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
22 #include "xfs_trans.h"
23 #include "xfs_mount.h"
24 #include "xfs_bmap_btree.h"
25 #include "xfs_dinode.h"
26 #include "xfs_inode.h"
27 #include "xfs_inode_item.h"
28 #include "xfs_alloc.h"
29 #include "xfs_error.h"
30 #include "xfs_iomap.h"
31 #include "xfs_vnodeops.h"
32 #include "xfs_trace.h"
34 #include <linux/gfp.h>
35 #include <linux/mpage.h>
36 #include <linux/pagevec.h>
37 #include <linux/writeback.h>
45 struct buffer_head
*bh
, *head
;
47 *delalloc
= *unwritten
= 0;
49 bh
= head
= page_buffers(page
);
51 if (buffer_unwritten(bh
))
53 else if (buffer_delay(bh
))
55 } while ((bh
= bh
->b_this_page
) != head
);
58 STATIC
struct block_device
*
59 xfs_find_bdev_for_inode(
62 struct xfs_inode
*ip
= XFS_I(inode
);
63 struct xfs_mount
*mp
= ip
->i_mount
;
65 if (XFS_IS_REALTIME_INODE(ip
))
66 return mp
->m_rtdev_targp
->bt_bdev
;
68 return mp
->m_ddev_targp
->bt_bdev
;
72 * We're now finished for good with this ioend structure.
73 * Update the page state via the associated buffer_heads,
74 * release holds on the inode and bio, and finally free
75 * up memory. Do not use the ioend after this.
81 struct buffer_head
*bh
, *next
;
83 for (bh
= ioend
->io_buffer_head
; bh
; bh
= next
) {
85 bh
->b_end_io(bh
, !ioend
->io_error
);
89 if (ioend
->io_isasync
) {
90 aio_complete(ioend
->io_iocb
, ioend
->io_error
?
91 ioend
->io_error
: ioend
->io_result
, 0);
93 inode_dio_done(ioend
->io_inode
);
96 mempool_free(ioend
, xfs_ioend_pool
);
100 * Fast and loose check if this write could update the on-disk inode size.
102 static inline bool xfs_ioend_is_append(struct xfs_ioend
*ioend
)
104 return ioend
->io_offset
+ ioend
->io_size
>
105 XFS_I(ioend
->io_inode
)->i_d
.di_size
;
109 xfs_setfilesize_trans_alloc(
110 struct xfs_ioend
*ioend
)
112 struct xfs_mount
*mp
= XFS_I(ioend
->io_inode
)->i_mount
;
113 struct xfs_trans
*tp
;
116 tp
= xfs_trans_alloc(mp
, XFS_TRANS_FSYNC_TS
);
118 error
= xfs_trans_reserve(tp
, 0, XFS_FSYNC_TS_LOG_RES(mp
), 0, 0, 0);
120 xfs_trans_cancel(tp
, 0);
124 ioend
->io_append_trans
= tp
;
127 * We will pass freeze protection with a transaction. So tell lockdep
130 rwsem_release(&ioend
->io_inode
->i_sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
133 * We hand off the transaction to the completion thread now, so
134 * clear the flag here.
136 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
141 * Update on-disk file size now that data has been written to disk.
145 struct xfs_ioend
*ioend
)
147 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
148 struct xfs_trans
*tp
= ioend
->io_append_trans
;
152 * The transaction was allocated in the I/O submission thread,
153 * thus we need to mark ourselves as beeing in a transaction
156 current_set_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
158 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
159 isize
= xfs_new_eof(ip
, ioend
->io_offset
+ ioend
->io_size
);
161 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
162 xfs_trans_cancel(tp
, 0);
166 trace_xfs_setfilesize(ip
, ioend
->io_offset
, ioend
->io_size
);
168 ip
->i_d
.di_size
= isize
;
169 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
170 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
172 return xfs_trans_commit(tp
, 0);
176 * Schedule IO completion handling on the final put of an ioend.
178 * If there is no work to do we might as well call it a day and free the
183 struct xfs_ioend
*ioend
)
185 if (atomic_dec_and_test(&ioend
->io_remaining
)) {
186 struct xfs_mount
*mp
= XFS_I(ioend
->io_inode
)->i_mount
;
188 if (ioend
->io_type
== XFS_IO_UNWRITTEN
)
189 queue_work(mp
->m_unwritten_workqueue
, &ioend
->io_work
);
190 else if (ioend
->io_append_trans
)
191 queue_work(mp
->m_data_workqueue
, &ioend
->io_work
);
193 xfs_destroy_ioend(ioend
);
198 * IO write completion.
202 struct work_struct
*work
)
204 xfs_ioend_t
*ioend
= container_of(work
, xfs_ioend_t
, io_work
);
205 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
208 if (ioend
->io_append_trans
) {
210 * We've got freeze protection passed with the transaction.
211 * Tell lockdep about it.
