Thomas Gleixner | 457c899 | 2019-05-19 13:08:55 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | /* |
| 3 | * mm/readahead.c - address_space-level file readahead. |
| 4 | * |
| 5 | * Copyright (C) 2002, Linus Torvalds |
| 6 | * |
Francois Cami | e1f8e87 | 2008-10-15 22:01:59 -0700 | [diff] [blame] | 7 | * 09Apr2002 Andrew Morton |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 8 | * Initial version. |
| 9 | */ |
| 10 | |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 11 | /** |
| 12 | * DOC: Readahead Overview |
| 13 | * |
| 14 | * Readahead is used to read content into the page cache before it is |
| 15 | * explicitly requested by the application. Readahead only ever |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 16 | * attempts to read folios that are not yet in the page cache. If a |
| 17 | * folio is present but not up-to-date, readahead will not try to read |
Matthew Wilcox (Oracle) | 5efe744 | 2022-04-29 08:43:23 -0400 | [diff] [blame] | 18 | * it. In that case a simple ->read_folio() will be requested. |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 19 | * |
| 20 | * Readahead is triggered when an application read request (whether a |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 21 | * system call or a page fault) finds that the requested folio is not in |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 22 | * the page cache, or that it is in the page cache and has the |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 23 | * readahead flag set. This flag indicates that the folio was read |
| 24 | * as part of a previous readahead request and now that it has been |
| 25 | * accessed, it is time for the next readahead. |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 26 | * |
| 27 | * Each readahead request is partly synchronous read, and partly async |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 28 | * readahead. This is reflected in the struct file_ra_state which |
| 29 | * contains ->size being the total number of pages, and ->async_size |
| 30 | * which is the number of pages in the async section. The readahead |
| 31 | * flag will be set on the first folio in this async section to trigger |
| 32 | * a subsequent readahead. Once a series of sequential reads has been |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 33 | * established, there should be no need for a synchronous component and |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 34 | * all readahead request will be fully asynchronous. |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 35 | * |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 36 | * When either of the triggers causes a readahead, three numbers need |
| 37 | * to be determined: the start of the region to read, the size of the |
| 38 | * region, and the size of the async tail. |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 39 | * |
| 40 | * The start of the region is simply the first page address at or after |
| 41 | * the accessed address, which is not currently populated in the page |
| 42 | * cache. This is found with a simple search in the page cache. |
| 43 | * |
| 44 | * The size of the async tail is determined by subtracting the size that |
| 45 | * was explicitly requested from the determined request size, unless |
| 46 | * this would be less than zero - then zero is used. NOTE THIS |
| 47 | * CALCULATION IS WRONG WHEN THE START OF THE REGION IS NOT THE ACCESSED |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 48 | * PAGE. ALSO THIS CALCULATION IS NOT USED CONSISTENTLY. |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 49 | * |
| 50 | * The size of the region is normally determined from the size of the |
| 51 | * previous readahead which loaded the preceding pages. This may be |
| 52 | * discovered from the struct file_ra_state for simple sequential reads, |
| 53 | * or from examining the state of the page cache when multiple |
| 54 | * sequential reads are interleaved. Specifically: where the readahead |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 55 | * was triggered by the readahead flag, the size of the previous |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 56 | * readahead is assumed to be the number of pages from the triggering |
| 57 | * page to the start of the new readahead. In these cases, the size of |
| 58 | * the previous readahead is scaled, often doubled, for the new |
| 59 | * readahead, though see get_next_ra_size() for details. |
| 60 | * |
| 61 | * If the size of the previous read cannot be determined, the number of |
| 62 | * preceding pages in the page cache is used to estimate the size of |
| 63 | * a previous read. This estimate could easily be misled by random |
| 64 | * reads being coincidentally adjacent, so it is ignored unless it is |
| 65 | * larger than the current request, and it is not scaled up, unless it |
| 66 | * is at the start of file. |
| 67 | * |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 68 | * In general readahead is accelerated at the start of the file, as |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 69 | * reads from there are often sequential. There are other minor |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 70 | * adjustments to the readahead size in various special cases and these |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 71 | * are best discovered by reading the code. |
| 72 | * |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 73 | * The above calculation, based on the previous readahead size, |
| 74 | * determines the size of the readahead, to which any requested read |
| 75 | * size may be added. |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 76 | * |
| 77 | * Readahead requests are sent to the filesystem using the ->readahead() |
| 78 | * address space operation, for which mpage_readahead() is a canonical |
| 79 | * implementation. ->readahead() should normally initiate reads on all |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 80 | * folios, but may fail to read any or all folios without causing an I/O |
Matthew Wilcox (Oracle) | 5efe744 | 2022-04-29 08:43:23 -0400 | [diff] [blame] | 81 | * error. The page cache reading code will issue a ->read_folio() request |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 82 | * for any folio which ->readahead() did not read, and only an error |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 83 | * from this will be final. |
| 84 | * |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 85 | * ->readahead() will generally call readahead_folio() repeatedly to get |
| 86 | * each folio from those prepared for readahead. It may fail to read a |
| 87 | * folio by: |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 88 | * |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 89 | * * not calling readahead_folio() sufficiently many times, effectively |
| 90 | * ignoring some folios, as might be appropriate if the path to |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 91 | * storage is congested. |
| 92 | * |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 93 | * * failing to actually submit a read request for a given folio, |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 94 | * possibly due to insufficient resources, or |
