Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | /* |
| 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 Cami | e1f8e87 | 2008-10-15 22:01:59 -0700 | [diff] [blame] | 10 | * 15May2002 Andrew Morton |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 11 | * 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 Gortmaker | 630d9c4 | 2011-11-16 23:57:37 -0500 | [diff] [blame] | 17 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include <linux/mm.h> |
| 19 | #include <linux/kdev_t.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 20 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | #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 Morton | 02c4363 | 2016-03-15 14:55:15 -0700 | [diff] [blame] | 28 | #include <linux/mm_inline.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include <linux/writeback.h> |
| 30 | #include <linux/backing-dev.h> |
| 31 | #include <linux/pagevec.h> |
Akinobu Mita | 4db96b7 | 2014-10-09 15:26:55 -0700 | [diff] [blame] | 32 | #include "internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | |
| 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) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 39 | * back to block_read_full_folio(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | * |
| 41 | * Why is this? If a page's completion depends on a number of different BIOs |
| 42 | * which can complete in any order (or at the same time) then determining the |
| 43 | * status of that page is hard. See end_buffer_async_read() for the details. |
| 44 | * There is no point in duplicating all that complexity. |
| 45 | */ |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 46 | static void mpage_end_io(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | { |
Kent Overstreet | 2c30c71 | 2013-11-07 12:20:26 -0800 | [diff] [blame] | 48 | struct bio_vec *bv; |
Ming Lei | 6dc4f10 | 2019-02-15 19:13:19 +0800 | [diff] [blame] | 49 | struct bvec_iter_all iter_all; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | |
Christoph Hellwig | 2b070cf | 2019-04-25 09:03:00 +0200 | [diff] [blame] | 51 | bio_for_each_segment_all(bv, bio, iter_all) { |
Kent Overstreet | 2c30c71 | 2013-11-07 12:20:26 -0800 | [diff] [blame] | 52 | struct page *page = bv->bv_page; |
Tejun Heo | 3f289dc | 2018-07-18 04:47:36 -0700 | [diff] [blame] | 53 | page_endio(page, bio_op(bio), |
| 54 | blk_status_to_errno(bio->bi_status)); |
Kent Overstreet | 2c30c71 | 2013-11-07 12:20:26 -0800 | [diff] [blame] | 55 | } |
| 56 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | bio_put(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | } |
| 59 | |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 60 | static struct bio *mpage_bio_submit(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | { |
Hai Shan | c32b0d4 | 2011-01-13 15:45:51 -0800 | [diff] [blame] | 62 | bio->bi_end_io = mpage_end_io; |
Ming Lei | 83c9c54 | 2020-01-05 09:41:14 +0800 | [diff] [blame] | 63 | guard_bio_eod(bio); |
Mike Christie | 4e49ea4 | 2016-06-05 14:31:41 -0500 | [diff] [blame] | 64 | submit_bio(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | return NULL; |
| 66 | } |
| 67 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | /* |
Matthew Wilcox (Oracle) | d438834 | 2020-06-01 21:47:02 -0700 | [diff] [blame] | 69 | * support function for mpage_readahead. The fs supplied get_block might |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | * return an up to date buffer. This is used to map that buffer into |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 71 | * the page, which allows read_folio to avoid triggering a duplicate call |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 72 | * 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) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 78 | static void map_buffer_to_folio(struct folio *folio, struct buffer_head *bh, |
| 79 | int page_block) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 80 | { |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 81 | struct inode *inode = folio->mapping->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | struct buffer_head *page_bh, *head; |
| 83 | int block = 0; |
| 84 | |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 85 | head = folio_buffers(folio); |
| 86 | if (!head) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 87 | /* |
| 88 | * don't make any buffers if there is only one buffer on |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 89 | * the folio and the folio just needs to be set up to date |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 90 | */ |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 91 | if (inode->i_blkbits == PAGE_SHIFT && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 92 | buffer_uptodate(bh)) { |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 93 | folio_mark_uptodate(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | return; |
| 95 | } |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 96 | create_empty_buffers(&folio->page, i_blocksize(inode), 0); |
| 97 | head = folio_buffers(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 98 | } |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 99 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | 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 Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 113 | struct mpage_readpage_args { |
| 114 | struct bio *bio; |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 115 | struct folio *folio; |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 116 | unsigned int nr_pages; |
