blob: b046d8f147e2083243f12cb67c5013550b5157be [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;
135 if (node)
136 workingset_node_shadows_dec(node);
137 } else {
138 /* DAX can replace empty locked entry with a hole */
139 WARN_ON_ONCE(p !=
140 (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY |
141 RADIX_DAX_ENTRY_LOCK));
142 /* DAX accounts exceptional entries as normal pages */
143 if (node)
144 workingset_node_pages_dec(node);
145 /* Wakeup waiters for exceptional entry lock */
146 dax_wake_mapping_entry_waiter(mapping, page->index,
Ross Zwisler87fa6f32017-01-10 16:57:15 -0800147 true);
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700148 }
149 }
150 radix_tree_replace_slot(slot, page);
151 mapping->nrpages++;
152 if (node) {
153 workingset_node_pages_inc(node);
154 /*
155 * Don't track node that contains actual pages.
156 *
157 * Avoid acquiring the list_lru lock if already
158 * untracked. The list_empty() test is safe as
159 * node->private_list is protected by
160 * mapping->tree_lock.
161 */
162 if (!list_empty(&node->private_list))
163 list_lru_del(&workingset_shadow_nodes,
164 &node->private_list);
165 }
166 return 0;
167}
168
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700169static void page_cache_tree_delete(struct address_space *mapping,
170 struct page *page, void *shadow)
171{
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700172 int i, nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700173
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700174 VM_BUG_ON_PAGE(!PageLocked(page), page);
175 VM_BUG_ON_PAGE(PageTail(page), page);
176 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
Johannes Weiner449dd692014-04-03 14:47:56 -0700177
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700178 for (i = 0; i < nr; i++) {
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200179 struct radix_tree_node *node;
180 void **slot;
181
182 __radix_tree_lookup(&mapping->page_tree, page->index + i,
183 &node, &slot);
184
185 radix_tree_clear_tags(&mapping->page_tree, node, slot);
186
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700187 if (!node) {
188 VM_BUG_ON_PAGE(nr != 1, page);
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200189 /*
190 * We need a node to properly account shadow
191 * entries. Don't plant any without. XXX
192 */
193 shadow = NULL;
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700194 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700195
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200196 radix_tree_replace_slot(slot, shadow);
197
198 if (!node)
199 break;
200
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700201 workingset_node_pages_dec(node);
202 if (shadow)
203 workingset_node_shadows_inc(node);
204 else
205 if (__radix_tree_delete_node(&mapping->page_tree, node))
206 continue;
207
208 /*
209 * Track node that only contains shadow entries. DAX mappings
210 * contain no shadow entries and may contain other exceptional
211 * entries so skip those.
212 *
213 * Avoid acquiring the list_lru lock if already tracked.
214 * The list_empty() test is safe as node->private_list is
215 * protected by mapping->tree_lock.
216 */
217 if (!dax_mapping(mapping) && !workingset_node_pages(node) &&
218 list_empty(&node->private_list)) {
219 node->private_data = mapping;
220 list_lru_add(&workingset_shadow_nodes,
221 &node->private_list);
222 }
Johannes Weiner449dd692014-04-03 14:47:56 -0700223 }
Johannes Weinerd3798ae2016-10-04 22:02:08 +0200224
225 if (shadow) {
226 mapping->nrexceptional += nr;
227 /*
228 * Make sure the nrexceptional update is committed before
229 * the nrpages update so that final truncate racing
230 * with reclaim does not see both counters 0 at the
231 * same time and miss a shadow entry.
232 */
233 smp_wmb();
234 }
235 mapping->nrpages -= nr;
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700236}
237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238/*
Minchan Kime64a7822011-03-22 16:32:44 -0700239 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 * sure the page is locked and that nobody else uses it - or that usage
Johannes Weinerfdf1cdb2016-03-15 14:57:25 -0700241 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 */
Johannes Weiner62cccb82016-03-15 14:57:22 -0700243void __delete_from_page_cache(struct page *page, void *shadow)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
245 struct address_space *mapping = page->mapping;
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700246 int nr = hpage_nr_pages(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700248 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600249 /*
250 * if we're uptodate, flush out into the cleancache, otherwise
251 * invalidate any existing cleancache entries. We can't leave
252 * stale data around in the cleancache once our page is gone
253 */
254 if (PageUptodate(page) && PageMappedToDisk(page))
255 cleancache_put_page(page);
256 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400257 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600258
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700259 VM_BUG_ON_PAGE(PageTail(page), page);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800260 VM_BUG_ON_PAGE(page_mapped(page), page);
261 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
262 int mapcount;
263
264 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
265 current->comm, page_to_pfn(page));
266 dump_page(page, "still mapped when deleted");
267 dump_stack();
268 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
269
270 mapcount = page_mapcount(page);
271 if (mapping_exiting(mapping) &&
272 page_count(page) >= mapcount + 2) {
273 /*
274 * All vmas have already been torn down, so it's
275 * a good bet that actually the page is unmapped,
276 * and we'd prefer not to leak it: if we're wrong,
277 * some other bad page check should catch it later.
278 */
279 page_mapcount_reset(page);
Joonsoo Kim6d061f92016-05-19 17:10:46 -0700280 page_ref_sub(page, mapcount);
Hugh Dickins06b241f2016-03-09 14:08:07 -0800281 }
282 }
283
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700284 page_cache_tree_delete(mapping, page, shadow);
285
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700287 /* Leave page->index set: truncation lookup relies upon it */
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700288
Michal Hocko4165b9b2015-06-24 16:57:24 -0700289 /* hugetlb pages do not participate in page cache accounting. */
290 if (!PageHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700291 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700292 if (PageSwapBacked(page)) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700293 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700294 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700295 __dec_node_page_state(page, NR_SHMEM_THPS);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700296 } else {
297 VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
298 }
Linus Torvalds3a692792007-12-19 14:05:13 -0800299
300 /*
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700301 * At this point page must be either written or cleaned by truncate.
302 * Dirty page here signals a bug and loss of unwritten data.
Linus Torvalds3a692792007-12-19 14:05:13 -0800303 *
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700304 * This fixes dirty accounting after removing the page entirely but
305 * leaves PageDirty set: it has no effect for truncated page and
306 * anyway will be cleared before returning page into buddy allocator.
Linus Torvalds3a692792007-12-19 14:05:13 -0800307 */
Konstantin Khlebnikovb9ea2512015-04-14 15:45:27 -0700308 if (WARN_ON_ONCE(PageDirty(page)))
Johannes Weiner62cccb82016-03-15 14:57:22 -0700309 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310}
311
Minchan Kim702cfbf2011-03-22 16:32:43 -0700312/**
313 * delete_from_page_cache - delete page from page cache
314 * @page: the page which the kernel is trying to remove from page cache
315 *
316 * This must be called only on pages that have been verified to be in the page
317 * cache and locked. It will never put the page into the free list, the caller
318 * has a reference on the page.
319 */
320void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700322 struct address_space *mapping = page_mapping(page);
Greg Thelenc4843a72015-05-22 17:13:16 -0400323 unsigned long flags;
Linus Torvalds6072d132010-12-01 13:35:19 -0500324 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
Matt Mackallcd7619d2005-05-01 08:59:01 -0700326 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
Linus Torvalds6072d132010-12-01 13:35:19 -0500328 freepage = mapping->a_ops->freepage;
Greg Thelenc4843a72015-05-22 17:13:16 -0400329
Greg Thelenc4843a72015-05-22 17:13:16 -0400330 spin_lock_irqsave(&mapping->tree_lock, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700331 __delete_from_page_cache(page, NULL);
Greg Thelenc4843a72015-05-22 17:13:16 -0400332 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Linus Torvalds6072d132010-12-01 13:35:19 -0500333
334 if (freepage)
335 freepage(page);
Kirill A. Shutemov83929372016-07-26 15:26:04 -0700336
337 if (PageTransHuge(page) && !PageHuge(page)) {
338 page_ref_sub(page, HPAGE_PMD_NR);
339 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
340 } else {
341 put_page(page);
342 }
Minchan Kim97cecb52011-03-22 16:30:53 -0700343}
344EXPORT_SYMBOL(delete_from_page_cache);
345
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200346int filemap_check_errors(struct address_space *mapping)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700347{
348 int ret = 0;
349 /* Check for outstanding write errors */
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700350 if (test_bit(AS_ENOSPC, &mapping->flags) &&
351 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700352 ret = -ENOSPC;
Jens Axboe7fcbbaf2014-05-22 11:54:16 -0700353 if (test_bit(AS_EIO, &mapping->flags) &&
354 test_and_clear_bit(AS_EIO, &mapping->flags))
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700355 ret = -EIO;
356 return ret;
357}
Miklos Szeredid72d9e22016-07-29 14:10:57 +0200358EXPORT_SYMBOL(filemap_check_errors);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700359
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700361 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700362 * @mapping: address space structure to write
363 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800364 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700365 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700367 * Start writeback against all of a mapping's dirty pages that lie
368 * within the byte offsets <start, end> inclusive.
369 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700371 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 * these two operations is that if a dirty page/buffer is encountered, it must
373 * be waited upon, and not just skipped over.
374 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800375int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
376 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377{
378 int ret;
379 struct writeback_control wbc = {
380 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800381 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700382 .range_start = start,
383 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 };
385
Konstantin Khlebnikov063620b2019-09-23 15:34:45 -0700386 if (!mapping_cap_writeback_dirty(mapping) ||
387 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 return 0;
389
Tejun Heob16b1de2015-06-02 08:39:48 -0600390 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 ret = do_writepages(mapping, &wbc);
Tejun Heob16b1de2015-06-02 08:39:48 -0600392 wbc_detach_inode(&wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 return ret;
394}
395
396static inline int __filemap_fdatawrite(struct address_space *mapping,
397 int sync_mode)
398{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700399 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400}
401
402int filemap_fdatawrite(struct address_space *mapping)
403{
404 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
405}
406EXPORT_SYMBOL(filemap_fdatawrite);
407
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400408int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800409 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410{
411 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
412}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400413EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Randy Dunlap485bb992006-06-23 02:03:49 -0700415/**
416 * filemap_flush - mostly a non-blocking flush
417 * @mapping: target address_space
418 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 * This is a mostly non-blocking flush. Not suitable for data-integrity
420 * purposes - I/O may not be started against all dirty pages.
421 */
422int filemap_flush(struct address_space *mapping)
423{
424 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
425}
426EXPORT_SYMBOL(filemap_flush);
427
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800428static int __filemap_fdatawait_range(struct address_space *mapping,
429 loff_t start_byte, loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300431 pgoff_t index = start_byte >> PAGE_SHIFT;
432 pgoff_t end = end_byte >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 struct pagevec pvec;
434 int nr_pages;
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800435 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200437 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700438 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
440 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 while ((index <= end) &&
442 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
443 PAGECACHE_TAG_WRITEBACK,
444 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
445 unsigned i;
446
447 for (i = 0; i < nr_pages; i++) {
448 struct page *page = pvec.pages[i];
449
450 /* until radix tree lookup accepts end_index */
451 if (page->index > end)
452 continue;
453
454 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800455 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 ret = -EIO;
457 }
458 pagevec_release(&pvec);
459 cond_resched();
460 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700461out:
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800462 return ret;
463}
464
465/**
466 * filemap_fdatawait_range - wait for writeback to complete
467 * @mapping: address space structure to wait for
468 * @start_byte: offset in bytes where the range starts
469 * @end_byte: offset in bytes where the range ends (inclusive)
470 *
471 * Walk the list of under-writeback pages of the given address space
472 * in the given range and wait for all of them. Check error status of
473 * the address space and return it.
474 *
475 * Since the error status of the address space is cleared by this function,
476 * callers are responsible for checking the return value and handling and/or
477 * reporting the error.
478 */
479int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
480 loff_t end_byte)
481{
482 int ret, ret2;
483
484 ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700485 ret2 = filemap_check_errors(mapping);
486 if (!ret)
487 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
489 return ret;
490}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200491EXPORT_SYMBOL(filemap_fdatawait_range);
492
493/**
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800494 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
495 * @mapping: address space structure to wait for
496 *
497 * Walk the list of under-writeback pages of the given address space
498 * and wait for all of them. Unlike filemap_fdatawait(), this function
499 * does not clear error status of the address space.
