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>
43 * Prime number of hash buckets since address is used as the key.
46 #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC])
47 static wait_queue_head_t xfs_ioend_wq
[NVSYNC
];
54 for (i
= 0; i
< NVSYNC
; i
++)
55 init_waitqueue_head(&xfs_ioend_wq
[i
]);
62 wait_queue_head_t
*wq
= to_ioend_wq(ip
);
64 wait_event(*wq
, (atomic_read(&ip
->i_iocount
) == 0));
71 if (atomic_dec_and_test(&ip
->i_iocount
))
72 wake_up(to_ioend_wq(ip
));
81 struct buffer_head
*bh
, *head
;
83 *delalloc
= *unwritten
= 0;
85 bh
= head
= page_buffers(page
);
87 if (buffer_unwritten(bh
))
89 else if (buffer_delay(bh
))
91 } while ((bh
= bh
->b_this_page
) != head
);
94 STATIC
struct block_device
*
95 xfs_find_bdev_for_inode(
98 struct xfs_inode
*ip
= XFS_I(inode
);
99 struct xfs_mount
*mp
= ip
->i_mount
;
101 if (XFS_IS_REALTIME_INODE(ip
))
102 return mp
->m_rtdev_targp
->bt_bdev
;
104 return mp
->m_ddev_targp
->bt_bdev
;
108 * We're now finished for good with this ioend structure.
109 * Update the page state via the associated buffer_heads,
110 * release holds on the inode and bio, and finally free
111 * up memory. Do not use the ioend after this.
117 struct buffer_head
*bh
, *next
;
118 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
120 for (bh
= ioend
->io_buffer_head
; bh
; bh
= next
) {
121 next
= bh
->b_private
;
122 bh
->b_end_io(bh
, !ioend
->io_error
);
126 * Volume managers supporting multiple paths can send back ENODEV
127 * when the final path disappears. In this case continuing to fill
128 * the page cache with dirty data which cannot be written out is
129 * evil, so prevent that.
131 if (unlikely(ioend
->io_error
== -ENODEV
)) {
132 xfs_do_force_shutdown(ip
->i_mount
, SHUTDOWN_DEVICE_REQ
,
137 mempool_free(ioend
, xfs_ioend_pool
);
141 * If the end of the current ioend is beyond the current EOF,
142 * return the new EOF value, otherwise zero.
148 xfs_inode_t
*ip
= XFS_I(ioend
->io_inode
);
152 bsize
= ioend
->io_offset
+ ioend
->io_size
;
153 isize
= MAX(ip
->i_size
, ip
->i_new_size
);
154 isize
= MIN(isize
, bsize
);
155 return isize
> ip
->i_d
.di_size
? isize
: 0;
159 * Update on-disk file size now that data has been written to disk. The
160 * current in-memory file size is i_size. If a write is beyond eof i_new_size
161 * will be the intended file size until i_size is updated. If this write does
162 * not extend all the way to the valid file size then restrict this update to
163 * the end of the write.
165 * This function does not block as blocking on the inode lock in IO completion
166 * can lead to IO completion order dependency deadlocks.. If it can't get the
167 * inode ilock it will return EAGAIN. Callers must handle this.
173 xfs_inode_t
*ip
= XFS_I(ioend
->io_inode
);
176 if (unlikely(ioend
->io_error
))
179 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_EXCL
))
182 isize
= xfs_ioend_new_eof(ioend
);
184 ip
->i_d
.di_size
= isize
;
185 xfs_mark_inode_dirty(ip
);
188 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
193 * Schedule IO completion handling on the final put of an ioend.
197 struct xfs_ioend
*ioend
)
199 if (atomic_dec_and_test(&ioend
->io_remaining
)) {
200 if (ioend
->io_type
== IO_UNWRITTEN
)
201 queue_work(xfsconvertd_workqueue
, &ioend
->io_work
);
203 queue_work(xfsdatad_workqueue
, &ioend
->io_work
);
208 * IO write completion.
212 struct work_struct
*work
)
214 xfs_ioend_t
*ioend
= container_of(work
, xfs_ioend_t
, io_work
);
215 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
219 * For unwritten extents we need to issue transactions to convert a
220 * range to normal written extens after the data I/O has finished.
