blob: 3083ded98b15f32b68c58d744a5c3abb07d94c96 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040012#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/compiler.h>
Ross Zwislerf9fe48b2016-01-22 15:10:40 -080014#include <linux/dax.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070016#include <linux/uaccess.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080032#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070033#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Johannes Weiner00501b52014-08-08 14:19:20 -070034#include <linux/hugetlb.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060036#include <linux/cleancache.h>
Kirill A. Shutemovf1820362014-04-07 15:37:19 -070037#include <linux/rmap.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080038#include "internal.h"
39
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070040#define CREATE_TRACE_POINTS
41#include <trace/events/filemap.h>
42
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 * FIXME: remove all knowledge of the buffer layer from the core VM
45 */
Jan Kara148f9482009-08-17 19:52:36 +020046#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <asm/mman.h>
49
50/*
51 * Shared mappings implemented 30.11.1994. It's not fully working yet,
52 * though.
53 *
54 * Shared mappings now work. 15.8.1995 Bruno.
55 *
56 * finished 'unifying' the page and buffer cache and SMP-threaded the
57 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
58 *
59 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
60 */
61
62/*
63 * Lock ordering:
64 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080065 * ->i_mmap_rwsem (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070067 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080070 * ->i_mutex
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080071 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 *
73 * ->mmap_sem
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080074 * ->i_mmap_rwsem
Hugh Dickinsb8072f02005-10-29 18:16:41 -070075 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 *
78 * ->mmap_sem
79 * ->lock_page (access_process_vm)
80 *
Al Viroccad2362014-02-11 22:36:48 -050081 * ->i_mutex (generic_perform_write)
Nick Piggin82591e62006-10-19 23:29:10 -070082 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060084 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110085 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * ->mapping->tree_lock (__sync_single_inode)
87 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -080088 * ->i_mmap_rwsem
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 * ->anon_vma.lock (vma_adjust)
90 *
91 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070092 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070094 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070095 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
Mel Gormana52633d2016-07-28 15:45:28 -070098 * ->zone_lru_lock(zone) (follow_page->mark_page_accessed)
99 * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600102 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700104 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600105 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100106 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
108 *
Davidlohr Buesoc8c06ef2014-12-12 16:54:24 -0800109 * ->i_mmap_rwsem
Andi Kleen9a3c5312012-03-21 16:34:09 -0700110 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 */
112
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700113static void page_cache_tree_delete(struct address_space *mapping,
114 struct page *page, void *shadow)
115{
Johannes Weiner449dd692014-04-03 14:47:56 -0700116 struct radix_tree_node *node;
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700117 int i, nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700118
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700119 VM_BUG_ON_PAGE(!PageLocked(page), page);
120 VM_BUG_ON_PAGE(PageTail(page), page);
121 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
Johannes Weiner449dd692014-04-03 14:47:56 -0700122
123 if (shadow) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700124 mapping->nrexceptional += nr;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700125 /*
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800126 * Make sure the nrexceptional update is committed before
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700127 * the nrpages update so that final truncate racing
128 * with reclaim does not see both counters 0 at the
129 * same time and miss a shadow entry.
130 */
131 smp_wmb();
Johannes Weiner449dd692014-04-03 14:47:56 -0700132 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700133 mapping->nrpages -= nr;
Johannes Weiner449dd692014-04-03 14:47:56 -0700134
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700135 for (i = 0; i < nr; i++) {
136 node = radix_tree_replace_clear_tags(&mapping->page_tree,
137 page->index + i, shadow);
138 if (!node) {
139 VM_BUG_ON_PAGE(nr != 1, page);
Johannes Weiner449dd692014-04-03 14:47:56 -0700140 return;
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700141 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700142
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700143 workingset_node_pages_dec(node);
144 if (shadow)
145 workingset_node_shadows_inc(node);
146 else
147 if (__radix_tree_delete_node(&mapping->page_tree, node))
148 continue;
149
150 /*
151 * Track node that only contains shadow entries. DAX mappings
152 * contain no shadow entries and may contain other exceptional
153 * entries so skip those.
154 *
155 * Avoid acquiring the list_lru lock if already tracked.
156 * The list_empty() test is safe as node->private_list is
157 * protected by mapping->tree_lock.
158 */
159 if (!dax_mapping(mapping) && !workingset_node_pages(node) &&
160 list_empty(&node->private_list)) {
161 node->private_data = mapping;
162 list_lru_add(&workingset_shadow_nodes,
163 &node->private_list);
164 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700165 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700166}
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168/*
Minchan Kime64a7822011-03-22 16:32:44 -0700169 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 * sure the page is locked and that nobody else uses it - or that usage
Johannes Weinerfdf1cdb2016-03-15 14:57:25 -0700171 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 */
Johannes Weiner62cccb82016-03-15 14:57:22 -0700173void __delete_from_page_cache(struct page *page, void *shadow)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174{
175 struct address_space *mapping = page->mapping;
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700176 int nr = hpage_nr_pages(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700178 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600179 /*
180 * if we're uptodate, flush out into the cleancache, otherwise
181 * invalidate any existing cleancache entries. We can't leave
182 * stale data around in the cleancache once our page is gone
183 */
184 if (PageUptodate(page) && PageMappedToDisk(page))
185 cleancache_put_page(page);
186 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400187 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600188
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700189 VM_BUG_ON_PAGE(PageTail(page), page);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800190 VM_BUG_ON_PAGE(page_mapped(page), page);
191 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
192 int mapcount;
193
194 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
195 current->comm, page_to_pfn(page));
196 dump_page(page, "still mapped when deleted");
197 dump_stack();
198 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
199
200 mapcount = page_mapcount(page);
201 if (mapping_exiting(mapping) &&
202 page_count(page) >= mapcount + 2) {
203 /*
204 * All vmas have already been torn down, so it's
205 * a good bet that actually the page is unmapped,
206 * and we'd prefer not to leak it: if we're wrong,
207 * some other bad page check should catch it later.
208 */
209 page_mapcount_reset(page);
Joonsoo Kim6d061f92016-05-19 17:10:46 -0700210 page_ref_sub(page, mapcount);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800211 }
212 }
213
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700214 page_cache_tree_delete(mapping, page, shadow);
215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700217 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700218
Michal Hocko4165b9b2015-06-24 16:57:24 -0700219 /* hugetlb pages do not participate in page cache accounting. */
220 if (!PageHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700221 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700222 if (PageSwapBacked(page)) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700223 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700224 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700225 __dec_node_page_state(page, NR_SHMEM_THPS);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700226 } else {
227 VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
228 }
Linus Torvalds3a692792007-12-19 14:05:13 -0800229
230 /*
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700231 * At this point page must be either written or cleaned by truncate.
232 * Dirty page here signals a bug and loss of unwritten data.
Linus Torvalds3a692792007-12-19 14:05:13 -0800233 *
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700234 * This fixes dirty accounting after removing the page entirely but
235 * leaves PageDirty set: it has no effect for truncated page and
236 * anyway will be cleared before returning page into buddy allocator.
Linus Torvalds3a692792007-12-19 14:05:13 -0800237 */
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700238 if (WARN_ON_ONCE(PageDirty(page)))
Johannes Weiner62cccb82016-03-15 14:57:22 -0700239 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240}
241
Minchan Kim702cfbf2011-03-22 16:32:43 -0700242/**
243 * delete_from_page_cache - delete page from page cache
244 * @page: the page which the kernel is trying to remove from page cache
245 *
246 * This must be called only on pages that have been verified to be in the page
247 * cache and locked. It will never put the page into the free list, the caller
248 * has a reference on the page.
249 */
250void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700252 struct address_space *mapping = page_mapping(page);
Greg Thelenc4843a72015-05-22 17:13:16 -0400253 unsigned long flags;
Linus Torvalds6072d132010-12-01 13:35:19 -0500254 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
Matt Mackallcd7619d2005-05-01 08:59:01 -0700256 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
Linus Torvalds6072d132010-12-01 13:35:19 -0500258 freepage = mapping->a_ops->freepage;
Greg Thelenc4843a72015-05-22 17:13:16 -0400259
Greg Thelenc4843a72015-05-22 17:13:16 -0400260 spin_lock_irqsave(&mapping->tree_lock, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700261 __delete_from_page_cache(page, NULL);
Greg Thelenc4843a72015-05-22 17:13:16 -0400262 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500263
264 if (freepage)
265 freepage(page);
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700266
267 if (PageTransHuge(page) && !PageHuge(page)) {
268 page_ref_sub(page, HPAGE_PMD_NR);
269 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
270 } else {
271 put_page(page);
272 }
Minchan Kim97cecb52011-03-22 16:30:53 -0700273}
274EXPORT_SYMBOL(delete_from_page_cache);
275
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200276int filemap_check_errors(struct address_space *mapping)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700277{
278 int ret = 0;
279 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700280 if (test_bit(AS_ENOSPC, &mapping->flags) &&
281 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700282 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700283 if (test_bit(AS_EIO, &mapping->flags) &&
284 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700285 ret = -EIO;
286 return ret;
287}
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200288EXPORT_SYMBOL(filemap_check_errors);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700289
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700291 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700292 * @mapping: address space structure to write
293 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800294 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700295 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700297 * Start writeback against all of a mapping's dirty pages that lie
298 * within the byte offsets <start, end> inclusive.
299 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700301 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 * these two operations is that if a dirty page/buffer is encountered, it must
303 * be waited upon, and not just skipped over.
304 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800305int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
306 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
308 int ret;
309 struct writeback_control wbc = {
310 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800311 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700312 .range_start = start,
313 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 };
315
316 if (!mapping_cap_writeback_dirty(mapping))
317 return 0;
318
Tejun Heob16b1de2015-06-02 08:39:48 -0600319 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 ret = do_writepages(mapping, &wbc);
Tejun Heob16b1de2015-06-02 08:39:48 -0600321 wbc_detach_inode(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 return ret;
323}
324
325static inline int __filemap_fdatawrite(struct address_space *mapping,
326 int sync_mode)
327{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700328 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329}
330
331int filemap_fdatawrite(struct address_space *mapping)
332{
333 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
334}
335EXPORT_SYMBOL(filemap_fdatawrite);
336
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400337int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800338 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339{
340 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
341}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400342EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
Randy Dunlap485bb992006-06-23 02:03:49 -0700344/**
345 * filemap_flush - mostly a non-blocking flush
346 * @mapping: target address_space
347 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 * This is a mostly non-blocking flush. Not suitable for data-integrity
349 * purposes - I/O may not be started against all dirty pages.
