1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 2002, Linus Torvalds.
7 * Contains functions related to preparing and submitting BIOs which contain
8 * multiple pagecache pages.
10 * 15May2002 Andrew Morton
12 * 27Jun2002 axboe@suse.de
13 * use bio_add_page() to build bio's just the right size
16 #include <linux/kernel.h>
17 #include <linux/export.h>
19 #include <linux/kdev_t.h>
20 #include <linux/gfp.h>
21 #include <linux/bio.h>
23 #include <linux/buffer_head.h>
24 #include <linux/blkdev.h>
25 #include <linux/highmem.h>
26 #include <linux/prefetch.h>
27 #include <linux/mpage.h>
28 #include <linux/mm_inline.h>
29 #include <linux/writeback.h>
30 #include <linux/backing-dev.h>
31 #include <linux/pagevec.h>
35 * I/O completion handler for multipage BIOs.
37 * The mpage code never puts partial pages into a BIO (except for end-of-file).
38 * If a page does not map to a contiguous run of blocks then it simply falls
39 * back to block_read_full_folio().
41 * Why is this? If a page's completion depends on a number of different BIOs
42 * which can complete in any order (or at the same time) then determining the
43 * status of that page is hard. See end_buffer_async_read() for the details.
44 * There is no point in duplicating all that complexity.
46 static void mpage_end_io(struct bio
*bio
)
49 struct bvec_iter_all iter_all
;
51 bio_for_each_segment_all(bv
, bio
, iter_all
) {
52 struct page
*page
= bv
->bv_page
;
53 page_endio(page
, bio_op(bio
),
54 blk_status_to_errno(bio
->bi_status
));
60 static struct bio
*mpage_bio_submit(struct bio
*bio
)
62 bio
->bi_end_io
= mpage_end_io
;
69 * support function for mpage_readahead. The fs supplied get_block might
70 * return an up to date buffer. This is used to map that buffer into
71 * the page, which allows read_folio to avoid triggering a duplicate call
74 * The idea is to avoid adding buffers to pages that don't already have
75 * them. So when the buffer is up to date and the page size == block size,
76 * this marks the page up to date instead of adding new buffers.
79 map_buffer_to_page(struct page
*page
, struct buffer_head
*bh
, int page_block
)
81 struct inode
*inode
= page
->mapping
->host
;
82 struct buffer_head
*page_bh
, *head
;
85 if (!page_has_buffers(page
)) {
87 * don't make any buffers if there is only one buffer on
88 * the page and the page just needs to be set up to date
90 if (inode
->i_blkbits
== PAGE_SHIFT
&&
91 buffer_uptodate(bh
)) {
92 SetPageUptodate(page
);
95 create_empty_buffers(page
, i_blocksize(inode
), 0);
97 head
= page_buffers(page
);
100 if (block
== page_block
) {
101 page_bh
->b_state
= bh
->b_state
;
102 page_bh
->b_bdev
= bh
->b_bdev
;
103 page_bh
->b_blocknr
= bh
->b_blocknr
;
106 page_bh
= page_bh
->b_this_page
;
108 } while (page_bh
!= head
);
111 struct mpage_readpage_args
{
114 unsigned int nr_pages
;
116 sector_t last_block_in_bio
;
117 struct buffer_head map_bh
;
118 unsigned long first_logical_block
;
119 get_block_t
*get_block
;
123 * This is the worker routine which does all the work of mapping the disk
124 * blocks and constructs largest possible bios, submits them for IO if the
125 * blocks are not contiguous on the disk.