222 if (ioend
->io_type
== IO_UNWRITTEN
&&
223 likely(!ioend
->io_error
&& !XFS_FORCED_SHUTDOWN(ip
->i_mount
))) {
225 error
= xfs_iomap_write_unwritten(ip
, ioend
->io_offset
,
228 ioend
->io_error
= error
;
232 * We might have to update the on-disk file size after extending
235 error
= xfs_setfilesize(ioend
);
236 ASSERT(!error
|| error
== EAGAIN
);
239 * If we didn't complete processing of the ioend, requeue it to the
240 * tail of the workqueue for another attempt later. Otherwise destroy
243 if (error
== EAGAIN
) {
244 atomic_inc(&ioend
->io_remaining
);
245 xfs_finish_ioend(ioend
);
246 /* ensure we don't spin on blocked ioends */
250 aio_complete(ioend
->io_iocb
, ioend
->io_result
, 0);
251 xfs_destroy_ioend(ioend
);
256 * Call IO completion handling in caller context on the final put of an ioend.
259 xfs_finish_ioend_sync(
260 struct xfs_ioend
*ioend
)
262 if (atomic_dec_and_test(&ioend
->io_remaining
))
263 xfs_end_io(&ioend
->io_work
);
267 * Allocate and initialise an IO completion structure.
268 * We need to track unwritten extent write completion here initially.
269 * We'll need to extend this for updating the ondisk inode size later
279 ioend
= mempool_alloc(xfs_ioend_pool
, GFP_NOFS
);
282 * Set the count to 1 initially, which will prevent an I/O
283 * completion callback from happening before we have started
284 * all the I/O from calling the completion routine too early.
286 atomic_set(&ioend
->io_remaining
, 1);
288 ioend
->io_list
= NULL
;
289 ioend
->io_type
= type
;
290 ioend
->io_inode
= inode
;
291 ioend
->io_buffer_head
= NULL
;
292 ioend
->io_buffer_tail
= NULL
;
293 atomic_inc(&XFS_I(ioend
->io_inode
)->i_iocount
);
294 ioend
->io_offset
= 0;
296 ioend
->io_iocb
= NULL
;
297 ioend
->io_result
= 0;
299 INIT_WORK(&ioend
->io_work
, xfs_end_io
);
307 struct xfs_bmbt_irec
*imap
,
311 struct xfs_inode
*ip
= XFS_I(inode
);
312 struct xfs_mount
*mp
= ip
->i_mount
;
313 ssize_t count
= 1 << inode
->i_blkbits
;
314 xfs_fileoff_t offset_fsb
, end_fsb
;
316 int bmapi_flags
= XFS_BMAPI_ENTIRE
;
319 if (XFS_FORCED_SHUTDOWN(mp
))
320 return -XFS_ERROR(EIO
);
322 if (type
== IO_UNWRITTEN
)
323 bmapi_flags
|= XFS_BMAPI_IGSTATE
;
325 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_SHARED
)) {
327 return -XFS_ERROR(EAGAIN
);
328 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
331 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
332 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
));
333 ASSERT(offset
<= mp
->m_maxioffset
);
335 if (offset
+ count
> mp
->m_maxioffset
)
336 count
= mp
->m_maxioffset
- offset
;
337 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ count
);
338 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
339 error
= xfs_bmapi(NULL
, ip
, offset_fsb
, end_fsb
- offset_fsb
,
340 bmapi_flags
, NULL
, 0, imap
, &nimaps
, NULL
);
341 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
344 return -XFS_ERROR(error
);
346 if (type
== IO_DELALLOC
&&
347 (!nimaps
|| isnullstartblock(imap
->br_startblock
))) {
348 error
= xfs_iomap_write_allocate(ip
, offset
, count
, imap
);
350 trace_xfs_map_blocks_alloc(ip
, offset
, count
, type
, imap
);
351 return -XFS_ERROR(error
);
355 if (type
== IO_UNWRITTEN
) {
357 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
358 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
362 trace_xfs_map_blocks_found(ip
, offset
, count
, type
, imap
);
369 struct xfs_bmbt_irec
*imap
,
372 offset
>>= inode
->i_blkbits
;
374 return offset
>= imap
->br_startoff
&&
375 offset
< imap
->br_startoff
+ imap
->br_blockcount
;
379 * BIO completion handler for buffered IO.
386 xfs_ioend_t
*ioend
= bio
->bi_private
;
388 ASSERT(atomic_read(&bio
->bi_cnt
) >= 1);
389 ioend
->io_error
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
) ? 0 : error
;
391 /* Toss bio and pass work off to an xfsdatad thread */
392 bio
->bi_private
= NULL
;
393 bio
->bi_end_io
= NULL
;
396 xfs_finish_ioend(ioend
);
400 xfs_submit_ioend_bio(
401 struct writeback_control
*wbc
,
405 atomic_inc(&ioend
->io_remaining
);
406 bio
->bi_private
= ioend
;
407 bio
->bi_end_io
= xfs_end_bio
;
410 * If the I/O is beyond EOF we mark the inode dirty immediately
411 * but don't update the inode size until I/O completion.
