blob: dc3b78db079bf2ee555fdf9927cadb713d6d7e14 [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>
Nick Piggin0f8053a2006-03-22 00:08:33 -080044#include "internal.h"
45
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070046#define CREATE_TRACE_POINTS
47#include <trace/events/filemap.h>
48
Linus Torvalds1da177e2005-04-16 15:20:36 -070049/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 * FIXME: remove all knowledge of the buffer layer from the core VM
51 */
Jan Kara148f9482009-08-17 19:52:36 +020052#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <asm/mman.h>
55
56/*
57 * Shared mappings implemented 30.11.1994. It's not fully working yet,
58 * though.
59 *
60 * Shared mappings now work. 15.8.1995 Bruno.
61 *
62 * finished 'unifying' the page and buffer cache and SMP-threaded the
63 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
64 *
65 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
66 */
67
68/*
69 * Lock ordering:
70 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080071 * ->i_mmap_rwsem (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070073 * ->swap_lock (exclusive_swap_page, others)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070074 * ->i_pages lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080076 * ->i_mutex
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080077 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 *
79 * ->mmap_sem
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080080 * ->i_mmap_rwsem
Hugh Dickinsb8072f02005-10-29 18:16:41 -070081 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070082 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 *
84 * ->mmap_sem
85 * ->lock_page (access_process_vm)
86 *
Al Viroccad2362014-02-11 22:36:48 -050087 * ->i_mutex (generic_perform_write)
Nick Piggin82591e62006-10-19 23:29:10 -070088 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060090 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110091 * sb_lock (fs/fs-writeback.c)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -070092 * ->i_pages lock (__sync_single_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080094 * ->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * ->anon_vma.lock (vma_adjust)
96 *
97 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070098 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700100 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -0700101 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 * ->private_lock (try_to_unmap_one)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700103 * ->i_pages lock (try_to_unmap_one)
Andrey Ryabininf4b7e272019-03-05 15:49:39 -0800104 * ->pgdat->lru_lock (follow_page->mark_page_accessed)
105 * ->pgdat->lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 * ->private_lock (page_remove_rmap->set_page_dirty)
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700107 * ->i_pages lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600108 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100109 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700110 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600111 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100112 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
114 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800115 * ->i_mmap_rwsem
Andi Kleen9a3c5312012-03-21 16:34:09 -0700116 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 */
118
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500119static void page_cache_delete(struct address_space *mapping,
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700120 struct page *page, void *shadow)
121{
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500122 XA_STATE(xas, &mapping->i_pages, page->index);
123 unsigned int nr = 1;
Kirill A. Shutemovc70b6472016-12-12 16:43:17 -0800124
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500125 mapping_set_update(&xas, mapping);
126
127 /* hugetlb pages are represented by a single entry in the xarray */
128 if (!PageHuge(page)) {
129 xas_set_order(&xas, page->index, compound_order(page));
Matthew Wilcox (Oracle)d8c65462019-09-23 15:34:30 -0700130 nr = compound_nr(page);
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500131 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700132
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700133 VM_BUG_ON_PAGE(!PageLocked(page), page);
134 VM_BUG_ON_PAGE(PageTail(page), page);
135 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
Johannes Weiner449dd692014-04-03 14:47:56 -0700136
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500137 xas_store(&xas, shadow);
138 xas_init_marks(&xas);
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200139
Jan Kara23006382017-11-15 17:37:26 -0800140 page->mapping = NULL;
141 /* Leave page->index set: truncation lookup relies upon it */
142
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200143 if (shadow) {
144 mapping->nrexceptional += nr;
145 /*
146 * Make sure the nrexceptional update is committed before
147 * the nrpages update so that final truncate racing
148 * with reclaim does not see both counters 0 at the
149 * same time and miss a shadow entry.
150 */
151 smp_wmb();
152 }
153 mapping->nrpages -= nr;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700154}
155
Jan Kara5ecc4d82017-11-15 17:37:29 -0800156static void unaccount_page_cache_page(struct address_space *mapping,
157 struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158{
Jan Kara5ecc4d82017-11-15 17:37:29 -0800159 int nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600161 /*
162 * if we're uptodate, flush out into the cleancache, otherwise
163 * invalidate any existing cleancache entries. We can't leave
164 * stale data around in the cleancache once our page is gone
165 */
166 if (PageUptodate(page) && PageMappedToDisk(page))
167 cleancache_put_page(page);
168 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400169 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600170
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700171 VM_BUG_ON_PAGE(PageTail(page), page);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800172 VM_BUG_ON_PAGE(page_mapped(page), page);
173 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
174 int mapcount;
175
176 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
177 current->comm, page_to_pfn(page));
178 dump_page(page, "still mapped when deleted");
179 dump_stack();
180 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
181
182 mapcount = page_mapcount(page);
183 if (mapping_exiting(mapping) &&
184 page_count(page) >= mapcount + 2) {
185 /*
186 * All vmas have already been torn down, so it's
187 * a good bet that actually the page is unmapped,
188 * and we'd prefer not to leak it: if we're wrong,
189 * some other bad page check should catch it later.
190 */
191 page_mapcount_reset(page);
Joonsoo Kim6d061f92016-05-19 17:10:46 -0700192 page_ref_sub(page, mapcount);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800193 }
194 }
195
Jan Kara76253fb2017-11-15 17:37:22 -0800196 /* hugetlb pages do not participate in page cache accounting. */
Jan Kara5ecc4d82017-11-15 17:37:29 -0800197 if (PageHuge(page))
198 return;
Jan Kara76253fb2017-11-15 17:37:22 -0800199
Jan Kara5ecc4d82017-11-15 17:37:29 -0800200 nr = hpage_nr_pages(page);
201
202 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
203 if (PageSwapBacked(page)) {
204 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
205 if (PageTransHuge(page))
206 __dec_node_page_state(page, NR_SHMEM_THPS);
Song Liu99cb0db2019-09-23 15:38:00 -0700207 } else if (PageTransHuge(page)) {
208 __dec_node_page_state(page, NR_FILE_THPS);
Song Liu09d91cd2019-09-23 15:38:03 -0700209 filemap_nr_thps_dec(mapping);
Jan Kara76253fb2017-11-15 17:37:22 -0800210 }
Jan Kara5ecc4d82017-11-15 17:37:29 -0800211
212 /*
213 * At this point page must be either written or cleaned by
214 * truncate. Dirty page here signals a bug and loss of
215 * unwritten data.
216 *
217 * This fixes dirty accounting after removing the page entirely
218 * but leaves PageDirty set: it has no effect for truncated
219 * page and anyway will be cleared before returning page into
220 * buddy allocator.
221 */
222 if (WARN_ON_ONCE(PageDirty(page)))
223 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
224}
225
226/*
227 * Delete a page from the page cache and free it. Caller has to make
228 * sure the page is locked and that nobody else uses it - or that usage
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700229 * is safe. The caller must hold the i_pages lock.
Jan Kara5ecc4d82017-11-15 17:37:29 -0800230 */
231void __delete_from_page_cache(struct page *page, void *shadow)
232{
233 struct address_space *mapping = page->mapping;
234
235 trace_mm_filemap_delete_from_page_cache(page);
236
237 unaccount_page_cache_page(mapping, page);
Matthew Wilcox5c024e6a2017-11-21 09:17:59 -0500238 page_cache_delete(mapping, page, shadow);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
240
Jan Kara59c66c52017-11-15 17:37:18 -0800241static void page_cache_free_page(struct address_space *mapping,
242 struct page *page)
243{
244 void (*freepage)(struct page *);
245
246 freepage = mapping->a_ops->freepage;
247 if (freepage)
248 freepage(page);
249
250 if (PageTransHuge(page) && !PageHuge(page)) {
251 page_ref_sub(page, HPAGE_PMD_NR);
252 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
253 } else {
254 put_page(page);
255 }
256}
257
Minchan Kim702cfbf2011-03-22 16:32:43 -0700258/**
259 * delete_from_page_cache - delete page from page cache
260 * @page: the page which the kernel is trying to remove from page cache
261 *
262 * This must be called only on pages that have been verified to be in the page
263 * cache and locked. It will never put the page into the free list, the caller
264 * has a reference on the page.
265 */
266void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700268 struct address_space *mapping = page_mapping(page);
Greg Thelenc4843a72015-05-22 17:13:16 -0400269 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
Matt Mackallcd7619d2005-05-01 08:59:01 -0700271 BUG_ON(!PageLocked(page));
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700272 xa_lock_irqsave(&mapping->i_pages, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700273 __delete_from_page_cache(page, NULL);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700274 xa_unlock_irqrestore(&mapping->i_pages, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500275
Jan Kara59c66c52017-11-15 17:37:18 -0800276 page_cache_free_page(mapping, page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700277}
278EXPORT_SYMBOL(delete_from_page_cache);
279
Jan Karaaa65c292017-11-15 17:37:33 -0800280/*
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500281 * page_cache_delete_batch - delete several pages from page cache
Jan Karaaa65c292017-11-15 17:37:33 -0800282 * @mapping: the mapping to which pages belong
283 * @pvec: pagevec with pages to delete
284 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700285 * The function walks over mapping->i_pages and removes pages passed in @pvec
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700286 * from the mapping. The function expects @pvec to be sorted by page index
287 * and is optimised for it to be dense.
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700288 * It tolerates holes in @pvec (mapping entries at those indices are not
Jan Karaaa65c292017-11-15 17:37:33 -0800289 * modified). The function expects only THP head pages to be present in the
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700290 * @pvec.
Jan Karaaa65c292017-11-15 17:37:33 -0800291 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700292 * The function expects the i_pages lock to be held.
Jan Karaaa65c292017-11-15 17:37:33 -0800293 */
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500294static void page_cache_delete_batch(struct address_space *mapping,
Jan Karaaa65c292017-11-15 17:37:33 -0800295 struct pagevec *pvec)
296{
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500297 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
Jan Karaaa65c292017-11-15 17:37:33 -0800298 int total_pages = 0;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700299 int i = 0;
Jan Karaaa65c292017-11-15 17:37:33 -0800300 struct page *page;
Jan Karaaa65c292017-11-15 17:37:33 -0800301
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500302 mapping_set_update(&xas, mapping);
303 xas_for_each(&xas, page, ULONG_MAX) {
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700304 if (i >= pagevec_count(pvec))
Jan Karaaa65c292017-11-15 17:37:33 -0800305 break;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700306
307 /* A swap/dax/shadow entry got inserted? Skip it. */
Matthew Wilcox3159f942017-11-03 13:30:42 -0400308 if (xa_is_value(page))
Jan Karaaa65c292017-11-15 17:37:33 -0800309 continue;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700310 /*
311 * A page got inserted in our range? Skip it. We have our
312 * pages locked so they are protected from being removed.
313 * If we see a page whose index is higher than ours, it
314 * means our page has been removed, which shouldn't be
315 * possible because we're holding the PageLock.
316 */
317 if (page != pvec->pages[i]) {
318 VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
319 page);
320 continue;
Linus Torvalds69bf4b62019-07-05 19:55:18 -0700321 }
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700322
323 WARN_ON_ONCE(!PageLocked(page));
324
325 if (page->index == xas.xa_index)
326 page->mapping = NULL;
327 /* Leave page->index set: truncation lookup relies on it */
328
329 /*
330 * Move to the next page in the vector if this is a regular
331 * page or the index is of the last sub-page of this compound
332 * page.
333 */
334 if (page->index + compound_nr(page) - 1 == xas.xa_index)
335 i++;
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500336 xas_store(&xas, NULL);
Jan Karaaa65c292017-11-15 17:37:33 -0800337 total_pages++;
338 }
339 mapping->nrpages -= total_pages;
340}
341
342void delete_from_page_cache_batch(struct address_space *mapping,
343 struct pagevec *pvec)
344{
345 int i;
346 unsigned long flags;
347
348 if (!pagevec_count(pvec))
349 return;
350
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700351 xa_lock_irqsave(&mapping->i_pages, flags);
Jan Karaaa65c292017-11-15 17:37:33 -0800352 for (i = 0; i < pagevec_count(pvec); i++) {
353 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
354
355 unaccount_page_cache_page(mapping, pvec->pages[i]);
356 }
Matthew Wilcoxef8e5712017-12-04 03:59:45 -0500357 page_cache_delete_batch(mapping, pvec);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700358 xa_unlock_irqrestore(&mapping->i_pages, flags);
Jan Karaaa65c292017-11-15 17:37:33 -0800359
360 for (i = 0; i < pagevec_count(pvec); i++)
361 page_cache_free_page(mapping, pvec->pages[i]);
362}
363
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200364int filemap_check_errors(struct address_space *mapping)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700365{
366 int ret = 0;
367 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700368 if (test_bit(AS_ENOSPC, &mapping->flags) &&
369 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700370 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700371 if (test_bit(AS_EIO, &mapping->flags) &&
372 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700373 ret = -EIO;
374 return ret;
375}
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200376EXPORT_SYMBOL(filemap_check_errors);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700377
Jeff Layton76341ca2017-07-06 07:02:22 -0400378static int filemap_check_and_keep_errors(struct address_space *mapping)
379{
380 /* Check for outstanding write errors */
381 if (test_bit(AS_EIO, &mapping->flags))
382 return -EIO;
383 if (test_bit(AS_ENOSPC, &mapping->flags))
384 return -ENOSPC;
385 return 0;
386}
387
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700389 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700390 * @mapping: address space structure to write
391 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800392 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700393 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700395 * Start writeback against all of a mapping's dirty pages that lie
396 * within the byte offsets <start, end> inclusive.
397 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700399 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 * these two operations is that if a dirty page/buffer is encountered, it must
401 * be waited upon, and not just skipped over.
Mike Rapoporta862f682019-03-05 15:48:42 -0800402 *
403 * Return: %0 on success, negative error code otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800405int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
406 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
408 int ret;
409 struct writeback_control wbc = {
410 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800411 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700412 .range_start = start,
413 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 };
415
Konstantin Khlebnikovc3aab9a02019-09-23 15:34:45 -0700416 if (!mapping_cap_writeback_dirty(mapping) ||
417 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 return 0;
419
Tejun Heob16b1de2015-06-02 08:39:48 -0600420 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 ret = do_writepages(mapping, &wbc);
Tejun Heob16b1de2015-06-02 08:39:48 -0600422 wbc_detach_inode(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 return ret;
424}
425
426static inline int __filemap_fdatawrite(struct address_space *mapping,
427 int sync_mode)
428{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700429 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430}
431
432int filemap_fdatawrite(struct address_space *mapping)
433{
434 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
435}
436EXPORT_SYMBOL(filemap_fdatawrite);
437
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400438int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800439 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440{
441 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
442}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400443EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444
Randy Dunlap485bb992006-06-23 02:03:49 -0700445/**
446 * filemap_flush - mostly a non-blocking flush
447 * @mapping: target address_space
448 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 * This is a mostly non-blocking flush. Not suitable for data-integrity
450 * purposes - I/O may not be started against all dirty pages.
Mike Rapoporta862f682019-03-05 15:48:42 -0800451 *
452 * Return: %0 on success, negative error code otherwise.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 */
454int filemap_flush(struct address_space *mapping)
455{
456 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
457}
458EXPORT_SYMBOL(filemap_flush);
459
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500460/**
461 * filemap_range_has_page - check if a page exists in range.
462 * @mapping: address space within which to check
463 * @start_byte: offset in bytes where the range starts
464 * @end_byte: offset in bytes where the range ends (inclusive)
465 *
466 * Find at least one page in the range supplied, usually used to check if
467 * direct writing in this range will trigger a writeback.
Mike Rapoporta862f682019-03-05 15:48:42 -0800468 *
469 * Return: %true if at least one page exists in the specified range,
470 * %false otherwise.
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500471 */
472bool filemap_range_has_page(struct address_space *mapping,
473 loff_t start_byte, loff_t end_byte)
474{
Jan Karaf7b68042017-09-06 16:21:40 -0700475 struct page *page;
Matthew Wilcox8fa8e532018-01-16 06:26:49 -0500476 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
477 pgoff_t max = end_byte >> PAGE_SHIFT;
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500478
479 if (end_byte < start_byte)
480 return false;
481
Matthew Wilcox8fa8e532018-01-16 06:26:49 -0500482 rcu_read_lock();
483 for (;;) {
484 page = xas_find(&xas, max);
485 if (xas_retry(&xas, page))
486 continue;
487 /* Shadow entries don't count */
488 if (xa_is_value(page))
489 continue;
490 /*
491 * We don't need to try to pin this page; we're about to
492 * release the RCU lock anyway. It is enough to know that
493 * there was a page here recently.
