blob: 5f9ed622274f411128752d7efd472f954f669a6e [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>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050016#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#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>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060030#include <linux/cleancache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32/*
33 * I/O completion handler for multipage BIOs.
34 *
35 * The mpage code never puts partial pages into a BIO (except for end-of-file).
36 * If a page does not map to a contiguous run of blocks then it simply falls
37 * back to block_read_full_page().
38 *
39 * Why is this? If a page's completion depends on a number of different BIOs
40 * which can complete in any order (or at the same time) then determining the
41 * status of that page is hard. See end_buffer_async_read() for the details.
42 * There is no point in duplicating all that complexity.
43 */
Hai Shanc32b0d42011-01-13 15:45:51 -080044static void mpage_end_io(struct bio *bio, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -070045{
Kent Overstreet2c30c712013-11-07 12:20:26 -080046 struct bio_vec *bv;
47 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Kent Overstreet2c30c712013-11-07 12:20:26 -080049 bio_for_each_segment_all(bv, bio, i) {
50 struct page *page = bv->bv_page;
Matthew Wilcox57d99842014-06-04 16:07:45 -070051 page_endio(page, bio_data_dir(bio), err);
Kent Overstreet2c30c712013-11-07 12:20:26 -080052 }
53
Linus Torvalds1da177e2005-04-16 15:20:36 -070054 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055}
56
Dmitri Vorobievced117c2009-03-31 00:41:20 +030057static struct bio *mpage_bio_submit(int rw, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070058{
Hai Shanc32b0d42011-01-13 15:45:51 -080059 bio->bi_end_io = mpage_end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 submit_bio(rw, bio);
61 return NULL;
62}
63
64static struct bio *
65mpage_alloc(struct block_device *bdev,
66 sector_t first_sector, int nr_vecs,
Al Virodd0fc662005-10-07 07:46:04 +010067 gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070068{
69 struct bio *bio;
70
71 bio = bio_alloc(gfp_flags, nr_vecs);
72
73 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
74 while (!bio && (nr_vecs /= 2))
75 bio = bio_alloc(gfp_flags, nr_vecs);
76 }
77
78 if (bio) {
79 bio->bi_bdev = bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -070080 bio->bi_iter.bi_sector = first_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 }
82 return bio;
83}
84
85/*
86 * support function for mpage_readpages. The fs supplied get_block might
87 * return an up to date buffer. This is used to map that buffer into
88 * the page, which allows readpage to avoid triggering a duplicate call
89 * to get_block.
90 *
91 * The idea is to avoid adding buffers to pages that don't already have
92 * them. So when the buffer is up to date and the page size == block size,
93 * this marks the page up to date instead of adding new buffers.
94 */
95static void
96map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
97{
98 struct inode *inode = page->mapping->host;
99 struct buffer_head *page_bh, *head;
100 int block = 0;
101
102 if (!page_has_buffers(page)) {
103 /*
104 * don't make any buffers if there is only one buffer on
105 * the page and the page just needs to be set up to date
106 */
107 if (inode->i_blkbits == PAGE_CACHE_SHIFT &&
108 buffer_uptodate(bh)) {
109 SetPageUptodate(page);
110 return;
111 }
112 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
113 }
114 head = page_buffers(page);
115 page_bh = head;
116 do {
117 if (block == page_block) {
118 page_bh->b_state = bh->b_state;
119 page_bh->b_bdev = bh->b_bdev;
120 page_bh->b_blocknr = bh->b_blocknr;
121 break;
122 }
123 page_bh = page_bh->b_this_page;
124 block++;
125 } while (page_bh != head);
126}
127
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800128/*
129 * This is the worker routine which does all the work of mapping the disk
130 * blocks and constructs largest possible bios, submits them for IO if the
131 * blocks are not contiguous on the disk.
132 *
133 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
134 * represent the validity of its disk mapping and to decide when to do the next
135 * get_block() call.
