Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/mpage.c | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds. | |
5 | * | |
6 | * Contains functions related to preparing and submitting BIOs which contain | |
7 | * multiple pagecache pages. | |
8 | * | |
e1f8e874 | 9 | * 15May2002 Andrew Morton |
1da177e4 LT |
10 | * Initial version |
11 | * 27Jun2002 axboe@suse.de | |
12 | * use bio_add_page() to build bio's just the right size | |
13 | */ | |
14 | ||
15 | #include <linux/kernel.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/mm.h> | |
18 | #include <linux/kdev_t.h> | |
19 | #include <linux/bio.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/buffer_head.h> | |
22 | #include <linux/blkdev.h> | |
23 | #include <linux/highmem.h> | |
24 | #include <linux/prefetch.h> | |
25 | #include <linux/mpage.h> | |
26 | #include <linux/writeback.h> | |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/pagevec.h> | |
29 | ||
30 | /* | |
31 | * I/O completion handler for multipage BIOs. | |
32 | * | |
33 | * The mpage code never puts partial pages into a BIO (except for end-of-file). | |
34 | * If a page does not map to a contiguous run of blocks then it simply falls | |
35 | * back to block_read_full_page(). | |
36 | * | |
37 | * Why is this? If a page's completion depends on a number of different BIOs | |
38 | * which can complete in any order (or at the same time) then determining the | |
39 | * status of that page is hard. See end_buffer_async_read() for the details. | |
40 | * There is no point in duplicating all that complexity. | |
41 | */ | |
6712ecf8 | 42 | static void mpage_end_io_read(struct bio *bio, int err) |
1da177e4 LT |
43 | { |
44 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
45 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
46 | ||
1da177e4 LT |
47 | do { |
48 | struct page *page = bvec->bv_page; | |
49 | ||
50 | if (--bvec >= bio->bi_io_vec) | |
51 | prefetchw(&bvec->bv_page->flags); | |
52 | ||
53 | if (uptodate) { | |
54 | SetPageUptodate(page); | |
55 | } else { | |
56 | ClearPageUptodate(page); | |
57 | SetPageError(page); | |
58 | } | |
59 | unlock_page(page); | |
60 | } while (bvec >= bio->bi_io_vec); | |
61 | bio_put(bio); | |
1da177e4 LT |
62 | } |
63 | ||
6712ecf8 | 64 | static void mpage_end_io_write(struct bio *bio, int err) |
1da177e4 LT |
65 | { |
66 | const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags); | |
67 | struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1; | |
68 | ||
1da177e4 LT |
69 | do { |
70 | struct page *page = bvec->bv_page; | |
71 | ||
72 | if (--bvec >= bio->bi_io_vec) | |
73 | prefetchw(&bvec->bv_page->flags); | |
74 | ||
854715be | 75 | if (!uptodate){ |
1da177e4 | 76 | SetPageError(page); |
854715be QF |
77 | if (page->mapping) |
78 | set_bit(AS_EIO, &page->mapping->flags); | |
79 | } | |
1da177e4 LT |
80 | end_page_writeback(page); |
81 | } while (bvec >= bio->bi_io_vec); | |
82 | bio_put(bio); | |
1da177e4 LT |
83 | } |
84 | ||
29a814d2 | 85 | struct bio *mpage_bio_submit(int rw, struct bio *bio) |
1da177e4 LT |
86 | { |
87 | bio->bi_end_io = mpage_end_io_read; | |
88 | if (rw == WRITE) | |
89 | bio->bi_end_io = mpage_end_io_write; | |
90 | submit_bio(rw, bio); | |
91 | return NULL; | |
92 | } | |
29a814d2 | 93 | EXPORT_SYMBOL(mpage_bio_submit); |
1da177e4 LT |
94 | |
95 | static struct bio * | |
96 | mpage_alloc(struct block_device *bdev, | |
