blob: a5265828a2f21e59677296c0711093abd73e6857 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * fs/mpage.c
4 *
5 * Copyright (C) 2002, Linus Torvalds.
6 *
7 * Contains functions related to preparing and submitting BIOs which contain
8 * multiple pagecache pages.
9 *
Francois Camie1f8e872008-10-15 22:01:59 -070010 * 15May2002 Andrew Morton
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Initial version
12 * 27Jun2002 axboe@suse.de
13 * use bio_add_page() to build bio's just the right size
14 */
15
16#include <linux/kernel.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050017#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/mm.h>
19#include <linux/kdev_t.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/bio.h>
22#include <linux/fs.h>
23#include <linux/buffer_head.h>
24#include <linux/blkdev.h>
25#include <linux/highmem.h>
26#include <linux/prefetch.h>
27#include <linux/mpage.h>
Andrew Morton02c43632016-03-15 14:55:15 -070028#include <linux/mm_inline.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/writeback.h>
30#include <linux/backing-dev.h>
31#include <linux/pagevec.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060032#include <linux/cleancache.h>
Akinobu Mita4db96b72014-10-09 15:26:55 -070033#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Mohan Srinivasand859afe2016-12-14 16:39:51 -080035#define CREATE_TRACE_POINTS
36#include <trace/events/android_fs.h>
37
38EXPORT_TRACEPOINT_SYMBOL(android_fs_datawrite_start);
39EXPORT_TRACEPOINT_SYMBOL(android_fs_datawrite_end);
40EXPORT_TRACEPOINT_SYMBOL(android_fs_dataread_start);
41EXPORT_TRACEPOINT_SYMBOL(android_fs_dataread_end);
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/*
44 * I/O completion handler for multipage BIOs.
45 *
46 * The mpage code never puts partial pages into a BIO (except for end-of-file).
47 * If a page does not map to a contiguous run of blocks then it simply falls
48 * back to block_read_full_page().
49 *
50 * Why is this? If a page's completion depends on a number of different BIOs
51 * which can complete in any order (or at the same time) then determining the
52 * status of that page is hard. See end_buffer_async_read() for the details.
53 * There is no point in duplicating all that complexity.
54 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020055static void mpage_end_io(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070056{
Kent Overstreet2c30c712013-11-07 12:20:26 -080057 struct bio_vec *bv;
58 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Mohan Srinivasand859afe2016-12-14 16:39:51 -080060 if (trace_android_fs_dataread_end_enabled() &&
61 (bio_data_dir(bio) == READ)) {
62 struct page *first_page = bio->bi_io_vec[0].bv_page;
63
64 if (first_page != NULL)
65 trace_android_fs_dataread_end(first_page->mapping->host,
66 page_offset(first_page),
67 bio->bi_iter.bi_size);
68 }
69
Kent Overstreet2c30c712013-11-07 12:20:26 -080070 bio_for_each_segment_all(bv, bio, i) {
71 struct page *page = bv->bv_page;
Tejun Heo3f289dc2018-07-18 04:47:36 -070072 page_endio(page, bio_op(bio),
73 blk_status_to_errno(bio->bi_status));
Kent Overstreet2c30c712013-11-07 12:20:26 -080074 }
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077}
78
Mike Christieeed25cd2016-06-05 14:31:59 -050079static struct bio *mpage_bio_submit(int op, int op_flags, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080{
Mohan Srinivasand859afe2016-12-14 16:39:51 -080081 if (trace_android_fs_dataread_start_enabled() && (op == REQ_OP_READ)) {
82 struct page *first_page = bio->bi_io_vec[0].bv_page;
83
84 if (first_page != NULL) {
85 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
86
87 path = android_fstrace_get_pathname(pathbuf,
88 MAX_TRACE_PATHBUF_LEN,
89 first_page->mapping->host);
90 trace_android_fs_dataread_start(
91 first_page->mapping->host,
92 page_offset(first_page),
93 bio->bi_iter.bi_size,
94 current->pid,
95 path,
96 current->comm);
97 }
98 }
Hai Shanc32b0d42011-01-13 15:45:51 -080099 bio->bi_end_io = mpage_end_io;
Mike Christieeed25cd2016-06-05 14:31:59 -0500100 bio_set_op_attrs(bio, op, op_flags);
101 guard_bio_eod(op, bio);
Mike Christie4e49ea42016-06-05 14:31:41 -0500102 submit_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 return NULL;
104}
105
106static struct bio *
107mpage_alloc(struct block_device *bdev,
108 sector_t first_sector, int nr_vecs,
Al Virodd0fc662005-10-07 07:46:04 +0100109 gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110{
111 struct bio *bio;
112
Michal Hocko8a5c7432016-07-26 15:24:53 -0700113 /* Restrict the given (page cache) mask for slab allocations */
114 gfp_flags &= GFP_KERNEL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 bio = bio_alloc(gfp_flags, nr_vecs);
116
117 if (bio == NULL && (current->flags & PF_MEMALLOC)) {
118 while (!bio && (nr_vecs /= 2))
119 bio = bio_alloc(gfp_flags, nr_vecs);
120 }
121
122 if (bio) {
Christoph Hellwig74d46992017-08-23 19:10:32 +0200123 bio_set_dev(bio, bdev);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700124 bio->bi_iter.bi_sector = first_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 }
126 return bio;
127}
128
129/*
130 * support function for mpage_readpages. The fs supplied get_block might
131 * return an up to date buffer. This is used to map that buffer into
132 * the page, which allows readpage to avoid triggering a duplicate call
133 * to get_block.
