4 * Copyright (C) 2002, Linus Torvalds.
6 * Contains functions related to preparing and submitting BIOs which contain
7 * multiple pagecache pages.
9 * 15May2002 Andrew Morton
11 * 27Jun2002 axboe@suse.de
12 * use bio_add_page() to build bio's just the right size
15 #include <linux/kernel.h>
16 #include <linux/export.h>
18 #include <linux/kdev_t.h>
19 #include <linux/gfp.h>
20 #include <linux/bio.h>
22 #include <linux/buffer_head.h>
23 #include <linux/blkdev.h>
24 #include <linux/highmem.h>
25 #include <linux/prefetch.h>
26 #include <linux/mpage.h>
27 #include <linux/mm_inline.h>
28 #include <linux/writeback.h>
29 #include <linux/backing-dev.h>
30 #include <linux/pagevec.h>
31 #include <linux/cleancache.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_page().
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
)
51 bio_for_each_segment_all(bv
, bio
, i
) {
52 struct page
*page
= bv
->bv_page
;
53 page_endio(page
, op_is_write(bio_op(bio
)),
54 blk_status_to_errno(bio
->bi_status
));
60 static struct bio
*mpage_bio_submit(int op
, int op_flags
, struct bio
*bio
)
62 bio
->bi_end_io
= mpage_end_io
;
63 bio_set_op_attrs(bio
, op
, op_flags
);
64 guard_bio_eod(op
, bio
);
70 mpage_alloc(struct block_device
*bdev
,
71 sector_t first_sector
, int nr_vecs
,
76 /* Restrict the given (page cache) mask for slab allocations */
77 gfp_flags
&= GFP_KERNEL
;
78 bio
= bio_alloc(gfp_flags
, nr_vecs
);
80 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
81 while (!bio
&& (nr_vecs
/= 2))
82 bio
= bio_alloc(gfp_flags
, nr_vecs
);
87 bio
->bi_iter
.bi_sector
= first_sector
;
93 * support function for mpage_readpages. The fs supplied get_block might
94 * return an up to date buffer. This is used to map that buffer into
95 * the page, which allows readpage to avoid triggering a duplicate call
98 * The idea is to avoid adding buffers to pages that don't already have
99 * them. So when the buffer is up to date and the page size == block size,
100 * this marks the page up to date instead of adding new buffers.
103 map_buffer_to_page(struct page
*page
, struct buffer_head
*bh
, int page_block
)
105 struct inode
*inode
= page
->mapping
->host
;
106 struct buffer_head
*page_bh
, *head
;
109 if (!page_has_buffers(page
)) {
111 * don't make any buffers if there is only one buffer on
112 * the page and the page just needs to be set up to date
114 if (inode
->i_blkbits
== PAGE_SHIFT
&&
115 buffer_uptodate(bh
)) {
116 SetPageUptodate(page
);
119 create_empty_buffers(page
, i_blocksize(inode
), 0);
121 head
= page_buffers(page
);
124 if (block
== page_block
) {
125 page_bh
->b_state
= bh
->b_state
;
126 page_bh
->b_bdev
= bh
->b_bdev
;
127 page_bh
->b_blocknr
= bh
->b_blocknr
;
130 page_bh
= page_bh
->b_this_page
;
132 } while (page_bh
!= head
);
136 * This is the worker routine which does all the work of mapping the disk
137 * blocks and constructs largest possible bios, submits them for IO if the
138 * blocks are not contiguous on the disk.
140 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
141 * represent the validity of its disk mapping and to decide when to do the next
145 do_mpage_readpage(struct bio
*bio
, struct page
*page
, unsigned nr_pages
,
146 sector_t
*last_block_in_bio
, struct buffer_head
*map_bh
,
147 unsigned long *first_logical_block
, get_block_t get_block
,
150 struct inode
*inode
= page
->mapping
->host
;
151 const unsigned blkbits
= inode
->i_blkbits
;
152 const unsigned blocks_per_page
= PAGE_SIZE
>> blkbits
;
153 const unsigned blocksize
= 1 << blkbits
;
154 sector_t block_in_file
;
156 sector_t last_block_in_file
;
157 sector_t blocks
[MAX_BUF_PER_PAGE
];
159 unsigned first_hole
= blocks_per_page
;
160 struct block_device
*bdev
= NULL
;
162 int fully_mapped
= 1;
164 unsigned relative_block
;
166 if (page_has_buffers(page
))
169 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
170 last_block
= block_in_file
+ nr_pages
* blocks_per_page
;
171 last_block_in_file
= (i_size_read(inode
) + blocksize
- 1) >> blkbits
;
172 if (last_block
> last_block_in_file
)
173 last_block
= last_block_in_file
;
177 * Map blocks using the result from the previous get_blocks call first.
