blob: fd56ca2ea55611314a841c9a65abba0c88b6ffa6 [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 */
NeilBrown6712ecf2007-09-27 12:47:43 +020043static void mpage_end_io_read(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);
53
54 if (uptodate) {
55 SetPageUptodate(page);
56 } else {
57 ClearPageUptodate(page);
58 SetPageError(page);
59 }
60 unlock_page(page);
61 } while (bvec >= bio->bi_io_vec);
62 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063}
64
NeilBrown6712ecf2007-09-27 12:47:43 +020065static void mpage_end_io_write(struct bio *bio, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -070066{
67 const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
68 struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
69
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 do {
71 struct page *page = bvec->bv_page;
72
73 if (--bvec >= bio->bi_io_vec)
74 prefetchw(&bvec->bv_page->flags);
75
Qu Fuping854715b2005-06-04 15:43:29 -070076 if (!uptodate){
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 SetPageError(page);
Qu Fuping854715b2005-06-04 15:43:29 -070078 if (page->mapping)
79 set_bit(AS_EIO, &page->mapping->flags);
80 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 end_page_writeback(page);
82 } while (bvec >= bio->bi_io_vec);
83 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084}
85
Dmitri Vorobievced117c2009-03-31 00:41:20 +030086static struct bio *mpage_bio_submit(int rw, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070087{
88 bio->bi_end_io = mpage_end_io_read;
89 if (rw == WRITE)
90 bio->bi_end_io = mpage_end_io_write;
91 submit_bio(rw, bio);
92 return NULL;
93}
94
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
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400383 map_bh.b_state = 0;
384 map_bh.b_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
386 struct page *page = list_entry(pages->prev, struct page, lru);
387
388 prefetchw(&page->flags);
389 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -0700390 if (!add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 page->index, GFP_KERNEL)) {
392 bio = do_mpage_readpage(bio, page,
393 nr_pages - page_idx,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800394 &last_block_in_bio, &map_bh,
395 &first_logical_block,
396 get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 }
Nick Piggineb2be182007-10-16 01:24:57 -0700398 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 BUG_ON(!list_empty(pages));
401 if (bio)
402 mpage_bio_submit(READ, bio);
403 return 0;
404}
405EXPORT_SYMBOL(mpage_readpages);
406
407/*
408 * This isn't called much at all
409 */
410int mpage_readpage(struct page *page, get_block_t get_block)
411{
412 struct bio *bio = NULL;
413 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800414 struct buffer_head map_bh;
415 unsigned long first_logical_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400417 map_bh.b_state = 0;
418 map_bh.b_size = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800419 bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
420 &map_bh, &first_logical_block, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 if (bio)
422 mpage_bio_submit(READ, bio);
423 return 0;
424}
425EXPORT_SYMBOL(mpage_readpage);
426
427/*
428 * Writing is not so simple.
429 *
430 * If the page has buffers then they will be used for obtaining the disk
431 * mapping. We only support pages which are fully mapped-and-dirty, with a
432 * special case for pages which are unmapped at the end: end-of-file.
433 *
434 * If the page has no buffers (preferred) then the page is mapped here.
435 *
436 * If all blocks are found to be contiguous then the page can go into the
437 * BIO. Otherwise fall back to the mapping's writepage().
438 *
439 * FIXME: This code wants an estimate of how many pages are still to be
440 * written, so it can intelligently allocate a suitably-sized BIO. For now,
441 * just allocate full-size (16-page) BIOs.
