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>
34 * I/O completion handler for multipage BIOs.
36 * The mpage code never puts partial pages into a BIO (except for end-of-file).
37 * If a page does not map to a contiguous run of blocks then it simply falls
38 * back to block_read_full_page().
40 * Why is this? If a page's completion depends on a number of different BIOs
41 * which can complete in any order (or at the same time) then determining the
42 * status of that page is hard. See end_buffer_async_read() for the details.
43 * There is no point in duplicating all that complexity.
45 static void mpage_end_io(struct bio
*bio
)
50 bio_for_each_segment_all(bv
, bio
, i
) {
51 struct page
*page
= bv
->bv_page
;
52 page_endio(page
, bio_data_dir(bio
), bio
->bi_error
);
58 static struct bio
*mpage_bio_submit(int rw
, struct bio
*bio
)
60 bio
->bi_end_io
= mpage_end_io
;
61 guard_bio_eod(rw
, bio
);
67 mpage_alloc(struct block_device
*bdev
,
68 sector_t first_sector
, int nr_vecs
,
73 bio
= bio_alloc(gfp_flags
, nr_vecs
);
75 if (bio
== NULL
&& (current
->flags
& PF_MEMALLOC
)) {
76 while (!bio
&& (nr_vecs
/= 2))
77 bio
= bio_alloc(gfp_flags
, nr_vecs
);
82 bio
->bi_iter
.bi_sector
= first_sector
;
88 * support function for mpage_readpages. The fs supplied get_block might
89 * return an up to date buffer. This is used to map that buffer into
90 * the page, which allows readpage to avoid triggering a duplicate call
93 * The idea is to avoid adding buffers to pages that don't already have
94 * them. So when the buffer is up to date and the page size == block size,
95 * this marks the page up to date instead of adding new buffers.
98 map_buffer_to_page(struct page
*page
, struct buffer_head
*bh
, int page_block
)
100 struct inode
*inode
= page
->mapping
->host
;
101 struct buffer_head
*page_bh
, *head
;
104 if (!page_has_buffers(page
)) {
106 * don't make any buffers if there is only one buffer on
107 * the page and the page just needs to be set up to date
109 if (inode
->i_blkbits
== PAGE_CACHE_SHIFT
&&
110 buffer_uptodate(bh
)) {
111 SetPageUptodate(page
);
114 create_empty_buffers(page
, 1 << inode
->i_blkbits
, 0);
116 head
= page_buffers(page
);
119 if (block
== page_block
) {
120 page_bh
->b_state
= bh
->b_state
;
121 page_bh
->b_bdev
= bh
->b_bdev
;
122 page_bh
->b_blocknr
= bh
->b_blocknr
;
125 page_bh
= page_bh
->b_this_page
;
127 } while (page_bh
!= head
);
131 * This is the worker routine which does all the work of mapping the disk
132 * blocks and constructs largest possible bios, submits them for IO if the
133 * blocks are not contiguous on the disk.
135 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
136 * represent the validity of its disk mapping and to decide when to do the next
140 do_mpage_readpage(struct bio
*bio
, struct page
*page
, unsigned nr_pages
,
141 sector_t
*last_block_in_bio
, struct buffer_head
*map_bh
,
142 unsigned long *first_logical_block
, get_block_t get_block
)
144 struct inode
*inode
= page
->mapping
->host
;
145 const unsigned blkbits
= inode
->i_blkbits
;
146 const unsigned blocks_per_page
= PAGE_CACHE_SIZE
>> blkbits
;
147 const unsigned blocksize
= 1 << blkbits
;
148 sector_t block_in_file
;
150 sector_t last_block_in_file
;
151 sector_t blocks
[MAX_BUF_PER_PAGE
];
153 unsigned first_hole
= blocks_per_page
;
154 struct block_device
*bdev
= NULL
;
156 int fully_mapped
= 1;
158 unsigned relative_block
;
160 if (page_has_buffers(page
))
163 block_in_file
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- blkbits
);
164 last_block
= block_in_file
+ nr_pages
* blocks_per_page
;
165 last_block_in_file
= (i_size_read(inode
) + blocksize
- 1) >> blkbits
;
166 if (last_block
> last_block_in_file
)
167 last_block
= last_block_in_file
;
171 * Map blocks using the result from the previous get_blocks call first.
