blob: 778a4ddef77a21844b08af82058d3b188371dc01 [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>
Akinobu Mita4db96b72014-10-09 15:26:55 -070031#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33/*
34 * I/O completion handler for multipage BIOs.
35 *
36 * The mpage code never puts partial pages into a BIO (except for end-of-file).
37 * If a page does not map to a contiguous run of blocks then it simply falls
38 * back to block_read_full_page().
39 *
40 * Why is this? If a page's completion depends on a number of different BIOs
41 * which can complete in any order (or at the same time) then determining the
42 * status of that page is hard. See end_buffer_async_read() for the details.
43 * There is no point in duplicating all that complexity.
44 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020045static void mpage_end_io(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070046{
Kent Overstreet2c30c712013-11-07 12:20:26 -080047 struct bio_vec *bv;
48 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Kent Overstreet2c30c712013-11-07 12:20:26 -080050 bio_for_each_segment_all(bv, bio, i) {
51 struct page *page = bv->bv_page;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020052 page_endio(page, bio_data_dir(bio), bio->bi_error);
Kent Overstreet2c30c712013-11-07 12:20:26 -080053 }
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056}
57
Dmitri Vorobievced117c2009-03-31 00:41:20 +030058static struct bio *mpage_bio_submit(int rw, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070059{
Hai Shanc32b0d42011-01-13 15:45:51 -080060 bio->bi_end_io = mpage_end_io;
Akinobu Mita4db96b72014-10-09 15:26:55 -070061 guard_bio_eod(rw, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 submit_bio(rw, bio);
63 return NULL;
64}
65
66static struct bio *
67mpage_alloc(struct block_device *bdev,
68 sector_t first_sector, int nr_vecs,
Al Virodd0fc662005-10-07 07:46:04 +010069 gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070070{
71 struct bio *bio;
72
73 bio = bio_alloc(gfp_flags, nr_vecs);
74
75 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
76 while (!bio && (nr_vecs /= 2))
77 bio = bio_alloc(gfp_flags, nr_vecs);
78 }
79
80 if (bio) {
81 bio->bi_bdev = bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -070082 bio->bi_iter.bi_sector = first_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 }
84 return bio;
85}
86
87/*
88 * support function for mpage_readpages. The fs supplied get_block might
89 * return an up to date buffer. This is used to map that buffer into
90 * the page, which allows readpage to avoid triggering a duplicate call
91 * to get_block.
92 *
93 * The idea is to avoid adding buffers to pages that don't already have
94 * them. So when the buffer is up to date and the page size == block size,
95 * this marks the page up to date instead of adding new buffers.
96 */
97static void
98map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
99{
100 struct inode *inode = page->mapping->host;
101 struct buffer_head *page_bh, *head;
102 int block = 0;
103
104 if (!page_has_buffers(page)) {
105 /*
106 * don't make any buffers if there is only one buffer on
107 * the page and the page just needs to be set up to date
108 */
109 if (inode->i_blkbits == PAGE_CACHE_SHIFT &&
110 buffer_uptodate(bh)) {
111 SetPageUptodate(page);
112 return;
113 }
114 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
115 }
116 head = page_buffers(page);
117 page_bh = head;
118 do {
119 if (block == page_block) {
120 page_bh->b_state = bh->b_state;
121 page_bh->b_bdev = bh->b_bdev;
122 page_bh->b_blocknr = bh->b_blocknr;
123 break;
124 }
125 page_bh = page_bh->b_this_page;
126 block++;
127 } while (page_bh != head);
128}
129
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800130/*
131 * This is the worker routine which does all the work of mapping the disk
132 * blocks and constructs largest possible bios, submits them for IO if the
133 * blocks are not contiguous on the disk.
134 *
135 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
136 * represent the validity of its disk mapping and to decide when to do the next
137 * get_block() call.
138 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139static struct bio *
140do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800141 sector_t *last_block_in_bio, struct buffer_head *map_bh,
142 unsigned long *first_logical_block, get_block_t get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143{
144 struct inode *inode = page->mapping->host;
145 const unsigned blkbits = inode->i_blkbits;
146 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
147 const unsigned blocksize = 1 << blkbits;
148 sector_t block_in_file;
149 sector_t last_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800150 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 sector_t blocks[MAX_BUF_PER_PAGE];
152 unsigned page_block;
153 unsigned first_hole = blocks_per_page;
154 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 int length;
156 int fully_mapped = 1;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800157 unsigned nblocks;
158 unsigned relative_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
160 if (page_has_buffers(page))
161 goto confused;
162
Andrew Morton54b21a72006-01-08 01:03:05 -0800163 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800164 last_block = block_in_file + nr_pages * blocks_per_page;
165 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
166 if (last_block > last_block_in_file)
167 last_block = last_block_in_file;
168 page_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800170 /*
171 * Map blocks using the result from the previous get_blocks call first.
