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 "xfs_bmap_util.h"
35 #include <linux/aio.h>
36 #include <linux/gfp.h>
37 #include <linux/mpage.h>
38 #include <linux/pagevec.h>
39 #include <linux/writeback.h>
47 struct buffer_head
*bh
, *head
;
49 *delalloc
= *unwritten
= 0;
51 bh
= head
= page_buffers(page
);
53 if (buffer_unwritten(bh
))
55 else if (buffer_delay(bh
))
57 } while ((bh
= bh
->b_this_page
) != head
);
60 STATIC
struct block_device
*
61 xfs_find_bdev_for_inode(
64 struct xfs_inode
*ip
= XFS_I(inode
);
65 struct xfs_mount
*mp
= ip
->i_mount
;
67 if (XFS_IS_REALTIME_INODE(ip
))
68 return mp
->m_rtdev_targp
->bt_bdev
;
70 return mp
->m_ddev_targp
->bt_bdev
;
74 * We're now finished for good with this ioend structure.
75 * Update the page state via the associated buffer_heads,
76 * release holds on the inode and bio, and finally free
77 * up memory. Do not use the ioend after this.
83 struct buffer_head
*bh
, *next
;
85 for (bh
= ioend
->io_buffer_head
; bh
; bh
= next
) {
87 bh
->b_end_io(bh
, !ioend
->io_error
);
91 inode_dio_done(ioend
->io_inode
);
92 if (ioend
->io_isasync
) {
93 aio_complete(ioend
->io_iocb
, ioend
->io_error
?
94 ioend
->io_error
: ioend
->io_result
, 0);
98 mempool_free(ioend
, xfs_ioend_pool
);
102 * Fast and loose check if this write could update the on-disk inode size.
104 static inline bool xfs_ioend_is_append(struct xfs_ioend
*ioend
)
106 return ioend
->io_offset
+ ioend
->io_size
>
107 XFS_I(ioend
->io_inode
)->i_d
.di_size
;
111 xfs_setfilesize_trans_alloc(
112 struct xfs_ioend
*ioend
)
114 struct xfs_mount
*mp
= XFS_I(ioend
->io_inode
)->i_mount
;
115 struct xfs_trans
*tp
;
118 tp
= xfs_trans_alloc(mp
, XFS_TRANS_FSYNC_TS
);
120 error
= xfs_trans_reserve(tp
, 0, XFS_FSYNC_TS_LOG_RES(mp
), 0, 0, 0);
122 xfs_trans_cancel(tp
, 0);
126 ioend
->io_append_trans
= tp
;
129 * We may pass freeze protection with a transaction. So tell lockdep
132 rwsem_release(&ioend
->io_inode
->i_sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
135 * We hand off the transaction to the completion thread now, so
136 * clear the flag here.
138 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
143 * Update on-disk file size now that data has been written to disk.
147 struct xfs_ioend
*ioend
)
149 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
150 struct xfs_trans
*tp
= ioend
->io_append_trans
;
154 * The transaction may have been allocated in the I/O submission thread,
155 * thus we need to mark ourselves as beeing in a transaction manually.
156 * Similarly for freeze protection.
158 current_set_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
159 rwsem_acquire_read(&VFS_I(ip
)->i_sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
162 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
163 isize
= xfs_new_eof(ip
, ioend
->io_offset
+ ioend
->io_size
);
165 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
166 xfs_trans_cancel(tp
, 0);
170 trace_xfs_setfilesize(ip
, ioend
->io_offset
, ioend
->io_size
);
172 ip
->i_d
.di_size
= isize
;
173 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
174 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
176 return xfs_trans_commit(tp
, 0);
180 * Schedule IO completion handling on the final put of an ioend.
182 * If there is no work to do we might as well call it a day and free the
187 struct xfs_ioend
*ioend
)
189 if (atomic_dec_and_test(&ioend
->io_remaining
)) {
190 struct xfs_mount
*mp
= XFS_I(ioend
->io_inode
)->i_mount
;
192 if (ioend
->io_type
== XFS_IO_UNWRITTEN
)
193 queue_work(mp
->m_unwritten_workqueue
, &ioend
->io_work
);
194 else if (ioend
->io_append_trans
||
195 (ioend
->io_isdirect
&& xfs_ioend_is_append(ioend
)))
196 queue_work(mp
->m_data_workqueue
, &ioend
->io_work
);
198 xfs_destroy_ioend(ioend
);
203 * IO write completion.
207 struct work_struct
*work
)
209 xfs_ioend_t
*ioend
= container_of(work
, xfs_ioend_t
, io_work
);
210 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
213 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
214 ioend
->io_error
= -EIO
;
221 * For unwritten extents we need to issue transactions to convert a
222 * range to normal written extens after the data I/O has finished.
224 if (ioend
->io_type
== XFS_IO_UNWRITTEN
) {
225 error
= xfs_iomap_write_unwritten(ip
, ioend
->io_offset
,
227 } else if (ioend
->io_isdirect
&& xfs_ioend_is_append(ioend
)) {
229 * For direct I/O we do not know if we need to allocate blocks
230 * or not so we can't preallocate an append transaction as that
231 * results in nested reservations and log space deadlocks. Hence
232 * allocate the transaction here. While this is sub-optimal and
233 * can block IO completion for some time, we're stuck with doing
234 * it this way until we can pass the ioend to the direct IO
235 * allocation callbacks and avoid nesting that way.
