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
19 #include "xfs_shared.h"
20 #include "xfs_format.h"
21 #include "xfs_log_format.h"
22 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_inode.h"
27 #include "xfs_trans.h"
28 #include "xfs_inode_item.h"
29 #include "xfs_alloc.h"
30 #include "xfs_error.h"
31 #include "xfs_iomap.h"
32 #include "xfs_trace.h"
34 #include "xfs_bmap_util.h"
35 #include "xfs_bmap_btree.h"
36 #include "xfs_dinode.h"
37 #include <linux/aio.h>
38 #include <linux/gfp.h>
39 #include <linux/mpage.h>
40 #include <linux/pagevec.h>
41 #include <linux/writeback.h>
49 struct buffer_head
*bh
, *head
;
51 *delalloc
= *unwritten
= 0;
53 bh
= head
= page_buffers(page
);
55 if (buffer_unwritten(bh
))
57 else if (buffer_delay(bh
))
59 } while ((bh
= bh
->b_this_page
) != head
);
62 STATIC
struct block_device
*
63 xfs_find_bdev_for_inode(
66 struct xfs_inode
*ip
= XFS_I(inode
);
67 struct xfs_mount
*mp
= ip
->i_mount
;
69 if (XFS_IS_REALTIME_INODE(ip
))
70 return mp
->m_rtdev_targp
->bt_bdev
;
72 return mp
->m_ddev_targp
->bt_bdev
;
76 * We're now finished for good with this ioend structure.
77 * Update the page state via the associated buffer_heads,
78 * release holds on the inode and bio, and finally free
79 * up memory. Do not use the ioend after this.
85 struct buffer_head
*bh
, *next
;
87 for (bh
= ioend
->io_buffer_head
; bh
; bh
= next
) {
89 bh
->b_end_io(bh
, !ioend
->io_error
);
92 mempool_free(ioend
, xfs_ioend_pool
);
96 * Fast and loose check if this write could update the on-disk inode size.
98 static inline bool xfs_ioend_is_append(struct xfs_ioend
*ioend
)
100 return ioend
->io_offset
+ ioend
->io_size
>
101 XFS_I(ioend
->io_inode
)->i_d
.di_size
;
105 xfs_setfilesize_trans_alloc(
106 struct xfs_ioend
*ioend
)
108 struct xfs_mount
*mp
= XFS_I(ioend
->io_inode
)->i_mount
;
109 struct xfs_trans
*tp
;
112 tp
= xfs_trans_alloc(mp
, XFS_TRANS_FSYNC_TS
);
114 error
= xfs_trans_reserve(tp
, &M_RES(mp
)->tr_fsyncts
, 0, 0);
116 xfs_trans_cancel(tp
, 0);
120 ioend
->io_append_trans
= tp
;
123 * We may pass freeze protection with a transaction. So tell lockdep
126 rwsem_release(&ioend
->io_inode
->i_sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
129 * We hand off the transaction to the completion thread now, so
130 * clear the flag here.
132 current_restore_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
137 * Update on-disk file size now that data has been written to disk.
141 struct xfs_ioend
*ioend
)
143 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
144 struct xfs_trans
*tp
= ioend
->io_append_trans
;
148 * The transaction may have been allocated in the I/O submission thread,
149 * thus we need to mark ourselves as beeing in a transaction manually.
150 * Similarly for freeze protection.
152 current_set_flags_nested(&tp
->t_pflags
, PF_FSTRANS
);
153 rwsem_acquire_read(&VFS_I(ip
)->i_sb
->s_writers
.lock_map
[SB_FREEZE_FS
-1],
156 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
157 isize
= xfs_new_eof(ip
, ioend
->io_offset
+ ioend
->io_size
);
159 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
160 xfs_trans_cancel(tp
, 0);
164 trace_xfs_setfilesize(ip
, ioend
->io_offset
, ioend
->io_size
);
166 ip
->i_d
.di_size
= isize
;
167 xfs_trans_ijoin(tp
, ip
, XFS_ILOCK_EXCL
);
168 xfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
170 return xfs_trans_commit(tp
, 0);
174 * Schedule IO completion handling on the final put of an ioend.
176 * If there is no work to do we might as well call it a day and free the
181 struct xfs_ioend
*ioend
)
183 if (atomic_dec_and_test(&ioend
->io_remaining
)) {
184 struct xfs_mount
*mp
= XFS_I(ioend
->io_inode
)->i_mount
;
186 if (ioend
->io_type
== XFS_IO_UNWRITTEN
)
187 queue_work(mp
->m_unwritten_workqueue
, &ioend
->io_work
);
188 else if (ioend
->io_append_trans
||
189 (ioend
->io_isdirect
&& xfs_ioend_is_append(ioend
)))
190 queue_work(mp
->m_data_workqueue
, &ioend
->io_work
);
192 xfs_destroy_ioend(ioend
);
197 * IO write completion.
201 struct work_struct
*work
)
203 xfs_ioend_t
*ioend
= container_of(work
, xfs_ioend_t
, io_work
);
204 struct xfs_inode
*ip
= XFS_I(ioend
->io_inode
);
207 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
208 ioend
->io_error
= -EIO
;
215 * For unwritten extents we need to issue transactions to convert a
216 * range to normal written extens after the data I/O has finished.
218 if (ioend
->io_type
== XFS_IO_UNWRITTEN
) {
219 error
= xfs_iomap_write_unwritten(ip
, ioend
->io_offset
,
221 } else if (ioend
->io_isdirect
&& xfs_ioend_is_append(ioend
)) {
223 * For direct I/O we do not know if we need to allocate blocks
224 * or not so we can't preallocate an append transaction as that
225 * results in nested reservations and log space deadlocks. Hence
226 * allocate the transaction here. While this is sub-optimal and
227 * can block IO completion for some time, we're stuck with doing
228 * it this way until we can pass the ioend to the direct IO
229 * allocation callbacks and avoid nesting that way.
231 error
= xfs_setfilesize_trans_alloc(ioend
);
234 error
= xfs_setfilesize(ioend
);
235 } else if (ioend
->io_append_trans
) {
236 error
= xfs_setfilesize(ioend
);
238 ASSERT(!xfs_ioend_is_append(ioend
));
243 ioend
->io_error
= error
;
244 xfs_destroy_ioend(ioend
);
248 * Call IO completion handling in caller context on the final put of an ioend.
251 xfs_finish_ioend_sync(
252 struct xfs_ioend
*ioend
)
254 if (atomic_dec_and_test(&ioend
->io_remaining
))
255 xfs_end_io(&ioend
->io_work
);
259 * Allocate and initialise an IO completion structure.