Jens Axboe | 74c8164 | 2018-08-17 15:45:36 -0700 | [diff] [blame] | 117 | bool is_readahead; |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 118 | 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 Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 122 | }; |
| 123 | |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 124 | /* |
| 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 Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 133 | static struct bio *do_mpage_readpage(struct mpage_readpage_args *args) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | { |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 135 | struct folio *folio = args->folio; |
| 136 | struct inode *inode = folio->mapping->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | const unsigned blkbits = inode->i_blkbits; |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 138 | const unsigned blocks_per_page = PAGE_SIZE >> blkbits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 139 | const unsigned blocksize = 1 << blkbits; |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 140 | struct buffer_head *map_bh = &args->map_bh; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | sector_t block_in_file; |
| 142 | sector_t last_block; |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 143 | sector_t last_block_in_file; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 148 | int length; |
| 149 | int fully_mapped = 1; |
Bart Van Assche | f84c94a | 2022-07-14 11:07:15 -0700 | [diff] [blame] | 150 | blk_opf_t opf = REQ_OP_READ; |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 151 | unsigned nblocks; |
| 152 | unsigned relative_block; |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 153 | 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 Axboe | 74c8164 | 2018-08-17 15:45:36 -0700 | [diff] [blame] | 157 | |
| 158 | if (args->is_readahead) { |
Bart Van Assche | f84c94a | 2022-07-14 11:07:15 -0700 | [diff] [blame] | 159 | opf |= REQ_RAHEAD; |
Christoph Hellwig | 61285ff | 2022-03-23 16:39:52 +0100 | [diff] [blame] | 160 | gfp |= __GFP_NORETRY | __GFP_NOWARN; |
Jens Axboe | 74c8164 | 2018-08-17 15:45:36 -0700 | [diff] [blame] | 161 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 162 | |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 163 | if (folio_buffers(folio)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 164 | goto confused; |
| 165 | |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 166 | block_in_file = (sector_t)folio->index << (PAGE_SHIFT - blkbits); |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 167 | last_block = block_in_file + args->nr_pages * blocks_per_page; |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 168 | 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 173 | /* |
| 174 | * Map blocks using the result from the previous get_blocks call first. |
| 175 | */ |
| 176 | nblocks = map_bh->b_size >> blkbits; |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 177 | 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 Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 181 | 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) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 199 | * Then do more get_blocks calls until we are done with this folio. |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 200 | */ |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 201 | map_bh->b_page = &folio->page; |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 202 | while (page_block < blocks_per_page) { |
| 203 | map_bh->b_state = 0; |
| 204 | map_bh->b_size = 0; |
| 205 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 206 | if (block_in_file < last_block) { |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 207 | map_bh->b_size = (last_block-block_in_file) << blkbits; |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 208 | if (args->get_block(inode, block_in_file, map_bh, 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 209 | goto confused; |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 210 | args->first_logical_block = block_in_file; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | } |
| 212 | |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 213 | if (!buffer_mapped(map_bh)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 214 | fully_mapped = 0; |
| 215 | if (first_hole == blocks_per_page) |
| 216 | first_hole = page_block; |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 217 | page_block++; |
| 218 | block_in_file++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | 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) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 224 | * 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | */ |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 228 | if (buffer_uptodate(map_bh)) { |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 229 | map_buffer_to_folio(folio, map_bh, page_block); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 230 | goto confused; |
| 231 | } |
| 232 | |
| 233 | if (first_hole != blocks_per_page) |
| 234 | goto confused; /* hole -> non-hole */ |
| 235 | |
| 236 | /* Contiguous blocks? */ |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 237 | if (page_block && blocks[page_block-1] != map_bh->b_blocknr-1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 238 | goto confused; |
Badari Pulavarty | fa30bd0 | 2006-03-26 01:38:01 -0800 | [diff] [blame] | 239 | 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | } |
| 252 | |