| 95 | * |
| 96 | * * getting an error during subsequent processing of a request. |
| 97 | * |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 98 | * In the last two cases, the folio should be unlocked by the filesystem |
| 99 | * to indicate that the read attempt has failed. In the first case the |
| 100 | * folio will be unlocked by the VFS. |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 101 | * |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 102 | * Those folios not in the final ``async_size`` of the request should be |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 103 | * considered to be important and ->readahead() should not fail them due |
| 104 | * to congestion or temporary resource unavailability, but should wait |
| 105 | * for necessary resources (e.g. memory or indexing information) to |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 106 | * become available. Folios in the final ``async_size`` may be |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 107 | * considered less urgent and failure to read them is more acceptable. |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 108 | * In this case it is best to use filemap_remove_folio() to remove the |
| 109 | * folios from the page cache as is automatically done for folios that |
| 110 | * were not fetched with readahead_folio(). This will allow a |
| 111 | * subsequent synchronous readahead request to try them again. If they |
NeilBrown | 9fd472a | 2022-03-22 14:38:54 -0700 | [diff] [blame] | 112 | * are left in the page cache, then they will be read individually using |
Matthew Wilcox (Oracle) | 5efe744 | 2022-04-29 08:43:23 -0400 | [diff] [blame] | 113 | * ->read_folio() which may be less efficient. |
NeilBrown | 84dacdb | 2022-03-22 14:38:51 -0700 | [diff] [blame] | 114 | */ |
| 115 | |
Christoph Hellwig | c97ab27 | 2022-04-20 06:27:19 +0200 | [diff] [blame] | 116 | #include <linux/blkdev.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | #include <linux/kernel.h> |
Ross Zwisler | 11bd969 | 2016-08-25 15:17:17 -0700 | [diff] [blame] | 118 | #include <linux/dax.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 119 | #include <linux/gfp.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 120 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 121 | #include <linux/backing-dev.h> |
Andrew Morton | 8bde37f | 2006-12-10 02:19:40 -0800 | [diff] [blame] | 122 | #include <linux/task_io_accounting_ops.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | #include <linux/pagevec.h> |
Jens Axboe | f5ff842 | 2007-09-21 09:19:54 +0200 | [diff] [blame] | 124 | #include <linux/pagemap.h> |
Christoph Hellwig | 1760424 | 2022-09-15 10:41:56 +0100 | [diff] [blame] | 125 | #include <linux/psi.h> |
Cong Wang | 782182e | 2012-05-29 15:06:43 -0700 | [diff] [blame] | 126 | #include <linux/syscalls.h> |
| 127 | #include <linux/file.h> |
Geliang Tang | d72ee91 | 2016-01-14 15:22:01 -0800 | [diff] [blame] | 128 | #include <linux/mm_inline.h> |
Josef Bacik | ca47e8c | 2018-07-03 11:15:03 -0400 | [diff] [blame] | 129 | #include <linux/blk-cgroup.h> |
Amir Goldstein | 3d8f761 | 2018-08-29 08:41:29 +0300 | [diff] [blame] | 130 | #include <linux/fadvise.h> |
Matthew Wilcox (Oracle) | f2c817b | 2020-06-01 21:46:58 -0700 | [diff] [blame] | 131 | #include <linux/sched/mm.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 132 | |
Fabian Frederick | 29f175d | 2014-04-07 15:37:55 -0700 | [diff] [blame] | 133 | #include "internal.h" |
| 134 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | /* |
| 136 | * Initialise a struct file's readahead state. Assumes that the caller has |
| 137 | * memset *ra to zero. |
| 138 | */ |
| 139 | void |
| 140 | file_ra_state_init(struct file_ra_state *ra, struct address_space *mapping) |
| 141 | { |
Christoph Hellwig | de1414a | 2015-01-14 10:42:36 +0100 | [diff] [blame] | 142 | ra->ra_pages = inode_to_bdi(mapping->host)->ra_pages; |
Fengguang Wu | f4e6b49 | 2007-10-16 01:24:33 -0700 | [diff] [blame] | 143 | ra->prev_pos = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | } |
Steven Whitehouse | d41cc70 | 2006-01-30 08:53:33 +0000 | [diff] [blame] | 145 | EXPORT_SYMBOL_GPL(file_ra_state_init); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 146 | |
Christoph Hellwig | b4e089d7 | 2022-03-31 05:35:55 -0700 | [diff] [blame] | 147 | static void read_pages(struct readahead_control *rac) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 148 | { |
Matthew Wilcox (Oracle) | a4d9653 | 2020-06-01 21:46:25 -0700 | [diff] [blame] | 149 | const struct address_space_operations *aops = rac->mapping->a_ops; |
Matthew Wilcox (Oracle) | a42634a | 2022-03-31 14:15:59 -0400 | [diff] [blame] | 150 | struct folio *folio; |
Jens Axboe | 5b417b1 | 2010-04-19 10:04:38 +0200 | [diff] [blame] | 151 | struct blk_plug plug; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | |
Matthew Wilcox (Oracle) | a4d9653 | 2020-06-01 21:46:25 -0700 | [diff] [blame] | 153 | if (!readahead_count(rac)) |
Christoph Hellwig | b4e089d7 | 2022-03-31 05:35:55 -0700 | [diff] [blame] | 154 | return; |
Matthew Wilcox (Oracle) | ad4ae1c | 2020-06-01 21:46:18 -0700 | [diff] [blame] | 155 | |
Christoph Hellwig | 1760424 | 2022-09-15 10:41:56 +0100 | [diff] [blame] | 156 | if (unlikely(rac->_workingset)) |
| 157 | psi_memstall_enter(&rac->_pflags); |
Jens Axboe | 5b417b1 | 2010-04-19 10:04:38 +0200 | [diff] [blame] | 158 | blk_start_plug(&plug); |
| 159 | |
Matthew Wilcox (Oracle) | 8151b4c | 2020-06-01 21:46:44 -0700 | [diff] [blame] | 160 | if (aops->readahead) { |
| 161 | aops->readahead(rac); |
NeilBrown | 9fd472a | 2022-03-22 14:38:54 -0700 | [diff] [blame] | 162 | /* |
Matthew Wilcox (Oracle) | a42634a | 2022-03-31 14:15:59 -0400 | [diff] [blame] | 163 | * Clean up the remaining folios. The sizes in ->ra |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 164 | * may be used to size the next readahead, so make sure |
NeilBrown | 9fd472a | 2022-03-22 14:38:54 -0700 | [diff] [blame] | 165 | * they accurately reflect what happened. |
| 166 | */ |
Matthew Wilcox (Oracle) | a42634a | 2022-03-31 14:15:59 -0400 | [diff] [blame] | 167 | while ((folio = readahead_folio(rac)) != NULL) { |
| 168 | unsigned long nr = folio_nr_pages(folio); |
| 169 | |
Matthew Wilcox (Oracle) | 6bf74cd | 2022-06-07 15:45:53 -0400 | [diff] [blame] | 170 | folio_get(folio); |
Matthew Wilcox (Oracle) | a42634a | 2022-03-31 14:15:59 -0400 | [diff] [blame] | 171 | rac->ra->size -= nr; |
| 172 | if (rac->ra->async_size >= nr) { |
| 173 | rac->ra->async_size -= nr; |
| 174 | filemap_remove_folio(folio); |
NeilBrown | 9fd472a | 2022-03-22 14:38:54 -0700 | [diff] [blame] | 175 | } |
Matthew Wilcox (Oracle) | a42634a | 2022-03-31 14:15:59 -0400 | [diff] [blame] | 176 | folio_unlock(folio); |
Matthew Wilcox (Oracle) | 6bf74cd | 2022-06-07 15:45:53 -0400 | [diff] [blame] | 177 | folio_put(folio); |
Matthew Wilcox (Oracle) | 8151b4c | 2020-06-01 21:46:44 -0700 | [diff] [blame] | 178 | } |
Matthew Wilcox (Oracle) | c1f6925 | 2020-06-01 21:46:40 -0700 | [diff] [blame] | 179 | } else { |
Matthew Wilcox (Oracle) | 5efe744 | 2022-04-29 08:43:23 -0400 | [diff] [blame] | 180 | while ((folio = readahead_folio(rac)) != NULL) |
