2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
24 #include "xfs_trans.h"
25 #include "xfs_mount.h"
26 #include "xfs_bmap_btree.h"
27 #include "xfs_dinode.h"
28 #include "xfs_inode.h"
29 #include "xfs_alloc.h"
30 #include "xfs_error.h"
32 #include "xfs_iomap.h"
33 #include "xfs_vnodeops.h"
34 #include "xfs_trace.h"
36 #include <linux/gfp.h>
37 #include <linux/mpage.h>
38 #include <linux/pagevec.h>
39 #include <linux/writeback.h>
42 * Types of I/O for bmap clustering and I/O completion tracking.
45 IO_READ
, /* mapping for a read */
46 IO_DELAY
, /* mapping covers delalloc region */
47 IO_UNWRITTEN
, /* mapping covers allocated but uninitialized data */
48 IO_NEW
/* just allocated */
52 * Prime number of hash buckets since address is used as the key.
55 #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC])
56 static wait_queue_head_t xfs_ioend_wq
[NVSYNC
];
63 for (i
= 0; i
< NVSYNC
; i
++)
64 init_waitqueue_head(&xfs_ioend_wq
[i
]);
71 wait_queue_head_t
*wq
= to_ioend_wq(ip
);
73 wait_event(*wq
, (atomic_read(&ip
->i_iocount
) == 0));
80 if (atomic_dec_and_test(&ip
->i_iocount
))
81 wake_up(to_ioend_wq(ip
));
90 struct buffer_head
*bh
, *head
;
92 *delalloc
= *unwritten
= 0;
94 bh
= head
= page_buffers(page
);
96 if (buffer_unwritten(bh
))
98 else if (buffer_delay(bh
))
100 } while ((bh
= bh
->b_this_page
) != head
);
103 STATIC
struct block_device
*
104 xfs_find_bdev_for_inode(
107 struct xfs_inode
*ip
= XFS_I(inode
);
108 struct xfs_mount
*mp
= ip
->i_mount
;
110 if (XFS_IS_REALTIME_INODE(ip
))
111 return mp
->m_rtdev_targp
->bt_bdev
;
113 return mp
->m_ddev_targp
->bt_bdev
;
117 * We're now finished for good with this ioend structure.
118 * Update the page state via the associated buffer_heads,
119 * release holds on the inode and bio, and finally free
120 * up memory. Do not use the ioend after this.
126 struct buffer_head
*bh
, *next
;
127 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
129 for (bh
= ioend
->io_buffer_head
; bh
; bh
= next
) {
130 next
= bh
->b_private
;
131 bh
->b_end_io(bh
, !ioend
->io_error
);
135 * Volume managers supporting multiple paths can send back ENODEV
136 * when the final path disappears. In this case continuing to fill
137 * the page cache with dirty data which cannot be written out is
138 * evil, so prevent that.
140 if (unlikely(ioend
->io_error
== -ENODEV
)) {
141 xfs_do_force_shutdown(ip
->i_mount
, SHUTDOWN_DEVICE_REQ
,
146 mempool_free(ioend
, xfs_ioend_pool
);
150 * If the end of the current ioend is beyond the current EOF,
151 * return the new EOF value, otherwise zero.
157 xfs_inode_t
*ip
= XFS_I(ioend
->io_inode
);
161 bsize
= ioend
->io_offset
+ ioend
->io_size
;
162 isize
= MAX(ip
->i_size
, ip
->i_new_size
);
163 isize
= MIN(isize
, bsize
);
164 return isize
> ip
->i_d
.di_size
? isize
: 0;
168 * Update on-disk file size now that data has been written to disk. The
169 * current in-memory file size is i_size. If a write is beyond eof i_new_size
170 * will be the intended file size until i_size is updated. If this write does
171 * not extend all the way to the valid file size then restrict this update to
172 * the end of the write.
174 * This function does not block as blocking on the inode lock in IO completion
175 * can lead to IO completion order dependency deadlocks.. If it can't get the
176 * inode ilock it will return EAGAIN. Callers must handle this.
182 xfs_inode_t
*ip
= XFS_I(ioend
->io_inode
);
185 ASSERT((ip
->i_d
.di_mode
& S_IFMT
) == S_IFREG
);
186 ASSERT(ioend
->io_type
!= IO_READ
);
188 if (unlikely(ioend
->io_error
))
191 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
))
194 isize
= xfs_ioend_new_eof(ioend
);
196 ip
->i_d
.di_size
= isize
;
197 xfs_mark_inode_dirty(ip
);
200 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
205 * Schedule IO completion handling on a xfsdatad if this was
206 * the final hold on this ioend. If we are asked to wait,
207 * flush the workqueue.
