blob: d4e17c88ce08679645e38125f85fdb16b7de72b2 [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
Mohan Srinivasan25cc70f2016-12-14 16:39:51 -080034#define CREATE_TRACE_POINTS
35#include <trace/events/android_fs.h>
36
37EXPORT_TRACEPOINT_SYMBOL(android_fs_datawrite_start);
38EXPORT_TRACEPOINT_SYMBOL(android_fs_datawrite_end);
39EXPORT_TRACEPOINT_SYMBOL(android_fs_dataread_start);
40EXPORT_TRACEPOINT_SYMBOL(android_fs_dataread_end);
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/*
43 * I/O completion handler for multipage BIOs.
44 *
45 * The mpage code never puts partial pages into a BIO (except for end-of-file).
46 * If a page does not map to a contiguous run of blocks then it simply falls
47 * back to block_read_full_page().
48 *
49 * Why is this? If a page's completion depends on a number of different BIOs
50 * which can complete in any order (or at the same time) then determining the
51 * status of that page is hard. See end_buffer_async_read() for the details.
52 * There is no point in duplicating all that complexity.
53 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020054static void mpage_end_io(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070055{
Kent Overstreet2c30c712013-11-07 12:20:26 -080056 struct bio_vec *bv;
57 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Mohan Srinivasan25cc70f2016-12-14 16:39:51 -080059 if (trace_android_fs_dataread_end_enabled() &&
60 (bio_data_dir(bio) == READ)) {
61 struct page *first_page = bio->bi_io_vec[0].bv_page;
62
63 if (first_page != NULL)
64 trace_android_fs_dataread_end(first_page->mapping->host,
65 page_offset(first_page),
66 bio->bi_iter.bi_size);
67 }
68
Kent Overstreet2c30c712013-11-07 12:20:26 -080069 bio_for_each_segment_all(bv, bio, i) {
70 struct page *page = bv->bv_page;
Jens Axboec11f0c02016-08-05 08:11:04 -060071 page_endio(page, op_is_write(bio_op(bio)), bio->bi_error);
Kent Overstreet2c30c712013-11-07 12:20:26 -080072 }
73
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075}
76
Mike Christieeed25cd2016-06-05 14:31:59 -050077static struct bio *mpage_bio_submit(int op, int op_flags, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070078{
Mohan Srinivasan25cc70f2016-12-14 16:39:51 -080079 if (trace_android_fs_dataread_start_enabled() && (op == REQ_OP_READ)) {
80 struct page *first_page = bio->bi_io_vec[0].bv_page;
81
82 if (first_page != NULL) {
Mohan Srinivasan009e6082017-02-10 14:26:23 -080083 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
84
85 path = android_fstrace_get_pathname(pathbuf,
86 MAX_TRACE_PATHBUF_LEN,
87 first_page->mapping->host);
Mohan Srinivasan25cc70f2016-12-14 16:39:51 -080088 trace_android_fs_dataread_start(
89 first_page->mapping->host,
90 page_offset(first_page),
91 bio->bi_iter.bi_size,
92 current->pid,
Mohan Srinivasan009e6082017-02-10 14:26:23 -080093 path,
Mohan Srinivasan25cc70f2016-12-14 16:39:51 -080094 current->comm);
95 }
96 }
Hai Shanc32b0d42011-01-13 15:45:51 -080097 bio->bi_end_io = mpage_end_io;
Mike Christieeed25cd2016-06-05 14:31:59 -050098 bio_set_op_attrs(bio, op, op_flags);
99 guard_bio_eod(op, bio);
Mike Christie4e49ea42016-06-05 14:31:41 -0500100 submit_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 return NULL;
102}
103
104static struct bio *
105mpage_alloc(struct block_device *bdev,
106 sector_t first_sector, int nr_vecs,
Al Virodd0fc662005-10-07 07:46:04 +0100107 gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108{
109 struct bio *bio;
110
Michal Hocko8a5c7432016-07-26 15:24:53 -0700111 /* Restrict the given (page cache) mask for slab allocations */
112 gfp_flags &= GFP_KERNEL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 bio = bio_alloc(gfp_flags, nr_vecs);
114
115 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
116 while (!bio && (nr_vecs /= 2))
117 bio = bio_alloc(gfp_flags, nr_vecs);
118 }
119
120 if (bio) {
121 bio->bi_bdev = bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700122 bio->bi_iter.bi_sector = first_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 }
124 return bio;
125}
126
127/*
128 * support function for mpage_readpages. The fs supplied get_block might
129 * return an up to date buffer. This is used to map that buffer into
130 * the page, which allows readpage to avoid triggering a duplicate call
131 * to get_block.
