blob: 84b7813badf1aef4ab7be6b47d42182eece267d4 [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>
24#include <linux/mman.h>
25#include <linux/pagemap.h>
26#include <linux/file.h>
27#include <linux/uio.h>
Josef Bacikcfcbfb12019-05-13 17:21:04 -070028#include <linux/error-injection.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/hash.h>
30#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070031#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/pagevec.h>
33#include <linux/blkdev.h>
34#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>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060038#include <linux/cleancache.h>
Mel Gormanc7df8ad2017-11-15 17:37:41 -080039#include <linux/shmem_fs.h>
Kirill A. Shutemovf1820362014-04-07 15:37:19 -070040#include <linux/rmap.h>
Johannes Weinerb1d29ba2018-10-26 15:06:08 -070041#include <linux/delayacct.h>
Johannes Weinereb414682018-10-26 15:06:27 -070042#include <linux/psi.h>
Ben Dooksd0e6a582019-10-18 20:20:20 -070043#include <linux/ramfs.h>
Yang Shib9306a72020-08-06 23:19:55 -070044#include <linux/page_idle.h>
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +030045#include <asm/pgalloc.h>
Will Deaconde591a82021-02-10 11:15:11 +000046#include <asm/tlbflush.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080047#include "internal.h"
48
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070049#define CREATE_TRACE_POINTS
50#include <trace/events/filemap.h>
51
Linus Torvalds1da177e2005-04-16 15:20:36 -070052/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 * FIXME: remove all knowledge of the buffer layer from the core VM
54 */
Jan Kara148f9482009-08-17 19:52:36 +020055#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <asm/mman.h>
58
59/*
60 * Shared mappings implemented 30.11.1994. It's not fully working yet,
61 * though.
62 *
63 * Shared mappings now work. 15.8.1995 Bruno.
64 *
65 * finished 'unifying' the page and buffer cache and SMP-threaded the
66 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
67 *
68 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
69 */
70
71/*
72 * Lock ordering:
73 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080074 * ->i_mmap_rwsem (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070076 * ->swap_lock (exclusive_swap_page, others)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070077 * ->i_pages lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080079 * ->i_mutex
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080080 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -070082 * ->mmap_lock
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080083 * ->i_mmap_rwsem
Hugh Dickinsb8072f02005-10-29 18:16:41 -070084 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070085 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -070087 * ->mmap_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * ->lock_page (access_process_vm)
89 *
Al Viroccad2362014-02-11 22:36:48 -050090 * ->i_mutex (generic_perform_write)
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -070091 * ->mmap_lock (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060093 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110094 * sb_lock (fs/fs-writeback.c)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070095 * ->i_pages lock (__sync_single_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080097 * ->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 * ->anon_vma.lock (vma_adjust)
99 *
100 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700101 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700103 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -0700104 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 * ->private_lock (try_to_unmap_one)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700106 * ->i_pages lock (try_to_unmap_one)
Hugh Dickins15b44732020-12-15 14:21:31 -0800107 * ->lruvec->lru_lock (follow_page->mark_page_accessed)
108 * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 * ->private_lock (page_remove_rmap->set_page_dirty)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700110 * ->i_pages lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600111 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100112 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700113 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600114 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100115 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
117 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800118 * ->i_mmap_rwsem
Andi Kleen9a3c5312012-03-21 16:34:09 -0700119 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 */
121
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500122static void page_cache_delete(struct address_space *mapping,
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700123 struct page *page, void *shadow)
124{
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500125 XA_STATE(xas, &mapping->i_pages, page->index);
126 unsigned int nr = 1;
Kirill A. Shutemovc70b6472016-12-12 16:43:17 -0800127
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500128 mapping_set_update(&xas, mapping);
129
130 /* hugetlb pages are represented by a single entry in the xarray */
131 if (!PageHuge(page)) {
132 xas_set_order(&xas, page->index, compound_order(page));
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -0700133 nr = compound_nr(page);
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500134 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700135
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700136 VM_BUG_ON_PAGE(!PageLocked(page), page);
137 VM_BUG_ON_PAGE(PageTail(page), page);
138 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
Johannes Weiner449dd692014-04-03 14:47:56 -0700139
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500140 xas_store(&xas, shadow);
141 xas_init_marks(&xas);
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200142
Jan Kara23006382017-11-15 17:37:26 -0800143 page->mapping = NULL;
144 /* Leave page->index set: truncation lookup relies upon it */
145
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200146 if (shadow) {
147 mapping->nrexceptional += nr;
148 /*
149 * Make sure the nrexceptional update is committed before
150 * the nrpages update so that final truncate racing
151 * with reclaim does not see both counters 0 at the
152 * same time and miss a shadow entry.
153 */
154 smp_wmb();
155 }
156 mapping->nrpages -= nr;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700157}
158
Jan Kara5ecc4d82017-11-15 17:37:29 -0800159static void unaccount_page_cache_page(struct address_space *mapping,
160 struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161{
Jan Kara5ecc4d82017-11-15 17:37:29 -0800162 int nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600164 /*
165 * if we're uptodate, flush out into the cleancache, otherwise
166 * invalidate any existing cleancache entries. We can't leave
167 * stale data around in the cleancache once our page is gone
168 */
169 if (PageUptodate(page) && PageMappedToDisk(page))
170 cleancache_put_page(page);
171 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400172 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600173
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700174 VM_BUG_ON_PAGE(PageTail(page), page);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800175 VM_BUG_ON_PAGE(page_mapped(page), page);
176 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
177 int mapcount;
178
179 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
180 current->comm, page_to_pfn(page));
181 dump_page(page, "still mapped when deleted");
182 dump_stack();
183 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
184
185 mapcount = page_mapcount(page);
186 if (mapping_exiting(mapping) &&
187 page_count(page) >= mapcount + 2) {
188 /*
189 * All vmas have already been torn down, so it's
190 * a good bet that actually the page is unmapped,
191 * and we'd prefer not to leak it: if we're wrong,
192 * some other bad page check should catch it later.
193 */
194 page_mapcount_reset(page);
Joonsoo Kim6d061f92016-05-19 17:10:46 -0700195 page_ref_sub(page, mapcount);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800196 }
197 }
198
Jan Kara76253fb2017-11-15 17:37:22 -0800199 /* hugetlb pages do not participate in page cache accounting. */
Jan Kara5ecc4d82017-11-15 17:37:29 -0800200 if (PageHuge(page))
201 return;
Jan Kara76253fb2017-11-15 17:37:22 -0800202
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -0700203 nr = thp_nr_pages(page);
Jan Kara5ecc4d82017-11-15 17:37:29 -0800204
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700205 __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr);
Jan Kara5ecc4d82017-11-15 17:37:29 -0800206 if (PageSwapBacked(page)) {
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700207 __mod_lruvec_page_state(page, NR_SHMEM, -nr);
Jan Kara5ecc4d82017-11-15 17:37:29 -0800208 if (PageTransHuge(page))
Muchun Song57b28472021-02-24 12:03:31 -0800209 __mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr);
Song Liu99cb0db2019-09-23 15:38:00 -0700210 } else if (PageTransHuge(page)) {
Muchun Songbf9ecea2021-02-24 12:03:27 -0800211 __mod_lruvec_page_state(page, NR_FILE_THPS, -nr);
Song Liu09d91cd2019-09-23 15:38:03 -0700212 filemap_nr_thps_dec(mapping);
Jan Kara76253fb2017-11-15 17:37:22 -0800213 }
Jan Kara5ecc4d82017-11-15 17:37:29 -0800214
215 /*
216 * At this point page must be either written or cleaned by
217 * truncate. Dirty page here signals a bug and loss of
218 * unwritten data.
219 *
220 * This fixes dirty accounting after removing the page entirely
221 * but leaves PageDirty set: it has no effect for truncated
222 * page and anyway will be cleared before returning page into
223 * buddy allocator.
224 */
225 if (WARN_ON_ONCE(PageDirty(page)))
226 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
227}
228
229/*
230 * Delete a page from the page cache and free it. Caller has to make
231 * sure the page is locked and that nobody else uses it - or that usage
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700232 * is safe. The caller must hold the i_pages lock.
Jan Kara5ecc4d82017-11-15 17:37:29 -0800233 */
234void __delete_from_page_cache(struct page *page, void *shadow)
235{
236 struct address_space *mapping = page->mapping;
237
238 trace_mm_filemap_delete_from_page_cache(page);
239
240 unaccount_page_cache_page(mapping, page);
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500241 page_cache_delete(mapping, page, shadow);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
243
Jan Kara59c66c52017-11-15 17:37:18 -0800244static void page_cache_free_page(struct address_space *mapping,
245 struct page *page)
246{
247 void (*freepage)(struct page *);
248
249 freepage = mapping->a_ops->freepage;
250 if (freepage)
251 freepage(page);
252
253 if (PageTransHuge(page) && !PageHuge(page)) {
Matthew Wilcox (Oracle)887b22c2020-10-15 20:05:23 -0700254 page_ref_sub(page, thp_nr_pages(page));
Jan Kara59c66c52017-11-15 17:37:18 -0800255 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
256 } else {
257 put_page(page);
258 }
259}
260
Minchan Kim702cfbf2011-03-22 16:32:43 -0700261/**
262 * delete_from_page_cache - delete page from page cache
263 * @page: the page which the kernel is trying to remove from page cache
264 *
265 * This must be called only on pages that have been verified to be in the page
266 * cache and locked. It will never put the page into the free list, the caller
267 * has a reference on the page.
268 */
269void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700271 struct address_space *mapping = page_mapping(page);
Greg Thelenc4843a72015-05-22 17:13:16 -0400272 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
Matt Mackallcd7619d2005-05-01 08:59:01 -0700274 BUG_ON(!PageLocked(page));
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700275 xa_lock_irqsave(&mapping->i_pages, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700276 __delete_from_page_cache(page, NULL);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700277 xa_unlock_irqrestore(&mapping->i_pages, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500278
Jan Kara59c66c52017-11-15 17:37:18 -0800279 page_cache_free_page(mapping, page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700280}
281EXPORT_SYMBOL(delete_from_page_cache);
282
Jan Karaaa65c292017-11-15 17:37:33 -0800283/*
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500284 * page_cache_delete_batch - delete several pages from page cache
Jan Karaaa65c292017-11-15 17:37:33 -0800285 * @mapping: the mapping to which pages belong
286 * @pvec: pagevec with pages to delete
287 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700288 * The function walks over mapping->i_pages and removes pages passed in @pvec
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700289 * from the mapping. The function expects @pvec to be sorted by page index
290 * and is optimised for it to be dense.
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700291 * It tolerates holes in @pvec (mapping entries at those indices are not
Jan Karaaa65c292017-11-15 17:37:33 -0800292 * modified). The function expects only THP head pages to be present in the
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700293 * @pvec.
Jan Karaaa65c292017-11-15 17:37:33 -0800294 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700295 * The function expects the i_pages lock to be held.
Jan Karaaa65c292017-11-15 17:37:33 -0800296 */
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500297static void page_cache_delete_batch(struct address_space *mapping,
Jan Karaaa65c292017-11-15 17:37:33 -0800298 struct pagevec *pvec)
299{
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500300 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
Jan Karaaa65c292017-11-15 17:37:33 -0800301 int total_pages = 0;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700302 int i = 0;
Jan Karaaa65c292017-11-15 17:37:33 -0800303 struct page *page;
Jan Karaaa65c292017-11-15 17:37:33 -0800304
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500305 mapping_set_update(&xas, mapping);
306 xas_for_each(&xas, page, ULONG_MAX) {
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700307 if (i >= pagevec_count(pvec))
Jan Karaaa65c292017-11-15 17:37:33 -0800308 break;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700309
310 /* A swap/dax/shadow entry got inserted? Skip it. */
Matthew Wilcox3159f942017-11-03 13:30:42 -0400311 if (xa_is_value(page))
Jan Karaaa65c292017-11-15 17:37:33 -0800312 continue;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700313 /*
314 * A page got inserted in our range? Skip it. We have our
315 * pages locked so they are protected from being removed.
316 * If we see a page whose index is higher than ours, it
317 * means our page has been removed, which shouldn't be
318 * possible because we're holding the PageLock.
319 */
320 if (page != pvec->pages[i]) {
321 VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
322 page);
323 continue;
Linus Torvalds69bf4b62019-07-05 19:55:18 -0700324 }
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700325
326 WARN_ON_ONCE(!PageLocked(page));
327
328 if (page->index == xas.xa_index)
329 page->mapping = NULL;
330 /* Leave page->index set: truncation lookup relies on it */
331
332 /*
333 * Move to the next page in the vector if this is a regular
334 * page or the index is of the last sub-page of this compound
335 * page.
336 */
337 if (page->index + compound_nr(page) - 1 == xas.xa_index)
338 i++;
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500339 xas_store(&xas, NULL);
Jan Karaaa65c292017-11-15 17:37:33 -0800340 total_pages++;
341 }
342 mapping->nrpages -= total_pages;
343}
344
345void delete_from_page_cache_batch(struct address_space *mapping,
346 struct pagevec *pvec)
347{
348 int i;
349 unsigned long flags;
350
351 if (!pagevec_count(pvec))
352 return;
353
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700354 xa_lock_irqsave(&mapping->i_pages, flags);
Jan Karaaa65c292017-11-15 17:37:33 -0800355 for (i = 0; i < pagevec_count(pvec); i++) {
356 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
357
358 unaccount_page_cache_page(mapping, pvec->pages[i]);
359 }
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500360 page_cache_delete_batch(mapping, pvec);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700361 xa_unlock_irqrestore(&mapping->i_pages, flags);
Jan Karaaa65c292017-11-15 17:37:33 -0800362
363 for (i = 0; i < pagevec_count(pvec); i++)
364 page_cache_free_page(mapping, pvec->pages[i]);
365}
366
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200367int filemap_check_errors(struct address_space *mapping)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700368{
369 int ret = 0;
370 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700371 if (test_bit(AS_ENOSPC, &mapping->flags) &&
372 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700373 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700374 if (test_bit(AS_EIO, &mapping->flags) &&
375 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700376 ret = -EIO;
377 return ret;
378}
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200379EXPORT_SYMBOL(filemap_check_errors);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700380
Jeff Layton76341ca2017-07-06 07:02:22 -0400381static int filemap_check_and_keep_errors(struct address_space *mapping)
382{
383 /* Check for outstanding write errors */
384 if (test_bit(AS_EIO, &mapping->flags))
385 return -EIO;
386 if (test_bit(AS_ENOSPC, &mapping->flags))
387 return -ENOSPC;
388 return 0;
389}
390
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700392 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700393 * @mapping: address space structure to write
394 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800395 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700396 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700398 * Start writeback against all of a mapping's dirty pages that lie
399 * within the byte offsets <start, end> inclusive.
400 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700402 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 * these two operations is that if a dirty page/buffer is encountered, it must
404 * be waited upon, and not just skipped over.
Mike Rapoporta862f682019-03-05 15:48:42 -0800405 *
406 * Return: %0 on success, negative error code otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800408int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
409 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410{
411 int ret;
412 struct writeback_control wbc = {
413 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800414 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700415 .range_start = start,
416 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 };
418
Christoph Hellwigf56753a2020-09-24 08:51:40 +0200419 if (!mapping_can_writeback(mapping) ||
Konstantin Khlebnikovc3aab9a02019-09-23 15:34:45 -0700420 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 return 0;
422
Tejun Heob16b1de2015-06-02 08:39:48 -0600423 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 ret = do_writepages(mapping, &wbc);
Tejun Heob16b1de2015-06-02 08:39:48 -0600425 wbc_detach_inode(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 return ret;
427}
428
429static inline int __filemap_fdatawrite(struct address_space *mapping,
430 int sync_mode)
431{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700432 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434
435int filemap_fdatawrite(struct address_space *mapping)
436{
437 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
438}
439EXPORT_SYMBOL(filemap_fdatawrite);
440
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400441int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800442 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443{
444 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
445}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400446EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447
Randy Dunlap485bb992006-06-23 02:03:49 -0700448/**
449 * filemap_flush - mostly a non-blocking flush
450 * @mapping: target address_space
451 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 * This is a mostly non-blocking flush. Not suitable for data-integrity
453 * purposes - I/O may not be started against all dirty pages.
Mike Rapoporta862f682019-03-05 15:48:42 -0800454 *
455 * Return: %0 on success, negative error code otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 */
457int filemap_flush(struct address_space *mapping)
458{
459 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
460}
461EXPORT_SYMBOL(filemap_flush);
462
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500463/**
464 * filemap_range_has_page - check if a page exists in range.
465 * @mapping: address space within which to check
466 * @start_byte: offset in bytes where the range starts
467 * @end_byte: offset in bytes where the range ends (inclusive)
468 *
469 * Find at least one page in the range supplied, usually used to check if
470 * direct writing in this range will trigger a writeback.
Mike Rapoporta862f682019-03-05 15:48:42 -0800471 *
472 * Return: %true if at least one page exists in the specified range,
473 * %false otherwise.
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500474 */
475bool filemap_range_has_page(struct address_space *mapping,
476 loff_t start_byte, loff_t end_byte)
477{
Jan Karaf7b68042017-09-06 16:21:40 -0700478 struct page *page;
Matthew Wilcox8fa8e532018-01-16 06:26:49 -0500479 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
480 pgoff_t max = end_byte >> PAGE_SHIFT;
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500481
482 if (end_byte < start_byte)
483 return false;
484
Matthew Wilcox8fa8e532018-01-16 06:26:49 -0500485 rcu_read_lock();
486 for (;;) {
487 page = xas_find(&xas, max);
488 if (xas_retry(&xas, page))
489 continue;
490 /* Shadow entries don't count */
491 if (xa_is_value(page))
492 continue;
493 /*
494 * We don't need to try to pin this page; we're about to
495 * release the RCU lock anyway. It is enough to know that
496 * there was a page here recently.
497 */
498 break;
499 }
500 rcu_read_unlock();
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500501
Matthew Wilcox8fa8e532018-01-16 06:26:49 -0500502 return page != NULL;
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500503}
504EXPORT_SYMBOL(filemap_range_has_page);
505
Jeff Layton5e8fcc12017-07-06 07:02:24 -0400506static void __filemap_fdatawait_range(struct address_space *mapping,
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800507 loff_t start_byte, loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300509 pgoff_t index = start_byte >> PAGE_SHIFT;
510 pgoff_t end = end_byte >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 struct pagevec pvec;
512 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200514 if (end_byte < start_byte)
Jeff Layton5e8fcc12017-07-06 07:02:24 -0400515 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516
Mel Gorman86679822017-11-15 17:37:52 -0800517 pagevec_init(&pvec);
Jan Kara312e9d22017-11-15 17:35:05 -0800518 while (index <= end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 unsigned i;
520
Jan Kara312e9d22017-11-15 17:35:05 -0800521 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
Jan Kara67fd7072017-11-15 17:35:19 -0800522 end, PAGECACHE_TAG_WRITEBACK);
Jan Kara312e9d22017-11-15 17:35:05 -0800523 if (!nr_pages)
524 break;
525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 for (i = 0; i < nr_pages; i++) {
527 struct page *page = pvec.pages[i];
528
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 wait_on_page_writeback(page);
Jeff Layton5e8fcc12017-07-06 07:02:24 -0400530 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 }
532 pagevec_release(&pvec);
533 cond_resched();
534 }
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800535}
536
537/**
538 * filemap_fdatawait_range - wait for writeback to complete
539 * @mapping: address space structure to wait for
540 * @start_byte: offset in bytes where the range starts
541 * @end_byte: offset in bytes where the range ends (inclusive)
542 *
543 * Walk the list of under-writeback pages of the given address space
544 * in the given range and wait for all of them. Check error status of
545 * the address space and return it.