134 *
135 * The idea is to avoid adding buffers to pages that don't already have
136 * them. So when the buffer is up to date and the page size == block size,
137 * this marks the page up to date instead of adding new buffers.
138 */
139static void
140map_buffer_to_page(struct page *page, struct buffer_head *bh, int page_block)
141{
142 struct inode *inode = page->mapping->host;
143 struct buffer_head *page_bh, *head;
144 int block = 0;
145
146 if (!page_has_buffers(page)) {
147 /*
148 * don't make any buffers if there is only one buffer on
149 * the page and the page just needs to be set up to date
150 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300151 if (inode->i_blkbits == PAGE_SHIFT &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 buffer_uptodate(bh)) {
153 SetPageUptodate(page);
154 return;
155 }
Fabian Frederick93407472017-02-27 14:28:32 -0800156 create_empty_buffers(page, i_blocksize(inode), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 }
158 head = page_buffers(page);
159 page_bh = head;
160 do {
161 if (block == page_block) {
162 page_bh->b_state = bh->b_state;
163 page_bh->b_bdev = bh->b_bdev;
164 page_bh->b_blocknr = bh->b_blocknr;
165 break;
166 }
167 page_bh = page_bh->b_this_page;
168 block++;
169 } while (page_bh != head);
170}
171
Jens Axboe357c1202018-08-17 15:45:32 -0700172struct mpage_readpage_args {
173 struct bio *bio;
174 struct page *page;
175 unsigned int nr_pages;
Jens Axboe74c81642018-08-17 15:45:36 -0700176 bool is_readahead;
Jens Axboe357c1202018-08-17 15:45:32 -0700177 sector_t last_block_in_bio;
178 struct buffer_head map_bh;
179 unsigned long first_logical_block;
180 get_block_t *get_block;
Jens Axboe357c1202018-08-17 15:45:32 -0700181};
182
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800183/*
184 * This is the worker routine which does all the work of mapping the disk
185 * blocks and constructs largest possible bios, submits them for IO if the
186 * blocks are not contiguous on the disk.
187 *
188 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
189 * represent the validity of its disk mapping and to decide when to do the next
190 * get_block() call.
191 */
Jens Axboe357c1202018-08-17 15:45:32 -0700192static struct bio *do_mpage_readpage(struct mpage_readpage_args *args)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193{
Jens Axboe357c1202018-08-17 15:45:32 -0700194 struct page *page = args->page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 struct inode *inode = page->mapping->host;
196 const unsigned blkbits = inode->i_blkbits;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300197 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 const unsigned blocksize = 1 << blkbits;
Jens Axboe357c1202018-08-17 15:45:32 -0700199 struct buffer_head *map_bh = &args->map_bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 sector_t block_in_file;
201 sector_t last_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800202 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 sector_t blocks[MAX_BUF_PER_PAGE];
204 unsigned page_block;
205 unsigned first_hole = blocks_per_page;
206 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 int length;
208 int fully_mapped = 1;
Jens Axboe74c81642018-08-17 15:45:36 -0700209 int op_flags;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800210 unsigned nblocks;
211 unsigned relative_block;
Jens Axboe74c81642018-08-17 15:45:36 -0700212 gfp_t gfp;
213
214 if (args->is_readahead) {
215 op_flags = REQ_RAHEAD;
216 gfp = readahead_gfp_mask(page->mapping);
217 } else {
218 op_flags = 0;
219 gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
220 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
222 if (page_has_buffers(page))
223 goto confused;
224
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300225 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Jens Axboe357c1202018-08-17 15:45:32 -0700226 last_block = block_in_file + args->nr_pages * blocks_per_page;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800227 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
228 if (last_block > last_block_in_file)
229 last_block = last_block_in_file;
230 page_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800232 /*
233 * Map blocks using the result from the previous get_blocks call first.
