blob: 242e213ee0644e82bb5b0de9e2c1e32a1a76a46f [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>
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
Matthew Wilcox (Oracle)2c69e202022-04-29 10:40:40 -040039 * back to block_read_full_folio().
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 *
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 */
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +020046static void mpage_read_end_io(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070047{
Pankaj Raghav09a607c2023-04-11 14:29:20 +020048 struct folio_iter fi;
49 int err = blk_status_to_errno(bio->bi_status);
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Pankaj Raghav09a607c2023-04-11 14:29:20 +020051 bio_for_each_folio_all(fi, bio) {
52 if (err)
53 folio_set_error(fi.folio);
54 else
55 folio_mark_uptodate(fi.folio);
56 folio_unlock(fi.folio);
57 }
Kent Overstreet2c30c712013-11-07 12:20:26 -080058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060}
61
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +020062static void mpage_write_end_io(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070063{
Pankaj Raghav09a607c2023-04-11 14:29:20 +020064 struct folio_iter fi;
65 int err = blk_status_to_errno(bio->bi_status);
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +020066
Pankaj Raghav09a607c2023-04-11 14:29:20 +020067 bio_for_each_folio_all(fi, bio) {
68 if (err) {
69 folio_set_error(fi.folio);
70 mapping_set_error(fi.folio->mapping, err);
71 }
72 folio_end_writeback(fi.folio);
73 }
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +020074
75 bio_put(bio);
76}
77
78static struct bio *mpage_bio_submit_read(struct bio *bio)
79{
80 bio->bi_end_io = mpage_read_end_io;
81 guard_bio_eod(bio);
82 submit_bio(bio);
83 return NULL;
84}
85
86static struct bio *mpage_bio_submit_write(struct bio *bio)
87{
88 bio->bi_end_io = mpage_write_end_io;
Ming Lei83c9c542020-01-05 09:41:14 +080089 guard_bio_eod(bio);
Mike Christie4e49ea42016-06-05 14:31:41 -050090 submit_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 return NULL;
92}
93
Linus Torvalds1da177e2005-04-16 15:20:36 -070094/*
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -070095 * support function for mpage_readahead. The fs supplied get_block might
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * return an up to date buffer. This is used to map that buffer into
Matthew Wilcox (Oracle)2c69e202022-04-29 10:40:40 -040097 * the page, which allows read_folio to avoid triggering a duplicate call
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 * to get_block.
99 *
100 * The idea is to avoid adding buffers to pages that don't already have
101 * them. So when the buffer is up to date and the page size == block size,
102 * this marks the page up to date instead of adding new buffers.
103 */
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400104static void map_buffer_to_folio(struct folio *folio, struct buffer_head *bh,
105 int page_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106{
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400107 struct inode *inode = folio->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 struct buffer_head *page_bh, *head;
109 int block = 0;
110
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400111 head = folio_buffers(folio);
112 if (!head) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 /*
114 * don't make any buffers if there is only one buffer on
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400115 * the folio and the folio just needs to be set up to date
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300117 if (inode->i_blkbits == PAGE_SHIFT &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 buffer_uptodate(bh)) {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400119 folio_mark_uptodate(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 return;
121 }
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400122 create_empty_buffers(&folio->page, i_blocksize(inode), 0);
123 head = folio_buffers(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 }
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400125
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 page_bh = head;
127 do {
128 if (block == page_block) {
129 page_bh->b_state = bh->b_state;
130 page_bh->b_bdev = bh->b_bdev;
131 page_bh->b_blocknr = bh->b_blocknr;
132 break;
133 }
134 page_bh = page_bh->b_this_page;
135 block++;
136 } while (page_bh != head);
137}
138
Jens Axboe357c1202018-08-17 15:45:32 -0700139struct mpage_readpage_args {
140 struct bio *bio;
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400141 struct folio *folio;
Jens Axboe357c1202018-08-17 15:45:32 -0700142 unsigned int nr_pages;
Jens Axboe74c81642018-08-17 15:45:36 -0700143 bool is_readahead;
Jens Axboe357c1202018-08-17 15:45:32 -0700144 sector_t last_block_in_bio;
145 struct buffer_head map_bh;
146 unsigned long first_logical_block;
147 get_block_t *get_block;
Jens Axboe357c1202018-08-17 15:45:32 -0700148};
149
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800150/*
151 * This is the worker routine which does all the work of mapping the disk
152 * blocks and constructs largest possible bios, submits them for IO if the
153 * blocks are not contiguous on the disk.