97 | sector_t first_sector, int nr_vecs, | |
dd0fc66f | 98 | gfp_t gfp_flags) |
1da177e4 LT |
99 | { |
100 | struct bio *bio; | |
101 | ||
102 | bio = bio_alloc(gfp_flags, nr_vecs); | |
103 | ||
104 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { | |
105 | while (!bio && (nr_vecs /= 2)) | |
106 | bio = bio_alloc(gfp_flags, nr_vecs); | |
107 | } | |
108 | ||
109 | if (bio) { | |
110 | bio->bi_bdev = bdev; | |
111 | bio->bi_sector = first_sector; | |
112 | } | |
113 | return bio; | |
114 | } | |
115 | ||
116 | /* | |
117 | * support function for mpage_readpages. The fs supplied get_block might | |
118 | * return an up to date buffer. This is used to map that buffer into | |
119 | * the page, which allows readpage to avoid triggering a duplicate call | |
120 | * to get_block. | |
121 | * | |
122 | * The idea is to avoid adding buffers to pages that don't already have | |
123 | * them. So when the buffer is up to date and the page size == block size, | |
124 | * this marks the page up to date instead of adding new buffers. | |
125 | */ | |
126 | static void | |
127 | map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block) | |
128 | { | |
129 | struct inode *inode = page->mapping->host; | |
130 | struct buffer_head *page_bh, *head; | |
131 | int block = 0; | |
132 | ||
133 | if (!page_has_buffers(page)) { | |
134 | /* | |
135 | * don't make any buffers if there is only one buffer on | |
136 | * the page and the page just needs to be set up to date | |
137 | */ | |
138 | if (inode->i_blkbits == PAGE_CACHE_SHIFT && | |
139 | buffer_uptodate(bh)) { | |
140 | SetPageUptodate(page); | |
141 | return; | |
142 | } | |
143 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
144 | } | |
145 | head = page_buffers(page); | |
146 | page_bh = head; | |
147 | do { | |
148 | if (block == page_block) { | |
149 | page_bh->b_state = bh->b_state; | |
150 | page_bh->b_bdev = bh->b_bdev; | |
151 | page_bh->b_blocknr = bh->b_blocknr; | |
152 | break; | |
153 | } | |
154 | page_bh = page_bh->b_this_page; | |
155 | block++; | |
156 | } while (page_bh != head); | |
157 | } | |
158 | ||
fa30bd05 BP |
159 | /* |
160 | * This is the worker routine which does all the work of mapping the disk | |
161 | * blocks and constructs largest possible bios, submits them for IO if the | |
162 | * blocks are not contiguous on the disk. | |
163 | * | |
164 | * We pass a buffer_head back and forth and use its buffer_mapped() flag to | |
165 | * represent the validity of its disk mapping and to decide when to do the next | |
166 | * get_block() call. | |
167 | */ | |
1da177e4 LT |
168 | static struct bio * |
169 | do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages, | |
fa30bd05 BP |
170 | sector_t *last_block_in_bio, struct buffer_head *map_bh, |
171 | unsigned long *first_logical_block, get_block_t get_block) | |
1da177e4 LT |
172 | { |
173 | struct inode *inode = page->mapping->host; | |
174 | const unsigned blkbits = inode->i_blkbits; | |
175 | const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits; | |
176 | const unsigned blocksize = 1 << blkbits; | |
177 | sector_t block_in_file; | |
178 | sector_t last_block; | |
fa30bd05 | 179 | sector_t last_block_in_file; |
1da177e4 LT |
180 | sector_t blocks[MAX_BUF_PER_PAGE]; |
181 | unsigned page_block; | |
182 | unsigned first_hole = blocks_per_page; | |
183 | struct block_device *bdev = NULL; | |
