blob: 16c3ef37eae348ed97ad90229d427449699315a3 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Francois Camie1f8e872008-10-15 22:01:59 -07009 * 15May2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 * 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 */
NeilBrown6712ecf2007-09-27 12:47:43 +020042static void mpage_end_io_read(struct bio *bio, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -070043{
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062}
63
NeilBrown6712ecf2007-09-27 12:47:43 +020064static void mpage_end_io_write(struct bio *bio, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -070065{
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 do {
70 struct page *page = bvec->bv_page;
71
72 if (--bvec >= bio->bi_io_vec)
73 prefetchw(&bvec->bv_page->flags);
74
Qu Fuping854715b2005-06-04 15:43:29 -070075 if (!uptodate){
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 SetPageError(page);
Qu Fuping854715b2005-06-04 15:43:29 -070077 if (page->mapping)
78 set_bit(AS_EIO, &page->mapping->flags);
79 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 end_page_writeback(page);
81 } while (bvec >= bio->bi_io_vec);
82 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083}
84
Alex Tomas29a814d2008-07-11 19:27:31 -040085struct bio *mpage_bio_submit(int rw, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086{
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}
Alex Tomas29a814d2008-07-11 19:27:31 -040093EXPORT_SYMBOL(mpage_bio_submit);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
95static struct bio *
96mpage_alloc(struct block_device *bdev,
97 sector_t first_sector, int nr_vecs,
Al Virodd0fc662005-10-07 07:46:04 +010098 gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099{
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 */
126static void
127map_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
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800159/*
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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168static struct bio *
169do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800170 sector_t *last_block_in_bio, struct buffer_head *map_bh,
171 unsigned long *first_logical_block, get_block_t get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172{
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;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800179 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 sector_t blocks[MAX_BUF_PER_PAGE];
181 unsigned page_block;
182 unsigned first_hole = blocks_per_page;
183 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 int length;
185 int fully_mapped = 1;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800186 unsigned nblocks;
187 unsigned relative_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 if (page_has_buffers(page))
190 goto confused;
191
Andrew Morton54b21a72006-01-08 01:03:05 -0800192 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800193 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800199 /*
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;
230
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800232 map_bh->b_size = (last_block-block_in_file) << blkbits;
233 if (get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800235 *first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 }
237
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800238 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 fully_mapped = 0;
240 if (first_hole == blocks_per_page)
241 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800242 page_block++;
243 block_in_file++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 continue;
245 }
246
247 /* some filesystems will copy data into the page during
248 * the get_block call, in which case we don't want to
249 * read it again. map_buffer_to_page copies the data
250 * we just collected from get_block into the page's buffers
251 * so readpage doesn't have to repeat the get_block call
252 */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800253 if (buffer_uptodate(map_bh)) {
254 map_buffer_to_page(page, map_bh, page_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 goto confused;
256 }
257
258 if (first_hole != blocks_per_page)
259 goto confused; /* hole -> non-hole */
260
261 /* Contiguous blocks? */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800262 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800264 nblocks = map_bh->b_size >> blkbits;
265 for (relative_block = 0; ; relative_block++) {
266 if (relative_block == nblocks) {
267 clear_buffer_mapped(map_bh);
268 break;
269 } else if (page_block == blocks_per_page)
270 break;
271 blocks[page_block] = map_bh->b_blocknr+relative_block;
272 page_block++;
273 block_in_file++;
274 }
275 bdev = map_bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 }
277
278 if (first_hole != blocks_per_page) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800279 zero_user_segment(page, first_hole << blkbits, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 if (first_hole == 0) {
281 SetPageUptodate(page);
282 unlock_page(page);
283 goto out;
284 }
285 } else if (fully_mapped) {
286 SetPageMappedToDisk(page);
287 }
288
289 /*
290 * This page will go to BIO. Do we need to send this BIO off first?