260 * We need to track unwritten extent write completion here initially.
261 * We'll need to extend this for updating the ondisk inode size later
271 ioend
= mempool_alloc(xfs_ioend_pool
, GFP_NOFS
);
274 * Set the count to 1 initially, which will prevent an I/O
275 * completion callback from happening before we have started
276 * all the I/O from calling the completion routine too early.
278 atomic_set(&ioend
->io_remaining
, 1);
279 ioend
->io_isdirect
= 0;
281 ioend
->io_list
= NULL
;
282 ioend
->io_type
= type
;
283 ioend
->io_inode
= inode
;
284 ioend
->io_buffer_head
= NULL
;
285 ioend
->io_buffer_tail
= NULL
;
286 ioend
->io_offset
= 0;
288 ioend
->io_append_trans
= NULL
;
290 INIT_WORK(&ioend
->io_work
, xfs_end_io
);
298 struct xfs_bmbt_irec
*imap
,
302 struct xfs_inode
*ip
= XFS_I(inode
);
303 struct xfs_mount
*mp
= ip
->i_mount
;
304 ssize_t count
= 1 << inode
->i_blkbits
;
305 xfs_fileoff_t offset_fsb
, end_fsb
;
307 int bmapi_flags
= XFS_BMAPI_ENTIRE
;
310 if (XFS_FORCED_SHUTDOWN(mp
))
313 if (type
== XFS_IO_UNWRITTEN
)
314 bmapi_flags
|= XFS_BMAPI_IGSTATE
;
316 if (!xfs_ilock_nowait(ip
, XFS_ILOCK_SHARED
)) {
319 xfs_ilock(ip
, XFS_ILOCK_SHARED
);
322 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_BTREE
||
323 (ip
->i_df
.if_flags
& XFS_IFEXTENTS
));
324 ASSERT(offset
<= mp
->m_super
->s_maxbytes
);
326 if (offset
+ count
> mp
->m_super
->s_maxbytes
)
327 count
= mp
->m_super
->s_maxbytes
- offset
;
328 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ count
);
329 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
330 error
= xfs_bmapi_read(ip
, offset_fsb
, end_fsb
- offset_fsb
,
331 imap
, &nimaps
, bmapi_flags
);
332 xfs_iunlock(ip
, XFS_ILOCK_SHARED
);
337 if (type
== XFS_IO_DELALLOC
&&
338 (!nimaps
|| isnullstartblock(imap
->br_startblock
))) {
339 error
= xfs_iomap_write_allocate(ip
, offset
, imap
);
341 trace_xfs_map_blocks_alloc(ip
, offset
, count
, type
, imap
);
346 if (type
== XFS_IO_UNWRITTEN
) {
348 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
349 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
353 trace_xfs_map_blocks_found(ip
, offset
, count
, type
, imap
);
360 struct xfs_bmbt_irec
*imap
,
363 offset
>>= inode
->i_blkbits
;
365 return offset
>= imap
->br_startoff
&&
366 offset
< imap
->br_startoff
+ imap
->br_blockcount
;
370 * BIO completion handler for buffered IO.
377 xfs_ioend_t
*ioend
= bio
->bi_private
;
379 ASSERT(atomic_read(&bio
->bi_cnt
) >= 1);
380 ioend
->io_error
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
) ? 0 : error
;
382 /* Toss bio and pass work off to an xfsdatad thread */
383 bio
->bi_private
= NULL
;
384 bio
->bi_end_io
= NULL
;
387 xfs_finish_ioend(ioend
);
391 xfs_submit_ioend_bio(
392 struct writeback_control
*wbc
,
396 atomic_inc(&ioend
->io_remaining
);
397 bio
->bi_private
= ioend
;
398 bio
->bi_end_io
= xfs_end_bio
;
399 submit_bio(wbc
->sync_mode
== WB_SYNC_ALL
? WRITE_SYNC
: WRITE
, bio
);
404 struct buffer_head
*bh
)
406 int nvecs
= bio_get_nr_vecs(bh
->b_bdev
);
407 struct bio
*bio
= bio_alloc(GFP_NOIO
, nvecs
);
409 ASSERT(bio
->bi_private
== NULL
);
410 bio
->bi_iter
.bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
411 bio
->bi_bdev
= bh
->b_bdev
;
416 xfs_start_buffer_writeback(
417 struct buffer_head
*bh
)
419 ASSERT(buffer_mapped(bh
));
420 ASSERT(buffer_locked(bh
));
421 ASSERT(!buffer_delay(bh
));
422 ASSERT(!buffer_unwritten(bh
));
424 mark_buffer_async_write(bh
);
425 set_buffer_uptodate(bh
);
426 clear_buffer_dirty(bh
);
430 xfs_start_page_writeback(
435 ASSERT(PageLocked(page
));
436 ASSERT(!PageWriteback(page
));
439 * if the page was not fully cleaned, we need to ensure that the higher
440 * layers come back to it correctly. That means we need to keep the page
441 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
442 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
443 * write this page in this writeback sweep will be made.
446 clear_page_dirty_for_io(page
);
447 set_page_writeback(page
);
449 set_page_writeback_keepwrite(page
);
453 /* If no buffers on the page are to be written, finish it here */
455 end_page_writeback(page
);
458 static inline int xfs_bio_add_buffer(struct bio
*bio
, struct buffer_head
*bh
)
460 return bio_add_page(bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
464 * Submit all of the bios for all of the ioends we have saved up, covering the
465 * initial writepage page and also any probed pages.
467 * Because we may have multiple ioends spanning a page, we need to start
468 * writeback on all the buffers before we submit them for I/O. If we mark the
469 * buffers as we got, then we can end up with a page that only has buffers
470 * marked async write and I/O complete on can occur before we mark the other
471 * buffers async write.
473 * The end result of this is that we trip a bug in end_page_writeback() because
474 * we call it twice for the one page as the code in end_buffer_async_write()
475 * assumes that all buffers on the page are started at the same time.
477 * The fix is two passes across the ioend list - one to start writeback on the
478 * buffer_heads, and then submit them for I/O on the second pass.
480 * If @fail is non-zero, it means that we have a situation where some part of
481 * the submission process has failed after we have marked paged for writeback
482 * and unlocked them. In this situation, we need to fail the ioend chain rather
483 * than submit it to IO. This typically only happens on a filesystem shutdown.