237 error
= xfs_setfilesize_trans_alloc(ioend
);
240 error
= xfs_setfilesize(ioend
);
241 } else if (ioend
->io_append_trans
) {
242 error
= xfs_setfilesize(ioend
);
244 ASSERT(!xfs_ioend_is_append(ioend
));
249 ioend
->io_error
= -error
;
250 xfs_destroy_ioend(ioend
);
254 * Call IO completion handling in caller context on the final put of an ioend.
257 xfs_finish_ioend_sync(
258 struct xfs_ioend
*ioend
)
260 if (atomic_dec_and_test(&ioend
->io_remaining
))
261 xfs_end_io(&ioend
->io_work
);
265 * Allocate and initialise an IO completion structure.
266 * We need to track unwritten extent write completion here initially.
267 * We'll need to extend this for updating the ondisk inode size later
277 ioend
= mempool_alloc(xfs_ioend_pool
, GFP_NOFS
);
280 * Set the count to 1 initially, which will prevent an I/O
281 * completion callback from happening before we have started
282 * all the I/O from calling the completion routine too early.
284 atomic_set(&ioend
->io_remaining
, 1);
285 ioend
->io_isasync
= 0;
286 ioend
->io_isdirect
= 0;
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 ioend
->io_offset
= 0;
295 ioend
->io_iocb
= NULL
;
296 ioend
->io_result
= 0;
297 ioend
->io_append_trans
= NULL
;
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
== XFS_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_super
->s_maxbytes
);
335 if (offset
+ count
> mp
->m_super
->s_maxbytes
)
336 count
= mp
->m_super
->s_maxbytes
- 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_read(ip
, offset_fsb
, end_fsb
- offset_fsb
,
340 imap
, &nimaps
, bmapi_flags
);
341 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
344 return -XFS_ERROR(error
);
346 if (type
== XFS_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
== XFS_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
;
408 submit_bio(wbc
->sync_mode
== WB_SYNC_ALL
? WRITE_SYNC
: WRITE
, bio
);
413 struct buffer_head
*bh
)
415 int nvecs
= bio_get_nr_vecs(bh
->b_bdev
);
416 struct bio
*bio
= bio_alloc(GFP_NOIO
, nvecs
);
418 ASSERT(bio
->bi_private
== NULL
);
419 bio
->bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
420 bio
->bi_bdev
= bh
->b_bdev
;
425 xfs_start_buffer_writeback(
426 struct buffer_head
*bh
)
428 ASSERT(buffer_mapped(bh
));
429 ASSERT(buffer_locked(bh
));
430 ASSERT(!buffer_delay(bh
));
431 ASSERT(!buffer_unwritten(bh
));
433 mark_buffer_async_write(bh
);
434 set_buffer_uptodate(bh
);
435 clear_buffer_dirty(bh
);
439 xfs_start_page_writeback(
444 ASSERT(PageLocked(page
));
445 ASSERT(!PageWriteback(page
));
447 clear_page_dirty_for_io(page
);
448 set_page_writeback(page
);
450 /* If no buffers on the page are to be written, finish it here */
452 end_page_writeback(page
);
455 static inline int bio_add_buffer(struct bio
*bio
, struct buffer_head
*bh
)
457 return bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
461 * Submit all of the bios for all of the ioends we have saved up, covering the
462 * initial writepage page and also any probed pages.
464 * Because we may have multiple ioends spanning a page, we need to start
465 * writeback on all the buffers before we submit them for I/O. If we mark the
466 * buffers as we got, then we can end up with a page that only has buffers
467 * marked async write and I/O complete on can occur before we mark the other
468 * buffers async write.
470 * The end result of this is that we trip a bug in end_page_writeback() because
471 * we call it twice for the one page as the code in end_buffer_async_write()
472 * assumes that all buffers on the page are started at the same time.
474 * The fix is two passes across the ioend list - one to start writeback on the
475 * buffer_heads, and then submit them for I/O on the second pass.
477 * If @fail is non-zero, it means that we have a situation where some part of
478 * the submission process has failed after we have marked paged for writeback
479 * and unlocked them. In this situation, we need to fail the ioend chain rather
480 * than submit it to IO. This typically only happens on a filesystem shutdown.
484 struct writeback_control
*wbc
,
488 xfs_ioend_t
*head
= ioend
;
490 struct buffer_head
*bh
;
492 sector_t lastblock
= 0;
494 /* Pass 1 - start writeback */
496 next
= ioend
->io_list
;
497 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
)
498 xfs_start_buffer_writeback(bh
);
499 } while ((ioend
= next
) != NULL
);
501 /* Pass 2 - submit I/O */
504 next
= ioend
->io_list
;
508 * If we are failing the IO now, just mark the ioend with an
509 * error and finish it. This will run IO completion immediately
510 * as there is only one reference to the ioend at this point in
514 ioend
->io_error
= -fail
;
515 xfs_finish_ioend(ioend
);
519 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
523 bio
= xfs_alloc_ioend_bio(bh
);
524 } else if (bh
->b_blocknr
!= lastblock
+ 1) {
525 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
529 if (bio_add_buffer(bio
, bh
) != bh
->b_size
) {
530 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
534 lastblock
= bh
->b_blocknr
;
537 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
538 xfs_finish_ioend(ioend
);
539 } while ((ioend
= next
) != NULL
);
543 * Cancel submission of all buffer_heads so far in this endio.