1da177e4 LT |
184 | int length; |
185 | int fully_mapped = 1; | |
fa30bd05 BP |
186 | unsigned nblocks; |
187 | unsigned relative_block; | |
1da177e4 LT |
188 | |
189 | if (page_has_buffers(page)) | |
190 | goto confused; | |
191 | ||
54b21a79 | 192 | block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits); |
fa30bd05 BP |
193 | last_block = block_in_file + nr_pages * blocks_per_page; |
194 | last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits; | |
195 | if (last_block > last_block_in_file) | |
196 | last_block = last_block_in_file; | |
197 | page_block = 0; | |
198 | ||
199 | /* | |
200 | * Map blocks using the result from the previous get_blocks call first. | |
201 | */ | |
202 | nblocks = map_bh->b_size >> blkbits; | |
203 | if (buffer_mapped(map_bh) && block_in_file > *first_logical_block && | |
204 | block_in_file < (*first_logical_block + nblocks)) { | |
205 | unsigned map_offset = block_in_file - *first_logical_block; | |
206 | unsigned last = nblocks - map_offset; | |
207 | ||
208 | for (relative_block = 0; ; relative_block++) { | |
209 | if (relative_block == last) { | |
210 | clear_buffer_mapped(map_bh); | |
211 | break; | |
212 | } | |
213 | if (page_block == blocks_per_page) | |
214 | break; | |
215 | blocks[page_block] = map_bh->b_blocknr + map_offset + | |
216 | relative_block; | |
217 | page_block++; | |
218 | block_in_file++; | |
219 | } | |
220 | bdev = map_bh->b_bdev; | |
221 | } | |
222 | ||
223 | /* | |
224 | * Then do more get_blocks calls until we are done with this page. | |
225 | */ | |
226 | map_bh->b_page = page; | |
227 | while (page_block < blocks_per_page) { | |
228 | map_bh->b_state = 0; | |
229 | map_bh->b_size = 0; | |
1da177e4 | 230 | |
1da177e4 | 231 | if (block_in_file < last_block) { |
fa30bd05 BP |
232 | map_bh->b_size = (last_block-block_in_file) << blkbits; |
233 | if (get_block(inode, block_in_file, map_bh, 0)) | |
1da177e4 | 234 | goto confused; |
fa30bd05 | 235 | *first_logical_block = block_in_file; |
1da177e4 LT |
236 | } |
237 | ||
fa30bd05 | 238 | if (!buffer_mapped(map_bh)) { |
1da177e4 LT |
239 | fully_mapped = 0; |
240 | if (first_hole == blocks_per_page) | |
241 | first_hole = page_block; | |
fa30bd05 BP |
242 | page_block++; |
243 | block_in_file++; | |
244 | clear_buffer_mapped(map_bh); | |
1da177e4 LT |
245 | continue; |
246 | } | |
247 | ||
248 | /* some filesystems will copy data into the page during | |
249 | * the get_block call, in which case we don't want to | |
250 | * read it again. map_buffer_to_page copies the data | |
251 | * we just collected from get_block into the page's buffers | |
252 | * so readpage doesn't have to repeat the get_block call | |
253 | */ | |
fa30bd05 BP |
254 | if (buffer_uptodate(map_bh)) { |
255 | map_buffer_to_page(page, map_bh, page_block); | |
1da177e4 LT |
256 | goto confused; |
257 | } | |
258 | ||
259 | if (first_hole != blocks_per_page) | |
260 | goto confused; /* hole -> non-hole */ | |
261 | ||
262 | /* Contiguous blocks? */ | |
fa30bd05 | 263 | if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1) |
1da177e4 | 264 | goto confused; |
fa30bd05 BP |
265 | nblocks = map_bh->b_size >> blkbits; |
266 | for (relative_block = 0; ; relative_block++) { | |
267 | if (relative_block == nblocks) { | |
268 | clear_buffer_mapped(map_bh); | |
269 | break; | |
270 | } else if (page_block == blocks_per_page) | |
271 | break; | |