487 struct writeback_control
*wbc
,
491 xfs_ioend_t
*head
= ioend
;
493 struct buffer_head
*bh
;
495 sector_t lastblock
= 0;
497 /* Pass 1 - start writeback */
499 next
= ioend
->io_list
;
500 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
)
501 xfs_start_buffer_writeback(bh
);
502 } while ((ioend
= next
) != NULL
);
504 /* Pass 2 - submit I/O */
507 next
= ioend
->io_list
;
511 * If we are failing the IO now, just mark the ioend with an
512 * error and finish it. This will run IO completion immediately
513 * as there is only one reference to the ioend at this point in
517 ioend
->io_error
= fail
;
518 xfs_finish_ioend(ioend
);
522 for (bh
= ioend
->io_buffer_head
; bh
; bh
= bh
->b_private
) {
526 bio
= xfs_alloc_ioend_bio(bh
);
527 } else if (bh
->b_blocknr
!= lastblock
+ 1) {
528 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
532 if (xfs_bio_add_buffer(bio
, bh
) != bh
->b_size
) {
533 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
537 lastblock
= bh
->b_blocknr
;
540 xfs_submit_ioend_bio(wbc
, ioend
, bio
);
541 xfs_finish_ioend(ioend
);
542 } while ((ioend
= next
) != NULL
);
546 * Cancel submission of all buffer_heads so far in this endio.
547 * Toss the endio too. Only ever called for the initial page
548 * in a writepage request, so only ever one page.
555 struct buffer_head
*bh
, *next_bh
;
558 next
= ioend
->io_list
;
559 bh
= ioend
->io_buffer_head
;
561 next_bh
= bh
->b_private
;
562 clear_buffer_async_write(bh
);
564 * The unwritten flag is cleared when added to the
565 * ioend. We're not submitting for I/O so mark the
566 * buffer unwritten again for next time around.
568 if (ioend
->io_type
== XFS_IO_UNWRITTEN
)
569 set_buffer_unwritten(bh
);
571 } while ((bh
= next_bh
) != NULL
);
573 mempool_free(ioend
, xfs_ioend_pool
);
574 } while ((ioend
= next
) != NULL
);
578 * Test to see if we've been building up a completion structure for
579 * earlier buffers -- if so, we try to append to this ioend if we
580 * can, otherwise we finish off any current ioend and start another.
581 * Return true if we've finished the given ioend.
586 struct buffer_head
*bh
,
589 xfs_ioend_t
**result
,
592 xfs_ioend_t
*ioend
= *result
;
594 if (!ioend
|| need_ioend
|| type
!= ioend
->io_type
) {
595 xfs_ioend_t
*previous
= *result
;
597 ioend
= xfs_alloc_ioend(inode
, type
);
598 ioend
->io_offset
= offset
;
599 ioend
->io_buffer_head
= bh
;
600 ioend
->io_buffer_tail
= bh
;
602 previous
->io_list
= ioend
;
605 ioend
->io_buffer_tail
->b_private
= bh
;
606 ioend
->io_buffer_tail
= bh
;
609 bh
->b_private
= NULL
;
610 ioend
->io_size
+= bh
->b_size
;
616 struct buffer_head
*bh
,
617 struct xfs_bmbt_irec
*imap
,
621 struct xfs_mount
*m
= XFS_I(inode
)->i_mount
;
622 xfs_off_t iomap_offset
= XFS_FSB_TO_B(m
, imap
->br_startoff
);
623 xfs_daddr_t iomap_bn
= xfs_fsb_to_db(XFS_I(inode
), imap
->br_startblock
);
625 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
626 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
628 bn
= (iomap_bn
>> (inode
->i_blkbits
- BBSHIFT
)) +
629 ((offset
- iomap_offset
) >> inode
->i_blkbits
);
631 ASSERT(bn
|| XFS_IS_REALTIME_INODE(XFS_I(inode
)));
634 set_buffer_mapped(bh
);
640 struct buffer_head
*bh
,
641 struct xfs_bmbt_irec
*imap
,
644 ASSERT(imap
->br_startblock
!= HOLESTARTBLOCK
);
645 ASSERT(imap
->br_startblock
!= DELAYSTARTBLOCK
);
647 xfs_map_buffer(inode
, bh
, imap
, offset
);
648 set_buffer_mapped(bh
);
649 clear_buffer_delay(bh
);
650 clear_buffer_unwritten(bh
);
654 * Test if a given page contains at least one buffer of a given @type.
655 * If @check_all_buffers is true, then we walk all the buffers in the page to
656 * try to find one of the type passed in. If it is not set, then the caller only
657 * needs to check the first buffer on the page for a match.
663 bool check_all_buffers
)
665 struct buffer_head
*bh
;
666 struct buffer_head
*head
;
668 if (PageWriteback(page
))
672 if (!page_has_buffers(page
))
675 bh
= head
= page_buffers(page
);
677 if (buffer_unwritten(bh
)) {
678 if (type
== XFS_IO_UNWRITTEN
)
680 } else if (buffer_delay(bh
)) {
681 if (type
== XFS_IO_DELALLOC
)
683 } else if (buffer_dirty(bh
) && buffer_mapped(bh
)) {
684 if (type
== XFS_IO_OVERWRITE
)
688 /* If we are only checking the first buffer, we are done now. */
689 if (!check_all_buffers
)
691 } while ((bh
= bh
->b_this_page
) != head
);
697 * Allocate & map buffers for page given the extent map. Write it out.
698 * except for the original page of a writepage, this is called on
699 * delalloc/unwritten pages only, for the original page it is possible
700 * that the page has no mapping at all.
707 struct xfs_bmbt_irec
*imap
,
708 xfs_ioend_t
**ioendp
,
709 struct writeback_control
*wbc
)
711 struct buffer_head
*bh
, *head
;
712 xfs_off_t end_offset
;
713 unsigned long p_offset
;
716 int count
= 0, done
= 0, uptodate
= 1;
717 xfs_off_t offset
= page_offset(page
);
719 if (page
->index
!= tindex
)
721 if (!trylock_page(page
))
723 if (PageWriteback(page
))
724 goto fail_unlock_page
;
725 if (page
->mapping
!= inode
->i_mapping
)
726 goto fail_unlock_page
;
727 if (!xfs_check_page_type(page
, (*ioendp
)->io_type
, false))
728 goto fail_unlock_page
;
731 * page_dirty is initially a count of buffers on the page before
732 * EOF and is decremented as we move each into a cleanable state.
736 * End offset is the highest offset that this page should represent.
