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
)
50 struct bvec_iter_all iter_all
;
52 bio_for_each_segment_all(bv
, bio
, iter_all
) {
53 struct page
*page
= bv
->bv_page
;
54 page_endio(page
, bio_op(bio
),
55 blk_status_to_errno(bio
->bi_status
));
61 static struct bio
*mpage_bio_submit(int op
, int op_flags
, struct bio
*bio
)
63 bio
->bi_end_io
= mpage_end_io
;
64 bio_set_op_attrs(bio
, op
, op_flags
);
71 mpage_alloc(struct block_device
*bdev
,
72 sector_t first_sector
, int nr_vecs
,
77 /* Restrict the given (page cache) mask for slab allocations */
78 gfp_flags
&= GFP_KERNEL
;
79 bio
= bio_alloc(gfp_flags
, nr_vecs
);
81 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
82 while (!bio
&& (nr_vecs
/= 2))
83 bio
= bio_alloc(gfp_flags
, nr_vecs
);
87 bio_set_dev(bio
, bdev
);
88 bio
->bi_iter
.bi_sector
= first_sector
;
94 * support function for mpage_readahead. The fs supplied get_block might
95 * return an up to date buffer. This is used to map that buffer into
96 * the page, which allows readpage to avoid triggering a duplicate call
99 * The idea is to avoid adding buffers to pages that don't already have
100 * them. So when the buffer is up to date and the page size == block size,
101 * this marks the page up to date instead of adding new buffers.
104 map_buffer_to_page(struct page
*page
, struct buffer_head
*bh
, int page_block
)
106 struct inode
*inode
= page
->mapping
->host
;
107 struct buffer_head
*page_bh
, *head
;
110 if (!page_has_buffers(page
)) {
112 * don't make any buffers if there is only one buffer on
113 * the page and the page just needs to be set up to date
115 if (inode
->i_blkbits
== PAGE_SHIFT
&&
116 buffer_uptodate(bh
)) {
117 SetPageUptodate(page
);
120 create_empty_buffers(page
, i_blocksize(inode
), 0);
122 head
= page_buffers(page
);
125 if (block
== page_block
) {
126 page_bh
->b_state
= bh
->b_state
;
127 page_bh
->b_bdev
= bh
->b_bdev
;
128 page_bh
->b_blocknr
= bh
->b_blocknr
;
131 page_bh
= page_bh
->b_this_page
;
133 } while (page_bh
!= head
);
136 struct mpage_readpage_args
{
139 unsigned int nr_pages
;
141 sector_t last_block_in_bio
;
142 struct buffer_head map_bh
;
143 unsigned long first_logical_block
;
144 get_block_t
*get_block
;
148 * This is the worker routine which does all the work of mapping the disk
149 * blocks and constructs largest possible bios, submits them for IO if the
150 * blocks are not contiguous on the disk.
152 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
153 * represent the validity of its disk mapping and to decide when to do the next
156 static struct bio
*do_mpage_readpage(struct mpage_readpage_args
*args
)
158 struct page
*page
= args
->page
;
159 struct inode
*inode
= page
->mapping
->host
;
160 const unsigned blkbits
= inode
->i_blkbits
;
161 const unsigned blocks_per_page
= PAGE_SIZE
>> blkbits
;
162 const unsigned blocksize
= 1 << blkbits
;
163 struct buffer_head
*map_bh
= &args
->map_bh
;
164 sector_t block_in_file
;
166 sector_t last_block_in_file
;
167 sector_t blocks
[MAX_BUF_PER_PAGE
];
169 unsigned first_hole
= blocks_per_page
;
170 struct block_device
*bdev
= NULL
;
172 int fully_mapped
= 1;
175 unsigned relative_block
;
178 if (args
->is_readahead
) {
179 op_flags
= REQ_RAHEAD
;
180 gfp
= readahead_gfp_mask(page
->mapping
);
183 gfp
= mapping_gfp_constraint(page
->mapping
, GFP_KERNEL
);
186 if (page_has_buffers(page
))
189 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
190 last_block
= block_in_file
+ args
->nr_pages
* blocks_per_page
;
191 last_block_in_file
= (i_size_read(inode
) + blocksize
- 1) >> blkbits
;
192 if (last_block
> last_block_in_file
)
193 last_block
= last_block_in_file
;
197 * Map blocks using the result from the previous get_blocks call first.