214 &ioend
->io_inode
->i_sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
217 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
218 ioend
->io_error
= -EIO
;
225 * For unwritten extents we need to issue transactions to convert a
226 * range to normal written extens after the data I/O has finished.
228 if (ioend
->io_type
== XFS_IO_UNWRITTEN
) {
230 * For buffered I/O we never preallocate a transaction when
231 * doing the unwritten extent conversion, but for direct I/O
232 * we do not know if we are converting an unwritten extent
233 * or not at the point where we preallocate the transaction.
235 if (ioend
->io_append_trans
) {
236 ASSERT(ioend
->io_isdirect
);
238 current_set_flags_nested(
239 &ioend
->io_append_trans
->t_pflags
, PF_FSTRANS
);
240 xfs_trans_cancel(ioend
->io_append_trans
, 0);
243 error
= xfs_iomap_write_unwritten(ip
, ioend
->io_offset
,
246 ioend
->io_error
= -error
;
249 } else if (ioend
->io_append_trans
) {
250 error
= xfs_setfilesize(ioend
);
252 ioend
->io_error
= -error
;
254 ASSERT(!xfs_ioend_is_append(ioend
));
258 xfs_destroy_ioend(ioend
);
262 * Call IO completion handling in caller context on the final put of an ioend.
265 xfs_finish_ioend_sync(
266 struct xfs_ioend
*ioend
)
268 if (atomic_dec_and_test(&ioend
->io_remaining
))
269 xfs_end_io(&ioend
->io_work
);
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);
293 ioend
->io_isasync
= 0;
294 ioend
->io_isdirect
= 0;
296 ioend
->io_list
= NULL
;
297 ioend
->io_type
= type
;
298 ioend
->io_inode
= inode
;
299 ioend
->io_buffer_head
= NULL
;
300 ioend
->io_buffer_tail
= NULL
;
301 ioend
->io_offset
= 0;
303 ioend
->io_iocb
= NULL
;
304 ioend
->io_result
= 0;
305 ioend
->io_append_trans
= NULL
;
307 INIT_WORK(&ioend
->io_work
, xfs_end_io
);
315 struct xfs_bmbt_irec
*imap
,
319 struct xfs_inode
*ip
= XFS_I(inode
);
320 struct xfs_mount
*mp
= ip
->i_mount
;
321 ssize_t count
= 1 << inode
->i_blkbits
;
322 xfs_fileoff_t offset_fsb
, end_fsb
;
324 int bmapi_flags
= XFS_BMAPI_ENTIRE
;
327 if (XFS_FORCED_SHUTDOWN(mp
))
328 return -XFS_ERROR(EIO
);
330 if (type
== XFS_IO_UNWRITTEN
)
331 bmapi_flags
|= XFS_BMAPI_IGSTATE
;
333 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_SHARED
)) {
335 return -XFS_ERROR(EAGAIN
);
336 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
339 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
340 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
));
341 ASSERT(offset
<= mp
->m_super
->s_maxbytes
);
343 if (offset
+ count
> mp
->m_super
->s_maxbytes
)
344 count
= mp
->m_super
->s_maxbytes
- offset
;
345 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ count
);
346 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
347 error
= xfs_bmapi_read(ip
, offset_fsb
, end_fsb
- offset_fsb
,
348 imap
, &nimaps
, bmapi_flags
);
349 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
352 return -XFS_ERROR(error
);
354 if (type
== XFS_IO_DELALLOC
&&
355 (!nimaps
|| isnullstartblock(imap
->br_startblock
))) {
356 error
= xfs_iomap_write_allocate(ip
, offset
, count
, imap
);
358 trace_xfs_map_blocks_alloc(ip
, offset
, count
, type
, imap
);
359 return -XFS_ERROR(error
);
363 if (type
== XFS_IO_UNWRITTEN
) {
365 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
366 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
370 trace_xfs_map_blocks_found(ip
, offset
, count
, type
, imap
);
377 struct xfs_bmbt_irec
*imap
,
380 offset
>>= inode
->i_blkbits
;
382 return offset
>= imap
->br_startoff
&&
383 offset
< imap
->br_startoff
+ imap
->br_blockcount
;
387 * BIO completion handler for buffered IO.
394 xfs_ioend_t
*ioend
= bio
->bi_private
;
396 ASSERT(atomic_read(&bio
->bi_cnt
) >= 1);
397 ioend
->io_error
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
) ? 0 : error
;
399 /* Toss bio and pass work off to an xfsdatad thread */
400 bio
->bi_private
= NULL
;
401 bio
->bi_end_io
= NULL
;
404 xfs_finish_ioend(ioend
);
408 xfs_submit_ioend_bio(
409 struct writeback_control
*wbc
,
413 atomic_inc(&ioend
->io_remaining
);
414 bio
->bi_private
= ioend
;
415 bio
->bi_end_io
= xfs_end_bio
;
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 mempool_free(ioend
, xfs_ioend_pool
);
554 } while ((ioend
= next
) != NULL
);
558 * Test to see if we've been building up a completion structure for
559 * earlier buffers -- if so, we try to append to this ioend if we
560 * can, otherwise we finish off any current ioend and start another.