132 *
133 * The idea is to avoid adding buffers to pages that don't already have
134 * them. So when the buffer is up to date and the page size == block size,
135 * this marks the page up to date instead of adding new buffers.
136 */
137static void
138map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
139{
140 struct inode *inode = page->mapping->host;
141 struct buffer_head *page_bh, *head;
142 int block = 0;
143
144 if (!page_has_buffers(page)) {
145 /*
146 * don't make any buffers if there is only one buffer on
147 * the page and the page just needs to be set up to date
148 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300149 if (inode->i_blkbits == PAGE_SHIFT &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 buffer_uptodate(bh)) {
151 SetPageUptodate(page);
152 return;
153 }
Fabian Frederick61604a22017-02-27 14:28:32 -0800154 create_empty_buffers(page, i_blocksize(inode), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 }
156 head = page_buffers(page);
157 page_bh = head;
158 do {
159 if (block == page_block) {
160 page_bh->b_state = bh->b_state;
161 page_bh->b_bdev = bh->b_bdev;
162 page_bh->b_blocknr = bh->b_blocknr;
163 break;
164 }
165 page_bh = page_bh->b_this_page;
166 block++;
167 } while (page_bh != head);
168}
169
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800170/*
171 * This is the worker routine which does all the work of mapping the disk
172 * blocks and constructs largest possible bios, submits them for IO if the
173 * blocks are not contiguous on the disk.
174 *
175 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
176 * represent the validity of its disk mapping and to decide when to do the next
177 * get_block() call.
178 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179static struct bio *
180do_mpage_readpage(struct bio *bio, struct page *page, unsigned nr_pages,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800181 sector_t *last_block_in_bio, struct buffer_head *map_bh,
Michal Hocko063d99b2015-10-15 15:28:24 -0700182 unsigned long *first_logical_block, get_block_t get_block,
183 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
185 struct inode *inode = page->mapping->host;
186 const unsigned blkbits = inode->i_blkbits;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300187 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 const unsigned blocksize = 1 << blkbits;
189 sector_t block_in_file;
190 sector_t last_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800191 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 sector_t blocks[MAX_BUF_PER_PAGE];
193 unsigned page_block;
194 unsigned first_hole = blocks_per_page;
195 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 int length;
197 int fully_mapped = 1;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800198 unsigned nblocks;
199 unsigned relative_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
201 if (page_has_buffers(page))
202 goto confused;
203
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300204 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800205 last_block = block_in_file + nr_pages * blocks_per_page;
206 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
207 if (last_block > last_block_in_file)
208 last_block = last_block_in_file;
209 page_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800211 /*
212 * Map blocks using the result from the previous get_blocks call first.
213 */
214 nblocks = map_bh->b_size >> blkbits;
215 if (buffer_mapped(map_bh) && block_in_file > *first_logical_block &&
216 block_in_file < (*first_logical_block + nblocks)) {
217 unsigned map_offset = block_in_file - *first_logical_block;
218 unsigned last = nblocks - map_offset;
219
220 for (relative_block = 0; ; relative_block++) {
221 if (relative_block == last) {
222 clear_buffer_mapped(map_bh);
223 break;
224 }
225 if (page_block == blocks_per_page)
226 break;
227 blocks[page_block] = map_bh->b_blocknr + map_offset +
228 relative_block;
229 page_block++;
230 block_in_file++;
231 }
232 bdev = map_bh->b_bdev;
233 }
234
235 /*
236 * Then do more get_blocks calls until we are done with this page.
