blob: 2723104cc06a12477a002fff87fe7fe7ec8ccd66 [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
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 Gortmakerb95f1b312011-10-16 02:01:52 -040013#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/compiler.h>
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080015#include <linux/dax.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/fs.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010017#include <linux/sched/signal.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070018#include <linux/uaccess.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080019#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090021#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/mm.h>
23#include <linux/swap.h>
Alistair Poppleffa65752022-01-21 22:10:46 -080024#include <linux/swapops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/mman.h>
26#include <linux/pagemap.h>
27#include <linux/file.h>
28#include <linux/uio.h>
Josef Bacikcfcbfb12019-05-13 17:21:04 -070029#include <linux/error-injection.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/hash.h>
31#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070032#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/security.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080035#include <linux/cpuset.h>
Johannes Weiner00501b52014-08-08 14:19:20 -070036#include <linux/hugetlb.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080037#include <linux/memcontrol.h>
Mel Gormanc7df8ad2017-11-15 17:37:41 -080038#include <linux/shmem_fs.h>
Kirill A. Shutemovf1820362014-04-07 15:37:19 -070039#include <linux/rmap.h>
Johannes Weinerb1d29ba2018-10-26 15:06:08 -070040#include <linux/delayacct.h>
Johannes Weinereb414682018-10-26 15:06:27 -070041#include <linux/psi.h>
Ben Dooksd0e6a582019-10-18 20:20:20 -070042#include <linux/ramfs.h>
Yang Shib9306a72020-08-06 23:19:55 -070043#include <linux/page_idle.h>
Alistair Poppleffa65752022-01-21 22:10:46 -080044#include <linux/migrate.h>
David Howells07073eb2023-02-14 15:01:42 +000045#include <linux/pipe_fs_i.h>
46#include <linux/splice.h>
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +030047#include <asm/pgalloc.h>
Will Deaconde591a82021-02-10 11:15:11 +000048#include <asm/tlbflush.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080049#include "internal.h"
50
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070051#define CREATE_TRACE_POINTS
52#include <trace/events/filemap.h>
53
Linus Torvalds1da177e2005-04-16 15:20:36 -070054/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 * FIXME: remove all knowledge of the buffer layer from the core VM
56 */
Jan Kara148f9482009-08-17 19:52:36 +020057#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#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 Buesoc8c06ef2014-12-12 16:54:24 -080076 * ->i_mmap_rwsem (truncate_pagecache)
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +000077 * ->private_lock (__free_pte->block_dirty_folio)
Hugh Dickins5d337b92005-09-03 15:54:41 -070078 * ->swap_lock (exclusive_swap_page, others)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070079 * ->i_pages lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 *
Jan Kara96087032021-04-12 15:50:21 +020081 * ->i_rwsem
Jan Kara730633f2021-01-28 19:19:45 +010082 * ->invalidate_lock (acquired by fs in truncate path)
83 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -070085 * ->mmap_lock
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080086 * ->i_mmap_rwsem
Hugh Dickinsb8072f02005-10-29 18:16:41 -070087 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070088 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -070090 * ->mmap_lock
Jan Kara730633f2021-01-28 19:19:45 +010091 * ->invalidate_lock (filemap_fault)
92 * ->lock_page (filemap_fault, access_process_vm)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 *
Jan Kara96087032021-04-12 15:50:21 +020094 * ->i_rwsem (generic_perform_write)
Andreas Gruenbacherbb523b42021-08-02 13:44:20 +020095 * ->mmap_lock (fault_in_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060097 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110098 * sb_lock (fs/fs-writeback.c)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070099 * ->i_pages lock (__sync_single_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800101 * ->i_mmap_rwsem
Liam R. Howlett0503ea82023-01-20 11:26:49 -0500102 * ->anon_vma.lock (vma_merge)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 *
104 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700105 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700107 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -0700108 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 * ->private_lock (try_to_unmap_one)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700110 * ->i_pages lock (try_to_unmap_one)
Hugh Dickins15b44732020-12-15 14:21:31 -0800111 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
112 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 * ->private_lock (page_remove_rmap->set_page_dirty)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700114 * ->i_pages lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600115 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100116 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700117 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600118 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100119 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +0000120 * ->private_lock (zap_pte_range->block_dirty_folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800122 * ->i_mmap_rwsem
Andi Kleen9a3c5312012-03-21 16:34:09 -0700123 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 */
125
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500126static void page_cache_delete(struct address_space *mapping,
Matthew Wilcox (Oracle)a548b612021-05-08 00:35:49 -0400127 struct folio *folio, void *shadow)
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700128{
Matthew Wilcox (Oracle)a548b612021-05-08 00:35:49 -0400129 XA_STATE(xas, &mapping->i_pages, folio->index);
130 long nr = 1;
Kirill A. Shutemovc70b6472016-12-12 16:43:17 -0800131
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500132 mapping_set_update(&xas, mapping);
133
134 /* hugetlb pages are represented by a single entry in the xarray */
Matthew Wilcox (Oracle)a548b612021-05-08 00:35:49 -0400135 if (!folio_test_hugetlb(folio)) {
136 xas_set_order(&xas, folio->index, folio_order(folio));
137 nr = folio_nr_pages(folio);
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500138 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700139
Matthew Wilcox (Oracle)a548b612021-05-08 00:35:49 -0400140 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
Johannes Weiner449dd692014-04-03 14:47:56 -0700141
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500142 xas_store(&xas, shadow);
143 xas_init_marks(&xas);
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200144
Matthew Wilcox (Oracle)a548b612021-05-08 00:35:49 -0400145 folio->mapping = NULL;
Jan Kara23006382017-11-15 17:37:26 -0800146 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200147 mapping->nrpages -= nr;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700148}
149
Matthew Wilcox (Oracle)621db482021-05-08 20:04:05 -0400150static void filemap_unaccount_folio(struct address_space *mapping,
151 struct folio *folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152{
Matthew Wilcox (Oracle)621db482021-05-08 20:04:05 -0400153 long nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
Matthew Wilcox (Oracle)621db482021-05-08 20:04:05 -0400155 VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
156 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
Hugh Dickins06b241f2016-03-09 14:08:07 -0800157 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
Matthew Wilcox (Oracle)621db482021-05-08 20:04:05 -0400158 current->comm, folio_pfn(folio));
159 dump_page(&folio->page, "still mapped when deleted");
Hugh Dickins06b241f2016-03-09 14:08:07 -0800160 dump_stack();
161 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
162
Hugh Dickins85207ad2022-03-24 18:09:52 -0700163 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 Dickins06b241f2016-03-09 14:08:07 -0800176 }
177 }
178
Matthew Wilcox (Oracle)621db482021-05-08 20:04:05 -0400179 /* hugetlb folios do not participate in page cache accounting. */
180 if (folio_test_hugetlb(folio))
Jan Kara5ecc4d82017-11-15 17:37:29 -0800181 return;
Jan Kara76253fb2017-11-15 17:37:22 -0800182
Matthew Wilcox (Oracle)621db482021-05-08 20:04:05 -0400183 nr = folio_nr_pages(folio);
Jan Kara5ecc4d82017-11-15 17:37:29 -0800184
Matthew Wilcox (Oracle)621db482021-05-08 20:04:05 -0400185 __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 Liu09d91cd2019-09-23 15:38:03 -0700192 filemap_nr_thps_dec(mapping);
Jan Kara76253fb2017-11-15 17:37:22 -0800193 }
Jan Kara5ecc4d82017-11-15 17:37:29 -0800194
195 /*
Matthew Wilcox (Oracle)621db482021-05-08 20:04:05 -0400196 * At this point folio must be either written or cleaned by
197 * truncate. Dirty folio here signals a bug and loss of
Hugh Dickins566d3362022-03-24 18:13:59 -0700198 * unwritten data - on ordinary filesystems.
Jan Kara5ecc4d82017-11-15 17:37:29 -0800199 *
Hugh Dickins566d3362022-03-24 18:13:59 -0700200 * 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)621db482021-05-08 20:04:05 -0400205 * but leaves the dirty flag set: it has no effect for truncated
206 * folio and anyway will be cleared before returning folio to
Jan Kara5ecc4d82017-11-15 17:37:29 -0800207 * buddy allocator.
208 */
Hugh Dickins566d3362022-03-24 18:13:59 -0700209 if (WARN_ON_ONCE(folio_test_dirty(folio) &&
210 mapping_can_writeback(mapping)))
211 folio_account_cleaned(folio, inode_to_wb(mapping->host));
Jan Kara5ecc4d82017-11-15 17:37:29 -0800212}
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 Wilcoxb93b0162018-04-10 16:36:56 -0700217 * is safe. The caller must hold the i_pages lock.
Jan Kara5ecc4d82017-11-15 17:37:29 -0800218 */
Matthew Wilcox (Oracle)452e9e62021-05-09 09:33:42 -0400219void __filemap_remove_folio(struct folio *folio, void *shadow)
Jan Kara5ecc4d82017-11-15 17:37:29 -0800220{
Matthew Wilcox (Oracle)452e9e62021-05-09 09:33:42 -0400221 struct address_space *mapping = folio->mapping;
Jan Kara5ecc4d82017-11-15 17:37:29 -0800222
Matthew Wilcox (Oracle)a0580c62021-07-23 09:29:46 -0400223 trace_mm_filemap_delete_from_page_cache(folio);
Matthew Wilcox (Oracle)621db482021-05-08 20:04:05 -0400224 filemap_unaccount_folio(mapping, folio);
Matthew Wilcox (Oracle)a548b612021-05-08 00:35:49 -0400225 page_cache_delete(mapping, folio, shadow);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
227
Matthew Wilcox (Oracle)78f42662021-07-28 15:52:34 -0400228void filemap_free_folio(struct address_space *mapping, struct folio *folio)
Jan Kara59c66c52017-11-15 17:37:18 -0800229{
Matthew Wilcox (Oracle)d2329aa2022-05-01 07:35:31 -0400230 void (*free_folio)(struct folio *);
Matthew Wilcox (Oracle)3abb28e2022-01-07 13:03:48 -0500231 int refs = 1;
Jan Kara59c66c52017-11-15 17:37:18 -0800232
Matthew Wilcox (Oracle)d2329aa2022-05-01 07:35:31 -0400233 free_folio = mapping->a_ops->free_folio;
234 if (free_folio)
235 free_folio(folio);
Jan Kara59c66c52017-11-15 17:37:18 -0800236
Matthew Wilcox (Oracle)3abb28e2022-01-07 13:03:48 -0500237 if (folio_test_large(folio) && !folio_test_hugetlb(folio))
238 refs = folio_nr_pages(folio);
239 folio_put_refs(folio, refs);
Jan Kara59c66c52017-11-15 17:37:18 -0800240}
241
Minchan Kim702cfbf2011-03-22 16:32:43 -0700242/**
Matthew Wilcox (Oracle)452e9e62021-05-09 09:33:42 -0400243 * filemap_remove_folio - Remove folio from page cache.
244 * @folio: The folio.
Minchan Kim702cfbf2011-03-22 16:32:43 -0700245 *
Matthew Wilcox (Oracle)452e9e62021-05-09 09:33:42 -0400246 * 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 Kim702cfbf2011-03-22 16:32:43 -0700249 */
Matthew Wilcox (Oracle)452e9e62021-05-09 09:33:42 -0400250void filemap_remove_folio(struct folio *folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
Matthew Wilcox (Oracle)452e9e62021-05-09 09:33:42 -0400252 struct address_space *mapping = folio->mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Matthew Wilcox (Oracle)452e9e62021-05-09 09:33:42 -0400254 BUG_ON(!folio_test_locked(folio));
Johannes Weiner51b8c1f2021-11-08 18:31:24 -0800255 spin_lock(&mapping->host->i_lock);
Johannes Weiner30472502021-09-02 14:53:18 -0700256 xa_lock_irq(&mapping->i_pages);
Matthew Wilcox (Oracle)452e9e62021-05-09 09:33:42 -0400257 __filemap_remove_folio(folio, NULL);
Johannes Weiner30472502021-09-02 14:53:18 -0700258 xa_unlock_irq(&mapping->i_pages);
Johannes Weiner51b8c1f2021-11-08 18:31:24 -0800259 if (mapping_shrinkable(mapping))
260 inode_add_lru(mapping->host);
261 spin_unlock(&mapping->host->i_lock);
Linus Torvalds6072d132010-12-01 13:35:19 -0500262
Matthew Wilcox (Oracle)452e9e62021-05-09 09:33:42 -0400263 filemap_free_folio(mapping, folio);
Minchan Kim97cecb52011-03-22 16:30:53 -0700264}
Minchan Kim97cecb52011-03-22 16:30:53 -0700265
Jan Karaaa65c292017-11-15 17:37:33 -0800266/*
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500267 * 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 Karaaa65c292017-11-15 17:37:33 -0800270 *
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500271 * 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 Karaaa65c292017-11-15 17:37:33 -0800276 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700277 * The function expects the i_pages lock to be held.
Jan Karaaa65c292017-11-15 17:37:33 -0800278 */
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500279static void page_cache_delete_batch(struct address_space *mapping,
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500280 struct folio_batch *fbatch)
Jan Karaaa65c292017-11-15 17:37:33 -0800281{
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500282 XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
Matthew Wilcox (Oracle)6b24ca42020-06-27 22:19:08 -0400283 long total_pages = 0;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700284 int i = 0;
Matthew Wilcox (Oracle)1afd7ae2021-03-12 23:13:46 -0500285 struct folio *folio;
Jan Karaaa65c292017-11-15 17:37:33 -0800286
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500287 mapping_set_update(&xas, mapping);
Matthew Wilcox (Oracle)1afd7ae2021-03-12 23:13:46 -0500288 xas_for_each(&xas, folio, ULONG_MAX) {
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500289 if (i >= folio_batch_count(fbatch))
Jan Karaaa65c292017-11-15 17:37:33 -0800290 break;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700291
292 /* A swap/dax/shadow entry got inserted? Skip it. */
Matthew Wilcox (Oracle)1afd7ae2021-03-12 23:13:46 -0500293 if (xa_is_value(folio))
Jan Karaaa65c292017-11-15 17:37:33 -0800294 continue;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700295 /*
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)51dcbda2021-12-07 14:15:07 -0500302 if (folio != fbatch->folios[i]) {
Matthew Wilcox (Oracle)1afd7ae2021-03-12 23:13:46 -0500303 VM_BUG_ON_FOLIO(folio->index >
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500304 fbatch->folios[i]->index, folio);
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700305 continue;
Linus Torvalds69bf4b62019-07-05 19:55:18 -0700306 }
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700307
Matthew Wilcox (Oracle)1afd7ae2021-03-12 23:13:46 -0500308 WARN_ON_ONCE(!folio_test_locked(folio));
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700309
Matthew Wilcox (Oracle)6b24ca42020-06-27 22:19:08 -0400310 folio->mapping = NULL;
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500311 /* Leave folio->index set: truncation lookup relies on it */
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700312
Matthew Wilcox (Oracle)6b24ca42020-06-27 22:19:08 -0400313 i++;
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500314 xas_store(&xas, NULL);
Matthew Wilcox (Oracle)6b24ca42020-06-27 22:19:08 -0400315 total_pages += folio_nr_pages(folio);
Jan Karaaa65c292017-11-15 17:37:33 -0800316 }
317 mapping->nrpages -= total_pages;
318}
319
320void delete_from_page_cache_batch(struct address_space *mapping,
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500321 struct folio_batch *fbatch)
Jan Karaaa65c292017-11-15 17:37:33 -0800322{
323 int i;
Jan Karaaa65c292017-11-15 17:37:33 -0800324
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500325 if (!folio_batch_count(fbatch))
Jan Karaaa65c292017-11-15 17:37:33 -0800326 return;
327
Johannes Weiner51b8c1f2021-11-08 18:31:24 -0800328 spin_lock(&mapping->host->i_lock);
Johannes Weiner30472502021-09-02 14:53:18 -0700329 xa_lock_irq(&mapping->i_pages);
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500330 for (i = 0; i < folio_batch_count(fbatch); i++) {
331 struct folio *folio = fbatch->folios[i];
Jan Karaaa65c292017-11-15 17:37:33 -0800332
Matthew Wilcox (Oracle)a0580c62021-07-23 09:29:46 -0400333 trace_mm_filemap_delete_from_page_cache(folio);
334 filemap_unaccount_folio(mapping, folio);
Jan Karaaa65c292017-11-15 17:37:33 -0800335 }
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500336 page_cache_delete_batch(mapping, fbatch);
Johannes Weiner30472502021-09-02 14:53:18 -0700337 xa_unlock_irq(&mapping->i_pages);
Johannes Weiner51b8c1f2021-11-08 18:31:24 -0800338 if (mapping_shrinkable(mapping))
339 inode_add_lru(mapping->host);
340 spin_unlock(&mapping->host->i_lock);
Jan Karaaa65c292017-11-15 17:37:33 -0800341
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -0500342 for (i = 0; i < folio_batch_count(fbatch); i++)
343 filemap_free_folio(mapping, fbatch->folios[i]);
Jan Karaaa65c292017-11-15 17:37:33 -0800344}
345
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200346int filemap_check_errors(struct address_space *mapping)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700347{
348 int ret = 0;
349 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700350 if (test_bit(AS_ENOSPC, &mapping->flags) &&
351 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700352 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700353 if (test_bit(AS_EIO, &mapping->flags) &&
354 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700355 ret = -EIO;
356 return ret;
357}
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200358EXPORT_SYMBOL(filemap_check_errors);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700359
Jeff Layton76341ca2017-07-06 07:02:22 -0400360static 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 Torvalds1da177e2005-04-16 15:20:36 -0700370/**
Josef Bacik5a798492021-07-14 14:47:22 -0400371 * 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 */
380int 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}
394EXPORT_SYMBOL(filemap_fdatawrite_wbc);
395
396/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700397 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700398 * @mapping: address space structure to write
399 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800400 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700401 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700403 * Start writeback against all of a mapping's dirty pages that lie
404 * within the byte offsets <start, end> inclusive.
405 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700407 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 * these two operations is that if a dirty page/buffer is encountered, it must
409 * be waited upon, and not just skipped over.
Mike Rapoporta862f682019-03-05 15:48:42 -0800410 *
411 * Return: %0 on success, negative error code otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800413int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
414 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 struct writeback_control wbc = {
417 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800418 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700419 .range_start = start,
420 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 };
422
Josef Bacik5a798492021-07-14 14:47:22 -0400423 return filemap_fdatawrite_wbc(mapping, &wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425
426static inline int __filemap_fdatawrite(struct address_space *mapping,
427 int sync_mode)
428{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700429 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430}
431
432int filemap_fdatawrite(struct address_space *mapping)
433{
434 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
435}
436EXPORT_SYMBOL(filemap_fdatawrite);
437
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400438int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800439 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
441 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
442}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400443EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Randy Dunlap485bb992006-06-23 02:03:49 -0700445/**
446 * filemap_flush - mostly a non-blocking flush
447 * @mapping: target address_space
448 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 * 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 Rapoporta862f682019-03-05 15:48:42 -0800451 *
452 * Return: %0 on success, negative error code otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 */
454int filemap_flush(struct address_space *mapping)
455{
456 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
457}
458EXPORT_SYMBOL(filemap_flush);
459
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500460/**
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 Rapoporta862f682019-03-05 15:48:42 -0800468 *
469 * Return: %true if at least one page exists in the specified range,
470 * %false otherwise.
