Root/
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 | * |
9 | * 15May2002 Andrew Morton |
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 | */ |
42 | static void mpage_end_io_read(struct bio *bio, int err) |
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 | |
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); |
62 | } |
63 | |
64 | static void mpage_end_io_write(struct bio *bio, int err) |
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 | |
69 | do { |
70 | struct page *page = bvec->bv_page; |
71 | |
72 | if (--bvec >= bio->bi_io_vec) |
73 | prefetchw(&bvec->bv_page->flags); |
74 | |
75 | if (!uptodate){ |
76 | SetPageError(page); |
77 | if (page->mapping) |
78 | set_bit(AS_EIO, &page->mapping->flags); |
79 | } |
80 | end_page_writeback(page); |
81 | } while (bvec >= bio->bi_io_vec); |
82 | bio_put(bio); |
83 | } |
84 | |
85 | static struct bio *mpage_bio_submit(int rw, struct bio *bio) |
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 | } |
93 | |
94 | static struct bio * |
95 | mpage_alloc(struct block_device *bdev, |
96 | sector_t first_sector, int nr_vecs, |
97 | gfp_t gfp_flags) |
98 | { |
99 | struct bio *bio; |
100 | |
101 | bio = bio_alloc(gfp_flags, nr_vecs); |
102 | |
103 | if (bio == NULL && (current->flags & PF_MEMALLOC)) { |
104 | while (!bio && (nr_vecs /= 2)) |
105 | bio = bio_alloc(gfp_flags, nr_vecs); |
106 | } |
107 | |
108 | if (bio) { |
109 | bio->bi_bdev = bdev; |
110 | bio->bi_sector = first_sector; |
111 | } |
112 | return bio; |
113 | } |
114 | |
115 | /* |
116 | * support function for mpage_readpages. The fs supplied get_block might |
117 | * return an up to date buffer. This is used to map that buffer into |
118 | * the page, which allows readpage to avoid triggering a duplicate call |
119 | * to get_block. |
120 | * |
121 | * The idea is to avoid adding buffers to pages that don't already have |
122 | * them. So when the buffer is up to date and the page size == block size, |
123 | * this marks the page up to date instead of adding new buffers. |
124 | */ |
125 | static void |
126 | map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block) |
127 | { |
128 | struct inode *inode = page->mapping->host; |
129 | struct buffer_head *page_bh, *head; |
130 | int block = 0; |
131 | |
132 | if (!page_has_buffers(page)) { |
133 | /* |
134 | * don't make any buffers if there is only one buffer on |
135 | * the page and the page just needs to be set up to date |
136 | */ |
137 | if (inode->i_blkbits == PAGE_CACHE_SHIFT && |
138 | buffer_uptodate(bh)) { |
139 | SetPageUptodate(page); |
140 | return; |
141 | } |
142 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); |
143 | } |
144 | head = page_buffers(page); |
145 | page_bh = head; |
146 | do { |
147 | if (block == page_block) { |
148 | page_bh->b_state = bh->b_state; |
149 | page_bh->b_bdev = bh->b_bdev; |
150 | page_bh->b_blocknr = bh->b_blocknr; |
151 | break; |
152 | } |
153 | page_bh = page_bh->b_this_page; |
154 | block++; |
155 | } while (page_bh != head); |
156 | } |
157 | |
158 | /* |
159 | * This is the worker routine which does all the work of mapping the disk |
160 | * blocks and constructs largest possible bios, submits them for IO if the |
161 | * blocks are not contiguous on the disk. |
162 | * |
163 | * We pass a buffer_head back and forth and use its buffer_mapped() flag to |
164 | * represent the validity of its disk mapping and to decide when to do the next |
165 | * get_block() call. |
166 | */ |
167 | static struct bio * |
168 | do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages, |
169 | sector_t *last_block_in_bio, struct buffer_head *map_bh, |
170 | unsigned long *first_logical_block, get_block_t get_block) |
171 | { |
172 | struct inode *inode = page->mapping->host; |
173 | const unsigned blkbits = inode->i_blkbits; |
174 | const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits; |
175 | const unsigned blocksize = 1 << blkbits; |
176 | sector_t block_in_file; |
177 | sector_t last_block; |
178 | sector_t last_block_in_file; |
