]> git.ipfire.org Git - people/ms/linux.git/blob - fs/mpage.c
Merge tag 'random-5.19-rc4-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[people/ms/linux.git] / fs / mpage.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * fs/mpage.c
4 *
5 * Copyright (C) 2002, Linus Torvalds.
6 *
7 * Contains functions related to preparing and submitting BIOs which contain
8 * multiple pagecache pages.
9 *
10 * 15May2002 Andrew Morton
11 * Initial version
12 * 27Jun2002 axboe@suse.de
13 * use bio_add_page() to build bio's just the right size
14 */
15
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/mm.h>
19 #include <linux/kdev_t.h>
20 #include <linux/gfp.h>
21 #include <linux/bio.h>
22 #include <linux/fs.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 "internal.h"
33
34 /*
35 * I/O completion handler for multipage BIOs.
36 *
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().
40 *
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.
45 */
46 static void mpage_end_io(struct bio *bio)
47 {
48 struct bio_vec *bv;
49 struct bvec_iter_all iter_all;
50
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));
55 }
56
57 bio_put(bio);
58 }
59
60 static struct bio *mpage_bio_submit(struct bio *bio)
61 {
62 bio->bi_end_io = mpage_end_io;
63 guard_bio_eod(bio);
64 submit_bio(bio);
65 return NULL;
66 }
67
68 /*
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
72 * to get_block.
73 *
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.
77 */
78 static void
79 map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
80 {
81 struct inode *inode = page->mapping->host;
82 struct buffer_head *page_bh, *head;
83 int block = 0;
84
85 if (!page_has_buffers(page)) {
86 /*
87 * don't make any buffers if there is only one buffer on
88 * the page and the page just needs to be set up to date
89 */
90 if (inode->i_blkbits == PAGE_SHIFT &&
91 buffer_uptodate(bh)) {
92 SetPageUptodate(page);
93 return;
94 }
95 create_empty_buffers(page, i_blocksize(inode), 0);
96 }
97 head = page_buffers(page);
98 page_bh = head;
99 do {
100 if (block == page_block) {
101 page_bh->b_state = bh->b_state;
102 page_bh->b_bdev = bh->b_bdev;
103 page_bh->b_blocknr = bh->b_blocknr;
104 break;
105 }
106 page_bh = page_bh->b_this_page;
107 block++;
108 } while (page_bh != head);
109 }
110
111 struct mpage_readpage_args {
112 struct bio *bio;
113 struct page *page;
114 unsigned int nr_pages;
115 bool is_readahead;
116 sector_t last_block_in_bio;
117 struct buffer_head map_bh;
118 unsigned long first_logical_block;
119 get_block_t *get_block;
120 };
121
122 /*
123 * This is the worker routine which does all the work of mapping the disk
124 * blocks and constructs largest possible bios, submits them for IO if the
125 * blocks are not contiguous on the disk.
126 *
127 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
128 * represent the validity of its disk mapping and to decide when to do the next
129 * get_block() call.
130 */
131 static struct bio *do_mpage_readpage(struct mpage_readpage_args *args)
132 {
133 struct page *page = args->page;
134 struct inode *inode = page->mapping->host;
135 const unsigned blkbits = inode->i_blkbits;
136 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
137 const unsigned blocksize = 1 << blkbits;
138 struct buffer_head *map_bh = &args->map_bh;
139 sector_t block_in_file;
140 sector_t last_block;
141 sector_t last_block_in_file;
142 sector_t blocks[MAX_BUF_PER_PAGE];
143 unsigned page_block;
144 unsigned first_hole = blocks_per_page;
145 struct block_device *bdev = NULL;
146 int length;
147 int fully_mapped = 1;
148 int op = REQ_OP_READ;
149 unsigned nblocks;
150 unsigned relative_block;
151 gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
152
153 if (args->is_readahead) {
154 op |= REQ_RAHEAD;
155 gfp |= __GFP_NORETRY | __GFP_NOWARN;
156 }
157
158 if (page_has_buffers(page))
159 goto confused;
160
161 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
162 last_block = block_in_file + args->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;
166 page_block = 0;
167
168 /*
169 * Map blocks using the result from the previous get_blocks call first.