561 * Return true if we've finished the given ioend.
566 struct buffer_head
*bh
,
569 xfs_ioend_t
**result
,
572 xfs_ioend_t
*ioend
= *result
;
574 if (!ioend
|| need_ioend
|| type
!= ioend
->io_type
) {
575 xfs_ioend_t
*previous
= *result
;
577 ioend
= xfs_alloc_ioend(inode
, type
);
578 ioend
->io_offset
= offset
;
579 ioend
->io_buffer_head
= bh
;
580 ioend
->io_buffer_tail
= bh
;
582 previous
->io_list
= ioend
;
585 ioend
->io_buffer_tail
->b_private
= bh
;
586 ioend
->io_buffer_tail
= bh
;
589 bh
->b_private
= NULL
;
590 ioend
->io_size
+= bh
->b_size
;
596 struct buffer_head
*bh
,
597 struct xfs_bmbt_irec
*imap
,
601 struct xfs_mount
*m
= XFS_I(inode
)->i_mount
;
602 xfs_off_t iomap_offset
= XFS_FSB_TO_B(m
, imap
->br_startoff
);
603 xfs_daddr_t iomap_bn
= xfs_fsb_to_db(XFS_I(inode
), imap
->br_startblock
);
605 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
606 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
608 bn
= (iomap_bn
>> (inode
->i_blkbits
- BBSHIFT
)) +
609 ((offset
- iomap_offset
) >> inode
->i_blkbits
);
611 ASSERT(bn
|| XFS_IS_REALTIME_INODE(XFS_I(inode
)));
614 set_buffer_mapped(bh
);
620 struct buffer_head
*bh
,
621 struct xfs_bmbt_irec
*imap
,
624 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
625 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
627 xfs_map_buffer(inode
, bh
, imap
, offset
);
628 set_buffer_mapped(bh
);
629 clear_buffer_delay(bh
);
630 clear_buffer_unwritten(bh
);
634 * Test if a given page is suitable for writing as part of an unwritten
635 * or delayed allocate extent.
642 if (PageWriteback(page
))
645 if (page
->mapping
&& page_has_buffers(page
)) {
646 struct buffer_head
*bh
, *head
;
649 bh
= head
= page_buffers(page
);
651 if (buffer_unwritten(bh
))
652 acceptable
+= (type
== XFS_IO_UNWRITTEN
);
653 else if (buffer_delay(bh
))
654 acceptable
+= (type
== XFS_IO_DELALLOC
);
655 else if (buffer_dirty(bh
) && buffer_mapped(bh
))
656 acceptable
+= (type
== XFS_IO_OVERWRITE
);
659 } while ((bh
= bh
->b_this_page
) != head
);
669 * Allocate & map buffers for page given the extent map. Write it out.
670 * except for the original page of a writepage, this is called on
671 * delalloc/unwritten pages only, for the original page it is possible
672 * that the page has no mapping at all.
679 struct xfs_bmbt_irec
*imap
,
680 xfs_ioend_t
**ioendp
,
681 struct writeback_control
*wbc
)
683 struct buffer_head
*bh
, *head
;
684 xfs_off_t end_offset
;
685 unsigned long p_offset
;
688 int count
= 0, done
= 0, uptodate
= 1;
689 xfs_off_t offset
= page_offset(page
);
691 if (page
->index
!= tindex
)
693 if (!trylock_page(page
))
695 if (PageWriteback(page
))
696 goto fail_unlock_page
;
697 if (page
->mapping
!= inode
->i_mapping
)
698 goto fail_unlock_page
;
699 if (!xfs_check_page_type(page
, (*ioendp
)->io_type
))
700 goto fail_unlock_page
;
703 * page_dirty is initially a count of buffers on the page before
704 * EOF and is decremented as we move each into a cleanable state.
708 * End offset is the highest offset that this page should represent.
709 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
710 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
711 * hence give us the correct page_dirty count. On any other page,
712 * it will be zero and in that case we need page_dirty to be the
713 * count of buffers on the page.
