
<= tlast
) {
786 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
788 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
791 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
792 done
= xfs_convert_page(inode
, pvec
.pages
[i
], tindex
++,
798 pagevec_release(&pvec
);
804 xfs_vm_invalidatepage(
806 unsigned long offset
)
808 trace_xfs_invalidatepage(page
->mapping
->host
, page
, offset
);
809 block_invalidatepage(page
, offset
);
813 * If the page has delalloc buffers on it, we need to punch them out before we
814 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
815 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
816 * is done on that same region - the delalloc extent is returned when none is
817 * supposed to be there.
819 * We prevent this by truncating away the delalloc regions on the page before
820 * invalidating it. Because they are delalloc, we can do this without needing a
821 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
822 * truncation without a transaction as there is no space left for block
823 * reservation (typically why we see a ENOSPC in writeback).
825 * This is not a performance critical path, so for now just do the punching a
826 * buffer head at a time.
829 xfs_aops_discard_page(
832 struct inode
*inode
= page
->mapping
->host
;
833 struct xfs_inode
*ip
= XFS_I(inode
);
834 struct buffer_head
*bh
, *head
;
835 loff_t offset
= page_offset(page
);
837 if (!xfs_is_delayed_page(page
, IO_DELALLOC
))
840 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
843 xfs_alert(ip
->i_mount
,
844 "page discard on page %p, inode 0x%llx, offset %llu.",
845 page
, ip
->i_ino
, offset
);
847 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
848 bh
= head
= page_buffers(page
);
851 xfs_fileoff_t start_fsb
;
853 if (!buffer_delay(bh
))
856 start_fsb
= XFS_B_TO_FSBT(ip
->i_mount
, offset
);
857 error
= xfs_bmap_punch_delalloc_range(ip
, start_fsb
, 1);
859 /* something screwed, just bail */
860 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
861 xfs_alert(ip
->i_mount
,
862 "page discard unable to remove delalloc mapping.");
867 offset
+= 1 << inode
->i_blkbits
;
869 } while ((bh
= bh
->b_this_page
) != head
);
871 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
873 xfs_vm_invalidatepage(page
, 0);
878 * Write out a dirty page.
880 * For delalloc space on the page we need to allocate space and flush it.
881 * For unwritten space on the page we need to start the conversion to
882 * regular allocated space.
883 * For any other dirty buffer heads on the page we should flush them.
888 struct writeback_control
*wbc
)
890 struct inode
*inode
= page
->mapping
->host
;
891 struct buffer_head
*bh
, *head
;
892 struct xfs_bmbt_irec imap
;
893 xfs_ioend_t
*ioend
= NULL
, *iohead
= NULL
;
896 __uint64_t end_offset
;
897 pgoff_t end_index
, last_index
;
899 int err
, imap_valid
= 0, uptodate
= 1;
903 trace_xfs_writepage(inode
, page
, 0);
905 ASSERT(page_has_buffers(page
));
908 * Refuse to write the page out if we are called from reclaim context.
910 * This avoids stack overflows when called from deeply used stacks in
911 * random callers for direct reclaim or memcg reclaim. We explicitly
912 * allow reclaim from kswapd as the stack usage there is relatively low.
914 * This should never happen except in the case of a VM regression so
917 if (WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
)) ==
922 * Given that we do not allow direct reclaim to call us, we should
923 * never be called while in a filesystem transaction.
925 if (WARN_ON(current
->flags
& PF_FSTRANS
))
928 /* Is this page beyond the end of the file? */
929 offset
= i_size_read(inode
);
930 end_index
= offset
>> PAGE_CACHE_SHIFT
;
931 last_index
= (offset
- 1) >> PAGE_CACHE_SHIFT
;
932 if (page
->index
>= end_index
) {
933 if ((page
->index
>= end_index
+ 1) ||
934 !(i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1))) {
940 end_offset
= min_t(unsigned long long,
941 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
943 len
= 1 << inode
->i_blkbits
;
945 bh
= head
= page_buffers(page
);
946 offset
= page_offset(page
);
949 if (wbc
->sync_mode
== WB_SYNC_NONE
)
955 if (offset
>= end_offset
)
957 if (!buffer_uptodate(bh
))
961 * set_page_dirty dirties all buffers in a page, independent
962 * of their state. The dirty state however is entirely
963 * meaningless for holes (!mapped && uptodate), so skip
964 * buffers covering holes here.
966 if (!buffer_mapped(bh
) && buffer_uptodate(bh
)) {
971 if (buffer_unwritten(bh
)) {
972 if (type
!= IO_UNWRITTEN
) {
976 } else if (buffer_delay(bh
)) {
977 if (type
!= IO_DELALLOC
) {
981 } else if (buffer_uptodate(bh
)) {
982 if (type
!= IO_OVERWRITE
) {
987 if (PageUptodate(page
)) {
988 ASSERT(buffer_mapped(bh
));
995 imap_valid
= xfs_imap_valid(inode
, &imap
, offset
);
998 * If we didn't have a valid mapping then we need to
999 * put the new mapping into a separate ioend structure.
