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
26 #include "xfs_trans.h"
27 #include "xfs_dmapi.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_dir_sf.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_alloc.h"
38 #include "xfs_btree.h"
39 #include "xfs_error.h"
41 #include "xfs_iomap.h"
42 #include <linux/mpage.h>
43 #include <linux/pagevec.h>
44 #include <linux/writeback.h>
53 struct buffer_head
*bh
, *head
;
55 *delalloc
= *unmapped
= *unwritten
= 0;
57 bh
= head
= page_buffers(page
);
59 if (buffer_uptodate(bh
) && !buffer_mapped(bh
))
61 else if (buffer_unwritten(bh
) && !buffer_delay(bh
))
62 clear_buffer_unwritten(bh
);
63 else if (buffer_unwritten(bh
))
65 else if (buffer_delay(bh
))
67 } while ((bh
= bh
->b_this_page
) != head
);
70 #if defined(XFS_RW_TRACE)
79 vnode_t
*vp
= vn_from_inode(inode
);
80 loff_t isize
= i_size_read(inode
);
81 loff_t offset
= page_offset(page
);
82 int delalloc
= -1, unmapped
= -1, unwritten
= -1;
84 if (page_has_buffers(page
))
85 xfs_count_page_state(page
, &delalloc
, &unmapped
, &unwritten
);
91 ktrace_enter(ip
->i_rwtrace
,
92 (void *)((unsigned long)tag
),
96 (void *)((unsigned long)mask
),
97 (void *)((unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff)),
98 (void *)((unsigned long)(ip
->i_d
.di_size
& 0xffffffff)),
99 (void *)((unsigned long)((isize
>> 32) & 0xffffffff)),
100 (void *)((unsigned long)(isize
& 0xffffffff)),
101 (void *)((unsigned long)((offset
>> 32) & 0xffffffff)),
102 (void *)((unsigned long)(offset
& 0xffffffff)),
103 (void *)((unsigned long)delalloc
),
104 (void *)((unsigned long)unmapped
),
105 (void *)((unsigned long)unwritten
),
106 (void *)((unsigned long)current_pid()),
110 #define xfs_page_trace(tag, inode, page, mask)
114 * Schedule IO completion handling on a xfsdatad if this was
115 * the final hold on this ioend.
121 if (atomic_dec_and_test(&ioend
->io_remaining
))
122 queue_work(xfsdatad_workqueue
, &ioend
->io_work
);
126 * We're now finished for good with this ioend structure.
127 * Update the page state via the associated buffer_heads,
128 * release holds on the inode and bio, and finally free
129 * up memory. Do not use the ioend after this.
135 struct buffer_head
*bh
, *next
;
137 for (bh
= ioend
->io_buffer_head
; bh
; bh
= next
) {
138 next
= bh
->b_private
;
139 bh
->b_end_io(bh
, ioend
->io_uptodate
);
142 vn_iowake(ioend
->io_vnode
);
143 mempool_free(ioend
, xfs_ioend_pool
);
147 * Buffered IO write completion for delayed allocate extents.
148 * TODO: Update ondisk isize now that we know the file data
149 * has been flushed (i.e. the notorious "NULL file" problem).
152 xfs_end_bio_delalloc(
155 xfs_ioend_t
*ioend
= data
;
157 xfs_destroy_ioend(ioend
);
161 * Buffered IO write completion for regular, written extents.
167 xfs_ioend_t
*ioend
= data
;
169 xfs_destroy_ioend(ioend
);
173 * IO write completion for unwritten extents.
175 * Issue transactions to convert a buffer range from unwritten
176 * to written extents.
179 xfs_end_bio_unwritten(
182 xfs_ioend_t
*ioend
= data
;
183 vnode_t
*vp
= ioend
->io_vnode
;
184 xfs_off_t offset
= ioend
->io_offset
;
185 size_t size
= ioend
->io_size
;
188 if (ioend
->io_uptodate
)
189 VOP_BMAP(vp
, offset
, size
, BMAPI_UNWRITTEN
, NULL
, NULL
, error
);
190 xfs_destroy_ioend(ioend
);
194 * Allocate and initialise an IO completion structure.
195 * We need to track unwritten extent write completion here initially.
196 * We'll need to extend this for updating the ondisk inode size later
206 ioend
= mempool_alloc(xfs_ioend_pool
, GFP_NOFS
);
209 * Set the count to 1 initially, which will prevent an I/O
210 * completion callback from happening before we have started
211 * all the I/O from calling the completion routine too early.
