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
25 #include "xfs_trans.h"
26 #include "xfs_dmapi.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dir2_sf.h"
32 #include "xfs_attr_sf.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_alloc.h"
36 #include "xfs_btree.h"
37 #include "xfs_error.h"
39 #include "xfs_iomap.h"
40 #include <linux/mpage.h>
41 #include <linux/pagevec.h>
42 #include <linux/writeback.h>
51 struct buffer_head
*bh
, *head
;
53 *delalloc
= *unmapped
= *unwritten
= 0;
55 bh
= head
= page_buffers(page
);
57 if (buffer_uptodate(bh
) && !buffer_mapped(bh
))
59 else if (buffer_unwritten(bh
))
61 else if (buffer_delay(bh
))
63 } while ((bh
= bh
->b_this_page
) != head
);
66 #if defined(XFS_RW_TRACE)
75 bhv_vnode_t
*vp
= vn_from_inode(inode
);
76 loff_t isize
= i_size_read(inode
);
77 loff_t offset
= page_offset(page
);
78 int delalloc
= -1, unmapped
= -1, unwritten
= -1;
80 if (page_has_buffers(page
))
81 xfs_count_page_state(page
, &delalloc
, &unmapped
, &unwritten
);
87 ktrace_enter(ip
->i_rwtrace
,
88 (void *)((unsigned long)tag
),
93 (void *)((unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff)),
94 (void *)((unsigned long)(ip
->i_d
.di_size
& 0xffffffff)),
95 (void *)((unsigned long)((isize
>> 32) & 0xffffffff)),
96 (void *)((unsigned long)(isize
& 0xffffffff)),
97 (void *)((unsigned long)((offset
>> 32) & 0xffffffff)),
98 (void *)((unsigned long)(offset
& 0xffffffff)),
99 (void *)((unsigned long)delalloc
),
100 (void *)((unsigned long)unmapped
),
101 (void *)((unsigned long)unwritten
),
102 (void *)((unsigned long)current_pid()),
106 #define xfs_page_trace(tag, inode, page, pgoff)
110 * Schedule IO completion handling on a xfsdatad if this was
111 * the final hold on this ioend.
117 if (atomic_dec_and_test(&ioend
->io_remaining
))
118 queue_work(xfsdatad_workqueue
, &ioend
->io_work
);
122 * We're now finished for good with this ioend structure.
123 * Update the page state via the associated buffer_heads,
124 * release holds on the inode and bio, and finally free
125 * up memory. Do not use the ioend after this.
131 struct buffer_head
*bh
, *next
;
133 for (bh
= ioend
->io_buffer_head
; bh
; bh
= next
) {
134 next
= bh
->b_private
;
135 bh
->b_end_io(bh
, !ioend
->io_error
);
137 if (unlikely(ioend
->io_error
))
138 vn_ioerror(ioend
->io_vnode
, ioend
->io_error
, __FILE__
,__LINE__
);
139 vn_iowake(ioend
->io_vnode
);
140 mempool_free(ioend
, xfs_ioend_pool
);
144 * Buffered IO write completion for delayed allocate extents.
145 * TODO: Update ondisk isize now that we know the file data
146 * has been flushed (i.e. the notorious "NULL file" problem).
149 xfs_end_bio_delalloc(
150 struct work_struct
*work
)
153 container_of(work
, xfs_ioend_t
, io_work
);
155 xfs_destroy_ioend(ioend
);
159 * Buffered IO write completion for regular, written extents.
163 struct work_struct
*work
)
166 container_of(work
, xfs_ioend_t
, io_work
);
168 xfs_destroy_ioend(ioend
);
172 * IO write completion for unwritten extents.
174 * Issue transactions to convert a buffer range from unwritten
175 * to written extents.
178 xfs_end_bio_unwritten(
179 struct work_struct
*work
)
182 container_of(work
, xfs_ioend_t
, io_work
);
183 bhv_vnode_t
*vp
= ioend
->io_vnode
;
184 xfs_off_t offset
= ioend
->io_offset
;
185 size_t size
= ioend
->io_size
;
187 if (likely(!ioend
->io_error
))
188 bhv_vop_bmap(vp
, offset
, size
, BMAPI_UNWRITTEN
, NULL
, NULL
);
189 xfs_destroy_ioend(ioend
);
193 * Allocate and initialise an IO completion structure.
194 * We need to track unwritten extent write completion here initially.
195 * We'll need to extend this for updating the ondisk inode size later
205 ioend
= mempool_alloc(xfs_ioend_pool
, GFP_NOFS
);
208 * Set the count to 1 initially, which will prevent an I/O
209 * completion callback from happening before we have started
210 * all the I/O from calling the completion routine too early.
