]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
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 | * | |
e1f8e874 | 10 | * 15May2002 Andrew Morton |
1da177e4 LT |
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> | |
630d9c47 | 17 | #include <linux/export.h> |
1da177e4 LT |
18 | #include <linux/mm.h> |
19 | #include <linux/kdev_t.h> | |
5a0e3ad6 | 20 | #include <linux/gfp.h> |
1da177e4 LT |
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> | |
02c43638 | 28 | #include <linux/mm_inline.h> |
1da177e4 LT |
29 | #include <linux/writeback.h> |
30 | #include <linux/backing-dev.h> | |
31 | #include <linux/pagevec.h> | |
4db96b71 | 32 | #include "internal.h" |
1da177e4 LT |
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 | */ | |
4246a0b6 | 46 | static void mpage_end_io(struct bio *bio) |
1da177e4 | 47 | { |
2c30c71b | 48 | struct bio_vec *bv; |
6dc4f100 | 49 | struct bvec_iter_all iter_all; |
1da177e4 | 50 | |
2b070cfe | 51 | bio_for_each_segment_all(bv, bio, iter_all) { |
2c30c71b | 52 | struct page *page = bv->bv_page; |
3f289dcb TH |
53 | page_endio(page, bio_op(bio), |
54 | blk_status_to_errno(bio->bi_status)); | |
2c30c71b KO |
55 | } |
56 | ||
1da177e4 | 57 | bio_put(bio); |
1da177e4 LT |
58 | } |
59 | ||
eed25cd5 | 60 | static struct bio *mpage_bio_submit(int op, int op_flags, struct bio *bio) |
1da177e4 | 61 | { |
c32b0d4b | 62 | bio->bi_end_io = mpage_end_io; |
eed25cd5 | 63 | bio_set_op_attrs(bio, op, op_flags); |
83c9c547 | 64 | guard_bio_eod(bio); |
4e49ea4a | 65 | submit_bio(bio); |
1da177e4 LT |
66 | return NULL; |
67 | } | |
68 | ||
1da177e4 | 69 | /* |
d4388340 | 70 | * support function for mpage_readahead. The fs supplied get_block might |
1da177e4 LT |
71 | * return an up to date buffer. This is used to map that buffer into |
72 | * the page, which allows readpage to avoid triggering a duplicate call | |
73 | * to get_block. | |
74 | * | |
75 | * The idea is to avoid adding buffers to pages that don't already have | |
76 | * them. So when the buffer is up to date and the page size == block size, | |
77 | * this marks the page up to date instead of adding new buffers. | |
78 | */ | |
79 | static void | |
80 | map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block) | |
81 | { | |
82 | struct inode *inode = page->mapping->host; | |
83 | struct buffer_head *page_bh, *head; | |
84 | int block = 0; | |
85 | ||
86 | if (!page_has_buffers(page)) { | |
87 | /* | |
88 | * don't make any buffers if there is only one buffer on | |
89 | * the page and the page just needs to be set up to date | |
90 | */ | |
09cbfeaf | 91 | if (inode->i_blkbits == PAGE_SHIFT && |
1da177e4 LT |
92 | buffer_uptodate(bh)) { |
93 | SetPageUptodate(page); | |
94 | return; | |
95 | } | |
93407472 | 96 | create_empty_buffers(page, i_blocksize(inode), 0); |
1da177e4 LT |
97 | } |
98 | head = page_buffers(page); | |
99 | page_bh = head; | |
100 | do { | |
101 | if (block == page_block) { | |
102 | page_bh->b_state = bh->b_state; | |
103 | page_bh->b_bdev = bh->b_bdev; | |
104 | page_bh->b_blocknr = bh->b_blocknr; | |
105 | break; | |
106 | } | |
107 | page_bh = page_bh->b_this_page; | |
108 | block++; | |
109 | } while (page_bh != head); | |
110 | } | |
111 | ||
357c1206 JA |
112 | struct mpage_readpage_args { |
113 | struct bio *bio; | |
114 | struct page *page; | |
115 | unsigned int nr_pages; | |
74c8164e | 116 | bool is_readahead; |