544 * Toss the endio too. Only ever called for the initial page
545 * in a writepage request, so only ever one page.
552 struct buffer_head
*bh
, *next_bh
;
555 next
= ioend
->io_list
;
556 bh
= ioend
->io_buffer_head
;
558 next_bh
= bh
->b_private
;
559 clear_buffer_async_write(bh
);
561 } while ((bh
= next_bh
) != NULL
);
563 mempool_free(ioend
, xfs_ioend_pool
);
564 } while ((ioend
= next
) != NULL
);
568 * Test to see if we've been building up a completion structure for
569 * earlier buffers -- if so, we try to append to this ioend if we
570 * can, otherwise we finish off any current ioend and start another.
571 * Return true if we've finished the given ioend.
576 struct buffer_head
*bh
,
579 xfs_ioend_t
**result
,
582 xfs_ioend_t
*ioend
= *result
;
584 if (!ioend
|| need_ioend
|| type
!= ioend
->io_type
) {
585 xfs_ioend_t
*previous
= *result
;
587 ioend
= xfs_alloc_ioend(inode
, type
);
588 ioend
->io_offset
= offset
;
589 ioend
->io_buffer_head
= bh
;
590 ioend
->io_buffer_tail
= bh
;
592 previous
->io_list
= ioend
;
595 ioend
->io_buffer_tail
->b_private
= bh
;
596 ioend
->io_buffer_tail
= bh
;
599 bh
->b_private
= NULL
;
600 ioend
->io_size
+= bh
->b_size
;
606 struct buffer_head
*bh
,
607 struct xfs_bmbt_irec
*imap
,
611 struct xfs_mount
*m
= XFS_I(inode
)->i_mount
;
612 xfs_off_t iomap_offset
= XFS_FSB_TO_B(m
, imap
->br_startoff
);
613 xfs_daddr_t iomap_bn
= xfs_fsb_to_db(XFS_I(inode
), imap
->br_startblock
);
615 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
616 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
618 bn
= (iomap_bn
>> (inode
->i_blkbits
- BBSHIFT
)) +
619 ((offset
- iomap_offset
) >> inode
->i_blkbits
);
621 ASSERT(bn
|| XFS_IS_REALTIME_INODE(XFS_I(inode
)));
624 set_buffer_mapped(bh
);
630 struct buffer_head
*bh
,
631 struct xfs_bmbt_irec
*imap
,
634 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
635 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
637 xfs_map_buffer(inode
, bh
, imap
, offset
);
638 set_buffer_mapped(bh
);
639 clear_buffer_delay(bh
);
640 clear_buffer_unwritten(bh
);
644 * Test if a given page is suitable for writing as part of an unwritten
645 * or delayed allocate extent.
652 if (PageWriteback(page
))
655 if (page
->mapping
&& page_has_buffers(page
)) {
656 struct buffer_head
*bh
, *head
;
659 bh
= head
= page_buffers(page
);
661 if (buffer_unwritten(bh
))
662 acceptable
+= (type
== XFS_IO_UNWRITTEN
);
663 else if (buffer_delay(bh
))
664 acceptable
+= (type
== XFS_IO_DELALLOC
);
665 else if (buffer_dirty(bh
) && buffer_mapped(bh
))
666 acceptable
+= (type
== XFS_IO_OVERWRITE
);
669 } while ((bh
= bh
->b_this_page
) != head
);
679 * Allocate & map buffers for page given the extent map. Write it out.
680 * except for the original page of a writepage, this is called on
681 * delalloc/unwritten pages only, for the original page it is possible
682 * that the page has no mapping at all.
689 struct xfs_bmbt_irec
*imap
,
690 xfs_ioend_t
**ioendp
,
691 struct writeback_control
*wbc
)
693 struct buffer_head
*bh
, *head
;
694 xfs_off_t end_offset
;
695 unsigned long p_offset
;
698 int count
= 0, done
= 0, uptodate
= 1;
699 xfs_off_t offset
= page_offset(page
);
701 if (page
->index
!= tindex
)
703 if (!trylock_page(page
))
705 if (PageWriteback(page
))
706 goto fail_unlock_page
;
707 if (page
->mapping
!= inode
->i_mapping
)
708 goto fail_unlock_page
;
709 if (!xfs_check_page_type(page
, (*ioendp
)->io_type
))
710 goto fail_unlock_page
;
713 * page_dirty is initially a count of buffers on the page before
714 * EOF and is decremented as we move each into a cleanable state.
718 * End offset is the highest offset that this page should represent.
719 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
720 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
721 * hence give us the correct page_dirty count. On any other page,
722 * it will be zero and in that case we need page_dirty to be the
723 * count of buffers on the page.