1000 * This ensures non-contiguous extents always have
1001 * separate ioends, which is particularly important
1002 * for unwritten extent conversion at I/O completion
1006 err
= xfs_map_blocks(inode
, offset
, &imap
, type
,
1010 imap_valid
= xfs_imap_valid(inode
, &imap
, offset
);
1014 if (type
!= IO_OVERWRITE
)
1015 xfs_map_at_offset(inode
, bh
, &imap
, offset
);
1016 xfs_add_to_ioend(inode
, bh
, offset
, type
, &ioend
,
1024 } while (offset
+= len
, ((bh
= bh
->b_this_page
) != head
));
1026 if (uptodate
&& bh
== head
)
1027 SetPageUptodate(page
);
1029 xfs_start_page_writeback(page
, 1, count
);
1031 if (ioend
&& imap_valid
) {
1032 xfs_off_t end_index
;
1034 end_index
= imap
.br_startoff
+ imap
.br_blockcount
;
1037 end_index
<<= inode
->i_blkbits
;
1040 end_index
= (end_index
- 1) >> PAGE_CACHE_SHIFT
;
1042 /* check against file size */
1043 if (end_index
> last_index
)
1044 end_index
= last_index
;
1046 xfs_cluster_write(inode
, page
->index
+ 1, &imap
, &ioend
,
1052 * Reserve log space if we might write beyond the on-disk
1055 if (ioend
->io_type
!= IO_UNWRITTEN
&&
1056 xfs_ioend_is_append(ioend
)) {
1057 err
= xfs_setfilesize_trans_alloc(ioend
);
1062 xfs_submit_ioend(wbc
, iohead
);
1069 xfs_cancel_ioend(iohead
);
1074 xfs_aops_discard_page(page
);
1075 ClearPageUptodate(page
);
1080 redirty_page_for_writepage(wbc
, page
);
1087 struct address_space
*mapping
,
1088 struct writeback_control
*wbc
)
1090 xfs_iflags_clear(XFS_I(mapping
->host
), XFS_ITRUNCATED
);
1091 return generic_writepages(mapping
, wbc
);
1095 * Called to move a page into cleanable state - and from there
1096 * to be released. The page should already be clean. We always
1097 * have buffer heads in this call.
1099 * Returns 1 if the page is ok to release, 0 otherwise.
1106 int delalloc
, unwritten
;
1108 trace_xfs_releasepage(page
->mapping
->host
, page
, 0);
1110 xfs_count_page_state(page
, &delalloc
, &unwritten
);
1112 if (WARN_ON(delalloc
))
1114 if (WARN_ON(unwritten
))
1117 return try_to_free_buffers(page
);
1122 struct inode
*inode
,
1124 struct buffer_head
*bh_result
,
1128 struct xfs_inode
*ip
= XFS_I(inode
);
1129 struct xfs_mount
*mp
= ip
->i_mount
;
1130 xfs_fileoff_t offset_fsb
, end_fsb
;
1133 struct xfs_bmbt_irec imap
;
1139 if (XFS_FORCED_SHUTDOWN(mp
))
1140 return -XFS_ERROR(EIO
);
1142 offset
= (xfs_off_t
)iblock
<< inode
->i_blkbits
;
1143 ASSERT(bh_result
->b_size
>= (1 << inode
->i_blkbits
));
1144 size
= bh_result
->b_size
;
1146 if (!create
&& direct
&& offset
>= i_size_read(inode
))
1150 lockmode
= XFS_ILOCK_EXCL
;
1151 xfs_ilock(ip
, lockmode
);
1153 lockmode
= xfs_ilock_map_shared(ip
);
1156 ASSERT(offset
<= mp
->m_maxioffset
);
1157 if (offset
+ size
> mp
->m_maxioffset
)
1158 size
= mp
->m_maxioffset
- offset
;
1159 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ size
);
1160 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1162 error
= xfs_bmapi_read(ip
, offset_fsb
, end_fsb
- offset_fsb
,
1163 &imap
, &nimaps
, XFS_BMAPI_ENTIRE
);
1169 (imap
.br_startblock
== HOLESTARTBLOCK
||
1170 imap
.br_startblock
== DELAYSTARTBLOCK
))) {
1172 error
= xfs_iomap_write_direct(ip
, offset
, size
,
1175 error
= xfs_iomap_write_delay(ip
, offset
, size
, &imap
);
1180 trace_xfs_get_blocks_alloc(ip
, offset
, size
, 0, &imap
);
1181 } else if (nimaps
) {
1182 trace_xfs_get_blocks_found(ip
, offset
, size
, 0, &imap
);
1184 trace_xfs_get_blocks_notfound(ip
, offset
, size
);
1187 xfs_iunlock(ip
, lockmode
);
1189 if (imap
.br_startblock
!= HOLESTARTBLOCK
&&
1190 imap
.br_startblock
!= DELAYSTARTBLOCK
) {
1192 * For unwritten extents do not report a disk address on
1193 * the read case (treat as if we're reading into a hole).