179 | sector_t blocks[MAX_BUF_PER_PAGE]; |
180 | unsigned page_block; |
181 | unsigned first_hole = blocks_per_page; |
182 | struct block_device *bdev = NULL; |
183 | int length; |
184 | int fully_mapped = 1; |
185 | unsigned nblocks; |
186 | unsigned relative_block; |
187 | |
188 | if (page_has_buffers(page)) |
189 | goto confused; |
190 | |
191 | block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits); |
192 | last_block = block_in_file + nr_pages * blocks_per_page; |
193 | last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits; |
194 | if (last_block > last_block_in_file) |
195 | last_block = last_block_in_file; |
196 | page_block = 0; |
197 | |
198 | /* |
199 | * Map blocks using the result from the previous get_blocks call first. |
200 | */ |
201 | nblocks = map_bh->b_size >> blkbits; |
202 | if (buffer_mapped(map_bh) && block_in_file > *first_logical_block && |
203 | block_in_file < (*first_logical_block + nblocks)) { |
204 | unsigned map_offset = block_in_file - *first_logical_block; |
205 | unsigned last = nblocks - map_offset; |
206 | |
207 | for (relative_block = 0; ; relative_block++) { |
208 | if (relative_block == last) { |
209 | clear_buffer_mapped(map_bh); |
210 | break; |
211 | } |
212 | if (page_block == blocks_per_page) |
213 | break; |
214 | blocks[page_block] = map_bh->b_blocknr + map_offset + |
215 | relative_block; |
216 | page_block++; |
217 | block_in_file++; |
218 | } |
219 | bdev = map_bh->b_bdev; |
220 | } |
221 | |
222 | /* |
223 | * Then do more get_blocks calls until we are done with this page. |
224 | */ |
225 | map_bh->b_page = page; |
226 | while (page_block < blocks_per_page) { |
227 | map_bh->b_state = 0; |
228 | map_bh->b_size = 0; |
229 | |
230 | if (block_in_file < last_block) { |
231 | map_bh->b_size = (last_block-block_in_file) << blkbits; |
232 | if (get_block(inode, block_in_file, map_bh, 0)) |
233 | goto confused; |
234 | *first_logical_block = block_in_file; |
235 | } |
236 | |
237 | if (!buffer_mapped(map_bh)) { |
238 | fully_mapped = 0; |
239 | if (first_hole == blocks_per_page) |
240 | first_hole = page_block; |
241 | page_block++; |
242 | block_in_file++; |
243 | continue; |
244 | } |
245 | |
246 | /* some filesystems will copy data into the page during |
247 | * the get_block call, in which case we don't want to |
248 | * read it again. map_buffer_to_page copies the data |
249 | * we just collected from get_block into the page's buffers |
250 | * so readpage doesn't have to repeat the get_block call |
251 | */ |
252 | if (buffer_uptodate(map_bh)) { |
253 | map_buffer_to_page(page, map_bh, page_block); |
254 | goto confused; |
255 | } |
256 | |
257 | if (first_hole != blocks_per_page) |
258 | goto confused; /* hole -> non-hole */ |
259 | |
260 | /* Contiguous blocks? */ |
261 | if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1) |
262 | goto confused; |
263 | nblocks = map_bh->b_size >> blkbits; |
264 | for (relative_block = 0; ; relative_block++) { |
265 | if (relative_block == nblocks) { |
266 | clear_buffer_mapped(map_bh); |
267 | break; |
268 | } else if (page_block == blocks_per_page) |
269 | break; |
270 | blocks[page_block] = map_bh->b_blocknr+relative_block; |
271 | page_block++; |
272 | block_in_file++; |
273 | } |
274 | bdev = map_bh->b_bdev; |
275 | } |
276 | |
277 | if (first_hole != blocks_per_page) { |
278 | zero_user_segment(page, first_hole << blkbits, PAGE_CACHE_SIZE); |
279 | if (first_hole == 0) { |
280 | SetPageUptodate(page); |
281 | unlock_page(page); |
282 | goto out; |
283 | } |
284 | } else if (fully_mapped) { |
285 | SetPageMappedToDisk(page); |
286 | } |
287 | |
288 | /* |
289 | * This page will go to BIO. Do we need to send this BIO off first? |
290 | */ |
291 | if (bio && (*last_block_in_bio != blocks[0] - 1)) |
292 | bio = mpage_bio_submit(READ, bio); |
293 | |
294 | alloc_new: |
295 | if (bio == NULL) { |
296 | bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9), |
297 | min_t(int, nr_pages, bio_get_nr_vecs(bdev)), |
298 | GFP_KERNEL); |
299 | if (bio == NULL) |
300 | goto confused; |
301 | } |
302 | |