170 */
171 nblocks = map_bh->b_size >> blkbits;
172 if (buffer_mapped(map_bh) &&
173 block_in_file > args->first_logical_block &&
174 block_in_file < (args->first_logical_block + nblocks)) {
175 unsigned map_offset = block_in_file - args->first_logical_block;
176 unsigned last = nblocks - map_offset;
177
178 for (relative_block = 0; ; relative_block++) {
179 if (relative_block == last) {
180 clear_buffer_mapped(map_bh);
181 break;
182 }
183 if (page_block == blocks_per_page)
184 break;
185 blocks[page_block] = map_bh->b_blocknr + map_offset +
186 relative_block;
187 page_block++;
188 block_in_file++;
189 }
190 bdev = map_bh->b_bdev;
191 }
192
193 /*
194 * Then do more get_blocks calls until we are done with this page.
195 */
196 map_bh->b_page = page;
197 while (page_block < blocks_per_page) {
198 map_bh->b_state = 0;
199 map_bh->b_size = 0;
200
201 if (block_in_file < last_block) {
202 map_bh->b_size = (last_block-block_in_file) << blkbits;
203 if (args->get_block(inode, block_in_file, map_bh, 0))
204 goto confused;
205 args->first_logical_block = block_in_file;
206 }
207
208 if (!buffer_mapped(map_bh)) {
209 fully_mapped = 0;
210 if (first_hole == blocks_per_page)
211 first_hole = page_block;
212 page_block++;
213 block_in_file++;
214 continue;
215 }
216
217 /* some filesystems will copy data into the page during
218 * the get_block call, in which case we don't want to
219 * read it again. map_buffer_to_page copies the data
220 * we just collected from get_block into the page's buffers
221 * so readpage doesn't have to repeat the get_block call
222 */
223 if (buffer_uptodate(map_bh)) {
224 map_buffer_to_page(page, map_bh, page_block);
225 goto confused;
226 }
227
228 if (first_hole != blocks_per_page)
229 goto confused; /* hole -> non-hole */
230
231 /* Contiguous blocks? */
232 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
233 goto confused;
234 nblocks = map_bh->b_size >> blkbits;
235 for (relative_block = 0; ; relative_block++) {
236 if (relative_block == nblocks) {
237 clear_buffer_mapped(map_bh);
238 break;
239 } else if (page_block == blocks_per_page)
240 break;
241 blocks[page_block] = map_bh->b_blocknr+relative_block;
242 page_block++;
243 block_in_file++;
244 }
245 bdev = map_bh->b_bdev;
246 }
247
248 if (first_hole != blocks_per_page) {
249 zero_user_segment(page, first_hole << blkbits, PAGE_SIZE);
250 if (first_hole == 0) {
251 SetPageUptodate(page);
252 unlock_page(page);
253 goto out;
254 }
255 } else if (fully_mapped) {
256 SetPageMappedToDisk(page);
257 }
258
259 /*
260 * This page will go to BIO. Do we need to send this BIO off first?
261 */
262 if (args->bio && (args->last_block_in_bio != blocks[0] - 1))
263 args->bio = mpage_bio_submit(args->bio);
264
265 alloc_new:
266 if (args->bio == NULL) {
267 if (first_hole == blocks_per_page) {
268 if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
269 page))
270 goto out;
271 }
272 args->bio = bio_alloc(bdev, bio_max_segs(args->nr_pages), op,
273 gfp);
274 if (args->bio == NULL)
275 goto confused;
276 args->bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
277 }
278
279 length = first_hole << blkbits;
280 if (bio_add_page(args->bio, page, length, 0) < length) {
281 args->bio = mpage_bio_submit(args->bio);
282 goto alloc_new;
283 }
284
285 relative_block = block_in_file - args->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 args->bio = mpage_bio_submit(args->bio);
290 else
291 args->last_block_in_bio = blocks[blocks_per_page - 1];
292 out:
293 return args->bio;
294
295 confused:
296 if (args->bio)
297 args->bio = mpage_bio_submit(args->bio);
298 if (!PageUptodate(page))
299 block_read_full_folio(page_folio(page), args->get_block);
300 else
301 unlock_page(page);
302 goto out;
303 }
304
305 /**
306 * mpage_readahead - start reads against pages
307 * @rac: Describes which pages to read.