| 253 | if (first_hole != blocks_per_page) { |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 254 | folio_zero_segment(folio, first_hole << blkbits, PAGE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 255 | if (first_hole == 0) { |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 256 | folio_mark_uptodate(folio); |
| 257 | folio_unlock(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | goto out; |
| 259 | } |
| 260 | } else if (fully_mapped) { |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 261 | folio_set_mappedtodisk(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 262 | } |
| 263 | |
| 264 | /* |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 265 | * This folio will go to BIO. Do we need to send this BIO off first? |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | */ |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 267 | if (args->bio && (args->last_block_in_bio != blocks[0] - 1)) |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 268 | args->bio = mpage_bio_submit(args->bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 269 | |
| 270 | alloc_new: |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 271 | if (args->bio == NULL) { |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 272 | if (first_hole == blocks_per_page) { |
| 273 | if (!bdev_read_page(bdev, blocks[0] << (blkbits - 9), |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 274 | &folio->page)) |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 275 | goto out; |
| 276 | } |
Bart Van Assche | f84c94a | 2022-07-14 11:07:15 -0700 | [diff] [blame] | 277 | args->bio = bio_alloc(bdev, bio_max_segs(args->nr_pages), opf, |
Christoph Hellwig | 07888c66 | 2022-01-24 10:11:05 +0100 | [diff] [blame] | 278 | gfp); |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 279 | if (args->bio == NULL) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | goto confused; |
Christoph Hellwig | d5f68a4 | 2022-01-24 10:10:49 +0100 | [diff] [blame] | 281 | args->bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 282 | } |
| 283 | |
| 284 | length = first_hole << blkbits; |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 285 | if (!bio_add_folio(args->bio, folio, length, 0)) { |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 286 | args->bio = mpage_bio_submit(args->bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | goto alloc_new; |
| 288 | } |
| 289 | |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 290 | relative_block = block_in_file - args->first_logical_block; |
Miquel van Smoorenburg | 38c8e61 | 2009-01-06 14:39:02 -0800 | [diff] [blame] | 291 | nblocks = map_bh->b_size >> blkbits; |
| 292 | if ((buffer_boundary(map_bh) && relative_block == nblocks) || |
| 293 | (first_hole != blocks_per_page)) |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 294 | args->bio = mpage_bio_submit(args->bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 295 | else |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 296 | args->last_block_in_bio = blocks[blocks_per_page - 1]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 297 | out: |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 298 | return args->bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | |
| 300 | confused: |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 301 | if (args->bio) |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 302 | args->bio = mpage_bio_submit(args->bio); |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 303 | if (!folio_test_uptodate(folio)) |
| 304 | block_read_full_folio(folio, args->get_block); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | else |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 306 | folio_unlock(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | goto out; |
| 308 | } |
| 309 | |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 310 | /** |
Matthew Wilcox (Oracle) | d438834 | 2020-06-01 21:47:02 -0700 | [diff] [blame] | 311 | * mpage_readahead - start reads against pages |
| 312 | * @rac: Describes which pages to read. |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 313 | * @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. Shutemov | ea1754a | 2016-04-01 15:29:48 +0300 | [diff] [blame] | 326 | * the end-of-file on blocksize < PAGE_SIZE setups. |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 327 | * |
| 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 Chehab | 0117d42 | 2017-05-12 07:45:42 -0300 | [diff] [blame] | 336 | * |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 337 | * 12 0 1 2 3 4 5 6 7 8 9 10 11 13 14 15 16 |
Randy Dunlap | 78a4a50 | 2008-02-29 22:02:31 -0800 | [diff] [blame] | 338 | * |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 339 | * 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) | d438834 | 2020-06-01 21:47:02 -0700 | [diff] [blame] | 349 | void mpage_readahead(struct readahead_control *rac, get_block_t get_block) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 350 | { |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 351 | struct folio *folio; |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 352 | struct mpage_readpage_args args = { |
| 353 | .get_block = get_block, |
Jens Axboe | 74c8164 | 2018-08-17 15:45:36 -0700 | [diff] [blame] | 354 | .is_readahead = true, |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 355 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 357 | while ((folio = readahead_folio(rac))) { |
| 358 | prefetchw(&folio->flags); |
| 359 | args.folio = folio; |
Matthew Wilcox (Oracle) | d438834 | 2020-06-01 21:47:02 -0700 | [diff] [blame] | 360 | args.nr_pages = readahead_count(rac); |