154 *
155 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
156 * represent the validity of its disk mapping and to decide when to do the next
157 * get_block() call.
158 */
Jens Axboe357c1202018-08-17 15:45:32 -0700159static struct bio *do_mpage_readpage(struct mpage_readpage_args *args)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160{
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400161 struct folio *folio = args->folio;
162 struct inode *inode = folio->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 const unsigned blkbits = inode->i_blkbits;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300164 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 const unsigned blocksize = 1 << blkbits;
Jens Axboe357c1202018-08-17 15:45:32 -0700166 struct buffer_head *map_bh = &args->map_bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 sector_t block_in_file;
168 sector_t last_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800169 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 sector_t blocks[MAX_BUF_PER_PAGE];
171 unsigned page_block;
172 unsigned first_hole = blocks_per_page;
173 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 int length;
175 int fully_mapped = 1;
Bart Van Asschef84c94a2022-07-14 11:07:15 -0700176 blk_opf_t opf = REQ_OP_READ;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800177 unsigned nblocks;
178 unsigned relative_block;
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400179 gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
180
181 /* MAX_BUF_PER_PAGE, for example */
182 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
Jens Axboe74c81642018-08-17 15:45:36 -0700183
184 if (args->is_readahead) {
Bart Van Asschef84c94a2022-07-14 11:07:15 -0700185 opf |= REQ_RAHEAD;
Christoph Hellwig61285ff2022-03-23 16:39:52 +0100186 gfp |= __GFP_NORETRY | __GFP_NOWARN;
Jens Axboe74c81642018-08-17 15:45:36 -0700187 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400189 if (folio_buffers(folio))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 goto confused;
191
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400192 block_in_file = (sector_t)folio->index << (PAGE_SHIFT - blkbits);
Jens Axboe357c1202018-08-17 15:45:32 -0700193 last_block = block_in_file + args->nr_pages * blocks_per_page;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800194 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
195 if (last_block > last_block_in_file)
196 last_block = last_block_in_file;
197 page_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800199 /*
200 * Map blocks using the result from the previous get_blocks call first.
201 */
202 nblocks = map_bh->b_size >> blkbits;
Jens Axboe357c1202018-08-17 15:45:32 -0700203 if (buffer_mapped(map_bh) &&
204 block_in_file > args->first_logical_block &&
205 block_in_file < (args->first_logical_block + nblocks)) {
206 unsigned map_offset = block_in_file - args->first_logical_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800207 unsigned last = nblocks - map_offset;
208
209 for (relative_block = 0; ; relative_block++) {
210 if (relative_block == last) {
211 clear_buffer_mapped(map_bh);
212 break;
213 }
214 if (page_block == blocks_per_page)
215 break;
216 blocks[page_block] = map_bh->b_blocknr + map_offset +
217 relative_block;
218 page_block++;
219 block_in_file++;
220 }
221 bdev = map_bh->b_bdev;
222 }
223
224 /*
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400225 * Then do more get_blocks calls until we are done with this folio.