212 atomic_set(&ioend
->io_remaining
, 1);
214 ioend
->io_list
= NULL
;
215 ioend
->io_type
= type
;
216 ioend
->io_vnode
= vn_from_inode(inode
);
217 ioend
->io_buffer_head
= NULL
;
218 ioend
->io_buffer_tail
= NULL
;
219 atomic_inc(&ioend
->io_vnode
->v_iocount
);
220 ioend
->io_offset
= 0;
223 if (type
== IOMAP_UNWRITTEN
)
224 INIT_WORK(&ioend
->io_work
, xfs_end_bio_unwritten
);
225 else if (type
== IOMAP_DELAY
)
226 INIT_WORK(&ioend
->io_work
, xfs_end_bio_delalloc
);
228 INIT_WORK(&ioend
->io_work
, xfs_end_bio_written
);
241 bhv_vnode_t
*vp
= vn_from_inode(inode
);
242 int error
, nmaps
= 1;
244 error
= bhv_vop_bmap(vp
, offset
, count
, flags
, mapp
, &nmaps
);
245 if (!error
&& (flags
& (BMAPI_WRITE
|BMAPI_ALLOCATE
)))
255 return offset
>= iomapp
->iomap_offset
&&
256 offset
< iomapp
->iomap_offset
+ iomapp
->iomap_bsize
;
260 * BIO completion handler for buffered IO.
265 unsigned int bytes_done
,
268 xfs_ioend_t
*ioend
= bio
->bi_private
;
273 ASSERT(atomic_read(&bio
->bi_cnt
) >= 1);
274 ioend
->io_error
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
) ? 0 : error
;
276 /* Toss bio and pass work off to an xfsdatad thread */
277 bio
->bi_private
= NULL
;
278 bio
->bi_end_io
= NULL
;
281 xfs_finish_ioend(ioend
);
286 xfs_submit_ioend_bio(
290 atomic_inc(&ioend
->io_remaining
);
292 bio
->bi_private
= ioend
;
293 bio
->bi_end_io
= xfs_end_bio
;
295 submit_bio(WRITE
, bio
);
296 ASSERT(!bio_flagged(bio
, BIO_EOPNOTSUPP
));
302 struct buffer_head
*bh
)
305 int nvecs
= bio_get_nr_vecs(bh
->b_bdev
);
308 bio
= bio_alloc(GFP_NOIO
, nvecs
);
312 ASSERT(bio
->bi_private
== NULL
);
313 bio
->bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
314 bio
->bi_bdev
= bh
->b_bdev
;
320 xfs_start_buffer_writeback(
321 struct buffer_head
*bh
)
323 ASSERT(buffer_mapped(bh
));
324 ASSERT(buffer_locked(bh
));
325 ASSERT(!buffer_delay(bh
));
326 ASSERT(!buffer_unwritten(bh
));
328 mark_buffer_async_write(bh
);
329 set_buffer_uptodate(bh
);
330 clear_buffer_dirty(bh
);
334 xfs_start_page_writeback(
336 struct writeback_control
*wbc
,
340 ASSERT(PageLocked(page
));
341 ASSERT(!PageWriteback(page
));
343 clear_page_dirty_for_io(page
);
344 set_page_writeback(page
);
347 end_page_writeback(page
);
348 wbc
->pages_skipped
++; /* We didn't write this page */
352 static inline int bio_add_buffer(struct bio
*bio
, struct buffer_head
*bh
)
354 return bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
358 * Submit all of the bios for all of the ioends we have saved up, covering the
359 * initial writepage page and also any probed pages.
361 * Because we may have multiple ioends spanning a page, we need to start
362 * writeback on all the buffers before we submit them for I/O. If we mark the
363 * buffers as we got, then we can end up with a page that only has buffers
364 * marked async write and I/O complete on can occur before we mark the other
365 * buffers async write.
367 * The end result of this is that we trip a bug in end_page_writeback() because
368 * we call it twice for the one page as the code in end_buffer_async_write()
369 * assumes that all buffers on the page are started at the same time.
371 * The fix is two passes across the ioend list - one to start writeback on the
372 * buffer_heads, and then submit them for I/O on the second pass.