546 *
547 * Since the error status of the address space is cleared by this function,
548 * callers are responsible for checking the return value and handling and/or
549 * reporting the error.
Mike Rapoporta862f682019-03-05 15:48:42 -0800550 *
551 * Return: error status of the address space.
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800552 */
553int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
554 loff_t end_byte)
555{
Jeff Layton5e8fcc12017-07-06 07:02:24 -0400556 __filemap_fdatawait_range(mapping, start_byte, end_byte);
557 return filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200559EXPORT_SYMBOL(filemap_fdatawait_range);
560
561/**
Ross Zwisleraa0bfcd2019-06-20 17:05:37 -0400562 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
563 * @mapping: address space structure to wait for
564 * @start_byte: offset in bytes where the range starts
565 * @end_byte: offset in bytes where the range ends (inclusive)
566 *
567 * Walk the list of under-writeback pages of the given address space in the
568 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
569 * this function does not clear error status of the address space.
570 *
571 * Use this function if callers don't handle errors themselves. Expected
572 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
573 * fsfreeze(8)
574 */
575int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
576 loff_t start_byte, loff_t end_byte)
577{
578 __filemap_fdatawait_range(mapping, start_byte, end_byte);
579 return filemap_check_and_keep_errors(mapping);
580}
581EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
582
583/**
Jeff Laytona823e452017-07-28 07:24:43 -0400584 * file_fdatawait_range - wait for writeback to complete
585 * @file: file pointing to address space structure to wait for
586 * @start_byte: offset in bytes where the range starts
587 * @end_byte: offset in bytes where the range ends (inclusive)
588 *
589 * Walk the list of under-writeback pages of the address space that file
590 * refers to, in the given range and wait for all of them. Check error
591 * status of the address space vs. the file->f_wb_err cursor and return it.
592 *
593 * Since the error status of the file is advanced by this function,
594 * callers are responsible for checking the return value and handling and/or
595 * reporting the error.
Mike Rapoporta862f682019-03-05 15:48:42 -0800596 *
597 * Return: error status of the address space vs. the file->f_wb_err cursor.
Jeff Laytona823e452017-07-28 07:24:43 -0400598 */
599int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
600{
601 struct address_space *mapping = file->f_mapping;
602
603 __filemap_fdatawait_range(mapping, start_byte, end_byte);
604 return file_check_and_advance_wb_err(file);
605}
606EXPORT_SYMBOL(file_fdatawait_range);
607
608/**
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800609 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
610 * @mapping: address space structure to wait for
611 *
612 * Walk the list of under-writeback pages of the given address space
613 * and wait for all of them. Unlike filemap_fdatawait(), this function
614 * does not clear error status of the address space.
615 *
616 * Use this function if callers don't handle errors themselves. Expected
617 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
618 * fsfreeze(8)
Mike Rapoporta862f682019-03-05 15:48:42 -0800619 *
620 * Return: error status of the address space.
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800621 */
Jeff Layton76341ca2017-07-06 07:02:22 -0400622int filemap_fdatawait_keep_errors(struct address_space *mapping)
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800623{
Jeff Laytonffb959b2017-07-31 10:29:38 -0400624 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
Jeff Layton76341ca2017-07-06 07:02:22 -0400625 return filemap_check_and_keep_errors(mapping);
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800626}
Jeff Layton76341ca2017-07-06 07:02:22 -0400627EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800628
Konstantin Khlebnikov875d91b2019-09-23 15:34:48 -0700629/* Returns true if writeback might be needed or already in progress. */
Jeff Layton9326c9b2017-07-26 10:21:11 -0400630static bool mapping_needs_writeback(struct address_space *mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631{
Konstantin Khlebnikov875d91b2019-09-23 15:34:48 -0700632 if (dax_mapping(mapping))
633 return mapping->nrexceptional;
634
635 return mapping->nrpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637
Randy Dunlap485bb992006-06-23 02:03:49 -0700638/**
639 * filemap_write_and_wait_range - write out & wait on a file range
640 * @mapping: the address_space for the pages
641 * @lstart: offset in bytes where the range starts
642 * @lend: offset in bytes where the range ends (inclusive)
643 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800644 * Write out and wait upon file offsets lstart->lend, inclusive.
645 *
mchehab@s-opensource.com0e056eb2017-03-30 17:11:36 -0300646 * Note that @lend is inclusive (describes the last byte to be written) so
Andrew Morton469eb4d2006-03-24 03:17:45 -0800647 * that this function can be used to write to the very end-of-file (end = -1).
Mike Rapoporta862f682019-03-05 15:48:42 -0800648 *
649 * Return: error status of the address space.
Andrew Morton469eb4d2006-03-24 03:17:45 -0800650 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651int filemap_write_and_wait_range(struct address_space *mapping,
652 loff_t lstart, loff_t lend)
653{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800654 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Jeff Layton9326c9b2017-07-26 10:21:11 -0400656 if (mapping_needs_writeback(mapping)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800657 err = __filemap_fdatawrite_range(mapping, lstart, lend,
658 WB_SYNC_ALL);
Ira Weinyddf8f372020-01-30 22:12:07 -0800659 /*
660 * Even if the above returned error, the pages may be
661 * written partially (e.g. -ENOSPC), so we wait for it.
662 * But the -EIO is special case, it may indicate the worst
663 * thing (e.g. bug) happened, so we avoid waiting for it.
664 */
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800665 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200666 int err2 = filemap_fdatawait_range(mapping,
667 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800668 if (!err)
669 err = err2;
Jeff Laytoncbeaf952017-07-06 07:02:23 -0400670 } else {
671 /* Clear any previously stored errors */
672 filemap_check_errors(mapping);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800673 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700674 } else {
675 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800677 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678}
Chris Masonf6995582009-04-15 13:22:37 -0400679EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Jeff Layton5660e132017-07-06 07:02:25 -0400681void __filemap_set_wb_err(struct address_space *mapping, int err)
682{
Jeff Layton3acdfd22017-07-24 06:22:15 -0400683 errseq_t eseq = errseq_set(&mapping->wb_err, err);
Jeff Layton5660e132017-07-06 07:02:25 -0400684
685 trace_filemap_set_wb_err(mapping, eseq);
686}
687EXPORT_SYMBOL(__filemap_set_wb_err);
688
689/**
690 * file_check_and_advance_wb_err - report wb error (if any) that was previously
691 * and advance wb_err to current one
692 * @file: struct file on which the error is being reported
693 *
694 * When userland calls fsync (or something like nfsd does the equivalent), we
695 * want to report any writeback errors that occurred since the last fsync (or
696 * since the file was opened if there haven't been any).
697 *
698 * Grab the wb_err from the mapping. If it matches what we have in the file,
699 * then just quickly return 0. The file is all caught up.
700 *
701 * If it doesn't match, then take the mapping value, set the "seen" flag in
702 * it and try to swap it into place. If it works, or another task beat us
703 * to it with the new value, then update the f_wb_err and return the error
704 * portion. The error at this point must be reported via proper channels
705 * (a'la fsync, or NFS COMMIT operation, etc.).
706 *
707 * While we handle mapping->wb_err with atomic operations, the f_wb_err
708 * value is protected by the f_lock since we must ensure that it reflects
709 * the latest value swapped in for this file descriptor.
Mike Rapoporta862f682019-03-05 15:48:42 -0800710 *
711 * Return: %0 on success, negative error code otherwise.
Jeff Layton5660e132017-07-06 07:02:25 -0400712 */
713int file_check_and_advance_wb_err(struct file *file)
714{
715 int err = 0;
716 errseq_t old = READ_ONCE(file->f_wb_err);
717 struct address_space *mapping = file->f_mapping;
718
719 /* Locklessly handle the common case where nothing has changed */
720 if (errseq_check(&mapping->wb_err, old)) {
721 /* Something changed, must use slow path */
722 spin_lock(&file->f_lock);
723 old = file->f_wb_err;
724 err = errseq_check_and_advance(&mapping->wb_err,
725 &file->f_wb_err);
726 trace_file_check_and_advance_wb_err(file, old);
727 spin_unlock(&file->f_lock);
728 }
Jeff Laytonf4e222c2017-10-03 16:15:25 -0700729
730 /*
731 * We're mostly using this function as a drop in replacement for
732 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
733 * that the legacy code would have had on these flags.
734 */
735 clear_bit(AS_EIO, &mapping->flags);
736 clear_bit(AS_ENOSPC, &mapping->flags);
Jeff Layton5660e132017-07-06 07:02:25 -0400737 return err;
738}
739EXPORT_SYMBOL(file_check_and_advance_wb_err);
740
741/**
742 * file_write_and_wait_range - write out & wait on a file range
743 * @file: file pointing to address_space with pages
744 * @lstart: offset in bytes where the range starts
745 * @lend: offset in bytes where the range ends (inclusive)
746 *
747 * Write out and wait upon file offsets lstart->lend, inclusive.
748 *
749 * Note that @lend is inclusive (describes the last byte to be written) so
750 * that this function can be used to write to the very end-of-file (end = -1).
751 *
752 * After writing out and waiting on the data, we check and advance the
753 * f_wb_err cursor to the latest value, and return any errors detected there.
Mike Rapoporta862f682019-03-05 15:48:42 -0800754 *
755 * Return: %0 on success, negative error code otherwise.
Jeff Layton5660e132017-07-06 07:02:25 -0400756 */
757int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
758{
759 int err = 0, err2;
760 struct address_space *mapping = file->f_mapping;
761
Jeff Layton9326c9b2017-07-26 10:21:11 -0400762 if (mapping_needs_writeback(mapping)) {
Jeff Layton5660e132017-07-06 07:02:25 -0400763 err = __filemap_fdatawrite_range(mapping, lstart, lend,
764 WB_SYNC_ALL);
765 /* See comment of filemap_write_and_wait() */
766 if (err != -EIO)
767 __filemap_fdatawait_range(mapping, lstart, lend);
768 }
769 err2 = file_check_and_advance_wb_err(file);
770 if (!err)
771 err = err2;
772 return err;
773}
774EXPORT_SYMBOL(file_write_and_wait_range);
775
Randy Dunlap485bb992006-06-23 02:03:49 -0700776/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700777 * replace_page_cache_page - replace a pagecache page with a new one
778 * @old: page to be replaced
779 * @new: page to replace with
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700780 *
781 * This function replaces a page in the pagecache with a new one. On
782 * success it acquires the pagecache reference for the new page and
783 * drops it for the old page. Both the old and new pages must be
784 * locked. This function does not add the new page to the LRU, the
785 * caller must do that.
786 *
Matthew Wilcox74d60952017-11-17 10:01:45 -0500787 * The remove + add is atomic. This function cannot fail.
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700788 */
Baolin Wang1f7ef652021-02-24 12:01:42 -0800789void replace_page_cache_page(struct page *old, struct page *new)
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700790{
Matthew Wilcox74d60952017-11-17 10:01:45 -0500791 struct address_space *mapping = old->mapping;
792 void (*freepage)(struct page *) = mapping->a_ops->freepage;
793 pgoff_t offset = old->index;
794 XA_STATE(xas, &mapping->i_pages, offset);
795 unsigned long flags;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700796
Sasha Levin309381fea2014-01-23 15:52:54 -0800797 VM_BUG_ON_PAGE(!PageLocked(old), old);
798 VM_BUG_ON_PAGE(!PageLocked(new), new);
799 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700800
Matthew Wilcox74d60952017-11-17 10:01:45 -0500801 get_page(new);
802 new->mapping = mapping;
803 new->index = offset;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700804
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700805 mem_cgroup_migrate(old, new);
806
Matthew Wilcox74d60952017-11-17 10:01:45 -0500807 xas_lock_irqsave(&xas, flags);
808 xas_store(&xas, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700809
Matthew Wilcox74d60952017-11-17 10:01:45 -0500810 old->mapping = NULL;
811 /* hugetlb pages do not participate in page cache accounting. */
812 if (!PageHuge(old))
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700813 __dec_lruvec_page_state(old, NR_FILE_PAGES);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500814 if (!PageHuge(new))
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700815 __inc_lruvec_page_state(new, NR_FILE_PAGES);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500816 if (PageSwapBacked(old))
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700817 __dec_lruvec_page_state(old, NR_SHMEM);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500818 if (PageSwapBacked(new))
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700819 __inc_lruvec_page_state(new, NR_SHMEM);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500820 xas_unlock_irqrestore(&xas, flags);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500821 if (freepage)
822 freepage(old);
823 put_page(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700824}
825EXPORT_SYMBOL_GPL(replace_page_cache_page);
826
Andrew Morton16c0cc02020-12-11 13:36:27 -0800827noinline int __add_to_page_cache_locked(struct page *page,
Alexei Starovoitov76cd6172020-08-27 15:01:10 -0700828 struct address_space *mapping,
Linus Torvaldsc4cf4982020-10-16 11:31:55 -0700829 pgoff_t offset, gfp_t gfp,
Alexei Starovoitov76cd6172020-08-27 15:01:10 -0700830 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831{
Matthew Wilcox74d60952017-11-17 10:01:45 -0500832 XA_STATE(xas, &mapping->i_pages, offset);
Johannes Weiner00501b52014-08-08 14:19:20 -0700833 int huge = PageHuge(page);
Nick Piggine2867812008-07-25 19:45:30 -0700834 int error;
Waiman Longda742402021-02-04 18:32:45 -0800835 bool charged = false;
Nick Piggine2867812008-07-25 19:45:30 -0700836
Sasha Levin309381fea2014-01-23 15:52:54 -0800837 VM_BUG_ON_PAGE(!PageLocked(page), page);
838 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500839 mapping_set_update(&xas, mapping);
Nick Piggine2867812008-07-25 19:45:30 -0700840
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300841 get_page(page);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700842 page->mapping = mapping;
843 page->index = offset;
844
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700845 if (!huge) {
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700846 error = mem_cgroup_charge(page, current->mm, gfp);
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700847 if (error)
848 goto error;
Waiman Longda742402021-02-04 18:32:45 -0800849 charged = true;
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700850 }
851
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700852 gfp &= GFP_RECLAIM_MASK;
853
Matthew Wilcox74d60952017-11-17 10:01:45 -0500854 do {
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700855 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
856 void *entry, *old = NULL;
857
858 if (order > thp_order(page))
859 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
860 order, gfp);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500861 xas_lock_irq(&xas);
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700862 xas_for_each_conflict(&xas, entry) {
863 old = entry;
864 if (!xa_is_value(entry)) {
865 xas_set_err(&xas, -EEXIST);
866 goto unlock;
867 }
868 }
869
870 if (old) {
871 if (shadowp)
872 *shadowp = old;
873 /* entry may have been split before we acquired lock */
874 order = xa_get_order(xas.xa, xas.xa_index);
875 if (order > thp_order(page)) {
876 xas_split(&xas, old, order);
877 xas_reset(&xas);
878 }
879 }
880
Matthew Wilcox74d60952017-11-17 10:01:45 -0500881 xas_store(&xas, page);
882 if (xas_error(&xas))
883 goto unlock;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700884
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700885 if (old)
Matthew Wilcox74d60952017-11-17 10:01:45 -0500886 mapping->nrexceptional--;
Matthew Wilcox74d60952017-11-17 10:01:45 -0500887 mapping->nrpages++;
888
889 /* hugetlb pages do not participate in page cache accounting */
890 if (!huge)
Johannes Weiner0d1c2072020-06-03 16:01:54 -0700891 __inc_lruvec_page_state(page, NR_FILE_PAGES);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500892unlock:
893 xas_unlock_irq(&xas);
Matthew Wilcox (Oracle)198b62f2020-10-15 20:05:20 -0700894 } while (xas_nomem(&xas, gfp));
Matthew Wilcox74d60952017-11-17 10:01:45 -0500895
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700896 if (xas_error(&xas)) {
897 error = xas_error(&xas);
Waiman Longda742402021-02-04 18:32:45 -0800898 if (charged)
899 mem_cgroup_uncharge(page);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500900 goto error;
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700901 }
Matthew Wilcox74d60952017-11-17 10:01:45 -0500902
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700903 trace_mm_filemap_add_to_page_cache(page);
904 return 0;
Matthew Wilcox74d60952017-11-17 10:01:45 -0500905error:
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700906 page->mapping = NULL;
907 /* Leave page->index set: truncation relies upon it */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300908 put_page(page);
Johannes Weiner3fea5a42020-06-03 16:01:41 -0700909 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910}
Josef Bacikcfcbfb12019-05-13 17:21:04 -0700911ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
Johannes Weinera5289102014-04-03 14:47:51 -0700912
913/**
914 * add_to_page_cache_locked - add a locked page to the pagecache
915 * @page: page to add
916 * @mapping: the page's address_space
917 * @offset: page index
918 * @gfp_mask: page allocation mode
919 *
920 * This function is used to add a page to the pagecache. It must be locked.
921 * This function does not add the page to the LRU. The caller must do that.
Mike Rapoporta862f682019-03-05 15:48:42 -0800922 *
923 * Return: %0 on success, negative error code otherwise.
Johannes Weinera5289102014-04-03 14:47:51 -0700924 */
925int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
926 pgoff_t offset, gfp_t gfp_mask)
927{
928 return __add_to_page_cache_locked(page, mapping, offset,
929 gfp_mask, NULL);
930}
Nick Piggine2867812008-07-25 19:45:30 -0700931EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
933int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400934 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935{
Johannes Weinera5289102014-04-03 14:47:51 -0700936 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700937 int ret;
938
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800939 __SetPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700940 ret = __add_to_page_cache_locked(page, mapping, offset,
941 gfp_mask, &shadow);
942 if (unlikely(ret))
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800943 __ClearPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700944 else {
945 /*
946 * The page might have been evicted from cache only
947 * recently, in which case it should be activated like
948 * any other repeatedly accessed page.
Rik van Rielf0281a02016-05-20 16:56:25 -0700949 * The exception is pages getting rewritten; evicting other
950 * data from the working set, only to cache data that will
951 * get overwritten with something else, is a waste of memory.