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500471 */
472bool filemap_range_has_page(struct address_space *mapping,
473 loff_t start_byte, loff_t end_byte)
474{
Matthew Wilcox (Oracle)eff3b362023-01-16 19:39:40 +0000475 struct folio *folio;
Matthew Wilcox8fa8e532018-01-16 06:26:49 -0500476 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
477 pgoff_t max = end_byte >> PAGE_SHIFT;
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500478
479 if (end_byte < start_byte)
480 return false;
481
Matthew Wilcox8fa8e532018-01-16 06:26:49 -0500482 rcu_read_lock();
483 for (;;) {
Matthew Wilcox (Oracle)eff3b362023-01-16 19:39:40 +0000484 folio = xas_find(&xas, max);
485 if (xas_retry(&xas, folio))
Matthew Wilcox8fa8e532018-01-16 06:26:49 -0500486 continue;
487 /* Shadow entries don't count */
Matthew Wilcox (Oracle)eff3b362023-01-16 19:39:40 +0000488 if (xa_is_value(folio))
Matthew Wilcox8fa8e532018-01-16 06:26:49 -0500489 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 Rodrigues7fc9e472017-06-20 07:05:41 -0500498
Matthew Wilcox (Oracle)eff3b362023-01-16 19:39:40 +0000499 return folio != NULL;
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500500}
501EXPORT_SYMBOL(filemap_range_has_page);
502
Jeff Layton5e8fcc12017-07-06 07:02:24 -0400503static void __filemap_fdatawait_range(struct address_space *mapping,
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800504 loff_t start_byte, loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300506 pgoff_t index = start_byte >> PAGE_SHIFT;
507 pgoff_t end = end_byte >> PAGE_SHIFT;
Vishal Moola (Oracle)6817ef52023-01-04 13:14:28 -0800508 struct folio_batch fbatch;
509 unsigned nr_folios;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Vishal Moola (Oracle)6817ef52023-01-04 13:14:28 -0800511 folio_batch_init(&fbatch);
512
Jan Kara312e9d22017-11-15 17:35:05 -0800513 while (index <= end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 unsigned i;
515
Vishal Moola (Oracle)6817ef52023-01-04 13:14:28 -0800516 nr_folios = filemap_get_folios_tag(mapping, &index, end,
517 PAGECACHE_TAG_WRITEBACK, &fbatch);
518
519 if (!nr_folios)
Jan Kara312e9d22017-11-15 17:35:05 -0800520 break;
521
Vishal Moola (Oracle)6817ef52023-01-04 13:14:28 -0800522 for (i = 0; i < nr_folios; i++) {
523 struct folio *folio = fbatch.folios[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Vishal Moola (Oracle)6817ef52023-01-04 13:14:28 -0800525 folio_wait_writeback(folio);
526 folio_clear_error(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 }
Vishal Moola (Oracle)6817ef52023-01-04 13:14:28 -0800528 folio_batch_release(&fbatch);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 cond_resched();
530 }
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800531}
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 Rapoporta862f682019-03-05 15:48:42 -0800546 *
547 * Return: error status of the address space.
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800548 */
549int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
550 loff_t end_byte)
551{
Jeff Layton5e8fcc12017-07-06 07:02:24 -0400552 __filemap_fdatawait_range(mapping, start_byte, end_byte);
553 return filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200555EXPORT_SYMBOL(filemap_fdatawait_range);
556
557/**
Ross Zwisleraa0bfcd2019-06-20 17:05:37 -0400558 * 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 */
571int 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}
577EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
578
579/**
Jeff Laytona823e452017-07-28 07:24:43 -0400580 * 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 Rapoporta862f682019-03-05 15:48:42 -0800592 *
593 * Return: error status of the address space vs. the file->f_wb_err cursor.
Jeff Laytona823e452017-07-28 07:24:43 -0400594 */
595int 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}
602EXPORT_SYMBOL(file_fdatawait_range);
603
604/**
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800605 * 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 Rapoporta862f682019-03-05 15:48:42 -0800615 *
616 * Return: error status of the address space.
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800617 */
Jeff Layton76341ca2017-07-06 07:02:22 -0400618int filemap_fdatawait_keep_errors(struct address_space *mapping)
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800619{
Jeff Laytonffb959b2017-07-31 10:29:38 -0400620 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
Jeff Layton76341ca2017-07-06 07:02:22 -0400621 return filemap_check_and_keep_errors(mapping);
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800622}
Jeff Layton76341ca2017-07-06 07:02:22 -0400623EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800624
Konstantin Khlebnikov875d91b2019-09-23 15:34:48 -0700625/* Returns true if writeback might be needed or already in progress. */
Jeff Layton9326c9b2017-07-26 10:21:11 -0400626static bool mapping_needs_writeback(struct address_space *mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627{
Konstantin Khlebnikov875d91b2019-09-23 15:34:48 -0700628 return mapping->nrpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630
Jens Axboe4bdcd1d2021-10-28 08:47:05 -0600631bool filemap_range_has_writeback(struct address_space *mapping,
632 loff_t start_byte, loff_t end_byte)
Jens Axboef8ee8902021-11-05 13:37:13 -0700633{
634 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
635 pgoff_t max = end_byte >> PAGE_SHIFT;
Vishal Moola (Oracle)b05f41a2022-09-05 14:45:57 -0700636 struct folio *folio;
Jens Axboef8ee8902021-11-05 13:37:13 -0700637
638 if (end_byte < start_byte)
639 return false;
640
641 rcu_read_lock();
Vishal Moola (Oracle)b05f41a2022-09-05 14:45:57 -0700642 xas_for_each(&xas, folio, max) {
643 if (xas_retry(&xas, folio))
Jens Axboef8ee8902021-11-05 13:37:13 -0700644 continue;
Vishal Moola (Oracle)b05f41a2022-09-05 14:45:57 -0700645 if (xa_is_value(folio))
Jens Axboef8ee8902021-11-05 13:37:13 -0700646 continue;
Vishal Moola (Oracle)b05f41a2022-09-05 14:45:57 -0700647 if (folio_test_dirty(folio) || folio_test_locked(folio) ||
648 folio_test_writeback(folio))
Jens Axboef8ee8902021-11-05 13:37:13 -0700649 break;
650 }
651 rcu_read_unlock();
Vishal Moola (Oracle)b05f41a2022-09-05 14:45:57 -0700652 return folio != NULL;
Jens Axboef8ee8902021-11-05 13:37:13 -0700653}
Jens Axboe4bdcd1d2021-10-28 08:47:05 -0600654EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
Jens Axboe63135aa2021-04-29 22:55:18 -0700655
656/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700657 * 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 Morton469eb4d2006-03-24 03:17:45 -0800662 * Write out and wait upon file offsets lstart->lend, inclusive.
663 *
mchehab@s-opensource.com0e056eb2017-03-30 17:11:36 -0300664 * Note that @lend is inclusive (describes the last byte to be written) so
Andrew Morton469eb4d2006-03-24 03:17:45 -0800665 * that this function can be used to write to the very end-of-file (end = -1).
Mike Rapoporta862f682019-03-05 15:48:42 -0800666 *
667 * Return: error status of the address space.
Andrew Morton469eb4d2006-03-24 03:17:45 -0800668 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669int filemap_write_and_wait_range(struct address_space *mapping,
670 loff_t lstart, loff_t lend)
671{
Miaohe Linccac11d2022-06-27 21:23:51 +0800672 int err = 0, err2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Brian Fosterfeeb9b22022-11-28 10:56:31 -0500674 if (lend < lstart)
675 return 0;
676
Jeff Layton9326c9b2017-07-26 10:21:11 -0400677 if (mapping_needs_writeback(mapping)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800678 err = __filemap_fdatawrite_range(mapping, lstart, lend,
679 WB_SYNC_ALL);
Ira Weinyddf8f372020-01-30 22:12:07 -0800680 /*
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 Linccac11d2022-06-27 21:23:51 +0800686 if (err != -EIO)
687 __filemap_fdatawait_range(mapping, lstart, lend);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 }
Miaohe Linccac11d2022-06-27 21:23:51 +0800689 err2 = filemap_check_errors(mapping);
690 if (!err)
691 err = err2;
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800692 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
Chris Masonf6995582009-04-15 13:22:37 -0400694EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
Jeff Layton5660e132017-07-06 07:02:25 -0400696void __filemap_set_wb_err(struct address_space *mapping, int err)
697{
Jeff Layton3acdfd22017-07-24 06:22:15 -0400698 errseq_t eseq = errseq_set(&mapping->wb_err, err);
Jeff Layton5660e132017-07-06 07:02:25 -0400699
700 trace_filemap_set_wb_err(mapping, eseq);
701}
702EXPORT_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 Rapoporta862f682019-03-05 15:48:42 -0800725 *
726 * Return: %0 on success, negative error code otherwise.
Jeff Layton5660e132017-07-06 07:02:25 -0400727 */
728int 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 Laytonf4e222c2017-10-03 16:15:25 -0700744
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 Layton5660e132017-07-06 07:02:25 -0400752 return err;
753}
754EXPORT_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 Rapoporta862f682019-03-05 15:48:42 -0800769 *
770 * Return: %0 on success, negative error code otherwise.
Jeff Layton5660e132017-07-06 07:02:25 -0400771 */
772int 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 Fosterfeeb9b22022-11-28 10:56:31 -0500777 if (lend < lstart)
778 return 0;
779
Jeff Layton9326c9b2017-07-26 10:21:11 -0400780 if (mapping_needs_writeback(mapping)) {
Jeff Layton5660e132017-07-06 07:02:25 -0400781 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}
792EXPORT_SYMBOL(file_write_and_wait_range);
793
Randy Dunlap485bb992006-06-23 02:03:49 -0700794/**
Vishal Moola (Oracle)3720dd62022-11-01 10:53:22 -0700795 * 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 Szeredief6a3c62011-03-22 16:30:52 -0700798 *
Vishal Moola (Oracle)3720dd62022-11-01 10:53:22 -0700799 * 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 Szeredief6a3c62011-03-22 16:30:52 -0700803 * caller must do that.
804 *
Matthew Wilcox74d60952017-11-17 10:01:45 -0500805 * The remove + add is atomic. This function cannot fail.
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700806 */
Vishal Moola (Oracle)3720dd62022-11-01 10:53:22 -0700807void replace_page_cache_folio(struct folio *old, struct folio *new)
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700808{
Matthew Wilcox74d60952017-11-17 10:01:45 -0500809 struct address_space *mapping = old->mapping;
Matthew Wilcox (Oracle)d2329aa2022-05-01 07:35:31 -0400810 void (*free_folio)(struct folio *) = mapping->a_ops->free_folio;
Matthew Wilcox74d60952017-11-17 10:01:45 -0500811 pgoff_t offset = old->index;
812 XA_STATE(xas, &mapping->i_pages, offset);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700813
Vishal Moola (Oracle)3720dd62022-11-01 10:53:22 -0700814 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 Szeredief6a3c62011-03-22 16:30:52 -0700817
Vishal Moola (Oracle)3720dd62022-11-01 10:53:22 -0700818 folio_get(new);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500819 new->mapping = mapping;
820 new->index = offset;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700821
Vishal Moola (Oracle)3720dd62022-11-01 10:53:22 -0700822 mem_cgroup_migrate(old, new);
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700823
Johannes Weiner30472502021-09-02 14:53:18 -0700824 xas_lock_irq(&xas);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500825 xas_store(&xas, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700826
Matthew Wilcox74d60952017-11-17 10:01:45 -0500827 old->mapping = NULL;
828 /* hugetlb pages do not participate in page cache accounting. */
Vishal Moola (Oracle)3720dd62022-11-01 10:53:22 -0700829 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 Weiner30472502021-09-02 14:53:18 -0700837 xas_unlock_irq(&xas);
Matthew Wilcox (Oracle)d2329aa2022-05-01 07:35:31 -0400838 if (free_folio)
Vishal Moola (Oracle)3720dd62022-11-01 10:53:22 -0700839 free_folio(old);
840 folio_put(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700841}
Vishal Moola (Oracle)3720dd62022-11-01 10:53:22 -0700842EXPORT_SYMBOL_GPL(replace_page_cache_folio);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700843
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500844noinline int __filemap_add_folio(struct address_space *mapping,
845 struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846{
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500847 XA_STATE(xas, &mapping->i_pages, index);
848 int huge = folio_test_hugetlb(folio);
Waiman Longda742402021-02-04 18:32:45 -0800849 bool charged = false;
Matthew Wilcox (Oracle)d68ecca2019-09-05 14:03:12 -0400850 long nr = 1;
Nick Piggine2867812008-07-25 19:45:30 -0700851
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500852 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
853 VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500854 mapping_set_update(&xas, mapping);
Nick Piggine2867812008-07-25 19:45:30 -0700855
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700856 if (!huge) {
Matthew Wilcox (Oracle)d68ecca2019-09-05 14:03:12 -0400857 int error = mem_cgroup_charge(folio, NULL, gfp);
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500858 VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700859 if (error)
Matthew Wilcox (Oracle)d68ecca2019-09-05 14:03:12 -0400860 return error;
Waiman Longda742402021-02-04 18:32:45 -0800861 charged = true;
Matthew Wilcox (Oracle)d68ecca2019-09-05 14:03:12 -0400862 xas_set_order(&xas, index, folio_order(folio));
863 nr = folio_nr_pages(folio);
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700864 }
865
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700866 gfp &= GFP_RECLAIM_MASK;
Matthew Wilcox (Oracle)d68ecca2019-09-05 14:03:12 -0400867 folio_ref_add(folio, nr);
868 folio->mapping = mapping;
869 folio->index = xas.xa_index;
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700870
Matthew Wilcox74d60952017-11-17 10:01:45 -0500871 do {
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700872 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
873 void *entry, *old = NULL;
874
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500875 if (order > folio_order(folio))
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700876 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
877 order, gfp);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500878 xas_lock_irq(&xas);
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700879 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)9dd3d062020-12-08 08:56:28 -0500892 if (order > folio_order(folio)) {
Matthew Wilcox (Oracle)d68ecca2019-09-05 14:03:12 -0400893 /* How to handle large swap entries? */
894 BUG_ON(shmem_mapping(mapping));
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700895 xas_split(&xas, old, order);
896 xas_reset(&xas);
897 }
898 }
899
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500900 xas_store(&xas, folio);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500901 if (xas_error(&xas))
902 goto unlock;
Michal Hocko4165b9b42015-06-24 16:57:24 -0700903
Matthew Wilcox (Oracle)d68ecca2019-09-05 14:03:12 -0400904 mapping->nrpages += nr;
Matthew Wilcox74d60952017-11-17 10:01:45 -0500905
906 /* hugetlb pages do not participate in page cache accounting */
Matthew Wilcox (Oracle)d68ecca2019-09-05 14:03:12 -0400907 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 Wilcox74d60952017-11-17 10:01:45 -0500913unlock:
914 xas_unlock_irq(&xas);
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700915 } while (xas_nomem(&xas, gfp));
Matthew Wilcox74d60952017-11-17 10:01:45 -0500916
Matthew Wilcox (Oracle)d68ecca2019-09-05 14:03:12 -0400917 if (xas_error(&xas))
Matthew Wilcox74d60952017-11-17 10:01:45 -0500918 goto error;
919
Matthew Wilcox (Oracle)a0580c62021-07-23 09:29:46 -0400920 trace_mm_filemap_add_to_page_cache(folio);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700921 return 0;
Matthew Wilcox74d60952017-11-17 10:01:45 -0500922error:
Matthew Wilcox (Oracle)d68ecca2019-09-05 14:03:12 -0400923 if (charged)
924 mem_cgroup_uncharge(folio);
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500925 folio->mapping = NULL;
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700926 /* Leave page->index set: truncation relies upon it */
Matthew Wilcox (Oracle)d68ecca2019-09-05 14:03:12 -0400927 folio_put_refs(folio, nr);
928 return xas_error(&xas);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929}
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500930ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
Johannes Weinera5289102014-04-03 14:47:51 -0700931
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500932int filemap_add_folio(struct address_space *mapping, struct folio *folio,
933 pgoff_t index, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934{
Johannes Weinera5289102014-04-03 14:47:51 -0700935 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700936 int ret;
937
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500938 __folio_set_locked(folio);
939 ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
Johannes Weinera5289102014-04-03 14:47:51 -0700940 if (unlikely(ret))
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500941 __folio_clear_locked(folio);
Johannes Weinera5289102014-04-03 14:47:51 -0700942 else {
943 /*
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500944 * The folio might have been evicted from cache only
Johannes Weinera5289102014-04-03 14:47:51 -0700945 * recently, in which case it should be activated like
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500946 * any other repeatedly accessed folio.
947 * The exception is folios getting rewritten; evicting other
Rik van Rielf0281a02016-05-20 16:56:25 -0700948 * data from the working set, only to cache data that will
949 * get overwritten with something else, is a waste of memory.
Johannes Weinera5289102014-04-03 14:47:51 -0700950 */
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500951 WARN_ON_ONCE(folio_test_active(folio));
952 if (!(gfp & __GFP_WRITE) && shadow)
953 workingset_refault(folio, shadow);
954 folio_add_lru(folio);
Johannes Weinera5289102014-04-03 14:47:51 -0700955 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 return ret;
957}
Matthew Wilcox (Oracle)9dd3d062020-12-08 08:56:28 -0500958EXPORT_SYMBOL_GPL(filemap_add_folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
Paul Jackson44110fe2006-03-24 03:16:04 -0800960#ifdef CONFIG_NUMA
Matthew Wilcox (Oracle)bb3c5792020-12-15 23:11:07 -0500961struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
Paul Jackson44110fe2006-03-24 03:16:04 -0800962{
Miao Xiec0ff7452010-05-24 14:32:08 -0700963 int n;
Matthew Wilcox (Oracle)bb3c5792020-12-15 23:11:07 -0500964 struct folio *folio;
Miao Xiec0ff7452010-05-24 14:32:08 -0700965
Paul Jackson44110fe2006-03-24 03:16:04 -0800966 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700967 unsigned int cpuset_mems_cookie;
968 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700969 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700970 n = cpuset_mem_spread_node();
Matthew Wilcox (Oracle)bb3c5792020-12-15 23:11:07 -0500971 folio = __folio_alloc_node(gfp, order, n);
972 } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700973
Matthew Wilcox (Oracle)bb3c5792020-12-15 23:11:07 -0500974 return folio;
Paul Jackson44110fe2006-03-24 03:16:04 -0800975 }
Matthew Wilcox (Oracle)bb3c5792020-12-15 23:11:07 -0500976 return folio_alloc(gfp, order);
Paul Jackson44110fe2006-03-24 03:16:04 -0800977}
Matthew Wilcox (Oracle)bb3c5792020-12-15 23:11:07 -0500978EXPORT_SYMBOL(filemap_alloc_folio);
Paul Jackson44110fe2006-03-24 03:16:04 -0800979#endif
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981/*
Jan Kara7506ae62021-05-24 13:02:30 +0200982 * 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 */
989void 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}
999EXPORT_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 */
1009void 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}
1017EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1018
1019/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 * 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 Piggin62906022016-12-25 13:00:30 +10001029#define PAGE_WAIT_TABLE_BITS 8
1030#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001031static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
Nicholas Piggin62906022016-12-25 13:00:30 +10001032
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001033static wait_queue_head_t *folio_waitqueue(struct folio *folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034{
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001035 return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036}
Nicholas Piggin62906022016-12-25 13:00:30 +10001037
1038void __init pagecache_init(void)
1039{
1040 int i;
1041
1042 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001043 init_waitqueue_head(&folio_wait_table[i]);
Nicholas Piggin62906022016-12-25 13:00:30 +10001044
1045 page_writeback_init();
1046}
1047
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001048/*
1049 * The page wait code treats the "wait->flags" somewhat unusually, because
Linus Torvalds5868ec22020-09-20 10:38:47 -07001050 * we have multiple different kinds of waits, not just the usual "exclusive"
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001051 * 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 Torvalds5868ec22020-09-20 10:38:47 -07001071 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001072 *
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 Molnarac6424b2017-06-20 12:06:13 +02001082static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
Nicholas Piggin62906022016-12-25 13:00:30 +10001083{
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001084 unsigned int flags;
Nicholas Piggin62906022016-12-25 13:00:30 +10001085 struct wait_page_key *key = arg;
1086 struct wait_page_queue *wait_page
1087 = container_of(wait, struct wait_page_queue, wait);
1088
Linus Torvaldscdc8fcb2020-08-03 13:01:22 -07001089 if (!wake_page_match(wait_page, key))
Nicholas Piggin62906022016-12-25 13:00:30 +10001090 return 0;
Linus Torvalds3510ca22017-08-27 13:55:12 -07001091
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001092 /*
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001093 * 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 Dickins9a1ea432018-12-28 00:36:14 -08001095 */
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001096 flags = wait->flags;
1097 if (flags & WQ_FLAG_EXCLUSIVE) {
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001098 if (test_bit(key->bit_nr, &key->folio->flags))
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001099 return -1;
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001100 if (flags & WQ_FLAG_CUSTOM) {
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001101 if (test_and_set_bit(key->bit_nr, &key->folio->flags))
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001102 return -1;
1103 flags |= WQ_FLAG_DONE;
1104 }
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001105 }
Nicholas Piggin62906022016-12-25 13:00:30 +10001106
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001107 /*
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)101c0bf2021-03-04 12:02:54 -05001114 * with the load-acquire in folio_wait_bit_common().
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001115 */
1116 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001117 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 Torvalds5ef64cc2020-09-13 14:05:35 -07001123 * 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 Torvalds2a9127f2020-07-23 10:16:49 -07001126 * might be de-allocated and the process might even have
1127 * exited.