303 | length = first_hole << blkbits; |
304 | if (bio_add_page(bio, page, length, 0) < length) { |
305 | bio = mpage_bio_submit(READ, bio); |
306 | goto alloc_new; |
307 | } |
308 | |
309 | relative_block = block_in_file - *first_logical_block; |
310 | nblocks = map_bh->b_size >> blkbits; |
311 | if ((buffer_boundary(map_bh) && relative_block == nblocks) || |
312 | (first_hole != blocks_per_page)) |
313 | bio = mpage_bio_submit(READ, bio); |
314 | else |
315 | *last_block_in_bio = blocks[blocks_per_page - 1]; |
316 | out: |
317 | return bio; |
318 | |
319 | confused: |
320 | if (bio) |
321 | bio = mpage_bio_submit(READ, bio); |
322 | if (!PageUptodate(page)) |
323 | block_read_full_page(page, get_block); |
324 | else |
325 | unlock_page(page); |
326 | goto out; |
327 | } |
328 | |
329 | /** |
330 | * mpage_readpages - populate an address space with some pages & start reads against them |
331 | * @mapping: the address_space |
332 | * @pages: The address of a list_head which contains the target pages. These |
333 | * pages have their ->index populated and are otherwise uninitialised. |
334 | * The page at @pages->prev has the lowest file offset, and reads should be |
335 | * issued in @pages->prev to @pages->next order. |
336 | * @nr_pages: The number of pages at *@pages |
337 | * @get_block: The filesystem's block mapper function. |
338 | * |
339 | * This function walks the pages and the blocks within each page, building and |
340 | * emitting large BIOs. |
341 | * |
342 | * If anything unusual happens, such as: |
343 | * |
344 | * - encountering a page which has buffers |
345 | * - encountering a page which has a non-hole after a hole |
346 | * - encountering a page with non-contiguous blocks |
347 | * |
348 | * then this code just gives up and calls the buffer_head-based read function. |
349 | * It does handle a page which has holes at the end - that is a common case: |
350 | * the end-of-file on blocksize < PAGE_CACHE_SIZE setups. |
351 | * |
352 | * BH_Boundary explanation: |
353 | * |
354 | * There is a problem. The mpage read code assembles several pages, gets all |
355 | * their disk mappings, and then submits them all. That's fine, but obtaining |
356 | * the disk mappings may require I/O. Reads of indirect blocks, for example. |
357 | * |
358 | * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be |
359 | * submitted in the following order: |
360 | * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16 |
361 | * |
362 | * because the indirect block has to be read to get the mappings of blocks |
363 | * 13,14,15,16. Obviously, this impacts performance. |
364 | * |
365 | * So what we do it to allow the filesystem's get_block() function to set |
366 | * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block |
367 | * after this one will require I/O against a block which is probably close to |
368 | * this one. So you should push what I/O you have currently accumulated. |
369 | * |
370 | * This all causes the disk requests to be issued in the correct order. |
371 | */ |
372 | int |
373 | mpage_readpages(struct address_space *mapping, struct list_head *pages, |
374 | unsigned nr_pages, get_block_t get_block) |
375 | { |
376 | struct bio *bio = NULL; |
377 | unsigned page_idx; |
378 | sector_t last_block_in_bio = 0; |
379 | struct buffer_head map_bh; |
380 | unsigned long first_logical_block = 0; |
381 | |
382 | map_bh.b_state = 0; |
383 | map_bh.b_size = 0; |
384 | for (page_idx = 0; page_idx < nr_pages; page_idx++) { |
385 | struct page *page = list_entry(pages->prev, struct page, lru); |
386 | |
387 | prefetchw(&page->flags); |
388 | list_del(&page->lru); |
389 | if (!add_to_page_cache_lru(page, mapping, |
390 | page->index, GFP_KERNEL)) { |
391 | bio = do_mpage_readpage(bio, page, |
392 | nr_pages - page_idx, |
393 | &last_block_in_bio, &map_bh, |
394 | &first_logical_block, |
395 | get_block); |
396 | } |
397 | page_cache_release(page); |
398 | } |
399 | BUG_ON(!list_empty(pages)); |
400 | if (bio) |
401 | mpage_bio_submit(READ, bio); |
402 | return 0; |
403 | } |