Johannes Weinera5289102014-04-03 14:47:51 -0700952 */
Johannes Weiner1899ad12018-10-26 15:06:04 -0700953 WARN_ON_ONCE(PageActive(page));
954 if (!(gfp_mask & __GFP_WRITE) && shadow)
955 workingset_refault(page, shadow);
Johannes Weinera5289102014-04-03 14:47:51 -0700956 lru_cache_add(page);
957 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 return ret;
959}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300960EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961
Paul Jackson44110fe2006-03-24 03:16:04 -0800962#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700963struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800964{
Miao Xiec0ff7452010-05-24 14:32:08 -0700965 int n;
966 struct page *page;
967
Paul Jackson44110fe2006-03-24 03:16:04 -0800968 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700969 unsigned int cpuset_mems_cookie;
970 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700971 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700972 n = cpuset_mem_spread_node();
Vlastimil Babka96db8002015-09-08 15:03:50 -0700973 page = __alloc_pages_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700974 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700975
Miao Xiec0ff7452010-05-24 14:32:08 -0700976 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800977 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700978 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800979}
Nick Piggin2ae88142006-10-28 10:38:23 -0700980EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800981#endif
982
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983/*
984 * In order to wait for pages to become available there must be
985 * waitqueues associated with pages. By using a hash table of
986 * waitqueues where the bucket discipline is to maintain all
987 * waiters on the same queue and wake all when any of the pages
988 * become available, and for the woken contexts to check to be
989 * sure the appropriate page became available, this saves space
990 * at a cost of "thundering herd" phenomena during rare hash
991 * collisions.
992 */
Nicholas Piggin62906022016-12-25 13:00:30 +1000993#define PAGE_WAIT_TABLE_BITS 8
994#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
995static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
996
997static wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998{
Nicholas Piggin62906022016-12-25 13:00:30 +1000999 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000}
Nicholas Piggin62906022016-12-25 13:00:30 +10001001
1002void __init pagecache_init(void)
1003{
1004 int i;
1005
1006 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1007 init_waitqueue_head(&page_wait_table[i]);
1008
1009 page_writeback_init();
1010}
1011
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001012/*
1013 * The page wait code treats the "wait->flags" somewhat unusually, because
Linus Torvalds5868ec22020-09-20 10:38:47 -07001014 * we have multiple different kinds of waits, not just the usual "exclusive"
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001015 * one.
1016 *
1017 * We have:
1018 *
1019 * (a) no special bits set:
1020 *
1021 * We're just waiting for the bit to be released, and when a waker
1022 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1023 * and remove it from the wait queue.
1024 *
1025 * Simple and straightforward.
1026 *
1027 * (b) WQ_FLAG_EXCLUSIVE:
1028 *
1029 * The waiter is waiting to get the lock, and only one waiter should
1030 * be woken up to avoid any thundering herd behavior. We'll set the
1031 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1032 *
1033 * This is the traditional exclusive wait.
1034 *
Linus Torvalds5868ec22020-09-20 10:38:47 -07001035 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001036 *
1037 * The waiter is waiting to get the bit, and additionally wants the
1038 * lock to be transferred to it for fair lock behavior. If the lock
1039 * cannot be taken, we stop walking the wait queue without waking
1040 * the waiter.
1041 *
1042 * This is the "fair lock handoff" case, and in addition to setting
1043 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1044 * that it now has the lock.
1045 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02001046static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
Nicholas Piggin62906022016-12-25 13:00:30 +10001047{
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001048 unsigned int flags;
Nicholas Piggin62906022016-12-25 13:00:30 +10001049 struct wait_page_key *key = arg;
1050 struct wait_page_queue *wait_page
1051 = container_of(wait, struct wait_page_queue, wait);
1052
Linus Torvaldscdc8fcb2020-08-03 13:01:22 -07001053 if (!wake_page_match(wait_page, key))
Nicholas Piggin62906022016-12-25 13:00:30 +10001054 return 0;
Linus Torvalds3510ca22017-08-27 13:55:12 -07001055
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001056 /*
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001057 * If it's a lock handoff wait, we get the bit for it, and
1058 * stop walking (and do not wake it up) if we can't.
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001059 */
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001060 flags = wait->flags;
1061 if (flags & WQ_FLAG_EXCLUSIVE) {
1062 if (test_bit(key->bit_nr, &key->page->flags))
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001063 return -1;
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001064 if (flags & WQ_FLAG_CUSTOM) {
1065 if (test_and_set_bit(key->bit_nr, &key->page->flags))
1066 return -1;
1067 flags |= WQ_FLAG_DONE;
1068 }
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001069 }
Nicholas Piggin62906022016-12-25 13:00:30 +10001070
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001071 /*
1072 * We are holding the wait-queue lock, but the waiter that
1073 * is waiting for this will be checking the flags without
1074 * any locking.
1075 *
1076 * So update the flags atomically, and wake up the waiter
1077 * afterwards to avoid any races. This store-release pairs
1078 * with the load-acquire in wait_on_page_bit_common().
1079 */
1080 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001081 wake_up_state(wait->private, mode);
1082
1083 /*
1084 * Ok, we have successfully done what we're waiting for,
1085 * and we can unconditionally remove the wait entry.
1086 *
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001087 * Note that this pairs with the "finish_wait()" in the
1088 * waiter, and has to be the absolute last thing we do.
1089 * After this list_del_init(&wait->entry) the wait entry
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001090 * might be de-allocated and the process might even have
1091 * exited.
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001092 */
Linus Torvaldsc6fe44d2020-07-23 12:33:41 -07001093 list_del_init_careful(&wait->entry);
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001094 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
Nicholas Piggin62906022016-12-25 13:00:30 +10001095}
1096
Nicholas Piggin74d81bf2017-02-22 15:44:41 -08001097static void wake_up_page_bit(struct page *page, int bit_nr)
Nicholas Piggin62906022016-12-25 13:00:30 +10001098{
1099 wait_queue_head_t *q = page_waitqueue(page);
1100 struct wait_page_key key;
1101 unsigned long flags;
Tim Chen11a19c72017-08-25 09:13:55 -07001102 wait_queue_entry_t bookmark;
Nicholas Piggin62906022016-12-25 13:00:30 +10001103
1104 key.page = page;
1105 key.bit_nr = bit_nr;
1106 key.page_match = 0;
1107
Tim Chen11a19c72017-08-25 09:13:55 -07001108 bookmark.flags = 0;
1109 bookmark.private = NULL;
1110 bookmark.func = NULL;
1111 INIT_LIST_HEAD(&bookmark.entry);
1112
Nicholas Piggin62906022016-12-25 13:00:30 +10001113 spin_lock_irqsave(&q->lock, flags);
Tim Chen11a19c72017-08-25 09:13:55 -07001114 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1115
1116 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1117 /*
1118 * Take a breather from holding the lock,
1119 * allow pages that finish wake up asynchronously
1120 * to acquire the lock and remove themselves
1121 * from wait queue
1122 */
1123 spin_unlock_irqrestore(&q->lock, flags);
1124 cpu_relax();
1125 spin_lock_irqsave(&q->lock, flags);
1126 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1127 }
1128
Nicholas Piggin62906022016-12-25 13:00:30 +10001129 /*
1130 * It is possible for other pages to have collided on the waitqueue
1131 * hash, so in that case check for a page match. That prevents a long-
1132 * term waiter
1133 *
1134 * It is still possible to miss a case here, when we woke page waiters
1135 * and removed them from the waitqueue, but there are still other
1136 * page waiters.
1137 */
1138 if (!waitqueue_active(q) || !key.page_match) {
1139 ClearPageWaiters(page);
1140 /*
1141 * It's possible to miss clearing Waiters here, when we woke
1142 * our page waiters, but the hashed waitqueue has waiters for
1143 * other pages on it.
1144 *
1145 * That's okay, it's a rare case. The next waker will clear it.
1146 */
1147 }
1148 spin_unlock_irqrestore(&q->lock, flags);
1149}
Nicholas Piggin74d81bf2017-02-22 15:44:41 -08001150
1151static void wake_up_page(struct page *page, int bit)
1152{
1153 if (!PageWaiters(page))
1154 return;
1155 wake_up_page_bit(page, bit);
1156}
Nicholas Piggin62906022016-12-25 13:00:30 +10001157
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001158/*
1159 * A choice of three behaviors for wait_on_page_bit_common():
1160 */
1161enum behavior {
1162 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1163 * __lock_page() waiting on then setting PG_locked.
1164 */
1165 SHARED, /* Hold ref to page and check the bit when woken, like
1166 * wait_on_page_writeback() waiting on PG_writeback.
1167 */
1168 DROP, /* Drop ref to page before wait, no check when woken,
1169 * like put_and_wait_on_page_locked() on PG_locked.
1170 */
1171};
1172
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001173/*
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001174 * Attempt to check (or get) the page bit, and mark us done
1175 * if successful.
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001176 */
1177static inline bool trylock_page_bit_common(struct page *page, int bit_nr,
1178 struct wait_queue_entry *wait)
1179{
1180 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1181 if (test_and_set_bit(bit_nr, &page->flags))
1182 return false;
1183 } else if (test_bit(bit_nr, &page->flags))
1184 return false;
1185
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001186 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001187 return true;
1188}
1189
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001190/* How many times do we accept lock stealing from under a waiter? */
1191int sysctl_page_lock_unfairness = 5;
1192
Nicholas Piggin62906022016-12-25 13:00:30 +10001193static inline int wait_on_page_bit_common(wait_queue_head_t *q,
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001194 struct page *page, int bit_nr, int state, enum behavior behavior)
Nicholas Piggin62906022016-12-25 13:00:30 +10001195{
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001196 int unfairness = sysctl_page_lock_unfairness;
Nicholas Piggin62906022016-12-25 13:00:30 +10001197 struct wait_page_queue wait_page;
Ingo Molnarac6424b2017-06-20 12:06:13 +02001198 wait_queue_entry_t *wait = &wait_page.wait;
Johannes Weinerb1d29ba2018-10-26 15:06:08 -07001199 bool thrashing = false;
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001200 bool delayacct = false;
Johannes Weinereb414682018-10-26 15:06:27 -07001201 unsigned long pflags;
Nicholas Piggin62906022016-12-25 13:00:30 +10001202
Johannes Weinereb414682018-10-26 15:06:27 -07001203 if (bit_nr == PG_locked &&
Johannes Weinerb1d29ba2018-10-26 15:06:08 -07001204 !PageUptodate(page) && PageWorkingset(page)) {
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001205 if (!PageSwapBacked(page)) {
Johannes Weinereb414682018-10-26 15:06:27 -07001206 delayacct_thrashing_start();
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001207 delayacct = true;
1208 }
Johannes Weinereb414682018-10-26 15:06:27 -07001209 psi_memstall_enter(&pflags);
Johannes Weinerb1d29ba2018-10-26 15:06:08 -07001210 thrashing = true;
1211 }
1212
Nicholas Piggin62906022016-12-25 13:00:30 +10001213 init_wait(wait);
1214 wait->func = wake_page_function;
1215 wait_page.page = page;
1216 wait_page.bit_nr = bit_nr;
1217
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001218repeat:
1219 wait->flags = 0;
1220 if (behavior == EXCLUSIVE) {
1221 wait->flags = WQ_FLAG_EXCLUSIVE;
1222 if (--unfairness < 0)
1223 wait->flags |= WQ_FLAG_CUSTOM;
1224 }
1225
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001226 /*
1227 * Do one last check whether we can get the
1228 * page bit synchronously.
1229 *
1230 * Do the SetPageWaiters() marking before that
1231 * to let any waker we _just_ missed know they
1232 * need to wake us up (otherwise they'll never
1233 * even go to the slow case that looks at the
1234 * page queue), and add ourselves to the wait
1235 * queue if we need to sleep.
1236 *
1237 * This part needs to be done under the queue
1238 * lock to avoid races.
1239 */
1240 spin_lock_irq(&q->lock);
1241 SetPageWaiters(page);
1242 if (!trylock_page_bit_common(page, bit_nr, wait))
1243 __add_wait_queue_entry_tail(q, wait);
1244 spin_unlock_irq(&q->lock);
1245
1246 /*
1247 * From now on, all the logic will be based on
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001248 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1249 * see whether the page bit testing has already
1250 * been done by the wake function.
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001251 *
1252 * We can drop our reference to the page.
1253 */
1254 if (behavior == DROP)
1255 put_page(page);
1256
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001257 /*
1258 * Note that until the "finish_wait()", or until
1259 * we see the WQ_FLAG_WOKEN flag, we need to
1260 * be very careful with the 'wait->flags', because
1261 * we may race with a waker that sets them.
1262 */
Nicholas Piggin62906022016-12-25 13:00:30 +10001263 for (;;) {
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001264 unsigned int flags;
1265
Nicholas Piggin62906022016-12-25 13:00:30 +10001266 set_current_state(state);
1267
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001268 /* Loop until we've been woken or interrupted */
1269 flags = smp_load_acquire(&wait->flags);
1270 if (!(flags & WQ_FLAG_WOKEN)) {
1271 if (signal_pending_state(state, current))
1272 break;
1273
1274 io_schedule();
1275 continue;
1276 }
1277
1278 /* If we were non-exclusive, we're done */
1279 if (behavior != EXCLUSIVE)
Linus Torvaldsa8b169a2017-08-27 16:25:09 -07001280 break;
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001281
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001282 /* If the waker got the lock for us, we're done */
1283 if (flags & WQ_FLAG_DONE)
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001284 break;
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001285
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001286 /*
1287 * Otherwise, if we're getting the lock, we need to
1288 * try to get it ourselves.
1289 *
1290 * And if that fails, we'll have to retry this all.
1291 */
1292 if (unlikely(test_and_set_bit(bit_nr, &page->flags)))
1293 goto repeat;
1294
1295 wait->flags |= WQ_FLAG_DONE;
1296 break;
Nicholas Piggin62906022016-12-25 13:00:30 +10001297 }
1298
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001299 /*
1300 * If a signal happened, this 'finish_wait()' may remove the last
1301 * waiter from the wait-queues, but the PageWaiters bit will remain
1302 * set. That's ok. The next wakeup will take care of it, and trying
1303 * to do it here would be difficult and prone to races.
1304 */
Nicholas Piggin62906022016-12-25 13:00:30 +10001305 finish_wait(q, wait);
1306
Johannes Weinereb414682018-10-26 15:06:27 -07001307 if (thrashing) {
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001308 if (delayacct)
Johannes Weinereb414682018-10-26 15:06:27 -07001309 delayacct_thrashing_end();
1310 psi_memstall_leave(&pflags);
1311 }
Johannes Weinerb1d29ba2018-10-26 15:06:08 -07001312
Nicholas Piggin62906022016-12-25 13:00:30 +10001313 /*
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001314 * NOTE! The wait->flags weren't stable until we've done the
1315 * 'finish_wait()', and we could have exited the loop above due
1316 * to a signal, and had a wakeup event happen after the signal
1317 * test but before the 'finish_wait()'.
1318 *
1319 * So only after the finish_wait() can we reliably determine
1320 * if we got woken up or not, so we can now figure out the final
1321 * return value based on that state without races.
1322 *
1323 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1324 * waiter, but an exclusive one requires WQ_FLAG_DONE.
Nicholas Piggin62906022016-12-25 13:00:30 +10001325 */
Linus Torvalds5ef64cc2020-09-13 14:05:35 -07001326 if (behavior == EXCLUSIVE)
1327 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
Nicholas Piggin62906022016-12-25 13:00:30 +10001328
Linus Torvalds2a9127f2020-07-23 10:16:49 -07001329 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
Nicholas Piggin62906022016-12-25 13:00:30 +10001330}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
Harvey Harrison920c7a52008-02-04 22:29:26 -08001332void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
Nicholas Piggin62906022016-12-25 13:00:30 +10001334 wait_queue_head_t *q = page_waitqueue(page);
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001335 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336}
1337EXPORT_SYMBOL(wait_on_page_bit);
1338
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -07001339int wait_on_page_bit_killable(struct page *page, int bit_nr)
1340{
Nicholas Piggin62906022016-12-25 13:00:30 +10001341 wait_queue_head_t *q = page_waitqueue(page);
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001342 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -07001343}
David Howells4343d002017-11-02 15:27:52 +00001344EXPORT_SYMBOL(wait_on_page_bit_killable);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -07001345
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346/**
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001347 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1348 * @page: The page to wait for.
Matthew Wilcox (Oracle)48054622021-02-24 12:02:02 -08001349 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001350 *
1351 * The caller should hold a reference on @page. They expect the page to
1352 * become unlocked relatively soon, but do not wish to hold up migration
1353 * (for example) by holding the reference while waiting for the page to
1354 * come unlocked. After this function returns, the caller should not
1355 * dereference @page.
Matthew Wilcox (Oracle)48054622021-02-24 12:02:02 -08001356 *
1357 * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal.
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001358 */
Matthew Wilcox (Oracle)48054622021-02-24 12:02:02 -08001359int put_and_wait_on_page_locked(struct page *page, int state)
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001360{
1361 wait_queue_head_t *q;
1362
1363 page = compound_head(page);
1364 q = page_waitqueue(page);
Matthew Wilcox (Oracle)48054622021-02-24 12:02:02 -08001365 return wait_on_page_bit_common(q, page, PG_locked, state, DROP);
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001366}
1367
1368/**
David Howells385e1ca5f2009-04-03 16:42:39 +01001369 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -07001370 * @page: Page defining the wait queue of interest
1371 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +01001372 *
1373 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1374 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02001375void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
David Howells385e1ca5f2009-04-03 16:42:39 +01001376{
1377 wait_queue_head_t *q = page_waitqueue(page);
1378 unsigned long flags;
1379
1380 spin_lock_irqsave(&q->lock, flags);
Linus Torvalds9c3a8152017-08-28 16:45:40 -07001381 __add_wait_queue_entry_tail(q, waiter);
Nicholas Piggin62906022016-12-25 13:00:30 +10001382 SetPageWaiters(page);
David Howells385e1ca5f2009-04-03 16:42:39 +01001383 spin_unlock_irqrestore(&q->lock, flags);
1384}
1385EXPORT_SYMBOL_GPL(add_page_wait_queue);
1386
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001387#ifndef clear_bit_unlock_is_negative_byte
1388
1389/*
1390 * PG_waiters is the high bit in the same byte as PG_lock.
1391 *
1392 * On x86 (and on many other architectures), we can clear PG_lock and
1393 * test the sign bit at the same time. But if the architecture does
1394 * not support that special operation, we just do this all by hand
1395 * instead.
1396 *
1397 * The read of PG_waiters has to be after (or concurrently with) PG_locked
Ethon Paulffceeb62020-06-04 16:49:22 -07001398 * being cleared, but a memory barrier should be unnecessary since it is
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001399 * in the same byte as PG_locked.