237 */
238 map_bh->b_page = page;
239 while (page_block < blocks_per_page) {
240 map_bh->b_state = 0;
241 map_bh->b_size = 0;
242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800244 map_bh->b_size = (last_block-block_in_file) << blkbits;
245 if (get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800247 *first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 }
249
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800250 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 fully_mapped = 0;
252 if (first_hole == blocks_per_page)
253 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800254 page_block++;
255 block_in_file++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 continue;
257 }
258
259 /* some filesystems will copy data into the page during
260 * the get_block call, in which case we don't want to
261 * read it again. map_buffer_to_page copies the data
262 * we just collected from get_block into the page's buffers
263 * so readpage doesn't have to repeat the get_block call
264 */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800265 if (buffer_uptodate(map_bh)) {
266 map_buffer_to_page(page, map_bh, page_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 goto confused;
268 }
269
270 if (first_hole != blocks_per_page)
271 goto confused; /* hole -> non-hole */
272
273 /* Contiguous blocks? */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800274 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800276 nblocks = map_bh->b_size >> blkbits;
277 for (relative_block = 0; ; relative_block++) {
278 if (relative_block == nblocks) {
279 clear_buffer_mapped(map_bh);
280 break;
281 } else if (page_block == blocks_per_page)
282 break;
283 blocks[page_block] = map_bh->b_blocknr+relative_block;
284 page_block++;
285 block_in_file++;
286 }
287 bdev = map_bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 }
289
290 if (first_hole != blocks_per_page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300291 zero_user_segment(page, first_hole << blkbits, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 if (first_hole == 0) {
293 SetPageUptodate(page);
294 unlock_page(page);
295 goto out;
296 }
297 } else if (fully_mapped) {
298 SetPageMappedToDisk(page);
299 }
300
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600301 if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
302 cleancache_get_page(page) == 0) {
303 SetPageUptodate(page);
304 goto confused;
305 }
306
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 /*
308 * This page will go to BIO. Do we need to send this BIO off first?
309 */
310 if (bio && (*last_block_in_bio != blocks[0] - 1))
Mike Christieeed25cd2016-06-05 14:31:59 -0500311 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
313alloc_new:
314 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700315 if (first_hole == blocks_per_page) {
316 if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
317 page))
318 goto out;
319 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
Michal Hocko063d99b2015-10-15 15:28:24 -0700321 min_t(int, nr_pages, BIO_MAX_PAGES), gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 if (bio == NULL)
323 goto confused;
324 }
325
326 length = first_hole << blkbits;
327 if (bio_add_page(bio, page, length, 0) < length) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500328 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 goto alloc_new;
330 }
331
Miquel van Smoorenburg38c8e612009-01-06 14:39:02 -0800332 relative_block = block_in_file - *first_logical_block;
333 nblocks = map_bh->b_size >> blkbits;
334 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
335 (first_hole != blocks_per_page))
Mike Christieeed25cd2016-06-05 14:31:59 -0500336 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 else
338 *last_block_in_bio = blocks[blocks_per_page - 1];
339out:
340 return bio;
341
342confused:
343 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500344 bio = mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 if (!PageUptodate(page))
346 block_read_full_page(page, get_block);
347 else
348 unlock_page(page);
349 goto out;
350}
351
Martin Waitz67be2dd2005-05-01 08:59:26 -0700352/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800353 * mpage_readpages - populate an address space with some pages & start reads against them
Martin Waitz67be2dd2005-05-01 08:59:26 -0700354 * @mapping: the address_space
355 * @pages: The address of a list_head which contains the target pages. These
356 * pages have their ->index populated and are otherwise uninitialised.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700357 * The page at @pages->prev has the lowest file offset, and reads should be
358 * issued in @pages->prev to @pages->next order.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700359 * @nr_pages: The number of pages at *@pages
360 * @get_block: The filesystem's block mapper function.
361 *
362 * This function walks the pages and the blocks within each page, building and
363 * emitting large BIOs.
364 *
365 * If anything unusual happens, such as:
366 *
367 * - encountering a page which has buffers
368 * - encountering a page which has a non-hole after a hole
369 * - encountering a page with non-contiguous blocks
370 *
371 * then this code just gives up and calls the buffer_head-based read function.
372 * It does handle a page which has holes at the end - that is a common case:
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300373 * the end-of-file on blocksize < PAGE_SIZE setups.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700374 *
375 * BH_Boundary explanation:
376 *
377 * There is a problem. The mpage read code assembles several pages, gets all
378 * their disk mappings, and then submits them all. That's fine, but obtaining
379 * the disk mappings may require I/O. Reads of indirect blocks, for example.