404 | EXPORT_SYMBOL(mpage_readpages); |
405 | |
406 | /* |
407 | * This isn't called much at all |
408 | */ |
409 | int mpage_readpage(struct page *page, get_block_t get_block) |
410 | { |
411 | struct bio *bio = NULL; |
412 | sector_t last_block_in_bio = 0; |
413 | struct buffer_head map_bh; |
414 | unsigned long first_logical_block = 0; |
415 | |
416 | map_bh.b_state = 0; |
417 | map_bh.b_size = 0; |
418 | bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio, |
419 | &map_bh, &first_logical_block, get_block); |
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 | */ |
442 | |
443 | struct mpage_data { |
444 | struct bio *bio; |
445 | sector_t last_block_in_bio; |
446 | get_block_t *get_block; |
447 | unsigned use_writepage; |
448 | }; |
449 | |
450 | static int __mpage_writepage(struct page *page, struct writeback_control *wbc, |
451 | void *data) |
452 | { |
453 | struct mpage_data *mpd = data; |
454 | struct bio *bio = mpd->bio; |
455 | struct address_space *mapping = page->mapping; |
456 | struct inode *inode = page->mapping->host; |
457 | const unsigned blkbits = inode->i_blkbits; |
458 | unsigned long end_index; |
459 | const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits; |
460 | sector_t last_block; |
461 | sector_t block_in_file; |
462 | sector_t blocks[MAX_BUF_PER_PAGE]; |
463 | unsigned page_block; |
464 | unsigned first_unmapped = blocks_per_page; |
465 | struct block_device *bdev = NULL; |
466 | int boundary = 0; |
467 | sector_t boundary_block = 0; |
468 | struct block_device *boundary_bdev = NULL; |
469 | int length; |
470 | struct buffer_head map_bh; |
471 | loff_t i_size = i_size_read(inode); |
472 | int ret = 0; |
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 |
519 | * using mpage_readpages then this can rarely happen. |
520 | */ |
521 | goto confused; |
522 | } |
523 | |
524 | /* |
525 | * The page has no buffers: map it to disk |
526 | */ |
527 | BUG_ON(!PageUptodate(page)); |
528 | block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits); |
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; |
534 | map_bh.b_size = 1 << blkbits; |
535 | if (mpd->get_block(inode, block_in_file, &map_bh, 1)) |
536 | goto confused; |
537 | if (buffer_new(&map_bh)) |
538 | unmap_underlying_metadata(map_bh.b_bdev, |
539 | map_bh.b_blocknr); |
540 | if (buffer_boundary(&map_bh)) { |
541 | boundary_block = map_bh.b_blocknr; |
542 | boundary_bdev = map_bh.b_bdev; |
543 | } |
544 | if (page_block) { |
545 | if (map_bh.b_blocknr != blocks[page_block-1] + 1) |
546 | goto confused; |
547 | } |
548 | blocks[page_block++] = map_bh.b_blocknr; |
549 | boundary = buffer_boundary(&map_bh); |
550 | bdev = map_bh.b_bdev; |
551 | if (block_in_file == last_block) |
552 | break; |
553 | block_in_file++; |
554 | } |
555 | BUG_ON(page_block == 0); |
556 | |
557 | first_unmapped = page_block; |
558 | |
559 | page_is_mapped: |
560 | end_index = i_size >> PAGE_CACHE_SHIFT; |
561 | if (page->index >= end_index) { |
562 | /* |
563 | * The page straddles i_size. It must be zeroed out on each |
564 | * and every writepage invokation because it may be mmapped. |
565 | * "A file is mapped in multiples of the page size. For a file |
566 | * that is not a multiple of the page size, the remaining memory |
567 | * is zeroed when mapped, and writes to that region are not |
568 | * written out to the file." |
569 | */ |
570 | unsigned offset = i_size & (PAGE_CACHE_SIZE - 1); |
571 | |
572 | if (page->index > end_index || !offset) |
573 | goto confused; |
574 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); |
575 | } |
576 | |
577 | /* |
578 | * This page will go to BIO. Do we need to send this BIO off first? |
579 | */ |
580 | if (bio && mpd->last_block_in_bio != blocks[0] - 1) |
581 | bio = mpage_bio_submit(WRITE, bio); |
582 | |
583 | alloc_new: |
584 | if (bio == NULL) { |
585 | bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9), |
586 | bio_get_nr_vecs(bdev), GFP_NOFS|__GFP_HIGH); |
587 | if (bio == NULL) |
588 | goto confused; |
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 | length = first_unmapped << blkbits; |