725 end_offset
= min_t(unsigned long long,
726 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
730 * If the current map does not span the entire page we are about to try
731 * to write, then give up. The only way we can write a page that spans
732 * multiple mappings in a single writeback iteration is via the
733 * xfs_vm_writepage() function. Data integrity writeback requires the
734 * entire page to be written in a single attempt, otherwise the part of
735 * the page we don't write here doesn't get written as part of the data
738 * For normal writeback, we also don't attempt to write partial pages
739 * here as it simply means that write_cache_pages() will see it under
740 * writeback and ignore the page until some point in the future, at
741 * which time this will be the only page in the file that needs
742 * writeback. Hence for more optimal IO patterns, we should always
743 * avoid partial page writeback due to multiple mappings on a page here.
745 if (!xfs_imap_valid(inode
, imap
, end_offset
))
746 goto fail_unlock_page
;
748 len
= 1 << inode
->i_blkbits
;
749 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
751 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
752 page_dirty
= p_offset
/ len
;
754 bh
= head
= page_buffers(page
);
756 if (offset
>= end_offset
)
758 if (!buffer_uptodate(bh
))
760 if (!(PageUptodate(page
) || buffer_uptodate(bh
))) {
765 if (buffer_unwritten(bh
) || buffer_delay(bh
) ||
767 if (buffer_unwritten(bh
))
768 type
= XFS_IO_UNWRITTEN
;
769 else if (buffer_delay(bh
))
770 type
= XFS_IO_DELALLOC
;
772 type
= XFS_IO_OVERWRITE
;
774 if (!xfs_imap_valid(inode
, imap
, offset
)) {
780 if (type
!= XFS_IO_OVERWRITE
)
781 xfs_map_at_offset(inode
, bh
, imap
, offset
);
782 xfs_add_to_ioend(inode
, bh
, offset
, type
,
790 } while (offset
+= len
, (bh
= bh
->b_this_page
) != head
);
792 if (uptodate
&& bh
== head
)
793 SetPageUptodate(page
);
796 if (--wbc
->nr_to_write
<= 0 &&
797 wbc
->sync_mode
== WB_SYNC_NONE
)
800 xfs_start_page_writeback(page
, !page_dirty
, count
);
810 * Convert & write out a cluster of pages in the same extent as defined
811 * by mp and following the start page.
817 struct xfs_bmbt_irec
*imap
,
818 xfs_ioend_t
**ioendp
,
819 struct writeback_control
*wbc
,
825 pagevec_init(&pvec
, 0);
826 while (!done
&& tindex
<= tlast
) {
827 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
829 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
832 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
833 done
= xfs_convert_page(inode
, pvec
.pages
[i
], tindex
++,
839 pagevec_release(&pvec
);
845 xfs_vm_invalidatepage(
850 trace_xfs_invalidatepage(page
->mapping
->host
, page
, offset
,
852 block_invalidatepage(page
, offset
, length
);
856 * If the page has delalloc buffers on it, we need to punch them out before we
857 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
858 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
859 * is done on that same region - the delalloc extent is returned when none is
860 * supposed to be there.
862 * We prevent this by truncating away the delalloc regions on the page before
863 * invalidating it. Because they are delalloc, we can do this without needing a
864 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
865 * truncation without a transaction as there is no space left for block
866 * reservation (typically why we see a ENOSPC in writeback).
868 * This is not a performance critical path, so for now just do the punching a
869 * buffer head at a time.
872 xfs_aops_discard_page(
875 struct inode
*inode
= page
->mapping
->host
;
876 struct xfs_inode
*ip
= XFS_I(inode
);
877 struct buffer_head
*bh
, *head
;
878 loff_t offset
= page_offset(page
);
880 if (!xfs_check_page_type(page
, XFS_IO_DELALLOC
))
883 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
886 xfs_alert(ip
->i_mount
,
887 "page discard on page %p, inode 0x%llx, offset %llu.",
888 page
, ip
->i_ino
, offset
);
890 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
891 bh
= head
= page_buffers(page
);
894 xfs_fileoff_t start_fsb
;
896 if (!buffer_delay(bh
))
899 start_fsb
= XFS_B_TO_FSBT(ip
->i_mount
, offset
);
900 error
= xfs_bmap_punch_delalloc_range(ip
, start_fsb
, 1);
902 /* something screwed, just bail */
903 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
904 xfs_alert(ip
->i_mount
,
905 "page discard unable to remove delalloc mapping.");
910 offset
+= 1 << inode
->i_blkbits
;
912 } while ((bh
= bh
->b_this_page
) != head
);
914 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
916 xfs_vm_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
921 * Write out a dirty page.
923 * For delalloc space on the page we need to allocate space and flush it.
924 * For unwritten space on the page we need to start the conversion to
925 * regular allocated space.
926 * For any other dirty buffer heads on the page we should flush them.
931 struct writeback_control
*wbc
)
933 struct inode
*inode
= page
->mapping
->host
;
934 struct buffer_head
*bh
, *head
;
935 struct xfs_bmbt_irec imap
;
936 xfs_ioend_t
*ioend
= NULL
, *iohead
= NULL
;
939 __uint64_t end_offset
;
940 pgoff_t end_index
, last_index
;
942 int err
, imap_valid
= 0, uptodate
= 1;
946 trace_xfs_writepage(inode
, page
, 0, 0);
948 ASSERT(page_has_buffers(page
));
951 * Refuse to write the page out if we are called from reclaim context.
