blob: 0f8ae954a579035253109dba0aa9592527be0964 [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 */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +020046static void mpage_end_io(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070047{
Kent Overstreet2c30c712013-11-07 12:20:26 -080048 struct bio_vec *bv;
Ming Lei6dc4f102019-02-15 19:13:19 +080049 struct bvec_iter_all iter_all;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Christoph Hellwig2b070cf2019-04-25 09:03:00 +020051 bio_for_each_segment_all(bv, bio, iter_all) {
Kent Overstreet2c30c712013-11-07 12:20:26 -080052 struct page *page = bv->bv_page;
Tejun Heo3f289dc2018-07-18 04:47:36 -070053 page_endio(page, bio_op(bio),
54 blk_status_to_errno(bio->bi_status));
Kent Overstreet2c30c712013-11-07 12:20:26 -080055 }
56
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058}
59
Christoph Hellwig77c436d2022-02-22 16:46:32 +010060static struct bio *mpage_bio_submit(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -070061{
Hai Shanc32b0d42011-01-13 15:45:51 -080062 bio->bi_end_io = mpage_end_io;
Ming Lei83c9c542020-01-05 09:41:14 +080063 guard_bio_eod(bio);
Mike Christie4e49ea42016-06-05 14:31:41 -050064 submit_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 return NULL;
66}
67
Linus Torvalds1da177e2005-04-16 15:20:36 -070068/*
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -070069 * support function for mpage_readahead. The fs supplied get_block might
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 * return an up to date buffer. This is used to map that buffer into
Matthew Wilcox (Oracle)2c69e202022-04-29 10:40:40 -040071 * the page, which allows read_folio to avoid triggering a duplicate call
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 * to get_block.
73 *
74 * The idea is to avoid adding buffers to pages that don't already have
75 * them. So when the buffer is up to date and the page size == block size,
76 * this marks the page up to date instead of adding new buffers.
77 */
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -040078static void map_buffer_to_folio(struct folio *folio, struct buffer_head *bh,
79 int page_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080{
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -040081 struct inode *inode = folio->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 struct buffer_head *page_bh, *head;
83 int block = 0;
84
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -040085 head = folio_buffers(folio);
86 if (!head) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 /*
88 * don't make any buffers if there is only one buffer on
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -040089 * the folio and the folio just needs to be set up to date
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030091 if (inode->i_blkbits == PAGE_SHIFT &&
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 buffer_uptodate(bh)) {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -040093 folio_mark_uptodate(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 return;
95 }
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -040096 create_empty_buffers(&folio->page, i_blocksize(inode), 0);
97 head = folio_buffers(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 }
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -040099
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 page_bh = head;
101 do {
102 if (block == page_block) {
103 page_bh->b_state = bh->b_state;
104 page_bh->b_bdev = bh->b_bdev;
105 page_bh->b_blocknr = bh->b_blocknr;
106 break;
107 }
108 page_bh = page_bh->b_this_page;
109 block++;
110 } while (page_bh != head);
111}
112
Jens Axboe357c1202018-08-17 15:45:32 -0700113struct mpage_readpage_args {
114 struct bio *bio;
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400115 struct folio *folio;
Jens Axboe357c1202018-08-17 15:45:32 -0700116 unsigned int nr_pages;
Jens Axboe74c81642018-08-17 15:45:36 -0700117 bool is_readahead;
Jens Axboe357c1202018-08-17 15:45:32 -0700118 sector_t last_block_in_bio;
119 struct buffer_head map_bh;
120 unsigned long first_logical_block;
121 get_block_t *get_block;
Jens Axboe357c1202018-08-17 15:45:32 -0700122};
123
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800124/*
125 * This is the worker routine which does all the work of mapping the disk
126 * blocks and constructs largest possible bios, submits them for IO if the
127 * blocks are not contiguous on the disk.
128 *
129 * We pass a buffer_head back and forth and use its buffer_mapped() flag to
130 * represent the validity of its disk mapping and to decide when to do the next
131 * get_block() call.
