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 "xfs_vnodeops.h"
41 #include <linux/mpage.h>
42 #include <linux/pagevec.h>
43 #include <linux/writeback.h>
52 struct buffer_head
*bh
, *head
;
54 *delalloc
= *unmapped
= *unwritten
= 0;
56 bh
= head
= page_buffers(page
);
58 if (buffer_uptodate(bh
) && !buffer_mapped(bh
))
60 else if (buffer_unwritten(bh
))
62 else if (buffer_delay(bh
))
64 } while ((bh
= bh
->b_this_page
) != head
);
67 #if defined(XFS_RW_TRACE)
76 bhv_vnode_t
*vp
= vn_from_inode(inode
);
77 loff_t isize
= i_size_read(inode
);
78 loff_t offset
= page_offset(page
);
79 int delalloc
= -1, unmapped
= -1, unwritten
= -1;
81 if (page_has_buffers(page
))
82 xfs_count_page_state(page
, &delalloc
, &unmapped
, &unwritten
);
88 ktrace_enter(ip
->i_rwtrace
,
89 (void *)((unsigned long)tag
),
94 (void *)((unsigned long)((ip
->i_d
.di_size
>> 32) & 0xffffffff)),
95 (void *)((unsigned long)(ip
->i_d
.di_size
& 0xffffffff)),
96 (void *)((unsigned long)((isize
>> 32) & 0xffffffff)),
97 (void *)((unsigned long)(isize
& 0xffffffff)),
98 (void *)((unsigned long)((offset
>> 32) & 0xffffffff)),
99 (void *)((unsigned long)(offset
& 0xffffffff)),
100 (void *)((unsigned long)delalloc
),
101 (void *)((unsigned long)unmapped
),
102 (void *)((unsigned long)unwritten
),
103 (void *)((unsigned long)current_pid()),
107 #define xfs_page_trace(tag, inode, page, pgoff)
111 * Schedule IO completion handling on a xfsdatad if this was
112 * the final hold on this ioend. If we are asked to wait,
113 * flush the workqueue.
120 if (atomic_dec_and_test(&ioend
->io_remaining
)) {
121 queue_work(xfsdatad_workqueue
, &ioend
->io_work
);
123 flush_workqueue(xfsdatad_workqueue
);
128 * We're now finished for good with this ioend structure.
129 * Update the page state via the associated buffer_heads,
130 * release holds on the inode and bio, and finally free
131 * up memory. Do not use the ioend after this.
137 struct buffer_head
*bh
, *next
;
139 for (bh
= ioend
->io_buffer_head
; bh
; bh
= next
) {
140 next
= bh
->b_private
;
141 bh
->b_end_io(bh
, !ioend
->io_error
);
143 if (unlikely(ioend
->io_error
))
144 vn_ioerror(ioend
->io_vnode
, ioend
->io_error
, __FILE__
,__LINE__
);
145 vn_iowake(ioend
->io_vnode
);
146 mempool_free(ioend
, xfs_ioend_pool
);
150 * Update on-disk file size now that data has been written to disk.
151 * The current in-memory file size is i_size. If a write is beyond
152 * eof io_new_size will be the intended file size until i_size is
153 * updated. If this write does not extend all the way to the valid
154 * file size then restrict this update to the end of the write.
164 ip
= xfs_vtoi(ioend
->io_vnode
);
168 ASSERT((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
);
169 ASSERT(ioend
->io_type
!= IOMAP_READ
);
171 if (unlikely(ioend
->io_error
))
174 bsize
= ioend
->io_offset
+ ioend
->io_size
;
176 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
178 isize
= MAX(ip
->i_size
, ip
->i_iocore
.io_new_size
);
179 isize
= MIN(isize
, bsize
);
181 if (ip
->i_d
.di_size
< isize
) {
182 ip
->i_d
.di_size
= isize
;
183 ip
->i_update_core
= 1;
184 ip
->i_update_size
= 1;
185 mark_inode_dirty_sync(vn_to_inode(ioend
->io_vnode
));
188 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
192 * Buffered IO write completion for delayed allocate extents.
195 xfs_end_bio_delalloc(
196 struct work_struct
*work
)
199 container_of(work
, xfs_ioend_t
, io_work
);
201 xfs_setfilesize(ioend
);
202 xfs_destroy_ioend(ioend
);
206 * Buffered IO write completion for regular, written extents.
210 struct work_struct
*work
)
213 container_of(work
, xfs_ioend_t
, io_work
);
215 xfs_setfilesize(ioend
);
216 xfs_destroy_ioend(ioend
);
220 * IO write completion for unwritten extents.
222 * Issue transactions to convert a buffer range from unwritten
223 * to written extents.
