blob: d2413af0823a24202560eb73ee363dcb980c939c [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;
Jens Axboec11f0c02016-08-05 08:11:04 -060053 page_endio(page, op_is_write(bio_op(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
Michal Hocko8a5c7432016-07-26 15:24:53 -070075 /* Restrict the given (page cache) mask for slab allocations */
76 gfp_flags &= GFP_KERNEL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 bio = bio_alloc(gfp_flags, nr_vecs);
78
79 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
80 while (!bio && (nr_vecs /= 2))
81 bio = bio_alloc(gfp_flags, nr_vecs);
82 }
83
84 if (bio) {
85 bio->bi_bdev = bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -070086 bio->bi_iter.bi_sector = first_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 }
88 return bio;
89}
90
91/*
92 * support function for mpage_readpages. The fs supplied get_block might
93 * return an up to date buffer. This is used to map that buffer into
94 * the page, which allows readpage to avoid triggering a duplicate call
95 * to get_block.
96 *
97 * The idea is to avoid adding buffers to pages that don't already have
98 * them. So when the buffer is up to date and the page size == block size,
99 * this marks the page up to date instead of adding new buffers.
100 */
101static void
102map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
103{
104 struct inode *inode = page->mapping->host;
105 struct buffer_head *page_bh, *head;
106 int block = 0;
107
108 if (!page_has_buffers(page)) {
109 /*
110 * don't make any buffers if there is only one buffer on
111 * the page and the page just needs to be set up to date
112 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300113 if (inode->i_blkbits == PAGE_SHIFT &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 buffer_uptodate(bh)) {
115 SetPageUptodate(page);
116 return;
117 }
118 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
119 }
120 head = page_buffers(page);
121 page_bh = head;
122 do {
123 if (block == page_block) {
124 page_bh->b_state = bh->b_state;
125 page_bh->b_bdev = bh->b_bdev;
126 page_bh->b_blocknr = bh->b_blocknr;
127 break;
128 }
129 page_bh = page_bh->b_this_page;
130 block++;
131 } while (page_bh != head);
132}
133
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800134/*
135 * This is the worker routine which does all the work of mapping the disk
136 * blocks and constructs largest possible bios, submits them for IO if the
137 * blocks are not contiguous on the disk.
138 *
139 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
140 * represent the validity of its disk mapping and to decide when to do the next
141 * get_block() call.
142 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static struct bio *
144do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800145 sector_t *last_block_in_bio, struct buffer_head *map_bh,
Michal Hocko063d99b2015-10-15 15:28:24 -0700146 unsigned long *first_logical_block, get_block_t get_block,
147 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148{
149 struct inode *inode = page->mapping->host;
150 const unsigned blkbits = inode->i_blkbits;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300151 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 const unsigned blocksize = 1 << blkbits;
153 sector_t block_in_file;
154 sector_t last_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800155 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 sector_t blocks[MAX_BUF_PER_PAGE];
157 unsigned page_block;
158 unsigned first_hole = blocks_per_page;
159 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 int length;
161 int fully_mapped = 1;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800162 unsigned nblocks;
163 unsigned relative_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165 if (page_has_buffers(page))
166 goto confused;
167
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300168 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800169 last_block = block_in_file + nr_pages * blocks_per_page;
170 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
171 if (last_block > last_block_in_file)
172 last_block = last_block_in_file;
173 page_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800175 /*
176 * Map blocks using the result from the previous get_blocks call first.
177 */
178 nblocks = map_bh->b_size >> blkbits;
179 if (buffer_mapped(map_bh) && block_in_file > *first_logical_block &&
180 block_in_file < (*first_logical_block + nblocks)) {
181 unsigned map_offset = block_in_file - *first_logical_block;
182 unsigned last = nblocks - map_offset;
183
184 for (relative_block = 0; ; relative_block++) {
185 if (relative_block == last) {
186 clear_buffer_mapped(map_bh);
187 break;
188 }
189 if (page_block == blocks_per_page)
190 break;
191 blocks[page_block] = map_bh->b_blocknr + map_offset +
192 relative_block;
193 page_block++;
194 block_in_file++;
195 }
196 bdev = map_bh->b_bdev;
197 }
198
199 /*
200 * Then do more get_blocks calls until we are done with this page.