308 * @get_block: The filesystem's block mapper function.
309 *
310 * This function walks the pages and the blocks within each page, building and
311 * emitting large BIOs.
312 *
313 * If anything unusual happens, such as:
314 *
315 * - encountering a page which has buffers
316 * - encountering a page which has a non-hole after a hole
317 * - encountering a page with non-contiguous blocks
318 *
319 * then this code just gives up and calls the buffer_head-based read function.
320 * It does handle a page which has holes at the end - that is a common case:
321 * the end-of-file on blocksize < PAGE_SIZE setups.
322 *
323 * BH_Boundary explanation:
324 *
325 * There is a problem. The mpage read code assembles several pages, gets all
326 * their disk mappings, and then submits them all. That's fine, but obtaining
327 * the disk mappings may require I/O. Reads of indirect blocks, for example.
328 *
329 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
330 * submitted in the following order:
331 *
332 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
333 *
334 * because the indirect block has to be read to get the mappings of blocks
335 * 13,14,15,16. Obviously, this impacts performance.
336 *
337 * So what we do it to allow the filesystem's get_block() function to set
338 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
339 * after this one will require I/O against a block which is probably close to
340 * this one. So you should push what I/O you have currently accumulated.
341 *
342 * This all causes the disk requests to be issued in the correct order.
343 */
344 void mpage_readahead(struct readahead_control *rac, get_block_t get_block)
345 {
346 struct page *page;
347 struct mpage_readpage_args args = {
348 .get_block = get_block,
349 .is_readahead = true,
350 };
351
352 while ((page = readahead_page(rac))) {
353 prefetchw(&page->flags);
354 args.page = page;
355 args.nr_pages = readahead_count(rac);
356 args.bio = do_mpage_readpage(&args);
357 put_page(page);
358 }
359 if (args.bio)
360 mpage_bio_submit(args.bio);
361 }
362 EXPORT_SYMBOL(mpage_readahead);
363
364 /*
365 * This isn't called much at all
366 */
367 int mpage_read_folio(struct folio *folio, get_block_t get_block)
368 {
369 struct mpage_readpage_args args = {
370 .page = &folio->page,
371 .nr_pages = 1,
372 .get_block = get_block,
373 };
374
375 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
376
377 args.bio = do_mpage_readpage(&args);
378 if (args.bio)
379 mpage_bio_submit(args.bio);
380 return 0;
381 }
382 EXPORT_SYMBOL(mpage_read_folio);
383
384 /*
385 * Writing is not so simple.
386 *
387 * If the page has buffers then they will be used for obtaining the disk
388 * mapping. We only support pages which are fully mapped-and-dirty, with a
389 * special case for pages which are unmapped at the end: end-of-file.
390 *
391 * If the page has no buffers (preferred) then the page is mapped here.
392 *
393 * If all blocks are found to be contiguous then the page can go into the
394 * BIO. Otherwise fall back to the mapping's writepage().
395 *
396 * FIXME: This code wants an estimate of how many pages are still to be
397 * written, so it can intelligently allocate a suitably-sized BIO. For now,
398 * just allocate full-size (16-page) BIOs.
399 */
400
401 struct mpage_data {
402 struct bio *bio;
403 sector_t last_block_in_bio;
404 get_block_t *get_block;
405 unsigned use_writepage;
406 };
407
408 /*
409 * We have our BIO, so we can now mark the buffers clean. Make
410 * sure to only clean buffers which we know we'll be writing.
411 */
412 static void clean_buffers(struct page *page, unsigned first_unmapped)
413 {
414 unsigned buffer_counter = 0;
415 struct buffer_head *bh, *head;
416 if (!page_has_buffers(page))
417 return;
418 head = page_buffers(page);
419 bh = head;
420
421 do {
422 if (buffer_counter++ == first_unmapped)
423 break;
424 clear_buffer_dirty(bh);
425 bh = bh->b_this_page;
426 } while (bh != head);
427
428 /*
429 * we cannot drop the bh if the page is not uptodate or a concurrent
430 * read_folio would fail to serialize with the bh and it would read from
431 * disk before we reach the platter.
