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 <linux/cleancache.h>
33 : #include "internal.h"
34 :
35 : /*
36 : * I/O completion handler for multipage BIOs.
37 : *
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().
41 : *
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.
46 : */
47 72 : static void mpage_end_io(struct bio *bio)
48 : {
49 72 : struct bio_vec *bv;
50 72 : struct bvec_iter_all iter_all;
51 :
52 589 : bio_for_each_segment_all(bv, bio, iter_all) {
53 517 : struct page *page = bv->bv_page;
54 517 : page_endio(page, bio_op(bio),
55 517 : blk_status_to_errno(bio->bi_status));
56 : }
57 :
58 72 : bio_put(bio);
59 72 : }
60 :
61 72 : static struct bio *mpage_bio_submit(int op, int op_flags, struct bio *bio)
62 : {
63 72 : bio->bi_end_io = mpage_end_io;
64 72 : bio_set_op_attrs(bio, op, op_flags);
65 72 : guard_bio_eod(bio);
66 0 : submit_bio(bio);
67 72 : return NULL;
68 : }
69 :
70 : static struct bio *
71 72 : mpage_alloc(struct block_device *bdev,
72 : sector_t first_sector, int nr_vecs,
73 : gfp_t gfp_flags)
74 : {
75 72 : struct bio *bio;
76 :
77 : /* Restrict the given (page cache) mask for slab allocations */
78 72 : gfp_flags &= GFP_KERNEL;
79 72 : bio = bio_alloc(gfp_flags, nr_vecs);
80 :
81 72 : if (bio == NULL && (current->flags & PF_MEMALLOC)) {
82 0 : while (!bio && (nr_vecs /= 2))
83 0 : bio = bio_alloc(gfp_flags, nr_vecs);
84 : }
85 :
86 72 : if (bio) {
87 72 : bio_set_dev(bio, bdev);
88 72 : bio->bi_iter.bi_sector = first_sector;
89 : }
90 72 : return bio;
91 : }
92 :
93 : /*
94 : * support function for mpage_readahead. 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
97 : * to get_block.
98 : *
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.
102 : */
103 : static void
104 0 : map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
105 : {
106 0 : struct inode *inode = page->mapping->host;
107 0 : struct buffer_head *page_bh, *head;
108 0 : int block = 0;
109 :
110 0 : if (!page_has_buffers(page)) {
111 : /*
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
114 : */
115 0 : if (inode->i_blkbits == PAGE_SHIFT &&
116 0 : buffer_uptodate(bh)) {
117 0 : SetPageUptodate(page);
118 0 : return;
119 : }
120 0 : create_empty_buffers(page, i_blocksize(inode), 0);
121 : }
122 0 : head = page_buffers(page);
123 0 : page_bh = head;
124 0 : do {
125 0 : if (block == page_block) {
126 0 : page_bh->b_state = bh->b_state;
127 0 : page_bh->b_bdev = bh->b_bdev;
128 0 : page_bh->b_blocknr = bh->b_blocknr;
129 0 : break;
130 : }
131 0 : page_bh = page_bh->b_this_page;
132 0 : block++;
133 0 : } while (page_bh != head);
134 : }
135 :
136 : struct mpage_readpage_args {
137 : struct bio *bio;
138 : struct page *page;
139 : unsigned int nr_pages;
140 : bool is_readahead;
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;
145 : };
146 :
147 : /*
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.
151 : *
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
154 : * get_block() call.