201 */
202 map_bh->b_page = page;
203 while (page_block < blocks_per_page) {
204 map_bh->b_state = 0;
205 map_bh->b_size = 0;
206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800208 map_bh->b_size = (last_block-block_in_file) << blkbits;
209 if (get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800211 *first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 }
213
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800214 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 fully_mapped = 0;
216 if (first_hole == blocks_per_page)
217 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800218 page_block++;
219 block_in_file++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 continue;
221 }
222
223 /* some filesystems will copy data into the page during
224 * the get_block call, in which case we don't want to
225 * read it again. map_buffer_to_page copies the data
226 * we just collected from get_block into the page's buffers
227 * so readpage doesn't have to repeat the get_block call
228 */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800229 if (buffer_uptodate(map_bh)) {
230 map_buffer_to_page(page, map_bh, page_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 goto confused;
232 }
233
234 if (first_hole != blocks_per_page)
235 goto confused; /* hole -> non-hole */
236
237 /* Contiguous blocks? */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800238 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800240 nblocks = map_bh->b_size >> blkbits;
241 for (relative_block = 0; ; relative_block++) {
242 if (relative_block == nblocks) {
243 clear_buffer_mapped(map_bh);
244 break;
245 } else if (page_block == blocks_per_page)
246 break;
247 blocks[page_block] = map_bh->b_blocknr+relative_block;
248 page_block++;
249 block_in_file++;
250 }
251 bdev = map_bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 }
253
254 if (first_hole != blocks_per_page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300255 zero_user_segment(page, first_hole << blkbits, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 if (first_hole == 0) {
257 SetPageUptodate(page);
258 unlock_page(page);
259 goto out;
260 }
261 } else if (fully_mapped) {
262 SetPageMappedToDisk(page);
263 }
264
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600265 if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
266 cleancache_get_page(page) == 0) {
267 SetPageUptodate(page);
268 goto confused;
269 }
270
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 /*
272 * This page will go to BIO. Do we need to send this BIO off first?
273 */
274 if (bio && (*last_block_in_bio != blocks[0] - 1))
Mike Christieeed25cd2016-06-05 14:31:59 -0500275 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
277alloc_new:
278 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700279 if (first_hole == blocks_per_page) {
280 if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
281 page))
282 goto out;
283 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
Michal Hocko063d99b2015-10-15 15:28:24 -0700285 min_t(int, nr_pages, BIO_MAX_PAGES), gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 if (bio == NULL)
287 goto confused;
288 }
289
290 length = first_hole << blkbits;
291 if (bio_add_page(bio, page, length, 0) < length) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500292 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 goto alloc_new;
294 }
295
Miquel van Smoorenburg38c8e612009-01-06 14:39:02 -0800296 relative_block = block_in_file - *first_logical_block;
297 nblocks = map_bh->b_size >> blkbits;
298 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
299 (first_hole != blocks_per_page))
Mike Christieeed25cd2016-06-05 14:31:59 -0500300 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 else
302 *last_block_in_bio = blocks[blocks_per_page - 1];
303out:
304 return bio;
305
306confused:
307 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500308 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 if (!PageUptodate(page))
310 block_read_full_page(page, get_block);
311 else
312 unlock_page(page);
313 goto out;
314}
315
Martin Waitz67be2dd2005-05-01 08:59:26 -0700316/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800317 * mpage_readpages - populate an address space with some pages & start reads against them
Martin Waitz67be2dd2005-05-01 08:59:26 -0700318 * @mapping: the address_space
319 * @pages: The address of a list_head which contains the target pages. These
320 * pages have their ->index populated and are otherwise uninitialised.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700321 * The page at @pages->prev has the lowest file offset, and reads should be
322 * issued in @pages->prev to @pages->next order.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700323 * @nr_pages: The number of pages at *@pages
324 * @get_block: The filesystem's block mapper function.
325 *
326 * This function walks the pages and the blocks within each page, building and
327 * emitting large BIOs.
328 *
329 * If anything unusual happens, such as:
330 *
331 * - encountering a page which has buffers
332 * - encountering a page which has a non-hole after a hole
333 * - encountering a page with non-contiguous blocks
334 *
335 * then this code just gives up and calls the buffer_head-based read function.
