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 | * linux/mm/filemap.c |
| 4 | * |
| 5 | * Copyright (C) 1994-1999 Linus Torvalds |
| 6 | */ |
| 7 | |
| 8 | /* |
| 9 | * This file handles the generic file mmap semantics used by |
| 10 | * most "normal" filesystems (but you don't /have/ to use this: |
| 11 | * the NFS filesystem used to do this differently, for example) |
| 12 | */ |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 13 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/compiler.h> |
Ross Zwisler | f9fe48b | 2016-01-22 15:10:40 -0800 | [diff] [blame] | 15 | #include <linux/dax.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 16 | #include <linux/fs.h> |
Ingo Molnar | 3f07c01 | 2017-02-08 18:51:30 +0100 | [diff] [blame] | 17 | #include <linux/sched/signal.h> |
Hiro Yoshioka | c22ce14 | 2006-06-23 02:04:16 -0700 | [diff] [blame] | 18 | #include <linux/uaccess.h> |
Randy.Dunlap | c59ede7 | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 19 | #include <linux/capability.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include <linux/kernel_stat.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 21 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | #include <linux/mm.h> |
| 23 | #include <linux/swap.h> |
Alistair Popple | ffa6575 | 2022-01-21 22:10:46 -0800 | [diff] [blame] | 24 | #include <linux/swapops.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/mman.h> |
| 26 | #include <linux/pagemap.h> |
| 27 | #include <linux/file.h> |
| 28 | #include <linux/uio.h> |
Josef Bacik | cfcbfb1 | 2019-05-13 17:21:04 -0700 | [diff] [blame] | 29 | #include <linux/error-injection.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include <linux/hash.h> |
| 31 | #include <linux/writeback.h> |
Linus Torvalds | 5325338 | 2007-10-18 14:47:32 -0700 | [diff] [blame] | 32 | #include <linux/backing-dev.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | #include <linux/pagevec.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | #include <linux/security.h> |
Paul Jackson | 44110fe | 2006-03-24 03:16:04 -0800 | [diff] [blame] | 35 | #include <linux/cpuset.h> |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 36 | #include <linux/hugetlb.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 37 | #include <linux/memcontrol.h> |
Mel Gorman | c7df8ad | 2017-11-15 17:37:41 -0800 | [diff] [blame] | 38 | #include <linux/shmem_fs.h> |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 39 | #include <linux/rmap.h> |
Johannes Weiner | b1d29ba | 2018-10-26 15:06:08 -0700 | [diff] [blame] | 40 | #include <linux/delayacct.h> |
Johannes Weiner | eb41468 | 2018-10-26 15:06:27 -0700 | [diff] [blame] | 41 | #include <linux/psi.h> |
Ben Dooks | d0e6a58 | 2019-10-18 20:20:20 -0700 | [diff] [blame] | 42 | #include <linux/ramfs.h> |
Yang Shi | b9306a7 | 2020-08-06 23:19:55 -0700 | [diff] [blame] | 43 | #include <linux/page_idle.h> |
Alistair Popple | ffa6575 | 2022-01-21 22:10:46 -0800 | [diff] [blame] | 44 | #include <linux/migrate.h> |
David Howells | 07073eb | 2023-02-14 15:01:42 +0000 | [diff] [blame] | 45 | #include <linux/pipe_fs_i.h> |
| 46 | #include <linux/splice.h> |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 47 | #include <asm/pgalloc.h> |
Will Deacon | de591a8 | 2021-02-10 11:15:11 +0000 | [diff] [blame] | 48 | #include <asm/tlbflush.h> |
Nick Piggin | 0f8053a | 2006-03-22 00:08:33 -0800 | [diff] [blame] | 49 | #include "internal.h" |
| 50 | |
Robert Jarzmik | fe0bfaa | 2013-04-29 15:06:10 -0700 | [diff] [blame] | 51 | #define CREATE_TRACE_POINTS |
| 52 | #include <trace/events/filemap.h> |
| 53 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 55 | * FIXME: remove all knowledge of the buffer layer from the core VM |
| 56 | */ |
Jan Kara | 148f948 | 2009-08-17 19:52:36 +0200 | [diff] [blame] | 57 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | #include <asm/mman.h> |
| 60 | |
| 61 | /* |
| 62 | * Shared mappings implemented 30.11.1994. It's not fully working yet, |
| 63 | * though. |
| 64 | * |
| 65 | * Shared mappings now work. 15.8.1995 Bruno. |
| 66 | * |
| 67 | * finished 'unifying' the page and buffer cache and SMP-threaded the |
| 68 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> |
| 69 | * |
| 70 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> |
| 71 | */ |
| 72 | |
| 73 | /* |
| 74 | * Lock ordering: |
| 75 | * |
Davidlohr Bueso | c8c06ef | 2014-12-12 16:54:24 -0800 | [diff] [blame] | 76 | * ->i_mmap_rwsem (truncate_pagecache) |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 77 | * ->private_lock (__free_pte->block_dirty_folio) |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 78 | * ->swap_lock (exclusive_swap_page, others) |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 79 | * ->i_pages lock |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 80 | * |
Jan Kara | 9608703 | 2021-04-12 15:50:21 +0200 | [diff] [blame] | 81 | * ->i_rwsem |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 82 | * ->invalidate_lock (acquired by fs in truncate path) |
| 83 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | * |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 85 | * ->mmap_lock |
Davidlohr Bueso | c8c06ef | 2014-12-12 16:54:24 -0800 | [diff] [blame] | 86 | * ->i_mmap_rwsem |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 87 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 88 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | * |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 90 | * ->mmap_lock |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 91 | * ->invalidate_lock (filemap_fault) |
| 92 | * ->lock_page (filemap_fault, access_process_vm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | * |
Jan Kara | 9608703 | 2021-04-12 15:50:21 +0200 | [diff] [blame] | 94 | * ->i_rwsem (generic_perform_write) |
Andreas Gruenbacher | bb523b4 | 2021-08-02 13:44:20 +0200 | [diff] [blame] | 95 | * ->mmap_lock (fault_in_readable->do_page_fault) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | * |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 97 | * bdi->wb.list_lock |
Dave Chinner | a66979a | 2011-03-22 22:23:41 +1100 | [diff] [blame] | 98 | * sb_lock (fs/fs-writeback.c) |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 99 | * ->i_pages lock (__sync_single_inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | * |
Davidlohr Bueso | c8c06ef | 2014-12-12 16:54:24 -0800 | [diff] [blame] | 101 | * ->i_mmap_rwsem |
Liam R. Howlett | 0503ea8 | 2023-01-20 11:26:49 -0500 | [diff] [blame] | 102 | * ->anon_vma.lock (vma_merge) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | * |
| 104 | * ->anon_vma.lock |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 105 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | * |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 107 | * ->page_table_lock or pte_lock |
Hugh Dickins | 5d337b9 | 2005-09-03 15:54:41 -0700 | [diff] [blame] | 108 | * ->swap_lock (try_to_unmap_one) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | * ->private_lock (try_to_unmap_one) |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 110 | * ->i_pages lock (try_to_unmap_one) |
Hugh Dickins | 15b4473 | 2020-12-15 14:21:31 -0800 | [diff] [blame] | 111 | * ->lruvec->lru_lock (follow_page->mark_page_accessed) |
| 112 | * ->lruvec->lru_lock (check_pte_range->isolate_lru_page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 113 | * ->private_lock (page_remove_rmap->set_page_dirty) |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 114 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 115 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 116 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 117 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
Christoph Hellwig | f758eea | 2011-04-21 18:19:44 -0600 | [diff] [blame] | 118 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
Dave Chinner | 250df6e | 2011-03-22 22:23:36 +1100 | [diff] [blame] | 119 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 120 | * ->private_lock (zap_pte_range->block_dirty_folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 121 | * |
Davidlohr Bueso | c8c06ef | 2014-12-12 16:54:24 -0800 | [diff] [blame] | 122 | * ->i_mmap_rwsem |
Andi Kleen | 9a3c531 | 2012-03-21 16:34:09 -0700 | [diff] [blame] | 123 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | */ |
| 125 | |
Matthew Wilcox | 5c024e6a | 2017-11-21 09:17:59 -0500 | [diff] [blame] | 126 | static void page_cache_delete(struct address_space *mapping, |
Matthew Wilcox (Oracle) | a548b61 | 2021-05-08 00:35:49 -0400 | [diff] [blame] | 127 | struct folio *folio, void *shadow) |
Johannes Weiner | 91b0abe | 2014-04-03 14:47:49 -0700 | [diff] [blame] | 128 | { |
Matthew Wilcox (Oracle) | a548b61 | 2021-05-08 00:35:49 -0400 | [diff] [blame] | 129 | XA_STATE(xas, &mapping->i_pages, folio->index); |
| 130 | long nr = 1; |
Kirill A. Shutemov | c70b647 | 2016-12-12 16:43:17 -0800 | [diff] [blame] | 131 | |
Matthew Wilcox | 5c024e6a | 2017-11-21 09:17:59 -0500 | [diff] [blame] | 132 | mapping_set_update(&xas, mapping); |
| 133 | |
| 134 | /* hugetlb pages are represented by a single entry in the xarray */ |
Matthew Wilcox (Oracle) | a548b61 | 2021-05-08 00:35:49 -0400 | [diff] [blame] | 135 | if (!folio_test_hugetlb(folio)) { |
| 136 | xas_set_order(&xas, folio->index, folio_order(folio)); |
| 137 | nr = folio_nr_pages(folio); |
Matthew Wilcox | 5c024e6a | 2017-11-21 09:17:59 -0500 | [diff] [blame] | 138 | } |
Johannes Weiner | 91b0abe | 2014-04-03 14:47:49 -0700 | [diff] [blame] | 139 | |
Matthew Wilcox (Oracle) | a548b61 | 2021-05-08 00:35:49 -0400 | [diff] [blame] | 140 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
Johannes Weiner | 449dd69 | 2014-04-03 14:47:56 -0700 | [diff] [blame] | 141 | |
Matthew Wilcox | 5c024e6a | 2017-11-21 09:17:59 -0500 | [diff] [blame] | 142 | xas_store(&xas, shadow); |
| 143 | xas_init_marks(&xas); |
Johannes Weiner | d3798ae | 2016-10-04 22:02:08 +0200 | [diff] [blame] | 144 | |
Matthew Wilcox (Oracle) | a548b61 | 2021-05-08 00:35:49 -0400 | [diff] [blame] | 145 | folio->mapping = NULL; |
Jan Kara | 2300638 | 2017-11-15 17:37:26 -0800 | [diff] [blame] | 146 | /* Leave page->index set: truncation lookup relies upon it */ |
Johannes Weiner | d3798ae | 2016-10-04 22:02:08 +0200 | [diff] [blame] | 147 | mapping->nrpages -= nr; |
Johannes Weiner | 91b0abe | 2014-04-03 14:47:49 -0700 | [diff] [blame] | 148 | } |
| 149 | |
Matthew Wilcox (Oracle) | 621db48 | 2021-05-08 20:04:05 -0400 | [diff] [blame] | 150 | static void filemap_unaccount_folio(struct address_space *mapping, |
| 151 | struct folio *folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | { |
Matthew Wilcox (Oracle) | 621db48 | 2021-05-08 20:04:05 -0400 | [diff] [blame] | 153 | long nr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | |
Matthew Wilcox (Oracle) | 621db48 | 2021-05-08 20:04:05 -0400 | [diff] [blame] | 155 | VM_BUG_ON_FOLIO(folio_mapped(folio), folio); |
| 156 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) { |
Hugh Dickins | 06b241f | 2016-03-09 14:08:07 -0800 | [diff] [blame] | 157 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", |
Matthew Wilcox (Oracle) | 621db48 | 2021-05-08 20:04:05 -0400 | [diff] [blame] | 158 | current->comm, folio_pfn(folio)); |
| 159 | dump_page(&folio->page, "still mapped when deleted"); |
Hugh Dickins | 06b241f | 2016-03-09 14:08:07 -0800 | [diff] [blame] | 160 | dump_stack(); |
| 161 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
| 162 | |
Hugh Dickins | 85207ad | 2022-03-24 18:09:52 -0700 | [diff] [blame] | 163 | if (mapping_exiting(mapping) && !folio_test_large(folio)) { |
| 164 | int mapcount = page_mapcount(&folio->page); |
| 165 | |
| 166 | if (folio_ref_count(folio) >= mapcount + 2) { |
| 167 | /* |
| 168 | * All vmas have already been torn down, so it's |
| 169 | * a good bet that actually the page is unmapped |
| 170 | * and we'd rather not leak it: if we're wrong, |
| 171 | * another bad page check should catch it later. |
| 172 | */ |
| 173 | page_mapcount_reset(&folio->page); |
| 174 | folio_ref_sub(folio, mapcount); |
| 175 | } |
Hugh Dickins | 06b241f | 2016-03-09 14:08:07 -0800 | [diff] [blame] | 176 | } |
| 177 | } |
| 178 | |
Matthew Wilcox (Oracle) | 621db48 | 2021-05-08 20:04:05 -0400 | [diff] [blame] | 179 | /* hugetlb folios do not participate in page cache accounting. */ |
| 180 | if (folio_test_hugetlb(folio)) |
Jan Kara | 5ecc4d8 | 2017-11-15 17:37:29 -0800 | [diff] [blame] | 181 | return; |
Jan Kara | 76253fb | 2017-11-15 17:37:22 -0800 | [diff] [blame] | 182 | |
Matthew Wilcox (Oracle) | 621db48 | 2021-05-08 20:04:05 -0400 | [diff] [blame] | 183 | nr = folio_nr_pages(folio); |
Jan Kara | 5ecc4d8 | 2017-11-15 17:37:29 -0800 | [diff] [blame] | 184 | |
Matthew Wilcox (Oracle) | 621db48 | 2021-05-08 20:04:05 -0400 | [diff] [blame] | 185 | __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr); |
| 186 | if (folio_test_swapbacked(folio)) { |
| 187 | __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr); |
| 188 | if (folio_test_pmd_mappable(folio)) |
| 189 | __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr); |
| 190 | } else if (folio_test_pmd_mappable(folio)) { |
| 191 | __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr); |
Song Liu | 09d91cd | 2019-09-23 15:38:03 -0700 | [diff] [blame] | 192 | filemap_nr_thps_dec(mapping); |
Jan Kara | 76253fb | 2017-11-15 17:37:22 -0800 | [diff] [blame] | 193 | } |
Jan Kara | 5ecc4d8 | 2017-11-15 17:37:29 -0800 | [diff] [blame] | 194 | |
| 195 | /* |
Matthew Wilcox (Oracle) | 621db48 | 2021-05-08 20:04:05 -0400 | [diff] [blame] | 196 | * At this point folio must be either written or cleaned by |
| 197 | * truncate. Dirty folio here signals a bug and loss of |
Hugh Dickins | 566d336 | 2022-03-24 18:13:59 -0700 | [diff] [blame] | 198 | * unwritten data - on ordinary filesystems. |
Jan Kara | 5ecc4d8 | 2017-11-15 17:37:29 -0800 | [diff] [blame] | 199 | * |
Hugh Dickins | 566d336 | 2022-03-24 18:13:59 -0700 | [diff] [blame] | 200 | * But it's harmless on in-memory filesystems like tmpfs; and can |
| 201 | * occur when a driver which did get_user_pages() sets page dirty |
| 202 | * before putting it, while the inode is being finally evicted. |
| 203 | * |
| 204 | * Below fixes dirty accounting after removing the folio entirely |
Matthew Wilcox (Oracle) | 621db48 | 2021-05-08 20:04:05 -0400 | [diff] [blame] | 205 | * but leaves the dirty flag set: it has no effect for truncated |
| 206 | * folio and anyway will be cleared before returning folio to |
Jan Kara | 5ecc4d8 | 2017-11-15 17:37:29 -0800 | [diff] [blame] | 207 | * buddy allocator. |
| 208 | */ |
Hugh Dickins | 566d336 | 2022-03-24 18:13:59 -0700 | [diff] [blame] | 209 | if (WARN_ON_ONCE(folio_test_dirty(folio) && |
| 210 | mapping_can_writeback(mapping))) |
| 211 | folio_account_cleaned(folio, inode_to_wb(mapping->host)); |
Jan Kara | 5ecc4d8 | 2017-11-15 17:37:29 -0800 | [diff] [blame] | 212 | } |
| 213 | |
| 214 | /* |
| 215 | * Delete a page from the page cache and free it. Caller has to make |
| 216 | * sure the page is locked and that nobody else uses it - or that usage |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 217 | * is safe. The caller must hold the i_pages lock. |
Jan Kara | 5ecc4d8 | 2017-11-15 17:37:29 -0800 | [diff] [blame] | 218 | */ |
Matthew Wilcox (Oracle) | 452e9e6 | 2021-05-09 09:33:42 -0400 | [diff] [blame] | 219 | void __filemap_remove_folio(struct folio *folio, void *shadow) |
Jan Kara | 5ecc4d8 | 2017-11-15 17:37:29 -0800 | [diff] [blame] | 220 | { |
Matthew Wilcox (Oracle) | 452e9e6 | 2021-05-09 09:33:42 -0400 | [diff] [blame] | 221 | struct address_space *mapping = folio->mapping; |
Jan Kara | 5ecc4d8 | 2017-11-15 17:37:29 -0800 | [diff] [blame] | 222 | |
Matthew Wilcox (Oracle) | a0580c6 | 2021-07-23 09:29:46 -0400 | [diff] [blame] | 223 | trace_mm_filemap_delete_from_page_cache(folio); |
Matthew Wilcox (Oracle) | 621db48 | 2021-05-08 20:04:05 -0400 | [diff] [blame] | 224 | filemap_unaccount_folio(mapping, folio); |
Matthew Wilcox (Oracle) | a548b61 | 2021-05-08 00:35:49 -0400 | [diff] [blame] | 225 | page_cache_delete(mapping, folio, shadow); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 226 | } |
| 227 | |
Matthew Wilcox (Oracle) | 78f4266 | 2021-07-28 15:52:34 -0400 | [diff] [blame] | 228 | void filemap_free_folio(struct address_space *mapping, struct folio *folio) |
Jan Kara | 59c66c5 | 2017-11-15 17:37:18 -0800 | [diff] [blame] | 229 | { |
Matthew Wilcox (Oracle) | d2329aa | 2022-05-01 07:35:31 -0400 | [diff] [blame] | 230 | void (*free_folio)(struct folio *); |
Matthew Wilcox (Oracle) | 3abb28e | 2022-01-07 13:03:48 -0500 | [diff] [blame] | 231 | int refs = 1; |
Jan Kara | 59c66c5 | 2017-11-15 17:37:18 -0800 | [diff] [blame] | 232 | |
Matthew Wilcox (Oracle) | d2329aa | 2022-05-01 07:35:31 -0400 | [diff] [blame] | 233 | free_folio = mapping->a_ops->free_folio; |
| 234 | if (free_folio) |
| 235 | free_folio(folio); |
Jan Kara | 59c66c5 | 2017-11-15 17:37:18 -0800 | [diff] [blame] | 236 | |
Matthew Wilcox (Oracle) | 3abb28e | 2022-01-07 13:03:48 -0500 | [diff] [blame] | 237 | if (folio_test_large(folio) && !folio_test_hugetlb(folio)) |
| 238 | refs = folio_nr_pages(folio); |
| 239 | folio_put_refs(folio, refs); |
Jan Kara | 59c66c5 | 2017-11-15 17:37:18 -0800 | [diff] [blame] | 240 | } |
| 241 | |
Minchan Kim | 702cfbf | 2011-03-22 16:32:43 -0700 | [diff] [blame] | 242 | /** |
Matthew Wilcox (Oracle) | 452e9e6 | 2021-05-09 09:33:42 -0400 | [diff] [blame] | 243 | * filemap_remove_folio - Remove folio from page cache. |
| 244 | * @folio: The folio. |
Minchan Kim | 702cfbf | 2011-03-22 16:32:43 -0700 | [diff] [blame] | 245 | * |
Matthew Wilcox (Oracle) | 452e9e6 | 2021-05-09 09:33:42 -0400 | [diff] [blame] | 246 | * This must be called only on folios that are locked and have been |
| 247 | * verified to be in the page cache. It will never put the folio into |
| 248 | * the free list because the caller has a reference on the page. |
Minchan Kim | 702cfbf | 2011-03-22 16:32:43 -0700 | [diff] [blame] | 249 | */ |
Matthew Wilcox (Oracle) | 452e9e6 | 2021-05-09 09:33:42 -0400 | [diff] [blame] | 250 | void filemap_remove_folio(struct folio *folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | { |
Matthew Wilcox (Oracle) | 452e9e6 | 2021-05-09 09:33:42 -0400 | [diff] [blame] | 252 | struct address_space *mapping = folio->mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 253 | |
Matthew Wilcox (Oracle) | 452e9e6 | 2021-05-09 09:33:42 -0400 | [diff] [blame] | 254 | BUG_ON(!folio_test_locked(folio)); |
Johannes Weiner | 51b8c1f | 2021-11-08 18:31:24 -0800 | [diff] [blame] | 255 | spin_lock(&mapping->host->i_lock); |
Johannes Weiner | 3047250 | 2021-09-02 14:53:18 -0700 | [diff] [blame] | 256 | xa_lock_irq(&mapping->i_pages); |
Matthew Wilcox (Oracle) | 452e9e6 | 2021-05-09 09:33:42 -0400 | [diff] [blame] | 257 | __filemap_remove_folio(folio, NULL); |
Johannes Weiner | 3047250 | 2021-09-02 14:53:18 -0700 | [diff] [blame] | 258 | xa_unlock_irq(&mapping->i_pages); |
Johannes Weiner | 51b8c1f | 2021-11-08 18:31:24 -0800 | [diff] [blame] | 259 | if (mapping_shrinkable(mapping)) |
| 260 | inode_add_lru(mapping->host); |
| 261 | spin_unlock(&mapping->host->i_lock); |
Linus Torvalds | 6072d13 | 2010-12-01 13:35:19 -0500 | [diff] [blame] | 262 | |
Matthew Wilcox (Oracle) | 452e9e6 | 2021-05-09 09:33:42 -0400 | [diff] [blame] | 263 | filemap_free_folio(mapping, folio); |
Minchan Kim | 97cecb5 | 2011-03-22 16:30:53 -0700 | [diff] [blame] | 264 | } |
Minchan Kim | 97cecb5 | 2011-03-22 16:30:53 -0700 | [diff] [blame] | 265 | |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 266 | /* |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 267 | * page_cache_delete_batch - delete several folios from page cache |
| 268 | * @mapping: the mapping to which folios belong |
| 269 | * @fbatch: batch of folios to delete |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 270 | * |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 271 | * The function walks over mapping->i_pages and removes folios passed in |
| 272 | * @fbatch from the mapping. The function expects @fbatch to be sorted |
| 273 | * by page index and is optimised for it to be dense. |
| 274 | * It tolerates holes in @fbatch (mapping entries at those indices are not |
| 275 | * modified). |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 276 | * |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 277 | * The function expects the i_pages lock to be held. |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 278 | */ |
Matthew Wilcox | ef8e571 | 2017-12-04 03:59:45 -0500 | [diff] [blame] | 279 | static void page_cache_delete_batch(struct address_space *mapping, |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 280 | struct folio_batch *fbatch) |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 281 | { |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 282 | XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index); |
Matthew Wilcox (Oracle) | 6b24ca4 | 2020-06-27 22:19:08 -0400 | [diff] [blame] | 283 | long total_pages = 0; |
Matthew Wilcox (Oracle) | 4101196 | 2019-09-23 15:34:52 -0700 | [diff] [blame] | 284 | int i = 0; |
Matthew Wilcox (Oracle) | 1afd7ae | 2021-03-12 23:13:46 -0500 | [diff] [blame] | 285 | struct folio *folio; |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 286 | |
Matthew Wilcox | ef8e571 | 2017-12-04 03:59:45 -0500 | [diff] [blame] | 287 | mapping_set_update(&xas, mapping); |
Matthew Wilcox (Oracle) | 1afd7ae | 2021-03-12 23:13:46 -0500 | [diff] [blame] | 288 | xas_for_each(&xas, folio, ULONG_MAX) { |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 289 | if (i >= folio_batch_count(fbatch)) |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 290 | break; |
Matthew Wilcox (Oracle) | 4101196 | 2019-09-23 15:34:52 -0700 | [diff] [blame] | 291 | |
| 292 | /* A swap/dax/shadow entry got inserted? Skip it. */ |
Matthew Wilcox (Oracle) | 1afd7ae | 2021-03-12 23:13:46 -0500 | [diff] [blame] | 293 | if (xa_is_value(folio)) |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 294 | continue; |
Matthew Wilcox (Oracle) | 4101196 | 2019-09-23 15:34:52 -0700 | [diff] [blame] | 295 | /* |
| 296 | * A page got inserted in our range? Skip it. We have our |
| 297 | * pages locked so they are protected from being removed. |
| 298 | * If we see a page whose index is higher than ours, it |
| 299 | * means our page has been removed, which shouldn't be |
| 300 | * possible because we're holding the PageLock. |
| 301 | */ |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 302 | if (folio != fbatch->folios[i]) { |
Matthew Wilcox (Oracle) | 1afd7ae | 2021-03-12 23:13:46 -0500 | [diff] [blame] | 303 | VM_BUG_ON_FOLIO(folio->index > |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 304 | fbatch->folios[i]->index, folio); |
Matthew Wilcox (Oracle) | 4101196 | 2019-09-23 15:34:52 -0700 | [diff] [blame] | 305 | continue; |
Linus Torvalds | 69bf4b6 | 2019-07-05 19:55:18 -0700 | [diff] [blame] | 306 | } |
Matthew Wilcox (Oracle) | 4101196 | 2019-09-23 15:34:52 -0700 | [diff] [blame] | 307 | |
Matthew Wilcox (Oracle) | 1afd7ae | 2021-03-12 23:13:46 -0500 | [diff] [blame] | 308 | WARN_ON_ONCE(!folio_test_locked(folio)); |
Matthew Wilcox (Oracle) | 4101196 | 2019-09-23 15:34:52 -0700 | [diff] [blame] | 309 | |
Matthew Wilcox (Oracle) | 6b24ca4 | 2020-06-27 22:19:08 -0400 | [diff] [blame] | 310 | folio->mapping = NULL; |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 311 | /* Leave folio->index set: truncation lookup relies on it */ |
Matthew Wilcox (Oracle) | 4101196 | 2019-09-23 15:34:52 -0700 | [diff] [blame] | 312 | |
Matthew Wilcox (Oracle) | 6b24ca4 | 2020-06-27 22:19:08 -0400 | [diff] [blame] | 313 | i++; |
Matthew Wilcox | ef8e571 | 2017-12-04 03:59:45 -0500 | [diff] [blame] | 314 | xas_store(&xas, NULL); |
Matthew Wilcox (Oracle) | 6b24ca4 | 2020-06-27 22:19:08 -0400 | [diff] [blame] | 315 | total_pages += folio_nr_pages(folio); |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 316 | } |
| 317 | mapping->nrpages -= total_pages; |
| 318 | } |
| 319 | |
| 320 | void delete_from_page_cache_batch(struct address_space *mapping, |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 321 | struct folio_batch *fbatch) |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 322 | { |
| 323 | int i; |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 324 | |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 325 | if (!folio_batch_count(fbatch)) |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 326 | return; |
| 327 | |
Johannes Weiner | 51b8c1f | 2021-11-08 18:31:24 -0800 | [diff] [blame] | 328 | spin_lock(&mapping->host->i_lock); |
Johannes Weiner | 3047250 | 2021-09-02 14:53:18 -0700 | [diff] [blame] | 329 | xa_lock_irq(&mapping->i_pages); |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 330 | for (i = 0; i < folio_batch_count(fbatch); i++) { |
| 331 | struct folio *folio = fbatch->folios[i]; |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 332 | |
Matthew Wilcox (Oracle) | a0580c6 | 2021-07-23 09:29:46 -0400 | [diff] [blame] | 333 | trace_mm_filemap_delete_from_page_cache(folio); |
| 334 | filemap_unaccount_folio(mapping, folio); |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 335 | } |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 336 | page_cache_delete_batch(mapping, fbatch); |
Johannes Weiner | 3047250 | 2021-09-02 14:53:18 -0700 | [diff] [blame] | 337 | xa_unlock_irq(&mapping->i_pages); |
Johannes Weiner | 51b8c1f | 2021-11-08 18:31:24 -0800 | [diff] [blame] | 338 | if (mapping_shrinkable(mapping)) |
| 339 | inode_add_lru(mapping->host); |
| 340 | spin_unlock(&mapping->host->i_lock); |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 341 | |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 342 | for (i = 0; i < folio_batch_count(fbatch); i++) |
| 343 | filemap_free_folio(mapping, fbatch->folios[i]); |
Jan Kara | aa65c29 | 2017-11-15 17:37:33 -0800 | [diff] [blame] | 344 | } |
| 345 | |
Miklos Szeredi | d72d9e2 | 2016-07-29 14:10:57 +0200 | [diff] [blame] | 346 | int filemap_check_errors(struct address_space *mapping) |
Dmitry Monakhov | 865ffef3 | 2013-04-29 15:08:42 -0700 | [diff] [blame] | 347 | { |
| 348 | int ret = 0; |
| 349 | /* Check for outstanding write errors */ |
Jens Axboe | 7fcbbaf | 2014-05-22 11:54:16 -0700 | [diff] [blame] | 350 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
| 351 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) |
Dmitry Monakhov | 865ffef3 | 2013-04-29 15:08:42 -0700 | [diff] [blame] | 352 | ret = -ENOSPC; |
Jens Axboe | 7fcbbaf | 2014-05-22 11:54:16 -0700 | [diff] [blame] | 353 | if (test_bit(AS_EIO, &mapping->flags) && |
| 354 | test_and_clear_bit(AS_EIO, &mapping->flags)) |
Dmitry Monakhov | 865ffef3 | 2013-04-29 15:08:42 -0700 | [diff] [blame] | 355 | ret = -EIO; |
| 356 | return ret; |
| 357 | } |
Miklos Szeredi | d72d9e2 | 2016-07-29 14:10:57 +0200 | [diff] [blame] | 358 | EXPORT_SYMBOL(filemap_check_errors); |
Dmitry Monakhov | 865ffef3 | 2013-04-29 15:08:42 -0700 | [diff] [blame] | 359 | |
Jeff Layton | 76341ca | 2017-07-06 07:02:22 -0400 | [diff] [blame] | 360 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
| 361 | { |
| 362 | /* Check for outstanding write errors */ |
| 363 | if (test_bit(AS_EIO, &mapping->flags)) |
| 364 | return -EIO; |
| 365 | if (test_bit(AS_ENOSPC, &mapping->flags)) |
| 366 | return -ENOSPC; |
| 367 | return 0; |
| 368 | } |
| 369 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | /** |
Josef Bacik | 5a79849 | 2021-07-14 14:47:22 -0400 | [diff] [blame] | 371 | * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range |
| 372 | * @mapping: address space structure to write |
| 373 | * @wbc: the writeback_control controlling the writeout |
| 374 | * |
| 375 | * Call writepages on the mapping using the provided wbc to control the |
| 376 | * writeout. |
| 377 | * |
| 378 | * Return: %0 on success, negative error code otherwise. |
| 379 | */ |
| 380 | int filemap_fdatawrite_wbc(struct address_space *mapping, |
| 381 | struct writeback_control *wbc) |
| 382 | { |
| 383 | int ret; |
| 384 | |
| 385 | if (!mapping_can_writeback(mapping) || |
| 386 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
| 387 | return 0; |
| 388 | |
| 389 | wbc_attach_fdatawrite_inode(wbc, mapping->host); |
| 390 | ret = do_writepages(mapping, wbc); |
| 391 | wbc_detach_inode(wbc); |
| 392 | return ret; |
| 393 | } |
| 394 | EXPORT_SYMBOL(filemap_fdatawrite_wbc); |
| 395 | |
| 396 | /** |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 397 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 398 | * @mapping: address space structure to write |
| 399 | * @start: offset in bytes where the range starts |
Andrew Morton | 469eb4d | 2006-03-24 03:17:45 -0800 | [diff] [blame] | 400 | * @end: offset in bytes where the range ends (inclusive) |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 401 | * @sync_mode: enable synchronous operation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 402 | * |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 403 | * Start writeback against all of a mapping's dirty pages that lie |
| 404 | * within the byte offsets <start, end> inclusive. |
| 405 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 406 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 407 | * opposed to a regular memory cleansing writeback. The difference between |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 408 | * these two operations is that if a dirty page/buffer is encountered, it must |
| 409 | * be waited upon, and not just skipped over. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 410 | * |
| 411 | * Return: %0 on success, negative error code otherwise. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 412 | */ |
Andrew Morton | ebcf28e | 2006-03-24 03:18:04 -0800 | [diff] [blame] | 413 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
| 414 | loff_t end, int sync_mode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 415 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | struct writeback_control wbc = { |
| 417 | .sync_mode = sync_mode, |
Nick Piggin | 05fe478 | 2009-01-06 14:39:08 -0800 | [diff] [blame] | 418 | .nr_to_write = LONG_MAX, |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 419 | .range_start = start, |
| 420 | .range_end = end, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | }; |
| 422 | |
Josef Bacik | 5a79849 | 2021-07-14 14:47:22 -0400 | [diff] [blame] | 423 | return filemap_fdatawrite_wbc(mapping, &wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 424 | } |
| 425 | |
| 426 | static inline int __filemap_fdatawrite(struct address_space *mapping, |
| 427 | int sync_mode) |
| 428 | { |
OGAWA Hirofumi | 111ebb6 | 2006-06-23 02:03:26 -0700 | [diff] [blame] | 429 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 430 | } |
| 431 | |
| 432 | int filemap_fdatawrite(struct address_space *mapping) |
| 433 | { |
| 434 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); |
| 435 | } |
| 436 | EXPORT_SYMBOL(filemap_fdatawrite); |
| 437 | |
Jan Kara | f4c0a0f | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 438 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
Andrew Morton | ebcf28e | 2006-03-24 03:18:04 -0800 | [diff] [blame] | 439 | loff_t end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 440 | { |
| 441 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); |
| 442 | } |
Jan Kara | f4c0a0f | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 443 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 444 | |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 445 | /** |
| 446 | * filemap_flush - mostly a non-blocking flush |
| 447 | * @mapping: target address_space |
| 448 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 449 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
| 450 | * purposes - I/O may not be started against all dirty pages. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 451 | * |
| 452 | * Return: %0 on success, negative error code otherwise. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | */ |
| 454 | int filemap_flush(struct address_space *mapping) |
| 455 | { |
| 456 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); |
| 457 | } |
| 458 | EXPORT_SYMBOL(filemap_flush); |
| 459 | |
Goldwyn Rodrigues | 7fc9e47 | 2017-06-20 07:05:41 -0500 | [diff] [blame] | 460 | /** |
| 461 | * filemap_range_has_page - check if a page exists in range. |
| 462 | * @mapping: address space within which to check |
| 463 | * @start_byte: offset in bytes where the range starts |
| 464 | * @end_byte: offset in bytes where the range ends (inclusive) |
| 465 | * |
| 466 | * Find at least one page in the range supplied, usually used to check if |
| 467 | * direct writing in this range will trigger a writeback. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 468 | * |
| 469 | * Return: %true if at least one page exists in the specified range, |
| 470 | * %false otherwise. |
Goldwyn Rodrigues | 7fc9e47 | 2017-06-20 07:05:41 -0500 | [diff] [blame] | 471 | */ |
| 472 | bool filemap_range_has_page(struct address_space *mapping, |
| 473 | loff_t start_byte, loff_t end_byte) |
| 474 | { |
Matthew Wilcox (Oracle) | eff3b36 | 2023-01-16 19:39:40 +0000 | [diff] [blame] | 475 | struct folio *folio; |
Matthew Wilcox | 8fa8e53 | 2018-01-16 06:26:49 -0500 | [diff] [blame] | 476 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
| 477 | pgoff_t max = end_byte >> PAGE_SHIFT; |
Goldwyn Rodrigues | 7fc9e47 | 2017-06-20 07:05:41 -0500 | [diff] [blame] | 478 | |
| 479 | if (end_byte < start_byte) |
| 480 | return false; |
| 481 | |
Matthew Wilcox | 8fa8e53 | 2018-01-16 06:26:49 -0500 | [diff] [blame] | 482 | rcu_read_lock(); |
| 483 | for (;;) { |
Matthew Wilcox (Oracle) | eff3b36 | 2023-01-16 19:39:40 +0000 | [diff] [blame] | 484 | folio = xas_find(&xas, max); |
| 485 | if (xas_retry(&xas, folio)) |
Matthew Wilcox | 8fa8e53 | 2018-01-16 06:26:49 -0500 | [diff] [blame] | 486 | continue; |
| 487 | /* Shadow entries don't count */ |
Matthew Wilcox (Oracle) | eff3b36 | 2023-01-16 19:39:40 +0000 | [diff] [blame] | 488 | if (xa_is_value(folio)) |
Matthew Wilcox | 8fa8e53 | 2018-01-16 06:26:49 -0500 | [diff] [blame] | 489 | continue; |
| 490 | /* |
| 491 | * We don't need to try to pin this page; we're about to |
| 492 | * release the RCU lock anyway. It is enough to know that |
| 493 | * there was a page here recently. |
| 494 | */ |
| 495 | break; |
| 496 | } |
| 497 | rcu_read_unlock(); |
Goldwyn Rodrigues | 7fc9e47 | 2017-06-20 07:05:41 -0500 | [diff] [blame] | 498 | |
Matthew Wilcox (Oracle) | eff3b36 | 2023-01-16 19:39:40 +0000 | [diff] [blame] | 499 | return folio != NULL; |
Goldwyn Rodrigues | 7fc9e47 | 2017-06-20 07:05:41 -0500 | [diff] [blame] | 500 | } |
| 501 | EXPORT_SYMBOL(filemap_range_has_page); |
| 502 | |
Jeff Layton | 5e8fcc1 | 2017-07-06 07:02:24 -0400 | [diff] [blame] | 503 | static void __filemap_fdatawait_range(struct address_space *mapping, |