213 atomic_set(&ioend
->io_remaining
, 1);
214 ioend
->io_uptodate
= 1; /* cleared if any I/O fails */
215 ioend
->io_list
= NULL
;
216 ioend
->io_type
= type
;
217 ioend
->io_vnode
= vn_from_inode(inode
);
218 ioend
->io_buffer_head
= NULL
;
219 ioend
->io_buffer_tail
= NULL
;
220 atomic_inc(&ioend
->io_vnode
->v_iocount
);
221 ioend
->io_offset
= 0;
224 if (type
== IOMAP_UNWRITTEN
)
225 INIT_WORK(&ioend
->io_work
, xfs_end_bio_unwritten
, ioend
);
226 else if (type
== IOMAP_DELAY
)
227 INIT_WORK(&ioend
->io_work
, xfs_end_bio_delalloc
, ioend
);
229 INIT_WORK(&ioend
->io_work
, xfs_end_bio_written
, ioend
);
242 vnode_t
*vp
= vn_from_inode(inode
);
243 int error
, nmaps
= 1;
245 VOP_BMAP(vp
, offset
, count
, flags
, mapp
, &nmaps
, error
);
246 if (!error
&& (flags
& (BMAPI_WRITE
|BMAPI_ALLOCATE
)))
256 return offset
>= iomapp
->iomap_offset
&&
257 offset
< iomapp
->iomap_offset
+ iomapp
->iomap_bsize
;
261 * BIO completion handler for buffered IO.
266 unsigned int bytes_done
,
269 xfs_ioend_t
*ioend
= bio
->bi_private
;
275 ASSERT(atomic_read(&bio
->bi_cnt
) >= 1);
277 /* Toss bio and pass work off to an xfsdatad thread */
278 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
279 ioend
->io_uptodate
= 0;
280 bio
->bi_private
= NULL
;
281 bio
->bi_end_io
= NULL
;
284 xfs_finish_ioend(ioend
);
289 xfs_submit_ioend_bio(
293 atomic_inc(&ioend
->io_remaining
);
295 bio
->bi_private
= ioend
;
296 bio
->bi_end_io
= xfs_end_bio
;
298 submit_bio(WRITE
, bio
);
299 ASSERT(!bio_flagged(bio
, BIO_EOPNOTSUPP
));
305 struct buffer_head
*bh
)
308 int nvecs
= bio_get_nr_vecs(bh
->b_bdev
);
311 bio
= bio_alloc(GFP_NOIO
, nvecs
);
315 ASSERT(bio
->bi_private
== NULL
);
316 bio
->bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
317 bio
->bi_bdev
= bh
->b_bdev
;
323 xfs_start_buffer_writeback(
324 struct buffer_head
*bh
)
326 ASSERT(buffer_mapped(bh
));
327 ASSERT(buffer_locked(bh
));
328 ASSERT(!buffer_delay(bh
));
329 ASSERT(!buffer_unwritten(bh
));
331 mark_buffer_async_write(bh
);
332 set_buffer_uptodate(bh
);
333 clear_buffer_dirty(bh
);
337 xfs_start_page_writeback(
339 struct writeback_control
*wbc
,
343 ASSERT(PageLocked(page
));
344 ASSERT(!PageWriteback(page
));
345 set_page_writeback(page
);
347 clear_page_dirty(page
);
350 end_page_writeback(page
);
351 wbc
->pages_skipped
++; /* We didn't write this page */
355 static inline int bio_add_buffer(struct bio
*bio
, struct buffer_head
*bh
)
357 return bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
361 * Submit all of the bios for all of the ioends we have saved up, covering the
362 * initial writepage page and also any probed pages.
364 * Because we may have multiple ioends spanning a page, we need to start
365 * writeback on all the buffers before we submit them for I/O. If we mark the
366 * buffers as we got, then we can end up with a page that only has buffers
367 * marked async write and I/O complete on can occur before we mark the other
368 * buffers async write.
370 * The end result of this is that we trip a bug in end_page_writeback() because
371 * we call it twice for the one page as the code in end_buffer_async_write()
372 * assumes that all buffers on the page are started at the same time.
374 * The fix is two passes across the ioend list - one to start writeback on the
375 * buffer_heads, and then submit them for I/O on the second pass.
