blob: 6bd9fd90964e2670808ee1d586b4080d68f2a1f1 [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>
Andrew Morton02c43632016-03-15 14:55:15 -070027#include <linux/mm_inline.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/writeback.h>
29#include <linux/backing-dev.h>
30#include <linux/pagevec.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060031#include <linux/cleancache.h>
Akinobu Mita4db96b72014-10-09 15:26:55 -070032#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
34/*
35 * I/O completion handler for multipage BIOs.
36 *
37 * The mpage code never puts partial pages into a BIO (except for end-of-file).
38 * If a page does not map to a contiguous run of blocks then it simply falls
39 * back to block_read_full_page().
40 *
41 * Why is this? If a page's completion depends on a number of different BIOs
42 * which can complete in any order (or at the same time) then determining the
43 * status of that page is hard. See end_buffer_async_read() for the details.
44 * There is no point in duplicating all that complexity.
45 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020046static void mpage_end_io(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070047{
Kent Overstreet2c30c712013-11-07 12:20:26 -080048 struct bio_vec *bv;
49 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Kent Overstreet2c30c712013-11-07 12:20:26 -080051 bio_for_each_segment_all(bv, bio, i) {
52 struct page *page = bv->bv_page;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020053 page_endio(page, bio_data_dir(bio), bio->bi_error);
Kent Overstreet2c30c712013-11-07 12:20:26 -080054 }
55
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057}
58
Dmitri Vorobievced117c2009-03-31 00:41:20 +030059static struct bio *mpage_bio_submit(int rw, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070060{
Hai Shanc32b0d42011-01-13 15:45:51 -080061 bio->bi_end_io = mpage_end_io;
Akinobu Mita4db96b72014-10-09 15:26:55 -070062 guard_bio_eod(rw, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 submit_bio(rw, bio);
64 return NULL;
65}
66
67static struct bio *
68mpage_alloc(struct block_device *bdev,
69 sector_t first_sector, int nr_vecs,
Al Virodd0fc662005-10-07 07:46:04 +010070 gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071{
72 struct bio *bio;
73
74 bio = bio_alloc(gfp_flags, nr_vecs);
75
76 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
77 while (!bio && (nr_vecs /= 2))
78 bio = bio_alloc(gfp_flags, nr_vecs);
79 }
80
81 if (bio) {
82 bio->bi_bdev = bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -070083 bio->bi_iter.bi_sector = first_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 }
85 return bio;
86}
87
88/*
89 * support function for mpage_readpages. The fs supplied get_block might
90 * return an up to date buffer. This is used to map that buffer into
91 * the page, which allows readpage to avoid triggering a duplicate call
92 * to get_block.
93 *
94 * The idea is to avoid adding buffers to pages that don't already have
95 * them. So when the buffer is up to date and the page size == block size,
96 * this marks the page up to date instead of adding new buffers.
97 */
98static void
99map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
100{
101 struct inode *inode = page->mapping->host;
102 struct buffer_head *page_bh, *head;
103 int block = 0;
104
105 if (!page_has_buffers(page)) {
106 /*
107 * don't make any buffers if there is only one buffer on
108 * the page and the page just needs to be set up to date
109 */
110 if (inode->i_blkbits == PAGE_CACHE_SHIFT &&
111 buffer_uptodate(bh)) {
112 SetPageUptodate(page);
113 return;
114 }
115 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
116 }
117 head = page_buffers(page);
118 page_bh = head;
119 do {
120 if (block == page_block) {
121 page_bh->b_state = bh->b_state;
122 page_bh->b_bdev = bh->b_bdev;
123 page_bh->b_blocknr = bh->b_blocknr;
124 break;
125 }
126 page_bh = page_bh->b_this_page;
127 block++;
128 } while (page_bh != head);
129}
130
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800131/*
132 * This is the worker routine which does all the work of mapping the disk
133 * blocks and constructs largest possible bios, submits them for IO if the
134 * blocks are not contiguous on the disk.