Junichi Nomura | aa750fd | 2015-11-05 18:47:23 -0800 | [diff] [blame] | 504 | loff_t start_byte, loff_t end_byte) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 505 | { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 506 | pgoff_t index = start_byte >> PAGE_SHIFT; |
| 507 | pgoff_t end = end_byte >> PAGE_SHIFT; |
Vishal Moola (Oracle) | 6817ef5 | 2023-01-04 13:14:28 -0800 | [diff] [blame] | 508 | struct folio_batch fbatch; |
| 509 | unsigned nr_folios; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 510 | |
Vishal Moola (Oracle) | 6817ef5 | 2023-01-04 13:14:28 -0800 | [diff] [blame] | 511 | folio_batch_init(&fbatch); |
| 512 | |
Jan Kara | 312e9d2 | 2017-11-15 17:35:05 -0800 | [diff] [blame] | 513 | while (index <= end) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 514 | unsigned i; |
| 515 | |
Vishal Moola (Oracle) | 6817ef5 | 2023-01-04 13:14:28 -0800 | [diff] [blame] | 516 | nr_folios = filemap_get_folios_tag(mapping, &index, end, |
| 517 | PAGECACHE_TAG_WRITEBACK, &fbatch); |
| 518 | |
| 519 | if (!nr_folios) |
Jan Kara | 312e9d2 | 2017-11-15 17:35:05 -0800 | [diff] [blame] | 520 | break; |
| 521 | |
Vishal Moola (Oracle) | 6817ef5 | 2023-01-04 13:14:28 -0800 | [diff] [blame] | 522 | for (i = 0; i < nr_folios; i++) { |
| 523 | struct folio *folio = fbatch.folios[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 524 | |
Vishal Moola (Oracle) | 6817ef5 | 2023-01-04 13:14:28 -0800 | [diff] [blame] | 525 | folio_wait_writeback(folio); |
| 526 | folio_clear_error(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | } |
Vishal Moola (Oracle) | 6817ef5 | 2023-01-04 13:14:28 -0800 | [diff] [blame] | 528 | folio_batch_release(&fbatch); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 529 | cond_resched(); |
| 530 | } |
Junichi Nomura | aa750fd | 2015-11-05 18:47:23 -0800 | [diff] [blame] | 531 | } |
| 532 | |
| 533 | /** |
| 534 | * filemap_fdatawait_range - wait for writeback to complete |
| 535 | * @mapping: address space structure to wait for |
| 536 | * @start_byte: offset in bytes where the range starts |
| 537 | * @end_byte: offset in bytes where the range ends (inclusive) |
| 538 | * |
| 539 | * Walk the list of under-writeback pages of the given address space |
| 540 | * in the given range and wait for all of them. Check error status of |
| 541 | * the address space and return it. |
| 542 | * |
| 543 | * Since the error status of the address space is cleared by this function, |
| 544 | * callers are responsible for checking the return value and handling and/or |
| 545 | * reporting the error. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 546 | * |
| 547 | * Return: error status of the address space. |
Junichi Nomura | aa750fd | 2015-11-05 18:47:23 -0800 | [diff] [blame] | 548 | */ |
| 549 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
| 550 | loff_t end_byte) |
| 551 | { |
Jeff Layton | 5e8fcc1 | 2017-07-06 07:02:24 -0400 | [diff] [blame] | 552 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
| 553 | return filemap_check_errors(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 554 | } |
Jan Kara | d3bccb6f | 2009-08-17 19:30:27 +0200 | [diff] [blame] | 555 | EXPORT_SYMBOL(filemap_fdatawait_range); |
| 556 | |
| 557 | /** |
Ross Zwisler | aa0bfcd | 2019-06-20 17:05:37 -0400 | [diff] [blame] | 558 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete |
| 559 | * @mapping: address space structure to wait for |
| 560 | * @start_byte: offset in bytes where the range starts |
| 561 | * @end_byte: offset in bytes where the range ends (inclusive) |
| 562 | * |
| 563 | * Walk the list of under-writeback pages of the given address space in the |
| 564 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), |
| 565 | * this function does not clear error status of the address space. |
| 566 | * |
| 567 | * Use this function if callers don't handle errors themselves. Expected |
| 568 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), |
| 569 | * fsfreeze(8) |
| 570 | */ |
| 571 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, |
| 572 | loff_t start_byte, loff_t end_byte) |
| 573 | { |
| 574 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
| 575 | return filemap_check_and_keep_errors(mapping); |
| 576 | } |
| 577 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); |
| 578 | |
| 579 | /** |
Jeff Layton | a823e45 | 2017-07-28 07:24:43 -0400 | [diff] [blame] | 580 | * file_fdatawait_range - wait for writeback to complete |
| 581 | * @file: file pointing to address space structure to wait for |
| 582 | * @start_byte: offset in bytes where the range starts |
| 583 | * @end_byte: offset in bytes where the range ends (inclusive) |
| 584 | * |
| 585 | * Walk the list of under-writeback pages of the address space that file |
| 586 | * refers to, in the given range and wait for all of them. Check error |
| 587 | * status of the address space vs. the file->f_wb_err cursor and return it. |
| 588 | * |
| 589 | * Since the error status of the file is advanced by this function, |
| 590 | * callers are responsible for checking the return value and handling and/or |
| 591 | * reporting the error. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 592 | * |
| 593 | * Return: error status of the address space vs. the file->f_wb_err cursor. |
Jeff Layton | a823e45 | 2017-07-28 07:24:43 -0400 | [diff] [blame] | 594 | */ |
| 595 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) |
| 596 | { |
| 597 | struct address_space *mapping = file->f_mapping; |
| 598 | |
| 599 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
| 600 | return file_check_and_advance_wb_err(file); |
| 601 | } |
| 602 | EXPORT_SYMBOL(file_fdatawait_range); |
| 603 | |
| 604 | /** |
Junichi Nomura | aa750fd | 2015-11-05 18:47:23 -0800 | [diff] [blame] | 605 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors |
| 606 | * @mapping: address space structure to wait for |
| 607 | * |
| 608 | * Walk the list of under-writeback pages of the given address space |
| 609 | * and wait for all of them. Unlike filemap_fdatawait(), this function |
| 610 | * does not clear error status of the address space. |
| 611 | * |
| 612 | * Use this function if callers don't handle errors themselves. Expected |
| 613 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), |
| 614 | * fsfreeze(8) |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 615 | * |
| 616 | * Return: error status of the address space. |
Junichi Nomura | aa750fd | 2015-11-05 18:47:23 -0800 | [diff] [blame] | 617 | */ |
Jeff Layton | 76341ca | 2017-07-06 07:02:22 -0400 | [diff] [blame] | 618 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
Junichi Nomura | aa750fd | 2015-11-05 18:47:23 -0800 | [diff] [blame] | 619 | { |
Jeff Layton | ffb959b | 2017-07-31 10:29:38 -0400 | [diff] [blame] | 620 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
Jeff Layton | 76341ca | 2017-07-06 07:02:22 -0400 | [diff] [blame] | 621 | return filemap_check_and_keep_errors(mapping); |
Junichi Nomura | aa750fd | 2015-11-05 18:47:23 -0800 | [diff] [blame] | 622 | } |
Jeff Layton | 76341ca | 2017-07-06 07:02:22 -0400 | [diff] [blame] | 623 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
Junichi Nomura | aa750fd | 2015-11-05 18:47:23 -0800 | [diff] [blame] | 624 | |
Konstantin Khlebnikov | 875d91b | 2019-09-23 15:34:48 -0700 | [diff] [blame] | 625 | /* Returns true if writeback might be needed or already in progress. */ |
Jeff Layton | 9326c9b | 2017-07-26 10:21:11 -0400 | [diff] [blame] | 626 | static bool mapping_needs_writeback(struct address_space *mapping) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 627 | { |
Konstantin Khlebnikov | 875d91b | 2019-09-23 15:34:48 -0700 | [diff] [blame] | 628 | return mapping->nrpages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 629 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 630 | |
Jens Axboe | 4bdcd1d | 2021-10-28 08:47:05 -0600 | [diff] [blame] | 631 | bool filemap_range_has_writeback(struct address_space *mapping, |
| 632 | loff_t start_byte, loff_t end_byte) |
Jens Axboe | f8ee890 | 2021-11-05 13:37:13 -0700 | [diff] [blame] | 633 | { |
| 634 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
| 635 | pgoff_t max = end_byte >> PAGE_SHIFT; |
Vishal Moola (Oracle) | b05f41a | 2022-09-05 14:45:57 -0700 | [diff] [blame] | 636 | struct folio *folio; |
Jens Axboe | f8ee890 | 2021-11-05 13:37:13 -0700 | [diff] [blame] | 637 | |
| 638 | if (end_byte < start_byte) |
| 639 | return false; |
| 640 | |
| 641 | rcu_read_lock(); |
Vishal Moola (Oracle) | b05f41a | 2022-09-05 14:45:57 -0700 | [diff] [blame] | 642 | xas_for_each(&xas, folio, max) { |
| 643 | if (xas_retry(&xas, folio)) |
Jens Axboe | f8ee890 | 2021-11-05 13:37:13 -0700 | [diff] [blame] | 644 | continue; |
Vishal Moola (Oracle) | b05f41a | 2022-09-05 14:45:57 -0700 | [diff] [blame] | 645 | if (xa_is_value(folio)) |
Jens Axboe | f8ee890 | 2021-11-05 13:37:13 -0700 | [diff] [blame] | 646 | continue; |
Vishal Moola (Oracle) | b05f41a | 2022-09-05 14:45:57 -0700 | [diff] [blame] | 647 | if (folio_test_dirty(folio) || folio_test_locked(folio) || |
| 648 | folio_test_writeback(folio)) |
Jens Axboe | f8ee890 | 2021-11-05 13:37:13 -0700 | [diff] [blame] | 649 | break; |
| 650 | } |
| 651 | rcu_read_unlock(); |
Vishal Moola (Oracle) | b05f41a | 2022-09-05 14:45:57 -0700 | [diff] [blame] | 652 | return folio != NULL; |
Jens Axboe | f8ee890 | 2021-11-05 13:37:13 -0700 | [diff] [blame] | 653 | } |
Jens Axboe | 4bdcd1d | 2021-10-28 08:47:05 -0600 | [diff] [blame] | 654 | EXPORT_SYMBOL_GPL(filemap_range_has_writeback); |
Jens Axboe | 63135aa | 2021-04-29 22:55:18 -0700 | [diff] [blame] | 655 | |
| 656 | /** |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 657 | * filemap_write_and_wait_range - write out & wait on a file range |
| 658 | * @mapping: the address_space for the pages |
| 659 | * @lstart: offset in bytes where the range starts |
| 660 | * @lend: offset in bytes where the range ends (inclusive) |
| 661 | * |
Andrew Morton | 469eb4d | 2006-03-24 03:17:45 -0800 | [diff] [blame] | 662 | * Write out and wait upon file offsets lstart->lend, inclusive. |
| 663 | * |
mchehab@s-opensource.com | 0e056eb | 2017-03-30 17:11:36 -0300 | [diff] [blame] | 664 | * Note that @lend is inclusive (describes the last byte to be written) so |
Andrew Morton | 469eb4d | 2006-03-24 03:17:45 -0800 | [diff] [blame] | 665 | * that this function can be used to write to the very end-of-file (end = -1). |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 666 | * |
| 667 | * Return: error status of the address space. |
Andrew Morton | 469eb4d | 2006-03-24 03:17:45 -0800 | [diff] [blame] | 668 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 669 | int filemap_write_and_wait_range(struct address_space *mapping, |
| 670 | loff_t lstart, loff_t lend) |
| 671 | { |
Miaohe Lin | ccac11d | 2022-06-27 21:23:51 +0800 | [diff] [blame] | 672 | int err = 0, err2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | |
Brian Foster | feeb9b2 | 2022-11-28 10:56:31 -0500 | [diff] [blame] | 674 | if (lend < lstart) |
| 675 | return 0; |
| 676 | |
Jeff Layton | 9326c9b | 2017-07-26 10:21:11 -0400 | [diff] [blame] | 677 | if (mapping_needs_writeback(mapping)) { |
OGAWA Hirofumi | 28fd129 | 2006-01-08 01:02:14 -0800 | [diff] [blame] | 678 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
| 679 | WB_SYNC_ALL); |
Ira Weiny | ddf8f37 | 2020-01-30 22:12:07 -0800 | [diff] [blame] | 680 | /* |
| 681 | * Even if the above returned error, the pages may be |
| 682 | * written partially (e.g. -ENOSPC), so we wait for it. |
| 683 | * But the -EIO is special case, it may indicate the worst |
| 684 | * thing (e.g. bug) happened, so we avoid waiting for it. |
| 685 | */ |
Miaohe Lin | ccac11d | 2022-06-27 21:23:51 +0800 | [diff] [blame] | 686 | if (err != -EIO) |
| 687 | __filemap_fdatawait_range(mapping, lstart, lend); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | } |
Miaohe Lin | ccac11d | 2022-06-27 21:23:51 +0800 | [diff] [blame] | 689 | err2 = filemap_check_errors(mapping); |
| 690 | if (!err) |
| 691 | err = err2; |
OGAWA Hirofumi | 28fd129 | 2006-01-08 01:02:14 -0800 | [diff] [blame] | 692 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | } |
Chris Mason | f699558 | 2009-04-15 13:22:37 -0400 | [diff] [blame] | 694 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 695 | |
Jeff Layton | 5660e13 | 2017-07-06 07:02:25 -0400 | [diff] [blame] | 696 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
| 697 | { |
Jeff Layton | 3acdfd2 | 2017-07-24 06:22:15 -0400 | [diff] [blame] | 698 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
Jeff Layton | 5660e13 | 2017-07-06 07:02:25 -0400 | [diff] [blame] | 699 | |
| 700 | trace_filemap_set_wb_err(mapping, eseq); |
| 701 | } |
| 702 | EXPORT_SYMBOL(__filemap_set_wb_err); |
| 703 | |
| 704 | /** |
| 705 | * file_check_and_advance_wb_err - report wb error (if any) that was previously |
| 706 | * and advance wb_err to current one |
| 707 | * @file: struct file on which the error is being reported |
| 708 | * |
| 709 | * When userland calls fsync (or something like nfsd does the equivalent), we |
| 710 | * want to report any writeback errors that occurred since the last fsync (or |
| 711 | * since the file was opened if there haven't been any). |
| 712 | * |
| 713 | * Grab the wb_err from the mapping. If it matches what we have in the file, |
| 714 | * then just quickly return 0. The file is all caught up. |
| 715 | * |
| 716 | * If it doesn't match, then take the mapping value, set the "seen" flag in |
| 717 | * it and try to swap it into place. If it works, or another task beat us |
| 718 | * to it with the new value, then update the f_wb_err and return the error |
| 719 | * portion. The error at this point must be reported via proper channels |
| 720 | * (a'la fsync, or NFS COMMIT operation, etc.). |
| 721 | * |
| 722 | * While we handle mapping->wb_err with atomic operations, the f_wb_err |
| 723 | * value is protected by the f_lock since we must ensure that it reflects |
| 724 | * the latest value swapped in for this file descriptor. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 725 | * |
| 726 | * Return: %0 on success, negative error code otherwise. |
Jeff Layton | 5660e13 | 2017-07-06 07:02:25 -0400 | [diff] [blame] | 727 | */ |
| 728 | int file_check_and_advance_wb_err(struct file *file) |
| 729 | { |
| 730 | int err = 0; |
| 731 | errseq_t old = READ_ONCE(file->f_wb_err); |
| 732 | struct address_space *mapping = file->f_mapping; |
| 733 | |
| 734 | /* Locklessly handle the common case where nothing has changed */ |
| 735 | if (errseq_check(&mapping->wb_err, old)) { |
| 736 | /* Something changed, must use slow path */ |
| 737 | spin_lock(&file->f_lock); |
| 738 | old = file->f_wb_err; |
| 739 | err = errseq_check_and_advance(&mapping->wb_err, |
| 740 | &file->f_wb_err); |
| 741 | trace_file_check_and_advance_wb_err(file, old); |
| 742 | spin_unlock(&file->f_lock); |
| 743 | } |
Jeff Layton | f4e222c | 2017-10-03 16:15:25 -0700 | [diff] [blame] | 744 | |
| 745 | /* |
| 746 | * We're mostly using this function as a drop in replacement for |
| 747 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect |
| 748 | * that the legacy code would have had on these flags. |
| 749 | */ |
| 750 | clear_bit(AS_EIO, &mapping->flags); |
| 751 | clear_bit(AS_ENOSPC, &mapping->flags); |
Jeff Layton | 5660e13 | 2017-07-06 07:02:25 -0400 | [diff] [blame] | 752 | return err; |
| 753 | } |
| 754 | EXPORT_SYMBOL(file_check_and_advance_wb_err); |
| 755 | |
| 756 | /** |
| 757 | * file_write_and_wait_range - write out & wait on a file range |
| 758 | * @file: file pointing to address_space with pages |
| 759 | * @lstart: offset in bytes where the range starts |
| 760 | * @lend: offset in bytes where the range ends (inclusive) |
| 761 | * |
| 762 | * Write out and wait upon file offsets lstart->lend, inclusive. |
| 763 | * |
| 764 | * Note that @lend is inclusive (describes the last byte to be written) so |
| 765 | * that this function can be used to write to the very end-of-file (end = -1). |
| 766 | * |
| 767 | * After writing out and waiting on the data, we check and advance the |
| 768 | * f_wb_err cursor to the latest value, and return any errors detected there. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 769 | * |
| 770 | * Return: %0 on success, negative error code otherwise. |
Jeff Layton | 5660e13 | 2017-07-06 07:02:25 -0400 | [diff] [blame] | 771 | */ |
| 772 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) |
| 773 | { |
| 774 | int err = 0, err2; |
| 775 | struct address_space *mapping = file->f_mapping; |
| 776 | |
Brian Foster | feeb9b2 | 2022-11-28 10:56:31 -0500 | [diff] [blame] | 777 | if (lend < lstart) |
| 778 | return 0; |
| 779 | |
Jeff Layton | 9326c9b | 2017-07-26 10:21:11 -0400 | [diff] [blame] | 780 | if (mapping_needs_writeback(mapping)) { |
Jeff Layton | 5660e13 | 2017-07-06 07:02:25 -0400 | [diff] [blame] | 781 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
| 782 | WB_SYNC_ALL); |
| 783 | /* See comment of filemap_write_and_wait() */ |
| 784 | if (err != -EIO) |
| 785 | __filemap_fdatawait_range(mapping, lstart, lend); |
| 786 | } |
| 787 | err2 = file_check_and_advance_wb_err(file); |
| 788 | if (!err) |
| 789 | err = err2; |
| 790 | return err; |
| 791 | } |
| 792 | EXPORT_SYMBOL(file_write_and_wait_range); |
| 793 | |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 794 | /** |
Vishal Moola (Oracle) | 3720dd6 | 2022-11-01 10:53:22 -0700 | [diff] [blame] | 795 | * replace_page_cache_folio - replace a pagecache folio with a new one |
| 796 | * @old: folio to be replaced |
| 797 | * @new: folio to replace with |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 798 | * |
Vishal Moola (Oracle) | 3720dd6 | 2022-11-01 10:53:22 -0700 | [diff] [blame] | 799 | * This function replaces a folio in the pagecache with a new one. On |
| 800 | * success it acquires the pagecache reference for the new folio and |
| 801 | * drops it for the old folio. Both the old and new folios must be |
| 802 | * locked. This function does not add the new folio to the LRU, the |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 803 | * caller must do that. |
| 804 | * |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 805 | * The remove + add is atomic. This function cannot fail. |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 806 | */ |
Vishal Moola (Oracle) | 3720dd6 | 2022-11-01 10:53:22 -0700 | [diff] [blame] | 807 | void replace_page_cache_folio(struct folio *old, struct folio *new) |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 808 | { |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 809 | struct address_space *mapping = old->mapping; |
Matthew Wilcox (Oracle) | d2329aa | 2022-05-01 07:35:31 -0400 | [diff] [blame] | 810 | void (*free_folio)(struct folio *) = mapping->a_ops->free_folio; |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 811 | pgoff_t offset = old->index; |
| 812 | XA_STATE(xas, &mapping->i_pages, offset); |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 813 | |
Vishal Moola (Oracle) | 3720dd6 | 2022-11-01 10:53:22 -0700 | [diff] [blame] | 814 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old); |
| 815 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new); |
| 816 | VM_BUG_ON_FOLIO(new->mapping, new); |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 817 | |
Vishal Moola (Oracle) | 3720dd6 | 2022-11-01 10:53:22 -0700 | [diff] [blame] | 818 | folio_get(new); |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 819 | new->mapping = mapping; |
| 820 | new->index = offset; |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 821 | |
Vishal Moola (Oracle) | 3720dd6 | 2022-11-01 10:53:22 -0700 | [diff] [blame] | 822 | mem_cgroup_migrate(old, new); |
Johannes Weiner | 0d1c207 | 2020-06-03 16:01:54 -0700 | [diff] [blame] | 823 | |
Johannes Weiner | 3047250 | 2021-09-02 14:53:18 -0700 | [diff] [blame] | 824 | xas_lock_irq(&xas); |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 825 | xas_store(&xas, new); |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 826 | |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 827 | old->mapping = NULL; |
| 828 | /* hugetlb pages do not participate in page cache accounting. */ |
Vishal Moola (Oracle) | 3720dd6 | 2022-11-01 10:53:22 -0700 | [diff] [blame] | 829 | if (!folio_test_hugetlb(old)) |
| 830 | __lruvec_stat_sub_folio(old, NR_FILE_PAGES); |
| 831 | if (!folio_test_hugetlb(new)) |
| 832 | __lruvec_stat_add_folio(new, NR_FILE_PAGES); |
| 833 | if (folio_test_swapbacked(old)) |
| 834 | __lruvec_stat_sub_folio(old, NR_SHMEM); |
| 835 | if (folio_test_swapbacked(new)) |
| 836 | __lruvec_stat_add_folio(new, NR_SHMEM); |
Johannes Weiner | 3047250 | 2021-09-02 14:53:18 -0700 | [diff] [blame] | 837 | xas_unlock_irq(&xas); |
Matthew Wilcox (Oracle) | d2329aa | 2022-05-01 07:35:31 -0400 | [diff] [blame] | 838 | if (free_folio) |
Vishal Moola (Oracle) | 3720dd6 | 2022-11-01 10:53:22 -0700 | [diff] [blame] | 839 | free_folio(old); |
| 840 | folio_put(old); |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 841 | } |
Vishal Moola (Oracle) | 3720dd6 | 2022-11-01 10:53:22 -0700 | [diff] [blame] | 842 | EXPORT_SYMBOL_GPL(replace_page_cache_folio); |
Miklos Szeredi | ef6a3c6 | 2011-03-22 16:30:52 -0700 | [diff] [blame] | 843 | |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 844 | noinline int __filemap_add_folio(struct address_space *mapping, |
| 845 | struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | { |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 847 | XA_STATE(xas, &mapping->i_pages, index); |
| 848 | int huge = folio_test_hugetlb(folio); |
Waiman Long | da74240 | 2021-02-04 18:32:45 -0800 | [diff] [blame] | 849 | bool charged = false; |
Matthew Wilcox (Oracle) | d68ecca | 2019-09-05 14:03:12 -0400 | [diff] [blame] | 850 | long nr = 1; |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 851 | |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 852 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
| 853 | VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio); |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 854 | mapping_set_update(&xas, mapping); |
Nick Piggin | e286781 | 2008-07-25 19:45:30 -0700 | [diff] [blame] | 855 | |
Johannes Weiner | 3fea5a4 | 2020-06-03 16:01:41 -0700 | [diff] [blame] | 856 | if (!huge) { |
Matthew Wilcox (Oracle) | d68ecca | 2019-09-05 14:03:12 -0400 | [diff] [blame] | 857 | int error = mem_cgroup_charge(folio, NULL, gfp); |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 858 | VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio); |
Johannes Weiner | 3fea5a4 | 2020-06-03 16:01:41 -0700 | [diff] [blame] | 859 | if (error) |
Matthew Wilcox (Oracle) | d68ecca | 2019-09-05 14:03:12 -0400 | [diff] [blame] | 860 | return error; |
Waiman Long | da74240 | 2021-02-04 18:32:45 -0800 | [diff] [blame] | 861 | charged = true; |
Matthew Wilcox (Oracle) | d68ecca | 2019-09-05 14:03:12 -0400 | [diff] [blame] | 862 | xas_set_order(&xas, index, folio_order(folio)); |
| 863 | nr = folio_nr_pages(folio); |
Johannes Weiner | 3fea5a4 | 2020-06-03 16:01:41 -0700 | [diff] [blame] | 864 | } |
| 865 | |
Matthew Wilcox (Oracle) | 198b62f | 2020-10-15 20:05:20 -0700 | [diff] [blame] | 866 | gfp &= GFP_RECLAIM_MASK; |
Matthew Wilcox (Oracle) | d68ecca | 2019-09-05 14:03:12 -0400 | [diff] [blame] | 867 | folio_ref_add(folio, nr); |
| 868 | folio->mapping = mapping; |
| 869 | folio->index = xas.xa_index; |
Matthew Wilcox (Oracle) | 198b62f | 2020-10-15 20:05:20 -0700 | [diff] [blame] | 870 | |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 871 | do { |
Matthew Wilcox (Oracle) | 198b62f | 2020-10-15 20:05:20 -0700 | [diff] [blame] | 872 | unsigned int order = xa_get_order(xas.xa, xas.xa_index); |
| 873 | void *entry, *old = NULL; |
| 874 | |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 875 | if (order > folio_order(folio)) |
Matthew Wilcox (Oracle) | 198b62f | 2020-10-15 20:05:20 -0700 | [diff] [blame] | 876 | xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), |
| 877 | order, gfp); |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 878 | xas_lock_irq(&xas); |
Matthew Wilcox (Oracle) | 198b62f | 2020-10-15 20:05:20 -0700 | [diff] [blame] | 879 | xas_for_each_conflict(&xas, entry) { |
| 880 | old = entry; |
| 881 | if (!xa_is_value(entry)) { |
| 882 | xas_set_err(&xas, -EEXIST); |
| 883 | goto unlock; |
| 884 | } |
| 885 | } |
| 886 | |
| 887 | if (old) { |
| 888 | if (shadowp) |
| 889 | *shadowp = old; |
| 890 | /* entry may have been split before we acquired lock */ |
| 891 | order = xa_get_order(xas.xa, xas.xa_index); |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 892 | if (order > folio_order(folio)) { |
Matthew Wilcox (Oracle) | d68ecca | 2019-09-05 14:03:12 -0400 | [diff] [blame] | 893 | /* How to handle large swap entries? */ |
| 894 | BUG_ON(shmem_mapping(mapping)); |
Matthew Wilcox (Oracle) | 198b62f | 2020-10-15 20:05:20 -0700 | [diff] [blame] | 895 | xas_split(&xas, old, order); |
| 896 | xas_reset(&xas); |
| 897 | } |
| 898 | } |
| 899 | |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 900 | xas_store(&xas, folio); |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 901 | if (xas_error(&xas)) |
| 902 | goto unlock; |
Michal Hocko | 4165b9b4 | 2015-06-24 16:57:24 -0700 | [diff] [blame] | 903 | |
Matthew Wilcox (Oracle) | d68ecca | 2019-09-05 14:03:12 -0400 | [diff] [blame] | 904 | mapping->nrpages += nr; |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 905 | |
| 906 | /* hugetlb pages do not participate in page cache accounting */ |
Matthew Wilcox (Oracle) | d68ecca | 2019-09-05 14:03:12 -0400 | [diff] [blame] | 907 | if (!huge) { |
| 908 | __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr); |
| 909 | if (folio_test_pmd_mappable(folio)) |
| 910 | __lruvec_stat_mod_folio(folio, |
| 911 | NR_FILE_THPS, nr); |
| 912 | } |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 913 | unlock: |
| 914 | xas_unlock_irq(&xas); |
Matthew Wilcox (Oracle) | 198b62f | 2020-10-15 20:05:20 -0700 | [diff] [blame] | 915 | } while (xas_nomem(&xas, gfp)); |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 916 | |
Matthew Wilcox (Oracle) | d68ecca | 2019-09-05 14:03:12 -0400 | [diff] [blame] | 917 | if (xas_error(&xas)) |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 918 | goto error; |
| 919 | |
Matthew Wilcox (Oracle) | a0580c6 | 2021-07-23 09:29:46 -0400 | [diff] [blame] | 920 | trace_mm_filemap_add_to_page_cache(folio); |
Kirill A. Shutemov | 66a0c8e | 2013-09-12 15:13:59 -0700 | [diff] [blame] | 921 | return 0; |
Matthew Wilcox | 74d6095 | 2017-11-17 10:01:45 -0500 | [diff] [blame] | 922 | error: |
Matthew Wilcox (Oracle) | d68ecca | 2019-09-05 14:03:12 -0400 | [diff] [blame] | 923 | if (charged) |
| 924 | mem_cgroup_uncharge(folio); |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 925 | folio->mapping = NULL; |
Kirill A. Shutemov | 66a0c8e | 2013-09-12 15:13:59 -0700 | [diff] [blame] | 926 | /* Leave page->index set: truncation relies upon it */ |
Matthew Wilcox (Oracle) | d68ecca | 2019-09-05 14:03:12 -0400 | [diff] [blame] | 927 | folio_put_refs(folio, nr); |
| 928 | return xas_error(&xas); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 929 | } |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 930 | ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO); |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 931 | |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 932 | int filemap_add_folio(struct address_space *mapping, struct folio *folio, |
| 933 | pgoff_t index, gfp_t gfp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 934 | { |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 935 | void *shadow = NULL; |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 936 | int ret; |
| 937 | |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 938 | __folio_set_locked(folio); |
| 939 | ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow); |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 940 | if (unlikely(ret)) |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 941 | __folio_clear_locked(folio); |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 942 | else { |
| 943 | /* |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 944 | * The folio might have been evicted from cache only |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 945 | * recently, in which case it should be activated like |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 946 | * any other repeatedly accessed folio. |
| 947 | * The exception is folios getting rewritten; evicting other |
Rik van Riel | f0281a0 | 2016-05-20 16:56:25 -0700 | [diff] [blame] | 948 | * data from the working set, only to cache data that will |
| 949 | * get overwritten with something else, is a waste of memory. |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 950 | */ |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 951 | WARN_ON_ONCE(folio_test_active(folio)); |
| 952 | if (!(gfp & __GFP_WRITE) && shadow) |
| 953 | workingset_refault(folio, shadow); |
| 954 | folio_add_lru(folio); |
Johannes Weiner | a528910 | 2014-04-03 14:47:51 -0700 | [diff] [blame] | 955 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 956 | return ret; |
| 957 | } |
Matthew Wilcox (Oracle) | 9dd3d06 | 2020-12-08 08:56:28 -0500 | [diff] [blame] | 958 | EXPORT_SYMBOL_GPL(filemap_add_folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 959 | |
Paul Jackson | 44110fe | 2006-03-24 03:16:04 -0800 | [diff] [blame] | 960 | #ifdef CONFIG_NUMA |
Matthew Wilcox (Oracle) | bb3c579 | 2020-12-15 23:11:07 -0500 | [diff] [blame] | 961 | struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order) |
Paul Jackson | 44110fe | 2006-03-24 03:16:04 -0800 | [diff] [blame] | 962 | { |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 963 | int n; |
Matthew Wilcox (Oracle) | bb3c579 | 2020-12-15 23:11:07 -0500 | [diff] [blame] | 964 | struct folio *folio; |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 965 | |
Paul Jackson | 44110fe | 2006-03-24 03:16:04 -0800 | [diff] [blame] | 966 | if (cpuset_do_page_mem_spread()) { |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 967 | unsigned int cpuset_mems_cookie; |
| 968 | do { |
Mel Gorman | d26914d | 2014-04-03 14:47:24 -0700 | [diff] [blame] | 969 | cpuset_mems_cookie = read_mems_allowed_begin(); |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 970 | n = cpuset_mem_spread_node(); |
Matthew Wilcox (Oracle) | bb3c579 | 2020-12-15 23:11:07 -0500 | [diff] [blame] | 971 | folio = __folio_alloc_node(gfp, order, n); |
| 972 | } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie)); |
Mel Gorman | cc9a6c8 | 2012-03-21 16:34:11 -0700 | [diff] [blame] | 973 | |
Matthew Wilcox (Oracle) | bb3c579 | 2020-12-15 23:11:07 -0500 | [diff] [blame] | 974 | return folio; |
Paul Jackson | 44110fe | 2006-03-24 03:16:04 -0800 | [diff] [blame] | 975 | } |
Matthew Wilcox (Oracle) | bb3c579 | 2020-12-15 23:11:07 -0500 | [diff] [blame] | 976 | return folio_alloc(gfp, order); |
Paul Jackson | 44110fe | 2006-03-24 03:16:04 -0800 | [diff] [blame] | 977 | } |
Matthew Wilcox (Oracle) | bb3c579 | 2020-12-15 23:11:07 -0500 | [diff] [blame] | 978 | EXPORT_SYMBOL(filemap_alloc_folio); |
Paul Jackson | 44110fe | 2006-03-24 03:16:04 -0800 | [diff] [blame] | 979 | #endif |
| 980 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 981 | /* |
Jan Kara | 7506ae6 | 2021-05-24 13:02:30 +0200 | [diff] [blame] | 982 | * filemap_invalidate_lock_two - lock invalidate_lock for two mappings |
| 983 | * |
| 984 | * Lock exclusively invalidate_lock of any passed mapping that is not NULL. |
| 985 | * |
| 986 | * @mapping1: the first mapping to lock |
| 987 | * @mapping2: the second mapping to lock |
| 988 | */ |
| 989 | void filemap_invalidate_lock_two(struct address_space *mapping1, |
| 990 | struct address_space *mapping2) |
| 991 | { |
| 992 | if (mapping1 > mapping2) |
| 993 | swap(mapping1, mapping2); |
| 994 | if (mapping1) |
| 995 | down_write(&mapping1->invalidate_lock); |
| 996 | if (mapping2 && mapping1 != mapping2) |
| 997 | down_write_nested(&mapping2->invalidate_lock, 1); |
| 998 | } |
| 999 | EXPORT_SYMBOL(filemap_invalidate_lock_two); |
| 1000 | |
| 1001 | /* |
| 1002 | * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings |
| 1003 | * |
| 1004 | * Unlock exclusive invalidate_lock of any passed mapping that is not NULL. |
| 1005 | * |
| 1006 | * @mapping1: the first mapping to unlock |
| 1007 | * @mapping2: the second mapping to unlock |
| 1008 | */ |
| 1009 | void filemap_invalidate_unlock_two(struct address_space *mapping1, |
| 1010 | struct address_space *mapping2) |
| 1011 | { |
| 1012 | if (mapping1) |
| 1013 | up_write(&mapping1->invalidate_lock); |
| 1014 | if (mapping2 && mapping1 != mapping2) |
| 1015 | up_write(&mapping2->invalidate_lock); |
| 1016 | } |
| 1017 | EXPORT_SYMBOL(filemap_invalidate_unlock_two); |
| 1018 | |
| 1019 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1020 | * In order to wait for pages to become available there must be |
| 1021 | * waitqueues associated with pages. By using a hash table of |
| 1022 | * waitqueues where the bucket discipline is to maintain all |
| 1023 | * waiters on the same queue and wake all when any of the pages |
| 1024 | * become available, and for the woken contexts to check to be |
| 1025 | * sure the appropriate page became available, this saves space |
| 1026 | * at a cost of "thundering herd" phenomena during rare hash |
| 1027 | * collisions. |
| 1028 | */ |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1029 | #define PAGE_WAIT_TABLE_BITS 8 |
| 1030 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1031 | static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1032 | |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1033 | static wait_queue_head_t *folio_waitqueue(struct folio *folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | { |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1035 | return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1036 | } |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1037 | |
| 1038 | void __init pagecache_init(void) |
| 1039 | { |
| 1040 | int i; |
| 1041 | |
| 1042 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1043 | init_waitqueue_head(&folio_wait_table[i]); |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1044 | |
| 1045 | page_writeback_init(); |
| 1046 | } |
| 1047 | |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1048 | /* |
| 1049 | * The page wait code treats the "wait->flags" somewhat unusually, because |
Linus Torvalds | 5868ec2 | 2020-09-20 10:38:47 -0700 | [diff] [blame] | 1050 | * we have multiple different kinds of waits, not just the usual "exclusive" |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1051 | * one. |
| 1052 | * |
| 1053 | * We have: |
| 1054 | * |
| 1055 | * (a) no special bits set: |
| 1056 | * |
| 1057 | * We're just waiting for the bit to be released, and when a waker |
| 1058 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, |
| 1059 | * and remove it from the wait queue. |
| 1060 | * |
| 1061 | * Simple and straightforward. |
| 1062 | * |
| 1063 | * (b) WQ_FLAG_EXCLUSIVE: |
| 1064 | * |
