blob: 3f9afded581be1a013bda4db2c0ec3a721323364 [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 Weiner22f2ac52016-09-30 15:11:29 -0700113static int page_cache_tree_insert(struct address_space *mapping,
114 struct page *page, void **shadowp)
115{
116 struct radix_tree_node *node;
117 void **slot;
118 int error;
119
120 error = __radix_tree_create(&mapping->page_tree, page->index, 0,
121 &node, &slot);
122 if (error)
123 return error;
124 if (*slot) {
125 void *p;
126
127 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
128 if (!radix_tree_exceptional_entry(p))
129 return -EEXIST;
130
131 mapping->nrexceptional--;
132 if (!dax_mapping(mapping)) {
133 if (shadowp)
134 *shadowp = p;
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700135 } else {
136 /* DAX can replace empty locked entry with a hole */
137 WARN_ON_ONCE(p !=
Ross Zwisler642261a2016-11-08 11:34:45 +1100138 dax_radix_locked_entry(0, RADIX_DAX_EMPTY));
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700139 /* Wakeup waiters for exceptional entry lock */
Ross Zwisler63e95b52016-11-08 11:32:20 +1100140 dax_wake_mapping_entry_waiter(mapping, page->index, p,
Ross Zwisler965d0042017-01-10 16:57:15 -0800141 true);
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700142 }
143 }
Johannes Weiner14b46872016-12-12 16:43:52 -0800144 __radix_tree_replace(&mapping->page_tree, node, slot, page,
145 workingset_update_node, mapping);
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700146 mapping->nrpages++;
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700147 return 0;
148}
149
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700150static void page_cache_tree_delete(struct address_space *mapping,
151 struct page *page, void *shadow)
152{
Kirill A. Shutemovc70b6472016-12-12 16:43:17 -0800153 int i, nr;
154
155 /* hugetlb pages are represented by one entry in the radix tree */
156 nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700157
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700158 VM_BUG_ON_PAGE(!PageLocked(page), page);
159 VM_BUG_ON_PAGE(PageTail(page), page);
160 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
Johannes Weiner449dd692014-04-03 14:47:56 -0700161
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700162 for (i = 0; i < nr; i++) {
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200163 struct radix_tree_node *node;
164 void **slot;
165
166 __radix_tree_lookup(&mapping->page_tree, page->index + i,
167 &node, &slot);
168
Johannes Weinerdbc446b2016-12-12 16:43:55 -0800169 VM_BUG_ON_PAGE(!node && nr != 1, page);
Johannes Weiner449dd692014-04-03 14:47:56 -0700170
Johannes Weiner14b46872016-12-12 16:43:52 -0800171 radix_tree_clear_tags(&mapping->page_tree, node, slot);
172 __radix_tree_replace(&mapping->page_tree, node, slot, shadow,
173 workingset_update_node, mapping);
Johannes Weiner449dd692014-04-03 14:47:56 -0700174 }
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200175
176 if (shadow) {
177 mapping->nrexceptional += nr;
178 /*
179 * Make sure the nrexceptional update is committed before
180 * the nrpages update so that final truncate racing
181 * with reclaim does not see both counters 0 at the
182 * same time and miss a shadow entry.
183 */
184 smp_wmb();
185 }
186 mapping->nrpages -= nr;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700187}
188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189/*
Minchan Kime64a7822011-03-22 16:32:44 -0700190 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 * sure the page is locked and that nobody else uses it - or that usage
Johannes Weinerfdf1cdb2016-03-15 14:57:25 -0700192 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 */
Johannes Weiner62cccb82016-03-15 14:57:22 -0700194void __delete_from_page_cache(struct page *page, void *shadow)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195{
196 struct address_space *mapping = page->mapping;
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700197 int nr = hpage_nr_pages(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700199 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600200 /*
201 * if we're uptodate, flush out into the cleancache, otherwise
202 * invalidate any existing cleancache entries. We can't leave
203 * stale data around in the cleancache once our page is gone
204 */
205 if (PageUptodate(page) && PageMappedToDisk(page))
206 cleancache_put_page(page);
207 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400208 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600209
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700210 VM_BUG_ON_PAGE(PageTail(page), page);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800211 VM_BUG_ON_PAGE(page_mapped(page), page);
212 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
213 int mapcount;
214
215 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
216 current->comm, page_to_pfn(page));
217 dump_page(page, "still mapped when deleted");
218 dump_stack();
219 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
220
221 mapcount = page_mapcount(page);
222 if (mapping_exiting(mapping) &&
223 page_count(page) >= mapcount + 2) {
224 /*
225 * All vmas have already been torn down, so it's
226 * a good bet that actually the page is unmapped,
227 * and we'd prefer not to leak it: if we're wrong,
228 * some other bad page check should catch it later.
229 */
230 page_mapcount_reset(page);
Joonsoo Kim6d061f92016-05-19 17:10:46 -0700231 page_ref_sub(page, mapcount);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800232 }
233 }
234
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700235 page_cache_tree_delete(mapping, page, shadow);
236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700238 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700239
Michal Hocko4165b9b2015-06-24 16:57:24 -0700240 /* hugetlb pages do not participate in page cache accounting. */
241 if (!PageHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700242 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700243 if (PageSwapBacked(page)) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700244 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700245 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700246 __dec_node_page_state(page, NR_SHMEM_THPS);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700247 } else {
248 VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
249 }
Linus Torvalds3a692792007-12-19 14:05:13 -0800250
251 /*
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700252 * At this point page must be either written or cleaned by truncate.
253 * Dirty page here signals a bug and loss of unwritten data.
Linus Torvalds3a692792007-12-19 14:05:13 -0800254 *
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700255 * This fixes dirty accounting after removing the page entirely but
256 * leaves PageDirty set: it has no effect for truncated page and
257 * anyway will be cleared before returning page into buddy allocator.
Linus Torvalds3a692792007-12-19 14:05:13 -0800258 */
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700259 if (WARN_ON_ONCE(PageDirty(page)))
Johannes Weiner62cccb82016-03-15 14:57:22 -0700260 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261}
262
Minchan Kim702cfbf2011-03-22 16:32:43 -0700263/**
264 * delete_from_page_cache - delete page from page cache
265 * @page: the page which the kernel is trying to remove from page cache
266 *
267 * This must be called only on pages that have been verified to be in the page
268 * cache and locked. It will never put the page into the free list, the caller
269 * has a reference on the page.
270 */
271void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700273 struct address_space *mapping = page_mapping(page);
Greg Thelenc4843a72015-05-22 17:13:16 -0400274 unsigned long flags;
Linus Torvalds6072d132010-12-01 13:35:19 -0500275 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276
Matt Mackallcd7619d2005-05-01 08:59:01 -0700277 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
Linus Torvalds6072d132010-12-01 13:35:19 -0500279 freepage = mapping->a_ops->freepage;
Greg Thelenc4843a72015-05-22 17:13:16 -0400280
Greg Thelenc4843a72015-05-22 17:13:16 -0400281 spin_lock_irqsave(&mapping->tree_lock, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700282 __delete_from_page_cache(page, NULL);
Greg Thelenc4843a72015-05-22 17:13:16 -0400283 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500284
285 if (freepage)
286 freepage(page);
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700287
288 if (PageTransHuge(page) && !PageHuge(page)) {
289 page_ref_sub(page, HPAGE_PMD_NR);
290 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
291 } else {
292 put_page(page);
293 }
Minchan Kim97cecb52011-03-22 16:30:53 -0700294}
295EXPORT_SYMBOL(delete_from_page_cache);
296
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200297int filemap_check_errors(struct address_space *mapping)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700298{
299 int ret = 0;
300 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700301 if (test_bit(AS_ENOSPC, &mapping->flags) &&
302 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700303 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700304 if (test_bit(AS_EIO, &mapping->flags) &&
305 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700306 ret = -EIO;
307 return ret;
308}
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200309EXPORT_SYMBOL(filemap_check_errors);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700312 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700313 * @mapping: address space structure to write
314 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800315 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700316 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700318 * Start writeback against all of a mapping's dirty pages that lie
319 * within the byte offsets <start, end> inclusive.
320 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700322 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 * these two operations is that if a dirty page/buffer is encountered, it must
324 * be waited upon, and not just skipped over.
325 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800326int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
327 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328{
329 int ret;
330 struct writeback_control wbc = {
331 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800332 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700333 .range_start = start,
334 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 };
336
337 if (!mapping_cap_writeback_dirty(mapping))
338 return 0;
339
Tejun Heob16b1de2015-06-02 08:39:48 -0600340 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 ret = do_writepages(mapping, &wbc);
Tejun Heob16b1de2015-06-02 08:39:48 -0600342 wbc_detach_inode(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 return ret;
344}
345
346static inline int __filemap_fdatawrite(struct address_space *mapping,
347 int sync_mode)
348{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700349 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350}
351
352int filemap_fdatawrite(struct address_space *mapping)
353{
354 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
355}
356EXPORT_SYMBOL(filemap_fdatawrite);
357
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400358int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800359 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360{
361 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
362}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400363EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
Randy Dunlap485bb992006-06-23 02:03:49 -0700365/**
366 * filemap_flush - mostly a non-blocking flush
367 * @mapping: target address_space
368 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 * This is a mostly non-blocking flush. Not suitable for data-integrity
370 * purposes - I/O may not be started against all dirty pages.
371 */
372int filemap_flush(struct address_space *mapping)
373{
374 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
375}
376EXPORT_SYMBOL(filemap_flush);
377
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800378static int __filemap_fdatawait_range(struct address_space *mapping,
379 loff_t start_byte, loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300381 pgoff_t index = start_byte >> PAGE_SHIFT;
382 pgoff_t end = end_byte >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 struct pagevec pvec;
384 int nr_pages;
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800385 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200387 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700388 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
390 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 while ((index <= end) &&
392 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
393 PAGECACHE_TAG_WRITEBACK,
394 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
395 unsigned i;
396
397 for (i = 0; i < nr_pages; i++) {
398 struct page *page = pvec.pages[i];
399
400 /* until radix tree lookup accepts end_index */
401 if (page->index > end)
402 continue;
403
404 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800405 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 ret = -EIO;
407 }
408 pagevec_release(&pvec);
409 cond_resched();
410 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700411out:
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800412 return ret;
413}
414
415/**
416 * filemap_fdatawait_range - wait for writeback to complete
417 * @mapping: address space structure to wait for
418 * @start_byte: offset in bytes where the range starts
419 * @end_byte: offset in bytes where the range ends (inclusive)
420 *
421 * Walk the list of under-writeback pages of the given address space
422 * in the given range and wait for all of them. Check error status of
423 * the address space and return it.
424 *
425 * Since the error status of the address space is cleared by this function,
426 * callers are responsible for checking the return value and handling and/or
427 * reporting the error.
428 */
429int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
430 loff_t end_byte)
431{
432 int ret, ret2;
433
434 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700435 ret2 = filemap_check_errors(mapping);
436 if (!ret)
437 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
439 return ret;
440}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200441EXPORT_SYMBOL(filemap_fdatawait_range);
442
443/**
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800444 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
445 * @mapping: address space structure to wait for
446 *
447 * Walk the list of under-writeback pages of the given address space
448 * and wait for all of them. Unlike filemap_fdatawait(), this function
449 * does not clear error status of the address space.
450 *
451 * Use this function if callers don't handle errors themselves. Expected
452 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
453 * fsfreeze(8)
454 */
455void filemap_fdatawait_keep_errors(struct address_space *mapping)
456{
457 loff_t i_size = i_size_read(mapping->host);
458
459 if (i_size == 0)
460 return;
461
462 __filemap_fdatawait_range(mapping, 0, i_size - 1);
463}
464
465/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700466 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700468 *
469 * Walk the list of under-writeback pages of the given address space
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800470 * and wait for all of them. Check error status of the address space
471 * and return it.