127 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
128 * represent the validity of its disk mapping and to decide when to do the next
131 static struct bio
*do_mpage_readpage(struct mpage_readpage_args
*args
)
133 struct page
*page
= args
->page
;
134 struct inode
*inode
= page
->mapping
->host
;
135 const unsigned blkbits
= inode
->i_blkbits
;
136 const unsigned blocks_per_page
= PAGE_SIZE
>> blkbits
;
137 const unsigned blocksize
= 1 << blkbits
;
138 struct buffer_head
*map_bh
= &args
->map_bh
;
139 sector_t block_in_file
;
141 sector_t last_block_in_file
;
142 sector_t blocks
[MAX_BUF_PER_PAGE
];
144 unsigned first_hole
= blocks_per_page
;
145 struct block_device
*bdev
= NULL
;
147 int fully_mapped
= 1;
148 int op
= REQ_OP_READ
;
150 unsigned relative_block
;
151 gfp_t gfp
= mapping_gfp_constraint(page
->mapping
, GFP_KERNEL
);
153 if (args
->is_readahead
) {
155 gfp
|= __GFP_NORETRY
| __GFP_NOWARN
;
158 if (page_has_buffers(page
))
161 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
162 last_block
= block_in_file
+ args
->nr_pages
* blocks_per_page
;
163 last_block_in_file
= (i_size_read(inode
) + blocksize
- 1) >> blkbits
;
164 if (last_block
> last_block_in_file
)
165 last_block
= last_block_in_file
;
169 * Map blocks using the result from the previous get_blocks call first.
171 nblocks
= map_bh
->b_size
>> blkbits
;
172 if (buffer_mapped(map_bh
) &&
173 block_in_file
> args
->first_logical_block
&&
174 block_in_file
< (args
->first_logical_block
+ nblocks
)) {
175 unsigned map_offset
= block_in_file
- args
->first_logical_block
;
176 unsigned last
= nblocks
- map_offset
;
178 for (relative_block
= 0; ; relative_block
++) {
179 if (relative_block
== last
) {
180 clear_buffer_mapped(map_bh
);
183 if (page_block
== blocks_per_page
)
185 blocks
[page_block
] = map_bh
->b_blocknr
+ map_offset
+
190 bdev
= map_bh
->b_bdev
;
194 * Then do more get_blocks calls until we are done with this page.
196 map_bh
->b_page
= page
;
197 while (page_block
< blocks_per_page
) {
201 if (block_in_file
< last_block
) {
202 map_bh
->b_size
= (last_block
-block_in_file
) << blkbits
;
203 if (args
->get_block(inode
, block_in_file
, map_bh
, 0))
205 args
->first_logical_block
= block_in_file
;
208 if (!buffer_mapped(map_bh
)) {
210 if (first_hole
== blocks_per_page
)
211 first_hole
= page_block
;
217 /* some filesystems will copy data into the page during
218 * the get_block call, in which case we don't want to
219 * read it again. map_buffer_to_page copies the data
220 * we just collected from get_block into the page's buffers
221 * so readpage doesn't have to repeat the get_block call
223 if (buffer_uptodate(map_bh
)) {
224 map_buffer_to_page(page
, map_bh
, page_block
);
228 if (first_hole
!= blocks_per_page
)
229 goto confused
; /* hole -> non-hole */
231 /* Contiguous blocks? */
232 if (page_block
&& blocks
[page_block
-1] != map_bh
->b_blocknr
-1)
234 nblocks
= map_bh
->b_size
>> blkbits
;
235 for (relative_block
= 0; ; relative_block
++) {
236 if (relative_block
== nblocks
) {
237 clear_buffer_mapped(map_bh
);
239 } else if (page_block
== blocks_per_page
)
241 blocks
[page_block
] = map_bh
->b_blocknr
+relative_block
;
245 bdev
= map_bh
->b_bdev
;
248 if (first_hole
!= blocks_per_page
) {
249 zero_user_segment(page
, first_hole
<< blkbits
, PAGE_SIZE
);
250 if (first_hole
== 0) {
251 SetPageUptodate(page
);
255 } else if (fully_mapped
) {
256 SetPageMappedToDisk(page
);
260 * This page will go to BIO. Do we need to send this BIO off first?