350 */
351int filemap_flush(struct address_space *mapping)
352{
353 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
354}
355EXPORT_SYMBOL(filemap_flush);
356
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800357static int __filemap_fdatawait_range(struct address_space *mapping,
358 loff_t start_byte, loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300360 pgoff_t index = start_byte >> PAGE_SHIFT;
361 pgoff_t end = end_byte >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 struct pagevec pvec;
363 int nr_pages;
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800364 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200366 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700367 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
369 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 while ((index <= end) &&
371 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
372 PAGECACHE_TAG_WRITEBACK,
373 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
374 unsigned i;
375
376 for (i = 0; i < nr_pages; i++) {
377 struct page *page = pvec.pages[i];
378
379 /* until radix tree lookup accepts end_index */
380 if (page->index > end)
381 continue;
382
383 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800384 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 ret = -EIO;
386 }
387 pagevec_release(&pvec);
388 cond_resched();
389 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700390out:
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800391 return ret;
392}
393
394/**
395 * filemap_fdatawait_range - wait for writeback to complete
396 * @mapping: address space structure to wait for
397 * @start_byte: offset in bytes where the range starts
398 * @end_byte: offset in bytes where the range ends (inclusive)
399 *
400 * Walk the list of under-writeback pages of the given address space
401 * in the given range and wait for all of them. Check error status of
402 * the address space and return it.
403 *
404 * Since the error status of the address space is cleared by this function,
405 * callers are responsible for checking the return value and handling and/or
406 * reporting the error.
407 */
408int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
409 loff_t end_byte)
410{
411 int ret, ret2;
412
413 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700414 ret2 = filemap_check_errors(mapping);
415 if (!ret)
416 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
418 return ret;
419}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200420EXPORT_SYMBOL(filemap_fdatawait_range);
421
422/**
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800423 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
424 * @mapping: address space structure to wait for
425 *
426 * Walk the list of under-writeback pages of the given address space
427 * and wait for all of them. Unlike filemap_fdatawait(), this function
428 * does not clear error status of the address space.
429 *
430 * Use this function if callers don't handle errors themselves. Expected
431 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
432 * fsfreeze(8)
433 */
434void filemap_fdatawait_keep_errors(struct address_space *mapping)
435{
436 loff_t i_size = i_size_read(mapping->host);
437
438 if (i_size == 0)
439 return;
440
441 __filemap_fdatawait_range(mapping, 0, i_size - 1);
442}
443
444/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700445 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700447 *
448 * Walk the list of under-writeback pages of the given address space
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800449 * and wait for all of them. Check error status of the address space
450 * and return it.
451 *
452 * Since the error status of the address space is cleared by this function,
453 * callers are responsible for checking the return value and handling and/or
454 * reporting the error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 */
456int filemap_fdatawait(struct address_space *mapping)
457{
458 loff_t i_size = i_size_read(mapping->host);
459
460 if (i_size == 0)
461 return 0;
462
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200463 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464}
465EXPORT_SYMBOL(filemap_fdatawait);
466
467int filemap_write_and_wait(struct address_space *mapping)
468{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800469 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800471 if ((!dax_mapping(mapping) && mapping->nrpages) ||
472 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800473 err = filemap_fdatawrite(mapping);
474 /*
475 * Even if the above returned error, the pages may be
476 * written partially (e.g. -ENOSPC), so we wait for it.
477 * But the -EIO is special case, it may indicate the worst
478 * thing (e.g. bug) happened, so we avoid waiting for it.
479 */
480 if (err != -EIO) {
481 int err2 = filemap_fdatawait(mapping);
482 if (!err)
483 err = err2;
484 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700485 } else {
486 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800488 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800490EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Randy Dunlap485bb992006-06-23 02:03:49 -0700492/**
493 * filemap_write_and_wait_range - write out & wait on a file range
494 * @mapping: the address_space for the pages
495 * @lstart: offset in bytes where the range starts
496 * @lend: offset in bytes where the range ends (inclusive)
497 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800498 * Write out and wait upon file offsets lstart->lend, inclusive.
499 *
500 * Note that `lend' is inclusive (describes the last byte to be written) so
501 * that this function can be used to write to the very end-of-file (end = -1).
502 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503int filemap_write_and_wait_range(struct address_space *mapping,
504 loff_t lstart, loff_t lend)
505{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800506 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800508 if ((!dax_mapping(mapping) && mapping->nrpages) ||
509 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800510 err = __filemap_fdatawrite_range(mapping, lstart, lend,
511 WB_SYNC_ALL);
512 /* See comment of filemap_write_and_wait() */
513 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200514 int err2 = filemap_fdatawait_range(mapping,
515 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800516 if (!err)
517 err = err2;
518 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700519 } else {
520 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800522 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523}
Chris Masonf6995582009-04-15 13:22:37 -0400524EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
Randy Dunlap485bb992006-06-23 02:03:49 -0700526/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700527 * replace_page_cache_page - replace a pagecache page with a new one
528 * @old: page to be replaced
529 * @new: page to replace with
530 * @gfp_mask: allocation mode
531 *
532 * This function replaces a page in the pagecache with a new one. On
533 * success it acquires the pagecache reference for the new page and
534 * drops it for the old page. Both the old and new pages must be
535 * locked. This function does not add the new page to the LRU, the
536 * caller must do that.
537 *
538 * The remove + add is atomic. The only way this function can fail is
539 * memory allocation failure.
540 */
541int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
542{
543 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700544
Sasha Levin309381fea2014-01-23 15:52:54 -0800545 VM_BUG_ON_PAGE(!PageLocked(old), old);
546 VM_BUG_ON_PAGE(!PageLocked(new), new);
547 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700548
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700549 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
550 if (!error) {
551 struct address_space *mapping = old->mapping;
552 void (*freepage)(struct page *);
Greg Thelenc4843a72015-05-22 17:13:16 -0400553 unsigned long flags;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700554
555 pgoff_t offset = old->index;
556 freepage = mapping->a_ops->freepage;
557
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300558 get_page(new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700559 new->mapping = mapping;
560 new->index = offset;
561
Greg Thelenc4843a72015-05-22 17:13:16 -0400562 spin_lock_irqsave(&mapping->tree_lock, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700563 __delete_from_page_cache(old, NULL);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700564 error = radix_tree_insert(&mapping->page_tree, offset, new);
565 BUG_ON(error);
566 mapping->nrpages++;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700567
568 /*
569 * hugetlb pages do not participate in page cache accounting.
570 */
571 if (!PageHuge(new))
Mel Gorman11fb9982016-07-28 15:46:20 -0700572 __inc_node_page_state(new, NR_FILE_PAGES);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700573 if (PageSwapBacked(new))
Mel Gorman11fb9982016-07-28 15:46:20 -0700574 __inc_node_page_state(new, NR_SHMEM);
Greg Thelenc4843a72015-05-22 17:13:16 -0400575 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700576 mem_cgroup_migrate(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700577 radix_tree_preload_end();
578 if (freepage)
579 freepage(old);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300580 put_page(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700581 }
582
583 return error;
584}
585EXPORT_SYMBOL_GPL(replace_page_cache_page);
586
Johannes Weiner0cd61442014-04-03 14:47:46 -0700587static int page_cache_tree_insert(struct address_space *mapping,
Johannes Weinera5289102014-04-03 14:47:51 -0700588 struct page *page, void **shadowp)
Johannes Weiner0cd61442014-04-03 14:47:46 -0700589{
Johannes Weiner449dd692014-04-03 14:47:56 -0700590 struct radix_tree_node *node;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700591 void **slot;
592 int error;
593
Matthew Wilcoxe6145232016-03-17 14:21:54 -0700594 error = __radix_tree_create(&mapping->page_tree, page->index, 0,
Johannes Weiner449dd692014-04-03 14:47:56 -0700595 &node, &slot);
596 if (error)
597 return error;
598 if (*slot) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700599 void *p;
600
601 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
602 if (!radix_tree_exceptional_entry(p))
603 return -EEXIST;
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800604
Ross Zwislerf9fe48b2016-01-22 15:10:40 -0800605 mapping->nrexceptional--;
Jan Kara4f622932016-05-12 18:29:17 +0200606 if (!dax_mapping(mapping)) {
607 if (shadowp)
608 *shadowp = p;
609 if (node)
610 workingset_node_shadows_dec(node);
611 } else {
612 /* DAX can replace empty locked entry with a hole */
613 WARN_ON_ONCE(p !=
614 (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY |
615 RADIX_DAX_ENTRY_LOCK));
616 /* DAX accounts exceptional entries as normal pages */
617 if (node)
618 workingset_node_pages_dec(node);
Jan Karaac401cc2016-05-12 18:29:18 +0200619 /* Wakeup waiters for exceptional entry lock */
620 dax_wake_mapping_entry_waiter(mapping, page->index,
621 false);
Jan Kara4f622932016-05-12 18:29:17 +0200622 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700623 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700624 radix_tree_replace_slot(slot, page);
625 mapping->nrpages++;
626 if (node) {
627 workingset_node_pages_inc(node);
628 /*
629 * Don't track node that contains actual pages.
630 *
631 * Avoid acquiring the list_lru lock if already
632 * untracked. The list_empty() test is safe as
633 * node->private_list is protected by
634 * mapping->tree_lock.
635 */
636 if (!list_empty(&node->private_list))
637 list_lru_del(&workingset_shadow_nodes,
638 &node->private_list);
639 }
640 return 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700641}
642
Johannes Weinera5289102014-04-03 14:47:51 -0700643static int __add_to_page_cache_locked(struct page *page,
644 struct address_space *mapping,
645 pgoff_t offset, gfp_t gfp_mask,
646 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647{
Johannes Weiner00501b52014-08-08 14:19:20 -0700648 int huge = PageHuge(page);
649 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700650 int error;
651
Sasha Levin309381fea2014-01-23 15:52:54 -0800652 VM_BUG_ON_PAGE(!PageLocked(page), page);
653 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700654
Johannes Weiner00501b52014-08-08 14:19:20 -0700655 if (!huge) {
656 error = mem_cgroup_try_charge(page, current->mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800657 gfp_mask, &memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -0700658 if (error)
659 return error;
660 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Jan Kara5e4c0d972013-09-11 14:26:05 -0700662 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700663 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700664 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800665 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700666 return error;
667 }
668
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300669 get_page(page);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700670 page->mapping = mapping;
671 page->index = offset;
672
673 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700674 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700675 radix_tree_preload_end();
676 if (unlikely(error))
677 goto err_insert;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700678
679 /* hugetlb pages do not participate in page cache accounting. */
680 if (!huge)
Mel Gorman11fb9982016-07-28 15:46:20 -0700681 __inc_node_page_state(page, NR_FILE_PAGES);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700682 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700683 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800684 mem_cgroup_commit_charge(page, memcg, false, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700685 trace_mm_filemap_add_to_page_cache(page);
686 return 0;
687err_insert:
688 page->mapping = NULL;
689 /* Leave page->index set: truncation relies upon it */
690 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700691 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800692 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300693 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 return error;
695}
Johannes Weinera5289102014-04-03 14:47:51 -0700696
697/**
698 * add_to_page_cache_locked - add a locked page to the pagecache
699 * @page: page to add
700 * @mapping: the page's address_space
701 * @offset: page index
702 * @gfp_mask: page allocation mode
703 *
704 * This function is used to add a page to the pagecache. It must be locked.
705 * This function does not add the page to the LRU. The caller must do that.