737 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
738 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
739 * hence give us the correct page_dirty count. On any other page,
740 * it will be zero and in that case we need page_dirty to be the
741 * count of buffers on the page.
743 end_offset
= min_t(unsigned long long,
744 (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
,
748 * If the current map does not span the entire page we are about to try
749 * to write, then give up. The only way we can write a page that spans
750 * multiple mappings in a single writeback iteration is via the
751 * xfs_vm_writepage() function. Data integrity writeback requires the
752 * entire page to be written in a single attempt, otherwise the part of
753 * the page we don't write here doesn't get written as part of the data
756 * For normal writeback, we also don't attempt to write partial pages
757 * here as it simply means that write_cache_pages() will see it under
758 * writeback and ignore the page until some point in the future, at
759 * which time this will be the only page in the file that needs
760 * writeback. Hence for more optimal IO patterns, we should always
761 * avoid partial page writeback due to multiple mappings on a page here.
763 if (!xfs_imap_valid(inode
, imap
, end_offset
))
764 goto fail_unlock_page
;
766 len
= 1 << inode
->i_blkbits
;
767 p_offset
= min_t(unsigned long, end_offset
& (PAGE_CACHE_SIZE
- 1),
769 p_offset
= p_offset
? roundup(p_offset
, len
) : PAGE_CACHE_SIZE
;
770 page_dirty
= p_offset
/ len
;
773 * The moment we find a buffer that doesn't match our current type
774 * specification or can't be written, abort the loop and start
775 * writeback. As per the above xfs_imap_valid() check, only
776 * xfs_vm_writepage() can handle partial page writeback fully - we are
777 * limited here to the buffers that are contiguous with the current
778 * ioend, and hence a buffer we can't write breaks that contiguity and
779 * we have to defer the rest of the IO to xfs_vm_writepage().
781 bh
= head
= page_buffers(page
);
783 if (offset
>= end_offset
)
785 if (!buffer_uptodate(bh
))
787 if (!(PageUptodate(page
) || buffer_uptodate(bh
))) {
792 if (buffer_unwritten(bh
) || buffer_delay(bh
) ||
794 if (buffer_unwritten(bh
))
795 type
= XFS_IO_UNWRITTEN
;
796 else if (buffer_delay(bh
))
797 type
= XFS_IO_DELALLOC
;
799 type
= XFS_IO_OVERWRITE
;
802 * imap should always be valid because of the above
803 * partial page end_offset check on the imap.
805 ASSERT(xfs_imap_valid(inode
, imap
, offset
));
808 if (type
!= XFS_IO_OVERWRITE
)
809 xfs_map_at_offset(inode
, bh
, imap
, offset
);
810 xfs_add_to_ioend(inode
, bh
, offset
, type
,
819 } while (offset
+= len
, (bh
= bh
->b_this_page
) != head
);
821 if (uptodate
&& bh
== head
)
822 SetPageUptodate(page
);
825 if (--wbc
->nr_to_write
<= 0 &&
826 wbc
->sync_mode
== WB_SYNC_NONE
)
829 xfs_start_page_writeback(page
, !page_dirty
, count
);
839 * Convert & write out a cluster of pages in the same extent as defined
840 * by mp and following the start page.
846 struct xfs_bmbt_irec
*imap
,
847 xfs_ioend_t
**ioendp
,
848 struct writeback_control
*wbc
,
854 pagevec_init(&pvec
, 0);
855 while (!done
&& tindex
<= tlast
) {
856 unsigned len
= min_t(pgoff_t
, PAGEVEC_SIZE
, tlast
- tindex
+ 1);
858 if (!pagevec_lookup(&pvec
, inode
->i_mapping
, tindex
, len
))
861 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
862 done
= xfs_convert_page(inode
, pvec
.pages
[i
], tindex
++,
868 pagevec_release(&pvec
);
874 xfs_vm_invalidatepage(
879 trace_xfs_invalidatepage(page
->mapping
->host
, page
, offset
,
881 block_invalidatepage(page
, offset
, length
);
885 * If the page has delalloc buffers on it, we need to punch them out before we
886 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
887 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
888 * is done on that same region - the delalloc extent is returned when none is
889 * supposed to be there.
891 * We prevent this by truncating away the delalloc regions on the page before
892 * invalidating it. Because they are delalloc, we can do this without needing a
893 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
894 * truncation without a transaction as there is no space left for block
895 * reservation (typically why we see a ENOSPC in writeback).
897 * This is not a performance critical path, so for now just do the punching a
898 * buffer head at a time.
901 xfs_aops_discard_page(
904 struct inode
*inode
= page
->mapping
->host
;
905 struct xfs_inode
*ip
= XFS_I(inode
);
906 struct buffer_head
*bh
, *head
;
907 loff_t offset
= page_offset(page
);
909 if (!xfs_check_page_type(page
, XFS_IO_DELALLOC
, true))
912 if (XFS_FORCED_SHUTDOWN(ip
->i_mount
))
915 xfs_alert(ip
->i_mount
,
916 "page discard on page %p, inode 0x%llx, offset %llu.",
917 page
, ip
->i_ino
, offset
);
919 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
920 bh
= head
= page_buffers(page
);
923 xfs_fileoff_t start_fsb
;
925 if (!buffer_delay(bh
))
928 start_fsb
= XFS_B_TO_FSBT(ip
->i_mount
, offset
);
929 error
= xfs_bmap_punch_delalloc_range(ip
, start_fsb
, 1);
931 /* something screwed, just bail */
932 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
933 xfs_alert(ip
->i_mount
,
934 "page discard unable to remove delalloc mapping.");
939 offset
+= 1 << inode
->i_blkbits
;
941 } while ((bh
= bh
->b_this_page
) != head
);
943 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
945 xfs_vm_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
950 * Write out a dirty page.
952 * For delalloc space on the page we need to allocate space and flush it.
953 * For unwritten space on the page we need to start the conversion to
954 * regular allocated space.
955 * For any other dirty buffer heads on the page we should flush them.
960 struct writeback_control
*wbc
)
962 struct inode
*inode
= page
->mapping
->host
;
963 struct buffer_head
*bh
, *head
;
964 struct xfs_bmbt_irec imap
;
965 xfs_ioend_t
*ioend
= NULL
, *iohead
= NULL
;
968 __uint64_t end_offset
;
969 pgoff_t end_index
, last_index
;
971 int err
, imap_valid
= 0, uptodate
= 1;
975 trace_xfs_writepage(inode
, page
, 0, 0);
977 ASSERT(page_has_buffers(page
));
980 * Refuse to write the page out if we are called from reclaim context.