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800226 */
Matthew Wilcox (Oracle)a5fd8392022-12-15 21:44:02 +0000227 map_bh->b_folio = folio;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800228 while (page_block < blocks_per_page) {
229 map_bh->b_state = 0;
230 map_bh->b_size = 0;
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800233 map_bh->b_size = (last_block-block_in_file) << blkbits;
Jens Axboe357c1202018-08-17 15:45:32 -0700234 if (args->get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 goto confused;
Jens Axboe357c1202018-08-17 15:45:32 -0700236 args->first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 }
238
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800239 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 fully_mapped = 0;
241 if (first_hole == blocks_per_page)
242 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800243 page_block++;
244 block_in_file++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 continue;
246 }
247
248 /* some filesystems will copy data into the page during
249 * the get_block call, in which case we don't want to
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400250 * read it again. map_buffer_to_folio copies the data
251 * we just collected from get_block into the folio's buffers
252 * so read_folio doesn't have to repeat the get_block call
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800254 if (buffer_uptodate(map_bh)) {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400255 map_buffer_to_folio(folio, map_bh, page_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 goto confused;
257 }
258
259 if (first_hole != blocks_per_page)
260 goto confused; /* hole -> non-hole */
261
262 /* Contiguous blocks? */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800263 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800265 nblocks = map_bh->b_size >> blkbits;
266 for (relative_block = 0; ; relative_block++) {
267 if (relative_block == nblocks) {
268 clear_buffer_mapped(map_bh);
269 break;
270 } else if (page_block == blocks_per_page)
271 break;
272 blocks[page_block] = map_bh->b_blocknr+relative_block;
273 page_block++;
274 block_in_file++;
275 }
276 bdev = map_bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 }
278
279 if (first_hole != blocks_per_page) {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400280 folio_zero_segment(folio, first_hole << blkbits, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 if (first_hole == 0) {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400282 folio_mark_uptodate(folio);
283 folio_unlock(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 goto out;
285 }
286 } else if (fully_mapped) {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400287 folio_set_mappedtodisk(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 }
289
290 /*
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400291 * This folio will go to BIO. Do we need to send this BIO off first?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 */
Jens Axboe357c1202018-08-17 15:45:32 -0700293 if (args->bio && (args->last_block_in_bio != blocks[0] - 1))
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +0200294 args->bio = mpage_bio_submit_read(args->bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295
296alloc_new:
Jens Axboe357c1202018-08-17 15:45:32 -0700297 if (args->bio == NULL) {
Bart Van Asschef84c94a2022-07-14 11:07:15 -0700298 args->bio = bio_alloc(bdev, bio_max_segs(args->nr_pages), opf,
Christoph Hellwig07888c662022-01-24 10:11:05 +0100299 gfp);
Jens Axboe357c1202018-08-17 15:45:32 -0700300 if (args->bio == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 goto confused;
Christoph Hellwigd5f68a42022-01-24 10:10:49 +0100302 args->bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 }
304
305 length = first_hole << blkbits;
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400306 if (!bio_add_folio(args->bio, folio, length, 0)) {
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +0200307 args->bio = mpage_bio_submit_read(args->bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 goto alloc_new;
309 }
310
Jens Axboe357c1202018-08-17 15:45:32 -0700311 relative_block = block_in_file - args->first_logical_block;
Miquel van Smoorenburg38c8e612009-01-06 14:39:02 -0800312 nblocks = map_bh->b_size >> blkbits;
313 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
314 (first_hole != blocks_per_page))
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +0200315 args->bio = mpage_bio_submit_read(args->bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 else
Jens Axboe357c1202018-08-17 15:45:32 -0700317 args->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318out:
Jens Axboe357c1202018-08-17 15:45:32 -0700319 return args->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
321confused:
Jens Axboe357c1202018-08-17 15:45:32 -0700322 if (args->bio)
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +0200323 args->bio = mpage_bio_submit_read(args->bio);
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400324 if (!folio_test_uptodate(folio))
325 block_read_full_folio(folio, args->get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 else
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400327 folio_unlock(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 goto out;
329}
330
Martin Waitz67be2dd2005-05-01 08:59:26 -0700331/**
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700332 * mpage_readahead - start reads against pages
333 * @rac: Describes which pages to read.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700334 * @get_block: The filesystem's block mapper function.
335 *
336 * This function walks the pages and the blocks within each page, building and
337 * emitting large BIOs.
338 *
339 * If anything unusual happens, such as:
340 *
341 * - encountering a page which has buffers
342 * - encountering a page which has a non-hole after a hole
343 * - encountering a page with non-contiguous blocks
344 *
345 * then this code just gives up and calls the buffer_head-based read function.
346 * It does handle a page which has holes at the end - that is a common case:
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300347 * the end-of-file on blocksize < PAGE_SIZE setups.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700348 *
349 * BH_Boundary explanation:
350 *
351 * There is a problem. The mpage read code assembles several pages, gets all
352 * their disk mappings, and then submits them all. That's fine, but obtaining
353 * the disk mappings may require I/O. Reads of indirect blocks, for example.