199 nblocks
= map_bh
->b_size
>> blkbits
;
200 if (buffer_mapped(map_bh
) &&
201 block_in_file
> args
->first_logical_block
&&
202 block_in_file
< (args
->first_logical_block
+ nblocks
)) {
203 unsigned map_offset
= block_in_file
- args
->first_logical_block
;
204 unsigned last
= nblocks
- map_offset
;
206 for (relative_block
= 0; ; relative_block
++) {
207 if (relative_block
== last
) {
208 clear_buffer_mapped(map_bh
);
211 if (page_block
== blocks_per_page
)
213 blocks
[page_block
] = map_bh
->b_blocknr
+ map_offset
+
218 bdev
= map_bh
->b_bdev
;
222 * Then do more get_blocks calls until we are done with this page.
224 map_bh
->b_page
= page
;
225 while (page_block
< blocks_per_page
) {
229 if (block_in_file
< last_block
) {
230 map_bh
->b_size
= (last_block
-block_in_file
) << blkbits
;
231 if (args
->get_block(inode
, block_in_file
, map_bh
, 0))
233 args
->first_logical_block
= block_in_file
;
236 if (!buffer_mapped(map_bh
)) {
238 if (first_hole
== blocks_per_page
)
239 first_hole
= page_block
;
245 /* some filesystems will copy data into the page during
246 * the get_block call, in which case we don't want to
247 * read it again. map_buffer_to_page copies the data
248 * we just collected from get_block into the page's buffers
249 * so readpage doesn't have to repeat the get_block call
251 if (buffer_uptodate(map_bh
)) {
252 map_buffer_to_page(page
, map_bh
, page_block
);
256 if (first_hole
!= blocks_per_page
)
257 goto confused
; /* hole -> non-hole */
259 /* Contiguous blocks? */
260 if (page_block
&& blocks
[page_block
-1] != map_bh
->b_blocknr
-1)
262 nblocks
= map_bh
->b_size
>> blkbits
;
263 for (relative_block
= 0; ; relative_block
++) {
264 if (relative_block
== nblocks
) {
265 clear_buffer_mapped(map_bh
);
267 } else if (page_block
== blocks_per_page
)
269 blocks
[page_block
] = map_bh
->b_blocknr
+relative_block
;
273 bdev
= map_bh
->b_bdev
;
276 if (first_hole
!= blocks_per_page
) {
277 zero_user_segment(page
, first_hole
<< blkbits
, PAGE_SIZE
);
278 if (first_hole
== 0) {
279 SetPageUptodate(page
);
283 } else if (fully_mapped
) {
284 SetPageMappedToDisk(page
);
287 if (fully_mapped
&& blocks_per_page
== 1 && !PageUptodate(page
) &&
288 cleancache_get_page(page
) == 0) {
289 SetPageUptodate(page
);
294 * This page will go to BIO. Do we need to send this BIO off first?
296 if (args
->bio
&& (args
->last_block_in_bio
!= blocks
[0] - 1))
297 args
->bio
= mpage_bio_submit(REQ_OP_READ
, op_flags
, args
->bio
);
300 if (args
->bio
== NULL
) {
301 if (first_hole
== blocks_per_page
) {
302 if (!bdev_read_page(bdev
, blocks
[0] << (blkbits
- 9),
306 args
->bio
= mpage_alloc(bdev
, blocks
[0] << (blkbits
- 9),
307 bio_max_segs(args
->nr_pages
), gfp
);
308 if (args
->bio
== NULL
)
312 length
= first_hole
<< blkbits
;
313 if (bio_add_page(args
->bio
, page
, length
, 0) < length
) {
314 args
->bio
= mpage_bio_submit(REQ_OP_READ
, op_flags
, args
->bio
);
318 relative_block
= block_in_file
- args
->first_logical_block
;
319 nblocks
= map_bh
->b_size
>> blkbits
;
320 if ((buffer_boundary(map_bh
) && relative_block
== nblocks
) ||
321 (first_hole
!= blocks_per_page
))
322 args
->bio
= mpage_bio_submit(REQ_OP_READ
, op_flags
, args
->bio
);
324 args
->last_block_in_bio
= blocks
[blocks_per_page
- 1];
330 args
->bio
= mpage_bio_submit(REQ_OP_READ
, op_flags
, args
->bio
);
331 if (!PageUptodate(page
))
332 block_read_full_page(page
, args
->get_block
);
339 * mpage_readahead - start reads against pages
340 * @rac: Describes which pages to read.