494 */
495 break;
496 }
497 rcu_read_unlock();
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500498
Matthew Wilcox8fa8e532018-01-16 06:26:49 -0500499 return page != NULL;
Goldwyn Rodrigues7fc9e472017-06-20 07:05:41 -0500500}
501EXPORT_SYMBOL(filemap_range_has_page);
502
Jeff Layton5e8fcc12017-07-06 07:02:24 -0400503static void __filemap_fdatawait_range(struct address_space *mapping,
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800504 loff_t start_byte, loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300506 pgoff_t index = start_byte >> PAGE_SHIFT;
507 pgoff_t end = end_byte >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 struct pagevec pvec;
509 int nr_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200511 if (end_byte < start_byte)
Jeff Layton5e8fcc12017-07-06 07:02:24 -0400512 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Mel Gorman86679822017-11-15 17:37:52 -0800514 pagevec_init(&pvec);
Jan Kara312e9d22017-11-15 17:35:05 -0800515 while (index <= end) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 unsigned i;
517
Jan Kara312e9d22017-11-15 17:35:05 -0800518 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
Jan Kara67fd7072017-11-15 17:35:19 -0800519 end, PAGECACHE_TAG_WRITEBACK);
Jan Kara312e9d22017-11-15 17:35:05 -0800520 if (!nr_pages)
521 break;
522
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 for (i = 0; i < nr_pages; i++) {
524 struct page *page = pvec.pages[i];
525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 wait_on_page_writeback(page);
Jeff Layton5e8fcc12017-07-06 07:02:24 -0400527 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 }
529 pagevec_release(&pvec);
530 cond_resched();
531 }
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800532}
533
534/**
535 * filemap_fdatawait_range - wait for writeback to complete
536 * @mapping: address space structure to wait for
537 * @start_byte: offset in bytes where the range starts
538 * @end_byte: offset in bytes where the range ends (inclusive)
539 *
540 * Walk the list of under-writeback pages of the given address space
541 * in the given range and wait for all of them. Check error status of
542 * the address space and return it.
543 *
544 * Since the error status of the address space is cleared by this function,
545 * callers are responsible for checking the return value and handling and/or
546 * reporting the error.
Mike Rapoporta862f682019-03-05 15:48:42 -0800547 *
548 * Return: error status of the address space.
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800549 */
550int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
551 loff_t end_byte)
552{
Jeff Layton5e8fcc12017-07-06 07:02:24 -0400553 __filemap_fdatawait_range(mapping, start_byte, end_byte);
554 return filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200556EXPORT_SYMBOL(filemap_fdatawait_range);
557
558/**
Ross Zwisleraa0bfcd2019-06-20 17:05:37 -0400559 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
560 * @mapping: address space structure to wait for
561 * @start_byte: offset in bytes where the range starts
562 * @end_byte: offset in bytes where the range ends (inclusive)
563 *
564 * Walk the list of under-writeback pages of the given address space in the
565 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
566 * this function does not clear error status of the address space.
567 *
568 * Use this function if callers don't handle errors themselves. Expected
569 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
570 * fsfreeze(8)
571 */
572int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
573 loff_t start_byte, loff_t end_byte)
574{
575 __filemap_fdatawait_range(mapping, start_byte, end_byte);
576 return filemap_check_and_keep_errors(mapping);
577}
578EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
579
580/**
Jeff Laytona823e452017-07-28 07:24:43 -0400581 * file_fdatawait_range - wait for writeback to complete
582 * @file: file pointing to address space structure to wait for
583 * @start_byte: offset in bytes where the range starts
584 * @end_byte: offset in bytes where the range ends (inclusive)
585 *
586 * Walk the list of under-writeback pages of the address space that file
587 * refers to, in the given range and wait for all of them. Check error
588 * status of the address space vs. the file->f_wb_err cursor and return it.
589 *
590 * Since the error status of the file is advanced by this function,
591 * callers are responsible for checking the return value and handling and/or
592 * reporting the error.
Mike Rapoporta862f682019-03-05 15:48:42 -0800593 *
594 * Return: error status of the address space vs. the file->f_wb_err cursor.
Jeff Laytona823e452017-07-28 07:24:43 -0400595 */
596int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
597{
598 struct address_space *mapping = file->f_mapping;
599
600 __filemap_fdatawait_range(mapping, start_byte, end_byte);
601 return file_check_and_advance_wb_err(file);
602}
603EXPORT_SYMBOL(file_fdatawait_range);
604
605/**
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800606 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
607 * @mapping: address space structure to wait for
608 *
609 * Walk the list of under-writeback pages of the given address space
610 * and wait for all of them. Unlike filemap_fdatawait(), this function
611 * does not clear error status of the address space.
612 *
613 * Use this function if callers don't handle errors themselves. Expected
614 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
615 * fsfreeze(8)
Mike Rapoporta862f682019-03-05 15:48:42 -0800616 *
617 * Return: error status of the address space.
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800618 */
Jeff Layton76341ca2017-07-06 07:02:22 -0400619int filemap_fdatawait_keep_errors(struct address_space *mapping)
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800620{
Jeff Laytonffb959b2017-07-31 10:29:38 -0400621 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
Jeff Layton76341ca2017-07-06 07:02:22 -0400622 return filemap_check_and_keep_errors(mapping);
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800623}
Jeff Layton76341ca2017-07-06 07:02:22 -0400624EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800625
Konstantin Khlebnikov875d91b2019-09-23 15:34:48 -0700626/* Returns true if writeback might be needed or already in progress. */
Jeff Layton9326c9b2017-07-26 10:21:11 -0400627static bool mapping_needs_writeback(struct address_space *mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
Konstantin Khlebnikov875d91b2019-09-23 15:34:48 -0700629 if (dax_mapping(mapping))
630 return mapping->nrexceptional;
631
632 return mapping->nrpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
635int filemap_write_and_wait(struct address_space *mapping)
636{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800637 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
Jeff Layton9326c9b2017-07-26 10:21:11 -0400639 if (mapping_needs_writeback(mapping)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800640 err = filemap_fdatawrite(mapping);
641 /*
642 * Even if the above returned error, the pages may be
643 * written partially (e.g. -ENOSPC), so we wait for it.
644 * But the -EIO is special case, it may indicate the worst
645 * thing (e.g. bug) happened, so we avoid waiting for it.
646 */
647 if (err != -EIO) {
648 int err2 = filemap_fdatawait(mapping);
649 if (!err)
650 err = err2;
Jeff Laytoncbeaf952017-07-06 07:02:23 -0400651 } else {
652 /* Clear any previously stored errors */
653 filemap_check_errors(mapping);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800654 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700655 } else {
656 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800658 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800660EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Randy Dunlap485bb992006-06-23 02:03:49 -0700662/**
663 * filemap_write_and_wait_range - write out & wait on a file range
664 * @mapping: the address_space for the pages
665 * @lstart: offset in bytes where the range starts
666 * @lend: offset in bytes where the range ends (inclusive)
667 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800668 * Write out and wait upon file offsets lstart->lend, inclusive.
669 *
mchehab@s-opensource.com0e056eb2017-03-30 17:11:36 -0300670 * Note that @lend is inclusive (describes the last byte to be written) so
Andrew Morton469eb4d2006-03-24 03:17:45 -0800671 * that this function can be used to write to the very end-of-file (end = -1).
Mike Rapoporta862f682019-03-05 15:48:42 -0800672 *
673 * Return: error status of the address space.
Andrew Morton469eb4d2006-03-24 03:17:45 -0800674 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675int filemap_write_and_wait_range(struct address_space *mapping,
676 loff_t lstart, loff_t lend)
677{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800678 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Jeff Layton9326c9b2017-07-26 10:21:11 -0400680 if (mapping_needs_writeback(mapping)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800681 err = __filemap_fdatawrite_range(mapping, lstart, lend,
682 WB_SYNC_ALL);
683 /* See comment of filemap_write_and_wait() */
684 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200685 int err2 = filemap_fdatawait_range(mapping,
686 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800687 if (!err)
688 err = err2;
Jeff Laytoncbeaf952017-07-06 07:02:23 -0400689 } else {
690 /* Clear any previously stored errors */
691 filemap_check_errors(mapping);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800692 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700693 } else {
694 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800696 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697}
Chris Masonf6995582009-04-15 13:22:37 -0400698EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Jeff Layton5660e132017-07-06 07:02:25 -0400700void __filemap_set_wb_err(struct address_space *mapping, int err)
701{
Jeff Layton3acdfd22017-07-24 06:22:15 -0400702 errseq_t eseq = errseq_set(&mapping->wb_err, err);
Jeff Layton5660e132017-07-06 07:02:25 -0400703
704 trace_filemap_set_wb_err(mapping, eseq);
705}
706EXPORT_SYMBOL(__filemap_set_wb_err);
707
708/**
709 * file_check_and_advance_wb_err - report wb error (if any) that was previously
710 * and advance wb_err to current one
711 * @file: struct file on which the error is being reported
712 *
713 * When userland calls fsync (or something like nfsd does the equivalent), we
714 * want to report any writeback errors that occurred since the last fsync (or
715 * since the file was opened if there haven't been any).
716 *
717 * Grab the wb_err from the mapping. If it matches what we have in the file,
718 * then just quickly return 0. The file is all caught up.
719 *
720 * If it doesn't match, then take the mapping value, set the "seen" flag in
721 * it and try to swap it into place. If it works, or another task beat us
722 * to it with the new value, then update the f_wb_err and return the error
723 * portion. The error at this point must be reported via proper channels
724 * (a'la fsync, or NFS COMMIT operation, etc.).
725 *
726 * While we handle mapping->wb_err with atomic operations, the f_wb_err
727 * value is protected by the f_lock since we must ensure that it reflects
728 * the latest value swapped in for this file descriptor.
Mike Rapoporta862f682019-03-05 15:48:42 -0800729 *
730 * Return: %0 on success, negative error code otherwise.
Jeff Layton5660e132017-07-06 07:02:25 -0400731 */
732int file_check_and_advance_wb_err(struct file *file)
733{
734 int err = 0;
735 errseq_t old = READ_ONCE(file->f_wb_err);
736 struct address_space *mapping = file->f_mapping;
737
738 /* Locklessly handle the common case where nothing has changed */
739 if (errseq_check(&mapping->wb_err, old)) {
740 /* Something changed, must use slow path */
741 spin_lock(&file->f_lock);
742 old = file->f_wb_err;
743 err = errseq_check_and_advance(&mapping->wb_err,
744 &file->f_wb_err);
745 trace_file_check_and_advance_wb_err(file, old);
746 spin_unlock(&file->f_lock);
747 }
Jeff Laytonf4e222c2017-10-03 16:15:25 -0700748
749 /*
750 * We're mostly using this function as a drop in replacement for
751 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
752 * that the legacy code would have had on these flags.
753 */
754 clear_bit(AS_EIO, &mapping->flags);
755 clear_bit(AS_ENOSPC, &mapping->flags);
Jeff Layton5660e132017-07-06 07:02:25 -0400756 return err;
757}
758EXPORT_SYMBOL(file_check_and_advance_wb_err);
759
760/**
761 * file_write_and_wait_range - write out & wait on a file range
762 * @file: file pointing to address_space with pages
763 * @lstart: offset in bytes where the range starts
764 * @lend: offset in bytes where the range ends (inclusive)
765 *
766 * Write out and wait upon file offsets lstart->lend, inclusive.
767 *
768 * Note that @lend is inclusive (describes the last byte to be written) so
769 * that this function can be used to write to the very end-of-file (end = -1).
770 *
771 * After writing out and waiting on the data, we check and advance the
772 * f_wb_err cursor to the latest value, and return any errors detected there.
Mike Rapoporta862f682019-03-05 15:48:42 -0800773 *
774 * Return: %0 on success, negative error code otherwise.
Jeff Layton5660e132017-07-06 07:02:25 -0400775 */
776int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
777{
778 int err = 0, err2;
779 struct address_space *mapping = file->f_mapping;
780
Jeff Layton9326c9b2017-07-26 10:21:11 -0400781 if (mapping_needs_writeback(mapping)) {
Jeff Layton5660e132017-07-06 07:02:25 -0400782 err = __filemap_fdatawrite_range(mapping, lstart, lend,
783 WB_SYNC_ALL);
784 /* See comment of filemap_write_and_wait() */
785 if (err != -EIO)
786 __filemap_fdatawait_range(mapping, lstart, lend);
787 }
788 err2 = file_check_and_advance_wb_err(file);
789 if (!err)
790 err = err2;
791 return err;
792}
793EXPORT_SYMBOL(file_write_and_wait_range);
794
Randy Dunlap485bb992006-06-23 02:03:49 -0700795/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700796 * replace_page_cache_page - replace a pagecache page with a new one
797 * @old: page to be replaced
798 * @new: page to replace with
799 * @gfp_mask: allocation mode
800 *
801 * This function replaces a page in the pagecache with a new one. On
802 * success it acquires the pagecache reference for the new page and
803 * drops it for the old page. Both the old and new pages must be
804 * locked. This function does not add the new page to the LRU, the
805 * caller must do that.
806 *
Matthew Wilcox74d60952017-11-17 10:01:45 -0500807 * The remove + add is atomic. This function cannot fail.
Mike Rapoporta862f682019-03-05 15:48:42 -0800808 *
809 * Return: %0
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700810 */
811int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
812{
Matthew Wilcox74d60952017-11-17 10:01:45 -0500813 struct address_space *mapping = old->mapping;
814 void (*freepage)(struct page *) = mapping->a_ops->freepage;
815 pgoff_t offset = old->index;
816 XA_STATE(xas, &mapping->i_pages, offset);
817 unsigned long flags;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700818
Sasha Levin309381fea2014-01-23 15:52:54 -0800819 VM_BUG_ON_PAGE(!PageLocked(old), old);
820 VM_BUG_ON_PAGE(!PageLocked(new), new);
821 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700822
Matthew Wilcox74d60952017-11-17 10:01:45 -0500823 get_page(new);
824 new->mapping = mapping;
825 new->index = offset;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700826
Matthew Wilcox74d60952017-11-17 10:01:45 -0500827 xas_lock_irqsave(&xas, flags);
828 xas_store(&xas, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700829
Matthew Wilcox74d60952017-11-17 10:01:45 -0500830 old->mapping = NULL;
831 /* hugetlb pages do not participate in page cache accounting. */
832 if (!PageHuge(old))
833 __dec_node_page_state(new, NR_FILE_PAGES);
834 if (!PageHuge(new))
835 __inc_node_page_state(new, NR_FILE_PAGES);
836 if (PageSwapBacked(old))
837 __dec_node_page_state(new, NR_SHMEM);
838 if (PageSwapBacked(new))
839 __inc_node_page_state(new, NR_SHMEM);
840 xas_unlock_irqrestore(&xas, flags);
841 mem_cgroup_migrate(old, new);
842 if (freepage)
843 freepage(old);
844 put_page(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700845
Matthew Wilcox74d60952017-11-17 10:01:45 -0500846 return 0;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700847}
848EXPORT_SYMBOL_GPL(replace_page_cache_page);
849
Johannes Weinera5289102014-04-03 14:47:51 -0700850static int __add_to_page_cache_locked(struct page *page,
851 struct address_space *mapping,
852 pgoff_t offset, gfp_t gfp_mask,
853 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854{
Matthew Wilcox74d60952017-11-17 10:01:45 -0500855 XA_STATE(xas, &mapping->i_pages, offset);
Johannes Weiner00501b52014-08-08 14:19:20 -0700856 int huge = PageHuge(page);
857 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700858 int error;
Matthew Wilcox74d60952017-11-17 10:01:45 -0500859 void *old;
Nick Piggine2867812008-07-25 19:45:30 -0700860
Sasha Levin309381fea2014-01-23 15:52:54 -0800861 VM_BUG_ON_PAGE(!PageLocked(page), page);
862 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500863 mapping_set_update(&xas, mapping);
Nick Piggine2867812008-07-25 19:45:30 -0700864
Johannes Weiner00501b52014-08-08 14:19:20 -0700865 if (!huge) {
866 error = mem_cgroup_try_charge(page, current->mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800867 gfp_mask, &memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -0700868 if (error)
869 return error;
870 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300872 get_page(page);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700873 page->mapping = mapping;
874 page->index = offset;
875
Matthew Wilcox74d60952017-11-17 10:01:45 -0500876 do {
877 xas_lock_irq(&xas);
878 old = xas_load(&xas);
879 if (old && !xa_is_value(old))
880 xas_set_err(&xas, -EEXIST);
881 xas_store(&xas, page);
882 if (xas_error(&xas))
883 goto unlock;
Michal Hocko4165b9b42015-06-24 16:57:24 -0700884
Matthew Wilcox74d60952017-11-17 10:01:45 -0500885 if (xa_is_value(old)) {
886 mapping->nrexceptional--;
887 if (shadowp)
888 *shadowp = old;
889 }
890 mapping->nrpages++;
891
892 /* hugetlb pages do not participate in page cache accounting */
893 if (!huge)
894 __inc_node_page_state(page, NR_FILE_PAGES);
895unlock:
896 xas_unlock_irq(&xas);
897 } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
898
899 if (xas_error(&xas))
900 goto error;
901
Johannes Weiner00501b52014-08-08 14:19:20 -0700902 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800903 mem_cgroup_commit_charge(page, memcg, false, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700904 trace_mm_filemap_add_to_page_cache(page);
905 return 0;
Matthew Wilcox74d60952017-11-17 10:01:45 -0500906error:
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700907 page->mapping = NULL;
908 /* Leave page->index set: truncation relies upon it */
Johannes Weiner00501b52014-08-08 14:19:20 -0700909 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800910 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300911 put_page(page);
Matthew Wilcox74d60952017-11-17 10:01:45 -0500912 return xas_error(&xas);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913}
Josef Bacikcfcbfb12019-05-13 17:21:04 -0700914ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
Johannes Weinera5289102014-04-03 14:47:51 -0700915
916/**
917 * add_to_page_cache_locked - add a locked page to the pagecache
918 * @page: page to add
919 * @mapping: the page's address_space
920 * @offset: page index
921 * @gfp_mask: page allocation mode
922 *
923 * This function is used to add a page to the pagecache. It must be locked.