Matthew Wilcox (Oracle) | 7e0a126 | 2022-04-29 11:53:28 -0400 | [diff] [blame] | 181 | aops->read_folio(rac->file, folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | } |
Jens Axboe | 5b417b1 | 2010-04-19 10:04:38 +0200 | [diff] [blame] | 183 | |
Jens Axboe | 5b417b1 | 2010-04-19 10:04:38 +0200 | [diff] [blame] | 184 | blk_finish_plug(&plug); |
Christoph Hellwig | 1760424 | 2022-09-15 10:41:56 +0100 | [diff] [blame] | 185 | if (unlikely(rac->_workingset)) |
| 186 | psi_memstall_leave(&rac->_pflags); |
| 187 | rac->_workingset = false; |
Matthew Wilcox (Oracle) | ad4ae1c | 2020-06-01 21:46:18 -0700 | [diff] [blame] | 188 | |
Matthew Wilcox (Oracle) | c1f6925 | 2020-06-01 21:46:40 -0700 | [diff] [blame] | 189 | BUG_ON(readahead_count(rac)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | } |
| 191 | |
Matthew Wilcox (Oracle) | 2c68423 | 2020-06-01 21:46:51 -0700 | [diff] [blame] | 192 | /** |
Matthew Wilcox (Oracle) | 73bb49d | 2020-10-15 20:06:14 -0700 | [diff] [blame] | 193 | * page_cache_ra_unbounded - Start unchecked readahead. |
| 194 | * @ractl: Readahead control. |
Matthew Wilcox (Oracle) | 2c68423 | 2020-06-01 21:46:51 -0700 | [diff] [blame] | 195 | * @nr_to_read: The number of pages to read. |
| 196 | * @lookahead_size: Where to start the next readahead. |
| 197 | * |
| 198 | * This function is for filesystems to call when they want to start |
| 199 | * readahead beyond a file's stated i_size. This is almost certainly |
| 200 | * not the function you want to call. Use page_cache_async_readahead() |
| 201 | * or page_cache_sync_readahead() instead. |
| 202 | * |
| 203 | * Context: File is referenced by caller. Mutexes may be held by caller. |
| 204 | * May sleep, but will not reenter filesystem to reclaim memory. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | */ |
Matthew Wilcox (Oracle) | 73bb49d | 2020-10-15 20:06:14 -0700 | [diff] [blame] | 206 | void page_cache_ra_unbounded(struct readahead_control *ractl, |
| 207 | unsigned long nr_to_read, unsigned long lookahead_size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | { |
Matthew Wilcox (Oracle) | 73bb49d | 2020-10-15 20:06:14 -0700 | [diff] [blame] | 209 | struct address_space *mapping = ractl->mapping; |
| 210 | unsigned long index = readahead_index(ractl); |
Michal Hocko | 8a5c743 | 2016-07-26 15:24:53 -0700 | [diff] [blame] | 211 | gfp_t gfp_mask = readahead_gfp_mask(mapping); |
Matthew Wilcox (Oracle) | c2c7ad7 | 2020-06-01 21:46:32 -0700 | [diff] [blame] | 212 | unsigned long i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | |
| 214 | /* |
Matthew Wilcox (Oracle) | f2c817b | 2020-06-01 21:46:58 -0700 | [diff] [blame] | 215 | * Partway through the readahead operation, we will have added |
| 216 | * locked pages to the page cache, but will not yet have submitted |
| 217 | * them for I/O. Adding another page may need to allocate memory, |
| 218 | * which can trigger memory reclaim. Telling the VM we're in |
| 219 | * the middle of a filesystem operation will cause it to not |
| 220 | * touch file-backed pages, preventing a deadlock. Most (all?) |
| 221 | * filesystems already specify __GFP_NOFS in their mapping's |
| 222 | * gfp_mask, but let's be explicit here. |
| 223 | */ |
| 224 | unsigned int nofs = memalloc_nofs_save(); |
| 225 | |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 226 | filemap_invalidate_lock_shared(mapping); |
Matthew Wilcox (Oracle) | f2c817b | 2020-06-01 21:46:58 -0700 | [diff] [blame] | 227 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 228 | * Preallocate as many pages as we will need. |
| 229 | */ |
Matthew Wilcox (Oracle) | c2c7ad7 | 2020-06-01 21:46:32 -0700 | [diff] [blame] | 230 | for (i = 0; i < nr_to_read; i++) { |
Matthew Wilcox (Oracle) | 0387df1 | 2021-03-10 16:06:51 -0500 | [diff] [blame] | 231 | struct folio *folio = xa_load(&mapping->i_pages, index + i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | |
Matthew Wilcox (Oracle) | 0387df1 | 2021-03-10 16:06:51 -0500 | [diff] [blame] | 233 | if (folio && !xa_is_value(folio)) { |
Christoph Hellwig | b3751e6 | 2018-06-01 09:03:06 -0700 | [diff] [blame] | 234 | /* |
Matthew Wilcox (Oracle) | 2d8163e | 2020-06-01 21:46:54 -0700 | [diff] [blame] | 235 | * Page already present? Kick off the current batch |
| 236 | * of contiguous pages before continuing with the |
| 237 | * next batch. This page may be the one we would |
| 238 | * have intended to mark as Readahead, but we don't |
| 239 | * have a stable reference to this page, and it's |
| 240 | * not worth getting one just for that. |
Christoph Hellwig | b3751e6 | 2018-06-01 09:03:06 -0700 | [diff] [blame] | 241 | */ |
Christoph Hellwig | b4e089d7 | 2022-03-31 05:35:55 -0700 | [diff] [blame] | 242 | read_pages(ractl); |
| 243 | ractl->_index++; |
Matthew Wilcox (Oracle) | f615bd5 | 2021-04-21 18:09:23 +0100 | [diff] [blame] | 244 | i = ractl->_index + ractl->_nr_pages - index - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 245 | continue; |
Christoph Hellwig | b3751e6 | 2018-06-01 09:03:06 -0700 | [diff] [blame] | 246 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 247 | |
Matthew Wilcox (Oracle) | 0387df1 | 2021-03-10 16:06:51 -0500 | [diff] [blame] | 248 | folio = filemap_alloc_folio(gfp_mask, 0); |
| 249 | if (!folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 250 | break; |
Matthew Wilcox (Oracle) | 704528d | 2022-03-23 21:29:04 -0400 | [diff] [blame] | 251 | if (filemap_add_folio(mapping, folio, index + i, |
Matthew Wilcox (Oracle) | c1f6925 | 2020-06-01 21:46:40 -0700 | [diff] [blame] | 252 | gfp_mask) < 0) { |
Matthew Wilcox (Oracle) | 0387df1 | 2021-03-10 16:06:51 -0500 | [diff] [blame] | 253 | folio_put(folio); |
Christoph Hellwig | b4e089d7 | 2022-03-31 05:35:55 -0700 | [diff] [blame] | 254 | read_pages(ractl); |
| 255 | ractl->_index++; |
Matthew Wilcox (Oracle) | f615bd5 | 2021-04-21 18:09:23 +0100 | [diff] [blame] | 256 | i = ractl->_index + ractl->_nr_pages - index - 1; |
Matthew Wilcox (Oracle) | c1f6925 | 2020-06-01 21:46:40 -0700 | [diff] [blame] | 257 | continue; |
| 258 | } |
Matthew Wilcox (Oracle) | c2c7ad7 | 2020-06-01 21:46:32 -0700 | [diff] [blame] | 259 | if (i == nr_to_read - lookahead_size) |
Matthew Wilcox (Oracle) | 0387df1 | 2021-03-10 16:06:51 -0500 | [diff] [blame] | 260 | folio_set_readahead(folio); |
Christoph Hellwig | 1760424 | 2022-09-15 10:41:56 +0100 | [diff] [blame] | 261 | ractl->_workingset |= folio_test_workingset(folio); |
Matthew Wilcox (Oracle) | 73bb49d | 2020-10-15 20:06:14 -0700 | [diff] [blame] | 262 | ractl->_nr_pages++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 263 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 264 | |
| 265 | /* |
Matthew Wilcox (Oracle) | 7e0a126 | 2022-04-29 11:53:28 -0400 | [diff] [blame] | 266 | * Now start the IO. We ignore I/O errors - if the folio is not |
| 267 | * uptodate then the caller will launch read_folio again, and |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 268 | * will then handle the error. |
| 269 | */ |
Christoph Hellwig | b4e089d7 | 2022-03-31 05:35:55 -0700 | [diff] [blame] | 270 | read_pages(ractl); |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 271 | filemap_invalidate_unlock_shared(mapping); |