| 1065 | * The waiter is waiting to get the lock, and only one waiter should |
| 1066 | * be woken up to avoid any thundering herd behavior. We'll set the |
| 1067 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. |
| 1068 | * |
| 1069 | * This is the traditional exclusive wait. |
| 1070 | * |
Linus Torvalds | 5868ec2 | 2020-09-20 10:38:47 -0700 | [diff] [blame] | 1071 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1072 | * |
| 1073 | * The waiter is waiting to get the bit, and additionally wants the |
| 1074 | * lock to be transferred to it for fair lock behavior. If the lock |
| 1075 | * cannot be taken, we stop walking the wait queue without waking |
| 1076 | * the waiter. |
| 1077 | * |
| 1078 | * This is the "fair lock handoff" case, and in addition to setting |
| 1079 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see |
| 1080 | * that it now has the lock. |
| 1081 | */ |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 1082 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1083 | { |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1084 | unsigned int flags; |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1085 | struct wait_page_key *key = arg; |
| 1086 | struct wait_page_queue *wait_page |
| 1087 | = container_of(wait, struct wait_page_queue, wait); |
| 1088 | |
Linus Torvalds | cdc8fcb | 2020-08-03 13:01:22 -0700 | [diff] [blame] | 1089 | if (!wake_page_match(wait_page, key)) |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1090 | return 0; |
Linus Torvalds | 3510ca2 | 2017-08-27 13:55:12 -0700 | [diff] [blame] | 1091 | |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1092 | /* |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1093 | * If it's a lock handoff wait, we get the bit for it, and |
| 1094 | * stop walking (and do not wake it up) if we can't. |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1095 | */ |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1096 | flags = wait->flags; |
| 1097 | if (flags & WQ_FLAG_EXCLUSIVE) { |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1098 | if (test_bit(key->bit_nr, &key->folio->flags)) |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1099 | return -1; |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1100 | if (flags & WQ_FLAG_CUSTOM) { |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1101 | if (test_and_set_bit(key->bit_nr, &key->folio->flags)) |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1102 | return -1; |
| 1103 | flags |= WQ_FLAG_DONE; |
| 1104 | } |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1105 | } |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1106 | |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1107 | /* |
| 1108 | * We are holding the wait-queue lock, but the waiter that |
| 1109 | * is waiting for this will be checking the flags without |
| 1110 | * any locking. |
| 1111 | * |
| 1112 | * So update the flags atomically, and wake up the waiter |
| 1113 | * afterwards to avoid any races. This store-release pairs |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1114 | * with the load-acquire in folio_wait_bit_common(). |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1115 | */ |
| 1116 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1117 | wake_up_state(wait->private, mode); |
| 1118 | |
| 1119 | /* |
| 1120 | * Ok, we have successfully done what we're waiting for, |
| 1121 | * and we can unconditionally remove the wait entry. |
| 1122 | * |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1123 | * Note that this pairs with the "finish_wait()" in the |
| 1124 | * waiter, and has to be the absolute last thing we do. |
| 1125 | * After this list_del_init(&wait->entry) the wait entry |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1126 | * might be de-allocated and the process might even have |
| 1127 | * exited. |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1128 | */ |
Linus Torvalds | c6fe44d | 2020-07-23 12:33:41 -0700 | [diff] [blame] | 1129 | list_del_init_careful(&wait->entry); |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1130 | return (flags & WQ_FLAG_EXCLUSIVE) != 0; |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1131 | } |
| 1132 | |
Matthew Wilcox (Oracle) | 6974d7c | 2021-01-15 17:14:48 -0500 | [diff] [blame] | 1133 | static void folio_wake_bit(struct folio *folio, int bit_nr) |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1134 | { |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1135 | wait_queue_head_t *q = folio_waitqueue(folio); |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1136 | struct wait_page_key key; |
| 1137 | unsigned long flags; |
Tim Chen | 11a19c7 | 2017-08-25 09:13:55 -0700 | [diff] [blame] | 1138 | wait_queue_entry_t bookmark; |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1139 | |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1140 | key.folio = folio; |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1141 | key.bit_nr = bit_nr; |
| 1142 | key.page_match = 0; |
| 1143 | |
Tim Chen | 11a19c7 | 2017-08-25 09:13:55 -0700 | [diff] [blame] | 1144 | bookmark.flags = 0; |
| 1145 | bookmark.private = NULL; |
| 1146 | bookmark.func = NULL; |
| 1147 | INIT_LIST_HEAD(&bookmark.entry); |
| 1148 | |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1149 | spin_lock_irqsave(&q->lock, flags); |
Tim Chen | 11a19c7 | 2017-08-25 09:13:55 -0700 | [diff] [blame] | 1150 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
| 1151 | |
| 1152 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { |
| 1153 | /* |
| 1154 | * Take a breather from holding the lock, |
| 1155 | * allow pages that finish wake up asynchronously |
| 1156 | * to acquire the lock and remove themselves |
| 1157 | * from wait queue |
| 1158 | */ |
| 1159 | spin_unlock_irqrestore(&q->lock, flags); |
| 1160 | cpu_relax(); |
| 1161 | spin_lock_irqsave(&q->lock, flags); |
| 1162 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
| 1163 | } |
| 1164 | |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1165 | /* |
Hugh Dickins | bb43b14 | 2022-03-24 18:09:49 -0700 | [diff] [blame] | 1166 | * It's possible to miss clearing waiters here, when we woke our page |
| 1167 | * waiters, but the hashed waitqueue has waiters for other pages on it. |
| 1168 | * That's okay, it's a rare case. The next waker will clear it. |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1169 | * |
Hugh Dickins | bb43b14 | 2022-03-24 18:09:49 -0700 | [diff] [blame] | 1170 | * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE, |
| 1171 | * other), the flag may be cleared in the course of freeing the page; |
| 1172 | * but that is not required for correctness. |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1173 | */ |
Hugh Dickins | bb43b14 | 2022-03-24 18:09:49 -0700 | [diff] [blame] | 1174 | if (!waitqueue_active(q) || !key.page_match) |
Matthew Wilcox (Oracle) | 6974d7c | 2021-01-15 17:14:48 -0500 | [diff] [blame] | 1175 | folio_clear_waiters(folio); |
Hugh Dickins | bb43b14 | 2022-03-24 18:09:49 -0700 | [diff] [blame] | 1176 | |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1177 | spin_unlock_irqrestore(&q->lock, flags); |
| 1178 | } |
Nicholas Piggin | 74d81bf | 2017-02-22 15:44:41 -0800 | [diff] [blame] | 1179 | |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1180 | static void folio_wake(struct folio *folio, int bit) |
Nicholas Piggin | 74d81bf | 2017-02-22 15:44:41 -0800 | [diff] [blame] | 1181 | { |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1182 | if (!folio_test_waiters(folio)) |
Nicholas Piggin | 74d81bf | 2017-02-22 15:44:41 -0800 | [diff] [blame] | 1183 | return; |
Matthew Wilcox (Oracle) | 6974d7c | 2021-01-15 17:14:48 -0500 | [diff] [blame] | 1184 | folio_wake_bit(folio, bit); |
Nicholas Piggin | 74d81bf | 2017-02-22 15:44:41 -0800 | [diff] [blame] | 1185 | } |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1186 | |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1187 | /* |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1188 | * A choice of three behaviors for folio_wait_bit_common(): |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1189 | */ |
| 1190 | enum behavior { |
| 1191 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like |
Matthew Wilcox (Oracle) | 7c23c78 | 2021-03-01 19:38:25 -0500 | [diff] [blame] | 1192 | * __folio_lock() waiting on then setting PG_locked. |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1193 | */ |
| 1194 | SHARED, /* Hold ref to page and check the bit when woken, like |
Matthew Wilcox (Oracle) | 9f2b04a | 2021-08-16 23:36:31 -0400 | [diff] [blame] | 1195 | * folio_wait_writeback() waiting on PG_writeback. |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1196 | */ |
| 1197 | DROP, /* Drop ref to page before wait, no check when woken, |
Matthew Wilcox (Oracle) | 9f2b04a | 2021-08-16 23:36:31 -0400 | [diff] [blame] | 1198 | * like folio_put_wait_locked() on PG_locked. |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1199 | */ |
| 1200 | }; |
| 1201 | |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1202 | /* |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1203 | * Attempt to check (or get) the folio flag, and mark us done |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1204 | * if successful. |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1205 | */ |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1206 | static inline bool folio_trylock_flag(struct folio *folio, int bit_nr, |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1207 | struct wait_queue_entry *wait) |
| 1208 | { |
| 1209 | if (wait->flags & WQ_FLAG_EXCLUSIVE) { |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1210 | if (test_and_set_bit(bit_nr, &folio->flags)) |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1211 | return false; |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1212 | } else if (test_bit(bit_nr, &folio->flags)) |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1213 | return false; |
| 1214 | |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1215 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1216 | return true; |
| 1217 | } |
| 1218 | |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1219 | /* How many times do we accept lock stealing from under a waiter? */ |
| 1220 | int sysctl_page_lock_unfairness = 5; |
| 1221 | |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1222 | static inline int folio_wait_bit_common(struct folio *folio, int bit_nr, |
| 1223 | int state, enum behavior behavior) |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1224 | { |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1225 | wait_queue_head_t *q = folio_waitqueue(folio); |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1226 | int unfairness = sysctl_page_lock_unfairness; |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1227 | struct wait_page_queue wait_page; |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 1228 | wait_queue_entry_t *wait = &wait_page.wait; |
Johannes Weiner | b1d29ba | 2018-10-26 15:06:08 -0700 | [diff] [blame] | 1229 | bool thrashing = false; |
Johannes Weiner | eb41468 | 2018-10-26 15:06:27 -0700 | [diff] [blame] | 1230 | unsigned long pflags; |
Yang Yang | aa1cf99 | 2022-08-15 07:11:35 +0000 | [diff] [blame] | 1231 | bool in_thrashing; |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1232 | |
Johannes Weiner | eb41468 | 2018-10-26 15:06:27 -0700 | [diff] [blame] | 1233 | if (bit_nr == PG_locked && |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1234 | !folio_test_uptodate(folio) && folio_test_workingset(folio)) { |
Yang Yang | aa1cf99 | 2022-08-15 07:11:35 +0000 | [diff] [blame] | 1235 | delayacct_thrashing_start(&in_thrashing); |
Johannes Weiner | eb41468 | 2018-10-26 15:06:27 -0700 | [diff] [blame] | 1236 | psi_memstall_enter(&pflags); |
Johannes Weiner | b1d29ba | 2018-10-26 15:06:08 -0700 | [diff] [blame] | 1237 | thrashing = true; |
| 1238 | } |
| 1239 | |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1240 | init_wait(wait); |
| 1241 | wait->func = wake_page_function; |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1242 | wait_page.folio = folio; |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1243 | wait_page.bit_nr = bit_nr; |
| 1244 | |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1245 | repeat: |
| 1246 | wait->flags = 0; |
| 1247 | if (behavior == EXCLUSIVE) { |
| 1248 | wait->flags = WQ_FLAG_EXCLUSIVE; |
| 1249 | if (--unfairness < 0) |
| 1250 | wait->flags |= WQ_FLAG_CUSTOM; |
| 1251 | } |
| 1252 | |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1253 | /* |
| 1254 | * Do one last check whether we can get the |
| 1255 | * page bit synchronously. |
| 1256 | * |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1257 | * Do the folio_set_waiters() marking before that |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1258 | * to let any waker we _just_ missed know they |
| 1259 | * need to wake us up (otherwise they'll never |
| 1260 | * even go to the slow case that looks at the |
| 1261 | * page queue), and add ourselves to the wait |
| 1262 | * queue if we need to sleep. |
| 1263 | * |
| 1264 | * This part needs to be done under the queue |
| 1265 | * lock to avoid races. |
| 1266 | */ |
| 1267 | spin_lock_irq(&q->lock); |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1268 | folio_set_waiters(folio); |
| 1269 | if (!folio_trylock_flag(folio, bit_nr, wait)) |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1270 | __add_wait_queue_entry_tail(q, wait); |
| 1271 | spin_unlock_irq(&q->lock); |
| 1272 | |
| 1273 | /* |
| 1274 | * From now on, all the logic will be based on |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1275 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to |
| 1276 | * see whether the page bit testing has already |
| 1277 | * been done by the wake function. |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1278 | * |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1279 | * We can drop our reference to the folio. |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1280 | */ |
| 1281 | if (behavior == DROP) |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1282 | folio_put(folio); |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1283 | |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1284 | /* |
| 1285 | * Note that until the "finish_wait()", or until |
| 1286 | * we see the WQ_FLAG_WOKEN flag, we need to |
| 1287 | * be very careful with the 'wait->flags', because |
| 1288 | * we may race with a waker that sets them. |
| 1289 | */ |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1290 | for (;;) { |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1291 | unsigned int flags; |
| 1292 | |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1293 | set_current_state(state); |
| 1294 | |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1295 | /* Loop until we've been woken or interrupted */ |
| 1296 | flags = smp_load_acquire(&wait->flags); |
| 1297 | if (!(flags & WQ_FLAG_WOKEN)) { |
| 1298 | if (signal_pending_state(state, current)) |
| 1299 | break; |
| 1300 | |
| 1301 | io_schedule(); |
| 1302 | continue; |
| 1303 | } |
| 1304 | |
| 1305 | /* If we were non-exclusive, we're done */ |
| 1306 | if (behavior != EXCLUSIVE) |
Linus Torvalds | a8b169a | 2017-08-27 16:25:09 -0700 | [diff] [blame] | 1307 | break; |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1308 | |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1309 | /* If the waker got the lock for us, we're done */ |
| 1310 | if (flags & WQ_FLAG_DONE) |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1311 | break; |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1312 | |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1313 | /* |
| 1314 | * Otherwise, if we're getting the lock, we need to |
| 1315 | * try to get it ourselves. |
| 1316 | * |
| 1317 | * And if that fails, we'll have to retry this all. |
| 1318 | */ |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1319 | if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0)))) |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1320 | goto repeat; |
| 1321 | |
| 1322 | wait->flags |= WQ_FLAG_DONE; |
| 1323 | break; |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1324 | } |
| 1325 | |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1326 | /* |
| 1327 | * If a signal happened, this 'finish_wait()' may remove the last |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1328 | * waiter from the wait-queues, but the folio waiters bit will remain |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1329 | * set. That's ok. The next wakeup will take care of it, and trying |
| 1330 | * to do it here would be difficult and prone to races. |
| 1331 | */ |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1332 | finish_wait(q, wait); |
| 1333 | |
Johannes Weiner | eb41468 | 2018-10-26 15:06:27 -0700 | [diff] [blame] | 1334 | if (thrashing) { |
Yang Yang | aa1cf99 | 2022-08-15 07:11:35 +0000 | [diff] [blame] | 1335 | delayacct_thrashing_end(&in_thrashing); |
Johannes Weiner | eb41468 | 2018-10-26 15:06:27 -0700 | [diff] [blame] | 1336 | psi_memstall_leave(&pflags); |
| 1337 | } |
Johannes Weiner | b1d29ba | 2018-10-26 15:06:08 -0700 | [diff] [blame] | 1338 | |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1339 | /* |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1340 | * NOTE! The wait->flags weren't stable until we've done the |
| 1341 | * 'finish_wait()', and we could have exited the loop above due |
| 1342 | * to a signal, and had a wakeup event happen after the signal |
| 1343 | * test but before the 'finish_wait()'. |
| 1344 | * |
| 1345 | * So only after the finish_wait() can we reliably determine |
| 1346 | * if we got woken up or not, so we can now figure out the final |
| 1347 | * return value based on that state without races. |
| 1348 | * |
| 1349 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive |
| 1350 | * waiter, but an exclusive one requires WQ_FLAG_DONE. |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1351 | */ |
Linus Torvalds | 5ef64cc | 2020-09-13 14:05:35 -0700 | [diff] [blame] | 1352 | if (behavior == EXCLUSIVE) |
| 1353 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1354 | |
Linus Torvalds | 2a9127f | 2020-07-23 10:16:49 -0700 | [diff] [blame] | 1355 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; |
Nicholas Piggin | 6290602 | 2016-12-25 13:00:30 +1000 | [diff] [blame] | 1356 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | |
Alistair Popple | ffa6575 | 2022-01-21 22:10:46 -0800 | [diff] [blame] | 1358 | #ifdef CONFIG_MIGRATION |
| 1359 | /** |
| 1360 | * migration_entry_wait_on_locked - Wait for a migration entry to be removed |
| 1361 | * @entry: migration swap entry. |
| 1362 | * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required |
| 1363 | * for pte entries, pass NULL for pmd entries. |
| 1364 | * @ptl: already locked ptl. This function will drop the lock. |
| 1365 | * |
| 1366 | * Wait for a migration entry referencing the given page to be removed. This is |
| 1367 | * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except |
| 1368 | * this can be called without taking a reference on the page. Instead this |
| 1369 | * should be called while holding the ptl for the migration entry referencing |
| 1370 | * the page. |
| 1371 | * |
| 1372 | * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock(). |
| 1373 | * |
| 1374 | * This follows the same logic as folio_wait_bit_common() so see the comments |
| 1375 | * there. |
| 1376 | */ |
| 1377 | void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep, |
| 1378 | spinlock_t *ptl) |
| 1379 | { |
| 1380 | struct wait_page_queue wait_page; |
| 1381 | wait_queue_entry_t *wait = &wait_page.wait; |
| 1382 | bool thrashing = false; |
Alistair Popple | ffa6575 | 2022-01-21 22:10:46 -0800 | [diff] [blame] | 1383 | unsigned long pflags; |
Yang Yang | aa1cf99 | 2022-08-15 07:11:35 +0000 | [diff] [blame] | 1384 | bool in_thrashing; |
Alistair Popple | ffa6575 | 2022-01-21 22:10:46 -0800 | [diff] [blame] | 1385 | wait_queue_head_t *q; |
| 1386 | struct folio *folio = page_folio(pfn_swap_entry_to_page(entry)); |
| 1387 | |
| 1388 | q = folio_waitqueue(folio); |
| 1389 | if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) { |
Yang Yang | aa1cf99 | 2022-08-15 07:11:35 +0000 | [diff] [blame] | 1390 | delayacct_thrashing_start(&in_thrashing); |
Alistair Popple | ffa6575 | 2022-01-21 22:10:46 -0800 | [diff] [blame] | 1391 | psi_memstall_enter(&pflags); |
| 1392 | thrashing = true; |
| 1393 | } |
| 1394 | |
| 1395 | init_wait(wait); |
| 1396 | wait->func = wake_page_function; |
| 1397 | wait_page.folio = folio; |
| 1398 | wait_page.bit_nr = PG_locked; |
| 1399 | wait->flags = 0; |
| 1400 | |
| 1401 | spin_lock_irq(&q->lock); |
| 1402 | folio_set_waiters(folio); |
| 1403 | if (!folio_trylock_flag(folio, PG_locked, wait)) |
| 1404 | __add_wait_queue_entry_tail(q, wait); |
| 1405 | spin_unlock_irq(&q->lock); |
| 1406 | |
| 1407 | /* |
| 1408 | * If a migration entry exists for the page the migration path must hold |
| 1409 | * a valid reference to the page, and it must take the ptl to remove the |
| 1410 | * migration entry. So the page is valid until the ptl is dropped. |
| 1411 | */ |
| 1412 | if (ptep) |
| 1413 | pte_unmap_unlock(ptep, ptl); |
| 1414 | else |
| 1415 | spin_unlock(ptl); |
| 1416 | |
| 1417 | for (;;) { |
| 1418 | unsigned int flags; |
| 1419 | |
| 1420 | set_current_state(TASK_UNINTERRUPTIBLE); |
| 1421 | |
| 1422 | /* Loop until we've been woken or interrupted */ |
| 1423 | flags = smp_load_acquire(&wait->flags); |
| 1424 | if (!(flags & WQ_FLAG_WOKEN)) { |
| 1425 | if (signal_pending_state(TASK_UNINTERRUPTIBLE, current)) |
| 1426 | break; |
| 1427 | |
| 1428 | io_schedule(); |
| 1429 | continue; |
| 1430 | } |
| 1431 | break; |
| 1432 | } |
| 1433 | |
| 1434 | finish_wait(q, wait); |
| 1435 | |
| 1436 | if (thrashing) { |
Yang Yang | aa1cf99 | 2022-08-15 07:11:35 +0000 | [diff] [blame] | 1437 | delayacct_thrashing_end(&in_thrashing); |
Alistair Popple | ffa6575 | 2022-01-21 22:10:46 -0800 | [diff] [blame] | 1438 | psi_memstall_leave(&pflags); |
| 1439 | } |
| 1440 | } |
| 1441 | #endif |
| 1442 | |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1443 | void folio_wait_bit(struct folio *folio, int bit_nr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1444 | { |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1445 | folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1446 | } |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1447 | EXPORT_SYMBOL(folio_wait_bit); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1448 | |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1449 | int folio_wait_bit_killable(struct folio *folio, int bit_nr) |
KOSAKI Motohiro | f62e00c | 2011-05-24 17:11:29 -0700 | [diff] [blame] | 1450 | { |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1451 | return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED); |
KOSAKI Motohiro | f62e00c | 2011-05-24 17:11:29 -0700 | [diff] [blame] | 1452 | } |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1453 | EXPORT_SYMBOL(folio_wait_bit_killable); |
KOSAKI Motohiro | f62e00c | 2011-05-24 17:11:29 -0700 | [diff] [blame] | 1454 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1455 | /** |
Matthew Wilcox (Oracle) | 9f2b04a | 2021-08-16 23:36:31 -0400 | [diff] [blame] | 1456 | * folio_put_wait_locked - Drop a reference and wait for it to be unlocked |
| 1457 | * @folio: The folio to wait for. |
Matthew Wilcox (Oracle) | 4805462 | 2021-02-24 12:02:02 -0800 | [diff] [blame] | 1458 | * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc). |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1459 | * |
Matthew Wilcox (Oracle) | 9f2b04a | 2021-08-16 23:36:31 -0400 | [diff] [blame] | 1460 | * The caller should hold a reference on @folio. They expect the page to |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1461 | * become unlocked relatively soon, but do not wish to hold up migration |
Matthew Wilcox (Oracle) | 9f2b04a | 2021-08-16 23:36:31 -0400 | [diff] [blame] | 1462 | * (for example) by holding the reference while waiting for the folio to |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1463 | * come unlocked. After this function returns, the caller should not |
Matthew Wilcox (Oracle) | 9f2b04a | 2021-08-16 23:36:31 -0400 | [diff] [blame] | 1464 | * dereference @folio. |
Matthew Wilcox (Oracle) | 4805462 | 2021-02-24 12:02:02 -0800 | [diff] [blame] | 1465 | * |
Matthew Wilcox (Oracle) | 9f2b04a | 2021-08-16 23:36:31 -0400 | [diff] [blame] | 1466 | * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal. |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1467 | */ |
Ke Sun | c195c32 | 2022-09-14 10:17:38 +0800 | [diff] [blame] | 1468 | static int folio_put_wait_locked(struct folio *folio, int state) |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1469 | { |
Matthew Wilcox (Oracle) | 9f2b04a | 2021-08-16 23:36:31 -0400 | [diff] [blame] | 1470 | return folio_wait_bit_common(folio, PG_locked, state, DROP); |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1471 | } |
| 1472 | |
| 1473 | /** |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1474 | * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue |
| 1475 | * @folio: Folio defining the wait queue of interest |
Randy Dunlap | 697f619 | 2009-04-13 14:39:54 -0700 | [diff] [blame] | 1476 | * @waiter: Waiter to add to the queue |
David Howells | 385e1ca5f | 2009-04-03 16:42:39 +0100 | [diff] [blame] | 1477 | * |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1478 | * Add an arbitrary @waiter to the wait queue for the nominated @folio. |
David Howells | 385e1ca5f | 2009-04-03 16:42:39 +0100 | [diff] [blame] | 1479 | */ |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1480 | void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter) |
David Howells | 385e1ca5f | 2009-04-03 16:42:39 +0100 | [diff] [blame] | 1481 | { |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1482 | wait_queue_head_t *q = folio_waitqueue(folio); |
David Howells | 385e1ca5f | 2009-04-03 16:42:39 +0100 | [diff] [blame] | 1483 | unsigned long flags; |
| 1484 | |
| 1485 | spin_lock_irqsave(&q->lock, flags); |
Linus Torvalds | 9c3a815 | 2017-08-28 16:45:40 -0700 | [diff] [blame] | 1486 | __add_wait_queue_entry_tail(q, waiter); |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1487 | folio_set_waiters(folio); |
David Howells | 385e1ca5f | 2009-04-03 16:42:39 +0100 | [diff] [blame] | 1488 | spin_unlock_irqrestore(&q->lock, flags); |
| 1489 | } |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1490 | EXPORT_SYMBOL_GPL(folio_add_wait_queue); |
David Howells | 385e1ca5f | 2009-04-03 16:42:39 +0100 | [diff] [blame] | 1491 | |
Linus Torvalds | b91e130 | 2016-12-27 11:40:38 -0800 | [diff] [blame] | 1492 | #ifndef clear_bit_unlock_is_negative_byte |
| 1493 | |
| 1494 | /* |
| 1495 | * PG_waiters is the high bit in the same byte as PG_lock. |
| 1496 | * |
| 1497 | * On x86 (and on many other architectures), we can clear PG_lock and |
| 1498 | * test the sign bit at the same time. But if the architecture does |
| 1499 | * not support that special operation, we just do this all by hand |
| 1500 | * instead. |
| 1501 | * |
| 1502 | * The read of PG_waiters has to be after (or concurrently with) PG_locked |
Ethon Paul | ffceeb6 | 2020-06-04 16:49:22 -0700 | [diff] [blame] | 1503 | * being cleared, but a memory barrier should be unnecessary since it is |
Linus Torvalds | b91e130 | 2016-12-27 11:40:38 -0800 | [diff] [blame] | 1504 | * in the same byte as PG_locked. |
| 1505 | */ |
| 1506 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) |
| 1507 | { |
| 1508 | clear_bit_unlock(nr, mem); |
| 1509 | /* smp_mb__after_atomic(); */ |
Olof Johansson | 98473f9 | 2016-12-29 14:16:07 -0800 | [diff] [blame] | 1510 | return test_bit(PG_waiters, mem); |
Linus Torvalds | b91e130 | 2016-12-27 11:40:38 -0800 | [diff] [blame] | 1511 | } |
| 1512 | |
| 1513 | #endif |
| 1514 | |
David Howells | 385e1ca5f | 2009-04-03 16:42:39 +0100 | [diff] [blame] | 1515 | /** |
Matthew Wilcox (Oracle) | 4e13642 | 2020-12-07 15:44:35 -0500 | [diff] [blame] | 1516 | * folio_unlock - Unlock a locked folio. |
| 1517 | * @folio: The folio. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1518 | * |
Matthew Wilcox (Oracle) | 4e13642 | 2020-12-07 15:44:35 -0500 | [diff] [blame] | 1519 | * Unlocks the folio and wakes up any thread sleeping on the page lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1520 | * |
Matthew Wilcox (Oracle) | 4e13642 | 2020-12-07 15:44:35 -0500 | [diff] [blame] | 1521 | * Context: May be called from interrupt or process context. May not be |
| 1522 | * called from NMI context. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1523 | */ |
Matthew Wilcox (Oracle) | 4e13642 | 2020-12-07 15:44:35 -0500 | [diff] [blame] | 1524 | void folio_unlock(struct folio *folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1525 | { |
Matthew Wilcox (Oracle) | 4e13642 | 2020-12-07 15:44:35 -0500 | [diff] [blame] | 1526 | /* Bit 7 allows x86 to check the byte's sign bit */ |
Linus Torvalds | b91e130 | 2016-12-27 11:40:38 -0800 | [diff] [blame] | 1527 | BUILD_BUG_ON(PG_waiters != 7); |
Matthew Wilcox (Oracle) | 4e13642 | 2020-12-07 15:44:35 -0500 | [diff] [blame] | 1528 | BUILD_BUG_ON(PG_locked > 7); |
| 1529 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
| 1530 | if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0))) |
Matthew Wilcox (Oracle) | 6974d7c | 2021-01-15 17:14:48 -0500 | [diff] [blame] | 1531 | folio_wake_bit(folio, PG_locked); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1532 | } |
Matthew Wilcox (Oracle) | 4e13642 | 2020-12-07 15:44:35 -0500 | [diff] [blame] | 1533 | EXPORT_SYMBOL(folio_unlock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1534 | |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 1535 | /** |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1536 | * folio_end_private_2 - Clear PG_private_2 and wake any waiters. |
| 1537 | * @folio: The folio. |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1538 | * |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1539 | * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for |
| 1540 | * it. The folio reference held for PG_private_2 being set is released. |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1541 | * |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1542 | * This is, for example, used when a netfs folio is being written to a local |
| 1543 | * disk cache, thereby allowing writes to the cache for the same folio to be |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1544 | * serialised. |
| 1545 | */ |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1546 | void folio_end_private_2(struct folio *folio) |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1547 | { |
Matthew Wilcox (Oracle) | 6974d7c | 2021-01-15 17:14:48 -0500 | [diff] [blame] | 1548 | VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio); |
| 1549 | clear_bit_unlock(PG_private_2, folio_flags(folio, 0)); |
| 1550 | folio_wake_bit(folio, PG_private_2); |
| 1551 | folio_put(folio); |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1552 | } |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1553 | EXPORT_SYMBOL(folio_end_private_2); |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1554 | |
| 1555 | /** |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1556 | * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio. |
| 1557 | * @folio: The folio to wait on. |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1558 | * |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1559 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio. |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1560 | */ |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1561 | void folio_wait_private_2(struct folio *folio) |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1562 | { |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1563 | while (folio_test_private_2(folio)) |
| 1564 | folio_wait_bit(folio, PG_private_2); |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1565 | } |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1566 | EXPORT_SYMBOL(folio_wait_private_2); |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1567 | |
| 1568 | /** |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1569 | * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio. |
| 1570 | * @folio: The folio to wait on. |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1571 | * |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1572 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1573 | * fatal signal is received by the calling task. |
| 1574 | * |
| 1575 | * Return: |
| 1576 | * - 0 if successful. |
| 1577 | * - -EINTR if a fatal signal was encountered. |
| 1578 | */ |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1579 | int folio_wait_private_2_killable(struct folio *folio) |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1580 | { |
| 1581 | int ret = 0; |
| 1582 | |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1583 | while (folio_test_private_2(folio)) { |
| 1584 | ret = folio_wait_bit_killable(folio, PG_private_2); |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1585 | if (ret < 0) |
| 1586 | break; |
| 1587 | } |
| 1588 | |
| 1589 | return ret; |
| 1590 | } |
Matthew Wilcox (Oracle) | b47393f | 2021-04-22 22:58:32 -0400 | [diff] [blame] | 1591 | EXPORT_SYMBOL(folio_wait_private_2_killable); |
David Howells | 73e10de | 2020-02-10 10:00:21 +0000 | [diff] [blame] | 1592 | |
| 1593 | /** |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1594 | * folio_end_writeback - End writeback against a folio. |
| 1595 | * @folio: The folio. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1596 | */ |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1597 | void folio_end_writeback(struct folio *folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1598 | { |
Mel Gorman | 888cf2d | 2014-06-04 16:10:34 -0700 | [diff] [blame] | 1599 | /* |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1600 | * folio_test_clear_reclaim() could be used here but it is an |
| 1601 | * atomic operation and overkill in this particular case. Failing |
| 1602 | * to shuffle a folio marked for immediate reclaim is too mild |
| 1603 | * a gain to justify taking an atomic operation penalty at the |
| 1604 | * end of every folio writeback. |
Mel Gorman | 888cf2d | 2014-06-04 16:10:34 -0700 | [diff] [blame] | 1605 | */ |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1606 | if (folio_test_reclaim(folio)) { |
| 1607 | folio_clear_reclaim(folio); |
Matthew Wilcox (Oracle) | 575ced1 | 2020-12-08 01:25:39 -0500 | [diff] [blame] | 1608 | folio_rotate_reclaimable(folio); |
Mel Gorman | 888cf2d | 2014-06-04 16:10:34 -0700 | [diff] [blame] | 1609 | } |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 1610 | |