381 xfs_ioend_t
*head
= ioend
;
383 struct buffer_head
*bh
;
385 sector_t lastblock
= 0;
387 /* Pass 1 - start writeback */
389 next
= ioend
->io_list
;
390 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
391 xfs_start_buffer_writeback(bh
);
393 } while ((ioend
= next
) != NULL
);
395 /* Pass 2 - submit I/O */
398 next
= ioend
->io_list
;
401 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
405 bio
= xfs_alloc_ioend_bio(bh
);
406 } else if (bh
->b_blocknr
!= lastblock
+ 1) {
407 xfs_submit_ioend_bio(ioend
, bio
);
411 if (bio_add_buffer(bio
, bh
) != bh
->b_size
) {
412 xfs_submit_ioend_bio(ioend
, bio
);
416 lastblock
= bh
->b_blocknr
;
419 xfs_submit_ioend_bio(ioend
, bio
);
420 xfs_finish_ioend(ioend
);
421 } while ((ioend
= next
) != NULL
);
425 * Cancel submission of all buffer_heads so far in this endio.
426 * Toss the endio too. Only ever called for the initial page
427 * in a writepage request, so only ever one page.
434 struct buffer_head
*bh
, *next_bh
;
437 next
= ioend
->io_list
;
438 bh
= ioend
->io_buffer_head
;
440 next_bh
= bh
->b_private
;
441 clear_buffer_async_write(bh
);
443 } while ((bh
= next_bh
) != NULL
);
445 vn_iowake(ioend
->io_vnode
);
446 mempool_free(ioend
, xfs_ioend_pool
);
447 } while ((ioend
= next
) != NULL
);
451 * Test to see if we've been building up a completion structure for
452 * earlier buffers -- if so, we try to append to this ioend if we
453 * can, otherwise we finish off any current ioend and start another.
454 * Return true if we've finished the given ioend.
459 struct buffer_head
*bh
,
462 xfs_ioend_t
**result
,
465 xfs_ioend_t
*ioend
= *result
;
467 if (!ioend
|| need_ioend
|| type
!= ioend
->io_type
) {
468 xfs_ioend_t
*previous
= *result
;
470 ioend
= xfs_alloc_ioend(inode
, type
);
471 ioend
->io_offset
= offset
;
472 ioend
->io_buffer_head
= bh
;
473 ioend
->io_buffer_tail
= bh
;
475 previous
->io_list
= ioend
;
478 ioend
->io_buffer_tail
->b_private
= bh
;
479 ioend
->io_buffer_tail
= bh
;
482 bh
->b_private
= NULL
;
483 ioend
->io_size
+= bh
->b_size
;
488 struct buffer_head
*bh
,
495 ASSERT(mp
->iomap_bn
!= IOMAP_DADDR_NULL
);
497 bn
= (mp
->iomap_bn
>> (block_bits
- BBSHIFT
)) +
498 ((offset
- mp
->iomap_offset
) >> block_bits
);
500 ASSERT(bn
|| (mp
->iomap_flags
& IOMAP_REALTIME
));
503 set_buffer_mapped(bh
);
508 struct buffer_head
*bh
,
513 ASSERT(!(iomapp
->iomap_flags
& IOMAP_HOLE
));
514 ASSERT(!(iomapp
->iomap_flags
& IOMAP_DELAY
));
517 xfs_map_buffer(bh
, iomapp
, offset
, block_bits
);
518 bh
->b_bdev
= iomapp
->iomap_target
->bt_bdev
;
519 set_buffer_mapped(bh
);
520 clear_buffer_delay(bh
);
521 clear_buffer_unwritten(bh
);
525 * Look for a page at index that is suitable for clustering.
530 unsigned int pg_offset
,
535 if (PageWriteback(page
))
538 if (page
->mapping
&& PageDirty(page
)) {
539 if (page_has_buffers(page
)) {
540 struct buffer_head
*bh
, *head
;
542 bh
= head
= page_buffers(page
);
544 if (!buffer_uptodate(bh
))
546 if (mapped
!= buffer_mapped(bh
))
549 if (ret
>= pg_offset
)
551 } while ((bh
= bh
->b_this_page
) != head
);
553 ret
= mapped
? 0 : PAGE_CACHE_SIZE
;
562 struct page
*startpage
,
563 struct buffer_head
*bh
,
564 struct buffer_head
*head
,
568 pgoff_t tindex
, tlast
, tloff
;
572 /* First sum forwards in this page */
574 if (!buffer_uptodate(bh
) || (mapped
!= buffer_mapped(bh
)))
577 } while ((bh
= bh
->b_this_page
) != head
);
579 /* if we reached the end of the page, sum forwards in following pages */
580 tlast
= i_size_read(inode
) >> PAGE_CACHE_SHIFT
;
581 tindex
= startpage
->index
+ 1;
583 /* Prune this back to avoid pathological behavior */
584 tloff
= min(tlast
, startpage
->index
+ 64);
586 pagevec_init(&pvec
, 0);
587 while (!done
&& tindex
<= tloff
) {
588 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
590 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
593 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
594 struct page
*page
= pvec
.pages
[i
];
595 size_t pg_offset
, len
= 0;
597 if (tindex
== tlast
) {
599 i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1);
605 pg_offset
= PAGE_CACHE_SIZE
;
607 if (page
->index
== tindex
&& !TestSetPageLocked(page
)) {
608 len
= xfs_probe_page(page
, pg_offset
, mapped
);
621 pagevec_release(&pvec
);
629 * Test if a given page is suitable for writing as part of an unwritten
630 * or delayed allocate extent.