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001128 */
Linus Torvaldsc6fe44d2020-07-23 12:33:41 -07001129 list_del_init_careful(&wait->entry);
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001130 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
Nicholas Piggin62906022016-12-25 13:00:30 +10001131}
1132
Matthew Wilcox (Oracle)6974d7c2021-01-15 17:14:48 -05001133static void folio_wake_bit(struct folio *folio, int bit_nr)
Nicholas Piggin62906022016-12-25 13:00:30 +10001134{
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001135 wait_queue_head_t *q = folio_waitqueue(folio);
Nicholas Piggin62906022016-12-25 13:00:30 +10001136 struct wait_page_key key;
1137 unsigned long flags;
Tim Chen11a19c72017-08-25 09:13:55 -07001138 wait_queue_entry_t bookmark;
Nicholas Piggin62906022016-12-25 13:00:30 +10001139
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001140 key.folio = folio;
Nicholas Piggin62906022016-12-25 13:00:30 +10001141 key.bit_nr = bit_nr;
1142 key.page_match = 0;
1143
Tim Chen11a19c72017-08-25 09:13:55 -07001144 bookmark.flags = 0;
1145 bookmark.private = NULL;
1146 bookmark.func = NULL;
1147 INIT_LIST_HEAD(&bookmark.entry);
1148
Nicholas Piggin62906022016-12-25 13:00:30 +10001149 spin_lock_irqsave(&q->lock, flags);
Tim Chen11a19c72017-08-25 09:13:55 -07001150 __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 Piggin62906022016-12-25 13:00:30 +10001165 /*
Hugh Dickinsbb43b142022-03-24 18:09:49 -07001166 * 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 Piggin62906022016-12-25 13:00:30 +10001169 *
Hugh Dickinsbb43b142022-03-24 18:09:49 -07001170 * 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 Piggin62906022016-12-25 13:00:30 +10001173 */
Hugh Dickinsbb43b142022-03-24 18:09:49 -07001174 if (!waitqueue_active(q) || !key.page_match)
Matthew Wilcox (Oracle)6974d7c2021-01-15 17:14:48 -05001175 folio_clear_waiters(folio);
Hugh Dickinsbb43b142022-03-24 18:09:49 -07001176
Nicholas Piggin62906022016-12-25 13:00:30 +10001177 spin_unlock_irqrestore(&q->lock, flags);
1178}
Nicholas Piggin74d81bf2017-02-22 15:44:41 -08001179
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001180static void folio_wake(struct folio *folio, int bit)
Nicholas Piggin74d81bf2017-02-22 15:44:41 -08001181{
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001182 if (!folio_test_waiters(folio))
Nicholas Piggin74d81bf2017-02-22 15:44:41 -08001183 return;
Matthew Wilcox (Oracle)6974d7c2021-01-15 17:14:48 -05001184 folio_wake_bit(folio, bit);
Nicholas Piggin74d81bf2017-02-22 15:44:41 -08001185}
Nicholas Piggin62906022016-12-25 13:00:30 +10001186
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001187/*
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001188 * A choice of three behaviors for folio_wait_bit_common():
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001189 */
1190enum behavior {
1191 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
Matthew Wilcox (Oracle)7c23c782021-03-01 19:38:25 -05001192 * __folio_lock() waiting on then setting PG_locked.
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001193 */
1194 SHARED, /* Hold ref to page and check the bit when woken, like
Matthew Wilcox (Oracle)9f2b04a2021-08-16 23:36:31 -04001195 * folio_wait_writeback() waiting on PG_writeback.
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001196 */
1197 DROP, /* Drop ref to page before wait, no check when woken,
Matthew Wilcox (Oracle)9f2b04a2021-08-16 23:36:31 -04001198 * like folio_put_wait_locked() on PG_locked.
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001199 */
1200};
1201
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001202/*
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001203 * Attempt to check (or get) the folio flag, and mark us done
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001204 * if successful.
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001205 */
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001206static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001207 struct wait_queue_entry *wait)
1208{
1209 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001210 if (test_and_set_bit(bit_nr, &folio->flags))
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001211 return false;
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001212 } else if (test_bit(bit_nr, &folio->flags))
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001213 return false;
1214
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001215 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001216 return true;
1217}
1218
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001219/* How many times do we accept lock stealing from under a waiter? */
1220int sysctl_page_lock_unfairness = 5;
1221
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001222static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1223 int state, enum behavior behavior)
Nicholas Piggin62906022016-12-25 13:00:30 +10001224{
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001225 wait_queue_head_t *q = folio_waitqueue(folio);
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001226 int unfairness = sysctl_page_lock_unfairness;
Nicholas Piggin62906022016-12-25 13:00:30 +10001227 struct wait_page_queue wait_page;
Ingo Molnarac6424b2017-06-20 12:06:13 +02001228 wait_queue_entry_t *wait = &wait_page.wait;
Johannes Weinerb1d29ba2018-10-26 15:06:08 -07001229 bool thrashing = false;
Johannes Weinereb414682018-10-26 15:06:27 -07001230 unsigned long pflags;
Yang Yangaa1cf992022-08-15 07:11:35 +00001231 bool in_thrashing;
Nicholas Piggin62906022016-12-25 13:00:30 +10001232
Johannes Weinereb414682018-10-26 15:06:27 -07001233 if (bit_nr == PG_locked &&
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001234 !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
Yang Yangaa1cf992022-08-15 07:11:35 +00001235 delayacct_thrashing_start(&in_thrashing);
Johannes Weinereb414682018-10-26 15:06:27 -07001236 psi_memstall_enter(&pflags);
Johannes Weinerb1d29ba2018-10-26 15:06:08 -07001237 thrashing = true;
1238 }
1239
Nicholas Piggin62906022016-12-25 13:00:30 +10001240 init_wait(wait);
1241 wait->func = wake_page_function;
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001242 wait_page.folio = folio;
Nicholas Piggin62906022016-12-25 13:00:30 +10001243 wait_page.bit_nr = bit_nr;
1244
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001245repeat:
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 Torvalds2a9127f2020-07-23 10:16:49 -07001253 /*
1254 * Do one last check whether we can get the
1255 * page bit synchronously.
1256 *
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001257 * Do the folio_set_waiters() marking before that
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001258 * 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)101c0bf2021-03-04 12:02:54 -05001268 folio_set_waiters(folio);
1269 if (!folio_trylock_flag(folio, bit_nr, wait))
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001270 __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 Torvalds5ef64cc2020-09-13 14:05:35 -07001275 * 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 Torvalds2a9127f2020-07-23 10:16:49 -07001278 *
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001279 * We can drop our reference to the folio.
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001280 */
1281 if (behavior == DROP)
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001282 folio_put(folio);
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001283
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001284 /*
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 Piggin62906022016-12-25 13:00:30 +10001290 for (;;) {
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001291 unsigned int flags;
1292
Nicholas Piggin62906022016-12-25 13:00:30 +10001293 set_current_state(state);
1294
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001295 /* 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 Torvaldsa8b169a2017-08-27 16:25:09 -07001307 break;
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001308
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001309 /* If the waker got the lock for us, we're done */
1310 if (flags & WQ_FLAG_DONE)
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001311 break;
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001312
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001313 /*
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)101c0bf2021-03-04 12:02:54 -05001319 if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001320 goto repeat;
1321
1322 wait->flags |= WQ_FLAG_DONE;
1323 break;
Nicholas Piggin62906022016-12-25 13:00:30 +10001324 }
1325
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001326 /*
1327 * If a signal happened, this 'finish_wait()' may remove the last
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001328 * waiter from the wait-queues, but the folio waiters bit will remain
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001329 * 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 Piggin62906022016-12-25 13:00:30 +10001332 finish_wait(q, wait);
1333
Johannes Weinereb414682018-10-26 15:06:27 -07001334 if (thrashing) {
Yang Yangaa1cf992022-08-15 07:11:35 +00001335 delayacct_thrashing_end(&in_thrashing);
Johannes Weinereb414682018-10-26 15:06:27 -07001336 psi_memstall_leave(&pflags);
1337 }
Johannes Weinerb1d29ba2018-10-26 15:06:08 -07001338
Nicholas Piggin62906022016-12-25 13:00:30 +10001339 /*
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001340 * 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 Piggin62906022016-12-25 13:00:30 +10001351 */
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001352 if (behavior == EXCLUSIVE)
1353 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
Nicholas Piggin62906022016-12-25 13:00:30 +10001354
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001355 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
Nicholas Piggin62906022016-12-25 13:00:30 +10001356}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
Alistair Poppleffa65752022-01-21 22:10:46 -08001358#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 */
1377void 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 Poppleffa65752022-01-21 22:10:46 -08001383 unsigned long pflags;
Yang Yangaa1cf992022-08-15 07:11:35 +00001384 bool in_thrashing;
Alistair Poppleffa65752022-01-21 22:10:46 -08001385 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 Yangaa1cf992022-08-15 07:11:35 +00001390 delayacct_thrashing_start(&in_thrashing);
Alistair Poppleffa65752022-01-21 22:10:46 -08001391 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 Yangaa1cf992022-08-15 07:11:35 +00001437 delayacct_thrashing_end(&in_thrashing);
Alistair Poppleffa65752022-01-21 22:10:46 -08001438 psi_memstall_leave(&pflags);
1439 }
1440}
1441#endif
1442
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001443void folio_wait_bit(struct folio *folio, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444{
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001445 folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446}
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001447EXPORT_SYMBOL(folio_wait_bit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001449int folio_wait_bit_killable(struct folio *folio, int bit_nr)
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -07001450{
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001451 return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -07001452}
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001453EXPORT_SYMBOL(folio_wait_bit_killable);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -07001454
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455/**
Matthew Wilcox (Oracle)9f2b04a2021-08-16 23:36:31 -04001456 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1457 * @folio: The folio to wait for.
Matthew Wilcox (Oracle)48054622021-02-24 12:02:02 -08001458 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001459 *
Matthew Wilcox (Oracle)9f2b04a2021-08-16 23:36:31 -04001460 * The caller should hold a reference on @folio. They expect the page to
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001461 * become unlocked relatively soon, but do not wish to hold up migration
Matthew Wilcox (Oracle)9f2b04a2021-08-16 23:36:31 -04001462 * (for example) by holding the reference while waiting for the folio to
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001463 * come unlocked. After this function returns, the caller should not
Matthew Wilcox (Oracle)9f2b04a2021-08-16 23:36:31 -04001464 * dereference @folio.
Matthew Wilcox (Oracle)48054622021-02-24 12:02:02 -08001465 *
Matthew Wilcox (Oracle)9f2b04a2021-08-16 23:36:31 -04001466 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001467 */
Ke Sunc195c322022-09-14 10:17:38 +08001468static int folio_put_wait_locked(struct folio *folio, int state)
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001469{
Matthew Wilcox (Oracle)9f2b04a2021-08-16 23:36:31 -04001470 return folio_wait_bit_common(folio, PG_locked, state, DROP);
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001471}
1472
1473/**
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001474 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1475 * @folio: Folio defining the wait queue of interest
Randy Dunlap697f6192009-04-13 14:39:54 -07001476 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +01001477 *
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001478 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
David Howells385e1ca5f2009-04-03 16:42:39 +01001479 */
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001480void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
David Howells385e1ca5f2009-04-03 16:42:39 +01001481{
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001482 wait_queue_head_t *q = folio_waitqueue(folio);
David Howells385e1ca5f2009-04-03 16:42:39 +01001483 unsigned long flags;
1484
1485 spin_lock_irqsave(&q->lock, flags);
Linus Torvalds9c3a8152017-08-28 16:45:40 -07001486 __add_wait_queue_entry_tail(q, waiter);
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001487 folio_set_waiters(folio);
David Howells385e1ca5f2009-04-03 16:42:39 +01001488 spin_unlock_irqrestore(&q->lock, flags);
1489}
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001490EXPORT_SYMBOL_GPL(folio_add_wait_queue);
David Howells385e1ca5f2009-04-03 16:42:39 +01001491
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001492#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 Paulffceeb62020-06-04 16:49:22 -07001503 * being cleared, but a memory barrier should be unnecessary since it is
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001504 * in the same byte as PG_locked.
1505 */
1506static 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 Johansson98473f92016-12-29 14:16:07 -08001510 return test_bit(PG_waiters, mem);
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001511}
1512
1513#endif
1514
David Howells385e1ca5f2009-04-03 16:42:39 +01001515/**
Matthew Wilcox (Oracle)4e136422020-12-07 15:44:35 -05001516 * folio_unlock - Unlock a locked folio.
1517 * @folio: The folio.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 *
Matthew Wilcox (Oracle)4e136422020-12-07 15:44:35 -05001519 * Unlocks the folio and wakes up any thread sleeping on the page lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 *
Matthew Wilcox (Oracle)4e136422020-12-07 15:44:35 -05001521 * Context: May be called from interrupt or process context. May not be
1522 * called from NMI context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 */
Matthew Wilcox (Oracle)4e136422020-12-07 15:44:35 -05001524void folio_unlock(struct folio *folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525{
Matthew Wilcox (Oracle)4e136422020-12-07 15:44:35 -05001526 /* Bit 7 allows x86 to check the byte's sign bit */
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001527 BUILD_BUG_ON(PG_waiters != 7);
Matthew Wilcox (Oracle)4e136422020-12-07 15:44:35 -05001528 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)6974d7c2021-01-15 17:14:48 -05001531 folio_wake_bit(folio, PG_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532}
Matthew Wilcox (Oracle)4e136422020-12-07 15:44:35 -05001533EXPORT_SYMBOL(folio_unlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
Randy Dunlap485bb992006-06-23 02:03:49 -07001535/**
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001536 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1537 * @folio: The folio.
David Howells73e10de2020-02-10 10:00:21 +00001538 *
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001539 * 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 Howells73e10de2020-02-10 10:00:21 +00001541 *
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001542 * 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 Howells73e10de2020-02-10 10:00:21 +00001544 * serialised.
1545 */
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001546void folio_end_private_2(struct folio *folio)
David Howells73e10de2020-02-10 10:00:21 +00001547{
Matthew Wilcox (Oracle)6974d7c2021-01-15 17:14:48 -05001548 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 Howells73e10de2020-02-10 10:00:21 +00001552}
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001553EXPORT_SYMBOL(folio_end_private_2);
David Howells73e10de2020-02-10 10:00:21 +00001554
1555/**
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001556 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1557 * @folio: The folio to wait on.
David Howells73e10de2020-02-10 10:00:21 +00001558 *
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001559 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
David Howells73e10de2020-02-10 10:00:21 +00001560 */
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001561void folio_wait_private_2(struct folio *folio)
David Howells73e10de2020-02-10 10:00:21 +00001562{
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001563 while (folio_test_private_2(folio))
1564 folio_wait_bit(folio, PG_private_2);
David Howells73e10de2020-02-10 10:00:21 +00001565}
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001566EXPORT_SYMBOL(folio_wait_private_2);
David Howells73e10de2020-02-10 10:00:21 +00001567
1568/**
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001569 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1570 * @folio: The folio to wait on.
David Howells73e10de2020-02-10 10:00:21 +00001571 *
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001572 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
David Howells73e10de2020-02-10 10:00:21 +00001573 * 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)b47393f2021-04-22 22:58:32 -04001579int folio_wait_private_2_killable(struct folio *folio)
David Howells73e10de2020-02-10 10:00:21 +00001580{
1581 int ret = 0;
1582
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001583 while (folio_test_private_2(folio)) {
1584 ret = folio_wait_bit_killable(folio, PG_private_2);
David Howells73e10de2020-02-10 10:00:21 +00001585 if (ret < 0)
1586 break;
1587 }
1588
1589 return ret;
1590}
Matthew Wilcox (Oracle)b47393f2021-04-22 22:58:32 -04001591EXPORT_SYMBOL(folio_wait_private_2_killable);
David Howells73e10de2020-02-10 10:00:21 +00001592
1593/**
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001594 * folio_end_writeback - End writeback against a folio.
1595 * @folio: The folio.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 */
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001597void folio_end_writeback(struct folio *folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598{
Mel Gorman888cf2d2014-06-04 16:10:34 -07001599 /*
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001600 * 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 Gorman888cf2d2014-06-04 16:10:34 -07001605 */
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001606 if (folio_test_reclaim(folio)) {
1607 folio_clear_reclaim(folio);
Matthew Wilcox (Oracle)575ced12020-12-08 01:25:39 -05001608 folio_rotate_reclaimable(folio);
Mel Gorman888cf2d2014-06-04 16:10:34 -07001609 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -07001610
Hugh Dickins073861e2020-11-24 08:46:43 -08001611 /*
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001612 * Writeback does not hold a folio reference of its own, relying
Hugh Dickins073861e2020-11-24 08:46:43 -08001613 * on truncation to wait for the clearing of PG_writeback.
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001614 * But here we must make sure that the folio is not freed and
1615 * reused before the folio_wake().
Hugh Dickins073861e2020-11-24 08:46:43 -08001616 */
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001617 folio_get(folio);
Matthew Wilcox (Oracle)269ccca32021-01-15 23:34:16 -05001618 if (!__folio_end_writeback(folio))
Miklos Szerediac6aadb2008-04-28 02:12:38 -07001619 BUG();
1620
Peter Zijlstra4e857c52014-03-17 18:06:10 +01001621 smp_mb__after_atomic();
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001622 folio_wake(folio, PG_writeback);
Linus Torvalds512b79312021-11-06 14:08:17 -07001623 acct_reclaim_writeback(folio);
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001624 folio_put(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625}
Matthew Wilcox (Oracle)4268b482021-03-03 15:21:55 -05001626EXPORT_SYMBOL(folio_end_writeback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
Matthew Wilcox57d99842014-06-04 16:07:45 -07001628/*
1629 * After completing I/O on a page, call this routine to update the page
1630 * flags appropriately
1631 */
Jens Axboec11f0c02016-08-05 08:11:04 -06001632void page_endio(struct page *page, bool is_write, int err)
Matthew Wilcox57d99842014-06-04 16:07:45 -07001633{
Shaoqin Huang223ce492022-08-09 10:32:56 +08001634 struct folio *folio = page_folio(page);
1635
Jens Axboec11f0c02016-08-05 08:11:04 -06001636 if (!is_write) {
Matthew Wilcox57d99842014-06-04 16:07:45 -07001637 if (!err) {
Shaoqin Huang223ce492022-08-09 10:32:56 +08001638 folio_mark_uptodate(folio);
Matthew Wilcox57d99842014-06-04 16:07:45 -07001639 } else {
Shaoqin Huang223ce492022-08-09 10:32:56 +08001640 folio_clear_uptodate(folio);
1641 folio_set_error(folio);
Matthew Wilcox57d99842014-06-04 16:07:45 -07001642 }
Shaoqin Huang223ce492022-08-09 10:32:56 +08001643 folio_unlock(folio);
Mike Christieabf54542016-08-04 14:23:34 -06001644 } else {
Matthew Wilcox57d99842014-06-04 16:07:45 -07001645 if (err) {
Minchan Kimdd8416c2017-02-24 14:59:59 -08001646 struct address_space *mapping;
1647
Shaoqin Huang223ce492022-08-09 10:32:56 +08001648 folio_set_error(folio);
1649 mapping = folio_mapping(folio);
Minchan Kimdd8416c2017-02-24 14:59:59 -08001650 if (mapping)
1651 mapping_set_error(mapping, err);
Matthew Wilcox57d99842014-06-04 16:07:45 -07001652 }
Shaoqin Huang223ce492022-08-09 10:32:56 +08001653 folio_end_writeback(folio);
Matthew Wilcox57d99842014-06-04 16:07:45 -07001654 }
1655}
1656EXPORT_SYMBOL_GPL(page_endio);
1657
Randy Dunlap485bb992006-06-23 02:03:49 -07001658/**
Matthew Wilcox (Oracle)7c23c782021-03-01 19:38:25 -05001659 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1660 * @folio: The folio to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 */
Matthew Wilcox (Oracle)7c23c782021-03-01 19:38:25 -05001662void __folio_lock(struct folio *folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663{
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001664 folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001665 EXCLUSIVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666}
Matthew Wilcox (Oracle)7c23c782021-03-01 19:38:25 -05001667EXPORT_SYMBOL(__folio_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
Matthew Wilcox (Oracle)af7f29d2020-12-08 00:07:31 -05001669int __folio_lock_killable(struct folio *folio)
Matthew Wilcox2687a352007-12-06 11:18:49 -05001670{
Matthew Wilcox (Oracle)101c0bf2021-03-04 12:02:54 -05001671 return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001672 EXCLUSIVE);
Matthew Wilcox2687a352007-12-06 11:18:49 -05001673}
Matthew Wilcox (Oracle)af7f29d2020-12-08 00:07:31 -05001674EXPORT_SYMBOL_GPL(__folio_lock_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -05001675
Matthew Wilcox (Oracle)ffdc8da2020-12-30 17:58:40 -05001676static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
Jens Axboedd3e6d52020-05-22 09:12:09 -06001677{
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001678 struct wait_queue_head *q = folio_waitqueue(folio);
Matthew Wilcox (Oracle)f32b5dd2021-02-24 12:02:09 -08001679 int ret = 0;
1680
Matthew Wilcox (Oracle)df4d4f122021-01-16 11:22:14 -05001681 wait->folio = folio;
Matthew Wilcox (Oracle)f32b5dd2021-02-24 12:02:09 -08001682 wait->bit_nr = PG_locked;
1683
1684 spin_lock_irq(&q->lock);
1685 __add_wait_queue_entry_tail(q, &wait->wait);
Matthew Wilcox (Oracle)ffdc8da2020-12-30 17:58:40 -05001686 folio_set_waiters(folio);
1687 ret = !folio_trylock(folio);
Matthew Wilcox (Oracle)f32b5dd2021-02-24 12:02:09 -08001688 /*
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 Axboedd3e6d52020-05-22 09:12:09 -06001700}
1701
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001702/*
1703 * Return values:
Matthew Wilcox (Oracle)9138e472021-03-18 21:39:45 -04001704 * true - folio is locked; mmap_lock is still held.
