Line data Source code
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_page().
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 0 : static void mpage_end_io(struct bio *bio)
47 : {
48 : struct bio_vec *bv;
49 : struct bvec_iter_all iter_all;
50 :
51 0 : bio_for_each_segment_all(bv, bio, iter_all) {
52 0 : struct page *page = bv->bv_page;
53 0 : page_endio(page, bio_op(bio),
54 0 : blk_status_to_errno(bio->bi_status));
55 : }
56 :
57 0 : bio_put(bio);
58 0 : }
59 :
60 : static struct bio *mpage_bio_submit(struct bio *bio)
61 : {
62 0 : bio->bi_end_io = mpage_end_io;
63 0 : guard_bio_eod(bio);
64 0 : 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 readpage 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 0 : map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
80 : {
81 0 : struct inode *inode = page->mapping->host;
82 : struct buffer_head *page_bh, *head;
83 0 : int block = 0;
84 :
85 0 : 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 0 : if (inode->i_blkbits == PAGE_SHIFT &&
91 0 : buffer_uptodate(bh)) {
92 : SetPageUptodate(page);
93 : return;
94 : }
95 0 : create_empty_buffers(page, i_blocksize(inode), 0);
96 : }
97 0 : head = page_buffers(page);
98 0 : page_bh = head;
99 : do {
100 0 : if (block == page_block) {
101 0 : page_bh->b_state = bh->b_state;
102 0 : page_bh->b_bdev = bh->b_bdev;
103 0 : page_bh->b_blocknr = bh->b_blocknr;
104 0 : break;
105 : }
106 0 : page_bh = page_bh->b_this_page;
107 0 : block++;
108 0 : } 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 0 : static struct bio *do_mpage_readpage(struct mpage_readpage_args *args)
132 : {
133 0 : struct page *page = args->page;
134 0 : struct inode *inode = page->mapping->host;
135 0 : const unsigned blkbits = inode->i_blkbits;
136 0 : const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
137 0 : const unsigned blocksize = 1 << blkbits;
138 0 : 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 0 : unsigned first_hole = blocks_per_page;
145 0 : struct block_device *bdev = NULL;
146 : int length;
147 0 : int fully_mapped = 1;
148 0 : int op = REQ_OP_READ;
149 : unsigned nblocks;
150 : unsigned relative_block;
151 0 : gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
152 :
153 0 : if (args->is_readahead) {
154 0 : op |= REQ_RAHEAD;
155 0 : gfp |= __GFP_NORETRY | __GFP_NOWARN;
156 : }
157 :
158 0 : if (page_has_buffers(page))
159 : goto confused;
160 :
161 0 : block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
162 0 : last_block = block_in_file + args->nr_pages * blocks_per_page;
163 0 : last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
164 0 : if (last_block > last_block_in_file)
165 0 : last_block = last_block_in_file;
166 0 : page_block = 0;
167 :
168 : /*
169 : * Map blocks using the result from the previous get_blocks call first.
170 : */
171 0 : nblocks = map_bh->b_size >> blkbits;
172 0 : if (buffer_mapped(map_bh) &&
173 0 : block_in_file > args->first_logical_block &&
174 0 : block_in_file < (args->first_logical_block + nblocks)) {
175 0 : unsigned map_offset = block_in_file - args->first_logical_block;
176 0 : unsigned last = nblocks - map_offset;
177 :
178 0 : for (relative_block = 0; ; relative_block++) {
179 0 : if (relative_block == last) {
180 : clear_buffer_mapped(map_bh);
181 : break;
182 : }
183 0 : if (page_block == blocks_per_page)
184 : break;
185 0 : blocks[page_block] = map_bh->b_blocknr + map_offset +
186 : relative_block;
187 0 : page_block++;
188 0 : block_in_file++;
189 : }
190 0 : bdev = map_bh->b_bdev;
191 : }
192 :
193 : /*
194 : * Then do more get_blocks calls until we are done with this page.