472 *
473 * Since the error status of the address space is cleared by this function,
474 * callers are responsible for checking the return value and handling and/or
475 * reporting the error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 */
477int filemap_fdatawait(struct address_space *mapping)
478{
479 loff_t i_size = i_size_read(mapping->host);
480
481 if (i_size == 0)
482 return 0;
483
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200484 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485}
486EXPORT_SYMBOL(filemap_fdatawait);
487
488int filemap_write_and_wait(struct address_space *mapping)
489{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800490 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800492 if ((!dax_mapping(mapping) && mapping->nrpages) ||
493 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800494 err = filemap_fdatawrite(mapping);
495 /*
496 * Even if the above returned error, the pages may be
497 * written partially (e.g. -ENOSPC), so we wait for it.
498 * But the -EIO is special case, it may indicate the worst
499 * thing (e.g. bug) happened, so we avoid waiting for it.
500 */
501 if (err != -EIO) {
502 int err2 = filemap_fdatawait(mapping);
503 if (!err)
504 err = err2;
505 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700506 } else {
507 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800509 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800511EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Randy Dunlap485bb992006-06-23 02:03:49 -0700513/**
514 * filemap_write_and_wait_range - write out & wait on a file range
515 * @mapping: the address_space for the pages
516 * @lstart: offset in bytes where the range starts
517 * @lend: offset in bytes where the range ends (inclusive)
518 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800519 * Write out and wait upon file offsets lstart->lend, inclusive.
520 *
521 * Note that `lend' is inclusive (describes the last byte to be written) so
522 * that this function can be used to write to the very end-of-file (end = -1).
523 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524int filemap_write_and_wait_range(struct address_space *mapping,
525 loff_t lstart, loff_t lend)
526{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800527 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800529 if ((!dax_mapping(mapping) && mapping->nrpages) ||
530 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800531 err = __filemap_fdatawrite_range(mapping, lstart, lend,
532 WB_SYNC_ALL);
533 /* See comment of filemap_write_and_wait() */
534 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200535 int err2 = filemap_fdatawait_range(mapping,
536 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800537 if (!err)
538 err = err2;
539 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700540 } else {
541 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800543 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544}
Chris Masonf6995582009-04-15 13:22:37 -0400545EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546
Randy Dunlap485bb992006-06-23 02:03:49 -0700547/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700548 * replace_page_cache_page - replace a pagecache page with a new one
549 * @old: page to be replaced
550 * @new: page to replace with
551 * @gfp_mask: allocation mode
552 *
553 * This function replaces a page in the pagecache with a new one. On
554 * success it acquires the pagecache reference for the new page and
555 * drops it for the old page. Both the old and new pages must be
556 * locked. This function does not add the new page to the LRU, the
557 * caller must do that.
558 *
559 * The remove + add is atomic. The only way this function can fail is
560 * memory allocation failure.
561 */
562int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
563{
564 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700565
Sasha Levin309381fea2014-01-23 15:52:54 -0800566 VM_BUG_ON_PAGE(!PageLocked(old), old);
567 VM_BUG_ON_PAGE(!PageLocked(new), new);
568 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700569
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700570 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
571 if (!error) {
572 struct address_space *mapping = old->mapping;
573 void (*freepage)(struct page *);
Greg Thelenc4843a72015-05-22 17:13:16 -0400574 unsigned long flags;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700575
576 pgoff_t offset = old->index;
577 freepage = mapping->a_ops->freepage;
578
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300579 get_page(new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700580 new->mapping = mapping;
581 new->index = offset;
582
Greg Thelenc4843a72015-05-22 17:13:16 -0400583 spin_lock_irqsave(&mapping->tree_lock, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700584 __delete_from_page_cache(old, NULL);
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700585 error = page_cache_tree_insert(mapping, new, NULL);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700586 BUG_ON(error);
Michal Hocko4165b9b2015-06-24 16:57:24 -0700587
588 /*
589 * hugetlb pages do not participate in page cache accounting.
590 */
591 if (!PageHuge(new))
Mel Gorman11fb9982016-07-28 15:46:20 -0700592 __inc_node_page_state(new, NR_FILE_PAGES);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700593 if (PageSwapBacked(new))
Mel Gorman11fb9982016-07-28 15:46:20 -0700594 __inc_node_page_state(new, NR_SHMEM);
Greg Thelenc4843a72015-05-22 17:13:16 -0400595 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700596 mem_cgroup_migrate(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700597 radix_tree_preload_end();
598 if (freepage)
599 freepage(old);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300600 put_page(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700601 }
602
603 return error;
604}
605EXPORT_SYMBOL_GPL(replace_page_cache_page);
606
Johannes Weinera5289102014-04-03 14:47:51 -0700607static int __add_to_page_cache_locked(struct page *page,
608 struct address_space *mapping,
609 pgoff_t offset, gfp_t gfp_mask,
610 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
Johannes Weiner00501b52014-08-08 14:19:20 -0700612 int huge = PageHuge(page);
613 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700614 int error;
615
Sasha Levin309381fea2014-01-23 15:52:54 -0800616 VM_BUG_ON_PAGE(!PageLocked(page), page);
617 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700618
Johannes Weiner00501b52014-08-08 14:19:20 -0700619 if (!huge) {
620 error = mem_cgroup_try_charge(page, current->mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800621 gfp_mask, &memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -0700622 if (error)
623 return error;
624 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
Jan Kara5e4c0d972013-09-11 14:26:05 -0700626 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700627 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700628 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800629 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700630 return error;
631 }
632
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300633 get_page(page);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700634 page->mapping = mapping;
635 page->index = offset;
636
637 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700638 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700639 radix_tree_preload_end();
640 if (unlikely(error))
641 goto err_insert;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700642
643 /* hugetlb pages do not participate in page cache accounting. */
644 if (!huge)
Mel Gorman11fb9982016-07-28 15:46:20 -0700645 __inc_node_page_state(page, NR_FILE_PAGES);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700646 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700647 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800648 mem_cgroup_commit_charge(page, memcg, false, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700649 trace_mm_filemap_add_to_page_cache(page);
650 return 0;
651err_insert:
652 page->mapping = NULL;
653 /* Leave page->index set: truncation relies upon it */
654 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700655 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800656 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300657 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 return error;
659}
Johannes Weinera5289102014-04-03 14:47:51 -0700660
661/**
662 * add_to_page_cache_locked - add a locked page to the pagecache
663 * @page: page to add
664 * @mapping: the page's address_space
665 * @offset: page index
666 * @gfp_mask: page allocation mode
667 *
668 * This function is used to add a page to the pagecache. It must be locked.
669 * This function does not add the page to the LRU. The caller must do that.
670 */
671int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
672 pgoff_t offset, gfp_t gfp_mask)
673{
674 return __add_to_page_cache_locked(page, mapping, offset,
675 gfp_mask, NULL);
676}
Nick Piggine2867812008-07-25 19:45:30 -0700677EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
679int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400680 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
Johannes Weinera5289102014-04-03 14:47:51 -0700682 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700683 int ret;
684
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800685 __SetPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700686 ret = __add_to_page_cache_locked(page, mapping, offset,
687 gfp_mask, &shadow);
688 if (unlikely(ret))
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800689 __ClearPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700690 else {
691 /*
692 * The page might have been evicted from cache only
693 * recently, in which case it should be activated like
694 * any other repeatedly accessed page.
Rik van Rielf0281a02016-05-20 16:56:25 -0700695 * The exception is pages getting rewritten; evicting other
696 * data from the working set, only to cache data that will
697 * get overwritten with something else, is a waste of memory.
Johannes Weinera5289102014-04-03 14:47:51 -0700698 */
Rik van Rielf0281a02016-05-20 16:56:25 -0700699 if (!(gfp_mask & __GFP_WRITE) &&
700 shadow && workingset_refault(shadow)) {
Johannes Weinera5289102014-04-03 14:47:51 -0700701 SetPageActive(page);
702 workingset_activation(page);
703 } else
704 ClearPageActive(page);
705 lru_cache_add(page);
706 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 return ret;
708}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300709EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
Paul Jackson44110fe2006-03-24 03:16:04 -0800711#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700712struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800713{
Miao Xiec0ff7452010-05-24 14:32:08 -0700714 int n;
715 struct page *page;
716
Paul Jackson44110fe2006-03-24 03:16:04 -0800717 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700718 unsigned int cpuset_mems_cookie;
719 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700720 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700721 n = cpuset_mem_spread_node();
Vlastimil Babka96db8002015-09-08 15:03:50 -0700722 page = __alloc_pages_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700723 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700724
Miao Xiec0ff7452010-05-24 14:32:08 -0700725 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800726 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700727 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800728}
Nick Piggin2ae88142006-10-28 10:38:23 -0700729EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800730#endif
731
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732/*
733 * In order to wait for pages to become available there must be
734 * waitqueues associated with pages. By using a hash table of
735 * waitqueues where the bucket discipline is to maintain all
736 * waiters on the same queue and wake all when any of the pages
737 * become available, and for the woken contexts to check to be
738 * sure the appropriate page became available, this saves space
739 * at a cost of "thundering herd" phenomena during rare hash
740 * collisions.
741 */
Nicholas Piggin62906022016-12-25 13:00:30 +1000742#define PAGE_WAIT_TABLE_BITS 8
743#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
744static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
745
746static wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747{
Nicholas Piggin62906022016-12-25 13:00:30 +1000748 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749}
Nicholas Piggin62906022016-12-25 13:00:30 +1000750
751void __init pagecache_init(void)
752{
753 int i;
754
755 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
756 init_waitqueue_head(&page_wait_table[i]);
757
758 page_writeback_init();
759}
760
761struct wait_page_key {
762 struct page *page;
763 int bit_nr;
764 int page_match;
765};
766
767struct wait_page_queue {
768 struct page *page;
769 int bit_nr;
770 wait_queue_t wait;
771};
772
773static int wake_page_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
774{
775 struct wait_page_key *key = arg;
776 struct wait_page_queue *wait_page
777 = container_of(wait, struct wait_page_queue, wait);
778
779 if (wait_page->page != key->page)
780 return 0;
781 key->page_match = 1;
782
783 if (wait_page->bit_nr != key->bit_nr)
784 return 0;
785 if (test_bit(key->bit_nr, &key->page->flags))
786 return 0;
787
788 return autoremove_wake_function(wait, mode, sync, key);
789}
790
791void wake_up_page_bit(struct page *page, int bit_nr)
792{
793 wait_queue_head_t *q = page_waitqueue(page);
794 struct wait_page_key key;
795 unsigned long flags;
796
797 key.page = page;
798 key.bit_nr = bit_nr;
799 key.page_match = 0;
800
801 spin_lock_irqsave(&q->lock, flags);
802 __wake_up_locked_key(q, TASK_NORMAL, &key);
803 /*
804 * It is possible for other pages to have collided on the waitqueue
805 * hash, so in that case check for a page match. That prevents a long-
806 * term waiter
807 *
808 * It is still possible to miss a case here, when we woke page waiters
809 * and removed them from the waitqueue, but there are still other
810 * page waiters.
811 */
812 if (!waitqueue_active(q) || !key.page_match) {
813 ClearPageWaiters(page);
814 /*
815 * It's possible to miss clearing Waiters here, when we woke
816 * our page waiters, but the hashed waitqueue has waiters for
817 * other pages on it.