597 | if (bio_add_page(bio, page, length, 0) < length) { |
598 | bio = mpage_bio_submit(WRITE, bio); |
599 | goto alloc_new; |
600 | } |
601 | |
602 | /* |
603 | * OK, we have our BIO, so we can now mark the buffers clean. Make |
604 | * sure to only clean buffers which we know we'll be writing. |
605 | */ |
606 | if (page_has_buffers(page)) { |
607 | struct buffer_head *head = page_buffers(page); |
608 | struct buffer_head *bh = head; |
609 | unsigned buffer_counter = 0; |
610 | |
611 | do { |
612 | if (buffer_counter++ == first_unmapped) |
613 | break; |
614 | clear_buffer_dirty(bh); |
615 | bh = bh->b_this_page; |
616 | } while (bh != head); |
617 | |
618 | /* |
619 | * we cannot drop the bh if the page is not uptodate |
620 | * or a concurrent readpage would fail to serialize with the bh |
621 | * and it would read from disk before we reach the platter. |
622 | */ |
623 | if (buffer_heads_over_limit && PageUptodate(page)) |
624 | try_to_free_buffers(page); |
625 | } |
626 | |
627 | BUG_ON(PageWriteback(page)); |
628 | set_page_writeback(page); |
629 | unlock_page(page); |
630 | if (boundary || (first_unmapped != blocks_per_page)) { |
631 | bio = mpage_bio_submit(WRITE, bio); |
632 | if (boundary_block) { |
633 | write_boundary_block(boundary_bdev, |
634 | boundary_block, 1 << blkbits); |
635 | } |
636 | } else { |
637 | mpd->last_block_in_bio = blocks[blocks_per_page - 1]; |
638 | } |
639 | goto out; |
640 | |
641 | confused: |
642 | if (bio) |
643 | bio = mpage_bio_submit(WRITE, bio); |
644 | |
645 | if (mpd->use_writepage) { |
646 | ret = mapping->a_ops->writepage(page, wbc); |
647 | } else { |
648 | ret = -EAGAIN; |
649 | goto out; |
650 | } |
651 | /* |
652 | * The caller has a ref on the inode, so *mapping is stable |
653 | */ |
654 | mapping_set_error(mapping, ret); |
655 | out: |
656 | mpd->bio = bio; |
657 | return ret; |
658 | } |
659 | |
660 | /** |
661 | * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them |
662 | * @mapping: address space structure to write |
663 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write |
664 | * @get_block: the filesystem's block mapper function. |
665 | * If this is NULL then use a_ops->writepage. Otherwise, go |
666 | * direct-to-BIO. |
667 | * |
668 | * This is a library function, which implements the writepages() |
669 | * address_space_operation. |
670 | * |
671 | * If a page is already under I/O, generic_writepages() skips it, even |
672 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, |
673 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() |
674 | * and msync() need to guarantee that all the data which was dirty at the time |
675 | * the call was made get new I/O started against them. If wbc->sync_mode is |
676 | * WB_SYNC_ALL then we were called for data integrity and we must wait for |
677 | * existing IO to complete. |
678 | */ |
679 | int |
680 | mpage_writepages(struct address_space *mapping, |
681 | struct writeback_control *wbc, get_block_t get_block) |
682 | { |
683 | int ret; |
684 | |
685 | if (!get_block) |
686 | ret = generic_writepages(mapping, wbc); |
687 | else { |
688 | struct mpage_data mpd = { |
689 | .bio = NULL, |
690 | .last_block_in_bio = 0, |
691 | .get_block = get_block, |
692 | .use_writepage = 1, |
693 | }; |
694 | |
695 | ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd); |
696 | if (mpd.bio) |
697 | mpage_bio_submit(WRITE, mpd.bio); |
698 | } |
699 | return ret; |
700 | } |
701 | EXPORT_SYMBOL(mpage_writepages); |
702 | |
703 | int mpage_writepage(struct page *page, get_block_t get_block, |
704 | struct writeback_control *wbc) |
705 | { |
706 | struct mpage_data mpd = { |
707 | .bio = NULL, |
708 | .last_block_in_bio = 0, |
709 | .get_block = get_block, |
710 | .use_writepage = 0, |
711 | }; |
712 | int ret = __mpage_writepage(page, wbc, &mpd); |
713 | if (mpd.bio) |
714 | mpage_bio_submit(WRITE, mpd.bio); |
715 | return ret; |
716 | } |
717 | EXPORT_SYMBOL(mpage_writepage); |
718 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9