354 *
355 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
356 * submitted in the following order:
Mauro Carvalho Chehab0117d422017-05-12 07:45:42 -0300357 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700358 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
Randy Dunlap78a4a502008-02-29 22:02:31 -0800359 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700360 * because the indirect block has to be read to get the mappings of blocks
361 * 13,14,15,16. Obviously, this impacts performance.
362 *
363 * So what we do it to allow the filesystem's get_block() function to set
364 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
365 * after this one will require I/O against a block which is probably close to
366 * this one. So you should push what I/O you have currently accumulated.
367 *
368 * This all causes the disk requests to be issued in the correct order.
369 */
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700370void mpage_readahead(struct readahead_control *rac, get_block_t get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400372 struct folio *folio;
Jens Axboe357c1202018-08-17 15:45:32 -0700373 struct mpage_readpage_args args = {
374 .get_block = get_block,
Jens Axboe74c81642018-08-17 15:45:36 -0700375 .is_readahead = true,
Jens Axboe357c1202018-08-17 15:45:32 -0700376 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400378 while ((folio = readahead_folio(rac))) {
379 prefetchw(&folio->flags);
380 args.folio = folio;
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700381 args.nr_pages = readahead_count(rac);
382 args.bio = do_mpage_readpage(&args);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 }
Jens Axboe357c1202018-08-17 15:45:32 -0700384 if (args.bio)
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +0200385 mpage_bio_submit_read(args.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386}
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700387EXPORT_SYMBOL(mpage_readahead);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389/*
390 * This isn't called much at all
391 */
Matthew Wilcox (Oracle)f132ab72022-04-29 11:47:39 -0400392int mpage_read_folio(struct folio *folio, get_block_t get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393{
Jens Axboe357c1202018-08-17 15:45:32 -0700394 struct mpage_readpage_args args = {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400395 .folio = folio,
Jens Axboe357c1202018-08-17 15:45:32 -0700396 .nr_pages = 1,
397 .get_block = get_block,
Jens Axboe357c1202018-08-17 15:45:32 -0700398 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Jens Axboe357c1202018-08-17 15:45:32 -0700400 args.bio = do_mpage_readpage(&args);
401 if (args.bio)
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +0200402 mpage_bio_submit_read(args.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 return 0;
404}
Matthew Wilcox (Oracle)f132ab72022-04-29 11:47:39 -0400405EXPORT_SYMBOL(mpage_read_folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407/*
408 * Writing is not so simple.
409 *
410 * If the page has buffers then they will be used for obtaining the disk
411 * mapping. We only support pages which are fully mapped-and-dirty, with a
412 * special case for pages which are unmapped at the end: end-of-file.
413 *
414 * If the page has no buffers (preferred) then the page is mapped here.
415 *
416 * If all blocks are found to be contiguous then the page can go into the
417 * BIO. Otherwise fall back to the mapping's writepage().
418 *
419 * FIXME: This code wants an estimate of how many pages are still to be
420 * written, so it can intelligently allocate a suitably-sized BIO. For now,
421 * just allocate full-size (16-page) BIOs.
422 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700423
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300424struct mpage_data {
425 struct bio *bio;
426 sector_t last_block_in_bio;
427 get_block_t *get_block;
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300428};
429
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700430/*
431 * We have our BIO, so we can now mark the buffers clean. Make
432 * sure to only clean buffers which we know we'll be writing.
433 */
434static void clean_buffers(struct page *page, unsigned first_unmapped)
435{
436 unsigned buffer_counter = 0;
437 struct buffer_head *bh, *head;
438 if (!page_has_buffers(page))
439 return;
440 head = page_buffers(page);
441 bh = head;
442
443 do {
444 if (buffer_counter++ == first_unmapped)
445 break;
446 clear_buffer_dirty(bh);
447 bh = bh->b_this_page;
448 } while (bh != head);
449
450 /*
451 * we cannot drop the bh if the page is not uptodate or a concurrent
Matthew Wilcox (Oracle)2c69e202022-04-29 10:40:40 -0400452 * read_folio would fail to serialize with the bh and it would read from
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700453 * disk before we reach the platter.