818 *
819 * That's okay, it's a rare case. The next waker will clear it.
820 */
821 }
822 spin_unlock_irqrestore(&q->lock, flags);
823}
824EXPORT_SYMBOL(wake_up_page_bit);
825
826static inline int wait_on_page_bit_common(wait_queue_head_t *q,
827 struct page *page, int bit_nr, int state, bool lock)
828{
829 struct wait_page_queue wait_page;
830 wait_queue_t *wait = &wait_page.wait;
831 int ret = 0;
832
833 init_wait(wait);
834 wait->func = wake_page_function;
835 wait_page.page = page;
836 wait_page.bit_nr = bit_nr;
837
838 for (;;) {
839 spin_lock_irq(&q->lock);
840
841 if (likely(list_empty(&wait->task_list))) {
842 if (lock)
843 __add_wait_queue_tail_exclusive(q, wait);
844 else
845 __add_wait_queue(q, wait);
846 SetPageWaiters(page);
847 }
848
849 set_current_state(state);
850
851 spin_unlock_irq(&q->lock);
852
853 if (likely(test_bit(bit_nr, &page->flags))) {
854 io_schedule();
855 if (unlikely(signal_pending_state(state, current))) {
856 ret = -EINTR;
857 break;
858 }
859 }
860
861 if (lock) {
862 if (!test_and_set_bit_lock(bit_nr, &page->flags))
863 break;
864 } else {
865 if (!test_bit(bit_nr, &page->flags))
866 break;
867 }
868 }
869
870 finish_wait(q, wait);
871
872 /*
873 * A signal could leave PageWaiters set. Clearing it here if
874 * !waitqueue_active would be possible (by open-coding finish_wait),
875 * but still fail to catch it in the case of wait hash collision. We
876 * already can fail to clear wait hash collision cases, so don't
877 * bother with signals either.
878 */
879
880 return ret;
881}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
Harvey Harrison920c7a52008-02-04 22:29:26 -0800883void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884{
Nicholas Piggin62906022016-12-25 13:00:30 +1000885 wait_queue_head_t *q = page_waitqueue(page);
886 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887}
888EXPORT_SYMBOL(wait_on_page_bit);
889
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700890int wait_on_page_bit_killable(struct page *page, int bit_nr)
891{
Nicholas Piggin62906022016-12-25 13:00:30 +1000892 wait_queue_head_t *q = page_waitqueue(page);
893 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700894}
895
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100897 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700898 * @page: Page defining the wait queue of interest
899 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100900 *
901 * Add an arbitrary @waiter to the wait queue for the nominated @page.
902 */
903void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
904{
905 wait_queue_head_t *q = page_waitqueue(page);
906 unsigned long flags;
907
908 spin_lock_irqsave(&q->lock, flags);
909 __add_wait_queue(q, waiter);
Nicholas Piggin62906022016-12-25 13:00:30 +1000910 SetPageWaiters(page);
David Howells385e1ca5f2009-04-03 16:42:39 +0100911 spin_unlock_irqrestore(&q->lock, flags);
912}
913EXPORT_SYMBOL_GPL(add_page_wait_queue);
914
Linus Torvaldsb91e1302016-12-27 11:40:38 -0800915#ifndef clear_bit_unlock_is_negative_byte
916
917/*
918 * PG_waiters is the high bit in the same byte as PG_lock.
919 *
920 * On x86 (and on many other architectures), we can clear PG_lock and
921 * test the sign bit at the same time. But if the architecture does
922 * not support that special operation, we just do this all by hand
923 * instead.
924 *
925 * The read of PG_waiters has to be after (or concurrently with) PG_locked
926 * being cleared, but a memory barrier should be unneccssary since it is
927 * in the same byte as PG_locked.
928 */
929static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
930{
931 clear_bit_unlock(nr, mem);
932 /* smp_mb__after_atomic(); */
Olof Johansson98473f92016-12-29 14:16:07 -0800933 return test_bit(PG_waiters, mem);
Linus Torvaldsb91e1302016-12-27 11:40:38 -0800934}
935
936#endif
937
David Howells385e1ca5f2009-04-03 16:42:39 +0100938/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700939 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 * @page: the page
941 *
942 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
943 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +0900944 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
946 *
Linus Torvaldsb91e1302016-12-27 11:40:38 -0800947 * Note that this depends on PG_waiters being the sign bit in the byte
948 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
949 * clear the PG_locked bit and test PG_waiters at the same time fairly
950 * portably (architectures that do LL/SC can test any bit, while x86 can
951 * test the sign bit).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800953void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954{
Linus Torvaldsb91e1302016-12-27 11:40:38 -0800955 BUILD_BUG_ON(PG_waiters != 7);
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800956 page = compound_head(page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800957 VM_BUG_ON_PAGE(!PageLocked(page), page);
Linus Torvaldsb91e1302016-12-27 11:40:38 -0800958 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
959 wake_up_page_bit(page, PG_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960}
961EXPORT_SYMBOL(unlock_page);
962
Randy Dunlap485bb992006-06-23 02:03:49 -0700963/**
964 * end_page_writeback - end writeback against a page
965 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 */
967void end_page_writeback(struct page *page)
968{
Mel Gorman888cf2d2014-06-04 16:10:34 -0700969 /*
970 * TestClearPageReclaim could be used here but it is an atomic
971 * operation and overkill in this particular case. Failing to
972 * shuffle a page marked for immediate reclaim is too mild to
973 * justify taking an atomic operation penalty at the end of
974 * ever page writeback.
975 */
976 if (PageReclaim(page)) {
977 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700978 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -0700979 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700980
981 if (!test_clear_page_writeback(page))
982 BUG();
983
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100984 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 wake_up_page(page, PG_writeback);
986}
987EXPORT_SYMBOL(end_page_writeback);
988
Matthew Wilcox57d99842014-06-04 16:07:45 -0700989/*
990 * After completing I/O on a page, call this routine to update the page
991 * flags appropriately
992 */
Jens Axboec11f0c02016-08-05 08:11:04 -0600993void page_endio(struct page *page, bool is_write, int err)
Matthew Wilcox57d99842014-06-04 16:07:45 -0700994{
Jens Axboec11f0c02016-08-05 08:11:04 -0600995 if (!is_write) {
Matthew Wilcox57d99842014-06-04 16:07:45 -0700996 if (!err) {
997 SetPageUptodate(page);
998 } else {
999 ClearPageUptodate(page);
1000 SetPageError(page);
1001 }
1002 unlock_page(page);
Mike Christieabf54542016-08-04 14:23:34 -06001003 } else {
Matthew Wilcox57d99842014-06-04 16:07:45 -07001004 if (err) {
1005 SetPageError(page);
1006 if (page->mapping)
1007 mapping_set_error(page->mapping, err);
1008 }
1009 end_page_writeback(page);
1010 }
1011}
1012EXPORT_SYMBOL_GPL(page_endio);
1013
Randy Dunlap485bb992006-06-23 02:03:49 -07001014/**
1015 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1016 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 */
Nicholas Piggin62906022016-12-25 13:00:30 +10001018void __lock_page(struct page *__page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019{
Nicholas Piggin62906022016-12-25 13:00:30 +10001020 struct page *page = compound_head(__page);
1021 wait_queue_head_t *q = page_waitqueue(page);
1022 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023}
1024EXPORT_SYMBOL(__lock_page);
1025
Nicholas Piggin62906022016-12-25 13:00:30 +10001026int __lock_page_killable(struct page *__page)
Matthew Wilcox2687a352007-12-06 11:18:49 -05001027{
Nicholas Piggin62906022016-12-25 13:00:30 +10001028 struct page *page = compound_head(__page);
1029 wait_queue_head_t *q = page_waitqueue(page);
1030 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
Matthew Wilcox2687a352007-12-06 11:18:49 -05001031}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +03001032EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -05001033
Paul Cassella9a95f3c2014-08-06 16:07:24 -07001034/*
1035 * Return values:
1036 * 1 - page is locked; mmap_sem is still held.
1037 * 0 - page is not locked.
1038 * mmap_sem has been released (up_read()), unless flags had both
1039 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1040 * which case mmap_sem is still held.
1041 *
1042 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1043 * with the page locked and the mmap_sem unperturbed.
1044 */
Michel Lespinassed065bd82010-10-26 14:21:57 -07001045int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1046 unsigned int flags)
1047{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001048 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1049 /*
1050 * CAUTION! In this case, mmap_sem is not released
1051 * even though return 0.
1052 */
1053 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1054 return 0;
1055
1056 up_read(&mm->mmap_sem);
1057 if (flags & FAULT_FLAG_KILLABLE)
1058 wait_on_page_locked_killable(page);
1059 else
Gleb Natapov318b2752011-03-22 16:30:51 -07001060 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001061 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -07001062 } else {
1063 if (flags & FAULT_FLAG_KILLABLE) {
1064 int ret;
1065
1066 ret = __lock_page_killable(page);
1067 if (ret) {
1068 up_read(&mm->mmap_sem);
1069 return 0;
1070 }
1071 } else
1072 __lock_page(page);
1073 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -07001074 }
1075}
1076
Randy Dunlap485bb992006-06-23 02:03:49 -07001077/**
Johannes Weinere7b563b2014-04-03 14:47:44 -07001078 * page_cache_next_hole - find the next hole (not-present entry)
1079 * @mapping: mapping
1080 * @index: index
1081 * @max_scan: maximum range to search
1082 *
1083 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1084 * lowest indexed hole.
1085 *
1086 * Returns: the index of the hole if found, otherwise returns an index
1087 * outside of the set specified (in which case 'return - index >=
1088 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1089 * be returned.
1090 *
1091 * page_cache_next_hole may be called under rcu_read_lock. However,
1092 * like radix_tree_gang_lookup, this will not atomically search a
1093 * snapshot of the tree at a single point in time. For example, if a
1094 * hole is created at index 5, then subsequently a hole is created at
1095 * index 10, page_cache_next_hole covering both indexes may return 10
1096 * if called under rcu_read_lock.
1097 */
1098pgoff_t page_cache_next_hole(struct address_space *mapping,
1099 pgoff_t index, unsigned long max_scan)
1100{
1101 unsigned long i;
1102
1103 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001104 struct page *page;
1105
1106 page = radix_tree_lookup(&mapping->page_tree, index);
1107 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001108 break;
1109 index++;
1110 if (index == 0)
1111 break;
1112 }
1113
1114 return index;
1115}
1116EXPORT_SYMBOL(page_cache_next_hole);
1117
1118/**
1119 * page_cache_prev_hole - find the prev hole (not-present entry)
1120 * @mapping: mapping
1121 * @index: index
1122 * @max_scan: maximum range to search
1123 *
1124 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1125 * the first hole.
1126 *
1127 * Returns: the index of the hole if found, otherwise returns an index
1128 * outside of the set specified (in which case 'index - return >=
1129 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1130 * will be returned.
1131 *
1132 * page_cache_prev_hole may be called under rcu_read_lock. However,
1133 * like radix_tree_gang_lookup, this will not atomically search a
1134 * snapshot of the tree at a single point in time. For example, if a
1135 * hole is created at index 10, then subsequently a hole is created at
1136 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1137 * called under rcu_read_lock.
1138 */
1139pgoff_t page_cache_prev_hole(struct address_space *mapping,
1140 pgoff_t index, unsigned long max_scan)
1141{
1142 unsigned long i;
1143
1144 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001145 struct page *page;
1146
1147 page = radix_tree_lookup(&mapping->page_tree, index);
1148 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001149 break;
1150 index--;
1151 if (index == ULONG_MAX)
1152 break;
1153 }
1154
1155 return index;
1156}
1157EXPORT_SYMBOL(page_cache_prev_hole);
1158
1159/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001160 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -07001161 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -07001162 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -07001163 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001164 * Looks up the page cache slot at @mapping & @offset. If there is a
1165 * page cache page, it is returned with an increased refcount.
