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/writeback.h>
28 #include <linux/backing-dev.h>
29 #include <linux/pagevec.h>
30 #include <linux/cleancache.h>
33 * I/O completion handler for multipage BIOs.
35 * The mpage code never puts partial pages into a BIO (except for end-of-file).
36 * If a page does not map to a contiguous run of blocks then it simply falls
37 * back to block_read_full_page().
39 * Why is this? If a page's completion depends on a number of different BIOs
40 * which can complete in any order (or at the same time) then determining the
41 * status of that page is hard. See end_buffer_async_read() for the details.
42 * There is no point in duplicating all that complexity.
44 static void mpage_end_io(struct bio
*bio
, int err
)
49 bio_for_each_segment_all(bv
, bio
, i
) {
50 struct page
*page
= bv
->bv_page
;
51 page_endio(page
, bio_data_dir(bio
), err
);
57 static struct bio
*mpage_bio_submit(int rw
, struct bio
*bio
)
59 bio
->bi_end_io
= mpage_end_io
;
65 mpage_alloc(struct block_device
*bdev
,
66 sector_t first_sector
, int nr_vecs
,
71 bio
= bio_alloc(gfp_flags
, nr_vecs
);
73 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
74 while (!bio
&& (nr_vecs
/= 2))
75 bio
= bio_alloc(gfp_flags
, nr_vecs
);
80 bio
->bi_iter
.bi_sector
= first_sector
;
86 * support function for mpage_readpages. The fs supplied get_block might
87 * return an up to date buffer. This is used to map that buffer into
88 * the page, which allows readpage to avoid triggering a duplicate call
91 * The idea is to avoid adding buffers to pages that don't already have
92 * them. So when the buffer is up to date and the page size == block size,
93 * this marks the page up to date instead of adding new buffers.
96 map_buffer_to_page(struct page
*page
, struct buffer_head
*bh
, int page_block
)
98 struct inode
*inode
= page
->mapping
->host
;
99 struct buffer_head
*page_bh
, *head
;
102 if (!page_has_buffers(page
)) {
104 * don't make any buffers if there is only one buffer on
105 * the page and the page just needs to be set up to date
107 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
&&
108 buffer_uptodate(bh
)) {
109 SetPageUptodate(page
);
112 create_empty_buffers(page
, 1 << inode
->i_blkbits
, 0);
114 head
= page_buffers(page
);
117 if (block
== page_block
) {
118 page_bh
->b_state
= bh
->b_state
;
119 page_bh
->b_bdev
= bh
->b_bdev
;
120 page_bh
->b_blocknr
= bh
->b_blocknr
;
123 page_bh
= page_bh
->b_this_page
;
125 } while (page_bh
!= head
);
129 * This is the worker routine which does all the work of mapping the disk
130 * blocks and constructs largest possible bios, submits them for IO if the
131 * blocks are not contiguous on the disk.
133 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
134 * represent the validity of its disk mapping and to decide when to do the next
138 do_mpage_readpage(struct bio
*bio
, struct page
*page
, unsigned nr_pages
,
139 sector_t
*last_block_in_bio
, struct buffer_head
*map_bh
,
140 unsigned long *first_logical_block
, get_block_t get_block
)
142 struct inode
*inode
= page
->mapping
->host
;
143 const unsigned blkbits
= inode
->i_blkbits
;
144 const unsigned blocks_per_page
= PAGE_CACHE_SIZE
>> blkbits
;
145 const unsigned blocksize
= 1 << blkbits
;
146 sector_t block_in_file
;
148 sector_t last_block_in_file
;
149 sector_t blocks
[MAX_BUF_PER_PAGE
];
151 unsigned first_hole
= blocks_per_page
;
152 struct block_device
*bdev
= NULL
;
154 int fully_mapped
= 1;
156 unsigned relative_block
;
158 if (page_has_buffers(page
))
161 block_in_file
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- blkbits
);
162 last_block
= block_in_file
+ 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
) && block_in_file
> *first_logical_block
&&
173 block_in_file
< (*first_logical_block
+ nblocks
)) {
174 unsigned map_offset
= block_in_file
- *first_logical_block
;
175 unsigned last
= nblocks
- map_offset
;
177 for (relative_block
= 0; ; relative_block
++) {
178 if (relative_block
== last
) {
179 clear_buffer_mapped(map_bh
);
182 if (page_block
== blocks_per_page
)
184 blocks
[page_block
] = map_bh
->b_blocknr
+ map_offset
+
189 bdev
= map_bh
->b_bdev
;
193 * Then do more get_blocks calls until we are done with this page.