378 xfs_ioend_t
*head
= ioend
;
380 struct buffer_head
*bh
;
382 sector_t lastblock
= 0;
384 /* Pass 1 - start writeback */
386 next
= ioend
->io_list
;
387 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
388 xfs_start_buffer_writeback(bh
);
390 } while ((ioend
= next
) != NULL
);
392 /* Pass 2 - submit I/O */
395 next
= ioend
->io_list
;
398 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
402 bio
= xfs_alloc_ioend_bio(bh
);
403 } else if (bh
->b_blocknr
!= lastblock
+ 1) {
404 xfs_submit_ioend_bio(ioend
, bio
);
408 if (bio_add_buffer(bio
, bh
) != bh
->b_size
) {
409 xfs_submit_ioend_bio(ioend
, bio
);
413 lastblock
= bh
->b_blocknr
;
416 xfs_submit_ioend_bio(ioend
, bio
);
417 xfs_finish_ioend(ioend
);
418 } while ((ioend
= next
) != NULL
);
422 * Cancel submission of all buffer_heads so far in this endio.
423 * Toss the endio too. Only ever called for the initial page
424 * in a writepage request, so only ever one page.
431 struct buffer_head
*bh
, *next_bh
;
434 next
= ioend
->io_list
;
435 bh
= ioend
->io_buffer_head
;
437 next_bh
= bh
->b_private
;
438 clear_buffer_async_write(bh
);
440 } while ((bh
= next_bh
) != NULL
);
442 vn_iowake(ioend
->io_vnode
);
443 mempool_free(ioend
, xfs_ioend_pool
);
444 } while ((ioend
= next
) != NULL
);
448 * Test to see if we've been building up a completion structure for
449 * earlier buffers -- if so, we try to append to this ioend if we
450 * can, otherwise we finish off any current ioend and start another.
451 * Return true if we've finished the given ioend.
456 struct buffer_head
*bh
,
459 xfs_ioend_t
**result
,
462 xfs_ioend_t
*ioend
= *result
;
464 if (!ioend
|| need_ioend
|| type
!= ioend
->io_type
) {
465 xfs_ioend_t
*previous
= *result
;
467 ioend
= xfs_alloc_ioend(inode
, type
);
468 ioend
->io_offset
= offset
;
469 ioend
->io_buffer_head
= bh
;
470 ioend
->io_buffer_tail
= bh
;
472 previous
->io_list
= ioend
;
475 ioend
->io_buffer_tail
->b_private
= bh
;
476 ioend
->io_buffer_tail
= bh
;
479 bh
->b_private
= NULL
;
480 ioend
->io_size
+= bh
->b_size
;
485 struct buffer_head
*bh
,
492 ASSERT(mp
->iomap_bn
!= IOMAP_DADDR_NULL
);
494 bn
= (mp
->iomap_bn
>> (block_bits
- BBSHIFT
)) +
495 ((offset
- mp
->iomap_offset
) >> block_bits
);
497 ASSERT(bn
|| (mp
->iomap_flags
& IOMAP_REALTIME
));
500 set_buffer_mapped(bh
);
505 struct buffer_head
*bh
,
510 ASSERT(!(iomapp
->iomap_flags
& IOMAP_HOLE
));
511 ASSERT(!(iomapp
->iomap_flags
& IOMAP_DELAY
));
514 xfs_map_buffer(bh
, iomapp
, offset
, block_bits
);
515 bh
->b_bdev
= iomapp
->iomap_target
->bt_bdev
;
516 set_buffer_mapped(bh
);
517 clear_buffer_delay(bh
);
518 clear_buffer_unwritten(bh
);
522 * Look for a page at index that is suitable for clustering.
527 unsigned int pg_offset
,
532 if (PageWriteback(page
))
535 if (page
->mapping
&& PageDirty(page
)) {
536 if (page_has_buffers(page
)) {
537 struct buffer_head
*bh
, *head
;
539 bh
= head
= page_buffers(page
);
541 if (!buffer_uptodate(bh
))
543 if (mapped
!= buffer_mapped(bh
))
546 if (ret
>= pg_offset
)
548 } while ((bh
= bh
->b_this_page
) != head
);
550 ret
= mapped
? 0 : PAGE_CACHE_SIZE
;
559 struct page
*startpage
,
560 struct buffer_head
*bh
,
561 struct buffer_head
*head
,
565 pgoff_t tindex
, tlast
, tloff
;
569 /* First sum forwards in this page */
571 if (!buffer_uptodate(bh
) || (mapped
!= buffer_mapped(bh
)))
574 } while ((bh
= bh
->b_this_page
) != head
);
576 /* if we reached the end of the page, sum forwards in following pages */
577 tlast
= i_size_read(inode
) >> PAGE_CACHE_SHIFT
;
578 tindex
= startpage
->index
+ 1;
580 /* Prune this back to avoid pathological behavior */
581 tloff
= min(tlast
, startpage
->index
+ 64);
583 pagevec_init(&pvec
, 0);
584 while (!done
&& tindex
<= tloff
) {
585 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
587 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
590 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
591 struct page
*page
= pvec
.pages
[i
];
592 size_t pg_offset
, len
= 0;
594 if (tindex
== tlast
) {
596 i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1);
602 pg_offset
= PAGE_CACHE_SIZE
;
604 if (page
->index
== tindex
&& !TestSetPageLocked(page
)) {
605 len
= xfs_probe_page(page
, pg_offset
, mapped
);
618 pagevec_release(&pvec
);
626 * Test if a given page is suitable for writing as part of an unwritten
627 * or delayed allocate extent.