173 nblocks
= map_bh
->b_size
>> blkbits
;
174 if (buffer_mapped(map_bh
) && block_in_file
> *first_logical_block
&&
175 block_in_file
< (*first_logical_block
+ nblocks
)) {
176 unsigned map_offset
= block_in_file
- *first_logical_block
;
177 unsigned last
= nblocks
- map_offset
;
179 for (relative_block
= 0; ; relative_block
++) {
180 if (relative_block
== last
) {
181 clear_buffer_mapped(map_bh
);
184 if (page_block
== blocks_per_page
)
186 blocks
[page_block
] = map_bh
->b_blocknr
+ map_offset
+
191 bdev
= map_bh
->b_bdev
;
195 * Then do more get_blocks calls until we are done with this page.
197 map_bh
->b_page
= page
;
198 while (page_block
< blocks_per_page
) {
202 if (block_in_file
< last_block
) {
203 map_bh
->b_size
= (last_block
-block_in_file
) << blkbits
;
204 if (get_block(inode
, block_in_file
, map_bh
, 0))
206 *first_logical_block
= block_in_file
;
209 if (!buffer_mapped(map_bh
)) {
211 if (first_hole
== blocks_per_page
)
212 first_hole
= page_block
;
218 /* some filesystems will copy data into the page during
219 * the get_block call, in which case we don't want to
220 * read it again. map_buffer_to_page copies the data
221 * we just collected from get_block into the page's buffers
222 * so readpage doesn't have to repeat the get_block call
224 if (buffer_uptodate(map_bh
)) {
225 map_buffer_to_page(page
, map_bh
, page_block
);
229 if (first_hole
!= blocks_per_page
)
230 goto confused
; /* hole -> non-hole */
232 /* Contiguous blocks? */
233 if (page_block
&& blocks
[page_block
-1] != map_bh
->b_blocknr
-1)
235 nblocks
= map_bh
->b_size
>> blkbits
;
236 for (relative_block
= 0; ; relative_block
++) {
237 if (relative_block
== nblocks
) {
238 clear_buffer_mapped(map_bh
);
240 } else if (page_block
== blocks_per_page
)
242 blocks
[page_block
] = map_bh
->b_blocknr
+relative_block
;
246 bdev
= map_bh
->b_bdev
;
249 if (first_hole
!= blocks_per_page
) {
250 zero_user_segment(page
, first_hole
<< blkbits
, PAGE_CACHE_SIZE
);
251 if (first_hole
== 0) {
252 SetPageUptodate(page
);
256 } else if (fully_mapped
) {
257 SetPageMappedToDisk(page
);
260 if (fully_mapped
&& blocks_per_page
== 1 && !PageUptodate(page
) &&
261 cleancache_get_page(page
) == 0) {
262 SetPageUptodate(page
);
267 * This page will go to BIO. Do we need to send this BIO off first?
269 if (bio
&& (*last_block_in_bio
!= blocks
[0] - 1))
270 bio
= mpage_bio_submit(READ
, bio
);
274 if (first_hole
== blocks_per_page
) {
275 if (!bdev_read_page(bdev
, blocks
[0] << (blkbits
- 9),
279 bio
= mpage_alloc(bdev
, blocks
[0] << (blkbits
- 9),
280 min_t(int, nr_pages
, bio_get_nr_vecs(bdev
)),
286 length
= first_hole
<< blkbits
;
287 if (bio_add_page(bio
, page
, length
, 0) < length
) {
288 bio
= mpage_bio_submit(READ
, bio
);
292 relative_block
= block_in_file
- *first_logical_block
;
293 nblocks
= map_bh
->b_size
>> blkbits
;
294 if ((buffer_boundary(map_bh
) && relative_block
== nblocks
) ||
295 (first_hole
!= blocks_per_page
))
296 bio
= mpage_bio_submit(READ
, bio
);
298 *last_block_in_bio
= blocks
[blocks_per_page
- 1];
304 bio
= mpage_bio_submit(READ
, bio
);
305 if (!PageUptodate(page
))
306 block_read_full_page(page
, get_block
);
313 * mpage_readpages - populate an address space with some pages & start reads against them
314 * @mapping: the address_space
315 * @pages: The address of a list_head which contains the target pages. These
316 * pages have their ->index populated and are otherwise uninitialised.