155 : */
156 517 : static struct bio *do_mpage_readpage(struct mpage_readpage_args *args)
157 : {
158 517 : struct page *page = args->page;
159 517 : struct inode *inode = page->mapping->host;
160 517 : const unsigned blkbits = inode->i_blkbits;
161 517 : const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
162 517 : const unsigned blocksize = 1 << blkbits;
163 517 : struct buffer_head *map_bh = &args->map_bh;
164 517 : sector_t block_in_file;
165 517 : sector_t last_block;
166 517 : sector_t last_block_in_file;
167 517 : sector_t blocks[MAX_BUF_PER_PAGE];
168 517 : unsigned page_block;
169 517 : unsigned first_hole = blocks_per_page;
170 517 : struct block_device *bdev = NULL;
171 517 : int length;
172 517 : int fully_mapped = 1;
173 517 : int op_flags;
174 517 : unsigned nblocks;
175 517 : unsigned relative_block;
176 517 : gfp_t gfp;
177 :
178 517 : if (args->is_readahead) {
179 517 : op_flags = REQ_RAHEAD;
180 517 : gfp = readahead_gfp_mask(page->mapping);
181 : } else {
182 0 : op_flags = 0;
183 0 : gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
184 : }
185 :
186 517 : if (page_has_buffers(page))
187 0 : goto confused;
188 :
189 517 : block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
190 517 : last_block = block_in_file + args->nr_pages * blocks_per_page;
191 517 : last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
192 517 : if (last_block > last_block_in_file)
193 : last_block = last_block_in_file;
194 517 : page_block = 0;
195 :
196 : /*
197 : * Map blocks using the result from the previous get_blocks call first.
198 : */
199 517 : nblocks = map_bh->b_size >> blkbits;
200 517 : if (buffer_mapped(map_bh) &&
201 445 : block_in_file > args->first_logical_block &&
202 445 : block_in_file < (args->first_logical_block + nblocks)) {
203 445 : unsigned map_offset = block_in_file - args->first_logical_block;
204 445 : unsigned last = nblocks - map_offset;
205 :
206 445 : for (relative_block = 0; ; relative_block++) {
207 890 : if (relative_block == last) {
208 26 : clear_buffer_mapped(map_bh);
209 : break;
210 : }
211 864 : if (page_block == blocks_per_page)
212 : break;
213 445 : blocks[page_block] = map_bh->b_blocknr + map_offset +
214 : relative_block;
215 445 : page_block++;
216 445 : block_in_file++;
217 : }
218 445 : bdev = map_bh->b_bdev;
219 : }
220 :
221 : /*
222 : * Then do more get_blocks calls until we are done with this page.
223 : */
224 517 : map_bh->b_page = page;
225 589 : while (page_block < blocks_per_page) {
226 72 : map_bh->b_state = 0;
227 72 : map_bh->b_size = 0;
228 :
229 72 : if (block_in_file < last_block) {
230 72 : map_bh->b_size = (last_block-block_in_file) << blkbits;
231 72 : if (args->get_block(inode, block_in_file, map_bh, 0))
232 0 : goto confused;
233 72 : args->first_logical_block = block_in_file;
234 : }
235 :
236 72 : if (!buffer_mapped(map_bh)) {
237 0 : fully_mapped = 0;
238 0 : if (first_hole == blocks_per_page)
239 0 : first_hole = page_block;
240 0 : page_block++;
241 0 : block_in_file++;
242 0 : continue;
243 : }
244 :
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
250 : */
251 72 : if (buffer_uptodate(map_bh)) {
252 0 : map_buffer_to_page(page, map_bh, page_block);
253 0 : goto confused;
254 : }
255 :
256 72 : if (first_hole != blocks_per_page)
257 0 : goto confused; /* hole -> non-hole */
258 :
259 : /* Contiguous blocks? */
260 72 : if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
261 0 : goto confused;
262 72 : nblocks = map_bh->b_size >> blkbits;
263 72 : for (relative_block = 0; ; relative_block++) {
264 144 : if (relative_block == nblocks) {
265 46 : clear_buffer_mapped(map_bh);
266 : break;
267 98 : } else if (page_block == blocks_per_page)
268 : break;
269 72 : blocks[page_block] = map_bh->b_blocknr+relative_block;
270 72 : page_block++;
271 72 : block_in_file++;
272 : }
273 72 : bdev = map_bh->b_bdev;
274 : }
275 :
276 517 : if (first_hole != blocks_per_page) {
277 0 : zero_user_segment(page, first_hole << blkbits, PAGE_SIZE);
278 0 : if (first_hole == 0) {
279 0 : SetPageUptodate(page);
280 0 : unlock_page(page);
281 0 : goto out;
282 : }
283 517 : } else if (fully_mapped) {
284 517 : SetPageMappedToDisk(page);
285 : }
286 :
287 1034 : if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
288 517 : cleancache_get_page(page) == 0) {
289 0 : SetPageUptodate(page);
290 0 : goto confused;
291 : }
292 :
293 : /*
294 : * This page will go to BIO. Do we need to send this BIO off first?