1705 * false - folio is not locked.
Michel Lespinasse3e4e28c2020-06-08 21:33:51 -07001706 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001707 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07001708 * which case mmap_lock is still held.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001709 *
Matthew Wilcox (Oracle)9138e472021-03-18 21:39:45 -04001710 * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1711 * with the folio locked and the mmap_lock unperturbed.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001712 */
Matthew Wilcox (Oracle)9138e472021-03-18 21:39:45 -04001713bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
Michel Lespinassed065bd82010-10-26 14:21:57 -07001714 unsigned int flags)
1715{
Peter Xu4064b982020-04-01 21:08:45 -07001716 if (fault_flag_allow_retry_first(flags)) {
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001717 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07001718 * CAUTION! In this case, mmap_lock is not released
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001719 * even though return 0.
1720 */
1721 if (flags & FAULT_FLAG_RETRY_NOWAIT)
Matthew Wilcox (Oracle)9138e472021-03-18 21:39:45 -04001722 return false;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001723
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07001724 mmap_read_unlock(mm);
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001725 if (flags & FAULT_FLAG_KILLABLE)
Matthew Wilcox (Oracle)6baa8d62021-03-04 10:21:02 -05001726 folio_wait_locked_killable(folio);
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001727 else
Matthew Wilcox (Oracle)6baa8d62021-03-04 10:21:02 -05001728 folio_wait_locked(folio);
Matthew Wilcox (Oracle)9138e472021-03-18 21:39:45 -04001729 return false;
Michel Lespinassed065bd82010-10-26 14:21:57 -07001730 }
Hailong Liu800bca72020-12-14 19:05:02 -08001731 if (flags & FAULT_FLAG_KILLABLE) {
Matthew Wilcox (Oracle)9138e472021-03-18 21:39:45 -04001732 bool ret;
Hailong Liu800bca72020-12-14 19:05:02 -08001733
Matthew Wilcox (Oracle)af7f29d2020-12-08 00:07:31 -05001734 ret = __folio_lock_killable(folio);
Hailong Liu800bca72020-12-14 19:05:02 -08001735 if (ret) {
1736 mmap_read_unlock(mm);
Matthew Wilcox (Oracle)9138e472021-03-18 21:39:45 -04001737 return false;
Hailong Liu800bca72020-12-14 19:05:02 -08001738 }
1739 } else {
Matthew Wilcox (Oracle)af7f29d2020-12-08 00:07:31 -05001740 __folio_lock(folio);
Hailong Liu800bca72020-12-14 19:05:02 -08001741 }
Hailong Liu800bca72020-12-14 19:05:02 -08001742
Matthew Wilcox (Oracle)9138e472021-03-18 21:39:45 -04001743 return true;
Michel Lespinassed065bd82010-10-26 14:21:57 -07001744}
1745
Randy Dunlap485bb992006-06-23 02:03:49 -07001746/**
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001747 * 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 Weinere7b563b2014-04-03 14:47:44 -07001751 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001752 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1753 * gap with the lowest index.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001754 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001755 * 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 Weinere7b563b2014-04-03 14:47:44 -07001760 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001761 * 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 Weinere7b563b2014-04-03 14:47:44 -07001764 */
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001765pgoff_t page_cache_next_miss(struct address_space *mapping,
Johannes Weinere7b563b2014-04-03 14:47:44 -07001766 pgoff_t index, unsigned long max_scan)
1767{
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001768 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001769
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001770 while (max_scan--) {
1771 void *entry = xas_next(&xas);
1772 if (!entry || xa_is_value(entry))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001773 break;
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001774 if (xas.xa_index == 0)
Johannes Weinere7b563b2014-04-03 14:47:44 -07001775 break;
1776 }
1777
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001778 return xas.xa_index;
Johannes Weinere7b563b2014-04-03 14:47:44 -07001779}
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001780EXPORT_SYMBOL(page_cache_next_miss);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001781
1782/**
Laurent Dufour2346a562019-05-13 17:21:29 -07001783 * page_cache_prev_miss() - Find the previous gap in the page cache.
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001784 * @mapping: Mapping.
1785 * @index: Index.
1786 * @max_scan: Maximum range to search.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001787 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001788 * Search the range [max(index - max_scan + 1, 0), index] for the
1789 * gap with the highest index.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001790 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001791 * 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 Weinere7b563b2014-04-03 14:47:44 -07001796 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001797 * 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 Weinere7b563b2014-04-03 14:47:44 -07001800 */
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001801pgoff_t page_cache_prev_miss(struct address_space *mapping,
Johannes Weinere7b563b2014-04-03 14:47:44 -07001802 pgoff_t index, unsigned long max_scan)
1803{
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001804 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001805
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001806 while (max_scan--) {
1807 void *entry = xas_prev(&xas);
1808 if (!entry || xa_is_value(entry))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001809 break;
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001810 if (xas.xa_index == ULONG_MAX)
Johannes Weinere7b563b2014-04-03 14:47:44 -07001811 break;
1812 }
1813
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001814 return xas.xa_index;
Johannes Weinere7b563b2014-04-03 14:47:44 -07001815}
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001816EXPORT_SYMBOL(page_cache_prev_miss);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001817
Matthew Wilcox (Oracle)44835d22021-02-25 17:15:36 -08001818/*
Matthew Wilcox (Oracle)020853b2021-05-10 16:33:22 -04001819 * 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)bc5a3012021-02-25 17:15:40 -08001839 * mapping_get_entry - Get a page cache entry.
Randy Dunlap485bb992006-06-23 02:03:49 -07001840 * @mapping: the address_space to search
Matthew Wilcox (Oracle)a6de4b42020-10-13 16:51:34 -07001841 * @index: The page cache index.
Randy Dunlap485bb992006-06-23 02:03:49 -07001842 *
Matthew Wilcox (Oracle)bca65ee2020-12-15 23:22:38 -05001843 * 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 Weiner0cd61442014-04-03 14:47:46 -07001847 *
Matthew Wilcox (Oracle)bca65ee2020-12-15 23:22:38 -05001848 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 */
Matthew Wilcox (Oracle)bca65ee2020-12-15 23:22:38 -05001850static void *mapping_get_entry(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851{
Matthew Wilcox (Oracle)a6de4b42020-10-13 16:51:34 -07001852 XA_STATE(xas, &mapping->i_pages, index);
Matthew Wilcox (Oracle)bca65ee2020-12-15 23:22:38 -05001853 struct folio *folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
Nick Piggina60637c2008-07-25 19:45:31 -07001855 rcu_read_lock();
1856repeat:
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001857 xas_reset(&xas);
Matthew Wilcox (Oracle)bca65ee2020-12-15 23:22:38 -05001858 folio = xas_load(&xas);
1859 if (xas_retry(&xas, folio))
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001860 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)bca65ee2020-12-15 23:22:38 -05001865 if (!folio || xa_is_value(folio))
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001866 goto out;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001867
Matthew Wilcox (Oracle)bca65ee2020-12-15 23:22:38 -05001868 if (!folio_try_get_rcu(folio))
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001869 goto repeat;
Nick Piggina60637c2008-07-25 19:45:31 -07001870
Matthew Wilcox (Oracle)bca65ee2020-12-15 23:22:38 -05001871 if (unlikely(folio != xas_reload(&xas))) {
1872 folio_put(folio);
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001873 goto repeat;
Nick Piggina60637c2008-07-25 19:45:31 -07001874 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001875out:
Nick Piggina60637c2008-07-25 19:45:31 -07001876 rcu_read_unlock();
1877
Matthew Wilcox (Oracle)bca65ee2020-12-15 23:22:38 -05001878 return folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Randy Dunlap485bb992006-06-23 02:03:49 -07001881/**
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001882 * __filemap_get_folio - Find and get a reference to a folio.
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001883 * @mapping: The address_space to search.
1884 * @index: The page index.
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001885 * @fgp_flags: %FGP flags modify how the folio is returned.
1886 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001887 *
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001888 * Looks up the page cache entry at @mapping & @index.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001889 *
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001890 * @fgp_flags can be zero or more of these flags:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001891 *
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001892 * * %FGP_ACCESSED - The folio will be marked accessed.
1893 * * %FGP_LOCK - The folio is returned locked.
Matthew Wilcox (Oracle)44835d22021-02-25 17:15:36 -08001894 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001895 * instead of allocating a new folio to replace it.
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001896 * * %FGP_CREAT - If no page is present then a new page is allocated using
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001897 * @gfp and added to the page cache and the VM's LRU list.
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001898 * 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)3f0c6a02021-03-08 11:45:35 -05001902 * * %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)b27652d92020-12-24 12:55:56 -05001905 * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
mchehab@s-opensource.com0e056eb2017-03-30 17:11:36 -03001906 *
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001907 * 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 Gorman2457aec2014-06-04 16:10:31 -07001909 *
1910 * If there is a page cache page, it is returned with an increased refcount.
Mike Rapoporta862f682019-03-05 15:48:42 -08001911 *
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001912 * Return: The found folio or %NULL otherwise.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001913 */
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001914struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1915 int fgp_flags, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916{
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001917 struct folio *folio;
Mel Gorman2457aec2014-06-04 16:10:31 -07001918
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919repeat:
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001920 folio = mapping_get_entry(mapping, index);
1921 if (xa_is_value(folio)) {
Matthew Wilcox (Oracle)44835d22021-02-25 17:15:36 -08001922 if (fgp_flags & FGP_ENTRY)
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001923 return folio;
1924 folio = NULL;
Matthew Wilcox (Oracle)44835d22021-02-25 17:15:36 -08001925 }
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001926 if (!folio)
Mel Gorman2457aec2014-06-04 16:10:31 -07001927 goto no_page;
1928
1929 if (fgp_flags & FGP_LOCK) {
1930 if (fgp_flags & FGP_NOWAIT) {
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001931 if (!folio_trylock(folio)) {
1932 folio_put(folio);
Mel Gorman2457aec2014-06-04 16:10:31 -07001933 return NULL;
1934 }
1935 } else {
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001936 folio_lock(folio);
Mel Gorman2457aec2014-06-04 16:10:31 -07001937 }
1938
1939 /* Has the page been truncated? */
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001940 if (unlikely(folio->mapping != mapping)) {
1941 folio_unlock(folio);
1942 folio_put(folio);
Mel Gorman2457aec2014-06-04 16:10:31 -07001943 goto repeat;
1944 }
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001945 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
Mel Gorman2457aec2014-06-04 16:10:31 -07001946 }
1947
Kirill Tkhaic16eb002018-12-28 00:37:35 -08001948 if (fgp_flags & FGP_ACCESSED)
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001949 folio_mark_accessed(folio);
Yang Shib9306a72020-08-06 23:19:55 -07001950 else if (fgp_flags & FGP_WRITE) {
1951 /* Clear idle flag for buffer write */
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001952 if (folio_test_idle(folio))
1953 folio_clear_idle(folio);
Yang Shib9306a72020-08-06 23:19:55 -07001954 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001955
Matthew Wilcox (Oracle)b27652d92020-12-24 12:55:56 -05001956 if (fgp_flags & FGP_STABLE)
1957 folio_wait_stable(folio);
Mel Gorman2457aec2014-06-04 16:10:31 -07001958no_page:
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001959 if (!folio && (fgp_flags & FGP_CREAT)) {
Mel Gorman2457aec2014-06-04 16:10:31 -07001960 int err;
Christoph Hellwigf56753a2020-09-24 08:51:40 +02001961 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001962 gfp |= __GFP_WRITE;
Michal Hocko45f87de2014-12-29 20:30:35 +01001963 if (fgp_flags & FGP_NOFS)
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001964 gfp &= ~__GFP_FS;
Jens Axboe0dd316b2022-07-01 14:04:43 -06001965 if (fgp_flags & FGP_NOWAIT) {
1966 gfp &= ~GFP_KERNEL;
1967 gfp |= GFP_NOWAIT | __GFP_NOWARN;
1968 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001969
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001970 folio = filemap_alloc_folio(gfp, 0);
1971 if (!folio)
Nick Piggineb2be182007-10-16 01:24:57 -07001972 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001973
Josef Bacika75d4c32019-03-13 11:44:14 -07001974 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
Mel Gorman2457aec2014-06-04 16:10:31 -07001975 fgp_flags |= FGP_LOCK;
1976
Hugh Dickinseb39d612014-08-06 16:06:43 -07001977 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001978 if (fgp_flags & FGP_ACCESSED)
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001979 __folio_set_referenced(folio);
Mel Gorman2457aec2014-06-04 16:10:31 -07001980
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001981 err = filemap_add_folio(mapping, folio, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07001982 if (unlikely(err)) {
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001983 folio_put(folio);
1984 folio = NULL;
Nick Piggineb2be182007-10-16 01:24:57 -07001985 if (err == -EEXIST)
1986 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 }
Josef Bacika75d4c32019-03-13 11:44:14 -07001988
1989 /*
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001990 * filemap_add_folio locks the page, and for mmap
1991 * we expect an unlocked page.
Josef Bacika75d4c32019-03-13 11:44:14 -07001992 */
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001993 if (folio && (fgp_flags & FGP_FOR_MMAP))
1994 folio_unlock(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001996
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001997 return folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998}
Matthew Wilcox (Oracle)3f0c6a02021-03-08 11:45:35 -05001999EXPORT_SYMBOL(__filemap_get_folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002001static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
Matthew Wilcox (Oracle)c7bad632021-02-25 17:15:44 -08002002 xa_mark_t mark)
2003{
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002004 struct folio *folio;
Matthew Wilcox (Oracle)c7bad632021-02-25 17:15:44 -08002005
2006retry:
2007 if (mark == XA_PRESENT)
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002008 folio = xas_find(xas, max);
Matthew Wilcox (Oracle)c7bad632021-02-25 17:15:44 -08002009 else
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002010 folio = xas_find_marked(xas, max, mark);
Matthew Wilcox (Oracle)c7bad632021-02-25 17:15:44 -08002011
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002012 if (xas_retry(xas, folio))
Matthew Wilcox (Oracle)c7bad632021-02-25 17:15:44 -08002013 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)f5e64292020-12-17 00:12:26 -05002019 if (!folio || xa_is_value(folio))
2020 return folio;
Matthew Wilcox (Oracle)c7bad632021-02-25 17:15:44 -08002021
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002022 if (!folio_try_get_rcu(folio))
Matthew Wilcox (Oracle)c7bad632021-02-25 17:15:44 -08002023 goto reset;
2024
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002025 if (unlikely(folio != xas_reload(xas))) {
2026 folio_put(folio);
Matthew Wilcox (Oracle)c7bad632021-02-25 17:15:44 -08002027 goto reset;
2028 }
2029
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002030 return folio;
Matthew Wilcox (Oracle)c7bad632021-02-25 17:15:44 -08002031reset:
2032 xas_reset(xas);
2033 goto retry;
2034}
2035
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07002037 * find_get_entries - gang pagecache lookup
2038 * @mapping: The address_space to search
2039 * @start: The starting page cache index
Matthew Wilcox (Oracle)ca122fe42021-02-25 17:16:00 -08002040 * @end: The final page index (inclusive).
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04002041 * @fbatch: Where the resulting entries are placed.
Johannes Weiner0cd61442014-04-03 14:47:46 -07002042 * @indices: The cache indices corresponding to the entries in @entries
2043 *
Matthew Wilcox (Oracle)cf2039a2021-02-25 17:16:11 -08002044 * find_get_entries() will search for and return a batch of entries in
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04002045 * the mapping. The entries are placed in @fbatch. find_get_entries()
2046 * takes a reference on any actual folios it returns.
Johannes Weiner0cd61442014-04-03 14:47:46 -07002047 *
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04002048 * The entries have ascending indexes. The indices may not be consecutive
2049 * due to not-present entries or large folios.
Johannes Weiner0cd61442014-04-03 14:47:46 -07002050 *
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04002051 * Any shadow entries of evicted folios, or swap entries from
Johannes Weiner139b6a62014-05-06 12:50:05 -07002052 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07002053 *
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04002054 * Return: The number of entries which were found.
Johannes Weiner0cd61442014-04-03 14:47:46 -07002055 */
Vishal Moola (Oracle)9fb6bee2022-10-17 09:18:00 -07002056unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04002057 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
Johannes Weiner0cd61442014-04-03 14:47:46 -07002058{
Vishal Moola (Oracle)9fb6bee2022-10-17 09:18:00 -07002059 XA_STATE(xas, &mapping->i_pages, *start);
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002060 struct folio *folio;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002061
2062 rcu_read_lock();
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002063 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
Matthew Wilcox (Oracle)0e499ed2020-09-01 23:17:50 -04002064 indices[fbatch->nr] = xas.xa_index;
2065 if (!folio_batch_add(fbatch, folio))
Johannes Weiner0cd61442014-04-03 14:47:46 -07002066 break;
2067 }
2068 rcu_read_unlock();
Matthew Wilcox (Oracle)cf2039a2021-02-25 17:16:11 -08002069
Vishal Moola (Oracle)9fb6bee2022-10-17 09:18:00 -07002070 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)0e499ed2020-09-01 23:17:50 -04002079 return folio_batch_count(fbatch);
Johannes Weiner0cd61442014-04-03 14:47:46 -07002080}
2081
2082/**
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002083 * 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)51dcbda2021-12-07 14:15:07 -05002087 * @fbatch: Where the resulting entries are placed.
2088 * @indices: The cache indices of the entries in @fbatch.
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002089 *
2090 * find_lock_entries() will return a batch of entries from @mapping.
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002091 * 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)5c211ba2021-02-25 17:15:56 -08002095 *
2096 * The entries have ascending indexes. The indices may not be consecutive
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002097 * due to not-present entries, large folios, folios which could not be
2098 * locked or folios under writeback.
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002099 *
2100 * Return: The number of entries which were found.
2101 */
Vishal Moola (Oracle)3392ca122022-10-17 09:17:59 -07002102unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -05002103 pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002104{
Vishal Moola (Oracle)3392ca122022-10-17 09:17:59 -07002105 XA_STATE(xas, &mapping->i_pages, *start);
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002106 struct folio *folio;
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002107
2108 rcu_read_lock();
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002109 while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2110 if (!xa_is_value(folio)) {
Vishal Moola (Oracle)3392ca122022-10-17 09:17:59 -07002111 if (folio->index < *start)
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002112 goto put;
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002113 if (folio->index + folio_nr_pages(folio) - 1 > end)
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002114 goto put;
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002115 if (!folio_trylock(folio))
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002116 goto put;
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002117 if (folio->mapping != mapping ||
2118 folio_test_writeback(folio))
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002119 goto unlock;
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002120 VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2121 folio);
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002122 }
Matthew Wilcox (Oracle)51dcbda2021-12-07 14:15:07 -05002123 indices[fbatch->nr] = xas.xa_index;
2124 if (!folio_batch_add(fbatch, folio))
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002125 break;
Matthew Wilcox (Oracle)6b24ca42020-06-27 22:19:08 -04002126 continue;
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002127unlock:
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002128 folio_unlock(folio);
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002129put:
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002130 folio_put(folio);
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002131 }
2132 rcu_read_unlock();
2133
Vishal Moola (Oracle)3392ca122022-10-17 09:17:59 -07002134 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)51dcbda2021-12-07 14:15:07 -05002143 return folio_batch_count(fbatch);
Matthew Wilcox (Oracle)5c211ba2021-02-25 17:15:56 -08002144}
2145
2146/**
Matthew Wilcox (Oracle)be0ced52022-06-03 15:30:25 -04002147 * filemap_get_folios - Get a batch of folios
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 * @mapping: The address_space to search
2149 * @start: The starting page index
Jan Karab947cee2017-09-06 16:21:21 -07002150 * @end: The final page index (inclusive)
Matthew Wilcox (Oracle)be0ced52022-06-03 15:30:25 -04002151 * @fbatch: The batch to fill.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 *
Matthew Wilcox (Oracle)be0ced52022-06-03 15:30:25 -04002153 * 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 Torvalds1da177e2005-04-16 15:20:36 -07002156 *
Matthew Wilcox (Oracle)be0ced52022-06-03 15:30:25 -04002157 * 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 Torvalds1da177e2005-04-16 15:20:36 -07002163 *
Matthew Wilcox (Oracle)be0ced52022-06-03 15:30:25 -04002164 * Return: The number of folios which were found.