262 if (args
->bio
&& (args
->last_block_in_bio
!= blocks
[0] - 1))
263 args
->bio
= mpage_bio_submit(args
->bio
);
266 if (args
->bio
== NULL
) {
267 if (first_hole
== blocks_per_page
) {
268 if (!bdev_read_page(bdev
, blocks
[0] << (blkbits
- 9),
272 args
->bio
= bio_alloc(bdev
, bio_max_segs(args
->nr_pages
), op
,
274 if (args
->bio
== NULL
)
276 args
->bio
->bi_iter
.bi_sector
= blocks
[0] << (blkbits
- 9);
279 length
= first_hole
<< blkbits
;
280 if (bio_add_page(args
->bio
, page
, length
, 0) < length
) {
281 args
->bio
= mpage_bio_submit(args
->bio
);
285 relative_block
= block_in_file
- args
->first_logical_block
;
286 nblocks
= map_bh
->b_size
>> blkbits
;
287 if ((buffer_boundary(map_bh
) && relative_block
== nblocks
) ||
288 (first_hole
!= blocks_per_page
))
289 args
->bio
= mpage_bio_submit(args
->bio
);
291 args
->last_block_in_bio
= blocks
[blocks_per_page
- 1];
297 args
->bio
= mpage_bio_submit(args
->bio
);
298 if (!PageUptodate(page
))
299 block_read_full_folio(page_folio(page
), args
->get_block
);
306 * mpage_readahead - start reads against pages
307 * @rac: Describes which pages to read.
308 * @get_block: The filesystem's block mapper function.
310 * This function walks the pages and the blocks within each page, building and
311 * emitting large BIOs.
313 * If anything unusual happens, such as:
315 * - encountering a page which has buffers
316 * - encountering a page which has a non-hole after a hole
317 * - encountering a page with non-contiguous blocks
319 * then this code just gives up and calls the buffer_head-based read function.
320 * It does handle a page which has holes at the end - that is a common case:
321 * the end-of-file on blocksize < PAGE_SIZE setups.
323 * BH_Boundary explanation:
325 * There is a problem. The mpage read code assembles several pages, gets all
326 * their disk mappings, and then submits them all. That's fine, but obtaining
327 * the disk mappings may require I/O. Reads of indirect blocks, for example.
329 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
330 * submitted in the following order:
332 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
334 * because the indirect block has to be read to get the mappings of blocks
335 * 13,14,15,16. Obviously, this impacts performance.
337 * So what we do it to allow the filesystem's get_block() function to set
338 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
339 * after this one will require I/O against a block which is probably close to
340 * this one. So you should push what I/O you have currently accumulated.
342 * This all causes the disk requests to be issued in the correct order.
344 void mpage_readahead(struct readahead_control
*rac
, get_block_t get_block
)
347 struct mpage_readpage_args args
= {
348 .get_block
= get_block
,
349 .is_readahead
= true,
352 while ((page
= readahead_page(rac
))) {
353 prefetchw(&page
->flags
);
355 args
.nr_pages
= readahead_count(rac
);
356 args
.bio
= do_mpage_readpage(&args
);
360 mpage_bio_submit(args
.bio
);
362 EXPORT_SYMBOL(mpage_readahead
);
365 * This isn't called much at all
367 int mpage_read_folio(struct folio
*folio
, get_block_t get_block
)
369 struct mpage_readpage_args args
= {
370 .page
= &folio
->page
,
372 .get_block
= get_block
,
375 VM_BUG_ON_FOLIO(folio_test_large(folio
), folio
);
377 args
.bio
= do_mpage_readpage(&args
);
379 mpage_bio_submit(args
.bio
);
382 EXPORT_SYMBOL(mpage_read_folio
);
385 * Writing is not so simple.
387 * If the page has buffers then they will be used for obtaining the disk
388 * mapping. We only support pages which are fully mapped-and-dirty, with a
389 * special case for pages which are unmapped at the end: end-of-file.
391 * If the page has no buffers (preferred) then the page is mapped here.
393 * If all blocks are found to be contiguous then the page can go into the
394 * BIO. Otherwise fall back to the mapping's writepage().
396 * FIXME: This code wants an estimate of how many pages are still to be
397 * written, so it can intelligently allocate a suitably-sized BIO. For now,
398 * just allocate full-size (16-page) BIOs.
403 sector_t last_block_in_bio
;
404 get_block_t
*get_block
;
405 unsigned use_writepage
;
409 * We have our BIO, so we can now mark the buffers clean. Make
410 * sure to only clean buffers which we know we'll be writing.