500 *
501 * Use this function if callers don't handle errors themselves. Expected
502 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
503 * fsfreeze(8)
504 */
505void filemap_fdatawait_keep_errors(struct address_space *mapping)
506{
507 loff_t i_size = i_size_read(mapping->host);
508
509 if (i_size == 0)
510 return;
511
512 __filemap_fdatawait_range(mapping, 0, i_size - 1);
513}
514
515/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700516 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700518 *
519 * Walk the list of under-writeback pages of the given address space
Junichi Nomuraaa750fd2015-11-05 18:47:23 -0800520 * and wait for all of them. Check error status of the address space
521 * and return it.
522 *
523 * Since the error status of the address space is cleared by this function,
524 * callers are responsible for checking the return value and handling and/or
525 * reporting the error.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 */
527int filemap_fdatawait(struct address_space *mapping)
528{
529 loff_t i_size = i_size_read(mapping->host);
530
531 if (i_size == 0)
532 return 0;
533
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200534 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535}
536EXPORT_SYMBOL(filemap_fdatawait);
537
538int filemap_write_and_wait(struct address_space *mapping)
539{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800540 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800542 if ((!dax_mapping(mapping) && mapping->nrpages) ||
543 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800544 err = filemap_fdatawrite(mapping);
545 /*
546 * Even if the above returned error, the pages may be
547 * written partially (e.g. -ENOSPC), so we wait for it.
548 * But the -EIO is special case, it may indicate the worst
549 * thing (e.g. bug) happened, so we avoid waiting for it.
550 */
551 if (err != -EIO) {
552 int err2 = filemap_fdatawait(mapping);
553 if (!err)
554 err = err2;
555 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700556 } else {
557 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800559 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800561EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
Randy Dunlap485bb992006-06-23 02:03:49 -0700563/**
564 * filemap_write_and_wait_range - write out & wait on a file range
565 * @mapping: the address_space for the pages
566 * @lstart: offset in bytes where the range starts
567 * @lend: offset in bytes where the range ends (inclusive)
568 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800569 * Write out and wait upon file offsets lstart->lend, inclusive.
570 *
571 * Note that `lend' is inclusive (describes the last byte to be written) so
572 * that this function can be used to write to the very end-of-file (end = -1).
573 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574int filemap_write_and_wait_range(struct address_space *mapping,
575 loff_t lstart, loff_t lend)
576{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800577 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578
Ross Zwisler7f6d5b52016-02-26 15:19:55 -0800579 if ((!dax_mapping(mapping) && mapping->nrpages) ||
580 (dax_mapping(mapping) && mapping->nrexceptional)) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800581 err = __filemap_fdatawrite_range(mapping, lstart, lend,
582 WB_SYNC_ALL);
583 /* See comment of filemap_write_and_wait() */
584 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200585 int err2 = filemap_fdatawait_range(mapping,
586 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800587 if (!err)
588 err = err2;
589 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700590 } else {
591 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800593 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594}
Chris Masonf6995582009-04-15 13:22:37 -0400595EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
Randy Dunlap485bb992006-06-23 02:03:49 -0700597/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700598 * replace_page_cache_page - replace a pagecache page with a new one
599 * @old: page to be replaced
600 * @new: page to replace with
601 * @gfp_mask: allocation mode
602 *
603 * This function replaces a page in the pagecache with a new one. On
604 * success it acquires the pagecache reference for the new page and
605 * drops it for the old page. Both the old and new pages must be
606 * locked. This function does not add the new page to the LRU, the
607 * caller must do that.
608 *
609 * The remove + add is atomic. The only way this function can fail is
610 * memory allocation failure.
611 */
612int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
613{
614 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700615
Sasha Levin309381fea2014-01-23 15:52:54 -0800616 VM_BUG_ON_PAGE(!PageLocked(old), old);
617 VM_BUG_ON_PAGE(!PageLocked(new), new);
618 VM_BUG_ON_PAGE(new->mapping, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700619
Matthew Wilcoxf4c86fa2018-04-20 14:56:20 -0700620 error = radix_tree_preload(gfp_mask & GFP_RECLAIM_MASK);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700621 if (!error) {
622 struct address_space *mapping = old->mapping;
623 void (*freepage)(struct page *);
Greg Thelenc4843a72015-05-22 17:13:16 -0400624 unsigned long flags;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700625
626 pgoff_t offset = old->index;
627 freepage = mapping->a_ops->freepage;
628
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300629 get_page(new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700630 new->mapping = mapping;
631 new->index = offset;
632
Greg Thelenc4843a72015-05-22 17:13:16 -0400633 spin_lock_irqsave(&mapping->tree_lock, flags);
Johannes Weiner62cccb82016-03-15 14:57:22 -0700634 __delete_from_page_cache(old, NULL);
Johannes Weiner22f2ac52016-09-30 15:11:29 -0700635 error = page_cache_tree_insert(mapping, new, NULL);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700636 BUG_ON(error);
Michal Hocko4165b9b2015-06-24 16:57:24 -0700637
638 /*
639 * hugetlb pages do not participate in page cache accounting.
640 */
641 if (!PageHuge(new))
Mel Gorman11fb9982016-07-28 15:46:20 -0700642 __inc_node_page_state(new, NR_FILE_PAGES);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700643 if (PageSwapBacked(new))
Mel Gorman11fb9982016-07-28 15:46:20 -0700644 __inc_node_page_state(new, NR_SHMEM);
Greg Thelenc4843a72015-05-22 17:13:16 -0400645 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700646 mem_cgroup_migrate(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700647 radix_tree_preload_end();
648 if (freepage)
649 freepage(old);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300650 put_page(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700651 }
652
653 return error;
654}
655EXPORT_SYMBOL_GPL(replace_page_cache_page);
656
Johannes Weinera5289102014-04-03 14:47:51 -0700657static int __add_to_page_cache_locked(struct page *page,
658 struct address_space *mapping,
659 pgoff_t offset, gfp_t gfp_mask,
660 void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661{
Johannes Weiner00501b52014-08-08 14:19:20 -0700662 int huge = PageHuge(page);
663 struct mem_cgroup *memcg;
Nick Piggine2867812008-07-25 19:45:30 -0700664 int error;
665
Sasha Levin309381fea2014-01-23 15:52:54 -0800666 VM_BUG_ON_PAGE(!PageLocked(page), page);
667 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
Nick Piggine2867812008-07-25 19:45:30 -0700668
Johannes Weiner00501b52014-08-08 14:19:20 -0700669 if (!huge) {
670 error = mem_cgroup_try_charge(page, current->mm,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800671 gfp_mask, &memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -0700672 if (error)
673 return error;
674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
Matthew Wilcoxf4c86fa2018-04-20 14:56:20 -0700676 error = radix_tree_maybe_preload(gfp_mask & GFP_RECLAIM_MASK);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700677 if (error) {
Johannes Weiner00501b52014-08-08 14:19:20 -0700678 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800679 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700680 return error;
681 }
682
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300683 get_page(page);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700684 page->mapping = mapping;
685 page->index = offset;
686
687 spin_lock_irq(&mapping->tree_lock);
Johannes Weinera5289102014-04-03 14:47:51 -0700688 error = page_cache_tree_insert(mapping, page, shadowp);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700689 radix_tree_preload_end();
690 if (unlikely(error))
691 goto err_insert;
Michal Hocko4165b9b2015-06-24 16:57:24 -0700692
693 /* hugetlb pages do not participate in page cache accounting. */
694 if (!huge)
Mel Gorman11fb9982016-07-28 15:46:20 -0700695 __inc_node_page_state(page, NR_FILE_PAGES);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700696 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700697 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800698 mem_cgroup_commit_charge(page, memcg, false, false);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700699 trace_mm_filemap_add_to_page_cache(page);
700 return 0;
701err_insert:
702 page->mapping = NULL;
703 /* Leave page->index set: truncation relies upon it */
704 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner00501b52014-08-08 14:19:20 -0700705 if (!huge)
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -0800706 mem_cgroup_cancel_charge(page, memcg, false);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300707 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 return error;
709}
Johannes Weinera5289102014-04-03 14:47:51 -0700710
711/**
712 * add_to_page_cache_locked - add a locked page to the pagecache
713 * @page: page to add
714 * @mapping: the page's address_space
715 * @offset: page index
716 * @gfp_mask: page allocation mode
717 *
718 * This function is used to add a page to the pagecache. It must be locked.
719 * This function does not add the page to the LRU. The caller must do that.
720 */
721int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
722 pgoff_t offset, gfp_t gfp_mask)
723{
724 return __add_to_page_cache_locked(page, mapping, offset,
725 gfp_mask, NULL);
726}
Nick Piggine2867812008-07-25 19:45:30 -0700727EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
729int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400730 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
Johannes Weinera5289102014-04-03 14:47:51 -0700732 void *shadow = NULL;
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700733 int ret;
734
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800735 __SetPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700736 ret = __add_to_page_cache_locked(page, mapping, offset,
737 gfp_mask, &shadow);
738 if (unlikely(ret))
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800739 __ClearPageLocked(page);
Johannes Weinera5289102014-04-03 14:47:51 -0700740 else {
741 /*
742 * The page might have been evicted from cache only
743 * recently, in which case it should be activated like
744 * any other repeatedly accessed page.
Rik van Rielf0281a02016-05-20 16:56:25 -0700745 * The exception is pages getting rewritten; evicting other
746 * data from the working set, only to cache data that will
747 * get overwritten with something else, is a waste of memory.
Johannes Weinera5289102014-04-03 14:47:51 -0700748 */
Rik van Rielf0281a02016-05-20 16:56:25 -0700749 if (!(gfp_mask & __GFP_WRITE) &&
750 shadow && workingset_refault(shadow)) {
Johannes Weinera5289102014-04-03 14:47:51 -0700751 SetPageActive(page);
752 workingset_activation(page);
753 } else
754 ClearPageActive(page);
755 lru_cache_add(page);
756 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 return ret;
758}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300759EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Paul Jackson44110fe2006-03-24 03:16:04 -0800761#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700762struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800763{
Miao Xiec0ff7452010-05-24 14:32:08 -0700764 int n;
765 struct page *page;
766
Paul Jackson44110fe2006-03-24 03:16:04 -0800767 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700768 unsigned int cpuset_mems_cookie;
769 do {
Mel Gormand26914d2014-04-03 14:47:24 -0700770 cpuset_mems_cookie = read_mems_allowed_begin();
Mel Gormancc9a6c82012-03-21 16:34:11 -0700771 n = cpuset_mem_spread_node();
Vlastimil Babka96db8002015-09-08 15:03:50 -0700772 page = __alloc_pages_node(n, gfp, 0);
Mel Gormand26914d2014-04-03 14:47:24 -0700773 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
Mel Gormancc9a6c82012-03-21 16:34:11 -0700774
Miao Xiec0ff7452010-05-24 14:32:08 -0700775 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800776 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700777 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800778}
Nick Piggin2ae88142006-10-28 10:38:23 -0700779EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800780#endif
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782/*
783 * In order to wait for pages to become available there must be
784 * waitqueues associated with pages. By using a hash table of
785 * waitqueues where the bucket discipline is to maintain all
786 * waiters on the same queue and wake all when any of the pages
787 * become available, and for the woken contexts to check to be
788 * sure the appropriate page became available, this saves space
789 * at a cost of "thundering herd" phenomena during rare hash
790 * collisions.
791 */
NeilBrowna4796e32014-09-24 11:28:32 +1000792wait_queue_head_t *page_waitqueue(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
Linus Torvalds9dcb8b62016-10-26 10:15:30 -0700794 return bit_waitqueue(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795}
NeilBrowna4796e32014-09-24 11:28:32 +1000796EXPORT_SYMBOL(page_waitqueue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
Harvey Harrison920c7a52008-02-04 22:29:26 -0800798void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
800 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
801
802 if (test_bit(bit_nr, &page->flags))
NeilBrown74316202014-07-07 15:16:04 +1000803 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 TASK_UNINTERRUPTIBLE);
805}
806EXPORT_SYMBOL(wait_on_page_bit);
807
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700808int wait_on_page_bit_killable(struct page *page, int bit_nr)
809{
810 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
811
812 if (!test_bit(bit_nr, &page->flags))
813 return 0;
814
815 return __wait_on_bit(page_waitqueue(page), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000816 bit_wait_io, TASK_KILLABLE);
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700817}
818
NeilBrowncbbce822014-09-25 13:55:19 +1000819int wait_on_page_bit_killable_timeout(struct page *page,
820 int bit_nr, unsigned long timeout)
821{
822 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
823
824 wait.key.timeout = jiffies + timeout;
825 if (!test_bit(bit_nr, &page->flags))
826 return 0;
827 return __wait_on_bit(page_waitqueue(page), &wait,
828 bit_wait_io_timeout, TASK_KILLABLE);
829}
830EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
831
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100833 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700834 * @page: Page defining the wait queue of interest
835 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100836 *
837 * Add an arbitrary @waiter to the wait queue for the nominated @page.