715 end_offset
= min_t(unsigned long long,
716 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
719 len
= 1 << inode
->i_blkbits
;
720 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
722 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
723 page_dirty
= p_offset
/ len
;
725 bh
= head
= page_buffers(page
);
727 if (offset
>= end_offset
)
729 if (!buffer_uptodate(bh
))
731 if (!(PageUptodate(page
) || buffer_uptodate(bh
))) {
736 if (buffer_unwritten(bh
) || buffer_delay(bh
) ||
738 if (buffer_unwritten(bh
))
739 type
= XFS_IO_UNWRITTEN
;
740 else if (buffer_delay(bh
))
741 type
= XFS_IO_DELALLOC
;
743 type
= XFS_IO_OVERWRITE
;
745 if (!xfs_imap_valid(inode
, imap
, offset
)) {
751 if (type
!= XFS_IO_OVERWRITE
)
752 xfs_map_at_offset(inode
, bh
, imap
, offset
);
753 xfs_add_to_ioend(inode
, bh
, offset
, type
,
761 } while (offset
+= len
, (bh
= bh
->b_this_page
) != head
);
763 if (uptodate
&& bh
== head
)
764 SetPageUptodate(page
);
767 if (--wbc
->nr_to_write
<= 0 &&
768 wbc
->sync_mode
== WB_SYNC_NONE
)
771 xfs_start_page_writeback(page
, !page_dirty
, count
);
781 * Convert & write out a cluster of pages in the same extent as defined
782 * by mp and following the start page.
788 struct xfs_bmbt_irec
*imap
,
789 xfs_ioend_t
**ioendp
,
790 struct writeback_control
*wbc
,
796 pagevec_init(&pvec
, 0);
797 while (!done
&& tindex
<= tlast
) {
798 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
800 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
803 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
804 done
= xfs_convert_page(inode
, pvec
.pages
[i
], tindex
++,
810 pagevec_release(&pvec
);
816 xfs_vm_invalidatepage(
818 unsigned long offset
)
820 trace_xfs_invalidatepage(page
->mapping
->host
, page
, offset
);
821 block_invalidatepage(page
, offset
);
825 * If the page has delalloc buffers on it, we need to punch them out before we
826 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
827 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
828 * is done on that same region - the delalloc extent is returned when none is
829 * supposed to be there.
831 * We prevent this by truncating away the delalloc regions on the page before
832 * invalidating it. Because they are delalloc, we can do this without needing a
833 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
834 * truncation without a transaction as there is no space left for block
835 * reservation (typically why we see a ENOSPC in writeback).
837 * This is not a performance critical path, so for now just do the punching a
838 * buffer head at a time.
841 xfs_aops_discard_page(
844 struct inode
*inode
= page
->mapping
->host
;
845 struct xfs_inode
*ip
= XFS_I(inode
);
846 struct buffer_head
*bh
, *head
;
847 loff_t offset
= page_offset(page
);
849 if (!xfs_check_page_type(page
, XFS_IO_DELALLOC
))
852 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
855 xfs_alert(ip
->i_mount
,
856 "page discard on page %p, inode 0x%llx, offset %llu.",
857 page
, ip
->i_ino
, offset
);
859 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
860 bh
= head
= page_buffers(page
);
863 xfs_fileoff_t start_fsb
;
865 if (!buffer_delay(bh
))
868 start_fsb
= XFS_B_TO_FSBT(ip
->i_mount
, offset
);
869 error
= xfs_bmap_punch_delalloc_range(ip
, start_fsb
, 1);
871 /* something screwed, just bail */
872 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
873 xfs_alert(ip
->i_mount
,
874 "page discard unable to remove delalloc mapping.");
879 offset
+= 1 << inode
->i_blkbits
;
881 } while ((bh
= bh
->b_this_page
) != head
);
883 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
885 xfs_vm_invalidatepage(page
, 0);
890 * Write out a dirty page.
892 * For delalloc space on the page we need to allocate space and flush it.
893 * For unwritten space on the page we need to start the conversion to
894 * regular allocated space.
895 * For any other dirty buffer heads on the page we should flush them.
900 struct writeback_control
*wbc
)
902 struct inode
*inode
= page
->mapping
->host
;
903 struct buffer_head
*bh
, *head
;
904 struct xfs_bmbt_irec imap
;
905 xfs_ioend_t
*ioend
= NULL
, *iohead
= NULL
;
908 __uint64_t end_offset
;
909 pgoff_t end_index
, last_index
;
911 int err
, imap_valid
= 0, uptodate
= 1;
915 trace_xfs_writepage(inode
, page
, 0);
917 ASSERT(page_has_buffers(page
));
920 * Refuse to write the page out if we are called from reclaim context.
922 * This avoids stack overflows when called from deeply used stacks in
923 * random callers for direct reclaim or memcg reclaim. We explicitly
924 * allow reclaim from kswapd as the stack usage there is relatively low.
926 * This should never happen except in the case of a VM regression so
929 if (WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
)) ==
934 * Given that we do not allow direct reclaim to call us, we should
935 * never be called while in a filesystem transaction.
937 if (WARN_ON(current
->flags
& PF_FSTRANS
))
940 /* Is this page beyond the end of the file? */
941 offset
= i_size_read(inode
);
942 end_index
= offset
>> PAGE_CACHE_SHIFT
;
943 last_index
= (offset
- 1) >> PAGE_CACHE_SHIFT
;
944 if (page
->index
>= end_index
) {
945 unsigned offset_into_page
= offset
& (PAGE_CACHE_SIZE
- 1);
948 * Just skip the page if it is fully outside i_size, e.g. due
949 * to a truncate operation that is in progress.