295 : */
296 517 : if (args->bio && (args->last_block_in_bio != blocks[0] - 1))
297 0 : args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
298 :
299 517 : alloc_new:
300 517 : if (args->bio == NULL) {
301 72 : if (first_hole == blocks_per_page) {
302 72 : if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
303 : page))
304 0 : goto out;
305 : }
306 144 : args->bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
307 72 : bio_max_segs(args->nr_pages), gfp);
308 72 : if (args->bio == NULL)
309 0 : goto confused;
310 : }
311 :
312 517 : length = first_hole << blkbits;
313 517 : if (bio_add_page(args->bio, page, length, 0) < length) {
314 0 : args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
315 0 : goto alloc_new;
316 : }
317 :
318 517 : relative_block = block_in_file - args->first_logical_block;
319 517 : nblocks = map_bh->b_size >> blkbits;
320 517 : if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
321 : (first_hole != blocks_per_page))
322 0 : args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
323 : else
324 517 : args->last_block_in_bio = blocks[blocks_per_page - 1];
325 517 : out:
326 517 : return args->bio;
327 :
328 0 : confused:
329 0 : if (args->bio)
330 0 : args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
331 0 : if (!PageUptodate(page))
332 0 : block_read_full_page(page, args->get_block);
333 : else
334 0 : unlock_page(page);
335 0 : goto out;
336 : }
337 :
338 : /**
339 : * mpage_readahead - start reads against pages
340 : * @rac: Describes which pages to read.
341 : * @get_block: The filesystem's block mapper function.
342 : *
343 : * This function walks the pages and the blocks within each page, building and
344 : * emitting large BIOs.
345 : *
346 : * If anything unusual happens, such as:
347 : *
348 : * - encountering a page which has buffers
349 : * - encountering a page which has a non-hole after a hole
350 : * - encountering a page with non-contiguous blocks
351 : *
352 : * then this code just gives up and calls the buffer_head-based read function.
353 : * It does handle a page which has holes at the end - that is a common case:
354 : * the end-of-file on blocksize < PAGE_SIZE setups.
355 : *
356 : * BH_Boundary explanation:
357 : *
358 : * There is a problem. The mpage read code assembles several pages, gets all
359 : * their disk mappings, and then submits them all. That's fine, but obtaining
360 : * the disk mappings may require I/O. Reads of indirect blocks, for example.
361 : *
362 : * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
363 : * submitted in the following order:
364 : *
365 : * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
366 : *
367 : * because the indirect block has to be read to get the mappings of blocks
368 : * 13,14,15,16. Obviously, this impacts performance.
369 : *
370 : * So what we do it to allow the filesystem's get_block() function to set
371 : * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
372 : * after this one will require I/O against a block which is probably close to
373 : * this one. So you should push what I/O you have currently accumulated.
374 : *
375 : * This all causes the disk requests to be issued in the correct order.
376 : */
377 72 : void mpage_readahead(struct readahead_control *rac, get_block_t get_block)
378 : {
379 72 : struct page *page;
380 72 : struct mpage_readpage_args args = {
381 : .get_block = get_block,
382 : .is_readahead = true,
383 : };
384 :
385 589 : while ((page = readahead_page(rac))) {
386 517 : prefetchw(&page->flags);
387 517 : args.page = page;
388 517 : args.nr_pages = readahead_count(rac);
389 517 : args.bio = do_mpage_readpage(&args);
390 517 : put_page(page);
391 : }
392 72 : if (args.bio)
393 72 : mpage_bio_submit(REQ_OP_READ, REQ_RAHEAD, args.bio);
394 72 : }
395 : EXPORT_SYMBOL(mpage_readahead);
396 :
397 : /*
398 : * This isn't called much at all
399 : */
400 0 : int mpage_readpage(struct page *page, get_block_t get_block)
401 : {
402 0 : struct mpage_readpage_args args = {
403 : .page = page,
404 : .nr_pages = 1,
405 : .get_block = get_block,
406 : };
407 :
408 0 : args.bio = do_mpage_readpage(&args);
409 0 : if (args.bio)
410 0 : mpage_bio_submit(REQ_OP_READ, 0, args.bio);
411 0 : return 0;
412 : }
413 : EXPORT_SYMBOL(mpage_readpage);
414 :
415 : /*
416 : * Writing is not so simple.