Matthew Wilcox (Oracle) | f2c817b | 2020-06-01 21:46:58 -0700 | [diff] [blame] | 272 | memalloc_nofs_restore(nofs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | } |
Matthew Wilcox (Oracle) | 73bb49d | 2020-10-15 20:06:14 -0700 | [diff] [blame] | 274 | EXPORT_SYMBOL_GPL(page_cache_ra_unbounded); |
Matthew Wilcox (Oracle) | 2c68423 | 2020-06-01 21:46:51 -0700 | [diff] [blame] | 275 | |
| 276 | /* |
Matthew Wilcox (Oracle) | 8238287 | 2020-10-15 20:06:17 -0700 | [diff] [blame] | 277 | * do_page_cache_ra() actually reads a chunk of disk. It allocates |
Matthew Wilcox (Oracle) | 2c68423 | 2020-06-01 21:46:51 -0700 | [diff] [blame] | 278 | * the pages first, then submits them for I/O. This avoids the very bad |
| 279 | * behaviour which would occur if page allocations are causing VM writeback. |
| 280 | * We really don't want to intermingle reads and writes like that. |
| 281 | */ |
Matthew Wilcox (Oracle) | 56a4d67 | 2021-07-24 23:26:14 -0400 | [diff] [blame] | 282 | static void do_page_cache_ra(struct readahead_control *ractl, |
Matthew Wilcox (Oracle) | 8238287 | 2020-10-15 20:06:17 -0700 | [diff] [blame] | 283 | unsigned long nr_to_read, unsigned long lookahead_size) |
Matthew Wilcox (Oracle) | 2c68423 | 2020-06-01 21:46:51 -0700 | [diff] [blame] | 284 | { |
Matthew Wilcox (Oracle) | 8238287 | 2020-10-15 20:06:17 -0700 | [diff] [blame] | 285 | struct inode *inode = ractl->mapping->host; |
| 286 | unsigned long index = readahead_index(ractl); |
Matthew Wilcox (Oracle) | 2c68423 | 2020-06-01 21:46:51 -0700 | [diff] [blame] | 287 | loff_t isize = i_size_read(inode); |
| 288 | pgoff_t end_index; /* The last page we want to read */ |
| 289 | |
| 290 | if (isize == 0) |
| 291 | return; |
| 292 | |
| 293 | end_index = (isize - 1) >> PAGE_SHIFT; |
| 294 | if (index > end_index) |
| 295 | return; |
| 296 | /* Don't read past the page containing the last byte of the file */ |
| 297 | if (nr_to_read > end_index - index) |
| 298 | nr_to_read = end_index - index + 1; |
| 299 | |
Matthew Wilcox (Oracle) | 8238287 | 2020-10-15 20:06:17 -0700 | [diff] [blame] | 300 | page_cache_ra_unbounded(ractl, nr_to_read, lookahead_size); |
Matthew Wilcox (Oracle) | 2c68423 | 2020-06-01 21:46:51 -0700 | [diff] [blame] | 301 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 302 | |
| 303 | /* |
| 304 | * Chunk the readahead into 2 megabyte units, so that we don't pin too much |
| 305 | * memory at once. |
| 306 | */ |
David Howells | 7b3df3b | 2020-10-15 20:06:24 -0700 | [diff] [blame] | 307 | void force_page_cache_ra(struct readahead_control *ractl, |
Matthew Wilcox (Oracle) | fcd9ae4 | 2021-04-07 21:18:55 +0100 | [diff] [blame] | 308 | unsigned long nr_to_read) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 309 | { |
David Howells | 7b3df3b | 2020-10-15 20:06:24 -0700 | [diff] [blame] | 310 | struct address_space *mapping = ractl->mapping; |
Matthew Wilcox (Oracle) | fcd9ae4 | 2021-04-07 21:18:55 +0100 | [diff] [blame] | 311 | struct file_ra_state *ra = ractl->ra; |
Jens Axboe | 9491ae4 | 2016-12-12 16:43:26 -0800 | [diff] [blame] | 312 | struct backing_dev_info *bdi = inode_to_bdi(mapping->host); |
David Howells | 7b3df3b | 2020-10-15 20:06:24 -0700 | [diff] [blame] | 313 | unsigned long max_pages, index; |
Jens Axboe | 9491ae4 | 2016-12-12 16:43:26 -0800 | [diff] [blame] | 314 | |
Matthew Wilcox (Oracle) | 7e0a126 | 2022-04-29 11:53:28 -0400 | [diff] [blame] | 315 | if (unlikely(!mapping->a_ops->read_folio && !mapping->a_ops->readahead)) |
Matthew Wilcox (Oracle) | 9a42823 | 2020-06-01 21:46:10 -0700 | [diff] [blame] | 316 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 317 | |
Jens Axboe | 9491ae4 | 2016-12-12 16:43:26 -0800 | [diff] [blame] | 318 | /* |
| 319 | * If the request exceeds the readahead window, allow the read to |
| 320 | * be up to the optimal hardware IO size |
| 321 | */ |
David Howells | 7b3df3b | 2020-10-15 20:06:24 -0700 | [diff] [blame] | 322 | index = readahead_index(ractl); |
Jens Axboe | 9491ae4 | 2016-12-12 16:43:26 -0800 | [diff] [blame] | 323 | max_pages = max_t(unsigned long, bdi->io_pages, ra->ra_pages); |
David Howells | 7b3df3b | 2020-10-15 20:06:24 -0700 | [diff] [blame] | 324 | nr_to_read = min_t(unsigned long, nr_to_read, max_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 325 | while (nr_to_read) { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 326 | unsigned long this_chunk = (2 * 1024 * 1024) / PAGE_SIZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 327 | |
| 328 | if (this_chunk > nr_to_read) |
| 329 | this_chunk = nr_to_read; |
David Howells | 7b3df3b | 2020-10-15 20:06:24 -0700 | [diff] [blame] | 330 | ractl->_index = index; |
| 331 | do_page_cache_ra(ractl, this_chunk, 0); |
Mark Rutland | 58d5640 | 2014-01-29 14:05:51 -0800 | [diff] [blame] | 332 | |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 333 | index += this_chunk; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | nr_to_read -= this_chunk; |
| 335 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 336 | } |
| 337 | |
Fengguang Wu | 5ce1110 | 2007-07-19 01:47:59 -0700 | [diff] [blame] | 338 | /* |
Fengguang Wu | c743d96 | 2007-07-19 01:48:04 -0700 | [diff] [blame] | 339 | * Set the initial window size, round to next power of 2 and square |
| 340 | * for small size, x 4 for medium, and x 2 for large |
| 341 | * for 128k (32 page) max ra |
Lin Feng | fb25a77 | 2021-11-05 13:43:47 -0700 | [diff] [blame] | 342 | * 1-2 page = 16k, 3-4 page 32k, 5-8 page = 64k, > 8 page = 128k initial |
Fengguang Wu | c743d96 | 2007-07-19 01:48:04 -0700 | [diff] [blame] | 343 | */ |
| 344 | static unsigned long get_init_ra_size(unsigned long size, unsigned long max) |
| 345 | { |
| 346 | unsigned long newsize = roundup_pow_of_two(size); |
| 347 | |
| 348 | if (newsize <= max / 32) |
| 349 | newsize = newsize * 4; |
| 350 | else if (newsize <= max / 4) |
| 351 | newsize = newsize * 2; |
| 352 | else |
| 353 | newsize = max; |
| 354 | |
| 355 | return newsize; |
| 356 | } |
| 357 | |
| 358 | /* |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 359 | * Get the previous window size, ramp it up, and |
| 360 | * return it as the new window size. |
| 361 | */ |
Fengguang Wu | c743d96 | 2007-07-19 01:48:04 -0700 | [diff] [blame] | 362 | static unsigned long get_next_ra_size(struct file_ra_state *ra, |
Gao Xiang | 20ff1c9 | 2018-12-28 00:33:34 -0800 | [diff] [blame] | 363 | unsigned long max) |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 364 | { |
Fengguang Wu | f9acc8c | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 365 | unsigned long cur = ra->size; |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 366 | |
| 367 | if (cur < max / 16) |
Gao Xiang | 20ff1c9 | 2018-12-28 00:33:34 -0800 | [diff] [blame] | 368 | return 4 * cur; |
| 369 | if (cur <= max / 2) |
| 370 | return 2 * cur; |
| 371 | return max; |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 372 | } |
| 373 | |
| 374 | /* |
| 375 | * On-demand readahead design. |
| 376 | * |
| 377 | * The fields in struct file_ra_state represent the most-recently-executed |
| 378 | * readahead attempt: |
| 379 | * |
Fengguang Wu | f9acc8c | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 380 | * |<----- async_size ---------| |