380 *
381 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
382 * submitted in the following order:
383 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
Randy Dunlap78a4a502008-02-29 22:02:31 -0800384 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700385 * because the indirect block has to be read to get the mappings of blocks
386 * 13,14,15,16. Obviously, this impacts performance.
387 *
388 * So what we do it to allow the filesystem's get_block() function to set
389 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
390 * after this one will require I/O against a block which is probably close to
391 * this one. So you should push what I/O you have currently accumulated.
392 *
393 * This all causes the disk requests to be issued in the correct order.
394 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395int
396mpage_readpages(struct address_space *mapping, struct list_head *pages,
397 unsigned nr_pages, get_block_t get_block)
398{
399 struct bio *bio = NULL;
400 unsigned page_idx;
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 Hocko8a5c7432016-07-26 15:24:53 -0700404 gfp_t gfp = readahead_gfp_mask(mapping);
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
Andrew Morton02c43632016-03-15 14:55:15 -0700409 struct page *page = lru_to_page(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
411 prefetchw(&page->flags);
412 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -0700413 if (!add_to_page_cache_lru(page, mapping,
Michal Hocko063d99b2015-10-15 15:28:24 -0700414 page->index,
415 gfp)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 bio = do_mpage_readpage(bio, page,
417 nr_pages - page_idx,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800418 &last_block_in_bio, &map_bh,
419 &first_logical_block,
Michal Hocko063d99b2015-10-15 15:28:24 -0700420 get_block, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300422 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 BUG_ON(!list_empty(pages));
425 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500426 mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 return 0;
428}
429EXPORT_SYMBOL(mpage_readpages);
430
431/*
432 * This isn't called much at all
433 */
434int mpage_readpage(struct page *page, get_block_t get_block)
435{
436 struct bio *bio = NULL;
437 sector_t last_block_in_bio = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800438 struct buffer_head map_bh;
439 unsigned long first_logical_block = 0;
Michal Hockoc62d2552015-11-06 16:28:49 -0800440 gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -0400442 map_bh.b_state = 0;
443 map_bh.b_size = 0;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800444 bio = do_mpage_readpage(bio, page, 1, &last_block_in_bio,
Michal Hocko063d99b2015-10-15 15:28:24 -0700445 &map_bh, &first_logical_block, get_block, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500447 mpage_bio_submit(REQ_OP_READ, 0, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 return 0;
449}
450EXPORT_SYMBOL(mpage_readpage);
451
452/*
453 * Writing is not so simple.
454 *
455 * If the page has buffers then they will be used for obtaining the disk
456 * mapping. We only support pages which are fully mapped-and-dirty, with a
457 * special case for pages which are unmapped at the end: end-of-file.
458 *
459 * If the page has no buffers (preferred) then the page is mapped here.
460 *
461 * If all blocks are found to be contiguous then the page can go into the
462 * BIO. Otherwise fall back to the mapping's writepage().
463 *
464 * FIXME: This code wants an estimate of how many pages are still to be
465 * written, so it can intelligently allocate a suitably-sized BIO. For now,
466 * just allocate full-size (16-page) BIOs.
467 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700468
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300469struct mpage_data {
470 struct bio *bio;
471 sector_t last_block_in_bio;
472 get_block_t *get_block;
473 unsigned use_writepage;
474};
475
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700476/*
477 * We have our BIO, so we can now mark the buffers clean. Make
478 * sure to only clean buffers which we know we'll be writing.
479 */
480static void clean_buffers(struct page *page, unsigned first_unmapped)
481{
482 unsigned buffer_counter = 0;
483 struct buffer_head *bh, *head;
484 if (!page_has_buffers(page))
485 return;
486 head = page_buffers(page);
487 bh = head;
488
489 do {
490 if (buffer_counter++ == first_unmapped)
491 break;
492 clear_buffer_dirty(bh);
493 bh = bh->b_this_page;
494 } while (bh != head);
495
496 /*
497 * we cannot drop the bh if the page is not uptodate or a concurrent
498 * readpage would fail to serialize with the bh and it would read from
499 * disk before we reach the platter.
500 */
501 if (buffer_heads_over_limit && PageUptodate(page))
502 try_to_free_buffers(page);
503}
504
Matthew Wilcox133ca5c2017-10-13 15:58:15 -0700505/*
506 * For situations where we want to clean all buffers attached to a page.