634 if (PageWriteback(page
))
637 if (page
->mapping
&& page_has_buffers(page
)) {
638 struct buffer_head
*bh
, *head
;
641 bh
= head
= page_buffers(page
);
643 if (buffer_unwritten(bh
))
644 acceptable
= (type
== IOMAP_UNWRITTEN
);
645 else if (buffer_delay(bh
))
646 acceptable
= (type
== IOMAP_DELAY
);
647 else if (buffer_dirty(bh
) && buffer_mapped(bh
))
648 acceptable
= (type
== 0);
651 } while ((bh
= bh
->b_this_page
) != head
);
661 * Allocate & map buffers for page given the extent map. Write it out.
662 * except for the original page of a writepage, this is called on
663 * delalloc/unwritten pages only, for the original page it is possible
664 * that the page has no mapping at all.
672 xfs_ioend_t
**ioendp
,
673 struct writeback_control
*wbc
,
677 struct buffer_head
*bh
, *head
;
678 xfs_off_t end_offset
;
679 unsigned long p_offset
;
681 int bbits
= inode
->i_blkbits
;
683 int count
= 0, done
= 0, uptodate
= 1;
684 xfs_off_t offset
= page_offset(page
);
686 if (page
->index
!= tindex
)
688 if (TestSetPageLocked(page
))
690 if (PageWriteback(page
))
691 goto fail_unlock_page
;
692 if (page
->mapping
!= inode
->i_mapping
)
693 goto fail_unlock_page
;
694 if (!xfs_is_delayed_page(page
, (*ioendp
)->io_type
))
695 goto fail_unlock_page
;
698 * page_dirty is initially a count of buffers on the page before
699 * EOF and is decremented as we move each into a cleanable state.
703 * End offset is the highest offset that this page should represent.
704 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
705 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
706 * hence give us the correct page_dirty count. On any other page,
707 * it will be zero and in that case we need page_dirty to be the
708 * count of buffers on the page.
710 end_offset
= min_t(unsigned long long,
711 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
714 len
= 1 << inode
->i_blkbits
;
715 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
717 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
718 page_dirty
= p_offset
/ len
;
720 bh
= head
= page_buffers(page
);
722 if (offset
>= end_offset
)
724 if (!buffer_uptodate(bh
))
726 if (!(PageUptodate(page
) || buffer_uptodate(bh
))) {
731 if (buffer_unwritten(bh
) || buffer_delay(bh
)) {
732 if (buffer_unwritten(bh
))
733 type
= IOMAP_UNWRITTEN
;
737 if (!xfs_iomap_valid(mp
, offset
)) {
742 ASSERT(!(mp
->iomap_flags
& IOMAP_HOLE
));
743 ASSERT(!(mp
->iomap_flags
& IOMAP_DELAY
));
745 xfs_map_at_offset(bh
, offset
, bbits
, mp
);
747 xfs_add_to_ioend(inode
, bh
, offset
,
750 set_buffer_dirty(bh
);
752 mark_buffer_dirty(bh
);
758 if (buffer_mapped(bh
) && all_bh
&& startio
) {
760 xfs_add_to_ioend(inode
, bh
, offset
,
768 } while (offset
+= len
, (bh
= bh
->b_this_page
) != head
);
770 if (uptodate
&& bh
== head
)
771 SetPageUptodate(page
);
775 struct backing_dev_info
*bdi
;
777 bdi
= inode
->i_mapping
->backing_dev_info
;
779 if (bdi_write_congested(bdi
)) {
780 wbc
->encountered_congestion
= 1;
782 } else if (wbc
->nr_to_write
<= 0) {
786 xfs_start_page_writeback(page
, wbc
, !page_dirty
, count
);
797 * Convert & write out a cluster of pages in the same extent as defined
798 * by mp and following the start page.