| 381 | * |------------------- size -------------------->| |
| 382 | * |==================#===========================| |
| 383 | * ^start ^page marked with PG_readahead |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 384 | * |
| 385 | * To overlap application thinking time and disk I/O time, we do |
| 386 | * `readahead pipelining': Do not wait until the application consumed all |
| 387 | * readahead pages and stalled on the missing page at readahead_index; |
Fengguang Wu | f9acc8c | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 388 | * Instead, submit an asynchronous readahead I/O as soon as there are |
| 389 | * only async_size pages left in the readahead window. Normally async_size |
| 390 | * will be equal to size, for maximum pipelining. |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 391 | * |
| 392 | * In interleaved sequential reads, concurrent streams on the same fd can |
| 393 | * be invalidating each other's readahead state. So we flag the new readahead |
Fengguang Wu | f9acc8c | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 394 | * page at (start+size-async_size) with PG_readahead, and use it as readahead |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 395 | * indicator. The flag won't be set on already cached pages, to avoid the |
| 396 | * readahead-for-nothing fuss, saving pointless page cache lookups. |
| 397 | * |
Fengguang Wu | f4e6b49 | 2007-10-16 01:24:33 -0700 | [diff] [blame] | 398 | * prev_pos tracks the last visited byte in the _previous_ read request. |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 399 | * It should be maintained by the caller, and will be used for detecting |
| 400 | * small random reads. Note that the readahead algorithm checks loosely |
| 401 | * for sequential patterns. Hence interleaved reads might be served as |
| 402 | * sequential ones. |
| 403 | * |
| 404 | * There is a special-case: if the first page which the application tries to |
| 405 | * read happens to be the first page of the file, it is assumed that a linear |
| 406 | * read is about to happen and the window is immediately set to the initial size |
| 407 | * based on I/O request size and the max_readahead. |
| 408 | * |
| 409 | * The code ramps up the readahead size aggressively at first, but slow down as |
| 410 | * it approaches max_readhead. |
| 411 | */ |
| 412 | |
| 413 | /* |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 414 | * Count contiguously cached pages from @index-1 to @index-@max, |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 415 | * this count is a conservative estimation of |
| 416 | * - length of the sequential read sequence, or |
| 417 | * - thrashing threshold in memory tight systems |
| 418 | */ |
| 419 | static pgoff_t count_history_pages(struct address_space *mapping, |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 420 | pgoff_t index, unsigned long max) |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 421 | { |
| 422 | pgoff_t head; |
| 423 | |
| 424 | rcu_read_lock(); |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 425 | head = page_cache_prev_miss(mapping, index - 1, max); |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 426 | rcu_read_unlock(); |
| 427 | |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 428 | return index - 1 - head; |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 429 | } |
| 430 | |
| 431 | /* |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 432 | * page cache context based readahead |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 433 | */ |
| 434 | static int try_context_readahead(struct address_space *mapping, |
| 435 | struct file_ra_state *ra, |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 436 | pgoff_t index, |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 437 | unsigned long req_size, |
| 438 | unsigned long max) |
| 439 | { |
| 440 | pgoff_t size; |
| 441 | |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 442 | size = count_history_pages(mapping, index, max); |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 443 | |
| 444 | /* |
Fengguang Wu | 2cad401 | 2013-09-11 14:21:47 -0700 | [diff] [blame] | 445 | * not enough history pages: |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 446 | * it could be a random read |
| 447 | */ |
Fengguang Wu | 2cad401 | 2013-09-11 14:21:47 -0700 | [diff] [blame] | 448 | if (size <= req_size) |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 449 | return 0; |
| 450 | |
| 451 | /* |
| 452 | * starts from beginning of file: |
| 453 | * it is a strong indication of long-run stream (or whole-file-read) |
| 454 | */ |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 455 | if (size >= index) |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 456 | size *= 2; |
| 457 | |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 458 | ra->start = index; |
Fengguang Wu | 2cad401 | 2013-09-11 14:21:47 -0700 | [diff] [blame] | 459 | ra->size = min(size + req_size, max); |
| 460 | ra->async_size = 1; |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 461 | |
| 462 | return 1; |
| 463 | } |
| 464 | |
| 465 | /* |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 466 | * There are some parts of the kernel which assume that PMD entries |
| 467 | * are exactly HPAGE_PMD_ORDER. Those should be fixed, but until then, |
| 468 | * limit the maximum allocation order to PMD size. I'm not aware of any |
| 469 | * assumptions about maximum order if THP are disabled, but 8 seems like |
| 470 | * a good order (that's 1MB if you're using 4kB pages) |
| 471 | */ |
| 472 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 473 | #define MAX_PAGECACHE_ORDER HPAGE_PMD_ORDER |
| 474 | #else |
| 475 | #define MAX_PAGECACHE_ORDER 8 |
| 476 | #endif |
| 477 | |
| 478 | static inline int ra_alloc_folio(struct readahead_control *ractl, pgoff_t index, |
| 479 | pgoff_t mark, unsigned int order, gfp_t gfp) |
| 480 | { |
| 481 | int err; |
| 482 | struct folio *folio = filemap_alloc_folio(gfp, order); |
| 483 | |
| 484 | if (!folio) |
| 485 | return -ENOMEM; |
Matthew Wilcox (Oracle) | b9ff43d | 2022-04-27 17:01:28 -0400 | [diff] [blame] | 486 | mark = round_up(mark, 1UL << order); |
| 487 | if (index == mark) |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 488 | folio_set_readahead(folio); |
| 489 | err = filemap_add_folio(ractl->mapping, folio, index, gfp); |
Christoph Hellwig | 1760424 | 2022-09-15 10:41:56 +0100 | [diff] [blame] | 490 | if (err) { |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 491 | folio_put(folio); |
Christoph Hellwig | 1760424 | 2022-09-15 10:41:56 +0100 | [diff] [blame] | 492 | return err; |
| 493 | } |
| 494 | |
| 495 | ractl->_nr_pages += 1UL << order; |
| 496 | ractl->_workingset |= folio_test_workingset(folio); |
| 497 | return 0; |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 498 | } |
| 499 | |
Matthew Wilcox (Oracle) | 56a4d67 | 2021-07-24 23:26:14 -0400 | [diff] [blame] | 500 | void page_cache_ra_order(struct readahead_control *ractl, |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 501 | struct file_ra_state *ra, unsigned int new_order) |
| 502 | { |
| 503 | struct address_space *mapping = ractl->mapping; |