179 nblocks
= map_bh
->b_size
>> blkbits
;
180 if (buffer_mapped(map_bh
) && block_in_file
> *first_logical_block
&&
181 block_in_file
< (*first_logical_block
+ nblocks
)) {
182 unsigned map_offset
= block_in_file
- *first_logical_block
;
183 unsigned last
= nblocks
- map_offset
;
185 for (relative_block
= 0; ; relative_block
++) {
186 if (relative_block
== last
) {
187 clear_buffer_mapped(map_bh
);
190 if (page_block
== blocks_per_page
)
192 blocks
[page_block
] = map_bh
->b_blocknr
+ map_offset
+
197 bdev
= map_bh
->b_bdev
;
201 * Then do more get_blocks calls until we are done with this page.
203 map_bh
->b_page
= page
;
204 while (page_block
< blocks_per_page
) {
208 if (block_in_file
< last_block
) {
209 map_bh
->b_size
= (last_block
-block_in_file
) << blkbits
;
210 if (get_block(inode
, block_in_file
, map_bh
, 0))
212 *first_logical_block
= block_in_file
;
215 if (!buffer_mapped(map_bh
)) {
217 if (first_hole
== blocks_per_page
)
218 first_hole
= page_block
;
224 /* some filesystems will copy data into the page during
225 * the get_block call, in which case we don't want to
226 * read it again. map_buffer_to_page copies the data
227 * we just collected from get_block into the page's buffers
228 * so readpage doesn't have to repeat the get_block call
230 if (buffer_uptodate(map_bh
)) {
231 map_buffer_to_page(page
, map_bh
, page_block
);
235 if (first_hole
!= blocks_per_page
)
236 goto confused
; /* hole -> non-hole */
238 /* Contiguous blocks? */
239 if (page_block
&& blocks
[page_block
-1] != map_bh
->b_blocknr
-1)
241 nblocks
= map_bh
->b_size
>> blkbits
;
242 for (relative_block
= 0; ; relative_block
++) {
243 if (relative_block
== nblocks
) {
244 clear_buffer_mapped(map_bh
);
246 } else if (page_block
== blocks_per_page
)
248 blocks
[page_block
] = map_bh
->b_blocknr
+relative_block
;
252 bdev
= map_bh
->b_bdev
;
255 if (first_hole
!= blocks_per_page
) {
256 zero_user_segment(page
, first_hole
<< blkbits
, PAGE_SIZE
);
257 if (first_hole
== 0) {
258 SetPageUptodate(page
);
262 } else if (fully_mapped
) {
263 SetPageMappedToDisk(page
);
266 if (fully_mapped
&& blocks_per_page
== 1 && !PageUptodate(page
) &&
267 cleancache_get_page(page
) == 0) {
268 SetPageUptodate(page
);
273 * This page will go to BIO. Do we need to send this BIO off first?
275 if (bio
&& (*last_block_in_bio
!= blocks
[0] - 1))
276 bio
= mpage_bio_submit(REQ_OP_READ
, 0, bio
);
280 if (first_hole
== blocks_per_page
) {
281 if (!bdev_read_page(bdev
, blocks
[0] << (blkbits
- 9),
285 bio
= mpage_alloc(bdev
, blocks
[0] << (blkbits
- 9),
286 min_t(int, nr_pages
, BIO_MAX_PAGES
), gfp
);
291 length
= first_hole
<< blkbits
;
292 if (bio_add_page(bio
, page
, length
, 0) < length
) {
293 bio
= mpage_bio_submit(REQ_OP_READ
, 0, bio
);
297 relative_block
= block_in_file
- *first_logical_block
;
298 nblocks
= map_bh
->b_size
>> blkbits
;
299 if ((buffer_boundary(map_bh
) && relative_block
== nblocks
) ||
300 (first_hole
!= blocks_per_page
))
301 bio
= mpage_bio_submit(REQ_OP_READ
, 0, bio
);
303 *last_block_in_bio
= blocks
[blocks_per_page
- 1];
309 bio
= mpage_bio_submit(REQ_OP_READ
, 0, bio
);
310 if (!PageUptodate(page
))
311 block_read_full_page(page
, get_block
);
318 * mpage_readpages - populate an address space with some pages & start reads against them
319 * @mapping: the address_space
320 * @pages: The address of a list_head which contains the target pages. These
321 * pages have their ->index populated and are otherwise uninitialised.