172 */
173 nblocks = map_bh->b_size >> blkbits;
174 if (buffer_mapped(map_bh) && block_in_file > *first_logical_block &&
175 block_in_file < (*first_logical_block + nblocks)) {
176 unsigned map_offset = block_in_file - *first_logical_block;
177 unsigned last = nblocks - map_offset;
178
179 for (relative_block = 0; ; relative_block++) {
180 if (relative_block == last) {
181 clear_buffer_mapped(map_bh);
182 break;
183 }
184 if (page_block == blocks_per_page)
185 break;
186 blocks[page_block] = map_bh->b_blocknr + map_offset +
187 relative_block;
188 page_block++;
189 block_in_file++;
190 }
191 bdev = map_bh->b_bdev;
192 }
193
194 /*
195 * Then do more get_blocks calls until we are done with this page.
196 */
197 map_bh->b_page = page;
198 while (page_block < blocks_per_page) {
199 map_bh->b_state = 0;
200 map_bh->b_size = 0;
201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800203 map_bh->b_size = (last_block-block_in_file) << blkbits;
204 if (get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800206 *first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 }
208
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800209 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 fully_mapped = 0;
211 if (first_hole == blocks_per_page)
212 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800213 page_block++;
214 block_in_file++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 continue;
216 }
217
218 /* some filesystems will copy data into the page during
219 * the get_block call, in which case we don't want to
220 * read it again. map_buffer_to_page copies the data
221 * we just collected from get_block into the page's buffers
222 * so readpage doesn't have to repeat the get_block call
223 */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800224 if (buffer_uptodate(map_bh)) {
225 map_buffer_to_page(page, map_bh, page_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 goto confused;
227 }
228
229 if (first_hole != blocks_per_page)
230 goto confused; /* hole -> non-hole */
231
232 /* Contiguous blocks? */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800233 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800235 nblocks = map_bh->b_size >> blkbits;
236 for (relative_block = 0; ; relative_block++) {
237 if (relative_block == nblocks) {
238 clear_buffer_mapped(map_bh);
239 break;
240 } else if (page_block == blocks_per_page)
241 break;
242 blocks[page_block] = map_bh->b_blocknr+relative_block;
243 page_block++;
244 block_in_file++;
245 }
246 bdev = map_bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 }
248
249 if (first_hole != blocks_per_page) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800250 zero_user_segment(page, first_hole << blkbits, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 if (first_hole == 0) {
252 SetPageUptodate(page);
253 unlock_page(page);
254 goto out;
255 }
256 } else if (fully_mapped) {
257 SetPageMappedToDisk(page);
258 }
259
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600260 if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
261 cleancache_get_page(page) == 0) {
262 SetPageUptodate(page);
263 goto confused;
264 }
265
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 /*
267 * This page will go to BIO. Do we need to send this BIO off first?
268 */
269 if (bio && (*last_block_in_bio != blocks[0] - 1))
270 bio = mpage_bio_submit(READ, bio);
271
272alloc_new:
273 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700274 if (first_hole == blocks_per_page) {
275 if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
276 page))
277 goto out;
278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
Kent Overstreetb54ffb72015-05-19 14:31:01 +0200280 min_t(int, nr_pages, BIO_MAX_PAGES),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 GFP_KERNEL);
282 if (bio == NULL)
283 goto confused;
284 }
285
286 length = first_hole << blkbits;
287 if (bio_add_page(bio, page, length, 0) < length) {
288 bio = mpage_bio_submit(READ, bio);
289 goto alloc_new;
290 }
291
Miquel van Smoorenburg38c8e612009-01-06 14:39:02 -0800292 relative_block = block_in_file - *first_logical_block;
293 nblocks = map_bh->b_size >> blkbits;
294 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
295 (first_hole != blocks_per_page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 bio = mpage_bio_submit(READ, bio);
297 else
298 *last_block_in_bio = blocks[blocks_per_page - 1];
299out:
300 return bio;
301
302confused:
303 if (bio)
304 bio = mpage_bio_submit(READ, bio);
305 if (!PageUptodate(page))
306 block_read_full_page(page, get_block);
307 else
308 unlock_page(page);
309 goto out;
310}
311
Martin Waitz67be2dd2005-05-01 08:59:26 -0700312/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800313 * mpage_readpages - populate an address space with some pages & start reads against them
Martin Waitz67be2dd2005-05-01 08:59:26 -0700314 * @mapping: the address_space
315 * @pages: The address of a list_head which contains the target pages. These
316 * pages have their ->index populated and are otherwise uninitialised.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700317 * The page at @pages->prev has the lowest file offset, and reads should be
318 * issued in @pages->prev to @pages->next order.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700319 * @nr_pages: The number of pages at *@pages
320 * @get_block: The filesystem's block mapper function.