357c1206 JA |
117 | sector_t last_block_in_bio; |
118 | struct buffer_head map_bh; | |
119 | unsigned long first_logical_block; | |
120 | get_block_t *get_block; | |
357c1206 JA |
121 | }; |
122 | ||
fa30bd05 BP |
123 | /* |
124 | * This is the worker routine which does all the work of mapping the disk | |
125 | * blocks and constructs largest possible bios, submits them for IO if the | |
126 | * blocks are not contiguous on the disk. | |
127 | * | |
128 | * We pass a buffer_head back and forth and use its buffer_mapped() flag to | |
129 | * represent the validity of its disk mapping and to decide when to do the next | |
130 | * get_block() call. | |
131 | */ | |
357c1206 | 132 | static struct bio *do_mpage_readpage(struct mpage_readpage_args *args) |
1da177e4 | 133 | { |
357c1206 | 134 | struct page *page = args->page; |
1da177e4 LT |
135 | struct inode *inode = page->mapping->host; |
136 | const unsigned blkbits = inode->i_blkbits; | |
09cbfeaf | 137 | const unsigned blocks_per_page = PAGE_SIZE >> blkbits; |
1da177e4 | 138 | const unsigned blocksize = 1 << blkbits; |
357c1206 | 139 | struct buffer_head *map_bh = &args->map_bh; |
1da177e4 LT |
140 | sector_t block_in_file; |
141 | sector_t last_block; | |
fa30bd05 | 142 | sector_t last_block_in_file; |
1da177e4 LT |
143 | sector_t blocks[MAX_BUF_PER_PAGE]; |
144 | unsigned page_block; | |
145 | unsigned first_hole = blocks_per_page; | |
146 | struct block_device *bdev = NULL; | |
1da177e4 LT |
147 | int length; |
148 | int fully_mapped = 1; | |
74c8164e | 149 | int op_flags; |
fa30bd05 BP |
150 | unsigned nblocks; |
151 | unsigned relative_block; | |
74c8164e JA |
152 | gfp_t gfp; |
153 | ||
154 | if (args->is_readahead) { | |
155 | op_flags = REQ_RAHEAD; | |
156 | gfp = readahead_gfp_mask(page->mapping); | |
157 | } else { | |
158 | op_flags = 0; | |
159 | gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL); | |
160 | } | |
1da177e4 LT |
161 | |
162 | if (page_has_buffers(page)) | |
163 | goto confused; | |
164 | ||
09cbfeaf | 165 | block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits); |
357c1206 | 166 | last_block = block_in_file + args->nr_pages * blocks_per_page; |
fa30bd05 BP |
167 | last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits; |
168 | if (last_block > last_block_in_file) | |
169 | last_block = last_block_in_file; | |
170 | page_block = 0; | |
171 | ||
172 | /* | |
173 | * Map blocks using the result from the previous get_blocks call first. | |
174 | */ | |
175 | nblocks = map_bh->b_size >> blkbits; | |
357c1206 JA |
176 | if (buffer_mapped(map_bh) && |
177 | block_in_file > args->first_logical_block && | |
178 | block_in_file < (args->first_logical_block + nblocks)) { | |
179 | unsigned map_offset = block_in_file - args->first_logical_block; | |
fa30bd05 BP |
180 | unsigned last = nblocks - map_offset; |
181 | ||
182 | for (relative_block = 0; ; relative_block++) { | |
183 | if (relative_block == last) { | |
184 | clear_buffer_mapped(map_bh); | |
185 | break; | |
186 | } | |
187 | if (page_block == blocks_per_page) | |
188 | break; | |
189 | blocks[page_block] = map_bh->b_blocknr + map_offset + | |
190 | relative_block; | |
191 | page_block++; | |
192 | block_in_file++; | |
193 | } | |
194 | bdev = map_bh->b_bdev; | |
195 | } | |
196 | ||
197 | /* | |
198 | * Then do more get_blocks calls until we are done with this page. | |