1166 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001167 * If the slot holds a shadow entry of a previously evicted page, or a
1168 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001169 *
1170 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 */
Johannes Weiner0cd61442014-04-03 14:47:46 -07001172struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173{
Nick Piggina60637c2008-07-25 19:45:31 -07001174 void **pagep;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001175 struct page *head, *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Nick Piggina60637c2008-07-25 19:45:31 -07001177 rcu_read_lock();
1178repeat:
1179 page = NULL;
1180 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1181 if (pagep) {
1182 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -08001183 if (unlikely(!page))
1184 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001185 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001186 if (radix_tree_deref_retry(page))
1187 goto repeat;
1188 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001189 * A shadow entry of a recently evicted page,
1190 * or a swap entry from shmem/tmpfs. Return
1191 * it without attempting to raise page count.
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001192 */
1193 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001194 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001195
1196 head = compound_head(page);
1197 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001198 goto repeat;
1199
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001200 /* The page was split under us? */
1201 if (compound_head(page) != head) {
1202 put_page(head);
1203 goto repeat;
1204 }
1205
Nick Piggina60637c2008-07-25 19:45:31 -07001206 /*
1207 * Has the page moved?
1208 * This is part of the lockless pagecache protocol. See
1209 * include/linux/pagemap.h for details.
1210 */
1211 if (unlikely(page != *pagep)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001212 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001213 goto repeat;
1214 }
1215 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001216out:
Nick Piggina60637c2008-07-25 19:45:31 -07001217 rcu_read_unlock();
1218
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 return page;
1220}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001221EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222
Randy Dunlap485bb992006-06-23 02:03:49 -07001223/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001224 * find_lock_entry - locate, pin and lock a page cache entry
1225 * @mapping: the address_space to search
1226 * @offset: the page cache index
1227 *
1228 * Looks up the page cache slot at @mapping & @offset. If there is a
1229 * page cache page, it is returned locked and with an increased
1230 * refcount.
1231 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001232 * If the slot holds a shadow entry of a previously evicted page, or a
1233 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001234 *
1235 * Otherwise, %NULL is returned.
1236 *
1237 * find_lock_entry() may sleep.
1238 */
1239struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240{
1241 struct page *page;
1242
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001244 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001245 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001246 lock_page(page);
1247 /* Has the page been truncated? */
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001248 if (unlikely(page_mapping(page) != mapping)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001249 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001250 put_page(page);
Nick Piggina60637c2008-07-25 19:45:31 -07001251 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001253 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 return page;
1256}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001257EXPORT_SYMBOL(find_lock_entry);
1258
1259/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001260 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001261 * @mapping: the address_space to search
1262 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001263 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001264 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001265 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001266 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001267 *
Randy Dunlap75325182014-07-30 16:08:37 -07001268 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001269 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001270 * FGP_ACCESSED: the page will be marked accessed
1271 * FGP_LOCK: Page is return locked
1272 * FGP_CREAT: If page is not present then a new page is allocated using
Michal Hocko45f87de2014-12-29 20:30:35 +01001273 * @gfp_mask and added to the page cache and the VM's LRU
1274 * list. The page is returned locked and with an increased
1275 * refcount. Otherwise, %NULL is returned.
Mel Gorman2457aec2014-06-04 16:10:31 -07001276 *
1277 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1278 * if the GFP flags specified for FGP_CREAT are atomic.
1279 *
1280 * If there is a page cache page, it is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001281 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001282struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001283 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284{
Nick Piggineb2be182007-10-16 01:24:57 -07001285 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001286
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001288 page = find_get_entry(mapping, offset);
1289 if (radix_tree_exceptional_entry(page))
1290 page = NULL;
1291 if (!page)
1292 goto no_page;
1293
1294 if (fgp_flags & FGP_LOCK) {
1295 if (fgp_flags & FGP_NOWAIT) {
1296 if (!trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001297 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001298 return NULL;
1299 }
1300 } else {
1301 lock_page(page);
1302 }
1303
1304 /* Has the page been truncated? */
1305 if (unlikely(page->mapping != mapping)) {
1306 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001307 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001308 goto repeat;
1309 }
1310 VM_BUG_ON_PAGE(page->index != offset, page);
1311 }
1312
1313 if (page && (fgp_flags & FGP_ACCESSED))
1314 mark_page_accessed(page);
1315
1316no_page:
1317 if (!page && (fgp_flags & FGP_CREAT)) {
1318 int err;
1319 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001320 gfp_mask |= __GFP_WRITE;
1321 if (fgp_flags & FGP_NOFS)
1322 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001323
Michal Hocko45f87de2014-12-29 20:30:35 +01001324 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001325 if (!page)
1326 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001327
1328 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1329 fgp_flags |= FGP_LOCK;
1330
Hugh Dickinseb39d612014-08-06 16:06:43 -07001331 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001332 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001333 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001334
Michal Hocko45f87de2014-12-29 20:30:35 +01001335 err = add_to_page_cache_lru(page, mapping, offset,
1336 gfp_mask & GFP_RECLAIM_MASK);
Nick Piggineb2be182007-10-16 01:24:57 -07001337 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001338 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07001339 page = NULL;
1340 if (err == -EEXIST)
1341 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001344
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 return page;
1346}
Mel Gorman2457aec2014-06-04 16:10:31 -07001347EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
1349/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001350 * find_get_entries - gang pagecache lookup
1351 * @mapping: The address_space to search
1352 * @start: The starting page cache index
1353 * @nr_entries: The maximum number of entries
1354 * @entries: Where the resulting entries are placed
1355 * @indices: The cache indices corresponding to the entries in @entries
1356 *
1357 * find_get_entries() will search for and return a group of up to
1358 * @nr_entries entries in the mapping. The entries are placed at
1359 * @entries. find_get_entries() takes a reference against any actual
1360 * pages it returns.
1361 *
1362 * The search returns a group of mapping-contiguous page cache entries
1363 * with ascending indexes. There may be holes in the indices due to
1364 * not-present pages.
1365 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001366 * Any shadow entries of evicted pages, or swap entries from
1367 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001368 *
1369 * find_get_entries() returns the number of pages and shadow entries
1370 * which were found.
1371 */
1372unsigned find_get_entries(struct address_space *mapping,
1373 pgoff_t start, unsigned int nr_entries,
1374 struct page **entries, pgoff_t *indices)
1375{
1376 void **slot;
1377 unsigned int ret = 0;
1378 struct radix_tree_iter iter;
1379
1380 if (!nr_entries)
1381 return 0;
1382
1383 rcu_read_lock();
Johannes Weiner0cd61442014-04-03 14:47:46 -07001384 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001385 struct page *head, *page;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001386repeat:
1387 page = radix_tree_deref_slot(slot);
1388 if (unlikely(!page))
1389 continue;
1390 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001391 if (radix_tree_deref_retry(page)) {
1392 slot = radix_tree_iter_retry(&iter);
1393 continue;
1394 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001395 /*
Ross Zwislerf9fe48b2016-01-22 15:10:40 -08001396 * A shadow entry of a recently evicted page, a swap
1397 * entry from shmem/tmpfs or a DAX entry. Return it
1398 * without attempting to raise page count.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001399 */
1400 goto export;
1401 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001402
1403 head = compound_head(page);
1404 if (!page_cache_get_speculative(head))
Johannes Weiner0cd61442014-04-03 14:47:46 -07001405 goto repeat;
1406
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001407 /* The page was split under us? */
1408 if (compound_head(page) != head) {
1409 put_page(head);
1410 goto repeat;
1411 }
1412
Johannes Weiner0cd61442014-04-03 14:47:46 -07001413 /* Has the page moved? */
1414 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001415 put_page(head);
Johannes Weiner0cd61442014-04-03 14:47:46 -07001416 goto repeat;
1417 }
1418export:
1419 indices[ret] = iter.index;
1420 entries[ret] = page;
1421 if (++ret == nr_entries)
1422 break;
1423 }
1424 rcu_read_unlock();
1425 return ret;
1426}
1427
1428/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 * find_get_pages - gang pagecache lookup
1430 * @mapping: The address_space to search
1431 * @start: The starting page index
1432 * @nr_pages: The maximum number of pages
1433 * @pages: Where the resulting pages are placed
1434 *
1435 * find_get_pages() will search for and return a group of up to
1436 * @nr_pages pages in the mapping. The pages are placed at @pages.
1437 * find_get_pages() takes a reference against the returned pages.
1438 *
1439 * The search returns a group of mapping-contiguous pages with ascending
1440 * indexes. There may be holes in the indices due to not-present pages.
1441 *
1442 * find_get_pages() returns the number of pages which were found.
1443 */
1444unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1445 unsigned int nr_pages, struct page **pages)
1446{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001447 struct radix_tree_iter iter;
1448 void **slot;
1449 unsigned ret = 0;
1450
1451 if (unlikely(!nr_pages))
1452 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453
Nick Piggina60637c2008-07-25 19:45:31 -07001454 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001455 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001456 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001457repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001458 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001459 if (unlikely(!page))
1460 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001461
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001462 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001463 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001464 slot = radix_tree_iter_retry(&iter);
1465 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001466 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001467 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001468 * A shadow entry of a recently evicted page,
1469 * or a swap entry from shmem/tmpfs. Skip
1470 * over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001471 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001472 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001473 }
Nick Piggina60637c2008-07-25 19:45:31 -07001474
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001475 head = compound_head(page);
1476 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001477 goto repeat;
1478
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001479 /* The page was split under us? */
1480 if (compound_head(page) != head) {
1481 put_page(head);
1482 goto repeat;
1483 }
1484
Nick Piggina60637c2008-07-25 19:45:31 -07001485 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001486 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001487 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001488 goto repeat;
1489 }
1490
1491 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001492 if (++ret == nr_pages)
1493 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001494 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001495
Nick Piggina60637c2008-07-25 19:45:31 -07001496 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 return ret;
1498}
1499
Jens Axboeebf43502006-04-27 08:46:01 +02001500/**
1501 * find_get_pages_contig - gang contiguous pagecache lookup
1502 * @mapping: The address_space to search
1503 * @index: The starting page index
1504 * @nr_pages: The maximum number of pages
1505 * @pages: Where the resulting pages are placed
1506 *
1507 * find_get_pages_contig() works exactly like find_get_pages(), except
1508 * that the returned number of pages are guaranteed to be contiguous.
1509 *
1510 * find_get_pages_contig() returns the number of pages which were found.
1511 */
1512unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1513 unsigned int nr_pages, struct page **pages)
1514{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001515 struct radix_tree_iter iter;
1516 void **slot;
1517 unsigned int ret = 0;
1518
1519 if (unlikely(!nr_pages))
1520 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001521
Nick Piggina60637c2008-07-25 19:45:31 -07001522 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001523 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001524 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001525repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001526 page = radix_tree_deref_slot(slot);
1527 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001528 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001529 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001530
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001531 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001532 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001533 slot = radix_tree_iter_retry(&iter);
1534 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001535 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001536 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001537 * A shadow entry of a recently evicted page,
1538 * or a swap entry from shmem/tmpfs. Stop
1539 * looking for contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001540 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001541 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001542 }
Nick Piggina60637c2008-07-25 19:45:31 -07001543
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001544 head = compound_head(page);
1545 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001546 goto repeat;
1547
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001548 /* The page was split under us? */
1549 if (compound_head(page) != head) {
1550 put_page(head);
1551 goto repeat;
1552 }
1553
Nick Piggina60637c2008-07-25 19:45:31 -07001554 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001555 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001556 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001557 goto repeat;
1558 }
1559
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001560 /*
1561 * must check mapping and index after taking the ref.
