blob: b1c3e58905086eb920cafd4df806b7d70dd6ef36 [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 *
9 * 15May2002 akpm@zip.com.au
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 */
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
Adrian Bunk75c96f82005-05-05 16:16:09 -070085static struct 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}
93
94static struct bio *
95mpage_alloc(struct block_device *bdev,
96 sector_t first_sector, int nr_vecs,
Al Virodd0fc662005-10-07 07:46:04 +010097 gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098{
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 */
125static void
126map_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
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800158/*
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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167static struct bio *
168do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800169 sector_t *last_block_in_bio, struct buffer_head *map_bh,
170 unsigned long *first_logical_block, get_block_t get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171{
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;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800178 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 sector_t blocks[MAX_BUF_PER_PAGE];
180 unsigned page_block;
181 unsigned first_hole = blocks_per_page;
182 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 int length;
184 int fully_mapped = 1;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800185 unsigned nblocks;
186 unsigned relative_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
188 if (page_has_buffers(page))
189 goto confused;
190
Andrew Morton54b21a72006-01-08 01:03:05 -0800191 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800192 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800198 /*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800231 map_bh->b_size = (last_block-block_in_file) << blkbits;
232 if (get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800234 *first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 }
236
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800237 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 fully_mapped = 0;
239 if (first_hole == blocks_per_page)
240 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800241 page_block++;
242 block_in_file++;
243 clear_buffer_mapped(map_bh);
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) {
Nate Diller01f27052007-05-09 02:35:07 -0700279 zero_user_page(page, first_hole << blkbits,
280 PAGE_CACHE_SIZE - (first_hole << blkbits),
281 KM_USER0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 if (first_hole == 0) {
283 SetPageUptodate(page);
284 unlock_page(page);
285 goto out;
286 }
287 } else if (fully_mapped) {
288 SetPageMappedToDisk(page);
289 }
290
291 /*
292 * This page will go to BIO. Do we need to send this BIO off first?
293 */
294 if (bio && (*last_block_in_bio != blocks[0] - 1))
295 bio = mpage_bio_submit(READ, bio);
296
297alloc_new:
298 if (bio == NULL) {
299 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
300 min_t(int, nr_pages, bio_get_nr_vecs(bdev)),
301 GFP_KERNEL);
302 if (bio == NULL)
303 goto confused;
304 }
305
306 length = first_hole << blkbits;
307 if (bio_add_page(bio, page, length, 0) < length) {
308 bio = mpage_bio_submit(READ, bio);
309 goto alloc_new;
310 }
311
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800312 if (buffer_boundary(map_bh) || (first_hole != blocks_per_page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 bio = mpage_bio_submit(READ, bio);
314 else
315 *last_block_in_bio = blocks[blocks_per_page - 1];
316out:
317 return bio;
318
319confused:
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
Martin Waitz67be2dd2005-05-01 08:59:26 -0700329/**
330 * mpage_readpages - populate an address space with some pages, and
331 * start reads against them.
332 *
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.
336 *
337 * The page at @pages->prev has the lowest file offset, and reads should be
338 * issued in @pages->prev to @pages->next order.
339 *
340 * @nr_pages: The number of pages at *@pages
341 * @get_block: The filesystem's block mapper function.
342 *
343 * This function walks the pages and the blocks within each page, building and
344 * emitting large BIOs.
345 *
346 * If anything unusual happens, such as:
347 *
348 * - encountering a page which has buffers
349 * - encountering a page which has a non-hole after a hole
350 * - encountering a page with non-contiguous blocks
351 *
352 * then this code just gives up and calls the buffer_head-based read function.
353 * It does handle a page which has holes at the end - that is a common case:
354 * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
355 *
356 * BH_Boundary explanation:
357 *
358 * There is a problem. The mpage read code assembles several pages, gets all
359 * their disk mappings, and then submits them all. That's fine, but obtaining
360 * the disk mappings may require I/O. Reads of indirect blocks, for example.