838 */
839void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
840{
841 wait_queue_head_t *q = page_waitqueue(page);
842 unsigned long flags;
843
844 spin_lock_irqsave(&q->lock, flags);
845 __add_wait_queue(q, waiter);
846 spin_unlock_irqrestore(&q->lock, flags);
847}
848EXPORT_SYMBOL_GPL(add_page_wait_queue);
849
850/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700851 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 * @page: the page
853 *
854 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
855 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
Masanari Iidada3dae52014-09-09 01:27:23 +0900856 * mechanism between PageLocked pages and PageWriteback pages is shared.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
858 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700859 * The mb is necessary to enforce ordering between the clear_bit and the read
860 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800862void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800864 page = compound_head(page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800865 VM_BUG_ON_PAGE(!PageLocked(page), page);
Nick Piggin8413ac92008-10-18 20:26:59 -0700866 clear_bit_unlock(PG_locked, &page->flags);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100867 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 wake_up_page(page, PG_locked);
869}
870EXPORT_SYMBOL(unlock_page);
871
Randy Dunlap485bb992006-06-23 02:03:49 -0700872/**
873 * end_page_writeback - end writeback against a page
874 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 */
876void end_page_writeback(struct page *page)
877{
Mel Gorman888cf2d2014-06-04 16:10:34 -0700878 /*
879 * TestClearPageReclaim could be used here but it is an atomic
880 * operation and overkill in this particular case. Failing to
881 * shuffle a page marked for immediate reclaim is too mild to
882 * justify taking an atomic operation penalty at the end of
883 * ever page writeback.
884 */
885 if (PageReclaim(page)) {
886 ClearPageReclaim(page);
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700887 rotate_reclaimable_page(page);
Mel Gorman888cf2d2014-06-04 16:10:34 -0700888 }
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700889
890 if (!test_clear_page_writeback(page))
891 BUG();
892
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100893 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 wake_up_page(page, PG_writeback);
895}
896EXPORT_SYMBOL(end_page_writeback);
897
Matthew Wilcox57d99842014-06-04 16:07:45 -0700898/*
899 * After completing I/O on a page, call this routine to update the page
900 * flags appropriately
901 */
Jens Axboec11f0c02016-08-05 08:11:04 -0600902void page_endio(struct page *page, bool is_write, int err)
Matthew Wilcox57d99842014-06-04 16:07:45 -0700903{
Jens Axboec11f0c02016-08-05 08:11:04 -0600904 if (!is_write) {
Matthew Wilcox57d99842014-06-04 16:07:45 -0700905 if (!err) {
906 SetPageUptodate(page);
907 } else {
908 ClearPageUptodate(page);
909 SetPageError(page);
910 }
911 unlock_page(page);
Mike Christieabf54542016-08-04 14:23:34 -0600912 } else {
Matthew Wilcox57d99842014-06-04 16:07:45 -0700913 if (err) {
Minchan Kim2c290ee2017-02-24 14:59:59 -0800914 struct address_space *mapping;
915
Matthew Wilcox57d99842014-06-04 16:07:45 -0700916 SetPageError(page);
Minchan Kim2c290ee2017-02-24 14:59:59 -0800917 mapping = page_mapping(page);
918 if (mapping)
919 mapping_set_error(mapping, err);
Matthew Wilcox57d99842014-06-04 16:07:45 -0700920 }
921 end_page_writeback(page);
922 }
923}
924EXPORT_SYMBOL_GPL(page_endio);
925
Randy Dunlap485bb992006-06-23 02:03:49 -0700926/**
927 * __lock_page - get a lock on the page, assuming we need to sleep to get it
928 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800930void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800932 struct page *page_head = compound_head(page);
933 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800935 __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 TASK_UNINTERRUPTIBLE);
937}
938EXPORT_SYMBOL(__lock_page);
939
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800940int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500941{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800942 struct page *page_head = compound_head(page);
943 DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500944
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800945 return __wait_on_bit_lock(page_waitqueue(page_head), &wait,
NeilBrown74316202014-07-07 15:16:04 +1000946 bit_wait_io, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500947}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300948EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500949
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700950/*
951 * Return values:
952 * 1 - page is locked; mmap_sem is still held.
953 * 0 - page is not locked.
954 * mmap_sem has been released (up_read()), unless flags had both
955 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
956 * which case mmap_sem is still held.
957 *
958 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
959 * with the page locked and the mmap_sem unperturbed.
960 */
Michel Lespinassed065bd82010-10-26 14:21:57 -0700961int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
962 unsigned int flags)
963{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700964 if (flags & FAULT_FLAG_ALLOW_RETRY) {
965 /*
966 * CAUTION! In this case, mmap_sem is not released
967 * even though return 0.
968 */
969 if (flags & FAULT_FLAG_RETRY_NOWAIT)
970 return 0;
971
972 up_read(&mm->mmap_sem);
973 if (flags & FAULT_FLAG_KILLABLE)
974 wait_on_page_locked_killable(page);
975 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700976 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700977 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700978 } else {
979 if (flags & FAULT_FLAG_KILLABLE) {
980 int ret;
981
982 ret = __lock_page_killable(page);
983 if (ret) {
984 up_read(&mm->mmap_sem);
985 return 0;
986 }
987 } else
988 __lock_page(page);
989 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700990 }
991}
992
Randy Dunlap485bb992006-06-23 02:03:49 -0700993/**
Johannes Weinere7b563b2014-04-03 14:47:44 -0700994 * page_cache_next_hole - find the next hole (not-present entry)
995 * @mapping: mapping
996 * @index: index
997 * @max_scan: maximum range to search
998 *
999 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1000 * lowest indexed hole.
1001 *
1002 * Returns: the index of the hole if found, otherwise returns an index
1003 * outside of the set specified (in which case 'return - index >=
1004 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1005 * be returned.
1006 *
1007 * page_cache_next_hole may be called under rcu_read_lock. However,
1008 * like radix_tree_gang_lookup, this will not atomically search a
1009 * snapshot of the tree at a single point in time. For example, if a
1010 * hole is created at index 5, then subsequently a hole is created at
1011 * index 10, page_cache_next_hole covering both indexes may return 10
1012 * if called under rcu_read_lock.
1013 */
1014pgoff_t page_cache_next_hole(struct address_space *mapping,
1015 pgoff_t index, unsigned long max_scan)
1016{
1017 unsigned long i;
1018
1019 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001020 struct page *page;
1021
1022 page = radix_tree_lookup(&mapping->page_tree, index);
1023 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001024 break;
1025 index++;
1026 if (index == 0)
1027 break;
1028 }
1029
1030 return index;
1031}
1032EXPORT_SYMBOL(page_cache_next_hole);
1033
1034/**
1035 * page_cache_prev_hole - find the prev hole (not-present entry)
1036 * @mapping: mapping
1037 * @index: index
1038 * @max_scan: maximum range to search
1039 *
1040 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1041 * the first hole.
1042 *
1043 * Returns: the index of the hole if found, otherwise returns an index
1044 * outside of the set specified (in which case 'index - return >=
1045 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1046 * will be returned.
1047 *
1048 * page_cache_prev_hole may be called under rcu_read_lock. However,
1049 * like radix_tree_gang_lookup, this will not atomically search a
1050 * snapshot of the tree at a single point in time. For example, if a
1051 * hole is created at index 10, then subsequently a hole is created at
1052 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1053 * called under rcu_read_lock.
1054 */
1055pgoff_t page_cache_prev_hole(struct address_space *mapping,
1056 pgoff_t index, unsigned long max_scan)
1057{
1058 unsigned long i;
1059
1060 for (i = 0; i < max_scan; i++) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07001061 struct page *page;
1062
1063 page = radix_tree_lookup(&mapping->page_tree, index);
1064 if (!page || radix_tree_exceptional_entry(page))
Johannes Weinere7b563b2014-04-03 14:47:44 -07001065 break;
1066 index--;
1067 if (index == ULONG_MAX)
1068 break;
1069 }
1070
1071 return index;
1072}
1073EXPORT_SYMBOL(page_cache_prev_hole);
1074
1075/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001076 * find_get_entry - find and get a page cache entry
Randy Dunlap485bb992006-06-23 02:03:49 -07001077 * @mapping: the address_space to search
Johannes Weiner0cd61442014-04-03 14:47:46 -07001078 * @offset: the page cache index
Randy Dunlap485bb992006-06-23 02:03:49 -07001079 *
Johannes Weiner0cd61442014-04-03 14:47:46 -07001080 * Looks up the page cache slot at @mapping & @offset. If there is a
1081 * page cache page, it is returned with an increased refcount.
1082 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001083 * If the slot holds a shadow entry of a previously evicted page, or a
1084 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001085 *
1086 * Otherwise, %NULL is returned.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 */
Johannes Weiner0cd61442014-04-03 14:47:46 -07001088struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089{
Nick Piggina60637c2008-07-25 19:45:31 -07001090 void **pagep;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001091 struct page *head, *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Nick Piggina60637c2008-07-25 19:45:31 -07001093 rcu_read_lock();
1094repeat:
1095 page = NULL;
1096 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1097 if (pagep) {
1098 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -08001099 if (unlikely(!page))
1100 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001101 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001102 if (radix_tree_deref_retry(page))
1103 goto repeat;
1104 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001105 * A shadow entry of a recently evicted page,
1106 * or a swap entry from shmem/tmpfs. Return
1107 * it without attempting to raise page count.
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001108 */
1109 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001110 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001111
1112 head = compound_head(page);
1113 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001114 goto repeat;
1115
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001116 /* The page was split under us? */
1117 if (compound_head(page) != head) {
1118 put_page(head);
1119 goto repeat;
1120 }
1121
Nick Piggina60637c2008-07-25 19:45:31 -07001122 /*
1123 * Has the page moved?
1124 * This is part of the lockless pagecache protocol. See
1125 * include/linux/pagemap.h for details.
1126 */
1127 if (unlikely(page != *pagep)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001128 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001129 goto repeat;
1130 }
1131 }
Nick Piggin27d20fd2010-11-11 14:05:19 -08001132out:
Nick Piggina60637c2008-07-25 19:45:31 -07001133 rcu_read_unlock();
1134
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 return page;
1136}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001137EXPORT_SYMBOL(find_get_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Randy Dunlap485bb992006-06-23 02:03:49 -07001139/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001140 * find_lock_entry - locate, pin and lock a page cache entry
1141 * @mapping: the address_space to search
1142 * @offset: the page cache index
1143 *
1144 * Looks up the page cache slot at @mapping & @offset. If there is a
1145 * page cache page, it is returned locked and with an increased
1146 * refcount.
1147 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001148 * If the slot holds a shadow entry of a previously evicted page, or a
1149 * swap entry from shmem/tmpfs, it is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001150 *
1151 * Otherwise, %NULL is returned.
1152 *
1153 * find_lock_entry() may sleep.
1154 */
1155struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156{
1157 struct page *page;
1158
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159repeat:
Johannes Weiner0cd61442014-04-03 14:47:46 -07001160 page = find_get_entry(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001161 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001162 lock_page(page);
1163 /* Has the page been truncated? */
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001164 if (unlikely(page_mapping(page) != mapping)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001165 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001166 put_page(page);
Nick Piggina60637c2008-07-25 19:45:31 -07001167 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001169 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 return page;
1172}
Johannes Weiner0cd61442014-04-03 14:47:46 -07001173EXPORT_SYMBOL(find_lock_entry);
1174
1175/**
Mel Gorman2457aec2014-06-04 16:10:31 -07001176 * pagecache_get_page - find and get a page reference
Johannes Weiner0cd61442014-04-03 14:47:46 -07001177 * @mapping: the address_space to search
1178 * @offset: the page index
Mel Gorman2457aec2014-06-04 16:10:31 -07001179 * @fgp_flags: PCG flags
Michal Hocko45f87de2014-12-29 20:30:35 +01001180 * @gfp_mask: gfp mask to use for the page cache data page allocation
Johannes Weiner0cd61442014-04-03 14:47:46 -07001181 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001182 * Looks up the page cache slot at @mapping & @offset.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001183 *
Randy Dunlap75325182014-07-30 16:08:37 -07001184 * PCG flags modify how the page is returned.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001185 *
Mel Gorman2457aec2014-06-04 16:10:31 -07001186 * FGP_ACCESSED: the page will be marked accessed
1187 * FGP_LOCK: Page is return locked
1188 * FGP_CREAT: If page is not present then a new page is allocated using
Michal Hocko45f87de2014-12-29 20:30:35 +01001189 * @gfp_mask and added to the page cache and the VM's LRU
1190 * list. The page is returned locked and with an increased
1191 * refcount. Otherwise, %NULL is returned.