982 * This avoids stack overflows when called from deeply used stacks in
983 * random callers for direct reclaim or memcg reclaim. We explicitly
984 * allow reclaim from kswapd as the stack usage there is relatively low.
986 * This should never happen except in the case of a VM regression so
989 if (WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
)) ==
994 * Given that we do not allow direct reclaim to call us, we should
995 * never be called while in a filesystem transaction.
997 if (WARN_ON_ONCE(current
->flags
& PF_FSTRANS
))
1000 /* Is this page beyond the end of the file? */
1001 offset
= i_size_read(inode
);
1002 end_index
= offset
>> PAGE_CACHE_SHIFT
;
1003 last_index
= (offset
- 1) >> PAGE_CACHE_SHIFT
;
1006 * The page index is less than the end_index, adjust the end_offset
1007 * to the highest offset that this page should represent.
1008 * -----------------------------------------------------
1009 * | file mapping | <EOF> |
1010 * -----------------------------------------------------
1011 * | Page ... | Page N-2 | Page N-1 | Page N | |
1012 * ^--------------------------------^----------|--------
1013 * | desired writeback range | see else |
1014 * ---------------------------------^------------------|
1016 if (page
->index
< end_index
)
1017 end_offset
= (xfs_off_t
)(page
->index
+ 1) << PAGE_CACHE_SHIFT
;
1020 * Check whether the page to write out is beyond or straddles
1022 * -------------------------------------------------------
1023 * | file mapping | <EOF> |
1024 * -------------------------------------------------------
1025 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1026 * ^--------------------------------^-----------|---------
1028 * ---------------------------------^-----------|--------|
1030 unsigned offset_into_page
= offset
& (PAGE_CACHE_SIZE
- 1);
1033 * Skip the page if it is fully outside i_size, e.g. due to a
1034 * truncate operation that is in progress. We must redirty the
1035 * page so that reclaim stops reclaiming it. Otherwise
1036 * xfs_vm_releasepage() is called on it and gets confused.
1038 * Note that the end_index is unsigned long, it would overflow
1039 * if the given offset is greater than 16TB on 32-bit system
1040 * and if we do check the page is fully outside i_size or not
1041 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1042 * will be evaluated to 0. Hence this page will be redirtied
1043 * and be written out repeatedly which would result in an
1044 * infinite loop, the user program that perform this operation
1045 * will hang. Instead, we can verify this situation by checking
1046 * if the page to write is totally beyond the i_size or if it's
1047 * offset is just equal to the EOF.
1049 if (page
->index
> end_index
||
1050 (page
->index
== end_index
&& offset_into_page
== 0))
1054 * The page straddles i_size. It must be zeroed out on each
1055 * and every writepage invocation because it may be mmapped.
1056 * "A file is mapped in multiples of the page size. For a file
1057 * that is not a multiple of the page size, the remaining
1058 * memory is zeroed when mapped, and writes to that region are
1059 * not written out to the file."
1061 zero_user_segment(page
, offset_into_page
, PAGE_CACHE_SIZE
);
1063 /* Adjust the end_offset to the end of file */
1064 end_offset
= offset
;
1067 len
= 1 << inode
->i_blkbits
;
1069 bh
= head
= page_buffers(page
);
1070 offset
= page_offset(page
);
1071 type
= XFS_IO_OVERWRITE
;
1073 if (wbc
->sync_mode
== WB_SYNC_NONE
)
1079 if (offset
>= end_offset
)
1081 if (!buffer_uptodate(bh
))
1085 * set_page_dirty dirties all buffers in a page, independent
1086 * of their state. The dirty state however is entirely
1087 * meaningless for holes (!mapped && uptodate), so skip
1088 * buffers covering holes here.
1090 if (!buffer_mapped(bh
) && buffer_uptodate(bh
)) {
1095 if (buffer_unwritten(bh
)) {
1096 if (type
!= XFS_IO_UNWRITTEN
) {
1097 type
= XFS_IO_UNWRITTEN
;
1100 } else if (buffer_delay(bh
)) {
1101 if (type
!= XFS_IO_DELALLOC
) {
1102 type
= XFS_IO_DELALLOC
;
1105 } else if (buffer_uptodate(bh
)) {
1106 if (type
!= XFS_IO_OVERWRITE
) {
1107 type
= XFS_IO_OVERWRITE
;
1111 if (PageUptodate(page
))
1112 ASSERT(buffer_mapped(bh
));
1114 * This buffer is not uptodate and will not be
1115 * written to disk. Ensure that we will put any
1116 * subsequent writeable buffers into a new
1124 imap_valid
= xfs_imap_valid(inode
, &imap
, offset
);
1127 * If we didn't have a valid mapping then we need to
1128 * put the new mapping into a separate ioend structure.
1129 * This ensures non-contiguous extents always have
1130 * separate ioends, which is particularly important
1131 * for unwritten extent conversion at I/O completion
1135 err
= xfs_map_blocks(inode
, offset
, &imap
, type
,
1139 imap_valid
= xfs_imap_valid(inode
, &imap
, offset
);
1143 if (type
!= XFS_IO_OVERWRITE
)
1144 xfs_map_at_offset(inode
, bh
, &imap
, offset
);
1145 xfs_add_to_ioend(inode
, bh
, offset
, type
, &ioend
,
1153 } while (offset
+= len
, ((bh
= bh
->b_this_page
) != head
));
1155 if (uptodate
&& bh
== head
)
1156 SetPageUptodate(page
);
1158 xfs_start_page_writeback(page
, 1, count
);
1160 /* if there is no IO to be submitted for this page, we are done */
1167 * Any errors from this point onwards need tobe reported through the IO
1168 * completion path as we have marked the initial page as under writeback
1172 xfs_off_t end_index
;
1174 end_index
= imap
.br_startoff
+ imap
.br_blockcount
;
1177 end_index
<<= inode
->i_blkbits
;
1180 end_index
= (end_index
- 1) >> PAGE_CACHE_SHIFT
;
1182 /* check against file size */
1183 if (end_index
> last_index
)
1184 end_index
= last_index
;
1186 xfs_cluster_write(inode
, page
->index
+ 1, &imap
, &ioend
,
1192 * Reserve log space if we might write beyond the on-disk inode size.