924 * This function does not add the page to the LRU. The caller must do that.
Mike Rapoporta862f682019-03-05 15:48:42 -0800925 *
926 * Return: %0 on success, negative error code otherwise.
Johannes Weinera5289102014-04-03 14:47:51 -0700927 */
928int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
929 pgoff_t offset, gfp_t gfp_mask)
930{
931 return __add_to_page_cache_locked(page, mapping, offset,
932 gfp_mask, NULL);
933}
Nick Piggine2867812008-07-25 19:45:30 -0700934EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
936int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400937 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
Johannes Weinera5289102014-04-03 14:47:51 -0700939 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700940 int ret;
941
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800942 __SetPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700943 ret = __add_to_page_cache_locked(page, mapping, offset,
944 gfp_mask, &shadow);
945 if (unlikely(ret))
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800946 __ClearPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700947 else {
948 /*
949 * The page might have been evicted from cache only
950 * recently, in which case it should be activated like
951 * any other repeatedly accessed page.
Rik van Rielf0281a02016-05-20 16:56:25 -0700952 * The exception is pages getting rewritten; evicting other
953 * data from the working set, only to cache data that will
954 * get overwritten with something else, is a waste of memory.
Johannes Weinera5289102014-04-03 14:47:51 -0700955 */
Johannes Weiner1899ad12018-10-26 15:06:04 -0700956 WARN_ON_ONCE(PageActive(page));
957 if (!(gfp_mask & __GFP_WRITE) && shadow)
958 workingset_refault(page, shadow);
Johannes Weinera5289102014-04-03 14:47:51 -0700959 lru_cache_add(page);
960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 return ret;
962}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300963EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
Paul Jackson44110fe2006-03-24 03:16:04 -0800965#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700966struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800967{
Miao Xiec0ff7452010-05-24 14:32:08 -0700968 int n;
969 struct page *page;
970
Paul Jackson44110fe2006-03-24 03:16:04 -0800971 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700972 unsigned int cpuset_mems_cookie;
973 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700974 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700975 n = cpuset_mem_spread_node();
Vlastimil Babka96db8002015-09-08 15:03:50 -0700976 page = __alloc_pages_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700977 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700978
Miao Xiec0ff7452010-05-24 14:32:08 -0700979 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800980 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700981 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800982}
Nick Piggin2ae88142006-10-28 10:38:23 -0700983EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800984#endif
985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986/*
987 * In order to wait for pages to become available there must be
988 * waitqueues associated with pages. By using a hash table of
989 * waitqueues where the bucket discipline is to maintain all
990 * waiters on the same queue and wake all when any of the pages
991 * become available, and for the woken contexts to check to be
992 * sure the appropriate page became available, this saves space
993 * at a cost of "thundering herd" phenomena during rare hash
994 * collisions.
995 */
Nicholas Piggin62906022016-12-25 13:00:30 +1000996#define PAGE_WAIT_TABLE_BITS 8
997#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
998static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
999
1000static wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001{
Nicholas Piggin62906022016-12-25 13:00:30 +10001002 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003}
Nicholas Piggin62906022016-12-25 13:00:30 +10001004
1005void __init pagecache_init(void)
1006{
1007 int i;
1008
1009 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1010 init_waitqueue_head(&page_wait_table[i]);
1011
1012 page_writeback_init();
1013}
1014
Linus Torvalds3510ca22017-08-27 13:55:12 -07001015/* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
Nicholas Piggin62906022016-12-25 13:00:30 +10001016struct wait_page_key {
1017 struct page *page;
1018 int bit_nr;
1019 int page_match;
1020};
1021
1022struct wait_page_queue {
1023 struct page *page;
1024 int bit_nr;
Ingo Molnarac6424b2017-06-20 12:06:13 +02001025 wait_queue_entry_t wait;
Nicholas Piggin62906022016-12-25 13:00:30 +10001026};
1027
Ingo Molnarac6424b2017-06-20 12:06:13 +02001028static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
Nicholas Piggin62906022016-12-25 13:00:30 +10001029{
1030 struct wait_page_key *key = arg;
1031 struct wait_page_queue *wait_page
1032 = container_of(wait, struct wait_page_queue, wait);
1033
1034 if (wait_page->page != key->page)
1035 return 0;
1036 key->page_match = 1;
1037
1038 if (wait_page->bit_nr != key->bit_nr)
1039 return 0;
Linus Torvalds3510ca22017-08-27 13:55:12 -07001040
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001041 /*
1042 * Stop walking if it's locked.
1043 * Is this safe if put_and_wait_on_page_locked() is in use?
1044 * Yes: the waker must hold a reference to this page, and if PG_locked
1045 * has now already been set by another task, that task must also hold
1046 * a reference to the *same usage* of this page; so there is no need
1047 * to walk on to wake even the put_and_wait_on_page_locked() callers.
1048 */
Nicholas Piggin62906022016-12-25 13:00:30 +10001049 if (test_bit(key->bit_nr, &key->page->flags))
Linus Torvalds3510ca22017-08-27 13:55:12 -07001050 return -1;
Nicholas Piggin62906022016-12-25 13:00:30 +10001051
1052 return autoremove_wake_function(wait, mode, sync, key);
1053}
1054
Nicholas Piggin74d81bf2017-02-22 15:44:41 -08001055static void wake_up_page_bit(struct page *page, int bit_nr)
Nicholas Piggin62906022016-12-25 13:00:30 +10001056{
1057 wait_queue_head_t *q = page_waitqueue(page);
1058 struct wait_page_key key;
1059 unsigned long flags;
Tim Chen11a19c72017-08-25 09:13:55 -07001060 wait_queue_entry_t bookmark;
Nicholas Piggin62906022016-12-25 13:00:30 +10001061
1062 key.page = page;
1063 key.bit_nr = bit_nr;
1064 key.page_match = 0;
1065
Tim Chen11a19c72017-08-25 09:13:55 -07001066 bookmark.flags = 0;
1067 bookmark.private = NULL;
1068 bookmark.func = NULL;
1069 INIT_LIST_HEAD(&bookmark.entry);
1070
Nicholas Piggin62906022016-12-25 13:00:30 +10001071 spin_lock_irqsave(&q->lock, flags);
Tim Chen11a19c72017-08-25 09:13:55 -07001072 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1073
1074 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1075 /*
1076 * Take a breather from holding the lock,
1077 * allow pages that finish wake up asynchronously
1078 * to acquire the lock and remove themselves
1079 * from wait queue
1080 */
1081 spin_unlock_irqrestore(&q->lock, flags);
1082 cpu_relax();
1083 spin_lock_irqsave(&q->lock, flags);
1084 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1085 }
1086
Nicholas Piggin62906022016-12-25 13:00:30 +10001087 /*
1088 * It is possible for other pages to have collided on the waitqueue
1089 * hash, so in that case check for a page match. That prevents a long-
1090 * term waiter
1091 *
1092 * It is still possible to miss a case here, when we woke page waiters
1093 * and removed them from the waitqueue, but there are still other
1094 * page waiters.
1095 */
1096 if (!waitqueue_active(q) || !key.page_match) {
1097 ClearPageWaiters(page);
1098 /*
1099 * It's possible to miss clearing Waiters here, when we woke
1100 * our page waiters, but the hashed waitqueue has waiters for
1101 * other pages on it.
1102 *
1103 * That's okay, it's a rare case. The next waker will clear it.
1104 */
1105 }
1106 spin_unlock_irqrestore(&q->lock, flags);
1107}
Nicholas Piggin74d81bf2017-02-22 15:44:41 -08001108
1109static void wake_up_page(struct page *page, int bit)
1110{
1111 if (!PageWaiters(page))
1112 return;
1113 wake_up_page_bit(page, bit);
1114}
Nicholas Piggin62906022016-12-25 13:00:30 +10001115
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001116/*
1117 * A choice of three behaviors for wait_on_page_bit_common():
1118 */
1119enum behavior {
1120 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1121 * __lock_page() waiting on then setting PG_locked.
1122 */
1123 SHARED, /* Hold ref to page and check the bit when woken, like
1124 * wait_on_page_writeback() waiting on PG_writeback.
1125 */
1126 DROP, /* Drop ref to page before wait, no check when woken,
1127 * like put_and_wait_on_page_locked() on PG_locked.
1128 */
1129};
1130
Nicholas Piggin62906022016-12-25 13:00:30 +10001131static inline int wait_on_page_bit_common(wait_queue_head_t *q,
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001132 struct page *page, int bit_nr, int state, enum behavior behavior)
Nicholas Piggin62906022016-12-25 13:00:30 +10001133{
1134 struct wait_page_queue wait_page;
Ingo Molnarac6424b2017-06-20 12:06:13 +02001135 wait_queue_entry_t *wait = &wait_page.wait;
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001136 bool bit_is_set;
Johannes Weinerb1d29ba2018-10-26 15:06:08 -07001137 bool thrashing = false;
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001138 bool delayacct = false;
Johannes Weinereb414682018-10-26 15:06:27 -07001139 unsigned long pflags;
Nicholas Piggin62906022016-12-25 13:00:30 +10001140 int ret = 0;
1141
Johannes Weinereb414682018-10-26 15:06:27 -07001142 if (bit_nr == PG_locked &&
Johannes Weinerb1d29ba2018-10-26 15:06:08 -07001143 !PageUptodate(page) && PageWorkingset(page)) {
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001144 if (!PageSwapBacked(page)) {
Johannes Weinereb414682018-10-26 15:06:27 -07001145 delayacct_thrashing_start();
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001146 delayacct = true;
1147 }
Johannes Weinereb414682018-10-26 15:06:27 -07001148 psi_memstall_enter(&pflags);
Johannes Weinerb1d29ba2018-10-26 15:06:08 -07001149 thrashing = true;
1150 }
1151
Nicholas Piggin62906022016-12-25 13:00:30 +10001152 init_wait(wait);
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001153 wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0;
Nicholas Piggin62906022016-12-25 13:00:30 +10001154 wait->func = wake_page_function;
1155 wait_page.page = page;
1156 wait_page.bit_nr = bit_nr;
1157
1158 for (;;) {
1159 spin_lock_irq(&q->lock);
1160
Ingo Molnar2055da92017-06-20 12:06:46 +02001161 if (likely(list_empty(&wait->entry))) {
Linus Torvalds3510ca22017-08-27 13:55:12 -07001162 __add_wait_queue_entry_tail(q, wait);
Nicholas Piggin62906022016-12-25 13:00:30 +10001163 SetPageWaiters(page);
1164 }
1165
1166 set_current_state(state);
1167
1168 spin_unlock_irq(&q->lock);
1169
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001170 bit_is_set = test_bit(bit_nr, &page->flags);
1171 if (behavior == DROP)
1172 put_page(page);
Nicholas Piggin62906022016-12-25 13:00:30 +10001173
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001174 if (likely(bit_is_set))
1175 io_schedule();
1176
1177 if (behavior == EXCLUSIVE) {
Nicholas Piggin62906022016-12-25 13:00:30 +10001178 if (!test_and_set_bit_lock(bit_nr, &page->flags))
1179 break;
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001180 } else if (behavior == SHARED) {
Nicholas Piggin62906022016-12-25 13:00:30 +10001181 if (!test_bit(bit_nr, &page->flags))
1182 break;
1183 }
Linus Torvaldsa8b169a2017-08-27 16:25:09 -07001184
Davidlohr Buesofa45f112019-01-03 15:28:55 -08001185 if (signal_pending_state(state, current)) {
Linus Torvaldsa8b169a2017-08-27 16:25:09 -07001186 ret = -EINTR;
1187 break;
1188 }
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001189
1190 if (behavior == DROP) {
1191 /*
1192 * We can no longer safely access page->flags:
1193 * even if CONFIG_MEMORY_HOTREMOVE is not enabled,
1194 * there is a risk of waiting forever on a page reused
1195 * for something that keeps it locked indefinitely.
1196 * But best check for -EINTR above before breaking.
1197 */
1198 break;
1199 }
Nicholas Piggin62906022016-12-25 13:00:30 +10001200 }
1201
1202 finish_wait(q, wait);
1203
Johannes Weinereb414682018-10-26 15:06:27 -07001204 if (thrashing) {
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001205 if (delayacct)
Johannes Weinereb414682018-10-26 15:06:27 -07001206 delayacct_thrashing_end();
1207 psi_memstall_leave(&pflags);
1208 }
Johannes Weinerb1d29ba2018-10-26 15:06:08 -07001209
Nicholas Piggin62906022016-12-25 13:00:30 +10001210 /*
1211 * A signal could leave PageWaiters set. Clearing it here if
1212 * !waitqueue_active would be possible (by open-coding finish_wait),
1213 * but still fail to catch it in the case of wait hash collision. We
1214 * already can fail to clear wait hash collision cases, so don't
1215 * bother with signals either.
1216 */
1217
1218 return ret;
1219}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
Harvey Harrison920c7a52008-02-04 22:29:26 -08001221void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222{
Nicholas Piggin62906022016-12-25 13:00:30 +10001223 wait_queue_head_t *q = page_waitqueue(page);
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001224 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225}
1226EXPORT_SYMBOL(wait_on_page_bit);
1227
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -07001228int wait_on_page_bit_killable(struct page *page, int bit_nr)
1229{
Nicholas Piggin62906022016-12-25 13:00:30 +10001230 wait_queue_head_t *q = page_waitqueue(page);
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001231 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -07001232}
David Howells4343d002017-11-02 15:27:52 +00001233EXPORT_SYMBOL(wait_on_page_bit_killable);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -07001234
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235/**
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001236 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1237 * @page: The page to wait for.
1238 *
1239 * The caller should hold a reference on @page. They expect the page to
1240 * become unlocked relatively soon, but do not wish to hold up migration
1241 * (for example) by holding the reference while waiting for the page to
1242 * come unlocked. After this function returns, the caller should not
1243 * dereference @page.
1244 */
1245void put_and_wait_on_page_locked(struct page *page)
1246{
1247 wait_queue_head_t *q;
1248
1249 page = compound_head(page);
1250 q = page_waitqueue(page);
1251 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1252}
1253
1254/**
David Howells385e1ca5f2009-04-03 16:42:39 +01001255 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -07001256 * @page: Page defining the wait queue of interest
1257 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +01001258 *
1259 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1260 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02001261void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
David Howells385e1ca5f2009-04-03 16:42:39 +01001262{
1263 wait_queue_head_t *q = page_waitqueue(page);
1264 unsigned long flags;
1265
1266 spin_lock_irqsave(&q->lock, flags);
Linus Torvalds9c3a8152017-08-28 16:45:40 -07001267 __add_wait_queue_entry_tail(q, waiter);
Nicholas Piggin62906022016-12-25 13:00:30 +10001268 SetPageWaiters(page);
David Howells385e1ca5f2009-04-03 16:42:39 +01001269 spin_unlock_irqrestore(&q->lock, flags);
1270}
1271EXPORT_SYMBOL_GPL(add_page_wait_queue);
1272
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001273#ifndef clear_bit_unlock_is_negative_byte
1274
1275/*
1276 * PG_waiters is the high bit in the same byte as PG_lock.
1277 *
1278 * On x86 (and on many other architectures), we can clear PG_lock and
1279 * test the sign bit at the same time. But if the architecture does
1280 * not support that special operation, we just do this all by hand
1281 * instead.
1282 *
1283 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1284 * being cleared, but a memory barrier should be unneccssary since it is
1285 * in the same byte as PG_locked.
1286 */
1287static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1288{
1289 clear_bit_unlock(nr, mem);
1290 /* smp_mb__after_atomic(); */
Olof Johansson98473f92016-12-29 14:16:07 -08001291 return test_bit(PG_waiters, mem);
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001292}
1293
1294#endif
1295
David Howells385e1ca5f2009-04-03 16:42:39 +01001296/**
Randy Dunlap485bb992006-06-23 02:03:49 -07001297 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 * @page: the page
1299 *
1300 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1301 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +09001302 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1304 *
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001305 * Note that this depends on PG_waiters being the sign bit in the byte
1306 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1307 * clear the PG_locked bit and test PG_waiters at the same time fairly
1308 * portably (architectures that do LL/SC can test any bit, while x86 can
1309 * test the sign bit).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001311void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312{
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001313 BUILD_BUG_ON(PG_waiters != 7);
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -08001314 page = compound_head(page);
Sasha Levin309381fea2014-01-23 15:52:54 -08001315 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvaldsb91e1302016-12-27 11:40:38 -08001316 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1317 wake_up_page_bit(page, PG_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318}
1319EXPORT_SYMBOL(unlock_page);
1320
Randy Dunlap485bb992006-06-23 02:03:49 -07001321/**
1322 * end_page_writeback - end writeback against a page
1323 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 */
1325void end_page_writeback(struct page *page)
1326{
Mel Gorman888cf2d2014-06-04 16:10:34 -07001327 /*
1328 * TestClearPageReclaim could be used here but it is an atomic
1329 * operation and overkill in this particular case. Failing to
1330 * shuffle a page marked for immediate reclaim is too mild to
1331 * justify taking an atomic operation penalty at the end of
1332 * ever page writeback.