2165 * We also update @start to index the next folio for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 */
Matthew Wilcox (Oracle)be0ced52022-06-03 15:30:25 -04002167unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
2168 pgoff_t end, struct folio_batch *fbatch)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169{
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04002170 XA_STATE(xas, &mapping->i_pages, *start);
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002171 struct folio *folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172
Nick Piggina60637c2008-07-25 19:45:31 -07002173 rcu_read_lock();
Matthew Wilcox (Oracle)be0ced52022-06-03 15:30:25 -04002174 while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04002175 /* Skip over shadow, swap and DAX entries */
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002176 if (xa_is_value(folio))
Hugh Dickins8079b1c2011-08-03 16:21:28 -07002177 continue;
Matthew Wilcox (Oracle)be0ced52022-06-03 15:30:25 -04002178 if (!folio_batch_add(fbatch, folio)) {
2179 unsigned long nr = folio_nr_pages(folio);
Nick Piggina60637c2008-07-25 19:45:31 -07002180
Matthew Wilcox (Oracle)be0ced52022-06-03 15:30:25 -04002181 if (folio_test_hugetlb(folio))
2182 nr = 1;
2183 *start = folio->index + nr;
Jan Karab947cee2017-09-06 16:21:21 -07002184 goto out;
2185 }
Nick Piggina60637c2008-07-25 19:45:31 -07002186 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07002187
Jan Karab947cee2017-09-06 16:21:21 -07002188 /*
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 Wilcoxfd1b3ce2018-05-16 17:38:56 -04002191 * breaks the iteration when there is a page at index -1 but that is
Jan Karab947cee2017-09-06 16:21:21 -07002192 * already broken anyway.
2193 */
2194 if (end == (pgoff_t)-1)
2195 *start = (pgoff_t)-1;
2196 else
2197 *start = end + 1;
2198out:
Nick Piggina60637c2008-07-25 19:45:31 -07002199 rcu_read_unlock();
Jan Karad72dc8a2017-09-06 16:21:18 -07002200
Matthew Wilcox (Oracle)be0ced52022-06-03 15:30:25 -04002201 return folio_batch_count(fbatch);
2202}
2203EXPORT_SYMBOL(filemap_get_folios);
2204
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205static inline
Jens Axboeebf43502006-04-27 08:46:01 +02002206bool 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 Torvalds1da177e2005-04-16 15:20:36 -07002213}
2214
Jens Axboeebf43502006-04-27 08:46:01 +02002215/**
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002216 * filemap_get_folios_contig - Get a batch of contiguous folios
Jens Axboeebf43502006-04-27 08:46:01 +02002217 * @mapping: The address_space to search
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002218 * @start: The starting page index
2219 * @end: The final page index (inclusive)
2220 * @fbatch: The batch to fill
Jens Axboeebf43502006-04-27 08:46:01 +02002221 *
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002222 * 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 Axboeebf43502006-04-27 08:46:01 +02002225 *
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002226 * Return: The number of folios found.
2227 * Also update @start to be positioned for traversal of the next folio.
Jens Axboeebf43502006-04-27 08:46:01 +02002228 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07002229
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002230unsigned 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 Axboeebf43502006-04-27 08:46:01 +02002236
Nick Piggina60637c2008-07-25 19:45:31 -07002237 rcu_read_lock();
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002238
2239 for (folio = xas_load(&xas); folio && xas.xa_index <= end;
2240 folio = xas_next(&xas)) {
Matthew Wilcox (Oracle)e1c37722021-03-06 16:38:38 -05002241 if (xas_retry(&xas, folio))
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04002242 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)e1c37722021-03-06 16:38:38 -05002247 if (xa_is_value(folio))
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002248 goto update_start;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07002249
Matthew Wilcox (Oracle)e1c37722021-03-06 16:38:38 -05002250 if (!folio_try_get_rcu(folio))
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04002251 goto retry;
Nick Piggina60637c2008-07-25 19:45:31 -07002252
Matthew Wilcox (Oracle)e1c37722021-03-06 16:38:38 -05002253 if (unlikely(folio != xas_reload(&xas)))
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002254 goto put_folio;
Nick Piggina60637c2008-07-25 19:45:31 -07002255
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002256 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)6b24ca42020-06-27 22:19:08 -04002263 }
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04002264 continue;
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002265put_folio:
Matthew Wilcox (Oracle)e1c37722021-03-06 16:38:38 -05002266 folio_put(folio);
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002267
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04002268retry:
2269 xas_reset(&xas);
Jens Axboeebf43502006-04-27 08:46:01 +02002270 }
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002271
2272update_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 }
2282out:
Nick Piggina60637c2008-07-25 19:45:31 -07002283 rcu_read_unlock();
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002284 return folio_batch_count(fbatch);
Jens Axboeebf43502006-04-27 08:46:01 +02002285}
Vishal Moola (Oracle)35b47142022-08-23 17:40:17 -07002286EXPORT_SYMBOL(filemap_get_folios_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02002287
Randy Dunlap485bb992006-06-23 02:03:49 -07002288/**
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002289 * 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 Dunlap485bb992006-06-23 02:03:49 -07002295 *
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002296 * Same as filemap_get_folios(), but only returning folios tagged with @tag.
Mike Rapoporta862f682019-03-05 15:48:42 -08002297 *
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002298 * Return: The number of folios found.
2299 * Also update @start to index the next folio for traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 */
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002301unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start,
2302 pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303{
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002304 XA_STATE(xas, &mapping->i_pages, *start);
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002305 struct folio *folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306
Nick Piggina60637c2008-07-25 19:45:31 -07002307 rcu_read_lock();
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002308 while ((folio = find_get_entry(&xas, end, tag)) != NULL) {
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002309 /*
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)247f9e12023-01-04 13:14:27 -08002312 * a page we saw tagged. Skip over it.
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002313 */
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002314 if (xa_is_value(folio))
Johannes Weiner139b6a62014-05-06 12:50:05 -07002315 continue;
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002316 if (!folio_batch_add(fbatch, folio)) {
2317 unsigned long nr = folio_nr_pages(folio);
Nick Piggina60637c2008-07-25 19:45:31 -07002318
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002319 if (folio_test_hugetlb(folio))
2320 nr = 1;
2321 *start = folio->index + nr;
Jan Kara72b045a2017-11-15 17:34:33 -08002322 goto out;
2323 }
Nick Piggina60637c2008-07-25 19:45:31 -07002324 }
Jan Kara72b045a2017-11-15 17:34:33 -08002325 /*
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002326 * 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 Kara72b045a2017-11-15 17:34:33 -08002330 */
2331 if (end == (pgoff_t)-1)
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002332 *start = (pgoff_t)-1;
Jan Kara72b045a2017-11-15 17:34:33 -08002333 else
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002334 *start = end + 1;
Jan Kara72b045a2017-11-15 17:34:33 -08002335out:
Nick Piggina60637c2008-07-25 19:45:31 -07002336 rcu_read_unlock();
2337
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002338 return folio_batch_count(fbatch);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339}
Vishal Moola (Oracle)247f9e12023-01-04 13:14:27 -08002340EXPORT_SYMBOL(filemap_get_folios_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Wu Fengguang76d42bd2006-06-25 05:48:43 -07002342/*
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 Joarder0f8e2db2020-04-01 21:04:50 -07002357static void shrink_readahead_size_eio(struct file_ra_state *ra)
Wu Fengguang76d42bd2006-06-25 05:48:43 -07002358{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07002359 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07002360}
2361
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002362/*
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002363 * filemap_get_read_batch - Get a batch of folios for read
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002364 *
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002365 * 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)cbd59c42021-02-24 12:01:59 -08002370 */
2371static void filemap_get_read_batch(struct address_space *mapping,
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002372 pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002373{
2374 XA_STATE(xas, &mapping->i_pages, index);
Matthew Wilcox (Oracle)bdb72932021-03-05 10:29:41 -05002375 struct folio *folio;
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002376
2377 rcu_read_lock();
Matthew Wilcox (Oracle)bdb72932021-03-05 10:29:41 -05002378 for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2379 if (xas_retry(&xas, folio))
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002380 continue;
Matthew Wilcox (Oracle)bdb72932021-03-05 10:29:41 -05002381 if (xas.xa_index > max || xa_is_value(folio))
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002382 break;
Matthew Wilcox (Oracle)cb995f42022-06-17 20:00:17 -04002383 if (xa_is_sibling(folio))
2384 break;
Matthew Wilcox (Oracle)bdb72932021-03-05 10:29:41 -05002385 if (!folio_try_get_rcu(folio))
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002386 goto retry;
2387
Matthew Wilcox (Oracle)bdb72932021-03-05 10:29:41 -05002388 if (unlikely(folio != xas_reload(&xas)))
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002389 goto put_folio;
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002390
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002391 if (!folio_batch_add(fbatch, folio))
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002392 break;
Matthew Wilcox (Oracle)bdb72932021-03-05 10:29:41 -05002393 if (!folio_test_uptodate(folio))
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002394 break;
Matthew Wilcox (Oracle)bdb72932021-03-05 10:29:41 -05002395 if (folio_test_readahead(folio))
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002396 break;
Matthew Wilcox (Oracle)6b24ca42020-06-27 22:19:08 -04002397 xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1);
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002398 continue;
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002399put_folio:
Matthew Wilcox (Oracle)bdb72932021-03-05 10:29:41 -05002400 folio_put(folio);
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002401retry:
2402 xas_reset(&xas);
2403 }
2404 rcu_read_unlock();
2405}
2406
Matthew Wilcox (Oracle)290e1a32022-05-12 17:37:01 -04002407static int filemap_read_folio(struct file *file, filler_t filler,
Matthew Wilcox (Oracle)9d427b42021-03-10 10:19:30 -05002408 struct folio *folio)
Kent Overstreet723ef242020-12-14 19:04:52 -08002409{
Christoph Hellwig17604242022-09-15 10:41:56 +01002410 bool workingset = folio_test_workingset(folio);
2411 unsigned long pflags;
Kent Overstreet723ef242020-12-14 19:04:52 -08002412 int error;
2413
Kent Overstreet723ef242020-12-14 19:04:52 -08002414 /*
Matthew Wilcox (Oracle)68430302021-02-24 12:02:15 -08002415 * A previous I/O error may have been due to temporary failures,
Matthew Wilcox (Oracle)7e0a1262022-04-29 11:53:28 -04002416 * eg. multipath errors. PG_error will be set again if read_folio
Matthew Wilcox (Oracle)68430302021-02-24 12:02:15 -08002417 * fails.
Kent Overstreet723ef242020-12-14 19:04:52 -08002418 */
Matthew Wilcox (Oracle)9d427b42021-03-10 10:19:30 -05002419 folio_clear_error(folio);
Christoph Hellwig17604242022-09-15 10:41:56 +01002420
Kent Overstreet723ef242020-12-14 19:04:52 -08002421 /* Start the actual read. The read will unlock the page. */
Christoph Hellwig17604242022-09-15 10:41:56 +01002422 if (unlikely(workingset))
2423 psi_memstall_enter(&pflags);
Matthew Wilcox (Oracle)290e1a32022-05-12 17:37:01 -04002424 error = filler(file, folio);
Christoph Hellwig17604242022-09-15 10:41:56 +01002425 if (unlikely(workingset))
2426 psi_memstall_leave(&pflags);
Matthew Wilcox (Oracle)68430302021-02-24 12:02:15 -08002427 if (error)
2428 return error;
Kent Overstreet723ef242020-12-14 19:04:52 -08002429
Matthew Wilcox (Oracle)9d427b42021-03-10 10:19:30 -05002430 error = folio_wait_locked_killable(folio);
Matthew Wilcox (Oracle)68430302021-02-24 12:02:15 -08002431 if (error)
2432 return error;
Matthew Wilcox (Oracle)9d427b42021-03-10 10:19:30 -05002433 if (folio_test_uptodate(folio))
Matthew Wilcox (Oracle)aa1ec2f2021-02-24 12:02:38 -08002434 return 0;
Matthew Wilcox (Oracle)290e1a32022-05-12 17:37:01 -04002435 if (file)
2436 shrink_readahead_size_eio(&file->f_ra);
Matthew Wilcox (Oracle)aa1ec2f2021-02-24 12:02:38 -08002437 return -EIO;
Kent Overstreet723ef242020-12-14 19:04:52 -08002438}
2439
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002440static bool filemap_range_uptodate(struct address_space *mapping,
David Howellsdd5b9d02023-02-08 18:18:17 +00002441 loff_t pos, size_t count, struct folio *folio,
2442 bool need_uptodate)
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002443{
Matthew Wilcox (Oracle)2fa4eeb82021-03-10 11:04:19 -05002444 if (folio_test_uptodate(folio))
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002445 return true;
2446 /* pipes can't handle partially uptodate pages */
David Howellsdd5b9d02023-02-08 18:18:17 +00002447 if (need_uptodate)
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002448 return false;
2449 if (!mapping->a_ops->is_partially_uptodate)
2450 return false;
Matthew Wilcox (Oracle)2fa4eeb82021-03-10 11:04:19 -05002451 if (mapping->host->i_blkbits >= folio_shift(folio))
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002452 return false;
2453
Matthew Wilcox (Oracle)2fa4eeb82021-03-10 11:04:19 -05002454 if (folio_pos(folio) > pos) {
2455 count -= folio_pos(folio) - pos;
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002456 pos = 0;
2457 } else {
Matthew Wilcox (Oracle)2fa4eeb82021-03-10 11:04:19 -05002458 pos -= folio_pos(folio);
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002459 }
2460
Matthew Wilcox (Oracle)2e7e80f2022-02-09 20:21:27 +00002461 return mapping->a_ops->is_partially_uptodate(folio, pos, count);
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002462}
2463
Matthew Wilcox (Oracle)4612aee2021-02-24 12:02:22 -08002464static int filemap_update_page(struct kiocb *iocb,
David Howellsdd5b9d02023-02-08 18:18:17 +00002465 struct address_space *mapping, size_t count,
2466 struct folio *folio, bool need_uptodate)
Kent Overstreet723ef242020-12-14 19:04:52 -08002467{
Kent Overstreet723ef242020-12-14 19:04:52 -08002468 int error;
2469
Jan Kara730633f2021-01-28 19:19:45 +01002470 if (iocb->ki_flags & IOCB_NOWAIT) {
2471 if (!filemap_invalidate_trylock_shared(mapping))
Matthew Wilcox (Oracle)87d1d7b2021-02-24 12:02:25 -08002472 return -EAGAIN;
Jan Kara730633f2021-01-28 19:19:45 +01002473 } else {
2474 filemap_invalidate_lock_shared(mapping);
2475 }
2476
Matthew Wilcox (Oracle)ffdc8da2020-12-30 17:58:40 -05002477 if (!folio_trylock(folio)) {
Jan Kara730633f2021-01-28 19:19:45 +01002478 error = -EAGAIN;
2479 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2480 goto unlock_mapping;
Matthew Wilcox (Oracle)87d1d7b2021-02-24 12:02:25 -08002481 if (!(iocb->ki_flags & IOCB_WAITQ)) {
Jan Kara730633f2021-01-28 19:19:45 +01002482 filemap_invalidate_unlock_shared(mapping);
Matthew Wilcox (Oracle)9f2b04a2021-08-16 23:36:31 -04002483 /*
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)4612aee2021-02-24 12:02:22 -08002488 return AOP_TRUNCATED_PAGE;
Matthew Wilcox (Oracle)bd8a1f32021-02-24 12:02:05 -08002489 }
Matthew Wilcox (Oracle)ffdc8da2020-12-30 17:58:40 -05002490 error = __folio_lock_async(folio, iocb->ki_waitq);
Matthew Wilcox (Oracle)87d1d7b2021-02-24 12:02:25 -08002491 if (error)
Jan Kara730633f2021-01-28 19:19:45 +01002492 goto unlock_mapping;
Kent Overstreet723ef242020-12-14 19:04:52 -08002493 }
Kent Overstreet723ef242020-12-14 19:04:52 -08002494
Jan Kara730633f2021-01-28 19:19:45 +01002495 error = AOP_TRUNCATED_PAGE;
Matthew Wilcox (Oracle)ffdc8da2020-12-30 17:58:40 -05002496 if (!folio->mapping)
Jan Kara730633f2021-01-28 19:19:45 +01002497 goto unlock;
Kent Overstreet723ef242020-12-14 19:04:52 -08002498
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002499 error = 0;
David Howellsdd5b9d02023-02-08 18:18:17 +00002500 if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio,
2501 need_uptodate))
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002502 goto unlock;
2503
2504 error = -EAGAIN;
2505 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2506 goto unlock;
2507
Matthew Wilcox (Oracle)290e1a32022-05-12 17:37:01 -04002508 error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
2509 folio);
Jan Kara730633f2021-01-28 19:19:45 +01002510 goto unlock_mapping;
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002511unlock:
Matthew Wilcox (Oracle)ffdc8da2020-12-30 17:58:40 -05002512 folio_unlock(folio);
Jan Kara730633f2021-01-28 19:19:45 +01002513unlock_mapping:
2514 filemap_invalidate_unlock_shared(mapping);
2515 if (error == AOP_TRUNCATED_PAGE)
Matthew Wilcox (Oracle)ffdc8da2020-12-30 17:58:40 -05002516 folio_put(folio);
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002517 return error;
Kent Overstreet723ef242020-12-14 19:04:52 -08002518}
2519
Matthew Wilcox (Oracle)a5d4ad02021-03-10 10:34:00 -05002520static int filemap_create_folio(struct file *file,
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002521 struct address_space *mapping, pgoff_t index,
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002522 struct folio_batch *fbatch)
Kent Overstreet723ef242020-12-14 19:04:52 -08002523{
Matthew Wilcox (Oracle)a5d4ad02021-03-10 10:34:00 -05002524 struct folio *folio;
Kent Overstreet723ef242020-12-14 19:04:52 -08002525 int error;
2526
Matthew Wilcox (Oracle)a5d4ad02021-03-10 10:34:00 -05002527 folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2528 if (!folio)
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002529 return -ENOMEM;
Kent Overstreet723ef242020-12-14 19:04:52 -08002530
Jan Kara730633f2021-01-28 19:19:45 +01002531 /*
Matthew Wilcox (Oracle)a5d4ad02021-03-10 10:34:00 -05002532 * 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)704528d2022-03-23 21:29:04 -04002540 * pages or ->readahead() that need to hold invalidate_lock
Matthew Wilcox (Oracle)a5d4ad02021-03-10 10:34:00 -05002541 * while mapping blocks for IO so let's hold the lock here as
2542 * well to keep locking rules simple.