199 | */ | |
200 | map_bh->b_page = page; | |
201 | while (page_block < blocks_per_page) { | |
202 | map_bh->b_state = 0; | |
203 | map_bh->b_size = 0; | |
1da177e4 | 204 | |
1da177e4 | 205 | if (block_in_file < last_block) { |
fa30bd05 | 206 | map_bh->b_size = (last_block-block_in_file) << blkbits; |
357c1206 | 207 | if (args->get_block(inode, block_in_file, map_bh, 0)) |
1da177e4 | 208 | goto confused; |
357c1206 | 209 | args->first_logical_block = block_in_file; |
1da177e4 LT |
210 | } |
211 | ||
fa30bd05 | 212 | if (!buffer_mapped(map_bh)) { |
1da177e4 LT |
213 | fully_mapped = 0; |
214 | if (first_hole == blocks_per_page) | |
215 | first_hole = page_block; | |
fa30bd05 BP |
216 | page_block++; |
217 | block_in_file++; | |
1da177e4 LT |
218 | continue; |
219 | } | |
220 | ||
221 | /* some filesystems will copy data into the page during | |
222 | * the get_block call, in which case we don't want to | |
223 | * read it again. map_buffer_to_page copies the data | |
224 | * we just collected from get_block into the page's buffers | |
225 | * so readpage doesn't have to repeat the get_block call | |
226 | */ | |
fa30bd05 BP |
227 | if (buffer_uptodate(map_bh)) { |
228 | map_buffer_to_page(page, map_bh, page_block); | |
1da177e4 LT |
229 | goto confused; |
230 | } | |
231 | ||
232 | if (first_hole != blocks_per_page) | |
233 | goto confused; /* hole -> non-hole */ | |
234 | ||
235 | /* Contiguous blocks? */ | |
fa30bd05 | 236 | if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1) |
1da177e4 | 237 | goto confused; |
fa30bd05 BP |
238 | nblocks = map_bh->b_size >> blkbits; |
239 | for (relative_block = 0; ; relative_block++) { | |
240 | if (relative_block == nblocks) { | |
241 | clear_buffer_mapped(map_bh); | |
242 | break; | |
243 | } else if (page_block == blocks_per_page) | |
244 | break; | |
245 | blocks[page_block] = map_bh->b_blocknr+relative_block; | |
246 | page_block++; | |
247 | block_in_file++; | |
248 | } | |
249 | bdev = map_bh->b_bdev; | |
1da177e4 LT |
250 | } |
251 | ||
252 | if (first_hole != blocks_per_page) { | |
09cbfeaf | 253 | zero_user_segment(page, first_hole << blkbits, PAGE_SIZE); |
1da177e4 LT |
254 | if (first_hole == 0) { |
255 | SetPageUptodate(page); | |
256 | unlock_page(page); | |
257 | goto out; | |
258 | } | |
259 | } else if (fully_mapped) { | |
260 | SetPageMappedToDisk(page); | |
261 | } | |
262 | ||
263 | /* | |
264 | * This page will go to BIO. Do we need to send this BIO off first? | |
265 | */ | |
357c1206 | 266 | if (args->bio && (args->last_block_in_bio != blocks[0] - 1)) |
74c8164e | 267 | args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio); |
1da177e4 LT |
268 | |
269 | alloc_new: | |
357c1206 | 270 | if (args->bio == NULL) { |
47a191fd MW |
271 | if (first_hole == blocks_per_page) { |
272 | if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9), | |
273 | page)) | |
274 | goto out; | |
275 | } | |
07888c66 CH |
276 | args->bio = bio_alloc(bdev, bio_max_segs(args->nr_pages), 0, |
277 | gfp); | |
357c1206 | 278 | if (args->bio == NULL) |
1da177e4 | 279 | goto confused; |
d5f68a42 | 280 | args->bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9); |
1da177e4 LT |
281 | } |
282 | ||
283 | length = first_hole << blkbits; | |
357c1206 | 284 | if (bio_add_page(args->bio, page, length, 0) < length) { |
74c8164e | 285 | args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio); |