412 static void clean_buffers(struct page
*page
, unsigned first_unmapped
)
414 unsigned buffer_counter
= 0;
415 struct buffer_head
*bh
, *head
;
416 if (!page_has_buffers(page
))
418 head
= page_buffers(page
);
422 if (buffer_counter
++ == first_unmapped
)
424 clear_buffer_dirty(bh
);
425 bh
= bh
->b_this_page
;
426 } while (bh
!= head
);
429 * we cannot drop the bh if the page is not uptodate or a concurrent
430 * read_folio would fail to serialize with the bh and it would read from
431 * disk before we reach the platter.
433 if (buffer_heads_over_limit
&& PageUptodate(page
))
434 try_to_free_buffers(page_folio(page
));
438 * For situations where we want to clean all buffers attached to a page.
439 * We don't need to calculate how many buffers are attached to the page,
440 * we just need to specify a number larger than the maximum number of buffers.
442 void clean_page_buffers(struct page
*page
)
444 clean_buffers(page
, ~0U);
447 static int __mpage_writepage(struct page
*page
, struct writeback_control
*wbc
,
450 struct mpage_data
*mpd
= data
;
451 struct bio
*bio
= mpd
->bio
;
452 struct address_space
*mapping
= page
->mapping
;
453 struct inode
*inode
= page
->mapping
->host
;
454 const unsigned blkbits
= inode
->i_blkbits
;
455 unsigned long end_index
;
456 const unsigned blocks_per_page
= PAGE_SIZE
>> blkbits
;
458 sector_t block_in_file
;
459 sector_t blocks
[MAX_BUF_PER_PAGE
];
461 unsigned first_unmapped
= blocks_per_page
;
462 struct block_device
*bdev
= NULL
;
464 sector_t boundary_block
= 0;
465 struct block_device
*boundary_bdev
= NULL
;
467 struct buffer_head map_bh
;
468 loff_t i_size
= i_size_read(inode
);
471 if (page_has_buffers(page
)) {
472 struct buffer_head
*head
= page_buffers(page
);
473 struct buffer_head
*bh
= head
;
475 /* If they're all mapped and dirty, do it */
478 BUG_ON(buffer_locked(bh
));
479 if (!buffer_mapped(bh
)) {
481 * unmapped dirty buffers are created by
482 * block_dirty_folio -> mmapped data
484 if (buffer_dirty(bh
))
486 if (first_unmapped
== blocks_per_page
)
487 first_unmapped
= page_block
;
491 if (first_unmapped
!= blocks_per_page
)
492 goto confused
; /* hole -> non-hole */
494 if (!buffer_dirty(bh
) || !buffer_uptodate(bh
))
497 if (bh
->b_blocknr
!= blocks
[page_block
-1] + 1)
500 blocks
[page_block
++] = bh
->b_blocknr
;
501 boundary
= buffer_boundary(bh
);
503 boundary_block
= bh
->b_blocknr
;
504 boundary_bdev
= bh
->b_bdev
;
507 } while ((bh
= bh
->b_this_page
) != head
);
513 * Page has buffers, but they are all unmapped. The page was
514 * created by pagein or read over a hole which was handled by
515 * block_read_full_folio(). If this address_space is also
516 * using mpage_readahead then this can rarely happen.
522 * The page has no buffers: map it to disk
524 BUG_ON(!PageUptodate(page
));
525 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
526 last_block
= (i_size
- 1) >> blkbits
;
527 map_bh
.b_page
= page
;
528 for (page_block
= 0; page_block
< blocks_per_page
; ) {
531 map_bh
.b_size
= 1 << blkbits
;
532 if (mpd
->get_block(inode
, block_in_file
, &map_bh
, 1))
534 if (buffer_new(&map_bh
))
535 clean_bdev_bh_alias(&map_bh
);
536 if (buffer_boundary(&map_bh
)) {
537 boundary_block
= map_bh
.b_blocknr
;
538 boundary_bdev
= map_bh
.b_bdev
;
541 if (map_bh
.b_blocknr
!= blocks
[page_block
-1] + 1)
544 blocks
[page_block
++] = map_bh
.b_blocknr
;
545 boundary
= buffer_boundary(&map_bh
);
546 bdev
= map_bh
.b_bdev
;
547 if (block_in_file
== last_block
)
551 BUG_ON(page_block
== 0);
553 first_unmapped
= page_block
;
556 end_index
= i_size
>> PAGE_SHIFT
;
557 if (page
->index
>= end_index
) {
559 * The page straddles i_size. It must be zeroed out on each
560 * and every writepage invocation because it may be mmapped.