Mel Gorman2457aec2014-06-04 16:10:31 -07001192 *
1193 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1194 * if the GFP flags specified for FGP_CREAT are atomic.
1195 *
1196 * If there is a page cache page, it is returned with an increased refcount.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001197 */
Mel Gorman2457aec2014-06-04 16:10:31 -07001198struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
Michal Hocko45f87de2014-12-29 20:30:35 +01001199 int fgp_flags, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200{
Nick Piggineb2be182007-10-16 01:24:57 -07001201 struct page *page;
Mel Gorman2457aec2014-06-04 16:10:31 -07001202
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203repeat:
Mel Gorman2457aec2014-06-04 16:10:31 -07001204 page = find_get_entry(mapping, offset);
1205 if (radix_tree_exceptional_entry(page))
1206 page = NULL;
1207 if (!page)
1208 goto no_page;
1209
1210 if (fgp_flags & FGP_LOCK) {
1211 if (fgp_flags & FGP_NOWAIT) {
1212 if (!trylock_page(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001213 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001214 return NULL;
1215 }
1216 } else {
1217 lock_page(page);
1218 }
1219
1220 /* Has the page been truncated? */
1221 if (unlikely(page->mapping != mapping)) {
1222 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001223 put_page(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001224 goto repeat;
1225 }
1226 VM_BUG_ON_PAGE(page->index != offset, page);
1227 }
1228
1229 if (page && (fgp_flags & FGP_ACCESSED))
1230 mark_page_accessed(page);
1231
1232no_page:
1233 if (!page && (fgp_flags & FGP_CREAT)) {
1234 int err;
1235 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
Michal Hocko45f87de2014-12-29 20:30:35 +01001236 gfp_mask |= __GFP_WRITE;
1237 if (fgp_flags & FGP_NOFS)
1238 gfp_mask &= ~__GFP_FS;
Mel Gorman2457aec2014-06-04 16:10:31 -07001239
Michal Hocko45f87de2014-12-29 20:30:35 +01001240 page = __page_cache_alloc(gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001241 if (!page)
1242 return NULL;
Mel Gorman2457aec2014-06-04 16:10:31 -07001243
1244 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1245 fgp_flags |= FGP_LOCK;
1246
Hugh Dickinseb39d612014-08-06 16:06:43 -07001247 /* Init accessed so avoid atomic mark_page_accessed later */
Mel Gorman2457aec2014-06-04 16:10:31 -07001248 if (fgp_flags & FGP_ACCESSED)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001249 __SetPageReferenced(page);
Mel Gorman2457aec2014-06-04 16:10:31 -07001250
Matthew Wilcoxf4c86fa2018-04-20 14:56:20 -07001251 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
Nick Piggineb2be182007-10-16 01:24:57 -07001252 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001253 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07001254 page = NULL;
1255 if (err == -EEXIST)
1256 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 }
Mel Gorman2457aec2014-06-04 16:10:31 -07001259
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 return page;
1261}
Mel Gorman2457aec2014-06-04 16:10:31 -07001262EXPORT_SYMBOL(pagecache_get_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
1264/**
Johannes Weiner0cd61442014-04-03 14:47:46 -07001265 * find_get_entries - gang pagecache lookup
1266 * @mapping: The address_space to search
1267 * @start: The starting page cache index
1268 * @nr_entries: The maximum number of entries
1269 * @entries: Where the resulting entries are placed
1270 * @indices: The cache indices corresponding to the entries in @entries
1271 *
1272 * find_get_entries() will search for and return a group of up to
1273 * @nr_entries entries in the mapping. The entries are placed at
1274 * @entries. find_get_entries() takes a reference against any actual
1275 * pages it returns.
1276 *
1277 * The search returns a group of mapping-contiguous page cache entries
1278 * with ascending indexes. There may be holes in the indices due to
1279 * not-present pages.
1280 *
Johannes Weiner139b6a62014-05-06 12:50:05 -07001281 * Any shadow entries of evicted pages, or swap entries from
1282 * shmem/tmpfs, are included in the returned array.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001283 *
1284 * find_get_entries() returns the number of pages and shadow entries
1285 * which were found.
1286 */
1287unsigned find_get_entries(struct address_space *mapping,
1288 pgoff_t start, unsigned int nr_entries,
1289 struct page **entries, pgoff_t *indices)
1290{
1291 void **slot;
1292 unsigned int ret = 0;
1293 struct radix_tree_iter iter;
1294
1295 if (!nr_entries)
1296 return 0;
1297
1298 rcu_read_lock();
Johannes Weiner0cd61442014-04-03 14:47:46 -07001299 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001300 struct page *head, *page;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001301repeat:
1302 page = radix_tree_deref_slot(slot);
1303 if (unlikely(!page))
1304 continue;
1305 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001306 if (radix_tree_deref_retry(page)) {
1307 slot = radix_tree_iter_retry(&iter);
1308 continue;
1309 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07001310 /*
Ross Zwislerf9fe48b2016-01-22 15:10:40 -08001311 * A shadow entry of a recently evicted page, a swap
1312 * entry from shmem/tmpfs or a DAX entry. Return it
1313 * without attempting to raise page count.
Johannes Weiner0cd61442014-04-03 14:47:46 -07001314 */
1315 goto export;
1316 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001317
1318 head = compound_head(page);
1319 if (!page_cache_get_speculative(head))
Johannes Weiner0cd61442014-04-03 14:47:46 -07001320 goto repeat;
1321
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001322 /* The page was split under us? */
1323 if (compound_head(page) != head) {
1324 put_page(head);
1325 goto repeat;
1326 }
1327
Johannes Weiner0cd61442014-04-03 14:47:46 -07001328 /* Has the page moved? */
1329 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001330 put_page(head);
Johannes Weiner0cd61442014-04-03 14:47:46 -07001331 goto repeat;
1332 }
1333export:
1334 indices[ret] = iter.index;
1335 entries[ret] = page;
1336 if (++ret == nr_entries)
1337 break;
1338 }
1339 rcu_read_unlock();
1340 return ret;
1341}
1342
1343/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 * find_get_pages - gang pagecache lookup
1345 * @mapping: The address_space to search
1346 * @start: The starting page index
1347 * @nr_pages: The maximum number of pages
1348 * @pages: Where the resulting pages are placed
1349 *
1350 * find_get_pages() will search for and return a group of up to
1351 * @nr_pages pages in the mapping. The pages are placed at @pages.
1352 * find_get_pages() takes a reference against the returned pages.
1353 *
1354 * The search returns a group of mapping-contiguous pages with ascending
1355 * indexes. There may be holes in the indices due to not-present pages.
1356 *
1357 * find_get_pages() returns the number of pages which were found.
1358 */
1359unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1360 unsigned int nr_pages, struct page **pages)
1361{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001362 struct radix_tree_iter iter;
1363 void **slot;
1364 unsigned ret = 0;
1365
1366 if (unlikely(!nr_pages))
1367 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
Nick Piggina60637c2008-07-25 19:45:31 -07001369 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001370 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001371 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001372repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001373 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001374 if (unlikely(!page))
1375 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001376
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001377 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001378 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001379 slot = radix_tree_iter_retry(&iter);
1380 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001381 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001382 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001383 * A shadow entry of a recently evicted page,
1384 * or a swap entry from shmem/tmpfs. Skip
1385 * over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001386 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001387 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -08001388 }
Nick Piggina60637c2008-07-25 19:45:31 -07001389
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001390 head = compound_head(page);
1391 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001392 goto repeat;
1393
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001394 /* The page was split under us? */
1395 if (compound_head(page) != head) {
1396 put_page(head);
1397 goto repeat;
1398 }
1399
Nick Piggina60637c2008-07-25 19:45:31 -07001400 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001401 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001402 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001403 goto repeat;
1404 }
1405
1406 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001407 if (++ret == nr_pages)
1408 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001409 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001410
Nick Piggina60637c2008-07-25 19:45:31 -07001411 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 return ret;
1413}
1414
Jens Axboeebf43502006-04-27 08:46:01 +02001415/**
1416 * find_get_pages_contig - gang contiguous pagecache lookup
1417 * @mapping: The address_space to search
1418 * @index: The starting page index
1419 * @nr_pages: The maximum number of pages
1420 * @pages: Where the resulting pages are placed
1421 *
1422 * find_get_pages_contig() works exactly like find_get_pages(), except
1423 * that the returned number of pages are guaranteed to be contiguous.
1424 *
1425 * find_get_pages_contig() returns the number of pages which were found.
1426 */
1427unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1428 unsigned int nr_pages, struct page **pages)
1429{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001430 struct radix_tree_iter iter;
1431 void **slot;
1432 unsigned int ret = 0;
1433
1434 if (unlikely(!nr_pages))
1435 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +02001436
Nick Piggina60637c2008-07-25 19:45:31 -07001437 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001438 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001439 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001440repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001441 page = radix_tree_deref_slot(slot);
1442 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -07001443 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001444 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001445
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001446 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001447 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001448 slot = radix_tree_iter_retry(&iter);
1449 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001450 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001451 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001452 * A shadow entry of a recently evicted page,
1453 * or a swap entry from shmem/tmpfs. Stop
1454 * looking for contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001455 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001456 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001457 }
Nick Piggina60637c2008-07-25 19:45:31 -07001458
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001459 head = compound_head(page);
1460 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001461 goto repeat;
1462
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001463 /* The page was split under us? */
1464 if (compound_head(page) != head) {
1465 put_page(head);
1466 goto repeat;
1467 }
1468
Nick Piggina60637c2008-07-25 19:45:31 -07001469 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001470 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001471 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001472 goto repeat;
1473 }
1474
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001475 /*
1476 * must check mapping and index after taking the ref.
1477 * otherwise we can get both false positives and false
1478 * negatives, which is just confusing to the caller.
1479 */
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001480 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001481 put_page(page);
Nick Piggin9cbb4cb2011-01-13 15:45:51 -08001482 break;
1483 }
1484
Nick Piggina60637c2008-07-25 19:45:31 -07001485 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001486 if (++ret == nr_pages)
1487 break;
Jens Axboeebf43502006-04-27 08:46:01 +02001488 }
Nick Piggina60637c2008-07-25 19:45:31 -07001489 rcu_read_unlock();
1490 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +02001491}
David Howellsef71c152007-05-09 02:33:44 -07001492EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +02001493
Randy Dunlap485bb992006-06-23 02:03:49 -07001494/**
1495 * find_get_pages_tag - find and return pages that match @tag
1496 * @mapping: the address_space to search
1497 * @index: the starting page index
1498 * @tag: the tag index
1499 * @nr_pages: the maximum number of pages
1500 * @pages: where the resulting pages are placed
1501 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -07001503 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 */
1505unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1506 int tag, unsigned int nr_pages, struct page **pages)
1507{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001508 struct radix_tree_iter iter;
1509 void **slot;
1510 unsigned ret = 0;
1511
1512 if (unlikely(!nr_pages))
1513 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
Nick Piggina60637c2008-07-25 19:45:31 -07001515 rcu_read_lock();
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001516 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1517 &iter, *index, tag) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001518 struct page *head, *page;
Nick Piggina60637c2008-07-25 19:45:31 -07001519repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001520 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -07001521 if (unlikely(!page))
1522 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -07001523
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001524 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001525 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001526 slot = radix_tree_iter_retry(&iter);
1527 continue;
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001528 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001529 /*
Johannes Weiner139b6a62014-05-06 12:50:05 -07001530 * A shadow entry of a recently evicted page.
1531 *
1532 * Those entries should never be tagged, but
1533 * this tree walk is lockless and the tags are
1534 * looked up in bulk, one radix tree node at a
1535 * time, so there is a sizable window for page
1536 * reclaim to evict a page we saw tagged.