214 if (atomic_dec_and_test(&ioend
->io_remaining
)) {
215 struct workqueue_struct
*wq
;
217 wq
= (ioend
->io_type
== IO_UNWRITTEN
) ?
218 xfsconvertd_workqueue
: xfsdatad_workqueue
;
219 queue_work(wq
, &ioend
->io_work
);
226 * IO write completion.
230 struct work_struct
*work
)
232 xfs_ioend_t
*ioend
= container_of(work
, xfs_ioend_t
, io_work
);
233 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
237 * For unwritten extents we need to issue transactions to convert a
238 * range to normal written extens after the data I/O has finished.
240 if (ioend
->io_type
== IO_UNWRITTEN
&&
241 likely(!ioend
->io_error
&& !XFS_FORCED_SHUTDOWN(ip
->i_mount
))) {
243 error
= xfs_iomap_write_unwritten(ip
, ioend
->io_offset
,
246 ioend
->io_error
= error
;
250 * We might have to update the on-disk file size after extending
253 if (ioend
->io_type
!= IO_READ
) {
254 error
= xfs_setfilesize(ioend
);
255 ASSERT(!error
|| error
== EAGAIN
);
259 * If we didn't complete processing of the ioend, requeue it to the
260 * tail of the workqueue for another attempt later. Otherwise destroy
263 if (error
== EAGAIN
) {
264 atomic_inc(&ioend
->io_remaining
);
265 xfs_finish_ioend(ioend
, 0);
266 /* ensure we don't spin on blocked ioends */
269 xfs_destroy_ioend(ioend
);
273 * Allocate and initialise an IO completion structure.
274 * We need to track unwritten extent write completion here initially.
275 * We'll need to extend this for updating the ondisk inode size later
285 ioend
= mempool_alloc(xfs_ioend_pool
, GFP_NOFS
);
288 * Set the count to 1 initially, which will prevent an I/O
289 * completion callback from happening before we have started
290 * all the I/O from calling the completion routine too early.
292 atomic_set(&ioend
->io_remaining
, 1);
294 ioend
->io_list
= NULL
;
295 ioend
->io_type
= type
;
296 ioend
->io_inode
= inode
;
297 ioend
->io_buffer_head
= NULL
;
298 ioend
->io_buffer_tail
= NULL
;
299 atomic_inc(&XFS_I(ioend
->io_inode
)->i_iocount
);
300 ioend
->io_offset
= 0;
303 INIT_WORK(&ioend
->io_work
, xfs_end_io
);
312 struct xfs_bmbt_irec
*imap
,
318 return -xfs_iomap(XFS_I(inode
), offset
, count
, flags
, imap
, &nmaps
, &new);
324 struct xfs_bmbt_irec
*imap
,
327 offset
>>= inode
->i_blkbits
;
329 return offset
>= imap
->br_startoff
&&
330 offset
< imap
->br_startoff
+ imap
->br_blockcount
;
334 * BIO completion handler for buffered IO.
341 xfs_ioend_t
*ioend
= bio
->bi_private
;
343 ASSERT(atomic_read(&bio
->bi_cnt
) >= 1);
344 ioend
->io_error
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
) ? 0 : error
;
346 /* Toss bio and pass work off to an xfsdatad thread */
347 bio
->bi_private
= NULL
;
348 bio
->bi_end_io
= NULL
;
351 xfs_finish_ioend(ioend
, 0);
355 xfs_submit_ioend_bio(
356 struct writeback_control
*wbc
,
360 atomic_inc(&ioend
->io_remaining
);
361 bio
->bi_private
= ioend
;
362 bio
->bi_end_io
= xfs_end_bio
;
365 * If the I/O is beyond EOF we mark the inode dirty immediately
366 * but don't update the inode size until I/O completion.
368 if (xfs_ioend_new_eof(ioend
))
369 xfs_mark_inode_dirty(XFS_I(ioend
->io_inode
));
371 submit_bio(wbc
->sync_mode
== WB_SYNC_ALL
?
372 WRITE_SYNC_PLUG
: WRITE
, bio
);
373 ASSERT(!bio_flagged(bio
, BIO_EOPNOTSUPP
));
379 struct buffer_head
*bh
)
382 int nvecs
= bio_get_nr_vecs(bh
->b_bdev
);
385 bio
= bio_alloc(GFP_NOIO
, nvecs
);
389 ASSERT(bio
->bi_private
== NULL
);
390 bio
->bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
391 bio
->bi_bdev
= bh
->b_bdev
;
397 xfs_start_buffer_writeback(
398 struct buffer_head
*bh
)
400 ASSERT(buffer_mapped(bh
));
401 ASSERT(buffer_locked(bh
));
402 ASSERT(!buffer_delay(bh
));
403 ASSERT(!buffer_unwritten(bh
));
405 mark_buffer_async_write(bh
);
406 set_buffer_uptodate(bh
);
407 clear_buffer_dirty(bh
);
411 xfs_start_page_writeback(
416 ASSERT(PageLocked(page
));
417 ASSERT(!PageWriteback(page
));
419 clear_page_dirty_for_io(page
);
420 set_page_writeback(page
);
422 /* If no buffers on the page are to be written, finish it here */
424 end_page_writeback(page
);
427 static inline int bio_add_buffer(struct bio
*bio
, struct buffer_head
*bh
)
429 return bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
433 * Submit all of the bios for all of the ioends we have saved up, covering the
434 * initial writepage page and also any probed pages.