Hugh Dickins | 073861e | 2020-11-24 08:46:43 -0800 | [diff] [blame] | 1611 | /* |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1612 | * Writeback does not hold a folio reference of its own, relying |
Hugh Dickins | 073861e | 2020-11-24 08:46:43 -0800 | [diff] [blame] | 1613 | * on truncation to wait for the clearing of PG_writeback. |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1614 | * But here we must make sure that the folio is not freed and |
| 1615 | * reused before the folio_wake(). |
Hugh Dickins | 073861e | 2020-11-24 08:46:43 -0800 | [diff] [blame] | 1616 | */ |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1617 | folio_get(folio); |
Matthew Wilcox (Oracle) | 269ccca3 | 2021-01-15 23:34:16 -0500 | [diff] [blame] | 1618 | if (!__folio_end_writeback(folio)) |
Miklos Szeredi | ac6aadb | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 1619 | BUG(); |
| 1620 | |
Peter Zijlstra | 4e857c5 | 2014-03-17 18:06:10 +0100 | [diff] [blame] | 1621 | smp_mb__after_atomic(); |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1622 | folio_wake(folio, PG_writeback); |
Linus Torvalds | 512b7931 | 2021-11-06 14:08:17 -0700 | [diff] [blame] | 1623 | acct_reclaim_writeback(folio); |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1624 | folio_put(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1625 | } |
Matthew Wilcox (Oracle) | 4268b48 | 2021-03-03 15:21:55 -0500 | [diff] [blame] | 1626 | EXPORT_SYMBOL(folio_end_writeback); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1627 | |
Matthew Wilcox | 57d9984 | 2014-06-04 16:07:45 -0700 | [diff] [blame] | 1628 | /* |
| 1629 | * After completing I/O on a page, call this routine to update the page |
| 1630 | * flags appropriately |
| 1631 | */ |
Jens Axboe | c11f0c0 | 2016-08-05 08:11:04 -0600 | [diff] [blame] | 1632 | void page_endio(struct page *page, bool is_write, int err) |
Matthew Wilcox | 57d9984 | 2014-06-04 16:07:45 -0700 | [diff] [blame] | 1633 | { |
Shaoqin Huang | 223ce49 | 2022-08-09 10:32:56 +0800 | [diff] [blame] | 1634 | struct folio *folio = page_folio(page); |
| 1635 | |
Jens Axboe | c11f0c0 | 2016-08-05 08:11:04 -0600 | [diff] [blame] | 1636 | if (!is_write) { |
Matthew Wilcox | 57d9984 | 2014-06-04 16:07:45 -0700 | [diff] [blame] | 1637 | if (!err) { |
Shaoqin Huang | 223ce49 | 2022-08-09 10:32:56 +0800 | [diff] [blame] | 1638 | folio_mark_uptodate(folio); |
Matthew Wilcox | 57d9984 | 2014-06-04 16:07:45 -0700 | [diff] [blame] | 1639 | } else { |
Shaoqin Huang | 223ce49 | 2022-08-09 10:32:56 +0800 | [diff] [blame] | 1640 | folio_clear_uptodate(folio); |
| 1641 | folio_set_error(folio); |
Matthew Wilcox | 57d9984 | 2014-06-04 16:07:45 -0700 | [diff] [blame] | 1642 | } |
Shaoqin Huang | 223ce49 | 2022-08-09 10:32:56 +0800 | [diff] [blame] | 1643 | folio_unlock(folio); |
Mike Christie | abf5454 | 2016-08-04 14:23:34 -0600 | [diff] [blame] | 1644 | } else { |
Matthew Wilcox | 57d9984 | 2014-06-04 16:07:45 -0700 | [diff] [blame] | 1645 | if (err) { |
Minchan Kim | dd8416c | 2017-02-24 14:59:59 -0800 | [diff] [blame] | 1646 | struct address_space *mapping; |
| 1647 | |
Shaoqin Huang | 223ce49 | 2022-08-09 10:32:56 +0800 | [diff] [blame] | 1648 | folio_set_error(folio); |
| 1649 | mapping = folio_mapping(folio); |
Minchan Kim | dd8416c | 2017-02-24 14:59:59 -0800 | [diff] [blame] | 1650 | if (mapping) |
| 1651 | mapping_set_error(mapping, err); |
Matthew Wilcox | 57d9984 | 2014-06-04 16:07:45 -0700 | [diff] [blame] | 1652 | } |
Shaoqin Huang | 223ce49 | 2022-08-09 10:32:56 +0800 | [diff] [blame] | 1653 | folio_end_writeback(folio); |
Matthew Wilcox | 57d9984 | 2014-06-04 16:07:45 -0700 | [diff] [blame] | 1654 | } |
| 1655 | } |
| 1656 | EXPORT_SYMBOL_GPL(page_endio); |
| 1657 | |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 1658 | /** |
Matthew Wilcox (Oracle) | 7c23c78 | 2021-03-01 19:38:25 -0500 | [diff] [blame] | 1659 | * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it. |
| 1660 | * @folio: The folio to lock |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1661 | */ |
Matthew Wilcox (Oracle) | 7c23c78 | 2021-03-01 19:38:25 -0500 | [diff] [blame] | 1662 | void __folio_lock(struct folio *folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1663 | { |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1664 | folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE, |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1665 | EXCLUSIVE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1666 | } |
Matthew Wilcox (Oracle) | 7c23c78 | 2021-03-01 19:38:25 -0500 | [diff] [blame] | 1667 | EXPORT_SYMBOL(__folio_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1668 | |
Matthew Wilcox (Oracle) | af7f29d | 2020-12-08 00:07:31 -0500 | [diff] [blame] | 1669 | int __folio_lock_killable(struct folio *folio) |
Matthew Wilcox | 2687a35 | 2007-12-06 11:18:49 -0500 | [diff] [blame] | 1670 | { |
Matthew Wilcox (Oracle) | 101c0bf | 2021-03-04 12:02:54 -0500 | [diff] [blame] | 1671 | return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE, |
Hugh Dickins | 9a1ea43 | 2018-12-28 00:36:14 -0800 | [diff] [blame] | 1672 | EXCLUSIVE); |
Matthew Wilcox | 2687a35 | 2007-12-06 11:18:49 -0500 | [diff] [blame] | 1673 | } |
Matthew Wilcox (Oracle) | af7f29d | 2020-12-08 00:07:31 -0500 | [diff] [blame] | 1674 | EXPORT_SYMBOL_GPL(__folio_lock_killable); |
Matthew Wilcox | 2687a35 | 2007-12-06 11:18:49 -0500 | [diff] [blame] | 1675 | |
Matthew Wilcox (Oracle) | ffdc8da | 2020-12-30 17:58:40 -0500 | [diff] [blame] | 1676 | static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait) |
Jens Axboe | dd3e6d5 | 2020-05-22 09:12:09 -0600 | [diff] [blame] | 1677 | { |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1678 | struct wait_queue_head *q = folio_waitqueue(folio); |
Matthew Wilcox (Oracle) | f32b5dd | 2021-02-24 12:02:09 -0800 | [diff] [blame] | 1679 | int ret = 0; |
| 1680 | |
Matthew Wilcox (Oracle) | df4d4f12 | 2021-01-16 11:22:14 -0500 | [diff] [blame] | 1681 | wait->folio = folio; |
Matthew Wilcox (Oracle) | f32b5dd | 2021-02-24 12:02:09 -0800 | [diff] [blame] | 1682 | wait->bit_nr = PG_locked; |
| 1683 | |
| 1684 | spin_lock_irq(&q->lock); |
| 1685 | __add_wait_queue_entry_tail(q, &wait->wait); |
Matthew Wilcox (Oracle) | ffdc8da | 2020-12-30 17:58:40 -0500 | [diff] [blame] | 1686 | folio_set_waiters(folio); |
| 1687 | ret = !folio_trylock(folio); |
Matthew Wilcox (Oracle) | f32b5dd | 2021-02-24 12:02:09 -0800 | [diff] [blame] | 1688 | /* |
| 1689 | * If we were successful now, we know we're still on the |
| 1690 | * waitqueue as we're still under the lock. This means it's |
| 1691 | * safe to remove and return success, we know the callback |
| 1692 | * isn't going to trigger. |
| 1693 | */ |
| 1694 | if (!ret) |
| 1695 | __remove_wait_queue(q, &wait->wait); |
| 1696 | else |
| 1697 | ret = -EIOCBQUEUED; |
| 1698 | spin_unlock_irq(&q->lock); |
| 1699 | return ret; |
Jens Axboe | dd3e6d5 | 2020-05-22 09:12:09 -0600 | [diff] [blame] | 1700 | } |
| 1701 | |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 1702 | /* |
| 1703 | * Return values: |
Matthew Wilcox (Oracle) | 9138e47 | 2021-03-18 21:39:45 -0400 | [diff] [blame] | 1704 | * true - folio is locked; mmap_lock is still held. |
| 1705 | * false - folio is not locked. |
Michel Lespinasse | 3e4e28c | 2020-06-08 21:33:51 -0700 | [diff] [blame] | 1706 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 1707 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 1708 | * which case mmap_lock is still held. |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 1709 | * |
Matthew Wilcox (Oracle) | 9138e47 | 2021-03-18 21:39:45 -0400 | [diff] [blame] | 1710 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return true |
| 1711 | * with the folio locked and the mmap_lock unperturbed. |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 1712 | */ |
Matthew Wilcox (Oracle) | 9138e47 | 2021-03-18 21:39:45 -0400 | [diff] [blame] | 1713 | bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm, |
Michel Lespinasse | d065bd8 | 2010-10-26 14:21:57 -0700 | [diff] [blame] | 1714 | unsigned int flags) |
| 1715 | { |
Peter Xu | 4064b98 | 2020-04-01 21:08:45 -0700 | [diff] [blame] | 1716 | if (fault_flag_allow_retry_first(flags)) { |
KOSAKI Motohiro | 37b23e0 | 2011-05-24 17:11:30 -0700 | [diff] [blame] | 1717 | /* |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 1718 | * CAUTION! In this case, mmap_lock is not released |
KOSAKI Motohiro | 37b23e0 | 2011-05-24 17:11:30 -0700 | [diff] [blame] | 1719 | * even though return 0. |
| 1720 | */ |
| 1721 | if (flags & FAULT_FLAG_RETRY_NOWAIT) |
Matthew Wilcox (Oracle) | 9138e47 | 2021-03-18 21:39:45 -0400 | [diff] [blame] | 1722 | return false; |
KOSAKI Motohiro | 37b23e0 | 2011-05-24 17:11:30 -0700 | [diff] [blame] | 1723 | |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 1724 | mmap_read_unlock(mm); |
KOSAKI Motohiro | 37b23e0 | 2011-05-24 17:11:30 -0700 | [diff] [blame] | 1725 | if (flags & FAULT_FLAG_KILLABLE) |
Matthew Wilcox (Oracle) | 6baa8d6 | 2021-03-04 10:21:02 -0500 | [diff] [blame] | 1726 | folio_wait_locked_killable(folio); |
KOSAKI Motohiro | 37b23e0 | 2011-05-24 17:11:30 -0700 | [diff] [blame] | 1727 | else |
Matthew Wilcox (Oracle) | 6baa8d6 | 2021-03-04 10:21:02 -0500 | [diff] [blame] | 1728 | folio_wait_locked(folio); |
Matthew Wilcox (Oracle) | 9138e47 | 2021-03-18 21:39:45 -0400 | [diff] [blame] | 1729 | return false; |
Michel Lespinasse | d065bd8 | 2010-10-26 14:21:57 -0700 | [diff] [blame] | 1730 | } |
Hailong Liu | 800bca7 | 2020-12-14 19:05:02 -0800 | [diff] [blame] | 1731 | if (flags & FAULT_FLAG_KILLABLE) { |
Matthew Wilcox (Oracle) | 9138e47 | 2021-03-18 21:39:45 -0400 | [diff] [blame] | 1732 | bool ret; |
Hailong Liu | 800bca7 | 2020-12-14 19:05:02 -0800 | [diff] [blame] | 1733 | |
Matthew Wilcox (Oracle) | af7f29d | 2020-12-08 00:07:31 -0500 | [diff] [blame] | 1734 | ret = __folio_lock_killable(folio); |
Hailong Liu | 800bca7 | 2020-12-14 19:05:02 -0800 | [diff] [blame] | 1735 | if (ret) { |
| 1736 | mmap_read_unlock(mm); |
Matthew Wilcox (Oracle) | 9138e47 | 2021-03-18 21:39:45 -0400 | [diff] [blame] | 1737 | return false; |
Hailong Liu | 800bca7 | 2020-12-14 19:05:02 -0800 | [diff] [blame] | 1738 | } |
| 1739 | } else { |
Matthew Wilcox (Oracle) | af7f29d | 2020-12-08 00:07:31 -0500 | [diff] [blame] | 1740 | __folio_lock(folio); |
Hailong Liu | 800bca7 | 2020-12-14 19:05:02 -0800 | [diff] [blame] | 1741 | } |
Hailong Liu | 800bca7 | 2020-12-14 19:05:02 -0800 | [diff] [blame] | 1742 | |
Matthew Wilcox (Oracle) | 9138e47 | 2021-03-18 21:39:45 -0400 | [diff] [blame] | 1743 | return true; |
Michel Lespinasse | d065bd8 | 2010-10-26 14:21:57 -0700 | [diff] [blame] | 1744 | } |
| 1745 | |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 1746 | /** |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1747 | * page_cache_next_miss() - Find the next gap in the page cache. |
| 1748 | * @mapping: Mapping. |
| 1749 | * @index: Index. |
| 1750 | * @max_scan: Maximum range to search. |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1751 | * |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1752 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
| 1753 | * gap with the lowest index. |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1754 | * |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1755 | * This function may be called under the rcu_read_lock. However, this will |
| 1756 | * not atomically search a snapshot of the cache at a single point in time. |
| 1757 | * For example, if a gap is created at index 5, then subsequently a gap is |
| 1758 | * created at index 10, page_cache_next_miss covering both indices may |
| 1759 | * return 10 if called under the rcu_read_lock. |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1760 | * |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1761 | * Return: The index of the gap if found, otherwise an index outside the |
| 1762 | * range specified (in which case 'return - index >= max_scan' will be true). |
| 1763 | * In the rare case of index wrap-around, 0 will be returned. |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1764 | */ |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1765 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1766 | pgoff_t index, unsigned long max_scan) |
| 1767 | { |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1768 | XA_STATE(xas, &mapping->i_pages, index); |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1769 | |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1770 | while (max_scan--) { |
| 1771 | void *entry = xas_next(&xas); |
| 1772 | if (!entry || xa_is_value(entry)) |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1773 | break; |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1774 | if (xas.xa_index == 0) |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1775 | break; |
| 1776 | } |
| 1777 | |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1778 | return xas.xa_index; |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1779 | } |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1780 | EXPORT_SYMBOL(page_cache_next_miss); |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1781 | |
| 1782 | /** |
Laurent Dufour | 2346a56 | 2019-05-13 17:21:29 -0700 | [diff] [blame] | 1783 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1784 | * @mapping: Mapping. |
| 1785 | * @index: Index. |
| 1786 | * @max_scan: Maximum range to search. |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1787 | * |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1788 | * Search the range [max(index - max_scan + 1, 0), index] for the |
| 1789 | * gap with the highest index. |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1790 | * |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1791 | * This function may be called under the rcu_read_lock. However, this will |
| 1792 | * not atomically search a snapshot of the cache at a single point in time. |
| 1793 | * For example, if a gap is created at index 10, then subsequently a gap is |
| 1794 | * created at index 5, page_cache_prev_miss() covering both indices may |
| 1795 | * return 5 if called under the rcu_read_lock. |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1796 | * |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1797 | * Return: The index of the gap if found, otherwise an index outside the |
| 1798 | * range specified (in which case 'index - return >= max_scan' will be true). |
| 1799 | * In the rare case of wrap-around, ULONG_MAX will be returned. |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1800 | */ |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1801 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1802 | pgoff_t index, unsigned long max_scan) |
| 1803 | { |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1804 | XA_STATE(xas, &mapping->i_pages, index); |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1805 | |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1806 | while (max_scan--) { |
| 1807 | void *entry = xas_prev(&xas); |
| 1808 | if (!entry || xa_is_value(entry)) |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1809 | break; |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1810 | if (xas.xa_index == ULONG_MAX) |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1811 | break; |
| 1812 | } |
| 1813 | |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1814 | return xas.xa_index; |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1815 | } |
Matthew Wilcox | 0d3f929 | 2017-11-21 14:07:06 -0500 | [diff] [blame] | 1816 | EXPORT_SYMBOL(page_cache_prev_miss); |
Johannes Weiner | e7b563b | 2014-04-03 14:47:44 -0700 | [diff] [blame] | 1817 | |
Matthew Wilcox (Oracle) | 44835d2 | 2021-02-25 17:15:36 -0800 | [diff] [blame] | 1818 | /* |
Matthew Wilcox (Oracle) | 020853b | 2021-05-10 16:33:22 -0400 | [diff] [blame] | 1819 | * Lockless page cache protocol: |
| 1820 | * On the lookup side: |
| 1821 | * 1. Load the folio from i_pages |
| 1822 | * 2. Increment the refcount if it's not zero |
| 1823 | * 3. If the folio is not found by xas_reload(), put the refcount and retry |
| 1824 | * |
| 1825 | * On the removal side: |
| 1826 | * A. Freeze the page (by zeroing the refcount if nobody else has a reference) |
| 1827 | * B. Remove the page from i_pages |
| 1828 | * C. Return the page to the page allocator |
| 1829 | * |
| 1830 | * This means that any page may have its reference count temporarily |
| 1831 | * increased by a speculative page cache (or fast GUP) lookup as it can |
| 1832 | * be allocated by another user before the RCU grace period expires. |
| 1833 | * Because the refcount temporarily acquired here may end up being the |
| 1834 | * last refcount on the page, any page allocation must be freeable by |
| 1835 | * folio_put(). |
| 1836 | */ |
| 1837 | |
| 1838 | /* |
Matthew Wilcox (Oracle) | bc5a301 | 2021-02-25 17:15:40 -0800 | [diff] [blame] | 1839 | * mapping_get_entry - Get a page cache entry. |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 1840 | * @mapping: the address_space to search |
Matthew Wilcox (Oracle) | a6de4b4 | 2020-10-13 16:51:34 -0700 | [diff] [blame] | 1841 | * @index: The page cache index. |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 1842 | * |
Matthew Wilcox (Oracle) | bca65ee | 2020-12-15 23:22:38 -0500 | [diff] [blame] | 1843 | * Looks up the page cache entry at @mapping & @index. If it is a folio, |
| 1844 | * it is returned with an increased refcount. If it is a shadow entry |
| 1845 | * of a previously evicted folio, or a swap entry from shmem/tmpfs, |
| 1846 | * it is returned without further action. |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 1847 | * |
Matthew Wilcox (Oracle) | bca65ee | 2020-12-15 23:22:38 -0500 | [diff] [blame] | 1848 | * Return: The folio, swap or shadow entry, %NULL if nothing is found. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1849 | */ |
Matthew Wilcox (Oracle) | bca65ee | 2020-12-15 23:22:38 -0500 | [diff] [blame] | 1850 | static void *mapping_get_entry(struct address_space *mapping, pgoff_t index) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1851 | { |
Matthew Wilcox (Oracle) | a6de4b4 | 2020-10-13 16:51:34 -0700 | [diff] [blame] | 1852 | XA_STATE(xas, &mapping->i_pages, index); |
Matthew Wilcox (Oracle) | bca65ee | 2020-12-15 23:22:38 -0500 | [diff] [blame] | 1853 | struct folio *folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1854 | |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 1855 | rcu_read_lock(); |
| 1856 | repeat: |
Matthew Wilcox | 4c7472c | 2018-05-16 16:12:50 -0400 | [diff] [blame] | 1857 | xas_reset(&xas); |
Matthew Wilcox (Oracle) | bca65ee | 2020-12-15 23:22:38 -0500 | [diff] [blame] | 1858 | folio = xas_load(&xas); |
| 1859 | if (xas_retry(&xas, folio)) |
Matthew Wilcox | 4c7472c | 2018-05-16 16:12:50 -0400 | [diff] [blame] | 1860 | goto repeat; |
| 1861 | /* |
| 1862 | * A shadow entry of a recently evicted page, or a swap entry from |
| 1863 | * shmem/tmpfs. Return it without attempting to raise page count. |
| 1864 | */ |
Matthew Wilcox (Oracle) | bca65ee | 2020-12-15 23:22:38 -0500 | [diff] [blame] | 1865 | if (!folio || xa_is_value(folio)) |
Matthew Wilcox | 4c7472c | 2018-05-16 16:12:50 -0400 | [diff] [blame] | 1866 | goto out; |
Kirill A. Shutemov | 8392937 | 2016-07-26 15:26:04 -0700 | [diff] [blame] | 1867 | |
Matthew Wilcox (Oracle) | bca65ee | 2020-12-15 23:22:38 -0500 | [diff] [blame] | 1868 | if (!folio_try_get_rcu(folio)) |
Matthew Wilcox | 4c7472c | 2018-05-16 16:12:50 -0400 | [diff] [blame] | 1869 | goto repeat; |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 1870 | |
Matthew Wilcox (Oracle) | bca65ee | 2020-12-15 23:22:38 -0500 | [diff] [blame] | 1871 | if (unlikely(folio != xas_reload(&xas))) { |
| 1872 | folio_put(folio); |
Matthew Wilcox | 4c7472c | 2018-05-16 16:12:50 -0400 | [diff] [blame] | 1873 | goto repeat; |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 1874 | } |
Nick Piggin | 27d20fd | 2010-11-11 14:05:19 -0800 | [diff] [blame] | 1875 | out: |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 1876 | rcu_read_unlock(); |
| 1877 | |
Matthew Wilcox (Oracle) | bca65ee | 2020-12-15 23:22:38 -0500 | [diff] [blame] | 1878 | return folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1879 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1880 | |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 1881 | /** |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1882 | * __filemap_get_folio - Find and get a reference to a folio. |
Matthew Wilcox (Oracle) | 2294b32 | 2020-04-01 21:05:07 -0700 | [diff] [blame] | 1883 | * @mapping: The address_space to search. |
| 1884 | * @index: The page index. |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1885 | * @fgp_flags: %FGP flags modify how the folio is returned. |
| 1886 | * @gfp: Memory allocation flags to use if %FGP_CREAT is specified. |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 1887 | * |
Matthew Wilcox (Oracle) | 2294b32 | 2020-04-01 21:05:07 -0700 | [diff] [blame] | 1888 | * Looks up the page cache entry at @mapping & @index. |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 1889 | * |
Matthew Wilcox (Oracle) | 2294b32 | 2020-04-01 21:05:07 -0700 | [diff] [blame] | 1890 | * @fgp_flags can be zero or more of these flags: |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 1891 | * |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1892 | * * %FGP_ACCESSED - The folio will be marked accessed. |
| 1893 | * * %FGP_LOCK - The folio is returned locked. |
Matthew Wilcox (Oracle) | 44835d2 | 2021-02-25 17:15:36 -0800 | [diff] [blame] | 1894 | * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1895 | * instead of allocating a new folio to replace it. |
Matthew Wilcox (Oracle) | 2294b32 | 2020-04-01 21:05:07 -0700 | [diff] [blame] | 1896 | * * %FGP_CREAT - If no page is present then a new page is allocated using |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1897 | * @gfp and added to the page cache and the VM's LRU list. |
Matthew Wilcox (Oracle) | 2294b32 | 2020-04-01 21:05:07 -0700 | [diff] [blame] | 1898 | * The page is returned locked and with an increased refcount. |
| 1899 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the |
| 1900 | * page is already in cache. If the page was allocated, unlock it before |
| 1901 | * returning so the caller can do the same dance. |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1902 | * * %FGP_WRITE - The page will be written to by the caller. |
| 1903 | * * %FGP_NOFS - __GFP_FS will get cleared in gfp. |
| 1904 | * * %FGP_NOWAIT - Don't get blocked by page lock. |
Matthew Wilcox (Oracle) | b27652d9 | 2020-12-24 12:55:56 -0500 | [diff] [blame] | 1905 | * * %FGP_STABLE - Wait for the folio to be stable (finished writeback) |
mchehab@s-opensource.com | 0e056eb | 2017-03-30 17:11:36 -0300 | [diff] [blame] | 1906 | * |
Matthew Wilcox (Oracle) | 2294b32 | 2020-04-01 21:05:07 -0700 | [diff] [blame] | 1907 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
| 1908 | * if the %GFP flags specified for %FGP_CREAT are atomic. |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1909 | * |
| 1910 | * If there is a page cache page, it is returned with an increased refcount. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 1911 | * |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1912 | * Return: The found folio or %NULL otherwise. |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 1913 | */ |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1914 | struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index, |
| 1915 | int fgp_flags, gfp_t gfp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1916 | { |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1917 | struct folio *folio; |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1918 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1919 | repeat: |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1920 | folio = mapping_get_entry(mapping, index); |
| 1921 | if (xa_is_value(folio)) { |
Matthew Wilcox (Oracle) | 44835d2 | 2021-02-25 17:15:36 -0800 | [diff] [blame] | 1922 | if (fgp_flags & FGP_ENTRY) |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1923 | return folio; |
| 1924 | folio = NULL; |
Matthew Wilcox (Oracle) | 44835d2 | 2021-02-25 17:15:36 -0800 | [diff] [blame] | 1925 | } |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1926 | if (!folio) |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1927 | goto no_page; |
| 1928 | |
| 1929 | if (fgp_flags & FGP_LOCK) { |
| 1930 | if (fgp_flags & FGP_NOWAIT) { |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1931 | if (!folio_trylock(folio)) { |
| 1932 | folio_put(folio); |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1933 | return NULL; |
| 1934 | } |
| 1935 | } else { |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1936 | folio_lock(folio); |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1937 | } |
| 1938 | |
| 1939 | /* Has the page been truncated? */ |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1940 | if (unlikely(folio->mapping != mapping)) { |
| 1941 | folio_unlock(folio); |
| 1942 | folio_put(folio); |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1943 | goto repeat; |
| 1944 | } |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1945 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1946 | } |
| 1947 | |
Kirill Tkhai | c16eb00 | 2018-12-28 00:37:35 -0800 | [diff] [blame] | 1948 | if (fgp_flags & FGP_ACCESSED) |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1949 | folio_mark_accessed(folio); |
Yang Shi | b9306a7 | 2020-08-06 23:19:55 -0700 | [diff] [blame] | 1950 | else if (fgp_flags & FGP_WRITE) { |
| 1951 | /* Clear idle flag for buffer write */ |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1952 | if (folio_test_idle(folio)) |
| 1953 | folio_clear_idle(folio); |
Yang Shi | b9306a7 | 2020-08-06 23:19:55 -0700 | [diff] [blame] | 1954 | } |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1955 | |
Matthew Wilcox (Oracle) | b27652d9 | 2020-12-24 12:55:56 -0500 | [diff] [blame] | 1956 | if (fgp_flags & FGP_STABLE) |
| 1957 | folio_wait_stable(folio); |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1958 | no_page: |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1959 | if (!folio && (fgp_flags & FGP_CREAT)) { |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1960 | int err; |
Christoph Hellwig | f56753a | 2020-09-24 08:51:40 +0200 | [diff] [blame] | 1961 | if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping)) |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1962 | gfp |= __GFP_WRITE; |
Michal Hocko | 45f87de | 2014-12-29 20:30:35 +0100 | [diff] [blame] | 1963 | if (fgp_flags & FGP_NOFS) |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1964 | gfp &= ~__GFP_FS; |
Jens Axboe | 0dd316b | 2022-07-01 14:04:43 -0600 | [diff] [blame] | 1965 | if (fgp_flags & FGP_NOWAIT) { |
| 1966 | gfp &= ~GFP_KERNEL; |
| 1967 | gfp |= GFP_NOWAIT | __GFP_NOWARN; |
| 1968 | } |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1969 | |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1970 | folio = filemap_alloc_folio(gfp, 0); |
| 1971 | if (!folio) |
Nick Piggin | eb2be18 | 2007-10-16 01:24:57 -0700 | [diff] [blame] | 1972 | return NULL; |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1973 | |
Josef Bacik | a75d4c3 | 2019-03-13 11:44:14 -0700 | [diff] [blame] | 1974 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1975 | fgp_flags |= FGP_LOCK; |
| 1976 | |
Hugh Dickins | eb39d61 | 2014-08-06 16:06:43 -0700 | [diff] [blame] | 1977 | /* Init accessed so avoid atomic mark_page_accessed later */ |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1978 | if (fgp_flags & FGP_ACCESSED) |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1979 | __folio_set_referenced(folio); |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1980 | |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1981 | err = filemap_add_folio(mapping, folio, index, gfp); |
Nick Piggin | eb2be18 | 2007-10-16 01:24:57 -0700 | [diff] [blame] | 1982 | if (unlikely(err)) { |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1983 | folio_put(folio); |
| 1984 | folio = NULL; |
Nick Piggin | eb2be18 | 2007-10-16 01:24:57 -0700 | [diff] [blame] | 1985 | if (err == -EEXIST) |
| 1986 | goto repeat; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1987 | } |
Josef Bacik | a75d4c3 | 2019-03-13 11:44:14 -0700 | [diff] [blame] | 1988 | |
| 1989 | /* |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1990 | * filemap_add_folio locks the page, and for mmap |
| 1991 | * we expect an unlocked page. |
Josef Bacik | a75d4c3 | 2019-03-13 11:44:14 -0700 | [diff] [blame] | 1992 | */ |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1993 | if (folio && (fgp_flags & FGP_FOR_MMAP)) |
| 1994 | folio_unlock(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1995 | } |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1996 | |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1997 | return folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1998 | } |
Matthew Wilcox (Oracle) | 3f0c6a0 | 2021-03-08 11:45:35 -0500 | [diff] [blame] | 1999 | EXPORT_SYMBOL(__filemap_get_folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2000 | |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2001 | static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max, |
Matthew Wilcox (Oracle) | c7bad63 | 2021-02-25 17:15:44 -0800 | [diff] [blame] | 2002 | xa_mark_t mark) |
| 2003 | { |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2004 | struct folio *folio; |
Matthew Wilcox (Oracle) | c7bad63 | 2021-02-25 17:15:44 -0800 | [diff] [blame] | 2005 | |
| 2006 | retry: |
| 2007 | if (mark == XA_PRESENT) |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2008 | folio = xas_find(xas, max); |
Matthew Wilcox (Oracle) | c7bad63 | 2021-02-25 17:15:44 -0800 | [diff] [blame] | 2009 | else |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2010 | folio = xas_find_marked(xas, max, mark); |
Matthew Wilcox (Oracle) | c7bad63 | 2021-02-25 17:15:44 -0800 | [diff] [blame] | 2011 | |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2012 | if (xas_retry(xas, folio)) |
Matthew Wilcox (Oracle) | c7bad63 | 2021-02-25 17:15:44 -0800 | [diff] [blame] | 2013 | goto retry; |
| 2014 | /* |
| 2015 | * A shadow entry of a recently evicted page, a swap |
| 2016 | * entry from shmem/tmpfs or a DAX entry. Return it |
| 2017 | * without attempting to raise page count. |
| 2018 | */ |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2019 | if (!folio || xa_is_value(folio)) |
| 2020 | return folio; |
Matthew Wilcox (Oracle) | c7bad63 | 2021-02-25 17:15:44 -0800 | [diff] [blame] | 2021 | |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2022 | if (!folio_try_get_rcu(folio)) |
Matthew Wilcox (Oracle) | c7bad63 | 2021-02-25 17:15:44 -0800 | [diff] [blame] | 2023 | goto reset; |
| 2024 | |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2025 | if (unlikely(folio != xas_reload(xas))) { |
| 2026 | folio_put(folio); |
Matthew Wilcox (Oracle) | c7bad63 | 2021-02-25 17:15:44 -0800 | [diff] [blame] | 2027 | goto reset; |
| 2028 | } |
| 2029 | |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2030 | return folio; |
Matthew Wilcox (Oracle) | c7bad63 | 2021-02-25 17:15:44 -0800 | [diff] [blame] | 2031 | reset: |
| 2032 | xas_reset(xas); |
| 2033 | goto retry; |
| 2034 | } |
| 2035 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2036 | /** |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 2037 | * find_get_entries - gang pagecache lookup |
| 2038 | * @mapping: The address_space to search |
| 2039 | * @start: The starting page cache index |
Matthew Wilcox (Oracle) | ca122fe4 | 2021-02-25 17:16:00 -0800 | [diff] [blame] | 2040 | * @end: The final page index (inclusive). |
Matthew Wilcox (Oracle) | 0e499ed | 2020-09-01 23:17:50 -0400 | [diff] [blame] | 2041 | * @fbatch: Where the resulting entries are placed. |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 2042 | * @indices: The cache indices corresponding to the entries in @entries |
| 2043 | * |
Matthew Wilcox (Oracle) | cf2039a | 2021-02-25 17:16:11 -0800 | [diff] [blame] | 2044 | * find_get_entries() will search for and return a batch of entries in |
Matthew Wilcox (Oracle) | 0e499ed | 2020-09-01 23:17:50 -0400 | [diff] [blame] | 2045 | * the mapping. The entries are placed in @fbatch. find_get_entries() |
| 2046 | * takes a reference on any actual folios it returns. |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 2047 | * |
Matthew Wilcox (Oracle) | 0e499ed | 2020-09-01 23:17:50 -0400 | [diff] [blame] | 2048 | * The entries have ascending indexes. The indices may not be consecutive |
| 2049 | * due to not-present entries or large folios. |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 2050 | * |
Matthew Wilcox (Oracle) | 0e499ed | 2020-09-01 23:17:50 -0400 | [diff] [blame] | 2051 | * Any shadow entries of evicted folios, or swap entries from |
Johannes Weiner | 139b6a6 | 2014-05-06 12:50:05 -0700 | [diff] [blame] | 2052 | * shmem/tmpfs, are included in the returned array. |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 2053 | * |
Matthew Wilcox (Oracle) | 0e499ed | 2020-09-01 23:17:50 -0400 | [diff] [blame] | 2054 | * Return: The number of entries which were found. |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 2055 | */ |
Vishal Moola (Oracle) | 9fb6bee | 2022-10-17 09:18:00 -0700 | [diff] [blame] | 2056 | unsigned find_get_entries(struct address_space *mapping, pgoff_t *start, |
Matthew Wilcox (Oracle) | 0e499ed | 2020-09-01 23:17:50 -0400 | [diff] [blame] | 2057 | pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices) |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 2058 | { |
Vishal Moola (Oracle) | 9fb6bee | 2022-10-17 09:18:00 -0700 | [diff] [blame] | 2059 | XA_STATE(xas, &mapping->i_pages, *start); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2060 | struct folio *folio; |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 2061 | |
| 2062 | rcu_read_lock(); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2063 | while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) { |
Matthew Wilcox (Oracle) | 0e499ed | 2020-09-01 23:17:50 -0400 | [diff] [blame] | 2064 | indices[fbatch->nr] = xas.xa_index; |
| 2065 | if (!folio_batch_add(fbatch, folio)) |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 2066 | break; |
| 2067 | } |
| 2068 | rcu_read_unlock(); |
Matthew Wilcox (Oracle) | cf2039a | 2021-02-25 17:16:11 -0800 | [diff] [blame] | 2069 | |
Vishal Moola (Oracle) | 9fb6bee | 2022-10-17 09:18:00 -0700 | [diff] [blame] | 2070 | if (folio_batch_count(fbatch)) { |
| 2071 | unsigned long nr = 1; |
| 2072 | int idx = folio_batch_count(fbatch) - 1; |
| 2073 | |
| 2074 | folio = fbatch->folios[idx]; |
| 2075 | if (!xa_is_value(folio) && !folio_test_hugetlb(folio)) |
| 2076 | nr = folio_nr_pages(folio); |
| 2077 | *start = indices[idx] + nr; |
| 2078 | } |
Matthew Wilcox (Oracle) | 0e499ed | 2020-09-01 23:17:50 -0400 | [diff] [blame] | 2079 | return folio_batch_count(fbatch); |
Johannes Weiner | 0cd6144 | 2014-04-03 14:47:46 -0700 | [diff] [blame] | 2080 | } |
| 2081 | |