1537 *
1538 * Skip over it.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001539 */
Johannes Weiner139b6a62014-05-06 12:50:05 -07001540 continue;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001541 }
Nick Piggina60637c2008-07-25 19:45:31 -07001542
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001543 head = compound_head(page);
1544 if (!page_cache_get_speculative(head))
Nick Piggina60637c2008-07-25 19:45:31 -07001545 goto repeat;
1546
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001547 /* The page was split under us? */
1548 if (compound_head(page) != head) {
1549 put_page(head);
1550 goto repeat;
1551 }
1552
Nick Piggina60637c2008-07-25 19:45:31 -07001553 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001554 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001555 put_page(head);
Nick Piggina60637c2008-07-25 19:45:31 -07001556 goto repeat;
1557 }
1558
1559 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001560 if (++ret == nr_pages)
1561 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001562 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001563
Nick Piggina60637c2008-07-25 19:45:31 -07001564 rcu_read_unlock();
1565
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 if (ret)
1567 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001568
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 return ret;
1570}
David Howellsef71c152007-05-09 02:33:44 -07001571EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001573/**
1574 * find_get_entries_tag - find and return entries that match @tag
1575 * @mapping: the address_space to search
1576 * @start: the starting page cache index
1577 * @tag: the tag index
1578 * @nr_entries: the maximum number of entries
1579 * @entries: where the resulting entries are placed
1580 * @indices: the cache indices corresponding to the entries in @entries
1581 *
1582 * Like find_get_entries, except we only return entries which are tagged with
1583 * @tag.
1584 */
1585unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1586 int tag, unsigned int nr_entries,
1587 struct page **entries, pgoff_t *indices)
1588{
1589 void **slot;
1590 unsigned int ret = 0;
1591 struct radix_tree_iter iter;
1592
1593 if (!nr_entries)
1594 return 0;
1595
1596 rcu_read_lock();
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001597 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1598 &iter, start, tag) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001599 struct page *head, *page;
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001600repeat:
1601 page = radix_tree_deref_slot(slot);
1602 if (unlikely(!page))
1603 continue;
1604 if (radix_tree_exception(page)) {
1605 if (radix_tree_deref_retry(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07001606 slot = radix_tree_iter_retry(&iter);
1607 continue;
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001608 }
1609
1610 /*
1611 * A shadow entry of a recently evicted page, a swap
1612 * entry from shmem/tmpfs or a DAX entry. Return it
1613 * without attempting to raise page count.
1614 */
1615 goto export;
1616 }
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001617
1618 head = compound_head(page);
1619 if (!page_cache_get_speculative(head))
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001620 goto repeat;
1621
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001622 /* The page was split under us? */
1623 if (compound_head(page) != head) {
1624 put_page(head);
1625 goto repeat;
1626 }
1627
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001628 /* Has the page moved? */
1629 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07001630 put_page(head);
Ross Zwisler7e7f7742016-01-22 15:10:44 -08001631 goto repeat;
1632 }
1633export:
1634 indices[ret] = iter.index;
1635 entries[ret] = page;
1636 if (++ret == nr_entries)
1637 break;
1638 }
1639 rcu_read_unlock();
1640 return ret;
1641}
1642EXPORT_SYMBOL(find_get_entries_tag);
1643
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001644/*
1645 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1646 * a _large_ part of the i/o request. Imagine the worst scenario:
1647 *
1648 * ---R__________________________________________B__________
1649 * ^ reading here ^ bad block(assume 4k)
1650 *
1651 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1652 * => failing the whole request => read(R) => read(R+1) =>
1653 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1654 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1655 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1656 *
1657 * It is going insane. Fix it by quickly scaling down the readahead size.
1658 */
1659static void shrink_readahead_size_eio(struct file *filp,
1660 struct file_ra_state *ra)
1661{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001662 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001663}
1664
Randy Dunlap485bb992006-06-23 02:03:49 -07001665/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001666 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001667 * @filp: the file to read
1668 * @ppos: current file position
Al Viro6e58e792014-02-03 17:07:03 -05001669 * @iter: data destination
1670 * @written: already copied
Randy Dunlap485bb992006-06-23 02:03:49 -07001671 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001673 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 *
1675 * This is really ugly. But the goto's actually try to clarify some
1676 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 */
Al Viro6e58e792014-02-03 17:07:03 -05001678static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1679 struct iov_iter *iter, ssize_t written)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001681 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001683 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001684 pgoff_t index;
1685 pgoff_t last_index;
1686 pgoff_t prev_index;
1687 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001688 unsigned int prev_offset;
Al Viro6e58e792014-02-03 17:07:03 -05001689 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
Wei Fangc2a97372016-10-07 17:01:52 -07001691 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
Linus Torvalds48466c42016-12-14 12:45:25 -08001692 return 0;
Wei Fangc2a97372016-10-07 17:01:52 -07001693 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1694
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001695 index = *ppos >> PAGE_SHIFT;
1696 prev_index = ra->prev_pos >> PAGE_SHIFT;
1697 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1698 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1699 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 for (;;) {
1702 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001703 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001704 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 unsigned long nr, ret;
1706
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708find_page:
Michal Hockob67c7d32017-02-03 13:13:29 -08001709 if (fatal_signal_pending(current)) {
1710 error = -EINTR;
1711 goto out;
1712 }
1713
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001715 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001716 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001717 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001718 index, last_index - index);
1719 page = find_get_page(mapping, index);
1720 if (unlikely(page == NULL))
1721 goto no_cached_page;
1722 }
1723 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001724 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001725 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001726 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001728 if (!PageUptodate(page)) {
Mel Gormanebded022016-03-15 14:55:39 -07001729 /*
1730 * See comment in do_read_cache_page on why
1731 * wait_on_page_locked is used to avoid unnecessarily
1732 * serialisations and why it's safe.
1733 */
Bart Van Asschec4b209a2016-10-07 16:58:33 -07001734 error = wait_on_page_locked_killable(page);
1735 if (unlikely(error))
1736 goto readpage_error;
Mel Gormanebded022016-03-15 14:55:39 -07001737 if (PageUptodate(page))
1738 goto page_ok;
1739
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001740 if (inode->i_blkbits == PAGE_SHIFT ||
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001741 !mapping->a_ops->is_partially_uptodate)
1742 goto page_not_up_to_date;
Eryu Guan6d6d36b2016-11-01 15:43:07 +08001743 /* pipes can't handle partially uptodate pages */
1744 if (unlikely(iter->type & ITER_PIPE))
1745 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001746 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001747 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001748 /* Did it get truncated before we got the lock? */
1749 if (!page->mapping)
1750 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001751 if (!mapping->a_ops->is_partially_uptodate(page,
Al Viro6e58e792014-02-03 17:07:03 -05001752 offset, iter->count))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001753 goto page_not_up_to_date_locked;
1754 unlock_page(page);
1755 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001757 /*
1758 * i_size must be checked after we know the page is Uptodate.
1759 *
1760 * Checking i_size after the check allows us to calculate
1761 * the correct value for "nr", which means the zero-filled
1762 * part of the page is not copied back to userspace (unless
1763 * another truncate extends the file - this is desired though).
1764 */
1765
1766 isize = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001767 end_index = (isize - 1) >> PAGE_SHIFT;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001768 if (unlikely(!isize || index > end_index)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001769 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07001770 goto out;
1771 }
1772
1773 /* nr is the maximum number of bytes to copy from this page */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001774 nr = PAGE_SIZE;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001775 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001776 nr = ((isize - 1) & ~PAGE_MASK) + 1;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001777 if (nr <= offset) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001778 put_page(page);
NeilBrowna32ea1e2007-07-17 04:03:04 -07001779 goto out;
1780 }
1781 }
1782 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783
1784 /* If users can be writing to this page using arbitrary
1785 * virtual addresses, take care about potential aliasing
1786 * before reading the page on the kernel side.
1787 */
1788 if (mapping_writably_mapped(mapping))
1789 flush_dcache_page(page);
1790
1791 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001792 * When a sequential read accesses a page several times,
1793 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 */
Jan Karaec0f1632007-05-06 14:49:25 -07001795 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 mark_page_accessed(page);
1797 prev_index = index;
1798
1799 /*
1800 * Ok, we have the page, and it's up-to-date, so
1801 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 */
Al Viro6e58e792014-02-03 17:07:03 -05001803
1804 ret = copy_page_to_iter(page, offset, nr, iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001806 index += offset >> PAGE_SHIFT;
1807 offset &= ~PAGE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001808 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001810 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001811 written += ret;
1812 if (!iov_iter_count(iter))
1813 goto out;
1814 if (ret < nr) {
1815 error = -EFAULT;
1816 goto out;
1817 }
1818 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819
1820page_not_up_to_date:
1821 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001822 error = lock_page_killable(page);
1823 if (unlikely(error))
1824 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001826page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001827 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 if (!page->mapping) {
1829 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001830 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 continue;
1832 }
1833
1834 /* Did somebody else fill it already? */
1835 if (PageUptodate(page)) {
1836 unlock_page(page);
1837 goto page_ok;
1838 }
1839
1840readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001841 /*
1842 * A previous I/O error may have been due to temporary
1843 * failures, eg. multipath errors.
1844 * PG_error will be set again if readpage fails.
1845 */
1846 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 /* Start the actual read. The read will unlock the page. */
1848 error = mapping->a_ops->readpage(filp, page);
1849
Zach Brown994fc28c2005-12-15 14:28:17 -08001850 if (unlikely(error)) {
1851 if (error == AOP_TRUNCATED_PAGE) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001852 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001853 error = 0;
Zach Brown994fc28c2005-12-15 14:28:17 -08001854 goto find_page;
1855 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001857 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001860 error = lock_page_killable(page);
1861 if (unlikely(error))
1862 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 if (!PageUptodate(page)) {
1864 if (page->mapping == NULL) {
1865 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001866 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 */
1868 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001869 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 goto find_page;
1871 }
1872 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001873 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001874 error = -EIO;
1875 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 }
1877 unlock_page(page);
1878 }
1879
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 goto page_ok;
1881
1882readpage_error:
1883 /* UHHUH! A synchronous read error occurred. Report it */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001884 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 goto out;
1886
1887no_cached_page:
1888 /*
1889 * Ok, it wasn't cached, so we need to create a new
1890 * page..
1891 */
Nick Piggineb2be182007-10-16 01:24:57 -07001892 page = page_cache_alloc_cold(mapping);
1893 if (!page) {
Al Viro6e58e792014-02-03 17:07:03 -05001894 error = -ENOMEM;
Nick Piggineb2be182007-10-16 01:24:57 -07001895 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 }
Michal Hocko6afdb852015-06-24 16:58:06 -07001897 error = add_to_page_cache_lru(page, mapping, index,
Michal Hockoc62d2552015-11-06 16:28:49 -08001898 mapping_gfp_constraint(mapping, GFP_KERNEL));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 if (error) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001900 put_page(page);
Al Viro6e58e792014-02-03 17:07:03 -05001901 if (error == -EEXIST) {
1902 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 goto find_page;
Al Viro6e58e792014-02-03 17:07:03 -05001904 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 goto out;
1906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 goto readpage;
1908 }
1909
1910out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001911 ra->prev_pos = prev_index;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001912 ra->prev_pos <<= PAGE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001913 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001915 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001916 file_accessed(filp);
Al Viro6e58e792014-02-03 17:07:03 -05001917 return written ? written : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918}
1919
Randy Dunlap485bb992006-06-23 02:03:49 -07001920/**
Al Viro6abd2322014-04-04 14:20:57 -04001921 * generic_file_read_iter - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001922 * @iocb: kernel I/O control block
Al Viro6abd2322014-04-04 14:20:57 -04001923 * @iter: destination for the data read
Randy Dunlap485bb992006-06-23 02:03:49 -07001924 *
Al Viro6abd2322014-04-04 14:20:57 -04001925 * This is the "read_iter()" routine for all filesystems
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 * that can use the page cache directly.
1927 */
1928ssize_t
Al Viroed978a82014-03-05 22:53:04 -05001929generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930{
Al Viroed978a82014-03-05 22:53:04 -05001931 struct file *file = iocb->ki_filp;
1932 ssize_t retval = 0;
Nicolai Stangee7080a42016-03-25 14:22:14 -07001933 size_t count = iov_iter_count(iter);
1934
1935 if (!count)
1936 goto out; /* skip atime */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
Al Viro2ba48ce2015-04-09 13:52:01 -04001938 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viroed978a82014-03-05 22:53:04 -05001939 struct address_space *mapping = file->f_mapping;
1940 struct inode *inode = mapping->host;
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11001941 struct iov_iter data = *iter;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001942 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 size = i_size_read(inode);
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001945 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
1946 iocb->ki_pos + count - 1);
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11001947 if (retval < 0)
1948 goto out;
Al Viroed978a82014-03-05 22:53:04 -05001949
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11001950 file_accessed(file);
1951
1952 retval = mapping->a_ops->direct_IO(iocb, &data);
Al Viroc3a69022016-10-10 13:26:27 -04001953 if (retval >= 0) {
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001954 iocb->ki_pos += retval;
Al Viroed978a82014-03-05 22:53:04 -05001955 iov_iter_advance(iter, retval);
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001956 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001957
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001958 /*
1959 * Btrfs can have a short DIO read if we encounter
1960 * compressed extents, so if there was an error, or if
1961 * we've already read everything we wanted to, or if
1962 * there was a short read because we hit EOF, go ahead
1963 * and return. Otherwise fallthrough to buffered io for
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08001964 * the rest of the read. Buffered reads will not work for
1965 * DAX files, so don't bother trying.