317 * The page at @pages->prev has the lowest file offset, and reads should be
318 * issued in @pages->prev to @pages->next order.
319 * @nr_pages: The number of pages at *@pages
320 * @get_block: The filesystem's block mapper function.
322 * This function walks the pages and the blocks within each page, building and
323 * emitting large BIOs.
325 * If anything unusual happens, such as:
327 * - encountering a page which has buffers
328 * - encountering a page which has a non-hole after a hole
329 * - encountering a page with non-contiguous blocks
331 * then this code just gives up and calls the buffer_head-based read function.
332 * It does handle a page which has holes at the end - that is a common case:
333 * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
335 * BH_Boundary explanation:
337 * There is a problem. The mpage read code assembles several pages, gets all
338 * their disk mappings, and then submits them all. That's fine, but obtaining
339 * the disk mappings may require I/O. Reads of indirect blocks, for example.
341 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
342 * submitted in the following order:
343 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
345 * because the indirect block has to be read to get the mappings of blocks
346 * 13,14,15,16. Obviously, this impacts performance.
348 * So what we do it to allow the filesystem's get_block() function to set
349 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
350 * after this one will require I/O against a block which is probably close to
351 * this one. So you should push what I/O you have currently accumulated.
353 * This all causes the disk requests to be issued in the correct order.
356 mpage_readpages(struct address_space
*mapping
, struct list_head
*pages
,
357 unsigned nr_pages
, get_block_t get_block
)
359 struct bio
*bio
= NULL
;
361 sector_t last_block_in_bio
= 0;
362 struct buffer_head map_bh
;
363 unsigned long first_logical_block
= 0;
367 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
368 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
370 prefetchw(&page
->flags
);
371 list_del(&page
->lru
);
372 if (!add_to_page_cache_lru(page
, mapping
,
373 page
->index
, GFP_KERNEL
)) {
374 bio
= do_mpage_readpage(bio
, page
,
376 &last_block_in_bio
, &map_bh
,
377 &first_logical_block
,
380 page_cache_release(page
);
382 BUG_ON(!list_empty(pages
));
384 mpage_bio_submit(READ
, bio
);
387 EXPORT_SYMBOL(mpage_readpages
);
390 * This isn't called much at all
392 int mpage_readpage(struct page
*page
, get_block_t get_block
)
394 struct bio
*bio
= NULL
;
395 sector_t last_block_in_bio
= 0;
396 struct buffer_head map_bh
;
397 unsigned long first_logical_block
= 0;
401 bio
= do_mpage_readpage(bio
, page
, 1, &last_block_in_bio
,
402 &map_bh
, &first_logical_block
, get_block
);
404 mpage_bio_submit(READ
, bio
);
407 EXPORT_SYMBOL(mpage_readpage
);
410 * Writing is not so simple.
412 * If the page has buffers then they will be used for obtaining the disk
413 * mapping. We only support pages which are fully mapped-and-dirty, with a
414 * special case for pages which are unmapped at the end: end-of-file.
416 * If the page has no buffers (preferred) then the page is mapped here.
418 * If all blocks are found to be contiguous then the page can go into the
419 * BIO. Otherwise fall back to the mapping's writepage().
421 * FIXME: This code wants an estimate of how many pages are still to be
422 * written, so it can intelligently allocate a suitably-sized BIO. For now,
423 * just allocate full-size (16-page) BIOs.
428 sector_t last_block_in_bio
;
429 get_block_t
*get_block
;
430 unsigned use_writepage
;
434 * We have our BIO, so we can now mark the buffers clean. Make
435 * sure to only clean buffers which we know we'll be writing.