1333 */
1334 if (PageReclaim(page)) {
1335 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -07001336 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -07001337 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -07001338
1339 if (!test_clear_page_writeback(page))
1340 BUG();
1341
Peter Zijlstra4e857c52014-03-17 18:06:10 +01001342 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 wake_up_page(page, PG_writeback);
1344}
1345EXPORT_SYMBOL(end_page_writeback);
1346
Matthew Wilcox57d99842014-06-04 16:07:45 -07001347/*
1348 * After completing I/O on a page, call this routine to update the page
1349 * flags appropriately
1350 */
Jens Axboec11f0c02016-08-05 08:11:04 -06001351void page_endio(struct page *page, bool is_write, int err)
Matthew Wilcox57d99842014-06-04 16:07:45 -07001352{
Jens Axboec11f0c02016-08-05 08:11:04 -06001353 if (!is_write) {
Matthew Wilcox57d99842014-06-04 16:07:45 -07001354 if (!err) {
1355 SetPageUptodate(page);
1356 } else {
1357 ClearPageUptodate(page);
1358 SetPageError(page);
1359 }
1360 unlock_page(page);
Mike Christieabf54542016-08-04 14:23:34 -06001361 } else {
Matthew Wilcox57d99842014-06-04 16:07:45 -07001362 if (err) {
Minchan Kimdd8416c2017-02-24 14:59:59 -08001363 struct address_space *mapping;
1364
Matthew Wilcox57d99842014-06-04 16:07:45 -07001365 SetPageError(page);
Minchan Kimdd8416c2017-02-24 14:59:59 -08001366 mapping = page_mapping(page);
1367 if (mapping)
1368 mapping_set_error(mapping, err);
Matthew Wilcox57d99842014-06-04 16:07:45 -07001369 }
1370 end_page_writeback(page);
1371 }
1372}
1373EXPORT_SYMBOL_GPL(page_endio);
1374
Randy Dunlap485bb992006-06-23 02:03:49 -07001375/**
1376 * __lock_page - get a lock on the page, assuming we need to sleep to get it
Randy Dunlap87066752017-02-22 15:44:44 -08001377 * @__page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 */
Nicholas Piggin62906022016-12-25 13:00:30 +10001379void __lock_page(struct page *__page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380{
Nicholas Piggin62906022016-12-25 13:00:30 +10001381 struct page *page = compound_head(__page);
1382 wait_queue_head_t *q = page_waitqueue(page);
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001383 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1384 EXCLUSIVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385}
1386EXPORT_SYMBOL(__lock_page);
1387
Nicholas Piggin62906022016-12-25 13:00:30 +10001388int __lock_page_killable(struct page *__page)
Matthew Wilcox2687a352007-12-06 11:18:49 -05001389{
Nicholas Piggin62906022016-12-25 13:00:30 +10001390 struct page *page = compound_head(__page);
1391 wait_queue_head_t *q = page_waitqueue(page);
Hugh Dickins9a1ea432018-12-28 00:36:14 -08001392 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1393 EXCLUSIVE);
Matthew Wilcox2687a352007-12-06 11:18:49 -05001394}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +03001395EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -05001396
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001397/*
1398 * Return values:
1399 * 1 - page is locked; mmap_sem is still held.
1400 * 0 - page is not locked.
1401 * mmap_sem has been released (up_read()), unless flags had both
1402 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1403 * which case mmap_sem is still held.
1404 *
1405 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1406 * with the page locked and the mmap_sem unperturbed.
1407 */
Michel Lespinassed065bd82010-10-26 14:21:57 -07001408int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1409 unsigned int flags)
1410{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001411 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1412 /*
1413 * CAUTION! In this case, mmap_sem is not released
1414 * even though return 0.
1415 */
1416 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1417 return 0;
1418
1419 up_read(&mm->mmap_sem);
1420 if (flags & FAULT_FLAG_KILLABLE)
1421 wait_on_page_locked_killable(page);
1422 else
Gleb Natapov318b2752011-03-22 16:30:51 -07001423 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001424 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001425 } else {
1426 if (flags & FAULT_FLAG_KILLABLE) {
1427 int ret;
1428
1429 ret = __lock_page_killable(page);
1430 if (ret) {
1431 up_read(&mm->mmap_sem);
1432 return 0;
1433 }
1434 } else
1435 __lock_page(page);
1436 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -07001437 }
1438}
1439
Randy Dunlap485bb992006-06-23 02:03:49 -07001440/**
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001441 * page_cache_next_miss() - Find the next gap in the page cache.
1442 * @mapping: Mapping.
1443 * @index: Index.
1444 * @max_scan: Maximum range to search.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001445 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001446 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1447 * gap with the lowest index.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001448 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001449 * This function may be called under the rcu_read_lock. However, this will
1450 * not atomically search a snapshot of the cache at a single point in time.
1451 * For example, if a gap is created at index 5, then subsequently a gap is
1452 * created at index 10, page_cache_next_miss covering both indices may
1453 * return 10 if called under the rcu_read_lock.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001454 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001455 * Return: The index of the gap if found, otherwise an index outside the
1456 * range specified (in which case 'return - index >= max_scan' will be true).
1457 * In the rare case of index wrap-around, 0 will be returned.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001458 */
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001459pgoff_t page_cache_next_miss(struct address_space *mapping,
Johannes Weinere7b563b2014-04-03 14:47:44 -07001460 pgoff_t index, unsigned long max_scan)
1461{
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001462 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001463
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001464 while (max_scan--) {
1465 void *entry = xas_next(&xas);
1466 if (!entry || xa_is_value(entry))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001467 break;
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001468 if (xas.xa_index == 0)
Johannes Weinere7b563b2014-04-03 14:47:44 -07001469 break;
1470 }
1471
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001472 return xas.xa_index;
Johannes Weinere7b563b2014-04-03 14:47:44 -07001473}
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001474EXPORT_SYMBOL(page_cache_next_miss);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001475
1476/**
Laurent Dufour2346a562019-05-13 17:21:29 -07001477 * page_cache_prev_miss() - Find the previous gap in the page cache.
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001478 * @mapping: Mapping.
1479 * @index: Index.
1480 * @max_scan: Maximum range to search.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001481 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001482 * Search the range [max(index - max_scan + 1, 0), index] for the
1483 * gap with the highest index.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001484 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001485 * This function may be called under the rcu_read_lock. However, this will
1486 * not atomically search a snapshot of the cache at a single point in time.
1487 * For example, if a gap is created at index 10, then subsequently a gap is
1488 * created at index 5, page_cache_prev_miss() covering both indices may
1489 * return 5 if called under the rcu_read_lock.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001490 *
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001491 * Return: The index of the gap if found, otherwise an index outside the
1492 * range specified (in which case 'index - return >= max_scan' will be true).
1493 * In the rare case of wrap-around, ULONG_MAX will be returned.
Johannes Weinere7b563b2014-04-03 14:47:44 -07001494 */
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001495pgoff_t page_cache_prev_miss(struct address_space *mapping,
Johannes Weinere7b563b2014-04-03 14:47:44 -07001496 pgoff_t index, unsigned long max_scan)
1497{
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001498 XA_STATE(xas, &mapping->i_pages, index);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001499
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001500 while (max_scan--) {
1501 void *entry = xas_prev(&xas);
1502 if (!entry || xa_is_value(entry))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001503 break;
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001504 if (xas.xa_index == ULONG_MAX)
Johannes Weinere7b563b2014-04-03 14:47:44 -07001505 break;
1506 }
1507
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001508 return xas.xa_index;
Johannes Weinere7b563b2014-04-03 14:47:44 -07001509}
Matthew Wilcox0d3f9292017-11-21 14:07:06 -05001510EXPORT_SYMBOL(page_cache_prev_miss);
Johannes Weinere7b563b2014-04-03 14:47:44 -07001511
1512/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001513 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -07001514 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -07001515 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -07001516 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001517 * Looks up the page cache slot at @mapping & @offset. If there is a
1518 * page cache page, it is returned with an increased refcount.
1519 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001520 * If the slot holds a shadow entry of a previously evicted page, or a
1521 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001522 *
Mike Rapoporta862f682019-03-05 15:48:42 -08001523 * Return: the found page or shadow entry, %NULL if nothing is found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 */
Johannes Weiner0cd61442014-04-03 14:47:46 -07001525struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526{
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001527 XA_STATE(xas, &mapping->i_pages, offset);
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001528 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Nick Piggina60637c2008-07-25 19:45:31 -07001530 rcu_read_lock();
1531repeat:
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001532 xas_reset(&xas);
1533 page = xas_load(&xas);
1534 if (xas_retry(&xas, page))
1535 goto repeat;
1536 /*
1537 * A shadow entry of a recently evicted page, or a swap entry from
1538 * shmem/tmpfs. Return it without attempting to raise page count.
1539 */
1540 if (!page || xa_is_value(page))
1541 goto out;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001542
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001543 if (!page_cache_get_speculative(page))
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001544 goto repeat;
Nick Piggina60637c2008-07-25 19:45:31 -07001545
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001546 /*
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001547 * Has the page moved or been split?
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001548 * This is part of the lockless pagecache protocol. See
1549 * include/linux/pagemap.h for details.
1550 */
1551 if (unlikely(page != xas_reload(&xas))) {
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001552 put_page(page);
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001553 goto repeat;
Nick Piggina60637c2008-07-25 19:45:31 -07001554 }
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001555 page = find_subpage(page, offset);
Nick Piggin27d20fd2010-11-11 14:05:19 -08001556out:
Nick Piggina60637c2008-07-25 19:45:31 -07001557 rcu_read_unlock();
1558
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 return page;
1560}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001561EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562
Randy Dunlap485bb992006-06-23 02:03:49 -07001563/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001564 * find_lock_entry - locate, pin and lock a page cache entry
1565 * @mapping: the address_space to search
1566 * @offset: the page cache index
1567 *
1568 * Looks up the page cache slot at @mapping & @offset. If there is a
1569 * page cache page, it is returned locked and with an increased
1570 * refcount.
1571 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001572 * If the slot holds a shadow entry of a previously evicted page, or a
1573 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001574 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001575 * find_lock_entry() may sleep.
Mike Rapoporta862f682019-03-05 15:48:42 -08001576 *
1577 * Return: the found page or shadow entry, %NULL if nothing is found.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001578 */
1579struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580{
1581 struct page *page;
1582
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001584 page = find_get_entry(mapping, offset);
Matthew Wilcox4c7472c2018-05-16 16:12:50 -04001585 if (page && !xa_is_value(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001586 lock_page(page);
1587 /* Has the page been truncated? */
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001588 if (unlikely(page_mapping(page) != mapping)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001589 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001590 put_page(page);
Nick Piggina60637c2008-07-25 19:45:31 -07001591 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001593 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 return page;
1596}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001597EXPORT_SYMBOL(find_lock_entry);
1598
1599/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001600 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001601 * @mapping: the address_space to search
1602 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001603 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001604 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001605 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001606 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001607 *
Randy Dunlap75325182014-07-30 16:08:37 -07001608 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001609 *
mchehab@s-opensource.com0e056eb2017-03-30 17:11:36 -03001610 * @fgp_flags can be:
1611 *
1612 * - FGP_ACCESSED: the page will be marked accessed
1613 * - FGP_LOCK: Page is return locked
1614 * - FGP_CREAT: If page is not present then a new page is allocated using
1615 * @gfp_mask and added to the page cache and the VM's LRU
1616 * list. The page is returned locked and with an increased
Mike Rapoporta862f682019-03-05 15:48:42 -08001617 * refcount.
Josef Bacika75d4c32019-03-13 11:44:14 -07001618 * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do
1619 * its own locking dance if the page is already in cache, or unlock the page
1620 * before returning if we had to add the page to pagecache.
Mel Gorman2457aec2014-06-04 16:10:31 -07001621 *
1622 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1623 * if the GFP flags specified for FGP_CREAT are atomic.
1624 *
1625 * If there is a page cache page, it is returned with an increased refcount.
Mike Rapoporta862f682019-03-05 15:48:42 -08001626 *
1627 * Return: the found page or %NULL otherwise.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001628 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001629struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001630 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631{
Nick Piggineb2be182007-10-16 01:24:57 -07001632 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001633
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001635 page = find_get_entry(mapping, offset);
Matthew Wilcox3159f942017-11-03 13:30:42 -04001636 if (xa_is_value(page))
Mel Gorman2457aec2014-06-04 16:10:31 -07001637 page = NULL;
1638 if (!page)
1639 goto no_page;
1640
1641 if (fgp_flags & FGP_LOCK) {
1642 if (fgp_flags & FGP_NOWAIT) {
1643 if (!trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001644 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001645 return NULL;
1646 }
1647 } else {
1648 lock_page(page);
1649 }
1650
1651 /* Has the page been truncated? */
Song Liu31895432019-09-23 15:37:47 -07001652 if (unlikely(compound_head(page)->mapping != mapping)) {
Mel Gorman2457aec2014-06-04 16:10:31 -07001653 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001654 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001655 goto repeat;
1656 }
1657 VM_BUG_ON_PAGE(page->index != offset, page);
1658 }
1659
Kirill Tkhaic16eb002018-12-28 00:37:35 -08001660 if (fgp_flags & FGP_ACCESSED)
Mel Gorman2457aec2014-06-04 16:10:31 -07001661 mark_page_accessed(page);
1662
1663no_page:
1664 if (!page && (fgp_flags & FGP_CREAT)) {
1665 int err;
1666 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001667 gfp_mask |= __GFP_WRITE;
1668 if (fgp_flags & FGP_NOFS)
1669 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001670
Michal Hocko45f87de2014-12-29 20:30:35 +01001671 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001672 if (!page)
1673 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001674
Josef Bacika75d4c32019-03-13 11:44:14 -07001675 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
Mel Gorman2457aec2014-06-04 16:10:31 -07001676 fgp_flags |= FGP_LOCK;
1677
Hugh Dickinseb39d612014-08-06 16:06:43 -07001678 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001679 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001680 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001681
Matthew Wilcoxabc1be12018-04-20 14:56:20 -07001682 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001683 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001684 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07001685 page = NULL;
1686 if (err == -EEXIST)
1687 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 }
Josef Bacika75d4c32019-03-13 11:44:14 -07001689
1690 /*
1691 * add_to_page_cache_lru locks the page, and for mmap we expect
1692 * an unlocked page.
1693 */
1694 if (page && (fgp_flags & FGP_FOR_MMAP))
1695 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001697
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 return page;
1699}
Mel Gorman2457aec2014-06-04 16:10:31 -07001700EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
1702/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001703 * find_get_entries - gang pagecache lookup
1704 * @mapping: The address_space to search
1705 * @start: The starting page cache index
1706 * @nr_entries: The maximum number of entries
1707 * @entries: Where the resulting entries are placed
1708 * @indices: The cache indices corresponding to the entries in @entries
1709 *
1710 * find_get_entries() will search for and return a group of up to
1711 * @nr_entries entries in the mapping. The entries are placed at
1712 * @entries. find_get_entries() takes a reference against any actual
1713 * pages it returns.
1714 *
1715 * The search returns a group of mapping-contiguous page cache entries
1716 * with ascending indexes. There may be holes in the indices due to
1717 * not-present pages.
1718 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001719 * Any shadow entries of evicted pages, or swap entries from
1720 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001721 *
Mike Rapoporta862f682019-03-05 15:48:42 -08001722 * Return: the number of pages and shadow entries which were found.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001723 */
1724unsigned find_get_entries(struct address_space *mapping,
1725 pgoff_t start, unsigned int nr_entries,
1726 struct page **entries, pgoff_t *indices)
1727{
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001728 XA_STATE(xas, &mapping->i_pages, start);
1729 struct page *page;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001730 unsigned int ret = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001731
1732 if (!nr_entries)
1733 return 0;
1734
1735 rcu_read_lock();
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001736 xas_for_each(&xas, page, ULONG_MAX) {
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001737 if (xas_retry(&xas, page))
Johannes Weiner0cd61442014-04-03 14:47:46 -07001738 continue;
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001739 /*
1740 * A shadow entry of a recently evicted page, a swap
1741 * entry from shmem/tmpfs or a DAX entry. Return it
1742 * without attempting to raise page count.
1743 */
1744 if (xa_is_value(page))
Johannes Weiner0cd61442014-04-03 14:47:46 -07001745 goto export;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001746
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001747 if (!page_cache_get_speculative(page))
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001748 goto retry;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001749
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001750 /* Has the page moved or been split? */
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001751 if (unlikely(page != xas_reload(&xas)))
1752 goto put_page;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001753 page = find_subpage(page, xas.xa_index);
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001754
Johannes Weiner0cd61442014-04-03 14:47:46 -07001755export:
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001756 indices[ret] = xas.xa_index;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001757 entries[ret] = page;
1758 if (++ret == nr_entries)
1759 break;
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001760 continue;
1761put_page:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001762 put_page(page);
Matthew Wilcoxf280bf02018-05-16 17:20:45 -04001763retry:
1764 xas_reset(&xas);
Johannes Weiner0cd61442014-04-03 14:47:46 -07001765 }
1766 rcu_read_unlock();
1767 return ret;
1768}
1769
1770/**
Jan Karab947cee2017-09-06 16:21:21 -07001771 * find_get_pages_range - gang pagecache lookup
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 * @mapping: The address_space to search
1773 * @start: The starting page index
Jan Karab947cee2017-09-06 16:21:21 -07001774 * @end: The final page index (inclusive)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 * @nr_pages: The maximum number of pages
1776 * @pages: Where the resulting pages are placed
1777 *
Jan Karab947cee2017-09-06 16:21:21 -07001778 * find_get_pages_range() will search for and return a group of up to @nr_pages
1779 * pages in the mapping starting at index @start and up to index @end
1780 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1781 * a reference against the returned pages.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 *
1783 * The search returns a group of mapping-contiguous pages with ascending
1784 * indexes. There may be holes in the indices due to not-present pages.