341 * @get_block: The filesystem's block mapper function.
343 * This function walks the pages and the blocks within each page, building and
344 * emitting large BIOs.
346 * If anything unusual happens, such as:
348 * - encountering a page which has buffers
349 * - encountering a page which has a non-hole after a hole
350 * - encountering a page with non-contiguous blocks
352 * then this code just gives up and calls the buffer_head-based read function.
353 * It does handle a page which has holes at the end - that is a common case:
354 * the end-of-file on blocksize < PAGE_SIZE setups.
356 * BH_Boundary explanation:
358 * There is a problem. The mpage read code assembles several pages, gets all
359 * their disk mappings, and then submits them all. That's fine, but obtaining
360 * the disk mappings may require I/O. Reads of indirect blocks, for example.
362 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
363 * submitted in the following order:
365 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
367 * because the indirect block has to be read to get the mappings of blocks
368 * 13,14,15,16. Obviously, this impacts performance.
370 * So what we do it to allow the filesystem's get_block() function to set
371 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
372 * after this one will require I/O against a block which is probably close to
373 * this one. So you should push what I/O you have currently accumulated.
375 * This all causes the disk requests to be issued in the correct order.
377 void mpage_readahead(struct readahead_control
*rac
, get_block_t get_block
)
380 struct mpage_readpage_args args
= {
381 .get_block
= get_block
,
382 .is_readahead
= true,
385 while ((page
= readahead_page(rac
))) {
386 prefetchw(&page
->flags
);
388 args
.nr_pages
= readahead_count(rac
);
389 args
.bio
= do_mpage_readpage(&args
);
393 mpage_bio_submit(REQ_OP_READ
, REQ_RAHEAD
, args
.bio
);
395 EXPORT_SYMBOL(mpage_readahead
);
398 * This isn't called much at all
400 int mpage_readpage(struct page
*page
, get_block_t get_block
)
402 struct mpage_readpage_args args
= {
405 .get_block
= get_block
,
408 args
.bio
= do_mpage_readpage(&args
);
410 mpage_bio_submit(REQ_OP_READ
, 0, args
.bio
);
413 EXPORT_SYMBOL(mpage_readpage
);
416 * Writing is not so simple.
418 * If the page has buffers then they will be used for obtaining the disk
419 * mapping. We only support pages which are fully mapped-and-dirty, with a
420 * special case for pages which are unmapped at the end: end-of-file.
422 * If the page has no buffers (preferred) then the page is mapped here.
424 * If all blocks are found to be contiguous then the page can go into the
425 * BIO. Otherwise fall back to the mapping's writepage().
427 * FIXME: This code wants an estimate of how many pages are still to be
428 * written, so it can intelligently allocate a suitably-sized BIO. For now,
429 * just allocate full-size (16-page) BIOs.
434 sector_t last_block_in_bio
;
435 get_block_t
*get_block
;
436 unsigned use_writepage
;
440 * We have our BIO, so we can now mark the buffers clean. Make
441 * sure to only clean buffers which we know we'll be writing.
443 static void clean_buffers(struct page
*page
, unsigned first_unmapped
)
445 unsigned buffer_counter
= 0;
446 struct buffer_head
*bh
, *head
;
447 if (!page_has_buffers(page
))
449 head
= page_buffers(page
);
453 if (buffer_counter
++ == first_unmapped
)
455 clear_buffer_dirty(bh
);
456 bh
= bh
->b_this_page
;
457 } while (bh
!= head
);
460 * we cannot drop the bh if the page is not uptodate or a concurrent
461 * readpage would fail to serialize with the bh and it would read from
462 * disk before we reach the platter.