Jan Kara730633f2021-01-28 19:19:45 +01002543 */
2544 filemap_invalidate_lock_shared(mapping);
Matthew Wilcox (Oracle)a5d4ad02021-03-10 10:34:00 -05002545 error = filemap_add_folio(mapping, folio, index,
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002546 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)290e1a32022-05-12 17:37:01 -04002552 error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002553 if (error)
2554 goto error;
2555
Jan Kara730633f2021-01-28 19:19:45 +01002556 filemap_invalidate_unlock_shared(mapping);
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002557 folio_batch_add(fbatch, folio);
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002558 return 0;
2559error:
Jan Kara730633f2021-01-28 19:19:45 +01002560 filemap_invalidate_unlock_shared(mapping);
Matthew Wilcox (Oracle)a5d4ad02021-03-10 10:34:00 -05002561 folio_put(folio);
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002562 return error;
Kent Overstreet723ef242020-12-14 19:04:52 -08002563}
2564
Matthew Wilcox (Oracle)5963fe02021-02-24 12:02:32 -08002565static int filemap_readahead(struct kiocb *iocb, struct file *file,
Matthew Wilcox (Oracle)65bca532021-03-10 14:01:22 -05002566 struct address_space *mapping, struct folio *folio,
Matthew Wilcox (Oracle)5963fe02021-02-24 12:02:32 -08002567 pgoff_t last_index)
2568{
Matthew Wilcox (Oracle)65bca532021-03-10 14:01:22 -05002569 DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2570
Matthew Wilcox (Oracle)5963fe02021-02-24 12:02:32 -08002571 if (iocb->ki_flags & IOCB_NOIO)
2572 return -EAGAIN;
Matthew Wilcox (Oracle)65bca532021-03-10 14:01:22 -05002573 page_cache_async_ra(&ractl, folio, last_index - folio->index);
Matthew Wilcox (Oracle)5963fe02021-02-24 12:02:32 -08002574 return 0;
2575}
2576
David Howellsdd5b9d02023-02-08 18:18:17 +00002577static int filemap_get_pages(struct kiocb *iocb, size_t count,
2578 struct folio_batch *fbatch, bool need_uptodate)
Kent Overstreet06c04442020-12-14 19:04:56 -08002579{
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)cbd59c42021-02-24 12:01:59 -08002584 pgoff_t last_index;
Matthew Wilcox (Oracle)65bca532021-03-10 14:01:22 -05002585 struct folio *folio;
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002586 int err = 0;
Kent Overstreet06c04442020-12-14 19:04:56 -08002587
Qian Yingjin59565922023-02-08 10:24:00 +08002588 /* "last_index" is the index of the page beyond the end of the read */
David Howellsdd5b9d02023-02-08 18:18:17 +00002589 last_index = DIV_ROUND_UP(iocb->ki_pos + count, PAGE_SIZE);
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002590retry:
Kent Overstreet06c04442020-12-14 19:04:56 -08002591 if (fatal_signal_pending(current))
2592 return -EINTR;
2593
Qian Yingjin59565922023-02-08 10:24:00 +08002594 filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002595 if (!folio_batch_count(fbatch)) {
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002596 if (iocb->ki_flags & IOCB_NOIO)
2597 return -EAGAIN;
2598 page_cache_sync_readahead(mapping, ra, filp, index,
2599 last_index - index);
Qian Yingjin59565922023-02-08 10:24:00 +08002600 filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002601 }
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002602 if (!folio_batch_count(fbatch)) {
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002603 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2604 return -EAGAIN;
Matthew Wilcox (Oracle)a5d4ad02021-03-10 10:34:00 -05002605 err = filemap_create_folio(filp, mapping,
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002606 iocb->ki_pos >> PAGE_SHIFT, fbatch);
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002607 if (err == AOP_TRUNCATED_PAGE)
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002608 goto retry;
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002609 return err;
2610 }
Kent Overstreet06c04442020-12-14 19:04:56 -08002611
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002612 folio = fbatch->folios[folio_batch_count(fbatch) - 1];
Matthew Wilcox (Oracle)65bca532021-03-10 14:01:22 -05002613 if (folio_test_readahead(folio)) {
2614 err = filemap_readahead(iocb, filp, mapping, folio, last_index);
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002615 if (err)
2616 goto err;
2617 }
Matthew Wilcox (Oracle)65bca532021-03-10 14:01:22 -05002618 if (!folio_test_uptodate(folio)) {
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002619 if ((iocb->ki_flags & IOCB_WAITQ) &&
2620 folio_batch_count(fbatch) > 1)
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002621 iocb->ki_flags |= IOCB_NOWAIT;
David Howellsdd5b9d02023-02-08 18:18:17 +00002622 err = filemap_update_page(iocb, mapping, count, folio,
2623 need_uptodate);
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002624 if (err)
2625 goto err;
Kent Overstreet06c04442020-12-14 19:04:56 -08002626 }
2627
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002628 return 0;
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002629err:
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002630 if (err < 0)
Matthew Wilcox (Oracle)65bca532021-03-10 14:01:22 -05002631 folio_put(folio);
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002632 if (likely(--fbatch->nr))
Matthew Wilcox (Oracle)ff993ba2021-02-24 12:01:55 -08002633 return 0;
Matthew Wilcox (Oracle)4612aee2021-02-24 12:02:22 -08002634 if (err == AOP_TRUNCATED_PAGE)
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002635 goto retry;
2636 return err;
Kent Overstreet06c04442020-12-14 19:04:56 -08002637}
2638
Matthew Wilcox (Oracle)5ccc9442022-06-10 14:44:41 -04002639static 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 Dunlap485bb992006-06-23 02:03:49 -07002646/**
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002647 * 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 Dunlap485bb992006-06-23 02:03:49 -07002651 *
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002652 * Copies data from the page cache. If the data is not currently present,
Matthew Wilcox (Oracle)7e0a1262022-04-29 11:53:28 -04002653 * uses the readahead and read_folio address_space operations to fetch it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 *
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002655 * 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 Torvalds1da177e2005-04-16 15:20:36 -07002658 */
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002659ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2660 ssize_t already_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661{
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002662 struct file *filp = iocb->ki_filp;
Kent Overstreet06c04442020-12-14 19:04:56 -08002663 struct file_ra_state *ra = &filp->f_ra;
Christoph Hellwig36e78912008-02-08 04:21:24 -08002664 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 struct inode *inode = mapping->host;
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002666 struct folio_batch fbatch;
Matthew Wilcox (Oracle)ff993ba2021-02-24 12:01:55 -08002667 int i, error = 0;
Kent Overstreet06c04442020-12-14 19:04:56 -08002668 bool writably_mapped;
2669 loff_t isize, end_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670
Kent Overstreet723ef242020-12-14 19:04:52 -08002671 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
Linus Torvaldsd05c5f72016-12-14 12:45:25 -08002672 return 0;
Kent Overstreet3644e2d2020-12-18 04:07:11 -05002673 if (unlikely(!iov_iter_count(iter)))
2674 return 0;
2675
Wei Fangc2a97372016-10-07 17:01:52 -07002676 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002677 folio_batch_init(&fbatch);
Wei Fangc2a97372016-10-07 17:01:52 -07002678
Kent Overstreet06c04442020-12-14 19:04:56 -08002679 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 cond_resched();
Kent Overstreet06c04442020-12-14 19:04:56 -08002681
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 Hellwig87fa0f32021-02-24 12:02:42 -08002687 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
Kent Overstreet06c04442020-12-14 19:04:56 -08002688 iocb->ki_flags |= IOCB_NOWAIT;
2689
David Howells8c8387e2021-11-05 13:36:49 -07002690 if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2691 break;
2692
David Howellsdd5b9d02023-02-08 18:18:17 +00002693 error = filemap_get_pages(iocb, iter->count, &fbatch,
2694 iov_iter_is_pipe(iter));
Matthew Wilcox (Oracle)ff993ba2021-02-24 12:01:55 -08002695 if (error < 0)
Kent Overstreet06c04442020-12-14 19:04:56 -08002696 break;
Michal Hocko5abf1862017-02-03 13:13:29 -08002697
Kent Overstreet723ef242020-12-14 19:04:52 -08002698 /*
Kent Overstreet06c04442020-12-14 19:04:56 -08002699 * 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 Overstreet723ef242020-12-14 19:04:52 -08002705 */
Kent Overstreet06c04442020-12-14 19:04:56 -08002706 isize = i_size_read(inode);
2707 if (unlikely(iocb->ki_pos >= isize))
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002708 goto put_folios;
Kent Overstreet06c04442020-12-14 19:04:56 -08002709 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2710
Kent Overstreet06c04442020-12-14 19:04:56 -08002711 /*
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)5ccc9442022-06-10 14:44:41 -04002718 * When a read accesses the same folio several times, only
Kent Overstreet06c04442020-12-14 19:04:56 -08002719 * mark it as accessed the first time.
2720 */
Matthew Wilcox (Oracle)5ccc9442022-06-10 14:44:41 -04002721 if (!pos_same_folio(iocb->ki_pos, ra->prev_pos - 1,
2722 fbatch.folios[0]))
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002723 folio_mark_accessed(fbatch.folios[0]);
Kent Overstreet06c04442020-12-14 19:04:56 -08002724
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002725 for (i = 0; i < folio_batch_count(&fbatch); i++) {
2726 struct folio *folio = fbatch.folios[i];
Matthew Wilcox (Oracle)d996fc72021-10-31 22:22:19 -04002727 size_t fsize = folio_size(folio);
2728 size_t offset = iocb->ki_pos & (fsize - 1);
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002729 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
Matthew Wilcox (Oracle)d996fc72021-10-31 22:22:19 -04002730 fsize - offset);
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002731 size_t copied;
Kent Overstreet06c04442020-12-14 19:04:56 -08002732
Matthew Wilcox (Oracle)d996fc72021-10-31 22:22:19 -04002733 if (end_offset < folio_pos(folio))
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002734 break;
2735 if (i > 0)
Matthew Wilcox (Oracle)d996fc72021-10-31 22:22:19 -04002736 folio_mark_accessed(folio);
Kent Overstreet06c04442020-12-14 19:04:56 -08002737 /*
Matthew Wilcox (Oracle)d996fc72021-10-31 22:22:19 -04002738 * 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 Overstreet06c04442020-12-14 19:04:56 -08002741 */
Matthew Wilcox (Oracle)d996fc72021-10-31 22:22:19 -04002742 if (writably_mapped)
2743 flush_dcache_folio(folio);
Kent Overstreet06c04442020-12-14 19:04:56 -08002744
Matthew Wilcox (Oracle)d996fc72021-10-31 22:22:19 -04002745 copied = copy_folio_to_iter(folio, offset, bytes, iter);
Kent Overstreet06c04442020-12-14 19:04:56 -08002746
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002747 already_read += copied;
Kent Overstreet06c04442020-12-14 19:04:56 -08002748 iocb->ki_pos += copied;
2749 ra->prev_pos = iocb->ki_pos;
2750
2751 if (copied < bytes) {
2752 error = -EFAULT;
2753 break;
Kent Overstreet723ef242020-12-14 19:04:52 -08002754 }
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07002755 }
Matthew Wilcox (Oracle)25d6a232021-12-06 15:25:33 -05002756put_folios:
2757 for (i = 0; i < folio_batch_count(&fbatch); i++)
2758 folio_put(fbatch.folios[i]);
2759 folio_batch_init(&fbatch);
Kent Overstreet06c04442020-12-14 19:04:56 -08002760 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761
Krishna Kumar0c6aa262008-10-15 22:01:13 -07002762 file_accessed(filp);
Kent Overstreet06c04442020-12-14 19:04:56 -08002763
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002764 return already_read ? already_read : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765}
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002766EXPORT_SYMBOL_GPL(filemap_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767
Randy Dunlap485bb992006-06-23 02:03:49 -07002768/**
Al Viro6abd2322014-04-04 14:20:57 -04002769 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07002770 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04002771 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07002772 *
Al Viro6abd2322014-04-04 14:20:57 -04002773 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 * that can use the page cache directly.
Andreas Gruenbacher41da51b2019-11-21 23:25:07 +00002775 *
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 Rapoporta862f682019-03-05 15:48:42 -08002785 * Return:
2786 * * number of bytes copied, even for partial reads
Andreas Gruenbacher41da51b2019-11-21 23:25:07 +00002787 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 */
2789ssize_t
Al Viroed978a82014-03-05 22:53:04 -05002790generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791{
Nicolai Stangee7080a42016-03-25 14:22:14 -07002792 size_t count = iov_iter_count(iter);
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002793 ssize_t retval = 0;
Nicolai Stangee7080a42016-03-25 14:22:14 -07002794
2795 if (!count)
Christoph Hellwig826ea862021-02-24 12:02:45 -08002796 return 0; /* skip atime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797
Al Viro2ba48ce2015-04-09 13:52:01 -04002798 if (iocb->ki_flags & IOCB_DIRECT) {
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002799 struct file *file = iocb->ki_filp;
Al Viroed978a82014-03-05 22:53:04 -05002800 struct address_space *mapping = file->f_mapping;
2801 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05002803 if (iocb->ki_flags & IOCB_NOWAIT) {
Jens Axboe7a60d6d2021-04-29 22:55:21 -07002804 if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2805 iocb->ki_pos + count - 1))
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05002806 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 Hellwig826ea862021-02-24 12:02:45 -08002812 return retval;
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05002813 }
Al Viroed978a82014-03-05 22:53:04 -05002814
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002815 file_accessed(file);
2816
Al Viro5ecda132017-04-13 14:13:36 -04002817 retval = mapping->a_ops->direct_IO(iocb, iter);
Al Viroc3a69022016-10-10 13:26:27 -04002818 if (retval >= 0) {
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07002819 iocb->ki_pos += retval;
Al Viro5ecda132017-04-13 14:13:36 -04002820 count -= retval;
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002821 }
Pavel Begunkovab2125d2021-02-24 12:01:45 -08002822 if (retval != -EIOCBQUEUED)
2823 iov_iter_revert(iter, count - iov_iter_count(iter));
Josef Bacik66f998f2010-05-23 11:00:54 -04002824
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002825 /*
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 Wilcoxfbbbad42015-02-16 15:58:53 -08002831 * the rest of the read. Buffered reads will not work for
2832 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002833 */
Jens Axboe61d00172021-11-05 13:37:07 -07002834 if (retval < 0 || !count || IS_DAX(inode))
2835 return retval;
2836 if (iocb->ki_pos >= i_size_read(inode))
Christoph Hellwig826ea862021-02-24 12:02:45 -08002837 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 }
2839
Christoph Hellwig826ea862021-02-24 12:02:45 -08002840 return filemap_read(iocb, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841}
Al Viroed978a82014-03-05 22:53:04 -05002842EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843
David Howells07073eb2023-02-14 15:01:42 +00002844/*
2845 * Splice subpages from a folio into a pipe.
2846 */
2847size_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 */
2882ssize_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
2964out:
2965 folio_batch_release(&fbatch);
2966 file_accessed(in);
2967
2968 return total_spliced ? total_spliced : error;
2969}
David Howells7c8e01e2023-02-15 08:00:31 +00002970EXPORT_SYMBOL(filemap_splice_read);
David Howells07073eb2023-02-14 15:01:42 +00002971
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002972static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2973 struct address_space *mapping, struct folio *folio,
Matthew Wilcox (Oracle)54fa39a2021-02-25 17:15:52 -08002974 loff_t start, loff_t end, bool seek_data)
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08002975{
Matthew Wilcox (Oracle)54fa39a2021-02-25 17:15:52 -08002976 const struct address_space_operations *ops = mapping->a_ops;
2977 size_t offset, bsz = i_blocksize(mapping->host);
2978
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002979 if (xa_is_value(folio) || folio_test_uptodate(folio))
Matthew Wilcox (Oracle)54fa39a2021-02-25 17:15:52 -08002980 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)f5e64292020-12-17 00:12:26 -05002986 folio_lock(folio);
2987 if (unlikely(folio->mapping != mapping))
Matthew Wilcox (Oracle)54fa39a2021-02-25 17:15:52 -08002988 goto unlock;
2989
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002990 offset = offset_in_folio(folio, start) & ~(bsz - 1);
Matthew Wilcox (Oracle)54fa39a2021-02-25 17:15:52 -08002991
2992 do {
Matthew Wilcox (Oracle)2e7e80f2022-02-09 20:21:27 +00002993 if (ops->is_partially_uptodate(folio, offset, bsz) ==
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002994 seek_data)
Matthew Wilcox (Oracle)54fa39a2021-02-25 17:15:52 -08002995 break;
2996 start = (start + bsz) & ~(bsz - 1);
2997 offset += bsz;
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05002998 } while (offset < folio_size(folio));
Matthew Wilcox (Oracle)54fa39a2021-02-25 17:15:52 -08002999unlock:
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05003000 folio_unlock(folio);
Matthew Wilcox (Oracle)54fa39a2021-02-25 17:15:52 -08003001 rcu_read_lock();
3002 return start;
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003003}
3004
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05003005static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003006{
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05003007 if (xa_is_value(folio))
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003008 return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05003009 return folio_size(folio);
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003010}
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 Molnarf0953a12021-05-06 18:06:47 -07003025 * Return: The requested offset on success, or -ENXIO if @whence specifies
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003026 * 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 */
3030loff_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 Dickinsed98b012021-04-23 14:29:00 -07003034 pgoff_t max = (end - 1) >> PAGE_SHIFT;
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003035 bool seek_data = (whence == SEEK_DATA);
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05003036 struct folio *folio;
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003037
3038 if (end <= start)
3039 return -ENXIO;
3040
3041 rcu_read_lock();
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05003042 while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
Hugh Dickinsed98b012021-04-23 14:29:00 -07003043 loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05003044 size_t seek_size;
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003045
3046 if (start < pos) {
3047 if (!seek_data)
3048 goto unlock;
3049 start = pos;
3050 }
3051
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05003052 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)54fa39a2021-02-25 17:15:52 -08003055 seek_data);
3056 if (start < pos)
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003057 goto unlock;
Hugh Dickinsed98b012021-04-23 14:29:00 -07003058 if (start >= end)
3059 break;
3060 if (seek_size > PAGE_SIZE)
3061 xas_set(&xas, pos >> PAGE_SHIFT);
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05003062 if (!xa_is_value(folio))
3063 folio_put(folio);
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003064 }
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003065 if (seek_data)
Hugh Dickinsed98b012021-04-23 14:29:00 -07003066 start = -ENXIO;
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003067unlock:
3068 rcu_read_unlock();
Matthew Wilcox (Oracle)f5e64292020-12-17 00:12:26 -05003069 if (folio && !xa_is_value(folio))
3070 folio_put(folio);
Matthew Wilcox (Oracle)41139aa2021-02-25 17:15:48 -08003071 if (start > end)
3072 return end;
3073 return start;
3074}
3075
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076#ifdef CONFIG_MMU
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077#define MMAP_LOTSAMISS (100)
Linus Torvaldsef00e082009-06-16 15:31:25 -07003078/*
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003079 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003080 * @vmf - the vm_fault for this fault.
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003081 * @folio - the folio to lock.
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003082 * @fpin - the pointer to the file we may pin (or is already pinned).
3083 *
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003084 * 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 Torvaldsef00e082009-06-16 15:31:25 -07003089 */
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003090static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003091 struct file **fpin)
3092{
Matthew Wilcox (Oracle)7c23c782021-03-01 19:38:25 -05003093 if (folio_trylock(folio))
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003094 return 1;
3095
Linus Torvalds8b0f9fa2019-03-15 11:26:07 -07003096 /*
3097 * NOTE! This will make us return with VM_FAULT_RETRY, but with
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07003098 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
Linus Torvalds8b0f9fa2019-03-15 11:26:07 -07003099 * is supposed to work. We have way too many special cases..
3100 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003101 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)af7f29d2020-12-08 00:07:31 -05003106 if (__folio_lock_killable(folio)) {
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003107 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07003108 * We didn't have the right flags to drop the mmap_lock,
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003109 * but all fault_handlers only check for fatal signals
3110 * if we return VM_FAULT_RETRY, so we need to drop the
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07003111 * mmap_lock here and return 0 if we don't have a fpin.
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003112 */
3113 if (*fpin == NULL)
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07003114 mmap_read_unlock(vmf->vma->vm_mm);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003115 return 0;
3116 }
3117 } else
Matthew Wilcox (Oracle)7c23c782021-03-01 19:38:25 -05003118 __folio_lock(folio);
3119
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003120 return 1;
3121}
3122
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003123/*
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 */
3130static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
Linus Torvaldsef00e082009-06-16 15:31:25 -07003131{
Josef Bacik2a1180f2019-03-13 11:44:18 -07003132 struct file *file = vmf->vma->vm_file;
3133 struct file_ra_state *ra = &file->f_ra;
Linus Torvaldsef00e082009-06-16 15:31:25 -07003134 struct address_space *mapping = file->f_mapping;
Matthew Wilcox (Oracle)fcd9ae42021-04-07 21:18:55 +01003135 DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003136 struct file *fpin = NULL;
Matthew Wilcox (Oracle)dcfa24b2022-05-25 14:23:45 -04003137 unsigned long vm_flags = vmf->vma->vm_flags;
Kirill A. Shutemove630bfac2020-08-14 17:31:27 -07003138 unsigned int mmap_miss;
Linus Torvaldsef00e082009-06-16 15:31:25 -07003139
Matthew Wilcox (Oracle)4687fdb2021-07-24 23:37:13 -04003140#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3141 /* Use the readahead code, even if readahead is disabled */
Matthew Wilcox (Oracle)dcfa24b2022-05-25 14:23:45 -04003142 if (vm_flags & VM_HUGEPAGE) {
Matthew Wilcox (Oracle)4687fdb2021-07-24 23:37:13 -04003143 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)dcfa24b2022-05-25 14:23:45 -04003150 if (!(vm_flags & VM_RAND_READ))
Matthew Wilcox (Oracle)4687fdb2021-07-24 23:37:13 -04003151 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 Torvaldsef00e082009-06-16 15:31:25 -07003158 /* If we don't want any read-ahead, don't bother */
Matthew Wilcox (Oracle)dcfa24b2022-05-25 14:23:45 -04003159 if (vm_flags & VM_RAND_READ)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003160 return fpin;
Wu Fengguang275b12b2011-05-24 17:12:28 -07003161 if (!ra->ra_pages)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003162 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07003163
Matthew Wilcox (Oracle)dcfa24b2022-05-25 14:23:45 -04003164 if (vm_flags & VM_SEQ_READ) {
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003165 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
Matthew Wilcox (Oracle)fcd9ae42021-04-07 21:18:55 +01003166 page_cache_sync_ra(&ractl, ra->ra_pages);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003167 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07003168 }
3169
Andi Kleen207d04b2011-05-24 17:12:29 -07003170 /* Avoid banging the cache line if not needed */
Kirill A. Shutemove630bfac2020-08-14 17:31:27 -07003171 mmap_miss = READ_ONCE(ra->mmap_miss);
3172 if (mmap_miss < MMAP_LOTSAMISS * 10)
3173 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
Linus Torvaldsef00e082009-06-16 15:31:25 -07003174
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. Shutemove630bfac2020-08-14 17:31:27 -07003179 if (mmap_miss > MMAP_LOTSAMISS)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003180 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07003181
Wu Fengguangd30a1102009-06-16 15:31:30 -07003182 /*
3183 * mmap read-around
3184 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003185 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
David Howellsdb660d42020-10-15 20:06:31 -07003186 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
Roman Gushchin600e19a2015-11-05 18:47:08 -08003187 ra->size = ra->ra_pages;
3188 ra->async_size = ra->ra_pages / 4;
David Howellsdb660d42020-10-15 20:06:31 -07003189 ractl._index = ra->start;
Matthew Wilcox (Oracle)56a4d672021-07-24 23:26:14 -04003190 page_cache_ra_order(&ractl, ra, 0);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003191 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07003192}
3193
3194/*
3195 * Asynchronous readahead happens when we find the page and PG_readahead,
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003196 * so we want to possibly extend the readahead further. We return the file that
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07003197 * was pinned if we have to drop the mmap_lock in order to do IO.