1195 if (create
|| !ISUNWRITTEN(&imap
))
1196 xfs_map_buffer(inode
, bh_result
, &imap
, offset
);
1197 if (create
&& ISUNWRITTEN(&imap
)) {
1199 bh_result
->b_private
= inode
;
1200 set_buffer_unwritten(bh_result
);
1205 * If this is a realtime file, data may be on a different device.
1206 * to that pointed to from the buffer_head b_bdev currently.
1208 bh_result
->b_bdev
= xfs_find_bdev_for_inode(inode
);
1211 * If we previously allocated a block out beyond eof and we are now
1212 * coming back to use it then we will need to flag it as new even if it
1213 * has a disk address.
1215 * With sub-block writes into unwritten extents we also need to mark
1216 * the buffer as new so that the unwritten parts of the buffer gets
1220 ((!buffer_mapped(bh_result
) && !buffer_uptodate(bh_result
)) ||
1221 (offset
>= i_size_read(inode
)) ||
1222 (new || ISUNWRITTEN(&imap
))))
1223 set_buffer_new(bh_result
);
1225 if (imap
.br_startblock
== DELAYSTARTBLOCK
) {
1228 set_buffer_uptodate(bh_result
);
1229 set_buffer_mapped(bh_result
);
1230 set_buffer_delay(bh_result
);
1235 * If this is O_DIRECT or the mpage code calling tell them how large
1236 * the mapping is, so that we can avoid repeated get_blocks calls.
1238 if (direct
|| size
> (1 << inode
->i_blkbits
)) {
1239 xfs_off_t mapping_size
;
1241 mapping_size
= imap
.br_startoff
+ imap
.br_blockcount
- iblock
;
1242 mapping_size
<<= inode
->i_blkbits
;
1244 ASSERT(mapping_size
> 0);
1245 if (mapping_size
> size
)
1246 mapping_size
= size
;
1247 if (mapping_size
> LONG_MAX
)
1248 mapping_size
= LONG_MAX
;
1250 bh_result
->b_size
= mapping_size
;
1256 xfs_iunlock(ip
, lockmode
);
1262 struct inode
*inode
,
1264 struct buffer_head
*bh_result
,
1267 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 0);
1271 xfs_get_blocks_direct(
1272 struct inode
*inode
,
1274 struct buffer_head
*bh_result
,
1277 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 1);
1281 * Complete a direct I/O write request.
1283 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1284 * need to issue a transaction to convert the range from unwritten to written
1285 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1286 * to do this and we are done. But in case this was a successful AIO
1287 * request this handler is called from interrupt context, from which we
1288 * can't start transactions. In that case offload the I/O completion to
1289 * the workqueues we also use for buffered I/O completion.
1292 xfs_end_io_direct_write(
1300 struct xfs_ioend
*ioend
= iocb
->private;
1303 * While the generic direct I/O code updates the inode size, it does
1304 * so only after the end_io handler is called, which means our
1305 * end_io handler thinks the on-disk size is outside the in-core
1306 * size. To prevent this just update it a little bit earlier here.
1308 if (offset
+ size
> i_size_read(ioend
->io_inode
))
1309 i_size_write(ioend
->io_inode
, offset
+ size
);
1312 * blockdev_direct_IO can return an error even after the I/O
1313 * completion handler was called. Thus we need to protect
1314 * against double-freeing.
1316 iocb
->private = NULL
;
1318 ioend
->io_offset
= offset
;
1319 ioend
->io_size
= size
;
1320 ioend
->io_iocb
= iocb
;
1321 ioend
->io_result
= ret
;
1322 if (private && size
> 0)
1323 ioend
->io_type
= IO_UNWRITTEN
;
1326 ioend
->io_isasync
= 1;
1327 xfs_finish_ioend(ioend
);
1329 xfs_finish_ioend_sync(ioend
);
1337 const struct iovec
*iov
,
1339 unsigned long nr_segs
)
1341 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
1342 struct block_device
*bdev
= xfs_find_bdev_for_inode(inode
);
1343 struct xfs_ioend
*ioend
= NULL
;
1347 size_t size
= iov_length(iov
, nr_segs
);
1350 * We need to preallocate a transaction for a size update
1351 * here. In the case that this write both updates the size
1352 * and converts at least on unwritten extent we will cancel
1353 * the still clean transaction after the I/O has finished.