1da177e4 LT |
286 | goto alloc_new; |
287 | } | |
288 | ||
357c1206 | 289 | relative_block = block_in_file - args->first_logical_block; |
38c8e618 MS |
290 | nblocks = map_bh->b_size >> blkbits; |
291 | if ((buffer_boundary(map_bh) && relative_block == nblocks) || | |
292 | (first_hole != blocks_per_page)) | |
74c8164e | 293 | args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio); |
1da177e4 | 294 | else |
357c1206 | 295 | args->last_block_in_bio = blocks[blocks_per_page - 1]; |
1da177e4 | 296 | out: |
357c1206 | 297 | return args->bio; |
1da177e4 LT |
298 | |
299 | confused: | |
357c1206 | 300 | if (args->bio) |
74c8164e | 301 | args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio); |
1da177e4 | 302 | if (!PageUptodate(page)) |
357c1206 | 303 | block_read_full_page(page, args->get_block); |
1da177e4 LT |
304 | else |
305 | unlock_page(page); | |
306 | goto out; | |
307 | } | |
308 | ||
67be2dd1 | 309 | /** |
d4388340 MWO |
310 | * mpage_readahead - start reads against pages |
311 | * @rac: Describes which pages to read. | |
67be2dd1 MW |
312 | * @get_block: The filesystem's block mapper function. |
313 | * | |
314 | * This function walks the pages and the blocks within each page, building and | |
315 | * emitting large BIOs. | |
316 | * | |
317 | * If anything unusual happens, such as: | |
318 | * | |
319 | * - encountering a page which has buffers | |
320 | * - encountering a page which has a non-hole after a hole | |
321 | * - encountering a page with non-contiguous blocks | |
322 | * | |
323 | * then this code just gives up and calls the buffer_head-based read function. | |
324 | * It does handle a page which has holes at the end - that is a common case: | |
ea1754a0 | 325 | * the end-of-file on blocksize < PAGE_SIZE setups. |
67be2dd1 MW |
326 | * |
327 | * BH_Boundary explanation: | |
328 | * | |
329 | * There is a problem. The mpage read code assembles several pages, gets all | |
330 | * their disk mappings, and then submits them all. That's fine, but obtaining | |
331 | * the disk mappings may require I/O. Reads of indirect blocks, for example. | |
332 | * | |
333 | * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be | |
334 | * submitted in the following order: | |
0117d427 | 335 | * |
67be2dd1 | 336 | * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16 |
78a4a50a | 337 | * |
67be2dd1 MW |
338 | * because the indirect block has to be read to get the mappings of blocks |
339 | * 13,14,15,16. Obviously, this impacts performance. | |
340 | * | |
341 | * So what we do it to allow the filesystem's get_block() function to set | |
342 | * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block | |
343 | * after this one will require I/O against a block which is probably close to | |
344 | * this one. So you should push what I/O you have currently accumulated. | |
345 | * | |
346 | * This all causes the disk requests to be issued in the correct order. | |
347 | */ | |
d4388340 | 348 | void mpage_readahead(struct readahead_control *rac, get_block_t get_block) |
1da177e4 | 349 | { |
d4388340 | 350 | struct page *page; |
357c1206 JA |
351 | struct mpage_readpage_args args = { |
352 | .get_block = get_block, | |
74c8164e | 353 | .is_readahead = true, |
357c1206 | 354 | }; |
1da177e4 | 355 | |
d4388340 | 356 | while ((page = readahead_page(rac))) { |
1da177e4 | 357 | prefetchw(&page->flags); |