442 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700443
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300444struct mpage_data {
445 struct bio *bio;
446 sector_t last_block_in_bio;
447 get_block_t *get_block;
448 unsigned use_writepage;
449};
450
451static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
Alex Tomas29a814d2008-07-11 19:27:31 -0400452 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700454 struct mpage_data *mpd = data;
455 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 struct address_space *mapping = page->mapping;
457 struct inode *inode = page->mapping->host;
458 const unsigned blkbits = inode->i_blkbits;
459 unsigned long end_index;
460 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
461 sector_t last_block;
462 sector_t block_in_file;
463 sector_t blocks[MAX_BUF_PER_PAGE];
464 unsigned page_block;
465 unsigned first_unmapped = blocks_per_page;
466 struct block_device *bdev = NULL;
467 int boundary = 0;
468 sector_t boundary_block = 0;
469 struct block_device *boundary_bdev = NULL;
470 int length;
471 struct buffer_head map_bh;
472 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700473 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
475 if (page_has_buffers(page)) {
476 struct buffer_head *head = page_buffers(page);
477 struct buffer_head *bh = head;
478
479 /* If they're all mapped and dirty, do it */
480 page_block = 0;
481 do {
482 BUG_ON(buffer_locked(bh));
483 if (!buffer_mapped(bh)) {
484 /*
485 * unmapped dirty buffers are created by
486 * __set_page_dirty_buffers -> mmapped data
487 */
488 if (buffer_dirty(bh))
489 goto confused;
490 if (first_unmapped == blocks_per_page)
491 first_unmapped = page_block;
492 continue;
493 }
494
495 if (first_unmapped != blocks_per_page)
496 goto confused; /* hole -> non-hole */
497
498 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
499 goto confused;
500 if (page_block) {
501 if (bh->b_blocknr != blocks[page_block-1] + 1)
502 goto confused;
503 }
504 blocks[page_block++] = bh->b_blocknr;
505 boundary = buffer_boundary(bh);
506 if (boundary) {
507 boundary_block = bh->b_blocknr;
508 boundary_bdev = bh->b_bdev;
509 }
510 bdev = bh->b_bdev;
511 } while ((bh = bh->b_this_page) != head);
512
513 if (first_unmapped)
514 goto page_is_mapped;
515
516 /*
517 * Page has buffers, but they are all unmapped. The page was
518 * created by pagein or read over a hole which was handled by
519 * block_read_full_page(). If this address_space is also
520 * using mpage_readpages then this can rarely happen.
521 */
522 goto confused;
523 }
524
525 /*
526 * The page has no buffers: map it to disk
527 */
528 BUG_ON(!PageUptodate(page));
Andrew Morton54b21a72006-01-08 01:03:05 -0800529 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 last_block = (i_size - 1) >> blkbits;
531 map_bh.b_page = page;
532 for (page_block = 0; page_block < blocks_per_page; ) {
533
534 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800535 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700536 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 goto confused;
538 if (buffer_new(&map_bh))
539 unmap_underlying_metadata(map_bh.b_bdev,
540 map_bh.b_blocknr);
541 if (buffer_boundary(&map_bh)) {
542 boundary_block = map_bh.b_blocknr;
543 boundary_bdev = map_bh.b_bdev;
544 }
545 if (page_block) {
546 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
547 goto confused;
548 }
549 blocks[page_block++] = map_bh.b_blocknr;
550 boundary = buffer_boundary(&map_bh);
551 bdev = map_bh.b_bdev;
552 if (block_in_file == last_block)
553 break;
554 block_in_file++;
555 }
556 BUG_ON(page_block == 0);
557
558 first_unmapped = page_block;
559
560page_is_mapped:
561 end_index = i_size >> PAGE_CACHE_SHIFT;
562 if (page->index >= end_index) {
563 /*
564 * The page straddles i_size. It must be zeroed out on each
Adam Buchbinder2a61aa42009-12-11 16:35:40 -0500565 * and every writepage invocation because it may be mmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 * "A file is mapped in multiples of the page size. For a file
567 * that is not a multiple of the page size, the remaining memory
568 * is zeroed when mapped, and writes to that region are not
569 * written out to the file."
570 */
571 unsigned offset = i_size & (PAGE_CACHE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 if (page->index > end_index || !offset)
574 goto confused;
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800575 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 }
577
578 /*
579 * This page will go to BIO. Do we need to send this BIO off first?