561 * "A file is mapped in multiples of the page size. For a file
562 * that is not a multiple of the page size, the remaining memory
563 * is zeroed when mapped, and writes to that region are not
564 * written out to the file."
566 unsigned offset
= i_size
& (PAGE_SIZE
- 1);
568 if (page
->index
> end_index
|| !offset
)
570 zero_user_segment(page
, offset
, PAGE_SIZE
);
574 * This page will go to BIO. Do we need to send this BIO off first?
576 if (bio
&& mpd
->last_block_in_bio
!= blocks
[0] - 1)
577 bio
= mpage_bio_submit(bio
);
581 if (first_unmapped
== blocks_per_page
) {
582 if (!bdev_write_page(bdev
, blocks
[0] << (blkbits
- 9),
586 bio
= bio_alloc(bdev
, BIO_MAX_VECS
,
587 REQ_OP_WRITE
| wbc_to_write_flags(wbc
),
589 bio
->bi_iter
.bi_sector
= blocks
[0] << (blkbits
- 9);
590 wbc_init_bio(wbc
, bio
);
594 * Must try to add the page before marking the buffer clean or
595 * the confused fail path above (OOM) will be very confused when
596 * it finds all bh marked clean (i.e. it will not write anything)
598 wbc_account_cgroup_owner(wbc
, page
, PAGE_SIZE
);
599 length
= first_unmapped
<< blkbits
;
600 if (bio_add_page(bio
, page
, length
, 0) < length
) {
601 bio
= mpage_bio_submit(bio
);
605 clean_buffers(page
, first_unmapped
);
607 BUG_ON(PageWriteback(page
));
608 set_page_writeback(page
);
610 if (boundary
|| (first_unmapped
!= blocks_per_page
)) {
611 bio
= mpage_bio_submit(bio
);
612 if (boundary_block
) {
613 write_boundary_block(boundary_bdev
,
614 boundary_block
, 1 << blkbits
);
617 mpd
->last_block_in_bio
= blocks
[blocks_per_page
- 1];
623 bio
= mpage_bio_submit(bio
);
625 if (mpd
->use_writepage
) {
626 ret
= mapping
->a_ops
->writepage(page
, wbc
);
632 * The caller has a ref on the inode, so *mapping is stable
634 mapping_set_error(mapping
, ret
);
641 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
642 * @mapping: address space structure to write
643 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
644 * @get_block: the filesystem's block mapper function.
645 * If this is NULL then use a_ops->writepage. Otherwise, go
648 * This is a library function, which implements the writepages()
649 * address_space_operation.
651 * If a page is already under I/O, generic_writepages() skips it, even
652 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
653 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
654 * and msync() need to guarantee that all the data which was dirty at the time
655 * the call was made get new I/O started against them. If wbc->sync_mode is
656 * WB_SYNC_ALL then we were called for data integrity and we must wait for
657 * existing IO to complete.
660 mpage_writepages(struct address_space
*mapping
,
661 struct writeback_control
*wbc
, get_block_t get_block
)
663 struct blk_plug plug
;
666 blk_start_plug(&plug
);
669 ret
= generic_writepages(mapping
, wbc
);
671 struct mpage_data mpd
= {
673 .last_block_in_bio
= 0,
674 .get_block
= get_block
,
678 ret
= write_cache_pages(mapping
, wbc
, __mpage_writepage
, &mpd
);
680 mpage_bio_submit(mpd
.bio
);
682 blk_finish_plug(&plug
);
685 EXPORT_SYMBOL(mpage_writepages
);
687 int mpage_writepage(struct page
*page
, get_block_t get_block
,
688 struct writeback_control
*wbc
)
690 struct mpage_data mpd
= {
692 .last_block_in_bio
= 0,
693 .get_block
= get_block
,
696 int ret
= __mpage_writepage(page
, wbc
, &mpd
);
698 mpage_bio_submit(mpd
.bio
);
701 EXPORT_SYMBOL(mpage_writepage
);