953 * This avoids stack overflows when called from deeply used stacks in
954 * random callers for direct reclaim or memcg reclaim. We explicitly
955 * allow reclaim from kswapd as the stack usage there is relatively low.
957 * This should never happen except in the case of a VM regression so
960 if (WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
)) ==
965 * Given that we do not allow direct reclaim to call us, we should
966 * never be called while in a filesystem transaction.
968 if (WARN_ON(current
->flags
& PF_FSTRANS
))
971 /* Is this page beyond the end of the file? */
972 offset
= i_size_read(inode
);
973 end_index
= offset
>> PAGE_CACHE_SHIFT
;
974 last_index
= (offset
- 1) >> PAGE_CACHE_SHIFT
;
975 if (page
->index
>= end_index
) {
976 unsigned offset_into_page
= offset
& (PAGE_CACHE_SIZE
- 1);
979 * Skip the page if it is fully outside i_size, e.g. due to a
980 * truncate operation that is in progress. We must redirty the
981 * page so that reclaim stops reclaiming it. Otherwise
982 * xfs_vm_releasepage() is called on it and gets confused.
984 if (page
->index
>= end_index
+ 1 || offset_into_page
== 0)
988 * The page straddles i_size. It must be zeroed out on each
989 * and every writepage invocation because it may be mmapped.
990 * "A file is mapped in multiples of the page size. For a file
991 * that is not a multiple of the page size, the remaining
992 * memory is zeroed when mapped, and writes to that region are
993 * not written out to the file."
995 zero_user_segment(page
, offset_into_page
, PAGE_CACHE_SIZE
);
998 end_offset
= min_t(unsigned long long,
999 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
1001 len
= 1 << inode
->i_blkbits
;
1003 bh
= head
= page_buffers(page
);
1004 offset
= page_offset(page
);
1005 type
= XFS_IO_OVERWRITE
;
1007 if (wbc
->sync_mode
== WB_SYNC_NONE
)
1013 if (offset
>= end_offset
)
1015 if (!buffer_uptodate(bh
))
1019 * set_page_dirty dirties all buffers in a page, independent
1020 * of their state. The dirty state however is entirely
1021 * meaningless for holes (!mapped && uptodate), so skip
1022 * buffers covering holes here.
1024 if (!buffer_mapped(bh
) && buffer_uptodate(bh
)) {
1029 if (buffer_unwritten(bh
)) {
1030 if (type
!= XFS_IO_UNWRITTEN
) {
1031 type
= XFS_IO_UNWRITTEN
;
1034 } else if (buffer_delay(bh
)) {
1035 if (type
!= XFS_IO_DELALLOC
) {
1036 type
= XFS_IO_DELALLOC
;
1039 } else if (buffer_uptodate(bh
)) {
1040 if (type
!= XFS_IO_OVERWRITE
) {
1041 type
= XFS_IO_OVERWRITE
;
1045 if (PageUptodate(page
))
1046 ASSERT(buffer_mapped(bh
));
1048 * This buffer is not uptodate and will not be
1049 * written to disk. Ensure that we will put any
1050 * subsequent writeable buffers into a new
1058 imap_valid
= xfs_imap_valid(inode
, &imap
, offset
);
1061 * If we didn't have a valid mapping then we need to
1062 * put the new mapping into a separate ioend structure.
1063 * This ensures non-contiguous extents always have
1064 * separate ioends, which is particularly important
1065 * for unwritten extent conversion at I/O completion
1069 err
= xfs_map_blocks(inode
, offset
, &imap
, type
,
1073 imap_valid
= xfs_imap_valid(inode
, &imap
, offset
);
1077 if (type
!= XFS_IO_OVERWRITE
)
1078 xfs_map_at_offset(inode
, bh
, &imap
, offset
);
1079 xfs_add_to_ioend(inode
, bh
, offset
, type
, &ioend
,
1087 } while (offset
+= len
, ((bh
= bh
->b_this_page
) != head
));
1089 if (uptodate
&& bh
== head
)
1090 SetPageUptodate(page
);
1092 xfs_start_page_writeback(page
, 1, count
);
1094 /* if there is no IO to be submitted for this page, we are done */
1101 * Any errors from this point onwards need tobe reported through the IO
1102 * completion path as we have marked the initial page as under writeback
1106 xfs_off_t end_index
;
1108 end_index
= imap
.br_startoff
+ imap
.br_blockcount
;
1111 end_index
<<= inode
->i_blkbits
;
1114 end_index
= (end_index
- 1) >> PAGE_CACHE_SHIFT
;
1116 /* check against file size */
1117 if (end_index
> last_index
)
1118 end_index
= last_index
;
1120 xfs_cluster_write(inode
, page
->index
+ 1, &imap
, &ioend
,
1126 * Reserve log space if we might write beyond the on-disk inode size.