454 */
455 if (buffer_heads_over_limit && PageUptodate(page))
Matthew Wilcox (Oracle)68189fe2022-05-01 01:08:08 -0400456 try_to_free_buffers(page_folio(page));
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700457}
458
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700459/*
460 * For situations where we want to clean all buffers attached to a page.
461 * We don't need to calculate how many buffers are attached to the page,
462 * we just need to specify a number larger than the maximum number of buffers.
463 */
464void clean_page_buffers(struct page *page)
465{
466 clean_buffers(page, ~0U);
467}
468
Matthew Wilcox (Oracle)d585bdb2023-01-26 20:12:54 +0000469static int __mpage_writepage(struct folio *folio, 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;
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000474 struct address_space *mapping = folio->mapping;
475 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 const unsigned blkbits = inode->i_blkbits;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300477 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 sector_t last_block;
479 sector_t block_in_file;
480 sector_t blocks[MAX_BUF_PER_PAGE];
481 unsigned page_block;
482 unsigned first_unmapped = blocks_per_page;
483 struct block_device *bdev = NULL;
484 int boundary = 0;
485 sector_t boundary_block = 0;
486 struct block_device *boundary_bdev = NULL;
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000487 size_t length;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 struct buffer_head map_bh;
489 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700490 int ret = 0;
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000491 struct buffer_head *head = folio_buffers(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000493 if (head) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 struct buffer_head *bh = head;
495
496 /* If they're all mapped and dirty, do it */
497 page_block = 0;
498 do {
499 BUG_ON(buffer_locked(bh));
500 if (!buffer_mapped(bh)) {
501 /*
502 * unmapped dirty buffers are created by
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +0000503 * block_dirty_folio -> mmapped data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 */
505 if (buffer_dirty(bh))
506 goto confused;
507 if (first_unmapped == blocks_per_page)
508 first_unmapped = page_block;
509 continue;
510 }
511
512 if (first_unmapped != blocks_per_page)
513 goto confused; /* hole -> non-hole */
514
515 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
516 goto confused;
517 if (page_block) {
518 if (bh->b_blocknr != blocks[page_block-1] + 1)
519 goto confused;
520 }
521 blocks[page_block++] = bh->b_blocknr;
522 boundary = buffer_boundary(bh);
523 if (boundary) {
524 boundary_block = bh->b_blocknr;
525 boundary_bdev = bh->b_bdev;
526 }
527 bdev = bh->b_bdev;
528 } while ((bh = bh->b_this_page) != head);
529
530 if (first_unmapped)
531 goto page_is_mapped;
532
533 /*
534 * Page has buffers, but they are all unmapped. The page was
535 * created by pagein or read over a hole which was handled by
Matthew Wilcox (Oracle)2c69e202022-04-29 10:40:40 -0400536 * block_read_full_folio(). If this address_space is also
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700537 * using mpage_readahead then this can rarely happen.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 */
539 goto confused;
540 }
541
542 /*
543 * The page has no buffers: map it to disk
544 */
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000545 BUG_ON(!folio_test_uptodate(folio));
546 block_in_file = (sector_t)folio->index << (PAGE_SHIFT - blkbits);
Jan Kara4b89a372023-01-03 11:44:30 +0100547 /*
548 * Whole page beyond EOF? Skip allocating blocks to avoid leaking
549 * space.