291 */
292 if (bio && (*last_block_in_bio != blocks[0] - 1))
293 bio = mpage_bio_submit(READ, bio);
294
295alloc_new:
296 if (bio == NULL) {
297 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
298 min_t(int, nr_pages, bio_get_nr_vecs(bdev)),
299 GFP_KERNEL);
300 if (bio == NULL)
301 goto confused;
302 }
303
304 length = first_hole << blkbits;
305 if (bio_add_page(bio, page, length, 0) < length) {
306 bio = mpage_bio_submit(READ, bio);
307 goto alloc_new;
308 }
309
Miquel van Smoorenburg38c8e612009-01-06 14:39:02 -0800310 relative_block = block_in_file - *first_logical_block;
311 nblocks = map_bh->b_size >> blkbits;
312 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
313 (first_hole != blocks_per_page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 bio = mpage_bio_submit(READ, bio);
315 else
316 *last_block_in_bio = blocks[blocks_per_page - 1];
317out:
318 return bio;
319
320confused:
321 if (bio)
322 bio = mpage_bio_submit(READ, bio);
323 if (!PageUptodate(page))
324 block_read_full_page(page, get_block);
325 else
326 unlock_page(page);
327 goto out;
328}
329
Martin Waitz67be2dd2005-05-01 08:59:26 -0700330/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800331 * mpage_readpages - populate an address space with some pages & start reads against them
Martin Waitz67be2dd2005-05-01 08:59:26 -0700332 * @mapping: the address_space
333 * @pages: The address of a list_head which contains the target pages. These
334 * pages have their ->index populated and are otherwise uninitialised.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700335 * The page at @pages->prev has the lowest file offset, and reads should be
336 * issued in @pages->prev to @pages->next order.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700337 * @nr_pages: The number of pages at *@pages
338 * @get_block: The filesystem's block mapper function.
339 *
340 * This function walks the pages and the blocks within each page, building and
341 * emitting large BIOs.
342 *
343 * If anything unusual happens, such as:
344 *
345 * - encountering a page which has buffers
346 * - encountering a page which has a non-hole after a hole
347 * - encountering a page with non-contiguous blocks
348 *
349 * then this code just gives up and calls the buffer_head-based read function.
350 * It does handle a page which has holes at the end - that is a common case:
351 * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
352 *
353 * BH_Boundary explanation:
354 *
355 * There is a problem. The mpage read code assembles several pages, gets all
356 * their disk mappings, and then submits them all. That's fine, but obtaining
357 * the disk mappings may require I/O. Reads of indirect blocks, for example.
358 *
359 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
360 * submitted in the following order:
361 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
Randy Dunlap78a4a502008-02-29 22:02:31 -0800362 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700363 * because the indirect block has to be read to get the mappings of blocks
364 * 13,14,15,16. Obviously, this impacts performance.
365 *
366 * So what we do it to allow the filesystem's get_block() function to set
367 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
368 * after this one will require I/O against a block which is probably close to
369 * this one. So you should push what I/O you have currently accumulated.
370 *
371 * This all causes the disk requests to be issued in the correct order.
372 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373int
374mpage_readpages(struct address_space *mapping, struct list_head *pages,
375 unsigned nr_pages, get_block_t get_block)
376{
377 struct bio *bio = NULL;
378 unsigned page_idx;
379 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800380 struct buffer_head map_bh;
381 unsigned long first_logical_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800383 clear_buffer_mapped(&map_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 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);
Nick Piggineb2be182007-10-16 01:24:57 -0700389 if (!add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 page->index, GFP_KERNEL)) {
391 bio = do_mpage_readpage(bio, page,
392 nr_pages - page_idx,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800393 &last_block_in_bio, &map_bh,
394 &first_logical_block,
395 get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 }
Nick Piggineb2be182007-10-16 01:24:57 -0700397 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 BUG_ON(!list_empty(pages));
400 if (bio)
401 mpage_bio_submit(READ, bio);
402 return 0;
403}
404EXPORT_SYMBOL(mpage_readpages);
405
406/*
407 * This isn't called much at all
408 */
409int mpage_readpage(struct page *page, get_block_t get_block)
410{
411 struct bio *bio = NULL;
412 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800413 struct buffer_head map_bh;
414 unsigned long first_logical_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800416 clear_buffer_mapped(&map_bh);
417 bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
418 &map_bh, &first_logical_block, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 if (bio)
420 mpage_bio_submit(READ, bio);
421 return 0;
422}
423EXPORT_SYMBOL(mpage_readpage);
424
425/*
426 * Writing is not so simple.
427 *
428 * If the page has buffers then they will be used for obtaining the disk
429 * mapping. We only support pages which are fully mapped-and-dirty, with a
430 * special case for pages which are unmapped at the end: end-of-file.
431 *
432 * If the page has no buffers (preferred) then the page is mapped here.