132 */
Jens Axboe357c1202018-08-17 15:45:32 -0700133static struct bio *do_mpage_readpage(struct mpage_readpage_args *args)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134{
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400135 struct folio *folio = args->folio;
136 struct inode *inode = folio->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 const unsigned blkbits = inode->i_blkbits;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300138 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 const unsigned blocksize = 1 << blkbits;
Jens Axboe357c1202018-08-17 15:45:32 -0700140 struct buffer_head *map_bh = &args->map_bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 sector_t block_in_file;
142 sector_t last_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800143 sector_t last_block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 sector_t blocks[MAX_BUF_PER_PAGE];
145 unsigned page_block;
146 unsigned first_hole = blocks_per_page;
147 struct block_device *bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 int length;
149 int fully_mapped = 1;
Bart Van Asschef84c94a2022-07-14 11:07:15 -0700150 blk_opf_t opf = REQ_OP_READ;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800151 unsigned nblocks;
152 unsigned relative_block;
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400153 gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
154
155 /* MAX_BUF_PER_PAGE, for example */
156 VM_BUG_ON_FOLIO(folio_test_large(folio), folio);
Jens Axboe74c81642018-08-17 15:45:36 -0700157
158 if (args->is_readahead) {
Bart Van Asschef84c94a2022-07-14 11:07:15 -0700159 opf |= REQ_RAHEAD;
Christoph Hellwig61285ff2022-03-23 16:39:52 +0100160 gfp |= __GFP_NORETRY | __GFP_NOWARN;
Jens Axboe74c81642018-08-17 15:45:36 -0700161 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400163 if (folio_buffers(folio))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 goto confused;
165
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400166 block_in_file = (sector_t)folio->index << (PAGE_SHIFT - blkbits);
Jens Axboe357c1202018-08-17 15:45:32 -0700167 last_block = block_in_file + args->nr_pages * blocks_per_page;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800168 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
169 if (last_block > last_block_in_file)
170 last_block = last_block_in_file;
171 page_block = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800173 /*
174 * Map blocks using the result from the previous get_blocks call first.
175 */
176 nblocks = map_bh->b_size >> blkbits;
Jens Axboe357c1202018-08-17 15:45:32 -0700177 if (buffer_mapped(map_bh) &&
178 block_in_file > args->first_logical_block &&
179 block_in_file < (args->first_logical_block + nblocks)) {
180 unsigned map_offset = block_in_file - args->first_logical_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800181 unsigned last = nblocks - map_offset;
182
183 for (relative_block = 0; ; relative_block++) {
184 if (relative_block == last) {
185 clear_buffer_mapped(map_bh);
186 break;
187 }
188 if (page_block == blocks_per_page)
189 break;
190 blocks[page_block] = map_bh->b_blocknr + map_offset +
191 relative_block;
192 page_block++;
193 block_in_file++;
194 }
195 bdev = map_bh->b_bdev;
196 }
197
198 /*
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400199 * Then do more get_blocks calls until we are done with this folio.
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800200 */
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400201 map_bh->b_page = &folio->page;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800202 while (page_block < blocks_per_page) {
203 map_bh->b_state = 0;
204 map_bh->b_size = 0;
205
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 if (block_in_file < last_block) {
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800207 map_bh->b_size = (last_block-block_in_file) << blkbits;
Jens Axboe357c1202018-08-17 15:45:32 -0700208 if (args->get_block(inode, block_in_file, map_bh, 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 goto confused;
Jens Axboe357c1202018-08-17 15:45:32 -0700210 args->first_logical_block = block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 }
212
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800213 if (!buffer_mapped(map_bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 fully_mapped = 0;
215 if (first_hole == blocks_per_page)
216 first_hole = page_block;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800217 page_block++;
218 block_in_file++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 continue;
220 }
221
222 /* some filesystems will copy data into the page during
223 * the get_block call, in which case we don't want to
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400224 * read it again. map_buffer_to_folio copies the data
225 * we just collected from get_block into the folio's buffers
226 * so read_folio doesn't have to repeat the get_block call
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800228 if (buffer_uptodate(map_bh)) {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400229 map_buffer_to_folio(folio, map_bh, page_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 goto confused;
231 }
232
233 if (first_hole != blocks_per_page)
234 goto confused; /* hole -> non-hole */
235
236 /* Contiguous blocks? */
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800237 if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 goto confused;
Badari Pulavartyfa30bd02006-03-26 01:38:01 -0800239 nblocks = map_bh->b_size >> blkbits;
240 for (relative_block = 0; ; relative_block++) {
241 if (relative_block == nblocks) {
242 clear_buffer_mapped(map_bh);
243 break;