437 static void clean_buffers(struct page
*page
, unsigned first_unmapped
)
439 unsigned buffer_counter
= 0;
440 struct buffer_head
*bh
, *head
;
441 if (!page_has_buffers(page
))
443 head
= page_buffers(page
);
447 if (buffer_counter
++ == first_unmapped
)
449 clear_buffer_dirty(bh
);
450 bh
= bh
->b_this_page
;
451 } while (bh
!= head
);
454 * we cannot drop the bh if the page is not uptodate or a concurrent
455 * readpage would fail to serialize with the bh and it would read from
456 * disk before we reach the platter.
458 if (buffer_heads_over_limit
&& PageUptodate(page
))
459 try_to_free_buffers(page
);
462 static int __mpage_writepage(struct page
*page
, struct writeback_control
*wbc
,
465 struct mpage_data
*mpd
= data
;
466 struct bio
*bio
= mpd
->bio
;
467 struct address_space
*mapping
= page
->mapping
;
468 struct inode
*inode
= page
->mapping
->host
;
469 const unsigned blkbits
= inode
->i_blkbits
;
470 unsigned long end_index
;
471 const unsigned blocks_per_page
= PAGE_CACHE_SIZE
>> blkbits
;
473 sector_t block_in_file
;
474 sector_t blocks
[MAX_BUF_PER_PAGE
];
476 unsigned first_unmapped
= blocks_per_page
;
477 struct block_device
*bdev
= NULL
;
479 sector_t boundary_block
= 0;
480 struct block_device
*boundary_bdev
= NULL
;
482 struct buffer_head map_bh
;
483 loff_t i_size
= i_size_read(inode
);
486 if (page_has_buffers(page
)) {
487 struct buffer_head
*head
= page_buffers(page
);
488 struct buffer_head
*bh
= head
;
490 /* If they're all mapped and dirty, do it */
493 BUG_ON(buffer_locked(bh
));
494 if (!buffer_mapped(bh
)) {
496 * unmapped dirty buffers are created by
497 * __set_page_dirty_buffers -> mmapped data
499 if (buffer_dirty(bh
))
501 if (first_unmapped
== blocks_per_page
)
502 first_unmapped
= page_block
;
506 if (first_unmapped
!= blocks_per_page
)
507 goto confused
; /* hole -> non-hole */
509 if (!buffer_dirty(bh
) || !buffer_uptodate(bh
))
512 if (bh
->b_blocknr
!= blocks
[page_block
-1] + 1)
515 blocks
[page_block
++] = bh
->b_blocknr
;
516 boundary
= buffer_boundary(bh
);
518 boundary_block
= bh
->b_blocknr
;
519 boundary_bdev
= bh
->b_bdev
;
522 } while ((bh
= bh
->b_this_page
) != head
);
528 * Page has buffers, but they are all unmapped. The page was
529 * created by pagein or read over a hole which was handled by
530 * block_read_full_page(). If this address_space is also
531 * using mpage_readpages then this can rarely happen.
537 * The page has no buffers: map it to disk
539 BUG_ON(!PageUptodate(page
));
540 block_in_file
= (sector_t
)page
->index
<< (PAGE_CACHE_SHIFT
- blkbits
);
541 last_block
= (i_size
- 1) >> blkbits
;
542 map_bh
.b_page
= page
;
543 for (page_block
= 0; page_block
< blocks_per_page
; ) {
546 map_bh
.b_size
= 1 << blkbits
;
547 if (mpd
->get_block(inode
, block_in_file
, &map_bh
, 1))
549 if (buffer_new(&map_bh
))
550 unmap_underlying_metadata(map_bh
.b_bdev
,
552 if (buffer_boundary(&map_bh
)) {
553 boundary_block
= map_bh
.b_blocknr
;
554 boundary_bdev
= map_bh
.b_bdev
;
557 if (map_bh
.b_blocknr
!= blocks
[page_block
-1] + 1)
560 blocks
[page_block
++] = map_bh
.b_blocknr
;
561 boundary
= buffer_boundary(&map_bh
);
562 bdev
= map_bh
.b_bdev
;
563 if (block_in_file
== last_block
)
567 BUG_ON(page_block
== 0);
569 first_unmapped
= page_block
;
572 end_index
= i_size
>> PAGE_CACHE_SHIFT
;
573 if (page
->index
>= end_index
) {
575 * The page straddles i_size. It must be zeroed out on each
576 * and every writepage invocation because it may be mmapped.