234 */
235 nblocks = map_bh->b_size >> blkbits;
Jens Axboe357c1202018-08-17 15:45:32 -0700236 if (buffer_mapped(map_bh) &&
237 block_in_file > args->first_logical_block &&
238 block_in_file < (args->first_logical_block + nblocks)) {
239 unsigned map_offset = block_in_file - args->first_logical_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800240 unsigned last = nblocks - map_offset;
241
242 for (relative_block = 0; ; relative_block++) {
243 if (relative_block == last) {
244 clear_buffer_mapped(map_bh);
245 break;
246 }
247 if (page_block == blocks_per_page)
248 break;
249 blocks[page_block] = map_bh->b_blocknr + map_offset +
250 relative_block;
251 page_block++;
252 block_in_file++;
253 }
254 bdev = map_bh->b_bdev;
255 }
256
257 /*
258 * Then do more get_blocks calls until we are done with this page.
259 */
260 map_bh->b_page = page;
261 while (page_block < blocks_per_page) {
262 map_bh->b_state = 0;
263 map_bh->b_size = 0;
264
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800266 map_bh->b_size = (last_block-block_in_file) << blkbits;
Jens Axboe357c1202018-08-17 15:45:32 -0700267 if (args->get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 goto confused;
Jens Axboe357c1202018-08-17 15:45:32 -0700269 args->first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 }
271
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800272 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 fully_mapped = 0;
274 if (first_hole == blocks_per_page)
275 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800276 page_block++;
277 block_in_file++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 continue;
279 }
280
281 /* some filesystems will copy data into the page during
282 * the get_block call, in which case we don't want to
283 * read it again. map_buffer_to_page copies the data
284 * we just collected from get_block into the page's buffers
285 * so readpage doesn't have to repeat the get_block call
286 */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800287 if (buffer_uptodate(map_bh)) {
288 map_buffer_to_page(page, map_bh, page_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 goto confused;
290 }
291
292 if (first_hole != blocks_per_page)
293 goto confused; /* hole -> non-hole */
294
295 /* Contiguous blocks? */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800296 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800298 nblocks = map_bh->b_size >> blkbits;
299 for (relative_block = 0; ; relative_block++) {
300 if (relative_block == nblocks) {
301 clear_buffer_mapped(map_bh);
302 break;
303 } else if (page_block == blocks_per_page)
304 break;
305 blocks[page_block] = map_bh->b_blocknr+relative_block;
306 page_block++;
307 block_in_file++;
308 }
309 bdev = map_bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 }
311
312 if (first_hole != blocks_per_page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300313 zero_user_segment(page, first_hole << blkbits, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 if (first_hole == 0) {
315 SetPageUptodate(page);
316 unlock_page(page);
317 goto out;
318 }
319 } else if (fully_mapped) {
320 SetPageMappedToDisk(page);
321 }
322
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600323 if (fully_mapped && blocks_per_page == 1 && !PageUptodate(page) &&
324 cleancache_get_page(page) == 0) {
325 SetPageUptodate(page);
326 goto confused;
327 }
328
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 /*
330 * This page will go to BIO. Do we need to send this BIO off first?