195 map_bh
->b_page
= page
;
196 while (page_block
< blocks_per_page
) {
200 if (block_in_file
< last_block
) {
201 map_bh
->b_size
= (last_block
-block_in_file
) << blkbits
;
202 if (get_block(inode
, block_in_file
, map_bh
, 0))
204 *first_logical_block
= block_in_file
;
207 if (!buffer_mapped(map_bh
)) {
209 if (first_hole
== blocks_per_page
)
210 first_hole
= page_block
;
216 /* some filesystems will copy data into the page during
217 * the get_block call, in which case we don't want to
218 * read it again. map_buffer_to_page copies the data
219 * we just collected from get_block into the page's buffers
220 * so readpage doesn't have to repeat the get_block call
222 if (buffer_uptodate(map_bh
)) {
223 map_buffer_to_page(page
, map_bh
, page_block
);
227 if (first_hole
!= blocks_per_page
)
228 goto confused
; /* hole -> non-hole */
230 /* Contiguous blocks? */
231 if (page_block
&& blocks
[page_block
-1] != map_bh
->b_blocknr
-1)
233 nblocks
= map_bh
->b_size
>> blkbits
;
234 for (relative_block
= 0; ; relative_block
++) {
235 if (relative_block
== nblocks
) {
236 clear_buffer_mapped(map_bh
);
238 } else if (page_block
== blocks_per_page
)
240 blocks
[page_block
] = map_bh
->b_blocknr
+relative_block
;
244 bdev
= map_bh
->b_bdev
;
247 if (first_hole
!= blocks_per_page
) {
248 zero_user_segment(page
, first_hole
<< blkbits
, PAGE_CACHE_SIZE
);
249 if (first_hole
== 0) {
250 SetPageUptodate(page
);
254 } else if (fully_mapped
) {
255 SetPageMappedToDisk(page
);
258 if (fully_mapped
&& blocks_per_page
== 1 && !PageUptodate(page
) &&
259 cleancache_get_page(page
) == 0) {
260 SetPageUptodate(page
);
265 * This page will go to BIO. Do we need to send this BIO off first?
267 if (bio
&& (*last_block_in_bio
!= blocks
[0] - 1))
268 bio
= mpage_bio_submit(READ
, bio
);
272 bio
= mpage_alloc(bdev
, blocks
[0] << (blkbits
- 9),
273 min_t(int, nr_pages
, bio_get_nr_vecs(bdev
)),
279 length
= first_hole
<< blkbits
;
280 if (bio_add_page(bio
, page
, length
, 0) < length
) {
281 bio
= mpage_bio_submit(READ
, bio
);
285 relative_block
= block_in_file
- *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 bio
= mpage_bio_submit(READ
, bio
);
291 *last_block_in_bio
= blocks
[blocks_per_page
- 1];
297 bio
= mpage_bio_submit(READ
, bio
);
298 if (!PageUptodate(page
))
299 block_read_full_page(page
, get_block
);
306 * mpage_readpages - populate an address space with some pages & start reads against them
307 * @mapping: the address_space
308 * @pages: The address of a list_head which contains the target pages. These
309 * pages have their ->index populated and are otherwise uninitialised.
310 * The page at @pages->prev has the lowest file offset, and reads should be
311 * issued in @pages->prev to @pages->next order.
312 * @nr_pages: The number of pages at *@pages
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_CACHE_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:
336 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
338 * because the indirect block has to be read to get the mappings of blocks
339 * 13,14,15,16. Obviously, this impacts performance.
341 * So what we do it to allow the filesystem's get_block() function to set
342 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
343 * after this one will require I/O against a block which is probably close to
344 * this one. So you should push what I/O you have currently accumulated.
346 * This all causes the disk requests to be issued in the correct order.
349 mpage_readpages(struct address_space
*mapping
, struct list_head
*pages
,
350 unsigned nr_pages
, get_block_t get_block
)
352 struct bio
*bio
= NULL
;
354 sector_t last_block_in_bio
= 0;
355 struct buffer_head map_bh
;
356 unsigned long first_logical_block
= 0;
360 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
361 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
363 prefetchw(&page
->flags
);
364 list_del(&page
->lru
);
365 if (!add_to_page_cache_lru(page
, mapping
,
366 page
->index
, GFP_KERNEL
)) {
367 bio
= do_mpage_readpage(bio
, page
,
369 &last_block_in_bio
, &map_bh
,
370 &first_logical_block
,
373 page_cache_release(page
);
375 BUG_ON(!list_empty(pages
));
377 mpage_bio_submit(READ
, bio
);
380 EXPORT_SYMBOL(mpage_readpages
);
383 * This isn't called much at all
385 int mpage_readpage(struct page
*page
, get_block_t get_block
)
387 struct bio
*bio
= NULL
;
388 sector_t last_block_in_bio
= 0;
389 struct buffer_head map_bh
;
390 unsigned long first_logical_block
= 0;
394 bio
= do_mpage_readpage(bio
, page
, 1, &last_block_in_bio
,
395 &map_bh
, &first_logical_block
, get_block
);
397 mpage_bio_submit(READ
, bio
);
400 EXPORT_SYMBOL(mpage_readpage
);
403 * Writing is not so simple.