706 */
707int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
708 pgoff_t offset, gfp_t gfp_mask)
709{
710 return __add_to_page_cache_locked(page, mapping, offset,
711 gfp_mask, NULL);
712}
Nick Piggine2867812008-07-25 19:45:30 -0700713EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
715int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400716 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717{
Johannes Weinera5289102014-04-03 14:47:51 -0700718 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700719 int ret;
720
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800721 __SetPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700722 ret = __add_to_page_cache_locked(page, mapping, offset,
723 gfp_mask, &shadow);
724 if (unlikely(ret))
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800725 __ClearPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700726 else {
727 /*
728 * The page might have been evicted from cache only
729 * recently, in which case it should be activated like
730 * any other repeatedly accessed page.
Rik van Rielf0281a02016-05-20 16:56:25 -0700731 * The exception is pages getting rewritten; evicting other
732 * data from the working set, only to cache data that will
733 * get overwritten with something else, is a waste of memory.
Johannes Weinera5289102014-04-03 14:47:51 -0700734 */
Rik van Rielf0281a02016-05-20 16:56:25 -0700735 if (!(gfp_mask & __GFP_WRITE) &&
736 shadow && workingset_refault(shadow)) {
Johannes Weinera5289102014-04-03 14:47:51 -0700737 SetPageActive(page);
738 workingset_activation(page);
739 } else
740 ClearPageActive(page);
741 lru_cache_add(page);
742 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 return ret;
744}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300745EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746
Paul Jackson44110fe2006-03-24 03:16:04 -0800747#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700748struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800749{
Miao Xiec0ff7452010-05-24 14:32:08 -0700750 int n;
751 struct page *page;
752
Paul Jackson44110fe2006-03-24 03:16:04 -0800753 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700754 unsigned int cpuset_mems_cookie;
755 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700756 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700757 n = cpuset_mem_spread_node();
Vlastimil Babka96db8002015-09-08 15:03:50 -0700758 page = __alloc_pages_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700759 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700760
Miao Xiec0ff7452010-05-24 14:32:08 -0700761 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800762 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700763 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800764}
Nick Piggin2ae88142006-10-28 10:38:23 -0700765EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800766#endif
767
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768/*
769 * In order to wait for pages to become available there must be
770 * waitqueues associated with pages. By using a hash table of
771 * waitqueues where the bucket discipline is to maintain all
772 * waiters on the same queue and wake all when any of the pages
773 * become available, and for the woken contexts to check to be
774 * sure the appropriate page became available, this saves space
775 * at a cost of "thundering herd" phenomena during rare hash
776 * collisions.
777 */
NeilBrowna4796e32014-09-24 11:28:32 +1000778wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779{
780 const struct zone *zone = page_zone(page);
781
782 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
783}
NeilBrowna4796e32014-09-24 11:28:32 +1000784EXPORT_SYMBOL(page_waitqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
Harvey Harrison920c7a52008-02-04 22:29:26 -0800786void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787{
788 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
789
790 if (test_bit(bit_nr, &page->flags))
NeilBrown74316202014-07-07 15:16:04 +1000791 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 TASK_UNINTERRUPTIBLE);
793}
794EXPORT_SYMBOL(wait_on_page_bit);
795
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700796int wait_on_page_bit_killable(struct page *page, int bit_nr)
797{
798 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
799
800 if (!test_bit(bit_nr, &page->flags))
801 return 0;
802
803 return __wait_on_bit(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000804 bit_wait_io, TASK_KILLABLE);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700805}
806
NeilBrowncbbce822014-09-25 13:55:19 +1000807int wait_on_page_bit_killable_timeout(struct page *page,
808 int bit_nr, unsigned long timeout)
809{
810 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
811
812 wait.key.timeout = jiffies + timeout;
813 if (!test_bit(bit_nr, &page->flags))
814 return 0;
815 return __wait_on_bit(page_waitqueue(page), &wait,
816 bit_wait_io_timeout, TASK_KILLABLE);
817}
818EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
819
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100821 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700822 * @page: Page defining the wait queue of interest
823 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100824 *
825 * Add an arbitrary @waiter to the wait queue for the nominated @page.
826 */
827void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
828{
829 wait_queue_head_t *q = page_waitqueue(page);
830 unsigned long flags;
831
832 spin_lock_irqsave(&q->lock, flags);
833 __add_wait_queue(q, waiter);
834 spin_unlock_irqrestore(&q->lock, flags);
835}
836EXPORT_SYMBOL_GPL(add_page_wait_queue);
837
838/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700839 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 * @page: the page
841 *
842 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
843 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +0900844 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
846 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700847 * The mb is necessary to enforce ordering between the clear_bit and the read
848 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800850void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800852 page = compound_head(page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800853 VM_BUG_ON_PAGE(!PageLocked(page), page);
Nick Piggin8413ac92008-10-18 20:26:59 -0700854 clear_bit_unlock(PG_locked, &page->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100855 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 wake_up_page(page, PG_locked);
857}
858EXPORT_SYMBOL(unlock_page);
859
Randy Dunlap485bb992006-06-23 02:03:49 -0700860/**
861 * end_page_writeback - end writeback against a page
862 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 */
864void end_page_writeback(struct page *page)
865{
Mel Gorman888cf2d2014-06-04 16:10:34 -0700866 /*
867 * TestClearPageReclaim could be used here but it is an atomic
868 * operation and overkill in this particular case. Failing to
869 * shuffle a page marked for immediate reclaim is too mild to
870 * justify taking an atomic operation penalty at the end of
871 * ever page writeback.
872 */
873 if (PageReclaim(page)) {
874 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700875 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -0700876 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700877
878 if (!test_clear_page_writeback(page))
879 BUG();
880
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100881 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 wake_up_page(page, PG_writeback);
883}
884EXPORT_SYMBOL(end_page_writeback);
885
Matthew Wilcox57d99842014-06-04 16:07:45 -0700886/*
887 * After completing I/O on a page, call this routine to update the page
888 * flags appropriately
889 */
890void page_endio(struct page *page, int rw, int err)
891{
892 if (rw == READ) {
893 if (!err) {
894 SetPageUptodate(page);
895 } else {
896 ClearPageUptodate(page);
897 SetPageError(page);
898 }
899 unlock_page(page);
900 } else { /* rw == WRITE */
901 if (err) {
902 SetPageError(page);
903 if (page->mapping)
904 mapping_set_error(page->mapping, err);
905 }
906 end_page_writeback(page);
907 }
908}
909EXPORT_SYMBOL_GPL(page_endio);
910
Randy Dunlap485bb992006-06-23 02:03:49 -0700911/**
912 * __lock_page - get a lock on the page, assuming we need to sleep to get it
913 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800915void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800917 struct page *page_head = compound_head(page);
918 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800920 __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 TASK_UNINTERRUPTIBLE);
922}
923EXPORT_SYMBOL(__lock_page);
924
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800925int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500926{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800927 struct page *page_head = compound_head(page);
928 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500929
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800930 return __wait_on_bit_lock(page_waitqueue(page_head), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000931 bit_wait_io, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500932}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300933EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500934
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700935/*
936 * Return values:
937 * 1 - page is locked; mmap_sem is still held.
938 * 0 - page is not locked.
939 * mmap_sem has been released (up_read()), unless flags had both
940 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
941 * which case mmap_sem is still held.
942 *
943 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
944 * with the page locked and the mmap_sem unperturbed.
945 */
Michel Lespinassed065bd82010-10-26 14:21:57 -0700946int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
947 unsigned int flags)
948{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700949 if (flags & FAULT_FLAG_ALLOW_RETRY) {
950 /*
951 * CAUTION! In this case, mmap_sem is not released
952 * even though return 0.
953 */
954 if (flags & FAULT_FLAG_RETRY_NOWAIT)
955 return 0;
956
957 up_read(&mm->mmap_sem);
958 if (flags & FAULT_FLAG_KILLABLE)
959 wait_on_page_locked_killable(page);
960 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700961 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700962 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700963 } else {
964 if (flags & FAULT_FLAG_KILLABLE) {
965 int ret;
966
967 ret = __lock_page_killable(page);
968 if (ret) {
969 up_read(&mm->mmap_sem);
970 return 0;
971 }
972 } else
973 __lock_page(page);
974 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700975 }
976}
977
Randy Dunlap485bb992006-06-23 02:03:49 -0700978/**
Johannes Weinere7b563b2014-04-03 14:47:44 -0700979 * page_cache_next_hole - find the next hole (not-present entry)
980 * @mapping: mapping
981 * @index: index
982 * @max_scan: maximum range to search
983 *
984 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
985 * lowest indexed hole.
986 *
987 * Returns: the index of the hole if found, otherwise returns an index
988 * outside of the set specified (in which case 'return - index >=
989 * max_scan' will be true). In rare cases of index wrap-around, 0 will
990 * be returned.
991 *
992 * page_cache_next_hole may be called under rcu_read_lock. However,
993 * like radix_tree_gang_lookup, this will not atomically search a
994 * snapshot of the tree at a single point in time. For example, if a
995 * hole is created at index 5, then subsequently a hole is created at
996 * index 10, page_cache_next_hole covering both indexes may return 10
997 * if called under rcu_read_lock.
998 */
999pgoff_t page_cache_next_hole(struct address_space *mapping,
1000 pgoff_t index, unsigned long max_scan)
1001{
1002 unsigned long i;
1003
1004 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001005 struct page *page;
1006
1007 page = radix_tree_lookup(&mapping->page_tree, index);
1008 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001009 break;
1010 index++;
1011 if (index == 0)
1012 break;
1013 }
1014
1015 return index;
1016}
1017EXPORT_SYMBOL(page_cache_next_hole);
1018
1019/**
1020 * page_cache_prev_hole - find the prev hole (not-present entry)
1021 * @mapping: mapping
1022 * @index: index
1023 * @max_scan: maximum range to search
1024 *
1025 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1026 * the first hole.
1027 *
1028 * Returns: the index of the hole if found, otherwise returns an index
1029 * outside of the set specified (in which case 'index - return >=
1030 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1031 * will be returned.
1032 *
1033 * page_cache_prev_hole may be called under rcu_read_lock. However,
1034 * like radix_tree_gang_lookup, this will not atomically search a
1035 * snapshot of the tree at a single point in time. For example, if a
1036 * hole is created at index 10, then subsequently a hole is created at
1037 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1038 * called under rcu_read_lock.
1039 */
1040pgoff_t page_cache_prev_hole(struct address_space *mapping,
1041 pgoff_t index, unsigned long max_scan)
1042{
1043 unsigned long i;
1044
1045 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001046 struct page *page;
1047
1048 page = radix_tree_lookup(&mapping->page_tree, index);
1049 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001050 break;
1051 index--;
1052 if (index == ULONG_MAX)
1053 break;
1054 }
1055
1056 return index;
1057}
1058EXPORT_SYMBOL(page_cache_prev_hole);
1059
1060/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001061 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -07001062 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -07001063 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -07001064 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001065 * Looks up the page cache slot at @mapping & @offset. If there is a
1066 * page cache page, it is returned with an increased refcount.