135 *
136 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
137 * represent the validity of its disk mapping and to decide when to do the next
138 * get_block() call.
139 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140static struct bio *
141do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800142 sector_t *last_block_in_bio, struct buffer_head *map_bh,
Michal Hocko063d99b2015-10-15 15:28:24 -0700143 unsigned long *first_logical_block, get_block_t get_block,
144 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145{
146 struct inode *inode = page->mapping->host;
147 const unsigned blkbits = inode->i_blkbits;
148 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
149 const unsigned blocksize = 1 << blkbits;
150 sector_t block_in_file;
151 sector_t last_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800152 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 sector_t blocks[MAX_BUF_PER_PAGE];
154 unsigned page_block;
155 unsigned first_hole = blocks_per_page;
156 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 int length;
158 int fully_mapped = 1;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800159 unsigned nblocks;
160 unsigned relative_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 if (page_has_buffers(page))
163 goto confused;
164
Andrew Morton54b21a72006-01-08 01:03:05 -0800165 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800166 last_block = block_in_file + nr_pages * blocks_per_page;
167 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
168 if (last_block > last_block_in_file)
169 last_block = last_block_in_file;
170 page_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800172 /*
173 * Map blocks using the result from the previous get_blocks call first.
174 */
175 nblocks = map_bh->b_size >> blkbits;
176 if (buffer_mapped(map_bh) && block_in_file > *first_logical_block &&
177 block_in_file < (*first_logical_block + nblocks)) {
178 unsigned map_offset = block_in_file - *first_logical_block;
179 unsigned last = nblocks - map_offset;
180
181 for (relative_block = 0; ; relative_block++) {
182 if (relative_block == last) {
183 clear_buffer_mapped(map_bh);
184 break;
185 }
186 if (page_block == blocks_per_page)
187 break;
188 blocks[page_block] = map_bh->b_blocknr + map_offset +
189 relative_block;
190 page_block++;
191 block_in_file++;
192 }
193 bdev = map_bh->b_bdev;
194 }
195
196 /*
197 * Then do more get_blocks calls until we are done with this page.
198 */
199 map_bh->b_page = page;
200 while (page_block < blocks_per_page) {
201 map_bh->b_state = 0;
202 map_bh->b_size = 0;
203
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800205 map_bh->b_size = (last_block-block_in_file) << blkbits;
206 if (get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800208 *first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 }
210
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800211 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 fully_mapped = 0;
213 if (first_hole == blocks_per_page)
214 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800215 page_block++;
216 block_in_file++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 continue;
218 }
219
220 /* some filesystems will copy data into the page during
221 * the get_block call, in which case we don't want to
222 * read it again. map_buffer_to_page copies the data
223 * we just collected from get_block into the page's buffers
224 * so readpage doesn't have to repeat the get_block call
225 */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800226 if (buffer_uptodate(map_bh)) {
227 map_buffer_to_page(page, map_bh, page_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 goto confused;
229 }
230
231 if (first_hole != blocks_per_page)
232 goto confused; /* hole -> non-hole */
233
234 /* Contiguous blocks? */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800235 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800237 nblocks = map_bh->b_size >> blkbits;
238 for (relative_block = 0; ; relative_block++) {
239 if (relative_block == nblocks) {
240 clear_buffer_mapped(map_bh);
241 break;
242 } else if (page_block == blocks_per_page)
243 break;
244 blocks[page_block] = map_bh->b_blocknr+relative_block;
245 page_block++;
246 block_in_file++;
247 }
248 bdev = map_bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 }
250
251 if (first_hole != blocks_per_page) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800252 zero_user_segment(page, first_hole << blkbits, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 if (first_hole == 0) {
254 SetPageUptodate(page);
255 unlock_page(page);
256 goto out;
257 }
258 } else if (fully_mapped) {
259 SetPageMappedToDisk(page);
260 }
261
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600262 if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
263 cleancache_get_page(page) == 0) {
264 SetPageUptodate(page);
265 goto confused;
266 }
267
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 /*
269 * This page will go to BIO. Do we need to send this BIO off first?