1129 if (ioend
->io_type
!= XFS_IO_UNWRITTEN
&& xfs_ioend_is_append(ioend
))
1130 err
= xfs_setfilesize_trans_alloc(ioend
);
1132 xfs_submit_ioend(wbc
, iohead
, err
);
1138 xfs_cancel_ioend(iohead
);
1143 xfs_aops_discard_page(page
);
1144 ClearPageUptodate(page
);
1149 redirty_page_for_writepage(wbc
, page
);
1156 struct address_space
*mapping
,
1157 struct writeback_control
*wbc
)
1159 xfs_iflags_clear(XFS_I(mapping
->host
), XFS_ITRUNCATED
);
1160 return generic_writepages(mapping
, wbc
);
1164 * Called to move a page into cleanable state - and from there
1165 * to be released. The page should already be clean. We always
1166 * have buffer heads in this call.
1168 * Returns 1 if the page is ok to release, 0 otherwise.
1175 int delalloc
, unwritten
;
1177 trace_xfs_releasepage(page
->mapping
->host
, page
, 0, 0);
1179 xfs_count_page_state(page
, &delalloc
, &unwritten
);
1181 if (WARN_ON(delalloc
))
1183 if (WARN_ON(unwritten
))
1186 return try_to_free_buffers(page
);
1191 struct inode
*inode
,
1193 struct buffer_head
*bh_result
,
1197 struct xfs_inode
*ip
= XFS_I(inode
);
1198 struct xfs_mount
*mp
= ip
->i_mount
;
1199 xfs_fileoff_t offset_fsb
, end_fsb
;
1202 struct xfs_bmbt_irec imap
;
1208 if (XFS_FORCED_SHUTDOWN(mp
))
1209 return -XFS_ERROR(EIO
);
1211 offset
= (xfs_off_t
)iblock
<< inode
->i_blkbits
;
1212 ASSERT(bh_result
->b_size
>= (1 << inode
->i_blkbits
));
1213 size
= bh_result
->b_size
;
1215 if (!create
&& direct
&& offset
>= i_size_read(inode
))
1219 * Direct I/O is usually done on preallocated files, so try getting
1220 * a block mapping without an exclusive lock first. For buffered
1221 * writes we already have the exclusive iolock anyway, so avoiding
1222 * a lock roundtrip here by taking the ilock exclusive from the
1223 * beginning is a useful micro optimization.
1225 if (create
&& !direct
) {
1226 lockmode
= XFS_ILOCK_EXCL
;
1227 xfs_ilock(ip
, lockmode
);
1229 lockmode
= xfs_ilock_map_shared(ip
);
1232 ASSERT(offset
<= mp
->m_super
->s_maxbytes
);
1233 if (offset
+ size
> mp
->m_super
->s_maxbytes
)
1234 size
= mp
->m_super
->s_maxbytes
- offset
;
1235 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ size
);
1236 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1238 error
= xfs_bmapi_read(ip
, offset_fsb
, end_fsb
- offset_fsb
,
1239 &imap
, &nimaps
, XFS_BMAPI_ENTIRE
);
1245 (imap
.br_startblock
== HOLESTARTBLOCK
||
1246 imap
.br_startblock
== DELAYSTARTBLOCK
))) {
1247 if (direct
|| xfs_get_extsz_hint(ip
)) {
1249 * Drop the ilock in preparation for starting the block
1250 * allocation transaction. It will be retaken
1251 * exclusively inside xfs_iomap_write_direct for the
1252 * actual allocation.
1254 xfs_iunlock(ip
, lockmode
);
1255 error
= xfs_iomap_write_direct(ip
, offset
, size
,
1262 * Delalloc reservations do not require a transaction,
1263 * we can go on without dropping the lock here. If we
1264 * are allocating a new delalloc block, make sure that
1265 * we set the new flag so that we mark the buffer new so
1266 * that we know that it is newly allocated if the write
1269 if (nimaps
&& imap
.br_startblock
== HOLESTARTBLOCK
)
1271 error
= xfs_iomap_write_delay(ip
, offset
, size
, &imap
);
1275 xfs_iunlock(ip
, lockmode
);
1278 trace_xfs_get_blocks_alloc(ip
, offset
, size
, 0, &imap
);
1279 } else if (nimaps
) {
1280 trace_xfs_get_blocks_found(ip
, offset
, size
, 0, &imap
);
1281 xfs_iunlock(ip
, lockmode
);
1283 trace_xfs_get_blocks_notfound(ip
, offset
, size
);
1287 if (imap
.br_startblock
!= HOLESTARTBLOCK
&&
1288 imap
.br_startblock
!= DELAYSTARTBLOCK
) {
1290 * For unwritten extents do not report a disk address on
1291 * the read case (treat as if we're reading into a hole).
1293 if (create
|| !ISUNWRITTEN(&imap
))
1294 xfs_map_buffer(inode
, bh_result
, &imap
, offset
);
1295 if (create
&& ISUNWRITTEN(&imap
)) {
1297 bh_result
->b_private
= inode
;
1298 set_buffer_unwritten(bh_result
);
1303 * If this is a realtime file, data may be on a different device.
1304 * to that pointed to from the buffer_head b_bdev currently.
1306 bh_result
->b_bdev
= xfs_find_bdev_for_inode(inode
);
1309 * If we previously allocated a block out beyond eof and we are now
1310 * coming back to use it then we will need to flag it as new even if it
1311 * has a disk address.