336 * It does handle a page which has holes at the end - that is a common case:
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300337 * the end-of-file on blocksize < PAGE_SIZE setups.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700338 *
339 * BH_Boundary explanation:
340 *
341 * There is a problem. The mpage read code assembles several pages, gets all
342 * their disk mappings, and then submits them all. That's fine, but obtaining
343 * the disk mappings may require I/O. Reads of indirect blocks, for example.
344 *
345 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
346 * submitted in the following order:
347 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
Randy Dunlap78a4a502008-02-29 22:02:31 -0800348 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700349 * because the indirect block has to be read to get the mappings of blocks
350 * 13,14,15,16. Obviously, this impacts performance.
351 *
352 * So what we do it to allow the filesystem's get_block() function to set
353 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
354 * after this one will require I/O against a block which is probably close to
355 * this one. So you should push what I/O you have currently accumulated.
356 *
357 * This all causes the disk requests to be issued in the correct order.
358 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359int
360mpage_readpages(struct address_space *mapping, struct list_head *pages,
361 unsigned nr_pages, get_block_t get_block)
362{
363 struct bio *bio = NULL;
364 unsigned page_idx;
365 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800366 struct buffer_head map_bh;
367 unsigned long first_logical_block = 0;
Michal Hocko8a5c7432016-07-26 15:24:53 -0700368 gfp_t gfp = readahead_gfp_mask(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400370 map_bh.b_state = 0;
371 map_bh.b_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
Andrew Morton02c43632016-03-15 14:55:15 -0700373 struct page *page = lru_to_page(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
375 prefetchw(&page->flags);
376 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -0700377 if (!add_to_page_cache_lru(page, mapping,
Michal Hocko063d99b2015-10-15 15:28:24 -0700378 page->index,
379 gfp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 bio = do_mpage_readpage(bio, page,
381 nr_pages - page_idx,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800382 &last_block_in_bio, &map_bh,
383 &first_logical_block,
Michal Hocko063d99b2015-10-15 15:28:24 -0700384 get_block, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300386 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 BUG_ON(!list_empty(pages));
389 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500390 mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 return 0;
392}
393EXPORT_SYMBOL(mpage_readpages);
394
395/*
396 * This isn't called much at all
397 */
398int mpage_readpage(struct page *page, get_block_t get_block)
399{
400 struct bio *bio = NULL;
401 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800402 struct buffer_head map_bh;
403 unsigned long first_logical_block = 0;
Michal Hockoc62d2552015-11-06 16:28:49 -0800404 gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400406 map_bh.b_state = 0;
407 map_bh.b_size = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800408 bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
Michal Hocko063d99b2015-10-15 15:28:24 -0700409 &map_bh, &first_logical_block, get_block, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500411 mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 return 0;
413}
414EXPORT_SYMBOL(mpage_readpage);
415
416/*
417 * Writing is not so simple.
418 *
419 * If the page has buffers then they will be used for obtaining the disk
420 * mapping. We only support pages which are fully mapped-and-dirty, with a
421 * special case for pages which are unmapped at the end: end-of-file.
422 *
423 * If the page has no buffers (preferred) then the page is mapped here.
424 *
425 * If all blocks are found to be contiguous then the page can go into the
426 * BIO. Otherwise fall back to the mapping's writepage().
427 *
428 * FIXME: This code wants an estimate of how many pages are still to be
429 * written, so it can intelligently allocate a suitably-sized BIO. For now,
430 * just allocate full-size (16-page) BIOs.
431 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700432
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300433struct mpage_data {
434 struct bio *bio;
435 sector_t last_block_in_bio;
436 get_block_t *get_block;
437 unsigned use_writepage;
438};
439
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700440/*
441 * We have our BIO, so we can now mark the buffers clean. Make
442 * sure to only clean buffers which we know we'll be writing.
443 */
444static void clean_buffers(struct page *page, unsigned first_unmapped)
445{
446 unsigned buffer_counter = 0;
447 struct buffer_head *bh, *head;
448 if (!page_has_buffers(page))
449 return;
450 head = page_buffers(page);
451 bh = head;
452
453 do {
454 if (buffer_counter++ == first_unmapped)
455 break;
456 clear_buffer_dirty(bh);
457 bh = bh->b_this_page;
458 } while (bh != head);
459
460 /*
461 * we cannot drop the bh if the page is not uptodate or a concurrent
462 * readpage would fail to serialize with the bh and it would read from
463 * disk before we reach the platter.