436 * Because we may have multiple ioends spanning a page, we need to start
437 * writeback on all the buffers before we submit them for I/O. If we mark the
438 * buffers as we got, then we can end up with a page that only has buffers
439 * marked async write and I/O complete on can occur before we mark the other
440 * buffers async write.
442 * The end result of this is that we trip a bug in end_page_writeback() because
443 * we call it twice for the one page as the code in end_buffer_async_write()
444 * assumes that all buffers on the page are started at the same time.
446 * The fix is two passes across the ioend list - one to start writeback on the
447 * buffer_heads, and then submit them for I/O on the second pass.
451 struct writeback_control
*wbc
,
454 xfs_ioend_t
*head
= ioend
;
456 struct buffer_head
*bh
;
458 sector_t lastblock
= 0;
460 /* Pass 1 - start writeback */
462 next
= ioend
->io_list
;
463 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
464 xfs_start_buffer_writeback(bh
);
466 } while ((ioend
= next
) != NULL
);
468 /* Pass 2 - submit I/O */
471 next
= ioend
->io_list
;
474 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
478 bio
= xfs_alloc_ioend_bio(bh
);
479 } else if (bh
->b_blocknr
!= lastblock
+ 1) {
480 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
484 if (bio_add_buffer(bio
, bh
) != bh
->b_size
) {
485 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
489 lastblock
= bh
->b_blocknr
;
492 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
493 xfs_finish_ioend(ioend
, 0);
494 } while ((ioend
= next
) != NULL
);
498 * Cancel submission of all buffer_heads so far in this endio.
499 * Toss the endio too. Only ever called for the initial page
500 * in a writepage request, so only ever one page.
507 struct buffer_head
*bh
, *next_bh
;
510 next
= ioend
->io_list
;
511 bh
= ioend
->io_buffer_head
;
513 next_bh
= bh
->b_private
;
514 clear_buffer_async_write(bh
);
516 } while ((bh
= next_bh
) != NULL
);
518 xfs_ioend_wake(XFS_I(ioend
->io_inode
));
519 mempool_free(ioend
, xfs_ioend_pool
);
520 } while ((ioend
= next
) != NULL
);
524 * Test to see if we've been building up a completion structure for
525 * earlier buffers -- if so, we try to append to this ioend if we
526 * can, otherwise we finish off any current ioend and start another.
527 * Return true if we've finished the given ioend.
532 struct buffer_head
*bh
,
535 xfs_ioend_t
**result
,
538 xfs_ioend_t
*ioend
= *result
;
540 if (!ioend
|| need_ioend
|| type
!= ioend
->io_type
) {
541 xfs_ioend_t
*previous
= *result
;
543 ioend
= xfs_alloc_ioend(inode
, type
);
544 ioend
->io_offset
= offset
;
545 ioend
->io_buffer_head
= bh
;
546 ioend
->io_buffer_tail
= bh
;
548 previous
->io_list
= ioend
;
551 ioend
->io_buffer_tail
->b_private
= bh
;
552 ioend
->io_buffer_tail
= bh
;
555 bh
->b_private
= NULL
;
556 ioend
->io_size
+= bh
->b_size
;
562 struct buffer_head
*bh
,
563 struct xfs_bmbt_irec
*imap
,
567 struct xfs_mount
*m
= XFS_I(inode
)->i_mount
;
568 xfs_off_t iomap_offset
= XFS_FSB_TO_B(m
, imap
->br_startoff
);
569 xfs_daddr_t iomap_bn
= xfs_fsb_to_db(XFS_I(inode
), imap
->br_startblock
);
571 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
572 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
574 bn
= (iomap_bn
>> (inode
->i_blkbits
- BBSHIFT
)) +
575 ((offset
- iomap_offset
) >> inode
->i_blkbits
);
577 ASSERT(bn
|| XFS_IS_REALTIME_INODE(XFS_I(inode
)));
580 set_buffer_mapped(bh
);
586 struct buffer_head
*bh
,
587 struct xfs_bmbt_irec
*imap
,
590 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
591 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
594 xfs_map_buffer(inode
, bh
, imap
, offset
);
595 bh
->b_bdev
= xfs_find_bdev_for_inode(inode
);
596 set_buffer_mapped(bh
);
597 clear_buffer_delay(bh
);
598 clear_buffer_unwritten(bh
);
602 * Look for a page at index that is suitable for clustering.