1313 * With sub-block writes into unwritten extents we also need to mark
1314 * the buffer as new so that the unwritten parts of the buffer gets
1318 ((!buffer_mapped(bh_result
) && !buffer_uptodate(bh_result
)) ||
1319 (offset
>= i_size_read(inode
)) ||
1320 (new || ISUNWRITTEN(&imap
))))
1321 set_buffer_new(bh_result
);
1323 if (imap
.br_startblock
== DELAYSTARTBLOCK
) {
1326 set_buffer_uptodate(bh_result
);
1327 set_buffer_mapped(bh_result
);
1328 set_buffer_delay(bh_result
);
1333 * If this is O_DIRECT or the mpage code calling tell them how large
1334 * the mapping is, so that we can avoid repeated get_blocks calls.
1336 if (direct
|| size
> (1 << inode
->i_blkbits
)) {
1337 xfs_off_t mapping_size
;
1339 mapping_size
= imap
.br_startoff
+ imap
.br_blockcount
- iblock
;
1340 mapping_size
<<= inode
->i_blkbits
;
1342 ASSERT(mapping_size
> 0);
1343 if (mapping_size
> size
)
1344 mapping_size
= size
;
1345 if (mapping_size
> LONG_MAX
)
1346 mapping_size
= LONG_MAX
;
1348 bh_result
->b_size
= mapping_size
;
1354 xfs_iunlock(ip
, lockmode
);
1360 struct inode
*inode
,
1362 struct buffer_head
*bh_result
,
1365 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 0);
1369 xfs_get_blocks_direct(
1370 struct inode
*inode
,
1372 struct buffer_head
*bh_result
,
1375 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 1);
1379 * Complete a direct I/O write request.
1381 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1382 * need to issue a transaction to convert the range from unwritten to written
1383 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1384 * to do this and we are done. But in case this was a successful AIO
1385 * request this handler is called from interrupt context, from which we
1386 * can't start transactions. In that case offload the I/O completion to
1387 * the workqueues we also use for buffered I/O completion.
1390 xfs_end_io_direct_write(
1398 struct xfs_ioend
*ioend
= iocb
->private;
1401 * While the generic direct I/O code updates the inode size, it does
1402 * so only after the end_io handler is called, which means our
1403 * end_io handler thinks the on-disk size is outside the in-core
1404 * size. To prevent this just update it a little bit earlier here.
1406 if (offset
+ size
> i_size_read(ioend
->io_inode
))
1407 i_size_write(ioend
->io_inode
, offset
+ size
);
1410 * blockdev_direct_IO can return an error even after the I/O
1411 * completion handler was called. Thus we need to protect
1412 * against double-freeing.
1414 iocb
->private = NULL
;
1416 ioend
->io_offset
= offset
;
1417 ioend
->io_size
= size
;
1418 ioend
->io_iocb
= iocb
;
1419 ioend
->io_result
= ret
;
1420 if (private && size
> 0)
1421 ioend
->io_type
= XFS_IO_UNWRITTEN
;
1424 ioend
->io_isasync
= 1;
1425 xfs_finish_ioend(ioend
);
1427 xfs_finish_ioend_sync(ioend
);
1435 const struct iovec
*iov
,
1437 unsigned long nr_segs
)
1439 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
1440 struct block_device
*bdev
= xfs_find_bdev_for_inode(inode
);
1441 struct xfs_ioend
*ioend
= NULL
;
1445 size_t size
= iov_length(iov
, nr_segs
);
1448 * We cannot preallocate a size update transaction here as we
1449 * don't know whether allocation is necessary or not. Hence we
1450 * can only tell IO completion that one is necessary if we are
1451 * not doing unwritten extent conversion.
1453 iocb
->private = ioend
= xfs_alloc_ioend(inode
, XFS_IO_DIRECT
);
1454 if (offset
+ size
> XFS_I(inode
)->i_d
.di_size
)
1455 ioend
->io_isdirect
= 1;
1457 ret
= __blockdev_direct_IO(rw
, iocb
, inode
, bdev
, iov
,
1459 xfs_get_blocks_direct
,
1460 xfs_end_io_direct_write
, NULL
, 0);
1461 if (ret
!= -EIOCBQUEUED
&& iocb
->private)
1462 goto out_destroy_ioend
;
1464 ret
= __blockdev_direct_IO(rw
, iocb
, inode
, bdev
, iov
,
1466 xfs_get_blocks_direct
,
1473 xfs_destroy_ioend(ioend
);
1478 * Punch out the delalloc blocks we have already allocated.
1480 * Don't bother with xfs_setattr given that nothing can have made it to disk yet
1481 * as the page is still locked at this point.