637 if (PageWriteback(page
))
640 if (page
->mapping
&& page_has_buffers(page
)) {
641 struct buffer_head
*bh
, *head
;
644 bh
= head
= page_buffers(page
);
646 if (buffer_unwritten(bh
))
647 acceptable
= (type
== IOMAP_UNWRITTEN
);
648 else if (buffer_delay(bh
))
649 acceptable
= (type
== IOMAP_DELAY
);
650 else if (buffer_dirty(bh
) && buffer_mapped(bh
))
651 acceptable
= (type
== 0);
654 } while ((bh
= bh
->b_this_page
) != head
);
664 * Allocate & map buffers for page given the extent map. Write it out.
665 * except for the original page of a writepage, this is called on
666 * delalloc/unwritten pages only, for the original page it is possible
667 * that the page has no mapping at all.
675 xfs_ioend_t
**ioendp
,
676 struct writeback_control
*wbc
,
680 struct buffer_head
*bh
, *head
;
681 xfs_off_t end_offset
;
682 unsigned long p_offset
;
684 int bbits
= inode
->i_blkbits
;
686 int count
= 0, done
= 0, uptodate
= 1;
687 xfs_off_t offset
= page_offset(page
);
689 if (page
->index
!= tindex
)
691 if (TestSetPageLocked(page
))
693 if (PageWriteback(page
))
694 goto fail_unlock_page
;
695 if (page
->mapping
!= inode
->i_mapping
)
696 goto fail_unlock_page
;
697 if (!xfs_is_delayed_page(page
, (*ioendp
)->io_type
))
698 goto fail_unlock_page
;
701 * page_dirty is initially a count of buffers on the page before
702 * EOF and is decremented as we move each into a cleanable state.
706 * End offset is the highest offset that this page should represent.
707 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
708 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
709 * hence give us the correct page_dirty count. On any other page,
710 * it will be zero and in that case we need page_dirty to be the
711 * count of buffers on the page.
713 end_offset
= min_t(unsigned long long,
714 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
717 len
= 1 << inode
->i_blkbits
;
718 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
720 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
721 page_dirty
= p_offset
/ len
;
723 bh
= head
= page_buffers(page
);
725 if (offset
>= end_offset
)
727 if (!buffer_uptodate(bh
))
729 if (!(PageUptodate(page
) || buffer_uptodate(bh
))) {
734 if (buffer_unwritten(bh
) || buffer_delay(bh
)) {
735 if (buffer_unwritten(bh
))
736 type
= IOMAP_UNWRITTEN
;
740 if (!xfs_iomap_valid(mp
, offset
)) {
745 ASSERT(!(mp
->iomap_flags
& IOMAP_HOLE
));
746 ASSERT(!(mp
->iomap_flags
& IOMAP_DELAY
));
748 xfs_map_at_offset(bh
, offset
, bbits
, mp
);
750 xfs_add_to_ioend(inode
, bh
, offset
,
753 set_buffer_dirty(bh
);
755 mark_buffer_dirty(bh
);
761 if (buffer_mapped(bh
) && all_bh
&& startio
) {
763 xfs_add_to_ioend(inode
, bh
, offset
,
771 } while (offset
+= len
, (bh
= bh
->b_this_page
) != head
);
773 if (uptodate
&& bh
== head
)
774 SetPageUptodate(page
);
778 struct backing_dev_info
*bdi
;
780 bdi
= inode
->i_mapping
->backing_dev_info
;
782 if (bdi_write_congested(bdi
)) {
783 wbc
->encountered_congestion
= 1;
785 } else if (wbc
->nr_to_write
<= 0) {
789 xfs_start_page_writeback(page
, wbc
, !page_dirty
, count
);
800 * Convert & write out a cluster of pages in the same extent as defined
801 * by mp and following the start page.