507 * We don't need to calculate how many buffers are attached to the page,
508 * we just need to specify a number larger than the maximum number of buffers.
509 */
510void clean_page_buffers(struct page *page)
511{
512 clean_buffers(page, ~0U);
513}
514
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300515static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
Alex Tomas29a814d2008-07-11 19:27:31 -0400516 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700518 struct mpage_data *mpd = data;
519 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 struct address_space *mapping = page->mapping;
521 struct inode *inode = page->mapping->host;
522 const unsigned blkbits = inode->i_blkbits;
523 unsigned long end_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300524 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 sector_t last_block;
526 sector_t block_in_file;
527 sector_t blocks[MAX_BUF_PER_PAGE];
528 unsigned page_block;
529 unsigned first_unmapped = blocks_per_page;
530 struct block_device *bdev = NULL;
531 int boundary = 0;
532 sector_t boundary_block = 0;
533 struct block_device *boundary_bdev = NULL;
534 int length;
535 struct buffer_head map_bh;
536 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700537 int ret = 0;
Mike Christieeed25cd2016-06-05 14:31:59 -0500538 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539
540 if (page_has_buffers(page)) {
541 struct buffer_head *head = page_buffers(page);
542 struct buffer_head *bh = head;
543
544 /* If they're all mapped and dirty, do it */
545 page_block = 0;
546 do {
547 BUG_ON(buffer_locked(bh));
548 if (!buffer_mapped(bh)) {
549 /*
550 * unmapped dirty buffers are created by
551 * __set_page_dirty_buffers -> mmapped data
552 */
553 if (buffer_dirty(bh))
554 goto confused;
555 if (first_unmapped == blocks_per_page)
556 first_unmapped = page_block;
557 continue;
558 }
559
560 if (first_unmapped != blocks_per_page)
561 goto confused; /* hole -> non-hole */
562
563 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
564 goto confused;
565 if (page_block) {
566 if (bh->b_blocknr != blocks[page_block-1] + 1)
567 goto confused;
568 }
569 blocks[page_block++] = bh->b_blocknr;
570 boundary = buffer_boundary(bh);
571 if (boundary) {
572 boundary_block = bh->b_blocknr;
573 boundary_bdev = bh->b_bdev;
574 }
575 bdev = bh->b_bdev;
576 } while ((bh = bh->b_this_page) != head);
577
578 if (first_unmapped)
579 goto page_is_mapped;
580
581 /*
582 * Page has buffers, but they are all unmapped. The page was
583 * created by pagein or read over a hole which was handled by
584 * block_read_full_page(). If this address_space is also
585 * using mpage_readpages then this can rarely happen.
586 */
587 goto confused;
588 }
589
590 /*
591 * The page has no buffers: map it to disk
592 */
593 BUG_ON(!PageUptodate(page));
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300594 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 last_block = (i_size - 1) >> blkbits;
596 map_bh.b_page = page;
597 for (page_block = 0; page_block < blocks_per_page; ) {
598
599 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800600 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700601 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 goto confused;
603 if (buffer_new(&map_bh))
604 unmap_underlying_metadata(map_bh.b_bdev,
605 map_bh.b_blocknr);
606 if (buffer_boundary(&map_bh)) {
607 boundary_block = map_bh.b_blocknr;
608 boundary_bdev = map_bh.b_bdev;
609 }
610 if (page_block) {
611 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
612 goto confused;
613 }
614 blocks[page_block++] = map_bh.b_blocknr;
615 boundary = buffer_boundary(&map_bh);
616 bdev = map_bh.b_bdev;
617 if (block_in_file == last_block)
618 break;
619 block_in_file++;
620 }
621 BUG_ON(page_block == 0);
622
623 first_unmapped = page_block;
624
625page_is_mapped:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300626 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 if (page->index >= end_index) {
628 /*
629 * The page straddles i_size. It must be zeroed out on each
Adam Buchbinder2a61aa42009-12-11 16:35:40 -0500630 * and every writepage invocation because it may be mmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 * "A file is mapped in multiples of the page size. For a file
632 * that is not a multiple of the page size, the remaining memory
633 * is zeroed when mapped, and writes to that region are not
634 * written out to the file."