331 */
Jens Axboe357c1202018-08-17 15:45:32 -0700332 if (args->bio && (args->last_block_in_bio != blocks[0] - 1))
Jens Axboe74c81642018-08-17 15:45:36 -0700333 args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335alloc_new:
Jens Axboe357c1202018-08-17 15:45:32 -0700336 if (args->bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700337 if (first_hole == blocks_per_page) {
338 if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
339 page))
340 goto out;
341 }
Jens Axboe357c1202018-08-17 15:45:32 -0700342 args->bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
343 min_t(int, args->nr_pages,
344 BIO_MAX_PAGES),
Jens Axboe74c81642018-08-17 15:45:36 -0700345 gfp);
Jens Axboe357c1202018-08-17 15:45:32 -0700346 if (args->bio == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 goto confused;
348 }
349
350 length = first_hole << blkbits;
Jens Axboe357c1202018-08-17 15:45:32 -0700351 if (bio_add_page(args->bio, page, length, 0) < length) {
Jens Axboe74c81642018-08-17 15:45:36 -0700352 args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 goto alloc_new;
354 }
355
Jens Axboe357c1202018-08-17 15:45:32 -0700356 relative_block = block_in_file - args->first_logical_block;
Miquel van Smoorenburg38c8e612009-01-06 14:39:02 -0800357 nblocks = map_bh->b_size >> blkbits;
358 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
359 (first_hole != blocks_per_page))
Jens Axboe74c81642018-08-17 15:45:36 -0700360 args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 else
Jens Axboe357c1202018-08-17 15:45:32 -0700362 args->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363out:
Jens Axboe357c1202018-08-17 15:45:32 -0700364 return args->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
366confused:
Jens Axboe357c1202018-08-17 15:45:32 -0700367 if (args->bio)
Jens Axboe74c81642018-08-17 15:45:36 -0700368 args->bio = mpage_bio_submit(REQ_OP_READ, op_flags, args->bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 if (!PageUptodate(page))
Jens Axboe357c1202018-08-17 15:45:32 -0700370 block_read_full_page(page, args->get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 else
372 unlock_page(page);
373 goto out;
374}
375
Martin Waitz67be2dd2005-05-01 08:59:26 -0700376/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800377 * mpage_readpages - populate an address space with some pages & start reads against them
Martin Waitz67be2dd2005-05-01 08:59:26 -0700378 * @mapping: the address_space
379 * @pages: The address of a list_head which contains the target pages. These
380 * pages have their ->index populated and are otherwise uninitialised.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700381 * The page at @pages->prev has the lowest file offset, and reads should be
382 * issued in @pages->prev to @pages->next order.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700383 * @nr_pages: The number of pages at *@pages
384 * @get_block: The filesystem's block mapper function.
385 *
386 * This function walks the pages and the blocks within each page, building and
387 * emitting large BIOs.
388 *
389 * If anything unusual happens, such as:
390 *
391 * - encountering a page which has buffers
392 * - encountering a page which has a non-hole after a hole
393 * - encountering a page with non-contiguous blocks
394 *
395 * then this code just gives up and calls the buffer_head-based read function.
396 * It does handle a page which has holes at the end - that is a common case:
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300397 * the end-of-file on blocksize < PAGE_SIZE setups.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700398 *
399 * BH_Boundary explanation:
400 *
401 * There is a problem. The mpage read code assembles several pages, gets all
402 * their disk mappings, and then submits them all. That's fine, but obtaining
403 * the disk mappings may require I/O. Reads of indirect blocks, for example.
404 *
405 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
406 * submitted in the following order:
Mauro Carvalho Chehab0117d422017-05-12 07:45:42 -0300407 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700408 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
Randy Dunlap78a4a502008-02-29 22:02:31 -0800409 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700410 * because the indirect block has to be read to get the mappings of blocks
411 * 13,14,15,16. Obviously, this impacts performance.
412 *
413 * So what we do it to allow the filesystem's get_block() function to set
414 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
415 * after this one will require I/O against a block which is probably close to
416 * this one. So you should push what I/O you have currently accumulated.
417 *
418 * This all causes the disk requests to be issued in the correct order.