| 2082 | /** |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2083 | * find_lock_entries - Find a batch of pagecache entries. |
| 2084 | * @mapping: The address_space to search. |
| 2085 | * @start: The starting page cache index. |
| 2086 | * @end: The final page index (inclusive). |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 2087 | * @fbatch: Where the resulting entries are placed. |
| 2088 | * @indices: The cache indices of the entries in @fbatch. |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2089 | * |
| 2090 | * find_lock_entries() will return a batch of entries from @mapping. |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2091 | * Swap, shadow and DAX entries are included. Folios are returned |
| 2092 | * locked and with an incremented refcount. Folios which are locked |
| 2093 | * by somebody else or under writeback are skipped. Folios which are |
| 2094 | * partially outside the range are not returned. |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2095 | * |
| 2096 | * The entries have ascending indexes. The indices may not be consecutive |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2097 | * due to not-present entries, large folios, folios which could not be |
| 2098 | * locked or folios under writeback. |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2099 | * |
| 2100 | * Return: The number of entries which were found. |
| 2101 | */ |
Vishal Moola (Oracle) | 3392ca12 | 2022-10-17 09:17:59 -0700 | [diff] [blame] | 2102 | unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start, |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 2103 | pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices) |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2104 | { |
Vishal Moola (Oracle) | 3392ca12 | 2022-10-17 09:17:59 -0700 | [diff] [blame] | 2105 | XA_STATE(xas, &mapping->i_pages, *start); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2106 | struct folio *folio; |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2107 | |
| 2108 | rcu_read_lock(); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2109 | while ((folio = find_get_entry(&xas, end, XA_PRESENT))) { |
| 2110 | if (!xa_is_value(folio)) { |
Vishal Moola (Oracle) | 3392ca12 | 2022-10-17 09:17:59 -0700 | [diff] [blame] | 2111 | if (folio->index < *start) |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2112 | goto put; |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2113 | if (folio->index + folio_nr_pages(folio) - 1 > end) |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2114 | goto put; |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2115 | if (!folio_trylock(folio)) |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2116 | goto put; |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2117 | if (folio->mapping != mapping || |
| 2118 | folio_test_writeback(folio)) |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2119 | goto unlock; |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2120 | VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index), |
| 2121 | folio); |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2122 | } |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 2123 | indices[fbatch->nr] = xas.xa_index; |
| 2124 | if (!folio_batch_add(fbatch, folio)) |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2125 | break; |
Matthew Wilcox (Oracle) | 6b24ca4 | 2020-06-27 22:19:08 -0400 | [diff] [blame] | 2126 | continue; |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2127 | unlock: |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2128 | folio_unlock(folio); |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2129 | put: |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2130 | folio_put(folio); |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2131 | } |
| 2132 | rcu_read_unlock(); |
| 2133 | |
Vishal Moola (Oracle) | 3392ca12 | 2022-10-17 09:17:59 -0700 | [diff] [blame] | 2134 | if (folio_batch_count(fbatch)) { |
| 2135 | unsigned long nr = 1; |
| 2136 | int idx = folio_batch_count(fbatch) - 1; |
| 2137 | |
| 2138 | folio = fbatch->folios[idx]; |
| 2139 | if (!xa_is_value(folio) && !folio_test_hugetlb(folio)) |
| 2140 | nr = folio_nr_pages(folio); |
| 2141 | *start = indices[idx] + nr; |
| 2142 | } |
Matthew Wilcox (Oracle) | 51dcbda | 2021-12-07 14:15:07 -0500 | [diff] [blame] | 2143 | return folio_batch_count(fbatch); |
Matthew Wilcox (Oracle) | 5c211ba | 2021-02-25 17:15:56 -0800 | [diff] [blame] | 2144 | } |
| 2145 | |
| 2146 | /** |
Matthew Wilcox (Oracle) | be0ced5 | 2022-06-03 15:30:25 -0400 | [diff] [blame] | 2147 | * filemap_get_folios - Get a batch of folios |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2148 | * @mapping: The address_space to search |
| 2149 | * @start: The starting page index |
Jan Kara | b947cee | 2017-09-06 16:21:21 -0700 | [diff] [blame] | 2150 | * @end: The final page index (inclusive) |
Matthew Wilcox (Oracle) | be0ced5 | 2022-06-03 15:30:25 -0400 | [diff] [blame] | 2151 | * @fbatch: The batch to fill. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2152 | * |
Matthew Wilcox (Oracle) | be0ced5 | 2022-06-03 15:30:25 -0400 | [diff] [blame] | 2153 | * Search for and return a batch of folios in the mapping starting at |
| 2154 | * index @start and up to index @end (inclusive). The folios are returned |
| 2155 | * in @fbatch with an elevated reference count. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2156 | * |
Matthew Wilcox (Oracle) | be0ced5 | 2022-06-03 15:30:25 -0400 | [diff] [blame] | 2157 | * The first folio may start before @start; if it does, it will contain |
| 2158 | * @start. The final folio may extend beyond @end; if it does, it will |
| 2159 | * contain @end. The folios have ascending indices. There may be gaps |
| 2160 | * between the folios if there are indices which have no folio in the |
| 2161 | * page cache. If folios are added to or removed from the page cache |
| 2162 | * while this is running, they may or may not be found by this call. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2163 | * |
Matthew Wilcox (Oracle) | be0ced5 | 2022-06-03 15:30:25 -0400 | [diff] [blame] | 2164 | * Return: The number of folios which were found. |
| 2165 | * We also update @start to index the next folio for the traversal. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2166 | */ |
Matthew Wilcox (Oracle) | be0ced5 | 2022-06-03 15:30:25 -0400 | [diff] [blame] | 2167 | unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start, |
| 2168 | pgoff_t end, struct folio_batch *fbatch) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2169 | { |
Matthew Wilcox | fd1b3ce | 2018-05-16 17:38:56 -0400 | [diff] [blame] | 2170 | XA_STATE(xas, &mapping->i_pages, *start); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2171 | struct folio *folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2172 | |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2173 | rcu_read_lock(); |
Matthew Wilcox (Oracle) | be0ced5 | 2022-06-03 15:30:25 -0400 | [diff] [blame] | 2174 | while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) { |
Matthew Wilcox | fd1b3ce | 2018-05-16 17:38:56 -0400 | [diff] [blame] | 2175 | /* Skip over shadow, swap and DAX entries */ |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2176 | if (xa_is_value(folio)) |
Hugh Dickins | 8079b1c | 2011-08-03 16:21:28 -0700 | [diff] [blame] | 2177 | continue; |
Matthew Wilcox (Oracle) | be0ced5 | 2022-06-03 15:30:25 -0400 | [diff] [blame] | 2178 | if (!folio_batch_add(fbatch, folio)) { |
| 2179 | unsigned long nr = folio_nr_pages(folio); |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2180 | |
Matthew Wilcox (Oracle) | be0ced5 | 2022-06-03 15:30:25 -0400 | [diff] [blame] | 2181 | if (folio_test_hugetlb(folio)) |
| 2182 | nr = 1; |
| 2183 | *start = folio->index + nr; |
Jan Kara | b947cee | 2017-09-06 16:21:21 -0700 | [diff] [blame] | 2184 | goto out; |
| 2185 | } |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2186 | } |
Hugh Dickins | 5b280c0 | 2011-03-22 16:33:07 -0700 | [diff] [blame] | 2187 | |
Jan Kara | b947cee | 2017-09-06 16:21:21 -0700 | [diff] [blame] | 2188 | /* |
| 2189 | * We come here when there is no page beyond @end. We take care to not |
| 2190 | * overflow the index @start as it confuses some of the callers. This |
Matthew Wilcox | fd1b3ce | 2018-05-16 17:38:56 -0400 | [diff] [blame] | 2191 | * breaks the iteration when there is a page at index -1 but that is |
Jan Kara | b947cee | 2017-09-06 16:21:21 -0700 | [diff] [blame] | 2192 | * already broken anyway. |
| 2193 | */ |
| 2194 | if (end == (pgoff_t)-1) |
| 2195 | *start = (pgoff_t)-1; |
| 2196 | else |
| 2197 | *start = end + 1; |
| 2198 | out: |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2199 | rcu_read_unlock(); |
Jan Kara | d72dc8a | 2017-09-06 16:21:18 -0700 | [diff] [blame] | 2200 | |
Matthew Wilcox (Oracle) | be0ced5 | 2022-06-03 15:30:25 -0400 | [diff] [blame] | 2201 | return folio_batch_count(fbatch); |
| 2202 | } |
| 2203 | EXPORT_SYMBOL(filemap_get_folios); |
| 2204 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2205 | static inline |
Jens Axboe | ebf4350 | 2006-04-27 08:46:01 +0200 | [diff] [blame] | 2206 | bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max) |
| 2207 | { |
| 2208 | if (!folio_test_large(folio) || folio_test_hugetlb(folio)) |
| 2209 | return false; |
| 2210 | if (index >= max) |
| 2211 | return false; |
| 2212 | return index < folio->index + folio_nr_pages(folio) - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2213 | } |
| 2214 | |
Jens Axboe | ebf4350 | 2006-04-27 08:46:01 +0200 | [diff] [blame] | 2215 | /** |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2216 | * filemap_get_folios_contig - Get a batch of contiguous folios |
Jens Axboe | ebf4350 | 2006-04-27 08:46:01 +0200 | [diff] [blame] | 2217 | * @mapping: The address_space to search |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2218 | * @start: The starting page index |
| 2219 | * @end: The final page index (inclusive) |
| 2220 | * @fbatch: The batch to fill |
Jens Axboe | ebf4350 | 2006-04-27 08:46:01 +0200 | [diff] [blame] | 2221 | * |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2222 | * filemap_get_folios_contig() works exactly like filemap_get_folios(), |
| 2223 | * except the returned folios are guaranteed to be contiguous. This may |
| 2224 | * not return all contiguous folios if the batch gets filled up. |
Jens Axboe | ebf4350 | 2006-04-27 08:46:01 +0200 | [diff] [blame] | 2225 | * |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2226 | * Return: The number of folios found. |
| 2227 | * Also update @start to be positioned for traversal of the next folio. |
Jens Axboe | ebf4350 | 2006-04-27 08:46:01 +0200 | [diff] [blame] | 2228 | */ |
Konstantin Khlebnikov | 0fc9d10 | 2012-03-28 14:42:54 -0700 | [diff] [blame] | 2229 | |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2230 | unsigned filemap_get_folios_contig(struct address_space *mapping, |
| 2231 | pgoff_t *start, pgoff_t end, struct folio_batch *fbatch) |
| 2232 | { |
| 2233 | XA_STATE(xas, &mapping->i_pages, *start); |
| 2234 | unsigned long nr; |
| 2235 | struct folio *folio; |
Jens Axboe | ebf4350 | 2006-04-27 08:46:01 +0200 | [diff] [blame] | 2236 | |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2237 | rcu_read_lock(); |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2238 | |
| 2239 | for (folio = xas_load(&xas); folio && xas.xa_index <= end; |
| 2240 | folio = xas_next(&xas)) { |
Matthew Wilcox (Oracle) | e1c3772 | 2021-03-06 16:38:38 -0500 | [diff] [blame] | 2241 | if (xas_retry(&xas, folio)) |
Matthew Wilcox | 3ece58a | 2018-05-16 18:00:33 -0400 | [diff] [blame] | 2242 | continue; |
| 2243 | /* |
| 2244 | * If the entry has been swapped out, we can stop looking. |
| 2245 | * No current caller is looking for DAX entries. |
| 2246 | */ |
Matthew Wilcox (Oracle) | e1c3772 | 2021-03-06 16:38:38 -0500 | [diff] [blame] | 2247 | if (xa_is_value(folio)) |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2248 | goto update_start; |
Hugh Dickins | 9d8aa4e | 2011-03-22 16:33:06 -0700 | [diff] [blame] | 2249 | |
Matthew Wilcox (Oracle) | e1c3772 | 2021-03-06 16:38:38 -0500 | [diff] [blame] | 2250 | if (!folio_try_get_rcu(folio)) |
Matthew Wilcox | 3ece58a | 2018-05-16 18:00:33 -0400 | [diff] [blame] | 2251 | goto retry; |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2252 | |
Matthew Wilcox (Oracle) | e1c3772 | 2021-03-06 16:38:38 -0500 | [diff] [blame] | 2253 | if (unlikely(folio != xas_reload(&xas))) |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2254 | goto put_folio; |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2255 | |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2256 | if (!folio_batch_add(fbatch, folio)) { |
| 2257 | nr = folio_nr_pages(folio); |
| 2258 | |
| 2259 | if (folio_test_hugetlb(folio)) |
| 2260 | nr = 1; |
| 2261 | *start = folio->index + nr; |
| 2262 | goto out; |
Matthew Wilcox (Oracle) | 6b24ca4 | 2020-06-27 22:19:08 -0400 | [diff] [blame] | 2263 | } |
Matthew Wilcox | 3ece58a | 2018-05-16 18:00:33 -0400 | [diff] [blame] | 2264 | continue; |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2265 | put_folio: |
Matthew Wilcox (Oracle) | e1c3772 | 2021-03-06 16:38:38 -0500 | [diff] [blame] | 2266 | folio_put(folio); |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2267 | |
Matthew Wilcox | 3ece58a | 2018-05-16 18:00:33 -0400 | [diff] [blame] | 2268 | retry: |
| 2269 | xas_reset(&xas); |
Jens Axboe | ebf4350 | 2006-04-27 08:46:01 +0200 | [diff] [blame] | 2270 | } |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2271 | |
| 2272 | update_start: |
| 2273 | nr = folio_batch_count(fbatch); |
| 2274 | |
| 2275 | if (nr) { |
| 2276 | folio = fbatch->folios[nr - 1]; |
| 2277 | if (folio_test_hugetlb(folio)) |
| 2278 | *start = folio->index + 1; |
| 2279 | else |
| 2280 | *start = folio->index + folio_nr_pages(folio); |
| 2281 | } |
| 2282 | out: |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2283 | rcu_read_unlock(); |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2284 | return folio_batch_count(fbatch); |
Jens Axboe | ebf4350 | 2006-04-27 08:46:01 +0200 | [diff] [blame] | 2285 | } |
Vishal Moola (Oracle) | 35b4714 | 2022-08-23 17:40:17 -0700 | [diff] [blame] | 2286 | EXPORT_SYMBOL(filemap_get_folios_contig); |
Jens Axboe | ebf4350 | 2006-04-27 08:46:01 +0200 | [diff] [blame] | 2287 | |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 2288 | /** |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2289 | * filemap_get_folios_tag - Get a batch of folios matching @tag |
| 2290 | * @mapping: The address_space to search |
| 2291 | * @start: The starting page index |
| 2292 | * @end: The final page index (inclusive) |
| 2293 | * @tag: The tag index |
| 2294 | * @fbatch: The batch to fill |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 2295 | * |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2296 | * Same as filemap_get_folios(), but only returning folios tagged with @tag. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 2297 | * |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2298 | * Return: The number of folios found. |
| 2299 | * Also update @start to index the next folio for traversal. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2300 | */ |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2301 | unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start, |
| 2302 | pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2303 | { |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2304 | XA_STATE(xas, &mapping->i_pages, *start); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2305 | struct folio *folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2306 | |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2307 | rcu_read_lock(); |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2308 | while ((folio = find_get_entry(&xas, end, tag)) != NULL) { |
Matthew Wilcox | a690697 | 2018-05-16 18:12:54 -0400 | [diff] [blame] | 2309 | /* |
| 2310 | * Shadow entries should never be tagged, but this iteration |
| 2311 | * is lockless so there is a window for page reclaim to evict |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2312 | * a page we saw tagged. Skip over it. |
Matthew Wilcox | a690697 | 2018-05-16 18:12:54 -0400 | [diff] [blame] | 2313 | */ |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2314 | if (xa_is_value(folio)) |
Johannes Weiner | 139b6a6 | 2014-05-06 12:50:05 -0700 | [diff] [blame] | 2315 | continue; |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2316 | if (!folio_batch_add(fbatch, folio)) { |
| 2317 | unsigned long nr = folio_nr_pages(folio); |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2318 | |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2319 | if (folio_test_hugetlb(folio)) |
| 2320 | nr = 1; |
| 2321 | *start = folio->index + nr; |
Jan Kara | 72b045a | 2017-11-15 17:34:33 -0800 | [diff] [blame] | 2322 | goto out; |
| 2323 | } |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2324 | } |
Jan Kara | 72b045a | 2017-11-15 17:34:33 -0800 | [diff] [blame] | 2325 | /* |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2326 | * We come here when there is no page beyond @end. We take care to not |
| 2327 | * overflow the index @start as it confuses some of the callers. This |
| 2328 | * breaks the iteration when there is a page at index -1 but that is |
| 2329 | * already broke anyway. |
Jan Kara | 72b045a | 2017-11-15 17:34:33 -0800 | [diff] [blame] | 2330 | */ |
| 2331 | if (end == (pgoff_t)-1) |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2332 | *start = (pgoff_t)-1; |
Jan Kara | 72b045a | 2017-11-15 17:34:33 -0800 | [diff] [blame] | 2333 | else |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2334 | *start = end + 1; |
Jan Kara | 72b045a | 2017-11-15 17:34:33 -0800 | [diff] [blame] | 2335 | out: |
Nick Piggin | a60637c | 2008-07-25 19:45:31 -0700 | [diff] [blame] | 2336 | rcu_read_unlock(); |
| 2337 | |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2338 | return folio_batch_count(fbatch); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2339 | } |
Vishal Moola (Oracle) | 247f9e1 | 2023-01-04 13:14:27 -0800 | [diff] [blame] | 2340 | EXPORT_SYMBOL(filemap_get_folios_tag); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2341 | |
Wu Fengguang | 76d42bd | 2006-06-25 05:48:43 -0700 | [diff] [blame] | 2342 | /* |
| 2343 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail |
| 2344 | * a _large_ part of the i/o request. Imagine the worst scenario: |
| 2345 | * |
| 2346 | * ---R__________________________________________B__________ |
| 2347 | * ^ reading here ^ bad block(assume 4k) |
| 2348 | * |
| 2349 | * read(R) => miss => readahead(R...B) => media error => frustrating retries |
| 2350 | * => failing the whole request => read(R) => read(R+1) => |
| 2351 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => |
| 2352 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => |
| 2353 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... |
| 2354 | * |
| 2355 | * It is going insane. Fix it by quickly scaling down the readahead size. |
| 2356 | */ |
Souptick Joarder | 0f8e2db | 2020-04-01 21:04:50 -0700 | [diff] [blame] | 2357 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
Wu Fengguang | 76d42bd | 2006-06-25 05:48:43 -0700 | [diff] [blame] | 2358 | { |
Wu Fengguang | 76d42bd | 2006-06-25 05:48:43 -0700 | [diff] [blame] | 2359 | ra->ra_pages /= 4; |
Wu Fengguang | 76d42bd | 2006-06-25 05:48:43 -0700 | [diff] [blame] | 2360 | } |
| 2361 | |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2362 | /* |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2363 | * filemap_get_read_batch - Get a batch of folios for read |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2364 | * |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2365 | * Get a batch of folios which represent a contiguous range of bytes in |
| 2366 | * the file. No exceptional entries will be returned. If @index is in |
| 2367 | * the middle of a folio, the entire folio will be returned. The last |
| 2368 | * folio in the batch may have the readahead flag set or the uptodate flag |
| 2369 | * clear so that the caller can take the appropriate action. |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2370 | */ |
| 2371 | static void filemap_get_read_batch(struct address_space *mapping, |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2372 | pgoff_t index, pgoff_t max, struct folio_batch *fbatch) |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2373 | { |
| 2374 | XA_STATE(xas, &mapping->i_pages, index); |
Matthew Wilcox (Oracle) | bdb7293 | 2021-03-05 10:29:41 -0500 | [diff] [blame] | 2375 | struct folio *folio; |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2376 | |
| 2377 | rcu_read_lock(); |
Matthew Wilcox (Oracle) | bdb7293 | 2021-03-05 10:29:41 -0500 | [diff] [blame] | 2378 | for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) { |
| 2379 | if (xas_retry(&xas, folio)) |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2380 | continue; |
Matthew Wilcox (Oracle) | bdb7293 | 2021-03-05 10:29:41 -0500 | [diff] [blame] | 2381 | if (xas.xa_index > max || xa_is_value(folio)) |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2382 | break; |
Matthew Wilcox (Oracle) | cb995f4 | 2022-06-17 20:00:17 -0400 | [diff] [blame] | 2383 | if (xa_is_sibling(folio)) |
| 2384 | break; |
Matthew Wilcox (Oracle) | bdb7293 | 2021-03-05 10:29:41 -0500 | [diff] [blame] | 2385 | if (!folio_try_get_rcu(folio)) |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2386 | goto retry; |
| 2387 | |
Matthew Wilcox (Oracle) | bdb7293 | 2021-03-05 10:29:41 -0500 | [diff] [blame] | 2388 | if (unlikely(folio != xas_reload(&xas))) |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2389 | goto put_folio; |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2390 | |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2391 | if (!folio_batch_add(fbatch, folio)) |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2392 | break; |
Matthew Wilcox (Oracle) | bdb7293 | 2021-03-05 10:29:41 -0500 | [diff] [blame] | 2393 | if (!folio_test_uptodate(folio)) |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2394 | break; |
Matthew Wilcox (Oracle) | bdb7293 | 2021-03-05 10:29:41 -0500 | [diff] [blame] | 2395 | if (folio_test_readahead(folio)) |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2396 | break; |
Matthew Wilcox (Oracle) | 6b24ca4 | 2020-06-27 22:19:08 -0400 | [diff] [blame] | 2397 | xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1); |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2398 | continue; |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2399 | put_folio: |
Matthew Wilcox (Oracle) | bdb7293 | 2021-03-05 10:29:41 -0500 | [diff] [blame] | 2400 | folio_put(folio); |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2401 | retry: |
| 2402 | xas_reset(&xas); |
| 2403 | } |
| 2404 | rcu_read_unlock(); |
| 2405 | } |
| 2406 | |
Matthew Wilcox (Oracle) | 290e1a3 | 2022-05-12 17:37:01 -0400 | [diff] [blame] | 2407 | static int filemap_read_folio(struct file *file, filler_t filler, |
Matthew Wilcox (Oracle) | 9d427b4 | 2021-03-10 10:19:30 -0500 | [diff] [blame] | 2408 | struct folio *folio) |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2409 | { |
Christoph Hellwig | 1760424 | 2022-09-15 10:41:56 +0100 | [diff] [blame] | 2410 | bool workingset = folio_test_workingset(folio); |
| 2411 | unsigned long pflags; |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2412 | int error; |
| 2413 | |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2414 | /* |
Matthew Wilcox (Oracle) | 6843030 | 2021-02-24 12:02:15 -0800 | [diff] [blame] | 2415 | * A previous I/O error may have been due to temporary failures, |
Matthew Wilcox (Oracle) | 7e0a126 | 2022-04-29 11:53:28 -0400 | [diff] [blame] | 2416 | * eg. multipath errors. PG_error will be set again if read_folio |
Matthew Wilcox (Oracle) | 6843030 | 2021-02-24 12:02:15 -0800 | [diff] [blame] | 2417 | * fails. |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2418 | */ |
Matthew Wilcox (Oracle) | 9d427b4 | 2021-03-10 10:19:30 -0500 | [diff] [blame] | 2419 | folio_clear_error(folio); |
Christoph Hellwig | 1760424 | 2022-09-15 10:41:56 +0100 | [diff] [blame] | 2420 | |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2421 | /* Start the actual read. The read will unlock the page. */ |
Christoph Hellwig | 1760424 | 2022-09-15 10:41:56 +0100 | [diff] [blame] | 2422 | if (unlikely(workingset)) |
| 2423 | psi_memstall_enter(&pflags); |
Matthew Wilcox (Oracle) | 290e1a3 | 2022-05-12 17:37:01 -0400 | [diff] [blame] | 2424 | error = filler(file, folio); |
Christoph Hellwig | 1760424 | 2022-09-15 10:41:56 +0100 | [diff] [blame] | 2425 | if (unlikely(workingset)) |
| 2426 | psi_memstall_leave(&pflags); |
Matthew Wilcox (Oracle) | 6843030 | 2021-02-24 12:02:15 -0800 | [diff] [blame] | 2427 | if (error) |
| 2428 | return error; |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2429 | |
Matthew Wilcox (Oracle) | 9d427b4 | 2021-03-10 10:19:30 -0500 | [diff] [blame] | 2430 | error = folio_wait_locked_killable(folio); |
Matthew Wilcox (Oracle) | 6843030 | 2021-02-24 12:02:15 -0800 | [diff] [blame] | 2431 | if (error) |
| 2432 | return error; |
Matthew Wilcox (Oracle) | 9d427b4 | 2021-03-10 10:19:30 -0500 | [diff] [blame] | 2433 | if (folio_test_uptodate(folio)) |
Matthew Wilcox (Oracle) | aa1ec2f | 2021-02-24 12:02:38 -0800 | [diff] [blame] | 2434 | return 0; |
Matthew Wilcox (Oracle) | 290e1a3 | 2022-05-12 17:37:01 -0400 | [diff] [blame] | 2435 | if (file) |
| 2436 | shrink_readahead_size_eio(&file->f_ra); |
Matthew Wilcox (Oracle) | aa1ec2f | 2021-02-24 12:02:38 -0800 | [diff] [blame] | 2437 | return -EIO; |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2438 | } |
| 2439 | |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2440 | static bool filemap_range_uptodate(struct address_space *mapping, |
David Howells | dd5b9d0 | 2023-02-08 18:18:17 +0000 | [diff] [blame] | 2441 | loff_t pos, size_t count, struct folio *folio, |
| 2442 | bool need_uptodate) |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2443 | { |
Matthew Wilcox (Oracle) | 2fa4eeb8 | 2021-03-10 11:04:19 -0500 | [diff] [blame] | 2444 | if (folio_test_uptodate(folio)) |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2445 | return true; |
| 2446 | /* pipes can't handle partially uptodate pages */ |
David Howells | dd5b9d0 | 2023-02-08 18:18:17 +0000 | [diff] [blame] | 2447 | if (need_uptodate) |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2448 | return false; |
| 2449 | if (!mapping->a_ops->is_partially_uptodate) |
| 2450 | return false; |
Matthew Wilcox (Oracle) | 2fa4eeb8 | 2021-03-10 11:04:19 -0500 | [diff] [blame] | 2451 | if (mapping->host->i_blkbits >= folio_shift(folio)) |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2452 | return false; |
| 2453 | |
Matthew Wilcox (Oracle) | 2fa4eeb8 | 2021-03-10 11:04:19 -0500 | [diff] [blame] | 2454 | if (folio_pos(folio) > pos) { |
| 2455 | count -= folio_pos(folio) - pos; |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2456 | pos = 0; |
| 2457 | } else { |
Matthew Wilcox (Oracle) | 2fa4eeb8 | 2021-03-10 11:04:19 -0500 | [diff] [blame] | 2458 | pos -= folio_pos(folio); |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2459 | } |
| 2460 | |
Matthew Wilcox (Oracle) | 2e7e80f | 2022-02-09 20:21:27 +0000 | [diff] [blame] | 2461 | return mapping->a_ops->is_partially_uptodate(folio, pos, count); |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2462 | } |
| 2463 | |
Matthew Wilcox (Oracle) | 4612aee | 2021-02-24 12:02:22 -0800 | [diff] [blame] | 2464 | static int filemap_update_page(struct kiocb *iocb, |
David Howells | dd5b9d0 | 2023-02-08 18:18:17 +0000 | [diff] [blame] | 2465 | struct address_space *mapping, size_t count, |
| 2466 | struct folio *folio, bool need_uptodate) |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2467 | { |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2468 | int error; |
| 2469 | |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2470 | if (iocb->ki_flags & IOCB_NOWAIT) { |
| 2471 | if (!filemap_invalidate_trylock_shared(mapping)) |
Matthew Wilcox (Oracle) | 87d1d7b | 2021-02-24 12:02:25 -0800 | [diff] [blame] | 2472 | return -EAGAIN; |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2473 | } else { |
| 2474 | filemap_invalidate_lock_shared(mapping); |
| 2475 | } |
| 2476 | |
Matthew Wilcox (Oracle) | ffdc8da | 2020-12-30 17:58:40 -0500 | [diff] [blame] | 2477 | if (!folio_trylock(folio)) { |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2478 | error = -EAGAIN; |
| 2479 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) |
| 2480 | goto unlock_mapping; |
Matthew Wilcox (Oracle) | 87d1d7b | 2021-02-24 12:02:25 -0800 | [diff] [blame] | 2481 | if (!(iocb->ki_flags & IOCB_WAITQ)) { |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2482 | filemap_invalidate_unlock_shared(mapping); |
Matthew Wilcox (Oracle) | 9f2b04a | 2021-08-16 23:36:31 -0400 | [diff] [blame] | 2483 | /* |
| 2484 | * This is where we usually end up waiting for a |
| 2485 | * previously submitted readahead to finish. |
| 2486 | */ |
| 2487 | folio_put_wait_locked(folio, TASK_KILLABLE); |
Matthew Wilcox (Oracle) | 4612aee | 2021-02-24 12:02:22 -0800 | [diff] [blame] | 2488 | return AOP_TRUNCATED_PAGE; |
Matthew Wilcox (Oracle) | bd8a1f3 | 2021-02-24 12:02:05 -0800 | [diff] [blame] | 2489 | } |
Matthew Wilcox (Oracle) | ffdc8da | 2020-12-30 17:58:40 -0500 | [diff] [blame] | 2490 | error = __folio_lock_async(folio, iocb->ki_waitq); |
Matthew Wilcox (Oracle) | 87d1d7b | 2021-02-24 12:02:25 -0800 | [diff] [blame] | 2491 | if (error) |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2492 | goto unlock_mapping; |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2493 | } |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2494 | |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2495 | error = AOP_TRUNCATED_PAGE; |
Matthew Wilcox (Oracle) | ffdc8da | 2020-12-30 17:58:40 -0500 | [diff] [blame] | 2496 | if (!folio->mapping) |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2497 | goto unlock; |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2498 | |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2499 | error = 0; |
David Howells | dd5b9d0 | 2023-02-08 18:18:17 +0000 | [diff] [blame] | 2500 | if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio, |
| 2501 | need_uptodate)) |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2502 | goto unlock; |
| 2503 | |
| 2504 | error = -EAGAIN; |
| 2505 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ)) |
| 2506 | goto unlock; |
| 2507 | |
Matthew Wilcox (Oracle) | 290e1a3 | 2022-05-12 17:37:01 -0400 | [diff] [blame] | 2508 | error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio, |
| 2509 | folio); |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2510 | goto unlock_mapping; |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2511 | unlock: |
Matthew Wilcox (Oracle) | ffdc8da | 2020-12-30 17:58:40 -0500 | [diff] [blame] | 2512 | folio_unlock(folio); |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2513 | unlock_mapping: |
| 2514 | filemap_invalidate_unlock_shared(mapping); |
| 2515 | if (error == AOP_TRUNCATED_PAGE) |
Matthew Wilcox (Oracle) | ffdc8da | 2020-12-30 17:58:40 -0500 | [diff] [blame] | 2516 | folio_put(folio); |
Matthew Wilcox (Oracle) | fce70da | 2021-02-24 12:02:28 -0800 | [diff] [blame] | 2517 | return error; |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2518 | } |
| 2519 | |
Matthew Wilcox (Oracle) | a5d4ad0 | 2021-03-10 10:34:00 -0500 | [diff] [blame] | 2520 | static int filemap_create_folio(struct file *file, |
Matthew Wilcox (Oracle) | f253e18 | 2021-02-24 12:02:18 -0800 | [diff] [blame] | 2521 | struct address_space *mapping, pgoff_t index, |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2522 | struct folio_batch *fbatch) |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2523 | { |
Matthew Wilcox (Oracle) | a5d4ad0 | 2021-03-10 10:34:00 -0500 | [diff] [blame] | 2524 | struct folio *folio; |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2525 | int error; |
| 2526 | |
Matthew Wilcox (Oracle) | a5d4ad0 | 2021-03-10 10:34:00 -0500 | [diff] [blame] | 2527 | folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0); |
| 2528 | if (!folio) |
Matthew Wilcox (Oracle) | f253e18 | 2021-02-24 12:02:18 -0800 | [diff] [blame] | 2529 | return -ENOMEM; |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2530 | |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2531 | /* |
Matthew Wilcox (Oracle) | a5d4ad0 | 2021-03-10 10:34:00 -0500 | [diff] [blame] | 2532 | * Protect against truncate / hole punch. Grabbing invalidate_lock |
| 2533 | * here assures we cannot instantiate and bring uptodate new |
| 2534 | * pagecache folios after evicting page cache during truncate |
| 2535 | * and before actually freeing blocks. Note that we could |
| 2536 | * release invalidate_lock after inserting the folio into |
| 2537 | * the page cache as the locked folio would then be enough to |
| 2538 | * synchronize with hole punching. But there are code paths |
| 2539 | * such as filemap_update_page() filling in partially uptodate |
Matthew Wilcox (Oracle) | 704528d | 2022-03-23 21:29:04 -0400 | [diff] [blame] | 2540 | * pages or ->readahead() that need to hold invalidate_lock |
Matthew Wilcox (Oracle) | a5d4ad0 | 2021-03-10 10:34:00 -0500 | [diff] [blame] | 2541 | * while mapping blocks for IO so let's hold the lock here as |
| 2542 | * well to keep locking rules simple. |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2543 | */ |
| 2544 | filemap_invalidate_lock_shared(mapping); |
Matthew Wilcox (Oracle) | a5d4ad0 | 2021-03-10 10:34:00 -0500 | [diff] [blame] | 2545 | error = filemap_add_folio(mapping, folio, index, |
Matthew Wilcox (Oracle) | f253e18 | 2021-02-24 12:02:18 -0800 | [diff] [blame] | 2546 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
| 2547 | if (error == -EEXIST) |
| 2548 | error = AOP_TRUNCATED_PAGE; |
| 2549 | if (error) |
| 2550 | goto error; |
| 2551 | |
Matthew Wilcox (Oracle) | 290e1a3 | 2022-05-12 17:37:01 -0400 | [diff] [blame] | 2552 | error = filemap_read_folio(file, mapping->a_ops->read_folio, folio); |
Matthew Wilcox (Oracle) | f253e18 | 2021-02-24 12:02:18 -0800 | [diff] [blame] | 2553 | if (error) |
| 2554 | goto error; |
| 2555 | |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2556 | filemap_invalidate_unlock_shared(mapping); |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2557 | folio_batch_add(fbatch, folio); |
Matthew Wilcox (Oracle) | f253e18 | 2021-02-24 12:02:18 -0800 | [diff] [blame] | 2558 | return 0; |
| 2559 | error: |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 2560 | filemap_invalidate_unlock_shared(mapping); |
Matthew Wilcox (Oracle) | a5d4ad0 | 2021-03-10 10:34:00 -0500 | [diff] [blame] | 2561 | folio_put(folio); |
Matthew Wilcox (Oracle) | f253e18 | 2021-02-24 12:02:18 -0800 | [diff] [blame] | 2562 | return error; |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2563 | } |
