blob: 37b28280ad04065cebdebe73c309581fb256863b [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
Mike Christieeed25cd2016-06-05 14:31:59 -050059static struct bio *mpage_bio_submit(int op, int op_flags, 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;
Mike Christieeed25cd2016-06-05 14:31:59 -050062 bio_set_op_attrs(bio, op, op_flags);
63 guard_bio_eod(op, bio);
Mike Christie4e49ea42016-06-05 14:31:41 -050064 submit_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 return NULL;
66}
67
68static struct bio *
69mpage_alloc(struct block_device *bdev,
70 sector_t first_sector, int nr_vecs,
Al Virodd0fc662005-10-07 07:46:04 +010071 gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072{
73 struct bio *bio;
74
75 bio = bio_alloc(gfp_flags, nr_vecs);
76
77 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
78 while (!bio && (nr_vecs /= 2))
79 bio = bio_alloc(gfp_flags, nr_vecs);
80 }
81
82 if (bio) {
83 bio->bi_bdev = bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -070084 bio->bi_iter.bi_sector = first_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 }
86 return bio;
87}
88
89/*
90 * support function for mpage_readpages. The fs supplied get_block might
91 * return an up to date buffer. This is used to map that buffer into
92 * the page, which allows readpage to avoid triggering a duplicate call
93 * to get_block.
94 *
95 * The idea is to avoid adding buffers to pages that don't already have
96 * them. So when the buffer is up to date and the page size == block size,
97 * this marks the page up to date instead of adding new buffers.
98 */
99static void
100map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
101{
102 struct inode *inode = page->mapping->host;
103 struct buffer_head *page_bh, *head;
104 int block = 0;
105
106 if (!page_has_buffers(page)) {
107 /*
108 * don't make any buffers if there is only one buffer on
109 * the page and the page just needs to be set up to date
110 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300111 if (inode->i_blkbits == PAGE_SHIFT &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 buffer_uptodate(bh)) {
113 SetPageUptodate(page);
114 return;
115 }
116 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
117 }
118 head = page_buffers(page);
119 page_bh = head;
120 do {
121 if (block == page_block) {
122 page_bh->b_state = bh->b_state;
123 page_bh->b_bdev = bh->b_bdev;
124 page_bh->b_blocknr = bh->b_blocknr;
125 break;
126 }
127 page_bh = page_bh->b_this_page;
128 block++;
129 } while (page_bh != head);
130}
131
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800132/*
133 * This is the worker routine which does all the work of mapping the disk
134 * blocks and constructs largest possible bios, submits them for IO if the
135 * blocks are not contiguous on the disk.
136 *
137 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
138 * represent the validity of its disk mapping and to decide when to do the next
139 * get_block() call.
140 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141static struct bio *
142do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800143 sector_t *last_block_in_bio, struct buffer_head *map_bh,
Michal Hocko063d99b2015-10-15 15:28:24 -0700144 unsigned long *first_logical_block, get_block_t get_block,
145 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
147 struct inode *inode = page->mapping->host;
148 const unsigned blkbits = inode->i_blkbits;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300149 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 const unsigned blocksize = 1 << blkbits;
151 sector_t block_in_file;
152 sector_t last_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800153 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 sector_t blocks[MAX_BUF_PER_PAGE];
155 unsigned page_block;
156 unsigned first_hole = blocks_per_page;
157 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 int length;
159 int fully_mapped = 1;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800160 unsigned nblocks;
161 unsigned relative_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
163 if (page_has_buffers(page))
164 goto confused;
165
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300166 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800167 last_block = block_in_file + nr_pages * blocks_per_page;
168 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
169 if (last_block > last_block_in_file)
170 last_block = last_block_in_file;
171 page_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800173 /*
174 * Map blocks using the result from the previous get_blocks call first.
175 */
176 nblocks = map_bh->b_size >> blkbits;
177 if (buffer_mapped(map_bh) && block_in_file > *first_logical_block &&
178 block_in_file < (*first_logical_block + nblocks)) {
179 unsigned map_offset = block_in_file - *first_logical_block;
180 unsigned last = nblocks - map_offset;
181
182 for (relative_block = 0; ; relative_block++) {
183 if (relative_block == last) {
184 clear_buffer_mapped(map_bh);
185 break;
186 }
187 if (page_block == blocks_per_page)
188 break;
189 blocks[page_block] = map_bh->b_blocknr + map_offset +
190 relative_block;
191 page_block++;
192 block_in_file++;
193 }
194 bdev = map_bh->b_bdev;
195 }
196
197 /*
198 * Then do more get_blocks calls until we are done with this page.