1400 */
1401static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1402{
1403 clear_bit_unlock(nr, mem);
1404 /* smp_mb__after_atomic(); */
Olof Johansson98473f92016-12-29 14:16:07 -08001405 return test_bit(PG_waiters, mem);
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001406}
1407
1408#endif
1409
David Howells385e1ca5f2009-04-03 16:42:39 +01001410/**
Randy Dunlap485bb992006-06-23 02:03:49 -07001411 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 * @page: the page
1413 *
Miaohe Lin0e9aa672020-10-15 20:09:58 -07001414 * Unlocks the page and wakes up sleepers in wait_on_page_locked().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +09001416 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1418 *
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001419 * Note that this depends on PG_waiters being the sign bit in the byte
1420 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1421 * clear the PG_locked bit and test PG_waiters at the same time fairly
1422 * portably (architectures that do LL/SC can test any bit, while x86 can
1423 * test the sign bit).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001425void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426{
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001427 BUILD_BUG_ON(PG_waiters != 7);
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -08001428 page = compound_head(page);
Sasha Levin309381fea2014-01-23 15:52:54 -08001429 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001430 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1431 wake_up_page_bit(page, PG_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432}
1433EXPORT_SYMBOL(unlock_page);
1434
Randy Dunlap485bb992006-06-23 02:03:49 -07001435/**
1436 * end_page_writeback - end writeback against a page
1437 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 */
1439void end_page_writeback(struct page *page)
1440{
Mel Gorman888cf2d2014-06-04 16:10:34 -07001441 /*
1442 * TestClearPageReclaim could be used here but it is an atomic
1443 * operation and overkill in this particular case. Failing to
1444 * shuffle a page marked for immediate reclaim is too mild to
1445 * justify taking an atomic operation penalty at the end of
1446 * ever page writeback.
1447 */
1448 if (PageReclaim(page)) {
1449 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -07001450 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -07001451 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -07001452
Hugh Dickins073861e2020-11-24 08:46:43 -08001453 /*
1454 * Writeback does not hold a page reference of its own, relying
1455 * on truncation to wait for the clearing of PG_writeback.
1456 * But here we must make sure that the page is not freed and
1457 * reused before the wake_up_page().
1458 */
1459 get_page(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -07001460 if (!test_clear_page_writeback(page))
1461 BUG();
1462
Peter Zijlstra4e857c52014-03-17 18:06:10 +01001463 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 wake_up_page(page, PG_writeback);
Hugh Dickins073861e2020-11-24 08:46:43 -08001465 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466}
1467EXPORT_SYMBOL(end_page_writeback);
1468
Matthew Wilcox57d99842014-06-04 16:07:45 -07001469/*
1470 * After completing I/O on a page, call this routine to update the page
1471 * flags appropriately
1472 */
Jens Axboec11f0c02016-08-05 08:11:04 -06001473void page_endio(struct page *page, bool is_write, int err)
Matthew Wilcox57d99842014-06-04 16:07:45 -07001474{
Jens Axboec11f0c02016-08-05 08:11:04 -06001475 if (!is_write) {
Matthew Wilcox57d99842014-06-04 16:07:45 -07001476 if (!err) {
1477 SetPageUptodate(page);
1478 } else {
1479 ClearPageUptodate(page);
1480 SetPageError(page);
1481 }
1482 unlock_page(page);
Mike Christieabf54542016-08-04 14:23:34 -06001483 } else {
Matthew Wilcox57d99842014-06-04 16:07:45 -07001484 if (err) {
Minchan Kimdd8416c2017-02-24 14:59:59 -08001485 struct address_space *mapping;
1486
Matthew Wilcox57d99842014-06-04 16:07:45 -07001487 SetPageError(page);
Minchan Kimdd8416c2017-02-24 14:59:59 -08001488 mapping = page_mapping(page);
1489 if (mapping)
1490 mapping_set_error(mapping, err);
Matthew Wilcox57d99842014-06-04 16:07:45 -07001491 }
1492 end_page_writeback(page);
1493 }
1494}
1495EXPORT_SYMBOL_GPL(page_endio);
1496
Randy Dunlap485bb992006-06-23 02:03:49 -07001497/**
1498 * __lock_page - get a lock on the page, assuming we need to sleep to get it
Randy Dunlap87066752017-02-22 15:44:44 -08001499 * @__page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 */
Nicholas Piggin62906022016-12-25 13:00:30 +10001501void __lock_page(struct page *__page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502{
Nicholas Piggin62906022016-12-25 13:00:30 +10001503 struct page *page = compound_head(__page);
1504 wait_queue_head_t *q = page_waitqueue(page);
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001505 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1506 EXCLUSIVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507}
1508EXPORT_SYMBOL(__lock_page);
1509
Nicholas Piggin62906022016-12-25 13:00:30 +10001510int __lock_page_killable(struct page *__page)
Matthew Wilcox2687a352007-12-06 11:18:49 -05001511{
Nicholas Piggin62906022016-12-25 13:00:30 +10001512 struct page *page = compound_head(__page);
1513 wait_queue_head_t *q = page_waitqueue(page);
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001514 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1515 EXCLUSIVE);
Matthew Wilcox2687a352007-12-06 11:18:49 -05001516}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +03001517EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -05001518
Jens Axboedd3e6d52020-05-22 09:12:09 -06001519int __lock_page_async(struct page *page, struct wait_page_queue *wait)
1520{
Matthew Wilcox (Oracle)f32b5dd2021-02-24 12:02:09 -08001521 struct wait_queue_head *q = page_waitqueue(page);
1522 int ret = 0;
1523
1524 wait->page = page;
1525 wait->bit_nr = PG_locked;
1526
1527 spin_lock_irq(&q->lock);
1528 __add_wait_queue_entry_tail(q, &wait->wait);
1529 SetPageWaiters(page);
1530 ret = !trylock_page(page);
1531 /*
1532 * If we were successful now, we know we're still on the
1533 * waitqueue as we're still under the lock. This means it's
1534 * safe to remove and return success, we know the callback
1535 * isn't going to trigger.
1536 */
1537 if (!ret)
1538 __remove_wait_queue(q, &wait->wait);
1539 else
1540 ret = -EIOCBQUEUED;
1541 spin_unlock_irq(&q->lock);
1542 return ret;
Jens Axboedd3e6d52020-05-22 09:12:09 -06001543}
1544
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001545/*
1546 * Return values:
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07001547 * 1 - page is locked; mmap_lock is still held.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001548 * 0 - page is not locked.
Michel Lespinasse3e4e28c2020-06-08 21:33:51 -07001549 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001550 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07001551 * which case mmap_lock is still held.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001552 *
1553 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07001554 * with the page locked and the mmap_lock unperturbed.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001555 */
Michel Lespinassed065bd82010-10-26 14:21:57 -07001556int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1557 unsigned int flags)
1558{
Peter Xu4064b982020-04-01 21:08:45 -07001559 if (fault_flag_allow_retry_first(flags)) {
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001560 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07001561 * CAUTION! In this case, mmap_lock is not released
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001562 * even though return 0.
1563 */
1564 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1565 return 0;
1566
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07001567 mmap_read_unlock(mm);
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001568 if (flags & FAULT_FLAG_KILLABLE)
1569 wait_on_page_locked_killable(page);
1570 else
Gleb Natapov318b2752011-03-22 16:30:51 -07001571 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001572 return 0;
1573 }
Hailong Liu800bca72020-12-14 19:05:02 -08001574 if (flags & FAULT_FLAG_KILLABLE) {
1575 int ret;
1576
1577 ret = __lock_page_killable(page);
1578 if (ret) {
1579 mmap_read_unlock(mm);
1580 return 0;
1581 }
1582 } else {
1583 __lock_page(page);
1584 }
1585 return 1;
1586
Michel Lespinassed065bd82010-10-26 14:21:57 -07001587}
1588
Randy Dunlap485bb992006-06-23 02:03:49 -07001589/**
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001590 * page_cache_next_miss() - Find the next gap in the page cache.
1591 * @mapping: Mapping.
1592 * @index: Index.
1593 * @max_scan: Maximum range to search.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001594 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001595 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1596 * gap with the lowest index.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001597 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001598 * This function may be called under the rcu_read_lock. However, this will
1599 * not atomically search a snapshot of the cache at a single point in time.
1600 * For example, if a gap is created at index 5, then subsequently a gap is
1601 * created at index 10, page_cache_next_miss covering both indices may
1602 * return 10 if called under the rcu_read_lock.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001603 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001604 * Return: The index of the gap if found, otherwise an index outside the
1605 * range specified (in which case 'return - index >= max_scan' will be true).
1606 * In the rare case of index wrap-around, 0 will be returned.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001607 */
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001608pgoff_t page_cache_next_miss(struct address_space *mapping,
Johannes Weinere7b563b2014-04-03 14:47:44 -07001609 pgoff_t index, unsigned long max_scan)
1610{
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001611 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001612
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001613 while (max_scan--) {
1614 void *entry = xas_next(&xas);
1615 if (!entry || xa_is_value(entry))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001616 break;
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001617 if (xas.xa_index == 0)
Johannes Weinere7b563b2014-04-03 14:47:44 -07001618 break;
1619 }
1620
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001621 return xas.xa_index;
Johannes Weinere7b563b2014-04-03 14:47:44 -07001622}
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001623EXPORT_SYMBOL(page_cache_next_miss);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001624
1625/**
Laurent Dufour2346a562019-05-13 17:21:29 -07001626 * page_cache_prev_miss() - Find the previous gap in the page cache.
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001627 * @mapping: Mapping.
1628 * @index: Index.
1629 * @max_scan: Maximum range to search.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001630 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001631 * Search the range [max(index - max_scan + 1, 0), index] for the
1632 * gap with the highest index.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001633 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001634 * This function may be called under the rcu_read_lock. However, this will
1635 * not atomically search a snapshot of the cache at a single point in time.
1636 * For example, if a gap is created at index 10, then subsequently a gap is
1637 * created at index 5, page_cache_prev_miss() covering both indices may
1638 * return 5 if called under the rcu_read_lock.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001639 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001640 * Return: The index of the gap if found, otherwise an index outside the
1641 * range specified (in which case 'index - return >= max_scan' will be true).
1642 * In the rare case of wrap-around, ULONG_MAX will be returned.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001643 */
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001644pgoff_t page_cache_prev_miss(struct address_space *mapping,
Johannes Weinere7b563b2014-04-03 14:47:44 -07001645 pgoff_t index, unsigned long max_scan)
1646{
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001647 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001648
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001649 while (max_scan--) {
1650 void *entry = xas_prev(&xas);
1651 if (!entry || xa_is_value(entry))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001652 break;
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001653 if (xas.xa_index == ULONG_MAX)
Johannes Weinere7b563b2014-04-03 14:47:44 -07001654 break;
1655 }
1656
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001657 return xas.xa_index;
Johannes Weinere7b563b2014-04-03 14:47:44 -07001658}
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001659EXPORT_SYMBOL(page_cache_prev_miss);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001660
Matthew Wilcox (Oracle)44835d22021-02-25 17:15:36 -08001661/*
Johannes Weiner0cd61442014-04-03 14:47:46 -07001662 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -07001663 * @mapping: the address_space to search
Matthew Wilcox (Oracle)a6de4b42020-10-13 16:51:34 -07001664 * @index: The page cache index.
Randy Dunlap485bb992006-06-23 02:03:49 -07001665 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001666 * Looks up the page cache slot at @mapping & @offset. If there is a
Matthew Wilcox (Oracle)a6de4b42020-10-13 16:51:34 -07001667 * page cache page, the head page is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001668 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001669 * If the slot holds a shadow entry of a previously evicted page, or a
1670 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001671 *
Matthew Wilcox (Oracle)a6de4b42020-10-13 16:51:34 -07001672 * Return: The head page or shadow entry, %NULL if nothing is found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 */
Matthew Wilcox (Oracle)44835d22021-02-25 17:15:36 -08001674static struct page *find_get_entry(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675{
Matthew Wilcox (Oracle)a6de4b42020-10-13 16:51:34 -07001676 XA_STATE(xas, &mapping->i_pages, index);
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001677 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678
Nick Piggina60637c2008-07-25 19:45:31 -07001679 rcu_read_lock();
1680repeat:
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001681 xas_reset(&xas);
1682 page = xas_load(&xas);
1683 if (xas_retry(&xas, page))
1684 goto repeat;
1685 /*
1686 * A shadow entry of a recently evicted page, or a swap entry from
1687 * shmem/tmpfs. Return it without attempting to raise page count.
1688 */
1689 if (!page || xa_is_value(page))
1690 goto out;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001691
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001692 if (!page_cache_get_speculative(page))
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001693 goto repeat;
Nick Piggina60637c2008-07-25 19:45:31 -07001694
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001695 /*
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001696 * Has the page moved or been split?
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001697 * This is part of the lockless pagecache protocol. See
1698 * include/linux/pagemap.h for details.
1699 */
1700 if (unlikely(page != xas_reload(&xas))) {
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001701 put_page(page);
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001702 goto repeat;
Nick Piggina60637c2008-07-25 19:45:31 -07001703 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001704out:
Nick Piggina60637c2008-07-25 19:45:31 -07001705 rcu_read_unlock();
1706
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 return page;
1708}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709
Randy Dunlap485bb992006-06-23 02:03:49 -07001710/**
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001711 * pagecache_get_page - Find and get a reference to a page.
1712 * @mapping: The address_space to search.
1713 * @index: The page index.
1714 * @fgp_flags: %FGP flags modify how the page is returned.
1715 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001716 *
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001717 * Looks up the page cache entry at @mapping & @index.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001718 *
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001719 * @fgp_flags can be zero or more of these flags:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001720 *
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001721 * * %FGP_ACCESSED - The page will be marked accessed.
1722 * * %FGP_LOCK - The page is returned locked.
Matthew Wilcox (Oracle)a8cf7f22020-10-13 16:51:41 -07001723 * * %FGP_HEAD - If the page is present and a THP, return the head page
1724 * rather than the exact page specified by the index.
Matthew Wilcox (Oracle)44835d22021-02-25 17:15:36 -08001725 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1726 * instead of allocating a new page to replace it.
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001727 * * %FGP_CREAT - If no page is present then a new page is allocated using
1728 * @gfp_mask and added to the page cache and the VM's LRU list.
1729 * The page is returned locked and with an increased refcount.
1730 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1731 * page is already in cache. If the page was allocated, unlock it before
1732 * returning so the caller can do the same dance.
Yang Shi605cad82020-08-06 23:19:58 -07001733 * * %FGP_WRITE - The page will be written
1734 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1735 * * %FGP_NOWAIT - Don't get blocked by page lock
mchehab@s-opensource.com0e056eb2017-03-30 17:11:36 -03001736 *
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001737 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1738 * if the %GFP flags specified for %FGP_CREAT are atomic.
Mel Gorman2457aec2014-06-04 16:10:31 -07001739 *
1740 * If there is a page cache page, it is returned with an increased refcount.
Mike Rapoporta862f682019-03-05 15:48:42 -08001741 *
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001742 * Return: The found page or %NULL otherwise.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001743 */
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001744struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
1745 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746{
Nick Piggineb2be182007-10-16 01:24:57 -07001747 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001748
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749repeat:
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001750 page = find_get_entry(mapping, index);
Matthew Wilcox (Oracle)44835d22021-02-25 17:15:36 -08001751 if (xa_is_value(page)) {
1752 if (fgp_flags & FGP_ENTRY)
1753 return page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001754 page = NULL;
Matthew Wilcox (Oracle)44835d22021-02-25 17:15:36 -08001755 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001756 if (!page)
1757 goto no_page;
1758
1759 if (fgp_flags & FGP_LOCK) {
1760 if (fgp_flags & FGP_NOWAIT) {
1761 if (!trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001762 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001763 return NULL;
1764 }
1765 } else {
1766 lock_page(page);
1767 }
1768
1769 /* Has the page been truncated? */
Matthew Wilcox (Oracle)a8cf7f22020-10-13 16:51:41 -07001770 if (unlikely(page->mapping != mapping)) {
Mel Gorman2457aec2014-06-04 16:10:31 -07001771 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001772 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001773 goto repeat;
1774 }
Matthew Wilcox (Oracle)a8cf7f22020-10-13 16:51:41 -07001775 VM_BUG_ON_PAGE(!thp_contains(page, index), page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001776 }
1777
Kirill Tkhaic16eb002018-12-28 00:37:35 -08001778 if (fgp_flags & FGP_ACCESSED)
Mel Gorman2457aec2014-06-04 16:10:31 -07001779 mark_page_accessed(page);
Yang Shib9306a72020-08-06 23:19:55 -07001780 else if (fgp_flags & FGP_WRITE) {
1781 /* Clear idle flag for buffer write */
1782 if (page_is_idle(page))
1783 clear_page_idle(page);
1784 }
Matthew Wilcox (Oracle)a8cf7f22020-10-13 16:51:41 -07001785 if (!(fgp_flags & FGP_HEAD))
1786 page = find_subpage(page, index);
Mel Gorman2457aec2014-06-04 16:10:31 -07001787
1788no_page:
1789 if (!page && (fgp_flags & FGP_CREAT)) {
1790 int err;
Christoph Hellwigf56753a2020-09-24 08:51:40 +02001791 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001792 gfp_mask |= __GFP_WRITE;
1793 if (fgp_flags & FGP_NOFS)
1794 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001795
Michal Hocko45f87de2014-12-29 20:30:35 +01001796 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001797 if (!page)
1798 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001799
Josef Bacika75d4c32019-03-13 11:44:14 -07001800 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
Mel Gorman2457aec2014-06-04 16:10:31 -07001801 fgp_flags |= FGP_LOCK;
1802
Hugh Dickinseb39d612014-08-06 16:06:43 -07001803 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001804 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001805 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001806
Matthew Wilcox (Oracle)2294b322020-04-01 21:05:07 -07001807 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001808 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001809 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07001810 page = NULL;
1811 if (err == -EEXIST)
1812 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 }
Josef Bacika75d4c32019-03-13 11:44:14 -07001814
1815 /*
1816 * add_to_page_cache_lru locks the page, and for mmap we expect
1817 * an unlocked page.
1818 */
1819 if (page && (fgp_flags & FGP_FOR_MMAP))
1820 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001822
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 return page;
1824}
Mel Gorman2457aec2014-06-04 16:10:31 -07001825EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
1827/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001828 * find_get_entries - gang pagecache lookup
1829 * @mapping: The address_space to search
1830 * @start: The starting page cache index
1831 * @nr_entries: The maximum number of entries
1832 * @entries: Where the resulting entries are placed
1833 * @indices: The cache indices corresponding to the entries in @entries
1834 *
1835 * find_get_entries() will search for and return a group of up to
1836 * @nr_entries entries in the mapping. The entries are placed at
1837 * @entries. find_get_entries() takes a reference against any actual
1838 * pages it returns.
1839 *
1840 * The search returns a group of mapping-contiguous page cache entries
1841 * with ascending indexes. There may be holes in the indices due to
1842 * not-present pages.
1843 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001844 * Any shadow entries of evicted pages, or swap entries from
1845 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001846 *
Hugh Dickins71725ed2020-04-06 20:07:57 -07001847 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1848 * stops at that page: the caller is likely to have a better way to handle
1849 * the compound page as a whole, and then skip its extent, than repeatedly
1850 * calling find_get_entries() to return all its tails.