| 504 | pgoff_t index = readahead_index(ractl); |
| 505 | pgoff_t limit = (i_size_read(mapping->host) - 1) >> PAGE_SHIFT; |
| 506 | pgoff_t mark = index + ra->size - ra->async_size; |
| 507 | int err = 0; |
| 508 | gfp_t gfp = readahead_gfp_mask(mapping); |
| 509 | |
| 510 | if (!mapping_large_folio_support(mapping) || ra->size < 4) |
| 511 | goto fallback; |
| 512 | |
| 513 | limit = min(limit, index + ra->size - 1); |
| 514 | |
| 515 | if (new_order < MAX_PAGECACHE_ORDER) { |
| 516 | new_order += 2; |
| 517 | if (new_order > MAX_PAGECACHE_ORDER) |
| 518 | new_order = MAX_PAGECACHE_ORDER; |
| 519 | while ((1 << new_order) > ra->size) |
| 520 | new_order--; |
| 521 | } |
| 522 | |
Alistair Popple | 00fa15e | 2022-06-20 19:05:36 +1000 | [diff] [blame] | 523 | filemap_invalidate_lock_shared(mapping); |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 524 | while (index <= limit) { |
| 525 | unsigned int order = new_order; |
| 526 | |
| 527 | /* Align with smaller pages if needed */ |
| 528 | if (index & ((1UL << order) - 1)) { |
| 529 | order = __ffs(index); |
| 530 | if (order == 1) |
| 531 | order = 0; |
| 532 | } |
| 533 | /* Don't allocate pages past EOF */ |
| 534 | while (index + (1UL << order) - 1 > limit) { |
| 535 | if (--order == 1) |
| 536 | order = 0; |
| 537 | } |
| 538 | err = ra_alloc_folio(ractl, index, mark, order, gfp); |
| 539 | if (err) |
| 540 | break; |
| 541 | index += 1UL << order; |
| 542 | } |
| 543 | |
| 544 | if (index > limit) { |
| 545 | ra->size += index - limit - 1; |
| 546 | ra->async_size += index - limit - 1; |
| 547 | } |
| 548 | |
Christoph Hellwig | b4e089d7 | 2022-03-31 05:35:55 -0700 | [diff] [blame] | 549 | read_pages(ractl); |
Alistair Popple | 00fa15e | 2022-06-20 19:05:36 +1000 | [diff] [blame] | 550 | filemap_invalidate_unlock_shared(mapping); |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 551 | |
| 552 | /* |
| 553 | * If there were already pages in the page cache, then we may have |
| 554 | * left some gaps. Let the regular readahead code take care of this |
| 555 | * situation. |
| 556 | */ |
| 557 | if (!err) |
| 558 | return; |
| 559 | fallback: |
| 560 | do_page_cache_ra(ractl, ra->size, ra->async_size); |
| 561 | } |
| 562 | |
| 563 | /* |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 564 | * A minimal readahead algorithm for trivial sequential/random reads. |
| 565 | */ |
David Howells | 6e4af69 | 2020-10-15 20:06:21 -0700 | [diff] [blame] | 566 | static void ondemand_readahead(struct readahead_control *ractl, |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 567 | struct folio *folio, unsigned long req_size) |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 568 | { |
David Howells | 6e4af69 | 2020-10-15 20:06:21 -0700 | [diff] [blame] | 569 | struct backing_dev_info *bdi = inode_to_bdi(ractl->mapping->host); |
Matthew Wilcox (Oracle) | fcd9ae4 | 2021-04-07 21:18:55 +0100 | [diff] [blame] | 570 | struct file_ra_state *ra = ractl->ra; |
Jens Axboe | 9491ae4 | 2016-12-12 16:43:26 -0800 | [diff] [blame] | 571 | unsigned long max_pages = ra->ra_pages; |
Markus Stockhausen | dc30b96 | 2018-07-27 09:09:53 -0600 | [diff] [blame] | 572 | unsigned long add_pages; |
Matthew Wilcox (Oracle) | b9ff43d | 2022-04-27 17:01:28 -0400 | [diff] [blame] | 573 | pgoff_t index = readahead_index(ractl); |
| 574 | pgoff_t expected, prev_index; |
| 575 | unsigned int order = folio ? folio_order(folio) : 0; |
Wu Fengguang | 045a252 | 2009-06-16 15:31:33 -0700 | [diff] [blame] | 576 | |
| 577 | /* |
Jens Axboe | 9491ae4 | 2016-12-12 16:43:26 -0800 | [diff] [blame] | 578 | * If the request exceeds the readahead window, allow the read to |
| 579 | * be up to the optimal hardware IO size |
| 580 | */ |
| 581 | if (req_size > max_pages && bdi->io_pages > max_pages) |
| 582 | max_pages = min(req_size, bdi->io_pages); |
| 583 | |
| 584 | /* |
Wu Fengguang | 045a252 | 2009-06-16 15:31:33 -0700 | [diff] [blame] | 585 | * start of file |
| 586 | */ |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 587 | if (!index) |
Wu Fengguang | 045a252 | 2009-06-16 15:31:33 -0700 | [diff] [blame] | 588 | goto initial_readahead; |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 589 | |
| 590 | /* |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 591 | * It's the expected callback index, assume sequential access. |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 592 | * Ramp up sizes, and push forward the readahead window. |
| 593 | */ |
Matthew Wilcox (Oracle) | b9ff43d | 2022-04-27 17:01:28 -0400 | [diff] [blame] | 594 | expected = round_up(ra->start + ra->size - ra->async_size, |
| 595 | 1UL << order); |
| 596 | if (index == expected || index == (ra->start + ra->size)) { |
Fengguang Wu | f9acc8c | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 597 | ra->start += ra->size; |
Jens Axboe | 9491ae4 | 2016-12-12 16:43:26 -0800 | [diff] [blame] | 598 | ra->size = get_next_ra_size(ra, max_pages); |
Fengguang Wu | f9acc8c | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 599 | ra->async_size = ra->size; |
| 600 | goto readit; |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 601 | } |
| 602 | |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 603 | /* |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 604 | * Hit a marked folio without valid readahead state. |
Fengguang Wu | 6b10c6c | 2007-10-16 01:24:34 -0700 | [diff] [blame] | 605 | * E.g. interleaved reads. |
| 606 | * Query the pagecache for async_size, which normally equals to |
| 607 | * readahead size. Ramp it up and use it as the new readahead size. |
| 608 | */ |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 609 | if (folio) { |
Fengguang Wu | 6b10c6c | 2007-10-16 01:24:34 -0700 | [diff] [blame] | 610 | pgoff_t start; |
| 611 | |
Nick Piggin | 30002ed | 2008-07-25 19:45:28 -0700 | [diff] [blame] | 612 | rcu_read_lock(); |
David Howells | 6e4af69 | 2020-10-15 20:06:21 -0700 | [diff] [blame] | 613 | start = page_cache_next_miss(ractl->mapping, index + 1, |
| 614 | max_pages); |
Nick Piggin | 30002ed | 2008-07-25 19:45:28 -0700 | [diff] [blame] | 615 | rcu_read_unlock(); |
Fengguang Wu | 6b10c6c | 2007-10-16 01:24:34 -0700 | [diff] [blame] | 616 | |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 617 | if (!start || start - index > max_pages) |
Matthew Wilcox (Oracle) | 9a42823 | 2020-06-01 21:46:10 -0700 | [diff] [blame] | 618 | return; |
Fengguang Wu | 6b10c6c | 2007-10-16 01:24:34 -0700 | [diff] [blame] | 619 | |
| 620 | ra->start = start; |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 621 | ra->size = start - index; /* old async_size */ |
Wu Fengguang | 160334a | 2009-06-16 15:31:23 -0700 | [diff] [blame] | 622 | ra->size += req_size; |
Jens Axboe | 9491ae4 | 2016-12-12 16:43:26 -0800 | [diff] [blame] | 623 | ra->size = get_next_ra_size(ra, max_pages); |
Fengguang Wu | 6b10c6c | 2007-10-16 01:24:34 -0700 | [diff] [blame] | 624 | ra->async_size = ra->size; |
| 625 | goto readit; |
| 626 | } |
| 627 | |
| 628 | /* |
Wu Fengguang | 045a252 | 2009-06-16 15:31:33 -0700 | [diff] [blame] | 629 | * oversize read |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 630 | */ |
Jens Axboe | 9491ae4 | 2016-12-12 16:43:26 -0800 | [diff] [blame] | 631 | if (req_size > max_pages) |