635 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300636 unsigned offset = i_size & (PAGE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
638 if (page->index > end_index || !offset)
639 goto confused;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300640 zero_user_segment(page, offset, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 }
642
643 /*
644 * This page will go to BIO. Do we need to send this BIO off first?
645 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700646 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Mike Christieeed25cd2016-06-05 14:31:59 -0500647 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
649alloc_new:
650 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700651 if (first_unmapped == blocks_per_page) {
652 if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
Matthew Wilcox133ca5c2017-10-13 15:58:15 -0700653 page, wbc))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700654 goto out;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700655 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
Kent Overstreetb54ffb72015-05-19 14:31:01 +0200657 BIO_MAX_PAGES, GFP_NOFS|__GFP_HIGH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 if (bio == NULL)
659 goto confused;
Tejun Heo429b3fb2015-05-22 17:14:04 -0400660
Tejun Heob16b1de2015-06-02 08:39:48 -0600661 wbc_init_bio(wbc, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 }
663
664 /*
665 * Must try to add the page before marking the buffer clean or
666 * the confused fail path above (OOM) will be very confused when
667 * it finds all bh marked clean (i.e. it will not write anything)
668 */
Tejun Heo2a814902015-05-28 14:50:51 -0400669 wbc_account_io(wbc, page, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 length = first_unmapped << blkbits;
671 if (bio_add_page(bio, page, length, 0) < length) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500672 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 goto alloc_new;
674 }
675
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700676 clean_buffers(page, first_unmapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
678 BUG_ON(PageWriteback(page));
679 set_page_writeback(page);
680 unlock_page(page);
681 if (boundary || (first_unmapped != blocks_per_page)) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500682 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 if (boundary_block) {
684 write_boundary_block(boundary_bdev,
685 boundary_block, 1 << blkbits);
686 }
687 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700688 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 }
690 goto out;
691
692confused:
693 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500694 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700696 if (mpd->use_writepage) {
697 ret = mapping->a_ops->writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700699 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 goto out;
701 }
702 /*
703 * The caller has a ref on the inode, so *mapping is stable
704 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700705 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700707 mpd->bio = bio;
708 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709}
710
711/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800712 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 * @mapping: address space structure to write
714 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
715 * @get_block: the filesystem's block mapper function.
716 * If this is NULL then use a_ops->writepage. Otherwise, go
717 * direct-to-BIO.
718 *
719 * This is a library function, which implements the writepages()
720 * address_space_operation.
721 *
722 * If a page is already under I/O, generic_writepages() skips it, even
723 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
724 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
725 * and msync() need to guarantee that all the data which was dirty at the time
726 * the call was made get new I/O started against them. If wbc->sync_mode is
727 * WB_SYNC_ALL then we were called for data integrity and we must wait for
728 * existing IO to complete.
729 */
730int
731mpage_writepages(struct address_space *mapping,
732 struct writeback_control *wbc, get_block_t get_block)
733{
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200734 struct blk_plug plug;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700735 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200737 blk_start_plug(&plug);
738
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700739 if (!get_block)
740 ret = generic_writepages(mapping, wbc);
741 else {
742 struct mpage_data mpd = {
743 .bio = NULL,
744 .last_block_in_bio = 0,
745 .get_block = get_block,
746 .use_writepage = 1,
747 };
748
749 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
Roman Pen5948edb2015-09-15 08:27:25 -0600750 if (mpd.bio) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500751 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
752 WRITE_SYNC : 0);
753 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
Roman Pen5948edb2015-09-15 08:27:25 -0600754 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 }
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200756 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 return ret;
758}
759EXPORT_SYMBOL(mpage_writepages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
761int mpage_writepage(struct page *page, get_block_t get_block,
762 struct writeback_control *wbc)
763{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700764 struct mpage_data mpd = {
765 .bio = NULL,
766 .last_block_in_bio = 0,
767 .get_block = get_block,
768 .use_writepage = 0,
769 };
770 int ret = __mpage_writepage(page, wbc, &mpd);
Roman Pen5948edb2015-09-15 08:27:25 -0600771 if (mpd.bio) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500772 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
773 WRITE_SYNC : 0);
774 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
Roman Pen5948edb2015-09-15 08:27:25 -0600775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 return ret;
777}
778EXPORT_SYMBOL(mpage_writepage);