1067 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001068 * If the slot holds a shadow entry of a previously evicted page, or a
1069 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001070 *
1071 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 */
Johannes Weiner0cd61442014-04-03 14:47:46 -07001073struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074{
Nick Piggina60637c2008-07-25 19:45:31 -07001075 void **pagep;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001076 struct page *head, *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
Nick Piggina60637c2008-07-25 19:45:31 -07001078 rcu_read_lock();
1079repeat:
1080 page = NULL;
1081 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1082 if (pagep) {
1083 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -08001084 if (unlikely(!page))
1085 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001086 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001087 if (radix_tree_deref_retry(page))
1088 goto repeat;
1089 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001090 * A shadow entry of a recently evicted page,
1091 * or a swap entry from shmem/tmpfs. Return
1092 * it without attempting to raise page count.
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001093 */
1094 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001095 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001096
1097 head = compound_head(page);
1098 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001099 goto repeat;
1100
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001101 /* The page was split under us? */
1102 if (compound_head(page) != head) {
1103 put_page(head);
1104 goto repeat;
1105 }
1106
Nick Piggina60637c2008-07-25 19:45:31 -07001107 /*
1108 * Has the page moved?
1109 * This is part of the lockless pagecache protocol. See
1110 * include/linux/pagemap.h for details.
1111 */
1112 if (unlikely(page != *pagep)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001113 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001114 goto repeat;
1115 }
1116 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001117out:
Nick Piggina60637c2008-07-25 19:45:31 -07001118 rcu_read_unlock();
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 return page;
1121}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001122EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Randy Dunlap485bb992006-06-23 02:03:49 -07001124/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001125 * find_lock_entry - locate, pin and lock a page cache entry
1126 * @mapping: the address_space to search
1127 * @offset: the page cache index
1128 *
1129 * Looks up the page cache slot at @mapping & @offset. If there is a
1130 * page cache page, it is returned locked and with an increased
1131 * refcount.
1132 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001133 * If the slot holds a shadow entry of a previously evicted page, or a
1134 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001135 *
1136 * Otherwise, %NULL is returned.
1137 *
1138 * find_lock_entry() may sleep.
1139 */
1140struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141{
1142 struct page *page;
1143
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001145 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001146 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001147 lock_page(page);
1148 /* Has the page been truncated? */
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001149 if (unlikely(page_mapping(page) != mapping)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001150 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001151 put_page(page);
Nick Piggina60637c2008-07-25 19:45:31 -07001152 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001154 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 return page;
1157}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001158EXPORT_SYMBOL(find_lock_entry);
1159
1160/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001161 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001162 * @mapping: the address_space to search
1163 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001164 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001165 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001166 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001167 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001168 *
Randy Dunlap75325182014-07-30 16:08:37 -07001169 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001170 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001171 * FGP_ACCESSED: the page will be marked accessed
1172 * FGP_LOCK: Page is return locked
1173 * FGP_CREAT: If page is not present then a new page is allocated using
Michal Hocko45f87de2014-12-29 20:30:35 +01001174 * @gfp_mask and added to the page cache and the VM's LRU
1175 * list. The page is returned locked and with an increased
1176 * refcount. Otherwise, %NULL is returned.
Mel Gorman2457aec2014-06-04 16:10:31 -07001177 *
1178 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1179 * if the GFP flags specified for FGP_CREAT are atomic.
1180 *
1181 * If there is a page cache page, it is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001182 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001183struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001184 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185{
Nick Piggineb2be182007-10-16 01:24:57 -07001186 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001189 page = find_get_entry(mapping, offset);
1190 if (radix_tree_exceptional_entry(page))
1191 page = NULL;
1192 if (!page)
1193 goto no_page;
1194
1195 if (fgp_flags & FGP_LOCK) {
1196 if (fgp_flags & FGP_NOWAIT) {
1197 if (!trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001198 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001199 return NULL;
1200 }
1201 } else {
1202 lock_page(page);
1203 }
1204
1205 /* Has the page been truncated? */
1206 if (unlikely(page->mapping != mapping)) {
1207 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001208 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001209 goto repeat;
1210 }
1211 VM_BUG_ON_PAGE(page->index != offset, page);
1212 }
1213
1214 if (page && (fgp_flags & FGP_ACCESSED))
1215 mark_page_accessed(page);
1216
1217no_page:
1218 if (!page && (fgp_flags & FGP_CREAT)) {
1219 int err;
1220 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001221 gfp_mask |= __GFP_WRITE;
1222 if (fgp_flags & FGP_NOFS)
1223 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001224
Michal Hocko45f87de2014-12-29 20:30:35 +01001225 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001226 if (!page)
1227 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001228
1229 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1230 fgp_flags |= FGP_LOCK;
1231
Hugh Dickinseb39d612014-08-06 16:06:43 -07001232 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001233 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001234 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001235
Michal Hocko45f87de2014-12-29 20:30:35 +01001236 err = add_to_page_cache_lru(page, mapping, offset,
1237 gfp_mask & GFP_RECLAIM_MASK);
Nick Piggineb2be182007-10-16 01:24:57 -07001238 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001239 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07001240 page = NULL;
1241 if (err == -EEXIST)
1242 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001245
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 return page;
1247}
Mel Gorman2457aec2014-06-04 16:10:31 -07001248EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
1250/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001251 * find_get_entries - gang pagecache lookup
1252 * @mapping: The address_space to search
1253 * @start: The starting page cache index
1254 * @nr_entries: The maximum number of entries
1255 * @entries: Where the resulting entries are placed
1256 * @indices: The cache indices corresponding to the entries in @entries
1257 *
1258 * find_get_entries() will search for and return a group of up to
1259 * @nr_entries entries in the mapping. The entries are placed at
1260 * @entries. find_get_entries() takes a reference against any actual
1261 * pages it returns.
1262 *
1263 * The search returns a group of mapping-contiguous page cache entries
1264 * with ascending indexes. There may be holes in the indices due to
1265 * not-present pages.
1266 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001267 * Any shadow entries of evicted pages, or swap entries from
1268 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001269 *
1270 * find_get_entries() returns the number of pages and shadow entries
1271 * which were found.
1272 */
1273unsigned find_get_entries(struct address_space *mapping,
1274 pgoff_t start, unsigned int nr_entries,
1275 struct page **entries, pgoff_t *indices)
1276{
1277 void **slot;
1278 unsigned int ret = 0;
1279 struct radix_tree_iter iter;
1280
1281 if (!nr_entries)
1282 return 0;
1283
1284 rcu_read_lock();
Johannes Weiner0cd61442014-04-03 14:47:46 -07001285 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001286 struct page *head, *page;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001287repeat:
1288 page = radix_tree_deref_slot(slot);
1289 if (unlikely(!page))
1290 continue;
1291 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001292 if (radix_tree_deref_retry(page)) {
1293 slot = radix_tree_iter_retry(&iter);
1294 continue;
1295 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001296 /*
Ross Zwislerf9fe48b2016-01-22 15:10:40 -08001297 * A shadow entry of a recently evicted page, a swap
1298 * entry from shmem/tmpfs or a DAX entry. Return it
1299 * without attempting to raise page count.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001300 */
1301 goto export;
1302 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001303
1304 head = compound_head(page);
1305 if (!page_cache_get_speculative(head))
Johannes Weiner0cd61442014-04-03 14:47:46 -07001306 goto repeat;
1307
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001308 /* The page was split under us? */
1309 if (compound_head(page) != head) {
1310 put_page(head);
1311 goto repeat;
1312 }
1313
Johannes Weiner0cd61442014-04-03 14:47:46 -07001314 /* Has the page moved? */
1315 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001316 put_page(head);
Johannes Weiner0cd61442014-04-03 14:47:46 -07001317 goto repeat;
1318 }
1319export:
1320 indices[ret] = iter.index;
1321 entries[ret] = page;
1322 if (++ret == nr_entries)
1323 break;
1324 }
1325 rcu_read_unlock();
1326 return ret;
1327}
1328
1329/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 * find_get_pages - gang pagecache lookup
1331 * @mapping: The address_space to search
1332 * @start: The starting page index
1333 * @nr_pages: The maximum number of pages
1334 * @pages: Where the resulting pages are placed
1335 *
1336 * find_get_pages() will search for and return a group of up to
1337 * @nr_pages pages in the mapping. The pages are placed at @pages.
1338 * find_get_pages() takes a reference against the returned pages.
1339 *
1340 * The search returns a group of mapping-contiguous pages with ascending
1341 * indexes. There may be holes in the indices due to not-present pages.
1342 *
1343 * find_get_pages() returns the number of pages which were found.
1344 */
1345unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1346 unsigned int nr_pages, struct page **pages)
1347{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001348 struct radix_tree_iter iter;
1349 void **slot;
1350 unsigned ret = 0;
1351
1352 if (unlikely(!nr_pages))
1353 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
Nick Piggina60637c2008-07-25 19:45:31 -07001355 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001356 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001357 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001358repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001359 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001360 if (unlikely(!page))
1361 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001362
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001363 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001364 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001365 slot = radix_tree_iter_retry(&iter);
1366 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001367 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001368 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001369 * A shadow entry of a recently evicted page,
1370 * or a swap entry from shmem/tmpfs. Skip
1371 * over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001372 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001373 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001374 }
Nick Piggina60637c2008-07-25 19:45:31 -07001375
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001376 head = compound_head(page);
1377 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001378 goto repeat;
1379
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001380 /* The page was split under us? */
1381 if (compound_head(page) != head) {
1382 put_page(head);
1383 goto repeat;
1384 }
1385
Nick Piggina60637c2008-07-25 19:45:31 -07001386 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001387 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001388 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001389 goto repeat;
1390 }
1391
1392 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001393 if (++ret == nr_pages)
1394 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001395 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001396
Nick Piggina60637c2008-07-25 19:45:31 -07001397 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 return ret;
1399}
1400
Jens Axboeebf43502006-04-27 08:46:01 +02001401/**
1402 * find_get_pages_contig - gang contiguous pagecache lookup
1403 * @mapping: The address_space to search
1404 * @index: The starting page index
1405 * @nr_pages: The maximum number of pages
1406 * @pages: Where the resulting pages are placed
1407 *
1408 * find_get_pages_contig() works exactly like find_get_pages(), except
1409 * that the returned number of pages are guaranteed to be contiguous.
1410 *
1411 * find_get_pages_contig() returns the number of pages which were found.
1412 */
1413unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1414 unsigned int nr_pages, struct page **pages)
1415{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001416 struct radix_tree_iter iter;
1417 void **slot;
1418 unsigned int ret = 0;
1419
1420 if (unlikely(!nr_pages))
1421 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001422
Nick Piggina60637c2008-07-25 19:45:31 -07001423 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001424 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001425 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001426repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001427 page = radix_tree_deref_slot(slot);
1428 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001429 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001430 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001431
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001432 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001433 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001434 slot = radix_tree_iter_retry(&iter);
1435 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001436 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001437 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001438 * A shadow entry of a recently evicted page,
1439 * or a swap entry from shmem/tmpfs. Stop
1440 * looking for contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001441 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001442 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001443 }
Nick Piggina60637c2008-07-25 19:45:31 -07001444
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001445 head = compound_head(page);
1446 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001447 goto repeat;
1448
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001449 /* The page was split under us? */
1450 if (compound_head(page) != head) {
1451 put_page(head);
1452 goto repeat;
1453 }
1454
Nick Piggina60637c2008-07-25 19:45:31 -07001455 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001456 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001457 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001458 goto repeat;
1459 }
1460
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001461 /*
1462 * must check mapping and index after taking the ref.