Jan Karad72dc8a2017-09-06 16:21:18 -07001785 * We also update @start to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 *
Mike Rapoporta862f682019-03-05 15:48:42 -08001787 * Return: the number of pages which were found. If this number is
1788 * smaller than @nr_pages, the end of specified range has been
Jan Karab947cee2017-09-06 16:21:21 -07001789 * reached.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 */
Jan Karab947cee2017-09-06 16:21:21 -07001791unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1792 pgoff_t end, unsigned int nr_pages,
1793 struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794{
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001795 XA_STATE(xas, &mapping->i_pages, *start);
1796 struct page *page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001797 unsigned ret = 0;
1798
1799 if (unlikely(!nr_pages))
1800 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
Nick Piggina60637c2008-07-25 19:45:31 -07001802 rcu_read_lock();
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001803 xas_for_each(&xas, page, end) {
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001804 if (xas_retry(&xas, page))
Nick Piggina60637c2008-07-25 19:45:31 -07001805 continue;
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001806 /* Skip over shadow, swap and DAX entries */
1807 if (xa_is_value(page))
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001808 continue;
Nick Piggina60637c2008-07-25 19:45:31 -07001809
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001810 if (!page_cache_get_speculative(page))
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001811 goto retry;
Nick Piggina60637c2008-07-25 19:45:31 -07001812
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001813 /* Has the page moved or been split? */
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001814 if (unlikely(page != xas_reload(&xas)))
1815 goto put_page;
Nick Piggina60637c2008-07-25 19:45:31 -07001816
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001817 pages[ret] = find_subpage(page, xas.xa_index);
Jan Karab947cee2017-09-06 16:21:21 -07001818 if (++ret == nr_pages) {
Yu Zhao5d3ee422019-03-05 15:49:17 -08001819 *start = xas.xa_index + 1;
Jan Karab947cee2017-09-06 16:21:21 -07001820 goto out;
1821 }
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001822 continue;
1823put_page:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001824 put_page(page);
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001825retry:
1826 xas_reset(&xas);
Nick Piggina60637c2008-07-25 19:45:31 -07001827 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001828
Jan Karab947cee2017-09-06 16:21:21 -07001829 /*
1830 * We come here when there is no page beyond @end. We take care to not
1831 * overflow the index @start as it confuses some of the callers. This
Matthew Wilcoxfd1b3ce2018-05-16 17:38:56 -04001832 * breaks the iteration when there is a page at index -1 but that is
Jan Karab947cee2017-09-06 16:21:21 -07001833 * already broken anyway.
1834 */
1835 if (end == (pgoff_t)-1)
1836 *start = (pgoff_t)-1;
1837 else
1838 *start = end + 1;
1839out:
Nick Piggina60637c2008-07-25 19:45:31 -07001840 rcu_read_unlock();
Jan Karad72dc8a2017-09-06 16:21:18 -07001841
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 return ret;
1843}
1844
Jens Axboeebf43502006-04-27 08:46:01 +02001845/**
1846 * find_get_pages_contig - gang contiguous pagecache lookup
1847 * @mapping: The address_space to search
1848 * @index: The starting page index
1849 * @nr_pages: The maximum number of pages
1850 * @pages: Where the resulting pages are placed
1851 *
1852 * find_get_pages_contig() works exactly like find_get_pages(), except
1853 * that the returned number of pages are guaranteed to be contiguous.
1854 *
Mike Rapoporta862f682019-03-05 15:48:42 -08001855 * Return: the number of pages which were found.
Jens Axboeebf43502006-04-27 08:46:01 +02001856 */
1857unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1858 unsigned int nr_pages, struct page **pages)
1859{
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04001860 XA_STATE(xas, &mapping->i_pages, index);
1861 struct page *page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001862 unsigned int ret = 0;
1863
1864 if (unlikely(!nr_pages))
1865 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001866
Nick Piggina60637c2008-07-25 19:45:31 -07001867 rcu_read_lock();
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04001868 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04001869 if (xas_retry(&xas, page))
1870 continue;
1871 /*
1872 * If the entry has been swapped out, we can stop looking.
1873 * No current caller is looking for DAX entries.
1874 */
1875 if (xa_is_value(page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001876 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001877
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001878 if (!page_cache_get_speculative(page))
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04001879 goto retry;
Nick Piggina60637c2008-07-25 19:45:31 -07001880
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001881 /* Has the page moved or been split? */
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04001882 if (unlikely(page != xas_reload(&xas)))
1883 goto put_page;
Nick Piggina60637c2008-07-25 19:45:31 -07001884
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001885 pages[ret] = find_subpage(page, xas.xa_index);
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001886 if (++ret == nr_pages)
1887 break;
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04001888 continue;
1889put_page:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001890 put_page(page);
Matthew Wilcox3ece58a2018-05-16 18:00:33 -04001891retry:
1892 xas_reset(&xas);
Jens Axboeebf43502006-04-27 08:46:01 +02001893 }
Nick Piggina60637c2008-07-25 19:45:31 -07001894 rcu_read_unlock();
1895 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001896}
David Howellsef71c152007-05-09 02:33:44 -07001897EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001898
Randy Dunlap485bb992006-06-23 02:03:49 -07001899/**
Jan Kara72b045a2017-11-15 17:34:33 -08001900 * find_get_pages_range_tag - find and return pages in given range matching @tag
Randy Dunlap485bb992006-06-23 02:03:49 -07001901 * @mapping: the address_space to search
1902 * @index: the starting page index
Jan Kara72b045a2017-11-15 17:34:33 -08001903 * @end: The final page index (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -07001904 * @tag: the tag index
1905 * @nr_pages: the maximum number of pages
1906 * @pages: where the resulting pages are placed
1907 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001909 * @tag. We update @index to index the next page for the traversal.
Mike Rapoporta862f682019-03-05 15:48:42 -08001910 *
1911 * Return: the number of pages which were found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 */
Jan Kara72b045a2017-11-15 17:34:33 -08001913unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
Matthew Wilcoxa6906972018-05-16 18:12:54 -04001914 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
Jan Kara72b045a2017-11-15 17:34:33 -08001915 struct page **pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916{
Matthew Wilcoxa6906972018-05-16 18:12:54 -04001917 XA_STATE(xas, &mapping->i_pages, *index);
1918 struct page *page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001919 unsigned ret = 0;
1920
1921 if (unlikely(!nr_pages))
1922 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
Nick Piggina60637c2008-07-25 19:45:31 -07001924 rcu_read_lock();
Matthew Wilcoxa6906972018-05-16 18:12:54 -04001925 xas_for_each_marked(&xas, page, end, tag) {
Matthew Wilcoxa6906972018-05-16 18:12:54 -04001926 if (xas_retry(&xas, page))
Nick Piggina60637c2008-07-25 19:45:31 -07001927 continue;
Matthew Wilcoxa6906972018-05-16 18:12:54 -04001928 /*
1929 * Shadow entries should never be tagged, but this iteration
1930 * is lockless so there is a window for page reclaim to evict
1931 * a page we saw tagged. Skip over it.
1932 */
1933 if (xa_is_value(page))
Johannes Weiner139b6a62014-05-06 12:50:05 -07001934 continue;
Nick Piggina60637c2008-07-25 19:45:31 -07001935
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001936 if (!page_cache_get_speculative(page))
Matthew Wilcoxa6906972018-05-16 18:12:54 -04001937 goto retry;
Nick Piggina60637c2008-07-25 19:45:31 -07001938
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001939 /* Has the page moved or been split? */
Matthew Wilcoxa6906972018-05-16 18:12:54 -04001940 if (unlikely(page != xas_reload(&xas)))
1941 goto put_page;
Nick Piggina60637c2008-07-25 19:45:31 -07001942
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001943 pages[ret] = find_subpage(page, xas.xa_index);
Jan Kara72b045a2017-11-15 17:34:33 -08001944 if (++ret == nr_pages) {
Yu Zhao5d3ee422019-03-05 15:49:17 -08001945 *index = xas.xa_index + 1;
Jan Kara72b045a2017-11-15 17:34:33 -08001946 goto out;
1947 }
Matthew Wilcoxa6906972018-05-16 18:12:54 -04001948 continue;
1949put_page:
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07001950 put_page(page);
Matthew Wilcoxa6906972018-05-16 18:12:54 -04001951retry:
1952 xas_reset(&xas);
Nick Piggina60637c2008-07-25 19:45:31 -07001953 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001954
Jan Kara72b045a2017-11-15 17:34:33 -08001955 /*
Matthew Wilcoxa6906972018-05-16 18:12:54 -04001956 * We come here when we got to @end. We take care to not overflow the
Jan Kara72b045a2017-11-15 17:34:33 -08001957 * index @index as it confuses some of the callers. This breaks the
Matthew Wilcoxa6906972018-05-16 18:12:54 -04001958 * iteration when there is a page at index -1 but that is already
1959 * broken anyway.
Jan Kara72b045a2017-11-15 17:34:33 -08001960 */
1961 if (end == (pgoff_t)-1)
1962 *index = (pgoff_t)-1;
1963 else
1964 *index = end + 1;
1965out:
Nick Piggina60637c2008-07-25 19:45:31 -07001966 rcu_read_unlock();
1967
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 return ret;
1969}
Jan Kara72b045a2017-11-15 17:34:33 -08001970EXPORT_SYMBOL(find_get_pages_range_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001972/*
1973 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1974 * a _large_ part of the i/o request. Imagine the worst scenario:
1975 *
1976 * ---R__________________________________________B__________
1977 * ^ reading here ^ bad block(assume 4k)
1978 *
1979 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1980 * => failing the whole request => read(R) => read(R+1) =>
1981 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1982 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1983 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1984 *
1985 * It is going insane. Fix it by quickly scaling down the readahead size.
1986 */
1987static void shrink_readahead_size_eio(struct file *filp,
1988 struct file_ra_state *ra)
1989{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001990 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001991}
1992
Randy Dunlap485bb992006-06-23 02:03:49 -07001993/**
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02001994 * generic_file_buffered_read - generic file read routine
1995 * @iocb: the iocb to read
Al Viro6e58e792014-02-03 17:07:03 -05001996 * @iter: data destination
1997 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001998 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07002000 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 *
2002 * This is really ugly. But the goto's actually try to clarify some
2003 * of the logic when it comes to error handling etc.
Mike Rapoporta862f682019-03-05 15:48:42 -08002004 *
2005 * Return:
2006 * * total number of bytes copied, including those the were already @written
2007 * * negative error code if nothing was copied
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 */
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002009static ssize_t generic_file_buffered_read(struct kiocb *iocb,
Al Viro6e58e792014-02-03 17:07:03 -05002010 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011{
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002012 struct file *filp = iocb->ki_filp;
Christoph Hellwig36e78912008-02-08 04:21:24 -08002013 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08002015 struct file_ra_state *ra = &filp->f_ra;
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002016 loff_t *ppos = &iocb->ki_pos;
Fengguang Wu57f6b962007-10-16 01:24:37 -07002017 pgoff_t index;
2018 pgoff_t last_index;
2019 pgoff_t prev_index;
2020 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07002021 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05002022 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
Wei Fangc2a97372016-10-07 17:01:52 -07002024 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
Linus Torvaldsd05c5f72016-12-14 12:45:25 -08002025 return 0;
Wei Fangc2a97372016-10-07 17:01:52 -07002026 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2027
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002028 index = *ppos >> PAGE_SHIFT;
2029 prev_index = ra->prev_pos >> PAGE_SHIFT;
2030 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2031 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2032 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 for (;;) {
2035 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07002036 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07002037 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 unsigned long nr, ret;
2039
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041find_page:
Michal Hocko5abf1862017-02-03 13:13:29 -08002042 if (fatal_signal_pending(current)) {
2043 error = -EINTR;
2044 goto out;
2045 }
2046
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07002048 if (!page) {
Milosz Tanski3239d832017-08-29 16:13:19 +02002049 if (iocb->ki_flags & IOCB_NOWAIT)
2050 goto would_block;
Rusty Russellcf914a72007-07-19 01:48:08 -07002051 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07002052 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07002053 index, last_index - index);
2054 page = find_get_page(mapping, index);
2055 if (unlikely(page == NULL))
2056 goto no_cached_page;
2057 }
2058 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07002059 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07002060 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07002061 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002063 if (!PageUptodate(page)) {
Milosz Tanski3239d832017-08-29 16:13:19 +02002064 if (iocb->ki_flags & IOCB_NOWAIT) {
2065 put_page(page);
2066 goto would_block;
2067 }
2068
Mel Gormanebded022016-03-15 14:55:39 -07002069 /*
2070 * See comment in do_read_cache_page on why
2071 * wait_on_page_locked is used to avoid unnecessarily
2072 * serialisations and why it's safe.
2073 */
Bart Van Asschec4b209a2016-10-07 16:58:33 -07002074 error = wait_on_page_locked_killable(page);
2075 if (unlikely(error))
2076 goto readpage_error;
Mel Gormanebded022016-03-15 14:55:39 -07002077 if (PageUptodate(page))
2078 goto page_ok;
2079
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002080 if (inode->i_blkbits == PAGE_SHIFT ||
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002081 !mapping->a_ops->is_partially_uptodate)
2082 goto page_not_up_to_date;
Eryu Guan6d6d36b2016-11-01 15:43:07 +08002083 /* pipes can't handle partially uptodate pages */
David Howells00e23702018-10-22 13:07:28 +01002084 if (unlikely(iov_iter_is_pipe(iter)))
Eryu Guan6d6d36b2016-11-01 15:43:07 +08002085 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02002086 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002087 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08002088 /* Did it get truncated before we got the lock? */
2089 if (!page->mapping)
2090 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002091 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05002092 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002093 goto page_not_up_to_date_locked;
2094 unlock_page(page);
2095 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07002097 /*
2098 * i_size must be checked after we know the page is Uptodate.
2099 *
2100 * Checking i_size after the check allows us to calculate
2101 * the correct value for "nr", which means the zero-filled
2102 * part of the page is not copied back to userspace (unless
2103 * another truncate extends the file - this is desired though).
2104 */
2105
2106 isize = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002107 end_index = (isize - 1) >> PAGE_SHIFT;
NeilBrowna32ea1e2007-07-17 04:03:04 -07002108 if (unlikely(!isize || index > end_index)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002109 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07002110 goto out;
2111 }
2112
2113 /* nr is the maximum number of bytes to copy from this page */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002114 nr = PAGE_SIZE;
NeilBrowna32ea1e2007-07-17 04:03:04 -07002115 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002116 nr = ((isize - 1) & ~PAGE_MASK) + 1;
NeilBrowna32ea1e2007-07-17 04:03:04 -07002117 if (nr <= offset) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002118 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07002119 goto out;
2120 }
2121 }
2122 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123
2124 /* If users can be writing to this page using arbitrary
2125 * virtual addresses, take care about potential aliasing
2126 * before reading the page on the kernel side.
2127 */
2128 if (mapping_writably_mapped(mapping))
2129 flush_dcache_page(page);
2130
2131 /*
Jan Karaec0f1632007-05-06 14:49:25 -07002132 * When a sequential read accesses a page several times,
2133 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 */
Jan Karaec0f1632007-05-06 14:49:25 -07002135 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 mark_page_accessed(page);
2137 prev_index = index;
2138
2139 /*
2140 * Ok, we have the page, and it's up-to-date, so
2141 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 */
Al Viro6e58e792014-02-03 17:07:03 -05002143
2144 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002146 index += offset >> PAGE_SHIFT;
2147 offset &= ~PAGE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07002148 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002150 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05002151 written += ret;
2152 if (!iov_iter_count(iter))
2153 goto out;
2154 if (ret < nr) {
2155 error = -EFAULT;
2156 goto out;
2157 }
2158 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159
2160page_not_up_to_date:
2161 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04002162 error = lock_page_killable(page);
2163 if (unlikely(error))
2164 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002166page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07002167 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 if (!page->mapping) {
2169 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002170 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 continue;
2172 }
2173
2174 /* Did somebody else fill it already? */
2175 if (PageUptodate(page)) {
2176 unlock_page(page);
2177 goto page_ok;
2178 }
2179
2180readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04002181 /*
2182 * A previous I/O error may have been due to temporary
2183 * failures, eg. multipath errors.
2184 * PG_error will be set again if readpage fails.
