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.
78 static void map_buffer_to_folio(struct folio
*folio
, struct buffer_head
*bh
,
81 struct inode
*inode
= folio
->mapping
->host
;
82 struct buffer_head
*page_bh
, *head
;
85 head
= folio_buffers(folio
);
88 * don't make any buffers if there is only one buffer on
89 * the folio and the folio just needs to be set up to date
91 if (inode
->i_blkbits
== PAGE_SHIFT
&&
92 buffer_uptodate(bh
)) {
93 folio_mark_uptodate(folio
);
96 create_empty_buffers(&folio
->page
, i_blocksize(inode
), 0);
97 head
= folio_buffers(folio
);
102 if (block
== page_block
) {
103 page_bh
->b_state
= bh
->b_state
;
104 page_bh
->b_bdev
= bh
->b_bdev
;
105 page_bh
->b_blocknr
= bh
->b_blocknr
;
108 page_bh
= page_bh
->b_this_page
;
110 } while (page_bh
!= head
);
113 struct mpage_readpage_args
{
116 unsigned int nr_pages
;
118 sector_t last_block_in_bio
;
119 struct buffer_head map_bh
;
120 unsigned long first_logical_block
;
121 get_block_t
*get_block
;
125 * This is the worker routine which does all the work of mapping the disk
126 * blocks and constructs largest possible bios, submits them for IO if the
127 * blocks are not contiguous on the disk.
129 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
130 * represent the validity of its disk mapping and to decide when to do the next
133 static struct bio
*do_mpage_readpage(struct mpage_readpage_args
*args
)
135 struct folio
*folio
= args
->folio
;
136 struct inode
*inode
= folio
->mapping
->host
;
137 const unsigned blkbits
= inode
->i_blkbits
;
138 const unsigned blocks_per_page
= PAGE_SIZE
>> blkbits
;
139 const unsigned blocksize
= 1 << blkbits
;
140 struct buffer_head
*map_bh
= &args
->map_bh
;
141 sector_t block_in_file
;
143 sector_t last_block_in_file
;
144 sector_t blocks
[MAX_BUF_PER_PAGE
];
146 unsigned first_hole
= blocks_per_page
;
147 struct block_device
*bdev
= NULL
;
149 int fully_mapped
= 1;
150 blk_opf_t opf
= REQ_OP_READ
;
152 unsigned relative_block
;
153 gfp_t gfp
= mapping_gfp_constraint(folio
->mapping
, GFP_KERNEL
);
155 /* MAX_BUF_PER_PAGE, for example */
156 VM_BUG_ON_FOLIO(folio_test_large(folio
), folio
);
158 if (args
->is_readahead
) {
160 gfp
|= __GFP_NORETRY
| __GFP_NOWARN
;
163 if (folio_buffers(folio
))
166 block_in_file
= (sector_t
)folio
->index
<< (PAGE_SHIFT
- blkbits
);
167 last_block
= block_in_file
+ args
->nr_pages
* blocks_per_page
;
168 last_block_in_file
= (i_size_read(inode
) + blocksize
- 1) >> blkbits
;
169 if (last_block
> last_block_in_file
)
170 last_block
= last_block_in_file
;
174 * Map blocks using the result from the previous get_blocks call first.
176 nblocks
= map_bh
->b_size
>> blkbits
;
177 if (buffer_mapped(map_bh
) &&
178 block_in_file
> args
->first_logical_block
&&
179 block_in_file
< (args
->first_logical_block
+ nblocks
)) {
180 unsigned map_offset
= block_in_file
- args
->first_logical_block
;
181 unsigned last
= nblocks
- map_offset
;
183 for (relative_block
= 0; ; relative_block
++) {
184 if (relative_block
== last
) {
185 clear_buffer_mapped(map_bh
);
188 if (page_block
== blocks_per_page
)
190 blocks
[page_block
] = map_bh
->b_blocknr
+ map_offset
+
195 bdev
= map_bh
->b_bdev
;
199 * Then do more get_blocks calls until we are done with this folio.