226 xfs_end_bio_unwritten(
227 struct work_struct
*work
)
230 container_of(work
, xfs_ioend_t
, io_work
);
231 bhv_vnode_t
*vp
= ioend
->io_vnode
;
232 xfs_off_t offset
= ioend
->io_offset
;
233 size_t size
= ioend
->io_size
;
235 if (likely(!ioend
->io_error
)) {
236 xfs_bmap(xfs_vtoi(vp
), offset
, size
,
237 BMAPI_UNWRITTEN
, NULL
, NULL
);
238 xfs_setfilesize(ioend
);
240 xfs_destroy_ioend(ioend
);
244 * IO read completion for regular, written extents.
248 struct work_struct
*work
)
251 container_of(work
, xfs_ioend_t
, io_work
);
253 xfs_destroy_ioend(ioend
);
257 * Allocate and initialise an IO completion structure.
258 * We need to track unwritten extent write completion here initially.
259 * We'll need to extend this for updating the ondisk inode size later
269 ioend
= mempool_alloc(xfs_ioend_pool
, GFP_NOFS
);
272 * Set the count to 1 initially, which will prevent an I/O
273 * completion callback from happening before we have started
274 * all the I/O from calling the completion routine too early.
276 atomic_set(&ioend
->io_remaining
, 1);
278 ioend
->io_list
= NULL
;
279 ioend
->io_type
= type
;
280 ioend
->io_vnode
= vn_from_inode(inode
);
281 ioend
->io_buffer_head
= NULL
;
282 ioend
->io_buffer_tail
= NULL
;
283 atomic_inc(&ioend
->io_vnode
->v_iocount
);
284 ioend
->io_offset
= 0;
287 if (type
== IOMAP_UNWRITTEN
)
288 INIT_WORK(&ioend
->io_work
, xfs_end_bio_unwritten
);
289 else if (type
== IOMAP_DELAY
)
290 INIT_WORK(&ioend
->io_work
, xfs_end_bio_delalloc
);
291 else if (type
== IOMAP_READ
)
292 INIT_WORK(&ioend
->io_work
, xfs_end_bio_read
);
294 INIT_WORK(&ioend
->io_work
, xfs_end_bio_written
);
307 xfs_inode_t
*ip
= XFS_I(inode
);
308 int error
, nmaps
= 1;
310 error
= xfs_bmap(ip
, offset
, count
,
311 flags
, mapp
, &nmaps
);
312 if (!error
&& (flags
& (BMAPI_WRITE
|BMAPI_ALLOCATE
)))
313 xfs_iflags_set(ip
, XFS_IMODIFIED
);
322 return offset
>= iomapp
->iomap_offset
&&
323 offset
< iomapp
->iomap_offset
+ iomapp
->iomap_bsize
;
327 * BIO completion handler for buffered IO.
332 unsigned int bytes_done
,
335 xfs_ioend_t
*ioend
= bio
->bi_private
;
340 ASSERT(atomic_read(&bio
->bi_cnt
) >= 1);
341 ioend
->io_error
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
) ? 0 : error
;
343 /* Toss bio and pass work off to an xfsdatad thread */
344 bio
->bi_private
= NULL
;
345 bio
->bi_end_io
= NULL
;
348 xfs_finish_ioend(ioend
, 0);
353 xfs_submit_ioend_bio(
357 atomic_inc(&ioend
->io_remaining
);
359 bio
->bi_private
= ioend
;
360 bio
->bi_end_io
= xfs_end_bio
;
362 submit_bio(WRITE
, bio
);
363 ASSERT(!bio_flagged(bio
, BIO_EOPNOTSUPP
));
369 struct buffer_head
*bh
)
372 int nvecs
= bio_get_nr_vecs(bh
->b_bdev
);
375 bio
= bio_alloc(GFP_NOIO
, nvecs
);
379 ASSERT(bio
->bi_private
== NULL
);
380 bio
->bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
381 bio
->bi_bdev
= bh
->b_bdev
;
387 xfs_start_buffer_writeback(
388 struct buffer_head
*bh
)
390 ASSERT(buffer_mapped(bh
));
391 ASSERT(buffer_locked(bh
));
392 ASSERT(!buffer_delay(bh
));
393 ASSERT(!buffer_unwritten(bh
));
395 mark_buffer_async_write(bh
);
396 set_buffer_uptodate(bh
);
397 clear_buffer_dirty(bh
);
401 xfs_start_page_writeback(
403 struct writeback_control
*wbc
,
407 ASSERT(PageLocked(page
));
408 ASSERT(!PageWriteback(page
));
410 clear_page_dirty_for_io(page
);
411 set_page_writeback(page
);
414 end_page_writeback(page
);
415 wbc
->pages_skipped
++; /* We didn't write this page */
419 static inline int bio_add_buffer(struct bio
*bio
, struct buffer_head
*bh
)
421 return bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
425 * Submit all of the bios for all of the ioends we have saved up, covering the
426 * initial writepage page and also any probed pages.
428 * Because we may have multiple ioends spanning a page, we need to start
429 * writeback on all the buffers before we submit them for I/O. If we mark the
430 * buffers as we got, then we can end up with a page that only has buffers
431 * marked async write and I/O complete on can occur before we mark the other
432 * buffers async write.