d4388340 MWO |
358 | args.page = page; |
359 | args.nr_pages = readahead_count(rac); | |
360 | args.bio = do_mpage_readpage(&args); | |
09cbfeaf | 361 | put_page(page); |
1da177e4 | 362 | } |
357c1206 | 363 | if (args.bio) |
74c8164e | 364 | mpage_bio_submit(REQ_OP_READ, REQ_RAHEAD, args.bio); |
1da177e4 | 365 | } |
d4388340 | 366 | EXPORT_SYMBOL(mpage_readahead); |
1da177e4 LT |
367 | |
368 | /* | |
369 | * This isn't called much at all | |
370 | */ | |
371 | int mpage_readpage(struct page *page, get_block_t get_block) | |
372 | { | |
357c1206 JA |
373 | struct mpage_readpage_args args = { |
374 | .page = page, | |
375 | .nr_pages = 1, | |
376 | .get_block = get_block, | |
357c1206 JA |
377 | }; |
378 | ||
379 | args.bio = do_mpage_readpage(&args); | |
380 | if (args.bio) | |
381 | mpage_bio_submit(REQ_OP_READ, 0, args.bio); | |
1da177e4 LT |
382 | return 0; |
383 | } | |
384 | EXPORT_SYMBOL(mpage_readpage); | |
385 | ||
386 | /* | |
387 | * Writing is not so simple. | |
388 | * | |
389 | * If the page has buffers then they will be used for obtaining the disk | |
390 | * mapping. We only support pages which are fully mapped-and-dirty, with a | |
391 | * special case for pages which are unmapped at the end: end-of-file. | |
392 | * | |
393 | * If the page has no buffers (preferred) then the page is mapped here. | |
394 | * | |
395 | * If all blocks are found to be contiguous then the page can go into the | |
396 | * BIO. Otherwise fall back to the mapping's writepage(). | |
397 | * | |
398 | * FIXME: This code wants an estimate of how many pages are still to be | |
399 | * written, so it can intelligently allocate a suitably-sized BIO. For now, | |
400 | * just allocate full-size (16-page) BIOs. | |
401 | */ | |
0ea97180 | 402 | |
ced117c7 DV |
403 | struct mpage_data { |
404 | struct bio *bio; | |
405 | sector_t last_block_in_bio; | |
406 | get_block_t *get_block; | |
407 | unsigned use_writepage; | |
408 | }; | |
409 | ||
90768eee MW |
410 | /* |
411 | * We have our BIO, so we can now mark the buffers clean. Make | |
412 | * sure to only clean buffers which we know we'll be writing. | |
413 | */ | |
414 | static void clean_buffers(struct page *page, unsigned first_unmapped) | |
415 | { | |
416 | unsigned buffer_counter = 0; | |
417 | struct buffer_head *bh, *head; | |
418 | if (!page_has_buffers(page)) | |
419 | return; | |
420 | head = page_buffers(page); | |
421 | bh = head; | |
422 | ||
423 | do { | |
424 | if (buffer_counter++ == first_unmapped) | |
425 | break; | |
426 | clear_buffer_dirty(bh); | |
427 | bh = bh->b_this_page; | |
428 | } while (bh != head); | |
429 | ||
430 | /* | |
431 | * we cannot drop the bh if the page is not uptodate or a concurrent | |
432 | * readpage would fail to serialize with the bh and it would read from | |
433 | * disk before we reach the platter. | |
434 | */ | |
435 | if (buffer_heads_over_limit && PageUptodate(page)) | |
436 | try_to_free_buffers(page); | |
437 | } | |
438 | ||
f892760a MW |
439 | /* |
440 | * For situations where we want to clean all buffers attached to a page. | |
441 | * We don't need to calculate how many buffers are attached to the page, | |
442 | * we just need to specify a number larger than the maximum number of buffers. | |
443 | */ | |
444 | void clean_page_buffers(struct page *page) | |
445 | { | |
446 | clean_buffers(page, ~0U); | |
447 | } | |
448 | ||