Wu Fengguang | 045a252 | 2009-06-16 15:31:33 -0700 | [diff] [blame] | 632 | goto initial_readahead; |
| 633 | |
| 634 | /* |
| 635 | * sequential cache miss |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 636 | * trivial case: (index - prev_index) == 1 |
| 637 | * unaligned reads: (index - prev_index) == 0 |
Wu Fengguang | 045a252 | 2009-06-16 15:31:33 -0700 | [diff] [blame] | 638 | */ |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 639 | prev_index = (unsigned long long)ra->prev_pos >> PAGE_SHIFT; |
| 640 | if (index - prev_index <= 1UL) |
Wu Fengguang | 045a252 | 2009-06-16 15:31:33 -0700 | [diff] [blame] | 641 | goto initial_readahead; |
| 642 | |
| 643 | /* |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 644 | * Query the page cache and look for the traces(cached history pages) |
| 645 | * that a sequential stream would leave behind. |
| 646 | */ |
David Howells | 6e4af69 | 2020-10-15 20:06:21 -0700 | [diff] [blame] | 647 | if (try_context_readahead(ractl->mapping, ra, index, req_size, |
| 648 | max_pages)) |
Wu Fengguang | 10be0b3 | 2009-06-16 15:31:36 -0700 | [diff] [blame] | 649 | goto readit; |
| 650 | |
| 651 | /* |
Wu Fengguang | 045a252 | 2009-06-16 15:31:33 -0700 | [diff] [blame] | 652 | * standalone, small random read |
| 653 | * Read as is, and do not pollute the readahead state. |
| 654 | */ |
David Howells | 6e4af69 | 2020-10-15 20:06:21 -0700 | [diff] [blame] | 655 | do_page_cache_ra(ractl, req_size, 0); |
Matthew Wilcox (Oracle) | 9a42823 | 2020-06-01 21:46:10 -0700 | [diff] [blame] | 656 | return; |
Wu Fengguang | 045a252 | 2009-06-16 15:31:33 -0700 | [diff] [blame] | 657 | |
| 658 | initial_readahead: |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 659 | ra->start = index; |
Jens Axboe | 9491ae4 | 2016-12-12 16:43:26 -0800 | [diff] [blame] | 660 | ra->size = get_init_ra_size(req_size, max_pages); |
Fengguang Wu | f9acc8c | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 661 | ra->async_size = ra->size > req_size ? ra->size - req_size : ra->size; |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 662 | |
Fengguang Wu | f9acc8c | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 663 | readit: |
Wu Fengguang | 51daa88 | 2009-06-16 15:31:24 -0700 | [diff] [blame] | 664 | /* |
| 665 | * Will this read hit the readahead marker made by itself? |
| 666 | * If so, trigger the readahead marker hit now, and merge |
| 667 | * the resulted next readahead window into the current one. |
Markus Stockhausen | dc30b96 | 2018-07-27 09:09:53 -0600 | [diff] [blame] | 668 | * Take care of maximum IO pages as above. |
Wu Fengguang | 51daa88 | 2009-06-16 15:31:24 -0700 | [diff] [blame] | 669 | */ |
Matthew Wilcox (Oracle) | 08eb965 | 2020-06-01 21:46:29 -0700 | [diff] [blame] | 670 | if (index == ra->start && ra->size == ra->async_size) { |
Markus Stockhausen | dc30b96 | 2018-07-27 09:09:53 -0600 | [diff] [blame] | 671 | add_pages = get_next_ra_size(ra, max_pages); |
| 672 | if (ra->size + add_pages <= max_pages) { |
| 673 | ra->async_size = add_pages; |
| 674 | ra->size += add_pages; |
| 675 | } else { |
| 676 | ra->size = max_pages; |
| 677 | ra->async_size = max_pages >> 1; |
| 678 | } |
Wu Fengguang | 51daa88 | 2009-06-16 15:31:24 -0700 | [diff] [blame] | 679 | } |
| 680 | |
David Howells | 6e4af69 | 2020-10-15 20:06:21 -0700 | [diff] [blame] | 681 | ractl->_index = ra->start; |
Matthew Wilcox (Oracle) | b9ff43d | 2022-04-27 17:01:28 -0400 | [diff] [blame] | 682 | page_cache_ra_order(ractl, ra, order); |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 683 | } |
| 684 | |
Matthew Wilcox (Oracle) | fefa7c4 | 2020-10-15 20:06:28 -0700 | [diff] [blame] | 685 | void page_cache_sync_ra(struct readahead_control *ractl, |
Matthew Wilcox (Oracle) | fcd9ae4 | 2021-04-07 21:18:55 +0100 | [diff] [blame] | 686 | unsigned long req_count) |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 687 | { |
Jens Axboe | 324bcf5 | 2020-10-17 09:25:52 -0600 | [diff] [blame] | 688 | bool do_forced_ra = ractl->file && (ractl->file->f_mode & FMODE_RANDOM); |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 689 | |
Jens Axboe | 324bcf5 | 2020-10-17 09:25:52 -0600 | [diff] [blame] | 690 | /* |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 691 | * Even if readahead is disabled, issue this request as readahead |
Jens Axboe | 324bcf5 | 2020-10-17 09:25:52 -0600 | [diff] [blame] | 692 | * as we'll need it to satisfy the requested range. The forced |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 693 | * readahead will do the right thing and limit the read to just the |
Jens Axboe | 324bcf5 | 2020-10-17 09:25:52 -0600 | [diff] [blame] | 694 | * requested range, which we'll set to 1 page for this case. |
| 695 | */ |
Matthew Wilcox (Oracle) | fcd9ae4 | 2021-04-07 21:18:55 +0100 | [diff] [blame] | 696 | if (!ractl->ra->ra_pages || blk_cgroup_congested()) { |
Jens Axboe | 324bcf5 | 2020-10-17 09:25:52 -0600 | [diff] [blame] | 697 | if (!ractl->file) |
| 698 | return; |
| 699 | req_count = 1; |
| 700 | do_forced_ra = true; |
| 701 | } |
Josef Bacik | ca47e8c | 2018-07-03 11:15:03 -0400 | [diff] [blame] | 702 | |
Wu Fengguang | 0141450 | 2010-03-05 13:42:03 -0800 | [diff] [blame] | 703 | /* be dumb */ |
Jens Axboe | 324bcf5 | 2020-10-17 09:25:52 -0600 | [diff] [blame] | 704 | if (do_forced_ra) { |
Matthew Wilcox (Oracle) | fcd9ae4 | 2021-04-07 21:18:55 +0100 | [diff] [blame] | 705 | force_page_cache_ra(ractl, req_count); |
Wu Fengguang | 0141450 | 2010-03-05 13:42:03 -0800 | [diff] [blame] | 706 | return; |
| 707 | } |
| 708 | |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 709 | ondemand_readahead(ractl, NULL, req_count); |
Fengguang Wu | 122a21d | 2007-07-19 01:48:01 -0700 | [diff] [blame] | 710 | } |
Matthew Wilcox (Oracle) | fefa7c4 | 2020-10-15 20:06:28 -0700 | [diff] [blame] | 711 | EXPORT_SYMBOL_GPL(page_cache_sync_ra); |
Rusty Russell | cf914a7 | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 712 | |
Matthew Wilcox (Oracle) | fefa7c4 | 2020-10-15 20:06:28 -0700 | [diff] [blame] | 713 | void page_cache_async_ra(struct readahead_control *ractl, |
Matthew Wilcox (Oracle) | 7836d99 | 2021-05-27 12:30:54 -0400 | [diff] [blame] | 714 | struct folio *folio, unsigned long req_count) |
Rusty Russell | cf914a7 | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 715 | { |
Matthew Wilcox (Oracle) | 1e47028 | 2022-03-31 15:02:34 -0400 | [diff] [blame] | 716 | /* no readahead */ |
Matthew Wilcox (Oracle) | fcd9ae4 | 2021-04-07 21:18:55 +0100 | [diff] [blame] | 717 | if (!ractl->ra->ra_pages) |
Rusty Russell | cf914a7 | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 718 | return; |
| 719 | |
| 720 | /* |
| 721 | * Same bit is used for PG_readahead and PG_reclaim. |
| 722 | */ |
Matthew Wilcox (Oracle) | 7836d99 | 2021-05-27 12:30:54 -0400 | [diff] [blame] | 723 | if (folio_test_writeback(folio)) |
Rusty Russell | cf914a7 | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 724 | return; |
| 725 | |
Matthew Wilcox (Oracle) | 7836d99 | 2021-05-27 12:30:54 -0400 | [diff] [blame] | 726 | folio_clear_readahead(folio); |
Rusty Russell | cf914a7 | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 727 | |