434 * The end result of this is that we trip a bug in end_page_writeback() because
435 * we call it twice for the one page as the code in end_buffer_async_write()
436 * assumes that all buffers on the page are started at the same time.
438 * The fix is two passes across the ioend list - one to start writeback on the
439 * buffer_heads, and then submit them for I/O on the second pass.
445 xfs_ioend_t
*head
= ioend
;
447 struct buffer_head
*bh
;
449 sector_t lastblock
= 0;
451 /* Pass 1 - start writeback */
453 next
= ioend
->io_list
;
454 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
455 xfs_start_buffer_writeback(bh
);
457 } while ((ioend
= next
) != NULL
);
459 /* Pass 2 - submit I/O */
462 next
= ioend
->io_list
;
465 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
469 bio
= xfs_alloc_ioend_bio(bh
);
470 } else if (bh
->b_blocknr
!= lastblock
+ 1) {
471 xfs_submit_ioend_bio(ioend
, bio
);
475 if (bio_add_buffer(bio
, bh
) != bh
->b_size
) {
476 xfs_submit_ioend_bio(ioend
, bio
);
480 lastblock
= bh
->b_blocknr
;
483 xfs_submit_ioend_bio(ioend
, bio
);
484 xfs_finish_ioend(ioend
, 0);
485 } while ((ioend
= next
) != NULL
);
489 * Cancel submission of all buffer_heads so far in this endio.
490 * Toss the endio too. Only ever called for the initial page
491 * in a writepage request, so only ever one page.
498 struct buffer_head
*bh
, *next_bh
;
501 next
= ioend
->io_list
;
502 bh
= ioend
->io_buffer_head
;
504 next_bh
= bh
->b_private
;
505 clear_buffer_async_write(bh
);
507 } while ((bh
= next_bh
) != NULL
);
509 vn_iowake(ioend
->io_vnode
);
510 mempool_free(ioend
, xfs_ioend_pool
);
511 } while ((ioend
= next
) != NULL
);
515 * Test to see if we've been building up a completion structure for
516 * earlier buffers -- if so, we try to append to this ioend if we
517 * can, otherwise we finish off any current ioend and start another.
518 * Return true if we've finished the given ioend.
523 struct buffer_head
*bh
,
526 xfs_ioend_t
**result
,
529 xfs_ioend_t
*ioend
= *result
;
531 if (!ioend
|| need_ioend
|| type
!= ioend
->io_type
) {
532 xfs_ioend_t
*previous
= *result
;
534 ioend
= xfs_alloc_ioend(inode
, type
);
535 ioend
->io_offset
= offset
;
536 ioend
->io_buffer_head
= bh
;
537 ioend
->io_buffer_tail
= bh
;
539 previous
->io_list
= ioend
;
542 ioend
->io_buffer_tail
->b_private
= bh
;
543 ioend
->io_buffer_tail
= bh
;
546 bh
->b_private
= NULL
;
547 ioend
->io_size
+= bh
->b_size
;
552 struct buffer_head
*bh
,
559 ASSERT(mp
->iomap_bn
!= IOMAP_DADDR_NULL
);
561 bn
= (mp
->iomap_bn
>> (block_bits
- BBSHIFT
)) +
562 ((offset
- mp
->iomap_offset
) >> block_bits
);
564 ASSERT(bn
|| (mp
->iomap_flags
& IOMAP_REALTIME
));
567 set_buffer_mapped(bh
);
572 struct buffer_head
*bh
,
577 ASSERT(!(iomapp
->iomap_flags
& IOMAP_HOLE
));
578 ASSERT(!(iomapp
->iomap_flags
& IOMAP_DELAY
));
581 xfs_map_buffer(bh
, iomapp
, offset
, block_bits
);
582 bh
->b_bdev
= iomapp
->iomap_target
->bt_bdev
;
583 set_buffer_mapped(bh
);
584 clear_buffer_delay(bh
);
585 clear_buffer_unwritten(bh
);
589 * Look for a page at index that is suitable for clustering.