ced117c7 | 449 | static int __mpage_writepage(struct page *page, struct writeback_control *wbc, |
29a814d2 | 450 | void *data) |
1da177e4 | 451 | { |
0ea97180 MS |
452 | struct mpage_data *mpd = data; |
453 | struct bio *bio = mpd->bio; | |
1da177e4 LT |
454 | struct address_space *mapping = page->mapping; |
455 | struct inode *inode = page->mapping->host; | |
456 | const unsigned blkbits = inode->i_blkbits; | |
457 | unsigned long end_index; | |
09cbfeaf | 458 | const unsigned blocks_per_page = PAGE_SIZE >> blkbits; |
1da177e4 LT |
459 | sector_t last_block; |
460 | sector_t block_in_file; | |
461 | sector_t blocks[MAX_BUF_PER_PAGE]; | |
462 | unsigned page_block; | |
463 | unsigned first_unmapped = blocks_per_page; | |
464 | struct block_device *bdev = NULL; | |
465 | int boundary = 0; | |
466 | sector_t boundary_block = 0; | |
467 | struct block_device *boundary_bdev = NULL; | |
468 | int length; | |
469 | struct buffer_head map_bh; | |
470 | loff_t i_size = i_size_read(inode); | |
0ea97180 | 471 | int ret = 0; |
7637241e | 472 | int op_flags = wbc_to_write_flags(wbc); |
1da177e4 LT |
473 | |
474 | if (page_has_buffers(page)) { | |
475 | struct buffer_head *head = page_buffers(page); | |
476 | struct buffer_head *bh = head; | |
477 | ||
478 | /* If they're all mapped and dirty, do it */ | |
479 | page_block = 0; | |
480 | do { | |
481 | BUG_ON(buffer_locked(bh)); | |
482 | if (!buffer_mapped(bh)) { | |
483 | /* | |
484 | * unmapped dirty buffers are created by | |
485 | * __set_page_dirty_buffers -> mmapped data | |
486 | */ | |
487 | if (buffer_dirty(bh)) | |
488 | goto confused; | |
489 | if (first_unmapped == blocks_per_page) | |
490 | first_unmapped = page_block; | |
491 | continue; | |
492 | } | |
493 | ||
494 | if (first_unmapped != blocks_per_page) | |
495 | goto confused; /* hole -> non-hole */ | |
496 | ||
497 | if (!buffer_dirty(bh) || !buffer_uptodate(bh)) | |
498 | goto confused; | |
499 | if (page_block) { | |
500 | if (bh->b_blocknr != blocks[page_block-1] + 1) | |
501 | goto confused; | |
502 | } | |
503 | blocks[page_block++] = bh->b_blocknr; | |
504 | boundary = buffer_boundary(bh); | |
505 | if (boundary) { | |
506 | boundary_block = bh->b_blocknr; | |
507 | boundary_bdev = bh->b_bdev; | |
508 | } | |
509 | bdev = bh->b_bdev; | |
510 | } while ((bh = bh->b_this_page) != head); | |
511 | ||
512 | if (first_unmapped) | |
513 | goto page_is_mapped; | |
514 | ||
515 | /* | |
516 | * Page has buffers, but they are all unmapped. The page was | |
517 | * created by pagein or read over a hole which was handled by | |
518 | * block_read_full_page(). If this address_space is also | |
d4388340 | 519 | * using mpage_readahead then this can rarely happen. |
1da177e4 LT |
520 | */ |
521 | goto confused; | |
522 | } | |
523 | ||
524 | /* | |
525 | * The page has no buffers: map it to disk | |
526 | */ | |
527 | BUG_ON(!PageUptodate(page)); | |
09cbfeaf | 528 | block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits); |
1da177e4 LT |
529 | last_block = (i_size - 1) >> blkbits; |
530 | map_bh.b_page = page; | |
531 | for (page_block = 0; page_block < blocks_per_page; ) { | |
532 | ||
533 | map_bh.b_state = 0; | |
b0cf2321 | 534 | map_bh.b_size = 1 << blkbits; |
0ea97180 | 535 | if (mpd->get_block(inode, block_in_file, &map_bh, 1)) |
1da177e4 LT |
536 | goto confused; |
537 | if (buffer_new(&map_bh)) | |