272 | blocks[page_block] = map_bh->b_blocknr+relative_block; | |
273 | page_block++; | |
274 | block_in_file++; | |
275 | } | |
276 | bdev = map_bh->b_bdev; | |
1da177e4 LT |
277 | } |
278 | ||
279 | if (first_hole != blocks_per_page) { | |
eebd2aa3 | 280 | zero_user_segment(page, first_hole << blkbits, PAGE_CACHE_SIZE); |
1da177e4 LT |
281 | if (first_hole == 0) { |
282 | SetPageUptodate(page); | |
283 | unlock_page(page); | |
284 | goto out; | |
285 | } | |
286 | } else if (fully_mapped) { | |
287 | SetPageMappedToDisk(page); | |
288 | } | |
289 | ||
290 | /* | |
291 | * This page will go to BIO. Do we need to send this BIO off first? | |
292 | */ | |
293 | if (bio && (*last_block_in_bio != blocks[0] - 1)) | |
294 | bio = mpage_bio_submit(READ, bio); | |
295 | ||
296 | alloc_new: | |
297 | if (bio == NULL) { | |
298 | bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9), | |
299 | min_t(int, nr_pages, bio_get_nr_vecs(bdev)), | |
300 | GFP_KERNEL); | |
301 | if (bio == NULL) | |
302 | goto confused; | |
303 | } | |
304 | ||
305 | length = first_hole << blkbits; | |
306 | if (bio_add_page(bio, page, length, 0) < length) { | |
307 | bio = mpage_bio_submit(READ, bio); | |
308 | goto alloc_new; | |
309 | } | |
310 | ||
38c8e618 MS |
311 | relative_block = block_in_file - *first_logical_block; |
312 | nblocks = map_bh->b_size >> blkbits; | |
313 | if ((buffer_boundary(map_bh) && relative_block == nblocks) || | |
314 | (first_hole != blocks_per_page)) | |
1da177e4 LT |
315 | bio = mpage_bio_submit(READ, bio); |
316 | else | |
317 | *last_block_in_bio = blocks[blocks_per_page - 1]; | |
318 | out: | |
319 | return bio; | |
320 | ||
321 | confused: | |
322 | if (bio) | |
323 | bio = mpage_bio_submit(READ, bio); | |
324 | if (!PageUptodate(page)) | |
325 | block_read_full_page(page, get_block); | |
326 | else | |
327 | unlock_page(page); | |
328 | goto out; | |
329 | } | |
330 | ||
67be2dd1 | 331 | /** |
78a4a50a | 332 | * mpage_readpages - populate an address space with some pages & start reads against them |
67be2dd1 MW |
333 | * @mapping: the address_space |
334 | * @pages: The address of a list_head which contains the target pages. These | |
335 | * pages have their ->index populated and are otherwise uninitialised. | |
67be2dd1 MW |
336 | * The page at @pages->prev has the lowest file offset, and reads should be |
337 | * issued in @pages->prev to @pages->next order. | |
67be2dd1 MW |
338 | * @nr_pages: The number of pages at *@pages |
339 | * @get_block: The filesystem's block mapper function. | |
340 | * | |
341 | * This function walks the pages and the blocks within each page, building and | |
342 | * emitting large BIOs. | |
343 | * | |
344 | * If anything unusual happens, such as: | |
345 | * | |
346 | * - encountering a page which has buffers | |
347 | * - encountering a page which has a non-hole after a hole | |
348 | * - encountering a page with non-contiguous blocks | |
349 | * | |
350 | * then this code just gives up and calls the buffer_head-based read function. | |
351 | * It does handle a page which has holes at the end - that is a common case: | |
352 | * the end-of-file on blocksize < PAGE_CACHE_SIZE setups. | |
353 | * | |
354 | * BH_Boundary explanation: | |
355 | * | |
356 | * There is a problem. The mpage read code assembles several pages, gets all | |
357 | * their disk mappings, and then submits them all. That's fine, but obtaining | |