594 unsigned int pg_offset
,
599 if (PageWriteback(page
))
602 if (page
->mapping
&& PageDirty(page
)) {
603 if (page_has_buffers(page
)) {
604 struct buffer_head
*bh
, *head
;
606 bh
= head
= page_buffers(page
);
608 if (!buffer_uptodate(bh
))
610 if (mapped
!= buffer_mapped(bh
))
613 if (ret
>= pg_offset
)
615 } while ((bh
= bh
->b_this_page
) != head
);
617 ret
= mapped
? 0 : PAGE_CACHE_SIZE
;
626 struct page
*startpage
,
627 struct buffer_head
*bh
,
628 struct buffer_head
*head
,
632 pgoff_t tindex
, tlast
, tloff
;
636 /* First sum forwards in this page */
638 if (!buffer_uptodate(bh
) || (mapped
!= buffer_mapped(bh
)))
641 } while ((bh
= bh
->b_this_page
) != head
);
643 /* if we reached the end of the page, sum forwards in following pages */
644 tlast
= i_size_read(inode
) >> PAGE_CACHE_SHIFT
;
645 tindex
= startpage
->index
+ 1;
647 /* Prune this back to avoid pathological behavior */
648 tloff
= min(tlast
, startpage
->index
+ 64);
650 pagevec_init(&pvec
, 0);
651 while (!done
&& tindex
<= tloff
) {
652 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
654 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
657 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
658 struct page
*page
= pvec
.pages
[i
];
659 size_t pg_offset
, pg_len
= 0;
661 if (tindex
== tlast
) {
663 i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1);
669 pg_offset
= PAGE_CACHE_SIZE
;
671 if (page
->index
== tindex
&& !TestSetPageLocked(page
)) {
672 pg_len
= xfs_probe_page(page
, pg_offset
, mapped
);
685 pagevec_release(&pvec
);
693 * Test if a given page is suitable for writing as part of an unwritten
694 * or delayed allocate extent.
701 if (PageWriteback(page
))
704 if (page
->mapping
&& page_has_buffers(page
)) {
705 struct buffer_head
*bh
, *head
;
708 bh
= head
= page_buffers(page
);
710 if (buffer_unwritten(bh
))
711 acceptable
= (type
== IOMAP_UNWRITTEN
);
712 else if (buffer_delay(bh
))
713 acceptable
= (type
== IOMAP_DELAY
);
714 else if (buffer_dirty(bh
) && buffer_mapped(bh
))
715 acceptable
= (type
== IOMAP_NEW
);
718 } while ((bh
= bh
->b_this_page
) != head
);
728 * Allocate & map buffers for page given the extent map. Write it out.
729 * except for the original page of a writepage, this is called on
730 * delalloc/unwritten pages only, for the original page it is possible
731 * that the page has no mapping at all.
739 xfs_ioend_t
**ioendp
,
740 struct writeback_control
*wbc
,
744 struct buffer_head
*bh
, *head
;
745 xfs_off_t end_offset
;
746 unsigned long p_offset
;
748 int bbits
= inode
->i_blkbits
;
750 int count
= 0, done
= 0, uptodate
= 1;
751 xfs_off_t offset
= page_offset(page
);
753 if (page
->index
!= tindex
)
755 if (TestSetPageLocked(page
))
757 if (PageWriteback(page
))
758 goto fail_unlock_page
;
759 if (page
->mapping
!= inode
->i_mapping
)
760 goto fail_unlock_page
;
761 if (!xfs_is_delayed_page(page
, (*ioendp
)->io_type
))
762 goto fail_unlock_page
;
765 * page_dirty is initially a count of buffers on the page before
766 * EOF and is decremented as we move each into a cleanable state.
770 * End offset is the highest offset that this page should represent.
771 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
772 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
773 * hence give us the correct page_dirty count. On any other page,
774 * it will be zero and in that case we need page_dirty to be the
775 * count of buffers on the page.
777 end_offset
= min_t(unsigned long long,
778 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
781 len
= 1 << inode
->i_blkbits
;
782 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
784 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
785 page_dirty
= p_offset
/ len
;
787 bh
= head
= page_buffers(page
);
789 if (offset
>= end_offset
)
791 if (!buffer_uptodate(bh
))
793 if (!(PageUptodate(page
) || buffer_uptodate(bh
))) {
798 if (buffer_unwritten(bh
) || buffer_delay(bh
)) {
799 if (buffer_unwritten(bh
))
800 type
= IOMAP_UNWRITTEN
;
804 if (!xfs_iomap_valid(mp
, offset
)) {
809 ASSERT(!(mp
->iomap_flags
& IOMAP_HOLE
));
810 ASSERT(!(mp
->iomap_flags
& IOMAP_DELAY
));
812 xfs_map_at_offset(bh
, offset
, bbits
, mp
);
814 xfs_add_to_ioend(inode
, bh
, offset
,
817 set_buffer_dirty(bh
);
819 mark_buffer_dirty(bh
);
825 if (buffer_mapped(bh
) && all_bh
&& startio
) {
827 xfs_add_to_ioend(inode
, bh
, offset
,
835 } while (offset
+= len
, (bh
= bh
->b_this_page
) != head
);
837 if (uptodate
&& bh
== head
)
838 SetPageUptodate(page
);
842 struct backing_dev_info
*bdi
;
844 bdi
= inode
->i_mapping
->backing_dev_info
;
846 if (bdi_write_congested(bdi
)) {
847 wbc
->encountered_congestion
= 1;
849 } else if (wbc
->nr_to_write
<= 0) {
853 xfs_start_page_writeback(page
, wbc
, !page_dirty
, count
);
864 * Convert & write out a cluster of pages in the same extent as defined
865 * by mp and following the start page.