951 if (page
->index
>= end_index
+ 1 || offset_into_page
== 0) {
957 * The page straddles i_size. It must be zeroed out on each
958 * and every writepage invocation because it may be mmapped.
959 * "A file is mapped in multiples of the page size. For a file
960 * that is not a multiple of the page size, the remaining
961 * memory is zeroed when mapped, and writes to that region are
962 * not written out to the file."
964 zero_user_segment(page
, offset_into_page
, PAGE_CACHE_SIZE
);
967 end_offset
= min_t(unsigned long long,
968 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
970 len
= 1 << inode
->i_blkbits
;
972 bh
= head
= page_buffers(page
);
973 offset
= page_offset(page
);
974 type
= XFS_IO_OVERWRITE
;
976 if (wbc
->sync_mode
== WB_SYNC_NONE
)
982 if (offset
>= end_offset
)
984 if (!buffer_uptodate(bh
))
988 * set_page_dirty dirties all buffers in a page, independent
989 * of their state. The dirty state however is entirely
990 * meaningless for holes (!mapped && uptodate), so skip
991 * buffers covering holes here.
993 if (!buffer_mapped(bh
) && buffer_uptodate(bh
)) {
998 if (buffer_unwritten(bh
)) {
999 if (type
!= XFS_IO_UNWRITTEN
) {
1000 type
= XFS_IO_UNWRITTEN
;
1003 } else if (buffer_delay(bh
)) {
1004 if (type
!= XFS_IO_DELALLOC
) {
1005 type
= XFS_IO_DELALLOC
;
1008 } else if (buffer_uptodate(bh
)) {
1009 if (type
!= XFS_IO_OVERWRITE
) {
1010 type
= XFS_IO_OVERWRITE
;
1014 if (PageUptodate(page
))
1015 ASSERT(buffer_mapped(bh
));
1017 * This buffer is not uptodate and will not be
1018 * written to disk. Ensure that we will put any
1019 * subsequent writeable buffers into a new
1027 imap_valid
= xfs_imap_valid(inode
, &imap
, offset
);
1030 * If we didn't have a valid mapping then we need to
1031 * put the new mapping into a separate ioend structure.
1032 * This ensures non-contiguous extents always have
1033 * separate ioends, which is particularly important
1034 * for unwritten extent conversion at I/O completion
1038 err
= xfs_map_blocks(inode
, offset
, &imap
, type
,
1042 imap_valid
= xfs_imap_valid(inode
, &imap
, offset
);
1046 if (type
!= XFS_IO_OVERWRITE
)
1047 xfs_map_at_offset(inode
, bh
, &imap
, offset
);
1048 xfs_add_to_ioend(inode
, bh
, offset
, type
, &ioend
,
1056 } while (offset
+= len
, ((bh
= bh
->b_this_page
) != head
));
1058 if (uptodate
&& bh
== head
)
1059 SetPageUptodate(page
);
1061 xfs_start_page_writeback(page
, 1, count
);
1063 if (ioend
&& imap_valid
) {
1064 xfs_off_t end_index
;
1066 end_index
= imap
.br_startoff
+ imap
.br_blockcount
;
1069 end_index
<<= inode
->i_blkbits
;
1072 end_index
= (end_index
- 1) >> PAGE_CACHE_SHIFT
;
1074 /* check against file size */
1075 if (end_index
> last_index
)
1076 end_index
= last_index
;
1078 xfs_cluster_write(inode
, page
->index
+ 1, &imap
, &ioend
,
1084 * Reserve log space if we might write beyond the on-disk
1087 if (ioend
->io_type
!= XFS_IO_UNWRITTEN
&&
1088 xfs_ioend_is_append(ioend
)) {
1089 err
= xfs_setfilesize_trans_alloc(ioend
);
1094 xfs_submit_ioend(wbc
, iohead
);
1101 xfs_cancel_ioend(iohead
);
1106 xfs_aops_discard_page(page
);
1107 ClearPageUptodate(page
);
1112 redirty_page_for_writepage(wbc
, page
);
1119 struct address_space
*mapping
,
1120 struct writeback_control
*wbc
)
1122 xfs_iflags_clear(XFS_I(mapping
->host
), XFS_ITRUNCATED
);
1123 return generic_writepages(mapping
, wbc
);
1127 * Called to move a page into cleanable state - and from there
1128 * to be released. The page should already be clean. We always
1129 * have buffer heads in this call.
1131 * Returns 1 if the page is ok to release, 0 otherwise.