244 } else if (page_block == blocks_per_page)
245 break;
246 blocks[page_block] = map_bh->b_blocknr+relative_block;
247 page_block++;
248 block_in_file++;
249 }
250 bdev = map_bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 }
252
253 if (first_hole != blocks_per_page) {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400254 folio_zero_segment(folio, first_hole << blkbits, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 if (first_hole == 0) {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400256 folio_mark_uptodate(folio);
257 folio_unlock(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 goto out;
259 }
260 } else if (fully_mapped) {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400261 folio_set_mappedtodisk(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 }
263
264 /*
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400265 * This folio will go to BIO. Do we need to send this BIO off first?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 */
Jens Axboe357c1202018-08-17 15:45:32 -0700267 if (args->bio && (args->last_block_in_bio != blocks[0] - 1))
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100268 args->bio = mpage_bio_submit(args->bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
270alloc_new:
Jens Axboe357c1202018-08-17 15:45:32 -0700271 if (args->bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700272 if (first_hole == blocks_per_page) {
273 if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9),
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400274 &folio->page))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700275 goto out;
276 }
Bart Van Asschef84c94a2022-07-14 11:07:15 -0700277 args->bio = bio_alloc(bdev, bio_max_segs(args->nr_pages), opf,
Christoph Hellwig07888c662022-01-24 10:11:05 +0100278 gfp);
Jens Axboe357c1202018-08-17 15:45:32 -0700279 if (args->bio == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 goto confused;
Christoph Hellwigd5f68a42022-01-24 10:10:49 +0100281 args->bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 }
283
284 length = first_hole << blkbits;
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400285 if (!bio_add_folio(args->bio, folio, length, 0)) {
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100286 args->bio = mpage_bio_submit(args->bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 goto alloc_new;
288 }
289
Jens Axboe357c1202018-08-17 15:45:32 -0700290 relative_block = block_in_file - args->first_logical_block;
Miquel van Smoorenburg38c8e612009-01-06 14:39:02 -0800291 nblocks = map_bh->b_size >> blkbits;
292 if ((buffer_boundary(map_bh) && relative_block == nblocks) ||
293 (first_hole != blocks_per_page))
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100294 args->bio = mpage_bio_submit(args->bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 else
Jens Axboe357c1202018-08-17 15:45:32 -0700296 args->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297out:
Jens Axboe357c1202018-08-17 15:45:32 -0700298 return args->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
300confused:
Jens Axboe357c1202018-08-17 15:45:32 -0700301 if (args->bio)
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100302 args->bio = mpage_bio_submit(args->bio);
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400303 if (!folio_test_uptodate(folio))
304 block_read_full_folio(folio, args->get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 else
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400306 folio_unlock(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 goto out;
308}
309
Martin Waitz67be2dd2005-05-01 08:59:26 -0700310/**
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700311 * mpage_readahead - start reads against pages
312 * @rac: Describes which pages to read.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700313 * @get_block: The filesystem's block mapper function.
314 *
315 * This function walks the pages and the blocks within each page, building and
316 * emitting large BIOs.
317 *
318 * If anything unusual happens, such as:
319 *
320 * - encountering a page which has buffers
321 * - encountering a page which has a non-hole after a hole
322 * - encountering a page with non-contiguous blocks
323 *
324 * then this code just gives up and calls the buffer_head-based read function.
325 * It does handle a page which has holes at the end - that is a common case:
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +0300326 * the end-of-file on blocksize < PAGE_SIZE setups.
Martin Waitz67be2dd2005-05-01 08:59:26 -0700327 *
328 * BH_Boundary explanation:
329 *
330 * There is a problem. The mpage read code assembles several pages, gets all
331 * their disk mappings, and then submits them all. That's fine, but obtaining
332 * the disk mappings may require I/O. Reads of indirect blocks, for example.
333 *
334 * So an mpage read of the first 16 blocks of an ext2 file will cause I/O to be
335 * submitted in the following order:
Mauro Carvalho Chehab0117d422017-05-12 07:45:42 -0300336 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700337 * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16
Randy Dunlap78a4a502008-02-29 22:02:31 -0800338 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700339 * because the indirect block has to be read to get the mappings of blocks
340 * 13,14,15,16. Obviously, this impacts performance.
341 *
342 * So what we do it to allow the filesystem's get_block() function to set
343 * BH_Boundary when it maps block 11. BH_Boundary says: mapping of the block
344 * after this one will require I/O against a block which is probably close to
345 * this one. So you should push what I/O you have currently accumulated.