1195 if (ioend
->io_type
!= XFS_IO_UNWRITTEN
&& xfs_ioend_is_append(ioend
))
1196 err
= xfs_setfilesize_trans_alloc(ioend
);
1198 xfs_submit_ioend(wbc
, iohead
, err
);
1204 xfs_cancel_ioend(iohead
);
1209 xfs_aops_discard_page(page
);
1210 ClearPageUptodate(page
);
1215 redirty_page_for_writepage(wbc
, page
);
1222 struct address_space
*mapping
,
1223 struct writeback_control
*wbc
)
1225 xfs_iflags_clear(XFS_I(mapping
->host
), XFS_ITRUNCATED
);
1226 return generic_writepages(mapping
, wbc
);
1230 * Called to move a page into cleanable state - and from there
1231 * to be released. The page should already be clean. We always
1232 * have buffer heads in this call.
1234 * Returns 1 if the page is ok to release, 0 otherwise.
1241 int delalloc
, unwritten
;
1243 trace_xfs_releasepage(page
->mapping
->host
, page
, 0, 0);
1245 xfs_count_page_state(page
, &delalloc
, &unwritten
);
1247 if (WARN_ON_ONCE(delalloc
))
1249 if (WARN_ON_ONCE(unwritten
))
1252 return try_to_free_buffers(page
);
1257 struct inode
*inode
,
1259 struct buffer_head
*bh_result
,
1263 struct xfs_inode
*ip
= XFS_I(inode
);
1264 struct xfs_mount
*mp
= ip
->i_mount
;
1265 xfs_fileoff_t offset_fsb
, end_fsb
;
1268 struct xfs_bmbt_irec imap
;
1274 if (XFS_FORCED_SHUTDOWN(mp
))
1277 offset
= (xfs_off_t
)iblock
<< inode
->i_blkbits
;
1278 ASSERT(bh_result
->b_size
>= (1 << inode
->i_blkbits
));
1279 size
= bh_result
->b_size
;
1281 if (!create
&& direct
&& offset
>= i_size_read(inode
))
1285 * Direct I/O is usually done on preallocated files, so try getting
1286 * a block mapping without an exclusive lock first. For buffered
1287 * writes we already have the exclusive iolock anyway, so avoiding
1288 * a lock roundtrip here by taking the ilock exclusive from the
1289 * beginning is a useful micro optimization.
1291 if (create
&& !direct
) {
1292 lockmode
= XFS_ILOCK_EXCL
;
1293 xfs_ilock(ip
, lockmode
);
1295 lockmode
= xfs_ilock_data_map_shared(ip
);
1298 ASSERT(offset
<= mp
->m_super
->s_maxbytes
);
1299 if (offset
+ size
> mp
->m_super
->s_maxbytes
)
1300 size
= mp
->m_super
->s_maxbytes
- offset
;
1301 end_fsb
= XFS_B_TO_FSB(mp
, (xfs_ufsize_t
)offset
+ size
);
1302 offset_fsb
= XFS_B_TO_FSBT(mp
, offset
);
1304 error
= xfs_bmapi_read(ip
, offset_fsb
, end_fsb
- offset_fsb
,
1305 &imap
, &nimaps
, XFS_BMAPI_ENTIRE
);
1311 (imap
.br_startblock
== HOLESTARTBLOCK
||
1312 imap
.br_startblock
== DELAYSTARTBLOCK
))) {
1313 if (direct
|| xfs_get_extsz_hint(ip
)) {
1315 * Drop the ilock in preparation for starting the block
1316 * allocation transaction. It will be retaken
1317 * exclusively inside xfs_iomap_write_direct for the
1318 * actual allocation.
1320 xfs_iunlock(ip
, lockmode
);
1321 error
= xfs_iomap_write_direct(ip
, offset
, size
,
1328 * Delalloc reservations do not require a transaction,
1329 * we can go on without dropping the lock here. If we
1330 * are allocating a new delalloc block, make sure that
1331 * we set the new flag so that we mark the buffer new so
1332 * that we know that it is newly allocated if the write
1335 if (nimaps
&& imap
.br_startblock
== HOLESTARTBLOCK
)
1337 error
= xfs_iomap_write_delay(ip
, offset
, size
, &imap
);
1341 xfs_iunlock(ip
, lockmode
);
1344 trace_xfs_get_blocks_alloc(ip
, offset
, size
, 0, &imap
);
1345 } else if (nimaps
) {
1346 trace_xfs_get_blocks_found(ip
, offset
, size
, 0, &imap
);
1347 xfs_iunlock(ip
, lockmode
);
1349 trace_xfs_get_blocks_notfound(ip
, offset
, size
);
1353 if (imap
.br_startblock
!= HOLESTARTBLOCK
&&
1354 imap
.br_startblock
!= DELAYSTARTBLOCK
) {
1356 * For unwritten extents do not report a disk address on
1357 * the read case (treat as if we're reading into a hole).
1359 if (create
|| !ISUNWRITTEN(&imap
))
1360 xfs_map_buffer(inode
, bh_result
, &imap
, offset
);
1361 if (create
&& ISUNWRITTEN(&imap
)) {
1363 bh_result
->b_private
= inode
;
1364 set_buffer_defer_completion(bh_result
);
1366 set_buffer_unwritten(bh_result
);
1371 * If this is a realtime file, data may be on a different device.
1372 * to that pointed to from the buffer_head b_bdev currently.
1374 bh_result
->b_bdev
= xfs_find_bdev_for_inode(inode
);
1377 * If we previously allocated a block out beyond eof and we are now
1378 * coming back to use it then we will need to flag it as new even if it
1379 * has a disk address.
1381 * With sub-block writes into unwritten extents we also need to mark
1382 * the buffer as new so that the unwritten parts of the buffer gets
1386 ((!buffer_mapped(bh_result
) && !buffer_uptodate(bh_result
)) ||
1387 (offset
>= i_size_read(inode
)) ||
1388 (new || ISUNWRITTEN(&imap
))))
1389 set_buffer_new(bh_result
);
1391 if (imap
.br_startblock
== DELAYSTARTBLOCK
) {
1394 set_buffer_uptodate(bh_result
);
1395 set_buffer_mapped(bh_result
);
1396 set_buffer_delay(bh_result
);
1401 * If this is O_DIRECT or the mpage code calling tell them how large
1402 * the mapping is, so that we can avoid repeated get_blocks calls.
1404 * If the mapping spans EOF, then we have to break the mapping up as the
1405 * mapping for blocks beyond EOF must be marked new so that sub block
1406 * regions can be correctly zeroed. We can't do this for mappings within
1407 * EOF unless the mapping was just allocated or is unwritten, otherwise
1408 * the callers would overwrite existing data with zeros. Hence we have
1409 * to split the mapping into a range up to and including EOF, and a
1410 * second mapping for beyond EOF.