| 2564 | |
Matthew Wilcox (Oracle) | 5963fe0 | 2021-02-24 12:02:32 -0800 | [diff] [blame] | 2565 | static int filemap_readahead(struct kiocb *iocb, struct file *file, |
Matthew Wilcox (Oracle) | 65bca53 | 2021-03-10 14:01:22 -0500 | [diff] [blame] | 2566 | struct address_space *mapping, struct folio *folio, |
Matthew Wilcox (Oracle) | 5963fe0 | 2021-02-24 12:02:32 -0800 | [diff] [blame] | 2567 | pgoff_t last_index) |
| 2568 | { |
Matthew Wilcox (Oracle) | 65bca53 | 2021-03-10 14:01:22 -0500 | [diff] [blame] | 2569 | DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index); |
| 2570 | |
Matthew Wilcox (Oracle) | 5963fe0 | 2021-02-24 12:02:32 -0800 | [diff] [blame] | 2571 | if (iocb->ki_flags & IOCB_NOIO) |
| 2572 | return -EAGAIN; |
Matthew Wilcox (Oracle) | 65bca53 | 2021-03-10 14:01:22 -0500 | [diff] [blame] | 2573 | page_cache_async_ra(&ractl, folio, last_index - folio->index); |
Matthew Wilcox (Oracle) | 5963fe0 | 2021-02-24 12:02:32 -0800 | [diff] [blame] | 2574 | return 0; |
| 2575 | } |
| 2576 | |
David Howells | dd5b9d0 | 2023-02-08 18:18:17 +0000 | [diff] [blame] | 2577 | static int filemap_get_pages(struct kiocb *iocb, size_t count, |
| 2578 | struct folio_batch *fbatch, bool need_uptodate) |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2579 | { |
| 2580 | struct file *filp = iocb->ki_filp; |
| 2581 | struct address_space *mapping = filp->f_mapping; |
| 2582 | struct file_ra_state *ra = &filp->f_ra; |
| 2583 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2584 | pgoff_t last_index; |
Matthew Wilcox (Oracle) | 65bca53 | 2021-03-10 14:01:22 -0500 | [diff] [blame] | 2585 | struct folio *folio; |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2586 | int err = 0; |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2587 | |
Qian Yingjin | 5956592 | 2023-02-08 10:24:00 +0800 | [diff] [blame] | 2588 | /* "last_index" is the index of the page beyond the end of the read */ |
David Howells | dd5b9d0 | 2023-02-08 18:18:17 +0000 | [diff] [blame] | 2589 | last_index = DIV_ROUND_UP(iocb->ki_pos + count, PAGE_SIZE); |
Matthew Wilcox (Oracle) | 2642fca | 2021-02-24 12:02:35 -0800 | [diff] [blame] | 2590 | retry: |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2591 | if (fatal_signal_pending(current)) |
| 2592 | return -EINTR; |
| 2593 | |
Qian Yingjin | 5956592 | 2023-02-08 10:24:00 +0800 | [diff] [blame] | 2594 | filemap_get_read_batch(mapping, index, last_index - 1, fbatch); |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2595 | if (!folio_batch_count(fbatch)) { |
Matthew Wilcox (Oracle) | 2642fca | 2021-02-24 12:02:35 -0800 | [diff] [blame] | 2596 | if (iocb->ki_flags & IOCB_NOIO) |
| 2597 | return -EAGAIN; |
| 2598 | page_cache_sync_readahead(mapping, ra, filp, index, |
| 2599 | last_index - index); |
Qian Yingjin | 5956592 | 2023-02-08 10:24:00 +0800 | [diff] [blame] | 2600 | filemap_get_read_batch(mapping, index, last_index - 1, fbatch); |
Matthew Wilcox (Oracle) | 2642fca | 2021-02-24 12:02:35 -0800 | [diff] [blame] | 2601 | } |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2602 | if (!folio_batch_count(fbatch)) { |
Matthew Wilcox (Oracle) | f253e18 | 2021-02-24 12:02:18 -0800 | [diff] [blame] | 2603 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ)) |
| 2604 | return -EAGAIN; |
Matthew Wilcox (Oracle) | a5d4ad0 | 2021-03-10 10:34:00 -0500 | [diff] [blame] | 2605 | err = filemap_create_folio(filp, mapping, |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2606 | iocb->ki_pos >> PAGE_SHIFT, fbatch); |
Matthew Wilcox (Oracle) | f253e18 | 2021-02-24 12:02:18 -0800 | [diff] [blame] | 2607 | if (err == AOP_TRUNCATED_PAGE) |
Matthew Wilcox (Oracle) | 2642fca | 2021-02-24 12:02:35 -0800 | [diff] [blame] | 2608 | goto retry; |
Matthew Wilcox (Oracle) | f253e18 | 2021-02-24 12:02:18 -0800 | [diff] [blame] | 2609 | return err; |
| 2610 | } |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2611 | |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2612 | folio = fbatch->folios[folio_batch_count(fbatch) - 1]; |
Matthew Wilcox (Oracle) | 65bca53 | 2021-03-10 14:01:22 -0500 | [diff] [blame] | 2613 | if (folio_test_readahead(folio)) { |
| 2614 | err = filemap_readahead(iocb, filp, mapping, folio, last_index); |
Matthew Wilcox (Oracle) | 2642fca | 2021-02-24 12:02:35 -0800 | [diff] [blame] | 2615 | if (err) |
| 2616 | goto err; |
| 2617 | } |
Matthew Wilcox (Oracle) | 65bca53 | 2021-03-10 14:01:22 -0500 | [diff] [blame] | 2618 | if (!folio_test_uptodate(folio)) { |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2619 | if ((iocb->ki_flags & IOCB_WAITQ) && |
| 2620 | folio_batch_count(fbatch) > 1) |
Matthew Wilcox (Oracle) | 2642fca | 2021-02-24 12:02:35 -0800 | [diff] [blame] | 2621 | iocb->ki_flags |= IOCB_NOWAIT; |
David Howells | dd5b9d0 | 2023-02-08 18:18:17 +0000 | [diff] [blame] | 2622 | err = filemap_update_page(iocb, mapping, count, folio, |
| 2623 | need_uptodate); |
Matthew Wilcox (Oracle) | 2642fca | 2021-02-24 12:02:35 -0800 | [diff] [blame] | 2624 | if (err) |
| 2625 | goto err; |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2626 | } |
| 2627 | |
Matthew Wilcox (Oracle) | 2642fca | 2021-02-24 12:02:35 -0800 | [diff] [blame] | 2628 | return 0; |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2629 | err: |
Matthew Wilcox (Oracle) | 2642fca | 2021-02-24 12:02:35 -0800 | [diff] [blame] | 2630 | if (err < 0) |
Matthew Wilcox (Oracle) | 65bca53 | 2021-03-10 14:01:22 -0500 | [diff] [blame] | 2631 | folio_put(folio); |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2632 | if (likely(--fbatch->nr)) |
Matthew Wilcox (Oracle) | ff993ba | 2021-02-24 12:01:55 -0800 | [diff] [blame] | 2633 | return 0; |
Matthew Wilcox (Oracle) | 4612aee | 2021-02-24 12:02:22 -0800 | [diff] [blame] | 2634 | if (err == AOP_TRUNCATED_PAGE) |
Matthew Wilcox (Oracle) | 2642fca | 2021-02-24 12:02:35 -0800 | [diff] [blame] | 2635 | goto retry; |
| 2636 | return err; |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2637 | } |
| 2638 | |
Matthew Wilcox (Oracle) | 5ccc944 | 2022-06-10 14:44:41 -0400 | [diff] [blame] | 2639 | static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio) |
| 2640 | { |
| 2641 | unsigned int shift = folio_shift(folio); |
| 2642 | |
| 2643 | return (pos1 >> shift == pos2 >> shift); |
| 2644 | } |
| 2645 | |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 2646 | /** |
Christoph Hellwig | 87fa0f3 | 2021-02-24 12:02:42 -0800 | [diff] [blame] | 2647 | * filemap_read - Read data from the page cache. |
| 2648 | * @iocb: The iocb to read. |
| 2649 | * @iter: Destination for the data. |
| 2650 | * @already_read: Number of bytes already read by the caller. |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 2651 | * |
Christoph Hellwig | 87fa0f3 | 2021-02-24 12:02:42 -0800 | [diff] [blame] | 2652 | * Copies data from the page cache. If the data is not currently present, |
Matthew Wilcox (Oracle) | 7e0a126 | 2022-04-29 11:53:28 -0400 | [diff] [blame] | 2653 | * uses the readahead and read_folio address_space operations to fetch it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2654 | * |
Christoph Hellwig | 87fa0f3 | 2021-02-24 12:02:42 -0800 | [diff] [blame] | 2655 | * Return: Total number of bytes copied, including those already read by |
| 2656 | * the caller. If an error happens before any bytes are copied, returns |
| 2657 | * a negative error number. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2658 | */ |
Christoph Hellwig | 87fa0f3 | 2021-02-24 12:02:42 -0800 | [diff] [blame] | 2659 | ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter, |
| 2660 | ssize_t already_read) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2661 | { |
Christoph Hellwig | 47c27bc | 2017-08-29 16:13:18 +0200 | [diff] [blame] | 2662 | struct file *filp = iocb->ki_filp; |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2663 | struct file_ra_state *ra = &filp->f_ra; |
Christoph Hellwig | 36e7891 | 2008-02-08 04:21:24 -0800 | [diff] [blame] | 2664 | struct address_space *mapping = filp->f_mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2665 | struct inode *inode = mapping->host; |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2666 | struct folio_batch fbatch; |
Matthew Wilcox (Oracle) | ff993ba | 2021-02-24 12:01:55 -0800 | [diff] [blame] | 2667 | int i, error = 0; |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2668 | bool writably_mapped; |
| 2669 | loff_t isize, end_offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2670 | |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2671 | if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes)) |
Linus Torvalds | d05c5f7 | 2016-12-14 12:45:25 -0800 | [diff] [blame] | 2672 | return 0; |
Kent Overstreet | 3644e2d | 2020-12-18 04:07:11 -0500 | [diff] [blame] | 2673 | if (unlikely(!iov_iter_count(iter))) |
| 2674 | return 0; |
| 2675 | |
Wei Fang | c2a9737 | 2016-10-07 17:01:52 -0700 | [diff] [blame] | 2676 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2677 | folio_batch_init(&fbatch); |
Wei Fang | c2a9737 | 2016-10-07 17:01:52 -0700 | [diff] [blame] | 2678 | |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2679 | do { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2680 | cond_resched(); |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2681 | |
| 2682 | /* |
| 2683 | * If we've already successfully copied some data, then we |
| 2684 | * can no longer safely return -EIOCBQUEUED. Hence mark |
| 2685 | * an async read NOWAIT at that point. |
| 2686 | */ |
Christoph Hellwig | 87fa0f3 | 2021-02-24 12:02:42 -0800 | [diff] [blame] | 2687 | if ((iocb->ki_flags & IOCB_WAITQ) && already_read) |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2688 | iocb->ki_flags |= IOCB_NOWAIT; |
| 2689 | |
David Howells | 8c8387e | 2021-11-05 13:36:49 -0700 | [diff] [blame] | 2690 | if (unlikely(iocb->ki_pos >= i_size_read(inode))) |
| 2691 | break; |
| 2692 | |
David Howells | dd5b9d0 | 2023-02-08 18:18:17 +0000 | [diff] [blame] | 2693 | error = filemap_get_pages(iocb, iter->count, &fbatch, |
| 2694 | iov_iter_is_pipe(iter)); |
Matthew Wilcox (Oracle) | ff993ba | 2021-02-24 12:01:55 -0800 | [diff] [blame] | 2695 | if (error < 0) |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2696 | break; |
Michal Hocko | 5abf186 | 2017-02-03 13:13:29 -0800 | [diff] [blame] | 2697 | |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2698 | /* |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2699 | * i_size must be checked after we know the pages are Uptodate. |
| 2700 | * |
| 2701 | * Checking i_size after the check allows us to calculate |
| 2702 | * the correct value for "nr", which means the zero-filled |
| 2703 | * part of the page is not copied back to userspace (unless |
| 2704 | * another truncate extends the file - this is desired though). |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2705 | */ |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2706 | isize = i_size_read(inode); |
| 2707 | if (unlikely(iocb->ki_pos >= isize)) |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2708 | goto put_folios; |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2709 | end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count); |
| 2710 | |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2711 | /* |
| 2712 | * Once we start copying data, we don't want to be touching any |
| 2713 | * cachelines that might be contended: |
| 2714 | */ |
| 2715 | writably_mapped = mapping_writably_mapped(mapping); |
| 2716 | |
| 2717 | /* |
Matthew Wilcox (Oracle) | 5ccc944 | 2022-06-10 14:44:41 -0400 | [diff] [blame] | 2718 | * When a read accesses the same folio several times, only |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2719 | * mark it as accessed the first time. |
| 2720 | */ |
Matthew Wilcox (Oracle) | 5ccc944 | 2022-06-10 14:44:41 -0400 | [diff] [blame] | 2721 | if (!pos_same_folio(iocb->ki_pos, ra->prev_pos - 1, |
| 2722 | fbatch.folios[0])) |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2723 | folio_mark_accessed(fbatch.folios[0]); |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2724 | |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2725 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
| 2726 | struct folio *folio = fbatch.folios[i]; |
Matthew Wilcox (Oracle) | d996fc7 | 2021-10-31 22:22:19 -0400 | [diff] [blame] | 2727 | size_t fsize = folio_size(folio); |
| 2728 | size_t offset = iocb->ki_pos & (fsize - 1); |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2729 | size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos, |
Matthew Wilcox (Oracle) | d996fc7 | 2021-10-31 22:22:19 -0400 | [diff] [blame] | 2730 | fsize - offset); |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2731 | size_t copied; |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2732 | |
Matthew Wilcox (Oracle) | d996fc7 | 2021-10-31 22:22:19 -0400 | [diff] [blame] | 2733 | if (end_offset < folio_pos(folio)) |
Matthew Wilcox (Oracle) | cbd59c4 | 2021-02-24 12:01:59 -0800 | [diff] [blame] | 2734 | break; |
| 2735 | if (i > 0) |
Matthew Wilcox (Oracle) | d996fc7 | 2021-10-31 22:22:19 -0400 | [diff] [blame] | 2736 | folio_mark_accessed(folio); |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2737 | /* |
Matthew Wilcox (Oracle) | d996fc7 | 2021-10-31 22:22:19 -0400 | [diff] [blame] | 2738 | * If users can be writing to this folio using arbitrary |
| 2739 | * virtual addresses, take care of potential aliasing |
| 2740 | * before reading the folio on the kernel side. |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2741 | */ |
Matthew Wilcox (Oracle) | d996fc7 | 2021-10-31 22:22:19 -0400 | [diff] [blame] | 2742 | if (writably_mapped) |
| 2743 | flush_dcache_folio(folio); |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2744 | |
Matthew Wilcox (Oracle) | d996fc7 | 2021-10-31 22:22:19 -0400 | [diff] [blame] | 2745 | copied = copy_folio_to_iter(folio, offset, bytes, iter); |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2746 | |
Christoph Hellwig | 87fa0f3 | 2021-02-24 12:02:42 -0800 | [diff] [blame] | 2747 | already_read += copied; |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2748 | iocb->ki_pos += copied; |
| 2749 | ra->prev_pos = iocb->ki_pos; |
| 2750 | |
| 2751 | if (copied < bytes) { |
| 2752 | error = -EFAULT; |
| 2753 | break; |
Kent Overstreet | 723ef24 | 2020-12-14 19:04:52 -0800 | [diff] [blame] | 2754 | } |
Fengguang Wu | 3ea89ee | 2007-07-19 01:48:02 -0700 | [diff] [blame] | 2755 | } |
Matthew Wilcox (Oracle) | 25d6a23 | 2021-12-06 15:25:33 -0500 | [diff] [blame] | 2756 | put_folios: |
| 2757 | for (i = 0; i < folio_batch_count(&fbatch); i++) |
| 2758 | folio_put(fbatch.folios[i]); |
| 2759 | folio_batch_init(&fbatch); |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2760 | } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2761 | |
Krishna Kumar | 0c6aa26 | 2008-10-15 22:01:13 -0700 | [diff] [blame] | 2762 | file_accessed(filp); |
Kent Overstreet | 06c0444 | 2020-12-14 19:04:56 -0800 | [diff] [blame] | 2763 | |
Christoph Hellwig | 87fa0f3 | 2021-02-24 12:02:42 -0800 | [diff] [blame] | 2764 | return already_read ? already_read : error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2765 | } |
Christoph Hellwig | 87fa0f3 | 2021-02-24 12:02:42 -0800 | [diff] [blame] | 2766 | EXPORT_SYMBOL_GPL(filemap_read); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2767 | |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 2768 | /** |
Al Viro | 6abd232 | 2014-04-04 14:20:57 -0400 | [diff] [blame] | 2769 | * generic_file_read_iter - generic filesystem read routine |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 2770 | * @iocb: kernel I/O control block |
Al Viro | 6abd232 | 2014-04-04 14:20:57 -0400 | [diff] [blame] | 2771 | * @iter: destination for the data read |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 2772 | * |
Al Viro | 6abd232 | 2014-04-04 14:20:57 -0400 | [diff] [blame] | 2773 | * This is the "read_iter()" routine for all filesystems |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2774 | * that can use the page cache directly. |
Andreas Gruenbacher | 41da51b | 2019-11-21 23:25:07 +0000 | [diff] [blame] | 2775 | * |
| 2776 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall |
| 2777 | * be returned when no data can be read without waiting for I/O requests |
| 2778 | * to complete; it doesn't prevent readahead. |
| 2779 | * |
| 2780 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O |
| 2781 | * requests shall be made for the read or for readahead. When no data |
| 2782 | * can be read, -EAGAIN shall be returned. When readahead would be |
| 2783 | * triggered, a partial, possibly empty read shall be returned. |
| 2784 | * |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 2785 | * Return: |
| 2786 | * * number of bytes copied, even for partial reads |
Andreas Gruenbacher | 41da51b | 2019-11-21 23:25:07 +0000 | [diff] [blame] | 2787 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2788 | */ |
| 2789 | ssize_t |
Al Viro | ed978a8 | 2014-03-05 22:53:04 -0500 | [diff] [blame] | 2790 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2791 | { |
Nicolai Stange | e7080a4 | 2016-03-25 14:22:14 -0700 | [diff] [blame] | 2792 | size_t count = iov_iter_count(iter); |
Christoph Hellwig | 47c27bc | 2017-08-29 16:13:18 +0200 | [diff] [blame] | 2793 | ssize_t retval = 0; |
Nicolai Stange | e7080a4 | 2016-03-25 14:22:14 -0700 | [diff] [blame] | 2794 | |
| 2795 | if (!count) |
Christoph Hellwig | 826ea86 | 2021-02-24 12:02:45 -0800 | [diff] [blame] | 2796 | return 0; /* skip atime */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2797 | |
Al Viro | 2ba48ce | 2015-04-09 13:52:01 -0400 | [diff] [blame] | 2798 | if (iocb->ki_flags & IOCB_DIRECT) { |
Christoph Hellwig | 47c27bc | 2017-08-29 16:13:18 +0200 | [diff] [blame] | 2799 | struct file *file = iocb->ki_filp; |
Al Viro | ed978a8 | 2014-03-05 22:53:04 -0500 | [diff] [blame] | 2800 | struct address_space *mapping = file->f_mapping; |
| 2801 | struct inode *inode = mapping->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2802 | |
Goldwyn Rodrigues | 6be96d3 | 2017-06-20 07:05:44 -0500 | [diff] [blame] | 2803 | if (iocb->ki_flags & IOCB_NOWAIT) { |
Jens Axboe | 7a60d6d | 2021-04-29 22:55:21 -0700 | [diff] [blame] | 2804 | if (filemap_range_needs_writeback(mapping, iocb->ki_pos, |
| 2805 | iocb->ki_pos + count - 1)) |
Goldwyn Rodrigues | 6be96d3 | 2017-06-20 07:05:44 -0500 | [diff] [blame] | 2806 | return -EAGAIN; |
| 2807 | } else { |
| 2808 | retval = filemap_write_and_wait_range(mapping, |
| 2809 | iocb->ki_pos, |
| 2810 | iocb->ki_pos + count - 1); |
| 2811 | if (retval < 0) |
Christoph Hellwig | 826ea86 | 2021-02-24 12:02:45 -0800 | [diff] [blame] | 2812 | return retval; |
Goldwyn Rodrigues | 6be96d3 | 2017-06-20 07:05:44 -0500 | [diff] [blame] | 2813 | } |
Al Viro | ed978a8 | 2014-03-05 22:53:04 -0500 | [diff] [blame] | 2814 | |
Christoph Hellwig | 0d5b0cf | 2016-10-03 09:48:08 +1100 | [diff] [blame] | 2815 | file_accessed(file); |
| 2816 | |
Al Viro | 5ecda13 | 2017-04-13 14:13:36 -0400 | [diff] [blame] | 2817 | retval = mapping->a_ops->direct_IO(iocb, iter); |
Al Viro | c3a6902 | 2016-10-10 13:26:27 -0400 | [diff] [blame] | 2818 | if (retval >= 0) { |
Christoph Hellwig | c64fb5c | 2016-04-07 08:51:55 -0700 | [diff] [blame] | 2819 | iocb->ki_pos += retval; |
Al Viro | 5ecda13 | 2017-04-13 14:13:36 -0400 | [diff] [blame] | 2820 | count -= retval; |
Steven Whitehouse | 9fe55ee | 2014-01-24 14:42:22 +0000 | [diff] [blame] | 2821 | } |
Pavel Begunkov | ab2125d | 2021-02-24 12:01:45 -0800 | [diff] [blame] | 2822 | if (retval != -EIOCBQUEUED) |
| 2823 | iov_iter_revert(iter, count - iov_iter_count(iter)); |
Josef Bacik | 66f998f | 2010-05-23 11:00:54 -0400 | [diff] [blame] | 2824 | |
Steven Whitehouse | 9fe55ee | 2014-01-24 14:42:22 +0000 | [diff] [blame] | 2825 | /* |
| 2826 | * Btrfs can have a short DIO read if we encounter |
| 2827 | * compressed extents, so if there was an error, or if |
| 2828 | * we've already read everything we wanted to, or if |
| 2829 | * there was a short read because we hit EOF, go ahead |
| 2830 | * and return. Otherwise fallthrough to buffered io for |
Matthew Wilcox | fbbbad4 | 2015-02-16 15:58:53 -0800 | [diff] [blame] | 2831 | * the rest of the read. Buffered reads will not work for |
| 2832 | * DAX files, so don't bother trying. |
Steven Whitehouse | 9fe55ee | 2014-01-24 14:42:22 +0000 | [diff] [blame] | 2833 | */ |
Jens Axboe | 61d0017 | 2021-11-05 13:37:07 -0700 | [diff] [blame] | 2834 | if (retval < 0 || !count || IS_DAX(inode)) |
| 2835 | return retval; |
| 2836 | if (iocb->ki_pos >= i_size_read(inode)) |
Christoph Hellwig | 826ea86 | 2021-02-24 12:02:45 -0800 | [diff] [blame] | 2837 | return retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2838 | } |
| 2839 | |
Christoph Hellwig | 826ea86 | 2021-02-24 12:02:45 -0800 | [diff] [blame] | 2840 | return filemap_read(iocb, iter, retval); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2841 | } |
Al Viro | ed978a8 | 2014-03-05 22:53:04 -0500 | [diff] [blame] | 2842 | EXPORT_SYMBOL(generic_file_read_iter); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2843 | |
David Howells | 07073eb | 2023-02-14 15:01:42 +0000 | [diff] [blame] | 2844 | /* |
| 2845 | * Splice subpages from a folio into a pipe. |
| 2846 | */ |
| 2847 | size_t splice_folio_into_pipe(struct pipe_inode_info *pipe, |
| 2848 | struct folio *folio, loff_t fpos, size_t size) |
| 2849 | { |
| 2850 | struct page *page; |
| 2851 | size_t spliced = 0, offset = offset_in_folio(folio, fpos); |
| 2852 | |
| 2853 | page = folio_page(folio, offset / PAGE_SIZE); |
| 2854 | size = min(size, folio_size(folio) - offset); |
| 2855 | offset %= PAGE_SIZE; |
| 2856 | |
| 2857 | while (spliced < size && |
| 2858 | !pipe_full(pipe->head, pipe->tail, pipe->max_usage)) { |
| 2859 | struct pipe_buffer *buf = pipe_head_buf(pipe); |
| 2860 | size_t part = min_t(size_t, PAGE_SIZE - offset, size - spliced); |
| 2861 | |
| 2862 | *buf = (struct pipe_buffer) { |
| 2863 | .ops = &page_cache_pipe_buf_ops, |
| 2864 | .page = page, |
| 2865 | .offset = offset, |
| 2866 | .len = part, |
| 2867 | }; |
| 2868 | folio_get(folio); |
| 2869 | pipe->head++; |
| 2870 | page++; |
| 2871 | spliced += part; |
| 2872 | offset = 0; |
| 2873 | } |
| 2874 | |
| 2875 | return spliced; |
| 2876 | } |
| 2877 | |
| 2878 | /* |
| 2879 | * Splice folios from the pagecache of a buffered (ie. non-O_DIRECT) file into |
| 2880 | * a pipe. |
| 2881 | */ |
| 2882 | ssize_t filemap_splice_read(struct file *in, loff_t *ppos, |
| 2883 | struct pipe_inode_info *pipe, |
| 2884 | size_t len, unsigned int flags) |
| 2885 | { |
| 2886 | struct folio_batch fbatch; |
| 2887 | struct kiocb iocb; |
| 2888 | size_t total_spliced = 0, used, npages; |
| 2889 | loff_t isize, end_offset; |
| 2890 | bool writably_mapped; |
| 2891 | int i, error = 0; |
| 2892 | |
| 2893 | init_sync_kiocb(&iocb, in); |
| 2894 | iocb.ki_pos = *ppos; |
| 2895 | |
| 2896 | /* Work out how much data we can actually add into the pipe */ |
| 2897 | used = pipe_occupancy(pipe->head, pipe->tail); |
| 2898 | npages = max_t(ssize_t, pipe->max_usage - used, 0); |
| 2899 | len = min_t(size_t, len, npages * PAGE_SIZE); |
| 2900 | |
| 2901 | folio_batch_init(&fbatch); |
| 2902 | |
| 2903 | do { |
| 2904 | cond_resched(); |
| 2905 | |
| 2906 | if (*ppos >= i_size_read(file_inode(in))) |
| 2907 | break; |
| 2908 | |
| 2909 | iocb.ki_pos = *ppos; |
| 2910 | error = filemap_get_pages(&iocb, len, &fbatch, true); |
| 2911 | if (error < 0) |
| 2912 | break; |
| 2913 | |
| 2914 | /* |
| 2915 | * i_size must be checked after we know the pages are Uptodate. |
| 2916 | * |
| 2917 | * Checking i_size after the check allows us to calculate |
| 2918 | * the correct value for "nr", which means the zero-filled |
| 2919 | * part of the page is not copied back to userspace (unless |
| 2920 | * another truncate extends the file - this is desired though). |
| 2921 | */ |
| 2922 | isize = i_size_read(file_inode(in)); |
| 2923 | if (unlikely(*ppos >= isize)) |
| 2924 | break; |
| 2925 | end_offset = min_t(loff_t, isize, *ppos + len); |
| 2926 | |
| 2927 | /* |
| 2928 | * Once we start copying data, we don't want to be touching any |
| 2929 | * cachelines that might be contended: |
| 2930 | */ |
| 2931 | writably_mapped = mapping_writably_mapped(in->f_mapping); |
| 2932 | |
| 2933 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
| 2934 | struct folio *folio = fbatch.folios[i]; |
| 2935 | size_t n; |
| 2936 | |
| 2937 | if (folio_pos(folio) >= end_offset) |
| 2938 | goto out; |
| 2939 | folio_mark_accessed(folio); |
| 2940 | |
| 2941 | /* |
| 2942 | * If users can be writing to this folio using arbitrary |
| 2943 | * virtual addresses, take care of potential aliasing |
| 2944 | * before reading the folio on the kernel side. |
| 2945 | */ |
| 2946 | if (writably_mapped) |
| 2947 | flush_dcache_folio(folio); |
| 2948 | |
| 2949 | n = min_t(loff_t, len, isize - *ppos); |
| 2950 | n = splice_folio_into_pipe(pipe, folio, *ppos, n); |
| 2951 | if (!n) |
| 2952 | goto out; |
| 2953 | len -= n; |
| 2954 | total_spliced += n; |
| 2955 | *ppos += n; |
| 2956 | in->f_ra.prev_pos = *ppos; |
| 2957 | if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) |
| 2958 | goto out; |
| 2959 | } |
| 2960 | |
| 2961 | folio_batch_release(&fbatch); |
| 2962 | } while (len); |
| 2963 | |
| 2964 | out: |
| 2965 | folio_batch_release(&fbatch); |
| 2966 | file_accessed(in); |
| 2967 | |
| 2968 | return total_spliced ? total_spliced : error; |
| 2969 | } |
David Howells | 7c8e01e | 2023-02-15 08:00:31 +0000 | [diff] [blame] | 2970 | EXPORT_SYMBOL(filemap_splice_read); |
David Howells | 07073eb | 2023-02-14 15:01:42 +0000 | [diff] [blame] | 2971 | |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2972 | static inline loff_t folio_seek_hole_data(struct xa_state *xas, |
| 2973 | struct address_space *mapping, struct folio *folio, |
Matthew Wilcox (Oracle) | 54fa39a | 2021-02-25 17:15:52 -0800 | [diff] [blame] | 2974 | loff_t start, loff_t end, bool seek_data) |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 2975 | { |
Matthew Wilcox (Oracle) | 54fa39a | 2021-02-25 17:15:52 -0800 | [diff] [blame] | 2976 | const struct address_space_operations *ops = mapping->a_ops; |
| 2977 | size_t offset, bsz = i_blocksize(mapping->host); |
| 2978 | |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2979 | if (xa_is_value(folio) || folio_test_uptodate(folio)) |
Matthew Wilcox (Oracle) | 54fa39a | 2021-02-25 17:15:52 -0800 | [diff] [blame] | 2980 | return seek_data ? start : end; |
| 2981 | if (!ops->is_partially_uptodate) |
| 2982 | return seek_data ? end : start; |
| 2983 | |
| 2984 | xas_pause(xas); |
| 2985 | rcu_read_unlock(); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2986 | folio_lock(folio); |
| 2987 | if (unlikely(folio->mapping != mapping)) |
Matthew Wilcox (Oracle) | 54fa39a | 2021-02-25 17:15:52 -0800 | [diff] [blame] | 2988 | goto unlock; |
| 2989 | |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2990 | offset = offset_in_folio(folio, start) & ~(bsz - 1); |
Matthew Wilcox (Oracle) | 54fa39a | 2021-02-25 17:15:52 -0800 | [diff] [blame] | 2991 | |
| 2992 | do { |
Matthew Wilcox (Oracle) | 2e7e80f | 2022-02-09 20:21:27 +0000 | [diff] [blame] | 2993 | if (ops->is_partially_uptodate(folio, offset, bsz) == |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2994 | seek_data) |
Matthew Wilcox (Oracle) | 54fa39a | 2021-02-25 17:15:52 -0800 | [diff] [blame] | 2995 | break; |
| 2996 | start = (start + bsz) & ~(bsz - 1); |
| 2997 | offset += bsz; |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 2998 | } while (offset < folio_size(folio)); |
Matthew Wilcox (Oracle) | 54fa39a | 2021-02-25 17:15:52 -0800 | [diff] [blame] | 2999 | unlock: |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 3000 | folio_unlock(folio); |
Matthew Wilcox (Oracle) | 54fa39a | 2021-02-25 17:15:52 -0800 | [diff] [blame] | 3001 | rcu_read_lock(); |
| 3002 | return start; |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3003 | } |
| 3004 | |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 3005 | static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio) |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3006 | { |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 3007 | if (xa_is_value(folio)) |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3008 | return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 3009 | return folio_size(folio); |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3010 | } |
| 3011 | |
| 3012 | /** |
| 3013 | * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache. |
| 3014 | * @mapping: Address space to search. |
| 3015 | * @start: First byte to consider. |
| 3016 | * @end: Limit of search (exclusive). |
| 3017 | * @whence: Either SEEK_HOLE or SEEK_DATA. |
| 3018 | * |
| 3019 | * If the page cache knows which blocks contain holes and which blocks |
| 3020 | * contain data, your filesystem can use this function to implement |
| 3021 | * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are |
| 3022 | * entirely memory-based such as tmpfs, and filesystems which support |
| 3023 | * unwritten extents. |
| 3024 | * |
Ingo Molnar | f0953a1 | 2021-05-06 18:06:47 -0700 | [diff] [blame] | 3025 | * Return: The requested offset on success, or -ENXIO if @whence specifies |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3026 | * SEEK_DATA and there is no data after @start. There is an implicit hole |
| 3027 | * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start |
| 3028 | * and @end contain data. |
| 3029 | */ |
| 3030 | loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start, |
| 3031 | loff_t end, int whence) |
| 3032 | { |
| 3033 | XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT); |
Hugh Dickins | ed98b01 | 2021-04-23 14:29:00 -0700 | [diff] [blame] | 3034 | pgoff_t max = (end - 1) >> PAGE_SHIFT; |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3035 | bool seek_data = (whence == SEEK_DATA); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 3036 | struct folio *folio; |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3037 | |
| 3038 | if (end <= start) |
| 3039 | return -ENXIO; |
| 3040 | |
| 3041 | rcu_read_lock(); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 3042 | while ((folio = find_get_entry(&xas, max, XA_PRESENT))) { |
Hugh Dickins | ed98b01 | 2021-04-23 14:29:00 -0700 | [diff] [blame] | 3043 | loff_t pos = (u64)xas.xa_index << PAGE_SHIFT; |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 3044 | size_t seek_size; |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3045 | |
| 3046 | if (start < pos) { |
| 3047 | if (!seek_data) |
| 3048 | goto unlock; |
| 3049 | start = pos; |
| 3050 | } |
| 3051 | |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 3052 | seek_size = seek_folio_size(&xas, folio); |
| 3053 | pos = round_up((u64)pos + 1, seek_size); |
| 3054 | start = folio_seek_hole_data(&xas, mapping, folio, start, pos, |
Matthew Wilcox (Oracle) | 54fa39a | 2021-02-25 17:15:52 -0800 | [diff] [blame] | 3055 | seek_data); |
| 3056 | if (start < pos) |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3057 | goto unlock; |
Hugh Dickins | ed98b01 | 2021-04-23 14:29:00 -0700 | [diff] [blame] | 3058 | if (start >= end) |
| 3059 | break; |
| 3060 | if (seek_size > PAGE_SIZE) |
| 3061 | xas_set(&xas, pos >> PAGE_SHIFT); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 3062 | if (!xa_is_value(folio)) |
| 3063 | folio_put(folio); |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3064 | } |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3065 | if (seek_data) |
Hugh Dickins | ed98b01 | 2021-04-23 14:29:00 -0700 | [diff] [blame] | 3066 | start = -ENXIO; |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3067 | unlock: |
| 3068 | rcu_read_unlock(); |
Matthew Wilcox (Oracle) | f5e6429 | 2020-12-17 00:12:26 -0500 | [diff] [blame] | 3069 | if (folio && !xa_is_value(folio)) |
| 3070 | folio_put(folio); |
Matthew Wilcox (Oracle) | 41139aa | 2021-02-25 17:15:48 -0800 | [diff] [blame] | 3071 | if (start > end) |
| 3072 | return end; |
| 3073 | return start; |
| 3074 | } |
| 3075 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3076 | #ifdef CONFIG_MMU |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3077 | #define MMAP_LOTSAMISS (100) |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3078 | /* |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3079 | * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3080 | * @vmf - the vm_fault for this fault. |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3081 | * @folio - the folio to lock. |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3082 | * @fpin - the pointer to the file we may pin (or is already pinned). |
| 3083 | * |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3084 | * This works similar to lock_folio_or_retry in that it can drop the |
| 3085 | * mmap_lock. It differs in that it actually returns the folio locked |
| 3086 | * if it returns 1 and 0 if it couldn't lock the folio. If we did have |
| 3087 | * to drop the mmap_lock then fpin will point to the pinned file and |
| 3088 | * needs to be fput()'ed at a later point. |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3089 | */ |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3090 | static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio, |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3091 | struct file **fpin) |
| 3092 | { |
Matthew Wilcox (Oracle) | 7c23c78 | 2021-03-01 19:38:25 -0500 | [diff] [blame] | 3093 | if (folio_trylock(folio)) |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3094 | return 1; |
| 3095 | |
Linus Torvalds | 8b0f9fa | 2019-03-15 11:26:07 -0700 | [diff] [blame] | 3096 | /* |
| 3097 | * NOTE! This will make us return with VM_FAULT_RETRY, but with |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 3098 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
Linus Torvalds | 8b0f9fa | 2019-03-15 11:26:07 -0700 | [diff] [blame] | 3099 | * is supposed to work. We have way too many special cases.. |
| 3100 | */ |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3101 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
| 3102 | return 0; |
| 3103 | |
| 3104 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); |
| 3105 | if (vmf->flags & FAULT_FLAG_KILLABLE) { |
Matthew Wilcox (Oracle) | af7f29d | 2020-12-08 00:07:31 -0500 | [diff] [blame] | 3106 | if (__folio_lock_killable(folio)) { |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3107 | /* |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 3108 | * We didn't have the right flags to drop the mmap_lock, |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3109 | * but all fault_handlers only check for fatal signals |
| 3110 | * if we return VM_FAULT_RETRY, so we need to drop the |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 3111 | * mmap_lock here and return 0 if we don't have a fpin. |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3112 | */ |