419 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420int
421mpage_readpages(struct address_space *mapping, struct list_head *pages,
422 unsigned nr_pages, get_block_t get_block)
423{
Jens Axboe357c1202018-08-17 15:45:32 -0700424 struct mpage_readpage_args args = {
425 .get_block = get_block,
Jens Axboe74c81642018-08-17 15:45:36 -0700426 .is_readahead = true,
Jens Axboe357c1202018-08-17 15:45:32 -0700427 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 unsigned page_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
Andrew Morton02c43632016-03-15 14:55:15 -0700431 struct page *page = lru_to_page(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
433 prefetchw(&page->flags);
434 list_del(&page->lru);
Nick Piggineb2be182007-10-16 01:24:57 -0700435 if (!add_to_page_cache_lru(page, mapping,
Michal Hocko063d99b2015-10-15 15:28:24 -0700436 page->index,
Jens Axboe74c81642018-08-17 15:45:36 -0700437 readahead_gfp_mask(mapping))) {
Jens Axboe357c1202018-08-17 15:45:32 -0700438 args.page = page;
439 args.nr_pages = nr_pages - page_idx;
440 args.bio = do_mpage_readpage(&args);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300442 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 BUG_ON(!list_empty(pages));
Jens Axboe357c1202018-08-17 15:45:32 -0700445 if (args.bio)
Jens Axboe74c81642018-08-17 15:45:36 -0700446 mpage_bio_submit(REQ_OP_READ, REQ_RAHEAD, args.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 return 0;
448}
449EXPORT_SYMBOL(mpage_readpages);
450
451/*
452 * This isn't called much at all
453 */
454int mpage_readpage(struct page *page, get_block_t get_block)
455{
Jens Axboe357c1202018-08-17 15:45:32 -0700456 struct mpage_readpage_args args = {
457 .page = page,
458 .nr_pages = 1,
459 .get_block = get_block,
Jens Axboe357c1202018-08-17 15:45:32 -0700460 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Jens Axboe357c1202018-08-17 15:45:32 -0700462 args.bio = do_mpage_readpage(&args);
463 if (args.bio)
464 mpage_bio_submit(REQ_OP_READ, 0, args.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 return 0;
466}
467EXPORT_SYMBOL(mpage_readpage);
468
469/*
470 * Writing is not so simple.
471 *
472 * If the page has buffers then they will be used for obtaining the disk
473 * mapping. We only support pages which are fully mapped-and-dirty, with a
474 * special case for pages which are unmapped at the end: end-of-file.
475 *
476 * If the page has no buffers (preferred) then the page is mapped here.
477 *
478 * If all blocks are found to be contiguous then the page can go into the
479 * BIO. Otherwise fall back to the mapping's writepage().
480 *
481 * FIXME: This code wants an estimate of how many pages are still to be
482 * written, so it can intelligently allocate a suitably-sized BIO. For now,
483 * just allocate full-size (16-page) BIOs.
484 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700485
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300486struct mpage_data {
487 struct bio *bio;
488 sector_t last_block_in_bio;
489 get_block_t *get_block;
490 unsigned use_writepage;
491};
492
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700493/*
494 * We have our BIO, so we can now mark the buffers clean. Make
495 * sure to only clean buffers which we know we'll be writing.
496 */
497static void clean_buffers(struct page *page, unsigned first_unmapped)
498{
499 unsigned buffer_counter = 0;
500 struct buffer_head *bh, *head;
501 if (!page_has_buffers(page))
502 return;
503 head = page_buffers(page);
504 bh = head;
505
506 do {
507 if (buffer_counter++ == first_unmapped)
508 break;
509 clear_buffer_dirty(bh);
510 bh = bh->b_this_page;
511 } while (bh != head);
512
513 /*
514 * we cannot drop the bh if the page is not uptodate or a concurrent
515 * readpage would fail to serialize with the bh and it would read from
516 * disk before we reach the platter.
517 */
518 if (buffer_heads_over_limit && PageUptodate(page))
519 try_to_free_buffers(page);
520}
521
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700522/*
523 * For situations where we want to clean all buffers attached to a page.
524 * We don't need to calculate how many buffers are attached to the page,
525 * we just need to specify a number larger than the maximum number of buffers.