1463 * otherwise we can get both false positives and false
1464 * negatives, which is just confusing to the caller.
1465 */
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001466 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001467 put_page(page);
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001468 break;
1469 }
1470
Nick Piggina60637c2008-07-25 19:45:31 -07001471 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001472 if (++ret == nr_pages)
1473 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001474 }
Nick Piggina60637c2008-07-25 19:45:31 -07001475 rcu_read_unlock();
1476 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001477}
David Howellsef71c152007-05-09 02:33:44 -07001478EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001479
Randy Dunlap485bb992006-06-23 02:03:49 -07001480/**
1481 * find_get_pages_tag - find and return pages that match @tag
1482 * @mapping: the address_space to search
1483 * @index: the starting page index
1484 * @tag: the tag index
1485 * @nr_pages: the maximum number of pages
1486 * @pages: where the resulting pages are placed
1487 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001489 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 */
1491unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1492 int tag, unsigned int nr_pages, struct page **pages)
1493{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001494 struct radix_tree_iter iter;
1495 void **slot;
1496 unsigned ret = 0;
1497
1498 if (unlikely(!nr_pages))
1499 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
Nick Piggina60637c2008-07-25 19:45:31 -07001501 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001502 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1503 &iter, *index, tag) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001504 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001505repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001506 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001507 if (unlikely(!page))
1508 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001509
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001510 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001511 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001512 slot = radix_tree_iter_retry(&iter);
1513 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001514 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001515 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001516 * A shadow entry of a recently evicted page.
1517 *
1518 * Those entries should never be tagged, but
1519 * this tree walk is lockless and the tags are
1520 * looked up in bulk, one radix tree node at a
1521 * time, so there is a sizable window for page
1522 * reclaim to evict a page we saw tagged.
1523 *
1524 * Skip over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001525 */
Johannes Weiner139b6a62014-05-06 12:50:05 -07001526 continue;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001527 }
Nick Piggina60637c2008-07-25 19:45:31 -07001528
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001529 head = compound_head(page);
1530 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001531 goto repeat;
1532
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001533 /* The page was split under us? */
1534 if (compound_head(page) != head) {
1535 put_page(head);
1536 goto repeat;
1537 }
1538
Nick Piggina60637c2008-07-25 19:45:31 -07001539 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001540 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001541 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001542 goto repeat;
1543 }
1544
1545 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001546 if (++ret == nr_pages)
1547 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001548 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001549
Nick Piggina60637c2008-07-25 19:45:31 -07001550 rcu_read_unlock();
1551
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 if (ret)
1553 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001554
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 return ret;
1556}
David Howellsef71c152007-05-09 02:33:44 -07001557EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001559/**
1560 * find_get_entries_tag - find and return entries that match @tag
1561 * @mapping: the address_space to search
1562 * @start: the starting page cache index
1563 * @tag: the tag index
1564 * @nr_entries: the maximum number of entries
1565 * @entries: where the resulting entries are placed
1566 * @indices: the cache indices corresponding to the entries in @entries
1567 *
1568 * Like find_get_entries, except we only return entries which are tagged with
1569 * @tag.
1570 */
1571unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1572 int tag, unsigned int nr_entries,
1573 struct page **entries, pgoff_t *indices)
1574{
1575 void **slot;
1576 unsigned int ret = 0;
1577 struct radix_tree_iter iter;
1578
1579 if (!nr_entries)
1580 return 0;
1581
1582 rcu_read_lock();
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001583 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1584 &iter, start, tag) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001585 struct page *head, *page;
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001586repeat:
1587 page = radix_tree_deref_slot(slot);
1588 if (unlikely(!page))
1589 continue;
1590 if (radix_tree_exception(page)) {
1591 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001592 slot = radix_tree_iter_retry(&iter);
1593 continue;
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001594 }
1595
1596 /*
1597 * A shadow entry of a recently evicted page, a swap
1598 * entry from shmem/tmpfs or a DAX entry. Return it
1599 * without attempting to raise page count.
1600 */
1601 goto export;
1602 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001603
1604 head = compound_head(page);
1605 if (!page_cache_get_speculative(head))
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001606 goto repeat;
1607
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001608 /* The page was split under us? */
1609 if (compound_head(page) != head) {
1610 put_page(head);
1611 goto repeat;
1612 }
1613
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001614 /* Has the page moved? */
1615 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001616 put_page(head);
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001617 goto repeat;
1618 }
1619export:
1620 indices[ret] = iter.index;
1621 entries[ret] = page;
1622 if (++ret == nr_entries)
1623 break;
1624 }
1625 rcu_read_unlock();
1626 return ret;
1627}
1628EXPORT_SYMBOL(find_get_entries_tag);
1629
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001630/*
1631 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1632 * a _large_ part of the i/o request. Imagine the worst scenario:
1633 *
1634 * ---R__________________________________________B__________
1635 * ^ reading here ^ bad block(assume 4k)
1636 *
1637 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1638 * => failing the whole request => read(R) => read(R+1) =>
1639 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1640 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1641 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1642 *
1643 * It is going insane. Fix it by quickly scaling down the readahead size.
1644 */
1645static void shrink_readahead_size_eio(struct file *filp,
1646 struct file_ra_state *ra)
1647{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001648 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001649}
1650
Randy Dunlap485bb992006-06-23 02:03:49 -07001651/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001652 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001653 * @filp: the file to read
1654 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001655 * @iter: data destination
1656 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001657 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001659 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 *
1661 * This is really ugly. But the goto's actually try to clarify some
1662 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 */
Al Viro6e58e792014-02-03 17:07:03 -05001664static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1665 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001667 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001669 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001670 pgoff_t index;
1671 pgoff_t last_index;
1672 pgoff_t prev_index;
1673 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001674 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001675 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001677 index = *ppos >> PAGE_SHIFT;
1678 prev_index = ra->prev_pos >> PAGE_SHIFT;
1679 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1680 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1681 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 for (;;) {
1684 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001685 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001686 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 unsigned long nr, ret;
1688
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690find_page:
1691 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001692 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001693 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001694 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001695 index, last_index - index);
1696 page = find_get_page(mapping, index);
1697 if (unlikely(page == NULL))
1698 goto no_cached_page;
1699 }
1700 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001701 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001702 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001703 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001705 if (!PageUptodate(page)) {
Mel Gormanebded022016-03-15 14:55:39 -07001706 /*
1707 * See comment in do_read_cache_page on why
1708 * wait_on_page_locked is used to avoid unnecessarily
1709 * serialisations and why it's safe.
1710 */
1711 wait_on_page_locked_killable(page);
1712 if (PageUptodate(page))
1713 goto page_ok;
1714
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001715 if (inode->i_blkbits == PAGE_SHIFT ||
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001716 !mapping->a_ops->is_partially_uptodate)
1717 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001718 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001719 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001720 /* Did it get truncated before we got the lock? */
1721 if (!page->mapping)
1722 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001723 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001724 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001725 goto page_not_up_to_date_locked;
1726 unlock_page(page);
1727 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001729 /*
1730 * i_size must be checked after we know the page is Uptodate.
1731 *
1732 * Checking i_size after the check allows us to calculate
1733 * the correct value for "nr", which means the zero-filled
1734 * part of the page is not copied back to userspace (unless
1735 * another truncate extends the file - this is desired though).
1736 */
1737
1738 isize = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001739 end_index = (isize - 1) >> PAGE_SHIFT;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001740 if (unlikely(!isize || index > end_index)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001741 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07001742 goto out;
1743 }
1744
1745 /* nr is the maximum number of bytes to copy from this page */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001746 nr = PAGE_SIZE;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001747 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001748 nr = ((isize - 1) & ~PAGE_MASK) + 1;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001749 if (nr <= offset) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001750 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07001751 goto out;
1752 }
1753 }
1754 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
1756 /* If users can be writing to this page using arbitrary
1757 * virtual addresses, take care about potential aliasing
1758 * before reading the page on the kernel side.
1759 */
1760 if (mapping_writably_mapped(mapping))
1761 flush_dcache_page(page);
1762
1763 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001764 * When a sequential read accesses a page several times,
1765 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 */
Jan Karaec0f1632007-05-06 14:49:25 -07001767 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 mark_page_accessed(page);
1769 prev_index = index;
1770
1771 /*
1772 * Ok, we have the page, and it's up-to-date, so
1773 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 */
Al Viro6e58e792014-02-03 17:07:03 -05001775
1776 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001778 index += offset >> PAGE_SHIFT;
1779 offset &= ~PAGE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001780 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001782 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001783 written += ret;
1784 if (!iov_iter_count(iter))
1785 goto out;
1786 if (ret < nr) {
1787 error = -EFAULT;
1788 goto out;
1789 }
1790 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
1792page_not_up_to_date:
1793 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001794 error = lock_page_killable(page);
1795 if (unlikely(error))
1796 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001798page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001799 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 if (!page->mapping) {
1801 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001802 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 continue;
1804 }
1805
1806 /* Did somebody else fill it already? */
1807 if (PageUptodate(page)) {
1808 unlock_page(page);
1809 goto page_ok;
1810 }
1811
1812readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001813 /*
1814 * A previous I/O error may have been due to temporary
1815 * failures, eg. multipath errors.
1816 * PG_error will be set again if readpage fails.
1817 */
1818 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 /* Start the actual read. The read will unlock the page. */
1820 error = mapping->a_ops->readpage(filp, page);
1821
Zach Brown994fc28c2005-12-15 14:28:17 -08001822 if (unlikely(error)) {
1823 if (error == AOP_TRUNCATED_PAGE) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001824 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001825 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001826 goto find_page;
1827 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001829 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830
1831 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001832 error = lock_page_killable(page);
1833 if (unlikely(error))
1834 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 if (!PageUptodate(page)) {
1836 if (page->mapping == NULL) {
1837 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001838 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 */
1840 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001841 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 goto find_page;
1843 }
1844 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001845 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001846 error = -EIO;
1847 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 }
1849 unlock_page(page);
1850 }
1851
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 goto page_ok;
1853
1854readpage_error:
1855 /* UHHUH! A synchronous read error occurred. Report it */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001856 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 goto out;
1858
1859no_cached_page:
1860 /*
1861 * Ok, it wasn't cached, so we need to create a new
1862 * page..
1863 */
Nick Piggineb2be182007-10-16 01:24:57 -07001864 page = page_cache_alloc_cold(mapping);
1865 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05001866 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07001867 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 }
Michal Hocko6afdb852015-06-24 16:58:06 -07001869 error = add_to_page_cache_lru(page, mapping, index,
Michal Hockoc62d2552015-11-06 16:28:49 -08001870 mapping_gfp_constraint(mapping, GFP_KERNEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 if (error) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001872 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001873 if (error == -EEXIST) {
1874 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05001876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 goto out;
1878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 goto readpage;
1880 }
1881
1882out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001883 ra->prev_pos = prev_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001884 ra->prev_pos <<= PAGE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001885 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001887 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001888 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05001889 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890}
1891
Randy Dunlap485bb992006-06-23 02:03:49 -07001892/**
Al Viro6abd2322014-04-04 14:20:57 -04001893 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001894 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04001895 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07001896 *
Al Viro6abd2322014-04-04 14:20:57 -04001897 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 * that can use the page cache directly.