413 if (xfs_ioend_new_eof(ioend
))
414 xfs_mark_inode_dirty(XFS_I(ioend
->io_inode
));
416 submit_bio(wbc
->sync_mode
== WB_SYNC_ALL
? WRITE_SYNC
: WRITE
, bio
);
421 struct buffer_head
*bh
)
423 int nvecs
= bio_get_nr_vecs(bh
->b_bdev
);
424 struct bio
*bio
= bio_alloc(GFP_NOIO
, nvecs
);
426 ASSERT(bio
->bi_private
== NULL
);
427 bio
->bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
428 bio
->bi_bdev
= bh
->b_bdev
;
433 xfs_start_buffer_writeback(
434 struct buffer_head
*bh
)
436 ASSERT(buffer_mapped(bh
));
437 ASSERT(buffer_locked(bh
));
438 ASSERT(!buffer_delay(bh
));
439 ASSERT(!buffer_unwritten(bh
));
441 mark_buffer_async_write(bh
);
442 set_buffer_uptodate(bh
);
443 clear_buffer_dirty(bh
);
447 xfs_start_page_writeback(
452 ASSERT(PageLocked(page
));
453 ASSERT(!PageWriteback(page
));
455 clear_page_dirty_for_io(page
);
456 set_page_writeback(page
);
458 /* If no buffers on the page are to be written, finish it here */
460 end_page_writeback(page
);
463 static inline int bio_add_buffer(struct bio
*bio
, struct buffer_head
*bh
)
465 return bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
469 * Submit all of the bios for all of the ioends we have saved up, covering the
470 * initial writepage page and also any probed pages.
472 * Because we may have multiple ioends spanning a page, we need to start
473 * writeback on all the buffers before we submit them for I/O. If we mark the
474 * buffers as we got, then we can end up with a page that only has buffers
475 * marked async write and I/O complete on can occur before we mark the other
476 * buffers async write.
478 * The end result of this is that we trip a bug in end_page_writeback() because
479 * we call it twice for the one page as the code in end_buffer_async_write()
480 * assumes that all buffers on the page are started at the same time.
482 * The fix is two passes across the ioend list - one to start writeback on the
483 * buffer_heads, and then submit them for I/O on the second pass.
487 struct writeback_control
*wbc
,
490 xfs_ioend_t
*head
= ioend
;
492 struct buffer_head
*bh
;
494 sector_t lastblock
= 0;
496 /* Pass 1 - start writeback */
498 next
= ioend
->io_list
;
499 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
)
500 xfs_start_buffer_writeback(bh
);
501 } while ((ioend
= next
) != NULL
);
503 /* Pass 2 - submit I/O */
506 next
= ioend
->io_list
;
509 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
513 bio
= xfs_alloc_ioend_bio(bh
);
514 } else if (bh
->b_blocknr
!= lastblock
+ 1) {
515 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
519 if (bio_add_buffer(bio
, bh
) != bh
->b_size
) {
520 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
524 lastblock
= bh
->b_blocknr
;
527 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
528 xfs_finish_ioend(ioend
);
529 } while ((ioend
= next
) != NULL
);
533 * Cancel submission of all buffer_heads so far in this endio.
534 * Toss the endio too. Only ever called for the initial page
535 * in a writepage request, so only ever one page.
542 struct buffer_head
*bh
, *next_bh
;
545 next
= ioend
->io_list
;
546 bh
= ioend
->io_buffer_head
;
548 next_bh
= bh
->b_private
;
549 clear_buffer_async_write(bh
);
551 } while ((bh
= next_bh
) != NULL
);
553 xfs_ioend_wake(XFS_I(ioend
->io_inode
));
554 mempool_free(ioend
, xfs_ioend_pool
);
555 } while ((ioend
= next
) != NULL
);
559 * Test to see if we've been building up a completion structure for
560 * earlier buffers -- if so, we try to append to this ioend if we
561 * can, otherwise we finish off any current ioend and start another.
562 * Return true if we've finished the given ioend.