577 * "A file is mapped in multiples of the page size. For a file
578 * that is not a multiple of the page size, the remaining memory
579 * is zeroed when mapped, and writes to that region are not
580 * written out to the file."
582 unsigned offset
= i_size
& (PAGE_CACHE_SIZE
- 1);
584 if (page
->index
> end_index
|| !offset
)
586 zero_user_segment(page
, offset
, PAGE_CACHE_SIZE
);
590 * This page will go to BIO. Do we need to send this BIO off first?
592 if (bio
&& mpd
->last_block_in_bio
!= blocks
[0] - 1)
593 bio
= mpage_bio_submit(WRITE
, bio
);
597 if (first_unmapped
== blocks_per_page
) {
598 if (!bdev_write_page(bdev
, blocks
[0] << (blkbits
- 9),
600 clean_buffers(page
, first_unmapped
);
604 bio
= mpage_alloc(bdev
, blocks
[0] << (blkbits
- 9),
605 bio_get_nr_vecs(bdev
), GFP_NOFS
|__GFP_HIGH
);
609 wbc_init_bio(wbc
, bio
);
613 * Must try to add the page before marking the buffer clean or
614 * the confused fail path above (OOM) will be very confused when
615 * it finds all bh marked clean (i.e. it will not write anything)
617 wbc_account_io(wbc
, page
, PAGE_SIZE
);
618 length
= first_unmapped
<< blkbits
;
619 if (bio_add_page(bio
, page
, length
, 0) < length
) {
620 bio
= mpage_bio_submit(WRITE
, bio
);
624 clean_buffers(page
, first_unmapped
);
626 BUG_ON(PageWriteback(page
));
627 set_page_writeback(page
);
629 if (boundary
|| (first_unmapped
!= blocks_per_page
)) {
630 bio
= mpage_bio_submit(WRITE
, bio
);
631 if (boundary_block
) {
632 write_boundary_block(boundary_bdev
,
633 boundary_block
, 1 << blkbits
);
636 mpd
->last_block_in_bio
= blocks
[blocks_per_page
- 1];
642 bio
= mpage_bio_submit(WRITE
, bio
);
644 if (mpd
->use_writepage
) {
645 ret
= mapping
->a_ops
->writepage(page
, wbc
);
651 * The caller has a ref on the inode, so *mapping is stable
653 mapping_set_error(mapping
, ret
);
660 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
661 * @mapping: address space structure to write
662 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
663 * @get_block: the filesystem's block mapper function.
664 * If this is NULL then use a_ops->writepage. Otherwise, go
667 * This is a library function, which implements the writepages()
668 * address_space_operation.
670 * If a page is already under I/O, generic_writepages() skips it, even
671 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
672 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
673 * and msync() need to guarantee that all the data which was dirty at the time
674 * the call was made get new I/O started against them. If wbc->sync_mode is
675 * WB_SYNC_ALL then we were called for data integrity and we must wait for
676 * existing IO to complete.
679 mpage_writepages(struct address_space
*mapping
,
680 struct writeback_control
*wbc
, get_block_t get_block
)
682 struct blk_plug plug
;
685 blk_start_plug(&plug
);
688 ret
= generic_writepages(mapping
, wbc
);
690 struct mpage_data mpd
= {
692 .last_block_in_bio
= 0,
693 .get_block
= get_block
,
697 ret
= write_cache_pages(mapping
, wbc
, __mpage_writepage
, &mpd
);
699 mpage_bio_submit(WRITE
, mpd
.bio
);
701 blk_finish_plug(&plug
);
704 EXPORT_SYMBOL(mpage_writepages
);
706 int mpage_writepage(struct page
*page
, get_block_t get_block
,
707 struct writeback_control
*wbc
)
709 struct mpage_data mpd
= {
711 .last_block_in_bio
= 0,
712 .get_block
= get_block
,
715 int ret
= __mpage_writepage(page
, wbc
, &mpd
);
717 mpage_bio_submit(WRITE
, mpd
.bio
);
720 EXPORT_SYMBOL(mpage_writepage
);