405 * If the page has buffers then they will be used for obtaining the disk
406 * mapping. We only support pages which are fully mapped-and-dirty, with a
407 * special case for pages which are unmapped at the end: end-of-file.
409 * If the page has no buffers (preferred) then the page is mapped here.
411 * If all blocks are found to be contiguous then the page can go into the
412 * BIO. Otherwise fall back to the mapping's writepage().
414 * FIXME: This code wants an estimate of how many pages are still to be
415 * written, so it can intelligently allocate a suitably-sized BIO. For now,
416 * just allocate full-size (16-page) BIOs.
421 sector_t last_block_in_bio
;
422 get_block_t
*get_block
;
423 unsigned use_writepage
;
427 * We have our BIO, so we can now mark the buffers clean. Make
428 * sure to only clean buffers which we know we'll be writing.
430 static void clean_buffers(struct page
*page
, unsigned first_unmapped
)
432 unsigned buffer_counter
= 0;
433 struct buffer_head
*bh
, *head
;
434 if (!page_has_buffers(page
))
436 head
= page_buffers(page
);
440 if (buffer_counter
++ == first_unmapped
)
442 clear_buffer_dirty(bh
);
443 bh
= bh
->b_this_page
;
444 } while (bh
!= head
);
447 * we cannot drop the bh if the page is not uptodate or a concurrent
448 * readpage would fail to serialize with the bh and it would read from
449 * disk before we reach the platter.
451 if (buffer_heads_over_limit
&& PageUptodate(page
))
452 try_to_free_buffers(page
);
455 static int __mpage_writepage(struct page
*page
, struct writeback_control
*wbc
,
458 struct mpage_data
*mpd
= data
;
459 struct bio
*bio
= mpd
->bio
;
460 struct address_space
*mapping
= page
->mapping
;
461 struct inode
*inode
= page
->mapping
->host
;
462 const unsigned blkbits
= inode
->i_blkbits
;
463 unsigned long end_index
;
464 const unsigned blocks_per_page
= PAGE_CACHE_SIZE
>> blkbits
;
466 sector_t block_in_file
;
467 sector_t blocks
[MAX_BUF_PER_PAGE
];
469 unsigned first_unmapped
= blocks_per_page
;
470 struct block_device
*bdev
= NULL
;
472 sector_t boundary_block
= 0;
473 struct block_device
*boundary_bdev
= NULL
;
475 struct buffer_head map_bh
;
476 loff_t i_size
= i_size_read(inode
);
479 if (page_has_buffers(page
)) {
480 struct buffer_head
*head
= page_buffers(page
);
481 struct buffer_head
*bh
= head
;
483 /* If they're all mapped and dirty, do it */
486 BUG_ON(buffer_locked(bh
));
487 if (!buffer_mapped(bh
)) {
489 * unmapped dirty buffers are created by
490 * __set_page_dirty_buffers -> mmapped data
492 if (buffer_dirty(bh
))
494 if (first_unmapped
== blocks_per_page
)
495 first_unmapped
= page_block
;
499 if (first_unmapped
!= blocks_per_page
)
500 goto confused
; /* hole -> non-hole */
502 if (!buffer_dirty(bh
) || !buffer_uptodate(bh
))
505 if (bh
->b_blocknr
!= blocks
[page_block
-1] + 1)
508 blocks
[page_block
++] = bh
->b_blocknr
;
509 boundary
= buffer_boundary(bh
);
511 boundary_block
= bh
->b_blocknr
;
512 boundary_bdev
= bh
->b_bdev
;
515 } while ((bh
= bh
->b_this_page
) != head
);
521 * Page has buffers, but they are all unmapped. The page was
522 * created by pagein or read over a hole which was handled by
523 * block_read_full_page(). If this address_space is also
524 * using mpage_readpages then this can rarely happen.