195 : */
196 0 : map_bh->b_page = page;
197 0 : while (page_block < blocks_per_page) {
198 0 : map_bh->b_state = 0;
199 0 : map_bh->b_size = 0;
200 :
201 0 : if (block_in_file < last_block) {
202 0 : map_bh->b_size = (last_block-block_in_file) << blkbits;
203 0 : if (args->get_block(inode, block_in_file, map_bh, 0))
204 : goto confused;
205 0 : args->first_logical_block = block_in_file;
206 : }
207 :
208 0 : if (!buffer_mapped(map_bh)) {
209 0 : fully_mapped = 0;
210 0 : if (first_hole == blocks_per_page)
211 0 : first_hole = page_block;
212 0 : page_block++;
213 0 : block_in_file++;
214 0 : 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 0 : if (buffer_uptodate(map_bh)) {
224 0 : map_buffer_to_page(page, map_bh, page_block);
225 0 : goto confused;
226 : }
227 :
228 0 : if (first_hole != blocks_per_page)
229 : goto confused; /* hole -> non-hole */
230 :
231 : /* Contiguous blocks? */
232 0 : if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
233 : goto confused;
234 0 : nblocks = map_bh->b_size >> blkbits;
235 0 : for (relative_block = 0; ; relative_block++) {
236 0 : if (relative_block == nblocks) {
237 : clear_buffer_mapped(map_bh);
238 : break;
239 0 : } else if (page_block == blocks_per_page)
240 : break;
241 0 : blocks[page_block] = map_bh->b_blocknr+relative_block;
242 0 : page_block++;
243 0 : block_in_file++;
244 : }
245 0 : bdev = map_bh->b_bdev;
246 : }
247 :
248 0 : if (first_hole != blocks_per_page) {
249 0 : zero_user_segment(page, first_hole << blkbits, PAGE_SIZE);
250 0 : if (first_hole == 0) {
251 0 : SetPageUptodate(page);
252 0 : unlock_page(page);
253 0 : goto out;
254 : }
255 0 : } 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 0 : if (args->bio && (args->last_block_in_bio != blocks[0] - 1))
263 0 : args->bio = mpage_bio_submit(args->bio);
264 :
265 : alloc_new:
266 0 : if (args->bio == NULL) {
267 0 : if (first_hole == blocks_per_page) {
268 0 : if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
269 : page))
270 : goto out;
271 : }
272 0 : args->bio = bio_alloc(bdev, bio_max_segs(args->nr_pages), op,
273 : gfp);
274 0 : if (args->bio == NULL)
275 : goto confused;
276 0 : args->bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
277 : }
278 :
279 0 : length = first_hole << blkbits;
280 0 : if (bio_add_page(args->bio, page, length, 0) < length) {
281 0 : args->bio = mpage_bio_submit(args->bio);
282 0 : goto alloc_new;
283 : }
284 :
285 0 : relative_block = block_in_file - args->first_logical_block;
286 0 : nblocks = map_bh->b_size >> blkbits;
287 0 : if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
288 : (first_hole != blocks_per_page))
289 0 : args->bio = mpage_bio_submit(args->bio);
290 : else
291 0 : args->last_block_in_bio = blocks[blocks_per_page - 1];
292 : out:
293 0 : return args->bio;
294 :
295 : confused:
296 0 : if (args->bio)
297 0 : args->bio = mpage_bio_submit(args->bio);
298 0 : if (!PageUptodate(page))
299 0 : block_read_full_page(page, args->get_block);
300 : else
301 0 : 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 0 : void mpage_readahead(struct readahead_control *rac, get_block_t get_block)
345 : {
346 : struct page *page;
347 0 : struct mpage_readpage_args args = {
348 : .get_block = get_block,
349 : .is_readahead = true,
350 : };
351 :
352 0 : while ((page = readahead_page(rac))) {
353 0 : prefetchw(&page->flags);
354 0 : args.page = page;
355 0 : args.nr_pages = readahead_count(rac);
356 0 : args.bio = do_mpage_readpage(&args);
357 0 : put_page(page);
358 : }
359 0 : if (args.bio)
360 0 : mpage_bio_submit(args.bio);
361 0 : }
362 : EXPORT_SYMBOL(mpage_readahead);
363 :
364 : /*
365 : * This isn't called much at all
366 : */
367 0 : int mpage_readpage(struct page *page, get_block_t get_block)
368 : {
369 0 : struct mpage_readpage_args args = {
370 : .page = page,
371 : .nr_pages = 1,
372 : .get_block = get_block,
373 : };
374 :
375 0 : args.bio = do_mpage_readpage(&args);
376 0 : if (args.bio)
377 0 : mpage_bio_submit(args.bio);
378 0 : return 0;
379 : }
380 : EXPORT_SYMBOL(mpage_readpage);
381 :
382 : /*
383 : * Writing is not so simple.