201 map_bh
->b_page
= &folio
->page
;
202 while (page_block
< blocks_per_page
) {
206 if (block_in_file
< last_block
) {
207 map_bh
->b_size
= (last_block
-block_in_file
) << blkbits
;
208 if (args
->get_block(inode
, block_in_file
, map_bh
, 0))
210 args
->first_logical_block
= block_in_file
;
213 if (!buffer_mapped(map_bh
)) {
215 if (first_hole
== blocks_per_page
)
216 first_hole
= page_block
;
222 /* some filesystems will copy data into the page during
223 * the get_block call, in which case we don't want to
224 * read it again. map_buffer_to_folio copies the data
225 * we just collected from get_block into the folio's buffers
226 * so read_folio doesn't have to repeat the get_block call
228 if (buffer_uptodate(map_bh
)) {
229 map_buffer_to_folio(folio
, map_bh
, page_block
);
233 if (first_hole
!= blocks_per_page
)
234 goto confused
; /* hole -> non-hole */
236 /* Contiguous blocks? */
237 if (page_block
&& blocks
[page_block
-1] != map_bh
->b_blocknr
-1)
239 nblocks
= map_bh
->b_size
>> blkbits
;
240 for (relative_block
= 0; ; relative_block
++) {
241 if (relative_block
== nblocks
) {
242 clear_buffer_mapped(map_bh
);
244 } else if (page_block
== blocks_per_page
)
246 blocks
[page_block
] = map_bh
->b_blocknr
+relative_block
;
250 bdev
= map_bh
->b_bdev
;
253 if (first_hole
!= blocks_per_page
) {
254 folio_zero_segment(folio
, first_hole
<< blkbits
, PAGE_SIZE
);
255 if (first_hole
== 0) {
256 folio_mark_uptodate(folio
);
260 } else if (fully_mapped
) {
261 folio_set_mappedtodisk(folio
);
265 * This folio will go to BIO. Do we need to send this BIO off first?
267 if (args
->bio
&& (args
->last_block_in_bio
!= blocks
[0] - 1))
268 args
->bio
= mpage_bio_submit(args
->bio
);
271 if (args
->bio
== NULL
) {
272 if (first_hole
== blocks_per_page
) {
273 if (!bdev_read_page(bdev
, blocks
[0] << (blkbits
- 9),
277 args
->bio
= bio_alloc(bdev
, bio_max_segs(args
->nr_pages
), opf
,
279 if (args
->bio
== NULL
)
281 args
->bio
->bi_iter
.bi_sector
= blocks
[0] << (blkbits
- 9);
284 length
= first_hole
<< blkbits
;
285 if (!bio_add_folio(args
->bio
, folio
, length
, 0)) {
286 args
->bio
= mpage_bio_submit(args
->bio
);
290 relative_block
= block_in_file
- args
->first_logical_block
;
291 nblocks
= map_bh
->b_size
>> blkbits
;
292 if ((buffer_boundary(map_bh
) && relative_block
== nblocks
) ||
293 (first_hole
!= blocks_per_page
))
294 args
->bio
= mpage_bio_submit(args
->bio
);
296 args
->last_block_in_bio
= blocks
[blocks_per_page
- 1];
302 args
->bio
= mpage_bio_submit(args
->bio
);
303 if (!folio_test_uptodate(folio
))
304 block_read_full_folio(folio
, args
->get_block
);
311 * mpage_readahead - start reads against pages
312 * @rac: Describes which pages to read.
313 * @get_block: The filesystem's block mapper function.
315 * This function walks the pages and the blocks within each page, building and
316 * emitting large BIOs.
318 * If anything unusual happens, such as:
320 * - encountering a page which has buffers
321 * - encountering a page which has a non-hole after a hole
322 * - encountering a page with non-contiguous blocks
324 * then this code just gives up and calls the buffer_head-based read function.
325 * It does handle a page which has holes at the end - that is a common case:
326 * the end-of-file on blocksize < PAGE_SIZE setups.
328 * BH_Boundary explanation:
330 * There is a problem. The mpage read code assembles several pages, gets all
331 * their disk mappings, and then submits them all. That's fine, but obtaining
332 * the disk mappings may require I/O. Reads of indirect blocks, for example.
334 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
335 * submitted in the following order:
337 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
339 * because the indirect block has to be read to get the mappings of blocks
340 * 13,14,15,16. Obviously, this impacts performance.
342 * So what we do it to allow the filesystem's get_block() function to set
343 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
344 * after this one will require I/O against a block which is probably close to
345 * this one. So you should push what I/O you have currently accumulated.
347 * This all causes the disk requests to be issued in the correct order.
349 void mpage_readahead(struct readahead_control
*rac
, get_block_t get_block
)
352 struct mpage_readpage_args args
= {
353 .get_block
= get_block
,
354 .is_readahead
= true,
357 while ((folio
= readahead_folio(rac
))) {
358 prefetchw(&folio
->flags
);
360 args
.nr_pages
= readahead_count(rac
);
361 args
.bio
= do_mpage_readpage(&args
);
364 mpage_bio_submit(args
.bio
);
366 EXPORT_SYMBOL(mpage_readahead
);
369 * This isn't called much at all
371 int mpage_read_folio(struct folio
*folio
, get_block_t get_block
)
373 struct mpage_readpage_args args
= {
376 .get_block
= get_block
,
379 args
.bio
= do_mpage_readpage(&args
);
381 mpage_bio_submit(args
.bio
);
384 EXPORT_SYMBOL(mpage_read_folio
);
387 * Writing is not so simple.