321 *
322 * This function walks the pages and the blocks within each page, building and
323 * emitting large BIOs.
324 *
325 * If anything unusual happens, such as:
326 *
327 * - encountering a page which has buffers
328 * - encountering a page which has a non-hole after a hole
329 * - encountering a page with non-contiguous blocks
330 *
331 * then this code just gives up and calls the buffer_head-based read function.
332 * It does handle a page which has holes at the end - that is a common case:
333 * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
334 *
335 * BH_Boundary explanation:
336 *
337 * There is a problem. The mpage read code assembles several pages, gets all
338 * their disk mappings, and then submits them all. That's fine, but obtaining
339 * the disk mappings may require I/O. Reads of indirect blocks, for example.
340 *
341 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
342 * submitted in the following order:
343 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
Randy Dunlap78a4a502008-02-29 22:02:31 -0800344 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700345 * because the indirect block has to be read to get the mappings of blocks
346 * 13,14,15,16. Obviously, this impacts performance.
347 *
348 * So what we do it to allow the filesystem's get_block() function to set
349 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
350 * after this one will require I/O against a block which is probably close to
351 * this one. So you should push what I/O you have currently accumulated.
352 *
353 * This all causes the disk requests to be issued in the correct order.
354 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355int
356mpage_readpages(struct address_space *mapping, struct list_head *pages,
357 unsigned nr_pages, get_block_t get_block)
358{
359 struct bio *bio = NULL;
360 unsigned page_idx;
361 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800362 struct buffer_head map_bh;
363 unsigned long first_logical_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400365 map_bh.b_state = 0;
366 map_bh.b_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
368 struct page *page = list_entry(pages->prev, struct page, lru);
369
370 prefetchw(&page->flags);
371 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -0700372 if (!add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 page->index, GFP_KERNEL)) {
374 bio = do_mpage_readpage(bio, page,
375 nr_pages - page_idx,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800376 &last_block_in_bio, &map_bh,
377 &first_logical_block,
378 get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 }
Nick Piggineb2be182007-10-16 01:24:57 -0700380 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 BUG_ON(!list_empty(pages));
383 if (bio)
384 mpage_bio_submit(READ, bio);
385 return 0;
386}
387EXPORT_SYMBOL(mpage_readpages);
388
389/*
390 * This isn't called much at all
391 */
392int mpage_readpage(struct page *page, get_block_t get_block)
393{
394 struct bio *bio = NULL;
395 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800396 struct buffer_head map_bh;
397 unsigned long first_logical_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400399 map_bh.b_state = 0;
400 map_bh.b_size = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800401 bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
402 &map_bh, &first_logical_block, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 if (bio)
404 mpage_bio_submit(READ, bio);
405 return 0;
406}
407EXPORT_SYMBOL(mpage_readpage);
408
409/*
410 * Writing is not so simple.
411 *
412 * If the page has buffers then they will be used for obtaining the disk
413 * mapping. We only support pages which are fully mapped-and-dirty, with a
414 * special case for pages which are unmapped at the end: end-of-file.
415 *
416 * If the page has no buffers (preferred) then the page is mapped here.
417 *
418 * If all blocks are found to be contiguous then the page can go into the
419 * BIO. Otherwise fall back to the mapping's writepage().
420 *
421 * FIXME: This code wants an estimate of how many pages are still to be
422 * written, so it can intelligently allocate a suitably-sized BIO. For now,
423 * just allocate full-size (16-page) BIOs.
424 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700425
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300426struct mpage_data {
427 struct bio *bio;
428 sector_t last_block_in_bio;
429 get_block_t *get_block;
430 unsigned use_writepage;
431};
432
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700433/*
434 * We have our BIO, so we can now mark the buffers clean. Make
435 * sure to only clean buffers which we know we'll be writing.