872 xfs_ioend_t
**ioendp
,
873 struct writeback_control
*wbc
,
881 pagevec_init(&pvec
, 0);
882 while (!done
&& tindex
<= tlast
) {
883 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
885 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
888 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
889 done
= xfs_convert_page(inode
, pvec
.pages
[i
], tindex
++,
890 iomapp
, ioendp
, wbc
, startio
, all_bh
);
895 pagevec_release(&pvec
);
901 * Calling this without startio set means we are being asked to make a dirty
902 * page ready for freeing it's buffers. When called with startio set then
903 * we are coming from writepage.
905 * When called with startio set it is important that we write the WHOLE
907 * The bh->b_state's cannot know if any of the blocks or which block for
908 * that matter are dirty due to mmap writes, and therefore bh uptodate is
909 * only valid if the page itself isn't completely uptodate. Some layers
910 * may clear the page dirty flag prior to calling write page, under the
911 * assumption the entire page will be written out; by not writing out the
912 * whole page the page can be reused before all valid dirty data is
913 * written out. Note: in the case of a page that has been dirty'd by
914 * mapwrite and but partially setup by block_prepare_write the
915 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
916 * valid state, thus the whole page must be written out thing.
920 xfs_page_state_convert(
923 struct writeback_control
*wbc
,
925 int unmapped
) /* also implies page uptodate */
927 struct buffer_head
*bh
, *head
;
929 xfs_ioend_t
*ioend
= NULL
, *iohead
= NULL
;
931 unsigned long p_offset
= 0;
933 __uint64_t end_offset
;
934 pgoff_t end_index
, last_index
, tlast
;
936 int flags
, err
, iomap_valid
= 0, uptodate
= 1;
937 int page_dirty
, count
= 0;
939 int all_bh
= unmapped
;
942 if (wbc
->sync_mode
== WB_SYNC_NONE
&& wbc
->nonblocking
)
943 trylock
|= BMAPI_TRYLOCK
;
946 /* Is this page beyond the end of the file? */
947 offset
= i_size_read(inode
);
948 end_index
= offset
>> PAGE_CACHE_SHIFT
;
949 last_index
= (offset
- 1) >> PAGE_CACHE_SHIFT
;
950 if (page
->index
>= end_index
) {
951 if ((page
->index
>= end_index
+ 1) ||
952 !(i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1))) {
960 * page_dirty is initially a count of buffers on the page before
961 * EOF and is decremented as we move each into a cleanable state.
965 * End offset is the highest offset that this page should represent.
966 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
967 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
968 * hence give us the correct page_dirty count. On any other page,
969 * it will be zero and in that case we need page_dirty to be the
970 * count of buffers on the page.
972 end_offset
= min_t(unsigned long long,
973 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
, offset
);
974 len
= 1 << inode
->i_blkbits
;
975 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
977 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
978 page_dirty
= p_offset
/ len
;
980 bh
= head
= page_buffers(page
);
981 offset
= page_offset(page
);
985 /* TODO: cleanup count and page_dirty */
988 if (offset
>= end_offset
)
990 if (!buffer_uptodate(bh
))
992 if (!(PageUptodate(page
) || buffer_uptodate(bh
)) && !startio
) {
994 * the iomap is actually still valid, but the ioend
995 * isn't. shouldn't happen too often.
1002 iomap_valid
= xfs_iomap_valid(&iomap
, offset
);
1005 * First case, map an unwritten extent and prepare for
1006 * extent state conversion transaction on completion.
1008 * Second case, allocate space for a delalloc buffer.
1009 * We can return EAGAIN here in the release page case.
1011 * Third case, an unmapped buffer was found, and we are
1012 * in a path where we need to write the whole page out.
1014 if (buffer_unwritten(bh
) || buffer_delay(bh
) ||
1015 ((buffer_uptodate(bh
) || PageUptodate(page
)) &&
1016 !buffer_mapped(bh
) && (unmapped
|| startio
))) {
1020 * Make sure we don't use a read-only iomap
1022 if (flags
== BMAPI_READ
)
1025 if (buffer_unwritten(bh
)) {
1026 type
= IOMAP_UNWRITTEN
;
1027 flags
= BMAPI_WRITE
| BMAPI_IGNSTATE
;
1028 } else if (buffer_delay(bh
)) {
1030 flags
= BMAPI_ALLOCATE
| trylock
;
1033 flags
= BMAPI_WRITE
| BMAPI_MMAP
;
1038 * if we didn't have a valid mapping then we
1039 * need to ensure that we put the new mapping
1040 * in a new ioend structure. This needs to be
1041 * done to ensure that the ioends correctly
1042 * reflect the block mappings at io completion
1043 * for unwritten extent conversion.
1046 if (type
== IOMAP_NEW
) {
1047 size
= xfs_probe_cluster(inode
,
1053 err
= xfs_map_blocks(inode
, offset
, size
,
1057 iomap_valid
= xfs_iomap_valid(&iomap
, offset
);
1060 xfs_map_at_offset(bh
, offset
,
1061 inode
->i_blkbits
, &iomap
);
1063 xfs_add_to_ioend(inode
, bh
, offset
,
1067 set_buffer_dirty(bh
);
1069 mark_buffer_dirty(bh
);
1074 } else if (buffer_uptodate(bh
) && startio
) {
1076 * we got here because the buffer is already mapped.