322 * The page at @pages->prev has the lowest file offset, and reads should be
323 * issued in @pages->prev to @pages->next order.
324 * @nr_pages: The number of pages at *@pages
325 * @get_block: The filesystem's block mapper function.
327 * This function walks the pages and the blocks within each page, building and
328 * emitting large BIOs.
330 * If anything unusual happens, such as:
332 * - encountering a page which has buffers
333 * - encountering a page which has a non-hole after a hole
334 * - encountering a page with non-contiguous blocks
336 * then this code just gives up and calls the buffer_head-based read function.
337 * It does handle a page which has holes at the end - that is a common case:
338 * the end-of-file on blocksize < PAGE_SIZE setups.
340 * BH_Boundary explanation:
342 * There is a problem. The mpage read code assembles several pages, gets all
343 * their disk mappings, and then submits them all. That's fine, but obtaining
344 * the disk mappings may require I/O. Reads of indirect blocks, for example.
346 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
347 * submitted in the following order:
349 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
351 * because the indirect block has to be read to get the mappings of blocks
352 * 13,14,15,16. Obviously, this impacts performance.
354 * So what we do it to allow the filesystem's get_block() function to set
355 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
356 * after this one will require I/O against a block which is probably close to
357 * this one. So you should push what I/O you have currently accumulated.
359 * This all causes the disk requests to be issued in the correct order.
362 mpage_readpages(struct address_space
*mapping
, struct list_head
*pages
,
363 unsigned nr_pages
, get_block_t get_block
)
365 struct bio
*bio
= NULL
;
367 sector_t last_block_in_bio
= 0;
368 struct buffer_head map_bh
;
369 unsigned long first_logical_block
= 0;
370 gfp_t gfp
= readahead_gfp_mask(mapping
);
374 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
375 struct page
*page
= lru_to_page(pages
);
377 prefetchw(&page
->flags
);
378 list_del(&page
->lru
);
379 if (!add_to_page_cache_lru(page
, mapping
,
382 bio
= do_mpage_readpage(bio
, page
,
384 &last_block_in_bio
, &map_bh
,
385 &first_logical_block
,
390 BUG_ON(!list_empty(pages
));
392 mpage_bio_submit(REQ_OP_READ
, 0, bio
);
395 EXPORT_SYMBOL(mpage_readpages
);
398 * This isn't called much at all
400 int mpage_readpage(struct page
*page
, get_block_t get_block
)
402 struct bio
*bio
= NULL
;
403 sector_t last_block_in_bio
= 0;
404 struct buffer_head map_bh
;
405 unsigned long first_logical_block
= 0;
406 gfp_t gfp
= mapping_gfp_constraint(page
->mapping
, GFP_KERNEL
);
410 bio
= do_mpage_readpage(bio
, page
, 1, &last_block_in_bio
,
411 &map_bh
, &first_logical_block
, get_block
, gfp
);
413 mpage_bio_submit(REQ_OP_READ
, 0, bio
);
416 EXPORT_SYMBOL(mpage_readpage
);
419 * Writing is not so simple.
421 * If the page has buffers then they will be used for obtaining the disk
422 * mapping. We only support pages which are fully mapped-and-dirty, with a
423 * special case for pages which are unmapped at the end: end-of-file.
425 * If the page has no buffers (preferred) then the page is mapped here.
427 * If all blocks are found to be contiguous then the page can go into the
428 * BIO. Otherwise fall back to the mapping's writepage().
430 * FIXME: This code wants an estimate of how many pages are still to be
431 * written, so it can intelligently allocate a suitably-sized BIO. For now,
432 * just allocate full-size (16-page) BIOs.
437 sector_t last_block_in_bio
;
438 get_block_t
*get_block
;
439 unsigned use_writepage
;
443 * We have our BIO, so we can now mark the buffers clean. Make
444 * sure to only clean buffers which we know we'll be writing.