| 3113 | if (*fpin == NULL) |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 3114 | mmap_read_unlock(vmf->vma->vm_mm); |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3115 | return 0; |
| 3116 | } |
| 3117 | } else |
Matthew Wilcox (Oracle) | 7c23c78 | 2021-03-01 19:38:25 -0500 | [diff] [blame] | 3118 | __folio_lock(folio); |
| 3119 | |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3120 | return 1; |
| 3121 | } |
| 3122 | |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3123 | /* |
| 3124 | * Synchronous readahead happens when we don't even find a page in the page |
| 3125 | * cache at all. We don't want to perform IO under the mmap sem, so if we have |
| 3126 | * to drop the mmap sem we return the file that was pinned in order for us to do |
| 3127 | * that. If we didn't pin a file then we return NULL. The file that is |
| 3128 | * returned needs to be fput()'ed when we're done with it. |
| 3129 | */ |
| 3130 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3131 | { |
Josef Bacik | 2a1180f | 2019-03-13 11:44:18 -0700 | [diff] [blame] | 3132 | struct file *file = vmf->vma->vm_file; |
| 3133 | struct file_ra_state *ra = &file->f_ra; |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3134 | struct address_space *mapping = file->f_mapping; |
Matthew Wilcox (Oracle) | fcd9ae4 | 2021-04-07 21:18:55 +0100 | [diff] [blame] | 3135 | DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff); |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3136 | struct file *fpin = NULL; |
Matthew Wilcox (Oracle) | dcfa24b | 2022-05-25 14:23:45 -0400 | [diff] [blame] | 3137 | unsigned long vm_flags = vmf->vma->vm_flags; |
Kirill A. Shutemov | e630bfac | 2020-08-14 17:31:27 -0700 | [diff] [blame] | 3138 | unsigned int mmap_miss; |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3139 | |
Matthew Wilcox (Oracle) | 4687fdb | 2021-07-24 23:37:13 -0400 | [diff] [blame] | 3140 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 3141 | /* Use the readahead code, even if readahead is disabled */ |
Matthew Wilcox (Oracle) | dcfa24b | 2022-05-25 14:23:45 -0400 | [diff] [blame] | 3142 | if (vm_flags & VM_HUGEPAGE) { |
Matthew Wilcox (Oracle) | 4687fdb | 2021-07-24 23:37:13 -0400 | [diff] [blame] | 3143 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
| 3144 | ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1); |
| 3145 | ra->size = HPAGE_PMD_NR; |
| 3146 | /* |
| 3147 | * Fetch two PMD folios, so we get the chance to actually |
| 3148 | * readahead, unless we've been told not to. |
| 3149 | */ |
Matthew Wilcox (Oracle) | dcfa24b | 2022-05-25 14:23:45 -0400 | [diff] [blame] | 3150 | if (!(vm_flags & VM_RAND_READ)) |
Matthew Wilcox (Oracle) | 4687fdb | 2021-07-24 23:37:13 -0400 | [diff] [blame] | 3151 | ra->size *= 2; |
| 3152 | ra->async_size = HPAGE_PMD_NR; |
| 3153 | page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER); |
| 3154 | return fpin; |
| 3155 | } |
| 3156 | #endif |
| 3157 | |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3158 | /* If we don't want any read-ahead, don't bother */ |
Matthew Wilcox (Oracle) | dcfa24b | 2022-05-25 14:23:45 -0400 | [diff] [blame] | 3159 | if (vm_flags & VM_RAND_READ) |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3160 | return fpin; |
Wu Fengguang | 275b12b | 2011-05-24 17:12:28 -0700 | [diff] [blame] | 3161 | if (!ra->ra_pages) |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3162 | return fpin; |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3163 | |
Matthew Wilcox (Oracle) | dcfa24b | 2022-05-25 14:23:45 -0400 | [diff] [blame] | 3164 | if (vm_flags & VM_SEQ_READ) { |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3165 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
Matthew Wilcox (Oracle) | fcd9ae4 | 2021-04-07 21:18:55 +0100 | [diff] [blame] | 3166 | page_cache_sync_ra(&ractl, ra->ra_pages); |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3167 | return fpin; |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3168 | } |
| 3169 | |
Andi Kleen | 207d04b | 2011-05-24 17:12:29 -0700 | [diff] [blame] | 3170 | /* Avoid banging the cache line if not needed */ |
Kirill A. Shutemov | e630bfac | 2020-08-14 17:31:27 -0700 | [diff] [blame] | 3171 | mmap_miss = READ_ONCE(ra->mmap_miss); |
| 3172 | if (mmap_miss < MMAP_LOTSAMISS * 10) |
| 3173 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss); |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3174 | |
| 3175 | /* |
| 3176 | * Do we miss much more than hit in this file? If so, |
| 3177 | * stop bothering with read-ahead. It will only hurt. |
| 3178 | */ |
Kirill A. Shutemov | e630bfac | 2020-08-14 17:31:27 -0700 | [diff] [blame] | 3179 | if (mmap_miss > MMAP_LOTSAMISS) |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3180 | return fpin; |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3181 | |
Wu Fengguang | d30a110 | 2009-06-16 15:31:30 -0700 | [diff] [blame] | 3182 | /* |
| 3183 | * mmap read-around |
| 3184 | */ |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3185 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
David Howells | db660d4 | 2020-10-15 20:06:31 -0700 | [diff] [blame] | 3186 | ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2); |
Roman Gushchin | 600e19a | 2015-11-05 18:47:08 -0800 | [diff] [blame] | 3187 | ra->size = ra->ra_pages; |
| 3188 | ra->async_size = ra->ra_pages / 4; |
David Howells | db660d4 | 2020-10-15 20:06:31 -0700 | [diff] [blame] | 3189 | ractl._index = ra->start; |
Matthew Wilcox (Oracle) | 56a4d67 | 2021-07-24 23:26:14 -0400 | [diff] [blame] | 3190 | page_cache_ra_order(&ractl, ra, 0); |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3191 | return fpin; |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3192 | } |
| 3193 | |
| 3194 | /* |
| 3195 | * Asynchronous readahead happens when we find the page and PG_readahead, |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3196 | * so we want to possibly extend the readahead further. We return the file that |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 3197 | * was pinned if we have to drop the mmap_lock in order to do IO. |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3198 | */ |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3199 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
Matthew Wilcox (Oracle) | 79598ce | 2021-07-29 14:57:01 -0400 | [diff] [blame] | 3200 | struct folio *folio) |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3201 | { |
Josef Bacik | 2a1180f | 2019-03-13 11:44:18 -0700 | [diff] [blame] | 3202 | struct file *file = vmf->vma->vm_file; |
| 3203 | struct file_ra_state *ra = &file->f_ra; |
Matthew Wilcox (Oracle) | 79598ce | 2021-07-29 14:57:01 -0400 | [diff] [blame] | 3204 | DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff); |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3205 | struct file *fpin = NULL; |
Kirill A. Shutemov | e630bfac | 2020-08-14 17:31:27 -0700 | [diff] [blame] | 3206 | unsigned int mmap_miss; |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3207 | |
| 3208 | /* If we don't want any read-ahead, don't bother */ |
Jan Kara | 5c72fee | 2020-04-01 21:04:40 -0700 | [diff] [blame] | 3209 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3210 | return fpin; |
Matthew Wilcox (Oracle) | 79598ce | 2021-07-29 14:57:01 -0400 | [diff] [blame] | 3211 | |
Kirill A. Shutemov | e630bfac | 2020-08-14 17:31:27 -0700 | [diff] [blame] | 3212 | mmap_miss = READ_ONCE(ra->mmap_miss); |
| 3213 | if (mmap_miss) |
| 3214 | WRITE_ONCE(ra->mmap_miss, --mmap_miss); |
Matthew Wilcox (Oracle) | 79598ce | 2021-07-29 14:57:01 -0400 | [diff] [blame] | 3215 | |
| 3216 | if (folio_test_readahead(folio)) { |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3217 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
Matthew Wilcox (Oracle) | 79598ce | 2021-07-29 14:57:01 -0400 | [diff] [blame] | 3218 | page_cache_async_ra(&ractl, folio, ra->ra_pages); |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3219 | } |
| 3220 | return fpin; |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3221 | } |
| 3222 | |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 3223 | /** |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 3224 | * filemap_fault - read in file data for page fault handling |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 3225 | * @vmf: struct vm_fault containing details of the fault |
Randy Dunlap | 485bb99 | 2006-06-23 02:03:49 -0700 | [diff] [blame] | 3226 | * |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 3227 | * filemap_fault() is invoked via the vma operations vector for a |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3228 | * mapped memory region to read in file data during a page fault. |
| 3229 | * |
| 3230 | * The goto's are kind of ugly, but this streamlines the normal case of having |
| 3231 | * it in the page cache, and handles the special cases reasonably without |
| 3232 | * having a lot of duplicated code. |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 3233 | * |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 3234 | * vma->vm_mm->mmap_lock must be held on entry. |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 3235 | * |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 3236 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3237 | * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap(). |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 3238 | * |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 3239 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
Paul Cassella | 9a95f3c | 2014-08-06 16:07:24 -0700 | [diff] [blame] | 3240 | * has not been released. |
| 3241 | * |
| 3242 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 3243 | * |
| 3244 | * Return: bitwise-OR of %VM_FAULT_ codes. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3245 | */ |
Souptick Joarder | 2bcd645 | 2018-06-07 17:08:00 -0700 | [diff] [blame] | 3246 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3247 | { |
| 3248 | int error; |
Dave Jiang | 11bac80 | 2017-02-24 14:56:41 -0800 | [diff] [blame] | 3249 | struct file *file = vmf->vma->vm_file; |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3250 | struct file *fpin = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3251 | struct address_space *mapping = file->f_mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3252 | struct inode *inode = mapping->host; |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3253 | pgoff_t max_idx, index = vmf->pgoff; |
| 3254 | struct folio *folio; |
Souptick Joarder | 2bcd645 | 2018-06-07 17:08:00 -0700 | [diff] [blame] | 3255 | vm_fault_t ret = 0; |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3256 | bool mapping_locked = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3257 | |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3258 | max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
| 3259 | if (unlikely(index >= max_idx)) |
Linus Torvalds | 5307cc1 | 2007-10-31 09:19:46 -0700 | [diff] [blame] | 3260 | return VM_FAULT_SIGBUS; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3261 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3262 | /* |
Johannes Weiner | 49426420 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 3263 | * Do we have something in the page cache already? |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3264 | */ |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3265 | folio = filemap_get_folio(mapping, index); |
| 3266 | if (likely(folio)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3267 | /* |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3268 | * We found the page, so try async readahead before waiting for |
| 3269 | * the lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3270 | */ |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3271 | if (!(vmf->flags & FAULT_FLAG_TRIED)) |
Matthew Wilcox (Oracle) | 79598ce | 2021-07-29 14:57:01 -0400 | [diff] [blame] | 3272 | fpin = do_async_mmap_readahead(vmf, folio); |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3273 | if (unlikely(!folio_test_uptodate(folio))) { |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3274 | filemap_invalidate_lock_shared(mapping); |
| 3275 | mapping_locked = true; |
| 3276 | } |
| 3277 | } else { |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3278 | /* No page in the page cache at all */ |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3279 | count_vm_event(PGMAJFAULT); |
Roman Gushchin | 2262185 | 2017-07-06 15:40:25 -0700 | [diff] [blame] | 3280 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3281 | ret = VM_FAULT_MAJOR; |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3282 | fpin = do_sync_mmap_readahead(vmf); |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3283 | retry_find: |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3284 | /* |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3285 | * See comment in filemap_create_folio() why we need |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3286 | * invalidate_lock |
| 3287 | */ |
| 3288 | if (!mapping_locked) { |
| 3289 | filemap_invalidate_lock_shared(mapping); |
| 3290 | mapping_locked = true; |
| 3291 | } |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3292 | folio = __filemap_get_folio(mapping, index, |
Josef Bacik | a75d4c3 | 2019-03-13 11:44:14 -0700 | [diff] [blame] | 3293 | FGP_CREAT|FGP_FOR_MMAP, |
| 3294 | vmf->gfp_mask); |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3295 | if (!folio) { |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3296 | if (fpin) |
| 3297 | goto out_retry; |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3298 | filemap_invalidate_unlock_shared(mapping); |
Matthew Wilcox (Oracle) | e520e93 | 2020-04-01 21:04:53 -0700 | [diff] [blame] | 3299 | return VM_FAULT_OOM; |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3300 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3301 | } |
| 3302 | |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3303 | if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin)) |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3304 | goto out_retry; |
Michel Lespinasse | b522c94 | 2010-10-26 14:21:56 -0700 | [diff] [blame] | 3305 | |
| 3306 | /* Did it get truncated? */ |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3307 | if (unlikely(folio->mapping != mapping)) { |
| 3308 | folio_unlock(folio); |
| 3309 | folio_put(folio); |
Michel Lespinasse | b522c94 | 2010-10-26 14:21:56 -0700 | [diff] [blame] | 3310 | goto retry_find; |
| 3311 | } |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3312 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); |
Michel Lespinasse | b522c94 | 2010-10-26 14:21:56 -0700 | [diff] [blame] | 3313 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3314 | /* |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 3315 | * We have a locked page in the page cache, now we need to check |
| 3316 | * that it's up-to-date. If not, it is going to be due to an error. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3317 | */ |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3318 | if (unlikely(!folio_test_uptodate(folio))) { |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3319 | /* |
| 3320 | * The page was in cache and uptodate and now it is not. |
| 3321 | * Strange but possible since we didn't hold the page lock all |
| 3322 | * the time. Let's drop everything get the invalidate lock and |
| 3323 | * try again. |
| 3324 | */ |
| 3325 | if (!mapping_locked) { |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3326 | folio_unlock(folio); |
| 3327 | folio_put(folio); |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3328 | goto retry_find; |
| 3329 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3330 | goto page_not_uptodate; |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3331 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3332 | |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3333 | /* |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 3334 | * We've made it this far and we had to drop our mmap_lock, now is the |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3335 | * time to return to the upper layer and have it re-find the vma and |
| 3336 | * redo the fault. |
| 3337 | */ |
| 3338 | if (fpin) { |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3339 | folio_unlock(folio); |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3340 | goto out_retry; |
| 3341 | } |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3342 | if (mapping_locked) |
| 3343 | filemap_invalidate_unlock_shared(mapping); |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3344 | |
| 3345 | /* |
Linus Torvalds | ef00e08 | 2009-06-16 15:31:25 -0700 | [diff] [blame] | 3346 | * Found the page and have a reference on it. |
| 3347 | * We must recheck i_size under page lock. |
| 3348 | */ |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3349 | max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
| 3350 | if (unlikely(index >= max_idx)) { |
| 3351 | folio_unlock(folio); |
| 3352 | folio_put(folio); |
Linus Torvalds | 5307cc1 | 2007-10-31 09:19:46 -0700 | [diff] [blame] | 3353 | return VM_FAULT_SIGBUS; |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 3354 | } |
| 3355 | |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3356 | vmf->page = folio_file_page(folio, index); |
Nick Piggin | 83c5407 | 2007-07-19 01:47:05 -0700 | [diff] [blame] | 3357 | return ret | VM_FAULT_LOCKED; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3358 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3359 | page_not_uptodate: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3360 | /* |
| 3361 | * Umm, take care of errors if the page isn't up-to-date. |
| 3362 | * Try to re-read it _once_. We do this synchronously, |
| 3363 | * because there really aren't any performance issues here |
| 3364 | * and we need to check for errors. |
| 3365 | */ |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3366 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
Matthew Wilcox (Oracle) | 290e1a3 | 2022-05-12 17:37:01 -0400 | [diff] [blame] | 3367 | error = filemap_read_folio(file, mapping->a_ops->read_folio, folio); |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3368 | if (fpin) |
| 3369 | goto out_retry; |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3370 | folio_put(folio); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 3371 | |
| 3372 | if (!error || error == AOP_TRUNCATED_PAGE) |
| 3373 | goto retry_find; |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3374 | filemap_invalidate_unlock_shared(mapping); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 3375 | |
Nick Piggin | d0217ac | 2007-07-19 01:47:03 -0700 | [diff] [blame] | 3376 | return VM_FAULT_SIGBUS; |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3377 | |
| 3378 | out_retry: |
| 3379 | /* |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 3380 | * We dropped the mmap_lock, we need to return to the fault handler to |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3381 | * re-find the vma and come back and find our hopefully still populated |
| 3382 | * page. |
| 3383 | */ |
Matthew Wilcox (Oracle) | e292e6d | 2021-03-10 10:46:41 -0500 | [diff] [blame] | 3384 | if (folio) |
| 3385 | folio_put(folio); |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3386 | if (mapping_locked) |
| 3387 | filemap_invalidate_unlock_shared(mapping); |
Josef Bacik | 6b4c9f4 | 2019-03-13 11:44:22 -0700 | [diff] [blame] | 3388 | if (fpin) |
| 3389 | fput(fpin); |
| 3390 | return ret | VM_FAULT_RETRY; |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 3391 | } |
| 3392 | EXPORT_SYMBOL(filemap_fault); |
| 3393 | |
Matthew Wilcox (Oracle) | 8808eca | 2023-01-16 19:39:39 +0000 | [diff] [blame] | 3394 | static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio, |
| 3395 | pgoff_t start) |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 3396 | { |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3397 | struct mm_struct *mm = vmf->vma->vm_mm; |
| 3398 | |
| 3399 | /* Huge page is mapped? No need to proceed. */ |
| 3400 | if (pmd_trans_huge(*vmf->pmd)) { |
Matthew Wilcox (Oracle) | 8808eca | 2023-01-16 19:39:39 +0000 | [diff] [blame] | 3401 | folio_unlock(folio); |
| 3402 | folio_put(folio); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3403 | return true; |
| 3404 | } |
| 3405 | |
Matthew Wilcox (Oracle) | 8808eca | 2023-01-16 19:39:39 +0000 | [diff] [blame] | 3406 | if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) { |
| 3407 | struct page *page = folio_file_page(folio, start); |
Yang Shi | e0f43fa | 2021-11-05 13:41:04 -0700 | [diff] [blame] | 3408 | vm_fault_t ret = do_set_pmd(vmf, page); |
| 3409 | if (!ret) { |
| 3410 | /* The page is mapped successfully, reference consumed. */ |
Matthew Wilcox (Oracle) | 8808eca | 2023-01-16 19:39:39 +0000 | [diff] [blame] | 3411 | folio_unlock(folio); |
Yang Shi | e0f43fa | 2021-11-05 13:41:04 -0700 | [diff] [blame] | 3412 | return true; |
| 3413 | } |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3414 | } |
| 3415 | |
Qi Zheng | 03c4f20 | 2021-11-05 13:38:38 -0700 | [diff] [blame] | 3416 | if (pmd_none(*vmf->pmd)) |
| 3417 | pmd_install(mm, vmf->pmd, &vmf->prealloc_pte); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3418 | |
| 3419 | /* See comment in handle_pte_fault() */ |
| 3420 | if (pmd_devmap_trans_unstable(vmf->pmd)) { |
Matthew Wilcox (Oracle) | 8808eca | 2023-01-16 19:39:39 +0000 | [diff] [blame] | 3421 | folio_unlock(folio); |
| 3422 | folio_put(folio); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3423 | return true; |
| 3424 | } |
| 3425 | |
| 3426 | return false; |
| 3427 | } |
| 3428 | |
Matthew Wilcox (Oracle) | 820b05e | 2021-03-12 23:46:45 -0500 | [diff] [blame] | 3429 | static struct folio *next_uptodate_page(struct folio *folio, |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3430 | struct address_space *mapping, |
| 3431 | struct xa_state *xas, pgoff_t end_pgoff) |
| 3432 | { |
| 3433 | unsigned long max_idx; |
| 3434 | |
| 3435 | do { |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3436 | if (!folio) |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3437 | return NULL; |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3438 | if (xas_retry(xas, folio)) |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3439 | continue; |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3440 | if (xa_is_value(folio)) |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3441 | continue; |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3442 | if (folio_test_locked(folio)) |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3443 | continue; |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3444 | if (!folio_try_get_rcu(folio)) |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3445 | continue; |
| 3446 | /* Has the page moved or been split? */ |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3447 | if (unlikely(folio != xas_reload(xas))) |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3448 | goto skip; |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3449 | if (!folio_test_uptodate(folio) || folio_test_readahead(folio)) |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3450 | goto skip; |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3451 | if (!folio_trylock(folio)) |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3452 | goto skip; |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3453 | if (folio->mapping != mapping) |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3454 | goto unlock; |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3455 | if (!folio_test_uptodate(folio)) |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3456 | goto unlock; |
| 3457 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); |
| 3458 | if (xas->xa_index >= max_idx) |
| 3459 | goto unlock; |
Matthew Wilcox (Oracle) | 820b05e | 2021-03-12 23:46:45 -0500 | [diff] [blame] | 3460 | return folio; |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3461 | unlock: |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3462 | folio_unlock(folio); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3463 | skip: |
Matthew Wilcox (Oracle) | 9184a30 | 2021-03-12 23:33:43 -0500 | [diff] [blame] | 3464 | folio_put(folio); |
| 3465 | } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3466 | |
| 3467 | return NULL; |
| 3468 | } |
| 3469 | |
Matthew Wilcox (Oracle) | 820b05e | 2021-03-12 23:46:45 -0500 | [diff] [blame] | 3470 | static inline struct folio *first_map_page(struct address_space *mapping, |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3471 | struct xa_state *xas, |
| 3472 | pgoff_t end_pgoff) |
| 3473 | { |
| 3474 | return next_uptodate_page(xas_find(xas, end_pgoff), |
| 3475 | mapping, xas, end_pgoff); |
| 3476 | } |
| 3477 | |
Matthew Wilcox (Oracle) | 820b05e | 2021-03-12 23:46:45 -0500 | [diff] [blame] | 3478 | static inline struct folio *next_map_page(struct address_space *mapping, |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3479 | struct xa_state *xas, |
| 3480 | pgoff_t end_pgoff) |
| 3481 | { |
| 3482 | return next_uptodate_page(xas_next_entry(xas, end_pgoff), |
| 3483 | mapping, xas, end_pgoff); |
| 3484 | } |
| 3485 | |
| 3486 | vm_fault_t filemap_map_pages(struct vm_fault *vmf, |
| 3487 | pgoff_t start_pgoff, pgoff_t end_pgoff) |
| 3488 | { |
| 3489 | struct vm_area_struct *vma = vmf->vma; |
| 3490 | struct file *file = vma->vm_file; |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 3491 | struct address_space *mapping = file->f_mapping; |
Kirill A. Shutemov | bae473a | 2016-07-26 15:25:20 -0700 | [diff] [blame] | 3492 | pgoff_t last_pgoff = start_pgoff; |
Will Deacon | 9d3af4b | 2021-01-14 15:24:19 +0000 | [diff] [blame] | 3493 | unsigned long addr; |
Matthew Wilcox | 070e807c | 2018-05-17 00:08:30 -0400 | [diff] [blame] | 3494 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
Matthew Wilcox (Oracle) | 820b05e | 2021-03-12 23:46:45 -0500 | [diff] [blame] | 3495 | struct folio *folio; |
| 3496 | struct page *page; |
Kirill A. Shutemov | e630bfac | 2020-08-14 17:31:27 -0700 | [diff] [blame] | 3497 | unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3498 | vm_fault_t ret = 0; |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 3499 | |
| 3500 | rcu_read_lock(); |
Matthew Wilcox (Oracle) | 820b05e | 2021-03-12 23:46:45 -0500 | [diff] [blame] | 3501 | folio = first_map_page(mapping, &xas, end_pgoff); |
| 3502 | if (!folio) |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3503 | goto out; |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 3504 | |
Matthew Wilcox (Oracle) | 8808eca | 2023-01-16 19:39:39 +0000 | [diff] [blame] | 3505 | if (filemap_map_pmd(vmf, folio, start_pgoff)) { |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3506 | ret = VM_FAULT_NOPAGE; |
| 3507 | goto out; |
| 3508 | } |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 3509 | |
Will Deacon | 9d3af4b | 2021-01-14 15:24:19 +0000 | [diff] [blame] | 3510 | addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
| 3511 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3512 | do { |
Matthew Wilcox (Oracle) | 6b24ca4 | 2020-06-27 22:19:08 -0400 | [diff] [blame] | 3513 | again: |
Matthew Wilcox (Oracle) | 820b05e | 2021-03-12 23:46:45 -0500 | [diff] [blame] | 3514 | page = folio_file_page(folio, xas.xa_index); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3515 | if (PageHWPoison(page)) |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 3516 | goto unlock; |
| 3517 | |
Kirill A. Shutemov | e630bfac | 2020-08-14 17:31:27 -0700 | [diff] [blame] | 3518 | if (mmap_miss > 0) |
| 3519 | mmap_miss--; |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3520 | |
Will Deacon | 9d3af4b | 2021-01-14 15:24:19 +0000 | [diff] [blame] | 3521 | addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3522 | vmf->pte += xas.xa_index - last_pgoff; |
Matthew Wilcox | 070e807c | 2018-05-17 00:08:30 -0400 | [diff] [blame] | 3523 | last_pgoff = xas.xa_index; |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3524 | |
Peter Xu | 5c041f5 | 2022-05-12 20:22:52 -0700 | [diff] [blame] | 3525 | /* |
| 3526 | * NOTE: If there're PTE markers, we'll leave them to be |
| 3527 | * handled in the specific fault path, and it'll prohibit the |
| 3528 | * fault-around logic. |
| 3529 | */ |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3530 | if (!pte_none(*vmf->pte)) |
Kirill A. Shutemov | 7267ec00 | 2016-07-26 15:25:23 -0700 | [diff] [blame] | 3531 | goto unlock; |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3532 | |
Will Deacon | 46bdb42 | 2020-11-24 18:48:26 +0000 | [diff] [blame] | 3533 | /* We're about to handle the fault */ |
Will Deacon | 9d3af4b | 2021-01-14 15:24:19 +0000 | [diff] [blame] | 3534 | if (vmf->address == addr) |
Will Deacon | 46bdb42 | 2020-11-24 18:48:26 +0000 | [diff] [blame] | 3535 | ret = VM_FAULT_NOPAGE; |
Will Deacon | 46bdb42 | 2020-11-24 18:48:26 +0000 | [diff] [blame] | 3536 | |
Will Deacon | 9d3af4b | 2021-01-14 15:24:19 +0000 | [diff] [blame] | 3537 | do_set_pte(vmf, page, addr); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3538 | /* no need to invalidate: a not-present page won't be cached */ |
Will Deacon | 9d3af4b | 2021-01-14 15:24:19 +0000 | [diff] [blame] | 3539 | update_mmu_cache(vma, addr, vmf->pte); |
Matthew Wilcox (Oracle) | 6b24ca4 | 2020-06-27 22:19:08 -0400 | [diff] [blame] | 3540 | if (folio_more_pages(folio, xas.xa_index, end_pgoff)) { |
| 3541 | xas.xa_index++; |
| 3542 | folio_ref_inc(folio); |
| 3543 | goto again; |
| 3544 | } |
Matthew Wilcox (Oracle) | 820b05e | 2021-03-12 23:46:45 -0500 | [diff] [blame] | 3545 | folio_unlock(folio); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3546 | continue; |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 3547 | unlock: |
Matthew Wilcox (Oracle) | 6b24ca4 | 2020-06-27 22:19:08 -0400 | [diff] [blame] | 3548 | if (folio_more_pages(folio, xas.xa_index, end_pgoff)) { |
| 3549 | xas.xa_index++; |
| 3550 | goto again; |
| 3551 | } |
Matthew Wilcox (Oracle) | 820b05e | 2021-03-12 23:46:45 -0500 | [diff] [blame] | 3552 | folio_unlock(folio); |
| 3553 | folio_put(folio); |
| 3554 | } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3555 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
| 3556 | out: |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 3557 | rcu_read_unlock(); |
Kirill A. Shutemov | e630bfac | 2020-08-14 17:31:27 -0700 | [diff] [blame] | 3558 | WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); |
Kirill A. Shutemov | f9ce0be | 2020-12-19 15:19:23 +0300 | [diff] [blame] | 3559 | return ret; |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 3560 | } |
| 3561 | EXPORT_SYMBOL(filemap_map_pages); |
| 3562 | |
Souptick Joarder | 2bcd645 | 2018-06-07 17:08:00 -0700 | [diff] [blame] | 3563 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
Jan Kara | 4fcf1c6 | 2012-06-12 16:20:29 +0200 | [diff] [blame] | 3564 | { |
Christoph Hellwig | 5df1a67 | 2020-11-16 14:33:37 +0100 | [diff] [blame] | 3565 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
Matthew Wilcox (Oracle) | 960ea97 | 2021-03-12 23:57:44 -0500 | [diff] [blame] | 3566 | struct folio *folio = page_folio(vmf->page); |
Souptick Joarder | 2bcd645 | 2018-06-07 17:08:00 -0700 | [diff] [blame] | 3567 | vm_fault_t ret = VM_FAULT_LOCKED; |
Jan Kara | 4fcf1c6 | 2012-06-12 16:20:29 +0200 | [diff] [blame] | 3568 | |
Christoph Hellwig | 5df1a67 | 2020-11-16 14:33:37 +0100 | [diff] [blame] | 3569 | sb_start_pagefault(mapping->host->i_sb); |
Dave Jiang | 11bac80 | 2017-02-24 14:56:41 -0800 | [diff] [blame] | 3570 | file_update_time(vmf->vma->vm_file); |
Matthew Wilcox (Oracle) | 960ea97 | 2021-03-12 23:57:44 -0500 | [diff] [blame] | 3571 | folio_lock(folio); |
| 3572 | if (folio->mapping != mapping) { |
| 3573 | folio_unlock(folio); |
Jan Kara | 4fcf1c6 | 2012-06-12 16:20:29 +0200 | [diff] [blame] | 3574 | ret = VM_FAULT_NOPAGE; |
| 3575 | goto out; |
| 3576 | } |
Jan Kara | 14da920 | 2012-06-12 16:20:37 +0200 | [diff] [blame] | 3577 | /* |
Matthew Wilcox (Oracle) | 960ea97 | 2021-03-12 23:57:44 -0500 | [diff] [blame] | 3578 | * We mark the folio dirty already here so that when freeze is in |
Jan Kara | 14da920 | 2012-06-12 16:20:37 +0200 | [diff] [blame] | 3579 | * progress, we are guaranteed that writeback during freezing will |
Matthew Wilcox (Oracle) | 960ea97 | 2021-03-12 23:57:44 -0500 | [diff] [blame] | 3580 | * see the dirty folio and writeprotect it again. |
Jan Kara | 14da920 | 2012-06-12 16:20:37 +0200 | [diff] [blame] | 3581 | */ |
Matthew Wilcox (Oracle) | 960ea97 | 2021-03-12 23:57:44 -0500 | [diff] [blame] | 3582 | folio_mark_dirty(folio); |
| 3583 | folio_wait_stable(folio); |
Jan Kara | 4fcf1c6 | 2012-06-12 16:20:29 +0200 | [diff] [blame] | 3584 | out: |
Christoph Hellwig | 5df1a67 | 2020-11-16 14:33:37 +0100 | [diff] [blame] | 3585 | sb_end_pagefault(mapping->host->i_sb); |
Jan Kara | 4fcf1c6 | 2012-06-12 16:20:29 +0200 | [diff] [blame] | 3586 | return ret; |
| 3587 | } |
Jan Kara | 4fcf1c6 | 2012-06-12 16:20:29 +0200 | [diff] [blame] | 3588 | |
Alexey Dobriyan | f0f37e2f | 2009-09-27 22:29:37 +0400 | [diff] [blame] | 3589 | const struct vm_operations_struct generic_file_vm_ops = { |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 3590 | .fault = filemap_fault, |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 3591 | .map_pages = filemap_map_pages, |
Jan Kara | 4fcf1c6 | 2012-06-12 16:20:29 +0200 | [diff] [blame] | 3592 | .page_mkwrite = filemap_page_mkwrite, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3593 | }; |
| 3594 | |
| 3595 | /* This is used for a general mmap of a disk file */ |
| 3596 | |
Zhiyuan Dai | 68d68ff | 2021-05-04 18:40:12 -0700 | [diff] [blame] | 3597 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3598 | { |
| 3599 | struct address_space *mapping = file->f_mapping; |
| 3600 | |
Matthew Wilcox (Oracle) | 7e0a126 | 2022-04-29 11:53:28 -0400 | [diff] [blame] | 3601 | if (!mapping->a_ops->read_folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3602 | return -ENOEXEC; |
| 3603 | file_accessed(file); |
| 3604 | vma->vm_ops = &generic_file_vm_ops; |
| 3605 | return 0; |
| 3606 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3607 | |
| 3608 | /* |
| 3609 | * This is for filesystems which do not implement ->writepage. |
| 3610 | */ |
| 3611 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
| 3612 | { |
| 3613 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) |
| 3614 | return -EINVAL; |
| 3615 | return generic_file_mmap(file, vma); |
| 3616 | } |
| 3617 | #else |
Souptick Joarder | 4b96a37 | 2018-10-26 15:04:03 -0700 | [diff] [blame] | 3618 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
Arnd Bergmann | 45397228 | 2018-04-13 15:35:27 -0700 | [diff] [blame] | 3619 | { |
Souptick Joarder | 4b96a37 | 2018-10-26 15:04:03 -0700 | [diff] [blame] | 3620 | return VM_FAULT_SIGBUS; |
Arnd Bergmann | 45397228 | 2018-04-13 15:35:27 -0700 | [diff] [blame] | 3621 | } |
Zhiyuan Dai | 68d68ff | 2021-05-04 18:40:12 -0700 | [diff] [blame] | 3622 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3623 | { |
| 3624 | return -ENOSYS; |
