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>
32 #include <linux/cleancache.h>
36 * I/O completion handler for multipage BIOs.
38 * The mpage code never puts partial pages into a BIO (except for end-of-file).
39 * If a page does not map to a contiguous run of blocks then it simply falls
40 * back to block_read_full_page().
42 * Why is this? If a page's completion depends on a number of different BIOs
43 * which can complete in any order (or at the same time) then determining the
44 * status of that page is hard. See end_buffer_async_read() for the details.
45 * There is no point in duplicating all that complexity.
47 static void mpage_end_io(struct bio
*bio
)
51 struct bvec_iter_all iter_all
;
53 bio_for_each_segment_all(bv
, bio
, i
, iter_all
) {
54 struct page
*page
= bv
->bv_page
;
55 page_endio(page
, bio_op(bio
),
56 blk_status_to_errno(bio
->bi_status
));
62 static struct bio
*mpage_bio_submit(int op
, int op_flags
, struct bio
*bio
)
64 bio
->bi_end_io
= mpage_end_io
;
65 bio_set_op_attrs(bio
, op
, op_flags
);
66 guard_bio_eod(op
, bio
);
72 mpage_alloc(struct block_device
*bdev
,
73 sector_t first_sector
, int nr_vecs
,
78 /* Restrict the given (page cache) mask for slab allocations */
79 gfp_flags
&= GFP_KERNEL
;
80 bio
= bio_alloc(gfp_flags
, nr_vecs
);
82 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
83 while (!bio
&& (nr_vecs
/= 2))
84 bio
= bio_alloc(gfp_flags
, nr_vecs
);
88 bio_set_dev(bio
, bdev
);
89 bio
->bi_iter
.bi_sector
= first_sector
;
95 * support function for mpage_readpages. The fs supplied get_block might
96 * return an up to date buffer. This is used to map that buffer into
97 * the page, which allows readpage to avoid triggering a duplicate call
100 * The idea is to avoid adding buffers to pages that don't already have
101 * them. So when the buffer is up to date and the page size == block size,
102 * this marks the page up to date instead of adding new buffers.
105 map_buffer_to_page(struct page
*page
, struct buffer_head
*bh
, int page_block
)
107 struct inode
*inode
= page
->mapping
->host
;
108 struct buffer_head
*page_bh
, *head
;
111 if (!page_has_buffers(page
)) {
113 * don't make any buffers if there is only one buffer on
114 * the page and the page just needs to be set up to date
116 if (inode
->i_blkbits
== PAGE_SHIFT
&&
117 buffer_uptodate(bh
)) {
118 SetPageUptodate(page
);
121 create_empty_buffers(page
, i_blocksize(inode
), 0);
123 head
= page_buffers(page
);
126 if (block
== page_block
) {
127 page_bh
->b_state
= bh
->b_state
;
128 page_bh
->b_bdev
= bh
->b_bdev
;
129 page_bh
->b_blocknr
= bh
->b_blocknr
;
132 page_bh
= page_bh
->b_this_page
;
134 } while (page_bh
!= head
);
137 struct mpage_readpage_args
{
140 unsigned int nr_pages
;
142 sector_t last_block_in_bio
;
143 struct buffer_head map_bh
;
144 unsigned long first_logical_block
;
145 get_block_t
*get_block
;
149 * This is the worker routine which does all the work of mapping the disk
150 * blocks and constructs largest possible bios, submits them for IO if the
151 * blocks are not contiguous on the disk.
153 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
154 * represent the validity of its disk mapping and to decide when to do the next
157 static struct bio
*do_mpage_readpage(struct mpage_readpage_args
*args
)
159 struct page
*page
= args
->page
;
160 struct inode
*inode
= page
->mapping
->host
;
161 const unsigned blkbits
= inode
->i_blkbits
;
162 const unsigned blocks_per_page
= PAGE_SIZE
>> blkbits
;
163 const unsigned blocksize
= 1 << blkbits
;
164 struct buffer_head
*map_bh
= &args
->map_bh
;
165 sector_t block_in_file
;
167 sector_t last_block_in_file
;
168 sector_t blocks
[MAX_BUF_PER_PAGE
];
170 unsigned first_hole
= blocks_per_page
;
171 struct block_device
*bdev
= NULL
;
173 int fully_mapped
= 1;
176 unsigned relative_block
;
179 if (args
->is_readahead
) {
180 op_flags
= REQ_RAHEAD
;
181 gfp
= readahead_gfp_mask(page
->mapping
);
184 gfp
= mapping_gfp_constraint(page
->mapping
, GFP_KERNEL
);
187 if (page_has_buffers(page
))
190 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
191 last_block
= block_in_file
+ args
->nr_pages
* blocks_per_page
;
192 last_block_in_file
= (i_size_read(inode
) + blocksize
- 1) >> blkbits
;
193 if (last_block
> last_block_in_file
)
194 last_block
= last_block_in_file
;
198 * Map blocks using the result from the previous get_blocks call first.