446 static void clean_buffers(struct page
*page
, unsigned first_unmapped
)
448 unsigned buffer_counter
= 0;
449 struct buffer_head
*bh
, *head
;
450 if (!page_has_buffers(page
))
452 head
= page_buffers(page
);
456 if (buffer_counter
++ == first_unmapped
)
458 clear_buffer_dirty(bh
);
459 bh
= bh
->b_this_page
;
460 } while (bh
!= head
);
463 * we cannot drop the bh if the page is not uptodate or a concurrent
464 * readpage would fail to serialize with the bh and it would read from
465 * disk before we reach the platter.
467 if (buffer_heads_over_limit
&& PageUptodate(page
))
468 try_to_free_buffers(page
);
471 static int __mpage_writepage(struct page
*page
, struct writeback_control
*wbc
,
474 struct mpage_data
*mpd
= data
;
475 struct bio
*bio
= mpd
->bio
;
476 struct address_space
*mapping
= page
->mapping
;
477 struct inode
*inode
= page
->mapping
->host
;
478 const unsigned blkbits
= inode
->i_blkbits
;
479 unsigned long end_index
;
480 const unsigned blocks_per_page
= PAGE_SIZE
>> blkbits
;
482 sector_t block_in_file
;
483 sector_t blocks
[MAX_BUF_PER_PAGE
];
485 unsigned first_unmapped
= blocks_per_page
;
486 struct block_device
*bdev
= NULL
;
488 sector_t boundary_block
= 0;
489 struct block_device
*boundary_bdev
= NULL
;
491 struct buffer_head map_bh
;
492 loff_t i_size
= i_size_read(inode
);
494 int op_flags
= wbc_to_write_flags(wbc
);
496 if (page_has_buffers(page
)) {
497 struct buffer_head
*head
= page_buffers(page
);
498 struct buffer_head
*bh
= head
;
500 /* If they're all mapped and dirty, do it */
503 BUG_ON(buffer_locked(bh
));
504 if (!buffer_mapped(bh
)) {
506 * unmapped dirty buffers are created by
507 * __set_page_dirty_buffers -> mmapped data
509 if (buffer_dirty(bh
))
511 if (first_unmapped
== blocks_per_page
)
512 first_unmapped
= page_block
;
516 if (first_unmapped
!= blocks_per_page
)
517 goto confused
; /* hole -> non-hole */
519 if (!buffer_dirty(bh
) || !buffer_uptodate(bh
))
522 if (bh
->b_blocknr
!= blocks
[page_block
-1] + 1)
525 blocks
[page_block
++] = bh
->b_blocknr
;
526 boundary
= buffer_boundary(bh
);
528 boundary_block
= bh
->b_blocknr
;
529 boundary_bdev
= bh
->b_bdev
;
532 } while ((bh
= bh
->b_this_page
) != head
);
538 * Page has buffers, but they are all unmapped. The page was
539 * created by pagein or read over a hole which was handled by
540 * block_read_full_page(). If this address_space is also
541 * using mpage_readpages then this can rarely happen.
547 * The page has no buffers: map it to disk
549 BUG_ON(!PageUptodate(page
));
550 block_in_file
= (sector_t
)page
->index
<< (PAGE_SHIFT
- blkbits
);
551 last_block
= (i_size
- 1) >> blkbits
;
552 map_bh
.b_page
= page
;
553 for (page_block
= 0; page_block
< blocks_per_page
; ) {
556 map_bh
.b_size
= 1 << blkbits
;
557 if (mpd
->get_block(inode
, block_in_file
, &map_bh
, 1))
559 if (buffer_new(&map_bh
))
560 clean_bdev_bh_alias(&map_bh
);
561 if (buffer_boundary(&map_bh
)) {
562 boundary_block
= map_bh
.b_blocknr
;
563 boundary_bdev
= map_bh
.b_bdev
;
566 if (map_bh
.b_blocknr
!= blocks
[page_block
-1] + 1)
569 blocks
[page_block
++] = map_bh
.b_blocknr
;
570 boundary
= buffer_boundary(&map_bh
);
571 bdev
= map_bh
.b_bdev
;
572 if (block_in_file
== last_block
)
576 BUG_ON(page_block
== 0);
578 first_unmapped
= page_block
;
581 end_index
= i_size
>> PAGE_SHIFT
;
582 if (page
->index
>= end_index
) {
584 * The page straddles i_size. It must be zeroed out on each
585 * and every writepage invocation because it may be mmapped.