361 *
362 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
363 * submitted in the following order:
364 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
365 * because the indirect block has to be read to get the mappings of blocks
366 * 13,14,15,16. Obviously, this impacts performance.
367 *
368 * So what we do it to allow the filesystem's get_block() function to set
369 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
370 * after this one will require I/O against a block which is probably close to
371 * this one. So you should push what I/O you have currently accumulated.
372 *
373 * This all causes the disk requests to be issued in the correct order.
374 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375int
376mpage_readpages(struct address_space *mapping, struct list_head *pages,
377 unsigned nr_pages, get_block_t get_block)
378{
379 struct bio *bio = NULL;
380 unsigned page_idx;
381 sector_t last_block_in_bio = 0;
382 struct pagevec lru_pvec;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800383 struct buffer_head map_bh;
384 unsigned long first_logical_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800386 clear_buffer_mapped(&map_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 pagevec_init(&lru_pvec, 0);
388 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
389 struct page *page = list_entry(pages->prev, struct page, lru);
390
391 prefetchw(&page->flags);
392 list_del(&page->lru);
393 if (!add_to_page_cache(page, mapping,
394 page->index, GFP_KERNEL)) {
395 bio = do_mpage_readpage(bio, page,
396 nr_pages - page_idx,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800397 &last_block_in_bio, &map_bh,
398 &first_logical_block,
399 get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 if (!pagevec_add(&lru_pvec, page))
401 __pagevec_lru_add(&lru_pvec);
402 } else {
403 page_cache_release(page);
404 }
405 }
406 pagevec_lru_add(&lru_pvec);
407 BUG_ON(!list_empty(pages));
408 if (bio)
409 mpage_bio_submit(READ, bio);
410 return 0;
411}
412EXPORT_SYMBOL(mpage_readpages);
413
414/*
415 * This isn't called much at all
416 */
417int mpage_readpage(struct page *page, get_block_t get_block)
418{
419 struct bio *bio = NULL;
420 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800421 struct buffer_head map_bh;
422 unsigned long first_logical_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800424 clear_buffer_mapped(&map_bh);
425 bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
426 &map_bh, &first_logical_block, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (bio)
428 mpage_bio_submit(READ, bio);
429 return 0;
430}
431EXPORT_SYMBOL(mpage_readpage);
432
433/*
434 * Writing is not so simple.
435 *
436 * If the page has buffers then they will be used for obtaining the disk
437 * mapping. We only support pages which are fully mapped-and-dirty, with a
438 * special case for pages which are unmapped at the end: end-of-file.
439 *
440 * If the page has no buffers (preferred) then the page is mapped here.
441 *
442 * If all blocks are found to be contiguous then the page can go into the
443 * BIO. Otherwise fall back to the mapping's writepage().
444 *
445 * FIXME: This code wants an estimate of how many pages are still to be
446 * written, so it can intelligently allocate a suitably-sized BIO. For now,
447 * just allocate full-size (16-page) BIOs.