808 xfs_ioend_t
**ioendp
,
809 struct writeback_control
*wbc
,
817 pagevec_init(&pvec
, 0);
818 while (!done
&& tindex
<= tlast
) {
819 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
821 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
824 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
825 done
= xfs_convert_page(inode
, pvec
.pages
[i
], tindex
++,
826 iomapp
, ioendp
, wbc
, startio
, all_bh
);
831 pagevec_release(&pvec
);
837 * Calling this without startio set means we are being asked to make a dirty
838 * page ready for freeing it's buffers. When called with startio set then
839 * we are coming from writepage.
841 * When called with startio set it is important that we write the WHOLE
843 * The bh->b_state's cannot know if any of the blocks or which block for
844 * that matter are dirty due to mmap writes, and therefore bh uptodate is
845 * only valid if the page itself isn't completely uptodate. Some layers
846 * may clear the page dirty flag prior to calling write page, under the
847 * assumption the entire page will be written out; by not writing out the
848 * whole page the page can be reused before all valid dirty data is
849 * written out. Note: in the case of a page that has been dirty'd by
850 * mapwrite and but partially setup by block_prepare_write the
851 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
852 * valid state, thus the whole page must be written out thing.
856 xfs_page_state_convert(
859 struct writeback_control
*wbc
,
861 int unmapped
) /* also implies page uptodate */
863 struct buffer_head
*bh
, *head
;
865 xfs_ioend_t
*ioend
= NULL
, *iohead
= NULL
;
867 unsigned long p_offset
= 0;
869 __uint64_t end_offset
;
870 pgoff_t end_index
, last_index
, tlast
;
872 int flags
, err
, iomap_valid
= 0, uptodate
= 1;
873 int page_dirty
, count
= 0, trylock_flag
= 0;
874 int all_bh
= unmapped
;
876 /* wait for other IO threads? */
877 if (startio
&& (wbc
->sync_mode
== WB_SYNC_NONE
&& wbc
->nonblocking
))
878 trylock_flag
|= BMAPI_TRYLOCK
;
880 /* Is this page beyond the end of the file? */
881 offset
= i_size_read(inode
);
882 end_index
= offset
>> PAGE_CACHE_SHIFT
;
883 last_index
= (offset
- 1) >> PAGE_CACHE_SHIFT
;
884 if (page
->index
>= end_index
) {
885 if ((page
->index
>= end_index
+ 1) ||
886 !(i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1))) {
894 * page_dirty is initially a count of buffers on the page before
895 * EOF and is decremented as we move each into a cleanable state.
899 * End offset is the highest offset that this page should represent.
900 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
901 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
902 * hence give us the correct page_dirty count. On any other page,
903 * it will be zero and in that case we need page_dirty to be the
904 * count of buffers on the page.
906 end_offset
= min_t(unsigned long long,
907 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
, offset
);
908 len
= 1 << inode
->i_blkbits
;
909 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
911 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
912 page_dirty
= p_offset
/ len
;
914 bh
= head
= page_buffers(page
);
915 offset
= page_offset(page
);
919 /* TODO: cleanup count and page_dirty */
922 if (offset
>= end_offset
)
924 if (!buffer_uptodate(bh
))
926 if (!(PageUptodate(page
) || buffer_uptodate(bh
)) && !startio
) {
928 * the iomap is actually still valid, but the ioend
929 * isn't. shouldn't happen too often.
936 iomap_valid
= xfs_iomap_valid(&iomap
, offset
);
939 * First case, map an unwritten extent and prepare for
940 * extent state conversion transaction on completion.
942 * Second case, allocate space for a delalloc buffer.
943 * We can return EAGAIN here in the release page case.
945 * Third case, an unmapped buffer was found, and we are
946 * in a path where we need to write the whole page out.
948 if (buffer_unwritten(bh
) || buffer_delay(bh
) ||
949 ((buffer_uptodate(bh
) || PageUptodate(page
)) &&
950 !buffer_mapped(bh
) && (unmapped
|| startio
))) {
952 * Make sure we don't use a read-only iomap
954 if (flags
== BMAPI_READ
)
957 if (buffer_unwritten(bh
)) {
958 type
= IOMAP_UNWRITTEN
;
959 flags
= BMAPI_WRITE
|BMAPI_IGNSTATE
;
960 } else if (buffer_delay(bh
)) {
962 flags
= BMAPI_ALLOCATE
;
964 flags
|= trylock_flag
;
967 flags
= BMAPI_WRITE
|BMAPI_MMAP
;
971 if (type
== IOMAP_NEW
) {
972 size
= xfs_probe_cluster(inode
,
978 err
= xfs_map_blocks(inode
, offset
, size
,
982 iomap_valid
= xfs_iomap_valid(&iomap
, offset
);
985 xfs_map_at_offset(bh
, offset
,
986 inode
->i_blkbits
, &iomap
);
988 xfs_add_to_ioend(inode
, bh
, offset
,
992 set_buffer_dirty(bh
);
994 mark_buffer_dirty(bh
);
999 } else if (buffer_uptodate(bh
) && startio
) {
1001 * we got here because the buffer is already mapped.