464 */
465 if (buffer_heads_over_limit && PageUptodate(page))
466 try_to_free_buffers(page);
467}
468
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300469static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
Alex Tomas29a814d2008-07-11 19:27:31 -0400470 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700472 struct mpage_data *mpd = data;
473 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 struct address_space *mapping = page->mapping;
475 struct inode *inode = page->mapping->host;
476 const unsigned blkbits = inode->i_blkbits;
477 unsigned long end_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300478 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 sector_t last_block;
480 sector_t block_in_file;
481 sector_t blocks[MAX_BUF_PER_PAGE];
482 unsigned page_block;
483 unsigned first_unmapped = blocks_per_page;
484 struct block_device *bdev = NULL;
485 int boundary = 0;
486 sector_t boundary_block = 0;
487 struct block_device *boundary_bdev = NULL;
488 int length;
489 struct buffer_head map_bh;
490 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700491 int ret = 0;
Mike Christieeed25cd2016-06-05 14:31:59 -0500492 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
494 if (page_has_buffers(page)) {
495 struct buffer_head *head = page_buffers(page);
496 struct buffer_head *bh = head;
497
498 /* If they're all mapped and dirty, do it */
499 page_block = 0;
500 do {
501 BUG_ON(buffer_locked(bh));
502 if (!buffer_mapped(bh)) {
503 /*
504 * unmapped dirty buffers are created by
505 * __set_page_dirty_buffers -> mmapped data
506 */
507 if (buffer_dirty(bh))
508 goto confused;
509 if (first_unmapped == blocks_per_page)
510 first_unmapped = page_block;
511 continue;
512 }
513
514 if (first_unmapped != blocks_per_page)
515 goto confused; /* hole -> non-hole */
516
517 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
518 goto confused;
519 if (page_block) {
520 if (bh->b_blocknr != blocks[page_block-1] + 1)
521 goto confused;
522 }
523 blocks[page_block++] = bh->b_blocknr;
524 boundary = buffer_boundary(bh);
525 if (boundary) {
526 boundary_block = bh->b_blocknr;
527 boundary_bdev = bh->b_bdev;
528 }
529 bdev = bh->b_bdev;
530 } while ((bh = bh->b_this_page) != head);
531
532 if (first_unmapped)
533 goto page_is_mapped;
534
535 /*
536 * Page has buffers, but they are all unmapped. The page was
537 * created by pagein or read over a hole which was handled by
538 * block_read_full_page(). If this address_space is also
539 * using mpage_readpages then this can rarely happen.
540 */
541 goto confused;
542 }
543
544 /*
545 * The page has no buffers: map it to disk
546 */
547 BUG_ON(!PageUptodate(page));
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300548 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 last_block = (i_size - 1) >> blkbits;
550 map_bh.b_page = page;
551 for (page_block = 0; page_block < blocks_per_page; ) {
552
553 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800554 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700555 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 goto confused;
557 if (buffer_new(&map_bh))
558 unmap_underlying_metadata(map_bh.b_bdev,
559 map_bh.b_blocknr);
560 if (buffer_boundary(&map_bh)) {
561 boundary_block = map_bh.b_blocknr;
562 boundary_bdev = map_bh.b_bdev;
563 }
564 if (page_block) {
565 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
566 goto confused;
567 }
568 blocks[page_block++] = map_bh.b_blocknr;
569 boundary = buffer_boundary(&map_bh);
570 bdev = map_bh.b_bdev;
571 if (block_in_file == last_block)
572 break;
573 block_in_file++;
574 }
575 BUG_ON(page_block == 0);
576
577 first_unmapped = page_block;
578
579page_is_mapped:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300580 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 if (page->index >= end_index) {
582 /*
583 * The page straddles i_size. It must be zeroed out on each
Adam Buchbinder2a61aa42009-12-11 16:35:40 -0500584 * and every writepage invocation because it may be mmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 * "A file is mapped in multiples of the page size. For a file
586 * that is not a multiple of the page size, the remaining memory
587 * is zeroed when mapped, and writes to that region are not
588 * written out to the file."