448 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700449struct mpage_data {
450 struct bio *bio;
451 sector_t last_block_in_bio;
452 get_block_t *get_block;
453 unsigned use_writepage;
454};
455
456static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
457 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700459 struct mpage_data *mpd = data;
460 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 struct address_space *mapping = page->mapping;
462 struct inode *inode = page->mapping->host;
463 const unsigned blkbits = inode->i_blkbits;
464 unsigned long end_index;
465 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
466 sector_t last_block;
467 sector_t block_in_file;
468 sector_t blocks[MAX_BUF_PER_PAGE];
469 unsigned page_block;
470 unsigned first_unmapped = blocks_per_page;
471 struct block_device *bdev = NULL;
472 int boundary = 0;
473 sector_t boundary_block = 0;
474 struct block_device *boundary_bdev = NULL;
475 int length;
476 struct buffer_head map_bh;
477 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700478 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
480 if (page_has_buffers(page)) {
481 struct buffer_head *head = page_buffers(page);
482 struct buffer_head *bh = head;
483
484 /* If they're all mapped and dirty, do it */
485 page_block = 0;
486 do {
487 BUG_ON(buffer_locked(bh));
488 if (!buffer_mapped(bh)) {
489 /*
490 * unmapped dirty buffers are created by
491 * __set_page_dirty_buffers -> mmapped data
492 */
493 if (buffer_dirty(bh))
494 goto confused;
495 if (first_unmapped == blocks_per_page)
496 first_unmapped = page_block;
497 continue;
498 }
499
500 if (first_unmapped != blocks_per_page)
501 goto confused; /* hole -> non-hole */
502
503 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
504 goto confused;
505 if (page_block) {
506 if (bh->b_blocknr != blocks[page_block-1] + 1)
507 goto confused;
508 }
509 blocks[page_block++] = bh->b_blocknr;
510 boundary = buffer_boundary(bh);
511 if (boundary) {
512 boundary_block = bh->b_blocknr;
513 boundary_bdev = bh->b_bdev;
514 }
515 bdev = bh->b_bdev;
516 } while ((bh = bh->b_this_page) != head);
517
518 if (first_unmapped)
519 goto page_is_mapped;
520
521 /*
522 * Page has buffers, but they are all unmapped. The page was
523 * created by pagein or read over a hole which was handled by
524 * block_read_full_page(). If this address_space is also
525 * using mpage_readpages then this can rarely happen.
526 */
527 goto confused;
528 }
529
530 /*
531 * The page has no buffers: map it to disk
532 */
533 BUG_ON(!PageUptodate(page));
Andrew Morton54b21a72006-01-08 01:03:05 -0800534 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 last_block = (i_size - 1) >> blkbits;
536 map_bh.b_page = page;
537 for (page_block = 0; page_block < blocks_per_page; ) {
538
539 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800540 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700541 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 goto confused;
543 if (buffer_new(&map_bh))
544 unmap_underlying_metadata(map_bh.b_bdev,
545 map_bh.b_blocknr);
546 if (buffer_boundary(&map_bh)) {
547 boundary_block = map_bh.b_blocknr;
548 boundary_bdev = map_bh.b_bdev;
549 }
550 if (page_block) {
551 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
552 goto confused;
553 }
554 blocks[page_block++] = map_bh.b_blocknr;
555 boundary = buffer_boundary(&map_bh);
556 bdev = map_bh.b_bdev;
557 if (block_in_file == last_block)
558 break;
559 block_in_file++;
560 }
561 BUG_ON(page_block == 0);
562
563 first_unmapped = page_block;
564
565page_is_mapped:
566 end_index = i_size >> PAGE_CACHE_SHIFT;
567 if (page->index >= end_index) {
568 /*
569 * The page straddles i_size. It must be zeroed out on each
570 * and every writepage invokation because it may be mmapped.
571 * "A file is mapped in multiples of the page size. For a file
572 * that is not a multiple of the page size, the remaining memory
573 * is zeroed when mapped, and writes to that region are not
574 * written out to the file."
575 */
576 unsigned offset = i_size & (PAGE_CACHE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 if (page->index > end_index || !offset)
579 goto confused;
Nate Diller01f27052007-05-09 02:35:07 -0700580 zero_user_page(page, offset, PAGE_CACHE_SIZE - offset,
581 KM_USER0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 }
583
584 /*
585 * This page will go to BIO. Do we need to send this BIO off first?