1077 * That means it must already have extents allocated
1078 * underneath it. Map the extent by reading it.
1080 if (!iomap_valid
|| flags
!= BMAPI_READ
) {
1082 size
= xfs_probe_cluster(inode
, page
, bh
,
1084 err
= xfs_map_blocks(inode
, offset
, size
,
1088 iomap_valid
= xfs_iomap_valid(&iomap
, offset
);
1092 * We set the type to IOMAP_NEW in case we are doing a
1093 * small write at EOF that is extending the file but
1094 * without needing an allocation. We need to update the
1095 * file size on I/O completion in this case so it is
1096 * the same case as having just allocated a new extent
1097 * that we are writing into for the first time.
1100 if (!test_and_set_bit(BH_Lock
, &bh
->b_state
)) {
1101 ASSERT(buffer_mapped(bh
));
1104 xfs_add_to_ioend(inode
, bh
, offset
, type
,
1105 &ioend
, !iomap_valid
);
1111 } else if ((buffer_uptodate(bh
) || PageUptodate(page
)) &&
1112 (unmapped
|| startio
)) {
1119 } while (offset
+= len
, ((bh
= bh
->b_this_page
) != head
));
1121 if (uptodate
&& bh
== head
)
1122 SetPageUptodate(page
);
1125 xfs_start_page_writeback(page
, wbc
, 1, count
);
1127 if (ioend
&& iomap_valid
) {
1128 offset
= (iomap
.iomap_offset
+ iomap
.iomap_bsize
- 1) >>
1130 tlast
= min_t(pgoff_t
, offset
, last_index
);
1131 xfs_cluster_write(inode
, page
->index
+ 1, &iomap
, &ioend
,
1132 wbc
, startio
, all_bh
, tlast
);
1136 xfs_submit_ioend(iohead
);
1142 xfs_cancel_ioend(iohead
);
1145 * If it's delalloc and we have nowhere to put it,
1146 * throw it away, unless the lower layers told
1149 if (err
!= -EAGAIN
) {
1151 block_invalidatepage(page
, 0);
1152 ClearPageUptodate(page
);
1158 * writepage: Called from one of two places:
1160 * 1. we are flushing a delalloc buffer head.
1162 * 2. we are writing out a dirty page. Typically the page dirty
1163 * state is cleared before we get here. In this case is it
1164 * conceivable we have no buffer heads.
1166 * For delalloc space on the page we need to allocate space and
1167 * flush it. For unmapped buffer heads on the page we should
1168 * allocate space if the page is uptodate. For any other dirty
1169 * buffer heads on the page we should flush them.
1171 * If we detect that a transaction would be required to flush
1172 * the page, we have to check the process flags first, if we
1173 * are already in a transaction or disk I/O during allocations
1174 * is off, we need to fail the writepage and redirty the page.
1180 struct writeback_control
*wbc
)
1184 int delalloc
, unmapped
, unwritten
;
1185 struct inode
*inode
= page
->mapping
->host
;
1187 xfs_page_trace(XFS_WRITEPAGE_ENTER
, inode
, page
, 0);
1190 * We need a transaction if:
1191 * 1. There are delalloc buffers on the page
1192 * 2. The page is uptodate and we have unmapped buffers
1193 * 3. The page is uptodate and we have no buffers
1194 * 4. There are unwritten buffers on the page
1197 if (!page_has_buffers(page
)) {
1201 xfs_count_page_state(page
, &delalloc
, &unmapped
, &unwritten
);
1202 if (!PageUptodate(page
))
1204 need_trans
= delalloc
+ unmapped
+ unwritten
;
1208 * If we need a transaction and the process flags say
1209 * we are already in a transaction, or no IO is allowed
1210 * then mark the page dirty again and leave the page
1213 if (current_test_flags(PF_FSTRANS
) && need_trans
)
1217 * Delay hooking up buffer heads until we have
1218 * made our go/no-go decision.
1220 if (!page_has_buffers(page
))
1221 create_empty_buffers(page
, 1 << inode
->i_blkbits
, 0);
1224 * Convert delayed allocate, unwritten or unmapped space
1225 * to real space and flush out to disk.
1227 error
= xfs_page_state_convert(inode
, page
, wbc
, 1, unmapped
);
1228 if (error
== -EAGAIN
)
1230 if (unlikely(error
< 0))
1236 redirty_page_for_writepage(wbc
, page
);
1246 struct address_space
*mapping
,
1247 struct writeback_control
*wbc
)
1249 xfs_iflags_clear(XFS_I(mapping
->host
), XFS_ITRUNCATED
);
1250 return generic_writepages(mapping
, wbc
);
1254 * Called to move a page into cleanable state - and from there
1255 * to be released. Possibly the page is already clean. We always
1256 * have buffer heads in this call.