1484 xfs_vm_kill_delalloc_range(
1485 struct inode
*inode
,
1489 struct xfs_inode
*ip
= XFS_I(inode
);
1490 xfs_fileoff_t start_fsb
;
1491 xfs_fileoff_t end_fsb
;
1494 start_fsb
= XFS_B_TO_FSB(ip
->i_mount
, start
);
1495 end_fsb
= XFS_B_TO_FSB(ip
->i_mount
, end
);
1496 if (end_fsb
<= start_fsb
)
1499 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1500 error
= xfs_bmap_punch_delalloc_range(ip
, start_fsb
,
1501 end_fsb
- start_fsb
);
1503 /* something screwed, just bail */
1504 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
1505 xfs_alert(ip
->i_mount
,
1506 "xfs_vm_write_failed: unable to clean up ino %lld",
1510 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1514 xfs_vm_write_failed(
1515 struct inode
*inode
,
1520 loff_t block_offset
;
1523 loff_t from
= pos
& (PAGE_CACHE_SIZE
- 1);
1524 loff_t to
= from
+ len
;
1525 struct buffer_head
*bh
, *head
;
1528 * The request pos offset might be 32 or 64 bit, this is all fine
1529 * on 64-bit platform. However, for 64-bit pos request on 32-bit
1530 * platform, the high 32-bit will be masked off if we evaluate the
1531 * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is
1532 * 0xfffff000 as an unsigned long, hence the result is incorrect
1533 * which could cause the following ASSERT failed in most cases.
1534 * In order to avoid this, we can evaluate the block_offset of the
1535 * start of the page by using shifts rather than masks the mismatch
1538 block_offset
= (pos
>> PAGE_CACHE_SHIFT
) << PAGE_CACHE_SHIFT
;
1540 ASSERT(block_offset
+ from
== pos
);
1542 head
= page_buffers(page
);
1544 for (bh
= head
; bh
!= head
|| !block_start
;
1545 bh
= bh
->b_this_page
, block_start
= block_end
,
1546 block_offset
+= bh
->b_size
) {
1547 block_end
= block_start
+ bh
->b_size
;
1549 /* skip buffers before the write */
1550 if (block_end
<= from
)
1553 /* if the buffer is after the write, we're done */
1554 if (block_start
>= to
)
1557 if (!buffer_delay(bh
))
1560 if (!buffer_new(bh
) && block_offset
< i_size_read(inode
))
1563 xfs_vm_kill_delalloc_range(inode
, block_offset
,
1564 block_offset
+ bh
->b_size
);
1570 * This used to call block_write_begin(), but it unlocks and releases the page
1571 * on error, and we need that page to be able to punch stale delalloc blocks out
1572 * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
1573 * the appropriate point.
1578 struct address_space
*mapping
,
1582 struct page
**pagep
,
1585 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1589 ASSERT(len
<= PAGE_CACHE_SIZE
);
1591 page
= grab_cache_page_write_begin(mapping
, index
,
1592 flags
| AOP_FLAG_NOFS
);
1596 status
= __block_write_begin(page
, pos
, len
, xfs_get_blocks
);
1597 if (unlikely(status
)) {
1598 struct inode
*inode
= mapping
->host
;
1600 xfs_vm_write_failed(inode
, page
, pos
, len
);
1603 if (pos
+ len
> i_size_read(inode
))
1604 truncate_pagecache(inode
, pos
+ len
, i_size_read(inode
));
1606 page_cache_release(page
);
1615 * On failure, we only need to kill delalloc blocks beyond EOF because they
1616 * will never be written. For blocks within EOF, generic_write_end() zeros them
1617 * so they are safe to leave alone and be written with all the other valid data.
1622 struct address_space
*mapping
,
1631 ASSERT(len
<= PAGE_CACHE_SIZE
);
1633 ret
= generic_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1634 if (unlikely(ret
< len
)) {
1635 struct inode
*inode
= mapping
->host
;
1636 size_t isize
= i_size_read(inode
);
1637 loff_t to
= pos
+ len
;
1640 truncate_pagecache(inode
, to
, isize
);
1641 xfs_vm_kill_delalloc_range(inode
, isize
, to
);
1649 struct address_space
*mapping
,
1652 struct inode
*inode
= (struct inode
*)mapping
->host
;
1653 struct xfs_inode
*ip
= XFS_I(inode
);
1655 trace_xfs_vm_bmap(XFS_I(inode
));
1656 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
1657 filemap_write_and_wait(mapping
);
1658 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
1659 return generic_block_bmap(mapping
, block
, xfs_get_blocks
);
1664 struct file
*unused
,
1667 return mpage_readpage(page
, xfs_get_blocks
);
1672 struct file
*unused
,
1673 struct address_space
*mapping
,
1674 struct list_head
*pages
,
1677 return mpage_readpages(mapping
, pages
, nr_pages
, xfs_get_blocks
);
1680 const struct address_space_operations xfs_address_space_operations
= {
1681 .readpage
= xfs_vm_readpage
,
1682 .readpages
= xfs_vm_readpages
,
1683 .writepage
= xfs_vm_writepage
,
1684 .writepages
= xfs_vm_writepages
,
1685 .releasepage
= xfs_vm_releasepage
,
1686 .invalidatepage
= xfs_vm_invalidatepage
,
1687 .write_begin
= xfs_vm_write_begin
,
1688 .write_end
= xfs_vm_write_end
,
1689 .bmap
= xfs_vm_bmap
,
1690 .direct_IO
= xfs_vm_direct_IO
,
1691 .migratepage
= buffer_migrate_page
,
1692 .is_partially_uptodate
= block_is_partially_uptodate
,
1693 .error_remove_page
= generic_error_remove_page
,