580 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700581 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 bio = mpage_bio_submit(WRITE, bio);
583
584alloc_new:
585 if (bio == NULL) {
586 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
587 bio_get_nr_vecs(bdev), GFP_NOFS|__GFP_HIGH);
588 if (bio == NULL)
589 goto confused;
590 }
591
592 /*
593 * Must try to add the page before marking the buffer clean or
594 * the confused fail path above (OOM) will be very confused when
595 * it finds all bh marked clean (i.e. it will not write anything)
596 */
597 length = first_unmapped << blkbits;
598 if (bio_add_page(bio, page, length, 0) < length) {
599 bio = mpage_bio_submit(WRITE, bio);
600 goto alloc_new;
601 }
602
603 /*
604 * OK, we have our BIO, so we can now mark the buffers clean. Make
605 * sure to only clean buffers which we know we'll be writing.
606 */
607 if (page_has_buffers(page)) {
608 struct buffer_head *head = page_buffers(page);
609 struct buffer_head *bh = head;
610 unsigned buffer_counter = 0;
611
612 do {
613 if (buffer_counter++ == first_unmapped)
614 break;
615 clear_buffer_dirty(bh);
616 bh = bh->b_this_page;
617 } while (bh != head);
618
619 /*
620 * we cannot drop the bh if the page is not uptodate
621 * or a concurrent readpage would fail to serialize with the bh
622 * and it would read from disk before we reach the platter.
623 */
624 if (buffer_heads_over_limit && PageUptodate(page))
625 try_to_free_buffers(page);
626 }
627
628 BUG_ON(PageWriteback(page));
629 set_page_writeback(page);
630 unlock_page(page);
631 if (boundary || (first_unmapped != blocks_per_page)) {
632 bio = mpage_bio_submit(WRITE, bio);
633 if (boundary_block) {
634 write_boundary_block(boundary_bdev,
635 boundary_block, 1 << blkbits);
636 }
637 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700638 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 }
640 goto out;
641
642confused:
643 if (bio)
644 bio = mpage_bio_submit(WRITE, bio);
645
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700646 if (mpd->use_writepage) {
647 ret = mapping->a_ops->writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700649 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 goto out;
651 }
652 /*
653 * The caller has a ref on the inode, so *mapping is stable
654 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700655 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700657 mpd->bio = bio;
658 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659}
660
661/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800662 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 * @mapping: address space structure to write
664 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
665 * @get_block: the filesystem's block mapper function.
666 * If this is NULL then use a_ops->writepage. Otherwise, go
667 * direct-to-BIO.
668 *
669 * This is a library function, which implements the writepages()
670 * address_space_operation.
671 *
672 * If a page is already under I/O, generic_writepages() skips it, even
673 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
674 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
675 * and msync() need to guarantee that all the data which was dirty at the time
676 * the call was made get new I/O started against them. If wbc->sync_mode is
677 * WB_SYNC_ALL then we were called for data integrity and we must wait for
678 * existing IO to complete.
679 */
680int
681mpage_writepages(struct address_space *mapping,
682 struct writeback_control *wbc, get_block_t get_block)
683{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700684 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700686 if (!get_block)
687 ret = generic_writepages(mapping, wbc);
688 else {
689 struct mpage_data mpd = {
690 .bio = NULL,
691 .last_block_in_bio = 0,
692 .get_block = get_block,
693 .use_writepage = 1,
694 };
695
696 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
697 if (mpd.bio)
698 mpage_bio_submit(WRITE, mpd.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 return ret;
701}
702EXPORT_SYMBOL(mpage_writepages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
704int mpage_writepage(struct page *page, get_block_t get_block,
705 struct writeback_control *wbc)
706{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700707 struct mpage_data mpd = {
708 .bio = NULL,
709 .last_block_in_bio = 0,
710 .get_block = get_block,
711 .use_writepage = 0,
712 };
713 int ret = __mpage_writepage(page, wbc, &mpd);
714 if (mpd.bio)
715 mpage_bio_submit(WRITE, mpd.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 return ret;
717}
718EXPORT_SYMBOL(mpage_writepage);