567 struct buffer_head
*bh
,
570 xfs_ioend_t
**result
,
573 xfs_ioend_t
*ioend
= *result
;
575 if (!ioend
|| need_ioend
|| type
!= ioend
->io_type
) {
576 xfs_ioend_t
*previous
= *result
;
578 ioend
= xfs_alloc_ioend(inode
, type
);
579 ioend
->io_offset
= offset
;
580 ioend
->io_buffer_head
= bh
;
581 ioend
->io_buffer_tail
= bh
;
583 previous
->io_list
= ioend
;
586 ioend
->io_buffer_tail
->b_private
= bh
;
587 ioend
->io_buffer_tail
= bh
;
590 bh
->b_private
= NULL
;
591 ioend
->io_size
+= bh
->b_size
;
597 struct buffer_head
*bh
,
598 struct xfs_bmbt_irec
*imap
,
602 struct xfs_mount
*m
= XFS_I(inode
)->i_mount
;
603 xfs_off_t iomap_offset
= XFS_FSB_TO_B(m
, imap
->br_startoff
);
604 xfs_daddr_t iomap_bn
= xfs_fsb_to_db(XFS_I(inode
), imap
->br_startblock
);
606 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
607 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
609 bn
= (iomap_bn
>> (inode
->i_blkbits
- BBSHIFT
)) +
610 ((offset
- iomap_offset
) >> inode
->i_blkbits
);
612 ASSERT(bn
|| XFS_IS_REALTIME_INODE(XFS_I(inode
)));
615 set_buffer_mapped(bh
);
621 struct buffer_head
*bh
,
622 struct xfs_bmbt_irec
*imap
,
625 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
626 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
628 xfs_map_buffer(inode
, bh
, imap
, offset
);
629 set_buffer_mapped(bh
);
630 clear_buffer_delay(bh
);
631 clear_buffer_unwritten(bh
);
635 * Test if a given page is suitable for writing as part of an unwritten
636 * or delayed allocate extent.
643 if (PageWriteback(page
))
646 if (page
->mapping
&& page_has_buffers(page
)) {
647 struct buffer_head
*bh
, *head
;
650 bh
= head
= page_buffers(page
);
652 if (buffer_unwritten(bh
))
653 acceptable
= (type
== IO_UNWRITTEN
);
654 else if (buffer_delay(bh
))
655 acceptable
= (type
== IO_DELALLOC
);
656 else if (buffer_dirty(bh
) && buffer_mapped(bh
))
657 acceptable
= (type
== IO_OVERWRITE
);
660 } while ((bh
= bh
->b_this_page
) != head
);
670 * Allocate & map buffers for page given the extent map. Write it out.
671 * except for the original page of a writepage, this is called on
672 * delalloc/unwritten pages only, for the original page it is possible
673 * that the page has no mapping at all.
680 struct xfs_bmbt_irec
*imap
,
681 xfs_ioend_t
**ioendp
,
682 struct writeback_control
*wbc
)
684 struct buffer_head
*bh
, *head
;
685 xfs_off_t end_offset
;
686 unsigned long p_offset
;
689 int count
= 0, done
= 0, uptodate
= 1;
690 xfs_off_t offset
= page_offset(page
);
692 if (page
->index
!= tindex
)
694 if (!trylock_page(page
))
696 if (PageWriteback(page
))
697 goto fail_unlock_page
;
698 if (page
->mapping
!= inode
->i_mapping
)
699 goto fail_unlock_page
;
700 if (!xfs_is_delayed_page(page
, (*ioendp
)->io_type
))
701 goto fail_unlock_page
;
704 * page_dirty is initially a count of buffers on the page before
705 * EOF and is decremented as we move each into a cleanable state.
709 * End offset is the highest offset that this page should represent.
710 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
711 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
712 * hence give us the correct page_dirty count. On any other page,
713 * it will be zero and in that case we need page_dirty to be the
714 * count of buffers on the page.
716 end_offset
= min_t(unsigned long long,
717 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
720 len
= 1 << inode
->i_blkbits
;
721 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
723 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
724 page_dirty
= p_offset
/ len
;
726 bh
= head
= page_buffers(page
);
728 if (offset
>= end_offset
)
730 if (!buffer_uptodate(bh
))
732 if (!(PageUptodate(page
) || buffer_uptodate(bh
))) {
737 if (buffer_unwritten(bh
) || buffer_delay(bh
) ||
739 if (buffer_unwritten(bh
))
741 else if (buffer_delay(bh
))
746 if (!xfs_imap_valid(inode
, imap
, offset
)) {
752 if (type
!= IO_OVERWRITE
)
753 xfs_map_at_offset(inode
, bh
, imap
, offset
);
754 xfs_add_to_ioend(inode
, bh
, offset
, type
,
762 } while (offset
+= len
, (bh
= bh
->b_this_page
) != head
);
764 if (uptodate
&& bh
== head
)
765 SetPageUptodate(page
);
768 if (--wbc
->nr_to_write
<= 0 &&
769 wbc
->sync_mode
== WB_SYNC_NONE
)
772 xfs_start_page_writeback(page
, !page_dirty
, count
);
782 * Convert & write out a cluster of pages in the same extent as defined
783 * by mp and following the start page.
