blob: 0afc809e46e09b53cb767e56795fba3139c4b181 [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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/bio.h>
21#include <linux/fs.h>
22#include <linux/buffer_head.h>
23#include <linux/blkdev.h>
24#include <linux/highmem.h>
25#include <linux/prefetch.h>
26#include <linux/mpage.h>
27#include <linux/writeback.h>
28#include <linux/backing-dev.h>
29#include <linux/pagevec.h>
30
31/*
32 * I/O completion handler for multipage BIOs.
33 *
34 * The mpage code never puts partial pages into a BIO (except for end-of-file).
35 * If a page does not map to a contiguous run of blocks then it simply falls
36 * back to block_read_full_page().
37 *
38 * Why is this? If a page's completion depends on a number of different BIOs
39 * which can complete in any order (or at the same time) then determining the
40 * status of that page is hard. See end_buffer_async_read() for the details.
41 * There is no point in duplicating all that complexity.
42 */
Hai Shanc32b0d42011-01-13 15:45:51 -080043static void mpage_end_io(struct bio *bio, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -070044{
45 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
46 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
47
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 do {
49 struct page *page = bvec->bv_page;
50
51 if (--bvec >= bio->bi_io_vec)
52 prefetchw(&bvec->bv_page->flags);
Hai Shanc32b0d42011-01-13 15:45:51 -080053 if (bio_data_dir(bio) == READ) {
54 if (uptodate) {
55 SetPageUptodate(page);
56 } else {
57 ClearPageUptodate(page);
58 SetPageError(page);
59 }
60 unlock_page(page);
61 } else { /* bio_data_dir(bio) == WRITE */
62 if (!uptodate) {
63 SetPageError(page);
64 if (page->mapping)
65 set_bit(AS_EIO, &page->mapping->flags);
66 }
67 end_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 } while (bvec >= bio->bi_io_vec);
70 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071}
72
Dmitri Vorobievced117c2009-03-31 00:41:20 +030073static struct bio *mpage_bio_submit(int rw, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
Hai Shanc32b0d42011-01-13 15:45:51 -080075 bio->bi_end_io = mpage_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 submit_bio(rw, bio);
77 return NULL;
78}
79
80static struct bio *
81mpage_alloc(struct block_device *bdev,
82 sector_t first_sector, int nr_vecs,
Al Virodd0fc662005-10-07 07:46:04 +010083 gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084{
85 struct bio *bio;
86
87 bio = bio_alloc(gfp_flags, nr_vecs);
88
89 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
90 while (!bio && (nr_vecs /= 2))
91 bio = bio_alloc(gfp_flags, nr_vecs);
92 }
93
94 if (bio) {
95 bio->bi_bdev = bdev;
96 bio->bi_sector = first_sector;
97 }
98 return bio;
99}
100
101/*
102 * support function for mpage_readpages. The fs supplied get_block might
103 * return an up to date buffer. This is used to map that buffer into
104 * the page, which allows readpage to avoid triggering a duplicate call
105 * to get_block.
106 *
107 * The idea is to avoid adding buffers to pages that don't already have
108 * them. So when the buffer is up to date and the page size == block size,
109 * this marks the page up to date instead of adding new buffers.
110 */
111static void
112map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
113{
114 struct inode *inode = page->mapping->host;
115 struct buffer_head *page_bh, *head;
116 int block = 0;
117
118 if (!page_has_buffers(page)) {
119 /*
120 * don't make any buffers if there is only one buffer on
121 * the page and the page just needs to be set up to date
122 */
123 if (inode->i_blkbits == PAGE_CACHE_SHIFT &&
124 buffer_uptodate(bh)) {
125 SetPageUptodate(page);
126 return;
127 }
128 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
129 }
130 head = page_buffers(page);
131 page_bh = head;
132 do {
133 if (block == page_block) {
134 page_bh->b_state = bh->b_state;
135 page_bh->b_bdev = bh->b_bdev;
136 page_bh->b_blocknr = bh->b_blocknr;
137 break;
138 }
139 page_bh = page_bh->b_this_page;
140 block++;
141 } while (page_bh != head);
142}
143
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800144/*
145 * This is the worker routine which does all the work of mapping the disk
146 * blocks and constructs largest possible bios, submits them for IO if the
147 * blocks are not contiguous on the disk.