464 if (buffer_heads_over_limit
&& PageUptodate(page
))
465 try_to_free_buffers(page
);
469 * For situations where we want to clean all buffers attached to a page.
470 * We don't need to calculate how many buffers are attached to the page,
471 * we just need to specify a number larger than the maximum number of buffers.
473 void clean_page_buffers(struct page
*page
)
475 clean_buffers(page
, ~0U);
478 static int __mpage_writepage(struct page
*page
, struct writeback_control
*wbc
,
481 struct mpage_data
*mpd
= data
;
482 struct bio
*bio
= mpd
->bio
;
483 struct address_space
*mapping
= page
->mapping
;
484 struct inode
*inode
= page
->mapping
->host
;
485 const unsigned blkbits
= inode
->i_blkbits
;
486 unsigned long end_index
;
487 const unsigned blocks_per_page
= PAGE_SIZE
>> blkbits
;
489 sector_t block_in_file
;
490 sector_t blocks
[MAX_BUF_PER_PAGE
];
492 unsigned first_unmapped
= blocks_per_page
;
493 struct block_device
*bdev
= NULL
;
495 sector_t boundary_block
= 0;
496 struct block_device
*boundary_bdev
= NULL
;
498 struct buffer_head map_bh
;
499 loff_t i_size
= i_size_read(inode
);
501 int op_flags
= wbc_to_write_flags(wbc
);
503 if (page_has_buffers(page
)) {
504 struct buffer_head
*head
= page_buffers(page
);
505 struct buffer_head
*bh
= head
;
507 /* If they're all mapped and dirty, do it */
510 BUG_ON(buffer_locked(bh
));
511 if (!buffer_mapped(bh
)) {
513 * unmapped dirty buffers are created by
514 * __set_page_dirty_buffers -> mmapped data
516 if (buffer_dirty(bh
))
518 if (first_unmapped
== blocks_per_page
)
519 first_unmapped
= page_block
;
523 if (first_unmapped
!= blocks_per_page
)
524 goto confused
; /* hole -> non-hole */
526 if (!buffer_dirty(bh
) || !buffer_uptodate(bh
))
529 if (bh
->b_blocknr
!= blocks
[page_block
-1] + 1)
532 blocks
[page_block
++] = bh
->b_blocknr
;
533 boundary
= buffer_boundary(bh
);
535 boundary_block
= bh
->b_blocknr
;
536 boundary_bdev
= bh
->b_bdev
;
539 } while ((bh
= bh
->b_this_page
) != head
);
545 * Page has buffers, but they are all unmapped. The page was
546 * created by pagein or read over a hole which was handled by
547 * block_read_full_page(). If this address_space is also
548 * using mpage_readahead then this can rarely happen.
554 * The page has no buffers: map it to disk
556 BUG_ON(!PageUptodate(page
));
557 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
558 last_block
= (i_size
- 1) >> blkbits
;
559 map_bh
.b_page
= page
;
560 for (page_block
= 0; page_block
< blocks_per_page
; ) {
563 map_bh
.b_size
= 1 << blkbits
;
564 if (mpd
->get_block(inode
, block_in_file
, &map_bh
, 1))
566 if (buffer_new(&map_bh
))
567 clean_bdev_bh_alias(&map_bh
);
568 if (buffer_boundary(&map_bh
)) {
569 boundary_block
= map_bh
.b_blocknr
;
570 boundary_bdev
= map_bh
.b_bdev
;
573 if (map_bh
.b_blocknr
!= blocks
[page_block
-1] + 1)
576 blocks
[page_block
++] = map_bh
.b_blocknr
;
577 boundary
= buffer_boundary(&map_bh
);
578 bdev
= map_bh
.b_bdev
;
579 if (block_in_file
== last_block
)
583 BUG_ON(page_block
== 0);
585 first_unmapped
= page_block
;
588 end_index
= i_size
>> PAGE_SHIFT
;
589 if (page
->index
>= end_index
) {
591 * The page straddles i_size. It must be zeroed out on each
592 * and every writepage invocation because it may be mmapped.