417 : *
418 : * If the page has buffers then they will be used for obtaining the disk
419 : * mapping. We only support pages which are fully mapped-and-dirty, with a
420 : * special case for pages which are unmapped at the end: end-of-file.
421 : *
422 : * If the page has no buffers (preferred) then the page is mapped here.
423 : *
424 : * If all blocks are found to be contiguous then the page can go into the
425 : * BIO. Otherwise fall back to the mapping's writepage().
426 : *
427 : * FIXME: This code wants an estimate of how many pages are still to be
428 : * written, so it can intelligently allocate a suitably-sized BIO. For now,
429 : * just allocate full-size (16-page) BIOs.
430 : */
431 :
432 : struct mpage_data {
433 : struct bio *bio;
434 : sector_t last_block_in_bio;
435 : get_block_t *get_block;
436 : unsigned use_writepage;
437 : };
438 :
439 : /*
440 : * We have our BIO, so we can now mark the buffers clean. Make
441 : * sure to only clean buffers which we know we'll be writing.
442 : */
443 0 : static void clean_buffers(struct page *page, unsigned first_unmapped)
444 : {
445 0 : unsigned buffer_counter = 0;
446 0 : struct buffer_head *bh, *head;
447 0 : if (!page_has_buffers(page))
448 : return;
449 0 : head = page_buffers(page);
450 0 : bh = head;
451 :
452 0 : do {
453 0 : if (buffer_counter++ == first_unmapped)
454 : break;
455 0 : clear_buffer_dirty(bh);
456 0 : bh = bh->b_this_page;
457 0 : } while (bh != head);
458 :
459 : /*
460 : * we cannot drop the bh if the page is not uptodate or a concurrent
461 : * readpage would fail to serialize with the bh and it would read from
462 : * disk before we reach the platter.
463 : */
464 0 : if (buffer_heads_over_limit && PageUptodate(page))
465 0 : try_to_free_buffers(page);
466 : }
467 :
468 : /*
469 : * For situations where we want to clean all buffers attached to a page.
470 : * We don't need to calculate how many buffers are attached to the page,
471 : * we just need to specify a number larger than the maximum number of buffers.
472 : */
473 0 : void clean_page_buffers(struct page *page)
474 : {
475 0 : clean_buffers(page, ~0U);
476 0 : }
477 :
478 0 : static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
479 : void *data)
480 : {
481 0 : struct mpage_data *mpd = data;
482 0 : struct bio *bio = mpd->bio;
483 0 : struct address_space *mapping = page->mapping;
484 0 : struct inode *inode = page->mapping->host;
485 0 : const unsigned blkbits = inode->i_blkbits;
486 0 : unsigned long end_index;
487 0 : const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
488 0 : sector_t last_block;
489 0 : sector_t block_in_file;
490 0 : sector_t blocks[MAX_BUF_PER_PAGE];
491 0 : unsigned page_block;
492 0 : unsigned first_unmapped = blocks_per_page;
493 0 : struct block_device *bdev = NULL;
494 0 : int boundary = 0;
495 0 : sector_t boundary_block = 0;
496 0 : struct block_device *boundary_bdev = NULL;
497 0 : int length;
498 0 : struct buffer_head map_bh;
499 0 : loff_t i_size = i_size_read(inode);
500 0 : int ret = 0;
501 0 : int op_flags = wbc_to_write_flags(wbc);
502 :
503 0 : if (page_has_buffers(page)) {
504 0 : struct buffer_head *head = page_buffers(page);
505 0 : struct buffer_head *bh = head;
506 :