Josef Bacik | ca47e8c | 2018-07-03 11:15:03 -0400 | [diff] [blame] | 728 | if (blk_cgroup_congested()) |
| 729 | return; |
| 730 | |
Matthew Wilcox (Oracle) | 793917d | 2020-02-05 11:27:01 -0500 | [diff] [blame] | 731 | ondemand_readahead(ractl, folio, req_count); |
Rusty Russell | cf914a7 | 2007-07-19 01:48:08 -0700 | [diff] [blame] | 732 | } |
Matthew Wilcox (Oracle) | fefa7c4 | 2020-10-15 20:06:28 -0700 | [diff] [blame] | 733 | EXPORT_SYMBOL_GPL(page_cache_async_ra); |
Cong Wang | 782182e | 2012-05-29 15:06:43 -0700 | [diff] [blame] | 734 | |
Dominik Brodowski | c7b95d5 | 2018-03-19 17:51:36 +0100 | [diff] [blame] | 735 | ssize_t ksys_readahead(int fd, loff_t offset, size_t count) |
Cong Wang | 782182e | 2012-05-29 15:06:43 -0700 | [diff] [blame] | 736 | { |
| 737 | ssize_t ret; |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 738 | struct fd f; |
Cong Wang | 782182e | 2012-05-29 15:06:43 -0700 | [diff] [blame] | 739 | |
| 740 | ret = -EBADF; |
Al Viro | 2903ff0 | 2012-08-28 12:52:22 -0400 | [diff] [blame] | 741 | f = fdget(fd); |
Amir Goldstein | 3d8f761 | 2018-08-29 08:41:29 +0300 | [diff] [blame] | 742 | if (!f.file || !(f.file->f_mode & FMODE_READ)) |
| 743 | goto out; |
| 744 | |
| 745 | /* |
| 746 | * The readahead() syscall is intended to run only on files |
| 747 | * that can execute readahead. If readahead is not possible |
| 748 | * on this file, then we must return -EINVAL. |
| 749 | */ |
| 750 | ret = -EINVAL; |
| 751 | if (!f.file->f_mapping || !f.file->f_mapping->a_ops || |
| 752 | !S_ISREG(file_inode(f.file)->i_mode)) |
| 753 | goto out; |
| 754 | |
| 755 | ret = vfs_fadvise(f.file, offset, count, POSIX_FADV_WILLNEED); |
| 756 | out: |
| 757 | fdput(f); |
Cong Wang | 782182e | 2012-05-29 15:06:43 -0700 | [diff] [blame] | 758 | return ret; |
| 759 | } |
Dominik Brodowski | c7b95d5 | 2018-03-19 17:51:36 +0100 | [diff] [blame] | 760 | |
| 761 | SYSCALL_DEFINE3(readahead, int, fd, loff_t, offset, size_t, count) |
| 762 | { |
| 763 | return ksys_readahead(fd, offset, count); |
| 764 | } |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 765 | |
Guo Ren | 59c10c5 | 2022-04-05 15:13:05 +0800 | [diff] [blame] | 766 | #if defined(CONFIG_COMPAT) && defined(__ARCH_WANT_COMPAT_READAHEAD) |
| 767 | COMPAT_SYSCALL_DEFINE4(readahead, int, fd, compat_arg_u64_dual(offset), size_t, count) |
| 768 | { |
| 769 | return ksys_readahead(fd, compat_arg_u64_glue(offset), count); |
| 770 | } |
| 771 | #endif |
| 772 | |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 773 | /** |
| 774 | * readahead_expand - Expand a readahead request |
| 775 | * @ractl: The request to be expanded |
| 776 | * @new_start: The revised start |
| 777 | * @new_len: The revised size of the request |
| 778 | * |
| 779 | * Attempt to expand a readahead request outwards from the current size to the |
| 780 | * specified size by inserting locked pages before and after the current window |
| 781 | * to increase the size to the new window. This may involve the insertion of |
| 782 | * THPs, in which case the window may get expanded even beyond what was |
| 783 | * requested. |
| 784 | * |
| 785 | * The algorithm will stop if it encounters a conflicting page already in the |
| 786 | * pagecache and leave a smaller expansion than requested. |
| 787 | * |
| 788 | * The caller must check for this by examining the revised @ractl object for a |
| 789 | * different expansion than was requested. |
| 790 | */ |
| 791 | void readahead_expand(struct readahead_control *ractl, |
| 792 | loff_t new_start, size_t new_len) |
| 793 | { |
| 794 | struct address_space *mapping = ractl->mapping; |
| 795 | struct file_ra_state *ra = ractl->ra; |
| 796 | pgoff_t new_index, new_nr_pages; |
| 797 | gfp_t gfp_mask = readahead_gfp_mask(mapping); |
| 798 | |
| 799 | new_index = new_start / PAGE_SIZE; |
| 800 | |
| 801 | /* Expand the leading edge downwards */ |
| 802 | while (ractl->_index > new_index) { |
| 803 | unsigned long index = ractl->_index - 1; |
Matthew Wilcox (Oracle) | 11a9804 | 2023-01-16 19:39:41 +0000 | [diff] [blame] | 804 | struct folio *folio = xa_load(&mapping->i_pages, index); |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 805 | |
Matthew Wilcox (Oracle) | 11a9804 | 2023-01-16 19:39:41 +0000 | [diff] [blame] | 806 | if (folio && !xa_is_value(folio)) |
| 807 | return; /* Folio apparently present */ |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 808 | |
Matthew Wilcox (Oracle) | 11a9804 | 2023-01-16 19:39:41 +0000 | [diff] [blame] | 809 | folio = filemap_alloc_folio(gfp_mask, 0); |
| 810 | if (!folio) |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 811 | return; |
Matthew Wilcox (Oracle) | 11a9804 | 2023-01-16 19:39:41 +0000 | [diff] [blame] | 812 | if (filemap_add_folio(mapping, folio, index, gfp_mask) < 0) { |
| 813 | folio_put(folio); |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 814 | return; |
| 815 | } |
Matthew Wilcox (Oracle) | 11a9804 | 2023-01-16 19:39:41 +0000 | [diff] [blame] | 816 | if (unlikely(folio_test_workingset(folio)) && |
| 817 | !ractl->_workingset) { |
| 818 | ractl->_workingset = true; |
| 819 | psi_memstall_enter(&ractl->_pflags); |
| 820 | } |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 821 | ractl->_nr_pages++; |
Matthew Wilcox (Oracle) | 11a9804 | 2023-01-16 19:39:41 +0000 | [diff] [blame] | 822 | ractl->_index = folio->index; |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 823 | } |
| 824 | |
| 825 | new_len += new_start - readahead_pos(ractl); |
| 826 | new_nr_pages = DIV_ROUND_UP(new_len, PAGE_SIZE); |
| 827 | |
| 828 | /* Expand the trailing edge upwards */ |
| 829 | while (ractl->_nr_pages < new_nr_pages) { |
| 830 | unsigned long index = ractl->_index + ractl->_nr_pages; |
Matthew Wilcox (Oracle) | 11a9804 | 2023-01-16 19:39:41 +0000 | [diff] [blame] | 831 | struct folio *folio = xa_load(&mapping->i_pages, index); |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 832 | |
Matthew Wilcox (Oracle) | 11a9804 | 2023-01-16 19:39:41 +0000 | [diff] [blame] | 833 | if (folio && !xa_is_value(folio)) |
| 834 | return; /* Folio apparently present */ |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 835 | |
Matthew Wilcox (Oracle) | 11a9804 | 2023-01-16 19:39:41 +0000 | [diff] [blame] | 836 | folio = filemap_alloc_folio(gfp_mask, 0); |
| 837 | if (!folio) |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 838 | return; |
Matthew Wilcox (Oracle) | 11a9804 | 2023-01-16 19:39:41 +0000 | [diff] [blame] | 839 | if (filemap_add_folio(mapping, folio, index, gfp_mask) < 0) { |
| 840 | folio_put(folio); |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 841 | return; |
| 842 | } |
Matthew Wilcox (Oracle) | 11a9804 | 2023-01-16 19:39:41 +0000 | [diff] [blame] | 843 | if (unlikely(folio_test_workingset(folio)) && |
| 844 | !ractl->_workingset) { |
Christoph Hellwig | 1760424 | 2022-09-15 10:41:56 +0100 | [diff] [blame] | 845 | ractl->_workingset = true; |
| 846 | psi_memstall_enter(&ractl->_pflags); |
| 847 | } |
David Howells | 3ca2364 | 2020-09-10 14:03:27 +0100 | [diff] [blame] | 848 | ractl->_nr_pages++; |
| 849 | if (ra) { |
| 850 | ra->size++; |
| 851 | ra->async_size++; |
| 852 | } |
| 853 | } |
| 854 | } |
| 855 | EXPORT_SYMBOL(readahead_expand); |