550 */
551 if (block_in_file >= (i_size + (1 << blkbits) - 1) >> blkbits)
552 goto page_is_mapped;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 last_block = (i_size - 1) >> blkbits;
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000554 map_bh.b_folio = folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 for (page_block = 0; page_block < blocks_per_page; ) {
556
557 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800558 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700559 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 goto confused;
Jan Kara70108392023-01-25 14:59:14 +0100561 if (!buffer_mapped(&map_bh))
562 goto confused;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 if (buffer_new(&map_bh))
Jan Karae64855c2016-11-04 18:08:15 +0100564 clean_bdev_bh_alias(&map_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 if (buffer_boundary(&map_bh)) {
566 boundary_block = map_bh.b_blocknr;
567 boundary_bdev = map_bh.b_bdev;
568 }
569 if (page_block) {
570 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
571 goto confused;
572 }
573 blocks[page_block++] = map_bh.b_blocknr;
574 boundary = buffer_boundary(&map_bh);
575 bdev = map_bh.b_bdev;
576 if (block_in_file == last_block)
577 break;
578 block_in_file++;
579 }
580 BUG_ON(page_block == 0);
581
582 first_unmapped = page_block;
583
584page_is_mapped:
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000585 /* Don't bother writing beyond EOF, truncate will discard the folio */
586 if (folio_pos(folio) >= i_size)
587 goto confused;
588 length = folio_size(folio);
589 if (folio_pos(folio) + length > i_size) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 /*
591 * The page straddles i_size. It must be zeroed out on each
Adam Buchbinder2a61aa42009-12-11 16:35:40 -0500592 * and every writepage invocation because it may be mmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 * "A file is mapped in multiples of the page size. For a file
594 * that is not a multiple of the page size, the remaining memory
595 * is zeroed when mapped, and writes to that region are not
596 * written out to the file."
597 */
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000598 length = i_size - folio_pos(folio);
599 folio_zero_segment(folio, length, folio_size(folio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 }
601
602 /*
603 * This page will go to BIO. Do we need to send this BIO off first?
604 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700605 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +0200606 bio = mpage_bio_submit_write(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
608alloc_new:
609 if (bio == NULL) {
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100610 bio = bio_alloc(bdev, BIO_MAX_VECS,
611 REQ_OP_WRITE | wbc_to_write_flags(wbc),
612 GFP_NOFS);
Christoph Hellwigd5f68a42022-01-24 10:10:49 +0100613 bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
Tejun Heob16b1de2015-06-02 08:39:48 -0600614 wbc_init_bio(wbc, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 }
616
617 /*
618 * Must try to add the page before marking the buffer clean or
619 * the confused fail path above (OOM) will be very confused when
620 * it finds all bh marked clean (i.e. it will not write anything)
621 */
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000622 wbc_account_cgroup_owner(wbc, &folio->page, folio_size(folio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 length = first_unmapped << blkbits;
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000624 if (!bio_add_folio(bio, folio, length, 0)) {
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +0200625 bio = mpage_bio_submit_write(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 goto alloc_new;
627 }
628
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000629 clean_buffers(&folio->page, first_unmapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000631 BUG_ON(folio_test_writeback(folio));
632 folio_start_writeback(folio);
633 folio_unlock(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 if (boundary || (first_unmapped != blocks_per_page)) {
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +0200635 bio = mpage_bio_submit_write(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 if (boundary_block) {
637 write_boundary_block(boundary_bdev,
638 boundary_block, 1 << blkbits);
639 }
640 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700641 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 }
643 goto out;
644
645confused:
646 if (bio)
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +0200647 bio = mpage_bio_submit_write(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 /*
650 * The caller has a ref on the inode, so *mapping is stable
651 */
Matthew Wilcox (Oracle)9160cff2023-01-26 20:12:55 +0000652 ret = block_write_full_page(&folio->page, mpd->get_block, wbc);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700653 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700655 mpd->bio = bio;
656 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657}
658
659/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800660 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 * @mapping: address space structure to write
662 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
663 * @get_block: the filesystem's block mapper function.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 *
665 * This is a library function, which implements the writepages()
666 * address_space_operation.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 */
668int
669mpage_writepages(struct address_space *mapping,
670 struct writeback_control *wbc, get_block_t get_block)
671{
Christoph Hellwigcf5e7a62022-06-13 07:37:15 +0200672 struct mpage_data mpd = {
673 .get_block = get_block,
674 };
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200675 struct blk_plug plug;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700676 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200678 blk_start_plug(&plug);
Christoph Hellwigcf5e7a62022-06-13 07:37:15 +0200679 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
680 if (mpd.bio)
Pankaj Raghavf0d6ca42023-04-11 14:29:19 +0200681 mpage_bio_submit_write(mpd.bio);
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200682 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 return ret;
684}
685EXPORT_SYMBOL(mpage_writepages);