507 : /* If they're all mapped and dirty, do it */
508 0 : page_block = 0;
509 0 : do {
510 0 : BUG_ON(buffer_locked(bh));
511 0 : if (!buffer_mapped(bh)) {
512 : /*
513 : * unmapped dirty buffers are created by
514 : * __set_page_dirty_buffers -> mmapped data
515 : */
516 0 : if (buffer_dirty(bh))
517 0 : goto confused;
518 0 : if (first_unmapped == blocks_per_page)
519 0 : first_unmapped = page_block;
520 0 : continue;
521 : }
522 :
523 0 : if (first_unmapped != blocks_per_page)
524 0 : goto confused; /* hole -> non-hole */
525 :
526 0 : if (!buffer_dirty(bh) || !buffer_uptodate(bh))
527 0 : goto confused;
528 0 : if (page_block) {
529 0 : if (bh->b_blocknr != blocks[page_block-1] + 1)
530 0 : goto confused;
531 : }
532 0 : blocks[page_block++] = bh->b_blocknr;
533 0 : boundary = buffer_boundary(bh);
534 0 : if (boundary) {
535 0 : boundary_block = bh->b_blocknr;
536 0 : boundary_bdev = bh->b_bdev;
537 : }
538 0 : bdev = bh->b_bdev;
539 0 : } while ((bh = bh->b_this_page) != head);
540 :
541 0 : if (first_unmapped)
542 0 : goto page_is_mapped;
543 :
544 : /*
545 : * Page has buffers, but they are all unmapped. The page was
546 : * created by pagein or read over a hole which was handled by
547 : * block_read_full_page(). If this address_space is also
548 : * using mpage_readahead then this can rarely happen.
549 : */
550 0 : goto confused;
551 : }
552 :
553 : /*
554 : * The page has no buffers: map it to disk
555 : */
556 0 : BUG_ON(!PageUptodate(page));
557 0 : block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
558 0 : last_block = (i_size - 1) >> blkbits;
559 0 : map_bh.b_page = page;
560 0 : for (page_block = 0; page_block < blocks_per_page; ) {
561 :
562 0 : map_bh.b_state = 0;
563 0 : map_bh.b_size = 1 << blkbits;
564 0 : if (mpd->get_block(inode, block_in_file, &map_bh, 1))
565 0 : goto confused;
566 0 : if (buffer_new(&map_bh))
567 0 : clean_bdev_bh_alias(&map_bh);
568 0 : if (buffer_boundary(&map_bh)) {
569 0 : boundary_block = map_bh.b_blocknr;
570 0 : boundary_bdev = map_bh.b_bdev;
571 : }
572 0 : if (page_block) {
573 0 : if (map_bh.b_blocknr != blocks[page_block-1] + 1)
574 0 : goto confused;
575 : }
576 0 : blocks[page_block++] = map_bh.b_blocknr;
577 0 : boundary = buffer_boundary(&map_bh);
578 0 : bdev = map_bh.b_bdev;
579 0 : if (block_in_file == last_block)
580 : break;
581 0 : block_in_file++;
582 : }
583 0 : BUG_ON(page_block == 0);
584 :
585 : first_unmapped = page_block;
586 :
587 0 : page_is_mapped:
588 0 : end_index = i_size >> PAGE_SHIFT;
589 0 : if (page->index >= end_index) {
590 : /*
591 : * The page straddles i_size. It must be zeroed out on each
592 : * and every writepage invocation because it may be mmapped.
593 : * "A file is mapped in multiples of the page size. For a file
594 : * that is not a multiple of the page size, the remaining memory
595 : * is zeroed when mapped, and writes to that region are not
596 : * written out to the file."
597 : */
598 0 : unsigned offset = i_size & (PAGE_SIZE - 1);
599 :
600 0 : if (page->index > end_index || !offset)
601 0 : goto confused;
602 0 : zero_user_segment(page, offset, PAGE_SIZE);
603 : }
604 :
605 : /*
606 : * This page will go to BIO. Do we need to send this BIO off first?