200 nblocks
= map_bh
->b_size
>> blkbits
;
201 if (buffer_mapped(map_bh
) &&
202 block_in_file
> args
->first_logical_block
&&
203 block_in_file
< (args
->first_logical_block
+ nblocks
)) {
204 unsigned map_offset
= block_in_file
- args
->first_logical_block
;
205 unsigned last
= nblocks
- map_offset
;
207 for (relative_block
= 0; ; relative_block
++) {
208 if (relative_block
== last
) {
209 clear_buffer_mapped(map_bh
);
212 if (page_block
== blocks_per_page
)
214 blocks
[page_block
] = map_bh
->b_blocknr
+ map_offset
+
219 bdev
= map_bh
->b_bdev
;
223 * Then do more get_blocks calls until we are done with this page.
225 map_bh
->b_page
= page
;
226 while (page_block
< blocks_per_page
) {
230 if (block_in_file
< last_block
) {
231 map_bh
->b_size
= (last_block
-block_in_file
) << blkbits
;
232 if (args
->get_block(inode
, block_in_file
, map_bh
, 0))
234 args
->first_logical_block
= block_in_file
;
237 if (!buffer_mapped(map_bh
)) {
239 if (first_hole
== blocks_per_page
)
240 first_hole
= page_block
;
246 /* some filesystems will copy data into the page during
247 * the get_block call, in which case we don't want to
248 * read it again. map_buffer_to_page copies the data
249 * we just collected from get_block into the page's buffers
250 * so readpage doesn't have to repeat the get_block call
252 if (buffer_uptodate(map_bh
)) {
253 map_buffer_to_page(page
, map_bh
, page_block
);
257 if (first_hole
!= blocks_per_page
)
258 goto confused
; /* hole -> non-hole */
260 /* Contiguous blocks? */
261 if (page_block
&& blocks
[page_block
-1] != map_bh
->b_blocknr
-1)
263 nblocks
= map_bh
->b_size
>> blkbits
;
264 for (relative_block
= 0; ; relative_block
++) {
265 if (relative_block
== nblocks
) {
266 clear_buffer_mapped(map_bh
);
268 } else if (page_block
== blocks_per_page
)
270 blocks
[page_block
] = map_bh
->b_blocknr
+relative_block
;
274 bdev
= map_bh
->b_bdev
;
277 if (first_hole
!= blocks_per_page
) {
278 zero_user_segment(page
, first_hole
<< blkbits
, PAGE_SIZE
);
279 if (first_hole
== 0) {
280 SetPageUptodate(page
);
284 } else if (fully_mapped
) {
285 SetPageMappedToDisk(page
);
288 if (fully_mapped
&& blocks_per_page
== 1 && !PageUptodate(page
) &&
289 cleancache_get_page(page
) == 0) {
290 SetPageUptodate(page
);
295 * This page will go to BIO. Do we need to send this BIO off first?