358 | * the disk mappings may require I/O. Reads of indirect blocks, for example. | |
359 | * | |
360 | * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be | |
361 | * submitted in the following order: | |
362 | * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16 | |
78a4a50a | 363 | * |
67be2dd1 MW |
364 | * because the indirect block has to be read to get the mappings of blocks |
365 | * 13,14,15,16. Obviously, this impacts performance. | |
366 | * | |
367 | * So what we do it to allow the filesystem's get_block() function to set | |
368 | * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block | |
369 | * after this one will require I/O against a block which is probably close to | |
370 | * this one. So you should push what I/O you have currently accumulated. | |
371 | * | |
372 | * This all causes the disk requests to be issued in the correct order. | |
373 | */ | |
1da177e4 LT |
374 | int |
375 | mpage_readpages(struct address_space *mapping, struct list_head *pages, | |
376 | unsigned nr_pages, get_block_t get_block) | |
377 | { | |
378 | struct bio *bio = NULL; | |
379 | unsigned page_idx; | |
380 | sector_t last_block_in_bio = 0; | |
fa30bd05 BP |
381 | struct buffer_head map_bh; |
382 | unsigned long first_logical_block = 0; | |
1da177e4 | 383 | |
fa30bd05 | 384 | clear_buffer_mapped(&map_bh); |
1da177e4 LT |
385 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
386 | struct page *page = list_entry(pages->prev, struct page, lru); | |
387 | ||
388 | prefetchw(&page->flags); | |
389 | list_del(&page->lru); | |
eb2be189 | 390 | if (!add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
391 | page->index, GFP_KERNEL)) { |
392 | bio = do_mpage_readpage(bio, page, | |
393 | nr_pages - page_idx, | |
fa30bd05 BP |
394 | &last_block_in_bio, &map_bh, |
395 | &first_logical_block, | |
396 | get_block); | |
1da177e4 | 397 | } |
eb2be189 | 398 | page_cache_release(page); |
1da177e4 | 399 | } |
1da177e4 LT |
400 | BUG_ON(!list_empty(pages)); |
401 | if (bio) | |
402 | mpage_bio_submit(READ, bio); | |
403 | return 0; | |
404 | } | |
405 | EXPORT_SYMBOL(mpage_readpages); | |
406 | ||
407 | /* | |
408 | * This isn't called much at all | |
409 | */ | |
410 | int mpage_readpage(struct page *page, get_block_t get_block) | |
411 | { | |
412 | struct bio *bio = NULL; | |
413 | sector_t last_block_in_bio = 0; | |
fa30bd05 BP |
414 | struct buffer_head map_bh; |
415 | unsigned long first_logical_block = 0; | |
1da177e4 | 416 | |
fa30bd05 BP |
417 | clear_buffer_mapped(&map_bh); |
418 | bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio, | |
419 | &map_bh, &first_logical_block, get_block); | |
1da177e4 LT |
420 | if (bio) |
421 | mpage_bio_submit(READ, bio); | |
422 | return 0; | |
423 | } | |
424 | EXPORT_SYMBOL(mpage_readpage); | |
425 | ||
426 | /* | |
427 | * Writing is not so simple. | |
428 | * | |
429 | * If the page has buffers then they will be used for obtaining the disk | |
430 | * mapping. We only support pages which are fully mapped-and-dirty, with a | |
431 | * special case for pages which are unmapped at the end: end-of-file. | |
432 | * | |
433 | * If the page has no buffers (preferred) then the page is mapped here. | |
434 | * | |
435 | * If all blocks are found to be contiguous then the page can go into the | |
436 | * BIO. Otherwise fall back to the mapping's writepage(). | |