432 */
433 if (buffer_heads_over_limit && PageUptodate(page))
434 try_to_free_buffers(page_folio(page));
435 }
436
437 /*
438 * For situations where we want to clean all buffers attached to a page.
439 * We don't need to calculate how many buffers are attached to the page,
440 * we just need to specify a number larger than the maximum number of buffers.
441 */
442 void clean_page_buffers(struct page *page)
443 {
444 clean_buffers(page, ~0U);
445 }
446
447 static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
448 void *data)
449 {
450 struct mpage_data *mpd = data;
451 struct bio *bio = mpd->bio;
452 struct address_space *mapping = page->mapping;
453 struct inode *inode = page->mapping->host;
454 const unsigned blkbits = inode->i_blkbits;
455 unsigned long end_index;
456 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
457 sector_t last_block;
458 sector_t block_in_file;
459 sector_t blocks[MAX_BUF_PER_PAGE];
460 unsigned page_block;
461 unsigned first_unmapped = blocks_per_page;
462 struct block_device *bdev = NULL;
463 int boundary = 0;
464 sector_t boundary_block = 0;
465 struct block_device *boundary_bdev = NULL;
466 int length;
467 struct buffer_head map_bh;
468 loff_t i_size = i_size_read(inode);
469 int ret = 0;
470
471 if (page_has_buffers(page)) {
472 struct buffer_head *head = page_buffers(page);
473 struct buffer_head *bh = head;
474
475 /* If they're all mapped and dirty, do it */
476 page_block = 0;
477 do {
478 BUG_ON(buffer_locked(bh));
479 if (!buffer_mapped(bh)) {
480 /*
481 * unmapped dirty buffers are created by
482 * block_dirty_folio -> mmapped data
483 */
484 if (buffer_dirty(bh))
485 goto confused;
486 if (first_unmapped == blocks_per_page)
487 first_unmapped = page_block;
488 continue;
489 }
490
491 if (first_unmapped != blocks_per_page)
492 goto confused; /* hole -> non-hole */
493
494 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
495 goto confused;
496 if (page_block) {
497 if (bh->b_blocknr != blocks[page_block-1] + 1)
498 goto confused;
499 }
500 blocks[page_block++] = bh->b_blocknr;
501 boundary = buffer_boundary(bh);
502 if (boundary) {
503 boundary_block = bh->b_blocknr;
504 boundary_bdev = bh->b_bdev;
505 }
506 bdev = bh->b_bdev;
507 } while ((bh = bh->b_this_page) != head);
508
509 if (first_unmapped)
510 goto page_is_mapped;
511
512 /*
513 * Page has buffers, but they are all unmapped. The page was
514 * created by pagein or read over a hole which was handled by
515 * block_read_full_folio(). If this address_space is also
516 * using mpage_readahead then this can rarely happen.
517 */
518 goto confused;
519 }
520
521 /*
522 * The page has no buffers: map it to disk
523 */
524 BUG_ON(!PageUptodate(page));
525 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
526 last_block = (i_size - 1) >> blkbits;
527 map_bh.b_page = page;
528 for (page_block = 0; page_block < blocks_per_page; ) {
529
530 map_bh.b_state = 0;
531 map_bh.b_size = 1 << blkbits;
532 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
533 goto confused;
534 if (buffer_new(&map_bh))
535 clean_bdev_bh_alias(&map_bh);
536 if (buffer_boundary(&map_bh)) {
537 boundary_block = map_bh.b_blocknr;
538 boundary_bdev = map_bh.b_bdev;
539 }
540 if (page_block) {
541 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
542 goto confused;
543 }
544 blocks[page_block++] = map_bh.b_blocknr;
545 boundary = buffer_boundary(&map_bh);
546 bdev = map_bh.b_bdev;
547 if (block_in_file == last_block)
548 break;
549 block_in_file++;
550 }
551 BUG_ON(page_block == 0);
552
553 first_unmapped = page_block;
554
555 page_is_mapped:
556 end_index = i_size >> PAGE_SHIFT;
557 if (page->index >= end_index) {
558 /*
559 * The page straddles i_size. It must be zeroed out on each
560 * and every writepage invocation because it may be mmapped.
561 * "A file is mapped in multiples of the page size. For a file
562 * that is not a multiple of the page size, the remaining memory
563 * is zeroed when mapped, and writes to that region are not
564 * written out to the file."