1002 * That means it must already have extents allocated
1003 * underneath it. Map the extent by reading it.
1005 if (!iomap_valid
|| type
!= 0) {
1007 size
= xfs_probe_cluster(inode
, page
, bh
,
1009 err
= xfs_map_blocks(inode
, offset
, size
,
1013 iomap_valid
= xfs_iomap_valid(&iomap
, offset
);
1017 if (!test_and_set_bit(BH_Lock
, &bh
->b_state
)) {
1018 ASSERT(buffer_mapped(bh
));
1021 xfs_add_to_ioend(inode
, bh
, offset
, type
,
1022 &ioend
, !iomap_valid
);
1028 } else if ((buffer_uptodate(bh
) || PageUptodate(page
)) &&
1029 (unmapped
|| startio
)) {
1036 } while (offset
+= len
, ((bh
= bh
->b_this_page
) != head
));
1038 if (uptodate
&& bh
== head
)
1039 SetPageUptodate(page
);
1042 xfs_start_page_writeback(page
, wbc
, 1, count
);
1044 if (ioend
&& iomap_valid
) {
1045 offset
= (iomap
.iomap_offset
+ iomap
.iomap_bsize
- 1) >>
1047 tlast
= min_t(pgoff_t
, offset
, last_index
);
1048 xfs_cluster_write(inode
, page
->index
+ 1, &iomap
, &ioend
,
1049 wbc
, startio
, all_bh
, tlast
);
1053 xfs_submit_ioend(iohead
);
1059 xfs_cancel_ioend(iohead
);
1062 * If it's delalloc and we have nowhere to put it,
1063 * throw it away, unless the lower layers told
1066 if (err
!= -EAGAIN
) {
1068 block_invalidatepage(page
, 0);
1069 ClearPageUptodate(page
);
1075 * writepage: Called from one of two places:
1077 * 1. we are flushing a delalloc buffer head.
1079 * 2. we are writing out a dirty page. Typically the page dirty
1080 * state is cleared before we get here. In this case is it
1081 * conceivable we have no buffer heads.
1083 * For delalloc space on the page we need to allocate space and
1084 * flush it. For unmapped buffer heads on the page we should
1085 * allocate space if the page is uptodate. For any other dirty
1086 * buffer heads on the page we should flush them.
1088 * If we detect that a transaction would be required to flush
1089 * the page, we have to check the process flags first, if we
1090 * are already in a transaction or disk I/O during allocations
1091 * is off, we need to fail the writepage and redirty the page.
1097 struct writeback_control
*wbc
)
1101 int delalloc
, unmapped
, unwritten
;
1102 struct inode
*inode
= page
->mapping
->host
;
1104 xfs_page_trace(XFS_WRITEPAGE_ENTER
, inode
, page
, 0);
1107 * We need a transaction if:
1108 * 1. There are delalloc buffers on the page
1109 * 2. The page is uptodate and we have unmapped buffers
1110 * 3. The page is uptodate and we have no buffers
1111 * 4. There are unwritten buffers on the page
1114 if (!page_has_buffers(page
)) {
1118 xfs_count_page_state(page
, &delalloc
, &unmapped
, &unwritten
);
1119 if (!PageUptodate(page
))
1121 need_trans
= delalloc
+ unmapped
+ unwritten
;
1125 * If we need a transaction and the process flags say
1126 * we are already in a transaction, or no IO is allowed
1127 * then mark the page dirty again and leave the page
1130 if (PFLAGS_TEST_FSTRANS() && need_trans
)
1134 * Delay hooking up buffer heads until we have
1135 * made our go/no-go decision.
1137 if (!page_has_buffers(page
))
1138 create_empty_buffers(page
, 1 << inode
->i_blkbits
, 0);
1141 * Convert delayed allocate, unwritten or unmapped space
1142 * to real space and flush out to disk.