297 if (args
->bio
&& (args
->last_block_in_bio
!= blocks
[0] - 1))
298 args
->bio
= mpage_bio_submit(REQ_OP_READ
, op_flags
, args
->bio
);
301 if (args
->bio
== NULL
) {
302 if (first_hole
== blocks_per_page
) {
303 if (!bdev_read_page(bdev
, blocks
[0] << (blkbits
- 9),
307 args
->bio
= mpage_alloc(bdev
, blocks
[0] << (blkbits
- 9),
308 min_t(int, args
->nr_pages
,
311 if (args
->bio
== NULL
)
315 length
= first_hole
<< blkbits
;
316 if (bio_add_page(args
->bio
, page
, length
, 0) < length
) {
317 args
->bio
= mpage_bio_submit(REQ_OP_READ
, op_flags
, args
->bio
);
321 relative_block
= block_in_file
- args
->first_logical_block
;
322 nblocks
= map_bh
->b_size
>> blkbits
;
323 if ((buffer_boundary(map_bh
) && relative_block
== nblocks
) ||
324 (first_hole
!= blocks_per_page
))
325 args
->bio
= mpage_bio_submit(REQ_OP_READ
, op_flags
, args
->bio
);
327 args
->last_block_in_bio
= blocks
[blocks_per_page
- 1];
333 args
->bio
= mpage_bio_submit(REQ_OP_READ
, op_flags
, args
->bio
);
334 if (!PageUptodate(page
))
335 block_read_full_page(page
, args
->get_block
);
342 * mpage_readpages - populate an address space with some pages & start reads against them
343 * @mapping: the address_space
344 * @pages: The address of a list_head which contains the target pages. These
345 * pages have their ->index populated and are otherwise uninitialised.
346 * The page at @pages->prev has the lowest file offset, and reads should be
347 * issued in @pages->prev to @pages->next order.
348 * @nr_pages: The number of pages at *@pages
349 * @get_block: The filesystem's block mapper function.
351 * This function walks the pages and the blocks within each page, building and
352 * emitting large BIOs.
354 * If anything unusual happens, such as:
356 * - encountering a page which has buffers
357 * - encountering a page which has a non-hole after a hole
358 * - encountering a page with non-contiguous blocks
360 * then this code just gives up and calls the buffer_head-based read function.
361 * It does handle a page which has holes at the end - that is a common case:
362 * the end-of-file on blocksize < PAGE_SIZE setups.
364 * BH_Boundary explanation:
366 * There is a problem. The mpage read code assembles several pages, gets all
367 * their disk mappings, and then submits them all. That's fine, but obtaining
368 * the disk mappings may require I/O. Reads of indirect blocks, for example.
370 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
371 * submitted in the following order:
373 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
375 * because the indirect block has to be read to get the mappings of blocks
376 * 13,14,15,16. Obviously, this impacts performance.
378 * So what we do it to allow the filesystem's get_block() function to set
379 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
380 * after this one will require I/O against a block which is probably close to
381 * this one. So you should push what I/O you have currently accumulated.
383 * This all causes the disk requests to be issued in the correct order.
386 mpage_readpages(struct address_space
*mapping
, struct list_head
*pages
,
387 unsigned nr_pages
, get_block_t get_block
)
389 struct mpage_readpage_args args
= {
390 .get_block
= get_block
,
391 .is_readahead
= true,
395 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
396 struct page
*page
= lru_to_page(pages
);
398 prefetchw(&page
->flags
);
399 list_del(&page
->lru
);
400 if (!add_to_page_cache_lru(page
, mapping
,
402 readahead_gfp_mask(mapping
))) {
404 args
.nr_pages
= nr_pages
- page_idx
;
405 args
.bio
= do_mpage_readpage(&args
);
409 BUG_ON(!list_empty(pages
));
411 mpage_bio_submit(REQ_OP_READ
, REQ_RAHEAD
, args
.bio
);
414 EXPORT_SYMBOL(mpage_readpages
);
417 * This isn't called much at all
419 int mpage_readpage(struct page
*page
, get_block_t get_block
)
421 struct mpage_readpage_args args
= {
424 .get_block
= get_block
,
427 args
.bio
= do_mpage_readpage(&args
);
429 mpage_bio_submit(REQ_OP_READ
, 0, args
.bio
);
432 EXPORT_SYMBOL(mpage_readpage
);
435 * Writing is not so simple.
437 * If the page has buffers then they will be used for obtaining the disk
438 * mapping. We only support pages which are fully mapped-and-dirty, with a
439 * special case for pages which are unmapped at the end: end-of-file.
441 * If the page has no buffers (preferred) then the page is mapped here.
443 * If all blocks are found to be contiguous then the page can go into the
444 * BIO. Otherwise fall back to the mapping's writepage().
446 * FIXME: This code wants an estimate of how many pages are still to be
447 * written, so it can intelligently allocate a suitably-sized BIO. For now,
448 * just allocate full-size (16-page) BIOs.
453 sector_t last_block_in_bio
;
454 get_block_t
*get_block
;
455 unsigned use_writepage
;
459 * We have our BIO, so we can now mark the buffers clean. Make
460 * sure to only clean buffers which we know we'll be writing.