805 xfs_ioend_t
**ioendp
,
806 struct writeback_control
*wbc
,
814 pagevec_init(&pvec
, 0);
815 while (!done
&& tindex
<= tlast
) {
816 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
818 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
821 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
822 done
= xfs_convert_page(inode
, pvec
.pages
[i
], tindex
++,
823 iomapp
, ioendp
, wbc
, startio
, all_bh
);
828 pagevec_release(&pvec
);
834 * Calling this without startio set means we are being asked to make a dirty
835 * page ready for freeing it's buffers. When called with startio set then
836 * we are coming from writepage.
838 * When called with startio set it is important that we write the WHOLE
840 * The bh->b_state's cannot know if any of the blocks or which block for
841 * that matter are dirty due to mmap writes, and therefore bh uptodate is
842 * only valid if the page itself isn't completely uptodate. Some layers
843 * may clear the page dirty flag prior to calling write page, under the
844 * assumption the entire page will be written out; by not writing out the
845 * whole page the page can be reused before all valid dirty data is
846 * written out. Note: in the case of a page that has been dirty'd by
847 * mapwrite and but partially setup by block_prepare_write the
848 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
849 * valid state, thus the whole page must be written out thing.
853 xfs_page_state_convert(
856 struct writeback_control
*wbc
,
858 int unmapped
) /* also implies page uptodate */
860 struct buffer_head
*bh
, *head
;
862 xfs_ioend_t
*ioend
= NULL
, *iohead
= NULL
;
864 unsigned long p_offset
= 0;
866 __uint64_t end_offset
;
867 pgoff_t end_index
, last_index
, tlast
;
869 int flags
, err
, iomap_valid
= 0, uptodate
= 1;
870 int page_dirty
, count
= 0;
872 int all_bh
= unmapped
;
875 if (wbc
->sync_mode
== WB_SYNC_NONE
&& wbc
->nonblocking
)
876 trylock
|= BMAPI_TRYLOCK
;
879 /* Is this page beyond the end of the file? */
880 offset
= i_size_read(inode
);
881 end_index
= offset
>> PAGE_CACHE_SHIFT
;
882 last_index
= (offset
- 1) >> PAGE_CACHE_SHIFT
;
883 if (page
->index
>= end_index
) {
884 if ((page
->index
>= end_index
+ 1) ||
885 !(i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1))) {
893 * page_dirty is initially a count of buffers on the page before
894 * EOF and is decremented as we move each into a cleanable state.
898 * End offset is the highest offset that this page should represent.
899 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
900 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
901 * hence give us the correct page_dirty count. On any other page,
902 * it will be zero and in that case we need page_dirty to be the
903 * count of buffers on the page.
905 end_offset
= min_t(unsigned long long,
906 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
, offset
);
907 len
= 1 << inode
->i_blkbits
;
908 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
910 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
911 page_dirty
= p_offset
/ len
;
913 bh
= head
= page_buffers(page
);
914 offset
= page_offset(page
);
918 /* TODO: cleanup count and page_dirty */
921 if (offset
>= end_offset
)
923 if (!buffer_uptodate(bh
))
925 if (!(PageUptodate(page
) || buffer_uptodate(bh
)) && !startio
) {
927 * the iomap is actually still valid, but the ioend
928 * isn't. shouldn't happen too often.
935 iomap_valid
= xfs_iomap_valid(&iomap
, offset
);
938 * First case, map an unwritten extent and prepare for
939 * extent state conversion transaction on completion.
941 * Second case, allocate space for a delalloc buffer.
942 * We can return EAGAIN here in the release page case.
944 * Third case, an unmapped buffer was found, and we are
945 * in a path where we need to write the whole page out.
947 if (buffer_unwritten(bh
) || buffer_delay(bh
) ||
948 ((buffer_uptodate(bh
) || PageUptodate(page
)) &&
949 !buffer_mapped(bh
) && (unmapped
|| startio
))) {
951 * Make sure we don't use a read-only iomap
953 if (flags
== BMAPI_READ
)
956 if (buffer_unwritten(bh
)) {
957 type
= IOMAP_UNWRITTEN
;
958 flags
= BMAPI_WRITE
| BMAPI_IGNSTATE
;
959 } else if (buffer_delay(bh
)) {
961 flags
= BMAPI_ALLOCATE
| trylock
;
964 flags
= BMAPI_WRITE
| BMAPI_MMAP
;
968 if (type
== IOMAP_NEW
) {
969 size
= xfs_probe_cluster(inode
,
975 err
= xfs_map_blocks(inode
, offset
, size
,
979 iomap_valid
= xfs_iomap_valid(&iomap
, offset
);
982 xfs_map_at_offset(bh
, offset
,
983 inode
->i_blkbits
, &iomap
);
985 xfs_add_to_ioend(inode
, bh
, offset
,
989 set_buffer_dirty(bh
);
991 mark_buffer_dirty(bh
);
996 } else if (buffer_uptodate(bh
) && startio
) {
998 * we got here because the buffer is already mapped.