136 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137static struct bio *
138do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800139 sector_t *last_block_in_bio, struct buffer_head *map_bh,
140 unsigned long *first_logical_block, get_block_t get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
142 struct inode *inode = page->mapping->host;
143 const unsigned blkbits = inode->i_blkbits;
144 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
145 const unsigned blocksize = 1 << blkbits;
146 sector_t block_in_file;
147 sector_t last_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800148 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 sector_t blocks[MAX_BUF_PER_PAGE];
150 unsigned page_block;
151 unsigned first_hole = blocks_per_page;
152 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 int length;
154 int fully_mapped = 1;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800155 unsigned nblocks;
156 unsigned relative_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
158 if (page_has_buffers(page))
159 goto confused;
160
Andrew Morton54b21a72006-01-08 01:03:05 -0800161 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800162 last_block = block_in_file + nr_pages * blocks_per_page;
163 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
164 if (last_block > last_block_in_file)
165 last_block = last_block_in_file;
166 page_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800168 /*
169 * Map blocks using the result from the previous get_blocks call first.
170 */
171 nblocks = map_bh->b_size >> blkbits;
172 if (buffer_mapped(map_bh) && block_in_file > *first_logical_block &&
173 block_in_file < (*first_logical_block + nblocks)) {
174 unsigned map_offset = block_in_file - *first_logical_block;
175 unsigned last = nblocks - map_offset;
176
177 for (relative_block = 0; ; relative_block++) {
178 if (relative_block == last) {
179 clear_buffer_mapped(map_bh);
180 break;
181 }
182 if (page_block == blocks_per_page)
183 break;
184 blocks[page_block] = map_bh->b_blocknr + map_offset +
185 relative_block;
186 page_block++;
187 block_in_file++;
188 }
189 bdev = map_bh->b_bdev;
190 }
191
192 /*
193 * Then do more get_blocks calls until we are done with this page.
194 */
195 map_bh->b_page = page;
196 while (page_block < blocks_per_page) {
197 map_bh->b_state = 0;
198 map_bh->b_size = 0;
199
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800201 map_bh->b_size = (last_block-block_in_file) << blkbits;
202 if (get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800204 *first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 }
206
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800207 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 fully_mapped = 0;
209 if (first_hole == blocks_per_page)
210 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800211 page_block++;
212 block_in_file++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 continue;
214 }
215
216 /* some filesystems will copy data into the page during
217 * the get_block call, in which case we don't want to
218 * read it again. map_buffer_to_page copies the data
219 * we just collected from get_block into the page's buffers
220 * so readpage doesn't have to repeat the get_block call
221 */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800222 if (buffer_uptodate(map_bh)) {
223 map_buffer_to_page(page, map_bh, page_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 goto confused;
225 }
226
227 if (first_hole != blocks_per_page)
228 goto confused; /* hole -> non-hole */
229
230 /* Contiguous blocks? */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800231 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800233 nblocks = map_bh->b_size >> blkbits;
234 for (relative_block = 0; ; relative_block++) {
235 if (relative_block == nblocks) {
236 clear_buffer_mapped(map_bh);
237 break;
238 } else if (page_block == blocks_per_page)
239 break;
240 blocks[page_block] = map_bh->b_blocknr+relative_block;
241 page_block++;
242 block_in_file++;
243 }
244 bdev = map_bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 }
246
247 if (first_hole != blocks_per_page) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800248 zero_user_segment(page, first_hole << blkbits, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 if (first_hole == 0) {
250 SetPageUptodate(page);
251 unlock_page(page);
252 goto out;
253 }
254 } else if (fully_mapped) {
255 SetPageMappedToDisk(page);
256 }
257
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600258 if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
259 cleancache_get_page(page) == 0) {
260 SetPageUptodate(page);
261 goto confused;
262 }
263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 /*
265 * This page will go to BIO. Do we need to send this BIO off first?
266 */
267 if (bio && (*last_block_in_bio != blocks[0] - 1))
268 bio = mpage_bio_submit(READ, bio);
269
270alloc_new:
271 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700272 if (first_hole == blocks_per_page) {
273 if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
274 page))
275 goto out;
276 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
278 min_t(int, nr_pages, bio_get_nr_vecs(bdev)),
279 GFP_KERNEL);
280 if (bio == NULL)
281 goto confused;
282 }
283
284 length = first_hole << blkbits;
285 if (bio_add_page(bio, page, length, 0) < length) {
286 bio = mpage_bio_submit(READ, bio);
287 goto alloc_new;
288 }
289
Miquel van Smoorenburg38c8e612009-01-06 14:39:02 -0800290 relative_block = block_in_file - *first_logical_block;
291 nblocks = map_bh->b_size >> blkbits;
292 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
293 (first_hole != blocks_per_page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 bio = mpage_bio_submit(READ, bio);
295 else
296 *last_block_in_bio = blocks[blocks_per_page - 1];
297out:
298 return bio;
299
300confused:
301 if (bio)
302 bio = mpage_bio_submit(READ, bio);
303 if (!PageUptodate(page))
304 block_read_full_page(page, get_block);
305 else
306 unlock_page(page);
307 goto out;
308}
309
Martin Waitz67be2dd2005-05-01 08:59:26 -0700310/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800311 * mpage_readpages - populate an address space with some pages & start reads against them
Martin Waitz67be2dd2005-05-01 08:59:26 -0700312 * @mapping: the address_space
313 * @pages: The address of a list_head which contains the target pages. These
314 * pages have their ->index populated and are otherwise uninitialised.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700315 * The page at @pages->prev has the lowest file offset, and reads should be
316 * issued in @pages->prev to @pages->next order.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700317 * @nr_pages: The number of pages at *@pages
318 * @get_block: The filesystem's block mapper function.