| 3625 | } |
Zhiyuan Dai | 68d68ff | 2021-05-04 18:40:12 -0700 | [diff] [blame] | 3626 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3627 | { |
| 3628 | return -ENOSYS; |
| 3629 | } |
| 3630 | #endif /* CONFIG_MMU */ |
| 3631 | |
Arnd Bergmann | 45397228 | 2018-04-13 15:35:27 -0700 | [diff] [blame] | 3632 | EXPORT_SYMBOL(filemap_page_mkwrite); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3633 | EXPORT_SYMBOL(generic_file_mmap); |
| 3634 | EXPORT_SYMBOL(generic_file_readonly_mmap); |
| 3635 | |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3636 | static struct folio *do_read_cache_folio(struct address_space *mapping, |
Matthew Wilcox (Oracle) | e9b5b23 | 2022-05-01 21:39:29 -0400 | [diff] [blame] | 3637 | pgoff_t index, filler_t filler, struct file *file, gfp_t gfp) |
Sasha Levin | 67f9fd9 | 2014-04-03 14:48:18 -0700 | [diff] [blame] | 3638 | { |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3639 | struct folio *folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3640 | int err; |
Matthew Wilcox (Oracle) | 0795000 | 2022-05-08 15:07:11 -0400 | [diff] [blame] | 3641 | |
| 3642 | if (!filler) |
| 3643 | filler = mapping->a_ops->read_folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3644 | repeat: |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3645 | folio = filemap_get_folio(mapping, index); |
| 3646 | if (!folio) { |
| 3647 | folio = filemap_alloc_folio(gfp, 0); |
| 3648 | if (!folio) |
Nick Piggin | eb2be18 | 2007-10-16 01:24:57 -0700 | [diff] [blame] | 3649 | return ERR_PTR(-ENOMEM); |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3650 | err = filemap_add_folio(mapping, folio, index, gfp); |
Nick Piggin | eb2be18 | 2007-10-16 01:24:57 -0700 | [diff] [blame] | 3651 | if (unlikely(err)) { |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3652 | folio_put(folio); |
Nick Piggin | eb2be18 | 2007-10-16 01:24:57 -0700 | [diff] [blame] | 3653 | if (err == -EEXIST) |
| 3654 | goto repeat; |
Matthew Wilcox | 22ecdb4 | 2017-12-04 04:02:00 -0500 | [diff] [blame] | 3655 | /* Presumably ENOMEM for xarray node */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3656 | return ERR_PTR(err); |
| 3657 | } |
Mel Gorman | 32b6352 | 2016-03-15 14:55:36 -0700 | [diff] [blame] | 3658 | |
Matthew Wilcox (Oracle) | 9bc3e86 | 2022-05-12 17:12:21 -0400 | [diff] [blame] | 3659 | goto filler; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3660 | } |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3661 | if (folio_test_uptodate(folio)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3662 | goto out; |
| 3663 | |
Matthew Wilcox (Oracle) | 81f4c03 | 2021-12-23 15:17:28 -0500 | [diff] [blame] | 3664 | if (!folio_trylock(folio)) { |
| 3665 | folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE); |
| 3666 | goto repeat; |
| 3667 | } |
Mel Gorman | ebded02 | 2016-03-15 14:55:39 -0700 | [diff] [blame] | 3668 | |
Matthew Wilcox (Oracle) | 81f4c03 | 2021-12-23 15:17:28 -0500 | [diff] [blame] | 3669 | /* Folio was truncated from mapping */ |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3670 | if (!folio->mapping) { |
| 3671 | folio_unlock(folio); |
| 3672 | folio_put(folio); |
Mel Gorman | 32b6352 | 2016-03-15 14:55:36 -0700 | [diff] [blame] | 3673 | goto repeat; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3674 | } |
Mel Gorman | ebded02 | 2016-03-15 14:55:39 -0700 | [diff] [blame] | 3675 | |
| 3676 | /* Someone else locked and filled the page in a very small window */ |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3677 | if (folio_test_uptodate(folio)) { |
| 3678 | folio_unlock(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3679 | goto out; |
| 3680 | } |
Xianting Tian | faffdfa | 2020-04-01 21:04:47 -0700 | [diff] [blame] | 3681 | |
Matthew Wilcox (Oracle) | 9bc3e86 | 2022-05-12 17:12:21 -0400 | [diff] [blame] | 3682 | filler: |
Matthew Wilcox (Oracle) | 290e1a3 | 2022-05-12 17:37:01 -0400 | [diff] [blame] | 3683 | err = filemap_read_folio(file, filler, folio); |
Matthew Wilcox (Oracle) | 1dfa24a | 2022-05-12 17:47:06 -0400 | [diff] [blame] | 3684 | if (err) { |
Matthew Wilcox (Oracle) | 9bc3e86 | 2022-05-12 17:12:21 -0400 | [diff] [blame] | 3685 | folio_put(folio); |
Matthew Wilcox (Oracle) | 1dfa24a | 2022-05-12 17:47:06 -0400 | [diff] [blame] | 3686 | if (err == AOP_TRUNCATED_PAGE) |
| 3687 | goto repeat; |
Matthew Wilcox (Oracle) | 9bc3e86 | 2022-05-12 17:12:21 -0400 | [diff] [blame] | 3688 | return ERR_PTR(err); |
| 3689 | } |
Mel Gorman | 32b6352 | 2016-03-15 14:55:36 -0700 | [diff] [blame] | 3690 | |
David Howells | c855ff3 | 2007-05-09 13:42:20 +0100 | [diff] [blame] | 3691 | out: |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3692 | folio_mark_accessed(folio); |
| 3693 | return folio; |
Nick Piggin | 6fe6900 | 2007-05-06 14:49:04 -0700 | [diff] [blame] | 3694 | } |
Linus Torvalds | 0531b2a | 2010-01-27 09:20:03 -0800 | [diff] [blame] | 3695 | |
| 3696 | /** |
Matthew Wilcox (Oracle) | e9b5b23 | 2022-05-01 21:39:29 -0400 | [diff] [blame] | 3697 | * read_cache_folio - Read into page cache, fill it if needed. |
| 3698 | * @mapping: The address_space to read from. |
| 3699 | * @index: The index to read. |
| 3700 | * @filler: Function to perform the read, or NULL to use aops->read_folio(). |
| 3701 | * @file: Passed to filler function, may be NULL if not required. |
Linus Torvalds | 0531b2a | 2010-01-27 09:20:03 -0800 | [diff] [blame] | 3702 | * |
Matthew Wilcox (Oracle) | e9b5b23 | 2022-05-01 21:39:29 -0400 | [diff] [blame] | 3703 | * Read one page into the page cache. If it succeeds, the folio returned |
| 3704 | * will contain @index, but it may not be the first page of the folio. |
Linus Torvalds | 0531b2a | 2010-01-27 09:20:03 -0800 | [diff] [blame] | 3705 | * |
Matthew Wilcox (Oracle) | e9b5b23 | 2022-05-01 21:39:29 -0400 | [diff] [blame] | 3706 | * If the filler function returns an error, it will be returned to the |
| 3707 | * caller. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 3708 | * |
Matthew Wilcox (Oracle) | e9b5b23 | 2022-05-01 21:39:29 -0400 | [diff] [blame] | 3709 | * Context: May sleep. Expects mapping->invalidate_lock to be held. |
| 3710 | * Return: An uptodate folio on success, ERR_PTR() on failure. |
Linus Torvalds | 0531b2a | 2010-01-27 09:20:03 -0800 | [diff] [blame] | 3711 | */ |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3712 | struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index, |
Matthew Wilcox (Oracle) | e9b5b23 | 2022-05-01 21:39:29 -0400 | [diff] [blame] | 3713 | filler_t filler, struct file *file) |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3714 | { |
Matthew Wilcox (Oracle) | e9b5b23 | 2022-05-01 21:39:29 -0400 | [diff] [blame] | 3715 | return do_read_cache_folio(mapping, index, filler, file, |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3716 | mapping_gfp_mask(mapping)); |
| 3717 | } |
| 3718 | EXPORT_SYMBOL(read_cache_folio); |
| 3719 | |
Matthew Wilcox (Oracle) | 3e62959 | 2023-02-06 16:25:19 +0000 | [diff] [blame] | 3720 | /** |
| 3721 | * mapping_read_folio_gfp - Read into page cache, using specified allocation flags. |
| 3722 | * @mapping: The address_space for the folio. |
| 3723 | * @index: The index that the allocated folio will contain. |
| 3724 | * @gfp: The page allocator flags to use if allocating. |
| 3725 | * |
| 3726 | * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with |
| 3727 | * any new memory allocations done using the specified allocation flags. |
| 3728 | * |
| 3729 | * The most likely error from this function is EIO, but ENOMEM is |
| 3730 | * possible and so is EINTR. If ->read_folio returns another error, |
| 3731 | * that will be returned to the caller. |
| 3732 | * |
| 3733 | * The function expects mapping->invalidate_lock to be already held. |
| 3734 | * |
| 3735 | * Return: Uptodate folio on success, ERR_PTR() on failure. |
| 3736 | */ |
| 3737 | struct folio *mapping_read_folio_gfp(struct address_space *mapping, |
| 3738 | pgoff_t index, gfp_t gfp) |
| 3739 | { |
| 3740 | return do_read_cache_folio(mapping, index, NULL, NULL, gfp); |
| 3741 | } |
| 3742 | EXPORT_SYMBOL(mapping_read_folio_gfp); |
| 3743 | |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3744 | static struct page *do_read_cache_page(struct address_space *mapping, |
Matthew Wilcox (Oracle) | e9b5b23 | 2022-05-01 21:39:29 -0400 | [diff] [blame] | 3745 | pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp) |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3746 | { |
| 3747 | struct folio *folio; |
| 3748 | |
Matthew Wilcox (Oracle) | e9b5b23 | 2022-05-01 21:39:29 -0400 | [diff] [blame] | 3749 | folio = do_read_cache_folio(mapping, index, filler, file, gfp); |
Matthew Wilcox (Oracle) | 539a332 | 2020-12-16 11:45:30 -0500 | [diff] [blame] | 3750 | if (IS_ERR(folio)) |
| 3751 | return &folio->page; |
| 3752 | return folio_file_page(folio, index); |
| 3753 | } |
| 3754 | |
Sasha Levin | 67f9fd9 | 2014-04-03 14:48:18 -0700 | [diff] [blame] | 3755 | struct page *read_cache_page(struct address_space *mapping, |
Matthew Wilcox (Oracle) | e9b5b23 | 2022-05-01 21:39:29 -0400 | [diff] [blame] | 3756 | pgoff_t index, filler_t *filler, struct file *file) |
Linus Torvalds | 0531b2a | 2010-01-27 09:20:03 -0800 | [diff] [blame] | 3757 | { |
Matthew Wilcox (Oracle) | e9b5b23 | 2022-05-01 21:39:29 -0400 | [diff] [blame] | 3758 | return do_read_cache_page(mapping, index, filler, file, |
Christoph Hellwig | d322a8e | 2019-07-11 20:55:17 -0700 | [diff] [blame] | 3759 | mapping_gfp_mask(mapping)); |
Linus Torvalds | 0531b2a | 2010-01-27 09:20:03 -0800 | [diff] [blame] | 3760 | } |
Sasha Levin | 67f9fd9 | 2014-04-03 14:48:18 -0700 | [diff] [blame] | 3761 | EXPORT_SYMBOL(read_cache_page); |
Linus Torvalds | 0531b2a | 2010-01-27 09:20:03 -0800 | [diff] [blame] | 3762 | |
| 3763 | /** |
| 3764 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. |
| 3765 | * @mapping: the page's address_space |
| 3766 | * @index: the page index |
| 3767 | * @gfp: the page allocator flags to use if allocating |
| 3768 | * |
| 3769 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with |
Dave Kleikamp | e6f67b8 | 2011-12-21 11:05:48 -0600 | [diff] [blame] | 3770 | * any new page allocations done using the specified allocation flags. |
Linus Torvalds | 0531b2a | 2010-01-27 09:20:03 -0800 | [diff] [blame] | 3771 | * |
| 3772 | * If the page does not get brought uptodate, return -EIO. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 3773 | * |
Jan Kara | 730633f | 2021-01-28 19:19:45 +0100 | [diff] [blame] | 3774 | * The function expects mapping->invalidate_lock to be already held. |
| 3775 | * |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 3776 | * Return: up to date page on success, ERR_PTR() on failure. |
Linus Torvalds | 0531b2a | 2010-01-27 09:20:03 -0800 | [diff] [blame] | 3777 | */ |
| 3778 | struct page *read_cache_page_gfp(struct address_space *mapping, |
| 3779 | pgoff_t index, |
| 3780 | gfp_t gfp) |
| 3781 | { |
Christoph Hellwig | 6c45b45 | 2019-07-11 20:55:20 -0700 | [diff] [blame] | 3782 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
Linus Torvalds | 0531b2a | 2010-01-27 09:20:03 -0800 | [diff] [blame] | 3783 | } |
| 3784 | EXPORT_SYMBOL(read_cache_page_gfp); |
| 3785 | |
Konstantin Khlebnikov | a92853b | 2019-11-30 17:49:44 -0800 | [diff] [blame] | 3786 | /* |
| 3787 | * Warn about a page cache invalidation failure during a direct I/O write. |
| 3788 | */ |
| 3789 | void dio_warn_stale_pagecache(struct file *filp) |
| 3790 | { |
| 3791 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); |
| 3792 | char pathname[128]; |
Konstantin Khlebnikov | a92853b | 2019-11-30 17:49:44 -0800 | [diff] [blame] | 3793 | char *path; |
| 3794 | |
Christoph Hellwig | 5df1a67 | 2020-11-16 14:33:37 +0100 | [diff] [blame] | 3795 | errseq_set(&filp->f_mapping->wb_err, -EIO); |
Konstantin Khlebnikov | a92853b | 2019-11-30 17:49:44 -0800 | [diff] [blame] | 3796 | if (__ratelimit(&_rs)) { |
| 3797 | path = file_path(filp, pathname, sizeof(pathname)); |
| 3798 | if (IS_ERR(path)) |
| 3799 | path = "(unknown)"; |
| 3800 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); |
| 3801 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, |
| 3802 | current->comm); |
| 3803 | } |
| 3804 | } |
| 3805 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3806 | ssize_t |
Christoph Hellwig | 1af5bb4 | 2016-04-07 08:51:56 -0700 | [diff] [blame] | 3807 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3808 | { |
| 3809 | struct file *file = iocb->ki_filp; |
| 3810 | struct address_space *mapping = file->f_mapping; |
| 3811 | struct inode *inode = mapping->host; |
Christoph Hellwig | 1af5bb4 | 2016-04-07 08:51:56 -0700 | [diff] [blame] | 3812 | loff_t pos = iocb->ki_pos; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3813 | ssize_t written; |
Christoph Hellwig | a969e90 | 2008-07-23 21:27:04 -0700 | [diff] [blame] | 3814 | size_t write_len; |
| 3815 | pgoff_t end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3816 | |
Al Viro | 0c94933 | 2014-03-22 06:51:37 -0400 | [diff] [blame] | 3817 | write_len = iov_iter_count(from); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 3818 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
Christoph Hellwig | a969e90 | 2008-07-23 21:27:04 -0700 | [diff] [blame] | 3819 | |
Goldwyn Rodrigues | 6be96d3 | 2017-06-20 07:05:44 -0500 | [diff] [blame] | 3820 | if (iocb->ki_flags & IOCB_NOWAIT) { |
| 3821 | /* If there are pages to writeback, return */ |
Christoph Hellwig | 5df1a67 | 2020-11-16 14:33:37 +0100 | [diff] [blame] | 3822 | if (filemap_range_has_page(file->f_mapping, pos, |
zhengbin | 35f12f0 | 2019-03-05 15:44:21 -0800 | [diff] [blame] | 3823 | pos + write_len - 1)) |
Goldwyn Rodrigues | 6be96d3 | 2017-06-20 07:05:44 -0500 | [diff] [blame] | 3824 | return -EAGAIN; |
| 3825 | } else { |
| 3826 | written = filemap_write_and_wait_range(mapping, pos, |
| 3827 | pos + write_len - 1); |
| 3828 | if (written) |
| 3829 | goto out; |
| 3830 | } |
Christoph Hellwig | a969e90 | 2008-07-23 21:27:04 -0700 | [diff] [blame] | 3831 | |
| 3832 | /* |
| 3833 | * After a write we want buffered reads to be sure to go to disk to get |
| 3834 | * the new data. We invalidate clean cached page from the region we're |
| 3835 | * about to write. We do this *before* the write so that we can return |
Hisashi Hifumi | 6ccfa80 | 2008-09-02 14:35:40 -0700 | [diff] [blame] | 3836 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
Christoph Hellwig | a969e90 | 2008-07-23 21:27:04 -0700 | [diff] [blame] | 3837 | */ |
Andrey Ryabinin | 55635ba | 2017-05-03 14:55:59 -0700 | [diff] [blame] | 3838 | written = invalidate_inode_pages2_range(mapping, |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 3839 | pos >> PAGE_SHIFT, end); |
Andrey Ryabinin | 55635ba | 2017-05-03 14:55:59 -0700 | [diff] [blame] | 3840 | /* |
| 3841 | * If a page can not be invalidated, return 0 to fall back |
| 3842 | * to buffered write. |
| 3843 | */ |
| 3844 | if (written) { |
| 3845 | if (written == -EBUSY) |
| 3846 | return 0; |
| 3847 | goto out; |
Christoph Hellwig | a969e90 | 2008-07-23 21:27:04 -0700 | [diff] [blame] | 3848 | } |
| 3849 | |
Al Viro | 639a93a5 | 2017-04-13 14:10:15 -0400 | [diff] [blame] | 3850 | written = mapping->a_ops->direct_IO(iocb, from); |
Christoph Hellwig | a969e90 | 2008-07-23 21:27:04 -0700 | [diff] [blame] | 3851 | |
| 3852 | /* |
| 3853 | * Finally, try again to invalidate clean pages which might have been |
| 3854 | * cached by non-direct readahead, or faulted in by get_user_pages() |
| 3855 | * if the source of the write was an mmap'ed region of the file |
| 3856 | * we're writing. Either one is a pretty crazy thing to do, |
| 3857 | * so we don't support it 100%. If this invalidation |
| 3858 | * fails, tough, the write still worked... |
Lukas Czerner | 332391a | 2017-09-21 08:16:29 -0600 | [diff] [blame] | 3859 | * |
| 3860 | * Most of the time we do not need this since dio_complete() will do |
| 3861 | * the invalidation for us. However there are some file systems that |
| 3862 | * do not end up with dio_complete() being called, so let's not break |
Konstantin Khlebnikov | 80c1fe9 | 2019-11-30 17:49:41 -0800 | [diff] [blame] | 3863 | * them by removing it completely. |
| 3864 | * |
Konstantin Khlebnikov | 9266a14 | 2019-11-30 17:49:47 -0800 | [diff] [blame] | 3865 | * Noticeable example is a blkdev_direct_IO(). |
| 3866 | * |
Konstantin Khlebnikov | 80c1fe9 | 2019-11-30 17:49:41 -0800 | [diff] [blame] | 3867 | * Skip invalidation for async writes or if mapping has no pages. |
Christoph Hellwig | a969e90 | 2008-07-23 21:27:04 -0700 | [diff] [blame] | 3868 | */ |
Konstantin Khlebnikov | 9266a14 | 2019-11-30 17:49:47 -0800 | [diff] [blame] | 3869 | if (written > 0 && mapping->nrpages && |
| 3870 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) |
| 3871 | dio_warn_stale_pagecache(file); |
Christoph Hellwig | a969e90 | 2008-07-23 21:27:04 -0700 | [diff] [blame] | 3872 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3873 | if (written > 0) { |
Namhyung Kim | 0116651 | 2010-10-26 14:21:58 -0700 | [diff] [blame] | 3874 | pos += written; |
Al Viro | 639a93a5 | 2017-04-13 14:10:15 -0400 | [diff] [blame] | 3875 | write_len -= written; |
Namhyung Kim | 0116651 | 2010-10-26 14:21:58 -0700 | [diff] [blame] | 3876 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
| 3877 | i_size_write(inode, pos); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3878 | mark_inode_dirty(inode); |
| 3879 | } |
Al Viro | 5cb6c6c | 2014-02-11 20:58:20 -0500 | [diff] [blame] | 3880 | iocb->ki_pos = pos; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3881 | } |
Pavel Begunkov | ab2125d | 2021-02-24 12:01:45 -0800 | [diff] [blame] | 3882 | if (written != -EIOCBQUEUED) |
| 3883 | iov_iter_revert(from, write_len - iov_iter_count(from)); |
Christoph Hellwig | a969e90 | 2008-07-23 21:27:04 -0700 | [diff] [blame] | 3884 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3885 | return written; |
| 3886 | } |
| 3887 | EXPORT_SYMBOL(generic_file_direct_write); |
| 3888 | |
Matthew Wilcox (Oracle) | 800ba29 | 2022-02-19 23:19:49 -0500 | [diff] [blame] | 3889 | ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i) |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3890 | { |
Matthew Wilcox (Oracle) | 800ba29 | 2022-02-19 23:19:49 -0500 | [diff] [blame] | 3891 | struct file *file = iocb->ki_filp; |
| 3892 | loff_t pos = iocb->ki_pos; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3893 | struct address_space *mapping = file->f_mapping; |
| 3894 | const struct address_space_operations *a_ops = mapping->a_ops; |
| 3895 | long status = 0; |
| 3896 | ssize_t written = 0; |
Nick Piggin | 674b892 | 2007-10-16 01:25:03 -0700 | [diff] [blame] | 3897 | |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3898 | do { |
| 3899 | struct page *page; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3900 | unsigned long offset; /* Offset into pagecache page */ |
| 3901 | unsigned long bytes; /* Bytes to write to page */ |
| 3902 | size_t copied; /* Bytes copied from user */ |
Alexander Potapenko | 1468c6f | 2022-09-15 17:04:16 +0200 | [diff] [blame] | 3903 | void *fsdata = NULL; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3904 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 3905 | offset = (pos & (PAGE_SIZE - 1)); |
| 3906 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3907 | iov_iter_count(i)); |
| 3908 | |
| 3909 | again: |
Linus Torvalds | 00a3d66 | 2015-10-07 08:32:38 +0100 | [diff] [blame] | 3910 | /* |
| 3911 | * Bring in the user page that we will copy from _first_. |
| 3912 | * Otherwise there's a nasty deadlock on copying from the |
| 3913 | * same page as we're writing to, without it being marked |
| 3914 | * up-to-date. |
Linus Torvalds | 00a3d66 | 2015-10-07 08:32:38 +0100 | [diff] [blame] | 3915 | */ |
Andreas Gruenbacher | 631f871 | 2021-11-09 12:56:06 +0100 | [diff] [blame] | 3916 | if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) { |
Linus Torvalds | 00a3d66 | 2015-10-07 08:32:38 +0100 | [diff] [blame] | 3917 | status = -EFAULT; |
| 3918 | break; |
| 3919 | } |
| 3920 | |
Jan Kara | 296291c | 2015-10-22 13:32:21 -0700 | [diff] [blame] | 3921 | if (fatal_signal_pending(current)) { |
| 3922 | status = -EINTR; |
| 3923 | break; |
| 3924 | } |
| 3925 | |
Matthew Wilcox (Oracle) | 9d6b0cd | 2022-02-22 14:31:43 -0500 | [diff] [blame] | 3926 | status = a_ops->write_begin(file, mapping, pos, bytes, |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3927 | &page, &fsdata); |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 3928 | if (unlikely(status < 0)) |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3929 | break; |
| 3930 | |
anfei zhou | 931e80e | 2010-02-02 13:44:02 -0800 | [diff] [blame] | 3931 | if (mapping_writably_mapped(mapping)) |
| 3932 | flush_dcache_page(page); |
Linus Torvalds | 00a3d66 | 2015-10-07 08:32:38 +0100 | [diff] [blame] | 3933 | |
Al Viro | f0b65f3 | 2021-04-30 10:26:41 -0400 | [diff] [blame] | 3934 | copied = copy_page_from_iter_atomic(page, offset, bytes, i); |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3935 | flush_dcache_page(page); |
| 3936 | |
| 3937 | status = a_ops->write_end(file, mapping, pos, bytes, copied, |
| 3938 | page, fsdata); |
Al Viro | f0b65f3 | 2021-04-30 10:26:41 -0400 | [diff] [blame] | 3939 | if (unlikely(status != copied)) { |
| 3940 | iov_iter_revert(i, copied - max(status, 0L)); |
| 3941 | if (unlikely(status < 0)) |
| 3942 | break; |
| 3943 | } |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3944 | cond_resched(); |
| 3945 | |
Al Viro | bc1bb41 | 2021-05-31 00:32:44 -0400 | [diff] [blame] | 3946 | if (unlikely(status == 0)) { |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3947 | /* |
Al Viro | bc1bb41 | 2021-05-31 00:32:44 -0400 | [diff] [blame] | 3948 | * A short copy made ->write_end() reject the |
| 3949 | * thing entirely. Might be memory poisoning |
| 3950 | * halfway through, might be a race with munmap, |
| 3951 | * might be severe memory pressure. |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3952 | */ |
Al Viro | bc1bb41 | 2021-05-31 00:32:44 -0400 | [diff] [blame] | 3953 | if (copied) |
| 3954 | bytes = copied; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3955 | goto again; |
| 3956 | } |
Al Viro | f0b65f3 | 2021-04-30 10:26:41 -0400 | [diff] [blame] | 3957 | pos += status; |
| 3958 | written += status; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3959 | |
| 3960 | balance_dirty_pages_ratelimited(mapping); |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 3961 | } while (iov_iter_count(i)); |
| 3962 | |
| 3963 | return written ? written : status; |
| 3964 | } |
Al Viro | 3b93f91 | 2014-02-11 21:34:08 -0500 | [diff] [blame] | 3965 | EXPORT_SYMBOL(generic_perform_write); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3966 | |
Jan Kara | e4dd9de | 2009-08-17 18:10:06 +0200 | [diff] [blame] | 3967 | /** |
Al Viro | 8174202 | 2014-04-03 03:17:43 -0400 | [diff] [blame] | 3968 | * __generic_file_write_iter - write data to a file |
Jan Kara | e4dd9de | 2009-08-17 18:10:06 +0200 | [diff] [blame] | 3969 | * @iocb: IO state structure (file, offset, etc.) |
Al Viro | 8174202 | 2014-04-03 03:17:43 -0400 | [diff] [blame] | 3970 | * @from: iov_iter with data to write |
Jan Kara | e4dd9de | 2009-08-17 18:10:06 +0200 | [diff] [blame] | 3971 | * |
| 3972 | * This function does all the work needed for actually writing data to a |
| 3973 | * file. It does all basic checks, removes SUID from the file, updates |
| 3974 | * modification times and calls proper subroutines depending on whether we |
| 3975 | * do direct IO or a standard buffered write. |
| 3976 | * |
Jan Kara | 9608703 | 2021-04-12 15:50:21 +0200 | [diff] [blame] | 3977 | * It expects i_rwsem to be grabbed unless we work on a block device or similar |
Jan Kara | e4dd9de | 2009-08-17 18:10:06 +0200 | [diff] [blame] | 3978 | * object which does not need locking at all. |
| 3979 | * |
| 3980 | * This function does *not* take care of syncing data in case of O_SYNC write. |
| 3981 | * A caller has to handle it. This is mainly due to the fact that we want to |
Jan Kara | 9608703 | 2021-04-12 15:50:21 +0200 | [diff] [blame] | 3982 | * avoid syncing under i_rwsem. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 3983 | * |
| 3984 | * Return: |
| 3985 | * * number of bytes written, even for truncated writes |
| 3986 | * * negative error code if no data has been written at all |
Jan Kara | e4dd9de | 2009-08-17 18:10:06 +0200 | [diff] [blame] | 3987 | */ |
Al Viro | 8174202 | 2014-04-03 03:17:43 -0400 | [diff] [blame] | 3988 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3989 | { |
| 3990 | struct file *file = iocb->ki_filp; |
Zhiyuan Dai | 68d68ff | 2021-05-04 18:40:12 -0700 | [diff] [blame] | 3991 | struct address_space *mapping = file->f_mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3992 | struct inode *inode = mapping->host; |
Al Viro | 3b93f91 | 2014-02-11 21:34:08 -0500 | [diff] [blame] | 3993 | ssize_t written = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3994 | ssize_t err; |
Al Viro | 3b93f91 | 2014-02-11 21:34:08 -0500 | [diff] [blame] | 3995 | ssize_t status; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3996 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3997 | /* We can write back this queue in page reclaim */ |
Christoph Hellwig | de1414a | 2015-01-14 10:42:36 +0100 | [diff] [blame] | 3998 | current->backing_dev_info = inode_to_bdi(inode); |
Jan Kara | 5fa8e0a | 2015-05-21 16:05:53 +0200 | [diff] [blame] | 3999 | err = file_remove_privs(file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4000 | if (err) |
| 4001 | goto out; |
| 4002 | |
Josef Bacik | c3b2da3 | 2012-03-26 09:59:21 -0400 | [diff] [blame] | 4003 | err = file_update_time(file); |
| 4004 | if (err) |
| 4005 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4006 | |
Al Viro | 2ba48ce | 2015-04-09 13:52:01 -0400 | [diff] [blame] | 4007 | if (iocb->ki_flags & IOCB_DIRECT) { |
Al Viro | 0b8def9 | 2015-04-07 10:22:53 -0400 | [diff] [blame] | 4008 | loff_t pos, endbyte; |
Jeff Moyer | fb5527e | 2006-10-19 23:28:13 -0700 | [diff] [blame] | 4009 | |
Christoph Hellwig | 1af5bb4 | 2016-04-07 08:51:56 -0700 | [diff] [blame] | 4010 | written = generic_file_direct_write(iocb, from); |
Matthew Wilcox | fbbbad4 | 2015-02-16 15:58:53 -0800 | [diff] [blame] | 4011 | /* |
| 4012 | * If the write stopped short of completing, fall back to |
| 4013 | * buffered writes. Some filesystems do this for writes to |
| 4014 | * holes, for example. For DAX files, a buffered write will |
| 4015 | * not succeed (even if it did, DAX does not handle dirty |
| 4016 | * page-cache pages correctly). |
| 4017 | */ |
Al Viro | 0b8def9 | 2015-04-07 10:22:53 -0400 | [diff] [blame] | 4018 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4019 | goto out; |
Al Viro | 3b93f91 | 2014-02-11 21:34:08 -0500 | [diff] [blame] | 4020 | |
Matthew Wilcox (Oracle) | 800ba29 | 2022-02-19 23:19:49 -0500 | [diff] [blame] | 4021 | pos = iocb->ki_pos; |
| 4022 | status = generic_perform_write(iocb, from); |
Jeff Moyer | fb5527e | 2006-10-19 23:28:13 -0700 | [diff] [blame] | 4023 | /* |
Al Viro | 3b93f91 | 2014-02-11 21:34:08 -0500 | [diff] [blame] | 4024 | * If generic_perform_write() returned a synchronous error |
Jeff Moyer | fb5527e | 2006-10-19 23:28:13 -0700 | [diff] [blame] | 4025 | * then we want to return the number of bytes which were |
| 4026 | * direct-written, or the error code if that was zero. Note |
| 4027 | * that this differs from normal direct-io semantics, which |
| 4028 | * will return -EFOO even if some bytes were written. |
| 4029 | */ |
Al Viro | 60bb452 | 2014-08-08 12:39:16 -0400 | [diff] [blame] | 4030 | if (unlikely(status < 0)) { |
Al Viro | 3b93f91 | 2014-02-11 21:34:08 -0500 | [diff] [blame] | 4031 | err = status; |
Jeff Moyer | fb5527e | 2006-10-19 23:28:13 -0700 | [diff] [blame] | 4032 | goto out; |
| 4033 | } |
Jeff Moyer | fb5527e | 2006-10-19 23:28:13 -0700 | [diff] [blame] | 4034 | /* |
| 4035 | * We need to ensure that the page cache pages are written to |
| 4036 | * disk and invalidated to preserve the expected O_DIRECT |
| 4037 | * semantics. |
| 4038 | */ |
Al Viro | 3b93f91 | 2014-02-11 21:34:08 -0500 | [diff] [blame] | 4039 | endbyte = pos + status - 1; |
Al Viro | 0b8def9 | 2015-04-07 10:22:53 -0400 | [diff] [blame] | 4040 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
Jeff Moyer | fb5527e | 2006-10-19 23:28:13 -0700 | [diff] [blame] | 4041 | if (err == 0) { |
Al Viro | 0b8def9 | 2015-04-07 10:22:53 -0400 | [diff] [blame] | 4042 | iocb->ki_pos = endbyte + 1; |
Al Viro | 3b93f91 | 2014-02-11 21:34:08 -0500 | [diff] [blame] | 4043 | written += status; |
Jeff Moyer | fb5527e | 2006-10-19 23:28:13 -0700 | [diff] [blame] | 4044 | invalidate_mapping_pages(mapping, |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 4045 | pos >> PAGE_SHIFT, |
| 4046 | endbyte >> PAGE_SHIFT); |
Jeff Moyer | fb5527e | 2006-10-19 23:28:13 -0700 | [diff] [blame] | 4047 | } else { |
| 4048 | /* |
| 4049 | * We don't know how much we wrote, so just return |
| 4050 | * the number of bytes which were direct-written |
| 4051 | */ |
| 4052 | } |
| 4053 | } else { |
Matthew Wilcox (Oracle) | 800ba29 | 2022-02-19 23:19:49 -0500 | [diff] [blame] | 4054 | written = generic_perform_write(iocb, from); |
Al Viro | 0b8def9 | 2015-04-07 10:22:53 -0400 | [diff] [blame] | 4055 | if (likely(written > 0)) |
| 4056 | iocb->ki_pos += written; |
Jeff Moyer | fb5527e | 2006-10-19 23:28:13 -0700 | [diff] [blame] | 4057 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4058 | out: |
| 4059 | current->backing_dev_info = NULL; |
| 4060 | return written ? written : err; |
| 4061 | } |
Al Viro | 8174202 | 2014-04-03 03:17:43 -0400 | [diff] [blame] | 4062 | EXPORT_SYMBOL(__generic_file_write_iter); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4063 | |
Jan Kara | e4dd9de | 2009-08-17 18:10:06 +0200 | [diff] [blame] | 4064 | /** |
Al Viro | 8174202 | 2014-04-03 03:17:43 -0400 | [diff] [blame] | 4065 | * generic_file_write_iter - write data to a file |
Jan Kara | e4dd9de | 2009-08-17 18:10:06 +0200 | [diff] [blame] | 4066 | * @iocb: IO state structure |
Al Viro | 8174202 | 2014-04-03 03:17:43 -0400 | [diff] [blame] | 4067 | * @from: iov_iter with data to write |
Jan Kara | e4dd9de | 2009-08-17 18:10:06 +0200 | [diff] [blame] | 4068 | * |
Al Viro | 8174202 | 2014-04-03 03:17:43 -0400 | [diff] [blame] | 4069 | * This is a wrapper around __generic_file_write_iter() to be used by most |
Jan Kara | e4dd9de | 2009-08-17 18:10:06 +0200 | [diff] [blame] | 4070 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
Jan Kara | 9608703 | 2021-04-12 15:50:21 +0200 | [diff] [blame] | 4071 | * and acquires i_rwsem as needed. |
Mike Rapoport | a862f68 | 2019-03-05 15:48:42 -0800 | [diff] [blame] | 4072 | * Return: |
| 4073 | * * negative error code if no data has been written at all of |
| 4074 | * vfs_fsync_range() failed for a synchronous write |
| 4075 | * * number of bytes written, even for truncated writes |
Jan Kara | e4dd9de | 2009-08-17 18:10:06 +0200 | [diff] [blame] | 4076 | */ |
Al Viro | 8174202 | 2014-04-03 03:17:43 -0400 | [diff] [blame] | 4077 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4078 | { |
| 4079 | struct file *file = iocb->ki_filp; |
Jan Kara | 148f948 | 2009-08-17 19:52:36 +0200 | [diff] [blame] | 4080 | struct inode *inode = file->f_mapping->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4081 | ssize_t ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4082 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 4083 | inode_lock(inode); |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 4084 | ret = generic_write_checks(iocb, from); |
| 4085 | if (ret > 0) |
Al Viro | 5f380c7 | 2015-04-07 11:28:12 -0400 | [diff] [blame] | 4086 | ret = __generic_file_write_iter(iocb, from); |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 4087 | inode_unlock(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4088 | |
Christoph Hellwig | e259221 | 2016-04-07 08:52:01 -0700 | [diff] [blame] | 4089 | if (ret > 0) |
| 4090 | ret = generic_write_sync(iocb, ret); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4091 | return ret; |
| 4092 | } |
Al Viro | 8174202 | 2014-04-03 03:17:43 -0400 | [diff] [blame] | 4093 | EXPORT_SYMBOL(generic_file_write_iter); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4094 | |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 4095 | /** |
Matthew Wilcox (Oracle) | 82c50f8 | 2021-07-28 15:14:48 -0400 | [diff] [blame] | 4096 | * filemap_release_folio() - Release fs-specific metadata on a folio. |
| 4097 | * @folio: The folio which the kernel is trying to free. |
| 4098 | * @gfp: Memory allocation flags (and I/O mode). |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 4099 | * |
Matthew Wilcox (Oracle) | 82c50f8 | 2021-07-28 15:14:48 -0400 | [diff] [blame] | 4100 | * The address_space is trying to release any data attached to a folio |
| 4101 | * (presumably at folio->private). |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 4102 | * |
Matthew Wilcox (Oracle) | 82c50f8 | 2021-07-28 15:14:48 -0400 | [diff] [blame] | 4103 | * This will also be called if the private_2 flag is set on a page, |
| 4104 | * indicating that the folio has other metadata associated with it. |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 4105 | * |
Matthew Wilcox (Oracle) | 82c50f8 | 2021-07-28 15:14:48 -0400 | [diff] [blame] | 4106 | * The @gfp argument specifies whether I/O may be performed to release |
| 4107 | * this page (__GFP_IO), and whether the call may block |
| 4108 | * (__GFP_RECLAIM & __GFP_FS). |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 4109 | * |
Matthew Wilcox (Oracle) | 82c50f8 | 2021-07-28 15:14:48 -0400 | [diff] [blame] | 4110 | * Return: %true if the release was successful, otherwise %false. |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 4111 | */ |
Matthew Wilcox (Oracle) | 82c50f8 | 2021-07-28 15:14:48 -0400 | [diff] [blame] | 4112 | bool filemap_release_folio(struct folio *folio, gfp_t gfp) |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 4113 | { |
Matthew Wilcox (Oracle) | 82c50f8 | 2021-07-28 15:14:48 -0400 | [diff] [blame] | 4114 | struct address_space * const mapping = folio->mapping; |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 4115 | |
Matthew Wilcox (Oracle) | 82c50f8 | 2021-07-28 15:14:48 -0400 | [diff] [blame] | 4116 | BUG_ON(!folio_test_locked(folio)); |
| 4117 | if (folio_test_writeback(folio)) |
| 4118 | return false; |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 4119 | |
Matthew Wilcox (Oracle) | fa29000 | 2022-04-29 17:00:05 -0400 | [diff] [blame] | 4120 | if (mapping && mapping->a_ops->release_folio) |
| 4121 | return mapping->a_ops->release_folio(folio, gfp); |
Matthew Wilcox (Oracle) | 68189fe | 2022-05-01 01:08:08 -0400 | [diff] [blame] | 4122 | return try_to_free_buffers(folio); |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 4123 | } |
Matthew Wilcox (Oracle) | 82c50f8 | 2021-07-28 15:14:48 -0400 | [diff] [blame] | 4124 | EXPORT_SYMBOL(filemap_release_folio); |