Linus Torvaldsef00e082009-06-16 15:31:25 -07003198 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003199static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
Matthew Wilcox (Oracle)79598ce2021-07-29 14:57:01 -04003200 struct folio *folio)
Linus Torvaldsef00e082009-06-16 15:31:25 -07003201{
Josef Bacik2a1180f2019-03-13 11:44:18 -07003202 struct file *file = vmf->vma->vm_file;
3203 struct file_ra_state *ra = &file->f_ra;
Matthew Wilcox (Oracle)79598ce2021-07-29 14:57:01 -04003204 DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003205 struct file *fpin = NULL;
Kirill A. Shutemove630bfac2020-08-14 17:31:27 -07003206 unsigned int mmap_miss;
Linus Torvaldsef00e082009-06-16 15:31:25 -07003207
3208 /* If we don't want any read-ahead, don't bother */
Jan Kara5c72fee2020-04-01 21:04:40 -07003209 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003210 return fpin;
Matthew Wilcox (Oracle)79598ce2021-07-29 14:57:01 -04003211
Kirill A. Shutemove630bfac2020-08-14 17:31:27 -07003212 mmap_miss = READ_ONCE(ra->mmap_miss);
3213 if (mmap_miss)
3214 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
Matthew Wilcox (Oracle)79598ce2021-07-29 14:57:01 -04003215
3216 if (folio_test_readahead(folio)) {
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003217 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
Matthew Wilcox (Oracle)79598ce2021-07-29 14:57:01 -04003218 page_cache_async_ra(&ractl, folio, ra->ra_pages);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003219 }
3220 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07003221}
3222
Randy Dunlap485bb992006-06-23 02:03:49 -07003223/**
Nick Piggin54cb8822007-07-19 01:46:59 -07003224 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07003225 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07003226 *
Nick Piggin54cb8822007-07-19 01:46:59 -07003227 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228 * 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 Cassella9a95f3c2014-08-06 16:07:24 -07003233 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07003234 * vma->vm_mm->mmap_lock must be held on entry.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003235 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07003236 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003237 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003238 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07003239 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
Paul Cassella9a95f3c2014-08-06 16:07:24 -07003240 * has not been released.
3241 *
3242 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Mike Rapoporta862f682019-03-05 15:48:42 -08003243 *
3244 * Return: bitwise-OR of %VM_FAULT_ codes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003245 */
Souptick Joarder2bcd6452018-06-07 17:08:00 -07003246vm_fault_t filemap_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247{
3248 int error;
Dave Jiang11bac802017-02-24 14:56:41 -08003249 struct file *file = vmf->vma->vm_file;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003250 struct file *fpin = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251 struct address_space *mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252 struct inode *inode = mapping->host;
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003253 pgoff_t max_idx, index = vmf->pgoff;
3254 struct folio *folio;
Souptick Joarder2bcd6452018-06-07 17:08:00 -07003255 vm_fault_t ret = 0;
Jan Kara730633f2021-01-28 19:19:45 +01003256 bool mapping_locked = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003258 max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3259 if (unlikely(index >= max_idx))
Linus Torvalds5307cc12007-10-31 09:19:46 -07003260 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262 /*
Johannes Weiner494264202013-10-16 13:46:59 -07003263 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264 */
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003265 folio = filemap_get_folio(mapping, index);
3266 if (likely(folio)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267 /*
Jan Kara730633f2021-01-28 19:19:45 +01003268 * We found the page, so try async readahead before waiting for
3269 * the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003270 */
Jan Kara730633f2021-01-28 19:19:45 +01003271 if (!(vmf->flags & FAULT_FLAG_TRIED))
Matthew Wilcox (Oracle)79598ce2021-07-29 14:57:01 -04003272 fpin = do_async_mmap_readahead(vmf, folio);
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003273 if (unlikely(!folio_test_uptodate(folio))) {
Jan Kara730633f2021-01-28 19:19:45 +01003274 filemap_invalidate_lock_shared(mapping);
3275 mapping_locked = true;
3276 }
3277 } else {
Linus Torvaldsef00e082009-06-16 15:31:25 -07003278 /* No page in the page cache at all */
Linus Torvaldsef00e082009-06-16 15:31:25 -07003279 count_vm_event(PGMAJFAULT);
Roman Gushchin22621852017-07-06 15:40:25 -07003280 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07003281 ret = VM_FAULT_MAJOR;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003282 fpin = do_sync_mmap_readahead(vmf);
Linus Torvaldsef00e082009-06-16 15:31:25 -07003283retry_find:
Jan Kara730633f2021-01-28 19:19:45 +01003284 /*
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003285 * See comment in filemap_create_folio() why we need
Jan Kara730633f2021-01-28 19:19:45 +01003286 * invalidate_lock
3287 */
3288 if (!mapping_locked) {
3289 filemap_invalidate_lock_shared(mapping);
3290 mapping_locked = true;
3291 }
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003292 folio = __filemap_get_folio(mapping, index,
Josef Bacika75d4c32019-03-13 11:44:14 -07003293 FGP_CREAT|FGP_FOR_MMAP,
3294 vmf->gfp_mask);
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003295 if (!folio) {
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003296 if (fpin)
3297 goto out_retry;
Jan Kara730633f2021-01-28 19:19:45 +01003298 filemap_invalidate_unlock_shared(mapping);
Matthew Wilcox (Oracle)e520e932020-04-01 21:04:53 -07003299 return VM_FAULT_OOM;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003300 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301 }
3302
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003303 if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003304 goto out_retry;
Michel Lespinasseb522c942010-10-26 14:21:56 -07003305
3306 /* Did it get truncated? */
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003307 if (unlikely(folio->mapping != mapping)) {
3308 folio_unlock(folio);
3309 folio_put(folio);
Michel Lespinasseb522c942010-10-26 14:21:56 -07003310 goto retry_find;
3311 }
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003312 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
Michel Lespinasseb522c942010-10-26 14:21:56 -07003313
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314 /*
Nick Piggind00806b2007-07-19 01:46:57 -07003315 * 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 Torvalds1da177e2005-04-16 15:20:36 -07003317 */
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003318 if (unlikely(!folio_test_uptodate(folio))) {
Jan Kara730633f2021-01-28 19:19:45 +01003319 /*
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)e292e6d2021-03-10 10:46:41 -05003326 folio_unlock(folio);
3327 folio_put(folio);
Jan Kara730633f2021-01-28 19:19:45 +01003328 goto retry_find;
3329 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330 goto page_not_uptodate;
Jan Kara730633f2021-01-28 19:19:45 +01003331 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332
Linus Torvaldsef00e082009-06-16 15:31:25 -07003333 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07003334 * We've made it this far and we had to drop our mmap_lock, now is the
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003335 * 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)e292e6d2021-03-10 10:46:41 -05003339 folio_unlock(folio);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003340 goto out_retry;
3341 }
Jan Kara730633f2021-01-28 19:19:45 +01003342 if (mapping_locked)
3343 filemap_invalidate_unlock_shared(mapping);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003344
3345 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07003346 * Found the page and have a reference on it.
3347 * We must recheck i_size under page lock.
3348 */
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003349 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 Torvalds5307cc12007-10-31 09:19:46 -07003353 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07003354 }
3355
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003356 vmf->page = folio_file_page(folio, index);
Nick Piggin83c54072007-07-19 01:47:05 -07003357 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 /*
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 Bacik6b4c9f42019-03-13 11:44:22 -07003366 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
Matthew Wilcox (Oracle)290e1a32022-05-12 17:37:01 -04003367 error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003368 if (fpin)
3369 goto out_retry;
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003370 folio_put(folio);
Nick Piggind00806b2007-07-19 01:46:57 -07003371
3372 if (!error || error == AOP_TRUNCATED_PAGE)
3373 goto retry_find;
Jan Kara730633f2021-01-28 19:19:45 +01003374 filemap_invalidate_unlock_shared(mapping);
Nick Piggind00806b2007-07-19 01:46:57 -07003375
Nick Piggind0217ac2007-07-19 01:47:03 -07003376 return VM_FAULT_SIGBUS;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003377
3378out_retry:
3379 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07003380 * We dropped the mmap_lock, we need to return to the fault handler to
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003381 * re-find the vma and come back and find our hopefully still populated
3382 * page.
3383 */
Matthew Wilcox (Oracle)e292e6d2021-03-10 10:46:41 -05003384 if (folio)
3385 folio_put(folio);
Jan Kara730633f2021-01-28 19:19:45 +01003386 if (mapping_locked)
3387 filemap_invalidate_unlock_shared(mapping);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07003388 if (fpin)
3389 fput(fpin);
3390 return ret | VM_FAULT_RETRY;
Nick Piggin54cb8822007-07-19 01:46:59 -07003391}
3392EXPORT_SYMBOL(filemap_fault);
3393
Matthew Wilcox (Oracle)8808eca2023-01-16 19:39:39 +00003394static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio,
3395 pgoff_t start)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07003396{
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003397 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)8808eca2023-01-16 19:39:39 +00003401 folio_unlock(folio);
3402 folio_put(folio);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003403 return true;
3404 }
3405
Matthew Wilcox (Oracle)8808eca2023-01-16 19:39:39 +00003406 if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
3407 struct page *page = folio_file_page(folio, start);
Yang Shie0f43fa2021-11-05 13:41:04 -07003408 vm_fault_t ret = do_set_pmd(vmf, page);
3409 if (!ret) {
3410 /* The page is mapped successfully, reference consumed. */
Matthew Wilcox (Oracle)8808eca2023-01-16 19:39:39 +00003411 folio_unlock(folio);
Yang Shie0f43fa2021-11-05 13:41:04 -07003412 return true;
3413 }
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003414 }
3415
Qi Zheng03c4f202021-11-05 13:38:38 -07003416 if (pmd_none(*vmf->pmd))
3417 pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003418
3419 /* See comment in handle_pte_fault() */
3420 if (pmd_devmap_trans_unstable(vmf->pmd)) {
Matthew Wilcox (Oracle)8808eca2023-01-16 19:39:39 +00003421 folio_unlock(folio);
3422 folio_put(folio);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003423 return true;
3424 }
3425
3426 return false;
3427}
3428
Matthew Wilcox (Oracle)820b05e2021-03-12 23:46:45 -05003429static struct folio *next_uptodate_page(struct folio *folio,
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003430 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)9184a302021-03-12 23:33:43 -05003436 if (!folio)
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003437 return NULL;
Matthew Wilcox (Oracle)9184a302021-03-12 23:33:43 -05003438 if (xas_retry(xas, folio))
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003439 continue;
Matthew Wilcox (Oracle)9184a302021-03-12 23:33:43 -05003440 if (xa_is_value(folio))
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003441 continue;
Matthew Wilcox (Oracle)9184a302021-03-12 23:33:43 -05003442 if (folio_test_locked(folio))
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003443 continue;
Matthew Wilcox (Oracle)9184a302021-03-12 23:33:43 -05003444 if (!folio_try_get_rcu(folio))
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003445 continue;
3446 /* Has the page moved or been split? */
Matthew Wilcox (Oracle)9184a302021-03-12 23:33:43 -05003447 if (unlikely(folio != xas_reload(xas)))
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003448 goto skip;
Matthew Wilcox (Oracle)9184a302021-03-12 23:33:43 -05003449 if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003450 goto skip;
Matthew Wilcox (Oracle)9184a302021-03-12 23:33:43 -05003451 if (!folio_trylock(folio))
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003452 goto skip;
Matthew Wilcox (Oracle)9184a302021-03-12 23:33:43 -05003453 if (folio->mapping != mapping)
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003454 goto unlock;
Matthew Wilcox (Oracle)9184a302021-03-12 23:33:43 -05003455 if (!folio_test_uptodate(folio))
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003456 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)820b05e2021-03-12 23:46:45 -05003460 return folio;
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003461unlock:
Matthew Wilcox (Oracle)9184a302021-03-12 23:33:43 -05003462 folio_unlock(folio);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003463skip:
Matthew Wilcox (Oracle)9184a302021-03-12 23:33:43 -05003464 folio_put(folio);
3465 } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003466
3467 return NULL;
3468}
3469
Matthew Wilcox (Oracle)820b05e2021-03-12 23:46:45 -05003470static inline struct folio *first_map_page(struct address_space *mapping,
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003471 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)820b05e2021-03-12 23:46:45 -05003478static inline struct folio *next_map_page(struct address_space *mapping,
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003479 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
3486vm_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. Shutemovf1820362014-04-07 15:37:19 -07003491 struct address_space *mapping = file->f_mapping;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07003492 pgoff_t last_pgoff = start_pgoff;
Will Deacon9d3af4b2021-01-14 15:24:19 +00003493 unsigned long addr;
Matthew Wilcox070e807c2018-05-17 00:08:30 -04003494 XA_STATE(xas, &mapping->i_pages, start_pgoff);
Matthew Wilcox (Oracle)820b05e2021-03-12 23:46:45 -05003495 struct folio *folio;
3496 struct page *page;
Kirill A. Shutemove630bfac2020-08-14 17:31:27 -07003497 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003498 vm_fault_t ret = 0;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07003499
3500 rcu_read_lock();
Matthew Wilcox (Oracle)820b05e2021-03-12 23:46:45 -05003501 folio = first_map_page(mapping, &xas, end_pgoff);
3502 if (!folio)
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003503 goto out;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07003504
Matthew Wilcox (Oracle)8808eca2023-01-16 19:39:39 +00003505 if (filemap_map_pmd(vmf, folio, start_pgoff)) {
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003506 ret = VM_FAULT_NOPAGE;
3507 goto out;
3508 }
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07003509
Will Deacon9d3af4b2021-01-14 15:24:19 +00003510 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. Shutemovf9ce0be2020-12-19 15:19:23 +03003512 do {
Matthew Wilcox (Oracle)6b24ca42020-06-27 22:19:08 -04003513again:
Matthew Wilcox (Oracle)820b05e2021-03-12 23:46:45 -05003514 page = folio_file_page(folio, xas.xa_index);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003515 if (PageHWPoison(page))
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07003516 goto unlock;
3517
Kirill A. Shutemove630bfac2020-08-14 17:31:27 -07003518 if (mmap_miss > 0)
3519 mmap_miss--;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003520
Will Deacon9d3af4b2021-01-14 15:24:19 +00003521 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003522 vmf->pte += xas.xa_index - last_pgoff;
Matthew Wilcox070e807c2018-05-17 00:08:30 -04003523 last_pgoff = xas.xa_index;
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003524
Peter Xu5c041f52022-05-12 20:22:52 -07003525 /*
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. Shutemovf9ce0be2020-12-19 15:19:23 +03003530 if (!pte_none(*vmf->pte))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07003531 goto unlock;
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003532
Will Deacon46bdb422020-11-24 18:48:26 +00003533 /* We're about to handle the fault */
Will Deacon9d3af4b2021-01-14 15:24:19 +00003534 if (vmf->address == addr)
Will Deacon46bdb422020-11-24 18:48:26 +00003535 ret = VM_FAULT_NOPAGE;
Will Deacon46bdb422020-11-24 18:48:26 +00003536
Will Deacon9d3af4b2021-01-14 15:24:19 +00003537 do_set_pte(vmf, page, addr);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003538 /* no need to invalidate: a not-present page won't be cached */
Will Deacon9d3af4b2021-01-14 15:24:19 +00003539 update_mmu_cache(vma, addr, vmf->pte);
Matthew Wilcox (Oracle)6b24ca42020-06-27 22:19:08 -04003540 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)820b05e2021-03-12 23:46:45 -05003545 folio_unlock(folio);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003546 continue;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07003547unlock:
Matthew Wilcox (Oracle)6b24ca42020-06-27 22:19:08 -04003548 if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3549 xas.xa_index++;
3550 goto again;
3551 }
Matthew Wilcox (Oracle)820b05e2021-03-12 23:46:45 -05003552 folio_unlock(folio);
3553 folio_put(folio);
3554 } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003555 pte_unmap_unlock(vmf->pte, vmf->ptl);
3556out:
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07003557 rcu_read_unlock();
Kirill A. Shutemove630bfac2020-08-14 17:31:27 -07003558 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03003559 return ret;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07003560}
3561EXPORT_SYMBOL(filemap_map_pages);
3562
Souptick Joarder2bcd6452018-06-07 17:08:00 -07003563vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
Jan Kara4fcf1c62012-06-12 16:20:29 +02003564{
Christoph Hellwig5df1a672020-11-16 14:33:37 +01003565 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
Matthew Wilcox (Oracle)960ea972021-03-12 23:57:44 -05003566 struct folio *folio = page_folio(vmf->page);
Souptick Joarder2bcd6452018-06-07 17:08:00 -07003567 vm_fault_t ret = VM_FAULT_LOCKED;
Jan Kara4fcf1c62012-06-12 16:20:29 +02003568
Christoph Hellwig5df1a672020-11-16 14:33:37 +01003569 sb_start_pagefault(mapping->host->i_sb);
Dave Jiang11bac802017-02-24 14:56:41 -08003570 file_update_time(vmf->vma->vm_file);
Matthew Wilcox (Oracle)960ea972021-03-12 23:57:44 -05003571 folio_lock(folio);
3572 if (folio->mapping != mapping) {
3573 folio_unlock(folio);
Jan Kara4fcf1c62012-06-12 16:20:29 +02003574 ret = VM_FAULT_NOPAGE;
3575 goto out;
3576 }
Jan Kara14da9202012-06-12 16:20:37 +02003577 /*
Matthew Wilcox (Oracle)960ea972021-03-12 23:57:44 -05003578 * We mark the folio dirty already here so that when freeze is in
Jan Kara14da9202012-06-12 16:20:37 +02003579 * progress, we are guaranteed that writeback during freezing will
Matthew Wilcox (Oracle)960ea972021-03-12 23:57:44 -05003580 * see the dirty folio and writeprotect it again.
Jan Kara14da9202012-06-12 16:20:37 +02003581 */
Matthew Wilcox (Oracle)960ea972021-03-12 23:57:44 -05003582 folio_mark_dirty(folio);
3583 folio_wait_stable(folio);
Jan Kara4fcf1c62012-06-12 16:20:29 +02003584out:
Christoph Hellwig5df1a672020-11-16 14:33:37 +01003585 sb_end_pagefault(mapping->host->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02003586 return ret;
3587}
Jan Kara4fcf1c62012-06-12 16:20:29 +02003588
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04003589const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003590 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07003591 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02003592 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593};
3594
3595/* This is used for a general mmap of a disk file */
3596
Zhiyuan Dai68d68ff2021-05-04 18:40:12 -07003597int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598{
3599 struct address_space *mapping = file->f_mapping;
3600
Matthew Wilcox (Oracle)7e0a1262022-04-29 11:53:28 -04003601 if (!mapping->a_ops->read_folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003602 return -ENOEXEC;
3603 file_accessed(file);
3604 vma->vm_ops = &generic_file_vm_ops;
3605 return 0;
3606}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003607
3608/*
3609 * This is for filesystems which do not implement ->writepage.