148 *
149 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
150 * represent the validity of its disk mapping and to decide when to do the next
151 * get_block() call.
152 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153static struct bio *
154do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800155 sector_t *last_block_in_bio, struct buffer_head *map_bh,
156 unsigned long *first_logical_block, get_block_t get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157{
158 struct inode *inode = page->mapping->host;
159 const unsigned blkbits = inode->i_blkbits;
160 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
161 const unsigned blocksize = 1 << blkbits;
162 sector_t block_in_file;
163 sector_t last_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800164 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 sector_t blocks[MAX_BUF_PER_PAGE];
166 unsigned page_block;
167 unsigned first_hole = blocks_per_page;
168 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 int length;
170 int fully_mapped = 1;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800171 unsigned nblocks;
172 unsigned relative_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
174 if (page_has_buffers(page))
175 goto confused;
176
Andrew Morton54b21a72006-01-08 01:03:05 -0800177 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800178 last_block = block_in_file + nr_pages * blocks_per_page;
179 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
180 if (last_block > last_block_in_file)
181 last_block = last_block_in_file;
182 page_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800184 /*
185 * Map blocks using the result from the previous get_blocks call first.
186 */
187 nblocks = map_bh->b_size >> blkbits;
188 if (buffer_mapped(map_bh) && block_in_file > *first_logical_block &&
189 block_in_file < (*first_logical_block + nblocks)) {
190 unsigned map_offset = block_in_file - *first_logical_block;
191 unsigned last = nblocks - map_offset;
192
193 for (relative_block = 0; ; relative_block++) {
194 if (relative_block == last) {
195 clear_buffer_mapped(map_bh);
196 break;
197 }
198 if (page_block == blocks_per_page)
199 break;
200 blocks[page_block] = map_bh->b_blocknr + map_offset +
201 relative_block;
202 page_block++;
203 block_in_file++;
204 }
205 bdev = map_bh->b_bdev;
206 }
207
208 /*
209 * Then do more get_blocks calls until we are done with this page.
210 */
211 map_bh->b_page = page;
212 while (page_block < blocks_per_page) {
213 map_bh->b_state = 0;
214 map_bh->b_size = 0;
215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800217 map_bh->b_size = (last_block-block_in_file) << blkbits;
218 if (get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800220 *first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 }
222
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800223 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 fully_mapped = 0;
225 if (first_hole == blocks_per_page)
226 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800227 page_block++;
228 block_in_file++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 continue;
230 }
231
232 /* some filesystems will copy data into the page during
233 * the get_block call, in which case we don't want to
234 * read it again. map_buffer_to_page copies the data
235 * we just collected from get_block into the page's buffers
236 * so readpage doesn't have to repeat the get_block call
237 */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800238 if (buffer_uptodate(map_bh)) {
239 map_buffer_to_page(page, map_bh, page_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 goto confused;
241 }
242
243 if (first_hole != blocks_per_page)
244 goto confused; /* hole -> non-hole */
245
246 /* Contiguous blocks? */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800247 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800249 nblocks = map_bh->b_size >> blkbits;
250 for (relative_block = 0; ; relative_block++) {
251 if (relative_block == nblocks) {
252 clear_buffer_mapped(map_bh);
253 break;
254 } else if (page_block == blocks_per_page)
255 break;
256 blocks[page_block] = map_bh->b_blocknr+relative_block;
257 page_block++;
258 block_in_file++;
259 }
260 bdev = map_bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 }
262
263 if (first_hole != blocks_per_page) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800264 zero_user_segment(page, first_hole << blkbits, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 if (first_hole == 0) {
266 SetPageUptodate(page);
267 unlock_page(page);
268 goto out;
269 }
270 } else if (fully_mapped) {
271 SetPageMappedToDisk(page);
272 }
273
274 /*
275 * This page will go to BIO. Do we need to send this BIO off first?