1562 * otherwise we can get both false positives and false
1563 * negatives, which is just confusing to the caller.
1564 */
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001565 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001566 put_page(page);
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001567 break;
1568 }
1569
Nick Piggina60637c2008-07-25 19:45:31 -07001570 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001571 if (++ret == nr_pages)
1572 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001573 }
Nick Piggina60637c2008-07-25 19:45:31 -07001574 rcu_read_unlock();
1575 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001576}
David Howellsef71c152007-05-09 02:33:44 -07001577EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001578
Randy Dunlap485bb992006-06-23 02:03:49 -07001579/**
1580 * find_get_pages_tag - find and return pages that match @tag
1581 * @mapping: the address_space to search
1582 * @index: the starting page index
1583 * @tag: the tag index
1584 * @nr_pages: the maximum number of pages
1585 * @pages: where the resulting pages are placed
1586 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001588 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 */
1590unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1591 int tag, unsigned int nr_pages, struct page **pages)
1592{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001593 struct radix_tree_iter iter;
1594 void **slot;
1595 unsigned ret = 0;
1596
1597 if (unlikely(!nr_pages))
1598 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
Nick Piggina60637c2008-07-25 19:45:31 -07001600 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001601 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1602 &iter, *index, tag) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001603 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001604repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001605 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001606 if (unlikely(!page))
1607 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001608
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001609 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001610 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001611 slot = radix_tree_iter_retry(&iter);
1612 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001613 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001614 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001615 * A shadow entry of a recently evicted page.
1616 *
1617 * Those entries should never be tagged, but
1618 * this tree walk is lockless and the tags are
1619 * looked up in bulk, one radix tree node at a
1620 * time, so there is a sizable window for page
1621 * reclaim to evict a page we saw tagged.
1622 *
1623 * Skip over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001624 */
Johannes Weiner139b6a62014-05-06 12:50:05 -07001625 continue;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001626 }
Nick Piggina60637c2008-07-25 19:45:31 -07001627
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001628 head = compound_head(page);
1629 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001630 goto repeat;
1631
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001632 /* The page was split under us? */
1633 if (compound_head(page) != head) {
1634 put_page(head);
1635 goto repeat;
1636 }
1637
Nick Piggina60637c2008-07-25 19:45:31 -07001638 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001639 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001640 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001641 goto repeat;
1642 }
1643
1644 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001645 if (++ret == nr_pages)
1646 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001647 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001648
Nick Piggina60637c2008-07-25 19:45:31 -07001649 rcu_read_unlock();
1650
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 if (ret)
1652 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001653
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 return ret;
1655}
David Howellsef71c152007-05-09 02:33:44 -07001656EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001658/**
1659 * find_get_entries_tag - find and return entries that match @tag
1660 * @mapping: the address_space to search
1661 * @start: the starting page cache index
1662 * @tag: the tag index
1663 * @nr_entries: the maximum number of entries
1664 * @entries: where the resulting entries are placed
1665 * @indices: the cache indices corresponding to the entries in @entries
1666 *
1667 * Like find_get_entries, except we only return entries which are tagged with
1668 * @tag.
1669 */
1670unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1671 int tag, unsigned int nr_entries,
1672 struct page **entries, pgoff_t *indices)
1673{
1674 void **slot;
1675 unsigned int ret = 0;
1676 struct radix_tree_iter iter;
1677
1678 if (!nr_entries)
1679 return 0;
1680
1681 rcu_read_lock();
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001682 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1683 &iter, start, tag) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001684 struct page *head, *page;
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001685repeat:
1686 page = radix_tree_deref_slot(slot);
1687 if (unlikely(!page))
1688 continue;
1689 if (radix_tree_exception(page)) {
1690 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001691 slot = radix_tree_iter_retry(&iter);
1692 continue;
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001693 }
1694
1695 /*
1696 * A shadow entry of a recently evicted page, a swap
1697 * entry from shmem/tmpfs or a DAX entry. Return it
1698 * without attempting to raise page count.
1699 */
1700 goto export;
1701 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001702
1703 head = compound_head(page);
1704 if (!page_cache_get_speculative(head))
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001705 goto repeat;
1706
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001707 /* The page was split under us? */
1708 if (compound_head(page) != head) {
1709 put_page(head);
1710 goto repeat;
1711 }
1712
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001713 /* Has the page moved? */
1714 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001715 put_page(head);
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001716 goto repeat;
1717 }
1718export:
1719 indices[ret] = iter.index;
1720 entries[ret] = page;
1721 if (++ret == nr_entries)
1722 break;
1723 }
1724 rcu_read_unlock();
1725 return ret;
1726}
1727EXPORT_SYMBOL(find_get_entries_tag);
1728
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001729/*
1730 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1731 * a _large_ part of the i/o request. Imagine the worst scenario:
1732 *
1733 * ---R__________________________________________B__________
1734 * ^ reading here ^ bad block(assume 4k)
1735 *
1736 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1737 * => failing the whole request => read(R) => read(R+1) =>
1738 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1739 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1740 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1741 *
1742 * It is going insane. Fix it by quickly scaling down the readahead size.
1743 */
1744static void shrink_readahead_size_eio(struct file *filp,
1745 struct file_ra_state *ra)
1746{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001747 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001748}
1749
Randy Dunlap485bb992006-06-23 02:03:49 -07001750/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001751 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001752 * @filp: the file to read
1753 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001754 * @iter: data destination
1755 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001756 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001758 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 *
1760 * This is really ugly. But the goto's actually try to clarify some
1761 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 */
Al Viro6e58e792014-02-03 17:07:03 -05001763static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1764 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001766 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001768 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001769 pgoff_t index;
1770 pgoff_t last_index;
1771 pgoff_t prev_index;
1772 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001773 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001774 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775
Wei Fangc2a97372016-10-07 17:01:52 -07001776 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
Linus Torvaldsd05c5f72016-12-14 12:45:25 -08001777 return 0;
Wei Fangc2a97372016-10-07 17:01:52 -07001778 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1779
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001780 index = *ppos >> PAGE_SHIFT;
1781 prev_index = ra->prev_pos >> PAGE_SHIFT;
1782 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1783 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1784 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 for (;;) {
1787 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001788 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001789 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 unsigned long nr, ret;
1791
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793find_page:
Michal Hocko5abf1862017-02-03 13:13:29 -08001794 if (fatal_signal_pending(current)) {
1795 error = -EINTR;
1796 goto out;
1797 }
1798
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001800 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001801 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001802 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001803 index, last_index - index);
1804 page = find_get_page(mapping, index);
1805 if (unlikely(page == NULL))
1806 goto no_cached_page;
1807 }
1808 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001809 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001810 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001811 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001813 if (!PageUptodate(page)) {
Mel Gormanebded022016-03-15 14:55:39 -07001814 /*
1815 * See comment in do_read_cache_page on why
1816 * wait_on_page_locked is used to avoid unnecessarily
1817 * serialisations and why it's safe.
1818 */
Bart Van Asschec4b209a2016-10-07 16:58:33 -07001819 error = wait_on_page_locked_killable(page);
1820 if (unlikely(error))
1821 goto readpage_error;
Mel Gormanebded022016-03-15 14:55:39 -07001822 if (PageUptodate(page))
1823 goto page_ok;
1824
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001825 if (inode->i_blkbits == PAGE_SHIFT ||
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001826 !mapping->a_ops->is_partially_uptodate)
1827 goto page_not_up_to_date;
Eryu Guan6d6d36b2016-11-01 15:43:07 +08001828 /* pipes can't handle partially uptodate pages */
1829 if (unlikely(iter->type & ITER_PIPE))
1830 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001831 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001832 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001833 /* Did it get truncated before we got the lock? */
1834 if (!page->mapping)
1835 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001836 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001837 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001838 goto page_not_up_to_date_locked;
1839 unlock_page(page);
1840 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001842 /*
1843 * i_size must be checked after we know the page is Uptodate.
1844 *
1845 * Checking i_size after the check allows us to calculate
1846 * the correct value for "nr", which means the zero-filled
1847 * part of the page is not copied back to userspace (unless
1848 * another truncate extends the file - this is desired though).
1849 */
1850
1851 isize = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001852 end_index = (isize - 1) >> PAGE_SHIFT;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001853 if (unlikely(!isize || index > end_index)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001854 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07001855 goto out;
1856 }
1857
1858 /* nr is the maximum number of bytes to copy from this page */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001859 nr = PAGE_SIZE;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001860 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001861 nr = ((isize - 1) & ~PAGE_MASK) + 1;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001862 if (nr <= offset) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001863 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07001864 goto out;
1865 }
1866 }
1867 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868
1869 /* If users can be writing to this page using arbitrary
1870 * virtual addresses, take care about potential aliasing
1871 * before reading the page on the kernel side.
1872 */
1873 if (mapping_writably_mapped(mapping))
1874 flush_dcache_page(page);
1875
1876 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001877 * When a sequential read accesses a page several times,
1878 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 */
Jan Karaec0f1632007-05-06 14:49:25 -07001880 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 mark_page_accessed(page);
1882 prev_index = index;
1883
1884 /*
1885 * Ok, we have the page, and it's up-to-date, so
1886 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 */
Al Viro6e58e792014-02-03 17:07:03 -05001888
1889 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001891 index += offset >> PAGE_SHIFT;
1892 offset &= ~PAGE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001893 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001895 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001896 written += ret;
1897 if (!iov_iter_count(iter))
1898 goto out;
1899 if (ret < nr) {
1900 error = -EFAULT;
1901 goto out;
1902 }
1903 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904
1905page_not_up_to_date:
1906 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001907 error = lock_page_killable(page);
1908 if (unlikely(error))
1909 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001911page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001912 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 if (!page->mapping) {
1914 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001915 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 continue;
1917 }
1918
1919 /* Did somebody else fill it already? */
1920 if (PageUptodate(page)) {
1921 unlock_page(page);
1922 goto page_ok;
1923 }
1924
1925readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001926 /*
1927 * A previous I/O error may have been due to temporary
1928 * failures, eg. multipath errors.
1929 * PG_error will be set again if readpage fails.
1930 */
1931 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 /* Start the actual read. The read will unlock the page. */
1933 error = mapping->a_ops->readpage(filp, page);
1934
Zach Brown994fc28c2005-12-15 14:28:17 -08001935 if (unlikely(error)) {
1936 if (error == AOP_TRUNCATED_PAGE) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001937 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001938 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001939 goto find_page;
1940 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001942 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
1944 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001945 error = lock_page_killable(page);
1946 if (unlikely(error))
1947 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 if (!PageUptodate(page)) {
1949 if (page->mapping == NULL) {
1950 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001951 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 */
1953 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001954 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 goto find_page;
1956 }
1957 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001958 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001959 error = -EIO;
1960 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 }
1962 unlock_page(page);
1963 }
1964
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 goto page_ok;
1966
1967readpage_error:
1968 /* UHHUH! A synchronous read error occurred. Report it */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001969 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 goto out;
1971
1972no_cached_page:
1973 /*
1974 * Ok, it wasn't cached, so we need to create a new
1975 * page..