1144 error
= xfs_page_state_convert(inode
, page
, wbc
, 1, unmapped
);
1145 if (error
== -EAGAIN
)
1147 if (unlikely(error
< 0))
1153 redirty_page_for_writepage(wbc
, page
);
1162 * Called to move a page into cleanable state - and from there
1163 * to be released. Possibly the page is already clean. We always
1164 * have buffer heads in this call.
1166 * Returns 0 if the page is ok to release, 1 otherwise.
1168 * Possible scenarios are:
1170 * 1. We are being called to release a page which has been written
1171 * to via regular I/O. buffer heads will be dirty and possibly
1172 * delalloc. If no delalloc buffer heads in this case then we
1173 * can just return zero.
1175 * 2. We are called to release a page which has been written via
1176 * mmap, all we need to do is ensure there is no delalloc
1177 * state in the buffer heads, if not we can let the caller
1178 * free them and we should come back later via writepage.
1185 struct inode
*inode
= page
->mapping
->host
;
1186 int dirty
, delalloc
, unmapped
, unwritten
;
1187 struct writeback_control wbc
= {
1188 .sync_mode
= WB_SYNC_ALL
,
1192 xfs_page_trace(XFS_RELEASEPAGE_ENTER
, inode
, page
, gfp_mask
);
1194 if (!page_has_buffers(page
))
1197 xfs_count_page_state(page
, &delalloc
, &unmapped
, &unwritten
);
1198 if (!delalloc
&& !unwritten
)
1201 if (!(gfp_mask
& __GFP_FS
))
1204 /* If we are already inside a transaction or the thread cannot
1205 * do I/O, we cannot release this page.
1207 if (PFLAGS_TEST_FSTRANS())
1211 * Convert delalloc space to real space, do not flush the
1212 * data out to disk, that will be done by the caller.
1213 * Never need to allocate space here - we will always
1214 * come back to writepage in that case.
1216 dirty
= xfs_page_state_convert(inode
, page
, &wbc
, 0, 0);
1217 if (dirty
== 0 && !unwritten
)
1222 return try_to_free_buffers(page
);
1227 struct inode
*inode
,
1229 unsigned long blocks
,
1230 struct buffer_head
*bh_result
,
1233 bmapi_flags_t flags
)
1235 vnode_t
*vp
= vn_from_inode(inode
);
1242 offset
= (xfs_off_t
)iblock
<< inode
->i_blkbits
;
1244 size
= (ssize_t
) min_t(xfs_off_t
, LONG_MAX
,
1245 (xfs_off_t
)blocks
<< inode
->i_blkbits
);
1247 size
= 1 << inode
->i_blkbits
;
1249 VOP_BMAP(vp
, offset
, size
,
1250 create
? flags
: BMAPI_READ
, &iomap
, &retpbbm
, error
);
1257 if (iomap
.iomap_bn
!= IOMAP_DADDR_NULL
) {
1259 * For unwritten extents do not report a disk address on
1260 * the read case (treat as if we're reading into a hole).
1262 if (create
|| !(iomap
.iomap_flags
& IOMAP_UNWRITTEN
)) {
1263 xfs_map_buffer(bh_result
, &iomap
, offset
,
1266 if (create
&& (iomap
.iomap_flags
& IOMAP_UNWRITTEN
)) {
1268 bh_result
->b_private
= inode
;
1269 set_buffer_unwritten(bh_result
);
1270 set_buffer_delay(bh_result
);
1274 /* If this is a realtime file, data might be on a new device */
1275 bh_result
->b_bdev
= iomap
.iomap_target
->bt_bdev
;
1277 /* If we previously allocated a block out beyond eof and
1278 * we are now coming back to use it then we will need to
1279 * flag it as new even if it has a disk address.