526 */
527void clean_page_buffers(struct page *page)
528{
529 clean_buffers(page, ~0U);
530}
531
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300532static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
Alex Tomas29a814d2008-07-11 19:27:31 -0400533 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700535 struct mpage_data *mpd = data;
536 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 struct address_space *mapping = page->mapping;
538 struct inode *inode = page->mapping->host;
539 const unsigned blkbits = inode->i_blkbits;
540 unsigned long end_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300541 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 sector_t last_block;
543 sector_t block_in_file;
544 sector_t blocks[MAX_BUF_PER_PAGE];
545 unsigned page_block;
546 unsigned first_unmapped = blocks_per_page;
547 struct block_device *bdev = NULL;
548 int boundary = 0;
549 sector_t boundary_block = 0;
550 struct block_device *boundary_bdev = NULL;
551 int length;
552 struct buffer_head map_bh;
553 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700554 int ret = 0;
Jens Axboe76372412016-11-01 10:00:38 -0600555 int op_flags = wbc_to_write_flags(wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
557 if (page_has_buffers(page)) {
558 struct buffer_head *head = page_buffers(page);
559 struct buffer_head *bh = head;
560
561 /* If they're all mapped and dirty, do it */
562 page_block = 0;
563 do {
564 BUG_ON(buffer_locked(bh));
565 if (!buffer_mapped(bh)) {
566 /*
567 * unmapped dirty buffers are created by
568 * __set_page_dirty_buffers -> mmapped data
569 */
570 if (buffer_dirty(bh))
571 goto confused;
572 if (first_unmapped == blocks_per_page)
573 first_unmapped = page_block;
574 continue;
575 }
576
577 if (first_unmapped != blocks_per_page)
578 goto confused; /* hole -> non-hole */
579
580 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
581 goto confused;
582 if (page_block) {
583 if (bh->b_blocknr != blocks[page_block-1] + 1)
584 goto confused;
585 }
586 blocks[page_block++] = bh->b_blocknr;
587 boundary = buffer_boundary(bh);
588 if (boundary) {
589 boundary_block = bh->b_blocknr;
590 boundary_bdev = bh->b_bdev;
591 }
592 bdev = bh->b_bdev;
593 } while ((bh = bh->b_this_page) != head);
594
595 if (first_unmapped)
596 goto page_is_mapped;
597
598 /*
599 * Page has buffers, but they are all unmapped. The page was
600 * created by pagein or read over a hole which was handled by
601 * block_read_full_page(). If this address_space is also
602 * using mpage_readpages then this can rarely happen.
603 */
604 goto confused;
605 }
606
607 /*
608 * The page has no buffers: map it to disk
609 */
610 BUG_ON(!PageUptodate(page));
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300611 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 last_block = (i_size - 1) >> blkbits;
613 map_bh.b_page = page;
614 for (page_block = 0; page_block < blocks_per_page; ) {
615
616 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800617 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700618 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 goto confused;
620 if (buffer_new(&map_bh))
Jan Karae64855c2016-11-04 18:08:15 +0100621 clean_bdev_bh_alias(&map_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 if (buffer_boundary(&map_bh)) {
623 boundary_block = map_bh.b_blocknr;
624 boundary_bdev = map_bh.b_bdev;
625 }
626 if (page_block) {
627 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
628 goto confused;
629 }
630 blocks[page_block++] = map_bh.b_blocknr;
631 boundary = buffer_boundary(&map_bh);
632 bdev = map_bh.b_bdev;
633 if (block_in_file == last_block)
634 break;
635 block_in_file++;
636 }
637 BUG_ON(page_block == 0);
638
639 first_unmapped = page_block;
640
641page_is_mapped:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300642 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 if (page->index >= end_index) {
644 /*
645 * The page straddles i_size. It must be zeroed out on each
Adam Buchbinder2a61aa42009-12-11 16:35:40 -0500646 * and every writepage invocation because it may be mmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 * "A file is mapped in multiples of the page size. For a file
648 * that is not a multiple of the page size, the remaining memory
649 * is zeroed when mapped, and writes to that region are not
650 * written out to the file."
651 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300652 unsigned offset = i_size & (PAGE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
654 if (page->index > end_index || !offset)
655 goto confused;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300656 zero_user_segment(page, offset, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 }
658
659 /*
660 * This page will go to BIO. Do we need to send this BIO off first?