346 *
347 * This all causes the disk requests to be issued in the correct order.
348 */
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700349void mpage_readahead(struct readahead_control *rac, get_block_t get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350{
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400351 struct folio *folio;
Jens Axboe357c1202018-08-17 15:45:32 -0700352 struct mpage_readpage_args args = {
353 .get_block = get_block,
Jens Axboe74c81642018-08-17 15:45:36 -0700354 .is_readahead = true,
Jens Axboe357c1202018-08-17 15:45:32 -0700355 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400357 while ((folio = readahead_folio(rac))) {
358 prefetchw(&folio->flags);
359 args.folio = folio;
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700360 args.nr_pages = readahead_count(rac);
361 args.bio = do_mpage_readpage(&args);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 }
Jens Axboe357c1202018-08-17 15:45:32 -0700363 if (args.bio)
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100364 mpage_bio_submit(args.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365}
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700366EXPORT_SYMBOL(mpage_readahead);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
368/*
369 * This isn't called much at all
370 */
Matthew Wilcox (Oracle)f132ab72022-04-29 11:47:39 -0400371int mpage_read_folio(struct folio *folio, get_block_t get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372{
Jens Axboe357c1202018-08-17 15:45:32 -0700373 struct mpage_readpage_args args = {
Matthew Wilcox (Oracle)211d0442022-06-10 22:58:24 -0400374 .folio = folio,
Jens Axboe357c1202018-08-17 15:45:32 -0700375 .nr_pages = 1,
376 .get_block = get_block,
Jens Axboe357c1202018-08-17 15:45:32 -0700377 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
Jens Axboe357c1202018-08-17 15:45:32 -0700379 args.bio = do_mpage_readpage(&args);
380 if (args.bio)
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100381 mpage_bio_submit(args.bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 return 0;
383}
Matthew Wilcox (Oracle)f132ab72022-04-29 11:47:39 -0400384EXPORT_SYMBOL(mpage_read_folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
386/*
387 * Writing is not so simple.
388 *
389 * If the page has buffers then they will be used for obtaining the disk
390 * mapping. We only support pages which are fully mapped-and-dirty, with a
391 * special case for pages which are unmapped at the end: end-of-file.
392 *
393 * If the page has no buffers (preferred) then the page is mapped here.
394 *
395 * If all blocks are found to be contiguous then the page can go into the
396 * BIO. Otherwise fall back to the mapping's writepage().
397 *
398 * FIXME: This code wants an estimate of how many pages are still to be
399 * written, so it can intelligently allocate a suitably-sized BIO. For now,
400 * just allocate full-size (16-page) BIOs.
401 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700402
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300403struct mpage_data {
404 struct bio *bio;
405 sector_t last_block_in_bio;
406 get_block_t *get_block;
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300407};
408
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700409/*
410 * We have our BIO, so we can now mark the buffers clean. Make
411 * sure to only clean buffers which we know we'll be writing.
412 */
413static void clean_buffers(struct page *page, unsigned first_unmapped)
414{
415 unsigned buffer_counter = 0;
416 struct buffer_head *bh, *head;
417 if (!page_has_buffers(page))
418 return;
419 head = page_buffers(page);
420 bh = head;
421
422 do {
423 if (buffer_counter++ == first_unmapped)
424 break;
425 clear_buffer_dirty(bh);
426 bh = bh->b_this_page;
427 } while (bh != head);
428
429 /*
430 * we cannot drop the bh if the page is not uptodate or a concurrent
Matthew Wilcox (Oracle)2c69e202022-04-29 10:40:40 -0400431 * read_folio would fail to serialize with the bh and it would read from
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700432 * disk before we reach the platter.
433 */
434 if (buffer_heads_over_limit && PageUptodate(page))
Matthew Wilcox (Oracle)68189fe2022-05-01 01:08:08 -0400435 try_to_free_buffers(page_folio(page));
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700436}
437
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700438/*
439 * For situations where we want to clean all buffers attached to a page.
440 * We don't need to calculate how many buffers are attached to the page,
441 * we just need to specify a number larger than the maximum number of buffers.