436 */
437static void clean_buffers(struct page *page, unsigned first_unmapped)
438{
439 unsigned buffer_counter = 0;
440 struct buffer_head *bh, *head;
441 if (!page_has_buffers(page))
442 return;
443 head = page_buffers(page);
444 bh = head;
445
446 do {
447 if (buffer_counter++ == first_unmapped)
448 break;
449 clear_buffer_dirty(bh);
450 bh = bh->b_this_page;
451 } while (bh != head);
452
453 /*
454 * we cannot drop the bh if the page is not uptodate or a concurrent
455 * readpage would fail to serialize with the bh and it would read from
456 * disk before we reach the platter.
457 */
458 if (buffer_heads_over_limit && PageUptodate(page))
459 try_to_free_buffers(page);
460}
461
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300462static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
Alex Tomas29a814d2008-07-11 19:27:31 -0400463 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700465 struct mpage_data *mpd = data;
466 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 struct address_space *mapping = page->mapping;
468 struct inode *inode = page->mapping->host;
469 const unsigned blkbits = inode->i_blkbits;
470 unsigned long end_index;
471 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
472 sector_t last_block;
473 sector_t block_in_file;
474 sector_t blocks[MAX_BUF_PER_PAGE];
475 unsigned page_block;
476 unsigned first_unmapped = blocks_per_page;
477 struct block_device *bdev = NULL;
478 int boundary = 0;
479 sector_t boundary_block = 0;
480 struct block_device *boundary_bdev = NULL;
481 int length;
482 struct buffer_head map_bh;
483 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700484 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
486 if (page_has_buffers(page)) {
487 struct buffer_head *head = page_buffers(page);
488 struct buffer_head *bh = head;
489
490 /* If they're all mapped and dirty, do it */
491 page_block = 0;
492 do {
493 BUG_ON(buffer_locked(bh));
494 if (!buffer_mapped(bh)) {
495 /*
496 * unmapped dirty buffers are created by
497 * __set_page_dirty_buffers -> mmapped data
498 */
499 if (buffer_dirty(bh))
500 goto confused;
501 if (first_unmapped == blocks_per_page)
502 first_unmapped = page_block;
503 continue;
504 }
505
506 if (first_unmapped != blocks_per_page)
507 goto confused; /* hole -> non-hole */
508
509 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
510 goto confused;
511 if (page_block) {
512 if (bh->b_blocknr != blocks[page_block-1] + 1)
513 goto confused;
514 }
515 blocks[page_block++] = bh->b_blocknr;
516 boundary = buffer_boundary(bh);
517 if (boundary) {
518 boundary_block = bh->b_blocknr;
519 boundary_bdev = bh->b_bdev;
520 }
521 bdev = bh->b_bdev;
522 } while ((bh = bh->b_this_page) != head);
523
524 if (first_unmapped)
525 goto page_is_mapped;
526
527 /*
528 * Page has buffers, but they are all unmapped. The page was
529 * created by pagein or read over a hole which was handled by
530 * block_read_full_page(). If this address_space is also
531 * using mpage_readpages then this can rarely happen.
532 */
533 goto confused;
534 }
535
536 /*
537 * The page has no buffers: map it to disk
538 */
539 BUG_ON(!PageUptodate(page));
Andrew Morton54b21a72006-01-08 01:03:05 -0800540 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 last_block = (i_size - 1) >> blkbits;
542 map_bh.b_page = page;
543 for (page_block = 0; page_block < blocks_per_page; ) {
544
545 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800546 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700547 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 goto confused;
549 if (buffer_new(&map_bh))
550 unmap_underlying_metadata(map_bh.b_bdev,
551 map_bh.b_blocknr);
552 if (buffer_boundary(&map_bh)) {
553 boundary_block = map_bh.b_blocknr;
554 boundary_bdev = map_bh.b_bdev;
555 }
556 if (page_block) {
557 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
558 goto confused;
559 }
560 blocks[page_block++] = map_bh.b_blocknr;
561 boundary = buffer_boundary(&map_bh);
562 bdev = map_bh.b_bdev;
563 if (block_in_file == last_block)
564 break;
565 block_in_file++;
566 }
567 BUG_ON(page_block == 0);
568
569 first_unmapped = page_block;
570
571page_is_mapped:
572 end_index = i_size >> PAGE_CACHE_SHIFT;
573 if (page->index >= end_index) {
574 /*
575 * The page straddles i_size. It must be zeroed out on each
Adam Buchbinder2a61aa42009-12-11 16:35:40 -0500576 * and every writepage invocation because it may be mmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 * "A file is mapped in multiples of the page size. For a file
578 * that is not a multiple of the page size, the remaining memory
579 * is zeroed when mapped, and writes to that region are not
580 * written out to the file."
