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_readpages. 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 min_t(int, args->nr_pages,
310 if (args->bio == NULL)
314 length = first_hole << blkbits;
315 if (bio_add_page(args->bio, page, length, 0) < length) {
316 args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
320 relative_block = block_in_file - args->first_logical_block;
321 nblocks = map_bh->b_size >> blkbits;
322 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
323 (first_hole != blocks_per_page))
324 args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
326 args->last_block_in_bio = blocks[blocks_per_page - 1];
332 args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
333 if (!PageUptodate(page))
334 block_read_full_page(page, args->get_block);
341 * mpage_readpages - populate an address space with some pages & start reads against them
342 * @mapping: the address_space
343 * @pages: The address of a list_head which contains the target pages. These
344 * pages have their ->index populated and are otherwise uninitialised.
345 * The page at @pages->prev has the lowest file offset, and reads should be
346 * issued in @pages->prev to @pages->next order.
347 * @nr_pages: The number of pages at *@pages
348 * @get_block: The filesystem's block mapper function.
350 * This function walks the pages and the blocks within each page, building and
351 * emitting large BIOs.
353 * If anything unusual happens, such as:
355 * - encountering a page which has buffers
356 * - encountering a page which has a non-hole after a hole
357 * - encountering a page with non-contiguous blocks
359 * then this code just gives up and calls the buffer_head-based read function.
360 * It does handle a page which has holes at the end - that is a common case:
361 * the end-of-file on blocksize < PAGE_SIZE setups.
363 * BH_Boundary explanation:
365 * There is a problem. The mpage read code assembles several pages, gets all
366 * their disk mappings, and then submits them all. That's fine, but obtaining
367 * the disk mappings may require I/O. Reads of indirect blocks, for example.
369 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
370 * submitted in the following order:
372 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
374 * because the indirect block has to be read to get the mappings of blocks
375 * 13,14,15,16. Obviously, this impacts performance.
377 * So what we do it to allow the filesystem's get_block() function to set
378 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
379 * after this one will require I/O against a block which is probably close to
380 * this one. So you should push what I/O you have currently accumulated.
382 * This all causes the disk requests to be issued in the correct order.
385 mpage_readpages(struct address_space *mapping, struct list_head *pages,
386 unsigned nr_pages, get_block_t get_block)
388 struct mpage_readpage_args args = {
389 .get_block = get_block,
390 .is_readahead = true,
394 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
395 struct page *page = lru_to_page(pages);
397 prefetchw(&page->flags);
398 list_del(&page->lru);
399 if (!add_to_page_cache_lru(page, mapping,
401 readahead_gfp_mask(mapping))) {
403 args.nr_pages = nr_pages - page_idx;
404 args.bio = do_mpage_readpage(&args);
408 BUG_ON(!list_empty(pages));
410 mpage_bio_submit(REQ_OP_READ, REQ_RAHEAD, args.bio);
413 EXPORT_SYMBOL(mpage_readpages);
416 * This isn't called much at all
418 int mpage_readpage(struct page *page, get_block_t get_block)
420 struct mpage_readpage_args args = {
423 .get_block = get_block,
426 args.bio = do_mpage_readpage(&args);
428 mpage_bio_submit(REQ_OP_READ, 0, args.bio);
431 EXPORT_SYMBOL(mpage_readpage);
434 * Writing is not so simple.
436 * If the page has buffers then they will be used for obtaining the disk
437 * mapping. We only support pages which are fully mapped-and-dirty, with a
438 * special case for pages which are unmapped at the end: end-of-file.
440 * If the page has no buffers (preferred) then the page is mapped here.
442 * If all blocks are found to be contiguous then the page can go into the
443 * BIO. Otherwise fall back to the mapping's writepage().
445 * FIXME: This code wants an estimate of how many pages are still to be
446 * written, so it can intelligently allocate a suitably-sized BIO. For now,
447 * just allocate full-size (16-page) BIOs.
452 sector_t last_block_in_bio;
453 get_block_t *get_block;
454 unsigned use_writepage;
458 * We have our BIO, so we can now mark the buffers clean. Make
459 * sure to only clean buffers which we know we'll be writing.
461 static void clean_buffers(struct page *page, unsigned first_unmapped)
463 unsigned buffer_counter = 0;
464 struct buffer_head *bh, *head;
465 if (!page_has_buffers(page))
467 head = page_buffers(page);
471 if (buffer_counter++ == first_unmapped)
473 clear_buffer_dirty(bh);
474 bh = bh->b_this_page;
475 } while (bh != head);
478 * we cannot drop the bh if the page is not uptodate or a concurrent
479 * readpage would fail to serialize with the bh and it would read from
480 * disk before we reach the platter.
482 if (buffer_heads_over_limit && PageUptodate(page))
483 try_to_free_buffers(page);
487 * For situations where we want to clean all buffers attached to a page.
488 * We don't need to calculate how many buffers are attached to the page,
489 * we just need to specify a number larger than the maximum number of buffers.