199 */
200 map_bh->b_page = page;
201 while (page_block < blocks_per_page) {
202 map_bh->b_state = 0;
203 map_bh->b_size = 0;
204
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800206 map_bh->b_size = (last_block-block_in_file) << blkbits;
207 if (get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800209 *first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 }
211
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800212 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 fully_mapped = 0;
214 if (first_hole == blocks_per_page)
215 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800216 page_block++;
217 block_in_file++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 continue;
219 }
220
221 /* some filesystems will copy data into the page during
222 * the get_block call, in which case we don't want to
223 * read it again. map_buffer_to_page copies the data
224 * we just collected from get_block into the page's buffers
225 * so readpage doesn't have to repeat the get_block call
226 */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800227 if (buffer_uptodate(map_bh)) {
228 map_buffer_to_page(page, map_bh, page_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 goto confused;
230 }
231
232 if (first_hole != blocks_per_page)
233 goto confused; /* hole -> non-hole */
234
235 /* Contiguous blocks? */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800236 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800238 nblocks = map_bh->b_size >> blkbits;
239 for (relative_block = 0; ; relative_block++) {
240 if (relative_block == nblocks) {
241 clear_buffer_mapped(map_bh);
242 break;
243 } else if (page_block == blocks_per_page)
244 break;
245 blocks[page_block] = map_bh->b_blocknr+relative_block;
246 page_block++;
247 block_in_file++;
248 }
249 bdev = map_bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 }
251
252 if (first_hole != blocks_per_page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300253 zero_user_segment(page, first_hole << blkbits, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 if (first_hole == 0) {
255 SetPageUptodate(page);
256 unlock_page(page);
257 goto out;
258 }
259 } else if (fully_mapped) {
260 SetPageMappedToDisk(page);
261 }
262
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600263 if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
264 cleancache_get_page(page) == 0) {
265 SetPageUptodate(page);
266 goto confused;
267 }
268
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 /*
270 * This page will go to BIO. Do we need to send this BIO off first?
271 */
272 if (bio && (*last_block_in_bio != blocks[0] - 1))
Mike Christieeed25cd2016-06-05 14:31:59 -0500273 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274
275alloc_new:
276 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700277 if (first_hole == blocks_per_page) {
278 if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
279 page))
280 goto out;
281 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
Michal Hocko063d99b2015-10-15 15:28:24 -0700283 min_t(int, nr_pages, BIO_MAX_PAGES), gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 if (bio == NULL)
285 goto confused;
286 }
287
288 length = first_hole << blkbits;
289 if (bio_add_page(bio, page, length, 0) < length) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500290 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 goto alloc_new;
292 }
293
Miquel van Smoorenburg38c8e612009-01-06 14:39:02 -0800294 relative_block = block_in_file - *first_logical_block;
295 nblocks = map_bh->b_size >> blkbits;
296 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
297 (first_hole != blocks_per_page))
Mike Christieeed25cd2016-06-05 14:31:59 -0500298 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 else
300 *last_block_in_bio = blocks[blocks_per_page - 1];
301out:
302 return bio;
303
304confused:
305 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500306 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 if (!PageUptodate(page))
308 block_read_full_page(page, get_block);
309 else
310 unlock_page(page);
311 goto out;
312}
313
Martin Waitz67be2dd2005-05-01 08:59:26 -0700314/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800315 * mpage_readpages - populate an address space with some pages & start reads against them
Martin Waitz67be2dd2005-05-01 08:59:26 -0700316 * @mapping: the address_space
317 * @pages: The address of a list_head which contains the target pages. These
318 * pages have their ->index populated and are otherwise uninitialised.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700319 * The page at @pages->prev has the lowest file offset, and reads should be
320 * issued in @pages->prev to @pages->next order.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700321 * @nr_pages: The number of pages at *@pages
322 * @get_block: The filesystem's block mapper function.
323 *
324 * This function walks the pages and the blocks within each page, building and
325 * emitting large BIOs.