1138 int delalloc
, unwritten
;
1140 trace_xfs_releasepage(page
->mapping
->host
, page
, 0);
1142 xfs_count_page_state(page
, &delalloc
, &unwritten
);
1144 if (WARN_ON(delalloc
))
1146 if (WARN_ON(unwritten
))
1149 return try_to_free_buffers(page
);
1154 struct inode
*inode
,
1156 struct buffer_head
*bh_result
,
1160 struct xfs_inode
*ip
= XFS_I(inode
);
1161 struct xfs_mount
*mp
= ip
->i_mount
;
1162 xfs_fileoff_t offset_fsb
, end_fsb
;
1165 struct xfs_bmbt_irec imap
;
1171 if (XFS_FORCED_SHUTDOWN(mp
))
1172 return -XFS_ERROR(EIO
);
1174 offset
= (xfs_off_t
)iblock
<< inode
->i_blkbits
;
1175 ASSERT(bh_result
->b_size
>= (1 << inode
->i_blkbits
));
1176 size
= bh_result
->b_size
;
1178 if (!create
&& direct
&& offset
>= i_size_read(inode
))
1182 * Direct I/O is usually done on preallocated files, so try getting
1183 * a block mapping without an exclusive lock first. For buffered
1184 * writes we already have the exclusive iolock anyway, so avoiding
1185 * a lock roundtrip here by taking the ilock exclusive from the
1186 * beginning is a useful micro optimization.
1188 if (create
&& !direct
) {
1189 lockmode
= XFS_ILOCK_EXCL
;
1190 xfs_ilock(ip
, lockmode
);
1192 lockmode
= xfs_ilock_map_shared(ip
);
1195 ASSERT(offset
<= mp
->m_super
->s_maxbytes
);
1196 if (offset
+ size
> mp
->m_super
->s_maxbytes
)
1197 size
= mp
->m_super
->s_maxbytes
- offset
;
1198 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ size
);
1199 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1201 error
= xfs_bmapi_read(ip
, offset_fsb
, end_fsb
- offset_fsb
,
1202 &imap
, &nimaps
, XFS_BMAPI_ENTIRE
);
1208 (imap
.br_startblock
== HOLESTARTBLOCK
||
1209 imap
.br_startblock
== DELAYSTARTBLOCK
))) {
1210 if (direct
|| xfs_get_extsz_hint(ip
)) {
1212 * Drop the ilock in preparation for starting the block
1213 * allocation transaction. It will be retaken
1214 * exclusively inside xfs_iomap_write_direct for the
1215 * actual allocation.
1217 xfs_iunlock(ip
, lockmode
);
1218 error
= xfs_iomap_write_direct(ip
, offset
, size
,
1225 * Delalloc reservations do not require a transaction,
1226 * we can go on without dropping the lock here. If we
1227 * are allocating a new delalloc block, make sure that
1228 * we set the new flag so that we mark the buffer new so
1229 * that we know that it is newly allocated if the write
1232 if (nimaps
&& imap
.br_startblock
== HOLESTARTBLOCK
)
1234 error
= xfs_iomap_write_delay(ip
, offset
, size
, &imap
);
1238 xfs_iunlock(ip
, lockmode
);
1241 trace_xfs_get_blocks_alloc(ip
, offset
, size
, 0, &imap
);
1242 } else if (nimaps
) {
1243 trace_xfs_get_blocks_found(ip
, offset
, size
, 0, &imap
);
1244 xfs_iunlock(ip
, lockmode
);
1246 trace_xfs_get_blocks_notfound(ip
, offset
, size
);
1250 if (imap
.br_startblock
!= HOLESTARTBLOCK
&&
1251 imap
.br_startblock
!= DELAYSTARTBLOCK
) {
1253 * For unwritten extents do not report a disk address on
1254 * the read case (treat as if we're reading into a hole).
1256 if (create
|| !ISUNWRITTEN(&imap
))
1257 xfs_map_buffer(inode
, bh_result
, &imap
, offset
);
1258 if (create
&& ISUNWRITTEN(&imap
)) {
1260 bh_result
->b_private
= inode
;
1261 set_buffer_unwritten(bh_result
);
1266 * If this is a realtime file, data may be on a different device.
1267 * to that pointed to from the buffer_head b_bdev currently.
1269 bh_result
->b_bdev
= xfs_find_bdev_for_inode(inode
);
1272 * If we previously allocated a block out beyond eof and we are now
1273 * coming back to use it then we will need to flag it as new even if it
1274 * has a disk address.
1276 * With sub-block writes into unwritten extents we also need to mark
1277 * the buffer as new so that the unwritten parts of the buffer gets
1281 ((!buffer_mapped(bh_result
) && !buffer_uptodate(bh_result
)) ||
1282 (offset
>= i_size_read(inode
)) ||
1283 (new || ISUNWRITTEN(&imap
))))
1284 set_buffer_new(bh_result
);
1286 if (imap
.br_startblock
== DELAYSTARTBLOCK
) {
1289 set_buffer_uptodate(bh_result
);
1290 set_buffer_mapped(bh_result
);
1291 set_buffer_delay(bh_result
);
1296 * If this is O_DIRECT or the mpage code calling tell them how large
1297 * the mapping is, so that we can avoid repeated get_blocks calls.