789 struct xfs_bmbt_irec
*imap
,
790 xfs_ioend_t
**ioendp
,
791 struct writeback_control
*wbc
,
797 pagevec_init(&pvec
, 0);
798 while (!done
&& tindex
<= tlast
) {
799 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
801 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
804 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
805 done
= xfs_convert_page(inode
, pvec
.pages
[i
], tindex
++,
811 pagevec_release(&pvec
);
817 xfs_vm_invalidatepage(
819 unsigned long offset
)
821 trace_xfs_invalidatepage(page
->mapping
->host
, page
, offset
);
822 block_invalidatepage(page
, offset
);
826 * If the page has delalloc buffers on it, we need to punch them out before we
827 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
828 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
829 * is done on that same region - the delalloc extent is returned when none is
830 * supposed to be there.
832 * We prevent this by truncating away the delalloc regions on the page before
833 * invalidating it. Because they are delalloc, we can do this without needing a
834 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
835 * truncation without a transaction as there is no space left for block
836 * reservation (typically why we see a ENOSPC in writeback).
838 * This is not a performance critical path, so for now just do the punching a
839 * buffer head at a time.
842 xfs_aops_discard_page(
845 struct inode
*inode
= page
->mapping
->host
;
846 struct xfs_inode
*ip
= XFS_I(inode
);
847 struct buffer_head
*bh
, *head
;
848 loff_t offset
= page_offset(page
);
850 if (!xfs_is_delayed_page(page
, IO_DELALLOC
))
853 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
856 xfs_alert(ip
->i_mount
,
857 "page discard on page %p, inode 0x%llx, offset %llu.",
858 page
, ip
->i_ino
, offset
);
860 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
861 bh
= head
= page_buffers(page
);
864 xfs_fileoff_t start_fsb
;
866 if (!buffer_delay(bh
))
869 start_fsb
= XFS_B_TO_FSBT(ip
->i_mount
, offset
);
870 error
= xfs_bmap_punch_delalloc_range(ip
, start_fsb
, 1);
872 /* something screwed, just bail */
873 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
874 xfs_alert(ip
->i_mount
,
875 "page discard unable to remove delalloc mapping.");
880 offset
+= 1 << inode
->i_blkbits
;
882 } while ((bh
= bh
->b_this_page
) != head
);
884 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
886 xfs_vm_invalidatepage(page
, 0);
891 * Write out a dirty page.
893 * For delalloc space on the page we need to allocate space and flush it.
894 * For unwritten space on the page we need to start the conversion to
895 * regular allocated space.
896 * For any other dirty buffer heads on the page we should flush them.
898 * If we detect that a transaction would be required to flush the page, we
899 * have to check the process flags first, if we are already in a transaction
900 * or disk I/O during allocations is off, we need to fail the writepage and
906 struct writeback_control
*wbc
)
908 struct inode
*inode
= page
->mapping
->host
;
909 int delalloc
, unwritten
;
910 struct buffer_head
*bh
, *head
;
911 struct xfs_bmbt_irec imap
;
912 xfs_ioend_t
*ioend
= NULL
, *iohead
= NULL
;
915 __uint64_t end_offset
;
916 pgoff_t end_index
, last_index
;
918 int err
, imap_valid
= 0, uptodate
= 1;
922 trace_xfs_writepage(inode
, page
, 0);
924 ASSERT(page_has_buffers(page
));
927 * Refuse to write the page out if we are called from reclaim context.
929 * This avoids stack overflows when called from deeply used stacks in
930 * random callers for direct reclaim or memcg reclaim. We explicitly
931 * allow reclaim from kswapd as the stack usage there is relatively low.
933 * This should really be done by the core VM, but until that happens
934 * filesystems like XFS, btrfs and ext4 have to take care of this
937 if ((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
)) == PF_MEMALLOC
)
941 * We need a transaction if there are delalloc or unwritten buffers
944 * If we need a transaction and the process flags say we are already
945 * in a transaction, or no IO is allowed then mark the page dirty
946 * again and leave the page as is.
948 xfs_count_page_state(page
, &delalloc
, &unwritten
);
949 if ((current
->flags
& PF_FSTRANS
) && (delalloc
|| unwritten
))
952 /* Is this page beyond the end of the file? */
953 offset
= i_size_read(inode
);
954 end_index
= offset
>> PAGE_CACHE_SHIFT
;
955 last_index
= (offset
- 1) >> PAGE_CACHE_SHIFT
;
956 if (page
->index
>= end_index
) {
957 if ((page
->index
>= end_index
+ 1) ||
958 !(i_size_read(inode
) & (PAGE_CACHE_SIZE
- 1))) {
964 end_offset
= min_t(unsigned long long,
965 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
967 len
= 1 << inode
->i_blkbits
;
969 bh
= head
= page_buffers(page
);
970 offset
= page_offset(page
);
973 if (wbc
->sync_mode
== WB_SYNC_NONE
&& wbc
->nonblocking
)
979 if (offset
>= end_offset
)
981 if (!buffer_uptodate(bh
))
985 * set_page_dirty dirties all buffers in a page, independent
986 * of their state. The dirty state however is entirely
987 * meaningless for holes (!mapped && uptodate), so skip
988 * buffers covering holes here.