384 : *
385 : * If the page has buffers then they will be used for obtaining the disk
386 : * mapping. We only support pages which are fully mapped-and-dirty, with a
387 : * special case for pages which are unmapped at the end: end-of-file.
388 : *
389 : * If the page has no buffers (preferred) then the page is mapped here.
390 : *
391 : * If all blocks are found to be contiguous then the page can go into the
392 : * BIO. Otherwise fall back to the mapping's writepage().
393 : *
394 : * FIXME: This code wants an estimate of how many pages are still to be
395 : * written, so it can intelligently allocate a suitably-sized BIO. For now,
396 : * just allocate full-size (16-page) BIOs.
397 : */
398 :
399 : struct mpage_data {
400 : struct bio *bio;
401 : sector_t last_block_in_bio;
402 : get_block_t *get_block;
403 : unsigned use_writepage;
404 : };
405 :
406 : /*
407 : * We have our BIO, so we can now mark the buffers clean. Make
408 : * sure to only clean buffers which we know we'll be writing.
409 : */
410 0 : static void clean_buffers(struct page *page, unsigned first_unmapped)
411 : {
412 0 : unsigned buffer_counter = 0;
413 : struct buffer_head *bh, *head;
414 0 : if (!page_has_buffers(page))
415 : return;
416 0 : head = page_buffers(page);
417 0 : bh = head;
418 :
419 : do {
420 0 : if (buffer_counter++ == first_unmapped)
421 : break;
422 0 : clear_buffer_dirty(bh);
423 0 : bh = bh->b_this_page;
424 0 : } while (bh != head);
425 :
426 : /*
427 : * we cannot drop the bh if the page is not uptodate or a concurrent
428 : * readpage would fail to serialize with the bh and it would read from
429 : * disk before we reach the platter.
430 : */
431 0 : if (buffer_heads_over_limit && PageUptodate(page))
432 0 : try_to_free_buffers(page);
433 : }
434 :
435 : /*
436 : * For situations where we want to clean all buffers attached to a page.
437 : * We don't need to calculate how many buffers are attached to the page,
438 : * we just need to specify a number larger than the maximum number of buffers.
439 : */
440 0 : void clean_page_buffers(struct page *page)
441 : {
442 0 : clean_buffers(page, ~0U);
443 0 : }
444 :
445 0 : static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
446 : void *data)
447 : {
448 0 : struct mpage_data *mpd = data;
449 0 : struct bio *bio = mpd->bio;
450 0 : struct address_space *mapping = page->mapping;
451 0 : struct inode *inode = page->mapping->host;
452 0 : const unsigned blkbits = inode->i_blkbits;
453 : unsigned long end_index;
454 0 : const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
455 : sector_t last_block;
456 : sector_t block_in_file;
457 : sector_t blocks[MAX_BUF_PER_PAGE];
458 : unsigned page_block;
459 0 : unsigned first_unmapped = blocks_per_page;
460 0 : struct block_device *bdev = NULL;
461 0 : int boundary = 0;
462 0 : sector_t boundary_block = 0;
463 0 : struct block_device *boundary_bdev = NULL;
464 : int length;
465 : struct buffer_head map_bh;
466 0 : loff_t i_size = i_size_read(inode);
467 0 : int ret = 0;
468 :
469 0 : if (page_has_buffers(page)) {
470 0 : struct buffer_head *head = page_buffers(page);
471 0 : struct buffer_head *bh = head;
472 :