1258 * Returns 0 if the page is ok to release, 1 otherwise.
1260 * Possible scenarios are:
1262 * 1. We are being called to release a page which has been written
1263 * to via regular I/O. buffer heads will be dirty and possibly
1264 * delalloc. If no delalloc buffer heads in this case then we
1265 * can just return zero.
1267 * 2. We are called to release a page which has been written via
1268 * mmap, all we need to do is ensure there is no delalloc
1269 * state in the buffer heads, if not we can let the caller
1270 * free them and we should come back later via writepage.
1277 struct inode
*inode
= page
->mapping
->host
;
1278 int dirty
, delalloc
, unmapped
, unwritten
;
1279 struct writeback_control wbc
= {
1280 .sync_mode
= WB_SYNC_ALL
,
1284 xfs_page_trace(XFS_RELEASEPAGE_ENTER
, inode
, page
, 0);
1286 if (!page_has_buffers(page
))
1289 xfs_count_page_state(page
, &delalloc
, &unmapped
, &unwritten
);
1290 if (!delalloc
&& !unwritten
)
1293 if (!(gfp_mask
& __GFP_FS
))
1296 /* If we are already inside a transaction or the thread cannot
1297 * do I/O, we cannot release this page.
1299 if (current_test_flags(PF_FSTRANS
))
1303 * Convert delalloc space to real space, do not flush the
1304 * data out to disk, that will be done by the caller.
1305 * Never need to allocate space here - we will always
1306 * come back to writepage in that case.
1308 dirty
= xfs_page_state_convert(inode
, page
, &wbc
, 0, 0);
1309 if (dirty
== 0 && !unwritten
)
1314 return try_to_free_buffers(page
);
1319 struct inode
*inode
,
1321 struct buffer_head
*bh_result
,
1324 bmapi_flags_t flags
)
1332 offset
= (xfs_off_t
)iblock
<< inode
->i_blkbits
;
1333 ASSERT(bh_result
->b_size
>= (1 << inode
->i_blkbits
));
1334 size
= bh_result
->b_size
;
1335 error
= xfs_bmap(XFS_I(inode
), offset
, size
,
1336 create
? flags
: BMAPI_READ
, &iomap
, &niomap
);
1342 if (iomap
.iomap_bn
!= IOMAP_DADDR_NULL
) {
1344 * For unwritten extents do not report a disk address on
1345 * the read case (treat as if we're reading into a hole).
1347 if (create
|| !(iomap
.iomap_flags
& IOMAP_UNWRITTEN
)) {
1348 xfs_map_buffer(bh_result
, &iomap
, offset
,
1351 if (create
&& (iomap
.iomap_flags
& IOMAP_UNWRITTEN
)) {
1353 bh_result
->b_private
= inode
;
1354 set_buffer_unwritten(bh_result
);
1359 * If this is a realtime file, data may be on a different device.
1360 * to that pointed to from the buffer_head b_bdev currently.
1362 bh_result
->b_bdev
= iomap
.iomap_target
->bt_bdev
;
1365 * If we previously allocated a block out beyond eof and we are now
1366 * coming back to use it then we will need to flag it as new even if it
1367 * has a disk address.
1369 * With sub-block writes into unwritten extents we also need to mark
1370 * the buffer as new so that the unwritten parts of the buffer gets
1374 ((!buffer_mapped(bh_result
) && !buffer_uptodate(bh_result
)) ||
1375 (offset
>= i_size_read(inode
)) ||
1376 (iomap
.iomap_flags
& (IOMAP_NEW
|IOMAP_UNWRITTEN
))))
1377 set_buffer_new(bh_result
);
1379 if (iomap
.iomap_flags
& IOMAP_DELAY
) {
1382 set_buffer_uptodate(bh_result
);
1383 set_buffer_mapped(bh_result
);
1384 set_buffer_delay(bh_result
);
1388 if (direct
|| size
> (1 << inode
->i_blkbits
)) {
1389 ASSERT(iomap
.iomap_bsize
- iomap
.iomap_delta
> 0);
1390 offset
= min_t(xfs_off_t
,
1391 iomap
.iomap_bsize
- iomap
.iomap_delta
, size
);
1392 bh_result
->b_size
= (ssize_t
)min_t(xfs_off_t
, LONG_MAX
, offset
);
1400 struct inode
*inode
,
1402 struct buffer_head
*bh_result
,
1405 return __xfs_get_blocks(inode
, iblock
,
1406 bh_result
, create
, 0, BMAPI_WRITE
);
1410 xfs_get_blocks_direct(
1411 struct inode
*inode
,
1413 struct buffer_head
*bh_result
,
1416 return __xfs_get_blocks(inode
, iblock
,
1417 bh_result
, create
, 1, BMAPI_WRITE
|BMAPI_DIRECT
);
1427 xfs_ioend_t
*ioend
= iocb
->private;
1430 * Non-NULL private data means we need to issue a transaction to
1431 * convert a range from unwritten to written extents. This needs
1432 * to happen from process context but aio+dio I/O completion
1433 * happens from irq context so we need to defer it to a workqueue.