1851 *
Mike Rapoporta862f682019-03-05 15:48:42 -08001852 * Return: the number of pages and shadow entries which were found.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001853 */
1854unsigned find_get_entries(struct address_space *mapping,
1855 pgoff_t start, unsigned int nr_entries,
1856 struct page **entries, pgoff_t *indices)
1857{
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001858 XA_STATE(xas, &mapping->i_pages, start);
1859 struct page *page;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001860 unsigned int ret = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001861
1862 if (!nr_entries)
1863 return 0;
1864
1865 rcu_read_lock();
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001866 xas_for_each(&xas, page, ULONG_MAX) {
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001867 if (xas_retry(&xas, page))
Johannes Weiner0cd61442014-04-03 14:47:46 -07001868 continue;
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001869 /*
1870 * A shadow entry of a recently evicted page, a swap
1871 * entry from shmem/tmpfs or a DAX entry. Return it
1872 * without attempting to raise page count.
1873 */
1874 if (xa_is_value(page))
Johannes Weiner0cd61442014-04-03 14:47:46 -07001875 goto export;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001876
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001877 if (!page_cache_get_speculative(page))
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001878 goto retry;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001879
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001880 /* Has the page moved or been split? */
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001881 if (unlikely(page != xas_reload(&xas)))
1882 goto put_page;
1883
Hugh Dickins71725ed2020-04-06 20:07:57 -07001884 /*
1885 * Terminate early on finding a THP, to allow the caller to
1886 * handle it all at once; but continue if this is hugetlbfs.
1887 */
1888 if (PageTransHuge(page) && !PageHuge(page)) {
1889 page = find_subpage(page, xas.xa_index);
1890 nr_entries = ret + 1;
1891 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001892export:
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001893 indices[ret] = xas.xa_index;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001894 entries[ret] = page;
1895 if (++ret == nr_entries)
1896 break;
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001897 continue;
1898put_page:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001899 put_page(page);
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001900retry:
1901 xas_reset(&xas);
Johannes Weiner0cd61442014-04-03 14:47:46 -07001902 }
1903 rcu_read_unlock();
1904 return ret;
1905}
1906
1907/**
Jan Karab947cee2017-09-06 16:21:21 -07001908 * find_get_pages_range - gang pagecache lookup
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 * @mapping: The address_space to search
1910 * @start: The starting page index
Jan Karab947cee2017-09-06 16:21:21 -07001911 * @end: The final page index (inclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 * @nr_pages: The maximum number of pages
1913 * @pages: Where the resulting pages are placed
1914 *
Jan Karab947cee2017-09-06 16:21:21 -07001915 * find_get_pages_range() will search for and return a group of up to @nr_pages
1916 * pages in the mapping starting at index @start and up to index @end
1917 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1918 * a reference against the returned pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 *
1920 * The search returns a group of mapping-contiguous pages with ascending
1921 * indexes. There may be holes in the indices due to not-present pages.
Jan Karad72dc8a2017-09-06 16:21:18 -07001922 * We also update @start to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 *
Mike Rapoporta862f682019-03-05 15:48:42 -08001924 * Return: the number of pages which were found. If this number is
1925 * smaller than @nr_pages, the end of specified range has been
Jan Karab947cee2017-09-06 16:21:21 -07001926 * reached.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 */
Jan Karab947cee2017-09-06 16:21:21 -07001928unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1929 pgoff_t end, unsigned int nr_pages,
1930 struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931{
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001932 XA_STATE(xas, &mapping->i_pages, *start);
1933 struct page *page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001934 unsigned ret = 0;
1935
1936 if (unlikely(!nr_pages))
1937 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
Nick Piggina60637c2008-07-25 19:45:31 -07001939 rcu_read_lock();
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001940 xas_for_each(&xas, page, end) {
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001941 if (xas_retry(&xas, page))
Nick Piggina60637c2008-07-25 19:45:31 -07001942 continue;
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001943 /* Skip over shadow, swap and DAX entries */
1944 if (xa_is_value(page))
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001945 continue;
Nick Piggina60637c2008-07-25 19:45:31 -07001946
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001947 if (!page_cache_get_speculative(page))
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001948 goto retry;
Nick Piggina60637c2008-07-25 19:45:31 -07001949
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001950 /* Has the page moved or been split? */
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001951 if (unlikely(page != xas_reload(&xas)))
1952 goto put_page;
Nick Piggina60637c2008-07-25 19:45:31 -07001953
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001954 pages[ret] = find_subpage(page, xas.xa_index);
Jan Karab947cee2017-09-06 16:21:21 -07001955 if (++ret == nr_pages) {
Yu Zhao5d3ee422019-03-05 15:49:17 -08001956 *start = xas.xa_index + 1;
Jan Karab947cee2017-09-06 16:21:21 -07001957 goto out;
1958 }
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001959 continue;
1960put_page:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001961 put_page(page);
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001962retry:
1963 xas_reset(&xas);
Nick Piggina60637c2008-07-25 19:45:31 -07001964 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001965
Jan Karab947cee2017-09-06 16:21:21 -07001966 /*
1967 * We come here when there is no page beyond @end. We take care to not
1968 * overflow the index @start as it confuses some of the callers. This
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001969 * breaks the iteration when there is a page at index -1 but that is
Jan Karab947cee2017-09-06 16:21:21 -07001970 * already broken anyway.
1971 */
1972 if (end == (pgoff_t)-1)
1973 *start = (pgoff_t)-1;
1974 else
1975 *start = end + 1;
1976out:
Nick Piggina60637c2008-07-25 19:45:31 -07001977 rcu_read_unlock();
Jan Karad72dc8a2017-09-06 16:21:18 -07001978
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 return ret;
1980}
1981
Jens Axboeebf43502006-04-27 08:46:01 +02001982/**
1983 * find_get_pages_contig - gang contiguous pagecache lookup
1984 * @mapping: The address_space to search
1985 * @index: The starting page index
1986 * @nr_pages: The maximum number of pages
1987 * @pages: Where the resulting pages are placed
1988 *
1989 * find_get_pages_contig() works exactly like find_get_pages(), except
1990 * that the returned number of pages are guaranteed to be contiguous.
1991 *
Mike Rapoporta862f682019-03-05 15:48:42 -08001992 * Return: the number of pages which were found.
Jens Axboeebf43502006-04-27 08:46:01 +02001993 */
1994unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1995 unsigned int nr_pages, struct page **pages)
1996{
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04001997 XA_STATE(xas, &mapping->i_pages, index);
1998 struct page *page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001999 unsigned int ret = 0;
2000
2001 if (unlikely(!nr_pages))
2002 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02002003
Nick Piggina60637c2008-07-25 19:45:31 -07002004 rcu_read_lock();
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04002005 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04002006 if (xas_retry(&xas, page))
2007 continue;
2008 /*
2009 * If the entry has been swapped out, we can stop looking.
2010 * No current caller is looking for DAX entries.
2011 */
2012 if (xa_is_value(page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07002013 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07002014
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002015 if (!page_cache_get_speculative(page))
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04002016 goto retry;
Nick Piggina60637c2008-07-25 19:45:31 -07002017
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002018 /* Has the page moved or been split? */
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04002019 if (unlikely(page != xas_reload(&xas)))
2020 goto put_page;
Nick Piggina60637c2008-07-25 19:45:31 -07002021
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002022 pages[ret] = find_subpage(page, xas.xa_index);
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07002023 if (++ret == nr_pages)
2024 break;
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04002025 continue;
2026put_page:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002027 put_page(page);
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04002028retry:
2029 xas_reset(&xas);
Jens Axboeebf43502006-04-27 08:46:01 +02002030 }
Nick Piggina60637c2008-07-25 19:45:31 -07002031 rcu_read_unlock();
2032 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02002033}
David Howellsef71c152007-05-09 02:33:44 -07002034EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02002035
Randy Dunlap485bb992006-06-23 02:03:49 -07002036/**
Matthew Wilcox (Oracle)c49f50d2021-02-25 17:15:25 -08002037 * find_get_pages_range_tag - Find and return head pages matching @tag.
Randy Dunlap485bb992006-06-23 02:03:49 -07002038 * @mapping: the address_space to search
2039 * @index: the starting page index
Jan Kara72b045a2017-11-15 17:34:33 -08002040 * @end: The final page index (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -07002041 * @tag: the tag index
2042 * @nr_pages: the maximum number of pages
2043 * @pages: where the resulting pages are placed
2044 *
Matthew Wilcox (Oracle)c49f50d2021-02-25 17:15:25 -08002045 * Like find_get_pages(), except we only return head pages which are tagged
2046 * with @tag. @index is updated to the index immediately after the last
2047 * page we return, ready for the next iteration.
Mike Rapoporta862f682019-03-05 15:48:42 -08002048 *
2049 * Return: the number of pages which were found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 */
Jan Kara72b045a2017-11-15 17:34:33 -08002051unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002052 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
Jan Kara72b045a2017-11-15 17:34:33 -08002053 struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054{
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002055 XA_STATE(xas, &mapping->i_pages, *index);
2056 struct page *page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07002057 unsigned ret = 0;
2058
2059 if (unlikely(!nr_pages))
2060 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061
Nick Piggina60637c2008-07-25 19:45:31 -07002062 rcu_read_lock();
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002063 xas_for_each_marked(&xas, page, end, tag) {
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002064 if (xas_retry(&xas, page))
Nick Piggina60637c2008-07-25 19:45:31 -07002065 continue;
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002066 /*
2067 * Shadow entries should never be tagged, but this iteration
2068 * is lockless so there is a window for page reclaim to evict
2069 * a page we saw tagged. Skip over it.
2070 */
2071 if (xa_is_value(page))
Johannes Weiner139b6a62014-05-06 12:50:05 -07002072 continue;
Nick Piggina60637c2008-07-25 19:45:31 -07002073
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002074 if (!page_cache_get_speculative(page))
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002075 goto retry;
Nick Piggina60637c2008-07-25 19:45:31 -07002076
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002077 /* Has the page moved or been split? */
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002078 if (unlikely(page != xas_reload(&xas)))
2079 goto put_page;
Nick Piggina60637c2008-07-25 19:45:31 -07002080
Matthew Wilcox (Oracle)c49f50d2021-02-25 17:15:25 -08002081 pages[ret] = page;
Jan Kara72b045a2017-11-15 17:34:33 -08002082 if (++ret == nr_pages) {
Matthew Wilcox (Oracle)c49f50d2021-02-25 17:15:25 -08002083 *index = page->index + thp_nr_pages(page);
Jan Kara72b045a2017-11-15 17:34:33 -08002084 goto out;
2085 }
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002086 continue;
2087put_page:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002088 put_page(page);
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002089retry:
2090 xas_reset(&xas);
Nick Piggina60637c2008-07-25 19:45:31 -07002091 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07002092
Jan Kara72b045a2017-11-15 17:34:33 -08002093 /*
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002094 * We come here when we got to @end. We take care to not overflow the
Jan Kara72b045a2017-11-15 17:34:33 -08002095 * index @index as it confuses some of the callers. This breaks the
Matthew Wilcoxa6906972018-05-16 18:12:54 -04002096 * iteration when there is a page at index -1 but that is already
2097 * broken anyway.
Jan Kara72b045a2017-11-15 17:34:33 -08002098 */
2099 if (end == (pgoff_t)-1)
2100 *index = (pgoff_t)-1;
2101 else
2102 *index = end + 1;
2103out:
Nick Piggina60637c2008-07-25 19:45:31 -07002104 rcu_read_unlock();
2105
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 return ret;
2107}
Jan Kara72b045a2017-11-15 17:34:33 -08002108EXPORT_SYMBOL(find_get_pages_range_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109
Wu Fengguang76d42bd2006-06-25 05:48:43 -07002110/*
2111 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2112 * a _large_ part of the i/o request. Imagine the worst scenario:
2113 *
2114 * ---R__________________________________________B__________
2115 * ^ reading here ^ bad block(assume 4k)
2116 *
2117 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2118 * => failing the whole request => read(R) => read(R+1) =>
2119 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2120 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2121 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2122 *
2123 * It is going insane. Fix it by quickly scaling down the readahead size.
2124 */
Souptick Joarder0f8e2db2020-04-01 21:04:50 -07002125static void shrink_readahead_size_eio(struct file_ra_state *ra)
Wu Fengguang76d42bd2006-06-25 05:48:43 -07002126{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07002127 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07002128}
2129
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002130/*
2131 * filemap_get_read_batch - Get a batch of pages for read
2132 *
2133 * Get a batch of pages which represent a contiguous range of bytes
2134 * in the file. No tail pages will be returned. If @index is in the
2135 * middle of a THP, the entire THP will be returned. The last page in
2136 * the batch may have Readahead set or be not Uptodate so that the
2137 * caller can take the appropriate action.
2138 */
2139static void filemap_get_read_batch(struct address_space *mapping,
2140 pgoff_t index, pgoff_t max, struct pagevec *pvec)
2141{
2142 XA_STATE(xas, &mapping->i_pages, index);
2143 struct page *head;
2144
2145 rcu_read_lock();
2146 for (head = xas_load(&xas); head; head = xas_next(&xas)) {
2147 if (xas_retry(&xas, head))
2148 continue;
2149 if (xas.xa_index > max || xa_is_value(head))
2150 break;
2151 if (!page_cache_get_speculative(head))
2152 goto retry;
2153
2154 /* Has the page moved or been split? */
2155 if (unlikely(head != xas_reload(&xas)))
2156 goto put_page;
2157
2158 if (!pagevec_add(pvec, head))
2159 break;
2160 if (!PageUptodate(head))
2161 break;
2162 if (PageReadahead(head))
2163 break;
2164 xas.xa_index = head->index + thp_nr_pages(head) - 1;
2165 xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK;
2166 continue;
2167put_page:
2168 put_page(head);
2169retry:
2170 xas_reset(&xas);
2171 }
2172 rcu_read_unlock();
2173}
2174
Matthew Wilcox (Oracle)68430302021-02-24 12:02:15 -08002175static int filemap_read_page(struct file *file, struct address_space *mapping,
2176 struct page *page)
Kent Overstreet723ef242020-12-14 19:04:52 -08002177{
Kent Overstreet723ef242020-12-14 19:04:52 -08002178 int error;
2179
Kent Overstreet723ef242020-12-14 19:04:52 -08002180 /*
Matthew Wilcox (Oracle)68430302021-02-24 12:02:15 -08002181 * A previous I/O error may have been due to temporary failures,
2182 * eg. multipath errors. PG_error will be set again if readpage
2183 * fails.