2185 */
2186 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 /* Start the actual read. The read will unlock the page. */
2188 error = mapping->a_ops->readpage(filp, page);
2189
Zach Brown994fc28c2005-12-15 14:28:17 -08002190 if (unlikely(error)) {
2191 if (error == AOP_TRUNCATED_PAGE) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002192 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05002193 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08002194 goto find_page;
2195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08002197 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198
2199 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04002200 error = lock_page_killable(page);
2201 if (unlikely(error))
2202 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 if (!PageUptodate(page)) {
2204 if (page->mapping == NULL) {
2205 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01002206 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 */
2208 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002209 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 goto find_page;
2211 }
2212 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07002213 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04002214 error = -EIO;
2215 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 }
2217 unlock_page(page);
2218 }
2219
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 goto page_ok;
2221
2222readpage_error:
2223 /* UHHUH! A synchronous read error occurred. Report it */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002224 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 goto out;
2226
2227no_cached_page:
2228 /*
2229 * Ok, it wasn't cached, so we need to create a new
2230 * page..
2231 */
Mel Gorman453f85d2017-11-15 17:38:03 -08002232 page = page_cache_alloc(mapping);
Nick Piggineb2be182007-10-16 01:24:57 -07002233 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05002234 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07002235 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 }
Michal Hocko6afdb852015-06-24 16:58:06 -07002237 error = add_to_page_cache_lru(page, mapping, index,
Michal Hockoc62d2552015-11-06 16:28:49 -08002238 mapping_gfp_constraint(mapping, GFP_KERNEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 if (error) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002240 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05002241 if (error == -EEXIST) {
2242 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05002244 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245 goto out;
2246 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 goto readpage;
2248 }
2249
Milosz Tanski3239d832017-08-29 16:13:19 +02002250would_block:
2251 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07002253 ra->prev_pos = prev_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002254 ra->prev_pos <<= PAGE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07002255 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002257 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07002258 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05002259 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260}
2261
Randy Dunlap485bb992006-06-23 02:03:49 -07002262/**
Al Viro6abd2322014-04-04 14:20:57 -04002263 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07002264 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04002265 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07002266 *
Al Viro6abd2322014-04-04 14:20:57 -04002267 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 * that can use the page cache directly.
Mike Rapoporta862f682019-03-05 15:48:42 -08002269 * Return:
2270 * * number of bytes copied, even for partial reads
2271 * * negative error code if nothing was read
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 */
2273ssize_t
Al Viroed978a82014-03-05 22:53:04 -05002274generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275{
Nicolai Stangee7080a42016-03-25 14:22:14 -07002276 size_t count = iov_iter_count(iter);
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002277 ssize_t retval = 0;
Nicolai Stangee7080a42016-03-25 14:22:14 -07002278
2279 if (!count)
2280 goto out; /* skip atime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281
Al Viro2ba48ce2015-04-09 13:52:01 -04002282 if (iocb->ki_flags & IOCB_DIRECT) {
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002283 struct file *file = iocb->ki_filp;
Al Viroed978a82014-03-05 22:53:04 -05002284 struct address_space *mapping = file->f_mapping;
2285 struct inode *inode = mapping->host;
Badari Pulavarty543ade12006-09-30 23:28:48 -07002286 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 size = i_size_read(inode);
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05002289 if (iocb->ki_flags & IOCB_NOWAIT) {
2290 if (filemap_range_has_page(mapping, iocb->ki_pos,
2291 iocb->ki_pos + count - 1))
2292 return -EAGAIN;
2293 } else {
2294 retval = filemap_write_and_wait_range(mapping,
2295 iocb->ki_pos,
2296 iocb->ki_pos + count - 1);
2297 if (retval < 0)
2298 goto out;
2299 }
Al Viroed978a82014-03-05 22:53:04 -05002300
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002301 file_accessed(file);
2302
Al Viro5ecda132017-04-13 14:13:36 -04002303 retval = mapping->a_ops->direct_IO(iocb, iter);
Al Viroc3a69022016-10-10 13:26:27 -04002304 if (retval >= 0) {
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07002305 iocb->ki_pos += retval;
Al Viro5ecda132017-04-13 14:13:36 -04002306 count -= retval;
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002307 }
Al Viro5b47d592017-05-08 13:54:47 -04002308 iov_iter_revert(iter, count - iov_iter_count(iter));
Josef Bacik66f998f2010-05-23 11:00:54 -04002309
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002310 /*
2311 * Btrfs can have a short DIO read if we encounter
2312 * compressed extents, so if there was an error, or if
2313 * we've already read everything we wanted to, or if
2314 * there was a short read because we hit EOF, go ahead
2315 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002316 * the rest of the read. Buffered reads will not work for
2317 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002318 */
Al Viro5ecda132017-04-13 14:13:36 -04002319 if (retval < 0 || !count || iocb->ki_pos >= size ||
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002320 IS_DAX(inode))
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002321 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 }
2323
Christoph Hellwig47c27bc2017-08-29 16:13:18 +02002324 retval = generic_file_buffered_read(iocb, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325out:
2326 return retval;
2327}
Al Viroed978a82014-03-05 22:53:04 -05002328EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330#ifdef CONFIG_MMU
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331#define MMAP_LOTSAMISS (100)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002332static struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
2333 struct file *fpin)
2334{
2335 int flags = vmf->flags;
2336
2337 if (fpin)
2338 return fpin;
2339
2340 /*
2341 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
2342 * anything, so we only pin the file and drop the mmap_sem if only
2343 * FAULT_FLAG_ALLOW_RETRY is set.
2344 */
2345 if ((flags & (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT)) ==
2346 FAULT_FLAG_ALLOW_RETRY) {
2347 fpin = get_file(vmf->vma->vm_file);
2348 up_read(&vmf->vma->vm_mm->mmap_sem);
2349 }
2350 return fpin;
2351}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352
Linus Torvaldsef00e082009-06-16 15:31:25 -07002353/*
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002354 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_sem
2355 * @vmf - the vm_fault for this fault.
2356 * @page - the page to lock.
2357 * @fpin - the pointer to the file we may pin (or is already pinned).
2358 *
2359 * This works similar to lock_page_or_retry in that it can drop the mmap_sem.
2360 * It differs in that it actually returns the page locked if it returns 1 and 0
2361 * if it couldn't lock the page. If we did have to drop the mmap_sem then fpin
2362 * will point to the pinned file and needs to be fput()'ed at a later point.
Linus Torvaldsef00e082009-06-16 15:31:25 -07002363 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002364static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2365 struct file **fpin)
2366{
2367 if (trylock_page(page))
2368 return 1;
2369
Linus Torvalds8b0f9fa2019-03-15 11:26:07 -07002370 /*
2371 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2372 * the mmap_sem still held. That's how FAULT_FLAG_RETRY_NOWAIT
2373 * is supposed to work. We have way too many special cases..
2374 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002375 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2376 return 0;
2377
2378 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2379 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2380 if (__lock_page_killable(page)) {
2381 /*
2382 * We didn't have the right flags to drop the mmap_sem,
2383 * but all fault_handlers only check for fatal signals
2384 * if we return VM_FAULT_RETRY, so we need to drop the
2385 * mmap_sem here and return 0 if we don't have a fpin.
2386 */
2387 if (*fpin == NULL)
2388 up_read(&vmf->vma->vm_mm->mmap_sem);
2389 return 0;
2390 }
2391 } else
2392 __lock_page(page);
2393 return 1;
2394}
2395
2396
2397/*
2398 * Synchronous readahead happens when we don't even find a page in the page
2399 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2400 * to drop the mmap sem we return the file that was pinned in order for us to do
2401 * that. If we didn't pin a file then we return NULL. The file that is
2402 * returned needs to be fput()'ed when we're done with it.
2403 */
2404static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002405{
Josef Bacik2a1180f2019-03-13 11:44:18 -07002406 struct file *file = vmf->vma->vm_file;
2407 struct file_ra_state *ra = &file->f_ra;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002408 struct address_space *mapping = file->f_mapping;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002409 struct file *fpin = NULL;
Josef Bacik2a1180f2019-03-13 11:44:18 -07002410 pgoff_t offset = vmf->pgoff;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002411
2412 /* If we don't want any read-ahead, don't bother */
Josef Bacik2a1180f2019-03-13 11:44:18 -07002413 if (vmf->vma->vm_flags & VM_RAND_READ)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002414 return fpin;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002415 if (!ra->ra_pages)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002416 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002417
Josef Bacik2a1180f2019-03-13 11:44:18 -07002418 if (vmf->vma->vm_flags & VM_SEQ_READ) {
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002419 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07002420 page_cache_sync_readahead(mapping, ra, file, offset,
2421 ra->ra_pages);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002422 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002423 }
2424
Andi Kleen207d04b2011-05-24 17:12:29 -07002425 /* Avoid banging the cache line if not needed */
2426 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002427 ra->mmap_miss++;
2428
2429 /*
2430 * Do we miss much more than hit in this file? If so,
2431 * stop bothering with read-ahead. It will only hurt.
2432 */
2433 if (ra->mmap_miss > MMAP_LOTSAMISS)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002434 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002435
Wu Fengguangd30a1102009-06-16 15:31:30 -07002436 /*
2437 * mmap read-around
2438 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002439 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
Roman Gushchin600e19a2015-11-05 18:47:08 -08002440 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2441 ra->size = ra->ra_pages;
2442 ra->async_size = ra->ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002443 ra_submit(ra, mapping, file);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002444 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002445}
2446
2447/*
2448 * Asynchronous readahead happens when we find the page and PG_readahead,
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002449 * so we want to possibly extend the readahead further. We return the file that
2450 * was pinned if we have to drop the mmap_sem in order to do IO.
Linus Torvaldsef00e082009-06-16 15:31:25 -07002451 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002452static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2453 struct page *page)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002454{
Josef Bacik2a1180f2019-03-13 11:44:18 -07002455 struct file *file = vmf->vma->vm_file;
2456 struct file_ra_state *ra = &file->f_ra;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002457 struct address_space *mapping = file->f_mapping;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002458 struct file *fpin = NULL;
Josef Bacik2a1180f2019-03-13 11:44:18 -07002459 pgoff_t offset = vmf->pgoff;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002460
2461 /* If we don't want any read-ahead, don't bother */
Josef Bacik2a1180f2019-03-13 11:44:18 -07002462 if (vmf->vma->vm_flags & VM_RAND_READ)
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002463 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002464 if (ra->mmap_miss > 0)
2465 ra->mmap_miss--;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002466 if (PageReadahead(page)) {
2467 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
Wu Fengguang2fad6f52009-06-16 15:31:29 -07002468 page_cache_async_readahead(mapping, ra, file,
2469 page, offset, ra->ra_pages);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002470 }
2471 return fpin;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002472}
2473
Randy Dunlap485bb992006-06-23 02:03:49 -07002474/**
Nick Piggin54cb8822007-07-19 01:46:59 -07002475 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07002476 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07002477 *
Nick Piggin54cb8822007-07-19 01:46:59 -07002478 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 * mapped memory region to read in file data during a page fault.
2480 *
2481 * The goto's are kind of ugly, but this streamlines the normal case of having
2482 * it in the page cache, and handles the special cases reasonably without
2483 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002484 *
2485 * vma->vm_mm->mmap_sem must be held on entry.
2486 *
Yang Shia4985832019-07-11 20:55:29 -07002487 * If our return value has VM_FAULT_RETRY set, it's because the mmap_sem
2488 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002489 *
2490 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2491 * has not been released.
2492 *
2493 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Mike Rapoporta862f682019-03-05 15:48:42 -08002494 *
2495 * Return: bitwise-OR of %VM_FAULT_ codes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 */
Souptick Joarder2bcd6452018-06-07 17:08:00 -07002497vm_fault_t filemap_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498{
2499 int error;
Dave Jiang11bac802017-02-24 14:56:41 -08002500 struct file *file = vmf->vma->vm_file;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002501 struct file *fpin = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 struct address_space *mapping = file->f_mapping;
2503 struct file_ra_state *ra = &file->f_ra;
2504 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002505 pgoff_t offset = vmf->pgoff;
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002506 pgoff_t max_off;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 struct page *page;
Souptick Joarder2bcd6452018-06-07 17:08:00 -07002508 vm_fault_t ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002510 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2511 if (unlikely(offset >= max_off))
Linus Torvalds5307cc12007-10-31 09:19:46 -07002512 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002515 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002517 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002518 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07002520 * We found the page, so try async readahead before
2521 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 */
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002523 fpin = do_async_mmap_readahead(vmf, page);
Shaohua Li45cac652012-10-08 16:32:19 -07002524 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07002525 /* No page in the page cache at all */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002526 count_vm_event(PGMAJFAULT);
Roman Gushchin22621852017-07-06 15:40:25 -07002527 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002528 ret = VM_FAULT_MAJOR;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002529 fpin = do_sync_mmap_readahead(vmf);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002530retry_find:
Josef Bacika75d4c32019-03-13 11:44:14 -07002531 page = pagecache_get_page(mapping, offset,
2532 FGP_CREAT|FGP_FOR_MMAP,
2533 vmf->gfp_mask);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002534 if (!page) {
2535 if (fpin)
2536 goto out_retry;
Josef Bacika75d4c32019-03-13 11:44:14 -07002537 return vmf_error(-ENOMEM);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002538 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 }
2540
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002541 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2542 goto out_retry;
Michel Lespinasseb522c942010-10-26 14:21:56 -07002543
2544 /* Did it get truncated? */
Song Liu585e5a72019-09-23 15:37:44 -07002545 if (unlikely(compound_head(page)->mapping != mapping)) {
Michel Lespinasseb522c942010-10-26 14:21:56 -07002546 unlock_page(page);
2547 put_page(page);
2548 goto retry_find;
2549 }
Song Liu520e5ba2019-09-23 15:37:50 -07002550 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07002551
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 /*
Nick Piggind00806b2007-07-19 01:46:57 -07002553 * We have a locked page in the page cache, now we need to check
2554 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 */
Nick Piggind00806b2007-07-19 01:46:57 -07002556 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 goto page_not_uptodate;
2558
Linus Torvaldsef00e082009-06-16 15:31:25 -07002559 /*
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002560 * We've made it this far and we had to drop our mmap_sem, now is the
2561 * time to return to the upper layer and have it re-find the vma and
2562 * redo the fault.
2563 */
2564 if (fpin) {
2565 unlock_page(page);
2566 goto out_retry;
2567 }
2568
2569 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07002570 * Found the page and have a reference on it.
2571 * We must recheck i_size under page lock.
2572 */
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002573 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2574 if (unlikely(offset >= max_off)) {
Nick Piggind00806b2007-07-19 01:46:57 -07002575 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002576 put_page(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07002577 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07002578 }
2579
Nick Piggind0217ac2007-07-19 01:47:03 -07002580 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07002581 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 /*
2585 * Umm, take care of errors if the page isn't up-to-date.
2586 * Try to re-read it _once_. We do this synchronously,
2587 * because there really aren't any performance issues here
2588 * and we need to check for errors.
2589 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 ClearPageError(page);
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002591 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
Zach Brown994fc28c2005-12-15 14:28:17 -08002592 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07002593 if (!error) {
2594 wait_on_page_locked(page);
2595 if (!PageUptodate(page))
2596 error = -EIO;
2597 }
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002598 if (fpin)
2599 goto out_retry;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002600 put_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002601
2602 if (!error || error == AOP_TRUNCATED_PAGE)
2603 goto retry_find;
2604
2605 /* Things didn't work out. Return zero to tell the mm layer so. */
2606 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002607 return VM_FAULT_SIGBUS;
Josef Bacik6b4c9f42019-03-13 11:44:22 -07002608
2609out_retry:
2610 /*
2611 * We dropped the mmap_sem, we need to return to the fault handler to
2612 * re-find the vma and come back and find our hopefully still populated
2613 * page.
2614 */
2615 if (page)
2616 put_page(page);
2617 if (fpin)
2618 fput(fpin);
2619 return ret | VM_FAULT_RETRY;
Nick Piggin54cb8822007-07-19 01:46:59 -07002620}
2621EXPORT_SYMBOL(filemap_fault);
2622
Jan Kara82b0f8c2016-12-14 15:06:58 -08002623void filemap_map_pages(struct vm_fault *vmf,
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002624 pgoff_t start_pgoff, pgoff_t end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002625{
Jan Kara82b0f8c2016-12-14 15:06:58 -08002626 struct file *file = vmf->vma->vm_file;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002627 struct address_space *mapping = file->f_mapping;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002628 pgoff_t last_pgoff = start_pgoff;
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002629 unsigned long max_idx;
Matthew Wilcox070e8072018-05-17 00:08:30 -04002630 XA_STATE(xas, &mapping->i_pages, start_pgoff);
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002631 struct page *page;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002632
2633 rcu_read_lock();
Matthew Wilcox070e8072018-05-17 00:08:30 -04002634 xas_for_each(&xas, page, end_pgoff) {
2635 if (xas_retry(&xas, page))
2636 continue;
2637 if (xa_is_value(page))
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002638 goto next;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002639
Michal Hockoe0975b22018-12-28 00:38:36 -08002640 /*
2641 * Check for a locked page first, as a speculative
2642 * reference may adversely influence page migration.