1412 if (direct
|| size
> (1 << inode
->i_blkbits
)) {
1413 xfs_off_t mapping_size
;
1415 mapping_size
= imap
.br_startoff
+ imap
.br_blockcount
- iblock
;
1416 mapping_size
<<= inode
->i_blkbits
;
1418 ASSERT(mapping_size
> 0);
1419 if (mapping_size
> size
)
1420 mapping_size
= size
;
1421 if (offset
< i_size_read(inode
) &&
1422 offset
+ mapping_size
>= i_size_read(inode
)) {
1423 /* limit mapping to block that spans EOF */
1424 mapping_size
= roundup_64(i_size_read(inode
) - offset
,
1425 1 << inode
->i_blkbits
);
1427 if (mapping_size
> LONG_MAX
)
1428 mapping_size
= LONG_MAX
;
1430 bh_result
->b_size
= mapping_size
;
1436 xfs_iunlock(ip
, lockmode
);
1442 struct inode
*inode
,
1444 struct buffer_head
*bh_result
,
1447 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 0);
1451 xfs_get_blocks_direct(
1452 struct inode
*inode
,
1454 struct buffer_head
*bh_result
,
1457 return __xfs_get_blocks(inode
, iblock
, bh_result
, create
, 1);
1461 * Complete a direct I/O write request.
1463 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1464 * need to issue a transaction to convert the range from unwritten to written
1465 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1466 * to do this and we are done. But in case this was a successful AIO
1467 * request this handler is called from interrupt context, from which we
1468 * can't start transactions. In that case offload the I/O completion to
1469 * the workqueues we also use for buffered I/O completion.
1472 xfs_end_io_direct_write(
1478 struct xfs_ioend
*ioend
= iocb
->private;
1481 * While the generic direct I/O code updates the inode size, it does
1482 * so only after the end_io handler is called, which means our
1483 * end_io handler thinks the on-disk size is outside the in-core
1484 * size. To prevent this just update it a little bit earlier here.
1486 if (offset
+ size
> i_size_read(ioend
->io_inode
))
1487 i_size_write(ioend
->io_inode
, offset
+ size
);
1490 * blockdev_direct_IO can return an error even after the I/O
1491 * completion handler was called. Thus we need to protect
1492 * against double-freeing.
1494 iocb
->private = NULL
;
1496 ioend
->io_offset
= offset
;
1497 ioend
->io_size
= size
;
1498 if (private && size
> 0)
1499 ioend
->io_type
= XFS_IO_UNWRITTEN
;
1501 xfs_finish_ioend_sync(ioend
);
1508 struct iov_iter
*iter
,
1511 struct inode
*inode
= iocb
->ki_filp
->f_mapping
->host
;
1512 struct block_device
*bdev
= xfs_find_bdev_for_inode(inode
);
1513 struct xfs_ioend
*ioend
= NULL
;
1517 size_t size
= iov_iter_count(iter
);
1520 * We cannot preallocate a size update transaction here as we
1521 * don't know whether allocation is necessary or not. Hence we
1522 * can only tell IO completion that one is necessary if we are
1523 * not doing unwritten extent conversion.
1525 iocb
->private = ioend
= xfs_alloc_ioend(inode
, XFS_IO_DIRECT
);
1526 if (offset
+ size
> XFS_I(inode
)->i_d
.di_size
)
1527 ioend
->io_isdirect
= 1;
1529 ret
= __blockdev_direct_IO(rw
, iocb
, inode
, bdev
, iter
,
1530 offset
, xfs_get_blocks_direct
,
1531 xfs_end_io_direct_write
, NULL
,
1533 if (ret
!= -EIOCBQUEUED
&& iocb
->private)
1534 goto out_destroy_ioend
;
1536 ret
= __blockdev_direct_IO(rw
, iocb
, inode
, bdev
, iter
,
1537 offset
, xfs_get_blocks_direct
,
1544 xfs_destroy_ioend(ioend
);
1549 * Punch out the delalloc blocks we have already allocated.
1551 * Don't bother with xfs_setattr given that nothing can have made it to disk yet
1552 * as the page is still locked at this point.
1555 xfs_vm_kill_delalloc_range(
1556 struct inode
*inode
,
1560 struct xfs_inode
*ip
= XFS_I(inode
);
1561 xfs_fileoff_t start_fsb
;
1562 xfs_fileoff_t end_fsb
;
1565 start_fsb
= XFS_B_TO_FSB(ip
->i_mount
, start
);
1566 end_fsb
= XFS_B_TO_FSB(ip
->i_mount
, end
);
1567 if (end_fsb
<= start_fsb
)
1570 xfs_ilock(ip
, XFS_ILOCK_EXCL
);
1571 error
= xfs_bmap_punch_delalloc_range(ip
, start_fsb
,
1572 end_fsb
- start_fsb
);
1574 /* something screwed, just bail */
1575 if (!XFS_FORCED_SHUTDOWN(ip
->i_mount
)) {
1576 xfs_alert(ip
->i_mount
,
1577 "xfs_vm_write_failed: unable to clean up ino %lld",
1581 xfs_iunlock(ip
, XFS_ILOCK_EXCL
);
1585 xfs_vm_write_failed(
1586 struct inode
*inode
,
1591 loff_t block_offset
;
1594 loff_t from
= pos
& (PAGE_CACHE_SIZE
- 1);
1595 loff_t to
= from
+ len
;
1596 struct buffer_head
*bh
, *head
;
1599 * The request pos offset might be 32 or 64 bit, this is all fine
1600 * on 64-bit platform. However, for 64-bit pos request on 32-bit
1601 * platform, the high 32-bit will be masked off if we evaluate the
1602 * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is
1603 * 0xfffff000 as an unsigned long, hence the result is incorrect
1604 * which could cause the following ASSERT failed in most cases.
1605 * In order to avoid this, we can evaluate the block_offset of the
1606 * start of the page by using shifts rather than masks the mismatch
1609 block_offset
= (pos
>> PAGE_CACHE_SHIFT
) << PAGE_CACHE_SHIFT
;
1611 ASSERT(block_offset
+ from
== pos
);
1613 head
= page_buffers(page
);
1615 for (bh
= head
; bh
!= head
|| !block_start
;
1616 bh
= bh
->b_this_page
, block_start
= block_end
,
1617 block_offset
+= bh
->b_size
) {
1618 block_end
= block_start
+ bh
->b_size
;
1620 /* skip buffers before the write */
1621 if (block_end
<= from
)
1624 /* if the buffer is after the write, we're done */
1625 if (block_start
>= to
)
1628 if (!buffer_delay(bh
))
1631 if (!buffer_new(bh
) && block_offset
< i_size_read(inode
))
1634 xfs_vm_kill_delalloc_range(inode
, block_offset
,
1635 block_offset
+ bh
->b_size
);
1638 * This buffer does not contain data anymore. make sure anyone
1639 * who finds it knows that for certain.