1976 */
Nick Piggineb2be182007-10-16 01:24:57 -07001977 page = page_cache_alloc_cold(mapping);
1978 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05001979 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07001980 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 }
Michal Hocko6afdb852015-06-24 16:58:06 -07001982 error = add_to_page_cache_lru(page, mapping, index,
Michal Hockoc62d2552015-11-06 16:28:49 -08001983 mapping_gfp_constraint(mapping, GFP_KERNEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 if (error) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001985 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001986 if (error == -EEXIST) {
1987 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05001989 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 goto out;
1991 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 goto readpage;
1993 }
1994
1995out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001996 ra->prev_pos = prev_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001997 ra->prev_pos <<= PAGE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001998 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002000 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07002001 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05002002 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003}
2004
Randy Dunlap485bb992006-06-23 02:03:49 -07002005/**
Al Viro6abd2322014-04-04 14:20:57 -04002006 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07002007 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04002008 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07002009 *
Al Viro6abd2322014-04-04 14:20:57 -04002010 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 * that can use the page cache directly.
2012 */
2013ssize_t
Al Viroed978a82014-03-05 22:53:04 -05002014generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015{
Al Viroed978a82014-03-05 22:53:04 -05002016 struct file *file = iocb->ki_filp;
2017 ssize_t retval = 0;
Nicolai Stangee7080a42016-03-25 14:22:14 -07002018 size_t count = iov_iter_count(iter);
2019
2020 if (!count)
2021 goto out; /* skip atime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
Al Viro2ba48ce2015-04-09 13:52:01 -04002023 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viroed978a82014-03-05 22:53:04 -05002024 struct address_space *mapping = file->f_mapping;
2025 struct inode *inode = mapping->host;
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002026 struct iov_iter data = *iter;
Badari Pulavarty543ade12006-09-30 23:28:48 -07002027 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 size = i_size_read(inode);
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07002030 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
2031 iocb->ki_pos + count - 1);
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002032 if (retval < 0)
2033 goto out;
Al Viroed978a82014-03-05 22:53:04 -05002034
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002035 file_accessed(file);
2036
2037 retval = mapping->a_ops->direct_IO(iocb, &data);
Al Viroc3a69022016-10-10 13:26:27 -04002038 if (retval >= 0) {
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07002039 iocb->ki_pos += retval;
Al Viroed978a82014-03-05 22:53:04 -05002040 iov_iter_advance(iter, retval);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002041 }
Josef Bacik66f998f2010-05-23 11:00:54 -04002042
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002043 /*
2044 * Btrfs can have a short DIO read if we encounter
2045 * compressed extents, so if there was an error, or if
2046 * we've already read everything we wanted to, or if
2047 * there was a short read because we hit EOF, go ahead
2048 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002049 * the rest of the read. Buffered reads will not work for
2050 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002051 */
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07002052 if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size ||
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11002053 IS_DAX(inode))
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00002054 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 }
2056
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07002057 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058out:
2059 return retval;
2060}
Al Viroed978a82014-03-05 22:53:04 -05002061EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07002064/**
2065 * page_cache_read - adds requested page to the page cache if not already there
2066 * @file: file to read
2067 * @offset: page index
Randy Dunlap62eb3202016-02-11 16:12:58 -08002068 * @gfp_mask: memory allocation flags
Randy Dunlap485bb992006-06-23 02:03:49 -07002069 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 * This adds the requested page to the page cache if it isn't already there,
2071 * and schedules an I/O to read in its contents from disk.
2072 */
Michal Hockoc20cd452016-01-14 15:20:12 -08002073static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074{
2075 struct address_space *mapping = file->f_mapping;
Paul McQuade99dadfd2014-10-09 15:29:03 -07002076 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08002077 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078
Zach Brown994fc28c2005-12-15 14:28:17 -08002079 do {
Michal Hockoc20cd452016-01-14 15:20:12 -08002080 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
Zach Brown994fc28c2005-12-15 14:28:17 -08002081 if (!page)
2082 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083
Michal Hockoc20cd452016-01-14 15:20:12 -08002084 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
Zach Brown994fc28c2005-12-15 14:28:17 -08002085 if (ret == 0)
2086 ret = mapping->a_ops->readpage(file, page);
2087 else if (ret == -EEXIST)
2088 ret = 0; /* losing race to add is OK */
2089
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002090 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091
Zach Brown994fc28c2005-12-15 14:28:17 -08002092 } while (ret == AOP_TRUNCATED_PAGE);
Paul McQuade99dadfd2014-10-09 15:29:03 -07002093
Zach Brown994fc28c2005-12-15 14:28:17 -08002094 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095}
2096
2097#define MMAP_LOTSAMISS (100)
2098
Linus Torvaldsef00e082009-06-16 15:31:25 -07002099/*
2100 * Synchronous readahead happens when we don't even find
2101 * a page in the page cache at all.
2102 */
2103static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2104 struct file_ra_state *ra,
2105 struct file *file,
2106 pgoff_t offset)
2107{
Linus Torvaldsef00e082009-06-16 15:31:25 -07002108 struct address_space *mapping = file->f_mapping;
2109
2110 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07002111 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002112 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002113 if (!ra->ra_pages)
2114 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002115
Joe Perches64363aa2013-07-08 16:00:18 -07002116 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07002117 page_cache_sync_readahead(mapping, ra, file, offset,
2118 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002119 return;
2120 }
2121
Andi Kleen207d04b2011-05-24 17:12:29 -07002122 /* Avoid banging the cache line if not needed */
2123 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002124 ra->mmap_miss++;
2125
2126 /*
2127 * Do we miss much more than hit in this file? If so,
2128 * stop bothering with read-ahead. It will only hurt.
2129 */
2130 if (ra->mmap_miss > MMAP_LOTSAMISS)
2131 return;
2132
Wu Fengguangd30a1102009-06-16 15:31:30 -07002133 /*
2134 * mmap read-around
2135 */
Roman Gushchin600e19a2015-11-05 18:47:08 -08002136 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2137 ra->size = ra->ra_pages;
2138 ra->async_size = ra->ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002139 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002140}
2141
2142/*
2143 * Asynchronous readahead happens when we find the page and PG_readahead,
2144 * so we want to possibly extend the readahead further..
2145 */
2146static void do_async_mmap_readahead(struct vm_area_struct *vma,
2147 struct file_ra_state *ra,
2148 struct file *file,
2149 struct page *page,
2150 pgoff_t offset)
2151{
2152 struct address_space *mapping = file->f_mapping;
2153
2154 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07002155 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002156 return;
2157 if (ra->mmap_miss > 0)
2158 ra->mmap_miss--;
2159 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07002160 page_cache_async_readahead(mapping, ra, file,
2161 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002162}
2163
Randy Dunlap485bb992006-06-23 02:03:49 -07002164/**
Nick Piggin54cb8822007-07-19 01:46:59 -07002165 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07002166 * @vma: vma in which the fault was taken
2167 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07002168 *
Nick Piggin54cb8822007-07-19 01:46:59 -07002169 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 * mapped memory region to read in file data during a page fault.
2171 *
2172 * The goto's are kind of ugly, but this streamlines the normal case of having
2173 * it in the page cache, and handles the special cases reasonably without
2174 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002175 *
2176 * vma->vm_mm->mmap_sem must be held on entry.
2177 *
2178 * If our return value has VM_FAULT_RETRY set, it's because
2179 * lock_page_or_retry() returned 0.
2180 * The mmap_sem has usually been released in this case.
2181 * See __lock_page_or_retry() for the exception.
2182 *
2183 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2184 * has not been released.
2185 *
2186 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002188int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189{
2190 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07002191 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 struct address_space *mapping = file->f_mapping;
2193 struct file_ra_state *ra = &file->f_ra;
2194 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002195 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 struct page *page;
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002197 loff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07002198 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002200 size = round_up(i_size_read(inode), PAGE_SIZE);
2201 if (offset >= size >> PAGE_SHIFT)
Linus Torvalds5307cc12007-10-31 09:19:46 -07002202 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002205 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002207 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002208 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07002210 * We found the page, so try async readahead before
2211 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002213 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002214 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07002215 /* No page in the page cache at all */
2216 do_sync_mmap_readahead(vma, ra, file, offset);
2217 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07002218 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002219 ret = VM_FAULT_MAJOR;
2220retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07002221 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 if (!page)
2223 goto no_cached_page;
2224 }
2225
Michel Lespinassed88c0922010-11-02 13:05:18 -07002226 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002227 put_page(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002228 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07002229 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07002230
2231 /* Did it get truncated? */
2232 if (unlikely(page->mapping != mapping)) {
2233 unlock_page(page);
2234 put_page(page);
2235 goto retry_find;
2236 }
Sasha Levin309381fea2014-01-23 15:52:54 -08002237 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07002238
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 /*
Nick Piggind00806b2007-07-19 01:46:57 -07002240 * We have a locked page in the page cache, now we need to check
2241 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 */
Nick Piggind00806b2007-07-19 01:46:57 -07002243 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 goto page_not_uptodate;
2245
Linus Torvaldsef00e082009-06-16 15:31:25 -07002246 /*
2247 * Found the page and have a reference on it.
2248 * We must recheck i_size under page lock.
2249 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002250 size = round_up(i_size_read(inode), PAGE_SIZE);
2251 if (unlikely(offset >= size >> PAGE_SHIFT)) {
Nick Piggind00806b2007-07-19 01:46:57 -07002252 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002253 put_page(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07002254 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07002255 }
2256
Nick Piggind0217ac2007-07-19 01:47:03 -07002257 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07002258 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260no_cached_page:
2261 /*
2262 * We're only likely to ever get here if MADV_RANDOM is in
2263 * effect.
2264 */
Michal Hockoc20cd452016-01-14 15:20:12 -08002265 error = page_cache_read(file, offset, vmf->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266
2267 /*
2268 * The page we want has now been added to the page cache.
2269 * In the unlikely event that someone removed it in the
2270 * meantime, we'll just come back here and read it again.
2271 */
2272 if (error >= 0)
2273 goto retry_find;
2274
2275 /*
2276 * An error return from page_cache_read can result if the
2277 * system is low on memory, or a problem occurs while trying
2278 * to schedule I/O.
2279 */
2280 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07002281 return VM_FAULT_OOM;
2282 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283
2284page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 /*
2286 * Umm, take care of errors if the page isn't up-to-date.
2287 * Try to re-read it _once_. We do this synchronously,
2288 * because there really aren't any performance issues here
2289 * and we need to check for errors.