276 */
277 if (bio && (*last_block_in_bio != blocks[0] - 1))
278 bio = mpage_bio_submit(READ, bio);
279
280alloc_new:
281 if (bio == NULL) {
282 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
283 min_t(int, nr_pages, bio_get_nr_vecs(bdev)),
284 GFP_KERNEL);
285 if (bio == NULL)
286 goto confused;
287 }
288
289 length = first_hole << blkbits;
290 if (bio_add_page(bio, page, length, 0) < length) {
291 bio = mpage_bio_submit(READ, bio);
292 goto alloc_new;
293 }
294
Miquel van Smoorenburg38c8e612009-01-06 14:39:02 -0800295 relative_block = block_in_file - *first_logical_block;
296 nblocks = map_bh->b_size >> blkbits;
297 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
298 (first_hole != blocks_per_page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 bio = mpage_bio_submit(READ, bio);
300 else
301 *last_block_in_bio = blocks[blocks_per_page - 1];
302out:
303 return bio;
304
305confused:
306 if (bio)
307 bio = mpage_bio_submit(READ, bio);
308 if (!PageUptodate(page))
309 block_read_full_page(page, get_block);
310 else
311 unlock_page(page);
312 goto out;
313}
314
Martin Waitz67be2dd2005-05-01 08:59:26 -0700315/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800316 * mpage_readpages - populate an address space with some pages & start reads against them
Martin Waitz67be2dd2005-05-01 08:59:26 -0700317 * @mapping: the address_space
318 * @pages: The address of a list_head which contains the target pages. These
319 * pages have their ->index populated and are otherwise uninitialised.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700320 * The page at @pages->prev has the lowest file offset, and reads should be
321 * issued in @pages->prev to @pages->next order.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700322 * @nr_pages: The number of pages at *@pages
323 * @get_block: The filesystem's block mapper function.
324 *
325 * This function walks the pages and the blocks within each page, building and
326 * emitting large BIOs.
327 *
328 * If anything unusual happens, such as:
329 *
330 * - encountering a page which has buffers
331 * - encountering a page which has a non-hole after a hole
332 * - encountering a page with non-contiguous blocks
333 *
334 * then this code just gives up and calls the buffer_head-based read function.
335 * It does handle a page which has holes at the end - that is a common case:
336 * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
337 *
338 * BH_Boundary explanation:
339 *
340 * There is a problem. The mpage read code assembles several pages, gets all
341 * their disk mappings, and then submits them all. That's fine, but obtaining
342 * the disk mappings may require I/O. Reads of indirect blocks, for example.
343 *
344 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
345 * submitted in the following order:
346 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
Randy Dunlap78a4a502008-02-29 22:02:31 -0800347 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700348 * because the indirect block has to be read to get the mappings of blocks
349 * 13,14,15,16. Obviously, this impacts performance.
350 *
351 * So what we do it to allow the filesystem's get_block() function to set
352 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
353 * after this one will require I/O against a block which is probably close to
354 * this one. So you should push what I/O you have currently accumulated.
355 *
356 * This all causes the disk requests to be issued in the correct order.
357 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358int
359mpage_readpages(struct address_space *mapping, struct list_head *pages,
360 unsigned nr_pages, get_block_t get_block)
361{
362 struct bio *bio = NULL;
363 unsigned page_idx;
364 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800365 struct buffer_head map_bh;
366 unsigned long first_logical_block = 0;
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200367 struct blk_plug plug;
368
369 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400371 map_bh.b_state = 0;
372 map_bh.b_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
374 struct page *page = list_entry(pages->prev, struct page, lru);
375
376 prefetchw(&page->flags);
377 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -0700378 if (!add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 page->index, GFP_KERNEL)) {
380 bio = do_mpage_readpage(bio, page,
381 nr_pages - page_idx,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800382 &last_block_in_bio, &map_bh,
383 &first_logical_block,
384 get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 }
Nick Piggineb2be182007-10-16 01:24:57 -0700386 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 BUG_ON(!list_empty(pages));
389 if (bio)
390 mpage_bio_submit(READ, bio);
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200391 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 return 0;
393}
394EXPORT_SYMBOL(mpage_readpages);
395
396/*
397 * This isn't called much at all
398 */
399int mpage_readpage(struct page *page, get_block_t get_block)
400{
401 struct bio *bio = NULL;
402 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800403 struct buffer_head map_bh;
404 unsigned long first_logical_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400406 map_bh.b_state = 0;
407 map_bh.b_size = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800408 bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
409 &map_bh, &first_logical_block, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 if (bio)
411 mpage_bio_submit(READ, bio);
412 return 0;
413}
414EXPORT_SYMBOL(mpage_readpage);
415
416/*
417 * Writing is not so simple.