437 | * | |
438 | * FIXME: This code wants an estimate of how many pages are still to be | |
439 | * written, so it can intelligently allocate a suitably-sized BIO. For now, | |
440 | * just allocate full-size (16-page) BIOs. | |
441 | */ | |
0ea97180 | 442 | |
29a814d2 AT |
443 | int __mpage_writepage(struct page *page, struct writeback_control *wbc, |
444 | void *data) | |
1da177e4 | 445 | { |
0ea97180 MS |
446 | struct mpage_data *mpd = data; |
447 | struct bio *bio = mpd->bio; | |
1da177e4 LT |
448 | struct address_space *mapping = page->mapping; |
449 | struct inode *inode = page->mapping->host; | |
450 | const unsigned blkbits = inode->i_blkbits; | |
451 | unsigned long end_index; | |
452 | const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits; | |
453 | sector_t last_block; | |
454 | sector_t block_in_file; | |
455 | sector_t blocks[MAX_BUF_PER_PAGE]; | |
456 | unsigned page_block; | |
457 | unsigned first_unmapped = blocks_per_page; | |
458 | struct block_device *bdev = NULL; | |
459 | int boundary = 0; | |
460 | sector_t boundary_block = 0; | |
461 | struct block_device *boundary_bdev = NULL; | |
462 | int length; | |
463 | struct buffer_head map_bh; | |
464 | loff_t i_size = i_size_read(inode); | |
0ea97180 | 465 | int ret = 0; |
1da177e4 LT |
466 | |
467 | if (page_has_buffers(page)) { | |
468 | struct buffer_head *head = page_buffers(page); | |
469 | struct buffer_head *bh = head; | |
470 | ||
471 | /* If they're all mapped and dirty, do it */ | |
472 | page_block = 0; | |
473 | do { | |
474 | BUG_ON(buffer_locked(bh)); | |
475 | if (!buffer_mapped(bh)) { | |
476 | /* | |
477 | * unmapped dirty buffers are created by | |
478 | * __set_page_dirty_buffers -> mmapped data | |
479 | */ | |
480 | if (buffer_dirty(bh)) | |
481 | goto confused; | |
482 | if (first_unmapped == blocks_per_page) | |
483 | first_unmapped = page_block; | |
484 | continue; | |
485 | } | |
486 | ||
487 | if (first_unmapped != blocks_per_page) | |
488 | goto confused; /* hole -> non-hole */ | |
489 | ||
490 | if (!buffer_dirty(bh) || !buffer_uptodate(bh)) | |
491 | goto confused; | |
492 | if (page_block) { | |
493 | if (bh->b_blocknr != blocks[page_block-1] + 1) | |
494 | goto confused; | |
495 | } | |
496 | blocks[page_block++] = bh->b_blocknr; | |
497 | boundary = buffer_boundary(bh); | |
498 | if (boundary) { | |
499 | boundary_block = bh->b_blocknr; | |
500 | boundary_bdev = bh->b_bdev; | |
501 | } | |
502 | bdev = bh->b_bdev; | |
503 | } while ((bh = bh->b_this_page) != head); | |
504 | ||
505 | if (first_unmapped) | |
506 | goto page_is_mapped; | |
507 | ||
508 | /* | |
509 | * Page has buffers, but they are all unmapped. The page was | |
510 | * created by pagein or read over a hole which was handled by | |
511 | * block_read_full_page(). If this address_space is also | |
512 | * using mpage_readpages then this can rarely happen. | |
513 | */ | |
514 | goto confused; | |
515 | } | |
516 | ||
517 | /* | |
518 | * The page has no buffers: map it to disk | |
519 | */ | |
520 | BUG_ON(!PageUptodate(page)); | |
54b21a79 | 521 | block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits); |
1da177e4 LT |
522 | last_block = (i_size - 1) >> blkbits; |
523 | map_bh.b_page = page; | |
524 | for (page_block = 0; page_block < blocks_per_page; ) { | |
525 | ||
526 | map_bh.b_state = 0; | |
b0cf2321 | 527 | map_bh.b_size = 1 << blkbits; |