1899 */
1900ssize_t
Al Viroed978a82014-03-05 22:53:04 -05001901generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902{
Al Viroed978a82014-03-05 22:53:04 -05001903 struct file *file = iocb->ki_filp;
1904 ssize_t retval = 0;
Nicolai Stangee7080a42016-03-25 14:22:14 -07001905 size_t count = iov_iter_count(iter);
1906
1907 if (!count)
1908 goto out; /* skip atime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909
Al Viro2ba48ce2015-04-09 13:52:01 -04001910 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viroed978a82014-03-05 22:53:04 -05001911 struct address_space *mapping = file->f_mapping;
1912 struct inode *inode = mapping->host;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001913 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 size = i_size_read(inode);
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001916 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
1917 iocb->ki_pos + count - 1);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001918 if (!retval) {
Al Viroed978a82014-03-05 22:53:04 -05001919 struct iov_iter data = *iter;
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07001920 retval = mapping->a_ops->direct_IO(iocb, &data);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001921 }
Al Viroed978a82014-03-05 22:53:04 -05001922
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001923 if (retval > 0) {
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001924 iocb->ki_pos += retval;
Al Viroed978a82014-03-05 22:53:04 -05001925 iov_iter_advance(iter, retval);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001926 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001927
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001928 /*
1929 * Btrfs can have a short DIO read if we encounter
1930 * compressed extents, so if there was an error, or if
1931 * we've already read everything we wanted to, or if
1932 * there was a short read because we hit EOF, go ahead
1933 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001934 * the rest of the read. Buffered reads will not work for
1935 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001936 */
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001937 if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size ||
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001938 IS_DAX(inode)) {
Al Viroed978a82014-03-05 22:53:04 -05001939 file_accessed(file);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001940 goto out;
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001941 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 }
1943
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001944 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945out:
1946 return retval;
1947}
Al Viroed978a82014-03-05 22:53:04 -05001948EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001951/**
1952 * page_cache_read - adds requested page to the page cache if not already there
1953 * @file: file to read
1954 * @offset: page index
Randy Dunlap62eb3202016-02-11 16:12:58 -08001955 * @gfp_mask: memory allocation flags
Randy Dunlap485bb992006-06-23 02:03:49 -07001956 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 * This adds the requested page to the page cache if it isn't already there,
1958 * and schedules an I/O to read in its contents from disk.
1959 */
Michal Hockoc20cd452016-01-14 15:20:12 -08001960static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961{
1962 struct address_space *mapping = file->f_mapping;
Paul McQuade99dadfd2014-10-09 15:29:03 -07001963 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001964 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
Zach Brown994fc28c2005-12-15 14:28:17 -08001966 do {
Michal Hockoc20cd452016-01-14 15:20:12 -08001967 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
Zach Brown994fc28c2005-12-15 14:28:17 -08001968 if (!page)
1969 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
Michal Hockoc20cd452016-01-14 15:20:12 -08001971 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
Zach Brown994fc28c2005-12-15 14:28:17 -08001972 if (ret == 0)
1973 ret = mapping->a_ops->readpage(file, page);
1974 else if (ret == -EEXIST)
1975 ret = 0; /* losing race to add is OK */
1976
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001977 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
Zach Brown994fc28c2005-12-15 14:28:17 -08001979 } while (ret == AOP_TRUNCATED_PAGE);
Paul McQuade99dadfd2014-10-09 15:29:03 -07001980
Zach Brown994fc28c2005-12-15 14:28:17 -08001981 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982}
1983
1984#define MMAP_LOTSAMISS (100)
1985
Linus Torvaldsef00e082009-06-16 15:31:25 -07001986/*
1987 * Synchronous readahead happens when we don't even find
1988 * a page in the page cache at all.
1989 */
1990static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1991 struct file_ra_state *ra,
1992 struct file *file,
1993 pgoff_t offset)
1994{
Linus Torvaldsef00e082009-06-16 15:31:25 -07001995 struct address_space *mapping = file->f_mapping;
1996
1997 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001998 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001999 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002000 if (!ra->ra_pages)
2001 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002002
Joe Perches64363aa2013-07-08 16:00:18 -07002003 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07002004 page_cache_sync_readahead(mapping, ra, file, offset,
2005 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002006 return;
2007 }
2008
Andi Kleen207d04b2011-05-24 17:12:29 -07002009 /* Avoid banging the cache line if not needed */
2010 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002011 ra->mmap_miss++;
2012
2013 /*
2014 * Do we miss much more than hit in this file? If so,
2015 * stop bothering with read-ahead. It will only hurt.
2016 */
2017 if (ra->mmap_miss > MMAP_LOTSAMISS)
2018 return;
2019
Wu Fengguangd30a1102009-06-16 15:31:30 -07002020 /*
2021 * mmap read-around
2022 */
Roman Gushchin600e19a2015-11-05 18:47:08 -08002023 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2024 ra->size = ra->ra_pages;
2025 ra->async_size = ra->ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002026 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002027}
2028
2029/*
2030 * Asynchronous readahead happens when we find the page and PG_readahead,
2031 * so we want to possibly extend the readahead further..
2032 */
2033static void do_async_mmap_readahead(struct vm_area_struct *vma,
2034 struct file_ra_state *ra,
2035 struct file *file,
2036 struct page *page,
2037 pgoff_t offset)
2038{
2039 struct address_space *mapping = file->f_mapping;
2040
2041 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07002042 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002043 return;
2044 if (ra->mmap_miss > 0)
2045 ra->mmap_miss--;
2046 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07002047 page_cache_async_readahead(mapping, ra, file,
2048 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002049}
2050
Randy Dunlap485bb992006-06-23 02:03:49 -07002051/**
Nick Piggin54cb8822007-07-19 01:46:59 -07002052 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07002053 * @vma: vma in which the fault was taken
2054 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07002055 *
Nick Piggin54cb8822007-07-19 01:46:59 -07002056 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 * mapped memory region to read in file data during a page fault.
2058 *
2059 * The goto's are kind of ugly, but this streamlines the normal case of having
2060 * it in the page cache, and handles the special cases reasonably without
2061 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002062 *
2063 * vma->vm_mm->mmap_sem must be held on entry.
2064 *
2065 * If our return value has VM_FAULT_RETRY set, it's because
2066 * lock_page_or_retry() returned 0.
2067 * The mmap_sem has usually been released in this case.
2068 * See __lock_page_or_retry() for the exception.
2069 *
2070 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2071 * has not been released.
2072 *
2073 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002075int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076{
2077 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07002078 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 struct address_space *mapping = file->f_mapping;
2080 struct file_ra_state *ra = &file->f_ra;
2081 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002082 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 struct page *page;
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002084 loff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07002085 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002087 size = round_up(i_size_read(inode), PAGE_SIZE);
2088 if (offset >= size >> PAGE_SHIFT)
Linus Torvalds5307cc12007-10-31 09:19:46 -07002089 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002092 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002094 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002095 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07002097 * We found the page, so try async readahead before
2098 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002100 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002101 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07002102 /* No page in the page cache at all */
2103 do_sync_mmap_readahead(vma, ra, file, offset);
2104 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07002105 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002106 ret = VM_FAULT_MAJOR;
2107retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07002108 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 if (!page)
2110 goto no_cached_page;
2111 }
2112
Michel Lespinassed88c0922010-11-02 13:05:18 -07002113 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002114 put_page(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002115 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07002116 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07002117
2118 /* Did it get truncated? */
2119 if (unlikely(page->mapping != mapping)) {
2120 unlock_page(page);
2121 put_page(page);
2122 goto retry_find;
2123 }
Sasha Levin309381fea2014-01-23 15:52:54 -08002124 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07002125
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 /*
Nick Piggind00806b2007-07-19 01:46:57 -07002127 * We have a locked page in the page cache, now we need to check
2128 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 */
Nick Piggind00806b2007-07-19 01:46:57 -07002130 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 goto page_not_uptodate;
2132
Linus Torvaldsef00e082009-06-16 15:31:25 -07002133 /*
2134 * Found the page and have a reference on it.
2135 * We must recheck i_size under page lock.
2136 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002137 size = round_up(i_size_read(inode), PAGE_SIZE);
2138 if (unlikely(offset >= size >> PAGE_SHIFT)) {
Nick Piggind00806b2007-07-19 01:46:57 -07002139 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002140 put_page(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07002141 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07002142 }
2143
Nick Piggind0217ac2007-07-19 01:47:03 -07002144 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07002145 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147no_cached_page:
2148 /*
2149 * We're only likely to ever get here if MADV_RANDOM is in
2150 * effect.
2151 */
Michal Hockoc20cd452016-01-14 15:20:12 -08002152 error = page_cache_read(file, offset, vmf->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153
2154 /*
2155 * The page we want has now been added to the page cache.
2156 * In the unlikely event that someone removed it in the
2157 * meantime, we'll just come back here and read it again.
2158 */
2159 if (error >= 0)
2160 goto retry_find;
2161
2162 /*
2163 * An error return from page_cache_read can result if the
2164 * system is low on memory, or a problem occurs while trying
2165 * to schedule I/O.
2166 */
2167 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07002168 return VM_FAULT_OOM;
2169 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
2171page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 /*
2173 * Umm, take care of errors if the page isn't up-to-date.
2174 * Try to re-read it _once_. We do this synchronously,
2175 * because there really aren't any performance issues here
2176 * and we need to check for errors.