581 */
582 unsigned offset = i_size & (PAGE_CACHE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584 if (page->index > end_index || !offset)
585 goto confused;
Christoph Lametereebd2aa2008-02-04 22:28:29 -0800586 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 }
588
589 /*
590 * This page will go to BIO. Do we need to send this BIO off first?
591 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700592 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 bio = mpage_bio_submit(WRITE, bio);
594
595alloc_new:
596 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700597 if (first_unmapped == blocks_per_page) {
598 if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
599 page, wbc)) {
600 clean_buffers(page, first_unmapped);
601 goto out;
602 }
603 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
Kent Overstreetb54ffb72015-05-19 14:31:01 +0200605 BIO_MAX_PAGES, GFP_NOFS|__GFP_HIGH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 if (bio == NULL)
607 goto confused;
Tejun Heo429b3fb2015-05-22 17:14:04 -0400608
Tejun Heob16b1de2015-06-02 08:39:48 -0600609 wbc_init_bio(wbc, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 }
611
612 /*
613 * Must try to add the page before marking the buffer clean or
614 * the confused fail path above (OOM) will be very confused when
615 * it finds all bh marked clean (i.e. it will not write anything)
616 */
Tejun Heo2a814902015-05-28 14:50:51 -0400617 wbc_account_io(wbc, page, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 length = first_unmapped << blkbits;
619 if (bio_add_page(bio, page, length, 0) < length) {
620 bio = mpage_bio_submit(WRITE, bio);
621 goto alloc_new;
622 }
623
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700624 clean_buffers(page, first_unmapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
626 BUG_ON(PageWriteback(page));
627 set_page_writeback(page);
628 unlock_page(page);
629 if (boundary || (first_unmapped != blocks_per_page)) {
630 bio = mpage_bio_submit(WRITE, bio);
631 if (boundary_block) {
632 write_boundary_block(boundary_bdev,
633 boundary_block, 1 << blkbits);
634 }
635 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700636 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 }
638 goto out;
639
640confused:
641 if (bio)
642 bio = mpage_bio_submit(WRITE, bio);
643
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700644 if (mpd->use_writepage) {
645 ret = mapping->a_ops->writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700647 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 goto out;
649 }
650 /*
651 * The caller has a ref on the inode, so *mapping is stable
652 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700653 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700655 mpd->bio = bio;
656 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657}
658
659/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800660 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 * @mapping: address space structure to write
662 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
663 * @get_block: the filesystem's block mapper function.
664 * If this is NULL then use a_ops->writepage. Otherwise, go
665 * direct-to-BIO.
666 *
667 * This is a library function, which implements the writepages()
668 * address_space_operation.
669 *
670 * If a page is already under I/O, generic_writepages() skips it, even
671 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
672 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
673 * and msync() need to guarantee that all the data which was dirty at the time
674 * the call was made get new I/O started against them. If wbc->sync_mode is
675 * WB_SYNC_ALL then we were called for data integrity and we must wait for
676 * existing IO to complete.
677 */
678int
679mpage_writepages(struct address_space *mapping,
680 struct writeback_control *wbc, get_block_t get_block)
681{
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200682 struct blk_plug plug;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700683 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200685 blk_start_plug(&plug);
686
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700687 if (!get_block)
688 ret = generic_writepages(mapping, wbc);
689 else {
690 struct mpage_data mpd = {
691 .bio = NULL,
692 .last_block_in_bio = 0,
693 .get_block = get_block,
694 .use_writepage = 1,
695 };
696
697 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
698 if (mpd.bio)
699 mpage_bio_submit(WRITE, mpd.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 }
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200701 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 return ret;
703}
704EXPORT_SYMBOL(mpage_writepages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
706int mpage_writepage(struct page *page, get_block_t get_block,
707 struct writeback_control *wbc)
708{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700709 struct mpage_data mpd = {
710 .bio = NULL,
711 .last_block_in_bio = 0,
712 .get_block = get_block,
713 .use_writepage = 0,
714 };
715 int ret = __mpage_writepage(page, wbc, &mpd);
716 if (mpd.bio)
717 mpage_bio_submit(WRITE, mpd.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 return ret;
719}
720EXPORT_SYMBOL(mpage_writepage);