999 * That means it must already have extents allocated
1000 * underneath it. Map the extent by reading it.
1002 if (!iomap_valid
|| type
!= 0) {
1004 size
= xfs_probe_cluster(inode
, page
, bh
,
1006 err
= xfs_map_blocks(inode
, offset
, size
,
1010 iomap_valid
= xfs_iomap_valid(&iomap
, offset
);
1014 if (!test_and_set_bit(BH_Lock
, &bh
->b_state
)) {
1015 ASSERT(buffer_mapped(bh
));
1018 xfs_add_to_ioend(inode
, bh
, offset
, type
,
1019 &ioend
, !iomap_valid
);
1025 } else if ((buffer_uptodate(bh
) || PageUptodate(page
)) &&
1026 (unmapped
|| startio
)) {
1033 } while (offset
+= len
, ((bh
= bh
->b_this_page
) != head
));
1035 if (uptodate
&& bh
== head
)
1036 SetPageUptodate(page
);
1039 xfs_start_page_writeback(page
, wbc
, 1, count
);
1041 if (ioend
&& iomap_valid
) {
1042 offset
= (iomap
.iomap_offset
+ iomap
.iomap_bsize
- 1) >>
1044 tlast
= min_t(pgoff_t
, offset
, last_index
);
1045 xfs_cluster_write(inode
, page
->index
+ 1, &iomap
, &ioend
,
1046 wbc
, startio
, all_bh
, tlast
);
1050 xfs_submit_ioend(iohead
);
1056 xfs_cancel_ioend(iohead
);
1059 * If it's delalloc and we have nowhere to put it,
1060 * throw it away, unless the lower layers told
1063 if (err
!= -EAGAIN
) {
1065 block_invalidatepage(page
, 0);
1066 ClearPageUptodate(page
);
1072 * writepage: Called from one of two places:
1074 * 1. we are flushing a delalloc buffer head.
1076 * 2. we are writing out a dirty page. Typically the page dirty
1077 * state is cleared before we get here. In this case is it
1078 * conceivable we have no buffer heads.
1080 * For delalloc space on the page we need to allocate space and
1081 * flush it. For unmapped buffer heads on the page we should
1082 * allocate space if the page is uptodate. For any other dirty
1083 * buffer heads on the page we should flush them.
1085 * If we detect that a transaction would be required to flush
1086 * the page, we have to check the process flags first, if we
1087 * are already in a transaction or disk I/O during allocations
1088 * is off, we need to fail the writepage and redirty the page.
1094 struct writeback_control
*wbc
)
1098 int delalloc
, unmapped
, unwritten
;
1099 struct inode
*inode
= page
->mapping
->host
;
1101 xfs_page_trace(XFS_WRITEPAGE_ENTER
, inode
, page
, 0);
1104 * We need a transaction if:
1105 * 1. There are delalloc buffers on the page
1106 * 2. The page is uptodate and we have unmapped buffers
1107 * 3. The page is uptodate and we have no buffers
1108 * 4. There are unwritten buffers on the page
1111 if (!page_has_buffers(page
)) {
1115 xfs_count_page_state(page
, &delalloc
, &unmapped
, &unwritten
);
1116 if (!PageUptodate(page
))
1118 need_trans
= delalloc
+ unmapped
+ unwritten
;
1122 * If we need a transaction and the process flags say
1123 * we are already in a transaction, or no IO is allowed
1124 * then mark the page dirty again and leave the page
1127 if (current_test_flags(PF_FSTRANS
) && need_trans
)
1131 * Delay hooking up buffer heads until we have
1132 * made our go/no-go decision.
1134 if (!page_has_buffers(page
))
1135 create_empty_buffers(page
, 1 << inode
->i_blkbits
, 0);
1138 * Convert delayed allocate, unwritten or unmapped space
1139 * to real space and flush out to disk.
1141 error
= xfs_page_state_convert(inode
, page
, wbc
, 1, unmapped
);
1142 if (error
== -EAGAIN
)
1144 if (unlikely(error
< 0))
1150 redirty_page_for_writepage(wbc
, page
);
1160 struct address_space
*mapping
,
1161 struct writeback_control
*wbc
)
1163 struct bhv_vnode
*vp
= vn_from_inode(mapping
->host
);
1167 return generic_writepages(mapping
, wbc
);
1171 * Called to move a page into cleanable state - and from there
1172 * to be released. Possibly the page is already clean. We always
1173 * have buffer heads in this call.