442 */
443void clean_page_buffers(struct page *page)
444{
445 clean_buffers(page, ~0U);
446}
447
Dmitri Vorobievced117c2009-03-31 00:41:20 +0300448static int __mpage_writepage(struct page *page, struct writeback_control *wbc,
Alex Tomas29a814d2008-07-11 19:27:31 -0400449 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450{
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700451 struct mpage_data *mpd = data;
452 struct bio *bio = mpd->bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 struct address_space *mapping = page->mapping;
454 struct inode *inode = page->mapping->host;
455 const unsigned blkbits = inode->i_blkbits;
456 unsigned long end_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300457 const unsigned blocks_per_page = PAGE_SIZE >> blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 sector_t last_block;
459 sector_t block_in_file;
460 sector_t blocks[MAX_BUF_PER_PAGE];
461 unsigned page_block;
462 unsigned first_unmapped = blocks_per_page;
463 struct block_device *bdev = NULL;
464 int boundary = 0;
465 sector_t boundary_block = 0;
466 struct block_device *boundary_bdev = NULL;
467 int length;
468 struct buffer_head map_bh;
469 loff_t i_size = i_size_read(inode);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700470 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472 if (page_has_buffers(page)) {
473 struct buffer_head *head = page_buffers(page);
474 struct buffer_head *bh = head;
475
476 /* If they're all mapped and dirty, do it */
477 page_block = 0;
478 do {
479 BUG_ON(buffer_locked(bh));
480 if (!buffer_mapped(bh)) {
481 /*
482 * unmapped dirty buffers are created by
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +0000483 * block_dirty_folio -> mmapped data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 */
485 if (buffer_dirty(bh))
486 goto confused;
487 if (first_unmapped == blocks_per_page)
488 first_unmapped = page_block;
489 continue;
490 }
491
492 if (first_unmapped != blocks_per_page)
493 goto confused; /* hole -> non-hole */
494
495 if (!buffer_dirty(bh) || !buffer_uptodate(bh))
496 goto confused;
497 if (page_block) {
498 if (bh->b_blocknr != blocks[page_block-1] + 1)
499 goto confused;
500 }
501 blocks[page_block++] = bh->b_blocknr;
502 boundary = buffer_boundary(bh);
503 if (boundary) {
504 boundary_block = bh->b_blocknr;
505 boundary_bdev = bh->b_bdev;
506 }
507 bdev = bh->b_bdev;
508 } while ((bh = bh->b_this_page) != head);
509
510 if (first_unmapped)
511 goto page_is_mapped;
512
513 /*
514 * Page has buffers, but they are all unmapped. The page was
515 * created by pagein or read over a hole which was handled by
Matthew Wilcox (Oracle)2c69e202022-04-29 10:40:40 -0400516 * block_read_full_folio(). If this address_space is also
Matthew Wilcox (Oracle)d4388342020-06-01 21:47:02 -0700517 * using mpage_readahead then this can rarely happen.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 */
519 goto confused;
520 }
521
522 /*
523 * The page has no buffers: map it to disk
524 */
525 BUG_ON(!PageUptodate(page));
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300526 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 last_block = (i_size - 1) >> blkbits;
528 map_bh.b_page = page;
529 for (page_block = 0; page_block < blocks_per_page; ) {
530
531 map_bh.b_state = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -0800532 map_bh.b_size = 1 << blkbits;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700533 if (mpd->get_block(inode, block_in_file, &map_bh, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 goto confused;
535 if (buffer_new(&map_bh))
Jan Karae64855c2016-11-04 18:08:15 +0100536 clean_bdev_bh_alias(&map_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 if (buffer_boundary(&map_bh)) {
538 boundary_block = map_bh.b_blocknr;
539 boundary_bdev = map_bh.b_bdev;
540 }
541 if (page_block) {
542 if (map_bh.b_blocknr != blocks[page_block-1] + 1)
543 goto confused;
544 }
545 blocks[page_block++] = map_bh.b_blocknr;
546 boundary = buffer_boundary(&map_bh);
547 bdev = map_bh.b_bdev;
548 if (block_in_file == last_block)
549 break;
550 block_in_file++;
551 }
552 BUG_ON(page_block == 0);
553
554 first_unmapped = page_block;
555
556page_is_mapped:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300557 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 if (page->index >= end_index) {
559 /*
560 * The page straddles i_size. It must be zeroed out on each
Adam Buchbinder2a61aa42009-12-11 16:35:40 -0500561 * and every writepage invocation because it may be mmapped.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 * "A file is mapped in multiples of the page size. For a file
563 * that is not a multiple of the page size, the remaining memory
564 * is zeroed when mapped, and writes to that region are not
565 * written out to the file."