1434 * This is not necessary for synchronous direct I/O, but we do
1435 * it anyway to keep the code uniform and simpler.
1437 * Well, if only it were that simple. Because synchronous direct I/O
1438 * requires extent conversion to occur *before* we return to userspace,
1439 * we have to wait for extent conversion to complete. Look at the
1440 * iocb that has been passed to us to determine if this is AIO or
1441 * not. If it is synchronous, tell xfs_finish_ioend() to kick the
1442 * workqueue and wait for it to complete.
1444 * The core direct I/O code might be changed to always call the
1445 * completion handler in the future, in which case all this can
1448 ioend
->io_offset
= offset
;
1449 ioend
->io_size
= size
;
1450 if (ioend
->io_type
== IOMAP_READ
) {
1451 xfs_finish_ioend(ioend
, 0);
1452 } else if (private && size
> 0) {
1453 xfs_finish_ioend(ioend
, is_sync_kiocb(iocb
));
1456 * A direct I/O write ioend starts it's life in unwritten
1457 * state in case they map an unwritten extent. This write
1458 * didn't map an unwritten extent so switch it's completion
1461 INIT_WORK(&ioend
->io_work
, xfs_end_bio_written
);
1462 xfs_finish_ioend(ioend
, 0);
1466 * blockdev_direct_IO can return an error even after the I/O
1467 * completion handler was called. Thus we need to protect
1468 * against double-freeing.
1470 iocb
->private = NULL
;
1477 const struct iovec
*iov
,
1479 unsigned long nr_segs
)
1481 struct file
*file
= iocb
->ki_filp
;
1482 struct inode
*inode
= file
->f_mapping
->host
;
1488 error
= xfs_bmap(XFS_I(inode
), offset
, 0,
1489 BMAPI_DEVICE
, &iomap
, &maps
);
1494 iocb
->private = xfs_alloc_ioend(inode
, IOMAP_UNWRITTEN
);
1495 ret
= blockdev_direct_IO_own_locking(rw
, iocb
, inode
,
1496 iomap
.iomap_target
->bt_bdev
,
1497 iov
, offset
, nr_segs
,
1498 xfs_get_blocks_direct
,
1501 iocb
->private = xfs_alloc_ioend(inode
, IOMAP_READ
);
1502 ret
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
,
1503 iomap
.iomap_target
->bt_bdev
,
1504 iov
, offset
, nr_segs
,
1505 xfs_get_blocks_direct
,
1509 if (unlikely(ret
!= -EIOCBQUEUED
&& iocb
->private))
1510 xfs_destroy_ioend(iocb
->private);
1515 xfs_vm_prepare_write(
1521 return block_prepare_write(page
, from
, to
, xfs_get_blocks
);
1526 struct address_space
*mapping
,
1529 struct inode
*inode
= (struct inode
*)mapping
->host
;
1530 struct xfs_inode
*ip
= XFS_I(inode
);
1532 vn_trace_entry(vn_from_inode(inode
), __FUNCTION__
,
1533 (inst_t
*)__return_address
);
1534 xfs_rwlock(ip
, VRWLOCK_READ
);
1535 xfs_flush_pages(ip
, (xfs_off_t
)0, -1, 0, FI_REMAPF
);
1536 xfs_rwunlock(ip
, VRWLOCK_READ
);
1537 return generic_block_bmap(mapping
, block
, xfs_get_blocks
);
1542 struct file
*unused
,
1545 return mpage_readpage(page
, xfs_get_blocks
);
1550 struct file
*unused
,
1551 struct address_space
*mapping
,
1552 struct list_head
*pages
,
1555 return mpage_readpages(mapping
, pages
, nr_pages
, xfs_get_blocks
);
1559 xfs_vm_invalidatepage(
1561 unsigned long offset
)
1563 xfs_page_trace(XFS_INVALIDPAGE_ENTER
,
1564 page
->mapping
->host
, page
, offset
);
1565 block_invalidatepage(page
, offset
);
1568 const struct address_space_operations xfs_address_space_operations
= {
1569 .readpage
= xfs_vm_readpage
,
1570 .readpages
= xfs_vm_readpages
,
1571 .writepage
= xfs_vm_writepage
,
1572 .writepages
= xfs_vm_writepages
,
1573 .sync_page
= block_sync_page
,
1574 .releasepage
= xfs_vm_releasepage
,
1575 .invalidatepage
= xfs_vm_invalidatepage
,
1576 .prepare_write
= xfs_vm_prepare_write
,
1577 .commit_write
= generic_commit_write
,
1578 .bmap
= xfs_vm_bmap
,
1579 .direct_IO
= xfs_vm_direct_IO
,
1580 .migratepage
= buffer_migrate_page
,