Kent Overstreet723ef242020-12-14 19:04:52 -08002184 */
2185 ClearPageError(page);
2186 /* Start the actual read. The read will unlock the page. */
Matthew Wilcox (Oracle)68430302021-02-24 12:02:15 -08002187 error = mapping->a_ops->readpage(file, page);
2188 if (error)
2189 return error;
Kent Overstreet723ef242020-12-14 19:04:52 -08002190
Matthew Wilcox (Oracle)aa1ec2f2021-02-24 12:02:38 -08002191 error = wait_on_page_locked_killable(page);
Matthew Wilcox (Oracle)68430302021-02-24 12:02:15 -08002192 if (error)
2193 return error;
Matthew Wilcox (Oracle)aa1ec2f2021-02-24 12:02:38 -08002194 if (PageUptodate(page))
2195 return 0;
2196 if (!page->mapping) /* page truncated */
2197 return AOP_TRUNCATED_PAGE;
2198 shrink_readahead_size_eio(&file->f_ra);
2199 return -EIO;
Kent Overstreet723ef242020-12-14 19:04:52 -08002200}
2201
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002202static bool filemap_range_uptodate(struct address_space *mapping,
2203 loff_t pos, struct iov_iter *iter, struct page *page)
2204{
2205 int count;
2206
2207 if (PageUptodate(page))
2208 return true;
2209 /* pipes can't handle partially uptodate pages */
2210 if (iov_iter_is_pipe(iter))
2211 return false;
2212 if (!mapping->a_ops->is_partially_uptodate)
2213 return false;
2214 if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page)))
2215 return false;
2216
2217 count = iter->count;
2218 if (page_offset(page) > pos) {
2219 count -= page_offset(page) - pos;
2220 pos = 0;
2221 } else {
2222 pos -= page_offset(page);
2223 }
2224
2225 return mapping->a_ops->is_partially_uptodate(page, pos, count);
2226}
2227
Matthew Wilcox (Oracle)4612aee2021-02-24 12:02:22 -08002228static int filemap_update_page(struct kiocb *iocb,
2229 struct address_space *mapping, struct iov_iter *iter,
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002230 struct page *page)
Kent Overstreet723ef242020-12-14 19:04:52 -08002231{
Kent Overstreet723ef242020-12-14 19:04:52 -08002232 int error;
2233
Matthew Wilcox (Oracle)87d1d7b2021-02-24 12:02:25 -08002234 if (!trylock_page(page)) {
2235 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2236 return -EAGAIN;
2237 if (!(iocb->ki_flags & IOCB_WAITQ)) {
Matthew Wilcox (Oracle)bd8a1f32021-02-24 12:02:05 -08002238 put_and_wait_on_page_locked(page, TASK_KILLABLE);
Matthew Wilcox (Oracle)4612aee2021-02-24 12:02:22 -08002239 return AOP_TRUNCATED_PAGE;
Matthew Wilcox (Oracle)bd8a1f32021-02-24 12:02:05 -08002240 }
Matthew Wilcox (Oracle)87d1d7b2021-02-24 12:02:25 -08002241 error = __lock_page_async(page, iocb->ki_waitq);
2242 if (error)
2243 return error;
Kent Overstreet723ef242020-12-14 19:04:52 -08002244 }
Kent Overstreet723ef242020-12-14 19:04:52 -08002245
Matthew Wilcox (Oracle)bd8a1f32021-02-24 12:02:05 -08002246 if (!page->mapping)
2247 goto truncated;
Kent Overstreet723ef242020-12-14 19:04:52 -08002248
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002249 error = 0;
2250 if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, page))
2251 goto unlock;
2252
2253 error = -EAGAIN;
2254 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2255 goto unlock;
2256
Matthew Wilcox (Oracle)68430302021-02-24 12:02:15 -08002257 error = filemap_read_page(iocb->ki_filp, mapping, page);
Matthew Wilcox (Oracle)68430302021-02-24 12:02:15 -08002258 if (error == AOP_TRUNCATED_PAGE)
Matthew Wilcox (Oracle)4612aee2021-02-24 12:02:22 -08002259 put_page(page);
2260 return error;
Matthew Wilcox (Oracle)bd8a1f32021-02-24 12:02:05 -08002261truncated:
2262 unlock_page(page);
2263 put_page(page);
Matthew Wilcox (Oracle)4612aee2021-02-24 12:02:22 -08002264 return AOP_TRUNCATED_PAGE;
Matthew Wilcox (Oracle)fce70da2021-02-24 12:02:28 -08002265unlock:
2266 unlock_page(page);
2267 return error;
Kent Overstreet723ef242020-12-14 19:04:52 -08002268}
2269
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002270static int filemap_create_page(struct file *file,
2271 struct address_space *mapping, pgoff_t index,
2272 struct pagevec *pvec)
Kent Overstreet723ef242020-12-14 19:04:52 -08002273{
Kent Overstreet723ef242020-12-14 19:04:52 -08002274 struct page *page;
2275 int error;
2276
Kent Overstreet723ef242020-12-14 19:04:52 -08002277 page = page_cache_alloc(mapping);
2278 if (!page)
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002279 return -ENOMEM;
Kent Overstreet723ef242020-12-14 19:04:52 -08002280
2281 error = add_to_page_cache_lru(page, mapping, index,
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002282 mapping_gfp_constraint(mapping, GFP_KERNEL));
2283 if (error == -EEXIST)
2284 error = AOP_TRUNCATED_PAGE;
2285 if (error)
2286 goto error;
2287
2288 error = filemap_read_page(file, mapping, page);
2289 if (error)
2290 goto error;
2291
2292 pagevec_add(pvec, page);
2293 return 0;
2294error:
Matthew Wilcox (Oracle)68430302021-02-24 12:02:15 -08002295 put_page(page);
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002296 return error;
Kent Overstreet723ef242020-12-14 19:04:52 -08002297}
2298
Matthew Wilcox (Oracle)5963fe02021-02-24 12:02:32 -08002299static int filemap_readahead(struct kiocb *iocb, struct file *file,
2300 struct address_space *mapping, struct page *page,
2301 pgoff_t last_index)
2302{
2303 if (iocb->ki_flags & IOCB_NOIO)
2304 return -EAGAIN;
2305 page_cache_async_readahead(mapping, &file->f_ra, file, page,
2306 page->index, last_index - page->index);
2307 return 0;
2308}
2309
Matthew Wilcox (Oracle)3a6bae42021-02-24 12:01:49 -08002310static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
Matthew Wilcox (Oracle)ff993ba2021-02-24 12:01:55 -08002311 struct pagevec *pvec)
Kent Overstreet06c04442020-12-14 19:04:56 -08002312{
2313 struct file *filp = iocb->ki_filp;
2314 struct address_space *mapping = filp->f_mapping;
2315 struct file_ra_state *ra = &filp->f_ra;
2316 pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002317 pgoff_t last_index;
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002318 struct page *page;
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002319 int err = 0;
Kent Overstreet06c04442020-12-14 19:04:56 -08002320
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002321 last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002322retry:
Kent Overstreet06c04442020-12-14 19:04:56 -08002323 if (fatal_signal_pending(current))
2324 return -EINTR;
2325
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002326 filemap_get_read_batch(mapping, index, last_index, pvec);
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002327 if (!pagevec_count(pvec)) {
2328 if (iocb->ki_flags & IOCB_NOIO)
2329 return -EAGAIN;
2330 page_cache_sync_readahead(mapping, ra, filp, index,
2331 last_index - index);
2332 filemap_get_read_batch(mapping, index, last_index, pvec);
2333 }
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002334 if (!pagevec_count(pvec)) {
2335 if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2336 return -EAGAIN;
2337 err = filemap_create_page(filp, mapping,
2338 iocb->ki_pos >> PAGE_SHIFT, pvec);
2339 if (err == AOP_TRUNCATED_PAGE)
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002340 goto retry;
Matthew Wilcox (Oracle)f253e182021-02-24 12:02:18 -08002341 return err;
2342 }
Kent Overstreet06c04442020-12-14 19:04:56 -08002343
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002344 page = pvec->pages[pagevec_count(pvec) - 1];
2345 if (PageReadahead(page)) {
2346 err = filemap_readahead(iocb, filp, mapping, page, last_index);
2347 if (err)
2348 goto err;
2349 }
2350 if (!PageUptodate(page)) {
2351 if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1)
2352 iocb->ki_flags |= IOCB_NOWAIT;
2353 err = filemap_update_page(iocb, mapping, iter, page);
2354 if (err)
2355 goto err;
Kent Overstreet06c04442020-12-14 19:04:56 -08002356 }
2357
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002358 return 0;
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002359err:
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002360 if (err < 0)
2361 put_page(page);
2362 if (likely(--pvec->nr))
Matthew Wilcox (Oracle)ff993ba2021-02-24 12:01:55 -08002363 return 0;
Matthew Wilcox (Oracle)4612aee2021-02-24 12:02:22 -08002364 if (err == AOP_TRUNCATED_PAGE)
Matthew Wilcox (Oracle)2642fca2021-02-24 12:02:35 -08002365 goto retry;
2366 return err;
Kent Overstreet06c04442020-12-14 19:04:56 -08002367}
2368
Randy Dunlap485bb992006-06-23 02:03:49 -07002369/**
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002370 * filemap_read - Read data from the page cache.
2371 * @iocb: The iocb to read.
2372 * @iter: Destination for the data.
2373 * @already_read: Number of bytes already read by the caller.
Randy Dunlap485bb992006-06-23 02:03:49 -07002374 *
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002375 * Copies data from the page cache. If the data is not currently present,
2376 * uses the readahead and readpage address_space operations to fetch it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 *
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002378 * Return: Total number of bytes copied, including those already read by
2379 * the caller. If an error happens before any bytes are copied, returns
2380 * a negative error number.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 */
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002382ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2383 ssize_t already_read)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384{
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002385 struct file *filp = iocb->ki_filp;
Kent Overstreet06c04442020-12-14 19:04:56 -08002386 struct file_ra_state *ra = &filp->f_ra;
Christoph Hellwig36e78912008-02-08 04:21:24 -08002387 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 struct inode *inode = mapping->host;
Matthew Wilcox (Oracle)ff993ba2021-02-24 12:01:55 -08002389 struct pagevec pvec;
2390 int i, error = 0;
Kent Overstreet06c04442020-12-14 19:04:56 -08002391 bool writably_mapped;
2392 loff_t isize, end_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393
Kent Overstreet723ef242020-12-14 19:04:52 -08002394 if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
Linus Torvaldsd05c5f72016-12-14 12:45:25 -08002395 return 0;
Kent Overstreet3644e2d2020-12-18 04:07:11 -05002396 if (unlikely(!iov_iter_count(iter)))
2397 return 0;
2398
Wei Fangc2a97372016-10-07 17:01:52 -07002399 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002400 pagevec_init(&pvec);
Wei Fangc2a97372016-10-07 17:01:52 -07002401
Kent Overstreet06c04442020-12-14 19:04:56 -08002402 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 cond_resched();
Kent Overstreet06c04442020-12-14 19:04:56 -08002404
2405 /*
2406 * If we've already successfully copied some data, then we
2407 * can no longer safely return -EIOCBQUEUED. Hence mark
2408 * an async read NOWAIT at that point.
2409 */
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002410 if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
Kent Overstreet06c04442020-12-14 19:04:56 -08002411 iocb->ki_flags |= IOCB_NOWAIT;
2412
Matthew Wilcox (Oracle)ff993ba2021-02-24 12:01:55 -08002413 error = filemap_get_pages(iocb, iter, &pvec);
2414 if (error < 0)
Kent Overstreet06c04442020-12-14 19:04:56 -08002415 break;
Michal Hocko5abf1862017-02-03 13:13:29 -08002416
Kent Overstreet723ef242020-12-14 19:04:52 -08002417 /*
Kent Overstreet06c04442020-12-14 19:04:56 -08002418 * i_size must be checked after we know the pages are Uptodate.
2419 *
2420 * Checking i_size after the check allows us to calculate
2421 * the correct value for "nr", which means the zero-filled
2422 * part of the page is not copied back to userspace (unless
2423 * another truncate extends the file - this is desired though).
Kent Overstreet723ef242020-12-14 19:04:52 -08002424 */
Kent Overstreet06c04442020-12-14 19:04:56 -08002425 isize = i_size_read(inode);
2426 if (unlikely(iocb->ki_pos >= isize))
2427 goto put_pages;
Kent Overstreet06c04442020-12-14 19:04:56 -08002428 end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2429
Kent Overstreet06c04442020-12-14 19:04:56 -08002430 /*
2431 * Once we start copying data, we don't want to be touching any
2432 * cachelines that might be contended:
2433 */
2434 writably_mapped = mapping_writably_mapped(mapping);
2435
2436 /*
2437 * When a sequential read accesses a page several times, only
2438 * mark it as accessed the first time.
2439 */
2440 if (iocb->ki_pos >> PAGE_SHIFT !=
2441 ra->prev_pos >> PAGE_SHIFT)
Matthew Wilcox (Oracle)ff993ba2021-02-24 12:01:55 -08002442 mark_page_accessed(pvec.pages[0]);
Kent Overstreet06c04442020-12-14 19:04:56 -08002443
Matthew Wilcox (Oracle)ff993ba2021-02-24 12:01:55 -08002444 for (i = 0; i < pagevec_count(&pvec); i++) {
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002445 struct page *page = pvec.pages[i];
2446 size_t page_size = thp_size(page);
2447 size_t offset = iocb->ki_pos & (page_size - 1);
2448 size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2449 page_size - offset);
2450 size_t copied;
Kent Overstreet06c04442020-12-14 19:04:56 -08002451
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002452 if (end_offset < page_offset(page))
2453 break;
2454 if (i > 0)
2455 mark_page_accessed(page);
Kent Overstreet06c04442020-12-14 19:04:56 -08002456 /*
2457 * If users can be writing to this page using arbitrary
2458 * virtual addresses, take care about potential aliasing
2459 * before reading the page on the kernel side.
2460 */
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002461 if (writably_mapped) {
2462 int j;
Kent Overstreet06c04442020-12-14 19:04:56 -08002463
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002464 for (j = 0; j < thp_nr_pages(page); j++)
2465 flush_dcache_page(page + j);
2466 }
2467
2468 copied = copy_page_to_iter(page, offset, bytes, iter);
Kent Overstreet06c04442020-12-14 19:04:56 -08002469
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002470 already_read += copied;
Kent Overstreet06c04442020-12-14 19:04:56 -08002471 iocb->ki_pos += copied;
2472 ra->prev_pos = iocb->ki_pos;
2473
2474 if (copied < bytes) {
2475 error = -EFAULT;
2476 break;
Kent Overstreet723ef242020-12-14 19:04:52 -08002477 }
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07002478 }
Kent Overstreet06c04442020-12-14 19:04:56 -08002479put_pages:
Matthew Wilcox (Oracle)ff993ba2021-02-24 12:01:55 -08002480 for (i = 0; i < pagevec_count(&pvec); i++)
2481 put_page(pvec.pages[i]);
Matthew Wilcox (Oracle)cbd59c42021-02-24 12:01:59 -08002482 pagevec_reinit(&pvec);
Kent Overstreet06c04442020-12-14 19:04:56 -08002483 } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
Krishna Kumar0c6aa262008-10-15 22:01:13 -07002485 file_accessed(filp);
Kent Overstreet06c04442020-12-14 19:04:56 -08002486
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002487 return already_read ? already_read : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488}
Christoph Hellwig87fa0f32021-02-24 12:02:42 -08002489EXPORT_SYMBOL_GPL(filemap_read);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490
Randy Dunlap485bb992006-06-23 02:03:49 -07002491/**
Al Viro6abd2322014-04-04 14:20:57 -04002492 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07002493 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04002494 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07002495 *
Al Viro6abd2322014-04-04 14:20:57 -04002496 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 * that can use the page cache directly.
Andreas Gruenbacher41da51b2019-11-21 23:25:07 +00002498 *
2499 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2500 * be returned when no data can be read without waiting for I/O requests
2501 * to complete; it doesn't prevent readahead.
2502 *
2503 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2504 * requests shall be made for the read or for readahead. When no data
2505 * can be read, -EAGAIN shall be returned. When readahead would be
2506 * triggered, a partial, possibly empty read shall be returned.
2507 *
Mike Rapoporta862f682019-03-05 15:48:42 -08002508 * Return:
2509 * * number of bytes copied, even for partial reads
Andreas Gruenbacher41da51b2019-11-21 23:25:07 +00002510 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 */
2512ssize_t
Al Viroed978a82014-03-05 22:53:04 -05002513generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514{
Nicolai Stangee7080a42016-03-25 14:22:14 -07002515 size_t count = iov_iter_count(iter);
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002516 ssize_t retval = 0;
Nicolai Stangee7080a42016-03-25 14:22:14 -07002517
2518 if (!count)
Christoph Hellwig826ea862021-02-24 12:02:45 -08002519 return 0; /* skip atime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520
Al Viro2ba48ce2015-04-09 13:52:01 -04002521 if (iocb->ki_flags & IOCB_DIRECT) {
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002522 struct file *file = iocb->ki_filp;
Al Viroed978a82014-03-05 22:53:04 -05002523 struct address_space *mapping = file->f_mapping;
2524 struct inode *inode = mapping->host;
Badari Pulavarty543ade12006-09-30 23:28:48 -07002525 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 size = i_size_read(inode);
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05002528 if (iocb->ki_flags & IOCB_NOWAIT) {
2529 if (filemap_range_has_page(mapping, iocb->ki_pos,
2530 iocb->ki_pos + count - 1))
2531 return -EAGAIN;
2532 } else {
2533 retval = filemap_write_and_wait_range(mapping,
2534 iocb->ki_pos,
2535 iocb->ki_pos + count - 1);
2536 if (retval < 0)
Christoph Hellwig826ea862021-02-24 12:02:45 -08002537 return retval;
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05002538 }
Al Viroed978a82014-03-05 22:53:04 -05002539
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002540 file_accessed(file);
2541
Al Viro5ecda132017-04-13 14:13:36 -04002542 retval = mapping->a_ops->direct_IO(iocb, iter);
Al Viroc3a69022016-10-10 13:26:27 -04002543 if (retval >= 0) {
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07002544 iocb->ki_pos += retval;
Al Viro5ecda132017-04-13 14:13:36 -04002545 count -= retval;
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002546 }
Pavel Begunkovab2125d2021-02-24 12:01:45 -08002547 if (retval != -EIOCBQUEUED)
2548 iov_iter_revert(iter, count - iov_iter_count(iter));
Josef Bacik66f998f2010-05-23 11:00:54 -04002549
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002550 /*
2551 * Btrfs can have a short DIO read if we encounter
2552 * compressed extents, so if there was an error, or if
2553 * we've already read everything we wanted to, or if
2554 * there was a short read because we hit EOF, go ahead
2555 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002556 * the rest of the read. Buffered reads will not work for
2557 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002558 */
Al Viro5ecda132017-04-13 14:13:36 -04002559 if (retval < 0 || !count || iocb->ki_pos >= size ||
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002560 IS_DAX(inode))
Christoph Hellwig826ea862021-02-24 12:02:45 -08002561 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 }
2563
Christoph Hellwig826ea862021-02-24 12:02:45 -08002564 return filemap_read(iocb, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565}
Al Viroed978a82014-03-05 22:53:04 -05002566EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568#ifdef CONFIG_MMU
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569#define MMAP_LOTSAMISS (100)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002570/*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002571 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002572 * @vmf - the vm_fault for this fault.
2573 * @page - the page to lock.
2574 * @fpin - the pointer to the file we may pin (or is already pinned).
2575 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002576 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002577 * It differs in that it actually returns the page locked if it returns 1 and 0
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002578 * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002579 * will point to the pinned file and needs to be fput()'ed at a later point.
Linus Torvaldsef00e082009-06-16 15:31:25 -07002580 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002581static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2582 struct file **fpin)
2583{
2584 if (trylock_page(page))
2585 return 1;
2586
Linus Torvalds8b0f9fa2019-03-15 11:26:07 -07002587 /*
2588 * NOTE! This will make us return with VM_FAULT_RETRY, but with
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002589 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
Linus Torvalds8b0f9fa2019-03-15 11:26:07 -07002590 * is supposed to work. We have way too many special cases..
2591 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002592 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2593 return 0;
2594
2595 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2596 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2597 if (__lock_page_killable(page)) {
2598 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002599 * We didn't have the right flags to drop the mmap_lock,
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002600 * but all fault_handlers only check for fatal signals
2601 * if we return VM_FAULT_RETRY, so we need to drop the
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002602 * mmap_lock here and return 0 if we don't have a fpin.
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002603 */
2604 if (*fpin == NULL)
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07002605 mmap_read_unlock(vmf->vma->vm_mm);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002606 return 0;
2607 }
2608 } else
2609 __lock_page(page);
2610 return 1;
2611}
2612
2613
2614/*
2615 * Synchronous readahead happens when we don't even find a page in the page
2616 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2617 * to drop the mmap sem we return the file that was pinned in order for us to do
2618 * that. If we didn't pin a file then we return NULL. The file that is
2619 * returned needs to be fput()'ed when we're done with it.
2620 */
2621static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002622{
Josef Bacik2a1180f2019-03-13 11:44:18 -07002623 struct file *file = vmf->vma->vm_file;
2624 struct file_ra_state *ra = &file->f_ra;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002625 struct address_space *mapping = file->f_mapping;
David Howellsdb660d42020-10-15 20:06:31 -07002626 DEFINE_READAHEAD(ractl, file, mapping, vmf->pgoff);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002627 struct file *fpin = NULL;
Kirill A. Shutemove630bfa2020-08-14 17:31:27 -07002628 unsigned int mmap_miss;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002629
2630 /* If we don't want any read-ahead, don't bother */
Josef Bacik2a1180f2019-03-13 11:44:18 -07002631 if (vmf->vma->vm_flags & VM_RAND_READ)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002632 return fpin;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002633 if (!ra->ra_pages)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002634 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002635
Josef Bacik2a1180f2019-03-13 11:44:18 -07002636 if (vmf->vma->vm_flags & VM_SEQ_READ) {
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002637 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
David Howellsdb660d42020-10-15 20:06:31 -07002638 page_cache_sync_ra(&ractl, ra, ra->ra_pages);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002639 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002640 }
2641
Andi Kleen207d04b2011-05-24 17:12:29 -07002642 /* Avoid banging the cache line if not needed */
Kirill A. Shutemove630bfa2020-08-14 17:31:27 -07002643 mmap_miss = READ_ONCE(ra->mmap_miss);
2644 if (mmap_miss < MMAP_LOTSAMISS * 10)
2645 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002646
2647 /*
2648 * Do we miss much more than hit in this file? If so,
2649 * stop bothering with read-ahead. It will only hurt.
2650 */
Kirill A. Shutemove630bfa2020-08-14 17:31:27 -07002651 if (mmap_miss > MMAP_LOTSAMISS)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002652 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002653
Wu Fengguangd30a1102009-06-16 15:31:30 -07002654 /*
2655 * mmap read-around
2656 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002657 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
David Howellsdb660d42020-10-15 20:06:31 -07002658 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
Roman Gushchin600e19a2015-11-05 18:47:08 -08002659 ra->size = ra->ra_pages;
2660 ra->async_size = ra->ra_pages / 4;
David Howellsdb660d42020-10-15 20:06:31 -07002661 ractl._index = ra->start;
2662 do_page_cache_ra(&ractl, ra->size, ra->async_size);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002663 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002664}
2665
2666/*
2667 * Asynchronous readahead happens when we find the page and PG_readahead,
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002668 * so we want to possibly extend the readahead further. We return the file that
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002669 * was pinned if we have to drop the mmap_lock in order to do IO.