| 361 | args.bio = do_mpage_readpage(&args); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | } |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 363 | if (args.bio) |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 364 | mpage_bio_submit(args.bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | } |
Matthew Wilcox (Oracle) | d438834 | 2020-06-01 21:47:02 -0700 | [diff] [blame] | 366 | EXPORT_SYMBOL(mpage_readahead); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 367 | |
| 368 | /* |
| 369 | * This isn't called much at all |
| 370 | */ |
Matthew Wilcox (Oracle) | f132ab7 | 2022-04-29 11:47:39 -0400 | [diff] [blame] | 371 | int mpage_read_folio(struct folio *folio, get_block_t get_block) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | { |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 373 | struct mpage_readpage_args args = { |
Matthew Wilcox (Oracle) | 211d044 | 2022-06-10 22:58:24 -0400 | [diff] [blame] | 374 | .folio = folio, |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 375 | .nr_pages = 1, |
| 376 | .get_block = get_block, |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 377 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 378 | |
Jens Axboe | 357c120 | 2018-08-17 15:45:32 -0700 | [diff] [blame] | 379 | args.bio = do_mpage_readpage(&args); |
| 380 | if (args.bio) |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 381 | mpage_bio_submit(args.bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | return 0; |
| 383 | } |
Matthew Wilcox (Oracle) | f132ab7 | 2022-04-29 11:47:39 -0400 | [diff] [blame] | 384 | EXPORT_SYMBOL(mpage_read_folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | |
| 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 Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 402 | |
Dmitri Vorobiev | ced117c | 2009-03-31 00:41:20 +0300 | [diff] [blame] | 403 | struct mpage_data { |
| 404 | struct bio *bio; |
| 405 | sector_t last_block_in_bio; |
| 406 | get_block_t *get_block; |
Dmitri Vorobiev | ced117c | 2009-03-31 00:41:20 +0300 | [diff] [blame] | 407 | }; |
| 408 | |
Matthew Wilcox | 90768ee | 2014-06-04 16:07:44 -0700 | [diff] [blame] | 409 | /* |
| 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 | */ |
| 413 | static 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) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 431 | * read_folio would fail to serialize with the bh and it would read from |
Matthew Wilcox | 90768ee | 2014-06-04 16:07:44 -0700 | [diff] [blame] | 432 | * disk before we reach the platter. |
| 433 | */ |
| 434 | if (buffer_heads_over_limit && PageUptodate(page)) |
Matthew Wilcox (Oracle) | 68189fe | 2022-05-01 01:08:08 -0400 | [diff] [blame] | 435 | try_to_free_buffers(page_folio(page)); |
Matthew Wilcox | 90768ee | 2014-06-04 16:07:44 -0700 | [diff] [blame] | 436 | } |
| 437 | |
Matthew Wilcox | f892760 | 2017-10-13 15:58:15 -0700 | [diff] [blame] | 438 | /* |
| 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 | */ |
| 443 | void clean_page_buffers(struct page *page) |
| 444 | { |
| 445 | clean_buffers(page, ~0U); |
| 446 | } |
| 447 | |
Dmitri Vorobiev | ced117c | 2009-03-31 00:41:20 +0300 | [diff] [blame] | 448 | static int __mpage_writepage(struct page *page, struct writeback_control *wbc, |
Alex Tomas | 29a814d | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 449 | void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 450 | { |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 451 | struct mpage_data *mpd = data; |
| 452 | struct bio *bio = mpd->bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | 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. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 457 | const unsigned blocks_per_page = PAGE_SIZE >> blkbits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 458 | 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 Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 470 | int ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 471 | |
| 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) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 483 | * block_dirty_folio -> mmapped data |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | */ |
| 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) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 516 | * block_read_full_folio(). If this address_space is also |
Matthew Wilcox (Oracle) | d438834 | 2020-06-01 21:47:02 -0700 | [diff] [blame] | 517 | * using mpage_readahead then this can rarely happen. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 518 | */ |
| 519 | goto confused; |
| 520 | } |
| 521 | |
| 522 | /* |
| 523 | * The page has no buffers: map it to disk |
| 524 | */ |
| 525 | BUG_ON(!PageUptodate(page)); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 526 | block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | 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 Pulavarty | b0cf232 | 2006-03-26 01:38:00 -0800 | [diff] [blame] | 532 | map_bh.b_size = 1 << blkbits; |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 533 | if (mpd->get_block(inode, block_in_file, &map_bh, 1)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 534 | goto confused; |
| 535 | if (buffer_new(&map_bh)) |
Jan Kara | e64855c | 2016-11-04 18:08:15 +0100 | [diff] [blame] | 536 | clean_bdev_bh_alias(&map_bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 537 | 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 | |
| 556 | page_is_mapped: |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 557 | end_index = i_size >> PAGE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | if (page->index >= end_index) { |