2290 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08002292 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07002293 if (!error) {
2294 wait_on_page_locked(page);
2295 if (!PageUptodate(page))
2296 error = -EIO;
2297 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002298 put_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002299
2300 if (!error || error == AOP_TRUNCATED_PAGE)
2301 goto retry_find;
2302
2303 /* Things didn't work out. Return zero to tell the mm layer so. */
2304 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002305 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07002306}
2307EXPORT_SYMBOL(filemap_fault);
2308
Jan Kara82b0f8c2016-12-14 15:06:58 -08002309void filemap_map_pages(struct vm_fault *vmf,
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002310 pgoff_t start_pgoff, pgoff_t end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002311{
2312 struct radix_tree_iter iter;
2313 void **slot;
Jan Kara82b0f8c2016-12-14 15:06:58 -08002314 struct file *file = vmf->vma->vm_file;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002315 struct address_space *mapping = file->f_mapping;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002316 pgoff_t last_pgoff = start_pgoff;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002317 loff_t size;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002318 struct page *head, *page;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002319
2320 rcu_read_lock();
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002321 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2322 start_pgoff) {
2323 if (iter.index > end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002324 break;
2325repeat:
2326 page = radix_tree_deref_slot(slot);
2327 if (unlikely(!page))
2328 goto next;
2329 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002330 if (radix_tree_deref_retry(page)) {
2331 slot = radix_tree_iter_retry(&iter);
2332 continue;
2333 }
2334 goto next;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002335 }
2336
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002337 head = compound_head(page);
2338 if (!page_cache_get_speculative(head))
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002339 goto repeat;
2340
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002341 /* The page was split under us? */
2342 if (compound_head(page) != head) {
2343 put_page(head);
2344 goto repeat;
2345 }
2346
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002347 /* Has the page moved? */
2348 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002349 put_page(head);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002350 goto repeat;
2351 }
2352
2353 if (!PageUptodate(page) ||
2354 PageReadahead(page) ||
2355 PageHWPoison(page))
2356 goto skip;
2357 if (!trylock_page(page))
2358 goto skip;
2359
2360 if (page->mapping != mapping || !PageUptodate(page))
2361 goto unlock;
2362
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002363 size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2364 if (page->index >= size >> PAGE_SHIFT)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002365 goto unlock;
2366
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002367 if (file->f_ra.mmap_miss > 0)
2368 file->f_ra.mmap_miss--;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002369
Jan Kara82b0f8c2016-12-14 15:06:58 -08002370 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2371 if (vmf->pte)
2372 vmf->pte += iter.index - last_pgoff;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002373 last_pgoff = iter.index;
Jan Kara82b0f8c2016-12-14 15:06:58 -08002374 if (alloc_set_pte(vmf, NULL, page))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002375 goto unlock;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002376 unlock_page(page);
2377 goto next;
2378unlock:
2379 unlock_page(page);
2380skip:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002381 put_page(page);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002382next:
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002383 /* Huge page is mapped? No need to proceed. */
Jan Kara82b0f8c2016-12-14 15:06:58 -08002384 if (pmd_trans_huge(*vmf->pmd))
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002385 break;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002386 if (iter.index == end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002387 break;
2388 }
2389 rcu_read_unlock();
2390}
2391EXPORT_SYMBOL(filemap_map_pages);
2392
Jan Kara4fcf1c62012-06-12 16:20:29 +02002393int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2394{
2395 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002396 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002397 int ret = VM_FAULT_LOCKED;
2398
Jan Kara14da9202012-06-12 16:20:37 +02002399 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002400 file_update_time(vma->vm_file);
2401 lock_page(page);
2402 if (page->mapping != inode->i_mapping) {
2403 unlock_page(page);
2404 ret = VM_FAULT_NOPAGE;
2405 goto out;
2406 }
Jan Kara14da9202012-06-12 16:20:37 +02002407 /*
2408 * We mark the page dirty already here so that when freeze is in
2409 * progress, we are guaranteed that writeback during freezing will
2410 * see the dirty page and writeprotect it again.
2411 */
2412 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002413 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002414out:
Jan Kara14da9202012-06-12 16:20:37 +02002415 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002416 return ret;
2417}
2418EXPORT_SYMBOL(filemap_page_mkwrite);
2419
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002420const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002421 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002422 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002423 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424};
2425
2426/* This is used for a general mmap of a disk file */
2427
2428int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2429{
2430 struct address_space *mapping = file->f_mapping;
2431
2432 if (!mapping->a_ops->readpage)
2433 return -ENOEXEC;
2434 file_accessed(file);
2435 vma->vm_ops = &generic_file_vm_ops;
2436 return 0;
2437}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438
2439/*
2440 * This is for filesystems which do not implement ->writepage.
2441 */
2442int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2443{
2444 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2445 return -EINVAL;
2446 return generic_file_mmap(file, vma);
2447}
2448#else
2449int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2450{
2451 return -ENOSYS;
2452}
2453int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2454{
2455 return -ENOSYS;
2456}
2457#endif /* CONFIG_MMU */
2458
2459EXPORT_SYMBOL(generic_file_mmap);
2460EXPORT_SYMBOL(generic_file_readonly_mmap);
2461
Sasha Levin67f9fd92014-04-03 14:48:18 -07002462static struct page *wait_on_page_read(struct page *page)
2463{
2464 if (!IS_ERR(page)) {
2465 wait_on_page_locked(page);
2466 if (!PageUptodate(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002467 put_page(page);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002468 page = ERR_PTR(-EIO);
2469 }
2470 }
2471 return page;
2472}
2473
Mel Gorman32b63522016-03-15 14:55:36 -07002474static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002475 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002476 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002477 void *data,
2478 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479{
Nick Piggineb2be182007-10-16 01:24:57 -07002480 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 int err;
2482repeat:
2483 page = find_get_page(mapping, index);
2484 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002485 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002486 if (!page)
2487 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002488 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002489 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002490 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002491 if (err == -EEXIST)
2492 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 return ERR_PTR(err);
2495 }
Mel Gorman32b63522016-03-15 14:55:36 -07002496
2497filler:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 err = filler(data, page);
2499 if (err < 0) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002500 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002501 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 }
Mel Gorman32b63522016-03-15 14:55:36 -07002503
2504 page = wait_on_page_read(page);
2505 if (IS_ERR(page))
2506 return page;
2507 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 if (PageUptodate(page))
2510 goto out;
2511
Mel Gormanebded022016-03-15 14:55:39 -07002512 /*
2513 * Page is not up to date and may be locked due one of the following
2514 * case a: Page is being filled and the page lock is held
2515 * case b: Read/write error clearing the page uptodate status
2516 * case c: Truncation in progress (page locked)
2517 * case d: Reclaim in progress
2518 *
2519 * Case a, the page will be up to date when the page is unlocked.
2520 * There is no need to serialise on the page lock here as the page
2521 * is pinned so the lock gives no additional protection. Even if the
2522 * the page is truncated, the data is still valid if PageUptodate as
2523 * it's a race vs truncate race.
2524 * Case b, the page will not be up to date
2525 * Case c, the page may be truncated but in itself, the data may still
2526 * be valid after IO completes as it's a read vs truncate race. The
2527 * operation must restart if the page is not uptodate on unlock but
2528 * otherwise serialising on page lock to stabilise the mapping gives
2529 * no additional guarantees to the caller as the page lock is
2530 * released before return.
2531 * Case d, similar to truncation. If reclaim holds the page lock, it
2532 * will be a race with remove_mapping that determines if the mapping
2533 * is valid on unlock but otherwise the data is valid and there is
2534 * no need to serialise with page lock.
2535 *
2536 * As the page lock gives no additional guarantee, we optimistically
2537 * wait on the page to be unlocked and check if it's up to date and
2538 * use the page if it is. Otherwise, the page lock is required to
2539 * distinguish between the different cases. The motivation is that we
2540 * avoid spurious serialisations and wakeups when multiple processes
2541 * wait on the same page for IO to complete.
2542 */
2543 wait_on_page_locked(page);
2544 if (PageUptodate(page))
2545 goto out;
2546
2547 /* Distinguish between all the cases under the safety of the lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 lock_page(page);
Mel Gormanebded022016-03-15 14:55:39 -07002549
2550 /* Case c or d, restart the operation */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 if (!page->mapping) {
2552 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002553 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002554 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 }
Mel Gormanebded022016-03-15 14:55:39 -07002556
2557 /* Someone else locked and filled the page in a very small window */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 if (PageUptodate(page)) {
2559 unlock_page(page);
2560 goto out;
2561 }
Mel Gorman32b63522016-03-15 14:55:36 -07002562 goto filler;
2563
David Howellsc855ff32007-05-09 13:42:20 +01002564out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002565 mark_page_accessed(page);
2566 return page;
2567}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002568
2569/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002570 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002571 * @mapping: the page's address_space
2572 * @index: the page index
2573 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002574 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002575 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002576 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002577 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002578 *
2579 * If the page does not get brought uptodate, return -EIO.
2580 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002581struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002582 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002583 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002584 void *data)
2585{
2586 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2587}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002588EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002589
2590/**
2591 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2592 * @mapping: the page's address_space
2593 * @index: the page index
2594 * @gfp: the page allocator flags to use if allocating
2595 *
2596 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002597 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002598 *
2599 * If the page does not get brought uptodate, return -EIO.
2600 */
2601struct page *read_cache_page_gfp(struct address_space *mapping,
2602 pgoff_t index,
2603 gfp_t gfp)
2604{
2605 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2606
Sasha Levin67f9fd92014-04-03 14:48:18 -07002607 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002608}
2609EXPORT_SYMBOL(read_cache_page_gfp);
2610
Nick Piggin2f718ff2007-10-16 01:24:59 -07002611/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 * Performs necessary checks before doing a write
2613 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002614 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 * Returns appropriate error code that caller should return or
2616 * zero in case that write should be allowed.
2617 */
Al Viro3309dd02015-04-09 12:55:47 -04002618inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619{
Al Viro3309dd02015-04-09 12:55:47 -04002620 struct file *file = iocb->ki_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002622 unsigned long limit = rlimit(RLIMIT_FSIZE);
Al Viro3309dd02015-04-09 12:55:47 -04002623 loff_t pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624
Al Viro3309dd02015-04-09 12:55:47 -04002625 if (!iov_iter_count(from))
2626 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627
Al Viro0fa6b002015-04-04 04:05:48 -04002628 /* FIXME: this is for backwards compatibility with 2.4 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002629 if (iocb->ki_flags & IOCB_APPEND)
Al Viro3309dd02015-04-09 12:55:47 -04002630 iocb->ki_pos = i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631
Al Viro3309dd02015-04-09 12:55:47 -04002632 pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633
Al Viro0fa6b002015-04-04 04:05:48 -04002634 if (limit != RLIM_INFINITY) {
Al Viro3309dd02015-04-09 12:55:47 -04002635 if (iocb->ki_pos >= limit) {
Al Viro0fa6b002015-04-04 04:05:48 -04002636 send_sig(SIGXFSZ, current, 0);
2637 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 }
Al Viro3309dd02015-04-09 12:55:47 -04002639 iov_iter_truncate(from, limit - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 }
2641
2642 /*
2643 * LFS rule
2644 */
Al Viro3309dd02015-04-09 12:55:47 -04002645 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646 !(file->f_flags & O_LARGEFILE))) {
Al Viro3309dd02015-04-09 12:55:47 -04002647 if (pos >= MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 return -EFBIG;
Al Viro3309dd02015-04-09 12:55:47 -04002649 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650 }
2651
2652 /*
2653 * Are we about to exceed the fs block limit ?
2654 *
2655 * If we have written data it becomes a short write. If we have
2656 * exceeded without writing data we send a signal and return EFBIG.
2657 * Linus frestrict idea will clean these up nicely..