270 */
271 if (bio && (*last_block_in_bio != blocks[0] - 1))
272 bio = mpage_bio_submit(READ, bio);
273
274alloc_new:
275 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700276 if (first_hole == blocks_per_page) {
277 if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
278 page))
279 goto out;
280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
Michal Hocko063d99b2015-10-15 15:28:24 -0700282 min_t(int, nr_pages, BIO_MAX_PAGES), gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 if (bio == NULL)
284 goto confused;
285 }
286
287 length = first_hole << blkbits;
288 if (bio_add_page(bio, page, length, 0) < length) {
289 bio = mpage_bio_submit(READ, bio);
290 goto alloc_new;
291 }
292
Miquel van Smoorenburg38c8e612009-01-06 14:39:02 -0800293 relative_block = block_in_file - *first_logical_block;
294 nblocks = map_bh->b_size >> blkbits;
295 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
296 (first_hole != blocks_per_page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 bio = mpage_bio_submit(READ, bio);
298 else
299 *last_block_in_bio = blocks[blocks_per_page - 1];
300out:
301 return bio;
302
303confused:
304 if (bio)
305 bio = mpage_bio_submit(READ, bio);
306 if (!PageUptodate(page))
307 block_read_full_page(page, get_block);
308 else
309 unlock_page(page);
310 goto out;
311}
312
Martin Waitz67be2dd2005-05-01 08:59:26 -0700313/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800314 * mpage_readpages - populate an address space with some pages & start reads against them
Martin Waitz67be2dd2005-05-01 08:59:26 -0700315 * @mapping: the address_space
316 * @pages: The address of a list_head which contains the target pages. These
317 * pages have their ->index populated and are otherwise uninitialised.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700318 * The page at @pages->prev has the lowest file offset, and reads should be
319 * issued in @pages->prev to @pages->next order.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700320 * @nr_pages: The number of pages at *@pages
321 * @get_block: The filesystem's block mapper function.
322 *
323 * This function walks the pages and the blocks within each page, building and
324 * emitting large BIOs.
325 *
326 * If anything unusual happens, such as:
327 *
328 * - encountering a page which has buffers
329 * - encountering a page which has a non-hole after a hole
330 * - encountering a page with non-contiguous blocks
331 *
332 * then this code just gives up and calls the buffer_head-based read function.
333 * It does handle a page which has holes at the end - that is a common case:
334 * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
335 *
336 * BH_Boundary explanation:
337 *
338 * There is a problem. The mpage read code assembles several pages, gets all
339 * their disk mappings, and then submits them all. That's fine, but obtaining
340 * the disk mappings may require I/O. Reads of indirect blocks, for example.
341 *
342 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
343 * submitted in the following order:
344 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
Randy Dunlap78a4a502008-02-29 22:02:31 -0800345 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700346 * because the indirect block has to be read to get the mappings of blocks
347 * 13,14,15,16. Obviously, this impacts performance.
348 *
349 * So what we do it to allow the filesystem's get_block() function to set
350 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
351 * after this one will require I/O against a block which is probably close to
352 * this one. So you should push what I/O you have currently accumulated.
353 *
354 * This all causes the disk requests to be issued in the correct order.