566 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300567 unsigned offset = i_size & (PAGE_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568
569 if (page->index > end_index || !offset)
570 goto confused;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300571 zero_user_segment(page, offset, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 }
573
574 /*
575 * This page will go to BIO. Do we need to send this BIO off first?
576 */
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700577 if (bio && mpd->last_block_in_bio != blocks[0] - 1)
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100578 bio = mpage_bio_submit(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580alloc_new:
581 if (bio == NULL) {
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700582 if (first_unmapped == blocks_per_page) {
583 if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9),
Matthew Wilcoxf8927602017-10-13 15:58:15 -0700584 page, wbc))
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700585 goto out;
Matthew Wilcox47a191f2014-06-04 16:07:46 -0700586 }
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100587 bio = bio_alloc(bdev, BIO_MAX_VECS,
588 REQ_OP_WRITE | wbc_to_write_flags(wbc),
589 GFP_NOFS);
Christoph Hellwigd5f68a42022-01-24 10:10:49 +0100590 bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
Tejun Heob16b1de2015-06-02 08:39:48 -0600591 wbc_init_bio(wbc, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 }
593
594 /*
595 * Must try to add the page before marking the buffer clean or
596 * the confused fail path above (OOM) will be very confused when
597 * it finds all bh marked clean (i.e. it will not write anything)
598 */
Tejun Heo34e51a52019-06-27 13:39:49 -0700599 wbc_account_cgroup_owner(wbc, page, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 length = first_unmapped << blkbits;
601 if (bio_add_page(bio, page, length, 0) < length) {
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100602 bio = mpage_bio_submit(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 goto alloc_new;
604 }
605
Matthew Wilcox90768ee2014-06-04 16:07:44 -0700606 clean_buffers(page, first_unmapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
608 BUG_ON(PageWriteback(page));
609 set_page_writeback(page);
610 unlock_page(page);
611 if (boundary || (first_unmapped != blocks_per_page)) {
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100612 bio = mpage_bio_submit(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 if (boundary_block) {
614 write_boundary_block(boundary_bdev,
615 boundary_block, 1 << blkbits);
616 }
617 } else {
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700618 mpd->last_block_in_bio = blocks[blocks_per_page - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 }
620 goto out;
621
622confused:
623 if (bio)
Christoph Hellwig77c436d2022-02-22 16:46:32 +0100624 bio = mpage_bio_submit(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 /*
627 * The caller has a ref on the inode, so *mapping is stable
628 */
Christoph Hellwigf2d3e572022-06-13 07:37:14 +0200629 ret = block_write_full_page(page, mpd->get_block, wbc);
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700630 mapping_set_error(mapping, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631out:
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700632 mpd->bio = bio;
633 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634}
635
636/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800637 * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 * @mapping: address space structure to write
639 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
640 * @get_block: the filesystem's block mapper function.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 *
642 * This is a library function, which implements the writepages()
643 * address_space_operation.
644 *
645 * If a page is already under I/O, generic_writepages() skips it, even
646 * if it's dirty. This is desirable behaviour for memory-cleaning writeback,
647 * but it is INCORRECT for data-integrity system calls such as fsync(). fsync()
648 * and msync() need to guarantee that all the data which was dirty at the time
649 * the call was made get new I/O started against them. If wbc->sync_mode is
650 * WB_SYNC_ALL then we were called for data integrity and we must wait for
651 * existing IO to complete.
652 */
653int
654mpage_writepages(struct address_space *mapping,
655 struct writeback_control *wbc, get_block_t get_block)
656{
Christoph Hellwigcf5e7a62022-06-13 07:37:15 +0200657 struct mpage_data mpd = {
658 .get_block = get_block,
659 };
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200660 struct blk_plug plug;
Miklos Szeredi0ea97182007-05-10 22:22:51 -0700661 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200663 blk_start_plug(&plug);
Christoph Hellwigcf5e7a62022-06-13 07:37:15 +0200664 ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd);
665 if (mpd.bio)
666 mpage_bio_submit(mpd.bio);
Jens Axboe2ed1a6b2010-06-22 12:52:14 +0200667 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 return ret;
669}
670EXPORT_SYMBOL(mpage_writepages);