593 * "A file is mapped in multiples of the page size. For a file
594 * that is not a multiple of the page size, the remaining memory
595 * is zeroed when mapped, and writes to that region are not
596 * written out to the file."
598 unsigned offset
= i_size
& (PAGE_SIZE
- 1);
600 if (page
->index
> end_index
|| !offset
)
602 zero_user_segment(page
, offset
, PAGE_SIZE
);
606 * This page will go to BIO. Do we need to send this BIO off first?
608 if (bio
&& mpd
->last_block_in_bio
!= blocks
[0] - 1)
609 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
613 if (first_unmapped
== blocks_per_page
) {
614 if (!bdev_write_page(bdev
, blocks
[0] << (blkbits
- 9),
618 bio
= mpage_alloc(bdev
, blocks
[0] << (blkbits
- 9),
619 BIO_MAX_VECS
, GFP_NOFS
|__GFP_HIGH
);
623 wbc_init_bio(wbc
, bio
);
624 bio
->bi_write_hint
= inode
->i_write_hint
;
628 * Must try to add the page before marking the buffer clean or
629 * the confused fail path above (OOM) will be very confused when
630 * it finds all bh marked clean (i.e. it will not write anything)
632 wbc_account_cgroup_owner(wbc
, page
, PAGE_SIZE
);
633 length
= first_unmapped
<< blkbits
;
634 if (bio_add_page(bio
, page
, length
, 0) < length
) {
635 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
639 clean_buffers(page
, first_unmapped
);
641 BUG_ON(PageWriteback(page
));
642 set_page_writeback(page
);
644 if (boundary
|| (first_unmapped
!= blocks_per_page
)) {
645 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
646 if (boundary_block
) {
647 write_boundary_block(boundary_bdev
,
648 boundary_block
, 1 << blkbits
);
651 mpd
->last_block_in_bio
= blocks
[blocks_per_page
- 1];
657 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
659 if (mpd
->use_writepage
) {
660 ret
= mapping
->a_ops
->writepage(page
, wbc
);
666 * The caller has a ref on the inode, so *mapping is stable
668 mapping_set_error(mapping
, ret
);
675 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
676 * @mapping: address space structure to write
677 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
678 * @get_block: the filesystem's block mapper function.
679 * If this is NULL then use a_ops->writepage. Otherwise, go
682 * This is a library function, which implements the writepages()
683 * address_space_operation.
685 * If a page is already under I/O, generic_writepages() skips it, even
686 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
687 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
688 * and msync() need to guarantee that all the data which was dirty at the time
689 * the call was made get new I/O started against them. If wbc->sync_mode is
690 * WB_SYNC_ALL then we were called for data integrity and we must wait for
691 * existing IO to complete.
694 mpage_writepages(struct address_space
*mapping
,
695 struct writeback_control
*wbc
, get_block_t get_block
)
697 struct blk_plug plug
;
700 blk_start_plug(&plug
);
703 ret
= generic_writepages(mapping
, wbc
);
705 struct mpage_data mpd
= {
707 .last_block_in_bio
= 0,
708 .get_block
= get_block
,
712 ret
= write_cache_pages(mapping
, wbc
, __mpage_writepage
, &mpd
);
714 int op_flags
= (wbc
->sync_mode
== WB_SYNC_ALL
?
716 mpage_bio_submit(REQ_OP_WRITE
, op_flags
, mpd
.bio
);
719 blk_finish_plug(&plug
);
722 EXPORT_SYMBOL(mpage_writepages
);
724 int mpage_writepage(struct page
*page
, get_block_t get_block
,
725 struct writeback_control
*wbc
)
727 struct mpage_data mpd
= {
729 .last_block_in_bio
= 0,
730 .get_block
= get_block
,
733 int ret
= __mpage_writepage(page
, wbc
, &mpd
);
735 int op_flags
= (wbc
->sync_mode
== WB_SYNC_ALL
?
737 mpage_bio_submit(REQ_OP_WRITE
, op_flags
, mpd
.bio
);
741 EXPORT_SYMBOL(mpage_writepage
);