1175 * Returns 0 if the page is ok to release, 1 otherwise.
1177 * Possible scenarios are:
1179 * 1. We are being called to release a page which has been written
1180 * to via regular I/O. buffer heads will be dirty and possibly
1181 * delalloc. If no delalloc buffer heads in this case then we
1182 * can just return zero.
1184 * 2. We are called to release a page which has been written via
1185 * mmap, all we need to do is ensure there is no delalloc
1186 * state in the buffer heads, if not we can let the caller
1187 * free them and we should come back later via writepage.
1194 struct inode
*inode
= page
->mapping
->host
;
1195 int dirty
, delalloc
, unmapped
, unwritten
;
1196 struct writeback_control wbc
= {
1197 .sync_mode
= WB_SYNC_ALL
,
1201 xfs_page_trace(XFS_RELEASEPAGE_ENTER
, inode
, page
, 0);
1203 if (!page_has_buffers(page
))
1206 xfs_count_page_state(page
, &delalloc
, &unmapped
, &unwritten
);
1207 if (!delalloc
&& !unwritten
)
1210 if (!(gfp_mask
& __GFP_FS
))
1213 /* If we are already inside a transaction or the thread cannot
1214 * do I/O, we cannot release this page.
1216 if (current_test_flags(PF_FSTRANS
))
1220 * Convert delalloc space to real space, do not flush the
1221 * data out to disk, that will be done by the caller.
1222 * Never need to allocate space here - we will always
1223 * come back to writepage in that case.
1225 dirty
= xfs_page_state_convert(inode
, page
, &wbc
, 0, 0);
1226 if (dirty
== 0 && !unwritten
)
1231 return try_to_free_buffers(page
);
1236 struct inode
*inode
,
1238 struct buffer_head
*bh_result
,
1241 bmapi_flags_t flags
)
1243 bhv_vnode_t
*vp
= vn_from_inode(inode
);
1250 offset
= (xfs_off_t
)iblock
<< inode
->i_blkbits
;
1251 ASSERT(bh_result
->b_size
>= (1 << inode
->i_blkbits
));
1252 size
= bh_result
->b_size
;
1253 error
= bhv_vop_bmap(vp
, offset
, size
,
1254 create
? flags
: BMAPI_READ
, &iomap
, &niomap
);
1260 if (iomap
.iomap_bn
!= IOMAP_DADDR_NULL
) {
1262 * For unwritten extents do not report a disk address on
1263 * the read case (treat as if we're reading into a hole).
1265 if (create
|| !(iomap
.iomap_flags
& IOMAP_UNWRITTEN
)) {
1266 xfs_map_buffer(bh_result
, &iomap
, offset
,
1269 if (create
&& (iomap
.iomap_flags
& IOMAP_UNWRITTEN
)) {
1271 bh_result
->b_private
= inode
;
1272 set_buffer_unwritten(bh_result
);
1277 * If this is a realtime file, data may be on a different device.
1278 * to that pointed to from the buffer_head b_bdev currently.
1280 bh_result
->b_bdev
= iomap
.iomap_target
->bt_bdev
;
1283 * If we previously allocated a block out beyond eof and we are now
1284 * coming back to use it then we will need to flag it as new even if it
1285 * has a disk address.
1287 * With sub-block writes into unwritten extents we also need to mark
1288 * the buffer as new so that the unwritten parts of the buffer gets
1292 ((!buffer_mapped(bh_result
) && !buffer_uptodate(bh_result
)) ||
1293 (offset
>= i_size_read(inode
)) ||
1294 (iomap
.iomap_flags
& (IOMAP_NEW
|IOMAP_UNWRITTEN
))))
1295 set_buffer_new(bh_result
);
1297 if (iomap
.iomap_flags
& IOMAP_DELAY
) {
1300 set_buffer_uptodate(bh_result
);
1301 set_buffer_mapped(bh_result
);
1302 set_buffer_delay(bh_result
);
1306 if (direct
|| size
> (1 << inode
->i_blkbits
)) {
1307 ASSERT(iomap
.iomap_bsize
- iomap
.iomap_delta
> 0);
1308 offset
= min_t(xfs_off_t
,
1309 iomap
.iomap_bsize
- iomap
.iomap_delta
, size
);
1310 bh_result
->b_size
= (ssize_t
)min_t(xfs_off_t
, LONG_MAX
, offset
);
1318 struct inode
*inode
,
1320 struct buffer_head
*bh_result
,
1323 return __xfs_get_blocks(inode
, iblock
,
1324 bh_result
, create
, 0, BMAPI_WRITE
);
1328 xfs_get_blocks_direct(
1329 struct inode
*inode
,
1331 struct buffer_head
*bh_result
,
1334 return __xfs_get_blocks(inode
, iblock
,
1335 bh_result
, create
, 1, BMAPI_WRITE
|BMAPI_DIRECT
);
1345 xfs_ioend_t
*ioend
= iocb
->private;
1348 * Non-NULL private data means we need to issue a transaction to
1349 * convert a range from unwritten to written extents. This needs
1350 * to happen from process context but aio+dio I/O completion
1351 * happens from irq context so we need to defer it to a workqueue.