319 *
320 * This function walks the pages and the blocks within each page, building and
321 * emitting large BIOs.
322 *
323 * If anything unusual happens, such as:
324 *
325 * - encountering a page which has buffers
326 * - encountering a page which has a non-hole after a hole
327 * - encountering a page with non-contiguous blocks
328 *
329 * then this code just gives up and calls the buffer_head-based read function.
330 * It does handle a page which has holes at the end - that is a common case:
331 * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
332 *
333 * BH_Boundary explanation:
334 *
335 * There is a problem. The mpage read code assembles several pages, gets all
336 * their disk mappings, and then submits them all. That's fine, but obtaining
337 * the disk mappings may require I/O. Reads of indirect blocks, for example.
338 *
339 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
340 * submitted in the following order:
341 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
Randy Dunlap78a4a502008-02-29 22:02:31 -0800342 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700343 * because the indirect block has to be read to get the mappings of blocks
344 * 13,14,15,16. Obviously, this impacts performance.
345 *
346 * So what we do it to allow the filesystem's get_block() function to set
347 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
348 * after this one will require I/O against a block which is probably close to
349 * this one. So you should push what I/O you have currently accumulated.
350 *
351 * This all causes the disk requests to be issued in the correct order.
352 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353int
354mpage_readpages(struct address_space *mapping, struct list_head *pages,
355 unsigned nr_pages, get_block_t get_block)
356{
357 struct bio *bio = NULL;
358 unsigned page_idx;
359 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800360 struct buffer_head map_bh;
361 unsigned long first_logical_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400363 map_bh.b_state = 0;
364 map_bh.b_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
366 struct page *page = list_entry(pages->prev, struct page, lru);
367
368 prefetchw(&page->flags);
369 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -0700370 if (!add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 page->index, GFP_KERNEL)) {
372 bio = do_mpage_readpage(bio, page,
373 nr_pages - page_idx,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800374 &last_block_in_bio, &map_bh,
375 &first_logical_block,
376 get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 }
Nick Piggineb2be182007-10-16 01:24:57 -0700378 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 BUG_ON(!list_empty(pages));
381 if (bio)
382 mpage_bio_submit(READ, bio);
383 return 0;
384}
385EXPORT_SYMBOL(mpage_readpages);
386
387/*
388 * This isn't called much at all
389 */
390int mpage_readpage(struct page *page, get_block_t get_block)
391{
392 struct bio *bio = NULL;
393 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800394 struct buffer_head map_bh;
395 unsigned long first_logical_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400397 map_bh.b_state = 0;
398 map_bh.b_size = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800399 bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
400 &map_bh, &first_logical_block, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 if (bio)
402 mpage_bio_submit(READ, bio);
403 return 0;
404}
405EXPORT_SYMBOL(mpage_readpage);
406
407/*
408 * Writing is not so simple.
409 *
410 * If the page has buffers then they will be used for obtaining the disk
411 * mapping. We only support pages which are fully mapped-and-dirty, with a
412 * special case for pages which are unmapped at the end: end-of-file.
413 *
414 * If the page has no buffers (preferred) then the page is mapped here.
415 *
416 * If all blocks are found to be contiguous then the page can go into the
417 * BIO. Otherwise fall back to the mapping's writepage().
418 *
419 * FIXME: This code wants an estimate of how many pages are still to be
420 * written, so it can intelligently allocate a suitably-sized BIO. For now,
421 * just allocate full-size (16-page) BIOs.