589 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300590 unsigned offset = i_size & (PAGE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 if (page->index > end_index || !offset)
593 goto confused;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300594 zero_user_segment(page, offset, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 }
596
597 /*
598 * This page will go to BIO. Do we need to send this BIO off first?
599 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700600 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Mike Christieeed25cd2016-06-05 14:31:59 -0500601 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
603alloc_new:
604 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700605 if (first_unmapped == blocks_per_page) {
606 if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
607 page, wbc)) {
608 clean_buffers(page, first_unmapped);
609 goto out;
610 }
611 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
Kent Overstreetb54ffb72015-05-19 14:31:01 +0200613 BIO_MAX_PAGES, GFP_NOFS|__GFP_HIGH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 if (bio == NULL)
615 goto confused;
Tejun Heo429b3fb2015-05-22 17:14:04 -0400616
Tejun Heob16b1de2015-06-02 08:39:48 -0600617 wbc_init_bio(wbc, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 }
619
620 /*
621 * Must try to add the page before marking the buffer clean or
622 * the confused fail path above (OOM) will be very confused when
623 * it finds all bh marked clean (i.e. it will not write anything)
624 */
Tejun Heo2a814902015-05-28 14:50:51 -0400625 wbc_account_io(wbc, page, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 length = first_unmapped << blkbits;
627 if (bio_add_page(bio, page, length, 0) < length) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500628 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 goto alloc_new;
630 }
631
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700632 clean_buffers(page, first_unmapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
634 BUG_ON(PageWriteback(page));
635 set_page_writeback(page);
636 unlock_page(page);
637 if (boundary || (first_unmapped != blocks_per_page)) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500638 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 if (boundary_block) {
640 write_boundary_block(boundary_bdev,
641 boundary_block, 1 << blkbits);
642 }
643 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700644 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 }
646 goto out;
647
648confused:
649 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500650 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700652 if (mpd->use_writepage) {
653 ret = mapping->a_ops->writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700655 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 goto out;
657 }
658 /*
659 * The caller has a ref on the inode, so *mapping is stable
660 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700661 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700663 mpd->bio = bio;
664 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665}
666
667/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800668 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 * @mapping: address space structure to write
670 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
671 * @get_block: the filesystem's block mapper function.
672 * If this is NULL then use a_ops->writepage. Otherwise, go
673 * direct-to-BIO.
674 *
675 * This is a library function, which implements the writepages()
676 * address_space_operation.
677 *
678 * If a page is already under I/O, generic_writepages() skips it, even
679 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
680 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
681 * and msync() need to guarantee that all the data which was dirty at the time
682 * the call was made get new I/O started against them. If wbc->sync_mode is
683 * WB_SYNC_ALL then we were called for data integrity and we must wait for
684 * existing IO to complete.
685 */
686int
687mpage_writepages(struct address_space *mapping,
688 struct writeback_control *wbc, get_block_t get_block)
689{
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200690 struct blk_plug plug;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700691 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200693 blk_start_plug(&plug);
694
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700695 if (!get_block)
696 ret = generic_writepages(mapping, wbc);
697 else {
698 struct mpage_data mpd = {
699 .bio = NULL,
700 .last_block_in_bio = 0,
701 .get_block = get_block,
702 .use_writepage = 1,
703 };
704
705 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
Roman Pen5948edb2015-09-15 08:27:25 -0600706 if (mpd.bio) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500707 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
708 WRITE_SYNC : 0);
709 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
Roman Pen5948edb2015-09-15 08:27:25 -0600710 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 }
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200712 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 return ret;
714}
715EXPORT_SYMBOL(mpage_writepages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
717int mpage_writepage(struct page *page, get_block_t get_block,
718 struct writeback_control *wbc)
719{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700720 struct mpage_data mpd = {
721 .bio = NULL,
722 .last_block_in_bio = 0,
723 .get_block = get_block,
724 .use_writepage = 0,
725 };
726 int ret = __mpage_writepage(page, wbc, &mpd);
Roman Pen5948edb2015-09-15 08:27:25 -0600727 if (mpd.bio) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500728 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
729 WRITE_SYNC : 0);
730 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
Roman Pen5948edb2015-09-15 08:27:25 -0600731 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 return ret;
733}
734EXPORT_SYMBOL(mpage_writepage);