418 *
419 * If the page has buffers then they will be used for obtaining the disk
420 * mapping. We only support pages which are fully mapped-and-dirty, with a
421 * special case for pages which are unmapped at the end: end-of-file.
422 *
423 * If the page has no buffers (preferred) then the page is mapped here.
424 *
425 * If all blocks are found to be contiguous then the page can go into the
426 * BIO. Otherwise fall back to the mapping's writepage().
427 *
428 * FIXME: This code wants an estimate of how many pages are still to be
429 * written, so it can intelligently allocate a suitably-sized BIO. For now,
430 * just allocate full-size (16-page) BIOs.
431 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700432
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300433struct mpage_data {
434 struct bio *bio;
435 sector_t last_block_in_bio;
436 get_block_t *get_block;
437 unsigned use_writepage;
438};
439
440static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
Alex Tomas29a814d2008-07-11 19:27:31 -0400441 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700443 struct mpage_data *mpd = data;
444 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 struct address_space *mapping = page->mapping;
446 struct inode *inode = page->mapping->host;
447 const unsigned blkbits = inode->i_blkbits;
448 unsigned long end_index;
449 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
450 sector_t last_block;
451 sector_t block_in_file;
452 sector_t blocks[MAX_BUF_PER_PAGE];
453 unsigned page_block;
454 unsigned first_unmapped = blocks_per_page;
455 struct block_device *bdev = NULL;
456 int boundary = 0;
457 sector_t boundary_block = 0;
458 struct block_device *boundary_bdev = NULL;
459 int length;
460 struct buffer_head map_bh;
461 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700462 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
464 if (page_has_buffers(page)) {
465 struct buffer_head *head = page_buffers(page);
466 struct buffer_head *bh = head;
467
468 /* If they're all mapped and dirty, do it */
469 page_block = 0;
470 do {
471 BUG_ON(buffer_locked(bh));
472 if (!buffer_mapped(bh)) {
473 /*
474 * unmapped dirty buffers are created by
475 * __set_page_dirty_buffers -> mmapped data
476 */
477 if (buffer_dirty(bh))
478 goto confused;
479 if (first_unmapped == blocks_per_page)
480 first_unmapped = page_block;
481 continue;
482 }
483
484 if (first_unmapped != blocks_per_page)
485 goto confused; /* hole -> non-hole */
486
487 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
488 goto confused;
489 if (page_block) {
490 if (bh->b_blocknr != blocks[page_block-1] + 1)
491 goto confused;
492 }
493 blocks[page_block++] = bh->b_blocknr;
494 boundary = buffer_boundary(bh);
495 if (boundary) {
496 boundary_block = bh->b_blocknr;
497 boundary_bdev = bh->b_bdev;
498 }
499 bdev = bh->b_bdev;
500 } while ((bh = bh->b_this_page) != head);
501
502 if (first_unmapped)
503 goto page_is_mapped;
504
505 /*
506 * Page has buffers, but they are all unmapped. The page was
507 * created by pagein or read over a hole which was handled by
508 * block_read_full_page(). If this address_space is also
509 * using mpage_readpages then this can rarely happen.
510 */
511 goto confused;
512 }
513
514 /*
515 * The page has no buffers: map it to disk
516 */
517 BUG_ON(!PageUptodate(page));
Andrew Morton54b21a72006-01-08 01:03:05 -0800518 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 last_block = (i_size - 1) >> blkbits;
520 map_bh.b_page = page;
521 for (page_block = 0; page_block < blocks_per_page; ) {
522
523 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800524 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700525 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 goto confused;
527 if (buffer_new(&map_bh))
528 unmap_underlying_metadata(map_bh.b_bdev,
529 map_bh.b_blocknr);
530 if (buffer_boundary(&map_bh)) {
531 boundary_block = map_bh.b_blocknr;
532 boundary_bdev = map_bh.b_bdev;
533 }
534 if (page_block) {
535 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
536 goto confused;
537 }
538 blocks[page_block++] = map_bh.b_blocknr;
539 boundary = buffer_boundary(&map_bh);
540 bdev = map_bh.b_bdev;
541 if (block_in_file == last_block)
542 break;
543 block_in_file++;
544 }
545 BUG_ON(page_block == 0);
546
547 first_unmapped = page_block;
548
549page_is_mapped:
550 end_index = i_size >> PAGE_CACHE_SHIFT;
551 if (page->index >= end_index) {
552 /*
553 * The page straddles i_size. It must be zeroed out on each
Adam Buchbinder2a61aa42009-12-11 16:35:40 -0500554 * and every writepage invocation because it may be mmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 * "A file is mapped in multiples of the page size. For a file
556 * that is not a multiple of the page size, the remaining memory
557 * is zeroed when mapped, and writes to that region are not
558 * written out to the file."