530 * The page has no buffers: map it to disk
532 BUG_ON(!PageUptodate(page
));
533 block_in_file
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- blkbits
);
534 last_block
= (i_size
- 1) >> blkbits
;
535 map_bh
.b_page
= page
;
536 for (page_block
= 0; page_block
< blocks_per_page
; ) {
539 map_bh
.b_size
= 1 << blkbits
;
540 if (mpd
->get_block(inode
, block_in_file
, &map_bh
, 1))
542 if (buffer_new(&map_bh
))
543 unmap_underlying_metadata(map_bh
.b_bdev
,
545 if (buffer_boundary(&map_bh
)) {
546 boundary_block
= map_bh
.b_blocknr
;
547 boundary_bdev
= map_bh
.b_bdev
;
550 if (map_bh
.b_blocknr
!= blocks
[page_block
-1] + 1)
553 blocks
[page_block
++] = map_bh
.b_blocknr
;
554 boundary
= buffer_boundary(&map_bh
);
555 bdev
= map_bh
.b_bdev
;
556 if (block_in_file
== last_block
)
560 BUG_ON(page_block
== 0);
562 first_unmapped
= page_block
;
565 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
566 if (page
->index
>= end_index
) {
568 * The page straddles i_size. It must be zeroed out on each
569 * and every writepage invocation because it may be mmapped.
570 * "A file is mapped in multiples of the page size. For a file
571 * that is not a multiple of the page size, the remaining memory
572 * is zeroed when mapped, and writes to that region are not
573 * written out to the file."
575 unsigned offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
577 if (page
->index
> end_index
|| !offset
)
579 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
583 * This page will go to BIO. Do we need to send this BIO off first?
585 if (bio
&& mpd
->last_block_in_bio
!= blocks
[0] - 1)
586 bio
= mpage_bio_submit(WRITE
, bio
);
590 bio
= mpage_alloc(bdev
, blocks
[0] << (blkbits
- 9),
591 bio_get_nr_vecs(bdev
), GFP_NOFS
|__GFP_HIGH
);
597 * Must try to add the page before marking the buffer clean or
598 * the confused fail path above (OOM) will be very confused when
599 * it finds all bh marked clean (i.e. it will not write anything)
601 length
= first_unmapped
<< blkbits
;
602 if (bio_add_page(bio
, page
, length
, 0) < length
) {
603 bio
= mpage_bio_submit(WRITE
, bio
);
607 clean_buffers(page
, first_unmapped
);
609 BUG_ON(PageWriteback(page
));
610 set_page_writeback(page
);
612 if (boundary
|| (first_unmapped
!= blocks_per_page
)) {
613 bio
= mpage_bio_submit(WRITE
, bio
);
614 if (boundary_block
) {
615 write_boundary_block(boundary_bdev
,
616 boundary_block
, 1 << blkbits
);
619 mpd
->last_block_in_bio
= blocks
[blocks_per_page
- 1];
625 bio
= mpage_bio_submit(WRITE
, bio
);
627 if (mpd
->use_writepage
) {
628 ret
= mapping
->a_ops
->writepage(page
, wbc
);
634 * The caller has a ref on the inode, so *mapping is stable
636 mapping_set_error(mapping
, ret
);
643 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
644 * @mapping: address space structure to write
645 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
646 * @get_block: the filesystem's block mapper function.
647 * If this is NULL then use a_ops->writepage. Otherwise, go
650 * This is a library function, which implements the writepages()
651 * address_space_operation.
653 * If a page is already under I/O, generic_writepages() skips it, even
654 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
655 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
656 * and msync() need to guarantee that all the data which was dirty at the time
657 * the call was made get new I/O started against them. If wbc->sync_mode is
658 * WB_SYNC_ALL then we were called for data integrity and we must wait for
659 * existing IO to complete.
662 mpage_writepages(struct address_space
*mapping
,
663 struct writeback_control
*wbc
, get_block_t get_block
)
665 struct blk_plug plug
;
668 blk_start_plug(&plug
);
671 ret
= generic_writepages(mapping
, wbc
);
673 struct mpage_data mpd
= {
675 .last_block_in_bio
= 0,
676 .get_block
= get_block
,
680 ret
= write_cache_pages(mapping
, wbc
, __mpage_writepage
, &mpd
);
682 mpage_bio_submit(WRITE
, mpd
.bio
);
684 blk_finish_plug(&plug
);
687 EXPORT_SYMBOL(mpage_writepages
);
689 int mpage_writepage(struct page
*page
, get_block_t get_block
,
690 struct writeback_control
*wbc
)
692 struct mpage_data mpd
= {
694 .last_block_in_bio
= 0,
695 .get_block
= get_block
,
698 int ret
= __mpage_writepage(page
, wbc
, &mpd
);
700 mpage_bio_submit(WRITE
, mpd
.bio
);
703 EXPORT_SYMBOL(mpage_writepage
);