326 *
327 * If anything unusual happens, such as:
328 *
329 * - encountering a page which has buffers
330 * - encountering a page which has a non-hole after a hole
331 * - encountering a page with non-contiguous blocks
332 *
333 * then this code just gives up and calls the buffer_head-based read function.
334 * It does handle a page which has holes at the end - that is a common case:
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300335 * the end-of-file on blocksize < PAGE_SIZE setups.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700336 *
337 * BH_Boundary explanation:
338 *
339 * There is a problem. The mpage read code assembles several pages, gets all
340 * their disk mappings, and then submits them all. That's fine, but obtaining
341 * the disk mappings may require I/O. Reads of indirect blocks, for example.
342 *
343 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
344 * submitted in the following order:
345 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
Randy Dunlap78a4a502008-02-29 22:02:31 -0800346 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700347 * because the indirect block has to be read to get the mappings of blocks
348 * 13,14,15,16. Obviously, this impacts performance.
349 *
350 * So what we do it to allow the filesystem's get_block() function to set
351 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
352 * after this one will require I/O against a block which is probably close to
353 * this one. So you should push what I/O you have currently accumulated.
354 *
355 * This all causes the disk requests to be issued in the correct order.
356 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357int
358mpage_readpages(struct address_space *mapping, struct list_head *pages,
359 unsigned nr_pages, get_block_t get_block)
360{
361 struct bio *bio = NULL;
362 unsigned page_idx;
363 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800364 struct buffer_head map_bh;
365 unsigned long first_logical_block = 0;
Michal Hockoc62d2552015-11-06 16:28:49 -0800366 gfp_t gfp = mapping_gfp_constraint(mapping, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400368 map_bh.b_state = 0;
369 map_bh.b_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
Andrew Morton02c43632016-03-15 14:55:15 -0700371 struct page *page = lru_to_page(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
373 prefetchw(&page->flags);
374 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -0700375 if (!add_to_page_cache_lru(page, mapping,
Michal Hocko063d99b2015-10-15 15:28:24 -0700376 page->index,
377 gfp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 bio = do_mpage_readpage(bio, page,
379 nr_pages - page_idx,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800380 &last_block_in_bio, &map_bh,
381 &first_logical_block,
Michal Hocko063d99b2015-10-15 15:28:24 -0700382 get_block, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300384 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 BUG_ON(!list_empty(pages));
387 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500388 mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 return 0;
390}
391EXPORT_SYMBOL(mpage_readpages);
392
393/*
394 * This isn't called much at all
395 */
396int mpage_readpage(struct page *page, get_block_t get_block)
397{
398 struct bio *bio = NULL;
399 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800400 struct buffer_head map_bh;
401 unsigned long first_logical_block = 0;
Michal Hockoc62d2552015-11-06 16:28:49 -0800402 gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400404 map_bh.b_state = 0;
405 map_bh.b_size = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800406 bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
Michal Hocko063d99b2015-10-15 15:28:24 -0700407 &map_bh, &first_logical_block, get_block, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500409 mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 return 0;
411}
412EXPORT_SYMBOL(mpage_readpage);
413
414/*
415 * Writing is not so simple.
416 *
417 * If the page has buffers then they will be used for obtaining the disk
418 * mapping. We only support pages which are fully mapped-and-dirty, with a
419 * special case for pages which are unmapped at the end: end-of-file.
420 *
421 * If the page has no buffers (preferred) then the page is mapped here.
422 *
423 * If all blocks are found to be contiguous then the page can go into the
424 * BIO. Otherwise fall back to the mapping's writepage().
425 *
426 * FIXME: This code wants an estimate of how many pages are still to be
427 * written, so it can intelligently allocate a suitably-sized BIO. For now,
428 * just allocate full-size (16-page) BIOs.
429 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700430
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300431struct mpage_data {
432 struct bio *bio;
433 sector_t last_block_in_bio;
434 get_block_t *get_block;
435 unsigned use_writepage;
436};
437
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700438/*
439 * We have our BIO, so we can now mark the buffers clean. Make
440 * sure to only clean buffers which we know we'll be writing.
441 */
442static void clean_buffers(struct page *page, unsigned first_unmapped)
443{
444 unsigned buffer_counter = 0;
445 struct buffer_head *bh, *head;
446 if (!page_has_buffers(page))
447 return;
448 head = page_buffers(page);
449 bh = head;
450
451 do {
452 if (buffer_counter++ == first_unmapped)
453 break;
454 clear_buffer_dirty(bh);
455 bh = bh->b_this_page;
456 } while (bh != head);
457
458 /*
459 * we cannot drop the bh if the page is not uptodate or a concurrent
460 * readpage would fail to serialize with the bh and it would read from
461 * disk before we reach the platter.