| 559 | /* |
| 560 | * The page straddles i_size. It must be zeroed out on each |
Adam Buchbinder | 2a61aa4 | 2009-12-11 16:35:40 -0500 | [diff] [blame] | 561 | * and every writepage invocation because it may be mmapped. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | * "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. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 567 | unsigned offset = i_size & (PAGE_SIZE - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 568 | |
| 569 | if (page->index > end_index || !offset) |
| 570 | goto confused; |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 571 | zero_user_segment(page, offset, PAGE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 572 | } |
| 573 | |
| 574 | /* |
| 575 | * This page will go to BIO. Do we need to send this BIO off first? |
| 576 | */ |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 577 | if (bio && mpd->last_block_in_bio != blocks[0] - 1) |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 578 | bio = mpage_bio_submit(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | |
| 580 | alloc_new: |
| 581 | if (bio == NULL) { |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 582 | if (first_unmapped == blocks_per_page) { |
| 583 | if (!bdev_write_page(bdev, blocks[0] << (blkbits - 9), |
Matthew Wilcox | f892760 | 2017-10-13 15:58:15 -0700 | [diff] [blame] | 584 | page, wbc)) |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 585 | goto out; |
Matthew Wilcox | 47a191f | 2014-06-04 16:07:46 -0700 | [diff] [blame] | 586 | } |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 587 | bio = bio_alloc(bdev, BIO_MAX_VECS, |
| 588 | REQ_OP_WRITE | wbc_to_write_flags(wbc), |
| 589 | GFP_NOFS); |
Christoph Hellwig | d5f68a4 | 2022-01-24 10:10:49 +0100 | [diff] [blame] | 590 | bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9); |
Tejun Heo | b16b1de | 2015-06-02 08:39:48 -0600 | [diff] [blame] | 591 | wbc_init_bio(wbc, bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 592 | } |
| 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 Heo | 34e51a5 | 2019-06-27 13:39:49 -0700 | [diff] [blame] | 599 | wbc_account_cgroup_owner(wbc, page, PAGE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | length = first_unmapped << blkbits; |
| 601 | if (bio_add_page(bio, page, length, 0) < length) { |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 602 | bio = mpage_bio_submit(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 603 | goto alloc_new; |
| 604 | } |
| 605 | |
Matthew Wilcox | 90768ee | 2014-06-04 16:07:44 -0700 | [diff] [blame] | 606 | clean_buffers(page, first_unmapped); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | |
| 608 | BUG_ON(PageWriteback(page)); |
| 609 | set_page_writeback(page); |
| 610 | unlock_page(page); |
| 611 | if (boundary || (first_unmapped != blocks_per_page)) { |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 612 | bio = mpage_bio_submit(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 613 | if (boundary_block) { |
| 614 | write_boundary_block(boundary_bdev, |
| 615 | boundary_block, 1 << blkbits); |
| 616 | } |
| 617 | } else { |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 618 | mpd->last_block_in_bio = blocks[blocks_per_page - 1]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 619 | } |
| 620 | goto out; |
| 621 | |
| 622 | confused: |
| 623 | if (bio) |
Christoph Hellwig | 77c436d | 2022-02-22 16:46:32 +0100 | [diff] [blame] | 624 | bio = mpage_bio_submit(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 626 | /* |
| 627 | * The caller has a ref on the inode, so *mapping is stable |
| 628 | */ |
Christoph Hellwig | f2d3e57 | 2022-06-13 07:37:14 +0200 | [diff] [blame] | 629 | ret = block_write_full_page(page, mpd->get_block, wbc); |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 630 | mapping_set_error(mapping, ret); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 631 | out: |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 632 | mpd->bio = bio; |
| 633 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 634 | } |
| 635 | |
| 636 | /** |
Randy Dunlap | 78a4a50 | 2008-02-29 22:02:31 -0800 | [diff] [blame] | 637 | * mpage_writepages - walk the list of dirty pages of the given address space & writepage() all of them |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 638 | * @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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | * |
| 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 | */ |
| 653 | int |
| 654 | mpage_writepages(struct address_space *mapping, |
| 655 | struct writeback_control *wbc, get_block_t get_block) |
| 656 | { |
Christoph Hellwig | cf5e7a6 | 2022-06-13 07:37:15 +0200 | [diff] [blame] | 657 | struct mpage_data mpd = { |
| 658 | .get_block = get_block, |
| 659 | }; |
Jens Axboe | 2ed1a6b | 2010-06-22 12:52:14 +0200 | [diff] [blame] | 660 | struct blk_plug plug; |
Miklos Szeredi | 0ea9718 | 2007-05-10 22:22:51 -0700 | [diff] [blame] | 661 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 662 | |
Jens Axboe | 2ed1a6b | 2010-06-22 12:52:14 +0200 | [diff] [blame] | 663 | blk_start_plug(&plug); |
Christoph Hellwig | cf5e7a6 | 2022-06-13 07:37:15 +0200 | [diff] [blame] | 664 | ret = write_cache_pages(mapping, wbc, __mpage_writepage, &mpd); |
| 665 | if (mpd.bio) |
| 666 | mpage_bio_submit(mpd.bio); |
Jens Axboe | 2ed1a6b | 2010-06-22 12:52:14 +0200 | [diff] [blame] | 667 | blk_finish_plug(&plug); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | return ret; |
| 669 | } |
| 670 | EXPORT_SYMBOL(mpage_writepages); |