0ea97180 | 528 | if (mpd->get_block(inode, block_in_file, &map_bh, 1)) |
1da177e4 LT |
529 | goto confused; |
530 | if (buffer_new(&map_bh)) | |
531 | unmap_underlying_metadata(map_bh.b_bdev, | |
532 | map_bh.b_blocknr); | |
533 | if (buffer_boundary(&map_bh)) { | |
534 | boundary_block = map_bh.b_blocknr; | |
535 | boundary_bdev = map_bh.b_bdev; | |
536 | } | |
537 | if (page_block) { | |
538 | if (map_bh.b_blocknr != blocks[page_block-1] + 1) | |
539 | goto confused; | |
540 | } | |
541 | blocks[page_block++] = map_bh.b_blocknr; | |
542 | boundary = buffer_boundary(&map_bh); | |
543 | bdev = map_bh.b_bdev; | |
544 | if (block_in_file == last_block) | |
545 | break; | |
546 | block_in_file++; | |
547 | } | |
548 | BUG_ON(page_block == 0); | |
549 | ||
550 | first_unmapped = page_block; | |
551 | ||
552 | page_is_mapped: | |
553 | end_index = i_size >> PAGE_CACHE_SHIFT; | |
554 | if (page->index >= end_index) { | |
555 | /* | |
556 | * The page straddles i_size. It must be zeroed out on each | |
557 | * and every writepage invokation because it may be mmapped. | |
558 | * "A file is mapped in multiples of the page size. For a file | |
559 | * that is not a multiple of the page size, the remaining memory | |
560 | * is zeroed when mapped, and writes to that region are not | |
561 | * written out to the file." | |
562 | */ | |
563 | unsigned offset = i_size & (PAGE_CACHE_SIZE - 1); | |
1da177e4 LT |
564 | |
565 | if (page->index > end_index || !offset) | |
566 | goto confused; | |
eebd2aa3 | 567 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); |
1da177e4 LT |
568 | } |
569 | ||
570 | /* | |
571 | * This page will go to BIO. Do we need to send this BIO off first? | |
572 | */ | |
0ea97180 | 573 | if (bio && mpd->last_block_in_bio != blocks[0] - 1) |
1da177e4 LT |
574 | bio = mpage_bio_submit(WRITE, bio); |
575 | ||
576 | alloc_new: | |
577 | if (bio == NULL) { | |
578 | bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9), | |
579 | bio_get_nr_vecs(bdev), GFP_NOFS|__GFP_HIGH); | |
580 | if (bio == NULL) | |
581 | goto confused; | |
582 | } | |
583 | ||
584 | /* | |
585 | * Must try to add the page before marking the buffer clean or | |
586 | * the confused fail path above (OOM) will be very confused when | |
587 | * it finds all bh marked clean (i.e. it will not write anything) | |
588 | */ | |
589 | length = first_unmapped << blkbits; | |
590 | if (bio_add_page(bio, page, length, 0) < length) { | |
591 | bio = mpage_bio_submit(WRITE, bio); | |
592 | goto alloc_new; | |
593 | } | |
594 | ||
595 | /* | |
596 | * OK, we have our BIO, so we can now mark the buffers clean. Make | |
597 | * sure to only clean buffers which we know we'll be writing. | |
598 | */ | |
599 | if (page_has_buffers(page)) { | |
600 | struct buffer_head *head = page_buffers(page); | |
601 | struct buffer_head *bh = head; | |
602 | unsigned buffer_counter = 0; | |
603 | ||
604 | do { | |
605 | if (buffer_counter++ == first_unmapped) | |
606 | break; | |
607 | clear_buffer_dirty(bh); | |
608 | bh = bh->b_this_page; | |
609 | } while (bh != head); | |
610 | ||
611 | /* | |
612 | * we cannot drop the bh if the page is not uptodate | |
613 | * or a concurrent readpage would fail to serialize with the bh | |
614 | * and it would read from disk before we reach the platter. | |
615 | */ | |
616 | if (buffer_heads_over_limit && PageUptodate(page)) | |