1352 * This is not necessary for synchronous direct I/O, but we do
1353 * it anyway to keep the code uniform and simpler.
1355 * The core direct I/O code might be changed to always call the
1356 * completion handler in the future, in which case all this can
1359 if (private && size
> 0) {
1360 ioend
->io_offset
= offset
;
1361 ioend
->io_size
= size
;
1362 xfs_finish_ioend(ioend
);
1364 xfs_destroy_ioend(ioend
);
1368 * blockdev_direct_IO can return an error even after the I/O
1369 * completion handler was called. Thus we need to protect
1370 * against double-freeing.
1372 iocb
->private = NULL
;
1379 const struct iovec
*iov
,
1381 unsigned long nr_segs
)
1383 struct file
*file
= iocb
->ki_filp
;
1384 struct inode
*inode
= file
->f_mapping
->host
;
1385 bhv_vnode_t
*vp
= vn_from_inode(inode
);
1391 error
= bhv_vop_bmap(vp
, offset
, 0, BMAPI_DEVICE
, &iomap
, &maps
);
1395 iocb
->private = xfs_alloc_ioend(inode
, IOMAP_UNWRITTEN
);
1398 ret
= blockdev_direct_IO_own_locking(rw
, iocb
, inode
,
1399 iomap
.iomap_target
->bt_bdev
,
1400 iov
, offset
, nr_segs
,
1401 xfs_get_blocks_direct
,
1404 ret
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
,
1405 iomap
.iomap_target
->bt_bdev
,
1406 iov
, offset
, nr_segs
,
1407 xfs_get_blocks_direct
,
1411 if (unlikely(ret
!= -EIOCBQUEUED
&& iocb
->private))
1412 xfs_destroy_ioend(iocb
->private);
1417 xfs_vm_prepare_write(
1423 return block_prepare_write(page
, from
, to
, xfs_get_blocks
);
1428 struct address_space
*mapping
,
1431 struct inode
*inode
= (struct inode
*)mapping
->host
;
1432 bhv_vnode_t
*vp
= vn_from_inode(inode
);
1434 vn_trace_entry(vp
, __FUNCTION__
, (inst_t
*)__return_address
);
1435 bhv_vop_rwlock(vp
, VRWLOCK_READ
);
1436 bhv_vop_flush_pages(vp
, (xfs_off_t
)0, -1, 0, FI_REMAPF
);
1437 bhv_vop_rwunlock(vp
, VRWLOCK_READ
);
1438 return generic_block_bmap(mapping
, block
, xfs_get_blocks
);
1443 struct file
*unused
,
1446 return mpage_readpage(page
, xfs_get_blocks
);
1451 struct file
*unused
,
1452 struct address_space
*mapping
,
1453 struct list_head
*pages
,
1456 return mpage_readpages(mapping
, pages
, nr_pages
, xfs_get_blocks
);
1460 xfs_vm_invalidatepage(
1462 unsigned long offset
)
1464 xfs_page_trace(XFS_INVALIDPAGE_ENTER
,
1465 page
->mapping
->host
, page
, offset
);
1466 block_invalidatepage(page
, offset
);
1469 const struct address_space_operations xfs_address_space_operations
= {
1470 .readpage
= xfs_vm_readpage
,
1471 .readpages
= xfs_vm_readpages
,
1472 .writepage
= xfs_vm_writepage
,
1473 .writepages
= xfs_vm_writepages
,
1474 .sync_page
= block_sync_page
,
1475 .releasepage
= xfs_vm_releasepage
,
1476 .invalidatepage
= xfs_vm_invalidatepage
,
1477 .prepare_write
= xfs_vm_prepare_write
,
1478 .commit_write
= generic_commit_write
,
1479 .bmap
= xfs_vm_bmap
,
1480 .direct_IO
= xfs_vm_direct_IO
,
1481 .migratepage
= buffer_migrate_page
,