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001966 */
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001967 if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size ||
Christoph Hellwig0d5b0cf2016-10-03 09:48:08 +11001968 IS_DAX(inode))
Steven Whitehouse9fe55ee2014-01-24 14:42:22 +00001969 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 }
1971
Christoph Hellwigc64fb5c2016-04-07 08:51:55 -07001972 retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973out:
1974 return retval;
1975}
Al Viroed978a82014-03-05 22:53:04 -05001976EXPORT_SYMBOL(generic_file_read_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001979/**
1980 * page_cache_read - adds requested page to the page cache if not already there
1981 * @file: file to read
1982 * @offset: page index
Randy Dunlap62eb3202016-02-11 16:12:58 -08001983 * @gfp_mask: memory allocation flags
Randy Dunlap485bb992006-06-23 02:03:49 -07001984 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 * This adds the requested page to the page cache if it isn't already there,
1986 * and schedules an I/O to read in its contents from disk.
1987 */
Michal Hockoc20cd452016-01-14 15:20:12 -08001988static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989{
1990 struct address_space *mapping = file->f_mapping;
Paul McQuade99dadfd2014-10-09 15:29:03 -07001991 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001992 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993
Zach Brown994fc28c2005-12-15 14:28:17 -08001994 do {
Michal Hockoc20cd452016-01-14 15:20:12 -08001995 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
Zach Brown994fc28c2005-12-15 14:28:17 -08001996 if (!page)
1997 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
Matthew Wilcoxf4c86fa2018-04-20 14:56:20 -07001999 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
Zach Brown994fc28c2005-12-15 14:28:17 -08002000 if (ret == 0)
2001 ret = mapping->a_ops->readpage(file, page);
2002 else if (ret == -EEXIST)
2003 ret = 0; /* losing race to add is OK */
2004
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002005 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
Zach Brown994fc28c2005-12-15 14:28:17 -08002007 } while (ret == AOP_TRUNCATED_PAGE);
Paul McQuade99dadfd2014-10-09 15:29:03 -07002008
Zach Brown994fc28c2005-12-15 14:28:17 -08002009 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010}
2011
2012#define MMAP_LOTSAMISS (100)
2013
Linus Torvaldsef00e082009-06-16 15:31:25 -07002014/*
2015 * Synchronous readahead happens when we don't even find
2016 * a page in the page cache at all.
2017 */
2018static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2019 struct file_ra_state *ra,
2020 struct file *file,
2021 pgoff_t offset)
2022{
Linus Torvaldsef00e082009-06-16 15:31:25 -07002023 struct address_space *mapping = file->f_mapping;
2024
2025 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07002026 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002027 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002028 if (!ra->ra_pages)
2029 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002030
Joe Perches64363aa2013-07-08 16:00:18 -07002031 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07002032 page_cache_sync_readahead(mapping, ra, file, offset,
2033 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002034 return;
2035 }
2036
Andi Kleen207d04b2011-05-24 17:12:29 -07002037 /* Avoid banging the cache line if not needed */
2038 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002039 ra->mmap_miss++;
2040
2041 /*
2042 * Do we miss much more than hit in this file? If so,
2043 * stop bothering with read-ahead. It will only hurt.
2044 */
2045 if (ra->mmap_miss > MMAP_LOTSAMISS)
2046 return;
2047
Wu Fengguangd30a1102009-06-16 15:31:30 -07002048 /*
2049 * mmap read-around
2050 */
Roman Gushchin600e19a2015-11-05 18:47:08 -08002051 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2052 ra->size = ra->ra_pages;
2053 ra->async_size = ra->ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07002054 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002055}
2056
2057/*
2058 * Asynchronous readahead happens when we find the page and PG_readahead,
2059 * so we want to possibly extend the readahead further..
2060 */
2061static void do_async_mmap_readahead(struct vm_area_struct *vma,
2062 struct file_ra_state *ra,
2063 struct file *file,
2064 struct page *page,
2065 pgoff_t offset)
2066{
2067 struct address_space *mapping = file->f_mapping;
2068
2069 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07002070 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07002071 return;
2072 if (ra->mmap_miss > 0)
2073 ra->mmap_miss--;
2074 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07002075 page_cache_async_readahead(mapping, ra, file,
2076 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002077}
2078
Randy Dunlap485bb992006-06-23 02:03:49 -07002079/**
Nick Piggin54cb8822007-07-19 01:46:59 -07002080 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07002081 * @vma: vma in which the fault was taken
2082 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07002083 *
Nick Piggin54cb8822007-07-19 01:46:59 -07002084 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 * mapped memory region to read in file data during a page fault.
2086 *
2087 * The goto's are kind of ugly, but this streamlines the normal case of having
2088 * it in the page cache, and handles the special cases reasonably without
2089 * having a lot of duplicated code.
Paul Cassella9a95f3c2014-08-06 16:07:24 -07002090 *
2091 * vma->vm_mm->mmap_sem must be held on entry.
2092 *
2093 * If our return value has VM_FAULT_RETRY set, it's because
2094 * lock_page_or_retry() returned 0.
2095 * The mmap_sem has usually been released in this case.
2096 * See __lock_page_or_retry() for the exception.
2097 *
2098 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2099 * has not been released.
2100 *
2101 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 */
Nick Piggind0217ac2007-07-19 01:47:03 -07002103int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104{
2105 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07002106 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 struct address_space *mapping = file->f_mapping;
2108 struct file_ra_state *ra = &file->f_ra;
2109 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07002110 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 struct page *page;
Kirill A. Shutemov99e3e532014-04-07 15:37:21 -07002112 loff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07002113 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002115 size = round_up(i_size_read(inode), PAGE_SIZE);
2116 if (offset >= size >> PAGE_SHIFT)
Linus Torvalds5307cc12007-10-31 09:19:46 -07002117 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 /*
Johannes Weiner49426422013-10-16 13:46:59 -07002120 * Do we have something in the page cache already?
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002122 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002123 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07002125 * We found the page, so try async readahead before
2126 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07002128 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07002129 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07002130 /* No page in the page cache at all */
2131 do_sync_mmap_readahead(vma, ra, file, offset);
2132 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07002133 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07002134 ret = VM_FAULT_MAJOR;
2135retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07002136 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 if (!page)
2138 goto no_cached_page;
2139 }
2140
Michel Lespinassed88c0922010-11-02 13:05:18 -07002141 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002142 put_page(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07002143 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07002144 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07002145
2146 /* Did it get truncated? */
2147 if (unlikely(page->mapping != mapping)) {
2148 unlock_page(page);
2149 put_page(page);
2150 goto retry_find;
2151 }
Sasha Levin309381fea2014-01-23 15:52:54 -08002152 VM_BUG_ON_PAGE(page->index != offset, page);
Michel Lespinasseb522c942010-10-26 14:21:56 -07002153
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 /*
Nick Piggind00806b2007-07-19 01:46:57 -07002155 * We have a locked page in the page cache, now we need to check
2156 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 */
Nick Piggind00806b2007-07-19 01:46:57 -07002158 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 goto page_not_uptodate;
2160
Linus Torvaldsef00e082009-06-16 15:31:25 -07002161 /*
2162 * Found the page and have a reference on it.
2163 * We must recheck i_size under page lock.
2164 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002165 size = round_up(i_size_read(inode), PAGE_SIZE);
2166 if (unlikely(offset >= size >> PAGE_SHIFT)) {
Nick Piggind00806b2007-07-19 01:46:57 -07002167 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002168 put_page(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07002169 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07002170 }
2171
Nick Piggind0217ac2007-07-19 01:47:03 -07002172 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07002173 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175no_cached_page:
2176 /*
2177 * We're only likely to ever get here if MADV_RANDOM is in
2178 * effect.
2179 */
Michal Hockoc20cd452016-01-14 15:20:12 -08002180 error = page_cache_read(file, offset, vmf->gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181
2182 /*
2183 * The page we want has now been added to the page cache.
2184 * In the unlikely event that someone removed it in the
2185 * meantime, we'll just come back here and read it again.
2186 */
2187 if (error >= 0)
2188 goto retry_find;
2189
2190 /*
2191 * An error return from page_cache_read can result if the
2192 * system is low on memory, or a problem occurs while trying
2193 * to schedule I/O.
2194 */
2195 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07002196 return VM_FAULT_OOM;
2197 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198
2199page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 /*
2201 * Umm, take care of errors if the page isn't up-to-date.
2202 * Try to re-read it _once_. We do this synchronously,
2203 * because there really aren't any performance issues here
2204 * and we need to check for errors.
2205 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08002207 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07002208 if (!error) {
2209 wait_on_page_locked(page);
2210 if (!PageUptodate(page))
2211 error = -EIO;
2212 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002213 put_page(page);
Nick Piggind00806b2007-07-19 01:46:57 -07002214
2215 if (!error || error == AOP_TRUNCATED_PAGE)
2216 goto retry_find;
2217
2218 /* Things didn't work out. Return zero to tell the mm layer so. */
2219 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07002220 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07002221}
2222EXPORT_SYMBOL(filemap_fault);
2223
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002224void filemap_map_pages(struct fault_env *fe,
2225 pgoff_t start_pgoff, pgoff_t end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002226{
2227 struct radix_tree_iter iter;
2228 void **slot;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002229 struct file *file = fe->vma->vm_file;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002230 struct address_space *mapping = file->f_mapping;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002231 pgoff_t last_pgoff = start_pgoff;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002232 loff_t size;
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002233 struct page *head, *page;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002234
2235 rcu_read_lock();
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002236 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2237 start_pgoff) {
2238 if (iter.index > end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002239 break;
2240repeat:
2241 page = radix_tree_deref_slot(slot);
2242 if (unlikely(!page))
2243 goto next;
2244 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002245 if (radix_tree_deref_retry(page)) {
2246 slot = radix_tree_iter_retry(&iter);
2247 continue;
2248 }
2249 goto next;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002250 }
2251
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002252 head = compound_head(page);
2253 if (!page_cache_get_speculative(head))
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002254 goto repeat;
2255
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002256 /* The page was split under us? */
2257 if (compound_head(page) != head) {
2258 put_page(head);
2259 goto repeat;
2260 }
2261
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002262 /* Has the page moved? */
2263 if (unlikely(page != *slot)) {
Kirill A. Shutemov83929372016-07-26 15:26:04 -07002264 put_page(head);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002265 goto repeat;
2266 }
2267
2268 if (!PageUptodate(page) ||
2269 PageReadahead(page) ||
2270 PageHWPoison(page))
2271 goto skip;
2272 if (!trylock_page(page))
2273 goto skip;
2274
2275 if (page->mapping != mapping || !PageUptodate(page))
2276 goto unlock;
2277
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002278 size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2279 if (page->index >= size >> PAGE_SHIFT)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002280 goto unlock;
2281
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002282 if (file->f_ra.mmap_miss > 0)
2283 file->f_ra.mmap_miss--;
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002284
2285 fe->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2286 if (fe->pte)
2287 fe->pte += iter.index - last_pgoff;
2288 last_pgoff = iter.index;
2289 if (alloc_set_pte(fe, NULL, page))
2290 goto unlock;
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002291 unlock_page(page);
2292 goto next;
2293unlock:
2294 unlock_page(page);
2295skip:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002296 put_page(page);
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002297next:
Kirill A. Shutemov7267ec002016-07-26 15:25:23 -07002298 /* Huge page is mapped? No need to proceed. */
2299 if (pmd_trans_huge(*fe->pmd))
2300 break;
Kirill A. Shutemovbae473a2016-07-26 15:25:20 -07002301 if (iter.index == end_pgoff)
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002302 break;
2303 }
2304 rcu_read_unlock();
2305}
2306EXPORT_SYMBOL(filemap_map_pages);
2307
Jan Kara4fcf1c62012-06-12 16:20:29 +02002308int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2309{
2310 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002311 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002312 int ret = VM_FAULT_LOCKED;
2313
Jan Kara14da9202012-06-12 16:20:37 +02002314 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002315 file_update_time(vma->vm_file);
2316 lock_page(page);
2317 if (page->mapping != inode->i_mapping) {
2318 unlock_page(page);
2319 ret = VM_FAULT_NOPAGE;
2320 goto out;
2321 }
Jan Kara14da9202012-06-12 16:20:37 +02002322 /*
2323 * We mark the page dirty already here so that when freeze is in
2324 * progress, we are guaranteed that writeback during freezing will
2325 * see the dirty page and writeprotect it again.