473 : /* If they're all mapped and dirty, do it */
474 0 : page_block = 0;
475 : do {
476 0 : BUG_ON(buffer_locked(bh));
477 0 : if (!buffer_mapped(bh)) {
478 : /*
479 : * unmapped dirty buffers are created by
480 : * block_dirty_folio -> mmapped data
481 : */
482 0 : if (buffer_dirty(bh))
483 : goto confused;
484 0 : if (first_unmapped == blocks_per_page)
485 0 : first_unmapped = page_block;
486 0 : continue;
487 : }
488 :
489 0 : if (first_unmapped != blocks_per_page)
490 : goto confused; /* hole -> non-hole */
491 :
492 0 : if (!buffer_dirty(bh) || !buffer_uptodate(bh))
493 : goto confused;
494 0 : if (page_block) {
495 0 : if (bh->b_blocknr != blocks[page_block-1] + 1)
496 : goto confused;
497 : }
498 0 : blocks[page_block++] = bh->b_blocknr;
499 0 : boundary = buffer_boundary(bh);
500 0 : if (boundary) {
501 0 : boundary_block = bh->b_blocknr;
502 0 : boundary_bdev = bh->b_bdev;
503 : }
504 0 : bdev = bh->b_bdev;
505 0 : } while ((bh = bh->b_this_page) != head);
506 :
507 0 : if (first_unmapped)
508 : goto page_is_mapped;
509 :
510 : /*
511 : * Page has buffers, but they are all unmapped. The page was
512 : * created by pagein or read over a hole which was handled by
513 : * block_read_full_page(). If this address_space is also
514 : * using mpage_readahead then this can rarely happen.
515 : */
516 : goto confused;
517 : }
518 :
519 : /*
520 : * The page has no buffers: map it to disk
521 : */
522 0 : BUG_ON(!PageUptodate(page));
523 0 : block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
524 0 : last_block = (i_size - 1) >> blkbits;
525 0 : map_bh.b_page = page;
526 0 : for (page_block = 0; page_block < blocks_per_page; ) {
527 :
528 0 : map_bh.b_state = 0;
529 0 : map_bh.b_size = 1 << blkbits;
530 0 : if (mpd->get_block(inode, block_in_file, &map_bh, 1))
531 : goto confused;
532 0 : if (buffer_new(&map_bh))
533 0 : clean_bdev_bh_alias(&map_bh);
534 0 : if (buffer_boundary(&map_bh)) {
535 0 : boundary_block = map_bh.b_blocknr;
536 0 : boundary_bdev = map_bh.b_bdev;
537 : }
538 0 : if (page_block) {
539 0 : if (map_bh.b_blocknr != blocks[page_block-1] + 1)
540 : goto confused;
541 : }
542 0 : blocks[page_block++] = map_bh.b_blocknr;
543 0 : boundary = buffer_boundary(&map_bh);
544 0 : bdev = map_bh.b_bdev;
545 0 : if (block_in_file == last_block)
546 : break;
547 0 : block_in_file++;
548 : }
549 0 : BUG_ON(page_block == 0);
550 :
551 : first_unmapped = page_block;
552 :
553 : page_is_mapped:
554 0 : end_index = i_size >> PAGE_SHIFT;
555 0 : if (page->index >= end_index) {
556 : /*
557 : * The page straddles i_size. It must be zeroed out on each
558 : * and every writepage invocation because it may be mmapped.
559 : * "A file is mapped in multiples of the page size. For a file
560 : * that is not a multiple of the page size, the remaining memory
561 : * is zeroed when mapped, and writes to that region are not
562 : * written out to the file."
563 : */
564 0 : unsigned offset = i_size & (PAGE_SIZE - 1);
565 :
566 0 : if (page->index > end_index || !offset)
567 : goto confused;
568 : zero_user_segment(page, offset, PAGE_SIZE);
569 : }
570 :
571 : /*
572 : * This page will go to BIO. Do we need to send this BIO off first?