491 void clean_page_buffers(struct page *page)
493 clean_buffers(page, ~0U);
496 static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
499 struct mpage_data *mpd = data;
500 struct bio *bio = mpd->bio;
501 struct address_space *mapping = page->mapping;
502 struct inode *inode = page->mapping->host;
503 const unsigned blkbits = inode->i_blkbits;
504 unsigned long end_index;
505 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
507 sector_t block_in_file;
508 sector_t blocks[MAX_BUF_PER_PAGE];
510 unsigned first_unmapped = blocks_per_page;
511 struct block_device *bdev = NULL;
513 sector_t boundary_block = 0;
514 struct block_device *boundary_bdev = NULL;
516 struct buffer_head map_bh;
517 loff_t i_size = i_size_read(inode);
519 int op_flags = wbc_to_write_flags(wbc);
521 if (page_has_buffers(page)) {
522 struct buffer_head *head = page_buffers(page);
523 struct buffer_head *bh = head;
525 /* If they're all mapped and dirty, do it */
528 BUG_ON(buffer_locked(bh));
529 if (!buffer_mapped(bh)) {
531 * unmapped dirty buffers are created by
532 * __set_page_dirty_buffers -> mmapped data
534 if (buffer_dirty(bh))
536 if (first_unmapped == blocks_per_page)
537 first_unmapped = page_block;
541 if (first_unmapped != blocks_per_page)
542 goto confused; /* hole -> non-hole */
544 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
547 if (bh->b_blocknr != blocks[page_block-1] + 1)
550 blocks[page_block++] = bh->b_blocknr;
551 boundary = buffer_boundary(bh);
553 boundary_block = bh->b_blocknr;
554 boundary_bdev = bh->b_bdev;
557 } while ((bh = bh->b_this_page) != head);
563 * Page has buffers, but they are all unmapped. The page was
564 * created by pagein or read over a hole which was handled by
565 * block_read_full_page(). If this address_space is also
566 * using mpage_readpages then this can rarely happen.
572 * The page has no buffers: map it to disk
574 BUG_ON(!PageUptodate(page));
575 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
576 last_block = (i_size - 1) >> blkbits;
577 map_bh.b_page = page;
578 for (page_block = 0; page_block < blocks_per_page; ) {
581 map_bh.b_size = 1 << blkbits;
582 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
584 if (buffer_new(&map_bh))
585 clean_bdev_bh_alias(&map_bh);
586 if (buffer_boundary(&map_bh)) {
587 boundary_block = map_bh.b_blocknr;
588 boundary_bdev = map_bh.b_bdev;
591 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
594 blocks[page_block++] = map_bh.b_blocknr;
595 boundary = buffer_boundary(&map_bh);
596 bdev = map_bh.b_bdev;
597 if (block_in_file == last_block)
601 BUG_ON(page_block == 0);
603 first_unmapped = page_block;
606 end_index = i_size >> PAGE_SHIFT;
607 if (page->index >= end_index) {
609 * The page straddles i_size. It must be zeroed out on each
610 * and every writepage invocation because it may be mmapped.
611 * "A file is mapped in multiples of the page size. For a file
612 * that is not a multiple of the page size, the remaining memory
613 * is zeroed when mapped, and writes to that region are not
614 * written out to the file."
616 unsigned offset = i_size & (PAGE_SIZE - 1);
618 if (page->index > end_index || !offset)
620 zero_user_segment(page, offset, PAGE_SIZE);
624 * This page will go to BIO. Do we need to send this BIO off first?
626 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
627 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
631 if (first_unmapped == blocks_per_page) {
632 if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
636 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
637 BIO_MAX_PAGES, GFP_NOFS|__GFP_HIGH);
641 wbc_init_bio(wbc, bio);
642 bio->bi_write_hint = inode->i_write_hint;
646 * Must try to add the page before marking the buffer clean or
647 * the confused fail path above (OOM) will be very confused when
648 * it finds all bh marked clean (i.e. it will not write anything)
650 wbc_account_cgroup_owner(wbc, page, PAGE_SIZE);
651 length = first_unmapped << blkbits;
652 if (bio_add_page(bio, page, length, 0) < length) {
653 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
657 clean_buffers(page, first_unmapped);
659 BUG_ON(PageWriteback(page));
660 set_page_writeback(page);
662 if (boundary || (first_unmapped != blocks_per_page)) {
663 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
664 if (boundary_block) {
665 write_boundary_block(boundary_bdev,
666 boundary_block, 1 << blkbits);
669 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
675 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
677 if (mpd->use_writepage) {
678 ret = mapping->a_ops->writepage(page, wbc);
684 * The caller has a ref on the inode, so *mapping is stable
686 mapping_set_error(mapping, ret);
693 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
694 * @mapping: address space structure to write
695 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
696 * @get_block: the filesystem's block mapper function.
697 * If this is NULL then use a_ops->writepage. Otherwise, go
700 * This is a library function, which implements the writepages()
701 * address_space_operation.
703 * If a page is already under I/O, generic_writepages() skips it, even
704 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
705 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
706 * and msync() need to guarantee that all the data which was dirty at the time
707 * the call was made get new I/O started against them. If wbc->sync_mode is
708 * WB_SYNC_ALL then we were called for data integrity and we must wait for
709 * existing IO to complete.
712 mpage_writepages(struct address_space *mapping,
713 struct writeback_control *wbc, get_block_t get_block)
715 struct blk_plug plug;
718 blk_start_plug(&plug);
721 ret = generic_writepages(mapping, wbc);
723 struct mpage_data mpd = {
725 .last_block_in_bio = 0,
726 .get_block = get_block,
730 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
732 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
734 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
737 blk_finish_plug(&plug);
740 EXPORT_SYMBOL(mpage_writepages);
742 int mpage_writepage(struct page *page, get_block_t get_block,
743 struct writeback_control *wbc)
745 struct mpage_data mpd = {
747 .last_block_in_bio = 0,
748 .get_block = get_block,
751 int ret = __mpage_writepage(page, wbc, &mpd);
753 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
755 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
759 EXPORT_SYMBOL(mpage_writepage);