2326 */
2327 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002328 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002329out:
Jan Kara14da9202012-06-12 16:20:37 +02002330 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02002331 return ret;
2332}
2333EXPORT_SYMBOL(filemap_page_mkwrite);
2334
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002335const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002336 .fault = filemap_fault,
Kirill A. Shutemovf1820362014-04-07 15:37:19 -07002337 .map_pages = filemap_map_pages,
Jan Kara4fcf1c62012-06-12 16:20:29 +02002338 .page_mkwrite = filemap_page_mkwrite,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339};
2340
2341/* This is used for a general mmap of a disk file */
2342
2343int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2344{
2345 struct address_space *mapping = file->f_mapping;
2346
2347 if (!mapping->a_ops->readpage)
2348 return -ENOEXEC;
2349 file_accessed(file);
2350 vma->vm_ops = &generic_file_vm_ops;
2351 return 0;
2352}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353
2354/*
2355 * This is for filesystems which do not implement ->writepage.
2356 */
2357int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2358{
2359 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2360 return -EINVAL;
2361 return generic_file_mmap(file, vma);
2362}
2363#else
2364int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2365{
2366 return -ENOSYS;
2367}
2368int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2369{
2370 return -ENOSYS;
2371}
2372#endif /* CONFIG_MMU */
2373
2374EXPORT_SYMBOL(generic_file_mmap);
2375EXPORT_SYMBOL(generic_file_readonly_mmap);
2376
Sasha Levin67f9fd92014-04-03 14:48:18 -07002377static struct page *wait_on_page_read(struct page *page)
2378{
2379 if (!IS_ERR(page)) {
2380 wait_on_page_locked(page);
2381 if (!PageUptodate(page)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002382 put_page(page);
Sasha Levin67f9fd92014-04-03 14:48:18 -07002383 page = ERR_PTR(-EIO);
2384 }
2385 }
2386 return page;
2387}
2388
Mel Gorman32b63522016-03-15 14:55:36 -07002389static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07002390 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002391 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002392 void *data,
2393 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394{
Nick Piggineb2be182007-10-16 01:24:57 -07002395 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 int err;
2397repeat:
2398 page = find_get_page(mapping, index);
2399 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002400 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07002401 if (!page)
2402 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002403 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07002404 if (unlikely(err)) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002405 put_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002406 if (err == -EEXIST)
2407 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 return ERR_PTR(err);
2410 }
Mel Gorman32b63522016-03-15 14:55:36 -07002411
2412filler:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 err = filler(data, page);
2414 if (err < 0) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002415 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002416 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 }
Mel Gorman32b63522016-03-15 14:55:36 -07002418
2419 page = wait_on_page_read(page);
2420 if (IS_ERR(page))
2421 return page;
2422 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 if (PageUptodate(page))
2425 goto out;
2426
Mel Gormanebded022016-03-15 14:55:39 -07002427 /*
2428 * Page is not up to date and may be locked due one of the following
2429 * case a: Page is being filled and the page lock is held
2430 * case b: Read/write error clearing the page uptodate status
2431 * case c: Truncation in progress (page locked)
2432 * case d: Reclaim in progress
2433 *
2434 * Case a, the page will be up to date when the page is unlocked.
2435 * There is no need to serialise on the page lock here as the page
2436 * is pinned so the lock gives no additional protection. Even if the
2437 * the page is truncated, the data is still valid if PageUptodate as
2438 * it's a race vs truncate race.
2439 * Case b, the page will not be up to date
2440 * Case c, the page may be truncated but in itself, the data may still
2441 * be valid after IO completes as it's a read vs truncate race. The
2442 * operation must restart if the page is not uptodate on unlock but
2443 * otherwise serialising on page lock to stabilise the mapping gives
2444 * no additional guarantees to the caller as the page lock is
2445 * released before return.
2446 * Case d, similar to truncation. If reclaim holds the page lock, it
2447 * will be a race with remove_mapping that determines if the mapping
2448 * is valid on unlock but otherwise the data is valid and there is
2449 * no need to serialise with page lock.
2450 *
2451 * As the page lock gives no additional guarantee, we optimistically
2452 * wait on the page to be unlocked and check if it's up to date and
2453 * use the page if it is. Otherwise, the page lock is required to
2454 * distinguish between the different cases. The motivation is that we
2455 * avoid spurious serialisations and wakeups when multiple processes
2456 * wait on the same page for IO to complete.
2457 */
2458 wait_on_page_locked(page);
2459 if (PageUptodate(page))
2460 goto out;
2461
2462 /* Distinguish between all the cases under the safety of the lock */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 lock_page(page);
Mel Gormanebded022016-03-15 14:55:39 -07002464
2465 /* Case c or d, restart the operation */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 if (!page->mapping) {
2467 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002468 put_page(page);
Mel Gorman32b63522016-03-15 14:55:36 -07002469 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 }
Mel Gormanebded022016-03-15 14:55:39 -07002471
2472 /* Someone else locked and filled the page in a very small window */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 if (PageUptodate(page)) {
2474 unlock_page(page);
2475 goto out;
2476 }
Mel Gorman32b63522016-03-15 14:55:36 -07002477 goto filler;
2478
David Howellsc855ff32007-05-09 13:42:20 +01002479out:
Nick Piggin6fe69002007-05-06 14:49:04 -07002480 mark_page_accessed(page);
2481 return page;
2482}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002483
2484/**
Sasha Levin67f9fd92014-04-03 14:48:18 -07002485 * read_cache_page - read into page cache, fill it if needed
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002486 * @mapping: the page's address_space
2487 * @index: the page index
2488 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002489 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002490 *
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002491 * Read into the page cache. If a page already exists, and PageUptodate() is
Sasha Levin67f9fd92014-04-03 14:48:18 -07002492 * not set, try to fill the page and wait for it to become unlocked.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002493 *
2494 * If the page does not get brought uptodate, return -EIO.
2495 */
Sasha Levin67f9fd92014-04-03 14:48:18 -07002496struct page *read_cache_page(struct address_space *mapping,
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002497 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07002498 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002499 void *data)
2500{
2501 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2502}
Sasha Levin67f9fd92014-04-03 14:48:18 -07002503EXPORT_SYMBOL(read_cache_page);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002504
2505/**
2506 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2507 * @mapping: the page's address_space
2508 * @index: the page index
2509 * @gfp: the page allocator flags to use if allocating
2510 *
2511 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06002512 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002513 *
2514 * If the page does not get brought uptodate, return -EIO.
2515 */
2516struct page *read_cache_page_gfp(struct address_space *mapping,
2517 pgoff_t index,
2518 gfp_t gfp)
2519{
2520 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2521
Sasha Levin67f9fd92014-04-03 14:48:18 -07002522 return do_read_cache_page(mapping, index, filler, NULL, gfp);
Linus Torvalds0531b2a2010-01-27 09:20:03 -08002523}
2524EXPORT_SYMBOL(read_cache_page_gfp);
2525
Nick Piggin2f718ff2007-10-16 01:24:59 -07002526/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 * Performs necessary checks before doing a write
2528 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002529 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 * Returns appropriate error code that caller should return or
2531 * zero in case that write should be allowed.
2532 */
Al Viro3309dd02015-04-09 12:55:47 -04002533inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534{
Al Viro3309dd02015-04-09 12:55:47 -04002535 struct file *file = iocb->ki_filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002537 unsigned long limit = rlimit(RLIMIT_FSIZE);
Al Viro3309dd02015-04-09 12:55:47 -04002538 loff_t pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539
Al Viro3309dd02015-04-09 12:55:47 -04002540 if (!iov_iter_count(from))
2541 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542
Al Viro0fa6b002015-04-04 04:05:48 -04002543 /* FIXME: this is for backwards compatibility with 2.4 */
Al Viro2ba48ce2015-04-09 13:52:01 -04002544 if (iocb->ki_flags & IOCB_APPEND)
Al Viro3309dd02015-04-09 12:55:47 -04002545 iocb->ki_pos = i_size_read(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
Al Viro3309dd02015-04-09 12:55:47 -04002547 pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548
Al Viro0fa6b002015-04-04 04:05:48 -04002549 if (limit != RLIM_INFINITY) {
Al Viro3309dd02015-04-09 12:55:47 -04002550 if (iocb->ki_pos >= limit) {
Al Viro0fa6b002015-04-04 04:05:48 -04002551 send_sig(SIGXFSZ, current, 0);
2552 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 }
Al Viro3309dd02015-04-09 12:55:47 -04002554 iov_iter_truncate(from, limit - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 }
2556
2557 /*
2558 * LFS rule
2559 */
Al Viro3309dd02015-04-09 12:55:47 -04002560 if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 !(file->f_flags & O_LARGEFILE))) {
Al Viro3309dd02015-04-09 12:55:47 -04002562 if (pos >= MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 return -EFBIG;
Al Viro3309dd02015-04-09 12:55:47 -04002564 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 }
2566
2567 /*
2568 * Are we about to exceed the fs block limit ?
2569 *
2570 * If we have written data it becomes a short write. If we have
2571 * exceeded without writing data we send a signal and return EFBIG.
2572 * Linus frestrict idea will clean these up nicely..
2573 */
Al Viro3309dd02015-04-09 12:55:47 -04002574 if (unlikely(pos >= inode->i_sb->s_maxbytes))
2575 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576
Al Viro3309dd02015-04-09 12:55:47 -04002577 iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2578 return iov_iter_count(from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579}
2580EXPORT_SYMBOL(generic_write_checks);
2581
Nick Pigginafddba42007-10-16 01:25:01 -07002582int pagecache_write_begin(struct file *file, struct address_space *mapping,
2583 loff_t pos, unsigned len, unsigned flags,
2584 struct page **pagep, void **fsdata)
2585{
2586 const struct address_space_operations *aops = mapping->a_ops;
2587
Nick Piggin4e02ed42008-10-29 14:00:55 -07002588 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002589 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002590}
2591EXPORT_SYMBOL(pagecache_write_begin);
2592
2593int pagecache_write_end(struct file *file, struct address_space *mapping,
2594 loff_t pos, unsigned len, unsigned copied,
2595 struct page *page, void *fsdata)
2596{
2597 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002598
Nick Piggin4e02ed42008-10-29 14:00:55 -07002599 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002600}
2601EXPORT_SYMBOL(pagecache_write_end);
2602
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603ssize_t
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002604generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605{
2606 struct file *file = iocb->ki_filp;
2607 struct address_space *mapping = file->f_mapping;
2608 struct inode *inode = mapping->host;
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002609 loff_t pos = iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002611 size_t write_len;
2612 pgoff_t end;
Al Viro26978b82014-03-10 14:08:45 -04002613 struct iov_iter data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614
Al Viro0c949332014-03-22 06:51:37 -04002615 write_len = iov_iter_count(from);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002616 end = (pos + write_len - 1) >> PAGE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002617
Nick Piggin48b47c52009-01-06 14:40:22 -08002618 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002619 if (written)
2620 goto out;
2621
2622 /*
2623 * After a write we want buffered reads to be sure to go to disk to get
2624 * the new data. We invalidate clean cached page from the region we're
2625 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002626 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002627 */
2628 if (mapping->nrpages) {
2629 written = invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002630 pos >> PAGE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002631 /*
2632 * If a page can not be invalidated, return 0 to fall back
2633 * to buffered write.
2634 */
2635 if (written) {
2636 if (written == -EBUSY)
2637 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002638 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002639 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002640 }
2641
Al Viro26978b82014-03-10 14:08:45 -04002642 data = *from;
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07002643 written = mapping->a_ops->direct_IO(iocb, &data);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002644
2645 /*
2646 * Finally, try again to invalidate clean pages which might have been
2647 * cached by non-direct readahead, or faulted in by get_user_pages()
2648 * if the source of the write was an mmap'ed region of the file
2649 * we're writing. Either one is a pretty crazy thing to do,
2650 * so we don't support it 100%. If this invalidation
2651 * fails, tough, the write still worked...