462 static void clean_buffers(struct page
*page
, unsigned first_unmapped
)
464 unsigned buffer_counter
= 0;
465 struct buffer_head
*bh
, *head
;
466 if (!page_has_buffers(page
))
468 head
= page_buffers(page
);
472 if (buffer_counter
++ == first_unmapped
)
474 clear_buffer_dirty(bh
);
475 bh
= bh
->b_this_page
;
476 } while (bh
!= head
);
479 * we cannot drop the bh if the page is not uptodate or a concurrent
480 * readpage would fail to serialize with the bh and it would read from
481 * disk before we reach the platter.
483 if (buffer_heads_over_limit
&& PageUptodate(page
))
484 try_to_free_buffers(page
);
488 * For situations where we want to clean all buffers attached to a page.
489 * We don't need to calculate how many buffers are attached to the page,
490 * we just need to specify a number larger than the maximum number of buffers.
492 void clean_page_buffers(struct page
*page
)
494 clean_buffers(page
, ~0U);
497 static int __mpage_writepage(struct page
*page
, struct writeback_control
*wbc
,
500 struct mpage_data
*mpd
= data
;
501 struct bio
*bio
= mpd
->bio
;
502 struct address_space
*mapping
= page
->mapping
;
503 struct inode
*inode
= page
->mapping
->host
;
504 const unsigned blkbits
= inode
->i_blkbits
;
505 unsigned long end_index
;
506 const unsigned blocks_per_page
= PAGE_SIZE
>> blkbits
;
508 sector_t block_in_file
;
509 sector_t blocks
[MAX_BUF_PER_PAGE
];
511 unsigned first_unmapped
= blocks_per_page
;
512 struct block_device
*bdev
= NULL
;
514 sector_t boundary_block
= 0;
515 struct block_device
*boundary_bdev
= NULL
;
517 struct buffer_head map_bh
;
518 loff_t i_size
= i_size_read(inode
);
520 int op_flags
= wbc_to_write_flags(wbc
);
522 if (page_has_buffers(page
)) {
523 struct buffer_head
*head
= page_buffers(page
);
524 struct buffer_head
*bh
= head
;
526 /* If they're all mapped and dirty, do it */
529 BUG_ON(buffer_locked(bh
));
530 if (!buffer_mapped(bh
)) {
532 * unmapped dirty buffers are created by
533 * __set_page_dirty_buffers -> mmapped data
535 if (buffer_dirty(bh
))
537 if (first_unmapped
== blocks_per_page
)
538 first_unmapped
= page_block
;
542 if (first_unmapped
!= blocks_per_page
)
543 goto confused
; /* hole -> non-hole */
545 if (!buffer_dirty(bh
) || !buffer_uptodate(bh
))
548 if (bh
->b_blocknr
!= blocks
[page_block
-1] + 1)
551 blocks
[page_block
++] = bh
->b_blocknr
;
552 boundary
= buffer_boundary(bh
);
554 boundary_block
= bh
->b_blocknr
;
555 boundary_bdev
= bh
->b_bdev
;
558 } while ((bh
= bh
->b_this_page
) != head
);
564 * Page has buffers, but they are all unmapped. The page was
565 * created by pagein or read over a hole which was handled by
566 * block_read_full_page(). If this address_space is also
567 * using mpage_readpages then this can rarely happen.
573 * The page has no buffers: map it to disk
575 BUG_ON(!PageUptodate(page
));
576 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
577 last_block
= (i_size
- 1) >> blkbits
;
578 map_bh
.b_page
= page
;
579 for (page_block
= 0; page_block
< blocks_per_page
; ) {
582 map_bh
.b_size
= 1 << blkbits
;
583 if (mpd
->get_block(inode
, block_in_file
, &map_bh
, 1))
585 if (buffer_new(&map_bh
))
586 clean_bdev_bh_alias(&map_bh
);
587 if (buffer_boundary(&map_bh
)) {
588 boundary_block
= map_bh
.b_blocknr
;
589 boundary_bdev
= map_bh
.b_bdev
;
592 if (map_bh
.b_blocknr
!= blocks
[page_block
-1] + 1)
595 blocks
[page_block
++] = map_bh
.b_blocknr
;
596 boundary
= buffer_boundary(&map_bh
);
597 bdev
= map_bh
.b_bdev
;
598 if (block_in_file
== last_block
)
602 BUG_ON(page_block
== 0);
604 first_unmapped
= page_block
;
607 end_index
= i_size
>> PAGE_SHIFT
;
608 if (page
->index
>= end_index
) {
610 * The page straddles i_size. It must be zeroed out on each
611 * and every writepage invocation because it may be mmapped.