565 */
566 unsigned offset = i_size & (PAGE_SIZE - 1);
567
568 if (page->index > end_index || !offset)
569 goto confused;
570 zero_user_segment(page, offset, PAGE_SIZE);
571 }
572
573 /*
574 * This page will go to BIO. Do we need to send this BIO off first?
575 */
576 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
577 bio = mpage_bio_submit(bio);
578
579 alloc_new:
580 if (bio == NULL) {
581 if (first_unmapped == blocks_per_page) {
582 if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
583 page, wbc))
584 goto out;
585 }
586 bio = bio_alloc(bdev, BIO_MAX_VECS,
587 REQ_OP_WRITE | wbc_to_write_flags(wbc),
588 GFP_NOFS);
589 bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
590 wbc_init_bio(wbc, bio);
591 }
592
593 /*
594 * Must try to add the page before marking the buffer clean or
595 * the confused fail path above (OOM) will be very confused when
596 * it finds all bh marked clean (i.e. it will not write anything)
597 */
598 wbc_account_cgroup_owner(wbc, page, PAGE_SIZE);
599 length = first_unmapped << blkbits;
600 if (bio_add_page(bio, page, length, 0) < length) {
601 bio = mpage_bio_submit(bio);
602 goto alloc_new;
603 }
604
605 clean_buffers(page, first_unmapped);
606
607 BUG_ON(PageWriteback(page));
608 set_page_writeback(page);
609 unlock_page(page);
610 if (boundary || (first_unmapped != blocks_per_page)) {
611 bio = mpage_bio_submit(bio);
612 if (boundary_block) {
613 write_boundary_block(boundary_bdev,
614 boundary_block, 1 << blkbits);
615 }
616 } else {
617 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
618 }
619 goto out;
620
621 confused:
622 if (bio)
623 bio = mpage_bio_submit(bio);
624
625 if (mpd->use_writepage) {
626 ret = mapping->a_ops->writepage(page, wbc);
627 } else {
628 ret = -EAGAIN;
629 goto out;
630 }
631 /*
632 * The caller has a ref on the inode, so *mapping is stable
633 */
634 mapping_set_error(mapping, ret);
635 out:
636 mpd->bio = bio;
637 return ret;
638 }
639
640 /**
641 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
642 * @mapping: address space structure to write
643 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
644 * @get_block: the filesystem's block mapper function.
645 * If this is NULL then use a_ops->writepage. Otherwise, go
646 * direct-to-BIO.
647 *
648 * This is a library function, which implements the writepages()
649 * address_space_operation.
650 *
651 * If a page is already under I/O, generic_writepages() skips it, even
652 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
653 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
654 * and msync() need to guarantee that all the data which was dirty at the time
655 * the call was made get new I/O started against them. If wbc->sync_mode is
656 * WB_SYNC_ALL then we were called for data integrity and we must wait for
657 * existing IO to complete.
658 */
659 int
660 mpage_writepages(struct address_space *mapping,
661 struct writeback_control *wbc, get_block_t get_block)
662 {
663 struct blk_plug plug;
664 int ret;
665
666 blk_start_plug(&plug);
667
668 if (!get_block)
669 ret = generic_writepages(mapping, wbc);
670 else {
671 struct mpage_data mpd = {
672 .bio = NULL,
673 .last_block_in_bio = 0,
674 .get_block = get_block,
675 .use_writepage = 1,
676 };
677
678 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
679 if (mpd.bio)
680 mpage_bio_submit(mpd.bio);
681 }
682 blk_finish_plug(&plug);
683 return ret;
684 }
685 EXPORT_SYMBOL(mpage_writepages);
686
687 int mpage_writepage(struct page *page, get_block_t get_block,
688 struct writeback_control *wbc)
689 {
690 struct mpage_data mpd = {
691 .bio = NULL,
692 .last_block_in_bio = 0,
693 .get_block = get_block,
694 .use_writepage = 0,
695 };
696 int ret = __mpage_writepage(page, wbc, &mpd);
697 if (mpd.bio)
698 mpage_bio_submit(mpd.bio);
699 return ret;
700 }
701 EXPORT_SYMBOL(mpage_writepage);