2652 */
2653 if (mapping->nrpages) {
2654 invalidate_inode_pages2_range(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002655 pos >> PAGE_SHIFT, end);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002656 }
2657
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002659 pos += written;
Al Virof8579f82014-03-03 22:03:20 -05002660 iov_iter_advance(from, written);
Namhyung Kim01166512010-10-26 14:21:58 -07002661 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2662 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 mark_inode_dirty(inode);
2664 }
Al Viro5cb6c6c2014-02-11 20:58:20 -05002665 iocb->ki_pos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002667out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 return written;
2669}
2670EXPORT_SYMBOL(generic_file_direct_write);
2671
Nick Piggineb2be182007-10-16 01:24:57 -07002672/*
2673 * Find or create a page at the given pagecache position. Return the locked
2674 * page. This function is specifically for buffered writes.
2675 */
Nick Piggin54566b22009-01-04 12:00:53 -08002676struct page *grab_cache_page_write_begin(struct address_space *mapping,
2677 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002678{
Nick Piggineb2be182007-10-16 01:24:57 -07002679 struct page *page;
Johannes Weinerbbddabe2016-05-20 16:56:28 -07002680 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002681
Nick Piggin54566b22009-01-04 12:00:53 -08002682 if (flags & AOP_FLAG_NOFS)
Mel Gorman2457aec2014-06-04 16:10:31 -07002683 fgp_flags |= FGP_NOFS;
Nick Piggineb2be182007-10-16 01:24:57 -07002684
Mel Gorman2457aec2014-06-04 16:10:31 -07002685 page = pagecache_get_page(mapping, index, fgp_flags,
Michal Hocko45f87de2014-12-29 20:30:35 +01002686 mapping_gfp_mask(mapping));
Mel Gorman2457aec2014-06-04 16:10:31 -07002687 if (page)
2688 wait_for_stable_page(page);
2689
Nick Piggineb2be182007-10-16 01:24:57 -07002690 return page;
2691}
Nick Piggin54566b22009-01-04 12:00:53 -08002692EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002693
Al Viro3b93f912014-02-11 21:34:08 -05002694ssize_t generic_perform_write(struct file *file,
Nick Pigginafddba42007-10-16 01:25:01 -07002695 struct iov_iter *i, loff_t pos)
2696{
2697 struct address_space *mapping = file->f_mapping;
2698 const struct address_space_operations *a_ops = mapping->a_ops;
2699 long status = 0;
2700 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002701 unsigned int flags = 0;
2702
2703 /*
2704 * Copies from kernel address space cannot fail (NFSD is a big user).
2705 */
Al Viro777eda22014-12-17 04:46:46 -05002706 if (!iter_is_iovec(i))
Nick Piggin674b8922007-10-16 01:25:03 -07002707 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002708
2709 do {
2710 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002711 unsigned long offset; /* Offset into pagecache page */
2712 unsigned long bytes; /* Bytes to write to page */
2713 size_t copied; /* Bytes copied from user */
2714 void *fsdata;
2715
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002716 offset = (pos & (PAGE_SIZE - 1));
2717 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002718 iov_iter_count(i));
2719
2720again:
Linus Torvalds00a3d662015-10-07 08:32:38 +01002721 /*
2722 * Bring in the user page that we will copy from _first_.
2723 * Otherwise there's a nasty deadlock on copying from the
2724 * same page as we're writing to, without it being marked
2725 * up-to-date.
2726 *
2727 * Not only is this an optimisation, but it is also required
2728 * to check that the address is actually valid, when atomic
2729 * usercopies are used, below.
2730 */
2731 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2732 status = -EFAULT;
2733 break;
2734 }
2735
Jan Kara296291c2015-10-22 13:32:21 -07002736 if (fatal_signal_pending(current)) {
2737 status = -EINTR;
2738 break;
2739 }
2740
Nick Piggin674b8922007-10-16 01:25:03 -07002741 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002742 &page, &fsdata);
Mel Gorman2457aec2014-06-04 16:10:31 -07002743 if (unlikely(status < 0))
Nick Pigginafddba42007-10-16 01:25:01 -07002744 break;
2745
anfei zhou931e80e2010-02-02 13:44:02 -08002746 if (mapping_writably_mapped(mapping))
2747 flush_dcache_page(page);
Linus Torvalds00a3d662015-10-07 08:32:38 +01002748
Nick Pigginafddba42007-10-16 01:25:01 -07002749 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
Nick Pigginafddba42007-10-16 01:25:01 -07002750 flush_dcache_page(page);
2751
2752 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2753 page, fsdata);
2754 if (unlikely(status < 0))
2755 break;
2756 copied = status;
2757
2758 cond_resched();
2759
Nick Piggin124d3b72008-02-02 15:01:17 +01002760 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002761 if (unlikely(copied == 0)) {
2762 /*
2763 * If we were unable to copy any data at all, we must
2764 * fall back to a single segment length write.
2765 *
2766 * If we didn't fallback here, we could livelock
2767 * because not all segments in the iov can be copied at
2768 * once without a pagefault.
2769 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002770 bytes = min_t(unsigned long, PAGE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002771 iov_iter_single_seg_count(i));
2772 goto again;
2773 }
Nick Pigginafddba42007-10-16 01:25:01 -07002774 pos += copied;
2775 written += copied;
2776
2777 balance_dirty_pages_ratelimited(mapping);
Nick Pigginafddba42007-10-16 01:25:01 -07002778 } while (iov_iter_count(i));
2779
2780 return written ? written : status;
2781}
Al Viro3b93f912014-02-11 21:34:08 -05002782EXPORT_SYMBOL(generic_perform_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783
Jan Karae4dd9de2009-08-17 18:10:06 +02002784/**
Al Viro81742022014-04-03 03:17:43 -04002785 * __generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002786 * @iocb: IO state structure (file, offset, etc.)
Al Viro81742022014-04-03 03:17:43 -04002787 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002788 *
2789 * This function does all the work needed for actually writing data to a
2790 * file. It does all basic checks, removes SUID from the file, updates
2791 * modification times and calls proper subroutines depending on whether we
2792 * do direct IO or a standard buffered write.
2793 *
2794 * It expects i_mutex to be grabbed unless we work on a block device or similar
2795 * object which does not need locking at all.
2796 *
2797 * This function does *not* take care of syncing data in case of O_SYNC write.
2798 * A caller has to handle it. This is mainly due to the fact that we want to
2799 * avoid syncing under i_mutex.
2800 */
Al Viro81742022014-04-03 03:17:43 -04002801ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802{
2803 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002804 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 struct inode *inode = mapping->host;
Al Viro3b93f912014-02-11 21:34:08 -05002806 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 ssize_t err;
Al Viro3b93f912014-02-11 21:34:08 -05002808 ssize_t status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +01002811 current->backing_dev_info = inode_to_bdi(inode);
Jan Kara5fa8e0a2015-05-21 16:05:53 +02002812 err = file_remove_privs(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 if (err)
2814 goto out;
2815
Josef Bacikc3b2da32012-03-26 09:59:21 -04002816 err = file_update_time(file);
2817 if (err)
2818 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819
Al Viro2ba48ce2015-04-09 13:52:01 -04002820 if (iocb->ki_flags & IOCB_DIRECT) {
Al Viro0b8def92015-04-07 10:22:53 -04002821 loff_t pos, endbyte;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002822
Christoph Hellwig1af5bb42016-04-07 08:51:56 -07002823 written = generic_file_direct_write(iocb, from);
Matthew Wilcoxfbbbad42015-02-16 15:58:53 -08002824 /*
2825 * If the write stopped short of completing, fall back to
2826 * buffered writes. Some filesystems do this for writes to
2827 * holes, for example. For DAX files, a buffered write will
2828 * not succeed (even if it did, DAX does not handle dirty
2829 * page-cache pages correctly).
2830 */
Al Viro0b8def92015-04-07 10:22:53 -04002831 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 goto out;
Al Viro3b93f912014-02-11 21:34:08 -05002833
Al Viro0b8def92015-04-07 10:22:53 -04002834 status = generic_perform_write(file, from, pos = iocb->ki_pos);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002835 /*
Al Viro3b93f912014-02-11 21:34:08 -05002836 * If generic_perform_write() returned a synchronous error
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002837 * then we want to return the number of bytes which were
2838 * direct-written, or the error code if that was zero. Note
2839 * that this differs from normal direct-io semantics, which
2840 * will return -EFOO even if some bytes were written.
2841 */
Al Viro60bb4522014-08-08 12:39:16 -04002842 if (unlikely(status < 0)) {
Al Viro3b93f912014-02-11 21:34:08 -05002843 err = status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002844 goto out;
2845 }
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002846 /*
2847 * We need to ensure that the page cache pages are written to
2848 * disk and invalidated to preserve the expected O_DIRECT
2849 * semantics.
2850 */
Al Viro3b93f912014-02-11 21:34:08 -05002851 endbyte = pos + status - 1;
Al Viro0b8def92015-04-07 10:22:53 -04002852 err = filemap_write_and_wait_range(mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002853 if (err == 0) {
Al Viro0b8def92015-04-07 10:22:53 -04002854 iocb->ki_pos = endbyte + 1;
Al Viro3b93f912014-02-11 21:34:08 -05002855 written += status;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002856 invalidate_mapping_pages(mapping,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002857 pos >> PAGE_SHIFT,
2858 endbyte >> PAGE_SHIFT);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002859 } else {
2860 /*
2861 * We don't know how much we wrote, so just return
2862 * the number of bytes which were direct-written
2863 */
2864 }
2865 } else {
Al Viro0b8def92015-04-07 10:22:53 -04002866 written = generic_perform_write(file, from, iocb->ki_pos);
2867 if (likely(written > 0))
2868 iocb->ki_pos += written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002869 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870out:
2871 current->backing_dev_info = NULL;
2872 return written ? written : err;
2873}
Al Viro81742022014-04-03 03:17:43 -04002874EXPORT_SYMBOL(__generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875
Jan Karae4dd9de2009-08-17 18:10:06 +02002876/**
Al Viro81742022014-04-03 03:17:43 -04002877 * generic_file_write_iter - write data to a file
Jan Karae4dd9de2009-08-17 18:10:06 +02002878 * @iocb: IO state structure
Al Viro81742022014-04-03 03:17:43 -04002879 * @from: iov_iter with data to write
Jan Karae4dd9de2009-08-17 18:10:06 +02002880 *
Al Viro81742022014-04-03 03:17:43 -04002881 * This is a wrapper around __generic_file_write_iter() to be used by most
Jan Karae4dd9de2009-08-17 18:10:06 +02002882 * filesystems. It takes care of syncing the file in case of O_SYNC file
2883 * and acquires i_mutex as needed.
2884 */
Al Viro81742022014-04-03 03:17:43 -04002885ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886{
2887 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002888 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890
Al Viro59551022016-01-22 15:40:57 -05002891 inode_lock(inode);
Al Viro3309dd02015-04-09 12:55:47 -04002892 ret = generic_write_checks(iocb, from);
2893 if (ret > 0)
Al Viro5f380c72015-04-07 11:28:12 -04002894 ret = __generic_file_write_iter(iocb, from);
Al Viro59551022016-01-22 15:40:57 -05002895 inode_unlock(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896
Christoph Hellwige2592212016-04-07 08:52:01 -07002897 if (ret > 0)
2898 ret = generic_write_sync(iocb, ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 return ret;
2900}
Al Viro81742022014-04-03 03:17:43 -04002901EXPORT_SYMBOL(generic_file_write_iter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902
David Howellscf9a2ae2006-08-29 19:05:54 +01002903/**
2904 * try_to_release_page() - release old fs-specific metadata on a page
2905 *
2906 * @page: the page which the kernel is trying to free
2907 * @gfp_mask: memory allocation flags (and I/O mode)
2908 *
2909 * The address_space is to try to release any data against the page
2910 * (presumably at page->private). If the release was successful, return `1'.
2911 * Otherwise return zero.
2912 *
David Howells266cf652009-04-03 16:42:36 +01002913 * This may also be called if PG_fscache is set on a page, indicating that the
2914 * page is known to the local caching routines.
2915 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002916 * The @gfp_mask argument specifies whether I/O may be performed to release
Mel Gorman71baba42015-11-06 16:28:28 -08002917 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002918 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002919 */
2920int try_to_release_page(struct page *page, gfp_t gfp_mask)
2921{
2922 struct address_space * const mapping = page->mapping;
2923
2924 BUG_ON(!PageLocked(page));
2925 if (PageWriteback(page))
2926 return 0;
2927
2928 if (mapping && mapping->a_ops->releasepage)
2929 return mapping->a_ops->releasepage(page, gfp_mask);
2930 return try_to_free_buffers(page);
2931}
2932
2933EXPORT_SYMBOL(try_to_release_page);