612 * "A file is mapped in multiples of the page size. For a file
613 * that is not a multiple of the page size, the remaining memory
614 * is zeroed when mapped, and writes to that region are not
615 * written out to the file."
617 unsigned offset
= i_size
& (PAGE_SIZE
- 1);
619 if (page
->index
> end_index
|| !offset
)
621 zero_user_segment(page
, offset
, PAGE_SIZE
);
625 * This page will go to BIO. Do we need to send this BIO off first?
627 if (bio
&& mpd
->last_block_in_bio
!= blocks
[0] - 1)
628 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
632 if (first_unmapped
== blocks_per_page
) {
633 if (!bdev_write_page(bdev
, blocks
[0] << (blkbits
- 9),
637 bio
= mpage_alloc(bdev
, blocks
[0] << (blkbits
- 9),
638 BIO_MAX_PAGES
, GFP_NOFS
|__GFP_HIGH
);
642 wbc_init_bio(wbc
, bio
);
643 bio
->bi_write_hint
= inode
->i_write_hint
;
647 * Must try to add the page before marking the buffer clean or
648 * the confused fail path above (OOM) will be very confused when
649 * it finds all bh marked clean (i.e. it will not write anything)
651 wbc_account_io(wbc
, page
, PAGE_SIZE
);
652 length
= first_unmapped
<< blkbits
;
653 if (bio_add_page(bio
, page
, length
, 0) < length
) {
654 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
658 clean_buffers(page
, first_unmapped
);
660 BUG_ON(PageWriteback(page
));
661 set_page_writeback(page
);
663 if (boundary
|| (first_unmapped
!= blocks_per_page
)) {
664 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
665 if (boundary_block
) {
666 write_boundary_block(boundary_bdev
,
667 boundary_block
, 1 << blkbits
);
670 mpd
->last_block_in_bio
= blocks
[blocks_per_page
- 1];
676 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
678 if (mpd
->use_writepage
) {
679 ret
= mapping
->a_ops
->writepage(page
, wbc
);
685 * The caller has a ref on the inode, so *mapping is stable
687 mapping_set_error(mapping
, ret
);
694 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
695 * @mapping: address space structure to write
696 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
697 * @get_block: the filesystem's block mapper function.
698 * If this is NULL then use a_ops->writepage. Otherwise, go
701 * This is a library function, which implements the writepages()
702 * address_space_operation.
704 * If a page is already under I/O, generic_writepages() skips it, even
705 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
706 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
707 * and msync() need to guarantee that all the data which was dirty at the time
708 * the call was made get new I/O started against them. If wbc->sync_mode is
709 * WB_SYNC_ALL then we were called for data integrity and we must wait for
710 * existing IO to complete.
713 mpage_writepages(struct address_space
*mapping
,
714 struct writeback_control
*wbc
, get_block_t get_block
)
716 struct blk_plug plug
;
719 blk_start_plug(&plug
);
722 ret
= generic_writepages(mapping
, wbc
);
724 struct mpage_data mpd
= {
726 .last_block_in_bio
= 0,
727 .get_block
= get_block
,
731 ret
= write_cache_pages(mapping
, wbc
, __mpage_writepage
, &mpd
);
733 int op_flags
= (wbc
->sync_mode
== WB_SYNC_ALL
?
735 mpage_bio_submit(REQ_OP_WRITE
, op_flags
, mpd
.bio
);
738 blk_finish_plug(&plug
);
741 EXPORT_SYMBOL(mpage_writepages
);
743 int mpage_writepage(struct page
*page
, get_block_t get_block
,
744 struct writeback_control
*wbc
)
746 struct mpage_data mpd
= {
748 .last_block_in_bio
= 0,
749 .get_block
= get_block
,
752 int ret
= __mpage_writepage(page
, wbc
, &mpd
);
754 int op_flags
= (wbc
->sync_mode
== WB_SYNC_ALL
?
756 mpage_bio_submit(REQ_OP_WRITE
, op_flags
, mpd
.bio
);
760 EXPORT_SYMBOL(mpage_writepage
);