661 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700662 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Mike Christieeed25cd2016-06-05 14:31:59 -0500663 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
665alloc_new:
666 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700667 if (first_unmapped == blocks_per_page) {
668 if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700669 page, wbc))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700670 goto out;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700671 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 bio = mpage_alloc(bdev, blocks[0] << (blkbits - 9),
Kent Overstreetb54ffb72015-05-19 14:31:01 +0200673 BIO_MAX_PAGES, GFP_NOFS|__GFP_HIGH);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 if (bio == NULL)
675 goto confused;
Tejun Heo429b3fb2015-05-22 17:14:04 -0400676
Tejun Heob16b1de2015-06-02 08:39:48 -0600677 wbc_init_bio(wbc, bio);
Jens Axboe8e8f9292017-06-27 09:30:05 -0600678 bio->bi_write_hint = inode->i_write_hint;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 }
680
681 /*
682 * Must try to add the page before marking the buffer clean or
683 * the confused fail path above (OOM) will be very confused when
684 * it finds all bh marked clean (i.e. it will not write anything)
685 */
Tejun Heo2a814902015-05-28 14:50:51 -0400686 wbc_account_io(wbc, page, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 length = first_unmapped << blkbits;
688 if (bio_add_page(bio, page, length, 0) < length) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500689 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 goto alloc_new;
691 }
692
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700693 clean_buffers(page, first_unmapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
695 BUG_ON(PageWriteback(page));
696 set_page_writeback(page);
697 unlock_page(page);
698 if (boundary || (first_unmapped != blocks_per_page)) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500699 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 if (boundary_block) {
701 write_boundary_block(boundary_bdev,
702 boundary_block, 1 << blkbits);
703 }
704 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700705 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
707 goto out;
708
709confused:
710 if (bio)
Mike Christieeed25cd2016-06-05 14:31:59 -0500711 bio = mpage_bio_submit(REQ_OP_WRITE, op_flags, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700713 if (mpd->use_writepage) {
714 ret = mapping->a_ops->writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700716 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 goto out;
718 }
719 /*
720 * The caller has a ref on the inode, so *mapping is stable
721 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700722 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700724 mpd->bio = bio;
725 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726}
727
728/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800729 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 * @mapping: address space structure to write
731 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
732 * @get_block: the filesystem's block mapper function.
733 * If this is NULL then use a_ops->writepage. Otherwise, go
734 * direct-to-BIO.
735 *
736 * This is a library function, which implements the writepages()
737 * address_space_operation.
738 *
739 * If a page is already under I/O, generic_writepages() skips it, even
740 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
741 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
742 * and msync() need to guarantee that all the data which was dirty at the time
743 * the call was made get new I/O started against them. If wbc->sync_mode is
744 * WB_SYNC_ALL then we were called for data integrity and we must wait for
745 * existing IO to complete.
746 */
747int
748mpage_writepages(struct address_space *mapping,
749 struct writeback_control *wbc, get_block_t get_block)
750{
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200751 struct blk_plug plug;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700752 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200754 blk_start_plug(&plug);
755
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700756 if (!get_block)
757 ret = generic_writepages(mapping, wbc);
758 else {
759 struct mpage_data mpd = {
760 .bio = NULL,
761 .last_block_in_bio = 0,
762 .get_block = get_block,
763 .use_writepage = 1,
764 };
765
766 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
Roman Pen5948edb2015-09-15 08:27:25 -0600767 if (mpd.bio) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500768 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
Christoph Hellwig70fd7612016-11-01 07:40:10 -0600769 REQ_SYNC : 0);
Mike Christieeed25cd2016-06-05 14:31:59 -0500770 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
Roman Pen5948edb2015-09-15 08:27:25 -0600771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 }
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200773 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 return ret;
775}
776EXPORT_SYMBOL(mpage_writepages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778int mpage_writepage(struct page *page, get_block_t get_block,
779 struct writeback_control *wbc)
780{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700781 struct mpage_data mpd = {
782 .bio = NULL,
783 .last_block_in_bio = 0,
784 .get_block = get_block,
785 .use_writepage = 0,
786 };
787 int ret = __mpage_writepage(page, wbc, &mpd);
Roman Pen5948edb2015-09-15 08:27:25 -0600788 if (mpd.bio) {
Mike Christieeed25cd2016-06-05 14:31:59 -0500789 int op_flags = (wbc->sync_mode == WB_SYNC_ALL ?
Christoph Hellwig70fd7612016-11-01 07:40:10 -0600790 REQ_SYNC : 0);
Mike Christieeed25cd2016-06-05 14:31:59 -0500791 mpage_bio_submit(REQ_OP_WRITE, op_flags, mpd.bio);
Roman Pen5948edb2015-09-15 08:27:25 -0600792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return ret;
794}
795EXPORT_SYMBOL(mpage_writepage);