3610 */
3611int 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 Joarder4b96a372018-10-26 15:04:03 -07003618vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
Arnd Bergmann453972282018-04-13 15:35:27 -07003619{
Souptick Joarder4b96a372018-10-26 15:04:03 -07003620 return VM_FAULT_SIGBUS;
Arnd Bergmann453972282018-04-13 15:35:27 -07003621}
Zhiyuan Dai68d68ff2021-05-04 18:40:12 -07003622int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623{
3624 return -ENOSYS;
3625}
Zhiyuan Dai68d68ff2021-05-04 18:40:12 -07003626int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003627{
3628 return -ENOSYS;
3629}
3630#endif /* CONFIG_MMU */
3631
Arnd Bergmann453972282018-04-13 15:35:27 -07003632EXPORT_SYMBOL(filemap_page_mkwrite);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003633EXPORT_SYMBOL(generic_file_mmap);
3634EXPORT_SYMBOL(generic_file_readonly_mmap);
3635
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003636static struct folio *do_read_cache_folio(struct address_space *mapping,
Matthew Wilcox (Oracle)e9b5b232022-05-01 21:39:29 -04003637 pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
Sasha Levin67f9fd92014-04-03 14:48:18 -07003638{
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003639 struct folio *folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003640 int err;
Matthew Wilcox (Oracle)07950002022-05-08 15:07:11 -04003641
3642 if (!filler)
3643 filler = mapping->a_ops->read_folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003644repeat:
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003645 folio = filemap_get_folio(mapping, index);
3646 if (!folio) {
3647 folio = filemap_alloc_folio(gfp, 0);
3648 if (!folio)
Nick Piggineb2be182007-10-16 01:24:57 -07003649 return ERR_PTR(-ENOMEM);
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003650 err = filemap_add_folio(mapping, folio, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07003651 if (unlikely(err)) {
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003652 folio_put(folio);
Nick Piggineb2be182007-10-16 01:24:57 -07003653 if (err == -EEXIST)
3654 goto repeat;
Matthew Wilcox22ecdb42017-12-04 04:02:00 -05003655 /* Presumably ENOMEM for xarray node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656 return ERR_PTR(err);
3657 }
Mel Gorman32b63522016-03-15 14:55:36 -07003658
Matthew Wilcox (Oracle)9bc3e862022-05-12 17:12:21 -04003659 goto filler;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660 }
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003661 if (folio_test_uptodate(folio))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662 goto out;
3663
Matthew Wilcox (Oracle)81f4c032021-12-23 15:17:28 -05003664 if (!folio_trylock(folio)) {
3665 folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3666 goto repeat;
3667 }
Mel Gormanebded022016-03-15 14:55:39 -07003668
Matthew Wilcox (Oracle)81f4c032021-12-23 15:17:28 -05003669 /* Folio was truncated from mapping */
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003670 if (!folio->mapping) {
3671 folio_unlock(folio);
3672 folio_put(folio);
Mel Gorman32b63522016-03-15 14:55:36 -07003673 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674 }
Mel Gormanebded022016-03-15 14:55:39 -07003675
3676 /* Someone else locked and filled the page in a very small window */
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003677 if (folio_test_uptodate(folio)) {
3678 folio_unlock(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679 goto out;
3680 }
Xianting Tianfaffdfa2020-04-01 21:04:47 -07003681
Matthew Wilcox (Oracle)9bc3e862022-05-12 17:12:21 -04003682filler:
Matthew Wilcox (Oracle)290e1a32022-05-12 17:37:01 -04003683 err = filemap_read_folio(file, filler, folio);
Matthew Wilcox (Oracle)1dfa24a2022-05-12 17:47:06 -04003684 if (err) {
Matthew Wilcox (Oracle)9bc3e862022-05-12 17:12:21 -04003685 folio_put(folio);
Matthew Wilcox (Oracle)1dfa24a2022-05-12 17:47:06 -04003686 if (err == AOP_TRUNCATED_PAGE)
3687 goto repeat;
Matthew Wilcox (Oracle)9bc3e862022-05-12 17:12:21 -04003688 return ERR_PTR(err);
3689 }
Mel Gorman32b63522016-03-15 14:55:36 -07003690
David Howellsc855ff32007-05-09 13:42:20 +01003691out:
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003692 folio_mark_accessed(folio);
3693 return folio;
Nick Piggin6fe69002007-05-06 14:49:04 -07003694}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003695
3696/**
Matthew Wilcox (Oracle)e9b5b232022-05-01 21:39:29 -04003697 * 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 Torvalds0531b2a2010-01-27 09:20:03 -08003702 *
Matthew Wilcox (Oracle)e9b5b232022-05-01 21:39:29 -04003703 * 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 Torvalds0531b2a2010-01-27 09:20:03 -08003705 *
Matthew Wilcox (Oracle)e9b5b232022-05-01 21:39:29 -04003706 * If the filler function returns an error, it will be returned to the
3707 * caller.
Mike Rapoporta862f682019-03-05 15:48:42 -08003708 *
Matthew Wilcox (Oracle)e9b5b232022-05-01 21:39:29 -04003709 * Context: May sleep. Expects mapping->invalidate_lock to be held.
3710 * Return: An uptodate folio on success, ERR_PTR() on failure.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003711 */
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003712struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
Matthew Wilcox (Oracle)e9b5b232022-05-01 21:39:29 -04003713 filler_t filler, struct file *file)
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003714{
Matthew Wilcox (Oracle)e9b5b232022-05-01 21:39:29 -04003715 return do_read_cache_folio(mapping, index, filler, file,
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003716 mapping_gfp_mask(mapping));
3717}
3718EXPORT_SYMBOL(read_cache_folio);
3719
Matthew Wilcox (Oracle)3e629592023-02-06 16:25:19 +00003720/**
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 */
3737struct 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}
3742EXPORT_SYMBOL(mapping_read_folio_gfp);
3743
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003744static struct page *do_read_cache_page(struct address_space *mapping,
Matthew Wilcox (Oracle)e9b5b232022-05-01 21:39:29 -04003745 pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003746{
3747 struct folio *folio;
3748
Matthew Wilcox (Oracle)e9b5b232022-05-01 21:39:29 -04003749 folio = do_read_cache_folio(mapping, index, filler, file, gfp);
Matthew Wilcox (Oracle)539a3322020-12-16 11:45:30 -05003750 if (IS_ERR(folio))
3751 return &folio->page;
3752 return folio_file_page(folio, index);
3753}
3754
Sasha Levin67f9fd92014-04-03 14:48:18 -07003755struct page *read_cache_page(struct address_space *mapping,
Matthew Wilcox (Oracle)e9b5b232022-05-01 21:39:29 -04003756 pgoff_t index, filler_t *filler, struct file *file)
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003757{
Matthew Wilcox (Oracle)e9b5b232022-05-01 21:39:29 -04003758 return do_read_cache_page(mapping, index, filler, file,
Christoph Hellwigd322a8e2019-07-11 20:55:17 -07003759 mapping_gfp_mask(mapping));
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003760}
Sasha Levin67f9fd92014-04-03 14:48:18 -07003761EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003762
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 Kleikampe6f67b82011-12-21 11:05:48 -06003770 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003771 *
3772 * If the page does not get brought uptodate, return -EIO.
Mike Rapoporta862f682019-03-05 15:48:42 -08003773 *
Jan Kara730633f2021-01-28 19:19:45 +01003774 * The function expects mapping->invalidate_lock to be already held.
3775 *
Mike Rapoporta862f682019-03-05 15:48:42 -08003776 * Return: up to date page on success, ERR_PTR() on failure.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003777 */
3778struct page *read_cache_page_gfp(struct address_space *mapping,
3779 pgoff_t index,
3780 gfp_t gfp)
3781{
Christoph Hellwig6c45b452019-07-11 20:55:20 -07003782 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003783}
3784EXPORT_SYMBOL(read_cache_page_gfp);
3785
Konstantin Khlebnikova92853b2019-11-30 17:49:44 -08003786/*
3787 * Warn about a page cache invalidation failure during a direct I/O write.
3788 */
3789void dio_warn_stale_pagecache(struct file *filp)
3790{
3791 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3792 char pathname[128];
Konstantin Khlebnikova92853b2019-11-30 17:49:44 -08003793 char *path;
3794
Christoph Hellwig5df1a672020-11-16 14:33:37 +01003795 errseq_set(&filp->f_mapping->wb_err, -EIO);
Konstantin Khlebnikova92853b2019-11-30 17:49:44 -08003796 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 Torvalds1da177e2005-04-16 15:20:36 -07003806ssize_t
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07003807generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808{
3809 struct file *file = iocb->ki_filp;
3810 struct address_space *mapping = file->f_mapping;
3811 struct inode *inode = mapping->host;
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07003812 loff_t pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003813 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07003814 size_t write_len;
3815 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816
Al Viro0c949332014-03-22 06:51:37 -04003817 write_len = iov_iter_count(from);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003818 end = (pos + write_len - 1) >> PAGE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07003819
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05003820 if (iocb->ki_flags & IOCB_NOWAIT) {
3821 /* If there are pages to writeback, return */
Christoph Hellwig5df1a672020-11-16 14:33:37 +01003822 if (filemap_range_has_page(file->f_mapping, pos,
zhengbin35f12f02019-03-05 15:44:21 -08003823 pos + write_len - 1))
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05003824 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 Hellwiga969e902008-07-23 21:27:04 -07003831
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 Hifumi6ccfa802008-09-02 14:35:40 -07003836 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07003837 */
Andrey Ryabinin55635ba2017-05-03 14:55:59 -07003838 written = invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003839 pos >> PAGE_SHIFT, end);
Andrey Ryabinin55635ba2017-05-03 14:55:59 -07003840 /*
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 Hellwiga969e902008-07-23 21:27:04 -07003848 }
3849
Al Viro639a93a52017-04-13 14:10:15 -04003850 written = mapping->a_ops->direct_IO(iocb, from);
Christoph Hellwiga969e902008-07-23 21:27:04 -07003851
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 Czerner332391a2017-09-21 08:16:29 -06003859 *
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 Khlebnikov80c1fe92019-11-30 17:49:41 -08003863 * them by removing it completely.
3864 *
Konstantin Khlebnikov9266a142019-11-30 17:49:47 -08003865 * Noticeable example is a blkdev_direct_IO().
3866 *
Konstantin Khlebnikov80c1fe92019-11-30 17:49:41 -08003867 * Skip invalidation for async writes or if mapping has no pages.
Christoph Hellwiga969e902008-07-23 21:27:04 -07003868 */
Konstantin Khlebnikov9266a142019-11-30 17:49:47 -08003869 if (written > 0 && mapping->nrpages &&
3870 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3871 dio_warn_stale_pagecache(file);
Christoph Hellwiga969e902008-07-23 21:27:04 -07003872
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07003874 pos += written;
Al Viro639a93a52017-04-13 14:10:15 -04003875 write_len -= written;
Namhyung Kim01166512010-10-26 14:21:58 -07003876 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3877 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878 mark_inode_dirty(inode);
3879 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05003880 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003881 }
Pavel Begunkovab2125d2021-02-24 12:01:45 -08003882 if (written != -EIOCBQUEUED)
3883 iov_iter_revert(from, write_len - iov_iter_count(from));
Christoph Hellwiga969e902008-07-23 21:27:04 -07003884out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003885 return written;
3886}
3887EXPORT_SYMBOL(generic_file_direct_write);
3888
Matthew Wilcox (Oracle)800ba292022-02-19 23:19:49 -05003889ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
Nick Pigginafddba42007-10-16 01:25:01 -07003890{
Matthew Wilcox (Oracle)800ba292022-02-19 23:19:49 -05003891 struct file *file = iocb->ki_filp;
3892 loff_t pos = iocb->ki_pos;
Nick Pigginafddba42007-10-16 01:25:01 -07003893 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 Piggin674b8922007-10-16 01:25:03 -07003897
Nick Pigginafddba42007-10-16 01:25:01 -07003898 do {
3899 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003900 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 Potapenko1468c6f2022-09-15 17:04:16 +02003903 void *fsdata = NULL;
Nick Pigginafddba42007-10-16 01:25:01 -07003904
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003905 offset = (pos & (PAGE_SIZE - 1));
3906 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07003907 iov_iter_count(i));
3908
3909again:
Linus Torvalds00a3d662015-10-07 08:32:38 +01003910 /*
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 Torvalds00a3d662015-10-07 08:32:38 +01003915 */
Andreas Gruenbacher631f8712021-11-09 12:56:06 +01003916 if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) {
Linus Torvalds00a3d662015-10-07 08:32:38 +01003917 status = -EFAULT;
3918 break;
3919 }
3920
Jan Kara296291c2015-10-22 13:32:21 -07003921 if (fatal_signal_pending(current)) {
3922 status = -EINTR;
3923 break;
3924 }
3925
Matthew Wilcox (Oracle)9d6b0cd2022-02-22 14:31:43 -05003926 status = a_ops->write_begin(file, mapping, pos, bytes,
Nick Pigginafddba42007-10-16 01:25:01 -07003927 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07003928 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07003929 break;
3930
anfei zhou931e80e2010-02-02 13:44:02 -08003931 if (mapping_writably_mapped(mapping))
3932 flush_dcache_page(page);
Linus Torvalds00a3d662015-10-07 08:32:38 +01003933
Al Virof0b65f32021-04-30 10:26:41 -04003934 copied = copy_page_from_iter_atomic(page, offset, bytes, i);
Nick Pigginafddba42007-10-16 01:25:01 -07003935 flush_dcache_page(page);
3936
3937 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3938 page, fsdata);
Al Virof0b65f32021-04-30 10:26:41 -04003939 if (unlikely(status != copied)) {
3940 iov_iter_revert(i, copied - max(status, 0L));
3941 if (unlikely(status < 0))
3942 break;
3943 }
Nick Pigginafddba42007-10-16 01:25:01 -07003944 cond_resched();
3945
Al Virobc1bb412021-05-31 00:32:44 -04003946 if (unlikely(status == 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07003947 /*
Al Virobc1bb412021-05-31 00:32:44 -04003948 * 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 Pigginafddba42007-10-16 01:25:01 -07003952 */
Al Virobc1bb412021-05-31 00:32:44 -04003953 if (copied)
3954 bytes = copied;
Nick Pigginafddba42007-10-16 01:25:01 -07003955 goto again;
3956 }
Al Virof0b65f32021-04-30 10:26:41 -04003957 pos += status;
3958 written += status;
Nick Pigginafddba42007-10-16 01:25:01 -07003959
3960 balance_dirty_pages_ratelimited(mapping);
Nick Pigginafddba42007-10-16 01:25:01 -07003961 } while (iov_iter_count(i));
3962
3963 return written ? written : status;
3964}
Al Viro3b93f912014-02-11 21:34:08 -05003965EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966
Jan Karae4dd9de2009-08-17 18:10:06 +02003967/**
Al Viro81742022014-04-03 03:17:43 -04003968 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02003969 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04003970 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02003971 *
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 Kara96087032021-04-12 15:50:21 +02003977 * It expects i_rwsem to be grabbed unless we work on a block device or similar
Jan Karae4dd9de2009-08-17 18:10:06 +02003978 * 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 Kara96087032021-04-12 15:50:21 +02003982 * avoid syncing under i_rwsem.
Mike Rapoporta862f682019-03-05 15:48:42 -08003983 *
3984 * Return:
3985 * * number of bytes written, even for truncated writes
3986 * * negative error code if no data has been written at all
Jan Karae4dd9de2009-08-17 18:10:06 +02003987 */
Al Viro81742022014-04-03 03:17:43 -04003988ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989{
3990 struct file *file = iocb->ki_filp;
Zhiyuan Dai68d68ff2021-05-04 18:40:12 -07003991 struct address_space *mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003992 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05003993 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003994 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05003995 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003996
Linus Torvalds1da177e2005-04-16 15:20:36 -07003997 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01003998 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02003999 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004000 if (err)
4001 goto out;
4002
Josef Bacikc3b2da32012-03-26 09:59:21 -04004003 err = file_update_time(file);
4004 if (err)
4005 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004006
Al Viro2ba48ce2015-04-09 13:52:01 -04004007 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04004008 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07004009
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07004010 written = generic_file_direct_write(iocb, from);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08004011 /*
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 Viro0b8def92015-04-07 10:22:53 -04004018 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05004020
Matthew Wilcox (Oracle)800ba292022-02-19 23:19:49 -05004021 pos = iocb->ki_pos;
4022 status = generic_perform_write(iocb, from);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07004023 /*
Al Viro3b93f912014-02-11 21:34:08 -05004024 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07004025 * 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 Viro60bb4522014-08-08 12:39:16 -04004030 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05004031 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07004032 goto out;
4033 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07004034 /*
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 Viro3b93f912014-02-11 21:34:08 -05004039 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04004040 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07004041 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04004042 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05004043 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07004044 invalidate_mapping_pages(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03004045 pos >> PAGE_SHIFT,
4046 endbyte >> PAGE_SHIFT);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07004047 } 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)800ba292022-02-19 23:19:49 -05004054 written = generic_perform_write(iocb, from);
Al Viro0b8def92015-04-07 10:22:53 -04004055 if (likely(written > 0))
4056 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07004057 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058out:
4059 current->backing_dev_info = NULL;
4060 return written ? written : err;
4061}
Al Viro81742022014-04-03 03:17:43 -04004062EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063
Jan Karae4dd9de2009-08-17 18:10:06 +02004064/**
Al Viro81742022014-04-03 03:17:43 -04004065 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02004066 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04004067 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02004068 *
Al Viro81742022014-04-03 03:17:43 -04004069 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02004070 * filesystems. It takes care of syncing the file in case of O_SYNC file
Jan Kara96087032021-04-12 15:50:21 +02004071 * and acquires i_rwsem as needed.
Mike Rapoporta862f682019-03-05 15:48:42 -08004072 * 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 Karae4dd9de2009-08-17 18:10:06 +02004076 */
Al Viro81742022014-04-03 03:17:43 -04004077ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004078{
4079 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02004080 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004081 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004082
Al Viro59551022016-01-22 15:40:57 -05004083 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -04004084 ret = generic_write_checks(iocb, from);
4085 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04004086 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -05004087 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004088
Christoph Hellwige2592212016-04-07 08:52:01 -07004089 if (ret > 0)
4090 ret = generic_write_sync(iocb, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004091 return ret;
4092}
Al Viro81742022014-04-03 03:17:43 -04004093EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094
David Howellscf9a2ae2006-08-29 19:05:54 +01004095/**
Matthew Wilcox (Oracle)82c50f82021-07-28 15:14:48 -04004096 * 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 Howellscf9a2ae2006-08-29 19:05:54 +01004099 *
Matthew Wilcox (Oracle)82c50f82021-07-28 15:14:48 -04004100 * The address_space is trying to release any data attached to a folio
4101 * (presumably at folio->private).
David Howellscf9a2ae2006-08-29 19:05:54 +01004102 *
Matthew Wilcox (Oracle)82c50f82021-07-28 15:14:48 -04004103 * 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 Howellscf9a2ae2006-08-29 19:05:54 +01004105 *
Matthew Wilcox (Oracle)82c50f82021-07-28 15:14:48 -04004106 * 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 Howells266cf652009-04-03 16:42:36 +01004109 *
Matthew Wilcox (Oracle)82c50f82021-07-28 15:14:48 -04004110 * Return: %true if the release was successful, otherwise %false.
David Howellscf9a2ae2006-08-29 19:05:54 +01004111 */
Matthew Wilcox (Oracle)82c50f82021-07-28 15:14:48 -04004112bool filemap_release_folio(struct folio *folio, gfp_t gfp)
David Howellscf9a2ae2006-08-29 19:05:54 +01004113{
Matthew Wilcox (Oracle)82c50f82021-07-28 15:14:48 -04004114 struct address_space * const mapping = folio->mapping;
David Howellscf9a2ae2006-08-29 19:05:54 +01004115
Matthew Wilcox (Oracle)82c50f82021-07-28 15:14:48 -04004116 BUG_ON(!folio_test_locked(folio));
4117 if (folio_test_writeback(folio))
4118 return false;
David Howellscf9a2ae2006-08-29 19:05:54 +01004119
Matthew Wilcox (Oracle)fa290002022-04-29 17:00:05 -04004120 if (mapping && mapping->a_ops->release_folio)
4121 return mapping->a_ops->release_folio(folio, gfp);
Matthew Wilcox (Oracle)68189fe2022-05-01 01:08:08 -04004122 return try_to_free_buffers(folio);
David Howellscf9a2ae2006-08-29 19:05:54 +01004123}
Matthew Wilcox (Oracle)82c50f82021-07-28 15:14:48 -04004124EXPORT_SYMBOL(filemap_release_folio);