1299 if (direct
|| size
> (1 << inode
->i_blkbits
)) {
1300 xfs_off_t mapping_size
;
1302 mapping_size
= imap
.br_startoff
+ imap
.br_blockcount
- iblock
;
1303 mapping_size
<<= inode
->i_blkbits
;
1305 ASSERT(mapping_size
> 0);
1306 if (mapping_size
> size
)
1307 mapping_size
= size
;
1308 if (mapping_size
> LONG_MAX
)
1309 mapping_size
= LONG_MAX
;
1311 bh_result
->b_size
= mapping_size
;
1317 xfs_iunlock(ip
, lockmode
);
1323 struct inode
*inode
,
1325 struct buffer_head
*bh_result
,
1328 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 0);
1332 xfs_get_blocks_direct(
1333 struct inode
*inode
,
1335 struct buffer_head
*bh_result
,
1338 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 1);
1342 * Complete a direct I/O write request.
1344 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1345 * need to issue a transaction to convert the range from unwritten to written
1346 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1347 * to do this and we are done. But in case this was a successful AIO
1348 * request this handler is called from interrupt context, from which we
1349 * can't start transactions. In that case offload the I/O completion to
1350 * the workqueues we also use for buffered I/O completion.
1353 xfs_end_io_direct_write(
1361 struct xfs_ioend
*ioend
= iocb
->private;
1364 * While the generic direct I/O code updates the inode size, it does
1365 * so only after the end_io handler is called, which means our
1366 * end_io handler thinks the on-disk size is outside the in-core
1367 * size. To prevent this just update it a little bit earlier here.
1369 if (offset
+ size
> i_size_read(ioend
->io_inode
))
1370 i_size_write(ioend
->io_inode
, offset
+ size
);
1373 * blockdev_direct_IO can return an error even after the I/O
1374 * completion handler was called. Thus we need to protect
1375 * against double-freeing.
1377 iocb
->private = NULL
;
1379 ioend
->io_offset
= offset
;
1380 ioend
->io_size
= size
;
1381 ioend
->io_iocb
= iocb
;
1382 ioend
->io_result
= ret
;
1383 if (private && size
> 0)
1384 ioend
->io_type
= XFS_IO_UNWRITTEN
;
1387 ioend
->io_isasync
= 1;
1388 xfs_finish_ioend(ioend
);
1390 xfs_finish_ioend_sync(ioend
);
1398 const struct iovec
*iov
,
1400 unsigned long nr_segs
)
1402 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
1403 struct block_device
*bdev
= xfs_find_bdev_for_inode(inode
);
1404 struct xfs_ioend
*ioend
= NULL
;
1408 size_t size
= iov_length(iov
, nr_segs
);
1411 * We need to preallocate a transaction for a size update
1412 * here. In the case that this write both updates the size
1413 * and converts at least on unwritten extent we will cancel
1414 * the still clean transaction after the I/O has finished.
1416 iocb
->private = ioend
= xfs_alloc_ioend(inode
, XFS_IO_DIRECT
);
1417 if (offset
+ size
> XFS_I(inode
)->i_d
.di_size
) {
1418 ret
= xfs_setfilesize_trans_alloc(ioend
);
1420 goto out_destroy_ioend
;
1421 ioend
->io_isdirect
= 1;
1424 ret
= __blockdev_direct_IO(rw
, iocb
, inode
, bdev
, iov
,
1426 xfs_get_blocks_direct
,
1427 xfs_end_io_direct_write
, NULL
, 0);
1428 if (ret
!= -EIOCBQUEUED
&& iocb
->private)
1429 goto out_trans_cancel
;
1431 ret
= __blockdev_direct_IO(rw
, iocb
, inode
, bdev
, iov
,
1433 xfs_get_blocks_direct
,
1440 if (ioend
->io_append_trans
) {
1441 current_set_flags_nested(&ioend
->io_append_trans
->t_pflags
,
1444 &inode
->i_sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
1446 xfs_trans_cancel(ioend
->io_append_trans
, 0);
1449 xfs_destroy_ioend(ioend
);
1454 * Punch out the delalloc blocks we have already allocated.
1456 * Don't bother with xfs_setattr given that nothing can have made it to disk yet
1457 * as the page is still locked at this point.