2658 */
Al Viro3309dd02015-04-09 12:55:47 -04002659 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2660 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661
Al Viro3309dd02015-04-09 12:55:47 -04002662 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2663 return iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664}
2665EXPORT_SYMBOL(generic_write_checks);
2666
Nick Pigginafddba42007-10-16 01:25:01 -07002667int pagecache_write_begin(struct file *file, struct address_space *mapping,
2668 loff_t pos, unsigned len, unsigned flags,
2669 struct page **pagep, void **fsdata)
2670{
2671 const struct address_space_operations *aops = mapping->a_ops;
2672
Nick Piggin4e02ed42008-10-29 14:00:55 -07002673 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002674 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002675}
2676EXPORT_SYMBOL(pagecache_write_begin);
2677
2678int pagecache_write_end(struct file *file, struct address_space *mapping,
2679 loff_t pos, unsigned len, unsigned copied,
2680 struct page *page, void *fsdata)
2681{
2682 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002683
Nick Piggin4e02ed42008-10-29 14:00:55 -07002684 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002685}
2686EXPORT_SYMBOL(pagecache_write_end);
2687
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688ssize_t
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002689generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690{
2691 struct file *file = iocb->ki_filp;
2692 struct address_space *mapping = file->f_mapping;
2693 struct inode *inode = mapping->host;
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002694 loff_t pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002696 size_t write_len;
2697 pgoff_t end;
Al Viro26978b82014-03-10 14:08:45 -04002698 struct iov_iter data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699
Al Viro0c949332014-03-22 06:51:37 -04002700 write_len = iov_iter_count(from);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002701 end = (pos + write_len - 1) >> PAGE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002702
Nick Piggin48b47c52009-01-06 14:40:22 -08002703 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002704 if (written)
2705 goto out;
2706
2707 /*
2708 * After a write we want buffered reads to be sure to go to disk to get
2709 * the new data. We invalidate clean cached page from the region we're
2710 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002711 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002712 */
2713 if (mapping->nrpages) {
2714 written = invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002715 pos >> PAGE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002716 /*
2717 * If a page can not be invalidated, return 0 to fall back
2718 * to buffered write.
2719 */
2720 if (written) {
2721 if (written == -EBUSY)
2722 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002723 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002724 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002725 }
2726
Al Viro26978b82014-03-10 14:08:45 -04002727 data = *from;
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07002728 written = mapping->a_ops->direct_IO(iocb, &data);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002729
2730 /*
2731 * Finally, try again to invalidate clean pages which might have been
2732 * cached by non-direct readahead, or faulted in by get_user_pages()
2733 * if the source of the write was an mmap'ed region of the file
2734 * we're writing. Either one is a pretty crazy thing to do,
2735 * so we don't support it 100%. If this invalidation
2736 * fails, tough, the write still worked...
2737 */
2738 if (mapping->nrpages) {
2739 invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002740 pos >> PAGE_SHIFT, end);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002741 }
2742
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002744 pos += written;
Al Virof8579f82014-03-03 22:03:20 -05002745 iov_iter_advance(from, written);
Namhyung Kim01166512010-10-26 14:21:58 -07002746 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2747 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 mark_inode_dirty(inode);
2749 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002750 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002752out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 return written;
2754}
2755EXPORT_SYMBOL(generic_file_direct_write);
2756
Nick Piggineb2be182007-10-16 01:24:57 -07002757/*
2758 * Find or create a page at the given pagecache position. Return the locked
2759 * page. This function is specifically for buffered writes.
2760 */
Nick Piggin54566b22009-01-04 12:00:53 -08002761struct page *grab_cache_page_write_begin(struct address_space *mapping,
2762 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002763{
Nick Piggineb2be182007-10-16 01:24:57 -07002764 struct page *page;
Johannes Weinerbbddabe2016-05-20 16:56:28 -07002765 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002766
Nick Piggin54566b22009-01-04 12:00:53 -08002767 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07002768 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07002769
Mel Gorman2457aec2014-06-04 16:10:31 -07002770 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01002771 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07002772 if (page)
2773 wait_for_stable_page(page);
2774
Nick Piggineb2be182007-10-16 01:24:57 -07002775 return page;
2776}
Nick Piggin54566b22009-01-04 12:00:53 -08002777EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002778
Al Viro3b93f912014-02-11 21:34:08 -05002779ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002780 struct iov_iter *i, loff_t pos)
2781{
2782 struct address_space *mapping = file->f_mapping;
2783 const struct address_space_operations *a_ops = mapping->a_ops;
2784 long status = 0;
2785 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002786 unsigned int flags = 0;
2787
2788 /*
2789 * Copies from kernel address space cannot fail (NFSD is a big user).
2790 */
Al Viro777eda22014-12-17 04:46:46 -05002791 if (!iter_is_iovec(i))
Nick Piggin674b8922007-10-16 01:25:03 -07002792 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002793
2794 do {
2795 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002796 unsigned long offset; /* Offset into pagecache page */
2797 unsigned long bytes; /* Bytes to write to page */
2798 size_t copied; /* Bytes copied from user */
2799 void *fsdata;
2800
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002801 offset = (pos & (PAGE_SIZE - 1));
2802 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002803 iov_iter_count(i));
2804
2805again:
Linus Torvalds00a3d662015-10-07 08:32:38 +01002806 /*
2807 * Bring in the user page that we will copy from _first_.
2808 * Otherwise there's a nasty deadlock on copying from the
2809 * same page as we're writing to, without it being marked
2810 * up-to-date.
2811 *
2812 * Not only is this an optimisation, but it is also required
2813 * to check that the address is actually valid, when atomic
2814 * usercopies are used, below.
2815 */
2816 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2817 status = -EFAULT;
2818 break;
2819 }
2820
Jan Kara296291c2015-10-22 13:32:21 -07002821 if (fatal_signal_pending(current)) {
2822 status = -EINTR;
2823 break;
2824 }
2825
Nick Piggin674b8922007-10-16 01:25:03 -07002826 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002827 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07002828 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07002829 break;
2830
anfei zhou931e80e2010-02-02 13:44:02 -08002831 if (mapping_writably_mapped(mapping))
2832 flush_dcache_page(page);
Linus Torvalds00a3d662015-10-07 08:32:38 +01002833
Nick Pigginafddba42007-10-16 01:25:01 -07002834 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07002835 flush_dcache_page(page);
2836
2837 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2838 page, fsdata);
2839 if (unlikely(status < 0))
2840 break;
2841 copied = status;
2842
2843 cond_resched();
2844
Nick Piggin124d3b72008-02-02 15:01:17 +01002845 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002846 if (unlikely(copied == 0)) {
2847 /*
2848 * If we were unable to copy any data at all, we must
2849 * fall back to a single segment length write.
2850 *
2851 * If we didn't fallback here, we could livelock
2852 * because not all segments in the iov can be copied at
2853 * once without a pagefault.
2854 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002855 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002856 iov_iter_single_seg_count(i));
2857 goto again;
2858 }
Nick Pigginafddba42007-10-16 01:25:01 -07002859 pos += copied;
2860 written += copied;
2861
2862 balance_dirty_pages_ratelimited(mapping);
Nick Pigginafddba42007-10-16 01:25:01 -07002863 } while (iov_iter_count(i));
2864
2865 return written ? written : status;
2866}
Al Viro3b93f912014-02-11 21:34:08 -05002867EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868
Jan Karae4dd9de2009-08-17 18:10:06 +02002869/**
Al Viro81742022014-04-03 03:17:43 -04002870 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002871 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04002872 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002873 *
2874 * This function does all the work needed for actually writing data to a
2875 * file. It does all basic checks, removes SUID from the file, updates
2876 * modification times and calls proper subroutines depending on whether we
2877 * do direct IO or a standard buffered write.
2878 *
2879 * It expects i_mutex to be grabbed unless we work on a block device or similar
2880 * object which does not need locking at all.
2881 *
2882 * This function does *not* take care of syncing data in case of O_SYNC write.
2883 * A caller has to handle it. This is mainly due to the fact that we want to
2884 * avoid syncing under i_mutex.
2885 */
Al Viro81742022014-04-03 03:17:43 -04002886ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887{
2888 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002889 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05002891 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002893 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01002896 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002897 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 if (err)
2899 goto out;
2900
Josef Bacikc3b2da32012-03-26 09:59:21 -04002901 err = file_update_time(file);
2902 if (err)
2903 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904
Al Viro2ba48ce2015-04-09 13:52:01 -04002905 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04002906 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002907
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002908 written = generic_file_direct_write(iocb, from);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002909 /*
2910 * If the write stopped short of completing, fall back to
2911 * buffered writes. Some filesystems do this for writes to
2912 * holes, for example. For DAX files, a buffered write will
2913 * not succeed (even if it did, DAX does not handle dirty
2914 * page-cache pages correctly).
2915 */
Al Viro0b8def92015-04-07 10:22:53 -04002916 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002918
Al Viro0b8def92015-04-07 10:22:53 -04002919 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002920 /*
Al Viro3b93f912014-02-11 21:34:08 -05002921 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002922 * then we want to return the number of bytes which were
2923 * direct-written, or the error code if that was zero. Note
2924 * that this differs from normal direct-io semantics, which
2925 * will return -EFOO even if some bytes were written.
2926 */
Al Viro60bb4522014-08-08 12:39:16 -04002927 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05002928 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002929 goto out;
2930 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002931 /*
2932 * We need to ensure that the page cache pages are written to
2933 * disk and invalidated to preserve the expected O_DIRECT
2934 * semantics.
2935 */
Al Viro3b93f912014-02-11 21:34:08 -05002936 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04002937 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002938 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04002939 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05002940 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002941 invalidate_mapping_pages(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002942 pos >> PAGE_SHIFT,
2943 endbyte >> PAGE_SHIFT);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002944 } else {
2945 /*
2946 * We don't know how much we wrote, so just return
2947 * the number of bytes which were direct-written
2948 */
2949 }
2950 } else {
Al Viro0b8def92015-04-07 10:22:53 -04002951 written = generic_perform_write(file, from, iocb->ki_pos);
2952 if (likely(written > 0))
2953 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002954 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955out:
2956 current->backing_dev_info = NULL;
2957 return written ? written : err;
2958}
Al Viro81742022014-04-03 03:17:43 -04002959EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960
Jan Karae4dd9de2009-08-17 18:10:06 +02002961/**
Al Viro81742022014-04-03 03:17:43 -04002962 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002963 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04002964 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002965 *
Al Viro81742022014-04-03 03:17:43 -04002966 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02002967 * filesystems. It takes care of syncing the file in case of O_SYNC file
2968 * and acquires i_mutex as needed.
2969 */
Al Viro81742022014-04-03 03:17:43 -04002970ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971{
2972 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002973 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975
Al Viro59551022016-01-22 15:40:57 -05002976 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -04002977 ret = generic_write_checks(iocb, from);
2978 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04002979 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -05002980 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981
Christoph Hellwige2592212016-04-07 08:52:01 -07002982 if (ret > 0)
2983 ret = generic_write_sync(iocb, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984 return ret;
2985}
Al Viro81742022014-04-03 03:17:43 -04002986EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987
David Howellscf9a2ae2006-08-29 19:05:54 +01002988/**
2989 * try_to_release_page() - release old fs-specific metadata on a page
2990 *
2991 * @page: the page which the kernel is trying to free
2992 * @gfp_mask: memory allocation flags (and I/O mode)
2993 *
2994 * The address_space is to try to release any data against the page
2995 * (presumably at page->private). If the release was successful, return `1'.
2996 * Otherwise return zero.
2997 *
David Howells266cf652009-04-03 16:42:36 +01002998 * This may also be called if PG_fscache is set on a page, indicating that the
2999 * page is known to the local caching routines.
3000 *
David Howellscf9a2ae2006-08-29 19:05:54 +01003001 * The @gfp_mask argument specifies whether I/O may be performed to release
Mel Gorman71baba42015-11-06 16:28:28 -08003002 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01003003 *
David Howellscf9a2ae2006-08-29 19:05:54 +01003004 */
3005int try_to_release_page(struct page *page, gfp_t gfp_mask)
3006{
3007 struct address_space * const mapping = page->mapping;
3008
3009 BUG_ON(!PageLocked(page));
3010 if (PageWriteback(page))
3011 return 0;
3012
3013 if (mapping && mapping->a_ops->releasepage)
3014 return mapping->a_ops->releasepage(page, gfp_mask);
3015 return try_to_free_buffers(page);
3016}
3017
3018EXPORT_SYMBOL(try_to_release_page);