433 *
434 * If all blocks are found to be contiguous then the page can go into the
435 * BIO. Otherwise fall back to the mapping's writepage().
436 *
437 * FIXME: This code wants an estimate of how many pages are still to be
438 * written, so it can intelligently allocate a suitably-sized BIO. For now,
439 * just allocate full-size (16-page) BIOs.
440 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700441
Alex Tomas29a814d2008-07-11 19:27:31 -0400442int __mpage_writepage(struct page *page, struct writeback_control *wbc,
443 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700445 struct mpage_data *mpd = data;
446 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 struct address_space *mapping = page->mapping;
448 struct inode *inode = page->mapping->host;
449 const unsigned blkbits = inode->i_blkbits;
450 unsigned long end_index;
451 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
452 sector_t last_block;
453 sector_t block_in_file;
454 sector_t blocks[MAX_BUF_PER_PAGE];
455 unsigned page_block;
456 unsigned first_unmapped = blocks_per_page;
457 struct block_device *bdev = NULL;
458 int boundary = 0;
459 sector_t boundary_block = 0;
460 struct block_device *boundary_bdev = NULL;
461 int length;
462 struct buffer_head map_bh;
463 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700464 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
466 if (page_has_buffers(page)) {
467 struct buffer_head *head = page_buffers(page);
468 struct buffer_head *bh = head;
469
470 /* If they're all mapped and dirty, do it */
471 page_block = 0;
472 do {
473 BUG_ON(buffer_locked(bh));
474 if (!buffer_mapped(bh)) {
475 /*
476 * unmapped dirty buffers are created by
477 * __set_page_dirty_buffers -> mmapped data
478 */
479 if (buffer_dirty(bh))
480 goto confused;
481 if (first_unmapped == blocks_per_page)
482 first_unmapped = page_block;
483 continue;
484 }
485
486 if (first_unmapped != blocks_per_page)
487 goto confused; /* hole -> non-hole */
488
489 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
490 goto confused;
491 if (page_block) {
492 if (bh->b_blocknr != blocks[page_block-1] + 1)
493 goto confused;
494 }
495 blocks[page_block++] = bh->b_blocknr;
496 boundary = buffer_boundary(bh);
497 if (boundary) {
498 boundary_block = bh->b_blocknr;
499 boundary_bdev = bh->b_bdev;
500 }
501 bdev = bh->b_bdev;
502 } while ((bh = bh->b_this_page) != head);
503
504 if (first_unmapped)
505 goto page_is_mapped;
506
507 /*
508 * Page has buffers, but they are all unmapped. The page was
509 * created by pagein or read over a hole which was handled by
510 * block_read_full_page(). If this address_space is also
511 * using mpage_readpages then this can rarely happen.
512 */
513 goto confused;
514 }
515
516 /*
517 * The page has no buffers: map it to disk
518 */
519 BUG_ON(!PageUptodate(page));
Andrew Morton54b21a72006-01-08 01:03:05 -0800520 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 last_block = (i_size - 1) >> blkbits;
522 map_bh.b_page = page;
523 for (page_block = 0; page_block < blocks_per_page; ) {
524
525 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800526 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700527 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 goto confused;
529 if (buffer_new(&map_bh))
530 unmap_underlying_metadata(map_bh.b_bdev,
531 map_bh.b_blocknr);
532 if (buffer_boundary(&map_bh)) {
533 boundary_block = map_bh.b_blocknr;
534 boundary_bdev = map_bh.b_bdev;
535 }
536 if (page_block) {
537 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
538 goto confused;
539 }
540 blocks[page_block++] = map_bh.b_blocknr;
541 boundary = buffer_boundary(&map_bh);
542 bdev = map_bh.b_bdev;
543 if (block_in_file == last_block)
544 break;
545 block_in_file++;
546 }
547 BUG_ON(page_block == 0);
548
549 first_unmapped = page_block;
550
551page_is_mapped:
552 end_index = i_size >> PAGE_CACHE_SHIFT;
553 if (page->index >= end_index) {
554 /*
555 * The page straddles i_size. It must be zeroed out on each
556 * and every writepage invokation because it may be mmapped.
557 * "A file is mapped in multiples of the page size. For a file
558 * that is not a multiple of the page size, the remaining memory
559 * is zeroed when mapped, and writes to that region are not
560 * written out to the file."