559 */
560 unsigned offset = i_size & (PAGE_CACHE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
562 if (page->index > end_index || !offset)
563 goto confused;
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800564 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 }
566
567 /*
568 * This page will go to BIO. Do we need to send this BIO off first?
569 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700570 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 bio = mpage_bio_submit(WRITE, bio);
572
573alloc_new:
574 if (bio == NULL) {
575 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
576 bio_get_nr_vecs(bdev), GFP_NOFS|__GFP_HIGH);
577 if (bio == NULL)
578 goto confused;
579 }
580
581 /*
582 * Must try to add the page before marking the buffer clean or
583 * the confused fail path above (OOM) will be very confused when
584 * it finds all bh marked clean (i.e. it will not write anything)
585 */
586 length = first_unmapped << blkbits;
587 if (bio_add_page(bio, page, length, 0) < length) {
588 bio = mpage_bio_submit(WRITE, bio);
589 goto alloc_new;
590 }
591
592 /*
593 * OK, we have our BIO, so we can now mark the buffers clean. Make
594 * sure to only clean buffers which we know we'll be writing.
595 */
596 if (page_has_buffers(page)) {
597 struct buffer_head *head = page_buffers(page);
598 struct buffer_head *bh = head;
599 unsigned buffer_counter = 0;
600
601 do {
602 if (buffer_counter++ == first_unmapped)
603 break;
604 clear_buffer_dirty(bh);
605 bh = bh->b_this_page;
606 } while (bh != head);
607
608 /*
609 * we cannot drop the bh if the page is not uptodate
610 * or a concurrent readpage would fail to serialize with the bh
611 * and it would read from disk before we reach the platter.
612 */
613 if (buffer_heads_over_limit && PageUptodate(page))
614 try_to_free_buffers(page);
615 }
616
617 BUG_ON(PageWriteback(page));
618 set_page_writeback(page);
619 unlock_page(page);
620 if (boundary || (first_unmapped != blocks_per_page)) {
621 bio = mpage_bio_submit(WRITE, bio);
622 if (boundary_block) {
623 write_boundary_block(boundary_bdev,
624 boundary_block, 1 << blkbits);
625 }
626 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700627 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 }
629 goto out;
630
631confused:
632 if (bio)
633 bio = mpage_bio_submit(WRITE, bio);
634
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700635 if (mpd->use_writepage) {
636 ret = mapping->a_ops->writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700638 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 goto out;
640 }
641 /*
642 * The caller has a ref on the inode, so *mapping is stable
643 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700644 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700646 mpd->bio = bio;
647 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648}
649
650/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800651 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 * @mapping: address space structure to write
653 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
654 * @get_block: the filesystem's block mapper function.
655 * If this is NULL then use a_ops->writepage. Otherwise, go
656 * direct-to-BIO.
657 *
658 * This is a library function, which implements the writepages()
659 * address_space_operation.
660 *
661 * If a page is already under I/O, generic_writepages() skips it, even
662 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
663 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
664 * and msync() need to guarantee that all the data which was dirty at the time
665 * the call was made get new I/O started against them. If wbc->sync_mode is
666 * WB_SYNC_ALL then we were called for data integrity and we must wait for
667 * existing IO to complete.
668 */
669int
670mpage_writepages(struct address_space *mapping,
671 struct writeback_control *wbc, get_block_t get_block)
672{
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200673 struct blk_plug plug;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700674 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200676 blk_start_plug(&plug);
677
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 }
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200692 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 return ret;
694}
695EXPORT_SYMBOL(mpage_writepages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
697int mpage_writepage(struct page *page, get_block_t get_block,
698 struct writeback_control *wbc)
699{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700700 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 return ret;
710}
711EXPORT_SYMBOL(mpage_writepage);