1641 clear_buffer_delay(bh
);
1642 clear_buffer_uptodate(bh
);
1643 clear_buffer_mapped(bh
);
1644 clear_buffer_new(bh
);
1645 clear_buffer_dirty(bh
);
1651 * This used to call block_write_begin(), but it unlocks and releases the page
1652 * on error, and we need that page to be able to punch stale delalloc blocks out
1653 * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
1654 * the appropriate point.
1659 struct address_space
*mapping
,
1663 struct page
**pagep
,
1666 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1670 ASSERT(len
<= PAGE_CACHE_SIZE
);
1672 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1676 status
= __block_write_begin(page
, pos
, len
, xfs_get_blocks
);
1677 if (unlikely(status
)) {
1678 struct inode
*inode
= mapping
->host
;
1679 size_t isize
= i_size_read(inode
);
1681 xfs_vm_write_failed(inode
, page
, pos
, len
);
1685 * If the write is beyond EOF, we only want to kill blocks
1686 * allocated in this write, not blocks that were previously
1687 * written successfully.
1689 if (pos
+ len
> isize
) {
1690 ssize_t start
= max_t(ssize_t
, pos
, isize
);
1692 truncate_pagecache_range(inode
, start
, pos
+ len
);
1695 page_cache_release(page
);
1704 * On failure, we only need to kill delalloc blocks beyond EOF in the range of
1705 * this specific write because they will never be written. Previous writes
1706 * beyond EOF where block allocation succeeded do not need to be trashed, so
1707 * only new blocks from this write should be trashed. For blocks within
1708 * EOF, generic_write_end() zeros them so they are safe to leave alone and be
1709 * written with all the other valid data.
1714 struct address_space
*mapping
,
1723 ASSERT(len
<= PAGE_CACHE_SIZE
);
1725 ret
= generic_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1726 if (unlikely(ret
< len
)) {
1727 struct inode
*inode
= mapping
->host
;
1728 size_t isize
= i_size_read(inode
);
1729 loff_t to
= pos
+ len
;
1732 /* only kill blocks in this write beyond EOF */
1735 xfs_vm_kill_delalloc_range(inode
, isize
, to
);
1736 truncate_pagecache_range(inode
, isize
, to
);
1744 struct address_space
*mapping
,
1747 struct inode
*inode
= (struct inode
*)mapping
->host
;
1748 struct xfs_inode
*ip
= XFS_I(inode
);
1750 trace_xfs_vm_bmap(XFS_I(inode
));
1751 xfs_ilock(ip
, XFS_IOLOCK_SHARED
);
1752 filemap_write_and_wait(mapping
);
1753 xfs_iunlock(ip
, XFS_IOLOCK_SHARED
);
1754 return generic_block_bmap(mapping
, block
, xfs_get_blocks
);
1759 struct file
*unused
,
1762 return mpage_readpage(page
, xfs_get_blocks
);
1767 struct file
*unused
,
1768 struct address_space
*mapping
,
1769 struct list_head
*pages
,
1772 return mpage_readpages(mapping
, pages
, nr_pages
, xfs_get_blocks
);
1776 * This is basically a copy of __set_page_dirty_buffers() with one
1777 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1778 * dirty, we'll never be able to clean them because we don't write buffers
1779 * beyond EOF, and that means we can't invalidate pages that span EOF
1780 * that have been marked dirty. Further, the dirty state can leak into
1781 * the file interior if the file is extended, resulting in all sorts of
1782 * bad things happening as the state does not match the underlying data.
1784 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1785 * this only exist because of bufferheads and how the generic code manages them.
1788 xfs_vm_set_page_dirty(
1791 struct address_space
*mapping
= page
->mapping
;
1792 struct inode
*inode
= mapping
->host
;
1797 if (unlikely(!mapping
))
1798 return !TestSetPageDirty(page
);
1800 end_offset
= i_size_read(inode
);
1801 offset
= page_offset(page
);
1803 spin_lock(&mapping
->private_lock
);
1804 if (page_has_buffers(page
)) {
1805 struct buffer_head
*head
= page_buffers(page
);
1806 struct buffer_head
*bh
= head
;
1809 if (offset
< end_offset
)
1810 set_buffer_dirty(bh
);
1811 bh
= bh
->b_this_page
;
1812 offset
+= 1 << inode
->i_blkbits
;
1813 } while (bh
!= head
);
1815 newly_dirty
= !TestSetPageDirty(page
);
1816 spin_unlock(&mapping
->private_lock
);
1819 /* sigh - __set_page_dirty() is static, so copy it here, too */
1820 unsigned long flags
;
1822 spin_lock_irqsave(&mapping
->tree_lock
, flags
);
1823 if (page
->mapping
) { /* Race with truncate? */
1824 WARN_ON_ONCE(!PageUptodate(page
));
1825 account_page_dirtied(page
, mapping
);
1826 radix_tree_tag_set(&mapping
->page_tree
,
1827 page_index(page
), PAGECACHE_TAG_DIRTY
);
1829 spin_unlock_irqrestore(&mapping
->tree_lock
, flags
);
1830 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
1835 const struct address_space_operations xfs_address_space_operations
= {
1836 .readpage
= xfs_vm_readpage
,
1837 .readpages
= xfs_vm_readpages
,
1838 .writepage
= xfs_vm_writepage
,
1839 .writepages
= xfs_vm_writepages
,
1840 .set_page_dirty
= xfs_vm_set_page_dirty
,
1841 .releasepage
= xfs_vm_releasepage
,
1842 .invalidatepage
= xfs_vm_invalidatepage
,
1843 .write_begin
= xfs_vm_write_begin
,
1844 .write_end
= xfs_vm_write_end
,
1845 .bmap
= xfs_vm_bmap
,
1846 .direct_IO
= xfs_vm_direct_IO
,
1847 .migratepage
= buffer_migrate_page
,
1848 .is_partially_uptodate
= block_is_partially_uptodate
,
1849 .error_remove_page
= generic_error_remove_page
,