389 * If the page has buffers then they will be used for obtaining the disk
390 * mapping. We only support pages which are fully mapped-and-dirty, with a
391 * special case for pages which are unmapped at the end: end-of-file.
393 * If the page has no buffers (preferred) then the page is mapped here.
395 * If all blocks are found to be contiguous then the page can go into the
396 * BIO. Otherwise fall back to the mapping's writepage().
398 * FIXME: This code wants an estimate of how many pages are still to be
399 * written, so it can intelligently allocate a suitably-sized BIO. For now,
400 * just allocate full-size (16-page) BIOs.
405 sector_t last_block_in_bio
;
406 get_block_t
*get_block
;
410 * We have our BIO, so we can now mark the buffers clean. Make
411 * sure to only clean buffers which we know we'll be writing.
413 static void clean_buffers(struct page
*page
, unsigned first_unmapped
)
415 unsigned buffer_counter
= 0;
416 struct buffer_head
*bh
, *head
;
417 if (!page_has_buffers(page
))
419 head
= page_buffers(page
);
423 if (buffer_counter
++ == first_unmapped
)
425 clear_buffer_dirty(bh
);
426 bh
= bh
->b_this_page
;
427 } while (bh
!= head
);
430 * we cannot drop the bh if the page is not uptodate or a concurrent
431 * read_folio would fail to serialize with the bh and it would read from
432 * disk before we reach the platter.
434 if (buffer_heads_over_limit
&& PageUptodate(page
))
435 try_to_free_buffers(page_folio(page
));
439 * For situations where we want to clean all buffers attached to a page.
440 * We don't need to calculate how many buffers are attached to the page,
441 * we just need to specify a number larger than the maximum number of buffers.
443 void clean_page_buffers(struct page
*page
)
445 clean_buffers(page
, ~0U);
448 static int __mpage_writepage(struct page
*page
, struct writeback_control
*wbc
,
451 struct mpage_data
*mpd
= data
;
452 struct bio
*bio
= mpd
->bio
;
453 struct address_space
*mapping
= page
->mapping
;
454 struct inode
*inode
= page
->mapping
->host
;
455 const unsigned blkbits
= inode
->i_blkbits
;
456 unsigned long end_index
;
457 const unsigned blocks_per_page
= PAGE_SIZE
>> blkbits
;
459 sector_t block_in_file
;
460 sector_t blocks
[MAX_BUF_PER_PAGE
];
462 unsigned first_unmapped
= blocks_per_page
;
463 struct block_device
*bdev
= NULL
;
465 sector_t boundary_block
= 0;
466 struct block_device
*boundary_bdev
= NULL
;
468 struct buffer_head map_bh
;
469 loff_t i_size
= i_size_read(inode
);
472 if (page_has_buffers(page
)) {
473 struct buffer_head
*head
= page_buffers(page
);
474 struct buffer_head
*bh
= head
;
476 /* If they're all mapped and dirty, do it */
479 BUG_ON(buffer_locked(bh
));
480 if (!buffer_mapped(bh
)) {
482 * unmapped dirty buffers are created by
483 * block_dirty_folio -> mmapped data
485 if (buffer_dirty(bh
))
487 if (first_unmapped
== blocks_per_page
)
488 first_unmapped
= page_block
;
492 if (first_unmapped
!= blocks_per_page
)
493 goto confused
; /* hole -> non-hole */
495 if (!buffer_dirty(bh
) || !buffer_uptodate(bh
))
498 if (bh
->b_blocknr
!= blocks
[page_block
-1] + 1)
501 blocks
[page_block
++] = bh
->b_blocknr
;
502 boundary
= buffer_boundary(bh
);
504 boundary_block
= bh
->b_blocknr
;
505 boundary_bdev
= bh
->b_bdev
;
508 } while ((bh
= bh
->b_this_page
) != head
);
514 * Page has buffers, but they are all unmapped. The page was
515 * created by pagein or read over a hole which was handled by
516 * block_read_full_folio(). If this address_space is also
517 * using mpage_readahead then this can rarely happen.