462 */
463 if (buffer_heads_over_limit && PageUptodate(page))
464 try_to_free_buffers(page);
465}
466
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300467static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
Alex Tomas29a814d2008-07-11 19:27:31 -0400468 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700470 struct mpage_data *mpd = data;
471 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 struct address_space *mapping = page->mapping;
473 struct inode *inode = page->mapping->host;
474 const unsigned blkbits = inode->i_blkbits;
475 unsigned long end_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300476 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 sector_t last_block;
478 sector_t block_in_file;
479 sector_t blocks[MAX_BUF_PER_PAGE];
480 unsigned page_block;
481 unsigned first_unmapped = blocks_per_page;
482 struct block_device *bdev = NULL;
483 int boundary = 0;
484 sector_t boundary_block = 0;
485 struct block_device *boundary_bdev = NULL;
486 int length;
487 struct buffer_head map_bh;
488 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700489 int ret = 0;
Mike Christieeed25cd2016-06-05 14:31:59 -0500490 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
492 if (page_has_buffers(page)) {
493 struct buffer_head *head = page_buffers(page);
494 struct buffer_head *bh = head;
495
496 /* If they're all mapped and dirty, do it */
497 page_block = 0;
498 do {
499 BUG_ON(buffer_locked(bh));
500 if (!buffer_mapped(bh)) {
501 /*
502 * unmapped dirty buffers are created by
503 * __set_page_dirty_buffers -> mmapped data
504 */
505 if (buffer_dirty(bh))
506 goto confused;
507 if (first_unmapped == blocks_per_page)
508 first_unmapped = page_block;
509 continue;
510 }
511
512 if (first_unmapped != blocks_per_page)
513 goto confused; /* hole -> non-hole */
514
515 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
516 goto confused;
517 if (page_block) {
518 if (bh->b_blocknr != blocks[page_block-1] + 1)
519 goto confused;
520 }
521 blocks[page_block++] = bh->b_blocknr;
522 boundary = buffer_boundary(bh);
523 if (boundary) {
524 boundary_block = bh->b_blocknr;
525 boundary_bdev = bh->b_bdev;
526 }
527 bdev = bh->b_bdev;
528 } while ((bh = bh->b_this_page) != head);
529
530 if (first_unmapped)
531 goto page_is_mapped;
532
533 /*
534 * Page has buffers, but they are all unmapped. The page was
535 * created by pagein or read over a hole which was handled by
536 * block_read_full_page(). If this address_space is also
537 * using mpage_readpages then this can rarely happen.
538 */
539 goto confused;
540 }
541
542 /*
543 * The page has no buffers: map it to disk
544 */
545 BUG_ON(!PageUptodate(page));
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300546 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 last_block = (i_size - 1) >> blkbits;
548 map_bh.b_page = page;
549 for (page_block = 0; page_block < blocks_per_page; ) {
550
551 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800552 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700553 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 goto confused;
555 if (buffer_new(&map_bh))
556 unmap_underlying_metadata(map_bh.b_bdev,
557 map_bh.b_blocknr);
558 if (buffer_boundary(&map_bh)) {
559 boundary_block = map_bh.b_blocknr;
560 boundary_bdev = map_bh.b_bdev;
561 }
562 if (page_block) {
563 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
564 goto confused;
565 }
566 blocks[page_block++] = map_bh.b_blocknr;
567 boundary = buffer_boundary(&map_bh);
568 bdev = map_bh.b_bdev;
569 if (block_in_file == last_block)
570 break;
571 block_in_file++;
572 }
573 BUG_ON(page_block == 0);
574
575 first_unmapped = page_block;
576
577page_is_mapped:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300578 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 if (page->index >= end_index) {
580 /*
581 * The page straddles i_size. It must be zeroed out on each
Adam Buchbinder2a61aa42009-12-11 16:35:40 -0500582 * and every writepage invocation because it may be mmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 * "A file is mapped in multiples of the page size. For a file
584 * that is not a multiple of the page size, the remaining memory
585 * is zeroed when mapped, and writes to that region are not
586 * written out to the file."