607 unsigned int pg_offset
)
609 struct buffer_head
*bh
, *head
;
612 if (PageWriteback(page
))
614 if (!PageDirty(page
))
618 if (!page_has_buffers(page
))
621 bh
= head
= page_buffers(page
);
623 if (!buffer_uptodate(bh
))
625 if (!buffer_mapped(bh
))
628 if (ret
>= pg_offset
)
630 } while ((bh
= bh
->b_this_page
) != head
);
638 struct page
*startpage
,
639 struct buffer_head
*bh
,
640 struct buffer_head
*head
)
643 pgoff_t tindex
, tlast
, tloff
;
647 /* First sum forwards in this page */
649 if (!buffer_uptodate(bh
) || !buffer_mapped(bh
))
652 } while ((bh
= bh
->b_this_page
) != head
);
654 /* if we reached the end of the page, sum forwards in following pages */
655 tlast
= i_size_read(inode
) >> PAGE_CACHE_SHIFT
;
656 tindex
= startpage
->index
+ 1;
658 /* Prune this back to avoid pathological behavior */
659 tloff
= min(tlast
, startpage
->index
+ 64);
661 pagevec_init(&pvec
, 0);
662 while (!done
&& tindex
<= tloff
) {
663 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
665 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
668 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
669 struct page
*page
= pvec
.pages
[i
];
670 size_t pg_offset
, pg_len
= 0;
672 if (tindex
== tlast
) {
674 i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1);
680 pg_offset
= PAGE_CACHE_SIZE
;
682 if (page
->index
== tindex
&& trylock_page(page
)) {
683 pg_len
= xfs_probe_page(page
, pg_offset
);
696 pagevec_release(&pvec
);
704 * Test if a given page is suitable for writing as part of an unwritten
705 * or delayed allocate extent.
712 if (PageWriteback(page
))
715 if (page
->mapping
&& page_has_buffers(page
)) {
716 struct buffer_head
*bh
, *head
;
719 bh
= head
= page_buffers(page
);
721 if (buffer_unwritten(bh
))
722 acceptable
= (type
== IO_UNWRITTEN
);
723 else if (buffer_delay(bh
))
724 acceptable
= (type
== IO_DELAY
);
725 else if (buffer_dirty(bh
) && buffer_mapped(bh
))
726 acceptable
= (type
== IO_NEW
);
729 } while ((bh
= bh
->b_this_page
) != head
);
739 * Allocate & map buffers for page given the extent map. Write it out.
740 * except for the original page of a writepage, this is called on
741 * delalloc/unwritten pages only, for the original page it is possible
742 * that the page has no mapping at all.
749 struct xfs_bmbt_irec
*imap
,
750 xfs_ioend_t
**ioendp
,
751 struct writeback_control
*wbc
,
754 struct buffer_head
*bh
, *head
;
755 xfs_off_t end_offset
;
756 unsigned long p_offset
;
759 int count
= 0, done
= 0, uptodate
= 1;
760 xfs_off_t offset
= page_offset(page
);
762 if (page
->index
!= tindex
)
764 if (!trylock_page(page
))
766 if (PageWriteback(page
))
767 goto fail_unlock_page
;
768 if (page
->mapping
!= inode
->i_mapping
)
769 goto fail_unlock_page
;
770 if (!xfs_is_delayed_page(page
, (*ioendp
)->io_type
))
771 goto fail_unlock_page
;
774 * page_dirty is initially a count of buffers on the page before
775 * EOF and is decremented as we move each into a cleanable state.
779 * End offset is the highest offset that this page should represent.
780 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
781 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
782 * hence give us the correct page_dirty count. On any other page,
783 * it will be zero and in that case we need page_dirty to be the
784 * count of buffers on the page.
786 end_offset
= min_t(unsigned long long,
787 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
790 len
= 1 << inode
->i_blkbits
;
791 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
793 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
794 page_dirty
= p_offset
/ len
;
796 bh
= head
= page_buffers(page
);
798 if (offset
>= end_offset
)
800 if (!buffer_uptodate(bh
))
802 if (!(PageUptodate(page
) || buffer_uptodate(bh
))) {
807 if (buffer_unwritten(bh
) || buffer_delay(bh
)) {
808 if (buffer_unwritten(bh
))
813 if (!xfs_imap_valid(inode
, imap
, offset
)) {
818 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
819 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
821 xfs_map_at_offset(inode
, bh
, imap
, offset
);
822 xfs_add_to_ioend(inode
, bh
, offset
, type
,
829 if (buffer_mapped(bh
) && all_bh
) {
831 xfs_add_to_ioend(inode
, bh
, offset
,
839 } while (offset
+= len
, (bh
= bh
->b_this_page
) != head
);
841 if (uptodate
&& bh
== head
)
842 SetPageUptodate(page
);
846 if (wbc
->nr_to_write
<= 0)
849 xfs_start_page_writeback(page
, !page_dirty
, count
);
859 * Convert & write out a cluster of pages in the same extent as defined
860 * by mp and following the start page.