1460 xfs_vm_kill_delalloc_range(
1461 struct inode
*inode
,
1465 struct xfs_inode
*ip
= XFS_I(inode
);
1466 xfs_fileoff_t start_fsb
;
1467 xfs_fileoff_t end_fsb
;
1470 start_fsb
= XFS_B_TO_FSB(ip
->i_mount
, start
);
1471 end_fsb
= XFS_B_TO_FSB(ip
->i_mount
, end
);
1472 if (end_fsb
<= start_fsb
)
1475 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1476 error
= xfs_bmap_punch_delalloc_range(ip
, start_fsb
,
1477 end_fsb
- start_fsb
);
1479 /* something screwed, just bail */
1480 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
1481 xfs_alert(ip
->i_mount
,
1482 "xfs_vm_write_failed: unable to clean up ino %lld",
1486 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1490 xfs_vm_write_failed(
1491 struct inode
*inode
,
1496 loff_t block_offset
= pos
& PAGE_MASK
;
1499 loff_t from
= pos
& (PAGE_CACHE_SIZE
- 1);
1500 loff_t to
= from
+ len
;
1501 struct buffer_head
*bh
, *head
;
1503 ASSERT(block_offset
+ from
== pos
);
1505 head
= page_buffers(page
);
1507 for (bh
= head
; bh
!= head
|| !block_start
;
1508 bh
= bh
->b_this_page
, block_start
= block_end
,
1509 block_offset
+= bh
->b_size
) {
1510 block_end
= block_start
+ bh
->b_size
;
1512 /* skip buffers before the write */
1513 if (block_end
<= from
)
1516 /* if the buffer is after the write, we're done */
1517 if (block_start
>= to
)
1520 if (!buffer_delay(bh
))
1523 if (!buffer_new(bh
) && block_offset
< i_size_read(inode
))
1526 xfs_vm_kill_delalloc_range(inode
, block_offset
,
1527 block_offset
+ bh
->b_size
);
1533 * This used to call block_write_begin(), but it unlocks and releases the page
1534 * on error, and we need that page to be able to punch stale delalloc blocks out
1535 * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
1536 * the appropriate point.
1541 struct address_space
*mapping
,
1545 struct page
**pagep
,
1548 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1552 ASSERT(len
<= PAGE_CACHE_SIZE
);
1554 page
= grab_cache_page_write_begin(mapping
, index
,
1555 flags
| AOP_FLAG_NOFS
);
1559 status
= __block_write_begin(page
, pos
, len
, xfs_get_blocks
);
1560 if (unlikely(status
)) {
1561 struct inode
*inode
= mapping
->host
;
1563 xfs_vm_write_failed(inode
, page
, pos
, len
);
1566 if (pos
+ len
> i_size_read(inode
))
1567 truncate_pagecache(inode
, pos
+ len
, i_size_read(inode
));
1569 page_cache_release(page
);
1578 * On failure, we only need to kill delalloc blocks beyond EOF because they
1579 * will never be written. For blocks within EOF, generic_write_end() zeros them
1580 * so they are safe to leave alone and be written with all the other valid data.
1585 struct address_space
*mapping
,
1594 ASSERT(len
<= PAGE_CACHE_SIZE
);
1596 ret
= generic_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1597 if (unlikely(ret
< len
)) {
1598 struct inode
*inode
= mapping
->host
;
1599 size_t isize
= i_size_read(inode
);
1600 loff_t to
= pos
+ len
;
1603 truncate_pagecache(inode
, to
, isize
);
1604 xfs_vm_kill_delalloc_range(inode
, isize
, to
);
1612 struct address_space
*mapping
,
1615 struct inode
*inode
= (struct inode
*)mapping
->host
;
1616 struct xfs_inode
*ip
= XFS_I(inode
);
1618 trace_xfs_vm_bmap(XFS_I(inode
));
1619 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
1620 xfs_flush_pages(ip
, (xfs_off_t
)0, -1, 0, FI_REMAPF
);
1621 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
1622 return generic_block_bmap(mapping
, block
, xfs_get_blocks
);
1627 struct file
*unused
,
1630 return mpage_readpage(page
, xfs_get_blocks
);
1635 struct file
*unused
,
1636 struct address_space
*mapping
,
1637 struct list_head
*pages
,
1640 return mpage_readpages(mapping
, pages
, nr_pages
, xfs_get_blocks
);
1643 const struct address_space_operations xfs_address_space_operations
= {
1644 .readpage
= xfs_vm_readpage
,
1645 .readpages
= xfs_vm_readpages
,
1646 .writepage
= xfs_vm_writepage
,
1647 .writepages
= xfs_vm_writepages
,
1648 .releasepage
= xfs_vm_releasepage
,
1649 .invalidatepage
= xfs_vm_invalidatepage
,
1650 .write_begin
= xfs_vm_write_begin
,
1651 .write_end
= xfs_vm_write_end
,
1652 .bmap
= xfs_vm_bmap
,
1653 .direct_IO
= xfs_vm_direct_IO
,
1654 .migratepage
= buffer_migrate_page
,
1655 .is_partially_uptodate
= block_is_partially_uptodate
,
1656 .error_remove_page
= generic_error_remove_page
,