1282 ((!buffer_mapped(bh_result
) && !buffer_uptodate(bh_result
)) ||
1283 (offset
>= i_size_read(inode
)) || (iomap
.iomap_flags
& IOMAP_NEW
)))
1284 set_buffer_new(bh_result
);
1286 if (iomap
.iomap_flags
& IOMAP_DELAY
) {
1289 set_buffer_uptodate(bh_result
);
1290 set_buffer_mapped(bh_result
);
1291 set_buffer_delay(bh_result
);
1296 ASSERT(iomap
.iomap_bsize
- iomap
.iomap_delta
> 0);
1297 offset
= min_t(xfs_off_t
,
1298 iomap
.iomap_bsize
- iomap
.iomap_delta
,
1299 (xfs_off_t
)blocks
<< inode
->i_blkbits
);
1300 bh_result
->b_size
= (u32
) min_t(xfs_off_t
, UINT_MAX
, offset
);
1308 struct inode
*inode
,
1310 struct buffer_head
*bh_result
,
1313 return __xfs_get_block(inode
, iblock
,
1314 bh_result
->b_size
>> inode
->i_blkbits
,
1315 bh_result
, create
, 0, BMAPI_WRITE
);
1319 xfs_get_blocks_direct(
1320 struct inode
*inode
,
1322 struct buffer_head
*bh_result
,
1325 return __xfs_get_block(inode
, iblock
,
1326 bh_result
->b_size
>> inode
->i_blkbits
,
1327 bh_result
, create
, 1, BMAPI_WRITE
|BMAPI_DIRECT
);
1337 xfs_ioend_t
*ioend
= iocb
->private;
1340 * Non-NULL private data means we need to issue a transaction to
1341 * convert a range from unwritten to written extents. This needs
1342 * to happen from process context but aio+dio I/O completion
1343 * happens from irq context so we need to defer it to a workqueue.
1344 * This is not necessary for synchronous direct I/O, but we do
1345 * it anyway to keep the code uniform and simpler.
1347 * The core direct I/O code might be changed to always call the
1348 * completion handler in the future, in which case all this can
1351 if (private && size
> 0) {
1352 ioend
->io_offset
= offset
;
1353 ioend
->io_size
= size
;
1354 xfs_finish_ioend(ioend
);
1357 xfs_destroy_ioend(ioend
);
1361 * blockdev_direct_IO can return an error even after the I/O
1362 * completion handler was called. Thus we need to protect
1363 * against double-freeing.
1365 iocb
->private = NULL
;
1372 const struct iovec
*iov
,
1374 unsigned long nr_segs
)
1376 struct file
*file
= iocb
->ki_filp
;
1377 struct inode
*inode
= file
->f_mapping
->host
;
1378 vnode_t
*vp
= vn_from_inode(inode
);
1384 VOP_BMAP(vp
, offset
, 0, BMAPI_DEVICE
, &iomap
, &maps
, error
);
1388 iocb
->private = xfs_alloc_ioend(inode
, IOMAP_UNWRITTEN
);
1390 ret
= blockdev_direct_IO_own_locking(rw
, iocb
, inode
,
1391 iomap
.iomap_target
->bt_bdev
,
1392 iov
, offset
, nr_segs
,
1393 xfs_get_blocks_direct
,
1396 if (unlikely(ret
<= 0 && iocb
->private))
1397 xfs_destroy_ioend(iocb
->private);
1402 xfs_vm_prepare_write(
1408 return block_prepare_write(page
, from
, to
, xfs_get_block
);
1413 struct address_space
*mapping
,
1416 struct inode
*inode
= (struct inode
*)mapping
->host
;
1417 vnode_t
*vp
= vn_from_inode(inode
);
1420 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
1422 VOP_RWLOCK(vp
, VRWLOCK_READ
);
1423 VOP_FLUSH_PAGES(vp
, (xfs_off_t
)0, -1, 0, FI_REMAPF
, error
);
1424 VOP_RWUNLOCK(vp
, VRWLOCK_READ
);
1425 return generic_block_bmap(mapping
, block
, xfs_get_block
);
1430 struct file
*unused
,
1433 return mpage_readpage(page
, xfs_get_block
);
1438 struct file
*unused
,
1439 struct address_space
*mapping
,
1440 struct list_head
*pages
,
1443 return mpage_readpages(mapping
, pages
, nr_pages
, xfs_get_block
);
1447 xfs_vm_invalidatepage(
1449 unsigned long offset
)
1451 xfs_page_trace(XFS_INVALIDPAGE_ENTER
,
1452 page
->mapping
->host
, page
, offset
);
1453 block_invalidatepage(page
, offset
);
1456 struct address_space_operations xfs_address_space_operations
= {
1457 .readpage
= xfs_vm_readpage
,
1458 .readpages
= xfs_vm_readpages
,
1459 .writepage
= xfs_vm_writepage
,
1460 .sync_page
= block_sync_page
,
1461 .releasepage
= xfs_vm_releasepage
,
1462 .invalidatepage
= xfs_vm_invalidatepage
,
1463 .prepare_write
= xfs_vm_prepare_write
,
1464 .commit_write
= generic_commit_write
,
1465 .bmap
= xfs_vm_bmap
,
1466 .direct_IO
= xfs_vm_direct_IO
,
1467 .migratepage
= buffer_migrate_page
,