586 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700587 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 bio = mpage_bio_submit(WRITE, bio);
589
590alloc_new:
591 if (bio == NULL) {
592 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
593 bio_get_nr_vecs(bdev), GFP_NOFS|__GFP_HIGH);
594 if (bio == NULL)
595 goto confused;
596 }
597
598 /*
599 * Must try to add the page before marking the buffer clean or
600 * the confused fail path above (OOM) will be very confused when
601 * it finds all bh marked clean (i.e. it will not write anything)
602 */
603 length = first_unmapped << blkbits;
604 if (bio_add_page(bio, page, length, 0) < length) {
605 bio = mpage_bio_submit(WRITE, bio);
606 goto alloc_new;
607 }
608
609 /*
610 * OK, we have our BIO, so we can now mark the buffers clean. Make
611 * sure to only clean buffers which we know we'll be writing.
612 */
613 if (page_has_buffers(page)) {
614 struct buffer_head *head = page_buffers(page);
615 struct buffer_head *bh = head;
616 unsigned buffer_counter = 0;
617
618 do {
619 if (buffer_counter++ == first_unmapped)
620 break;
621 clear_buffer_dirty(bh);
622 bh = bh->b_this_page;
623 } while (bh != head);
624
625 /*
626 * we cannot drop the bh if the page is not uptodate
627 * or a concurrent readpage would fail to serialize with the bh
628 * and it would read from disk before we reach the platter.
629 */
630 if (buffer_heads_over_limit && PageUptodate(page))
631 try_to_free_buffers(page);
632 }
633
634 BUG_ON(PageWriteback(page));
635 set_page_writeback(page);
636 unlock_page(page);
637 if (boundary || (first_unmapped != blocks_per_page)) {
638 bio = mpage_bio_submit(WRITE, bio);
639 if (boundary_block) {
640 write_boundary_block(boundary_bdev,
641 boundary_block, 1 << blkbits);
642 }
643 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700644 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 }
646 goto out;
647
648confused:
649 if (bio)
650 bio = mpage_bio_submit(WRITE, bio);
651
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700652 if (mpd->use_writepage) {
653 ret = mapping->a_ops->writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700655 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 goto out;
657 }
658 /*
659 * The caller has a ref on the inode, so *mapping is stable
660 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700661 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700663 mpd->bio = bio;
664 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
667/**
668 * mpage_writepages - walk the list of dirty pages of the given
669 * address space and writepage() all of them.
670 *
671 * @mapping: address space structure to write
672 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
673 * @get_block: the filesystem's block mapper function.
674 * If this is NULL then use a_ops->writepage. Otherwise, go
675 * direct-to-BIO.
676 *
677 * This is a library function, which implements the writepages()
678 * address_space_operation.
679 *
680 * If a page is already under I/O, generic_writepages() skips it, even
681 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
682 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
683 * and msync() need to guarantee that all the data which was dirty at the time
684 * the call was made get new I/O started against them. If wbc->sync_mode is
685 * WB_SYNC_ALL then we were called for data integrity and we must wait for
686 * existing IO to complete.
687 */
688int
689mpage_writepages(struct address_space *mapping,
690 struct writeback_control *wbc, get_block_t get_block)
691{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700692 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700694 if (!get_block)
695 ret = generic_writepages(mapping, wbc);
696 else {
697 struct mpage_data mpd = {
698 .bio = NULL,
699 .last_block_in_bio = 0,
700 .get_block = get_block,
701 .use_writepage = 1,
702 };
703
704 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
705 if (mpd.bio)
706 mpage_bio_submit(WRITE, mpd.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 return ret;
709}
710EXPORT_SYMBOL(mpage_writepages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
712int mpage_writepage(struct page *page, get_block_t get_block,
713 struct writeback_control *wbc)
714{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700715 struct mpage_data mpd = {
716 .bio = NULL,
717 .last_block_in_bio = 0,
718 .get_block = get_block,
719 .use_writepage = 0,
720 };
721 int ret = __mpage_writepage(page, wbc, &mpd);
722 if (mpd.bio)
723 mpage_bio_submit(WRITE, mpd.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 return ret;
725}
726EXPORT_SYMBOL(mpage_writepage);