2177 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08002179 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07002180 if (!error) {
2181 wait_on_page_locked(page);
2182 if (!PageUptodate(page))
2183 error = -EIO;
2184 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002185 put_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002186
2187 if (!error || error == AOP_TRUNCATED_PAGE)
2188 goto retry_find;
2189
2190 /* Things didn't work out. Return zero to tell the mm layer so. */
2191 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002192 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07002193}
2194EXPORT_SYMBOL(filemap_fault);
2195
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002196void filemap_map_pages(struct fault_env *fe,
2197 pgoff_t start_pgoff, pgoff_t end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002198{
2199 struct radix_tree_iter iter;
2200 void **slot;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002201 struct file *file = fe->vma->vm_file;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002202 struct address_space *mapping = file->f_mapping;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002203 pgoff_t last_pgoff = start_pgoff;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002204 loff_t size;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002205 struct page *head, *page;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002206
2207 rcu_read_lock();
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002208 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2209 start_pgoff) {
2210 if (iter.index > end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002211 break;
2212repeat:
2213 page = radix_tree_deref_slot(slot);
2214 if (unlikely(!page))
2215 goto next;
2216 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002217 if (radix_tree_deref_retry(page)) {
2218 slot = radix_tree_iter_retry(&iter);
2219 continue;
2220 }
2221 goto next;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002222 }
2223
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002224 head = compound_head(page);
2225 if (!page_cache_get_speculative(head))
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002226 goto repeat;
2227
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002228 /* The page was split under us? */
2229 if (compound_head(page) != head) {
2230 put_page(head);
2231 goto repeat;
2232 }
2233
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002234 /* Has the page moved? */
2235 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002236 put_page(head);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002237 goto repeat;
2238 }
2239
2240 if (!PageUptodate(page) ||
2241 PageReadahead(page) ||
2242 PageHWPoison(page))
2243 goto skip;
2244 if (!trylock_page(page))
2245 goto skip;
2246
2247 if (page->mapping != mapping || !PageUptodate(page))
2248 goto unlock;
2249
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002250 size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2251 if (page->index >= size >> PAGE_SHIFT)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002252 goto unlock;
2253
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002254 if (file->f_ra.mmap_miss > 0)
2255 file->f_ra.mmap_miss--;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002256
2257 fe->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2258 if (fe->pte)
2259 fe->pte += iter.index - last_pgoff;
2260 last_pgoff = iter.index;
2261 if (alloc_set_pte(fe, NULL, page))
2262 goto unlock;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002263 unlock_page(page);
2264 goto next;
2265unlock:
2266 unlock_page(page);
2267skip:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002268 put_page(page);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002269next:
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002270 /* Huge page is mapped? No need to proceed. */
2271 if (pmd_trans_huge(*fe->pmd))
2272 break;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002273 if (iter.index == end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002274 break;
2275 }
2276 rcu_read_unlock();
2277}
2278EXPORT_SYMBOL(filemap_map_pages);
2279
Jan Kara4fcf1c62012-06-12 16:20:29 +02002280int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2281{
2282 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002283 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002284 int ret = VM_FAULT_LOCKED;
2285
Jan Kara14da9202012-06-12 16:20:37 +02002286 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002287 file_update_time(vma->vm_file);
2288 lock_page(page);
2289 if (page->mapping != inode->i_mapping) {
2290 unlock_page(page);
2291 ret = VM_FAULT_NOPAGE;
2292 goto out;
2293 }
Jan Kara14da9202012-06-12 16:20:37 +02002294 /*
2295 * We mark the page dirty already here so that when freeze is in
2296 * progress, we are guaranteed that writeback during freezing will
2297 * see the dirty page and writeprotect it again.
2298 */
2299 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002300 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002301out:
Jan Kara14da9202012-06-12 16:20:37 +02002302 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002303 return ret;
2304}
2305EXPORT_SYMBOL(filemap_page_mkwrite);
2306
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002307const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002308 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002309 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002310 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311};
2312
2313/* This is used for a general mmap of a disk file */
2314
2315int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2316{
2317 struct address_space *mapping = file->f_mapping;
2318
2319 if (!mapping->a_ops->readpage)
2320 return -ENOEXEC;
2321 file_accessed(file);
2322 vma->vm_ops = &generic_file_vm_ops;
2323 return 0;
2324}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
2326/*
2327 * This is for filesystems which do not implement ->writepage.
2328 */
2329int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2330{
2331 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2332 return -EINVAL;
2333 return generic_file_mmap(file, vma);
2334}
2335#else
2336int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2337{
2338 return -ENOSYS;
2339}
2340int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2341{
2342 return -ENOSYS;
2343}
2344#endif /* CONFIG_MMU */
2345
2346EXPORT_SYMBOL(generic_file_mmap);
2347EXPORT_SYMBOL(generic_file_readonly_mmap);
2348
Sasha Levin67f9fd92014-04-03 14:48:18 -07002349static struct page *wait_on_page_read(struct page *page)
2350{
2351 if (!IS_ERR(page)) {
2352 wait_on_page_locked(page);
2353 if (!PageUptodate(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002354 put_page(page);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002355 page = ERR_PTR(-EIO);
2356 }
2357 }
2358 return page;
2359}
2360
Mel Gorman32b63522016-03-15 14:55:36 -07002361static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002362 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002363 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002364 void *data,
2365 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366{
Nick Piggineb2be182007-10-16 01:24:57 -07002367 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 int err;
2369repeat:
2370 page = find_get_page(mapping, index);
2371 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002372 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002373 if (!page)
2374 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002375 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002376 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002377 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002378 if (err == -EEXIST)
2379 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 return ERR_PTR(err);
2382 }
Mel Gorman32b63522016-03-15 14:55:36 -07002383
2384filler:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 err = filler(data, page);
2386 if (err < 0) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002387 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002388 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 }
Mel Gorman32b63522016-03-15 14:55:36 -07002390
2391 page = wait_on_page_read(page);
2392 if (IS_ERR(page))
2393 return page;
2394 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 if (PageUptodate(page))
2397 goto out;
2398
Mel Gormanebded022016-03-15 14:55:39 -07002399 /*
2400 * Page is not up to date and may be locked due one of the following
2401 * case a: Page is being filled and the page lock is held
2402 * case b: Read/write error clearing the page uptodate status
2403 * case c: Truncation in progress (page locked)
2404 * case d: Reclaim in progress
2405 *
2406 * Case a, the page will be up to date when the page is unlocked.
2407 * There is no need to serialise on the page lock here as the page
2408 * is pinned so the lock gives no additional protection. Even if the
2409 * the page is truncated, the data is still valid if PageUptodate as
2410 * it's a race vs truncate race.
2411 * Case b, the page will not be up to date
2412 * Case c, the page may be truncated but in itself, the data may still
2413 * be valid after IO completes as it's a read vs truncate race. The
2414 * operation must restart if the page is not uptodate on unlock but
2415 * otherwise serialising on page lock to stabilise the mapping gives
2416 * no additional guarantees to the caller as the page lock is
2417 * released before return.
2418 * Case d, similar to truncation. If reclaim holds the page lock, it
2419 * will be a race with remove_mapping that determines if the mapping
2420 * is valid on unlock but otherwise the data is valid and there is
2421 * no need to serialise with page lock.
2422 *
2423 * As the page lock gives no additional guarantee, we optimistically
2424 * wait on the page to be unlocked and check if it's up to date and
2425 * use the page if it is. Otherwise, the page lock is required to
2426 * distinguish between the different cases. The motivation is that we
2427 * avoid spurious serialisations and wakeups when multiple processes
2428 * wait on the same page for IO to complete.
2429 */
2430 wait_on_page_locked(page);
2431 if (PageUptodate(page))
2432 goto out;
2433
2434 /* Distinguish between all the cases under the safety of the lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 lock_page(page);
Mel Gormanebded022016-03-15 14:55:39 -07002436
2437 /* Case c or d, restart the operation */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 if (!page->mapping) {
2439 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002440 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002441 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 }
Mel Gormanebded022016-03-15 14:55:39 -07002443
2444 /* Someone else locked and filled the page in a very small window */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 if (PageUptodate(page)) {
2446 unlock_page(page);
2447 goto out;
2448 }
Mel Gorman32b63522016-03-15 14:55:36 -07002449 goto filler;
2450
David Howellsc855ff32007-05-09 13:42:20 +01002451out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002452 mark_page_accessed(page);
2453 return page;
2454}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002455
2456/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002457 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002458 * @mapping: the page's address_space
2459 * @index: the page index
2460 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002461 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002462 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002463 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002464 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002465 *
2466 * If the page does not get brought uptodate, return -EIO.
2467 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002468struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002469 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002470 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002471 void *data)
2472{
2473 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2474}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002475EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002476
2477/**
2478 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2479 * @mapping: the page's address_space
2480 * @index: the page index
2481 * @gfp: the page allocator flags to use if allocating
2482 *
2483 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002484 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002485 *
2486 * If the page does not get brought uptodate, return -EIO.
2487 */
2488struct page *read_cache_page_gfp(struct address_space *mapping,
2489 pgoff_t index,
2490 gfp_t gfp)
2491{
2492 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2493
Sasha Levin67f9fd92014-04-03 14:48:18 -07002494 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002495}
2496EXPORT_SYMBOL(read_cache_page_gfp);
2497
Nick Piggin2f718ff2007-10-16 01:24:59 -07002498/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 * Performs necessary checks before doing a write
2500 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002501 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 * Returns appropriate error code that caller should return or
2503 * zero in case that write should be allowed.
2504 */
Al Viro3309dd02015-04-09 12:55:47 -04002505inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506{
Al Viro3309dd02015-04-09 12:55:47 -04002507 struct file *file = iocb->ki_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002509 unsigned long limit = rlimit(RLIMIT_FSIZE);
Al Viro3309dd02015-04-09 12:55:47 -04002510 loff_t pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
Al Viro3309dd02015-04-09 12:55:47 -04002512 if (!iov_iter_count(from))
2513 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514
Al Viro0fa6b002015-04-04 04:05:48 -04002515 /* FIXME: this is for backwards compatibility with 2.4 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002516 if (iocb->ki_flags & IOCB_APPEND)
Al Viro3309dd02015-04-09 12:55:47 -04002517 iocb->ki_pos = i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518
Al Viro3309dd02015-04-09 12:55:47 -04002519 pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520
Al Viro0fa6b002015-04-04 04:05:48 -04002521 if (limit != RLIM_INFINITY) {
Al Viro3309dd02015-04-09 12:55:47 -04002522 if (iocb->ki_pos >= limit) {
Al Viro0fa6b002015-04-04 04:05:48 -04002523 send_sig(SIGXFSZ, current, 0);
2524 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 }
Al Viro3309dd02015-04-09 12:55:47 -04002526 iov_iter_truncate(from, limit - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 }
2528
2529 /*
2530 * LFS rule
2531 */
Al Viro3309dd02015-04-09 12:55:47 -04002532 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002533 !(file->f_flags & O_LARGEFILE))) {
Al Viro3309dd02015-04-09 12:55:47 -04002534 if (pos >= MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 return -EFBIG;
Al Viro3309dd02015-04-09 12:55:47 -04002536 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 }
2538
2539 /*
2540 * Are we about to exceed the fs block limit ?
2541 *
2542 * If we have written data it becomes a short write. If we have
2543 * exceeded without writing data we send a signal and return EFBIG.
2544 * Linus frestrict idea will clean these up nicely..
2545 */
Al Viro3309dd02015-04-09 12:55:47 -04002546 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2547 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548
Al Viro3309dd02015-04-09 12:55:47 -04002549 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2550 return iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551}
2552EXPORT_SYMBOL(generic_write_checks);
2553
Nick Pigginafddba42007-10-16 01:25:01 -07002554int pagecache_write_begin(struct file *file, struct address_space *mapping,
2555 loff_t pos, unsigned len, unsigned flags,
2556 struct page **pagep, void **fsdata)
2557{
2558 const struct address_space_operations *aops = mapping->a_ops;
2559
Nick Piggin4e02ed42008-10-29 14:00:55 -07002560 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002561 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002562}
2563EXPORT_SYMBOL(pagecache_write_begin);
2564
2565int pagecache_write_end(struct file *file, struct address_space *mapping,
2566 loff_t pos, unsigned len, unsigned copied,
2567 struct page *page, void *fsdata)
2568{
2569 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002570
Nick Piggin4e02ed42008-10-29 14:00:55 -07002571 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002572}
2573EXPORT_SYMBOL(pagecache_write_end);
2574
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575ssize_t
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002576generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577{
2578 struct file *file = iocb->ki_filp;
2579 struct address_space *mapping = file->f_mapping;
2580 struct inode *inode = mapping->host;
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002581 loff_t pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002583 size_t write_len;
2584 pgoff_t end;
Al Viro26978b82014-03-10 14:08:45 -04002585 struct iov_iter data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586
Al Viro0c949332014-03-22 06:51:37 -04002587 write_len = iov_iter_count(from);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002588 end = (pos + write_len - 1) >> PAGE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002589
Nick Piggin48b47c52009-01-06 14:40:22 -08002590 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002591 if (written)
2592 goto out;
2593
2594 /*
2595 * After a write we want buffered reads to be sure to go to disk to get
2596 * the new data. We invalidate clean cached page from the region we're
2597 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002598 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002599 */
2600 if (mapping->nrpages) {
2601 written = invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002602 pos >> PAGE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002603 /*
2604 * If a page can not be invalidated, return 0 to fall back
2605 * to buffered write.