2643 */
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002644 if (PageLocked(page))
Michal Hockoe0975b22018-12-28 00:38:36 -08002645 goto next;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002646 if (!page_cache_get_speculative(page))
Matthew Wilcox070e8072018-05-17 00:08:30 -04002647 goto next;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002648
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002649 /* Has the page moved or been split? */
Matthew Wilcox070e8072018-05-17 00:08:30 -04002650 if (unlikely(page != xas_reload(&xas)))
2651 goto skip;
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -07002652 page = find_subpage(page, xas.xa_index);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002653
2654 if (!PageUptodate(page) ||
2655 PageReadahead(page) ||
2656 PageHWPoison(page))
2657 goto skip;
2658 if (!trylock_page(page))
2659 goto skip;
2660
2661 if (page->mapping != mapping || !PageUptodate(page))
2662 goto unlock;
2663
Matthew Wilcox9ab25942017-05-03 14:53:29 -07002664 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2665 if (page->index >= max_idx)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002666 goto unlock;
2667
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002668 if (file->f_ra.mmap_miss > 0)
2669 file->f_ra.mmap_miss--;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002670
Matthew Wilcox070e8072018-05-17 00:08:30 -04002671 vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
Jan Kara82b0f8c2016-12-14 15:06:58 -08002672 if (vmf->pte)
Matthew Wilcox070e8072018-05-17 00:08:30 -04002673 vmf->pte += xas.xa_index - last_pgoff;
2674 last_pgoff = xas.xa_index;
Jan Kara82b0f8c2016-12-14 15:06:58 -08002675 if (alloc_set_pte(vmf, NULL, page))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002676 goto unlock;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002677 unlock_page(page);
2678 goto next;
2679unlock:
2680 unlock_page(page);
2681skip:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002682 put_page(page);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002683next:
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002684 /* Huge page is mapped? No need to proceed. */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002685 if (pmd_trans_huge(*vmf->pmd))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002686 break;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002687 }
2688 rcu_read_unlock();
2689}
2690EXPORT_SYMBOL(filemap_map_pages);
2691
Souptick Joarder2bcd6452018-06-07 17:08:00 -07002692vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
Jan Kara4fcf1c62012-06-12 16:20:29 +02002693{
2694 struct page *page = vmf->page;
Dave Jiang11bac802017-02-24 14:56:41 -08002695 struct inode *inode = file_inode(vmf->vma->vm_file);
Souptick Joarder2bcd6452018-06-07 17:08:00 -07002696 vm_fault_t ret = VM_FAULT_LOCKED;
Jan Kara4fcf1c62012-06-12 16:20:29 +02002697
Jan Kara14da9202012-06-12 16:20:37 +02002698 sb_start_pagefault(inode->i_sb);
Dave Jiang11bac802017-02-24 14:56:41 -08002699 file_update_time(vmf->vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002700 lock_page(page);
2701 if (page->mapping != inode->i_mapping) {
2702 unlock_page(page);
2703 ret = VM_FAULT_NOPAGE;
2704 goto out;
2705 }
Jan Kara14da9202012-06-12 16:20:37 +02002706 /*
2707 * We mark the page dirty already here so that when freeze is in
2708 * progress, we are guaranteed that writeback during freezing will
2709 * see the dirty page and writeprotect it again.
2710 */
2711 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002712 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002713out:
Jan Kara14da9202012-06-12 16:20:37 +02002714 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002715 return ret;
2716}
Jan Kara4fcf1c62012-06-12 16:20:29 +02002717
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04002718const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002719 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002720 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002721 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722};
2723
2724/* This is used for a general mmap of a disk file */
2725
2726int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2727{
2728 struct address_space *mapping = file->f_mapping;
2729
2730 if (!mapping->a_ops->readpage)
2731 return -ENOEXEC;
2732 file_accessed(file);
2733 vma->vm_ops = &generic_file_vm_ops;
2734 return 0;
2735}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736
2737/*
2738 * This is for filesystems which do not implement ->writepage.
2739 */
2740int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2741{
2742 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2743 return -EINVAL;
2744 return generic_file_mmap(file, vma);
2745}
2746#else
Souptick Joarder4b96a372018-10-26 15:04:03 -07002747vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
Arnd Bergmann453972282018-04-13 15:35:27 -07002748{
Souptick Joarder4b96a372018-10-26 15:04:03 -07002749 return VM_FAULT_SIGBUS;
Arnd Bergmann453972282018-04-13 15:35:27 -07002750}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2752{
2753 return -ENOSYS;
2754}
2755int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2756{
2757 return -ENOSYS;
2758}
2759#endif /* CONFIG_MMU */
2760
Arnd Bergmann453972282018-04-13 15:35:27 -07002761EXPORT_SYMBOL(filemap_page_mkwrite);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762EXPORT_SYMBOL(generic_file_mmap);
2763EXPORT_SYMBOL(generic_file_readonly_mmap);
2764
Sasha Levin67f9fd92014-04-03 14:48:18 -07002765static struct page *wait_on_page_read(struct page *page)
2766{
2767 if (!IS_ERR(page)) {
2768 wait_on_page_locked(page);
2769 if (!PageUptodate(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002770 put_page(page);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002771 page = ERR_PTR(-EIO);
2772 }
2773 }
2774 return page;
2775}
2776
Mel Gorman32b63522016-03-15 14:55:36 -07002777static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002778 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002779 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002780 void *data,
2781 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782{
Nick Piggineb2be182007-10-16 01:24:57 -07002783 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 int err;
2785repeat:
2786 page = find_get_page(mapping, index);
2787 if (!page) {
Mel Gorman453f85d2017-11-15 17:38:03 -08002788 page = __page_cache_alloc(gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002789 if (!page)
2790 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002791 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002792 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002793 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002794 if (err == -EEXIST)
2795 goto repeat;
Matthew Wilcox22ecdb42017-12-04 04:02:00 -05002796 /* Presumably ENOMEM for xarray node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 return ERR_PTR(err);
2798 }
Mel Gorman32b63522016-03-15 14:55:36 -07002799
2800filler:
Christoph Hellwig6c45b452019-07-11 20:55:20 -07002801 if (filler)
2802 err = filler(data, page);
2803 else
2804 err = mapping->a_ops->readpage(data, page);
2805
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806 if (err < 0) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002807 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002808 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 }
Mel Gorman32b63522016-03-15 14:55:36 -07002810
2811 page = wait_on_page_read(page);
2812 if (IS_ERR(page))
2813 return page;
2814 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 if (PageUptodate(page))
2817 goto out;
2818
Mel Gormanebded022016-03-15 14:55:39 -07002819 /*
2820 * Page is not up to date and may be locked due one of the following
2821 * case a: Page is being filled and the page lock is held
2822 * case b: Read/write error clearing the page uptodate status
2823 * case c: Truncation in progress (page locked)
2824 * case d: Reclaim in progress
2825 *
2826 * Case a, the page will be up to date when the page is unlocked.
2827 * There is no need to serialise on the page lock here as the page
2828 * is pinned so the lock gives no additional protection. Even if the
2829 * the page is truncated, the data is still valid if PageUptodate as
2830 * it's a race vs truncate race.
2831 * Case b, the page will not be up to date
2832 * Case c, the page may be truncated but in itself, the data may still
2833 * be valid after IO completes as it's a read vs truncate race. The
2834 * operation must restart if the page is not uptodate on unlock but
2835 * otherwise serialising on page lock to stabilise the mapping gives
2836 * no additional guarantees to the caller as the page lock is
2837 * released before return.
2838 * Case d, similar to truncation. If reclaim holds the page lock, it
2839 * will be a race with remove_mapping that determines if the mapping
2840 * is valid on unlock but otherwise the data is valid and there is
2841 * no need to serialise with page lock.
2842 *
2843 * As the page lock gives no additional guarantee, we optimistically
2844 * wait on the page to be unlocked and check if it's up to date and
2845 * use the page if it is. Otherwise, the page lock is required to
2846 * distinguish between the different cases. The motivation is that we
2847 * avoid spurious serialisations and wakeups when multiple processes
2848 * wait on the same page for IO to complete.
2849 */
2850 wait_on_page_locked(page);
2851 if (PageUptodate(page))
2852 goto out;
2853
2854 /* Distinguish between all the cases under the safety of the lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 lock_page(page);
Mel Gormanebded022016-03-15 14:55:39 -07002856
2857 /* Case c or d, restart the operation */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 if (!page->mapping) {
2859 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002860 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002861 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 }
Mel Gormanebded022016-03-15 14:55:39 -07002863
2864 /* Someone else locked and filled the page in a very small window */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865 if (PageUptodate(page)) {
2866 unlock_page(page);
2867 goto out;
2868 }
Mel Gorman32b63522016-03-15 14:55:36 -07002869 goto filler;
2870
David Howellsc855ff32007-05-09 13:42:20 +01002871out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002872 mark_page_accessed(page);
2873 return page;
2874}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002875
2876/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002877 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002878 * @mapping: the page's address_space
2879 * @index: the page index
2880 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002881 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002882 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002883 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002884 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002885 *
2886 * If the page does not get brought uptodate, return -EIO.
Mike Rapoporta862f682019-03-05 15:48:42 -08002887 *
2888 * Return: up to date page on success, ERR_PTR() on failure.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002889 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002890struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002891 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002892 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002893 void *data)
2894{
Christoph Hellwigd322a8e2019-07-11 20:55:17 -07002895 return do_read_cache_page(mapping, index, filler, data,
2896 mapping_gfp_mask(mapping));
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002897}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002898EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002899
2900/**
2901 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2902 * @mapping: the page's address_space
2903 * @index: the page index
2904 * @gfp: the page allocator flags to use if allocating
2905 *
2906 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002907 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002908 *
2909 * If the page does not get brought uptodate, return -EIO.
Mike Rapoporta862f682019-03-05 15:48:42 -08002910 *
2911 * Return: up to date page on success, ERR_PTR() on failure.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002912 */
2913struct page *read_cache_page_gfp(struct address_space *mapping,
2914 pgoff_t index,
2915 gfp_t gfp)
2916{
Christoph Hellwig6c45b452019-07-11 20:55:20 -07002917 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002918}
2919EXPORT_SYMBOL(read_cache_page_gfp);
2920
Nick Piggin2f718ff2007-10-16 01:24:59 -07002921/*
Darrick J. Wong9fd91a90cb2018-10-30 10:40:46 +11002922 * Don't operate on ranges the page cache doesn't support, and don't exceed the
2923 * LFS limits. If pos is under the limit it becomes a short access. If it
2924 * exceeds the limit we return -EFBIG.
2925 */
Darrick J. Wong9fd91a90cb2018-10-30 10:40:46 +11002926static int generic_write_check_limits(struct file *file, loff_t pos,
2927 loff_t *count)
2928{
Amir Goldstein646955c2019-06-05 08:04:48 -07002929 struct inode *inode = file->f_mapping->host;
2930 loff_t max_size = inode->i_sb->s_maxbytes;
Darrick J. Wong9fd91a90cb2018-10-30 10:40:46 +11002931 loff_t limit = rlimit(RLIMIT_FSIZE);
2932
2933 if (limit != RLIM_INFINITY) {
2934 if (pos >= limit) {
2935 send_sig(SIGXFSZ, current, 0);
2936 return -EFBIG;
2937 }
2938 *count = min(*count, limit - pos);
2939 }
2940
Amir Goldstein646955c2019-06-05 08:04:48 -07002941 if (!(file->f_flags & O_LARGEFILE))
2942 max_size = MAX_NON_LFS;
2943
2944 if (unlikely(pos >= max_size))
2945 return -EFBIG;
2946
2947 *count = min(*count, max_size - pos);
2948
2949 return 0;
Darrick J. Wong9fd91a90cb2018-10-30 10:40:46 +11002950}
2951
2952/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953 * Performs necessary checks before doing a write
2954 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002955 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 * Returns appropriate error code that caller should return or
2957 * zero in case that write should be allowed.
2958 */
Al Viro3309dd02015-04-09 12:55:47 -04002959inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960{
Al Viro3309dd02015-04-09 12:55:47 -04002961 struct file *file = iocb->ki_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 struct inode *inode = file->f_mapping->host;
Darrick J. Wong9fd91a90cb2018-10-30 10:40:46 +11002963 loff_t count;
2964 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965
Darrick J. Wongdc617f22019-08-20 07:55:16 -07002966 if (IS_SWAPFILE(inode))
2967 return -ETXTBSY;
2968
Al Viro3309dd02015-04-09 12:55:47 -04002969 if (!iov_iter_count(from))
2970 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971
Al Viro0fa6b002015-04-04 04:05:48 -04002972 /* FIXME: this is for backwards compatibility with 2.4 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002973 if (iocb->ki_flags & IOCB_APPEND)
Al Viro3309dd02015-04-09 12:55:47 -04002974 iocb->ki_pos = i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05002976 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2977 return -EINVAL;
2978
Darrick J. Wong9fd91a90cb2018-10-30 10:40:46 +11002979 count = iov_iter_count(from);
2980 ret = generic_write_check_limits(file, iocb->ki_pos, &count);
2981 if (ret)
2982 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983
Darrick J. Wong9fd91a90cb2018-10-30 10:40:46 +11002984 iov_iter_truncate(from, count);
Al Viro3309dd02015-04-09 12:55:47 -04002985 return iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986}
2987EXPORT_SYMBOL(generic_write_checks);
2988
Darrick J. Wong1383a7e2018-10-30 10:40:31 +11002989/*
2990 * Performs necessary checks before doing a clone.
2991 *
Amir Goldstein646955c2019-06-05 08:04:48 -07002992 * Can adjust amount of bytes to clone via @req_count argument.
Darrick J. Wong1383a7e2018-10-30 10:40:31 +11002993 * Returns appropriate error code that caller should return or
2994 * zero in case the clone should be allowed.
2995 */
2996int generic_remap_checks(struct file *file_in, loff_t pos_in,
2997 struct file *file_out, loff_t pos_out,
Darrick J. Wong42ec3d42018-10-30 10:41:49 +11002998 loff_t *req_count, unsigned int remap_flags)
Darrick J. Wong1383a7e2018-10-30 10:40:31 +11002999{
3000 struct inode *inode_in = file_in->f_mapping->host;
3001 struct inode *inode_out = file_out->f_mapping->host;
3002 uint64_t count = *req_count;
3003 uint64_t bcount;
3004 loff_t size_in, size_out;
3005 loff_t bs = inode_out->i_sb->s_blocksize;
Darrick J. Wong9fd91a90cb2018-10-30 10:40:46 +11003006 int ret;
Darrick J. Wong1383a7e2018-10-30 10:40:31 +11003007
3008 /* The start of both ranges must be aligned to an fs block. */
3009 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
3010 return -EINVAL;
3011
3012 /* Ensure offsets don't wrap. */
3013 if (pos_in + count < pos_in || pos_out + count < pos_out)
3014 return -EINVAL;
3015
3016 size_in = i_size_read(inode_in);
3017 size_out = i_size_read(inode_out);
3018
3019 /* Dedupe requires both ranges to be within EOF. */
Darrick J. Wong3d281932018-10-30 10:41:34 +11003020 if ((remap_flags & REMAP_FILE_DEDUP) &&
Darrick J. Wong1383a7e2018-10-30 10:40:31 +11003021 (pos_in >= size_in || pos_in + count > size_in ||
3022 pos_out >= size_out || pos_out + count > size_out))
3023 return -EINVAL;
3024
3025 /* Ensure the infile range is within the infile. */
3026 if (pos_in >= size_in)
3027 return -EINVAL;
3028 count = min(count, size_in - (uint64_t)pos_in);
3029
Darrick J. Wong9fd91a90cb2018-10-30 10:40:46 +11003030 ret = generic_write_check_limits(file_out, pos_out, &count);
3031 if (ret)
3032 return ret;
3033
Darrick J. Wong1383a7e2018-10-30 10:40:31 +11003034 /*
3035 * If the user wanted us to link to the infile's EOF, round up to the
3036 * next block boundary for this check.
3037 *
3038 * Otherwise, make sure the count is also block-aligned, having
3039 * already confirmed the starting offsets' block alignment.
3040 */
3041 if (pos_in + count == size_in) {
3042 bcount = ALIGN(size_in, bs) - pos_in;
3043 } else {
3044 if (!IS_ALIGNED(count, bs))
Darrick J. Wongeca36542018-10-30 10:42:10 +11003045 count = ALIGN_DOWN(count, bs);
Darrick J. Wong1383a7e2018-10-30 10:40:31 +11003046 bcount = count;
3047 }
3048
3049 /* Don't allow overlapped cloning within the same file. */
3050 if (inode_in == inode_out &&
3051 pos_out + bcount > pos_in &&
3052 pos_out < pos_in + bcount)
3053 return -EINVAL;
3054
Darrick J. Wongeca36542018-10-30 10:42:10 +11003055 /*
3056 * We shortened the request but the caller can't deal with that, so
3057 * bounce the request back to userspace.
3058 */
3059 if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
Darrick J. Wong1383a7e2018-10-30 10:40:31 +11003060 return -EINVAL;
3061
Darrick J. Wongeca36542018-10-30 10:42:10 +11003062 *req_count = count;
Darrick J. Wong1383a7e2018-10-30 10:40:31 +11003063 return 0;
3064}
3065
Amir Goldsteina3171352019-06-05 08:04:48 -07003066
3067/*
3068 * Performs common checks before doing a file copy/clone
3069 * from @file_in to @file_out.