355 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356int
357mpage_readpages(struct address_space *mapping, struct list_head *pages,
358 unsigned nr_pages, get_block_t get_block)
359{
360 struct bio *bio = NULL;
361 unsigned page_idx;
362 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800363 struct buffer_head map_bh;
364 unsigned long first_logical_block = 0;
Michal Hockoc62d2552015-11-06 16:28:49 -0800365 gfp_t gfp = mapping_gfp_constraint(mapping, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400367 map_bh.b_state = 0;
368 map_bh.b_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
Andrew Morton02c43632016-03-15 14:55:15 -0700370 struct page *page = lru_to_page(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
372 prefetchw(&page->flags);
373 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -0700374 if (!add_to_page_cache_lru(page, mapping,
Michal Hocko063d99b2015-10-15 15:28:24 -0700375 page->index,
376 gfp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 bio = do_mpage_readpage(bio, page,
378 nr_pages - page_idx,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800379 &last_block_in_bio, &map_bh,
380 &first_logical_block,
Michal Hocko063d99b2015-10-15 15:28:24 -0700381 get_block, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 }
Nick Piggineb2be182007-10-16 01:24:57 -0700383 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 BUG_ON(!list_empty(pages));
386 if (bio)
387 mpage_bio_submit(READ, bio);
388 return 0;
389}
390EXPORT_SYMBOL(mpage_readpages);
391
392/*
393 * This isn't called much at all
394 */
395int mpage_readpage(struct page *page, get_block_t get_block)
396{
397 struct bio *bio = NULL;
398 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800399 struct buffer_head map_bh;
400 unsigned long first_logical_block = 0;
Michal Hockoc62d2552015-11-06 16:28:49 -0800401 gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400403 map_bh.b_state = 0;
404 map_bh.b_size = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800405 bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
Michal Hocko063d99b2015-10-15 15:28:24 -0700406 &map_bh, &first_logical_block, get_block, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 if (bio)
408 mpage_bio_submit(READ, bio);
409 return 0;
410}
411EXPORT_SYMBOL(mpage_readpage);
412
413/*
414 * Writing is not so simple.
415 *
416 * If the page has buffers then they will be used for obtaining the disk
417 * mapping. We only support pages which are fully mapped-and-dirty, with a
418 * special case for pages which are unmapped at the end: end-of-file.
419 *
420 * If the page has no buffers (preferred) then the page is mapped here.
421 *
422 * If all blocks are found to be contiguous then the page can go into the
423 * BIO. Otherwise fall back to the mapping's writepage().
424 *
425 * FIXME: This code wants an estimate of how many pages are still to be
426 * written, so it can intelligently allocate a suitably-sized BIO. For now,
427 * just allocate full-size (16-page) BIOs.
428 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700429
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300430struct mpage_data {
431 struct bio *bio;
432 sector_t last_block_in_bio;
433 get_block_t *get_block;
434 unsigned use_writepage;
435};
436
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700437/*
438 * We have our BIO, so we can now mark the buffers clean. Make
439 * sure to only clean buffers which we know we'll be writing.
440 */
441static void clean_buffers(struct page *page, unsigned first_unmapped)
442{
443 unsigned buffer_counter = 0;
444 struct buffer_head *bh, *head;
445 if (!page_has_buffers(page))
446 return;
447 head = page_buffers(page);
448 bh = head;
449
450 do {
451 if (buffer_counter++ == first_unmapped)
452 break;
453 clear_buffer_dirty(bh);
454 bh = bh->b_this_page;
455 } while (bh != head);
456
457 /*
458 * we cannot drop the bh if the page is not uptodate or a concurrent
459 * readpage would fail to serialize with the bh and it would read from
460 * disk before we reach the platter.