422 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700423
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300424struct mpage_data {
425 struct bio *bio;
426 sector_t last_block_in_bio;
427 get_block_t *get_block;
428 unsigned use_writepage;
429};
430
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700431/*
432 * We have our BIO, so we can now mark the buffers clean. Make
433 * sure to only clean buffers which we know we'll be writing.
434 */
435static void clean_buffers(struct page *page, unsigned first_unmapped)
436{
437 unsigned buffer_counter = 0;
438 struct buffer_head *bh, *head;
439 if (!page_has_buffers(page))
440 return;
441 head = page_buffers(page);
442 bh = head;
443
444 do {
445 if (buffer_counter++ == first_unmapped)
446 break;
447 clear_buffer_dirty(bh);
448 bh = bh->b_this_page;
449 } while (bh != head);
450
451 /*
452 * we cannot drop the bh if the page is not uptodate or a concurrent
453 * readpage would fail to serialize with the bh and it would read from
454 * disk before we reach the platter.
455 */
456 if (buffer_heads_over_limit && PageUptodate(page))
457 try_to_free_buffers(page);
458}
459
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300460static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
Alex Tomas29a814d2008-07-11 19:27:31 -0400461 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700463 struct mpage_data *mpd = data;
464 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 struct address_space *mapping = page->mapping;
466 struct inode *inode = page->mapping->host;
467 const unsigned blkbits = inode->i_blkbits;
468 unsigned long end_index;
469 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
470 sector_t last_block;
471 sector_t block_in_file;
472 sector_t blocks[MAX_BUF_PER_PAGE];
473 unsigned page_block;
474 unsigned first_unmapped = blocks_per_page;
475 struct block_device *bdev = NULL;
476 int boundary = 0;
477 sector_t boundary_block = 0;
478 struct block_device *boundary_bdev = NULL;
479 int length;
480 struct buffer_head map_bh;
481 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700482 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
484 if (page_has_buffers(page)) {
485 struct buffer_head *head = page_buffers(page);
486 struct buffer_head *bh = head;
487
488 /* If they're all mapped and dirty, do it */
489 page_block = 0;
490 do {
491 BUG_ON(buffer_locked(bh));
492 if (!buffer_mapped(bh)) {
493 /*
494 * unmapped dirty buffers are created by
495 * __set_page_dirty_buffers -> mmapped data
496 */
497 if (buffer_dirty(bh))
498 goto confused;
499 if (first_unmapped == blocks_per_page)
500 first_unmapped = page_block;
501 continue;
502 }
503
504 if (first_unmapped != blocks_per_page)
505 goto confused; /* hole -> non-hole */
506
507 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
508 goto confused;
509 if (page_block) {
510 if (bh->b_blocknr != blocks[page_block-1] + 1)
511 goto confused;
512 }
513 blocks[page_block++] = bh->b_blocknr;
514 boundary = buffer_boundary(bh);
515 if (boundary) {
516 boundary_block = bh->b_blocknr;
517 boundary_bdev = bh->b_bdev;
518 }
519 bdev = bh->b_bdev;
520 } while ((bh = bh->b_this_page) != head);
521
522 if (first_unmapped)
523 goto page_is_mapped;
524
525 /*
526 * Page has buffers, but they are all unmapped. The page was
527 * created by pagein or read over a hole which was handled by
528 * block_read_full_page(). If this address_space is also
529 * using mpage_readpages then this can rarely happen.
530 */
531 goto confused;
532 }
533
534 /*
535 * The page has no buffers: map it to disk
536 */
537 BUG_ON(!PageUptodate(page));
Andrew Morton54b21a72006-01-08 01:03:05 -0800538 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 last_block = (i_size - 1) >> blkbits;
540 map_bh.b_page = page;
541 for (page_block = 0; page_block < blocks_per_page; ) {
542
543 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800544 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700545 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 goto confused;
547 if (buffer_new(&map_bh))
548 unmap_underlying_metadata(map_bh.b_bdev,
549 map_bh.b_blocknr);
550 if (buffer_boundary(&map_bh)) {
551 boundary_block = map_bh.b_blocknr;
552 boundary_bdev = map_bh.b_bdev;
553 }
554 if (page_block) {
555 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
556 goto confused;
557 }
558 blocks[page_block++] = map_bh.b_blocknr;
559 boundary = buffer_boundary(&map_bh);
560 bdev = map_bh.b_bdev;
561 if (block_in_file == last_block)
562 break;
563 block_in_file++;
564 }
565 BUG_ON(page_block == 0);
566
567 first_unmapped = page_block;
568
569page_is_mapped:
570 end_index = i_size >> PAGE_CACHE_SHIFT;
571 if (page->index >= end_index) {
572 /*
573 * The page straddles i_size. It must be zeroed out on each
Adam Buchbinder2a61aa42009-12-11 16:35:40 -0500574 * and every writepage invocation because it may be mmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 * "A file is mapped in multiples of the page size. For a file
576 * that is not a multiple of the page size, the remaining memory
577 * is zeroed when mapped, and writes to that region are not
578 * written out to the file."