866 struct xfs_bmbt_irec
*imap
,
867 xfs_ioend_t
**ioendp
,
868 struct writeback_control
*wbc
,
875 pagevec_init(&pvec
, 0);
876 while (!done
&& tindex
<= tlast
) {
877 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
879 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
882 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
883 done
= xfs_convert_page(inode
, pvec
.pages
[i
], tindex
++,
884 imap
, ioendp
, wbc
, all_bh
);
889 pagevec_release(&pvec
);
895 xfs_vm_invalidatepage(
897 unsigned long offset
)
899 trace_xfs_invalidatepage(page
->mapping
->host
, page
, offset
);
900 block_invalidatepage(page
, offset
);
904 * If the page has delalloc buffers on it, we need to punch them out before we
905 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
906 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
907 * is done on that same region - the delalloc extent is returned when none is
908 * supposed to be there.
910 * We prevent this by truncating away the delalloc regions on the page before
911 * invalidating it. Because they are delalloc, we can do this without needing a
912 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
913 * truncation without a transaction as there is no space left for block
914 * reservation (typically why we see a ENOSPC in writeback).
916 * This is not a performance critical path, so for now just do the punching a
917 * buffer head at a time.
920 xfs_aops_discard_page(
923 struct inode
*inode
= page
->mapping
->host
;
924 struct xfs_inode
*ip
= XFS_I(inode
);
925 struct buffer_head
*bh
, *head
;
926 loff_t offset
= page_offset(page
);
927 ssize_t len
= 1 << inode
->i_blkbits
;
929 if (!xfs_is_delayed_page(page
, IO_DELAY
))
932 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
935 xfs_fs_cmn_err(CE_ALERT
, ip
->i_mount
,
936 "page discard on page %p, inode 0x%llx, offset %llu.",
937 page
, ip
->i_ino
, offset
);
939 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
940 bh
= head
= page_buffers(page
);
943 xfs_fileoff_t offset_fsb
;
944 xfs_bmbt_irec_t imap
;
947 xfs_fsblock_t firstblock
;
948 xfs_bmap_free_t flist
;
950 if (!buffer_delay(bh
))
953 offset_fsb
= XFS_B_TO_FSBT(ip
->i_mount
, offset
);
956 * Map the range first and check that it is a delalloc extent
957 * before trying to unmap the range. Otherwise we will be
958 * trying to remove a real extent (which requires a
959 * transaction) or a hole, which is probably a bad idea...
961 error
= xfs_bmapi(NULL
, ip
, offset_fsb
, 1,
962 XFS_BMAPI_ENTIRE
, NULL
, 0, &imap
,
966 /* something screwed, just bail */
967 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
968 xfs_fs_cmn_err(CE_ALERT
, ip
->i_mount
,
969 "page discard failed delalloc mapping lookup.");
977 if (imap
.br_startblock
!= DELAYSTARTBLOCK
) {
978 /* been converted, ignore */
981 WARN_ON(imap
.br_blockcount
== 0);
984 * Note: while we initialise the firstblock/flist pair, they
985 * should never be used because blocks should never be
986 * allocated or freed for a delalloc extent and hence we need
987 * don't cancel or finish them after the xfs_bunmapi() call.
989 xfs_bmap_init(&flist
, &firstblock
);
990 error
= xfs_bunmapi(NULL
, ip
, offset_fsb
, 1, 0, 1, &firstblock
,
993 ASSERT(!flist
.xbf_count
&& !flist
.xbf_first
);
995 /* something screwed, just bail */
996 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
997 xfs_fs_cmn_err(CE_ALERT
, ip
->i_mount
,
998 "page discard unable to remove delalloc mapping.");
1005 } while ((bh
= bh
->b_this_page
) != head
);
1007 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1009 xfs_vm_invalidatepage(page
, 0);
1014 * Write out a dirty page.
1016 * For delalloc space on the page we need to allocate space and flush it.
1017 * For unwritten space on the page we need to start the conversion to
1018 * regular allocated space.
1019 * For any other dirty buffer heads on the page we should flush them.