990 if (!buffer_mapped(bh
) && buffer_uptodate(bh
)) {
995 if (buffer_unwritten(bh
)) {
996 if (type
!= IO_UNWRITTEN
) {
1000 } else if (buffer_delay(bh
)) {
1001 if (type
!= IO_DELALLOC
) {
1005 } else if (buffer_uptodate(bh
)) {
1006 if (type
!= IO_OVERWRITE
) {
1007 type
= IO_OVERWRITE
;
1011 if (PageUptodate(page
)) {
1012 ASSERT(buffer_mapped(bh
));
1019 imap_valid
= xfs_imap_valid(inode
, &imap
, offset
);
1022 * If we didn't have a valid mapping then we need to
1023 * put the new mapping into a separate ioend structure.
1024 * This ensures non-contiguous extents always have
1025 * separate ioends, which is particularly important
1026 * for unwritten extent conversion at I/O completion
1030 err
= xfs_map_blocks(inode
, offset
, &imap
, type
,
1034 imap_valid
= xfs_imap_valid(inode
, &imap
, offset
);
1038 if (type
!= IO_OVERWRITE
)
1039 xfs_map_at_offset(inode
, bh
, &imap
, offset
);
1040 xfs_add_to_ioend(inode
, bh
, offset
, type
, &ioend
,
1048 } while (offset
+= len
, ((bh
= bh
->b_this_page
) != head
));
1050 if (uptodate
&& bh
== head
)
1051 SetPageUptodate(page
);
1053 xfs_start_page_writeback(page
, 1, count
);
1055 if (ioend
&& imap_valid
) {
1056 xfs_off_t end_index
;
1058 end_index
= imap
.br_startoff
+ imap
.br_blockcount
;
1061 end_index
<<= inode
->i_blkbits
;
1064 end_index
= (end_index
- 1) >> PAGE_CACHE_SHIFT
;
1066 /* check against file size */
1067 if (end_index
> last_index
)
1068 end_index
= last_index
;
1070 xfs_cluster_write(inode
, page
->index
+ 1, &imap
, &ioend
,
1075 xfs_submit_ioend(wbc
, iohead
);
1081 xfs_cancel_ioend(iohead
);
1086 xfs_aops_discard_page(page
);
1087 ClearPageUptodate(page
);
1092 redirty_page_for_writepage(wbc
, page
);
1099 struct address_space
*mapping
,
1100 struct writeback_control
*wbc
)
1102 xfs_iflags_clear(XFS_I(mapping
->host
), XFS_ITRUNCATED
);
1103 return generic_writepages(mapping
, wbc
);
1107 * Called to move a page into cleanable state - and from there
1108 * to be released. The page should already be clean. We always
1109 * have buffer heads in this call.
1111 * Returns 1 if the page is ok to release, 0 otherwise.
1118 int delalloc
, unwritten
;
1120 trace_xfs_releasepage(page
->mapping
->host
, page
, 0);
1122 xfs_count_page_state(page
, &delalloc
, &unwritten
);
1124 if (WARN_ON(delalloc
))
1126 if (WARN_ON(unwritten
))
1129 return try_to_free_buffers(page
);
1134 struct inode
*inode
,
1136 struct buffer_head
*bh_result
,
1140 struct xfs_inode
*ip
= XFS_I(inode
);
1141 struct xfs_mount
*mp
= ip
->i_mount
;
1142 xfs_fileoff_t offset_fsb
, end_fsb
;
1145 struct xfs_bmbt_irec imap
;
1151 if (XFS_FORCED_SHUTDOWN(mp
))
1152 return -XFS_ERROR(EIO
);
1154 offset
= (xfs_off_t
)iblock
<< inode
->i_blkbits
;
1155 ASSERT(bh_result
->b_size
>= (1 << inode
->i_blkbits
));
1156 size
= bh_result
->b_size
;
1158 if (!create
&& direct
&& offset
>= i_size_read(inode
))
1162 lockmode
= XFS_ILOCK_EXCL
;
1163 xfs_ilock(ip
, lockmode
);
1165 lockmode
= xfs_ilock_map_shared(ip
);
1168 ASSERT(offset
<= mp
->m_maxioffset
);
1169 if (offset
+ size
> mp
->m_maxioffset
)
1170 size
= mp
->m_maxioffset
- offset
;
1171 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ size
);
1172 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1174 error
= xfs_bmapi(NULL
, ip
, offset_fsb
, end_fsb
- offset_fsb
,
1175 XFS_BMAPI_ENTIRE
, NULL
, 0, &imap
, &nimaps
, NULL
);
1181 (imap
.br_startblock
== HOLESTARTBLOCK
||
1182 imap
.br_startblock
== DELAYSTARTBLOCK
))) {
1184 error
= xfs_iomap_write_direct(ip
, offset
, size
,
1187 error
= xfs_iomap_write_delay(ip
, offset
, size
, &imap
);
1192 trace_xfs_get_blocks_alloc(ip
, offset
, size
, 0, &imap
);
1193 } else if (nimaps
) {
1194 trace_xfs_get_blocks_found(ip
, offset
, size
, 0, &imap
);
1196 trace_xfs_get_blocks_notfound(ip
, offset
, size
);
1199 xfs_iunlock(ip
, lockmode
);
1201 if (imap
.br_startblock
!= HOLESTARTBLOCK
&&
1202 imap
.br_startblock
!= DELAYSTARTBLOCK
) {
1204 * For unwritten extents do not report a disk address on
1205 * the read case (treat as if we're reading into a hole).