e64855c6 | 538 | clean_bdev_bh_alias(&map_bh); |
1da177e4 LT |
539 | if (buffer_boundary(&map_bh)) { |
540 | boundary_block = map_bh.b_blocknr; | |
541 | boundary_bdev = map_bh.b_bdev; | |
542 | } | |
543 | if (page_block) { | |
544 | if (map_bh.b_blocknr != blocks[page_block-1] + 1) | |
545 | goto confused; | |
546 | } | |
547 | blocks[page_block++] = map_bh.b_blocknr; | |
548 | boundary = buffer_boundary(&map_bh); | |
549 | bdev = map_bh.b_bdev; | |
550 | if (block_in_file == last_block) | |
551 | break; | |
552 | block_in_file++; | |
553 | } | |
554 | BUG_ON(page_block == 0); | |
555 | ||
556 | first_unmapped = page_block; | |
557 | ||
558 | page_is_mapped: | |
09cbfeaf | 559 | end_index = i_size >> PAGE_SHIFT; |
1da177e4 LT |
560 | if (page->index >= end_index) { |
561 | /* | |
562 | * The page straddles i_size. It must be zeroed out on each | |
2a61aa40 | 563 | * and every writepage invocation because it may be mmapped. |
1da177e4 LT |
564 | * "A file is mapped in multiples of the page size. For a file |
565 | * that is not a multiple of the page size, the remaining memory | |
566 | * is zeroed when mapped, and writes to that region are not | |
567 | * written out to the file." | |
568 | */ | |
09cbfeaf | 569 | unsigned offset = i_size & (PAGE_SIZE - 1); |
1da177e4 LT |
570 | |
571 | if (page->index > end_index || !offset) | |
572 | goto confused; | |
09cbfeaf | 573 | zero_user_segment(page, offset, PAGE_SIZE); |
1da177e4 LT |
574 | } |
575 | ||
576 | /* | |
577 | * This page will go to BIO. Do we need to send this BIO off first? | |
578 | */ | |
0ea97180 | 579 | if (bio && mpd->last_block_in_bio != blocks[0] - 1) |
eed25cd5 | 580 | bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio); |
1da177e4 LT |
581 | |
582 | alloc_new: | |
583 | if (bio == NULL) { | |
47a191fd MW |
584 | if (first_unmapped == blocks_per_page) { |
585 | if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9), | |
f892760a | 586 | page, wbc)) |
47a191fd | 587 | goto out; |
47a191fd | 588 | } |
07888c66 | 589 | bio = bio_alloc(bdev, BIO_MAX_VECS, 0, GFP_NOFS); |
d5f68a42 | 590 | bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9); |
429b3fb0 | 591 | |
b16b1deb | 592 | wbc_init_bio(wbc, bio); |
8e8f9298 | 593 | bio->bi_write_hint = inode->i_write_hint; |
1da177e4 LT |
594 | } |
595 | ||
596 | /* | |
597 | * Must try to add the page before marking the buffer clean or | |
598 | * the confused fail path above (OOM) will be very confused when | |
599 | * it finds all bh marked clean (i.e. it will not write anything) | |
600 | */ | |
34e51a5e | 601 | wbc_account_cgroup_owner(wbc, page, PAGE_SIZE); |
1da177e4 LT |
602 | length = first_unmapped << blkbits; |
603 | if (bio_add_page(bio, page, length, 0) < length) { | |
eed25cd5 | 604 | bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio); |
1da177e4 LT |
605 | goto alloc_new; |
606 | } | |
607 | ||
90768eee | 608 | clean_buffers(page, first_unmapped); |
1da177e4 LT |
609 | |
610 | BUG_ON(PageWriteback(page)); | |
611 | set_page_writeback(page); | |
612 | unlock_page(page); | |
613 | if (boundary || (first_unmapped != blocks_per_page)) { | |
eed25cd5 | 614 | bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio); |
1da177e4 LT |
615 | if (boundary_block) { |
616 | write_boundary_block(boundary_bdev, | |
617 | boundary_block, 1 << blkbits); | |
618 | } | |
619 | } else { | |
0ea97180 | 620 | mpd->last_block_in_bio = blocks[blocks_per_page - 1]; |