1021 * If we detect that a transaction would be required to flush the page, we
1022 * have to check the process flags first, if we are already in a transaction
1023 * or disk I/O during allocations is off, we need to fail the writepage and
1029 struct writeback_control
*wbc
)
1031 struct inode
*inode
= page
->mapping
->host
;
1032 int delalloc
, unwritten
;
1033 struct buffer_head
*bh
, *head
;
1034 struct xfs_bmbt_irec imap
;
1035 xfs_ioend_t
*ioend
= NULL
, *iohead
= NULL
;
1038 __uint64_t end_offset
;
1039 pgoff_t end_index
, last_index
;
1041 int flags
, err
, imap_valid
= 0, uptodate
= 1;
1045 trace_xfs_writepage(inode
, page
, 0);
1047 ASSERT(page_has_buffers(page
));
1050 * Refuse to write the page out if we are called from reclaim context.
1052 * This avoids stack overflows when called from deeply used stacks in
1053 * random callers for direct reclaim or memcg reclaim. We explicitly
1054 * allow reclaim from kswapd as the stack usage there is relatively low.
1056 * This should really be done by the core VM, but until that happens
1057 * filesystems like XFS, btrfs and ext4 have to take care of this
1060 if ((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
)) == PF_MEMALLOC
)
1064 * We need a transaction if there are delalloc or unwritten buffers
1067 * If we need a transaction and the process flags say we are already
1068 * in a transaction, or no IO is allowed then mark the page dirty
1069 * again and leave the page as is.
1071 xfs_count_page_state(page
, &delalloc
, &unwritten
);
1072 if ((current
->flags
& PF_FSTRANS
) && (delalloc
|| unwritten
))
1076 * Delay hooking up buffer heads until we have
1077 * made our go/no-go decision.
1079 if (!page_has_buffers(page
))
1080 create_empty_buffers(page
, 1 << inode
->i_blkbits
, 0);
1082 /* Is this page beyond the end of the file? */
1083 offset
= i_size_read(inode
);
1084 end_index
= offset
>> PAGE_CACHE_SHIFT
;
1085 last_index
= (offset
- 1) >> PAGE_CACHE_SHIFT
;
1086 if (page
->index
>= end_index
) {
1087 if ((page
->index
>= end_index
+ 1) ||
1088 !(i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1))) {
1094 end_offset
= min_t(unsigned long long,
1095 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
1097 len
= 1 << inode
->i_blkbits
;
1099 bh
= head
= page_buffers(page
);
1100 offset
= page_offset(page
);
1105 if (offset
>= end_offset
)
1107 if (!buffer_uptodate(bh
))
1111 * A hole may still be marked uptodate because discard_buffer
1112 * leaves the flag set.
1114 if (!buffer_mapped(bh
) && buffer_uptodate(bh
)) {
1115 ASSERT(!buffer_dirty(bh
));
1121 imap_valid
= xfs_imap_valid(inode
, &imap
, offset
);
1123 if (buffer_unwritten(bh
) || buffer_delay(bh
)) {
1127 * Make sure we don't use a read-only iomap
1129 if (flags
== BMAPI_READ
)
1132 if (buffer_unwritten(bh
)) {
1133 type
= IO_UNWRITTEN
;
1134 flags
= BMAPI_WRITE
| BMAPI_IGNSTATE
;
1135 } else if (buffer_delay(bh
)) {
1137 flags
= BMAPI_ALLOCATE
;
1139 if (wbc
->sync_mode
== WB_SYNC_NONE
&&
1141 flags
|= BMAPI_TRYLOCK
;
1146 * If we didn't have a valid mapping then we
1147 * need to ensure that we put the new mapping
1148 * in a new ioend structure. This needs to be
1149 * done to ensure that the ioends correctly
1150 * reflect the block mappings at io completion
1151 * for unwritten extent conversion.
1154 err
= xfs_map_blocks(inode
, offset
, len
,
1158 imap_valid
= xfs_imap_valid(inode
, &imap
,
1162 xfs_map_at_offset(inode
, bh
, &imap
, offset
);
1163 xfs_add_to_ioend(inode
, bh
, offset
, type
,
1167 } else if (buffer_uptodate(bh
)) {
1169 * we got here because the buffer is already mapped.
1170 * That means it must already have extents allocated
1171 * underneath it. Map the extent by reading it.
1173 if (!imap_valid
|| flags
!= BMAPI_READ
) {
1175 size
= xfs_probe_cluster(inode
, page
, bh
, head
);
1176 err
= xfs_map_blocks(inode
, offset
, size
,
1180 imap_valid
= xfs_imap_valid(inode
, &imap
,
1185 * We set the type to IO_NEW in case we are doing a
1186 * small write at EOF that is extending the file but
1187 * without needing an allocation. We need to update the
1188 * file size on I/O completion in this case so it is
1189 * the same case as having just allocated a new extent
1190 * that we are writing into for the first time.