523 * The page has no buffers: map it to disk
525 BUG_ON(!PageUptodate(page
));
526 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
527 last_block
= (i_size
- 1) >> blkbits
;
528 map_bh
.b_page
= page
;
529 for (page_block
= 0; page_block
< blocks_per_page
; ) {
532 map_bh
.b_size
= 1 << blkbits
;
533 if (mpd
->get_block(inode
, block_in_file
, &map_bh
, 1))
535 if (buffer_new(&map_bh
))
536 clean_bdev_bh_alias(&map_bh
);
537 if (buffer_boundary(&map_bh
)) {
538 boundary_block
= map_bh
.b_blocknr
;
539 boundary_bdev
= map_bh
.b_bdev
;
542 if (map_bh
.b_blocknr
!= blocks
[page_block
-1] + 1)
545 blocks
[page_block
++] = map_bh
.b_blocknr
;
546 boundary
= buffer_boundary(&map_bh
);
547 bdev
= map_bh
.b_bdev
;
548 if (block_in_file
== last_block
)
552 BUG_ON(page_block
== 0);
554 first_unmapped
= page_block
;
557 end_index
= i_size
>> PAGE_SHIFT
;
558 if (page
->index
>= end_index
) {
560 * The page straddles i_size. It must be zeroed out on each
561 * and every writepage invocation because it may be mmapped.
562 * "A file is mapped in multiples of the page size. For a file
563 * that is not a multiple of the page size, the remaining memory
564 * is zeroed when mapped, and writes to that region are not
565 * written out to the file."
567 unsigned offset
= i_size
& (PAGE_SIZE
- 1);
569 if (page
->index
> end_index
|| !offset
)
571 zero_user_segment(page
, offset
, PAGE_SIZE
);
575 * This page will go to BIO. Do we need to send this BIO off first?
577 if (bio
&& mpd
->last_block_in_bio
!= blocks
[0] - 1)
578 bio
= mpage_bio_submit(bio
);
582 if (first_unmapped
== blocks_per_page
) {
583 if (!bdev_write_page(bdev
, blocks
[0] << (blkbits
- 9),
587 bio
= bio_alloc(bdev
, BIO_MAX_VECS
,
588 REQ_OP_WRITE
| wbc_to_write_flags(wbc
),
590 bio
->bi_iter
.bi_sector
= blocks
[0] << (blkbits
- 9);
591 wbc_init_bio(wbc
, bio
);
595 * Must try to add the page before marking the buffer clean or
596 * the confused fail path above (OOM) will be very confused when
597 * it finds all bh marked clean (i.e. it will not write anything)
599 wbc_account_cgroup_owner(wbc
, page
, PAGE_SIZE
);
600 length
= first_unmapped
<< blkbits
;
601 if (bio_add_page(bio
, page
, length
, 0) < length
) {
602 bio
= mpage_bio_submit(bio
);
606 clean_buffers(page
, first_unmapped
);
608 BUG_ON(PageWriteback(page
));
609 set_page_writeback(page
);
611 if (boundary
|| (first_unmapped
!= blocks_per_page
)) {
612 bio
= mpage_bio_submit(bio
);
613 if (boundary_block
) {
614 write_boundary_block(boundary_bdev
,
615 boundary_block
, 1 << blkbits
);
618 mpd
->last_block_in_bio
= blocks
[blocks_per_page
- 1];
624 bio
= mpage_bio_submit(bio
);
627 * The caller has a ref on the inode, so *mapping is stable
629 ret
= block_write_full_page(page
, mpd
->get_block
, wbc
);
630 mapping_set_error(mapping
, ret
);
637 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
638 * @mapping: address space structure to write
639 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
640 * @get_block: the filesystem's block mapper function.
642 * This is a library function, which implements the writepages()
643 * address_space_operation.
645 * If a page is already under I/O, generic_writepages() skips it, even
646 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
647 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
648 * and msync() need to guarantee that all the data which was dirty at the time
649 * the call was made get new I/O started against them. If wbc->sync_mode is
650 * WB_SYNC_ALL then we were called for data integrity and we must wait for
651 * existing IO to complete.
654 mpage_writepages(struct address_space
*mapping
,
655 struct writeback_control
*wbc
, get_block_t get_block
)
657 struct mpage_data mpd
= {
658 .get_block
= get_block
,
660 struct blk_plug plug
;
663 blk_start_plug(&plug
);
664 ret
= write_cache_pages(mapping
, wbc
, __mpage_writepage
, &mpd
);
666 mpage_bio_submit(mpd
.bio
);
667 blk_finish_plug(&plug
);
670 EXPORT_SYMBOL(mpage_writepages
);