1207 if (create
|| !ISUNWRITTEN(&imap
))
1208 xfs_map_buffer(inode
, bh_result
, &imap
, offset
);
1209 if (create
&& ISUNWRITTEN(&imap
)) {
1211 bh_result
->b_private
= inode
;
1212 set_buffer_unwritten(bh_result
);
1217 * If this is a realtime file, data may be on a different device.
1218 * to that pointed to from the buffer_head b_bdev currently.
1220 bh_result
->b_bdev
= xfs_find_bdev_for_inode(inode
);
1223 * If we previously allocated a block out beyond eof and we are now
1224 * coming back to use it then we will need to flag it as new even if it
1225 * has a disk address.
1227 * With sub-block writes into unwritten extents we also need to mark
1228 * the buffer as new so that the unwritten parts of the buffer gets
1232 ((!buffer_mapped(bh_result
) && !buffer_uptodate(bh_result
)) ||
1233 (offset
>= i_size_read(inode
)) ||
1234 (new || ISUNWRITTEN(&imap
))))
1235 set_buffer_new(bh_result
);
1237 if (imap
.br_startblock
== DELAYSTARTBLOCK
) {
1240 set_buffer_uptodate(bh_result
);
1241 set_buffer_mapped(bh_result
);
1242 set_buffer_delay(bh_result
);
1247 * If this is O_DIRECT or the mpage code calling tell them how large
1248 * the mapping is, so that we can avoid repeated get_blocks calls.
1250 if (direct
|| size
> (1 << inode
->i_blkbits
)) {
1251 xfs_off_t mapping_size
;
1253 mapping_size
= imap
.br_startoff
+ imap
.br_blockcount
- iblock
;
1254 mapping_size
<<= inode
->i_blkbits
;
1256 ASSERT(mapping_size
> 0);
1257 if (mapping_size
> size
)
1258 mapping_size
= size
;
1259 if (mapping_size
> LONG_MAX
)
1260 mapping_size
= LONG_MAX
;
1262 bh_result
->b_size
= mapping_size
;
1268 xfs_iunlock(ip
, lockmode
);
1274 struct inode
*inode
,
1276 struct buffer_head
*bh_result
,
1279 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 0);
1283 xfs_get_blocks_direct(
1284 struct inode
*inode
,
1286 struct buffer_head
*bh_result
,
1289 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 1);
1293 * Complete a direct I/O write request.
1295 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1296 * need to issue a transaction to convert the range from unwritten to written
1297 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1298 * to do this and we are done. But in case this was a successfull AIO
1299 * request this handler is called from interrupt context, from which we
1300 * can't start transactions. In that case offload the I/O completion to
1301 * the workqueues we also use for buffered I/O completion.
1304 xfs_end_io_direct_write(
1312 struct xfs_ioend
*ioend
= iocb
->private;
1315 * blockdev_direct_IO can return an error even after the I/O
1316 * completion handler was called. Thus we need to protect
1317 * against double-freeing.
1319 iocb
->private = NULL
;
1321 ioend
->io_offset
= offset
;
1322 ioend
->io_size
= size
;
1323 if (private && size
> 0)
1324 ioend
->io_type
= IO_UNWRITTEN
;
1328 * If we are converting an unwritten extent we need to delay
1329 * the AIO completion until after the unwrittent extent
1330 * conversion has completed, otherwise do it ASAP.