579 */
580 unsigned offset = i_size & (PAGE_CACHE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581
582 if (page->index > end_index || !offset)
583 goto confused;
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800584 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 }
586
587 /*
588 * This page will go to BIO. Do we need to send this BIO off first?
589 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700590 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 bio = mpage_bio_submit(WRITE, bio);
592
593alloc_new:
594 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700595 if (first_unmapped == blocks_per_page) {
596 if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
597 page, wbc)) {
598 clean_buffers(page, first_unmapped);
599 goto out;
600 }
601 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
603 bio_get_nr_vecs(bdev), GFP_NOFS|__GFP_HIGH);
604 if (bio == NULL)
605 goto confused;
606 }
607
608 /*
609 * Must try to add the page before marking the buffer clean or
610 * the confused fail path above (OOM) will be very confused when
611 * it finds all bh marked clean (i.e. it will not write anything)
612 */
613 length = first_unmapped << blkbits;
614 if (bio_add_page(bio, page, length, 0) < length) {
615 bio = mpage_bio_submit(WRITE, bio);
616 goto alloc_new;
617 }
618
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700619 clean_buffers(page, first_unmapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 BUG_ON(PageWriteback(page));
622 set_page_writeback(page);
623 unlock_page(page);
624 if (boundary || (first_unmapped != blocks_per_page)) {
625 bio = mpage_bio_submit(WRITE, bio);
626 if (boundary_block) {
627 write_boundary_block(boundary_bdev,
628 boundary_block, 1 << blkbits);
629 }
630 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700631 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 }
633 goto out;
634
635confused:
636 if (bio)
637 bio = mpage_bio_submit(WRITE, bio);
638
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700639 if (mpd->use_writepage) {
640 ret = mapping->a_ops->writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700642 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 goto out;
644 }
645 /*
646 * The caller has a ref on the inode, so *mapping is stable
647 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700648 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700650 mpd->bio = bio;
651 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652}
653
654/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800655 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 * @mapping: address space structure to write
657 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
658 * @get_block: the filesystem's block mapper function.
659 * If this is NULL then use a_ops->writepage. Otherwise, go
660 * direct-to-BIO.
661 *
662 * This is a library function, which implements the writepages()
663 * address_space_operation.
664 *
665 * If a page is already under I/O, generic_writepages() skips it, even
666 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
667 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
668 * and msync() need to guarantee that all the data which was dirty at the time
669 * the call was made get new I/O started against them. If wbc->sync_mode is
670 * WB_SYNC_ALL then we were called for data integrity and we must wait for
671 * existing IO to complete.
672 */
673int
674mpage_writepages(struct address_space *mapping,
675 struct writeback_control *wbc, get_block_t get_block)
676{
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200677 struct blk_plug plug;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700678 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200680 blk_start_plug(&plug);
681
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700682 if (!get_block)
683 ret = generic_writepages(mapping, wbc);
684 else {
685 struct mpage_data mpd = {
686 .bio = NULL,
687 .last_block_in_bio = 0,
688 .get_block = get_block,
689 .use_writepage = 1,
690 };
691
692 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
693 if (mpd.bio)
694 mpage_bio_submit(WRITE, mpd.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 }
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200696 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 return ret;
698}
699EXPORT_SYMBOL(mpage_writepages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
701int mpage_writepage(struct page *page, get_block_t get_block,
702 struct writeback_control *wbc)
703{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700704 struct mpage_data mpd = {
705 .bio = NULL,
706 .last_block_in_bio = 0,
707 .get_block = get_block,
708 .use_writepage = 0,
709 };
710 int ret = __mpage_writepage(page, wbc, &mpd);
711 if (mpd.bio)
712 mpage_bio_submit(WRITE, mpd.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 return ret;
714}
715EXPORT_SYMBOL(mpage_writepage);