1da177e4 LT |
621 | } |
622 | goto out; | |
623 | ||
624 | confused: | |
625 | if (bio) | |
eed25cd5 | 626 | bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio); |
1da177e4 | 627 | |
0ea97180 MS |
628 | if (mpd->use_writepage) { |
629 | ret = mapping->a_ops->writepage(page, wbc); | |
1da177e4 | 630 | } else { |
0ea97180 | 631 | ret = -EAGAIN; |
1da177e4 LT |
632 | goto out; |
633 | } | |
634 | /* | |
635 | * The caller has a ref on the inode, so *mapping is stable | |
636 | */ | |
0ea97180 | 637 | mapping_set_error(mapping, ret); |
1da177e4 | 638 | out: |
0ea97180 MS |
639 | mpd->bio = bio; |
640 | return ret; | |
1da177e4 LT |
641 | } |
642 | ||
643 | /** | |
78a4a50a | 644 | * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them |
1da177e4 LT |
645 | * @mapping: address space structure to write |
646 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
647 | * @get_block: the filesystem's block mapper function. | |
648 | * If this is NULL then use a_ops->writepage. Otherwise, go | |
649 | * direct-to-BIO. | |
650 | * | |
651 | * This is a library function, which implements the writepages() | |
652 | * address_space_operation. | |
653 | * | |
654 | * If a page is already under I/O, generic_writepages() skips it, even | |
655 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
656 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
657 | * and msync() need to guarantee that all the data which was dirty at the time | |
658 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
659 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
660 | * existing IO to complete. | |
661 | */ | |
662 | int | |
663 | mpage_writepages(struct address_space *mapping, | |
664 | struct writeback_control *wbc, get_block_t get_block) | |
1da177e4 | 665 | { |
2ed1a6bc | 666 | struct blk_plug plug; |
0ea97180 MS |
667 | int ret; |
668 | ||
2ed1a6bc JA |
669 | blk_start_plug(&plug); |
670 | ||
0ea97180 MS |
671 | if (!get_block) |
672 | ret = generic_writepages(mapping, wbc); | |
673 | else { | |
674 | struct mpage_data mpd = { | |
675 | .bio = NULL, | |
676 | .last_block_in_bio = 0, | |
677 | .get_block = get_block, | |
678 | .use_writepage = 1, | |
679 | }; | |
680 | ||
681 | ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd); | |
5948edbc | 682 | if (mpd.bio) { |
eed25cd5 | 683 | int op_flags = (wbc->sync_mode == WB_SYNC_ALL ? |
70fd7614 | 684 | REQ_SYNC : 0); |
eed25cd5 | 685 | mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio); |
5948edbc | 686 | } |
1da177e4 | 687 | } |
2ed1a6bc | 688 | blk_finish_plug(&plug); |
1da177e4 LT |
689 | return ret; |
690 | } | |
691 | EXPORT_SYMBOL(mpage_writepages); | |
1da177e4 LT |
692 | |
693 | int mpage_writepage(struct page *page, get_block_t get_block, | |
694 | struct writeback_control *wbc) | |
695 | { | |
0ea97180 MS |
696 | struct mpage_data mpd = { |
697 | .bio = NULL, | |
698 | .last_block_in_bio = 0, | |
699 | .get_block = get_block, | |
700 | .use_writepage = 0, | |
701 | }; | |
702 | int ret = __mpage_writepage(page, wbc, &mpd); | |
5948edbc | 703 | if (mpd.bio) { |
eed25cd5 | 704 | int op_flags = (wbc->sync_mode == WB_SYNC_ALL ? |
70fd7614 | 705 | REQ_SYNC : 0); |
eed25cd5 | 706 | mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio); |
5948edbc | 707 | } |
1da177e4 LT |
708 | return ret; |
709 | } | |
710 | EXPORT_SYMBOL(mpage_writepage); |