1355 iocb
->private = ioend
= xfs_alloc_ioend(inode
, IO_DIRECT
);
1356 if (offset
+ size
> XFS_I(inode
)->i_d
.di_size
) {
1357 ret
= xfs_setfilesize_trans_alloc(ioend
);
1359 goto out_destroy_ioend
;
1360 ioend
->io_isdirect
= 1;
1363 ret
= __blockdev_direct_IO(rw
, iocb
, inode
, bdev
, iov
,
1365 xfs_get_blocks_direct
,
1366 xfs_end_io_direct_write
, NULL
, 0);
1367 if (ret
!= -EIOCBQUEUED
&& iocb
->private)
1368 goto out_trans_cancel
;
1370 ret
= __blockdev_direct_IO(rw
, iocb
, inode
, bdev
, iov
,
1372 xfs_get_blocks_direct
,
1379 if (ioend
->io_append_trans
) {
1380 current_set_flags_nested(&ioend
->io_append_trans
->t_pflags
,
1382 xfs_trans_cancel(ioend
->io_append_trans
, 0);
1385 xfs_destroy_ioend(ioend
);
1390 xfs_vm_write_failed(
1391 struct address_space
*mapping
,
1394 struct inode
*inode
= mapping
->host
;
1396 if (to
> inode
->i_size
) {
1398 * Punch out the delalloc blocks we have already allocated.
1400 * Don't bother with xfs_setattr given that nothing can have
1401 * made it to disk yet as the page is still locked at this
1404 struct xfs_inode
*ip
= XFS_I(inode
);
1405 xfs_fileoff_t start_fsb
;
1406 xfs_fileoff_t end_fsb
;
1409 truncate_pagecache(inode
, to
, inode
->i_size
);
1412 * Check if there are any blocks that are outside of i_size
1413 * that need to be trimmed back.
1415 start_fsb
= XFS_B_TO_FSB(ip
->i_mount
, inode
->i_size
) + 1;
1416 end_fsb
= XFS_B_TO_FSB(ip
->i_mount
, to
);
1417 if (end_fsb
<= start_fsb
)
1420 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1421 error
= xfs_bmap_punch_delalloc_range(ip
, start_fsb
,
1422 end_fsb
- start_fsb
);
1424 /* something screwed, just bail */
1425 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
1426 xfs_alert(ip
->i_mount
,
1427 "xfs_vm_write_failed: unable to clean up ino %lld",
1431 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1438 struct address_space
*mapping
,
1442 struct page
**pagep
,
1447 ret
= block_write_begin(mapping
, pos
, len
, flags
| AOP_FLAG_NOFS
,
1448 pagep
, xfs_get_blocks
);
1450 xfs_vm_write_failed(mapping
, pos
+ len
);
1457 struct address_space
*mapping
,
1466 ret
= generic_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1467 if (unlikely(ret
< len
))
1468 xfs_vm_write_failed(mapping
, pos
+ len
);
1474 struct address_space
*mapping
,
1477 struct inode
*inode
= (struct inode
*)mapping
->host
;
1478 struct xfs_inode
*ip
= XFS_I(inode
);
1480 trace_xfs_vm_bmap(XFS_I(inode
));
1481 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
1482 xfs_flush_pages(ip
, (xfs_off_t
)0, -1, 0, FI_REMAPF
);
1483 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
1484 return generic_block_bmap(mapping
, block
, xfs_get_blocks
);
1489 struct file
*unused
,
1492 return mpage_readpage(page
, xfs_get_blocks
);
1497 struct file
*unused
,
1498 struct address_space
*mapping
,
1499 struct list_head
*pages
,
1502 return mpage_readpages(mapping
, pages
, nr_pages
, xfs_get_blocks
);
1505 const struct address_space_operations xfs_address_space_operations
= {
1506 .readpage
= xfs_vm_readpage
,
1507 .readpages
= xfs_vm_readpages
,
1508 .writepage
= xfs_vm_writepage
,
1509 .writepages
= xfs_vm_writepages
,
1510 .releasepage
= xfs_vm_releasepage
,
1511 .invalidatepage
= xfs_vm_invalidatepage
,
1512 .write_begin
= xfs_vm_write_begin
,
1513 .write_end
= xfs_vm_write_end
,
1514 .bmap
= xfs_vm_bmap
,
1515 .direct_IO
= xfs_vm_direct_IO
,
1516 .migratepage
= buffer_migrate_page
,
1517 .is_partially_uptodate
= block_is_partially_uptodate
,
1518 .error_remove_page
= generic_error_remove_page
,