617 | try_to_free_buffers(page); | |
618 | } | |
619 | ||
620 | BUG_ON(PageWriteback(page)); | |
621 | set_page_writeback(page); | |
622 | unlock_page(page); | |
623 | if (boundary || (first_unmapped != blocks_per_page)) { | |
624 | bio = mpage_bio_submit(WRITE, bio); | |
625 | if (boundary_block) { | |
626 | write_boundary_block(boundary_bdev, | |
627 | boundary_block, 1 << blkbits); | |
628 | } | |
629 | } else { | |
0ea97180 | 630 | mpd->last_block_in_bio = blocks[blocks_per_page - 1]; |
1da177e4 LT |
631 | } |
632 | goto out; | |
633 | ||
634 | confused: | |
635 | if (bio) | |
636 | bio = mpage_bio_submit(WRITE, bio); | |
637 | ||
0ea97180 MS |
638 | if (mpd->use_writepage) { |
639 | ret = mapping->a_ops->writepage(page, wbc); | |
1da177e4 | 640 | } else { |
0ea97180 | 641 | ret = -EAGAIN; |
1da177e4 LT |
642 | goto out; |
643 | } | |
644 | /* | |
645 | * The caller has a ref on the inode, so *mapping is stable | |
646 | */ | |
0ea97180 | 647 | mapping_set_error(mapping, ret); |
1da177e4 | 648 | out: |
0ea97180 MS |
649 | mpd->bio = bio; |
650 | return ret; | |
1da177e4 | 651 | } |
29a814d2 | 652 | EXPORT_SYMBOL(__mpage_writepage); |
1da177e4 LT |
653 | |
654 | /** | |
78a4a50a | 655 | * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them |
1da177e4 LT |
656 | * @mapping: address space structure to write |
657 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
658 | * @get_block: the filesystem's block mapper function. | |
659 | * If this is NULL then use a_ops->writepage. Otherwise, go | |
660 | * direct-to-BIO. | |
661 | * | |
662 | * This is a library function, which implements the writepages() | |
663 | * address_space_operation. | |
664 | * | |
665 | * If a page is already under I/O, generic_writepages() skips it, even | |
666 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
667 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
668 | * and msync() need to guarantee that all the data which was dirty at the time | |
669 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
670 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
671 | * existing IO to complete. | |
672 | */ | |
673 | int | |
674 | mpage_writepages(struct address_space *mapping, | |
675 | struct writeback_control *wbc, get_block_t get_block) | |
1da177e4 | 676 | { |
0ea97180 MS |
677 | int ret; |
678 | ||
679 | if (!get_block) | |
680 | ret = generic_writepages(mapping, wbc); | |
681 | else { | |
682 | struct mpage_data mpd = { | |
683 | .bio = NULL, | |
684 | .last_block_in_bio = 0, | |
685 | .get_block = get_block, | |
686 | .use_writepage = 1, | |
687 | }; | |
688 | ||
689 | ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd); | |
690 | if (mpd.bio) | |
691 | mpage_bio_submit(WRITE, mpd.bio); | |
1da177e4 | 692 | } |
1da177e4 LT |
693 | return ret; |
694 | } | |
695 | EXPORT_SYMBOL(mpage_writepages); | |
1da177e4 LT |
696 | |
697 | int mpage_writepage(struct page *page, get_block_t get_block, | |
698 | struct writeback_control *wbc) | |
699 | { | |
0ea97180 MS |
700 | struct mpage_data mpd = { |
701 | .bio = NULL, | |
702 | .last_block_in_bio = 0, | |
703 | .get_block = get_block, | |
704 | .use_writepage = 0, | |
705 | }; | |
706 | int ret = __mpage_writepage(page, wbc, &mpd); | |
707 | if (mpd.bio) | |
708 | mpage_bio_submit(WRITE, mpd.bio); | |
1da177e4 LT |
709 | return ret; |
710 | } | |
711 | EXPORT_SYMBOL(mpage_writepage); |