1193 if (trylock_buffer(bh
)) {
1196 xfs_add_to_ioend(inode
, bh
, offset
, type
,
1197 &ioend
, !imap_valid
);
1202 } else if (PageUptodate(page
)) {
1203 ASSERT(buffer_mapped(bh
));
1210 } while (offset
+= len
, ((bh
= bh
->b_this_page
) != head
));
1212 if (uptodate
&& bh
== head
)
1213 SetPageUptodate(page
);
1215 xfs_start_page_writeback(page
, 1, count
);
1217 if (ioend
&& imap_valid
) {
1218 xfs_off_t end_index
;
1220 end_index
= imap
.br_startoff
+ imap
.br_blockcount
;
1223 end_index
<<= inode
->i_blkbits
;
1226 end_index
= (end_index
- 1) >> PAGE_CACHE_SHIFT
;
1228 /* check against file size */
1229 if (end_index
> last_index
)
1230 end_index
= last_index
;
1232 xfs_cluster_write(inode
, page
->index
+ 1, &imap
, &ioend
,
1233 wbc
, all_bh
, end_index
);
1237 xfs_submit_ioend(wbc
, iohead
);
1243 xfs_cancel_ioend(iohead
);
1245 xfs_aops_discard_page(page
);
1246 ClearPageUptodate(page
);
1251 redirty_page_for_writepage(wbc
, page
);
1258 struct address_space
*mapping
,
1259 struct writeback_control
*wbc
)
1261 xfs_iflags_clear(XFS_I(mapping
->host
), XFS_ITRUNCATED
);
1262 return generic_writepages(mapping
, wbc
);
1266 * Called to move a page into cleanable state - and from there
1267 * to be released. The page should already be clean. We always
1268 * have buffer heads in this call.
1270 * Returns 1 if the page is ok to release, 0 otherwise.
1277 int delalloc
, unwritten
;
1279 trace_xfs_releasepage(page
->mapping
->host
, page
, 0);
1281 xfs_count_page_state(page
, &delalloc
, &unwritten
);
1283 if (WARN_ON(delalloc
))
1285 if (WARN_ON(unwritten
))
1288 return try_to_free_buffers(page
);
1293 struct inode
*inode
,
1295 struct buffer_head
*bh_result
,
1299 int flags
= create
? BMAPI_WRITE
: BMAPI_READ
;
1300 struct xfs_bmbt_irec imap
;
1307 offset
= (xfs_off_t
)iblock
<< inode
->i_blkbits
;
1308 ASSERT(bh_result
->b_size
>= (1 << inode
->i_blkbits
));
1309 size
= bh_result
->b_size
;
1311 if (!create
&& direct
&& offset
>= i_size_read(inode
))
1314 if (direct
&& create
)
1315 flags
|= BMAPI_DIRECT
;
1317 error
= xfs_iomap(XFS_I(inode
), offset
, size
, flags
, &imap
, &nimap
,
1324 if (imap
.br_startblock
!= HOLESTARTBLOCK
&&
1325 imap
.br_startblock
!= DELAYSTARTBLOCK
) {
1327 * For unwritten extents do not report a disk address on
1328 * the read case (treat as if we're reading into a hole).
1330 if (create
|| !ISUNWRITTEN(&imap
))
1331 xfs_map_buffer(inode
, bh_result
, &imap
, offset
);
1332 if (create
&& ISUNWRITTEN(&imap
)) {
1334 bh_result
->b_private
= inode
;
1335 set_buffer_unwritten(bh_result
);
1340 * If this is a realtime file, data may be on a different device.
1341 * to that pointed to from the buffer_head b_bdev currently.
1343 bh_result
->b_bdev
= xfs_find_bdev_for_inode(inode
);
1346 * If we previously allocated a block out beyond eof and we are now
1347 * coming back to use it then we will need to flag it as new even if it
1348 * has a disk address.
1350 * With sub-block writes into unwritten extents we also need to mark
1351 * the buffer as new so that the unwritten parts of the buffer gets
1355 ((!buffer_mapped(bh_result
) && !buffer_uptodate(bh_result
)) ||
1356 (offset
>= i_size_read(inode
)) ||
1357 (new || ISUNWRITTEN(&imap
))))
1358 set_buffer_new(bh_result
);
1360 if (imap
.br_startblock
== DELAYSTARTBLOCK
) {
1363 set_buffer_uptodate(bh_result
);
1364 set_buffer_mapped(bh_result
);
1365 set_buffer_delay(bh_result
);
1370 * If this is O_DIRECT or the mpage code calling tell them how large
1371 * the mapping is, so that we can avoid repeated get_blocks calls.