Linus Torvaldsef00e082009-06-16 15:31:25 -07002670 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002671static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2672 struct page *page)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002673{
Josef Bacik2a1180f2019-03-13 11:44:18 -07002674 struct file *file = vmf->vma->vm_file;
2675 struct file_ra_state *ra = &file->f_ra;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002676 struct address_space *mapping = file->f_mapping;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002677 struct file *fpin = NULL;
Kirill A. Shutemove630bfa2020-08-14 17:31:27 -07002678 unsigned int mmap_miss;
Josef Bacik2a1180f2019-03-13 11:44:18 -07002679 pgoff_t offset = vmf->pgoff;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002680
2681 /* If we don't want any read-ahead, don't bother */
Jan Kara5c72fee2020-04-01 21:04:40 -07002682 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002683 return fpin;
Kirill A. Shutemove630bfa2020-08-14 17:31:27 -07002684 mmap_miss = READ_ONCE(ra->mmap_miss);
2685 if (mmap_miss)
2686 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002687 if (PageReadahead(page)) {
2688 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
Wu Fengguang2fad6f52009-06-16 15:31:29 -07002689 page_cache_async_readahead(mapping, ra, file,
2690 page, offset, ra->ra_pages);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002691 }
2692 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002693}
2694
Randy Dunlap485bb992006-06-23 02:03:49 -07002695/**
Nick Piggin54cb8822007-07-19 01:46:59 -07002696 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07002697 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07002698 *
Nick Piggin54cb8822007-07-19 01:46:59 -07002699 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 * mapped memory region to read in file data during a page fault.
2701 *
2702 * The goto's are kind of ugly, but this streamlines the normal case of having
2703 * it in the page cache, and handles the special cases reasonably without
2704 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002705 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002706 * vma->vm_mm->mmap_lock must be held on entry.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002707 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002708 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
Yang Shia4985832019-07-11 20:55:29 -07002709 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002710 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002711 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002712 * has not been released.
2713 *
2714 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Mike Rapoporta862f682019-03-05 15:48:42 -08002715 *
2716 * Return: bitwise-OR of %VM_FAULT_ codes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002717 */
Souptick Joarder2bcd6452018-06-07 17:08:00 -07002718vm_fault_t filemap_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719{
2720 int error;
Dave Jiang11bac802017-02-24 14:56:41 -08002721 struct file *file = vmf->vma->vm_file;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002722 struct file *fpin = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 struct address_space *mapping = file->f_mapping;
2724 struct file_ra_state *ra = &file->f_ra;
2725 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002726 pgoff_t offset = vmf->pgoff;
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002727 pgoff_t max_off;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 struct page *page;
Souptick Joarder2bcd6452018-06-07 17:08:00 -07002729 vm_fault_t ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002731 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2732 if (unlikely(offset >= max_off))
Linus Torvalds5307cc12007-10-31 09:19:46 -07002733 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002736 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002738 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002739 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07002741 * We found the page, so try async readahead before
2742 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002744 fpin = do_async_mmap_readahead(vmf, page);
Shaohua Li45cac652012-10-08 16:32:19 -07002745 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07002746 /* No page in the page cache at all */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002747 count_vm_event(PGMAJFAULT);
Roman Gushchin22621852017-07-06 15:40:25 -07002748 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002749 ret = VM_FAULT_MAJOR;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002750 fpin = do_sync_mmap_readahead(vmf);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002751retry_find:
Josef Bacika75d4c32019-03-13 11:44:14 -07002752 page = pagecache_get_page(mapping, offset,
2753 FGP_CREAT|FGP_FOR_MMAP,
2754 vmf->gfp_mask);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002755 if (!page) {
2756 if (fpin)
2757 goto out_retry;
Matthew Wilcox (Oracle)e520e932020-04-01 21:04:53 -07002758 return VM_FAULT_OOM;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 }
2761
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002762 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2763 goto out_retry;
Michel Lespinasseb522c942010-10-26 14:21:56 -07002764
2765 /* Did it get truncated? */
Song Liu585e5a72019-09-23 15:37:44 -07002766 if (unlikely(compound_head(page)->mapping != mapping)) {
Michel Lespinasseb522c942010-10-26 14:21:56 -07002767 unlock_page(page);
2768 put_page(page);
2769 goto retry_find;
2770 }
Song Liu520e5ba2019-09-23 15:37:50 -07002771 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07002772
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 /*
Nick Piggind00806b2007-07-19 01:46:57 -07002774 * We have a locked page in the page cache, now we need to check
2775 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 */
Nick Piggind00806b2007-07-19 01:46:57 -07002777 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 goto page_not_uptodate;
2779
Linus Torvaldsef00e082009-06-16 15:31:25 -07002780 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002781 * We've made it this far and we had to drop our mmap_lock, now is the
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002782 * time to return to the upper layer and have it re-find the vma and
2783 * redo the fault.
2784 */
2785 if (fpin) {
2786 unlock_page(page);
2787 goto out_retry;
2788 }
2789
2790 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07002791 * Found the page and have a reference on it.
2792 * We must recheck i_size under page lock.
2793 */
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002794 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2795 if (unlikely(offset >= max_off)) {
Nick Piggind00806b2007-07-19 01:46:57 -07002796 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002797 put_page(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07002798 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07002799 }
2800
Nick Piggind0217ac2007-07-19 01:47:03 -07002801 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07002802 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 /*
2806 * Umm, take care of errors if the page isn't up-to-date.
2807 * Try to re-read it _once_. We do this synchronously,
2808 * because there really aren't any performance issues here
2809 * and we need to check for errors.
2810 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 ClearPageError(page);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002812 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
Zach Brown994fc28c2005-12-15 14:28:17 -08002813 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07002814 if (!error) {
2815 wait_on_page_locked(page);
2816 if (!PageUptodate(page))
2817 error = -EIO;
2818 }
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002819 if (fpin)
2820 goto out_retry;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002821 put_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002822
2823 if (!error || error == AOP_TRUNCATED_PAGE)
2824 goto retry_find;
2825
Souptick Joarder0f8e2db2020-04-01 21:04:50 -07002826 shrink_readahead_size_eio(ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002827 return VM_FAULT_SIGBUS;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002828
2829out_retry:
2830 /*
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -07002831 * We dropped the mmap_lock, we need to return to the fault handler to
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002832 * re-find the vma and come back and find our hopefully still populated
2833 * page.
2834 */
2835 if (page)
2836 put_page(page);
2837 if (fpin)
2838 fput(fpin);
2839 return ret | VM_FAULT_RETRY;
Nick Piggin54cb8822007-07-19 01:46:59 -07002840}
2841EXPORT_SYMBOL(filemap_fault);
2842
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002843static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002844{
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002845 struct mm_struct *mm = vmf->vma->vm_mm;
2846
2847 /* Huge page is mapped? No need to proceed. */
2848 if (pmd_trans_huge(*vmf->pmd)) {
2849 unlock_page(page);
2850 put_page(page);
2851 return true;
2852 }
2853
2854 if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
2855 vm_fault_t ret = do_set_pmd(vmf, page);
2856 if (!ret) {
2857 /* The page is mapped successfully, reference consumed. */
2858 unlock_page(page);
2859 return true;
2860 }
2861 }
2862
2863 if (pmd_none(*vmf->pmd)) {
2864 vmf->ptl = pmd_lock(mm, vmf->pmd);
2865 if (likely(pmd_none(*vmf->pmd))) {
2866 mm_inc_nr_ptes(mm);
2867 pmd_populate(mm, vmf->pmd, vmf->prealloc_pte);
2868 vmf->prealloc_pte = NULL;
2869 }
2870 spin_unlock(vmf->ptl);
2871 }
2872
2873 /* See comment in handle_pte_fault() */
2874 if (pmd_devmap_trans_unstable(vmf->pmd)) {
2875 unlock_page(page);
2876 put_page(page);
2877 return true;
2878 }
2879
2880 return false;
2881}
2882
2883static struct page *next_uptodate_page(struct page *page,
2884 struct address_space *mapping,
2885 struct xa_state *xas, pgoff_t end_pgoff)
2886{
2887 unsigned long max_idx;
2888
2889 do {
2890 if (!page)
2891 return NULL;
2892 if (xas_retry(xas, page))
2893 continue;
2894 if (xa_is_value(page))
2895 continue;
2896 if (PageLocked(page))
2897 continue;
2898 if (!page_cache_get_speculative(page))
2899 continue;
2900 /* Has the page moved or been split? */
2901 if (unlikely(page != xas_reload(xas)))
2902 goto skip;
2903 if (!PageUptodate(page) || PageReadahead(page))
2904 goto skip;
2905 if (PageHWPoison(page))
2906 goto skip;
2907 if (!trylock_page(page))
2908 goto skip;
2909 if (page->mapping != mapping)
2910 goto unlock;
2911 if (!PageUptodate(page))
2912 goto unlock;
2913 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2914 if (xas->xa_index >= max_idx)
2915 goto unlock;
2916 return page;
2917unlock:
2918 unlock_page(page);
2919skip:
2920 put_page(page);
2921 } while ((page = xas_next_entry(xas, end_pgoff)) != NULL);
2922
2923 return NULL;
2924}
2925
2926static inline struct page *first_map_page(struct address_space *mapping,
2927 struct xa_state *xas,
2928 pgoff_t end_pgoff)
2929{
2930 return next_uptodate_page(xas_find(xas, end_pgoff),
2931 mapping, xas, end_pgoff);
2932}
2933
2934static inline struct page *next_map_page(struct address_space *mapping,
2935 struct xa_state *xas,
2936 pgoff_t end_pgoff)
2937{
2938 return next_uptodate_page(xas_next_entry(xas, end_pgoff),
2939 mapping, xas, end_pgoff);
2940}
2941
2942vm_fault_t filemap_map_pages(struct vm_fault *vmf,
2943 pgoff_t start_pgoff, pgoff_t end_pgoff)
2944{
2945 struct vm_area_struct *vma = vmf->vma;
2946 struct file *file = vma->vm_file;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002947 struct address_space *mapping = file->f_mapping;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002948 pgoff_t last_pgoff = start_pgoff;
Will Deacon9d3af4b2021-01-14 15:24:19 +00002949 unsigned long addr;
Matthew Wilcox070e8072018-05-17 00:08:30 -04002950 XA_STATE(xas, &mapping->i_pages, start_pgoff);
Matthew Wilcox (Oracle)27a83a62020-10-13 16:51:44 -07002951 struct page *head, *page;
Kirill A. Shutemove630bfa2020-08-14 17:31:27 -07002952 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002953 vm_fault_t ret = 0;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002954
2955 rcu_read_lock();
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002956 head = first_map_page(mapping, &xas, end_pgoff);
2957 if (!head)
2958 goto out;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002959
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002960 if (filemap_map_pmd(vmf, head)) {
2961 ret = VM_FAULT_NOPAGE;
2962 goto out;
2963 }
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002964
Will Deacon9d3af4b2021-01-14 15:24:19 +00002965 addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
2966 vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002967 do {
Matthew Wilcox (Oracle)27a83a62020-10-13 16:51:44 -07002968 page = find_subpage(head, xas.xa_index);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002969 if (PageHWPoison(page))
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002970 goto unlock;
2971
Kirill A. Shutemove630bfa2020-08-14 17:31:27 -07002972 if (mmap_miss > 0)
2973 mmap_miss--;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002974
Will Deacon9d3af4b2021-01-14 15:24:19 +00002975 addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002976 vmf->pte += xas.xa_index - last_pgoff;
Matthew Wilcox070e8072018-05-17 00:08:30 -04002977 last_pgoff = xas.xa_index;
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002978
2979 if (!pte_none(*vmf->pte))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002980 goto unlock;
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002981
Will Deacon46bdb422020-11-24 18:48:26 +00002982 /* We're about to handle the fault */
Will Deacon9d3af4b2021-01-14 15:24:19 +00002983 if (vmf->address == addr)
Will Deacon46bdb422020-11-24 18:48:26 +00002984 ret = VM_FAULT_NOPAGE;
Will Deacon46bdb422020-11-24 18:48:26 +00002985
Will Deacon9d3af4b2021-01-14 15:24:19 +00002986 do_set_pte(vmf, page, addr);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002987 /* no need to invalidate: a not-present page won't be cached */
Will Deacon9d3af4b2021-01-14 15:24:19 +00002988 update_mmu_cache(vma, addr, vmf->pte);
Matthew Wilcox (Oracle)27a83a62020-10-13 16:51:44 -07002989 unlock_page(head);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002990 continue;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002991unlock:
Matthew Wilcox (Oracle)27a83a62020-10-13 16:51:44 -07002992 unlock_page(head);
Matthew Wilcox (Oracle)27a83a62020-10-13 16:51:44 -07002993 put_page(head);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002994 } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL);
2995 pte_unmap_unlock(vmf->pte, vmf->ptl);
2996out:
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002997 rcu_read_unlock();
Kirill A. Shutemove630bfa2020-08-14 17:31:27 -07002998 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
Kirill A. Shutemovf9ce0be2020-12-19 15:19:23 +03002999 return ret;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07003000}
3001EXPORT_SYMBOL(filemap_map_pages);
3002
Souptick Joarder2bcd6452018-06-07 17:08:00 -07003003vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
Jan Kara4fcf1c62012-06-12 16:20:29 +02003004{
Christoph Hellwig5df1a672020-11-16 14:33:37 +01003005 struct address_space *mapping = vmf->vma->vm_file->f_mapping;
Jan Kara4fcf1c62012-06-12 16:20:29 +02003006 struct page *page = vmf->page;
Souptick Joarder2bcd6452018-06-07 17:08:00 -07003007 vm_fault_t ret = VM_FAULT_LOCKED;
Jan Kara4fcf1c62012-06-12 16:20:29 +02003008
Christoph Hellwig5df1a672020-11-16 14:33:37 +01003009 sb_start_pagefault(mapping->host->i_sb);
Dave Jiang11bac802017-02-24 14:56:41 -08003010 file_update_time(vmf->vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02003011 lock_page(page);
Christoph Hellwig5df1a672020-11-16 14:33:37 +01003012 if (page->mapping != mapping) {
Jan Kara4fcf1c62012-06-12 16:20:29 +02003013 unlock_page(page);
3014 ret = VM_FAULT_NOPAGE;
3015 goto out;
3016 }
Jan Kara14da9202012-06-12 16:20:37 +02003017 /*
3018 * We mark the page dirty already here so that when freeze is in
3019 * progress, we are guaranteed that writeback during freezing will
3020 * see the dirty page and writeprotect it again.
3021 */
3022 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08003023 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02003024out:
Christoph Hellwig5df1a672020-11-16 14:33:37 +01003025 sb_end_pagefault(mapping->host->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02003026 return ret;
3027}
Jan Kara4fcf1c62012-06-12 16:20:29 +02003028
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04003029const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003030 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07003031 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02003032 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033};
3034
3035/* This is used for a general mmap of a disk file */
3036
3037int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3038{
3039 struct address_space *mapping = file->f_mapping;
3040
3041 if (!mapping->a_ops->readpage)
3042 return -ENOEXEC;
3043 file_accessed(file);
3044 vma->vm_ops = &generic_file_vm_ops;
3045 return 0;
3046}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047
3048/*
3049 * This is for filesystems which do not implement ->writepage.
3050 */
3051int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3052{
3053 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3054 return -EINVAL;
3055 return generic_file_mmap(file, vma);
3056}
3057#else
Souptick Joarder4b96a372018-10-26 15:04:03 -07003058vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
Arnd Bergmann453972282018-04-13 15:35:27 -07003059{
Souptick Joarder4b96a372018-10-26 15:04:03 -07003060 return VM_FAULT_SIGBUS;
Arnd Bergmann453972282018-04-13 15:35:27 -07003061}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3063{
3064 return -ENOSYS;
3065}
3066int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
3067{
3068 return -ENOSYS;
3069}
3070#endif /* CONFIG_MMU */
3071
Arnd Bergmann453972282018-04-13 15:35:27 -07003072EXPORT_SYMBOL(filemap_page_mkwrite);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073EXPORT_SYMBOL(generic_file_mmap);
3074EXPORT_SYMBOL(generic_file_readonly_mmap);
3075
Sasha Levin67f9fd92014-04-03 14:48:18 -07003076static struct page *wait_on_page_read(struct page *page)
3077{
3078 if (!IS_ERR(page)) {
3079 wait_on_page_locked(page);
3080 if (!PageUptodate(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003081 put_page(page);
Sasha Levin67f9fd92014-04-03 14:48:18 -07003082 page = ERR_PTR(-EIO);
3083 }
3084 }
3085 return page;
3086}
3087
Mel Gorman32b63522016-03-15 14:55:36 -07003088static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07003089 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07003090 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003091 void *data,
3092 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093{
Nick Piggineb2be182007-10-16 01:24:57 -07003094 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 int err;
3096repeat:
3097 page = find_get_page(mapping, index);
3098 if (!page) {
Mel Gorman453f85d2017-11-15 17:38:03 -08003099 page = __page_cache_alloc(gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07003100 if (!page)
3101 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06003102 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07003103 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003104 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07003105 if (err == -EEXIST)
3106 goto repeat;
Matthew Wilcox22ecdb42017-12-04 04:02:00 -05003107 /* Presumably ENOMEM for xarray node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 return ERR_PTR(err);
3109 }
Mel Gorman32b63522016-03-15 14:55:36 -07003110
3111filler:
Christoph Hellwig6c45b452019-07-11 20:55:20 -07003112 if (filler)
3113 err = filler(data, page);
3114 else
3115 err = mapping->a_ops->readpage(data, page);
3116
Linus Torvalds1da177e2005-04-16 15:20:36 -07003117 if (err < 0) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003118 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07003119 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 }
Mel Gorman32b63522016-03-15 14:55:36 -07003121
3122 page = wait_on_page_read(page);
3123 if (IS_ERR(page))
3124 return page;
3125 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003127 if (PageUptodate(page))
3128 goto out;
3129
Mel Gormanebded022016-03-15 14:55:39 -07003130 /*
Miaohe Lin0e9aa672020-10-15 20:09:58 -07003131 * Page is not up to date and may be locked due to one of the following
Mel Gormanebded022016-03-15 14:55:39 -07003132 * case a: Page is being filled and the page lock is held
3133 * case b: Read/write error clearing the page uptodate status
3134 * case c: Truncation in progress (page locked)
3135 * case d: Reclaim in progress
3136 *
3137 * Case a, the page will be up to date when the page is unlocked.
3138 * There is no need to serialise on the page lock here as the page
3139 * is pinned so the lock gives no additional protection. Even if the
Randy Dunlapce89fddf2020-08-11 18:32:53 -07003140 * page is truncated, the data is still valid if PageUptodate as
Mel Gormanebded022016-03-15 14:55:39 -07003141 * it's a race vs truncate race.