586 * "A file is mapped in multiples of the page size. For a file
587 * that is not a multiple of the page size, the remaining memory
588 * is zeroed when mapped, and writes to that region are not
589 * written out to the file."
591 unsigned offset
= i_size
& (PAGE_SIZE
- 1);
593 if (page
->index
> end_index
|| !offset
)
595 zero_user_segment(page
, offset
, PAGE_SIZE
);
599 * This page will go to BIO. Do we need to send this BIO off first?
601 if (bio
&& mpd
->last_block_in_bio
!= blocks
[0] - 1)
602 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
606 if (first_unmapped
== blocks_per_page
) {
607 if (!bdev_write_page(bdev
, blocks
[0] << (blkbits
- 9),
609 clean_buffers(page
, first_unmapped
);
613 bio
= mpage_alloc(bdev
, blocks
[0] << (blkbits
- 9),
614 BIO_MAX_PAGES
, GFP_NOFS
|__GFP_HIGH
);
618 wbc_init_bio(wbc
, bio
);
619 bio
->bi_write_hint
= inode
->i_write_hint
;
623 * Must try to add the page before marking the buffer clean or
624 * the confused fail path above (OOM) will be very confused when
625 * it finds all bh marked clean (i.e. it will not write anything)
627 wbc_account_io(wbc
, page
, PAGE_SIZE
);
628 length
= first_unmapped
<< blkbits
;
629 if (bio_add_page(bio
, page
, length
, 0) < length
) {
630 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
634 clean_buffers(page
, first_unmapped
);
636 BUG_ON(PageWriteback(page
));
637 set_page_writeback(page
);
639 if (boundary
|| (first_unmapped
!= blocks_per_page
)) {
640 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
641 if (boundary_block
) {
642 write_boundary_block(boundary_bdev
,
643 boundary_block
, 1 << blkbits
);
646 mpd
->last_block_in_bio
= blocks
[blocks_per_page
- 1];
652 bio
= mpage_bio_submit(REQ_OP_WRITE
, op_flags
, bio
);
654 if (mpd
->use_writepage
) {
655 ret
= mapping
->a_ops
->writepage(page
, wbc
);
661 * The caller has a ref on the inode, so *mapping is stable
663 mapping_set_error(mapping
, ret
);
670 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
671 * @mapping: address space structure to write
672 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
673 * @get_block: the filesystem's block mapper function.
674 * If this is NULL then use a_ops->writepage. Otherwise, go
677 * This is a library function, which implements the writepages()
678 * address_space_operation.
680 * If a page is already under I/O, generic_writepages() skips it, even
681 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
682 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
683 * and msync() need to guarantee that all the data which was dirty at the time
684 * the call was made get new I/O started against them. If wbc->sync_mode is
685 * WB_SYNC_ALL then we were called for data integrity and we must wait for
686 * existing IO to complete.
689 mpage_writepages(struct address_space
*mapping
,
690 struct writeback_control
*wbc
, get_block_t get_block
)
692 struct blk_plug plug
;
695 blk_start_plug(&plug
);
698 ret
= generic_writepages(mapping
, wbc
);
700 struct mpage_data mpd
= {
702 .last_block_in_bio
= 0,
703 .get_block
= get_block
,
707 ret
= write_cache_pages(mapping
, wbc
, __mpage_writepage
, &mpd
);
709 int op_flags
= (wbc
->sync_mode
== WB_SYNC_ALL
?
711 mpage_bio_submit(REQ_OP_WRITE
, op_flags
, mpd
.bio
);
714 blk_finish_plug(&plug
);
717 EXPORT_SYMBOL(mpage_writepages
);
719 int mpage_writepage(struct page
*page
, get_block_t get_block
,
720 struct writeback_control
*wbc
)
722 struct mpage_data mpd
= {
724 .last_block_in_bio
= 0,
725 .get_block
= get_block
,
728 int ret
= __mpage_writepage(page
, wbc
, &mpd
);
730 int op_flags
= (wbc
->sync_mode
== WB_SYNC_ALL
?
732 mpage_bio_submit(REQ_OP_WRITE
, op_flags
, mpd
.bio
);
736 EXPORT_SYMBOL(mpage_writepage
);