1332 if (ioend
->io_type
== IO_UNWRITTEN
) {
1333 ioend
->io_iocb
= iocb
;
1334 ioend
->io_result
= ret
;
1336 aio_complete(iocb
, ret
, 0);
1338 xfs_finish_ioend(ioend
);
1340 xfs_finish_ioend_sync(ioend
);
1348 const struct iovec
*iov
,
1350 unsigned long nr_segs
)
1352 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
1353 struct block_device
*bdev
= xfs_find_bdev_for_inode(inode
);
1357 iocb
->private = xfs_alloc_ioend(inode
, IO_DIRECT
);
1359 ret
= __blockdev_direct_IO(rw
, iocb
, inode
, bdev
, iov
,
1361 xfs_get_blocks_direct
,
1362 xfs_end_io_direct_write
, NULL
, 0);
1363 if (ret
!= -EIOCBQUEUED
&& iocb
->private)
1364 xfs_destroy_ioend(iocb
->private);
1366 ret
= __blockdev_direct_IO(rw
, iocb
, inode
, bdev
, iov
,
1368 xfs_get_blocks_direct
,
1376 xfs_vm_write_failed(
1377 struct address_space
*mapping
,
1380 struct inode
*inode
= mapping
->host
;
1382 if (to
> inode
->i_size
) {
1384 * punch out the delalloc blocks we have already allocated. We
1385 * don't call xfs_setattr() to do this as we may be in the
1386 * middle of a multi-iovec write and so the vfs inode->i_size
1387 * will not match the xfs ip->i_size and so it will zero too
1388 * much. Hence we jus truncate the page cache to zero what is
1389 * necessary and punch the delalloc blocks directly.
1391 struct xfs_inode
*ip
= XFS_I(inode
);
1392 xfs_fileoff_t start_fsb
;
1393 xfs_fileoff_t end_fsb
;
1396 truncate_pagecache(inode
, to
, inode
->i_size
);
1399 * Check if there are any blocks that are outside of i_size
1400 * that need to be trimmed back.
1402 start_fsb
= XFS_B_TO_FSB(ip
->i_mount
, inode
->i_size
) + 1;
1403 end_fsb
= XFS_B_TO_FSB(ip
->i_mount
, to
);
1404 if (end_fsb
<= start_fsb
)
1407 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1408 error
= xfs_bmap_punch_delalloc_range(ip
, start_fsb
,
1409 end_fsb
- start_fsb
);
1411 /* something screwed, just bail */
1412 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
1413 xfs_alert(ip
->i_mount
,
1414 "xfs_vm_write_failed: unable to clean up ino %lld",
1418 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1425 struct address_space
*mapping
,
1429 struct page
**pagep
,
1434 ret
= block_write_begin(mapping
, pos
, len
, flags
| AOP_FLAG_NOFS
,
1435 pagep
, xfs_get_blocks
);
1437 xfs_vm_write_failed(mapping
, pos
+ len
);
1444 struct address_space
*mapping
,
1453 ret
= generic_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1454 if (unlikely(ret
< len
))
1455 xfs_vm_write_failed(mapping
, pos
+ len
);
1461 struct address_space
*mapping
,
1464 struct inode
*inode
= (struct inode
*)mapping
->host
;
1465 struct xfs_inode
*ip
= XFS_I(inode
);
1467 trace_xfs_vm_bmap(XFS_I(inode
));
1468 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
1469 xfs_flush_pages(ip
, (xfs_off_t
)0, -1, 0, FI_REMAPF
);
1470 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
1471 return generic_block_bmap(mapping
, block
, xfs_get_blocks
);
1476 struct file
*unused
,
1479 return mpage_readpage(page
, xfs_get_blocks
);
1484 struct file
*unused
,
1485 struct address_space
*mapping
,
1486 struct list_head
*pages
,
1489 return mpage_readpages(mapping
, pages
, nr_pages
, xfs_get_blocks
);
1492 const struct address_space_operations xfs_address_space_operations
= {
1493 .readpage
= xfs_vm_readpage
,
1494 .readpages
= xfs_vm_readpages
,
1495 .writepage
= xfs_vm_writepage
,
1496 .writepages
= xfs_vm_writepages
,
1497 .releasepage
= xfs_vm_releasepage
,
1498 .invalidatepage
= xfs_vm_invalidatepage
,
1499 .write_begin
= xfs_vm_write_begin
,
1500 .write_end
= xfs_vm_write_end
,
1501 .bmap
= xfs_vm_bmap
,
1502 .direct_IO
= xfs_vm_direct_IO
,
1503 .migratepage
= buffer_migrate_page
,
1504 .is_partially_uptodate
= block_is_partially_uptodate
,
1505 .error_remove_page
= generic_error_remove_page
,