561 */
562 unsigned offset = i_size & (PAGE_CACHE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564 if (page->index > end_index || !offset)
565 goto confused;
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800566 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 }
568
569 /*
570 * This page will go to BIO. Do we need to send this BIO off first?
571 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700572 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 bio = mpage_bio_submit(WRITE, bio);
574
575alloc_new:
576 if (bio == NULL) {
577 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
578 bio_get_nr_vecs(bdev), GFP_NOFS|__GFP_HIGH);
579 if (bio == NULL)
580 goto confused;
581 }
582
583 /*
584 * Must try to add the page before marking the buffer clean or
585 * the confused fail path above (OOM) will be very confused when
586 * it finds all bh marked clean (i.e. it will not write anything)
587 */
588 length = first_unmapped << blkbits;
589 if (bio_add_page(bio, page, length, 0) < length) {
590 bio = mpage_bio_submit(WRITE, bio);
591 goto alloc_new;
592 }
593
594 /*
595 * OK, we have our BIO, so we can now mark the buffers clean. Make
596 * sure to only clean buffers which we know we'll be writing.
597 */
598 if (page_has_buffers(page)) {
599 struct buffer_head *head = page_buffers(page);
600 struct buffer_head *bh = head;
601 unsigned buffer_counter = 0;
602
603 do {
604 if (buffer_counter++ == first_unmapped)
605 break;
606 clear_buffer_dirty(bh);
607 bh = bh->b_this_page;
608 } while (bh != head);
609
610 /*
611 * we cannot drop the bh if the page is not uptodate
612 * or a concurrent readpage would fail to serialize with the bh
613 * and it would read from disk before we reach the platter.
614 */
615 if (buffer_heads_over_limit && PageUptodate(page))
616 try_to_free_buffers(page);
617 }
618
619 BUG_ON(PageWriteback(page));
620 set_page_writeback(page);
621 unlock_page(page);
622 if (boundary || (first_unmapped != blocks_per_page)) {
623 bio = mpage_bio_submit(WRITE, bio);
624 if (boundary_block) {
625 write_boundary_block(boundary_bdev,
626 boundary_block, 1 << blkbits);
627 }
628 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700629 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 }
631 goto out;
632
633confused:
634 if (bio)
635 bio = mpage_bio_submit(WRITE, bio);
636
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700637 if (mpd->use_writepage) {
638 ret = mapping->a_ops->writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700640 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 goto out;
642 }
643 /*
644 * The caller has a ref on the inode, so *mapping is stable
645 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700646 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700648 mpd->bio = bio;
649 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650}
Alex Tomas29a814d2008-07-11 19:27:31 -0400651EXPORT_SYMBOL(__mpage_writepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
653/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800654 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 * @mapping: address space structure to write
656 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
657 * @get_block: the filesystem's block mapper function.
658 * If this is NULL then use a_ops->writepage. Otherwise, go
659 * direct-to-BIO.
660 *
661 * This is a library function, which implements the writepages()
662 * address_space_operation.
663 *
664 * If a page is already under I/O, generic_writepages() skips it, even
665 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
666 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
667 * and msync() need to guarantee that all the data which was dirty at the time
668 * the call was made get new I/O started against them. If wbc->sync_mode is
669 * WB_SYNC_ALL then we were called for data integrity and we must wait for
670 * existing IO to complete.
671 */
672int
673mpage_writepages(struct address_space *mapping,
674 struct writeback_control *wbc, get_block_t get_block)
675{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700676 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700678 if (!get_block)
679 ret = generic_writepages(mapping, wbc);
680 else {
681 struct mpage_data mpd = {
682 .bio = NULL,
683 .last_block_in_bio = 0,
684 .get_block = get_block,
685 .use_writepage = 1,
686 };
687
688 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
689 if (mpd.bio)
690 mpage_bio_submit(WRITE, mpd.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 return ret;
693}
694EXPORT_SYMBOL(mpage_writepages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
696int mpage_writepage(struct page *page, get_block_t get_block,
697 struct writeback_control *wbc)
698{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700699 struct mpage_data mpd = {
700 .bio = NULL,
701 .last_block_in_bio = 0,
702 .get_block = get_block,
703 .use_writepage = 0,
704 };
705 int ret = __mpage_writepage(page, wbc, &mpd);
706 if (mpd.bio)
707 mpage_bio_submit(WRITE, mpd.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 return ret;
709}
710EXPORT_SYMBOL(mpage_writepage);