3070 */
3071int generic_file_rw_checks(struct file *file_in, struct file *file_out)
3072{
3073 struct inode *inode_in = file_inode(file_in);
3074 struct inode *inode_out = file_inode(file_out);
3075
3076 /* Don't copy dirs, pipes, sockets... */
3077 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
3078 return -EISDIR;
3079 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
3080 return -EINVAL;
3081
3082 if (!(file_in->f_mode & FMODE_READ) ||
3083 !(file_out->f_mode & FMODE_WRITE) ||
3084 (file_out->f_flags & O_APPEND))
3085 return -EBADF;
3086
3087 return 0;
3088}
3089
Amir Goldstein96e6e8f2019-06-05 08:04:49 -07003090/*
3091 * Performs necessary checks before doing a file copy
3092 *
3093 * Can adjust amount of bytes to copy via @req_count argument.
3094 * Returns appropriate error code that caller should return or
3095 * zero in case the copy should be allowed.
3096 */
3097int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
3098 struct file *file_out, loff_t pos_out,
3099 size_t *req_count, unsigned int flags)
3100{
3101 struct inode *inode_in = file_inode(file_in);
3102 struct inode *inode_out = file_inode(file_out);
3103 uint64_t count = *req_count;
3104 loff_t size_in;
3105 int ret;
3106
3107 ret = generic_file_rw_checks(file_in, file_out);
3108 if (ret)
3109 return ret;
3110
3111 /* Don't touch certain kinds of inodes */
3112 if (IS_IMMUTABLE(inode_out))
3113 return -EPERM;
3114
3115 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
3116 return -ETXTBSY;
3117
3118 /* Ensure offsets don't wrap. */
3119 if (pos_in + count < pos_in || pos_out + count < pos_out)
3120 return -EOVERFLOW;
3121
3122 /* Shorten the copy to EOF */
3123 size_in = i_size_read(inode_in);
3124 if (pos_in >= size_in)
3125 count = 0;
3126 else
3127 count = min(count, size_in - (uint64_t)pos_in);
3128
3129 ret = generic_write_check_limits(file_out, pos_out, &count);
3130 if (ret)
3131 return ret;
3132
3133 /* Don't allow overlapped copying within the same file. */
3134 if (inode_in == inode_out &&
3135 pos_out + count > pos_in &&
3136 pos_out < pos_in + count)
3137 return -EINVAL;
3138
3139 *req_count = count;
3140 return 0;
3141}
3142
Nick Pigginafddba42007-10-16 01:25:01 -07003143int pagecache_write_begin(struct file *file, struct address_space *mapping,
3144 loff_t pos, unsigned len, unsigned flags,
3145 struct page **pagep, void **fsdata)
3146{
3147 const struct address_space_operations *aops = mapping->a_ops;
3148
Nick Piggin4e02ed42008-10-29 14:00:55 -07003149 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07003150 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07003151}
3152EXPORT_SYMBOL(pagecache_write_begin);
3153
3154int pagecache_write_end(struct file *file, struct address_space *mapping,
3155 loff_t pos, unsigned len, unsigned copied,
3156 struct page *page, void *fsdata)
3157{
3158 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07003159
Nick Piggin4e02ed42008-10-29 14:00:55 -07003160 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07003161}
3162EXPORT_SYMBOL(pagecache_write_end);
3163
Konstantin Khlebnikova92853b2019-11-30 17:49:44 -08003164/*
3165 * Warn about a page cache invalidation failure during a direct I/O write.
3166 */
3167void dio_warn_stale_pagecache(struct file *filp)
3168{
3169 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3170 char pathname[128];
3171 struct inode *inode = file_inode(filp);
3172 char *path;
3173
3174 errseq_set(&inode->i_mapping->wb_err, -EIO);
3175 if (__ratelimit(&_rs)) {
3176 path = file_path(filp, pathname, sizeof(pathname));
3177 if (IS_ERR(path))
3178 path = "(unknown)";
3179 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3180 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3181 current->comm);
3182 }
3183}
3184
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185ssize_t
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07003186generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187{
3188 struct file *file = iocb->ki_filp;
3189 struct address_space *mapping = file->f_mapping;
3190 struct inode *inode = mapping->host;
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07003191 loff_t pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003192 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07003193 size_t write_len;
3194 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003195
Al Viro0c949332014-03-22 06:51:37 -04003196 write_len = iov_iter_count(from);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003197 end = (pos + write_len - 1) >> PAGE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07003198
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05003199 if (iocb->ki_flags & IOCB_NOWAIT) {
3200 /* If there are pages to writeback, return */
3201 if (filemap_range_has_page(inode->i_mapping, pos,
zhengbin35f12f02019-03-05 15:44:21 -08003202 pos + write_len - 1))
Goldwyn Rodrigues6be96d32017-06-20 07:05:44 -05003203 return -EAGAIN;
3204 } else {
3205 written = filemap_write_and_wait_range(mapping, pos,
3206 pos + write_len - 1);
3207 if (written)
3208 goto out;
3209 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07003210
3211 /*
3212 * After a write we want buffered reads to be sure to go to disk to get
3213 * the new data. We invalidate clean cached page from the region we're
3214 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07003215 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07003216 */
Andrey Ryabinin55635ba2017-05-03 14:55:59 -07003217 written = invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003218 pos >> PAGE_SHIFT, end);
Andrey Ryabinin55635ba2017-05-03 14:55:59 -07003219 /*
3220 * If a page can not be invalidated, return 0 to fall back
3221 * to buffered write.
3222 */
3223 if (written) {
3224 if (written == -EBUSY)
3225 return 0;
3226 goto out;
Christoph Hellwiga969e902008-07-23 21:27:04 -07003227 }
3228
Al Viro639a93a52017-04-13 14:10:15 -04003229 written = mapping->a_ops->direct_IO(iocb, from);
Christoph Hellwiga969e902008-07-23 21:27:04 -07003230
3231 /*
3232 * Finally, try again to invalidate clean pages which might have been
3233 * cached by non-direct readahead, or faulted in by get_user_pages()
3234 * if the source of the write was an mmap'ed region of the file
3235 * we're writing. Either one is a pretty crazy thing to do,
3236 * so we don't support it 100%. If this invalidation
3237 * fails, tough, the write still worked...
Lukas Czerner332391a2017-09-21 08:16:29 -06003238 *
3239 * Most of the time we do not need this since dio_complete() will do
3240 * the invalidation for us. However there are some file systems that
3241 * do not end up with dio_complete() being called, so let's not break
Konstantin Khlebnikov80c1fe92019-11-30 17:49:41 -08003242 * them by removing it completely.
3243 *
Konstantin Khlebnikov9266a142019-11-30 17:49:47 -08003244 * Noticeable example is a blkdev_direct_IO().
3245 *
Konstantin Khlebnikov80c1fe92019-11-30 17:49:41 -08003246 * Skip invalidation for async writes or if mapping has no pages.
Christoph Hellwiga969e902008-07-23 21:27:04 -07003247 */
Konstantin Khlebnikov9266a142019-11-30 17:49:47 -08003248 if (written > 0 && mapping->nrpages &&
3249 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3250 dio_warn_stale_pagecache(file);
Christoph Hellwiga969e902008-07-23 21:27:04 -07003251
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07003253 pos += written;
Al Viro639a93a52017-04-13 14:10:15 -04003254 write_len -= written;
Namhyung Kim01166512010-10-26 14:21:58 -07003255 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3256 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257 mark_inode_dirty(inode);
3258 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05003259 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260 }
Al Viro639a93a52017-04-13 14:10:15 -04003261 iov_iter_revert(from, write_len - iov_iter_count(from));
Christoph Hellwiga969e902008-07-23 21:27:04 -07003262out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 return written;
3264}
3265EXPORT_SYMBOL(generic_file_direct_write);
3266
Nick Piggineb2be182007-10-16 01:24:57 -07003267/*
3268 * Find or create a page at the given pagecache position. Return the locked
3269 * page. This function is specifically for buffered writes.
3270 */
Nick Piggin54566b22009-01-04 12:00:53 -08003271struct page *grab_cache_page_write_begin(struct address_space *mapping,
3272 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07003273{
Nick Piggineb2be182007-10-16 01:24:57 -07003274 struct page *page;
Johannes Weinerbbddabe2016-05-20 16:56:28 -07003275 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08003276
Nick Piggin54566b22009-01-04 12:00:53 -08003277 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07003278 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07003279
Mel Gorman2457aec2014-06-04 16:10:31 -07003280 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01003281 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07003282 if (page)
3283 wait_for_stable_page(page);
3284
Nick Piggineb2be182007-10-16 01:24:57 -07003285 return page;
3286}
Nick Piggin54566b22009-01-04 12:00:53 -08003287EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07003288
Al Viro3b93f912014-02-11 21:34:08 -05003289ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07003290 struct iov_iter *i, loff_t pos)
3291{
3292 struct address_space *mapping = file->f_mapping;
3293 const struct address_space_operations *a_ops = mapping->a_ops;
3294 long status = 0;
3295 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07003296 unsigned int flags = 0;
3297
Nick Pigginafddba42007-10-16 01:25:01 -07003298 do {
3299 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07003300 unsigned long offset; /* Offset into pagecache page */
3301 unsigned long bytes; /* Bytes to write to page */
3302 size_t copied; /* Bytes copied from user */
3303 void *fsdata;
3304
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003305 offset = (pos & (PAGE_SIZE - 1));
3306 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07003307 iov_iter_count(i));
3308
3309again:
Linus Torvalds00a3d662015-10-07 08:32:38 +01003310 /*
3311 * Bring in the user page that we will copy from _first_.
3312 * Otherwise there's a nasty deadlock on copying from the
3313 * same page as we're writing to, without it being marked
3314 * up-to-date.
3315 *
3316 * Not only is this an optimisation, but it is also required
3317 * to check that the address is actually valid, when atomic
3318 * usercopies are used, below.
3319 */
3320 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3321 status = -EFAULT;
3322 break;
3323 }
3324
Jan Kara296291c2015-10-22 13:32:21 -07003325 if (fatal_signal_pending(current)) {
3326 status = -EINTR;
3327 break;
3328 }
3329
Nick Piggin674b8922007-10-16 01:25:03 -07003330 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07003331 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07003332 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07003333 break;
3334
anfei zhou931e80e2010-02-02 13:44:02 -08003335 if (mapping_writably_mapped(mapping))
3336 flush_dcache_page(page);
Linus Torvalds00a3d662015-10-07 08:32:38 +01003337
Nick Pigginafddba42007-10-16 01:25:01 -07003338 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07003339 flush_dcache_page(page);
3340
3341 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3342 page, fsdata);
3343 if (unlikely(status < 0))
3344 break;
3345 copied = status;
3346
3347 cond_resched();
3348
Nick Piggin124d3b72008-02-02 15:01:17 +01003349 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07003350 if (unlikely(copied == 0)) {
3351 /*
3352 * If we were unable to copy any data at all, we must
3353 * fall back to a single segment length write.
3354 *
3355 * If we didn't fallback here, we could livelock
3356 * because not all segments in the iov can be copied at
3357 * once without a pagefault.
3358 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003359 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07003360 iov_iter_single_seg_count(i));
3361 goto again;
3362 }
Nick Pigginafddba42007-10-16 01:25:01 -07003363 pos += copied;
3364 written += copied;
3365
3366 balance_dirty_pages_ratelimited(mapping);
Nick Pigginafddba42007-10-16 01:25:01 -07003367 } while (iov_iter_count(i));
3368
3369 return written ? written : status;
3370}
Al Viro3b93f912014-02-11 21:34:08 -05003371EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003372
Jan Karae4dd9de2009-08-17 18:10:06 +02003373/**
Al Viro81742022014-04-03 03:17:43 -04003374 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02003375 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04003376 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02003377 *
3378 * This function does all the work needed for actually writing data to a
3379 * file. It does all basic checks, removes SUID from the file, updates
3380 * modification times and calls proper subroutines depending on whether we
3381 * do direct IO or a standard buffered write.
3382 *
3383 * It expects i_mutex to be grabbed unless we work on a block device or similar
3384 * object which does not need locking at all.
3385 *
3386 * This function does *not* take care of syncing data in case of O_SYNC write.
3387 * A caller has to handle it. This is mainly due to the fact that we want to
3388 * avoid syncing under i_mutex.
Mike Rapoporta862f682019-03-05 15:48:42 -08003389 *
3390 * Return:
3391 * * number of bytes written, even for truncated writes
3392 * * negative error code if no data has been written at all
Jan Karae4dd9de2009-08-17 18:10:06 +02003393 */
Al Viro81742022014-04-03 03:17:43 -04003394ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003395{
3396 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003397 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003398 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05003399 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05003401 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01003404 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02003405 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003406 if (err)
3407 goto out;
3408
Josef Bacikc3b2da32012-03-26 09:59:21 -04003409 err = file_update_time(file);
3410 if (err)
3411 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003412
Al Viro2ba48ce2015-04-09 13:52:01 -04003413 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04003414 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003415
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07003416 written = generic_file_direct_write(iocb, from);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08003417 /*
3418 * If the write stopped short of completing, fall back to
3419 * buffered writes. Some filesystems do this for writes to
3420 * holes, for example. For DAX files, a buffered write will
3421 * not succeed (even if it did, DAX does not handle dirty
3422 * page-cache pages correctly).
3423 */
Al Viro0b8def92015-04-07 10:22:53 -04003424 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05003426
Al Viro0b8def92015-04-07 10:22:53 -04003427 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003428 /*
Al Viro3b93f912014-02-11 21:34:08 -05003429 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003430 * then we want to return the number of bytes which were
3431 * direct-written, or the error code if that was zero. Note
3432 * that this differs from normal direct-io semantics, which
3433 * will return -EFOO even if some bytes were written.
3434 */
Al Viro60bb4522014-08-08 12:39:16 -04003435 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05003436 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003437 goto out;
3438 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003439 /*
3440 * We need to ensure that the page cache pages are written to
3441 * disk and invalidated to preserve the expected O_DIRECT
3442 * semantics.
3443 */
Al Viro3b93f912014-02-11 21:34:08 -05003444 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04003445 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003446 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04003447 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05003448 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003449 invalidate_mapping_pages(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003450 pos >> PAGE_SHIFT,
3451 endbyte >> PAGE_SHIFT);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003452 } else {
3453 /*
3454 * We don't know how much we wrote, so just return
3455 * the number of bytes which were direct-written
3456 */
3457 }
3458 } else {
Al Viro0b8def92015-04-07 10:22:53 -04003459 written = generic_perform_write(file, from, iocb->ki_pos);
3460 if (likely(written > 0))
3461 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07003462 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003463out:
3464 current->backing_dev_info = NULL;
3465 return written ? written : err;
3466}
Al Viro81742022014-04-03 03:17:43 -04003467EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003468
Jan Karae4dd9de2009-08-17 18:10:06 +02003469/**
Al Viro81742022014-04-03 03:17:43 -04003470 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02003471 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04003472 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02003473 *
Al Viro81742022014-04-03 03:17:43 -04003474 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02003475 * filesystems. It takes care of syncing the file in case of O_SYNC file
3476 * and acquires i_mutex as needed.
Mike Rapoporta862f682019-03-05 15:48:42 -08003477 * Return:
3478 * * negative error code if no data has been written at all of
3479 * vfs_fsync_range() failed for a synchronous write
3480 * * number of bytes written, even for truncated writes
Jan Karae4dd9de2009-08-17 18:10:06 +02003481 */
Al Viro81742022014-04-03 03:17:43 -04003482ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003483{
3484 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02003485 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003486 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487
Al Viro59551022016-01-22 15:40:57 -05003488 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -04003489 ret = generic_write_checks(iocb, from);
3490 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04003491 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -05003492 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493
Christoph Hellwige2592212016-04-07 08:52:01 -07003494 if (ret > 0)
3495 ret = generic_write_sync(iocb, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003496 return ret;
3497}
Al Viro81742022014-04-03 03:17:43 -04003498EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499
David Howellscf9a2ae2006-08-29 19:05:54 +01003500/**
3501 * try_to_release_page() - release old fs-specific metadata on a page
3502 *
3503 * @page: the page which the kernel is trying to free
3504 * @gfp_mask: memory allocation flags (and I/O mode)
3505 *
3506 * The address_space is to try to release any data against the page
Mike Rapoporta862f682019-03-05 15:48:42 -08003507 * (presumably at page->private).
David Howellscf9a2ae2006-08-29 19:05:54 +01003508 *
David Howells266cf652009-04-03 16:42:36 +01003509 * This may also be called if PG_fscache is set on a page, indicating that the
3510 * page is known to the local caching routines.
3511 *
David Howellscf9a2ae2006-08-29 19:05:54 +01003512 * The @gfp_mask argument specifies whether I/O may be performed to release
Mel Gorman71baba42015-11-06 16:28:28 -08003513 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01003514 *
Mike Rapoporta862f682019-03-05 15:48:42 -08003515 * Return: %1 if the release was successful, otherwise return zero.
David Howellscf9a2ae2006-08-29 19:05:54 +01003516 */
3517int try_to_release_page(struct page *page, gfp_t gfp_mask)
3518{
3519 struct address_space * const mapping = page->mapping;
3520
3521 BUG_ON(!PageLocked(page));
3522 if (PageWriteback(page))
3523 return 0;
3524
3525 if (mapping && mapping->a_ops->releasepage)
3526 return mapping->a_ops->releasepage(page, gfp_mask);
3527 return try_to_free_buffers(page);
3528}
3529
3530EXPORT_SYMBOL(try_to_release_page);