3142 * Case b, the page will not be up to date
3143 * Case c, the page may be truncated but in itself, the data may still
3144 * be valid after IO completes as it's a read vs truncate race. The
3145 * operation must restart if the page is not uptodate on unlock but
3146 * otherwise serialising on page lock to stabilise the mapping gives
3147 * no additional guarantees to the caller as the page lock is
3148 * released before return.
3149 * Case d, similar to truncation. If reclaim holds the page lock, it
3150 * will be a race with remove_mapping that determines if the mapping
3151 * is valid on unlock but otherwise the data is valid and there is
3152 * no need to serialise with page lock.
3153 *
3154 * As the page lock gives no additional guarantee, we optimistically
3155 * wait on the page to be unlocked and check if it's up to date and
3156 * use the page if it is. Otherwise, the page lock is required to
3157 * distinguish between the different cases. The motivation is that we
3158 * avoid spurious serialisations and wakeups when multiple processes
3159 * wait on the same page for IO to complete.
3160 */
3161 wait_on_page_locked(page);
3162 if (PageUptodate(page))
3163 goto out;
3164
3165 /* Distinguish between all the cases under the safety of the lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166 lock_page(page);
Mel Gormanebded022016-03-15 14:55:39 -07003167
3168 /* Case c or d, restart the operation */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169 if (!page->mapping) {
3170 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003171 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07003172 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173 }
Mel Gormanebded022016-03-15 14:55:39 -07003174
3175 /* Someone else locked and filled the page in a very small window */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003176 if (PageUptodate(page)) {
3177 unlock_page(page);
3178 goto out;
3179 }
Xianting Tianfaffdfa2020-04-01 21:04:47 -07003180
3181 /*
3182 * A previous I/O error may have been due to temporary
3183 * failures.
3184 * Clear page error before actual read, PG_error will be
3185 * set again if read page fails.
3186 */
3187 ClearPageError(page);
Mel Gorman32b63522016-03-15 14:55:36 -07003188 goto filler;
3189
David Howellsc855ff32007-05-09 13:42:20 +01003190out:
Nick Piggin6fe69002007-05-06 14:49:04 -07003191 mark_page_accessed(page);
3192 return page;
3193}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003194
3195/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07003196 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003197 * @mapping: the page's address_space
3198 * @index: the page index
3199 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07003200 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003201 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003202 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07003203 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003204 *
3205 * If the page does not get brought uptodate, return -EIO.
Mike Rapoporta862f682019-03-05 15:48:42 -08003206 *
3207 * Return: up to date page on success, ERR_PTR() on failure.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003208 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07003209struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003210 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07003211 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003212 void *data)
3213{
Christoph Hellwigd322a8e2019-07-11 20:55:17 -07003214 return do_read_cache_page(mapping, index, filler, data,
3215 mapping_gfp_mask(mapping));
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003216}
Sasha Levin67f9fd92014-04-03 14:48:18 -07003217EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003218
3219/**
3220 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3221 * @mapping: the page's address_space
3222 * @index: the page index
3223 * @gfp: the page allocator flags to use if allocating
3224 *
3225 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06003226 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003227 *
3228 * If the page does not get brought uptodate, return -EIO.
Mike Rapoporta862f682019-03-05 15:48:42 -08003229 *
3230 * Return: up to date page on success, ERR_PTR() on failure.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003231 */
3232struct page *read_cache_page_gfp(struct address_space *mapping,
3233 pgoff_t index,
3234 gfp_t gfp)
3235{
Christoph Hellwig6c45b452019-07-11 20:55:20 -07003236 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08003237}
3238EXPORT_SYMBOL(read_cache_page_gfp);
3239
Nick Pigginafddba42007-10-16 01:25:01 -07003240int pagecache_write_begin(struct file *file, struct address_space *mapping,
3241 loff_t pos, unsigned len, unsigned flags,
3242 struct page **pagep, void **fsdata)
3243{
3244 const struct address_space_operations *aops = mapping->a_ops;
3245
Nick Piggin4e02ed42008-10-29 14:00:55 -07003246 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07003247 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07003248}
3249EXPORT_SYMBOL(pagecache_write_begin);
3250
3251int pagecache_write_end(struct file *file, struct address_space *mapping,
3252 loff_t pos, unsigned len, unsigned copied,
3253 struct page *page, void *fsdata)
3254{
3255 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07003256
Nick Piggin4e02ed42008-10-29 14:00:55 -07003257 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07003258}
3259EXPORT_SYMBOL(pagecache_write_end);
3260
Konstantin Khlebnikova92853b2019-11-30 17:49:44 -08003261/*
3262 * Warn about a page cache invalidation failure during a direct I/O write.
3263 */
3264void dio_warn_stale_pagecache(struct file *filp)
3265{
3266 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3267 char pathname[128];
Konstantin Khlebnikova92853b2019-11-30 17:49:44 -08003268 char *path;
3269
Christoph Hellwig5df1a672020-11-16 14:33:37 +01003270 errseq_set(&filp->f_mapping->wb_err, -EIO);
Konstantin Khlebnikova92853b2019-11-30 17:49:44 -08003271 if (__ratelimit(&_rs)) {
3272 path = file_path(filp, pathname, sizeof(pathname));
3273 if (IS_ERR(path))
3274 path = "(unknown)";
3275 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3276 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3277 current->comm);
3278 }
3279}
3280
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281ssize_t
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07003282generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283{
3284 struct file *file = iocb->ki_filp;
3285 struct address_space *mapping = file->f_mapping;
3286 struct inode *inode = mapping->host;
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07003287 loff_t pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003288 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07003289 size_t write_len;
3290 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291
Al Viro0c949332014-03-22 06:51:37 -04003292 write_len = iov_iter_count(from);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003293 end = (pos + write_len - 1) >> PAGE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07003294
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05003295 if (iocb->ki_flags & IOCB_NOWAIT) {
3296 /* If there are pages to writeback, return */
Christoph Hellwig5df1a672020-11-16 14:33:37 +01003297 if (filemap_range_has_page(file->f_mapping, pos,
zhengbin35f12f02019-03-05 15:44:21 -08003298 pos + write_len - 1))
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05003299 return -EAGAIN;
3300 } else {
3301 written = filemap_write_and_wait_range(mapping, pos,
3302 pos + write_len - 1);
3303 if (written)
3304 goto out;
3305 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07003306
3307 /*
3308 * After a write we want buffered reads to be sure to go to disk to get
3309 * the new data. We invalidate clean cached page from the region we're
3310 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07003311 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07003312 */
Andrey Ryabinin55635ba2017-05-03 14:55:59 -07003313 written = invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003314 pos >> PAGE_SHIFT, end);
Andrey Ryabinin55635ba2017-05-03 14:55:59 -07003315 /*
3316 * If a page can not be invalidated, return 0 to fall back
3317 * to buffered write.
3318 */
3319 if (written) {
3320 if (written == -EBUSY)
3321 return 0;
3322 goto out;
Christoph Hellwiga969e902008-07-23 21:27:04 -07003323 }
3324
Al Viro639a93a52017-04-13 14:10:15 -04003325 written = mapping->a_ops->direct_IO(iocb, from);
Christoph Hellwiga969e902008-07-23 21:27:04 -07003326
3327 /*
3328 * Finally, try again to invalidate clean pages which might have been
3329 * cached by non-direct readahead, or faulted in by get_user_pages()
3330 * if the source of the write was an mmap'ed region of the file
3331 * we're writing. Either one is a pretty crazy thing to do,
3332 * so we don't support it 100%. If this invalidation
3333 * fails, tough, the write still worked...
Lukas Czerner332391a2017-09-21 08:16:29 -06003334 *
3335 * Most of the time we do not need this since dio_complete() will do
3336 * the invalidation for us. However there are some file systems that
3337 * do not end up with dio_complete() being called, so let's not break
Konstantin Khlebnikov80c1fe92019-11-30 17:49:41 -08003338 * them by removing it completely.
3339 *
Konstantin Khlebnikov9266a142019-11-30 17:49:47 -08003340 * Noticeable example is a blkdev_direct_IO().
3341 *
Konstantin Khlebnikov80c1fe92019-11-30 17:49:41 -08003342 * Skip invalidation for async writes or if mapping has no pages.
Christoph Hellwiga969e902008-07-23 21:27:04 -07003343 */
Konstantin Khlebnikov9266a142019-11-30 17:49:47 -08003344 if (written > 0 && mapping->nrpages &&
3345 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3346 dio_warn_stale_pagecache(file);
Christoph Hellwiga969e902008-07-23 21:27:04 -07003347
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07003349 pos += written;
Al Viro639a93a52017-04-13 14:10:15 -04003350 write_len -= written;
Namhyung Kim01166512010-10-26 14:21:58 -07003351 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3352 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353 mark_inode_dirty(inode);
3354 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05003355 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003356 }
Pavel Begunkovab2125d2021-02-24 12:01:45 -08003357 if (written != -EIOCBQUEUED)
3358 iov_iter_revert(from, write_len - iov_iter_count(from));
Christoph Hellwiga969e902008-07-23 21:27:04 -07003359out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360 return written;
3361}
3362EXPORT_SYMBOL(generic_file_direct_write);
3363
Nick Piggineb2be182007-10-16 01:24:57 -07003364/*
3365 * Find or create a page at the given pagecache position. Return the locked
3366 * page. This function is specifically for buffered writes.
3367 */
Nick Piggin54566b22009-01-04 12:00:53 -08003368struct page *grab_cache_page_write_begin(struct address_space *mapping,
3369 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07003370{
Nick Piggineb2be182007-10-16 01:24:57 -07003371 struct page *page;
Johannes Weinerbbddabe2016-05-20 16:56:28 -07003372 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08003373
Nick Piggin54566b22009-01-04 12:00:53 -08003374 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07003375 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07003376
Mel Gorman2457aec2014-06-04 16:10:31 -07003377 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01003378 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07003379 if (page)
3380 wait_for_stable_page(page);
3381
Nick Piggineb2be182007-10-16 01:24:57 -07003382 return page;
3383}
Nick Piggin54566b22009-01-04 12:00:53 -08003384EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07003385
Al Viro3b93f912014-02-11 21:34:08 -05003386ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07003387 struct iov_iter *i, loff_t pos)
3388{
3389 struct address_space *mapping = file->f_mapping;
3390 const struct address_space_operations *a_ops = mapping->a_ops;
3391 long status = 0;
3392 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07003393 unsigned int flags = 0;
3394
Nick Pigginafddba42007-10-16 01:25:01 -07003395 do {
3396 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003397 unsigned long offset; /* Offset into pagecache page */
3398 unsigned long bytes; /* Bytes to write to page */
3399 size_t copied; /* Bytes copied from user */
3400 void *fsdata;
3401
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003402 offset = (pos & (PAGE_SIZE - 1));
3403 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07003404 iov_iter_count(i));
3405
3406again:
Linus Torvalds00a3d662015-10-07 08:32:38 +01003407 /*
3408 * Bring in the user page that we will copy from _first_.
3409 * Otherwise there's a nasty deadlock on copying from the
3410 * same page as we're writing to, without it being marked
3411 * up-to-date.
3412 *
3413 * Not only is this an optimisation, but it is also required
3414 * to check that the address is actually valid, when atomic
3415 * usercopies are used, below.
3416 */
3417 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3418 status = -EFAULT;
3419 break;
3420 }
3421
Jan Kara296291c2015-10-22 13:32:21 -07003422 if (fatal_signal_pending(current)) {
3423 status = -EINTR;
3424 break;
3425 }
3426
Nick Piggin674b8922007-10-16 01:25:03 -07003427 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07003428 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07003429 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07003430 break;
3431
anfei zhou931e80e2010-02-02 13:44:02 -08003432 if (mapping_writably_mapped(mapping))
3433 flush_dcache_page(page);
Linus Torvalds00a3d662015-10-07 08:32:38 +01003434
Nick Pigginafddba42007-10-16 01:25:01 -07003435 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07003436 flush_dcache_page(page);
3437
3438 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3439 page, fsdata);
3440 if (unlikely(status < 0))
3441 break;
3442 copied = status;
3443
3444 cond_resched();
3445
Nick Piggin124d3b72008-02-02 15:01:17 +01003446 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07003447 if (unlikely(copied == 0)) {
3448 /*
3449 * If we were unable to copy any data at all, we must
3450 * fall back to a single segment length write.
3451 *
3452 * If we didn't fallback here, we could livelock
3453 * because not all segments in the iov can be copied at
3454 * once without a pagefault.
3455 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003456 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07003457 iov_iter_single_seg_count(i));
3458 goto again;
3459 }
Nick Pigginafddba42007-10-16 01:25:01 -07003460 pos += copied;
3461 written += copied;
3462
3463 balance_dirty_pages_ratelimited(mapping);
Nick Pigginafddba42007-10-16 01:25:01 -07003464 } while (iov_iter_count(i));
3465
3466 return written ? written : status;
3467}
Al Viro3b93f912014-02-11 21:34:08 -05003468EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003469
Jan Karae4dd9de2009-08-17 18:10:06 +02003470/**
Al Viro81742022014-04-03 03:17:43 -04003471 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02003472 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04003473 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02003474 *
3475 * This function does all the work needed for actually writing data to a
3476 * file. It does all basic checks, removes SUID from the file, updates
3477 * modification times and calls proper subroutines depending on whether we
3478 * do direct IO or a standard buffered write.
3479 *
3480 * It expects i_mutex to be grabbed unless we work on a block device or similar
3481 * object which does not need locking at all.
3482 *
3483 * This function does *not* take care of syncing data in case of O_SYNC write.
3484 * A caller has to handle it. This is mainly due to the fact that we want to
3485 * avoid syncing under i_mutex.
Mike Rapoporta862f682019-03-05 15:48:42 -08003486 *
3487 * Return:
3488 * * number of bytes written, even for truncated writes
3489 * * negative error code if no data has been written at all
Jan Karae4dd9de2009-08-17 18:10:06 +02003490 */
Al Viro81742022014-04-03 03:17:43 -04003491ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492{
3493 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003494 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003495 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05003496 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003497 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05003498 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01003501 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02003502 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503 if (err)
3504 goto out;
3505
Josef Bacikc3b2da32012-03-26 09:59:21 -04003506 err = file_update_time(file);
3507 if (err)
3508 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003509
Al Viro2ba48ce2015-04-09 13:52:01 -04003510 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04003511 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003512
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07003513 written = generic_file_direct_write(iocb, from);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08003514 /*
3515 * If the write stopped short of completing, fall back to
3516 * buffered writes. Some filesystems do this for writes to
3517 * holes, for example. For DAX files, a buffered write will
3518 * not succeed (even if it did, DAX does not handle dirty
3519 * page-cache pages correctly).
3520 */
Al Viro0b8def92015-04-07 10:22:53 -04003521 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003522 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05003523
Al Viro0b8def92015-04-07 10:22:53 -04003524 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003525 /*
Al Viro3b93f912014-02-11 21:34:08 -05003526 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003527 * then we want to return the number of bytes which were
3528 * direct-written, or the error code if that was zero. Note
3529 * that this differs from normal direct-io semantics, which
3530 * will return -EFOO even if some bytes were written.
3531 */
Al Viro60bb4522014-08-08 12:39:16 -04003532 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05003533 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003534 goto out;
3535 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003536 /*
3537 * We need to ensure that the page cache pages are written to
3538 * disk and invalidated to preserve the expected O_DIRECT
3539 * semantics.
3540 */
Al Viro3b93f912014-02-11 21:34:08 -05003541 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04003542 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003543 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04003544 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05003545 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003546 invalidate_mapping_pages(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003547 pos >> PAGE_SHIFT,
3548 endbyte >> PAGE_SHIFT);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003549 } else {
3550 /*
3551 * We don't know how much we wrote, so just return
3552 * the number of bytes which were direct-written
3553 */
3554 }
3555 } else {
Al Viro0b8def92015-04-07 10:22:53 -04003556 written = generic_perform_write(file, from, iocb->ki_pos);
3557 if (likely(written > 0))
3558 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003559 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560out:
3561 current->backing_dev_info = NULL;
3562 return written ? written : err;
3563}
Al Viro81742022014-04-03 03:17:43 -04003564EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565
Jan Karae4dd9de2009-08-17 18:10:06 +02003566/**
Al Viro81742022014-04-03 03:17:43 -04003567 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02003568 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04003569 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02003570 *
Al Viro81742022014-04-03 03:17:43 -04003571 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02003572 * filesystems. It takes care of syncing the file in case of O_SYNC file
3573 * and acquires i_mutex as needed.
Mike Rapoporta862f682019-03-05 15:48:42 -08003574 * Return:
3575 * * negative error code if no data has been written at all of
3576 * vfs_fsync_range() failed for a synchronous write
3577 * * number of bytes written, even for truncated writes
Jan Karae4dd9de2009-08-17 18:10:06 +02003578 */
Al Viro81742022014-04-03 03:17:43 -04003579ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003580{
3581 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02003582 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584
Al Viro59551022016-01-22 15:40:57 -05003585 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -04003586 ret = generic_write_checks(iocb, from);
3587 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04003588 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -05003589 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590
Christoph Hellwige2592212016-04-07 08:52:01 -07003591 if (ret > 0)
3592 ret = generic_write_sync(iocb, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003593 return ret;
3594}
Al Viro81742022014-04-03 03:17:43 -04003595EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003596
David Howellscf9a2ae2006-08-29 19:05:54 +01003597/**
3598 * try_to_release_page() - release old fs-specific metadata on a page
3599 *
3600 * @page: the page which the kernel is trying to free
3601 * @gfp_mask: memory allocation flags (and I/O mode)
3602 *
3603 * The address_space is to try to release any data against the page
Mike Rapoporta862f682019-03-05 15:48:42 -08003604 * (presumably at page->private).
David Howellscf9a2ae2006-08-29 19:05:54 +01003605 *
David Howells266cf652009-04-03 16:42:36 +01003606 * This may also be called if PG_fscache is set on a page, indicating that the
3607 * page is known to the local caching routines.
3608 *
David Howellscf9a2ae2006-08-29 19:05:54 +01003609 * The @gfp_mask argument specifies whether I/O may be performed to release
Mel Gorman71baba42015-11-06 16:28:28 -08003610 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01003611 *
Mike Rapoporta862f682019-03-05 15:48:42 -08003612 * Return: %1 if the release was successful, otherwise return zero.
David Howellscf9a2ae2006-08-29 19:05:54 +01003613 */
3614int try_to_release_page(struct page *page, gfp_t gfp_mask)
3615{
3616 struct address_space * const mapping = page->mapping;
3617
3618 BUG_ON(!PageLocked(page));
3619 if (PageWriteback(page))
3620 return 0;
3621
3622 if (mapping && mapping->a_ops->releasepage)
3623 return mapping->a_ops->releasepage(page, gfp_mask);
3624 return try_to_free_buffers(page);
3625}
3626
3627EXPORT_SYMBOL(try_to_release_page);