1373 if (direct
|| size
> (1 << inode
->i_blkbits
)) {
1374 xfs_off_t mapping_size
;
1376 mapping_size
= imap
.br_startoff
+ imap
.br_blockcount
- iblock
;
1377 mapping_size
<<= inode
->i_blkbits
;
1379 ASSERT(mapping_size
> 0);
1380 if (mapping_size
> size
)
1381 mapping_size
= size
;
1382 if (mapping_size
> LONG_MAX
)
1383 mapping_size
= LONG_MAX
;
1385 bh_result
->b_size
= mapping_size
;
1393 struct inode
*inode
,
1395 struct buffer_head
*bh_result
,
1398 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 0);
1402 xfs_get_blocks_direct(
1403 struct inode
*inode
,
1405 struct buffer_head
*bh_result
,
1408 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 1);
1418 xfs_ioend_t
*ioend
= iocb
->private;
1421 * Non-NULL private data means we need to issue a transaction to
1422 * convert a range from unwritten to written extents. This needs
1423 * to happen from process context but aio+dio I/O completion
1424 * happens from irq context so we need to defer it to a workqueue.
1425 * This is not necessary for synchronous direct I/O, but we do
1426 * it anyway to keep the code uniform and simpler.
1428 * Well, if only it were that simple. Because synchronous direct I/O
1429 * requires extent conversion to occur *before* we return to userspace,
1430 * we have to wait for extent conversion to complete. Look at the
1431 * iocb that has been passed to us to determine if this is AIO or
1432 * not. If it is synchronous, tell xfs_finish_ioend() to kick the
1433 * workqueue and wait for it to complete.
1435 * The core direct I/O code might be changed to always call the
1436 * completion handler in the future, in which case all this can
1439 ioend
->io_offset
= offset
;
1440 ioend
->io_size
= size
;
1441 if (ioend
->io_type
== IO_READ
) {
1442 xfs_finish_ioend(ioend
, 0);
1443 } else if (private && size
> 0) {
1444 xfs_finish_ioend(ioend
, is_sync_kiocb(iocb
));
1447 * A direct I/O write ioend starts it's life in unwritten
1448 * state in case they map an unwritten extent. This write
1449 * didn't map an unwritten extent so switch it's completion
1452 ioend
->io_type
= IO_NEW
;
1453 xfs_finish_ioend(ioend
, 0);
1457 * blockdev_direct_IO can return an error even after the I/O
1458 * completion handler was called. Thus we need to protect
1459 * against double-freeing.
1461 iocb
->private = NULL
;
1468 const struct iovec
*iov
,
1470 unsigned long nr_segs
)
1472 struct file
*file
= iocb
->ki_filp
;
1473 struct inode
*inode
= file
->f_mapping
->host
;
1474 struct block_device
*bdev
;
1477 bdev
= xfs_find_bdev_for_inode(inode
);
1479 iocb
->private = xfs_alloc_ioend(inode
, rw
== WRITE
?
1480 IO_UNWRITTEN
: IO_READ
);
1482 ret
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
, bdev
, iov
,
1484 xfs_get_blocks_direct
,
1487 if (unlikely(ret
!= -EIOCBQUEUED
&& iocb
->private))
1488 xfs_destroy_ioend(iocb
->private);
1495 struct address_space
*mapping
,
1499 struct page
**pagep
,
1503 return block_write_begin(file
, mapping
, pos
, len
, flags
, pagep
, fsdata
,
1509 struct address_space
*mapping
,
1512 struct inode
*inode
= (struct inode
*)mapping
->host
;
1513 struct xfs_inode
*ip
= XFS_I(inode
);
1515 trace_xfs_vm_bmap(XFS_I(inode
));
1516 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
1517 xfs_flush_pages(ip
, (xfs_off_t
)0, -1, 0, FI_REMAPF
);
1518 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
1519 return generic_block_bmap(mapping
, block
, xfs_get_blocks
);
1524 struct file
*unused
,
1527 return mpage_readpage(page
, xfs_get_blocks
);
1532 struct file
*unused
,
1533 struct address_space
*mapping
,
1534 struct list_head
*pages
,
1537 return mpage_readpages(mapping
, pages
, nr_pages
, xfs_get_blocks
);
1540 const struct address_space_operations xfs_address_space_operations
= {
1541 .readpage
= xfs_vm_readpage
,
1542 .readpages
= xfs_vm_readpages
,
1543 .writepage
= xfs_vm_writepage
,
1544 .writepages
= xfs_vm_writepages
,
1545 .sync_page
= block_sync_page
,
1546 .releasepage
= xfs_vm_releasepage
,
1547 .invalidatepage
= xfs_vm_invalidatepage
,
1548 .write_begin
= xfs_vm_write_begin
,
1549 .write_end
= generic_write_end
,
1550 .bmap
= xfs_vm_bmap
,
1551 .direct_IO
= xfs_vm_direct_IO
,
1552 .migratepage
= buffer_migrate_page
,
1553 .is_partially_uptodate
= block_is_partially_uptodate
,
1554 .error_remove_page
= generic_error_remove_page
,