573 : */
574 0 : if (bio && mpd->last_block_in_bio != blocks[0] - 1)
575 0 : bio = mpage_bio_submit(bio);
576 :
577 : alloc_new:
578 0 : if (bio == NULL) {
579 0 : if (first_unmapped == blocks_per_page) {
580 0 : if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
581 : page, wbc))
582 : goto out;
583 : }
584 0 : bio = bio_alloc(bdev, BIO_MAX_VECS,
585 0 : REQ_OP_WRITE | wbc_to_write_flags(wbc),
586 : GFP_NOFS);
587 0 : bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
588 0 : wbc_init_bio(wbc, bio);
589 : }
590 :
591 : /*
592 : * Must try to add the page before marking the buffer clean or
593 : * the confused fail path above (OOM) will be very confused when
594 : * it finds all bh marked clean (i.e. it will not write anything)
595 : */
596 0 : wbc_account_cgroup_owner(wbc, page, PAGE_SIZE);
597 0 : length = first_unmapped << blkbits;
598 0 : if (bio_add_page(bio, page, length, 0) < length) {
599 0 : bio = mpage_bio_submit(bio);
600 0 : goto alloc_new;
601 : }
602 :
603 0 : clean_buffers(page, first_unmapped);
604 :
605 0 : BUG_ON(PageWriteback(page));
606 0 : set_page_writeback(page);
607 0 : unlock_page(page);
608 0 : if (boundary || (first_unmapped != blocks_per_page)) {
609 0 : bio = mpage_bio_submit(bio);
610 0 : if (boundary_block) {
611 0 : write_boundary_block(boundary_bdev,
612 0 : boundary_block, 1 << blkbits);
613 : }
614 : } else {
615 0 : mpd->last_block_in_bio = blocks[blocks_per_page - 1];
616 : }
617 : goto out;
618 :
619 : confused:
620 0 : if (bio)
621 0 : bio = mpage_bio_submit(bio);
622 :
623 0 : if (mpd->use_writepage) {
624 0 : ret = mapping->a_ops->writepage(page, wbc);
625 : } else {
626 : ret = -EAGAIN;
627 : goto out;
628 : }
629 : /*
630 : * The caller has a ref on the inode, so *mapping is stable
631 : */
632 0 : mapping_set_error(mapping, ret);
633 : out:
634 0 : mpd->bio = bio;
635 0 : return ret;
636 : }
637 :
638 : /**
639 : * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
640 : * @mapping: address space structure to write
641 : * @wbc: subtract the number of written pages from *@wbc->nr_to_write
642 : * @get_block: the filesystem's block mapper function.
643 : * If this is NULL then use a_ops->writepage. Otherwise, go
644 : * direct-to-BIO.
645 : *
646 : * This is a library function, which implements the writepages()
647 : * address_space_operation.
648 : *
649 : * If a page is already under I/O, generic_writepages() skips it, even
650 : * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
651 : * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
652 : * and msync() need to guarantee that all the data which was dirty at the time
653 : * the call was made get new I/O started against them. If wbc->sync_mode is
654 : * WB_SYNC_ALL then we were called for data integrity and we must wait for
655 : * existing IO to complete.
656 : */
657 : int
658 0 : mpage_writepages(struct address_space *mapping,
659 : struct writeback_control *wbc, get_block_t get_block)
660 : {
661 : struct blk_plug plug;
662 : int ret;
663 :
664 0 : blk_start_plug(&plug);
665 :
666 0 : if (!get_block)
667 0 : ret = generic_writepages(mapping, wbc);
668 : else {
669 0 : struct mpage_data mpd = {
670 : .bio = NULL,
671 : .last_block_in_bio = 0,
672 : .get_block = get_block,
673 : .use_writepage = 1,
674 : };
675 :
676 0 : ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
677 0 : if (mpd.bio)
678 0 : mpage_bio_submit(mpd.bio);
679 : }
680 0 : blk_finish_plug(&plug);
681 0 : return ret;
682 : }
683 : EXPORT_SYMBOL(mpage_writepages);
684 :
685 0 : int mpage_writepage(struct page *page, get_block_t get_block,
686 : struct writeback_control *wbc)
687 : {
688 0 : struct mpage_data mpd = {
689 : .bio = NULL,
690 : .last_block_in_bio = 0,
691 : .get_block = get_block,
692 : .use_writepage = 0,
693 : };
694 0 : int ret = __mpage_writepage(page, wbc, &mpd);
695 0 : if (mpd.bio)
696 0 : mpage_bio_submit(mpd.bio);
697 0 : return ret;
698 : }
699 : EXPORT_SYMBOL(mpage_writepage);
|