607 : */
608 0 : if (bio && mpd->last_block_in_bio != blocks[0] - 1)
609 0 : bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
610 :
611 0 : alloc_new:
612 0 : if (bio == NULL) {
613 0 : if (first_unmapped == blocks_per_page) {
614 0 : if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
615 : page, wbc))
616 0 : goto out;
617 : }
618 0 : bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
619 : BIO_MAX_VECS, GFP_NOFS|__GFP_HIGH);
620 0 : if (bio == NULL)
621 0 : goto confused;
622 :
623 0 : wbc_init_bio(wbc, bio);
624 0 : bio->bi_write_hint = inode->i_write_hint;
625 : }
626 :
627 : /*
628 : * Must try to add the page before marking the buffer clean or
629 : * the confused fail path above (OOM) will be very confused when
630 : * it finds all bh marked clean (i.e. it will not write anything)
631 : */
632 0 : wbc_account_cgroup_owner(wbc, page, PAGE_SIZE);
633 0 : length = first_unmapped << blkbits;
634 0 : if (bio_add_page(bio, page, length, 0) < length) {
635 0 : bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
636 0 : goto alloc_new;
637 : }
638 :
639 0 : clean_buffers(page, first_unmapped);
640 :
641 0 : BUG_ON(PageWriteback(page));
642 0 : set_page_writeback(page);
643 0 : unlock_page(page);
644 0 : if (boundary || (first_unmapped != blocks_per_page)) {
645 0 : bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
646 0 : if (boundary_block) {
647 0 : write_boundary_block(boundary_bdev,
648 0 : boundary_block, 1 << blkbits);
649 : }
650 : } else {
651 0 : mpd->last_block_in_bio = blocks[blocks_per_page - 1];
652 : }
653 0 : goto out;
654 :
655 0 : confused:
656 0 : if (bio)
657 0 : bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
658 :
659 0 : if (mpd->use_writepage) {
660 0 : ret = mapping->a_ops->writepage(page, wbc);
661 : } else {
662 0 : ret = -EAGAIN;
663 0 : goto out;
664 : }
665 : /*
666 : * The caller has a ref on the inode, so *mapping is stable
667 : */
668 0 : mapping_set_error(mapping, ret);
669 0 : out:
670 0 : mpd->bio = bio;
671 0 : return ret;
672 : }
673 :
674 : /**
675 : * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
676 : * @mapping: address space structure to write
677 : * @wbc: subtract the number of written pages from *@wbc->nr_to_write
678 : * @get_block: the filesystem's block mapper function.
679 : * If this is NULL then use a_ops->writepage. Otherwise, go
680 : * direct-to-BIO.
681 : *
682 : * This is a library function, which implements the writepages()
683 : * address_space_operation.
684 : *
685 : * If a page is already under I/O, generic_writepages() skips it, even
686 : * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
687 : * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
688 : * and msync() need to guarantee that all the data which was dirty at the time
689 : * the call was made get new I/O started against them. If wbc->sync_mode is
690 : * WB_SYNC_ALL then we were called for data integrity and we must wait for
691 : * existing IO to complete.
692 : */
693 : int
694 0 : mpage_writepages(struct address_space *mapping,
695 : struct writeback_control *wbc, get_block_t get_block)
696 : {
697 0 : struct blk_plug plug;
698 0 : int ret;
699 :
700 0 : blk_start_plug(&plug);
701 :
702 0 : if (!get_block)
703 0 : ret = generic_writepages(mapping, wbc);
704 : else {
705 0 : struct mpage_data mpd = {
706 : .bio = NULL,
707 : .last_block_in_bio = 0,
708 : .get_block = get_block,
709 : .use_writepage = 1,
710 : };
711 :
712 0 : ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
713 0 : if (mpd.bio) {
714 0 : int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
715 : REQ_SYNC : 0);
716 0 : mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
717 : }
718 : }
719 0 : blk_finish_plug(&plug);
720 0 : return ret;
721 : }
722 : EXPORT_SYMBOL(mpage_writepages);
723 :
724 0 : int mpage_writepage(struct page *page, get_block_t get_block,
725 : struct writeback_control *wbc)
726 : {
727 0 : struct mpage_data mpd = {
728 : .bio = NULL,
729 : .last_block_in_bio = 0,
730 : .get_block = get_block,
731 : .use_writepage = 0,
732 : };
733 0 : int ret = __mpage_writepage(page, wbc, &mpd);
734 0 : if (mpd.bio) {
735 0 : int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
736 : REQ_SYNC : 0);
737 0 : mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
738 : }
739 0 : return ret;
740 : }
741 : EXPORT_SYMBOL(mpage_writepage);
|