587 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300588 unsigned offset = i_size & (PAGE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 if (page->index > end_index || !offset)
591 goto confused;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300592 zero_user_segment(page, offset, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 }
594
595 /*
596 * This page will go to BIO. Do we need to send this BIO off first?
597 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700598 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Mike Christieeed25cd2016-06-05 14:31:59 -0500599 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
601alloc_new:
602 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700603 if (first_unmapped == blocks_per_page) {
604 if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
605 page, wbc)) {
606 clean_buffers(page, first_unmapped);
607 goto out;
608 }
609 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
Kent Overstreetb54ffb72015-05-19 14:31:01 +0200611 BIO_MAX_PAGES, GFP_NOFS|__GFP_HIGH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 if (bio == NULL)
613 goto confused;
Tejun Heo429b3fb2015-05-22 17:14:04 -0400614
Tejun Heob16b1de2015-06-02 08:39:48 -0600615 wbc_init_bio(wbc, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 }
617
618 /*
619 * Must try to add the page before marking the buffer clean or
620 * the confused fail path above (OOM) will be very confused when
621 * it finds all bh marked clean (i.e. it will not write anything)
622 */
Tejun Heo2a814902015-05-28 14:50:51 -0400623 wbc_account_io(wbc, page, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 length = first_unmapped << blkbits;
625 if (bio_add_page(bio, page, length, 0) < length) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500626 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 goto alloc_new;
628 }
629
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700630 clean_buffers(page, first_unmapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
632 BUG_ON(PageWriteback(page));
633 set_page_writeback(page);
634 unlock_page(page);
635 if (boundary || (first_unmapped != blocks_per_page)) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500636 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 if (boundary_block) {
638 write_boundary_block(boundary_bdev,
639 boundary_block, 1 << blkbits);
640 }
641 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700642 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 }
644 goto out;
645
646confused:
647 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500648 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700650 if (mpd->use_writepage) {
651 ret = mapping->a_ops->writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700653 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 goto out;
655 }
656 /*
657 * The caller has a ref on the inode, so *mapping is stable
658 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700659 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700661 mpd->bio = bio;
662 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
665/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800666 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 * @mapping: address space structure to write
668 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
669 * @get_block: the filesystem's block mapper function.
670 * If this is NULL then use a_ops->writepage. Otherwise, go
671 * direct-to-BIO.
672 *
673 * This is a library function, which implements the writepages()
674 * address_space_operation.
675 *
676 * If a page is already under I/O, generic_writepages() skips it, even
677 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
678 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
679 * and msync() need to guarantee that all the data which was dirty at the time
680 * the call was made get new I/O started against them. If wbc->sync_mode is
681 * WB_SYNC_ALL then we were called for data integrity and we must wait for
682 * existing IO to complete.
683 */
684int
685mpage_writepages(struct address_space *mapping,
686 struct writeback_control *wbc, get_block_t get_block)
687{
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200688 struct blk_plug plug;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700689 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200691 blk_start_plug(&plug);
692
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700693 if (!get_block)
694 ret = generic_writepages(mapping, wbc);
695 else {
696 struct mpage_data mpd = {
697 .bio = NULL,
698 .last_block_in_bio = 0,
699 .get_block = get_block,
700 .use_writepage = 1,
701 };
702
703 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
Roman Pen5948edb2015-09-15 08:27:25 -0600704 if (mpd.bio) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500705 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
706 WRITE_SYNC : 0);
707 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
Roman Pen5948edb2015-09-15 08:27:25 -0600708 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 }
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200710 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 return ret;
712}
713EXPORT_SYMBOL(mpage_writepages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
715int mpage_writepage(struct page *page, get_block_t get_block,
716 struct writeback_control *wbc)
717{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700718 struct mpage_data mpd = {
719 .bio = NULL,
720 .last_block_in_bio = 0,
721 .get_block = get_block,
722 .use_writepage = 0,
723 };
724 int ret = __mpage_writepage(page, wbc, &mpd);
Roman Pen5948edb2015-09-15 08:27:25 -0600725 if (mpd.bio) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500726 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
727 WRITE_SYNC : 0);
728 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
Roman Pen5948edb2015-09-15 08:27:25 -0600729 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 return ret;
731}
732EXPORT_SYMBOL(mpage_writepage);