461 */
462 if (buffer_heads_over_limit && PageUptodate(page))
463 try_to_free_buffers(page);
464}
465
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300466static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
Alex Tomas29a814d2008-07-11 19:27:31 -0400467 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700469 struct mpage_data *mpd = data;
470 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 struct address_space *mapping = page->mapping;
472 struct inode *inode = page->mapping->host;
473 const unsigned blkbits = inode->i_blkbits;
474 unsigned long end_index;
475 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
476 sector_t last_block;
477 sector_t block_in_file;
478 sector_t blocks[MAX_BUF_PER_PAGE];
479 unsigned page_block;
480 unsigned first_unmapped = blocks_per_page;
481 struct block_device *bdev = NULL;
482 int boundary = 0;
483 sector_t boundary_block = 0;
484 struct block_device *boundary_bdev = NULL;
485 int length;
486 struct buffer_head map_bh;
487 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700488 int ret = 0;
Roman Pen5948edb2015-09-15 08:27:25 -0600489 int wr = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
491 if (page_has_buffers(page)) {
492 struct buffer_head *head = page_buffers(page);
493 struct buffer_head *bh = head;
494
495 /* If they're all mapped and dirty, do it */
496 page_block = 0;
497 do {
498 BUG_ON(buffer_locked(bh));
499 if (!buffer_mapped(bh)) {
500 /*
501 * unmapped dirty buffers are created by
502 * __set_page_dirty_buffers -> mmapped data
503 */
504 if (buffer_dirty(bh))
505 goto confused;
506 if (first_unmapped == blocks_per_page)
507 first_unmapped = page_block;
508 continue;
509 }
510
511 if (first_unmapped != blocks_per_page)
512 goto confused; /* hole -> non-hole */
513
514 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
515 goto confused;
516 if (page_block) {
517 if (bh->b_blocknr != blocks[page_block-1] + 1)
518 goto confused;
519 }
520 blocks[page_block++] = bh->b_blocknr;
521 boundary = buffer_boundary(bh);
522 if (boundary) {
523 boundary_block = bh->b_blocknr;
524 boundary_bdev = bh->b_bdev;
525 }
526 bdev = bh->b_bdev;
527 } while ((bh = bh->b_this_page) != head);
528
529 if (first_unmapped)
530 goto page_is_mapped;
531
532 /*
533 * Page has buffers, but they are all unmapped. The page was
534 * created by pagein or read over a hole which was handled by
535 * block_read_full_page(). If this address_space is also
536 * using mpage_readpages then this can rarely happen.
537 */
538 goto confused;
539 }
540
541 /*
542 * The page has no buffers: map it to disk
543 */
544 BUG_ON(!PageUptodate(page));
Andrew Morton54b21a72006-01-08 01:03:05 -0800545 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 last_block = (i_size - 1) >> blkbits;
547 map_bh.b_page = page;
548 for (page_block = 0; page_block < blocks_per_page; ) {
549
550 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800551 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700552 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 goto confused;
554 if (buffer_new(&map_bh))
555 unmap_underlying_metadata(map_bh.b_bdev,
556 map_bh.b_blocknr);
557 if (buffer_boundary(&map_bh)) {
558 boundary_block = map_bh.b_blocknr;
559 boundary_bdev = map_bh.b_bdev;
560 }
561 if (page_block) {
562 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
563 goto confused;
564 }
565 blocks[page_block++] = map_bh.b_blocknr;
566 boundary = buffer_boundary(&map_bh);
567 bdev = map_bh.b_bdev;
568 if (block_in_file == last_block)
569 break;
570 block_in_file++;
571 }
572 BUG_ON(page_block == 0);
573
574 first_unmapped = page_block;
575
576page_is_mapped:
577 end_index = i_size >> PAGE_CACHE_SHIFT;
578 if (page->index >= end_index) {
579 /*
580 * The page straddles i_size. It must be zeroed out on each
Adam Buchbinder2a61aa42009-12-11 16:35:40 -0500581 * and every writepage invocation because it may be mmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 * "A file is mapped in multiples of the page size. For a file
583 * that is not a multiple of the page size, the remaining memory
584 * is zeroed when mapped, and writes to that region are not
585 * written out to the file."
586 */
587 unsigned offset = i_size & (PAGE_CACHE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
589 if (page->index > end_index || !offset)
590 goto confused;
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800591 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 }
593
594 /*
595 * This page will go to BIO. Do we need to send this BIO off first?