2606 */
2607 if (written) {
2608 if (written == -EBUSY)
2609 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002610 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002611 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002612 }
2613
Al Viro26978b82014-03-10 14:08:45 -04002614 data = *from;
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07002615 written = mapping->a_ops->direct_IO(iocb, &data);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002616
2617 /*
2618 * Finally, try again to invalidate clean pages which might have been
2619 * cached by non-direct readahead, or faulted in by get_user_pages()
2620 * if the source of the write was an mmap'ed region of the file
2621 * we're writing. Either one is a pretty crazy thing to do,
2622 * so we don't support it 100%. If this invalidation
2623 * fails, tough, the write still worked...
2624 */
2625 if (mapping->nrpages) {
2626 invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002627 pos >> PAGE_SHIFT, end);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002628 }
2629
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002631 pos += written;
Al Virof8579f82014-03-03 22:03:20 -05002632 iov_iter_advance(from, written);
Namhyung Kim01166512010-10-26 14:21:58 -07002633 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2634 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 mark_inode_dirty(inode);
2636 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002637 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002639out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 return written;
2641}
2642EXPORT_SYMBOL(generic_file_direct_write);
2643
Nick Piggineb2be182007-10-16 01:24:57 -07002644/*
2645 * Find or create a page at the given pagecache position. Return the locked
2646 * page. This function is specifically for buffered writes.
2647 */
Nick Piggin54566b22009-01-04 12:00:53 -08002648struct page *grab_cache_page_write_begin(struct address_space *mapping,
2649 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002650{
Nick Piggineb2be182007-10-16 01:24:57 -07002651 struct page *page;
Johannes Weinerbbddabe2016-05-20 16:56:28 -07002652 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002653
Nick Piggin54566b22009-01-04 12:00:53 -08002654 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07002655 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07002656
Mel Gorman2457aec2014-06-04 16:10:31 -07002657 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01002658 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07002659 if (page)
2660 wait_for_stable_page(page);
2661
Nick Piggineb2be182007-10-16 01:24:57 -07002662 return page;
2663}
Nick Piggin54566b22009-01-04 12:00:53 -08002664EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002665
Al Viro3b93f912014-02-11 21:34:08 -05002666ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002667 struct iov_iter *i, loff_t pos)
2668{
2669 struct address_space *mapping = file->f_mapping;
2670 const struct address_space_operations *a_ops = mapping->a_ops;
2671 long status = 0;
2672 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002673 unsigned int flags = 0;
2674
2675 /*
2676 * Copies from kernel address space cannot fail (NFSD is a big user).
2677 */
Al Viro777eda22014-12-17 04:46:46 -05002678 if (!iter_is_iovec(i))
Nick Piggin674b8922007-10-16 01:25:03 -07002679 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002680
2681 do {
2682 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002683 unsigned long offset; /* Offset into pagecache page */
2684 unsigned long bytes; /* Bytes to write to page */
2685 size_t copied; /* Bytes copied from user */
2686 void *fsdata;
2687
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002688 offset = (pos & (PAGE_SIZE - 1));
2689 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002690 iov_iter_count(i));
2691
2692again:
Linus Torvalds00a3d662015-10-07 08:32:38 +01002693 /*
2694 * Bring in the user page that we will copy from _first_.
2695 * Otherwise there's a nasty deadlock on copying from the
2696 * same page as we're writing to, without it being marked
2697 * up-to-date.
2698 *
2699 * Not only is this an optimisation, but it is also required
2700 * to check that the address is actually valid, when atomic
2701 * usercopies are used, below.
2702 */
2703 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2704 status = -EFAULT;
2705 break;
2706 }
2707
Jan Kara296291c2015-10-22 13:32:21 -07002708 if (fatal_signal_pending(current)) {
2709 status = -EINTR;
2710 break;
2711 }
2712
Nick Piggin674b8922007-10-16 01:25:03 -07002713 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002714 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07002715 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07002716 break;
2717
anfei zhou931e80e2010-02-02 13:44:02 -08002718 if (mapping_writably_mapped(mapping))
2719 flush_dcache_page(page);
Linus Torvalds00a3d662015-10-07 08:32:38 +01002720
Nick Pigginafddba42007-10-16 01:25:01 -07002721 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07002722 flush_dcache_page(page);
2723
2724 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2725 page, fsdata);
2726 if (unlikely(status < 0))
2727 break;
2728 copied = status;
2729
2730 cond_resched();
2731
Nick Piggin124d3b72008-02-02 15:01:17 +01002732 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002733 if (unlikely(copied == 0)) {
2734 /*
2735 * If we were unable to copy any data at all, we must
2736 * fall back to a single segment length write.
2737 *
2738 * If we didn't fallback here, we could livelock
2739 * because not all segments in the iov can be copied at
2740 * once without a pagefault.
2741 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002742 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002743 iov_iter_single_seg_count(i));
2744 goto again;
2745 }
Nick Pigginafddba42007-10-16 01:25:01 -07002746 pos += copied;
2747 written += copied;
2748
2749 balance_dirty_pages_ratelimited(mapping);
Nick Pigginafddba42007-10-16 01:25:01 -07002750 } while (iov_iter_count(i));
2751
2752 return written ? written : status;
2753}
Al Viro3b93f912014-02-11 21:34:08 -05002754EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755
Jan Karae4dd9de2009-08-17 18:10:06 +02002756/**
Al Viro81742022014-04-03 03:17:43 -04002757 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002758 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04002759 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002760 *
2761 * This function does all the work needed for actually writing data to a
2762 * file. It does all basic checks, removes SUID from the file, updates
2763 * modification times and calls proper subroutines depending on whether we
2764 * do direct IO or a standard buffered write.
2765 *
2766 * It expects i_mutex to be grabbed unless we work on a block device or similar
2767 * object which does not need locking at all.
2768 *
2769 * This function does *not* take care of syncing data in case of O_SYNC write.
2770 * A caller has to handle it. This is mainly due to the fact that we want to
2771 * avoid syncing under i_mutex.
2772 */
Al Viro81742022014-04-03 03:17:43 -04002773ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774{
2775 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002776 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05002778 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002780 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01002783 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002784 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 if (err)
2786 goto out;
2787
Josef Bacikc3b2da32012-03-26 09:59:21 -04002788 err = file_update_time(file);
2789 if (err)
2790 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791
Al Viro2ba48ce2015-04-09 13:52:01 -04002792 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04002793 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002794
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002795 written = generic_file_direct_write(iocb, from);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002796 /*
2797 * If the write stopped short of completing, fall back to
2798 * buffered writes. Some filesystems do this for writes to
2799 * holes, for example. For DAX files, a buffered write will
2800 * not succeed (even if it did, DAX does not handle dirty
2801 * page-cache pages correctly).
2802 */
Al Viro0b8def92015-04-07 10:22:53 -04002803 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002805
Al Viro0b8def92015-04-07 10:22:53 -04002806 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002807 /*
Al Viro3b93f912014-02-11 21:34:08 -05002808 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002809 * then we want to return the number of bytes which were
2810 * direct-written, or the error code if that was zero. Note
2811 * that this differs from normal direct-io semantics, which
2812 * will return -EFOO even if some bytes were written.
2813 */
Al Viro60bb4522014-08-08 12:39:16 -04002814 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05002815 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002816 goto out;
2817 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002818 /*
2819 * We need to ensure that the page cache pages are written to
2820 * disk and invalidated to preserve the expected O_DIRECT
2821 * semantics.
2822 */
Al Viro3b93f912014-02-11 21:34:08 -05002823 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04002824 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002825 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04002826 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05002827 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002828 invalidate_mapping_pages(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002829 pos >> PAGE_SHIFT,
2830 endbyte >> PAGE_SHIFT);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002831 } else {
2832 /*
2833 * We don't know how much we wrote, so just return
2834 * the number of bytes which were direct-written
2835 */
2836 }
2837 } else {
Al Viro0b8def92015-04-07 10:22:53 -04002838 written = generic_perform_write(file, from, iocb->ki_pos);
2839 if (likely(written > 0))
2840 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002841 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842out:
2843 current->backing_dev_info = NULL;
2844 return written ? written : err;
2845}
Al Viro81742022014-04-03 03:17:43 -04002846EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847
Jan Karae4dd9de2009-08-17 18:10:06 +02002848/**
Al Viro81742022014-04-03 03:17:43 -04002849 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002850 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04002851 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002852 *
Al Viro81742022014-04-03 03:17:43 -04002853 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02002854 * filesystems. It takes care of syncing the file in case of O_SYNC file
2855 * and acquires i_mutex as needed.
2856 */
Al Viro81742022014-04-03 03:17:43 -04002857ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858{
2859 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002860 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862
Al Viro59551022016-01-22 15:40:57 -05002863 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -04002864 ret = generic_write_checks(iocb, from);
2865 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04002866 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -05002867 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868
Christoph Hellwige2592212016-04-07 08:52:01 -07002869 if (ret > 0)
2870 ret = generic_write_sync(iocb, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 return ret;
2872}
Al Viro81742022014-04-03 03:17:43 -04002873EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874
David Howellscf9a2ae2006-08-29 19:05:54 +01002875/**
2876 * try_to_release_page() - release old fs-specific metadata on a page
2877 *
2878 * @page: the page which the kernel is trying to free
2879 * @gfp_mask: memory allocation flags (and I/O mode)
2880 *
2881 * The address_space is to try to release any data against the page
2882 * (presumably at page->private). If the release was successful, return `1'.
2883 * Otherwise return zero.
2884 *
David Howells266cf652009-04-03 16:42:36 +01002885 * This may also be called if PG_fscache is set on a page, indicating that the
2886 * page is known to the local caching routines.
2887 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002888 * The @gfp_mask argument specifies whether I/O may be performed to release
Mel Gorman71baba42015-11-06 16:28:28 -08002889 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002890 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002891 */
2892int try_to_release_page(struct page *page, gfp_t gfp_mask)
2893{
2894 struct address_space * const mapping = page->mapping;
2895
2896 BUG_ON(!PageLocked(page));
2897 if (PageWriteback(page))
2898 return 0;
2899
2900 if (mapping && mapping->a_ops->releasepage)
2901 return mapping->a_ops->releasepage(page, gfp_mask);
2902 return try_to_free_buffers(page);
2903}
2904
2905EXPORT_SYMBOL(try_to_release_page);