596 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700597 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Roman Pen5948edb2015-09-15 08:27:25 -0600598 bio = mpage_bio_submit(wr, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
600alloc_new:
601 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700602 if (first_unmapped == blocks_per_page) {
603 if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
604 page, wbc)) {
605 clean_buffers(page, first_unmapped);
606 goto out;
607 }
608 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
Kent Overstreetb54ffb72015-05-19 14:31:01 +0200610 BIO_MAX_PAGES, GFP_NOFS|__GFP_HIGH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 if (bio == NULL)
612 goto confused;
Tejun Heo429b3fb2015-05-22 17:14:04 -0400613
Tejun Heob16b1de2015-06-02 08:39:48 -0600614 wbc_init_bio(wbc, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 }
616
617 /*
618 * Must try to add the page before marking the buffer clean or
619 * the confused fail path above (OOM) will be very confused when
620 * it finds all bh marked clean (i.e. it will not write anything)
621 */
Tejun Heo2a814902015-05-28 14:50:51 -0400622 wbc_account_io(wbc, page, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 length = first_unmapped << blkbits;
624 if (bio_add_page(bio, page, length, 0) < length) {
Roman Pen5948edb2015-09-15 08:27:25 -0600625 bio = mpage_bio_submit(wr, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 goto alloc_new;
627 }
628
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700629 clean_buffers(page, first_unmapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
631 BUG_ON(PageWriteback(page));
632 set_page_writeback(page);
633 unlock_page(page);
634 if (boundary || (first_unmapped != blocks_per_page)) {
Roman Pen5948edb2015-09-15 08:27:25 -0600635 bio = mpage_bio_submit(wr, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 if (boundary_block) {
637 write_boundary_block(boundary_bdev,
638 boundary_block, 1 << blkbits);
639 }
640 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700641 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 }
643 goto out;
644
645confused:
646 if (bio)
Roman Pen5948edb2015-09-15 08:27:25 -0600647 bio = mpage_bio_submit(wr, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700649 if (mpd->use_writepage) {
650 ret = mapping->a_ops->writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700652 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 goto out;
654 }
655 /*
656 * The caller has a ref on the inode, so *mapping is stable
657 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700658 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700660 mpd->bio = bio;
661 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662}
663
664/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800665 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 * @mapping: address space structure to write
667 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
668 * @get_block: the filesystem's block mapper function.
669 * If this is NULL then use a_ops->writepage. Otherwise, go
670 * direct-to-BIO.
671 *
672 * This is a library function, which implements the writepages()
673 * address_space_operation.
674 *
675 * If a page is already under I/O, generic_writepages() skips it, even
676 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
677 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
678 * and msync() need to guarantee that all the data which was dirty at the time
679 * the call was made get new I/O started against them. If wbc->sync_mode is
680 * WB_SYNC_ALL then we were called for data integrity and we must wait for
681 * existing IO to complete.
682 */
683int
684mpage_writepages(struct address_space *mapping,
685 struct writeback_control *wbc, get_block_t get_block)
686{
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200687 struct blk_plug plug;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700688 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200690 blk_start_plug(&plug);
691
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700692 if (!get_block)
693 ret = generic_writepages(mapping, wbc);
694 else {
695 struct mpage_data mpd = {
696 .bio = NULL,
697 .last_block_in_bio = 0,
698 .get_block = get_block,
699 .use_writepage = 1,
700 };
701
702 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
Roman Pen5948edb2015-09-15 08:27:25 -0600703 if (mpd.bio) {
704 int wr = (wbc->sync_mode == WB_SYNC_ALL ?
705 WRITE_SYNC : WRITE);
706 mpage_bio_submit(wr, mpd.bio);
707 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 }
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200709 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 return ret;
711}
712EXPORT_SYMBOL(mpage_writepages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
714int mpage_writepage(struct page *page, get_block_t get_block,
715 struct writeback_control *wbc)
716{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700717 struct mpage_data mpd = {
718 .bio = NULL,
719 .last_block_in_bio = 0,
720 .get_block = get_block,
721 .use_writepage = 0,
722 };
723 int ret = __mpage_writepage(page, wbc, &mpd);
Roman Pen5948edb2015-09-15 08:27:25 -0600724 if (mpd.bio) {
725 int wr = (wbc->sync_mode == WB_SYNC_ALL ?
726 WRITE_SYNC : WRITE);
727 mpage_bio_submit(wr, mpd.bio);
728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 return ret;
730}
731EXPORT_SYMBOL(mpage_writepage);